Spherical nucleic acids. Volume 2 9781003056690, 9789814800358, 9780429200151, 9789814877220, 9781000092363, 1000092364, 9781000092400, 1000092402, 9781000092448, 1000092445, 1003056695

Spherical Nucleic Acids (SNAs) are typically comprised of a nanoparticle core and a densely packed and highly oriented n

170 117 37MB

English Pages xii, 442 [455] Year 2020

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Spherical nucleic acids. Volume 2
 9781003056690, 9789814800358, 9780429200151, 9789814877220, 9781000092363, 1000092364, 9781000092400, 1000092402, 9781000092448, 1000092445, 1003056695

Table of contents :
Cover......Page 1
Half Title......Page 3
Title Page......Page 5
Copyright Page......Page 6
Table of Contents......Page 7
Part 3: Design Rules for Colloidal Crystallization......Page 15
20: The Structural Characterization of Oligonucleotide- Modified Gold Nanoparticle Networks Formed by DNA Hybridization......Page 17
20.1 Introduction......Page 18
20.2.3 SAXS Measurements......Page 21
20.3.1 Scattering from Dispersed DNA-Modified Nanoparticles......Page 22
20.3.2 Nanoparticle Aggregates Formed from Different-Length DNA Interconnects and Different-Sized Nanoparticles......Page 24
20.3.3 Particle Packing within DNA-Linked Nanoparticle Assemblies......Page 27
20.3.4 The Effect of Ionic Strength......Page 28
20.3.5 The Effect of Single-Strand Spacer DNA......Page 29
20.4 Conclusions......Page 31
21: DNA-Programmable Nanoparticle Crystallization......Page 35
22: Establishing the Design Rules for DNA-Mediated Programmable Colloidal Crystallization......Page 47
23: Nanoparticle Superlattice Engineering with DNA......Page 59
24: Modeling the Crystallization of Spherical Nucleic Acid Nanoparticle Conjugates with Molecular Dynamics Simulations......Page 75
Part 4: Building Blocks for Crystal Engineering......Page 91
25: Synthetically Programmable Nanoparticle Superlattices Using a Hollow Three-Dimensional Spacer Approach......Page 93
25.1 Introduction......Page 94
25.2.2 Synthesis of Hollow SNAs......Page 101
25.2.5 Transmission Electron Microscopy and Electron Tomography......Page 103
26: A General Approach to DNA-Programmable Atom Equivalents......Page 107
26.1 Introduction......Page 108
26.2.2 Nanoparticle Crystallization......Page 117
26.2.5 Gel Electrophoresis and DLS Measurements......Page 118
27: DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks......Page 121
28: Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization......Page 135
28.1 Methods......Page 148
29: Programming Colloidal Crystal Habit with Anisotropic Nanoparticle Building Blocks and DNA Bonds......Page 153
30: Exploring the Zone of Anisotropy and Broken Symmetries in DNA-Mediated Nanoparticle Crystallization......Page 163
30.1 Introduction......Page 164
30.2 Materials and Methods......Page 167
30.3 Results and Discussion......Page 170
31: Clathrate Colloidal Crystals......Page 179
31.1 Introduction......Page 180
31.3 Results and Discussion......Page 184
32: General and Direct Method for Preparing Oligonucleotide-Functionalized Metal-Organic Framework Nanoparticles......Page 191
32.1 Introduction......Page 192
32.2 Materials and Methods......Page 194
32.3 Results and Discussion......Page 199
33: DNA-Mediated Engineering of Multicomponent Enzyme Crystals......Page 203
33.1 Introduction......Page 204
33.2 Significance......Page 206
33.3 Results and Discussion......Page 207
33.5 Materials and Methods......Page 214
34: Altering DNA-Programmable Colloidal Crystallization Paths by Modulating Particle Repulsion......Page 223
34.1 Introduction......Page 224
34.2 Materials and Methods......Page 227
34.3 Summary......Page 235
35: Modulating Nanoparticle Superlattice Structure Using Proteins with Tunable Bond Distributions......Page 241
35.1 Introduction......Page 242
35.2 Materials and Methods......Page 244
35.3 Summary......Page 249
36: DNA-Functionalized, Bivalent Proteins......Page 253
36.1 Introduction......Page 254
36.2 Methods and Discussion......Page 255
36.3 Summary......Page 261
37: DNA-Encoded Protein Janus Nanoparticles......Page 265
37.1 Introduction......Page 266
37.2 Results and Discussion......Page 268
37.3 Colloidal Crystallization......Page 271
37.4 Conclusions......Page 277
Part 5: DNA and RNA as Programmable "Bonds"......Page 281
38: Controlling the Lattice Parameters of Gold Nanoparticle FCC Crystals with Duplex DNA Linkers......Page 283
38.1 Introduction......Page 284
38.2 Methods and Discussion......Page 285
38.3 Summary......Page 291
39: Importance of the DNA "Bond" in Programmable Nanoparticle Crystallization......Page 295
39.1 Introduction......Page 296
39.2 Results and Discussion......Page 297
39.2.1 Significance......Page 299
39.2.2 Free DNA Analogy......Page 300
39.2.3 DNA Sticky-End Sequence......Page 301
39.2.4 Number of Linkers......Page 304
39.2.5 Salt Concentration......Page 306
39.3 Conclusions......Page 310
40: Modular and Chemically Responsive Oligonucleotide "Bonds" in Nanoparticle Superlattices......Page 315
40.1 Introduction......Page 316
40.2.1 Design of Nanoparticle Superlattices with Different Oligonucleotide Bonds......Page 318
40.2.2 Synthesis and Characterization of DNA and RNA Nanoparticle Superlattices......Page 319
40.2.3 Isostructural Nanoparticle Superlattices Exhibit Tunable Responsiveness to Enzymes......Page 323
40.3 Conclusion......Page 328
41: Enzymatically Controlled Vacancies in Nanoparticle Crystals......Page 333
41.1 Introduction......Page 334
41.2 Methods......Page 336
41.3 Results and Discussion......Page 339
42: Modulating the Bond Strength of DNA-Nanoparticle Superlattices......Page 347
42.1 Introduction......Page 348
42.2.1 Thermal Stabilization of the DNA-NP Superlattice “Bond”......Page 350
42.2.2 Structural Effects of Intercalation on the DNA-NP Superlattice......Page 355
42.2.3 Synthesis of Core-Shell Superlattice......Page 357
42.4.2 Nanoparticle Functionalization......Page 359
42.4.2.2 Quantum dot NP functionalization......Page 360
42.4.4 Core-Shell Synthesis......Page 361
42.4.5 Fluorescence Binding Assay......Page 362
42.4.6 Absorbance Binding Assay......Page 363
42.4.8 Small-Angle X-ray Scattering......Page 364
42.4.9 Isotropic Strain upon Intercalation......Page 365
42.4.10 Williamson-Hall Analysis......Page 366
42.4.11 Scanning Electron Microscopy......Page 367
43: The Significance of Multivalent Bonding Motifs and "Bond Order" in DNA-Directed Nanoparticle Crystallization......Page 371
43.1 Introduction......Page 372
43.2 Methods......Page 374
43.3 Results and Discussion......Page 375
44: Oligonucleotide Flexibility Dictates Crystal Quality in DNA-Programmable Nanoparticle Superlattices......Page 383
44.1 Introduction......Page 384
44.2 Methods and Discussion......Page 386
44.3.1 Nanoparticle Crystallization......Page 394
44.3.2 Transmission Electron Microscopy......Page 395
45. Entropy-Driven Crystallization Behavior in DNA-Mediated Nanoparticle Assembly......Page 399
45.1 Introduction......Page 400
45.2 Methods......Page 402
45.3 Discussion......Page 406
45.4 Conclusion......Page 412
46. Electrolyte-Mediated Assembly of Charged Nanoparticles......Page 417
46.1 Introduction......Page 418
46.2.1 SAXS Studies of DNA-Coated AuNP Assembly......Page 420
46.2.2 MD Simulations for Potential of Mean Force between DNA-Coated AuNPs......Page 424
46.2.3 Liquid-State Theory for Like-Charged Attraction......Page 426
46.3 Conclusions......Page 430
47. The Role of Repulsion in Colloidal Crystal Engineering with DNA......Page 433
47.1 Introduction......Page 434
47.2.1 Assembly of Colloidal PAEs through DNA Hybridization Interactions......Page 438
47.2.2 Assembly of Colloidal PAEs through Depletion Forces......Page 440
47.3.1 The Role of Repulsion in PAE Assembly......Page 442
47.3.1.1 Excluded volume repulsion......Page 443
47.3.1.2 Elastic repulsion......Page 445
47.3.1.3 Repulsion from entropic effects due to counterions......Page 446
47.3.2 Potential Energy Calculation for PAE Superlattices......Page 447
47.4 Conclusion......Page 451

Citation preview

edited by

Chad A. Mirkin

Spherical Nucleic Acids Volume 2

Spherical Nucleic Acids

Spherical Nucleic Acids Volume 2

edited by

Chad A. Mirkin

Published by Jenny Stanford Publishing Pte. Ltd. Level 34, Centennial Tower 3 Temasek Avenue Singapore 039190

Email: [email protected] Web: www.jennystanford.com British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

Spherical Nucleic Acids, Volume 2 Copyright © 2020 by Jenny Stanford Publishing Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher. ISBN ISBN ISBN ISBN

978-981-4800-35-8 (Set) (Hardcover) 978-0-429-20015-1 (Set) (eBook) 978-981-4877-22-0 (Volume 2) (Hardcover) 978-1-003-05669-0 (Volume 2) (eBook)

Contents

Part 3

Design Rules for Colloidal Crystallization

20. The Structural Characterization of OligonucleotideModified Gold Nanoparticle Networks Formed by DNA Hybridization 497 So-Jung Park, Anne A. Lazarides, James J. Storhoff, Lorenzo Pesce, and Chad A. Mirkin 20.1 Introduction 498 20.2 Experimental Section 501 20.2.1 Preparation of OligonucleotideModified Gold Nanoparticles and Linker DNA 501 20.2.2 Preparation of Nanoparticle Assemblies 501 20.2.3 SAXS Measurements 501 20.3 Results and Discussion 502 20.3.1 Scattering from Dispersed DNA-Modified Nanoparticles 502 20.3.2 Nanoparticle Aggregates Formed from Different-Length DNA Interconnects and Different-Sized Nanoparticles 504 20.3.3 Particle Packing within DNA-Linked Nanoparticle Assemblies 507 20.3.4 The Effect of Ionic Strength 508 20.3.5 The Effect of Single-Strand Spacer DNA 509 20.4 Conclusions 511 21. DNA-Programmable Nanoparticle Crystallization Sung Yong Park, Abigail K. R. Lytton-Jean, Byeongdu Lee, Steven Weigand, George C. Schatz, and Chad A. Mirkin

515

vi

Contents

22. Establishing the Design Rules for DNA-Mediated Programmable Colloidal Crystallization Robert J. Macfarlane, Matthew R. Jones, Andrew J. Senesi, Kaylie L. Young, Byeongdu Lee, Jinsong Wu, and Chad A. Mirkin 23. Nanoparticle Superlattice Engineering with DNA Robert J. Macfarlane, Byeongdu Lee, Matthew R. Jones, Nadine Harris, George C. Schatz, and Chad A. Mirkin 24. Modeling the Crystallization of Spherical Nucleic Acid Nanoparticle Conjugates with Molecular Dynamics Simulations Ting I. N. G. Li, Rastko Sknepnek, Robert J. Macfarlane, Chad A. Mirkin, and Monica Olvera de la Cruz

Part 4

527

539

555

Building Blocks for Crystal Engineering

25. Synthetically Programmable Nanoparticle Superlattices Using a Hollow Three-Dimensional Spacer Approach 573 Evelyn Auyeung, Joshua I. Cutler, Robert J. Macfarlane, Matthew R. Jones, Jinsong Wu, George Liu, Ke Zhang, Kyle D. Osberg, and Chad A. Mirkin 25.1 Introduction 574 25.2 Methods 581 25.2.1 DNA Synthesis and Nanoparticle Functionalization 581 25.2.2 Synthesis of Hollow SNAs 581 25.2.3 Nanoparticle Crystallization 583 25.2.4 Small-Angle X-ray Scattering 583 25.2.5 Transmission Electron Microscopy and Electron Tomography 583 26. A General Approach to DNA-Programmable Atom Equivalents 587 Chuan Zhang, Robert J. Macfarlane, Kaylie L. Young, Chung Hang J. Choi, Liangliang Hao, Evelyn Auyeung, Guoliang Liu, Xiaozhu Zhou, and Chad A. Mirkin 26.1 Introduction 588

Contents



26.2 Methods 597 26.2.1 DNA Synthesis and Nanoparticle Functionalization 597 26.2.2 Nanoparticle Crystallization 597 26.2.3 SAXS 598 26.2.4 STEM and EDX 598 26.2.5 Gel Electrophoresis and DLS Measurements 598

27. DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks Matthew R. Jones, Robert J. Macfarlane, Byeongdu Lee, Jian Zhang, Kaylie L. Young, Andrew J. Senesi, and Chad A. Mirkin 28. Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization Matthew N. O’Brien, Matthew R. Jones, Byeongdu Lee, and Chad A. Mirkin 28.1 Methods 29. Programming Colloidal Crystal Habit with Anisotropic Nanoparticle Building Blocks and DNA Bonds Matthew N. O’Brien, Hai-Xin Lin, Martin Girard, Monica Olvera de la Cruz, and Chad A. Mirkin

601

615

628 633

30. Exploring the Zone of Anisotropy and Broken Symmetries in DNA-Mediated Nanoparticle Crystallization 643 Matthew N. O’Brien, Martin Girard, Hai-Xin Lin, Jaime A. Millan, Monica Olvera de la Cruz, Byeongdu Lee, and Chad A. Mirkin 30.1 Introduction 644 30.2 Materials and Methods 647 30.3 Results and Discussion 650 32. Clathrate Colloidal Crystals Haixin Lin, Sangmin Lee, Lin Sun, Matthew Spellings, Michael Engel, Sharon C. Glotzer, and Chad A. Mirkin

659

vii

viii

Contents



31.1 Introduction 31.2 Materials and Methods 31.3 Results and Discussion

32. General and Direct Method for Preparing Oligonucleotide-Functionalized Metal−Organic Framework Nanoparticles Shunzhi Wang, C. Michael McGuirk, Michael B. Ross, Shuya Wang, Pengcheng Chen, Hang Xing, Yuan Liu, and Chad A. Mirkin 32.1 Introduction 32.2 Materials and Methods 32.3 Results and Discussion 33. DNA-Mediated Engineering of Multicomponent Enzyme Crystals

Jeffrey D. Brodin, Evelyn Auyeung, and Chad A. Mirkin 33.1 Introduction 33.2 Significance 33.3 Results and Discussion 33.4 Conclusions 33.5 Materials and Methods

34. Altering DNA-Programmable Colloidal Crystallization Paths by Modulating Particle Repulsion Mary X. Wang, Jeffrey D. Brodin, Jaime A. Millan, Soyoung E. Seo, Martin Girard, Monica Olvera de la Cruz, Byeongdu Lee, and Chad A. Mirkin 34.1 Introduction 34.2 Materials and Methods 34.3 Summary

35. Modulating Nanoparticle Superlattice Structure Using Proteins with Tunable Bond Distributions Janet R. McMillan, Jeffrey D. Brodin, Jaime A. Millan, Byeongdu Lee, Monica Olvera de la Cruz, and Chad A. Mirkin 35.1 Introduction 35.2 Materials and Methods

660 664 664 671

672 674 679 683 684 686 687 694 694 703

704 707 715 721

722 724

Contents



36.

35.3 Summary

DNA-Functionalized, Bivalent Proteins Janet R. McMillan and Chad A. Mirkin 36.1 Introduction 36.2 Methods and Discussion 36.3 Summary

37. DNA-Encoded Protein Janus Nanoparticles Oliver G. Hayes, Janet R. McMillan, Byeongdu Lee, and Chad A. Mirkin



37.1 Introduction 37.2 Results and Discussion 37.3 Colloidal Crystallization 37.4 Conclusions

Part 5

DNA and RNA as Programmable “Bonds”

38. Controlling the Lattice Parameters of Gold Nanoparticle FCC Crystals with Duplex DNA Linkers Haley D. Hill, Robert J. Macfarlane, Andrew J. Senesi, Byeongdu Lee, Sung Yong Park, and Chad A. Mirkin 38.1 Introduction 38.2 Methods and Discussion 38.3 Summary

39. Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization Robert J. Macfarlane, Ryan V. Thaner, Keith A. Brown, Jian Zhang, Byeongdu Lee, SonBinh T. Nguyen, and Chad A. Mirkin 39.1 Introduction 39.2 Results and Discussion 39.2.1 Significance 39.2.2 Free DNA Analogy 39.2.3 DNA Sticky-End Sequence 39.2.4 Number of Linkers 39.2.5 Salt Concentration 39.3 Conclusions

729 733

734 735 741 745

746 748 751 757 763

764 765 771 775

776 777 779 780 781 784 786 790

ix

x

Contents

40. Modular and Chemically Responsive Oligonucleotide “Bonds” in Nanoparticle Superlattices Stacey N. Barnaby, Ryan V. Thaner, Michael B. Ross, Keith A. Brown, George C. Schatz, and Chad A. Mirkin 40.1 Introduction 40.2 Results and Discussion 40.2.1 Design of Nanoparticle Superlattices with Different Oligonucleotide Bonds 40.2.2 Synthesis and Characterization of DNA and RNA Nanoparticle Superlattices 40.2.3 Isostructural Nanoparticle Superlattices Exhibit Tunable Responsiveness to Enzymes 40.3 Conclusion

795

796 798 798 799 803 808

41. Enzymatically Controlled Vacancies in Nanoparticle Crystals 813 Stacey N. Barnaby, Michael B. Ross, Ryan V. Thaner, Byeongdu Lee, George C. Schatz, and Chad A. Mirkin 41.1 Introduction 814 41.2 Methods 816 41.3 Results and Discussion 819

42. Modulating the Bond Strength of DNA-Nanoparticle Superlattices 827 Soyoung E. Seo, Mary X. Wang, Chad M. Shade, Jessica L. Rouge, Keith A. Brown, and Chad A. Mirkin 42.1 Introduction 828 42.2 Results and Discussion 830 42.2.1 Thermal Stabilization of the 830 DNA-NP Superlattice “Bond” 42.2.2 Structural Effects of Intercalation on the DNA-NP Superlattice 835 42.2.3 Synthesis of Core−Shell Superlattice 837 42.3 Conclusion 839 42.4 Methods 839 42.4.1 RuII Complex Synthesis 839 42.4.2 Nanoparticle Functionalization 839 42.4.2.1 AuNP functionalization 840

Contents





42.4.2.2 Quantum dot NP functionalization 840 42.4.3 Superlattice Assembly 841 841 42.4.4 Core−Shell Synthesis 42.4.5 Fluorescence Binding Assay 842 42.4.6 Absorbance Binding Assay 843 42.4.7 Melting Transition Measurements 844 42.4.8 Small-Angle X-ray Scattering 844 42.4.9 Isotropic Strain upon Intercalation 845 846 42.4.10 Williamson−Hall Analysis 42.4.11 Scanning Electron Microscopy 847

43. The Significance of Multivalent Bonding Motifs and “Bond Order” in DNA-Directed Nanoparticle Crystallization Ryan V. Thaner, Ibrahim Eryazici, Robert J. Macfarlane, Keith A. Brown, Byeongdu Lee, SonBinh T. Nguyen, and Chad A. Mirkin 43.1 Introduction 43.2 Methods 43.3 Results and Discussion 44. Oligonucleotide Flexibility Dictates Crystal Quality in DNA-Programmable Nanoparticle Superlattices Andrew J. Senesi, Daniel J. Eichelsdoerfer, Keith A. Brown, Byeongdu Lee, Evelyn Auyeung, Chung Hang J. Choi, Robert J. Macfarlane, Kaylie L. Young, and Chad A. Mirkin 44.1 Introduction 44.2 Methods and Discussion 44.3 Experimental Section 44.3.1 Nanoparticle Crystallization 44.3.2 Transmission Electron Microscopy 45. Entropy-Driven Crystallization Behavior in DNA-Mediated Nanoparticle Assembly Ryan V. Thaner, Youngeun Kim, Ting I. N. G. Li, Robert J. Macfarlane, SonBinh T. Nguyen, Monica Olvera de la Cruz, and Chad A. Mirkin 45.1 Introduction

851

852 854 855 863

864 866 874 874 875 879

880

xi

xii

Contents



45.2 Methods 45.3 Discussion 45.4 Conclusion

882 886 892

46. Electrolyte-Mediated Assembly of Charged Nanoparticles 897 Sumit Kewalramani, Guillermo I. Guerrero-García, Liane M. Moreau, Jos W. Zwanikken, Chad A. Mirkin, Monica Olvera de la Cruz, and Michael J. Bedzyk 46.1 Introduction 898 46.2 Results and Discussion 900 46.2.1 SAXS Studies of DNA-Coated AuNP Assembly 900 46.2.2 MD Simulations for Potential of Mean Force between DNA-Coated AuNPs 904 46.2.3 Liquid-State Theory for Like-Charged Attraction 906 46.3 Conclusions 910

47. The Role of Repulsion in Colloidal Crystal Engineering with DNA 913 Soyoung E. Seo, Tao Li, Andrew J. Senesi, Chad A. Mirkin, and Byeongdu Lee 47.1 Introduction 914 47.2 Results 918 47.2.1 Assembly of Colloidal PAEs through 918 DNA Hybridization Interactions 47.2.2 Assembly of Colloidal PAEs through 920 Depletion Forces 47.3 Discussion 922 47.3.1 The Role of Repulsion in PAE Assembly 922 47.3.1.1 Excluded volume repulsion 923 47.3.1.2 Elastic repulsion 925 47.3.1.3 Repulsion from entropic effects due to counterions 926 47.3.2 Potential Energy Calculation for PAE Superlattices 927 47.4 Conclusion 931

Part 3

Design Rules for Colloidal Crystallization

Chapter 20

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks Formed by DNA Hybridization*

So-Jung Park,a,b Anne A. Lazarides,c James J. Storhoff,a,b Lorenzo Pesce,a,b and Chad A. Mirkina,b aDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cDepartment of Mechanical Engineering and Materials Science, Duke University, Durham, NC 27708, USA [email protected]

So-Jung Park and Anne A. Lazarides contributed equally to this work.

The structural properties of DNA-linked gold nanoparticle materials were examined using synchrotron small-angle X-ray scattering. The materials are composed of 12 or 19 nm diameter gold particles *Reprinted with permission from Park, S.-J., Lazarides, A. A., Storhoff, J. J., Pesce, L. and Mirkin, C. A. (2004). The structural characterization of oligonucleotide-modified gold nanoparticle networks formed by DNA hybridization, J. Phys. Chem. B 108, 12375– 12380. Copyright (2004) American Chemical Society.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

498

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

modified with 3´ or 5´ alkylthiol-capped 12-base oligonucleotides and linked with complementary oligonucleotides. Structure factors were derived from scattering intensities, and nearest-neighbor distances were determined from the primary peak in the pair distance distribution functions. The separation between particles was found to increase linearly with DNA linker length for 24, 48, and 72 base pair linkers. For assemblies formed in 0.3 M NaCl, 10 mM phosphate buffer solution, the increment in the interparticle distance was found to be 2.5 Å per base pair. Particle separations in assemblies at lower electrolyte concentration were larger, indicating that dielectric screening modulates the interactions. The effect of DNA sequence was studied with poly-adenine or polythymine spacer sequences incorporated between the alkylthiol and recognition sequences. The assemblies with poly-adenine spacer sequences showed significantly shorter particle separations than the assemblies involving poly-thymine spacers, a consequence of their different affinities for the gold surface. While the scattering data do not display evidence of long-range order, pair distance distribution functions indicate the presence of short-range order.

20.1  Introduction

Recently, there has been considerable effort aimed at developing strategies for assembling nanoparticles into two- and threedimensional functional structures [1]. Because such structures exhibit collective physical and chemical properties that are dependent upon their architectural parameters [1–8], the ability to deliberately organize nanoparticles into a preconceived structure is essential for achieving maximum control over desired properties. A variety of assembly schemes have been developed based upon DNA, proteins, small organic molecules, and synthetic polymers as particle linkers, and, thus far, insulators, semiconductors, and metallic nanoparticles as well as multivalent proteins have been used as particle building blocks [9–16]. DNA has been demonstrated to be a particularly versatile construction material due to its flexible length scale and rigid and

Introduction

chemically programmable duplex structure [17]. Over the past few years, researchers have demonstrated that one can use DNA to control the assembly of nanoparticles in solution in the form of aggregates (Scheme 20.1A) and small clusters (Scheme 20.1B) and off of surfaces in the form of multilayered structures (Scheme 20.1C) [9, 13, 18–20]. These materials have been used to develop a variety of biomolecule detection schemes based upon their collective optical, catalytic, or electrical properties [21–25]. The most extensively studied and characterized DNA-driven nanoparticle linking system is the three DNA strand system, Scheme 20.1A [9], applied to gold nanoparticles. In this system, two sets of oligonucleotide-modified nanoparticles are connected by complementary linking DNA to form extended structures. Those structures have been characterized by transmission electron microscopy (TEM) [9, 26], dynamic light scattering (DLS) [26], and ultrasmall-angle X-ray scattering experiments (USAXS) [27]. While TEM only provides two-dimensional information on sedimented, desiccated assemblies and DLS and USAXS provide only ensemble-averaged data, collectively they provide both direct imaging and in situ, aqueous phase evidence of extended assembly formation. From the USAXS data, the assemblies are understood to be space-filling as opposed to fractal. Herein, we describe a detailed study of the aqueous phase, three-dimensional structure of a variety of DNA-linked nanoparticle assemblies. To investigate interparticle structure, we use synchrotron small-angle X-ray scattering (SAXS) in a q-range that provides information on structure in the few to tens of nanometer length scale. The following issues have been addressed. First, how do interparticle spacings vary with number of bases in the DNA interconnect and size of the nanoparticle building blocks? Second, how do interparticle interactions and DNAnanoparticle interactions affect the resulting three-dimensional structure? To provide a basis for comparison, we have considered systems (Scheme 20.2) that have been extensively characterized by UV-vis spectroscopy, conductivity measurements, melting analyses, transmission electron microscopy, and light-scattering experiments [26, 28]. Some prior results on the interparticle spacing control provided by duplex DNA have been presented in Refs. [11, 26, 28].

499

500

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

Scheme 20.1

Scheme 20.2

Experimental Section

20.2  Experimental Section 20.2.1 Preparation of Oligonucleotide-Modified Gold Nanoparticles and Linker DNA The synthesis of alkylthiol-capped oligonucleotides (1, 2) and the immobilization of them on gold nanoparticles were described elsewhere [22]. Linker DNA strands (3–5) were synthesized and purified via literature methods [22]. Gold particles of 12 nm diameter were prepared by citrate reduction of HAuCl4 [29]. Gold particles of ~19 nm diameter were purchased from Ted Pella. The oligonucleotide-modified gold nanoparticles were indefinitely stable (over a year) to agglomeration in salt solution (0.3 M NaCl) as evidenced by no change in the UV-vis spectroscopy of the aqueous suspension of particles. The solutions containing oligonucleotidemodified nanoparticles were filtered through an acetate syringe filter (0.22 µm) prior to SAXS experiments. Longer duplex linkers (48-and 72-mer) were prepared by hybridizing 48- and 72-mer single-stranded oligonucleotides with 24- and 48-mer filler DNA, respectively, Scheme 20.2.

20.2.2  Preparation of Nanoparticle Assemblies

To form nanoparticle networks, equal amounts of oligonucleotidemodified nanoparticles, a and b (10 nM, 130 µL) were combined in a microcentrifuge tube, and linker DNA, 3-5 (10 µM, 6 µL), was added to the nanoparticle mixture. Except for the experiments aimed at probing the effect of particle size on aggregate structure, 12 nm particles were used for all other experiments. Unless otherwise mentioned, nanoparticle aggregates were formed in 0.3 M NaCl, 10 mM phosphate buffer at pH 7 (PBS) and were annealed at 45°C for 15 min.

20.2.3  SAXS Measurements

The SAXS experiments were performed at the Dupont-NorthwesternDow Collaborative Access Team (DND-CAT) Sector 5 of the Advanced Photon Source, Argonne National Laboratory with X-rays of wavelength 1.54 Å (8 keV) or 1.24 Å (10 keV). Aqueous samples

501

502

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

were placed in 0.8 or 1.2 mm flat cells between Kapton windows. Two sets of slits were used to define and collimate the X-ray beam, and a pinhole was used to remove parasitic scattering. Samples were irradiated with a 0.3 ¥ 0.3 mm2 beam, and scattered radiation was detected with a CCD area detector. The 2D scattering data were azimuthally averaged, and the resulting 1D profiles of scattered intensity as a function of scattering angle, 2θ, were transformed into profiles of scattered intensity as a function of scattering vector, s (=2 sin(θ)/ λ), using silver behenate [30] as a standard. All data were corrected for background scattering and sample absorption. In this system, scattering from DNA is negligible as compared to that from electron dense gold; scattering from the buffer and windows is weak relative to that from the gold, but the buffer data are nonetheless used to reduce the solution sample data to scattering representative of the particles alone.

20.3  Results and Discussion

20.3.1 Scattering from Dispersed DNA-Modified Nanoparticles SAXS data were collected from mixtures of dispersed DNA-modified nanoparticles (Fig. 20.1). The scattering patterns display fringes characteristic of dispersions of noninteracting particles with modest dispersity in size and shape [31, 32]. Particle sizing was accomplished by fitting the profiles with ensemble-averaged differential scattering cross sections, that is, profiles evaluated using the expression I(s) = IeÚP(ξ)Ne2(ξ)F2(s;ξ)dξ, where Ie is the “constant” differential scattering cross section for a single electron in the forward direction; P is the probability density for the shape and size distribution; Ne (=Vne) is the electron number equal to the particle volume, V, times the excess electron density, ne, of the nanoparticles relative to the solvent; and F 2(s;ξ) is the form factor of a single particle with shape and size indexed by ξ [31]. Initially, the profiles were calculated assuming spherical particles with Gaussian distributions of radii, r, for which the square root, F, of the form factor is proportional to the spherical Bessel function, j1(x) = (sin(x) – x cos(x))/x3, where x = 2πsr. These fits were compared to fits obtained from ensembles

Results and Discussion

of spheroidal particles with distributions in aspect ratio as well as size. Figure 20.1A shows an experimental profile for the 12 nm particles that is composed of short exposure (10 s) data at low angle and longer exposure (240 s) data at high angle [33]. An ensemble-averaged calculated profile is shown as well. The ensemble was modeled as a population of spheroidal particles with Gaussian distributions in size and aspect ratio. Fitting yielded a size distribution of 12.0 ± 1.0 nm and an aspect ratio distribution of 1.0 ± 0.1. Alternatively, an assumption of particle sphericity yielded a size distribution of 11.8 ± 1.2 nm. Figure 20.1B shows the experimental profile for the 19 nm particles along with an ensemble-averaged calculated profile. The ensemble was modeled as a population of spheroidal particles with an aspect ratio of 1.0 ± 0.2 and size of 19.0 ± 1.2 nm. The alternative fit assuming no eccentricity yielded a size distribution of 18.8 ± 1.9 nm.

Figure 20.1  Scattering patterns from dispersed particles (black) and ensembleaveraged calculated profiles (red) for (A) 12.0 ± 1 nm particles and (B) 19 ± 1.2 nm particles.

503

504

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

20.3.2 Nanoparticle Aggregates Formed from DifferentLength DNA Interconnects and Different-Sized Nanoparticles The oligonucleotide-modified nanoparticles (a, b) were assembled into extended networks using three different DNA interconnects, 24mer, 48-mer, and 72-mer strands, Scheme 20.2. Note that the 24-mer linker is a single-stranded oligonucleotide and the 48- and 72-mer linkers are duplex DNA with “sticky” ends that are complementary to the oligonucleotides on the nanoparticles. Upon aggregate formation, the red nanoparticle solution turned purple, and the nanoparticle aggregates precipitated from solution. The smallangle scattering patterns of the DNA-linked nanoparticle assemblies display interference peaks in addition to the fringe patterns of the dispersed particles. Given the monodispersity of the particles, the structure factor, S, of the nanoparticle assemblies can be determined by dividing the scattered intensity, Iagg, by the intensity from the noninteracting particle dispersions, Idisp (Fig. 20.1), after correcting for buffer, window, and parasitic scattering, as described in Eq. 20.1 32, 34].

Iagg (q) µ Idisp (q)S(q)

(20.1)

The diffraction peaks are well defined, and they shift to smaller angles with increasing base pair number in the DNA linkers, indicating larger interparticle spacing (Fig. 20.2). To examine the role of DNA duplex structure in defining assembly structure, nanoparticle aggregates were prepared using singlestranded 72-mer linker without the complementary 48-base “filler” strand. As compared to the aggregates formed from duplex linkers, the aggregates formed using single-stranded DNA linkers show a significantly less well-defined scattering pattern (Fig. 20.3). This result indicates that the rigid nature of DNA duplex is critical for the formation of a well-defined assembly structure in which DNA base pair number controls the interparticle spacing. To obtain interparticle distances within the assemblies, the pair distance distribution functions (PDDF), g(r) [32], were calculated from the structure factors, S(s), using Eq. 20.2, where r is the particle number density (Fig. 20.4).

Ú

g(r ) = 1 + 2rr s( S ( s ) - 1) sin(2p sr ) ds (20.2)

Results and Discussion

Figure 20.2  The structure factors of nanoparticle aggregates formed from 12 nm particles and three different linkers, 24-mer, 48-mer, and 72-mer, in 0.3 M PBS solutions.

Figure 20.3  The structure factors of nanoparticle aggregates formed from 72-mer duplex linkers (solid line) and nanoparticle aggregates formed from single-stranded 72-mer oligonucleotides without filler DNA (dashed line). Both samples were prepared from 12 nm particles in 0.3 M PBS solutions.

The PDDF gives the probability of finding a second particle at a given distance, r, from a given particle; thus, the low-r peak position of each PDDF corresponds to the nearest-neighbor center-tocenter distance, Dnn, for that assembly. When Dnn values for various

505

506

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

assemblies are plotted as a function of linker base pair number, they display a linear relationship (Fig. 20.5).

Figure 20.4  PDDFs of the scattering patterns from nanoparticle aggregates linked by three different-length DNA linkers presented in Fig. 20.2. The primary peak positions of the PDDFs indicate the nearest-neighbor distances.

Figure 20.5  Center-to-center interparticle distances obtained from PDDFs as a function of linker length. Nanoparticle aggregates were formed from two different sized particles, 12 and 19 nm particles in 0.3 M PBS solutions. The particle separation increases linearly with DNA linker length for both cases.

In addition to the 12 nm particles, 19 nm particles were used to form the same types of assemblies and examined by SAXS. Interparticle spacings as given by the PDDFs for the larger particle

Results and Discussion

assemblies also display a linear relationship with base pair number, with offsets of ~7 nm from the spacings found in assemblies of 12 nm particles (Fig. 20.5). Note that the offset of 7 nm matches the particle size difference. From the slopes of the plots, the increase of Dnn per one base pair is found to be 0.25 nm, which is shorter than the length per base pair of B-form DNA, 0.34 nm [35]. Possible explanations are that (1) interparticle spaces are determined by multiple DNA duplex linkers that connect regions of the particle surface offset from the point of closest approach, or (2) particle-linked DNA has a different conformation than solution-phase B form DNA. The latter possibility is not unreasonable given the high dielectric environment and the possibly condensed nature of the oligonucleotide environment. Nonetheless, the results demonstrate that particle spacings in DNAlinked nanoparticle assemblies can be tailored by using different DNA duplex interconnects. For comparison, 24-mer-linked nanoparticle aggregates were formed by a two-strand system in addition to the three-strand system described thus far. In the two-strand system, the aggregates were formed by mixing two sets of nanoparticles modified with complementary 24-mer oligonucleotides. The assemblies formed by the two different methods but with the same linker length showed scattering peaks at approximately the same positions. The effects of particle and linker concentrations and of annealing also were investigated; no significant effect on the interparticle distances was observed.

20.3.3 Particle Packing within DNA-Linked Nanoparticle Assemblies

Typically, monodisperse hard spheres crystallize into closely packed structures. Metal particles, however, interact through long-range attractive forces as well as hard repulsive forces, and charge or sterically stabilized particles interact through a complex interplay of distance-dependent attractive and repulsive interactions. Thus, depending on the types of particles and ligand molecules and the preparation conditions, chemically functionalized particles assemble into a variety of morphologies and may display a rich phase behavior [36, 37]. Oligonucleotide-modified nanoparticles provide a unique

507

508

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

system to study superstructure formation in the presence of both repulsive and attractive interactions [37]. DNA-linked nanoparticle assemblies have scattering patterns analogous to those of amorphous materials, that is, materials that display short-range, though not long-range order. A comparison of nearest-neighbor distances, revealed by the PDDFs, and the d spacings determined by peaks in the structure factors confirms the presence of short-range order. Specifically, for all linked assemblies, the scattering patterns display (1) an absence of a peak at the angle corresponding to the d spacing, Dnn, for the nearest-neighbor distance as determined by the PDDF and (2) a maximally intense, lowest angle peak at an angle corresponding to a d spacing of (√2/√3) Dnn. This behavior is characteristic of systems whose “atomic” bases null intensity in the scattering direction corresponding to nearest-neighbor planes such as body-centered cubic (BCC), facecentered cubic (FCC), or closely related body-centered tetragonal (BCT) structures. While these structural properties are shared by all linked assemblies considered here, that is, assemblies with particle size to linker lengths in the range of 0.4–2.0 particle radii, it is by no means universal among DNA-linked particle assemblies. Rather, other linking systems have been observed to result in alternative structures. For example, assemblies composed of structurally incommensurate component “particles,” such as the protein/nanoparticle system described in Ref. [11], have scattering patterns that are quantitatively different. Nonetheless, the extensive interference displayed in the scattering data for all of the DNA-linked nanoparticle systems that we have studied indicates local order and an absence of the fractal structure observed in aggregates of bare particles [38].

20.3.4  The Effect of Ionic Strength

In the absence of complementary linking DNA, the oligonucleotidemodified nanoparticles are uniformly dispersed in a buffer solution due to the repulsive interaction between like-charged nanoparticles. In DNA-linked assemblies, electrostatic forces are nonetheless expected to play a significant role in mediating the attractive interactions between complementary DNA strands and may also affect the assembly superstructure [39]. To examine this

Results and Discussion

effect, nanoparticle aggregates were formed in a series of buffer solutions with different ionic strengths. As the salt concentration of the buffer is lowered, the diffraction peaks shift to lower angles, indicating larger interparticle spacings (Fig. 20.6). When NaCl concentration is varied over the range of 0.05–0.7 M, the spacing between 24-mer-linked particles changes by ~1.7 nm (Fig. 20.6, inset). At high salt concentration, cations screen the negative charge of the oligonucleotides on the nanoparticles, allowing particles to assemble more closely, while, at low salt concentration, particle–particle repulsion becomes stronger and results in larger interparticle spacings.

Figure 20.6  The structural factors of 24-mer-linked 12 nm particle aggregates formed in buffer solutions of different salt concentration. Inset: interparticle distances as a function of NaCl concentration.

20.3.5  The Effect of Single-Strand Spacer DNA In the design of oligonucleotide sequences, we have employed spacer sequences in addition to the recognition strands (Fig. 20.7A). Interestingly, the type and length of spacer DNA affect the properties of the oligonucleotide-modified nanoparticles, such as their hybridization efficiencies and the thermal denaturation temperatures of the resulting duplex structures [40]. From a study of nucleoside binding on gold nanoparticles and on planar surfaces [41, 42], we have postulated that this effect is due to the different

509

510

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

affinity of the four DNA bases to the gold surfaces. We expect that differences in binding affinities of different oligonucleotide sequences should manifest themselves both as differences in spacer configuration and also as differences in aggregate superstructure. Here, we seek to demonstrate that the information of interest can be derived from scattering studies of DNA-linked nanoparticles.

Figure 20.7  (A) A drawing describing the spacer DNA. (B) The structure factors of 24-mer-linked aggregates formed from 12 nm particles with A10 (sequence: AAA AAA AAA A), T10 (sequence: TTT TTT TTT T), or no spacer sequence. (C) The interparticle distances as a function of spacer DNA length for poly-A and poly-T spacers. The two systems show distinct signatures due to their different binding strength to gold surface. For all samples, aggregates were prepared in 0.3 M PBS solutions.

For this purpose, nanoparticles were modified with oligonucleotides with poly-adenine or poly-thymine spacers incorporated between the surface-binding alkylthiol and the recognition sequence (Fig. 20.7A). Subsequently, the nanoparticles were assembled using 24-mer linkers. Interestingly, the assemblies involving A10 and T10 spacer units showed scattering peaks at quite different angles, despite their identical base number (Fig. 20.7B). When the particles with an A10 spacer were linked by 24-mer DNA, interparticle spacing increased by only about 1.6 nm, as compared to the spacer-free system, while the insertion of a T10 spacer increased the interparticle spacing by approximately 4.3 nm. These

Conclusions

results clearly indicate that ssDNA composed of poly-adenine has a high affinity for gold which is manifested by a tendency to lie on the surface, while poly-thymine has a low affinity and tends to stand upright and away from the surface [43]. This is consistent with previous spectroscopic studies aimed at determining such interactions [41, 42, 44]. between single-stranded Because the interactions oligonucleotides and a gold surface are complex and possibly surface area limited, the number of bases in the single-stranded spacer DNA and the interparticle distances do not display a simple linear relationship, such as that seen when varying base pair number in the duplex portion of the linker, Fig. 20.7C. In addition, assemblies involving different types of spacer sequences show distinct signatures due to the difference in their binding strength. In the case of poly-A, a major portion of the A10 spacer is coordinated to the particle surface due to its high affinity to gold. As the number of spacer bases increases, a portion will bind on the surface and the rest will stand out, as evidenced by the significant increase in interparticle spacing with longer poly-A spacer, A20, as compared to the system with an A10 spacer. In the case of poly-T, because it has a low affinity for gold, the interparticle spacing gradually increases with increasing spacer length.

20.4  Conclusions

We have studied a number of factors that control the structural parameters of DNA-linked nanoparticle assemblies. When oligonucleotide-modified nanoparticles are connected by duplex DNA linkers, the assemblies exhibit relatively sharp scattering peaks. While the assemblies are not crystalline, the scattering patterns and derived PDDFs together reveal that the assemblies have FCC-, BCC-, or BCT-like local structure. Importantly, the assemblies formed from different-length oligonucleotide linkers (24-, 48-, and 72-mer) show an increase in particle spacing as the base pair number of the linker DNA is increased. This result demonstrates that we can indeed take advantage of the flexible length scale and the rigid duplex structure of DNA to tailor interparticle spacings using different-length DNA interconnects.

511

512

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

Small-angle X-ray scattering has also been used to probe the effects of surface interaction on the assembly structure. Oligonucleotides composed of poly-A have high affinities for gold, and incorporation of a poly-A spacer unit increases the interparticle spacing by only a small amount. In contrast, poly-T has a relatively low affinity for gold, and a poly-T spacer significantly increases the interparticle spacing. This study demonstrates the importance of DNA/nanoparticle interactions in DNA-based assembly schemes. Even if the nanoparticles are modified with oligonucleotides of equal base number, assemblies with different structures can result depending on the choice of spacer DNA sequences. Because many interesting properties of nanoparticle-based materials are highly dependent upon their structural parameters, it is important to understand how assembly structure depends on the various building blocks, interconnects, and assembly conditions. The study described herein was designed to explore these relationships in DNA-linked assemblies and thereby provide insight into the structural basis for some of the important physical properties associated with these assemblies. Ultimately, an understanding of the mechanisms of structural control in hybrid systems of this type will enable the design of the materials with targeted properties.

Acknowledgments

C.A.M. acknowledges DARPA, AFOSR, and NSF. A.A.L. acknowledges the support of G. C. Schatz and the ARO. The SAXS experiments were performed with the support of the Dupont Northwestern Dow Collaborative Access Team (DND-CAT) at Sector 5 of the Advanced Photon Source. Preliminary experiments were performed with the support of the Biological SAXS/Diffraction staff of BL 4-2 at the Stanford Synchrotron Radiation Laboratory.

References

1. Storhoff, J. J., Mucic, R. C. and Mirkin, C. A. (1997). J. Cluster Sci., 8, 179. 2. Fendler, J. H. and Meldrum, F. C. (1995). Adv. Mater., 7, 607.

3. Bethell, D., Brust, M., Schiffrin, D. J. and Kiely, C. (1996). J. Electroanal. Chem., 409, 137.

References

4. Bethell, D. and Schiffrin, D. J. (1996). Nature, 382, 581.

5. Weller, H. (1996). Angew. Chem. Int. Ed. Engl., 35, 1079.

6. Alivisatos, A. P. (1996). Science, 271, 933. 7. Heath, J. R. (1995). Science, 270, 1315.

8. Schmid, G. and Chi, L. F. (1998). Adv. Mater., 10, 515.

9. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607. 10. Mitchell, G. P., Mirkin, C. A. and Letsinger, R. L. (1999). J. Am. Chem. Soc., 121, 8122.

11. Park, S.-J., Lazarides, A. A., Mirkin, C. A. and Letsinger, R. L. (2001). Angew. Chem. Int. Ed., 40, 2909.

12. Cao, Y. W., Jin, R. and Mirkin, C. A. (2001). J. Am. Chem. Soc., 123, 7961.

13. Alivisatos, A. P., Johnsson, K. P., Peng, X., Wilson, T. E., Loweth, C. J., Bruchez, M. P., Jr. and Schultz, P. G. (1996). Nature, 382, 609.

