Size-Tunable Nanoneedle Arrays for Influencing Stem Cell Morphology, Gene Expression, and Nuclear Membrane Curvature.

High-aspect-ratio nanostructures have emerged as versatile platforms for intracellular sensing and biomolecule delivery. Here, we present a microfabrication approach in which a combination of reactive ion etching protocols were used to produce high-aspect-ratio, nondegradable silicon nanoneedle arrays with tip diameters that could be finely tuned between 20 and 700 nm. We used these arrays to guide the long-term culture of human mesenchymal stem cells (hMSCs). Notably, we used changes in the nanoneedle tip diameter to control the morphology, nuclear size, and F-actin alignment of interfaced hMSCs and to regulate the expression of nuclear lamina genes, Yes-associated protein (YAP) target genes, and focal adhesion genes. These topography-driven changes were attributed to signaling by Rho-family GTPase pathways, differences in the effective stiffness of the nanoneedle arrays, and the degree of nuclear membrane impingement, with the latter clearly visualized using focused ion beam scanning electron microscopy (FIB-SEM). Our approach to design high-aspect-ratio nanostructures will be broadly applicable to design biomaterials and biomedical devices used for long-term cell stimulation and monitoring.

Many aspects of the microenvironment directly influence cell function and differentiation. Bioengineered systems often employ biomimetic physical cues arising from intrinsic properties of the substrate (e.g., chemical identity, mechanical properties, topography)[1−4] or externally applied forces (e.g., shear, compression, tension, electric, acoustic, magnetic).[5−8] In particular, micro- and nanoscale topographies are potent regulators of stem cell adhesion, morphology, migration, proliferation, and differentiation.[9−11] Nanoscale structures (e.g., dots, pits, grooves, roughened surfaces) can be created using a variety of fabrication strategies, including electron-beam lithography,[12] photolithography,[13] interference lithography,[14] nanoimprinting,[15] particle self-assembly,[16] and electrospinning.[17,18] Of these structures, vertically aligned nanowires, nanotubes, and nanostraws have shown great utility for recording intracellular electrical signals,[19] studying mechanosensing pathways,[20] and facilitating biomolecule delivery.[21−23] We have previously reported the fabrication of high-aspect-ratio porous silicon nanoneedle arrays for intracellular biochemical sensing[24,25] and in vivo transfection of plasmid DNA encoding for vascular endothelial growth factor.[26] Key design features of this platform include the material mesoporosity and the sharp tips of the nanoneedles (50 nm in diameter), which have been shown to enable cargo loading and promote endocytosis.[27] These mesoporous silicon nanoneedles were biodegradable in aqueous environments within 48 h. This collection of properties was ideal for delivery applications requiring a temporary cell–material interface; however, for long-term cell culture, nanoneedle arrays must remain stable for several days to weeks. Here, we describe the fabrication of nondegradable silicon nanoneedles that can provide a continuous topographical interface to human mesenchymal stem cells (hMSCs) for at least 5 weeks in culture. We used a combination of different reactive ion etching protocols to create solid silicon nanoneedles with tips that could be tuned from 20 to 700 nm in diameter. The diameter of the nanoneedle tips impacted the morphology, polarization, gene expression, Yes-associated protein (YAP) localization, and nuclear deformation of cultured hMSCs. These results are applicable to the design of biomedical devices, bioelectrodes, and platforms that seek to control the cell behavior using topographical cues and should provide insight into basic biology and cell–nanomaterial interactions.

Results and Discussion

Fabrication and Characterization of Nanoneedle Arrays with Different Tip Diameters

