The ability to naturally integrate state-of-the-art electronic materials and devices representsan essential, defining characteristic of these ap-proaches. A mechanically tunable inductor basedon a 3D toroidal structure with feed and groundlines, all constructed with polyimide encapsu-lation (1.2 mm) and Ni conducting layers (400 nm),provides an example. Here, the geometry issimilar to the circular helix III in Fig. 2D, withthe addition of contact pads located at the pe-riphery for electrical probing. The graph of Fig.4D shows measurements and modeling resultsfor the frequency dependence of the inductanceand the quality (Q) factor for a 2D closed-loopserpentine precursor and a single 3D toroidstructure in two different mechanically adjustedconfigurations. In both cases, the 3D cage struc-ture enhances the mutual inductance betweenadjacent twisted turns. The maximum Q factorsand resonant frequencies increase systematical-ly from 1.7 to 2.2 GHz and from 6.8 to 9.5 GHz,respectively, as the structure transforms from2D to two distinct 3D shapes associated with par-tial release (about half of the total initial prestrainof 54%) and then complete release of the prestrain.These trends arise from a systematic reduction insubstrate parasitic capacitance with increasingthree-dimensional character (40). The measuredresults correspond well to modeling that in-volves computation of the electromagnetic prop-erties associated with the predicted 3D structuregeometries from FEA, as shown in the rightpanels of Fig. 4D [see (33) and figs. S20 to S23].The ideas presented here combine precise,

lithographic control of the thicknesses, widths,and layouts of 2D structures with patternedsites of adhesion to the surfaces of high-elongationelastomer substrates to enable rapid assembly ofbroad classes of 3D mesostructures of relevanceto diverse microsystem technologies. The process,which can be implemented with any substratethat is capable of controlled, large-scale dimen-sional change, expands and complements thecapabilities of other approaches in 3D materialsassembly. Compatibility with the most advancedmaterials (e.g., monocrystalline inorganics), fab-rication methods (e.g., photolithography), andprocessing techniques (e.g., etching, deposition)that are available in the semiconductor and pho-tonics industries suggest many possibilities forachieving sophisticated classes of 3D electronic,optoelectronic, and electromagnetic devices.

REFERENCES AND NOTES

1. V. B. Shenoy, D. H. Gracias, MRS Bull. 37, 847854 (2012).2. F. Li, D. P. Josephson, A. Stein, Angew. Chem. Int. Ed. 50,

360388 (2011).3. N. B. Crane, O. Onen, J. Carballo, Q. Ni, R. Guldiken, Microfluid.

Nanofluid. 14, 383419 (2013).4. J. H. Jang et al., Adv. Funct. Mater. 17, 30273041 (2007).5. J. Fischer, M. Wegener, Laser Photonics Rev. 7, 2244 (2013).6. K. A. Arpin et al., Adv. Mater. 22, 10841101 (2010).7. W. L. Noorduin, A. Grinthal, L. Mahadevan, J. Aizenberg,

Science 340, 832837 (2013).8. P. X. Gao et al., Science 309, 17001704 (2005).9. M. Huang, F. Cavallo, F. Liu, M. G. Lagally, Nanoscale 3, 96120

(2011).10. B. Tian et al., Nat. Mater. 11, 986994 (2012).11. T. G. Leong et al., Proc. Natl. Acad. Sci. U.S.A. 106, 703708 (2009).12. M. Yu et al., ACS Nano 5, 24472457 (2011).

