Influence of nanopillar arrays on fibroblast motility, adhesion and migration mechanisms

Surfaces decorated with high aspect ratio nanostructures are a promising tool to study cellular processes and design novel devices to control cellular behaviour, perform intracellular sensing or deliver effector molecules to cells in culture. However, little is known about the dynamics of cellular phenomenon such as adhesion, spreading and migration on such surfaces. In particular, how these are influenced by the surface properties. In this work, we investigate fibroblast behaviour on regular arrays of 1 micrometer high, polymer nanopillars with varying pillar to pillar distance (array pitch). NIH-3T3 fibroblasts spread on all arrays, and on contact with the substrate engulf nanopillars independently of the array pitch. As the cells start to spread, different behaviour is observed. On dense arrays which have the pitch equal or below 1 micrometer, cells are suspended on top of the nanopillars, making only sporadic contact with the glass support. Cells stay attached to the glass support and fully engulf nanopillars during spreading and migration on the sparse arrays which are characterized by a pitch of 2 micrometers and above. These alternate states have a profound effect on cell migration rates, which are strongly reduced on nanopillar sparse arrays. Dynamic actin puncta colocalize with nanopillars during cell spreading and migration. Strong membrane association with engulfed nanopillars might explain the reduced migration rates on sparse arrays. This work reveals several interesting phenomenon of dynamical cell behaviour on nanopillar arrays, and provides important perspectives on design and applications of high aspect ratio nanostructures.


Introduction
Cells on vertically aligned high aspect ratio nanostructures (VA-HARNs) has been an area of great research interest in recent years. 1 Examples include surface-based delivery of molecules to cells aided by the nanostructures, 2-9 intracellular electrical measurements, 10-12 capture of circulating tumour cells, 13-15 induction of stem cell phenotypes, 16,17 intracellular sensing, [18][19][20] site-specific cellular imaging 11, 21,22 or controlling the geometry of in vitro neuronal networks. 23 Despite the range of applications, there is still much unknown about how cells are influenced by HARNs. High viability and reduced spreading area of adherent cells is often reported. 24 Arrays of nanowires have been shown to inhibit fibroblast migration with longer nanowires having a stronger effect, 25 and small patches of nanopillars inhibited neuronal cell migration. 26 Reduced spreading of cells on high aspect ratio nanostructures arrays has been reported, 16,27 as well as altered expression of genes related to the cytoskeleton and cell adhesion. 28,29 It has also been demonstrated that the cell membrane is able to wrap tightly over high aspect ratio nanostructures while maintaining membrane integrity. 30 Surface bound nanostructures including nanopillars were used to show that local membrane curvature can directly act as source of a biochemical signal for endocytic proteins. 33,34 Positively curved membranes were shown to be clathrin-mediated endocytosis (CME) hot spots, as determined by preferential accumulation of CME-related proteins at these sites.
Nanopillars were also applied for controlled probing of nuclear mechanical properties. 35 In that report, the mechanism of nuclear deformation was linked to adhesive actin patches associating with the nanopillars, pulling the nucleus down, as well as a stress-fiber linked actin "tent" at the apical membrane exerting a downwards force.
Depending on the HARN array pitch, two distinct cell adhesion states have been reported: cells can be suspended on top of the nanostructures in a "bed of nails" effect, or adhere to the substrate with the HARNs protruding into, and even through the cell body with the cell plasma membrane wrapped around nanostructures. 16,24,27,36 The critical array pitch for each state has been theoretically indicated to depend on nanostructure geometry, and has been experimentally shown to depend on surface chemistry, which influences cell-substrate adhesion. 36 We have previously shown that on 1µm high nanopillar arrays, interim and dynamic states are also possible, where only smaller regions of the cell maintains substrate contact at a given time. 27 Focal adhesions on nanowire arrays have been shown to be somewhat upregulated in one report, 37 however they localized mainly on the substrate between nanowires. Carefully selected nanostructures have the potential to facilitate targeting and high throughput studies of certain molecular events in living cells.
This can for example include endocytosis, formation of adhesion points, receptor clustering, dynamic of the cytoskeleton, cell membrane mechanics, etc. Much is still unknown about what type of structures are needed to activate certain biochemical signals and how these could be used in both screening and fundamental cell biology research. This is because most reported studies are performed on static, fixed cells. Substrates used for HARN are typically opaque, limiting possibilities of detailed dynamic studies of cell-nanostructure in-teractions with optical microscopy. Throughput of employed nanofabrication processes is often a significant bottleneck as well, as cell studies require many samples with relatively large area (∼cm 2 ). To better understand the cell responses to VA-HARNs and how they might be manipulated, we investigate the dynamics of NIH-3T3 fibroblasts adhesion, spreading and migration on arrays of high aspect ratio polymer nanopillars as a function of array pitch. Used cells are NIH-3T3 fibroblast stably transfected with LifeAct-mNeonGreen, 38 and PH-PLCd1-mScarlet 39,40 which allow for visualization of F-actin and membrane respectively. LifeAct is a 17-amino-acid long peptide, that stains filamentous actin structures in eukaryotic cells and tissues. 38 We employ nanopillar arrays fabricated with electron beam lithography (EBL) directly on microscopy-grade glass. The nanopillars are made from the stable, stiff and cell compatible polymer SU-8. 27 We show that lamellipodium-induced membrane and cytoskeleton configuration around nanopillars strongly depends on pillar spacing on the substrate (array pitch). Lamellipodium-induced cell adhesion occurs immediately once the cells encounter the nanopillars during settling. Subsequent spreading and resulting cell adherence states are strongly influenced by the nanopillar pitch. During cell migration, cells form highly dynamic F-actin bundles at nanopillar sites, while focal adhesions form on the substrate between the nanopillars. Cell migration is reduced on sparse nanopillar arrays, but unchanged on denser arrays displaying the "bed of nails" effect. Our detailed dynamic studies shed new light on the range of mechanobiological interactions that may occur between cells and nanostructures.

