REPORT◥

POLYMERS

Templated nanofiber synthesis viachemical vapor polymerization intoliquid crystalline filmsKenneth C. K. Cheng1,2*†, Marco A. Bedolla-Pantoja3*, Young-Ki Kim3,4*,Jason V. Gregory1,5, Fan Xie1,5, Alexander de France1,5, Christoph Hussal6, Kai Sun2,Nicholas L. Abbott3,4‡, Joerg Lahann1,2,5,6‡

Extrusion, electrospinning, and microdrawing are widely used to create fibrous polymermats, but these approaches offer limited access to oriented arrays of nanometer-scalefibers with controlled size, shape, and lateral organization. We show that chemical vaporpolymerization can be performed on surfaces coated with thin films of liquid crystalsto synthesize organized assemblies of end-attached polymer nanofibers. The process useslow concentrations of radical monomers formed initially in the vapor phase and thendiffused into the liquid-crystal template. This minimizes monomer-induced changes to theliquid-crystal phase and enables access to nanofiber arrays with complex yet preciselydefined structures and compositions. The nanofiber arrays permit tailoring of a wide rangeof functional properties, including adhesion that depends on nanofiber chirality.

Surfaces decorated with oriented arrays offibers are ubiquitous in the natural worldbecause they can provide functions suchas sensing [hair cells (1)], thermal insula-tion [polar bear fur (2)], enhanced mass

transport [microtubules (3)], extreme wettingproperties [lotus leaf (4)], and reversible adhe-

sion [gecko foot (5)]. However, re-creation ofthese functions in synthetic materials requiresmultiscale engineering of the composition, shape,and morphology of individual fibers, as well ascontrol of higher-order organization of fibers intoarrays. We address this challenge by buildingfrom studies reported in 1916 by T. Svedberg (6),

who used the long-range molecular order andfluidity inherent to liquid crystals (LCs) to con-trol chemical reactions, principles that have sincebeen exploited in awide range of transformationsbased on unimolecular reactions (7), molecularself-assembly (8), or polymerizations (9, 10). Akey limitation of LC-templated polymerization,however, has been perturbation of the LC phaseby monomers that are dissolved into the LCbefore the polymerization and then consumedduring polymerization (9, 10).Chemical vapor polymerization (CVP) is a versa-

tile process that is compatible with a range ofpolymerization modes (Fig. 1A), including reactionpathways using [2.2]paracyclophanes (11–13)(Gorham process) or halogenated xylene precursors(Gilch process) (14, 15) and free-radical ring-opening copolymerization (ROP) using 5,6-benzo-2-methylene-1,3-dioxepane and [2.2]paracyclophanes(16). CVP is widely used for fabrication of con-sumer products (e.g., electronics and packaging)because it enables rapid and inexpensive conformalcoating of large surface areas with polymer films (12).Whereas past studies used CVP to form continuous

RESEARCH

Cheng et al., Science 362, 804–808 (2018) 16 November 2018 1 of 4

1Biointerfaces Institute, University of Michigan, Ann Arbor,MI, USA. 2Department of Materials Science and Engineering,University of Michigan, Ann Arbor, MI, USA. 3Department ofChemical and Biological Engineering, University ofWisconsin–Madison, Madison, WI, USA. 4Robert FrederickSmith School of Chemical and Biomolecular Engineering,Cornell University, Ithaca, NY, USA. 5Department of ChemicalEngineering, University of Michigan, Ann Arbor, MI, USA.6Institute of Functional Interfaces, Karlsruhe Institute ofTechnology, Karlsruhe, Germany.*These authors contributed equally to this work.†Present address: IBM Research, Albany, NY, USA.‡Corresponding author. Email: nabbott@cornell.edu (N.L.A.);lahann@umich.edu (J.L.)