13. D. Bishop, F. Pardo, C. Bolle, R. Giles, V. Aksyuk, J. Low Temp.Phys. 169, 386399 (2012).

14. R. J. Wood, Am. Sci. 102, 124131 (2014).15. R. Songmuang, A. Rastelli, S. Mendach, O. G. Schmidt, Appl.

Phys. Lett. 90, 091905 (2007).16. J. H. Lee et al., Adv. Mater. 26, 532569 (2014).17. M. Schumann, T. Buckmann, N. Gruhler, M. Wegener,

W. Pernice, Light Sci. Appl. 3, e175 (2014).18. X. Zheng et al., Science 344, 13731377 (2014).19. T. A. Schaedler et al., Science 334, 962965 (2011).20. C. M. Soukoulis, M. Wegener, Nat. Photonics 5, 523530 (2011).21. J. H. Cho et al., Small 7, 19431948 (2011).22. B. Y. Ahn et al., Science 323, 15901593 (2009).23. W. Huang et al., Nano Lett. 12, 62836288 (2012).24. H. Zhang, X. Yu, P. V. Braun, Nat. Nanotechnol. 6, 277281 (2011).25. K. Sun et al., Adv. Mater. 25, 45394543 (2013).26. W. Zheng, H. O. Jacobs, Adv. Funct. Mater. 15, 732738 (2005).27. X. Guo et al., Proc. Natl. Acad. Sci. U.S.A. 106, 2014920154 (2009).28. V. Y. Prinz et al., Physica E 6, 828831 (2000).29. O. G. Schmidt, K. Eberl, Nature 410, 168168 (2001).30. L. Zhang et al., Microelectron. Eng. 83, 12371240 (2006).31. G. Hwang et al., Nano Lett. 9, 554561 (2009).32. W. Gao et al., Nano Lett. 14, 305310 (2014).33. See supplementary materials on Science Online.34. D. C. Duffy, J. C. McDonald, O. J. A. Schueller, G. M. Whitesides,

Anal. Chem. 70, 49744984 (1998).

35. Y. Sun, W. M. Choi, H. Jiang, Y. Y. Huang, J. A. Rogers, Nat.Nanotechnol. 1, 201207 (2006).

36. D. H. Kim et al., Science 333, 838843 (2011).37. S. Yang, K. Khare, P. C. Lin, Adv. Funct. Mater. 20, 25502564 (2010).38. S. Singamaneni, V. V. Tsukruk, Soft Matter 6, 56815692

(2010).39. D. H. Kim, N. S. Lu, Y. G. Huang, J. A. Rogers, MRS Bull. 37,

226235 (2012).40. C. P. Yue, S. S. Wong, IEEE Trans. Electron. Dev. 47, 560568

(2000).

ACKNOWLEDGMENTS

Supported by the U.S. Department of Energy, Office of Science, BasicEnergy Sciences, under award DE-FG02-07ER46741. We thank S. B. Gongfor providing the RF testing equipment in this study, and K. W. Nan,H. Z. Si, J. Mabon, J. H. Lee, Y. M. Song, and S. Xiang for technical supportand stimulating discussions. Full data are in the supplementary materials.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/347/6218/154/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S23

7 September 2014; accepted 17 November 201410.1126/science.1260960

BIOMATERIALS

Electronic dura mater for long-termmultimodal neural interfacesIvan R. Minev,1* Pavel Musienko,2,3* Arthur Hirsch,1 Quentin Barraud,2

Nikolaus Wenger,2 Eduardo Martin Moraud,4 Jrme Gandar,2 Marco Capogrosso,4

Tomislav Milekovic,2 Lonie Asboth,2 Rafael Fajardo Torres,2 Nicolas Vachicouras,1,2

Qihan Liu,5 Natalia Pavlova,2,3 Simone Duis,2 Alexandre Larmagnac,6 Janos Vrs,6

Silvestro Micera,4,7 Zhigang Suo,5 Grgoire Courtine,2 Stphanie P. Lacour1

The mechanical mismatch between soft neural tissues and stiff neural implants hinders thelong-term performance of implantable neuroprostheses. Here, we designed and fabricatedsoft neural implants with the shape and elasticity of dura mater, the protective membraneof the brain and spinal cord. The electronic dura mater, which we call e-dura, embedsinterconnects, electrodes, and chemotrodes that sustain millions of mechanical stretchcycles, electrical stimulation pulses, and chemical injections. These integrated modalitiesenable multiple neuroprosthetic applications. The soft implants extracted cortical states infreely behaving animals for brain-machine interface and delivered electrochemical spinalneuromodulation that restored locomotion after paralyzing spinal cord injury.

