Nature - The International Journal of Science / 1 February 2024

Table of contents :
7 Research funders must join the fight for equal access to medicines
8 How can scientists make the most of the public’s trust in them?
9 Academia needs radical change — mothers are ready to pave the way
11 Research Highlights
13 Pioneering nuclear-fusion reactor shuts down- what scientists will learn
14 AlphaFold found thousands of possible psychedelics. Will its predictions help drug discovery?
16 Dana-Farber retractions- meet the blogger who spotted problems in dozens of cancer papers
17 Science’s fake-paper problem- high-profile effort will tackle paper mills
18 Japan’s successful Moon landing was the most precise ever
19 Long-COVID signatures identified in huge analysis of blood proteins
20 How does chronic stress harm the gut? New clues emerge
22 How cancer hijacks the nervous system to grow and spread
25 John L. Heilbron (1934–2023), historian of ‘big science’
26 Forget lung, breast or prostate cancer- why tumour naming needs to change
30 Cervical cancer kills 300,000 people a year — here’s how to speed up its elimination
33 Correspondence
35 Ecosystem effects of sea otters limit coastal erosion
36 Mobile atoms enable efficient computation with logical qubits
38 Flexible fibres take fabrics into the information age
39 Bacterial prey turns the tables on predator
40 Snapshots of genetic copy-and-paste machinery in action
42 Contact-tracing app predicts risk of SARS-CoV-2 transmission
45 Designing a circular carbon and plastics economy for a sustainable future
Re-imagining a circular plastics economy
A bold system change is needed
The carbon and plastics life cycle
Key terms of a circular carbon and plastics economy
Evaluating carbon emissions and other metrics
Delivering a bold system change
Sustainable plastics through smart design
Design principles for sustainable plastics
Reduce plastics demand
Switch to renewably sourced plastics
Maximize recycling
Minimize broader environmental impacts
A roadmap towards sustainable plastics
Acknowledgements
Fig. 1 The circular carbon and plastics life cycle.
Fig. 2 Plastic industry scenarios and GHG emissions for 2050 based on an estimated global production of 1.
Fig. 3 The impacts of smart design, service lifespan and recoverability in the carbon footprint of plastics.
Fig. 4 Interventions roadmap for a bold-system-change scenario.
58 Logical quantum processor based on reconfigurable atom arrays
Logical quantum processor based on reconfigurable atom arrays
Logical processor based on atom arrays
Improving entangling gates with code distance
Fault-tolerant logical algorithms
Complex logical circuits using 3D codes
Quantum simulations with logical qubits
Outlook
Online content
Fig. 1 A programmable logical processor based on reconfigurable atom arrays.
Fig. 2 Transversal entangling gates between two surface codes.
Fig. 3 Fault-tolerant logical algorithms.
Fig. 4 Zoned logical processor: scaling and mid-circuit feedforward.
Fig. 5 Complex logical circuits using 3D codes.
Fig. 6 Logical two-copy measurement.
Extended Data Fig. 1 Neutral-atom quantum computer architecture.
Extended Data Fig. 2 Single-qubit Raman addressing.
Extended Data Fig. 3 Mid-circuit readout and feedforward.
Extended Data Fig. 4 Further surface-code data.
Extended Data Fig. 5 Surface-code preparation and decoding data.
Extended Data Fig. 6 [[8,3,2]] and hypercube encoding.
Extended Data Fig. 7 Further [[8,3,2]] circuit sampling data.
Extended Data Fig. 8 Theoretical exploration of hypercube IQP circuits.
Extended Data Fig. 9 Further Bell-basis measurement results.
66 Observation of interband Berry phase in laser-driven crystals
Observation of interband Berry phase in laser-driven crystals
Interband Berry-phase interferometry
Resolving the Berry curvature
Online content
Fig. 1 Interband Berry phase resolved using HHG spectroscopy.
Fig. 2 Berry-phase interferometry.
Fig. 3 Resolving the Berry curvature.
Fig. 4 Circular dichroism HHG spectroscopy.
72High-quality semiconductor fibres via mechanical design
High-quality semiconductor fibres via mechanical design
Stress formation
Capillary instability
Optoelectronic fibres
Applications
Discussion
Online content
Fig. 1 Design and fabrication of semiconductor optoelectronic fibres.
Fig. 2 Stress analysis and capillary instability in the molten-core method.
Fig. 3 Optoelectronic fibres, fabrics and representative applications.
Extended Data Fig. 1 Cracks in the Ge core of Ge/silica fibres.
Extended Data Fig. 2 Comparison of theoretical and FE results on the stress distributions.
Extended Data Fig. 3 Raman spectra for Si/silica and Ge/silica fibres.
Extended Data Fig. 4 Capillary instability in the molten core method.
Extended Data Fig. 5 Glass-clad semiconductor fibres and the removal of glass cladding.
Extended Data Fig. 6 Optoelectronic fibres.
Extended Data Fig. 7 Evolution of the maximum principal stress in the solidified semiconductor core of.
79 Structural transition and migration of incoherent twin boundary in diamond
Structural transition and migration of incoherent twin boundary in diamond
Coexistent multiple ITB configurations
ITB transitions
Configuration-dependent ITB migration
Driving force for ITB activities
Online content
Fig. 1 Coexistent multiple configurations of {112} ITBs in nt-diamond.
Fig. 2 In situ observation of ITB transitions at atomic resolution.
Fig. 3 Configuration-dependent ITB migration.
Fig. 4 Evidence for the stress-driven mechanism of ITB activities.
Extended Data Fig. 1 Microstructural features of nt-diamond.
Extended Data Fig. 2 Structure differences among three symmetrical ITB configurations.
Extended Data Fig. 3 Dislocation characteristics of intrinsic SF, extrinsic SF, and ITB with configuration I.
Extended Data Fig. 4 Dislocations, nanocracks, and boundary disordering associated with ITB transitions.
Extended Data Fig. 5 ITB transition and associated dislocation behaviours observed at the accelerating voltage of 200 kV.
Extended Data Fig. 6 Structural features and relative changes of six ITB configurations, viewed along the (upper) and [111] (lower) zone axes of the left twin domain.
Extended Data Fig. 7 RBD across asymmetric ITBs in diamond.
Extended Data Fig. 8 Mapping of irradiation-induced strains in an nt-diamond grain using PED.
Extended Data Fig. 9 Contour mappings of in-plane atomic strains during an ITB transition from asymmetric configuration V to symmetric configuration III.
Extended Data Fig. 10 Contour mappings of in-plane atomic strains during an ITB transition from mixed IV and VI segments to a hybrid state with multiple configurations f.
86 Durable CO2 conversion in the proton-exchange membrane system
Durable CO2 conversion in the proton-exchange membrane system
Online content
Fig. 1 Physical characterization.
Fig. 2 Electrochemical measurements.
Fig. 3 In situ characterization.
Fig. 4 Theoretical investigation.
92 Stereodivergent 1,3-difunctionalization of alkenes by charge relocation
Stereodivergent 1,3-difunctionalization of alkenes by charge relocation
Online content
Fig. 1 Examples of state-of-the-art remote functionalization reactions of alkenes.
Fig. 2 General reaction scheme, optimization and the scope of syn-selective 1,3-hydroxyacylation.
Fig. 3 Stereodivergence of the 1,3-difunctionalization of alkenes and extension to other product classes.
Fig. 4 Application and mechanistic investigation of the 1,3-difunctionalization of alkenes.
Extended Data Fig. 1 Additional products of syn-selective 1,3-hydroxyacylation.
98 Establishing reaction networks in the 16-electron sulfur reduction reaction
Establishing reaction networks in the 16-electron sulfur reduction reaction
Reaction network in SRR and CV results
In situ Raman study on SRR
The role of catalysis in the SRR network
Conclusion
Online content
Fig. 1 Polysulfide conversion reactions involved in the Li-S battery.
Fig. 2 Charge analysis and reaction network for the SRR.
Fig. 3 In situ Raman results during discharge with the N,S–HGF catalytic electrode.
Fig. 4 Comparison of different catalysts in SRR.
Fig. 5 Simulated site-specific output potential of Li2S4 → Li2S conversion.
105 Flexible silicon solar cells with high power-to-weight ratios
Flexible silicon solar cells with high power-to-weight ratios
Highly efficient flexible and thin SHJ solar cells
Epitaxy-preventing composite gradient passivation
Low-damage continuous-plasma CVD operation
Nanocrystalline sowing and contact vertical growth
Ce-doped indium oxide and laser transfer printing
Cell performance and certification
External quantum efficiency and stability test
Online content
Fig. 1 Schematic diagrams of the FT and SF SHJ solar cells.
Fig. 2 Passivation and nanocrystalline contacts.
Fig. 3 Parameter statistics and certification reports.
Fig. 4 Quantum efficiencies, loss elements and stabilities.
Extended Data Fig. 1 Cross-sectional HRTEM images.
Extended Data Fig. 2 Hydrogen content.
Extended Data Fig. 3 Self-restoring nanocrystalline sowing and vertical growth induction (NSVGI).
Extended Data Fig. 4 Self-restoring nanocrystalline sowing.
Extended Data Fig. 5 Contact-free laser transfer printing.
Extended Data Fig. 6 Endurance assessment.
Extended Data Fig. 7 Durability analysis.
Extended Data Fig. 8 Power recovery in light-induced degradation.
Extended Data Fig. 9 Visualization of the surface potential distribution during RF-PECVD.
Extended Data Table 1 Evaluation of the performance using various technologies.
111Top-predator recovery abates geomorphic decline of a coastal ecosystem
Top-predator recovery abates geomorphic decline of a coastal ecosystem
Long-term relationships
Predator-exclusion experiment
Pre- and post-expansion of sea otters
Spatial comparisons
Discussion
Online content
Fig. 1 Study system and long-term trends in sea otter abundance and creekbank erosion.
Fig. 2 Locations of experimental and observational studies and results from a predator-exclusion experiment.
Fig. 3 Results from an analysis of tidal creeks comparing pre- and post-expansion of sea otters, examining relationships between sea otters, salt marsh biomass and creekbank retreat.
Fig. 4 Relationships between sea otter abundance, shore crab consumption and creekbank erosion in Elkhorn Slough from 2013 to 2015.
Extended Data Fig. 1 Results from a three-year sea otter-exclusion experiment testing the effects of otters and shore crabs on salt marsh vegetation.
Extended Data Fig. 2 Changes in shore crab densities when compared to the first survey in May 2014.
Extended Data Fig. 3 Example of camera-trap data.
Extended Data Fig. 4 Results of a shore crab feeding experiment on pickleweed aboveground and belowground biomass.
Extended Data Fig. 5 Results of a survey examining the relationship between shore crabs and sea otters in tidal creeks, and leverage analysis between sea otter crab consumption and erosion.
Extended Data Fig. 6 Modelling erosion rates.
Extended Data Fig. 7 Illustration of the modelled reduction in the base rate of creek erosion as the number of otters increases.
Extended Data Fig. 8 Posterior parameter distributions from modelling relative creek width as a function of a base rate of widening and an adjustment given the abundance of sea otters.
Extended Data Fig. 9 Histograms of b and r posterior distributions from the second-stage model (while propagating uncertainty from the first model).
Extended Data Fig. 10 Model outputs describing creek changes in width relative to the starting year and erosion accounting for sea otters.
119A lethal mitonuclear incompatibility in complex I of natural hybrids
A lethal mitonuclear incompatibility in complex I of natural hybrids
Mapping mitonuclear incompatibilities
Interactions with X. birchmanni mitochondrial DNA
Lethal effects in early development
Physiology and complex fitness effects
Mitochondrial biology in heterozygotes
Mitonuclear substitutions in complex I
Rapid evolution of complex I proteins
Introgression of incompatibility genes
Discussion
Online content
Fig. 1 Admixture mapping pinpoints mitonuclear incompatibility in Xiphophorus.
Fig. 2 Effect of incompatibility on Xiphophorus hybrid embryos.
Fig. 3 Physiology and proteomics of viable heterozygotes.
Fig. 4 Predicted structures of Xiphophorus respiratory complex I and evolutionary rates of incompatible alleles.
Fig. 5 Phylogenetic analysis and ancestry mapping suggest that genes underlying the mitonuclear incompatibility have introgressed from X.
Extended Data Fig. 1 ABC inference of additional selection parameters.
Extended Data Fig. 2 Ancestry depletion in natural hybrid populations.
Extended Data Fig. 3 Chromosome 15 incompatibility.
Extended Data Fig. 4 Additional F2 embryo morphometrics by ndufs5 genotype.
Extended Data Fig. 5 Juvenile F2 heart morphology by ndufa13 genotype.
Extended Data Fig. 6 Interface between ndufa13, ndufs5, and nd6 in RaptorX model.
Extended Data Fig. 7 Complex I mtDNA gene trees.
Extended Data Table 1 Sensitivity of inter-residue distances to modelling approach.
Extended Data Table 2 CodeML results for Complex I genes.
Extended Data Table 3 SIFT analyses of Complex I genes.
128 Alternative splicing of latrophilin-3 controls synapse formation
Extensive alternative splicing of Lphn3
Lphn3 splicing controls Gαs coupling
Genetic manipulation of Lphn3 splicing
Synapse connection requires Gαs–LPHN3
Assembly of synaptic complexes by LPHN3
Synapse formation requires PBM of E31
Activity promotes E31 splicing
Summary
Online content
Fig. 1 Differentially expressed Lphn3 splice variants couple to different G proteins.
Fig. 2 CRISPR-mediated conversion of Lphn3 alternative splicing from E31 to E32 impairs neuronal network activity.
Fig. 3 Switching LPHN3 G-protein coupling from Gαs to Gα12/13 by deleting E31 suppresses synaptic connectivity of hippocampal neurons.
Fig. 4 The alternatively spliced LPHN3 E31 variant assembles phase-separated postsynaptic scaffold protein condensates.
Fig. 5 Neuronal activity promotes E31 inclusion and E32 exclusion in Lphn3 by alternative splicing, leading to increased expression of the synaptogenic LPHN3 E31 variant.
Extended Data Fig. 1 Alternative splicing of Lphn3 (Adgrl3) transcripts (a & b) and demonstration that a subset of the sites of alternative splicing of Lphn3 exhibits a high degree of cell type-specific expression as revealed by RNAseq analyses (c–e).
Extended Data Fig. 2 Alternative splicing of Lphn1 (Adgrl1), Lphn2 (Adgrl2), and regulation of Lphn1 and Lphn2 G-protein coupling by alternative splicing.
Extended Data Fig. 3 Diverse patterns of Lphn3 alternative splicing analysed by RT-PCR in different brain regions and at different times of postnatal development.
Extended Data Fig. 4 Detailed TRUPATH analyses of G-protein coupling mediated by six different Lphn3 splice variants.
Extended Data Fig. 5 Further data characterizing cAMP assays, RNAseq analyses and pseudorabies virus tracing experiments.
Extended Data Fig. 6 Characterization of purified proteins used for phase-transition experiments.
Extended Data Fig. 7 Independent replication of the Lphn3-E31 dependent recruitment of phase-separated scaffold protein condensates at a higher concentrations of scaffold proteins (38 µM GKAP, 30 µM Homer3, 19 µM PSD95, 32 µM Shank3).
Extended Data Fig. 8 Further characterization of phase-separated, Lphn3-E31 coated post-synaptic scaffold condensates and FRAP (fluorescence recovery after photobleaching) experiments.
Extended Data Fig. 9 Selective deletion of the PDZ-domain binding motif (PBM) in Exon31 of the Lphn3 gene decreases synapse numbers.
Extended Data Fig. 10 Additional analyses of the activity-dependent splicing of Lphn3 Exon31 and Exon32.
136 Cortical regulation of helping behaviour towards others in pain
Cortical regulation of helping behaviour towards others in pain
Helping responses to others in pain
Allolicking helps others cope with pain
Encoding of others’ pain versus stress
Encoding of self-pain versus others’ pain
Control of allolicking and allogrooming
Separable coding of helping and comforting
Discussion
Online content
Fig. 1 Mice exhibit targeted allolicking towards social partners in pain.
Fig. 2 Allolicking by observers reduces self-licking in demonstrators.
Fig. 3 Neural representations of others’ pain and stress in the ACC.
Fig. 4 The ACC bidirectionally regulates allolicking and allogrooming behaviours.
Fig. 5 Separable representation of allolicking and allogrooming in the ACC.
Extended Data Fig. 1 Behavioral responses of demonstrators and observers following melittin injection.
Extended Data Fig. 2 Behaviors of female observers towards female demonstrators in pain.
Extended Data Fig. 3 Observers’ behaviors towards demonstrators experiencing pain induced by formalin injection.
Extended Data Fig. 4 Observers display general allogrooming but not targeted allolicking towards demonstrators in a stress state induced by acute restraint.
Extended Data Fig. 5 Allolicking assists others in coping with pain.
Extended Data Fig. 6 Response of ACC neurons to different states of others across demonstrators.
Extended Data Fig. 7 Single-cell- and population-level representations of prosocial behaviors and different states of demonstrators.
Extended Data Fig. 8 Response of ACC neurons to others’ pain and stress states and during prosocial behaviors.
Extended Data Fig. 9 Behavioral effects of DREADD inhibition of ACC neurons.
Extended Data Fig. 10 Optogenetic activation of ACC neurons and control experiments.
145Digital measurement of SARS-CoV-2 transmission risk from 7 million contacts
Digital measurement of SARS-CoV-2 transmission risk from 7 million contacts
Discussion
Online content
Fig. 1 App risk score and duration of exposure correlate with probability of infection.
Fig. 2 The probability of transmission is affected by both duration and proximity as captured by risk score.
Fig. 3 Transmission probability per exposure window increases almost linearly with risk score.
Fig. 4 Short, intermediate and long exposures all contribute to SARS-CoV-2 transmissions in the population.
Extended Data Fig. 1 The app has more nuanced distance-duration rules than manual contact tracing.
Extended Data Fig. 2 The probability of transmission depends linearly on duration and cumulative risk for short exposures, then sublinearly.
Extended Data Fig. 3 The monotonic relationship between the risk score per window and the probability of transmission in that window is robust with respect to the inclusion of individual heterogeneities in the model.
Extended Data Fig. 4 The transmission probability per exposure window decreases for contacts located in conurbations and increases for low-risk exposures during the weekend.
Extended Data Fig. 5 Duration and cumulative risk are the best predictors of infection, only marginally improved by machine learning.
Extended Data Fig. 6 Illustration of optimal strategies to reduce social costs of contact tracing via amber/red alert notifications.
Extended Data Fig. 7 Illustration of optimal strategies to increase effectiveness of contact tracing via amber/red alert notifications.
Extended Data Table 1 Summary statistics for the NHS COVID-19 app exposure dataset.
Extended Data Table 2 Summary statistics for different types of contacts in our dataset.
151Affinity-optimizing enhancer variants disrupt development
Affinity-optimizing enhancer variants disrupt development
Redundant low-affinity ETS sites regulate the ZRS
Human polydactyly SNVs subtly increase affinity
Affinity-optimizing SNVs cause polydactyly
Predicting penetrance and severity
Affinity-optimizing SNVs prevalent across the ZRS
The enhanceosome contains affinity-optimizing SNVs
Other transcription factors and disease enhancers
Regulatory principles predict causal SNVs
Discussion
Online content
Fig. 1 An ETS-A site in the ZRS enhancer contains two human variants that are associated with polydactyly, both of which subtly increase ETS binding affinity.
Fig. 2 Synthetic changes to the ETS-A site that create a 0.
Fig. 3 All mice with the approximately 0.
Fig. 4 A greater increase in affinity at the ETS-A site causes more severe and penetrant polydactyly and long-bone defects.
Fig. 5 Affinity-optimizing SNVs drive GOF expression in the ZRS and IFNβ enhanceosome.
Fig. 6 Affinity-optimizing SNVs drive GOF expression in a wide variety of disease-associated enhancers.
Extended Data Fig. 1 Class I ETS family members have conserved DBDs.
Extended Data Fig. 2 PBM binding affinities correlate with the in vivo ETS-1 ChIP signal in various cell types.
Extended Data Fig. 3 Conservation of ZRS ETS sites between humans and mice.
Extended Data Fig. 4 EMSA shows the binding of human and mouse ETS-1 to the ETS-A site and ETS-A variants.
Extended Data Fig. 5 Ptch1 in situ hybridization in the hindlimb bud and forelimb bud of transgenic mice.
Extended Data Fig. 6 Syn 0.
Extended Data Fig. 7 Affinity-optimizing SNVs are significantly associated with GOF enhancer activity.
Extended Data Fig. 8 EMSA shows stronger binding of the human HOXA13 and HOXD13 DBDs to the Dutch 2 variant relative to the WT sequence.
Extended Data Fig. 9 Experimental details for MPRA performed with 11 disease-associated enhancers.
Extended Data Fig. 10 Affinity-optimizing eQTL variants are enriched in GOF target gene expression.
160 Autoreactive T cells target peripheral nerves in Guillain–Barré syndrome
Autoreactive T cells in patients with GBS
Cytotoxic TH1 signature of autoreactive T cells
Characterization of autoreactive T cell clones
TCRβ clonotypes in patients with GBS
Antigen recognition and HLA alleles
Autoreactivity in CSF and peripheral nerves
Discussion
Online content
Fig. 1 Ex vivo stimulation of memory CD4+ T cells from the blood of patients with GBS and healthy donors.
Fig. 2 scRNA-seq analysis of memory CD4+ T cells from patients with GBS.
Fig. 3 Characterization of autoreactive CD4+ T cell clones from patients with GBS.
Fig. 4 Clonotypic analysis of autoreactive T cells in patients with GBS.
Fig. 5 Identification of autoreactive CD4+ T cells in the CSF and peripheral nerves of patients with GBS.
Extended Data Fig. 1 Experimental approach for studying autoreactive T cells in patients with GBS.
Extended Data Fig. 2 Overview of the autoreactive memory CD4+ T cell response in patients with GBS, with respect to disease subtype and stage and previous SARS-CoV-2 infection.
Extended Data Fig. 3 Ex vivo stimulation of memory CD8+ T cells from the blood of patients with GBS and healthy donors.
Extended Data Fig. 4 Characterization of autoreactive T cell clones in patients with GBS.
Extended Data Fig. 5 CDR3β consensus motifs characterizing the GLIPH2 specificity clusters.
Extended Data Fig. 6 Association between HLA polymorphisms and public autoreactive TCR Vβ sequences in patients with GBS.
Extended Data Table 1 Patients included in this study.
Extended Data Table 2 List of experiments performed on samples from patients with AIDP.
169 Motion of VAPB molecules reveals ER–mitochondria contact site subdomains
Discussion
Online content
Fig. 1 Altered ER tether motion at ER–mitochondria contact sites.
Fig. 2 Dynamic interactions generate a variable VAPB diffusion landscape within single ER–mitochondria contact sites.
Fig. 3 VAPB contact sites dynamically reorganize according to tether availability and metabolic needs.
Fig. 4 The ALS-linked VAPB P56S mutation displays aberrant motility at ER–mitochondria contact sites.
Extended Data Fig. 1 FIB-SEM reveals association of ER-mitochondrial contact sites and known contact site-associated biology.
Extended Data Fig. 2 Comparisons of VAPB interactions with mitochondria-associated and non-mitochondria associated structures.
Extended Data Fig. 3 Individual ERMCS show variability in capacity to bind passing VAPB molecules.
Extended Data Fig. 4 Latent states can be inferred from trajectory segments with similar mobility profiles.
Extended Data Fig. 5 VAPB-mediated ERMCS contact site mobility is visible in VAPB trajectory analysis.
Extended Data Fig. 6 Stationary contact sites show spatially stable regions of common molecular behaviour.
Extended Data Fig. 7 Changes in ER curvature within ERMCSs.
Extended Data Fig. 8 Variability of VAPB contact site size and shape in well fed and starved cells.
Extended Data Fig. 9 Additional characterization of ERMCS interactions of P56S VAPB molecules.
177 Discovery of a structural class of antibiotics with explainable deep learning
Models for antibiotic activity
Models for human cell cytotoxicity
Filtering and visualizing chemical space
Rationales predict antibiotic classes
Novel filtered substructures
A structural class of antibiotics from rationales
Mechanism of action and resistance
Toxicology, chemical properties and in vivo efficacy
Discussion
Online content
Fig. 1 Ensembles of deep learning models for predicting antibiotic activity and human cell cytotoxicity.
Fig. 2 Filtering and visualizing chemical space.
Fig. 3 Graph-based rationales reveal scaffolds for prospective antibiotic classes.
Fig. 4 Resistance and mechanism of action of a structural class.
Fig. 5 In vivo efficacy.
Extended Data Fig. 1 Molecular weight distribution of the 39,312 compounds screened.
Extended Data Fig. 2 Comparison of deep learning models for predicting antibiotic activity.
Extended Data Fig. 3 Comparison of deep learning models for predicting human cell cytotoxicity.
Extended Data Fig. 4 Visualizing chemical space across different prediction score thresholds.
Extended Data Fig. 5 Examples of rationale calculations using Monte-Carlo tree search.
Extended Data Fig. 6 Maximal common substructure identification reveals known antibiotic classes, but are less predictive than Chemprop rationales across all hits.
Extended Data Fig. 7 Closest active training set compounds to, and selectivities of, four validated hits associated with rationale groups G1-G5.
Extended Data Fig. 8 Comparison of MICs of different compounds against methicillin-susceptible and methicillin-resistant S.
Extended Data Fig. 9 Toxicity, chemical properties, and in vivo efficacy of compounds 1 and 2.
Extended Data Fig. 10 Exploration of a structural class through structure-activity relationships.
186 Template and target-site recognition by human LINE-1 in retrotransposition
Reconstitution of L1 ORF2p-mediated TPRT
Structure of template-RNA-bound L1 ORF2p
Features of the catalytic core
Single-stranded RNA recognition
Novel roles for the C-terminal domain 
Target-site architecture for TPRT
Discussion
Adaptation for nucleic acid recognition
Implications for L1 and SINE lifecycles
Online content
Fig. 1 In vitro TPRT activity and cryo-EM structures of human L1 ORF2p RNPs.
Fig. 2 Recognition of the template RNA and its poly(A) tract.
Fig. 3 Engagement and unwinding of the template RNA by L1 ORF2p CTS.
Fig. 4 Target-site position and upstream single-stranded DNA determine the efficiency of nicking and TPRT.
Extended Data Fig. 1 Purification, electron microscopy and reverse transcriptase activity of human L1 ORF2p and mutants.
Extended Data Fig. 2 Cryo-EM of L1 ORF2p RNP with Alu RNA.
Extended Data Fig. 3 Cryo-EM data processing for L1 ORF2p RNP complex bound to synthetic template RNA.
Extended Data Fig. 4 Resolution estimation.
Extended Data Fig. 5 Active site conformation and supporting data for investigation of the poly(A) tract and stem-loop engagement.
Extended Data Fig. 6 Structure and sequence-based bioinformatics analysis on L1 ORF2p CTS domain.
Extended Data Fig. 7 Analysis of off-target cleavage by L1 ORF2p.
Extended Data Fig. 8 Analysis of target cleavage by ΔCTS L1 ORF2p.
Extended Data Fig. 9 Comparison between the L1 ORF2p RNP and related structures.
Extended Data Fig. 10 Proposed configuration of PCNA interaction with L1 ORF2p.
Extended Data Table 1 Cryo-EM data collection, refinement and validation statistics.
194 Structures, functions and adaptations of the human LINE-1 ORF2 protein
Purification of highly active ORF2p RT
A 2.1 Å crystal structure of the ORF2p core
Five ORF2p core domains all bind nucleic acid
Structure of the L1 wrist domain
ORF2p cryo-electron microscopy structures in three states
Structure of the L1 tower domain
ORF2p RT and polymerase activities
Requirements for ORF2p priming
ORF2p synthesizes cDNA in the cytosol
Synthesized cDNAs activate cGAS/STING
In vitro inhibition of ORF2p
Structural basis of inhibition of ORF2p
Structure of full-length ORF2p
Domain comparison of ORF2p and other RTs
Structural adaptations of ORF2p RT
Structural insight into L1 evolution
Discussion
Online content
Fig. 1 Pathogenic replication cycle of L1 and the 2.
Fig. 2 Cryo-EM structures of ORF2p core in apo, ssRNA and RNA:DNA hybrid-bound states.
Fig. 3 L1 biochemical activities, priming and cytoplasmic reverse transcription of L1.
Fig. 4 Inhibition and structure of full-length ORF2p.
Fig. 5 Structural evolutionary analysis of ORF2p.
Fig. 6 Revised L1 insertion model.
Extended Data Fig. 1 Purification and crystal structure of ORF2p core.
Extended Data Fig. 2 Comparison of cryo-EM maps and models.
Extended Data Fig. 3 Design and characterization of the ORF2p tower domain deletions reveal it is not required for RT.
Extended Data Fig. 4 Priming requirements and mismatch tolerance of ORF2p core.
Extended Data Fig. 5 Comparative enzymology of ORF2p RT with HIV-1 and HERV-K.
Extended Data Fig. 6 Cytoplasmic RT activity of ORF2p and activation of interferon.
Extended Data Fig. 7 Inhibition of ORF2p core by NRTI and NNRTI reverse transcriptase inhibitors.
Extended Data Fig. 8 Comparison of ORF2p with other RTs.
Extended Data Fig. 9 ORF2p and R2Bm structures show opposing topologies of target DNA relative to the active site.
Extended Data Table 1 Data collection and refinement statistics (molecular replacement).
Extended Data Table 2 Cryo-EM Data collection, refinement, and validation statistics.
207 Targeted design of synthetic enhancers for selected tissues in the Drosophila embryo
Online content
Fig. 1 Deep learning-based design of tissue-specific synthetic enhancers.
Fig. 2 Validation of synthetic enhancers in vivo.
Extended Data Fig. 1 Learning the cis-regulatory code of Drosophila embryo tissues with deep learning.
Extended Data Fig. 2 TF motifs predictive of DNA accessibility discovered by TF-Modisco.
Extended Data Fig. 3 Comparison of sequence-to-accessibility and sequence-to-activity models plus controls.
Extended Data Fig. 4 Metric evaluation of the different models.
Extended Data Fig. 5 Predictive value of DNA accessibility and enhancer-activity models for predicted accessible sequences.
Extended Data Fig. 6 Model evaluation on positive and negative control sequences.
Extended Data Fig. 7 Nucleotide contribution scores of synthetic enhancers.
Extended Data Fig. 8 Nucleotide contribution scores of synthetic enhancers.
Extended Data Fig. 9 All synthetic sequences experimentally tested as enhancers.
Extended Data Fig. 10 Predicted scores for synthetic sequences and quantitative validations.
212 Cell-type-directed design of synthetic enhancers
In silico evolutions
Multiple cell-type codes
Motif implantation
Human enhancer design
Discussion
Online content
Fig. 1 Deep learning-based enhancer design.
Fig. 2 In silico sequence evolution towards functional enhancers.
Fig. 3 Spatial expansion and restriction of enhancer activity.
Fig. 4 Motif implantation towards minimal enhancer design.
Fig. 5 Human enhancer design.
Extended Data Fig. 1 In silico sequence evolution from random sequences.
Extended Data Fig. 2 State space optimization, design of perineurial glia enhancers and modification of genomic sequences toward KC enhancers.
Extended Data Fig. 3 Enhancer design toward multiple cell type codes.
Extended Data Fig. 4 Enhancer design by motif implanting.
Extended Data Fig. 5 Human enhancer design by in silico evolution.
Extended Data Fig. 6 Intermediate steps of in silico evolution and generation of repressor sites in human generated enhancers.
Extended Data Fig. 7 Human enhancer design by in silico evolution.
Extended Data Fig. 8 ZEB2 repression of in silico evolved MEL enhancers.
Extended Data Fig. 9 Human enhancer rescue.
Extended Data Fig. 10 Human enhancer design by motif implantation.
221 How co-working labs reduce costs and accelerate progress for biotech start-ups
226 Spreading paws-itivity
e1 Author Correction- A genomic mutational constraint map using variation in 76,156 human genomes
e2 Author Correction- A dense ring of the trans-Neptunian object Quaoar outside its Roche limit
e3 Publisher Correction- Population genomics of post-glacial western Eurasia

Citation preview

The IrnemactonalJciijrnal of science / 1February 2024

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

nature

The international journal of science / 1 February 2024

Research funders must join the fight for equal access to medicines

part in the evolution of our understanding of SARS-CoV-2 and COVID-19 by openly sharing research findings. Shared data on genome sequences and protein structures was necessary to create the vaccines and drugs that eventually controlled the pandemic. LMICs are asking for the same spirit from research-funding agencies and companies that researchers work with.

Groundhog Day

Pandemic treaty is a rare opportunity to ensure pandemic-related technologies are accessible and affordable to all.

F

or almost a year, nations have been negotiating the terms of an international agreement to better prepare the world for future pandemics. The talks are due to conclude this year, but countries are poles apart on key issues. In a statement last week, the World Health Organization (WHO)’s director-general Tedros Adhanom Ghebreyesus acknowledged that the talks are in trouble, meaning that the deadline might not be met. The ideal outcome would be for high- and low-income countries to have the same access to life-saving vaccines, drugs and other tools to combat a global health emergency, at a fair and transparent price. Although memories of the COVID-19 pandemic are fading, many people in low- and middle-income countries (LMICs) will never forget that people died because they had to wait for scarce vaccines, while leaders of high-income nations paid large sums to ensure more than adequate supplies. LMIC negotiators have an idea for how to stop this from happening in the future. The research community should consider backing it. The best way to extinguish competitive behaviour in vaccine and drug procurement during a pandemic is to prevent such behaviour happening in the first place. During the COVID-19 pandemic, countries agreed to work with the WHO and with pharmaceutical companies to distribute drugs, vaccines, tools and technologies equitably through COVAX, a global vaccine-sharing scheme. But this scheme failed, because wealthy countries did not honour their pledges. As part of the treaty discussions, LMICs are asking for public funders of scientific research to require that any pandemic-related drugs, vaccines or life-saving technologies that result from those organizations’ grants be shared equitably during a global health emergency. Funders should agree to this. It would be a one-time move, with the potential to save many lives. Funders could, for example, require grantees to openly share study results. They could also require that products arising from those studies be priced affordably. Moreover, funders could retain certain intellectual property (IP) rights to be used only when there’s a necessity to develop and distribute products equitably. Researchers played, and continue to play, an important

It would be a one-time move, with the potential to save many lives.”

However, the latest version of the treaty text does not include such provisions. Some European countries say that the World Trade Organization (WTO), not the WHO, is the organization to host discussions relating to IP rights. However, this disregards how, during the pandemic, WTO member states failed to temporarily waive IP rights for COVID-19 vaccines and therapies, despite a focused campaign led by India and South Africa, which Nature supported. Other high-income countries say that it could be complicated to include such conditions in research-funding contracts. Some funders might view these stipulations as burdensome on researchers. Moreover, in the United States at least, such a provision will almost certainly struggle to win the necessary approval from elected lawmakers. The preferred approach of the United States and many European countries is to negotiate agreements without passing laws. But we know the limitations of the voluntary approach. The US government tried and failed to persuade the biotechnology company Moderna, based in Cambridge, Massachusetts, to license its COVID-19 vaccine to LMIC manufacturers, despite having given the company more than US$1 billion of public funding to support its vaccine research. Attaching conditions to public funding is, in itself, not new — and in this instance, it would be for pandemic emergencies only. One example is the Coalition for Epidemic Preparedness Innovations (CEPI), an international nonprofit organization based in Oslo that is a leading funder of vaccines against epidemic and pandemic threats. CEPI asks for commitments to data sharing and affordable pricing, among other things in its research-funding contracts. It did as much for the COVID-19 vaccines that it funded, including four that received WHO emergency-use listing. One of these was the Moderna vaccine, which CEPI supported with a modest grant of almost $1 million early in its development. But the company never returned to CEPI for support, instead turning to US government funding, which did not come with access conditions. That shows the limitations of such individual agreements, and why a global and legally binding approach is needed, Frederik Kristensen, CEPI’s deputy chief executive, told Nature.

Ticking clock Time is running out. A preliminary draft of the pandemic agreement, published in February 2023, proposed some conditions to be included in research-funding contracts, including on prices of products, data sharing and the transfer of technology during a pandemic. The latest draft, published in October, omits this and instead says that governments should “publish the terms of Nature | Vol 626 | 1 February 2024 | 7

Editorials

government-funded research and development agreements for pandemic-related products”. This move will at least make it possible to know which, if any, governments are including pandemic-related conditions in their research grants. The problem is that demanding that the terms of the contracts are made public, without specifying what these terms should be, is not enough. Suerie Moon, a global-health policy researcher at the Geneva Graduate Institute in Switzerland, rightly asks: “Do we want to take an approach that helps countries to structure their collaboration with each other? Or do we want to maintain the status quo, where countries are essentially competing with each other?” High-income countries might feel that they’re better off on their own, she says. “But for most countries in the world, there’s a huge advantage to collaborating and agreeing on the rules of that international collaboration.” An international treaty is a rare opportunity for countries, companies and researchers to commit to making pandemic-related technologies accessible and affordable to all. Funders should take this opportunity and play their part in making that happen.

Making the most of trust in scientists How can researchers capitalize on the public’s trust in them and help to address concerns about government interference in science?

P

eople around the world have high levels of trust in scientists, but are concerned about governments interfering in research. These are among the findings reported by the global communications giant Edelman in its Trust Barometer, an annual survey that, in its latest iteration, consulted more than 32,000 people across 28 countries, from Mexico to Japan (see go.nature.com/4bgsipa). The report, published in mid-January, shines a spotlight on public trust in science and innovation. It follows several tumultuous years dominated by the COVID-19 pandemic, impacts resulting from climate change, falling standards of living and increasing global instability — and comes as the world grapples with a new challenge from innovation, the explosive rise of artificial intelligence (AI). Scientists are among those most trusted by the survey’s respondents to tell the truth about innovations and new technologies, with 74% of respondents saying they trust scientists to tell the truth. A similar proportion said that they wanted the introduction of innovations to be led by scientists. By comparison, just 47% of respondents said that they trusted journalists and 45% trusted government leaders to tell the truth on innovations. However, the survey also hints at a growing challenge for 8 | Nature | Vol 626 | 1 February 2024【 】

Governments worldwide have long looked to science and innovation to boost economies.”

scientists and governments alike, with 53% of respondents saying that science in their country has become politicized, referring to interference in science by politicians. Globally, some 59% said that governments and research funders have too much influence on how science is done — with the proportion rising to 70% and 75% in India and China, respectively. And nearly 60% of all respondents think that their government lacks the competence to regulate emerging innovations. The findings suggest both an opportunity and a challenge for scientists. How can researchers leverage people’s trust in them to improve the likelihood of government policy and decisions being evidence-based, while helping to address the public’s concerns about government interference and the lack of confidence in regulatory processes? The report is certainly timely. Governments worldwide have long looked to science and innovation to boost economies, but the pandemic has added a sense of urgency. Approaches being tried include clustering universities in cities in the hope of yielding the next Amazon or Google; policies that encourage entrepreneurial ideas from faculty members and students; readily available finance for every stage of a business idea; and relatively light-touch regulation so products can quickly reach consumers. The latest such proposal came last week from the Tony Blair Institute for Global Change, an influential policyresearch think tank in London set up by the former UK prime minister. Its report on innovation in biosciences proposes a much bigger role for AI in medical science and clinical practice (see go.nature.com/3ugt3gh). To this end, the institute is urging the UK government to reform regulatory structures that govern how researchers and companies can access anonymized patient data. But if the Edelman report is correct, and people are concerned about governments interfering in science and having poor regulatory competence, then ways must be found to turn that around. In this context, the social sciences present an invaluable and underused tool. In January, a report by the UK Academy of Social Sciences rightly reminded governments of the need to embed social science in their science, technology, engineering and mathematics policymaking as one way to enhance public trust (see go.nature.com/4bioq0i). Data scientists, economists, ethicists, legal scholars and sociologists are among the social scientists who are skilled at studying the strengths and limitations of new technologies, as well as different economic and regulatory models — and communicating their findings, along with all the attendant uncertainties. If people think that science has become politicized and that governments are interfering too much in research, that is a problem not only for science, but also for society, because it could affect public confidence in governments’ ability to deliver the benefits of science and innovation, while simultaneously protecting people from harm. Scientists should make the most of the public’s trust in them as a source of information on innovation. And they should work with governments to dissuade them from overly politicizing science. Governments have an equal part to play in this — and Nature hopes they are listening.

A personal take on science and society

World view

By Fernanda Staniscuaski

Mothers are a force for radical change in academia The research system must lose its overly rigid attitude toward career progression — and mothers are uniting to make that happen.

GUSTAVO DIEHL/UFRGS

I

n December, an academic incident made headlines in Brazil. An ad hoc reviewer for the National Council for Scientific and Technological Development (CNPq), one of the country’s main federal funding agencies, told social scientist Maria Carlotto that her pregnancies had held back her career. Carlotto tweeted about the comment, sparking general outrage and broader discussion of the rigid mindset concerning career progression in academia. Such outdated views have serious consequences: for more than 20 years, women have received only 35% of CNPq’s prestigious research productivity scholarships, and Black and Indigenous women are completely absent from the top ranks of CNPq fellowships. Worldwide, mothers are too often pushed out of academia. But we are stepping up to push for a revolution in academia. Individuals and groups affected by the hostile academic environment should unite. By organizing to seek change, mothers put ourselves at the forefront of work to reshape academic culture. Our efforts extend beyond personal struggles — we are architects of a transformative movement. The struggle for change spearheaded by mothers yields benefits for the entire academic community. Carlotto’s experience illuminates the hostility of the academic system towards those who do not follow a direct career path. This harms many in academia, not just mothers: people with caring responsibilities; those with disabilities or chronic illnesses; and the list goes on. Motherhood can also intersect with factors such as ethnicity and race to amplify disparities. Fixation on a narrow idea of success often leads to burnout, mental-health issues and the abandonment of promising careers. I know this from personal experience. I am a mother of three as well as a molecular biologist at the Institute of Biosciences and the Centre for Biotechnology at the Federal University of Rio Grande do Sul in Porto Alegre, Brazil. Before motherhood, my career was making a steady ascent, marked by publications, grants and mentorship roles in graduate programmes. Everything changed after my first child was born in 2013. In 2014, I had my last grant approved in molecular biology. After that, all my applications were denied, with the comment that my productivity did not match that of my peers. But this was never a fair comparison: it included no recognition that I had taken breaks for parental leave. Pauses in my career were seen as impediments, conflicting with the myth of the ideal academic. Becoming a mother did not render me incapable of being

The struggle for change spearheaded by mothers yields benefits for the entire academic community.”

Fernanda Staniscuaski is a molecular biologist at the Federal University of Rio Grande do Sul in Porto Alegre, Brazil. She is the founder and coordinator of the Parent in Science Movement. e-mail: fernanda. [email protected]

a scientist, but the system nearly made me give up. Feeling unsupported, I contemplated leaving science. Against all odds, I persisted and eventually founded the Parent in Science Movement. The organization has established partnerships with many others worldwide, including the international non-profit group Mothers in Science. In January 2023, a global movement of mothers in science launched a global call to action, requesting that funders take measures to make science truly fair for us. These efforts have led to concrete changes in Brazil. In 2021, a new field was added to Lattes, a database of all Brazilian scientists’ CVs, recognizing that career breaks of all kinds — not just maternity leave — are part of the academic journey. In 2023, the Rio de Janeiro state funding agency FAPERJ introduced a career-restart grant for mothers in collaboration with the Parent in Science Movement and the Serrapilheira Institute, a private research-funding institute in Rio de Janeiro. This initiative — pioneering in Brazil and Latin America — will make its first call for grant applications this year. It aims to support at least 21 proposals from mothers who are advancing their careers after a hiatus. In response to the ad hoc reviewer’s comments about Carlotto’s pregnancies, the Parent in Science Movement wrote to the CNPq expressing concern and outlining a plan for a fair and inclusive academic space. We recommended actions such as setting concrete goals for increasing the representation of historically under-represented groups; creating a CNPq equity, diversity and inclusion committee; ensuring that all CNPq advisory committees have a diverse gender, racial and regional composition; and revising evaluation criteria for funding and scholarships. Under pressure from the academic community, the CNPq adopted maternity criteria in all committees. This means that when evaluating the scientific productivity of scholarship applicants, the evaluation period is extended by two years for each childbirth or adoption — a significant step. The agency also pledged to investigate further reports of prejudiced opinions being aired during the 2023 call for funding. These changes show how our advocacy can transform academia. We need a unified front, in which individuals and groups affected by the hostile academic environment — which is all of us — work towards a common goal. All university administrators and academics, particularly those in privileged positions, must engage in open dialogues, challenge existing norms, make concrete changes and recognize that people have lives and obligations outside work. Funders and institutions must adapt their structures to accommodate the multifaceted lives of individuals. A flexible and supportive environment is the only way academia can attract the scientists it needs. Nature | Vol 626 | 1 February 2024【 】 | 9

Selections from the scientific literature

L TO R: E. DODD, COURTESY OF THE MINISTERO DELLA CULTURA – PARCO ARCHEOLOGICO DI POMPEI; STEVE GSCHMEISSNER/SPL; THOMAS O’NEILL/NURPHOTO/GETTY

Research highlights ‘QUMODES’ SHINE IN AN ALTERNATIVE QUANTUM COMPUTER

NATURAL KILLER CELLS HARNESSED TO FIGHT BLOOD CANCER

Quantum computers are built from qubits: physical objects with properties that have one of two possible results when measured. The prototypical example is an electron, whose spin will always show as either clockwise or anticlockwise. Now physicists have demonstrated an alternative foundation for quantum computers called qumodes, which were first proposed in 2001. Qumodes’ measured properties — in this case, the brightness of a light pulse — can vary along a continuum instead of consisting of just two discrete possibilities. To create qumodes, Shunya Konno at the University of Tokyo and his collaborators carefully modified laser pulses by removing one photon at a time and creating interference between pairs of pulses. They then demonstrated that the resulting pulses had the properties that would be required, both to perform ‘digital’ quantum computations and to correct errors in those computations. This means that, with further work, a few photonic qumodes could perform the same kinds of quantum algorithm that require hundreds or even thousands of qubits in ordinary quantum computers. Theoretical studies have suggested that qumode-based photonic quantum computers could be faster than qubitbased machines, and easier to build at large scales without becoming impractically prone to computational errors.