14. Shenton, W., Davis, S. A. and Mann, S. (1999). Adv. Mater., 11, 449.

15. Connolly, S. and Fitzmaurice, D. (1999). Adv. Mater., 11, 1202.

16. Brust, M., Bethell, D., Kiely, C. J. and Schiffrin, D. J. (1998). Langmuir, 14, 5425. 17. Storhoff, J. J. and Mirkin, C. A. (1999). Chem. Rev., 99, 1849.

18. Taton, T. A., Mucic, R. C., Mirkin, C. A. and Letsinger, R. L. (2000). J. Am. Chem. Soc., 112, 6305.

19. Niemeyer, C. M., Burger, W. and Peplies, J. (1998). Angew. Chem. Int. Ed., 37, 2265.

20. Soto, C. M., Srinivasan, A. and Ratna, B. R. (2002). J. Am. Chem. Soc., 124, 8508. 21. Elghanian, R., Storhoff, J. J., Mucic, R. C., Letsinger, R. L. and Mirkin, C. A. (1997). Science, 277, 1078.

22. Storhoff, J. J., Elghanian, R., Mucic, R. C., Mirkin, C. A. and Letsinger, R. L. (1998). J. Am. Chem. Soc., 120, 1959. 23. Park, S.-J., Taton, T. A. and Mirkin, C. A. (2002). Science, 295, 1503.

24. Taton, T. A., Mirkin, C. A. and Letsinger, R. L. (2000). Science, 289, 1757.

25. Cao, Y. W. C., Jin, R. C. and Mirkin, C. A. (2002). Science, 297, 1536.

26. Storhoff, J. J., Lazarides, A. A., Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Schatz, G. C. (2000). J. Am. Chem. Soc., 122, 4640.

27. Lazarides, A. A. (unpublished results). USAXS experiments on 24 base pair-linked 12 nm Au particle assemblies not only indicated the

513

514

The Structural Characterization of Oligonucleotide-Modified Gold Nanoparticle Networks

presence of large assemblies, but also yielded a fractal dimension for these assemblies of 2.8 ( 0.2. From this we conclude that the assemblies are essentially space-filling, that is, not fractal.

28. Park, S.-J., Lazarides, A. A., Mirkin, C. A., Brazis, P. W., Kannewurf, C. R. and Letsinger, R. L. (2000). Angew. Chem. Int. Ed., 39, 3845.

29. Frens, G. (1973). Nat. Phys. Sci., 241, 20.

30. Huang, T. C., Toraya, H., Blanton, T. N. and Wu, Y. (1993). J. Appl. Crystallogr., 26, 180.

31. Glatter, O. and Kratky, O. (1982). Small-Angle X-ray Scattering (Academic, New York).

32. Feigin, L. A. and Svergun, D. I. (1997). Structure Analysis by Small-Angle X-ray and Neutron Scattering (Plenum Press, New York).

33. Composition was done using 1/2(1 + sin(x)) as a switch function. The relative normalization of the 10 and 240 s profiles was done to have the two profiles overlap for the largest possible extent of scattering vector values. 34. Equation 1 is exact only in the case of monodisperse systems. For systems with dispersity less than 10%, it provides a useful approximation.

35. Yanagi, K., Prive, G. G. and Dickerson, R. E. (1991). J. Mol. Biol., 217, 201. 36. Korgel, B. A. and Fitzmaurice, D. (1999). Phys. Rev. B, 59, 14191. 37. Tkachenko, A. V. (2001). Condens. Matter, 1.

38. Dimon, P., Sinha, S. K., Weitz, D. A., Safinya, C. R., Smith, G. S., Varady, W. A., Lindsay, H. M. (1986). Phys. Rev. Lett., 57, 595. 39. Cobbe, S., Connolly, S., Ryan, D., Nagle, L., Eritja, R. and Fitzmaurice, D. (2003). J. Phys. Chem. B, 107, 470.

40. Demers, L. M., Mirkin, C. A., Mucic, R. C., Reynolds, R. A., Letsinger, R. L., Elghanian, R. and Viswanadham, G. (2000). Anal. Chem., 72, 5535.

41. Storhoff, J. J., Elghanian, R., Mirkin, C. A. and Letsinger, R. L. (2002). Langmuir, 18, 6666.

42. Demers, L. M., Östblom, M., Jang, N.-H., Liedberg, B. and Mirkin, C. A. (2002). J. Am. Chem. Soc., 124, 11248. 43. For presumably the same reason, particles with poly-adenine spacer maintain their recognition properties for a long period of time (over a year), but particles with poly-thymine spacer tend to lose their recognition properties over time.

44. Kimura-Suda, H., Petrovykh, D. Y., Tarlov, M. J. and Whitman, L. J. (2003). J. Am. Chem. Soc., 125, 9014.

Chapter 21

DNA-Programmable Nanoparticle Crystallization*

Sung Yong Park,a,b,c Abigail K. R. Lytton-Jean,a,b Byeongdu Lee,d Steven Weigand,e George C. Schatz,a,b and Chad A. Mirkina,b aDepartment

of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cDepartment of Biostatistics and Computational Biology, University of Rochester, 601 Elmwood Avenue, Rochester, NY 14642, USA dX-ray Science Division, Advanced Photon Source, Argonne National Laboratory, 9700 S. Cass Ave., Argonne, IL 60439, USA eDND-CAT Synchrotron Research Center, Northwestern University, APS/ANL 432-A004, 9700 S. Cass Avenue, Argonne, IL 60439, USA [email protected] Sung Yong Park and Abigail K. R. Lytton-Jean contributed equally to this work.

It was first shown [1, 2] more than ten years ago that DNA oligonucleotides can be attached to gold nanoparticles rationally to direct the formation of larger assemblies. Since then, oligonucleotide-functionalized nanoparticles have been *Reproduced with permission from Park, S. Y., Lytton-Jean, A. K. R., Lee, B., Weigand, S., Schatz, G. C. and Mirkin, C. A. (2008). DNA-programmable nanoparticle crystallization, Nature 451, 553–556.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

516

DNA-Programmable Nanoparticle Crystallization

developed into powerful diagnostic tools [3, 4] for nucleic acids and proteins, and into intracellular probes [5] and gene regulators [6]. In contrast, the conceptually simple yet powerful idea that functionalized nanoparticles might serve as basic building blocks that can be rationally assembled through programmable basepairing interactions into highly ordered macroscopic materials remains poorly developed. So far, the approach has mainly resulted in polymerization, with modest control over the placement of, the periodicity in, and the distance between particles within the assembled material. That is, most of the materials obtained thus far are best classified as amorphous polymers [7–16], although a few examples of colloidal crystal formation exist [8, 16]. Here, we demonstrate that DNA can be used to control the crystallization of nanoparticle–oligonucleotide conjugates to the extent that different DNA sequences guide the assembly of the same type of inorganic nanoparticle into different crystalline states. We show that the choice of DNA sequences attached to the nanoparticle building blocks, the DNA linking molecules and the absence or presence of a non-bonding single-base flexor can be adjusted so that gold nanoparticles assemble into micrometre-sized face-centredcubic or body-centred-cubic crystal structures. Our findings thus clearly demonstrate that synthetically programmable colloidal crystallization is possible, and that a single-component system can be directed to form different structures. From a surface receptor standpoint, gold nanoparticles can be programmed to behave as a single-component or binary system by using the sequence-specific recognition properties of DNA (Fig. 21.1a) and designing DNA linkers with two different regions (Fig. 21.1b and c). In a typical experiment, gold nanoparticles (15 nm in diameter) are modified with synthetic oligonucleotides [17] and then linker DNA is introduced; the latter contains a region 1 complementary to the gold-nanoparticle-bound DNA, and a region 2 that acts as a dangling end and can be varied to control the interactions between the gold nanoparticles. In all cases, region 1 is significantly longer than region 2, and therefore the duplex formed from hybridization with region 1 is more stable than the duplex formed from hybridization with region 2. This allows region 2 to be thermally addressable without significantly perturbing region 1 (Ref. [18]). By designing a linker sequence in which region 2 is self-complementary, the nanoparticles will effectively behave as a

DNA-Programmable Nanoparticle Crystallization

single-component system (Fig. 21.1b). Alternatively, by designing a linker with a non-self-complementary region 2, an additional, different linker is required to achieve particle assembly (Fig. 21.1c). From a surface receptor standpoint, the latter design creates a binary system in which gold nanoparticles hybridized to linker-X (AuNP-X) can only bind to gold nanoparticles hybridized to linker-Y (AuNP-Y). Between region 1 and 2, a non-binding single DNA base, called a flexor, is added (typically adenosine, A). As discussed later, the flexor plays a crucial role in DNA-programmable nanoparticle crystallization. a

f.c.c.

Linker A Linker X b.c.c.

Linker Y

b 5¢



5¢ S-A10 -AAGACGAATATTTAACAA CGCG-A-TTGTTAAATATTCGTCTT 3¢ 3¢ TTCTGCTTATAAA TTGTT-A-GCGC 5¢ 3¢AACAATTTATAAGCAGAA-A -S 5¢ 10 Linker A Linker A T > Tm T < Tm



S-A10-AAGACGAATATTTAACAA CGCG-A-TTGTTAAATATTCGTCTT 3¢ 3¢ TTCTGCTTATAAATTGTT-A-GCGC AACAATTTATAAGCAGAA-A 10-S 5¢ Region 1 (18-mer)

Region 1 (18-mer)

Region 2

c 5¢ S-A10 -AAGACGAATATTTAACAA 3¢ 3¢′



TTCCTTT-X-TTGTTAAATATTCGTCTT 3¢ 3¢

TTCTGCTTATAAATTGTT-X-AAGGAAA′ 5¢ Linker X

T > Tm 5¢

S-A10-AAGACGAATATTTAACAA

AACAATTTATAAGCAGAA-A 10 -S Linker Y

5

T < Tm TTCCTTT-X-TTGTTAAATATTCGTCTT 3¢ 5¢ AACAATTTATAAGCAGAA-A 10 -S

3¢ TTCTGCTTATAAATTGTT-X-AAGGAAA

Region 1 (18-mer)

Region 2

Region 1 (18-mer)

Figure 21.1  Scheme of gold nanoparticle assembly method. (a) Gold nanoparticle–DNA conjugates can be programmed to assemble into different crystallographic arrangements by changing the sequence of the DNA linkers. (b) Single-component assembly system (f.c.c.) where gold nanoparticles are assembled using one DNA sequence, linker-A. (c) Binary-component assembly system (b.c.c.) in which gold nanoparticles are assembled using two different DNA linkers -X and -Y. X in the DNA sequence denotes the flexor region: A, PEG6 or no base. NP1 indicates that the same gold nanoparticle-DNA conjugates were used in all experiments.

517

518

DNA-Programmable Nanoparticle Crystallization

The ability to simulate a single-component or binary system without the irreversible chemical alteration of the gold nanoparticle– oligonucleotide conjugate is a unique aspect of this system. From an energy minimization standpoint, it is expected that the gold nanoparticle assemblies will maximize the number of hybridized DNA linkages by adopting a conformation that will maximize the number of nanoparticle nearest neighbors. In a single-component system, in which each particle can bind to every other particle with equal affinity, a close-packed face-centred-cubic (f.c.c.) structure is expected to form wherein each particle has 12 nearest neighbors. Alternatively, in a binary system, where AuNP-X can bind only to AuNP-Y, the maximum number of hybridization events is achieved in a non-close-packed body-centred-cubic (b.c.c.) structure wherein each particle has eight nearest neighbors (that is, a caesium chloride lattice). Should a binary system assemble into a closepacked structure, each particle will have, on average, less than eight compatible nearest neighbors through which DNA hybridization can occur. We begin by demonstrating the ability to form close-packed macroscopic single-crystalline domains using the single-component nanoparticle system (Fig. 21.1b). To create a well-defined and closepacked crystal, weak and reversible interactions are necessary [13, 19–23]. This is achieved by combining the gold nanoparticles and linker-A above the melting temperature (Tm) of region 2 (~44°C) followed by slow cooling (10 min/1°C) to room temperature, to ensure that crystal formation is thermodynamically and not kinetically controlled. Two-dimensional small-angle X-ray scattering (SAXS) data collected from the resultant particle assemblies display a scattering pattern specific to a f.c.c. structure (Fig. 21.2a). In addition to well-defined scattering rings, individual scattering spots are clearly seen in the first and second ring indicating the formation of many large crystallites (Fig. 21.2b). The majority of the high-intensity spots reside in the first ring and display nearly identical q values. The two-dimensional data were integrated and normalized based on the q value from the first ring (red line in Fig. 21.2c). The normalized spot positions were located at q/q0 ≈ 1, √4/3, √8/3, √11/3 and 2, identifying the crystalline the domains as possessing f.c.c. structure when

DNA-Programmable Nanoparticle Crystallization

compared to the theoretical spectrum (green line in Fig. 21.2c). Indeed, the averaged structure factor S(q) (red line in Fig. 21.2c) is very similar to a theoretical simulation (blue line in Fig. 21.2c) of the SAXS pattern for a f.c.c. configuration containing a small amount of disorder [24]. We note that the experimental data exhibit a small feature at q/q0 = √3, which could be due to a hexagonal closepacked crystalline domain, a compressed f.c.c. crystalline domain, or a less ordered random hexagonal close-packed domain. However, the dominant overall structure is clearly f.c.c. We use the Scherrer formula [25] to estimate from the spots in the scattering pattern a size of around 2.1 µm (~106 particles) for the single-crystal domains. The average interparticle distance can also be determined from the position of the first peak, which in q space is placed at √6π/ dAu, where dAu is the distance between the nanoparticle centres. This gives a measured interparticle distance of 27.9 nm, which falls within the range of the predicted interparticle distance (28.6–36.1 nm) that is based on the length of the DNA linkers. 6

b

a

qz (×10–2 Å–1)

4 2 0 4 c

–2

111

S(q)

200

–4 –6 –6

2

Experiment Theory: perfect f.c.c. Disordered f.c.c. 311 220

0 –4

–2

0

qy (×10–2 Å–1)

2

4

6

1

23 q/q0

Figure 21.2  f.c.c. gold nanoparticle SAXS pattern. (a) SAXS pattern of micrometre-size single-crystalline domains using a single-component system. The colour scale indicates the intensity, I. The image is in log scale. (b) A partial magnification of a displays individual spots of increased scattering intensity (red arrows). (c) The integrated data from a shows an f.c.c. crystal structure. The x-axis is normalized to the first peak from a (2.76 ¥ 10-2 Å-1). The entire spectrum from c is not shown in a.

519

520

DNA-Programmable Nanoparticle Crystallization

More interesting than the formation of a close-packed structure is the ability to program the assembly of the same nanoparticles into a non-close-packed structure. This was achieved by using linkers -X and -Y, to create a binary system (from the surface receptor standpoint) which drives the gold nanoparticle assembly into a non-close-packed b.c.c. structure to maximize the number of DNA hybridization events (Fig. 21.1c). The two-dimensional SAXS pattern for this system clearly indicates a b.c.c. structure (Fig. 21.3a and b). The averaged S(q), determined by integration and normalization based on the q value from the first ring, shows five peak positions at q/q0 = 1, √2, √3, 2 and √5, in agreement with the theoretical b.c.c. structure (green line in Fig. 21.3b). In addition to the peak position, the relative peak heights are consistent with the theoretical calculations. Using the Scherrer formula, the average size of a single-crystalline domain was estimated to be about 600 nm (~104 particles) with an average d-spacing of 31 nm (estimated range 29.6–37.1 nm). The binary system discussed above can also form a close-packed structure, by carefully controlling the temperature at which AuNP-X and AuNP-Y are combined. If the binary particles are treated in the same manner as the single-component system by combining AuNP-X and AuNP-Y above the Tm of region 2 (weak DNA attractive forces) followed by slow cooling, a substitutionally disordered f.c.c. structure [21] is formed, which presents an f.c.c. scattering pattern. Alternatively, a non-close-packed b.c.c. structure is achieved by combining AuNP-X and AuNP-Y at room temperature, below the Tm (~37°C) of region 2 (stronger DNA attractive forces) followed by annealing a few degrees below the Tm. The formation of the different crystal structures is attributed to a competition between the entropic and enthalpic contributions involved in the assembly process at different temperatures. From an entropic standpoint, a close-packed structure is favoured over a non-close-packed structure because the entropy of the entire system can be maximized if the aggregates possess the smallest possible volume fraction [22, 26]. Therefore, if gold nanoparticles begin to assemble near the DNA Tm, where the DNA binding strength is very weak and the enthalpic contribution is small, the entropic contribution will dominate the assembly process and a close-packed structure forms. However, if the gold nanoparticles are combined

DNA-Programmable Nanoparticle Crystallization

several degrees below the Tm, the enthalpic contribution associated with DNA hybridization will govern the assembly process and a nonclose-packed structure forms that maximizes the number of DNA hybridization events. 6

4

a

b

3

0

2

–2

1

–4

0

–6 –6

–4

–2

0

2

4

211 200

c

3

S(q) (arbitrary units)

1.0

13 min, T = 35.4, with PEG6 13 min, T = 37.0, with A 25 min, T = 33.4, no A

1.5

0.0

1.0

2.0 q/q 0

220

310

23

1

6

q y (×10–2 Å–1)

I(q)/Imax

Experiment Theory: perfect b.c.c.

110

2 S(q)

q z (×10–2 Å–1)

4

q/q 0 23 min, T = 38.3, with PEG6 33 min, T = 40.6, with A 32 min, T = 40.0, no A

d

2

1

0

1

q/q 0

2

Figure 21.3  b.c.c. gold nanoparticle SAXS pattern. (a) SAXS pattern of the binary gold nanoparticle system combined below the Tm of region 2. The colour scale indicates the intensity, I. The image is in log scale. (b) The integrated SAXS data from a shows a b.c.c. crystal structure. (c) Comparison of the first peak between three different binary samples containing different flexor regions as assembly is initiated. (d) Comparison of the entire SAXS pattern of three different binary samples containing different flexors (A, PEG6, no flexor). The times given indicate the time passed since the initiation of the experiment.

Gold nanoparticle–oligonucleotide conjugate systems have many variables that can be adjusted to affect the final structure. In addition to the DNA sequence as addressed above, DNA rigidity, DNA length and particle size can be manipulated to influence gold nanoparticle crystallization without changing the basic properties of the overall system. To probe the importance of these variables in the crystallization process, we began by changing the rigidity of the

521

522

DNA-Programmable Nanoparticle Crystallization

flexor region which is not involved in the DNA hybridization. Two variations of linker-X and -Y DNA were synthesized; one without the A-flexor and one with a polyethylene glycol oligomer (PEG6) in place of the A-flexor (Fig. 21.1c). The absence of the A-flexor should result in a rigid system while the PEG6-flexor should give a more flexible system. The first peak in the SAXS pattern from the sample with the PEG6-flexor is the sharpest, and the sample with no flexor is the broadest, Fig. 21.3c. This indicates that the sample with the flexible PEG6-flexor can grow larger crystals. Also, after similar crystallization times, a more well-defined crystalline structure arises from the PEG6-flexor (Fig. 21.3d). Hence, greater flexibility can enhance the assembly process and results in a more well-defined crystalline structure. Next, the effect of DNA length was interrogated by designing a gold nanoparticle with a shorter DNA sequence in region 1 (12mer versus 18-mer in Fig. 21.4a) while maintaining an elevated Tm compared to region 2. When both AuNP-X and AuNP-Y contain the shorter region 1 DNA sequence, the result is a b.c.c. structure, similar to before, only with a shorter interparticle distance. However, the combination of a short region 1 AuNP-X and a long region 1 AuNP-Y (Fig. 21.4b), results in a b.c.c. structure that is thermally more stable. This suggests that the aspect ratio between the effective radii of the binary DNA-linked gold nanoparticles is an important factor in the crystallization process. As the aspect ratio of the particles decreases from one, a b.c.c. structure becomes entropically more favoured because the volume fraction of the b.c.c. structure is reduced [22, 26] and the f.c.c. structure loses its entropic advantage as the particles become effectively polydisperse [27]. This was further addressed by using small gold nanoparticles (10 nm) with a short region 1 and large gold nanoparticles (15 nm) with a long region 1 (Fig. 21.4c). These samples formed crystals with a b.c.c. structure that exhibit even greater stability, such that the b.c.c. structure can be achieved independent of pathway (that is, slow cooling from above Tm versus combining and annealing below Tm). In all cases, to create well-defined programmable crystalline structures using DNA-linked gold nanoparticles, several conditions must be met. In addition to having control over the strength of the DNA attractive forces, it is important to have highly monodisperse particles ( koff. The weak, polyvalent nature of the DNA-AuNP hybridization scheme is critical for this reorganization process, as the weak binding of individual linker recognition units allows for high values of koff, while the high local concentration of DNA on the surface of the nanoparticles inflates the values of kon. The combined effect of these high kon and koff rates results in “mobile” nanoparticles within an aggregate, where repositioning of the DNA-AuNPs is possible. To determine the relative kon and koff values for these nanoparticle systems, an effective concentration (Ceff) for the DNA linker recognition units within an aggregate was determined. Ceff is defined as the number of linker recognition units per particle divided by the limited volume in which they exist, due to localized confinement of DNAs tethered to the surface of a AuNP. Values of 1/Tm were plotted against values of ln(Ceff), where Tm is the temperature at which an aggregate dissociates. These values exist in a linear relationship (Fig. 22.4c) and can be used to determine the thermodynamic constants (∆H° and ∆S°) associated with DNA duplex formation. Based on the data calculated in Fig. 22.4c, the ∆H° value for DNA hybridization is 141.9 kJ mol–1, which is within 6.7% variance from previously established literature values for this 5¢-CGCG-3¢ sequence [19], indicating that this is indeed an accurate model. Using these data, plots of calculated kon and koff values show reasonable agreement with experimental Tm values for the systems studied. When the kon of a system is less than 6 × 103 s-1, no crystals are observed. This indicates that, at the temperatures immediately below dehybridization of the linker–linker overlap for these systems, there is not enough thermal energy to induce restructuring

Establishing the Design Rules for DNA-Mediated Programmable Colloidal

of the nanoparticles on an appreciable timescale. However, when the kon of a system is greater than 1 × 104 s-1, the rates of DNA deand re-hybridization are fast enough to induce restructuring in the aggregate at temperatures slightly below Tm. (Between these values, some systems are able to restructure, while others are not.) These kinetic data explain the results at the bottom right of Fig. 22.3, where systems with large DNA length to AuNP diameter ratios are unable to transition from disordered aggregates to ordered crystals—it is in these systems that the lowest rates of kon are observed. In conclusion, we have determined that there is a definable relationship between DNA length and particle size in the DNAdirected assembly of nanoparticles. This discovery enables not only a better understanding of the fundamental forces driving the crystallization process, but also the formation of colloidal crystals with tailorable features, including interparticle distance, particle size, degree of filled space, and unit cell lattice parameters. The method we have developed to model DNA strands con-strained on surfaces of NPs and predict their assembly behavior provides a basis for future exploration into the interactions of DNA and nanoscale objects, and we project that these methods can be extended to studies of nanoparticles with different shapes and compositions. These discoveries will serve as a template for future nanoparticle crystallization efforts, allowing for the formation of crystals with controllable and tunable physical properties.

Acknowledgments

We acknowledge George Schatz for helpful discussions regarding the theoretical calculations of DNA flexibility and relative DNA concentrations in aggregates. C.A.M. acknowledges the NSF-NSEC and the AFOSR for grant support. He also is grateful for a NIH Director’s Pioneer Award and an NSSEF Fellowship from the DoD. Portions of this work were supported as part of the Non-Equilibrium Energy Research Center (NERC), an Energy Fontier Research Center funded by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences under Award Number DE-SC0000989. R.J.M. acknowledges Northwestern University for a Ryan Fellow-ship.

535

536

Establishing the Design Rules for DNA-Mediated Programmable Colloidal

M.R.J. acknowledges Northwestern University for a Ryan Fellowship and the NSF for a Graduate Research Fellowship. K.L.Y. acknowledges the NSF and the NDSEG for Graduate Research Fellowships. Portions of this work were performed at the DuPont-Northwestern-Dow Collaborative Access Team (DND-CAT) located at Sector 5 of the Advanced Photon Source (APS). DND-CAT is supported by E.I. DuPont de Nemours & Co., The Dow Chemical Company, and the State of Illinois. Use of the APS was supported by U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357.

References

1. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607. 2. Alivisatos, A. P., Johnsson, K. P., Peng, X. G., Wilson, T. E., Loweth, C. J., Bruchez, M. P. and Schultz, P. G. (1996). Nature, 382, 609. 3. Park, S. Y., Lytton-Jean, A. K. R., Lee, B., Weigand, S., Schatz, G. C. and Mirkin, C. A. (2008). Nature, 451, 553.

4. Hill, H. D., Macfarlane, R. J., Senesi, A. J., Lee, B., Park, S. Y. and Mirkin, C. A. (2008). Nano Lett., 8, 2341. 5. Macfarlane, R. J., Lee, B., Hill, Senesi, H. D. A. J., Seifert, S. and Mirkin, C. A. (2009). Proc. Natl. Acad. Sci. U.S.A., 106, 10493.

6. Storhoff, J. J., Lazarides, A. A., Mucic, R. C., Mirkin, C. A., Letsinger, R. L. and Schatz, G. C. (2000). J. Am. Chem. Soc., 122, 4640.

7. Park, S. Y., Lee, J.-S., Georganopoulou, D., Mirkin, C. A. and Schatz, G. C. (2006). J. Phys. Chem. B, 110, 12673.

8. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451, 549.

9. Xiong, H., van der Lelie, D. and Gang, O. (2009). Phys. Rev. Lett., 102, 015504. 10. Cheng, W., Hartman, M. R., Smilgies, D.-M., Long, R., Campolongo, M. J., Li, R., Sekar, K., Hui, C.-Y. and Luo, D. (2009). Angew. Chem. Int. Ed. Angew. Chem., 121, 6587; Angew. Chem. Int. Ed. Engl., 48, 6465.

11. Alivisatos, A. P. (1996). Science, 271, 933.

12. Nikoobakht, B. and El-Sayed, M. A. (2003). Chem. Mater., 15, 1957.

References

13. Kelly, K. L., Coronado, E., Zhao, L. L. and Schatz, G. C. (2003). J. Phys. Chem. B, 107, 668.

14. Millstone, J. E., Metraux, G. S. and Mirkin, C. A. (2006). Adv. Funct. Mater., 16, 1209.

15. Seo, W. S., Jo, H. H., Lee, K., Kim, B., Oh, S. J. and Park, J. T. (2004). Angew. Chem., 116, 1135; Angew. Chem. Int. Ed., 43, 1115.

16. Hill, H. D. and Mirkin, C. A. (2006). Nat. Protoc., 1, 324.

17. Rivetti, C., Walker, C. and Bustamante, C. (1998). J. Mol. Biol., 280, 41.

18. SantaLucia, J. and Hicks, D. (2004). Annu. Rev. Biophys. Biomol. Struct., 33, 415.

19. Freier, S. M., Sugimoto, N., Sinclair, A., Alkema, D., Neilson, T., Kierzek, R., Caruthers, M. H. and Turner, D. H. (1986). Biochemistry, 25, 3214.

537

Chapter 23

Nanoparticle Superlattice Engineering with DNA*

Robert J. Macfarlane,a,b Byeongdu Lee,c Matthew R. Jones,b,d Nadine Harris,a,b George C. Schatz,a,b and Chad A. Mirkina,b,d aDepartment

of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cX-ray Science Division, Advanced Photon Source, Argonne National Laboratory, 9700 S. Cass Ave., Argonne, IL 60439, USA dDepartment of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA [email protected]

A current limitation in nanoparticle superlattice engineering is that the identities of the particles being assembled often determine the structures that can be synthesized. Therefore, specific crystallographic symmetries or lattice parameters can only be achieved using specific nanoparticles as building blocks (and vice *From Macfarlane, R. J., Lee, B., Jones, M. R., Harris, N., Schatz, G. C. and Mirkin, C. A. (2011). Nanoparticle superlattice engineering with DNA, Science 334, 204–208. Reprinted with permission from AAAS.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

540

Nanoparticle Superlattice Engineering with DNA

versa). We present six design rules that can be used to deliberately prepare nine distinct colloidal crystal structures, with control over lattice parameters on the 25- to 150-nanometer length scale. These design rules outline a strategy to independently adjust each of the relevant crystallographic parameters, including particle size (5 to 60 nanometers), periodicity, and interparticle distance. As such, this work represents an advance in synthesizing tailorable macroscale architectures comprising nanoscale materials in a predictable fashion. The crystallographic lattice adopted by a given set of atomic and molecular components is often difficult to predict and control and is dependent on a large number of factors. For ionic solids, Pauling developed rules that explain the relative stabilities of different lattices of simple salts, but these rules do not allow for structure control [1]. This is because parameters such as size and charge for atoms (and small molecules) are not tunable; changing an atom’s size or charge inherently changes the electronic properties that affect relative lattice stability. In principle, nanoparticle-based superlattice materials should allow for more control over the types of crystal lattice that they adopt, given that one can tune multiple variables (such as nanoparticle size or the presence of different organic molecule layers on the nanoparticle surface) to control superlattice stability [2–14]. Although advances have been made using a variety of electrostatic forces [7–9], covalent and noncovalent molecular interactions [6, 11], and biologically driven assembly strategies [2–5, 12], predictable architectural control remains an elusive goal, regardless of the type of particle interconnect strategy chosen. In 1996, the use of oligonucleotides as particle-directing motifs to synthesize amorphous polymeric materials from polyvalent particles modified with nucleic acids was demonstrated [2]. Subsequent work showed that crystallization and lattice control were possible for face-centered cubic (fcc) and body-centered cubic (bcc) crystal structures simply by taking advantage of the programmable nature of DNA (both in base sequence and in overall oligonucleotide length) [3–5, 15–18]. Herein, we describe a set of rules for using programmable oligonucleotide interactions, elements of both thermodynamic and

Nanoparticle Superlattice Engineering with DNA

kinetic control, and an understanding of the dominant forces that are responsible for particle assembly to design and deliberately make a wide variety of crystal types. Like the rules for atomic lattices developed by Pauling, these are guidelines for determining relative nanoparticle superlattice stability, rather than rigorous mathematical descriptions. However, unlike Pauling’s rules, the set of rules below can be used not only to predict crystal stability but also to deliberately and independently control the nanoparticle sizes, interparticle spacings, and crystallographic symmetries of a superlattice (Fig. 23.1A). This methodology represents a major advance toward nanoparticle superlattice engineering, as it effectively separates the identity of a particle core (and thereby its physical properties) from the variables that control its assembly. We used polyvalent conjugates of DNA and gold nanoparticles (DNA-NPs) as the basic building blocks for assembling superlattices, for which programmable recognition and hybridization interactions between DNA strands drive the assembly process (Fig. 23.1B). The key hypothesis in this work is that the maximization of DNA hybridization events between adjacent particles is a more important factor in determining lattice stability than all other forces in the system. Synthetically controllable variations in nucleotide sequence allowed us to change the overall hydrodynamic size and coordination environment (and thus the hybridization behavior) of the particles, without the need to alter the structure of the inorganic nanoparticle core [2–5, 15–18]. We used synchrotron-based small-angle x-ray scattering (SAXS) to characterize all lattices reported herein, because it allows for in situ analysis of highly solvated materials. We also have developed a complementary method to embed these superlattices in a resin, which enables their characterization by transmission electron microscopy (TEM) [17]. However, we note that the embedding process results in a slight deformation and disordering of the lattices, and that it significantly reduces the crystal lattice parameters as determined with in situ SAXS measurements. In a typical experiment, we assembled DNA-NP superlattices using oligonucleotide linker strands that, upon binding to a DNA-NP, present a short, single-stranded DNA “sticky end” at a controllable

541

542

Nanoparticle Superlattice Engineering with DNA

distance from the nanoparticle surface [19]. This distance dictates the interparticle spacing in a programmable manner [16]. Because of the polyvalent nature of the DNA-NPs, each NP hybridizes to multiple linker strands and subsequently forms tens to hundreds of sticky-end duplexes to adjacent NPs, enabling the construction of lattices that are indefinitely stable under ambient conditions. However, because individual sticky-end connections are weak (a single sticky-end duplex is not stable on its own at room temperature) and therefore transient, upon thermal annealing, DNA-NPs can shift positions within the material to ultimately form ordered lattices [15]. Although all of the structures we describe are made with gold NPs, the assembly process should also be applicable to any other NP that can be densely functionalized with oligonucleotides. We determined structural characteristics for a total of 41 crystals that adopted one of nine crystal lattices. In addition to fcc and bcc structures, we also prepared the following lattices [19]: hexagonal close-packed (hcp); AB, isostructural with cesium chloride (CsCl); AB2, isostructural with aluminum diboride; AB3, isostructural with Cr3Si; AB6, isostructural with the alkali-fullerene complex Cs6C60; AB, isostructural with sodium chloride (NaCl); and simple cubic (sc). For each structure, we could tune lattice parameters by means of independent modifications to both oligonucleotide interconnect length and nanoparticle size. Rather than discuss each group of structures in turn [19], we describe a set of rules that constitute a design strategy for synthesizing a particular choice of one of the nine distinct crystallographic symmetries. Rule 1: When all DNA-NPs in a system possess equal hydrodynamic radii, each NP in the thermodynamic product will maximize the number of nearest neighbors to which it can form DNA connections. This occurs because maximizing the number of nearest neighbors in these systems in turn maximizes the number of potential DNA connections between nanoparticles, which we have hypothesized to be the driving force in forming ordered crystals. When using linkers with self-complementary sticky ends, where all particles can bind to all other particles in solution, the observed thermodynamic product is always an fcc lattice (Fig. 23.1C), a conclusion supported by theory

Nanoparticle Superlattice Engineering with DNA

[3]. When two sets of nanoparticles are functionalized with linkers that contain different but complementary sticky ends, particles can only bind to particles of the opposite type. A bcc lattice is therefore the most stable for these binary systems (Fig. 23.1D), rather than an fcc lattice, as each NP in a bcc lattice possesses more nearest neighbors of the opposite particle type. Note that this rule holds for a wide range of nanoparticle diameters and oligonucleotide lengths, and it can therefore be used to make many fcc and bcc lattices with well-defined and predictable lattice parameters over the 25- to 150-nm range. Rule 2: When two lattices are of similar stability, the kinetic product can be produced by slowing the rate at which individual DNA linkers dehybridize and subsequently rehybridize. For example, theoretical predictions show that, although they possess the same number of nearest neighbors, hcp lattices are slightly less stable than fcc lattices, and thus any hcp crystals observed would likely be kinetic products [20]. Indeed, we have observed hcp lattices in these systems, but only as metastable structures that reorganize into fcc lattices upon annealing [15]. Stable hcp lattices can be realized by annealing at lower solution temperatures and decreasing the local DNA density around a NP surface (Fig. 23.1E). These two changes both slow the DNA linker sticky-end release and rehybridization rates necessary for crystallization, and promote lattice growth over lattice reorganization, thereby stabilizing initial kinetic products. For example, by using long DNA strands (~30 nm) and NPs bearing a small number of linkers (7.2-nm NPs, 20 ± 3 DNA strands per particle) and annealing at 25° to 30°C, one can preferentially stabilize the growth of initial hcp-like lattices that form during early time points of the assembly process [15]. It is important to note that although this process can consistently be used to produce large (>1 µm) hcp lattices that are stable for extended periods of time (several weeks after formation), these structures are still kinetic products. Annealing hcp lattices at higher temperatures for several hours always results in the lattices reorganizing to an fcc structure.

543

544

Nanoparticle Superlattice Engineering with DNA

Figure 23.1  (A) Nanoparticle superlattice engineering with DNA, unlike conventional particle crystallization, allows for independent control of three important design parameters (particle size, lattice parameters, and crystallographic symmetry) by separating the identity of the particle from the variables that control its assembly. (B) The DNA strands that assemble these nanoparticle superlattices consist of (i) an alkyl-thiol moiety and 10-base nonbinding region, (ii) a recognition sequence that binds to a DNA linker, (iii) a spacer sequence of programmable length to control interparticle distances, and (iv) a “sticky end” sequence that drives nanoparticle assembly via DNA hybridization interactions. Although only a single linkage is shown schematically here, DNA-NPs typically contain tens to hundreds of DNA linkers per particle. (C to I) The superlattices reported herein are isostructural with (C) fcc, (D) bcc, (E) hcp, (F) CsCl, (G) AlB2, (H) Cr3Si, and (I) Cs6C60 lattices. From left to right, each panel contains a model unit cell (not to scale), 1D and 2D (inset) x-ray diffraction (SAXS) patterns, and a TEM image of resin-embedded superlattices, along with the unit cell viewed along the appropriate projection axis (inset). Lines in the model denote edges of the unit cell; individual DNA connections are omitted for clarity. SAXS data are plots of nanoparticle superlattice structure factor S(q) (y axis, arbitrary units) versus scattering vector q (x axis, Å−1). Black traces are experimental data; blue traces are modeled SAXS patterns for perfect lattices. All scale bars in the TEM images are 50 nm. See [19] for a complete list of particle sizes and lattice parameters.

Nanoparticle Superlattice Engineering with DNA

Rule 3: The overall hydrodynamic radius of a DNA-NP, rather than the sizes of its individual NP or oligonucleotide components, dictates its assembly and packing behavior. An important aspect of DNA-NP design is that the overall hydrodynamic radius of a DNA-NP is a combination of the NP diameter and the DNA length. As each of these parameters is independently controllable, one can easily synthesize two DNA-NPs with the same overall hydrodynamic radius but different NP core sizes (Fig. 23.2A). Thus, we could assemble NPs into three-dimensional (3D) structures with lattice parameters and interparticle distances that are not dictated solely by the sizes of the inorganic particle cores. This rule is well illustrated by the synthesis of CsCl lattices (Fig. 23.1F), which exhibit the same DNA-NP arrangement and connectivity as a bcc lattice but use two different NP core sizes. To create a range of CsCl lattices, we systematically changed the lengths of oligonucleotide linkers to obtain DNA-NPs with the appropriate hydrodynamic radii (Fig. 23.2B). Note that by simply changing the length of the oligonucleotide linkers, the inorganic particle radius and interparticle distance were independently programmed for nanoparticles ranging from 5 to 60 nm in diameter, with lattice parameters ranging from ~40 to ~140 nm. The inorganic NP core sizes in these lattices differed by as much as 30 nm and still exhibited equivalent packing and assembly behavior. Rule 4: In a binary system, the size ratio and DNA linker ratio between two particles dictate the thermodynamically favored crystal structure. For this rule, the “size ratio” is defined as the ratio of the DNA-NPs’ hydrodynamic radii (a sum of the inorganic particle radius and DNA linker length), and the DNA linker ratio is the ratio of the number of DNA linkers on the two different types of DNANPs. Size ratio can be predicted to affect the stability of different crystal symmetries because it determines the packing parameters of DNA-NPs within a lattice (i.e., the number and positions of adjacent particles to which a given DNA-NP can bind). The DNA linker ratio can also be expected to affect crystal stability, as it determines the number of DNA sticky ends available to form DNA connections with these adjacent particles. For example, by adjusting the size ratio of the DNA-NP components, lattices isostructural with AlB2 can be obtained (Fig. 23.1G; size ratio 0.64). By varying both the size ratio and the DNA linker ratio, lattices isostructural with Cr3Si can be

545

546

Nanoparticle Superlattice Engineering with DNA

made (Fig. 23.1H; size ratio 0.37, DNA linker ratio ~2)—an unusual example of a NP superlattice with this lower crystallographic symmetry. Finally, by using a DNA linker ratio of ~3, we synthesized a lattice that has no mineral equivalent but is isostructural with the alkali-fullerene complex Cs6C60 [21] (Fig. 23.1I; size ratio ~0.35). Note that the lattices in Fig. 23.1 are only individual examples of the many AlB2, Cr3Si, and Cs6C60 crystals synthesized with this method. These structures also have been constructed using multiple particle sizes (5 to 30 nm) and hydrodynamic radii (10 to 50 nm). By applying rule 3, one can tune the hydrodynamic radii of particles (and thus the hydrodynamic size ratio) to position particles into a specific crystallographic symmetry without being restricted to specific inorganic particle sizes or even to specific inorganic particle size ratios. Indeed, the hydrodynamic radii of the particles can even be tuned such that in a given system, the DNA-NP with the larger inorganic core size possesses the smaller hydrodynamic radius. In this way, one can position a given nanoparticle at any of the occupied Wyckoff positions within a given lattice type’s unit cell, regardless of the inorganic particle’s size (Fig. 23.2C). Rule 5: Two systems with the same size ratio and DNA linker ratio exhibit the same thermodynamic product. Note that crystal stability is determined by the ratio of the two variables discussed in rule 4, not their absolute values. A comparison of the lattices created with rule 4 shows that, regardless of the absolute values of DNA-NP size or the number of DNA linkers per particle, two systems with the same size ratios and DNA linker ratios form the same thermodynamic product. Consequently, the application of rule 5 as a guiding principle in superlattice assembly enables a large number of lattices to be synthesized without necessitating a complete reanalysis of the forces involved in assembly for each specific nanoparticle size or DNA length. Further, this result implies that one could construct a phase diagram that would predict the most stable crystal structure as a function of these two variables. As previously mentioned, the main hypothesis of this work states that the thermodynamic products in this assembly method are the ones that maximize DNA duplex formation. However, experimental verification of this hypothesis (and thus the development of a phase diagram) is challenging, as it is difficult to experimentally determine the number of DNA duplexes

Nanoparticle Superlattice Engineering with DNA

formed in a given lattice. Therefore, we have constructed a model that is based on the predictable and well-established properties of both DNA (persistence length, rise per base pair) [22] and DNA-NPs (number of DNA strands per particle, the hybridization behavior of sticky ends) [16] and used this model to calculate relative crystal stabilities.

Figure 23.2  (A) Two particles with the same hydrodynamic radius exhibit the same assembly behavior, regardless of the sizes of the inorganic nanoparticle cores. (B) SAXS patterns for CsCl lattices in binary systems where two particles have the same hydrodynamic radii but different inorganic core sizes. The inset and model show the relative sizes of the nanoparticles, DNA linkers, and assembled lattices, all drawn to scale. From top to bottom, the nanoparticle sizes are 60 and 40 nm, 40 and 20 nm, and 30 and 10 nm. (C) SAXS patterns for AlB2 lattices, demonstrating that crystallographic symmetry and lattice parameters can be controlled independently of the sizes and size ratios of the inorganic nanoparticle cores (inset and model, both drawn to scale). From top to bottom, the inorganic core sizes of the “big” and “small” nanoparticles (as defined by their overall hydrodynamic radii) are 10 and 10 nm, 20 and 10 nm, and 5 and 10 nm. See Ref. [19] for exact interparticle distances and lattice parameters for all structures.

547

548

Nanoparticle Superlattice Engineering with DNA

Figure 23.3  (A) Surface plot of modeled data, in which the percentage of DNA sticky ends that form duplexes (z axis) is calculated for different crystallographic arrangements as a function of experimentally controllable design parameters (DNA linker ratio, x axis; DNA-NP size ratio, y axis). (B) Phase diagram constructed as a top-down view of (A), where each dot on the graph represents a lattice that was synthesized experimentally. The color of each experimental data point denotes the identity of the lattice obtained. (C) Two-dimensional “slice” through the plot in (B), at a constant DNA linker ratio of 1.0. This plot demonstrates the relative stability of both lattices that have been constructed with DNA-programmed assembly (color traces) and other lattices that have been theoretically predicted or synthesized using other assembly methodologies (black traces). The inset indicates where this slice was taken from (B).

The foundations for this model (hereafter referred to as the complementary contact model, or CCM) are the assumptions that (i) DNA linker sticky ends must be able to physically contact one another to hybridize, and (ii) any sticky ends that can come into contact will eventually form a DNA duplex. The DNA linker strands on the surface of a DNA-NP are dynamic and can therefore be treated as a single collective entity [16]. This allows one to represent a DNANP as a “fuzzy sphere” rather than a particle with a discrete set of DNA linkers. Because DNA linker sticky ends must physically contact one another to form a DNA duplex, it is therefore assumed that a greater amount of surface contact between adjacent spheres that contain complementary sticky ends correlates to a larger number of DNA duplexes being formed. By using the physical characteristics of the DNA-NP building blocks mentioned above, one can design a model lattice of arbitrary symmetry that has the appropriate lattice parameters (as dictated by a given set of particle sizes and DNA lengths), and then use the CCM to determine how many complementary sticky ends are able to contact one another and subsequently hybridize. This process enables the prediction of the number of DNA duplexes in a given crystal structure as a function of size ratio and DNA linker ratio (Fig. 23.3A) [19]. If the main hypothesis of this work is correct, a larger number of DNA duplexes formed for

Nanoparticle Superlattice Engineering with DNA

a given crystal structure should correlate to a more stable lattice. Thus, the lattice with the most surface contact between adjacent complementary spheres should possess the greatest number of DNA duplexes and therefore should be the most stable phase for a given set of variables. Although the CCM is not intended to provide an explicit solution for determining the most stable crystal structure for a given set of design parameters, it should provide a suitable means to test both rule 5 and the hypothesis that maximization of DNA hybridization is the driving force for forming ordered crystals. A comparison of the modeled phase diagram to experimentally obtained data shows that the model correctly predicts the structures obtained for a wide range of DNA-NP size ratios and DNA linker ratios, confirming the predominant hypothesis of this work as well as rules 4 and 5 (Fig. 23.3B). The model was also used to confirm that the lattices obtained experimentally are more stable than a number of other structures that have been predicted by previous theoretical calculations or that have been assembled with other methodologies (Fig. 23.3C) [8, 23]. Although there are limitations to the predictive nature of the CCM as it currently is constructed [19], the vast majority of the data generated by the model are in complete agreement with the synthesized lattices. Given that all experimentally generated data points validate the six rules developed in this work, it is reasonable to assume that simplifications used to develop the CCM are the result of this discrepancy. Nonetheless, the strong agreement between experiment and theory demonstrates that the CCM should provide a solid basis for future computational work in this area. As a result, the control over experimental design parameters (hydrodynamic size ratios, inorganic particle radii, and DNA lengths) afforded by this DNA-based assembly method and coupled with the predictive nature of this phase diagram, allows one to determine the experimental variables necessary to create a diverse array of lattices a priori, with independent control over crystallographic symmetry, lattice parameters, and nanoparticle sizes. Rule 6: The most stable crystal structure will maximize all possible types of DNA sequence–specific hybridization interactions. The examples above examine relatively simple binary systems, where only a single type of DNA sticky-end duplex is created. However, because of the polyvalent nature of the DNA-NPs and the base sequence programmability of DNA, one is not necessarily restricted to a single type of favorable particle interaction in a given lattice. By

549

550

Nanoparticle Superlattice Engineering with DNA

cofunctionalizing a nanoparticle with different linkers that contain different base sequences, multiple sequence-specific DNA duplex interactions are possible (Fig. 23.4A). This is an inherent distinction and potential advantage of using a sequence-programmable linker such as DNA, as opposed to entropy- or charge-dominated assembly processes.

