Nanoscale Surface Topography Reshapes Neuronal Growth inCultureGhislain Bugnicourt,†,‡,§ Jacques Brocard,§ Alice Nicolas,∥ and Catherine Villard*,†,‡

†Institut Neel, Universite Grenoble-Alpes, F-38042 Grenoble, France and Institut Neel/CRETA, CNRS, F-38042 Grenoble, France‡CNRS, Inst NEEL and CRETA, F-38042 Grenoble, France§Institut National de la Sante et de la Recherche Medicale, U836-GIN, 38700 Grenoble Cedex 9, France∥CNRS-LTM-UJF, CEA-LETI 17 Av. des Martyrs, 38054 Grenoble France

*S Supporting Information

ABSTRACT: Neurons are sensitive to topographical cues provided eitherby in vivo or in vitro environments on the micrometric scale. We haveexplored the role of randomly distributed silicon nanopillars on primaryhippocampal neurite elongation and axonal differentiation. We observedthat neurons adhere on the upper part of nanopillars with a typical distancebetween adhesion points of about 500 nm. These neurons produce fewerneurites, elongate faster, and differentiate an axon earlier than those grownon flat silicon surfaces. Moreover, when confronted with a differentialsurface topography, neurons specify an axon preferentially on nanopillars. Asa whole, these results highlight the influence of the physical environment inmany aspects of neuronal growth.

■ INTRODUCTION

In vitro neurons are usually plated on glass coverslips or Petridishes. However, these substrates provide to developingneurons a flat and uniform environment that contrasts withthe complex tridimensional topography of the embryonicbrain.1 Moreover, cellular adhesion involves adhesive com-plexes mediated by transmembrane heterodimers, namedintegrins, that are established on the micrometric2 and evensubmicrometric scale.3 Understanding the mechanisms of celladhesion might therefore benefit from the use of micro-structured surfaces.In line with this reasoning, the specific interaction between

neurons and different micropillar geometries has been explored.A common feature observed on these pillared surfaces is aneurite channeling effect between pillars when the interpillarspacing is larger but close to the neurite width (∼1 to 2μm).4−6 Another remarkable effect is the accelerated neuriteelongation provided by micropillars.4,6 Interestingly, axons havebeen reported to specify preferentially within micropillars areascompared to control flat surfaces,6 a feature also observed whenmicropillars are replaced by submicroscale holes.7 Anothergeneric behavior is the transition between a neurite channelingeffect between pillars to a neurite growth on pillars. Thistransition seems to occur for interpillar distances shorter thanthe neurite width,5,6 possibly because of geometrical arguments.A stringent restriction of elementary adhesive areas up to the

nanometric scale was also explored. Spatz and Geiger usedregularly spaced adhesive gold nanoparticles to demonstratethat cell spreading is an active process controlled by density-

dependent integrin signals.8 The role of tridimentionaltopographies on the nanometer scale has also been studiedusing GaP vertical nanowires (GaP-NWs) with typicaldiameters in the range of 50−80 nm. Depending on theinter-NW distance, a channeling effect,9 neurite growth on topof GaP-NWs bidimentional networks,10 or neurite guidanceabove GaP-NWs rows was reported.11 Growth on top of GaP-NWs was observed when the distance between GaP-NWs wasabout half a micrometer.Beside these fundamental aspects of neurobiology that

ultimately might concern the issue of neuroregeneration,acquiring knowledge about the interaction between neuronsand nanostructured surfaces meets the requirements of recentintracellular recording techniques based on the insertion ofvertical nanostructures through cell membranes.12,13

Following the studies on the interaction between GaPnanowires and either retinal10 or cervical and dorsal rootganglia neurons,9,11 the present work explores in detail theeffect of randomly distributed silicon nanopillars in the primaryhippocampal neurite elongation rate and axonal polarization. Inthe course of our study, we were led to discriminate betweenthe role of topography and effective rigidity and to evaluate thedirectional choices performed by developing neurons accordingto the spatial distribution of nanopillars. How neurons behaveat the frontier between nanostructured and flat surfaces was

Received: January 14, 2014Revised: March 19, 2014Published: March 21, 2014

Article

pubs.acs.org/Langmuir

© 2014 American Chemical Society 4441 dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−4449

also explored. Our results reinforce the growing body ofevidence of the role of topographical cues on thesubmicrometer scale and show that spatially distributednanometric adhesive areas over submicrometer distancesinfluence many aspects of neuronal growth during the firststages of development.

■ MATERIALS AND METHODSStructuration of Silicon Surfaces. Nanopillared surfaces were

prepared by reactive ion etching performed on silicon substrates cutfrom 51 mm in-diameter wafers (four substrates per wafer). This 50 Wetching process involves SF6 gas (64% in volume) and O2 (36%). Theoxygen-based plasma forms a silicon oxide layer on the exposed Sisurface, and the SF6-based plasma etches both the silicon and thesilicon oxide (SiO2), the latter at a lower speed. The combination ofthese two gases gives rise to a competition between etching andpassivation. This complex phenomenon after about 15 min produces acarpet of silicon nanopillars capped by a SiO2 layer and spaced bysubmicrometric distances.Etching was not uniformly performed on the whole sample surface

but restricted to rectangular, millimeter-sized areas (Figure S1 in theSupporting Material). This large-scale modulation of the substratetopography was achieved by classical UV photolithography stepsincluding Shipley S1818 photoresist spinning (4000 rpm, 1.8 μmthickness, 115 °C annealing step for 1 min), insulation through amask, and development (Microposit concentrate 1:1, Shipley). Afterthe etching process, the remaining layer of photoresist protecting thesilicon surface was removed in pure acetone. Then, a pure oxygenplasma (2 min, 50 W) was used to obtain hydrophilic surfaceproperties suitable for the next step of cellular functionalization forboth pillared and flat silicon surfaces.Neuronal Culture and Labeling. Mouse hippocampal neurons

(E18.5) were prepared and seeded over poly-L-lysine-covered surfacesas previously described.14 Poly-L-lysine at a concentration of 1 mg/mLincubated overnight at room temperature was used for cell adhesion.In some experiments, a laminin coating was also used at a

concentration of 10 μg/mL incubated for 6 h at 37 °C after thedeposition of poly-L-lysine.