We fabricated arrays of nanoneedles with different tip diameters from silicon wafers using a top-down fabrication approach (Figure a). On a nitride-coated wafer, we first patterned a two-dimensional dot array using negative photoresist, then used reactive ion etching (RIE) to transfer this pattern into a hard silicon nitride etch mask.[26] We used deep reactive ion etching (DRIE),[28,29] with alternating etch and passivation steps, to anisotropically etch vertical silicon pillars. We then sharpened the pillars into nanoneedles using RIE, which isotropically etched the silicon nitride cap and the top of the pillar. The tip diameter (Dtip), defined as the diameter observed from the top view of the resulting structure, correlated linearly with the RIE process time, allowing us to finely tune the Dtip from 20 to 700 nm (Figure b). The other geometric properties were consistent across the surface of the 100 mm diameter silicon wafers, with 5 μm high nanoneedles arranged in uniform square arrays with 2 μm spacing. We tested the long-term stability of our nanoneedle arrays against degradation in cell culture medium (minimum essential medium alpha-modification) supplemented with 10% v/v fetal bovine serum (FBS) under standard cell culture conditions (37 °C, 5% CO2). Analysis by scanning electron microscopy (SEM) revealed no visible degradation over 4 weeks (Figure S1a), in contrast to the rapid degradation observed for porous silicon nanoneedles.[26,30] This aqueous stability enabled us to use the nanoneedle arrays as a continuous topographical interface to influence the long-term culture of hMSCs (Figure S1b). hMSCs readily adhered to the sharp nanoneedles (Dtip = 47 ± 7 nm), nanopillars (Dtip = 718 ± 32 nm), and flat controls without the need for any additional substrate coating. We assessed the proliferation of hMSCs cultured on the sharp nanoneedles, nanopillars, and flat controls by measuring the number of Ki-67 positive nuclei after 72 h and the relative gene expression of MKI67 after 6 and 24 h (Figure S2). We observed a slight reduction in Ki-67 positive nuclei on the nanoneedles and nanopillars compared to the flat controls and significant reduction in gene-level expression of MKI67 between the blunt and sharp nanostructures after both time points. The survival with reduced proliferation of hMSCs cultured on nanoneedles was also evidenced by LIVE/DEAD staining performed after 35 d, which showed that all substrates supported long-term hMSC viability (Figure S2c). SEM analysis revealed that the hMSCs on the nanopillar array had large, flattened cell bodies and relatively few protruding filopodia (Figure c), whereas cells on the sharp nanoneedles were highly polarized with extended filopodia (Figure d). In the latter case, we observed that nanoneedles in contact with hMSCs were clearly deformed. Previous studies have shown that silicon-based nanostructures can be thinned in order to reduce the effective material stiffness and increase the mechanical flexibility.[31,32] To understand the change in effective substrate stiffness as a function of Dtip, we modeled the flexural stiffness of the nanoneedles using Euler–Bernoulli beam theory.[33,34] This model was used to calculate the deflection profile when an external force of 300 nN (the maximum traction force reported for a single cell[35]) was applied orthogonally to the nanoneedles (Figure e). This analysis showed an increased deflection with increased tip sharpness and a calculated value of effective stiffness that was linearly dependent upon tip diameter (Figure f). This model suggested that the nanopillars had an effective stiffness approximately 15 times greater than the sharp nanoneedles. By using this geometric relationship to tune the effective stiffness of the nanoneedle arrays, it may be possible to control the morphological and phenotypic changes of interfaced cells through topography-driven mechanotransduction.

Figure 1

(a, b) Fabrication of vertically aligned nonporous nanoneedle arrays with systematic control over tip sharpness: (a) Schematic of the photolithography and dry silicon etching processes: (i) silicon nitride deposited as a hard mask onto a silicon wafer, (ii–iv) photolithography to produce dot arrays of photoresist on the nitride layer, (v) reactive ion etching (RIE) to pattern the silicon nitride, (vi) silicon etching via deep reactive ion etching (DRIE) to produce nanopillar arrays, (vii) sharpening of the nanopillars using RIE to produce nanoneedle arrays. Corresponding SEM images of the resulting structures from each step (scale bars: 2 μm). (b) Systematic control over the tip diameter and aspect ratio (height of structure divided by tip diameter) of structures by adjusting RIE process time, showing how tip sharpness can be controlled by varying the RIE process time (data shown as mean ± SD, R2 = 0.9886 for the linear extrapolation of tip diameter control, N = 3 for all image analysis). (c, d) hMSCs after 72 h culture on nanopillars and sharp nanoneedles, respectively. Scale bars: 5 μm. (e, f) Theoretical stiffness for a model of a point-loaded conical beam: (e) Deflection of nanopillars (Dtip = 718 ± 32 nm), blunt nanoneedles (Dtip = 316 ± 20 nm), and sharp nanoneedles (Dtip = 47 ± 7 nm) when 300 nN of traction force (F) was imposed at the apex of each structure. Note that only the upper 1.5 μm of tip deflection is shown, as this is the region in which the deflection profiles differ the most. (f) Stiffness calculated for the different tip diameters. For the calculation, the following variables were held constant: height of structure = 5.5 μm; base diameter = 0.8 μm; elastic modulus of silicon = 129.5 GPa. The height and base diameter of the nanostructures were measured from images taken by scanning electron microscopy, while the elastic modulus of silicon was provided by the manufacturer.

Tip Diameter of Nanoneedle Arrays Affects Cell Morphology

To further  nvestigate the changes in cell morphology, we cultured hMSCs on flat silicon (as a control), nanopillars (Dtip = 718 ± 32 nm), two different blunt nanoneedles (Dtip = 316 ± 20 nm and 172 ± 6 nm), and one set of sharp nanoneedles (Dtip = 47 ± 7 nm). We fixed the cells for immunostaining at four time points (6, 12, 24, and 72 h after seeding). We then used image-based cell profiling to analyze 5372 immunofluorescent microscopy images and extract single-cell morphological and protein localization features for over 100 000 cells (Figure a). From this high-content image analysis, we were able to quantify pronounced and systematic morphological changes as a function of nanoneedle sharpness (Figure b). In particular, decreasing the Dtip reduced the spread area of both cells and nuclei, promoted cell body elongation, and decreased the cell protrusion ratio, an indicator of the number of protrusions extending from the cell (exemplar raw data for a selection of measurements shown in Figure S3). In addition, nuclear solidity (an inverse measure of nuclear perimeter tortuosity) was visualized as a slight scalloping in the in-plane nuclear membrane around the nanoneedles, and decreased with increasing nanoneedle sharpness (Figure S4). Background-corrected and batch-normalized intensities of cytoskeletal proteins (F-actin, α-tubulin) were also influenced by changing Dtip. Local cell density, determined by Voronoï tessellation, was also observed to be greater on nanoneedles than flat surfaces or nanopillars.