Nanopillar arrays
Nanopillar arrays were fabricated similar to our previously described approach, 27 but with an 100keV Elionix GLS-100 EBL system, which allows to produce pillars with a more uniform tip/base diameter and with a much higher fabrication throughput (pattern writing speed >1 mm 2 /min). Figure 1 shows examples of fabricated nanopillar arrays illustrating uniform and reproducible nanopillar geometry. Nanopillar tip diameter is in the range of 90nm, and the diameter at the base is 130 nm for pillars 1µm in length used in this study. For fluorescence imaging, the nanopillars could be optionally doped with a fluorescent dye, for example oxazine-170 (far-red excitation and emission), as shown in Figure 1C. Hexagonal arrays of nanopillars with pitches of 0.75 -10µm were made in mm 2 areas on glass coverslips mounted into polystyrene cell culture dishes (35 mm dishes or 96-well plates). Nanopillar array pitches ≤ 1µm are considered dense, while pitches of ≥ 2µm are considered sparse. This setup for nanopillar arrays is ideal for high resolution microscopy of the interface between cells and nanostructures due to the possibility of imaging live cells in inverted optical microscope and with the use of low working distance, high numerical aperture objectives. Prior to investigating cell migration and attachment on the nanopillar arrays, we verify that cell membrane and actin filaments can be visualized using employed labelling strategy. Figure 2A and Figure 2B show micrographs of NIH-3T3 fibroblasts cultured on a glass substrate for 24h.
Membrane and actin configuration for NIH-3T3 fibroblasts expressing LifeAct-mNeonGreen (green) and PH-PLCd1-mScarlet (orange) on dense and sparse naopillar arrays are presented in Figure 2C-F. The figure shows single confocal plane images taken at around 50% pillar height from the glass support (500nm), as well as xz and yz projections from corresponding z-stacks (side panels). Fluorescent signal from the Oxazine-170 dye doped SU-8 is shown in blue. Membrane signal co-localizes with the pillar signal along the pillar length for sparse arrays, seen as a pink color in both xy, xz and yz images ( Figure 2F). As previously described, the signal enhancement in the membrane channel is due to the membrane wrapping around the nanopillars. 27 At some pillar locations F-actin signal is also clearly enhanced (F-actin puncta visible as bright green colour in the side panels, Figure 2E). For dense arrays, the plasma membrane is associated with pillars at the top of the pillar surface ( Figure 2D, side panels). Some membrane wrapped around pillars at cell periphery, where cells make contact with the glass support, is also observed ( Figure 2D, side panels). For these arrays, F-actin signal is only observed above the top surface of the pillar array, again indicating that most of the cell body is suspended on the top of nanopillar array ( Figure 2C, side panels).
These results are corroborated by live-cell TIRF microscopy using the fluorescence signal from the cell membrane of NIH-3T3 fibroblasts expressing PH-PLCd1-mScarlet ( Figure 3).
For the glass control ( Figure 3A), strong fluorescent signal is observed both in widefield fluorescence mode and in TIRF mode, indicating that the cell membrane is in immediate proximity to the glass substrate. The situation is different for cells on 0.75 and 1µm pitch nanopillar arrays ( Figure 3B and Figure 3C), where strong signal is observed in the EPI mode, but only small patches of cell membrane are visible in the TIRF mode. These patches located close to the cell periphery and indicate that the cell membrane is in contact with the substrate only in these areas. The rest of the cell body is suspended on top of the nanopillar array. For sparse arrays, illustrated here by TIRF data for nanopillar spacing of 2µm, a larger area of the cell membrane is in contact with the glass substrate, indicating that the cell is wrapping around the pillars to a larger extent ( Figure 3D).