Two types of fixation protocols were used depending on theimaging technique. For optical observations, we used paraformalde-hyde (PFA) according to the following protocol: (i) the plates wereincubated at 37 °C for about half an hour with a solution of PFA/sucrose (paraformaldehyde 4%, 120 mM sucrose, PBS), (ii) after beingwashed in PBS, cells were permeabilized for a few minutes in PBS/Triton X-100 0.1%, and (iii) Triton was then eliminated by rinsingwith PBS. For scanning electron microscopy, the PFA protocol wasused with the addition of glutaraldehyde 0.5% and without the step ofmembrane permeabilization (i.e., without Triton). The sample wasthen dehydrated by a successive few minutes of dipping in 50%(diluted with water) and then 100% acetone followed by immersion in50% (diluted with acetone) and then 100% hexamethyldisilazane(HMDS).15 Samples were finally allowed to dry slowly under a hood.

Primary antibodies were Tau (clone tau-1, Millipore), rat mAbagainst tubulin (clone YL1/2), and mouse antivinculin (Sigma).Secondary antibodies were Alexa488 or Cy3 coupled (MolecularProbes, USA). F-actin was stained with Texas Red-phalloidin (lifetechnologies). Neurons were observed with two different microscopes:a Zeiss Ultra scanning electron microscope and a BX51 opticalmicroscope (Olympus, Inc.) using 10×, 20×, or 40× dry objectivescombined with an F-View II camera.

Image Analysis. Automatic measurements were performed usingfree software ImageJ16 with custom-made plugins specificallydeveloped for this study.

Silicon Topography. Top-view scanning electron microscope(SEM) images from (magnification 24K×, polaroid reference) werefirst processed in ImageJ (smooth, threshold, remove outliers) andthen loaded into Gwyddion17 to extract pillar radius and centroidcoordinates. The coordinates were further analyzed in ImageJ bycalculating all of the distances for each pillar and sorting them inascending order to get the nth-neighbor distances. Note that these nth-neighbor distances were kept only for the pillars situated in the centerof the image (in a square whose surface is a fourth of the total squareimage) to avoid side effects.

Figure 1. Morphology and characteristic lengths of silicon nanopillared surfaces. (a) SEM micrographs (top view and section). Scale bars: 500 nm.(b) Histograms of the equivalent radius of the nanopillars (Req) and of the different characteristic distances between nanopillars. The mean values ofthe distributions are indicated in parentheses. Req (35 ± 17 nm, n = 350), first neighbor (210 ± 60 nm, n = 349), and square root of the distancebetween nanopillars in a given arbitrary direction (i.e in our case along an horizontal line, 29.67 ± 12.24 (nm)1/2, giving a representative length of29.672 ≈ 880 nm, n = 92) and under a neurite (adhesion points, 22.96 ± 5.21 (nm)1/2, giving a representative length of 22.962 ≈ 527 nm, n = 222).(For all values, we give the mean ± standard deviation of the distribution, with n denoting the number of measured distances.)

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494442

For straight-line distances, five horizontal lines (1 pixel thick) weredrawn arbitrarily on the image. Around four pillars were found alongeach line, but only the first interpillar distance was taken into account.The root square of the mean distances was calculated to obtain anormal distribution whose mean was used as the representative linedistance. A similar method using the mean value of the distribution ofthe root square distances was employed to compute the adhesionpoint distance.Other lengths, surfaces, and angles were measured manually using

the segmented line tool of ImageJ.Cellular Lengths. The automatic measurements developed for the

needs of this study consist of (i) thresholding the image ofmicrotubules in fluorescence, (ii) skeletonizing the signal by themethod18 to transform both neurites and somas into white lines of 1pixel thickness on a black background, and finally (iii) setting the lineswidth to 3 pixels with the ImageJ dilate function for binary images.The total length of cells in a given image is then obtained by dividingthe total white area by 3. Note this method was proven to make fewererrors in the evaluation of neurite length than 1 pixel skeletonizationcompared to manual measurements. The number of cells in the imageis determined by manual counting. Only optical micrographs withreasonable cell densities of between 7 and 16 neurons per field of 892× 673 μm2 (about 0.6 mm2) were used.Finally, the ratio of the total cell length of the number of cells gives

the average neurite length per cell. Note that although many cells haveneurites coming out of the image there are statistically as manyneurites entering the same optical field.Statistics. All percentage comparisons were performed using χ2

tests. Quantitative measurements were analyzed via a Kolmogorov−Smirnov test to assess normality and then compared using a standardunpaired t test. All calculations were performed in Excel (Microsoft).