Infusions of bioengineered natural killer cells reduced the number of cancerous cells in people with blood cancer. Immune cells can be genetically engineered to make proteins called chimeric antigen receptors (CARs), which help the cells to locate and destroy cancer cells (pictured, blue). Often, T cells are used, but one study found that CAR-equipped natural killer cells eradicated detectable cancer in 7 out of 11 people with lymphoma or chronic lymphocytic leukaemia. David Marin at the University of Texas MD Anderson Cancer Center in Houston and his colleagues did a larger study. They extracted natural killer cells (pictured, pink) from donated umbilical cord blood and used those cells to make CAR natural killer cells. Tests 100 days after treatment found that infusions of the engineered cells had reduced the number of cancer cells in roughly half of the 37 study participants. One year in, the number of cancer cells had not increased in about half of the 25 surviving participants. Cord blood samples frozen soon after collection and with lower numbers of nucleated red blood cells were linked to the best outcomes.

A GOOD MIXER: HOW ROMAN WINES GOT THEIR NUTTY AROMA

Wine was the lifeblood of Roman social life. By comparing ancient and modern winemaking jars, archaeologists have revealed how Roman wine looked, smelled and tasted. Ancient Romans fermented and aged wine in dolia — large, egg-shaped clay containers that were typically buried up to their necks in the ground (pictured). To investigate how these vessels influenced their boozy contents, Dimitri Van Limbergen at Ghent University in Belgium and Paulina Komar at the University of Warsaw compared Roman dolia with similar winemaking jars used in modern Georgia. The comparison suggests that the shape of the dolia, like that of the Georgian jars, helped to drive heat-driven currents in the crushed grapes — typically white varieties — inside the vessel. These currents promoted uniform fermentation by mixing the jars’ contents. The grape solids that remained at the bottom of the jar after fermentation gave the wine an amber colour. Burial of the jars promoted the formation of compounds that give wine aromas of toasted bread, apples and walnuts. The conditions the clay provided also created pleasant flavours fruit, the Science 383, 289–293 (2024【 】 authors say. Antiquity https://doi.org/ gtfrqp (2024 【 】)

Nature Med. https://doi.org/ mdqq (2024【淘宝唯一店铺：艾 米学社】)

TALL STORY: THE MOUNTAINS THAT LOST THEIR ROOTS Geologists are starting to answer a long-standing question about a puzzling mountain range in northern Colombia: how did it get so high? Typically, tall mountains sit on top of thick ‘roots’ of crustal rocks, which support their great weight. But gravitational studies of the Sierra Nevada de Santa Marta mountain range (pictured), which soars more than 5 kilometres high, suggest that it doesn’t have a root. To understand why, a team led by David Quiroga, then at the University of Alberta in Edmonton, Canada, simulated how the crust beneath the mountains might have evolved. The team found that the range once had a thick crustal root, but that it slowly disappeared over time. That’s because the rocks in the root were colder and denser than neighbouring rocks. Those dense rocks sank lower into Earth. That removed much of the crust that lay beneath the mountains and left them sitting high — even without a crustal root to support them. This could have happened as long ago as 40 million to 56 million years, or as recently as within the last 2 million years. J. Geophys. Res. Solid Earth 129, e2023JB027646 (2024【 】【淘 宝唯一店 铺：艾米学社】)

Nature | Vol 626 | 1 February 2024 | 11

The world this week

CEA-IRFM/EUROFUSION/ZUMA PRESS WIRE/SHUTTERSTOCK

News in focus

The metal tiles lining the inside cavity of the Joint European Torus are irradiated with tritium, a radioactive isotope of hydrogen.

PIONEERING NUCLEAR-FUSION REACTOR SHUTS DOWN: WHAT SCIENTISTS WILL LEARN The decommissioning of the Joint European Torus near Oxford, UK — a test bed for ITER — will take until 2040 and be studied in detail. By Elizabeth Gibney

S

cientists have begun to decommission one of the world’s foremost nuclear-fusion reactors, 40 years after it began operations. Researchers will study the 17-year process of dismantling the Joint European Torus ( JET) near Oxford, UK, in unprecedented detail — and use the knowledge to make sure future fusion power plants are safe and financially viable. “We are starting to think seriously about a fusion power plant,” says Rob Buckingham, who leads the decommissioning of the UK Atomic Energy Authority, which oversees JET.

“This means thinking about the whole plant life cycle.” Harnessing the fusion of atoms — the process that powers the Sun — could provide a near-limitless source of clean energy. Creating the conditions for fusion in power plants will require complex engineering that hasn’t yet been proved, meaning that commercial fusion power is still many decades away. But researchers are pushing ahead with designs for the first commercial reactors as excitement about fusion grows. In 2022, JET smashed the record for the amount of energy created through fusion. And the US National Ignition Facility (NIF) in Livermore,

California — the flagship US fusion facility — now routinely generates more energy from a fusion reaction than was put in. The NIF calculations do not include the entire energy requirements of running the facility, which fusion plants would need to exceed to truly ‘break even’, but physicists have celebrated the milestones.

Fuel remnants JET is important because the facility is a test bed for ITER, a US$22-billion fusion reactor being built near Saint-Paul-lez-Durance, France, that aims to prove the feasibility of fusion as an energy source in the 2030s. JET has Nature | Vol 626 | 1 February 2024 | 13

News in focus Studies at JET this year will recover and analyse 60 wall tiles — the first of more than 4,000 components. “We can use this information to move from lab-scale research to industrial-scale processes, to detritiate the many tonnes of tiles and components which will be removed from JET over the next few years,” he says.

The tokamak pit of ITER, the world’s largest fusion experiment, being built in southern France.

informed decisions on what materials to build and fuel ITER with, and it has been crucial to predicting the bigger experiment’s behaviour. The thorniest part of decommissioning the JET site will be dealing with its radioactive components. The process of fusion does not leave waste that is radioactive for millennia, unlike nuclear fission, which powers today’s nuclear reactors. But JET used significant amounts of tritium, a radioactive isotope of hydrogen. Tritium, which will be used as a fuel in future plants including ITER, has a half-life of 12.3 years, and its radiation, alongside the high-energy particles released during fusion, can leave components radioactive for decades. Decommissioning a fusion experiment doesn’t have to mean “bulldozing everything within sight into rubble and not letting anyone near the site for ages”, says Anne White, a plasma physicist at the Massachusetts Institute of Technology in Cambridge. Instead, engineers will prioritize reusing and recycling parts. This will include removing tritium where possible, says Buckingham. This reduces radioactivity and allows the tritium to be reused. Ultimately, physicists will use the knowledge gained from JET’s decommissioning to improve how they embed recycling into the design of the Spherical Tokamak for Energy Production (STEP), a prototype commercial reactor being planned in Britain.

Plasma Physics Laboratory in New Jersey shut down. Many parts, such as the equipment for injecting hot beams of gas into the reactor, were reused. But the tokamak had to be filled with concrete, cut up and buried. JET scientists hope to leave little overall waste. The main challenge, says Buckingham, is to understand where the tritium is and to remove it from materials, including from metal tiles that line the inside of the tokamak. JET engineers will use a robotic system to remove sample tiles for analysis. And they will use remotely operated lasers to measure how much tritium is in samples that remain inside. Tritium is a gas that “penetrates all materials, and we need to know exactly how deep the tritium has penetrated”, says Buckingham.

Radioactive doughnut

By Ewen Callaway

JET and ITER are both ‘tokamak’ reactors, which confine gas in their doughnut-shaped cavities. JET uses magnets to squeeze a plasma of hydrogen isotopes, ten times hotter than the Sun, until the nuclei fuse. The last time the fusion community decommissioned a comparable device was in 1997, when the Tokamak Fusion Test Reactor at Princeton 14 | Nature | Vol 626 | 1 February 2024【 】

To extract the tritium from metals, engineers will heat the components in a furnace before capturing the released isotope in water. Tritium can be removed from water and turned back into fuel; leftover materials become low-level waste, the classification given to radio­active waste made by universities and hospitals. Variations on this process are being tested for other materials, including resins and plastics. JET researchers are exploring how to dispose of the low-level waste, as well as the much smaller amount of intermediate-level radioactive waste — in which nuclear decay occurs more frequently. Options include re-treating the waste, removing it to special disposal sites or storing it until it decays to lower levels of radioactivity. Some unaffected parts of JET, such as diagnostic and test equipment, have already been repurposed in fusion experiments in France, Italy and Canada. In its final experiments last December, JET went out with a bang. Scientists explored inverting the shape of the plasma in a way that might more readily confine heat. They also deliberately damaged the facility by sending a high-energy beam of ‘runaway’ electrons — produced when plasma is disrupted — careering into the reactor’s inner wall. “Analysis of the damage will provide useful data,” says Joelle Mailloux, who leads the scientific programme at JET.

WILL ALPHAFOLD’S PREDICTIONS HELP DRUG DISCOVERY? Researchers are learning how to deploy the protein-structure tool effectively.

S

cientists have used the protein-structure-prediction tool AlphaFold to identify1 hundreds of thousands of potential psychedelic molecules, which could help to develop new kinds of antidepressant. The research shows, for the first time, that AlphaFold predictions

— available at the touch of a button — can be just as useful for drug discovery as experimentally derived protein structures, which can take months, or even years, to determine. The development is a boost for AlphaFold, the artificial intelligence (AI) tool developed by DeepMind in London that has been a game changer in biology. The public AlphaFold database holds structure predictions for nearly every

XINHUA/SHUTTERSTOCK

Detritiation nation

known protein. Protein structures of molecules implicated in disease are used in the pharmaceutical industry to identify and improve promising medicines. But some scientists were starting to doubt whether AlphaFold’s predictions could stand in for gold-standard experimental models in the hunt for drugs. “AlphaFold is an absolute revolution. If we have a good structure, we should be able to use it for drug design,” says Jens Carlsson, a computational chemist at the University of Uppsala in Sweden.

DEEPMIND

AlphaFold scepticism Efforts to apply AlphaFold to finding drugs have been met with considerable scepticism, says Brian Shoichet, a pharmaceutical chemist at the University of California, San Francisco. “There is a lot of hype. Whenever anybody says ‘such and such is going to revolutionize drug discovery’, it warrants some scepticism.” Shoichet counts more than ten studies that have found AlphaFold’s predictions to be less useful than protein structures obtained with experimental methods, such as X-ray crystallography, when used to identify potential drugs with a method called protein–ligand docking. This approach — common in the early stages of drug discovery — involves modelling how hundreds of millions or billions of chemicals interact with key regions of a target protein, in the hope of identifying compounds that alter the protein’s activity. Previous studies have tended to find that when AlphaFold-predicted structures are used, the models are poor at singling out drugs already known to bind to a particular protein. Researchers led by Shoichet and Bryan Roth, a structural biologist at the University of North Carolina at Chapel Hill, came to a similar conclusion when they checked AlphaFold structures of two proteins against known drugs for the neuropsychiatric conditions in which the proteins are implicated. The researchers wondered whether small differences from experimental structures might cause the predicted structures to miss certain compounds that bind to proteins — but also make them able to identify different ones that were no less promising. To test this idea, the team used experimental structures of the two proteins to virtually screen hundreds of millions of potential drugs. One protein, a receptor that senses the neurotransmitter serotonin, was previously determined using cryo-electron microscopy. The structure of the other protein, called the σ2 receptor, had been mapped using X-ray crystallography. They ran the same screen with models of the proteins plucked from the AlphaFold database. They then synthesized hundreds of the most promising compounds identified with either the predicted or the experimental structures, and measured their activity.

The screens with predicted structures yielded completely different drug candidates from those with experimental structures. “There were no two molecules that were the same,” says Shoichet. “They didn’t even resemble each other.” But to the team’s surprise, the ‘hit rates’ — the proportion of flagged compounds that actually altered protein activity in a meaningful way — were nearly identical. And AlphaFold structures

“You could advance the project by a couple of years and that’s huge.” identified the drugs that activated the serotonin receptor most potently. The psychedelic drug LSD works partly through this route, and many researchers are looking for non-hallucinogenic compounds that do the same thing, as potential antidepressants. “It’s a genuinely new result,” says Shoichet.

Prediction power In unpublished work, Carlsson’s team has found that AlphaFold structures are good at identifying drugs for a sought-after class of target called G-protein-coupled receptors, for which their hit rate is around 60%. Having confidence in predicted protein structures could be game-changing for drug discovery, says Carlsson. Determining structures experimentally isn’t trivial, and many would-be targets might not yield to existing experimental tools. “It would be very convenient if we could push the button and get a structure we can use for ligand discovery,” he says. The two proteins that Shoichet and Roth’s team picked are good candidates for relying

A protein structure predicted by AlphaFold.

on AlphaFold, says Sriram Subramaniam, a structural biologist at the University of British Columbia in Vancouver, Canada. Experimental models of related proteins — including detailed maps of the regions where drugs bind to them — are readily available. “If you stack the deck, AlphaFold is a paradigm shift. It changes the way we do things,” he adds. “This is not a panacea,” says Karen Akinsanya, president of research and development for therapeutics at Schrödinger, a drug-software company based in New York City that is using AlphaFold. Predicted structures are helpful for some drug targets, but not others, and it’s not always clear which applies. In about 10% of cases, predictions that AlphaFold deems highly accurate are substantially different from the experimental structure, a study2 found. And even when predicted structures can help to identify leads, more detailed experimental models are often needed to optimize the properties of a particular drug candidate, Akinsanya adds.

Big bet Shoichet agrees that AlphaFold predictions are not universally useful. “There were a lot of models that we didn’t even try because we thought they were so bad,” he says. But he estimates that in about one-third of cases, an AlphaFold structure could jump-start a project. “Compared to actually going out and getting a new structure, you could advance the project by a couple of years and that’s huge,” he says. That is the goal of Isomorphic Labs, DeepMind’s drug-discovery spin-off in London. On 7 January, the company announced deals worth a minimum of US$82.5 million — and up to $2.9 billion if business targets are met — to hunt for drugs on behalf of pharmaceutical giants Novartis and Eli Lilly using machine-learning tools such as AlphaFold. The company says that the work will be aided by a new version of AlphaFold that can predict the structures of proteins when they are bound to drugs and other interacting molecules. DeepMind has not yet said when — or whether — the update will be made available to researchers, as earlier versions of AlphaFold have been. A competing tool called RoseTTAFold All-Atom3 will be made available soon by its developers. Such tools won’t fully replace experiments, scientists say, but their potential to help find new drugs shouldn’t be discounted. “There’s a lot of people that want AlphaFold to do everything, and a lot of structural biologists want to find reasons to say we are still needed,” says Carlsson. “Finding the right balance is difficult.” 1. Lyu, J. et al. Preprint at bioRxiv https://doi. org/10.1101/2023.12.20.572662 (2023). 2. Terwilliger, T. et al. Nature Methods 21, 110–116 (2024). 3. Krishna, R. et al. Preprint at bioRxiv https://doi. org/10.1101/2023.10.09.561603 (2023).

Nature | Vol 626 | 1 February 2024 | 15

News in focus

The prestigious Dana-Farber Cancer Institute (DFCI) in Boston, Massachusetts, acknowledged last week that it would seek retractions for 6 papers and corrections for an additional 31 — some co-authored by DFCI chief executive Laurie Glimcher and other prominent cancer researchers. The news came after scientific-image sleuth Sholto David posted his concerns about more than 50 manuscripts to a blog on 2 January. In the papers, published in a range of journals including Blood, Cell and Nature Immunology, David found images from western blots — a common test for detecting proteins in biological samples — in which bands seemed to be spliced, stretched and copied and pasted. (Nature’s news team is editorially independent of its publisher, Springer Nature, and of other Nature-branded journals.) It was not the first time that some of these irregularities had been noted; some were flagged years ago on PubPeer, a website where researchers comment on and critique scientific papers. The DFCI, an affiliate of Harvard University, had already been investigating some of the papers in question before David’s blogpost was published, says the centre’s researchintegrity officer, Barrett Rollins. “Correcting the scientific record is a common practice of institutions with strong researchintegrity processes,” he adds. (Rollins is a co-author of three of the papers that David flagged and is not involved in investigations into them, says DFCI spokesperson Ellen Berlin.) David, based in Pontypridd, UK, spoke to Nature about how he uncovered the data irregularities at the DFCI and what scientists can do to prevent mix-ups in their own work. You’ve said that you’re doing data sleuthing full-time. How did you get into it? I’m not doing anything else. I did my PhD in cellular molecular biology at Newcastle University, UK, and I finished that in 2019. And then I went to work for Oxford Biomedica [a UK gene- and cell-therapy company]. I was there for three years and then I moved out here to Wales. Since then, I’ve been doing this image stuff. I’m not doing this nine to five, but I am pretty busy with it. I used to write letters to the editor [at

16 | Nature | Vol 626 | 1 February 2024【 】

CRAIG F. WALKER/THE BOSTON GLOBE VIA GETTY

Dana-Farber retractions: the blogger who spotted problems in dozens of cancer papers

Dana-Farber Cancer Institute in Boston, Massachusetts, is seeking retractions of six papers. journals], but it’s a very infuriating process. So, it’s through getting frustrated that I discovered PubPeer. You recently left your 2,000th comment on PubPeer. What keeps you coming back? I enjoy the ridiculous back and forth with the authors over e-mail. I care a lot about the animals [that are killed for life-sciences experiments] as well. The level of expectation we should have when we’re dealing with animals and high-profile institutions is that they’re super careful and that they get things right, so it’s frustrating when you see errors. What’s your usual process when you’re combing through a paper? It’s going to depend on the problem that I might expect to find. A few months ago, [the open-access journal] PLoS ONE retracted nine papers, and these were all to do with gastricdamage stuff. In that case, I was looking for image reuse between papers. I went and I got all of this guy’s papers, and I cropped all of the images out of all the papers, put it in a giant folder and then resized them all. And I used a script to feed it into Imagetwin [software that compares images with a database of more than 25 million others]. But for the DFCI stuff, a lot of what I found and what had been previously posted [on PubPeer] is duplicated

images within the same paper. Imagetwin is really useful for these things. I note and collect the errors on PubPeer, write a blog, send the blog to the publisher and university. What I’m hoping for is that the authors respond on PubPeer. If I see a really credible, active response on PubPeer, then I’d probably just leave it there. Thinking about the DFCI papers, what pattern of potential image manipulation stood out to you the most? There is one where there’s images of mice, and it looks like one of them has been copied lots of times, and there’s a bioluminescent signal that’s been superimposed on top. It’s got the ears in the same place. It’s almost certainly the same mouse in about five different pictures in different groups and different time points. In [another paper], you’ve got a western-blot figure, and the same band has appeared multiple times across the whole lot. Not just one splice or one clumsy copy and paste; it’s the same band that has been superimposed into that block quite carefully. What did you think of the DFCI’s acknowledgement that it would seek retractions for 6 papers and corrections for 31 others? I’ve flagged about 58 papers. In 16 or 17 of

those, they say the data were collected at other institutions. Three of them, they dispute. I accept that. But I’d like to know what the dispute is. [The DFCI did not respond to explain why it disagrees with the anomalies flagged by David. It also said that one further paper is still under examination.] So that seems like it’s pretty much all of them accounted for. In one sense, I’m relieved. They basically accepted that these are all errors. I stand by what’s on the blog and by what I post on PubPeer. It does leave a frustrated feeling because a lot of these comments have been on PubPeer for ages. Do you think journals are doing enough to correct the scientific record when these cases come up? The response is usually slow, if they respond at all. It’s a very painful process to try to report an accusation of image duplication. It’d be nice if there was an obvious way just to click a button and flag a paper. How can scientists avoid accusations of data impropriety? There’s a pretty simple way to prevent all of this: you should design some fileorganizing system that involves giving research images a sensible name. And then when it comes to checking your paper before you publish it, you need to trace back all the images to the raw data and check them against the metadata. For example, if you have a photo labelled as ‘Day three’, does that correspond to the date the photo was taken or that the experiment happened on? Do you have any recommendations for scientists whose work is questioned? I don’t want to make an environment where people feel harassed. The main thing I’d like to see is a polite response, and to acknowledge whether the error is there or not. Because it’s very frustrating if you say, ‘We’ll look into this,’ without acknowledging the errors on the page or giving a timeline. By Max Kozlov This interview has been edited for length and clarity.

JES2UFOTO/ALAMY

Q&A

Paper mills often sell authorships on nonsense papers to researchers.

HIGH-PROFILE EFFORT WILL TACKLE PAPER MILLS Fake studies are polluting the literature — a group will study the businesses that produce them. By Katharine Sanderson

A

high-profile group of academic ­publishers, research organizations and funders has launched an effort to tackle one of the thorniest problems in scientific integrity: paper mills, businesses that churn out fake or poor-quality journal papers and sell authorships. In a statement released on 19 January, the group outlined how it will address the problem through measures such as closely studying paper mills, including their regional and topic specialties, and improving author-verification methods. “There’ve been too many empty words. This is intended to actually deliver,” says Deborah Kahn, an elected council member of the Committee on Publication Ethics (COPE), a non-profit organization in Eastleigh, UK, and co-chair of the steering group of United2Act, which produced the consensus statement. “Paper milling isn’t an operation, it isn’t an organization: it’s a culture,” says data scientist Adam Day, who founded Clear Skies in London, which has developed a detection tool called The Papermill Alarm. Paper mills have been creating a problem for a long time, he says. “And it’s been ignored for a long time.” Estimates suggest that hundreds of thousands of paper-mill publications are polluting the scientific literature. Paper mills often sell authorships on bogus papers to researchers

trying to pad their CVs. One analysis indicates that some 2% of all scientific papers published in 2022 resembled paper-mill productions. Detecting these articles is difficult — although there are growing technological efforts to spot them — and shutting down the operations that produce them is even harder. Researchers are also concerned that the rise of generative ­artificial intelligence (AI) tools will exacerbate the problem by providing more ways to quickly generate fake papers that can dodge current detection methods.

Five-point plan United2Act’s statement is the outcome of a summit last May, convened by COPE and the International Association of Scientific, Technical and Medical Publishers (STM), based in Oxford, UK. Researchers, research-integrity analysts, publishers and funders attended the meeting and described five areas that need action, enshrined in the statement. Each point has an associated working group, which will: improve education and awareness of the problem; conduct detailed research into paper mills; improve post-publication corrections; support the development of tools to verify the identities of authors, editors and reviewers; and ensure that the groups across publishing that are tackling the issue communicate. Signatories to the statement include the prestigious funder the European Research Nature | Vol 626 | 1 February 2024 | 17

News in focus Council, the publishing-services company Clarivate and major publishers including Elsevier, Wiley and Springer Nature. (Nature is editorially independent of its publisher, Springer Nature.) “The consensus statement was really just the basis for what we’re going to do next,” says Kahn.

Most of the groups have already held their first meetings. Anna Abalkina, a social scientist at the Free University of Berlin and an independent research-integrity analyst, attended the summit. She sits on three of the five working groups, including the ‘research group’, which will take a scholarly approach to studying paper mills to fill in knowledge gaps about their status and scale. Abalkina hopes the group will publish two or three papers on its work. She adds that this group won’t produce results quickly, because its members are from a range of specialties, including publishers, researchers and sleuths — people, including Abalkina, who work out where the paper mills are and who’s running them. But “this is the most important advantage of these working groups, because they unite experts from different spheres”, says Abalkina. The ‘education and awareness group’ will develop educational aids on paper mills and how they operate, says Kahn. “It brings together those who already have training with people that need it,” she says. “Because of language and cultural differences, it is important to tailor the training to the audiences.” Paper-mill operations have been identified in south Asian countries, as well as in China, Russia and Iran.

Data for action To ensure that action follows, Kahn has asked each group for a progress report by June, when she will update attendees of the World Conference on Research Integrity in Athens, the most high-profile meeting on the subject. Day, who wasn’t involved in the statement, welcomes it. He would like to see more detail on what the working groups will be doing, so that people like him can apply it to their own work. His company has identified locations of paper mills and their leaders. But distributing that information is difficult, says Day. “We know where the problems are. We know who is responsible for a lot of it. And that seems like actionable data,” he says. “But it is going to be up to other stakeholders to work out how they want to act on that.” Several efforts will be needed to make a dent in the problem, say researchers. “Paper mills are very shape shifting. They anticipate what we’re doing,” says Kahn. “The thing that’s exciting about this is that we are actually starting to do the work.” 18 | Nature | Vol 626 | 1 February 2024【 】

JAXA

Deep study

Artist’s impression of the SLIM spacecraft landing on the Moon.

JAPAN’S SUCCESSFUL MOON LANDING WAS THE MOST PRECISE EVER Landing within 100 metres of its target zone, the craft has pioneered image-based automatic navigation. By Ling Xin

O

n 20 January, Japan became the fifth country to soft-land a spacecraft on the Moon, using precision technology that allowed the craft to touch down closer to its target landing site than any mission has before. But the spacecraft lost communication with Earth for several days, owing to a power failure, before re-establishing contact on 28 January. Telemetry data showed that the Smart Lander for Investigating Moon (SLIM) touched down in its target area near Shioli crater, south of the lunar equator. NASA’s Lunar Reconnaissance Orbiter confirmed SLIM’s location. It landed four months after lifting off from the Tanegashima Space Centre, south of Japan’s main islands. “SLIM has made it to the Moon’s surface. It has been communicating with our ground station and responding to commands from Earth accurately,” Hitoshi Kuninaka, vice-president of the Kanegawa-based Japan Aerospace Exploration Agency ( JAXA), told a press conference after the landing was completed. “However, it seems that the solar cells are not generating electricity at this point, and the spacecraft is operating solely on its battery,”

Kuninaka said. “The battery will last several more hours — those hours will be the remaining life of SLIM.” He said the agency would continue to monitor the lander, because there was a chance that the panels could start working again. More than a week later, they did. The successful landing came around two weeks after the launch of a commercial US spacecraft destined for the Moon — but a propellant problem will prevent that craft from landing as planned. It’s also been almost a year since a Japanese commercial lander experienced a failure and crashed into the Moon; lunar landings are notoriously difficult to pull off, and a commercial company has yet to do so. Namrata Goswami, a space-policy researcher at Arizona State University in Phoenix, says the successful landing is “a big win for Asia”. Only China, India and Japan have put a spacecraft on the Moon in the past decade. India successfully landed one last August.

Innovative technology SLIM achieved its main goal — to land on the Moon with an unprecedented accuracy of 100 metres, which is a big leap from previous ranges of a few to dozens of kilometres. SLIM used vision-based navigation technology,

which was intended to image the surface as it flew over the Moon, and could locate itself quickly by matching the images with onboard maps. It remains unclear whether the car-sized, 200-kilogram spacecraft actually touched down in the planned two-step manner. Previous craft landed on four legs simultaneously on a relatively flat area of the Moon. SLIM was designed to hit a 15-degree slope outside Shioli crater first with one leg at the back of the craft before tipping forward and stabilizing on the four front legs. Two small robots were intended to eject from SLIM before touchdown, says Kuninaka. Their purpose was to take images of the lander that they would send back to Earth. Images that one returned suggest that SLIM had rolled upside-down during its touchdown, preventing its solar cells from facing the Sun. However, more than a week later, enough sunlight had reached the solar cells for SLIM to re-establish communication with Earth. If SLIM has enough power, scientists plan to use a specialized camera — the only scientific instrument on board — to look for a mineral called olivine in the Moon’s mantle. “If we can detect the olivine’s components and compare it with its counterpart on Earth, it may offer new evidence to support the theory that the Moon was part of Earth long time ago,” says Shinichiro Sakai, the mission’s project manager at JAXA. The camera would also help to confirm the origin of the Apollo 16 Moon samples. The landing site is about 250 kilometres east of where Apollo 16 touched down in 1972, and west of an ancient lunar sea called Mare Nectaris. “In Apollo 16 samples, we found exotic basalts, which were most likely ejected from Mare Nectaris,” says Clive Neal, a planetary geologist at the University of Notre Dame, Indiana. By helping to confirm the source, SLIM could tell scientists a lot about impact dynamics and the chemistry of the ancient sea. “It would show that smaller missions can still be very productive and do important science,” Neal says.

LONG-COVID SIGNATURES IDENTIFIED IN ANALYSIS OF BLOOD PROTEINS Proteins involved in clotting and inflammation could help to unravel the complexity of long COVID. By Miryam Naddaf

R

esearchers have developed a computational model that predicts how likely a person is to develop long COVID, on the basis of an analysis of more than 6,500 proteins found in blood. In a study published on 18 January in Science, the team compared blood samples from people who tested positive for COVID-19 with ones from healthy adults. There were found notable differences in the composition of proteins in people with long COVID, those who had recovered and those who were never infected (C. Cervia‑Hasler et al. Science 383, eadg7942; 2024). The analysis suggests that proteins involved in immunity, blood clotting and inflammation could be key biomarkers in diagnosing and monitoring long COVID, which affects an estimated 65 million people worldwide. The condition has been linked to more than 200 symptoms, including brain fog, fatigue and shortness of breath, which can persist for months or years after a SARS-CoV-2 infection. The small study “will hopefully pave the way for further studies to try and develop therapies for what is, at the moment, pretty much an impossible thing to treat”, says Aran

Singanayagam, a respiratory physician at Imperial College London. The study included 39 healthy adults who had never tested positive for COVID-19 and 113 people who had, of whom 40 had long COVID, defined as having symptoms 6 months after initial infection. Of those, 22 still had symptoms 12 months after first testing positive. The researchers analysed 6,596 proteins across 268 blood samples, which were collected from participants once during the acute phase and again six months later. They found several differences in the blood of people with long COVID compared with those without it, including an imbalance in proteins involved in blood clotting and inflammation. Compared with healthy participants and those who had fully recovered from COVID-19, people with long COVID had lower levels of a protein called antithrombin III, which helps to prevent blood clots, and higher levels of the proteins thrombospondin-1 and von Willebrand factor, both of which are associated with clot formation. When they examined blood cells from a subset of participants, the researchers found that the expression of a protein called CD41 on white blood cells was lowest in healthy people and highest in people who had 12-month long COVID.

JOVELLE TAMAYO/THE WASHINGTON POST VIA GETTY

Moon rush Sakai and his team hope that SLIM’s pinpoint landing technology will give Japan a head-start in a future planned mission to the Moon. “This technology can be applied to many missions and may constitute a Japanese contribution in international cooperation,” says Sakai. India and Japan are already planning a joint mission. The Moon is experiencing an uptick in visitors. SLIM was the second Moon landing attempt this year, after the ill-fated US Peregrine spacecraft. This month, the US firm Intuitive Machines will continue its attempt to become the first private company to land a spacecraft on the Moon. Later this year, China will launch its Chang’e-6 mission to bring back samples from the far side of the Moon.

Long COVID is characterized by symptoms such as fatigue and brain fog.

Nature | Vol 626 | 1 February 2024 | 19

News in focus participant would develop long COVID on the basis of various protein levels, along with other factors such as age and body mass index. When applied to a separate data set, the model performed well in predicting which participants would have 12-month long COVID.

‘We are at the beginning’ Some of the team’s findings fit well with existing theories on the causes of long COVID, and “could open up new research regarding [therapies] that could help”, says Cervia-Hasler. But the analysis involved a relatively small number of participants, and it does not pinpoint the root cause of the condition, which has been a barrier to developing treatments. “We are at the beginning of the exploration of this emerging field,” says Chakrabarti. Singanayagam adds that, because long COVID involves such a broad range of symptoms, there are likely to be several underlying causes. “Bigger studies are needed,” he says. “It isn’t going to be a single mechanism underlying all of these symptoms.”

HOW DOES CHRONIC STRESS HARM THE GUT? NEW CLUES EMERGE A bacterium in the intestines of stressed mice interferes with cells that protect against pathogens. By Max Kozlov

M

ental stress has long been linked to flare-ups of gastrointestinal conditions such as irritable bowel syndrome (IBS). Now, researchers have uncovered exact details of one way that stress can harm the intestines — by setting off a biochemical cascade that reshapes the gut microbiome (W. Wei et al. Cell Metab. https://doi.org/md4v; 2024). Their study, published on 22 January, is nice, says Christoph Thaiss, a microbiologist and neuroscientist at the University of Pennsylvania in Philadelphia. It highlights how the brain — despite being far away from the gastrointestinal tract — can still influence it. IBS, which causes abdominal pain and diarrhoea, affects one in ten people. Up to ten million people worldwide have inflammatory bowel disease, which causes inflammation of the intestines and triggers similar symptoms. Study co-author Xiao Zheng, a metabolism researcher at the China Pharmaceutical University in Nanjing, wanted to understand what happens on the cellular level to trigger these

20 | Nature | Vol 626 | 1 February 2024 【】

conditions. He and his colleagues exposed mice to chronic stress for two weeks and observed the effects. The animals ended up with reduced

The small intestine’s lining has finger-like structures that help to absorb nutrients.

levels of cells that help to protect the intestines from pathogens, compared with mice that weren’t stressed. This is because the metabolism of intestinal stem cells that normally transform into these protector cells was malfunctioning. Seeking a reason, the researchers turned to the animals’ microbiomes — the bacteria and other microorganisms in their guts that aid digestion. Previous work2 showed

“When we suffer from stress, our gut microbiome is also suffering from stress.” that activation of the sympathetic nervous system, which is responsible for the body’s ‘fight or flight’ response and often triggered by mental stress, can reshape the microbiome (M. D. Gershon and K. G. Margolis J. Clin. Invest. 131, e143768; 2021). Some bacteria of the genus Lactobacillus, which occur naturally in the gut and proliferate under stressful conditions, make the chemical indole-3-acetate (IAA). The researchers found that a raised level of IAA, triggered by stress, prevented the mouse intestinal stem cells from becoming protector cells.

A piece of the puzzle Although this study was conducted in mice, the researchers gathered evidence that their findings might hold true for humans: the team found elevated levels of both Lactobacillus bacteria and IAA in the faeces of people with depression, compared with samples from people without it. “When we suffer from stress, our gut microbiome is also suffering from stress,” Zheng says. The authors also found a possible antidote, in mice at least. When they gave stressed mice a supplement called α-ketoglutarate, which is taken by some bodybuilders, it kick-started the metabolism of the impaired stem cells in their intestines. Thaiss warns, however, that more work is needed to understand the longterm effects of the supplement and whether it reduces the symptoms of gut dysfunction. Because stress triggers a raft of biochemical changes in the body, this study alone won’t tell the whole story of the stress–gut connection, Thaiss adds. He also points out that the IAA study tackled only the downstream effects of stress on the gut, and says that more work is needed to understand how the brain transmits signals that kick off the bacterial proliferation. Zheng responds that he and his colleagues plan to investigate these upstream effects next, in addition to further testing the safety and efficacy of α-ketoglutarate. The IAA study “is certainly a new piece of the puzzle”, says Gerard Clarke, a neurogastro­ enterologist at University College Cork in Ireland, “but how many pieces are in that puzzle is still an open question”.

STEVE GSCHMEISSNER/SPL VIA ALAMY

CD41 is typically found on platelets — cell fragments involved in clotting — and its presence on white blood cells indicates abnormal clumping of these cells. “That could contribute to microclots,” says Lisa Chakrabarti, an immunovirologist at the Pasteur Institute in Paris. Some scientists think that these tiny blood clots could be the cause of some long COVID symptoms, by blocking oxygen flow to tissues. The researchers also found increased activation of the complement system — part of the body’s immune defences, which normally help to clear infections — in people with long COVID, both during initial infection and six months later. People with six-month long COVID had reduced levels of some proteins involved in the system and elevated levels of others, compared with fully recovered or healthy people. An imbalance of these proteins could cause tissue damage, says study co-author Carlo CerviaHasler, a physician–scientist at the University of Zurich in Switzerland. Using machine learning, the researchers then created a model to predict whether a

JENNIFER SU, PETER WANG, NICOLE LESTER & WILLIAM L. HWANG

Feature

A 3D model system shows how nerve cells (magenta) interact with cancer cells (green).

HOW CANCER HIJACKS NEURONS TO GROW AND SPREAD

Researchers are exploring the links between cancer and the nervous system. Could blocking crosstalk help to treat the disease? By McKenzie Prillaman

L

ightning bolts of lime green flashed chaotically across the computer screen, a sight that stunned cancer neuroscientist Humsa Venkatesh. It was late 2017, and she was watching a storm of electrical activity in cells from a human brain tumour called a glioma. Venkatesh was expecting a little background chatter between the cancerous brain cells, just as there is between healthy ones. But the conversations were continuous, and rapid-fire. “I could see these tumour cells just lighting up,” says Venkatesh, who was then

22 | Nature | Vol 626 | 1 February 2024 【】

a postdoctoral researcher at Stanford University School of Medicine in Stanford, California. “They were so clearly electrically active.” She immediately began to think about the implications. Scientists just hadn’t considered that cancer cells — even those in the brain — could communicate with each other to this extent. Perhaps the tumour’s constant electrical communication was helping it to survive, or even to grow. “This is cancer that we’re working on — not neurons, not any other cell type.” To see the cells fizz with so much activity was “truly mind blowing,” says Venkatesh, who

is now at Harvard Medical School in Boston, Massachusetts. Venkatesh’s work formed part of a 2019 paper in Nature1, which was published alongside another article2 that came to the same conclusion: gliomas are electrically active. The tumours can even wire themselves into neural circuits and receive stimulation directly from neurons, which helps them to grow. The findings have been pivotal in the emerging field of cancer neuroscience, in which researchers are parsing the many ways in which cancer — even outside the brain — co-opts the

EYE OF SCIENCE/SPL

nervous system for its own benefit. In much the same way as tumours recruit blood vessels to feed themselves and grow, cancer relies on the nervous system for everything from initiation to spread. The interaction between oncology and neuro­science is just beginning to unravel in this once-overlooked part of the tumour’s environment. Scientists are starting to understand which neurons and signals are involved, but new-found interactions with the immune system are making the story even more complicated. As researchers dig deeper into the relationship between cancer and the nervous system, therapies that target the connections are emerging. Some of these treatments use existing drugs to improve outcomes in people with cancer. “Where we’re headed with this is helping patients,” says cancer biologist Erica Sloan at Monash University in Melbourne, Australia. “Yes, there’s the intellectual delight of understanding what goes on at the biology level. But the key goal is, ‘How do we translate this?’”

Invasion and persuasion Scientists first spotted liaisons between cancer cells and neurons almost 200 years ago. In the mid-nineteenth century, French anatomist and pathologist Jean Cruveilhier described a case in which breast cancer had invaded the cranial nerve responsible for facial movement and sensations. This was the first account of perineural invasion, in which cancer cells weave in and around nerves — and then spread. The phenomenon is a sign of an aggressive tumour and foreshadows poor health outcomes. For a long time, scientists and health professionals thought that nerves served passively as a highway to transport cancer and its associated pain. Many viewed the nervous system as “the victim — the structure that gets destroyed by or damaged by the cancer”, says neuro-oncologist Michelle Monje at Stanford University School of Medicine, who was Venkatesh’s adviser. But in the late 1990s, urological pathologist Gustavo Ayala, now at the University of Texas Health Science Center at Houston, started investigating the interaction a little more closely. He placed mouse nerves in dishes speckled with human prostate cancer cells. Within 24 hours, the nerves began growing little branches called neurites, which reached out towards the diseased cells. Once they made contact, the cancer travelled along the nerves until it reached the neuronal cell bodies3. Nerves weren’t just bystanders: they actively sought a connection with cancer. “I thought it was real, and I decided to make it my career,” says Ayala. He soon became known as ‘the nerve guy’. “People didn’t quite make fun, but they didn’t share my interest in the field,” he says. In 2008, Ayala reported another strange

An immune cell (left) next to a cell from a nervous-system cancer called an ependymoma.

phenomenon. Prostate-cancer tumours taken from people who underwent surgery contained more nerve fibres, known as axons, than did samples from healthy prostates4. Not everyone found this result odd, however. Some scientists were starting to view tumours as being organs themselves, because they contain multiple cell types, a scaffolding structure, blood vessels and other elements that distinguish them from being clumps of cancer cells. But “there was a piece missing in the landscape — it was nerves”, says Claire Magnon, a cancer biologist at the French National Institute of Health and Medical Research in Paris. That hunch led to a groundbreaking paper in 2013. She and her colleagues documented nerve fibres sprouting in and around prostate tumours in mice5. Moreover, severing the connections to the nervous system brought the disease to a standstill. In a few years, an avalanche of research demonstrated the same thing happening in cancers elsewhere, including in the stomach, pancreas and skin. Some of the severed nerves carry cancer-associated pain, and researchers already knew that blocking those paths in people with pancreatic cancer could bring some relief. “The stars were sort of aligned,” says neuroscientist Brian Davis at the University of Pittsburgh in Pennsylvania. The converging results showed “that this component of the tumour microenvironment, that had basically been ignored, was playing some role”.

Hitting a nerve But where these cancer-infiltrating nerves came from baffled researchers. Work conducted in the following years suggested that cells in the tumour can turn into neurons, or at least acquire neuron-like features. And in 2019, Magnon and her colleagues reported another origin6. They saw cells called neural progenitors travelling

through the blood to prostate tumours in mice, where they settled and fledged into neurons. Somehow, cancers were influencing the brain region that contains these cells — an area called the subventricular zone. In mice, these cells are known to help heal certain brain conditions, such as strokes. Some evidence suggests that the same region produces neurons in adult humans, although the idea is controversial. The following year, another team discovered that cancer can force neurons to change their identities. In a study of oral cancer in mice, researchers found that a group of nerves that relay sensations to the brain, called sensory neurons, acquired features of a different type of neuron that is usually rare in the oral cavity: sympathetic neurons, which are responsible for the ‘fight or flight’ response7. “Now they’re wearing two hats,” says cancer neuroscientist Moran Amit at the University of Texas MD Anderson Cancer Center in Houston, who co-led the study. The transformation might help tumour growth, because sympathetic nerves have been shown to benefit certain cancers. But the relationships between nerve types and their effects on tumours are complicated. In the pancreas, for instance, a push and pull exists between two types of nerve that have opposite effects on tumours. Sympathetic nerves participate in a vicious feedforward loop that aids the growth of cancer. They release signals that instruct diseased cells to secrete a protein called nerve growth factor, which draws in more nerve fibres. Their counterparts — para­ sympathetic nerves, which are responsible for the ‘rest and digest’ response — send chemical messages that thwart disease progression. But in stomach cancer, parasympathetic signals act in the opposite way, encouraging the tumour to grow. And in prostate cancer, both types of nerve aid tumours, with Nature | Vol 626 | 1 February 2024 | 23

Feature sympathetic nerves helping during the early stages of cancer development and parasympathetic nerves boosting later-stage spread. “Every cancer is a little bit different in how it interacts with the nervous system,” says gastroenterologist Timothy Wang at Columbia University in New York City. This means that treatment targets must be specific to the type of cancer and how the cancer connects with or uses the nervous system. Neurons can have direct effects on cancers, or they can act indirectly, by damping down the immune system so that it can’t fight tumours as effectively. A 2022 discovery hints at one such mechanism: a chemical called calcitonin gene-related peptide (CGRP), which is released by sensory nerves, can quell the activity of certain immune cells, making them ill-prepared to ward off cancer8. Neurons can suppress immune-cell activity to keep themselves safe, because too much inflammation can harm them. So, not only do nerves provide a route and scaffolding for cancer’s spread, says cancer neuroscientist Jami Saloman at the University of Pittsburgh, but they also seem to provide a safe harbour. A tumour can “tuck itself into the nerves”, Davis says, where it is protected from both the immune system and medication because drugs have a hard time entering nerves. “The cancer cells can hang out while they’re waiting for the storm of biologics and chemotherapy to pass,” he notes. “And then they can re-emerge.”

Central takeover Some of the most aggressive cancers affect the brain. As Venkatesh and others found, cancer cells even form direct synapses with neurons, the signals of which help them to grow. A paper published alongside the two 2019 brain-cancer papers showed that breast-cancer metastases in the brain could also form synapse-like connections9. And previous research has linked brain metastases with cognitive impairment. There are yet more ways in which brain cancers seem to act like brain cells. Last November, Monje’s laboratory reported that gliomas strengthen their neuronal input using a classic brain-signalling method10. When exposed to a protein that helps neurons to grow, called brain-derived neurotrophic factor, glioma cells respond by spawning more receptors that can receive signals from neurons. “It’s exactly the same mechanism that healthy neurons use in learning and memory,” Monje says. “Cancer doesn’t really invent anything new — it just hijacks processes that already exist.” Furthermore, just like in networks of neurons, some glioma cells can generate their own rhythmic waves of electrical activity11. “They are simply like little beating hearts,” says Frank Winkler, a neuro-oncologist at the German Cancer Research Center in Heidelberg, whose lab conducted the work. 24 | Nature | Vol 626 | 1 February 2024 【】

Those electrical surges radiate throughout the cancer cells using a network of thin, stringy bridges called tumour microtubes, which Winkler’s group started studying several years ago. The activity choreographs cancer-cell proliferation and survival — just as pacemaker neurons orchestrate activity during the formation of neural circuits. “Yet again, cancer is hijacking an important neural mechanism of neuro­development,” Winkler says. Brain cancers can even have effects on whole networks. A study last May found that gliomas can reshape entire functional circuits in the brain12. People with tumours that infiltrated speech-production areas were asked to name items described in audio or shown in pictures.

CANCER DOESN’T INVENT ANYTHING NEW. IT JUST HIJACKS PROCESSES THAT ALREADY EXIST.” Electrodes on the surface of their brains showed that the language task didn’t just stimulate those key language regions — the entire tumour-infiltrated area, including regions not usually involved in speech production, spiked in activity as well. The more functionally connected the tumour was to the rest of the brain, the worse people did on the task, and the less time they were expected to live. “The tumour had remodelled the functional language circuitry to feed itself,” says Monje, who co-authored the work. She remembers her horror when she looked at the results. “I get goosebumps when I think about the first time I saw that data.”

