Surface-templated hydrogel patterns prompt matrix-dependent migration of breast cancer cells towards chemokine-secreting cells

Abstract

This paper describes a novel technique to fabricate spatially-defined cell-laden collagen hydrogels, using patterned, non-adhesive polyacrylamide-coated polydimethylsiloxane (PDMS) surfaces as a template. Precisely patterned embedded co-cultures of breast cancer cells and chemokine-producing cells generated with this technique, revealed matrix- and chemokine isoform-dependent migration of cancer cells. CXCL12 chemokine-secreting cells induce significantly more chemotaxis of cancer cells when the 3D extracellular matrix includes components that bind the secreted CXCL12 chemokines. Experimental observations using cells that secrete CXCL12 isoforms with different matrix affinities together with computational simulations show that stronger ligand-matrix interactions sharpen chemoattractant gradients, leading to increased chemotaxis of the CXCL12 gradient-sensing CXCR4 receptor-expressing (CXCR4+) cells patterned in the hydrogel. These results extend our recent report on CXCL12 isoform-dependent chemotaxis studies from 2D to 3D environments and additionally reveal the important role of extracellular matrix composition. The developed technology is simple, versatile and robust; and as chemoattractant-matrix interactions are common, the methods described here should be broadly applicable for study of physiological migration of many different cell types in response to a variety of chemoattractants.

1. Introduction

Gradient formation and sensing is a complicated process involved in many physiologic and pathologic processes. Cells will change morphology and move toward a chemical gradient depending on the shape, dynamics, and magnitude of the gradient [1]. The simplest model of gradient formation involves diffusion of soluble factors away from cells that secrete them. However, in vivo there are two types of microenvironmental interactions that define gradient formation and sensing: ligand-matrix, and ligand-cell. Several studies have demonstrated the importance of complex gradient shape and dynamics in vivo, driven by ligand-matrix interactions [2, 3] and by active cell-dependent ligand scavenging [4]. Hence, to properly model and experimentally manipulate these complex chemotactic processes, we need experimental systems that recreate these environmental influences on gradient formation. In this work, we develop a technique that enables robust and versatile definition of in vitro multi-cellular / microenvironment interactions, in physiologically-relevant 3D environments, and utilize this technique to study the relationship between ligand-matrix and ligand-cell interactions on migration of breast cancer cells.

A few in vitro assays exist to recreate how ligand-matrix and ligand-cell interactions collectively guide gradient formation. Transwells [5], hydrogels [6–10], and microchannels [11–16] are typically used to spatially pattern cells, define morphogenetic and chemotactic gradients, and monitor cell morphology and chemotaxis. Although these assays isolate individual aspects of gradient formation and sensing, they fail to replicate how multiple cell types and matrix interactions together define a gradient. We previously developed an experimental source-sink system to replicate the formation of defined soluble gradients between spatially-patterned cells that secrete ligands (source) and cells that scavenge ligands (sink) [17, 18]. These previous studies capture the involvement of multiple cell types in source-sink gradient formation and the role of ligand binding to the device- and cell-surfaces. However, the relatively small surface area with limited amount of binding sites available in this simple two-dimensional assay is not sufficient to address the potential influence of ligand-matrix interactions, as compared to the in vivo situation in which 3D matrices provide a significantly greater concentration of binding sites.

In this work, we develop a novel patterning system to spatially pattern cells within an extracellular matrix (ECM), creating a model tissue-like environment for studies of directional cell migration. There are several techniques to create desired hydrogel patterns, such as laser lithography [19–25] and microchannel-guides [26–28], that often require significant expertise, specialized instruments and complicated processing. Our approach enables the precise positioning of multiple cell types within a 3D matrix, using relatively simple tools and expertise that should be accessible to most wet-labs. As a first application of this technology, we spatially pattern cells engineered to secrete the α- and β- isoforms of the CXCL12 chemokine and CXCR4+ cells that respond to CXCL12, while varying the composition of the surrounding matrix. Using this breast cancer model system, we demonstrate (1) the ability to systematically control ligand-matrix interactions via matrix composition; (2) the ability to spatially pattern multiple interacting cell types within a 3D matrix; and (3) the effects of ligand-matrix interactions on gradient formation, and on subsequent cell migration.

2. Materials and Methods

Unless otherwise stated, all chemicals and reagents for cell culture were purchased from Sigma- Aldrich, fluorescent dyes from Invitrogen and all other equipment and materials from Fisher Scientific.