■ RESULTS

Topographical Characteristics of Silicon Surfaces. Theetching process changed the initial flat silicon surface into ananopillared surface (Figure 1a, top) defined by an isotropic(Figure S2a−c) pillar density of about 3/μm2 (i.e., 300 × 106

nanopillars per cm2) and statistically distributed characteristicinterpillar distances (Figure 1b). We estimated these distancesaccording to two different methods. The first neighbor distanceis a usual physical parameter that gives the average distance tothe closest pillar. (See Figure S2d in the SupportingInformation for the example of a series of nth neighbordistances.) The line distance assesses the average distancebetween a pillar and the first pillar met along a straight lineevoking the directional persistency of neurites and of axons inparticular.19 The first-neighbor distance (210 ± 60 nm) is thesmallest one (with 880 ± 150 nm for the line distance, Figure1b). Typically, the nanopillars adopt a tapered shapecharacterized by a typical height of 700 nm and a main radiusof 35 ± 17 nm (Figure 1b). The formation of this surfacetopography is accompanied by a digging of the bulk silicon,leading to the presence of a step at the border between flat andnanostructured surfaces (Figure 1a, bottom).Three silicon substrates (Materials and Methods) were

analyzed. Twenty images were used for the determination of allthese lengths except for the adhesion-point distance (24images).

Neurons Form Adhesions on the Upper Parts ofNanopillars. We observed, as expected from the submicro-metric values of the interpillar distances, that neurons did notdevelop at the basis of the nanopillars. The cells rather display

Figure 2. Neurons adhere to the upper parts of nanopillars. (a) SEM micrographs of a neuron after fixation and dehydration. Right: high-magnification view of the area boxed in the left image. Scale bars: 5 μm (left) and 1 μm (right). (b) Actin (Phalloidin, red), microtubules (YL1/2,green), and nucleus (Hoechst, blue) immunolabeling. The details of the actin structure are displayed on the high-magnification views of the areasboxed in the left image. Scale bars: 10 μm (left) and 1 μm (right). (c) Actin (Phalloidin, red) and vinculin (antivinculin, green) immunolabelingalong a neurite. The two colored images are slightly shifted to reveal each staining separately. Scale bar: 2 μm.

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494443

adhesive contacts close to the free extremities of the nanopillarsby clinging either to their tops or to their top edges. This leadsto the discretization of their adhesion at both the soma andneurite levels (Figure 2a). The mean distance between neuriteadhesive contacts is 527 ± 27 nm, an intermediate valuebetween the first neighbor and line distances (Figure 1b). Thisindicates that growing neurites that cannot bend enough toselect the closest peak do not grow straight but instead makedirectional choices.We decided to build a probabilistic model to quantify this

directional choice. This model should answer the question,what would be the mean distance Lα between two adhesionpoints if the probability of finding the next nanopillar isrestricted to an angle α? Lα can be written as a series of termsexpressing the probability, weighted by the distance to the nth

neighbor (Dn), of finding a pillar in an arc α knowing that nopillar has been found for the previous (n − 1) neighbors. (SeeFigure 3a for a representation of these parameters.) Knowingthat the probability of finding a peak in an arc of angle α is theratio α/2π, the first term in the mathematical series is L1 =

a/2π ×D1. The second term can be written as L2 = (1 − a/2π)(

a/2π) ×D2 (i.e., as the probability (1 − a/2π) of not finding the firstneighbor in an arc of angle α multiplied by the probability a/2πto find the second neighbor in an arc of angle α. Similarly, thethird term is L3 = (1 − a/2π)

2 × a/2π × D3, leading to theexpression for the whole series

∑ ∑ απ

απ

= = − × ×α=

∞

=

∞ −⎜ ⎟⎛⎝

⎞⎠L L D1

2 2nn

n

n

n1 1

( 1)

(1)

Figure 3. Neurite directional choices. (Left) Enlarged binarized top view of the rough surface. The first-neighbor distance D1 is sketched togetherwith Lα, which represents the distance to the first neighbor in an angle α (in an arbitrary direction because the peak distribution is isotropic). Thisangle can be calculated on the basis of the Dn values and the experimental distances Lα between adhesive points (eq 1). (Right) Histogram of thedistribution of the angle α expressing the angular tolerance to bending of neurites developing on top of nanopillars. Each count corresponds to theanalysis of one SEM micrograph providing on average 20 values of distances between adhesion points (n = 44). The curve is a Gaussian fit of thedata, leading to α = 48 ± 20° (standard deviation).

Figure 4. Neuronal development at 3DIV on nanopillared and flat silicon surfaces. (a) Example of neuronal development for each condition.Microtubules (YL1/2, green) immunolabellings. Scale bars: 50 μm. (b) Distribution of the total neurite length, the neurite number, and the meanneurite length per cell (n = 397 for pillared surfaces and n = 353 for flat surfaces, three cultures, three silicon substrates per culture) obtained throughthe analysis of about 80 optical micrographs per condition. The mean neurite length is calculated, for each micrograph, from the total length dividedby the number of neurites (***, p < 0.001).