Fig. 1. Templated synthesis of nanofiber arraysvia CVP into anisotropic media. (A) CVP of 1a to1h yields polymers 2a to 2h. m, n, and l:copolymer repeat units; D: 250°C. (B andC) Representative chemical structures ofcyanobiphenyl-based (5CB and E7) (B) and halo-genated (TL205) (C) LCs. (D) Fabrication of polymernanofibers via CVP into a LC phase alignedperpendicular to the substrate. (i) CVP;(ii) LC removal. (E) Scanning electron microscopy(SEM) images of nanofibers polymerizedfrom 1a (10 mg) in 5CB. After the nanofibersynthesis, the LC template was removed.(F) Optical micrograph (crossed polars) of a nano-fiber. Orientations of the analyzer (A)and polarizer (P) are shown in the whitedouble-arrow cross. The yellow double arrow indi-cates the main axis of the nanofiber.(G and H) Micrographs (crossed polars) of thenanofiber with a quarter-wave plate with itsslow axis (g, green double arrow) perpendicular (G)or parallel (H) to the fiber axis; lower-order interfer-ence colors [yellow in (G)] indicate a decrease inretardance. (I) Analysis of interference colors of thenanofiber in (G) and (H) indicates that the polymerchains are aligned along the fiber axis.

CVP

CVP

CVP

CVP

on October 8, 2020

http://science.sciencem

ag.org/D

ownloaded from

polymeric films on surfaces (11–16), we foundthat CVP of compound 1a into micrometer-thicksupported films of nematic LCs (Fig. 1, B and C)resulted in the formation of surfaces decoratedwith aligned arrays of nanofibers (Fig. 1, D and E).When nematic 4′-pentyl-4-biphenylcarbonitrile(5CB) was anchored on a silane-functionalizedglass surface in a perpendicular orientation (i.e.,homeotropic anchoring, Fig. 1D), the nanofiberswere straight and aligned perpendicular to thesurface. Removal of the LC [confirmed by Fouriertransform infrared (FTIR) spectroscopy and solid-state 13C nuclear magnetic resonance (NMR)spectroscopy, figs. S1 and S2] revealed that thenanofibers were made of insoluble polymer, wereanchored at one end to the surface, and werestructurally amorphous (fig. S3) yet optically bire-fringent, as confirmed by cross-polarized lightmicroscopy (Fig. 1F). Insertion of a quarter-waveplate between crossed polarizers (Fig. 1, G and H)revealed that the refractive index was greatestalong the fiber axis, consistent with electrondiffraction patterns (fig. S3B), indicating align-ment of polymer chains along the main axis ofthe nanofibers (Fig. 1I) (17).CVP into films that lacked fluidity (crystal-

line solid 5CB, Fig. 2A) or long-range order (iso-tropic liquid 5CB, Fig. 2C; or silicone oil, Fig. 2D)

did not yield nanofibers, indicating that long-range order and fluidity are both necessary re-quirements for the shape-controlled synthesisof nanofiber arrays (Fig. 2B). Replacement of5CB with nematic E7 (a mixture of cyanobi-phenyls, Fig. 1B), TL205 (a mixture of halogen-ated molecules, Fig. 1C), or other nematic LCs(fig. S4) yielded organized assemblies of nano-fibers with well-defined yet distinct diameters(D) of 142 ± 11 nm in 5CB, 85 ± 9 nm in E7, and69 ± 7 nm in TL205 (Fig. 2E). We hypothesizedthat the nanofiber diameter is controlled by theextrapolation length x, which is defined as K/W,where K is the average Frank elastic constantfor splay and bend of the LC and W is thesurface anchoring energy density. If an inclu-sion (here, nanofiber) in a LC grows to a sizethat exceeds x, the orientation-dependent inter-facial energy associated with interaction of theLC with the surface of the inclusion (WD2, wherediameter D is the inclusion size) exceeds theenergetic cost of elastic deformation of the LC(KD), and the LC will elastically deform aroundthe inclusion (18, 19). We tested the hypothesisthat D ≈ K/W by using literature values of K attemperature (T) = 25°C (18, 20) and our exper-imental values of D to calculate W (Fig. 2E).Calculated values of W were ~10−4 J/m2 (21),

with WTL205 > WE7 > W5CB, a ranking that isconsistent with (i) the theoretical predictionthat W º (TNI − T)2b, where the material con-stant b = 0.4 to 0.5 (22, 23) and TNI is thenematic-isotropic phase-transition temperatureof the LCs (Fig. 2E), and (ii) independent ex-periments that measured the relative values ofW of these LCs (fig. S5). We also found theaverage diameters of the nanofibers to depend onthe temperature of the LC during CVP. For ex-ample, D increased with T (Fig. 2E and fig. S6), astheoretically predicted by x = K/Wº (TNI − T)−b