2.1 Cell culture

MDA-MB-231 (231, ATCC) cells were primarily used for these experiments, and were cultured in fully supplemented Dulbecco’s Modified Eagle Medium (DMEM, with 10% Fetal Bovine Serum (FBS), 1% antibiotics-antimycotics). Some demonstration experiments were conducted with NIH 3T3 murine fibroblasts (cultured in fully supplemented DMEM), and human bone marrow endothelial cells (HBMECs, were a gift from Irma de Jong, cultured in fully supplemented Endothelial Cell Growth media, EGM2, with 5% FBS; Lonza). Standard trypsinization-based sub-culture protocols were used to passage cells prior to the experiment. We previously described culture, lentiviral transduction, and migration of 231 cells expressing CXCR4 towards 231 cells secreting CXCL12.[17, 18] Briefly, we transduced 231 cells sequentially with a CXCR4-GFP fusion [29] and NLS-AcGFP to facilitate receptor-based migration and image-based tracking of nuclei [18], respectively. We expressed CXCL12-isoforms fused to Gaussia luciferase (GL) upstream of the fluorescent protein mCherry in a pLVX IRES vector, to facilitate proportional fluorescence sorting for CXCL12-expressing cells [18].

2.2 Preparation of PAA-coated PDMS surface

10:1 (w/w) degassed mixture of PDMS and a curing agent were diluted with toluene (PDMS:toluene = 1:3). 12 mm diameter glass slides were dipped into the mixed solution and baked at 120 °C for 30 min. 9:1 (v/v) mixture of prepolymer solution (18.9 w/w% acrylamide, 0.33 w/w% pluronic F108, and 80.77 w/w% water) and photoinitiator solution (0.3 w/w% benzophenone and 99.7 w/w% 2-propanol) were added to the PDMS-coated slides and polymerized under UV (CL-1000, UVP: 8W × 4min) followed by an extensive wash in water [30, 31]. The PAA-coated PDMS slides were stored at ambient conditions and used for following hydrogel patterning within a few days (Figure 1A).

Figure 1.

Patterning of hydrogels on a cell-repellent surface. (A) PAA was polymerized on PDMS-coated glass slides and oxidized along mask patterns. Hydrogel precursor solutions are added on oxidized regions and cured. Gel solution was overlaid the patterns and incubated for the measurement. (B) Top view illustration of a patterned slide. Whole slide 12 mm diameter and oxidized spot 4 mm diameter. (C) Side view illustration of the whole setup. (D) Fluorescent image of patterned collagen gels containing NIH3T3 cells overlaid by collagen gel with 400 µm pattern-to-pattern distance and fluorescent image of patterned collagen gels containing HBMEC cells overlaid by the mixed gel (60 v/v% matigel + 40 v/v% collagen gel) with 200 µm pattern-to-pattern distance. Scale bar 400 um for NIH3T3 and 200 µm for HBMEC.

2.3 Fabrication of oxidized patterns

The SU-8 master molds were fabricated by soft lithography. Detailed procedure for fabrication of master molds can be found elsewhere [32]. The PDMS replicas of patterns served as oxidation stencils were placed in conformal contact with the PAA-coated PDMS slides and oxidized (100W × 10 min) [33]. Oxidized PAA-coated PDMS slides were immediately used for hydrogel patterning (Figure 1B).

2.4 Surface characterization

Water contact angles in air were measured on PDMS-coated slides and PAA-coated PDMS slides before and after oxidation by the sessile drop technique (4 µL of water) using a goniometer and analyzed with ImageJ. Data are expressed as a mean ± standard deviation (n = 3). Surface topography of PAA-coated PDMS slides oxidized through patterns was measured by atomic force microscope in ScanAsyst mode (Veeco, Bruker) and analyzed with Nanoscope (Veeco, Bruker).