Figure 2

High-content screening of hMSCs cultured on flat controls (i), nanopillars (ii = 718 ± 32 nm), blunt nanoneedles (iii = 316 ± 20 nm, iv = 172 ± 6 nm), and sharp nanoneedles (v = 47 ± 7 nm): (a) Illustration of the image and data analysis pipeline, showing examples of imaging, cell segmentation, single-cell profiles, data transformation, collinear feature reduction, and model confusion matrix. (b) Normalized heatmap showing the change in population median on different substrates for a selected range of parameters. Note: Data shown here are transformed and normalized to flat substrates; hence (i) is blank (white). (c) Composite potency index of model features, where larger values indicate the relative influence of the feature on the overall model. Minimum and maximum values are indicated as error bars. (d) Scatter plot of hMSCs cultured on different substrates onto the first two linear discriminants (LD1 and LD2), to visualize the class separation. 95% confidence ellipses were added to each scatter.

Linear Discriminant Analysis Uncovers Complex Morphological Phenotypes That Are Impacted by Tip Diameter of Nanoneedle Arrays

While image analysis allowed us to intuitively explore features of interest, we also used linear discriminant analysis (LDA) to uncover less obvious features that were influenced by the nanoneedle tip diameter. While all measured features could be used to train this model, it was important to avoid multicollinearity between samples (interdependence between related measurements), since this negatively impacts the stability and robustness of LDA models.[36] Accordingly, we preselected 31 features of interest (Table S1) and then used an automated stepwise procedure to remove strongly correlated features (ρ > 0.75, see Supporting Information for details).[36] This filtering produced an LDA model with 18 features, which could each be described in terms of the composite potency index, a relative measure that enables interpretation of a model comprising four discriminant functions (Figure c).[37] This model revealed that α-tubulin intensity had the largest impact in separating cells from different substrates. Cell mean radius and cytoplasmic ratio of F-actin and α-tubulin were also observed to strongly influence the cell classification. Separation of the three different target classes (flat substrates, nanopillars, and sharp nanoneedles) was clearly visible when clustered using the two primary discriminant functions (LD1 and LD2) (Figure d).

Sharper Nanoneedles Guide Cell Alignment and Polarization by Activated Actomyosin Contractility

High-content image analysis also revealed a strong dependence on cell orientation as a function of nanoneedle tip sharpness (Figure S5). Alignment was clearly evident from fluorescence microscopy images, which showed hMSC bidirectionally polarized on the nanoneedle arrays (Figure a, c, and e). To investigate this effect further, we used an orientation analysis algorithm, OrientationJ,[38,39] to quantify the alignment of F-actin fibers in hMSCs on different substrates. The algorithm was used to calculate a tensor map of angular orientation for each pixel in the image and weight each tensor by the pixel intensity. This analysis allowed the proportion of image regions corresponding to a given orientation to be plotted as a function of angle. The measured angular distribution showed that F-actin fibers were randomly oriented on the flat substrate, but preferentially aligned along both major axes of the nanoneedle arrays (designated here as 0° and ±90°, Figure b, d, and f). The degree of alignment greatly increased from 6 to 24 h for the nanoneedles but not for the flat controls or the nanopillars. In all cases the degree of alignment plateaued between 24 and 72 h (Figure S6).

Figure 3

(a–f) Cytoskeletal alignment of hMSCs on nanostructured surfaces: (a, c, e) Representative immunofluorescence images of hMSCs after 6 and 72 h of culture on flat controls, nanopillars (Dtip = 718 ± 32 nm), and sharp nanoneedles (Dtip = 46 ± 7 nm) (scale bar: 100 μm). (b, d, f) Orientation analysis of F-actin fibers shown as radial plots of weighted distributions for the four groups at 6 and 72 h. (g) Schematic of RhoA and Rac1 signaling pathways. (h, i) Representative immunostained images of hMSCs cultured on nanoneedles after gene silencing: (h) RhoA-silenced and (i) Rac1-silenced hMSCs cultured on nanoneedles for 24, 72, and 120 h. Rho-A silenced hMSCs exhibited a less elongated morphology compared to Rac1-silenced cells, which exhibited elongation and alignment to the underlying nanoneedles. Scale bars: 25 μm.