(22, 23). Overall, these results are consistentwith a mechanism of growth in which the elasticenergy of the LC defines the nanofiber diameter(via x) and promotes preferential growth ofthe nanofibers along the alignment direction ofthe LCs.We performed CVP of 1a using homeotropi-

cally oriented E7 films with thicknesses rangingfrom 5 to 22 mm and found that the lengths ofthe fibers closely matched the LC film thick-nesses (Fig. 2F). The result confirms growth ofthe nanofibers along the LC (figs. S7 and S8).We also found that monomers with a widerange of chemical functional groups could bepolymerized in LCs by CVP (Gorham process),yielding (i) ethynyl-functionalized nanofibers

Cheng et al., Science 362, 804–808 (2018) 16 November 2018 2 of 4

Fig. 2. Templated CVP of nanofibers withprecise lengths, diameters, and surfacechemistries. (A to D) SEM images of nanofibersformed by CVP of 1a (10 mg) with the indicatedtemplates (see insets): crystalline 5CB at13°C (A), nematic 5CB at 25°C (B), isotropic5CB at 37°C (C), and isotropic silicone oil at 25°C(D). The LC thickness was 21.7 ± 0.5 mm.(E) Nanofiber diameters (left axis) obtained byCVP of 1a (6 mg) into E7 at 13°C (down triangle),25°C (circle), 30°C (up triangle), and 5CB andTL205 at 25°C (circles). Red X’s are thecalculated surface anchoring energy densities(W, right axis) for each LC at 25°C. The insettable shows elastic constants (K) at 25°Cand the nematic-isotropic phase transitiontemperatures (TNI) of TL205, E7, and 5CB.(F) Nanofiber length as a function of eithernematic E7 film thickness (black points) or massof polymerized 1a for a LC film with thicknessof 21.7 ± 0.5 mm (red points). Mean ± SD, n ≥10 measurements. (G to I) Representative SEMimages of 2b (G), 2c (H), and 2d (I) templatedinto TL205. (J to L) Representative FTIR spectraof 2b (J), 2c (K), and 2d (L) templated intoTL205. FTIR spectra of the nanofibers (red)are compared to polymer films synthesizedwithout the LC phase (blue). LCs were removedbefore imaging and FTIR spectroscopy.a.u., arbitrary units.

RESEARCH | REPORTon O

ctober 8, 2020

http://science.sciencemag.org/

Dow

nloaded from

for click-based reactions with azide derivatives(2b, Fig. 2G), (ii) nanofibers that simultaneouslypresent ethynyl and hydroxyl groups for reac-tion with azides and activated carboxylic acids(2c, Fig. 2H), (iii) nanofibers without func-tional groups (2d, Fig. 2I) as a nonreactive ref-erence, (iv) polycationic pyridine-functionalizednanofibers (2e, fig. S9A), and (v) water-repelling

perfluoro-functionalized nanofibers (2f, fig. S9B).Additionally, we used CVP into LCs to generatepolymeric nanofibers with distinct main chains,including (vi) biodegradable polyester nano-fibers (2g, ROP process, fig. S9C) and (vii) semi-conducting poly(phenylene vinylene) nanofibers(2h, Gilch process, fig. S9D). When templated bynematic TL205, nanofibers made of polymers

2b to 2h were morphologically similar (Fig. 2, Gto I, and fig. S9, A to D) and possessed infraredspectroscopic signatures of the constituent chem-ical groups (Fig. 2, J to L, and fig. S9, E to H).LC ordering within films is influenced by