2.5 Hydrogel patterning

2 mL of degassed PDMS was poured and cured into each well of a 6-well plate (Figure 1C). Each PDMS well was punched along the shape of a 12 mm glass slide and each oxidized PAA-coated PDMS slide was set at the bottom of the well. Trypsinized cells were mixed with neutralized type I bovine collagen (BD Biosciences) to create a suspension of 10 million cells / mL in 2 mg/mL of collagen. For cell-free experiments, 1 µm diameter fluorescent beads (Sigma) were added to the neutralized collagen solution. 6 µL of the collagen gel solution was dispensed over each adhesive pattern and allowed to polymerize for 45 min in a humidified cell incubator (37 °C, 5% CO2). 250 µL of either 2.5 mg/mL neutralized collagen or 2.5 mg/mL neutralized collagen supplemented with growth factor reduced matrigel (BD Biosciences: 61% laminin, 30% collagen IV, 7% entacin, and 2% other proteins including proteoglycans) mixture of collagen gel and matrigel solutions (75v/v% collagen I + 25v/v% marigel) was poured on each well and allowed to polymerize for 1 hour in the humidified incubator. 2 mL of cell culture media were added to each well and cultured at 37 °C. For observation of non-specific cellular movement, Latrunculin (10µM) was added to the media in control samples to prevent actin polymerization and hence migration.

2.6 Cell adhesion test

DMEM cell culture media containing MDA-MB-231 cells was cultured on tissue culture plastic, PAA-coated PDMS or oxidized PAA-coated PDMS slides in a 24-well plate and incubated for 24 hours. Cell culture media was aspirated and samples were rinsed twice in PBS. 1 mL of 4% paraformaldehyde solution was added to fix cells. Cellular actin cytoskeletal structures and the cell nucleus were labeled with Phalloidin and Hoechst dyes, following protocols obtained from the manufacturers. Images were collected with an epifluorescent microscope (TE-300, Nikon) and analyzed with ImageJ to determine cell density and spread area, expressed as a mean ± standard deviation (n = 6).

2.7 Modeling diffusion and binding kinetics

Finite element modelling of the diffusion and binding kinetics were conducted in COMSOL 4.2 (Burlington, MA), using a 1-D model geometry. Parameters for diffusion of CXCL12 (MW ≈ 10 kDa) were estimated based on the Stokes-Einstein equation relating hydrodynamic radius with diffusion rate, and published values in the literature for diffusion of large molecules in collagen [34]. Flux of CXCL12 molecules at one end of the model was calculated based on the estimated rate of CXCL12 production by cells [18] and on the geometry of the hydrogel system. A reaction-diffusion COMSOL module was used to simulate binding of the secreted soluble factors to uniformly distributed binding sites, representing matrigel proteoglycans in the collagen matrix. Kinetic binding parameters of CXCL12-α and CXCL12-β to matrigel proteoglycans were estimated as being similar to the binding kinetics of CXCL12 isoforms to heparan sulfate [35]. Results are reported as concentrations of soluble factors and as specific gradients across the length-scale of an individual cell. The specific gradient was determined by dividing the concentration difference across 50 µm by the average concentration over that distance [17].

2.8 Binding assay of CXCL12 to different matrix coatings

We previously described and characterized the production and activity of CXCL12- isoforms fused to Gaussia luciferase [18]. Measurement of Gaussia luciferase activity allows quantitative measurements of CXCL12 associated with cells, bound to extracellular matrix, and in cell culture supernatants. We collected supernatants from 10⁶ cells expressing secreted GL, CXCL12-α-GL, CXCL12-β-GL, or GL-negative that were plated in 60 mm dishes. After cells adhered in 10% FBS DMEM media overnight, cells were washed once in PBS, and media was replenished with DMEM with 0.2% albumin (Probumin Media Grade, Millipore). After 24 hours the supernatants were collected and imaged for GL activity and diluted in GL-negative cell supernatants to normalize GL flux activity to that of the lowest condition. Briefly, to measure GL flux, 1 µl of supernatant was queried in 1:1000 final dilution of the GL-substrate coelanterazine brought to a total volume of 100 µl of PBS, as we have described previously [18]. We added 50 µl of the GL flux-normalized supernatants to 96 well tissue culture plates that were pretreated with extracellular matrix (collagen, collagen and matrigel, matrigel only, or untreated tissue culture). Pre-treatment was achieved by incubating the 96 well plate with the appropriate solutions at 4 °C for 2 hours. The low temperature prevented polymerization but allowed adsorption of both collagen and matrigel components. Wells were then washed twice with PBS. Wells were grouped in quadruplicate for each supernatant type and extracellular matrix (ECM) type for three independent experimental setups. We incubated these plates for 30 minutes at 37°C to facilitate binding to the ECM. To remove unbound GL-species, we aspirated medium; washed each well in triplicate with an excess of PBS; and replaced medium with 50 µl of PBS per well. We imaged each well with the same substrate and PBS volumes as above. All bioluminescence images were acquired with the IVIS Lumina Series III (Caliper, LifeSciences), and data were analyzed with Living Image 4.3.1