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494444

Then, the series of the nth neighbor distances Dn and themean distance between adhesion points (expressed as Lα in eq1) were measured in the same areas, and the value of α wasassessed numerically by equalizing the two terms in eq 1. Fromthe analysis of a large number of images (n = 44, leading to nvalues of Lα, themselves calculated from an average of 20distances between adhesion points), we obtained the histogramshown in Figure 3b.The most probable value of α is 48 ± 20° (standard

deviation). On the scale of adhesions, the neurite thus toleratesan angle of about 25° to bend on either side of the meantrajectory. This means that the neurite would stretch andeventually unhook from nanopillars too distant from theaverage trajectory if the path followed by the growth conewould lead to a greater neurite curvature. Interestingly, thevalue of 25° is very close to the range of orientational changesin the growth direction of chick embryos and Xenopus axons ona time scale of 10 min, as reported by Katz.19

Growth cones themselves adhere on top of the nanopillars.To determine if this discontinuous adhesive surface had aneffect on their shape, we labeled actin filaments (F-actin) withphalloidin to reveal growth cones at the neurite tips.Interestingly, no differences were observed in the generalshape between nanopillars and flat surfaces: growth conespresented very similar surfaces and fluorescence intensities, andthe percentage of active growth cones (i.e., the percentage ofneurite tips terminating in a large actin structure evocative of agrowth cone) among all neurite extremities was unchanged(Figure S3).Of note, actin and vinculin labelings show a discrete

distribution of these proteins characterized by the presence ofspots separated by submicrometer distances (Figure 2b,c). Thismay indicate the existence of adhesion complexes around peaksas evidenced by the presence of vinculin, a signaling proteinthat couples integrin receptors to actin filaments and actin.11,20

In brief, we achieved a discontinuous adhesive environmentfor neuronal growth, characterized by pointlike adhesions of thetypical size of elementary integrin structures separated bydistances on the order of magnitude of the neurite diameter.Moreover, the macroscopic alternation of nanopillared and flatareas created frontiers along which individual neurons cansimultaneously explore two different physical environments.We therefore studied how neurons developed on nanopillaredcompared to flat silicon surfaces and how they located theiraxons when confronted with a differential surface topography.

Influence of Nanopillars on Neuronal Development.Neuronal Development at 3DIV (Days in Vitro). We focusedon three morphological parameters: the total neurite length, thenumber of neurites emerging from the soma, and the meanneuritic length that is obtained by dividing the total length bythe neurite number. The graphs in Figure 4, in which eachpoint results from the analysis of one field of about 0.6 mm2

(Figure 4a), present a comparison of neuronal growth onnanopillared and flat areas at 3DIV. It appears that neurons aremore developed on nanopillars (total neuritic length of 636 ±165 μm compared to 506 ± 142 μm, p < 0.001, ***), producefewer neurites (3.6 ± 0.5 compared to 4.6 ± 0.7, p < 0.001,***), and therefore present a longer mean neurite length (180± 47 μm compared to 112 ± 32 μm, p < 0.001, ***). Note thatour control condition (flat silicon surfaces covered by a nativesilicon oxide) gave results very similar to those in other studieson glass substrates,21 as expected from their similar surfacechemistry.Beside, the reduced neurite number on nanopillared surfaces

raises the question of what happens in earlier stages of cellspreading and development. We therefore conducted a similarstudy at 2DIV and at 10 h after plating.

Neuronal Development at Early Stages. We found that, asfor 3DIV, neurons developed longer neurites at 2DIV onnanopillared as compared to flat surfaces. The total neuritelength is now 438 ± 76 μm compared to 382 ± 72 μm (p <0.01, **), and the mean neurite length is 125 ± 22 μmcompared to 86 ± 17 μm (p < 0.001, ***) (Figure 5a) for aneurite number of 3.5 ± 1.4 as compared to 4.6 ± 1.4 (p <0.001, ***). The number of neurites is, as expected, notsignificantly different from that measured under the 3DIVcondition.To determine when the number of neurites diverges between

the two types of substrates, we looked at the very first stage ofdevelopment (i.e., 10 h after plating (h10 condition)). At thisstage, newly formed neurites still cohabit with lamellipodia. Onaverage, we counted 1.05 neurites and 1.2 lamellipodia onnanopillars (n = 86), and very similar values were obtained on aflat silicon surface (1.09 neurites and 1.26 lamellipodia, n = 99).We also measured the lamellipodia surface and obtained 71.6 ±39 μm2 on nanopillars compared to 81.2 ± 50.6 μm2 on flatsurfaces. In addition, the length of the longest neurite underboth conditions was not significantly different, with 25 ± 11 μm(nanopillars, n = 38) and 24 ± 14 μm (flat surfaces, n = 47). Nomorphological differences were thus found between the two

Figure 5. Neuronal growth at 2DIV on nanopillared and flat silicon surfaces. Distributions of (a) the mean neurite length and (b) the total lengthobtained at 2DIV; n = 397 (nanopillared surface) and n = 323 (flat conditions). Data were obtained from an analysis of 33 optical micrographs percondition (1 culture, 2 silicon substrates) (**, p < 0.01; ***, p < 0.001).

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494445

physical environments, or in other words, nothing distinguishesneurites or lamellipodia that formed on nanopillars or on flatsurfaces in a very early stage of development. This may indicatethat the differential development between nanopillared and flatsurfaces may occur only when all neurites are formed,presumably between h10 and 1DIV.Neuronal Development on Laminin. The above results

show that the surface topography can hasten neurite elongation,even without any morphological guidance. Growth enhance-ment was previously described in the case of multiple adhesionproteins, including laminin (LN).22 These proteins generatespecific signals that are very different from the physical signal ofnanopillared surfaces, so it seemed interesting to coat oursamples with LN on top of PLL to confront the effects of thetwo types of cell environments.On both nanopillared and flat topographies, neurons were

more developed with a PLL/LN coating compared to PLLalone, confirming that LN is a strong growth enhancer.However, neurons on the pillars + LN surface were not moredeveloped (660 ± 174 μm) than those on the flat + LN surface(773 ± 138 μm) (n ≈ 200 neurons per condition, 1 culture).Interestingly, the number of neurites was not significantlydifferent between the nanopillared (4.05 ± 0.6 neurites) andflat (3.85 ± 0.5 neurites) surfaces.Influence of Nanopillars on Axonal Polarization.

