Extracellular matrix alignment dictates the organization of focal adhesions and directs uniaxial cell migration

Physical features of the extracellular matrix (ECM) heavily influence cell migration strategies and efficiency. Migration in and on fibrous ECMs is of significant physiologic importance, but limitations in the ability to experimentally define the diameter, density, and alignment of native ECMs in vitro have hampered our understanding of how these properties affect this basic cell function. Here, we designed a high-throughput in vitro platform that models fibrous ECM as collections of lines of cell-adhesive fibronectin on a flat surface to eliminate effects of dimensionality and topography. Using a microcontact printing approach to orthogonally vary line alignment, density, and size, we determined each factor's individual influence on NIH3T3 fibroblast migration. High content imaging and statistical analyses revealed that ECM alignment is the most critical parameter in influencing cell morphology, polarization, and migratory behavior. Specifically, increasing ECM alignment led cells to adopt an elongated uniaxial morphology and migrate with enhanced speed and persistence. Intriguingly, migration speeds were tightly correlated with the organization of focal adhesions, where cells with the most aligned adhesions migrated fastest. Highly organized focal adhesions and associated actin stress fibers appeared to define the number and location of protrusive fronts, suggesting that ECM alignment influences active Rac1 localization. Utilizing a novel microcontact-printing approach that lacks confounding influences of substrate dimensionality, mechanics, or differences in the adhesive area, this work highlights the effect of ECM alignment on orchestrating the cytoskeletal machinery that governs directed uniaxial cell migration.


INTRODUCTION
Directed cell migration plays a fundamental role in numerous physiologic and pathologic processes, where biochemical and physical cues from the extracellular matrix (ECM) guide cells to their target destinations. [1][2][3] In wound healing, gradients of chemokines recruit a) Authors to whom correspondence should be addressed: bambren@umich.edu and chencs@bu.edu 2473-2877/2018/2(4)/046107/16 V C Author(s) 2018. 2, 046107-1 macrophages to wound sites to initiate the tissue repair process. 4 Similarly, during the progression of most solid cancers, gradients of growth factors produced by the tumor mass direct the invasion of endothelial cells required for angiogenesis and subsequent tumor growth. 5 While soluble cues clearly are important in mediating directed cell migration, physical features of the microenvironment have more recently been implicated in this process. For example, early work demonstrated that contact guidance produced by the orientation and spatial restriction of fibronectin fibrils during amphibian gastrulation direct mesodermal cell migration from the blastopore to the animal pole. 6 In the context of breast cancer, Provenzano et al. described how the collagenous stroma undergoes marked reorganization, resulting in radially aligned collagen tracts emanating from the tumor. This organization in turn appears to facilitate cancer cell escape from the primary tumor. 7,8 In both of these rather diverse processes, directional cues influencing cell migration arise from the fibrous ECM that cells negotiate during migration.
Given its ubiquity throughout the body, type I collagen gels have been widely used as a physiologically representative ECM model for in vitro cell migration studies. [9][10][11][12][13][14] These studies further implicate physical attributes of the ECM such as alignment, density, diameter, and stiffness as integral factors in directing cell migration. However, identifying the individual contribution of these factors proves to be quite challenging in natural materials such as collagen gels. For example, increasing the concentration of a collagen gel concurrently increases fibril density, matrix stiffness, and cell-adhesive ligand density while decreasing the pore size. Thus, bioengineered in vitro models that simplify the complexity of natural ECM and can decouple confounding factors have helped deepen our understanding of how the physical properties of the ECM regulate cell migration. [15][16][17][18][19] In particular, cell migration on micropatterned lines of adhesive ECM proteins has been suggested to recapitulate migration observed within in vivo 3D microenvironments composed of highly aligned fibers. [20][21][22] NIH3T3 fibroblasts migrating on single lines of patterned adhesive ECM proteins, termed "1D migration," exhibit a uniaxial cell morphology and undergo directed migration with similar speed as in 3D cell-derived matrices, whereas cells undergoing unconstrained migration on unpatterned 2D surfaces do not. 20 Similarly, 1D micropatterned substrates have also been used to reconstitute macrophage-tumor cell interactions, validating intravital observations in highly metastatic patient-derived orthotopic mammary tumors. 21 These and other studies 23 implicate the dimensionality of the ECM substratum in dictating the cell migration phenotype and suggest that 1D lines of ECM recapitulate key aspects of cell migration observed in 3D fibrillar tissue settings; however, orthogonal control over the adhesive area proved to be difficult in these studies. Furthermore, in addition to dimensionality, there may be other aspects of the ECM which are important in regulating the migration phenotype. Given that many mesenchymal tissues contain adhesive fibers of ECM proteins such as fibronectin and collagen, we wondered how restricting cell adhesion to linear tracks of ECM influences migration, and furthermore, if the geometry and organization of these patterns, independent of substrate dimensionality and mechanics, could dictate the cell migration mode.