Cell spreading and morphology are known to be regulated by Rho-family GTPase signaling pathways.[40,41] The small GTPase, RhoA, and its effector, Rho-associated, coiled-coil-containing protein kinase (ROCK), drive cell polarization in part by activating myosin light chain II (MLC2), which promotes actin filament assembly and cell contraction.[41] Another Rho-family protein, Ras-related C3 botulinum toxin substrate 1 (Rac1), regulates cell polarization and motility by promoting the formation of actin-rich protrusions (Figure g).[42,43] Thus, we sought to investigate how the polarization and elongation of cells on the sharp nanoneedle arrays might depend upon RhoA and Rac1 signaling. To test the effects of inhibiting Rho-mediated and Rac-mediated signaling, we performed a gene silencing study. Small interfering RNA (siRNA) targeting RhoA and Rac1 were transfected into hMSCs using Lipofectamine, and knockdown was validated by qRT-PCR (Figure S7). The RhoA-silenced hMSCs cultured on nanoneedles generally adopted a circular morphology with little elongation or alignment to the underlying substrate (Figure h). In contrast, Rac1-silenced hMSCs cultured on nanoneedles exhibited a well aligned, elongated morphology, similar to that observed with nontransfected hMSCs (Figure i). Similar behavior was observed when we used chemical inhibitors of MLC2 (blebbistatin) and Rac1 (NSC23766). Culturing hMSCs with 10 μM blebbistatin abolished the bidirectional alignment of hMSCs on sharp nanoneedles, whereas alignment was maintained in the presence of 10 μM NSC23766 (Figure S8b). This suggested that RhoA-mediated contractility had a more critical role than Rac1 signaling in the polarization of hMSCs on sharp nanoneedle arrays.

Nanoneedle Tip Diameter Affects the Expression of Nucleoskeleton, Cytoskeleton, Focal Adhesion, and YAP Target Genes

We next looked at the impact of nanoneedle tip diameter on the expression of relevant genes. hMSCs were cultured for 6 or 24 h on flat silicon substrates or nanoneedles with varying tip diameters. mRNA was isolated to measure the expression of a wide portfolio of genes, including those encoding cytoskeletal and focal adhesions proteins (ACTB, TUBA1A, DSTN, MACF, MAP4, PTK2, PXN), integrin subunits (ITGAV, ITGA2, ITGA5, ITGB1, ITGB3), YAP targets (ANKRD1, CTGF), nuclear lamins (LMNA, LMNB), nuclear envelopes (DNM2, SYNE2, SYNE3, SUN2), Rho-GTPases (RHOA, ROCK, RAC1, CDC42), proliferation markers (MK167, PCNA), and caveolin (CAV1) (Table S6). Interestingly, LMNA expression tended to be influenced by the presence of a nanostructured substrate and as a function of increasing nanoneedle tip diameter (Figure a). LMNA codes for lamin A, a major structural component of the nuclear lamina, and our observation is consistent with previous studies showing a strong correlation between nuclear deformation and lamin expression.[44−46] This finding is also consistent with reported increases in LMNA expression in cells on porous nanoneedles.[30] LMNB, the gene encoding lamin B, exhibited a slight downward trend with increasing sharpness, which was the opposite of the trend for LMNA (Figure b). Expression of PXN, a gene that codes for paxillin that is expressed at focal adhesions during cell attachment, was significantly reduced after 6 h of culture on sharp nanoneedles compared to the flat substrates (Figure c), and we observed similar trends for integrins and other cytoskeleton-related genes shown in Table S6. These changes in gene expression were consistent with previous studies performed on porous nanoneedles,[30] but our nondegradable arrays enabled us to investigate gene expression beyond 6 h. Analysis of later time points revealed that PXN returned to baseline expression levels after 24 h, although immunostaining for paxillin showed a reduced overall intensity and reduced focal adhesion puncta in the hMSCs cultured on sharp nanoneedles at 24 h (Figure c–f). Sharp nanoneedles were also observed to regulate integrin expression, as evidenced by downregulation of genes coding for integrin subunits (ITGAV, ITGA5, ITGB1) after 24 h of culture, and reduced immunofluorescent intensity of integrin β1 after 72 h of culture (Figure S9). This indicated that the nanoneedle geometry had profound effects on both the adhesion assembly and adhesion-related gene expression.