2.9 Chemotaxisis of MDA-MB-231 cells

CXCR4+ and source cells (secreted GL as GL, CXCL12-α-GL as CXCL12-α, and CXCL12-β-GL as CXCL12-β) in mixture of collagen and matrigel matrix with 2 mL DMEM media were incubated for 24 hours and imaged with a fluorescent microscope (TE-300, Nikon). The representative leading cell-to-cell distance at day 0 and day 1 were each measured at five points with ImageJ and expressed as a mean ± standard deviation (n = 6 for GL, n = 6 for CXCL12-α, n = 6 for CXCL12-β, each in the presence of matrigel, n = 5 for CXCL12-β in the absence of matrigel).

2.10 Statistical analysis

All statistics are reported as means ± standard deviation. ANOVA tests were performed using a commercially-available software package (SigmaStat 3.5; Systat Software Inc., San Jose, CA), using the Tukey test for post-hoc pairwise comparisons.

3. Results

3.1 Characterization of hydrogel patterns on PAA-coated PDMS

This system enables control of three parameters: spacing between groups of patterned cells, gel composition, and cell type (Figure 1). PAA-coated PDMS slides were fabricated as described (Figure 1A–C) and clusters of cells encapsulated in hydrogels were patterned adjacent to each other with defined spacing (Figure 1D and Figure S1). The pattern-to-pattern distance was controllable and hydrogel patterns were stable regardless of cell type and gel composition. Obtaining successful patterns depends on carefully balancing topographic and surface chemistry features of the patterned surface template, at the contact line between the hydrogel precursor solution and the underlying substrate. Surface chemical heterogeneity and physical topology are both known to define the contact line at liquid-solid and liquid-liquid interface [32, 36]. We characterized the surface properties of PAA-coated PDMS by measuring the contact angle of sessile droplets, and AFM topographical measurements. The contact angle measurement of native PDMS and PAA-coated PDMS illustrates a dramatic reduction of contact angle before and after plasma oxidation (Figure 2A). While intact PAA-coated PDMS showed lower contact angle compared to native PDMS due to the hydrophilic nature of PAA introduced on PDMS, both surfaces demonstrated similar contact angles after oxidation. Meanwhile, AFM measurement reveals distinct surface topology of patterned PAA-coated PDMS (Figure 2B). A clear boundary at the edge of the pattern was observed and the step was about 50 nm at the boundary. In contrast, there was no distinct difference between intact and oxidized PAA-coated PDMS regions (Figure S2). We confirmed that either reducing oxidation time or leaving substrates for a few days resulted in failure of hydrogel patterning (data not shown), indicating that both surface chemistry and physical topography boundaries are necessary to support the formation of stable hydrogel patterns. We note that the accuracy of the patterning with 250 µm spacing was 246 ± 4.97 µm (n = 3) and the height of a droplet was 0.76 ± 0.041 mm (n = 6). We examined the limitations where the minimum spacing is 100 µm and minimum size of a pattern is 1.5 mm in diameter. Below these dimensions, oxidation and surface treatment did not work properly and failed to sustain the droplet shape.

Figure 2.

Surface characterization of PAA-coated PDMS. A) Contact angle values of water droplet on PDMS, oxidized PDMS, PAA-coated PDMS, and oxidized PAA-coated surface. (B) Surface topography of oxidized PAA-coated PDMS. (a) Height profile at the boundary and (b) highlighted section by white bar. (C–D) Cell adhesion test of MDA-MB-231 cells. (C) Cell density and (D) cell spread area of CXCL12-GL on TCP, PAA-coated PDMS, and oxidized PAA-coated PDMS. Images of a cell on (E) TCP, (F) PAA-coated PDMS, and (G) oxidized PAA-coated PDMS, respectively. Nucleus and cytosol were stained by blue and green, respectively. Scale bar 50 µm.