To answer these questions, we designed a microcontact printing-based ECM parameter screening tool where we modeled fibrous ECM as arrays of cell-adhesive micron-scale linear elements. We then varied over graded steps three pertinent features reflecting the distribution of adhesive ligands in fibrillary microenvironments (line alignment, density, and width) and examined changes in the morphology, cytoskeletal architecture, and migratory behavior. Using these 2D micropatterned substrates to exclude the influence of substrate dimensionality and mechanics, we found that the alignment of ECM had a dominant effect on cell morphology and migration over other geometric factors, where high alignment induced a uniaxial phenotype and rapid cell migration in a directed fashion. Underlying directed cell migration is the matrix alignment correlated with highly organized focal adhesions (FAs) and actin stress fibers. Members of the Rho family of small GTPases have emerged as prominent players in cell motility, working to spatiotemporally modulate the signaling processes involved in cell adhesion and cytoskeletal dynamics. 79 In particular, Rac1 is necessary for lamellipodium extension preceding the formation of new adhesions to ECM and has been implicated in directed cell migration, where experimentally decreasing Rac1 activity switches cells from random to directionally persistent migration. 90 Given the observed influence of matrix alignment on directed cell migration, we further used two distinct molecular approaches to examine Rac1 localization and found that matrix alignment additionally influences the number and location of protrusive edges initiated by active Rac1.

RESULTS AND DISCUSSION
Multiparameter ECM screening arrays for high content imaging Given the ubiquity of micrometer-scale diameter fibrous proteins throughout mammalian extracellular matrices (ECM), we sought to develop a model that captured key aspects of adhesive ligand patterning in fibrous ECMs without confounding factors stemming from substrate dimensionality and mechanics. 15,16,29 To do so, we simplified fibrous ECMs into twodimensional (2D) patterns of microcontact-printed fibronectin lines where the angular dispersion (alignment), number (density), and size (width) of these cell-adhesive line elements could be orthogonally varied [ Fig. 1(a)]. Building upon previous work, 20 this microcontact printing-based approach allowed us to explore how the organization of adhesive ECM proteins influenced cell migration, while sidestepping confounding factors such as steric hindrance, 30-34 the requirement for proteolytic activity, [35][36][37][38] and the influence of substrate mechanics. 36,[39][40][41][42][43] Using custom MATLAB scripts, a prescribed number of initiation points were randomly selected in 2D space, and lines were extended in a direction selected randomly within prescribed bounds. To achieve submicron features, we employed a stepper with 5Â optical reduction resulting in silicon masters with SU8 photoresist line features ranging between 0.5 lm and 2 lm in width. In total, 252 unique ECM patterns were generated within a single 1.2 Â 1.2 cm 2 stamp, enabling highthroughput screening over a wide range of adhesive microenvironment parameters using high content tile-scan imaging [ Fig. 1(c)].