Polarization Rate. Neuronal elongation on nanopillars isaccelerated. Knowing that the nascent axon is, as described invitro, the longest process,23 an accelerated elongation rate mayaffect the rate of polarization. The majority of neurons grownon poly-L-lysine (PLL) are polarized within 48 h after plating.21

We thus counted the number of polarized neurons (assessed by

Tau staining) on nanopillars and on flat surfaces coated withPLL at 2DIV (one culture) and 3DIV (two cultures). Thedifference between the two adhesive conditions is dramatic at2DIV: a rate of 74.2% polarization was obtained on nanopillars(n = 260 cells) whereas flat surfaces generated significantlyfewer polarized neurons (57.9%, n = 235, p < 0.01, **).However, this disparity vanishes at 3DIV where an asymptoticpolarization is reached under both conditions, with 76.8% (n =164 cells) and 74.7% (n = 95) for the polarization rate onnanopillared and flat silicon surfaces, respectively.Nanopillars thus increase the rate of neurite elongation and

polarization. In this context, we looked at the polarization whenthe two physical environments were simultaneously availablefor a given cell.

Polarization at Frontiers. As the elongation rate is enhancedon nanopillars, the question of the selectivity of this weaklyadhesive environment toward axonal differentiation is an openand sound question. To answer it, the notion of theneighborhood with respect to the boundary between nano-pillars and flat surfaces must be specified to select the pertinentpopulation of boundary neurons.The mean neurite length before polarization has been

estimated to be 35 μm.22 In this context, a simple selection rulefor a boundary neuron is that at least one of its neurites musthave crossed that boundary before reaching a length of 35 μm(sketch in Figure 6b). In other words, neurons are selectedsuch that the distance between their soma and the border alongthe path followed by at least one neurite does not exceed 35μm. Neurons are then classified according to the side of theborder where somas and axons are found (one example of such

Figure 6. Axonal polarization at frontiers. (a) Optical micrograph of a neuron crossing the border between the flat and nanopillared surfaces.Microtubules (YL1/2, green), axon (Tau, red), and nucleus (Hoechst, blue) immunolabellings. Scale bar: 25 μm. (b) Sketch representing the rule ofselection of boundary neurons from the use of a threshold in neurite length of 35 μm between the soma center (soma are represented by greencircles) and the frontier between the nanopillared and flat surfaces (in the text). The abbreviations “check” (V) and “uncheck” (X) identify selectedor rejected neurons, respectively. (c) Percentage of axons at 2DIV and 3DIV found exclusively on nanopillared surfaces for soma localization oneither side of the frontier (i.e., on either flat or nanopillared (nano-P) surfaces). Dashed lines denote the theoretical percentage expected without anyinfluence of the frontier (p < 0.001, ***, one culture; NS, nonsignificant). See Table 1 for the numerical values attached to each condition.

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494446

neurons is given in Figure 6a), and the results are reported inTable 1 (see also Figure 6c).

Then, to estimate the theoretical percentage of axons onboth sides in the case in which the composite adhesiveenvironment would not affect the localization of axonalspecification, we first measured the average distance betweenthe soma and the frontier in the population of selected neurons.This experimental average distance was 10 μm. We then draw a35-μm-radius circle whose center, in modeling the position of asoma, is located 10 μm from the boundary (see the sketch inFigure 6b for a visual description of our procedure) whereas thecircle represents the border that a neurite should be the first tocross to become the axon. In the absence of any bias providedby the environment, the axon should go beyond that circle in arandom orientation. Thus, for a soma located on the flatsurface, the ratio of the perimeter included in the nanopillar(pNP) area on the total perimeter (pT) gives the theoreticalpercentage of axons expected in the weakly adhesive area in theabsence of any bias. Numerically, this percentage p can bewritten as (see Figure 6 for the signification of α)

απ π

= = × = ≈( )

pp

pR

R2

2

arcos0.40 or 40%

T

NP

1035

with R = 35 μm. Conversely, the percentage of axons expectedin the flat area in the absence of any bias is 60%. The samereasoning applies for a soma located on the nanopillared area.The comparison of experimental results to the theoretical

percentage (Table 1) demonstrates that there is a significantaxonal preference at 3 DIV for nanopillared areas for somabeing on either side of the boundary (p < 0.001, ***, Figure6c).Remark: Axonal versus Dendritic Length. Nanopillars

increase the rate of neurite elongation and polarization. Asimilar phenomenon was observed on flat coverslips usinglaminin instead of poly-L-lysine. However, laminin is known toenhance the axonal elongation selectively, and both the numberof neurites and the length of the other processes areunchanged.21 We therefore checked the relative length of theaxons that grow on nanopillared and on flat silicon surfaces andcompared these values to the total length of the neurites percell. On average, an axon on top of nanopillars displays thesame increase of length (around 22%) as the whole neuritenetwork, indicating a nonspecific effect of these nanostructuresregarding axonal elongation. In conclusion, the dendritic lengthseems to compensate for the reduction in the number of

neurites observed on nanopillars so that the total length of thetwo poles of the cell (axon/dendrites) retains the sameproportion of lengths.