Figure 23.4  (A) More complex nanoparticle assemblies can be created when programming multivalent DNA-NP interactions. For example, by encoding multiple distinct sticky-end sequences on the same particle, both self-complementary and non-self-complementary interactions can be used to assemble lattices. (B and C) This strategy can be used to create a NaCl lattice (B) when using two particles with different inorganic core sizes, or a simple cubic lattice (C) when using two particles with the same inorganic core size. From left to right, each panel shows a model unit cell, 1D and 2D (inset) SAXS data, and a TEM image with the unit cell viewed along the appropriate projection axis (inset). In (B), the SAXS data correspond to a NaCl lattice with 15-nm and 10-nm AuNPs and the TEM image is of a NaCl lattice with 30-nm and 15-nm AuNPs. In (C), the SAXS data correspond to a simple cubic lattice with 10-nm AuNPs and the TEM image shows a simple cubic lattice with 15-nm AuNPs. Scale bars, 50 nm.

Nanoparticle Superlattice Engineering with DNA

This rule was tested by cofunctionalizing a nanoparticle with two different linkers: one that bore a self-complementary sticky end, and one that bore a sticky-end sequence complementary to the sticky ends of a second particle. In this system, the cofunctionalized particle (blue particle, Fig. 23.4A) exhibited an attractive force with respect to all particles encountered in the system, whereas the second particle (red particle, Fig. 23.4A) was only attracted to the first particle type. When the hydrodynamic radius size ratio of the two NPs was ~0.3 to 0.4, the sticky ends were presented at the correct distances from the particle surface to form a NaCl lattice (Fig. 23.4B); that is, the selfcomplementary and non–self-complementary linkers were both at a position to form duplexes in this crystallographic arrangement. Furthermore, when the inorganic core sizes were the same on both DNA-NPs, the particles formed a simple cubic lattice, as defined by the positions of the inorganic cores (Fig. 23.4C). Although NaCl and simple cubic structures are presented as the first examples of this multivalent strategy, one can envision even more sophisticated and complex systems (such as lattices with three or more nanoparticle components) using multiple DNA-programmed NP interactions. We have presented a set of basic design rules for synthesizing a diverse array of nanoparticle superlattices using DNA as a synthetically programmable linker. These rules provide access to an easily tailorable, multifaceted design space in which one can independently dictate the crystallographic symmetry, lattice parameters, and particle sizes within a lattice. This in turn enables the synthesis of many different nanoparticle superlattices that cannot be achieved through other methodologies. Indeed, superlattices that do not follow the well-known hard-sphere packing parameter rules defined by Schiffrin and co-workers [6] and Murray and coworkers [8, 24] can easily be assembled as thermodynamically stable structures over a range of nanoparticle sizes and lattice parameters. The understanding gained from the use of these rules will both inform and enable future assembly efforts, allowing for the construction of new crystallographic arrangements that have emergent properties for use in the fields of plasmonics [14, 25, 26], photonics [27], catalysis [28, 29], and potentially many others.

551

552

Nanoparticle Superlattice Engineering with DNA

Acknowledgments Supported by the Defense Research & Engineering Multidisciplinary University Research Initiative of the Air Force Office of Scientific Research and by the U.S. Department of Energy Office of Basic Energy Sciences [award DE-SC0000989; Northwestern University (NU) Non-equilibrium Energy Research Center] (C.A.M. and G.C.S.); a National Security Science and Engineering Faculty Fellowship from the U.S. Department of Defense (C.A.M.); a NU Ryan Fellowship (R.J.M.); and a NU Ryan Fellowship and a NSF Graduate Research Fellowship (M.R.J.). Portions of this work were carried out at the DuPont-Northwestern-Dow Collaborative Access Team (DNDCAT) beamline located at Sector 5 of the Advanced Photon Source (APS). DND-CAT is supported by E. I. DuPont de Nemours & Co., Dow Chemical Company, and the state of Illinois. Use of the APS was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under contract DE-AC02-06CH11357. The transmission electron microscope work was carried out in the EPIC facility of the NU Atomic and Nanoscale Characterization Experimental Center, which is supported by NSF-NSEC, NSF-MRSEC, Keck Foundation, the state of Illinois, and NU. Ultrathin sectioning was carried out at the NU Biological Imaging Facility, supported by the NU Office for Research.

References

1. Pauling, L. (1960). The Nature of the Chemical Bond, 3rd ed. (Cornell Univ. Press, Ithaca, NY). 2. Mirkin, C. A. Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607.

3. Park, S.-J., Lazarides, A. A., Storhoff, J. J., Pesce, L. and Mirkin, C. A. (2004). J. Phys. Chem. B, 108, 12375. 4. Park, S. Y., et al. (2008). Nature, 451, 553.

5. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451, 549. 6. Kiely, C. J., Fink, J., Brust, M., Bethell, D. and Schiffrin, D. J. (1998). Nature, 396, 444. 7. Kalsin, A. M., et al. (2006). Science, 312, 420.

8. Shevchenko, E. V., Talapin, D. V., Kotov, N. A., O’Brien, S. and Murray, C. B. (2006). Nature, 439, 55.

References

9. Srivastava, S., et al. (2010). Science, 327, 1355.

10. Wong, S., Kitaev, V. and Ozin, G. A. (2003). J. Am. Chem. Soc., 125, 15589.

11. Zhao, Y., et al. (2009). Nat. Mater., 8, 979.

12. Chen, C.-L. and Rosi, N. L. (2010). Angew. Chem. Int. Ed., 49, 1924.

13. Nie, Z., Petukhova, A. and Kumacheva, E. (2010). Nat. Nanotechnol., 5, 15.

14. Jones, M. R., Osberg, K. D., Macfarlane, R. J., Langille, M. R. and Mirkin, C. A. (2011). Chem. Rev., 111, 3736.

15. Macfarlane, R. J., et al. (2009). Proc. Natl. Acad. Sci. U.S.A., 106, 10493. 16. Macfarlane, R. J., et al. (2010). Agnew. Chem. Int. Ed., 49, 4589.

17. Jones, M. R., et al. (2010). Nat. Mater., 9, 913.

18. Xiong, H., van der Lelie, D. and Gang, O. (2009). Phys. Rev. Lett., 102, 015504. 19. See supporting material on Science online.

20. Woodcock, L. V., et al. (1997). Nature, 385, 141. 21. Zhou, O., et al. (1991). Nature, 351, 462.

22. Bloomfield, V. A., Crothers, D. M. and Tinoco, I. (2000). Nucleic Acids: Structures, Properties, and Functions (University Science Books, Sausalito, CA). 23. Tkachenko, A. V. (2002). Phys. Rev. Lett., 89, 148303.

24. Bodnarchuk, M. I., Kovalenko, M. V., Heiss, W. and Talapin, D. V. (2011). J. Am. Chem. Soc., 132, 11967. 25. Kelly, K. L., Coronado, E., Zhao, L. L. and Schatz, G. C. (2002). J. Phys. Chem. B, 107, 668.

26. Fan, J. A., et al. (2010). Science, 328, 1135.

27. Stebe, K. J., Lewandowski, E. and Ghosh, M. (2009). Science, 325, 159. 28. Bell, A. T. (2003). Science, 299, 1688.

29. Grunes, J., Zhu, J., Anderson, E. A. and Somorjai, G. A. (2002). J. Phys. Chem. B, 106, 11463.

553

Chapter 24

Modeling the Crystallization of Spherical Nucleic Acid Nanoparticle Conjugates with Molecular Dynamics Simulations*

Ting I. N. G. Li,a Rastko Sknepnek,a Robert J. Macfarlane,b,c Chad A. Mirkin,a,b,c and Monica Olvera de la Cruza,b,c aDepartment of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA bDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA [email protected], [email protected]

We use molecular dynamics simulations to study the crystallization of spherical nucleic-acid (SNA) gold nanoparticle conjugates, guided by sequence-specific DNA hybridization events. Binary mixtures of SNA gold nanoparticle conjugates (inorganic core diameter in the 8−15 nm range) are shown to assemble into BCC, CsCl, AlB2, and Cr3Si crystalline structures, depending upon particle stoichiometry, *Reprinted with permission from Li, T. I. N. G., Sknepnek, R., Macfarlane, R. J., Mirkin, C. A. and Olvera de la Cruz, M. (2012). Modeling the crystallization of spherical nucleic acid nanoparticle conjugates with molecular dynamics simulations, Nano Lett. 12, 2509−2514. Copyright (2012) American Chemical Society. Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

556

Modeling the Crystallization of Spherical Nucleic Acid

number of immobilized strands of DNA per particle, DNA sequence length, and hydrodynamic size ratio of the conjugates involved in crystallization. These data have been used to construct phase diagrams that are in excellent agreement with experimental data from wet-laboratory studies. The strategy of using DNA to program the assembly of nanoparticles into macroscopic materials emerged in the mid1990s [1]. Over the past decade, substantial advances have been achieved in transforming this wet chemical technique from one that can be used to generate amorphous or pseudocrystalline architectures [2–7] into one that yields highly crystalline materials with a high degree of predictability in terms of crystal type and lattice parameters [8–13]. With this bottom-up approach, spherical nucleic acid gold nanoparticle conjugates (SNA-GNPs) can be used as artificial “atoms,” where the oligonucleotides connecting the SNAGNPs are “chemical bonds” that can be used to create novel one-, two-, and three-dimensional superlattices. In previous work, six rules for predicting superlattice structure have been developed [14]. These rules take into account the nanoparticle size, DNA length, overall number of DNA connections, and sequence complementarity to predict and understand the periodic packing of nanoparticles in discrete lattice types with lattice constants in the range of tens to hundreds of nanometers. This degree of tailorability provides the opportunity to control the properties of crystalline nanoparticle architectures with potential applications in a variety of fields, including medical diagnostics [15, 16], catalysis [17], energy conversion, and plasmonics.18 In order to better understand the complex phenomena associated with nanoparticle assembly, analytical [14, 19] and numerical models [20–26] have been proposed. To make them tractable, the theoretical models are greatly simplified and only consider factors such as the overlapping area of two complementary building blocks [14], or the volume fraction and the average coordination number of each type of building block [19], without addressing important aspects of the DNA chain conformations or dynamics. In order to explore these aspects, numerical models have been introduced. In recent simulations, crystal structures were predicted by computing the free energy [23] or comparing the stability of different crystals [24]. However, despite the great advances in our understanding

Modeling the Crystallization of Spherical Nucleic Acid

of these nanocomposite systems, there are still numerous open questions, such as the relative number of DNA connections formed for a given crystal phase, or the thermodynamics and kinetics of individual DNA binding events necessary to induce crystallization, which can in principle be addressed by simulations. Moreover, the majority of previous attempts to model DNA-guided crystallization processes have utilized DNA linkers that consist primarily of long, flexible single-stranded DNA (ssDNA) connections [20–26], but such structures have not yielded the diversity of structures that have been experimentally realized with double-stranded DNA (dsDNA) connections [14]. It is therefore necessary to model these types of systems to compare with experimental results in order to better understand the crystallization pathways and factors that determine the thermodynamically favored products for given SNAGNP designs. Unfortunately, due to a large number of DNA strands and nanoparticles that would need to be modeled to simulate the experimentally obtained structures, a full atomistic description is not computationally feasible. As a result, it is essential to develop a coarse-grained scheme that is detailed enough to capture relevant physical processes but still possible to simulate. In this study, we have modified the coarse-grained model of Knorowski, Burleigh and Travesset [25] developed for ssDNAnanoparticle assemblies to include effects of the chain stiffness that are more relevant for dsDNA. Using molecular dynamics (MD) simulations, we directly study the guided assembly and crystallization process of SNA-GNPs. This model faithfully mimics the relative size ratios and DNA coverage used in the experiments of Mirkin and co-workers [14] and is surprisingly robust, as arbitrary initial configurations of nanoparticles quickly self-assemble into several types of ordered crystals, where the crystal obtained is dictated by the initial design parameters. In particular, we have identified BCC, CsCl, AlB2, and Cr3Si structures. In addition, we have successfully simulated the assembly of crystals with nanoparticles of sizes up to 15 nm (with corresponding DNA coverage of around 100 strands per particle). This is, to the best of our knowledge, the largest crystal building block simulated thus far in terms of the number of simulated components per building block. Using nanoparticles in the 8−15 nm size range and DNA chain lengths from 8 to 25 nm, we have

557

558

Modeling the Crystallization of Spherical Nucleic Acid

obtained crystals with lattice constants in the 25−60 nm range. With this information, we have constructed a detailed phase diagram with over 100 data points for binary systems in an excellent agreement with the reported experiments [14]. In addition to obtaining structural information on the relative stability of different phases, the MD simulations also allow us to explore the dynamic aspects of the assembly process by tracking the nanoparticles’ positions in time and monitoring attachment and detachment events of the DNA linkers. This analysis provides valuable insight into the assembly routes and possible structures of these highly complex systems.

Figure 24.1  Cartoon of the coarse grained model. (a) A model DNA chain for a sequence used in the experiments [14]. The bead size for the dsDNA portion of the chain is σ, corresponding to a diameter of approximately 2 nm; the bead size for the ssDNA portion is 0.5σ, corresponding to an approximate diameter of 1 nm. (b) A model spherical GNP core. (c) An example of two SNA-GNPs: the particle on the left has an 8 nm gold core and 40 DNA chains; the particle on the right has a 10 nm core and 60 attached DNA chains.

Modeling the Crystallization of Spherical Nucleic Acid

Each SNA-GNP in this simulation method is modeled as a rigid spherical core with a fixed number of beads (designated as NC) and a fixed number of DNA chains (designated as NDNA) attached to it. We point out that although GNPs are typically faceted, at the resolution of our coarse-grained model a GNP can be well approximated by a sphere. In addition, previously published experiments [14] show that faceting of pseudospherical GNPs does not affect the symmetry of the final structure due to the flexibility of the SNA graft. Although particles with large shape anisotropy do show that particle shape has a strong influence on the symmetry of the assembled structure [13], the unique effects of particle shape on the assembly process are outside the scope of this work and will be examined in future studies. Systems with NC between 100 and 300 (corresponding to nanoparticle sizes between 8 and 15 nm), and NDNA between 40 and 100 were studied. A typical SNA-GNP is shown in Fig. 24.1. For the dsDNA part of a chain, approximately every five base pairs were represented with a single bead of size σ, where σ ≈ 2 nm. This represents the unit of length in our simulations and is also the approximate diameter of a dsDNA chain. In the ssDNA part, each bead of size σ/2 represents 2−3 bases. Each DNA chain is terminated with a sticky end used to model the directional hydrogen bond between complementary bases (A-T, C-G). Note that each bead of type X will only bind to a bead of its complementary type Y. The interaction between the sticky-end beads is modeled with the shifted LennardJones potential

Ï Ê Ê s ˆ 12 Ê s ˆ 6 ˆ ÔÔ 4e Á - Á ˜ ˜ - VXY (rc ) for r £ rc Ër¯ ¯ VXY (r ) = Ì Ë ÁË r ˜¯ (1) Ô for r > rc ÔÓ 0 



where r ∫ | ri - r j | is the distance between beads i and j and rc = 3σ is the cutoff distance. The unit of energy is set by ε and is on the order of the thermal energy at the experimental temperatures (~40°C); for simplicity, we assume εXY = 7ε. The active X- and Y-type beads were surrounded by flanking beads (FL) to mimic the directionality of a hydrogen bond (see Fig. 24.1), as introduced by Knorowski, Burleigh and Travesset [25]. Without the FL beads, more than two DNA linkers could easily bind together, which is not expected for dsDNA chains used in the experiments [14]. Interaction between

559

560

Modeling the Crystallization of Spherical Nucleic Acid

any other pairs of beads was modeled with the Weeks−Chandler− Andersen (WCA) soft-core repulsive potential [27] with rc = 21/6σ and εWCA ∫ ε. Covalent bonds along a DNA chain were modeled with a harmonic potential, Vbond = (1/2) k(r − r0)2 where k = 330ε/σ2 is the spring constant and r0 = 0.84σ is the native spring length. Stiffness of the dsDNA portion of the chain was modeled with an angle potential, Vangle = (1/2)kθ(θ − π)2 where kθ = 10ε and θ is the angle between three consecutive beads. kθ is chosen to provide a 0.255 nm rise per base pair as suggested by experiments [9]. For simplicity, as is common in coarse-grained simulations, all beads were assumed to have equal mass, m = 1. All temperatures are presented in the reduced, LennardJones units (i.e., reduced temperature T* is measured in terms of kBT/ε, where T is the real physical temperature). In this paper, we study binary systems, each containing two types of SNA-GNP building blocks; DNA-GNPs of one type have sticky ends complementary to the sticky ends of the SNA-GNPs of the other type. GNP cores with diameter between 8 and 15 nm were built by randomly but evenly placing beads on a sphere of radius R. A total of NDNA DNA chains were then attached at random sites on the GNP surface, creating a DNA graft. The DNA coverage varied with the GNP size, consistent with previously determined experimental values [28]. MD simulations were performed with the HOOMD-Blue [29, 30] package in the constant volume and temperature ensemble (NVT) with the temperature controlled via a Nosé–Hoover thermostat [31, 32]. GNP cores were treated as rigid bodies and simulated with the HOOMD-Blue rigid body package [33]. Between 36 and 64 SNA-GNPs were placed in a periodic simulation box in order to effectively simulate a bulk system. Instead of replicating a singular SNA-GNP template, each building block within a given simulation was generated independently and positioned randomly within the simulation box, avoiding any unphysical overlaps in the initial configuration. In this way, we ensured that the initial configuration of the system, which contained ~105 beads, had no bias toward any ordered structure. Although initial positions of the SNA-GNPs were random, the total number of the SNA-GNPs was chosen to be compatible with the number of nanoparticles needed to create the given number of unit cells (2 × 2 × 2 or 3 × 3 × 3) expected to form at that particular stoichiometric ratio of the two SNA-GNP types. Each simulation was carried over 6 × 107 time steps. With the unit of time,

Modeling the Crystallization of Spherical Nucleic Acid

τ, defined as τ = (mσ2/ε)1/2, a time step was chosen to be δτ = 2.5 × 10−3τ. A typical simulation took up to 48 h to complete on a single NVIDIA GTX 480 graphics card.

Figure 24.2  Snapshots of the crystal structures obtained from MD simulations. Left to right: The ideal crystal cell, a simulation snapshot after the system fully crystallized and the averaged positions of the SNA-GNPs over 2 × 106 time steps. CsCl: Each large (type A) SNA-GNP is built of a 15 nm GNP core and dressed with 100 DNA chains of length ≈ 19 nm; each small (type B) SNA-GNP is built with a 8 nm GNP core and dressed with 40 DNA chains of length ≈ 17 nm. AlB2: Each large (type A) SNA-GNP has a 10 nm GNP core and 60 DNA chains of length ≈ 24 nm; each small (type B) SNA-GNP has an 8 nm GNP core with 40 DNA chains of length ≈ 12 nm. Note that the simulation box is anisotropic and thus for clarity we show snapshots from three different angles. Cr3Si: Each large (type A) SNA-GNP contains a 13 nm GNP core and 80 DNA chains of length ≈ 17 nm; each small (type B) SNA-GNP contains an 8 nm GNP core with 40 DNA chains of length ≈ 10 nm. BCC lattices (not shown) occur in the special cases of the CsCl lattice when SNA-GNPs of both types have the same size GNP core. All snapshots were generated with the Visual Molecular Dynamics (VMD) package [34] and rendered with Tachyon ray-tracer [35].

561

562

Modeling the Crystallization of Spherical Nucleic Acid

By tuning the stoichiometric ratio of two SNA-GNP types, to 1:1, 1:2, and 1:3, we obtained CsCl, AlB2, and Cr3Si crystal lattices, respectively. In Fig. 24.2, we show simulation snapshots of representative cases for each of these crystal structures. For example, the unit cell of the CsCl structure (Fig. 24.2 shows a 3 × 3 × 3 lattice of unit cells) contains two atoms of types A and B with coordinates (0, 0, 0) (type A) and (a/2, a/2, a/2) (type B), where a is the lattice constant, determined at the end of the simulation. The obtained structure is clearly a CsCl lattice (also confirmed by computing the structure factor), although the SNA-GNP positions are not perfectly aligned due to thermal fluctuations around their equilibrium positions. In order to remove these thermal fluctuations, we averaged the positions of the centers of the SNA-GNPs over the last 2 × 106 time steps to obtain structures that almost perfectly match an ideal CsCl lattice. Note that in the special case when SNAGNPs of both types have equal nanoparticle core size, the structure generated at a 1:1 particle stoichiometry is a BCC lattice (as defined by the inorganic cores).

Figure 24.3  Phase diagrams as a function of the SNA-GNP size ratio and the DNA coverage ratios (large to small SNA-GNP) for three stoichiometric ratios, 1:1 (left), 1:2 (middle), and 1:3 (right). Filled symbols indicate that the corresponding crystal structure formed while the cross indicates that crystal structure did not show in the simulations; open diamonds are the experimental data points [14] with the same crystal structure as those obtained by simulations for a given stoichiometric ratio. Note that BCC structures are special cases of CsCl lattices where both GNP cores are the same size. Boundaries between different phases cannot be determined exactly from our simulations and thus are represented with dashed lines.

The predicted structure for 1:2 stoichiometry of particles is the AlB2 lattice, which has a hexagonal unit cell. It was therefore necessary to place 36 SNA-GNPs (1:2 stoichiometric ratio with 12 type A and 24 type B SNA-GNPs) into a noncubic orthogonal

Modeling the Crystallization of Spherical Nucleic Acid

simulation box, resulting in a crystal with 12 (2 × 2 × 3) unit cells. For lattices assembled at a particle stoichiometry of 1:3, 64 SNA-GNPs (16 type A and 48 type B SNA-GNPs) generated a 2 × 2 × 2 array of Cr3Si unit cells, as would be predicted based upon experimental results. Note that the experimentally observed Cs6C60 structure [14] has 14 SNA-GNPs per unit cell and obtaining it with our model is currently computationally not feasible. In the previous experimental study [14], the most stable crystal phase for different binary combinations of SNA-GNPs was determined to be dependent upon two factors: the hydrodynamic size ratio of the two particles and the relative number of DNA linkers attached to each particle type. The former variable dictated how nanoparticles packed within a given lattice type, thus determining the positions of the DNA strands within a lattice and how many of these DNA connections could physically reach one another and engage in hybridization. The latter variable dictated the relative number of each type of DNA linkage per unit cell, which determined the maximum number of DNA connections that could be formed. In Fig. 24.3, we show phase diagrams as a function of these variables. In order to ensure that the obtained crystalline structures correspond to the thermodynamic equilibrium, for each point shown in the phase diagrams, we performed up to 15 independent simulations. Each phase diagram was generated using an appropriate stoichiometric ratio of the two SNA-GNP types: 1:1 for the CsCl lattices, 1:2 for the AlB2 lattices, and 1:3 for the Cr3Si lattices, respectively. As the stoichiometric ratio within a lattice is determined at the beginning of each simulation, we construct separate phase diagrams for each of these three ratios. We note, however, that what happens if the initial stoichiometric ratio is not compatible with any crystalline structure (e.g., if noninteger stoichiometric ratios of particles are used) is still an open question. Although this has been examined briefly in recent experiments [14], a full investigation of the relative stabilities of different phases in these scenarios is still pending. In comparison to the experimental results reported by Mirkin [14], the majority of the previously collected experimental points fall within the phase diagrams obtained in our simulations. The minor discrepancies near the phase boundaries are attributed to the inherently insufficient resolution of the coarse grained model to accurately predict phase boundaries. In addition, there are clear

563

564

Modeling the Crystallization of Spherical Nucleic Acid

regions of the phase diagrams where different crystal structures can be obtained, depending upon the initial particle stoichiometry. This indicates that within these overlapping regions one can obtain different crystal structures by tuning the stoichiometric ratio of the two SNA-GNP types; this is also consistent with previous experimental results [14]. In order to ensure that the formation of the crystals is driven by the interactions between the DNA linkers rather than simply being a packing effect, we have performed a set of control simulations. Instead of using sticky ends, we applied the WCA repulsive potential between the end-groups to make them nonsticky. This was tested over a broad range of densities, and no crystals formed for any of the systems for which we obtained crystals when the sticky interactions were present. Clearly, the attractive interaction is crucial for the crystallization process to occur. We note that the coarse-grained model used in this study is not designed to quantitatively agree with the properties of the experimental system (e.g., with each system’s melting temperature, or the actual kinetics of DNA strand hybridization or particle reorganization). However, due to the accurate scale of the building blocks, it is able to successfully obtain assembled structures that are in very good agreement to those found experimentally with similar system designs. This reaffirms the notion that the assembly process is not sensitive to the atomistic details of the system but is rather robust and dictated by geometry of the building blocks. We highlight two central points essential for the crystallization process: (i) the slight flexibility of the DNA strands connecting particles together and (ii) the dynamic hybridization between the sticky ends. The small degree of flexibility allows for DNA strands to maximize hybridization interactions between particles, while the dynamic nature of the hybridization events allows the nanoparticle to sample several positions within a lattice and ultimately adopt a thermodynamically favored placement. Starting from the surface of the GNP core, a DNA chain is composed of a soft [36] ssDNA segment, followed by a stiffer dsDNA segment, and terminated with an ssDNA sticky end. The dsDNA imparts a tunability in both SNA-GNP hydrodynamic size and interparticle distance, but including small flexible portions provides enough flexibility in the DNA chains to allow adjacent strands of

Modeling the Crystallization of Spherical Nucleic Acid

complementary types to bend in order to hybridize to one another, even if their optimal configurations are not necessary directly parallel to one another. Therefore, when a SNA-GNP is surrounded by neighbors of the complementary type, its DNA chains stretch to reach and bind to as many DNA strands as possible. Conversely, a SNA-GNP will tend to avoid its neighbors that do not possess DNA strands by bending its DNA strands away from these neighboring particles to which it cannot hybridize. The dynamic nature of the hybridization between sticky ends is the other crucial ingredient for thermodynamic equilibrium to be reached. DNA hybridization (i.e., the formation of hydrogen bonds between complementary bases on two sticky ends) has a strength (~2−10 kBT/bond) between weak van der Waals forces (~kBT/ bond) and strong covalent (~102 kBT/ bond) or ionic bonds (~102 kBT/bond). Neither too strong nor too weak, hydrogen bonds can serve as a suitable stabilizing interaction between the SNA-GNPs and guide them to assemble into crystals. However, if the interaction between the complementary base pairs is strong enough to lead to the formation of too many interconnecting duplexes, the attraction between two SNA-GNPs becomes too strong and they remain locked into place, thus preventing reorganization of the assembled SNAGNPs into a thermodynamically favored structure [28]. On the other hand, if too few hybridized pairs form, there is not enough attraction to counteract the thermal motion of the SNA-GNPs, and the system remains liquid. Therefore, there is a rather narrow distribution of the number of hybridization pairs per SNA-GNP for which a stable equilibrium crystal can form. Outside this region, the system is either kinetically trapped in an amorphous state or it melts with concomitant dispersing of the particles. This accounts for the fact that the crystal structures form over only a relatively narrow range of temperatures, T* = 1.4−1.5, resulting in the corresponding bond strength between complementary beads of 4.5−5.0 kBT. Although the nature of the hybridization holds the key to understand the crystallization process, it is very hard to accurately measure its properties in an experiment. Guided by the observations from simulations, we identify two primary guiding principles in determining the appropriate strength of the hybridization for the crystallization of SNA-GNPs. First, optimal hybridization strength should lead to a sufficiently large but not excessive percentage of hybridizations when the system is at equilibrium. We find that the

565

566

Modeling the Crystallization of Spherical Nucleic Acid

optimum average percentage of hybridized pairs is in the 15−50% range. Second, linkers should be able to easily attach to and detach from their complementary counterparts in order to overcome kinetic traps. A direct tracking of the hybridization events reveals that a pair of complementary sticky ends is bound together for ~104 time steps, much shorter than the typical time it takes to form a crystal, which is ~107 time steps. Therefore, during the assembly process the system has a chance to explore a large region of the configuration space and is thus able to successfully equilibrate. We note that the lifetime of a hybridized pair is very sensitive to temperature with long-lived bonds at low T* which leads to the system being trapped in a metastable state. As evident from the MD simulations the dynamic binding of DNA linkers is a key feature for successful crystal assembly. It is important to note that this pathway of reorganization has been previously hypothesized and explored experimentally [28], and the results of the data presented here are consistent with the previous conclusions. However, these simulations provide the first examination of these hybridization and dehybridization events at the level of single DNA chains, rather than the previous experimental results which only examined collective effects of many DNA strands and SNA-GNPs in bulk measurements. In Fig. 24.4, we show the percentage of hybridized linkers during the crystallization process. As expected, this percentage increases during the assembly process and reaches a steady state once the crystal has been formed. We point out that in the steady state only the average number of the hybridizations is constant, and bonds between the sticky ends are dynamic, constantly breaking and reforming even once thermodynamic equilibrium has been achieved. In summary, we have performed MD simulations of a scaleaccurate coarse-grained model for gold nanoparticles grafted with dsDNA strands with short sticky ends. We have identified a number of ordered crystalline phases consisting of BCC, CsCl, AlB2 and Cr3Si lattices, respectively, and mapped a detailed phase diagram that is in an excellent agreement with the recent wet-laboratory experiments [14]. In addition, we have probed the crystallization process at the level of individual DNA hybridization and dehybridization events, confirming previous experimental data (examined only at the level of bulk, collective effects) that suggested that highly dynamic hybridization processes between semiflexible DNA linkers enables crystallization.

Modeling the Crystallization of Spherical Nucleic Acid

Figure 24.4  Top and middle: Mean square displacement, Msd(α)-( =   , α = A, B) as a function of the number of simulation steps for the larger SNA-GNPs of type A (top) and for the smaller SNA-GNPs of type  B (middle). ra (t) is the position of the center of a DNA-GNP of type α = A, B at time t and = (1/τmax)

Â

t max

t i =1

... is the average over τmax time steps. Bottom:

The percentage of hybridizations (p(H)) versus simulation time step. Time is measured in terms of the number of the simulation time steps δτ. All plots correspond to a 1:1 stoichiometric ratio of SNA-GNPs. The arrow indicates the time point at which the crystallization process has been completed. For time beyond the dashed line, the system is crystalline.

Acknowledgments We are grateful to Christopher Knorowski and Alex Travesset for sharing their scripts with us and to Joshua A. Anderson for sharing the HOOMD-Blue rigid body package prior to its release. This work was supported by the Defense Research and Engineering Multidisciplinary University Research Initiative of the Air Force Office of Scientific Research (FA9550-11-1-0275), the DoD NSSEFF program (FA955010-1-0167) and the Northwestern Nonequilibrium Energy Research Center from the DOE (DE-SC0000989). R.J.M. acknowledges Northwestern University for a Ryan Fellowship.

567

568

Modeling the Crystallization of Spherical Nucleic Acid

References 1. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607−609. 2. Mucic, R. C., Storhoff, J. J., Mirkin, C. A. and Letsinger, R. L. (1998). J. Am. Chem. Soc., 120, 12674−12675.

3. Jin, R. C., Wu, G. S., Li, Z., Mirkin, C. A. and Schatz, G. C. (2003). J. Am. Chem. Soc., 125, 1643−1654.

4. Valignat, M., Theodoly, O., Crocker, J. C., Russel, W. B. and Chaikin, P. M. (2005). Proc. Natl. Acad. Sci. U.S.A., 102, 4225−4229.

5. Maye, M. M., Nykypanchuk, D., van der Lelie, D. and Gang, O. (2006). J. Am. Chem. Soc., 128, 14020−14021.

6. Geerts, N., Schmatko, T. and Eiser, E. (2008). Langmuir, 24, 5118−5123. 7. Biancaniello, P. L., Kim, A. J. and Crocker, J. C. (2005). Phys. Rev. Lett., 94, 058302. 8. Park, S. J., Lazarides, A. A., Storhoff, J. J., Pesce, L. and Mirkin, C. A. (2004). J. Phys. Chem. B, 108, 12375−12380. 9. Hill, H. D., Macfarlane, R. J., Senesi, A. J., Lee, B., Park, S. Y. and Mirkin, C. A. (2008). Nano Lett., 8, 2341−2344.

10. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451, 549−552. 11. Park, S. Y., Lytton-Jean, A. K. R., Lee, B., Weigand, S., Schatz, G. C. and Mirkin, C. A. (2008). Nature, 451, 553−556.

12. Maye, M. M., Kumara, M. T., Nykypanchuk, D., Sherman, W. B. and Gang, O. (2010). Nat. Nanotechnol., 5, 116−120.

13. Jones, M. R., Macfarlane, R. J., Lee, B., Zhang, J., Young, K. L., Senesi, A. J. and Mirkin, C. A. (2010). Nat. Mater., 9, 913−917.

14. Macfarlane, R. J., Lee, B., Jones, M. R., Harris, N., Schatz, G. C. and Mirkin, C. A. (2011). Science, 334, 204−208.

15. Cutler, J. I., Auyeung, E. and Mirkin, C. A. (2012). J. Am. Chem. Soc., 134, 1376−1391.

16. Mirkin, C. A. (2010). MRS Bull., 35, 532−540.

17. Lin, L., Liu, Y., Tang, L. and Li, J. (2011). Analyst, 136, 4732−4737.

18. Sebba, D. S., Mock, J. J., Smith, D. R., LaBean, T. H. and Lazarides, A. A. (2008). Nano Lett., 8, 1803−1808. 19. Tkachenko, A. V. (2011). Phys. Rev. Lett., 106, 255501.

References

20. Lukatsky, D. B., Mulder, B. M. and Frenkel, D. J. (2006). Phys.: Condens. Matter, 18, S567−S580.

21. Largo, J., Starr, F. W. and Sciortino, F. (2007). Langmuir, 23, 5896−5905. 22. Lee, O. -S. and Schatz, G. C. (2009). J. Phys. Chem. C, 113, 2316−2321.

23. Martinez-Veracoechea, F. J., Mladek, B. M., Tkachenko, A. V. and Frenkel, D. (2011). Phys. Rev. Lett., 107, 045902.

24. Padovan-Merhar, O., Lara, F. V. and Starr, F. W. (2011). J. Chem. Phys., 134, 244701. 25. Knorowski, C., Burleigh, S. and Travesset, A. (2011). Phys. Rev. Lett., 106, 215501.

26. Fan, J. A., He, Y., Bao, K., Wu, C., Bao, J., Schade, N. B., Manoharan, V. N., Shvets, G., Nordlander, P., Liu, D. R. and Capasso, F. (2011). Nano Lett., 11, 4859−4864. 27. Weeks, J. D., Chandler, D. and Andersen, H. C. (1971). J. Chem. Phys., 54, 5237−5247. 28. Macfarlane, R. J., Jones, M. R., Senesi, A. J., Young, K. L., Lee, B., Wu, J. and Mirkin, C. A. (2010). Angew. Chem. Int. Ed., 49, 4589−4592.

29. Anderson, J. A., Lorenz, C. D. and Travesset, A. (2008). J. Comput. Phys., 227, 5342−5359. 30. http://codeblue.umich.edu/hoomd-blue/. 31. Nosé, S. (1984). Mol. Phys., 52, 255−268.

32. Hoover, W. G. (1985). Phys. Rev. A, 31, 1695−1697.

33. Nguyen, T. D., Phillips, C. L., Anderson, J. A. and Glotzer, S. C. (2011). Comput. Phys. Commun., 182, 2307−2313. 34. Humphrey, W., Dalke, A. and Schulten, K. (1996). J. Mol. Graphics, 14, 33−38. 35. Stone, J. E. (1998). An efficient library for parallel ray tracing and animation. M.Sc. Thesis, University of Missouri-Rolla, Rolla, Missouri. 36. Tinland, B., Pluen, A., Sturm, J. and Weill, G. (1997). Macromolecules, 30, 5763−5765.

569

Part 4

Building Blocks for Crystal Engineering

Chapter 25

Synthetically Programmable Nanoparticle Superlattices Using a Hollow Three-Dimensional Spacer Approach*

Evelyn Auyeung,a,b Joshua I. Cutler,b,c Robert J. Macfarlane,b,c Matthew R. Jones,a,b Jinsong Wu,d George Liu,d Ke Zhang,b,c Kyle D. Osberg,a,b and Chad A. Mirkina,b,c aDepartment of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA bInternational

Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA

cDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA dElectron

Probe Instrumentation Center, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA [email protected]

Evelyn Auyeung and Joshua I. Cutler contributed equally to this work.

Crystalline nanoparticle arrays and superlattices with well-defined geometries can be synthesized by using appropriate electrostatic

*Reprinted with permission from Auyeung, E., Cutler, J. I., Macfarlane, R. J., Jones, M. R., Wu, J., Liu, G., Zhang, K., Osberg, K. D. and Mirkin, C. A. (2012). Synthetically programmable nanoparticle superlattices using a hollow three-dimensional spacer approach, Nat. Nanotechnol. 7, 24–28.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

574

Synthetically Programmable Nanoparticle Superlattices

[1–3], hydrogen-bonding [4, 5] or biological recognition interactions [6–11]. Although superlattices with many distinct geometries can be produced using these approaches, the library of achievable lattices could be increased by developing a strategy that allows some of the nanoparticles within a binary lattice to be replaced with “spacer” entities that are constructed to mimic the behavior of the nanoparticles they replace, even though they do not contain an inorganic core. The inclusion of these spacer entities within a known binary superlattice would effectively delete one set of nanoparticles without affecting the positions of the other set. Here, we show how hollow DNA nanostructures can be used as “threedimensional spacers” within nanoparticle superlattices assembled through programmable DNA interactions [7, 11–16]. We show that this strategy can be used to form superlattices with five distinct symmetries, including one that has never before been observed in any crystalline material.

25.1  Introduction

The programmability of DNA makes it a powerful tool for the rational assembly of nanoparticles into one- [10, 13], two- [13] and three-dimensional arrays [6, 7, 11, 13–17]. In the context of threedimensional particle assembly based on spherical nucleic acid (SNA) nanoparticle conjugates (spherical structures with a layer of densely packed, highly oriented nucleic acids) [6, 18], DNA can serve as a directing ligand that controls the placement of nanoparticles into specific positions within an ordered lattice [7, 11, 13–16]. Applying the three-dimensional spacer approach to this particular assembly scheme requires a DNA-functionalized nanoparticle without a core that exhibits the same binding and assembly behavior as the SNA gold nanoparticle conjugate (SNA–AuNP) [14–16]. Recently, our group developed a method for synthesizing hollow SNAs composed almost exclusively of DNA [18]. These hollow SNAs, which can be fabricated to have a range of sizes, are synthesized by the surfacecatalysed crosslinking of alkyne-bearing oligonucleotides bound to the surface of gold nanoparticles (Fig. 25.1a) [18], and they exhibit many of the hallmark properties of SNA–AuNPs that are important

Introduction

for programmed assembly, such as cooperative binding [15]. The monodispersity and cooperative binding of the hollow SNAs allow them to be used as three-dimensional spacers that occupy specific lattice positions based on their DNA sequence (Fig. 25.1b,c). In this work, three-dimensional spacers were first used to prepare a simple cubic arrangement of gold nanoparticles from a body-centred cubic (bcc) precursor lattice. This technique was then extended to more complex binary AB2 and AB6 precursor lattices, with either the “A” or “B” particles being selectively replaced by the spacers to form multiple distinct superlattices. The simplest binary system, in which the particles are identical in size and added in equal ratios, has been shown to result in a bcc lattice [7, 16] (Fig. 25.2a). This structure is achieved using a binary linker system such that the DNA linker sticky-end sequence on the surface of one gold nanoparticle type (5¢-AAGGAA-3¢) is complementary to the sequence on the surface of the other particle type (5¢-TTCCTT-3¢). One can modify this strategy to create a simple cubic arrangement of gold nanoparticles (Fig. 25.2b) by substituting hollow SNAs for one of the nanoparticle components in the assembly process. The SNA–AuNPs and hollow SNAs were combined in a 1:1 ratio to form amorphous macroscopic aggregates that were subsequently annealed a few degrees below their melting temperature to form crystalline structures. Structural characterization of the resulting nanoparticle crystals was performed using small-angle X-ray scattering (SAXS). Because the fully organic hollow SNAs exhibit negligible X-ray scattering compared to gold [7], the SAXS data exhibit scattering solely due to the inorganic gold nanoparticles. As expected, when the gold is not dissolved and the SNA precursor is used in place of the hollow structure, a lattice with a bcc arrangement of gold nanoparticles was synthesized. However, when the nanoparticles were assembled using hollow SNAs functionalized with the same complementary linker sequence, a simple cubic lattice of gold nanoparticles formed, as confirmed by SAXS (Fig. 25.2a,b). Here, the hollow SNA occupies the position in the centre of each unit cell to produce an overall simple cubic arrangement of gold nanoparticles. The particles associate first, then reorganize into an ordered arrangement upon annealing, consistent with our previous work [14]. This implies

575

576

Synthetically Programmable Nanoparticle Superlattices

that the mechanical rigidity of the hollow SNAs, if it substantially differs from that of the SNA–AuNPs, does not significantly affect the crystallization process. (a)

AuNP-catalysed crosslinking

Dissolve AuNP template

AuNP with crosslinked DNA (c)

(b) AuNP

AuNP

No spacer (bcc lattice)

AuNP

3D spacer

3D spacer

Spacer incorporated (simple cubic lattice)

Figure 25.1  Use of a three-dimensional spacer in DNA-programmable crystallization of gold nanoparticles. (a) Schematic of the synthesis of threedimensional spacer particles by crosslinking alkyne-modified DNA on the gold nanoparticle surface and subsequent dissolution of the gold nanoparticle template to create hollow particles (shown in gray). (b, c) Assembly using a nonself-complementary binary DNA linker system (red and blue strands) results in a bcc lattice when only SNA–AuNPs are used (b), and in a simple cubic lattice when a spacer particle is used (c). The shaded region surrounding the gold nanoparticle and spacer particles denotes the crosslinked layer between the DNA binding region and the gold (or hollow) core.