Fibroblast spreading on nanopillar arrays
In addition to describing cell conformation on various nanopillar arrays, we have used live cell microscopy to study the dynamics of initial cell attachment and settling of NIH-3T3 cells ( Figure 4). Membrane and F-actin dynamics are further presented in Supplementary Movies 1 to 3. Flat, glass areas at the edge of the nanopillar arrays are shown to allow direct comparison of cell behaviour on these two surfaces. Cells were seeded onto nanopillar arrays after trypsination and imaged for t > 3h. On dense arrays, once making the initial contact with the substrate, the cells are able to engulf nanopillars and explore the glass surface (indicated cells in Figure 4A). However as soon as they start to spread, this contact is lost and the cells stay on the top of the arrays and only explore the glass surface at the cell periphery ( Figure 4A and MovieS01). Moreover cells settling on the glass surface lose contact with the glass substrate when they migrate onto the dense pillar array, attaining a floating confirmation with lamellipodium at the leading edge reaching down to the glass surface (MovieS01). On sparse arrays nanopillars are also engulfed early in the settling process ( Figure 4B and MovieS02 and MovieS03). However, in contrast to the dense arrays, this engulfment is stable during cell spreading and migration. F-actin and membrane signals are enhanced at the location of the pillars through the mechanism described above already after the initial contact with the surface. We observe a highly dynamic situation, where the F-actin puncta are forming and disappearing at various locations in the cell as a function of time (see below for more detailed description of this process).

Fibroblast migration
Once the initial adhesion and spreading has occurred, fibroblasts start migrating. We per-  Figure 5D presents a brief summary of the cell migration data. For arrays with 2µm pitch, migration data was recorded from a total area of 18mm 2 , a vary large area considering high resolution patterning. This area corresponds to 4.5×10 6 pillars for 2µm arrays and to 32×10 6 nanopillars for 0.75µm arrays.
Due to large variations in the migration rates for individual cells, cell trajectories were analyzed by a method proposed by Banigan et.al. 41