Figure 4

Relative expression of nuclear lamina genes: (a, b) Relative gene expression of LMNA and LMNB after 6 and 24 h. (c–f) Relative expression and distribution of focal adhesion proteins: (c) Relative expression of PXN after 6 and 24 h. Where Welch’s ANOVA detected a significant effect on gene expression, a Games–Howell post hoc test was conducted. Data have been normalized with the cells cultured for 6 h on the substrate. Asterisks (*) indicate significantly different (p < 0.05) groups (box plots, minimum/maximum, N = 3). (d) Representative confocal immunofluorescence images of paxillin-stained focal adhesions in hMSCs on different structures after 24 h of culture (scale bars: 25 μm) and (e) quantified fluorescence intensity of focal adhesions (box plots, minimum/maximum. N = 3). (f) Focal adhesion length and number per unit area: (i) flat substrate, (ii) nanopillars, (iii, iv) blunt nanoneedles, (v) sharp nanoneedles (mean ± SD, N = 3). *: p < 0.05 from one-way ANOVA test. p-Values were obtained from Tukey’s honestly significant difference (HSD) post hoc test. (g–i) Visualization of nuclear membrane–structure interfaces using FIB-SEM: (g) FIB-SEM images show the extent of plasma membrane and nuclear envelope deformation after 6 and 72 h of culture on nanopillars (Dtip = 718 ± 32 nm) and sharp nanoneedles (Dtip = 47 ± 7 nm), respectively. Scale bars: 2 μm. (h, i) 3D reconstructions of the plasma membrane and nuclear envelope overlaid on the SEM background image, nanopillar/nanoneedles (green), cell membrane outside (gray), inside (red), and nucleus (blue).

We next examined how nanoneedles impact the activity of YAP, a transcriptional cofactor that plays important roles in cancer, tissue regeneration, and organ size control.[47] YAP can translocate to the nucleus in response to mechanical tension, such as extracellular rigidity, applied stress, or cell spreading. We immunostained for YAP in hMSCs cultured on different substrates after 6 and 24 h, and calculated the nuclear-to-cytoplasmic ratio (N:C ratio). YAP N:C ratios were observed to be lower on nanostructures than on the flat silicon, and we subsequently found that the expression of two YAP-target genes, ANKRD1 and CTGF, were significantly reduced after 24 h on sharp nanoneedles compared to the flat substrates (Figure S10c and d). The inverse relationship between LMNA and YAP target gene expression may point to a feedback mechanism in which nuclear deformation induced by nanotopography is transduced into changes in gene expression.

Long-Term Cell Differentiation Studies Using Nanoneedles

Having demonstrated that the nondegradable nanoneedle arrays were capable of long-term cell culture, we next investigated whether they could be used to support hMSC differentiation (Figure S11). We cultured hMSCs for 3 weeks on flat substrates, nanopillars, and nanoneedles, using either basal media or media supplemented with adipogenic or osteogenic factors. Oil Red O staining for lipid vacuoles and Alizarin Red S staining for calcium deposits were used to assess adipogenesis and osteogenesis, respectively. Negligible staining was observed for all substrates using basal media, suggesting an absence of material-driven differentiation down these lineages. All substrates supported adipogenic differentiation; however, the nanoneedle substrate appeared to impair osteogenic differentiation. This is not surprising since osteogenesis requires generation of intracellular forces[9,12] and hMSCs on nanoneedles were observed to have limited paxillin staining, which indicates a limited ability to establish focal adhesions and generate tension. Further studies are needed to confirm whether this difference is driven by morphological and cytoskeletal network changes and whether nanoneedles can be used to modulate other differentiation processes and other stem cells.

Nanoneedle Impingement of Cell and Nuclear Membranes Visualized via Focused Ion Beam Scanning Electron Microscopy

Finally, we used focused ion beam scanning electron microscopy (FIB-SEM) to image cross sections through the cell–nanoneedle interface in order to visualize how biological membranes were perturbed by the underlying nanotopography.[48] For both the sharp nanoneedles (Dtip = 47 ± 7 nm) and nanopillars (Dtip = 718 ± 32 nm), the plasma membrane was deeply impinged by the vertical arrays, wrapping conformably around the silicon nanostructures after just 6 h (Figure g). The sharp nanoneedles induced a far greater degree of impingement of the plasma membrane and perturbation of the nuclear membrane than the nanopillars. These extreme deformations, which would put mechanical stress on the nucleus, could offer an explanation for the upregulation of LMNA shown in Figure a, which is consistent with previous findings in hMSCs on porous nanoneedles.[30] Moreover, the degree of cellular and nuclear membrane deformation increased with culture time, an important insight for the design of nondegradable nanoneedle arrays for long-term culture, intracellular delivery, and sensing. We further investigated these ultrastructural changes for the 12 h time point by reconstructing consecutive FIB-SEM slices into a volumetric map, which allowed us to fully visualize the cell–nanoneedle interface in 3D (Figure h,i and Movie M1). This reconstruction analysis validated that nanoneedle impingement was present across the entire cell and nuclear area.

Conclusion

In summary, we report the design and application of cytocompatible, solid silicon nanoneedle arrays with precisely tunable tip diameters. These nanoneedle arrays were stable against degradation in aqueous media, which enabled us to explore the long-term effects of nanoneedle geometry upon interfaced hMSCs. By varying the nanoneedle tip diameter, we were able to directly influence cell morphology, polarization, and gene expression in hMSCs. While previous studies have shown that cell protrusions can be controlled using nanostructures with different densities,[32,49,50] this study demonstrates that nanoneedle tip diameter can also be used to regulate bidirectional polarization, morphological heterogeneity, nuclear morphology, and gene expression. It should be noted, however, that we did not observe any significant changes in pluripotency or differentiation markers (e.g., SOX2, NANOG, AP2, OCN, RUNX2) within 24 h (data not shown), and we have not yet determined whether the observed morphological and gene-level changes could impact or direct long-term stem cell differentiation.