One advantage of DNA-programmed crystallization is that it allows one to easily and independently vary the assembly parameters, specifically particle size and DNA length, while maintaining the original superlattice symmetry [15, 16, 19]. Assembly strategies that rely on other interactions (for example, electrostatic or hydrophobic/hydrophilic) between ligand shells

Introduction

are generally sensitive to the surface area and/or volume occupied by the particles, so certain lattice symmetries can only be formed using a fixed particle size range or ligand composition [1, 2, 20–25]. The incorporation of a three-dimensional spacer into the assembly scheme adds an additional variable [16] (in addition to DNA linker length and particle size) that one can use to modulate the lattice parameters. This affords a level of spatial control over the placement of the inorganic particles that cannot be achieved using SNA–AuNPs (of a particular diameter) and linear linkers alone [15, 26]. Thus, by using hollow SNAs of varying diameters, a simple cubic lattice was easily formed using this new technique with 5, 10 and 20 nm gold nanoparticles, as demonstrated by SAXS (Fig. 25.3). The rise per base pair (bp) value as a function of linker length for a given particle size was found to be 0.247 ± 0.004 nm bp [21], demonstrating that this method maintains the nanometre precision in lattice parameter tunability that is one of the hallmarks of DNA-mediated nanoparticle assembly [15, 19]. More complex structures can also be obtained by varying the particle molar ratios and DNA lengths [16]. Specifically, a series of AB2-type crystals (isostructural with aluminium diboride; Fig. 25.2c) and AB6-type crystals (isostructural with the cesium fullerene complex Cs6C60; Fig. 25.2f ) have been synthesized, demonstrating the versatility of this DNA-based crystallization method and providing an opportunity to further expand the lattice symmetries available with a three-dimensional spacer approach. To achieve these structures, the hydrodynamic radii for the two particle types were adjusted by increasing the length of one of the linker strands (from 25 bases to 66 bases for the AB2-type lattices, and to 104 bases for the AB6-type lattices) as well as the sizes of the initial gold nanoparticles utilized (the A-type particle was changed from 10 nm to 20 nm for the AB6-type lattices). For the AB2 precursor, selectively replacing the A-type particles with hollow SNAs was predicted to result in a simple hexagonal lattice (Fig. 25.2d), while replacing the B-type particles with hollow SNAs was predicted to result in a graphite-like arrangement (Fig. 25.2e). SAXS data confirm that both lattices were synthesized, with the predicted crystal symmetry and lattice parameters. Although hexagonally close-

577

578

Synthetically Programmable Nanoparticle Superlattices

packed arrays of nanoparticles are quite commonly made by many assembly strategies [3, 4, 13, 16, 22, 25, 27], simple hexagonal lattices of nanoparticles (Fig. 25.2d) are more difficult to make because the close-packed structure is often energetically more favourable [27]. Furthermore, to the best of our knowledge, a graphite arrangement (Fig. 25.2e) of spherical nanoparticles has never been made by any assembly method. However, by applying the three-dimensional spacer approach to the AB2 lattice, both of these lattices are easily synthesized and exhibit crystalline order with micrometre-sized domains. In the case of the AB6 precursor structure, selectively replacing the A-type particles with spacers was predicted to form a structure that is not known to exist for any naturally occurring or synthetic material, and that has not been synthesized using any previously reported assembly strategy. The A-type hollow SNAs were added to an excess of B-type SNA–AuNPs to synthesize this structure, which we call “lattice X” (Fig. 25.2h); the synthesis of this lattice was confirmed by SAXS. The inverse procedure, adding A-type SNA– AuNPs to an excess of B-type hollow SNAs, results in the formation of a bcc lattice (Fig. 25.2g), as expected. To visualize these mixed superlattices and their related precursor structures, the nanoparticle crystals were embedded in a polymer resin, which was subsequently sectioned and imaged using transmission electron microscopy (TEM) and electron tomography (ET). The SAXS data unequivocally show the high degree of ordering in the formed superlattices, and any disorder shown in the TEM images is a result of the embedding procedure [13, 16]. The TEM images provide secondary evidence of the structures of the AB6-type superlattices (Fig. 25.4) by showing “snapshots” of the cross-sections of the lattices after sample embedding and lattice perturbation. In Fig. 25.4c, for instance, the 10 nm particles are disordered, but the hexagonal pattern of voids left by the spacer particles is clearly visible. ET was used to generate a three-dimensional reconstruction of one of the AB6-type lattices, which shows ordering along multiple zone axes (Fig. 25.4d–f). Surprisingly, as suggested by the TEM images and reconstruction, the hollow SNAs are stable during the embedding process, showing their ability to behave as a robust assembly tool.

Introduction

− (110)

0.04 q(Å-1)

− (220) − (310) − (222)

S(q)

− (200)

5

0.01

0.06

0.03

q(Å-1)

0.04

− (211)

− (220) − (310) − (222)

S(q)

− (220) − (310) − (222)

− (200)

S(q)

− (211)

7

− (200)

− (110)

h

0.02

− (210)

− (211)

0.06

6

0.05

q(Å-1) − (110)

q(Å-1)

g

4

− (211)

− (110)

− (001)− (100) − (101) − (110) − (111)

f

0.04

0.05

− (101)

0.02

S(q)

0.02

− (200)

− (111)

0.06

q(Å-1)

e

0.04

− (110) − (200) − (111) − (201)

S(q)

− (001)

3

− (111) − (201) − (112)

0.04

0.03

q(Å-1)

− (101) − (110) − (001)

S(q)

0.02

0.08

d

0.02

2

S(q)

− (211)

− (220) − (310) − (222) − (321)

0.04 0.06 q(Å-1)

− (100)

0.02

c

− (100)

− (110)

b 1

− (200)

S(q)

a

0.02

8

0.04 q(Å-1)

Figure 25.2  SAXS data for seven distinct gold nanoparticle superlattices and “lattice X.” (a–h) One- and two-dimensional X-ray scattering patterns (left and bottom right of each panel) and schematic unit cell (top right; crosslinked shell omitted) showing the formation of the following superlattices: bcc (a); simple cubic (b); AB2 (isostructural with AlB2) (c); simple hexagonal (d); graphitetype (e); AB6 (isostructural with Cs6C60) (f); bcc (g); “lattice X” (h). Five of the seven lattices (b,d,e,g,h) are made using a mixture of gold and hollow SNA particles. Red traces are the experimentally obtained scattering patterns and the black peaks are the theoretical scattering for each respective lattice. Higher reflections were not indexed for the purpose of clarity.

579

Synthetically Programmable Nanoparticle Superlattices

In summary, we have developed a new nanoparticle assembly strategy whereby voids are selectively programmed into a lattice of inorganic nanoparticles to produce superlattices with distinct symmetries that are difficult to obtain and, in some cases, have not been observed before. Although most assembly methods to date have focused on controlling particle density and symmetry through modifying the strength of interparticle forces, the use of a threedimensional spacer circumvents the need for direct connections between inorganic nanoparticles. Furthermore, although methods have been developed to create binary crystals of SNA–AuNPs and in-organic quantum dots that are transparent to X-rays [28], our method is the first to use hollow spacer nanoparticles made entirely of DNA to create lattices of inorganic nanoparticles. We have also shown that the reorganization process towards nanoparticle crystals is core-independent and readily applicable to other nanoparticle types, as long as they are densely functionalized with DNA. The additional structural diversity afforded by this technique will be useful in the development and study of novel materials with functionality determined by the periodic arrangement of their constituent nanoscale building blocks. (100)

(110)

(111)

(200)

(210)

(211) 20 nm

S(q)

580

10 nm

5 nm

1

2 q/q0

3

Figure 25.3  SAXS data for cubic lattices made from nanoparticles of different sizes. SAXS data with indexed reflections corresponding to a simple cubic lattice made from 5 nm gold nanoparticles (gray line), 10 nm gold nanoparticles (black) and 20 nm gold nanoparticles (red). A hollow SNA particle is incorporated into the centre of each unit cell. Crosslinked shells in the schematic unit cells are omitted for clarity. Lines between particles denote edges of the unit cell, not direct connections between gold nanoparticles.

Methods

25.2  Methods 25.2.1 DNA Synthesis and Nanoparticle Functionalization The DNA strands were either synthesized on a MM48 automated oligonucleotide synthesizer (BioAutomation) or purchased from Integrated DNA Technologies. DNA functionalization of nanoparticles was performed according to procedures described in the literature [7, 14]. Briefly, thiolated strands were first reduced with 100 mM dithiothreitol (DTT) (Sigma Aldrich) for 1 h, followed by removal of the DTT using a size-exclusion NAP5 column (GE Life Sciences). The purified thiolated strands were subsequently added to gold nanoparticles at a ratio of approximately 4 nmol per 1 ml gold nanoparticles, and sodium chloride (NaCl) was slowly added to the SNA–AuNP solution over the course of several hours until the final salt concentration reached 1.0 M NaCl. The SNA–AuNP solution was placed in an incubator shaker at 40°C for two days to allow the gold-catalysed crosslinking of the pendant alkyne moieties to occur. The SNA–AuNPs were then purified by three centrifugation cycles, each of which was followed by removal of the supernatant and resuspension in Nanopure 18.2 MΩ water. After the third centrifugation cycle and removal of the final supernatant, the purified SNA–AuNPs were resuspended in phosphate buffered saline with a final salt concentration of 0.5 M. Particle and DNA concentrations were calculated using measurements from UV-vis-spectroscopy.

25.2.2  Synthesis of Hollow SNAs

In a typical experiment, gold nanoparticles with diameters of 5, 10 or 20 nm were densely functionalized with 3¢-alkylthiolated DNA consisting of a region of ten alkyne-modified thymine bases immediately preceding a five-base spacer region, followed by a programmable binding region. The programmable binding region was hybridized to linker sequences containing sticky ends used to direct assembly, and the five-base spacer region was used to push this recognition sequence away from the crosslinked region of the DNA and decrease steric inhibition of linker binding. The alkynes were allowed to crosslink, according to chemistry previously established by our group [29], forming a shell that encapsulates

581

582

Synthetically Programmable Nanoparticle Superlattices

the gold nanoparticle template. This SNA–AuNP inorganic particle was used in all lattices synthesized in this work. To form the hollow SNAs, the gold cores were dissolved by the addition of KCN to a final concentration of ~60 mM, followed by thorough dialysis to remove the remaining gold salt and KCN. DNA loading studies involving the hollow SNAs have shown that the number of strands per particle is nearly identical to that of their SNA–AuNP counterparts, resulting in similar DNA hybridization behavior [18]. When SNA– AuNPs and hollow SNAs were assembled with DNA linker strands, a sharp melting profile (full-width at half-maximum, FWHM, of first derivative = 2.09 ± 0.270°C) was observed for the AuNP–SNA aggregate, indicating cooperative binding in this binary particle system. Such transitions are characteristic of aggregates formed from particles that are densely functionalized with DNA [6, 30]. a

b

c

d

e

f

Figure 25.4  TEM images of the AB6-type lattices. (a) AB6-type lattice formed from 20 nm and 10 nm gold nanoparticles. (b) A bcc lattice formed from 20 nm gold nanoparticles and 10 nm hollow SNA spacers. (c) Lattice X structure formed from 20 nm hollow SNA spacers and 10 nm gold nanoparticles. All scale bars, 200 nm. Lattice projections shown in the insets are outlined in the TEM images. A three-dimensional reconstruction of a thin (100 nm) section of the lattice in b was obtained from electron tomography. (d) Representative snapshot of TEM images obtained in the tilt series, where the hole was used as a reference point for alignment (scale bar: 200 nm). (e,f) Snapshots from the reconstructed lattice along the [100] zone axis (e) and [111] zone axis (f) of a bcc lattice. Insets: perfect bcc lattice along each respective zone axis. A unit cell in each reconstructed lattice is outlined in red for clarity.

Methods

25.2.3  Nanoparticle Crystallization The two particle types were combined in the specified ratios and shaken vigorously at room temperature for 1 min. Visible aggregates formed within minutes and the solution was allowed to sit for ~1 h to maximize aggregation. To initiate reorganization of the nanoparticles to form ordered lattices, the aggregates were annealed at a temperature a few degrees below their melting temperature.

25.2.4  Small-Angle X-ray Scattering

The annealed aggregates were transferred to quartz capillaries for characterization by SAXS. All SAXS data were collected at the DuPont–Northwestern–Dow Collaborative Access Team (DNDCAT) beamline of Argonne National Laboratory’s Advanced Photon Source (APS) with 10 keV (wavelength λ = 1.24 Å) collimated X-rays calibrated with a silver behenate standard. Exposure times of 0.1 and 0.2 s were used.

25.2.5 Transmission Electron Microscopy and Electron Tomography

The superlattices were transferred to a resin and sectioned into thin (90%) and a lack of aggregation during the functionalization process. In addition, the DNA-functionalized complexes migrated through the gel at a significantly slower rate than the N3-NPs, indicative of their increased size (Fig. 26.1b). This size increase was

Introduction

also confirmed by DLS measurements, which showed an increase in hydrodynamic radii both after coating the particles with N3-PMAO and after DNA functionalization. For example, 7.0 nm CdSe/ZnS QDs were determined to have hydrodynamic radii of 8.1 ± 0.5 nm before polymer coating, 11.7 ± 0.5 nm after polymer coating, and 18.2 ± 1.3 nm after DNA functionalization (Fig. 26.1c). A crucial design aspect of converting nanoparticles into PAEs is modifying particles with a dense layer of oriented oligonucleotides [1, 26]. The large number of DNA strands oriented perpendicular to the particle surface results in both a high local DNA concentration, as well as a high localized salt concentration; together, these endow the particle with enhanced binding constants for complementary DNA strands and polyvalent interactions between particles that are necessary for their use in the construction of colloidal crystals [26– 28]. The average number of DNA strands conjugated to the surface of each type of nanoparticle reported herein was determined using fluorophore-labelled oligonucleotides and a fluorescence assay [18], and in all cases, the coverage was between 12 and 20 pmol cm–2. For comparison purposes, a particle with a 5 nm core (when including the polymer shell, the particle diameter is equivalent to a 13 nm gold particle) has on average 55 strands per particle, a coverage that is approaching what one can realize with thiol adsorption on gold (80 strands per 13 nm particle, 24 pmol cm–2 [29]). Once the size and surface functionalization of the PAE-NPs were determined, the potential to assemble them into a series of superlattices using established DNA linking strategies was explored [3, 5]. In a typical experiment, PAEs were combined with DNA linker strands that, once hybridized to the particle, presented a large number of short DNA sticky-end sequences at the periphery of each PAE-NP. When two PAE-NPs with complementary sticky ends came into contact in solution, their sticky ends hybridized to one another, thereby linking the particles together. Crystalline arrangements were formed by annealing the duplex-interconnected nanoparticles at a temperature a few degrees below their collective melting temperature and allowing them to reorganize into thermodynamically favoured structures. Synchrotron-based small-angle X-ray scattering (SAXS) was then used to determine the 3D arrangement of particles formed in solution.

591

A General Approach to DNA-Programmable Atom Equivalents b

S (q)

S (q)

a

0.02

0.04

0.06 -1 q (Å )

0.08

0.02

0.10

0.04

0.06

-1

0.08

0.10

q (Å )

d

S

S (q)

(q)

c

0.02 0.04 0.06 0.08 0.10 0.12 -1 q (Å )

0.02 0.04 0.06 0.08 0.10 0.12 -1

q (Å )

f S (q)

e S (q)

592

0.04

0.06

0.08 -1

q (Å )

0.10

0.02

0.04

0.06

0.08

0.10

-1

q (Å )

Figure 26.2  Characterization of fcc and bcc colloidal superlattices assembled from QD-, DAu- and Fe3O4-PAEs. (a–f) 2D and radially averaged 1D SAXS patterns of: QD-PAEs in fcc (a) and bcc (b) arrangements; DAu-PAEs in fcc (c) and bcc (d) arrangements; Fe3O4-PAEs in fcc (e) and bcc (f) arrangements. Experimental data are shown in black, and predicted scattering patterns are shown in red. For each pattern, a unit cell is shown of the corresponding crystal symmetry and particle composition.

The new PAEs were first used to synthesize face-centred cubic (fcc) and body-centred cubic (bcc) crystal structures consisting of particles with only a single inorganic core type; fcc structures form when the PAEs possess DNA sticky ends that are self-complementary, and bcc structures form when using two sets of PAEs with equal hydrodynamic radii whose sticky ends are complementary to one another [3, 5]. On addition of the appropriate DNA linker strands and thermal annealing, the PAEs formed both fcc and bcc crystal structures using QD, DAu, Fe3O4 and Pt particles (Fig. 26.2), as determined by SAXS. All crystals yielded SAXS patterns with a large number of sharp diffraction peaks, indicative of high-quality superlattices with micrometre-sized crystalline domains. Each of the crystal structures presented in Fig. 26.2 used relatively small nanoparticles (≤10 nm in diameter), with short DNA linker sequences (~8 nm per linker strand). Previous studies have

Introduction

demonstrated that interparticle distances in these superlattices can be changed by controlling the nanoparticle diameter and/ or DNA length, where each additional nucleobase in a linker adds ~0.26 nm to the distance between particles, allowing for nanometrescale control in crystal lattice parameters [16, 30]. Therefore, to verify the generality of this new PAE synthesis strategy in constructing atom equivalents of different sizes, we increased the hydrodynamic radii of the PAEs by using larger particles and longer DNA linker sequences (Fig. 26.3a). Fe3O4 nanoparticles ranging from 5 to 25 nm in diameter were functionalized with DNA and assembled into fcc crystal structures, using DNA linker lengths of either 8 or 15 nm. For each system, a high-quality fcc diffraction pattern was obtained, from which lattice parameters and interparticle distances were calculated. This allowed us to confirm that the nanometrescale control over interparticle distances and lattice parameters was also exhibited in these new superlattices, as the calculated lattice parameters were within ~5% of the predicted values for each of the fcc crystals synthesized. These data also allowed us to determine that the thickness of the polymer shell was ~4 nm. To directly visualize the assembled superlattices with microscopy techniques, they were first embedded in amorphous silica and then a polymer resin [31]. The silica- and resin-embedded lattices were subsequently ultra-microtomed into ~100-nm-thick slices and imaged using both scanning transmission electron microscopy (STEM) and energy-dispersive X-ray spectroscopy (EDX). STEM images confirmed both the crystallographic arrangement and lattice parameters calculated from the SAXS data (Fig. 26.3c), and EDX elemental mapping (Fig. 26.3c) clearly showed the presence of elements corresponding to the appropriate nanoparticle type (Fe, Cd, or Au). The design rules that have been established for the conventional Au nanoparticle-based PAE crystals provide access to crystal symmetries besides fcc and bcc by assembling particles whose hydrodynamic sizes are inequivalent [5]. To demonstrate that these design rules are also applicable to this new class of PAEs, particles with inequivalent hydrodynamic radii were combined (20-nmdiameter Fe3O4 nanoparticles with ~15 nm DNA linkers, and 5-nmdiameter Fe3O4 nanoparticles with ~8 nm DNA linkers; Fig. 26.3b) at the appropriate stoichiometry to form AlB2-type binary structures,

593

A General Approach to DNA-Programmable Atom Equivalents

and SAXS was used to confirm the formation of the AlB2 lattice. This was also confirmed with STEM by embedding the lattice in silica and resin (Fig. 26.3d). Note that the positions of the nanoparticles b

S (q)

a

S (q)

594

0.02

0.04

0.06

0.08

0.10

0.01

q (Å-1)

c

0.02

0.03

0.04

q (Å-1)

d

Figure 26.3  Characterization of superlattices assembled from Fe3O4-PAEs of different size. (a) Fe3O4 nanoparticles of various sizes were assembled into fcc lattices with controlled lattice parameters (LP; blue, 25 nm nanoparticles with 80 nm LP; magenta, 20 nm nanoparticles with 73 nm LP; red, 10 nm nanoparticles with 43 nm LP; and black, 5 nm nanoparticles with 36 nm LP). (b) 20 nm Fe3O4 nanoparticles functionalized with 15 nm DNA strands and 8 nm Fe3O4 nanoparticles functionalized with 8 nm DNA strands were used to assemble an AlB2-type structure. Experimental data are shown in black, and predicted scattering patterns are shown in red. (c) STEM image (top) and EDX elemental map (bottom) of the fcc lattice of 25 nm Fe3O4 nanoparticles shown in (a). (d) STEM image (top) and EDX map (bottom) of the AlB2-type lattice in (b). In EDX maps, the signal from Fe is shown in red. All scale bars: 100 nm.

Introduction

are slightly perturbed in these images, with greater perturbation for particles smaller than 10 nm in diameter, a consequence of the preparative techniques necessary to obtain STEM images [31]. Therefore, although the STEM images presented here are a rough snapshot of the crystals post-synthesis, in all cases, SAXS was the definitive characterization technique and unequivocally showed the ordering of these structures in their native, solution-phase state. In principle, an infinite set of new colloidal crystals can be assembled from these PAEs, where core composition and size can be varied independently from crystal symmetry and lattice parameters. To evaluate this assertion, we synthesized CsCl-type, AlB2-type and Cs6C60-type superlattices using various combinations of PAEs. For example, CsCl-type lattices were prepared from two different sizes of QDs, as well as combinations of any two types of QD-, DAu-, Fe3O4- and Pt-PAE with similar sizes (Fig. 26.4a–d). In each case, the PAEs behaved in a manner that was dictated by the linker strands, and the composition of the inorganic core played no role in determining the thermodynamically preferred crystal structure. This can be noted in the SAXS data, specifically by observing the relative intensity of the diffraction peaks in the CsCl structures. Although the QD, DAu, Fe3O4 and Pt nanoparticles used in these lattices were all of approximately equal size, their electron densities were significantly different. Therefore, although the diffraction peaks for each of these CsCl structures occurred at the same relative positions, their intensities were noticeably different (Fig. 26.4a–d). In addition to these CsCl lattices, the construction of AlB2-type and Cs6C60-type lattices with different combinations of Fe3O4, DAu and QD nanoparticles demonstrated the ability to change both the nanoparticle stoichiometry and coordination environment for nanoparticles of different materials in these superlattices (Fig. 26.4e,f). STEM imaging of resin-embedded CsCl, AlB2 and Cs6C60 lattices also confirmed the identities of the structures assigned with the SAXS data, where the differences in electron density for nanoparticles of different chemical compositions were readily apparent (Fig. 26.4g–k). For example, the DAu, QD and Fe3O4 nanoparticles could be easily distinguished in the DAu-QD and DAuFe3O4 binary samples, in which Au was much brighter than CdSe and Fe3O4 because it more readily scatters electrons. When EDX elemental mapping was performed on these STEM samples, the

595

A General Approach to DNA-Programmable Atom Equivalents b

c

0.02

0.04

q (A-1)

0.06

0.08

0.02

0.04

0.06 q (A-1)

Fe3O4 8 nm Au 4.5 nm

0.02

0.04

q (A-1)

0.06

q (A-1)

0.06

Au 4.5 nm Fe3O4 10 nm

S (q)

0.08

0.02

0.04

0.06

q (A-1)

h

g

0.04

0.08

f

Fe3O4 8 nm QD 7 nm

0.02

0.08

e

d

Fe3O4

Au

QD 7 nm

0.08

Au

QD

j

QD 7 nm Au 4.5 nm

S (q)

S (q)

QD 7 nm

S (q)

QD 3 nm

Fe3O4 10 nm

S (q)

a

S (q)

596

0.01

i

0.02

0.03

0.04

q (A-1)

Au

Fe3O4

k

Figure 26.4  Binary superlattices assembled from arbitrary combinations of QD-, DAu- and Fe3O4-PAEs. Nanoparticles of different compositions and sizes were used to synthesize lattices where particle size, particle composition, interparticle distance and crystal symmetry could be independently controlled. (a–f) SAXS data for: 7 nm QD, 3 nm QD CsCl lattices (a); 7 nm QD, 4.5 nm DAu CsCl lattices (b); 8 nm Fe3O4, 4.5 nm DAu CsCl lattices (c); 8 nm Fe3O4, 7 nm QD CsCl lattices (d); 10 nm Fe3O4, 4.5 nm DAu AlB2-type lattices (e); and 20 nm Fe3O4, 7 nm QD Cs6C60-type lattices (f). Experimental data are shown in black, and predicted scattering patterns are shown in red. (g–i) STEM images of 7 nm QD, 4.5 nm DAu CsCl lattices (g); 8 nm Fe3O4, 4.5 nm DAu CsCl lattices (h); and 10 nm Fe3O4, 4.5 nm DAu AlB2-type lattices (i). The insets at the right corner are higher-magnification images with labels denoting particle composition. (j, k) STEM images and EDX maps of 8 nm Fe3O4, 7 nm QD CsCl lattices (j), and 20 nm Fe3O4, 7 nm QD Cs6C60 lattices (k). In EDX maps, the signal from Fe is shown in red, and the signal from Cd is shown in cyan. All scale bars: 50 nm.

Methods

differences in nanoparticle composition were even more apparent, as signals unique to Au, Fe, and Cd were clearly identifiable at different locations in the STEM samples (Fig. 26.4j,k). We have developed a general method for functionalizing a wide variety of nanoparticles with a dense monolayer of DNA. These structures represent a new class of programmable atom equivalents, and because the method used to prepare them is in principle applicable to any nanoparticle with a hydrophobic capping ligand, a large number of PAEs can be generated and used in the construction of single- and multi-component nanoparticle superlattice materials. Indeed, a simple literature search for nanoparticle syntheses carried out in organic solvents reveals hundreds of different synthesis protocols for nanoparticles of many different compositions, sizes and shapes. By combining this set of atom-equivalent precursors with the assembly strategy reported herein, the scientific community can forge the foundation for a new periodic table of PAEs defined by particle size, shape, composition and oligonucleotide shell.

26.2 Methods

26.2.1 DNA Synthesis and Nanoparticle Functionalization DNA strands were either synthesized on an MM48 automated oligonucleotide synthesizer (BioAutomation) or purchased from Integrated DNA Technologies. The nanoparticles used for PAE synthesis were either obtained commercially or synthesized according to the procedures described in the literature.

26.2.2  Nanoparticle Crystallization

Once the appropriate DNA linkers were added to a solution of PAEs, visible aggregates formed within minutes. The solutions were then allowed to sit for 1 h to maximize aggregation. To initiate reorganization of the nanoparticles to form ordered lattices, the aggregates were annealed at a temperature a few degrees below their respective melting temperatures.

597

598

A General Approach to DNA-Programmable Atom Equivalents

26.2.3 SAXS The annealed aggregates were transferred to quartz capillaries for characterization by SAXS. All SAXS data were collected at the DuPont–Northwestern–Dow Collaborative Access Team beamline of Argonne National Laboratory’s Advanced Photon Source with 10 keV (wavelength l = 1.24 Å) collimated X-rays calibrated with a silver behenate standard. Exposure times of 0.5 s were used.

26.2.4  STEM and EDX

The silica-embedded superlattices were transferred to a resin (Embed 812, Electron Microscopy Sciences) and sectioned into thin (~100 nm) samples for imaging by STEM. STEM and EDX mapping experiments were conducted on a Hitachi HD-2300 microscope in z-contrast mode.

26.2.5  Gel Electrophoresis and DLS Measurements

Native gels containing 1% agarose were run at room temperature. DLS measurements were performed on a Malvern Zetasizer Nano-ZS (Malvern Instruments) with five parallel measurements.

Acknowledgments

C.A.M. acknowledges the support of DoD/NSSEFF/NPS Awards N00244-09-1-0012 and N00244-09-1-0071, AFOSR Awards FA9550-11-1-0275, FA9550-12-1-0280, and FA9550-09-1-0294, the National Science Foundation’s MRSEC programme (DMR0520513) at the Materials Research Center of Northwestern University, the Defense Advanced Research Projects Agency (DARPA)/Microsystems Technology Office (MTO) under Award Nos HR0011-13-2-0002, HR0011-13-2-0018 and N66001-11-14189, and the Non-equilibrium Energy Research Center (NERC), an Energy Frontier Research Center funded by the US DOE, Office of Science, Office of Basic Energy Sciences Award DE-SC0000989 (nanoparticle synthesis). Any opinions, findings, and conclusion or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the views of the funding agencies, and no official endorsement should be inferred. R.J.M.

References

acknowledges a Ryan Fellowship from Northwestern University. K.L.Y. and E.A. acknowledge National Defense Science and Engineering Graduate Research Fellowships. C.H.J.C. acknowledges a postdoctoral research fellowship from the Croucher Foundation. L.H. acknowledges the HHMI for support from an international student research fellowship. SAXS experiments were carried out at the Dupont–Northwestern–Dow Collaborative Access Team beam line at the Advanced Photon Source (APS), Argonne National Laboratory, and use of the APS was supported by the DOE (DEAC02-06CH11357). The electron microscopy work was performed at the Biological Imaging Facility (BIF) and the Electron Probe Instrumentation Center (EPIC) at Northwestern University

References

1. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607–609. 2. Mirkin, C. A. (2010). Mater. Res. Soc. Bull., 35, 532–539. 3. Park, S. Y., et al. (2008). Nature, 451, 553–556.

4. Nykypanchuk, D., Maye, M. M., van der Lelie, D and Gang, O. (2008). Nature, 451, 549–552. 5. Macfarlane, R. J., et al. (2011). Science, 334, 204–208.

6. Maye, M. M., Kumara, M. T., Nykypanchuk, D., Sherman, W. B. and Gang, O. (2010). Nat. Nanotechnol., 5, 116–120. 7. Jones, M. R., et al. (2010). Nat. Mater., 9, 913–917.

8. Agard, N. J., Prescher, J. A. and Bertozzi, C. R. (2004). J. Am. Chem. Soc., 126, 15046–15047.

9. Alivisatos, A. P. (1996). Science, 271, 933–937.

10. Cutler, J. I., Auyeung, E. and Mirkin, C. A. (2012). J. Am. Chem. Soc., 134, 1376–1391.

11. Daniel, M. C. and Astruc, D. (2004). Chem. Rev., 104, 293–346.

12. Jeong, U., Teng, X., Wang, Y., Yang, H. and Xia, Y. (2007). Adv. Mater., 19, 33–60. 13. Sun, S. H. (2006). Adv. Mater., 18, 393–403.

14. Murray, C. B., Kagan, C. R. and Bawendi, M. G. (2000). Annu. Rev. Mater. Sci., 30, 545–610.

15. Talapin, D. V., Lee, J. S., Kovalenko, M. V. and Shevchenko, E. V. (2010). Chem. Rev., 110, 389–458.

599

600

A General Approach to DNA-Programmable Atom Equivalents

16. Macfarlane, R. J., et al. (2010). Angew. Chem. Int. Ed., 49, 4589–4592.

17. Cutler, J. I., Zheng, D., Xu, X. Y., Giljohann, D. A. and Mirkin, C. A. (2010). Nano Lett., 10, 1477–1480. 18. Sun, D. Z. and Gang, O. (2011). J. Am. Chem. Soc., 133, 5252–5254. 19. Tikhomirov, G., et al. (2011). Nat. Nanotechnol., 6, 485–490.

20. Gao, X. H., Cui, Y. Y., Levenson, R. M., Chung, L. W. K. and Nie, S. M. (2004). Nat. Biotechnol., 22, 969–976 (2004). 21. Pellegrino, T., et al. (2004). Nano Lett., 4, 703–707.

22. Di Corato, R., et al. (2008). J. Mater. Chem., 18, 1991–1996.

23. Shtykova, E. V., et al. (2008). J. Phys. Chem. C, 112, 16809–16817.

24. Lin, C. A. J., et al. (2008). Small, 4, 334–341.

25. Yu, W. W., et al. (2007). J. Am. Chem. Soc., 129, 2871–2879.

26. Jin, R., Wu, G., Li, Z., Mirkin, C. A. and Schatz, G. C. (2003). J. Am. Chem. Soc., 125, 1643–1654.

27. Macfarlane, R. J., et al. (2009). Proc. Natl. Acad. Sci. U.S.A., 106, 10493– 10498. 28. Elghanian, R., Storhoff, J. J., Mucic, R. C., Letsinger, R. L. and Mirkin, C. A. (1997). Science, 277, 1078–1081.

29. Hurst, S. J., Lytton-Jean, A. K. R. and Mirkin, C. A. (2006). Anal. Chem., 78, 8313–8318. 30. Hill, H. D., et al. (2008). Nano Lett., 8, 2341–2344.

31. Auyeung, E., Macfarlane, R. J., Choi, C. H. J., Cutler, J. I. and Mirkin, C. A. (2012). Adv. Mater., 24, 5181–5186.

Chapter 27

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks*

Matthew R. Jones,a,b Robert J. Macfarlane,b,c Byeongdu Lee,d Jian Zhang,b,c Kaylie L. Young,b,c Andrew J. Senesi,b,c and Chad A. Mirkina,b,c aDepartment of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA dX-ray Science Division, Advanced Photon Source, Argonne National Laboratory, 9700 S. Cass Ave., Argonne, IL 60439, USA [email protected]

Directional bonding interactions in solid-state atomic lattices dictate the unique symmetries of atomic crystals, resulting in a diverse and complex assortment of three-dimensional structures that exhibit a wide variety of material properties. Methods to create analogous *Reprinted with permission from Jones, M. R., Macfarlane, R. J., Lee, B., Zhang, J., Young, K. L., Senesi, A. J. and Mirkin, C. A. (2010). DNA-nanoparticle superlattices formed from anisotropic building blocks, Nat. Mater. 9, 913–917. Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

602

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

nanoparticle superlattices are beginning to be realized [1–5], but the concept of anisotropy is still largely underdeveloped in most particle assembly schemes [6]. Some examples provide interesting methods to take advantage of anisotropic effects [7–11], but most are able to make only small clusters or lattices that are limited in crystallinity and especially in lattice parameter programmability [12–17]. Anisotropic nanoparticles can be used to impart directional bonding interactions on the nanoscale [6, 18], both through faceselective functionalization of the particle with recognition elements to introduce the concept of valency [19–21], and through anisotropic interactions resulting from particle shape [13, 22]. In this work, we examine the concept of inherent shape-directed crystallization in the context of DNA-mediated nanoparticle assembly. Importantly, we show how the anisotropy of these particles can be used to synthesize one-, two- and three-dimensional structures that cannot be made through the assembly of spherical particles. Particle assembly is a rapidly developing field of research, because the properties of superlattices can be as different from their individual components as the physical properties of nanoparticles are from bulk materials [6, 14, 18]. Shape is an important parameter that affects the properties of a particle, and therefore the incorporation of anisotropic nanostructures into colloidal crystals should lead to materials with as yet undiscovered collective phenomena [6]. Maximum utility of these structures requires that a technique be developed to rationally integrate particles with nonspherical shapes into ordered assemblies. The use of DNA as a ligand for the three-dimensional (3D) crystallization of nanoparticles into colloidal superlattices has many advantages over other assembly techniques [1–4]. In particular, the synthetically programmable length and recognition properties of DNA have enabled researchers to generate well-defined superlattices from spherical nanoparticles that vary in size (5–80 nm; Ref. [23]). Moreover, the technique provides nanometre-scale precision in programming the resulting lattice parameters. By replacing the spherical cores that associate through isotropic hybridization interactions with anisotropic nanostructures, we hypothesized that directional bonding interactions could be facilitated by virtue of different particle shapes (Fig. 27.1a).

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks a

b

20

30 40 50 60 Temperature (°C)

20

30 40 50 60 Temperature (°C)

Extinction (a.u.)

Extinction (a.u.)

Extinction (a.u.)

Extinction (a.u.)

c

20

30 40 50 60 Temperature (°C)

20

30 40 50 60 Temperature (°C)

d

S

S n

n

Figure 27.1  Directional bonding interactions can be imparted to monodisperse, DNA-functionalized gold nanostructures through the introduction of shape anisotropy. (a) The curved surfaces of spherical particles (left) cannot support the same number of idealized oligonucleotide interactions without DNA deformation as the flat, faceted surfaces of anisotropic nanostructures (middle). This feature allows for the shape of a nanostructure to more strongly dictate the structural details of the assembled superlattice it composes (right). (b) Transmission electron microscopy images of (from left to right) rods, triangular prisms, rhombic dodecahedra and octahedra. The scale bars represent 50 nm. (c) Extinction monitored at surface plasmon resonance maximum (from left to right: rods—800 nm, prisms—1,200 nm, rhombic dodecahedra—618 nm, octahedra—550 nm) as a function of temperature for DNA-functionalized anisotropic nanostructures assembled with linker oligonucleotides. The sharp melting transitions are indicative of a dense surface coverage of DNA. (d) Schematic of the oligonucleotides used to assemble anisotropic nanostructures. Thiolated DNA strands (gray) anchored to the particle’s surface were hybridized to linker DNA (black), which contained modular blocks of a repeated 40-basepair sequence (labelled “n,” where n refers to the number of 40-base segments) and a short self-complementary (GCGC) recognition sequence (red) that induced particle assembly.

603

604

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

Dense DNA functionalization of anisotropic nanostructures (previously demonstrated only for a single anisotropic nanostructure [21]) was achieved through a combination of particle purification, overgrowth or appropriate surfactant mixtures. These procedures allow one to prepare monodisperse (95%) solutions of DNA-functionalized triangular nanoprisms [24], nanorods [25], rhombic dodecahedra [26] and octahedral [26] (Fig. 27.1b). Nanoparticle assembly occurred through the hybridization of DNA linkers (containing 23, 64, 105, 146 and 187 nucleobases) to the oligonucleotides anchored to the particles (Fig. 27.1d). On binding to the nanoparticle, these linkers present multiple “sticky ends” at a programmable distance from the nanoparticle surface, creating, in essence, a controllably sized “DNA shell” that directs the crystallization process [23]. Cooperative melting transitions (where “melting” refers to the dehybridization of DNA bases linking particles) have been measured for crystals formed in solution for each of these particle shapes, indicating a dense surface coverage of oligonucleotides [27], which is essential for the assembly and subsequent crystallization process [23] (Fig. 27.1c). Small-angle X-ray scattering (SAXS) was used to interrogate the colloidal crystal structures synthesized from anisotropic particles. SAXS represents a powerful characterization tool for this new class of solution-phase-assembled nanostructures, and can give the symmetries, lattice constants, particle orientation and domain size of a 3D ordered structure without requiring drying of the sample (see Table 27.1), which significantly affects the resulting lattices [23]. Modelling of nanoparticle superlattices has been carried out to corroborate these results by comparing modelled SAXS patterns to those obtained experimentally. Additionally, transmission electron micrographs of several ordered nanoparticle crystals, embedded in a resin, support the conclusions drawn from the SAXS data. The first systems examined were assemblies created from primarily 1D structures (nanorods), where a “1D” particle is defined as having a length significantly greater than its width or depth. As has been previously shown with spherical particles, the most stable crystal structure for a given system is typically the one with the largest number of DNA linker interactions [3]; one would therefore predict that these rods would preferentially assemble with their

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

long axes parallel to each other, to maximize the hybridization events between DNA sticky ends. It is important to note that there is a difference between the processes of particle “crystallization” and particle “assembly.” In this context, particle crystallization refers to positioning particles in an ordered formation that has translational symmetry, whereas assembly refers to the process of DNAhybridization-induced association, regardless of the structure of the aggregate. It has previously been demonstrated that the process of particle assembly occurs on a significantly faster timescale than the process of particle crystallization [28]. Therefore, in all cases presented here, one can visually observe the initial formation of large, disordered aggregates and differentiate the subsequent ordering events using SAXS. Table 27.1 Summary of crystallization parameters for DNA-functionalized anisotropic nanoparticle superlattices Average crystal domain sizes (nm)

Nanoparticle dimensions (nm)

Lattice parameters (nm)

Rods

14 (w) ¥ 55 (l)

34.3, 55.8, 77.9, 1,270 ± 310 97.9, 110

330 ± 210

Triangular prisms

60 (e) ¥ 7 (w)

26.2, 48.3, 71.8, 750 ± 230 94.3, 113

14 ± 4

19 (w) ¥ 52 (l)

34.3, 57.7, 77.6, 1,180 ± 650 93.9, 113

Average no. unit cells/ crystal

220 ± 120

95 (e) ¥ 7 (w)

23.6, 48.6, 72.1, 980 ± 400 94.9, 114

18 ± 6

Rhombic

39 (e), 64 (d)

181, 201

1,160 ± 810

Octahedra

59 (e), 83 (d)

110, 118 (bcc); 1,510 ± 170 135, 159 (fcc)

140 (e) ¥ 7 (w) 24.4, 48.9, 72.3, 1,040 ± 510 93.3, 114

dodecahedra 50 (e), 81 (d)

198, 220

2,550 ± 990 1,780 ± 490

19 ± 6

420 ± 120 910 ± 470

Gold nanorods were assembled using DNA linkers of varying lengths; on annealing, the resulting superlattices possessed long-

605

606

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

range hexagonal symmetry (Fig. 27.2a,b, Table 27.1). Most crystals exhibited scattering peaks corresponding only to 2D ordering. However, extended thermal annealing of some samples resulted in ordering between 2D sheets, with the resulting peaks indexing to a P63/mmc, hexagonal-close-packed, lattice. As nanorods were observed to order into 2D sheets in most samples studied, one can conclude that the particles favour interactions perpendicular to their long axis, demonstrating that the “1D” shape is directing their crystallization into primarily a 2D lattice, and these 2D lattices can subsequently reorganize into an ordered 3D structure. This was further probed through in situ monitoring of the crystallization process, wherein peaks corresponding to 2D order appear and begin to sharpen (indicating growth in crystal domain size) significantly before any peaks corresponding to 3D order appear. The assembly of triangular nanoprisms was then investigated as an example of a “2D” nanostructure, wherein the length and width of a prism are an order of magnitude greater than its depth. One would expect the most dominant crystallization force to be faceto-face interactions between the predominantly 2D nanoprisms. Scattering patterns from superstructures of assembled prisms indicate a lamellar (or columnar) 1D arrangement of particles stacked in a face-to-face configuration (Fig. 27.2c,d). This is consistent with the previous observation (see above) that particles associate in a manner that maximizes hybridization interactions. Unlike the primarily 2D hexagonally packed nanorods, no longrange DNA-mediated ordering was observed between 1D stacks of prisms. Some crystals that underwent significant thermal annealing exhibited short-range order between nanoprism superlattices, but these scattering patterns could not be attributed to a welldefined 3D lattice. This lack of 3D ordering can be explained by the inherent thinness of the prisms (7 nm) and the relative rigidity of double-stranded DNA (persistence length of ~50 nm [29]), resulting in very low DNA density along the side of a column. As the long axis of a 1D stack of prisms contains a relatively diffuse coating of DNA sticky ends, we project that hierarchical crystallization of these columns is not favourable enough to produce 3D structures in the timescales monitored herein.

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks c

e

b

d

f

0.01 0.02 0.03 0.04 0.05 0.06 q (1/A)

S (q)

S (q)

S (q)

a

0

0.05

0.10 q (1/A)

0.15

0.20

0.01

0.02

0.03

0.04

0.05

q (1/A)

Figure 27.2  SAXS characterization of anisotropic nanoparticle colloidal crystals. (a) Schematic of a hexagonal 2D layer (additional layers omitted for clarity) in assemblies of gold nanorods. (b) Experimental (blue) and simulated (red) SAXS patterns for gold nanorods (55 nm length, 14 nm diameter) assembled into a hexagonal-close-packed lattice with lattice constants of a = 76.0 nm and c = 176.5 nm. (c) Schematic of the primarily 1D lamellar assemblies of gold triangular nanoprisms. (d) Experimental (blue) and simulated (red) SAXS patterns for nanoprisms (95 nm edge length, 7 nm thickness) assembled into a columnar arrangement with a lattice spacing of 72.1 nm. (e) Schematic of the 3D fcc assemblies of gold rhombic dodecahedra. The lines denote the fcc unit cell, not interparticle interactions. (f) Experimental (blue) and simulated (red) SAXS patterns for rhombic dodecahedra (64 nm diameter) assembled into an fcc arrangement with a lattice constant of 201.4 nm.

A high degree of precision over the placement of prisms in 1D chains can be seen qualitatively by the large number of diffraction peaks present in the scattering pattern. Quantification of the rise per base pair (particle face-to-face distance divided by number of DNA bases) for three different particle sizes (edge lengths = 60, 95, 140 nm) and three different linker lengths (DNA lengths = 40.5, 63.5, 86.5 nm) yields a value of 0.281 ± 0.002 nm. This is a remarkable level of precision and programmability over the placement of nanomaterials along one dimension that would be difficult, if not impossible, to replicate by any lithographic or other directed-assembly method, illustrating one of the primary advantages of the DNA-directed crystallization methodology. In accordance with the previous definitions of particle dimensionality (see above), a “3D” object would have no physical

607

608

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

dimension significantly larger or smaller than the other two. The rhombic dodecahedron is an ideal particle to investigate the role of shape in 3D nanoparticle crystallization, as it naturally forms a facecentred-cubic (fcc) lattice with 100% packing efficiency, contrasted with previous fcc lattices of spherical particles [3, 23], which exhibit a packing density of only 74%. Unlike spheres, which interact in an isotropic fashion along a curved surface, maximum DNA interactions for the rhombic dodecahedron system would be obtained only when the particles associate face-to-face and are ordered in an fcc lattice with both positional and orientational order. Comparing the crystallization of rhombic dodecahedra to similarly sized spheres would give a good indication of how shape anisotropy affects positional and rotational order and crystal quality in the resulting colloidal superlattices. The SAXS data confirm that the rhombic dodecahedra crystallize into fcc lattices and retain the lattice parameter programmability imparted by variations in DNA linker length (Fig. 27.2e,f). Importantly, the lattices exhibit a significantly larger number of scattering peaks than fcc crystals consisting of similarly sized spherical particles [23], indicating greater positional order with respect to an ideal fcc lattice. Relative intensities of diffraction peaks correlate with modelled SAXS patterns for lattices wherein the rhombic dodecahedra retain orientational order. This ordering occurs even when the hydrodynamic size of the “DNA shell” is significantly larger than that of the particle, confirming that the rhombic dodecahedron shape does indeed have a strong influence on colloidal crystal formation. From these data, one can conclude that the incorporation of shape anisotropy provides significant benefit to packing precision over the assembly of isotropic spheres. It is important to note that the DNA linkers directing the crystallization process are not completely rigid. With increasing DNA length, one would expect the increasing flexibility of the DNA strands to change the sphericity of the DNA shell. Rhombic dodecahedra crystallized in fcc lattices retain their orientational order, indicating that the shape of their DNA shell is similar over all DNA lengths examined. Octahedra are another interesting class of particle worth probing in this manner. Experimentally and theoretically, octahedra pack most densely in lattices that do not maximize commensurate face-to-face interactions [30–32].

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

Therefore, octahedral nanoparticles allow one to test the hypothesis that DNA-driven particle assembly and crystallization favours the structure that maximizes hybridization events and, in the case of anisotropic nanostructures, face-to-face interactions. a

b

d

S (q)

S (q)

S (q)

c

0.01 0.02 0.03 0.04 0.05 0.06 q (1/A)

0.01

0.02 0.03 q (1/A)

0.04

0.005 0.010

0.015 0.020 0.025 q (1/A)

Figure 27.3  Nanoparticle colloidal crystals undergo phase transformations as a function of DNA length. (a) Schematic of disordered, bcc and fcc phases of crystallized octahedral nanoparticles, which are stable at short, intermediate and long DNA lengths, respectively. (b) Experimental (blue) SAXS characterization of gold octahedra (59 nm edge length) assembled with short DNA (~15 nm) resulting in an interparticle distance of 76.8 nm. Simulated SAXS patterns for bcc (red) and fcc (gray) crystals with lattice parameters that would be expected given the length of the linking DNA strands demonstrate that the particles do not correlate with either lattice. (c) Experimental (blue) and simulated (red) SAXS patterns for octahedra assembled with DNA linkers of an intermediate length (~45 nm) resulting in face-to-face orientational ordering within a bcc crystal with a lattice parameter of 118.4 nm. (d) Experimental (blue) and simulated (gray) SAXS patterns for octahedra assembled with long DNA (~90 nm) resulting in no orientational ordering within an fcc crystal with a lattice parameter of 194.7 nm.