Bench to bedside and beyond These initial discoveries are already pointing to potential cancer treatments. They also hint at why existing options often bring brain-draining side effects. Many people undergoing chemotherapy experience cognitive decline, or ‘chemo brain’, and degeneration of nerve fibres elsewhere in the body, says Venkatesh. Despite being an effective way to attack cancer, if chemotherapy destroys neurons elsewhere in the body, “that’s quite obviously not good for the patient”, she adds. One tactic is to target specific prongs of the nervous system. And existing therapies might be able to help. “We have the drugs to target almost every branch of the nervous system,” Amit says. “Most of those drugs have a very established safety profile.” Beta blockers, for instance, can disrupt

signals from sympathetic nerves that drive cancer progression in the breast, pancreas, prostate and elsewhere. These drugs have been used to treat heart problems such as high blood pressure, and sometimes also anxiety, since the 1960s. Sloan has wanted to repurpose the drugs for the past decade, but at first she faced resistance. People often remarked, “If beta blockers were going to do anything to cancer, we would know that already,” she recalls. To explore the connection, she led a phase II clinical trial, published in 2020, testing the beta blocker propranolol in people with breast cancer. Taking the medication for just one week reduced signs of the cancer’s potential to metastasize13. Another phase II trial, inspired by observational studies that have linked beta-blocker use to better health outcomes, demonstrated that it was safe to combine chemotherapy and propranolol in people being treated for breast cancer14. And last year, Sloan found that the drug enhances a common chemotherapy treatment15. Other researchers are repurposing drugs that interrupt neuronal communication, including medications developed for seizures and migraine. At least one clinical trial is aiming to block the synapses formed between neurons and cancer cells in gliomas using an anti-seizure drug, which calms hyperexcitable cells. Another trial in the planning stages will look at whether people receiving immunotherapy for skin or head-and-neck cancer would also benefit from taking a migraine medication. It’s thought that migraines can be triggered by high levels of CGRP, the molecule that can blunt the activity of some immune cells in cancer. So the medication, which blocks CGRP receptors, could counteract CGRP and allow immune cells to help fight cancer again. Venkatesh imagines that a cocktail of drugs with complementary effects will probably be needed to control the disease. “There is really no silver bullet,” she says. The field is only just beginning to unravel this insidious relationship, and questions abound. “I think I would need 50 lives to go after all of them,” Winkler says. McKenzie Prillaman is a freelance science journalist in Washington DC. 1. Venkatesh, H. S. et al. Nature 573, 539–545 (2019). 2. Venkataramani, V. et al. Nature 573, 532–538 (2019). 3. Ayala, G. E. et al. Prostate 49, 213–223 (2001). 4. Ayala, G. E. et al. Clin. Cancer Res. 14, 7593–7603 (2008). 5. Magnon, C. et al. Science 341, 1236361 (2013). 6. Mauffrey, P. et al. Nature 569, 672–678 (2019). 7. Amit, M. et al. Nature 578, 449–454 (2020). 8. Balood, M. et al. Nature 611, 405–412 (2022). 9. Zeng, Q. et al. Nature 573, 526–531 (2019). 10. Taylor, K. R. et al. Nature 623, 366–374 (2023). 11. Hausmann, D. et al. Nature 613, 179–186 (2023). 12. Krishna, S. et al. Nature 617, 599–607 (2023). 13. Hiller, J. G. et al. Clin. Cancer Res. 26, 1803–1811 (2020). 14. Hopson, M. B. et al. Breast Cancer Res. Treat. 188, 427–432 (2021). 15. Chang, A. et al. Sci. Transl. Med. 15, eadf1147 (2023).

Obituary

John L. Heilbron (1934–2023)

Historian of science who wove together scientists and institutions.

KEITH MORRIS/HAY FFOTOS/ALAMY

J

ohn Heilbron’s intellectual rigour transformed how scholars approach the history of science. The intersection of people, scientific ideas and institutions was his domain, ranging from early modern European astronomy to the revolution of twentieth-century physics. The author of more than 20 books, Heilbron was known for his razor-sharp analysis and perceptive discernment of human frailties and strengths. He has died aged 89. Heilbron understood institutions and how they channelled and sustained a scientist’s work. In Lawrence and his Laboratory, for instance, a 1990 book he wrote with fellow science historian Robert W. Seidel, he laid bare the world of physicist Ernest Lawrence, the inventor of the ‘cyclotron’ particle accelerator whose work led to the setting up of National Laboratories in Berkeley and Livermore, California. By putting discussions of nuclear science and radioisotopes cheek by jowl with descriptions of the building and financing of institutions, the book set the stage for similar histories in other areas of ‘big science’. And Heilbron’s 1999 study The Sun in the Church discussed how, from the sixteenth to the eighteenth century, the Catholic church sought to resolve astronomical questions regarding solar motion using meridian lines laid into cathedral floors. Turning the narrative of hostility between religion and science on its head, Heilbron built an empirical case for how church clerics had contributed to the triumph of heliocentric theory. And he did it with precision and verve. John Heilbron was born in San Francisco, California, in 1934 and he studied at the University of California, Berkeley. He did undergraduate and master’s degrees in physics in the heady time after the Second World War, before switching to the history of physics for his PhD, which he earned in 1964. Heilbron studied at Berkeley with Thomas S. Kuhn, whose work at the intersection of history and the philosophy of science introduced the world to the concepts of ‘scientific revolutions’ and ‘paradigm shifts’. Kuhn also underscored the crucial role of the scientific community as the arbiter of how well a theory fits the real world. The student took seriously the teacher’s demand that scholars consider the history of science as it really happened, rather than favouring the smoothed-out narratives that retrospection might bring. But Heilbron didn’t follow Kuhn’s philosophical

tack, because he considered history to be hard enough without engaging in abstract arguments about how science works. Heilbron’s early work examined the theory of atomic structure in the decades around 1900. For his PhD, he trawled through published papers that led to the model of the atom proposed by Danish physicist Niels Bohr and gathered archives and testimonies from the founding generation of quantum physicists. The field of quantum physics and the

“Heilbron was a driving force in making the history of science a discipline of its own.” transformation of scientific institutions that occurred on the cusp of the twentieth century were ripe for study. But he did not stop at that era. He also studied Europe’s early modern period (1450–1789), focusing on theories of electricity and natural philosophy. And he pursued broad questions in the physical sciences, including ideas and societal impacts that emerged from the field across centuries. Appointed Kuhn’s successor at the faculty in Berkeley in 1967, Heilbron remained there until his retirement in 1994. He was prodigiously productive in many arenas: as a historical scholar, research collaborator and trainer of young historians. Alongside

his detailed studies of electricity, quantum physics and Euclidean geometry, he wrote astute biographies. He produced portraits of German physicist Max Planck, Italian astronomer Galileo Galilei and others, homing in on how their science interacted with their human character. Noting Planck’s longing for abstracted harmony and Galileo’s literary finesse, he always found something new to say. Heilbron was a driving force in making the history of science a discipline of its own. For 27 years, he shaped the field as editor-in-chief of the journal then called Historical Studies in the Physical Sciences. He saw his editorial work as a commitment to teaching others how to rise to his high standards of rigorous argument, by removing what he saw as intellectual fluff and superfluous words. Scholars whose work he edited remember gasping at how vigorously he would cut into, rearrange and rewrite their prose. His own writing was entertaining and literate, witty and acerbic. When he spoke on public occasions, he did not shy away from expressing sharp views. However, both colleagues and students remember the affection and care he showed them. John was a man of institutions, even though he preferred to work on his own terms. He wove together international networks of historians of science, and, at Berkeley, rose to become the university’s vice-chancellor. His broad scientific knowledge and sharp insights into human behaviour helped him to work with the university’s competing power centres and to make deals that stuck. He received his profession’s highest awards, including the Sarton Medal in 1993, the 2001 Pfizer Award and the Wilkins Prize Lecture of the Royal Society of London in 2006. When John retired from his administrative post, he was free to pursue history again. He moved to live near Oxford, UK, and continued his editorial work, travels and research, accompanied by his wife Alison. A scholar’s scholar, John knew that people — individuals and their relationships — are at the heart of all intellectual ventures. Making a life in the history of science as the field emerged, he contributed profoundly to shaping it. Cathryn Carson is professor of history of science, past director of the Office for History of Science and Technology and chair of the Department of History at the University of California, Berkeley. e-mail: [email protected]

Nature | Vol 626 | 1 February 2024 | 25

Setting the agenda in research

ANNE-CHRISTINE POUJOULAT/AFP/GETTY

Comment

Getting access to samples will become increasingly important as approaches for the molecular profiling of tumours improve.

The way we name cancers needs to change Fabrice André, Elie Rassy, Aurélien Marabelle, Stefan Michiels & Benjamin Besse

Classifying metastatic cancers according to their organ of origin is hampering access to potentially life-saving drugs.

26 | Nature | Vol 626 | 1 February 2024【 】

O

ver the past century, the two main approaches to treating people with cancer — surgery and radiation — have focused on where in the body the tumour is. This has led to medical oncologists and other health-care providers, regulatory agencies, insurance companies, drug firms — and patients — categorizing cancers according to the organ in which the tumour originated. Yet there is a growing disconnect between classifying cancers in this way and developments in precision oncology, which uses the molecular profiling of tumour and immune cells to guide therapies.

More than ten years ago, for example, investigators in the United States showed in a clinical trial that the drug nivolumab could improve outcomes for certain individuals with cancer1. In the trial — which included people with different ‘types’ of cancer (as conventionally defined), from melanoma to kidney cancer — nivolumab shrank some people’s tumours by more than 30%, but it had little or no effect on the tumours of others. Nivolumab targets PD1. This is a receptor of a protein called PD-L1, which helps cancer cells to escape attack from the immune system. Of the 236 trial participants whose tumours could be assessed, 49 responded positively

SOURCES: BRCA–PARP: H. E. BRYANT ET AL. NATURE 434, 913–917 (2005); H. FARMER ET AL. NATURE 434, 917–921 (2005). OLAPARIB TRIAL: P. C. FONG ET AL. N. ENGL. J. MED. 361, 123–134 (2009). OVARIAN APPROVAL: G. KIM ET AL. CLIN. CANCER RES. 21, 4257–4261 (2015). BREAST: HTTPS://GO.NATURE.COM/3STQGSU. PANCREAS: HTTPS://GO.NATURE.COM/3SHNTEJ. PROSTATE: HTTPS://GO.NATURE.COM/4267RUJ

to the treatment. The key determinant was whether their tumour cells were expressing high levels of PD-L1. The logical next step would have been to conduct clinical trials that tested the effects of nivolumab and other PD1 inhibitors in people with metastatic tumours that strongly express PD-L1, regardless of the organ in which their cancer had originated. But because of the way cancers are classified as breast, kidney, lung and so on, researchers had to conduct clinical trials sequentially for each disease type. For about a decade, millions of people with tumours expressing high levels of PD-L1 were not able to access relevant drugs because trials had not yet been conducted for their type of cancer when they became unwell. Those with certain breast or gynaecological cancers expressing PD-L1 had to wait 7–10 years to access PD1 inhibitors. A similar story has played out with most of the drugs tested in clinical trials over the past decade. These include PARP inhibitors, which kill tumour cells carrying mutations in the breast cancer genes BRCA1 and BRCA2. These mutations are now known to occur in multiple tumour ‘types’ as conventionally defined (see ‘Losing lives’), not just in breast cancers. Metastatic cancers (those that have spread beyond the organ where they originated) account for around 67–90% of cancer deaths2,3, and are almost always treated systemically, meaning with drugs that enter the bloodstream. To improve treatments for people with metastatic cancer, the community urgently needs to shift from using organ-based classifications of cancer to using molecular-based ones. This will require radical changes in how medical oncology is structured, conducted and taught.

LOSING LIVES

Habits of a century

and pulmonology; and preclinical research, clinical trials and treatment protocols are often tailored to organ-specific specialties. This attachment to classifying cancer — and addressing it — on the basis of the organ in which it originated is stalling progress in multiple ways. First, it runs counter to the scientific understanding now emerging. The past two decades of cancer research, which have been dominated by efforts to characterize tumours at the cellular and molecular level, have shown that some of the molecular events driving their evolution are shared across different ‘types’ of cancer. Mutations in the tumour suppressor gene TP53, for example, are a feature of most types

In France and some other European countries, patients are not reimbursed if they take drugs that have been tested in trials in which cancers are not defined by the organ in which they originated4. Meanwhile, most of the scientific organizations in oncology, such as the American Society of Clinical Oncology and the European Society of Medical Oncology (ESMO), organize their meetings and issue their guidelines according to ‘organ of origin’. And over the past decade, cancer centres and universities have transformed oncology into multiple organ-specific subspecialities. Hospitals have breast cancer wards, lung cancer wards and so on; medical students are offered modules in subjects such as gastroenterology

Classifying metastatic cancers according to where they originate in the body is delaying treatment for millions of people, because trials must be conducted sequentially for each ‘disease type’. 2005

Studies show that cells with BRCA1/2 mutations can be killed by PARP inhibitors.

2009

Clinical trials begin for a drug called olaparib (a type of PARP inhibitor) involving participants with ovarian cancer.

2014 FDA* approves use of olaparib for ovarian cancer

2018 Use of olaparib for breast cancer is approved 2020 Use of olaparib for prostate cancer is approved

*FDA, US Food and Drug Administration.

Between 2014 and 2018, about 100,000 patients with breast cancer, who might have benefited from treatment with olaparib, died.

2019 Use of olaparib for pancreatic cancer is approved Between 2014 and 2020, about 200,000 patients with prostate or pancreatic cancer, who might have benefited from olaparib treatment, died.

of cancer, as defined by the organ in which the cancer originated. What’s more, most cancer types can be subdivided into different molecular subgroups. Some lung cancers have mutations in the epidermal growth factor receptor (EGFR) gene, some have mutations in the MET gene, others have translocations involving the ALK gene, and so on. Second — as already described — classifying cancer according to the organ in which it originated is making it harder for patients to obtain the drugs that could help them. In fact, when it comes to regulators approving the use of treatments, molecular-based classifications are likely to become ever more important as more drugs are developed using advanced biotechnologies. Antibody drug conjugates, for instance, are antibodies that target membrane proteins expressed by multiple types of cancer to deliver chemotherapy to tumour cells. The antibody drug conjugate trastuzumab deruxtecan has already shown promise in phase I and phase II trials in treating people whose cancers either overexpress the HER2 gene or have a mutated version of it, regardless of the organ in which their cancer originated5,6. Last, the conventional approach to classifying cancer is hampering medical education and patient understanding. Currently, students and practitioners have to memorize and digest an overwhelming amount of information; around 10,000 scientific articles that include the words ‘cancer’ and ‘randomized trial’ are published every year. Implementing a molecular-based classification would make it easier for students and physicians to learn. Students wouldn’t need to memorize the results of clinical trials conducted for each type because trials would be conducted across cancer types. And a knowledge of the molecular mechanisms underpinning the disease would make it easier for students to remember the outcomes of clinical trials. Take, for example, a family of enzymes called PI3Ks, which are involved in cellular processes such as cell growth and proliferation. After a student is taught that these are involved in regulating glucose levels, it should be easier for them to remember that PI3K inhibitors — which are used to treat some people with breast cancer — can lead to hyperglycaemia (high sugar levels in the blood). This means that people with diabetes either should not be given these drugs, or if they are, should have their blood sugar levels closely monitored. Molecular-based classifications could also Nature | Vol 626 | 1 February 2024【 】 | 27

Comment

The way to change Since the first trial of nivolumab in 2012, things have begun to move in a better direction, particularly when it comes to regulatory agencies approving drugs that are focused on the existence of a molecular target rather than the cancer’s organ of origin. In 2017, the US Food and Drug Administration (FDA) approved the use of a drug called pembrolizumab to treat people who have tumour cells with a deficiency in their DNA mismatch-repair system, regardless of which organ the cancer originated in. In 2020, the agency determined that pembrolizumab could also be used to treat people whose tumour cells have high numbers of mutations relative to healthy cells and other cancer cells. And in subsequent years, it has approved the use of several other drugs to treat cancer based on the biological targets of the drugs8. However, a much greater shift in mindset across other regulators and the cancer community at large is needed. Making this happen will require things to be done differently on at least four fronts. Improve guidance and methodologies. Regulatory agencies, scientific societies and insurance companies need to better define what preclinical and clinical evidence is required to determine whether — when it comes to treatment — a specific molecular alteration should be prioritized over the organ in which the cancer originated. Some scientific societies, such as ESMO, are already developing guidelines. And the FDA is working towards defining when a drug can be approved on the basis of a molecular marker, regardless of the organ in which the cancer originated9. Guidance for medical practice that has been developed in other contexts can help with this. For instance, a tool called the Magnitude of 28 | Nature | Vol 626 | 1 February 2024 【】

Clinical Benefit Scale allows clinicians to rank the efficacy of a drug according to various criteria. Each drug is given a score on the basis of how well people respond to the drug in clinical trials, its toxicity, its effects on the survival of trial participants and so on. Likewise, clinicians use the ESMO Scale for Clinical Actionability of Target to rank molecular alterations according to the strength of the evidence that the alteration is important when it comes to treating the person. But a key step in developing such guidelines will be to establish the methodological and statistical tools that would enable researchers to demonstrate that a drug is working in an organ-agnostic way. How many individuals representing how many types of cancer should be included in a clinical trial investigating the

“Innovations could make the widespread adoption of molecular testing feasible even in low- and middleincome countries.” effect of a drug across multiple cancer types, for instance? Or what methodology could prove that there is no difference between two tumour types in terms of responsiveness to a drug? Restructure oncology. The problem of hospital reorganization could first be addressed by cancer centres and university hospitals, given the expertise of these institutions in molecular oncology. Such organizations

could establish organ-agnostic teams that are centred around the interpretation of molecular analyses. In fact, several institutes, including the National PRecISion Medicine Cancer Center (PRISM) at the Gustave Roussy hospital in Villejuif, France, where we work, have already established teams that focus on analysing patients’ molecular profiles, regardless of cancer type. Taking this approach will be harder for small hospitals that do not have clinical departments focused on systemic therapies. But fellowships could help to transfer knowledge between institutions, and raise awareness about the benefits of prioritizing the molecular mechanisms driving cancers in treatment plans. Rethink education. Medical students must be equipped with a comprehensive molecular understanding of carcinogenesis early in their training. This could involve asking students to come up with treatment plans focused on the underlying molecular drivers of cancer — not just to memorize the characteristics of primary tumours and the results of phase III clinical trials. Improve access to molecular testing. The change in approach to the classification of metastatic cancer that we are calling for will not happen unless more people have access to the tests that reveal the molecular alterations in their tumour cells. Since 2020, societies such as ESMO have recommended that all individuals with advanced lung cancer undergo multigene testing 10. Yet, a study involving around 38,000 patients with this condition in the

GUSTAVE ROUSSY

improve people’s adherence to treatment. In our experience, the fact that any two people diagnosed with the same cancer type can be given different treatments causes confusion and misunderstanding. Most people are more familiar with body parts than with gene names. But each patient is affected by only around one to four molecular alterations, limiting the amount of new information that any one person would need to receive. And if patients are also told about the biological mechanisms driving their cancer, they will understand the rationale for treatment better. In support of this, studies from the past two decades7 have shown that telling people living with HIV why their treatment should be matched to their condition — as tracked by the count of CD4 cells (a type of white blood cell) in their blood — increases their adherence to treatment by 5%.

Physicians can identify molecular targets in individual cancers and decide on treatments.

FABRIZIO VILLA/GETTY

People with cancers that have spread beyond the organ of origin are usually treated with drugs that enter the bloodstream.

United States, who were diagnosed between 2010 and 2018, showed that only 22% had molecular test results in their medical record. This is consistent with findings from other studies, conducted both in the United States and elsewhere11,12. Ensuring that all individuals diagnosed with metastatic cancer receive molecular testing comes down to reducing the costs of those tests. Currently, the approach costs around US$3,000 per test in the United States and around $1,000 in Europe. But prices are falling fast: today, complete genome sequencing typically costs around $330, compared to more than $1,100 a few years ago. And some cancer centres are developing ways to perform molecular analyses themselves, so that they don’t have to rely on diagnostic companies13. Moreover, in the coming years, artificial intelligence could be used to identify genomic abnormalities from routine pathological slides at low cost14. Such innovations could make the widespread adoption of molecular testing feasible even in low- and middle-income countries.

Personalized care In the coming years and decades, numerous layers of information could be incorporated into comprehensive characterizations of cancer that are unique to each patient. These include the cancer’s organ of origin, which sometimes remains an important factor in deciding what treatment to try15; the number and size of tumours; and their aggressiveness, as measured by the expression levels of certain genes. Among other potentially useful information is genetic analysis of a person’s germline DNA, which can

provide information about their sensitivity to certain drugs or their chances of experiencing harmful side effects; and their general health, as tracked by levels of fatigue, weight loss and so on. Classifying cancers according to their molecular characteristics would expedite the access of millions of people to effective treatments; it is also the first step towards precision oncology and a deeper biological understanding of how cancer works.

The authors Fabrice André is head of the Research Division at Gustave Roussy, Department of Oncological Medicine, Villejuif, France, a full professor of medicine at Paris-Saclay University, Faculty of Medicine, Le KremlinBicêtre, France, and chair at the National Institute of Health and Medical Research (INSERM) U981, Villejuif, France. Elie Rassy is an oncologist at Gustave Roussy, Department of Oncological Medicine, Villejuif, France, at the Biostatistics and Epidemiology Office at Gustave Roussy, Clinical Research Division, Villejuif, France, and a member of the Oncostat team at the National Institute of Health and Medical Research 9 (INSERM) U1018 CESP, Paris-Saclay University, Villejuif, France. Aurélien Marabelle is principal investigator at Gustave Roussy, Department of Therapeutic Innovation and Early Trials, Villejuif, France, a full professor of medicine at Paris-Saclay University Faculty of Medicine, Le Kremlin-Bicêtre, France, chair at the National Institute of Health and Medical Research (INSERM),

CIC1428, Villejuif, France, and principal investigator at INSERM, U1015, Villejuif, France. Stefan Michiels is head of the Biostatistics and Epidemiology Office at Gustave Roussy, Clinical Research Division, Villejuif, France, and Oncostat team leader at the National Institute of Health and Medical Research (INSERM) U1018 CESP, Paris-Saclay University, Villejuif, France. Benjamin Besse is head of the Clinical Research Division at Gustave Roussy, Department of Oncological Medicine, Villejuif, France, a full professor of medicine at Paris-Saclay University, Faculty of Medicine, Le Kremlin-Bicêtre, France, and principal investigator at the National Institute of Health and Medical Research (INSERM) U981, Villejuif, France. e-mail: [email protected] 1. Topalian, S. L. et al. N. Engl. J. Med. 366, 2443–2454 (2012). 2. Dillekås, H., Rogers, M. S. & Straume, O. Cancer Med. 8, 5574–5576 (2019). 3. Hanahan, D. & Weinberg, R. A. Cell 100, 57–70 (2000). 4. Gill, J., Fontrier, A.-M., Miracolo, A. & Kanavos, P. Access to Personalised Oncology in Europe (London School of Economics and Political Science, 2020). 5. Li, B. T. et al. Ann. Oncol. 34, S459–S460 (2023). 6. Meric-Bernstam, F. et al. J. Clin. Oncol. 41, LBA3000 (2023). 7. Demonceau, J. et al. Drugs 73, 545–562 (2013). 8. Adashek, J. J., Kato, S., Sicklick, J. K., Lippman, S. M. & Kurzrock, R. Nature Cancer 4, 1622–1626 (2023). 9. US Food and Drug Administration. Tissue Agnostic Drug Development in Oncology (FDA, 2022). 10. Hendriks, L. E. et al. Ann. Oncol. 34, 339–357 (2023). 11. Behera, M. et al. J. Clin. Oncol. 40, 9128–9128 (2022). 12. Schilsky, R. L. & Longo, D. L. N. Engl. J. Med. 387, 2107–2110 (2022). 13. Andre, F. et al. Nature 610, 343–348 (2022). 14. Kather, J. N. et al. Nature Med. 25, 1054–1056 (2019). 15. Kopetz, S. et al. N. Engl. J. Med. 381, 1632–1643 (2019). The authors declare competing interests; see go.nature. com/48j8btx for details.

Nature | Vol 626 | 1 February 2024【 】 | 29

LYNETTE DENNY

Comment

Oncologist Lynette Denny has spent 29 years working in the field of cervical cancer prevention.

The world must tackle cervical cancer faster — here’s how Lynette Denny, Ishu Kataria, Lisa Huang & Kathleen M. Schmeler

Without rapid change, the World Health Organization’s goals for tackling cervical cancer by 2030 will be missed. Four specialists share ways to move the needle.

30 | Nature | Vol 626 | 1 February 2024【 】

C

ervical cancer can be prevented through vaccination and be cured if diagnosed early. Yet it still kills more than 300,000 people worldwide each year. Globally, only around 21% of women have had a vaccine against the human papillomaviruses (HPVs) that cause the disease. That number needs to rise to 90% by 2030, if cervical cancer is to be eliminated in the next century — as the World Health Organization (WHO) plans. Screening and treatment should also become routine worldwide, with 70% of people with a cervix checked by the age of 35 and again at 45, and 90% of those with signs of cervical cancer treated. The world is not on track to meet any of these targets. A step change is urgently needed. The tools to vaccinate, screen and treat people are available, and effective. But a lack of funding, staffing and infrastructure — coupled with vaccine hesitancy — are major obstacles. Here, four specialists highlight pockets of good practice that can help to buck the trend.

LYNETTE DENNY TARGET SCHOOLS FOR VACCINATION PROGRAMMES Schools are the most effective place to roll out national HPV vaccination programmes. As long as enrolment levels in education are high, it’s easier to reach young people at school than in health-care settings. Political will is crucial, as is collaboration between a government’s health and education departments — and close communication with schools. I’ve seen the benefits of school-based vaccination at first hand. In 2013, I helped to run a pilot project targeting girls in 31 primary schools in South Africa, in regions where poverty and lack of health-care provision are typically obstacles to high vaccination rates. Our pilot provided 97.8% of eligible girls with what was then the full course of three vaccines1. (In December 2022, the WHO advised that a single dose is sufficient to protect against cervical cancer.) Similar results were

EXPERTISE FRANCE

seen in other pilots, including in Bolivia, Uganda and Vietnam. Scaling these up to country-wide programmes requires determination. But lessons can be learnt from countries around the world. Take Rwanda. In 2011, it became the first low-income country to implement a national HPV immunization programme for girls in the sixth grade (mostly aged 11–12 years). By 2018, more than one million girls had received a vaccination — 98% of the target population2. To do this, Rwanda had to overcome a lack of resources — a common problem in low- and middle-income countries (LMICs) — and put cervical cancer at the apex of its health agenda. Merck provided free vaccines for three years and helped to prepare for the national rollout. Later funding came from GAVI, the Vaccine Alliance — an international organization focused on providing vaccines for children in LMICs. Multiple government departments3 collaborated to set up committees that would oversee all aspects of the programme. Together, these partners organized and delivered school-based vaccinations, rigorously monitored vaccination coverage and ran awareness campaigns4. Girls not enrolled in schools, or absent on vaccination days, were tracked by community health workers and vaccinated at health-care facilities instead. As Rwanda shows, strong, trustworthy and reliable collaboration between all stakeholders is key. We’ve found the same ingredients to be essential in South Africa, where we invested more than six months in regular meetings between health-care workers, education providers, technology specialists and the government to ensure that the roll-out was well coordinated. High-income countries, which typically have more resources and fewer barriers to introducing vaccination programmes, would do well to learn from Sweden. In 2012, the country rolled out a free, school-based HPV vaccination programme for girls as young as 10 — alongside a successful screening and treatment programme. Here again, planning and stakeholder cooperation was essential. By 2021, 90% of girls in the country had received one vaccine dose by age 15, and 84% had received two. Going forward, governments around the world must place prevention of cervical cancer high on the health agenda. Health and education departments must cooperate, and must allocate funding to all aspects of HPV vaccination — from vaccine procurement to infrastructure, awareness campaigns to human resources. Without this focus, roll-out will fail. Lynette Denny is a professor of special projects in the Department of Obstetrics and Gynaecology, Faculty of Health Sciences, University of Cape Town, Cape Town, South Africa. e-mail: [email protected]

ISHU KATARIA BUST MYTHS THROUGH COMMUNICATION CAMPAIGNS People in India are generally not hesitant about vaccines, especially for children. Yet, the Indian government has not included the HPV vaccine in its national immunization programme — even though one person dies from cervical cancer every eight minutes here. In 2019, I interviewed 32 physicians in Kolkata, to try and understand the hesitancy surrounding HPV vaccination5. The physicians’ foremost reason was that many parents associate HPV vaccination with promiscuity. Because HPV is transmitted through sexual intercourse, parents often assume that giving a young child the vaccine will be viewed in the community as a sign that they are sexually active. Physicians were also unclear about the benefits of recommending the HPV vaccine before a child becomes sexually active, and they did not want to risk their reputation by making a recommendation that could be controversial. Similar concerns and misunderstandings are common elsewhere, including in Eastern Mediterranean countries6.

“Raising general awareness — among schoolteachers, parents, children and adolescents — is crucial.” A campaign run by national health departments is needed to instil confidence in the vaccine among physicians. It should make clear that vaccination is most effective between the ages of 9 and 14, because that is when it produces the most robust immune response. The campaign should highlight that the vaccine is extremely safe. It should provide guidance on communicating the benefits to parents in a culturally sensitive way — as a vaccine to prevent cancer, rather than against a sexually transmitted infection. Raising general awareness — among schoolteachers, parents, children and adolescents — is also crucial. The campaign run by the health department for the northeastern state of Sikkim when it first rolled out the HPV vaccine in 2018 provides a blueprint for others to follow. Sikkim’s six-month-long campaign educated physicians, community leaders, government officials, the media and the public through workshops, written materials and television and radio broadcasts. It resulted in 97% HPV vaccine uptake among eligible girls7. Indian states cannot afford to roll out the vaccine unless it is part of the national immunization programme (in which case the government covers the cost of the vaccine).

The launch of an affordable, cost-effective, India-manufactured vaccine by the Serum Institute in September 2022 has put pressure on the Indian government to fund the HPV vaccine, with a decision expected after this year’s election. States should now lay the groundwork for roll-out, following Sikkim’s lead. Key first steps include communication with physicians and parents, along with logistical planning. Ishu Kataria is a senior public-health researcher at the Center for Global Noncommunicable Diseases at RTI International, New Delhi, India. e-mail: [email protected]

LISA HUANG INTEGRATE SCREENING INTO HEALTH-CARE SYSTEMS There is no one-size-fits-all approach to rolling out cervical cancer screening programmes. For LMICs, the best strategies focus on maximizing efficiency — and thereby reducing costs — for resource-poor countries. This can be achieved by integrating screening and treatment programmes into existing health-care systems and facilities. The SUCCESS project, of which I am a director, is trialling such an approach in Burkina Faso, Côte d’Ivoire, Guatemala and the Philippines. Performing screens in existing health-care settings minimizes the need for extra medical workers, who are scarce in LMICs. Screening

Public-health expert Lisa Huang.

Nature | Vol 626 | 1 February 2024【 】 | 31

programmes can be run in primary health-care settings, gynaecology clinics, family-planning services and — importantly — HIV clinics. The last is essential because the 20 million women living with HIV are six times more likely than other women to develop cervical cancer. Local contexts need to be considered. In many countries, staff members will need to be trained in screening, and supply chains and inventory management systems will need to be set up. Digital health-information systems are crucial, allowing patient information to be passed between departments and between health workers to aid follow-up. In the SUCCESS project, we’ve seen the benefits of such digital solutions. In Burkina Faso and Côte d’Ivoire we’ve made use of the DHIS2 Tracker, an app available as part of DHIS2 — an open-source health-information management platform widely used in LMICs. Using a tablet, a health-care worker can input patient information into the tracker along with information about any follow-up needed, which the patient can be told of either by instant messaging or when visiting another health centre. Although it is challenging to collect digital data in LMICs, I am confident that tracking screening will save lives. Governments should aim to implement tracking technologies as soon as possible. People often resist complex changes, so engagement with health-care workers is an essential first step in a move towards digitization, to gain support for the switch. Investment from local and international funders is key, and time must be taken to understand each country’s health-care ecosystem and ensure that new digital solutions are interoperable with those already in use. Lisa Huang is director of the SUCCESS project at Expertise France, Abidjan, Côte d’Ivoire. e-mail: [email protected]

KATHLEEN M. SCHMELER USE INTERNATIONAL MENTORS TO TRAIN DOCTORS LMICs face a shortage of medical providers. Just 4% of the global medical workforce is in Africa, for instance — yet the continent shoulders one-quarter of the global disease burden8. Most LMICs have no formal training programmes for cancer specialists, particularly surgeons. In these countries, more specialized nurses and physicians are urgently needed to diagnose and treat cervical cancer. Global collaboration can help to meet the need for training, as demonstrated by two international projects in which I’ve been involved. Both focused on Mozambique, a country that has no organized screening programme and few trained medical providers. 32 | Nature | Vol 626 | 1 February 2024 【】

SARAH BERGER

Comment

Kathleen Schmeler helps to train medical graduates in Mozambique.

“To scale up these efforts, institutes in high-income countries must coordinate with one another.” In Mozambique, 39 of every 100,000 women die from cervical cancer, compared with the global average of 7.2. First, I co-lead a collaboration between the Mozambique Ministry of Health, the MD Anderson Cancer Center in Houston, Texas, and five institutes in Brazil. The collaboration — which was initiated in 2014 at the request of the First Lady of Mozambique — aims to build capacity in Mozambique by teaching the nation’s medical providers to treat ‘pre-cancerous’ cells. Specialists from Texas and Brazil travel to Mozambique three or four times a year to provide lectures, hands-on training and mentoring to trainee doctors and nurses. We train 30–40 participants each time. Ongoing support is provided through monthly video conferences. Second, I co-chair a global gynaecological oncology fellowship run by the International Society of Gynecologic Cancer (see go.nature.com/4b6edzk) for institutions in LMICs that lack formal training in cancer care. The fellowship site is paired with a partner institute in a high-income country. Fellows — recently, graduates in obstetrics and gynaecology — spend two years undertaking a comprehensive education and training programme, mainly in their home country, but with a few months at the mentor institution. Maputo Central Hospital in Mozambique was

a pilot site when the programme first began in 2017. There are now fellowship sites in 22 countries. Each of these projects initially required a handful of very motivated international mentors. But training and mentoring is now being performed, at least in part, by programme graduates living in Mozambique. To scale up these efforts, institutes in high-income countries must coordinate with one another, and enhance collaborations with health ministries and training institutions in LMICs. Funding for our work has come from small grants, philanthropic, institution and foundation budgets, and often from the volunteers themselves. These types of donation can fund individual projects, but investment from governments, United Nations agencies and industry partners is needed to make the approach work on a global scale. Kathleen M. Schmeler is a professor of gynaecological oncology and associate vicepresident of the Global Oncology Program, MD Anderson Cancer Center, Houston, Texas. e-mail: [email protected] 1. Moodley, I., Tathiah, N., Mubaiwa, V. & Denny, L. S. Afr. Med. J. 103, 318–321 (2013). 2. Sayinzoga, F. et al. Vaccine 38, 4001–4005 (2020). 3. Binagwaho, A. et al. Bull. World Health Org. 90, 623–628 (2012). 4. Kabakama, S. et al. BMC Public Health 16, 834 (2016). 5. Kataria, I. et al. Vaccine X 12, 100228 (2022). 6. Hakimi, S., Lami, F., Allahqoli, L., Alkatout, I. J. Turk. Ger. Gynecol. Assoc. 24, 48–56 (2023). 7. Sankaranarayanan, R. et al. Lancet Oncol. 20, e637–e644 (2019). 8. Boniol, M. et al. BMJ Glob. Health. 7, e009316 (2022). The authors declare no competing interests.

Readers respond

Correspondence Rebuilding Saudi Arabia’s research reputation

Carbon neutrality requires energyefficient AI training

Respect ‘soilscapes’ Brazil must reverse gear on Amazon road to ensure soil health by 2050 development

The news that Saudi Arabia has the highest rate of article retractions among large research-producing nations (see Nature 624, 479–481; 2023) raises concerns about the credibility of the country’s research. Swift action must be taken to avoid this issue hindering its scientific advancement. Controls in institutions must be strengthened to ensure that the quality and ethical standards of research articles are met. Education on research ethics should be prioritized, and a reward system based on quality, not quantity, should be implemented (see M. Uddin & N. Khalaf Alharbi Dig. Health https://doi.org/mchh; 2023). Actions to rebuild credibility could include implementing a post-publication peer-review system that allows full and transparent access to research data for validation, revision and replication. Such a system would help to avoid bias and favouritism in conventional journals, facilitate dynamic discussion and leverage international expert opinions. Strengthened research protocols supported by the Saudi government and the help of specialist external advisers would provide better direction for researchers. Penalties for infringements would deter unethical practices and emphasize the importance of adhering to publication standards. Saudi Arabia has the potential to be a valuable member of the international research community, if it can show its commitment to resolving its current problems.

Artificial intelligence (AI) models have achieved remarkable success, but their training requires a huge amount of energy. The training of large language models in particular generates thousands of tonnes of carbon emissions (see H. Wang Nature 620, 731–732; 2023). This runs counter to the global goal of carbon neutrality by 2050. AI professionals and governments must do more to respond. AI researchers should prioritize the development of computationally efficient AI, including energy-saving AI chips and ‘sparse’ neuralnetwork models that require less energy and memory than current models do. Dedicated efforts are needed to reduce the training requirements of deep-learning algorithms, including speeding up iterations through more-efficient parallel algorithms, and using optimization methods to curtail the number of iterations. More attention should be directed towards efficient scheduling of extensive computing tasks to reduce power consumption in large-scale computing facilities such as data centres. Governments should serve as gatekeepers of carbon neutrality, formulating and implementing policies to enhance communication and collaboration between the energy sector and the AI industry. Furthermore, there is a need to invest in the cultivation of comprehensive talent with backgrounds in computer architecture, AI and energy efficiency and sustainability.

Protecting the environment is a science-based decision and should not be negotiable (see Nature 624, 225; 2023). Yet, two days after Brazilian President Luiz Lula da Silva’s speech at the COP28 climate summit in Dubai, United Arab Emirates, on 1 December 2023, his government announced the paving of highway BR-319. This connects Manaus, in the heart of the Amazon rainforest, to Porto Velho, in the ‘deforestation arc’ on its southern frontiers. On 13 December, Brazil’s national petroleum agency sold off oil-drilling rights in 602 areas, 21 of them in the Amazon basin accessed through the BR-319 and secondary roads. A week later, the Brazilian Congress approved a bill deeming the road project essential and allocating money to it from the Amazon Fund, intended to protect the rainforest. Amazon Fund donors must exercise their veto to prevent this, or be held co-responsible for a project that will thwart climate goals, speed up biodiversity loss and open up vast forest blocks that hold an incalculable diversity of pathogens, such as viruses, bacteria and prions. The Brazilian government is undermining its own claims that it is protecting the Amazon. To underscore Brazil’s commitment to mitigating climate change and fostering global well-being, it must reverse these harmful decisions.

Theeb Ayedh Alkahtani Waraqa Publishing House, Riyadh, Saudi Arabia. [email protected]

Lei Guan National University of Defense Technology, Changsha, China. [email protected]

C. Guilherme Becker The Pennsylvania State University, University Park, Pennsylvania, USA.

Lucas Ferrante Federal University of Amazonas (UFAM), Manaus, Brazil. [email protected]

The European Commission is proposing to create soil ‘districts’ to improve the evaluation and monitoring of soil health in European Union member states so that all EU soils are in a healthy condition by 2050 (go.nature. com/3ru949t). European scientists are collaborating with land users and national environmental agencies to provide an operational definition of such districts so that they are maximally effective. These districts of homogeneous soils must be large enough to be managed by a single authority, for soilhealth assessment and for implementing policy. Given the diversity of soils in the EU, using homogeneity with respect to soil class or property could lead to an unmanageable number of districts. Spatial units should instead be homogeneous in terms of soil-landscape relationships, or ‘soilscapes’ (P. Lagacherie et al. Geoderma 101, 105–118; 2001). A district would then comprise a limited sequence of soil types amenable to similar management and policy. An effective soil district definition must also consider the national borders of member states and local EU administrative units, ensuring that the spatial contiguity of the districts is maintained. Alexandre M. J.-C. Wadoux National Research Institute for Agriculture, Food and Environment, Montpellier, France. [email protected] The author declares competing interests; see go.nature.com/3kb1fgx for details.

Nature | Vol 626 | 1 February 2024【 】 | 33

Expert insight into current research

KILIII YUYAN

News & views

Figure 1 | A sea otter (Enhydra lutris) eating a crab.

Ecology

Ecosystem effects of sea otters limit coastal erosion Johan S. Eklöf

Conservation is bringing back certain predators that are high in the food chain, but how this affects an ecosystem overall is debated. Rigorous fieldwork provides strong evidence that sea otters help to mitigate coastal erosion. See p.111 Whether food webs are regulated by resources in a bottom-up manner or by consumers in a top-down way is a long-standing debate1 that is relevant to an even broader fundamental question in ecology. Namely: to what extent are ecosystems influenced by interactions between organisms (such as predation) compared with the effect of environmental conditions? On page 111, Hughes et al.2 report data that provide insights into the fundamental

effects of a predator in the wild and highlight a system that benefits plants and their influences on coastal landscapes. Over the past three decades, the rapid recovery of some populations of large ‘top’ predators — those high in food webs  — after hunting bans, pollution abatement or reintroduction programmes, is helping ecologists to investigate the role of predators in ecosystems. However, there has been

considerable debate about how strong the effects of predators on an ecosystem really are. Perhaps the most well-known example of this is the reintroduction of grey wolves (Canis lupus) to Yellowstone National Park in the United States. There, subsequent increases in plant cover and riverbank stability have been put forward as a landscape-wide example of a ‘trophic cascade’ — an indirect effect observed when predators, by reducing the density or behaviour of their prey, enhance the survival and activity of organisms at the next, lower level of the food web3,4. In the Yellowstone case, wolf predation of elk (Cervus canadensis) was proposed to reduce elk grazing pressure, resulting in taller and more-dense plant communities that stabilized riverbanks by reducing soil erosion, thereby altering the landscape5. Yet the lack of controlled experiments needed to provide rigorous scientific evidence of this complex cascade effect has made it impossible to determine whether wolves or other factors caused the observed changes4,5. Hughes and colleagues report results gathered from another type of ecosystem, which provide strong evidence for the idea that the recovery of top-predator populations can Nature | Vol 626 | 1 February 2024【 】 | 35

News & views benefit plant communities and aid ecological processes regulated by such plants, including shoreline protection. The authors conducted their study in salt marshes at Elkhorn Slough, one of California’s largest remaining coastal wetlands. At this site, intense land development, excess nutrient input (eutrophication) and sea-level rise has caused coastal erosion, and more than 60% of the marsh area found in 1870 has either been lost or converted into other habitat types6. Over the past 40 years, the number of sea otters (Enhydra lutris) — a top predator that was once hunted to near extinction — has gradually increased in the area, from a few individuals in the 1980s to more than 100 identified animals by the late 2000s, as the authors note. Hughes and colleagues were inspired by previous findings from their team indicating considerable effects from sea otter recovery on food webs in nearby seagrass beds7. Sea otters (Fig. 1) need to consume an amount of food equivalent to more than 20% of their body mass per day in these cold estuarine waters8, and their diet includesthe commonplace striped shore crab Pachygrapsus crassipes. The authors hypothesized that, in tidal marsh creeks where otters had become abundant, their intense predation of these crabs should reduce crab burrowing and feeding on roots of the dominant marsh plant, pickleweed (Salicornia pacifica). This plant is an effective ‘ecosystem engineer’ that stabilizes shorelines. Therefore, sea otter recovery should have triggered a trophic cascade that mitigates salt-marsh erosion, similar to the proposed effect of wolves on the landscape in Yellowstone. To test their hypothesis, the authors combined four approaches, each of which could have been a study in its own right. First, Hughes and colleagues used time-series data partly extracted from aerial and satellite imagery from the 1930s to the present day. They combined these data with advanced statistical modelling to assess the influence of sea otter abundance on tidal creek widening (a measure of creek-bank erosion). The model output suggested that, despite a sustained increase in factors known to cause erosion of the shorelines (such as eutrophication or sea-level rise), marsh erosion instead abated alongside the recovery of the sea otters. The second, and in my view major, feat was to experimentally test the effect of otters on the ecosystem at this site. This was done by excluding otters from fenced plots measuring 1 × 2 metres and comparing these exclosures with unfenced controls in five tidal creeks over the course of an impressive timespan of three years. This type of field experiment is usually run for just a couple of months because of the regular need for maintenance and the risk of damage to the exclosures — a period that can be too short to capture effects that build up over time. 36 | Nature | Vol 626 | 1 February 2024 【】

The authors’ results indicate that sea otter predation strongly suppressed crab numbers and crab burrowing, which increased pickleweed root biomass and soil density; factors known to reduce the risk of erosion on creek banks. The authors also demonstrate that common side effects of exclosures, such as shading or the alteration of water flow, did not affect their results. Consequently, this proves that the otters have an effect on coastal plants and soil stability through a trophic cascade. For the other two approaches, the authors used field surveys covering both time (comparing the periods before and after the otter population increased) and space (across 13  creeks) to scale up their experimental results. This involved more than three years of daily observations of sea otter foraging and diet composition by trained observers. As predicted, otter-predation rates on crabs rose over time with increasing otter abundances, whereas marsh-creek erosion decreased. Compared with creeks that had the highest predation rates, creeks with the lowest measured predation rates had more than twice as many crabs, half the amount of plant-root biomass and three times faster marsh-erosion rates — data that again support the trophic-cascade hypothesis. Hughes and colleagues’ study is notable for at least three reasons. First, it experimentally confirms the theory that abundant top predators can strongly influence both ecosystem structure and processes. This adds to a large body of work showing that predation,

similar to factors such as nutrients and temperature, matters for ecosystem functioning9. Second, the powerful combination of methods used raises the bar on the evidence needed to support claims of strong effects of organisms on ecosystem functioning in the wild. Finally, the findings should intensify discussions on the role of conservation of large animals to help mitigate the environmental effects of stressors such as eutrophication and global warming10. This is especially important in times of rapid climate change and increasing calls to again limit coastal top-predator populations as a way to reduce conflicts between wildlife and fisheries11. Johan S. Eklöf is in the Department of Ecology, Environment and Plant Sciences, Stockholm University, SE-106 91 Stockholm, Sweden. e-mail: [email protected] 1. Hairston, N. G., Smith, F. E. & Slobodkin, L. B. Am. Nat. 94, 421–425 (1960). 2. Hughes, B. B. et al. Nature 626, 111–118 (2024【 】). 3. Eriksson, B. K., Bergström, U., Govers, L. L. & Eklöf, J. S. in Reference Module in Earth Systems and Environmental   https://doi.org/10.1016/ Sciences B978-0-323-90798-9.00006-8 (Elsevier, 2023). 4. Ripple, W. J., Larsen, E. J., Renkin, R. A. & Smith, D. W. Biol. Conserv. 102, 227–234 (2001).   R. L. & Ripple, W. J. Ecohydrology 12, e2048 5. Beschta, (2019). 6. Van Dyke, E. & Wasson, K. Estuaries 28, 173–189 (2005). 7. Hughes, B. B. et al. Proc. Natl Acad. Sci. USA 110, 15313–15318 (2013). 8. Costa, P. Physiol. Zool. 55, 35–44 (1982). 9. Koske, A. K. et al. Integr. Comp. Biol. 51, 644–646 (2011). 10. Malhi, Y. et al. Curr. Biol. 32, R181–R196 (2022). 11. Tixier, P. et al. Fish Fish. 22, 31–53 (2021). The author declares no competing interests.

