Nature - The International Journal of Science / 8 February 2024

Table of contents :
233 Open science — embrace it before it’s too late
234 Cyberattacks on knowledge institutions are increasing- what can be done?
235 Why the mental cost of a STEM career can be too high for women and people of colour
237 Research Highlights
239 Trump’s presidential push renews fears for US science
240 Black-hole observations solve cosmic-ray mystery
241 Signs of ‘transmissible’ Alzheimer’s seen in people who received growth hormone
242 Leading US particle-physics lab faces uncertain future
244 First aircraft to fly on Mars dies — but leaves a legacy of science
245 CRISPR-edited crops break new ground in Africa
246 Obesity drugs have another superpower- taming inflammation
248 The new car batteries that could power the electric vehicle revolution
252 Santorini’s volcanic past- underwater clues reveal giant prehistoric eruption
254 It’s time to admit that genes are not the blueprint for life
256 Science and government- can the power struggle ever end?
258 No ‘easy’ weight loss- don’t overlook the social cost of anti-obesity drugs
261 Correspondence
263 Mimas’s surprise ocean prompts an update of the rule book for moons
264 Stone tools in northern Europe made by Homo sapiens 45,000 years ago
266 Resting restores performance of discharged lithium-metal batteries
267 The journey to understand previously unknown microbial genes
269 Natural inhibitor found for cell death by ferroptosis
271 A break in mitochondrial endosymbiosis as a basis for inflammatory diseases
Implications of endosymbiotic origin
Remnants of the endosymbiotic origin of mitochondria
Mitochondrial signals to the cytosol
Inheritance or exaptation?
Sterile and non-sterile inflammation overlap
Endosymbiosis and autoimmunity
Emerging aspects
Outlook
Acknowledgements
Fig. 1 Pathways of molecular signal release from mitochondria.
Fig. 2 How breakdown in endosymbiosis can lead to inflammation.
Fig. 3 Did nucleic acid-sensing PRRs evolve to sense mitochondrial nucleic acids? mtDNA has been shown to be sensed by the PRRs cGAS and TLR9 while mitochondrial dsRNA can be sensed by the PRRs MDA-5 and RIG-1.
280 A recently formed ocean inside Saturn’s moon Mimas
Online content
Fig. 1 Mimas measurements and ocean models.
Fig. 2 Mimas’s interior and orbital evolution.
Extended Data Fig. 1 Reprocessing of Mimas astrometry.
Extended Data Fig. 2 Core radius for solid model.
Extended Data Fig. 3 Sensitivity of shape parameters.
Extended Data Table 1 Estimation of Mimas gravity field.
283 Ultracold field-linked tetratomic molecules
Field-linked tetramers
Binding energy and lifetime
Association and dissociation processes
Imaging of the dissociated tetramers
Discussion
Conclusion
Online content
Fig. 1 Electroassociation of field-linked tetramers.
Fig. 2 Tetramer binding energy and lifetime.
Fig. 3 Association and dissociation processes.
Fig. 4 Momentum distributions of the dissociated tetramers.
Extended Data Fig. 1 Dimer loss near the FL resonance.
Extended Data Fig. 2 Conditions for efficient electroassociation.
Extended Data Fig. 3 Tetramer lifetime in trap and in time-of-flight.
Extended Data Fig. 4 Hyperfine transitions of NaK molecules in the modulation spectra.
Extended Data Fig. 5 Tetramer dissociation patterns and their angular distribution.
Extended Data Fig. 6 Theoretical tetramer decay rate.
288Observation and quantification of the pseudogap in unitary Fermi gases
Observation and quantification of the pseudogap in unitary Fermi gases
Experimental scheme and set-up
Fermion spectral function
Fermion self-energy and EDCs
Summary
Online content
Fig. 1 Experimental scheme.
Fig. 2 Microwave spectra at 0.
Fig. 3 Momentum-resolved microwave spectra at various temperatures across the superfluid transition.
Fig. 4 Temperature dependence of Δ, Γ and the EDCs.
Extended Data Fig. 1 Pair momentum distributions in the vicinity of the superfluid phase transition.
Extended Data Fig. 2 Schematic diagram for the magnetic field stabilization.
Extended Data Fig. 3 The residual 50 Hz noise and Rabi oscillations between the |3⟩ and |4⟩ hyperfine levels.
Extended Data Fig. 4 Density distribution after ballistic expansion and n(k, Δω) of the unitary Fermi gas at 0.
Extended Data Fig. 5 The contour plots of A(k, ω).
Extended Data Fig. 6 Analysis of the energy dispersion.
Extended Data Fig. 7 The evolution of EDC as a function of k for various T.
Extended Data Fig. 8 Temperature dependence of Δ, and Uh.
Extended Data Table 1 Temperature dependence of Δ, m*, and U.
294Evidence of superconducting Fermi arcs
Evidence of superconducting Fermi arcs
Three-dimensional band structure
Surface states on two terminations
Robust Fermi arcs from laser-ARPES
Superconductivity at the surface
Online content
Fig. 1 3D band structure of PtBi2.
Fig. 2 Fermi arcs.
Fig. 3 Laser-ARPES.
Fig. 4 Superconducting arcs.
Extended Data Fig. 1 Fermi surface maps.
Extended Data Fig. 2 Photon energy dependence of the Fermi arcs.
Extended Data Fig. 3 3D band structure.
Extended Data Fig. 4 EDC across the Fermi arc.
Extended Data Fig. 5 Theoretical calculations of gap opening in PtBi2.
Extended Data Fig. 6 Polarization dependent datasets.
300 Stable blue phosphorescent organic LEDs that use polariton-enhanced Purcell effects
PEP and SPP dispersion engineering
Purcell effect and energy transfer rates
PHOLED performance
Online content
Fig. 1 The PEP-enhanced Purcell effect.
Fig. 2 Polariton dispersion engineering.
Fig. 3 Optical engineering of blue PHOLEDs.
Fig. 4 Ir(dmp)3 device performance.
Fig. 5 Ir(dmp)3 device performance summary.
Extended Data Fig. 1 Molecular structural formulae of organic materials used in the EML, ETL and hosts.
Extended Data Fig. 2 Angle-resolved TM-mode reflectance of Al/BPyTP2.
Extended Data Fig. 3 Device structures used in this study.
Extended Data Fig. 4 Simulated and measured PFs for the devices studied.
Extended Data Fig. 5 Ir(dmp)3 device energy levels and performance.
Extended Data Fig. 6 Ir(cb)3 device energy levels and performance.
Table 1 Summary of Ir(dmp)3 PHOLED performance, in which device performances for control (C) and full (F) cavity devices are compared.
Extended Data Table 1 Summary of Ir(dmp)3 and Ir(cb)3 device performance.
Extended Data Table 2 Summary of stretched exponential model for Ir(dmp)3 and Ir(cb)3 devices.
306Recovery of isolated lithium through discharged state calendar ageing
Recovery of isolated lithium through discharged state calendar ageing
Discharged state capacity recovery
Operando optical cell setup
Proposed Li recovery mechanism
Online content
Fig. 1 Capacity recovery from discharged state rest illustrated by CE and TGC data.
Fig. 2 Operando optical microscopy of Li isolation and reconnection under continuous cycling.
Fig. 3 i-Li areal comparison between rested and continuously cycled optical cells.
Fig. 4 Rest-induced SEI dissolution and overpotential reduction.
Extended Data Fig. 1 Hybrid cycling protocol and corresponding CE.
Extended Data Fig. 2 Various charging capacity hybrid cycle performance.
Extended Data Fig. 3 Various discharge current density hybrid cycle performance.
Extended Data Fig. 4 Various rest time hybrid cycle performance.
Extended Data Fig. 5 Various rest cycle number hybrid cycle performance.
Extended Data Fig. 6 Optical cell assembly and cycle performance.
Extended Data Fig. 7 Optical cell colored areal maps of isolated and recovered Li.
Extended Data Fig. 8 LFP | |Cu pouch cell and long cycle Li | |Cu half-cells cycle performance.
Extended Data Fig. 9 SEI dissolution.
Extended Data Fig. 10 NMR and XPS analysis of discharge rested SEI and electrolyte.
313 A rechargeable calcium–oxygen battery that operates at room temperature
Online content
Fig. 1 Rechargeable Ca–O2 batteries at room temperature with CaO2 as the main discharge product.
Fig. 2 Cathode reaction of Ca–O2 batteries involves reversible two-electron O2/CaO2 chemistry.
Fig. 3 Optimized electrolyte facilitates stable operation of Ca–O2 batteries.
Fig. 4 Ca–O2 batteries are suitable for practical applications.
Extended Data Fig. 1 Characterization of CNT air cathode and cycling performance of the Ca-O2 battery.
Extended Data Fig. 2 Characterization of the discharge product in Ca-O2 batteries.
Extended Data Fig. 3 Reversibility of CaO2 formation/decomposition in Ca-O2 cell chemistry.
Extended Data Fig. 4 Characterization of Ca metal anode disassembled from Ca-O2 batteries.
Extended Data Fig. 5 Evaluation of oxidation stability of the electrolyte.
Extended Data Fig. 6 Properties of the optimized electrolyte containing DMSO.
Extended Data Fig. 7 Electrochemical performance of Ca metal anode in electrolytes with and without DMSO.
Extended Data Fig. 8 Improved Ca2+ de-solvation and Ca plating/stripping.
Extended Data Fig. 9 Electrochemical performance of Ca-O2 batteries under practical conditions.
Extended Data Fig. 10 Fabrication and electrochemical performances of fibre Ca-O2 batteries.
Extended Data Table 1 Atomic percentage (at%) of the elements observed in XPS spectra of Ca deposits in anode disassembled from our Ca-O2 batteries at different sputtering time.
Extended Data Table 2 The parameters measured by potentiostatic polarization and electrochemical impedance spectroscopy for calculating the Ca2+ transference number.
319Elevated Southern Hemisphere moisture availability during glacial periods
Elevated Southern Hemisphere moisture availability during glacial periods
Speleothem growth during glacial periods
A pervasive cool-moist glacial pattern
Extent of the cool-moist glacial pattern
Implications of a cool-moist subtropics
Online content
Fig. 1 CMI for the three main landmasses of the Southern Hemisphere subtropics and study sites.
Fig. 2 Two new speleothem proxy records of subtropical moisture in southern Australia show a cool-moist pattern.
Fig. 3 Naracoorte fossil pollen record and moisture reconstructions confirm Naracoorte KDE pattern.
Fig. 4 Southern Hemisphere subtropical hydroclimate proxy records show a widespread cool-moist pattern.
Fig. 5 Modelled responses of Southern Hemisphere hydroclimate proxy records to Southern Hemisphere temperature.
Fig. 6 Modelled responses of Southern Hemisphere hydroclimate proxy records to hemispheric temperature show the latitudinal limits of the subtropical cool-moist pattern.
Extended Data Fig. 1 Spectral analyses shows coherence and typically small phase lags with Southern Hemisphere summer insolation.
Extended Data Fig. 2 Pollen-based quantitative climate reconstructions for Naracoorte.
Extended Data Fig. 3 Climatically sensitive plant taxa’s geographic ranges within eastern Australia.
Extended Data Fig. 4 Climatically sensitive plant taxa’s climatic ranges in summer and winter seasons.
Extended Data Fig. 5 Correlation plots of HadCM3 simulations vs.
Extended Data Fig. 6 Three subtropical lake basin hydroclimate records show a cool-moist pattern.
Extended Data Fig. 7 Hydroclimate proxy records demonstrating the equatorward and poleward limits of the subtropical cool-wet pattern.
Extended Data Fig. 8 Power spectra with red noise threshold reveal significant precessional frequencies.
Extended Data Fig. 9 Naracoorte speleothem sample thickness/diameter versus sample age confirms there is no size-based preservation bias.
Extended Data Fig. 10 Southern Hemisphere subtropical DJF and JJA precipitation (mm).
Extended Data Fig. 11 Micrographs of selected pollen and spores from Naracoorte speleothems.
Extended Data Table 1 Modern climatic data for the Naracoorte Cave Complex and the Leeuwin-Naturaliste caves in southern Australia.
327 Country-specific net-zero strategies of the pulp and paper industry
Global and national paper-related GHG emissions
Scenario analysis on future GHG emissions
National strategies towards net-zero emissions by 2050
Decarbonization options through a systems approach
Priority for energy-related measures
Sustainable and diversified sourcing
Improved waste management and recycling strategies
Online content
Fig. 1 Global GHG emissions of paper-related sectors during 1961–2019.
Fig. 2 Total net GHG emissions of paper-related sectors in regions and countries.
Fig. 3 GHG emissions of paper-related sectors in 30 countries from 1961 to 2019.
Fig. 4 Effects of single measures and all scenarios in 30 countries.
Fig. 5 The analysis of strategies towards net-zero emissions in 2050.
Extended Data Fig. 1 System definition and GHG emissions inventory.
Extended Data Fig. 2 Diagram of global energy consumption of pulping, papermaking and printing accumulated in 1961–2019.
Extended Data Fig. 3 Global GHG emissions of all processes accumulated in 1961–2019.
Extended Data Fig. 4 Detailed information about forest carbon emissions in the first ten countries in Fig.
Extended Data Fig. 5 Net GHG emissions under three recycling measures in 30 countries when no other measures are taken.
Extended Data Fig. 6 Net GHG emissions of 30 countries in 2019 and 2050 under the BAU scenario and 16 single-measure scenarios.
Extended Data Fig. 7 Statistics of factor scenario settings of net-zero scenarios.
Extended Data Fig. 8 Distribution of net-zero scenarios by the number of best or medium measures.
Extended Data Fig. 9 Carbon intensity of energy consumption in S2 and S3 across 30 countries from 1961 to 2019.
Extended Data Fig. 10 Analytical framework of forest carbon emissions estimation.
Extended Data Table 1 The grouping principles by difficulty of net-zero achievement for 30 countries.
Extended Data Table 2 The setting of factors in scenario analysis.
335 Predator mass mortality events restructure food webs through trophic decoupling
Trophic biomass responses
Community structural responses
Community biomass dynamics
Online content
Fig. 1 Food-web biomass responses to predator removals, resource pulses and MMEs.
Fig. 2 Community-wide biomass, density and functional trait responses to predator removals, resource pulses and MMEs.
Fig. 3 Distinct zooplankton and microalgal community biomass trajectories after predator removals, resource pulses and MMEs.
Extended Data Fig. 1 Time series of mean zooplankton and microalgae density following ecological perturbations.
Extended Data Fig. 2 Time series of mean zooplankton body size and microalgae biovolume following ecological perturbations.
Extended Data Fig. 3 Time series of mean biomass across five major zooplankton families following ecological perturbations.
Extended Data Fig. 4 Time series of mean biomass across five major microalgae phyla following ecological perturbations.
Extended Data Fig. 5 Time series of mean density across five major zooplankton families following ecological perturbations.
Extended Data Fig. 6 Time series of mean density across five major microalgae phyla following ecological perturbations.
Extended Data Fig. 7 Time series of mean body size across five major zooplankton families following ecological perturbations.
Extended Data Fig. 8 Time series of mean biovolume across two major microalgae phyla following ecological perturbations.
Extended Data Fig. 9 Raw biomass estimates during sample periods before ecological perturbations were induced (i.
Extended Data Fig. 10 Raw (non-smoothed) and smoothed biomass estimates following ecological perturbations.
341 Homo sapiens reached the higher latitudes of Europe by 45,000 years ago
Online content
Fig. 1 Stratigraphy with location of H.
Fig. 2 Chronological comparison of Ranis with selected contemporary sites and directly dated human remains.
Fig. 3 Bayesian phylogenetic tree of the newly reconstructed mtDNA genomes with previously published ancient and recent modern human mtDNA genomes constructed with BEAST2.
Extended Data Fig. 1 Ranis site plan and main West profile of the 2016–2022 excavation.
Extended Data Fig. 2 Map of Ranis with the location of the newly identified hominin specimens and selected lithic artefacts.
Extended Data Fig. 3 Chronological site model of 2016-2022 material from Ranis.
Extended Data Fig. 4 Protein deamidation for all hominin specimens in Ranis.
Extended Data Fig. 5 Proteomic coverage for the seven hominin bone specimens analysed with SPIN.
Extended Data Table 1 Hominin specimens identified in Ranis.
347 A dedicated hypothalamic oxytocin circuit controls aversive social learning
One-day defeat induces avoidance and fear
Winner cues drive loser aVMHvlOXTR cells
The aVMHvlOXTR response is specific to social contexts
Avoidance expression requires aVMHvlOXTR cells
Social avoidance learning requires OXTR
The SOR provides oxytocin to aVMHvlOXTR cells
Oxytocin facilitates synaptic potentiation
Noxious stimuli activate SOROXT cells
SOROXT cells boost social avoidance learning
Discussion
Online content
Fig. 1 aVMHvlOXTR cells in male mice show increased responses to aggressors after defeat.
Fig. 2 aVMHvlOXTR cells bidirectionally modulate social avoidance.
Fig. 3 OXTRs in the aVMHvl are essential for defeat-induced social avoidance learning.
Fig. 4 The SOR is the primary source of oxytocin for aVMHvlOXTR cells.
Fig. 5 SOROXT cells are activated by noxious stimuli.
Fig. 6 SOROXT cells are essential for social avoidance learning.
Extended Data Fig. 1 One-day 10-min social defeat is sufficient to induce social avoidance of winner-like conspecifics.
Extended Data Fig. 2 Defeated animals do not avoid conspecifics with genetic backgrounds different from the aggressor.
Extended Data Fig. 3 The relationship between OXTR and defeat-induced c-Fos and Esr1 in the VMHvl.
Extended Data Fig. 4 aVMHvlOXTR cells increase responses to the aggressor after defeat in male mice.
Extended Data Fig. 5 Female aVMHvlOXTR cells increase responses to the aggressor after defeat.
Extended Data Fig. 6 aVMHvlOXTR cells increase response to the aggressor after defeat in female mice.
Extended Data Fig. 7 Defeat experience enhances aggressor cue-induced c-Fos in aVMHvlOXTR cells during subsequent encounters.
Extended Data Fig. 8 No change in excitability of aVMHvlOXTR cells one day after defeat.
Extended Data Fig. 9 aVMHvlOXTR cells do not respond to non-social aversive odors.
Extended Data Fig. 10 aVMHvlOXTR cells do not respond to non-social aversive cues or noxious somatosensory stimuli in head-fixed animals.
Extended Data Fig. 11 Behavior changes induced by optogenetic activation of aVMHvlOXTR and SOROXT cells.
Extended Data Fig. 12 SOROXT affects aVMHvl cell activity by activating OXTR, not glutamatergic synaptic transmission.
Extended Data Fig. 13 Overlap between OXT and defeat-induced c-Fos.
Extended Data Fig. 14 SOROXT cells do not increase responses to aggressors after defeat in male mice.
Extended Data Fig. 15 SOROXT cells in female mice are activated by noxious stimuli.
357 Hypoblast from human pluripotent stem cells regulates epiblast development-
Naive hPSC-induced hypoblast by GATA6
Hypoblast induced by signalling molecules
FGF/BMP for hypoblast specification
Generation of bilaminoids
Epiblast progression via TB-secreted IL-6
Mesoderm-like cells emerge in bilaminoids
Single-cell transcriptomics of bilaminoids
Anterior–posterior axis formation in bilaminoids
nHyCs support epiblast progression
Lineage specification in bilaminoids
Discussion
Online content
Fig. 1 Naive hPSC differentiation into the PDGFRA+ hypoblast by GATA6 overexpression.
Fig. 2 Essential signalling for human hypoblast specification.
Fig. 3 Naive hPSCs and nHyCs generate bilaminoids.
Fig. 4 TB enhances epiblast progression through IL-6 paracrine signalling.
Fig. 5 Global gene expression profiles of individual cells in bilaminoids.
Fig. 6 Bilaminoids recapitulate human pregastrulation.
Extended Data Fig. 1 Naïve and Primed hPSCs and GATA6 overexpression.
Extended Data Fig. 2 Transcriptome analysis after GATA6 overexpression in naïve and primed G6-PDGFRA+ cells.
Extended Data Fig. 3 Naïve hPSCs differentiate into hypoblast lineage with 7F without GATA6 overexpression.
Extended Data Fig. 4 Surface markers and signalling molecules of human hypoblast cells.
Extended Data Fig. 5 Signalling for hypoblast specification differs between humans and mice.
Extended Data Fig. 6 Bilaminoids generated by naïve hPSCs and nHyC.
Extended Data Fig. 7 Trophoblast enhances epiblast progression.
Extended Data Fig. 8 Global gene expression profiles and anterior-posterior axis formation of bilaminoids.
Extended Data Fig. 9 LAMB1 knockout nHyC and gene expression profiles of bilaminoids.
Extended Data Fig. 10 Bilaminoids on D9 and interspecies chimaera assays.
367Modelling post-implantation human development to yolk sac blood emergence
Modelling post-implantation human development to yolk sac blood emergence
Epiblast and hypoblast codevelopment
Intra- and/or extra-embryonic scRNA trajectories
Specification of amniotic ectoderm
Anterior hypoblast and posterior domain
Yolk sac mesoderm and blood progenitors
Yolk sac-like haematopoiesis
Cell composition of haematopoietic waves
Discussion
Online content
Fig. 1 Engineering codevelopment of embryonic and extra-embryonic endoderm tissues.
Fig. 2 Amniotic cavity formation and expansion.
Fig. 3 Anterior hypoblast domain and posterior pole in heX-embryoids.
Fig. 4 Haematopoietic lineages and haematopoietic foci structures in the heX-embryoids.
Fig. 5 Haematopoietic programme characterization in heX-embryoids.
Extended Data Fig. 1 Fate acquisition, sorting, and symmetry breaking following GATA6 induction.
Extended Data Fig. 2 Lumen development and optimization within WT cluster.
Extended Data Fig. 3 Single Cell RNA-seq analysis and clustering per day (day 0 to 5).
Extended Data Fig. 4 Hypergeometric statistical comparison of heX-embryoid time points to human and NHP embryo data.
Extended Data Fig. 5 Merged clustering of RNA-seq Day 0 – Day 5 of the embryoids.
Extended Data Fig. 6 Amnion and anteroposterior domains.
Extended Data Fig. 7 Identification of endothelial/hematopoietic populations in heX-embryoids.
Extended Data Fig. 8 GATA6-hi supplementation and structure of hematopoietic foci.
Extended Data Fig. 9 Hematopoietic foci in heX-embryoids.
Extended Data Fig. 10 Hematopoietic cell composition in heX-embryoids.
Extended Data Fig. 11 heX-embryoid formation from hiPSCs to model human early post-implantation development in vitro.
Extended Data Fig. 12 heX-embryoid development, passaging, cryostorage as well as engineering in a separate iPSC line.
377 Functional and evolutionary significance of unknown genes from uncultivated taxa
A curated catalogue of novel gene families
Functional predictions
Hypothesis-driven functional validations
Density of novel families per genome
Synapomorphies in uncultivated taxa
Habitat distribution of novel families
Discovery of new biomarkers
Discussion
Online content
Fig. 1 Gene family discovery pipeline and general statistics.
Fig. 2 Distribution of FESNov gene families confidently linked to KEGG pathways.
Fig. 3 FESNov gene families are spread across the entire microbial phylogeny, covering a variety of habitats.
Fig. 4 FESNov synapomorphic gene families found at high-level taxonomic rank.
Fig. 5 FESNov gene families contribute to CRC predictive power and include biomarkers for the disease.
Extended Data Fig. 1 Swimming chemotaxis assay of Escherichia coli W3110 strain expressing NOV3845Y.
Extended Data Fig. 2 Antimicrobial activity of NOVOQR9B peptide.
Extended Data Fig. 3 Structural similarity of NOV5WD8W with the copper metallochaperone CusF.
Extended Data Fig. 4 Schematic representation of the genomic context of four transmembrane FESNov gene families in Patescibacteria genomes.
Extended Data Fig. 5 Correlation between FESNov gene families mobility and ecological dispersion.
Extended Data Fig. 6 Number of FESNov gene families confined to each taxonomic rank.
Extended Data Fig. 7 Separation of habitats and human gut populations with FESNov families and KO relative abundances.
Extended Data Fig. 8 Examples of the genomic context of FESNov families over-abundant in CRC samples.
Extended Data Fig. 9 Performance of predictors built upon the relative abundance matrices of both FESNov gene families and KEGG Orthologs (KOs) families.
385 Mucosal boosting enhances vaccine protection against SARS-CoV-2 in macaques
Study design
Mucosal and peripheral humoral responses
Mucosal and peripheral T cell responses
Protective efficacy
Histopathology
Lung transcriptomics and cytokines
Discussion
Online content
Fig. 1 Study outline.
Fig. 2 Mucosal and peripheral SARS-CoV-2 NAb responses.
Fig. 3 Mucosal and peripheral IgA spike-specific binding antibody responses.
Fig. 4 Mucosal and peripheral T cell responses.
Fig. 5 Viral loads after SARS-CoV-2 BQ.
Fig. 6 Transcriptomics and cytokine analyses in the BAL.
Extended Data Fig. 1 Comparison of week 4 immune responses in BAL.
Extended Data Fig. 2 Mucosal and peripheral IgA spike-specific binding antibody responses by ECLA.
Extended Data Fig. 3 Mucosal and peripheral IgG spike-specific binding antibody responses by ELISA.
Extended Data Fig. 4 Mucosal and peripheral IgG spike-specific binding antibody responses by ECLA.
Extended Data Fig. 5 Sample flow cytometry gating.
Extended Data Fig. 6 Anamnestic SARS-CoV-2 neutralizing antibody responses following SARS-CoV-2 BQ.
Extended Data Fig. 7 Immune correlates of protection.
Extended Data Fig. 8 Histopathology.
Extended Data Fig. 9 Fibrosis scores.
Extended Data Fig. 10 Additional transcriptomics analyses in the BAL.
392 Prevention of respiratory virus transmission by resident memory CD8+ T cells
CD8+ TRM cells limit the transmission window
IFNγ is required to limit transmission
TRM cells limit susceptibility to infection
IFNγ alters epithelial cell programming
CD8+ TRM cells can provide durable protection
Discussion
Online content
Fig. 1 Respiratory tract CD8+ TRM cells can limit transmission of respiratory viruses.
Fig. 2 IFNγ signalling has an essential role in preventing transmission of respiratory viruses.
Fig. 3 Respiratory tract CD8+ TRM cells protect against viral propagation following transmission through IFNγ.
Fig. 4 IFNγ signalling induces antiviral gene expression and increases antigen presentation in nasal cavity epithelial cells.
Fig. 5 The number of respiratory tract CD8+ TRM cells is strongly linked to protection from transmission.
Extended Data Fig. 1 Distribution and characterization of tissue-resident Sendai-specific CD8+ T cells limit following intranasal and intraperitoneal infection with recombinant influenza virus.
Extended Data Fig. 2 Sendai-luciferase bioluminescence strongly correlates with viral titer.
Extended Data Fig. 3 Contact mice that show no indication of transmission by bioluminescence also fail to develop a Sendai-specific T cell response.
Extended Data Fig. 4 Number of SenNP-specific CD8+ TRM in knockout mouse strains following immunization.
Extended Data Fig. 5 Immunization does not alter influx of NK cells and monocytes following Sendai virus transmission.
Extended Data Fig. 6 Sendai-specific TRM numbers and assessment of transmission under different immunization strategies at 1- and 6-months post-immunization.
Extended Data Fig. 7 High viral burden in index mice does not correlate with increased viral burden in contact mice.
Extended Data Fig. 8 Pre-existing immunity to related influenza strains limits the efficacy of protective T cell immunity induced by LAIV-SenNP immunization but can be overcome by Ad-SenNP immunization.
Extended Data Fig. 9 Heterologous influenza prime-boost does not improve the durability of respiratory tract TRM.
401 7-Dehydrocholesterol is an endogenous suppressor of ferroptosis
DHCR7 is a proferroptotic gene
7-DHC is an antiferroptotic metabolite
7-DHC blocks phospholipid peroxidation
Truncated phospholipids drive cell lysis
7-DHC accumulation increases cell fitness
Discussion
Online content
Fig. 1 Identification and impact of DHCR7 deficiency on ferroptosis.
Fig. 2 7-DHC accumulation suppresses ferroptosis.
Fig. 3 7-DHC acts to suppress (phospho)lipid peroxidation.
Fig. 4 Phospholipid truncated species contribute to ferroptosis execution.
Fig. 5 Impact of 7-DHC accumulation on lymphoma growth.
Extended Data Fig. 1 Lipidomic characterization of DHCR7-deficient cells.
Extended Data Fig. 2 DHCR7 deficiency impact on ferroptosis and other cell death modalities.
Extended Data Fig. 3 Characterization of HT1080 DHCR7-deficient clonal cell line.
Extended Data Fig. 4 Impact of 7-DHC accumulation on ferroptosis.
Extended Data Fig. 5 Influence of cholesterol low conditions on the antiferroptotic activity of the 7-DHC/DHCR7 axis.
Extended Data Fig. 6 Role of B-ring unsaturated sterol in ferroptosis.
Extended Data Fig. 7 Impact and consequence of 7-DHC on phospholipid peroxidation.
Extended Data Fig. 8 Impact of ferroptosis inhibitors on oxidant mediated liposomal rupture.
Extended Data Fig. 9 Role of truncated phospholipid in membrane permeability.
Extended Data Fig. 10 Conjugation at the omega position affects ferroptosis sensitivity.
Extended Data Fig. 11 Impact of 7-DHC accumulation on BL growth.
Extended Data Fig. 12 Impact of DHCR7 loss in vivo.
411 7-Dehydrocholesterol dictates ferroptosis sensitivity
Distal cholesterol biosynthesis regulates ferroptosis
7-DHC suppresses ferroptosis
7-DHC suppresses phospholipid peroxidation
7-DHC regulates tumour ferroptosis
7-DHC protects the kidneys from IRI
Discussion
Online content
Fig. 1 The genes involved in distal cholesterol synthesis differentially regulate ferroptosis.
Fig. 2 7-DHC suppresses ferroptosis.
Fig. 3 7-DHC shields plasma and mitochondria membranes from autoxidation.
Fig. 4 Targeting 7-DHC biosynthesis regulates cancer cell sensitivity to ferroptosis.
Fig. 5 7-DHC attenuates IRI in vivo.
Extended Data Fig. 1 Identification of distal CB pathway including CYP51A1, MSMO1, EBP and SC5D as ferroptosis suppressors.
Extended Data Fig. 2 The distal cholesterol biosynthesis genes regulates ferroptosis.
Extended Data Fig. 3 Distal CB pathway regulates ferroptosis independent of cholesterol and known ferroptosis defence system.
Extended Data Fig. 4 7-DHC suppresses ferroptosis.
Extended Data Fig. 5 7-DHC is a general suppressor of ferroptosis.
Extended Data Fig. 6 7-DHC protects against phospholipid peroxidation.
Extended Data Fig. 7 Regulation of 7-DHC level during ferroptosis.
Extended Data Fig. 8 Targeting 7-DHC biosynthesis regulates cancer cell sensitivity to ferroptosis.
Extended Data Fig. 9 7-DHC promotes metastasis.
Extended Data Fig. 10 Cariprazine protects IRI in vivo.
419 Reverse metabolomics for the discovery of chemical structures from humans
Reverse metabolomics
Acyl amide and ester searches
Analysis of bile amidates
Validation of the IBD association
Microbial production of bile amides
Conclusion
Online content
Fig. 1 Overview of reverse metabolomics and the synthetic strategies used to obtain standards for MS/MS in this work.
Fig. 2 Repository-scale analysis of public MS data.
Fig. 3 IBD association of new conjugated bile acids.
Fig. 4 Bile acid conjugations observed in HMP isolates cultured in fecal growth medium containing CA and DCA.
Extended Data Fig. 1 Representative fragmentation of standard vs observed MS/MS in public data with key fragment ions shown.
Extended Data Fig. 2 Results of MASST searches for N-acyl amides.
Extended Data Fig. 3 Results of MASST searches for acyl esters.
Extended Data Fig. 4 Analysis of synthetic conjugated bile acid mixtures.
Extended Data Fig. 5 Independent validation of new conjugated bile acids in PRISM human IBD cohort.
Extended Data Fig. 6 Overview of conjugated bile acids in iHMP2 human IBD cohort.
Extended Data Fig. 7 PXR activity of conjugated bile acids.
Extended Data Fig. 8 LC-IMS-MS Analysis of Bacterial Cultures.
427Transport and inhibition mechanisms of human VMAT2
Transport and inhibition mechanisms of human VMAT2
Function and architecture of VMAT2
5-HT recognition by VMAT2
Titratable residues within the cavity
Molecular dynamics simulations of 5-HT binding
Essential role of E312 and R189
Inhibition of VMAT2 by reserpine
VMAT2 inhibition by TBZ
VMAT2 inhibition by KET
State transitions of VMAT2
Discussion
Online content
Fig. 1 Functional characterization and structure of human VMAT2.
Fig. 2 Structural basis of 5-HT recognition by VMAT2.
Fig. 3 Structural basis of VMAT2 inhibition by reserpine.
Fig. 4 Structural basis of VMAT2 inhibition by TBZ.
Fig. 5 Structural basis of VMAT2 inhibition by KET.
Fig. 6 Structural basis of VMAT2 state transitions.
Extended Data Fig. 1 Purification of human VMAT2.
Extended Data Fig. 2 Cryo-EM analysis of human VMAT25HT.
Extended Data Fig. 3 Cryo-EM data processing of VMAT2RES.
Extended Data Fig. 4 Cryo-EM data processing of VMAT2TBZ and VMAT2KET.
Extended Data Fig. 5 Structures of VMAT2 in three distinct states.
Extended Data Fig. 6 Comparison of the hydrogen bond network of VMAT2 in different states.
Extended Data Fig. 7 MD simulation of 5HT binding to the lumen-facing VMAT2.
Extended Data Fig. 8 Sequence alignment of SLC18 transporters.
Extended Data Fig. 9 Function and transport model of VMAT2.
Extended Data Table 1 Cryo-EM data collection, refinement and validation statistics.
435 De novo design of high-affinity binders of bioactive helical peptides
Design of peptide-binding scaffolds
Designing binders using Hallucination
Design refinement with RFdiffusion
De novo binder design using RFdiffusion
Human versus machine problem solving
Design of protein biosensors
Enrichment for LC–MS/MS detection
Discussion
Online content
Fig. 1 Design strategies for binding helical peptides.
Fig. 2 Peptide binder optimization with RFdiffusion.
Fig. 3 De novo peptide binder design with RFdiffusion.
Fig. 4 Application of designed binders to sensing and detection.
Extended Data Fig. 1 Low affinity RFjoint-Inpainted binders for NPY and GCG using extended parametric designs.
Extended Data Fig. 2 Additional binders made using threading and redesign.
Extended Data Fig. 3 Hallucinated Bid binders are stable and bind Bid peptide with high affinity.
Extended Data Fig. 4 Partial diffusion increases designability of native proteins.
Extended Data Fig. 5 PTH and GCG binders designed with RFdiffusion.
Extended Data Fig. 6 LC-MS/MS chromatograms for PTH and GCG binders.
Extended Data Table 1 Amino acid sequences of peptide binders.
Extended Data Table 2 Crystallographic data collection and refinement.
443 Organize your –80 °C freezer to save time and prevent frozen fingertips
444 How a peer network made my worst day as a grad student bearable
448 I started fossil hunting in my 60s — now I have more than 2,000 pieces
e4 Author Correction- Short tRNA anticodon stem and mutant eRF1 allow stop codon reassignment
e5 Author Correction- Aerial additive manufacturing with multiple autonomous robots
e6 Author Correction- R-loop-derived cytoplasmic RNA–DNA hybrids activate an immune response
e7 Publisher Correction- Intermediate conformations of CD4-bound HIV-1 Env heterotrimers

Citation preview

nature Tbfce JriemadunaljLiurnal of science / 8 February 202+

https:/ www.foxitsoftware.cn/editmac/?MD=Mshanchu

The international journal of science / 8 February 2024

Embrace open science — before it’s too late A UNESCO report laments the lack of progress towards making science more open. More awareness could aid efforts to achieve the UN’s Sustainable Development Goals.

T

he ‘open science’ concept is gaining more followers, not least through the efforts of the cultural organization UNESCO. Over the past several years, the organization has been consulting on how science can become more collaborative, transparent, accessible, equitable and inclusive, which are all attributes of open science. And in 2021, it published a framework for what a genuinely open science could look like. At the end of last year, UNESCO, which is headquartered in Paris, published a report on the current status of this endeavour. The report makes it clear that, although there are instances of good practice, there is still much work to do to fulfil the potential of open science globally. In 2021, UNESCO’s member states agreed on a definition of open science that includes open access to scientific knowledge (including the humanities and social sciences); open access to research infrastructure; open collaboration between scientists and ‘societal actors’ (essentially, all those who are not scientists); and open dialogue between different knowledge systems, including between scientific knowledge and Indigenous knowledge. Member states also pledged to incorporate the concept into their research systems, including using open-science principles in publicly funded research; supporting nonprofit and community-driven publishing; encouraging the publication of research in more languages; and incentivizing the private sector to join discussions about achieving open-science goals. UNESCO’s report describes several examples of positive initiatives, such as in research collaboration, open-access scientific publishing and public engagement in science. For example, in 2020, the Brazilian government launched the National Platform of Research Infrastructure, a digital platform in which scientific institutions can register their available infrastructure, and make it available to researchers outside their organization. This is an excellent way to spread access to expensive equipment across the research community. Meanwhile, South African policymakers are consulting researchers to help to create a national open-science policy for the whole country. The aim here is to build more transparency, scrutiny and reproducibility into the country’s research system. The policy will also include

Threequarters of publications in openaccess repositories are in just six languages.”

measures to monitor progress. The European Commission, based in Brussels, was an early proponent of open science. Between 2002 and 2020, it increased its funding for ‘societal engagement’ projects from €88 million to €462 million — an amount that is now equivalent to US$500 million. Moreover, a decade ago, all scientific publications arising from the European Union’s €80 billion Horizon 2020 programme needed to be published open-access. Citizen science is another growing area in open science with much promise, UNESCO notes. By 2018, half of all records in the Global Biodiversity Information Facility — an international open-access data repository based in Copenhagen — were from citizen scientists, up from around 10% in around 2007. Other indicators are less rosy, however. Around three-quarters (73%) of publications in open-access repositories are in just six languages — with nearly half (46%) being in English alone. And in spite of some of the progress mentioned, overall the report finds that scientific institutions, such as universities, national science academies and journals, are struggling to include communities, in all their diversity, in the process of creating scientific knowledge itself. Open science aligns with UNESCO’s founding mission for science and education to benefit all of humanity; and with the idea that access to science is a human right. But the organization’s interest in open science goes beyond these broad founding principles. The UN’s Sustainable Development Goals (SDGs), adopted in 2015, are humanity’s best attempt to map a pathway towards a better future — and a more open approach to science could have a larger part to play in achieving them. That effort needs as much help as it can get: only about 12% of the SDG targets are likely to be met by the 2030 deadline. Monitoring SDG indicators is one obvious way that citizen scientists can help. Some of the largest gaps in the collection of relevant SDG data are in lowand middle-income countries, which is where citizen research can really make a difference. In 2020, Dilek Fraisl, a data researcher at the International Institute for Applied Systems Analysis in Laxenburg, Austria, and her colleagues found that citizen-science projects were already helping to monitor at least five SDG indicators (D. Fraisl et al. Sustain. Sci. 15, 1735–1751; 2020). At the time, more than half of the data collected on indicators for sustainable cities, good health and well-being, and clean water and sanitation were provided by citizen scientists. There’s scope for citizen scientists to do more. UN agencies have also recognized the potential of connecting citizen scientists with official data bodies. The UN Statistical Commission and UN Women are working with researchers in civil society organizations to produce resources, such as toolkits for producers of citizen-generated data. The UNESCO report shines a much-needed light on some promising developments in open science. The challenge will be how to accumulate individual examples of good practice into something similar to a critical mass, so that, in cases such as monitoring the SDGs, they can be harnessed to get the world to where it needs to be. Nature | Vol 626 | 8 February 2024 | 233

Editorials

Research institutions must share knowledge of cyberattacks Ransomware attacks that have debilitated the British Library in London and Berlin’s natural history museum shows how vulnerable research institutions are to this kind of crime.

I

t has been more than three months since the British Library’s staff and users awoke to the news that its computer systems had been hijacked. After the attack on 28 October, anything that used the Internet — the library’s phone systems, its digital collections and website — became inaccessible. A hacking group called Rhysida had demanded a ransom, which the London-based library refused to pay. In November, Rhysida listed around half a million confidential files, including names and e-mail addresses of the library’s staff and users, for auction on the dark web, with bids starting at 20 bitcoins (US$800,000). Berlin’s natural history museum was also attacked in mid-October. In-person visits are continuing, but research is possible only “to a limited extent”. These attacks are not isolated cases. In one study, researchers analysed 58 cyberattacks between 1988 and 2022 on universities, schools and other organizations worldwide, and found that the frequency of attacks had increased since 2015 (H. Singh Lallie et al. Preprint at https://arxiv.org/abs/2307.07755; 2023). Information on the attacks was gleaned from publicly available online sources, such as media reports and the institutions’ own websites. The scientists concluded that research and education data are “a prime target for cyber criminals”. The study suggests that ransomware attacks — which permanently block access to data or systems until money is paid — were the most common form of cyberattack from an external source. Within an institution, students hacking the system to alter their grades were most often the cause. The vulnerability of educational and research institutions is not difficult to predict. All around the world, millions of members of staff, students and alumni log into institutional computer systems daily. Moreover, since the COVID-19 pandemic, remote access from personal devices with varying levels of protection has increased massively. Some of the biggest security risks come from the use of weak passwords and computer systems that can be accessed without multi-factor authentication — in which users verify their identity through two or more independent pieces of evidence. According to an annual survey by US technology giant IBM on data breaches, only four in ten organizations, including those in research and education, require users of computer systems 234 | Nature | Vol 626 | 8 February 2024

Collaboration between researchers who study computer security and those who investigate crime will bring wider benefits.”

to verify their identities regularly with such authentication methods (see bit.ly/4bfzamz). Research institutions are generally not short of information technology expertise — the British Library, for example, houses the UK national research centre for artificial intelligence and data science, the Alan Turing Institute. Yet there is a lack of in-depth, publicly available research on the extent and range of cyberattacks against educational institutions. Not all those that are attacked go public with details — the British Library did not reveal the attack was an instance of ransomware until 29 November. In many countries, organizations are required to report attacks to the relevant authorities, but governments, for understandable reasons, often do not publish this information. Some in national security circles consider such research, and the public scrutiny associated with it, a risk for producing or increasing vulnerabilities. However, collaboration between researchers who study computer security and those who investigate crime will bring wider benefits. It could help institutions to protect themselves against future attacks, and enable organizations to handle an attack effectively and minimize damage. Sharing knowledge on how to react to a ransom demand is one example. Institutions that are subject to ransomware attacks are advised not to pay, although some have done so. Everyone would benefit if these experiences were studied, peer reviewed and published in the open literature. Another important question is who should pay to recover and strengthen computer systems that are protecting national assets. In the case of the British Library, three months after the attack, some collections are available for people who visit in person, but it could be months more before its online records of books, journals, PhD theses and rare manuscripts are fully accessible to the library’s users all over the world. The organization also needs to find in the region of £6 million ($7.5 million) to £7 million from its own resources to repair the damage. So far, the UK government has not said whether it will underwrite the costs — a position that has left other librarians perplexed. The British Library is the United Kingdom’s national library. It is important to the nation’s businesses, colleges, research centres, schools and universities, and even more so to all those who do independent research. Library users are experiencing continued delays in a range of lending services, from ordering copies of books published over a span of more than three centuries, to accessing journal articles. The institution has one of the world’s largest collections of maps, along with archives of sound recordings and every UK PhD thesis published over the past century. By not contributing to the repairs, the government is disadvantaging researchers who cannot access other institutional libraries. This is not just a matter for the UK government, but for national and regional governments worldwide. Relevant authorities need to step up to support important institutions in times of crisis. And funders and researchers should consider how they can help — for example, by studying how to minimize the risk of cyberattacks happening in the future and what to do when they do take place.

A personal take on science and society

World view By Jean King

The mental cost of stress for women in science is too high Under-represented groups face chronic barriers, creating psychological ­— and physical ­— effects. The scientific community must ease this burden.

MATTHEW BURGOS

A

fter growing up on the tiny Caribbean islands of Aruba and Grenada, I entered my first collegiate chemistry laboratory in the late 1970s, at St. Francis College in Brooklyn, New York. I was so excited that I felt like I was floating. But there was no one like me in the class. The professor was a white man — ­ as was everyone else in the class, except for one white woman at the back of the room. When I went up to her and asked whether all the other women were in another section, she looked perplexed. We became fast friends, and from that day forward, we took every class together, to ensure that neither of us would be the only woman in the room. Decades later, women still comprise only 29% of the science, technology, engineering and mathematics (STEM) workforce — compared with 49% of non-STEM workers — in the 146 nations evaluated in the World Economic Forum’s 2023 Global Gender Gap Report (see go.nature. com/47xxa). The numbers are even more dire for people of colour: the 2021 US STEM workforce was only 15% Latinx and 9% Black. Why are there still alarmingly few women — particularly women of colour, like me — in STEM? I think that’s the wrong question. The real question is: what is the cost for women, particularly those of colour, to survive and thrive in STEM — and what can we do to reduce it? Factors behind the gender gap in STEM include not only lack of promotion, unequal pay and a dearth of meaningful work, but also stress, burnout and insufficient diversity. Role models who are women and people of colour are often hard to come by. I did not have a single Black female neuroscience instructor in my entire time at university. That is still the experience of many minoritized students attending predominantly white institutions in the United States. It’s a circular problem. Allostatic load is a term coined by physiological psychologist Eliot Stellar and neuroendocrinologist Bruce McEwen, one of my mentors. It describes how chronic adverse physical, psychological or social situations ­— including racial and gender-based oppression — ­ cause sustained activation of the body and brain’s stress response, resulting in cumulative wear and tear. Increased allostatic load is associated with depression and anxiety, which can impair motivation. Public-health researcher Arline Geronimus at the University of Michigan in Ann Arbor describes how this can affect minoritized groups in her 2023 book Weathering. Geronimus found that allostatic-load scores indicated that

Women and people of colour should not have to endure extra physical and neurological stress to apply their scientific talents.”

Jean King is the dean of arts and sciences and a professor of neuroscience at Worcester Polytechnic Institute in Massachusetts. e-mail: jaking@wpi. edu

Black women in the United States age faster than do white women. Socio-economic status could not explain the disparities — Black women from all socio-economic classes were likely to have high allostatic loads (A. T. Geronimus et al. Am. J. Public Health 96, 826–833; 2006). In my own experience, there is a cost for success in STEM academia. The few minoritized people who seem to thrive might be doing what psychologists call ‘high-effort coping’: working harder than others to succeed, because of anticipated or experienced hostility. We are still exposed to discrimination and an elevated risk of depression, increased allostatic load and accelerated ageing. As a society, how do we lift these burdens and make real, lasting progress? On an individual level, what is helpful is finding your people. In the words of US author and civil-rights activist Audre Lorde: “We must allow each other our differences at the same time as we recognize our sameness.” I encourage each person to reflect on their identity markers and find a group that works for them. Location matters. If you can choose where you are educated or work, take into account factors such as the prevalence of people of colour, the number of women and people of colour in leadership roles and how open the people around you are to having difficult conversations. The need to build communities might seem to add to cognitive load, but I think the opposite is true: seeing other minoritized people doing what we do gives us great strength. Alongside neuroscientist Emmeline Edwards at the US National Institutes of Health, I am a co-chair of World Women in Neuroscience, an independent consulting, mentoring and networking organization that promotes the careers and amplifies the expertise of female neuroscientists across the globe, especially in under-resourced regions. In more than a decade of experience, we have found that it is crucial to listen to women in a particular region to define their community’s needs. Female neuroscientists in Uganda and Brazil, for example, tell us that the lack of resources and mentorship are priorities. Women in Japan often say they need more networking opportunities and leadership roles. Allies, institutions and governments must do all it takes to decrease the cost for women and others in under-represented groups in STEM, both early on and throughout their careers. Immediate actions must be to champion pay equity; increase representation, including in leadership; and foster inclusive environments that support everyone. Countless women and people of colour have a passion for science. They should not have to endure extra physical and neurological stress to apply their scientific talents to the world’s most pressing problems. Nature | Vol 626 | 8 February 2024 | 235

Selections from the scientific literature

L TO R: CNR, CA’ FOSCARI UNIV./RICCARDO SELVATICO; METROPOLITAN MUSEUM OF ART, NEW YORK; SPL

Research highlights SOLVED: MYSTERY OF FEYNMAN’S SPRINKLER

ON THE SPECIES OF ORIGIN: IVORY’S PEPTIDES REVEAL ALL

An experiment has settled a controversy about how a type of sprinkler would spin if it sucked water in — a riddle made famous by the US physicist Richard Feynman. In normal operation, an S-shaped lawn sprinkler rotates because the water shooting from its nozzles in one direction pushes the device to spin in the opposite direction. But if the sprinkler is underwater and sucking in water, the nozzles do not simply act as ‘reverse jets’, because the water flows in from all possible directions. These complexities, together with confounding effects such as turbulence, mean that past experiments on such systems have given inconclusive or contradictory results. Kaizhe Wang at New York University in New York City and his collaborators carefully designed a sprinkler to remove confounding effects. They found that the reverse sprinkler rotates forwards instead of backwards, but unsteadily, and at only about one-fortieth of its speed under normal operation. Detailed observations backed up by mathematical modelling suggest that a weak jet effect inside the device dominates the sprinkler’s motion.

Ivory has long been used to make objects such as jewellery. A minimally invasive technique now offers a way to identify specific peptides in ancient ivory artefacts, revealing the animals that they came from. Knowing the exact origin of the bone or tooth used to make an ivory artefact has implications for tracking the trade in these materials, and for understanding past cultural norms. Catherine Gilbert at the University of Bordeaux in France and her colleagues sampled and analysed 16 objects aged between 800 and 6,000 years old at the Metropolitan Museum of Art in New York City. Rubbing small areas with a diamond microgrit film yielded samples from which the authors extracted peptides in collagens — proteins that are abundant in bone. Analysis identified the collagens in each sample, allowing the authors to pinpoint the source animal’s species, genus or family. The team showed that a thirteenth-century chess piece from Scandinavia was made from the bone of a sperm whale (Physeter macrocephalus), and an ancient Egyptian hippo figurine (pictured) was carved from a hippo tooth (Hippopotamus amphibius).

Phys. Rev. Lett. 132, 044003 (2024)

GLACIER’S MEMORY FADING AS EARTH’S CLIMATE WARMS Climate change is erasing the environmental history preserved in a Swiss mountain glacier. Ice sheets and glaciers are natural environmental archives: they record the pollutants that are captured each year in snow, which is eventually compressed into layers of glacier ice. Margit Schwikowski at the Paul Scherrer Institute in Villigen, Switzerland, and her colleagues wanted to assess how global warming is affecting these ice records. In 2018 and 2020, they drilled cores from the Corbassière Glacier (pictured) in the Swiss Alps, at an altitude of around 4,100 metres. Even at this great height, the glacier is melting and losing its ‘ice memory’. Throughout its length, the 2018 core contained ammonium, nitrate and sulfate ions — signatures of pollution lacing the snow that had fallen year after year. But the 2020 core had much less of those signals in its deepest layers. That’s presumably because warm air temperatures had melted the top of the glacier, and the melt water had percolated deep into the glacier and literally washed away the ions. The record was lost in just two years, the authors say. Nature Geosci. https://doi.org/ mfk7 (2024)

Sci Adv. 4, eadi9028 (2024)

BLOOD AND GUTS: THE INTESTINAL CELLS BOOSTING IMMUNITY A protein made by gut cells helps to destroy harmful bacteria in the intestines, according to experiments in mice. Roughly 50 proteins in the blood make up the complement system, which forms a crucial part of the immune response. These proteins, most of which are manufactured in the liver, help immune cells (pictured, artificially coloured) to track down invading microorganisms. Meng Wu at Harvard Medical School in Boston, Massachusetts, and her colleagues have found, for the first time, that certain gut cells in mice, called stromal cells, produce a few complement proteins. When the team infected mice with the diseasecausing bacterium Citrobacter rodentium, the animals’ gut levels of a crucial protein called C3 increased. It’s known that C3 attaches to the surface of bacteria, helping immune cells to gobble up and destroy the invaders. Twenty days after being infected with C. rodentium, only half of mice that had been bred to lack C3 survived. By contrast, all animals with normal C3 levels survived. Twelve days after infection, mice lacking C3 had lost, on average, more than 20% of their body weight, whereas those with normal C3 levels lost no weight. Cell https://doi.org/mfmc (2024)

Nature | Vol 626 | 8 February 2024 | 237

The world this week

DAVID BECKER/GETTY

News in focus

Donald Trump won the first two Republican presidential primaries, opening the path to the US presidency once again.

TRUMP’S PRESIDENTIAL PUSH RENEWS FEARS FOR US SCIENCE If he wins a second term, the former US president has promised to limit the authority of federal agencies and employees, including scientists. By Jeff Tollefson

D

onald Trump’s promise to “dismantle the deep state” moved one step closer to reality last month as he cruised to victory in New Hampshire’s Republican presidential primary. Faced with the possibility of Trump winning the US presidency for a second time, science advocates are gearing up to fight what they see as an existential threat to the future of science in the US government. If he wins, Trump, who now dominates the

Republican party with his far-right following, has promised to revive a plan to reclassify tens of thousands of federal employees. These include scientists and others who are currently shielded from politics in permanent professional positions. This plan, known as Schedule F, would allow his administration to more easily fire “rogue bureaucrats” — those who he says oppose his political agenda and are part of the ‘deep state’. The administration could then appoint replacements, regardless of their scientific or technical expertise, who are aligned with Trump politically.

“It’s sort of the ultimate attack on government,” says Betsy Southerland, a former environ­mental scientist at the US Environmental Protection Agency. She says many of the former Trump administration’s proposals that put politics ahead of science were blocked by the objections of professional civil servants. If Schedule F gets the green light, Southerland adds, “you would have nobody to report scientific-integrity violations, because anybody who objected would be fired”. One of Joe Biden’s first acts as US president was to call for stronger scientific-integrity Nature | Vol 626 | 8 February 2024 | 239

News in focus rules in the government that could help to thwart future efforts to politicize science. As the 2024 presidential election draws near, US agencies are rushing to complete their own rules, including policies for how to report and investigate violations. If Trump wins, however, many observers say that there’s nothing to stop him reversing those policies and following through on his agenda. “My message to the self-appointed global elites: Your time is up,” wrote Kevin Roberts in a 18 January post on the social-media platform X. Roberts is president of the Heritage Foundation, a conservative think tank in Washington DC that is helping to prepare for a potential Trump transition into the White House next year.

Transferring power Conservatives have long been critical of what they see as regulatory overreach by permanent, unelected government employees, and they have already sought to limit the power of regulatory agencies in the courts. During his tenure as president from 2017 to 2020, Trump appointed a trio of conservative judges to the Supreme Court. Now ruled by a 6–3 conservative majority, the court has stunned scientists with rulings overturning abortion rights and restricting environmental regulations. Legal specialists say this trend could intensify under a new Trump administration. On 17 January, the Supreme Court heard arguments in a pair of cases that could end the use of what is known as the Chevron doctrine, which allows judges nationwide to defer to the expertise of US agencies when implementing ambiguous federal laws, as long as their interpretations are reasonable. Both cases involve seemingly minor disputes about the fees that fishing-boat operators pay to the National Marine Fisheries Service to monitor for overfishing. But the companies have asked the court to overturn the underlying Chevron doctrine, which was established in 1984, arguing that it undermines the power of the courts. Supporters of the Chevron doctrine say that it gives agencies the flexibility to address new research and challenges — such as innovative technologies, pandemics and climate change — that lawmakers in Congress could not have foreseen when they wrote regulatory statutes, often decades ago. Giving agencies this authority is necessary, because it is not realistic to expect that Congress will step forward with laws that answer every regulatory question that arises in the courts, says Allison Larsen, a legal scholar at the William & Mary Law School in Williamsburg, Virginia. Trump’s plan to reclassify federal employees would limit agencies further by transferring power away from experts, such as scientists, who help to craft policies on issues such as pandemic preparedness and biosecurity, and pass it to political appointees, who serve the president. In effect, this would expand 240 | Nature | Vol 626 | 8 February 2024

a patronage system that encourages a winner-takes-all approach to government, says Max Stier, president of the Partnership for Public Service, an advocacy group based in ­Washington DC.

Temporary obstacles Political meddling with science can come from both sides of the political aisle, but the Trump administration came under fire for repeatedly ignoring scientific evidence and sidelining government scientists, especially during the COVID-19 pandemic. In response, the Biden administration has sought to update and expand scientific-integrity rules for all ­federal agencies. Science advocates say that early drafts of these rules, although imperfect, are a significant improvement. The Biden administration also plans to create an expert panel under the National Science and Technology Council that would have the authority to review agency policies and investigate alleged violations by appointees throughout the government. “They are taking steps forward and have already done a lot of really good things,” says Anita Desikan, an analyst at the Center for Science and Democracy at the Union of Concerned Scientists, an advocacy group in Cambridge, Massachusetts. “But if you do have bad-faith actors who are looking to harass scientists, it’s probably still not enough.”

The Biden administration has also sought to head off the threat to federal workers. In September, the US Office of Personnel Management proposed a rule that seeks to reinforce long-standing protections for some 2.2 million federal employees. It would clarify when and under what conditions these civil servants can be moved out of a merit-based system and into one that makes it easier for administrators to terminate their employment. The rule might slow down Trump’s plan if he wins a second term, but it’s probably a temporary obstacle, observers say. This is in part because agency policies are not laws, and they can be reversed easily by a new administration. One thing that would help, Stier says, is a statute that would make it illegal for future presidents to strip federal workers of their protections. Some Democratic lawmakers have also proposed legislation that would seek to enshrine scientific-integrity requirements in federal law. But given the current polarization in the US Congress, which would have to pass those laws, both efforts remain a long shot. For Blake Emerson, an administrative-law researcher at the University of California, Los Angeles, the cumulative threat from the courts and a future Trump administration is almost overwhelming. “If all of these efforts are successful,” he says, “you won’t have a real place for independent professional and scientific judgement in the government.”

BLACK-HOLE OBSERVATIONS SOLVE COSMIC-RAY MYSTERY Data show that jets from a collapsed star are capable of producing some of the Galaxy’s fastest particles. By Davide Castelvecchi

A

n array of telescopes in Namibia has pinpointed the origin of some of the most energetic particles the Galaxy can produce. The observations point to a place where particles of matter spewed by a black hole in a region known as the Manatee Nebula are accelerated to nearlight-speed (H.E.S.S. Collaboration. Science 383, 402–406; 2024). The findings, published on 25 January in Science by researchers at the High Energy Stereoscopic System (HESS), are a step forward in the century-old quest to understand the origins of cosmic rays — fast-moving atomic nuclei and other particles that are continually hitting Earth’s upper atmosphere.

“For people like me who want to model astrophysical jets, including their internal composition, propagation and evolution”, the information produced by HESS is “incredible”, says Sera Markoff, a theoretical astrophysicist at the University of Amsterdam.

Rain from space Cosmic rays can have a wide range of energies. The most abundant, lowest-energy cosmic rays consist of particles of solar wind that rain down on Earth’s atmosphere after spiralling in the planet’s magnetic field. Cosmic rays of much higher energies are thought to be produced by supernovae, the explosive deaths of massive stars. And yet-more-energetic cosmic rays originate outside the Galaxy, in particular from quasars — supermassive black holes that produce

B. SAXTON, (NRAO/AUI/NSF) FROM DATA PROVIDED BY M. GOSS ET AL.

The Manatee Nebula formed when a giant star exploded more than 10,000 years ago.

jets of plasma travelling at near-light-speed. These jets can have energies up to eight orders of magnitude higher than those produced in particle accelerators. Astrophysicists have proposed that plasma jets from black holes that are smaller than quasars — but still several times as massive as the Sun — could also contribute to the cosmic-ray population. The energies produced by these ‘microquasars’, which are also bright sources of X-rays and radio waves, could reach a range intermediate between those from supernovae and those from quasars. In the latest study, astrophysicist Laura Olivera-Nieto at the Max Planck Institute for Nuclear Physics in Heidelberg, Germany, and her collaborators studied a microquasar called SS 433. The black hole lies in the Aquila Constellation at around 5.5 kiloparsecs (18,000 light years) from the Solar System, and forms a binary system together with a large star. Matter ejected from the star swirls around the black hole, then spirals into it, generating highly energetic jets. The binary system is surrounded by a nebula nicknamed the Manatee owing to its elongated shape. The nebula is a shell of dust and gas left over from a supernova that occurred between 10,000 and 100,000 years ago, during which the core of an exploding star collapsed to form the black hole. The outflow of matter from the supernova would itself have produced cosmic rays for thousands of years after the event, activity that has long since quietened down. But some time between 10,000 and 30,000 years ago, the system lit up again, when the black hole formed its jets. The researchers think that this is when it started producing cosmic rays again.

Cosmic clues Any cosmic-ray particles originating from a microquasar would move across the Galaxy in spirals before reaching Earth, their trajectories

bent by magnetic fields. This makes it impossible to trace their paths back to a specific source. Instead, astrophysicists searching for the possible origins of cosmic rays look for γ-ray photons, which should be produced in the same processes that accelerate cosmic-ray particles, but travel to Earth in straight lines. Astronomers first observed γ-rays from SS 433 in 2018 using the High Altitude Water Cherenkov (HAWC) observatory in Mexico’s Pico de Orizaba National Park (A. U. Abeysekara et al. Nature 562, 82–85; 2018). But, unlike the team at HESS, they were unable to locate the exact source with precision. Both HAWC and HESS detect γ-ray photons

indirectly, but they use different approaches. When a γ-ray collides with an atomic nucleus in the upper atmosphere, it produces a shower of secondary particles, including electrons and their heavier siblings, muons. HAWC has waterfilled tanks that pick up these particles as they reach the ground, whereas HESS works by imaging flashes of light that the particles produce as they move down through the atmosphere. The five dishes of HESS can be pointed in a specific direction in the sky. This allowed HESS to precisely locate where in the Manatee Nebula the γ-rays were produced, and to focus on distinguishing those with particular energies. More than 200 hours of observations, made over 3 years, show that the γ-ray emission starts around halfway between the black hole and the supernova remnant, and slowly peters out. “The highest-energy photons only come from closer to the black hole,” says Olivera-Nieto. “This was really the crucial discovery.” This suggests that the γ-rays — and, by implication, cosmic rays — are produced by mechanisms internal to the jets, rather than by collisions with other matter, Olivera-Nieto explains. The space surrounding the black hole is otherwise empty, swept clean by the supernova’s expanding shockwave. The finding “strengthens the case that X-ray binaries are smaller analogues to supermassive black holes, and equally capable of accelerating cosmic rays”, says Markoff, who also praises Olivera-Nieto’s data analysis. “Her technique allowed the use of more data, and amplified the sensitivity enough to do this fantastic study, and so sets the stage for much more work like this.”