The desired production of micron-scale features ranging widely in spacing across a single stamp prevented us from using standard "forward" microcontact printing, where the featured stamp is protein inked prior to transfer. Indeed, when conventional "forward" stamping was employed, only patterns with greater than 400 lines/pattern could be successfully generated; at lower line densities, stamp collapse led to protein transfer from non-feature background areas. Thus, we employed an inverse microcontact printing or "stamp off" technique, 44 whereby a flat PDMS stamp is uniformly inked with adhesive protein and the activated inverse-featured stamp is first applied to remove background areas of protein before transferring the remaining protein pattern to the culture substrate [ Fig. 1(b)]. This method enabled the production of nearly all patterns over a range of line alignments (65 , 15 , 30 , 45 , 60 , and 90 ), densities (50,100,200, and 400 lines/pattern), and widths (0.5, 0.7, 1.0, and 2.0 lm) [ Fig. 1(d)]; arrays with 800 line/pattern at line widths of 1 and 2 lm possessed a larger line area than the total pattern area, resulting in a uniform region of fibronectin. Silicon masters were generated with 0.5 and 2.0 lm thickness photoresist, with 2.0 lm thick features generally leading to more consistent and faithful pattern reproduction. Although linear adhesive patterns below 0.5 lm have been produced previously by other means and may impact focal adhesion growth and dynamics, 45 the 0.5 lm wide lines achieved here match the limit previously reported for microcontact printing proteins using PDMS. 46 In summary, we successfully created multi-parametered micropatterned ECM screening arrays to model and study cell migration in fibrous microenvironments.
Array-wide parameter screen reveals that matrix alignment robustly dictates cell morphology and orientation influence mesenchymal cell migration. 47,48 High content tile-scan fluorescence imaging was performed on cells stained for their nucleus and actin cytoskeleton [ Fig. 2(a)], with subsequent semi-automated image analysis to extract an assortment of morphometric features. Of the features examined, we immediately identified two factors-cell aspect ratio and orientationwhich varied markedly across the array [ Fig. 2(b)]. In particular, when analyzing the entire dataset considering each parameter independently, we observed that decreasing line widths, decreasing line densities, and increasing alignment all led to an increase in the cell aspect ratio [Figs. 2(c) and 2(d)]. A high aspect ratio corresponds to a uniaxial cell shape and has been heavily implicated in directed cell migration, where the direction of the cell's long axis (or orientation) defines the direction of migration. 2 Further statistical analyses were performed to determine the relative strength of the effect of line alignment, density, and width on the cell aspect ratio and variation in cell orientation. When the degree of linear correlation between cell aspect ratio vs. alignment, density, and width was determined, we found that the aspect ratio had the highest correlation with alignment (r 2 ¼ 0.156, p < 0.0001). The aspect ratio was also independently correlated with density (r 2 ¼ 0.029, p < 0.0001) and width (r 2 ¼ 0.089, p < 0.0001). To understand the relationships among alignment, density, and width in explaining the variations in both the aspect ratio and the orientation, we performed multinomial multivariate linear regression model building using a dual-direction selection algorithm from a full interaction model, optimizing for the Akaike information criterion. The model selection proceeded for two rounds and eliminated two interaction terms from the model, without adding back either term. In our final model, alignment, density, width as well as the width by alignment and alignment by density terms were each significant for the multivariate outcome pair of cell orientation and aspect ratio ( Table I). The interaction term between alignment and density, for example, implies that at higher alignments, density more potently increases the aspect ratio. We also determined that the variation within cell orientation was significantly associated with alignment (p < 0.0001), but not with density or width. Taken together, the combination of high-throughput screening of matrix conditions and statistical analyses indicates that while line width and the density significantly altered the cell aspect ratio, only the alignment of adhesive ECM tracks strongly influenced cell shape towards a uniaxial phenotype and simultaneously exerted a pronounced effect in orienting cells towards the underlying central axis of pattern alignment. Given the effect of matrix alignment in these and numerous previous studies, our subsequent studies focused on the effect of matrix alignment, maintaining the line width constant at 1 lm.