interactions with confining surfaces, LC elasticmoduli, and molecular properties of LCs, includ-ing chirality (19), thus offering access to a di-verse range of LC templates for CVP. For example,a film of nematic 5CB prepared with planarand homeotropic anchoring at bottom and topLC surfaces, respectively, leads to a bent andsplayed internal ordering of the LC that tem-plates banana-shaped nanofibers (Fig. 3A). Smec-tic LCs, with a statistical layering of orientedmolecules (19), templated straight nanofiberswith broadened tips arranged in conical fan-like morphologies (Fig. 3B) (24). Chiral nematicphases (cholesteric) yielded shape-controlledand chiral nanofiber assemblies with micrometer-scale periodicities (11.5 ± 1.5 and 11.3 ± 1.5 mm forS- and R-templated nanofibers, respectively) andan organization consistent with the fingerprintpattern characteristic of cholesteric films (peri-odicity of 10.8 ± 1.2 and 10.7 ± 1.1 mm for S- andR-handed cholesteric phases, respectively; Fig.3C and fig. S10). The chirality of the LC alsoinfluenced the handedness of the nanofibers,as confirmed by circular dichroism spectroscopyof surface-immobilized and solvent-dispersednanofibers (figs. S11 and S12) (25). In contrastto all other LC phases, a three-dimensional net-work of helical nanofibers was formed by CVPinto blue-phase LCs, reflecting nanofiber growthtemplated by a three-dimensional network ofdouble-twisted LC and line defects (Fig. 3D) (26).Overall, these results reveal that the shape, in-terfacial orientations, and morphologies of thenanofiber arrays are templated by the structureof the LC phase. The LC transition temperaturesremain almost unaltered by CVP (fig. S13), con-sistent with our conclusion (fig. S14) that themonomer concentration in the templating LCphase during nanofiber formation is low. Thesefindings differ from conventional polymerizationsin LCs, in which the dynamic interplay betweenthe polymerization process and the LC templatephase behavior makes control of the resultingpolymeric nanostructures difficult and often leavesunreacted monomers in the sample (26–28).Figure 4, A to D, shows that CVP into con-

formal films of nematic E7 formed over theouter or inner surfaces of a hollow cylinderyielded arrays of nanofibers (97.5 ± 17.5 nm indiameter) anchored on the curved surfaces ofthe cylinder. On the inner surface, the densityof the nanofibers decreased with increasingdistance from the open end (Fig. 4D). Meso-scopic nanofiber islands were templated frommicrometer-sized LC droplets electrosprayedonto surfaces, thus providing a scalable approachfor fabrication of arrays (Fig. 4, E and F). We alsofound that free-standing LC films formed withinmetallic meshes or at the ends of capillaries(Fig. 4G) templated organized nanofiber assem-blies (Fig. 4, H and I). Additionally, microbeadsdispersed in LC phases before CVP supported

Cheng et al., Science 362, 804–808 (2018) 16 November 2018 3 of 4

Fig. 3. Influence of LC template on nanofiber morphology and organization. (A to D) The leftcolumn shows optical micrographs (top view, crossed polars) of LC templates; insets are schematicillustrations (side view) of molecular order within the LC templates. The right two columns showSEM images of nanofibers templated from the LCs. (A) Nematic film of E7 with hybrid anchoring andresulting banana-shaped nanofibers. (B) Homeotropically oriented film of a smectic A LC phaseand the resulting polymeric nanostructures. (C) Micrograph showing cholesteric LC phase ofE7 doped with a left-handed chiral dopant (S-811). SEM images in middle and right columns shownanofibers templated from E7 containing left-handed (S-811) and right-handed (R-811) dopants,respectively. The black and blue arrows in the inset indicate the helical axis and handedness of thetwist, respectively. (D) Blue phase LC (BP1) with a cubic lattice spacing of ~250 nm and the resultingpolymeric nanostructure. The inset in the far-right column shows a bundle of helical nanofibers.

RESEARCH | REPORTon O

ctober 8, 2020

http://science.sciencemag.org/

Dow

nloaded from

growth of nanofibers, and copolymerization wasused to create nanofibers for coimmobilizationof different biomolecules (fig. S15).Overall, our results reveal that CVP into LC

templates enables scalable fabrication of arraysof polymeric nanofibers with programmableshapes, chemistries, and long-range lateral or-ganization, yielding interfacial properties cur-rently unavailable by other techniques. Forexample, we found that it was possible to ma-nipulate nanofiber chirality to control adhe-sion between surfaces. As shown in Fig. 4, Jand K, we measured adhesion to be higher forsurfaces decorated with nanofibers than thecorresponding flat CVP films, with the relativechirality of the nanofibers presented by the twosurfaces also influencing the magnitude andselectivity of adhesion (Fig. 4K). Other proper-

ties designed into nanofiber arrays prepared byLC-templated CVP include wettability, intrinsicphotoluminescence, biodegradability, and sur-face charge (Fig. 4, J and K, and fig. S16). Weenvisage that additional functional propertiescan be realized by exploiting the full diversityof LC templates (17, 19) along with other polym-erization mechanisms using vapor-phase deliv-ery of monomers.