‘TRANSMISSIBLE’ ALZHEIMER’S HINTS SEEN IN BRAINS FOR FIRST TIME The findings support a controversial hypothesis that proteins related to the disease can be transferred. By Carissa Wong

R

esearchers say they have uncovered more evidence to support a controversial hypothesis that sticky proteins that are a signature of Alzheimer’s disease can be transmitted from person to person through certain surgical procedures. The authors and other scientists stress that the research is based on a small number of people and is related to medical practices that are no longer used. The study does not suggest that forms of dementia such as Alzheimer’s

disease can be contagious. Still, “we’d like to take precautions going forward to reduce even those rare cases occurring”, says neurologist John Collinge at University College London, who led the research1, which was published in Nature Medicine on 29 January. For the past decade, Collinge and his team have studied people in the United Kingdom who, during childhood, received growth hormone derived from the pituitary glands of cadavers to treat medical conditions such as short stature. The latest study finds that, Nature | Vol 626 | 8 February 2024 | 241

decades later, some of these people developed signs of early-onset dementia. The dementia symptoms, such as memory and language problems, were diagnosed clinically, and in some patients appeared alongside plaques of the sticky protein amyloid-β in the brain, a hallmark of Alzheimer’s disease. The authors suggest that this protein, which was present in the hormone preparations, was ‘seeded’ in the brains and caused the damage.

Contaminated hormone The work builds on the team’s previous studies of people who received cadaver-derived growth hormone, a practice that the United Kingdom stopped in 1985. In 2015, Collinge’s team described2 the discovery at post-mortem of amyloid-β deposits in the brains of four people who had been treated with the growth hormone. These people had died in middle age of the neurological condition Creutzfeldt–Jakob disease, which is caused by infectious, misfolded proteins called prions. The prions were present in batches of the growth hormone. The four people analysed in that study died before clinical signs related to the amyloid-β build-up might have been observed. But the presence of these amyloid plaques in blood vessels in their brains suggested that they would have developed a condition called cerebral amyloid angiopathy (CAA), which causes bleeding in the brain and is often a precursor to Alzheimer’s disease. Collinge’s team also located and studied archived batches of the cadaver-derived growth hormone. In a 2018 study 3, they reported that certain batches of the hormone preparation contained amyloid-β proteins, and that when such preparations were injected into mice, this led to the development of amyloid plaques and caused CAA in the animals. This led the team to wonder whether the contaminated hormone preparations might also have resulted in people who received it developing Alzheimer’s disease, in which amyloid plaques are thought to cause the loss of neurons and brain tissue.

Signs of dementia In the latest study, the researchers found that five out of eight people who had received the hormone treatment in childhood — but did not develop Creutzfeldt–Jakob disease — developed behavioural signs of early-onset dementia later in life, between the ages of 38 and 55. Collinge’s team argues that these five people — whom the researchers studied in the clinic or through medical records and brain scans — met the diagnostic criteria for early-onset Alzheimer’s disease. Early-onset Alzheimer’s is usually caused by certain genetic variants, but the researchers did not find these variants in three of the people who showed signs of Alzheimer’s and 242 | Nature | Vol 626 | 8 February 2024

A brain affected by Alzheimer’s disease.

whose DNA samples were available for testing. “This is consistent with these patients having developed a form of Alzheimer’s disease resulting from childhood treatment with this contaminated pituitary hormone,” says Collinge. Taken together, the studies suggest that, in rare cases, Alzheimer’s disease could be transmitted through the transfer of biological material, the authors argue. However, the study’s small size limits the strength of the findings, says neuroscientist Tara Spires-Jones at the UK Dementia Research Institute at the University of Edinburgh.

“Are the amyloid-β seeds from the hormone treatment playing a role in the development of dementia? It’s hard to know with just eight people,” she says. The possibility that some of the people might have developed dementia regardless of the hormone treatment cannot be excluded, says neuroscientist Mathias Jucker at the German Center for Neurodegenerative Diseases in Tübingen. “These people had many different medical conditions which could have increased the risk of developing a neurodegenerative disease like Alzheimer’s disease,” he says. Researchers including Spires-Jones also question whether the people with dementia had Alzheimer’s, despite the clinical diagnoses. “There are often errors in diagnosing the type of dementia someone has while they’re alive,” agrees neuroscience researcher Andrew Doig at the University of Manchester, UK. From a public-health perspective, there is no need to be concerned about ‘transmissible’ dementia today, says Spires-Jones. “This treatment doesn’t exist any more.” Despite the study’s limitations, the research furthers our understanding of neurodegenerative diseases, scientists say. “I’m glad that people are doing amazing research to help us better understand seeding of neurodegenerative disease by amyloid-β,” says Spires-Jones. 1. Banerjee, G. et al. Nature Med. https://doi.org/10.1038/ s41591-023-02729-2 (2024). 2. Jaunmuktane, Z. et al. Nature 525, 247–250 (2015). 3. Purro, S. A. et al. Nature 564, 415–419 (2018).

LEADING US PARTICLEPHYSICS LAB FACES UNCERTAIN FUTURE Several organizations are vying for the contract to manage Fermilab, after it received failing grades. By Dan Garisto

T

he future of US particle physics is in limbo as its central institution, Fermi National Accelerator Laboratory (Fermilab), continues to receive mixed grades and its management is put up for grabs. Conditions at the lab were called into question in 2021, when Fermilab failed an annual assessment by its overseer, the US Department of Energy (DoE). It received a B grade overall — a B+ was required to pass — and earned a C for its handling of the troubled multibillion-­dollar Deep Underground Neutrino Experiment (DUNE), the nation’s flagship particle-physics

project. Last year, the DoE took the unusual step of opening the bidding on the contract to run Fermilab, which since 2007 has been operated by the Fermi Research Alliance (FRA) — a partnership between the University of Chicago in Illinois and the Universities Research Association, a consortium of 90 universities, most of which are US-based. A University of Chicago spokesperson emphasized the university’s “deep and long-standing commitment” to Fermilab, and declined to comment further, but Nature has confirmed that the FRA is —reapplying to manage the lab. The FRA is not the only contender. Associated Universities, Inc. (AUI), which runs the Green Bank Observatory in West Virginia,

ZEPHYR/SPL

News in focus

LYNN JOHNSON, FERMILAB

has confirmed that it is also competing for the contract. “AUI excels in making scientific and technological breakthroughs possible by delivering on large science projects on time and on budget,” AUI chief executive Adam Cohen told Nature. “We are very interested in bringing our experience to the challenges we understand exist at Fermi.” In addition to AUI, a virtual meeting on 11 January about proposals for the management contract was attended by Paragon Systems, a private security firm based in Herndon, Virginia, which declined to comment. Proposals must be submitted by 4 March. A DoE board will make a selection by the end of the year, and the new contract will begin next year. Since the release of the critical report card in 2021, Fermilab has acquired new leadership. In April 2022, the FRA selected accelerator physicist Lia Merminga to be the director of the lab. Under her watch, the 2022 and 2023 assessments showed improvement, although some problems remain: last year’s report card gave Fermilab high marks for many of its scientific objectives, but it scored a failing B– on environment, health and safety. “Overall, the lab’s performance has been steadily improving,” Merminga told Nature. “We continue to build a strong workplace culture focused on safety.”

Key infrastructure Located about 65 kilometres west of Chicago, Fermilab was founded in 1967 and has many advances in fundamental science to its name. These include the discoveries of the bottom and top quarks, which are a pair of heavy subatomic particles essential to the current picture of particle physics, the standard model. Over the past decade, the lab has focused on neutrinos — mysterious, weakly interacting

Excavation at the South Dakota site of the DUNE neutrino project is nearing completion.

elementary particles that can pass through matter mostly unimpeded — with DUNE as its flagship programme. Starting in 2031, scientists at Fermilab will shoot a beam of neutrinos through the Earth to a former mine in South Dakota that is 1,300 kilometres away. At that site, some of the particles will hit giant vats of liquid argon, creating detectable flashes of light and electricity. DUNE’s primary goal is to discover whether the neutrino behaves differently from its antimatter mirror image, the antineutrino; if there is a difference, it could have far-reaching consequences for the evolution of the Universe. The project has faced serious setbacks in construction, however, and its price tag has

reached US$3.3 billion, roughly double its initial estimated cost. To ensure DUNE’s feasibility under a limited budget, researchers have had to make concessions, including reducing the number of argon detector modules from four to three. With the responsibility of DUNE on its shoulders, Fermilab is effectively the centre of US particle physics. Hundreds of scientists work there, and thousands of outside researchers use its facilities. Last December, an influential panel of physicists issued a road map — the Particle Physics Project Prioritization Panel (P5) report — describing the direction they would like their field to take. Fermilab drew special recognition in the report for its leadership in current and future experiments — including a potential collider that smashes together muons, the heavier cousins of electrons.

MICHELLE LITVIN/NEW YORK TIMES/REDUX/EYEVINE

Access issues

The Fermi National Accelerator Laboratory is located west of Chicago, Illinois.

唯一　

But along with its management woes, questions over access to Fermilab have also dogged the centre. In 2022, after restrictions related to the COVID-19 pandemic were rolled back at other national labs, Fermilab — which had long been open to the public because of its lack of classified research — kept limitations in place, owing to DoE regulations. Early last year, concerns reached a boiling point: a petition to reopen Fermilab gathered nearly 3,000 signatures and hundreds of testimonials. A graduate student who was living on site and had experienced a medical emergency wrote that her mother was not allowed into the lab to see her because of an expired passport. Even JoAnne Hewett, the director of Brookhaven National Laboratory (BNL) in Upton, New York, was held at the gate until Merminga called to ensure that she was allowed access. The controversy had scientists questioning Nature | Vol 626 | 8 February 2024 | 243

News in focus meeting that “more work needs to be done”, but added that Fermilab is “turning the corner in this matter, which is essential for world-leading science”. Wilson Hall, the iconic cathedral-like main building on the Fermilab campus, r­ eopened to the public on 23 January. Some of DUNE’s troubles might be resolving as well — the excavation of the South Dakota site is now nearly complete, and the installation of concrete floors there is under way. Although things have improved for the FRA, there’s no guarantee that it will win its bid for the Fermilab contract. National labs have changed hands before, although it is rare. In 1997, after worries about a radioactive tritium leak and other mishaps at the BNL, Federico Peña, who was then the US secretary of energy, pointed to “long-term mismanagement” at the lab, and terminated the contract with its operator, AUI.

FIRST AIRCRAFT TO FLY ON MARS DIES — BUT LEAVES A LEGACY OF SCIENCE The record-setting Mars helicopter Ingenuity broke during a final, fatal flight. By Alexandra Witze

N

ASA’s Ingenuity helicopter, the first aircraft to fly on another world, has died. It perished on 18 January ­during its 72nd flight in Jezero Crater on Mars. Ingenuity was nearly three years old

(counting just its time on the red planet). The helicopter, a box-shaped drone with a pair of 1.2-metre-long carbon-fibre blades, was a trailblazer for interplanetary spacecraft. NASA’s Jet Propulsion Laboratory ( JPL) in Pasadena, California, built it as a test to see whether powered flight was possible in the

The Mars helicopter Ingenuity propelled itself with two blades, each 1.2 metres in length.

244 | Nature | Vol 626 | 8 February 2024

唯一　

thin atmospheres of other worlds. Ingenuity accompanied NASA’s Perseverance rover to Mars, where both landed in February 2021 and began studying Jezero. “Ingenuity absolutely shattered our paradigm of exploration by introducing this new dimension of aerial mobility,” says Lori Glaze, head of NASA’s planetary sciences division in Washington DC. Ingenuity was supposed to make only five flights and last about a month, but ultimately it traversed 17 kilometres of the red planet and flew for a total of nearly 129 minutes between 2021 and 2024. During its final journey, something fatal happened — perhaps the rotor blades striking the ground — NASA announced on 25 January. An image that the helicopter took of the ground after the flight ended shows the shadow of one of the blades, with at least one-quarter of it missing. The helicopter can still communicate with Earth, at least for now, but it will not fly again.

Sidekick and explorer Future planetary missions might use the aerial lessons learnt from Ingenuity. “This type of mobility can take us to places we never dreamed we’d be able to explore,” says Laurie Leshin, director of JPL. NASA is already building an eight-rotor helicopter to explore ­Saturn’s moon Titan, a mission that will launch as early as 2028. And engineers at JPL have been working on advanced helicopter designs for some future Mars mission that could carry large payloads and explore places, such as cliffs and canyons, that other spacecraft ­cannot reach. Ingenuity enabled science while on Mars, in part by allowing researchers to study how its blades kicked up dust clouds into the red ­planet’s thin atmosphere. It also served as something of a sidekick to Perseverance, often flying above the rover’s planned path to scout the landscape for potential obstacles. Sometimes, however, the rover drove much faster than the helicopter could fly, and ­Perseverance had to wait for Ingenuity to catch up. Ingenuity died on top of an ancient river delta in Jezero, several hundred metres northwest of the rover’s current location. The goal for both spacecraft was to explore billions of years of history in the crater, in particular whether it was ever home to Martian life. Perseverance has travelled more than 24 kilometres while collecting cores of rock and dirt from geological settings in the crater. It has gathered 23 cores so far; 10 are sitting on the surface in a storage depot, awaiting a future mission to pick them up. The rest are still on board the rover. NASA and the European Space Agency hope to bring those cores back to Earth in the ­coming years for scientific study. But their plan is being revamped after an independent review estimated that the cost could be as high as US$11 billion.

NASA/JPL-CALTECH/ASU/MSSS

the lab’s fitness to serve as the US hub for particle physics. “Who would still consider organizing a scientific meeting at a lab where doing so means endless and senseless paperwork for the visitors?” wrote Joachim Kopp, a theorist at CERN, the European particle-physics lab outside Geneva, Switzerland, in a testimonial. At other labs, restrictions might be a mere inconvenience, but “for Fermilab, those changes are existential”, said Fernanda Psihas, a researcher at Fermilab and organizer of the petition. Last August, in response to the concerns about site access, the Fermilab site office, which manages building and construction projects, underwent an unprecedented review by site office managers at other national labs. Roger Snyder, the manager of Fermilab’s site office, did not respond to Nature’s enquiry about this peer review. Merminga said at an 11 December town-hall

that these gene-edited plants can bypass some of the heavy testing and other requirements imposed on genetically modified crops that contain foreign DNA. Nigeria and Malawi have similar policies, and other African countries, including Ethiopia and Uganda, are expected to follow suit, Runo says. Last year, Kenyan authorities gave Runo and his collaborators permission to grow the gene-edited seeds under those regulations, and he plans to launch field studies later this year. It is a significant step, Runo said at the conference, because Striga is not a problem in wealthier regions — meaning that large, multinational corporations have little incentive to develop solutions for it.

MARCPO/GETTY

Cattle that can beat the heat

Sorghum has been gene edited to resist a parasitic plant called witchweed.

CRISPR-EDITED CROPS BREAK NEW GROUND IN AFRICA Scientists in the global south use the popular technique to protect crops against local threats. By Heidi Ledford in San Diego, California

M

olecular biologist Steven Runo once thought that his team would make history as the first to plant gene-edited seeds in African soil. The competition turned out to be stiffer than he’d anticipated. A research group working on maize “beat us by two or three months”, says Runo, who works at Kenyatta University in Nairobi and whose gene-editing project focuses on sorghum. “But that’s good — African countries will see that this is actually possible.” The friendly rivalry is a sign of progress. Researchers have long hoped that the relative ease and low cost of CRISPR gene-editing systems will make it possible for scientists in low- and middle-income countries to produce crops with traits tailored to the needs of local farmers — who would no longer need to rely on seeds developed in foreign countries. Now scientists are overseeing at least a dozen efforts to develop such gene-edited crops. Among those projects is Runo’s effort to engineer sorghum to be resistant to Striga hermonthica, a troublesome species of a parasitic plant known as witchweed. Field trials

唯一　

of the new variety are scheduled for later this year, Runo said at the Plant and Animal Genome Conference in San Diego, California, on 16 January. “It’s not as easy as people make it out to be to do gene editing, but it is pretty accessible,” says Kevin Pixley, a research director at the International Maize and Wheat Improvement Center in Texcoco, Mexico. “Runo is a perfect example of that.”

Witchy weed Sorghum is a hardy crop that is used widely in Africa for feedstock and more. But more than 60% of African farmland is contaminated with species of Striga, which attaches itself to sorghum roots and siphons away nutrients and water. A witchweed infestation can wipe out an entire crop. Some wild varieties of sorghum are resistant to Striga because they carry mutations that alter the crop’s production of compounds that promote germination of Striga seeds. Runo and his collaborators have used CRISPR–Cas9 to mimic these mutations. Under Kenya’s 2022 regulations governing gene-edited crops, such plants are treated like conventionally bred crops because they do not contain DNA from another species. This means

Other gene-editing projects are under way to improve African agricultural products. Pixley and his collaborators, including researchers at the Kenya Agricultural and Livestock Research Organization in Nairobi, have developed ways to edit maize (corn) to make it resistant to maize lethal necrosis disease. They are also editing pearl millet to make its flour less prone to becoming rancid soon after milling, and groundnuts to make them more resistant to infection by the fungus that produces cancer-causing aflatoxins. African livestock are also being edited. At the Plant and Animal Genome Conference, Dan Carlson, chief scientific officer at Recombinetics in Eagan, Minnesota, described work to edit African breeds of cattle to improve their milk yields and tolerance to heat and disease. Although gene editing is relatively cheap to perform in the laboratory, there are still significant hurdles to bringing edited crops to the farm, says Klara Fischer, who studies rural development at the Swedish University of Agricultural Sciences in Uppsala. “Sometimes the discourse around this technology is overly enthusiastic,” she says. And because the market is unlikely to provide for poor small-scale farmers with limited purchasing power, government involvement would probably still be needed for the gene-edited products to benefit them.

Markets and money Runo has relied on funding from the US Agency for International Development and has collaborated with Corteva Agriscience, an agricultural company in Indianapolis, Indiana. Pixley and his team have received funds from the Bill & Melinda Gates Foundation in Seattle, Washington, and have also received technical assistance from Corteva. Runo is mindful that this support is not guaranteed. He and his team are working on cutting the cost of lab supplies and equipment and finding alternative funding sources. Also unknown, says Pixley, is how intellectual-property battles over CRISPR gene Nature | Vol 626 | 8 February 2024 | 245

News in focus more comfortable with crops developed by a local researcher than with seeds developed abroad. “This is not a multinational company. The people using the technology are people you have grown up with,” he says. “The narrative is very different.”

OBESITY DRUGS HAVE ANOTHER SUPERPOWER: TAMING INFLAMMATION Blockbuster weight-loss drugs also reduce inflammation in the brain and other organs. By Mariana Lenharo

T

he latest generation of anti-obesity drugs has taken the world by storm, thanks to their effectiveness at reducing weight. But these drugs also have a less well-known superpower: the ability to suppress inflammation. Evidence suggests that the drugs classified as GLP-1 receptor agonists can reduce inflammation in the liver, kidneys and heart. The drugs even seem to dial down inflammation in the brain, leading scientists to hope that the compounds could be used to treat Parkinson’s and Alzheimer’s diseases, both of which are characterized by brain inflammation. A 2022 review1 listed more than 20 clinical trials that are exploring the drugs as therapies for the two conditions. “The next generation of drugs could be even more targeted to reduce these new inflammation pathways that we’ve identified,” says Daniel Drucker, an endocrinologist at the University of Toronto in Canada, who co-authored a study2, published in December, investigating how the drugs dampen inflammation.

and the heart6. And GLP-1 reduces inflammation in fat tissue in obese mice7. “We know from animal studies and human studies that GLP-1 seems to reduce inflammation almost everywhere,” says Drucker. The reductions in body weight and blood sugar triggered by the drugs probably help to control inflammation. But some of the antiinflammatory effects start before meaningful weight loss is achieved, leading scientists to think a separate mechanism is at play.

Brain power Drucker and his colleagues noticed a potential clue: receptors for GLP-1 are scarce in immune cells in many tissues in which the hormone and its mimics reduce inflammation, but are abundant in the brain. To test the nervous system’s role, Drucker’s team began by inducing system-wide inflammation in mice. “Multiple GLP-1 drugs made those mice better and reduced inflammation,” Drucker says.

Body-wide effects The GLP-1 receptor agonists include semaglutide, which is marketed as Wegovy for obesity, and tirzepatide, marketed as Zepbound for obesity. The drugs mimic a gut hormone called glucagon-like peptide 1 (GLP-1), which controls blood sugar levels and dampens appetite. But a slew of findings showcase the ability of the hormone and its mimics to calm inflammation, which is caused by an onslaught of immune cells and immune-system chemicals. In one experiment, a GLP-1 receptor agonist called liraglutide alleviated liver inflammation in mice with a fatty liver3. A similar effect was observed in a pilot study in people4. In other experiments in mice, liraglutide demonstrated anti-inflammatory potential in the kidneys5 246 | Nature | Vol 626 | 8 February 2024

Immune cells (purple, artificially coloured) can contribute to chronic inflammation.

唯一　

But when the team blocked GLP-1 receptors in the animals’ brains, the GLP-1 drugs no longer reduced inflammation in multiple tissues2. The work demonstrates that, at least in mice, the drugs’ anti-inflammatory effects are achieved directly through GLP-1 receptors and are mediated by the brain, says Nigel Greig, a pharmacologist at the National Institutes of Health in Baltimore, Maryland. He notes that previous studies8 have established that only a small amount of these drugs can actually enter the brain. “It’s quite remarkable that the brain entry is so low, but it’s so important for anti-inflammatory action,” Greig says.

Targeting pathological proteins The GLP-1 drugs’ anti-inflammatory powers have promise for treating neurodegenerative diseases such as Parkinson’s and Alzheimer’s. Both are characterized by neuroinflammation. And in both disorders, pathological proteins — for example, amyloid-β in Alzheimer’s — interact with certain receptors in the brain to induce events that cause inflammation. Excessive inflammation can contribute to disease, Greig says. But GLP-1 receptor agonists seem to have the ability to knock back inflammation in the brain so that important processes, such as the birth of new neurons, can continue to occur, he notes. In one clinical trial, a GLP-1 receptor agonist called exenatide led to greater improvement in the motor abilities of people with Parkinson’s than did a placebo8. A trial is now assessing the medication in a larger population of people with Parkinson’s, and at least two clinical trials are testing semaglutide as a therapy for earlystage Alzheimer’s disease. The drugs’ anti-inflammatory action might also help to boost their effectiveness against diabetes and obesity, says Vinicius de Frias Carvalho, a biologist at the Inflammation Laboratory at the Oswaldo Cruz Institute in Rio de Janeiro, Brazil. Both conditions “are also inflammatory diseases”, he says. Semaglutide’s anti-inflammatory action might play a part in its ability to protect people with obesity against cardiovascular disease. The use of GLP-1 drugs to treat inflammation-related diseases could expand, Greig says, especially given the drugs’ lack of significant side effects. “There are so many systemic disorders where there’s an inflammatory component,” he says. It only makes sense, he says, to try the drugs against such disorders if there’s no effective treatment. 1. Kopp, K. O., Glotfelty, E. J., Li, Y. & Greig, N. H. Pharmacol. Res. 186, 106550 (2022). 2. Wong, C. K. et al. Cell Metab. 36, 130–143 (2024). 3. Somm, E. et al. Transl. Res. 227, 75–88 (2021). 4. Eguchi, Y. et al. Hepatol. Res. 45, 269–278 (2015). 5. Filippidou, F. M. et al. Am. J. Pathol. 190, 400–411 (2020). 6. McLean, B. A., Wong, C. K., Kabir, M. G. & Drucker, D. J. Mol. Metab. 66, 101641 (2022). 7. Lee, Y.-S. et al. Diabetologia 55, 2456–2468 (2012). 8. Athauda, D. et al. Lancet 390, 1664–1675 (2017).

STEVE GSCHMEISSNER/SPL

editing will ultimately affect efforts in Africa, and whether foreign markets — particularly in Europe — will be open to African-grown gene-edited crops. But as for local acceptance of the crops, Runo says the farmers he has spoken to feel

唯一　

Feature

Electric vehicles charge in a car park in the United Kingdom, which will ban the sale of petrol and diesel cars in 2035.

THE ELECTRIC-CAR BATTERY REVOLUTION

T

here’s a revolution brewing in batteries for electric cars. Japanese car maker Toyota said last year that it aims to release a car in 2027–28 that could travel 1,000 kilometres and recharge in just 10 minutes, using a battery type that swaps liquid components for solids. Chinese manufacturers have announced budget cars for 2024 featuring batteries based not on the lithium that powers today’s best electric vehicles (EVs), but on cheap sodium — one of the most abundant elements in Earth’s crust. And a US laboratory has surprised the world with a dream cell that runs in part on air1 and could 248 | Nature | Vol 626 | 8 February 2024

pack enough energy to power aeroplanes. These and other announcements rely on alternative designs to the conventional lithium-ion batteries that have dominated EVs for decades. Although lithium-ion is hard to beat, researchers think that a range of options will soon fill different niches of the market: some very cheap, others providing much more power. “We’re going to see the market diversify,” says Gerbrand Ceder, a materials scientist at the University of California, Berkeley. The pursuit of better car batteries is fierce, in large part because the market is skyrocketing. More than a dozen nations have declared that all new cars must be electric by 2035 or earlier.

唯一　

The International Energy Agency forecasts that the global stock of EVs on the road will rise from 16.5 million in 2021 to nearly 350 million by 2030 (see go.nature.com/42mpkqy), and that demand for energy from EV batteries will reach 14 terawatt hours (TWh) by 2050, which is 90 times more than in 2020 (ref. 2). Car batteries have a stiff list of requirements. They need to pack a lot of energy into as little material and weight as possible so that cars can go farther on a single charge. They need to provide enough power for acceleration, recharge fast, have a long lifespan (the common standard is to withstand 1,000 full recharging cycles, which should last a consumer 10–20 years),

CHRIS RATCLIFFE/BLOOMBERG/GETTY

Alternatives to lithium-ion cells could power future electric vehicles. By Nicola Jones

SOURCE: ADAPTED FROM G. HARPER ET AL. NATURE 575, 75–86 (2019) AND G. OFFER ET AL. NATURE 582, 485–487 (2020)

唯一　 work well across wide temperature ranges and be safe and affordable. “It’s very hard to optimize all these things at once,” says Linda Nazar, a battery researcher at the University of Waterloo, Canada. So researchers are pursuing a plethora of options, with different targets in mind. The US Department of Energy’s (DoE’s) Battery500 programme, launched in 2017, is aiming for a cell energy density of 500 watt-hours per kilogram (Wh kg–1), a 65% boost compared with today’s best products. The PROPEL-1K programme, launched last year by the US Advanced Research Projects Agency–Energy, is ambitiously aiming for a longer-term goal of 1,000 Wh kg–1. As for cost, the DoE’s Vehicle Technologies Office is aiming to hit US$60 per kilowatt hour by 2030, about half today’s prices, which it reckons will mean that the price of electric cars will break even with the cost of those powered by gas guzzling petrol engines (see ‘Powering up’). It’s hard to pin down where things stand. Commercial announcements about yet-to-bereleased batteries or cars sometimes emphasize one metric over others, and proprietary claims can be impossible to check until batteries have been tested for years in real-world cars. But it’s clear that decades of work on variants such as solid-state and sodium batteries are finally coming to fruition, says Nazar. As for the far future, plenty of battery chemistries remain tantalizing possibilities. “Now everyone has accepted battery development is really important, everyone is tripping over themselves to do it,” she says.

Electrode evolution Batteries are effectively chemical sandwiches, which work by shuttling charged ions from one side (the anode) to the other (the cathode) through some intermediate material (the electrolyte) while electrons flow in an outside circuit. Recharging the battery means shunting the ions back to the anode (see ‘How a battery works’). Today, most electric cars run on some variant of a lithium-ion battery. Lithium is the third-lightest element in the periodic table and has a reactive outer electron, making its ions great energy carriers. The lithium ions travel between an anode usually made from graphite and a cathode made from a metal oxide, both of which host lithium ions between atomic layers. The electrolyte is typically an organic liquid. Lithium-ion batteries have improved a lot since the first commercial product in 1991: cell energy densities have nearly tripled, while prices have dropped by an order of magnitude3. “Lithium-ion is a formidable competitor,” says Ceder. And with further scope for improvement, some say lithium-ion will be king for a long time. “I think lithium ion will for decades be the technology which powers electric cars, because it’s good enough,” says Winfried Wilcke, a recently retired scientist

HOW A BATTERY WORKS

Conventional lithium-ion batteries share some common features and materials. Lithium ions move from the anode through an electrolyte to the cathode; during recharging, they move back to the anode. Researchers are exploring ways to improve battery design, including six options shown here. Circuit

Electron

Electrolyte

Anode

–

1. Add silicon to the anode

Cathode

Lithium ion

6. Swap lithium to sodium to reduce cost

+

5. Use oxygen from the air as one of the cathode ingredients

1.

Metal oxides, including elements such as manganese, cobalt and nickel

Typically graphite

2. Swap to a lithium-metal anode

3. Use a solid electrolyte

in Los Altos, California, who headed an IBM Research battery project from 2009 to 2015. Most of the improvement in lithium-ion thus far has come from changes to the material of the cathode, resulting in multiple commercial cell types. One, popular in laptops, uses lithium cobalt oxide, which produces relatively light but expensive batteries. Others, popular in many cars, use a mix of nickel and cobalt with aluminium or manganese as a stabilizer (NCA and NCM). Then there’s lithium iron phosphate (LFP), which does without expensive cobalt and nickel but so far has relatively poor energy densities (see ‘Lithium-ion battery types’). LFP’s price has made it attractive and plenty of researchers and companies are working to improve it; US EV manufacturer Tesla notably decided in 2021 to swap to LFP batteries in its mid-range cars. There is scope for more tweaks to the cathode. In NCM batteries, researchers have been paring back more-expensive cobalt in favour of nickel, which also provides a higher energy density. That path has led to commercial NCM811 battery cathodes with 80% nickel, and researchers are now working on NCM955, with 90% nickel. Meanwhile, at the anode, one common option is to swap graphite for silicon, a mat­ erial that can store ten times more lithium atoms per weight. The challenge is that silicon expands and contracts by around 300% during charge–discharge cycles, putting a lot of structural strain on the battery and limiting its lifetime. Even better than a silicon anode is simply lithium itself. “You don’t have any wasted material,” says chemical engineer Brian Cunningham at the DoE’s Vehicle Technologies Office in Arlington, Virginia. In addition to cutting down on weight, this can speed up charging, because there is no waiting for lithium ions to slot in between any layers (this change, technically,

唯一　

4. Swap to a sulfur cathode

makes the design a lithium-metal rather than a lithium-ion battery). But a big problem with this strategy is that during recharging, lithium tends to redeposit on the anode unevenly, with hotspots that form tendrils called dendrites, which can reach out through the electrolyte and short-circuit the battery. Lithium-based batteries with better electrodes can, in theory, achieve huge energy densities, but often have trade-offs in terms of cell lifetimes or safety. Last year, one group of researchers in China reported a cell with a lithium-metal anode (and a type of lithium-rich cathode) that hit higher than 700 Wh kg–1 in the lab4. The group’s start-up firm, WeLion New Energy in Beijing, is aiming to develop and commercialize this battery, along with other options. Another aspirational idea offering high energy densities is a lithium sulfur (LiS) battery, with a lithium-metal anode and a sulfur cathode. But sulfur reacts with lithium to make soluble products that can deposit on the anode and kill the battery. LiS “has been tried for 30 years and it still has major challenges”, says Ceder. With such troubles plaguing batteries with better electrodes, many say the most enticing solution is to replace the liquid electrolyte with a solid.

Solid idea The idea of solid-state batteries is to use a ceramic or solid polymer as the electrolyte, which hosts the passage of lithium ions but helps to stem dendrite formation. Not only does this make it easier to use an all-lithium anode — with the attendant energy-density advantage — but getting rid of the flammable organic liquid also means removing a hazard that can cause fires. The cell architecture of solid-state batteries is simpler than that of liquid-based cells, says Nazar. And the solid batteries, in theory, work better both at low Nature | Vol 626 | 8 February 2024 | 249

Feature Meanwhile, plenty of researchers are pursuing ways to improve solid state. Chemist Jennifer Rupp at the Technical University of Munich in Germany has founded a company, QKera, also in Munich, that manufactures ceramic electrolytes at half the usual 1,000 °C temperature. That both helps to limit carbon dioxide emissions from the furnaces used in the manufacturing process and helps to resolve some issues over binding the electrolyte to the cathode. Another promising angle, says Nazar, is a new class of oxyhalide electrolytes for solid-state batteries. Some of these are ‘gooey’ and so more flexible, which should ease manufacturing and make them less vulnerable to cracking5. And some have extremely high conductivity, letting lithium ions zoom through as if through a liquid rather than a solid, with associated power benefits6. Other

“Some of my colleagues call it fairy-tale chemistry.” firms are working on a solid-state version of LiS, says Cunningham. The ‘pot of gold’ battery at the end of this solid-state rainbow, many say, would be a lithium–air design. This kind of battery uses a lithium-metal anode, and the cathode is based on lithium binding to oxygen that is pulled from the air and released again when the battery recharges. In part because a key cathode ingredient isn’t stored in the battery, this design can hold much more energy per kilogram. But the idea has long seemed speculative. “Some of my colleagues call it fairy-tale chemistry,” says Nazar. Materials scientist Larry Curtiss at Argonne National Laboratory in Lemont, Illinois, and his colleagues hit the headlines in 2023 with a surprising paper showing a solid-state, experimental lithium–air battery tested over 1,000 cycles in the lab1. The team says its coin-sized test cell runs at about 685 Wh kg–1 and should be able to reach 1,200 Wh kg–1, four times what’s achievable with lithium-ion now and roughly comparable with the energy density of petrol in cars. The experimental system works using a new chemistry that surprised even the team

studying it. Previous lithium–air battery projects, typically using liquid electrolytes, made lithium superoxide (LiO2) or lithium peroxide (Li2O2) at the cathode, which store one or two electrons per oxygen molecule. The new cell instead makes lithium oxide (Li2O), which can hold four. Those extra electrons translate to a higher energy density, and the system seems a lot more stable than previous efforts, which should lead to longer battery life. “It’s unbelievable what they did,” says Wilcke. “They can use ordinary dirty air with moisture and carbon dioxide and all the other crap that you find in unfiltered air. Not a problem,” says Wilcke. But many say they would like to see the effort replicated before getting too excited. And although it’s a great energy storage system, it’s unclear how it would work in practice — how you could get the air in and out, for example, and whether it can be built bigger and made to work with higher currents. “It’s definitely a much longer time horizon then than even lithium sulfur,” says Cunningham. Curtiss says the team is thinking about aviation as the best application for the technology, given that it’s so energy dense. Wilcke agrees. Energy density is a “huge, huge factor in aircraft”, says Wilcke, who is bullish in particular on electric vertical take-off and landing craft, expected to be used as ‘flying taxis’. If that sounds like science fiction, an electric air taxi was licensed to fly in China — even without a pilot — in October 2023, and several companies make craft that can go a couple of hundred kilometres on lithium-ion batteries. Air taxis that can skip the traffic taking you from the airport to your hotel, Wilcke says, are an emergent industry that’s about to take off.

Price drop As the quest continues for miracle batteries that pack in ever more energy, some scientists argue that the most pressing concern is the need to pick a battery chemistry that will be cheap and sustainable in the long run. “The biggest challenges are resource-related,” says Ceder, who calculates that the projected 14 TWh needed for cars by 2050 will require 14 million tonnes of total metal. That’s a lot; for comparison, today’s global mining of lithium is about 130,000 tonnes per year, whereas cobalt is nearly 200,000

LITHIUM-ION BATTERY TYPES

A variety of cathode materials are used, each with pros and cons. Nickel cobalt aluminium oxide (NCA) 8%

Lithium iron phosphate (LFP) 30%

NCM

Other 2%

BATTERY TYPES IN 2022

Nickel cobalt manganese oxide (NCM) 60%

LFP

Energy density

Affordability

Power

Life-span

Safety Performance at extreme temperatures

250 | Nature | Vol 626 | 8 February 2024

唯一　

NCA

SOURCES: IEA AMD Y. MIAO ET AL. ENERGIES 12, 1074 (2019)

temperatures (because there’s no liquid to get more viscous when it’s cold) and at high temperatures (because the interfaces with the electrodes don’t suffer so much when it’s hot). But there are challenges: in particular, how to manufacture a smooth, flawless interface between the layers. Also, the transport of ions through a solid tends to be slower than through a liquid, limiting power. And solid-state batteries require an entirely new manufacturing process. “From all we see, they will be more expensive,” says Ceder. “Solid state has a big future. No question. But it’s bloody difficult to make it happen,” says Wilcke. Some battery companies are moving forward with solid state. Colorado-based Solid Power in Louisville (partnered with car makers BMW and Ford), for example, has begun pilotscale production of a solid-state cell with a silicon-based anode that they say hits 390 Wh kg–1, and California-based QuantumScape (which has signed deals with manufacturers including Volkswagen) has a solid-state battery that gets the advantages of a lithium anode with an even lower-weight, anode-less design. Lithium metal gathers at the anode side, but there is no need for a lithium plate there to start with. Some of these battery details are proprietary. QuantumScape has released some prototype performance data, but won’t say what its electrolyte is made from or what the energy density is of its intended first commercial product. In general, the touted higher energy densities for solid-state batteries are “unproven today at any sort of commercial scale”, says Ceder. Actual cars powered by solid-state batteries seem to be perpetually on the horizon: Toyota’s original target date for commercializing them in the early 2020s has now slipped to the late 2020s, for example. When it comes to batteries, “Toyota has said a lot of things in the last ten years, none of which have come through,” cautions Ceder. But Nazar thinks the time frame in general is realistic. “I believe that in 2025, we’re probably going to see some market intrusion of some of these cells,” she says, especially given that there are some ambitious Chinese companies on the case. That includes the world’s largest battery manufacturer, Contemporary Amperex Technology (CATL), headquartered in Ningde.

SOURCE: REF. 3

唯一　 tonnes and nickel 3.3 million tonnes — that’s for all purposes, including non-EV batteries and, for nickel, stainless steel. The quantity needed makes it important to choose metals that are not scarce or expensive and do not cause excessive environmental damage when they are mined. Plenty of researchers and companies are trying to make batteries that don’t use nickel, cobalt or other expensive metals. QuantumScape, for example, says its batteries have this advantage, as do lithium–air concepts, LiS (if it can be made to work), other experimental materials7 and the already commercial LFP cathodes (although LFP might put a strain on phosphorus resources if that technology scales up a lot). Ceder is looking at alternative cathodes called disordered rocksalts (DRX)8. These rely on the idea that lithium ions can just meander through a crystalline cathode rather than taking an ordered path through layers, and thus the cathode can be made with almost any transition metals. Ceder’s team favours manganese and titanium. He expects the first batteries with DRX cathodes to be cheaper than current lithium-ion cells and to achieve comparable energy densities. Perhaps the ultimate goal is to get rid of the lithium itself — a metal that has seen wild price swings thanks to booming demand and supply pinchpoints. In 2022–23, for example, battery-grade lithium carbonate prices briefly spiked at six times higher than usual. Researchers have toyed with replacing lithium with plenty of other charge carriers, including magnesium, calcium, aluminium and zinc, but work on sodium is the most advanced. Sodium lies directly beneath lithium in the periodic table, making its atoms heavier and bigger, but with similar chemical properties. This means a lot of the lessons from lithium battery development and manufacturing can be copied over to sodium. And sodium is much easier to source: it’s about 1,000 times more plentiful in Earth’s crust than is lithium. “Sodium is just unbelievably abundant,” says Ceder, who thinks sodium batteries could end up costing around $50 per kilowatt hour. Sodium batteries are already in production (see go.nature.com/3tnwdgt). Chinese conglomerate BYD — which in early 2024 replaced Tesla as the world’s largest EV manufacturer — has broken ground on its first sodium-ion battery plant. And Chinese car makers Chery, JMEV and JAC have all announced budget cars powered by sodium-ion batteries in their line-up for China this year. List prices for these small cars are expected to be around $10,000. On the plus side, sodium’s larger atomic size opens up more options for the metals that can be used in the layered metal oxides at the cathode, says Ceder: “There’s a lot more chemical flexibility.” And researchers could make an anode-less solid-state battery with sodium, too — an enticing possibility, says Nazar.

POWERING UP

Research is pushing energy density ever higher in all battery types, with some extremely high possibilities. But these can come with some trade-offs on price and other performance measures.

Conventional lithium ion (LFP, NCM, NCA)

Original Sony lithium-ion battery (120)

Sodium ion

Energy density of petrol in cars (1,700)

Experimental result for lithium-metal liquid electrolyte cell (711)

Lithium metal Lithium sulfur

0

ARPA-E Propel 1k target (1,000)

CATL sodium battery (160)

Silicon anode

Solid state

DoE battery 500 target (500)

Existing batteries Future potential

Solid power (390)

200

400

Projection for lithium–air Experimental result for lithium–air solid-state cell (680) solid-state cell (1,200)

600 800 1,000 Energy density (Wh kg–1)

1,200

1,400

1,600

LFP, Lithium iron phosphate; NCM, nickel cobalt manganese; NCA, nickel cobalt aluminium oxide; CATL, Contemporary Amperex Technology Company; DoE, US Department of Energy; ARPA-E, Advanced Research Projects Agency–Energy.

But the heavier weight of sodium compared to lithium makes it fundamentally harder to get to high energy densities. There also hasn’t been as much time to develop the best electrodes and electrolytes — sodium-ion battery energy density now roughly matches that of the best lithium-ion batteries from a decade ago. CATL has a sodium battery that hit an advertised energy density of 160 Wh kg–1 in 2021 at a reported price of $77 per kilowatt hour; the company says that will ramp up to 200 Wh kg–1 in its next model. These lower energy densities mean that range is limited. The ultra-compact cars expected to run on sodium batteries have advertised ranges of around 250–300 km, compared with nearly 600 km for a lithium-powered Tesla Model S. “It’s going to need chemistry advances in order to get to the level that is necessary for the automotive market in the United States,” says Cunningham, where consumers are used to longer drives and bigger cars. Some companies, including UK-based Faradion and Swedish Northvolt, are promoting their sodium batteries (also both advertised at 160 Wh kg–1) to store excess renewable energy for electricity grids, where sodium’s weight problem is less of an issue.

help to explore more options more quickly. For example, the DoE’s Pacific Northwest National Laboratory in Richland, Washington, is working with Microsoft to rapidly come up with new battery materials; a lithium–sodium solid electrolyte found this way is now in initial tests. But these AI strategies are limited by the information that chemists have to feed into them, says Nazar. There are still plenty of unknowns, she says, about what’s actually going on at the atomic level at the interface of electrode and electrolyte materials. In the end, experts say we’re likely to see a range of batteries for our future cars — in much the same way that we have 2-, 4- and 6-cylinder engines today. We might see sodium batteries or LFP for lower-range cars, forklifts or specialist vehicles, for example. Then there might be improved lithium-ion batteries, maybe using silicon anodes or rocksalt cathodes, for midrange vehicles, or perhaps solid-state lithium batteries will take over that class. Then there might be LiS or even lithium–air cells for highend cars — or flying taxis. But there’s a lot of work yet to be done. “All of the different chemistries that aren’t commercialized today have their pros and cons,” says Cunningham. “Our job is to remove all those cons.”

Guess and test

Nicola Jones is a freelance journalist in Pemberton, Canada, and drives an electric car.

Battery development is onerous, because the behaviours of materials are not always predictable. Rupp says, for instance, that it currently takes researchers 8–15 years to come up with new solid-state electrolyte designs and optimize the specifications, including which additives to use and how to pack in high densities of lithium. “This gives me as a material scientist two-and-a-half more materials to work on” before retirement, says Rupp. “That’s too slow”. Assistance is coming from artificial intelligence (AI) and automated synthesis, which can

1. Kondori, A. et al. Science 379, 499–505 (2023). 2. International Energy Agency. Net Zero by 2050: A Roadmap for the Global Energy Sector (IEA, 2021). 3. Frith, J. T., Lacey, M. J. & Ulissi, U. Nature Commun. 14, 420 (2023). 4. Li, Q., Yang, Y., Yu, X. & Li, H. Chinese Phys. Lett. 40, 048201 (2023). 5. Jung, S-K. et al. ACS Energy Lett. 6, 2006–2015 (2021). 6. Dai, T. et al. Nature Energy 8, 1221–1228 (2023). 7. Chen, T. et al. ACS Cent. Sci. https://doi.org/10.1021/ acscentsci.3c01478 (2024). 8. Lun, Z. et al. Nature Mater. 20, 214–221 (2021).

Nature | Vol 626 | 8 February 2024 | 251

Santorini is a tourist haven and also the site of gigantic ancient eruptions.

CLUES OF GIANT PREHISTORIC ERUPTION FOUND AT SANTORINI

An expedition that drilled into the sea floor around the famous Greek island found signs of a gargantuan blast 520,000 years ago and more recent eruptions.

O

By Alexandra Witze

ne of the world’s most-studied volcanoes turns out to be hiding plenty of secrets. Geologists have unearthed major clues about past eruptions of the Greek island of Santorini by drilling into the sea floor around the partially submerged volcano.

252 | Nature | Vol 626 | 8 February 2024

Santorini is famous among volcanologists for its Bronze Age eruption in approximately 1600 bc, which might have contributed to the decline of the Minoan civilization on the island of Crete. Santorini is also home to more than 15,000 residents and attracts around 2 million tourists each year, who instagram their way around the white- and blue-washed

buildings set against the glittering sea. During an expedition between late 2022 and early 2023, researchers discovered evidence of a previously unknown cataclysm. Half a million years ago, the volcano erupted violently enough to blanket three nearby islands in debris, and it sent underwater currents racing for 70 kilometres. The eruption was much larger than the one in 1600 bc and was one of the biggest ever in this part of the Mediterranean. The expedition also pulled up evidence that Santorini erupted in the year ad 726 in a blast approximately the size of Mount St Helens’ in Washington in 1980. No one had understood the scale and scope of these eruptions until now. “The history of Santorini is being written again,” says Paraskevi Nomikou, a marine geologist at the National and Kapodistrian University of Athens in Greece, who was a researcher on the expedition. Although scientists aren’t expecting similar eruptions to happen any time soon, the findings add to the growing understanding of the volcanic risk at Santorini, which last erupted in 1950. A related volcano, Kolumbo, lies underwater just 7 kilometres away; it last erupted in 1650 and is also considered active. Both Santorini and Kolumbo are part of the Hellenic volcanic arc, a chain of mostly underwater volcanoes that sit at the junction where the plate of Earth’s crust that carries Africa dives beneath the Aegean Sea plate. With its explosive history and thriving tourist trade, Santorini is one of the most hazardous volcanoes in Europe. Researchers have pieced together much of its eruptive past, by gathering evidence from rocks on land and from cores that could be obtained fairly easily from the top few metres of the Mediterranean sea floor. But part of Santorini’s history is buried deep beneath the sea floor and had remained inaccessible. That is, until the drill ship JOIDES Resolution arrived in December 2022 for a 2-month expedition; the researchers drilled 12 holes into the sea floor and pulled up long cores of sediment and rock in and around Santorini and Kolumbo (see ‘Eruption clues’). “By going into the marine realm we can go further back in time,” says Timothy Druitt, a volcanologist at the University of Clermont Auvergne in Clermont-Ferrand, France, and co-chief scientist of the expedition. The drilling around Santorini wasn’t easy; several of the drill holes collapsed in a slurry of pumice and ash, which glommed onto drilling equipment like superglue. At one point, on New Year’s Eve, technicians had to sever the pipe used for drilling, leaving some of it in the hole. The many drilling challenges meant that “we had several days where we were just sitting there not knowing what to expect for the next days”, says Steffen Kutterolf, a volcanologist

MARCOS DEL MAZO/LIGHTROCKET VIA GETTY

Feature

唯一　 at the GEOMAR Helmholtz Centre for Ocean Research in Kiel, Germany, and the expedition’s other co-chief scientist. But ultimately, the team extracted an unprecedented data set about the region’s volcanic past. Some of that volcanic past involved submarine eruptions, in which most of the volcanic debris never reaches above the ocean surface. Learning about these eruptions is important because they can be powerful and are poorly understood: in January 2022, an underwater volcano exploded near Tonga in the most violent eruption in decades, yet nearly all of the evidence that it happened remains beneath the waves. Among the most significant discoveries at Santorini was a thick layer of the volcanic rock called tuff, which kept appearing in core after core, created by a huge prehistoric eruption. “Slowly it began to dawn on us that this was a major [geological] unit we didn’t know anything about,” says Druitt. The researchers named it the Archaeos tuff, after the Greek word for ‘ancient’. It formed around 520,000 years ago when Santorini erupted underwater, sending shards of ash and rock racing outwards like giant avalanches, the team reported in January in Communications Earth & Environment1. As measured by the size of those underwater flows, the eruption was 6 times larger than the 1600 bc eruption at Santorini and 10 times larger than the 2022 Tonga eruption. But Druitt says not to worry about such an ancient eruption: “There’s absolutely no

reason to think that Santorini is going to do anything like this in the near future.” More relevant to modern hazards is the discovery of how big the eruption in ad 726 was. Historical accounts relate that “the sea was seen to boil” that year, and that large blocks of pumice — lightweight rocks that often form during underwater eruptions — floated to the surface and travelled for hundreds of kilometres. But researchers had little information about the scale and nature of the eruption, Jonas Preine, a marine geophysicist at the University of Hamburg in Germany, said in December at a meeting of the American Geophysical Union in San Francisco. With the JOIDES Resolution, he said, the team aimed to “solve a historic mystery and find a lost eruption”. The researchers did this by comparing rocks drilled during the expedition with information about the layering in the rocks beneath Santorini that comes from studies of seismic waves passing beneath the island. Preine and his colleagues concluded that they had indeed found widespread evidence of an eruption in ad 726 (ref. 2). “The 726 eruption has always been used as a worst-case scenario” for a modern eruption at Santorini, says Druitt. “What’s interesting is, the worst-case scenario has just increased in magnitude quite a lot.” Among other things, Preine’s team found that the material from the ad 726 eruption is crumbly, meaning that a future eruption could destabilize that layer and lead to greater chances of underwater

ERUPTION CLUES

Amorgos

Researchers drilled into the sea floor at 12 sites near the Greek island of Santorini and pulled up geological evidence of past eruptions, including a giant blast around 520,000 years ago.

Ios

Sikinos Drill sites Folegandros

Submerged volcano Kolumbo

Aegean Sea

SOURCES: ELEVATION: NASA; BATHYMETRY: EMODNET

Santorini Anafi

Christiana volcano

Area of detail Greece

Hellenic volcanic arc

landslides, which could trigger tsunamis. The ocean-drilling expedition has added important chapters to the scientific understanding of Santorini, says Emilie Hooft, a geophysicist at the University of Oregon in Eugene, who did not go on the expedition but who has been working to map the magma chambers beneath Santorini and Kolumbo3. “The history of the volcano was perceived to be so well known from all the work that was done on land,” she says. And yet “there’s actually a lot to be discovered about the interaction with the eruptive system and the marine environment” — such as how those crumbly layers formed during the underwater eruption in ad 726. Greek authorities regularly monitor geological activity at Santorini, by tracking earthquakes, movement of the ground and other changes. The JOIDES Resolution expedition was a high-profile chance to inform residents about the volcano’s hazards, says Nomikou, who was born and raised on the island. She gave talks and virtual tours to school groups about how researchers are monitoring Santorini and neighbouring volcanoes. “There is no need for panic,” she says. Although magma continues to pool beneath Santorini, it also leaks out in minor eruptions such as those in the 1920s through to the 1940s that created small lava flows on uninhabited islands. The magma leakage takes some of the pressure off the system, Druitt says. Still, in 2011 and 2012, a months-long period of small earthquakes and ground shifts rattled Santorini, frightening the public and triggering emergency authorities to research evacuation scenarios. And the growing understanding of the threat of submarine eruptions means that researchers need to keep a particular eye on Kolumbo. Its crater floor is around 500 metres below sea level but the rim is around 20 metres beneath the waves. Kolumbo’s last eruption, in 1650, generated poisonous gases that killed around 70 people and many animals on Santorini. But Kolumbo is now monitored better than ever; over the past year, Nomikou has led an effort to install geochemical, seismic and other instruments on the sea floor around the volcano4. Early results show that Kolumbo has more than 300 active volcanic chimneys that spew carbon dioxide into the surrounding waters. “It’s a very toxic environment,” says Nomikou. And that’s why she spends so much time educating people on Santorini about its unique risks: “They need to know that they are living in an active volcano.” Alexandra Witze writes for Nature from Boulder, Colorado. 1. Druitt, T. et al. Commun. Earth Environ. 5, 24 (2024). 2. Preine, P. et al. Nature Geoscience (in the press). 3. Chrapkiewicz, K. et al. Geochem. Geophys. Geosys. 23, e2022GC10475 (2022). 4. Nomikou, P. et al. Front. Mar. Sci. 9, 796376 (2022).

Nature | Vol 626 | 8 February 2024 | 253

Science in culture 唯一　

PETER MENZEL/SPL

Books & arts

DNA sequencing has become routine, but the roles of individual genes can be hard to be pin.

Genes are not the blueprint for life The view of biology often presented to the public is oversimplified and out of date. By Denis Noble 254 | Nature | Vol 626 | 8 February 2024

F

or too long, scientists have been content in espousing the lazy metaphor of living systems operating simply like machines, says science writer Philip Ball in How Life Works. Yet, it’s important to be open about the complexity of biology — including what we don’t know — because public understanding affects policy, health care and trust in science. “So long as we insist that cells are computers and genes are their code,” writes Ball, life might as well be “sprinkled

唯一　

PHILIPPE PLAILLY/SPL

唯一　 with invisible magic”. But, reality “is far more interesting and wonderful”, as he explains in this must-read user’s guide for biologists and non-biologists alike. When the human genome was sequenced in 2001, many thought that it would prove to be an ‘instruction manual’ for life. But the genome turned out to be no blueprint. In fact, most genes don’t have a pre-set function that can be determined from their DNA sequence. Instead, genes’ activity — whether they are expressed or not, for instance, or the length of protein that they encode — depends on myriad external factors, from the diet to the environment in which the organism develops. And each trait can be influenced by many genes. For example, mutations in almost 300 genes have been identified as indicating a risk that a person will develop schizophrenia. It’s therefore a huge oversimplification, notes Ball, to say that genes cause this trait or that disease. The reality is that organisms are extremely robust, and a particular function can often be performed even when key genes are removed. For instance, although the HCN4 gene encodes a protein that acts as the heart’s primary pacemaker, the heart retains its rhythm even if the gene is mutated1. Another metaphor that Ball criticizes is that of a protein with a fixed shape binding to its target being similar to how a key fits into a lock. Many proteins, he points out, have disordered domains — sections whose shape is not fixed, but changes constantly. This “fuzziness and imprecision” is not sloppy design, but an essential feature of protein interactions. Being disordered makes proteins “versatile communicators”, able to respond rapidly to changes in the cell, binding to different partners and transmitting different signals depending on the circumstance. For example, the protein aconitase can switch from metabolizing sugar to promoting iron intake to red blood cells when iron is scarce. Almost 70% of protein domains might be disordered. Classic views of evolution should also be questioned. Evolution is often regarded as “a slow affair of letting random mutations change one amino acid for another and seeing what effect it produces”. But in fact, proteins are typically made up of several sections called modules — reshuffling, duplicating and tinkering with these modules is a common way to produce a useful new protein. Later in the book, Ball grapples with the

DNA alone cannot reveal how life works.

philosophical question of what makes an organism alive. Agency — the ability of an organism to bring about change to itself or its environment to achieve a goal — is the author’s central focus. Such agency, he argues, is attributable to whole organisms, not just to their genomes. Genes, proteins and processes such as evolution don’t have goals, but a person certainly does. So, too, do plants and bacteria, on more-simple levels — a bacterium might avoid some stimuli and be drawn to others, for instance. Dethroning the genome in this way contests the current standard thinking about biology, and I think that such a challenge is sorely needed. Ball is not alone in calling for a drastic rethink of how scientists discuss biology. There has been a flurry of publications in this vein in the past year, written by me and others2–4. All outline reasons to redefine what genes do. All highlight the physiological processes by which organisms control their genomes. And all argue that agency and purpose are definitive characteristics of life that have been overlooked in conventional, gene-centric views of biology. How Life Works: A User’s Guide to the New Biology Philip Ball Pan Macmillan (2024)

This burst of activity represents a frustrated thought that “it is time to become impatient with the old view”, as Ball says. Genetics alone cannot help us to understand and treat many of the diseases that cause the biggest health-care burdens, such as schizophrenia, cardiovascular diseases and cancer. These conditions are physiological at their core, the author points out — despite having genetic components, they are nonetheless caused by cellular processes going awry. Those holistic processes are what we must understand, if we are to find cures. Ultimately, Ball concludes that “we are at the beginning of a profound rethinking of how life works”. In my view, beginning is the key word here. Scientists must take care not to substitute an old set of dogmas with a new one. It’s time to stop pretending that, give or take a few bits and pieces, we know how life works. Instead, we must let our ideas evolve as more discoveries are made in the coming decades. Sitting in uncertainty, while working to make those discoveries, will be biology’s great task for the twenty-first century. Denis Noble is emeritus professor of physiology and biology at the University of Oxford, UK. 1. Noble, D. Prog. Biophys. Mol. Biol. 166, 3–11 (2021). 2. Noble, R. & Noble. D. Understanding Living Systems (Cambridge Univ. Press, 2023). 3. Vane-Wright, R. I. & Corning, P. A. Biol. J. Linn. Soc. 139, 341–356 (2023). 4. Corning, P. A. et al. (eds) Evolution “On Purpose”: Teleonomy in Living Systems (MIT Press, 2023).

Nature | Vol 626 | 8 February 2024 | 255

唯一　

VUK VALCIC/SOPA IMAGES/LIGHTROCKET/GETTY

Books & arts

Scientists joined climate activists in challenging the UK government’s fossil-fuel policies in London in September 2023.

Science and government: can the power struggle ever end? Similar goals but different strategies underlie tensions between science and the state, an in-depth analysis explains. By Rhona Mijumbi

R

elations between scientific communities and governments are often tense — think back to how science advisers were both given a platform and criticized during the COVID-19 pandemic. Although scientists and policymakers should be allies, working together and enhancing one another’s efforts to solve society’s problems, in reality they often compete for power and fight for superiority. 256 | Nature | Vol 626 | 8 February 2024

When Science meets Power explores that dynamic. This engaging book by Geoff Mulgan, a public-policy researcher at University College London, is packed with wisdom and is written with a deep analytical prowess that has been acquired over decades of work at this interface — including as director of policy to former UK prime minister Tony Blair and founder of the progressive think-tank Demos.

With his inside view, Mulgan’s account is a breath of fresh air that stands apart from others that only theorize on science’s role in political decision-making. The struggle between science and government, Mulgan explains, stems in part from their having similar objectives yet different approaches to achieving them. Both sides aim to provide solutions for society, but

DREW ANGERER/GETTY

唯一　 policymakers need answers immediately, whereas scientists can afford the luxury of waiting to gain perfect understanding. Mayors of cities, such as Mombasa, Kenya, that frequently face floods need to take informed steps quickly to limit damage. They cannot wait years for better computer models to offer optimal solutions. Science also walks a systematic journey, hypothesizing, theorizing and refining research questions and methods inside boundaries that can be controlled. However, a government’s path from problem to solution is often dictated by circumstances out of its control, such as when it is dealing with the complex aftermath of a natural disaster or trying to reduce crime. The definition and boundaries of the problem can shift, as can the resources that are available, leading scientists to view these pragmatic decision-making processes as subjective and irrational. What does this relationship look like? Mulgan describes times when the state was the dominant partner, and science had to fight for recognition. For example, Soviet physicist Peter Kapitsa discreetly wrote to then-leader Joseph Stalin in 1945 asking for respect for science alongside the state and the economy, with the view that these three pillars should be equally important in helping the country to move forward. Similarly, in Uganda in the mid1970s, scientists were at the mercy of the militarized state under then-president Idi Amin. Many lived in fear or fled the country, because their influence was perceived to rival that of the politicians. It seems to me that this relationship will always be a seesaw, with power rocking back and forth as times, places, situations and players change. Reading When Science meets Power also has me wondering about the role of culture in this dynamic, whether individual, group or jurisdictional. Chief US and UK science advisers, for example, might experience different atmospheres depending on who is president or prime minister at the time and the nature of the social, legal and institutional systems in which they operate. Each shift in shared ideas, social behaviours and systems of knowledge influences the value placed on science or government. For example, in today’s period of polarized politics, populism and fake news, the balance is tipping once again. Science is losing some of its power and politics is again gaining the upper hand. The question, then, is how can scientists and policymakers create mutual respect and understanding when power lies in one court or the other? Both might learn from occasions when science and the state have worked closely, if only for a while. For example, during the COVID-19 pandemic, many governments looked to scientists for advice in decision-making, and the scientific community expected the state to act on that advice. Even if there

Immunologist Anthony Fauci (left) advised US leaders after COVID-19 spread in 2020.

was friction, both worlds acknowledged each other’s value in running society. I think that one answer lies in what Mulgan also calls for: more collaboration and better links between science and governance. Another answer lies in harnessing opportunities to work synergistically. For example, governments can use artificial intelligence (AI) to increase the efficiency of their health,

“This relationship will always be a seesaw, with power rocking back and forth.” information and transport sectors, and scientists can help them to do that. But scientists and the public also need AI to be regulated to help realize its potential and manage its rapid growth while stopping it from causing harm. Maintaining public trust is another point of connection. Concerns over technologies such as nuclear weapons, gene editing and AI mean that many members of the public want politicians to keep the frontiers of science in check. When Science Meets Power Geoff Mulgan Polity (2023)

Similarly, people weary of politicians’ motives want scientists to hold them to account. Scientists and government working together effectively might comfort the public. In my favourite part of the book, Mulgan highlights the importance of brokers and intermediaries who work at the interface between science and government, developing structures, relationships and networks that improve access and linkages. Intermediaries’ ability to “think about thought and to be intelligent about intelligence” makes them crucial for raising awareness of each side’s issues and for promoting self-regulation, which would otherwise be difficult for either side. In conclusion, it is important to understand that the science–policy interface is not a meeting point but a continuum — a pendulum of power that swings from one side to the other, with a rare point of equilibrium at which science and government are mutually respectful allies. Extending the point of equilibrium and improving the power relationship between science and government requires each side investing in understanding how the other’s world works, and what, where and when the intersection of these two is. Investing in government science advisers and other knowledge brokers can ease the tension between the two worlds. Mulgan’s excellent analysis lays a strong foundation on which these essential investments can be made. Rhona Mijumbi is a public-policy researcher at the Liverpool School of Tropical Medicine, UK, and the Center for Rapid Evidence Synthesis, Kampala, Uganda. e-mail: [email protected]

Nature | Vol 626 | 8 February 2024 | 257

Setting the agenda in research

RICARDO RUBIO/EUROPA PRESS/GETTY

Comment

Injections of the diabetes drug Ozempic (semaglutide) decrease appetite, among other effects.