Quantum information

Mobile atoms power up logical qubits Barbara M. Terhal

Small groups of mobile neutral atoms have been manipulated with extraordinary control to form ‘logical’ quantum bits. These qubits can perform quantum computations more reliably than can individual atoms. See p.58 Over the past 20 years, scientists have been developing ways of using neutral atoms for quantum computing1. On page 58, Bluvstein et al.2 demonstrate how far these methods have come: the authors’ efficient optical techniques enabled them to control tens to hundreds of atoms in parallel, maintaining the quantum state of the atoms, and allowing them to execute logical operations on an unprecedented scale. Bluvstein and colleagues’ quantum-

computing platform uses lasers to trap atoms in arrays that are hundreds of micrometres wide. Two of the possible energy levels of the electrons in each atom form a quantum bit (qubit). Before any computation can begin, a cloud containing millions of extremely cold atoms is loaded into the optical array, and atoms are removed and reshuffled until they are positioned in an organized grid. The authors first subdivide the grid into three zones (Fig. 1). One section is designated

a storage zone, in which each qubit could hold its state for at least one second, or in which states of single qubits could be altered selectively or collectively. The second is an interacting zone, in which atoms could interact with their neighbours and become ‘entangled’, meaning that the state of one qubit depends on that of its neighbour. Finally, the authors specify a read-out zone, in which the state of all qubits could be measured in parallel in less than a millisecond. What is crucial and impressive about this approach is that the atoms can be shifted from one zone to the other with exquisite control: a whole row of atoms can be quickly moved sideways, and then upwards or downwards, essentially without any change to the qubits’ states. The platform is also extremely efficient at executing basic qubit operations simultaneously. For example, a single laser pulse lasting just 300 nanoseconds, and illuminating all the atoms in the entangling zone, can be used to make each qubit interact with its neighbour — and not with all the other atoms in the zone. Any quantum-computing platform is ultimately useless if operations on qubits cannot be performed with high accuracy and low error. Bluvstein et al. showed that basic operations, such as the one that entangles two neighbouring qubits, can be executed with an error rate of 0.5% or less. This rate is on a par with numbers achieved for other qubit platforms, such as those using superconducting qubits and trapped ions. However, error rates certainly need to be reduced much more to realize a truly ‘fault-tolerant’ quantum computer. One process for minimizing these error rates is known as quantum error correction, and the idea is to use redundancy: a set of many physical qubits (in this case, atoms) is used to make another type of quantum bit, known as a logical qubit. The error rates of logical qubits can be impressively lower than those of the underlying physical qubits because the physical qubits are repeatedly monitored for errors, and the errors that are found can be corrected. And this is where the neutral-atom platform shines: as Bluvstein et al. showed, neutral atoms can be controlled in an efficient way to perform computation on logical qubits on a larger scale than has been achieved with other platforms. Quantum error correction comes at a price, because the redundancy implies an overhead in the number of qubits required, and fault -tolerant operations on logical qubits can incur a further cost in terms of the time spent monitoring errors. One approach to error correction is known as the surface code, which involves representing a logical qubit by physical qubits that are arranged in a 2D array. This method has been implemented using superconducting qubits3, but such qubits are typically not mobile, which exacerbates the cost associated with needing many qubits.

Storage zone

Trapped atom

Logical qubit

Interacting zone

Entangled qubit

Read-out zone

Entangling laser beam

Imaging laser beam

Figure 1 | An efficient way of using neutral atoms for quantum logic operations.  Bluvstein et al.2 devised a method for simultaneously manipulating atoms that form quantum bits (qubits), and are grouped into structures called logical qubits (red and yellow), with which the authors performed a quantum computation. The atoms were trapped using lasers (not shown) in a grid that was subdivided into three zones — a storage zone, an interacting zone and a read-out zone — and the atoms could be rapidly shifted between zones. A laser beam illuminating the entire interacting zone was used to perform parallel operations that ‘entangled’ each atom with its neighbour, such that the state of one qubit depended on that of the other. A second laser was used to image atoms that had been moved to the read-out zone, providing a quantum measurement that determined each qubit’s state. (Adapted from Fig. 1a of ref. 1.)

Bluvstein and colleagues’ ability to move their qubits quickly and easily offers the opportunity to go beyond the limitations of 2D connectivity. Indeed, the authors performed logical operations for various small quantum-error-correcting codes in an extremely efficient manner. Whereas previous experiments4–7 on superconducting or trapped-ion qubits involved operations on one to three logical qubits, Bluvstein et al. played, for example, with 40 logical qubits, each comprising 7 atoms. They also showed that a single logical operation was more reliably executed using a surface code involving a large array than when using a smaller array, demonstrating that, when it comes to quantum error correction, bigger is better. The authors then showed that hundreds of two- and three-qubit logical operations could be executed using 48 logical qubits, each of which was represented by 8 atoms. To execute these logical operations efficiently, it was essential for Bluvstein et al. to use patterns of interactions between individual atoms that go beyond nearest-neighbour interactions in two dimensions, hence requiring mobile atoms. The goal of the quantum computation was to demonstrate quantum advantage: the time it takes a quantum computer to perform a quantum computation increases slowly with the number of qubits, whereas the time taken for a classical computer to carry out an equivalent computation explodes with the number of qubits3. The computation executed with the logical qubits achieved an accuracy that surpassed anything

attainable without quantum error correction. One disadvantage of working with neutral atoms is that a quantum measurement is destructive: in future experiments, atoms will need to be continuously resupplied throughout the course of lengthy error-monitoring measurements that stabilize a logical qubit. By contrast, superconducting qubits are measured non-destructively and can be used again immediately. For this reason, Bluvstein and colleagues’ results fall short of showing that logical qubits comprising more atoms can be sustained for longer than can those with fewer atoms, which has been achieved with superconducting qubits4, albeit for a smaller surface code than the one Bluvstein et al. studied. Another challenge is that qubits can ‘leak’ to other electronic states during two-qubit operations, although ideas for how to convert such qubit leakage into benefits for quantum error correction have already been proposed8. What is exciting is that there is still plenty of room for growth. Bluvstein et al. are already looking ahead to future experiments comprising 10,000 atoms, further reducing measurement and operation times, saving on laser power and enabling atoms to be loaded into the grid continuously. Implementations of quantum error correction could also be improved: newly developed codes9,10 with higher efficiency and thus a smaller qubit overhead could further enhance quantum computing with Bluvstein and colleagues’ impressive platform. Nature | Vol 626 | 1 February 2024【 】 | 37

News & views Barbara M. Terhal is in the Delft Institute of Applied Mathematics, Delft University of Technology, 2628 CD Delft, the Netherlands. e-mail: [email protected] 1. 2. 3. 4. 5.

Jaksch, D. et al. Phys. Rev. Lett. 85, 2208 (2000). Bluvstein, D. et al. Nature 626, 58–65 (2024【 】). Arute, F. et al. Nature 574, 505–510 (2019). Google Quantum AI. Nature 614, 676–681 (2023). Postler, L. et al. Nature 605, 675–680 (2022).

6. Honciuc Menendez, D., Ray, A. & Vasmer, M. Preprint at https://arxiv.org/abs/2309.08663 (2023). 7. Wang, Y. et al. Preprint at https://arxiv.org/ abs/2309.09893 (2023). 8. Wu, Y., Kolkowitz, S., Puri, S. & Thompson, J. D. Nature Commun. 13, 4657 (2022). 9. Xu, Q. et al. Preprint at https://arxiv.org/abs/2308.08648 (2023). 10. Bravyi, S. et al. Preprint at https://arxiv.org/ abs/2308.07915 (2023). The author declares no competing interests.

Engineering

Flexible fibres take fabrics into the information age Xiaoting Jia & Alex Parrott

A technique for embedding fibres with semiconductor devices produces defect-free strands that are hundreds of metres long. Garments woven with these threads offer a tantalizing glimpse of the wearable electronics of the future. See p.72 Imagine a washable hat that can help a blind person to sense changes in traffic lights, or a dress that can act as a tour guide as its wearer moves through a museum. These technologies can be realized using smart flexible fibres equipped with semiconductor devices that detect and process signals, and the performance of such fibres has advanced rapidly over the past few years1–8. However, existing fabrication methods can produce threads with fractured, defective semiconductor cores. On page 72, Wang et al.9 report an innovative approach in which tiny semiconductor components are fed into a fibre-pulling machine, resulting in continuous high-performance flexible fibres that can sense, communicate and interact with each other. Humans have been using fibres since the Stone Age10, but attempts to expand their functionality beyond simple thermal insulation are much more modern. One of the challenges associated with incorporating semiconductor

Glass cladding

Heating element

devices into fibres involves wearability: the fibres must be flexible and twistable, so that they can be woven; washable, for repeated use; and breathable, so that they can be worn ­comfortably. Many researchers have prioritized these features by creating smart fibre devices made from amorphous semiconductor materials. However, incorporating conventional silicon-based semiconductors is expected to lead to superior electronic ­properties and performances2. An alternative approach involves generating a glass-clad fibre with a silicon core, but this technique, known as the molten-core method, often leads to fibres that fracture ­easily and contain defects. Wang et al. overcame this problem by first performing a detailed mechanical analysis of the fabrication process to identify the sources of stress that induce fracture. The molten-core method begins with a semiconductor wire made of silicon or germanium being inserted inside a

Wire fed into a polymer tube

Polymer cladding

glass tube (Fig. 1). Both materials are heated to at least 1,000 °C until they are soft enough to be pulled into a thin strand, which is then cooled. Wang et al. identified stress formation in two stages: the point at which the core solidifies; and the subsequent phase, in which the fibre is cooled. Specifically, they found that mismatches in the viscous behaviour of glass and the melting point of the wire induced stresses in the core, as did differences in the thermal-expansion rates of the materials. The authors showed that both problems could be alleviated by choosing the right combination of materials: silicon cores worked well with cladding made from ultra-tough silica glass, whereas germanium cores performed better when clad in aluminosilicate glass. The resulting fibres are fracture-free and of high quality, but the glass cladding compromises their applicability. For this reason, the cooled fibres are immersed in hydro­fluoric acid to remove the outer layer of glass, leaving behind the semiconductor core. This core is then fed into a polymer tube together with electrically conductive wires. The whole structure is heated again, and pulled into a flexible fibre that can be hundreds of metres long. The resulting fibre takes the form of a p–n junction, which is the basic building block of modern electronics and can convert light signals into electrical currents. The process is reminiscent of making spun sugar — if the sugar were embedded with tiny, controllable, electronic components. Wang et al. showed that their fibres could be used to make several devices. In one example, the authors knitted fibres into a hat that could be used to sense signals from traffic lights. The signal received by the hat was sent to a mobile phone, which then buzzed when the lights turned red or green. In another case, the fibre was woven into a sweater that served as a light-fidelity (Li-Fi) device, a technology that transmits data at light frequencies instead of the radio frequencies used by wireless networks such as 5G. The sweater detected image signals that were encoded as light pulses, and then a second device decoded these signals to

Metal wire

Semiconductor Fibre heated and pulled

Hydrofluoric acid removes glass

Figure 1 | A method for fabricating digitized fibres. Wang et al.9 devised a way of embedding fibres with semiconductor materials to create long, flexible threads. A semiconductor wire is inserted inside a glass tube and both materials are heated until they are soft enough to be pulled into a thin strand, which is

38 | Nature | Vol 626 | 1 February 2024【 】

Fibre heated and pulled again

Embedded thread

then cooled. Hydrofluoric acid is used to remove the glass. The wire is then fed into a polymer tube together with metal wires. The structure is heated again, and pulled into a flexible thread that can be hundreds of metres long, and can be woven into fabrics that detect and process signals.

reconstruct the image. The authors also wove their smart fibres into a flexible wristband that outperforms similar devices for heart-rate monitoring. The devices that are currently available ­typically use a rigid sensor that doesn’t flex to the shape of the wrist, and can therefore produce ­inaccurate measurements. The performance of Wang and colleagues’ fibres is on a par with these commercial silicon devices, but they can also withstand high compression, such as that experienced at an underwater depth of 3,000 metres. The authors showed that their wristband could be used to detect visible light around a submarine. Another key advantage of this technology is its industrial readiness. The instrument that fabricates the fibres includes a fibre-­drawing device that is used to produce commercial optical fibres in the telecommunication industry. And once the fibres are generated, they can be knitted or woven into fabric using tools that are already used widely in the textile industry.

Wang and colleagues’ work takes a leap towards embedding micro-computers into everyday clothing. An exciting future direction would be to equip the fibres with more-complex devices, such as transistors, and to increase the density of these functional components. One limitation of the current approach is that it requires a post-processing step to incorporate exceptionally high-quality (single-crystal) semiconductors into the fibre. Finding a way of embedding these materials during fabrication would broaden the scope of the fibres’ electronic and optoelectronic applications. Finally, because the wires embedded in Wang and colleagues’ fibres are easily ­connected to existing computer hardware, this technology could prove useful in efforts to develop integrated human–machine systems. The work therefore allows us to imagine a generation of smart fibres and fabrics that enable individuals to engage seamlessly with their surroundings — and make their everyday experiences fully immersive.

Xiaoting Jia is in the Bradley Department of Electrical and Computer Engineering, the Department of Materials Science and Engineering, and the School of Neuroscience, Virginia Tech, Blacksburg, Virginia 24061, USA. Alex Parrott is in the Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, Virginia 24061, USA. e-mail: [email protected]

1. Qian, S., Liu, M., Dou, Y., Fink, Y. Yan, W. Natl Sci. Rev. 10, nwac202 (2023). 2. Abouraddy, A. et al. Nature Mater. 6, 336–347 (2007). 3. Shi, X. et al. Nature 591, 240–245 (2021). 4. Fang, B. et al. Nature Commun. 13, 2101 (2022). 5. Healy, N. et al. Nature Mater. 13, 1122–1127 (2014). 6. Ma, W. et al. Chem. Soc. Rev. 50, 7009–7061 (2021). 7. He, J. et al. Nature 597, 57–63 (2021). 8. Ballato, J. et al. Opt. Express 16, 18675–18683 (2008). 9. Wang, Z. et al. Nature 626, 72–78 (2024【 】). 10. Kvavadze, E. et al. Science 325, 1359 (2009). The authors declare no competing interests.

NICOLA MAYRHOFER

Bacterial prey turns the tables on predator The bacterium Myxococcus xanthus is a voracious predator of other microorganisms. M. xanthus can form fruiting bodies (black dots, pictured) and swarm through colonies of prey (right-hand circle) by forming ripplelike travelling waves. However, writing in PLoS Biology, Vasse et al. report that the prey bacterium Pseudomonas fluorescens can slaughter M. xanthus to extinction under certain conditions (M. Vasse et al. PLoS Biol. 22, e3002454; 2024【 】【淘宝唯一店铺：艾 米学社】【淘宝唯一 店铺：艾米学社】). Inspired by earlier observations, Vasse et al. investigated how the growth temperature of P. fluorescens affects its risk of predation. When this bacterium was grown in culture at 32 °C, the authors observed that it was largely killed by M. xanthus, as expected. But remarkably, when cultured at 22 °C, P. fluorescens became the predator — killing M. xanthus and growing on its remains. This is a rare example of the reversal of a microbial predator–prey relationship triggered by a specific abiotic environmental factor — in this case, temperature. Intriguingly, the reversal depends on the temperature experienced before, rather than during, the predator–prey interaction. The authors speculate that most microbes might engage in predation to some extent. Andrew Mitchinson

Nature | Vol 626 | 1 February 2024【 】 | 39

News & views Structural biology

Snapshots of genetic ­copy-and-paste machinery Gael Cristofari

LINE-1 DNA elements self-duplicate, inserting the copy into new regions of the genome — a key process in chromosome evolution. Structures of the machinery that performs this process in humans are now reported. See p.186 & p.194 DNA sequences known as LINE-1 (L1) elements replicate themselves and spread throughout mammalian genomes using a biological ‘copy and paste’ mechanism. On pages 186 and 194, respectively, Thawani et al.1 and Baldwin et al.2 report X-ray and cryoelectron-microscopy structures of the remarkable complex that enables L1 to propagate itself — providing insight into how and where this DNA element is duplicated. Enzymes known as reverse transcriptases (RTs) enable retroviruses, such as HIV, to complete their replication cycle. These enzymes convert the viral RNA genome into DNA, which is then integrated into the chromosomes of the infected host. The discovery of RTs in 1970 revolutionized molecular biology by revealing the existence of a reverse flow of genetic information (RNA to DNA), challenging the dogma that only the forward flow — DNA is transcribed into RNA, which is translated into proteins — can occur. Nearly two decades later, some cases of haemophilia A were attributed to a mutation caused by the insertion of an L1 DNA fragment into a gene that encodes a blood coagulation factor3. Given that many copies of L1 are present in human chromosomes, and that this fragment encodes an enzyme distantly related to retrovirus RTs, it was this discovery that suggested that L1s actively replicate in humans and spread themselves throughout the genome using a copy-and-paste mechanism. We now know that L1s not only cause genetic diseases, but are also involved in many cancers, in ageing and probably in several neurodegenerative diseases3. L1 sequences constitute one of the many families of DNA elements known as retrotransposons, which have invaded the genomes of all eukaryotes (organisms that include plants, animals and fungi). The L1 family has been remarkably successful throughout evolution, with the activity of these sequences accounting for roughly one-third of human chromosomes; by contrast, only about 1% of human chromosomes consist of DNA that encodes

other cellular proteins3. The copy-and-paste machinery for L1 is a ribonucleoprotein particle (RNP) — a complex that contains the L1 RNA (known as the template) and two proteins, named ORF1p and ORF2p. ORF1p binds to the template and probably aids the assembly or proper folding of the complex, whereas ORF2p acts as both an RT and an endonuclease (an enzyme that cuts DNA)3. ORF2p cuts one of the strands of chromosomal DNA in the cell nucleus, and initiates reverse transcription of the template, starting from the cut3 (Fig. 1a). The L1 RNP preferentially targets a short DNA sequence that is found in all types of genomic region4,5. The overall process is known as target-primed reverse transcription (TPRT), and differs in various respects from retroviral reverse transcription (which generally occurs in the cytoplasm)6. Phylogenetic studies indicate that the ­central catalytic domain of ORF2p is distantly a

Genomic DNA Cut

b

Template

c Replicating genomic DNA

ORF2p

Copy of L1

Figure 1 | Enzymatic activities mediated by LINE-1 elements.  a, DNA sequences known as LINE-1 (L1) elements, found in mammalian genomes, encode enzymes that enable a copy of the element to be inserted at a new position in the genome. One of these enzymes (ORF2p) forms a complex with the L1 transcript (the template) and binds to a target site in the genomic DNA, cutting one of the DNA strands. It then makes a copy of L1 (blue arrow) starting from the cut, using the sequence encoded in the template (a process known as target-primed reverse transcription). b, Thawani et al.1 report that, in vitro, the single-strand cleavage is enhanced close to junctions between single- and double-stranded DNA, which are commonly found at Core RTORF2p domaincan initiate reverse transcription in vitro from that sitesaof DNA replication. c, Baldwin et al.2 observe ‘hairpin’ templates that fold back — possibly how ORF2p Tower Fingers EN on themselves Thumb Wrist CTD synthesizes DNA in the Palm explaining cytoplasm, far from genomic DNA in the nucleus. ORF2p ORFp

40 | Nature | Vol 626 | 1 February 2024【 】

related to retroviral RTs. The structures reported by Baldwin et al. and Thawani et al. now confirm that this central domain adopts the hand-shaped structure typical of the retro­ viral enzymes, comprising fingers, palm and thumb subdomains (Fig. 2). However, L1 ORF2p also harbours amino- and carboxy-­terminus regions that extend the central domain. At the N-terminus, the RT domain is connected to the endonuclease domain by a flexible linker that Baldwin et al. call the tower, which extends and partially covers the fingers subdomain. At the C-terminus, the thumb is prolonged by a ‘wrist’ that contacts the template bound to the cut end of the chromosomal DNA. ORF2p ends with a C-terminal domain (CTD; also known as a C-terminal segment) that harbours a structural motif called a zinc finger (often found in proteins that bind to DNA or RNA). The function of the CTD has long been mysterious, but the new structures show that it makes contact with the template at a position distant from the RT active site, and might help to unwind the RNA, preventing it from becoming tangled and obstructing DNA synthesis. The flexibility of the tower allows the endonuclease to rotate relative to the CTD, thus switching the structure from an ‘open’ to a ‘closed’ form, in which the template is clamped2. Our understanding of TPRT stems mainly from studies of R2, a retrotransposon found in silkworms (Bombyx mori) that is related to L1, but is easier to purify and study7. The structural details of R2-mediated TPRT have been reported in the past year8,9. The two new structures of the L1 machinery therefore e ­ nable a comparison of how the key enzymes of L1 and R2 are used for TPRT. This reveals some ­notable differences.

DBD

Tower-like domain

First, the enzyme encoded by R2 (ORFp) has an N-terminal DNA-binding domain that guides integration of R2 DNA into one specific genomic site. By contrast, ORF2p inserts L1 DNA into a short DNA motif that is broadly distributed throughout the genome, and lacks the DNA-binding domain found in ORFp4,5. Second, ORFp harbours a ‘tower-like’ domain (NTE-1) that recruits and positions the R2 RNA, ready for reverse transcription, through binding of a structured region found only in that RNA. ORF2p, by comparison, associates with the tail of the RNA template10; this tail contains a sequence that is found in many RNA molecules. And third, the endonuclease domains of ORF2p and ORFp are at the N-terminal and C-terminal ends, respectively, of these proteins — a striking difference in 3D organization. Similarly to ORF2p, ORFp has a CTD-like domain with a zinc finger downstream of the RT. However, this zinc finger ‘unzips’ the target double-stranded DNA, allowing only one of the two strands to access the endonuclease active site, rather than unwinding entangled regions of the template, as it does in ORF2p1. Previous studies involving the in vitro reconstitution of the L1 RNP with model templates and DNA substrates have also offered mechanistic insights into the activity of ORF2p. The two new studies1,2 confirm that ORF2p is highly processive11,12 (capable of copying long RNA templates into DNA in a single step). They also confirm that reverse transcription by ORF2p requires a minimal base-pairing of four to six nucleotides between the DNA substrate and RNA template11, but can also tolerate some level of mispairing11,13. Thawani et al. successfully reconstituted TPRT reactions in vitro and demonstrated that the initial cleavage of a single strand of the DNA substrate is enhanced when the L1 target-site sequence is close to a junction between ­single- and double-stranded DNA (Fig. 1b). Such junctions are commonly found at structures called replication forks, which form during replication of the host DNA. This finding corroborates previous proposals that replication forks are preferred sites for retrotransposon integration4,5, and suggests that cleavage of the second strand is not essential to complete integration. Baldwin et al. found that ORF2p can initiate reverse transcription directly from very short single-stranded DNA or RNA molecules, or from ‘hairpin’ RNA substrates that fold back on themselves (Fig. 1c). These findings suggest a possible explanation for how L1 DNA is formed in the cytoplasm — something that can lead to the activation of inflammatory pathways during cellular senescence (a condition linked to ageing in which cells cease to proliferate) and in certain inflammatory diseases14,15. Consistent with this mechanism, RT activity — but not endonuclease activity — is essential for

a

Core RT domain EN ORF2p

Tower Fingers Palm Thumb Wrist

CTD

ORFp DBD

Tower-like domain

b EN active site

EN Fingers

CTD Thumb

Template RNA

RT active site

Tower-like domain

DBD

Host DNA

Tower

Palm Wrist

Figure 2 | Comparison of the ORF2p protein with the ORFp protein.  a, The ORFp protein, produced by silkworms (Bombyx mori), is related to ORF2p. Comparison of the amino-acid sequences of the two proteins shows that they have a similar ‘core’ (the reverse transcriptase (RT) domain) consisting of subdomains known as the fingers, palm and thumb, next to a wrist domain. ORF2p contains a ‘tower’ domain2 that connects the core to the endonuclease domain (EN, used to cleave one strand of target DNA), whereas ORFp has a shorter, tower-like domain. ORF2p lacks the DNA-binding domain (DBD) of ORFp, and its EN domain is found at the amino-terminus of the protein, rather than in the carboxy-terminus domain (CTD). b, New structures1,2 of the human ORF2p protein show that its 3D structure (left) differs substantially from that of ORFp8,9 (right). Active sites of the EN and RT domains are shown, where visible (the EN active site cannot be seen in this view of ORFp). The 3D structures are drawn from Protein Data Bank accessions 8GH6 (ref. 9) and 8UW3 (ref. 1).

the accumulation of these cytoplasmic DNA molecules2, ruling out the possibility that they originate from abortive TPRT products that escaped the nucleus. Overall, it seems that ORF2p potentially acts on a variety of DNA substrates beyond those in the conventional TPRT model. It remains to be seen whether replication forks — or other structures containing junctions of duplex DNA with single strands — are the main targets for L1 DNA insertion. In vitro experiments that test more DNA substrates, or that incorporate

“Overall, it seems that the ORF2p protein potentially acts on a variety of DNA substrates.” ORF1p and other factors known to intervene in L1 replication16,17, will help to refine our understanding of this retrotransposon’s activity. The discovery of RTs paved the way for methods now used to produce proteins for research and medical applications, including insulin, growth hormone and the hepatitis B vaccine. It also enabled the development of the RT-PCR and RNA-sequencing techniques, which are used to detect RNA-virus infections and to measure gene expression. Expanding the molecular-biology toolbox to include RTs that are highly processive (or exhibit other ­previously unavailable biochemical

properties) might benefit technologies used for genomics — for instance, by enabling the sequencing of full-length RNA molecules18. Elucidating the substrates and mechanisms of TPRT might also aid the design of genome-­ engineering tools9. Finally, RT inhibitors were among the first medicines for AIDS, and are still key to therapy regimens that have transformed this devastating disease into a manageable ­ chronic condition. Baldwin et al. report that some drugs designed to target HIV also inhibit ORF2p (albeit with moderate binding affinities), and modelled the mode of ORF2p inhibition in light of the structure they had determined. The new ORF2p structures will enable the design of more-specific inhibitors that target both the RT and the endonuclease activities of ORF2p; such inhibitors might be useful for cancer treatment and research into ageing19,20. Gael Cristofari is at the University Cote d’Azur, Inserm, CNRS, Institute for Research on Cancer and Aging of Nice (IRCAN), Nice 06107, France. e-mail: [email protected] 1. Thawani, A., Florez Ariza, A. J., Nogales, E. & Collins, K. Nature 626, 186–193 (2024【 】). 2. Baldwin, E. T. et al. Nature 626, 194–206 (2024【淘宝唯一 店铺：艾米学社】). 3. Kazazian, H. H. & Moran, J. V. N. Engl. J. Med. 377, 361–370 (2017). 4. Sultana, T. et al. Mol. Cell 74, 555–570 (2019). 5. Flasch, D. A. et al. Cell 177, 837–851 (2019). 6. Sultana, T., Zamborlini, A., Cristofari, G. & Lesage, P.

Nature | Vol 626 | 1 February 2024 | 41

News & views Nature Rev. Genet. 18, 292–308 (2017). 7. Luan, D. D., Korman, M. H., Jakubczak, J. L. & Eickbush, T. H. Cell 72, 595–605 (1993). 8. Deng, P. et al. Cell 186, 2865–2879 (2023). 9. Wilkinson, M. E., Frangieh, C. J., Macrae, R. K. & Zhang, F. Science 380, 301–308 (2023). 10. Doucet, A. J., Wilusz, J. E., Miyoshi, T., Liu, Y. & Moran, J. V. Mol. Cell 60, 728–741 (2015). 11. Monot, C. et al. PLoS Genet. 9, e1003499 (2013). 12. Piskareva, O. & Schmatchenko, V. FEBS Lett. 580, 661–668 (2006). 13. Kulpa, D. A. & Moran, J. V. Nature Struct. Mol. Biol. 13,

655–660 (2006). 14. De Cecco, M. et al. Nature 566, 73–78 (2019). 15. Thomas, C. A. et al. Cell Stem Cell 21, 319–331 (2017). 16. Taylor, M. S. et al. Cell 155, 1034–1048 (2013). 17. Benitez-Guijarro, M. et al. EMBO J. 37, e98506 (2018). 18. Zhao, C., Liu, F. & Pyle, A. M. RNA 24, 183–195 (2018). 19. Brochard, T. et al. Ageing Res. Rev. 92, 102132 (2023). 20. Rajurkar, M. et al. Cancer Discov. 12, 1462–1481 (2022). The author declares competing interests; see go.nature. com/3sfap2x for details. This article was published online on 15 January 2024 【 】.

Public health

Contact-tracing app predicts transmission risk Justus Benzler

The risk of catching COVID-19 as calculated by a smartphone app scales with the probability of subsequently testing positive for the coronavirus SARS-CoV-2, showing that digital contact tracing is a useful tool for fighting future pandemics. See p.145

42 | Nature | Vol 626 | 1 February 2024 【】

systems provided by Google and Apple, known as the Exposure Notification framework6. This feature relies on Bluetooth signals that

Living in the same household

0.2 Probability of reported infection

The COVID-19 pandemic was the first global outbreak in human history to unfold in a world where many countries had more than 70% smartphone coverage1. This generated many initiatives for smartphone applications that could complement or replace established measures to control the spread of infection in pandemics. For instance, apps were developed that could deliver test results, provide proof of vaccination2 or trace the recent contacts of an infected person. On page 145, Ferretti et al.3 report that data from a smartphone app used for contact tracing can provide valuable epidemiological information. Contact tracing is a well-established process that public-health authorities follow during outbreaks of diseases that are transmitted directly between humans. The aim is to find people who were in contact with infected individuals, so that those with potential exposures can receive recommendations or interventions, such as quarantine, to prevent further disease transmission. Manual contact tracing is particularly resource-intensive and is not easily scalable4. Hence, in a pandemic, it quickly reaches its limits. Digital contact tracing is an alternative solution that relies on data gathered by personal mobile devices. However, it can pose a particular threat to privacy, because it involves collecting sensitive information about an individual’s health status and relationships5. Various approaches to digital contact tracing were debated at the onset of the COVID-19 pandemic. Ultimately, the public-health authorities in many countries chose to base their contact-tracing apps on an integrated feature of smartphone operating

Brief encounter (15 minutes at 2-metre distance)

0.1

0

1

Spending the day together

10 100 Risk score calculated by the app (log scale)

1,000

Figure 1 | A smartphone app used for contact tracing during the COVID-19 pandemic can predict the probability of SARS-CoV-2 transmission.  Using anonymized data from the NHS COVID-19 app for England and Wales, Ferretti et al.3 show that a risk score for infection — calculated in the app on the basis of the amount of time spent with and proximity to an infected person, and how infectious that person is — scales with the probability of an infection subsequently being reported. A high risk score might result from living in the same household as an infectious person. A brief encounter at a distance of 2 metres for 15 minutes, the threshold for a ‘relevant’ contact defined in manual contact tracing during the COVID-19 pandemic in most countries, results in a low risk score. In future pandemics, harnessing data from contact-tracing apps might help public-health authorities to understand how infections spread.

are exchanged between participating smartphones when they are physically close to each other. The signals transmit unique, randomly generated codes that are then temporarily stored locally on the other phone. If a smartphone user tests positive for SARSCoV-2, they can opt to upload a set of codes to a server. These codes are renewed daily, and are used to generate the codes that are sent over Bluetooth. The smartphones of other users can then compare the codes on the server to the ones that they had received and stored locally. In the case of a match, the app notifies the second user about past encounters with a potentially infectious person. The threshold for a notification is based on a risk score calculated by the app from the estimated proximity and duration of these exposures, and the infectiousness level of the notifier. This is estimated from the date of the encounter in relation to the notifier’s test and the onset of their symptoms. There is ongoing controversy7 over the extent to which digital contact tracing — and other non-pharmaceutical interventions — actually contributed to slowing the spread of infections. The aim was to prevent health-care systems being overwhelmed, and to buy time for the development, production and delivery of vaccines. There was little information collected that could address this controversy, mainly because of privacy concerns and the decision to use a decentralized architecture for digital contact-tracing systems, which meant that contact data collected by the apps were not stored in centralized databases. Furthermore, in many countries it was not clear how widely the apps were adopted, because public-health authorities often used downloads as a (poor) proxy for an app’s use (see go.nature.com/42q6axc). For the same reasons, the systems also scarcely reached their potential for monitoring key epidemiological indicators for the spread of COVID-19 in the population. A notable exception is the NHS COVID-19 app rolled out by the National Health Service in England and Wales, which was pioneered by a strong partnership between app developers and academic institutions. Early in the pandemic, the team behind the app created a tool8 to model the impact of non-pharmaceutical interventions, including digital contact tracing, and published empirical evidence showing that the app helped to prevent COVID-19 cases and COVID-related hospitalizations and deaths9. In the current study3, researchers in the same team analysed data recorded by the app to answer a fundamental question that arose during the COVID-19 pandemic: how is the probability of SARS-CoV-2 transmission from one individual to another related to the proximity and duration of the exposure (Fig. 1)? Ferretti and colleagues analysed ‘packets’

of data that each active instance of an app automatically and anonymously sent to a central server. Data packets were sent daily, and on two further occasions: when the app notified its user of an exposure, and when a user submitted a positive COVID-19 test result through the app. From common data points in these packets, the researchers were able to match exposures to actual SARS-CoV-2 infections. Not all infections resulted from exposures that would have been recorded by the app. By modelling this background ‘noise’ of infections and removing them, the authors were able to estimate the probability of a notified exposure resulting in a reported transmission. They also further disentangled the risk score into its components, which enabled them to separately analyse the contribution of the instantaneous level of risk (the risk regardless of duration) and the contribution of the duration of exposure. Their main findings are that a clear dose–response relationship exists between the risk score and the probability of reported transmission, and that duration matters even more than proximity. They also confirm that the measurements made by smartphones and the calculations made by contact-tracing apps, despite their limitations, are valid predictors of transmission probability.

Digital contact tracing thus has its place in the toolkit of non-pharmaceutical interventions for future pandemics, and should be part of pandemic preparedness plans. Proximity estimation will probably be improved as smartphones move to using other types of radio technology for signalling, such as ultra-wideband, which enables the distances between devices to be measured more accurately than does Bluetooth. Future smartphones might also be able to take into account

“Digital contact tracing has its place in the toolkit of non-pharmaceutical interventions.” other factors that affect the probability of disease transmission, such as being indoors or outdoors. Furthermore, strategies need to be developed that allow epidemiologically relevant data to be collected while preserving privacy. Such strategies should be discussed with the general public to achieve wide acceptance before the next pandemic, when policy decisions will again need to be made urgently. In future pandemics, analyses such as those presented by Ferretti and colleagues should

ideally happen continuously and at the same time as the data are generated. This would enable health authorities to monitor a dynamic pandemic situation with appropriate spatial resolution, and to fine-tune non-pharmaceutical interventions to control the spread of disease. Justus Benzler is in the Department for Infectious Disease Epidemiology, Robert Koch Institute, Nordufer 20, 13353 Berlin, Germany. e-mail: [email protected]

1. 2. 3. 4.

5. 6. 7. 8. 9.

https://www.statista.com/statistics/539395 Cascini, F. et al. Front. Public Health 9, 744356 (2021). Ferretti, L. et al. Nature 626, 145–150 (2024【 】). European Centre for Disease Prevention and Control. Contact Tracing for COVID-19: Current Evidence, Options for Scale-up and an Assessment of Resources Needed (ECDC, 2020). Bradford, L., Aboy, M. & Liddell, K. J. Law Biosci. 7, lsaa034 (2020). https://www.macrumors.com/guide/exposurenotification https://algorithmwatch.org/en/analysis-digital-contacttracing-apps-2021 Hinch, R. et al. PLoS Comput. Biol. 17, e1009146 (2021). Wymant, C. et al. Nature 594, 408–412 (2021).

The author declares competing interests; see go.nature.com/46zpftb for details. This article was published online on 21 December 2023.

New to the Nature Portfolio

Nature Cities launched in January 2024【 】 and aims to deepen andintegrate basic and applied understanding of the character and dynamics of cities, including their roles, impacts and influences — past, present and future. We encourage research and views from and about the Global South and other geographies that have traditionally been marginalized from urban scholarship. 07 Endorsement

The journal is open for submissions and the full aims and scope can be found on our website. 07 Endorsement

Learn more about the journal nature.com/natcities @NatCities 031AJ

Designing a circular carbon and plastics economy for a sustainable future https://doi.org/10.1038/s41586-023-06939-z Received: 2 November 2021 Accepted: 5 December 2023 Published online: 31 January 2024【 】 Check for updates

Fernando Vidal1,2, Eva R. van der Marel3,4, Ryan W. F. Kerr1, Caitlin McElroy5, Nadia Schroeder5, Celia Mitchell5, Gloria Rosetto1, Thomas T. D. Chen1, Richard M. Bailey6, Cameron Hepburn5 ✉, Catherine Redgwell3 ✉ & Charlotte K. Williams1 ✉

The linear production and consumption of plastics today is unsustainable. It creates large amounts of unnecessary and mismanaged waste, pollution and carbon dioxide emissions, undermining global climate targets and the Sustainable Development Goals. This Perspective provides an integrated technological, economic and legal view on how to deliver a circular carbon and plastics economy that minimizes carbon dioxide emissions. Different pathways that maximize recirculation of carbon (dioxide) between plastics waste and feedstocks are outlined, including mechanical, chemical and biological recycling, and those involving the use of biomass and carbon dioxide. Four future scenarios are described, only one of which achieves sufficient greenhouse gas savings in line with global climate targets. Such a bold system change requires 50% reduction in future plastic demand, complete phase-out of fossil-derived plastics, 95% recycling rates of retrievable plastics and use of renewable energy. It is hard to overstate the challenge of achieving this goal. We therefore present a roadmap outlining the scale and timing of the economic and legal interventions that could possibly support this. Assessing the service lifespan and recoverability of plastic products, along with considerations of sufficiency and smart design, can moreover provide design principles to guide future manufacturing, use and disposal of plastics.

Plastics are extraordinarily useful materials. They play key roles across the global economy in important areas such as food production and preservation, transport, insulation, clothing, healthcare and medicine. Since their discovery more than a century ago, we have mastered their mass production at a phenomenal annual rate, exceeding 460 megatonnes (Mt) in 2019 (ref. 1). Driven by their low cost, light weight, stability, longevity and high performance across various economic sectors, our consumption of plastic has greatly accelerated2. However, plastics pollution is correspondingly pervasive and only expected to get worse3–5, with serious consequences for the environment and human health6–9. Moreover, the greenhouse gas (GHG), and in particular carbon dioxide (CO2), emissions associated with plastics production, use and end-of-life (EoL) remain a substantial barrier to keep global warming below 1.5 °C (ref. 10). Globally, awareness of these challenges has grown in recent years, providing scope for new research on mapping the problems and providing solutions. A growing body of academic literature has also emerged, offering important insights into the problems associated with the current economic model, including plastics life-cycle carbon emissions11–13, dependence on fossil resources14–16, inadequate recycling techniques and infrastructure17–19, the implications of plastic waste exports and mismanagement20–23, ecosystem pollution20,24,25 and health risks26,27. Although these studies often include mitigation strategies2,28–31, a future sustainable plastics economy requires a multidisciplinary and holistic

vision of the entire life cycle, as well as an economic and legal perspective on what stimulates (or hampers) change. In November 2020, several members of the World Trade Organization (WTO) launched an informal dialogue to explore the role of trade cooperation in reducing plastics pollution and transitioning to a more circular and environmentally sustainable global plastics economy, complementing discussions in other forums32. Moreover, following a resolution adopted unanimously at the fifth session of the United Nations Environment Assembly (UNEA) in May 2022, countries have started negotiating a legally binding international treaty on plastic pollution, addressing the full life cycle of plastics, with the ambition of completing negotiations by the end of 2024【 】 (refs. 33–35). In line with the resolution, the treaty couldinclude provisions to promote sustain-able production and consumption of plastics through, among other things, product design and environmentally sound waste management (Supplementary information sections 4.2 and 4.4). If successful, the new treaty will provide legal certainty on key sustainability approaches and bring greater coherence to international law, which —at present—addresses aspects of the plastics life cycle in a fragmented manner (chemicals, wastes, pollution etc.)34–39. In light of these continuing international discussions, this perspec-tive aims to propose a vision for a future circular carbon and plastics economy. This circular system can only be attained if four interlinked targets, centred on consumption reduction, uptake of renewable 1 2 Department of Chemistry, University of Oxford, Oxford, UK. POLYMAT, University of the Basque Country (UPV/EHU), Donostia-San Sebastian, Spain. 3Faculty of Law, University of Oxford, Oxford, UK. 4Faculty of Law, UiT The Arctic University of Norway, Tromsø, Norway. 5Smith School of Enterprise and the Environment, University of Oxford, Oxford, UK. 6School of Geography and the Environment, University of Oxford, Oxford, UK. ✉e-mail: [email protected]; [email protected]; [email protected]

Nature | Vol 626 | 1 February 2024【 】 | 45

plastics, greater recycling and elimination of burdens to the environment, are achieved at scale. Urgent adoption of key interventions is necessary to reach these four targets by 2050, a system change that, to be successful, needs to be guided by ‘smart design’.

Re-imagining a circular plastics economy A bold system change is needed The plastics economy remains stubbornly linear. In 2019, only 9% of global plastic waste was mechanically recycled into new products, that is, most (estimated at 320 Mt) was lost to the economy through landfilling, incineration or entered the terrestrial or marine ecosystems1. It is also estimated that the volume of plastics accumulated worldwide since the 1950s (the amount ever produced minus that incinerated) has reached 8.2 gigatonnes (Gt), of which only 2.2 Gt remains in use and 6.0 Gt is waste40. The mass of carbon trapped in these plastic wastes is approximately double the total amount stored in human and animal biomass on Earth41. Moreover, our plastic consumption contributes to climate change: the anthropogenic GHG emissions through the plastics life cycle, including the extraction of petrochemicals, production of virgin polymers and additives, manufacturing of products and common EoL options, were estimated at 1.8 GtCO2 in 2015 (ref. 42). This figure represents 3.8% of the entire current global CO2 budget, roughly the size of the combined national emissions of the three largest economies in Europe (Germany, UK and France)43. Both the increasing volumes of waste and GHG emissions emphasize the inadequacy of the current linear plastics economy (predicated on extract-make-use-waste) to respond to the demands of a growing world population and the expected rise of living standards in emerging economies. For instance, the amount of plastic waste generated could almost triple by 2060 (ref. 24) and, if not remediated, more than 50% of it could become unmanageable and end up in the oceans44. GHG emissions associated with the plastics life cycle are also predicted to rise by 2050 to 6.5 GtCO2 under business-as-usual practices using fossil energy (10–15% of the overall annual global CO2 budget in 2050)10, a 3.6-fold increase primarily driven by a net growth in plastic production and waste incineration levels42. Against this backdrop, it is crucial to redesign the plastics economy across future sectors, regardless of their market share, place in the economy or intended use. Hence, building on the circular economy model that is already shaping law, policy and action by industry worldwide (Supplementary information section 4.1)45–49, we propose a future circular carbon and plastics economy centred on four targets: 1. Reduce plastic demand: eliminate 50% of all plastic materials and products. 2. Switch to renewable plastics: replace all fossil-fuel-based plastics with those sourced from alternative feedstocks, accelerating carbon recirculation through use of biomass and CO2. 3. Maximize recycling: design plastic materials and products for circularity and ensure that 95% of plastics are recycled. 4. Minimize environmental impacts: remove all sources of hazards to organisms and pollution to the environment, as well as decrease the carbon footprint throughout the plastics life cycle. The implementation of these targets needs to be guided by smart design, as described in further detail in Supplementary information section 3.1.