ECM alignment polarizes cells and induces directed cell migration
The multiple regression parameter screen identified matrix alignment as the strongest factor in dictating cell morphology and orientation. Isolating alignment's individual effect on cell morphology, we held the line density and width constant and examined the cell aspect ratio and orientation (supplementary material Fig. 1). Confirming our initial findings, increasing matrix alignment resulted in an increased cell aspect ratio, causing cells to adopt an elongated morphology and orient preferentially towards the underlying pattern's direction of alignment [Figs. 3(a)-3(d)]. This trend was found to be consistent across all line density conditions examined (data not shown). Gross cell morphology alone, however, does not necessarily indicate cell polarity. Indeed, elongated cells on aligned ECM were occasionally observed to extend bidirectionally without evidence of a dominant leading edge. To better understand the influence of matrix alignment on cell polarization, we performed immunofluorescence staining for pericentrin to identify the microtubule organizing center (MTOC), as previous work suggests that the MTOC's location indicates cell polarity [ Fig. 3(e)]. 20,49 We first examined the orientation of the MTOC with respect to the pattern axis [ Fig. 3 of how ECM properties influence cell polarity remains incomplete, our study clearly demonstrates that the anisotropy of adhesive ECM patterning influences cell shape, intracellular organization, and polarity. Previous work has linked an elongated, uniaxial cell shape with increased migration efficiency, 20 and so, we next examined the functional consequence of ECM organization on migration speed and persistence using our two extremes of alignment: aligned (65 ) and non-aligned (690 ) patterns (supplementary material, Fig. 1). We performed live time-lapse imaging over 6 h, utilizing Hoechst-labelled nuclei to track cell movement with an automated image analysis algorithm. Representative temporal overlays of the cell shape on non-aligned patterns reveal a dynamic cell shape, while cells on aligned patterns maintained an elongated morphology throughout their migration track [ Fig. 4(a)]. Aligned patterns also produced more directionally persistent migration and faster speeds compared to cells migrating on non-aligned patterns, and furthermore, this effect remained consistently true over the full range of line densities examined degree of ECM anisotropy critically defines a uniaxial phenotype and directed cell migration. While many studies examining migration on aligned ECM report enhanced persistence and directionality, net migration speed has not consistently been observed to increase with directed migration. Our finding of increased speeds may stem from the comparison of non-aligned vs. aligned substrates with comparable underlying adhesive areas or the fact that proteolytic activity is dispensable for migration in this setting. each level of density examined [ Fig. 5(d)]. Non-aligned ECM resulted in a population of adhesions with high angular dispersion with respect to each other; in contrast, aligned ECM promoted the organization of FAs along the cell's long axis [ Fig. 5(a)]. This difference in adhesion angular deviation mirrored differences in migration speed across multiple ECM densities and proved to be significantly linearly correlated with an R 2 value of 0.916 [ Fig. 5(g), p < 0.0001]. Correlations between the other adhesion metrics quantified and migration speed proved to be non-significant [Figs. 5(e) and 5(f)]. These findings were additionally confirmed with HT1080 fibrosarcoma cells (supplementary material, Fig. 3). The observation that the global orientation of adhesions uniquely predicts migration speed is in contrast to recent reports highlighting FA size as a critical predictor. However, in contrast to the work at hand, these studies were performed on flat unpatterned substrates of varying stiffness, highlighting again a clear distinction between aligned uniaxial cell migration and unconstrained 2D migration. 71 F-actin organization reflected the organization of FAs, where cells on non-aligned ECM with randomly oriented adhesions also had disorganized F-actin stress fibers [ Fig. 5(h), top]. In contrast, uniaxial cells on aligned ECM possessed highly aligned stress fibers running predominantly in the direction of the long axis of the elongated cell body, mirroring the co-alignment of FAs [ Fig. 5(h), bottom]. Alignment of FAs and F-actin bundles supports observations that FA sites facilitate actin stress fiber assembly 72,73 and intracellular force transmission to the ECM occurs via actin engagement with FA proteins. 63,74,75 For cells on aligned ECMs, we observed intense F-actin staining at the cell walls running along the direction of matrix alignment and the putative direction of migration [ Fig. 5(h), arrows]. Cells on non-aligned ECM also possessed intense F-actin bundles along straight edges of the cell's periphery. Regardless of cell shape and orientation, FAs and active lamellipodial protrusions were seldom observed at such locations possessing robust stress fibers, echoing previous observations employing cellsized geometric micropatterns. 76,77 Given that the formation of F-actin stress fibers has been associated with augmented RhoA activity, [78][79][80][81] previous studies suggest a mutual exclusion between RhoA and Rac1 localization, [82][83][84][85][86][87] and Rac1 activity is heavily implicated in protrusion activity, 2,79,88,89 we hypothesized that highly elongated cells on aligned ECM would preferentially form protrusions at the distal ends of the cell in the direction of ECM alignment. Furthermore, we predicted that the stability of these protrusions would be enhanced compared to cells on disorganized (non-aligned) ECM.