No ‘easy’ weight loss: don’t overlook the social cost of anti-obesity drugs Alexandra Brewis & Sarah Trainer

Ideas of diet and exercise as the ‘best’ way to lose weight could stigmatize people taking Ozempic, WeGovy and other blockbuster drugs that affect appetite. Lessons from weight-loss surgery reveal ways to help.

 

 

 

 

 

 

258 | Nature | Vol 626 | 8 February 2024

T

he arrival of a new class of drugs that curb appetite — including semaglutide, sold as Ozempic and WeGovy by the Danish company Novo Nordisk — has triggered a surge in their use to accelerate weight loss. In 2023, around 1.7% of people in the United States were prescribed a semaglutide medication, and demand is growing fast around the world. These injectable drugs could help to prevent and treat diabetes and other chronic conditions often associated with obesity — classed as having a body mass index (BMI; weight divided by height squared) of more than 30. The World Obesity Federation in London estimated that 770 million adults worldwide could be classed

medically as obese in 2020, and predicts that the number could exceed one billion by 2030. Weight loss, however, is not just a medical phenomenon — it’s a social one, too. As anthropologists, we’re well aware that drastic weight loss can reshape people’s social lives and emotional well-being in negative as well as positive ways. We’ve seen it before, in the context of weight-loss surgery. Between 2013 and 2016, through in-depth interviews, we traced the journeys of 35 people undergoing bariatric surgery in the United States. We tracked the experiences of another 300 individuals through surveys1. Bariatric surgery restricts food intake and absorption by reducing the

Such cultural beliefs profoundly influenced the weight-loss experiences of the people with whom we spoke1. More than half of our US study participants

Nature | Vol 626 | 8 February 2024 | 259  

Judged for weight loss

 

tions to help lose weight7. Social-media posts provide myriad examples of how weight loss and staying slim through diet and exercise is viewed as a virtuous achievement. This general idea that success should be the outcome of personal effort is embedded in many aspects of contemporary life, including education and wealth8.

 

“People had lost their physical weight, but they still carried the stigma of being labelled as lazy and undisciplined.”

didn’t tell people other than close family that they were undergoing bariatric surgery, because they feared judgement. This fear was borne out in the remainder of the cohort — among those who did speak openly about their surgery, 90% were told by at least one person in their social networks that they were cheating at weight loss. Because they underwent a surgical intervention, they were viewed as not working hard enough, not showing enough discipline or not displaying enough moral fortitude to ‘deserve’ their weight loss. They had lost their physical weight, but they still carried the stigma of being labelled as lazy and undisciplined. We found the same pattern in Brazil, where women who underwent bariatric surgery were told by family, friends and strangers that they ‘took the easy way out’9. Such judgement has a cascade of emotional and behavioural effects. Women in Brazil reported a range of reactions to being judged, from anger and frustration to resigned acceptance and agreement. Stress around body weight was a common response in our US interviewees. Stress, in turn, can affect digestion, the immune system, cognitive function, emotional regulation and more — particularly unwelcome at a time when the body is undergoing profound physical changes from surgery. Interviewees worried that their eating would become uncontrolled,

 

Humans excel at judging one another. Bodies — weight, height, clothing, physical signs of disease — are often central to these assessments, in part because they are so visible. In a world of increasingly sedentary work and processed food, ‘thinness’ is difficult to maintain. And so, around the globe, slimmer bodies have become associated with a higher social status2. Thinness is also widely linked to good health in popular culture, even though science suggests a more complex picture. Although a high BMI has been linked to diabetes, for example, it is also a predictor of lower risk of stroke3. In 2023, the American Medical Association in Chicago, Illinois, acknowledged that BMI should not be used as a sole measure for assessing health (see go.nature.com/484j2dt). At the same time, stigmas abound around ‘fatness’ 4. Compared with people who are clinically defined as at a ‘healthy’ weight, those defined as obese report being treated by health-care providers with less care and compassion, having less choice of romantic partners and less access to educational and career-advancement opportunities5. Women seem to be particularly at risk of such discrimination2. For example, heavier-weight daughters tend to receive less financial support for university from their parents than do

lighter-weight female students from families in similar financial circumstances6. There is also a widespread sociocultural myth that becoming and staying thin should be achieved through arduous and morally valued dietary control and exercise. Being fat is seen as a sign of laziness and of a lack of self-discipline — but so is the use of medical interven-

 

Weight is a social issue

The idea that weight loss should be achieved through exercise is rife in popular culture.

 

ATT SAYLES/AP IMAGES FOR WEIGHT WATCHERS/ALAMY

size of the stomach and, often, the length of the intestine. The people we worked with experienced greater self-confidence, in addition to health benefits, after their surgery. But many also had to cope with both unpleasant physical side effects and harsh judgements from others about their choice to lose weight through surgery, rather than through diet and exercise. We anticipate that people who take drugs to lose weight will be similarly affected by side effects and judged by others, with consequences for well-being and mental health. Many people who qualify for the drugs on prescription because they have a BMI of more than 30 will have no weight-associated health complications — unlike those who undergo bariatric surgery in an accredited programme, who typically have at least one chronic health condition. Moreover, with the drugs increasingly sold off-prescription or on the black market, people with BMIs of less than 30 will try them, too, and might experience unnecessary adverse consequences. Here, we call for an urgent and realistic discussion about the social downsides of achieving substantial weight loss through drug use, informed by experiences around bariatric surgery.

Comment labelled by others as fat can shape decisions to use weight-loss medications, and with weight loss comes the expectation of a ‘better’ life with more social acceptance. But, as our research with people undergoing bariatric surgery shows, weight-loss trajectories are not that simple. It’s crucial that pharmaceutical companies, clinicians and researchers consider the emotional aspect of decision-making around anti-obesity drugs.

Next steps First, pharmaceutical companies should market drugs in ways that avoid promoting weight loss as an easy fix. The physical side effects of rapid weight loss should be made clear. Next, as part of this effort, drug firms and

“Clinics and physicians should support people who are taking weight-loss drugs.” researchers should focus on understanding and communicating the interconnected emotional and social ramifications of rapid weight loss. Gaining this knowledge will require tracking the experiences of people using Ozempic, WeGovy and similar drugs over years — by following their changing health, trajectories of weight loss and gain, and their attitudes towards and understanding of these changes. This research must account for the fact that a person’s emotional well-being is dependent on context, by examining how the

attitudes of social circles and broader society affect each individual’s experience. Finally, clinics and physicians should support people who are taking weight-loss drugs. Our bariatric-surgery cohort highlighted the clinic’s support programme as one of the biggest factors in post-operative satisfaction. Educational seminars before surgery enabled participants to make an informed decision about whether the benefits of surgery outweighed the costs for them. Individuals could also access a support group after surgery, to discuss issues and receive support and affirmation. Without this programme, our interviewees told us, they would have had a much harder time coping with negative feedback and judgement. Similar resources should be provided for those who are prescribed weight-loss drugs. These resources don’t need to be in-person — quality educational programmes can be provided online. But they do need accredited professionals and trained moderators, to prevent the risk that participants without medical background will share health ‘tips’ and information that are not medically sound. Taking such steps to ensure that people understand the physical and emotional pros and cons of weight-loss drugs is crucial, to prevent a cascade of unexpected negative social and emotional effects among the millions of people who will take anti-obesity drugs in the coming years.

The authors Alexandra Brewis is professor in the School of Human Evolution and Social Change at Arizona State University, Tempe, Arizona, and founding director of the Center for Global Health, Arizona State University, Tempe, Arizona, USA. Sarah Trainer is director of research oversight at Seattle University, Seattle, Washington, USA. e-mails: [email protected]; [email protected]



1.



2.





3. 4.





5. 6.



7.



8.





9. 10.

Bariatric surgery should not be considered an easy route to weight loss.

Trainer, S., Brewis, A. & Wutich, A. Extreme Weight Loss: Life Before and After Bariatric Surgery (NYU Press, 2021). StrutzSreetharan, S., Brewis, A., Hardin, J., Trainer, S. & Wutich, A. Fat in Four Cultures: A Global Ethnography of Weight (Univ. Toronto Press, 2021). Mehta, A. et al. Nutr. Rev. 80, 2275–2287 (2022). Brewis, A. A., Wutich, A., Falletta-Cowden, A. & Rodriguez-Soto, I. Curr. Anthropol. 52, 269–276 (2011). Puhl, R. M. & Heuer, C. A. Obesity 17, 941–964 (2009). Crandall, C. S. Person. Soc. Psychol. Bull. 21, 724–735 (1995). Brewis, A. & Wutich, A. Lazy, Crazy, and Disgusting: Stigma and the Undoing of Global Health (Johns Hopkins Univ. Press, 2019). Lerch, J. C., Bromley, P. & Meyer, J. W. Int. J. Sociol. 52, 97–127 (2022). Dimitrov Ulian, M. et al. PLoS ONE 18, e0287822 (2023). Raves, D. M., Brewis, A., Trainer, S., Han, S.-Y. & Wutich, A. Front. Psychol. 7, e1497 (2016).

The authors declare no competing interests.

 

 

 

 

 

 

260 | Nature | Vol 626 | 8 February 2024

KATHY YOUNG/AP/ALAMY

and that they would ‘stress eat’ in adverse situations, including when they felt judged10. Sixty-four per cent of our survey respondents reported that they were only ‘somewhat adherent’ to the strict diet required after surgery, and that this self-perceived failure increased their stress, creating a vicious cycle of negative emotions that can make dietary control even harder. And because many people hide the fact that they have had surgery, open discussion about their experiences is lacking in the public domain. Our interviewees often cited reality television as their main source of knowledge about bariatric surgery before coming to the clinic. In such TV programmes, surgery and the resulting weight loss are often portrayed as the successful end point, with ongoing challenges over the following decades mostly invisible. In real life, managing extreme weight loss is a life-long effort. Thirty-six per cent of our survey respondents reported chronic struggles with vitamin deficiencies, and 47% reported new food intolerances after surgery. Nausea was common — as it is for those who take anti-obesity drugs. Other side effects of rapid, extreme weight loss — by any means — include a drawn face and loose, sagging skin that can be painful and becomes infected easily1. As well as being physically challenging, such bodily features are viewed societally as unattractive. In our view, the widespread enthusiasm for weight-loss drugs is not just a solution to a medical problem — it’s also a response to deeply held, widespread fear and anxiety around body weight. The stigma of being

Readers respond

Correspondence Adopt best practice for capturing LGBTQ+ identity data Researchers who identify as members of gender or sexual minority groups face barriers in science, technology, engineering and mathematics (STEM) that others do not (E. A. Cech & T. J. Waidzunas Sci. Adv. 7, eabe0933; 2021). As younger generations increasingly identify as LGBTQ+, the need for inclusivity to maintain a thriving scientific community is growing. Yet, many broad initiatives to address equity, diversity and inclusion (EDI) in STEM fields do not capture LGBTQ+ identities (see Nature 613, 624; 2023). A likely reason for this is the rapidly evolving language around selfidentification, and concerns that surveys might be outdated before they are rolled out. LGBTQ+ STEM is a UK-based organization that supports LGBTQ+ researchers in STEM subjects and advocates for inclusive policies and practice. Recognizing that many STEM organizations might not have the resources to be fully aware of best practice for capturing LGBTQ+ demographic data, we have compiled guidance based on UK community feedback, experiences and wishes, and made this resource available to the broader research community (see go.nature.com/42ysm42). We hope this will empower organizations to inclusively capture data on LGBTQ+ identity, leading to improved EDI practices across STEM fields. Alexander L. Bond Natural History Museum, Tring, UK. [email protected] Tyler L. Kelly University of Birmingham, Birmingham, UK The authors declare competing interests; see go.nature.com/4bcrhgs for details.

Cultivate urban trees Protect wild bees not Japan’s drug-delay for nature-based just honeybees dilemma climate solutions Urban trees boost biodiversity, provide cooling effects and lessen climate-related impacts. Trees deliver ecosystem services that are crucial to the health and well-being of the majority of the global population who live in densely populated cities. But urban trees are threatened by climate change and human-centric cultivation practices, such as planting in extremely limited spaces and enclosing roots with pavement or compacted soil (M. EsperonRodriguez et al. Nature Clim. Change 12, 950–955; 2022). City authorities should encourage natural development of trees to maximize their beneficial potential. This should follow a ‘SETS’ framework that integrates social, ecological and technological considerations (see T. McPhearson et al. One Earth 5, 505–518; 2022), and take into account the environment, economics and equity. Urban tree planning must consider past inequities, such as ‘redlining’ policies that led to a lack of green spaces in Black US neighbourhoods, exacerbating urban heat-island effects (D. H. Locke et al. npj Urban Sustain. 1, 15; 2021). Clear standards for urban tree planting and management are needed, covering considerations such as infrastructure, species, density, climate resilience and human– nature interactions. Lina Tang Institute of Urban Environment, Xiamen, China. [email protected] Guofan Shao Purdue University, West Lafayette, Indiana, USA. Peter M. Groffman The City University of New York, New York, USA.

Tree-planting initiatives, including the World Economic Forum’s 1 Trillion Trees project (www.1t.org) launched in 2020, are a popular way to tackle climate change and biodiversity loss. But challenges such as quantifying carbon stored, the time it takes trees to grow and competing land uses mean that these projects are increasingly supplemented with a swifter solution: pollinator protection. Although the idea of ‘one trillion hives’ might sound positive, I worry that this remedy could be worse than the disease. As pollinator populations decline, the spotlight for conserving this crucial group has fallen on bees — specifically one managed species, the European honeybee (Apis mellifera). Companies from L’Oréal and Guerlain to IBM and Google have rallied around the honeybee. And numbers of honeybees have increased. Yet, the remaining 20,000 species of wild bee still face habitat loss, pesticide exposure and climate change. Two further threats to wild bees are competition for floral resources with managed honeybees (see A. Magrach et al. Nature Ecol. Evol. 1, 1299–1307; 2017) and the transmission of pathogens from managed to wild species (see J. M. Iwasaki and K. Hogendoorn Curr. Res. Insect Sci. 2, 100043; 2022). Conservation of pollinator diversity is nuanced and timeintensive, it involves restoring diverse floral and nesting resources and reducing the use of agrochemicals. Let’s protect all pollinators, not just achieve swift outcomes. Ainhoa Magrach BC3 Basque Centre for Climate Change, Leioa, Spain. ainhoa.magrach@bc3research. org

On 25 December 2023, Japan waived the requirement for domestic phase I clinical trials of drugs developed overseas before Japanese individuals participate in international phase III trials for pharmaceutical regulatory approval. This policy change aims to address Japan’s ‘drug lag’. For example, in 2020, 72% of drugs approved in the United States and the European Union still awaited approval in Japan. Although this step could fast-track drug approvals, it also raises safety concerns. Japan has a history of drug-induced lung diseases, in which some people experience inflammation and scarring of lung tissue when exposed to certain drugs. The prevalence of this is higher in Japan than in other countries, which underscores the risks of relying solely on international trials (K. Kubo et al. Respir. Investig. 51, 260–277; 2013). To mitigate these risks and safeguard participants’ welfare, international trials should analyse data on racial subgroups, especially underrepresented minority ethnic groups, and the medical community, including pharmaceutical companies, should track side effects after drugs reach the market. Mira Namba Keio University, Tokyo, Japan. [email protected] Yudai Kaneda Hokkaido University, Hokkaido, Japan. Akihiko Ozaki Jyoban Hospital of Tokiwa Foundation, Fukushima, Japan. Tetsuya Tanimoto Navitas Clinic Kawasaki, Kanagawa, Japan. The authors declare competing interests; see go.nature.com/3upfhih for details.

Nature | Vol 626 | 8 February 2024 | 261

Expert insight into current research

News & views Planetary science

Surprise ocean prompts update of rules for moons Matija Ćuk & Alyssa Rose Rhoden

The shifting orbit of one of Saturn’s moons indicates that the satellite has a subsurface ocean, contradicting theories that its interior is entirely solid. The finding calls for a fresh take on what constitutes an ocean moon. See p.280 The detection of liquid water oceans under the icy surfaces of outer Solar System moons suggests that these moons could provide abodes for life under conditions that differ markedly from those on Earth. However, it can be a challenge to detect subsurface oceans directly, so inferences about candidate ocean moons are typically drawn from comparison to moons known to harbour oceans, such as Jupiter’s Europa and Saturn’s Enceladus. These moons have many similarities in terms of both the conditions that sustain their oceans and the way that their surfaces indicate the existence of an internal ocean. If the criteria were set by these moons, the small Saturnian moon Mimas would easily be ruled out as an ocean moon. It therefore comes as a surprise to learn that Mimas must have an internal ocean, according to results reported by Lainey et al.1 on page 280. Mimas is a small body whose most distinctive feature is a crater so large that it gives the moon the appearance of the Death Star space station from the Star Wars franchise. It has a slightly egg-shaped form, which is common among planetary satellites that are in synchronous rotation (that is, those that keep the same side facing the parent planet). Identifying Mimas’s stealth ocean required Lainey et al. to analyse precise measurements of changes in the moon’s orbit and rotation, which are affected by the make-up of its interior. These changes can be tracked by measuring the moon’s moments of inertia, which measure its resistance to rotational acceleration, and depend on both the moon’s surface shape and how matter is distributed inside it. Mimas’s moments of inertia were previously probed2 by looking at rocking motions, known as librations, that the moon makes as it is tugged by Saturn’s gravity. These

measurements revealed that Mimas’s librations are much larger than would be expected from the shape of its surface. This could be explained by the moon having either a very elongated rocky core, which would enhance the difference between its moments of inertia, or an internal ocean, which would allow its outer shell to oscillate independently of its core. Because there was no other widely recognized evidence for an ocean, many planetary scientists preferred the elongated-core hypothesis. But the once-neglected — and just as plausible — ocean option2 now has support

from another corner. Moments of inertia can also be used to quantify a moon’s gravity field, which acts on the parent planet and on other bodies. An oblate (slightly flattened) body such as Earth or Saturn makes the orbits of its satellites precess forwards, meaning that the ellipses traced by the orbits rotate slowly in space, in the direction of the satellites’ (much faster) orbital motion. Intriguingly, the elongated shape of a moon in synchronous rotation induces the opposite effect: the mutual orbit of the moon and the planet precesses backwards, opposite to the direction of orbital motion. By analysing measurements of the position of Mimas made by NASA’s Cassini spacecraft, Lainey et al. concluded that the moon precesses backwards in this way (Fig. 1) — a tendency that must result from the elongation of its own gravity field. The big surprise is that, if Mimas is assumed to be frozen, the moments of inertia calculated from its librations do not match those required to explain its orbital precession. In fact, Lainey et al. showed that no internal distribution of mass in a solid body can explain these two data sets. The only viable conclusion is that Mimas has a subsurface ocean. There are many implications of Mimas being an ocean world. For starters, Mimas has a large orbital eccentricity, which means that its orbit

a Backwards precession

b Proposals for

Mimas's interior

Direction of precession

Solid core

Parent planet

Direction of orbit

Internal ocean

Satellite

Figure 1 | Evidence for a subsurface ocean.­  a, The orbits of some satellites can precess backwards, meaning that the orbital path rotates slowly in a direction opposite to that of the satellite’s orbit around its parent planet. b, Lainey et al.1 detected a small amount of backwards precession in the orbit of Mimas, one of Saturn’s moons, after removing other dynamical effects. The authors showed that this precession isn’t consistent with predictions that assume that Mimas is fully solid. The authors’ finding settles a debate about whether the moon’s interior comprises a very elongated rocky core or an interior ocean. Lainey et al. showed that no internal distribution of mass in a solid body can explain the existing data, and so conclude that Mimas must have a subsurface ocean.

Nature | Vol 626 | 8 February 2024 | 263

News & views traces an ellipse rather than a perfect circle. But this eccentricity would rapidly diminish if the moon’s interior could readily respond to gravitational forces exerted on Mimas by other bodies. This indicates that the ocean or the orbital eccentricity — or even both — must have been around for only a short time, of the order of tens of millions of years. A young ocean also matches constraints derived from Mimas’s geology; in particular, the large crater, known as Herschel, could not have formed in an ice shell that is as thin as Lainey and colleagues (and others2) predict. Rather, the ice shell must have thinned by tens of kilometres since Herschel formed3. A thinning ice shell might also explain why Mimas lacks the heavy fracturing observed on ocean moons such as Europa and Enceladus4. In this way, geological features can help researchers to pin down the timing of ocean formation and the orbital conditions that stimulated the growth of an ocean. The idea that Mimas’s ocean could have formed relatively recently also has implications for other features of the Saturnian system that remain mysteries, in spite of clues retrieved by the Cassini mission. Saturn’s bright icy rings are apparently young in geological terms5, but not all scientists agree6. The heavily cratered icy moons seem ancient, but the source of the bodies that made the craters is disputed7,8, and there are suggestions that the moons themselves are also geologically young9. The clues provided by Mimas and its ocean could help to resolve some of these conundrums. Finally, adding Mimas to the catalogue of ocean worlds changes the general picture of what these moons can look like. The idea that relatively small, icy moons can harbour young oceans is inspiring, as is the possibility that transformational processes have occurred even in the most recent history of these moons. Lainey and colleagues’ findings will motivate a thorough examination of midsized icy moons throughout the Solar System. Most notably, there is a suite of mid-sized icy moons orbiting Uranus, which was selected as the highest-priority target of a NASA flagship mission by the Planetary Science and Astrobiology Decadal Survey. Mimas also has an important lesson to teach scientists: intuition is excellent for generating hypotheses, but not sufficient for drawing conclusions. The Solar System will always have surprises in store, and researchers must be open enough to new ideas and unexpected possibilities to recognize them. Matija Ćuk is at the SETI Institute, Mountain View, California 94043, USA. Alyssa Rose Rhoden is at Southwest Research Institute, Boulder, Colorado 80302, USA. e-mails: [email protected]; [email protected]

264 | Nature | Vol 626 | 8 February 2024

1. Lainey, V. et al. Nature 626, 280–282 (2024). 2. Tajeddine, R. et al. Science 346, 322–324 (2014). 3. Denton, C. A. & Rhoden, A. R. Geophys. Res. Lett. 49, e2022GL100516 (2022). 4. Rhoden, A. R. Ann. Rev. Earth Planet. Sci. 51, 367–387 (2023). 5. Iess, L. et al. Science 364, eaat2965 (2019). 6. Crida, A., Charnoz, S., Hsu, H.-W. & Dones, L. Nature Astron. 3, 967–970 (2019).

7. Ferguson, S. N., Rhoden, A. R. & Kirchoff, M. R. J. Geophys. Res. Planets 125, e2020JE006400 (2020). 8. Bottke, W. F. et al. Planet. Sci. J. 4, 168 (2023). 9. Ćuk, M., Dones, L. & Nesvorný, D. Astrophys. J. 820, 97 (2016).

The authors declare no competing interests.

Archaeology

Ancient toolmakers in north Europe identified William E. Banks

DNA analyses of skeletal fragments from a site in Germany provide evidence that humans, rather than Neanderthals, were responsible for a particular stone-tool industry called the Lincombian–Ranisian–Jerzmanowician. See p.341 Analyses of stone tools and human skeletal remains can help to determine whether specific excavated levels at archaeological sites are associated with Neanderthals or anatomically modern humans (with a body form similar to ours) during the period when both groups were present in Europe. Mylopotamitaki et al.1, on page 341, and Pederzani et al.2 and Smith et al.3, writing in Nature Ecology & Evolution, report their analyses of Ilsenhöhle, an archaeological site from this period near Ranis, Germany. The findings shed light on the environmental conditions there and identify the inhabitants linked to a widespread stone-tool industry in the region, for which the associated population was previously unknown. The arrival of anatomically modern humans in Europe has long been of interest to scientists because this migration did not take place in a previously uninhabited landscape. ­Neanderthals (Homo neanderthalensis) with their classic anatomical form occupied Europe from at least 200,000 years ago4 and were related to individuals who had a subset of Neanderthal anatomical features and were present in Europe by approximately 400,000 years ago. Homo sapiens arrived in southeastern Europe by 46,000 years ago5. Neanderthals subsequently disappeared, leaving only H. sapiens. The reason why Neanderthals were replaced by humans is the source of oftenheated debate. To understand the factors in this population replacement, termed the Middle to Upper Palaeolithic (MUP) transition, we must learn exactly when H. sapiens arrived in Europe and for how long the two populations occupied the landscape. Mylopotamitaki and colleagues used a protein-analysis technique (a proteomic method) to determine that the remains of

individuals found at Ilsenhöhle (Fig. 1) were hominins (members of the genus Homo, which includes humans and close relatives). The authors also analysed DNA from mitochondrial organelles to demonstrate that these individuals were H. sapiens, and carried out statistical analyses of radiocarbon-dating evidence to show that the remains are approximately 45,000 years old. Studies by Pederzani et al. and Smith et al. provide a window into the environmental conditions at the site during its occupation, reveal the ecology of the recovered prey species and offer insights into how the human groups that frequented Ilsenhöhle incorporated this site into their use of the region. Archaeologists have defined a number of regionally specific stone-tool technological traditions for the MUP transition, but skeletal remains in association with these industries are sparse, and most sites have contextual problems — the archaeological relationships between the few remains and the MUP transitional stone tools with which they are associated are difficult to determine. As a result, it is unclear who created these technologies, but working this out is essential to infer the cultural dynamics of populations of late Neanderthals and early modern humans in Europe and to determine whether interactions between these groups, which are known to have occurred6, might have influenced the technologies and objects used by a culture. Two archaeological levels excavated by ­Mylopotamitaki and colleagues at I­ lsenhöhle are associated with one of these MUP transitional industries, termed the Lincombian– Ranisian–Jerzmanowician (LRJ). Mylopotamitaki et al. demonstrate that the LRJ-associated levels are clearly separated

Lincombian–Ranisian– Jerzmanowician Ilsenhöhle

Bohunician and Bachokirian

Châtelperronian

Uluzzian

Figure 1 | Regions associated with distinctive stone-tool industries in Europe during the period when Homo sapiens and Neanderthals were present.  Mylopotamitaki et al.1 present evidence from the Ilsenhöhle archaeological site indicating that the stone tools there, of a type described as Lincombian– Ranisian–Jerzmanowician, were associated with H. sapiens. H. sapiens is also associated with Bachokirian stone tools5. The identity of populations associated with other types of stone-tool industry — Bohunician, Châtelperronian and Uluzzian — remains to be determined. Circled regions reflect previous evidence about Châtelperronian14, Lincombian–Ranisian–Jerzmanowician15, Uluzzian15, Bachokirian15 and Bohunician16 sites.

from the underlying level, which is linked to a type of stone-tool industry termed ­Mousterian (made by Neanderthals), and from the overlying Upper Palaeolithic levels (the period that represents the end of the last ice age and is associated with modern humans). The authors established that the LRJ levels were intact and did not contain materials from adjacent ones, which would cause problems in terms of archaeological interpretations. Mylopotamitaki and colleagues performed proteomic analyses of a number of skeletal remains from these levels. Limitations in the existing proteomic reference databases prevented more precise identification than the remains being hominin. To remedy this, the authors turned to ancient mitochondrial DNA, and the results revealed that the remains are H. sapiens. The four remains of modern humans from the excavated LRJ levels had radiocarbon dates that are ‘stratigraphically coherent’, which means that the oldest specimen comes from the lower level and the younger specimens from the upper level. These dates fall roughly between 47,000 and 44,000 years ago. This timeframe indicates that H. sapiens LRJ occupations occurred sometime between the tail end of a period of cold conditions (described as Greenland Stadial 13) and the initial stages of the Greenland Interstadial 12, a period that was less cold than the previous one but still had a mean annual temperature of less than 5 °C. The site was situated in a steppe or tundra habitat, indicating that the human groups that occupied it were adept at preying on fauna in a cold, open environment. In 2020, scientists found that modern human remains were associated with another

MUP transitional stone-tool industry, the Bachokirian, in southeastern Europe5; therefore, with Ilsenhöhle, we now know of two sites from the MUP transition with H. sapiens remains that are associated with two different transitional industries. Mylopotamitaki et al. point out that the majority of the mitochondrial-DNA genomes from Ilsenhöhle show similarities to the mitochondrial DNA extracted from the Zlatý kůň H. sapiens skull from the Czech Republic, and point out that this skull’s purported age of about 45,000 years old7 overlaps with the date range of another MUP transitional industry called the Bohunician. This echoes the conclusion, based on technological characteristics, reached by others that the LRJ has its roots in the Bohunician8. However, in the absence of human remains, as well as the fact that the techniques used to produce projectile tips in the Bohunician are reminiscent of the technique used by Neanderthals to produce a type of projectile tip called a Levallois point, the Bohunician’s association with modern humans remains to be demonstrated with certainty. As an archaeologist, I find that my colleagues and I are conducting our research during an interesting time. We have at our disposal methods that the discipline did not possess just a couple of decades ago — which enables us to examine the archaeological record at a better level of detail and improves our capacity to pinpoint when specific levels at a site were occupied. Furthermore, there is now an appreciation for the often-regional nature of cultural trajectories, and the need to examine them as such9. Another development is the recognition

that Neanderthal groups were culturally complex10–12, so we should not necessarily assume that all MUP transitional industries, because they differ from those of the preceding Middle Palaeolithic, must have been made by modern humans. Although Mylopotamitaki and c ­ olleagues’ discoveries provide another important piece of the puzzle of this culturally and demographically complex period in Europe, we must be careful not to generalize findings from one or two sites — or even one or two transitional industries — to other contemporaneous, regional transitional industries. For example, a lack of consensus13 surrounds the identity of groups (Neanderthal or modern human) associated with two transitional industries — the Châtelperronian in present-day France and northern Spain, and the Uluzzian in the Italian peninsula and immediately adjacent areas — clearly indicating the need for further objective investigation. What is the way forward? I argue that at least two things are needed. First, researchers must rigorously examine archaeological association and archaeological context — factors that are the basis on which all of our inferences are founded. Second, we must assess all available data from reliable contexts as objectively as possible, and do our best to not try to fit data into a preconceived idea of how this culturally complex period played out. In so doing, I am convinced that, in a decade’s time, we’ll have a much clearer picture of the European MUP transition. William E. Banks is in the CNRS, University of Bordeaux, Pessac 33615, France. e-mail: [email protected] 1. Mylopotamitaki, D. et al. Nature 626, 341–346 (2024). 2. Pederzani, S. et al. Nature Ecol. Evol. https://doi. org/10.1038/s41559-023-02318-z (2024). 3. Smith, G. M. et al. Nature Ecol. Evol. https://doi. org/10.1038/s41559-023-02303-6 (2024). 4. Hublin, J. J. Proc. Natl Acad. Sci. USA 106, 16022–16027 (2009). 5. Hublin, J.-J. et al. Nature 581, 299–302 (2020). 6. Hajdinjak, M. et al. Nature 592, 253–257 (2021). 7. Prüfer, K. et al. Nature Ecol. Evol. 5, 820–825 (2021). 8. Demidenko, Y. E. & Škrdla, P. J. Paleo. Archaeol. 6, 17 (2023). 9. d’Errico, F. & Banks, W. E. Curr. Anthropol. 54, S371–S387 (2013). 10. Jaubert, J. et al. Nature 534, 111–114 (2016). 11. Niekus, M. J. L. T. et al. Proc. Natl Acad. Sci. USA 116, 22081–22087 (2019). 12. Tartar, E. et al. Paleoanthropology 2022, 211−236 (2022). 13. Teyssandier, N. J. Paleo. Archaeol. 7, 1 (2024). 14. Bachellerie, F. Quelle Unité pour le Châtelperronien?: Apport de l’Analyse Taphonomique et TechnoÉconomique des Industries Lithiques de Trois Gisements Aquitains de Plein Air: Le Basté, Bidart (PyrénéesAtlantiques) et Canaule II (Dordogne). PhD thesis, Univ. Bordeaux 1 (2011). 15. Hublin, J.-J. Quat. Sci. Rev. 118, 194–210 (2015). 16. Svoboda, J. in Rethinking the Human Revolution: New Behavioural and Biological Perspectives on the Origin and Dispersal of Modern Humans (eds. Mellars, P., Boyle, K., Bar-Yosef, O. & Stringer, C.) 329–339 (McDonald Institute for Archaeological Research, 2007). The author declares no competing interests. This article was published online on 31 January 2024.

Nature | Vol 626 | 8 February 2024 | 265

News & views

on the anode during charging to form lithium metal, which then dissolves into lithium ions during discharging. The efficiency of this electrodeposition–electrodissolution cycle is used as a metric with which to evaluate lithium-metal battery systems, and is measured in model systems in which lithium is cycled

on and off a copper substrate. One might imagine that the lithium anode increases and decreases in size uniformly as lithium is cycled. However, the anode instead often transforms into a mixture of lithium grains and electrolyte-decomposition products. The decomposition products are typically electrical insulators that can surround some of the lithium grains, thus electrically isolating the grains from the rest of the electrode, and resulting in loss of battery capacity. Calendar ageing is thought to cause irreversible lithium corrosion that is detrimental to battery performance2–6. It has been suggested, however, that electrically isolated lithium grains (usually referred to simply as isolated lithium) formed during calendar ageing after charging can reconnect to the anode during subsequent battery cycles7,8. An understanding of how calendar ageing affects the performance of lithium-metal batteries is necessary because it might occur in future applications — for example, when battery-powered electronic devices and electric vehicles are not in use. Zhang et al. investigated calendar ageing and the recovery of isolated lithium in discharged systems. In their experiments, lithium was cycled on and off copper substrates in model coin cells (which are similar to the disc-shaped batteries used in watches). Control cells were cycled without resting, and were compared with cells that were cycled and then calendar-aged for 12 hours after discharge. The authors report that the average Coulombic efficiency of the calendar-aged cells — a measure of the efficiency with which charge is transferred during lithium cycling — was 98.2%. This was higher than the average Coulombic efficiency of the control cells (96.9%). The results suggest that battery

b

c

Materials science

Rested batteries can recover lost performance Laura C. Merrill

In lithium-metal batteries, grains of lithium can become electrically isolated from the anode, lowering battery performance. Experiments reveal that rest periods after battery discharge might help to solve this problem. See p.306 Lithium-metal batteries, which use metallic lithium as the anode, show great promise as the next generation of rechargeable batteries. It is generally thought that ‘calendar ageing’ of these batteries — resting them without an applied current or voltage — results in degradation that permanently reduces the amount of charge that the battery can supply, lowering performance. This loss of battery capacity is typically caused by irreversible reactions occurring between lithium metal and the battery’s electrolyte, but it can also be due to lithium metal becoming electrically isolated from the rest of the anode. On page 306, Zhang et al.1 report that electrically isolated lithium metal can reconnect to the anode after calendar ageing of the discharged battery. To make next-generation batteries smaller and lighter, they should contain active materials that have a greater energy density (energy stored per unit mass) than those used in currently available batteries. Unlike the anodes of a

Lithium anode

lithium-ion batteries, which act as host materials for the active material (the lithium ions), lithium-metal anodes are entirely composed of the active material and therefore have a higher theoretical charge-storage capacity. In lithium-metal batteries, lithium ions are deposited

“Battery capacity lost through the formation of isolated lithium during cycling can be recovered dur­ing calendar ageing.”

Rest period

Electrolyte IDPs

Charging IDP dissolution

Copper electrode Electrically isolated lithium

Figure 1 | Proposed mechanism for how discharged lithium-metal batteries recover lost charge-storage capacity.  Zhang et al.1 studied cells that model electrochemical processes in lithium-metal batteries. The cells consisted of a liquid electrolyte sandwiched between a lithium anode and a copper electrode. During cell charging, transfer of lithium ions from the anode through the electrolyte results in electrodeposition of lithium metal on the copper; during cell discharging, the reverse process causes lithium metal on the copper to dissolve. The electrolyte partly degrades during charge–discharge cycles,

266 | Nature | Vol 626 | 8 February 2024

Reconnection of isolated lithium

forming electrically insulating decomposition products (IDPs) around grains of the deposited lithium metal. a, After discharge, IDPs can surround lithium grains at the copper–electrolyte interface. Such electrically isolated lithium grains cannot take part in charging cycles, lowering cell performance. b, Zhang et al. observe that, when the cell is allowed to rest, the IDPs dissolve into the electrolyte. c, During subsequent charging, the lithium grains can reconnect to lithium deposited on the copper, and thereby take part in charging cycles once again.

capacity lost through the formation of isolated lithium during cycling can be recovered during calendar ageing. This is the opposite of what has typically been reported (irreversible lithium corrosion and loss of capacity) during calendar ageing of charged batteries2–5. Zhang and colleagues observed similar capacity recovery during calendar ageing in discharged coin cells when they tested a range of battery electrolytes and cycling conditions, both of which influence the efficiency of lithium-metal cycling. This demonstrated that the phenomenon is general. However, cycling measurements alone cannot prove that the capacity recovery was caused by the recovery of isolated lithium. The authors therefore used several other methods to investigate the capacity recovery, of which operando measurements — in which an optical microscope was used to observe lithium electrodeposition–electrodissolution cycles on a copper mesh electrode in a coin cell over time — provided most insight into the mechanism. The optical time-lapse data showed that isolated lithium formed as early as the first cycle. However, when aged after discharge, the electrically insulating products of electrolyte decomposition surrounding the isolated lithium dissolved into the electrolyte (Fig. 1). This allowed the isolated lithium to become electrically reconnected to the electrode after subsequent charging. The small coin cells used by Zhang et al. were helpful models for evaluating lithium-metal cycling, but they did not contain a battery cathode. When the authors added a cathode, they observed the same benefits of calendar ageing. It should be noted, however, that addition of a cathode creates a more complex system, the behaviour of which can be difficult to interpret. The authors also studied capacity recovery in larger pouch cells, a type of battery characterized by soft packaging. These cells contained a cathode made of lithium iron phosphate (LiFePO4, a frequently used lithium-ion cathode material) and a copper substrate. The cells lacked a lithium-metal anode, which both decreases the total mass of the active material in the cells compared with cells that contain lithium-metal anodes and increases the energy density. During the first charging step of the pouch cells, lithium metal was electrodeposited onto the copper substrate using the cathode as the lithium source. Zhang et al. observed that, as in the smaller coin cells, the pouch cells showed signs of capacity recovery after ageing in the discharged state — although further characterization was not carried out to confirm that this was due to recovery of isolated lithium. Nevertheless, the finding suggests that isolated lithium could be recovered to mitigate capacity losses in the large-format, cathode-containing batteries that are likely to be developed for practical applications.

Although the study aids our understanding of calendar ageing and capacity loss in lithium-metal batteries, gaps in our knowledge still exist. For example, the authors implemented a carefully prescribed protocol that enabled recovery of isolated lithium, but it is too soon to say whether this can be translated to consumer applications in which the number of cycles and rest time will vary. The authors also investigated pouch cells that used a low-voltage cathode material, but high-voltage materials will be needed to make batteries that have high energy densities, and this might complicate or accelerate ageing mechanisms. Furthermore, this work did not study ageing in partly charged or partly discharged cells; this should be evaluated, given that ageing might occur in such states in practical applications. Nevertheless, Zhang et al. offer a crucial perspective: calendar ageing

might not be detrimental to the performance of lithium-metal batteries, and could be used to improve it. Laura C. Merrill is in the Nanoscale Sciences Department, Sandia National Laboratories, Albuquerque, New Mexico 87185, USA. e-mail: [email protected] 1. 2. 3. 4. 5. 6. 7.

8.

Zhang, W. et al. Nature 626, 306–312 (2024). Lin, D. et al. Nature Chem. 11, 382–389 (2019). Boyle, D. T. et al. Nature Energy 6, 487–494 (2021). Kolesnikov, A. et al. Adv. Energy Mater. 10, 2000017 (2020). Gao, Y. et al. Nature Energy 5, 534–542 (2020). Wood, S. M. et al. Adv. Energy Mater. 8, 1801427 (2018). Merrill, L. C., Rosenberg, S. G., Jungjohann, K. L. & Harrison, K. L. ACS Appl. Energy Mater. 4, 7589–7598 (2021). Harrison, K. L. et al. ACS Nano 11, 11194–11205 (2017).

The author declares no competing interests.

Forum: Microbiology

The journey to understand previously unknown genes The analysis of DNA sequences sheds light on microbial biology, but it is difficult to assess the function of genes that have little or no similarity to characterized genes. Here, scientists discuss this challenge from genomic and microbial perspectives. See p.377 The topic in brief •

•

•

Some aspects of microbiology remain mysterious because of a lack of information about the identity and role of many microbial genes and proteins. The ability to obtain and analyse microbial sequences at scale and across species, including those that cannot be grown under laboratory conditions, are providing insights and data to explore. Writing in Nature, Rodríguez del Río et al.1 report their analysis of

Jakob Wirbel & Ami S. Bhatt Bringing structure and context to gene mysteries The function of most microbial genes is unknown. Some of this microbial ‘dark matter’ might encode previously unknown types of enzyme or classes of antibiotic. As ever more genes of unknown function are discovered through sequencing of DNA from mixtures of multiple genomes, termed metagenomic

149,842 bacterial genomes sampled from a variety of habitats in the wild. • The data were used to select sequences to generate a catalogue of 404,085 previously unknown gene families that could be prioritized for further study. • The investigation of these previously unknown genes could lead to new clinical tools or offer fresh perspectives about how microorganisms evolved to survive in their natural environments.

sequencing, the difficulty of experimentally characterizing these enigmatic genes has led to a focus on computationally predicting their function2. Two publications in Nature, one on page 377 by Rodríguez del Río et al.1 , and one by Pavlopoulos et al.3 published last October, tackle this challenge by cleverly leveraging advances in clustering algorithms (computational tools that group genes on the basis of similarities in amino-acid sequence) and protein-structure prediction tools 4 such as AlphaFold. Despite distinct technical approaches, Nature | Vol 626 | 8 February 2024 | 267

News & views Bacterial orders Archaeal orders

400 200 0

Red bars represent the number of previously unknown gene families per order

Height of blue bars indicates the percentage of uncultivated species per order (maximum height is 100%)

Figure 1 | Previously unknown microbial gene families. The large-scale analysis of DNA sequences captured from microbial samples as reported by Rodríguez del Río et al.1 and by Pavlopoulos et al.3 has revealed hundreds of thousands of previously unknown gene families. These data — which were gathered from microbes in the wild and across different habitats, and include species that have not been cultivated in the laboratory — provide a starting point for gaining insights into unexplored aspects of the biology of bacterial and archaeal microorganisms. Figure adapted from Fig. 3a of ref. 1.

the core strategy used by Pavlopoulos et al. and Rodríguez del Río et al. was similar. Both clustered hundreds of millions of protein sequences from metagenomic data sets into previously unknown protein families. Rodríguez del Río and colleagues filtered their data to examine genes only from prokaryotes (organisms whose cells lack a nucleus), whereas Pavlopoulos et al. used data that also included sequences from eukaryotes (organisms whose cells have a nucleus) and viruses. With these catalogues of previously unknown families at hand, both teams set out to predict the function of their newly described families, capitalizing on genomic-context analysis, which involves examining adjacent genes for clues about function, as well as harnessing breakthroughs in methods to predict protein structures. In prokaryotic genomes, genes involved in the same pathway are often present close to one other. Genomic-context analysis, which proposes ‘guilt by association’, has been used effectively to predict previously unknown antiviral defence systems used by bacteria5. The second approach, comparing predicted protein structures to find similar (homologous) proteins, is more sensitive than simply comparing amino-acid sequences alone6. Both teams predicted structures for their protein families and compared them with databases of known structures, thereby generating informed predictions about the function of some of these enigmatic proteins. The sheer scale and computational 268 | Nature | Vol 626 | 8 February 2024

investment involved in these efforts, which yielded hundreds of thousands of newly discovered protein families (Fig. 1), is impressive. Yet, the number of previously unknown genes that have a functional prediction still remains relatively small. In both publications, only around 15% of the previously unknown protein families could be annotated on the basis of structural similarity; genomic-context analysis enabled functions to be proposed for 7.4% of families in Pavlopoulos et al. and 13% in Rodríguez del Río and co-workers. In addition,

“These two studies unlock a wealth of previously hidden knowledge.” some assigned functional categories (such as ‘ribosome’) lack detailed specificity and this might obscure the precise role of these genes. Ultimately, the reliability of these predictions will have to be determined experimentally. Indeed, Rodríguez del Río et al. took the first step towards this objective by experimentally verifying the annotation for two of their predicted families. By delving deeper into the microbial dark matter, these two studies unlock a wealth of previously hidden knowledge, paving the way for future discoveries in diverse fields from medicine to biotechnology. Follow-up

experiments might include the study of protein families with completely new protein folds, possibly revealing unexplored biological functions. Similarly, synapomorphic genes — corresponding to protein families that are specific to a group of organisms sharing a common ancestor but absent in others — might hold clues to key evolutionary processes. With further refinement and validation, these computational approaches offer a powerful tool for unlocking the functional secrets of the unseen microbial world. Jakob Wirbel and Ami S. Bhatt are in the Department of Medicine, Division of Hematology, Stanford University School of Medicine, Stanford, California 94305, USA. A.S.B. is also in the Department of Genetics, Stanford University School of Medicine. e-mail: [email protected]

Alexander J. Probst Microbial sequences reveal ecology and evolution Genes are the ultimate source of all biological information on Earth, from human eye colour to the cell shape of microorganisms. The proteins they encode can be grouped using bioinformatics into families, usually with shared functionality. The ensemble of all known proteins in databases is continuously expanding as genomes are sequenced and the functions of the encoded proteins are predicted. The greatest fraction of biological functional diversity on our planet is attributed to microbial proteins. With the advent of sequencing of mixed microbial genomes from the environment (an approach that explores multiple genomes and is called metagenomics7), the increase in the rate at which data are being added to genome and protein databases is striking. However, the functional capacity of most protein families is unknown and part of the microbial dark matter. Rodríguez del Río and colleagues’ work, as well as the study by Pavlopoulos et al., analysed large-scale metagenomic data and explored the potential function and distribution of unknown protein families, which might have evolutionary and ecological importance. Rodríguez del Río analysed nearly 150,000 microbial genomes (Fig. 1), and Pavlopoulos and colleagues investigated nearly 27,000 metagenomic data sets retrieved from diverse ecosystems with various bioinformatics approaches — going well beyond the scale of public-database entries used in previous such studies8. Surprisingly, a method called rarefaction analysis used by Pavlopoulos and colleagues revealed no slowing down in the detection of previously unknown protein

families as new metagenomes were added to their analysis. Instead, the detection of protein families increased exponentially, warranting an array of follow-on studies. The distribution of protein families across Earth’s categories of ecosystem (biomes) presented by Pavlopoulos and colleagues corroborates the findings of previous investigations regarding the distribution of microbial genes8. Some biological entities, however, were particularly rich sources of newly discovered protein families, including viruses, as Pavlopoulos et al. report, and microbes called Asgardarchaeota, as presented by Rodríguez del Río and colleagues. The latter are a group of microorganisms called archaea that are closely related to the first ancestor of eukaryotes. As such, studying their proteins might reveal new insights into the evolution of the eukaryotic cell9. One major challenge in exploring the wealth of previously unknown protein families encoded in genomes of natural samples is the identification of eukaryotic genes in metagenomes. Although certain algorithms exist for the recovery of eukaryotic genomes from metagenomes, accurately predicting eukaryotic genes in mixed DNA sequences — equivalent to Pavlopoulos and colleagues’ method of identifying microbial genes — is still not possible bioinformatically. Once this shortcoming is overcome with the development of new algorithms, scientists will substantially expand the protein ‘sequence space’ and will identify protein families of unknown function that drive the ecology and evolution of eukaryotes. The greatest advance in painstakingly organizing the protein families of nearly 27,000 metagenomes and across the tree of life lies in the identification of ecosystem-specific protein clusters that differ in terms of their presence or absence, or relative abundance between varying conditions of a given ecosystem — for example, between the contexts of health or disease. Applying this strategy to examine microbial data for healthy people and those with colorectal cancer, Rodríguez del Río and colleagues found that specific unknown protein families were enriched in the gut bacteria of people with cancer. These protein families were associated with microbial motility, adhesion and invasion potentially of human tissue, as revealed through genomic-context analysis. Harnessing this approach in other fields of research should be extremely helpful for deciphering the different functions of sample sets, in the hope of identifying new targets for biochemical analyses to shed light on a tiny fraction of the microbial dark matter. Identifying differences in microbial communities (microbiomes) that might explain, for example, the disease state of a person, rely heavily on comparing which species are

present and how abundant they are (the taxonomic composition), and examining genes that are associated with certain functions. Finding specific but differentially abundant protein families of unknown function, as demonstrated by Rodríguez del Río and co-workers, has the potential not only to replace current markergene-based approaches for differentiating microbiomes but also to advance microbiome research to a new and causality-driven level. Alexander J. Probst is in the Research Center One Health Ruhr, University Alliance Ruhr, Department of Chemistry at University of

Duisburg-Essen, Essen, 45141, Germany. e-mail: [email protected]

Rodríguez del Río, A. et al. Nature 626, 377–384 (2024). Vanni, C. et al. eLife 11, e67667 (2022). Pavlopoulos, G. A. et al. Nature 622, 594–602 (2023). Jumper, J. et al. Nature 596, 583–589 (2021). Doron, S. et al. Science 359, eaar4120 (2018). Illergård, K., Ardell, D. H. & Elofsson, A. Proteins 77, 499–508 (2009). 7. Tyson, G. W. et al. Nature 428, 37–43 (2004). 8. Coelho, L. P. et al. Nature 601, 252–256 (2022). 9. Eme, L. et al. Nature 618, 992–999 (2023). 1. 2. 3. 4. 5. 6.

The authors declare no competing interests. This article was published online on 30 January 2024.

Cancer

Natural inhibitor found for cell death by ferroptosis Donna D. Zhang

The discovery that an evolutionarily conserved molecule used to make cholesterol also acts as a defence against a cell-death mechanism called ferroptosis might lead to new ways to treat cancer and other clinical conditions. See p.401 & p.411 Biology remains nothing short of astonishing, as researchers unveil the underpinnings of its myriad systems, especially those that are involved in protecting against cell death. On pages 401 and 411, respectively, Freitas et al.1 and Li et al.2 shed light on a regulated form of cell death called ferroptosis, which is driven by an iron-dependent modification of lipids in cellular membranes. The results bring into sharp focus an unexpected hero, the molecule 7-dehydrocholesterol (7-DHC). The term ferroptosis was coined in 2012 (ref. 3). This cell death encompasses a variety of processes that include lipid oxidation by the action of reactive molecules called radicals (versions of molecules that have an unpaired electron) and the fragmentation of lipids at cellular membranes, culminating in membrane disruption, shrunken mitochondrial organelles and swelling of cells (‘ballooning’). Ferroptosis occurs when there are problems in the regulation of normal iron levels (iron homeostasis) and in the oxidation of lipids. Preventing ferroptosis might be beneficial in alleviating neurodegenerative and kidney diseases, and activation of ferroptosis can kill cancer cells4–7. Freitas et al. and Li et al. report that 7-DHC, a molecule in the cholesterol-synthesis pathway (Fig. 1), acts to suppress ferroptosis. Both teams independently discovered the anti-ferroptotic role of the cholesterol-synthesis pathway. The authors reveal that several enzymes in this pathway function as potential suppressors

of ferroptosis. However, one of the enzymes, DHCR7, which catalyses the reaction that converts 7-DHC to cholesterol, was found to promote ferroptosis. This indicates that 7-DHC, produced by the enzyme SC5D and used by DHCR7, operates as a key protection against ferroptosis. Both teams then explored the mechanism of action of 7-DHC in more detail. They highlighted a key characteristic of its structure — a part that is described as a conjugated double bond in the sterol B-ring. This component of its structure enables 7-DHC to absorb radicals, thereby reducing lipid fragmentation driven by an oxidation process called peroxidation. Both teams recognized that it is mainly lipid components called phospholipids — especially if these are fragmented into smaller pieces — that can initiate ferroptosis. These findings underscore the protective function of 7-DHC, especially in scenarios in which ferroptosis might otherwise occur if this molecule wasn’t present. Most remarkably, the two teams also found that the molecule ergosterol, which is found in yeast and fungi and has structural similarity to 7-DHC, also offers protection against ferroptosis. This finding suggests that the anti-ferroptotic effect of a molecule in the cholesterol-synthesis pathway might be evolutionarily conserved, serving to safeguard a variety of organisms from ferroptosis. Nature | Vol 626 | 8 February 2024 | 269

News & views Fragmented lipid leads to cell rupture and ferroptosis

HMG-CoA

Zymosterol

Radical

Intact phospholipid

Phospholipid

EBP Lathosterol SC5D 7-DHC

Cell membrane

ROS free radical

Oxidized phospholipid

Iron

Lipid fragmentation

DHCR7 Cholesterol

7-DHC Radical absorbed by 7-DHC Lipid fragmentation suppressed

ER Mitochondrion

Figure 1 | A molecule that can help to halt cell death by ferroptosis. A hallmark of ferroptosis is lipid modification (oxidation) mediated by iron and molecules called free radicals, such as reactive oxygen species (ROS). Oxidation of lipids, particularly of those called phospholipids, in cellular membranes can kill cells. Freitas et al.1 and Li et al.2 report a previously unknown natural inhibitor of ferroptosis. This molecule, 7-dehydrocholesterol (7-DHC), is made in an organelle called the endoplasmic reticulum (ER) and found on the cell

The discovery of a natural inhibitor of ferroptosis has profound therapeutic implications. By modulating the levels of 7-DHC, there is the potential to either induce or counteract ferroptosis. Investigating the role of 7-DHC in human cancers, Freitas and colleagues observed that 9.8% of individuals studied who had Burkitt’s lymphoma had mutations in the DHCR7 gene, which encodes the corresponding enzyme. The authors note a less-frequent occurrence of such mutations in people who have a brain cancer called neuroblastoma. The deletion of DHCR7 by Li and colleagues increased 7-DHC levels, bolstering the resistance of neuroblastoma cancer cells to ferroptosis when tested in vitro. In vivo experiments in mice showed that these engineered human neuroblastoma cells evaded ferroptotic death, leading to faster tumour growth, increased tumour spread (metastasis) and decreased survival time compared with results from mice in which the transplanted cells expressed DHCR7. Li et al. present evidence of the effect of manipulating 7-DHC levels. They report that inhibiting 7-DHC synthesis by targeting an enzyme called EBP in the cholesterol-synthesis pathway with the molecule TASIN-30 induced ferroptosis. This led to the suppression of tumour growth in vivo when human cancer cells with high levels of 7-DHC were injected into mice. It is notable that this cell-death suppression occurred even in the absence of other ferroptosis inducers besides TASIN-30, underscoring the key role of 7-DHC in providing protection against cell death for certain cancer types. This protective mechanism makes these cancer cells susceptible to drugs that inhibit 7-DHC production. 270 | Nature | Vol 626 | 8 February 2024

membrane and in another type of organelle, known as mitochondria. 7-DHC is generated in the pathway (not all steps of which are shown) that converts the molecule HMG-CoA to cholesterol through a route that includes the molecules zymosterol and lathosterol and depends on enzymes such as EBP, SC5D and DHCR7. When radicals attack phospholipids, the lipid is oxidized and it fragments. 7-DHC absorbs radicals, counteracting their ability to trigger lipid oxidation and ferroptosis.

Building on their initial findings, Li and colleagues examined 7-DHC’s role in metastasis. The authors hypothesized that before spreading to distant organs, cancer cells need to endure environments that have a condition (oxidative stress) that predisposes the cells to ferroptosis. Li et al. used a combination of approaches to investigate this. These included deleting the DHCR7 gene, inhibiting DHCR7 with the drug AY9944, pretreating melanoma cancer cells with 7-DHC to elevate 7-DHC levels and using TASIN-30 to hinder 7-DHC synthesis. These approaches establish that 7-DHC

“Harnessing this new understanding with available tools holds the potential to transform clinical treatments.” can fortify melanoma cells, enhancing their survival and accelerating metastasis. Using a mouse model of tissue injury, Li and colleagues demonstrated that the accumulation of 7-DHC could be amplified by inhibiting DHCR7, achieved through pre-injury injections of DHCR7 inhibitors. This intervention effectively shielded against injury by preventing ferroptotic death of kidney cells. The discovery of 7-DHC’s role in preventing ferroptosis is a key leap forwards in this area. The therapeutic potential of the finding is strikingly evident. Given the availability of drugs already in clinical use to target DHCR7, harnessing this new understanding with available tools holds the potential to transform clinical treatments for conditions influenced by ferroptosis, if this approach

is validated in clinical testing. However, previous research8,9 indicates that excessive levels of 7-DHC and the products arising from modifications (oxidation) of this molecule might be detrimental, particularly to neuronal and retinal cells. Thoroughly examining the implications for brain and eye health will be crucial before incorporating any 7-DHCbased therapeutic strategies into standard clinical practice. Donna D. Zhang is in the Center for Inflammation Science and Systems Medicine, The Herbert Wertheim University of Florida Scripps Institute for Biomedical Innovation and Technology, Jupiter, Florida 33458, USA. e-mail: [email protected]

1. 2. 3. 4. 5. 6. 7. 8. 9.

Freitas, F. P. et al. Nature 626, 401–410 (2024). Li, Y. et al. Nature 626, 411–418 (2024). Dixon, S. J. et al. Cell 149, 1060–1072 (2012). Stockwell, B. R. et al. Cell 171, 273–285 (2017). Galy, B., Conrad, M. & Muckenthaler, M. Nature Rev. Mol. Cell Biol. 25, 133–155 (2024). Anandhan, A. et al. Sci. Adv. 9, eade9585 (2023). Conrad, M., Angeli, J. P. F., Vandenabeele, P. & Stockwell, B. R. Nature Rev. Drug Discov. 15, 348–366 (2016). Xu, L. et al. Neurobiol. Dis. 45, 923–929 (2012). Pfeffer, B. A., Xu, L., Porter, N. A., Rao, S. R. & Fliesler, S. J. Exp. Eye Res. 145, 297–316 (2016).

The author declares no competing interests. This article was published online on 31 January 2024.

Review

A break in mitochondrial endosymbiosis as a basis for inflammatory diseases https://doi.org/10.1038/s41586-023-06866-z

Michael P. Murphy1,2 ✉ & Luke A. J. O’Neill3 ✉

Received: 23 July 2023 Accepted: 14 November 2023 Published online: 7 February 2024 Check for updates

Mitochondria retain bacterial traits due to their endosymbiotic origin, but host cells do not recognize them as foreign because the organelles are sequestered. However, the regulated release of mitochondrial factors into the cytosol can trigger cell death, innate immunity and inflammation. This selective breakdown in the 2-billion-year-old endosymbiotic relationship enables mitochondria to act as intracellular signalling hubs. Mitochondrial signals include proteins, nucleic acids, phospholipids, metabolites and reactive oxygen species, which have many modes of release from mitochondria, and of decoding in the cytosol and nucleus. Because these mitochondrial signals probably contribute to the homeostatic role of inflammation, dysregulation of these processes may lead to autoimmune and inflammatory diseases. A potential reason for the increased incidence of these diseases may be changes in mitochondrial function and signalling in response to such recent phenomena as obesity, dietary changes and other environmental factors. Focusing on the mixed heritage of mitochondria therefore leads to predictions for future insights, research paths and therapeutic opportunities. Thus, whereas mitochondria can be considered ‘the enemy within’ the cell, evolution has used this strained relationship in intriguing ways, with increasing evidence pointing to the recent failure of endosymbiosis being critical for the pathogenesis of inflammatory diseases.

It is now well accepted that mitochondria play many central roles within the cell, far beyond their essential energetic functions in oxidative phosphorylation, the Krebs cycle and fatty acid oxidation1–4. These non-canonical activities include signalling, biosynthesis and the regulation of cell fate3. Particularly intriguing is a deluge of findings into how mitochondria act in the immune and cell death signalling pathways that enable cells to respond to infection or damage5,6. A recent analysis showed that, since 2011, publications on mitochondria began to eclipse those on other organelles3, reflecting the upsurge in interest in mitochondria beyond bioenergetics. This raises the question of why so many pathways that react to the challenge of infection or tissue injury use mitochondria as a central signalling hub to integrate and transduce the cell’s response. One probable factor is that the endosymbiotic origin of mitochondria marks them apart from the rest of the cell in a way that can be co-opted to produce key messages pertaining to cell fate2,7–10. Here we discuss how the evolutionary origin of mitochondria has been used by cells to facilitate their response to injury and infection. In addition to providing an explanation for the critical role of mitochondria in cell fate, this hypothesis raises intriguing new research questions. One in particular that we raise here is whether the increase over the past few generations in inflammatory diseases, including systemic lupus erythematosus, multiple sclerosis, rheumatoid arthritis and inflammatory bowel disease11–16, could be due to a break in the endosymbiotic relationship between mitochondria and the cell, driven by such relatively recent phenomena as the rise in obesity, chronic stress, dietary changes towards processed food,

social changes such as lack of sleep, and synthetic chemicals in the environment. Might these insights provide new therapeutic options to treat chronic inflammatory diseases, many of which remain difficult to address and present a significant burden to humanity, contributing to over 50% of all deaths17?

Implications of endosymbiotic origin Although the details are still vigorously debated18–20, the broad outlines of the endosymbiotic origin of mitochondria and eukaryotic cells are now well accepted21. Despite the first suggestion of a bacterial ancestry for mitochondria late in the 19th century by Portier and Wallin18,19, this idea did not gain widespread acceptance until it was compellingly revived in 1967 by Lynn Margulis22. The consensus is that eukaryotic cells arose as a single event about 2 billion years ago when an Asgard archaeon entered into an endosymbiotic relationship with an α-proteobacterium19,21,23–26. Both cells must have benefitted from this arrangement, although the mechanical basis of these advantages is still disputed24,27,28. Over time, most α-proteobacterial DNA either relocated to the nascent nucleus or was eliminated, with its functions replaced by the host’s genome, leaving a small mitochondrial DNA (mtDNA) molecule, 16.5 kb in mammals, that encodes only 37 genes required for the assembly of the oxidative phosphorylation machinery29–31. The remaining 1,200 or so types of protein present within mammalian mitochondria are translated in the cytosol for subsequent import into the mitochondrion32,33.

1 MRC Mitochondrial Biology Unit, University of Cambridge, Cambridge, UK. 2Department of Medicine, University of Cambridge, Cambridge, UK. 3School of Biochemistry and Immunology, Trinity Biomedical Sciences Institute, Trinity College Dublin, Dublin, Ireland. ✉e-mail: [email protected]; [email protected]

Nature | Vol 626 | 8 February 2024 | 271

Review Box 1

Remnants of the endosymbiotic origin of mitochondria Mitochondria Bacteria/viruses Eukaryotic cell Unmethylated DNA

Y

Y

N

N-formylated peptides

Y

Y

N

ROS production

Y

Y

?

Cardiolipin

Y

Y

N

dsRNA

Y

Y

Y (but less)

Mitochondria within eukaryotic cells retain many bacterial remnants that are more typical of bacteria: mtDNA is present in many copies and is not methylated at CpG, in contrast to much of the nuclear DNA; it is more susceptible to oxidative damage; it has a slightly different genetic code; the 13 polypeptides translated within mitochondria are all initiated with an N-formylmethionine; ROS levels are elevated and can be driven by high membrane potential; and there is bidirectional transcription of mitochondrial genes, hence overlapping of messenger RNA with the potential for dsRNA.