Implantable neuroprostheses are engineeredsystems designed to study and treat the in-jured nervous system. Cochlear implantsrestore hearing in deaf children, deep brainstimulation alleviates Parkinsonian symptoms,

and spinal cord neuromodulation attenuateschronic neuropathic pain (1). New methods forrecording andmodulation of neural activity usingelectrical, chemical, and/or optical modalitiesopen promising therapeutic perspectives for neu-roprosthetic treatments. These advances havetriggered the development of myriad neural tech-nologies to design multimodal neural implants(25). However, the conversion of these sophis-ticated technologies into implantsmediating long-lasting therapeutic benefits has yet to be achieved.A recurring challenge restricting long-term bio-integration is the substantial biomechanical mis-match between implants and neural tissues (68).

Neural tissues are viscoelastic (9, 10) with elasticand shear moduli in the 100- to 1500-kPa range.They are mechanically heterogeneous (11, 12)and endure constant body dynamics (13, 14). Incontrast, most electrode implantseven thin,plastic interfacespresent high elasticmoduli inthe gigapascal range, thus are rigid comparedto neural tissues (3, 15). Consequently, their sur-gical insertion triggers both acute and long-termtissue responses (68, 14). Here, we tested thehypothesis that neural implants withmechanicalproperties matching the statics and dynamics ofhost tissues will display long-term biointegrationand functionality within the brain and spinal cord.We designed and engineered soft neural inter-

faces that mimic the shape and mechanical be-havior of the dura mater (Fig. 1, A and B, and fig.S1). The implant, which we called electronic duramater or e-dura, integrates a transparent silicone

SCIENCE sciencemag.org 9 JANUARY 2015 VOL 347 ISSUE 6218 159

RESEARCH | REPORTS

on

Janu

ary

8, 2

015

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Janu

ary

8, 2

015

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Janu

ary

8, 2

015

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Janu

ary

8, 2

015

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Janu

ary

8, 2

015

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

http://www.sciencemag.org/http://www.sciencemag.org/http://www.sciencemag.org/http://www.sciencemag.org/http://www.sciencemag.org/

substrate (120 mm in thickness), stretchable goldinterconnects (35 nm in thickness), soft electrodescoatedwith a platinum-silicone composite (300 mmin diameter), and a compliant fluidic microchan-nel (100 mm by 50 mm in cross section) (Fig. 1, Cand D, and figs. S2 to S4). The interconnects andelectrodes transmit electrical excitation and trans-fer electrophysiological signals. The microfluidicchannel, termed chemotrode (16), delivers drugslocally (Fig. 1C and fig. S4). The substrate, encap-sulation, and microchannel silicone layers arepreparedwith soft lithography and assembled bycovalent bonding following oxygen plasma activa-tion (Fig. 1D and fig. S2). Interconnects are ther-mally evaporated through a stencilmask,whereaselectrodes are coated with the soft composite byscreen-printing (fig. S2). Microcracks in the goldinterconnects (17) and soft platinum-silicone com-posite electrodes confer exceptional stretchabilityto the entire implant (Fig. 1D and movie S1).Most implants used experimentally or clin-

ically to assess and treat neurological disordersare placed above the dura mater (15, 1820). Thecompliance of the soft implant enables surgicalinsertion below the dura mater through a smallopening (Fig. 1, A and C, and fig. S5). This locationprovides an intimate interface between electrodesand targeted neural tissues (Fig. 1E) and allowsdirect delivery of drugs into the intrathecal space.To illustrate these properties, we fabricated im-plants tailored to the spinal cord, one of themostdemanding environments of the central nervoussystem. We developed a vertebral orthosis tosecure the connector (Fig. 1F) and dedicatedsurgical procedures for subdural implantation(fig. S5). The soft implant smoothly integratedthe subdural space along the entire extent oflumbosacral segments (2.5 cm in length and0.3 cm in width), conforming to the delicatespinal neural tissue (Fig. 1, E and F).We next tested the long-term biointegration of