By identifying ruffling edges in high resolution spatiotemporal time-lapse image series, we were able to determine the location and number of active protrusions on aligned (65 ) Fig.  4(c)] support the notion that the maintenance of a single location of protrusive activity underlies efficient migration 20,90 and clearly demonstrate that ECM organization dictates protrusive activity.
Given these striking distinctions in the location of active protrusions and Rac1's known role in driving lamellipodial protrusion, 90,91 we next examined whether ECM alignment also altered the localization of active Rac1 signaling. Glutathione-S-Transferase (GST)-tagged protein binding domain (PBD) and immunofluorescence staining were used to identify intracellular locations of enriched active Rac1 on aligned (65 ) and non-aligned (690 ) patterns with 100 lines/pattern [ Fig. 7(a), supplementary material, Fig. 1]. Cross-correlating fluorescence intensities from Rac1 and F-actin images allowed us to identify locations of high co-localization of these two proteins, both critical to protrusion activity [ Fig. 7(b)]. Qualitatively, we observed that cells on non-aligned ECM possessed multiple protrusions with co-localization of these two signals. Conversely, cells on aligned ECM possess fewer protrusions with hot spots, and these protrusions were additionally in the direction of ECM alignment and putative migration. These observations suggest that ECM alignment influences the number and location of active protrusions, contributing to the stable and directional protrusions underlying uniaxial directional migration. To further support these observations, we employed a F€ orster resonance energy transfer (FRET)-based approach to image Rac1 activity modified from a previously reported construct. Within this single chain bioactivity reporter, FRET between molecules occurs upon binding of PBD to Rac1. Live imaging using this reporter revealed similar localization of Rac1 activity as a function of ECM alignment, with preferential Rac1 activation in directions corresponding to the underlying pattern alignment [Figs. 7(c) and 7(d)].
We next transiently transfected 3T3s with a green fluorescent protein (GFP)-tagged constitutively active mutant copy of Rac1 (RacV12-GFP, abbreviated RacV) and plated cells onto aligned ECM (65 , 200 lines/pattern, 1 lm width). As expression of the plasmid was heterogeneous, we segregated transfected cells into three distinct populations based on GFP intensity: no, low, and high expression [Figs. 7(e) and 7(f)]. High RacV expression resulted in an increased frequency of active protrusions off-axis to the direction of ECM alignment relative to nonexpressing cells [25% vs. 8%, Fig. 7(g)]. Strikingly, the loss of protrusion directionality with increasing RacV expression paralleled decreases in migration speed where the highest expression of RacV resulted in active protrusions along the entire cell periphery and virtually no cell movement [Figs. 7(h)-7(i)]. These findings build on previous work by Pankov et al. showing how graded levels of Rac1 activity switch cells between directed and random migration on unpatterned 2D substrates. 90 In addition to total levels, it is now clear that localization of active Rho GTPase are critical to cell polarity and subsequent movement. 87,92 Our findings suggest that ECM alignment and therefore anisotropy of the local adhesive microenvironment can modulate cell shape and cytoskeletal architecture to spatially define the location of such signals.