Intriguingly, mitochondria retain a number of bacterial traits: metabolic pathways, unmethylated DNA, double-stranded (ds) RNA, N-formyl peptides, elevated reactive oxygen species (ROS) production and the phospholipid cardiolipin (CL) (Box 1). Importantly, these bacterial aspects of mitochondria are all associated with the matrix, inner membrane and intermembrane space (IMS)25,34. These compartments are surrounded by the mitochondrial outer membrane, which probably arose from both the outer membrane of the α-proteobacterium and the endocytic plasma membrane of the Archaeon host25,34. This mixed origin is indicated by the mitochondrial outer membrane containing β-barrel proteins that evolved from bacterial outer membrane precursors35, whereas the phospholipid composition of the mitochondrial outer membrane is similar to that of other cell membranes but distinct from the mitochondrial inner membrane25,34. Thus, although mitochondria are clearly now fully integrated into the eukaryotic cell, we can also consider mitochondria as a pseudobacterium ‘bricked in’ behind the mitochondrial outer membrane, which allows exchange of metabolites but—under most conditions—retains macromolecules. Any breach in the protective barrier provided by the mitochondrial outer membrane will allow entry of these pseudobacterial factors to the cytosol, which can potentially be recognized as ‘foreign’9, activating immune signalling and cell death pathways. Therefore, evolution has not erased the bacterial origins of mitochondria but has retained many of these traits because of their usefulness to the cell.

Mitochondrial signals to the cytosol This retained ‘otherness’ of mitochondria shows itself to the rest of the cell by the release of molecules into the cytosol by a number of mechanisms (Fig. 1) that activate multiple immune signalling or programmed cell death pathways (Fig. 2). Proteins derived from the mitochondrial IMS, such as cytochrome c, SMAC/Diablo and Omi, are released following mitochondrial outer membrane permeabilization (MOMP) to induce the intrinsic apoptosis pathway in response to cell stress36–38. Formation of the MOMP pore is still incompletely characterized, but occurs in response to the interaction of proapoptotic members of the Bcl-2 family proteins 272 | Nature | Vol 626 | 8 February 2024

such as Bax and Bak38–40. MOMP exposes the IMS to the cytosol without alteration of the permeability of the inner membrane, allowing proteins to exit39,40. Many IMS proteins such as cytochrome c are largely present within cristae and are therefore not in equilibrium with the space between the mitochondrial inner boundary membrane and outer membrane; consequently there is extensive further remodelling of the cristae by OPA-1 in concert with MOMP to enhance release of these proteins38,41,42. The peroxidation of unsaturated fatty acids of CL in the inner membrane is also proposed to enhance cytochrome c release43. Access and egress from the cristae to the space between the mitochondrial inner boundary and outer membranes is also modulated by the action of the mitochondrial contact site and cristae-organizing system complex44,45. Because the mitochondrial inner membrane is far greater in surface area than the outer membrane, swelling of the mitochondrial matrix can rupture the outer membrane with subsequent release of IMS proteins10,46,47. One mechanism of matrix swelling is by induction of the mitochondrial permeability transition pore (mPTP), which forms in the inner membrane, rendering it permeable to molecules up to roughly 1.5 kDa (refs. 46,48). Although the composition of the mPTP is disputed48 it is induced by a wide range of factors, notably by oxidative stress and mitochondrial calcium overload, and hence is often associated with necrotic cell death46. However, there are many other redox and stress modifiers of its function, including mitochondrial membrane potential, whereas the pore itself has a number of different permeability states whose (patho)physiological significance remains unclear48. Whether induction of the mPTP can be modulated for selective release of proteins from the intermembrane space, or whether it leads to rupture of the inner membrane and release of matrix macromolecules, is not certain. MtDNA is distinct from that in the nucleus, resembling bacterial DNA owing to its lack of CpG methylation49. Therefore, if mtDNA appears in the cytosol it will be detected by sensors of bacterial DNA such as Toll-like receptor 9 (TLR9) and cGAS, which generates the second messenger cGAMP50. These sensors will lead to the induction of multiple proinflammatory genes, notably cytokines including type I interferons (IFNs). There are a number of potential mechanisms by which mtDNA can be released from the mitochondrial matrix51. One is herniation of the mitochondrial inner membrane, which occurs as the mitochondrial matrix swells, pushing sections of the inner membrane encapsulating mtDNA and its binding proteins through Bax/ Bak pores in the mitochondrial outer membrane52. Furthermore, an unbiased screen has shown a potential role for cristae remodelling by OPA-1 in mtDNA release41. The released mtDNA seems to be intact and remains encapsulated within vesicles, which can then conjugate to endosomes enabling recognition of mtDNA by endosomal TLR9. There is also the direct release of fragmented mtDNA from the matrix into the cytosol, which seems to be associated with induction of the mPTP and oligomerization of the pore-forming voltage-dependent anion channel (VDAC) in the mitochondrial outer membrane53. Finally, there is also the release of newly synthesized and oxidized mtDNA molecules, which have been shown to interact with the NLRP3 inflammasome54. Activation of NLRP3 leads to the stimulation of caspase-1, which processes the precursors of inflammatory cytokines IL-1β and IL18, and the gasdermin family of proteins, to promote pyroptotic cell death55. Regulation of NLRP3 by mitochondria, however, remains complex, because it has also been shown that mitochondrial ATP is required for NLRP3 activation, generating phosphocreatine which, in turn, leads to cytosolic ATP production via creatine kinase B, the ATP being required for NLRP3 inflammasome assembly via direct binding to the NACHT domain of NLRP3 (ref. 56). In all these cases, considerable uncertainty remains regarding whether the released mtDNA is intact, fragmented or oxidized, or indeed whether the process for extrusion of the mtDNA is selective, occurs directly into the cytosol or via vesicles that can be further processed—or even extruded from the cell51.

a Mitochondrion

CytC

b

IMS MOMP Outer membrane

IMS mPTP

Inner membrane

Oxidative stressb + calcium

d

c

e

Metabolites

Vesicle Mitochondrial DNA CL translocation

Fragmentation oxidation

?

Carrier protein

b

f

g

2H+

O2•–

O2•–

e–

Δ Eh

IMS 4H+

QH2

Δp

Q

Complex I

4H+

Inner membrane

Q QH2

Complex III

2e– 4H+

Fig. 1 | Pathways of molecular signal release from mitochondria. Signals can be released from mitochondria by a number of mechanisms. a, Cytochrome c (CytC) can be released from the mitochondrial IMS by MOMP, a pore in the mitochondrial outer membrane formed by Bcl-2 family proteins. These proteins activate apoptosis. b, Induction of the mPTP by oxidative stress and elevated calcium leads to a pore in the inner membrane rendering it permeable to molecules of up to around 1.5 kDa, but can also lead to swelling of the mitochondrial inner membrane, potentially rupturing the outer membrane and, in some cases, the inner membrane. c, mtDNA and dsRNA can be released by incompletely characterized pathways. One involves oxidation of newly formed mtDNA fragments via the mPTP, and another is herniation of mitochondrial inner membrane vesicles containing mtDNA, possibly through MOMP-like structures. The mtDNA-containing vesicles are processed by the endosomal pathway to expose mtDNA to TLR4, although other potential fates of these vesicles remain uncertain. d, CL is normally within the mitochondrial inner

membrane but can be transferred to the outer membrane at contact sites. In addition, oxidation of polyunsaturated fatty acids in CL to oxCL and/or lysoCL can affect release of cytochrome c, and can also act as a marker for mitophagy. e, Polar metabolites are readily exchanged across the mitochondrial inner membrane by a series of carrier proteins and can thus signal to the cytosol, including itaconate, succinate and citrate. f,g, Mitochondrial respiratory chain complexes I and III can produce superoxide. Superoxide production by complex I can be by RET (f) whereas that by complex III is derived from a ubisemiquinone radical at the Qo site (g). Superoxide formation is greatly enhanced by a high protonmotive force. Superoxide can act on aconitase to release Fe and form hydrogen peroxide, or can dismutate to hydrogen peroxide by the action of manganese-dependent superoxide dismutase; once formed, hydrogen peroxide can act as a redox signal by either acting on thiols or inducing lipid peroxidation on CL. Eh, reduction potential.

The release of mtRNA into the cytosol also occurs in a range of scenarios57. Mammalian mtDNA encodes two ribosomal RNAs, 22 transfer RNAs and 11 mRNAs, two of which are bicistronic, thereby encoding 13 mitochondrial polypeptides. Because mtDNA encodes genes on both strands that overlap, this can lead to dsRNA molecules58 which, following release, can activate the viral sensing pathways through RIG-I-like receptors, including RIG-I itself and MDA-5, that interact with MAVS, which perhaps conveniently localizes to the surface of the mitochondrial outer membrane59. The mechanism(s) of dsRNA release are less certain but may be similar to those for mtDNA release.

The 13 mitochondrially translated polypeptides contain an N-terminal, N-formylated methionine (fMet) residue that originates from the N-formylation of a proportion of the methionine-charged tRNA used to initiate translation on mitochondrial ribosomes60. N-formylated Met is retained in 12 out of 13 of the mitochondrially encoded proteins in mammals, the one exception being the Cox III subunit of cytochrome oxidase61. Bacterial proteins also contain an N-terminal fMet and consequently mammalian cells contain a series of N-formyl peptide G-protein-coupled receptors on cell membranes that act as chemotaxis receptors to respond to bacterial fMet peptides62. Nature | Vol 626 | 8 February 2024 | 273

Review Obesity

Processed foodstuffs

Lack of exercise

Disrupted circadian rhythm

Environmental pollutants

Break in endosymbiosis

Mitochondria

Cytochrome c release

H2O2 production

Metabolite release

mtRNA release

mtDNA release

Apoptosis

Oxidative damage, antiparasitic and antibacterial response

Succinate Fumarate

MDA-5/RIG-I

cGAS–STING

Cell death

Phosphocreatine ATP NLRP3 activation

SUCNR1

Cytokines and IFNs

Pyroptosis, IL-1β, IL-18

Inflammation Rheumatoid arthritis Inflammatory bowel disease

Multiple sclerosis Asthma

Fig. 2 | How breakdown in endosymbiosis can lead to inflammation. A large number of factors are implicated, including obesity and the impact of environmental pollutants on mitochondria, disrupting their integrity and driving the release of a range of factors that, via specific sensors, drive inflammation, notable examples being NLRP3 and nucleic acid sensors.

Metabolites derived from mitochondria can also provoke inflammation and one, fumarate, when disrupted can drive release of mitochondrial dsRNA which, in turn, will drive type I IFN production. Could an increase in these various provoking factors be a reason for the rise in incidence of inflammation and autoimmune diseases?

In addition, a number of N-formyl peptides that originate from the N termini of mitochondrially translated proteins are released during various forms of inflammation63 and seem to act via N-formyl peptide G-protein-coupled receptors64. How these peptides are processed and released from mitochondria is unclear, but is yet another example of the bacterial origin of potential signals generated by mitochondria. Cardiolipin has a structure slightly different from that of other phospholipids, with two phosphatidic acids linked by a third glycerol molecule26,65. CL is widespread in bacteria but in eukaryotes is found primarily within the mitochondrial inner membrane, where it is synthesized and its fatty acid composition remodelled66,67. In both bacterial and mitochondrial membranes, the properties of CL are used to adjust membrane curvature and are also closely associated with, and essential for, the function of many membrane proteins, indicating the reason why it has been retained by mitochondria65. In a number of situations CL is translocated to the mitochondrial outer membrane, probably at contact sites between the mitochondrial inner and outer membranes67. The translocation of CL is often associated with peroxidation of the unsaturated fatty acids that predominate in CL and, once exposed on the mitochondrial outer membrane, peroxidized CL may contribute to the formation of the pore that enables MOMP67. Furthermore, exposure of CL on the cytosolic-facing surface of the mitochondrial outer membrane can facilitate the formation of protein-signalling assemblies, potentially including the recruitment and activation of NLRP3 (ref. 68). Furthermore, exposure of CL on the mitochondrial outer membrane can also mark the organelle for mitophagy67, which may have been retained since the origins of the endosymbiosis of the protomitochondrial α-proteobacterium26,28. Interestingly, Polly Matzinger, originator of the Danger Hypothesis that theorizes that the immune system recognizes various threats to become activated, considered CL recognition to be an excellent example of a ‘danger’ signal69. Current

questions include how the peroxidation of unsaturated fatty acids on CL is regulated and how the remodelling and translocation of CL to the outer membrane are coordinated. The mitochondrial respiratory chain can produce superoxide from complexes I and III70,71. The production of superoxide by complex I by reverse electron transport (RET), under conditions of high protonmotive force (Δp) and a reduced coenzyme Q pool, is of particular interest because it can be regulated physiologically in response to changes in these key mitochondrial parameters72,73 and is emerging as a potential mode of mitochondrial redox signalling74,75. Thus there is considerable interest in this process underlying a redox signal. For example, RET could be activated by turning off the consumption of Δp through ATP synthesis by FoF1-ATP synthase or by enhancing glutaminolysis as a source of succinate to drive RET76. The expectation is that superoxide produced within the mitochondrial matrix by either complex would be converted to H2O2 by the high local concentration of Mn-superoxide dismutase70. H2O2 can act as a potential redox signal, most probably by reversibly oxidizing cysteine residues on key signalling relay proteins77,78, such as those seen in the redox-sensitive OxyR transcription factor in bacteria79 and in similar redox-sensitive cysteines that are widespread in eukaryotic proteins77. Even so, there are uncertainties about the circumstances under which H2O2 can diffuse from mitochondria within cells, or whether elevated H2O2 may act within mitochondria and thereby contribute to the release of other factors into the cytosol—for example, by activation of the mPTP80. Similarly, peroxidation of CL is a further potential mode of signalling but whether CL is modulated directly by mitochondrial ROS is unclear. In some circumstances there are suggestions that elevated mitochondrial ROS can act as a means of killing invading bacteria within cells81. The sequestration of much of central metabolism within the mitochondrial matrix, coupled with the selective transport of polar

274 | Nature | Vol 626 | 8 February 2024

metabolites including α-ketoglutarate, citrate, itaconate and succinate across the mitochondrial inner membrane, enables the release of mitochondrial metabolites to act as signals6,82–84. In all cases, a signalling modality requires the regulated production and export of the metabolite to the cytosol/nucleus followed by a mode of response to this signal. Elevated generation of citrate and its export to the cytosol produces large amounts of acetylCoA, some of which drives fatty acid biosynthesis but which also alters levels of histone acetylation85. During inflammation, succinate production is enhanced by increased glutaminolysis in conjunction with reduction of the CoQ pool, which leads to succinate export via the dicarboxylate carrier to the cytosol75. Once in the cytosol, succinate can affect the activity of α-ketoglutarate-dependent dioxygenases, which act to introduce an oxygen atom into protein targets. This is important in the activation of hypoxia-inducible factor 1α, leading to the induction of cytokines such as interleukin (IL)-1β (refs. 76,86). Mitochondrial metabolites such as succinate and α-ketoglutarate can also alter DNA and histone methylation as potential epigenetic signals87–89. Similarly, following inflammatory activation of macrophages there is upregulation of aconitate decarboxylase-1 (encoded by the gene IRG-1) that produces itaconate from aconitate that is then exported from the matrix to the cytosol90. Itaconate is a metabolite now gaining intense interest, and has been shown to have antibacterial effects but also to modulate various inflammatory proteins, leading to a net anti-inflammatory effect91,92. In addition to individual metabolites signalling from mitochondria, there is also the well-established metabolic shift known as the Warburg effect93. The rationale for the Warburg effect is generally thought to be that it enables increased flux through glycolysis and the pentose phosphate pathway and hence greater production of metabolic building blocks for cell growth94. The far greater flux through glycolysis during immune cell activation and in cancer cells often leads to sufficient ATP production so that oxidative phosphorylation is no longer required. However, it is important to note that ATP production by oxidative phosphorylation is just one of many roles carried out by mitochondria. Many of these—notably the TCA cycle and the respiratory chain—are essential for biosynthesis and thus are still required for cells undergoing a shift to glycolytic metabolism95–98, even if mitochondria are no longer the main suppliers of ATP. An underappreciated aspect of this shift from ATP supply by oxidative phosphorylation to glycolysis is that it can also be brought about by blocking the ability of mitochondrial oxidative phosphorylation to supply ATP to the cytosol, thus forcing the cell to switch to glycolytic ATP production. In summary, the shift to aerobic glycolysis is not an indicator of a loss of mitochondrial function because most tumours and inflammatory conditions require active mitochondria with a functioning respiratory chain and Krebs cycle that can generate oncometabolites or intermediates for biosynthesis95. Thus, the sequestration of mitochondria from the rest of the cell in conjunction with their otherness enables the use of mitochondria as signalling hubs for multiple processes, notably in cell death and immunity. Nevertheless, this raises many questions and leaves numerous loose ends. Among these are the following: were there any antecedents to these pathways in bacteria that became mitochondria? And were there precursors of the receptors now active in the eukaryotic cells present in archaea, or did these pathways evolve after endosymbiosis?

Inheritance or exaptation? Sequestered mitochondrial compartments within eukaryotic cells contain components of bacterial origin and therefore probably facilitated the evolution of signalling pathways centred on the organelle26,28. Two complementary scenarios for this are probable. One is that mitochondrial signalling pathways developed from those already used before endosymbiosis. These could be from the protomitochondrial α-proteobacteria that were transferred to the host genome, or from the

archaeal host. The alternative is that endosymbiosis enabled many mitochondrial components to be repurposed for new functions that arose to take advantage of the new situation following endosymbiosis—a process known as exaptation99—defined as features acquiring new functions for which they were not originally selected. For example, the role of Bax/Bak in conjunction with OPA-1 in mtDNA release fits with the exaptation model, whereby that function might have evolved after endosymbiosis. Most probably, a combination of both processes contributes. Considering bacterial antecedents for mitochondrial RET, it is important to note that complex I from the α-proteobacterium Paracoccus denitrificans, which is often used as a model for the protomitochondrial endosymbiont25,100, can readily undergo RET101,102. Although the role of RET in P. denitrificans is unclear, it indicates that RET was probably available as a potential mode of signalling in the early prokaryotic cell. Bacteria such as P. denitrificans can elevate ROS production in response to stressors such as DNA damage leading to cell death103, and hence it is plausible that mitochondrial ROS production and lysis in the eukaryotic cell are derived from bacterial antecedents. The release of factors from mitochondria into the cytosol recapitulates a mechanism by which bacteria protect against phage infection by lysis of infected cells before the invader has an opportunity to fully replicate8,104. Furthermore, bacteria can release vesicles encapsulating intracellular components105. However, the mechanistic parallels between these bacterial processes and those of MOMP, vesicle release and mPTP induction in mitochondria are not fully clear. Furthermore, key components of eukaryotic cell-autonomous innate immune responses have evolved from bacterial precursors of the cGAS–STING pathway, gasdermins and Toll-IL-1 receptor-resistance (TIR) domain-containing proteins (such as occur in TLRs in mammals), that all protected bacteria against phage infection8. It is therefore possible that at least some of these pathways existed in the archaeon that originally hosted the α-proteobacterium, and were then co-opted into sensing the release of factors from the endosymbiont should the need arise106–108. In that original endosymbiont, nuclear gene expression might have been initiated by these systems to ensure homeostasis should the endosymbiont become damaged and show itself. Mitophagy could clear the damaged endosymbiont, perhaps by recognizing exposed CL26, although quite how mitophagy evolved in still uncertain109. In multicellular organisms, which took a further 1.4 billion years (approximately) to evolve, these same pathways, along with additional ones, might therefore provoke local or systemic inflammation to restore homeostasis following tissue injury or infection. What therefore began as the capacity of a single cell to survive in response to mitochondrial damage became an organismal inflammatory process, in both cases, the protective events being triggered by the recognition of components from the damaged mitochondria.

Sterile and non-sterile inflammation overlap Pattern recognition receptors (PRRs) respond to both sterile inflammation via danger-associated molecule patterns (DAMPS) and infection via pathogen-associated molecular patterns (PAMPs), with considerable overlap in those PRRs that respond to endogenous DAMPS from mitochondria with those that respond to external PAMPs from bacterial or viral infection9,110. This raises the question as to which came first. Did the innate immune system arise in response to cell damage that led to the release of DAMPS, and then subsequently respond to PAMPs, or was it originally designed to address infection and was then co-opted to respond to mitochondrial damage? Evidence is emerging of an interesting interplay between PAMP and DAMP sensing. RNA-containing viruses, such as the influenza virus, dengue virus or norovirus, can be detected by RNA sensors such as RIG-I that then promote the release of mtDNA, which amplifies the induction of type I IFNs via cGAS– STING111–113. DNA viruses such as the Herpes virus also drive the release of mtDNA to promote antiviral innate immunity114. It therefore appears Nature | Vol 626 | 8 February 2024 | 275

Review with systemic lupus erythematosus, a known interferonopathy, again potentially linking mitochondrial dsRNA to autoimmunity. Fumarate hydratase-deficient tumours have also been shown to release their mtDNA and promote type I IFN production, which further supports a role for fumarate hydratase in mitochondrial nucleic acid release123,124.

RNA viruses (for example, influenza, dengue, norovirus) RNA

PAMP

Emerging aspects

RIG-I RNA

MDA-5

Mitochondria RNA

RIG-I

Type I interferon

DAMP DNA

Antiviral

cGAS TLR9

Optimal response

Fig. 3 | Did nucleic acid-sensing PRRs evolve to sense mitochondrial nucleic acids? mtDNA has been shown to be sensed by the PRRs cGAS and TLR9 while mitochondrial dsRNA can be sensed by the PRRs MDA-5 and RIG-1. These PRRs were, however, discovered as sensors of microbially derived nucleic acids, so which came first? Recently it has been shown that RNA viruses, including influenza, dengue and norovirus, are sensed by RIG-I but also provoke the release of mitochondrial nucleic acids that are sensed in turn by the relevant PRR. This PAMP–DAMP combination would appear to be necessary for an optimal response to these viruses. Ultimately, might this apply to the sensing of all microbes whereby the response is initiated by the sensing of the microbial PAMP which, in turn, drives release of the mitochondrial DAMP for an optimum response?

that it is a PAMP–DAMP combination that is required for an optimal response, the DAMP originating in the mitochondria in the form of mtDNA (Fig. 3).

Endosymbiosis and autoimmunity The similarity in PRRs activated by DAMPs from mitochondria and PAMPs from infection raises the question of crosstalk between DAMP/ PAMP sensing contributing to autoimmunity. In this context it is important to note that in many autoimmune diseases antibodies against mitochondrial antigens are detected, including those against CL115, mitochondrial proteins116, mtDNA117 and mtRNA118. Furthermore, the overlap between responses to mitochondrial and bacterial DNA and mitochondrial and viral dsRNA creates another way in which an autoimmune response could arise. In addition there are situations where succinate arises from mitochondrial metabolism75,119, which may contribute to inflammatory signalling86. Supporting this, succinate is found in inflamed joints and promotes IL-1β production via the succinate receptor SUCNR1 (ref. 120). A recent study has also shown the prominence of dsRNA in multiple autoimmune and inflammatory diseases121. Strikingly, the authors report on how aberrant editing of dsRNA by adenosine deaminase acting on RNA, which converts adenosine to inosine, rendering dsRNA more like ‘self’ RNA, generates immunogenic dsRNA that is sensed by MDA-5, promoting inflammation. Because a large proportion of dsRNA in eukaryotic cells is likely to be of mitochondrial origin58, this suggests that aberrant sensing of mitochondrial dsRNA might be pathologic in a number of autoimmune conditions. Finally, a link between fumarate hydratase and the release of mitochondrial dsRNA has been made122. Activation of TLR4 with the Gram-negative bacterial product lipopolysaccharide leads to a decrease in fumarate hydratase in macrophages, which is linked to dsRNA release via an unknown mechanism. The dsRNA then drives type I IFN production via MDA-5 and RIG-I. Fumarate hydratase was also reported to be repressed in monocytes from patients 276 | Nature | Vol 626 | 8 February 2024

The insights gained from considering the bacterial origin of mitochondria suggest broadening our outlook to consider also whether other bacterial pathways might be used by mitochondria in eukaryotic cells. For example, bacteria use quorum sensing by which they self-produce extracellular chemical signals, which can accumulate in a local environment to levels that are required to activate transcription of specific genes125. Mitochondria regularly release metabolic signals such as succinate, citrate and itaconate82 and can also release ROS and other signals to modulate the action of adaptor proteins such as Miro that regulate the ways in which mitochondria interact with, and are moved by, the cytoskeleton126,127. Currently the focus is on the reading of these signals in the cytosol and nucleus, but these messengers can also be readily taken up by mitochondria in the same cell and indeed pass from cell to cell through plasma membrane transporters, or act on cell surface receptors75. Related to this is the generation of mitokines such as FGF21 and GDF15, which are generated in response to mitochondrial stress to signal to other tissues128, but much still needs to be resolved about whether these are mitochondrial-selective or general stress responses. Other potential modes of signalling worth exploring include the generation of lipid peroxidation signals from mitochondria, potentially from the peroxidation of unsaturated fatty acids on CL67, or of ROS-induced ROS release129. Whether these signals can affect local mitochondrial transcription/translation has not been explored but may be a fertile area, particularly in cells with mitochondria at a distance from the nucleus, such as in neurons. The chemotactic mobility of bacteria in response to signals is well established. Mitochondria readily move, divide and recombine throughout the cell and this process is also linked to mitophagy130–132. Mitochondrial dynamics involves the close interplay between mitochondria, cytoskeleton and other organelles, notably the endothelial reticulum. In addition, mitochondrial location within the cell can respond dynamically to a number of stimuli, moving close to the plasma membrane under conditions of ATP demand or in response to infection130–132. The current assumption is that mitochondria are moved around the cell by the cytoskeleton, but there may well be local signals generated by, and acting on, mitochondria intracellularly that contribute to these processes. Intriguingly, it is now clear that intact mitochondria can migrate from cell to cell within the body, including from one cell type to another133–137. This can occur through either tubular connections, transfer of vesicles or cell-to-cell connections such as gap junction channels, or by cell fusion135,136. It may also be the case that damaged mitochondria can trigger the activation of mitochondrial transfer from either healthy cells or stem cells135. Related to this, there is considerable interest in mitochondrial transfer/transplantation as a therapeutic strategy—for example, following ischaemia-reperfusion injury to the heart138,139. External mitochondria are readily endocytosed within cells, and the tacit assumption has been that the addition of new mitochondria enhances oxidative phosphorylation in target cells, although this is seldom demonstrated explicitly and many other interpretations of the positive effects are possible. This ability of mitochondria to migrate from cell to cell is of course a further reflection of their endosymbiotic origin.

Outlook A key question is: when does endosymbiosis break down? Because inflammation is a normal physiological process designed to restore

us to health following injury and infection, why does this process go rogue in so many diseases? The rise in incidence of these diseases has led to various culprits being implicated, including obesity, chronic stress, poor diet (potentially with a contribution from processed foods) and lifestyle changes such as disrupted sleep or environmental toxins17, most probably in the context of specific genetic backgrounds. Mitochondria provide us with a unifying target for all of these processes1,140–142 (Fig. 2). Might the pressure on mitochondria caused by obesity, chronic stress, certain foodstuffs, disrupted circadian rhythms or damage to mitochondria caused by various environmental toxins exacerbate a process whose homeostatic goal is to drive inflammation but in a beneficial way? All of these triggers could well give rise to enhanced release of the mitochondrial factors discussed here, be they ROS, metabolites, peptides or nucleic acids, with hyperactivation of the sensors designed to trigger beneficial inflammation leading to inflammatory diseases. It is intriguing to consider what the early pioneers of mitochondrial research, such as Hans Krebs or Peter Mitchell, might have made of the exciting developments discussed here. The theory of endosymbiosis provided us with a huge insight into the evolution of eukaryotic life; without this unlikely random event, we as a species would not have evolved to consider it. That the Krebs cycle is repurposed for inflammation, or that mitochondrial nucleic acids might be key triggers for immunity via induction of cytokines, gives us a whole new perspective on what mitochondria are doing in our cells. Might these new insights provide an explanation for the increased incidence in inflammatory and autoimmune diseases? And might these diseases be explained in part by a fracture in a 2-billion-year relationship? This insight could well give rise to new therapies to treat a rapidly growing and troubling group of diseases. Future studies should help clarify these issues and allow us to target these most intriguing of organelles for therapeutic gain. 1.

2.

3. 4.

5. 6. 7. 8.

9. 10. 11. 12. 13.

14.

15. 16.

17.

Murphy, M. P. & Hartley, R. C. Mitochondria as a therapeutic target for common pathologies. Nat. Rev. Drug Discov. 17, 865–886 (2018). The many pathogical roles of mitochondria are discussed. Chandel, N. S. Evolution of mitochondria as signaling organelles. Cell Metab. 22, 204–206 (2015). The key signalling roles of mitochondria are discussed in an evolutionary context. Picard, M. & Shirihai, O. S. Mitochondrial signal transduction. Cell Metab. 34, 1620–1653 (2022). Monzel, A. S., Enriquez, J. A. & Picard, M. Multifaceted mitochondria: moving mitochondrial science beyond function and dysfunction. Nat. Metab. 5, 546–562 (2023). A recent review that highlights the many emerging facets of mitochondrial biology. Tait, S. W. & Green, D. R. Mitochondria and cell signalling. J. Cell Sci. 125, 807–815 (2012). Bahat, A., MacVicar, T. & Langer, T. Metabolism and innate immunity meet at the mitochondria. Front. Cell Dev. Biol. 9, 720490 (2021). Marchi, S., Guilbaud, E., Tait, S. W. G., Yamazaki, T. & Galluzzi, L. Mitochondrial control of inflammation. Nat. Rev. Immunol. 23, 159–173 (2023). Wein, T. & Sorek, R. Bacterial origins of human cell-autonomous innate immune mechanisms. Nat. Rev. Immunol. 22, 629–638 (2022). Here the authors suggest how the origins of mitochondria can lead to innate immunity mechanisms. Krysko, D. V. et al. Emerging role of damage-associated molecular patterns derived from mitochondria in inflammation. Trends Immunol. 32, 157–164 (2011). Galluzzi, L., Kepp, O. & Kroemer, G. Mitochondria: master regulators of danger signalling. Nat. Rev. Mol. Cell Biol. 13, 780–788 (2012). Kaplan, G. G. & Windsor, J. W. The four epidemiological stages in the global evolution of inflammatory bowel disease. Nat. Rev. Gastroenterol. Hepatol. 18, 56–66 (2021). Dinse, G. E. et al. Increasing prevalence of antinuclear antibodies in the United States. Arthritis Rheumatol. 72, 1026–1035 (2020). Wang, R., Li, Z., Liu, S. & Zhang, D. Global, regional and national burden of inflammatory bowel disease in 204 countries and territories from 1990 to 2019: a systematic analysis based on the Global Burden of Disease Study 2019. BMJ Open 13, e065186 (2023). Duarte-Garcia, A. et al. Rising incidence and prevalence of systemic lupus erythematosus: a population-based study over four decades. Ann. Rheum. Dis. https://doi.org/10.1136/ annrheumdis-2022-222276 (2022). Walton, C. et al. Rising prevalence of multiple sclerosis worldwide: insights from the Atlas of MS, third edition. Mult. Scler. 26, 1816–1821 (2020). Shi, G. et al. Estimation of the global prevalence, incidence, years lived with disability of rheumatoid arthritis in 2019 and forecasted incidence in 2040: results from the Global Burden of Disease Study 2019. Clin. Rheumatol. 42, 2297–2309 (2023). Furman, D. et al. Chronic inflammation in the etiology of disease across the life span. Nat. Med. 25, 1822–1832 (2019). An excellent account of how chronic inflammation changes as we age and how environmental factors impact on inflammatory diseases.

18. Gontier, N. in Encyclopedia of Evolutionary Biology Vol. 4 (ed. Kliman, R. M.) 261–271 (Elsevier, 2016). 19. Garg, S., Zimorski, V. & Martin, W. F. in Encyclopedia of Evolutionary Biology Vol. 1 (ed. Kliman, R. M.) 511–517 (Elsevier, 2016). 20. Dacks, J. B. et al. The changing view of eukaryogenesis – fossils, cells, lineages and how they all come together. J. Cell Sci. 129, 3695–3703 (2016). 21. Roger, A. J., Munoz-Gomez, S. A. & Kamikawa, R. The origin and diversification of mitochondria. Curr. Biol. 27, R1177–R1192 (2017). 22. Sagan, L. On the origin of mitosing cells. J. Theor. Biol. 14, 255–274 (1967). A classic paper that led to the acceptance of endosymbiosis as the origin of mitochondria. 23. Zaremba-Niedzwiedzka, K. et al. Asgard archaea illuminate the origin of eukaryotic cellular complexity. Nature 541, 353–358 (2017). 24. Martin, W. F. & Mentel, M. The origin of mitochondria. Nat. Educ. 3, 58 (2010). 25. John, P. & Whatley, F. R. Paracoccus denitrificans and the evolutionary origin of the mitochondrion. Nature 254, 495–498 (1975). 26. Geiger, O., Sanchez-Flores, A., Padilla-Gomez, J. & Degli Esposti, M. Multiple approaches of cellular metabolism define the bacterial ancestry of mitochondria. Sci. Adv. 9, eadh0066 (2023). 27. Martin, W. & Muller, M. The hydrogen hypothesis for the first eukaryote. Nature 392, 37–41 (1998). 28. Raval, P. K., Martin, W. F. & Gould, S. B. Mitochondrial evolution: gene shuffling, endosymbiosis, and signaling. Sci. Adv. 9, eadj4493 (2023). 29. Wallace, D. C. Mitochondrial diseases in man and mouse. Science 283, 1482–1488 (1999). 30. Gorman, G. S. et al. Mitochondrial diseases. Nat. Rev. Dis. Primers 2, 16080 (2016). 31. Gustafsson, C. M., Falkenberg, M. & Larsson, N. G. Maintenance and expression of mammalian mitochondrial DNA. Annu. Rev. Biochem. 85, 133–160 (2016). 32. Rath, S. et al. MitoCarta3.0: an updated mitochondrial proteome now with sub-organelle localization and pathway annotations. Nucleic Acids Res. 49, D1541–D1547 (2021). 33. Morgenstern, M. et al. Quantitative high-confidence human mitochondrial proteome and its dynamics in cellular context. Cell Metab. 33, 2464–2483 (2021). 34. Gross, J. & Bhattacharya, D. Mitochondrial and plastid evolution in eukaryotes: an outsiders’ perspective. Nat. Rev. Genet. 10, 495–505 (2009). 35. Paschen, S. A., Neupert, W. & Rapaport, D. Biogenesis of beta-barrel membrane proteins of mitochondria. Trends Biochem. Sci. 30, 575–582 (2005). 36. Gross, A. et al. Caspase cleaved BID targets mitochondria and is required for cytochrome c release, while Bcl-Xl prevents this release but not tumor necrosis factor-R1/Fas death. J. Biol. Chem. 274, 1156–1163 (1999). 37. Liu, X. S., Kim, C. N., Yang, J., Jemmerson, R. & Wang, X. D. Induction of apoptotic program in cell-free extracts – requirement for datp and cytochrome c. Cell 86, 147–157 (1996). 38. Giacomello, M., Pyakurel, A., Glytsou, C. & Scorrano, L. The cell biology of mitochondrial membrane dynamics. Nat. Rev. Mol. Cell Biol. 21, 204–224 (2020). 39. Kalkavan, H. & Green, D. R. MOMP, cell suicide as a BCL-2 family business. Cell Death Differ. 25, 46–55 (2018). 40. Suhaili, S. H., Karimian, H., Stellato, M., Lee, T. H. & Aguilar, M. I. Mitochondrial outer membrane permeabilization: a focus on the role of mitochondrial membrane structural organization. Biophys. Rev. 9, 443–457 (2017). 41. He, B. et al. Mitochondrial cristae architecture protects against mtDNA release and inflammation. Cell Rep. 41, 111774 (2022). 42. Frezza, C. et al. OPA1 controls apoptotic cristae remodeling independently from mitochondrial fusion. Cell 126, 177–189 (2006). 43. Ott, M., Zhivotovsky, B. & Orrenius, S. Role of cardiolipin in cytochrome c release from mitochondria. Cell Death Differ. 14, 1243–1247 (2007). 44. Munoz-Gomez, S. A., Slamovits, C. H., Dacks, J. B. & Wideman, J. G. The evolution of MICOS: ancestral and derived functions and interactions. Commun. Integr. Biol. 8, e1094593 (2015). 45. Friedman, J. R., Mourier, A., Yamada, J., McCaffery, J. M. & Nunnari, J. MICOS coordinates with respiratory complexes and lipids to establish mitochondrial inner membrane architecture. eLife 4, e07739 (2015). 46. Bernardi, P. & Di Lisa, F. The mitochondrial permeability transition pore: molecular nature and role as a target in cardioprotection. J. Mol. Cell. Cardiol. 78, 100–106 (2015). 47. Scarlett, J. L. & Murphy, M. P. Release of apoptogenic proteins from the mitochondrial intermembrane space during the mitochondrial permeability transition. FEBS Lett. 418, 282–286 (1997). 48. Bernardi, P. et al. Identity, structure, and function of the mitochondrial permeability transition pore: controversies, consensus, recent advances, and future directions. Cell Death Differ. 30, 1869–1885 (2023). 49. Liu, B. et al. CpG methylation patterns of human mitochondrial DNA. Sci. Rep. 6, 23421 (2016). 50. Riley, J. S. & Tait, S. W. Mitochondrial DNA in inflammation and immunity. EMBO Rep. 21, e49799 (2020). 51. Kim, J., Kim, H. S. & Chung, J. H. Molecular mechanisms of mitochondrial DNA release and activation of the cGAS-STING pathway. Exp. Mol. Med. 55, 510–519 (2023). 52. McArthur, K. et al. BAK/BAX macropores facilitate mitochondrial herniation and mtDNA efflux during apoptosis. Science 359, 6378 (2018). First description of a role for BAK/BAX and mitochondrial herniation in the release of mtDNA. 53. Kim, J. et al. VDAC oligomers form mitochondrial pores to release mtDNA fragments and promote lupus-like disease. Science 366, 1531–1536 (2019). Evidence for oxidized mtDNA as an activator of the NLRP3 inflammasome. 54. Xian, H. et al. Oxidized DNA fragments exit mitochondria via mPTP- and VDAC-dependent channels to activate NLRP3 inflammasome and interferon signaling. Immunity 55, 1370–1385 (2022). Evidence for oxidized mtDNA as an activator of the NLRP3 inflammasome. 55. Mills, E. L., Kelly, B. & O’Neill, L. A. J. Mitochondria are the powerhouses of immunity. Nat. Immunol. 18, 488–498 (2017).

Nature | Vol 626 | 8 February 2024 | 277

Review 56. Billingham, L. K. et al. Mitochondrial electron transport chain is necessary for NLRP3 inflammasome activation. Nat. Immunol. 23, 692–704 (2022). Evidence that mitochondrial phosphocreatine generated from ATP derived from oxidative phosphorylation is required for ATP production in the cytosol by creatine kinase B, for NLRP3 activation. 57. Chowdhury, A., Witte, S. & Aich, A. Role of mitochondrial nucleic acid sensing pathways in health and patho-physiology. Front. Cell Dev. Biol. 10, 796066 (2022). 58. Chen, Y. G. & Hur, S. Cellular origins of dsRNA, their recognition and consequences. Nat. Rev. Mol. Cell Biol. 23, 286–301 (2022). 59. Seth, R. B., Sun, L., Ea, C. K. & Chen, Z. J. Identification and characterization of MAVS, a mitochondrial antiviral signaling protein that activates NF-kappaB and IRF 3. Cell 122, 669–682 (2005). 60. Wang, F., Zhang, D., Zhang, D., Li, P. & Gao, Y. Mitochondrial protein translation: emerging roles and clinical significance in disease. Front. Cell Dev. Biol. 9, 675465 (2021). 61. Walker, J. E., Carroll, J., Altman, M. C. & Fearnley, I. M. Chapter 6 mass spectrometric characterization of the thirteen subunits of bovine respiratory complexes that are encoded in mitochondrial DNA. Methods Enzymol. 456, 111–131 (2009). 62. Le, Y., Murphy, P. M. & Wang, J. M. Formyl-peptide receptors revisited. Trends Immunol. 23, 541–548 (2002). 63. Dorward, D. A. et al. Novel role for endogenous mitochondrial formylated peptide-driven formyl peptide receptor 1 signalling in acute respiratory distress syndrome. Thorax 72, 928–936 (2017). 64. Cai, N. et al. Mitochondrial DNA variants modulate N-formylmethionine, proteostasis and risk of late-onset human diseases. Nat. Med. 27, 1564–1575 (2021). A fascinating report linking mitochondrial N-formylmethionine formation and pathology. 65. Paradies, G., Paradies, V., Ruggiero, F. M. & Petrosillo, G. Role of cardiolipin in mitochondrial function and dynamics in health and disease: molecular and pharmacological aspects. Cells 8, 728 (2019). 66. Pizzuto, M. & Pelegrin, P. Cardiolipin in immune signaling and cell death. Trends Cell Biol. 30, 892–903 (2020). 67. Dudek, J. Role of cardiolipin in mitochondrial signaling pathways. Front. Cell Dev. Biol. 5, 90 (2017). 68. Iyer, S. S. et al. Mitochondrial cardiolipin is required for Nlrp3 inflammasome activation. Immunity 39, 311–323 (2013). 69. Matzinger, P. Tolerance, danger, and the extended family. Annu. Rev. Immunol. 12, 991–1045 (1994). 70. Murphy, M. P. How mitochondria produce reactive oxygen species. Biochem. J 417, 1–13 (2009). An overview of how mitochondrial redox signals may be generated. 71. Wong, H. S., Dighe, P. A., Mezera, V., Monternier, P. A. & Brand, M. D. Production of superoxide and hydrogen peroxide from specific mitochondrial sites under different bioenergetic conditions. J. Biol. Chem. 292, 16804–16809 (2017). 72. Robb, E. L. et al. Control of mitochondrial superoxide production by reverse electron transport at complex I. J. Biol. Chem. 293, 9869–9879 (2018). 73. Wright, J. J. et al. Reverse electron transfer by respiratory complex I catalyzed in a modular proteoliposome system. J. Am. Chem. Soc. 144, 6791–6801 (2022). 74. Roca, F. J., Whitworth, L. J., Prag, H. A., Murphy, M. P. & Ramakrishnan, L. Tumor necrosis factor induces pathogenic mitochondrial ROS in tuberculosis through reverse electron transport. Science. 376, eabh2841 (2022). 75. Murphy, M. P. & Chouchani, E. T. Why succinate? Physiological regulation by a mitochondrial coenzyme Q sentinel. Nat. Chem. Biol. 18, 461–469 (2022). 76. Mills, E. L. et al. Succinate dehydrogenase supports metabolic repurposing of mitochondria to drive inflammatory macrophages. Cell 167, 457–470 (2016). This paper describes how mitochondrial metabolism can be repurposed to generate succinate as a signal. 77. Xiao, H. et al. A quantitative tissue-specific landscape of protein redox regulation during aging. Cell 180, 968–983 (2020). 78. Holmstrom, K. M. & Finkel, T. Cellular mechanisms and physiological consequences of redox-dependent signalling. Nat. Rev. Mol. Cell Biol. 15, 411–421 (2014). 79. Christman, M. F., Storz, G. & Ames, B. N. OxyR, a positive regulator of hydrogen peroxideinducible genes in Escherichia coli and Salmonella typhimurium, is homologous to a family of bacterial regulatory proteins. Proc. Natl Acad. Sci. USA 86, 3484–3488 (1989). 80. Redza-Dutordoir, M. & Averill-Bates, D. A. Activation of apoptosis signalling pathways by reactive oxygen species. Biochim. Biophys. Acta 1863, 2977–2992 (2016). 81. West, A. P. et al. TLR signalling augments macrophage bactericidal activity through mitochondrial ROS. Nature 472, 476–480 (2011). 82. Ryan, D. G. et al. Coupling Krebs cycle metabolites to signalling in immunity and cancer. Nat. Metab. 1, 16–33 (2019). 83. Murphy, M. P. & O’Neill, L. A. J. Krebs cycle reimagined: the emerging roles of succinate and itaconate as signal transducers. Cell 174, 780–784 (2018). 84. Martinez-Reyes, I. & Chandel, N. S. Mitochondrial TCA cycle metabolites control physiology and disease. Nat. Commun. 11, 102 (2020). 85. Sivanand, S., Viney, I. & Wellen, K. E. Spatiotemporal control of acetyl-CoA metabolism in chromatin regulation. Trends Biochem. Sci. 43, 61–74 (2018). 86. Tannahill, G. M. et al. Succinate is an inflammatory signal that induces IL-1beta through HIF-1alpha. Nature 496, 238–242 (2013). Evidence for macrophage-derived succinate being a pro-inflammatory signal. 87. Selak, M. A. et al. Succinate links TCA cycle dysfunction to oncogenesis by inhibiting HIF-alpha prolyl hydroxylase. Cancer Cell 7, 77–85 (2005). A key paper linking succinate to HIF1-alpha activation. 88. Matilainen, O., Quiros, P. M. & Auwerx, J. Mitochondria and epigenetics – crosstalk in homeostasis and stress. Trends Cell Biol. 27, 453–463 (2017). 89. Santos, J. H. Mitochondria signaling to the epigenome: a novel role for an old organelle. Free Radic. Biol. Med. 170, 59–69 (2021). 90. Mills, E. L. et al. Itaconate is an anti-inflammatory metabolite that activates Nrf2 via alkylation of KEAP1. Nature 556, 113–117 (2018).

278 | Nature | Vol 626 | 8 February 2024

91. Day, E. A. & O’Neill, L. A. J. Protein targeting by the itaconate family in immunity and inflammation. Biochem. J. 479, 2499–2510 (2022). 92. McGettrick, A. F. & O’Neill, L. A. Two for the price of one: itaconate and its derivatives as an anti-infective and anti-inflammatory immunometabolite. Curr. Opin. Immunol. 80, 102268 (2023). 93. DeBerardinis, R. J. & Chandel, N. S. Fundamentals of cancer metabolism. Sci. Adv. 2, e1600200 (2016). 94. Liberti, M. V. & Locasale, J. W. The Warburg effect: how does it benefit cancer cells? Trends Biochem. Sci. 41, 211–218 (2016). 95. DeBerardinis, R. J. & Chandel, N. S. We need to talk about the Warburg effect. Nat. Metab. 2, 127–129 (2020). 96. Weinberg, F. et al. Mitochondrial metabolism and ROS generation are essential for Kras-mediated tumorigenicity. Proc. Natl Acad. Sci. USA 107, 8788–8793 (2010). 97. Martinez-Reyes, I. et al. Mitochondrial ubiquinol oxidation is necessary for tumour growth. Nature 585, 288–292 (2020). 98. Sena, L. A. et al. Mitochondria are required for antigen-specific T cell activation through reactive oxygen species signaling. Immunity 38, 225–236 (2013). 99. Frenkel-Pinter, M. et al. Adaptation and exaptation: from small molecules to feathers. J. Mol. Evol. 90, 166–175 (2022). 100. Jarman, O. D., Biner, O., Wright, J. J. & Hirst, J. Paracoccus denitrificans: a genetically tractable model system for studying respiratory complex I. Sci. Rep. 11, 10143 (2021). 101. Henry, M. F. & Vignais, P. M. Production of superoxide anions in Paracoccus denitrificans. Arch. Biochem. Biophys. 203, 365–371 (1980). 102. Kotlyar, A. B. & Borovok, N. NADH oxidation and NAD+ reduction catalysed by tightly coupled inside-out vesicles from Paracoccus denitrificans. Eur. J. Biochem. 269, 4020–4024 (2002). 103. Hong, Y., Zeng, J., Wang, X., Drlica, K. & Zhao, X. Post-stress bacterial cell death mediated by reactive oxygen species. Proc. Natl Acad. Sci. USA 116, 10064–10071 (2019). 104. Lopatina, A., Tal, N. & Sorek, R. Abortive infection: bacterial suicide as an antiviral immune strategy. Annu. Rev. Virol. 7, 371–384 (2020). 105. Toyofuku, M., Schild, S., Kaparakis-Liaskos, M. & Eberl, L. Composition and functions of bacterial membrane vesicles. Nat. Rev. Microbiol. 21, 415–430 (2023). 106. Horvath, P. & Barrangou, R. CRISPR/Cas, the immune system of bacteria and archaea. Science 327, 167–170 (2010). 107. Georjon, H. & Bernheim, A. The highly diverse antiphage defence systems of bacteria. Nat. Rev. Microbiol. 21, 686–700 (2023). 108. Li, Y. et al. cGLRs are a diverse family of pattern recognition receptors in innate immunity. Cell 186, 3261–3276 (2023). 109. Wu, X. et al. Phylogenetic and molecular evolutionary analysis of mitophagy receptors under hypoxic conditions. Front. Physiol. 8, 539 (2017). 110. Zhang, Q. et al. Circulating mitochondrial DAMPs cause inflammatory responses to injury. Nature 464, 104–107 (2010). 111. Moriyama, M., Koshiba, T. & Ichinohe, T. Influenza A virus M2 protein triggers mitochondrial DNA-mediated antiviral immune responses. Nat. Commun. 10, 4624 (2019). A role for mitochondrial DNA in the induction of anti-viral immunity in response to an RNA virus (influenza). 112. Jahun, A. S. et al. Leaked genomic and mitochondrial DNA contribute to the host response to noroviruses in a STING-dependent manner. Cell Rep. 42, 112179 (2023). 113. Sun, B. et al. Dengue virus activates cGAS through the release of mitochondrial DNA. Sci. Rep. 7, 3594 (2017). 114. West, A. P. et al. Mitochondrial DNA stress primes the antiviral innate immune response. Nature 520, 553–557 (2015). 115. Colaco, C. B., Scadding, G. K. & Lockhart, S. Anti-cardiolipin antibodies in neurological disorders: cross-reaction with anti-single stranded DNA activity. Clin. Exp. Immunol. 68, 313–319 (1987). 116. Colapietro, F., Lleo, A. & Generali, E. Antimitochondrial antibodies: from bench to bedside. Clin. Rev. Allergy Immunol. 63, 166–177 (2022). 117. Chen, P. M. & Tsokos, G. C. Mitochondria in the pathogenesis of systemic lupus erythematosus. Curr. Rheumatol. Rep. 24, 88–95 (2022). 118. Becker, Y. et al. Autoantibodies in systemic lupus erythematosus target mitochondrial RNA. Front. Immunol. 10, 1026 (2019). 119. Chouchani, E. T. et al. Ischaemic accumulation of succinate controls reperfusion injury through mitochondrial ROS. Nature 515, 431–435 (2014). 120. Littlewood-Evans, A. et al. GPR91 senses extracellular succinate released from inflammatory macrophages and exacerbates rheumatoid arthritis. J. Exp. Med. 213, 1655–1662 (2016). 121. Li, Q. et al. RNA editing underlies genetic risk of common inflammatory diseases. Nature 608, 569–577 (2022). 122. Hooftman, A. et al. Macrophage fumarate hydratase restrains mtRNA-mediated interferon production. Nature 615, 490–498 (2023). 123. Zecchini, V. et al. Fumarate induces vesicular release of mtDNA to drive innate immunity. Nature 615, 499–506 (2023). 124. Sciacovelli, M. et al. Fumarate is an epigenetic modifier that elicits epithelial-tomesenchymal transition. Nature 537, 544–547 (2016). 125. Whiteley, M., Diggle, S. P. & Greenberg, E. P. Progress in and promise of bacterial quorum sensing research. Nature 551, 313–320 (2017). 126. Lopez-Domenech, G. et al. Miro proteins coordinate microtubule- and actin-dependent mitochondrial transport and distribution. EMBO J. 37, 321–336 (2018). 127. Debattisti, V., Gerencser, A. A., Saotome, M., Das, S. & Hajnoczky, G. ROS control mitochondrial motility through p38 and the motor adaptor Miro/Trak. Cell Rep. 21, 1667–1680 (2017). 128. Croon, M. et al. FGF21 modulates mitochondrial stress response in cardiomyocytes only under mild mitochondrial dysfunction. Sci. Adv. 8, eabn7105 (2022). 129. Zorov, D. B., Juhaszova, M. & Sollott, S. J. Mitochondrial reactive oxygen species (ROS) and ROS-induced ROS release. Physiol. Rev. 94, 909–950 (2014). 130. Campello, S. et al. Orchestration of lymphocyte chemotaxis by mitochondrial dynamics. J. Exp. Med. 203, 2879–2886 (2006). 131. Grafstein, B. & Forman, D. S. Intracellular transport in neurons. Physiol. Rev. 60, 1167–1283 (1980).

132. Eisner, V., Picard, M. & Hajnoczky, G. Mitochondrial dynamics in adaptive and maladaptive cellular stress responses. Nat. Cell Biol. 20, 755–765 (2018). 133. Spees, J. L., Olson, S. D., Whitney, M. J. & Prockop, D. J. Mitochondrial transfer between cells can rescue aerobic respiration. Proc. Natl Acad. Sci. USA 103, 1283–1288 (2006). 134. Rustom, A., Saffrich, R., Markovic, I., Walther, P. & Gerdes, H. H. Nanotubular highways for intercellular organelle transport. Science. 303, 1007–1010 (2004). 135. Liu, D. et al. Intercellular mitochondrial transfer as a means of tissue revitalization. Signal Transduct. Target Ther. 6, 65 (2021). 136. Liu, Z., Sun, Y., Qi, Z., Cao, L. & Ding, S. Mitochondrial transfer/transplantation: an emerging therapeutic approach for multiple diseases. Cell Biosci. 12, 66 (2022). 137. Dong, L. F. et al. Horizontal transfer of whole mitochondria restores tumorigenic potential in mitochondrial DNA-deficient cancer cells. eLife 6, e22187 (2017). 138. McCully, J. D., Levitsky, S., Del Nido, P. J. & Cowan, D. B. Mitochondrial transplantation for therapeutic use. Clin. Transl. Med. 5, 16 (2016). 139. Hayashida, K. et al. Mitochondrial transplantation therapy for ischemia reperfusion injury: a systematic review of animal and human studies. J. Transl. Med. 19, 214 (2021). 140. Sardon Puig, L., Valera-Alberni, M., Canto, C. & Pillon, N. J. Circadian rhythms and mitochondria: connecting the dots. Front. Genet. 9, 452 (2018). 141. Chang, E. M., Chao, C. C., Wang, M. T., Hsu, C. L. & Chen, P. C. PM(2.5) promotes pulmonary fibrosis by mitochondrial dysfunction. Environ. Toxicol. 38, 1905–1913 (2023). 142. Gioscia-Ryan, R. A. et al. Lifelong voluntary aerobic exercise prevents age- and Western diet- induced vascular dysfunction, mitochondrial oxidative stress and inflammation in mice. J. Physiol. 599, 911–925 (2021).

Acknowledgements We thank E. L. Mills, D. G. Ryan and H. A. Prag for helpful discussions. Author contributions Both authors contributed equally to the writing of the manuscript. Competing interests The authors declare no competing interests. Additional information Correspondence and requests for materials should be addressed to Michael P. Murphy or Luke A. J. O’Neill. Peer review information Nature thanks Navdeep Chandel, Zhijian (James) Chen, Luca Scorrano and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law. © Springer Nature Limited 2024

Nature | Vol 626 | 8 February 2024 | 279

A recently formed ocean inside Saturn’s moon Mimas https://doi.org/10.1038/s41586-023-06975-9

V. Lainey1 ✉, N. Rambaux1, G. Tobie2, N. Cooper3, Q. Zhang4, B. Noyelles5 & K. Baillié1

Received: 28 February 2023 Accepted: 14 December 2023 Published online: 7 February 2024 Check for updates

Moons potentially harbouring a global ocean are tending to become relatively common objects in the Solar System1. The presence of these long-lived global oceans is generally betrayed by surface modification owing to internal dynamics2. Hence, Mimas would be the most unlikely place to look for the presence of a global ocean3. Here, from detailed analysis of Mimas’s orbital motion based on Cassini data, with a particular focus on Mimas’s periapsis drift, we show that its heavily cratered icy shell hides a global ocean, at a depth of 20–30 kilometres. Eccentricity damping implies that the ocean is likely to be less than 25 million years old and still evolving. Our simulations show that the ocean–ice interface reached a depth of less than 30 kilometres only recently (less than 2–3 million years ago), a time span too short for signs of activity at Mimas’s surface to have appeared.

Analysing the rotational motion of Mimas from Cassini Imaging Science Subsystem (ISS) images, a measured libration amplitude of −50.3 ± 1.0 arcmin was found for the Mimas orbital frequency4. It was deduced that Mimas should harbour either a highly elongated silicate core or a global ocean. The libration measurement was based on the stereophotogrammetry control point network method4. Hence, Cassini ISS data were used to investigate the rotation of Mimas from its surface motion only. To discriminate between the two interior models (a prominent distorted silicate core or a global internal ocean), it is necessary to have a further constraint on the internal state. However, these two interior assumptions imply very different gravitational potentials, corresponding to a rigid body in the case of the elongated scenario, or a body composed of different layers in relative motion in the case of the scenario with an ocean. Both models would induce a different gravitational pull received by Mimas from Saturn, implying a slightly different orbital trajectory over time. The study of such an orbital effect was proposed theoretically5 and recently applied with success to the inner Saturnian moons6. In the case of a completely solid body, the libration amplitude depends primarily on the gravity coefficient C22 for Mimas as the reasonable range of C/MR2 is small, where C, M and R are the principal moment of inertia about the rotation axis, the mass of Mimas and the radius of Mimas, respectively. However, the orbital effect depends on both C22 and J2, where J2 is the only other degree-two gravity harmonic for Mimas if we assume rotational stability. Accordingly, the two measurements (libration and orbital change) can be used to solve for both J2 and C22 provided the body is solid. The measured orbital shift is −9.4 ± 0.9 km over the duration of the Cassini mission (Methods), and we find that this is compatible with both measurements provided that J2/C22 is of the order of 10. For a more typical J2/C22 of the order of three (that is, close to hydrostatic interior), the predicted orbital shift is about a factor of two larger in magnitude. Although unusually large, this ratio is a possible value if we assume that the silicate-core component of Mimas takes the

form of triaxial ellipsoid that is an elliptical pancake in the equatorial plane, elongated along the line to Saturn. The problem with such a solution is that the total volume of the silicate core is constrained by the mean density of Mimas, and only a markedly flattened shape can satisfy this constraint along with the required large C22, and even larger J2, if the elongation of the pancake causes its extremity to pierce the surface of Mimas. This is incompatible with observations (Methods). As the solid Mimas hypothesis leads to a dead end in terms of interior modelling, we tested the influence of a subsurface ocean. Starting from the one-layer formulation of the periapsis drift, we extended the theoretical approach5 to the case of a body containing an internal ocean. The derived expression is provided in Methods. The new expression depends essentially on the flattening for each layer that controls the gravity coefficients and libration amplitude. Here we took into account the libration amplitudes of the icy crust and the silicate mantle4,7. Figure 1 shows the set of models explored as a function of the libration amplitude in longitude and the periapsis drift. The librational model is based on three deformed layers7 where the outer ice shell is viscoelastic4. Here the surface polar and equatorial flattenings are computed based on the best-fit ellipsoids8, and the interior interfaces are assumed to be at hydrostatic values. The intersection of the two sets of measurements (in grey) allows us to exclude several interior models and to constrain the ice-sheet thickness to 20–30 km to be consistent with the measurements. The thickness of Mimas’s icy shell that we obtained is in close agreement with the one deduced recently3, assuming a realistic model for dissipation in the satellite. This confirms that Mimas may be close to thermal equilibrium at the present time. The amount of heat released at the surface of Mimas was estimated to be typically3 25 mW m−2, corresponding to a total dissipated power of 12 GW. Nevertheless, such a large energy loss should be accompanied by a damping of Mimas’s eccentricity, which, at this dissipation rate, should be reduced by a factor of 2 in 4–5 Myr.

IMCCE, Observatoire de Paris, PSL Research University, Sorbonne Université, CNRS, Université Lille, Paris, France. 2LPG, UMR-CNRS 6112, Nantes Université, Nantes, France. 3Department of Physics and Astronomy, Queen Mary University of London, London, UK. 4Department of Computer Science, Jinan University, Guangzhou, P. R. China. 5Institut UTINAM, CNRS UMR 6213, Université de Franche-Comté, OSU THETA, BP 1615, Besançon, France. ✉e-mail: [email protected] 1

280 | Nature | Vol 626 | 8 February 2024

50

–30

45

–40

40

–50

35

–60

30

–70

25

–80 –9

–8

–7

–6

–5

–4

–3

–2

evolved, the ocean evolution owing to heat balance and the progressive evolutionary decay9,10 (Methods). Figure 2 shows the evolution of Mimas’s eccentricity, tidal power, ice-shell thickness and resulting surface heat flux, for end-member assumptions for the mechanical properties for the rocky core and the ice shell. For all tested cases, the initial eccentricity has been adjusted to match the present-day eccentricity value and to have an ice-shell thickness ranging between 20 km and 30 km. The initial eccentricity at the time of ocean initiation ranges between 2.3 times the current value for a very dissipative rocky core, comparable to Enceladus, to 2.9 for a rigid core. Tests performed at higher initial eccentricity values result in a very thin ice shell (50–100 mW m−2), incompatible with the lack of surface activity. Our simulations show that the ocean appeared between 25 Ma for the less dissipative case and only 2–3 Ma for the most dissipative case. In all cases, a rapid growth of the internal ocean occurs during the past few million years. Interestingly, the most dissipative model results in a smaller surface heat flux, owing to very rapid ocean melting and the delay in the propagation of the heat wave to the surface. As a comparison, the surface heat flux of 20–25 mW m−2 obtained here corresponds to the estimated surface heat flux in the oldest cratered terrains observed in the equatorial region of Enceladus11. Our results clearly indicate that a hydrothermally active porous core inside Mimas comparable to Enceladus is possible even in the absence of surface activity. Moreover, for an ice-shell thickness ranging

hs (km)

Is (arcmin)

–20

20

ΔY × 107 (rad per day)

Fig. 1 | Mimas measurements and ocean models. The amplitude of libration in longitude ϕs and periapsis drift variation Δϖ for different internal structure models with an ocean. The colours represent the thickness of the ice crust hs. The grey areas correspond to the measured libration amplitude and perihelion longitude variation. The dispersion represents sensitivity to the crustal polar and equatorial flattenings (see additional tests in Methods).