and described in detail in Materials and
Methods. Briefly, for each data set, a probability density function of step sizes was generated and fitted to a double gaussian model. In this model, the heterogeneously migrating cell population is modelled as the sum of two (slow and fast migrating) randomly migrating cell populations with characteristic diffusion coefficients D 1 and D 2 as well as a weighting factors c 1 and c 2 for each population.
The resulting plots ( Figure 5E) for Experiment 1, and determined values for diffusion coefficients for both experiments ( Figure 5F and Figure 5G) demonstrate that cell migration is indeed inhibited on sparse nanopillar arrays compared to glass. Interestingly, the dependency on the nanopillar density is not directly proportional to nanopillar array pitch. On dense arrays (0.75-1µm) cell migration is not significantly hampered, but on sparse arrays migration is strongly reduced, with migration recovering on 10µm pitch. Comparing with the glass control, migration on the 2µm pitch substrates is reduced by 50% and 25% for the slow and fast migrating cell populations respectively. Based on not overlapping confidence intervals for data on glass and on 2µm pitch arrays, we can conclude that this difference is statistically significant with p-value which is at least p = 0.05. 42,43 The slow population has a diffusion coefficient that is 7-10 times lower than the coefficient for the fast cell population on all substrates. The weighting factor, plotted as c 1 /(c 1 + c 2 ) in Figure 5H, reveal that about 25% of the analyzed displacements are modelled in the less mobile D 1 population.

High resolution investigation of fibroblast migration
To gain insight into the processes occurring during cell migration, especially the origin of observed migration rate differences, live-cell confocal imaging was performed on migrating fibroblasts expressing LifeAct-mNeonGreen and PH-PLCd1-mScarlet. Selected time points from migrating fibroblasts on nanopillar arrays are shown in Figure 6. On dense nanopillar arrays, fast dynamics of densely spaced F-actin puncta is observed (see inserts in Figure 6A and Supplementary MovieS04 -MovieS08). Both F-actin and membrane signal are more intense close to cell periphery, indicating that the cells are close to the glass substrate in these regions and wrap along the full length of the pillars. Only a small enhancement of the membrane signal at the locations of nanopillars is observed ( Figure 6B). This indicates, that apart from the peripheral regions, only small and dynamic membrane indentations are formed for cells on dense arrays. On spares arrays, the membrane signal is highly enhanced and stable ( Figure 6D), indicating that the cell membrane wraps along the full length of the nanopillars. The cell body maintain a stable contact with the glass support (see also Supplementary MovieS09 -MovieS12). On sparse arrays membrane at the trailing edge of the migrating cell appears to remain associated with the nanopillars, which could be one reason for reduction in cell migration on sparse arrays. In contrast to the membrane signal ( Figure 6D) which is uniform and stable, F-actin puncta seen in Figure 6C