Summary and outlook
Many engineered models such as 1D patterned lines, nanopatterned ridges and grooves, and electrospun fibers have been utilized to understand how contact guidance cues provided by fibrous ECM influence cell migration. 20,23,60,93 Previous studies primarily compared aligned, adhesion-restrictive substrates versus uniformly coated ECM substrates, resulting in differences in ECM anisotropy as well as total adhesive area. In contrast, this work provides a microcontact printing-based platform modeling the ECM as collections of fibronectin lines with graded variations of alignment, density, and width to identify how combinations of these parameters influenced cell migration. With high content imaging and statistical analyses, we identified matrix  alignment as a critical parameter in influencing cell morphology, polarization, and migratory behavior. Interestingly, we find that cells on highly aligned ECM possess organized FAs. This alignment of FAs correlates strongly with directional, high speed migration in stark contrast to other morphometric features of FAs. Highly aligned FAs and associated F-actin stress fibers result in the formation and stabilization of cell protrusions in the direction of ECM alignment, and our data suggest that this control occurs through the localization of active Rac1. This work clearly indicated that anisotropy of the adhesive ECM, independent of substrate dimensionality, dictates the cytoskeletal architecture required for directed cell migration. Furthermore, these micropatterns could provide a high-throughput screen of ECM parameters on other cell functions and may aid in identifying novel therapeutics that selectively inhibit directionally invasive tumor cells. These angles reflect bounds between which each line angle was selected from a uniform distribution. Individual fields were exported in vector format, assembled en mass in Autocad (Autodesk, Mill Valley, CA) into arrays, and printed in chrome on a quartz reticle (Advance Reproductions, North Andover, MA). Using a Nikon G4 stepper enabling 5Â optical reduction in line widths (0.5, 0.7, 1.0, and 2.0 lm) (Penn Regional Nanotechnology Facility, Philadelphia, PA), patterns were transferred to a 5 in. silicon wafer spuncoat with 500 nm thickness Microposit S1813 photoresist (MicroChem, Westborough, MA). Following development, polydimethylsiloxane (PDMS) (Sylgard 184, Dow Corning, Midland, MI) was cast onto wafers and cured at 60 C to generate micropatterned stamps.
To enable pattern visualization, fibronectin from human plasma (Fn, Corning, Corning, NY) was fluorescently tagged with AlexaFluor555 succidinyl ester following the manufacturer's protocol (Invitrogen, Carlsbad, CA). Briefly, a 1 mg/ml solution of fibronectin in 1 M sodium bicarbonate reacted with AlexaFluor555 succidinyl ester at a 9-fold molar excess for 2 h at RT with continual agitation. Unconjugated fluorophores were removed by overnight dialysis (6.5 kDa cutoff), and the concentration of conjugated fibronectin (Fn555) was determined by absorbance at 280 and 555 nm using a spectrophotometer. For microcontact printing, Fn555 was diluted to 50 lg/ml in phosphate buffered saline (PBS) and adsorbed uniformly onto a 1.5 cm by 1.5 cm piece of flat PDMS cast from a cleaned silicon wafer. Patterned stamps containing the negative of the intended final features were activated via UV ozone and applied to the Fn-inked stamp to selectively remove background (non-feature) areas of Fn. The remaining patterned Fn was transferred to the final cell culture substrate by applying the inking stamp a UV ozone activated, PDMS coated coverslip (via spin coater, 5000 RPM). Following storage overnight to allow for recovery of hydrophobicity, substrates were incubated in Pluronics F-127 (0.2% w/v in deionized water, Sigma-Aldrich, St. Louis, MO) for 30 min at 25 C to prevent non-intended protein adsorption and undesired cell adhesion to non-printed regions.

Immunofluorescence
To examine cell morphology and the organization of the actin cytoskeleton, cells were fixed in 4% phosphate-buffered paraformaldehyde for 10 min and then permeabilized with 0.03% Triton X-100 for 10 min. Filamentous actin was stained with phalloidin-AlexaFluor488 (Life Technologies), and cell nuclei were stained with Hoechst33342, (1 lg/ml, Sigma-Aldrich) blocked in 2% bovine serum albumin. For immunofluorescence staining, fixation and permeabilization were performed as above (unless specified otherwise) followed by incubation in blocking solution (10% fetal bovine serum in PBS) for 1 h, primary antibody (below) dilution in blocking solution for 1 h, three PBS washes for 5 min each, a 1:1000 dilution of AlexaFluor conjugated IgG antibody (Life Technologies) in blocking solution for 1 h, and two PBS washes for 5 min each. To stain the microtubule organizing center (MTOC), a 1:500 dilution of rabbit anti-pericentrin was employed as the primary antibody (PRB-432C, Covance). To stain focal adhesions, samples were permeabilized and fixed simultaneously and a 1:500 dilution of monoclonal mouse anti-vinculin antibody was employed as the primary antibody (V9264, Sigma-Aldrich). To stain for Rac activity, samples were incubated with glutathione-eluted protein binding domain (PBD)-GST overnight at 4 C (Cytoskeleton), rinsed with PBS, and incubated with goat anti-GST antibody (27-4577-01, GE Healthcare). Imaging was performed on a Nikon Eclipse Ti (10Â) or on a Zeiss 710 laser scanning microscope (40Â), and images are presented as maximum intensity projections.