To predict the recent past evolution of Mimas leading to the present day, we computed the tidal dissipation inside Mimas as the interior a

b

3.0

70 2.5

60 50

Ptide (GW)

e/ea

2.0

1.5

1.0

40 30 20 10

0.5

0 –15.0

–12.5

–10.0

–7.5

–5.0

–2.5

0

2.5

5.0

Time before present (Ma)

c

–15.0

–12.5

–10.0

–7.5

–5.0

–2.5

0

2.5

5.0

0

2.5

5.0

Time before present (Ma)

d

100

40

90

35

80 30 Isurf (mW m–2)

bice (km)

70 60 50

25 20 15

40 10 30 5

20 –15.0

–12.5

–10.0

–7.5

–5.0

–2.5

0

2.5

Time before present (Ma)

5.0

–15.0

–12.5

–10.0

–7.5 –5.0 –2.5 Time before present (Ma)

Non-dissipative core Dissipative core Km = 1016 Pa s Km = 1015 Pa s

Km = 1014 Pa s

Fig. 2 | Mimas’s interior and orbital evolution. a–d, Time evolution of Mimas’s eccentricity (normalized to the present value ea; a), tidal dissipated power P tide (b), ice-shell thickness bice (c) and surface heat flux ϕsurf (d) for interior models with a rigid non-dissipative core or with a unconsolidated dissipative core comparable to Enceladus. The pink dashed lines indicate the upper and lower estimates of ice-shell thickness at present. The grey area

indicates predicted time evolution during the next 5 Myr. For the three tested values of ice viscosity at the melting point, ηm (1014 Pa s, 1015 Pa s and 1016 Pa s), the initial eccentricity was adjusted to match the present-day eccentricity value and the estimated ice-shell thickness. For a rigid core and ηm = 1016 Pa s, no solution could match the ice-shell thickness estimate and therefore no results are shown. Ma, million years ago.

Nature | Vol 626 | 8 February 2024 | 281

between 20 km and 30 km, we predict a Love number k2 of the order of 0.01, which might be measured by a future dedicated mission, thus confirming the presence of an ocean and its depth. The origin of a larger eccentricity in the past may well be the consequence of a resonant encounter with another main Saturnian moon. Although Mimas is currently in a mean motion resonance with Tethys, this interaction does not involve Mimas’s eccentricity. Hence, Mimas may have passed through another resonant interaction involving its eccentricity in the past. As recently demonstrated12, such a resonance could also involve three Saturnian moons. In particular, a larger eccentricity for Mimas could be obtained, up to 0.06, by considering a three-body resonance involving Mimas, Dione and Titan12. Another possibility could be that Mimas crossed a resonant encounter, but without capture. Such a hypothesis was studied13,14 to explain the formation of the Cassini Division. However, for a past eccentricity of 2.3–2.9 times the present value, Mimas alone cannot explain the opening of the Cassini Division anymore, suggesting that Enceladus should have had a higher eccentricity as well, and thus a more intense heating during the past millions of years. More recently, it has been suggested that the loss of a Saturnian moon could explain Saturn’s obliquity and young rings15 (with a typical age of 100 Myr). Under such a hypothesis, the mid-sized moon system might have been jostled, allowing for slight changes in their orbital elements, with possible increases in eccentricity for some of them in the recent past. All these rapid changes in orbital dyna­mics are triggered by the strong tidal dissipation in Saturn16–18. From that perspective, the temporary existence of Mimas’ ocean is another manifestation of strong Saturnian tides. Our results imply that Mimas has experienced during the past million years internal processes that may have been common in many icy worlds shortly after their formation. From this point of view, Mimas offers a unique opportunity to study ongoing melting-induced differentiation and extensive aqueous alteration driven by water–rock interactions. Mimas can thus provide a time window into the past evolution for Enceladus and other now-quiet icy worlds, including Uranian moons and Kuiper Belt objects. Aqueous mineral alteration in icy worlds is known to be geologically rapid, typically of the order of millions to tens of millions of years19,20. Mimas would be one of the few places where this could happen right now.

Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions

282 | Nature | Vol 626 | 8 February 2024

and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06975-9. 1.

2. 3. 4. 5. 6. 7.

8. 9. 10. 11. 12. 13.

14.

15. 16. 17. 18. 19. 20.

Castillo-Rogez, J. et al. Compositions and interior structures of the large moons of Uranus and implications for future spacecraft observations. J. Geophys. Res. Planets 128, e2022JE007432 (2023). Ćuk, M., Dones, L. & Nesvorný, D. Dynamical evidence for a late formation of Saturn’s moons. Astrophys. J. 820, 97 (2016). Rhoden, A. R. & Walker, M. E. The case for an ocean-bearing Mimas from tidal heating analysis. Icarus 376, 114872 (2022). Tajeddine, R. et al. Constraints on Mimas’ interior from Cassini ISS libration measurements. Science 346, 322–324 (2014). Borderies, N. & Yoder, C. F. Phobos’ gravity field and its influence on its orbit and physical librations. Astron. Astrophys. 233, 235–251 (1990). Lainey, V., Rambaux, N., Cooper, N. & Zhang, Q. Characterizing the interior of five inner Saturnian moons using Cassini ISS data. Astron. Astrophys. 670, L25 (2023). Viswanathan, V., Rambaux, N., Fienga, A., Laskar, J. & Gastineau, M. Observational constraint on the radius and oblateness of the lunar core–mantle boundary. Geophys. Res. Lett. 46, 7295–7303 (2019). Balmino, G. Gravitational potential harmonics from the shape of an homogeneous body. Celest. Mech. Dyn. Astron. 60, 331–364 (1994). Tobie, G., Grasset, O., Lunine, J. I., Mocquet, A. & Sotin, C. Titan’s internal structure inferred from a coupled thermal–orbital model. Icarus 175, 496–502 (2005). Tobie, G., Mocquet, A. & Sotin, C. Tidal dissipation within large icy satellites: applications to Europa and Titan. Icarus 177, 534–549 (2005). Cadek, O. et al. Long-term stability of Enceladus’ uneven ice shell. Icarus 319, 476–484 (2019). Ćuk, M. & El Moutamid, M. Three-body resonances in the Saturnian system. Astrophys. J. 926, L18 (2022). Baillié, K., Noyelles, B., Lainey, V., Charnoz, S. & Tobie, G. Formation of the Cassini Division—I. Shaping the rings by Mimas inward migration. Mon. Not. R. Astron. Soc. 486, 2933–2946 (2019). Noyelles, B., Baillié, K., Charnoz, S., Lainey, V. & Tobie, G. Formation of the Cassini Division—II. Possible histories of Mimas and Enceladus. Mon. Not. R. Astron. Soc. 486, 2947–2963 (2019). Wisdom, J. et al. Loss of a satellite could explain Saturn’s obliquity and young rings. Science 377, 1285–1289 (2022). Lainey, V. et al. Strong tidal dissipation in Saturn and constraints on Enceladus’ thermal state from astrometry. Astrophys. J. 752, 14 (2012). Lainey, V. et al. New constraints on Saturn’s interior from Cassini astrometric data. Icarus 281, 286–296 (2017). Lainey, V. et al. Resonance locking in giant planets indicated by the rapid orbital expansion of Titan. Nat. Astron. 4, 1053–1058 (2020). Zandanel, A. et al. Short lifespans of serpentinization in the rocky core of Enceladus: implications for hydrogen production. Icarus 364, 114461 (2021). Zandanel, A. et al. Geologically rapid aqueous mineral alteration at subfreezing temperatures in icy worlds. Nat. Astron. 6, 554–559 (2022).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law. © The Author(s), under exclusive licence to Springer Nature Limited 2024

Methods New astrometric reduction of Mimas and Tethys ISS images using three-dimensional complex-shape modelling To benefit from the most accurate astrometric data from ISS images, we reprocessed close images of Mimas and Tethys, using complex-shape modelling. Previously, all ISS data were astrometrically reduced using an ellipsoidal-shape model. But the limitation of using such a simple-shape modelling approach has recently been demonstrated21. Here we used a spherical harmonic representation for the topography of both moons, using a recently developed method22. The data were then split into two different subsets depending on the extension of the limb detection on the image. In Extended Data Fig. 1, most of the satellite edge is detectable on the image, with now an accurate modelling of Herschel crater. The typical difference between ellipsoidal-shape models versus those derived using spherical harmonics was found to be few tenths of a pixel. In addition, we looked for images that were acquired with a different filter combination than CL1/CL2 (clear filters). This allowed us to add several tens of extra images. No biases were found from the use of non-clear filters. All our new observations and astrometric residuals are fully available on request. Orbital modelling and astrometric fitting We challenged the elongated silicate-core hypothesis for Mimas’s interior by measuring the feedback of the associated physical libration on Mimas’s orbital motion. Analytical estimation suggests that the periapsis of Mimas should drift by 20.1 km over the 13 years of Cassini data (see Methods section ‘Periapsis drift for a fully rigid Mimas’), which is large enough compared with the ISS astrometric precision to be detectable. We used all the Cassini ISS astrometric data already considered in former work18. Moreover, we reprocessed all the close images of Mimas and Tethys, introducing an improved modelling of their shapes22 to increase the accuracy of our measurements (see previous section). Similarly, we improved the astrometric reduction of the small moons23, especially Methone and Anthe, whose motions are affected by commensurabilities with Mimas. Our model solved the equations of motion of the eight main moons of Saturn, with the addition of the five inner moons, the four Lagrangian moons of Dione and Tethys, as well as Methone, Anthe and Pallene. In addition to the initial state vectors of the moons, we fitted the masses of the moons and their primary, the gravity field of Saturn up to the order of ten, the orientation and precession of Saturn’s pole, the Saturnian Love number k2, and the physical libration of Prometheus, Pandora, Janus and Epimetheus. Mimas’s gravity field by means of the two gravity coefficients C20 and C22 and Mimas’s physical libration were simultaneously solved for, along with the physical parameters described above. The development was limited to degree two in shape, as the influence of higher terms is small on the C20 and C22 coefficients8. During the fitting procedure, Saturn’s gravity field was constrained to remain within the radio-science data solution24, while Mimas’s physical libration was constrained to be consistent at a one-sigma level with the direct measurement from Cassini images4. Our results are shown in Extended Data Table 1. Most of the parameter space for C20 and C22 is unphysical, as it implies positive C20 and negative C22. It appears that a solution remains possible allowing for negative C20 and positive C22. Owing to the tremendous anticorrelation between Mimas’s C20 and C22 at the 99% level, we double-checked the validity of our analysis by setting C22 to a fixed value and solving for C20 only. In such a way, no significant correlation is present anymore among all fitted parameters. Starting from two very low polar moment of inertia values, we computed C22 values (Extended Data Table 1) and restarted the fitting process. It should be noted that the chosen values of moment of inertia are small compared with other icy satellites (Titan25 0.3431; Enceladus26 0.335), and are closer to the gas giant planet Jupiter’s value27 (C/MR2 = 0.26939). Such low values are required to

avoid too low a negative C20 value. Our two estimations of C20 followed perfectly the more general solution where both C20 and C22 were solved for. This indicates that the gravity field of Mimas inferred from its orbit and rotational motions, and assuming a rigid interior, must have an extremely large ratio C20/C22 of about ten, which is three times larger than the hydrostatic case26. Extra tests were performed, using different dynamical modelling and adding Hubble Space Telescope data28, confirming the robustness of our results (see Methods section ‘Periapsis drift for a fully rigid Mimas’). We determined the periapsis drift, Δϖ , associated with Mimas’s libration to infer an orbital signal (see Methods section ‘Periapsis drift for a fully rigid Mimas’) of −5.4 ± 0.5 (10−7 rad per day, 3σ). This translates to −9.4 ± 0.9 km over the whole duration of the Cassini mission, a value which is smaller by a factor of 2 than that expected from the silicate elongated core hypothesis (−20.1 ± 0.2 km). To test the robustness of our solution under slightly different modelling scenarios, we performed many tests. In Extended Data Table 1, we also show the variation of the Mimas gravity field under a few extra different modelling assumptions. In addition, using, potential theory7, we have estimated the possible values of the core–mantle interface shape coefficients d20c and d22c from the C20 and C22 estimated. We looked at the strong values obtained translated in terms of polar and equatorial radii. Despite the fact that too large radii appeared often, all solutions imply one radius to be negative (Extended Data Fig. 2).

Periapsis drift for a fully rigid Mimas As already pointed out in 19905, the physical libration introduces a specific extra secular term in the precession of the periapsis, that can be barely masked within the fitting procedure. This term is equal to29: 2

Δϖ =

4A   3 R     J − 2C22 5 −  nt 2  a   2 e   

where a, e, n and t denote the semi-major axis, eccentricity, mean motion and time, respectively. In other words, the most important orbital signal that allows us to determine the physical libration of a moon from astrometry is this extra periapsis drift. As already experienced30, using the periapsis drift estimation above is a pretty convenient way to infer the physical libration A of a spinning and orbiting celestial object for any set of J2 (−C20) and C22 values, and vice versa, that is, any combination of the physical libration A, gravity parameters J2 and C22 should be consistent with the periapsis drift estimation. Once this periapsis drift and its uncertainty are determined from a fit of Mimas’s C20 and C22 assuming the physical libration value from ISS data4, we were then able to use it for discriminating between the various global ocean models.

Periapsis drift for three-layer Mimas In this section, we present the model used to compute the periapsis drift for a three-layer Mimas. The potential is calculated using a previous approach29, but extended to a body containing a solid crust, a fluid layer and a solid core. The solid crust and core are free to librate at different amplitudes. The librations are computed using the usual formalism31–33. The ocean potential has been computed following the method developed for the Earth34. The final expression is Δϖ =

2 ρ 4As 4As 3R  s s o ) + ( J2o − 2C22 )(5 − )(1 − s )    ( J2 − 2C22)(5 − e e ρo 2  a   ρ  4Ac c + ( J2c − 2C22 )(5 − )(1 − o )nt e ρc 

This expression depends on the librational amplitude of the shell As and core Ac, the density of each layer, the shell ρs, the ocean ρo, the core ρc,

and the Stokes coefficients of each layer that can be determined from the mean radii, densities and geometric flattening of each layer

J k2 =

8π 5 (r α − r 5 α ) 15 k k k −1 k −1

C k22 =

2π 5 (r β − r 5 β ) 15 k k k −1 k −1

the index k represents the current layer and k − 1 is the layer below the current layer where rk, αk and βk are the radius, polar flattening and equatorial flattening of the layer. Extended Data Fig. 3 shows the sensitivity of the periapsis drift to the variation of the geometric flattening. The main impact is from the α and β of the surface and then β of the ocean. The other parameters have a negligible contribution. Here we use extrema values of 7% and 8% for α and β of the crust and 10% for the other interfaces. The first two uncertainties were derived using the uncertainties on the shapes35.

Tidal dissipation and thermo-orbital evolution of Mimas Interior structure and rheology. Mimas’s interior is divided in three layers, from the centre to the surface: a rock-rich core assumed to be either a rigid and non-dissipative core or water-saturated unconsolidated dissipative core (similar to Enceladus36), an inviscid water ocean and a viscoelastic conductive ice shell. For simplicity, the rock core is assumed to have constant and uniform densities and mechanical properties (shear modulus and viscosity). In the outer ice layer, the viscosity is computed for the temperature profile solved assuming thermal diffusion only (no thermal convection). In the rock core, the complex shear modulus, μc, is determined from the effective shear modulus μeff, equal to the norm of μc and the dissipation function Q−1 equal to the μ ratio between the imaginary part and the norm of μc. Values between 107 Pa and 109 Pa and between 0.2 and 0.8 are tested for μeff and Q−1 , μ respectively. For the outer ice shell, an Andrade rheology37 is considered, characterized by constant values of elastic modulus, μE, and temperature-dependent viscosity, η(T). The thicknesses of the ice shell and ocean are computed from heat thermal balance. Computation of tidal dissipation. The viscoelastic deformation of Mimas under the action of periodic tidal forces is computed following the method of ref. 10. The Poisson equation and the equations of motion are solved for small perturbations in the frequency domain assuming a compressible viscoelastic rheology. The potential perturbation, associated displacement and stress are computed as a function of radius by integrating the radial functions associated with the radial and tangential displacements (y1 and y3, respectively), the radial and tangential stresses (y2 and y4, respectively), and the gravitational potential (y5), and a sixth radial function (y6) to account for the continuity of the gravitational potential in the elastic equivalent problem. For the deformation of the inviscid water ocean, the static simplified formulation38 is adopted relying on two radial functions, y5 and y7. The solution in the solid part (porous core and ice shell) is expressed as the linear combination of three independent solutions (yi = Ayi1 + Byi2 + Cyi3), which reduces to one solution in the liquid part. The integration of these three solutions is initiated at the centre using the analytical solutions of spheroidal oscillations for a compressible homogeneous sphere (equations 98, 99 and 100 in ref. 39). The system of six differential equations is solved by integrating the three independent solutions using a fifth-order Runge–Kutta method with adjustive step size control from the centre (radius, r = 0 km) to the surface (r = 252 km). The three coefficients, A, B and C, are determined at the surface by imposing the boundary conditions appropriate for forcing by an external tidal potential. The solutions for the six radial functions are then computed using the coefficients and the appropriate relationship at each liquid/ solid interface.

From the radial functions, y1, y2, y3 and y4, and the degree-two tidal potential at the surface, the tidal heating rate as a function of depth is computed. The global dissipation in each layer (Pcore and Pice) is then determined by integrating over the entire layer the dissipation rate computed at each radius. In addition, we also determine the glo­ bal dissipation Ptide directly from the imaginary part of the complex c Love number, I(k 2) (defined from the potential radial function, y5, c at the surface (r = Rs) : k 2 = y5 (Rs) − 1) using the classical formulation: Ptide = −

21 c (ωR ) I(k 2) Gs 2

5

e 2 where Rs is the surface radius, e the orbital ecc­

entricity, ω the orbital angular frequency and G the gravitational constant. The two approaches are used simultaneously to ensure that the computation is consistent. Heat balance and ocean evolution. The melting/cooling of the inter­nal ocean is controlled by heat production by tides in the inte­ rior and the heat transfer through the outer ice shell. The heat is transported by thermal diffusion through the outer ice shell, as the conditions for the onset of convection are not met during the melting phase. Conductive heat transfer through the ice shell is computed using temperature-dependent thermal conductivity and tidal heating. Tidal heating is self consistently computed from the viscosity profile determined from the temperature profile, for viscosity values ranging between 1014 Pa s and 1016 Pa s and activation energy equal to 50 kJ mol−1. The evolution of the ice–ocean interface is determined from the balance between heat flux from the rocky core (including tidal heating) and heat flux through the outer ice shell. The melting temperature takes into account the ammonia fraction in the liquid whose concentration involved with the ocean volume, following the para­ meterization of ref. 40. The initial ice-shell thickness is set to 100 km and the rock core radius to 95 km. The ammonia mass fraction relative to the total water mass is set to 1%, comparable to the value observed in Enceladus’s plume41. Orbital eccentricity evolution. The reduction in orbital eccentricity, e, due to tidal dissipation is computed from the total dissipated power Ptide as:

de − a(1 − e 2) = P dt eGMSMM tide with a the semi-major axis, MS Saturn’s mass and MM Mimas’s mass.

Data availability Most astrometric data are already available from refs. 6,21 and references therein. The extra astrometric data of Mimas and Tethys that were obtained from three-dimensional complex-shape modelling are available on IMCCE FTP server at ftp://ftp.imcce.fr/pub/psf.

Code availability All astrometric data derived from ISS images can be reproduced using our CAVIAR software available under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International. The software is available at www.imcce.fr/recherche/equipes/pegase/caviar. 21. Cooper, N. J. et al. The Caviar software package for the astrometric reduction of Cassini ISS images: description and examples. Astron. Astrophys. 610, A2 (2018). 22. Rambaux, N., Lainey, V., Cooper, N., Auzemery, L. & Zhang, Q. F. Spherical harmonic decomposition and interpretation of the shapes of the small Saturnian inner moons. Astron. Astrophys. 667, A78 (2022). 23. Zhang, Q. F. et al. A comparison of centring algorithms in the astrometry of Cassini imaging science subsystem images and Anthe’s astrometric reduction. Mon. Not. R. Astron. Soc. 505, 5253–5259 (2021). 24. Iess, L. et al. Measurement and implications of Saturn’s gravity field and ring mass. Science 364, aat2965 (2019). 25. Iess, L. et al. The tides of Titan. Science 337, 457–459 (2012).

26. Iess, L. et al. The gravity field and interior structure of Enceladus. Science 344, 78–80 (2014). 27. Militzer, B. & Hubbard, W. Relation of gravity, winds, and the moment of inertia of Jupiter and Saturn. Planet. Sci. J. 4, 95 (2023). 28. French, R. G. et al. Astrometry of Saturn’s satellites from the Hubble Space Telescope WFPC2. Publ. Astron. Soc. Pac. 118, 246–259 (2006). 29. Jacobson, R. A. The orbits and masses of the Martian satellites and the libration of Phobos. Astron. J 139, 668–679 (2010). 30. Lainey, V. et al. Interior properties of the inner Saturnian moons from space astrometry data. Icarus 326, 48–62 (2019). 31. Van Hoolst, T., Rambaux, N., Karatekin, Ö., Dehant, V. & Rivoldini, A. The librations, shape, and icy shell of Europa. Icarus 195, 386–399 (2008). 32. Rambaux, N., van Hoolst, T. & Karatekin, Ö. Librational response of Europa, Ganymede, and Callisto with an ocean for a non-Keplerian orbit. Astron. Astrophys. 527, A118 (2011). 33. Richard, A., Rambaux, N. & Charnay, B. Librational response of a deformed 3-layer Titan perturbed by non-Keplerian orbit and atmospheric couplings. Planet. Space Sci. 93, 22–34 (2014). 34. Xu, S. & Szeto, A. M. K. Gravitational coupling in the Earth’s interior revisited. Geophys. J. Int. 118, 94–100 (1994). 35. Thomas, P. C. et al. Shapes of the Saturnian icy satellites and their significance. Icarus 190, 573–584 (2007). 36. Choblet, G. et al. Powering prolonged hydrothermal activity inside Enceladus. Nat. Astron. 1, 841–847 (2017). 37. Castillo-Rogez, J. C., Efroimsky, M. & Lainey, V. The tidal history of Lapetus: spin dynamics in the light of a refined dissipation model. J. Geophys. Res. 116, E09008 (2011). 38. Saito, M. Some problems of static deformation of the Earth. J. Phys. Earth 22, 123–140 (1974). 39. Takeushi, H. & Saito, M. in Methods in Computational Physics Vol. 1 (ed. Bolt, B. A.) 217–295 (Academic Press, 1972).

40. Grasset, O. & Pargamin, J. The ammonia water system at high pressures: implications for the methane of Titan. Planet. Space Sci. 53, 371–384 (2005). 41. Waite, J. H. et al. Cassini finds molecular hydrogen in the Enceladus plume: evidence for hydrothermal processes. Science 356, 155–159 (2017). Acknowledgements V.L. and N.R. thank the FP7-ESPaCE European programme for funding under the agreement number 263466. G.T. acknowledges support from the ANR COLOSSe project. Q.Z. is supported by the Joint Research Fund in Astronomy (number U2031104) under cooperative agreement between the National Natural Science Foundation of China (NSFC) and Chinese Academy of Sciences (CAS). Author contributions V.L. developed and fitted to the observations the full numerical model presented for the astrometric approach. N.R. developed the librational model and provided the solutions as function of interior structure. G.T. developed the thermo-orbital model of Mimas and performed the simulations of past evolution. N.C., V.L. and Q.Z. provided extra astrometric data. B.N. ran the N-body simulations involving a high eccentric Mimas. All authors contributed to the writing of the paper. Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41586-023-06975-9. Correspondence and requests for materials should be addressed to V. Lainey. Peer review information Nature thanks David Stevenson and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Reprints and permissions information is available at http://www.nature.com/reprints.

Extended Data Fig. 1 | Reprocessing of Mimas astrometry. Here most of the satellite edge can be used for determining the center of figure of Mimas. The purple dots are the detected edge points on the image, while the turquoise

ones represent the expected shape from spherical harmonics computation. The orange curve represents Mimas’s equator. Using 3-D complex shape modelling allows a more accurate center of figure to be obtained for Mimas.

Extended Data Fig. 2 | Core radius for solid model. Solutions for the 3-D geometric axes, polar radius (rp) and equatorial radius at longitude = pi/2 (re2) as function of the equatorial radius at longitude = 0 (re1). Each point represents

an interior model where the core and mantle densities vary from [920–1100] kg m−3 and [1200–3600] kg m−3. In all cases, rp or re2 is negative. For this figure, Stoke’s coefficients are C20 = −0.101 and C22 = 0.0093.

Extended Data Fig. 3 | Sensitivity of shape parameters. This figure shows the Mimas’ libration and periapsis drift solutions for a range of equatorial and polar flattenings.

Extended Data Table 1 | Estimation of Mimas gravity field

Error bars are at 3 sigma level. The value of 50.3 arcmin in column 4 is from the control point network measurement4. When not solved for with all other parameters, expression of C22 coefficient was computed from C22 = CA/24eMR2 where C, e, M and R are the largest moment of inertia, Mimas’ eccentricity, mass and radius, respectively. On row 5, the orientation of Saturn was forced to the one of French et al. (2017). On the next line, we introduced the nutations of Saturn estimated from the JPL kernel sat427. The last line corresponds to the ocean case, where Mimas gravity field was set to its mean expected value under our ocean interior model, while no constraints was applied on the Mimas physical libration.

Ultracold field-linked tetratomic molecules https://doi.org/10.1038/s41586-023-06986-6 Received: 5 June 2023 Accepted: 15 December 2023 Published online: 31 January 2024 Open access Check for updates

Xing-Yan Chen1,2, Shrestha Biswas1,2, Sebastian Eppelt1,2, Andreas Schindewolf1,2, Fulin Deng3,4, Tao Shi4,5 ✉, Su Yi4,5,6, Timon A. Hilker1,2, Immanuel Bloch1,2,7 & Xin-Yu Luo1,2 ✉

Ultracold polyatomic molecules offer opportunities1 in cold chemistry2,3, precision measurements4 and quantum information processing5,6, because of their rich internal structure. However, their increased complexity compared with diatomic molecules presents a challenge in using conventional cooling techniques. Here we demonstrate an approach to create weakly bound ultracold polyatomic molecules by electroassociation7 (F.D. et al., manuscript in preparation) in a degenerate Fermi gas of microwave-dressed polar molecules through a field-linked resonance8–11. Starting from ground-state NaK molecules, we create around 1.1 × 103 weakly bound tetratomic (NaK)2 molecules, with a phase space density of 0.040(3) at a temperature of 134(3) nK, more than 3,000 times colder than previously realized tetratomic molecules12. We observe a maximum tetramer lifetime of 8(2) ms in free space without a notable change in the presence of an optical dipole trap, indicating that these tetramers are collisionally stable. Moreover, we directly image the dissociated tetramers through microwave-field modulation to probe the anisotropy of their wavefunction in momentum space. Our result demonstrates a universal tool for assembling weakly bound ultracold polyatomic molecules from smaller polar molecules, which is a crucial step towards Bose–Einstein condensation of polyatomic molecules and towards a new crossover from a dipolar Bardeen–Cooper–Schrieffer superfluid13–15 to a Bose–Einstein condensation of tetramers. Moreover, the long-lived field-linked state provides an ideal starting point for deterministic optical transfer to deeply bound tetramer states16–18.

Molecules exhibit a rich set of internal and external degrees of freedom, which can be fully controlled only under ultracold temperatures ( ħ /μ3 d 1 d 2, the cross-section σ can be estimated using the semiclassical formula given by51 σ=

2 d1d 2 μ . 3 ϵ 0ħ 2E

(2)

Here d1 and d2 are the dipole moments of the two colliding particles, μ is the reduced mass and E is the kinetic energy. We neglect the effect of a small ellipticity ξ and estimate the effective dipole moment of the dimers to be d 0 / 12(1 + (Δ/Ω) 2 ) . The dipole moment of tetramers is roughly twice as large as that of dimers. With that, the above formula provides an estimation for the elastic scattering rates to be 9.7 × 10−9 cm3 s−1 for dimer–tetramer and 1.9 × 10−8 cm3 s−1 for tetramer– tetramer. This implies that tens of elastic collisions can occur within the lifetime of tetramers.

Lifetime analysis For the measurements in time of flight, we verify in the absence of tetramers that the two-body loss between dimers is negligible during the hold time. Thus we fit an exponential decay with a constant offset given by the unpaired dimer number N(t) = 2NTe−Γt + ND. The offset ND is extracted from the data with ellipticity over 8°, in which the number undergoes a fast initial decay and stays constant afterwards. To investigate the collisional stability of tetramers, we also assess their lifetimes while the dipole trap remains active. Our observations indicate a combined one-body and two-body loss of the detected dimer number, and we confirm that the two-body loss arises from dimer– dimer collisions. Apart from the data near the collisional threshold ξ = 5(1)°, in which in-trap measurements are influenced by thermal dissociation, we do not detect notable additional loss of tetramers in in-trap measurements compared with those in time-of-flight experiments. The deduced inelastic collision rates are consistent with zero within the error bar. We estimate that more than ten elastic collisions can occur throughout the lifetime of tetramers, which suggests that collisions with tetramers are predominantly elastic. For measurements in a trap, we ramp up the trap depth by 50% simultaneously with the association, to compensate for the force from the inhomogeneous microwave field. The spatially varying microwave changes the dressed state energy, and thus exerts a force on the molecules that lowers the trap depth and leads to additional loss in the trapped lifetime measurements. We first measure the total number of tetramers and dimers, and then do a comparison measurement in which we remove the tetramers as described in the main text. As shown in Extended Data Fig. 3a, we observe a two-body decay in the dimer number, in contrast to the time-of-flight measurements. To account for this background loss, we first determine the two-body loss rate Γ2 and the initial dimer number ND,0 from the comparison measurement and then perform a fit of one-body plus two-body decay in which we fix Γ2 and ND,0. The fit function is given by ND(t) = 2NT,0e−Γt + ND,0/(1 + Γ2t). Extended Data Fig. 3b,c

shows that the tetramer decay in trap and in free space are similar. The extracted decay rates differ by 9(9) × 101 Hz, which we use to obtain an upper bound for the inelastic scattering rate coefficients. By assuming that the additional loss is either purely dimer–tetramer or tetramer– tetramer, we estimate the upper bounds of their inelastic collision rate coefficients to be 2(2) × 10−10 cm3 s−1 and 9(9) × 10−10 cm3 s−1, respectively. Both values are consistent with zero within the error bar. Even for the worst-case estimation, the inelastic collision rate coefficients remain orders of magnitude lower than the estimated elastic dipolar scattering rate coefficients. The lifetime of the long-range field-linked tetramers is much longer than that observed in polyatomic Feshbach molecules, which are either short lived ( Rc, we obtain α ′l ′m ′ the scattering amplitudes f αlm and the scattering cross sections α ′l ′m ′ σ αlm from the channel (αlm) to the channel (α ′l ′m′). All results are convergent for (l, ∣m∣) > 7 and Rc > 5 × 104a0. We note that a different position of the absorption boundary (for example, ra = 32a0 and ra = 64a0) does not affect the result because the wavefunction has a negligible component inside the shielding core. Without loss of generality, we concentrate on the cross-section σ 210 210 of the incident and outgoing molecules in the channel (210). When the incident energy is resonant with the tetramer state, a peak appears in

the cross-section σ 210 210, where the width of the peak is the decay rate of the tetramer. The cross-section σ 210 210 quantitatively agrees with the lineshape

σ (E ) =

2π 2

k2

2

∣ig 2 G(E ) + Sbg − 1∣ ,

(7)

where G(E) = 1/(E − Eb + iΓ/2) is the tetramer propagator, k2 = M (E − E 2) and Sbg are the incident momentum and the background scattering amplitude of molecules in the dressed state channel ∣2⟩, respectively. By fitting σ 210 210 and σ(E), we obtain the binding energy Eb and the decay rate Γ of the tetramer. We remark that for the incident and outgoing molecules in other channels α ≈ 3–7, the propagator G(E) in equation (7) ′l ′m ′ does not change. Therefore, fitting σ ααlm in a different scattering channel leads to the same binding energy Eb and decay rate Γ. For a tetramer with a small decay rate, its wavefunction ψb(r) can be obtained by solving the Schrödinger equation Heffψ b(r) = Ebψ b(r) . The single-channel model Heff = −Δ2/M + Veff(r) is determined by the effective potential15

Veff(r) =

C6

sin2 θ {1 − F ξ2(ϕ) + [1 − Fξ (ϕ)]2 cos 2θ} r6 C + 33 [3cos 2θ − 1 + 3Fξ (ϕ)sin2 θ] r

(8)

for two molecules in the dressed state channel ∣1⟩ , where Fξ (ϕ) = −sin2ξ cos2ϕ, θ and ϕ are the polar and azimuthal angles of r. The 2 2 strength C3 = d /[48πϵ 0(1 + δ r )] of the dipole–dipole interaction depends only on the relative detuning δr = ∣Δ∣/Ω, whereas the C6 term describes an anisotropic shielding potential that prevents destructive short-range collisions. Using the B-spline algorithm, we obtain the binding energy Eb and the wavefunction ψb(r) ≈ Y1−(r)φ1(r)/r of the first tetramer bound state, where Y1−(r ) = (Y11(r ) − Y1−1(r ))/ 2 . The binding energies Eb and Eb obtained from the single-channel model and the seven-channel scattering calculation agree with each other quantitatively for small ξ and Ω. For the largest ξ and Ω in Fig. 1, the relative error of Eb is less than 30%. The tetramer wavefunction in the momentum space is the Fourier transform ψb(k) = ∫dre−ik·rψb(r)/(2π)3/2 of ψb(r). For the modulation dissociation, the transition probability pk to the momentum state k is determined by the coupling strength gk = ∫ drψ *( r)∂ξ Veff(r)ψ b(r). Here, ψk(r) represents the wavefunction k of the scattering state. The coupling strength gk is primarily influenced by Y1−(k)̂ , which characterizes the angular distribution of ψb(k). This dominance arises because ∂ξVeff maintains mirror symmetry with respect to the x–y plane. Therefore, by measuring pk ≈ ∣Y1−(k)̂ ∣2, we can effectively probe the angular dependence of the tetramer state in the momentum space.

Data availability The experimental data that support the findings of this study are available from the corresponding authors upon request.

Code availability All relevant codes are available from the corresponding authors upon request. 49. Hodby, E. et al. Production efficiency of ultracold Feshbach molecules in bosonic and fermionic systems. Phys. Rev. Lett. 94, 120402 (2005). 50. Klempt, C. et al. Radio-frequency association of heteronuclear Feshbach molecules. Phys. Rev. A 78, 061602 (2008). 51. Bohn, J. L., Cavagnero, M. & Ticknor, C. Quasi-universal dipolar scattering in cold and ultracold gases. New J. Phys. 11, 055039 (2009). 52. Jochim, S. et al. Bose-Einstein condensation of molecules. Science 302, 2101–2103 (2003).

53. Avdeenkov, A. V., Bortolotti, D. C. E. & Bohn, J. L. Field-linked states of ultracold polar molecules. Phys. Rev. A 69, 012710 (2004). 54. Karman, T. Microwave shielding with far-from-circular polarization. Phys. Rev. A 101, 042702 (2020). 55. Żuchowski, P. S., Kosicki, M., Kodrycka, M. & Soldán, P. van der Waals coefficients for systems with ultracold polar alkali-metal molecules. Phys. Rev. A 87, 022706 (2013). 56. Johnson, B. R. The multichannel log-derivative method for scattering calculations. J. Comput. Phys. 13, 445–449 (1973).

Author contributions All authors contributed substantially to the work presented in this paper. X.-Y.C. and S.B. carried out the experiments and together with S.E. and A.S. improved the experimental setup. X.-Y.C., S.E. and S.B. analysed the data. F.D., T.S. and S.Y. performed the theoretical calculations. T.H., I.B. and X.-Y.L. supervised the study. All authors worked on the interpretation of the data and contributed to the final paper. Funding Open access funding provided by Max Planck Society. Competing interests The authors declare no competing interests.

Acknowledgements We thank G. Quéméner for stimulating discussions. We acknowledge support from the Max Planck Society and the Deutsche Forschungsgemeinschaft under Germany’s Excellence Strategy—EXC-2111—390814868 and under grant no. FOR 2247. F.D., T.S. and S.Y. acknowledge support from the National Key Research and Development Program of China (grant no. 2021YFA0718304), National Natural Science Foundation of China (grant nos. 11974363 and 12274331) and CAS Project for Young Scientists in Basic Research (grant no. YSBR-057).

Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41586-023-06986-6. Correspondence and requests for materials should be addressed to Tao Shi or Xin-Yu Luo. Peer review information Nature thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available. Reprints and permissions information is available at http://www.nature.com/reprints.

Extended Data Fig. 1 | Dimer loss near the FL resonance. Number ND of remaining dimers as a function of ellipticity ξ. The hold time is 100 ms. The Rabi frequency of the microwave field is Ω = 2π × 29(1) MHz and the detuning is Δ = 2π × 9.5 MHz. The error bars represent the standard error of the mean of four repetitions.

Extended Data Fig. 2 | Conditions for efficient electroassociation. a, Tetramer number NT as a function of the association time. The solid blue line is a fit to a double exponential function, which captures the formation and decay of the tetramers (Methods). The error bars represent the standard error of the mean of eight repetitions. b, Conversion efficiency η as a function of the initial T/TF of the dimer gas. We use a ramp speed of 7° ms−1 for the electroassociation. The initial T/TF are extracted separately, without performing electroassociation. The error bars represent the standard error of the mean of four repetitions.

Extended Data Fig. 3 | Tetramer lifetime in trap and in time-of-flight. a, Example loss of the molecule number with (orange) and without (blue) removal of tetramers in the trap at ξ = 7(1)°. b, Normalized tetramer decay measured in time-of-flight at the same ellipticity ξ. c, Extracted tetramer number from the data in a. No notable additional loss is observed compared to b. The error bars represent standard error of the mean of ten repetitions.

Extended Data Fig. 4 | Hyperfine transitions of NaK molecules in the modulation spectra. a, Tetramer dissociation spectrum (blue) compared to background loss (orange). The background loss is measured under the same condition as the dissociation spectrum, except with a fast ramp over the FL resonance so that no tetramers are formed. The absence of loss in the background measurement suggests that the dissociation spectrum is not affected by either the hyperfine transitions or the association. This is ensured by using a small modulation amplitude of 1. 4° over a duration of 2 ms and by taking the measurements away from known hyperfine transitions. The error bars represent standard error of the mean of ten repetitions. b, Hyperfine spectrum measured with Landau–Zener sweeps in the modulation frequency and a modulation amplitude of 11° (purple) and 3. 6° (green). The transitions with the larger modulation amplitude is power broadened compared to the lower amplitude ones. The modulation frequency that we use for the tetramer dissociation spectrum are marked as vertical dashed lines. The error bars represent standard error of the mean of four repetitions.

Extended Data Fig. 5 | Tetramer dissociation patterns and their angular distribution. a,b, Images of the modulation-dissociated tetramers at different microwave field orientations. For each image, we average over the areas between the two circles to obtain the angular distribution of the optical density as shown

in c. The dashed line marks the extracted orientation of the elliptical microwave polarization, which is ϕ0 = − 50(4)° (a) and ϕ0 = 27(4)° (b). c, Angular distribution of the optical density. The upper and lower panels show the data corresponding to the images of a and b, respectively.

Extended Data Fig. 6 | Theoretical tetramer decay rate. a,b, Decay rate Γ as a function of the binding energy (a) and Rabi frequency (b). Each curve in the figures represents a calculation with a fixed ξ while varying the Rabi frequencies, resulting in a range of binding energies. The blue (green) curve corresponds to circular polarization with angular momentum projection m = 1 (m = − 1). The orange curve corresponds to ξ = 5°.

Observation and quantification of the pseudogap in unitary Fermi gases https://doi.org/10.1038/s41586-023-06964-y Received: 7 May 2023

Xi Li1,2,3,5, Shuai Wang1,2,5, Xiang Luo1,2, Yu-Yang Zhou1,2, Ke Xie1,2, Hong-Chi Shen1,2, Yu-Zhao Nie1,2, Qijin Chen1,2,3, Hui Hu1,4, Yu-Ao Chen1,2,3 ✉, Xing-Can Yao1,2,3 ✉ & Jian-Wei Pan1,2,3 ✉

Accepted: 12 December 2023 Published online: 7 February 2024 Check for updates

The microscopic origin of high-temperature superconductivity in cuprates remains unknown. It is widely believed that substantial progress could be achieved by better understanding of the pseudogap phase, a normal non-superconducting state of cuprates1,2. In particular, a central issue is whether the pseudogap could originate from strong pairing fluctuations3. Unitary Fermi gases4,5, in which the pseudogap—if it exists—necessarily arises from many-body pairing, offer ideal quantum simulators to address this question. Here we report the observation of a pair-fluctuation-driven pseudogap in homogeneous unitary Fermi gases of lithium-6 atoms, by precisely measuring the fermion spectral function through momentum-resolved microwave spectroscopy and without spurious effects from final-state interactions. The temperature dependence of the pairing gap, inverse pair lifetime and single-particle scattering rate are quantitatively determined by analysing the spectra. We find a large pseudogap above the superfluid transition temperature. The inverse pair lifetime exhibits a thermally activated exponential behaviour, uncovering the microscopic virtual pair breaking and recombination mechanism. The obtained large, temperatureindependent single-particle scattering rate is comparable with that set by the Planckian limit6. Our findings quantitatively characterize the pseudogap in strongly interacting Fermi gases and they lend support for the role of preformed pairing as a precursor to superfluidity.

A pseudogap, manifested as a depletion of spectral weight in the normal-state single-particle energy spectrum7,8, would intuitively arise from strong pair fluctuations, as a precursor of coherent pair condensation beyond the standard Bardeen–Cooper–Schrieffer (BCS) theory3. However, in cuprate superconductors, the confirmation of such a simple scenario is hindered by possible competing quantum orders, such as the d-density wave9, the pair-density wave10 and the stripe phase11, particularly in the underdoped region12,13. Unitary Fermi gases, featuring a diverging s-wave scattering length5, provide an ideal platform and quantum simulator14 for observing the long-sought pair-fluctuation-driven pseudogap15, owing to their unprecedented controllability, purity and, in particular, the known short-range attractive interaction4,16,17. To this end, radio-frequency spectroscopy18–20 was developed to probe the pairing of fermions; in particular, with momentum resolution, the trap-averaged single-particle spectral function21–23 was measured. However, the experimental resolution was insufficient to resolve the two expected quasiparticle branches. Instead, only a very broad spectral response was observed, with the dispersion exhibiting a back-bending behaviour at wavenumber k larger than the Fermi wavenumber kF21,22. This back-bending behaviour was attributed to a BCS-like quasiparticle dispersion, and it provided a primitive indication of a pseudogap22. However, there has been an on-going debate regarding this interpretation16,17,24, because the back-bending at k > kF could

occur in any short-range interacting Fermi systems25 and the observed spectra have also been argued to be interpretable by a Fermi-liquid description16,24,26. In this work, we prepare a homogeneous unitary Fermi gas of 6Li atoms in a cylindrical box trap27–30, and we develop a new momentum-resolved microwave spectroscopy to probe the fermion spectral function A(k, ω). On the one hand, the box trap eliminates the trap inhomogeneity which prohibited the extraction of a homogeneous spectral function in previous experiments. On the other hand, the microwave spectroscopy removes the complications associated with final-state interactions19,31, such as spectral shift and broadening, in the widely adopted radio-frequency spectroscopy19,32, particularly for 6Li atoms. Here the final atoms in the high-lying hyperfine level are essentially free from interactions with the initial atoms in the many-body system. Although the microwave transition is sensitive to magnetic field variations, we are able to achieve an ultrahigh stability of the magnetic field and thus realize this unique microwave transition with high energy-resolution. We observe simultaneously the two BCS-like quasiparticle branches in A(k, ω), both below and above the superfluid transition temperature Tc, which is a direct manifestation of pairing. The obtained large Δ/EF above Tc unambiguously reveals the existence of a pair-fluctuationdriven pseudogap, where Δ is the pairing (pseudo)gap and EF is the Fermi energy. The pseudogap persists up to the highest temperature (1.51Tc)

1 Hefei National Research Center for Physical Sciences at the Microscale and School of Physical Sciences, University of Science and Technology of China, Hefei, China. 2Shanghai Research Center for Quantum Science and CAS Center for Excellence in Quantum Information and Quantum Physics, University of Science and Technology of China, Shanghai, China. 3Hefei National Laboratory, University of Science and Technology of China, Hefei, China. 4Centre for Quantum Technology Theory, Swinburne University of Technology, Melbourne, Victoria, Australia. 5 These authors contributed equally: Xi Li, Shuai Wang. ✉e-mail: [email protected]; [email protected]; [email protected]

288 | Nature | Vol 626 | 8 February 2024

a |4〉

State 1–3 1–4 3–4 1/(kFa) 0 ~230 ~61

0

–500

I˜(ΔZ)h/EF

F = 3/2

F = 1/2

MW |3〉 |1〉

0.4

0.2

–1,000 100

Z(B0)

|4〉 0.6

500 E/h (MHz)

trap with diameter 84 μm and height 41 μm, as depicted in Fig. 1b. For a homogeneous unitary Fermi gas at Tc ≈ 0.17TF, where TF = EF/kB is the Fermi temperature (see, for example, the in situ image in Fig. 1b), the achieved density is n ≈ 1.09 × 1013 cm−3, which yields a large EF = ħ2kF2/(2m) ≈ 2πħ × 39.5 kHz. Here m is the atomic mass and kF = (3π2n)1/3. In our momentum-resolved microwave spectroscopy, atoms in level |3⟩ are transferred to an initially unoccupied hyperfine level |4⟩ ≡ |F = 3/2, mF = −1/2⟩ using a Gaussian-shaped microwave pulse with

c

1,000

300 500 Magnetic field (G)

|3〉 |1〉 689.68

2

0 –8

–4

0 4 ΔZ/2π (kHz)

8

b Image

B MW

y Feshbach coils

Normalized OD

z x

1

0

Fig. 1 | Experimental scheme. a, Magnetic field dependence of the 2 2S1/2 ground state of 6Li atoms. The dashed line marks the applied |3⟩ → |4⟩ microwave transition at the Feshbach resonance B0 = 689.68 G. The inset table shows the interaction parameter 1/(k Fa) for |1⟩ – |3⟩, |1⟩ – |4⟩ and |3⟩ – |4⟩ spin mixtures. b, Sketch of the experimental set-up. Left, the cylindrical box trap consists of a tube and two sheets of 532 nm laser beams, which are generated by two spatial light modulators. The magnetic field is generated by a pair of Feshbach coils. The microwave pulse travelling along the horizontal plane is generated by an antenna. Right, the incident imaging laser along the vertical direction gives an in situ image of the unitary Fermi gas at Tc. c, The Fourier-limited microwave (MW) spectrum of a non-interacting Fermi gas. Each data point and error bar represents the mean and standard error of the mean, calculated from approximately 30 independent measurements. The dark green solid line represents the calculated Fourier-limited spectrum, determined by the microwave pulse. The inset shows a sketch of the measurement. OD, optical density.

that we have achieved, which indicates a sizable pseudogap window in the unitary Fermi gas between pairing onset temperature T* and Tc (refs. 33–35). Furthermore, by analysing the energy distribution curve (EDC) in A(k, ω), a thermally activated exponential T-dependence of the inverse pair lifetime Γ0 is revealed, with an activation energy of 2Δ0, where Δ0 is the low-T pairing gap. Owing to the diverging scattering length, we observe a large single-particle scattering rate Γ1 ≈ 0.3EF/ħ ≈  1.7kBT/ħ at Tc, where kB and ħ are the Boltzmann and reduced Planck constants, respectively. Our result settles the long-standing debate about the existence of a pseudogap in unitary Fermi gases, and it provides crucial information for establishing a proper theory of strongly interacting Fermi superfluids. It also supports pairing as a possible origin of the pseudogap in high-Tc superconductors, within the framework of preformed-pair superconductivity theory.

Experimental scheme and set-up The experimental procedure of creating a homogeneous unitary Fermi gas at a given temperature follows our previous work30, except for two main distinctions (Methods). First, 6Li atoms are equally prepared in the two hyperfine levels |1⟩ ≡ |F = 1/2, mF = 1/2⟩ and |3⟩ ≡ |F = 3/2, mF = −3/2⟩ at the magnetic field B0 = 689.68 G (that is, the unitary limit), where F is the total atomic spin and mF is the projection along the magnetic-field axis36. Second, we engineer a cylindrical box

2

an intensity of VI(t) = V0e−t /(2τ ) at the time interval −tp/2 ≤ t ≤ +tp/2. Here V0 = ħΩm/2 is the maximal coupling strength of the pulse, characterized by a peak Rabi frequency Ωm of approximately 2π × 1.35 kHz, tp = 850 μs is the pulse duration and τ = 180 μs represents the 1/ e half-width of the pulse. We then immediately turn off the box trap, let the gas expand ballistically in the residual magnetic curvature for 5 ms and finally measure the density distribution n2D(x, y) of atoms in level |4⟩ using absorption imaging with a high signal-to-noise ratio (SNR), which results in a high wavenumber resolution of 0.04 μm−1 (that is, approximately 0.006kF). We determine the transferred atom number N4(Δω) =  ∬ n2D(x, y) dxdy as a function of the microwave frequency offset Δω = ωMW − ω(B0), where ωMW is the absolute microwave frequency and ω(B0) ≈ 2π × 1,787.398 MHz corresponds to the Zeeman energy splitting E3–4(B0) = ħω(B0) between the |3⟩ and |4⟩ hyperfine levels. According to Fermi’s golden rule, within the linear response, N4(Δω) is 2 proportional to N3Ωm τ, where N3 is the initial atom number in level |3⟩ (Methods). Thus, we define a dimensionless microwave spectrum 2 Ĩ(Δω) = (N4(Δω)/N3)(2EF/(π3/2ħΩm τ)), which is normalized in accordance with ∫ Ĩ(Δω)d(ħΔω/EF) = 1. The use of hyperfine level |4⟩ as the final state brings two key advantages owing to the negligible s-wave scattering lengths a14 and a34: (1) the final-state interaction can be neglected, for example, kFa14 ≈ 0.0044 ≈ 0; and (2) the isotropic momentum information n(k, Δω) carried by the outcoupled atoms in level |4⟩ can be accurately extracted through ballistic expansion. More specifically, we perform an inverse Abel transform21 on the measured n2D(x, y) to reconstruct the full density distribution n3D(r) (Extended Data Fig. 4) and then obtain the momentum distribution by means of appropriate variable substitution (Methods), due to the scaling r ∝ k satisfied in the ballistic expansion. Owing to the negligible momentum of the microwave photon37, we have n(k, Δω) ∝ A(k, ω)f(ω) (Methods), where the single-particle excitation energy ω = ϵk/ħ − Δω is offset by the free-particle kinetic energy ϵk = ħ2k2/(2m), as a result of energy conservation. Thus, A(k, ω) can be determined by performing proper transformations on n(k, Δω) and dividing by the Fermi function f(ω) = 1/(e(ħω− μ)/(kBT )  + 1), where the chemical potential μ can be accurately obtained from the known thermodynamic equation of state (EoS)30,38 for the given reduced temperature T/TF. Unlike the radio-frequency transition |3⟩ → |2⟩ used in previous experiments20,32, the realization of a high energy-resolution microwave transition |3⟩ → |4⟩ represents a great experimental challenge. This is because the energy gradient dE3–4(B)/dB at B0 is approximately 2πħ × 2.79 MHz G−1. In comparison with dE3–2(B)/dB ≈ 2πħ × 8.48 kHz G−1, we need to significantly improve the magnetic field stability by at least two orders of magnitude to the level of 0.1 ppm (parts per million) within the short time interval of tp. This is achieved by developing and combining several techniques of magnetic field stabilization (Methods). To evaluate the magnetic field stability at B0, two complementary measurements are implemented for a non-interacting Fermi gas, where all atoms are initially prepared in level |3⟩. As shown in Extended Data Fig. 3, Rabi oscillations between levels |3⟩ and |4⟩ are measured, from which an unprecedented small uncertainty δB ≈ 20 μG is inferred (Methods). The measured microwave transition |3⟩ → |4⟩, as depicted in Fig. 1c, agrees well with the calculated Fourier-limited spectrum. Fitting the data to a Gaussian function yields a very high energy-resolution of 1.51(5) kHz (that is, approximately 0.038EF). Nature | Vol 626 | 8 February 2024 | 289

a

b |4〉 0.37

0.08

1.35 0

0.77Tc 0.5

1

1

0.60

0.99

0.1

0.5

0

1.51Tc 0

1.5

0.5

d 1

1 0

h ¯ ΔZ/EF

1.0

1.5

¯hZ/EF

0

1.0

0.15

0

A(k,Z)k2 (a.u.)

c

h ¯ ΔZ/EF

–0.17

|3〉 |1〉

OD 0.92

0.5

0

¯hZ/EF

MW 0.1

A(k,Z)k2 (a.u.)

I˜(ΔZ)

|3〉 |1〉

OD

MW

0

|4〉

1.0

I˜(ΔZ)

1.0

0

0

–1

0.77Tc 0

0.5

k/kF

1.0

Fig. 2 | Microwave spectra at 0.77Tc and 1.51Tc. a, The dimensionless microwave rate Ĩ(Δω) = (EF/(ħN3))Σk A(k, ω)f(ω) as a function of the normalized frequency offset ħΔω/EF at T = 0.77Tc. b, Ĩ(Δω) at 1.51Tc. Each data point (blue circle) represents an average of approximately 100 independent measurements with the standard statistical error being calculated. In each plot, the left inset shows the experimental scheme, where the broadened energy levels (continuous spectra) of |1⟩ and |3⟩ are due to strong interatomic interactions. The dashed ellipse denotes condensed pairs below Tc. The right inset exhibits the measured n 2D(x, y) at four exemplary normalized frequency offsets, to highlight a pronounced Δω dependence. c, Intensity plots of the momentum-resolved microwave spectra A(k, ω)k 2 at T = 0.77Tc as a function of the momentum k = |k|

Fermion spectral function The ability to create a homogeneous unitary Fermi gas with a large EF and the realization of momentum-resolved microwave spectroscopy with a negligible final-state effect enable us to measure A(k, ω) in previously inaccessible parameter regimes. We first consider two limiting cases, corresponding to the lowest (0.77Tc) and the highest temperatures (1.51Tc). In comparison with a similar recent measurement32 but using the |3⟩ → |2⟩ transition (with a stronger final-state interaction kFa23 ≈ 0.2), two distinctive features are observed in our momentum-integrated microwave spectra (Fig. 2a,b): (1) the spectral width Ew, that is, the full-width at half-maximum of the spectrum, is smaller than in previous results, which indicates a larger spectroscopic pair size ξ, by means of the relationship ξ2 ≈ ħ2/(mEw) (ref. 19); furthermore, our measured Ew increases by only approximately 0.2EF as T increases from 0.77Tc to 1.51Tc, which implies that the pair size does not change much over a wide temperature range, in particular at T > Tc; and (2) the peak position Ep of the spectrum39, which is mainly determined by Δ and the Hartree energy U, remains almost unchanged with Ep ≈ 0.56EF (0.53EF) at 0.77Tc (1.51Tc). This suggests that, as T increases, |U| may increase to compensate for the decrease in Δ. The weak temperature dependence of both Ew and Ep implies a smooth change in the 290 | Nature | Vol 626 | 8 February 2024

–1

1.51Tc 0

0.5

k/kF

1.0

1.5

(normalized to kF) and the single-particle excitation energy ω (normalized to EF). d, A(k, ω)k 2 at 1.51Tc. Note that we do not show the data point with signal intensity less than approximately 0.005 times the maximal spectral intensity across all temperatures, which is essentially the background noise and sets the boundary of the intensity plot. Here the circles are the maximal spectral response Emax(k). At T = 0.77Tc in c, where the two branches are clearly distinguishable, we show Emax(k) of the lower branch only. The grey solid lines are fitting curves to the loci of Emax(k) using E (±) k in c and E k in d. For comparison, the free-particle dispersion relation ϵ k is shown as the grey dashed line. Scale bars, 0.5 mm (a, b).

many-body pairing gap Δ across Tc. We note that the spectral response falls off as Ĩ(Δω → ∞) ∝ C/(Δω)3/2 at ħω ≫ EF, where C is Tan’s contact25,32,40 characterizing short-range pairing correlations. However, a quantitative verification of such a (Δω)−3/2 decay is difficult here because a high SNR measurement can be achieved only in the range of ħΔω  kF,

0.8

¯hZ/EF

0

A(k,Z)k2 (a.u.)

1

0

–0.8

0.90Tc

0.97Tc

1.00Tc

¯hZ/EF

0.8

0

–0.8

1.04Tc 0

0.5

1.0

1.11Tc 0

0.5

k/kF

1.0

1.23Tc 0

0.5

k/kF

1.0 k/kF

Fig. 3 | Momentum-resolved microwave spectra at various temperatures across the superfluid transition. In each panel, the spectral intensity is normalized to the maximal intensity. The grey dashed lines denote the free-particle dispersion. The grey circles represent the maximal spectral

response Emax(k). For T ≤ 1.11Tc, the two branches are well separated; thus, we show Emax(k) of the lower branch only. The grey solid lines are fitting curves, as described in the text.

which has caused serious doubts about the pairing interpretation of the back-bending behaviour25,26. With a much-improved SNR in a homogeneous setting, we observe the back-bending at k ≈ 0.93kF, which is lower than kF, as expected. Here we quantitatively extract the pairing gap Δ, by identifying the locus of the maximal spectral response Emax(k) of the lower excitation branch (see circles in Fig. 2c) with a BCS-like dispersion E k(−) (Methods)39

In Fig. 3, the back-bending of the lower branch can be observed for temperatures up to 1.11Tc, and the fitted E (−) k are shown as solid lines. Here the BCS-like dispersion of equation (1) is assumed to be applicable39 above Tc. The obtained pairing gap Δ(T) is plotted as the red triangles in Fig. 4a, while m* and U are listed in Extended Data Table 1 (Methods). These results are compared with quantum Monte Carlo calculations43 in Extended Data Fig. 8. We find a smooth evolution of Δ(T) across Tc, with Δ(Tc) = 0.280(3)EF and Δ(1.11Tc) = 0.195(3)EF. At 1.23Tc, we can no longer discern the back-bending in A(k, ω), due to the merging of the two branches and the thermal broadening and reduction in the spectral response of the lower branch. The two quasiparticle branches with BCS-like dispersions E (±) k still exist; as k increases across the Fermi surface, the maximal spectral response evolves from the lower to the upper branch, resulting in an S-shaped dispersion, which is more pronounced than that at 1.51Tc. The S-shaped dispersion can be described using a simple phenom(−) (−) enological model: Ek = αk E (+) k + βk E k with α k + βk = 1. We have Ek ≈ E k (+) at small k, and Ek ≈ E k at large k (Methods). Fitting Ek to the data (solid lines for 1.51Tc and 1.23Tc in Figs. 2 and 3) yields Δ(1.23Tc) = 0.15(4)EF and Δ(1.51Tc) = 0.05(4)EF. We emphasize that the existence of a pseudogap at T ≥ 1.23Tc is also manifested as the spectral weight suppression and the presence of a saddle point near the Fermi surface in the contour plot of A(k, ω) (Extended Data Fig. 5). Using linear extrapolation of Δ(T) down to zero, we roughly estimate an onset pairing temperature T* ≳ 1.7Tc ≈ 0.29TF.

2

E (±) k = μ±

  ħ 2k 2 2  ∗ + U − μ  +Δ 2 m  

(1)

(−) Here E (+) k and E k are the quasiparticle energy dispersions in the upper and lower branches, respectively, and m* is the effective mass. The Hartree energy U leads to a downshift of the spectrum from the free-particle dispersion (dashed line in Fig. 2c). The locus of Emax(k) can be fitted perfectly to equation (1), as shown by the lower solid line in Fig. 2c. We obtain Δ = 0.396(2)EF, m* = 1.118(6)m and U = −0.374(3)EF. Using these quantities, E (+) k is plotted as the upper solid line in Fig. 2c, which reasonably describes the observed upper quasiparticle branch. Note that our Δ is slightly smaller than the most recent experimental result of Δ = 0.47(1)EF from Bragg spectroscopy42. At sufficiently high T, one expects a single excitation branch with a quadratic dispersion39,41 in A(k, ω). We indeed observe a single upward branch for T = 1.51Tc, as shown in Fig. 2d. However, the locus of Emax(k) exhibits an S-shaped dispersion, which is particularly evident near k ≈ kF. We then investigate how A(k, ω) evolves across Tc and determine the T-dependent pairing gap Δ(T). As shown in Fig. 3, over the temperature range of 0.90Tc ≤ T ≤ 1.23Tc, the lower and upper branches are clearly distinguishable, with their relative spectral weights varying smoothly across Tc. More importantly, the suppression of spectral weight near the Fermi surface is unequivocally revealed for the normal state at T ≥ Tc. This is the smoking gun of a pseudogap in a unitary Fermi gas, without the need to invoke any specific microscopic theories, which rules out a possible Fermi-liquid description of the normal state of the unitary Fermi gas.

Fermion self-energy and EDCs Next, we extract from EDCs the fermion self-energy Σ(k, ω), which is a key quantity that characterizes the many-body interaction effects of a unitary Fermi gas (or any interacting system). The evolution of EDCs as a function of k at all T is shown in Extended Data Fig. 7, where the back-bending is identified for T ≤ 1.11Tc, with the turning point k*= 2m*(μ − U ) /ħ. In Fig. 4b, we analyse the normalized EDCs at k ≈ k* ≈ 0.93kF. Remarkably, there are two sharp peaks at ħω = E− and E+ in the EDCs for T  μ. The error bars represent uncertainties

induced by errors in μ and T through the factor of 1/f(ω). c, The inverse pair lifetime Γ0 (orange circles) and single-particle scattering rate Γ1 (green squares) as a function of T/Tc. The solid line corresponds to a thermally activated exponential fitting Γ0 ∝ exp(−Ea /(k BT)) to the data at T ≤ 1.04Tc. The inset shows the thermally activated behaviour over a broad temperature range. The vertical error bars in a and c represent one standard error, obtained from curve fitting, whereas the horizontal error bars are the temperature uncertainties.

quasiparticle energies. Thus, in the superfluid phase, the pairing gap Δ can be easily read off as the half separation of two peaks (E+ − E−)/2. Note that the maximal suppression of the spectral weight occurs at ħω ≈ μ. The temperature dependence of the EDC is evident as T increases: (1) the two quasiparticle peaks move towards each other, resulting in a slow decrease in Δ; and (2) the peak width increases gradually, accompanied by an increasing intragap spectral weight. At Tc, the two peaks are significantly broadened, leaving only a weak suppression of the spectral weight near the Fermi level. For higher temperatures T = 1.04Tc and 1.11Tc in the normal state, quasiparticle peaks cannot be identified in the EDCs, and we observe a flat top instead. We attribute the disappearance of the double-peak structure to the significantly increased peak width, which becomes comparable to or larger than the size of the pseudogap at T > Tc. The measured EDCs can be described using a minimal representation of the fermion self-energy44, which has been widely adopted in the analysis of A(k, ω) in high-Tc superconductors: that is,

formula and EDC fitting, yield nearly the same Δ at all temperatures. This ensures the quantitative accuracy of the obtained pseudogap size, and thus provides an experimental benchmark for many-body theories. We mention that, for T ≥ 1.23Tc, fitting the data to equation (2) becomes challenging; the spectral peaks become so broad and heavily overlapped that a crucial portion of the data is absent for both peaks at high ω. We present Γ0 (orange circles) and Γ1 (green squares) in Fig. 4c. The inverse pair lifetime Γ0 exhibits a rapid increase with T below Tc, following a thermally activated exponential behaviour, ħΓ0 ∝ exp(−Ea/ (kBT)), where Ea = 2Δ0 and Δ0 = 0.39(4)EF is the fitted low-T pairing gap. Note that the obtained Δ0 is consistent with our measured pairing gap at 0.77Tc and a many-body theoretical calculation45. This result can be physically understood within a simple picture: Γ0 is dominated by a virtual pair breaking and recombination process, which requires an excitation energy of 2Δ0. The single-particle scattering rate Γ1 exhibits a rather weak temperature dependence across Tc. This seems to suggest that the single-fermion scattering process is insensitive to the fermion pairing, and thus is largely unchanged in the experimental temperature regime. Indeed, a similar (but slightly stronger) T-dependence of Γ1 has also been observed in high-Tc superconductors46. We find Γ1 ≈ 0.3EF/ħ at Tc, approximately 1.7 times the Planckian scattering limit6 (that is, kBT/ħ). This large Γ1 is presumably due to the diverging scattering length in the unitary limit. Moreover, the viscous relaxation rate τη−1 of a unitary Fermi gas in the normal state is given by the ratio of pressure P to shear viscosity η. Near Tc, P ≈ 0.22nEF and η ≈ 1.2nħ, which leads to τη−1 ≈ 0.2EF/ħ, in reasonable agreement with Γ1 ≈ 0.3EF/ħ, where Γ1 is taken as the transport relaxation rate30.