soft implants compared to stiff, plastic implants(6 weeks of implantation). A stiff implant wasfabricated by means of a 25-mm-thick polyimidefilm, which corresponds to standard practices forflexible neural implants (21) and is robust enoughto withstand the surgical procedure. Both types ofimplants were inserted into the subdural space oflumbosacral segments in healthy rats. A sham-operated group of animals received the headstage,

connector, and vertebral orthosis but withoutspinal implant.To assess motor performance, we obtained

high-resolution kinematic recordings of whole-bodymovement during basicwalking and skilledlocomotion across a horizontal ladder. In thechronic stages, the behavior of rats with softimplantswas indistinguishable from that of sham-operated animals (Fig. 2A, fig. S6, and movie S2).By contrast, rats with stiff implants displayedsignificant motor deficits that emerged around1 to 2 weeks after implantation and deterioratedover time. They failed to accurately position their

paws onto the rungs of the ladder (Fig. 2A). Evenduring basic walking, rats with stiff implantsshowed pronounced gait impairments, includingaltered foot control, reduced leg movement, andpostural imbalance (fig. S6).The spinal cords were explanted after 6 weeks

of implantation. Both soft and stiff implants oc-cupied the targeted location within the subduralspace. Minimal connective tissue was observedaround the implants. To evaluate potential mac-roscopic damage to the spinal cord that may ex-plainmotor deficits, we reconstructed the explantedlumbosacral segments in three dimensions. A

160 9 JANUARY 2015 VOL 347 ISSUE 6218 sciencemag.org SCIENCE

1Bertarelli Foundation Chair in Neuroprosthetic Technology,Laboratory for Soft Bioelectronic Interfaces, Centre forNeuroprosthetics, Institute of Microengineering and Instituteof Bioengineering, Ecole Polytechnique Fdrale de Lausanne(EPFL), Switzerland. 2International Paraplegic FoundationChair in Spinal Cord Repair, Centre for Neuroprosthetics andBrain Mind Institute, EPFL, Switzerland. 3Pavlov Institute ofPhysiology, St. Petersburg, Russia. 4Translational NeuralEngineering Laboratory, Center for Neuroprosthetics andInstitute of Bioengineering, EPFL, Lausanne, Switzerland.5School of Engineering and Applied Sciences, Kavli Institutefor Bionano Science and Technology, Harvard University,Cambridge, MA, USA. 6Laboratory for Biosensors andBioelectronics, Institute for Biomedical Engineering,University and ETH Zurich, Switzerland. 7The BioRoboticsInstitute, Scuola Superiore SantAnna, Pisa 56025, Italy.*These authors contributed equally to this work. These authorscontributed equally to this work. Corresponding author. E-mail:gregoire.courtine@epfl.ch (G.C.); stephanie.lacour@epfl.ch (S.P.L.)

Fig. 1. Electronic dura mater, e-dura, tailored for the spinal cord. (A) Schematic cross section ofthe vertebral columnwith the soft implant inserted in the spinal subdural space. (B) Strain-stress curves ofspinal tissues, duramater, and implant materials. Plastics (polyimide), silicone, and duramater responsesare experimental data. Spinal tissue response is adapted from the literature (see supplementarymaterials). (C) Illustration of the e-dura implant inserted in the spinal subdural space of rats. (D) Opticalimage of an implant, and scanning electron micrographs of the gold film and the platinum-siliconecomposite. (E) Cross-section of an e-dura inserted for 6 weeks in the spinal subdural space. (F) Re-constructed 3D microcomputed tomography scans of the e-dura inserted in the spinal subdural spacecovering L2 to S1 spinal segments in rats.The scan was obtained in vivo at week 5 after implantation.