Rac1 FRET imaging
To examine Rac1 activity, we modified a previously designed RaichuEV-Rac1 FRET biosensor 24 by replacing the mTurquoise/YPet fluorophore pair with Clover/mRuby2 (RaichuEV-Rac1-CR) to reduce fluorophore bleed-through, improve the FRET dynamic range, and enable incorporation into lentiviral vectors. NIH3T3s were transiently transfected with the RaichuEV-Rac1-CR biosensor using Lipofectamine LTX with Plus Reagent (ThermoFisher Scientific) 24 h before seeding onto microcontact-printed substrates. Cells expressing RaichuEV-Rac1-CR were imaged 12 h after seeding. The binding of active Rac1 was detected by imaging the FRETdependent, intramolecular emission fluorophore (mRuby2) from RaichuEV-Rac1-CR as previously described. 12 Briefly, optimal FRET acquisition settings were determined for the Zeiss LSM 800 confocal microscope and strictly maintained during all subsequent FRET imaging; intensity levels of biosensor expression were similarly carefully controlled and maintained between selected cells. Images of mClover and mRuby2 were obtained for each z-plane under 488 nm and 567 nm illumination. Summed projections of confocal z-stacks were generated using ImageJ software. Images were first background subtracted, and a binary mask was applied by thresholding to the Acceptor (mRuby2) channel to isolate the cellular signal. Pixel-by-pixel FRET to Donor (Clover) ratio images were generated in ImageJ. All the resulting FRET ratio images were processed with a 3 Â 3 median filter to remove any hot pixels and presented in a scaled 16 color lookup table (ImageJ).

Microscopy and image analysis
For migration studies, sample media were supplemented with 1 lg/ml Hoechst33342 and samples were incubated for 30 min to label cell nuclei. Coverslips were then transferred to Attofluor chambers (ThermoFisher Scientific) and media were refreshed. Samples were imaged every 10 min for a duration of 6 h on a Nikon Eclipse Ti epifluorescence microscope equipped with a custom-built environmental chamber (37 C, 5% CO 2 ). Following raw image export, nuclear tracking was performed with a custom Matlab script predicated on the IDL Particle Tracking code. 25 Briefly, parameters to threshold and locate the centroids of cell nuclei were identified and applied uniformly across the entire dataset. Centroids of nuclei in serial images were linked using IDL to define migration tracks. Migration speed was calculated as the total tracked distance over the total tracking duration. Persistence was defined as the distance between initial and final positions normalized to the total tracked distance. Cells that underwent proliferation or were non-migratory over the tracked duration were not analyzed, except in Rac1 perturbation studies. Kymographs were generated in ImageJ.
Additional Matlab scripts were created to analyze cell morphology and focal adhesion characteristics. For cell morphometric data, images were acquired from 4 0 ,6-diamidino-2-phenylindole/phalloidin-stained samples. For focal adhesion analysis, images were acquired from samples immunostained for vinculin. In both cases, images were imported, background filtered, and manually thresholded, with identical threshold values applied across entire image sets. Outlines of individual features were extracted, and shape characteristics (via regionprops) including the area, aspect ratio, and orientation were exported.

Statistics
Significance was determined by one-way analysis of variance (ANOVA) with Bonferroni post hoc tests and generally established with p < 0.05, unless specified otherwise. For cell and adhesion orientation data, angular means were determined using circular statistics in Matlab (circstat). Correlation between variables was tested using Pearson's product-moment correlation and the correspondent test. 26,27 Multinomial multivariate linear regression model building was performed using a full linear regression model, and backward selection was performed using Ftest p-values. Comparisons between models of different sizes were performed using the Aikake Information Criteria. 28 Differences within variability between groups were tested using Levene's test. Analysis was performed using GraphPad Prism 6 or program R (v3.1.0).

Ethics approval
No ethics approval was required for the experiments described in this study.

SUPPLEMENTARY MATERIAL
See supplementary material for a substrate map depicting the patterns utilized in Figs. 1-7 and additional cell migration and focal adhesion data utilizing HT1080 fibrosarcoma cells.