REFERENCES AND NOTES

1. A. J. Hudspeth, Nat. Rev. Neurosci. 15, 600–614 (2014).2. H. E. M. Liwanag, A. Berta, D. P. Costa, S. M. Budge,

T. M. Williams, Biol. J. Linn. Soc. Lond. 107, 774–787(2012).

3. N. Hirokawa, Science 279, 519–526 (1998).4. W. Barthlott, C. Neinhuis, Planta 202, 1–8 (1997).5. K. Autumn et al., Nature 405, 681–685 (2000).6. T. Svedberg, Kolloid Z 18, 54–56 (1916).

7. A. Matsumoto, S. Nagahama, T. Odani, J. Am. Chem. Soc. 122,9109–9119 (2000).

8. X. Wang, D. S. Miller, E. Bukusoglu, J. J. de Pablo, N. L. Abbott,Nat. Mater. 15, 106–112 (2016).

9. T. J. White, D. J. Broer, Nat. Mater. 14, 1087–1098(2015).

10. K. Akagi, Chem. Rev. 109, 5354–5401 (2009).11. F. Bally-Le Gall et al., Chem. Eur. J. 23, 13342–13350 (2017).12. X. P. Deng, J. Lahann, J. Appl. Polym. Sci. 131, 40315 (2014).13. W. F. Gorham, J. Polym. Sci. A1 4, 3027–3039 (1966).14. K. M. Vaeth, K. F. Jensen, Macromolecules 31, 6789–6793

(1998).15. O. Schäfer et al., Synth. Met. 82, 1–9 (1996).16. F. Xie et al., Angew. Chem. Int. Ed. 56, 203–207 (2017).17. D. Demus, J. Goodby, G. W. Gray, H.-W. Spiess, V. Vill,

Handbook of Liquid Crystals (Wiley-VCH, 1998), vol. 2.18. Y.-K. Kim, X. Wang, P. Mondkar, E. Bukusoglu, N. L. Abbott,

Nature 557, 539–544 (2018).19. P. G. de Gennes, J. Prost, The Physics of Liquid Crystals

(Clarendon Press, 1993).20. N. Podoliak et al., RSC Advances 4, 46068–46074 (2014).21. W. Iglesias, N. L. Abbott, E. K. Mann, A. Jákli, ACS Appl. Mater.

Interfaces 4, 6884–6890 (2012).22. H. Wang, T. X. Wu, S. Gauza, J. R. Wu, S. Wu, Liq. Cryst. 33,

91–98 (2006).23. S. Faetti, M. Gatti, V. Palleschi, T. J. Sluckin, Phys. Rev. Lett.

55, 1681–1684 (1985).24. V. Designolle, S. Herminghaus, T. Pfohl, Ch. Bahr, Langmuir 22,

363–368 (2006).25. R. Kuroda, T. Honma, Chirality 12, 269–277 (2000).26. F. Castles et al., Nat. Mater. 11, 599–603 (2012).27. M. Mizusaki, K. Nakai, S. Enomoto, Y. Hara, S. Yusa, Polym. J.

49, 457–463 (2017).28. Y. Nakanishi, K. Okamoto, Jpn. J. Appl. Phys. 51, 041701

(2012).

ACKNOWLEDGMENTS

We thank Y. Dong, C. Lu, S. Rahmani, and B. Plummer forhelp with SEM and transmission electron microscopy;V. Subramanian and the University of Michigan Biophysics NMRcenter for help with the solid-state NMR analysis; and N. Kotovfor insightful comments on the manuscript. Funding: Thisstudy was supported by the Army Research Office(W911NF-11-1-0251 and W911NF-17-1-0575). We thank the MichiganCenter for Materials Characterization and University of WisconsinMaterials Research Science and Engineering Center (DMR-1720415). The experiments performed with blue-phase liquidcrystals at University of Wisconsin–Madison were supported by theU.S. Department of Energy (DE-SC0004025). J.L. and C.H.acknowledge support from the German Science Foundation(SFB 1176, Project B3), and F.X. acknowledges funding from theChina Scholarship Council and Northwestern PolytechnicalUniversity, China. Author contributions: N.L.A. and J.L. proposedand supervised the project; K.C.K.C., M.A.B.-P., and Y.-K.K.conducted experiments with assistance from J.V.G., F.X., A.d.F.,C.H., and K.S.; and K.C.K.C., M.A.B.-P., Y.-K.K., N.L.A., and J.L.analyzed the data and wrote the manuscript with input fromall coauthors. Competing interests: The University ofWisconsin–Madison and University of Michigan have filed apatent application (PCT/US17/27764) on the work described inthis manuscript. F.X. is also affiliated with the Department ofApplied Chemistry, School of Science, NorthwesternPolytechnical University, Xi’an 710072, China, and the College ofBioresources Chemical and Materials Engineering, ShaanxiUniversity of Science and Technology, Xi’an 710021, China.Data and materials availability: All data supporting thefindings of this study are available in the manuscript or thesupplementary materials.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/362/6416/804/suppl/DC1Materials and MethodsFigs. S1 to S16References (29–31)