The carbon and plastics life cycle A selective focus on the individual targets set out above is unlikely to succeed. Instead, an integrated and scalable approach is essential. To accomplish this, the plastics economy as a whole needs rethinking. But where and how to start? As a useful basis for such analysis, CO2 emissions from the global plastics life cycle provide quantitative information and an effective evaluation tool. Current life-cycle emissions 46 | Nature | Vol 626 | 1 February 2024【 】

for plastics are dominated by the production of virgin plastics from petrochemicals (61%), followed by emissions from product manufacturing (30%) and lower emissions attributed to EoL treatments (9%)42. The four targets for future sustainability must be applied synergistically across the plastics life cycle to minimize these emissions. For instance, substituting virgin petrochemicals as a plastics feedstock and ensuring the recycling of plastic wastes should both drive down GHG emissions during the life-cycle stage42,50,51. As we shall discuss later, interventions must be carefully and simultaneously balanced to not only cut GHG emissions but also to reduce other negative social and environmental impacts throughout the entire life cycle39,44. Proposing mechanisms for such system restructuring first requires a method to formulate global estimates for total material fluxes and life-cycle GHG emissions. One approach considers total carbon recirculation. Because carbon constitutes, on average, about 74 wt% of current commodity polymers41, tracing the carbon content of plastics serves as a useful proxy to track recirculation. Plastics feedstocks, products and wastes could be understood as mass vectors for carbon in all its oxidation states, including the most reduced in methane (CH4) and hydrocarbons, the rich structures of oxygenated bio-based raw materials and polymers, and its oxides, carbon monoxide (CO) and carbon dioxide (CO2). Some in the biogeochemical community already apply related carbon-accounting frameworks in tracing the distribution and impact of plastics in the biosphere41,52,53. As well as understanding where and how carbon in plastics accumulates on Earth, tracking carbon is useful to monitor and improve sustainability and circularity measures; for instance, towards reducing GHG emissions, preventing environmental pollution and limiting further carbon-feedstock depletion. We propose a circular carbon and plastics life cycle that considers current, future, natural and technological pathways by which carbon raw materials are produced from feedstocks, transformed into carbon-storing assets in the form of monomers and polymers, stocked while plastics remain in use and recycled into new plastics or feedstocks (Fig. 1). Crucially, a hierarchy of recycling pathways for plastic waste is needed to organize and optimize the carbon recirculation in the system. The framework gives priority to recycling routes that preserve material integrity, value and chemical structure, and that minimize energy and material losses. This is consistent with the well-known waste-management hierarchy, which is a common principle of waste management law and policy and which generally assigns priority to waste-management options in the following order: prevention, reuse, recycling, recovery and, finally, disposal (see also Supplementary information section 4.5)54. In this circularity-oriented life cycle, plastics are used for their intended application, providing economic value through function and sustainable performance55. Once their required service life is fulfilled, which can vary from days to decades, and all options for material reuse are exhausted, retrievable wastes are collected, sorted and managed using the most appropriate recycling routes31,56–60 (physical, chemical or biological; see Box 1 and Fig. 1). Such selective material recirculation enables the carbon embedded in plastics to efficiently flow ‘upwards’ from waste to resources, thereby regenerating feedstocks and preventing waste accumulation, leakage and consequent pollution. Plastic wastes that cannot be either physically or (bio)chemically recycled must follow a biological degradation route in specialized facilities, including industrial composting or anaerobic digestion, which allows for the capture of the released CO2 emissions61. Only intractable or contaminated waste mixtures should be considered for energy recovery by means of incineration as a last resort, owing to the energy inefficiency of such processes and the need to capture all CO2 emissions. In this system, all plastics are made from recycled C1 molecules, lignocellulosic biomass (second generation), waste biomass and captured CO2 (refs. 62–65). This avoids use of fossil carbon for plastic production and energy generation. C1 molecules such as CO, CO2, methanol (CH3OH) and CH4 are key building blocks that connect the

Atmospheric CO2 Circular carbon and plastics life cycle Photosynthesis Biomass

Hydrocarbons Naphtha Syngas

Catalysis Direct carbon capture

Fossil resources

Pyrolysis and gasification Chemically unbreakable (for example, polyolefins)

Carbon refining C1 chemicals CO, CO2, CH3OH, CH4

Monomers

(Bio)chemical depolymerization

Polymerization

Energy

Controlled incineration with carbon capture and energy recovery

Polymers

Anaerobic digestion Industrial composting

Conversion

Biodegradable (for example, PLA, PHAs, TPS)

Mechanical recycling

Cleavable polymer chains (for example, PET, PUR)

Reuse Recoverable thermoplastics

Products Discard

Renewables

Intractable, non-recyclable

Waste Collection and sorting

Production pathways Linear Sustainable CCU

Recirculation pathway hierarchy (1) Reuse (2) Physical recycling (3) Chemical recycling (4) Biodegradation (5) Energy recovery

Landfill

Leakage

Emissions and pollution

Incineration

Fig. 1 | The circular carbon and plastics life cycle. In the current system, which features linear flows of production and disposal (black pathway), demand for plastics is satisfied by consuming fossil resources, while the lack of perceived economic value in plastic waste leads to unsustainable rates of landfilling, incineration or, worse, environmental leakage. In this future system (grey box), a recirculation pathway hierarchy helps to maximize material recycling and carbon recirculation by prioritizing (1) reuse of products, (2) physical recycling and (3) (bio)chemical recycling (including depolymerization to monomer, pyrolysis and gasification) over (4) biodegradation and (5) energy recovery

(last resort and only with carbon capture). The recycling streams replace the need for more virgin raw materials for the plastics industry (polymers, monomers or unprocessed feedstock) and avoid extraction of petrochemicals. Any remaining production of virgin polymers uses carbon from captured CO2 or indirectly from (waste) biomass. The carbon-recirculation system must be powered using only renewable energy. Acronyms refer to common polymer classes, for example: PET, poly(ethylene terephthalate); PHAs, poly(hydroxy alkanoates); PLA, poly(lactic acid); PUR, poly(urethane); TPS, thermoplastic starch.

downstream and upstream cycles. This future system cannot operate using only biomass as the feedstock, but carbon capture and utilization (CCU) technologies are also essential66. Estimates to reach climate targets situate the efforts to scale current carbon-removal capacity, including CCU, by a factor of 1,300 by 2050, an increase of 4.8 GtCO2 per year from 2020 values67. Captured CO2 must be catalytically transformed into chemical intermediates, for example, syngas, alkenes, naphtha, aromatics, carbonates and alcohols68–70. In nature, CO2 is captured and converted into biochemicals by photosynthesis. Plant metabolic pathways can be exploited, or even enhanced, to eventually produce more complex monomers, including those from lignin, carbohydrates, triglycerides and terpenes71,72. These combined synthetic and natural CO2 use pathways must offset all GHG emissions from plastic production, manufacturing, recycling and any eventual

biodegradation in order for the system to reach net zero: a very tough challenge.

Evaluating carbon emissions and other metrics Addressing the impacts and biosphere burdens of the materials, processes and energy requirements for this future circular plastics economy requires careful life-cycle assessments accompanied by notable technological breakthroughs in the coming decades. GHG emissions are essential metrics to evaluate the effect on climate change, but there will be other negative environmental trade-offs that require careful consideration and minimization. A useful metric for comparing impacts is the global planetary boundary framework73–75. This defines boundary limits for human activities and sets a safe operating space against nine Earth-system processes, including biosphere integrity through Nature | Vol 626 | 1 February 2024【 】 | 47

Box 1

Key terms of a circular carbon and plastics economy See Supplementary information for further discussion.  Circular carbon and plastics life cycle is a framework that keeps the atomic carbon embedded in plastics in circulation for as long as possible and tracks all its transformations as materials move from waste to resources to products. It promotes the sustainable management of the carbon content in plastics, giving priority to efficient conversions with minimal energy inputs. It requires an effective waste collection infrastructure and coupling between recycling and chemical manufacturing industries. ● Renewable plastics are considered those obtained from recirculated carbon, with chemicals, monomers and plastics manufactured by CCU or indirectly through photosynthesis to biochemicals. ● C chemicals are key molecules, each containing one carbon 1 atom, from which many monomers and plastics are prepared in the plastic-carbon life cycle. They include methane (CH4), carbon monoxide (CO), methanol (CH3OH) and carbon dioxide (CO2). ● Bio-based plastics are derived from biomass feedstocks, co-products or wastes selectively chosen to minimize other inputs (for example, land use, fertilizer, energy and irrigation). Such plastics include new synthetic polymers produced by chemical transformations of biomass, for example, polylactic acid from starch, and drop-in polymers, analogous to existing petrochemical plastics but biomass-derived, such as biopolyethylene from bioethanol. Bio-based plastics need notable developments in five main areas: competitive economics, efficient transformations from sustainable resources, optimized performance characteristics, establishment of clear EoL options (including recycling) and global standardization. ● Reuse extends the life of plastic products by reintroducing them to the economy after use. In combination with other ‘r strategies’ (redistribute, repair, refurbish, repurpose and remanufacture), reuse is an appealing approach to reduce consumption of virgin polymers, as it requires minimal energy compared with recycling, and prevents waste. These strategies benefit from effective business models, such as refills for packaging. ● Physical recycling applies mechanical and/or thermal energy to recycle waste plastic into new products (for example, through melting, extrusion and moulding technologies). These methods are expected to be the dominant EoL treatment for thermoplastic materials, which constitute 86% of the global plastics market, owing to the low energy requirements and economic feasibility. However, challenges include deterioration in material properties after repeated recycles and the need to separate and purify mixed or contaminated wastes. ● Chemical recycling involves the controlled breakdown of plastics into monomers or precursors suitable for recycling to polymers using existing manufacturing processes. It is, at present, most ● ●

(in)direct land-use change, balanced nitrogen/phosphorus cycles and atmospheric aerosol loading10. It is essential to work together to articulate clear and commonly understood international sustainability criteria76 and accelerate research, including to efficiently use and transform biomass feedstocks that avoid habitat loss, overconsumption of freshwater and fertilizers, and limits other adverse environmental impacts. Providing solutions will not be easy, but priority fields for research include methods to transform organic wastes and agricultural by-products to 48 | Nature | Vol 626 | 1 February 2024【 】

widely applied to polyesters and may be accomplished by alcoholysis or hydrolysis reactions, often at high temperatures. Biochemical processes use natural or synthetic enzymes or whole organisms (bacteria) and typically operate under low-temperature conditions. Poly(ethylene terephthalate) (PET) can be recycled to its monomers using either of the (bio)chemical recycling processes. Equivalent processes for polyolefins are very challenging because of the high stability of the carbon–carbon bonds; their pyrolysis and gasification at high temperatures often yields product mixtures. Although new developments and infrastructure are required to implement chemical recycling, it may be useful to maximize carbon recirculation from mixed waste streams, thermosets and/or multilayered materials. ● Biodegradation occurs when plastics are decomposed to molecules using processes accelerated by microorganisms, either in the presence of oxygen (aerobic) or without it (anaerobic digestion). Polymers containing heteroatoms, such as polyesters, polycarbonates and polyamides, are most often investigated and can be biodegraded to C1 chemicals and water. ● Incineration with energy recovery is best applied to inseparable or contaminated plastic wastes that cannot be treated through other forms of recycling. Although energy is recovered, this option sits at the bottom of the recirculation pathway hierarchy because it requires capture of carbon emissions and destroys the material structure and value. ● Smart design refers to innovations in plastic production, manufacturing and recycling that are underpinned by material applications, EoL retrievability and any environmental impacts throughout the life cycle of the product. ● Service lifespan of plastic products indicates the length of time a particular material retains its function and is used, reused, repaired, refurbished or remanufactured. Long-lived plastics retain their properties for years and even decades (for example, in construction and automotive) and thus show high durability, repairability and adaptability. Short-lived plastics are disposed of within months or a few years (for example, packaging), contributing substantially to projected increases in demand for plastics. ● Recoverability of plastic wastes describes the feasibility to collect, sort, process and recycle a plastic after use. Recoverable plastics are used in applications and in geographies in which recovery and separations are physically and economically feasible and are best recycled. Irretrievable plastics describe products that are distributed in the environment after use; for example, uncollected wastes in rural areas of the developing world, waste fishing gear, agricultural mulch film plastics, microplastic sources, including from tyre dust and textile fibres, and formulated products. These cannot be recycled and are best designed to biodegrade.

polymers77 or application of crops that grow rapidly in marginal lands or increase the soil organic carbon content78. By comparison, the direct use of CO2 to make monomers and polymers does not usually result in burdens on arable lands and habitats, which are great concerns in any increased-biomass-consumption model1,10. Nonetheless, CCU in polymer production may increase demand for renewable energy and green hydrogen, particularly if reduced molecules such as methanol or syngas are the primary feedstocks10. Increased renewable energy production has planetary footprints, including increases to atmospheric

Virgin production and conversion Incineration

Landfilling

Recycling

Biodegradable

Petrochemical Bio-based Net GHG emissions in 2050 (GtCO2)

Plastic industry scenarios in 2050

a

b

0

670

57

0

c

0

Life cycle emissions (GtCO2)

100 EoL (share, %)

3.2

4.8

4.0

0% 4.5

0.8

0

1.6

2.4

3.2

4.0

4.8

39%

4.1

825

100

528

64 40

0

d

EoL (share, %)

1,100

0

Life cycle emissions (GtCO2)

Conservative commitments 0.8

0

1.6

100

4

EoL (share, %)

77

3.2

4.0

4.8

2.2

1,100

550

2.4

36%

Bold system change Life cycle emissions (GtCO2)

Manufacturing (MMT plastic)

2.4

Business as usual 1,100

Manufacturing (MMT plastic)

1.6

Life cycle emissions (GtCO2)

0

EoL (share, %)

70

0.8

0

100

Manufacturing (MMT plastic)

1,100

Manufacturing (MMT plastic)

Carbon recirculation

Baseline

0.8

0

1.6

1.8

3.2

4.0

4.8

86%

0.2

Net GHG emissions levels in 2020 (1.6 GtCO2)

Fig. 2 | Plastic industry scenarios and GHG emissions for 2050 based on an estimated global production of 1.1 Gt. a, Baseline scenario (all plastic waste is discarded through a combination of landfilling and incineration). b, Businessas-usual scenario (projected growth from current levels in recycling, up to 43%, and incineration, up to 50%). c, Conservative commitments scenario (only partially deploying key interventions: 25% demand reduction, 50% substitution

of fossil-based plastic and intermediate deployment of renewable energies). d, Bold system change (cut plastic demand in half, full-scale introduction of bio-based plastics and recycling technologies up to 95%, plus complete decarbonization of the energy system). In the left column, the dashed frames highlight the projected size of manufacturing and EoL in the 2050 baseline scenario. MMT, million metric tonnes.

aerosol loading and impacts on P-cycles. Hence, the implementation of larger-scale CCU technologies and biomass uptake must be carefully balanced10. Notably, plastic recirculation by physical and (bio) chemical recycling replaces the need for extracting virgin renewable resources, avoiding notable burden shifting and, therefore, represents a promising approach to reducing GHG emissions without other negative consequences10. Several excellent global studies have modelled and quantified the carbon footprint of the plastics life cycle, helping to visualize the consequences of implementing particular mitigation strategies. Notably, recent investigations led by Suh42, Bardow50 and Stegmann51 all show that isolated solutions cannot fix the plastics problem; it is only the interplay of four interlinked targets that effectively reduces GHG emissions towards net zero. Using these published models and datasets, we propose four different 2050 scenarios for the plastics economy, at a time when the market is expected to grow to 1.1 Gt plastic per year (Fig. 2): (1) baseline, (2) business-as-usual, (3) conservative commitments and (4) bold system change. For each scenario, estimates of the consequences for GHG emissions and recirculated carbon content are provided, capturing the possible impacts and outcomes, rather than any kind of absolute forecast. The scenarios apply the four mitigation strategies: reducing plastic consumption, increasing recycling rates, replacing fossil carbon with renewable alternatives and powering manufacturing and recycling processes with renewable electricity (Supplementary information section 1)42,50,79. It should be noted that, in these scenarios, there were many data gaps and, therefore, estimates had to be made, particularly given the diversity in global

waste infrastructure (where it exists) and recycling options. Another challenge associated with estimating the effects of increased use of renewable plastics is the earlier market stage of such materials, which affects manufacturing-process efficiency with knock-on consequences for GHG emissions. In the four scenarios, the current data for the top five bio-based plastics was applied, but we acknowledge that, in the future, their properties need to be improved and that their current life-cycle GHG emissions should serve as an upper bound to drive innovation. The current scenarios do not attempt to quantify or compare other negative environmental impacts, but it is worth emphasizing the critical importance of continuing with such assessments42,80. The baseline scenario involves all plastics being both fully fossil derived and disposed of in landfill or by incineration. Because no carbon is recirculated, future demand must be satisfied by manufacturing more virgin plastics from crude oil or gas. These upstream processes result in an estimate of 4.0 GtCO2, that is, roughly 90% of all life-cycle emissions. By comparison, the business-as-usual scenario linearly tracks the range of EoL options relevant today to 2050, with rates of incineration and recycling reaching up to 50% and 43%, respectively40. The climate cost of maintaining a business-as-usual scenario is massive: the carbon footprint remains >2.5-fold higher than the level of today and surpasses 4.0 GtCO2 a year, not far from the projected baseline. This scenario reveals a key message: augmenting recycling of plastics waste alone (approximately 40% of recirculated carbon) is simply not enough to tackle the emissions crisis. Breaking from the status quo is very challenging in complex global economies and requires coordinated action and investment. Nature | Vol 626 | 1 February 2024【 】 | 49

Perspective The conservative commitments scenario, which results in global GHG emissions of about 2.2 Gt per year and only moderate carbon recycling, illustrates the negative effects of such global barriers on life-cycle emissions (Fig. 2c and Supplementary information). For instance, slow changes in consumption behaviour and reorganization of power relationships could hamper a swift cut in the growth of plastics demand. The large-scale infrastructure investment needed to close the collection gap that exists in many parts of the world (about 60% of the planet either lacks or has an inadequate waste infrastructure) represents a substantial challenge to establishing a functional recycling industry. Moreover, the slow growth of recycling technologies that can preserve material quality, a difficulty in current mechanical recycling81, could perpetuate material loss. Feedstock markets, regional availability and costs to build/retrofit manufacturing and recycling infrastructure present further obstacles to a global scale-up of renewably sourced plastics production, prolonging reliance on fossil-carbon resources. Equally, the seasonal and regional intermittency of some renewable energy, its incorporation to mature energy grids and the overall system costs and complexity could slow down the deployment of decarbonized electricity82,83. Mitigation of the remaining GHG emissions from both petrochemical refining and power requires carbon capture and storage. Although the costs for such technologies remain unclear, a value of US$100 per tCO2 is the relevant order of magnitude66; compensating for the residual emissions in this scenario could add more than US$200 billion to the costs of the plastic system and would do nothing to address any other negative environmental consequences. The bold-system-change scenario has the potential to sufficiently reduce the CO2 emissions, but it requires some very substantial changes to current practices (Fig. 2d). First, future plastic demand must be cut by 50% from expected levels, mandating the decoupling of future economic growth from plastic consumption. Second, recycling must become the dominant managed EoL option, so that it can extract the maximum material and economic value from wastes, enabling recirculated carbon to deliver 86% of new plastics demand. Third, any remaining virgin raw materials must derive from renewable resources, including biomass and CCU, which fully replace petrochemicals. Finally, the energy must be entirely renewable to access the lowest CO2 emissions. Only this scenario has the potential to reduce global emissions to about 0.2 GtCO2 in 2050 and could reach net zero if the remaining emissions were subject to carbon capture and removal technologies. Achieving such a scenario must simultaneously address any possible risks, for instance, through the adoption of global and national measures to prevent regrettable substitutions and the recirculation of hazardous chemicals.

Delivering a bold system change At Stockholm+50 ( June 2022), which took stock of the human environment, states called for a “system-wide change in the way our current economic system works to contribute to a healthy planet”84. This reflects the fact that a bold system change requires a fundamental rethinking of current economic structures and laws. The different stages of the plastics life cycle are interconnected, subject to multifaceted and complex cross-border trade flows85. A coordinated approach is therefore needed to avoid the risk of new legal and economic measures inadvertently creating new (trade) barriers, disproportionally affecting certain groups in society or disincentivizing innovation86. The following sections highlight some of the interventions required, although noting that they will require continual evaluation to guide and scale a future circular carbon and plastics economy, and acknowledging the challenges in tracking progress towards targets. The plastics life cycle is only partly regulated by international law35–38 and states across the world have adopted specific national measures in relation to the plastics sector (Supplementary information section 4.6)87. In this section, examples from the EU are provided as a useful knowledge base, as it has been actively pursuing a 50 | Nature | Vol 626 | 1 February 2024【 】

more circular (plastics) economy since 2015 (refs. 47–49) and aims for a climate-neutral continent by 2050 (refs. 88,89). This does not, however, imply that these examples are universally pertinent or may be easily implemented across all economies. Interventions for system change, particularly in the Global South, face challenges related to rapid growth and urbanization, insufficient municipal solid waste collection and management, large quantities of high-income-country waste imports, substantial data gaps impeding development of effective policy aligning social and financial incentives and scarce data on the large and important informal sector of waste management5,90,91.

Sustainable plastics through smart design The future plastics economy should be centred on sustainable consumption and production practices, another important UN Sustainable Development Goal (SDG12)92. The transition from our current complex and interconnected economy must be upheld by principles of sufficiency rather than superfluous consumerism, while still meeting socio-ecological needs93,94. Plastics design and manufacturing choices require a process we term ‘smart design’. Guiding this process are four questions that help focus attention across the product life cycle. What is the origin of the raw materials? What is the intended product application? What EoL option is most appropriate to recirculate the material and carbon? And what are the environmental and health impacts throughout the product life cycle? To help answer these questions, a series of smart design principles can be formulated (Box 2). These should help to guide appropriate use of resources and production methods, deliver sufficient performances, ensure sound waste management and help to minimize broader environmental impacts. Because plastics comprise numerous different chemistries and are used across several application sectors, two other aspects of the product life cycle warrant further attention and discussion: their service lifespan and recoverability95,96. Service lifespan defines how long plastics within products remain in (re)use and varies enormously from short-lived packaging (about half a year, on average) to long-lived construction materials (approximately 35 years, on average)97. Recoverability describes the potential for plastics to be recirculated, either by reuse or recycling. It is affected by material technical characteristics, for example, composition, separation methods, product disassembly processes and polymer recyclability, as well as by the environmental impacts of any sorting, separation and treatment options. It is also influenced by economic factors, for example, waste infrastructure management, recycling business models, cost–benefits, asset devaluation and legal constraints that determine recoverability, as discussed below. To help in the (re)design process, a taxonomy for all plastic products is proposed on the basis of these distinctions and is exemplified here using some specific materials and applications (Fig. 3a). Plastic products with long lifespans and a high (potential) recoverability must be recycled in future. This includes those materials used in construction, textile or transport sectors at present in which polymers are widely used but recycling levels are rather low. Conversely, plastics used in short-lived applications and with low recoverability show inherently linear life cycles, and so will need to be eliminated, substituted or redesigned to improve recoverability; for instance, by redesigning multilayer packaging. In applications in which recoverability is impossible and environmental dissipation is a consequence of the application, polymers must be completely biodegradable, so that carbon recirculation occurs through natural carbon cycles. These applications include plastics in agriculture (mulch films), as well as fast-moving consumer goods (personal or home-care formulations). Figure 3b,c illustrates the potential reductions in carbon emissions, achievable by 2050, if such smart design decisions were implemented in four exemplar application sectors: recoverable packaging, complex packaging, construction and agriculture. These examples represent different quadrants in the proposed plastics taxonomy. In

Box 2

Design principles for sustainable plastics 1. Net-zero feedstocks. Maximize carbon recirculation by disengaging plastics feedstocks from fossil sources and using renewable carbon, such as biomass, industrial by-products, waste CO2 or recycled plastics. 2. Efficient production. Minimize energy input by optimization of production and conversion (manufacturing), the use of catalytic processes, the balance of conditions and reduction of the number of intermediates and stages. 3. By-product rejection. Preserve the value of carbon, and other elements, by applying atom-economical transformations, maximizing process selectivity and recycling or repurposing by-products, offcuts and scraps. 4. Essential purpose. Deliver the necessary performance (for example, flexibility, density, toughness, durability, gas permeability, optical clarity) without overengineering plastic materials and products. 5. Extended use. Increase product lifetimes by allowing the repair of damaged materials and giving priority to reuse models (for example, return and refill for packaging). 6. Competitive properties. Implement bio-based plastics with properties that match or exceed those of current fossil plastics, while minimizing manufacturing costs.

the bold-system-change scenario, implementation of smart design principles could help to greatly reduce GHG emissions (70–90%) compared with the business-as-usual scenario. In each sector, different interventions will be necessary to curtail emissions, examples of which include demand reduction for short-lived plastics, increased recycling of recoverable plastics and ensuring complete biodegradability for irretrievable plastics (see Supplementary information section 2). To implement smart design across all sectors, technical advances must be guided by future sustainability criteria98,99 and supported by a robust legal framework and economic incentives, as discussed next for each of the four key targets identified for a bold system change.

Reduce plastics demand Restrictions on the production and consumption of unnecessary plastics incentivize smart design for application and thereby help to reduce overall plastic demand. These may take the form of national bans on single-use items, with early examples including the prohibition on certain plastic bags in Meghalaya (India)100, South Africa101, Eritrea102 and Bangladesh103, and—more recently—the EU Single-Use Plastics (SUP) Directive104. It has also been suggested that the new international plastics treaty could, among other things, target both the phasing out/ reduction of primary plastics, of problematic and avoidable plastic products and of chemicals and polymers of concern105. This raises further questions about the most appropriate criteria for definition and the need to clearly distinguish between substances and products. An international phasing out of substances of concern in relation to plastics could draw inspiration from existing international restrictions of persistent organic pollutants106 or of ozone-depleting substances107. Furthermore, measures to reduce plastic demand could stipulate or prohibit plastic content, such as the US prohibition of microbeads in cosmetics108, the new EU ban on intentionally added microplastics109 or the ban on oxo-degradable plastics in the EU104. In the packaging sector alone, the elimination of unnecessary plastics combined with innovative product and packaging design is estimated to reduce by about 38% packaging demand in Europe by 2050 (ref. 110). A tax or fee on

7. Preserved value. Preserve energy and raw material value for the long term, which requires the conservation of polymer and monomer structures, if possible, during physical and chemical recycling. 8. Easy separation. Minimize the use of additives and other contaminants, design products for separation, sorting, disassembly and purification in recycling and replace multimaterials with homo-composites. 9. Optimized recycling. Maximize yield, value and quality of properties in recyclates; chemical recycling and upcycling should minimize energy inputs and preserve value. 10. Synergistic biocompatibility. Design materials for optimal compatibility with biological recycling plants (aerobic composting and anaerobic digestion) wherever recycling is unsuitable (for example, contaminated agricultural and food wastes). 11. Harmless biodegradation. Provide materials with embedded strategies for full degradation to non-toxic metabolites wherever polymers are environmentally distributed or dispersed (for example, water formulations). 12. Minimal hazards. Assess ecotoxicity and human toxicity of all plastics, additives and degradation products and analyse negative environmental impacts throughout the life cycle.

plastic products to reflect their social and environmental costs would also help to reduce demand by increasing the price of plastic per user. A tax on plastic products may have the added benefit of generating revenue to subsidize recycling and/or composting infrastructure, helping to make recycled materials more competitive111. Such so-called hypothecated tax revenues do not typically appeal to finance ministries, but experience suggests that they make environmental taxes more politically palatable112, provided inequitable distributive consequences are avoided or compensated for with an appropriate policy mix86.

Switch to renewably sourced plastics So far, global climate targets are driving aspirational policy objectives for sustainable carbon cycles. However, any efforts to introduce renewable feedstocks must also carefully consider and minimize the other environmental trade-offs, as mentioned previously. This is not an easy task, as increasing use of biomass and CO2 use might result in transfer of ecological burdens from climate change to other Earth-system processes1,10. On the other hand, the petrochemicals industry also has other negative environmental effects, including particulate pollution, ecosystem threat and release of volatile organic compounds, as well as sulfur-containing and nitrogen-containing contaminant gasses. Given the pressing need to reduce GHG emissions, the target to scale up future renewable-feedstock technologies must be guided by careful environmental-sustainability analyses. Economic and legal measures also help to drive this change, for example, in the EU, at least 20% of the carbon used in chemical and plastic products should be from renewable sources by 2030 (ref. 113). Realizing these objectives requires investment in both the infrastructure and markets for such products. States can facilitate the transition through implementing targets, for example, the Dutch Transition Agenda for Plastics plans to increase the percentage of recyclate and bio-based plastics to 41% and 15%, by 2030, respectively114; the USA has a bold goal to, in 20 years, “demonstrate and deploy cost-effective and sustainable routes to convert bio-based feedstocks into recyclable-by-design polymers Nature | Vol 626 | 1 February 2024【 】 | 51

a

Recoverability High

• ‘Drop-in’ • Durable • Recyclable for example, construction (pipes, window frames)

• ‘Drop-in’ • Reusable • Recyclable for example, recoverable packaging (rigids, bottles) Service lifespan Short

Long

• Biodegradable • Compostable • Reusable

• Biodegradable • Compostable • Durable

for example, complex packaging (flexibles, multimaterials)

for example, agriculture (mulching films)

Low

b

Business as usual

Bold system change

1.00

0.75

0.50

0.25

0

c CO2 savings of bold system change versus business as usual (%)

GHG emissions (GtCO2)

1.25

Recoverable Complex packaging packaging

Construction Agriculture

100

80

60

40

20

0

Recoverable Complex Construction Agriculture packaging packaging

Fig. 3 | The impacts of smart design, service lifespan and recoverability in the carbon footprint of plastics. a, Taxonomy for plastic products based on their inherent service lifespan and recoverability, with sector examples in each quadrant. b, Estimated GHG emissions for the business-as-usual and

bold-system-change scenarios for four illustrative application sectors. c, Corresponding GHG savings attained by applying smart design principles in a system change.

that can displace more than 90% of today’s plastics”115,116; and the EU proposes to mandate the use of compostable packaging for certain applications117. States can also facilitate these objectives indirectly through exemptions, for example, the exemption for certain degradable and soluble polymers from the restriction on intentionally added microplastics under EU chemicals law109. States can further create demand for bio-based goods by implementing green public procurement criteria when exercising large-scale purchasing power in contracts for goods, services and infrastructure development118. In the EU, public authorities spend approximately €2 trillion each year on public contracts, equivalent to roughly 14% of its gross domestic product (ref. 119). Such spending power serves as an effective tool for directing markets in a sustainable direction. Scaling renewable plastics requires cost-competitiveness, a current key barrier. States can provide financial subsidies to allow manufacturers to sell such plastics at lower cost, driving production at scale. Global growth of renewably sourced plastics could rapidly increase from about 4% to 10–20% if their adoption was subsidized and politically supported, similarly to contracts for difference used to grow low-carbon (emissions) energy120. Public–private partnerships that bring together the expertise and resources of several stakeholders can accelerate innovation and advance cost reduction for nascent biobased

industries. Further developing such partnerships is therefore flagged as critical for achieving the US biotechnology and biomanufacturing goals, including through databases, joint funding opportunity projects116 and user facilities such as the Advanced Biofuels and Bioproducts Process Development Unit (ABPDU), a scale-up facility that has already helped companies raise more than US$2 billion in private funding and transition 17 products to market. Similarly, the Circular Bio-based Europe Joint Undertaking (CBE JU) is a €2 billion public– private partnership that finances projects advancing competitive circular bio-based industries. Conversely, phasing out fossil-fuel subsidies could help to bridge the cost gap between fossil and renewably sourced plastics121–123. Pressure to phase out inefficient fossil-fuel subsidies will probably intensify in line with SDG12 (ref. 92) and the Glasgow Climate Pact124, and the initiatives of some countries already signal a change in this direction125. Further, a fossil-carbon-free system requires changing the investment from petrochemical plants to technologies for biomass and CCU to chemicals. This could be stimulated by including CCU in emissions-trading schemes, such as under the recently amended EU Emissions Trading System (EU ETS), which already supported innovation in CCU and other low-carbon technologies through its Innovation Fund, and where emissions allowances now no longer have to be surrendered for GHG that are captured and used where they become

52 | Nature | Vol 626 | 1 February 2024【 】

permanently chemically bonded in a wproduct (that is, obviating atmospheric contamination under normal use), including any normal activity occurring after the product EoL126. At present, however, the petrochemical industry continues to invest in new fossil-carbon plants, which risks petrochemical lock-in127,128. Building and maintaining demand for renewably sourced plastics also relies on effective and accurate communication to downstream users. In the EU, for example, there is at present no systematic certification scheme or label for such products, although voluntary standards have been developed129. This gap should be addressed through clear international standards (for example, through the International Organization for Standardization) and definitions (for example, through the new plastics treaty), including for terms such as bio-based, renewable carbon, biodegradable and compostable plastics, so as to harmonize domestic measures.

Maximize recycling Current technologies at scale for mechanical recycling are unlikely to be able to solely meet future needs for plastic waste recycling, as they are hampered by material degradation, losses and incompatibilities with mixed-waste streams, multilayers, additives and thermosets. Hence, notable technical improvements, infrastructure upgrades and worldwide expansion of matured technologies, in coordination with better waste management, are necessary. Moreover, to maximize efficient recycling, we need clear definitions, as has been discussed in the context of revising international technical guidelines on the environmentally sound management of plastic wastes under the Basel Convention (Supplementary information section 4.3)130,131. Defining what falls within/ outside the definition of plastics recycling has important legal implications, both for international waste-management obligations23 and for states to meet recycling targets, such as the EU target to recycle 55% of plastic packaging waste by 2030 (ref. 132). A variety of specific tools exist to promote recycling. These include the growing practice of extended producer responsibility (EPR) schemes specific to plastics, for example, packaging, or products containing them in, for instance, electronics or vehicles. EPR makes producers responsible physically or financially for their plastic products at EoL inclusive of collection and sorting. Many EPR schemes require fees, pricing-in environmental costs and charging producers more when selecting non-recyclable or harmful plastics. Shifting responsibilities and costs for the EoL of plastics is also intended to stimulate design change. EPR has been increasingly discussed as a means of implementing the polluter-pays principle, including within the context of the new plastics treaty133. By the end of 2024【 】, for exam-ple, EU member states must ensurethat EPR schemes are established for all packaging132. Althoughsome producers resist, more than 100 large-scale plastic-packagingbusinesses support the expansion of EPR globally, pointing out the benefits for effective and economic recy-cling134. More than 400 EPR instruments exist around the world, some mandatory through law and some not, with varying success in inducing design changes owing to challenges such as a lack of individual rather than joint producer responsibility, free-riding and implementation gaps135,136, although criteria for a successful EPR have been developed in international forums137–139. Producers also increasingly make voluntary commitments through the global network of Plastics Pacts. In the UK, pact participants agreed four targets by 2025, including 70% plastic packaging effectively recycled or composted, helping to drive demand for recycling facilities, and a target of increasing to 30% the average recycled content in plastic packaging, pushing demand for recycled feedstocks140. Other tools to increase recycling include fiscal incentives such as tax credits to promote research and development of plastics-recycling technology. In Colorado (USA), for example, a plastic recycling investment tax credit is available for expenditures made towards new plasticsrecycling technology in the state141. Implementing deposit-return

schemes to incentivize consumer behaviour change may be effective in enhancing consumer recycling rates and generating fewer emissions than direct recycling subsidies142. Innovative business-to-business solutions, such as digital trading platforms for recycled plastics, are showing potential to increase recycling in formal and informal economies (see Supplementary information). In developing economies, building waste-management facilities will also require supportive regulatory environments and investment to scale infrastructure projects143. Internationally, coordinating the investment for scaling recycling infrastructure with that of waste-collection systems is a continual challenge5,144. Finally, promoting the uptake of recycled plastics as a feedstock will require coherency between different areas of law, in particular, waste, products and chemicals law (see Supplementary information section 4.7)145. In the EU, key regulatory obstacles to a circular economy have already been identified, such as different rules on how waste becomes a new material; legacy substances hampering the uptake of recycled materials; a lack of information about complex waste streams; and substances being differently classified as hazardous or not falling under products and waste law146. Greater regulatory alignment between these sectors, and across countries, will be vital to generate confidence in the quality of recycled (plastic) materials. Similarly, policy coordination including subsidies for recycling and removal of fossil-fuel subsidies (see previous section titled ‘Switch to renewably sourced plastics’) are critical for the cost competitiveness of recycling against current virgin fossil-based production.

Minimize broader environmental impacts Eliminating pollutants will require continuous attention, as the plastics sector changes over time. Greater transparency is therefore key, from the content of waste streams to avoiding legacy pollution to identification of all additives used during manufacturing147,148. Legal measures for greater transparency include the active discussion on making certain polymers subject to registration and evaluation under EU chemicals law (REACH), as is the case in some other jurisdictions149. A comprehensive and transparent global knowledge base on the composition of plastics would benefit downstream users and allow regulators to assess and manage risks, similar to clearing-house mechanisms established in other areas150,151. Other measures could include making market access for products that contain plastics conditional on documenting environmental information (product passports)152. To be most effective, such measures should apply across the board and not only to specific products or categories. Further, a circular plastics economy requires careful management and evaluation of its flows to eliminate pollution, involving all stakeholders. Consumer education is key, which calls for clear marking specifications about appropriate waste management and the consequences of littering, such as foreseen under the EU SUP Directive104.

A roadmap towards sustainable plastics To inspire a shift towards a more circular carbon plastics economy and to stimulate discussion on how to achieve this, we propose an ambitious timescale for economic and legal measures, from the present until 2050 (Fig. 4 and see Supplementary information section 5 for further details). Although the roadmap proposes, sequences and tentatively suggests time periods for when their effects on the system may be felt on the basis of leading policy and academic literature, these interventions should be considered both cumulative and continuous over time. For example, a collaborative global approach to solving plastics pollution is already continuing in several multilateral forums; this includes the prospect of adopting the aforementioned plastics treaty. Moreover, some measures (for example, improving waste-management infrastructure) will be more readily achieved in economies with existing foundations for such changes, and others (for example, EPR or deposit-return schemes) may not be equally effective in all parts of the world nor across Nature | Vol 626 | 1 February 2024【 】 | 53

Bold-system-change interventions

Global waste-collection infrastructure Remove global fossil-fuel subsidies

Intervention type Financial mechanism

Global EPR schemes Halt global fossil-fuel infrastructure Clean energy in plastic system

Market-led Standards and targets

Compostables and biodegradables at scale

Restrictions and prohibitions

Early adopter recycling infrastructure for new renewable plastics

Infrastructure

Green public procurement Financial support for scaling renewable plastics production Early adopter EPR schemes Recycling targets Eco-design regulations Collaborative global approach to plastics Deposit-return schemes Tax ‘unnecessary’ plastics R&D of renewable plastics Ban ‘unnecessary’ plastics Ban ‘harmful’ plastics Producer voluntary commitments Sustainable biomass production

Emissions (MtCO2e)

5,000 Production Carbon removal End of life

Business-as-usual emissions (CO2e)

2,500 Bold-system-change emissions (CO2e)

0

2020

2025

2030

2035

2040

2045

2050

Year Fig. 4 | Interventions roadmap for a bold-system-change scenario. Timeline of interventions to prevent the projected GHG emissions of a business-as-usual pathway and their potential to curb the carbon footprint of the future plastics

economy. The interventions all support the four interlinked targets to reach the life-cycle emissions of the bold-system-change scenario outlined in Fig. 2. Further scaling efforts for carbon removal to attain net zero are also shown.

all economic sectors. As also acknowledged in the UNEA resolution for a new plastics treaty, a wide range of approaches exist to address the full life cycle of plastics, which further highlights the need for enhanced international collaboration to facilitate access to technology, capacity building and scientific and technical cooperation33. Successful system change will require global collaboration, investment and social and technical capacity building. Realizing the bold system change is demanding and requires integrated technical, legal and economic interventions, from the modernization of current laws to consistent implementation of economic measures across different markets and jurisdictions, and ensuring that responsibility for plastics is shared by both upstream and downstream stakeholders. Further, these technical, legal and economic frameworks must be adaptable and responsive to future breakthroughs, where possible avoiding sector lock-ins. Although the transition facing the plastics sector is very important for the chemicals industry, it is accompanied by a marked restructuring of the global energy system, which is already set to achieve 98% of the increase in global energy demand using renewables by 2025 (ref. 153). The challenge is substantial but our perspective is that re-engineering the plastics economy is achievable. Ongoing international dialogue and coordinated legal and economic actions are therefore essential to deliver change in time, and the international plastics treaty being negotiated at present has a vital role to play in empowering these efforts at a global scale.

1.

54 | Nature | Vol 626 | 1 February 2024【 】

Global Plastics Outlook: Economic Drivers, Environmental Impacts and Policy Options (OECD, 2022). 2. Schneiderman, D. K. & Hillmyer, M. A. 50th anniversary perspective: there is a great future in sustainable polymers. Macromolecules 50, 3733–3749 (2017). 3. Ostle, C. et al. The rise in ocean plastics evidenced from a 60-year time series. Nat. Commun. 10, 1622 (2019). 4. Borrelle, S. B. et al. Predicted growth in plastic waste exceeds efforts to mitigate plastic pollution. Science 369, 1515–1518 (2020). 5. Lau, W. W. Y. et al. Evaluating scenarios toward zero plastic pollution. Science 369, 1455–1461 (2020). 6. Brahney, J., Hallerud, M., Heim, E., Hahnenberger, M. & Sukumaran, S. Plastic rain in protected areas of the United States. Science 368, 1257–1260 (2020). 7. MacLeod, M., Arp, H. P. H., Tekman, M. B. & Jahnke, A. The global threat from plastic pollution. Science 373, 61–65 (2021). 8. Santos, R. G., Machovsky-Capuska, G. E. & Andrades, R. Plastic ingestion as an evolutionary trap: toward a holistic understanding. Science 373, 56–60 (2021). 9. Tamargo, A. et al. PET microplastics affect human gut microbiota communities during simulated gastrointestinal digestion, first evidence of plausible polymer biodegradation during human digestion. Sci. Rep. 12, 528 (2022). 10. Bachmann, M. et al. Towards circular plastics within planetary boundaries. Nat. Sustain. 6, 599–610 (2023). Evaluation of various pathways for a circular plastics economy in view of nine planetary boundaries beyond GHG emissions. 11. Carus, M., Dammer, L., Raschka, A. & Skoczinski, P. Renewable carbon: key to a sustainable and future‐oriented chemical and plastic industry: definition, strategy, measures and potential. Greenh. Gases Sci. Technol. 10, 488–505 (2020). 12. Pires da Mata Costa, L. et al. Capture and reuse of carbon dioxide (CO2) for a plastics circular economy: a review. Processes 9, 759 (2021). 13. Bauer, F. et al. Plastics and climate change—breaking carbon lock-ins through three mitigation pathways. One Earth 5, 361–376 (2022). Potential risks and consequences of carbon lock-in at each stage of the plastics life cycle and their relationship with innovations and policy to change the plastic system.

14. Spierling, S. et al. Bio-based plastics – a review of environmental, social and economic impact assessments. J. Clean. Prod. 185, 476–491 (2018). 15. Cywar, R. M., Rorrer, N. A., Hoyt, C. B., Beckham, G. T. & Chen, E. Y. X. Bio-based polymers with performance-advantaged properties. Nat. Rev. Mater. 7, 83–103 (2021). 16. Rosenboom, J.-G., Langer, R. & Traverso, G. Bioplastics for a circular economy. Nat. Rev. Mater. 7, 117–137 (2022). Prospects of integrating bio-based plastics in a future circular economy, with a critical look at all the stages of the life cycle, as well as the current state of their market and policy implementation worldwide. 17. Law, K. L. & Narayan, R. Reducing environmental plastic pollution by designing polymer materials for managed end-of-life. Nat. Rev. Mater. 7, 104–116 (2021). 18. Jehanno, C. et al. Critical advances and future opportunities in upcycling commodity polymers. Nature 603, 803–814 (2022). 19. Mangold, H. & von Vacano, B. The frontier of plastics recycling: rethinking waste as a resource for high‐value applications. Macromol. Chem. Phys. 223, 2100488 (2022). 20. Jambeck, J. R. et al. Plastic waste inputs from land into the ocean. Science 347, 768–771 (2015). 21. Brooks, A. L., Wang, S. & Jambeck, J. R. The Chinese import ban and its impact on global plastic waste trade. Sci. Adv. 4, eaat0131 (2018). 22. Wang, C., Zhao, L., Lim, M. K., Chen, W.-Q. & Sutherland, J. W. Structure of the global plastic waste trade network and the impact of China’s import ban. Resour. Conserv. Recycl. 153, 104591 (2020). 23. van der Marel, E. R. Trading plastic waste in a global economy: soundly regulated by the Basel Convention? J. Environ. Law 34, 477–497 (2022). 24. Lebreton, L. C. M. et al. River plastic emissions to the world’s oceans. Nat. Commun. 8, 15611 (2017). 25. Horejs, C. Solutions to plastic pollution. Nat. Rev. Mater. 5, 641–641 (2020). 26. Rochman, C. M. et al. Classify plastic waste as hazardous. Nature 494, 169–171 (2013). 27. Jung, Y. S. et al. Characterization and regulation of microplastic pollution for protecting planetary and human health. Environ. Pollut. 315, 120442 (2022). 28. Fortman, D. J. et al. Approaches to sustainable and continually recyclable cross-linked polymers. ACS Sustain. Chem. Eng. 6, 11145–11159 (2018). 29. Coates, G. W. & Getzler, Y. D. Y. L. Chemical recycling to monomer for an ideal, circular polymer economy. Nat. Rev. Mater. 5, 501–516 (2020). Advances in chemical depolymerization to monomers from polymers to revalorize plastic waste. 30. Vollmer, I. et al. Beyond mechanical recycling: giving new life to plastic waste. Angew. Chem. Int. Ed. 59, 15402–15423 (2020). 31. Ellis, L. D. et al. Chemical and biological catalysis for plastics recycling and upcycling. Nat. Catal. 4, 539–556 (2021). Challenges and opportunities for catalysis development in chemical and biological recycling. 32. World Trade Organization (WTO), Committee on Trade and Environment. WTO Informal Dialogue on Plastics Pollution and Environmentally Sustainable Plastics Trade. WT/CTE/ W/250/Rev.2 (WTO, 2021). 33. United Nations Environment Assembly (UNEA) of the United Nations Environment Programme. End Plastic Pollution: Towards an International Legally Binding Instrument. UNEP/PP/OEWG/1/INF/1 (UNEA, 2022). 34. Stöfen-O’Brien, A. The prospects of an international treaty on plastic pollution. Int. J. Mar. Coast. Law 37, 727–740 (2022). 35. Simon, N. et al. A binding global agreement to address the life cycle of plastics. Science 373, 43–47 (2021). 36. Kirk, E. A. & Popattanachai, N. Marine plastics: fragmentation, effectiveness and legitimacy in international lawmaking. Rev. Eur. Comp. Int. Environ. Law 27, 222–233 (2018). 37. Raubenheimer, K., McIlgorm, A. & Oral, N. Towards an improved international framework to govern the life cycle of plastics. Rev. Eur. Comp. Int. Environ. Law 27, 210–221 (2018). 38. United Nations Environment Assembly (UNEA) of the United Nations Environment Programme. Combating Marine Plastic Litter and Microplastics: An Assessment of the Effectiveness of Relevant International, Regional and Subregional Governance Strategies and Approaches. UNEP/EA.3/INF/5 (UNEA, 2018). 39. Landrigan, P. J. et al. The Minderoo-Monaco Commission on Plastics and Human Health. Ann. Glob. Health 89, 23 (2023). 40. Geyer, R. in Plastic Waste and Recycling: Environmental Impact, Societal Issues, Prevention, and Solutions (ed. Letcher, T. M.) 13–32 (Academic, 2020). 41. Stubbins, A., Law, K. L., Muñoz, S. E., Bianchi, T. S. & Zhu, L. Plastics in the Earth system. Science 373, 51–55 (2021). 42. Zheng, J. & Suh, S. Strategies to reduce the global carbon footprint of plastics. Nat. Clim. Change 9, 374–378 (2019). The effect of implementing decarbonized energy, reduced demand, high recycling and bio-based plastics in reducing the carbon footprint of plastics. 43. Friedlingstein, P. et al. Global Carbon Budget 2022. Earth Syst. Sci. Data 14, 4811–4900 (2022). 44. Breaking the Plastic Wave: A Comprehensive Assessment of Pathways Towards Stopping Ocean Plastic Pollution (The Pew Charitable Trusts and SYSTEMIQ, 2020). 45. The New Plastics Economy: Rethinking the Future of Plastics (Ellen MacArthur Foundation, 2016). 46. The New Plastics Economy: Catalysing Action (Ellen MacArthur Foundation, 2017). 47. Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions: A New Circular Economy Action Plan for a Cleaner and More Competitive Europe COM(2020) 98 final (European Commission, 2020). 48. Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions: Closing the Loop – An EU Action Plan for the Circular Economy COM(2015) 614 final (European Commission, 2015).

49. Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions: A European Strategy for Plastics in a Circular Economy COM(2018) 28 final (European Commission, 2018). 50. Meys, R. et al. Achieving net-zero greenhouse gas emission plastics by a circular carbon economy. Science 374, 71–76 (2021). Life-cycle assessments of circular pathways for the plastics sector, focused on recycling, biomass and CCU, and aimed towards carbon-footprint reductions. 51. Stegmann, P., Daioglou, V., Londo, M., van Vuuren, D. P. & Junginger, M. Plastic futures and their CO2 emissions. Nature 612, 272–276 (2022). Cumulative projections of plastic production and emissions towards 2100 and the effects of implementing circular mitigation strategies. 52. Bank, M. S. & Hansson, S. V. The plastic cycle: a novel and holistic paradigm for the Anthropocene. Environ. Sci. Technol. 53, 7177–7179 (2019). 53. Zhu, X. The plastic cycle – an unknown branch of the carbon cycle. Front. Mar. Sci. 7, 1227 (2021). 54. Global Waste Management Outlook (UNEP, 2015). 55. Zimmerman, J. B., Anastas, P. T., Erythropel, H. C. & Leitner, W. Designing for a green chemistry future. Science 367, 397–400 (2020). 56. Sinha, V., Patel, M. R. & Patel, J. V. Pet waste management by chemical recycling: a review. J. Polym. Environ. 18, 8–25 (2008). 57. Lopez, G., Artetxe, M., Amutio, M., Bilbao, J. & Olazar, M. Thermochemical routes for the valorization of waste polyolefinic plastics to produce fuels and chemicals. A review. Renew. Sustain. Energy Rev. 73, 346–368 (2017). 58. Rahimi, A. & García, J. M. Chemical recycling of waste plastics for new materials production. Nat. Rev. Chem. 1, 0046 (2017). 59. Lopez, G. et al. Recent advances in the gasification of waste plastics. A critical overview. Renew. Sustain. Energy Rev. 82, 576–596 (2018). 60. Schyns, Z. O. G. & Shaver, M. P. Mechanical recycling of packaging plastics: a review. Macromol. Rapid Commun. 42, 2000415 (2021). 61. Ghosh, K. & Jones, B. H. Roadmap to biodegradable plastics—current state and research needs. ACS Sustain. Chem. Eng. 9, 6170–6187 (2021). Analysis of current biodegradable plastics and description of future focus areas for development. 62. Zhang, X., Fevre, M., Jones, G. O. & Waymouth, R. M. Catalysis as an enabling science for sustainable polymers. Chem. Rev. 118, 839–885 (2017). 63. Sun, Z., Fridrich, B., de Santi, A., Elangovan, S. & Barta, K. Bright side of lignin depolymerization: toward new platform chemicals. Chem. Rev. 118, 614–678 (2018). Review on the catalytic processes to obtain chemicals and monomers from lignin. 64. Grignard, B., Gennen, S., Jérôme, C., Kleij, A. W. & Detrembleur, C. Advances in the use of CO2 as a renewable feedstock for the synthesis of polymers. Chem. Soc. Rev. 48, 4466–4514 (2019). Catalytic pathways for the chemical conversion of CO2 into polymers. 65. O’Dea, R. M., Willie, J. A. & Epps, T. H. 100th anniversary of macromolecular science viewpoint: polymers from lignocellulosic biomass. Current challenges and future opportunities. ACS Macro Lett. 9, 476–493 (2020). 66. Hepburn, C. et al. The technological and economic prospects for CO2 utilization and removal. Nature 575, 87–97 (2019). 67. Smith, S. M. et al. The State of Carbon Dioxide Removal 1st edn (The State of Carbon Dioxide Removal, 2023). 68. Centi, G., Quadrelli, E. A. & Perathoner, S. Catalysis for CO2 conversion: a key technology for rapid introduction of renewable energy in the value chain of chemical industries. Energy Environ. Sci. 6, 1711–1731 (2013). 69. Sternberg, A., Jens, C. M. & Bardow, A. Life cycle assessment of CO2-based C1-chemicals. Green Chem. 19, 2244–2259 (2017). 70. Zhong, J. et al. State of the art and perspectives in heterogeneous catalysis of CO2 hydrogenation to methanol. Chem. Soc. Rev. 49, 1385–1413 (2020). 71. Kaiser, S. & Bringezu, S. Use of carbon dioxide as raw material to close the carbon cycle for the German chemical and polymer industries. J. Clean. Prod. 271, 122775 (2020). 72. Schirmeister, C. G. & Mülhaupt, R. Closing the carbon loop in the circular plastics economy. Macromol. Rapid Commun. 43, 2200247 (2022). 73. Steffen, W. et al. Planetary boundaries: guiding human development on a changing planet. Science 347, 1259855 (2015). 74. Tulus, V., Pérez-Ramírez, J. & Guillén-Gosálbez, G. Planetary metrics for the absolute environmental sustainability assessment of chemicals. Green Chem. 23, 9881–9893 (2021). 75. Rockström, J. et al. Safe and just Earth system boundaries. Nature 619, 102–111 (2023). 76. Raubenheimer, K. & Urho, N. Possible elements of a new global agreement to prevent plastic pollution. Nordic Council of Ministers https://www.nordicreport2020.com/ (2020). 77. Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions: EU Policy Framework on Biobased, Biodegradable and Compostable Plastics COM(2022) 682 final (European Commission, 2022). 78. Brizga, J., Hubacek, K. & Feng, K. The unintended side effects of bioplastics: carbon, land, and water footprints. One Earth 3, 45–53 (2020). 79. Nicholson, S. R., Rorrer, N. A., Carpenter, A. C. & Beckham, G. T. Manufacturing energy and greenhouse gas emissions associated with plastics consumption. Joule 5, 673–686 (2021). 80. Miller, S. A. Five misperceptions surrounding the environmental impacts of single-use plastic. Environ. Sci. Technol. 54, 14143–14151 (2020). 81. Ragaert, K., Delva, L. & Van Geem, K. Mechanical and chemical recycling of solid plastic waste. Waste Manage. 69, 24–58 (2017). 82. Stram, B. N. Key challenges to expanding renewable energy. Energy Policy 96, 728–734 (2016).

Nature | Vol 626 | 1 February 2024【 】 | 55

83.

84. 85.

86. 87. 88.

89.

90. 91. 92. 93. 94. 95. 96. 97. 98. 99. 100.

101.

102.

103. 104.

105.

106. 107. 108. 109.

110. 111. 112. 113.

114. 115.

116.

117.

Posen, I. D., Jaramillo, P., Landis, A. E. & Griffin, W. M. Greenhouse gas mitigation for U.S. plastics production: energy first, feedstocks later. Environ. Res. Lett. 12, 034024 (2017). Stockholm+50: A Healthy Planet for the Prosperity of All–Our Responsibility, Our Opportunity A/CONF.238/9 (United Nations, 2022). Barrowclough, D., Birkbeck, C. D. & Christen, J. Global trade in plastics: insights from the first life-cycle trade database. UNCTAD Research Paper No. 53, UNCTAD/SER.RP/2020/12 (United Nations Conference on Trade and Development, 2020). Chiroleu-Assouline, M. & Fodha, M. From regressive pollution taxes to progressive environmental tax reforms. Eur. Econ. Rev. 69, 126–142 (2014). Kirk, E., Popattanachai, N., van der Marel, E. R. & Barnes, R. Research Handbook on Plastics Regulation: Law, Policy and the Environment (Edward Elgar, 2024【 】). Communication from the Commission to the European Parliament, the European Council, the Council, the European Economic and Social Committee and the Committee of the Regions: The European Green Deal COM(2019) 640 final (European Commission, 2019). The European Parliament and the Council of the European Union. Regulation (EU) 2021/1119 of the European Parliament and of the Council of 30 June 2021 establishing the framework for achieving climate neutrality and amending Regulations (EC) No 401/2009 and (EU) 2018/1999 (‘European Climate Law’) (European Commission, 2021). Ayeleru, O. O. et al. Challenges of plastic waste generation and management in sub-Saharan Africa: a review. Waste Manage. 110, 24–42 (2020). Velis, C. A. Circular economy and global secondary material supply chains. Waste Manag. Res. 33, 389–391 (2015). Transforming Our World: The 2030 Agenda for Sustainable Development A/RES/70/1 (UNGA, 2015). Bocken, N. M. P. & Short, S. W. in Handbook of the Circular Economy 250–265 (Edward Elgar, 2020). Mederake, L. Without a debate on sufficiency, a circular plastics economy will remain an illusion. Circ. Econ. Sustain. 3, 1425–1439 (2023). Murakami, S., Oguchi, M., Tasaki, T., Daigo, I. & Hashimoto, S. Lifespan of commodities, part I. The creation of a database and its review. J. Ind. Ecol. 14, 598–612 (2010). Leal, J. M., Pompidou, S., Charbuillet, C. & Perry, N. Product recoverability: a review of assessment methods. Procedia CIRP 69, 710–715 (2018). Geyer, R., Jambec, J. R. & Law, K. L. Production, use, and fate of all plastics ever made. Sci. Adv 3, e1700782 (2017). Fagnani, D. E. et al. 100th anniversary of macromolecular science viewpoint: redefining sustainable polymers. ACS Macro Lett. 10, 41–53 (2020). Haque, F. M. et al. Defining the macromolecules of tomorrow through synergistic sustainable polymer research. Chem. Rev. 122, 6322–6373 (2022). The Gazette of Meghalaya. The Meghalaya Prohibition of Manufacture, Sale, Use and Throwing of Low Density Plastic Bags Act, 2001 (Act No. 4 of 2001) (Meghalaya Legislative Assembly, 2001). Regulation Gazette. Regulation 543 under Section 24 (d) of the Environmental Conservation Act (Act No. 73 of 1989). No. 7348 Vol. 443 (South Africa Department of Environmental Affairs and Tourism, 2002). Gazette of Eritrean Laws: Proclamations and Legal Notices. Regulations to Prohibit the Production, Sale or Distribution of Plastic Bags in Eritrea (Legal Notice No. 63/2002). Vol. 11/2002 No. 7 (The Government of Eritrea, 2002). Bangladesh Gazette. The Bangladesh Environment Conservation Act, 1995 Act No. 1 of 1995, Sec. 6a, as amended by Sec. 5 of Act 9 of 2002 (The Government of Bangladesh, 2002). The European Parliament and the Council of the European Union. Directive (EU) 2019/904 of the European Parliament and of the Council of 5 June 2019 on the reduction of the impact of certain plastic products on the environment. L155/1 (European Commission, 2019). United Nations Environment Programme (UNEP). Potential options for elements towards an international legally binding instrument, based on a comprehensive approach that addresses the full life cycle of plastics as called for by United Nations Environment Assembly resolution 5/14. UNEP/PP/INC.2/4 (UNEP, 2023). United Nations. Stockholm Convention on Persistent Organic Pollutants. Treaty Series, Vol. 2256, p. 119 (United Nations, 2001). United Nations. Montreal Protocol on Substances that Deplete the Ozone Layer. Treaty Series, Vol. 1522, p. 3 (United Nations, 1987). United States Congress. Microbead-Free Waters Act of 2015. Public Law 114–114 (U.S. Congress, 2015). The European Parliament and the Council of the European Union. Commission Regulation (EU) 2023/2055 of 25 September 2023 amending Annex XVII to Regulation (EC) No 1907/2006 of the European Parliament and of the Council concerning the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH) as regards synthetic polymer microparticle. L238/67 (European Commission, 2023). ReShaping Plastics: Pathways to a Circular, Climate Neutral Plastics System in Europe (SYSTEMIQ, 2022). Desalegn, G. & Tangl, A. Banning vs taxing, reviewing the potential opportunities and challenges of plastic products. Sustainability 14, 7189 (2022). Klenert, D. et al. Making carbon pricing work for citizens. Nat. Clim. Change 8, 669–677 (2018). European Commission. Communication from the Commission to the European Parliament and the Council: Sustainable Carbon Cycles. COM(2021) 800 (European Commission, 2021). Bergsma, G., Broeren, M., Schep, E. & Warringa, G. Mandatory percentage of recycled or bio-based plastic in the European Union (CE Delft, 2022). The White House. Executive Order 14081: Advancing Biotechnology and Biomanufacturing Innovation for a Sustainable, Safe, and Secure American Bioeconomy (The White House, 2022). The White House Office of Science and Technology Policy. Bold Goals for U.S. Biotechnology and Biomanufacturing. Harnessing Research and Development to Further Societal Goals (The White House, 2023). European Commission. Art. 8 of the Proposal for a Regulation of the European Parliament and of the Council on packaging and packaging waste, amending Regulation (EU)

56 | Nature | Vol 626 | 1 February 2024         

2019/1020 and Directive (EU) 2019/904, and repealing Directive 94/62/EC, 30 November 2022 (European Commission, 2022). 118. Watkins, E., Schweitzer, J.-P., Leinala, E. & Börkey, P. Policy approaches to incentivise sustainable plastic design. OECD Environment Working Papers 149. OECD Publishing https://doi.org/10.1787/233ac351-en (2019). Contextualized examples of policies designed to enhance the prospects of a sustainable plastic economy. 119. European Commission. Public procurement. https://single-market-economy.ec.europa.eu/ single-market/public-procurement_en (European Commission, 2023). 120. Department for Business, Energy and Industrial Strategy, United Kingdom. Evaluation of the Contracts for Difference Scheme. Phase 1: Allocation Rounds 1 & 2. Final Report (Department for Business, Energy and Industrial Strategy, 2019). 121. Skovgaard, J. & van Asselt, H. The politics of fossil fuel subsidies and their reform: Implications for climate change mitigation. Wiley Interdiscip. Rev. Clim. Change 10, e581 (2019). 122. Döhler, N., Wellenreuther, C. & Wolf, A. Market Dynamics of Biodegradable Bio-based Plastics: Projections and Linkages to European Policies (Hamburg Institute of International Economics, 2020). 123. Barrowclough, D. & Birkbeck, C. Transforming the global plastics economy: the role of economic policies in the global governance of plastic pollution. Soc. Sci. 11, 26 (2022). 124. United Nations Framework Convention on Climate Change. Decision 1/CMA.3 Glasgow Climate Pact. Report of the Conference of the Parties serving as the meeting of the Parties to the Paris Agreement on its third session, held in Glasgow from 31 October to 13 November 2021. FCCC/PA/CMA/2021/10/Add.1 (United Nations, 2021). 125. World Trade Organization. Ministerial Statement on Fossil Fuel Subsidies. WT/MIN(21)/9/ Rev.2 (World Trade Organization, 2022). 126. The European Parliament and the Council of the European Union. Directive (EU) 2023/959 of the European Parliament and of the Council of 10 May 2023 amending Directive 2003/87/EC establishing a system for greenhouse gas emission allowance trading within the Union and Decision (EU) 2015/1814 concerning the establishment and operation of a market stability reserve for the Union greenhouse gas emission trading system. L130/134 (European Commission, 2023). 127. The Future of Petrochemicals: Towards More Sustainable Plastics and Fertilisers (IEA, 2018). A highlight of the petrochemical sector becoming the largest driver of global oil demand and its continuous investments in the production of fossil-fuel-based plastics. 128. Bauer, F. & Fontenit, G. Plastic dinosaurs – digging deep into the accelerating carbon lock-in of plastics. Energy Policy 156, 112418 (2021). 129. European Committee for Standardization (CEN). CEN/TC 411 - Bio-based products. https:// standards.cencenelec.eu/dyn/www/f?p=CEN:6 (CEN, 2021). 130. United Nations. Basel Convention on the Control of Transboundary Movements of Hazardous Wastes and their Disposal. Treaty Series, Vol. 1673, p. 57 (United Nations, 1989). 131. United Nations Environment Programme (UNEP). Technical guidelines on the environmentally sound management of plastic wastes. UNEP/CHW.16/6/Add.3/Rev.1 (UNEP, 2023). 132. The European Parliament and the Council of the European Union. Directive (EU) 2018/852 of 30 May 2018 amending Directive 94/62/EC on packaging and packaging waste. L150/141 (European Commission, 2018). 133. Raubenheimer, K. & Urho, N. Rethinking global governance of plastics – the role of industry. Mar. Policy 113, 103802 (2020). 134. Extended Producer Responsibility: A Necessary Part of the Solution to Packaging Waste and Pollution (Ellen MacArthur Foundation, 2021). 135. Dalhammar, C. in Elgar Encyclopedia of Environmental Law Ch. VI.16 (ed. Faure, M.) (2018). 136. Sachs, N. Planning the funeral at the birth: extended producer responsibility in the European Union and the United States. Harv. Envtl. L. Rev. 30, 51 (2006). 137. Extended Producer Responsibility (OECD, 2016). 138. United Nations Environment Programme (UNEP). Revised Draft Practical manual on Extended Producer Responsibility. UNEP/CHW.14/5/Add.1 (UNEP, 2019). 139. United Nations Environment Programme (UNEP). Report of the Conference of the Parties to the Basel Convention on the Control of Transboundary Movements of Hazardous Wastes and Their Disposal on the work of its fourteenth meeting. UNEP/CHW.14/28 (UNEP, 2019). 140. WRAP and the UK Plastics Pact. A Roadmap to 2025: The UK Plastics Pact https://wrap.org. uk/resources/guide/roadmap-2025-uk-plastics-pact (WRAP, 2022). 141. Colorado Department of Revenue. Plastic recycling investment tax credit. https://tax. colorado.gov/sites/tax/files/Income56.pdf (1989). 142. Larrain, M., Billen, P. & Van Passel, S. The effect of plastic packaging recycling policy interventions as a complement to extended producer responsibility schemes: a partial equilibrium model. Waste Manage. 153, 355–366 (2022). 143. Gao, W., Hundertmark, T., Pais, G., Ryba, A., & Wallach, J. Addressing the challenges of plastic waste: circularity and leakage. McKinsey https://www.mckinsey.com/industries/ chemicals/our-insights/addressing-the-challenges-of-plastic-waste-circularity-andleakage#/ (2022). 144. Gall, M., Wiener, M., Chagas de Oliveira, C., Lang, R. W. & Hansen, E. G. Building a circular plastics economy with informal waste pickers: recyclate quality, business model, and societal impacts. Resour. Conserv. Recycl. 156, 104685 (2020). 145. Alaranta, J. & Turunen, T. How to reach a safe circular economy?—Perspectives on reconciling the waste, product and chemicals regulation. J. Environ. Law 33, 113–136 (2021). 146. European Commission. Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions on the implementation of the circular economy package: options to address the interface between chemical, product and waste legislation. COM(2018) 32 final. (European Commission, 2018). 147. Schwarz, A. E., Lensen, S. M. C., Langeveld, E., Parker, L. A. & Urbanus, J. H. Plastics in the global environment assessed through material flow analysis, degradation and environmental transportation. Sci. Total Environ. 875, 162644 (2023). 148. Hahladakis, J. N., Velis, C. A., Weber, R., Iacovidou, E. & Purnell, P. An overview of chemical additives present in plastics: migration, release, fate and environmental impact during their use, disposal and recycling. J. Hazard. Mater. 344, 179–199 (2018).

149. European Commission. Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions. Chemicals Strategy for Sustainability Towards a Toxic-Free Environment. COM(2020) 667 final (European Commission, 2020). 150. United Nations. Cartagena Protocol on Biosafety to the Convention on Biological Diversity. Treaty Series, Vol. 2226, p. 208 (United Nations, 2000). 151. Basel Convention, Rotterdam Convention, and Stockholm Convention. Joint clearing house mechanism for information exchange: revised draft strategy. UNEP/CHW.13/1-UNEP/ FAO/RC/COP.8/1-UNEP/POPS/COP.8/1 (Basel Convention, Rotterdam Convention, and Stockholm Convention, 2017). 152. European Commission. Proposal for an EU Regulation on Ecodesign for Sustainable Products. COM(2022)142 final (European Commission, 2022). 153. CO2 Emissions in 2022 https://www.iea.org/reports/co2-emissions-in-2022 (IEA, 2023). Acknowledgements The Oxford Martin School (‘Future of Plastics’, C.K.W., all authors), European Union Horizon 2020 research and innovation programme (Marie Skłodowska-Curie no. 101018516) (F.V.), EPSRC (EP/S018603/1; EP/R027129/1; EP/V038117/1) (C.K.W.) and Research England (iCAST, RED, RE-P-2020-04) (C.K.W.) are acknowledged for funding. Author contributions C.K.W., C.H. and C.R. conceived the idea for the paper. F.V., E.R.v.d.M., R.W.F.K., C. McElroy and N.S. conducted the research and led the writing of the manuscript,

with further contributions from all the other authors. All authors approved submission of the article for publication. 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/s41586-023-06939-z. Correspondence and requests for materials should be addressed to Cameron Hepburn, Catherine Redgwell or Charlotte K. Williams. Peer review information Nature thanks Anna Schwarz 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 | 1 February 2024【 】 | 57

Article

Logical quantum processor based on reconfigurable atom arrays ­­­­­

https://doi.org/10.1038/s41586-023-06927-3 Received: 21 October 2023 Accepted: 1 December 2023 Published online: 6 December 2023 Open access Check for updates

Dolev Bluvstein1, Simon J. Evered1, Alexandra A. Geim1, Sophie H. Li1, Hengyun Zhou1,2, Tom Manovitz1, Sepehr Ebadi1, Madelyn Cain1, Marcin Kalinowski1, Dominik Hangleiter3, J. Pablo Bonilla Ataides1, Nishad Maskara1, Iris Cong1, Xun Gao1, Pedro Sales Rodriguez2, Thomas Karolyshyn2, Giulia Semeghini4, Michael J. Gullans3, Markus Greiner1, Vladan Vuletić5 & Mikhail D. Lukin1 ✉

Suppressing errors is the central challenge for useful quantum computing1, requiring quantum error correction (QEC)2–6 for large-scale processing. However, the overhead in the realization of error-corrected ‘logical’ qubits, in which information is encoded across many physical qubits for redundancy2–4, poses substantial challenges to large-scale logical quantum computing. Here we report the realization of a programmable quantum processor based on encoded logical qubits operating with up to 280 physical qubits. Using logical-level control and a zoned architecture in reconfigurable neutral-atom arrays7, our system combines high two-qubit gate fidelities8, arbitrary connectivity7,9, as well as fully programmable single-qubit rotations and mid-circuit readout10–15. Operating this logical processor with various types of encoding, we demonstrate improvement of a two-qubit logic gate by scaling surface-code6 distance from d = 3 to d = 7, preparation of colour-code qubits with break-even fidelities5, fault-tolerant creation of logical Greenberger– Horne–Zeilinger (GHZ) states and feedforward entanglement teleportation, as well as operation of 40 colour-code qubits. Finally, using 3D [[8,3,2]] code blocks16,17, we realize computationally complex sampling circuits18 with up to 48 logical qubits entangled with hypercube connectivity19 with 228 logical two-qubit gates and 48 logical CCZ gates20. We find that this logical encoding substantially improves algorithmic performance with error detection, outperforming physical-qubit fidelities at both cross-entropy benchmarking and quantum simulations of fast scrambling21,22. These results herald the advent of early error-corrected quantum computation and chart a path towards large-scale logical processors.

Quantum computers have the potential to substantially outperform their classical counterparts for solving certain problems1. However, executing large-scale, useful algorithms on quantum processors requires very low gate error rates (generally below about 10−10)23, far below those that will probably ever be achievable with any physical device2. The landmark development of QEC theory provides a conceptual solution to this challenge2–4. The key idea is to use entanglement to delocalize a logical qubit degree of freedom across many redundant physical qubits, such that, if any given physical qubit fails, it does not corrupt the underlying logical information. In principle, with sufficiently low physical error rates and sufficiently many qubits, a logical qubit can be made to operate with extremely high fidelity, providing a path to realizing large-scale algorithms4. However, in practice, useful QEC poses many challenges, ranging from large overhead in physical qubit numbers23 to highly complex gate operations between the delocalized logical degrees of freedom24.

Recent experiments have achieved milestone demonstrations of two logical qubits and one entangling gate5,6 and explorations of new encodings25–28. One specific challenge for realizing large-scale logical processors involves efficient control. Unlike modern classical processors that can efficiently access and manipulate many bits of information29, quantum devices are typically built such that each physical qubit requires several classical control lines. Although suitable for the implementation of physical qubit processors, this approach poses a substantial obstacle to the control of logical qubits redundantly encoded over many physical qubits. Here we describe the realization of a programmable quantum processor based on hardware-efficient control over logical qubits in reconfigurable neutral-atom arrays7. We use this logical processor to demonstrate key building blocks of QEC and realize programmable logical algorithms. In particular, we explore important features of

Department of Physics, Harvard University, Cambridge, MA, USA. 2QuEra Computing Inc., Boston, MA, USA. 3Joint Center for Quantum Information and Computer Science, NIST/University of

1

Maryland, College Park, MD, USA. 4John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA. 5Department of Physics and Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, MA, USA. ✉e-mail: [email protected]

58 | Nature | Vol 626 | 1 February 2024【 】

b

0

0

10

c Rydberg laser

FPGA

Camera Local imaging

50

60

0

20 40 Time (μs)

Feedforward

Mid-circuit measurement

0.5

0

Logical qubit decoding

0

With local imaging No local imaging

1.0

AOD Syndrome extraction

20 30 40 Time (μs)

Qubit state

Entangling zone

40

Qubit state

Logical 1Q gate

Readout zone

Neighbour

Qubit state

Addressed site

1

Logical 2Q gate

1

80

Site index

Ancilla qubit reservoir

Storage zone

Logical qubit storage

Number of events

a

0

0.5 Ramsey phase (2π)

1.0

1,000 500 0 1,750

2,000 2,250 Camera counts

Fig. 1 | A programmable logical processor based on reconfigurable atom arrays. a, Schematic of the logical processor, split into three zones: storage, entangling and readout (see Extended Data Fig. 1 for detailed layout). Logical single-qubit and two-qubit operations are realized transversally with efficient, parallel operations. Transversal CNOTs are realized by interlacing two logical qubit grids and performing a single global entangling pulse that excites atoms to Rydberg states. Physical qubits are encoded in hyperfine ground states of 87 Rb atoms trapped in optical tweezers. b, Fully programmable single-qubit

rotations are implemented using Raman excitation through a 2D AOD; parallel grid illumination delivers the same instruction to multiple atomic qubits. c, Mid-circuit readout and feedforward. The imaging histogram shows high-fidelity state discrimination (500 μs imaging time, readout fidelity approximately 99.8%; Methods) and the Ramsey fringe shows that qubit coherence is unaffected by measuring other qubits in the readout zone (error probability p ≈ 10 −3; Methods). The FPGA performs real-time image processing, state decoding and feedforward (Fig. 4).

logical operations and circuits, including scaling to large codes, fault tolerance and complex non-Clifford circuits.

large, dynamically programmable grids of light. Fully programmable local single-qubit rotations are realized through qubit-specific, parallel Raman excitation through an additional 2D AOD (ref. 34) (Fig. 1b and Extended Data Fig. 2). Mid-circuit readout is enabled by moving selected qubits about 100 μm away to a readout zone and illuminating with a focused imaging beam7,35, resulting in high-fidelity imaging, as well as negligible decoherence on stored qubits (Fig. 1c and Extended Data Fig. 3). The mid-circuit10–15 image is collected with a CMOS camera and sent to a field-programmable gate array (FPGA) for real-time decoding and feedforward. The central aspect of our logical processor is the control of individual logical qubits as the fundamental units, instead of individual physical qubits. To this end, we observe that, during most error-corrected operations, the physical qubits of a logical block are supposed to realize the same operation, and this instruction can be delivered in parallel with only a few control lines. This approach naturally multiplexes with optical techniques. For example, to realize a logical single-qubit gate2, we use the Raman 2D AOD (Fig. 1b) to create a grid of light beams and simultaneously illuminate the physical qubits of the logical block with the same instruction. Such a gate is transversal2, meaning that operations act on physical qubits of the code block independently. This transversal property further implies that the gate is inherently fault-tolerant2, meaning that errors cannot spread within the code block (see Methods), thereby preventing a physical error from spreading into a logical fault. Crucially, a similar approach can realize logical entangling gates2,4. Specifically, we use the grids generated by our moving 2D AOD to pick up two logical qubits, interlace them in the entangling

Logical processor based on atom arrays Our logical processor architecture, illustrated in Fig. 1a, is split into three zones (see also Extended Data Fig. 1). The storage zone is used for dense qubit storage, free from entangling-gate errors and featuring long coherence times. The entangling zone is used for parallel logical qubit encoding, stabilizer measurements and logical gate operations. Finally, the readout zone enables mid-circuit readout of desired logical or physical qubits, without disturbing the coherence of the computation qubits still in operation. This architecture is implemented using arrays of individual 87Rb atoms trapped in optical tweezers, which can be dynamically reconfigured in the middle of the computation while preserving qubit coherence7,9. Our experiments make use of the apparatus described previously in refs. 7,8,30, with key upgrades enabling universal digital operation. Physical qubits are encoded in clock states within the ground-state hyperfine manifold (T2 > 1s (ref. 7)) and stored in optical tweezer arrays created by a spatial light modulator (SLM)30,31. We use systems of up to 280 atomic qubits, combining high-fidelity two-qubit gates8, enabled by fast excitation into atomic Rydberg states interacting through robust Rydberg blockade32, with arbitrary connectivity enabled by atom transport by means of 2D acousto-optic deflectors (AODs)7. Central to our approach of scalable control, AODs10–15,31,33 use frequency multiplexing to take in just two voltage waveforms (one for each axis) to create

Nature | Vol 626 | 1 February 2024【 】 | 59

Article a

b Physical error propagation

|+L〉 |0L〉 |+L〉

X/Z

|\1L〉

X/Z

|\2L〉

|0L〉

X

X X

Control stabilizers

Target stabilizers

1. Encode logical qubits ?

2. Parallel move (interlace)

Hyperedge

c

XX

3. Pulse global laser (transversal CNOT)

CNOT

0

Logical Bell-pair error Logical Bell-pair error

d

Number of physical qubits per Bell pair 0.4

26

74

d=7

ZZ

Probability

0.5

++ +– –+ –– 00 01 10 11 Logical bitstring

146

d = 3 (four pairs)

0.3 d=5

Conventional decoding

0.2 0.11

Correlated decoding d=7

ZL

0.09

0.07 3

5 Surface-code distance d

7

Data Ancilla

XL

X stabilizer Z stabilizer

Fig. 2 | Transversal entangling gates between two surface codes. a, Illustration of transversal CNOT between two d = 7 surface codes based on parallel atom transport. b, The concept of correlated decoding. Physical errors propagate between physical qubit pairs during transversal CNOT gates, creating correlations that can be used for improved decoding. We account for these correlations, arising from deterministic error propagation (as opposed to correlated error events), by adding edges and hyperedges that connect the decoding graphs of the two logical qubits. c, Populations of entangled d = 7 surface codes measured in the XX and ZZ basis. d, Measured Bell-pair error as a function of code distance, for both conventional (top) and correlated (bottom) decoding. We estimate Bell error with the average of the ZZ populations and the XX parities (Methods). To reduce code distance, we simply remove selected atoms from the grid, as shown on the right, ensuring unchanged experimental conditions (for d = 3, four logical Bell pairs are generated in parallel). Error bars represent the standard error of the mean. See Extended Data Figs. 4 and 5 for further surface-code data.

zone and then pulse our single global Rydberg excitation laser to realize a physical entangling gate on each twin pair of the blocks (Figs. 1a and 2a). This process realizes a high-fidelity, fault-tolerant transversal CNOT in a single parallel step.

Improving entangling gates with code distance A key property of QEC codes is that, for error rates below some threshold, the performance should improve with system size, associated with a so-called code distance4,24. Recently, this property has been experimentally verified by reducing idling errors of a code6. Neutral-atom qubits can be idly stored for long times with low errors, and the central 60 | Nature | Vol 626 | 1 February 2024【淘宝唯一：艾米学社】【 】

challenge is to improve entangling operations with code distance. Thus motivated, we realize a transversal CNOT gate using logical qubits encoded in two surface codes (Fig. 2). Surface codes have stabilizers that are used for detecting and correcting errors without disrupting the logical state4,24. The stabilizers form a 2D lattice of four-body plaquettes of X and Z operators, which commute with the XL (ZL) logical operators that run horizontally (vertically) along the lattice (Fig. 2d). By measuring stabilizers, we can detect the presence of physical qubit errors, decode (infer what error occurred) and correct the error simply by applying a software ZL/XL correction24. Such a code can detect and correct a certain number of errors determined by the linear dimension of the system (the code distance d). To test the performance of our logical entangling gate, we first initialize the logical qubits by preparing physical qubits of two blocks in |+⟩ and |0⟩ states, respectively, and performing a single round of stabilizer measurements with parallel operations7. Although this state preparation is non-fault-tolerant (nFT) beyond d = 3, we are still able to study error suppression of the transversal CNOT (Methods). Specifically, we prepare the two logicals in state |+L⟩ and |0L⟩, perform the transversal CNOT and then projectively measure to evaluate the logical Bell-state stabilizers XL1XL2 and ZL1ZL2 (Fig. 2c). For decoding and correcting the logical state, we observe that there are strong correlations between the stabilizers of the two blocks (Extended Data Figs. 4 and 5) owing to propagation of physical errors between the codes during the transversal CNOT (ref. 36) (Fig. 2b). We use these correlations to improve performance by decoding the logical qubits jointly, realized by a joint decoding graph that includes edges and hyperedges connecting the stabilizers of the two logical qubits (Fig. 2b, Methods). Using this correlated decoding procedure, we measure roughly 0.95 populations in the XLXL and ZLZL bases (Fig. 2c), showing entanglement between the d = 7 logical qubits. Studying the performance as a function of code size (Fig. 2d) reveals that the logical Bell pair improves with larger code distance, demonstrating improvement of the entangling operation. By contrast, we note that, when conventional decoding, that is, independent minimum-weight perfect matching within both codes4, is used, the fidelity decreases with code distance. This is in part because of the nFT state preparation, whose effect is partially mitigated by the correlated decoding (Methods). We emphasize that, although these results demonstrate surpassing an effective threshold for the entire circuit (implying that we surpass the threshold of the transversal CNOT), such a threshold is higher owing to projective readout after the transversal CNOT. In practice, the transversal CNOT should be used in combination with many repeated rounds of noisy syndrome extraction6, which is expected to have a lower threshold and is an important goal for future research.

Fault-tolerant logical algorithms All logical algorithms we perform in this work are built from transversal gates, which are intrinsically fault-tolerant2. We now also use fault-tolerant state preparation to explore programmable logical algorithms. We use 2D d = 3 colour codes3,37, which are topological codes akin to the surface code, but with the useful capability of transversal operations of the full Clifford group: Hadamard (H), π/2 phase (S) gate and CNOT (ref. 37). This transversal gate set can realize any Clifford circuit fault-tolerantly. As a test case, here we create a logical GHZ state. Figure 3a shows the implementation of a ten-logical-qubit algorithm, in which all ten qubits are first encoded by a nFT encoding circuit (Methods). Then, five of the codes are used as ancilla logicals, performing parallel transversal CNOTs to fault-tolerantly detect errors on the computation logicals38, and are then moved into the storage zone, in which they are safely kept. Subsequently, four computation logicals are used to prepare the GHZ state and logical Clifford rotations

Fault-tolerant GHZ circuit |0L〉

3

1 Physical + + + + + + +

6

C

|0L〉

C

|0L〉 |0L〉

H H H

|0L〉 |0L〉 |0L〉

H

Real(Ulogical) Increasing error detection

0.9

0.6 0.4

0.8 0.7

0.2 0.6 nFT

FT

EDFT

0.001

Logical nFT

1.0

0.8

0.010

e

d

1.0

0

C

H

|0L〉 |0L〉

GHZ fidelity

Logical GHZ state fidelity

c

5 7

|0L〉

X/Z stabilizers

4

Move to storage

X

L /Y L /Z L

2

b C

SPAM infidelity

a Encoding circuit

FT 1 2 Max. flag 3 errors 4 0.2

0.4 0.6 0.8 Accepted fraction

Imag( Ulogical)

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0000 0.1

0 0000 1.0

1111

Physical Logical FT atom (with CNOT)

1111

0 0000

0000

1111

1111

Fig. 3 | Fault-tolerant logical algorithms. a, Circuit for preparation of logical GHZ state. Ten colour codes are encoded non-fault-tolerantly and then parallel transversal CNOTs between computation and ancilla logical qubits perform fault-tolerant initialization. The ancilla logical qubits are moved to storage and a four-logical-qubit GHZ state is created between the computation qubits. Logical Clifford operations are applied before readout to examine the GHZ state. b, SPAM infidelity of the logical qubits without (nFT) and with (FT) the transversal-CNOT-based flagged preparation, compared with physical

qubit SPAM. c, Logical GHZ fidelity without postselecting on flags (nFT), postselecting on flags (FT) and postselecting on flags and stabilizers of the computation logical qubits, corresponding to error detection (EDFT). d, GHZ fidelity as a function of sliding-scale error-detection threshold (converted into the probability of accepted repetitions) and of the number of successful flags in the circuit. e, Density matrix of the four-logical-qubit GHZ state (with at most three flag errors) measured by means of full-state tomography involving all 256 logical Pauli strings.

are used at the end of the circuit for direct fidelity estimation39 and full logical state tomography. We first benchmark our state initialization5,40,41 (Fig. 3b). Averaged over the five computation logicals, we find that, by using the fault-tolerant initialization (postselecting on the ancilla logical flag not detecting errors) our |0L⟩ initialization fidelity is 99.91+0.04 −0.09% , exceeding both our physical qubit |0⟩ initialization fidelity (99.32(4)% (ref. 8)) and physical two-qubit gate fidelity (99.5% (ref. 8)). Then, Fig. 3c shows that the resulting GHZ state fidelity obtained using the fault-tolerant algorithm is 72(2)% (again using correlated decoding), demonstrating genuine multipartite entanglement. Furthermore, we can postselect on all stabilizers of our computation logicals being correct; using this error-detection approach, the GHZ fidelity increases to 99.85+0.1 −1.0 %, at the cost of postselection overhead. Because not all nontrivial syndromes are equally likely to cause algorithmic failure, we can perform a partial postselection, in which syndrome events most likely to have caused algorithmic failure are discarded, given by the weight of the correlated matching in the whole algorithm. Figure 3d shows the measured GHZ fidelity as a function of this sliding threshold converted into a fraction of accepted experimental repetitions, continuously tuning the trade-off between the success probability of the algorithm and its fidelity; for example, discarding just 50% of the data improves GHZ fidelity to approximately 90%. (As discussed below, for certain applications, purifying samples can be advantageous in improving algorithmic performance.) Finally, fault-tolerantly measuring all 256 logical Pauli strings, we perform full GHZ state tomography (Fig. 3e).

The use of the zoned architecture directly allows scaling circuits to larger numbers, without increasing the number of controls, by encoding and operating on logical qubits, moving them to storage and then accessing storage as appropriate. This process is illustrated in Fig. 4a,b, in which ten colour codes are made and operated on with parallel transversal CNOTs, moved to storage and then more qubits are accessed from storage. Repeating this process four times, we create 40 colour codes with 280 physical qubits, at the cost of slow idling errors of roughly 1% logical decoherence per additional encoding step (Fig. 4c). These storage idling errors primarily originate from global Raman π pulses applied for dynamical decoupling of atoms in the entangling zone, which could be greatly reduced with zone-specific Raman controls. Because mid-circuit readout10–15 is an important component of logical algorithms, we next demonstrate a fault-tolerant entanglement teleportation circuit. We first create a three-logical-qubit GHZ state |0L0L0L⟩ + |1L1L1L⟩ (Fig. 4d,e) from fault-tolerantly prepared colour codes. Mid-circuit X-basis measurement of the middle logical creates |0L0L⟩ + |1L1L⟩ if measured as |+L⟩ and |0L0L⟩ − |1L1L⟩ if measured as |−L⟩. We recover |0L0L⟩ + |1L1L⟩ by applying a logical S gate to the first and third logicals conditioned in real time on the state of the middle logical, akin to the magic-state-teleportation circuit24. Measurements in Fig. 4e indicate that, although ⟨XLXL⟩ and ⟨YLYL⟩ indeed vanish without the feedforward step, by applying the feedforward correction, we recover a Bell-state fidelity of 77(2)%, limited by imperfections in the original underlying GHZ state. By repeating this experiment without mid-circuit readout and instead postselecting on the middle logical being in |+L⟩, Nature | Vol 626 | 1 February 2024【淘宝唯 一：艾米学社】 | 61

b

Entangling zone

Storage zone

Total number of colour codes 20 30 40

0.02

Physical Detection Correction

0 0

1 2 3 Additional encoded groups

e

d

|0L〉

S

|0L〉 H

Readout zone

No mid-circuit

No feedforward

–0.5

YY

Feedforward

0

XX

Storage zone

0.5

Feedforward

S

XX

S

S

YY

Entangling zone

Measured logical parity

|0L〉

XX

Logical decoherence

10

YY

c

|0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉 |0L〉

Access storage

a

Move to storage

Article

Fig. 4 | Zoned logical processor: scaling and mid-circuit feedforward. a, Atoms in storage and entangling zones and approach for creating and entangling 40 colour codes with 280 physical qubits. b,c, 40 colour codes are prepared with a nFT circuit and then 20 transversal CNOTs are used to fault-tolerantly prepare 20 of the 40 codes, whose fidelity is plotted. Logical decoherence is smaller than the physical idling decoherence experienced during the encoding steps. d, Mid-circuit measurement and feedforward for logical entanglement teleportation. The middle of three logical qubits is measured in the X basis and, by applying a mid-circuit conditional, locally pulsed logical S rotation on the other two logical qubits, the state |0L0L⟩ + |1L1L⟩ is prepared. e, Measured logical qubit parity with and without feedforward, showing that feedforward recovers the intended state with Bell fidelity of 77(2)% (ZZ parities of 83(4)% not plotted; Methods). No mid-circuit refers to turning off the mid-circuit readout and postselecting on the middle logical being in state |+L⟩ in the final readout. By postselecting on perfect stabilizers of only the two computation logicals (error detection in the final measurement), the feedforward Bell fidelity is 92(2)% (not plotted). In d, three of the extra blocks are flag qubits and the other four are prepared but unused for this circuit.

we find a similar Bell fidelity of 75(2)%, indicating high-fidelity performance of the readout and feedforward operations.