In addition to the ability to precisely pattern hydrogels, PAA serves as a non-adhesive cell-repellent material [37], and prevents migration of the cells over the PAA surface. To confirm this, the number of cells adhered to these surfaces were assessed using a standard cell-adhesion test, and indicates that adhesion was virtually eliminated, as compared to tissue culture plastic surfaces. Furthermore, cell spread area was also greatly reduced for the few adherent cells (Figure 2C–G), indicating that cells cannot attach sufficiently to spread, and therefore migrate. Cell adhesion and spreading were restored after oxidation of the PAA surface, but as this occurs only within the regions of patterned cells and not in the region of migration between cell patterns, these effects can be ignored in this migration assay. Hence, migrating cells are forced to interact with the hydrogel matrix.

3.2 Migration rates depend on matrix composition and CXCL12 production

We first confirmed that cells could migrate through the patterned matrix by monitoring radial movement of MBA-231 cells initially patterned in a single spot (Figure S3A). Radial dispersion of cells can be attributed to both random movement and growth of 231 cells (~24 hour doubling time). To confirm that soluble factors can be transported through the matrix, the cytoskeletal disruption compound Latrunculin A was added to the culture and significantly suppressed radial migration (Figure S3A–C). Confocal images of GL cells in collagen gel matrix at day3 indicates most of the cells settled down on the surface with a 15 µm-thick cell layer in the presence of latruncurin while the cells radially migrated through the gel matrix without latruncurin (data not shown). We seeded 6 × 10⁴ cells / hydrogel-droplet in theory and, based on confocal imaging, confirmed that over 99% (> 5.9 × 10⁴) of the cells settled down on the surface at t=0. We also observed the same trend for other cell types used in this manuscript, It indicates that the observed cell migration did not arise from the randomly distributed suspended cells in hydrogels. Thus we assume that most cells settled down on the surface at t=0 whereas they start to migrate in 3D over the course of time.

Next, to examine effects of cell-matrix and ligand-matrix interactions on cell migration in hydrogel matrices, we patterned encapsulated CXCR4+ migrating cells adjacent to cells that secrete its ligand, CXCL12-β, which we express as a fusion to Gaussia luciferase (GL). CXCL12-β cells were used because the cells secrete the most chemokine compared to other CXCL12-isoforms [18]. In control experiments, we replaced the CXCL12-β source cells with cells secreting GL unfused to a chemokine, which does not bind to CXCR4 or regulate signaling through this receptor. To assess migration in this system, we measured the gap spacing between the leading edge of the migrating and source cell populations after 24 hours and report the measured gap closure as the migration distance (Figure 3A and B). CXCL12-β drove more rapid migration than the GL control in the presence of matrigel. Without matrigel, migration rate was non-significant for CXCL12-β as compared to the GL control. This suggests that gradient generation and/or sensing of CXCL12-β requires interactions between CXCL12 and matrigel components (Figure 3C).

Figure 3.

Migration assay of CXCR4 cells in 3D gel matrix. (A) Fluorescent images of CXCR4 (green) and GL (red) at day0 and day1. (B) Fluorescent images of CXCR4 (green) and CXCL12- β (red) at day0 and day1. Scale bar 250 µm. (E) Migration of CXCR4-CXCL12 at day 1 in the absence and presence of matrigel (**p <0.01 and ***p < 0.001).

3.3 Matrix binding generates steep CXCL12 gradients

We hypothesized that matrigel provides binding sites for CXCL12 factors, which influences the formation of both soluble and bound gradients in the matrix. To evaluate the potential contribution of ligand-matrix interactions in shaping the gradient, we used computational models to simulate diffusion and diffusion with a matrix-binding component. In the diffusion model, unbound CXCL12-β rapidly diffuses through the gel matrix, and generates a negligible specific gradient to stimulate cell migration (Figure 4A). In contrast, ligand-matrix binding generates a steep specific gradient of unbound CXCL12-β, albeit at lower concentrations (Figure 4B). Bound gradients may have similar profiles, but cannot precede the soluble gradient as diffusion across surfaces is well-established to be significantly less than diffusion through an aqueous medium. These simulation results strongly indicate that interactions between the ligand and matrix facilitates gradient shaping and presents a possible mechanism explaining the matrix-dependent migratory differences observed in our chemotaxis experiments.

Figure 4.

Computational modeling of CXCL12-β gradients. Simulation of unbound CXCL12-β concentration as a function of distance and time and CXCL12-β specific gradient (defined as ΔC / C_avg / × across a 50 µm gap) as a function of distance and time in (A) diffusion only and (B) diffusion-matrix binding.