22 December 2017; resubmitted 9 June 2018Accepted 8 October 201810.1126/science.aar8449

Cheng et al., Science 362, 804–808 (2018) 16 November 2018 4 of 4

5 µm

Micrometer-scale LC droplets on glass surfaces

F

H I

J K

G

E

500 µm1 µmLC-coated metal mesh

Region with fibers

Region without fibers

LC

5 µm20 µm 1 µm

5 µm100 µm

** *

** *

N vs RF vs F0

0.2

0.4

0.6

0.8

1.0

Forc

e [m

N]

N vs N N vs S S vs S R vs R S vs R

Glass Substrate

FiberArrays

Glass Substrate

ass Subst

ass Subst

CVP

LC removal

CVP

LC removal

(2)D (1)C

LC removal

LC coated on the inner surface

(1)

(2)

LCTube

5 µm

BA

LC coated on the outer surface

LCTube

LC removal

Force

Force

CVP CVP

Fig. 4. Templated CVP of polymeric nanofibers in complex geometries. (A to D) CVP of 1a intonematic E7 films coated on either the exterior (A) or interior (C) surfaces of glass capillariesand SEM images [(B) and (D)] of corresponding nanofibers [(1) indicates the region closest tothe orifice and (2) indicates the region 1.5 mm from (1) inside the cylinder]. (E and F) CVP of 1a intoE7 microdroplets on a glass surface (E) and SEM images of the nanofiber assemblies (F). (G andH) CVP of 1a into free-standing films of E7 hosted within a stainless-steel mesh (G) and SEM imageof suspended nanofiber film (H). (I) SEM image of a nanofiber membrane spanning the tip of aglass capillary, which was initially intact but was opened during microscopy, revealing an ultrathinnanofiber array. (J) Schematic illustration of two substrates coated with nanofiber arrays preparedby CVP. (K) Adhesion forces between pairs of substrates decorated by flat CVP films (F) or nanofiberarrays templated from nematic (N), left-handed cholesteric (S) or right-handed cholesteric(R) LC phases. Data are means ± SD; n ≥ 5 measurements. Statistical analyses were performedbetween groups using Tukey’s test; * indicates statistically identical results, P > 0.4; ** indicatesstatistically different results, P < 0.002.

RESEARCH | REPORTon O

ctober 8, 2020

http://science.sciencemag.org/

Dow

nloaded from

Templated nanofiber synthesis via chemical vapor polymerization into liquid crystalline films

Hussal, Kai Sun, Nicholas L. Abbott and Joerg LahannKenneth C. K. Cheng, Marco A. Bedolla-Pantoja, Young-Ki Kim, Jason V. Gregory, Fan Xie, Alexander de France, Christoph

DOI: 10.1126/science.aar8449 (6416), 804-808.362Science 

, this issue p. 804Sciencesheets for potential use in sensing or filtration applications.conditions could be used to tweak the arrangement of the liquid crystals to generate a wide range of polymer mats ordirect the growth of nonliquid crystalline polymers into sheets of highly ordered fibers. Small changes to the processing

exploited this preponderance toward long-range ordering toet al.technologies to extremely tough polymer fibers. Cheng The ability of liquid crystalline materials to order spontaneously has driven many innovations, from display

Patterned fiber formation

ARTICLE TOOLS http://science.sciencemag.org/content/362/6416/804

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2018/11/14/362.6416.804.DC1

REFERENCES

http://science.sciencemag.org/content/362/6416/804#BIBLThis article cites 28 articles, 1 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science. No claim to original U.S. Government WorksCopyright © 2018 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on October 8, 2020

http://science.sciencem

ag.org/D

ownloaded from