Nevertheless, this investigation illustrates how nanoneedle sharpness can be used to precisely tune the mechanical microenvironment experienced by interfaced cells, regulate plasma membrane impingement and nuclear deformation, and instigate changes in cell phenotype. This system offers a long-term cell culture interface for applications in bioelectronic sensing and stimulation.

Experimental Method

Fabrication of Vertically Aligned Nonporous Nanoneedle Arrays

Nanoneedle arrays were fabricated on a 4-in.-diameter p-type doped Si wafer with 0.01 Ω cm resistivity (University Wafers, USA). A hard mask layer of low-stress silicon nitride was deposited onto the wafer to a thickness of 1200 Å using low-pressure chemical vapor deposition (Scottish Microelectronic Centre, The University of Edinburgh, UK). Dot arrays of 0.6 μm diameter and 2 μm pitch spacing were transferred to the hard mask via photolithography using an NR9-250P photoresist, RD6 developer (Futurrex, USA), and a MA6 mask aligner (Suss Microtech, Germany). RIE was performed on the wafer using an Oxford NGP80 (Oxford Instrument, UK) using 50 sccm of CF3 gas and 5 sccm of O2 gas with a process pressure of 55 mTorr and power of 140 W for 150 s. The patterned wafer was mounted on a 6-in.-diameter carrier wafer using a Crystalbond 555 adhesive stick for DRIE using a deep reactive ion etcher (Surface Technology Systems, UK). Each DRIE cycle consisted of (i) 130 sccm of SF6 gas and 6 sccm of O2 gas with a process pressure of 15 mTorr and power of 800 W for an 8 s etch phase and (ii) 85 sccm of C4F8 gas with a process pressure of 14 mTorr and power of 600 W for a deposition phase of 6.5 s. To produce structures of 5–6 μm height, between 30 and 35 cycles were conducted. The processed wafer was then released from the carrier wafer and diced into 8 × 8 mm squares for further use (DISCO Technologies, Japan) or sharpened into conical nanoneedle structures using a further RIE step. The parameters for this RIE step comprised 10 sccm of SF6 gas with a process pressure of 100 mTorr and power of 300 W. The nanoneedle tip sharpness could be controlled by adjusting the RIE run time between 0 and 9 min, as indicated in Figure b.

Cell Culture

Human mesenchymal stem cells (Lonza Ltd., Basel, Switzerland) were expanded in MesenPRO RS medium (Gibco, ThermoFisher Scientific, UK) and passaged using 0.05% v/v trypsin–EDTA at approximately 80% confluence. Prior to cell seeding, all nanoneedle and flat substrates were sterilized using two 10 min washes with 70% v/v ethanol (Sigma-Aldrich), rinsed with sterile phosphate-buffered saline (PBS), and then further sterilized under ultraviolet light for at least 10 min. hMSCs were seeded between passage 3 and 6 at a density of 10 000–15 000 viable cells/cm2 onto substrates using minimum essential medium alpha modification (Gibco, ThermoFisher Scientific, UK) with 10% v/v MSC-qualified fetal bovine serum (Gibco, ThermoFisher Scientific, UK) and 1% v/v penicillin/streptomycin (Gibco, ThermoFisher Scientific, UK).

LIVE/DEAD Assay

hMSCs were seeded at a density of 20 000 viable cells per substrate and cultured for 35 d. LIVE/DEAD staining was then performed by immersing each substrate into calcein-AM/ethidium homodimer-1 (Invitrogen) solution (each at 1 × 10–6 M in PBS) for 20 min, followed by gentle washing with PBS. Fluorescence microscopy (Invitrogen EVOS FL auto imaging system, Thermo Fisher Scientific) was used to capture images of viable and nonviable cells.