Σ (k, ω) =

Δ2 − iħΓ1 + (ξ (k) − ϵ k + μ) ħω − μ + ξ (k) + iħΓ0

(2)

where ξ(k) ≡ ħ2k2/(2m*) + U − μ with ξ(k*) = 0. Γ0 is responsible for the broadening of quasiparticle peaks due to the finite pair lifetime, and Γ1 is the single-particle scattering rate that characterizes the contribution from the ‘incoherent’ background37. Using equation (2), we fit the data with the explicit form of spectral function A(k, ω) = −ImΣ(k, ω)/ (π((ħω − ϵk − ReΣ(k, ω))2 + (ImΣ(k, ω))2)). As shown in Fig. 4b, the fitting curves (solid lines) agree well with the experimental data, which enables us to extract Δ, Γ0 and Γ1. The obtained Δ is denoted by blue squares in Fig. 4a. We find Δ(1.11Tc) = 0.195(3)EF, which agrees well with that extracted from the dispersion fitting (red triangles). Indeed, the two independent approaches, that is, dispersion fitting with the BCS-like 292 | Nature | Vol 626 | 8 February 2024

Summary We provide conclusive evidence for the existence of a pseudogap, which originates from strong pair fluctuations, in a homogeneous

unitary Fermi gas through momentum-resolved microwave spectroscopy, in the absence of the final-state effect. We obtain a quantitatively accurate T-dependent pairing gap Δ(T) by both analysing the quasiparticle energy dispersion and fitting the EDC with a minimal fermion self-energy model. Moreover, the inverse pair lifetime Γ0 and the single-particle scattering rate Γ1 are extracted, which are two essential quantities for characterizing the microscopic interaction processes in strongly interacting quantum systems. Our findings not only demonstrate that a many-body pairing pseudogap phase is a precursor to superfluidity, but they also provide valuable microscopic details of unitary Fermi gases in the superfluid and normal phases. The quantitative theoretical explanation of the observed pseudogap constitutes a challenge for microscopic quantum many-body theories. Furthermore, our momentum-resolved microwave spectroscopy offers a powerful tool for probing many elusive quantum phases of strongly interacting fermions, such as pseudogap and d-wave superconductivity in the doped repulsive Fermi–Hubbard model14,47–49 and the Fulde–Ferrell– Larkin–Ovchinnikov state50 in a spin-polarized Fermi gas.

Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06964-y. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17.

18. 19.

Ding, H. et al. Spectroscopic evidence for a pseudogap in the normal state of underdoped high-Tc superconductors. Nature 382, 51–54 (1996). Loeser, A. G. et al. Excitation gap in the normal state of underdoped Bi2Sr2CaCu2O8+δ. Science 273, 325–329 (1996). Chen, Q., Stajic, J., Tan, S. & Levin, K. BCS–BEC crossover: from high temperature superconductors to ultracold superfluids. Phys. Rep. 412, 1–88 (2005). Giorgini, S., Pitaevskii, L. P. & Stringari, S. Theory of ultracold atomic Fermi gases. Rev. Mod. Phys. 80, 1215–1274 (2008). Chin, C., Grimm, R., Julienne, P. & Tiesinga, E. Feshbach resonances in ultracold gases. Rev. Mod. Phys. 82, 1225–1286 (2010). Zaanen, J. Why the temperature is high. Nature 430, 512–513 (2004). Micnas, R., Ranninger, J. & Robaszkiewicz, S. Superconductivity in narrow-band systems with local nonretarded attractive interactions. Rev. Mod. Phys. 62, 113–171 (1990). Trivedi, N. & Randeria, M. Deviations from Fermi-liquid behavior above Tc in 2D short coherence length superconductors. Phys. Rev. Lett. 75, 312 (1995). Chakravarty, S., Laughlin, R. B., Morr, D. K. & Nayak, C. Hidden order in the cuprates. Phys. Rev. B 63, 094503 (2001). Fradkin, E., Kivelson, S. A. & Tranquada, J. M. Colloquium: theory of intertwined orders in high temperature superconductors. Rev. Mod. Phys. 87, 457–482 (2015). Kivelson, S. A. et al. How to detect fluctuating stripes in the high-temperature superconductors. Rev. Mod. Phys. 75, 1201–1241 (2003). Keimer, B., Kivelson, S. A., Norman, M. R., Uchida, S. & Zaanen, J. From quantum matter to high-temperature superconductivity in copper oxides. Nature 518, 179–186 (2015). Damascelli, A., Hussain, Z. & Shen, Z.-X. Angle-resolved photoemission studies of the cuprate superconductors. Rev. Mod. Phys. 75, 473–541 (2003). Bloch, I., Dalibard, J. & Nascimbene, S. Quantum simulations with ultracold quantum gases. Nat. Phys. 8, 267–276 (2012). Stajic, J. et al. Nature of superfluidity in ultracold Fermi gases near Feshbach resonances. Phys. Rev. A 69, 063610 (2004). Zwerger, W. (ed.) The BCS–BEC Crossover and the Unitary Fermi Gas (Springer, 2012). Randeria, M. & Taylor, E. Crossover from Bardeen–Cooper–Schrieffer to Bose–Einstein condensation and the unitary Fermi gas. Annu. Rev. Condens. Matter Phys. 5, 209–232 (2014). Chin, C. et al. Observation of the pairing gap in a strongly interacting Fermi gas. Science 305, 1128–1130 (2004). Schunck, C. H., Shin, Y., Schirotzek, A. & Ketterle, W. Determination of the fermion pair size in a resonantly interacting superfluid. Nature 454, 739–743 (2008).

20. Murthy, P. A. et al. High-temperature pairing in a strongly interacting two-dimensional Fermi gas. Science 359, 452–455 (2018). 21. Stewart, J. T., Gaebler, J. P. & Jin, D. S. Using photoemission spectroscopy to probe a strongly interacting Fermi gas. Nature 454, 744–747 (2008). 22. Gaebler, J. P. et al. Observation of pseudogap behaviour in a strongly interacting Fermi gas. Nat. Phys. 6, 569–573 (2010). 23. Feld, M., Fröhlich, B., Vogt, E., Koschorreck, M. & Köhl, M. Observation of a pairing pseudogap in a two-dimensional Fermi gas. Nature 480, 75–78 (2011). 24. Mueller, E. J. Review of pseudogaps in strongly interacting Fermi gases. Rep. Prog. Phys. 80, 104401 (2017). 25. Schneider, W. & Randeria, M. Universal short-distance structure of the single-particle spectral function of dilute Fermi gases. Phys. Rev. A 81, 021601 (2010). 26. Nascimbène, S. et al. Fermi-liquid behavior of the normal phase of a strongly interacting gas of cold atoms. Phys. Rev. Lett. 106, 215303 (2011). 27. Gaunt, A. L., Schmidutz, T. F., Gotlibovych, I., Smith, R. P. & Hadzibabic, Z. Bose–Einstein condensation of atoms in a uniform potential. Phys. Rev. Lett. 110, 200406 (2013). 28. Mukherjee, B. et al. Homogeneous atomic Fermi gases. Phys. Rev. Lett. 118, 123401 (2017). 29. Baird, L., Wang, X., Roof, S. & Thomas, J. E. Measuring the hydrodynamic linear response of a unitary Fermi gas. Phys. Rev. Lett. 123, 160402 (2019). 30. Li, X. et al. Second sound attenuation near quantum criticality. Science 375, 528–533 (2022). 31. Baym, G., Pethick, C. J., Yu, Z. & Zwierlein, M. W. Coherence and clock shifts in ultracold Fermi gases with resonant interactions. Phys. Rev. Lett. 99, 190407 (2007). 32. Mukherjee, B. et al. Spectral response and contact of the unitary Fermi gas. Phys. Rev. Lett. 122, 203402 (2019). 33. Robaszkiewicz, S., Micnas, R. & Chao, K. A. Thermodynamic properties of the extended Hubbard model with strong intra-atomic attraction and an arbitrary electron density. Phys. Rev. B 23, 1447 (1981). 34. Nozières, P. & Schmitt-Rink, S. Bose condensation in an attractive fermion gas: from weak to strong coupling superconductivity. J. Low Temp. Phys. 59, 195–211 (1985). 35. Sá de Melo, C. A. R., Randeria, M. & Engelbrecht, J. R. Crossover from BCS to Bose superconductivity: transition temperature and time-dependent Ginzburg–Landau theory. Phys. Rev. Lett. 71, 3202–3205 (1993). 36. Zürn, G. et al. Precise characterization of 6Li Feshbach resonances using trap-sidebandresolved RF spectroscopy of weakly bound molecules. Phys. Rev. Lett. 110, 135301 (2013). 37. Chen, Q., He, Y., Chien, C.-C. & Levin, K. Theory of radio frequency spectroscopy experiments in ultracold Fermi gases and their relation to photoemission in the cuprates. Rep. Prog. Phys. 72, 122501 (2009). 38. Ku, M. J. H., Sommer, A. T., Cheuk, L. W. & Zwierlein, M. W. Revealing the superfluid lambda transition in the universal thermodynamics of a unitary Fermi gas. Science 335, 563–567 (2012). 39. Haussmann, R., Punk, M. & Zwerger, W. Spectral functions and rf response of ultracold fermionic atoms. Phys. Rev. A 80, 063612 (2009). 40. Carcy, C. et al. Contact and sum rules in a near-uniform Fermi gas at unitarity. Phys. Rev. Lett. 122, 203401 (2019). 41. Chen, Q. & Levin, K. Momentum resolved radio frequency spectroscopy in trapped Fermi gases. Phys. Rev. Lett. 102, 190402 (2009). 42. Biss, H. et al. Excitation spectrum and superfluid gap of an ultracold Fermi gas. Phys. Rev. Lett. 128, 100401 (2022). 43. Magierski, P., Wlazłowski, G., Bulgac, A. & Drut, J. E. Finite-temperature pairing gap of a unitary Fermi gas by quantum Monte Carlo calculations. Phys. Rev. Lett. 103, 210403 (2009). 44. Norman, M. R., Randeria, M., Ding, H. & Campuzano, J. C. Phenomenology of the low-energy spectral function in high-Tc superconductors. Phys. Rev. B 57, R11093 (1998). 45. Haussmann, R., Rantner, W., Cerrito, S. & Zwerger, W. Thermodynamics of the BCS–BEC crossover. Phys. Rev. A 75, 023610 (2007). 46. Kondo, T. et al. Point nodes persisting far beyond Tc in Bi2212. Nat. Commun. 6, 7699 (2015). 47. Esslinger, T. Fermi–Hubbard physics with atoms in an optical lattice. Annu. Rev. Condens. Matter Phys. 1, 129–152 (2010). 48. Hart, R. A. et al. Observation of antiferromagnetic correlations in the Hubbard model with ultracold atoms. Nature 519, 211–214 (2015). 49. Mazurenko, A. et al. A cold-atom Fermi–Hubbard antiferromagnet. Nature 545, 462–466 (2017). 50. Kinnunen, J. J., Baarsma, J. E., Martikainen, J.-P. & Törmä, P. The Fulde–Ferrell–Larkin– Ovchinnikov state for ultracold fermions in lattice and harmonic potentials: a review. Rep. Prog. Phys. 81, 046401 (2018). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law. © The Author(s), under exclusive licence to Springer Nature Limited 2024

Nature | Vol 626 | 8 February 2024 | 293

Methods State preparation As described in our previous work30, we begin with approximately 2.7 × 107 6Li atoms at a temperature of 20 μK in a transport optical dipole trap. Here the atoms are equally populated in the lowest two hyperfine levels, that is, |1⟩ ≡ |F = 1/2, mF = 1/2⟩ and |2⟩ ≡ |F = 1/2, mF = −1/2⟩. We first ramp up the magnetic field to 320 G and perform first-stage evaporative cooling by lowering the laser power from 48 W to 8 W in 3 s. Then, the magnetic field is increased to 589 G, where the s-wave scattering length between atoms in the |2⟩ and |3⟩ ≡ |F = 3/2, mF = −3/2⟩ levels is close to zero. After applying an optimized hyperbolic secant radio-frequency pulse, almost all the atoms in the |2⟩ hyperfine level are transferred to the |3⟩ hyperfine level. To prepare a pure |1⟩ – |3⟩ mixture, a 10 μs resonant optical pulse is further implemented to remove the remaining atoms in level |2⟩. Next, the magnetic field is ramped up to B0 = 689.68 G in 30 ms, where the s-wave scattering length between atoms in the |1⟩ and |3⟩ hyperfine levels diverges36. After a 2 s evaporative cooling in the transport trap, the atoms are adiabatically transferred into a disk-like trap, where two elliptical laser beams with an aspect ratio of 3.82:1 (wavelength 1,064 nm, maximal laser power 1.5 W, 1/e2 radial (axial) radius 162.5 μm (42.5 μm)) are crossed perpendicularly51. In the disk-like trap, we further cooled the atoms by ramping down the laser power to 150 mW in 0.7 s. Subsequently, the ultracold cloud is transferred into the final box trap. In this experiment, the box trap consists of a hollow cylinder and two sheets of 532 nm laser beams, which are generated by two spatial light modulators30,52. The cylindrical geometry of the box trap (diameter of 84 μm, height of 41 μm) is specially designed to facilitate the measurement of isotropic momentum distribution. To achieve good spatial mode matching during the atomic cloud transfer, we adiabatically switch on an additional 1,064 nm laser beam along the gravity direction and simultaneously increase the power of the disk-like trap from 150 mW to 2.1 W. After this adiabatic compression and shaping process, the cylindrical box trap is gradually turned on to its maximal potential depth (which is restricted by the optical damage threshold of essential components, including polarization-maintaining optical fibres and spatial light modulators) with a magnetic gradient of levitation applied to compensate for gravity, and all the 1,064 nm dipole traps are switched off in the meantime. Taking advantages of carefully designed experimental procedures and optimized experimental parameters, we prepare a homogeneous unitary Fermi gas with approximately 2.7 × 106 atoms at the highest temperature of approximately 0.26TF in the cylindrical box trap (as restricted by the attainable trap depth). To prepare a low-T unitary Fermi gas, we linearly ramp down the laser power of the box trap to different final trap depths in 0.6 s and hold the trap for an additional 0.5 s to reach thermal equilibrium. In this way, homogeneous unitary Fermi gases at various temperatures T can be prepared. Thermometry To determine the system temperature, we measure the system energy E at each given trap depth through Bragg spectroscopy in the long wavelength limit, which was developed in our previous work30. By comparing the obtained E with the known equation of state (EoS), we extract the reduced temperature T/Tc at different trap depths. We mention that, for T ≤ 1.04Tc, we adopt our EoS30, whereas for T > 1.04Tc, we use the EoS from previous measurements38. To validate the EoS thermometry, pair momentum distributions of unitary Fermi gases are further probed through the combination of an interaction quench and a matter-wave focusing technique30,53. The resulting one-dimensional (1D) momentum distributions n1D(k) at 0.99Tc, 1.00Tc and 1.02Tc are reported in Extended Data Fig. 1; they have a one-to-one correspondence with the pair momentum distributions before the quench. It is observed that, when T ≤ Tc, the distributions

are notably enhanced in the vicinity of zero momentum. This enhancement is revealed in a clearer way54 by plotting log(n1D(k)) as a function of k2, as shown in the top right-hand panel of Extended Data Fig. 1. In comparison with the momentum distribution at T > Tc, which is well fitted by a Boltzmann distribution, we find that a pronounced fraction of the momentum density suddenly emerges above the Boltzmann fit (blue and red regions in the top right-hand panel of Extended Data Fig. 1) in the case of T ≤ Tc, indicating the onset of pair condensation. Because the temperature is independently determined from the EoS thermometry, the onset of pair condensation at Tc confirms the accuracy of our temperature measurement. We note that, experimentally, the wings of the pair momentum distributions, contributed by thermal bosons, were also found to align with Maxwell–Boltzmann distributions in previous experiments, after quenching strongly interacting Fermi gases to the deep BEC regime, in both two and three dimensions54,55.

Magnetic field stabilization The stabilization of the magnetic field is crucial in many applications. At low magnetic fields, high-performance fluxgate sensors can be used for direct measurement and active stabilization of the magnetic field using a proportional integral derivative loop: for example, a remarkable root mean square level of 0.28 ppm at 10.58 G has been achieved56. At high magnetic fields, the conventional Hall probes do not exhibit the same high sensitivity. Nevertheless, with the combination of multiple techniques, a root mean square noise of 0.29 ppm at 146 G and 2.3 ppm at 1,050 G have also been reported57,58. Here we focus on minimizing the magnetic field noise at B0 = 689.68 G within the microwave duration of tp = 850 μs. As shown in Extended Data Fig. 2, the experimental platform is first enclosed with several layers of mu-metal plates. This passive magnetic field shielding provides an optimal long-term stability and uniformity of the background magnetic field. Two temperature-stabilized current sensors with a low-frequency noise level of less than 0.1 ppm and high-frequency noise at approximately 1 ppm are employed to stabilize the main current. We integrate one of the sensors into a proportional integral derivative loop to achieve the ultrastable current for magnetic field generation, which yields a field stability of around 1 ppm. However, this stability level is insufficient to achieve a Fourier-limited spectral resolution. Thus, we develop additional methods to enhance magnetic field stability. First, we measure the current for 2 s using the other current sensor and an 8.5-digit multimeter before applying the microwave pulse. This process provides us with the average value of the generated magnetic field. Subsequently, a digital proportional integral loop governs a voltage-controlled bipolar current source, which, in turn, drives a pair of compensation coils responsible for producing the necessary compensation magnetic field. This approach effectively minimizes the influence of long-term drift (or extremely low-frequency noise) in the  magnetic field. Second, we apply a series of low-pass filters to suppress the remaining high-frequency noise from the ultrastable current source. Moreover, to mitigate the impact of 50 Hz noise, we synchronize the microwave pulse with the mains electricity. This approach is very effective because tp = 850 μs is much shorter compared with the period of the mains electricity of 20 ms. Therefore, once synchronized, the impact of 50 Hz noise is negligible. To quantitatively characterize the stability of the magnetic field, two measurements on the non-interacting Fermi gas are performed. To this end, we first prepare all atoms in level |3⟩ at 689.68 G. Then, we switch off the magnetic gradient of levitation to ensure that the Fermi gases experience a homogeneous magnetic field. Subsequently, a microwave pulse is applied to drive the transition between the |3⟩ and |4⟩ hyperfine levels. Extended Data Fig. 3a shows the resonant frequency of the |3⟩ → |4⟩ microwave transition as a function of delay time td. A remaining 50 Hz noise with a peak-to-peak value of 2.82 mG is clearly exhibited. Note that we attempted to cancel the 50 Hz noise

using a feed-forward method, but did not achieve satisfactory results because: (1) the complexity of feed-forward compensation introduces additional noise; and (2) this operation can not effectively mitigate the noise stemming from the spatial environment. Extended Data Fig. 3b shows the measured atom number N4 as a function of the pulse duration t. The slowly decaying Rabi oscillations are clearly observed. Ideally, when atoms initially in level |3⟩ are driven by a resonant microwave pulse, the normalized atom number in level |4⟩ varies coherently with time as N4 = sin2 (ΩRt/2), where ΩR is the Rabi frequency. Experimentally, the amplitude of the Rabi oscillations decays exponentially with the pulse duration, as a result of the magnetic field fluctuation . Furthermore, the low-frequency noise in the magnetic field can cause the microwave transition to deviate from resonance, leading to a variation in the Rabi frequency as ΩR2 + Δd2 (t) , where Δd(t) is an effective t-dependent detuning from the slowly varying noise. For these reasons, the normalized atom number in the |4⟩ state can be written as

1 1 N4 = − e−t /τ cos(t ΩR2 + Asin2 (2πfm t + φ) ) 2 2

(3)

where τ is the coherence time of the Rabi oscillation. Here the sinusoidal function with frequency fm = 50 Hz describes the effective detuning Δd(t) from the resonant microwave transition, which is due to the remaining 50 Hz noise mentioned above. A and φ denote the amplitude and initial phase of the 50 Hz noise, respectively. The solid line in Extended Data Fig. 3b is the fitting to equation (3), which agrees well with the experimental data. From the extracted coherence time τ = 3.0(5) ms, we obtain the spectral broadening 1/(2πτ) ≈ (dE3-4/dB)δB/h = (2.79 MHz G−1) × δB related to the magnetic field fluctuation δB. Thus, the fluctuation or uncertainty in the magnetic field is approximately given by δB = (2πτ  × 2.79 MHz G−1)−1 = 19(3) μG (ref. 59).

Momentum-resolved microwave spectroscopy In our microwave spectroscopy, all the initial atoms are equally populated in the |1i⟩ and |3i⟩ states, which constitute a resonantly interacting many-body system at B0 = 689.68 G. Here |1i⟩ and |3i⟩ denote the continuous energy states of atoms in the |1⟩ and |3⟩ Zeeman hyperfine levels, respectively, which experience strong many-body interactions. Note that we use the label i to explicitly indicate the degrees of freedom related to, for example, the momentum k. At ti = −T/2, atoms in the |3i⟩ state are excited into an initially unoccupied |4⟩ hyperfine level by a microwave pulse with frequency ωMW. After a duration T, at tf = +T/2, the number of atoms in the |4⟩ level N4(Δω) is determined as a function of the frequency detuning Δω = ωMW − ω(B0) ≡ ωMW − ω0. Note that, in the main text, we used the variable tp to denote the pulse duration T, to avoid confusion with temperature. The |3i⟩ state with momentum k and energy ω can be fully characterized by the single-particle spectral function A(k, ω). Thus, the total number of atoms N3 in the |3⟩ hyperfine level can be obtained by summing the occupied single-particle spectral function over all the |3i⟩ states, N3 = ∑k∫dωA(k, ω)f(ω), where f(ω) is the Fermi–Dirac distribution function. The microwave transition can then be well described within the framework of a two-level system, that is, N4(Δω) =

T

T

∑ ∫ dω ∣⟨4∣U + 2 , − 2  ∣3i⟩∣2 A(k, ω)f (ω)

(4)

k

where the factor A(k, ω)f(ω) represents the initial occupied spectral intensity of state |3i⟩, and ⟨4| U(tf = +T/2, ti = −T/2)|3i⟩ denotes the transition amplitude from state |3i⟩ to level |4⟩. Note that the momentum of the microwave photon is several orders of magnitude smaller than the momentum of the atoms. This implies that, once transferred, the momenta of excited atoms in the |4⟩ hyperfine level are the same as the initial momenta in the |3i⟩ states, so the kinetic energy of the

atoms in the |4⟩ hyperfine level is simply given by ϵk = ħ2k2/(2m). The frequency difference ωfi between level |4⟩ and state |3i⟩ can be written as ωfi = ω0 + (ϵk/ħ − ω). We now determine the transition amplitude ⟨4|U(t f = +T/2, ti = −T/2)|3i⟩. Because atoms in the |4⟩ hyperfine level have negligible interactions with atoms in both |1i⟩ and |3i⟩ states, the atomic Hamiltonian and atom-field coupling can be described by the equations

ħωfi (|4⟩⟨4| − |3i ⟩⟨3i|) 2

H0 =

(5)

HI(t) = VI(t)(|4⟩⟨3i|e−iωMWt + |3i ⟩⟨4|e+iωMWt )

(6)

where VI(t) = ħΩ(t)/2 is the strength of atom-field coupling and Ω(t) is the Rabi frequency. In the linear response regime, the evolution operator U(tf = T/2, ti = −T/2) is given to the first order in Ω(t) by

T T  T T  1 U  , −  = U0  , −  + 2 2 2  iħ   2

T /2

T

T

∫ −T /2 U0  2 , t  HI (t) U0 t , − 2  dt

(7)

where U0(tf, ti) = e−iH0(t f − t i)/ħ = e−iω fi(t f − t i)/2|4⟩⟨4| + e+iω fi(t f − t i)/2 |3i⟩⟨3i| is the evolution operator in the absence of the transfer coupling HI(t). According to equations (5)–(7), the amplitude of the microwave transition can be written as ⟨4|U (T /2, − T /2)|3i ⟩ = =

1 iħ 1 iħ

T /2

∫ −T /2 e−iω (T /2−t )/2VI(t )e−iω fi

T /2

∫ −T /2 VI(t )ei(ω −ω fi

MWt e+iω fi(t + T /2)/2 dt

(8)

MW)t dt

To minimize the side lobes generated by a rectangular microwave pulse, we adopt a Gaussian-shaped microwave pulse in the experiment: 2 2 that is, VI(t) = V0e−t /(2τ ) . Here V0 = ħΩm/2 is the maximal coupling strength during the excitation and τ represents the 1/ e half-width of the pulse. As a result, in the time integral of equation (8), we can safely extend the interval of the integral [T/2,−T/2] to ±∞. After the Fourier transform of the Gaussian-shaped pulse, we find that +∞ 2 2 1 V ∫ e−t /(2τ )ei(ω fi− ωMW)t dt iħ 0 −∞ 2π V0τ −(ω fi− ωMW) 2τ 2 /2 e = iħ

⟨4|U (T /2, − T /2)|3i ⟩ =

(9)

By substituting equation (9) into equation (4), N4(Δω) at a given frequency ωMW can be calculated as

N4(Δω) = ∑ ∫ dω

2πV 02τ 2 ħ

k

2

e−(ω fi− ωMW)

2 2

τ

A(k, ω)f (ω)

(10)

It is important to note that the full-width at half-maximum of the microwave transition is approximately 1.51(5) kHz (Fig. 1c), which is far less than the energy spectral width of A(k, ω) of the |3i⟩ state. Therefore, we can use the asymptotic form of the delta function 2 2 τ δ(ωfi − ωMW) = lim π e−(ω fi− ωMW) τ in equation (10). This leads to τ →∞

N4(Δω) = ∑ ∫ dω

2π 3/2V 02τ ħ

k

2 π 3/2 Ωm τ = 2

∑

2

δ(ωfi − ωMW) A(k, ω) f (ω) (11)

A(k, ω) f (ω)

k

where, in the last line, we have used V0 = ħΩm/2, and ω = ϵk/ħ − Δω is the solution of the expression ωMW = ωfi = ω0 + (ϵk/ħ − ω) due to the delta function. It is easy to see that N4 in level |4⟩ satisfies

∫ dΔωN4(Δω) =

2 π 3/2 Ωm τ 2

∑ ∫ dωA(k, ω)f (ω) = k

2 π 3/2 Ωm τ N3 2

(12)

Therefore, we define a dimensionless microwave spectrum

∼ I (Δω) ≡

2EF

2 π 3/2ħΩm τ

N4(Δω) N3

(13)

which obeys the sum rule31 ∫ Ĩ(Δω)d(ħΔω/EF) = 1. It is also instructive to rewrite N4(Δω) in equation (11) in terms of the measured momentum distribution n(k, Δω): that is, N4(Δω) = ∑kn (k, Δω), such that

n(k, Δω) ≡

(14)

Ballistic expansion of atoms in the residual magnetic curvature potential In the experiment, non-interacting atoms in level |4⟩ expanded ballistically in the residual magnetic curvature potential. In the meantime, an optimized magnetic field gradient is employed to compensate for gravitational effects on atoms in level |4⟩. The corresponding Hamiltonian can be written as 3

∑

j =1

 pˆ2 1   j + mω 2j xˆ2j   2m 2   

(15)

where pˆj and xˆj ( j = x, y, z) are the momentum and position operators of an atom in the jth direction, respectively, and ωj represents the trap frequency of the magnetic potential. Note that, in general, ωj can be a complex frequency, which describes both the attractive and repulsive potentials. In the Heisenberg picture, the momentum and position operators are time dependent and satisfy the Heisenberg equation of motion ^ ^ d x^ (t) = [x^j (t), H^ ] = eiHt /ħ[x^j , H^ ]e−iHt /ħ dt j d ^ ^ iħ p^j (t) = [p^j (t), H^ ] = eiHt /ħ[p^j , H^ ]e−iHt /ħ dt

iħ

(16)

In our case, atoms can be considered as wave packets. According to Ehrenfest’s theorem, the mean values of the position and momentum operators are similar to their classical counterparts, which should obey the equation60

d 1 ⟨x^ ⟩(t) = ⟨p^j ⟩(t) dt j m d ⟨p^ ⟩(t) = − mω 2j ⟨x^j ⟩(t) dt j

(17)

For atoms in the |4⟩ hyperfine level, the residual magnetic curvature potential is attractive in the z direction (gravity direction) and repulsive in the x and y directions. By solving equation (17) with real ωz and imaginary ωx (ωy), respectively, we have61

⟨x^j ⟩(t) = ⟨x^j ⟩(0)cos(|ωj |t) + ⟨p^j ⟩(0)

sin(|ωj |t)

⟨x^j ⟩(t) = ⟨x^j ⟩(0)cosh(|ωj |t) + ⟨p^j ⟩(0)

m|ωj |

, ( j = z)

sinh(|ωj |t) m|ωj |

n(x , y , z , t ) =

(18) , ( j = x , y)

∭ ∭ n(x ′, y ′ , z ′, t = 0)fp (p′x , p′y, p′z ) δ(Ax ′ + Bp′x − x )δ(Ay ′ + Bp′y − y ) δ(Cz ′ + D p′z − z )dp′x dp′y dp′z dx ′dy ′ dz ′

=

2 π 3/2 Ωm τ A(k, ω)f (ω) 2

In the next two sections, we describe how to extract the momentum distribution n(k, Δω) and hence the single-particle spectral function A(k, Δω), where ω = ϵk/ħ − Δω.

Hˆ =

Equation (18) allows us to access the initial momentum distribution of atoms by measuring their density distribution after ballistic expansion, as described below. At the beginning of the ballistic expansion, the density distribution and momentum distribution function of atoms in the |4⟩ hyperfine level are defined as n(x, y, z, t = 0) and fp(px, py, pz), respectively. After a given expansion time t, the density distribution n(x, y, z, t) can be written as

∭ N4(Δω)δ(x ′, y ′ , z ′)

 x − Ax ′ y − Ay ′ z − Cz ′  fp  , , dx ′dy ′ dz ′ B D   B x y z  = N4(Δω)fp  , ,  B B D = N4(Δω)fp (ħk x , ħk y , ħk z )

(19)

Here the scaling factors in the three delta functions are determined by equation  (18), where A = cosh (|ω∥|t), B = sinh (|ω∥|t)/(m|ω∥|), C = cos (|ω⊥|t) and D = sin (|ω⊥|t)/(m|ω⊥|). Note that ω∥ = ωx = ωy and ω⊥ = ωz, and the scaling factor B should not be confused with the magnetic field. In the second equality of equation (19), we assume that n(x, y, z, t = 0) ≈ N4(Δω)δ(x, y, z). The validation of this approximation is twofold. On the one hand, the density distribution of atoms in the box trap is homogeneous. On the other hand, after a long-time ballistic expansion, the size of the atomic cloud is approximately ten times larger than that in the box trap. This implies that the initial homogeneous density distribution plays a negligible role in the final density distribution of atoms. Thus, n(x, y, z, t) after ballistic expansion is precisely the momentum distribution n(k, Δω) that we wish to measure under the relationship ∫ fp(k)d3k = 1. The position and momentum parameters are simply related by a scaling transformation: kx = x/(ħB), ky = y/(ħB) and kz = z/(ħD). However, there is a little complication, because we cannot directly measure a three-dimensional (3D) density distribution n(x, y, z, t). Instead, a column density n2D(x, y) =  ∫ n(x, y, z, t) dz is measured by the absorption imaging. The solution is that we can reconstruct an isotropic three-dimensional distribution n3D(r = {x, y, z}) from n2D(x, y) using an inverse Abel transformation21, under the condition that the momentum distribution is isotropic. We then obtain n(k, Δω) = n3D(r), where k = r/ (ħB). Note that we do not need to take care of the different scaling factor D related to the z direction in the reconstruction, because kz is integrated out and is independent of kx and ky. We discuss the technical details of obtaining n(k, Δω) in the next section.

Obtaining the single-particle spectral function A(k, ω) To obtain the momentum distribution of atoms in the hyperfine level |4⟩, state-selective absorption imaging is applied to measure their two-dimensional (2D) density distribution n2D(x, y) =  ∫ n(x, y, z, t)dz after 5 ms ballistic expansion (that is, t = 5 ms). To improve the SNR of n2D(x, y), not only are the imaging optics and parameters carefully optimized62, but the acquired images are also processed with the optimized fringe removal algorithm63. For each ωMW or Δω = ωMW − ω0, approximately 100 experimental runs are performed with n2D(x, y) being probed. After averaging all the acquired images, the noise in n2D(x, y) is greatly reduced, as shown in the inset of Extended Data Fig. 4a. Moreover, benefiting from the rotational symmetry in the x–y plane, n2D(x, y) can be further averaged azimuthally, leading to a high-quality n2D(r), where r =  x 2 + y 2 (for example, see Extended Data Fig. 4a). We then perform an inverse Abel transform21 to extract the three-dimensional (3D) distribution n3D(r) from this high-quality n2D(r), as shown in Extended Data

Fig. 4b. Finally, the 3D isotropic momentum distribution n(k, Δω) can be obtained by scaling n3D(r) with k = r/(ħB). As an example, we present the obtained n(k, Δω) at 0.77Tc in Extended Data Fig. 4c. We mention that, at large k, the signals are buried in the background noise. Thus, the data points (noises) with intensities less than 0.005 times the maximal spectral intensity across all temperatures are removed along the contour line, as illustrated in Extended Data Fig. 4d, which also sets the boundary of our valid data. Furthermore, given equation (14) and ω = ϵk/ħ − Δω by means of energy conservation, A(k, ω) can be obtained through dividing n(k, Δω) by the Fermi function f(ω) = 1/(e(ħω− μ)/(kBT ) + 1 ). The obtained contour plots of A(k, ω) are presented in Extended Data Fig. 5 for all temperatures.

Analysis of the energy dispersion We employ the exemplary spectrum at Tc to illustrate the procedure for determining the energy dispersion Ek(ref. 39). To quantitatively determine Ek from A(k, ω) (Extended Data Fig. 6a), we first attempt to find the peak position of each EDC, that is, the Emax(k). This method has been generally used in previous experiments21,22. However, as shown in Extended Data Fig. 6c, the peak position of the lower branch in a typical EDC near kF can be hard to locate. This is because the coupling between the two branches is significant and the signal of the lower branch is relatively weak at large k. To solve this problem, we develop a new approach to extract Emax(k). First, to reduce the coupling between the two branches, we perform a translation transformation to the frequency in A(k, ω). The transformed spectrum A(k, Δω) with Δω = ϵk/ħ − ω is displayed in Extended Data Fig. 6b, in which the two branches are well separated. Second, to enhance the relative signal intensity of the lower branch in the distribution curve, we consider the cut lines in A(k, Δω) that are perpendicular to the intensity ridge of the lower branch (see the dashed lines in Extended Data Fig. 6b). We name these cuts momentum–energy distribution curves (MEDCs). Note that, in an MEDC, the momentum is no longer fixed; instead, it linearly increases with Δω. As shown in Extended Data Fig. 6d, two peaks of the upper and lower branches are well recognized. Next, the peak maxima in MEDCs can be readily identified at low and high k. However, for MEDCs with two sufficiently coupled branches, the determination of peak maxima is difficult, and thus we need to perform a double-peak fitting to extract the peak positions. Each peak possesses an asymmetric line shape given by the modified pseudo-Voigt function64 Vpseudo (ω) =

a 1 /γ ( ω )

2

1 + (b (ω − ωmax )/γ (ω)) a 2 − (ln2) (b (ω− ω max) /γ (ω)) 2 + e γ ( ω)

(20)

Here γ(ω) ≡ (1 + e s(ω− ω max))−1, where s measures the skewness of the line shape and ωmax corresponds to the peak position of the MEDC (red star in Extended Data Fig. 6d). This leads to Emax(k) = ϵk − ħωmax. We display EDCs for a series of k, with the extracted Emax(k) being labelled as red circles for all T in Extended Data Fig. 7. The back-bending of the lower branch can be seen for T ≤ 1.11Tc. This motivates us to describe the lower-branch energy dispersion of a unitary Fermi gas with a modified BCS form39 2

m  2 E (−) k = μ −  ∗ ϵ k + Uh − μ  + Δ m  h 

(21)

where m*h and Uh denote the effective mass and Hartree energy of the hole-like quasiparticle, respectively. The upper-branch energy dispersion of the particle-like quasiparticle may be similarly described as 2 2 ∗ E (+) k = μ + (mϵ k / mp + Up − μ) + Δ . Note that we have carefully used the subscripts p and h to distinguish the effective mass and Hartree

energy shift for the two branches. This is because, in the unitary limit, we may lose the particle–hole symmetry satisfied by the two branches in the weak-coupling BCS limit. For T ≤ 1.11Tc, we use only equation (21) to fit the data of the lower branches. This is because the obtained upper branch is less accurate than the results for the lower branch: the occupied densities are greatly suppressed by the Fermi function f(ω), which lead to extremely weak signals. For T = 1.23Tc and 1.51Tc, due to the decrease in the pairing gap, the two branches are no longer well separated and the back-bending of the lower branch can hardly be observed. We find that the fitting error for Emax(k) of the lower branch near kF suddenly becomes overly large (greater than EF), and thus it is impractical to determine the back-bending, even if it still exists. In this case, the extracted Emax(k) exhibits a distinctive S-shaped dispersion as the two branches merge (Fig. 3), which resembles a crossover from a BCS hole-like dispersion at low k to a particle-like dispersion at high k. In this picture, Emax(k) might be modelled by a combined form of BCS-like hole and particle dispersions (−) Ek = αk E (+) k + βk E k

 k − km  1 )μ + = 1 + tanh( 2 σ  

2   m 2  ∗ ϵ k + Up − μ  + Δ     mp 

(22)

2   k − km  m  1  )μ −  ∗ ϵ k + Uh − μ  + Δ2  + 1 − tanh( 2 m σ h     

where km is the crossover point and σ controls the crossover rate between the two branches. At k ≪ kF, α ≈ 0 and β ≈ 1, so that Ek reduces to E k(−) . Conversely, Ek ≈ E k(+) holds at large wavenumber. Moreover, at the crossover point, αk = βk = 1/2 indicates the equal contribution of the two branches to the maximal spectral response. The fitted curves obtained by using either equation (21) (E k(−) for T ≤ 1.11Tc) or equation (22) (Ek for T = 1.23Tc and 1.51Tc) are shown as solid grey lines in Figs. 2 and 3 and agree well with the experimental data. Moreover, the fitting allows us to quantitatively determine important quantities of the unitary Fermi gas, as summarized in Extended Data Table 1. Note that the chemical potentials μ listed in Extended Data Table 1 are determined from previous measurements of the EoS30,38. It is observed that the Hartree energy Uh monotonically decreases with increasing temperature, which suggests that the temperaturedependent Hartree energy should be considered in the many-body theory. Moreover, for T = 1.23Tc and 1.51Tc, we find that m*p ≈ mh*, while the Hartree energies seem to exhibit a weak particle–hole asymmetry. However, given the low signal strength for the upper branch, this result requires further investigation. Therefore, for simplicity, in the main text, we do not mention the difference in the effective mass and Hartree shift of the two excitation branches. Finally, we compare our results with Magierski and coworkers’ quantum Monte Carlo results in Extended Data Fig. 8. Although the results seem to be in reasonable agreement for m*h and Uh, our Δ appears to persist up to a higher temperature.

Data availability The data that support the findings of this study are available at https:// doi.org/10.5281/zenodo.10115338.

51. Yao, X.-C. et al. Observation of coupled vortex lattices in a mass-imbalance Bose and Fermi superfluid mixture. Phys. Rev. Lett. 117, 145301 (2016). 52. Pasienski, M. & DeMarco, B. A high-accuracy algorithm for designing arbitrary holographic atom traps. Opt. Express 16, 2176–2190 (2008). 53. Murthy, P. A. et al. Matter-wave Fourier optics with a strongly interacting two-dimensional Fermi gas. Phys. Rev. A 90, 043611 (2014).

54. Ries, M. G. et al. Observation of pair condensation in the quasi-2D BEC–BCS crossover. Phys. Rev. Lett. 114, 230401 (2015). 55. Ketterle, W. & Zwierlein, M. W. Making, probing and understanding ultracold Fermi gases. Riv. Nuovo Cim. 31, 247–422 (2008). 56. Duan, Z.-X., Wu, W.-T., Lin, Y.-T. & Yang, S.-J. Simple and active magnetic-field stabilization for cold atom experiments. Rev. Sci. Instrum. 93, 123201 (2022). 57. Merkel, B. et al. Magnetic field stabilization system for atomic physics experiments. Rev. Sci. Instrum. 90, 044702 (2019). 58. Borkowski, M. et al. Active stabilization of kilogauss magnetic fields to the ppm level for magnetoassociation on ultranarrow Feshbach resonances. Rev. Sci. Instrum. 94, 073202 (2023). 59. Xu, X.-T. et al. Ultra-low noise magnetic field for quantum gases. Rev. Sci. Instrum. 90, 054708 (2019). 60. Cohen-Tannoudji, C., Diu, B. & Laloë, F. Quantum Mechanics, Vol. I, 522–523 (Wiley-VCH, 2020). 61. Riou, J.-F. et al. Theoretical tools for atom-laser-beam propagation. Phys. Rev. A 77, 033630 (2008). 62. Horikoshi, M. et al. Appropriate probe condition for absorption imaging of ultracold 6Li atoms. J. Phys. Soc. Japan 86, 104301 (2017). 63. Ockeloen, C. F., Tauschinsky, A. F., Spreeuw, R. J. C. & Whitlock, S. Detection of small atom numbers through image processing. Phys. Rev. A 82, 061606 (2010). 64. Stancik, A. L. & Brauns, E. B. A simple asymmetric lineshape for fitting infrared absorption spectra. Vib. Spectrosc. 47, 66–69 (2008).

Acknowledgements We acknowledge R. Qi for sharing the scattering data of lithium-6 atoms. This work was supported by the National Key R&D Program of China (grant no. 2018YFA0306501), NSFC of China (grant no. 11874340), the Innovation Program for Quantum Science and Technology (grant no. 2021ZD0301900), the Chinese Academy of Sciences (CAS), the Anhui Initiative in Quantum Information Technologies and the Shanghai Municipal Science and Technology Major Project (grant no. 2019SHZDZX01). Y.-A.C. was supported by the XPLORER PRIZE from Tencent Foundation. Author contributions Y.-A.C., X.-C.Y. and J.-W.P. conceived the research. X. Luo, Y.-Y.Z. and X.C.Y. stabilized the magnetic field. X. Li, S.W., X. Luo, Y.-Y.Z., K.X., H.-C.S., Y.-Z.N. and X.-C.Y. performed the experiment and collected the data. X. Li, S.W., Q.C., H.H., Y.-A.C., X.-C.Y. and J.W.P. contributed to the data analysis and writing of the manuscript. Competing interests The authors declare no competing interests. Additional information Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41586-023-06964-y. Correspondence and requests for materials should be addressed to Yu-Ao Chen, Xing-Can Yao or Jian-Wei Pan. Peer review information Nature thanks the anonymous reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Reprints and permissions information is available at http://www.nature.com/reprints.

Extended Data Fig. 1 | Pair momentum distributions in the vicinity of the superfluid phase transition. After ballistic expansion for one quarter of the radial trap period, the one-dimensional momentum distribution n1D(k) maps out the original pair momentum distribution of a unitary Fermi gas before the interaction quench30. Each data point corresponds to the average of

approximately 15 individual measurements. In the subplot, we show log(n1D(k)) as a function of k 2 at low momentum. The solid lines are the Boltzmann distribution fitting to the thermal wings. Pair condensation is clearly observed for T ≤ Tc.

Extended Data Fig. 2 | Schematic diagram for the magnetic field stabilization. a, The experimental platform is enclosed with several layers of mu-metal plates. b, The ultrastable magnet power supply for generating the required magnetic field of 689.68 G, which is further stabilized with an analog proportional integral derivative controller. c, The active magnetic field compensation system that includes (1) a current measurement setup,

consisting of a high precision current sensor and an 8.5-digit multimeter; (2) a proportional integral controlled compensation current source, consisting of a digital proportional integral, a waveform generator, and a voltage-controlled bipolar current source; and (3) a pair of compensation coils. d, A series of low-pass filters, which are composed of several capacitors and inductors. e, The microwave pulse is synchronized to the mains electricity.

Extended Data Fig. 3 | The residual 50 Hz noise and Rabi oscillations between the |3⟩ and |4⟩ hyperfine levels. a, The data points (red circles) are obtained by measuring the resonant frequency of the |3⟩ to |4⟩ microwave transition as a function of delay time td, i.e., the time duration between the synchronization trigger and the microwave pulse. The red solid line is the sine fitting curve with 50 Hz frequency. The red arrow indicates the moment of the

synchronization trigger. b, Data points: the normalized atom number in level |4⟩ as a function of microwave duration. Every data point corresponds to an average value of approximately 10 independent measurements. The error bar represents one standard error. The solid line is the fitting curve described by equation (3).

Extended Data Fig. 4 | Density distribution after ballistic expansion and n(k, Δω) of the unitary Fermi gas at 0.77Tc. a, The azimuthally averaged 2D density distribution after 5 ms ballistic expansion at Δω = 2π × 35 kHz. The inset displays the average result of n2D(x, y) of approximately 100 raw images. b, The

3D reconstructed distribution n3D(r), obtained by performing an inverse Abel transform to the n2D(r) in a. c, Momentum-resolved microwave spectrum n(k, Δω). d, Plot of n(k, Δω) in a logarithmic scale. The blue dashed line denotes the cutoff contour line.

Extended Data Fig. 5 | The contour plots of A(k, ω). The black dashed circles in the panels at 1.23Tc and 1.51Tc highlight the saddle region.

Extended Data Fig. 6 | Analysis of the energy dispersion. a, The contour plot of A(k, ω) at Tc. The green line is a guideline indicating an EDC slice at ∼ k F that is presented in c. b, The contour plot of A(k, Δω) at Tc, where Δω = ϵ k /ħ-ω. The gray dashed lines are the cut lines of the MEDCs and the red circles are the peak

positions of these lines. The orange line shows the MEDC used in d near kF, with a red star highlighted for the peak position. c, d, The spectral slices along the green (EDC) and the orange (MEDC) lines illustrated in a and b. The solid line in d is the fitting result of the MEDC by equation (20).

Extended Data Fig. 7 | The evolution of EDC as a function of k for various T. The red circles denote the peak energies of the left peak in EDCs (that is, the lower branch) for T ≤ 1.11Tc and of the combined single branch for T ≥ 1.23Tc. The red dashed line represents the chemical potential μ.

Extended Data Fig. 8 | Temperature dependence of Δ, m∗h and Uh. Dark red circles: our fitting results by equations (21) and (22). Dark green squares: quantum Monte Carlo results from ref. 43. Error bars represent standard errors.

Extended Data Table 1 | Temperature dependence of Δ, m*, and U

The parameters are obtained from fitting the energy dispersions with equations (21) and (22).

Evidence of superconducting Fermi arcs https://doi.org/10.1038/s41586-023-06977-7 Received: 5 May 2023 Accepted: 14 December 2023

Andrii Kuibarov1 ✉, Oleksandr Suvorov1,2, Riccardo Vocaturo1, Alexander Fedorov1,3, Rui Lou1,3 ✉, Luise Merkwitz1, Vladimir Voroshnin3,7, Jorge I. Facio4, Klaus Koepernik1, Alexander Yaresko5, Grigory Shipunov1, Saicharan Aswartham1, Jeroen van den Brink1,6, Bernd Büchner1,6 & Sergey Borisenko1,6 ✉

Published online: 7 February 2024 Open access Check for updates

An essential ingredient for the production of Majorana fermions for use in quantum computing is topological superconductivity1,2. As bulk topological superconductors remain elusive, the most promising approaches exploit proximity-induced superconductivity3, making systems fragile and difficult to realize4–7. Due to their intrinsic topology8, Weyl semimetals are also potential candidates1,2, but have always been connected with bulk superconductivity, leaving the possibility of intrinsic superconductivity of their topological surface states, the Fermi arcs, practically without attention, even from the theory side. Here, by means of angle-resolved photoemission spectroscopy and ab initio calculations, we identify topological Fermi arcs on two opposing surfaces of the non-centrosymmetric Weyl material trigonal PtBi2 (ref. 9). We show these states become superconducting at temperatures around 10 K. Remarkably, the corresponding coherence peaks appear as the strongest and sharpest excitations ever detected by photoemission from solids. Our findings indicate that superconductivity in PtBi2 can occur exclusively at the surface, rendering it a possible platform to host Majorana modes in intrinsically topological superconductor–normal metal–superconductor Josephson junctions.

The realization of topological superconductivity in new materials, which leads to robust Majorana fermions, has so far been hindered by numerous experimental challenges. Among them are the sophisticated growth of nanowire single crystals and heterostructures, as well as fine-tuning of the composition of non-stoichiometric compounds. Additionally, the rarity of spin-triplet superconductors and extremely small inverted gaps in iron-based superconductors7, proposed as intrinsic heterostructures10, are responsible for the lack of success in existing materials. Weyl semimetals bear non-degenerate spin states both in the bulk and at the surface and the doped version of either time-reversal-breaking or non-centrosymmetric Weyl semimetals can become superconducting11. The search for topological superconductivity in such systems has been focused on finding bulk superconductivity, which would lead to Majorana fermion surface states. The possibility of intrinsic superconductivity of the arcs themselves, related to the topology of the band structure with Weyl nodes, has virtually not been considered. Although the arcs cannot support superconductivity in time-reversal-breaking Weyl semimetals1, the non-centrosymmetric varieties remain an option. Indeed, very recently, superconductivity associated with only the Fermi arcs of such systems has been predicted theoretically12. Trigonal PtBi2 has emerged recently as a type-I Weyl semimetal that reportedly exhibits superconductivity9,13, making it an attractive candidate for topological superconductivity. Scanning tunnelling spectroscopy experiments confirmed the presence of surface superconductivity by observing typical spectra of superconducting gaps13,14. These spectra provided the evidence that the topological

Fermi arcs bear the superconductivity in PtBi2. This occurs on both of the non-equivalent surfaces of PtBi2.

Three-dimensional band structure The electronic structure of trigonal PtBi2 has been studied both experimentally and theoretically9,15–19. The material crystallizes in the trigonal P31m space group13 and exposes two different surfaces upon cleaving, which we refer to as A and B below (Fig. 1a). The band structure (Fig. 1b) arises mostly due to hybridization of Bi 6p, Pt 5d and Pt 6s states. Two sets of Weyl points are located in momentum space as shown in Fig. 1c,d, having the energy of 47 meV above the Fermi level. To set a baseline from which the three-dimensional (3D) band structure can be resolved, we have recorded 16 angle-resolved photoemission spectroscopy (ARPES) datasets covering at least the first 3D Brillouin zone and approximately 1 eV in energy using the photon energies from 15 to 43 eV (see also Extended Data Fig. 1). This allowed us to identify high-symmetry points along the kz direction and find the value of the inner potential (V0 = 10.5 eV). In Fig. 1e we show the Fermi surface maps taken using the photon energies corresponding to high-symmetry points along the kz direction and approximately half-way in between them. The latter two (left column) are easy to recognize by the pronounced C3 symmetry of the pattern, rotated by 60° with respect to each other. The map taken with 19 eV photons has a higher degree of hexagonal shape, which corresponds to the A-point of the Brillouin zone. The map taken with 29 eV photons at the level of the Γ-point bears a certain resemblance to the almost featureless calculated intensity as suggested by Fig. 1c.

1 Leibniz Institute for Solid State and Materials Research, IFW Dresden, Dresden, Germany. 2Kyiv Academic University, Kyiv, Ukraine. 3Helmholtz-Zentrum Berlin für Materialien und Energie, Berlin, Germany. 4Centro Atómico Bariloche, Instituto de Nanociencia y Nanotecnología (CNEA-CONICET) and Instituto Balseiro, San Carlos de Bariloche, Argentina. 5Max Planck Institute for Solid State Research, Stuttgart, Germany. 6Würzburg-Dresden Cluster of Excellence ct.qmat, Dresden, Germany. 7Present address: Helmhotz-Zentrum Dresden-Rossendorf, Dresden, Germany. ✉e-mail: [email protected]; [email protected]; [email protected]

294 | Nature | Vol 626 | 8 February 2024

Pt

Bi2

c

Pt

Pt

Bi3

Bi3 Bi2

A

H

L

Γ

0.5 Bi2

d 1.00

K

M

1.0

Pt

Pt

Bi3

Bi3

b

Bi4

Bi4 Bi4

Bi4

Energy (eV)

Bi4 Bi4

Bi4

K

106 (m s–1)

a

Γ

WP

0

–0.5

M –1.0

0.11

Γ

A

K

M

WP

L

e 0.5 Momentum (Å–1)

Momentum (Å–1)

0.5

0

–0.5

H

19 eV

0.5 Momentum (Å–1)

Momentum (Å–1)

0.5

0

–0.5

K Γ

0

M

–0.5 21 eV

29 eV

–0.5 0 0.5 Momentum (Å–1)

–0.5 0 0.5 Momentum (Å–1)

–0.5 Max. g

0

–0.2

Energy (eV)

Energy (eV)

L

–0.5 17 eV

f

A

0

–0.4

Min.

–0.6 –0.4

0 Momentum (Å–1)

0.4

–0.4

0

0.4

Momentum (Å–1)

0 0.5 Momentum (Å–1)

–0.5

0 0.5 Momentum (Å–1) Max.

0

–0.2

–0.4

Min.

–0.6 –0.4

0

0.4

Momentum (Å–1)

–0.4

0

0.4

Momentum (Å–1)

Fig. 1 | 3D band structure of PtBi2 . a, Crystal structure of PtBi2. b, Fragment of the band structure. One Weyl point is included. c, Fermi surface, Weyl and high-symmetry points. Colour scale indicates Fermi velocity. d, ΓMK plane of the Brillouin zone with projections of the Weyl points. Magenta (blue) colours stand for positive (negative) chirality. e, Fermi surface maps taken using different photon energies and the corresponding results of the band structure calculations. We note that fixed photon energy probes a sphere of the large radius in the k-space, matching theoretical data formally only at one point in

the centre. Theoretical Fermi maps were averaged over a range of 1/10 of the Brillouin zone size in the k z direction to account for experimental uncertainties. The intensities of the theoretical Fermi maps were normalized to the density of states for different k z points. f,g, Left, energy-momentum intensity distributions at 21 eV (f) and 19 eV (g) along the cuts indicated by blue dashed arrows in e. Right, corresponding energy-momentum spectra taken from the band structure calculation.

The experimental pattern in this case is connected with the finite kz resolution of ARPES. In Fig. 1f,g, we also show the comparison of the dispersions along the lines indicated in Fig. 1e. The features look very similar, being shifted in energy or momentum without any signatures of strong renormalization, or similar manifestations (see also Extended

Data Fig. 3). These data suggest a reasonable general agreement between experiment and theory, which is in accord with previous ARPES studies15–18. The experimental confirmation of the main features of the band structure and Weyl points near the Fermi level thus implies that PtBi2 is indeed a Weyl semimetal, which we will fully confirm below. Nature | Vol 626 | 8 February 2024 | 295

a

Termination A

b

Termination B

Angle (°)

10

0.4 0.2

–10 15 eV

0 Arcs

–0.2

c

17 eV

19 eV

Arcs 10

–0.4 –0.2

0

0.2

0.4

Termination A

–0.4 –0.2

0

0.2

0.4

Termination B

Arcs

0 –10

0.4

23 eV

21 eV

0.2 10

0

Angle (°)

Momentum (Å–1)

0

Angle (°)

Momentum (Å–1)

0.6

Termination A

Arcs –0.2

Arcs

0 –10

–0.4

29 eV –0.4 –0.2 0 0.2 Momentum (Å–1)

d

0.4

–0.4 –0.2 0 0.2 Momentum (Å–1)

Arcs

0.4

–20 –10

e

Arcs

33 eV 0 10 Angle (°)

–20 –10

39 eV 0

10

–20 –10

Angle (°)

Arcs

0

Arcs

Max.

0

10

Angle (°)

Max. 0

Energy (eV)

Energy (eV)

–0.1

–0.2

–0.3

–0.1

–0.2 Termination A

Min.

–0.4 –0.4

0

0.4

–0.4

Momentum (Å–1)

0

0.4

Momentum (Å–1)

Fig. 2 | Fermi arcs. a, High-resolution Fermi surface maps (hν = 17 eV, T = 1.5 K) from both terminations. Arcs in the first Brillouin zone are indicated by the arrows. Note their presence in the equivalent positions in the first and repeated Brillouin zone. b, Fermi surface maps at different photon energies, all showing the presence of the arcs measured at 15 K. The sketch in the middle provides a

Surface states on two terminations In Fig. 2a we present high-resolution Fermi surface maps from both terminations. Although seemingly different, closer inspection suggests that they share mostly the same pattern, provided intensity variations are taken into account. The number of localized features can be clearly distinguished in the map from termination B, at approximately 3/4 of ΓM distance and equivalent locations. These features have been overlooked in earlier ARPES studies15–18. Since the calculated bulk continuum displayed in Fig. 1b does not contain any similar electronic states in this region, we consider those as originating from the surface. The termination A map also shows similarly located features, but they are more clearly seen in the second Brillouin zone. The underlying bulk-related intensity is higher, masking the surface states. To establish their presence unambiguously, we show eight Fermi surface maps taken using different photon energies in Fig. 2b. All maps exhibit all the above features at the same location—approximately 3/4 of the ΓM distance, as 296 | Nature | Vol 626 | 8 February 2024

Termination B –0.4

0 Momentum (Å–1)

Min. 0.4

–0.4

0 0.4 Momentum (Å–1)

visual reference for position of the arcs. c, Arcs as seen in the calculations. Blue dots show the projections of the Weyl points. d,e, Experimental and calculated energy-momentum intensity plots for terminations A (d) and B (e) along the cuts through the arcs highlighted by blue dashed arrows in a.

in the case with termination A. Since it is unlikely that a particular bulk feature would be present in all of the recorded Fermi surface maps for a material with a highly 3D electronic structure, we conclude that these features also represent the surface. The schematic plot (in the middle of Fig. 2b) summarizes our observations regarding the locations of the arcs made from considering the Fermi surface maps. Detected spots of intensity, which we identified above as surface states, remarkably coincide with the results of calculations which take into account the presence of the surface (Fig. 2c). Since PtBi2 is a Weyl semimetal, one does expect the presence of the topological Fermi arcs, different for terminations A and B. In an ideal type-I Weyl semimetal, the location of the starting and end points of the arcs should be identical as those are the projections of the Weyl points (Fig. 1d). The Weyl points, non-degenerate crossings of the bands in 3D k-space, are almost impossible to detect by ARPES directly because of the finite resolution, but the corresponding Fermi arcs have been repeatedly seen experimentally in various materials20–24.

a

b

0.1 Arcs

Min.

–0.1 –0.2

–0.1

–0.2

–0.3 –0.3 0 Momentum (Å–1)

EDC Fit

0.2

0

0.2

0 0.2 0 Momentum (Å–1)

Arc

0.2

0

0.2

0

0.2

Arc

Arc

0 Energy (eV)

Surface Bulk

0.8

–0.1

–0.2

0.6

–0.1

0

0.1

–0.1

f 0.4

0.2

0

0

0.1

g 0

Energy (eV)

Intensity (a.u.)

0

e

d 1.0

Arcs

Arc

0

Max.

0

c

Arc

Arc

Energy (eV)

Momentum (Å–1)

0.2

Arc

Arc

–0.1

h Arc

Arc

0

0.1

Min. Arc

0.2 Max.

–

Arc

–0.1 –0.2 –0.3

–100 –50 0 Energy (meV)

–40

–20 0 Energy (meV)

–0.2

0 0.2 Momentum (Å–1)

0.2

0 –0.2 Momentum (Å–1)

–0.2

0 0.2 Momentum (Å–1)

Fig. 3 | Laser-ARPES. a, Fermi surface map taken using hν = 5.9 eV at 3 K. Arcs are seen together with other bulk-originated features. b, Underlying dispersion along the momentum cuts indicated by arrows in a. c, Typical EDCs from b. Bulk EDC is taken close to zeroth momentum, while surface EDC corresponds to the arc. d, One of the narrowest and strongest EDCs detected in the present study. e, Arcs seen along the different cuts through the Brillouin zone in different

experimental geometries. f, Intensity distribution taken using horizontally polarized light along the path crossing two arcs. g, The same momentum and energy range as in f, from the calculations. Note, the surface states at around 200 meV binding energies are also reproduced. h, Circular dichroism from the same region of the k-space. Colour bar in pannel a also applies to panels b,e,f, and g.

Figure 2d,e demonstrates a comparison of the intensity distribution along the paths marked in Fig. 2a, which run through the arcs. The arcs are situated very close to the Fermi level and are well distinguished from the regions smeared out by kz-resolution bulk dispersions. Considering the discrepancies in the experimental and theoretical 3D band structure (Fig. 1), we do not expect exact correspondence between the calculated Fermi arcs and the ARPES data, but the observed agreement proves not only that the experimental features are indeed the topological Fermi arcs, but also that PtBi2 is a Weyl semimetal.