Complex logical circuits using 3D codes One important challenge in realizing complex algorithms with logical qubits is that universal computation cannot be implemented transversally42. For instance, when using 2D codes such as the surface code, non-Clifford operations cannot be easily performed37, and relatively expensive techniques are required for nontrivial computation24,43, as Clifford circuits can be easily simulated44. By contrast, 3D codes can transversally realize non-Clifford operations, but lose the transversal H (ref. 37). However, these constraints do not imply that classically hard or useful quantum circuits cannot be realized transversally or efficiently. Motivated by these considerations, we explore efficient realization of classically hard algorithms that are co-designed with a particular error-correcting code. Specifically, we implement fast 62 | Nature | Vol 626 | 1 February 2024【淘宝唯一：艾米学社】

scrambling circuits using small 3D codes, which are used for native non-Clifford operations (CCZ). We focus on small 3D [[8,3,2]] codes16,17,26,27 (Fig. 5a), which have various appealing features. They encode three logicals per block, feature d = 2 (d = 4) in the Z basis (X basis), implying error-detection (error-correction) capabilities for Z (X) errors and can realize a transversal CNOT between blocks. Most importantly, by using physical {T, S} rotations (T is π/4 phase gate), we can realize transversal {CCZ, CZ, Z} gates on the logical qubits encoded within each block, as well as intrablock CNOTs by physical permutation26,27 (Methods). This gate set allows us to transversally realize the circuits illustrated in Fig. 5a,c, alternating between layers of {CCZ, CZ, Z} within blocks and layers of CNOTs between blocks. Although transversal H is forbidden, initialization and measurement in either the X or the Z basis effectively allows H at the beginning and end of the circuit. We use these transversal operations to realize logical algorithms that are difficult to simulate classically45,46. More specifically, these circuits can be mapped to instantaneous quantum polynomial (IQP) circuits20,45,46. Sampling from the output distribution of such circuits is known to be classically hard in certain instances20, implying that a quantum device can be exponentially faster than a classical computer for this task. Figure 5b shows an example implementation of a 12-logical-qubit sampling circuit. Here we prepare all logical blocks in |+L⟩, implement a scrambling circuit with 28 logical entangling gates and then measure all logicals in the X basis. Figure 5b shows the probability of observing each of the 212 = 4,096 possible logical bitstring outcomes, showing that, as we progressively apply more error detection (that is, postselection) in post-processing, the distribution more closely reproduces the ideal theoretical distribution. To characterize the distribution overlap, we use the cross-entropy benchmark (XEB)18,47, which is a weighted sum between the measured probability distribution and the ideal calculated distribution, normalized such that XEB = 1 corresponds to perfectly reproducing the ideal distribution and XEB = 0 corresponds to the uniform distribution, which occurs when circuits are overwhelmed by noise. Consistent with Fig. 5b, the 12-logical-qubit circuit XEB increases from 0.156(2) to 0.616(7) when applying error detection (Fig. 5e). We note that the XEB should be a good fidelity benchmark for IQP circuits (Methods). We next explore scaling to larger systems and circuit depths. To ensure high complexity of our logical circuits, we use nonlocal connections to entangle the logical triplets on up to 4D hypercube graphs (Extended Data Fig. 6 and Supplementary Video 3), which results in fast scrambling19. Exploring entangled systems of 3, 6, 12, 24 and 48 logical qubits, in all cases, we find a finite XEB score, which improves with increased error detection (Fig. 5e,f). The finite XEB indicates successful sampling and the improvement with error detection shows the benefit of using logical qubits. Although this improvement comes at the cost of measurement time owing to error detection, improving the sample quality cannot be replaced by simply generating more samples. Thus, improving the XEB score yields substantial practical gains. We obtain an XEB of approximately 0.1 for 48 logical qubits and hundreds of nonlocal logical entangling gates, up to roughly an order of magnitude higher than previous physical qubit implementations of digital circuits of similar complexity18,48, showing the benefits of a logical encoding for this application. Assuming our best measured physical fidelities, the estimated upper bound for an optimized physical qubit implementation in our system is also greatly below the measured logical XEB (blue line in Fig. 5f; Methods). In attempting to run these complex physical circuits, in practice, we find that realizing non-vanishing XEB is much more challenging; we confirm with small physical instances that we measure values well below this upper bound (Methods). As well as the error-detecting benefits, it seems that the logical circuit is substantially more tolerant to coherent errors, exhibiting operation that is inherently digital,

a

c

3

1

XL1

Logical qubits

5 ZL1

8

ZL2

XL2

ZL3

6

4

Stabilizers

XL3

2

1

Z

Z

Z

H ZZ

Z

d 100 Measured Runtime per bitstring (s)

7

RY π/2 q Z

SZ1 = Z1Z2Z3Z4 SZ2 = Z1Z3Z5Z7 SZ3 = Z1Z2Z5Z6 SZ4 = Z1Z2Z3Z4Z5Z6Z7Z8 SX1 = X1X2X3X4X5X6X7X8

T

T†

T†

Runtime per bitstring (s)

S

106

Transversal CNOT

ap

S†

S†

T

T†

Permutation CNOTL1,L2

CZL1,L2 S

T†

Sw

T

1 2

T

1 2

3 4

3 4

5 6

5 6

7 8

7 8

48 logicals

1.00

Theory Postselection g n sin tio ea c cr te In r de ro er

4,096 Logical bitstring index

0.10

0.01

Hypercube dimension 5 6

7

1.00

Logical XEB

Logical XEB

Logical bitstring probability

3 4

0

102

f

12 logicals

Raw

Projected 48 logicals

0 1 2 3 Extra CNOT layers

e

b

0.01

104

100 q48

0

10–4 6 12 24 48 Number of logicals

Logical operations CCZL1,L2,L3

10–2

3 6 12 24 48 0.001

0.10

Raw Error detection Physical upper bound

0.01

0.010 0.100 Accepted fraction

1.000

3 6

12 24 Number of logical qubits

48

Fig. 5 | Complex logical circuits using 3D codes. a, [[8,3,2]] block codes can transversally realize {CCZ, CZ, Z, CNOT} gates within each block and transversal CNOTs between blocks. By preparing logical qubits in |+L⟩, performing layers of {CCZ, CZ, Z} alternated with inter-block CNOTs and measuring in the X basis, we realize classically hard sampling circuits with logical qubits. b, Measured sampling outcomes for a circuit with 12 logical qubits, eight logical CZs, 12 logical CNOTs and eight logical CCZs. By increasing error detection, the measured distribution converges towards the ideal distribution. c, Circuit involving 48 logical qubits with 228 logical CZ/CNOT gates and 48 logical CCZs. d, Classical simulation runtime for calculating an

individual bitstring probability; bottom plot is estimated on the basis of matrix multiplication complexity. e, Measured normalized XEB as a function of sliding-scale error detection for 3, 6, 12, 24 and 48 logical qubits. For all sizes, we observe a finite XEB score that improves with increased error detection. Diagram shows 48-logical connectivity, with logical triplets entangled on a 4D hypercube. f, Scaling of raw (red) and fully error-detected (black) XEB from e. Physical upper-bound fidelity (blue) is calculated using best measured physical gate fidelities (see Methods and Extended Data Fig. 7 for scaling discussion). Diagrams show physical connectivity. [[8,3,2]] cubes are entangled on 4D hypercubes, realizing physical connectivity of 7D hypercubes.

just with imperfect fidelity (see, for example, Extended Data Fig. 7a), consistent with theoretical predictions49. We also note that, for the logical algorithms, we optimize performance by optimizing the stabilizer expectation values (rather than the complex sampling output), providing further advantage for logical implementations. Our 48-logical circuit, corresponding to a physical qubit connectivity of a 7D hypercube, contains up to 228 logical two-qubit gates and 48 logical CCZ gates. Simulation of such logical circuits is challenging because of the high connectivity (rendering tensor networks inefficient) and large numbers of non-Cliffords50. To benchmark our circuits, we structure them such that we can use an efficient simulation method (Methods), which takes about 2 s to calculate the probability of each bitstring (Fig. 5d and Extended Data Fig. 8). Modelling noise in our logical circuits is even more complicated, as they are composed of 128 physical qubits and 384 T gates, thereby making experimentation with logical algorithms necessary to understand and optimize performance.

In particular, we use a Bell-basis measurement made on two copies of the quantum state (Fig. 6a), which is a powerful tool that can efficiently extract many properties of an unknown state21,22,52. With this two-copy technique, in Fig. 6b, we plot the measured entanglement entropy in the scrambled system. We observe a characteristic Page curve51 associated with a maximally entangled, highly scrambled, but globally pure state. These measurements also reveal a final state purity of 0.74(3), compared with the measured XEB of 0.616(7) in Fig. 5f, consistent with the XEB being a good proxy for the final state fidelity. Despite postselection overhead, we find that error detection greatly improves signal to noise here, as near-zero entropies are exponentially faster to measure (Extended Data Fig. 9). Two-copy measurements can also be used to simultaneously extract information about all 4N Pauli strings22. Using this property and an analysis technique known as Bell difference sampling53, we experimentally evaluate and directly verify the amount of additive Bell magic53 in our circuits as a function of the number of applied logical CCZs (Fig. 6c). This measurement of magic, associated with non-Clifford operations, quantifies the number of T gates (assuming decomposition into T) required to realize the quantum state by observing the probability that sampled Pauli strings commute with each other (see Methods). Moreover, combining encoded qubits and two-copy measurement

Quantum simulations with logical qubits Finally, we explore the use of logical qubits as a tool in quantum simulation, probing entanglement properties of our fast scrambling circuits, potentially related to complex systems such as black holes19,51.

Nature | Vol 626 | 1 February 2024【淘宝唯 一：艾米学社】 | 63

Article b

Scrambling circuit

Scrambling circuit

Bell-basis measurement

10

Logical entanglement entropy

a

X

Z

Increasing error detection

8 6 4 2 0 0

Additive Bell magic

8

Experiment Theory

d 0.25 Logical Pauli string expectation value

c

4

2 4 6 8 10 Logical subsystem size

12

Increasing error detection

0.20 0.15 0.10 0.05 0

0 0 1 2 3 Number of logical CCZ gates

0

0.2

0.4 0.6 Logical purity

0.8

1.0

Fig. 6 | Logical two-copy measurement. a, Identical scrambling circuits are performed on two copies of 12 logical qubits and then measured in the Bell basis to extract information about the state. Z-basis measurements are corrected with an [[8,3,2]] decoder (when full error detection is not applied). b, Measured entanglement entropy as a function of subsystem size, showing expected Page-curve behaviour19 for the highly scrambled state, improving with increased error detection. c, Measured and simulated magic (associated with non-Clifford operations) as a function of the number of CCZ gates applied, performed on two copies of scrambled six-logical-qubit systems. d, Pauli string measurement and zero-noise extrapolation using logical qubits. Plot shows the absolute values of all 412 Pauli string expectation values, which only have five discrete values for our digital circuit; Pauli strings with the same theory value are grouped. By analysing with sliding-scale error detection, we improve towards the theoretical expectation values (squares) while also improving towards a purity of 1. By extrapolating to perfect purity, we extrapolate the expectation values and better approximate the ideal values (shaded regions are statistical fit uncertainty).

Our observations open the door for exploration of large-scale logical qubit devices. A key future milestone would be to perform repetitive error correction6 during a logical quantum algorithm to greatly extend its accessible depth. This repetitive correction can be directly realized using the tools demonstrated here by repeating the stabilizer measurement (Fig. 2) in combination with mid-circuit readout (Fig. 4). The use of the zoned architecture and logical-level control should allow our techniques to be readily scaled to more than 10,000 physical qubits by increasing laser power and optimizing control methods, whereas QEC efficiency can be improved by reducing two-qubit gate errors to 0.1% (ref. 8). Deep computation will further require continuous reloading of atoms from a reservoir source11,15. Continued scaling will benefit from improving encoding efficiency, for example, by using quantum low-density-parity-check codes55,56, using erasure conversion13,33,57 or noise bias35 and optimizing the choice of (possibly several) atomic species11,14,47, as well as advanced optical controls34. Further advances could be enabled by connecting processors together in a modular fashion using photonic links or transport10,58 or more power-efficient trapping schemes such as optical lattices59. Although we do not expect clock speed to limit medium-scale logical systems, approaches to speed up processing in hardware60 or with nonlocal connectivity61 should also be explored. We expect that such experiments with early-generation logical devices will enable experimental and theoretical advances that greatly reduce anticipated costs of large-scale error-corrected systems, accelerating the development of practical applications of quantum computers.

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/s41586-023-06927-3. 1. 2. 3.

allows for further error-mitigation techniques. As an example, Fig. 6d shows the measured absolute expectation values of all 412 logical Pauli strings with sliding-scale error detection. Because in the two-copy measurements for each error-detection threshold we also measure the overall system purity, we can extrapolate our expectation values to the case of unit purity (zero noise)54. This procedure evaluates the averaged Pauli expectation values to about 10% relative precision of the ideal theoretical values spanning several orders of magnitude (Methods).

4. 5. 6. 7. 8. 9. 10.

Outlook These experiments demonstrate key ingredients of scalable error correction and quantum information processing with logical qubits. As well as implementing the key elements of logical processing, our approach demonstrates practical utility of encoding methods for improving sampling and quantum simulations of complex scrambling circuits. Future work can explore whether these methods can be generalized, for example, to more robust, higher-distance codes and if such highly entangled, non-Clifford states could be used in practical algorithms. We note that the demonstrated logical circuits are approaching the edge of exact simulation methods (Fig. 5d) and can readily be used for exploring error-corrected quantum advantage. These examples demonstrate that the use of new encoding schemes, co-designed with efficient implementations, can allow the implementation of particular logical algorithms at reduced cost. 64 | Nature | Vol 626 | 1 February 2024【淘宝唯一：艾米学社】

11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Preskill, J. Quantum computing in the NISQ era and beyond. Quantum 2, 79 (2018). Shor, P. W. in Proc. 37th Conference on Foundations of Computer Science 56–65 (IEEE, 1996). Steane, A. Multiple-particle interference and quantum error correction. Proc. R. Soc. Lond. A Math. Phys. Eng. Sci. 452, 2551–2577 (1996). Dennis, E., Kitaev, A., Landahl, A. & Preskill, J. Topological quantum memory. J. Math. Phys. 43, 4452–4505 (2002). Ryan-Anderson, C. et al. Implementing fault-tolerant entangling gates on the five-qubit code and the color code. Preprint at https://arxiv.org/abs/2208.01863 (2022). Google Quantum AI. Suppressing quantum errors by scaling a surface code logical qubit. Nature 614, 676–681 (2023). Bluvstein, D. et al. A quantum processor based on coherent transport of entangled atom arrays. Nature 604, 451–456 (2022). Evered, S. J. et al. High-fidelity parallel entangling gates on a neutral-atom quantum computer. Nature 622, 268–272 (2023). Beugnon, J. et al. Two-dimensional transport and transfer of a single atomic qubit in optical tweezers. Nat. Phys. 3, 696–699 (2007). Deist, E. et al. Mid-circuit cavity measurement in a neutral atom array. Phys. Rev. Lett. 129, 203602 (2022). Singh, K. et al. Mid-circuit correction of correlated phase errors using an array of spectator qubits. Science 380, 1265–1269 (2023). Graham, T. M. et al. Midcircuit measurements on a single-species neutral alkali atom quantum processor. Phys. Rev. X 13, 041051 (2023). Ma, S. et al. High-fidelity gates and mid-circuit erasure conversion in an atomic qubit. Nature 622, 279–284 (2023). Lis, J. W. et al. Midcircuit operations using the omg architecture in neutral atom arrays. Phys. Rev. X 13, 041035 (2023). Norcia, M. A. et al. Midcircuit qubit measurement and rearrangement in a 171Yb atomic array. Phys. Rev. X 13, 041034 (2023). Campbell, E. T. The smallest interesting colour code. Earl T. Campbell https:// earltcampbell.com/2016/09/26/the-smallest-interesting-colour-code/ (2016). Vasmer, M. & Kubica, A. Morphing quantum codes. Phys. Rev. Appl. 10, 030319 (2022). Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Nature 574, 505–510 (2019). Kuriyattil, S., Hashizume, T., Bentsen, G. & Daley, A. J. Onset of scrambling as a dynamical transition in tunable-range quantum circuits. PRX Quantum 4, 030325 (2023). Bremner, M. J., Montanaro, A. & Shepherd, D. J. Average-case complexity versus approximate simulation of commuting quantum computations. Phys. Rev. Lett. 117, 080501 (2016).

21. 22. 23. 24. 25.

26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42. 43. 44. 45.

Daley, A. J., Pichler, H., Schachenmayer, J. & Zoller, P. Measuring entanglement growth in quench dynamics of bosons in an optical lattice. Phys. Rev. Lett. 109, 020505 (2012). Huang, H. Y. et al. Quantum advantage in learning from experiments. Science 376, 1182–1186 (2022). Gidney, C. & Ekerå, M. How to factor 2048 bit RSA integers in 8 hours using 20 million noisy qubits. Quantum 5, 433 (2021). Fowler, A. G., Mariantoni, M., Martinis, J. M. & Cleland, A. N. Surface codes: towards practical large-scale quantum computation. Phys. Rev. A 86, 032324 (2012). Self, C. N., Benedetti, M. & Amaro, D. Protecting expressive circuits with a quantum error detection code. Nat. Phys. https://doi.org/10.1038/s41567-023-02282-2 (2024【淘宝唯 一：艾米学社】). Honciuc Menendez, D., Ray, A. & Vasmer, M. Implementing fault-tolerant non-Clifford gates using the [[8,3,2]] color code. Preprint at https://arxiv.org/abs/2309.08663 (2023). Wang, Y. et al. Fault-tolerant one-bit addition with the smallest interesting colour code. Preprint at https://arxiv.org/abs/2309.09893 (2023). Andersen, T. I. et al. Non-Abelian braiding of graph vertices in a superconducting processor. Nature 618, 264–269 (2023). Patterson, D. A. & Hennessy, J. L. Computer Organization and Design: The Hardware/ Software Interface. RISC-V Edition (Morgan Kaufmann, 2018). Ebadi, S. et al. Quantum phases of matter on a 256-atom programmable quantum simulator. Nature 595, 227–232 (2021). Scholl, P. et al. Quantum simulation of 2D antiferromagnets with hundreds of Rydberg atoms. Nature 595, 233–238 (2021). Jaksch, D. et al. Fast quantum gates for neutral atoms. Phys. Rev. Lett. 85, 2208–2211 (2000). Scholl, P. et al. Erasure conversion in a high-fidelity Rydberg quantum simulator. Nature 622, 273–278 (2023). Graham, T. M. et al. Multi-qubit entanglement and algorithms on a neutral-atom quantum computer. Nature 604, 457–462 (2022). Cong, I. et al. Hardware-efficient, fault-tolerant quantum computation with Rydberg atoms. Phys. Rev. X 12, 021049 (2022). Beverland, M. E., Kubica, A. & Svore, K. M. Cost of universality: a comparative study of the overhead of state distillation and code switching with color codes. PRX Quantum 2, 020341 (2021). Bombín, H. Gauge color codes: optimal transversal gates and gauge fixing in topological stabilizer codes. New J. Phys. 17, 083002 (2015). Goto, H. Minimizing resource overheads for fault-tolerant preparation of encoded states of the Steane code. Sci. Rep. 6, 19578 (2016). Flammia, S. T. & Liu, Y. K. Direct fidelity estimation from few Pauli measurements. Phys. Rev. Lett. 106, 230501 (2011). Egan, L. et al. Fault-tolerant control of an error-corrected qubit. Nature 598, 281–286 (2021). Postler, L. et al. Demonstration of fault-tolerant universal quantum gate operations. Nature 605, 675–680 (2022). Eastin, B. & Knill, E. Restrictions on transversal encoded quantum gate sets. Phys. Rev. Lett. 102, 110502 (2009). Brown, B. J. A fault-tolerant non-Clifford gate for the surface code in two dimensions. Sci. Adv. 6, eaay4929 (2020). Aaronson, S. & Gottesman, D. Improved simulation of stabilizer circuits. Phys. Rev. A 70, 052328 (2004). Mezher, R., Ghalbouni, J., Dgheim, J. & Markham, D. Fault-tolerant quantum speedup from constant depth quantum circuits. Phys. Rev. Res. 2, 033444 (2020).

46. Paletta, L., Leverrier, A., Sarlette, A., Mirrahimi, M. & Vuillot, C. Robust sparse IQP sampling in constant depth. Preprint at https://arxiv.org/abs/2307.10729 (2023). 47. Shaw, A. L. et al. Benchmarking highly entangled states on a 60-atom analog quantum simulator. Preprint at https://arxiv.org/abs/2308.07914 (2023). 48. Wu, Y. et al. Strong quantum computational advantage using a superconducting quantum processor. Phys. Rev. Lett. 127, 180501 (2021). 49. Bravyi, S., Englbrecht, M., König, R. & Peard, N. Correcting coherent errors with surface codes. npj Quantum Inf. 4, 55 (2018). 50. Bravyi, S. et al. Simulation of quantum circuits by low-rank stabilizer decompositions. Quantum 3, 181 (2019). 51. Sekino, Y. & Susskind, L. Fast scramblers. J. High Energy Phys. 2008, 065 (2008). 52. Hangleiter, D. & Gullans, M. J. Bell sampling from quantum circuits. Preprint at https:// arxiv.org/abs/2306.00083 (2023). 53. Haug, T. & Kim, M. S. Scalable measures of magic resource for quantum computers. PRX Quantum 4, 010301 (2023). 54. Kim, Y. et al. Evidence for the utility of quantum computing before fault tolerance. Nature 618, 500–505 (2023). 55. Bravyi, S. et al. High-threshold and low-overhead fault-tolerant quantum memory. Preprint at https://arxiv.org/abs/2308.07915 (2023). 56. Xu, Q. et al. Constant-overhead fault-tolerant quantum computation with reconfigurable atom arrays. Preprint at https://arxiv.org/abs/2308.08648 (2023). 57. Wu, Y., Kolkowitz, S., Puri, S. & Thompson, J. D. Erasure conversion for fault-tolerant quantum computing in alkaline earth Rydberg atom arrays. Nat. Commun. 13, 4657 (2022). 58. Dordević, T. et al. Entanglement transport and a nanophotonic interface for atoms in optical tweezers. Science 373, 1511–1514 (2021). 59. Tao, R., Ammenwerth, M., Gyger, F., Bloch, I. & Zeiher, J. High-fidelity detection of largescale atom arrays in an optical lattice. Preprint at https://arxiv.org/abs/2309.04717 (2023). 60. Xu, W. et al. Fast preparation and detection of a Rydberg qubit using atomic ensembles. Phys. Rev. Lett. 127, 050501 (2021). 61. Litinski, D. & Nickerson, N. Active volume: an architecture for efficient fault-tolerant quantum computers with limited non-local connections. Preprint at https://arxiv.org/ abs/2211.15465 (2022). 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) 2023

Nature | Vol 626 | 1 February 2024【淘宝唯 一：艾米学社】 | 65

Article Methods System overview Our experimental apparatus (Extended Data Fig. 1a) is described previously in refs. 7,8,30. To carry out these experiments, several key upgrades have been made enabling programmable quantum circuits on both physical and logical qubits. A cloud containing millions of cold 87Rb atoms is loaded in a magneto-optical trap inside a glass vacuum cell, which are then loaded stochastically into programmable, static arrangements of 852-nm traps generated with a SLM and rearranged with a set of 850-nm moving traps generated by a pair of crossed AODs (DTSX400, AA Opto-Electronic) to realize defect-free arrays30,31,62. Atoms are imaged with a 0.65-NA objective (Special Optics) onto a CMOS camera (Hamamatsu ORCA-Quest C15550-20UP), chosen for fast electronic readout times. The qubit state is encoded in mF = 0 hyperfine clock states in the 87Rb ground-state manifold, with T2 > 1s (ref. 7), and fast, high-fidelity single-qubit control is executed by two-photon Raman excitation7,63 (Extended Data Fig. 1b). A global Raman path illuminating the entire array is used for global rotations (Rabi frequency roughly 1 MHz, resulting in approximately 1-μs rotations with composite pulse techniques7), as well as for dynamical decoupling throughout the entire circuit (typically one global π pulse per movement). Fully programmable local single-qubit rotations are realized with the same Raman light but redirected through a local path, which is focused onto targeted atoms by an additional set of 2D AODs. Entangling gates (270-ns duration) between clock qubits are performed with fast two-photon excitation using 420-nm and 1,013-nm Rydberg beams to n = 53 Rydberg states, using a time-optimal two-qubit gate pulse64,65, detailed in ref. 8. During the computation, atoms are rearranged with the AOD traps to enable arbitrary connectivity7,66,67. Mid-circuit readout is carried out by illuminating from the side with a locally focused 780-nm imaging beam, with scattered photons collected on the CMOS camera and processed in real time by a FPGA (Xilinx ZCU102), with feedforward control signal outputs. The quantum circuits are programmed with a control infrastructure consisting of five arbitrary waveform generators (AWGs) (Spectrum Instrumentation), as illustrated in Extended Data Fig. 1c, synchronized to 1.0 Ω cm) and Ge (purity >99.999%) rods of 2 ± 0.127 mm diameter were purchased from Lattice Materials. P- and n-type Si rods (purity >99.999%, resistivity 20:1, unless otherwise indicated

O

AgSbF6, 70%

O

OH

Charge relocation

syn-selective, d.r. >20:1

b

O

4 53%

O

I

MeO

5 74%* O

O

OH

OH

OH

Me Me

F3C

Me

6 83%

7 86% Me

O Cl

8 75%

9 73%, 12:1 r.r.

O

O

OH

10 81%*, 17:1 r.r.

O

O

OH

OH

OH

Me Me

Cl

Me

11 75%*, 17:1 r.r.

O2N

Me

H3CO2C

12 41%a, 9:1 r.r.

O

13 58%*

14 70%, 14:1 r.r.

O

O

O OH

O

OH MeO

Me

15 86% Me OH

C9H19

OH

OH

O OH

Me

O

16 58%b, 17:1 r.r.

17 80%

18 77%

O

O

F 3C

O OH

Cl

Me

19 70%c

OH

Me

20 86%* O

O

OH

OH

OH

Me

Cl

Me

21 46%

22 91%, 14:1 r.r.

23 65%*

24 59%, 10:1 r.r.

O

O

O OH

OH

Ph

1

S

O

25 54%*

2

4

O 1

OH

3

27 61%, 13:1 r.r.

26 83%, 17:1 r.r.

4

1

Ph

3

4

2

28 60%

3

2

Norbornene

c

OH

C5H11

[Ni] or [Pd]

Unbiased alkene

C5H11 Not detected

Fig. 2 | General reaction scheme, optimization and the scope of syn-selective 1,3-hydroxyacylation. a, 1,3-Hydroacylation of cyclohexene. b, Scope of syn-alcohols. c, Comparison with transition-metal catalysis. All yields in the table of optimization correspond to nuclear magnetic resonance yields using 1,3,5-trimethoxybenzene as an internal standard. All products were obtained at greater than 20:1 d.r. and regioisomeric ratio (r.r.), unless otherwise

The reactions of more complex substrates also proved interesting. When 1-methylcyclohexene, a trisubstituted alkene, was used, the all-syn-hydroxyketone 54 was obtained as a single diastereomer (see X-ray structure to the right; Fig. 4d). This stereochemical outcome is probably governed by the preferred equatorial orientation of the only substituent not bound within a ring, the methyl group (Fig. 4d). 94 | Nature | Vol 626 | 1 February 2024

C5H11 Unbiased alkene

Charge relocation

C5H11

Ph

2 64%

O

mentioned. All yields correspond to isolated material. See Supplementary Information for further details and additional scope entries. *The reported yields correspond to averages over three runs. In cases in which competing pathways led to the observation of enone byproducts in the crude mixtures, the following percentages were observed: a37, b24, c30. Additional substrate scope is presented in Extended Data Fig. 1 and Supplementary Information.

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

a

b

Syn-alcohols

Anti-alcohols

O

O

O

OH

OH

Me Me

F 3C

F3C 10 83%

O OH

Me

Me

29 66%

Stereodivergent

OH Me Me

30 61%

O

31 76%*a O

O

Me

OH

OH OH S 32 80%*

33 66%*b, 14:1 r.r.

O

O

O

OH X-ra crystal structure X-ray of 10

c

34 68%

OH

OH

F

CF3O 35 71%*

– SbF6

Other nucleophiles

37 83% Stereoinvertive work-up

+ O

Charge relocation

O

36 92%*

Charge relocation

+

In situ divergence

O

d

1,4-Dicarbonyls O

O

Cl

Br

O

O F3C

38 83%

39 47%*

O

44 81%

O

45 55%*

O

I

O

O

O

O O S

Me 40 57%, 13:1 d.r.

41 66%, 13:1 r.r. O

O O

46 74% Me

O

O

Me

O Me

N

Me Me 42 68%

47 51%c O O

Me O

Me

Me

43 80%

48 38%

49 64% Kornblum-type oxidation

Fig. 3 | Stereodivergence of the 1,3-difunctionalization of alkenes and extension to other product classes. a, Selected example of syn-selective hydroxyacylation in contrast to anti-configured product 29. b, Scope of anti-selective 1,3-hydroxyacylation, achieved by the addition of DMSO at 35 °C and quenching with tetrabutylammonium bromide at room temperature. c, Additional 1,3-difunctionalization reactions using other nucleophiles— tetrabutylammonium halides, dimethylacetamide, dimethylformamide or

TEMPO—at 35 °C. d, 1,4-Dicarbonyl synthesis through Kornblum-type oxidation, employing DMSO at 35 °C, followed by NEt 3 at room temperature. All products were obtained at greater than 20:1 d.r. and 20:1 r.r., unless otherwise mentioned (Supplementary Information). *The reported yields correspond to averages over three runs. In cases in which competing pathways led to observation of enone byproducts in the crude mixtures, the following percentages were observed: a5, b10, c11.

code availability are available at https://doi.org/10.1038/s41586-02306938-0.

5.

1. 2. 3. 4.

Clayden, J., Greeves, N. & Warren S. Organic Chemistry (Oxford Univ. Press, 2012). Vasseur, A., Bruffaerts, J. & Marek, I. Remote functionalization through alkene isomerization. Nat. Chem. 8, 209–219 (2016). Massad, I. et al. Stereoselective synthesis through remote functionalization. Nat. Synth. 1, 37–48 (2022). Delcamp, J. H. & White, M. C. Sequential hydrocarbon functionalization: allylic C-H oxidation/vinylic C–H arylation. J. Am. Chem. Soc. 128, 15076–15077 (2006).

Wang, W. et al. Migratory arylboration of unactivated alkenes enabled by nickel catalysis. Angew. Chem. Int. Ed. 58, 4612–4616 (2019). 6. Li, Y., Wu, D., Cheng, H.-G. & Yin, G. Difunctionalization of alkenes involving metal migration. Angew. Chem. Int. Ed. 59, 7990–8003 (2020). 7. Han, C. et al. Palladium-catalyzed remote 1,n-arylamination of unactivated terminal alkenes. ACS Catal. 9, 4196–4202 (2019). 8. Thompson, R. C., Swan, S. H., Moore, C. J. & vom Saal, F. S. Our plastic age. Phil. Trans. R. Soc. Lond. B Biol. Sci. 364, 1973–1976 (2009). 9. Trost, B. M. Transition metal templates for selectivity in organic synthesis. Pure Appl. Chem. 53, 2357–2370 (1981). 10. Tsuji, J. Catalytic reactions via π-allylpalladium complexes. Pure Appl. Chem. 54, 197–206 (1982).

Nature | Vol 626 | 1 February 2024 | 95

Article a

One-step synthesis

O O

EtO

Me

O

Five steps OEt

Ref. 43

2

Charge relocation

3 2

Acylation

c

O

Me OH

O +

1

O

Br or

α

2

1

+O

O γ

or

β

Br

1

Hydrolysis

2

3

OH

2

H SbF6– rac-I O

Br – abstraction

OH

Charge relocation

Br – abstraction

Br – abstraction

O

Yields α, 70% from: β, 70% γ, 78%

Br – abstraction

O

O +

Charge relocation

+

OH 53 68%

O 1

H 3

Br

S

OH 52 56%

‘Locking’ event

3+

Me

Me

51 83%

O

O+

3

O

O

Me

4-Ipomeanol (50) 73% One step from 1-butene

b

Me

1-Butene

OH

O

O

1

Me

Derivatives of 4-ipomeanol O

Charge relocation

+

Hydrolysis

O+

Equilibration

2

To most stable isomer

O+ 2

1

3

1

O+ 3

1 3

2

‘Locked position’

d 1

2

3

Charge relocation

O+

3

Me 2

O

Ph

Ph

Me

or

Me

1

2

Ph

O+

3

Me 1

2

3

OH

Hydrolysis 1

54 40% Single diastereomer

Fig. 4 | Application and mechanistic investigation of the 1,3-difunctionalization of alkenes. a, Synthesis of 4-ipomeanol and other biologically active molecules 43. b, Proposed mechanism of 1,3-alkene difunctionalization, involving charge relocation. c, Mechanistic investigations

support the hypothesis that nascency of the charge does not affect the constitution of the obtained product (Supplementary Information). d, Complete stereocontrol for the formation of a 1,2,3-trisubstituted cyclohexane.

11.

22. Basnet, P. et al. Ni-catalyzed regioselective β,δ-diarylation of unactivated olefins in ketimines via ligand-enabled contraction of transient nickellacycles: rapid access to remotely diarylated ketones. J. Am. Chem. Soc. 140, 7782–7786 (2018). 23. Groves, J. K. The Friedel–Crafts acylation of alkenes. Chem. Soc. Rev. 1, 73–97 (1972). 24. Valko, J. T. & Wolinsky, J. Acylation-cycloalkylation. reaction of phenylacetyl chloride with cyclohexene. J. Org. Chem. 44, 1502–1508 (1979). 25. Gao, S., Gao, X., Yang, Z. & Zhang, F. Process research and impurity control strategy of esketamine. Org. Process Res. Dev. 24, 555–566 (2020). 26. Wang, Z. Comprehensive Organic Name Reactions and Reagents (Wiley, 2010). 27. Friess, S. L. & Pinson, R. Jr. A rearrangement in the Nenitzescu reaction of cycloheptene with acetyl chloride and aluminum chloride. J. Am. Chem. Soc. 73, 3512–3514 (1951). 28. Ahmad, M. S., Baddeley, G., Heaton, B. G. & Rasburn, J. W. Vinyl ethers from the reaction of decalin with Friedel–Crafts acylating agents. Proc. Chem. Soc. 395 (1959). 29. Lyall, C. L. et al. Lewis C–H functionalization of sp3 centers with aluminum: a computational and mechanistic study of the Baddeley reaction of decalin. J. Am. Chem. Soc. 136, 13745–13753 (2014). 30. Baddeley, G., Heaton, B. G. & Rasburn, J. W. The interaction of decalin and Friedel–Crafts acetylating agent. Part II. J. Chem. Soc. 4713–4719 (1960). 31. Zong, Y. et al. Enantioselective total syntheses of manginoids A and C and guignardones A and C. Angew. Chem. Int. Ed. 60, 15286–15290 (2021). 32. DeHovitz, J. S. et al. Static to inducibly dynamic stereocontrol: the convergent use of racemic β-substituted ketones. Science 369, 1113–1118 (2020). 33. Krautwald, S., Schafroth, M. A., Sarlah, D. & Carreira, E. M. Stereodivergent α-allylation of linear aldehydes with dual iridium and amine catalysis. J. Am. Chem. Soc. 136, 3020–3023 (2014). 34. Schafroth, M. A., Zuccarello, G., Krautwald, S., Sarlah, D. & Carreira, E. M. Stereodivergent total synthesis of Δ9-tetrahydrocannabinols. Angew. Chem. Int. Ed. 53, 13898–13901 (2014).

12. 13. 14. 15.

16. 17.

18. 19.

20. 21.

Beller, M., Seayad, J., Tillack, A. & Jiao, H. Catalytic Markovnikov and anti-Markovnikov functionalization of alkenes and alkynes: recent developments and trends. Angew. Chem. Int. Ed. 43, 3368–3398 (2004). Werner, E. W., Mei, T.-S., Burckle, A. J. & Sigman, M. S. Enantioselective Heck arylations of acyclic alkenyl alcohols using a redox-relay strategy. Science 338, 1455–1458 (2012). Jensen, K. H. & Sigman, M. S. Mechanistic approaches to palladium-catalyzed alkene difunctionalization reactions. Org. Biomol. Chem. 6, 4083–4088 (2008). Zhang, F.-G. et al. Remote fluorination and fluoroalkyl(thiol)ation reactions. Chem. Eur. J. 26, 15378–15396 (2020). Bruffaerts, J., Pierrot, D. & Marek, I. Zirconocene-assisted remote cleavage of C–C and C–O bonds: application to acyclic stereodefined metalated hydrocarbons. Org. Biomol. Chem. 14, 10325–10330 (2016). Sommer, H., Juliá-Hernández, F., Martin, R. & Marek, I. Walking metals for remote functionalization. ACS Cent. Sci. 4, 153–165 (2018). Vasseur, A., Perrin, L., Eisensteinc, O. & Marek, I. Remote functionalization of hydrocarbons with reversibility enhanced stereocontrol. Chem. Sci. 6, 2770–2776 (2015). Morcillo, S. P. et al. Photoinduced remote functionalization of amides and amines using electrophilic nitrogen radicals. Angew. Chem. Int. Ed. 57, 12945–12949 (2018). Zhao, Y. & Ge, S. Synergistic hydrocobaltation and borylcobaltation enable regioselective migratory triborylation of unactivated alkenes. Angew. Chem. Int. Ed. 61, e202116133 (2022). Kong, W. et al. Base-modulated 1,3-regio- and stereoselective carboboration of cyclohexenes. Angew. Chem. Int. Ed. 62, e202308041 (2023). Li, W., Boon, J. K. & Zhao, Y. Nickel-catalyzed difunctionalization of allyl moieties using organoboronic acids and halides with divergent regioselectivities. Chem. Sci. 9, 600–607 (2018).

96 | Nature | Vol 626 | 1 February 2024

35. Krautwald, S. & Carreira, E. M. Stereodivergence in asymmetric catalysis. J. Am. Chem. Soc. 139, 5627–5639 (2017). 36. Zhou, Q., Chin, M., Fu, Y., Liu, P. & Yang, Y. Stereodivergent atom-transfer radical cyclization by engineered cytochromes P450. Science 374, 1612–1616 (2021). 37. Kaldre, D., Klose, I. & Maulide, N. Stereodivergent synthesis of 1,4-dicarbonyls by traceless charge-accelerated sulfonium rearrangement. Science 361, 664–667 (2018). 38. Kaiser, D., Teskey, C. J., Adler, P. & Maulide, N. Chemoselective intermolecular cross-enolate-type coupling of amides. J. Am. Chem. Soc. 139, 16040–16043 (2017). 39. DeMartino, M. P., Chen, K. & Baran, P. S. Intermolecular enolate heterocoupling: scope, mechanism, and application. J. Am. Chem. Soc. 130, 11546–11560 (2008). 40. Parida, K. N., Pathe, G. K., Maksymenko, S. & Szpilman, A. M. Cross-coupling of dissimilar ketone enolates via enolonium species to afford non-symmetrical 1,4-diketones. Beilstein J. Org. Chem. 14, 992–997 (2018). 41. Christian, M. C. et al. 4-Ipomeanol: a novel investigational new drug for lung cancer. J. Natl. Cancer Inst. 81, 1133–1143 (1989). 42. Parkinson, O. T., Teitelbaum, A. M., Whittington, D., Kelly, E. J. & Rettie, A. E. Species differences in microsomal oxidation and glucuronidation of 4-ipomeanol: relationship to rarget organ toxicity. Drug Metab. Dispos. 44, 1598–1602 (2016). 43. Boyd, M. R., Wilson, B. J. & Harris, T. M. Confirmation by chemical synthesis of the structure of 4-ipomeanol, a lung-toxic metabolite of the sweet potato, Ipomoea batatas. Nat. New Biol. 236, 158–159 (1972). 44. Desai, D., Chang, L. & Amin, S. Synthesis and bioassay of 4-ipomeanol analogs as potential chemopreventive agents against 4-(methylnitrosamino)–1-(3-pyridyl)– 1-butanone (NNK)-induced tumorigenicity in A/J mice. Cancer Lett. 108, 263–270 (1996). 45. Lubinskaya, O. V., Shashkov, A. S., Chertkov, V. A. & Smit, W. A. Facile synthesis of cyclic carboxonium salts by acylation of alkenes. Synthesis 11, 742–745 (1976). 46. Olah, G. A., Kuhn, S. J., Tolgyesi, W. S. & Baker, E. B. Stable carbonium ions. II. 1a Oxocarbonium1b(acylium) tetrafluoroborates, hexafluorophosphates,

47.

48. 49.

50.

hexafluoroantimonates and hexafluoroarsenates. Structure and chemical reactivity of acyl fluoride: Lewis acid fluoride complexes1c. J. Am. Chem. Soc. 84, 2733–2740 (1962). Cresswell, A. J., Davies, S. G., Roberts, P. M. & Thomson, J. E. Beyond the Balz–Schiemann reaction: the utility of tetrafluoroborates and boron trifluoride as nucleophilic fluoride sources. Chem. Rev. 115, 566–611 (2015). Gockel, S. N., Buchanan, T. L. & Hull, K. L. Cu-catalyzed three-component carboamination of alkenes. J. Am. Chem. Soc. 143, 6019–6020 (2021). Lubinskaya, O. V. et al. Formation of carboxonium salts in the acylation of alkenes by acylium salts and some problems of the mechanism of acylation. Russ. Chem. Bull. 27, 343–351 (1978). Buchanan, T. L., Gockel, S. N., Veatch, A. M., Wang, Y.-N. & Hull, K. L. Copper-catalyzed three-component alkene carbofunctionalization: C–N, C–O, and C–C bond formation from a single reaction platform. Org. Lett. 23, 4538–4542 (2021).

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 | 1 February 2024 | 97

Article Data availability All data are available in the manuscript or Supplementary Information. Acknowledgements We thank A. Roller (University of Vienna) for X-ray crystallographic structure determination. We thank P. S. Grant, C. R. Gonçalves and N. Gillaizeau-Simonian for helpful discussions. This work has been supported by the Austrian Academy of Sciences (DOC Fellowship to B.R.B.) and the European Research Council (CoG VINCAT to N.M.). We thank the University of Vienna for generous support. Author contributions N.M. designed and directed the project. B.R.B., G.I., M.R. and D.K. performed and analysed the experiments. B.R.B., G.I., D.K. and N.M. cowrote 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/s41586-023-06938-0. Correspondence and requests for materials should be addressed to Nuno Maulide. 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 | Additional products of syn-selective 1,3-hydroxyacylation. *Percentages of observed enone by-products in the crude mixtures (see the Supplementary Information for additional details).

Article

Establishing reaction networks in the 16-electron sulfur reduction reaction https://doi.org/10.1038/s41586-023-06918-4 Received: 5 September 2022 Accepted: 30 November 2023 Published online: 31 January 2024 Check for updates

Rongli Liu1,7, Ziyang Wei1,7, Lele Peng1, Leyuan Zhang1,2, Arava Zohar3, Rachel Schoeppner4, Peiqi Wang1, Chengzhang Wan1, Dan Zhu1, Haotian Liu2, Zhaozong Wang1, Sarah H. Tolbert1,2,5, Bruce Dunn2,5, Yu Huang2,5, Philippe Sautet1,5,6 ✉ & Xiangfeng Duan1,5 ✉

The sulfur reduction reaction (SRR) plays a central role in high-capacity lithium sulfur (Li-S) batteries. The SRR involves an intricate, 16-electron conversion process featuring multiple lithium polysulfide intermediates and reaction branches1–3. Establishing the complex reaction network is essential for rational tailoring of the SRR for improved Li-S batteries, but represents a daunting challenge4–6. Herein we systematically investigate the electrocatalytic SRR to decipher its network using the nitrogen, sulfur, dual-doped holey graphene framework as a model electrode to understand the role of electrocatalysts in acceleration of conversion kinetics. Combining cyclic voltammetry, in situ Raman spectroscopy and density functional theory calculations, we identify and directly profile the key intermediates (S8, Li2S8, Li2S6, Li2S4 and Li2S) at varying potentials and elucidate their conversion pathways. Li2S4 and Li2S6 were predominantly observed, in which Li2S4 represents the key electrochemical intermediate dictating the overall SRR kinetics. Li2S6, generated (consumed) through a comproportionation (disproportionation) reaction, does not directly participate in electrochemical reactions but significantly contributes to the polysulfide shuttling process. We found that the nitrogen, sulfur dual-doped holey graphene framework catalyst could help accelerate polysulfide conversion kinetics, leading to faster depletion of soluble lithium polysulfides at higher potential and hence mitigating the polysulfide shuttling effect and boosting output potential. These results highlight the electrocatalytic approach as a promising strategy for tackling the fundamental challenges regarding Li-S batteries.

The lithium sulfur (Li-S) battery represents an attractive, nextgeneration energy storage device because of its exceptional theoretical capacity of 1,672 mAh g−1 and ultrahigh energy density of 2,600 Wh kg−1 (refs. 7,8). Despite extensive efforts devoted to improving the practical performance of Li-S batteries3,9–11, the fundamental reaction mechanism—particularly for the sulfur reduction reaction (SRR) during discharge—remains a topic of considerable debate4–6,12–15. The SRR involves a complex multistep, 16-electron conversion from S8 molecules to Li2S solid, with multiple potential interwoven branches among a series of soluble lithium polysulfide (LiPS) intermediates. Soluble LiPS can readily shuttle across the cathode and anode, leading to rapid capacity fading. Recent investigations suggest that the conversion from high-order polysulfides to insoluble Li2S2/Li2S represents the most difficult step, leading to an accumulation of soluble LiPS in the electrolyte and exacerbation of the shuttling issue16. An electrocatalytic process could help accelerate such conversion kinetics and reduce LiPS accumulation, and diverse electrocatalysts have shown promise in improving battery performance17–25. However, the exact role of such electrocatalysts in modification of the SRR mechanism remains

elusive. A comprehensive elucidation of the SRR network and the electrocatalytic effect is essential for the rational design of electrocatalysts that can target specific steps to fundamentally solve the polysulfide shuttling problem. Various approaches have been considered to elucidate the SRR mechanism, involving both experimental4,5,12–14,26–28 and computational efforts12,13,29–34. Detailed mechanistic studies based on stand-alone electrochemistry are generally challenging, owing to the complex convolution of multiple electrochemical reactions and non-electrochemical side reactions (for example, LiPS disproportionations4,5,28) at the same potential. Advanced in situ characterization, to identify and track different polysulfide species generated electrochemically or nonelectrochemically, is essential for the interpretion and corroboration of electrochemical characteristics5,13,14,27. Computational efforts have focused on predicting the energetics of different species and, furthermore, the reaction network12,34. However, current studies do not treat the overall reaction network equilibrium and lack a description of potential-dependent properties such as polysulfide concentrations along the SRR process. This hinders direct comparison with

1 Department of Chemistry and Biochemistry, University of California, Los Angeles, CA, USA. 2Department of Materials Science and Engineering, University of California, Los Angeles, CA, USA. 3Materials Department and Materials Research Laboratory, University of California, Santa Barbara, CA, USA. 4California NanoSystems Institute, University of California, Santa Barbara, CA, USA. 5California NanoSystems Institute, University of California, Los Angeles, CA, USA. 6Department of Chemical and Biomolecular Engineering, University of California, Los Angeles, CA, USA. 7These authors contributed equally: Rongli Liu, Ziyang Wei. ✉e-mail: [email protected]; [email protected]

98 | Nature | Vol 626 | 1 February 2024

a S cathode

Li anode

b

c

Jgeo (mA cm–2)

1.5

0

2.1

2.3

Blank + S Blank

–1.5 1.6

2.0 2.4 Voltage (V) versus Li/Li+

2.8

Fig. 1 | Polysulfide conversion reactions involved in the Li-S battery. a, Schematic illustration of the SRR network involved in the Li-S battery. Blue and yellow spheres represent lithium and sulfur, respectively. SRR encompasses a sophisticated 16-electron conversion process from S8 molecules to Li2S solid, involving multiple soluble LiPS intermediates. b, CV of the SRR in a Li-S battery with N,S–HGF catalysts. The CV data were collected in a coin cell at a scanning rate of 0.05 mV s−1 (see Methods for details). The black baseline was obtained

using the same N,S–HGF in the blank electrolyte without sulfur, indicating the negligible double-layer capacity contribution to overall capacity. Inset shows the plateau in the approximate voltage range 2.11–2.25 V, which originated from the delayed conversion of Li2S 8 to Li2S 4 due to Li2S 8 + Li2S 4 ⇄ 2 Li2S 6 comproportionation/disproportionation reactions. Jgeo, geometric current density. c, Schematic illustration of the in situ Raman technique used in this study.

experimental results and therefore it is highly desirable to determine the detailed mechanism based on ab initio energetics with validation from both electrochemical and in situ techniques. Here we report a systematic investigation of the electrocatalytic SRR mechanism. We chose the previously developed nitrogen, sulfur, dual-doped holey graphene framework (N,S–HGF) electrocatalysts and non-doped HGF as model systems to explore the impact of catalysts in modification of the reaction network and kinetics. Combining cyclic voltammetry (CV), in situ Raman spectroscopy and density functional theory (DFT) calculations, we establish (1) a detailed reaction network, (2) elucidate the dominant reaction pathway before and after the central Li2S4 intermediate, (3) identify the key species as S8, Li2S8, Li2S6, Li2S4 and Li2S and (4) determine that the non-electrochemical comproportionation/disproportionation reaction between Li2S8 and Li2S4 is the main path for formation or consumption of Li2S6. Comparison between N,S–HGF and HGF confirmed the same key species in the reaction network and the N,S–HGF catalyst accelerated LiPS conversion, leading to more rapid depletion of LiPSs at higher potential to mitigate the polysulfide shuttling effect and yield a higher output potential. These results emphasize electro­ catalysis as a promising strategy to address fundamental Li-S battery challenges.