Differentiation of hMSCs

hMSCs were seeded at a density of 40 000 viable cells per substrate and cultured for 1 d in basal media, before switching to adipogenic or osteogenic media prepared with StemXVivo adipogenic supplement (100×, from R&D Systems) and StemXVivo Osteogenic Supplement (10×, from R&D Systems). The media was changed twice weekly. After 21 d of differentiation (three replicates), each of the substrates was fixed with 4% w/v methanol-free formaldehyde (Pierce 16% formaldehyde (w/v), methanol-free, ThermoFisher Scientific, UK) for 30 min at room temperature, and then Oil Red O or Alizarin Red S staining followed. For Oil Red O staining, 5 mg/mL of Oil Red O stock solution was prepared in 100 % isopropanol as solvent, and then diluted once more to make Oil Red O solution in deionized water at a 3:2 (stock to water ratio). Each sample was stained with 500 μL of Oil Red O solution at room temperature for 30 min. After the staining, samples were imaged with fluorescence microscopy. For quantification, the stain was extracted in 500 μL of isopropanol and transferred into a 96-well plate (100 μL per well) to measure the absorbance at 492 nm using a plate reader, with 100% isopropanol used for the background subtraction. For Alizarin Red S staining, each sample was stained with 2% w/v Alizarin Red S (Sigma) for 30 min. Samples were viewed using a digital camera, and then dye extraction was followed for quantification. The stain was extracted in 200 μL of 10% v/v acetic acid followed by adding 75 μL of 10% v/v ammonium hydroxide to neutralize the acid. The solution was aliquoted and transferred into a 96-well plate (100 μL per well) to measure the absorbance at 405 nm using a plate reader.

Immunocytochemistry and Imaging

The hMSCs cultured on the different substrates were fixed for 15 min at room temperature using 4% w/v methanol-free formaldehyde (Pierce 16% formaldehyde (w/v), ThermoFisher Scientific, UK) and then washed twice with PBS. The fixed hMSCs were permeabilized for 10 min with 0.25% v/v Triton X-100 (Sigma-Aldrich), blocked for 1 h with 5% v/v donkey serum (heat-inactivated, Gibco, ThermoFisher Scientific, UK), then incubated overnight at 4 °C with primary antibodies diluted in 0.1% w/v bovine serum albumin (BSA, Sigma-Aldrich) in PBS. The blocked hMSCs were washed three times in PBS for 5 min, then incubated for 1 h at room temperature with secondary antibody (1:500 in 0.1% w/v BSA/PBS). The hMSCs were washed three more times with PBS for 5 min, and where applicable, samples were incubated with AlexaFluor-conjugated Phalloidin (1:100 in 0.1% w/v BSA/PBS) for 1 h. All samples were counterstained with DAPI (1:1000 in PBS) for 5 min. Full information on antibodies is listed in Table S7. Samples were mounted on glass-bottomed chamber slides using Fluoromount-GTM (Invitrogen, ThermoFisher Scientific, UK) and imaged using a Zeiss Axio Observer wide-field microscope (Zeiss, Germany).

Transfection of siRNA

Transfection was conducted using cationic liposome reagent, Lipofectamine RNAiMAX (Invitrogen, ThermoFisher Scientific, UK). RhoA and Rac siRNA were purchased from Santa Cruz Biotechnology, USA. For transfection, hMSCs were expanded in MesenPRO RS medium (Gibco, ThermoFisher Scientific, UK) in a six-well plate. When the cells were around 70% confluent, transfection was performed by adding RhoA siRNA (75 pmol)–Lipofectamine RNAiMAX (9 μL) or Rac siRNA (75 pmol)–Lipofectamine RNAiMAX (9 μL) complex formed in Opti-MEM medium (Gibco, ThermoFisher Scientific, UK) directly to the media. Lipofectamine RNAiMAX without siRNA was used as a control. Cells were incubated for 3 d, then the transfection was confirmed by checking gene downregulation via qRT-PCR.

Inhibitor Treatment

Y27632, blebbistatin, and NSC23766 (Sigma-Aldrich) were each diluted in dimethyl sulfoxide (DMSO, Sigma-Aldrich), and each mixture was added at a 1:1000 dilution to the hMSC suspension immediately prior to seeding on the substrates. A final concentration of 5–20 μM was used, and as a control, DMSO was added at the same ratio without any inhibitors. Cells were cultured for 24 h, prior to analysis.

Quantifying the Orientation of hMSCs

Population orientation of hMSCs under different conditions (substrates, inhibitors) was analyzed using ImageJ and the OrientationJ plug-in.[38,39] Images of actin-stained hMSCs were converted to 8-bit and the minimum brightness was raised to remove any background signal intensity. OrientationJ was run with a factor of 5 (corresponding to 6.55 μm when after pixel conversion) to produce a weighted angle distribution between −90° and 90°.

Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR)

hMSCs cultured on nanoneedles and flat substrates were incubated with Trizol (Life Technologies) for 5 min, mixed with chloroform (5:1 Trizol/chloroform), and separated by centrifugation (12000 g, 15 min, 4 °C). RNA contained within the clear aqueous phase was isolated and collected using a Direct-zol RNA MiniPrep kit (ZYMO Research, USA), according to the manufacturer’s instructions. The RNA was used to synthesize cDNA using high-capacity cDNA reverse transcription kits (Applied Biosystems, Life Technologies), assuming a 1:1 conversion from RNA to cDNA. qRT-PCR was performed with a PowerUP SYBR Green Master Mix (Applied Biosystems, Life Technologies) and QuantStudio6/StepOnePlus (Applied Biosystems) with 0.5 ng of cDNA and 500 nM of the forward and reverse primers per reaction. After each run, a melt curve was performed to ensure that a single amplicon was generated for each target gene. Cycles-to-threshold (Ct) values were used to generate fold change expression values. The expression of each gene of interest was normalized to the expression of housekeeping genes, PPIA, RPL13a, or HPRT, to generate ΔCt values. Expression of 2–ΔΔCt relative to the flat control was reported for at least three experimental replicates. Custom-designed primers were purchased from Invitrogen and tested for specificity prior to use, with sequences listed in Table S8.