SAXS data demonstrate that the octahedra crystallize into a disordered lattice, a body-centred-cubic (bcc) lattice and an fcc lattice, with short, intermediate and long DNA lengths, respectively (Fig. 27.3a). With short DNA, only short-range order is observed the corresponding scattering pattern does not index with either a bcc or fcc lattice, the two structures observed for longer DNA lengths (Fig. 27.3b). Although several possible lattices have been

609

610

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

proposed for the dense packing of octahedra without DNA [30– 32], with short, inflexible DNA, none of these would produce a stable structure, according to the hypothesis, because they all exhibit severely limited face-to-face overlap. At a DNA length of ~45 nm, however, the scattering peaks index to a bcc lattice; the intensities of these peaks are best fitted by a model that includes face-to-face orientational ordering between octahedra (Fig. 27.3c). As the DNA is more flexible, the shape of the DNA shell may appear more like a truncated octahedron, which is able to maximize face-to-face interactions in a bcc arrangement. At a DNA length of ~90 nm, the scattering pattern indexes to an fcc lattice, indicative of a particle that is more similar to a sphere than a welldefined octahedron (Fig. 27.3d). In this case, the intensities of the peaks indicate no orientational alignment of particles, as would be expected of a DNA shell that is approximately spherical. These data indicate that, at all DNA lengths, shape has a significant impact on the most stable structure, but, depending on DNA flexibility, the particle and the DNA have differing levels of importance in determining the anisotropy of the interactions between DNA-functionalized particles. Therefore, the interplay between the two components of this bionanoconjugate crystallization methodology provides a means to control the phase behavior of a colloidal superlattice constructed from a given set of anisotropic nanostructure building blocks. We have demonstrated that particle shape has a strong influence on the crystallization parameters of DNA-functionalized nanoparticles, affecting superlattice dimensionality, crystallographic symmetry and phase behavior. Furthermore, the use of DNA as a programmable linker imparts the ability to tune the lattice parameters of the resulting crystals while retaining the shape directing effects of the nanoparticles within the limits of DNA flexibility. Moreover, this work is consistent with the conclusion that nanoparticle superlattices that can maximize interparticle DNA hybridization events will be the most stable structures. Such crystals may find use in applications that take advantage of the ability to tune the unique physical properties of these structures, such as plasmonicbased circuitry or waveguides, photonic bandgap materials and energy harvesting or storage materials, all of which exhibit unique emergent properties that are dependent on interparticle distance and crystal symmetry. In particular, the precision with which we can

References

position particles is difficult to replicate using other assembly or lithographic techniques, indicating that this methodology provides a powerful means to realize these types of designer materials. Furthermore, we project that these results will enable fundamental insights into shape-directed hybridization effects and the influence of nanostructure valency on crystallographic parameters.

Acknowledgments

C.A.M. acknowledges the NSF-NSEC and the AFOSR for grant support, and the DOE Office (Award No. DE-SC0000989) for support through the NU Nonequilibrium Energy Research Center. He is also grateful for an NSSEF Fellowship from the DoD. M.R.J. acknowledges Northwestern University for a Ryan Fellowship and the NSF for a Graduate Research Fellowship. R.J.M. acknowledges Northwestern University for a Ryan Fellowship. K.L.Y. acknowledges the NSF and the NDSEG for Graduate Research Fellowships. Portions of this work were carried out at the DuPont-Northwestern-Dow Collaborative Access Team (DND-CAT) beamline located at Sector 5 of the Advanced Photon Source (APS). DND-CAT is supported by E.I. DuPont de Nemours & Co., The Dow Chemical Company and the State of Illinois. Use of the APS was supported by US Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DEAC02-06CH11357. The transmission electron microscope work was carried out in the EPIC facility of NUANCE Center at Northwestern University. NUANCE Center is supported by NSF-NSEC, NSF-MRSEC, Keck Foundation, the State of Illinois and Northwestern University. Ultrathin sectioning was carried out at the Northwestern University Biological Imaging Facility supported by the NU Office for Research.

References

1. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607–609. 2. Park, S-J., Lazarides, A. A., Storhoff, J. J., Pesce, L. and Mirkin, C. A. (2004). J. Phys. Chem. B, 108, 12375–12380. 3. Park, S. Y., et al. (2008). Nature, 451, 553–556.

4. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451, 549–552.

611

612

DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks

5. Shevchenko, E. V., Talapin, D. V., Kotov, N. A., O’Brien, S. and Murray, C. B. (2006). Nature, 439, 55–59.

6. Glotzer, S. C. and Solomon, M. J. (2007). Nat. Mater., 6, 557–562.

7. Lapointe, C. P., Mason, T. G. and Smalyukh, I. I. (2009). Science, 326, 1083–1086.

8. Sacanna, S., Irvine, W. T. M., Chaikin, P. M. and Pine, D. J. (2010). Nature, 464, 575–578.

9. Zerrouki, D., Baudry, J., Pine, D., Chaikin, P. and Bibette, J. (2008). Nature, 455, 380–382. 10. Srivastava, S., et al. (2010). Science, 327, 1355–1359. 11. DeVries, G. A., et al. (2007). Science, 315, 358–361. 12. Liu, Q., et al. (2010). Nano Lett., 10, 1347–1353.

13. Ming, T., et al. (2008). Angew. Chem. Int. Ed., 120, 9831–9836.

14. Tao, A., Sinsermsuksakul, P. and Yang, P. (2007). Nat. Nanotechnol., 2, 435–440. 15. Alivisatos, A. P., et al. (1996). Nature, 382, 609–611.

16. Wei, Y., Bishop, Kyle J. M., Kim, J., Soh, S. and Grzybowski, B. A. (2009). Angew. Chem. Int. Ed., 121, 9641–9644.

17. Li, M., Schnablegger, H. and Mann, S. (1999). Nature, 402, 393–395.

18. Stebe, K. J., Lewandowski, E. and Ghosh, M. (2009). Science, 325, 159– 160. 19. Nie, Z., et al. (2007). Nat. Mater., 6, 609–614.

20. Li, F., Yoo, W. C., Beernink, M. B. and Stein, A. (2009). J. Am. Chem. Soc., 131, 18548–18555. 21. Millstone, J. E., et al. (2008). Small, 4, 2176–2180.

22. Zhang, Z. and Glotzer, S. C. (2004). Nano Lett., 4, 1407–1413.

23. Macfarlane, R. J., et al. (2010). Angew. Chem. Int. Ed., 49, 4589–4592.

24. Millstone, J. E., et al. (2005). J. Am. Chem. Soc., 127, 5312–5313.

25. Nikoobakht, B. and El-Sayed, M. A. (2003). Chem. Mater., 15, 1957– 1962. 26. Niu, W., et al. (2008). J. Am. Chem. Soc., 131, 697–703.

27. Jin, R., Wu, G., Li, Z., Mirkin, C. A. and Schatz, G. C. (2003). J. Am. Chem. Soc., 125, 1643–1654.

References

28. Macfarlane, R. J., et al. (2009). Proc. Natl. Acad. Sci. U.S.A., 106, 10493– 10498.

29. Rivetti, C., Walker, C. and Bustamante, C. (1998). J. Mol. Biol., 280, 41– 59. 30. Lu, W., et al. (2008). J. Am. Chem. Soc., 130, 6983–6991.

31. Xie, S., et al. (2009). J. Phys. Chem. C, 113, 19107–19111. 32. Torquato, S. and Jiao, Y. (2009). Nature, 460, 876–879.

613

Chapter 28

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization*

Matthew N. O’Brien,a,b Matthew R. Jones,b,c Byeongdu Lee,d and Chad A. Mirkina,b,c aDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cDepartment of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA dX-ray Science Division, Advanced Photon Source, Argonne National Laboratory, 9700 S Cass Ave, Argonne, IL 60439, USA [email protected], [email protected]

Whether two species will co-crystallize depends on the chemical, physical and structural complementarity of the interacting components. Here, by using DNA as a surface ligand, we selectively *Reprinted with permission from O’Brien, M. N., Jones, M. R., Lee, B. and Mirkin, C. A. (2015). Anisotropic nanoparticle complementarity in DNA-mediated cocrystallization, Nat. Mater. 1–8.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

616

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

co-crystallize mixtures of two different anisotropic nanoparticles and systematically investigate the effects of nanoparticle size and shape complementarity on the resultant crystal symmetry, microstrain, and effective “DNA bond” length and strength. We then use these results to understand a more complicated system where both size and shape complementarity change, and where one nanoparticle can participate in multiple types of directional interactions. Our findings offer improved control of nonspherical nanoparticles as building blocks for the assembly of sophisticated macroscopic materials, and provide a framework to understand complementarity and directional interactions in DNA-mediated nanoparticle crystallization. The rational co-crystallization of two species (that is, the formation of a crystalline structure composed of two components that each possess well-defined symmetry and stoichiometric relationships in a unit cell) represents a significant challenge in chemistry, biology, pharmacology and materials science. This challenge arises because it is often more favourable for a multicomponent system to phase-separate or form an alloy—neither of which possesses a regular arrangement of both species within a crystalline lattice. Accordingly, successful co-crystallization can often only be achieved through a careful balance of the chemical, physical and/or structural complementarity between the two species. Whereas rational control over these forms of complementarity can be challenging for atomic and molecular systems, each can be tuned independently for nanoparticle systems [1–5]. Although nanoparticle co-crystallization has been demonstrated with spherical structures through a chemical complementarity of the surface-bound ligands (for example, electrostatics, DNA hybridization) [6–13], isotropic particles offer a limited set of building blocks to construct and study new materials. If instead, anisotropic nanoparticles are used, an additional structural complementarity must be satisfied. However, this further narrows the range of conditions over which cocrystals form [14–24]. As a result, examples in the literature based primarily on structural complementarity require extraordinary synthetic precision in nanoparticle size and shape to prevent phase segregation [24–29]. Consequently, our ability to understand and

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

control the co-crystallization of two nanoparticle building blocks with different sizes and shapes has heretofore been limited to a small number of “lock and key”-type systems. In this work, we use the sequence programmability of surface-bound DNA ligands to force the co-crystallization of a range of different anisotropic nanoparticles and systematically investigate how structural complementarity impacts the formation of co-crystalline materials. Specifically, we examine the effects of nanoparticle size and shape complementarity on crystal symmetry, crystallinity, effective “DNA bond” length and effective “DNA bond” strength, with nanoparticles capable of forming a single type of directional interaction. Building on this understanding, we then investigate the phase behavior of a system which includes particles of lower symmetry that are able to participate in two types of directional interactions—one that favours co-crystal formation and the other that does not. This work greatly advances our ability to understand and control nanoparticle co-crystallization, and thus offers a guide for the future assembly of sophisticated and hierarchical materials. Nanoparticles functionalized with a densely packed shell of double-stranded oligonucleotides can be analogized to “programmable atom equivalents” (PAEs), where the particle core represents the “atom” and the DNA shell represents the “bonds” to connect particles through complementary hybridization events [3, 30, 31]. With this design, the DNA sequence can be used to program both chemical complementarity between two species and DNA bond length, or the distance between nanoparticles, with subnanometer resolution based on the number of DNA base pairs in each duplex [11, 15, 32]. The nanoparticle acts as a scaffold to arrange the DNA in a surface-normal orientation, causing it to form a conformal shell that mimics the shape of the underlying structure. This allows the shape of the nanoparticle to dictate the symmetry of the DNA interactions in a superlattice [15]. For example, a cube-shaped nanoparticle will present DNA oriented perpendicular to each of its six facets and is thus capable of forming six directional interactions with neighboring nanoparticles, each consisting of many individual DNA hybridization events (Fig. 28.1a,b).

617

618

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization a

b

=

c

d

Size complementarity

Shape complementarity

Figure 28.1  The degree of size and shape complementarity between two collections of DNA-modified anisotropic nanoparticles dictate to what extent the hybridization of surface DNA ligands can drive co-crystallization. (a) Nanoparticle shape can be used to template the arrangement of surface-bound DNA ligands. The resultant shell of DNA (light green) interacts with other nanoparticles in a directional fashion (red arrows). (b) Many complementary DNA hybridization events between two interacting nanoparticles mediate crystallization. (c) Size complementarity between two nanoparticles is controlled by the differences in the characteristic dimension of the nanoparticles. (d) Shape complementarity between two nanoparticles is controlled by the alignment of directional interactions in three-dimensional space.

Whether two collections of PAEs with chemically complementary DNA will co-crystallize into an ordered structure is governed by a more complicated interplay of several factors, including the size and shape complementarity between the individual nanoparticles (Fig. 28.1). In this work, we define size complementarity as the dimensional similarity between the interacting facets of two nanoparticles (Fig. 28.1c), and shape complementarity as how efficiently the interacting facets of two nanoparticles pack together in three-dimensional space, analogous to a “lock and key” concept (Fig. 28.1d). To systematically investigate these effects, which we collectively term structural complementarity, we took advantage of several important advances from our group: a DNA design and length previously shown to form directional interactions between anisotropic nanoparticles

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

in single-component crystallization [15]; a thermal annealing procedure that allows access to equilibrium crystal structures [33]; and highly uniform nanoparticles with tunable sizes and shapes [34, 35]. To characterize the effects of nanoparticle complementarity on the structural properties of the resultant co-crystal, we used small-angle X-ray scattering (SAXS) and compared the results with modelled SAXS patterns [36]. SAXS allows the direct measurement of crystal symmetry, lattice parameters, domain size and defect density of nanoparticle superlattices in solution. These results were corroborated with electron microscopy (EM), wherein crystals were encapsulated in silica using a procedure previously shown to accurately preserve the symmetry of PAE superlattices [37]. Together, SAXS and EM provide complementary characterization at the ensemble and local levels, respectively. To characterize the effects of nanoparticle complementarity on DNA bond strength, we measured the thermodynamic temperature at which nanoparticles transition from an assembled state to a discrete state via ultravioletvisible spectroscopy measurements of extinction. To examine the effect of size complementarity on crystallization, we assembled two collections of cube PAEs in a 1:1 ratio (analytically determined with optical extinction coefficients) [34] but systematically changed the difference in cube edge length (∆L; Fig. 28.2a,b). This was accomplished through the use of one cube with a fixed average edge length (L) of 47 nm and a systematic variation of L for the second cube from 47 to 85 nm—a maximum difference of 38 nm (or 81%). Despite large values of ∆L, we observed a consistent NaCl-type crystal structure (Fm3m space group) for all samples, as confirmed by EM and SAXS, wherein the cubes align in a face-to-face manner (Fig. 28.2c). However, if one evaluates the longrange order in these samples through analysis of SAXS peak breadth, crystallinity decreases as ∆L increases. To understand the origins of this decrease in crystallinity, we used a Williamson–Hall analysis that allows one to separate the contributions of finite crystalline domain size and microstrain to SAXS peak broadening [38–40]. In this analysis, domain size (DWH) affects integral peak breadth (β) uniformly throughout reciprocal space, whereas microstrain (ε) increases integral peak breadth non-uniformly in reciprocal space

619

620

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

owing to a loss of coherent scattering. From this analysis, we found that as ∆L increases from 0 to 38 nm, the average crystalline domain size decreases by >30% from ~4,500 to ~3,000 nm (Fig. 28.2d) and the root-mean-squared microstrain (εr.m.s) increases by >95% from ~5% to ~10% (Fig. 28.2e). Furthermore, (εr.m.s). does not seem to have a preferred crystallographic axis, as plots of β versus q that include reflections from different families of planes show a linear relationship. The observed increase in microstrain with ∆L is consistent with positional error propagation associated with the difference in nanoparticle size. At an individual nanoparticle level, this can be imagined as a small cube that can interact in an energetically similar face-to-face fashion with a large cube over a wide range of offset positions. As the distance between the centre of each set of nearest-neighbor cube faces increases, each of the subsequent next-nearest-neighbor interactions will likewise be displaced to a greater extent, and thus the positional error propagates throughout the structure. The greater the difference in nanoparticle size, the more pronounced this effect becomes, resulting in greater lattice microstrain. To investigate the role of the DNA in the co-crystallization process, we determined crystal lattice parameters from both experimental and modelled SAXS data, and used these values to calculate a surface-to-surface distance between cubes, that is, the effective DNA bond length (lDNA; Fig. 28.2f). This analysis shows that despite the remarkable difference in nanoparticle size and lattice microstrain, lDNA does not vary by more than 5% throughout the samples, which is within the error of the measurement. This consistent lDNA suggests that the majority of the DNA does not adopt a significantly strained conformation to accommodate differences in nanoparticle size and thus supports our interpretation of positional-error-propagationrelated microstrain. Furthermore, the consistent lDNA observed here shows excellent agreement with previous investigations of anisotropic nanoparticle crystallization in single-component systems [15, 17] and therefore suggests that lDNA is controlled more by how the nanoparticles are oriented with respect to each other (that is, in a face-to-face manner), rather than size complementarity.

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

e

d

36 nm

0.01 q (Å−1 ) 80

f

10

IDNA (nm)

ε r.m.s.(%)

9 8 7 6 5 0

10

20 30 ∆L (nm)

28 nm 22 nm 18 nm 11 nm 6 nm 3 nm 0 nm

40

Estimated crystallite size (µm)

∆L

5

Williamson−Hall Scherrer

4 3 2 1

0

10

20 30 ∆L (nm)

40

0

10

20 30 ∆L (nm)

40

40

g

70 0

38 Tm(°C)

100 nm

c

I(q)

b

∆ L = 36nm ∆ L = 18nm ∆ L = 0nm

a

60

50

36 6

0

10

20 30 ∆L (nm)

40

34 4

Figure 28.2  Two collections of cube PAEs with complementary DNA but different edge length were co-crystallized to investigate the effect of size complementarity. (a) Cube edge length was varied from 47 to 85 nm, as shown by TEM images, where each cube size has a variation in edge length L, we hypothesized one-dimensional (1D) structures would form of alternating cubes and disks, with the DNA on the remaining four cube faces rendered inaccessible by the presence of the disks on the other two facets (Fig. 28.4c). The results of these experiments are summarized in a phase diagram based on the relative size of each structure (D/L) and the DNA length, where we used correlated SAXS and EM to determine the position of the phase boundary (Fig. 28.4d). It should be noted that, to the best of our knowledge, these are the first demonstrations of these crystal structures regardless of nanoparticle composition or assembly method. For the shortest DNA lengths, we indeed observed the formation of 3D structures when D/L < 1, a mixture of 3D and 1D structures when D/L ≈ 1, and 1D structures when D/L > 1. However, as the DNA on the disks and cubes was increased in length, we observed that the region of D/L values over which 3D structures formed increased drastically at the expense of 1D structure formation. For example, at the largest DNA length, we observed 3D structures for D = 1.7L and a mixture of 1D and 3D structures for D/L > 2, despite the drastic difference in size complementarity. This result is analogous to the size complementarity study described above, wherein the DNA was able to accommodate large differences

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

in ∆L. To further understand this, we determined lDNA from the comparison of experimental and simulated SAXS spectra, and observed a similar value for all structures with the same number of DNA bases. This result is again consistent with the observation from the size complementarity study described above and suggests that these findings are not exclusive to the cases explored here. Phase behavior was also investigated for the case of octahedra assembled with disks, wherein the octahedra can participate in eight directional interactions (Fig. 28.4e,f). This results in a transition from a 3D body-centred cubic structure of octahedra with interpenetrating disks at D/L values 1, due to a disruption in structure from the secondary (edge-based) disk interaction. To further investigate the role of this secondary disk interaction, we maintained a consistent DNA length on the octahedra, but varied the DNA length on the disk. We hypothesized that the DNA length could be used to control the relative influence of the face and edge interactions of the disk and, as a result influence co-crystallization. Indeed, as the DNA length on the disk was decreased, the effect of the secondary interaction was reduced and a greater phase space of 3D assembly was accessed, consistent with the observations above. In summary, we have elucidated how nanoparticle shape and size complementarity can be used to control the structural properties of anisotropic nanoparticle co-crystals. This work adds to the growing body of knowledge showing the power of DNA to make complex, ordered materials [3, 31, 44–46] and provides a framework to understand and predict directional interactions between nanoparticles. It is likely that these findings can be extended to the co-crystallization of nanoparticles with a range of different compositions, shapes and sizes. As such, these results represent a significant advance towards the utilization of the full library of nanoparticle shapes as building blocks to synthesize crystalline materials, wherein material properties can be designed at multiple levels—from the individual nanoparticle to the macroscopic crystal. In particular, the ability to control nanostructure registry, symmetry and separation distance hold important consequences for tuning the optoelectronic and mechanical properties of nanoparticle-based materials.

627

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization a

b e

L

c

100 nm

100 nm

D

d

Disordered 3D 3D/1D 1D

2.0

1.5

f

1.5

D/ L

D/L

628

Disordered 3D

1.0

1.0

0.5

0.5 43

127 211 DNA length (# DNA base pairs)

0.2

1.0 4.8 DNADisk/DNA Octahedron

Figure 28.4  The number and type of directional interactions each nanoparticle can participate in is related to nanoparticle shape. (a) A cube has one type of directional interaction repeated six times around the structure. A disk has two different types of interactions: one from the circular facets on top and bottom and the other from the sides. When these two nanoparticles are co-crystallized, the resultant structure will depend on the relative size of the disk diameter (D) and the cube edge length (L), as well as the DNA length. (b) Model and TEM image show a 3D structure formed with cubes that pack in a face-to-face oriented simple cubic lattice with interpenetrating disks. (c) Model and TEM image show a 1D, biperiodic lamellar structure with alternating cubes and disks. (d) SAXS and TEM analysis were used to determine a phase diagram for cube–disk co-crystals as a function of the number of duplex DNA base pairs per strand connecting nanoparticles and the relative size ratio of the disk diameter and cube edge length. (e) Model and TEM image show a 3D structure formed with octahedra that pack into a face-to-face oriented body-centred cubic lattice with interpenetrating disks. (f) SAXS and TEM analysis were used to determine a phase diagram for octahedron-disk co-crystals as a function of the relative DNA length on the disks and octahedra.

28.1 Methods Nanoparticles were synthesized according to published literature protocols [34, 35] and characterized by transmission

Methods

electron microscopy (TEM; Hitachi H8100). Oligonucleotides were synthesized on a solid-support MM48 synthesizer with reagents purchased from Glen Research. Nanoparticles were functionalized according to published literature protocols [15, 21]. Nanoparticle concentrations were determined using published extinction coefficient values [34, 35, 47] and Cary 5000 UV–vis spectrophotometer measurements. Crystal melting temperatures were determined with measurement of extinction as a function of temperature using a Cary 5000 spectrophotometer at a rate of 0.1°C/10 min. Nanoparticle crystallization was performed by slowly cooling nanoparticle mixtures through their crystallization temperature at a rate of 0.1°C/10 min using a Life Technologies temperature cycler. Crystals were characterized by synchrotron SAXS experiments conducted at the Advanced Photon Source at Argonne National Laboratory. Crystals were transferred to the solid state via silica embedding [37] for visualization by scanning transmission electron microscopy (STEM; Hitachi HD-2300). Silica-encapsulated superlattices were then embedded in EMBed-812 resin (Electron Microscopy Sciences) and microtomed to thicknesses between ~80 and ~200 nm.

Acknowledgments

C.A.M. acknowledges support from the following awards: the Air Force Office of Scientific Research (AFOSR) Multidisciplinary University Research Initiative (MURI) FA9550-11-1-0275, the Department of Defense National Security Science and Engineering Faculty Fellowship (NSSEFF) award N00014-15-1-0043, the National Science Foundation (NSF) Materials Research Science and Engineering Center program DMR-1121262 at the Materials Research Center of Northwestern University, and the Non-equilibrium Energy Research Center (NERC) an Energy Frontier Research Center funded by the Department of Energy (DoE), Office of Science, and Office of Basic Energy Sciences under Award DE-SC0000989. M.N.O. and M.R.J. are grateful to the NSF for Graduate Research Fellowships. SAXS experiments were carried out at the Dupont Northwestern Dow Collaborative Access Team beamline at the Advanced Photon Source (APS) at Argonne National Laboratory, and use of the APS was supported by the DoE (DE-AC02-06CH11357). This work made

629

630

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

use of the EPIC facility (NUANCE Center-Northwestern University), which has received support from the MRSEC programme (NSF DMR-1121262) at the Materials Research Center, and the Nanoscale Science and Engineering Center (EEC-0118025/003), both programmes of the National Science Foundation, the State of Illinois and Northwestern University. We thank K. A. Brown and A. J. Senesi for helpful discussions.

References

1. Bishop, K. J. M., Wilmer, C. E., Soh, S. and Grzybowski, B. A. (2009). Small, 5, 1600–1630. 2. Jones, M. R., Osberg, K. D., Macfarlane, R. J., Langille, M. R. and Mirkin, C. A. (2011). Chem. Rev., 111, 3736–3827.

3. Macfarlane, R. J., O’Brien, M. N., Petrosko, S. H. and Mirkin, C. A. (2013). Angew. Chem. Int. Ed., 52, 5688–5698.

4. Min, Y., Akbulut, M., Kristiansen, K., Golan, Y. and Israelachvili, J. (2008). Nat. Mater., 7, 527–538.

5. Grzelczak, M., Vermant, J., Furst, E. M. and Liz-Marzán, L. M. (2010). ACS Nano, 4, 3591–3605.

6. Shevchenko, E. V., Talapin, D. V., Kotov, N. A., O’Brien, S. and Murray, C. B. (2006). Nature, 439, 55–59.

7. Park, S. Y., et al. (2008). Nature, 451, 553–556.

8. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451, 549–552. 9. Talapin, D. V., et al. (2009). Nature, 461, 964–967.

10. Dong, A. G., Chen, J., Vora, P. M., Kikkawa, J. M. and Murray, C. B. (2010). Nature, 466, 474–477. 11. Macfarlane, R. J., et al. (2011). Science, 334, 204–208.

12. Li, T. I. N. G., Sknepnek, R., Macfarlane, R. J., Mirkin, C. A. and Olvera de la Cruz, M. (2012). Nano Lett., 12, 2509–2514. 13. Macfarlane, R. J., Jones, M. R., Lee, B., Auyeung, E. and Mirkin, C. A. (2013). Science, 341, 1222–1225. 14. Glotzer, S. C. and Solomon, M. J. (2007). Nat. Mater., 6, 557–562. 15. Jones, M. R., et al. (2010). Nat. Mater., 9, 913–917.

16. Quan, Z. and Fang, J. (2010). Nano Today, 5, 390–411.

References

17. Jones, M. R., Macfarlane, R. J., Prigodich, A. E., Patel, P. C. and Mirkin, C. A. (2011). J. Am. Chem. Soc., 133, 18865–18869.

18. Damasceno, P. F., Engel, M. and Glotzer, S. C. (2012). Science, 337, 453– 457. 19. Walker, D. A., Leitsch, E. K., Nap, R. J., Szleifer, I. and Grzybowski, B. A. (2013). Nat. Nanotechnol., 8, 676–681. 20. Ye, X., et al. (2013). Nat. Chem. 5, 466–473.

21. O’Brien, M. N., Radha, B., Brown, K. A., Jones, M. R. and Mirkin, C. A. (2014). Angew. Chem. Int. Ed., 53, 9532–9538.

22. Singh, G., et al. (2014). Science, 345, 1149–1153.

23. Boles, M. A. and Talapin, D. V. (2014). J. Am. Chem. Soc., 136, 5868– 5871. 24. Ming, T., et al. (2008). Angew. Chem. Int. Ed., 47, 9685–9690.

25. Sacanna, S., Irvine,W. T. M., Chaikin, P. M. and Pine, D. J. (2010). Nature, 464, 575–578. 26. Ye, X., et al. (2013). Nano Lett., 13, 4980–4988.

27. Paik, T. and Murray, C. B. (2013). Nano Lett., 13, 2952–2956. 28. Kang, Y., et al. (2012). J. Am. Chem. Soc., 135, 42–45.

29. Paik, T., Ko, D-K., Gordon, T. R., Doan-Nguyen, V. and Murray, C. B. (2011). ACS Nano, 5, 8322–8330.

30. Macfarlane, R. J., et al. (2014). Proc. Natl. Acad. Sci. U.S.A., 111, 14995– 15000. 31. Jones, M. R., Seeman, N. C. and Mirkin, C. A. (2015). Programmable materials and the nature of the DNA bond. Science, 347, 1260901. 32. Macfarlane, R. J., et al. (2010). Angew. Chem. Int. Ed., 49, 4589–4592.

33. Auyeung, E., et al. (2014). Nature, 505, 73–77.

34. O’Brien, M. N., Jones, M. R., Brown, K. A. and Mirkin, C. A. (2014). J. Am. Chem. Soc., 136, 7603–7606.

35. O’Brien, M. N., Jones, M. R., Kohlstedt, K. L., Schatz, G. C. and Mirkin, C. A. (2015). Nano Lett., 15, 1012–1017. 36. Senesi, A. and Lee, B. (2015). J. Appl. Cryst., 48, 565–577.

37. Auyeung, E., Macfarlane, R. J., Choi, C. H. J., Cutler, J. I. and Mirkin, C. A. (2012). Adv. Mater., 24, 5181–5186. 38. Williamson, G. K. and Hall,W. H. (1953). Acta Metall. Mater., 1, 22–31. 39. Weidenthaler, C. (2011). Nanoscale, 3, 792–810.

40. Senesi, A. J., et al. (2014). Adv. Mater., 26, 7235–7240.

631

632

Anisotropic Nanoparticle Complementarity in DNA-Mediated Co-crystallization

41. Jin, R. C., Wu, G. S., Li, Z., Mirkin, C. A. and Schatz, G. C. (2003). J. Am. Chem. Soc., 125, 1643–1654.

42. Storhoff, J. J., et al. (2000). J. Am. Chem. Soc., 122, 4640–4650.

43. Mergny, J. L. and Lacroix, L. (2003). Oligonucleotides, 13, 515–537. 44. Zheng, J., et al. (2006). Nano Lett., 6, 1502–1504. 45. Schreiber, R., et al. (2014). Nat. Nano, 9, 74–78.

46. Gerling, T., Wagebauer, K. F., Neuner, A. M. and Dietz, H. (2015). Science, 347, 1446–1452.

47. Jones, M. R. and Mirkin, C. A. (2013). Angew. Chem. Int. Ed., 52, 2886– 2891.

Chapter 29

Programming Colloidal Crystal Habit with Anisotropic Nanoparticle Building Blocks and DNA Bonds*

Matthew N. O’Brien,a,b Hai-Xin Lin,a,b Martin Girard,c Monica Olvera de la Cruz,a,b,c and Chad A. Mirkina,b,c aDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cDepartment of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA [email protected]

Matthew N. O’Brien and Hai-Xin Lin contributed equally to this work.

Colloidal crystallization can be programmed using building blocks consisting of a nanoparticle core and DNA bonds to form materials with controlled crystal symmetry, lattice parameters, stoichiometry, and dimensionality. Despite this diversity of colloidal

*Reprinted with permission from O’Brien, M. N., Lin, H.-X., Girard, M., Olvera de la Cruz, M. and Mirkin, C. A. (2016). Programming colloidal crystal habit with anisotropic nanoparticle building blocks and DNA bonds, J. Am. Chem. Soc. 138, 14562−14565. . Further permissions related to the material excerpted should be directed to the ACS. Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

634

Programming Colloidal Crystal Habit with Anisotropic Nanoparticle Building Blocks

crystal structures, only spherical nanoparticles crystallized with BCC symmetry experimentally yield single crystals with welldefined crystal habits. Here, we use low-symmetry, anisotropic nanoparticles to overcome this limitation and to access single crystals with different equilibrium Wulff shapes: a cubic habit from cube-shaped nanoparticles, a rhombic dodecahedron habit from octahedron-shaped nanoparticles, and an octahedron habit from rhombic dodecahedron-shaped nanoparticles. The observation that one can control the microscopic shape of single crystals based upon control of particle building block and crystal symmetry has important fundamental and technological implications for this novel class of colloidal matter. The equilibrium shape of a crystal can be predicted with a Wulff construction, which plots the surface energy (γ) along each direction of a crystalline lattice in order to identify local minima in γ [1−3]. The crystalline planes corresponding to these minima possess the most stable interactions, and thus one would expect crystals bound by these facets. However, experimental realization of equilibrium structures represents a significant challenge in many atomic, molecular, and nanoscale systems due to energetic fluctuations in the system greater than the differences in γ. In the context of DNAmediated nanoparticle crystallization, the most stable interactions often contain the greatest number of hybridization events, or “bonding” interactions, between the DNA ligands on neighboring particles [4 −11]. Control of crystal habit therefore relates to the relative number of hybridization events along different crystalline planes [12], which can be tuned based on nanoparticle size and shape, DNA length and density, and lattice symmetry [13−18]. Despite their widespread use, high symmetry spherical nanoparticles are particularly challenging building blocks to use in this endeavor. This challenge originates from the weak interaction strength between spheres along their curved surfaces, as evidenced by relatively low DNA dehybridization temperatures (Tm) [19, 20] and small fractions of hybridized DNA, and the rotational freedom of spheres within a lattice [12, 21]. In contrast, the reduced symmetry of polyhedral nanoparticles yields structures that can template an oriented array of densely packed DNA on each facet, which facilitates stronger, directional interactions. Recent work has introduced the concept of a “zone of anisotropy” for polyhedral nanoparticles, or the phase

Programming Colloidal Crystal Habit with Anisotropic Nanoparticle Building Blocks

space where directional interactions templated by the anisotropy of the particle core persist and result in correlated particle orientations [11, 13, 15, 16, 19, 20]. Importantly, the face-to-face interactions that occur within this zone of anisotropy possess greater fractions of hybridized DNA and greater rotational restrictions than spherical nanoparticles, which in principle, should enable experimental realization of equilibrium habits (Fig. 29.1).

Figure 29.1  Nanoparticle shape can be used to control crystal habit in DNAmediated nanoparticle crystallization. Each shape crystallizes into a lattice with a different closest-packed plane (top) and crystal habit (bottom). Cube, octahedron, and rhombic dodecahedron nanoparticles (left to right) are shown with cube, rhombic dodecahedron, and octahedron crystal habits, respectively. Scale bars: 1 μm.

To test this hypothesis, three different polyhedral nanoparticle shapes were investigated: cubes, octahedra, and rhombic dodecahedra (Fig. 29.1) [16, 22]. When particles are prepared with self-complementary DNA sequences, such that every particle can connect to its neighbors, each of these shapes should crystallize into lattice symmetries with different closest packed planes and thus different Wulff shapes [16]. Within the zone of anisotropy for each shape, one would expect that cubes should crystallize into lattices with simple cubic (SC) symmetry, {100} closest packed planes, and cube habits; octahedra should crystallize into lattices with BCC symmetry, {110} closest packed planes, and rhombic dodecahedron habits; and rhombic dodecahedra should crystallize into lattices with face-centered cubic (FCC) symmetry, {111} closest packed

635

636

Programming Colloidal Crystal Habit with Anisotropic Nanoparticle Building Blocks

planes, and truncated octahedron habits. Indeed, the expected crystal habits can be experimentally realized with careful control of the crystallization conditions (Fig. 29.1).

Figure 29.2  Yield of crystals with equilibrium facets can be controlled based on particle symmetry, particle size (shown here as surface area, SA), and DNA length (D). (A) Facet yields are plotted over the phase space encoded by SA and D for each shape, where yield is color-coded and determined from EM analysis of silica-embedded crystals. (B) Examples of the highest-quality crystals formed from the investigated phase space, all with the predicted Wulff shape. Scale bars: 500 nm.

Two recent advances are of particular importance to realize these structures. First, advances in seed-mediated nanoparticle synthesis enable the use of significantly more uniform building blocks (>95% shape yield with 95% of the desired shape with Tm resulting in predominantly free DNA and T < Tm resulting in predominantly duplexed DNA (Fig. 39.2A). The reorganization rate of PAEs is expected to depend on the rate at which DNA dehybridizes and rehybridizes, and thus reorganization will accelerate with increasing koff. This suggests that at the temperature range in

Results and Discussion

which koff ~ C0kon, reorganization will be fastest, provided that the superlattice remains the thermodynamically favored configuration (i.e., the superlattice does not completely dissociate), and that this temperature should be the optimal annealing condition. In the subsequent sections, this model is experimentally tested.

39.2.3  DNA Sticky-End Sequence

To study the effect of the DNA bond energy on reorganization kinetics, a series of PAE systems were synthesized with sticky ends with varying G of hybridization, based on the free energy of a single sticky-end interaction at 25°C. As with free DNA, ∆G can be varied by adjusting the length of the sticky ends and/or their GC content (increasing strand length and increasing GC content each make ∆G more negative). Although the quality and domain size of the final crystals in fcc-forming systems obtained after extended annealing (up to 24 h at temperatures within ~1°C of their melting temperature) did not vary with ∆G values for an individual connection weaker than −90 kJ/mol, we hypothesized that the kinetics of reorganization would depend strongly on ∆G owing to its effect on koff. (Samples with ∆G of hybridization < −90 kJ/mol did not form ordered crystals, this “upper limit” of sticky-end hybridization strength is discussed later in this section). Thus, four PAE systems were synthesized with individual sticky-end ∆G values between −29 and −69 kJ/mol [23] (Fig. 39.3) and SAXS experiments were performed to observe the aggregates as they transitioned from a disordered to ordered arrangement at a given temperature. In a typical reorganization experiment, a sample was assembled into a disordered state at low temperature, then placed in the path of the X-ray beam where the system was rapidly heated to a desired annealing temperature (either just below, at, or slightly above the Tm) and SAXS patterns were repeatedly taken at intervals of several seconds. In all cases where the system had reached a sufficiently high temperature, a clear transition from a disordered aggregate to an fcc lattice was observed (Fig. 39.2B). The time point at which each system could be considered an fcc crystal (i.e., displaying diffraction peaks that are typical of fcc lattices) was then plotted as a function of temperature. When the rates of

781

782

Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization

reorganization at different temperatures are compared for each of the four systems examined, it is clear that the “thermal window” of crystallization (Fig. 39.2C; i.e., the range of temperatures over which crystalline structures are achieved within a reasonable time scale) for these samples decreases with more negative G values (Fig. 39.3A). This effect is more obvious when each PAE system is plotted relative to their respective melting temperature (Fig. 39.3B).

Figure 39.2  (A) General heuristic image demonstrating the temperature dependences of both hybridization (C0kon) and dehybridization (koff). When C0kon >> koff, sticky ends exist primarily in the bound state. When C0kon and koff are approximately equal, but C0kon > koff, DNA connections between particles are readily broken and reformed, leading to reorganization within the aggregate. When koff > C0kon, strands are primarily in the unbound state, which leads to aggregate melting. (B) Comparison of disordered and face-centered cubic SAXS patterns. (C) Example of the metric referred to as the “thermal window,” wherein particles reorganize on a reasonable timescale (100 s is chosen as an arbitrary point for qualitative comparison between all samples).

Interestingly, these results are qualitatively consistent with our proposed analogy between the behavior of free DNA and nanoparticle superlattices. Examining the theoretical rates reveals two predictions (Fig. 39.3A, inset). (i) Because the intersection of koff and C0kon increases with stronger sticky-end interactions (i.e., more negative ∆G values), Tm is also expected to increase; and (ii) because the slope of koff becomes steeper at Tm for more negative G,

Results and Discussion

the region in which koff ~ C0kon becomes narrower as the magnitude of ∆G increases [19]. Indeed, both of these predictions match the experimental trends. It is significant to note that the failure of systems with the strongest individual sticky ends tested (∆G values ≤ −90 kJ/mol) to crystallize after extended annealing at temperatures ~1°C below their Tm can also be explained by this effect, as this temperature was likely outside the annealing window for these systems.

Figure 39.3  Effects and trends of varying sticky-end sequence on reorganization kinetics. Each sticky-end is designated with a different color: Black: TGCA (Tm = 30.5°C; ∆G = −28.9 kJ/mol); red: TAGCTA (Tm = 44.3°C; ∆G = −33.9 kJ/mol); blue: TGCGCA (Tm = 57.0°C; ∆G = −56.9 kJ/mol); and green: GCGCGC (Tm = 60.3°C; ∆G = −69.0 kJ/mol); [NaCl] = 0.5 M for all samples. Lines are not quantitative fits, but rather guides for the eye: the solid portion of each line is within the experimentally examined temperature range and the dotted portion is an extrapolation beyond this regime. (A) Plot of the time required for crystallization versus absolute temperature. (B) Plot of the time required for crystallization versus relative temperature (each system’s Tm is set as 0). (A, inset): Heuristic image showing the effects of increasing linker strength (light to dark traces: more negative ∆G of hybridization).

783

784

Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization

Strikingly, in many systems, reorganization was observed at temperatures above Tm. Although the melting temperature determined by UV−vis spectroscopy is not a perfect determinant of the melting temperature of an aggregate in a capillary during an SAXS experiment due to a difference in concentration aggregate reorganization was found to occur simultaneously with aggregate melting in many systems. A likely cause for this is that PAEs at the edge of an aggregate have fewer DNA linkages holding them in place, resulting in more rapid melting than particles within the center of the aggregate. This competition between melting and reorganization suggests that the “upper limit” of DNA bond strength beyond which no ordered crystals are formed is essentially a kinetic wall. In these systems, temperatures necessary to induce a disordered-to-ordered transition on an observable time-scale are sufficiently high that the aggregate melts before crystallization is achieved. Thus, crystals have not been observed with these systems [3].

39.2.4  Number of Linkers

Increasing the number of DNA strands that can participate in bonding on each particle raises C0kon by virtue of the increased DNA density, but koff is unaffected by DNA concentration [9, 16]. Therefore, by adding more linkers, the practically flat C0kon curves are simply shifted up and intersect the unchanged koff curve at higher values (Fig. 39.4A). Following this, one can expect slightly higher Tm and, because koff has a higher slope at higher temperatures, a slightly narrowed annealing window. In an initial set of experiments, different sets of particles with a range of linker loadings (from 20 to 100 equivalents per particle) were allowed to reach their maximum crystal qualities upon a 24-h incubation at their optimal annealing temperatures (Fig. 39.4C). Subsequent analysis with SAXS showed that samples in the low-loading regime (i.e., 20 and 30 equivalents) displayed broad features corresponding to a large number of disordered regions that are overlaid with weak but discernible narrow peaks attributable to small fcc domains. This mixed structure could be predicted considering that each particle in an fcc lattice has 12 neighbors; even a modest nonuniformity in the distribution of linker strands among

Results and Discussion

the particles could lead to defects in the lattice. Assuming that the region of the particle surface corresponding to each linkage has a Poissonian distribution of DNA where the mean is the linker number divided by 12, in a system with 20 linker equivalents, less than 10% of particles will have DNA that can bind to all neighbors. Even for 30 linker equivalents, only 35% of particles will have DNA that can bind to each neighbor, while for 40 linker equivalents, over 65% of particles are expected to be able to bind to all of their neighbors. This simple statistical argument provides a rationale for the inability of low-linker–number systems to form uniform crystals.

Figure 39.4  (A) Heuristic image showing the kinetic effects of adding more linker equivalents [light to dark traces: increasing equivalents (eq.) per particle]. (B) Plot of the time required for crystallization versus relative temperature (each system’s Tm is set as zero). (C) Comparison of normalized SAXS patterns for systems with different numbers of linker equivalents after overnight incubation at their optimal annealing conditions; [NaCl] = 0.5 M for all samples. The black trace is the predicted scattering pattern for a perfect fcc lattice.

Kinetics experiments were also carried out for three different DNA loadings within the regime that demonstrated reliable crystal formation (45, 60, and 80 equivalents of DNA per particle). The thermal windows for these systems narrow as the linker density increases, in agreement with our aforementioned hypothesis that reorganization is hindered when a larger number of DNA connections tether a particle in place (Fig. 39.4B). Furthermore, within the regime of reliable crystal formation, we attribute the observed broadening of peaks as the linker loading increases (Fig. 39.4C) to the presence of smaller and less-perfect crystal domains, suggesting that very

785

786

Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization

dense DNA shells hinder the reorganization processes and limit the ability of particles to arrange themselves into large, perfect domains.

39.2.5  Salt Concentration

Changing the solution salt concentration has multiple effects on reorganization because it affects both C0kon and koff (Fig. 39.5A, inset) [18, 20]. Specifically, raising the salt concentration is expected to increase kon because the additional charge screening minimizes the repulsion between negatively charged DNA strands, and it is also expected to lower koff because the increased counterion concentration will help to stabilize the linkages once formed. To probe the effects of varying the solution cation concentration, samples were prepared using a standard linker sequence and number of linker equivalents (60 equivalents TAGCTA, ∆G = −33.9 kJ/ mol) at 0.5 M NaCl. Once the particles were assembled, the solution salt concentration was adjusted to lie within the range of 150 mM to 2.0 M. The salt concentration was modified after assembling the particles to ensure that the physical characteristics and behavior of the particles (number of thiolated and linker strands per particle, etc.) during the preparation process were the same as the previous experiments that monitored linker sequence and number of linkers per particle. This range of salt concentrations represents the maximum range in which aggregates were found to be stable; lower values of salt concentration led to crystals that were not stable at room temperature and higher values resulted in aggregates that formed dense, black pellets that stuck to the walls of the containers in which they were stored. When these samples were annealed at their respective optimal temperatures for 24 h, each formed fcc crystals of comparable quality and crystal domain size, indicating that changing these salt concentrations had no discernible effect on the ability of these systems to form ordered crystals. When kinetics experiments were performed at four different salt concentrations, examination of their thermal windows showed that systems with increasing melting temperatures had broader thermal windows than those with lower melting temperatures (Fig. 39.5A and B). Although this result may seem counterintuitive given that the opposite was obtained upon increasing the ∆G of hybridization of the sticky ends, it is actually in

Results and Discussion

agreement with our model. Because increasing the salt concentration raises both C0kon and koff, the absolute rates of kon and koff at Tm (and thus the absolute rates of reorganization) are higher when a system with greater salt concentration is annealed at Tm. Thus, the annealing window is broadened at higher salt concentration.