Discussion
Cell adhesion, spreading and migration in vitro are highly studied processes, but have mainly been studied on flat substrates such as glass, or on gels or patterned surfaces. However, the geometry of the nanopillars introduce a new factor into the mechanobiology of cell adhesion and migration. To successfully manoeuvre the nanopillars, the main challenge for the cells appears to lie in shaping the membrane to conform to the nanopillars. Theoretical models of cell adhesion and membrane conformation on nanopillar arrays emphasis the balance between membrane bending and surface adhesion. 36,44,45 Bending the cell membrane around each nanopillar requires energy to overcome the membrane bending stiffness, while adhesion to the surface of both the nanopillars and the substrate is energetically preferred for adherent cells used in this study. Thus, on dense nanopillar array pitches, the cells adopt a suspended state while on sparser arrays the cells adopt a conformal state, engulfing the nanopillars and adhering to the substrate. In this work, these two states were indeed observed in fully spread fibroblasts, with the transition occurring between 1µm and 2µm nanopillar array pitch. In addition, the detailed imaging performed here allowed several nuances to be observed.
During initial cell adhesion, nanopillar engulfment was observed regardless of nanopillar array pitch, which was not predicted in the theoretical models or observed experimentally before. The engulfment of the nanopillars likely has two interrelated contributing factors: membrane tension and actin polymerization forces. In general, depending on cell state and conformation, cells maintain significant membrane reserves stored as microscopic bends and buckles in the membrane. 46,47 This could facilitate the initial engulfment of nanopillars regardless of array pitch. As the cell adhere and spread on the substrate, the membrane tension might increase, making the engulfment less preferred. The distinction between these states has a strong influence on cell migration rates. Previous work with silicon nanowires and nanopillars has indeed shown strong reduction of migration in the sparse array regime, while less of a reduction was observed on dense arrays. 25,26,49,50 The mechanisms for reduction in cell migration on sparse arrays are attributed to cytoskeletal and membrane entanglement by the nanostructures. Our results support these general descriptions, as in the sparse array regime both membrane engulfment and strong dynamic F-actin puncta association were observed. The requirement of continuous reorganization of the membrane and cytoskeleton in response to the moving cell likely causes a significant reduction of migration rates. A further effect of the nanopillars was observed where membrane residues were left behind on the nanopillars during retraction of the trailing edge. This indicates a strong adhesion between the membrane and nanopillars, sufficiently strong that the membrane does not release, but remains bound and forms membrane tethers.
Strong membrane adhesion of cells to silicon nanowires has been reported, 28 and extraction of membrane tethers is a well-known method of assessing membrane tension of cells. 51 Such tethers are produced when an outward force is applied to a small area of the cell membrane and requires the loss of association between the cytoskeleton and cell membrane. Thus, the membrane binding and tether extraction at the trailing edge of migrating cells is a further hindrance for cell migration.
There are several opportunities for further investigations of cell migration on this platform. The exact mechanism of recruitment and function of the F-actin puncta or bundles around the nanopillars are not known. Using superresolution imaging we have previously shown that in HeLa cells these puncta are in fact rings or cylinders formed around the nanopillars. 27 In a recent report by Hanson et. al., similar puncta and rings were observed and attributed to cell adhesion to nanopillars. 35 As F-actin puncta were observed to be highly dynamic in this work, disappearing and appearing over a few min, it does not seem likely that these F-actin enrichments are solely involved in adhesion. Especially, as membrane engulfment of nanopillars was generally observed both with and without F-actin enrichment at the nanopillar site, this is not a continuous requirement for cell adhesion to nanopillars. We consider it likely that the dynamic F-actin enrichments are involved in the reorganization of the cell membrane as the cells migrate on nanopillar arrays, in addition to adherence processes. In a recent study, Mettlen et al showed that membrane curvature induced by using liposomes with various sizes, combined with PI(4,5)P 2 , and PI(3)P signalling are needed to trigger actin polarization. 52 Some actin assemblies induced by membrane curvature on the sparse arrays (see Supplementary MovieS15) resembles actin comets induced in a cell-free assay for the role of phosphoinositide in actin polymerization. 52 The role of nanopillar surface chemistry also remains to be investigated. Although geometry was attributed the leading role in reduced neuronal migration on nanopillars (Si, SiO 2 and Pt were tested), 26 it cannot be ruled out that the surface chemistry of SU-8 could play a role in the strong membrane interactions observed, and should be investigated further.