16e− process (Supplementary Fig. 1 and Supplementary Table 1). By separation of the overall discharge process at the onset of the second peak (2.11 V), a charge transfer ratio of 4.08:11.92 (approximately equal to 1:3) was obtained, which was also validated by galvanostatic charge–discharge testing (Supplementary Fig. 2). This 1:3 charge transfer ratio suggests that Li2S4 is the primary intermediate separating these two reduction peaks, because the reaction S8 + 4Li+ + 4e− → 2Li2S4 involves 4 electrons out of the formally overall 16 electrons transferred, and the subsequent conversion 2Li2S4 + 12Li+ + 12e− → 8Li2S involves 12 electrons. Interestingly, a non-zero plateau was observed in the voltage range 2.11–2.25 V between the two peaks (inset in Fig. 1b). Charge integration results, as represented in Fig. 2a by coloured regions, illustrate the non-negligible contribution of the plateau region between the two major redox peaks—about one out of 16 electrons per S8 molecule in SRR. In comparison with a control group lacking sulfur (black baseline in Fig. 1b), the double-layer capacity contribution to this plateau has been eliminated. Considering the instability and complexity of polysulfides4,5,28, the voltage ranges of electrochemical reactions reflected on CV can be affected by comproportionation/disproportionation reactions, although such non-electrochemical processes are not directly detectable via CV measurement. To investigate the chemical origin of such a plateau we used firstprinciples calculations to explore the fundamental energetics among different polysulfide intermediates. Computational modelling of the SRR network began with the conversion from S8 to Li2S8, yielding a calculated output potential of 2.41 V, the highest among all steps. Further conversion of the Li2S8 molecule involved multiple possible branches whereas the calculated energetics show that the pathway forming two Li2S4 is the most exergonic and hence the favoured one, yielding an output potential of 2.24 V (Fig. 2d).

Reaction network in SRR and CV results It has previously been shown that N,S–HGF can markedly accelerate SRR kinetics in comparison with non-doped HGF16. The CV curve for SRR with N,S–HGF (red curve in Fig. 1b) exhibited two main peaks during discharge: one appearing around 2.2–2.5 V and a second around 1.9–2.1 V. The charge number calculated from the integrated area in CV was converted to the formal electron transfer number in a full

Nature | Vol 626 | 1 February 2024 | 99

Article a

d

S8

Li2S4

Li2S

11.92

1.10

2.98

–1.5

b

Exp.

Concentration normalized J (mA cm–2) relative to initial [S8] (%) geo

0

2. 24

–1.0

V

–0.5 2.41 V

Jgeo (mA cm–2)

0

C H

–0.5 –1.0

O 11.98

1.20

2.82

–1.5

c

100 80 60 40

S Li Sim.

Li2S8

S8 Li2S8 Li2S6 Li2S4 Li2S

Li2S6

Minor electrochemical reaction

20 0 1.6

2.0 2.4 Voltage (V) versus Li/Li+

Major electrochemical reaction

2.8

Li2S7

Li2S5

Chemical reaction

Fig. 2 | Charge analysis and reaction network for the SRR. a, Experimental CV curve (Exp.) with charge integration results, separated into three potential regions with a charge ratio of 2.98:1.10:11.92 from high to low potential. The CV data were collected in a coin cell at a scanning rate of 0.05 mV s−1 (see Methods for details). b, Simulated CV curve (Sim.) from first principles with the charge integration results, separated into the same three regions as in a, with charge ratio 2.82:1.20:11.98. c, Simulated voltage-dependent concentrations of the major species considered: S8, Li2S8, Li2S6, Li2S 4 and Li2S. Concentrations are normalized according to sulfur content. d, The dominant reaction mechanism

suggested by DFT energetics: S8 → Li2S8 → 2Li2S 4 → 8Li2S (Li2S8 + Li2S 4 ⇄ 2Li2S6), in which the chemical disproportionation part is shown in parentheses. Solid red and dotted yellow lines indicate major and minor electrochemical reactions, respectively, and blue lines indicate chemical reactions. Major products are indicated by red and blue boxes, corresponding to electrochemical and chemical origin, respectively. Thermodynamic output potentials are denoted for major electrochemical reactions. The catalytic site-dependent output potentials for Li2S 4 → Li2S are detailed in Fig. 5.

Because the experimentally observed plateau might have originated from the delayed electrochemical conversion of comproportionation or disproportionation products, we checked the possibilities starting with one Li2S8 molecule and one Li2S4 molecule, or two Li2S4 molecules, and found that the reaction Li2S8 + Li2S4 → 2Li2S6 was the only exergonic one, with a reaction Gibbs free energy of −0.16 eV, whereas the disproportionation reactions involving either Li2S8 and Li2S6 or Li2S4 and Li2S6 were found to be endergonic. These results suggest that Li2S6 formation by comproportionation of Li2S4 and Li2S8 is the only chemical elementary step that significantly competes with the electrochemical reaction network. Combining the aforementioned DFT balances (Fig. 2d) with voltage effects, the potential-dependent concentrations of different polysulfides were simulated, giving a sequence of dominant LiPS species as S8, Li2S8, Li2S6, Li2S4 and Li2S with reducing potential (Fig. 2c). The simulated CV curve was further derived from the concentrations, giving a charge ratio of 2.82:1.20:11.98 in the red, yellow and blue zones, respectively, of Fig. 2b, which matches well with the experimental ratio shown in Fig. 2a. Although Li2S6 appeared immediately after Li2S8, it was not produced by electrochemical reduction of Li2S8. Instead, a fraction of Li2S8 (about one-third) was electrochemically transformed into Li2S4, providing a fractional amount of charge in the red region shown in Fig. 2a,b (around two-thirds e), whereas the remainder underwent comproportionation with the produced Li2S4 to yield a large concentration of Li2S6 at roughly 2.25 V. Note that the exergonic nature of Li2S8 + Li2S4 → 2Li2S6 comproportionation provided additional driving force to initiate the electrochemical reduction of Li2S8 to Li2S4 at a potential higher (about 2.35 V) than its equilibrium (2.24 V). At lower potential, when Li2S8 is largely consumed, the comproportionation reaction operates backwards to disproportionate Li2S6 into Li2S4 and Li2S8 (2Li2S6  → Li2S8 + Li2S4) in which Li2S8 is electrochemically reduced to Li2S4, resulting in the discharge plateau seen in the yellow region. We found that the direct reduction

of Li2S6 to Li2S4, Li2S3 or other lower-order polysulfides cannot occur at a potential higher than 1.97 V, making the disproportionation reaction the only viable path in this potential regime (Supplementary Note 1). At even lower potential in the blue zone, Li2S4 is eventually reduced to Li2S, involving the 12 extra electrons of the electrochemical reduction reaction.

100 | Nature | Vol 626 | 1 February 2024

In situ Raman study on SRR We used in situ Raman spectroscopy (Supplementary Fig. 3) to probe the specific reaction intermediates along a discharge CV scan (Fig. 3a,b). Initially the typical S8 peak at 469 cm−1 confirmed the existence of elemental sulfur30. With reducing potential, the S8 signal gradually decreased and mostly disappeared at approximately 2.36 V, accompanied by the emergence of the Li2S8 signal at 508 cm−1 starting at about 2.44 V, along with the appearance of the Li2S6 peak at 399 cm−1. As discussed in the computation section, this occurs by electrochemical transformation of Li2S8 to Li2S4 and rapid comproportionation between the formed Li2S4 and remaining Li2S8 to yield Li2S6. The Li2S8 peak at 508 cm−1 reached maximum at around 2.32 V, at which a deconvoluted peak at 501 cm−1 for Li2S4 emerged. At 2.18 V, the 508 cm−1 Li2S8 peak largely disappeared and the Li2S4 peak at 501 cm−1 reached its maximum. As the potential decreased, Li2S4 became the main polysulfide species and Li2S6, formed through the comproportionation reaction (Li2S8 + Li2S4 ⇄ 2Li2S6), was also significantly present. The Li2S6 peak at 399 cm−1 began to decrease at roughly 2.30 V and almost disappeared at 2.02 V. Similarly, most Li2S4 disappeared at around 2.00 V, indicating the conversion from Li2S4 to Li2Sx (x = 1, 2, 3; Fig. 5). A small amount of Li2S4 remained at the low-voltage region below 1.90 V, probably due to incomplete conversion. Additionally, a minor Raman peak at 531 cm−1 was assigned to LiS3•. Our calculations show that LiS3• originated from an electrochemically passive branch (Li2S6 ⇄ 2LiS3•) and was present as a minor species (under 3% of Li2S6; Supplementary Fig. 4).

b

2.6

c

Discharge

a

469 S8

Voltage (V) versus Li/Li+

2.4 508 Li2S8

Li2S8 Li2S6 2.2 Li2S4

2.0

501 Li2S4

399 Li2S6

S8 Li2S8 Li2S6 Li2S4

1.8 0

–1 –2 Jgeo (mA cm–2)

400

500

Raman shift (cm–1)

0

100 0

100

Normalized concentration (%)

Fig. 3 | In situ Raman results during discharge with the N,S–HGF catalytic electrode. a,b, CV profile (a) and experimental in situ Raman spectra (b), with colours corresponding to voltages. The Raman cell was run with a discharge CV scan at 0.05 mV s−1 when data were being collected (see Methods for details). Characteristic peaks used to quantify intermediates are marked in corresponding colour shades. Small labels with darker colour indicate

computed frequency values. c, Comparison between voltage-dependent experimental concentrations (left) derived from in situ Raman spectra in b and simulated concentrations from DFT (right). The normalized concentration of each species is calculated by dividing the concentration at each voltage by the highest concentration of that particular species over the whole discharge process.

LiS3• had negligible impact on the equilibrium of other species or the overall reaction network, and is thus omitted in our subsequent analysis. The voltage-dependent concentration profile of each LiPS derived from the peak area (Fig. 3c, left) was further compared with the computational results (Fig. 3c, right). The comparison showed a similar sequence of concentration evolution for S8, Li2S8, Li2S6 and Li2S4 with decreasing potential. Notably, Li2S6 appeared at a potential similar to that of Li2S8 and was depleted at a potential similar to that of Li2S4, suggesting the dynamic balance among these three species through the comproportionation/disproportionation reactions. We also note that the experimental peak for Li2S4 appeared at a slightly higher potential value (by 0.09 V) than the theoretical prediction. Only thermodynamics was considered in the simulated voltage-dependent equilibrium concentration whereas the formation of Li2S6 could have been slow because of kinetics and diffusion barriers, and thus Li2S4 accumulation may have started at a slightly higher voltage. This further validates the origin of Li2S6 from the comproportionation reaction. In addition, the low solubility of Li2S4 in the electrolyte (10 mM for Li2S4 versus 1 M for Li2S6 and 0.5 M for Li2S8; Supplementary Note 2) may also have contributed to an apparent shift in the Li2S4 profile: taking the solubility limit into account shifts the onset of the simulated Li2S4 peak position from about 2.25 V to roughly 2.35 V, and a better match with experimental results (Supplementary Fig. 5). Together, these in situ Raman spectroscopy analyses provide semiquantitative polysulfide tracking, in agreement with DFT computed values and thus robustly validating the SRR molecular pathway obtained by theory: S8 → Li2S8 → 2Li2S4 (Li2S8 + Li2S4 ⇄ 2Li2S6) → 8Li2S. On the basis of Raman studies, we also estimated the relative concentration ratio of 1.7:5.5:0.4 for the maximum concentration point of Li2S8:Li2S6:Li2S4 (Supplementary Table 2). Thus, both theoretical

and experimental studies suggest that Li2S6 represents the dominant species during discharge and is a major contributor to the shuttling effect.

The role of catalysis in the SRR network To further understand the influence of electrocatalysts on the SRR network, the non-doped HGF was studied as a less active catalytic system for comparison with N,S–HGF (Supplementary Fig. 6). Overall, our in situ Raman spectra studies (Supplementary Fig. 7) show that HGF has a similar polysulfide evolution sequence, Li2S8 → Li2S6 → Li2S4, but with different voltage ranges for each species (Fig. 4). The first step in discharge at high potential was weakly modified between the two different catalysts, the peak centre (Supplementary Table 3) for Li2S8 being slightly delayed in HGF (2.29 V) compared with N,S–HGF (2.32 V). Transformation of Li2S6 was more sensitive and markedly delayed in HGF, with an average peak value at 2.19 V (compared with 2.27 V in N,S–HGF) and with about 20% (relative to the peak concentration of Li2S6) remaining at 1.80 V in HGF (compared with less than 3% remaining at 1.80 V in N,S–HGF). A similar delay was observed for Li2S4, from higher overpotential in the later steps, with an average peak value at 2.12 V (compared with 2.14 V in N,S–HGF) and with more than 30% remaining at 1.80 V in HGF (compared with 17% remaining in N,S–HGF). The delayed depletion of Li2S4 and Li2S6 until a much lower potential with non-doped HGF electrodes implies a more sluggish conversion kinetics to lower-order polysulfide, potentially worsening the shuttling effect. Our calculations described above indicate that direct electrochemical reduction of Li2S6 to lower-order LiPSs above 1.97 V is unfavourable. Rather, elimination of Li2S6 relies on a thermodynamically unfavourable disproportionation reaction (2Li2S6 → Li2S8 + Li2S4) that requires rapid Nature | Vol 626 | 1 February 2024 | 101

Article a Jgeo (mA cm–2)

0 N,S–HGF Li2S6

–0.8

Li2S8

Li2S4

–1.6

Normalized concentration (%)

b

Normalized concentration (%)

c

Jgeo (mA cm–2)

d

80

Exp.

40 0 80

S8 Li2S8

40

Li2S6 Li2S4

Sim.

0 0 HGF Li2S6

–0.8

Li2S8

Li2S4 –1.6

Normalized concentration (%)

e

Normalized concentration (%)

f

80

Exp.

40 0 80

Sim.

40 0 1.8

2.0

2.2 Voltage (V) versus Li/Li+

2.4

2.6

Fig. 4 | Comparison of different catalysts in SRR. a,d, Experimental CV curves for N,S–HGF (a) and HGF (d). b,e, Voltage-dependent concentration for each polysulfide species in N,S–HGF (b) and HGF (e) derived from experimental in situ Raman spectra. The Raman cell was run with a discharge CV scan at 0.05 mV s−1 when Raman data were being collected (see Methods for details). c,f, Simulated voltage-dependent concentration for each polysulfide species in N,S–HGF (c) and HGF (f). The normalized concentration of each species is

calculated by dividing the concentration at each voltage by the highest concentration of that particular species over the whole discharge process. b,c,e,f, Coloured bars represent the peak centres of voltage-dependent concentration curves (Supplementary Table 3) for each species: orange for Li2S8, yellow for Li2S6 and blue for Li2S 4. Solid lines and bars in b,e represent the experimental results, and dashed lines and bars in c,f represent the simulated results.

depletion of Li2S4 to proceed. In this case, a slower conversion kinetics of Li2S4 could seriously delay the consumption of Li2S6 until a much lower potential regime (under 1.97 V) in which direct electrochemical reduction of Li2S6 may also start to occur. The conversion of Li2S4 thus dominates electrocatalytic performance, especially at the start of the second stage of SRR, which is consistent with the previously reported critical role of Li2S4 in SRR overpotential20. Overall, the slower conversion kinetics of Li2S6 resulted in its wider accumulation potential range, which exacerbated the polysulfide shuttling problem. This is also reflected by the reduced charge number at high potential and within the overall SRR process: 2.36 and 9.24 electrons per S8 molecule in the HGF system compared with 3.61 and 14.17 in the N,S–HGF system (Supplementary Table 1). These CV and

in situ Raman studies showed distinct SRR kinetics between HGF and N,S–HGF, highlighting the fundamental benefits yielded by efficient catalysts. To understand the distinct potential ranges for these two electrocatalysts, we further investigated the reaction pathways of the second stage—that is, the conversion from Li2S4 to Li2S. Considering all possible 2e−, 4e− and 6e− steps starting from Li2S4 (Fig. 5), we examined a total of 12 different reaction pathways in the presence of various catalyst sites: armchair edge of graphene, zigzag edge and inner defects in the graphene plane with various doping situations (non-doped, S-doped, N-doped and N,S-doped; Supplementary Note 3). Two pathways were found to yield the highest output potential: (1) one 4e− step: Li2S4 + 4Li+ + 4e− → Li2S2 + 2Li2S, followed by one 2e− step: Li2S2 + 2Li+ + 2e− → 2Li2S

102 | Nature | Vol 626 | 1 February 2024

a

b

Li2S4 Li2S3 + Li2S

2Li2S2

Output potential (V)

2.3 2.2 2.1 2.0

c 2.3 2e– step

O

4e–

S

step

6e– step

Li 4Li2S

Output potential (V)

C H

A Z D

1.9 1.8 –3.0

Li2S2 + 2Li2S

Li2S4 + Li+ + e– → LiS2* + Li2S2 LiS2* + Li+ + e– → LiS* + LiS* LiS* + Li+ + e– → Li2S LiS* + Li+ + e– → Li2S Li2S2 + Li+ + e– → LiS* + Li2S LiS* + Li+ + e– → Li2S

2.2 2.1 2.0 1.9 1.8 –3.0

HGF S–HGF N–HGF N,S–HGF

–2.5 –2.0 –1.5 –1.0 LiS adsorption energy (eV) Li2S4 + Li+ + e– → LiS3* + Li2S LiS3* + Li+ + e– → LiS2* + LiS* LiS2* + Li+ + e– → LiS* + LiS* LiS* + Li+ + e– → Li2S LiS* + Li+ + e– → Li2S LiS* + Li+ + e– → Li2S A Z D

HGF S–HGF N–HGF N,S–HGF

–2.5 –2.0 –1.5 –1.0 LiS adsorption energy (eV)

Fig. 5 | Simulated site-specific output potential of Li2S 4 → Li2S conversion. a, Different potential combinations of 2e−, 4e− and 6e− steps considered for the second stage of SRR, the conversion of Li2S 4 to Li2S. Red, yellow and blue indicate 2e−, 4e− and 6e− steps, respectively. b,c, Simulated multistep output potential from Li2S 4 to Li2S for the two pathways with highest output potentials

considering various active sites on different catalytic electrode sites: armchair edge (A, triangles), zigzag edge (Z, squares) and inner defect sites (D, filled circles). Four types of dopant were considered: non-doped (black), S (orange), N (blue) and N,S (red). Asterisks indicate surface adsorbate species.

(Fig. 5b); and (2) a pathway consisting of one 6e− step—that is, at least one intermediate species is adsorbed on the surface during the reduction process: Li2S4 + 6Li+ + 6e− → 4Li2S (Fig. 5c). The results clearly show that, in both pathways, N,S-codoped sites exhibited higher output potential, 2.11 V compared with 2.03 V for non-doped sites, aligning with experimental results and showcasing N,S–HGF’s superior performance16. Interestingly, the inner defect sites (Supplementary Figs. 8 and 9) appear closer to the top of the output potential plot (circles in Fig. 5b,c) compared with armchair and zigzag edge sites (triangles and squares, respectively), confirming the effectiveness of defect engineering, together with heteroatom doping, in holey graphene for electrocatalysis. It is worth mentioning that conversion from LiS to Li2S solid is the potential limiting step for most of the sites with relatively large output potentials (approximately 2.1 V), with one site having the adsorption of LiS as the potential limiting step. This indicates that the final conversion from Li2S2 to Li2S is the potential limiting step, consistent with the experimental observation in which the final steps exhibited the largest overpotential16. Therefore, LiS adsorption energy can be used as a descriptor to classify site output potentials. Moreover, a smaller output potential in the second stage has far-reaching effects: the sluggish conversion of Li2S4 to lower-order polysulfides could markedly retard the thermodynamically unfavourable disproportionation reaction (2Li2S6 → Li2S8 + Li2S4), the essential path for consumption of Li2S6. Consistent with the experimental results, the simulated potential-dependent concentrations for the HGF electrode show more sluggish conversion: lower depletion potentials for Li2S4 and Li2S6 species (1.85 and 2.00 V for HGF versus 2.00 and 2.05 V for N,S–HGF; Fig. 4). The simulation of HGF and N,S–HGF effective output potential differs significantly only in the second stage, largely comparable to experimental results. Such a close correlation between experiment and theory further validates the electrocatalytic strategy for improved Li-S

batteries: an accelerated polysulfide conversion kinetics can not only produce a larger output potential, but also narrow the potential range in which LiPSs could appear and effectively mitigate the polysulfide shuttling effect.

Conclusion In conclusion, a combined experimental and theoretical investigation has allowed us to decipher and establish the complex reaction network for the 16-electron SRR, showing two stages separated by the central Li2S4 intermediate. Our studies indicate that Li2S4 and Li2S6 represent the dominant intermediates, in which Li2S4 is the key electrochemical intermediate controlling the overall SRR kinetics—particularly in the more sluggish second reduction stage; Li2S6 is generated by the comproportionation reaction between Li2S8 and Li2S4, does not directly participate in electrochemical reactions, but contributes substantially to the shuttling effect because of its high solubility and energetically favourable accumulation in the electrolyte. It is found that the optimized N,S–HGF electrocatalytic electrode markedly accelerates the conversion of high-order LiPSs, leading to more rapid depletion of soluble LiPSs at a higher potential regime and hence mitigating the polysulfide shuttling effect and boosting output potential. This study resolves the fundamental reaction network in SRR and offers valuable insights into electrocatalyst design for improved Li-S batteries. For example, considering the origin of Li2S6 (Li2S8 + Li2S4 → 2Li2S6), designing electrocatalysts with enhanced Li2S8 or Li2S4 adsorption could restrain these species on the catalyst surface, thereby suppressing comproportionation reactions and limiting Li2S6 formation. Furthermore, this methodology can be applied to understanding the sulfur evolution reaction to guide the design of bifunctional sulfur catalysts for acceleration of both SRR and sulfur evolution reaction processes, which is essential for the development of robust Li-S batteries. Nature | Vol 626 | 1 February 2024 | 103

Article 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/s41586-023-06918-4. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

11.

12. 13. 14.

15.

16. 17.

Mikhaylik, Y. V. & Akridge, J. R. Polysulfide shuttle study in the Li/S battery system. J. Electrochem. Soc. 151, A1969–A1976 (2004). Yin, Y. X., Xin, S., Guo, Y. G. & Wan, L. J. Lithium-sulfur batteries: electrochemistry, materials, and prospects. Angew. Chem. Int. Edn Engl. 52, 13186–13200 (2013). Pang, Q., Liang, X., Kwok, C. Y. & Nazar, L. F. Advances in lithium-sulfur batteries based on multifunctional cathodes and electrolytes. Nat. Energy 1, 16132 (2016). Barchasz, C. et al. Lithium/sulfur cell discharge mechanism: an original approach for intermediate species identification. Anal. Chem. 84, 3973–3980 (2012). Cuisinier, M. et al. Sulfur speciation in Li-S batteries determined by operando X-ray absorption spectroscopy. J. Phys. Chem. Lett. 4, 3227–3232 (2013). Wild, M. et al. Lithium sulfur batteries, a mechanistic review. Energy Environ. Sci. 8, 3477–3494 (2015). Bruce, P. G., Freunberger, S. A., Hardwick, L. J. & Tarascon, J. M. Li-O2 and Li-S batteries with high energy storage. Nat. Mater. 11, 19–29 (2011). Manthiram, A., Fu, Y., Chung, S. H., Zu, C. & Su, Y. S. Rechargeable lithium-sulfur batteries. Chem. Rev. 114, 11751–11787 (2014). Ji, X., Lee, K. T. & Nazar, L. F. A highly ordered nanostructured carbon-sulphur cathode for lithium-sulphur batteries. Nat. Mater. 8, 500–506 (2009). Jayaprakash, N., Shen, J., Moganty, S. S., Corona, A. & Archer, L. A. Porous hollow carbon@sulfur composites for high-power lithium-sulfur batteries. Angew. Chem. Int. Edn Engl. 50, 5904–5908 (2011). Wang, N. et al. Thickness-independent scalable high-performance Li-S batteries with high areal sulfur loading via electron-enriched carbon framework. Nat. Commun. 12, 4519 (2021). Wang, L. et al. A quantum-chemical study on the discharge reaction mechanism of lithiumsulfur batteries. J. Energy Chem. 22, 72–77 (2013). Zou, Q. & Lu, Y. C. Solvent-dictated lithium sulfur redox reactions: an operando UV-vis spectroscopic study. J. Phys. Chem. Lett. 7, 1518–1525 (2016). He, Q., Freiberg, A. T. S., Patel, M. U. M., Qian, S. & Gasteiger, H. A. Operando identification of liquid intermediates in lithium-sulfur batteries via transmission UV-vis spectroscopy. J. Electrochem. Soc. 167, 080508 (2020). Zheng, D. et al. Investigation of the Li-S battery mechanism by real-time monitoring of the changes of sulfur and polysulfide species during the discharge and charge. ACS Appl. Mater. Interfaces 9, 4326–4332 (2017). Peng, L. et al. A fundamental look at electrocatalytic sulfur reduction reaction. Nat. Catal. 3, 762–770 (2020). Zhou, G. M., Paek, E., Hwang, G. S. & Manthiram, A. Long-life Li/polysulphide batteries with high sulphur loading enabled by lightweight three-dimensional nitrogen/sulphurcodoped graphene sponge. Nat. Commun. 6, 7760 (2015).

104 | Nature | Vol 626 | 1 February 2024

18. Yuan, Z. et al. Powering lithium-sulfur battery performance by propelling polysulfide redox at sulfiphilic hosts. Nano Lett. 16, 519–527 (2016). 19. Peng, H. J. et al. Enhanced electrochemical kinetics on conductive polar mediators for lithium-sulfur batteries. Angew. Chem. Int. Edn Engl. 55, 12990–12995 (2016). 20. Wang, L. et al. Li2S4 anchoring governs the catalytic sulfur reduction on defective SmMn2O5 in lithium-sulfur battery. Adv. Energy Mater. 12, 2200340 (2022). 21. Zhou, T. et al. Twinborn TiO2-TiN heterostructures enabling smooth trapping-diffusionconversion of polysulfides towards ultralong life lithium-sulfur batteries. Energy Environ. Sci. 10, 1694–1703 (2017). 22. Zhou, J. et al. Deciphering the modulation essence of p bands in Co-based compounds on Li-S chemistry. Joule 2, 2681–2693 (2018). 23. Hua, W. et al. Selective catalysis remedies polysulfide shuttling in lithium‐sulfur batteries. Adv. Mater. 33, 2101006 (2021). 24. Zhou, G. et al. Theoretical calculation guided design of single-atom catalysts toward fast kinetic and long-life Li-S batteries. Nano Lett. 20, 1252–1261 (2019). 25. Wang, L. et al. Design rules of a sulfur redox electrocatalyst for lithium-sulfur batteries. Adv. Mater. 34, 2110279 (2022). 26. Lu, Y.-C., He, Q. & Gasteiger, H. A. Probing the lithium-sulfur redox reactions: a rotating-ring disk electrode study. J. Phys. Chem. C Nanomater. Interfaces 118, 5733–5741 (2014). 27. Conder, J. et al. Direct observation of lithium polysulfides in lithium-sulfur batteries using operando X-ray diffraction. Nat. Energy 2, 17069 (2017). 28. Wujcik, K. H. et al. Fingerprinting lithium-sulfur battery reaction products by X-ray absorption spectroscopy. J. Electrochem. Soc. 161, A1100–A1106 (2014). 29. Hou, T. Z. et al. Lithium bond chemistry in lithium-sulfur batteries. Angew. Chem. Int. Edn Engl. 129, 8290–8294 (2017). 30. Hagen, M. et al. In-situ Raman investigation of polysulfide formation in Li-S cells. J. Electrochem. Soc. 160, A1205–A1214 (2013). 31. Zhou, G. et al. Catalytic oxidation of Li2S on the surface of metal sulfides for Li-S batteries. Proc. Natl Acad. Sci. USA 114, 840–845 (2017). 32. Wujcik, K. H. et al. Characterization of polysulfide radicals present in an ether-based electrolyte of a lithium-sulfur battery during initial discharge using in situ X-ray absorption spectroscopy experiments and first‐principles calculations. Adv. Energy Mater. 5, 1500285 (2015). 33. Rajput, N. N. et al. Elucidating the solvation structure and dynamics of lithium polysulfides resulting from competitive salt and solvent interactions. Chem. Mater. 29, 3375–3379 (2017). 34. Assary, R. S., Curtiss, L. A. & Moore, J. S. Toward a molecular understanding of energetics in Li-S batteries using nonaqueous electrolytes: a high-level quantum chemical study. J. Phys. Chem. C Nanomater. Interfaces 118, 11545–11558 (2014). 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 selfarchiving 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 Synthesis of graphene oxide and heteroatom-doped holey graphene framework The heteroatom-doped HGF can function as a conductive scaffold with flexible pore structure and exhibit tunable electrocatalytic properties in SRR16,17, and hence serves as a good model system for investigation of the fundamental reaction mechanism. Furthermore, compared with metal oxides or sulfides that are usually in the form of nanoparticles and that require additional conductive additives and binders, the free-standing HGF electrode could offer a more robust and simplified binder-free model system for systematic investigation of the SRR process. The specific heteroatom-doped N,S–HGF was chosen for its superior SRR kinetics, markedly low activation energy and excellent performance in Li-S batteries16, and non-doped HGF was selected as a control for comparison to evaluate the role of the catalyst in modification of the kinetics along the complex SRR network. A typical hydrothermal process was used to construct the HGF structure, with interconnected micro- and nanopores. Graphene oxide was prepared using a modified Hummers’ method35. Briefly, 6 g of natural graphite (325 mesh, Sigma-Aldrich) was added to 140 ml of concentrated sulfuric acid under vigorous stirring in an ice-water bath followed by the slow addition of 3 g of sodium nitrate (Sigma-Aldrich) and 18 g of potassium permanganate (Sigma-Aldrich). Owing to the strong acidity of sulfuric acid and highly oxidative nature of sodium nitrate and potassium permanganate, it is necessary to keep the temperature near 0 °C to slow the oxidation process and avoid potential safety concerns. Following stirring for 30 min, the reaction system was transferred to a water bath at about 50 °C under continuous stirring until the mixture had formed a thick paste. The system was then transferred back to the ice-water bath followed by the dropwise addition of around 1 l of iced deionized water. The mixture was then centrifuged and washed three times with 1:10 HCl aqueous solution followed by repeated washing with deionized water. The final solution was dialysed for 1 week to remove extra H+ ions absorbed on graphene oxide surfaces. Heteroatom-doped HGFs were synthesized by reacting the dopant sources and H2O2 with graphene oxide aqueous dispersion through a typical hydrothermal method. N,S–HGF was synthesized by mixing 30 ml of 2 mg ml−1 graphene oxide aqueous dispersion solution, 45 μl of 30% H2O2 aqueous solution and 3.0 g of NH4SCN, followed by sonication and hydrothermal reaction at 180 °C for 6 h in an autoclave, to produce a free-standing N,S–HGF hydrogel. The hydrogel was then freeze-dried and annealed at 900 °C in Ar for 1 h to obtain the N,S–HGF aerogel. The control sample HGF was synthesized by substitution of dopant sources with ascorbic acid, following the same procedures. Preparation of electrolyte and Li2Sx solutions The electrolyte (denoted as blank electrolyte) comprised 1 M lithium bis(trifluoromethanesulfonyl)imide (Sigma-Aldrich) and 0.2 M lithium nitrate (Sigma-Aldrich) in the mixed 1,2-dimethoxyethane (Sigma-Aldrich) and 1,3-dioxolane (Sigma-Aldrich) solution (1:1 by volume). The Li2S6 catholyte (0.1 M) was prepared by reacting the sublimed sulfur (Sigma-Aldrich) with Li2S (Sigma-Aldrich) in stoichiometric proportion in the blank electrolyte. The mixture was vigorously stirred at 50 °C in an Ar-filled glovebox overnight to produce a brownish-red Li2S6 catholyte solution. Electrochemical measurements The electrochemical performance of the catalyst was conducted in CR2032 coin cells assembled in an Ar-filled glovebox. The catalyst electrode was prepared by direct pressing of the aerogel into a free-standing thin film. Next, the catholyte (0.1 M Li2S6 in the blank electrolyte) was used directly as the sulfur source for drop-casting in the catalyst electrode. In our experiment we specifically used a low mass ratio of sulfur (33%) in the cathodes to ensure complete conversion from S8 to Li2S for

mechanistic understanding. The sulfur cathodes were then directly assembled into a CR2032 coin cell with lithium foil, Celgard 2500 separator and blank electrolyte. To better probe the different stages in the SRR we performed CV measurements followed by quantitative charge analysis, by integration of the peak area in different potential regions. CV curves were recorded in the voltage range 1.6–2.8 V at a scanning rate of 0.05 mV s−1. The charge number transferred (Q) for the electrochemical steps in the SRR can be calculated from CV curves. Q is calculated by integrating the area enclosed in CV: t2

Q = ∫ i (t) dt = t1

1 v

V2

A

V2

∫ i (V ) dV = v ∫ j (V ) dV =

V1

V1

AS v

dV dt = v where Q is charge (C), i is current intensity (A), j is current density (A cm−2), A is the geometric area of the electrode (cm2), v is scan rate (V s−1), t is time (s) and S is integrated area.

In situ Raman spectroscopy Given the specific Raman activity of elemental sulfur and polysulfides, in situ Raman spectroscopy offers an attractive technique for identification and tracking of the polysulfide conversion process30,36–41. Raman spectroscopy measurements were collected using a liquid nitrogen-cooled, charge-coupled device array detector with a Horiba Jobin Yvon T64000 open-frame confocal microscope using the ×10 objective, followed by a triple monochromator resulting in high spectral resolution (down to 0.15 cm−1). The sample was subject to a 514 nm laser for 5 s and averaged 70 times. Data were collected with a 1,800 cm−1 grating and 500 μm slit. For in situ Raman spectroscopy, a regular coin cell was modified with a transparent window on the cathode side to allow ingress of the laser. The same procedure as described for electrochemical measurements was used to assemble the Raman cell. The laser was focused on the electrolyte near the boundary between HGF and electrolyte (Supplementary Fig. 3). The Raman cell was run with a discharge CV scan at 0.05 mV s−1 when data were being collected. The resulting Raman spectra (Supplementary Fig. 10) were carefully corrected and deconvolved. Because the blank electrolyte shows multiple peaks in our spectral range of interest, in situ data were first corrected by ‘blank’ subtraction. Specifically, a blank spectrum was collected using the same cell set-up but without active sulfur being added and subtracted from the in situ data, to eliminate the influence from electrolyte peaks. Subsequently the in situ data were corrected by subtracting the baseline. The blank/baseline subtraction was conducted systematically using adaptive iteratively reweighted penalized least squares42 fitting implemented in the pybaseline package. The corrected Raman spectra were then loaded in LabSpec software to conduct peak deconvolution and assignments for different polysulfides. We note that peak assignment for different polysulfides is complicated and divergent in various studies30,36,39,43, owing to the instability of various polysulfides and absence of pure polysulfide standard samples. Peaks used in our analysis were selected based on their behaviour during the SRR process, previous reports30,36,39,43 and our DFT simulated results. We deconvoluted all Raman peaks and analysed multiple peaks for each given LiPS to confirm the validity of the assignments (Supplementary Figs. 11 and 12), and primarily used the peaks at 501, 399, 508 and 469 cm−1 for quantification of Li2S4, Li2S6, Li2S8 and S8, respectively. DFT calculations Calculations were performed with DFT44,45 using the Vienna Ab initio Simulation Package46. The strongly constrained and appropriately normed functional47 was used, and further details are provided in

Article Supplementary Note 4 and Supplementary Table 4. Benchmarking and comparison between different functionals are provided in Supplementary Note 5 and Supplementary Tables 5 and 6. Solvation effects are described using a microsolvation model: the first solvation shell is described using explicit 1,3-dioxolane molecules and the remainder by an implicit dielectric model as implemented in the VASPsol48 add-on package (microsolvated structures are given in Supplementary Fig. 13 and Supplementary Tables 7 and 8). We note that the choice of the explicit solvent molecule is an approximation, and we also performed calculations with explicit 1,2-dimethoxyethane solvation for Li2S4, which showed the same Li-O coordination number with very similar energies (Supplementary Note 6 and Supplementary Table 9). Details of Gibbs free energies of polysulfide species are provided in Supplementary Note 7 and Supplementary Tables 10–15, and reaction Gibbs free energies in Supplementary Note 1. Details of potential-dependent concentration calculations and simulated CV curves are provided in Supplementary Note 8. The possible effects of solubility on the simulated potential-dependent concentration profile are discussed in Supplementary Note 2. Adsorption models (Supplementary Figs. 8, 9 and 14) are discussed in Supplementary Note 3. Volcano plot details (Supplementary Fig. 15) are discussed in Supplementary Note 9. Raman intensity (Supplementary Fig. 16) was calculated within the double-harmonic approximation, and details are provided in Supplementary Note 10 and Supplementary Table 16. A sensitivity analysis is provided in Supplementary Note 11 and Supplementary Fig. 17, showing that theoretical conclusions are unaffected by a reasonable change of calculated energies (up to 0.15 eV) arising from the considered approximations.

Data availability The data that support the findings of this study are available in the main text, figures and Supplementary Information files.  Source data are provided with this paper. All relevant data are available from the corresponding authors on request.

Code availability The code used for simulated voltage-dependent concentrations and CV curves is available at https://github.com/lophocinalis/ concentration_cv.

35. Xu, Y., Sheng, K., Li, C. & Shi, G. Self-assembled graphene hydrogel via a one-step hydrothermal process. ACS Nano 4, 4324–4330 (2010). 36. Chen, J.-J. et al. Conductive lewis base matrix to recover the missing link of Li2S8 during the sulfur redox cycle in Li-S battery. Chem. Mater. 27, 2048–2055 (2015). 37. Zhu, X. et al. A highly stretchable cross-linked polyacrylamide hydrogel as an effective binder for silicon and sulfur electrodes toward durable lithium-ion storage. Adv. Funct. Mater. 28, 1705015 (2018). 38. Wu, H. L., Huff, L. A. & Gewirth, A. A. In situ Raman spectroscopy of sulfur speciation in lithium-sulfur batteries. ACS Appl. Mater. Interfaces 7, 1709–1719 (2015). 39. Lei, T. et al. Inhibiting polysulfide shuttling with a graphene composite separator for highly robust lithium-sulfur batteries. Joule 2, 2091–2104 (2018). 40. Chen, W. et al. A new hydrophilic binder enabling strongly anchoring polysulfides for high-performance sulfur electrodes in lithium-sulfur battery. Adv. Energy Mater. 8, 1702889 (2018). 41. Hannauer, J. et al. The quest for polysulfides in lithium-sulfur battery electrolytes: an operando confocal Raman spectroscopy study. ChemPhysChem 16, 2709–2709 (2015). 42. Zhang, Z.-M., Chen, S. & Liang, Y.-Z. Baseline correction using adaptive iteratively reweighted penalized least squares. Analyst 135, 1138–1146 (2010). 43. Zhu, W. et al. Investigation of the reaction mechanism of lithium sulfur batteries in different electrolyte systems by in situ Raman spectroscopy and in situ X-ray diffraction. Sustain. Energy Fuels 1, 737–747 (2017). 44. Hohenberg, P. & Kohn, W. Inhomogeneous electron gas. Phys. Rev. 136, B864 (1964). 45. Kohn, W. & Sham, L. J. Self-consistent equations including exchange and correlation effects. Phys. Rev. 140, A1133 (1965). 46. Kresse, G. & Hafner, J. Ab initio molecular dynamics for liquid metals. Phys. Rev. B 47, 558 (1993). 47. Sun, J., Ruzsinszky, A. & Perdew, J. P. Strongly constrained and appropriately normed semilocal density functional. Phys. Rev. Lett. 115, 036402 (2015). 48. Mathew, K., Sundararaman, R., Letchworth-Weaver, K., Arias, T. & Hennig, R. G. Implicit solvation model for density-functional study of nanocrystal surfaces and reaction pathways. J. Chem. Phys. 140, 084106 (2014). Acknowledgements This work is supported by the Center for Synthetic Control Across Length-scales for Advancing Rechargeables, an Energy Frontier Research Center funded by the US Department of Energy, Office of Science Basic Energy Sciences programme under award no. DE-SC0019381. Author contributions X.D. conceived the research. X.D. and R.L. designed the experimental research. P.S. and Z. Wei designed and performed density functional theory (DFT) calculations. R.L. performed experiments and conducted data analysis, with contributions from Z. Wei, L.P., L.Z., P.W., C.W., D.Z., H.L., Z. Wang, S.T., B.D., Y.H., P.S. and X.D. A.Z. and R.S. contributed to Raman data collection. R.L. and Z. Wei wrote the original draft. X.D. and P.S. revised the manuscript. All authors discussed the results and commented on 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/s41586-023-06918-4. Correspondence and requests for materials should be addressed to Philippe Sautet or Xiangfeng Duan. Peer review information Nature thanks Mahbub Islam, Quan-Hong Yang 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.

Article

Flexible silicon solar cells with high power-to-weight ratios https://doi.org/10.1038/s41586-023-06948-y Received: 24 October 2022 Accepted: 6 December 2023 Published online: 31 January 2024 Check for updates

Yang Li1,2,3,5, Xiaoning Ru2,5, Miao Yang2,5, Yuhe Zheng1,5, Shi Yin2, Chengjian Hong2, Fuguo Peng2, Minghao Qu2, Chaowei Xue2, Junxiong Lu2, Liang Fang2, Chao Su1, Daifen Chen1 ✉, Junhua Xu3 ✉, Chao Yan3 ✉, Zhenguo Li2 ✉, Xixiang Xu2 ✉ & Zongping Shao4 ✉

Silicon solar cells are a mainstay of commercialized photovoltaics, and further improving the power conversion efficiency of large-area and flexible cells remains an important research objective1,2. Here we report a combined approach to improving the power conversion efficiency of silicon heterojunction solar cells, while at the same time rendering them flexible. We use low-damage continuous-plasma chemical vapour deposition to prevent epitaxy, self-restoring nanocrystalline sowing and vertical growth to develop doped contacts, and contact-free laser transfer printing to deposit low-shading grid lines. High-performance cells of various thicknesses (55–130 μm) are fabricated, with certified efficiencies of 26.06% (57 μm), 26.19% (74 μm), 26.50% (84 μm), 26.56% (106 μm) and 26.81% (125 μm). The wafer thinning not only lowers the weight and cost, but also facilitates the charge migration and separation. It is found that the 57-μm flexible and thin solar cell shows the highest power-to-weight ratio (1.9 W g−1) and open-circuit voltage (761 mV) compared to the thick ones. All of the solar cells characterized have an area of 274.4 cm2, and the cell components ensure reliability in potential-induced degradation and light-induced degradation ageing tests. This technological progress provides a practical basis for the commercialization of flexible, lightweight, low-cost and highly efficient solar cells, and the ability to bend or roll up crystalline silicon solar cells for travel is anticipated.

Crystalline silicon (c-Si) solar cells have been the mainstay of green and renewable energy3, accounting for 3.6% of global electricity generation and becoming the most cost-effective option for new electricity generation in most of the world4. Although c-Si solar cells now account for more than 95% of the market for solar cells, which usually have a wafer thickness of 150–180 μm, their use is unfeasible in some extreme application scenarios, such as satellites, spacecraft and unmanned aerial vehicles, and there is a need for further weight reduction and flexibility of solar cells5. Thus, reducing the thickness of the c-Si wafer to much thinner than that in typical c-Si solar cells, and thereby incorporating the advantages of ‘thin-film solar cells’ into c-Si solar cells, is the focus of much research1,6,7. However, the power conversion efficiencies (PCEs) of all of the thin c-Si solar cells (55–130 μm) studied have remained in the range of 23.27–24.70% for decades8–13. Recently, front-back contact silicon heterojunction (SHJ) solar cells have become a formidable contender for the next generation of photovoltaic devices owing to their advantages in double-sided power generation, low cost and scalable production, compared to the interdigitated back contact configurations14. To further improve the performance of front-back contact SHJ solar cells on the premise of bendable thicknesses (26% PCE with thicknesses in the range of 55–130 μm, possessing features of both high PCE and flexibility, can be produced. Therefore, flexibility must be taken into consideration as an important factor. We divided the c-Si solar cells into categories according to the minimum bending radius of curvature (rb): nonflexible cells (rb > 63 mm) with thicknesses of >150 μm; semiflexible (SF) cells (38 mm