RESEARCH | REPORTS

cross-sectional circularity index was calculatedto quantify changes in shape. All the rats withstiff implants displayed significant deformationof spinal segments under the implant (P < 0.05,Fig. 2B), ranging frommoderate to extreme com-pression (fig. S7 and movie S2).Neuroinflammatory responses at chronic stages

were visualized with antibodies against activatedastrocytes and microglia (Fig. 2C), two standardcellular markers for foreign-body reaction (7). Asanticipated from macroscopic damage, both celltypes massively accumulated in the vicinity ofstiff implants (P < 0.05; Fig. 2C and fig. S8). Inmarked contrast, no significant difference wasfound between rats with soft implants and sham-operated animals (Fig. 2C and fig. S8). These re-sults demonstrate the long-term biocompatibilityof the soft implants.Wemanufactured amodel of spinal cord using

a hydrogel core to simulate spinal tissues and a

silicone tube to simulate the duramater (fig. S9A).A soft or stiff implant was inserted into the mod-el (Fig. 2D). The stiff implant induced a pro-nounced flattening of the simulated spinal cord,whereas the soft implant did not alter the cir-cularity of the model (Fig. 2D and fig. S10). Toprovide themodel with realisticmetrics, we quan-tified the natural flexure of the spine in freelymoving rats (fig. S9B).When themodel was bent,the stiff implant formed wrinkles that inducedlocal compressions along the hydrogel core. Incontrast, the soft implant did not affect thesmoothness of simulated spinal tissues (fig. S11).When the model was stretched, the stiff implantslid relative to the hydrogel core, whereas thesoft implant elongated together with the entirespinal cord (Fig. 2D and fig. S11). Reducing thethickness of the plastic implant to 2.5 mm im-proved bending stiffness and conformability. How-ever, the ultrathin, plastic implant still failed to

deform during motion of the soft tissue (fig. S10and supplementary text).Patterning extremely thin films into web-like

systems offers alternative mechanical designs forelastic surfaces (2224). For example, fractal-likemeshes develop into out-of-plane structures dur-ingmechanical loading, which facilitates reversibleand local compliance. Medical devices preparedwith such three-dimensional (3D) topologies canconform the curvilinear surface of the heart (25)and skin (23). However, this type of interfacerequires complex, multistep processing and tran-sient packaging. In comparison, fabrication stepsof e-dura are remarkably simple. Moreover, theshape and unusual resilience of the soft implantgreatly facilitate surgical procedures.The composite electrodes of the soft implant

displayed low impedance (Z = 5.2 T 0.8 kilohm at1 kHz, n = 28 electrodes) and maintained theelectrochemical characteristics of platinum (Fig. 3,

SCIENCE sciencemag.org 9 JANUARY 2015 VOL 347 ISSUE 6218 161

Fig. 2. Biointegration. (A) Hindlimb kinematics during ladder walking 6 weeksafter implantation. Bar plots reporting mean percentage of missed stepsaveraged per animal onto the rungs of the ladder (n = 8 trials per rat, n = 4rats per group). (B) 3D spinal cord reconstructions, including enhancedviews, 6 weeks after implantation.The arrowheads indicate the entrance ofthe implant into the subdural space. Bar plots reporting mean values ofspinal cord circularity index (4p area/perimeter2). (C) Photographs showingmicroglia (Iba1) and astrocytes (GFAP, glial fibrillary acidic protein) staining

reflecting neuroinflammation. Scale bars: 30mm. Heat maps and bar plotsshowing normalized astrocyte and microglia density. (D) Spinal cord modelscanned using microcomputed tomography without and with a soft or stiffimplant. e-dura implant is 120 mm thick. The red line materializes the stiffimplant (25 mm thick), not visualized because of scanner resolution. Plotreporting local longitudinal strain as a function of global applied strain.Statistical test: Kruskal-Wallis one-way analysis of variance (*P < 0.05; **P