Scanning Electron Microscopy and Focused Ion Beam Scanning Electron Microscopy

hMSCs on nanoneedles or flat substrates were washed in PBS, fixed for 15 min with 4% w/v methanol-free formaldehyde (Pierce 16% formaldehyde (w/v), methanol-free, ThermoFisher Scientific, UK), and washed a further three times in PBS. Next, the samples were washed twice for 5 min in 0.1 M sodium cacodylate buffer (Electron Microscopy Sciences, PA, USA, 0.2 M stock diluted in Milli-Q water) and further fixed for 1 h in 2.5% v/v glutaraldehyde solution (Electron Microscopy Sciences) in 0.1 M sodium cacodylate buffer. hMSCs were washed twice for 5 min in pure deionized water, stained for 1 h with 1% v/v osmium tetroxide in 0.1 M sodium cacodylate buffer, and then washed with pure deionized water twice for 5 min. For SEM, the samples were dehydrated in a series of ethanol dilutions (20, 30, 50, 70, 80, 90% v/v ethanol in pure water) twice for 5 min, treated with 100% ethanol four times for 5 min, incubated for 5 min with hexamethyldisilazane (HMDS, 97%, Sigma-Aldrich), and then air-dried. Samples were mounted and sputtered with a 10 nm layer of chromium (Q150, Quorum) and imaged using a LEO Gemini 1525 FEGSEM (Zeiss, Germany) with an accelerating voltage of 5 keV. For FIB-SEM, the protocol varied after osmium tetroxide staining: the samples were stained further after washing twice for 5 min in pure deionized water for 1 h with 1% (w/v) tannic acid (Sigma-Aldrich) in pure deionized water, freshly made and filtered through a 0.2 μm syringe filter (Milli-Pore), washed twice with pure deionized water, stained for 2.5–3 h with 1% (v/v) uranyl acetate in deionized water, filtered through a 0.2 μm syringe filter, and then dehydrated as described above for the SEM samples. Samples were then infiltrated with serially diluted resin (1:3, 1:2, 1:1, 2:1 resin:ethanol), 3 h each and then kept overnight in the 2:1. All resins were prepared according to the manufacturer’s instructions (epoxy embedding medium kit, Sigma). The 2:1 resin was replaced by pure resin twice for 3 h each, after which the excess resin was removed by washing with ethanol for minimal resin embedding and cured for 72 h at 60 °C. Samples were mounted and sputtered with a 20 nm layer of chromium. FIB-SEM imaging was performed on an Auriga Crossbeam FEG-SEM (Zeiss, Germany) at 1.6 keV electron beam imaging current and milled using a 30 keV gallium ion beam at 4 nA for coarse milling and 1 nA for fine milling. Images were acquired using a backscattered electron detector, and for 3D reconstruction (serial milling) the interval was set to 90 nm. Images were analyzed using ImageJ, and 3D reconstructions were made using Amira 5.3.2. (FEI) following manual image alignment and segmentation.

Statistical Analysis

Statistics on biochemical data were performed with SPSS software (IBM Corporation, USA) and OriginPro (OriginLab Corporation, USA). All raw data were checked for normality (Kolmogorov–Smirnov’s test and Shapiro–Wilk’s test) and homogeneity of variance (Levene’s test). Normal and homogeneous data were analyzed with one-way ANOVAs. If significant effects were detected, Tukey’s method was used to identify significantly different groups. In normal but heterogeneous data, Welch’s ANOVA was chosen with a Games–Howell post hoc test for significantly different groups. Significance was set at p < 0.05 for all statistical tests.

For the qRT-PCR analysis, three to four experimental replicates were performed, with at least two biological replicates within each experiment. For each experimental replicate, the expression of the gene of interest (GOI) was normalized to the geometric mean of at least two housekeeping genes (HKGs), which generated the ΔCt value. For each cell type, the average was calculated for the normalized GOI expression on flat samples, and expression of all groups was then normalized to these values. This resulted in an expression value equal to 1 for flat samples but carried a nonzero standard deviation that reflected the intraexperimental heterogeneity of biological replicates. In order to propagate this error, the relative standard deviation (RSD) was calculated for all groups of interest, wherewhere N is the number of experimental replicates (N = 3) and SD_expRep represents the standard deviation of the normalized expression for each group within each experimental replicate.

For all box plots, the 25th and 75th quartiles are represented, the line is the median, and the whiskers extend to the minimum and maximum data points.