Figure 39.5  (A) Plot of the time required for systems with different NaCl concentrations (exact concentrations listed in B) to transition from disordered to ordered versus absolute temperature. (B) Plot of the time frame of reorganization versus relative temperature, with each system’s Tm set at 0. (A, inset) Heuristic image demonstrating the effect of increasing salt on both kon and koff (light to dark traces: increasing solution ionic strength).

The way in which salt concentration affects the annealing properties provides an interesting and useful contrast to tuning Tm through linker strength or density, where increasing Tm also narrows the annealing window. To further illustrate this, we studied

787

788

Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization

Figure 39.6  Comparison of thermal windows for two different stickyend sequences (TAGCTA, red traces, and TGCA, blue traces) whose melting temperatures have been modified by adjusting bulk solution salt concentration. Data are grouped into pairs of systems that have equivalent melting temperatures (denoted by matching data point shape and line type).

the kinetics of two systems with different sticky ends (and therefore different ∆G values) after tuning their salt concentrations such that their melting temperatures overlap. Because both experimental modifications of increasing the strength of individual sticky ends and lowering the salt concentration cause a narrowing of the thermal window (thus increasing the experimental difficulty of inducing crystallization), we hypothesized that the systems with the lowest magnitudes of ∆G and highest salt concentrations should prove easiest to anneal. Measurements of the reorganization rates for three such pairs of systems showed that the annealing window was lower (relative to Tm) and broader for the higher-salt–, lowerlinker–strength system of each pair, consistent with our hypothesis (Fig. 39.6). Again, the effect of increased thermal stability with elevated salt concentrations leads to higher absolute rates of reorganization and thus widens the range of temperatures at which

Results and Discussion

crystal formation can occur. More importantly, this demonstrates how modifying several design variables in a coordinated fashion can be used to tailor crystallization behavior in these systems, a feat that cannot be accomplished in traditional atomic- or molecularbased crystallization without altering the crystal type or symmetry achieved.

Figure 39.7  Implications of the proposed model of PAE reorganization. (A) Schematic showing the model in which particles can exist freely in solution (i), as singly bound intermediates (ii), or in a doubly bound state (iii). (B–D, top) The average number of particles that exist as singly bound intermediates are shown at different linker free energies (B), linker equivalents (C), and salt concentrations (D), with the color scale representing concentration of (ii). (Note: total particle concentration is 50 nM.) The white line indicates the computed melting temperature. (B–D, bottom) The computed annealing window at different sticky-end free energies (B), linker equivalents (C), and salt concentrations (D). These calculations are based on numerically solving the thermodynamic equations.

To gain further insight into the kinetics of the crystallization process, we have constructed a simplified model of DNA-mediated nanoparticle crystallization that includes multivalency. The multivalent effects can be captured in the simplest way possible by considering a three-state system where a pair of particles can either be free, bound to another particle by one DNA linkage, or bound

789

790

Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization

by two DNA linkages (Fig. 39.7A). At low temperatures, the system will be completely in the doubly bound state. As the temperature rises, some strands begin to dissociate and some particles enter the singly bound state. At the Tm, where half of the available DNA bonds have broken, a given particle will transition between doubly bound, singly bound, and free states with a rate that is related to the population of the singly bound state. At even higher temperatures, all DNA linkages break and all particles enter the free state. If there is reorganization occurring, singly bound particles must be present, as this state allows for particles to move between free and fully bound states. Therefore, the concentration of the singly bound intermediate may be considered a proxy for the ability of a system to reorganize. Interestingly, this means that the predicted reorganization in a system could be estimated from experimentally derived thermodynamic data with no free parameters. Specifically, we treat the process of free particles binding as having the same ∆G as that of the sticky end with a concentration equal to that of the particles. The process of the second strand binding is assumed to have the same ∆G as free DNA, but now includes the higher effective concentration of linker strands in the vicinity of the particle. It is interesting to note that the predicted variation of melting temperature and annealing window matches the observations discussed above (Fig. 39.7B–D), suggesting a simple extension of the free DNA model, even one based solely on thermodynamic parameters, could be very useful in predicted PAE crystallization.

39.3 

Conclusions

The experiments described above indicate that there are a large number of design variables that can be used to control the process of PAE crystallization, including solution temperature, DNA stickyend sequence, number of bound linker strands per particle, and salt concentration. Given that each of these variables can be controlled independently, it would therefore be useful to determine the optimal conditions under which one can most reliably synthesize colloidal crystals using PAEs as building blocks. Although the specific conditions under which crystals need to be assembled may vary depending on their application, a general set of principles that

Conclusions

make PAEs strong candidates for materials by design can still be established for this system. In essence, crystallization is easiest when there is a broad thermal window, and is fastest when annealing occurs at a relatively high temperature. From these observations, the experiments probing the effects of variations in sticky-end sequence provide the first principle of crystal design: the sticky ends of the DNA linkers should only be as strong as is necessary to stabilize an aggregate at the desired temperature range, thus maximizing the breadth of the thermal window. When considering the number of DNA linkers per particle, the trends at both extremes of linker density indicate that the highest-quality crystals are most readily achieved with smaller numbers of linker equivalents per particle, provided that sufficient linkers are present to stabilize an ordered structure. Thus, the second design principle is the minimum linker density necessary to obtain homogenous loading (where all particles are capable of forming a lattice as a thermodynamic product) facilitates the formation of the highest-quality crystals. Finally, tuning the salt concentration provides a facile means of controlling the crystallization process postsynthetically, without the need to prepare multiple sets of DNA strands or different batches of DNA-functionalized particles. The melting temperature of the system increased with increasing salt concentration up to 2.0 M NaCl; the thermal window in which reorganization occurred also broadened with increasing salt concentration up to a value of 1.0 M NaCl. This translates to the third design principle: the ionic strength of the solution should be as high as possible without altering the macroscopic behavior of the crystals (for example, by flocculation). In this work, we have analyzed the crystallization pathway for DNA-functionalized particles behaving as nanoscale PAEs, and found it to depend intricately on the thermodynamics of the DNA bonds. Importantly, this work has demonstrated that there are multiple synthetic and environmental handles that can be used to tune the interaction strength between nanoparticle building blocks, and that manipulating these variables allows us to control the relative ease with which crystals can be formed. These results should aid in the design of future crystals, as understanding the pathway by which a disordered aggregate becomes crystalline (as well as knowing

791

792

Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization

which design parameters and conditions facilitate this process) enables better control of the annealing procedure, allowing for the establishment of design processes that encourage reliable, highquality crystal syntheses. Moreover, the ability to stabilize a crystal under different environmental conditions (e.g., variations in solution ionic strength or temperature) improves their ability to be used in different contexts, such as epitaxial growth of crystals on surfaces [24], thermally addressable topotactic intercalation of multiple nanoparticle components [25], or assembly of Wulff polyhedra via slow cooling at high temperatures [15]. We also established that studies of systems where DNA bonds are formed via transient interactions between small sticky ends at the tip of primarily rigid DNA duplexes not only enable more complex superlattice design, but also facilitate fundamental investigations of their crystallization behavior. We therefore encourage both the experimental and theoretical scientific communities to further examine the behavior of these PAEs using techniques that have been previously applied to systems using long DNA overlaps between particles, or primarily single-stranded DNA linkers [9, 10, 12, 21, 22, 26]. Lastly, this work underscores the notion that PAEs present a highly program-mable means of studying and controlling crystallization in a more facile and directable manner than their atomic counterparts, and thus are a useful tool for materials synthesis.

Acknowledgments

This material is based upon work supported by the Air Force Office of Scientific Research (AFOSR) Awards FA9550-11-1-0275 and FA9550-12-1-0280. R.J.M. is a Ryan Fellow at Northwestern University. R.V.T. acknowledges a National Science Foundation Graduate Research Fellowship (DGE-1324585). K.A.B. acknowledges support from Northwestern University’s International Institute for Nanotechnology. Small-angle X-ray scattering was carried out at the DuPont–Northwestern–Dow Collaborative Access Team (DNDCAT) beamline located at Sector 5 of the Advanced Photon Source (APS). DND-CAT is supported by E. I. DuPont de Nemours & Co., The Dow Chemical Company, and the State of Illinois. Use of the APS was supported by the US Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract DE-AC02-06CH11357.

References

References 1. Cutler, J. I., Auyeung, E. and Mirkin, C. A. (2012). J. Am. Chem. Soc., 134(3), 1376–1391.

2. Macfarlane, R. J., O’Brien, M. N., Petrosko, S. H. and Mirkin, C. A. (2013). Angew. Chem. Int. Ed. Engl., 52(22), 5688–5698. 3. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382(6592), 607–609. 4. Park, S. Y., et al. (2008). Nature, 451(7178), 553–556.

5. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451(7178), 549–552. 6. Macfarlane, R. J., et al. (2011). Science, 334(6053), 204–208.

7. Jones, M. R., et al. (2010). Nat. Mater., 9(11), 913–917.

8. Macfarlane, R. J., et al. (2009). Proc. Natl. Acad. Sci. U.S.A., 106(26), 10493–10498. 9. Biancaniello, P. L., Kim, A. J. and Crocker, J. C. (2005). Phys. Rev. Lett., 94(5), 058302.

10. Hsu, C. W., Sciortino, F. and Starr, F. W. (2010). Phys. Rev. Lett., 105(5), 055502.

11. Li, T. I., Sknepnek, R., Macfarlane, R. J., Mirkin, C. A. and de la Cruz, M. O. (2012). Nano Lett., 12(5), 2509–2514. 12. Valignat, M.-P., Theodoly, O., Crocker, J. C., Russel, W. B. and Chaikin, P. M. (2005). Proc. Natl. Acad. Sci. U.S.A., 102(12), 4225–4229. 13. Senesi, A. J., et al. (2014). Adv. Mater., 26(42), 7235–7240.

14. Hill, H. D., et al. (2008). Nano Lett., 8(8), 2341–2344.

15. Auyeung, E., et al. (2014). Nature, 505(7481), 73–77.

16. Macfarlane, R. J., et al. (2010). Angew. Chem. Int. Ed. Engl., 49(27), 4589–4592. 17. Kewalramani, S., et al. (2013). ACS Nano, 7(12), 11301–11309.

18. Bloomfield, V. A., Crothers, D. M. and Tinoco, I. (2000). Nucleic Acids: Structures, Properties, and Functions (University Science Books, Sausalito, CA).

19. Craig, M. E., Crothers, D. M. and Doty, P. (1971). J. Mol. Biol., 62(2), 383– 401. 20. Okahata, Y., et al. (1998). Anal. Chem., 70(7), 1288–1296.

21. Xiong, H., van der Lelie, D. and Gang, O. (2009). Phys. Rev. Lett., 102(1), 015504.

793

794

Importance of the DNA “Bond” in Programmable Nanoparticle Crystallization

22. Tkachenko, A. V. (2002). Phys. Rev. Lett., 89(14), 148303.

23. Breslauer, K. J., Frank, R., Blöcker, H. and Marky, L. A. (1986). Proc. Natl. Acad. Sci. U.S.A., 83(11), 3746–3750.

24. Senesi, A. J., et al. (2013). Angew. Chem. Int. Ed. Engl., 52(26), 6624– 6628. 25. Macfarlane, R. J., Jones, M. R., Lee, B., Auyeung, E. and Mirkin, C. A. (2013). Science, 341(6151), 1222–1225.

26. Dreyfus, R., et al. (2010). Phys. Rev. E: Stat. Nonlinear Soft Matter Phys., 81(4 Pt 1), 041404.

Chapter 40

Modular and Chemically Responsive Oligonucleotide “Bonds” in Nanoparticle Superlattices*

Stacey N. Barnaby,a,b Ryan V. Thaner,a,b Michael B. Ross,a,b Keith A. Brown,a,b,c George C. Schatz,a,b and Chad A. Mirkina,b aDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cDepartment of Mechanical Engineering, Boston University, 110 Cummington Mall, Boston, MA 02215, USA [email protected]

If a solution of DNA-coated nanoparticles is allowed to crystallize, the thermodynamic structure can be predicted by a set of structural design rules analogous to Pauling’s rules for ionic crystallization. *Reprinted with permission from Barnaby, S. N., Thaner, R. V., Ross, M. B., Brown, K. A., Schatz, G. C. and Mirkin, C. A. (2015). Modular and chemically responsive oligonucleotide “bonds” in nanoparticle superlattices, J. Am. Chem. Soc. 137, 13566−13571. Copyright (2015) American Chemical Society.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

796

Modular and Chemically Responsive Oligonucleotide “Bonds”

The details of the crystallization process, however, have proved more difficult to characterize as they depend on a complex interplay of many factors. Here, we report that this crystallization process is dictated by the individual DNA bonds and that the effect of changing structural or environmental conditions can be understood by considering the effect of these parameters on free oligonucleotides. Specifically, we observed the reorganization of nanoparticle superlattices using time-resolved synchrotron smallangle X-ray scattering in systems with different DNA sequences, salt concentrations, and densities of DNA linkers on the surface of the nanoparticles. The agreement between bulk crystallization and the behavior of free oligonucleotides may bear important consequences for constructing novel classes of crystals and incorporating new interparticle bonds in a rational manner.

40.1 Introduction

DNA is a powerful ligand for programming the assembly of nanoparticles into superlattices with a vast number of crystallographic symmetries [1a−g]. This can be achieved by using a programmable atom equivalent (PAE), which consists of a nanoparticle core densely functionalized with geometrically defined oligonucleotides, where DNA mediates interactions between nanoparticles. The oligonucleotide density and rigid nanoparticle core impose a radial orientation of the DNA and valency to the nanoparticles. Initially, spherical gold nanoparticles (AuNPs) were studied as PAE cores, but subsequent work has found that this approach is core generalizable, as other inorganic [2a,b] and organic [1f] cores, anisotropic cores [3a,b], as well as biological materials, such as proteins [4], can be assembled using the same design rules [1d]. The unifying element of all these studies is the DNA “bond” that programs nanoparticle interactions and drives their assembly into ordered crystalline structures. While recent work has been dedicated to understanding the function of these materials including emergent plasmonic [2a, 5] and catalytic properties [4, 6], these properties are predominantly derived from the nanoparticle core. Studies of how the bond contributes to the functional properties of the crystalline superlattice are absent.

Introduction

When considering materials that could in principle be used as a programmable ligand to assemble nanoparticles, DNA is not the only candidate. Specifically, the incorporation of RNA into nanoparticle superlattices would enable new classes of functional and stimuliresponsive superstructures that are not achievable with DNA or solely by engineering the PAE building block core. Though RNA is chemically similar to DNA (the primary difference is the presence of a 2¢-hydroxyl (2¢-OH) group in RNA), it has a vast chemical, structural, and functional design space that exceeds that of DNA [7]. For example, in cells, while DNA is often found in the form of long double helices [8], RNA is generally composed of short helices surrounded by loops and bulges [9]. Notable forms of biofunctional RNA include small interfering RNA (siRNA) that can regulate gene expression [10], ribozymes (ribonucleic acid enzymes) which are catalytic RNA molecules [11], and riboswitches which are structures formed in mRNA that can regulate gene expression in bacteria [12] and even act as stimuli responsive sensors [13]. While the vast chemical and biological space that RNA occupies may appear to make it an ideal ligand for endowing PAEs with additional functionalities, its instability and vulnerability to nuclease-catalyzed hydrolysis [14] provides a substantial barrier to realizing biomaterials based upon RNA. Research on synthesizing RNA biomaterials has focused on the analogy to DNA hybridization, where rigidity is imposed by the DNA hybridization events, which leads to rigid structures and therefore valency [15]. This approach, based purely on DNA hybridization, has been extended to RNA for the synthesis of micrometer scale RNA filaments, molecular jigsaw puzzles [16], and square-shaped RNA particles [17a,b]. In order for these syntheses to work for RNA, however, a hierarchical multistep process is required, whereas DNA structures can typically be made in a “one pot” synthesis [18]. In addition, the DNA and RNA-based hybridization approaches require the use of both simulation and experiment to rationally design the 3D RNA architectures through initial computer modeling [19]. This strategy is conceptually related but different from the method discussed herein for forming nanoparticle-based templated bonds, where the rigid nanoparticle core leads to a radial upright orientation of the densely packed DNA, leading to valency imposed by the core.

797

798

Modular and Chemically Responsive Oligonucleotide “Bonds”

Therefore, the well-understood nature of DNA programmable assembly, through the established design rules for the rational construction of DNA nanoparticle superlattices [1d], provides the perfect platform for exploring the degree to which non-DNA oligonucleotides can serve as programmable “bonds.” Here, the conventional design space for DNA-programmable assembly is transformed by introducing oligonucleotide identity (i.e., PAEs held together by DNA/DNA, RNA/RNA, or DNA/RNA duplexes) as an important design parameter. Similar to conventional DNA-based assembly, the programmable nature of the oligonucleotide bond is the driving force, and is independent of the oligonucleotide identity such that DNA/DNA, RNA/RNA, and DNA/RNA duplexes are all suitable programmable ligands. However, the ability to tune the bond identity enables the rational design of responsive materials, whereby the oligonucleotide bond identity and interparticle distance dictate the response to enzymes.

40.2  Results and Discussion

40.2.1 Design of Nanoparticle Superlattices with Different Oligonucleotide Bonds Design rules for the synthesis of nanoparticle superlattices with a variety of crystallographic symmetries have been established, which allow one to independently adjust each of the relevant crystallographic parameters, including particle size, periodicity, and interparticle distance [1d]. Because these design rules are based upon explorations of DNA as the programmable ligand, one must first explore how the oligonucleotide bond identity affects the programmable assembly of nanoparticle superlattices. We hypothesize that since RNA/RNA and RNA/DNA binding proceeds in a similar fashion to DNA/DNA binding, the use of DNA, RNA, or a DNA/RNA heteroduplex will not significantly change the resulting nanoparticle superlattice crystal structure. As an initial proof-ofconcept study, a two-component system which is expected to yield superlattices with a body-centered cubic (bcc) crystallographic symmetry, was evaluated. Four binary sets of particles were functionalized with DNA or RNA with non-self-complementary sticky ends, such that particle A can only bind to particle B and vice

Results and Discussion

versa (Fig. 40.1a). The DNA and RNA design contains short overhang regions on the 3¢ end of the linkers, which facilitate the interactions between nanoparticles. For each sample, particle A and particle B were mixed in a 1:1 ratio so that each sample was allowed to form aggregates. The possible permutations of oligonucleotide bonds are A-DNA/B-DNA (red), A-RNA/B-RNA (blue), A-DNA/B-RNA (dark purple), and A-RNA/B-DNA (light purple).

Figure 40.1  Nanoparticle superlattices synthesized with modular oligonucleotide bonds. (a) Two-component system where particles A and B are linked by non-self-complementary sticky ends. Four types of oligonucleotide bonds are explored: DNA/DNA (red), RNA/RNA (blue), DNA/RNA (dark purple), and RNA/DNA (light purple). (b) UV−vis melts of aggregates linked with four different oligonucleotide bond compositions at short linker lengths (color scheme is the same as in (a)). Tunable melting transitions (Tm) emerge based on oligonucleotide bond composition.

40.2.2 Synthesis and Characterization of DNA and RNA Nanoparticle Superlattices It is well-known that the density of DNA affects the cooperative melting transition and crystallization of PAEs [20, 21]. In order to explore new oligonucleotide identities as bonding ligands, a method

799

800

Modular and Chemically Responsive Oligonucleotide “Bonds”

to functionalize particles with RNA at a density similar to that attainable with DNA must be developed [21]. In previous reports, RNA immobilized on gold nanoparticles (AuNPs) has been exclusively in the form of double-stranded RNA [22a,b], where a backfill molecule was added to passivate the remaining gold surface to account for the lower loading of RNA compared to what is observed with DNA [21]. But herein, RNA particles A and B need to be synthesized with single-stranded RNA without backfill molecules. Therefore, RNA particles A and B were synthesized using methods analogous to their DNA counterparts. This was found to significantly increase the density of thiolated RNA on particles A and B from 25−60 pmol/cm2 to 50−75 pmol/cm2, thus allowing RNA particle A and particle B to be analogous to their DNA counterparts in terms of oligonucleotide density. Once this has been accomplished, the only parameters that need to be optimized to synthesize nanoparticle superlattices with different bond compositions are the strength and length of the sticky end, the spacer unit between the nanoparticle surface, and the oligonucleotide recognition sequence. It has been widely observed that PAEs exhibit cooperative and sharp melting transitions (transition widths of 2−8°C) compared to linear nucleic acids, which exhibit broad melting transitions (transition width ~20°C) [20, 22a,b]. A typical melting experiment involves monitoring the optical extinction at 260 and 520 nm, which is dampened for room temperature assembled PAE aggregates, but increases as the temperature is raised and the nanoparticles begin to dissociate. All aforementioned PAE-oligonucleotide combinations (Fig. 40.1a) exhibited sharp and cooperative melting transitions (Fig. 40.1b). Notably, the characteristic melting transition (Tm) for the RNA-PAE aggregates occurred about 13°C higher than analogous DNA-PAE aggregates (full width at half-maximum (fwhm) = 2.7°C for both), where the only difference is the identity of the oligonucleotide. A similar stabilization effect exists in molecular duplexes of RNA, which exhibit greater thermal stability and higher melting temperatures than their DNA counterparts [7]. Additionally, the two PAE aggregates held together by heteroduplexes exhibited two distinct melting transitions (Tm = 39°C, fwhm = 3°C for aggregates with RNA-PAE A; Tm = 50°C, fwhm = 1.3°C for aggregates with DNAPAE A). While this result may appear surprising given the similarity of RNA and DNA, the position of the melting transitions for the two

Results and Discussion

heteroduplexes can be understood by examining the molecular counterparts for their sticky ends, where the same trend in melting temperatures was observed for molecular duplexes of similar sequences as the sticky ends [23a,b]. More specifically, the stability of homopurine-homopyrimidine oligomer duplexes mirrors the trend in superlattice melting temperatures when looking at stickyend identity [24a−c], thus demonstrating that the characteristics of hybrid molecular duplexes are maintained when they are used as programmable ligands. Additionally, adenine DNA/uracil RNA heteroduplexes are known to be exceptionally unstable [23b, 25a,b]. This heteroduplex is analogous to the sticky-end interaction in A-RNA/B-DNA and thus explains the lower melting temperature. Finally, this trend in melting temperatures was found to persist as the sticky end was increasingly moved away from the particle surface by utilizing longer linker oligonucleotides. Recent work has demonstrated that slowly cooling PAEs through their melting transition is an effective method for synthesizing micron-scale single crystals [26]. To test whether RNA-programmable assembly could also be used to form such large scale crystals, PAE aggregates were slowly cooled (0.01°C/min) from 5 to 10°C above their melting temperature down to room temperature. Small-angle X-ray scattering (SAXS) was used to confirm the bcc crystallographic symmetry [1d] (Fig. 40.2a) across three different oligonucleotide linker length scales and four different oligonucleotide bond compositions (Fig. 40.2b). Despite differences at the molecular level, both DNA and RNA can be used interchangeably with the same crystal design principles, as evidenced by SAXS and SEM. For example, the macroscopic crystallites formed by this slow cooling process were examined by SEM and determined to be rhombic dodecahedra (Fig. 40.2c), as observed in pure DNA systems [27]. Though the translated sequences used for assembly are identical (i.e., the DNA and RNA used on all A type particles had the same sequence just a different oligonucleotide identity), the A-DNA/B-DNA superlattices exhibited the largest interparticle distance, while the A-RNA/BRNA superlattices consistently exhibited the shortest interparticle distances (Table 40.1). These data can be understood by looking at the typical characteristics of the molecular duplexes and specifically the 0.275 nm rise per base pair for RNA (A-form) as compared with 0.34 nm for DNA (B-form) [7, 28]. DNA/RNA heteroduplexes are typically intermediate in pitch, however, it is difficult to predict the

801

802

Modular and Chemically Responsive Oligonucleotide “Bonds”

properties of a DNA-RNA heteroduplex compared to its homoduplex counterpart, as they are known to be highly sequence specific [23a].

Figure 40.2  Body-centered cubic nanoparticle superlattices. (a) Depictions of body-centered cubic (bcc) nanoparticle superlattices of four different oligonucleotide bond compositions (to-scale). Gold nanoparticles are shown in yellow and the oligonucleotide bond in red, blue, or purple. (b) Small-angle X-ray scattering (SAXS) of nanoparticle superlattices with four different “bond” compositions and three different interparticle distances (from bottom to top: short (46-base-pair (-bp)), medium (67-bp), and long (128-bp) linkers). (c) Scanning electron microscopy (SEM) images of nanoparticle superlattices in the solid state. In all cases, single-crystal rhombic dodecahedra are observed. Scale bars: 100 nm.

Use of the Scherrer equation allows one to calculate the mean crystallite size for a given sample, defined as the average diameter of a single-crystalline domain. These calculations show that the grain sizes are all very similar regardless of bond type (Table 40.1), thus demonstrating the power of programmable assembly for generating crystals of similar size but with different oligonucleotide constituents. Together, these data demonstrate that tuning the oligonucleotide bond is an important new handle for on-demand materials properties including melting temperature and interparticle distance in crystalline nanoparticle materials.

Results and Discussion

Finally, SAXS patterns of analogous DNA and RNA superlattices stored at 25°C were obtained throughout the course of 100 days. These data revealed that the superlattices remain well ordered, with the interparticle distance changing 80% particle B¢. (b) Scattering pattern for 50% particle B¢, showing the emergence of the (100) reflection, indicating the removal of the B¢ particles and the remaining nanoparticles in a simple cubic arrangement. (c) Illustrations of particles in a bcc (blue) and simple cubic (black) arrangement. (d) SEM images. Scale bars: 1 μm. (e) A figure of merit, site occupancy factor (SOF), used to measure the number of vacancies generated after particle B¢ was removed. Plot of one minus SOF (1 − SOF) versus fraction of particle B¢, fit to a linear regression. These data allow for a quantitative measure of vacancies in the nanoparticle crystals.

In order to quantify the number of vacancies present after enzymatic removal of particle B¢, the SAXS data from Fig. 41.5a were analyzed to generate a figure of merit called the site occupancy factor (SOF), which compares the intensity of the (100) reflection for simple cubic to that of the (110) reflection for a bcc lattice (Fig. 41.5c), such that SOF = 1 is the case where there are no vacancies. For particle A, lattice sites (within error) are expected to be filled at positions (0, 0, 0), and thus one minus SOF (1 − SOF) for particle A is

Results and Discussion

0 (meaning zero empty sites). For particles B and B¢, the number of sites at positions (1/2, 1/2, 1/2) can be varied from 0 to 1 by RNase A, and thus 1 − SOF for particle B, B¢ will range from 0 (no vacancies) to 1 (100% vacancies; no crystal). A plot of 1 − SOF for particles B, B¢ versus the fraction of particle B¢ fit to a linear regression has a slope of 0.389 (Fig. 41.5e). These data reveal that 30−40% of the B¢ particles that are added to the system are ultimately observed as vacancies. The equation of the line allows one to input a desired number of vacancies and calculate the ratio of particle B:B¢ required at synthesis. Taken together, these data suggest that RNA bonding elements can be hydrolyzed up to a certain point (80% particle B¢), after which the larger crystal can no longer stay intact. We note that these values are calculated based on the model for Structure 3. They may account for the fractions of homogeneously distributed vacancies and underestimate the fractions of overall vacancies including inhomogeneously distributed ones. When a B¢ particle rich region is hydrolyzed, there may not be an ordered array of particles (as expected from S(q) of samples with high percentages of B¢ particles; Fig. 41.5a), and thus remaining particles in the region do not contribute to the diffraction peaks. These data all confirm the presence of Structure 4 and also provide insight into the nucleation and growth processes of nanoparticle superlattices. Polycrystalline nanoparticle superlattices crystallize through rapid hybridization and dehybridiza-tion of sticky ends between adjacent nanoparticles [9]. Here, when the melting temperature of DNA and RNA bonding elements is well-matched (i.e., ≈3°C difference), such that reorganization of both particle types can occur simultaneously, and only one crystal forms (evident by SAXS) even though nucleation between the bonding elements of particles A and B¢ happens first. Future work investigating greater melting temperature differentials between the bonding elements may lead to different types of structures, such as core−shell and phase-separated crystals. In addition, we also demonstrate a strategy to manipulate nanoscale bonds without changing the identity of the nanoparticle atom, something not observed in atomic systems. These data lay the groundwork for new types of structures such as those with programmable amounts of defects and dopants. All atomic crystalline materials contain some type of intrinsic defect structure that is often responsible for tuning and controlling the material properties

823

824

Enzymatically Controlled Vacancies in Nanoparticle Crystals

beyond what is already present due to the atomic bonding and lattice structure [32−34]. We believe that methodically combining different oligonucleotide bonds can lead to defect structures in nanoparticle superlattices with enhanced properties, which would be of interest for catalytic and optical materials.

Acknowledgments

This material is based upon work supported by the following awards: Air Force Office of Scientific Research FA9550-11-1-0275; Department of Defense National Security Science and Engineering Faculty Fellowship N00014-15-1-0043; Department of the Navy, Office of Naval Research N00014-11-1-0729, and the National Science Foundation’s MRSEC program (DMR-1121262) at the Materials Research Center of Northwestern University. S.N.B. and R.V.T. acknowledge National Science Foundation Graduate Research Fellowships. S.N.B. also acknowledges a P.E.O. Scholar Award. M.B.R. acknowledges a National Defense and Science Engineering Graduate Research Fellowship. This work made use of the EPIC facility of the NUANCE Center at Northwestern University, which has received support from the Soft and Hybrid Nanotechnology Experimental (SHyNE) Resource (NSF NNCI-1542205); the MRSEC program (NSF DMR-1121262) at the Materials Research Center; the International Institute for Nanotechnology (IIN); the Keck Foundation; and the State of Illinois, through the IIN. Portions of this work were carried out at the DuPont-Northwestern-Dow Collaborative Access Team (DND-CAT) beamline located at Sector 5 of the Advanced Photon Source (APS). DND-CAT is supported by E. I. DuPont de Nemours & Co., Dow Chemical Company, and the state of Illinois. Use of the APS was supported by the U.S. DOE, Office of Science, Office of Basic Energy Sciences, under contract DE-AC02-06CH11357.

References

1. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382(15), 607−609. 2. Park, S. Y., Lytton-Jean, A. K. R., Lee, B., Weigand, S., Schatz, G. C. and Mirkin, C. A. (2008). Nature, 451(7178), 553−556.

References

3. Macfarlane, R. J., Lee, B., Jones, M. R., Harris, N., Schatz, G. C. and Mirkin, C. A. (2011). Science, 334(6053), 204−208.

4. Alivisatos, A. P., Johnsson, K. P., Peng, X., Wilson, T. E., Loweth, C. J., Bruchez, M. P. and Schultz, P. G. (1996). Nature, 382(6592), 609−611.

5. Tan, S. J., Campolongo, M. J., Luo, D. and Cheng, W. (2011). Nat. Nanotechnol., 6(5), 268−276.

6. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451(7178), 549−552.

7. Jones, M. R., Seeman, N. C. and Mirkin, C. A. (2015). Science, 347(6224), 1260901. 8. Hill, H. D., Macfarlane, R. J., Senesi, A. J., Lee, B., Park, S. Y. and Mirkin, C. A. (2008). Nano Lett., 8(8), 2341−2344.

9. Macfarlane, R. J., Thaner, R. V., Brown, K. A., Zhang, J., Lee, B., Nguyen, S. T. and Mirkin, C. A. (2014). Proc. Natl. Acad. Sci. U.S.A., 111(42), 14995−15000.

10. Seo, S. E., Wang, M. X., Shade, C. M., Rouge, J. L., Brown, K. A. and Mirkin, C. A. (2016). ACS Nano, 10(2), 1771−1779. 11. Pal, S., Zhang, Y., Kumar, S. K. and Gang, O. (2015). J. Am. Chem. Soc., 137(12), 4030−4033. 12. Maye, M. M., Kumara, M. T., Nykypanchuk, D., Sherman, W. B. and Gang, O. (2010). Nat. Nanotechnol., 5(2), 116−120.

13. Zhang, Y., Pal, S., Srinivasan, B., Vo, T., Kumar, S. and Gang, O. (2015). Nat. Mater., 14(8), 840−847.

14. Barnaby, S. N., Thaner, R. V., Ross, M. B., Brown, K. A., Schatz, G. C. and Mirkin, C. A. (2015). J. Am. Chem. Soc., 137(42), 13566−13571. 15. Zhang, C., Macfarlane, R. J., Young, K. L., Choi, C. H. J., Hao, L., Auyeung, E., Liu, G., Zhou, X. and Mirkin, C. A. (2013). Nat. Mater., 12(8), 741−746. 16. Sun, D. and Gang, O. (2011). J. Am. Chem. Soc., 133(14), 5252−5254.

17. Young, K. L., Ross, M. B., Blaber, M. G., Rycenga, M., Jones, M. R., Zhang, C., Senesi, A. J., Lee, B., Schatz, G. C. and Mirkin, C. A. (2014). Adv. Mater., 26(4), 653−659. 18. Jones, M. R., Macfarlane, R. J., Lee, B., Zhang, J., Young, K. L., Senesi, A. J. and Mirkin, C. A. (2010). Nat. Mater., 9(11), 913−917. 19. O’Brien, M. N., Jones, M. R., Lee, B. and Mirkin, C. A. (2015). Nat. Mater., 14(8), 833−839.

20. Lu, F., Yager, K. G., Zhang, Y., Xin, H. and Gang, O. (2015). Nat. Commun., 6, 6912.

825

826

Enzymatically Controlled Vacancies in Nanoparticle Crystals

21. Ross, M. B., Ku, J. C., Lee, B., Mirkin, C. A. and Schatz, G. C. (2016). Adv. Mater., 28(14), 2790−4.

22. Senesi, A. J., Eichelsdoerfer, D. J., Macfarlane, R. J., Jones, M. R., Auyeung, E., Lee, B. and Mirkin, C. A. (2013). Angew. Chem. Int. Ed., 52(26), 6624−6628. 23. Auyeung, E., Li, T. I. N. G., Senesi, A. J., Schmucker, A. L., Pals, B. C., de la Cruz, M. O. and Mirkin, C. A. (2013). Nature, 505(7481), 73−77.

24. Ross, M. B., Ku, J. C., Vaccarezza, V. M., Schatz, G. C. and Mirkin, C. A. (2015). Nat. Nanotechnol., 10(5), 453−458. 25. Macfarlane, R. J., Jones, M. R., Lee, B., Auyeung, E. and Mirkin, C. A. (2013). Science, 341(6151), 1222−1225.

26. Jin, R., Wu, G., Li, Z., Mirkin, C. A. and Schatz, G. C. (2003). J. Am. Chem. Soc., 125(6), 1643−1654. 27. Prytkova, T. R., Eryazici, I., Stepp, B., Nguyen, S. T. and Schatz, G. C. (2010). J. Phys. Chem. B, 114(8), 2627−2634.

28. Sugimoto, N., Nakano, S.-I., Katoh, M., Matsumura, A., Nakamuta, H., Ohmichi, T., Yoneyama, M. and Sasaki, M. (1995). Biochemistry, 34(35), 11211−11216.

29. Sugimoto, N., Nakano, S., Yoneyama, M. and Honda, K. (1996). Nucleic Acids Res., 24(22), 4501−4505. 30. Auyeung, E., Macfarlane, R. J., Choi, C. H., Cutler, J. I. and Mirkin, C. A. (2012). Adv. Mater., 24(38), 5181−5186. 31. Raines, R. T. (1998). Chem. Rev., 98(3), 1045−1066.

32. Banhart, F., Kotakoski, J. and Krasheninnikov, A. V. (2011). ACS Nano, 5(1), 26−41. 33. Watkins, G. D. (2000). Mater. Sci. Semicond. Process., 3(4), 227−235.

34. Tang, M., Colombo, L., Zhu, J. and Diaz de la Rubia, T. (1997). Phys. Rev. B: Condens. Matter Mater. Phys., 55(21), 14279−14289.

Chapter 42

Modulating the Bond Strength of DNA-Nanoparticle Superlattices*

Soyoung E. Seo,a,b Mary X. Wang,b,c Chad M. Shade,a,b Jessica L. Rouge,a,b Keith A. Brown,a,b and Chad A. Mirkina,b,c aDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cDepartment of Chemical and Biological Engineering, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208-3113, USA [email protected]

Soyoung E. Seo and Mary X. Wang contributed equally to this work.

A method is introduced for modulating the bond strength in DNA− programmable nanoparticle (NP) superlattice crystals. This method utilizes noncovalent interactions between a family of [Ru(dipyrido[2,3a:3¢,2¢-c]phenazine)(N−N)2]2+-based small molecule intercalators *Reprinted with permission from Seo, S. E., Wang, M. X., Shade, C. M., Rouge, J. L., Brown, K. A. and Mirkin, C. A. (2016). Modulating the bond strength of DNA-nanoparticle superlattices, ACS Nano 10, 1771−1779. Copyright (2016) American Chemical Society.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

828

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

and DNA duplexes to postsynthetically modify DNA-NP superlattices. This dramatically increases the strength of the DNA bonds that hold the nanoparticles together, thereby making the superlattices more resistant to thermal degradation. In this work, we systematically investigate the relationship between the structure of the intercalator and its binding affinity for DNA duplexes and determine how this translates to the increased thermal stability of the intercalated superlattices. We find that intercalator charge and steric profile serve as handles that give us a wide range of tunability and control over DNA-NP bond strength, with the resulting crystal lattices retaining their structure at temperatures more than 50°C above what nonintercalated structures can withstand. This allows us to subject DNA-NP superlattice crystals to conditions under which they would normally melt, enabling the construction of a core−shell (gold NP-quantum dot NP) superlattice crystal.

42.1 Introduction

The ability to build complex, hierarchical materials with precise control over the identity and placement of each component has long been a goal of materials science [1]. Nature provides many elegant examples of self-assembled functional materials with multiple levels of organization. Many synthetic routes for nanoparticle (NP) crystallization have been explored, yet the engineering of versatile building blocks that rationally assemble into complex macroscopic materials remains a challenge. DNA is an attractive candidate for nanoparticle assembly due to its length, sequence, and chemical programmability [2]. This allows for the crystallization of DNA-NP conjugates into extended threedimensional superlattices dictated by well-established design rules [3, 4]. Despite being constructed from semiflexible polymers (i.e., oligonucleotides), these materials have extraordinarily well-defined symmetry. To date, researchers have utilized DNA-mediated NP assembly to make lattices with over 24 different crystal symmetries [4−8] consisting of various nanoparticle compositions [9, 10] and interparticle distances [11]. In addition, this technique, in certain cases, allows one to control three-dimensional crystal habit (e.g., rhombic dodecahedra single crystals are formed from body centered cubic (bcc) superlattices) [12]. This method has been used to create

Introduction

a variety of functional materials, including catalysts [13] and optical devices [14]. While a great deal of work has been done in controlling architecture and building block identity, relatively little work has been conducted on deliberately tuning bond strength. Changing the interparticle bond strength has been explored via changing the identity of the nucleic acid (RNA vs. DNA) [15], varying counterion concentration, DNA sequence and number of oligonucleotides per particle [16], and introducing ethidium bromide, a well-known DNA intercalator [17]. Of these methods, intercalation is the only method that allows bond strength to be altered after superlattice assembly has occurred. This is advantageous since nanoparticle crystallization cannot occur when DNA hybridization strength is above a certain limit [16, 18, 19].

Figure 42.1 RuII complexes with extended ancillary ligands varying in steric profile and charge.

The complex chemical structure of the DNA helix facilitates many types of highly specific molecular interactions. Small molecule DNA intercalators insert between DNA bases and can have an effect on both the thermodynamic and structural properties of the duplex

829

830

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

[20, 21]. The degree to which intercalators affect the DNA duplex is dependent upon their molecular structure, charge, and composition [22]. In principle, a family of structurally different intercalator molecules could provide a way to rationally tailor oligonucleotide bond strength in DNA-NP superlattices. Herein, we evaluate the use of a family of structurally different intercalator molecules to noncovalently modulate the bond strength of oligonucleotides in a DNA-NP superlattice. We investigate the effect of interalator charge and steric profile on its binding affinity for double strand deoxyribose nucleic acid (dsDNA) and how this translates to the thermal stability and structure of intercalated bcc DNA-NP superlattices. As a proof of concept, we then selected the most suitable intercalator molecule to facilitate the creation of a core−shell material that has not previously been demonstrated with DNA-NP assembly techniques. This material consists of a DNA−gold NP (AuNP) single-crystal core coated via stepwise growth cycles in a shell of DNA−quantum dot NPs (QdNPs). The general architecture of the intercalator molecule is based on work by Barton and co-workers, which showed that [Ru(bpy)2(dppz)]2+ (dppz = dipyrido[3,2-a:2¢,3¢-c]phenazine, bpy = 2,2¢-bipyridine) effectively intercalates into dsDNA via the planar dppz ligand, whereupon π−π stacking interactions between this ligand and the base pairs of the DNA helix enables “molecular light switch” activity [23]. Recently, the effect of systematically changing the steric profile of the ancillary ligands and overall charge of this “parent molecule” (Complex 1, Fig. 42.1) on its fluorescence signal in the presence of dsDNA and ssDNA was reported [24]. It was hypothesized that these ancillary ligand modifications could be used as a handle to tune DNA-NP superlattice bond strength, reflected in the characteristic melting temperature (Tm) of the material.

42.2  Results and Discussion

42.2.1 Thermal Stabilization of the DNA-NP Superlattice “Bond” The potential of using intercalators to thermally stabilize the DNA bonds in a superlattice was studied by performing melt analyses on

Results and Discussion

superlattices incubated with RuII complexes 1−6. Using a family of molecules that possess the same intercalating ligand but different ancillary ligands allowed the relationship between ancillary ligand structure and DNA bond stabilization to be determined. Superlattices were assembled by coating 15 nm diameter AuNPs with a dense monolayer of diisopropylthiol-functionalized oligonucleotides (398 ± 10 strands per particle). These oligonucleotide-coated NPs were then hybridized with 350 equiv of DNA linker strand per particle, resulting in 18 base pair (bp) duplexed regions with unpaired 7 base “sticky ends” at the termini. These particles (25 nM) were then mixed with an equimolar amount of a second species of oligonucleotide-coated AuNP presenting a complementary sticky-end sequence. Upon hybridization of the sticky ends of neighboring particles, the conjugates settled out of solution as an amorphous aggregate. To remove excess unhybridized DNA linker strands, the aggregate was washed with 0.5 M NaCl followed by annealing for 30 min at a temperature slightly below Tm, yielding bcc superlattices. These nanoparticle superlattices were then incubated with the racemic RuII intercalators 1−6 (12.1 μM) in 0.5 M NaCl solution overnight, allowing ample time for intercalation to occur (Scheme 42.1). The number of intercalators bound per available DNA base pair (degree of association) was determined by quantifying the remaining free RuII complexes in the supernatant of the solution using UV−vis spectroscopy after a brief centrifugation. The difference was assumed to correspond with uptake by the DNANP superlattice.

Scheme 42.1 RuII intercalation into DNA-nanoparticle superlattice.a aDNA−functionalized

gold nanoparticles are assembled with linkers and annealed into a crystalline structure. RuII complex is added and the mixture is stored overnight to maximize intercalation.

831

832

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

Figure 42.2  DNA-NP superlattice melting temperatures increase after intercalation and correlate to binding affinity of different intercalators. (a) Superlattice samples (50 nM NP total) incubated in 12.1 μM of different RuII complexes exhibit unique shifts in melting transition. (b) Superlattice melting temperatures after intercalation with complexes 1, 3, and 6 as a function of degree of association. (c) The binding affinity of RuII complexes for one dsDNA bp plotted as a function of ∆Tm of DNA−programmable superlattices.