Discussion. About the Development of Neurites onNanopillars. Neurite elongation on nanopillars is accelerated.Despite the high rigidity of bulk silicon (in the range ofhundreds of GPa), strongly bent nanopillars are sometimesobserved. (See for example the pillars located in the boxed areaof Figure S3.) Therefore, we might consider that nanopillarswould display lower effective rigidities than bulk silicon becauseof their high aspect ratio. Cells in general are very sensitive totheir mechanical environment, and a dramatic change in neuritenumber and length below a rigidity of 100 Pa in the PC12neuronal cell line,24 as well as an increase of branching on verysoft gels,25 has been reported. The relationship between theYoung modulus of the bulk material that composes a rod (Ebulk)and the apparent Young modulus of the rod itself (Eeff) is afunction of its radius and length and takes the form

= ⎜ ⎟⎛⎝

⎞⎠E E

RH

2716eff bulk

3

with H being the height and R being the radius of the rod(details in Supporting Information S4).Nanopillars are composite structures made of silicon (ESi =

170 GPa) and silicon dioxyde (ESiO2= 50 GPa) produced by

the etching process. From the geometrical parameters ofnanopillars (i.e., H = 700 nm and R = 35 nm), we obtain Eeff =10.5 MPa using the lowest Young modulus (i.e., ESiO2

).Although this value is much lower than any bulk values, thisrigidity remains in a range known to have no effect on neurondevelopment.26 Therefore, even by considering the extremesituation where the nanopillars might display linear elasticbehavior in a large range of deformations, the changes weobserved in neuronal elongation probably do not result fromthe flexibility of nanopillars.Accelerated directional neurite elongation was reported on

submicrometric grooves formed on polyurethane acrylatesubstrates, providing a 1D anisotropic adhesive environmentfor cells.27 This phenomenon was associated with the presenceof a stable, fully adherent filopodia population aligned with thegrooves and a relative destabilization of the perpendicularfilopodia that experience the substrate striations transversally.On micropillared surfaces, the faster axonal elongation wascorrelated to a decrease in the overall growth cone size,probably as a result of a narrowing of this structure that occursbetween pillars. This result was explained in terms of possibleboosts in growth at pillar contacts.6 In our study, it isinteresting to note that, although directional choices are madeby neurites in their positioning on the top of nanopillars, theaccelerated elongation that occurs in this somehow pointlike,0D adhesive configuration is observed in the absence of anychanneling effect such as that provided by the 1D topographyof grooves or the presence of micropillars. Our results suggestthat the primary signal of accelerated elongation might be givenby the population of transverse, weakly adherent filopodia thatsense the presence of lateral topographies provided either bymicropillars or nanopillars and grooves and that this signal mayhave a significant effect without even altering the overall shapeof growth cones.Chemical and topographic environments are probably not

sensed by neurons the same way. Pillars produce a physicalsignal, intrinsically linked to their spatial organization and

Table 1. Axonal Localization of Boundary Neurons for SomaLocated on Each Side of the Frontier SeparatingNanopillared and Flat Surfaces

soma onnanopillars

soma onflat

surfaces

totalnumber of

cells

2DIV number of boundaryneurons

33 31 64

percentage (number) ofaxons on nanopillars

67% (22) 45% (14) 56% (36)

3DIV number of boundaryneurons

46 38 84

axons on nanopillars 85% (39) 66% (25) 76% (64)theoretical percentages of axons onnanopillars

60% 40%

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494447

therefore sensed “as a whole”. On the contrary, chemical cuessuch as laminin produce a specific signal on the molecular level,which allows it to induce a modulated response depending onits concentration21 or its surface density (in the present work,the effective coating density is reduced by an order ofmagnitude by nanopillared compared to flat surfaces).Considering furthermore that the two different signals mayalso activate competing signaling pathways, we face a rathercomplex situation ruling out the possibility of a simple additiveeffect between chemical and topographic cues. Interestingly,nanopillars combined with a laminin coating do not induce areduction in the number of neurites, in contrast to what isobserved under nonspecific adhesion conditions (i.e., with PLLcoating). This result is surprising knowing that lamininincreases only the length of the axon and therefore shouldnot affect the growth dynamics of undifferentiated neurites inthe first stage of development. Although puzzling, neuronalgrowth on nanopillars reveals that the mode of action oflaminin is more complex than expected because it may alsoregulate the number of neurites.Last, the clear reduction of the neurite number on PLL-

coated nanopillars combined with the absence of morphologicaldifferences after 10 h in vitro suggests that only the last stagesof neurite formation following the partial collapse of theprimary lamellipodium around the cell body28 are affected bytopographic cues.The mechanisms at the origin of the accelerated neurite

elongation provided by the fragmentation of adhesive surfacesare still unknown, and only speculative hypotheses could beformulated at this stage. Among them, an increase in neuritetension, resulting from the spacing between adhesion points,could be pertinent for two reasons: stretching neurites results inan accelerated elongation that eventually leads to theirdifferentiation into axons,29 and axons displayed periodicactin rings wrapped around their circumference spaced bysubmicrometric distances.30 These actin rings may play a role insustaining the mechanical strains to which axons are subjected.In our work, we have identified discrete actin/vinculinstructures whose spacing might be imposed by the distancebetween adhesive structures provided by the contact of neuriteswith the top of the nanopillars. An increase in neurite tensionmediated by these discrete actin structures is a hypothesisdeserving specific investigation in future work.Consequence of Nanopillars on Axonal Polarization. The