Nanopillar fabrication
All chemicals and reagents were purchased from Sigma-Aldrich (Oslo, Norway) unless otherwise specified. SU-8 nanopillar arrays on glass were prepared as previously described. 27 Briefly, 1µm thick SU-8 films were spin coated on 0.17mm (# 1.5) glass cover slips (Menzel-Gläser borosilicate glass) and soft-baked for 1min. Fluorescent SU-8 was made by including 100 µg/ml oxazine-170 in the SU-8 solution. Nanopillars were defined using an Elionix GLS-100 EBL-system. Samples were post-exposure baked for 3min and developed for 40s using mr-Dev 600 (Microchem, USA). The resulting nanopillars were 1µm high and had tip diameters in the range of 90 nm. All samples were treated with oxygen plasma before use. Probability density functions of cell displacements were analysed and fit to a double Gaussian function as described by Banigan et. al. 41 Probability density function of cell displacements was constructed by assigning a chosen number of displacements m to each bin (m = 700 was chosen by Banigan et.al 41 and was used in this work as well). The position of each bin was the average step size of that bin, while each bin was assigned a weight W i = 1/(r i max − r i min ), where r i max and r i min are the maximum and minimum displacements in that bin. This gives a normalized probability density function (which in our case summed to 2 as absolute values of all displacements were used). Note that we also pooled all x and y displacements. This probability density function was then fit to a double Gaussian of the form:

Cell migration
where c 1 + c 2 = 2 are the population weighting coefficients, D 1 and D 2 are the characteristic diffusion coefficient for each population, r is the cell displacement distance and ∆t is the used frame rate. The double Guassian was fit to the probability density functions for each sample using least squares fitting, and 95% confidence intervals for each parameter were extracted as error estimates. Data was analysed using custom Python scripts and the lmfit-py package for fitting. 53

High resolution cell imaging and immunofluorescence
NIH-3T3 cells were transduced with LifeAct-mNeonGreen targeting actin and the PH domain of PLCδ1 fused to mScarlet targeting the plasma membrane and stably selected on 10µg/mL blasticidine and 1µg/mL puromycin. The cells were imaged from 0-24h after seeding or during seeding after trypsination at 37 o C and 5% CO 2 with Zeiss LSM 800 Airyscan (Axiovert  Table 1: Values of fitted parameters D 1 , D 2 and c 1 weight coefficients from the double Gaussian model used to analyze the migration data (see Materials and Moethods). Weight coefficient c 2 for the second population of cells can be calculated from: c 2 = 2 − c 1 . Error bars (±) are 95% confidence intervals for each parameter extracted from the least squares fitting of the double Gaussian model to the migration data.

Supplementary Movies
• Examples of cells settling on dense and sparse nanopillar arrays.
In all movies, NIH-3T3 fibroblasts expressing LifeAct-mNeonGreen (green) and PH-PLCd1-mScarlet (orange) were used and the cells were imaged for around 4h, capturing the initial cell attachment and spreading on various arrays. Nanopillar arrays are visible in the bright field images and in the oxazine-170 channel (blue). In some movies, pillars cover only part of the imaged sample area, so the cell behaviour on glass can be directly compared with the behaviour on the pillar arrays. In all movies, NIH-3T3 fibroblasts expressing LifeAct-mNeonGreen (green) and PH-PLCd1-mScarlet (orange) were used and the cells were imaged at 2.5 to 5min intervals, capturing cell migration dynamics at the time scale of <1h.
• Example of a migration data set.
NIH-3T3 fibroblasts expressing LifeAct-mNeonGreen (green) were imaged for 14h at 37 o C, 5% CO 2 using a 10X objective of the EVOS FL Auto 2 microscope (Thermo Fisher Scientific).         . Actin recruitment is transient on both dense and sparse arrays, but on the sparse arrays, actin is assembled for longer time periods and on a higher percentage of nanopillars. Also, intensity of the F-actin signal recorded for the sparse array is higher. Figure 8: (A-B) Actin and vinculin organization in spreading fibroblasts. Fibroblasts were seeded for 2h and 4h before fixation and labelling Alexa488-phalloidin (red) and Alexa555labelled vinculin (green) on dense arrays (1µm pitch) or sparse arrays (2µm pitch). Yellow color indicates actin-vinculin overlap. All scale bars are 10µm. (C) Excerpts from a timelapse of a fibroblast transduced with talin-GFP (green) on a 2µm pitch nanopillar array (magenta), highlighting the formation of focal adhesions at the leading edge (cyan square) and trailing edge (white square) over a 11-min period. Scale bar overview image 10µm, excerpts 2µm.