The most striking characteristic of the arc states is their energy distribution. In Fig. 3c, we compare the energy distribution curves (EDCs) corresponding to the bulk and surface states. As the data are taken with very high resolution and at extremely low temperature, the Fermi momentum (kF) EDC representing bulk states has a well-defined maximum (full-width at half-maximum (FWHM), 30 meV) near the Fermi level and leading-edge width of approximately 5 meV. However, the sharpness and peak-to-background ratio of the EDC representing the Fermi arc is unprecedented. We have routinely observed the peaks having FWHM below 3 meV and a peak-to-background ratio of approximately 50 in numerous cleaves of many samples (Extended Data Fig. 4b and Methods). One such curve is shown in Fig. 3d. As far as we are aware, such a sharp peak has never before been observed in any photoemission experiment from solids. In Fig. 3e, we show further appearances of the arcs in the momentum– energy plots from different cleaves and different terminations. The sharpness and flatness retain the robust characteristics of the feature in all our experiments at the lowest temperatures. We noticed that for A and B surfaces, the arc states are supported by the strongly and weakly dispersing bulk states, respectively, exactly as expected from theory. Direct comparison with the calculations, considerating the presence of the topological surface states, is presented in Fig. 3f,g. The agreement

Robust Fermi arcs from laser-ARPES To study the detected spots of intensity in the Fermi surface maps in more detail, we carried out ARPES experiments using a laser setup. Because of the low kinetic energy of photoelectrons (approximately 1.7 eV), the part of the Brillouin zone that is accessible during these experiments is very limited. We have concentrated on detecting at least one arc in the portion of the k-space marked in the sketch of Fig. 3a. Several representative cuts through the key features seen in the map are shown in Fig. 3b. With that effort, the arc is better resolved, but still very localized in terms of both momentum and energy. We estimate the momentum extension to be of the order of 0.04 Å−1, which is in excellent agreement with theory (Fig. 2c).

Nature | Vol 626 | 8 February 2024 | 297

a

b 20

T = 30 K Energy (eV)

–0.05 –0.10

c Intensity (a.u.)

–0.05 –0.10

0.2

–0.15

Momentum (Å–1)

f

e

1.0

–0.5

Intensity (a.u.)

0 –1

–1.0 –2

–1.5

Sample 1 4

9

Sample 2

14

0

5

10

0

–0.5 –1

–1.0 –1.5

Sample 3

Sample 4

5 10 15 20 Temperature (K)

5 10 15 20 25 Temperature (K)

0 –2 –4 Sample 6

2.7 K 10 K

Sample 2

–0.17 –0.13 Momentum (Å–1)

g

0.5

Sample 2 0.15 0.19 Momentum (Å–1)

h 0

Sample 2 Surface

Bulk –0.05

–0.10

0

15

1.0 Intensity (a.u.)

Energy (meV)

0

Energy (meV)

–10 0 Energy (meV)

Momentum (Å–1)

2

–6

0

–0.05

Peak position

3K 10 K

Sample 4

0

0.5

–0.05

i

Sample 4

2Δ

Surface

Bulk

Superconductor s

0

Leading edge

nt

–0.2

Termination B

0.5

0.20 Momentum (Å–1)

oi

–0.15

Termination A

0.15

d

1.0

Energy (eV)

Energy (eV)

0

–0.10 Momentum (Å–1)

–1

–0.10

0

1

0 –15 –10 –5

0

Energy (meV)

5

0

0.1 Momentum

0.2

lp

T=3K

–0.15

Sample 2

ey

T=3K

Sample 6

–40

W

Termination B

–20

Metal

–0.15

Termination A

0

Energy (eV)

Energy (eV)

0

Energy (meV)

T = 30 K

Superconductor

(Å–1)

Fig. 4 | Superconducting arcs. a, Temperature dependence of the arcs’ dispersion from the terminations A and B. b, Zoomed-in datasets showing underlying dispersion of the arcs. c, EDCs corresponding to the coloured arrows in b. d, Leading edge and peak positions from b. e, Averaged values of the peak positions closest to the Fermi level as a function of temperature for different samples and terminations. Samples 1 and 3 correspond to

termination A and Samples 2 and 4 correspond to termination B. f, Shift of the EDCs with temperature. g, Difference plots showing the changes of the intensity as a function of temperature. h, Results of the calculated spectral weight, taking into account the superconductivity at the surface. i, Schematics of the electronic structure of PtBi2. Green contours represent the Majorana states suggested by the topological superconductivity at the surfaces. a.u., arbitrary units.

with the experiment is remarkable: bulk- and surface-related dispersions are captured not only qualitatively but also quantitatively. The difference between the spectra taken with right- and left-circularly polarized light (Fig. 3h) allows us to identify additional features in the intensity distribution, making the agreement with the theory even stronger (see also Extended Data Fig. 6). Despite the clear correspondence between laser-ARPES data and density functional theory calculations, there is one detail which remains unexplained—the striking flatness of the surface band without any signature of the Fermi level crossings.

characteristic features of superconductivity, we have carried out temperature dependent measurements. In Fig. 4a we show the datasets recorded at 3 and 30 K for both terminations. The comparison of the spectra taken at different temperatures underlines their flatness at the lowest temperature. The arcs clearly lose spectral weight and gain dispersion—just as is to be expected when the system enters the normal state. The apparent asymmetry of the arcs’ dispersion stems from the openness of the Fermi contour made by an arc—conventional electron-like Fermi surface pockets would be supported by the symmetric (with respect to the bottom) dispersions crossing the Fermi level. Extended Data Fig. 2b presents a zoomed-in picture of the calculated arcs, together with bulk bands. The arcs intersect the Fermi level on one side only and merge into the bulk zones on the other side, creating an open contour on the Fermi surface map. We have also reproducibly observed another peculiar aspect of the superconducting state behaviour. In Fig. 4b, where the arcs are

Superconductivity at the surface Record-high sharpness of the arc EDCs strongly resembles coherence peaks in ARPES data from superconductors (for example, ref. 25). To determine whether the electronic states in question bear any other 298 | Nature | Vol 626 | 8 February 2024

measured with the highest resolution, the typical back-bending of the dispersion from the side where the states most closely approach the Fermi level is clearly seen. This is illustrated in Fig. 4c,d where we plot EDCs at several momentum values as well as their peaks and their leading-edge positions. We further track the behaviour of the peak positions as a function of temperature in Fig. 4e. Actual EDCs corresponding to the broadest transition in sample 3 can be found in Extended Data Fig. 4b (Methods). Typical for superconductivity, shifts are observed when going through the critical temperatures Tc. Such shifts measured at the kF give a rather precise estimate of the superconducting gap Δ. Our measurements yield TcA = 14 ± 2 K and TcB = 8 ± 2 K, whereas the corresponding superconducting energy gaps are 1.4 ± 0.2 meV and 2 ± 0.2 meV, respectively. The transition for termination A seems to be broader and Tc higher compared to termination B, which suggests that slightly differing superconducting states set in on the opposing surfaces. Taking into account the different electronic structure of the two surfaces with Fermi arcs, it is not surprising that the superconducting orders are not fully equivalent. Signatures of Berezinskii–Kosterlitz– Thoules transition seen by transport9 may explain the unusual Δ/Tc ratios. In Fig. 4f,g, we present additional evidence for essentially surfacerelated superconductivity observed in two different samples. While the EDCs corresponding to the surface states are clearly shifted with varying temperature (Fig. 4f), the bulk-related spectral weight remains virtually intact, showing only weak changes caused by the slightly different width of the Fermi function. This is illustrated with the aid of two-dimensional difference plots (Fig. 4g), where the clearly stronger variations of spectral function occur in the region where the arcs are located. We can reproduce the experimental spectral function of PtBi2, including both the flatness and back-bending of the surface states, by switching on superconductivity only at the surface via a solution of the Bogoliubov–de Gennes (BdG) Hamiltonian for a semi-infinite solid with a gap function of V0 = 2 meV in the first three PtBi2 layers. The result is shown in Fig. 4h for the momentum and energy intervals corresponding to Fig. 4g (see also Extended Data Fig. 5 (Methods)). Note that only the electron–electron part of the BdG spectral density is plotted to model the ARPES signal and that only the states around the Fermi arcs acquire a gap at the Fermi level. The superconductivity of arcs in PtBi2 follows not only from the emergence of unusually strong and sharp coherence peaks at low temperatures, flatness and back-bending of the dispersion, as well as characteristic shifts of the EDCs; it follows also from the striking agreement with recent scanning tunnelling microscopy (STM) data14—results of another surface-sensitive experiment on the crystals from the same batch. There (see figure 3 in ref. 14) the authors observed typical tunnelling conductance of superconductors characterized by the superconducting gaps varying in space. Remarkably, the average value of the gap closely corresponds to the gap values determined by ARPES. The rather unusual considerable zero-biased conductance observed by Schimmel et al.14 now has a very natural explanation in terms of a bulk contribution which remains ungapped. As seen from the ARPES data, the integrated contribution from the states associated with the bulk can easily reach a noticeable fraction of the signal from the surface, despite the dominant intensity of the arcs. The spot size of the laser beam in our study is of the order of 0.1 mm. This explains the agreement between the determined gap values with averaged STM data and does not exclude the existence of higher Tc regions, the detection of which by ARPES would require an application of micro- or nano-variations of the technique. PtBi2 emerges as a stoichiometric Weyl semimetal with possible surface-only superconductivity (Fig. 4i) and thus opens up a plethora of possibilities to manipulate topological and superconducting phases in a single material. For instance, by varying the thickness of the single crystal, one can obtain a tunable Josephson junction that is intrinsically topological due to the Weyl semimetal forming the weak

link. Topological superconductivity at the surface also may generate Majorana states at the edges. In this context it is interesting to note that we observe that the momentum-independent spectral weight at the Fermi level (apparently seen, for example, in Fig. 2d), appears to be enhanced when probing regions of the surface with a number of terraces. Further studies are needed to unambiguously identify and control both the higher-Tc superconductivity and possible Majorana states in surfaces and edges of PtBi2 single crystals and nano-structures.

Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06977-7. 1. 2. 3. 4. 5. 6. 7. 8.

9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25.

Sato, M. & Ando, Y. Topological superconductors: a review. Rep. Prog. Phys. 80, 076501 (2017). Sharma, M. M., Sharma, P., Karn, N. K. & Awana, V. P. S. Comprehensive review on topological superconducting materials and interfaces. Supercond. Sci. Tech. 35, 083003 (2022). Fu, L. & Kane, C. L. Superconducting proximity effect and Majorana fermions at the surface of a topological insulator. Phys. Rev. Lett. 100, 096407 (2008). Mourik, V. et al. Signatures of Majorana fermions in hybrid superconductorsemiconductor nanowire devices. Science 336, 1003–1007 (2012). Frolov, S. Quantum computing’s reproducibility crisis: Majorana fermions. Nature 592, 350–352 (2021). Wang, D., Wiebe, J., Zhong, R., Gu, G. & Wiesendanger, R. Spin-polarized Yu-ShibaRusinov states in an iron-based superconductor. Phys. Rev. Lett. 126, 076802 (2021). Borisenko, S. et al. Strongly correlated superconductor with polytypic 3D Dirac points. npj Quantum Mater. 5, 67 (2020). Wan, X., Turner, A. M., Vishwanath, A. & Savrasov, S. Y. Topological semimetal and Fermi-arc surface states in the electronic structure of pyrochlore iridates. Phys. Rev. B 83, 205101 (2011). Veyrat, A. et al. Berezinskii–Kosterlitz–Thouless transition in the type-I Weyl semimetal PtBi2. Nano Lett. 23, 1229–1235 (2023). Zhang, P. et al. Observation of topological superconductivity on the surface of an iron-based superconductor. Science 360, 182–186 (2018). Meng, T. & Balents, L. Weyl superconductors. Phys. Rev. B 86, 054504 (2012). Nomani, A. & Hosur, P. Intrinsic surface superconducting instability in type-I Weyl semimetals. Phys. Rev. B 108, 165144 (2023). Shipunov, G. et al. Polymorphic PtBi2: growth, structure, and superconducting properties. Phys. Rev. Mater. 4, 124202 (2020). Schimmel, S. et al. High-TC surface superconductivity in topological Weyl semimetal t-PtBi2. Preprint at https://arxiv.org/abs/2302.08968 (2023). Yao, Q. et al. Bulk and surface electronic structure of hexagonal structured PtBi2 studied by angle-resolved photoemission spectroscopy. Phys. Rev. B 94, 235140 (2016). Thirupathaiah, S. et al. Possible origin of linear magnetoresistance: observation of Dirac surface states in layered PtBi2. Phys. Rev. B 97, 035133 (2018). Jiang, W. et al. Electronic structure of non-centrosymmetric PtBi2 studied by angle-resolved photoemission spectroscopy. J. Appl. Phys. 128, 135103 (2020). & Feng, Y. et al. Rashba-like spin splitting along three momentum directions in trigonal layered PtBi2. Nat. Commun. https://doi.org/10.1038/s41467-019-12805-2 (2019). & Gao, W. et al. A possible candidate for triply degenerate point fermions in trigonal layered PtBi2. Nat. Commun. https://doi.org/10.1038/s41467-018-05730-3 (2018). Yang, L. X. et al. Weyl semimetal phase in the non-centrosymmetric compound TaAs. Nat. Phys. 11, 728–732 (2015). Lv, B. Q. et al. Experimental discovery of Weyl semimetal TaAs. Phys. Rev. X 5, 031013 (2015). Deng, K. et al. Experimental observation of topological Fermi arcs in type-II Weyl semimetal MoTe2. Nat. Phys. 12, 1105–1110 (2016). Haubold, E. et al. Experimental realization of type-II Weyl state in noncentrosymmetric tairte4. Phys. Rev. B 95, 241108 (2017). Borisenko, S. et al. Time-reversal symmetry breaking type-II Weyl state in YbMnBi2. Nat. Commun. https://doi.org/10.1038/s41467-019-11393-5 (2019). Kushnirenko, Y. S. et al. Nematic superconductivity in LiFeAs. Phys. Rev. B 102, 184502 (2020).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. © The Author(s) 2024

Nature | Vol 626 | 8 February 2024 | 299

Methods ARPES measurements ARPES measurements were carried out on the 12 and 13ARPES endstations26 at BESSY II synchrotron (Helmholtz-Zentrum Berlin), as well as in the Leibniz-Institut für Festkörper und Werkstoffforschung Dresden (IFW) laboratory using the 5.9 eV laser light source. Samples were cleaved in situ at a pressure lower than 1 × 10−10 mbar and measured at the temperatures of 15 K and 1.5 K at BESSY II and 3-30 K in the IFW laboratory. The experimental data were obtained using the synchrotron light in the photon energy range from 15 to 50 eV with horizontal polarization and laser light with horizontal and circular polarizations. Angular resolution was set to 0.2–0.5° and energy resolution to 2–20 meV. The findings from the experiments were consistent and reproducible across multiple samples. The simultaneous presence of bulk non-superconducting and surface superconducting states hinders the detection of true coherence peaks with ARPES. Our experiments at the synchrotron, with energy resolution of the order of 5 meV, turned out to be insufficient to detect even the shifts of the leading edges of the corresponding arc peaks having FWHM of the order of 10 meV and peak-to-background ratio of approximately 5. This is because the arc states are always on top of the bulk continuum. Only by measuring with energy resolution of the order of 1–2 meV did we manage to observe sufficiently sharp peaks (Fig. 3c,d and Extended Data Fig. 4) and their sensitivity to temperature. The sharpest features need to be found on the surface. A superconducting gap on the arcs is most likely anisotropic. We included error bars in Fig. 4e to show the influence of a small shift of the beam spot and thus slightly different emission angle. Taking into account the very high localization in momentum space, this could lead to probing a different part of the arc and thus different kF, where the superconducting gap is slightly different. Bulk band structure and Fermi arc position In Extended Data Fig. 1, we show ARPES Fermi surface maps obtained using the photon energies from 15 eV to 43 eV. Relatively strong variation of the pattern suggests a reasonable kz-sensitivity of our experiment. We found the optimal value of the inner potential to be equal to 10.5 eV. This agrees with the previous study of Jiang et al.17. In Extended Data Fig. 2, we present further evidence that our assignment of the surface and bulk features is correct. Extended Data Fig. 2a shows EDCs taken across the Fermi arc for different photon energies (from synchrotron and laser sources), alongside the theoretical EDC for the fully integrated kz. The peak corresponding to the Fermi arc remains clearly visible without any noticeable dispersion for different values of kz, whereas the peaks located further below the Fermi level disperse. Such absence of the dispersion is peculiar to the surface states. In Extended Data Fig. 3, we show an analogue of Fig. 1e–g, but here we compare experimental data with the results of band structure calculations carried out using the linear muffin-tin orbital (LMTO) method in the atomic sphere approximation as implemented in PY LMTO computer code27. As is seen from the figure, the agreement is at the same level as earlier, underpinning the previous conclusion as regards the good agreement between experimental and theoretical 3D band structure. In Extended Data Fig. 4b, we present the sharpest EDCs from among the various samples and cleaves. Most have FWHM below 3 meV and a peak-to-background ratio of over 30. Band structure calculations We performed density functional theory calculations using the full-potential nonorthogonal local-orbital scheme of ref. 28 within the general gradient approximation29 and extracted a Wannier function

model. This allows determination of bulk projected spectral densities (without surface states) and the spectral densities of semi-infinite slabs via Green’s function techniques30. To model surface superconductivity of the semi-infinite slab, the Wannier model is extended into the BdG formalism with a zero-gap function except for a constant Wannier orbital diagonal singlet gap function matrix at the first three PtBi2 layers. A modification of the Green’s function method is used to accommodate this surface-specific term.

Surface superconductivity calculations To model a system which has a non-zero gap function only at the surface—in the first 30aB which is 3(PtBi2) layers—we modified the standard Green’s function technique for semi-infinite slabs. The system is built by a semi-infinite chain of identical blocks consisting of 3(PtBi2) layers, repeating indefinitely away from the surface. Each block has a Hamiltonian Hk for each pseudo momentum k in the plane perpendicular to the surface and a hopping matrix Vk, which couples neighbouring blocks. The blocks’ minimum size is determined by the condition that H and V describe all possible hoppings. To add superconductivity, the BdG formalism is used by extending the matrices in the following way:  Hk Δk  Hk ,BdG =  +  ,  Δk − H −* k  0  Vk Vk ,BdG =  ,  0 − V −* k   0 V0 where we choose Δk = δii ′   with i being a spinless Wannier − V0 0  function index and the 2 × 2 matrix to act in a single Wannier function’s spin subspace. This choice also leads to Δ [Vk,BdG] = 0, since V is an offdiagonal part of the full Hamiltonian. To model surface-only superconductivity, we let V0 = 0 for all (infinite) blocks, except the first one, which gets a finite V0 = 2 meV. The standard Green’s function solution for this problem consists of determining the propagator X which encompasses all diagrams that describe paths that start at a certain block, propagate anywhere towards the infinite side of that block and return to that block. X also describes the Green’s function G00 of the first block and the self-energy to be added to the Hamiltonian to obtain G00 (a self-consistency condition) G00 = X = (ω+ − H − Σ)−1, Σ = VXV + (in practice, however, self-consistency is obtained by an accelerated algorithm). From this recursion, relations can calculate all other Green’s-function blocks. These can be derived by subdividing propagation diagrams into irreducible parts using known components, in particular X. If the first block differs from all the others (as is the case due to Δk) one needs to modify the method in the following way. Let the first block have Hamiltonian h and hoppings to the second block v (while all other blocks are described by H and V). Then the irreducible subdivision of −1 the propagation diagrams for G00 results in g = (ω+ − h) . G00 = g + gvXv+g + ( gvXv +) g =

1 ω+ − h − vXv +

which contains the surface Hamiltonian and a modified self-energy depending on the X of the unmodified semi-infinite slab. From this we can derive the second block’s Green’s function

G11 = X + Xv +G00vX and all others

Gn +1, n +1 = X + XV +GnnVX ,

n>0

which can be used to obtain the spectral density up to a certain penetration depth. Note that in our BdG case H = Hk,BdG [V0 = 0],V = Vk,BdG [V0 = 0] and h = Hk,BdG [V0 ≠ 0] , v = V. The BdG spectral density is particle–hole symmetric and to obtain results that resemble ARPES data, one needs to use the particle–particle block Gee (the upper left quarter of the G matrix) only. Extended Data Fig. 5b shows the resulting spectra of this method along the path denoted in Extended Data Fig. 5a. Note that a gap is opened at the surface band pockets close to the Fermi energy, while the rest of the spectrum stays gapless (if we let V0 ≠ 0 for all blocks, we get a completely gapped spectrum). Extended Data Fig. 5c shows a zoomed-in region around the surface state. Note that the bulk bands are gapless (dark blue vertical features) while the surface state shows a gap and corresponding band back-bending. The particle–hole symmetry becomes apparent, although with a larger spectral weight for the occupied part because we use Gee only.

Further discussion One approach to demonstrate the existence of topologically protected states with a topological insulator is to perform spin-resolved ARPES. In this technique, the spin-locking effect determines the spin structure in the vicinity of the surface Dirac node. However, the situation is quite different for Weyl semimetals. Here, there is no specific spin structure or configuration associated with the Weyl nodes, which can occur at generic points in the Brillouin zone. As inversion is broken and spin-orbital coupling present, each band at a generic k-point naturally possesses a spin direction, but this spin texture is smooth. Consequently, spin-resolved ARPES measurements cannot directly reveal Weyl points. We would like to exclude the interpretation of our data based on density-wave order, which could, in principle, result in the similar features in the spectra. Charge density-waves require a redistribution of the spectral weight in the momentum space, characterized by the particular k-vector (vectors). We have always observed almost the same Fermi surface maps and underlying dispersions, independent of temperature. In line with these observations are the results of the STM studies which never detected any kind of a reconstruction. We have never observed any replica of the arcs or of the deeper lying surface states, such as a strong feature at (−0.2, −0.2) in Fig. 3f,g. It is also not clear which k-vector would be suitable for characterizing the density-wave order. If the arcs are simply superimposed in momentum, they all are of electron-like topology, so the opening of the hybridization gaps

seems very unlikely. Finally, the fundamental difference between the density-wave gaps and superconducting gaps is that the latter are always pinned to the Fermi level. This is the only energy interval where we observe the changes in the spectra of PtBi2 with temperature.

Data availability Source data are provided with this paper. Other data are available from the corresponding authors upon reasonable request.  26. Borisenko, S. V. “One-cubed” ARPES user facility at BESSY II. Synchrotron Radiat. News 25, 6–11 (2012). 27. Antonov, V., Harmon, B. & Yaresko, A. Electronic Structure and Magneto-Optical Properties of Solids (Springer, 2004). 28. Koepernik, K. & Eschrig, H. Full-potential nonorthogonal local-orbital minimum-basis band-structure scheme. Phys. Rev. B 59, 1743–1757 (1999). 29. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865–3868 (1996). 30. Sancho, M. P. L., Sancho, J. M. L., Sancho, J. M. L. & Rubio, J. Highly convergent schemes for the calculation of bulk and surface green functions. J. Phys., F Met. Phys. 15, 851–858 (1985). Acknowledgements We thank C. Fulga for enlightening discussions and the Helmholtz-Zentrum Berlin für Materialien und Energie for the allocation of synchrotron radiation beamtime. This work was supported within the Collaborative Research Center project, “Correlated Magnetism: From Frustration to Topology (SFB 1143)” and by the Dresden-Würzburg Cluster of Excellence project “EXC 2147: Complexity and Topology in Quantum Matter (CT.QMAT)”. S.A. acknowledges the support of Deutsche Forschungsgemeinschaft (DFG) via AS 523/4-1. S.A. and B.B. also acknowledge the support of DFG through Project No. 405940956. J.I.F. acknowledges the support of the Alexander von Humboldt Foundation via the Georg Forster Return Fellowship and Agencia Nacional de Promoción Científica y Tecnológica (ANPCyT) grants PICT 2018/01509 and PICT 2019/00371. A.K., O.S., and S.B. acknowledge the support of Bundesministerium für Bildung und Forschung (BMBF) through project “UKRATOP”. O.S. and B.B. acknowledge the support of BMBF through project “Instant micro-ARPES for in-operando tuning of material and device properties”. We thank U. Nitzsche for technical assistance. Author contributions A.K., O.S., R.L., A.F., L.M., V.V., S.B. and B.B. designed and carried out the ARPES experiments. R.V., K.K., J.I.F., A.Y. and J.v.d.B. developed theoretical aspects and performed calculations. G.S. and S.A. grew single crystals. S.B., A.K. and J.v.d.B. wrote the paper with contributions from all authors. Funding Open access funding provided by Leibniz-Institut für Festkörper- und Werkstoffforschung Dresden (IFW). Competing interests The authors declare no competing interests. Additional information Correspondence and requests for materials should be addressed to Andrii Kuibarov, Rui Lou or Sergey Borisenko. Peer review information Nature thanks Pavan Hosur, Andres Santander-Syro and Dawei Shen for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints.

Extended Data Fig. 1 | Fermi surface maps.Photoemission intensity integrated within a small energy region around the Fermi level. Data were recorded using 16 different photon energies.

Extended Data Fig. 2 | Photon energy dependence of the Fermi arcs. a, Energy distribution curves (EDC) across the Fermi arc as a function of photon energy (upper panel), EDC across the Fermi arc measured with 5.9 eV laser

(middle panel), theoretical EDC for fully integrated bulk (lower panel). b, Theoretical calculation for bulk (left) and bulk with surface (right) in ΓM direction.

Extended Data Fig. 3 | 3D band structure. a, Fermi surface maps taken using different photon energies and corresponding results of the band structure calculations. We note, that fixed photon energy probes a sphere of the large

radius in the k-space, matching theoretical data formally only at one point in the centre. b,c, Energy-momentum intensity distributions at 21 eV and 19 eV respectively along the cuts indicated by blue dashed arrows in panel e.

Extended Data Fig. 4 | EDC across the Fermi arc. a, Energy distribution curves corresponding to sample # 3 from Fig. 4e of the main manuscript and normalized to the maximum intensity. The opening of the superconducting gap is clearly visible as a displacement of the peaks and leading edges with temperature. b, Energy distribution curves taken close to k F of the arcs for different k, samples (S) and cleaves (C).

Extended Data Fig. 5 | Theoretical calculations of gap opening in PtBi2 . a The path in the BZ. b The surface-only superconducting spectral function of the (00-1)-surface for Δ = 2meV and penetration depth 30a B (G ee only). Note that

the gap is only open around the surface state pocket. c the blow-up of this pocket. Note, that the gap is open for the surface state but closed for the bulk bands.

Extended Data Fig. 6 | Polarization dependent datasets. ARPES spectra measured with horizontal (left), circular left (middle) and circular right (right) polarization of the 5.9 eV laser at 3.5 K.

Stable blue phosphorescent organic LEDs that use polariton-enhanced Purcell effects https://doi.org/10.1038/s41586-023-06976-8

Haonan Zhao1, Claire E. Arneson1, Dejiu Fan2 & Stephen R. Forrest1,2,3 ✉

Received: 15 April 2023 Accepted: 14 December 2023 Published online: 20 December 2023 Check for updates

Phosphorescent organic light-emitting diodes (PHOLEDs) feature high efficiency1,2, brightness and colour tunability suitable for both display and lighting applications3. However, overcoming the short operational lifetime of blue PHOLEDs remains one of the most challenging high-value problems in the field of organic electronics. Their short lifetimes originate from the annihilation of high-energy, long-lived blue triplets that leads to molecular dissociation4–7. The Purcell effect, the enhancement of the radiative decay rate in a microcavity, can reduce the triplet density and, hence, the probability of destructive high-energy triplet–polaron annihilation (TPA)5,6 and triplet–triplet annihilation (TTA) events4,5,7,8. Here we introduce the polariton-enhanced Purcell effect in blue PHOLEDs. We find that plasmon–exciton polaritons9 (PEPs) substantially increase the strength of the Purcell effect and achieve an average Purcell factor (PF) of 2.4 ± 0.2 over a 50-nm-thick emission layer (EML) in a blue PHOLED. A 5.3-fold improvement in LT90 (the time for the PHOLED luminance to decay to 90% of its initial value) of a cyan-emitting Ir-complex device is achieved compared with its use in a conventional PHOLED. Shifting the chromaticity coordinates to (0.14, 0.14) and (0.15, 0.20) into the deep blue, the Purcell-enhanced devices achieve 10–14 times improvement over similarly deep-blue PHOLEDs, with one structure reaching the longest Ir-complex device lifetime of LT90 = 140 ± 20 h reported so far10–21. The polariton-enhanced Purcell effect and microcavity engineering provide new possibilities for extending deep-blue PHOLED lifetimes.

PHOLEDs have been extensively used in both display and lighting applications owing to their vibrant colours and high efficiencies. However, because degradation is fundamentally energy-driven4–7, blue PHOLEDs used in displays have unacceptably short lifetimes10–22 compared with green and red PHOLEDs3. The primary, energetically driven mechanisms leading to short blue PHOLED lifetimes are TPA (refs. 5,6) and/or TTA (refs. 4,5,7). These reactions approximately double the energy of the excited states up to approximately 6.0 eV (refs. 4,11), which is sufficient to break intramolecular bonds and convert an organic molecule to a nonradiative quenching centre5,7,10,11. To minimize the probability for high-energy annihilation events while maintaining high efficiency, the triplet density should be reduced using rapid radiative energy transfer. An OLED is, by nature, a weak multimode microcavity23 comprising outcoupled and waveguided modes in the organic and substrate layers and surface plasmon polaritons (SPPs), among others4. The enhancement of the radiative decay rates using a microcavity, known as the Purcell effect, can reduce the triplet density to approximately the inverse of the PF (refs. 5,8) and thereby reduce the probability for TPA and/or TTA (see Methods). Here the PF is the triplet radiative decay rate in the OLED microcavity normalized by its natural radiative decay rate, that is, PF = kr/kr,0. However, as depicted in Fig. 1a, in a conventional PHOLED, the weak triplet energy transfer to conventional metal-cathode SPPs induces only a modest change in the decay rate, leaving a high triplet density at equilibrium. A promising

approach to enhance decay rates was reported by Fusella et al.24, who doubled the lifetime of a green PHOLED by means of energy transfer from triplets to SPPs in a thin top metal cathode containing a random array of Ag nanocubes for top-emission extraction. However, this technique introduces complexity that is incompatible with full-colour, stable and scalable displays manufacturing. In this work, we introduce the polariton-enhanced Purcell effect to extend the operational lifetime of blue PHOLEDs. We demonstrate that energy transfer to PEPs substantially reduces the triplet radiative lifetime and their density within the PHOLED EML. PEPs are a strongly coupled state at the metal–dielectric interface9,25 resulting from mixing of the SPP mode with excitons in the adjacent dielectric. Here the PEP strength is a function of the oscillator strengths of both the cathode and the electron transport layer (ETL) (see Fig. 1b). Combined with a low-quality-factor (Q) optical cavity comprising a Ag cathode + a distributed Bragg reflector (DBR) mirror, the light extraction efficiency and the emission colour saturation are increased. Three archetype devices are made to maximize the total deep-bluephoton output featuring long device lifetime, saturated emission and high external quantum efficiency (EQE). By engineering the PEP-enhanced Purcell effect, we demonstrate a deep-blue Ir(dmp)3based PHOLED with an average PF of 2.4 ± 0.2 across the 50-nm-thick EML, leading to a 5.3-fold increase in LT90 compared with a conventional PHOLED using this same phosphor. By optimizing the Ag/DBR

Department of Physics, University of Michigan, Ann Arbor, MI, USA. 2Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI, USA. 3Department of Materials Science and Engineering, University of Michigan, Ann Arbor, MI, USA. ✉e-mail: [email protected]

1

300 | Nature | Vol 626 | 8 February 2024

a

b e– TPA/TTA

e–

Reduced TPA/TTA

Ag ETL PEP P E EP--enhanced en Purcell elll e eff ffectt PEP-enhanced effect

ITO O ITO

Short-lived conventional cyan OLED Radiatively recombined triplets

Ag/DBR cavity mode

Al Weak energy energ gy tran gy tra transfer sf to SPPs sfer

SiNx/S /SiO2 DBR

Long-lived Purcell-enhanced blue OLED Triplets at equilibrium

e– Polaron

Fig. 1 | The PEP-enhanced Purcell effect. a, Cyan-emitting conventional devices. Triplets have a small energy transfer rate to the SPPs at the cathode surface and thus only a moderate change in the PL lifetime. The slow radiative decay of triplets results in a large density at equilibrium, inducing a high probability for TPA and TTA, leading to rapid degradation. b, Deep-blue-emitting

cavity devices using a Ag/ETL combination, leading to an enhanced Purcell effect by means of fast energy transfer to PEPs. Consequently, the triplet density is reduced, thereby reducing TPA and TTA. The Purcell-enhanced devices use DBRs to form an optical cavity with the metal cathode to improve light extraction and emission colour saturation.

cavity, the Commission Internationale d’Eclairage (CIE) coordinates of the conventional Ir(dmp)3 PHOLED shifted from cyan at (0.16, 0.26) to deep blue at (0.14, 0.14), gaining an almost threefold increase in LT90 using the Purcell effect enhanced by the strong Ag SPP, while maintaining the same EQE. Considering the prolonged device operational lifetime and saturated colour, the device achieves a 14 times enhancement in LT90 compared with other, similarly deep-blue Ir-complex-based PHOLEDs. By balancing the EQE and the PF, a PEP-enhanced device using Ag cathode/BPyTP2 ETL achieves the longest normalized LT90 of 140 ± 20 h at CIE = (0.15, 0.20) among Ir-complex-based PHOLEDs with CIEy  9 weeks), C57BL/6N male and female mice (> 8 weeks) originally purchased from Charles River and then bred in-house. Swiss Webster male and female mice (> 11 weeks) were purchased from Taconic, Charles River, or bred in house. All mice were group housed until adulthood. After surgery with fiber or cannula implantation, all test mice were single-housed. Animals were randomly assigned to control and test groups.

Wild animals

The study did not involve wild animals.

Reporting on sex

Stimulus animals in resident-intruder test were BALB/c male (>9 weeks), C57BL/6N male and female mice (>8 weeks) originally purchased from Charles River and then bred in house, and Swiss Webster male and female mice (>11 weeks) purchased from Taconic, Charles River, or bred in house. Both male and female mice from OxtrCre, OXTCre, and C57BL/6N colonies were used as test animals.

Field-collected samples

The study did not involve field collected samples.

March 2021

Laboratory animals

3

All procedures were approved by the NYULMC Institutional Animal Care and Use Committee (IACUC) in compliance with the National Institutes of Health (NIH) Guidelines for the Care and Use of Laboratory Animals.

Note that full information on the approval of the study protocol must also be provided in the manuscript.

nature portfolio | reporting summary

Ethics oversight

March 2021

4

Hypoblast from human pluripotent stem cells regulates epiblast development ­­­­­

https://doi.org/10.1038/s41586-023-06871-2 Received: 21 February 2020 Accepted: 15 November 2023

Takumi Okubo1, Nicolas Rivron2, Mio Kabata1, Hideki Masaki3,4, Keiko Kishimoto5, Katsunori Semi1, May Nakajima-Koyama1, Haruko Kunitomi1, Belinda Kaswandy1, Hideyuki Sato3,4, Hiromitsu Nakauchi3,4,6, Knut Woltjen1, Mitinori Saitou1,7,8, Erika Sasaki5, Takuya Yamamoto1,7,9 ✉ & Yasuhiro Takashima1 ✉

Published online: 5 December 2023 Open access Check for updates

Recently, several studies using cultures of human embryos together with single-cell RNA-seq analyses have revealed differences between humans and mice, necessitating the study of human embryos1–8. Despite the importance of human embryology, ethical and legal restrictions have limited post-implantation-stage studies. Thus, recent efforts have focused on developing in vitro self-organizing models using human stem cells9–17. Here, we report genetic and non-genetic approaches to generate authentic hypoblast cells (naive hPSC-derived hypoblast-like cells (nHyCs))—known to give rise to one of the two extraembryonic tissues essential for embryonic development—from naive human pluripotent stem cells (hPSCs). Our nHyCs spontaneously assemble with naive hPSCs to form a three-dimensional bilaminar structure (bilaminoids) with a pro-amniotic-like cavity. In the presence of additional naive hPSC-derived analogues of the second extraembryonic tissue, the trophectoderm, the efficiency of bilaminoid formation increases from 20% to 40%, and the epiblast within the bilaminoids continues to develop in response to trophectoderm-secreted IL-6. Furthermore, we show that bilaminoids robustly recapitulate the patterning of the anterior–posterior axis and the formation of cells reflecting the pregastrula stage, the emergence of which can be shaped by genetically manipulating the DKK1/OTX2 hypoblast-like domain. We have therefore successfully modelled and identified the mechanisms by which the two extraembryonic tissues efficiently guide the stage-specific growth and progression of the epiblast as it establishes the post-implantation landmarks of human embryogenesis.

Early blastocysts of the pre-implantation human embryos are composed of trophectoderm and inner cell mass (ICM). The ICM generates the epiblast (that is, future fetus) and hypoblast (that is, primitive endoderm, future yolk sac), a process completed in the late blastocyst stage. During implantation, these two tissues form a bilaminar disc that functions as a developmental template for the embryo. Despite the importance of early human development, our knowledge of human peri-implantation development is limited owing to ethical and legal restrictions. Thus, alternative approaches for analysing this developmentally critical period are necessary. To model human pre-implantation development, it is important to establish cells that correspond to pre-implantation embryos in vitro. In contrast to their mouse counterpart, naive human pluripotent stem cells (hPSCs), corresponding to the pre-implantation epiblast18–20, can generate blastocyst-like structures (blastoids)16,17 and differentiate into the trophectoderm of blastocysts21,22. Although hypoblast differentiation from naive hPSCs has been reported23, the molecular details remain unclear, and the capture of in vitro pre-implantation hypoblast has not been achieved. Thus, it remains unclear whether

extraembryonic tissues support the development of pre-implantation epiblast. Here we induced human pre-implantation hypoblast from naive hPSCs by either transgene overexpression or chemical induction, which guides the epiblast to form the first embryonic cavity, establishes the anterior–posterior axis and, together with the second extraembryonic tissue, the trophectoderm/trophoblast (TB), supports the establishment of the post-implantation embryonic state.

Naive hPSC-induced hypoblast by GATA6 To induce the pre-implantation hypoblast, we compared the potential of naive and primed hPSCs18–20 to differentiate into this tissue (Extended Data Fig. 1a–c). Gata6, Gata4 and Sox17 are expressed in the mouse hypoblast24, and their overexpression was shown to induce embryonic stem (ES) cells to hypoblasts25,26. As the human hypoblast also expresses GATA6, GATA4 and SOX172,3, we introduced doxycycline (DOX)-inducible GATA6, GATA4 or SOX17 transgenes into both naive and primed H9 ES cells by piggyBac (PB) (Fig. 1a). GATA6 overexpression induced the

1 Center for iPS Cell Research and Application, Kyoto University, Kyoto, Japan. 2Institute of Molecular Biotechnology of the Austrian Academy of Sciences (IMBA), Vienna BioCenter (VBC), Vienna, Austria. 3Institute of Medical Science, University of Tokyo, Tokyo, Japan. 4Advanced Research Institute, Tokyo Medical and Dental University, Tokyo, Japan. 5Central Institute for Experimental Animals, Kawasaki, Japan. 6Institute for Stem Cell Biology and Regenerative Medicine, Stanford University School of Medicine, Stanford, CA, USA. 7Institute for the Advanced Study of Human Biology (WPI-ASHBi), Kyoto University, Kyoto, Japan. 8Department of Anatomy and Cell Biology, Graduate School of Medicine, Kyoto University, Kyoto, Japan. 9Medical-risk Avoidance Based on iPS Cells Team, RIKEN Center for Advanced Intelligence Project (AIP), Kyoto, Japan. ✉e-mail: [email protected]; [email protected]

Nature | Vol 626 | 8 February 2024 | 357

a

b Naive or Primed GATA6

GATA6

GATA4

GATA4

SOX17

SOX17

Day 3

78%

101 103 105 101 103 105 101 103 105 PDGFRA–APC Control Sample Primed Day 0

Day 1

0.1%

Day 3

14%

78%

101 103 105 101 103 105 101 103 105 PDGFRA–APC

f

40 0 –40

N N-G6-D1 N-G6-D3

Primed

–50 0 50 PC1 (39.9%)

40 ) 0 3% –40 24. ( 2 PC

14,480 genes

PC3 (12.2%)

Primed Day 1

Day 3

33%

e

Naive

Day 0

Naive

PC2 loadings LEFTY2 NANOG FGF4 SOX2 PRDM14

RSPO3 COL4A1 PDGFRA NID2 FN1 MYL4 GATA6 APOA1

Primed PC3 loadings

14,480 genes

100 80 60 40 20 0

1.5%

Day 3

d

Naive Day 1

Day 0

Naive Day 1

N-D0-ex1 N-D0-ex2 N-G6-D1-ex1 N-G6-D1-ex2 N-G6-D3-ex1 N-G6-D3-ex2 P-D0-ex1 P-D0-ex2 P-G6-D1-ex1 P-G6-D1-ex2 P-G6-D3-ex1 P-G6-D3-ex2

Cell count (Modal)

100 80 60 40 20 0

Cell count (Modal)

c

+DOX

Day 0

CTSE CEACAM1

–4 –3 –2 –1 0 1 2 3 4 –3 –2 –1 0

HAND1 LRRC32 HOXB6 HOXB1 TNNT2 SNAI2 HOXB3

1

2

3

P P-G6-D1 P-G6-D3

Fig. 1 | Naive hPSC differentiation into the PDGFRA+ hypoblast by GATA6 overexpression. a, Schematic of the DOX-dependent induction of the GATA6, GATA4 or SOX17 transgene in hPSCs. b, Bright-field images of naive and primed H9 hPSCs (day 0 (D0)) and hPSC-derived cells with GATA6 overexpression at D1 and D3 under serum-containing conditions (Extended Data Fig. 1o). n = 10. c, Flow cytometry analysis of PDGFRA expression in naive and primed hPSCs after GATA6 induction under serum-free conditions (Extended Data Fig. 1o). n = 3. d, UHC analysis of the transcriptomes of naive hPSCs (N-D0), naive hPSC-derived GATA6-PDGFRA+ cells (N-G6-D1 and N-G6-D3), primed hPSCs (P-D0) and primed hPSC-derived G6-PDGFRA+ cells (P-G6-D1 and P-G6-D3) from two independent experiments (ex1 and ex2). PDGFRA+ cells were sorted on D1 and D3. e, PCA of naive and primed cells. f, PC2 and PC3 loadings of e. In total, 14,481 genes were ordered by their PC2 or PC3 loading scores (Supplementary Table 2). Representative genes among the top 50 are shown. n values show biologically independent experiments. Scale bars, 100 µm (b). Reproducibility is shown in the Methods.

endogenous hypoblast genes GATA6, GATA4, SOX17 and PDGFRA. GATA4 overexpression induced these genes only moderately, but SOX17 overexpression failed (Extended Data Fig. 1d). This suggests a hierarchy in propagating the human hypoblast program, like in mice. After 3 days of overexpression, characteristic naive hPSC morphologies disappeared (Fig. 1b and Extended Data Fig. 1e). Flow cytometry analysis confirmed that PDGFRA was expressed after GATA6 overexpression in naive and primed hPSC-derived cells (Extended Data Fig. 1f). PDGFRA+ cells from naive GATA6-induced hPSCs (naive G6-PDGFRA+) expressed hypoblast marker genes, whereas primed G6-PDGFRA+ cells expressed mesoderm marker genes (Extended Data Fig. 1g). GATA4 overexpression also induced PDGFRA+ cells, but SOX17 did not (Extended Data Fig. 1f). Naive and primed G4-PDGFRA+ cells expressed hypoblast and 358 | Nature | Vol 626 | 8 February 2024

mesoderm genes, respectively (Extended Data Fig. 1h). To characterize hypoblast specification from naive hPSCs further, we developed and optimized a serum-free induction system using N2B27 chemically defined medium (NDiff 227) as a basal medium. First, we observed that GATA6 overexpression in naive hPSCs induces PDGFRA+ cells under N2B27, and FGF4 addition further enhanced this induction (Extended Data Fig. 1i). GATA6 overexpression most efficiently induced PDGFRA expression and PDGFRA+ cells in naive and primed hPSCs after 48 and 72 h, respectively (Extended Data Fig. 1j,k). We observed that 0.1 μM DOX induced PDGFRA expression and PDGFRA+ cells more effectively than 10 μM DOX (Extended Data Fig. 1l–n). On the basis of these data, we defined a hypoblast induction protocol based on GATA6 overexpression (Extended Data Fig. 1o). With optimized induction, GATA6 overexpression reproducibly converted around 80% of naive hPSCs into PDGFRA+ cells on day 3 expressing hypoblast genes (five lines, n = 71; Fig. 1c, Extended Data Fig. 2a,b and Supplementary Fig. 1). GATA4 overexpression under the same induction protocol also induced PDGFRA+ cells, but less efficiently than GATA6 (Extended Data Fig. 2c,d and Supplementary Fig. 1). Hypoblast protein markers were observed after GATA6 overexpression, whereas pluripotency markers were downregulated (Extended Data Fig. 2e). We performed RNA-sequencing (RNA-seq) analysis during differentiation (Supplementary Table 1). Unsupervised hierarchical clustering (UHC) classified the samples on the basis of their origin (Fig. 1d). Principal component analysis (PCA) revealed that PC1 separated naive hPSCs and primed hPSCs even after differentiation (Fig. 1e). However, the similar directional transition along PC2 suggested that a common subset of genes was similarly up- or downregulated in both naive and primed G6-PDGFRA+ cells. During differentiation, naive hPSCs lost the expression of pre-implantation epiblast marker genes2,7 but upregulated hypoblast marker genes (Extended Data Fig. 2f,g). A subset of epiblast and hypoblast marker genes in primed cells also showed a similar expression pattern and strongly affected PC2 (Fig. 1f and Supplementary Table 2). Finally, PC3 revealed a directional, progressive, but opposite transition of cellular properties in naive and primed G6-PDGFRA+ cells. Specifically, mesoderm and body plan genes were enriched for negative PC3 loading values (primed G6-PDGFRA+) (Fig. 1f and Supplementary Table 2). Previous studies reported that PDGFRA is expressed in mesoderm progenitors27,28, and GATA6 is expressed in primitive streak/ gastrulating cells and the mesoderm8,29. Indeed, primed G6-PDGFRA+ cells expressed primitive streak, definitive endoderm and mesoderm genes (Extended Data Fig. 2h) and post-implantation late epiblast marker genes in cynomolgus monkey embryos29 (Extended Data Fig. 2i). Moreover, primed G6-PDGFRA+ cells expressed early primitive streak genes on day 1 and several gastrulation- and mesoderm-related genes on day 3 (Extended Data Fig. 2j). By contrast, naive G6-PDGFRA+ cells did not express these mesoderm genes aside from MIXL1, EOMES and HAND1 (Extended Data Fig. 2j), which were also detected in embryonic hypoblast cells (Extended Data Fig. 2k). Similarly, the hypoblast genes SOX17, APOA2, HNF4A and CTSE were strongly expressed only in naive G6-PDGFRA+ cells along with KLF4 and OTX2 (Extended Data Fig. 2l), which are also expressed in the hypoblast of human blastocysts (Extended Data Fig. 2m). Together, we concluded that GATA6 promotes naive hPSC differentiation into the hypoblast lineage, while primed hPSCs adopt a post-implantation embryonic fate.

Hypoblast induced by signalling molecules As GATA6 and FGF4 efficiently induced hypoblast formation, we investigated the signalling pathways affected by GATA6 overexpression that are vital for hypoblast induction. RNA-seq data showed the upregulation of BMP2/6, STAT3, FRZB and FGFR2 and the downregulation of WNT3 (Extended Data Fig. 3a). We therefore examined these signalling pathways using western blotting. While phosphorylated

a

b

Naive PSCs (H9, H1, AdiPS)

D3 (7F) H9 naive derived

c

e GATA6

GATA4

FOXA2

SOX17

OCT3/4

Naive D3

NANOG

ANPEP–PE

CEACAM1– Alexa488

103

45.3%

Primed 1.7%

101 103

103

101 103 105 101 103 105 PDGFRA–APC Correlation 0.4 0.6 0.8

Zygote 4 cell 8 cell Morula Early ICM Epi Hypoblast

Control D3

Ex1

Ex2

h

1.1% 101 103

103

105 101 103

105

5.8%

6.9%

101 101 103

105 101 103

FGF4 + BMP4

105

103

0 nTE

105

7F

36.0% 37.3% 105 101 103

105

Pre-Epi (Naive)

FGFi/aPKCi

RACL

Definitive Endoderm Primed G6-PDGFRA+ Primed PSCs Hypoblast nHyCs

nCT

105 101 103

101

Int-post-Epi

Post-CT

−50 7F

101 103 105

PSA-Epi

50

20.9%

i

Int-PSA

100 First CT

4F

19.8%

101

101 103

AME

PC3 (4.882%)

g

Blastocyst

D1

Primed-derived Primed PDGFRA+ cells PSCs

Naive-derived PDGFRA+ cells

4F

Control

BMP4

0.6%

105

103

Naive PSCs

4F

105

105

Ex4 Ex5 Ex2 Ex3 Ex1 Ex1 Ex2 Ex1 Ex2 Ex1 Ex2 Ex1 Ex2 Ex1 Ex2 Ex1 Ex2 Ex2 Ex2 Ex4 Ex3 Ex1 Ex3 Ex4 Ex1 Ex1 Ex2 Ex1 Ex2 Ex1 Ex2 Ex1 Ex2 H9 H9 Adi H1 H9 H9 H9 H9 H9 H9 H9 H9 H9 H9 H9 H9 Adi H9 Adi H1 7F 7F 4F G6 (D3) 2F G6 (D3) G6 (D3) G6 (D1) G6 (D1) 7F 7F 4F 2F

D0

105 101 103

101

d

f

0.4%

105

PDGFRA–APC

Naive

FGF4

0.2%

101 D3 (7F)

Cell count (Modal)

1: FGF4 2: BMP 3: IL-6 N2B27 4: PDGF-AA 5: A83-01 100 6: XAV939 80 7: RA 60 Day 3 40 Flow cytometry 20 0

D0

105

FGF/TGF-βi/BMP

FGFi/TGF-αi/(BMP) TE

Hyp FGF/TGF-β Post-Epi

Pre-TE −100

Naive PSCs −50

0 50 PC2 (7.59%)

Pre-Epi

Post-Epi (primed)

100

Fig. 2 | Essential signalling for human hypoblast specification. a, Schematic of the 7F induction of PDGFRA+ cells. b, Bright-field image and flow cytometry 3 days after 7F induction. n = 44. c, Immunofluorescence analysis of naive hPSCs at day 0 and day 3 in 7F medium. The indicated proteins are shown in red and green. Blue, DAPI. n = 2. d, Correlation coefficients of human pre-implantation embryos and naive hPSCs, primed hPSCs and PDGFRA+ cells in 7F, 4F or 2F, or with GATA6 overexpression. Adi, AdiPS cells; 4F, FGF4 and BMP4 with A83-01 and XAV939; 2F, FGF4 and BMP4. e, Minimum essential factors for hypoblast specification. n = 3. f, Bright-field images of marmoset ICM-derived cells. ICM cells were cultured in 4F or with MEK and BMP pathway inhibitors (PD0325901 and LDN-193189) and A83 + XAV (control). n = 2. g, Immunofluorescence images of the marmoset ICM at day 3. Green, SOX17; blue, DAPI. n = 2. h, PCA of bulk RNA-seq data from this study and published reports, and of scRNA-seq data from human embryos. The circles indicate cell types5: blue, pre-implantation;

light blue, post-implantation; pre-Epi, pre-implantation epiblast; post-Epi, post-implantation epiblast; PSA-Epi, primitive-streak anlage epiblast; int-PSA and int-post-Epi, intermediate state cells of primitive-streak anlage epiblast and post-implantation epiblast; AME, amnion; pre-TE, pre-implantation trophectoderm; post-CT, post-implantation cytotrophoblast. Bulk RNA-seq data: purple squares, naive hPSCs and nHyCs from this study; black squares, naive hPSC-derived trophectoderm (nTE) and CT (nCT)21; vermillion squares, naive hPSCs and RACL cells23; triangles, primed hPSCs and primed hPSC-derived G6 PDGFRA+ cells; crosses, primed hPSCs and definitive endoderm35; and diamonds, first-trimester primary CT21. i, Signalling pathways to specify the three cell types of blastocyst. Hyp, hypoblast; aPKCi, aPKC inhibitor; FGFi, FGF inhibitor; TGF-βi, TGFβ inhibitor. n values show biologically independent experiments. Scale bars, 100 µm (b) and 50 µm (c, f and g).

(p) SMAD1/5/9, pSTAT3 and pMAPK were upregulated, pSMAD2 was downregulated (Extended Data Fig. 3b). We therefore selected seven factors (7F) as candidates for chemical hypoblast specification: BMPs (a pSMAD1/5/9 activator), IL-6 (a pSTAT3 activator), FGF4, A83-01 (a pSMAD2 inhibitor and ALK4/5/7 inhibitor) and XAV939 (a WNT/β-catenin inhibitor and tankyrase inhibitor) along with PDGF-AA and retinoic acid, which work in mice for hypoblast specification30–32 (Fig. 2a). 7F induced the expression of PDGFRA and hypoblast genes in multiple naive hPSC cell lines (H9, H1, induced pluripotent stem cells (iPSCs)) but not in primed hPSCs (Fig. 2b,c, Extended Data Fig. 3c–g and Supplementary Fig. 1). The transcriptome of naive 7F-PDGFRA+ cells was consistent with naive G6-PDGFRA+ cells (Extended Data Fig. 3h). A correlation analysis

with human pre-implantation embryos7 revealed that they correlated most prominently (Fig. 2d). We concluded that naive hPSC-derived PDGFRA+ cells overexpressing GATA6 or manipulated chemically to activate relevant signalling pathways progress into a hypoblast-like state, and we refer to these cells as nHyCs. We identified that the transcription factors FOXA2, HNF4A, and SP8 (Supplementary Table 3 and Extended Data Fig. 3i,j) and cell surface markers ANPEP (also known as CD13) and CEACAM1 (Extended Data Fig. 4a,b) mark nHyCs. Flow cytometry confirmed that ANPEP and CEACAM1 were highly expressed in G6-nHyCs and 7F-nHyCs but not in naive hPSCs, primed cells, naive hPSCs in a primed medium (FGF2/ TGFβ), definitive endoderm cells or mouse hypoblast (Extended Data Fig. 4c–f). Nature | Vol 626 | 8 February 2024 | 359

a

Naive (WT)

D0 N2B27 +DOX D2 N2B27

Naive(WT) + Naive-GFP (G6-OE)

Naive-GFP (G6-OE)

N2B27 N2B27 +DOX +DOX +IL-6 No aggregation N2B27 +IL-6

b

Naive(WT) + Naive-GFP(G6-OE) GFP OCT3/4

PAR6

Naive(WT) GFP OCT3/4

Merge

PAR6

Merge

D0

D2

D4 N2B27

N2B27

D4

D6

HNF4A

0.024 0.012 N

G+ G– G– WT + G6 WT

N

P

G+ G– G– WT + G6 WT

Fig. 3 | Naive hPSCs and nHyCs generate bilaminoids. a, Bilaminar embryo-like aggregates (bilaminoids) generated by the mixture of naive hPSCs and nHyCs. Aggregates were cultured without Matrigel. DOX was added for the first 2 days. IL-6 was added from day 0 to 4 where indicated. Naive(WT), WT naive hPSCs; Naive-GFP(G6-OE), GFP-expressing naive hPSCs expressing GATA6 under DOX treatment. b, Immunofluorescence images of cell aggregates. Green, Naive-GFP(G6-OE); purple, OCT3/4; white, PAR6; blue, DAPI. n = 3. c, qPCR analysis of aggregates on days 2 and 4. Cell aggregates were sorted by GFP on

FGF/BMP for hypoblast specification During embryonic development, signalling pathways act in concert to promote specification. Accordingly, removing FGF4 or BMP4 from 7F medium substantially decreased PDGFRA expression (Extended Data Fig. 4g) and adding activin A or CHIR99021 abolished PDGFRA+ cells (Extended Data Fig. 4h). nHyCs were induced even when we removed vitamin A and retinoic acid (Extended Data Fig. 4i), suggesting that, contrary to the mouse hypoblast, the human hypoblast does not require retinoic acid for its specification. FGF4/BMP4 complemented with A83/ XAV (4F) or without A83/XAV (2F), albeit at a low efficiency, successfully induced hypoblast gene expression and nHyCs (Fig. 2e and Extended Data Fig. 4j–m), which had strong correlations with hypoblasts of the blastocyst stage, similar to G6-nHyCs and 7F-nHyCs (Fig. 2d). To assess the effects of these molecules on hypoblast specification directly from the ICM of blastocysts, non-human-primate common marmoset ICM was cultured using 7F or 4F medium, or inhibitors of the FGF/BMP pathways (PD0325901/LDN-193189) and A83/XAV as a control (Fig. 2f). On day 3 of culture, the 4F colonies were flatter and contained larger cuboidal cells (Fig. 2f). SOX17+ hypoblast-like cells formed in 4F and 7F medium but not in the control medium (Fig. 2g). These observations suggest a crucial role for BMP/FGF signalling in hypoblast specification from the marmoset ICM while, in mouse ES cells, 7F did not induce PDGFRA or Sox733, in contrast to activin A + CHIR99021/LIF (ACL)34 or activin A + retinoic acid30 (Extended Data Fig. 5a,b). These data indicate that, in contrast to transcription factors of which the hierarchy and functions appear to be conserved between humans and mice, signalling may be common between humans and marmosets but differs with mice. 360 | Nature | Vol 626 | 8 February 2024

PODXL F-actin GATA4

aPKC F-actin GATA4

0.00005 D2 D4 D2 D4 D2 D4

P

THY1

0.00010

0.003

G+ G– G– WT + G6 WT

P

G+ G– G– WT + G6 WT

KLF17

0.006

0.05 N

N

G+ G– G– WT + G6 WT

NANOG

0.10

P

D2 D4 D2 D4 D2 D4

P

Naive(WT)

Naive(WT) + Naive-GFP (G6-OE)

0.003 D2 D4 D2 D4 D2 D4

N

D2 D4 D2 D4 D2 D4

0.0008

d

FOXA2

0.006

N

D2 D4 D2 D4 D2 D4

SOX17

0.0016

D2 D4 D2 D4 D2 D4

Expression normalized to GAPDH

c

P

G+ G– G– WT + G6 WT

days 2 and 4. G+, GFP+; G−, GFP−; WT + G6-OE, mixed aggregates of Naive(WT) and Naive-GFP(G6-OE) cells; WT, aggregates of Naive(WT) cells only; N, naive hPSCs; P, primed hPSCs. n = 2. d, Immunofluorescence images of polarization markers in aggregates on day 4. Purple, PODXL; blue, F-actin; yellow, GATA4; white, DAPI (top); yellow, aPKC; white, F-actin; purple, GATA4; blue, DAPI (bottom). n = 2. n values show biologically independent experiments. Data are mean (c). Scale bars, 20 µm (b) and 50 µm (d).

Human hypoblast lineage cells are reported to be induced from naive hPSCs in RPMI with ACL (RACL)23. RACL induced PDGFRA+ cells by day 7 but not some other hypoblast markers (that is, CEACAM1, HNF4A, FOXA2, SP8, SOX17 or KLF4), in contrast to 7F-nHyCs and 4F-nHyCs (Extended Data Fig. 5c,d). The transcriptome of RACL cells23 appeared to be more like post-implantation-stage cells, like primed-derived cells (Extended Data Fig. 5e–g). Furthermore, while PCA combined with single-cell RNA-seq (scRNA-seq) data of human embryos5 indicated that nHyCs and hypoblasts had similar gene expression profiles, RACL and primed-derived cells had closer expression profiles with post-implantation cells35 (Fig. 2h and Extended Data Fig. 5h), suggesting that nHyCs closely resemble the pre-implantation, blastocyst-stage hypoblast, a tissue that supports the epiblast development.

Generation of bilaminoids During the peri-implantation period, non-polarized naive epiblast acquires apical–basal polarity, concomitantly loses naive pluripotency to create the pro-amniotic cavity and, finally, forms both the post-implantation epiblast and amnion cells. Meanwhile, the hypoblast differentiates into visceral endoderm and yolk sac endoderm cells. As the visceral endoderm and post-implantation epiblast, having lost the naive pluripotent state, generate the bilaminar disc together, we aimed to model their intertwined development by culturing naive hPSCs (naive, wild type (WT)) with naive hPSCs overexpressing GATA6 under DOX treatment (Naive(G6-OE)) on a microwell array36 (Fig. 3a). To mark aggregated cells, GFP or DsRed was introduced into naive hPSCs (Naive-GFP and Naive-DsRed, respectively). Aggregates generated by a mixture of Naive(WT) and Naive-GFP(G6-OE) cells were

more spherical, consistent with the epithelial nature of hypoblast tissues (Extended Data Fig. 6a,b). While a mixture of Naive(WT) and Naive-GFP(G6-OE) cells was observed on day 0, Naive-GFP(G6-OE) after DOX treatment(called nHyCs(G6-OE)) relocated to the outer edge on day 2, as is typically observed in late blastocysts after maturation (Fig. 3b and Extended Data Fig. 6c), such that half of the aggregates were surrounded completely (Extended Data Fig. 6d). Time-lapse experiments confirmed the progressive segregation of GFP (nHyCs(G6-OE)) and DsRed (naive hPSC-derived epiblast-like cells (nEpiCs)) cells (Extended Data Fig. 6e,f). Only a few GFP+ cells were inside the aggregates on day 4 but were probably not hypoblast-like cells given their lack of SOX17 expression (Extended Data Fig. 6g). Previous reports of human embryos suggested that, between days 7 and 10, epiblast and hypoblast cell numbers increase from around 20–40 to 80–100 and from about 20–50 to 60–90, respectively6,37,38. Similarly, nHyC(G6-OE) and nEpiC cell numbers and aggregate size increased during differentiation (Extended Data Fig. 6h,i). GATA6 total expression in nHyCs(G6-OE) on day 2 after DOX treatment, at a similar level to blastocysts7, was higher than in nEpiCs (Extended Data Fig. 6j,k). nHyCs(G6-OE) in day 2 aggregates upregulated hypoblast genes and downregulated pluripotency-related genes, whereas nEpiCs expressed naive or epiblast genes (Fig. 3c and Extended Data Fig. 6j,l). We therefore concluded that nHyCs and nEpiCs self-organize and express markers like the late human blastocysts. We next analysed the apical–basal polarity of nEpiCs. Consistent with a blastocyst-like stage, PAR6 had not accumulated on day 2, (Fig. 3b). However, by day 4, around 20% of aggregates surrounded by nHyCs(G6-OE) accumulated PAR6 at the centre (Fig. 3b and Extended Data Fig. 6d). Polarized nEpiCs on day 4 gradually formed a rosette-like structure, which we refer to as bilaminoids, wherein PODXL and aPKC were localized together with F-actin (Fig. 3d). Lifeact—a small peptide with an affinity for actin microfilaments (F-actin)39—accumulated in the middle of the aggregates around 64 h after GATA6 induction (Extended Data Fig. 6m). Consistent with a pre- to post-implantation transition, nEpiCs showed a gradual decrease in KLF17 expression (naive pluripotency gene) and increases in THY1, DNMT3B and SFRP2 expression (early post-implantation epiblast genes)5,29,40 (Fig. 3c and Extended Data Fig. 6j). We also observed bilaminoids made by naive hPSCs and sorted naive PDGFRA+ cells induced by GATA6, 7F or 4F on laminin511-E8 (Extended Data Fig. 6n). Although primed G6-PDGFRA+ cells, RACL cells and definitive endoderm cells with either naive or primed hPSCs also surrounded epiblast cells, none generated a polarized cavity (Extended Data Fig. 6n). 7F-PDGFRA+ and G6-PDGFRA+ cells together with Naive(WT) cells generated bilaminoids with similar efficiency but less effectively compared with the mixture of Naive(WT) and Naive-GFP(G6-OE) cells, probably due to damages from flow cytometry (Extended Data Fig. 6p).