The effect of intercalation on the melting behavior of the aggregate was studied by heating the solution at a rate of 0.25°C/ min while monitoring the extinction of the solution at 520 nm, the colorimetric signature associated with the gold nanoparticle plasmon resonance (Fig. 42.2a). The Tm of these superlattices was calculated as the inflection point of the melting transition. Addition of any of the complexes at a constant concentration of 12.1 μM noticeably increased Tm, which can be attributed to intercalation of the dppz moiety into the sticky ends of the DNA, as this is the weak point in the DNA duplex and therefore the lattice. Each of the complexes resulted in a slightly different Tm increase, from 1°C for complex 6 to 15°C for complex 1, indicating that intercalator charge and structure significantly influence superlattice thermal stability. This could be due to two factors: either there were a different number of intercalation events occurring at each sticky end or each intercalator complex stabilized dsDNA to a different extent due to its molecular architecture. To determine the contribution of the latter, the Tm values of the DNA-NP superlattice after intercalation with complexes 1, 3, and 6 were compared as a function of degree of association (Fig. 42.2b). This analysis shows that there is a slight contribution from the intrinsic architecture of the molecule on a per intercalator basis, especially at higher degrees of association. This indicates that the large differences in Tm observed in Fig. 42.2a are due to a difference in the number of molecules intercalating,

Results and Discussion

which correlates to the binding affinity of the complex for DNA. To further explore the relationship between intercalator structure and its binding affinity (K) for free duplex DNA, fluorescence titration was performed to measure K. If K values of members of this family of intercalators are compared in a systematic fashion, the relative contributions of the ancillary ligand could be determined. While the binding modes of different intercalating ligands are well characterized in the literature, the effect of extended ancillary ligands on the intercalation behavior of RuII complexes into the DNA duplex has yet to be explored extensively [22, 25, 26]. Our results revealed that the structure and electrostatic charge of the ancillary ligand had a dramatic effect on K. For this family of RuII intercalators, simply changing the number and the nature of the pendant group, which altered the overall charge of the complex, affected K by 2 orders of magnitude at low salt conditions. At higher salt concentrations, electrostatic interactions between the negatively charged DNA backbone and the positively charged complex are screened. A trend of decreasing binding affinity for complexes with more sterically hindered ancillary ligands was observed. The local charge of the pendant group had a more modest effect on binding affinity; complexes with more negatively charged pendant groups exhibited lower binding affinity for DNA. To determine if these trends translated to thermal stabilization of the superlattice, binding affinities were correlated with ∆Tm,superlattice upon intercalation in 12.1 μM complex solution (Fig. 42.2c). A positive, linear relationship between the binding affinity and ∆Tm was observed, revealing that complexes with high binding affinity stabilized superlattices to the greatest extent. This result illustrates that, by modulating the molecular structure of the intercalator, the stability of DNA bonds in superlattices can be tuned in a rational manner. Furthermore, binding affinity of a complex for free dsDNA can serve as an indicator for the degree of bond stabilization. Binding assays confirmed that the intercalators were equilibrated into the DNA-NP superlattice with comparable affinity as for free DNA, providing evidence that the environment of the superlattice does not prevent diffusion of the complexes.

833

834

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

Figure 42.3  Structural effects of intercalation emerge from SAXS and SEM. (a) Scattering patterns of the superlattice in increasing concentrations from 0 to 37.3 μM of complex 1. (b) Isotropic strain resulting from uniform lattice expansion and (c) RMS microstrain induced in superlattice upon intercalation. (d) DNA-NP superlattice single crystals retain structure upon intercalation. Before and after intercalation with complex 1. Scale bars: 2 μm.

As spectroscopic techniques can only provide information about whether particles are bound, small-angle X-ray scattering (SAXS) was used to gain insight into the positioning of nanoparticles within a superlattice after intercalation and during the heating and melting processes. SAXS is a powerful technique analogous to powder diffraction that provides information about the crystal symmetry,

Results and Discussion

interparticle distance (the distance between nearest nanoparticle neighbors), and domain size of a material. By comparing radially averaged one-dimensional SAXS data, any structural changes in a lattice can be determined. To probe whether intercalated superlattices retained their structure at higher temperatures, in situ SAXS was used. The annealed and intercalated samples (50 nM NP in 40 μM intercalator) were heated at a rate of 1°C/min and scans were taken every 2°C. The characteristic bcc diffraction peaks of the unintercalated control sample disappeared at 46°C (designated Tmax), indicating a dissociation of the nanoparticles due to dehybridization of the DNA bonds. Superlattices that had been incubated with intercalator remained intact at higher temperatures (close to the boiling temperature of water for complex 1). Importantly, the Tmax trends between the complexes mirrored those of Tm from the UV− vis melt data. The absolute differences between Tm and Tmax can be attributed to the different ramp rates used in these two techniques. These results confirm the idea that intercalation strengthens the DNA bonds between nanoparticles against thermal destabilization and that intercalator molecular structure is an important factor. In summary, we have demonstrated that increasing complex binding affinity for dsDNA, which changes depending on the charge and steric profile of the ancillary ligands of the intercalator, translates to increased bond strength. A wide range of thermal stabilities was achieved by intercalation with this family of complexes, suggesting the versatility of using this strategy in postsynthetic modification of DNA-NP superlattices.

42.2.2  S tructural Effects of Intercalation on the DNA-NP Superlattice

To fully characterize the structural effect of intercalation on both the superlattice bond and the overall material, SAXS and scanning electron microscopy (SEM) were used. First, nanoparticles were functionalized and aggregated as previously mentioned. After aggregation, the amorphous assembly was slow cooled from 55 to 25°C at a rate of 0.01°C/min. This yielded micrometer-sized bcc superlattice single crystals. Each of these superlattice samples was titrated with 0.1−40 μM of RuII complex and then stored overnight

835

836

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

(ambient temperature, shaking at 750 rpm) before being measured with SAXS. The data (Fig. 42.3a) revealed a gradual shift in the q0 peak (from which interparticle distance is calculated) with increasing quantities of complex 1. Notably, this peak shift occurred without changing the overall bcc character of the structure factor. These changes in interparticle distance (i.e., bond length) can be described in terms of isotropic strain within the superlattice material. Figure 42.3b shows the calculated isotropic strain (η = ∆L/L0), or change in interparticle distance (∆L) as a fraction of the original value (L0), after intercalation with complexes 1 and 6. Isotropic strain increased linearly at the same magnitude (2.9 Å/intercalation event) for both complexes before plateauing upon reaching 0.5 degrees of association, or the saturation point of possible intercalation sites. Since complexes 1 and 6 affect superlattice structure in an identical manner, it can be concluded that structural changes in duplexed DNA upon intercalation are solely dependent upon the intercalating dppz ligand and independent of the ancillary ligands. In this particular context, the lengthening bond can be correlated with an increase in bond strength, in contrast with atomic systems, where increased bond strength between any two atoms correlates with a decrease in bond length, e.g., a C−C triple bond is both shorter and stronger than C−C double bond.

Scheme 42.2  Modulation of superlattice bond strength with RuII intercalation enables stepwise synthesis of core−shell crystals.a aDNA−AuNP

superlattice single crystals are intercalated with complex 1 overnight before being combined with complementary DNA−QdNPs. Slow cooling allows the DNA−QdNPs to assemble into a shell around core crystals. The presence of the intercalator increases the melting temperature of the AuNP core crystal, allowing it to stay intact during the annealing process that leads to shell formation.

To probe local defects in the system introduced by intercalation, microstrain in the superlattices was quantified by analyzing peak

Results and Discussion

broadening in the scattering patterns. Deconvoluting the effects of microstrain from peak broadening due to domain size effects can be done by performing a Williamson–Hall peak shape analysis. In the DNA-NP superlattice, microstrain represents the distribution of DNA lengths throughout a crystal. Figure 42.3c shows the fractional change in root-mean-square (RMS) microstrain as a function of degree of association. It is evident that the initial addition of intercalators increases the microstrain up to a maximum (4.5× the initial microstrain), after which the strain decreases. One possible explanation is that the initial distribution of intercalators (and thus elongation of DNA) throughout the superlattice crystal is not entirely uniform, and the crystal exhibits some defects as the particles move relative to their positioning in a perfect bcc lattice. As additional intercalators are added, a slightly more uniform distribution of particle spacing is evidenced by a slight reduction in the microstrain, a result that we rationalize as arising from the smoothing of the statistical variation of intercalation events as the maximum loading is approached.

42.2.3  Synthesis of Core−Shell Superlattice

As a demonstration of the effectiveness of using intercalators to increase the bond strength of DNA-NP superlattices, a core− shell material was synthesized. Previously explored strategies for assembling these materials have been based on methods such as layer-by-layer adsorption using electrostatics [27−29]. The use of DNA-assembly to create such structures would enable programmable design and synthesis of materials where each component is positioned in a precise manner. In a method analogous to seed-mediated growth [30, 31], an intercalated superlattice crystal was coated with a shell of another type of DNA-nanoparticle superlattice through subsequent thermal annealing steps (Scheme 42.2). The presence of the intercalator was crucial for keeping the core superlattice intact during the thermal annealing steps of the secondary growth step. This postsynthetic modification circumvented the need to redesign the DNA strands which could prevent the nanoparticles from reorganizing into their thermodynamic product [16]. SEM confirmed that the rhombic dodecahedron crystal habit was preserved after postassembly intercalation (Fig. 42.3d).

837

838

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

Rhombic dodecahedron bcc superlattice single crystals were intercalated with complex 1 and then combined with complementary DNA−QdNPs. This mixture was slowly cooled from 45 to 25°C with additional cycling steps around the Tm of the DNA−QdNPs at a rate of 0.01°C/min (Scheme 42.2). The product was embedded in silica and studied using transmission electron microscopy (TEM). Figure 42.4 shows the resulting core−shell structures, which demonstrate the successful assembly of a superlattice shell of lower electron density (CdSe/ZnS) around a core single crystal of higher electron density (Au). These structures differ strikingly from TEM images of independently crystallized core and shell crystals. SAXS and fast Fourier transforms (FFT) of a TEM image reveal that the QdNP shell is, in fact, crystalline. When a large excess of DNA−QdNPs was added, single crystals of only DNA−QdNPs formed in addition to the core− shell structures.

Figure 42.4  Core Au−Shell QdNP crystals. TEM images of micron sized core− shell structures synthesized in a stepwise manner. A difference in electron densities between AuNPs (denser) and QdNPs (lighter) results in contrast between the core and shell portions. Scale bars: 1 μm.

Methods

42.3 Conclusion We have demonstrated the effectiveness of using a family of RuII intercalators to increase the DNA bond strength and length of DNA− programmable nanoparticle superlattices. Through rational design of the intercalator ancillary ligands and choice of concentration, the thermal stability of the superlattice was tunable over a wide range. This method was used to increase the strength of the DNA bonds holding DNA programmable materials together, enabling the synthesis of a core−shell superlattice structure. The ability to control the thermal stability of nanoparticle assemblies is important for the creation of tailorable and functional materials. Such tunability in bond strength greatly expands the synthetic capabilities of using DNA-functionalized nanoparticles as building blocks in bottom up assembly of functional materials and provides a foundation for the assembly of increasingly sophisticated structures using DNAmediated nanoparticle crystallization for plasmonic, photonic, and catalytic applications.

42.4 Methods

42.4.1 RuII Complex Synthesis The full synthesis, and chemical and photophysical characterization for the RuII complexes (Fig. 42.1) were performed as described by Shade and Kennedy et al. [24].

42.4.2  Nanoparticle Functionalization

The oligonucleotides used in this work were synthesized on an MM48 solid-support automated DNA synthesizer (BioAutomation) using standard phosphoramidite chemistry and reagents purchased from Glen Research. All DNA was purified using reverse phase HPLC (Varian ProStar 210) and characterized using matrix-assisted laser desorption ionization time-of-flight mass spectrometry to confirm that the molecular weights matched their theoretical mass.

839

840

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

42.4.2.1  AuNP functionalization Gold nanoparticles were DNA functionalized as described previously [3]. Briefly, citrate capped AuNPs from Ted Pella were functionalized with a dense nucleotide shell. The 5¢-dimethoxytritylhexyl-mercaptan protecting group of the thiolated DNA strand was treated for 45 min with 100 mM dithiolthrietol (DTT, pH 8, SigmaAldrich) and then desalted on a NAP5 size exclusion column (GE Life Sciences). The purified thiolated strands were added to 15 nm diameter AuNPs (4 nmol DNA per mL nanoparticles). After 30 min, the solution was brought up to a total concentration of 0.01% sodium dodecyl sulfate. To increase the density of DNA on the nanoparticles without causing them to lose colloidal stability, 5 M sodium chloride (NaCl) was added stepwise over the course of several hours to reach a final concentration of 0.5 M NaCl. Each addition of salt was followed by 10 s of sonication. Following the salting procedure, the solution was placed on a shaker at 140 rpm and 37°C overnight. The functionalized particles were purified by three rounds of centrifugation (14 000 rpm for 45 min). After each round, the supernatant was removed and the particles resuspended in Milli-Q water. Particle and DNA concentrations were measured on a Cary 5000 UV−vis spectrophotometer (Agilent). Extinction coefficients for the AuNPs and oligonucleotides were obtained from Ted Pella and the IDT Oligo Analyzer, respectively. Particle and DNA concentrations were measured on a Cary 5000 UV−vis spectrophotometer (Agilent). Extinction coefficients for the AuNPs and oligonucleotides were obtained from Ted Pella and the IDT Oligo Analyzer, respectively.

42.4.2.2  Quantum dot NP functionalization

CdSe/ZnS QdNPs (Ocean Nanotech) were polymer coated and DNA functionalized as described previously [10]. To polymer coat the QdNPs, nanocrystals were dissolved in chloroform (0.1 μM) and heated to 55°C, after which N3−PMAO in chloroform (0.2 M monomer units) was added. After a gentle shaking period of 2 min, the solution was allowed to cool to room temperature. Solvent was evaporated using a rotary evaporator. Sodium borate buffer (75 mM, pH 9) was added to cover the nanoparticle layer in the flask, and

Methods

then the mixture was sonicated in ice for 60 min. The solution was then filtered through 0.22 μm cellulose acetate syringe filter. This solution was then concentrated in a 100 kDa cellulose membrane spin filter (Amicon Ultra-15, Millipore) and then ultracentrifuged at 100 000 rcf in a continuous sucrose gradient (density 10% to 60% w/v) for 2 h, or until the nanoparticle band had separated from the floating layer of pure polymer. The layer of nanoparticles was carefully taken out using a syringe and then washed three times through 100 kDa spin filters using deionized water. These coated QdNPs were then functionalized with DBCO-modified DNA strands by combining them at 10-fold excess DNA at 0.5 M NaCl, shaking at room temperature (RT) and 750 rpm. After 5 h, the concentration of salt was increased to 0.5 M NaCl.

42.4.3  Superlattice Assembly

DNA−AuNPs were assembled into aggregates by combining 25 nM each of AuNPs functionalized with sequences A and B with 350 equiv per particle of each linker nucleotide in 100 μL of 0.5 M NaCl. After aggregation at room temperature, the amorphous assembly was heated to a temperature slightly below the melting temperature (44°C for this system) for 30 min. This led to the formation of polycrystalline bcc superlattices. To form single-crystal bcc rhombic dodecahedron superlattices, the amorphous aggregates were transferred to PCR tube strips and slowly cooled in a Life Technologies PCR Thermocycler from 55°C through the melting temperature to 25°C at a rate of 0.01°C/min [12].

42.4.4  Core−Shell Synthesis

DNA−AuNP superlattice rhombic dodecahedra single crystals (10 nM) were intercalated in 40 μM complex 1 overnight (0.5 M NaCl, 1× PBS). Aggregates of complementary bcc DNA−QdNPs were added (50 and 100 nM total) and subjected to a thermal cycling procedure that slowly decreased from a temperature at least 5°C above Tm,QdNP, and upon reaching Tm,QdNP, cycled five times to Tm,QdNP − 1°C before cooling to room temperature. The ramp rate stayed constant at 0.01°C/min.

841

842

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

42.4.5  Fluorescence Binding Assay Binding isotherms for each complex into free dsDNA were obtained by titrating a fixed quantity of intercalator complex with duplex DNA and measuring the fluorescence response. Duplex DNA was prepared by combining equimolar amounts of 18 bp sequence C with linker C in 5 mM Tris-HCl (pH 7.2) and 50 mM NaCl. This solution was heated to 70°C for 5 min before allowing it to cool down slowly through the melting temperature, allowing the strands to hybridize. The amount of intercalator complex was fixed at 1 μM and this was titrated with dsDNA ranging in concentration from 10−9 to 10−2 M bp (corresponding with 5.6 × 10−11 to 5.6 × 10 −4 M oligomer). For the assays performed at 0.5 M NaCl, additional salt was added. Each point in the titration curve was prepared as an individual sample and allowed 30 min to equilibrate before the measurement was taken. Performing a binding assay by titrating dsDNA into a single sample should yield the same results, as no photobleaching was observed for a given sample after running 20 consecutive scans. Fluorescence measurements were performed using a Horiba Jovin-Yvonne Fluorolog fluorometer. The sample temperature was maintained at 25°C and the Peltier chiller for the detector was set to −20°C. The sample was excited at the wavelength of 450 nm, and the emission was measured at 633 nm, followed by background subtraction. The observed fluorescence is assumed to be the sum of the weighted contributions of free and bound ligand: F = F0(Ct − Cb) + FbCb

where F is the apparent fluorescence at each DNA concentration measured at 633 nm, F0 is the fluorescence of the free ligand only, Fb is the fluorescence of the bound species, Ct is the concentration of total ligand, and Cb is the concentration of bound ligand. The following relationship is used to determine K of an intercalator for one DNA bp, derived from a study of the intercalation of DNA with the small molecule intercalator proflavine [32]:

F - F0 K [DNA ] = Fb - F0 1 + K [DNA ]

Fluorescence binding isotherms were fit to this equation using nonlinear least-squares, from which the binding affinities were calculated.

Methods

42.4.6  Absorbance Binding Assay To ensure that the environment of the DNA−programmable superlattice and its constituent DNA−functionalized nanoparticle did not drastically reduce the affinity of these complexes for dsDNA, we obtained binding isotherms of the complexes into both the superlattice and colloidal DNA−AuNP to compare to that of free DNA. RuII complex was added to solutions containing 0.5 pmol AuNPs total of either superlattices or AuNPs functionalized with sequence B and loaded with linker B in 0.5 M NaCl. The final concentrations of RuII complex in these solutions ranged from 2 to 60 μM. This mixture was stored overnight at RT shaking at 750 rpm. The samples were subsequently vortexed and centrifuged. The DNA-NP superlattice samples were centrifuged for 10 s at 6000 rpm using a microcentrifuge, while the DNA−AuNP samples were centrifuged for 1.5 h at 15 000 rpm. The absorbance of the supernatant of each of these samples was then measured using a Cary 5000 UV−vis-NIR spectrophotometer (Agilent) and the amount of complex free in solution (Cfree) was quantified. This calculation used an extinction coefficient of 12 400 M−1 cm−1 at 445 nm corresponding with the absorbance of the complex’s metal to ligand charge transfer [33]. The number of intercalators bound per available DNA base pair (degree of association) was calculated as follows:



Degree of association =

Rtotal = NP ¥

C total - C free Rtotal

350 oligomers 21.5 bp ¥ NP oligomer

where C refers to the amount of ligand (RuII complex), Rtotal refers to the total number of binding sites, and NP refers to the total moles of nanoparticle in a sample. We make the assumption that, due to the high salt concentration of the solution, all 350 linker oligonucleotides added to the sample hybridize to their complement sequence. Intercalation into dsDNA oriented and packed on a nanoparticle is a very different system than those described in existing models of DNA intercalation [34]. Since developing a model to accurately describe such system is beyond the scope of this work, we used the model for a single binding site with a univalent receptor in order to extract an average K that describes the affinity of a complex for one

843

844

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

bp averaged over all types of binding sites on these AuNPs. Thus, we fit the binding isotherm data to the following relationship using nonlinear least-squares:

B=

Ligand Bound K [L] = Receptor Bound 1 + K [L]

The resulting binding affinities of each of these complexes into DNA-programmable superlattices are very similar to their binding affinity into the SNAs.

42.4.7  Melting Transition Measurements

AuNPs (1 pmol total) were functionalized and assembled into a bcc superlattice as described above. A defined volume of RuII complex was added to these bcc superlattices in a total volume of 1 mL of 0.5 M NaCl aqueous solution. After the samples were stored overnight at RT shaking at 750 rpm, each solution was transferred to a quartz cuvette containing a stir bar. The melting behavior of the aggregate was studied using UV−vis spectrometer. The extinction of the solution was monitored at 520 and 260 nm while heating the solution from 25 to 85°C at a rate of 0.25°C/min. The melting temperature is calculated from the point of inflection on the curve where the extinction at 520 nm increases due to the aggregate dissociating.

42.4.8  Small-Angle X-ray Scattering

SAXS experiments were performed at sector 5-ID of the DuPontNorthwestern-Dow Collaborative Access Team (DND-CAT) beamline at the Advanced Photon Source (APS) at Argonne National Laboratory. X-rays of wavelength 1.24 Å (10 keV) were used to probe the sample and the scattering angle was calibrated against a silver behenate standard. All SAXS data was collected at RT. To perform a SAXS measurement, roughly 80 μL of sample was loaded into 1.5 mm quartz capillary (Charles Supper Co.) and placed into a sample stage in the X-ray beam path. Two sets of slits were used to define and collimate the X-ray beam, and a pinhole was used. The X-ray beam cross section measured 200 μm in diameter and a 0.5 s exposure time

Methods

was used. A CCD area detector was used and dark current frames were subtracted from all data. To obtain 1D data, two-dimensional SAXS patterns were azimuthally averaged and relative scattering intensity was plotted as a function of scattering vector q, where q is 4p sin(q ) l where θ is the scattering angle and the λ is the wavelength of X-ray radiation. The structure factor was determined using the relationship:





q=

I(q ) =

S (q ) F (q )

where I(q) is intensity, S(q) is structure factor, and F(q) is the form factor. The particle form factor is due to individual dispersed particles solution and can be determined from a suspension of free particles.

42.4.9  Isotropic Strain upon Intercalation

Nearest-neighbor distance (L, in nm) between the nanoparticles in the lattice is determined from the position of the first-order scattering peak, q0 [110] in a SAXS pattern. For bcc symmetry, this equation is used:



L=

1 6-1/2 p 10 q0

Rise per base pair is calculated based on the following equation:

Rise L0 - rA - rB - 0.75x - 0.8 = bp n where L0 represents the nearest-853 neighbor distance prior to intercalation, rA and rB are the core radii of particles functionalized with sequences A and B, respectively, x is the number of hexaethylene glycol phosphate flexors (sp18) multiplied by a calculated 0.75 nm per sp18 for our system, n is the number of base pairs in between adjacent particles, and 0.8 nm accounts for the combined contribution from the hexylalkyl-thiol moieties on DNA sequences A and B.

845

846

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

To determine the effect of intercalation on DNA elongation, RuII complex was added to DNA−programmable single-crystal superlattices (0.5 pmol NP total). The concentrations of RuII complex ranged from 0.1 to 40 μM and these samples were stored overnight (RT, 750 rpm). The scattering pattern was taken for each sample and the isotropic strain (ηisotropic) calculated from

hisotropic =

L - L0 L0

where L0 is as defined above and L is the interparticle distance after the intercalation event. When ηisotropic is plotted against degree of association (Fig. 42.3b), a linear regime is observed from 0 to 0.3 degrees of intercalation. This initial slope (k) represents the steady increase in strain as a function of degree of association. From this, we can calculate the average rise per intercalation event using the following equation:

k ¥ L0 Rise = Intercalation Event n

42.4.10  Williamson−Hall Analysis To perform a Williamson−Hall analysis, S(q) from scattering patterns was fit to Lorentzian profiles. The integral breadth of each peak (β) was obtained and microstrain was calculated using a relationship that combines the Scherrer equation (size broadening) and the Stokes−Wilson equation (strain broadening):

kl + e sin(q ) t where θ is the scattering angle, β is the integral breadth of the scattering peak, λ is the wavelength of X-ray radiation, τ is the crystal size in the direction perpendicular to the beam, ε is the apparent strain, and k is the shape factor, which is approximated as 0.9 for spherical particles. This can be rewritten as

b cos(q ) =

1 + eq D where β* = β cos θ/λ and q = 4 sin θ/λ. Plotting β* versus q should result in a straight line. The size of the crystallite (D) can be extracted from the y-intercept and the microstrain (ε) can be extracted from the slope [35−37].



b* =

Methods

42.4.11  Scanning Electron Microscopy DNA-NP superlattices were transferred to the solid state following the previous literature to allow for scanning electron microscopy (SEM) visualization [38]. This method encapsulates solution phase single-crystal superlattices in silica. First, the slow-cooled crystals were transferred to a 1.5 mL Eppendorf tube and the volume was brought up to 1 mL with 1× phosphate buffered saline. The quaternary ammonium salt N-trimethoxysilylpropyl-N,N,Ntrimethylammonium chloride (TMSPA, Gelest, Inc.) was added in 1000-fold excess relative to the amount of phosphates in the aggregates (2 μL) to the solution containing the crystals. The tube was placed on a thermomixer (Eppendorf) to shake for 10 min at 750 rpm (RT) to allow the quaternary ammonium to associate with the DNA phosphate backbone. Then, 4 μL of triethoxysilane (TES) was added to the solution to initiate silica growth. The mixture was left shaking at RT and 750 rpm for 4 days. At that point, the crystals had aggregated at the bottom of the tube with a cloudy precipitate of silica formed throughout the solution. This supernatant was carefully removed and discarded. The embedded crystals were then transferred to a new eppendorf tube and dispersed in Milli-Q water. This mixture was purified with three rounds of centrifugation (5 min, 3000 rpm), removal of supernatant, and resuspension in Milli-Q water. Following the final round of centrifugation, the supernatant was removed and the pellet was dried, resuspended, and vortexed in 300 μL of ethanol. This solution (10 μL) was drop-cast onto a silicon wafer (NOVA Electronic Materials) for imaging. SEM images were obtained at the Northwestern University Atomic and Nanoscale Characterization Experimental Center (NUANCE) using a Hitachi SEM SU8030. Images were acquired using an accelerating voltage of 10 kV.

Acknowledgments

We thank J. Griffin for helpful discussions and for providing coding expertise for curve fitting analysis. This material is based upon work supported by AFOSR Award FA9550-11-1-0275, and the Centers of Cancer Nanotechnology Excellence (CCNE) initiative of

847

848

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

the National Institutes of Health (NIH) under Award U54 CA151880. SAXS experiments were carried out at Sector 5-ID of the DuPontNorthwestern-Dow Collaborative Access Team at the Advanced Photon Source. Use of the Advanced Photon Source at Argonne National Laboratory was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. This work made use of the EPIC facility (NUANCE Center-Northwestern University), which has received support from the MRSEC program (NSF DMR-1121262) at the Materials Research Center; the International Institute for Nanotechnology (IIN); and the State of Illinois, through the IIN. M. Wang gratefully acknowledges a Graduate Research Fellowship from the National Science Foundation (NSFGRFP) and a Northwestern University Ryan Fellowship. J. Rouge acknowledges a postdoctoral fellowship from the PhRMA foundation. K. Brown gratefully acknowledges support from Northwestern University’s International Institute for Nanotechnology.

References

1. Desiraju, G. R. (1995). Angew. Chem. Int. Ed. Engl., 34, 2311−2327.

2. Jones, M. R., Seeman, N. C. and Mirkin, C. A. (2015). Science, 347, 204−208. 3. Macfarlane, R. J., Jones, M. R., Senesi, A. J., Young, K. L., Lee, B., Wu, J. and Mirkin, C. A. (2010). Angew. Chem., 122, 4693−4696.

4. Macfarlane, R. J., Lee, B., Jones, M. R., Harris, N., Schatz, G. C. and Mirkin, C. A. (2011). Science, 334, 204−208.

5. Macfarlane, R. J., Jones, M. R., Lee, B., Auyeung, E. and Mirkin, C. A. (2013). Science, 341, 1222−1225.

6. Jones, M. R., Macfarlane, R. J., Lee, B., Zhang, J., Young, K. L., Senesi, A. J. and Mirkin, C. A. (2010). Nat. Mater., 9, 913−917. 7. Auyeung, E., Cutler, J. I., Macfarlane, R. J., Jones, M. R., Wu, J., Liu, G., Zhang, K., Osberg, K. D. and Mirkin, C. A. (2012). Nat. Nanotechnol., 7, 24−28.

8. Senesi, A. J., Eichelsdoerfer, D. J., Macfarlane, R. J., Jones, M. R., Auyeung, E., Lee, B. and Mirkin, C. A. (2013). Angew. Chem. Int. Ed., 52, 6624−6628.

References

9. Zhang, Y., Lu, F., Yager, K. G., van der Lelie, D. and Gang, O. (2013). Nat. Nanotechnol., 8, 865−872.

10. Zhang, C., Macfarlane, R. J., Young, K. L., Choi, C. H. J., Hao, L., Auyeung, E., Liu, G., Zhou, X. and Mirkin, C. A. (2013). Nat. Mater., 12, 741−746.

11. Park, S. Y., Lytton-Jean, A. K., Lee, B., Weigand, S., Schatz, G. C. and Mirkin, C. A. (2008). Nature, 451, 553−556.

12. Auyeung, E., Li, T. I., Senesi, A. J., Schmucker, A. L., Pals, B. C., de La Cruz, M. O. and Mirkin, C. A. (2014). Nature, 505, 73−77. 13. Auyeung, E., Morris, W., Mondloch, J. E., Hupp, J. T., Farha, O. K. and Mirkin, C. A. (2015). J. Am. Chem. Soc., 137, 1658−1662. 14. Park, D. J., Zhang, C., Ku, J. C., Zhou, Y., Schatz, G. C. and Mirkin, C. A. (2015). Proc. Natl. Acad. Sci. U.S.A., 112, 977−981.

15. Barnaby, S. N., Thaner, R. V., Ross, M. B., Brown, K. A., Schatz, G. C. and Mirkin, C. A. (2015). J. Am. Chem. Soc., 137, 13566−13571. 16. Macfarlane, R. J., Thaner, R. V., Brown, K. A., Zhang, J., Lee, B., Nguyen, S. T. and Mirkin, C. A. (2014). Proc. Natl. Acad. Sci. U.S.A., 111, 14995−15000. 17. Pal, S., Zhang, Y., Kumar, S. K. and Gang, O. (2015). J. Am. Chem. Soc., 137, 4030−4033.

18. Li, T. I., Sknepnek, R., Macfarlane, R. J., Mirkin, C. A. and Olvera de la Cruz, M. (2012). Nano Lett., 12, 2509−2514. 19. Dhakal, S., Kohlstedt, K. L., Schatz, G. C., Mirkin, C. A. and Olvera de la Cruz, M. (2013). ACS Nano, 7, 10948−10959.

20. Zeglis, B. M., Pierre, V. C. and Barton, J. K. (2007). Chem. Commun., 4565−4579. 21. Song, H., Kaiser, J. T. and Barton, J. K. (2012). Nat. Chem., 4, 615−620.

22. Pyle, A., Rehmann, J., Meshoyrer, R., Kumar, C., Turro, N. and Barton, J. K. (1989). J. Am. Chem. Soc., 111, 3051−3058.

23. Friedman, A. E., Chambron, J. C., Sauvage, J. P., Turro, N. J. and Barton, J. K. (1990). J. Am. Chem. Soc., 112, 4960−4962.

24. Shade, C. M., Kennedy, R. D., Rouge, J. L., Rosen, M. S., Wang, M. X., Seo, S. E., Clingerman, D. J. and Mirkin, C. A. (2015). Chem. Eur. J., 21, 10983− 10987. 25. Zhen, Q. X., Ye, B. H., Zhang, Q. L., Liu, J. G., Li, H., Ji, L. N. and Wang, L. (1999). J. Inorg. Biochem., 76, 47−53.

26. Erkkila, K. E., Odom, D. T. and Barton, J. K. (1999). Chem. Rev., 99, 2777−2796.

849

850

Modulating the Bond Strength of DNA-Nanoparticle Superlattices

27. Liang, Z., Susha, A. and Caruso, F. (2003). Chem. Mater., 15, 3176−3183.

28. Graf, C. and van Blaaderen, A. (2002). Langmuir, 18, 524− 534.

29. Dokoutchaev, A., James, J. T., Koene, S. C., Pathak, S., Prakash, G. S. and Thompson, M. E. (1999). Chem. Mater., 11, 2389−2399.

30. Langille, M. R., Zhang, J., Personick, M. L., Li, S. and Mirkin, C. A. (2012). Science, 337, 954−957.

31. Lofton, C. and Sigmund, W. (2005). Adv. Funct. Mater., 15, 1197−1208. 32. Schmechel, D. and Crothers, D. (1971). Biopolymers, 10, 465−480.

33. Liu, J. G., Zhang, Q. L., Shi, X. F. and Ji, L. N. (2001). Inorg. Chem., 40, 5045−5050. 34. McGhee, J. D. and von Hippel, P. H. (1974). J. Mol. Biol., 86, 469−489.

35. Zhao, Y. and Zhang, J. (2008). J. Appl. Crystallogr., 41, 1095−1108.

36. Mittemeijer, E. J. and Welzel, U. (2008). Z. Kristallogr., 223, 552−560.

37. Senesi, A. J., Eichelsdoerfer, D. J., Brown, K. A., Lee, B., Auyeung, E., Choi, C. H. J., Macfarlane, R. J., Young, K. L. and Mirkin, C. A. (2014). Adv. Mater., 26, 7235−7240. 38. Auyeung, E., Macfarlane, R. J., Choi, C. H. J., Cutler, J. I. and Mirkin, C. A. (2012). Adv. Mater., 24, 5181−5186.

Chapter 43

The Significance of Multivalent Bonding Motifs and “Bond Order” in DNADirected Nanoparticle Crystallization*

Ryan V. Thaner,a,b Ibrahim Eryazici,a,b Robert J. Macfarlane,a,b Keith A. Brown,a,b Byeongdu Lee,c SonBinh T. Nguyen,a,b and Chad A. Mirkina,b aDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bInternational Institute for Nanotechnology, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA cX-ray Science Division, Advanced Photon Source, Argonne National Laboratory, 9700 S. Cass Ave., Argonne, IL 60439, USA [email protected]; [email protected]

Ryan V. Thaner and Ibrahim Eryazici contributed equally to this work.

Multivalent oligonucleotide-based bonding elements have been synthesized and studied for the assembly and crystallization of gold *Reprinted with permission from Thaner, R. V., Eryazici, I., Macfarlane, R. J., Brown, K. A., Lee, B., Nguyen, S. T. and Mirkin, C. A. (2016). The significance of multivalent bonding motifs and “bond order” in DNA-directed nanoparticle crystallization, J. Am. Chem. Soc. 138, 6119−6122. Copyright (2016) American Chemical Society.

Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

852

The Significance of Multivalent Bonding Motifs and “Bond Order”

nanoparticles. Through the use of organic branching points, divalent and trivalent DNA linkers were readily incorporated into the oligonucleotide shells that define DNA-nanoparticles and compared to monovalent linker systems. These multivalent bonding motifs enable the change of “bond strength” between particles and therefore modulate the effective “bond order.” In addition, the improved accessibility of strands between neighboring particles, either due to multivalency or modifications to increase strand flexibility, gives rise to superlattices with less strain in the crystallites compared to traditional designs. Furthermore, the increased availability and number of binding modes also provide a new variable that allows previously unobserved crystal structures to be synthesized, as evidenced by the formation of a thorium phosphide superlattice.

43.1 Introduction

Almost two decades ago, we reported that polyvalent, DNAfunctionalized gold nanoparticles (DNA-NPs) can be programmably assembled via sequence-specific DNA interactions to yield macroscopic aggregates [1a] and more recently highly crystalline structures [1b]. These materials exhibit plasmonic properties and cooperative melting behavior that are ideal for a variety of uses in molecular diagnostics [2]. By replacing the nanoparticle core with a rigid organic molecule, we and others have also systematically studied the cooperative binding characteristics of molecularly pure multivalent nucleic acid constructs [3]. While both of these types of DNA conjugates have illustrated the significance of multivalency in cooperative behavior, the effect is highly amplified in DNA-NP assemblies where hundreds of DNA interactions can occur between NPs. Remarkably, depending upon particle−particle contact area, even a single base pair (bp) can stabilize particle assemblies due to the effects of polyvalency [4]. Indeed, the combination of a polyvalent scaffold and a weak individual interparticle interaction strength allows DNA-NPs of arbitrary core composition to behave as “programmable atom equivalents” (PAEs) in the synthesis of hundreds of different crystalline lattices with unique combinations of spacing and symmetry [1b, 5]. Notably, both the NP “atom” and

Introduction

oligonucleotide “bond” can be independently manipulated such that building block identity and bonding behavior can be decoupled. This highly modular platform has prompted us to explore the possibility of creating multiple-bond analogs by incorporating branched organic molecules into the DNA-based bonding elements with the potential for discovering new properties and making new structures (Fig. 43.1A).

Figure 43.1  (A) A schematic representation of NP-based PAEs with tunable bond strengths via branching linkers: the gold NP core (blue and red spheres), a dense surface layer of hexanethiol-modified DNA comprising a [PEG6]2 nonbinding region (gold segment), and an 18-bp sequence (black segment) as well as a DNA linker (red and blue segments). The linker comprises an 18-bp complementary sequence attached to a branching point (gray sphere), a flexor region, and single-stranded “sticky ends” (3¢-TTTCCTT and 3¢-AAGGAAA) that drive NP assembly via DNA hybridization. (B) Increasing melting temperatures were observed for two sets of NPs with different diameters as the “bond order” or number of sticky ends per linker was increased. For each AuNP size, concentrations were identical, and linker loadings were found to be similar.

Herein, we report the synthesis of NP-based PAEs with branched binding motifs capable of forming DNA “multiple bonds” and their use in the formation of low-strain superlattices. By incorporating branched organic molecules into the traditional linker design, a diversity of bonding modalities can be accessed, providing an additional level of polyvalency at the periphery of PAEs. These complex binding motifs manifest in the formation of higher-quality crystals than those obtained via conventional linear designs as well as the previously unobserved thorium phosphide (Th3P4, I43d) lattice symmetry.

853

854

The Significance of Multivalent Bonding Motifs and “Bond Order”

By design, PAEs consist of three components: a NP core, a dense surface layer of thiol-modified DNA, and a “shell” of DNA linkers. In a typical assembly experiment, these DNA-NPs are combined with a DNA linker, which comprises a recognition sequence complementary to the particle-bound strand, an unpaired “flexor” region, and a short, single-stranded region termed the “sticky end” that engages in interparticle hybridization events. In our “multiple-bond” design, branching points are incorporated into the DNA linker after the recognition sequence such that each linker possesses two (doubler) or three (trebler) sticky ends (Fig. 43.1A) [6]. This design gives rise to a form of valency derived from the DNA itself as opposed to using NP shape and dense oligonucleotide packing to direct the DNA bonding elements [7]. It also should be noted that the multivalent linkers studied herein were typically synthesized with a “T3” flexor region consisting of three unpaired thymine bases, as opposed to the traditional single-base design, as this has been shown to be optimal in directing DNA interactions in analogous branched organic-DNA hybrid molecular assemblies [8]. We also investigated variations in this flexor region ranging from a standard unpaired A-base to up to 18 ethylene glycol chains, which have previously been shown to improve the extent of long-range order or relative crystal quality [9]. Both of these modifications to the traditional linker design should increase the accessibility of sticky ends at the periphery of PAEs, which in turn would provide optimal environments to maximize polyvalent interactions, resulting in improved crystal quality.

43.2 Methods

As a starting point, commercially available AuNPs with diameters of 10 and 15 nm were densely functionalized with 5¢-thiol modified ssDNA consisting of a [PEG6]2 nonbonding region and a 18-bp sequence using a previously established salt-aging method [10]. A predetermined amount of the appropriate type of linker was then added to each type of DNA-coated NP. Upon combining the complementary particle types (i.e., trebler with trebler), hybridization events between the sticky ends induced precipitation of the particles into a disordered aggregate. It is

Results and Discussion

important to note that while multivalent linkers are capable of having the sticky ends bind in the ideal fashion as depicted in Fig. 43.1A, there are other potential binding modes (e.g., hybridization between one or two sticky ends of a trebler but not necessarily all three), and it can be assumed that all of these potential interactions collectively contribute to the observed behavior presented herein.

43.3  Results and Discussion

The use of the doubler and trebler linkers greatly improved the thermal stability of the as-synthesized DNA-NP assemblies and dramatically sharpened the melting profiles compared to the traditional systems [11]. When the melting temperatures (Tm) are plotted for all three linker types, a clear trend is observed for each set of NP diameters: the Tm increases with the number of sticky ends per linker (Fig. 43.1B). These data suggest that multivalent stickyend designs can lead to appreciable increases in thermal stability that are analogous to the increase in bond strengths between two partner atoms in a molecule as the bond order increases. (Detailed data and analysis for both Tm and transition breadth can be found in the SI, including the effects of different flexor types. We note that von Kiedrowski and others have shown that multiple hybridization events in DNA-based molecular systems can greatly stabilize the resulting assembled structures and drawn a similar multiplebond analogy [12]. For the purposes of brevity and simplicity, the remainder of this manuscript will focus on the contrast between the linear and trebler systems where the differences in all observed behavior were the most prominent. The notable differences in melting behavior upon the introduction of either branching or flexible units into DNA linkers should have a significant impact on the NP crystallization behavior under annealing conditions. In addition, we hypothesized that having an “accommodating,” high density of sticky ends in the potential interaction volumes of neighboring PAEs would facilitate crystal formation and potentially improve crystal quality. Twocomponent superlattices composed of 10 and 15 nm AuNPs were prepared by annealing the aggregates around their characteristic

855

856

The Significance of Multivalent Bonding Motifs and “Bond Order”

Tm values, as determined in the abovementioned experiments, yielding body-centered cubic (bcc) superlattices when the AuNPs are the same or cesium chloride-type (CsCl) symmetries when the components are of different sizes [5b]. Confirmation of these lattice symmetries as well as other structural information was obtained using synchrotron-based small-angle X-ray scattering (SAXS). Differences in the extent of long-range order are manifested in the sharpness and number of the scattering peaks, where sharper and well-resolved peaks indicate a higher-quality crystal (i.e., those with less strain and where the NPs are not significantly deviated from the ideal positions in the lattices). The scattering pattern for the CsCl structure formed from the trebler system is visibly higher quality than that for the linear linkers (Fig. 43.2A), in accordance with our hypothesis. Such differences can be quantified using a Williamson−Hall [13] (W-H) analysis, which enables the two main sources of peak broadening, finite grain size and microstrain, to be deconvoluted through line-shape analysis. Remarkably, the superlattices composed of trebler-T3 linkers have 7× less strain (strain = 0.7 ± 0.4 × 10−3) than the analogous linear-A systems (strain = 5.1 ± 0.6 × 10−3), suggesting that the multivalent linkers enable PAEs to situate themselves closer to their ideal lattice positions than linear systems. For each of the three lattice symmetries (15 and 10 nm bcc; 15−10 nm CsCl) that we examined, a plot of the average strain as a function of the type of linker employed (Fig. 43.2B, a total 95 analyses) clearly illustrates the lower quality and broad variation of crystals assembled with the traditional linear-A design, where the amount of strain and deviations from sample-to-sample are the most varied compared to its more flexible counterparts. Notably, each data point represents the average of a variety of linker loadings for each system, further illustrating how the incorporation of flexible moieties imparts an “accommodating” environment for heterogeneities across systems. While the trebler branching unit provides a means to generate high-quality crystals in a consistent manner, previous work has demonstrated that the incorporation of flexibility into the PAE design through other chemical modifications also improves crystal quality

Results and Discussion

(see additional discussion in the SI) [9]. When considering the effect of flexibility on crystallization using strictly linear linkers, we also observe that a PEG flexor imparts less strain into the system while simultaneously minimizing the deviations among various samples. This similarity suggests that the incorporation of flexibility into the linker design, in other forms, provides enough “tolerance” to the system to facilitate the particles arranging themselves closer to their ideal positions as well as accommodates sources of polydispersity (e.g., NP size variations, differences in linker loading) without compromising superlattice quality. However, when considering the temperature at which each sample was annealed relative to its Tm, there are clear differences between these two sources of linker flexibility. Specifically, linear systems must be annealed right at, or in some cases, above their Tms in order to achieve high-quality crystals, while trebler systems can be annealed several degrees below their Tm values without any noticeable reduction in crystal quality (Fig. 43.2C) [4b, 9b]. Such a difference is an important practical consideration since aggregate melting typically occurs over a small temperature range and can completely disrupt the structural integrity of the superlattice. Finally, while attempting to assemble a CsCl superlattice with 10 and 15 nm AuNP cores (but comparable hydrodynamic radii) using trebler linkers, we observed a new crystalline phase that corresponds to the I43d space group (Fig. 43.3; the cubic unit cell is isostructural with Th3P4). This unexpected and complex lattice structure was initially observed in atomic systems when lithium was exposed to high-pressure conditions; to our knowledge, this lattice has never been observed in DNA-directed crystallization and only once previously in a colloidal system [14]. This new Th3P4-type structure was consistently observed (over 20× with slight variations in DNA design) and only when using treblers. Compared to the other binary superlattices previously formed using DNA, this structure is significantly more complex with a large unit cell (lattice parameter of 93.59 nm) containing twelve 15 nm and sixteen 10 nm particles as opposed to the lattice parameter of 38.24 nm and one-to-one particle ratio for the analogous CsCl unit cell [5b].

857

858

The Significance of Multivalent Bonding Motifs and “Bond Order”

Figure 43.2  (A) Comparison of SAXS patterns of CsCl-type superlattices (inset = a depiction of unit cell) formed from PAEs with 10 and 15 nm diameter cores. The traditional linear-A system is of lower quality than that assembled with the trebler-T3 linkers. (B) Plot of average amounts of microstrain for (Continued)

Results and Discussion

each lattice symmetry as a function of linker design where each data point encompasses all linker loadings for each system. The incorporation of flexibility enables consistent formation of higher-quality lattices relative to those with the linear-A linker design. (C) Plot of strain values for all CsCl systems based on the temperature at which each system was annealed relative to its Tm (i.e., the difference between the temperature determined by optical melting analysis and that used to anneal the system; in such a plot, systems that appear to have different strains at the same ∆T value are unique assemblies that were annealed at different absolute temperatures). Trebler linkers allow for the formation of high-quality crystals when annealed several degrees below the Tm. In contrast, systems containing linear linkers must be annealed right at or just above their respective Tm values for the crystals to form.

Figure 43.3  SAXS data for the Th3P4 structure observed when using trebler linkers (red, theoretical; black, observed). The unit cell is depicted in the inset with a perspective down the z-axis (blue spheres = 10 nm AuNPs, Wyckoff position of 16c with x = 0.087; red spheres = 15 nm AuNPs, Wyckoff position of 12a). One 15 nm particle is enclosed in a triangle-based distorted dodecahedron composed of eight 10 nm particles.