accelerated elongation rate on nanopillars is accompanied by apreferential axonal localization for cells located close to the flat/nanopillar frontiers. Note that this effect was not yet significantafter 2DIV, which might indicate that neurons that arepolarized later exhibit a greater sensitivity to their environment,presumably because they have already spent more time toexplore it.But neurites crossing this frontier are also confronted by a

step of 2 μm height, as shown in Figure 1a. The work of Li andFolch31 has shown that a step of 2.5 μm separating two flatareas is not sufficient to localize the axons. The step height hasto reach a value of 11 μm to prevent the passage of 50% of theaxons, regardless of the crossing direction. These results wereinterpreted as a resistance to curvature, leading to the choice ofa path that minimizes axonal bending: according to the stepheight, an axon may prefer to avoid crossing the step and bendalong the frontier. From these results, the 2 μm steps thatseparate the nanopillars from the flat areas should not have asignificant influence, especially because these steps are not

purely vertical, which reduces the curvature necessary for theircrossing. In addition, a possible influence of the 2 μm stepheight difference would be symmetrical: neurons on flat areasshould have more axons on this side of the frontier and viceversa. We observe on the contrary more axons on nanopillarscompared to the theoretical estimations, regardless of the somalocation. The effect of the steps themselves, if any, is thereforefully dominated by the nanopillars’ selective environment itself.Let us note that finding more axons on a surface that

increases the elongation rate is consistent with the notionproposed by Craig et al.32 and further explored with the use ofmicropatterns33 for the process of axonal specification triggeredwhen the first undifferentiated neurite exceeds a lengththreshold.

■ CONCLUSIONSWe have shown in this work that a change from uniform todiscontinuous adhesive conditions on the nanometric scaleinfluences many aspects of neuronal growth. Our results showthat neurite elongation is somehow guided but most of allaccelerated under the weak adhesion conditions provided bysilicon nanopillared surfaces, leading to a preferential local-ization of axonal specification.This major influence of the physical environment raises

important questions relative to the mechanism of neurito-genesis and neurite elongation that should motivate furtherstudies. So far, the nanopillared surfaces used in this study wereobtained using a simple plasma etching process. However, therecent work of Xu et al.,30 by revealing the existence of periodicactin rings in the axon, should motivate the search for periodiccellular adhesive complexes associated with these actin/vinculinstructures. Thus, surfaces characterized by the precise control ofthe inter-nano-pillar distance would be very useful, and theirimplementation on either silicon or PDMS (to explore a widerrange of effective rigidities) substrates is a natural perspective,although technologically challenging, of the present work.

■ ASSOCIATED CONTENT*S Supporting InformationSchematic aspect of the samples produced for the study. Nth-neighbor distances. Comparison of growth cones on nano-pillared and flat surfaces. Elastic analogy between thecontinuous elastic medium and pillars. This material is availablefree of charge via the Internet at http://pubs.acs.org.

■ AUTHOR INFORMATIONCorresponding Author*E-mail: catherine.villard@neel.cnrs.fr.

NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSWe thank A. Schweitzer and N. Collomb for excellent technicalsupport with neuronal cultures, the NanoFab team for theirsupport and advice concerning silicon etching, and C. Tombafor vinculin staining. This work has been funded by theFondation Nanoscience, Grenoble, France.

■ REFERENCES(1) Ghashghaei, H. T.; Lai, C.; Anton, E. S. Neuronal migration inthe adult brain: are we there yet? Nat. Rev. Neurosci. 2007, 8, 141−151.