Epiblast progression via TB-secreted IL-6 Naive hPSCs can differentiate into trophectoderm by blocking FGF and TGF-β/activin signalling pathways21,22 and can generate blastocyst-like structures (blastoids) under TB induction medium containing PD03 and A8316,17. Although we did not use PD03 and A83 for bilaminoid induction, we examined whether TBs appeared in the bilaminoids. Indeed, they were not found in bilaminoids, although a few GATA2+ cells were detected in incomplete aggregates without an amniotic cavity (Extended Data Fig. 6q). To quantify TB-like cells (nTBs), we performed flow cytometry and identified HAVCR1+ENPEP+ nTBs4,21,22. However, less than 1% were HAVCR1+ENPEP+ nTBs in bilaminoids, suggesting that they, in contrast to blastoids, do not contain TB-like cells (Extended Data Fig. 6r). These results were confirmed using two other independent iPSC lines (Extended Data Fig. 6s,t). We next analysed the role of TBs in epiblast development by co-culturing Naive(WT) + Naive(G6-OE) with nTBs that were separately cultured on a Transwell plate (Fig. 4a). nEpiC proliferation was enhanced

in the presence of nTBs (Fig. 4b and Extended Data Fig. 7a), resulting in larger bilaminoids (Extended Data Fig. 7b). Although the efficiency of generating aggregates surrounded by nHyCs was similar for bilaminoids with and without nTBs (around 50%; Extended Data Fig. 7c,d), the amniotic cavity formed more efficiently and to a larger size with nTBs (from 20% to 40%; Fig. 4c–e and Extended Data Fig. 7e). This effect was confirmed using two other naive hPSC lines (Extended Data Fig. 7f). As bilaminoids and nTBs were separately cultured, we hypothesized that TBs promote epiblast proliferation and accelerate pro-amniotic cavity formation through secreted factors. As previously reported, IL-6 and PDGFA are expressed in TBs41 (Extended Data Fig. 7g). When IL-6 or PDGFA were added to the culture of bilaminoids, they efficiently enhanced pro-amniotic cavity formation by day 4 (Extended Data Fig. 7h). Furthermore, JAK inhibitor treatment negated the positive effects of nTBs (Extended Data Fig. 7i). We next knocked out IL6 in naive hPSCs (Extended Data Fig. 7j,k). IL6-knockout (KO) naive hPSCs differentiated into trophectoderm (Extended Data Fig. 7l,m), but these cells did not enhance bilaminoid growth and cavitation (Fig. 4f–h and Extended Data Fig. 7n). Finally, to determine whether IL-6 acts on nEpiCs or nHyCs, we activated JAK/STAT3 signalling in nEpiCs or nHyCs using the GP130/GCSFR chimeric receptor (Y118F)42. Both cell types activated STAT3 signalling (Extended Data Fig. 7o,p) but bilaminoids formed pro-amniotic cavities more efficiently by day 4 when STAT3 signalling was specifically activated in the nEpiCs (Fig. 4i and Extended Data Fig. 7q,r). We concluded that nTB-secreted IL-6 activates STAT3 signalling in nEpiCs to support proliferation and pro-amniotic-like cavity formation. This positive effect by IL-6 was also observed in the bilaminoids generated by 7F-nHyCs (Extended Data Fig. 7u).

Mesoderm-like cells emerge in bilaminoids After forming the pro-amniotic cavity and bilaminar disc, a subset of epiblast cells engages in gastrulation. By day 6, nEpiCs surrounded by nHyCs expressed TBXT (T) and primitive-streak-related genes (Fig. 4j and Extended Data Fig. 7v). Importantly, without nHyCs, cavities did not form, and mesoderm genes were not induced even in the presence of IL-6 and nTBs (Fig. 4j and Extended Data Fig. 7v). By contrast, the aggregates surrounded by 7F-nHyCs also contained cavities and T+ cells at day 6 (Extended Data Fig. 7w). Moreover, nTBs increased the efficiency of bilaminoids generated by 7F-nHyCs and the pro-amniotic cavity volume (Extended Data Fig. 7w). To induce mesoderm, the amniotic ectoderm is essential in human12. Co-culturing with G6- or 7F-nHyCs on Transwell plates, we confirmed that primed hPSCs to differentiate into T+ mesoderm cells 2 days after amnion-like cells emerged (Extended Data Fig. 7x). Furthermore, we observed GATA3+, TFAP2A+ or ISL1+ cells (amnion markers) in day 6 bilaminoids (Extended Data Fig. 7y). We concluded that nHyCs have a crucial role in regulating the expression of gastrulation-related genes in nEpiCs.

Single-cell transcriptomics of bilaminoids We identified the cell types of bilaminoids using scRNA-seq (197 cells from 23 bilaminoids; Extended Data Fig. 8a and Supplementary Table 4) and benchmarked them against a reference human embryo dataset2,3,5,8,16 together with recently published human embryo models12,14–17. We generated an integrated uniform manifold approximation and projection (UMAP), as proposed previously43, which clustered each cell type of the embryos as hypoblast, epiblast, primitive streak, mesoderm, amnion, primordial germ cells (PGCs), extraembryonic mesoderm, TB and ICM (Fig. 5a and Extended Data Fig. 8b,c). We confirmed that our clusters match with reported annotations of embryos and embryo models (Extended Data Fig. 8d). As TBs and amnion cells share many common genes, we further analysed whether our clustering separated them properly. We observed that the amnion cell clusters correlate with the amnion strongly but not with the trophectoderm (Extended Data Nature | Vol 626 | 8 February 2024 | 361

N2B27 + Dox(D0-D2) D6

nTB(+)

100

40 20

25

nTB

0

– WT 1 2 IL6 KO n = 122 195 203 205

40

D0 D2

D4

D6

20 0

nTB(WT)

nTB(KO 1)

nTB(KO 2)

1,000 100 10 nTB –

+ D6 36

P < 0.0001 10,000

P = 0.002

1,000 100 10 nTB – WT 1 2 IL6 KO n = 22 48 23 32

i Aggregates with a cavity/ surrounded aggregates (%)

50

Cavity volume (μm3)

Aggregates with a cavity/ surrounded aggregates (%)

75

60

Bilaminoid Bilaminoid + nTB

10,000

h P < 0.0001

d

PAR6 SOX17 OCT3/4 Naive(WT) + Naive(G6-OE) nTB(–)

100,000

n = 16

g

f

P < 0.0001

nTB(+)

nTB(+)

60

PAR6 GFP Naive(WT) + Naive(G6-OE) nTB(–)

nTB(–)

80

nTB – + – + – + D2 D4 D6

Cavity volume (μm3)

e

PAR6 SOX17 OCT3/4 Naive(WT) Naive(WT) + Naive(G6-OE)

86 200

D0

120

c

150 243

nTB

Bilaminoid Bilaminoid + nTB

n = 559 199 185

b

Naive(WT) + Naive(G6-OE)

Aggregates with a cavity/ surrounded aggregates (%)

Naive(WT) + Naive(G6-OE)

Number of nEpiCs

a

j 45

P = 0.0003

T PAR6 + GATA6 Naive(WT) + Naive(WT) Naive(G6-OE)

30 15 0 nEpiC WT nHyC G6 n = 127

WT WT-Y G6-Y G6 64 71

Fig. 4 | TB enhances epiblast progression through IL-6 paracrine signalling. a, Co-cultures of bilaminoids with naive hPSC-derived TBs (nTB) on Transwell plates. Co-cultures were performed from day 0 to day 4. b, The cell number of nEpiCs in each aggregate. Ten aggregates on each day were counted. n = 2 biologically independent experiments. c, Immunofluorescence images of aggregates on day 6. Blue, PAR6; yellow, SOX17; purple, OCT3/4; white, DAPI. n = 2 biologically independent experiments. d, The efficiency of cavity formation. n = 2 biologically independent experiments. e, Three-dimensional images and volume of the amniotic cavity in each aggregate on day 6. Purple, amniotic cavity; yellow, PAR6; green, naive-GFP(GATA6); blue, DAPI. n = 3 biologically independent experiments. Statistical analysis was performed using two-tailed Mann–Whitney’s U-tests. f, Immunofluorescence images of aggregates using IL6-KO naive hPSCs on day 6. Blue, PAR6; purple, SOX17; yellow, OCT3/4; white, DAPI. nTB(−), bilaminoid without nTB; nTB(WT), bilaminoid with nTB(WT); nTB(KO 1 or 2), bilaminoid with nTB (IL6-KO two clones (1 or 2)) (Extended Data Fig. 7j). n = 4 biologically independent experiments. g, The

efficiency of cavity formation on day 6 as in f. n = 4 biologically independent experiments. Statistical analysis was performed using two-tailed Fisher’s exact tests. h, The volume of the amniotic cavity of each aggregate on day 6 as in f. n = 4 biologically independent experiments. Statistical analysis was performed using Kruskal–Wallis and Dunn’s multiple-comparisons test. i, The efficiency of cavity formation in bilaminoids after STAT3 activation on day 4. GP130/GCSFR chimeric gene (Y118F) activates STAT3 signalling by adding G-CSF. WT, Naive(WT); WT-Y, Naive(WT) with Y118F; G6, Naive(G6-OE); G6-Y, Naive(G6-OE) with Y118F. n = 3 biologically independent experiments. Two-tailed Fisher’s exact test. j, Immunofluorescence images of aggregates with nTB on day 6. Purple, T; blue, PAR6 and GATA6; white, DAPI. n = 5 biologically independent experiments. For d,e and g–i, the number of aggregates analysed for each group is shown. For the box plots in b,e and h, the centre line shows the median; the box limits show the 25th and 75th percentile range, and the whiskers show 1.5 × interquartile range (IQR). Data are mean ± s.e.m. (g and i) and mean (d). Scale bars, 50 µm (c,e,f and j).

Fig. 8e and Supplementary Table 5). Finally, we checked the annotation of our bilaminoids in this integrated UMAP (Fig. 5b and Extended Data Fig. 8f). Hypoblast, epiblast, primitive streak, mesoderm and amnion cells were reproducibly present on day 6, whereas TB cells were not. Each cluster expressed key cell-type-specific marker genes (Fig. 5c and Supplementary Table 6). Notably, we noticed that a subpopulation of nHyCs in bilaminoids classified by UHC expressed anterior visceral endoderm marker genes (Extended Data Fig. 8g). PCA and contributed genes also suggested that there were anterior-visceral-endoderm-like cells in the bilaminoids on day 6 (Extended Data Fig. 8h).

anterior visceral endoderm (Fig. 5e and Extended Data Fig. 8j). Further immunostaining of the anterior visceral endoderm markers DKK1 and LEFTY confirmed this positional information (Extended Data Fig. 8k). To check whether anterior visceral marker genes were functional, we overexpressed OTX2 or DKK1 in nHyCs, which reduced T expression along with other mesoderm genes (Fig. 5f and Extended Data Fig. 8l–n), indicating that nHyCs control anterior–posterior axis formation and patterns epiblast differentiation. We further concluded that a subpopulation of nHyCs inhibits and thereby patterns the expression of gastrulation-related genes in nEpiCs.

Anterior–posterior axis formation in bilaminoids

nHyCs support epiblast progression

During mouse embryogenesis, a subpopulation of hypoblasts secretes anteriorization factors to guide anterior–posterior axis formation by restricting gastrulation to the posterior epiblasts44. To track CER1 expression, one of the anteriorization factors, we generated CER1-H2B-GFP knockin naive hPSCs (Extended Data Fig. 8i). We detected CER1–H2B–GFP+ cells in a part of the nHyC(G6-OE) bilaminoids on day 6 and T+ cells located away from them in nEpiCs (Fig. 5d). Similarly, T+ cells did not contact OTX2+ cells in nHyCs, which also marks the

Next, we analysed the interaction between epiblast and hypoblast using the scRNA-seq data. In mice, GATA factors induce laminins in hypoblasts25, and basal lamina formation separates hypoblasts from epiblasts45. Our scRNA-seq data show that LAMA1, LAMB1 and LAMC1 were strongly expressed in nHyCs(G6-OE) (Extended Data Fig. 9a), and laminins formed at the boundary between nHyCs(G6-OE) and nEpiCs in the bilaminoids (Fig. 5g), therefore reflecting a basement membrane between the hypoblast and epiblast cells. Laminin is known to interact

362 | Nature | Vol 626 | 8 February 2024

a

b

c

Integrated UMAP

Hyp

Epi

PriS Mes Am CTSE, GATA4, GATA6, CDH2, APOA2, COL4A1, PDGFRA, FOXA2

Bilaminoid D6

UMAP 2

Hyp Epi PriS Mes Am PGCs ExM TB ICM Prelineage

Hyp Epi PriS Mes Am PGCs ExM TB ICM Prelineage

NANOG, DPPA5, POU5F1, DPPA4, SFRP2, PMAIP1 TBXT, NNAT, DCLK1, GAL, RBP1, MESP1, KDR, VIM, HAND1, APLNR GABRP, VTCN1, EPAS1, GATA3

UMAP 1 Expression

f

Tamoxifen

Naive(WT) + Naive-GFP(G6-OE) OTX2-ERT2

OTX2 T GATA4

120 60 0

–

CER1 GFP+ nEpiC nHyC OTX2 T GFP

+

Laminin GATA4 F-actin

h Control

−2 0 2 PAR6 SOX17 OCT3/4 LAMB1 KO 1 LAMB1 KO 2

120 60

0°

T+

P < 0.0001

180

g

0

T+ OTX2+ nEpiC nHyC P < 0.0001

100 75 50

25 0 Tamoxifen

–

n = 55

Naive (WT)

Naive (WT) + Naive (G6-OE)

+ 78

Aggregates with a cavity/surrounded aggregates (%)

0°

P = 0.00127

Angle (º)

180

e

Bilaminoids containing T+ cells (%)

CER1–H2B–GFP T GATA6 + PAR6 Angle (º)

d

P < 0.0001

60 40 20 0

Control n = 209

KO 1 236

KO 2 216

Fig. 5 | Global gene expression profiles of individual cells in bilaminoids. a, UMAP analysis of integrated datasets of bilaminoids, published human embryos2,3,5,8,16 and stem-cell-based embryo models12,14–17 (Extended Data Fig. 8b–d). PriS, primitive streak; Mes, mesoderm; Am, amnion; ExM, extraembryonic mesoderm. b, Cells from day 6 bilaminoids highlighted on the UMAP shown in a. c, The relative expression values of the top 50 differentially expressed genes of bilaminoids and representative genes. d, The anterior– posterior axis of bilaminoids on day 6. Yellow, CER1–H2B–GFP (nuclei); purple, T; blue, GATA6 + PAR6; white, DAPI. Yellow and purple arrowheads indicate CER1–H2B–GFP+ and T+ nuclei, respectively. Ten aggregates expressing T and H2B–GFP were analysed. n = 3 biologically independent experiments. Statistical analysis was performed using two-tailed Welch’s t-tests. e, Anterior–posterior axis of bilaminoids on day 6. Yellow, OTX2; purple, T; blue, GATA4; white, DAPI. The yellow and purple arrowheads indicate OTX2+ and T+ nuclei, respectively. A total of 14 aggregates expressing T and OTX2 was analysed. n = 3 biologically independent experiments. Statistical analysis was performed using two-tailed

Mann–Whitney U-tests. f, OTX2 overexpression in nHyCs. Bilaminoids were generated by Naive(WT) + Naive-GFP(G6-OE) cells containing tamoxifeninducible OTX2 (OTX2-ERT2). Purple, OTX2; yellow, T; green, GATA6; white, DAPI. n = 2 biologically independent experiments. Statistical analysis was performed using two-tailed Fisher’s exact tests. g, Immunofluorescence images of LAMININ in bilaminoids on day 4. Yellow, laminin; green, GATA4; purple, F-actin; white, DAPI. n = 2 biologically independent experiments. h, Aggregates generated by Naive(WT) and Naive LAMB1 KO(G6-OE) (two clones (1 and 2)) cells (Extended Data Fig. 9c). Blue, PAR6; purple, SOX17; yellow, OCT3/4; white, DAPI. n = 3 biologically independent experiments. Statistical analysis was performed using two-tailed Fisher’s exact tests. For f and h, the number of aggregates analysed for each group is shown at the bottom. For the box plots in d and e, the centre line shows the median; the box limits show the 25th and 75th percentile range, and the whiskers show 1.5 × IQR. Data are mean ± s.e.m. (h) and mean (f). Scale bars, 50 µm (d–f) and 20 µm (g and h).

through integrin heterodimers on cell surface receptors46. We found that the integrin α6β1, which is required for the formation of rosette structure in mice47, is expressed in nEpiCs (Extended Data Fig. 9b), suggesting that, like in mice, laminin in nHyCs may act through integrins for rosette formation in humans. We therefore generated LAMB1-KO hPSC lines (Naive LAMB1-KO) (Extended Data Fig. 9c). Naive LAMB1-KO(G6-OE) cells differentiated into the hypoblast lineage (Extended Data Fig. 9d,e) but did not surround nEpiCs as a single cell layer nor did they support pro-amniotic cavity formation (Fig. 5h). We concluded that, like in mice, laminins secreted by the human hypoblast support epiblast differentiation and morphogenesis. We also noticed that nHyCs expressed BMP genes, NODAL and WNT11 (Extended Data Fig. 9f) and nEpiCs expressed receptors related to BMP, FGF and WNT (Extended Data Fig. 9g). To examine how BMP, NODAL and WNT signalling affects mesoderm induction, we added activators and inhibitors from day 4 and found that BMP, WNT or activin inhibition reduces the appearance of gastrulation-related genes in nEpiCs on day 6 (Extended Data Fig. 9h).

flattened amniotic epithelium expressing the amnion markers ISL1 and GATA3 (Fig. 6a and Extended Data Fig. 10a–c). Notably, we also observed flattened epithelial cells expressing BLIMP1 and TFAP2C, markers for PGCs (Fig. 6b and Extended Data Fig. 10d), and CD34+ERG+ cells, markers for haematoendothelial progenitor (HEP) cells (Fig. 6c). We purified VTCN1+ cells, BLIMP1+TFAP2C+ (BTAG) cells and CD34+ cells as single cells using flow cytometry (Extended Data Fig. 10e). Integrated UMAP with human embryo cells showed that VTCN1+, CD34+ and BTAG cells clustered with embryonic amnion cells, PGCs and HEP cells, respectively (Fig. 6d–f). They also expressed embryonic amnion, PGC or HEP marker genes similar to published in vivo and in vitro controls (Fig. 6g). Although detailed characterization of these emerging cell types is necessary, this observation gives an early indication that bilaminoids support the progression of the epiblast from a blastocyst-like (naive state) to a post-implantation-like stage that is permissive for lineage specification (Fig. 6).

Lineage specification in bilaminoids Finally, we cultured the bilaminoids until day 9. The amniotic cavity of the bilaminoids enlarged, and nEpiCs partially differentiated into a

Discussion Here we highlight the crucial mechanistic roles of the two extraembryonic tissues—hypoblast and TB—to guide the progression and patterning of naive hPSCs into the post-implantation epiblast stage, thereby Nature | Vol 626 | 8 February 2024 | 363

b

Bilaminoid D9 ISL1 GATA3 GATA6

ISL1

GATA3

Bilaminoid D9 CD34 ERG GATA6

BLIMP1–tdTomato

GATA6

e

AmLCs (bilaminoid D9) Amnion (CS7) AMLCs (ref. 12) Transwell AMLCs (ref. 12) Amnion

g TFAP2C–GFP

f

GATA6

Bilaminoid D9: Naive(BTAG)+Naive(GATA6) TFAP2C–GFP BLIMP1–tdTomato

c

d

DAPI

CD34 ERG

−6

0 6 Value

PGCLCs (bilaminoid D9) PGCLC (ref. 12)

Pluripotency

PGCs

HEPLCs (bilaminoid D9) HEP (CS7)

Haematoendothelial

GABRP VTCN1 HEY1 HAND1 GATA3 ISL1 IGFBP3 EPAS1 POU5F1 NANOG SOX2 TFAP2C PRDM1 PRDM14 NANOS3 SOX17 TBXT CXCR4 CD34 KDR ERG MEF2C LMO2 TAL1 CDH5 PECAM1

a

AmLCs (bilaminoid D9) CS7 amnion (ref. 8) AMLCs (ref. 12) Transwell AMLCs (ref. 12) PGCLCs (bilaminoid D9) PGCLCs (ref. 12) HEPLCs (bilaminoid D9) CS7 HEPs (ref. 8) Naive (ref. 17) Primed (ref. 17)

Fig. 6 | Bilaminoids recapitulate human pregastrulation. a, Amnion marker expression in bilaminoids on day 9. Blue, ISL1; yellow, GATA3; purple, GATA6; white, DAPI. n = 3 biologically independent experiments. b, PGC marker expression in bilaminoids on day 9. Bilaminoids were generated by BLIMP1tdTomato and TFAP2C-GFP double-knockin Naive(BTAG) and Naive(G6-OE) cells. Green, TFAP2C–GFP; purple, BLIMP1–tdTomato; white, DAPI. The yellow arrowheads indicate BTAG double-positive cells. n = 4 biologically independent

experiments. c, HEP marker expression in bilaminoids on day 9. Blue, CD34; yellow, ERG; purple, GATA6; white, DAPI. n = 3 biologically independent experiments. d–f, Amnion and amnion-like cells (AmLCs from bilaminoids and AMLCs from the model in Zheng et al.12) (d), PGC-like cells (PGCLCs) (e) and HEP and HEP-like cells (f) on UMAP for integrated datasets of bilaminoids on day 9, and published data as in Fig. 5a. g, Relative expression values of each tissuespecific marker gene in each cell type. Scale bars, 50 µm (a–c).

enabling them to generate subsequent lineages (for example, PGC-like and HEP cells) in a manner mimicking human embryogenesis. Although naive hPSCs were reported to differentiate into the hypoblast lineage23, reanalysing the RNA-seq data revealed that they lack several pre-implantation hypoblast markers, suggesting that they resemble extraembryonic endoderm or mesoderm cells at the post-implantation stage. Thus, our study demonstrates robust and reproducible induction of pre-implantation hypoblast-like cells. In particular, FGF and BMP plus inhibition of WNT and activin A signalling pathways were critical for inducing naive hPSCs to hypoblasts specific to the pre-implantation-stage blastocyst. Our findings extend our understanding of the signalling pathways essential to specifying all three cell types of the blastocyst. Namely, naive epiblast can be maintained with FGF and aPKC inhibition, the trophectoderm with FGF and TGFβ inhibition, and the hypoblast with FGF and BMP4 activation plus TGFβ inhibition (Fig. 2i). These data also reveal that the signalling pathways that are required to induce the hypoblast of blastocysts in humans differ significantly from those in mice using either ACL34 or activin A + retinoic acid30, akin to the differences in trophectoderm induction. However, hypoblast induction with the transcription factors GATA6 and GATA4 induces naive hPSCs to hypoblast, similar to in mice. Although transgene copy numbers and insertion sites may be variable because we used the PB system, we reproducibly obtained more than 80% PDGFRA+ cells from five independently established DOX-inducible GATA6 H9 hPSCs (Extended Data Fig. 2a). At the same time, our data show that the levels and duration of GATA6 overexpression are critical. To recapitulate a more in vivo like scenario and determine the in vivo function and contribution of G6-nHyCs and 7F-nHyCs, we performed mouse–human interspecies chimera assays. Whereas naive hPSCs integrated into the ICM, injected 7F- and G6-nHyCs contacted the ICM

and expressed SOX17, similar to the late morulae–early blastocysts of mouse embryos, and never contributed to the epiblast lesion (Extended Data Fig. 10g–h). Furthermore, 7F- and G6-nHyCs contributed to the visceral endoderm and extraembryonic lesions in embryonic day 6.5 embryos, suggesting that both chemically and genetically induced nHyCs are functionally competent to form mouse–human chimera (Extended Data Fig. 10i–l). Notably, 7F-nHyCs contributed to the mouse visceral endoderm more efficiently than G6-nHyCs (Extended Data Fig. 10l). Although we titrated the DOX concentration, high levels of GATA6 mRNA may have resulted in off-target effects and caused some functional disadvantages. As 7F induction is a non-genetic chemical induction method, 7F may enable naive hPSCs to differentiate into hypoblast under more physiologically relevant conditions compared with GATA6 overexpression. The hypoblast-like cells that we generated efficiently and reproducibly assemble into bilaminoids, proceeding to mimic human peri-implantation development, including the formation of the pro-amniotic-like cavity and anterior-posterior patterning of the epiblast. We showed by genetic modulation that this patterning is caused by a DKK1/OTX2 hypoblast-like domain. Although naive hPSCs can differentiate into trophectoderm and TB, we did not detect TB-like cells in bilaminoids until day 6, except in incomplete aggregates without an amniotic cavity (Extended Data Fig. 6p), even though TBs may emerge in later stages. Moreover, as there has been no report about the early stages of in vivo human amnion just after implantation (Carnegie stage 5), we could only estimate the gene expression profiles of the emergent amnion from Carnegie stage 78, in vitro cultures of human embryos5 or primed hPSC-derived amnion-like cells12. Notably, the separated co-cultures of additional trophectoderm-like cells enhanced the formation of the pro-amniotic-like cavity and early

364 | Nature | Vol 626 | 8 February 2024

post-implantation epiblast growth (Extended Data Fig. 10f). This inducive effect by the trophectoderm is regulated in part by secreted molecules IL-6 and PDGF, as shown using both genetic- and chemical-based approaches. A recent report suggested that the in vitro early amnion expresses the AQP3 channel that may initiate amniotic cavity formation48. Furthermore, AQP3 is one of the STAT3-target genes predicted by transcription factor binding motif analysis49. Furthermore, assay for transposase-accessible chromatin with sequencing (ATAC–seq) data suggest that this predicted STAT3-binding site in AQP3 is open in both naive and primed hPSCs (Extended Data Fig. 7s). As our quantitative PCR (qPCR) data showed that AQP3 was upregulated in epiblast-like cells by co-culture with nTB (Extended Data Fig. 7t), further studies may confirm an IL-6 dependency. Recently, during revisions of this Article, stem-cell-based postimplantation models using in vitro epiblast- and hypoblast-like cells were reported50–53 (Supplementary Table 7). While the developmental window of our model extends from blastocyst to peri-gastrulation by starting with naive hPSCs that reflect day 5 pre-implantation epiblast and hypoblast, other models start from the post-implantation stage. Thus, our model covers a wider developmental time window from pre-implantation and precisely matches the natural developmental sequence and timing. Furthermore, considering that our bilaminoid model does not necessarily require genetic manipulation, it offers a flexible, alternative way for generating peri-implantation embryo models in vitro, with an efficiency that is comparable to the other models using RSeT and extended pluripotent stem cells (EPSCs)51,52 (Extended Data Figs. 6o and 7u,w and Supplementary Table 7). Importantly, functional assays with genetic modifications are almost impossible in human embryos but, using bilaminoids, we performed several lineage-specific gene modifications and identified interactions between these lineages. Finally, a limitation of our bilaminoids is that the amnion is covered by hypoblast when it should be in direct contact with the TB. Nevertheless, our study, together with the other human stem cell-based embryo models, will drive scientific discoveries in biomedical science.

Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-023-06871-2. 1.

Yan, L. et al. Single-cell RNA-seq profiling of human preimplantation embryos and embryonic stem cells. Nat. Struct. Mol. Biol. 20, 1131–1139 (2013). 2. Petropoulos, S. et al. Single-cell RNA-seq reveals lineage and X chromosome dynamics in human preimplantation embryos. Cell 165, 1012–1026 (2016). 3. Blakeley, P. et al. Defining the three cell lineages of the human blastocyst by single-cell RNA-seq. Development 142, 3151–3165 (2015). 4. Zhou, F. et al. Reconstituting the transcriptome and DNA methylome landscapes of human implantation. Nature 572, 660–664 (2019). 5. Xiang, L. et al. A developmental landscape of 3D-cultured human pre-gastrulation embryos. Nature 577, 537–542 (2020). 6. Deglincerti, A. et al. Self-organization of the in vitro attached human embryo. Nature 533, 251–254 (2016). 7. Stirparo, G. G. et al. Integrated analysis of single-cell embryo data yields a unified transcriptome signature for the human pre-implantation epiblast. Development 145, dev158501 (2018). 8. Tyser, R. C. V. et al. Single-cell transcriptomic characterization of a gastrulating human embryo. Nature https://doi.org/10.1038/s41586-021-04158-y (2021). 9. Warmflash, A., Sorre, B., Etoc, F., Siggia, E. D. & Brivanlou, A. H. A method to recapitulate early embryonic spatial patterning in human embryonic stem cells. Nat. Methods 11, 847–854 (2014). 10. Shahbazi, M. N. et al. Pluripotent state transitions coordinate morphogenesis in mouse and human embryos. Nature 552, 239–243 (2017). 11. Martyn, I., Kanno, T. Y., Ruzo, A., Siggia, E. D. & Brivanlou, A. H. Self-organization of a human organizer by combined Wnt and Nodal signalling. Nature 558, 132–135 (2018). 12. Zheng, Y. et al. Controlled modelling of human epiblast and amnion development using stem cells. Nature 573, 421–425 (2019).

13. Moris, N. et al. An in vitro model of early anteroposterior organization during human development. Nature 582, 410–415 (2020). 14. Liu, X. et al. Modelling human blastocysts by reprogramming fibroblasts into iBlastoids. Nature https://doi.org/10.1038/s41586-021-03372-y (2021). 15. Yu, L. et al. Blastocyst-like structures generated from human pluripotent stem cells. Nature 591, 620–626 (2021). 16. Yanagida, A. et al. Naive stem cell blastocyst model captures human embryo lineage segregation. Cell Stem Cell https://doi.org/10.1016/j.stem.2021.04.031 (2021). 17. Kagawa, H. et al. Human blastoids model blastocyst development and implantation. Nature https://doi.org/10.1038/s41586-021-04267-8 (2021). 18. Takashima, Y. et al. Resetting transcription factor control circuitry toward ground-state pluripotency in human. Cell 158, 1254–1269 (2014). 19. Theunissen, T. W. et al. Systematic identification of culture conditions for induction and maintenance of naive human pluripotency. Cell Stem Cell 15, 471–487 (2014). 20. Guo, G. et al. Epigenetic resetting of human pluripotency. Development 144, 2748–2763 (2017). 21. Io, S. et al. Capturing human trophoblast development with naive pluripotent stem cells in vitro. Cell Stem Cell 28, 1023–1039 (2021). 22. Guo, G. et al. Human naive epiblast cells possess unrestricted lineage potential. Cell Stem Cell 28, 1040–1056 (2021). 23. Linneberg-Agerholm, M. et al. Naive human pluripotent stem cells respond to Wnt, Nodal and LIF signalling to produce expandable naive extra-embryonic endoderm. Development https://doi.org/10.1242/dev.180620 (2019). 24. Schrode, N., Saiz, N., Di Talia, S. & Hadjantonakis, A. K. GATA6 levels modulate primitive endoderm cell fate choice and timing in the mouse blastocyst. Dev. Cell 29, 454–467 (2014). 25. Fujikura, J. et al. Differentiation of embryonic stem cells is induced by GATA factors. Genes Dev. 16, 784–789 (2002). 26. McDonald, A. C., Biechele, S., Rossant, J. & Stanford, W. L. Sox17-mediated XEN cell conversion identifies dynamic networks controlling cell-fate decisions in embryo-derived stem cells. Cell Rep. 9, 780–793 (2014). 27. Kataoka, H. et al. Expressions of PDGF receptor alpha, c-Kit and Flk1 genes clustering in mouse chromosome 5 define distinct subsets of nascent mesodermal cells. Dev. Growth Differ. 39, 729–740 (1997). 28. Murry, C. E. & Keller, G. Differentiation of embryonic stem cells to clinically relevant populations: lessons from embryonic development. Cell 132, 661–680 (2008). 29. Nakamura, T. et al. A developmental coordinate of pluripotency among mice, monkeys and humans. Nature 537, 57–62 (2016). 30. Cho, L. T. et al. Conversion from mouse embryonic to extra-embryonic endoderm stem cells reveals distinct differentiation capacities of pluripotent stem cell states. Development 139, 2866–2877 (2012). 31. Artus, J., Panthier, J. J. & Hadjantonakis, A. K. A role for PDGF signaling in expansion of the extra-embryonic endoderm lineage of the mouse blastocyst. Development 137, 3361–3372 (2010). 32. Vrij, E. J. et al. A pendulum of induction between the epiblast and extra-embryonic endoderm supports post-implantation progression. Development https://doi.org/10.1242/ dev.192310 (2022). 33. Artus, J., Piliszek, A. & Hadjantonakis, A. K. The primitive endoderm lineage of the mouse blastocyst: sequential transcription factor activation and regulation of differentiation by Sox17. Dev. Biol. 350, 393–404 (2011). 34. Anderson, K. G. V. et al. Insulin fine-tunes self-renewal pathways governing naive pluripotency and extra-embryonic endoderm. Nat. Cell Biol. https://doi.org/10.1038/ ncb3617 (2017). 35. Chu, L. F. et al. Single-cell RNA-seq reveals novel regulators of human embryonic stem cell differentiation to definitive endoderm. Genome Biol. 17, 173 (2016). 36. Rivron, N. C. et al. Blastocyst-like structures generated solely from stem cells. Nature 557, 106–111 (2018). 37. Shahbazi, M. N. et al. Self-organization of the human embryo in the absence of maternal tissues. Nat. Cell Biol. 18, 700–708 (2016). 38. Roode, M. et al. Human hypoblast formation is not dependent on FGF signalling. Dev. Biol. 361, 358–363 (2012). 39. Riedl, J. et al. Lifeact: a versatile marker to visualize F-actin. Nat. Methods 5, 605–607 (2008). 40. Di Stefano, B. et al. Reduced MEK inhibition preserves genomic stability in naive human embryonic stem cells. Nat. Methods 15, 732–740 (2018). 41. Meistermann, D. et al. Integrated pseudotime analysis of human pre-implantation embryo single-cell transcriptomes reveals the dynamics of lineage specification. Cell Stem Cell https://doi.org/10.1016/j.stem.2021.04.027 (2021). 42. Niwa, H., Burdon, T., Chambers, I. & Smith, A. Self-renewal of pluripotent embryonic stem cells is mediated via activation of STAT3. Genes Dev. 12, 2048–2060 (1998). 43. Zhao, C. et al. Reprogrammed iBlastoids contain amnion-like cells but not trophectoderm. Preprint at bioRxiv https://doi.org/10.1101/2021.05.07.442980 (2021). 44. Varlet, I., Collignon, J. & Robertson, E. J. Nodal expression in the primitive endoderm is required for specification of the anterior axis during mouse gastrulation. Development 124, 1033–1044 (1997). 45. Chazaud, C., Yamanaka, Y., Pawson, T. & Rossant, J. Early lineage segregation between epiblast and primitive endoderm in mouse blastocysts through the Grb2-MAPK pathway. Dev. Cell 10, 615–624 (2006). 46. Colognato, H. & Yurchenco, P. D. Form and function: the laminin family of heterotrimers. Dev. Dyn. 218, 213–234 (2000). 47. Bedzhov, I. & Zernicka-Goetz, M. Self-organizing properties of mouse pluripotent cells initiate morphogenesis upon implantation. Cell 156, 1032–1044 (2014). 48. Rostovskaya, M., Andrews, S., Reik, W. & Rugg-Gunn, P. J. Amniogenesis occurs in two independent waves in primates. Cell Stem Cell 29, 744–759 (2022). 49. Rouillard, A. D. et al. The harmonizome: a collection of processed datasets gathered to serve and mine knowledge about genes and proteins. Database https://doi.org/10.1093/ database/baw100 (2016).

Nature | Vol 626 | 8 February 2024 | 365

50. Weatherbee, B. A. T. et al. Pluripotent stem cell-derived model of the post-implantation human embryo. Nature https://doi.org/10.1038/s41586-023-06368-y (2023). 51. Pedroza, M. et al. Self-patterning of human stem cells into post-implantation lineages. Nature https://doi.org/10.1038/s41586-023-06354-4 (2023). 52. Liu, L. et al. Modeling post-implantation stages of human development into early organogenesis with stem-cell-derived peri-gastruloids. Cell https://doi.org/10.1016/j. cell.2023.07.018 (2023). 53. Oldak, B. et al. Complete human day 14 post-implantation embryo models from naive ES cells. Nature 622, 562–573 (2023). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

366 | Nature | Vol 626 | 8 February 2024

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. © The Author(s) 2023

Methods Data reporting The experiments were not randomized. The investigators were not blinded to the group allocation of experimental samples or the outcome assessment. No statistical methods were used to predetermine sample sizes. Ethics statement Our embryo model lacks TBs and does not intend to recapitulate the full conceptus. Thus, our models are considered to be non-integrated embryo models and are not considered to be human embryos according to the ISSCR. Our work fully complies with current ISSCR 2016 and 2021 guidelines and follows the Guidelines on the Utilization of Human Embryonic Stem Cells in Japan. The CiRA Ethics Committee, an internal committee at CiRA, approved our research plan for human ES cell research (CiRA08-08), human iPSC research (CiRA18-21) and recombinant DNA experiments (190438). The WiCell lines H1 and H9 were used under agreements 10-WO-0098 and 10-WO-0099 for a research program entitled “Understanding mechanisms of pluripotency”. Bilaminoid models were generated using H9 ES cells, 551B1 iPSCs and 1390G3 iPSCs. These cell lines were consented for use in this study. Human-to-mouse interspecies chimera research was approved by the Research Ethics Committee of the University of Tokyo, and was conducted after receiving approval from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) Japan after confirmation of compliance by the Specified Embryo Expert Committee. This approval includes the establishment of human iPSCs from peripheral blood samples. Signed informed consent was obtained from the volunteers before human peripheral blood samples were collected to establish iPSCs. The approved iPSC line, PB004, was used for interspecies chimera assays. Cell culture Cells were cultured under 5% O2 and 5% CO2. Human ES cell lines H1 and H9 (WiCell Research Institute) and human iPSCs (AdiPSCs18, 585B154 and 1390G355) were cultured on mouse embryonic fibroblasts (MEFs) (1 × 106 cells per six-well plate). Primed hPSCs were maintained in DMEM/F12 (08460-95, Nacalai Tesque) containing 20% Knockout Serum Replacement (10828028, Thermo Fisher Scientific), 1% non-essential amino acids (11140-050, Thermo Fisher Scientific), 4 ng ml−1 recombinant human basic fibroblast growth factor 2 (bFGF; NIB 47079000, Oriental Yeast) and 0.1 mM 2-mercaptoethanol (M3148, Sigma-Aldrich). Cultures were passaged every 5–7 days as small clumps using dissociation buffer containing 0.025% trypsin (15090-046, Thermo Fisher Scientific), 1 mg ml−1 collagenase IV (17104-019, Thermo Fisher Scientific), 20% Knockout Serum Replacement and 1 µM CaCl2. Naive hPSCs were maintained in t2iLGo medium, consisting of a chemically defined medium, N2B27 (NDiff 227, Y40002, Takara Bio) supplemented with 1 µM PD0325901 (PD03; 4192, Tocris), 1 µM CHIR99021 (CH; SML1046, Sigma-Aldrich), 10 ng ml−1 recombinant human LIF (hLIF; 300-05, Peprotech) and 3 µM Go6983 (Go; 2285, Tocris) as previously described18. The components of the N2B27 medium were DMEM/F12, Neurobasal medium, N2 and B2756. Naive hPSCs were passaged every 3–5 days using Accutase (A6964, Sigma-Aldrich). Resetting primed hPSCs to naive hPSCs by NANOG and KLF2 overexpression was performed as previously described18. In brief, PB vectors (2 µg) carrying DOX-inducible KLF2 or NANOG and a PB-M2rtTA expression vector (2 µg) were co-transfected with pBase helper plasmid (4 µg) using the Neon Transfection System (Program 14, Invitrogen). The medium was switched to t2iL plus DOX (1 µM) for resetting. Cells were split every 5–7 days after dissociation with Accutase. After 2 weeks, DOX was withdrawn, and the PKC inhibitor Go6983 (3 µM) was added (t2iLGo). Cells were maintained on MEF feeders t­hr­ou­gh­out.

Chemical conversion to naive hPSCs was performed as previously described20. Primed hPSCs (1 × 104 cells per cm2) were seeded onto MEF feeder cells under primed hPSC medium with 10 µM Y-27632. The medium was switched the next day to cRM-1 (N2B27, 1 µM PD03, 10 ng ml−1 hLIF, and 1 mM valproic acid sodium salt (P4543, Sigma-Aldrich)). On day 3, the medium was replaced with cRM-2 (N2B27, 1 µM PD03, 10 ng ml−1 hLIF, 2 µM Go and 2 µM XAV939; X3004, Sigma-Aldrich). Dome-shaped naive colonies were observed around 2 weeks after seeding. Reset cells were passaged and maintained on MEF feeders under t2iLGo. Chemical conversion to naive hPSCs using 5iLA was also performed as described previously19. Here, 2 × 105 cells per cm2 were seeded on MEF feeder cells under primed hPSC medium with 10 µM Y-27632. The medium was switched the next day to 5iLA medium (N2B27 plus 1 µM PD03, 1 µM CH, 1 µM WH-4-023 (H620061), 0.5 µM SB590885 (2650, R&D Systems), 10 µM Y-27632, 10 ng ml−1 hLIF and 20 ng ml−1 activin A (338-AC-010, R&D Systems). After conversion to naive hPSCs, the cells were maintained under t2iLGo on MEF feeder cells. Mouse ES cells were cultured on a gelatine-coated dish in 2iL (N2B27, 1 µM PD03, 3 µM CH and 10 ng ml−1 hLIF). Cells were passaged every 2–3 days using Accutase. Naive hPSCs form tightly packed small colonies and expressed GFP if carrying the EOS-GFP reporter, which consists of an OCT3/4 distal enhancer and an early transposon promoter18–20,57 (Extended Data Fig. 1a). Naive hPSCs expressed the naive-specific genes KLF17 and TFCP2L1 (Extended Data Fig. 1b,c) but primed and expanded PSCs did not58. All cell lines were routinely checked for mycoplasma contamination (Lonza–MycoAlert), and all samples analysed in this study were not contaminated.

GATA6 overexpression GATA6, GATA4 and SOX17 were cloned into a DOX-inducible PB vector coupled to a rtTA expression construct (KW110)59. PB-GATA6 vector (2 µg), PB-GATA4 vector (2 µg) or PB-SOX17 vector (2 µg), and pBase helper plasmid (2 µg) were transfected into naive or primed hPSCs using the Neon Transfection System (Program 20 for naive hPSCs; Program 14 for primed hPSCs). Then, 2 days later, G418 was added (200 µg ml−1) for about 2 weeks. Naive or primed hPSCs with inducible GATA6, GATA4 or SOX17 were maintained in naive or primed medium. For transgene induction, MEF feeder cells were removed by incubation on a gelatine-coated dish after dissociation to single cells. Then, 1 × 105 cells per cm2 were seeded into a dish coated with fibronectin (FC010, Millipore) or iMatrix-511 silk (Laminin511-E8) (892021, Matrixome). The serum medium consisted of GMEM (G5154, Sigma-Aldrich), FBS (10437028, Thermo Fisher Scientific), 2 mM l-glutamine (25030081 Thermo Fisher Scientific), 1 mM sodium pyruvate (11360-070, Thermo Fisher Scientific), NEAA and 0.1 mM 2-ME. Hypoblast induction by serum medium is shown in Fig. 1b and Extended Data Fig. 1d–h. Except for these experiments, all other analyses were performed in serum-free conditions. As a serum-free basal medium, we used the N2B27 medium (NDiff 227; Y40002, Takara Bio). The components of the N2B27 medium were DMEM/F12, Neurobasal, N2 and B2756. BSA is included in N2 and B27. For nHyC induction, 25 ng ml−1 recombinant human FGF4 (FGF4; 100-31) and 1 µg ml−1 heparin sodium (081-00131, Wako) were added to the basal medium. The medium was changed every day. Hypoblast specification using chemical components In brief, 5 × 104 per cm2 naive hPSCs were seeded onto laminin511-E8 in the N2B27 medium. Six factors, 25 ng ml−1 FGF4 (+1 µg ml−1 heparin sodium), 10 ng ml−1 recombinant human BMP4 (BMP4; 314-BP, R&D), 10 ng ml−1 recombinant human PDGF-AA (Peprotech, 100-13A), 1 µM XAV939, 3 µM A83-01 (2939, Tocris) and 0.1 µM retinoic acid (R2625, Sigma-Aldrich), were added on day 0. On day 2, the medium was switched to seven factors (six factors and 10 ng ml−1 recombinant human IL-6) (IL-6; 47066000, Oriental Yeast). In some experiments,

500 ng ml−1 recombinant human BMP2 (BMP2; 47304000, Oriental Yeast) or 50 ng ml−1 recombinant human BMP6 (BMP6; 120-06, Peprotech) was used instead of BMP4. N2B27 medium without vitamin A was made in house.

Hypoblast induction from mouse ES cells Two previously reported protocols were used for hypoblast induction from mouse ES cells30,34. Mouse ES cells were maintained under 2iL conditions. In the first protocol, 5 × 104per cm2 mouse ES cells were seeded onto gelatine under RPMI 1640 (12633012, Thermo Fisher Scientific) with 2 mM l-glutamine, B27 minus insulin (A1895601, Gibco), 20 ng ml−1 activin A, 3 µM CHIR and 10 ng ml−1 hLIF34. In the second protocol, 5 × 104 per cm2 mouse ES cells were seeded onto gelatine under 10 nM retinoic acid and 20 ng ml−1 activin A30. The medium was changed every day in both conditions. Marmoset embryo cultures All animal experiments were approved by the Animal Experiment Committee at CiRA and Kyoto University (Approval number 16-75-6) and the Institutional Animal Care and Use Committee of the Central Institute for Experimental Animals (CIEA: 17029A and 18031A). Naturally fertilized embryos were collected from the uterus by non-invasive flushing60. Embryos (morulae or blastocyst) were cultured under Sequential Blast (Origio, 83050010). When embryos reached the blastocyst stage, the zona pellucidae were removed using acidic Tyrode’s solution (Sigma-Aldrich), and the embryos were processed for immunosurgery using a custom rabbit polyclonal anti-marmoset antibody. ICM were seeded on laminin511-E8 under N2B27 plus 7F, 4F (FGF4, BMP4, A83, XAV) or control (PD03, LDN, A83, XAV) for 3 days, fixed and analysed using anti-SOX17 antibodies. Generation of bilaminoids Ten naive hPSCs (Naive(WT)) and 40 naive hPSCs or GFP-expressing naive hPSCs expressing GATA6 under DOX treatment (Naive(G6-OE) or naive-GFP(G6-OE)) were seeded in each well of a microwell array36 or Elplasia plate (4441, Corning) under t2iLGo plus 10 μM Y27632 without Matrigel or Geltrex. After 24–36 h of aggregation (day 0), the medium was switched to N2B27 with 0.1 μM DOX. On day 2, DOX was withdrawn. Bilaminoids were cultured under N2B27 until day 10. To identify the signalling pathways involved, 10 ng ml−1 BMP4, 300 nM LDN193189 (LDN, SML0559, Sigma-Aldrich), 3 µM A83-01, 10 ng ml−1 activin, 1 µM XAV and 1 µM CHIR were added from day 4 to day 6. In the experiments noted in the text, 10 ng ml−1 IL-6, 1 µM JAK inhibitor 1 ( JAKi, 420099, Sigma-Aldrich) or 10 ng ml−1 PDGF-AA was added from day 0 to day 4. The medium was changed every day. To collect PGCLCs, bilaminoids were cultured under N2B27 + 200 ng ml−1 BMP4 from day 5 to day 9.

Generation of LAMB1-KO lines To KO the LAMB1 gene, two gRNAs targeting exon 3 (gRNA 1)61 and exon 6 (gRNA 2) of human LAMB1 were designed and inserted into pSpCas9(BB)-2A-mCherry (Extended Data Fig. 9c): gRNA 1, 5′-GTCCTGG GCTCAAGTCGAT-3′; and gRNA 2, 5′-ATCTTGCTAGCAGGCTGAAA-3′. pSpCas9/gRNA plasmid (5 μg) was electroporated into primed H9 human ES cells (Neon Program 14). Then, 2 days later, mCherry+ cells were sorted by flow cytometry and seeded at a low density. About 10 colonies were picked 7–8 days after seeding, and genomic DNA was extracted. DNA was amplified and sequenced using the following primers: gRNA 1, Fw 5′-CCCCCGCTTGTTCGTTTTTTTCGG-3′, Rv 5′-TCACCTGCA AGTGGCTGACGATACAG-3′; and gRNA 2, Fw 5′-TCCGTGTCCTTC TCCTTTCG-3′, Rv 5′-CAGGAAATGTGTGGCGGATG-3′. The generated LAMB1-KO primed hPSCs were reset to naive hPSCs. Generation of CER1-knockin lines CER1-H2B-GFP reporter cells were generated from primed H9 human ES cells by replacing the endogenous stop codon of the CER1 gene with a T2A-H2B-GFP-LoxP-SV40-NeoR-LoxP cassette using CRISPR– Cas9 homology-directed repair (Extended Data Fig. 8i). H2B–GFP accumulates in the nucleus. gRNA targeting the stop codon of human CER1 was designed and inserted into pX330-U6-Chimaeric_ BB-CBh-hSpCas9: gRNA, 5′-TCCCAGGATTCCTTTATCCCAGG-3′. For the donor vector, approximately 1,000 bp upstream and downstream of the CRISP–Cas9 cleavage site was prepared by long PCR, fused with a T2A-H2B-GFP-LoxP-SV40-NeoR-LoxP cassette and cloned into a TOPO vector. pSpCas9/gRNA and the donor vector (1 μg each) were electroporated into primed H9 human ES cells (Neon Program 14). Then, 2 days later, G418 was added (200 µg ml−1) for about 2 weeks. The cells were collected and seeded on MEFs at a low density. Colonies were picked 7–8 days after seeding, and genomic DNA was extracted. DNA was amplified by PCR and sequenced. The SV40-NeoR gene was deleted from the CER1-H2B-GFP line by the transient introduction of a cre-expressing vector. The generated CER1-H2B-GFP-primed hPSCs were reset to naive hPSCs. The measure of the anterior–posterior axis of bilaminoids Angles between T+ nuclei and CER1–H2B–GFP, OTX2, LEFTY or DKK1 nuclei on sections of bilaminoids were analysed. The centre of the T+ nuclei was defined as 0°. Angles were averaged for each aggregate.

Co-culture with bilaminoid and nTB Bilaminoid and nTB were co-cultured using a cell culture insert (Transwell). nTB was induced from naive hPSCs on the Transwell. Bilaminoids were generated by culturing a mixture of 10 naive hPSCs (Naive(WT)) and 40 naive hPSCs or GFP-expressing naive hPSCs expressing GATA6 under DOX treatment (Naive(G6-OE) or Naive-GFP(G6-OE)) in each well of an Elplasia plate under t2iLGo plus 10 µM Y27632. After 24–36 h of aggregation (day 0), nTB on the Transwell was placed on the Elplasia plate under N2B27 with 0.1 µM DOX. On day 2, the DOX was withdrawn. Co-cultures continued until day 4.

Generation of IL6-KO lines To KO the IL6 gene, two sgRNAs that targeting exon 2 (sgRNA 1) and exon 3 (sgRNA 2) of human IL6 were designed and inserted into pSpCas9(BB)-2A-mCherry (Extended Data Fig. 7j): sgRNA 1, 5′-GAAGTCT TGCTTAACTGTTTG-3′; and gRNA 2, 5′-TAGACCTAAGTTACTCCATG-3′. pSpCas9/sgRNA plasmid (5 μg) was electroporated into primed H9 human ES cells (Neon Program 14). Then, 2 days later, mCherry+ cells were sorted by flow cytometry and seeded at a low density. Colonies were picked 7–8 days after seeding, and genomic DNA was extracted. DNA was amplified and sequenced using the following primers: sgRNA 1, Fw 5′-AGCCCACCGGGAACGAAAGAGAAGCT-3′, Rv 5′-GGCAGAACCAGAATTCGAGTGTGGGCTC-3′; and sgRNA 2, Fw 5′-G AACACAGGAGGGGAGATTGGGAGCCCA-3′, Rv 5′-GGGGATCCTTC TCTGATTGTCCCCCTTG-3′. The generated IL6-KO primed hPSCs were reset to naive hPSCs.

Aggregates generated by hPSCs and sorted cells A mixture of 100 naive or primed hPSCs and 100 sorted cells expressing GFP (naive 7F-, 4F-, G6-PDFRA+ cells, primed G6-PDGFRA+ cells, PDGFRA+ RACL cells, and CXCR4+CDH1+ definitive endoderm cells) were seeded in each well of an Elplasia plate under N2B27 plus 10 μM Y27632. The medium was changed every other day. Aggregates were evaluated on day 4.

Measurement of IL-6 Naive hPSCs were plated (1.5 × 105 cells per cm2) on iMatrix-coated Transwell plates and differentiated into nTB as described above (day 0). On day 3, the nTB induction medium was replaced with NDiff 227. As controls, hPSCs were plated (1.5 × 105 cells per cm2) on iMatrix-coated Transwell plates under each medium (naive hPSCs, t2iLGo; primed hPSCs, AK02N). On day 3, the hPSC medium was replaced with NDiff

227. The cell culture supernatants were collected on day 5 and centrifuged to remove debris. The levels of IL-6 were quantified using an IL-6 ELISA kit (Abcam, ab178013) according to the manufacturer’s protocol. The absorbance at 450 nm was measured using a plate reader (TECAN, Infinite 200 PRO). Each sample was analysed in duplicates.

Generation of naive hPSCs overexpressing OTX2, DKK1 and GP130/GCSFR For OTX2 overexpression, OTX2 fused to ERT2 was inserted into the PB vector (PB-OTX2-ERT2). The PB-OTX2-ERT2 vector and pBase helper plasmid were transfected into naive hPSCs expressing GATA6 under DOX treatment (Naive(G6-OE)). To generate bilaminoids, OTX2-ERT2 was activated by treatment with 100 nM 4-hydroxytamoxifen (tamoxifen) from day 4 to day 6. For DKK1 overexpression, DKK1 fused to destabilizing domain (DD) was cloned into the PB vector (PB-DD-DKK1). The PB-DD-DKK1 vector and pBase helper plasmid were transfected into naive hPSCs expressing GATA6 under DOX treatment (Naive(G6-OE)). To generate bilaminoids, DD-DKK1 was activated by treatment with 500 nM Shield1 (Takara, 632189) from day 4 to day 6. To activate JAK/STAT3 signalling, GP130/GCSFR chimeric receptor (Y118F) cDNA was inserted into the PB vector (PB-Y118F). The PB-Y118F vector and pBase helper plasmid were transfected into naive hPSCs (Naive(WT) or Naive(G6-OE)). To generate bilaminoids, STAT3 signalling was activated the treatment with G-CSF from day 0 to day 4. RACL induction from naive hPSCs Naive hPSCs (H9) were differentiated under RACL conditions as described previously23,62. The cells were plated (5 × 104 per cm2) onto MEF feeder cells and cultured under RACL medium, composed of RPMI 1640 medium with GlutaMAX (61870036, Thermo Fisher Scientific), B27 minus insulin (A1895601, Gibco), 100 ng ml−1 activin A, 3 μM CHIR, and 10 ng ml−1 LIF, for 7 days. The medium was changed every other day. On day 7, the cells were dissociated by Accutase, and PDGFRA+ cells were sorted. Anti-feeder antibody was used to remove the MEF feeder cells. Definitive endoderm induction Primed hPSCs were differentiated into definitive endoderm as described previously63. Primed hPSCs were seeded on an uncoated bacterial dish to form EBs under StemFit AK02N (AK02N, Ajinomoto) plus 10 μM Y27632. After 2 days, the EBs were washed and cultured under N2B27 with 200 ng ml−1 activin A and 3 µM CHIR. The next day, the medium was replaced with N2B27 and 200 ng ml−1 activin A and cultured for 2 more days. The EBs were dissociated using Accutase, and CXCR4+CDH1+ definitive endoderm cells were sorted and used for the experiments. TB induction from naive hPSCs Naive hPSC-derived TB-like cells (nTBs) were induced as described previously21,64. H9 naive hPSCs (5 × 104 cells per cm2) were plated onto laminin511-E8 (0.15 µg cm−2 iMatrix511 silk) under NDiff 227, 2 µM A83-01, 2 µM PD03, 10 ng ml−1 BMP4 and 10 µM Y27632. The next day, the medium was changed to NDiff 227, 2 µM A83-01, 2 µM PD03, and 1 µg ml−1 JAK inhibitor I ( JAKi, 420099, Sigma-Aldrich). On day 3, the cells were dissociated by Accutase, and HAVCR1+ENPEP+ (refs. 21,22,65,66) nTBs were sorted and recultured for further experiments. Transwell assay The Transwell assay was performed as previously described12 on Transwell 12-well plates with porous polyester membrane inserts (0.4 µm pore size; Corning). The membrane inserts were coated with 1% Geltrex diluted in DMEM/F12 for 1 h. For amnion-like cell induction, primed hPSCs were seeded on membrane inserts at a density of 3 × 104 cells per cm2 under mTeSR plus 10 μM Y27632. Then, 18 h after cell seeding, the medium was switched to E6 supplemented with bFGF (20 ng ml−1) and

BMP4 (50 ng ml−1) and cultured for 48 h. For G6-nHyCs and 7F-nHyCs, day 3 PDGFRA+ cells were sorted and recultured on membrane inserts at a density of 9 × 104 cells per cm2 overnight. Primed hPSCs were collected as small clumps and seeded onto the membrane inserts under E6 medium supplemented with bFGF (20 ng ml−1). The cells were cultured for another 48 or 96 h before analysis. The medium was changed every other day.

Flow cytometry and cell sorting Cells were dissociated into single cells by Accutase or trypsin, washed and blocked in HBSS (14185052, Thermo Fisher Scientific) with 1% BSA (A2153, Sigma-Aldrich) on ice for 30 min. Staining was performed on ice with the following: biotinylated PDGFRA antibodies (BAF322, R&D), CEACAM1 + CEACAM5 antibodies (Ab91213, Abcam) and directly conjugated antibodies in HBSS with 1% BSA for 30 min. After washing, Streptavidin-APC (405207, BioLegend) was used as the secondary antibody for PDGFRA–biotin. Alexa Fluor 488 was used for the CEACAM1 antibody. Flow cytometry and cell sorting were performed on the BD LSR Fortessa (BD) or FACS Aria II (BD) system. Data were analysed using FlowJo v.10.7.2. A list of the antibodies used is provided in Supplementary Table 7. qPCR with reverse transcription Total RNA was extracted using the RNeasy Kit (74106, Qiagen). Total RNA (0.5 µg) was reverse-transcribed into cDNA with an oligo-dT primer using SuperScriptIV (18090050, Thermo Fisher Scientific). qPCR was performed using QuantStudio3 (Thermo Fisher Scientific) and QuantStudio12K (Thermo Fisher Scientific) with TaqMan Fast Universal Master Mix (4364103, Thermo Fisher Scientific) and TaqMan probe or PowerUP SYBR Green Master Mix (A25743, Thermo Fisher Scientific) according to the manufacturer’s instructions. The results were analysed using QuantStudio Design & Analysis v.1.4.1 (Thermo Fisher Scientific). Immunostaining Cells were fixed in 4% paraformaldehyde (09154-85, Nacalai Tesque) for 10 min at room temperature. After fixation, the cells were washed with PBS, permeabilized in PBS plus 0.5% Triton X-100 for 1 h, and blocked in PBS plus 1% BSA and 0.05% Tween-20 (PBS-BT) for 2 h. Primary antibodies were diluted in PBS-BT and incubated at 4 °C overnight. After washing, secondary antibodies were diluted at 1:2,000 and incubated at room temperature for 2 h or at 4 °C overnight. Nuclei were stained with DAPI. Fluorescent images were obtained using the confocal laser scanning microscope TCS SP8 (Leica) or LSM710 (Zeiss). Cavity volume (Fig. 4e,h) was quantified from confocal z-stack images using Imaris software v.10.0.0 (Bitplane). PAR6 and F-actin images were used to quantify cavity volume with the Surfaces program. Western blot analysis Forwesternblotanalysis,1 × 106 cellswerelysedwithRIPAbuffer(08714-04, Nakalai Tesque). SDS sample buffer was added, and the mixture was incubated at 93 °C for 3 min. The extracted proteins were separated on Bollt 4–12%, Bis-Tris, 1.0 mm, Mini Protein Gel (NW04120BOX, Thermo Fisher Scientific) and blotted onto an Immobilon-P PVDF Membrane (IPVH00010, Merck) using a Mini PROTEAN Tetra Cell (Bio-Rad). The transferred membranes were incubated with the following primary antibodies: α-tubulin (ab7291, Abcam), pSMAD1/5/9 (9511, Cell Signaling Technology), pSMAD2 (3108, Cell Signaling Technology), pMAPK (4376, Cell Signaling Technology), pSTAT3 (9131, Cell Signaling Technology) and STAT3 (564533, BD Bioscience). The primary antibodies were detected with anti-rabbit IgG, HRP-linked antibodies (7074, Cell Signaling Technology) and anti-mouse IgG, HRP-linked antibodies (7076, Cell Signaling Technology), followed by detection using ECL Prime Western Blotting Detection Reagent (RPN2236, Amersham). Chemiluminescence images were acquired using the ImageQuant LAS 4000 (GE Healthcare) and Ambersham ImageQuant 800 (Cytiva)

systems. Uncropped western blot images are shown in Supplementary Figs. 2 and 3.

integrated analyses, clustering and visualization using the R Seurat package (v.4.0.4).

RNA-seq analysis For RNA-seq, samples were collected after removing MEFs by gelatine treatment. RNA was purified using the miRNeasy Mini Kit (217004, Qiagen), and 200 ng RNA and the TruSeq Stranded mRNA LT Sample Prep Kit (RS-122-2101, Illumina) were used for library construction. RNA-seq libraries were sequenced using the NextSeq 500 High Output v2 Kit (75 Cycles, FC-404-2005) (Illumina). The sequenced reads were trimmed to remove low-quality bases and adaptor sequences using cutadapt (v.1.15)67. The trimmed reads were mapped to the human reference genome (hg38) using TopHat268 with GENCODE v.2769. Uniquely mapped reads (MAPQ ≥ 20) were used for further analyses. Each gene expression level was calculated as reads per kilobase per million mapped reads (FPKM) using cufflinks (v.2.2.1)70. Genes expressed at low levels (defined as genes with FPKM