Interestingly, the inorganic volume fraction of PAEs in the new Th3P4-type lattice is lower than that for CsCl (0.036 vs. 0.041). However, when the volume fraction of only the 15 nm PAE is considered, the new structure is more dense than CsCl (0.035 vs. 0.032). While the previously mentioned melting transitions indicate that replacing linear linkers with branched units increases attractive interactions between individual DNA strands due to multivalent

859

860

The Significance of Multivalent Bonding Motifs and “Bond Order”

binding, this new phase with lower packing density suggests that it also increases repulsive interactions between PAEs as a whole. This is further suggested by the fact that the center-to-center distance between the 10 and 15 nm AuNPs in the new phase is smaller (32.39 nm, compared with the 33.11 nm distance observed in the CsCl structure), meaning the DNA is 3.5% shorter in length. An in situ crystallization experiment was carried out to determine the relative stabilities of these two lattices and the nature of their phase boundary. When monitoring the scattering pattern of the initial disordered aggregate as it transitions to a crystalline arrangement upon annealing, it was observed that both the new phase and CsCl initially formed simultaneously. However, over time, the peaks corresponding to CsCl disappeared and only the scattering pattern of the new phase remained, suggesting that it is the thermodynamically favored structure. Such a result is counter to what is typically observed in these systems where the densest arrangement of PAEs is favored, but is again consistent with the hypothesis of increased repulsive interactions between PAEs. In conclusion, the use of organic branching moieties in the linkers that define PAEs allows for the realization of new building blocks that provide a type of controllable “bond order” in this field. Such structures also allow one to modulate the DNA-based valency in NP assembly, which presents a means to control particle interactions beyond geometric parameters such as particle size or hydrodynamic radius, and has significant consequences on the observed crystallization path. Indeed, such building blocks have now led to the discovery of the Th3P4 lattice symmetry in nanoparticle assembly, one not previously observed in the programmable assembly field. The realization of this new structure is both exciting and challenging as it moves beyond the predictable realm established by previous design rules [5b]. This result points to the need to develop additional computational and experimental tools to accurately model and predict the formation of such complex architectures and, in turn, understand how multivalency changes the crystallization pathway. Moving forward, this new design paradigm will have important implications as research efforts shift to the use of PAEs with low linker density or minimal sites for DNA attachment as well as multitiered structures that necessitate gradients of interaction strengths using the same sequence-specific, sticky-end interactions.

References

Acknowledgments This material is based upon work supported by the following awards: Air Force Office of Scientific Research FA9550-11-1-0275 and FA9550-12-1-0280; the Center for Cancer Nanotechnology Excellence initiative of the National Institutes of Health U54CA151880 and U54CA199091. R.V.T. gratefully acknowledges an NSF Graduate Research Fellowship. This research used resources of the Advanced Photon Source, a U.S. Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory under contract no. DE-AC02-06CH11357.

References

1. (a) Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607. (b) Jones, M. R., Seeman, N. C. and Mirkin, C. A. (2015). Science, 347, 1260901. 2. (a) Cutler, J. I., Zhang, K., Zheng, D., Auyeung, E., Prigodich, A. E. and Mirkin, C. A. (2011). J. Am. Chem. Soc., 133, 9254. (b) Howes, P. D., Chandrawati, R. and Stevens, M. M. (2014). Science, 346, 1247390. (c) Rosi, N. L. and Mirkin, C. A. (2005). Chem. Rev., 105, 1547. (d) Anker, J. N., Hall, W. P., Lyandres, O., Shah, N. C., Zhao, J. and Van Duyne, R. P. (2008). Nat. Mater., 7, 442.

3. (a) Eryazici, I., Prytkova, T. R., Schatz, G. C. and Nguyen, S. T. (2010). J. Am. Chem. Soc., 132, 17068. (b) Prytkova, T. R., Eryazici, I., Stepp, B. R., Nguyen, S. T. and Schatz, G. C. (2010). J. Phys. Chem. B, 114, 2627. (c) Long, H., Kudlay, A. and Schatz, G. C. (2006). J. Phys. Chem. B, 110, 2918. (d) Kudlay, A., Gibbs, J. M., Schatz, G. C., Nguyen, S. T. and Olvera de la Cruz, M. (2007). J. Phys. Chem. B, 111, 1610. (e) Aldaye, F. A. and Sleiman, H. F. (2007). J. Am. Chem. Soc., 129, 13376. (f) McLaughlin, C. K., Hamblin, G. D. and Sleiman, H. F. (2011). Chem. Soc. Rev., 40, 5647. (g) McLaughlin, C. K., Hamblin, G. D., Aldaye, F. A., Yang, H. and Sleiman, H. F. (2011). Chem. Commun., 47, 8925. (h) Meng, M., Ahlborn, C., Bauer, M., Plietzsch, O., Soomro, S. A., Singh, A., Muller, T., Wenzel, W., Brase, S. and Richert, C. (2009). ChemBioChem, 10, 1335. (i) Greschner, A. A., Toader, V. and Sleiman, H. F. (2012). J. Am. Chem. Soc., 134, 14382. 4. (a) Jin, R., Wu, G., Li, Z., Mirkin, C. A. and Schatz, G. C. (2003). J. Am. Chem. Soc., 125, 1643. (b) Macfarlane, R. J., Thaner, R. V., Brown, K. A., Zhang, J., Lee, B., Nguyen, S. T. and Mirkin, C. A. (2014). Proc. Natl. Acad.

861

862

The Significance of Multivalent Bonding Motifs and “Bond Order”

Sci. U.S.A., 111, 14995. (c) Hurst, S. J., Hill, H. D. and Mirkin, C. A. (2008). J. Am. Chem. Soc., 130, 12192.

5. (a) Macfarlane, R. J., O’Brien, M. N., Petrosko, S. H. and Mirkin, C. A. (2013). Angew. Chem. Int. Ed., 52, 5688. (b) Macfarlane, R. J., Lee, B., Jones, M. R., Harris, N., Schatz, G. C. and Mirkin, C. A. (2011). Science, 334, 204. (c) Macfarlane, R. J., Jones, M. R., Lee, B., Auyeung, E. and Mirkin, C. A. (2013). Science, 341, 1222. (d) Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451, 549. (e) Liu, W., Tagawa, M., Xin, H. L., Wang, T., Emamy, H., Li, H., Yager, K. G., Starr, F. W., Tkachenko, A. V. and Gang, O. (2016). Science, 351, 582. (f) Brodin, J. D., Auyeung, E. and Mirkin, C. A. (2015). Proc. Natl. Acad. Sci. U.S.A., 112, 4564. 6. Shchepinov, M. S., Udalova, I. A., Bridgman, A. J. and Southern, E. M. (1997). Nucleic Acids Res., 25, 4447.

7. (a) O’Brien, M. N., Jones, M. R., Lee, B. and Mirkin, C. A. (2015). Nat. Mater., 14, 833. (b) Jones, M. R., Macfarlane, R. J., Lee, B., Zhang, J., Young, K. L., Senesi, A. J. and Mirkin, C. A. (2010). Nat. Mater., 9, 913. (c) Lu, F., Yager, K. G., Zhang, Y., Xin, H. and Gang, O. (2015). Nat. Commun., 6, 6912. 8. Eryazici, I., Yildirim, I., Schatz, G. C. and Nguyen, S. T. (2012). J. Am. Chem. Soc., 134, 7450.

9. (a) Park, S. Y., Lytton-Jean, A. K. R., Lee, B., Weigand, S., Schatz, G. C. and Mirkin, C. A. (2008). Nature, 451, 553. (b) Senesi, A. J., Eichelsdoerfer, D. J., Brown, K. A., Lee, B., Auyeung, E., Choi, C. H. J., Macfarlane, R. J., Young, K. L. and Mirkin, C. A. (2014). Adv. Mater., 26, 7235.

10. Hurst, S. J., Lytton-Jean, A. K. R. and Mirkin, C. A. (2006). Anal. Chem., 78, 8313. 11. (a) Diaz, J. A. and Gibbs-Davis, J. M. (2013). Small, 9, 2862. (b) Díaz, J. A., Grewer, D. M. and Gibbs-Davis, J. M. (2012). Small, 8, 873. 12. (a) Scheffler, M., Dorenbeck, A., Jordan, S., Wüstefeld, M. and von Kiedrowski, G. (1999). Angew. Chem. Int. Ed., 38, 3311. (b) Shchepinov, M. S., Mir, K. U., Elder, J. K., Frank-Kamenetskii, M. D. and Southern, E. M. (1999). Nucleic Acids Res., 27, 3035. (c) Hong, B. J., Cho, V. Y., Bleher, R., Schatz, G. C. and Nguyen, S. T. (2015). J. Am. Chem. Soc., 137, 13381.

13. (a) Scardi, P., Leoni, M. and Delhez, R. J. (2004). Appl. Crystallogr., 37, 381. (b) Williamson, G. K. and Hall, W. H. (1953). Acta Metall., 1, 22. 14. (a) Hanfland, M., Syassen, K., Christensen, N. E. and Novikov, D. L. (2000). Nature, 408, 174. (b) Henzie, J., Grunwald, M., Widmer-Cooper, A., Geissler, P. L. and Yang, P. D. (2012). Nat. Mater., 11, 131.

Chapter 44

Oligonucleotide Flexibility Dictates Crystal Quality in DNA-Programmable Nanoparticle Superlattices*

Andrew J. Senesi,a,b Daniel J. Eichelsdoerfer,a Keith A. Brown,a Byeongdu Lee,b Evelyn Auyeung,c Chung Hang J. Choi,a Robert J. Macfarlane,a Kaylie L. Young,a and Chad A. Mirkina,c aDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA bX-ray Science Division, Advanced Photon Source, Argonne National Laboratory, 9700 S. Cass Ave., Argonne, IL 60439, USA cDepartment of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA [email protected]

Andrew J. Senesi and Daniel J. Eichelsdoerfer contributed equally to this work.

*Reprinted with permission from Senesi, A. J., Eichelsdoerfer, D. J., Brown, K. A., Lee, B., Auyeung, E., Choi, C. H. J., Macfarlane, R. J., Young, K. L. and Mirkin, C. A. (2014). Oligonucleotide flexibility dictates crystal quality in DNA-programmable nanoparticle superlattices, Adv. Mater. 26, 7235–7240. Copyright © 2014, John Wiley and Sons. Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

864

Oligonucleotide Flexibility Dictates Crystal Quality in DNA-Programmable

44.1 Introduction Superlattices composed of noble metal nanoparticles display considerable promise for the development of optical metamaterials [1–5], and architectures with tunable permittivity [3], permeability [4], negative indices of refraction [2, 4, 5], and even some with chiral [5, 6] and active [7] responses have been explored. However, the realization of such architectures using nanoparticles has proven synthetically challenging because these structures typically require both high metallic volume fraction (approaching or exceeding 20%) [3, 8] and large diameter particles (i.e., 30–100 nm). While superlattices have been formed using electrostatic [9], entropic [10] and molecular interactions [11], DNA-mediated assembly [12, 13] offers unprecedented levels of control over the interparticle distance and superlattice symmetry. The development of metamaterials from DNA-nanoparticle superlattices, however, has been impeded by the difficulty in synthesizing periodic structures with suitably high metallic volume fractions, as they are more prone to kinetic jamming and often form non-crystalline arrangements regardless of annealing conditions. Here, we show that incorporating flexible spacers along with annealing DNA-nanoparticle systems close to their melting transitions improves crystalline quality in terms of domain size and microstrain. These insights allow for the crystallization of nanoparticle superlattices with up to 34% volume fraction, up from the previously reported maximum of 14% [14]. Importantly, this paper does not focus on metamaterials per se, but rather on the scientific underpinnings required to assemble large AuNPs into structures with high inorganic volume fractions. In DNA-mediated nanoparticle assembly, nanoparticles are coated with a dense shell of double-stranded DNA that binds specific particles together using programmable hybridization. In this sense, the particles act as “programmable atom equivalents” (PAEs), where each particle behaves as an “atom” with bonding behavior that can be tuned via the DNA interconnects. The synthetic tailorability afforded by DNA allows independent control over the superlattice connectivity and nanoparticle core, thereby enabling the design and synthesis of colloidal crystals with widely varying symmetry, scale,

Introduction

and composition [14–19]. While DNA-mediated assembly is a useful approach for spatially arranging noble metal nanoparticles, this method has not succeeded in forming crystals with short rigid DNA strands, especially for larger diameter particles [14]. This inability to crystallize large particles with small separations is likely a result of kinetic jamming behavior, which is often seen in colloidal systems [20] and may be related to the rigidity of the double-stranded DNA ligands.

Figure 44.1  (A) Scheme showing DNA crystallization employing a flexible oligomer with hexaethylene glycol phosphate (Spacer-18) repeat units in the spacer region. We propose that the flexibility could increase the interaction range of PAEs, which is known to prevent jamming in colloidal systems. The successive addition of a flexible oligomer to the DNA ligand (from B to C) will allow the DNA strands greater conformational freedom, thereby increasing the volume in which the sticky end could be located and widening the interaction range.

Initial evidence for the importance of flexibility came from the observation that certain PAE systems which form amorphous aggregates when linked with fully duplexed DNA will crystallize if a single unpaired base is added to the DNA strands that link the particles [12]. Given that single-stranded DNA has a shorter persistence length

865

866

Oligonucleotide Flexibility Dictates Crystal Quality in DNA-Programmable

than double-stranded DNA, this can be rationalized by considering that adding some degree of flexibility may be required to allow the packing of the inherently polydisperse particles into a regular lattice. Further, it is well known that the phase behavior of colloids depends on their range of interaction; long-range interactions result in a “softer” pair-potential that may allow for crystallization, while shortrange interactions often result in jamming and gel formation [21]. Based on these observations, we hypothesize that adding flexibility to the DNA ligand could allow each strand to explore a wider space, thereby increasing the interaction range and decreasing jamming during PAE crystallization. Herein, we explore the relationship between crystallization and linker flexibility by systematically adding various lengths of a flexible oligomer to the DNA ligand (Fig. 44.1). Through annealing at different temperatures, we find that the highest-quality crystals (defined here as crystals with the largest grain sizes and lowest microstrain as measured by X-ray diffraction) are formed at temperatures close to the melting temperature. Significantly, increasing ligand flexibility increases the temperature range under which crystallization occurs while simultaneously minimizing microstrain and maximizing domain size.

44.2  Methods and Discussion

In a typical assembly experiment, PAE superlattices were observed by correlated synchrotron small-angle X-ray scattering (SAXS) and UV–vis spectroscopy (Fig. 44.2), a set of measurements that allowed for the simultaneous determination of structural and thermodynamic information. Specifically, gold nanoparticles (AuNPs) were coated with a dense, oriented monolayer of an oligomer composed of a thiolated “spacer” region (either a dA10 sequence or a hexaethylene glycol phosphate oligomer), an 18 DNA base duplexed region, a hexaethylene glycol phosphate monomer (or “flexor”), and an unpaired 7 DNA base “sticky end” (Fig. 44.1A). A solution of these particles in phosphate buffered saline (0.5 M NaCl) was mixed with an equal concentration of a second particle species that had been coated with a similar oligomer with a complementary sticky-end sequence. After mixing, the system was allowed to fully

Methods and Discussion

aggregate at room temperature, forming an amorphous material. The samples were subsequently annealed at various temperatures (the “annealing temperature”) and for various times (the “annealing time”), typically 12 h. Annealing resulted in a variety of structural changes including a disorder-to-order transition, decrease in microstrain, and increase in grain size (vide infra); we note that our definitions of “annealing temperature” and “annealing time” refer to experimentally variable parameters and not transition temperatures associated with each structural change. Without lowering the temperature, the supernatant (including any free particles) was then removed from the aggregates. The supernatant particle concentration was determined by UV–vis spectroscopy to give the fraction of free particles, yielding a melting curve where each point was obtained from a distinct sample (Fig. 44.2B). The aggregates were subsequently examined by SAXS (Fig. 44.2C). Corroborating structural evidence was provided by transmission electron microscope (TEM) imaging of samples embedded in silica [22], a technique which preserves the configuration of the sample upon removal from solution. As an initial experiment, we examined the effect of temperature on the crystallinity of a body centered cubic (BCC) system, comprised of 8.7 nm diameter AuNP cores. Melting and SAXS data are presented in Fig. 44.2B and C, respectively. All the SAXS patterns correspond to BCC symmetry, with various degrees of ordering (Fig. 44.2C). As the annealing temperature increases, the scattering peaks appear sharper and more defined, indicating that the samples are forming larger, more ordered crystallites [24]. Interestingly, the scattering spectra showed no change for annealing times between 15 min and 12 h, indicating that structural changes occur at relatively short timescales (10°C); however, this range severely decreases as the nanoparticle size increases. For example, 38.1 nm diameter AuNPs with the same DNA sequence do not crystallize until the sample is annealed at 1°C below Tm (Fig. 44.3). We hypothesized that adding flexibility into the DNA system using a hexaethylene glycol phosphate group (Glen Research Corporation, “Spacer-18,” SP18) would widen the annealing window. Although SP18 contains a charged phosphate moiety, its persistence length in buffers with high ionic strength (e.g., 0.5 M NaCl used here) was found to be the same as polyethylene glycol [26] (Lp = 2.8 Å) which is substantially less than single [27] (Lp = 7.5 Å) or double [28] (Lp = 500 Å) stranded DNA. Ligand flexibility was explored by adding various subunits of SP18 to the DNA at the flexible spacer region (Fig. 44.1), annealing a series of samples at various temperatures near Tm, and examining the structures by SAXS. The effect of adding the flexible spacer to the DNA sequence can clearly be seen in the annealing curves (Fig. 44.3A). As up to 10 SP18 units are added to the spacer region, the annealing curves broaden and shift to lower temperatures relative to Tm. Grain size and microstrain for each sample type at Tm were compared by linearly interpolating data points of samples annealed just above and just below the melting temperature, using Williamson–Hall analysis (Fig. 44.3B). Generally, the grain size increases monotonically

Methods and Discussion

with flexibility while microstrain decreases. However, for samples with 15 SP18 units, an increase in microstrain was observed, and at 20 SP18 units, two phases could be distinguished consisting of crystal regions surrounded by an amorphous phase. This could be explained in light of the earlier observation that outlier particles will segregate at grain boundaries coupled with the fact that larger diameter particles are generally more polydisperse.

Figure 44.3  Adding flexibility to the DNA ligands connecting adjacent particles allows for the formation of higher-quality crystals. (A) Annealing curves for 38.1 nm diameter AuNP cores with varying ligand flexibility plotted as a function of Tm −T, determined by SAXS. (B) Grain size and microstrain for each system at Tm, linearly interpolated from Williamson–Hall analysis of samples annealed just below and just above Tm.

The flexible SP18 spacers could have several effects. First, the SP18 results in a denser DNA shell compared to a dA10 spacer [29], possibly allowing more polyvalent interparticle interactions during annealing. Second, the increased flexibility allows polydisperse particles to pack into a regular array. Third, flexible ligands should be able to explore a larger volume surrounding the particle (Fig. 44.1B,C), thereby widening the interaction pair potential. In addition, the local concentration of DNA sticky ends would decrease, which should depress the melting temperature. Indeed, as ligand flexibility increases, the melting temperature decreases. It should be noted that this is not just an effect of ligand length; the dA10, (SP18)1, and (SP18)5 systems all have similar lattice parameters, yet the increased flexibility widens the annealing window by ca. 1.5°C.

871

872

Oligonucleotide Flexibility Dictates Crystal Quality in DNA-Programmable

Furthermore, this is not solely an effect of the width of the melting transition. The same data plotted as ß vs. the singlet fraction, or the ratio of free over total particles, clearly shows a decrease in ß with increasing additions of SP18 up to 10 subunits. Thus, we postulate that flexibility itself is a major factor in the increased ability of the system to crystallize. Similar to colloidal hard-sphere packing, longer interparticle interactions will widen and thus soften the pair potential, which is known to decrease jamming for colloidal systems [21]. Using the principles outlined above, we next examined the crystallization of PAEs with larger diameter cores, up to approximately 100 nm in diameter (Fig. 44.4). The range of the annealing window for each system can be found in Fig. 44.4H. By using a (SP18)10 flexible spacer, 58.6 and 77.5 nm diameter AuNPs crystallized at approximately Tm and 0.5°C above Tm, respectively. Note, Tm is defined as the temperature at which the fraction of free particles equals the fraction of bound particles; thus a sample may be annealed slightly above Tm, though a significant fraction of particles will dissociate from the crystals. For this spacer length, PAEs with 99.9 nm diameter AuNP cores remained disordered at all temperatures that were thermodynamically stable; crystallization was not observed until longer (SP18)20 spacers were used, and even then not until almost 1°C above Tm. While a variety of metamaterials designs have been proposed using noble metal colloidal building blocks, all rely on the plasmonic coupling between adjacent particles. For two particles, coupling is predicted to occur when the gap/diameter ratio is less than 0.5 [30]. It is therefore expected that interesting plasmonic behavior will be present in PAE materials with these gap/diameter ratios ∆Hbcc; e.g., 7.9% difference in fH between fcc and bcc in system (b), as shown in Table 45.1), which results in the enthalpic term dominating the entropic one and therefore favoring the formation of an fcc lattice (i.e., ∆Gfcc > ∆Gbcc). However, when the enthalpic difference between the two lattices is not as significant (∆Hfcc > ∆Hbcc; e.g. 6.3% difference in fH between fcc and bcc in system (a), as shown in Table 45.1), the role of the “−T∆S” term becomes more important than in the previous case. The additional flexibility afforded in the DNA allows the linkers to access more conformational states, resulting in

891

892

Entropy-Driven Crystallization Behavior in DNA-Mediated Nanoparticle Assembly

a greater difference between the entropic values (i.e., ∆Sbcc >> ∆Sfcc) and bcc being favored over fcc for these specific systems (i.e., ∆Gbcc > ∆Gfcc). Therefore, the subtle difference in the enthalpies of forming fcc over bcc dictates the significance of the entropic contribution to the overall free energy. As the sum of these two energies determines the most thermodynamically stable structure, two different approaches need to be used when explaining crystallization behavior: when the system is enthalpically dominated, design rules and geometric calculations can be applied, but in a system with long, flexible strands, it is crucial to utilize computational models to accurately account for entropic contributions. It is important to note that an fcc−bcc transition is also seen in the assembly of micelle and star-polymer systems as well as silver nanoparticles capped with alkylthiols of varying length [32−38]. In these polymer systems, researchers reported fcc packing for “crewcut” micelles and bcc packing for “hairy” micelles [36]. Comparing this to our discussion above for the DNA-coated AuNP system, we find that “soft” shells typically lead to bcc lattice formation, whereas “hard” spheres tend to prefer fcc arrangements.

45.4 Conclusion

In conclusion, this study has demonstrated that under specific sets of conditions, contributions beyond the enthalpically centered “maximization of nearest-neighbor interactions” in DNA-directed crystallization of nanoparticles become significant. Specifically, the additional configurational states that are accessed by incorporating long DNA linkers with flexor “pivot points” lead to body-centered cubic superlattice formation in the typically fcc-forming, singlecomponent system. This concept can be illustrated through modular iterations to both the DNA linker length and AuNP size or the strikingly simple difference of a single base being hybridized or not. Molecular dynamics simulations elucidate the underlying subtleties of the thermodynamics of crystallization among these different systems. These results highlight the notion that the region of the linker traditionally considered to be a “passive spacer” for tuning interparticle distances can actually contribute to the energetics of the crystallization pathway. In the current state of the field, while this

References

outlines a region of the vast parameter space that perhaps should be avoided due to the loss of predictability, it also could potentially serve as another tunable design parameter in the programmable assembly of nanomaterials with DNA to determine new structures.

Acknowledgments

This material is based upon work supported by the Air Force Office of Scientific Research Awards FA9550-11-1-0275 and FA9550-121-0280; the Department of Defense National Security Science and Engineering Faculty Fellowship award N00014-15-1-0043; and the Center for Bio-Inspired Energy Science (CBES), an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences award DE-SC0000989-0002. R.V.T. acknowledges a National Science Foundation Graduate Research Fellowship (DGE-1324585). T.I.N.G.L. and Y.K. acknowledge the Ryan Fellowship at the Northwestern University International Institute for Nanotechnology. T.I.N.G.L. acknowledges the Chinese government for the Award for Outstanding Self-Financed Students Abroad.

References

1. Macfarlane, R. J., O’Brien, M. N., Petrosko, S. H. and Mirkin, C. A. (2013). Angew. Chem. Int. Ed., 52, 5688.

2. Mirkin, C. A., Letsinger, R. L., Mucic, R. C. and Storhoff, J. J. (1996). Nature, 382, 607.

3. Tan, S. J., Campolongo, M. J., Luo, D. and Cheng, W. (2011). Nat. Nanotechnol., 6, 268. 4. Park, S. Y., Lytton-Jean, A. K. R., Lee, B., Weigand, S., Schatz, G. C. and Mirkin, C. A. (2008). Nature, 451, 553.

5. Nykypanchuk, D., Maye, M. M., van der Lelie, D. and Gang, O. (2008). Nature, 451, 549. 6. Cutler, J. I., Auyeung, E. and Mirkin, C. A. (2012). J. Am. Chem. Soc., 134, 1376.

7. Brodin, J. D., Auyeung, E. and Mirkin, C. A. (2015). Proc. Natl. Acad. Sci. U.S.A., 112, 4564.

8. Auyeung, E., Cutler, J. I., Macfarlane, R. J., Jones, M. R., Wu, J., Liu, G., Zhang, K., Osberg, K. D. and Mirkin, C. A. (2012). Nat. Nanotechnol., 7, 24.

893

894

Entropy-Driven Crystallization Behavior in DNA-Mediated Nanoparticle Assembly

9. Zhang, C., Macfarlane, R. J., Young, K. L., Choi, C. H. J., Hao, L., Auyeung, E., Liu, G., Zhou, X. and Mirkin, C. A. (2013). Nat. Mater., 12, 741.

10. Macfarlane, R. J., Thaner, R. V., Brown, K. A., Zhang, J., Lee, B., Nguyen, S. T. and Mirkin, C. A. (2014). Proc. Natl. Acad. Sci. U.S.A., 111, 14995.

11. Kim, Y., Macfarlane, R. J. and Mirkin, C. A. (2013). J. Am. Chem. Soc., 135, 10342. 12. Auyeung, E., Li, T. I. N. G., Senesi, A. J., Schmucker, A. L., Pals, B. C., de la Cruz, M. O. and Mirkin, C. A. (2014). Nature, 505, 73. 13. Auyeung, E., Macfarlane, R. J., Choi, C. H. J., Cutler, J. I. and Mirkin, C. A. (2012). Adv. Mater., 24, 5181.

14. Macfarlane, R. J., Lee, B., Jones, M. R., Harris, N., Schatz, G. C. and Mirkin, C. A. (2011). Science, 334, 204.

15. Macfarlane, R. J., Jones, M. R., Senesi, A. J., Young, K. L., Lee, B., Wu, J. and Mirkin, C. A. (2010). Angew. Chem. Int. Ed., 49, 4589. 16. Li, T. I. N. G., Sknepnek, R., Macfarlane, R. J., Mirkin, C. A. and Olvera de la Cruz, M. (2012). Nano Lett., 12, 2509.

17. Senesi, A. J., Eichelsdoerfer, D. J., Brown, K. A., Lee, B., Auyeung, E., Choi, C. H. J., Macfarlane, R. J., Young, K. L. and Mirkin, C. A. (2014). Adv. Mater., 26, 7235.

18. Hurst, S. J., Lytton-Jean, A. K. R. and Mirkin, C. A. (2006). Anal. Chem., 78, 8313. 19. Bloomfield, V. A., Crothers, D. M. and Tinoco, I. (2000). Nucleic Acids: Structures, Properties, and Functions (University Science Books, Sausalito, CA). 20. Li, T. I. N. G., Sknepnek, R. and Olvera de la Cruz, M. (2013). J. Am. Chem. Soc., 135, 8535. 21. Knorowski, C. and Travesset, A. (2012). Soft Matter, 8, 12053.

22. Ortiz, V. and de Pablo, J. J. (2011). Phys. Rev. Lett., 106, 238107. 23. Zhang, J. and Liu, J. S. (2006). PLoS Comput. Biol., 2, e168.

24. Shevchenko, E. V., Talapin, D. V., Kotov, N. A., O’Brien, S. and Murray, C. B. (2006). Nature, 439, 55.

25. Tkachenko, A. V. (2011). Phys. Rev. Lett., 106, 255501.

26. Knorowski, C., Burleigh, S. and Travesset, A. (2011). Phys. Rev. Lett., 106, 215501. 27. Knorowski, C. and Travesset, A. (2011). Curr. Opin. Solid State Mater. Sci., 15, 262.

References

28. Anderson, J. A., Lorenz, C. D. and Travesset, A. (2008). J. Comput. Phys., 227, 5342.

29. Nguyen, T. D., Phillips, C. L., Anderson, J. A. and Glotzer, S. C. (2011). Comput. Phys. Commun., 182, 2307. 30. Welch, B. L. (1947). Biometrika, 34, 28.

31. Suzuki, H. (1993). Phase transition in gels of sub-millimeter size induced by interaction with stimuli. In: Dušek, K. (eds.) Responsive Gels: Volume Transitions II. Advances in Polymer Science, vol. 110 (Springer, Berlin, Heidelberg), pp. 149−240. 32. Korgel, B. A. and Fitzmaurice, D. (1999). Phys. Rev. B, 59, 14191. 33. Bang, J. and Lodge, T. (2008). Macromol. Res., 16, 51.

34. Bang, J. and Lodge, T. P. (2004). Phys. Rev. Lett., 93, 245701.

35. Bang, J., Viswanathan, K., Lodge, T. P., Park, M. J. and Char, K. J. (2004). Chem. Phys., 121, 11489. 36. Lodge, T. P., Bang, J., Park, M. J. and Char, K. (2004). Phys. Rev. Lett., 92, 145501.

37. Ziherl, P. and Kamien, R. D. (2001). J. Phys. Chem. B, 105, 10147.

38. Thomas, E. L., Kinning, D. J., Alward, D. B. and Henkee, C. S. (1987). Macromolecules, 20, 2934.

895

Chapter 46

Electrolyte-Mediated Assembly of Charged Nanoparticles*

Sumit Kewalramani,a Guillermo I. Guerrero-García,a,b Liane M. Moreau,a Jos W. Zwanikken,a Chad A. Mirkin,a,c Monica Olvera de la Cruz,a,c,d and Michael J. Bedzyka,d aDepartment

of Materials Science and Engineering, Northwestern University, 2220 Campus Drive, Evanston, IL 60208, USA bInstituto de Física, Universidad Autonoma de San Luis Potosí, Alvaro Obregon 64, 78000 San Luis Potosí, San Luis Potosí, Mexico cDepartment of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA dPhysics and Astronomy Department, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208, USA [email protected]; [email protected]

Solutions at high salt concentrations are used to crystallize or segregate charged colloids, including proteins and polyelectrolytes *Reprinted with permission from Kewalramani, S., Guerrero-García, G. I., Moreau, L. M., Zwanikken, J. W., Mirkin, C. A., Olvera de la Cruz, M. and Bedzyk, M. J. (2016). Electrolyte-mediated assembly of charged nanoparticles, ACS Cent. Sci. 2, 219−224. . Further permissions related to the material excerpted should be directed to the ACS. Spherical Nucleic Acids, Volume 2 Edited by Chad A. Mirkin Copyright © 2020 Jenny Stanford Publishing Pte. Ltd. ISBN 978-981-4877-22-0 (Hardcover), 978-1-003-05669-0 (eBook) www.jennystanford.com

898

Electrolyte-Mediated Assembly of Charged Nanoparticles

via a complex mechanism referred to as “salting-out.” Here, we combine small-angle X-ray scattering (SAXS), molecular dynamics (MD) simulations, and liquid-state theory to show that saltingout is a long-range interaction, which is controlled by electrolyte concentration and colloid charge density. As a model system, we analyze Au nanoparticles coated with noncomplementary DNA designed to prevent interparticle assembly via Watson−Crick hybridization. SAXS shows that these highly charged nanoparticles undergo “gas” to face-centered cubic (FCC) to “glass-like” transitions with increasing NaCl or CaCl2 concentration. MD simulations reveal that the crystallization is concomitant with interparticle interactions changing from purely repulsive to a “long-range potential well” condition. Liquid-state theory explains this attraction as a sum of cohesive and depletion forces that originate from the interelectrolyte ion and electrolyte−ion−nanoparticle positional correlations. Our work provides fundamental insights into the effect of ionic correlations in the salting-out mechanism and suggests new routes for the crystallization of colloids and proteins using concentrated salts.

46.1 

Introduction

Controlling the crystallization of colloids, including proteins, from solutions has been a scientific goal for decades [1−7]. The crystallization of charged colloids is often induced by using high salt concentrations, a process referred to as “salting-out” [7]. Colloids can also be concentrated and crystallized via the well-understood depletion forces induced by the addition of polymers [5, 8] or micelle forming surfactants [9]. However, colloidal crystallization in high ionic strength solutions is subtle and not understood. Crystallization via “salting-out” is observed for specific salts in a narrow range of salt concentrations, when the interparticle interactions are weakly attractive [10]. Above this salt concentration range, in the regime of stronger attractive interactions, amorphous precipitates are observed. It is generally believed that short-ranged attractions due to ionic correlations and solvation effects drive the colloidal assembly [11]. By contrast, the present study reveals that, in high ionic strength solutions, the interparticle attraction between like-

Introduction

charged nanoparticles extends a few nm from the colloidal surface. This “long-range” attraction is induced by the electrolyte ions, and is not an effect of van der Waals forces. Long-range interactions between like-charged colloids near surfaces [12, 13] have been explored for decades. These interactions are attractive near surfaces due to hydrodynamic effects [14], but in bulk solutions they are found to be purely repulsive [15]. Here, we show that electrolyte-mediated long-range interparticle attractions are possible in bulk solutions in the regime of high ionic strength. To enhance the electrostatic coupling between the nanoparticles and the electrolyte ions, our experimental design used highly charged (>2000 e−/nanoparticle) DNA coated spherical gold nanoparticles (AuNPs) in solutions containing high concentrations of NaCl or CaCl2. To avoid interparticle assembly via Watson−Crick hybridization [16, 17], we used DNAs that lacked self-complementary single-stranded sticky ends. Naively, one might expect that, in the absence of hybridization, the interactions between DNA coated AuNPs are purely repulsive. Here, small-angle X-ray scattering (SAXS) shows that, depending on the salt concentration and the DNA, FCC crystals are formed with nearest-neighbor distances (dNN) that are comparable with twice the nanoparticle hydrodynamic radius R. This demonstrates the emergence of concentrated electrolyte-mediated attractions. Various mean field theories have been developed to compute the effective interactions between charged colloids, for example, the Derjaguin−Landau−Verwey−Overbeek (DLVO) [18] theory and its extensions that include the renormalized charges of the colloids. However, at high ionic strengths these models cannot account for the correlations among ions surrounding strongly charged colloids. Recently, numerical techniques have elucidated that ionic correlations in confined concentrated electrolytes can induce attractions between like-charged surfaces at concentrations larger than 300 mM NaCl [19]. These attractions are distinct from the multivalent (Z ≥ 3) counterion-mediated attractions in DNA and other polyelectrolytes [20−22], which are observed at low ionic strengths (μM− mM), are short-ranged (a few Å corresponding to the multivalent ion diameter), and lead to unstable precipitates in the absence of specific short-range attractions as the salt concentration increases. Here, we find attractions at high ionic strengths (>100 mM) and even in monovalent salts, resembling the “salting-out” effect.

899

900

Electrolyte-Mediated Assembly of Charged Nanoparticles

By molecular dynamics (MD) simulations and liquid-state theory we provide evidence that the ionic correlations in the concentrated electrolyte induce interparticle long-range attractions and drive the assembly.

46.2  Results and Discussion

46.2.1  SAXS Studies of DNA-Coated AuNP Assembly To analyze the effect of charge density, DNA rigidity, and electrolyte concentration, we studied four sample sets. These sets correspond to two nanoparticle types: AuNPs (nominally, 10 nm diameter) functionalized with single-stranded (ss)-DNA (ssDNA-AuNP) or double-stranded (ds)-DNA (dsDNA-AuNP) (insets, Figs. 46.1A and C), each dispersed in two solution types, NaCl and CaCl2. For all samples, the nanoparticle concentration was ~50 nM, corresponding to an average center-to-center interparticle distance of ~400 nm in the gas phase. For each set, ionic strengths (μs) in the range ~30−2000 mM were examined. By definition, for NaCl solutions, μs = [NaCl] and for CaCl2 solutions, μs = 3 × [CaCl2]. In salt-free solutions, dynamic light scattering (DLS) yield hydrodynamic radii of R ª 19 nm and R ª 13 nm for ssDNA-AuNP and dsDNA-AuNP, respectively, corresponding to volume fractions of ~8.7 × 10−4 and ~2.7 × 10−4. These salt-free values mark the upper bounds for the volume fractions since the radial extension of the DNA on the nanoparticles, as expected [23], is found to decrease with increasing μs due to the enhanced screening of the intra-DNA electrostatic repulsions. For all the sample sets, Fig. 46.1 shows representative SAXS intensity profiles (I) as a function of the scattering vector magnitude q (= 4π sin θ/λ). Here, λ is the X-ray wavelength and 2θ is the scattering angle. For ssDNA-AuNPs in NaCl solutions, the main features of the intensity profiles are μs-independent. To illustrate, two extreme μs cases are shown in Fig. 46.1A. These SAXS profiles exhibit the characteristics of scattering from isolated DNA-coatedAuNPs, which is predominantly due to the electron-dense Au cores [24]. Based on SAXS from a solid homogeneous sphere [25], the position of the first minima (qmin ª 1 nm−1) corresponds to a Au core radius of RAu = ~4.5/qmin = 4.5 nm. Unlike ssDNA-AuNPs in NaCl

Results and Discussion

solutions, ssDNA-AuNPs in CaCl2 or dsDNA-AuNPs in NaCl or CaCl2 solutions aggregate into clusters above a threshold ionic strength μt, as evidenced by the appearance of sharp intensity modulations in the q < 1 nm−1 region (Fig. 46.1B−D). DLS measurements show that a typical cluster size is ~1.7 μm.

Figure 46.1  Ionic-strength-dependent assembly behavior of DNA coated AuNPs. 1D SAXS intensity profiles for ssDNA-AuNP and dsDNA-AuNP in NaCl (A, C) and CaCl2 (B, D) solutions. The data shown is the scattered intensity above the background scattering from empty capillary and pure water. The insets in panels A and C show the DNA-grafted-AuNP components. There are ~60 thiolatedDNA tethered to each AuNP. About 40% of the strands on dsDNA-AuNPs were in duplexed form. The ssDNA is a T40 strand. The DNA chain in panel C consists of a 10 base long ssDNA spacer A10 and an 18 base-pair long dsDNA segment. Therefore, the total charge on the nanoparticles is ~2400 e−/NP and ~2100 e−/NP for ssDNA-AuNP and dsDNA-AuNP, respectively. Solid red lines are the expected scattered intensities from isolated DNA-grafted-AuNPs.

Comparison of μt in different sample sets shows that Ca2+ induces aggregation of DNA-coated-AuNPs at much lower μs than Na+ (Figs. 46.1C and D). Similarly, dsDNA-AuNPs form aggregates at a much lower μs than ssDNA-AuNPs (Figs. 46.1B and D). Thus, the DNA-coated-AuNPs form aggregates more readily when the DNA charge density and the counterion valency are increased. These trends indicate that the responsible attractions cannot originate from van der Waals forces.

901

902

Electrolyte-Mediated Assembly of Charged Nanoparticles

Figure 46.1 shows the simulated intensities P(q) for isolated DNA-grafted-AuNPs (solid red lines). For all the cases where nanoparticle aggregation is not observed, the measured I(q) are well described by simulations based on mean Au core size RAu = 4.5 nm and polydispersity (PD) = 8.5% or RAu = 4.4 nm and PD = 7.7%, depending upon the nanoparticle batch used. This analysis allows extraction of the structure factor [S(q) = I(q)/P(q)] for nanoparticle aggregates (Figs. 46.2A and B).

Figure 46.2  Structure of DNA coated AuNP assemblies. (A, B) SAXS-derived S(q) for DNA-grafted-AuNP aggregates (circles) along with simulations based on FCC lattices (red lines). For reference, the expected peak positions and relative intensities for Bragg reflections from ideal FCC lattices are shown (A, vertical black lines). The labels ss and ds correspond to ssDNA-AuNP and dsDNA-AuNP, respectively. (C) Representative radial distribution functions for dsDNA-AuNPs in CaCl2 as a function of μs along with the expected positions and relative populations (P/12) for neighbors in a FCC lattice (black lines). For visual comparisons, g(r) is plotted against normalized radial distance r/r1. Here, r1 = dNN represents the nearest-neighbor interparticle distance. Monte Carlo simulations for g(r) based on random close packing (RCP) of hard spheres (blue lines) reasonably describe the experimental g(r) for μs much higher than μt. (D) Schematic of the observed changes in colloidal packing as a function of ionic strength.

Two types of S(q) profiles are observed. First, regardless of the DNA-coating and the salt solution, S(q) exhibits similar features at the threshold ionic strength (μt) for aggregation. These S(q) are

Results and Discussion

plotted against q/q1 (Fig. 46.2A), where, q1 is the position of the principal peak. Similarly, for μs >> μt, S(q) vs. q/q1 profiles are nearly identical (Fig. 46.2B), but subtly different from the profiles at μs = μt. The analysis of S(q) based on a formalism by Förster et al. [26] shows that, for μs = μt, DNA functionalized AuNPs are arranged on FCC lattices (Fig. 46.2A). The positions of the principal FCC (1 1 Ê 12p ˆ = 29.2, 36.7, and 1) peak yield lattice parameters aFCC Á = q1 ˜¯ Ë 34.4 nm for ssDNA-AuNPs in CaCl2, and dsDNA-AuNPs in NaCl and CaCl2 solutions, respectively. For the three cases in Fig. 46.2A, the widths of the (h k l) diffraction profiles yield average crystallite sizes of 200−300 nm. Taken together, the DLS-measured aggregate size (1.7 μm) and the SAXS-derived crystallite size imply that the DNAgrafted-AuNPs assemble into polycrystalline aggregates at ionic strengths equal to or slightly above μt. Therefore, under appropriate conditions, electrolyte-mediated interactions can induce crystalline order in DNA functionalized AuNPs even in the absence of Watson− Crick hybridization. Figure 46.2B shows that, for μs >> μt, the assembly does not consist of FCC crystallites. More information about the nanoparticle packing in these aggregates is gleaned from radial distribution function g(r). Figure 46.2C shows the μs-dependence of g(r) for dsDNA-AuNPs in CaCl2 solutions. For the 50 mM [Ca2+] case (μs = μt), the amplitudes and the positions of maxima in g(r) at r/r1 = 1, √2, √3, √4, √5, etc. are consistent with FCC lattices (Fig. 46.2C, bottom). With increasing μs, the r/r1 = √2 modulation smears out. Further, the g(r) exhibit a slightly split doublet with nearly equal amplitude maxima at r/r1 ª √3 and ª √4 (Fig. 46.2C, middle and top). This doublet is a signature of a glassy phase [27]. Specifically, the g(r) for [Ca2+] = 100 mM (Fig. 46.2C, middle) resembles the g(r) for the “metallic-glasslike” packing of spherical colloids [2]. Similarly, the g(r) for [Ca2+] = 250 mM, where the r/r1 = √2 feature is mostly smeared out, is reminiscent of the g(r) for random-close-packed (RCP) spheres [28]. These observations imply that the packing of DNA-grafted-AuNPs transforms from isolated particles (gas-like) to face centered cubic (FCC) to “glass-like” arrangement with increasing μs (Fig. 46.2D). The structural phase transition sequence is similar to that observed for protein crystallization [10]. Furthermore, similar to the case of

903

904

Electrolyte-Mediated Assembly of Charged Nanoparticles

proteins, the crystallization of DNA-coated AuNPs occurs in a narrow μs regime, for example, μs ª 1050−1500 mM for ssDNA-AuNP in CaCl2. Our results suggest that the electrolyte concentration induced “gas” to “crystalline” to “amorphous” transitions are a general feature of the assembly of charged colloids in high-ionic-strength solutions. Some insight into the assembly mechanism of DNA-graftedAuNPs is obtained from the (nearest-neighbor distance) dNN versus μs trends. First, the dNN continuously decreases with increasing μs to reach a constant value in the glassy state, which is ~94% of the dNN observed for FCC crystals at μs = μt. Second, the observed dNN are smaller than estimates for 2R that are based on the combination of modified Daoud−Cotton model parameters [23] for the ssDNA radial extension and the experimental values for the average inter-basepair separation for dsDNA in Watson−Crick hybridization-driven assemblies [29]. Both these observations suggest a dense packing of DNA-grafted-AuNPs that is driven by electrolyte-mediated attractions.

46.2.2 MD Simulations for Potential of Mean Force between DNA-Coated AuNPs

The hypothesis of electrolyte-mediated interparticle attractions was validated by MD simulations. Figure 46.3A shows the potential of mean force between two dsDNA-AuNPs as a function of the distance between their centers in the presence of an electrolyte with divalent cations and monovalent anions (2:1 electrolyte). Here, the two DNA-grafted-AuNPs interact only via short-ranged repulsive steric interactions, and long-ranged Coulomb potentials. Two values of μs were simulated: 15 mM (μs