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494448

(2) Robles, E.; Gomez, T. M. Focal adhesion kinase signaling at sitesof integrin- mediated adhesion controls axon pathfinding. NatureNeurosci. 2006, 9, 1274−1283.(3) Geiger, B.; Spatz, J. P.; Bershadsky, A. D. Environmental sensingthrough focal adhesions. Nat. Rev. Mol. Cell Biol. 2009, 10, 21−33.(4) Dowell-Mesfin, N. M.; Abdul-Karim, M.-A.; Turner, A. M. P.;Schanz, S.; Craighead, H. G.; Roysam, B.; Turner, J. N.; Shain, W.Topographically modified surfaces affect orientation and growth ofhippocampal neurons. J. Neural Eng. 2004, 1, 78−90.(5) Kundu, A.; Micholt, L.; Friedrich, S.; Rand, D. R.; Bartic, C.;Braeken, D.; Levchenko, A. Superimposed topographic and chemicalcues synergistically guide neurite outgrowth. Lab Chip 2013, 13,3070−3081.(6) Micholt, L.; Gartner, A.; Prodanov, D.; Braeken, D.; Dotti, C. G.;Bartic, C. Substrate topography determines neuronal polarization andgrowth in vitro. PLoS ONE 2013, 8, e66170.(7) Fozdar, D. Y.; Lee, J. Y.; Schmidt, C. E.; Chen, S. Selective axonalgrowth of embryonic hippocampal neurons according to topographicfeatures of various sizes and shapes. Int. J. Nanomed. 2010, 2011, 45−57.(8) Spatz, J. P.; Geiger, B. Molecular engineering of cellularenvironments: cell adhesion to nano-digital surfaces. Methods CellBiol. 2007, 83, 89−111.(9) Hallstrom, W.; Prinz, C. N.; Suyatin, D.; Samuelson, L.;Montelius, L.; Kanje, M. Rectifying and sorting of regenerating axonsby free-standing nanowire patterns: a highway for nerve fibers.Langmuir 2009, 25, 4343−4346.(10) Piret, G.; Perez, M. T.; Prinz, C. N. Neurite outgrowth andsynaptophysin expression of postnatal CNS neurons on GaP nanowirearrays in long-term retinal cell culture. Biomaterials 2012, 34 (4), 875−87.(11) Prinz, C.; Hallstrom, W.; Martensson, T.; Samuelson, L.;Montelius, L.; Kanje, M. Axonal guidance on patterned free-standingnanowire surfaces. Nanotechnology 2008, 19, 345101.(12) Robinson, J. T.; Jorgolli, M.; Shalek, A. K.; Yoon, M. H.;Gertner, R. S.; Park, H. Vertical nanowire electrode arrays as a scalableplatform for intracellular interfacing to neuronal circuits. Nat.Nanotechnol. 2012, 7 (), 180−184.(13) Xie, C.; Lin, Z.; Hanson, L.; Cui, Y.; Cui, B. Intracellularrecording of action potentials by nanopillar electroporation. Nat.Nanotechnol. 2012, 7, 185−190.(14) Roth, S.; Bugnicourt, G.; Bisbal, M.; Gory-Faure, S.; Brocard, J.;Villard, C. Neuronal architectures with axo-dendritic polarity abovesilicon nanowires. Small 2012, 8, 671−675.(15) Bray, D. F.; Bagu, J.; Koefler, P. Comparison ofhexamethyldisilazane (HMDS), Peldri II, and critical-point dryingmethods for scanning electron microscopy of biological specimens.Microsc. Res. Tech. 1993, 26, 489−495.(16) Abramoff, M. D.; Magalhaes, P. J.; Ram, S. J. Image processingwith imageJ. Biophoton. Int. 2004, 11, 36−42.(17) Necas, A.; Klapetek, P. Gwyddion: an open-source software forSPM data analysis. Cent. Eur. J. Phys. 2012, 10, 181−188.(18) Zhang, T. Y.; Suen, C. Y. A fast parallel algorithm for thinningdigital patterns. Commun. ACM 1984, 27, 236−239.(19) Katz, M. J. How straight do axons grow? J. Neurosci. 1985, 5 (),589−595.(20) Carlstrom, L. P.; Hines, J. H.; Henle, S. J.; Henley, J. R.Bidirectional remodeling of b1-integrin adhesions during chemotropicregulation of nerve growth BMC Biol. 2011 9, doi: 10.1186/1741-7007-9-82.(21) Lein, P. J.; Banker, G. A.; Higgins, D. Laminin selectivelyenhances axonal growth and accelerates the development of polarityby hippocampal neurons in culture. Brain Res. Dev. 1992, 69, 191−197.(22) Esch, T.; Lemmon, V.; Banker, G. Local presentation ofsubstrate molecules directs axon specification by cultured hippocampalneurons. J. Neurosci. 1999, 19, 6417−6426.(23) Dotti, C. G.; Sullivan, C. A.; Banker, G. A. The establishment ofpolarity by hippocampal neurons in culture. J. Neuroci. 1988, 8, 1454.

(24) Leach, J. B.; Brown, X. Q.; Jacot, J. G.; DiMilla, P. A.; Wong, J.Y. Neurite outgrowth and branching of PC12 cells on very softsubstrates sharply decreases below a threshold of substrate rigidity. J.Neural Eng. 2007, 4, 26−34.(25) Flanagan, L. A.; Ju, Y. E.; Marg, B.; Osterfield, M.; Janmey, P. A.Neurite branching on deformable substrates. Neuroreport 2002, 13,2411−2415.(26) Georges, P. C.; Miller, W. J.; Meaney, D. F.; Sawyer, E. S.;Janmey, P. A. Matrices with compliance comparable to that of braintissue select neuronal over glial growth in mixed cortical cultures.Biophys. J. 2006, 90, 3012−3018.(27) Jang, K. J.; Kim, M. S.; Feltrin, D.; Jeon, N. L.; Suh, K. Y.; Pertz,O. Two distinct filopodia populations at the growth cone allow tosense nanotopographical extracellular matrix cues to guide neuriteoutgrowth. PLoS One 2010, 5, 15966.(28) Dehmelt, L.; Halpain, S. Actin and microtubules in neuriteinitiation: are MAPs the missing link? J. Neurobiol. 2004, 58, 18−33.(29) Lamoureux, P.; Ruthel, G.; Buxbaum, R. E.; Heidemann, S. R.Mechanical tension can specify axonal fate in hippocampal neurons. J.Cell Biol. 2002, 159, 499−508.(30) Xu, K.; Zhong, G.; Zhuang, X. Actin, spectrin, and associatedproteins form a periodic cytoskeletal structure in axons. Science 2013,25, 452−456.(31) Li, N.; Folch, A. Integration of topographical and biochemicalcues by axons during growth on microfabricated 3-d substrates. Exp.Cell Res. 2005, 311 (), 307−316.(32) Craig, A. M.; Jareb, M.; Banker, G. Neuronal polarity. Curr.Opin. Neurobiol. 1992, 2, 602−606.(33) Yamamoto, H.; Demura, T.; Morita, M.; Banker, G. A.; Tanii,T.; Nakamura, S. Differential neurite outgrowth is required for axonspecification by cultured hippocampal neurons. J. Neurochem. 2013,123, 904−910.

Langmuir Article

dx.doi.org/10.1021/la5001683 | Langmuir 2014, 30, 4441−44494449