Additionally, our model suggests that when cell clusters are spaced more than 500 µm apart, chemotaxis cannot occur within the 24 hour time-frame of the experiment, as steep gradients cannot form across large distances rapidly enough. To confirm the importance of this spacing, we doubled the spacing to 500 µm, and observed no significant migratory differences prompted by GL-secreting cells and CXCL12-expressing cells in 24 hours (Figure S4). Hence, the ability of our technique to position the cell clusters close to each other is crucial parameter in observing these gradient-driven migratory responses.

3.4 Differential binding of CXCL12 isoforms to matrix components

To better understand the ligand-matrix interactions, we assessed the relative binding of CXCL12-isoforms to various matrix components. We used Gaussia luciferase activity to measure the relative binding of secreted GL or CXCL12-α and CXCL12-β fused to GL to wells coated with collagen I alone, collagen I-matrigel mixture, and matrigel alone (Figure 5A). Note that this measurement was conducted on the hydrogel-coated 2D well surface in contrast to other measurements in 3D hydrogel system. Collagen by itself provided no differential binding across all secreted factors (Figure 5B). However supplemented matrigel in collagen I caused a rankorder increase in binding, in which GL < CXCL12-α < CXCL12-β. Differential binding was statistically higher with matrigel by itself as compared to collagen only and the collagen-matrigel mixture. These findings suggest that matrigel increases capacity for CXCL12 binding and indicate that ligand-matrix interactions can be a dominant factor in shaping chemotactic gradients of CXCL12 isoforms. Furthermore, this suggests that matrix-binding may provide a mechanism by which migration of CXCR4 cells in response to different CXCL12-isoforms may be altered.

Figure 5.

Binding assay of CXCL12 isoforms in different gel matrices. (A) Representative heat map of flux plotted on the same log. (B) Fold change of flux relative to tissue culture plastic (TCP) for CXCL12-α, CXCL12-β, and GL treated with collagen, collagen-matrigel, and matrigel. Values are plotted from one of three representative experiments as mean ± S.E.M. for quadruplicate wells. Statistical demarcations above a bar indicate pairwise comparison to all others in that group. The bar indicates individual pairwise differences (*p <0.05, **p <0.01, and ****p <0.0001).

3.5 Gradients of CXCL12 isoforms affect migration rate

To further explore the hypothesis that matrix-binding effects generate migration-inducing gradients, we compared the migration behavior in response to GL, CXCL12-α and CXCL12-β in the presence of matrigel (Figure 6A). We observed significant differences in migration associated with these secreted factors, in which the migration rate order is GL < CXCL12-α < CXCL12-β. These results are consistent with the results from the binding assay of these molecules to matrigel presented in section 2.8 and 3.4. Simulations of gradient formation in the diffusive matrix-binding model revealed distinct gradient dynamics for CXCL12 isoforms, based on the secretion rate and binding kinetics of the isoforms (Figure S5). These simulations indicate that CXCR4-α gradients establish out from the source more rapidly than CXCR4-α gradients (Figure 6B and C). Hence, in a relatively short-term experiment, migratory cells are exposed to chemotaxis-inducing gradients for different time periods, which likely explains the differences in migration observed between α and β isoforms. While there may also be other effects of adding matrigel, such as altered gel morphology or additional macromolecular crowding effects, taken together, our findings strongly suggest that our model system is able to adequately explore the parameters involved in matrix-sequestering gradient generation effects.

Figure 6.

Migratory comparison of CXCL12 isoforms in the presence of matrigel. (A) Migration assay of CXCR4-CXCL12 (GL, CXCL12-α, and CXCL12-β) at day1 in the presence of matrigel with 250 µm spacing (*p <0.05, **p <0.01, and ***p < 0.001). GL^# represents the replication of the data in Figure 3C (+ matrigel / − CXCL12-β). (B–C) Computational simulation of unbound CXCL12-β and CXCL12-α gradient formation as a function of distance and time. Dotted line represents the spacing using between the patterned gels.

4. Discussion

We developed a novel patterning technique that exploits surface heterogeneity and cell-repellent properties to precisely pattern hydrogels for studies of gradient formation and cell migration. This system presents two core advantages over existing technologies for patterning cells: 1) broad applicability to a variety of ECM compositions and cell types; and 2) facile patterning of multiple cell types in defined geometric locations. We leverage our ability to pattern cell-laden hydrogels of varying composition to determine that formation of gradients of CXCL12 isoforms critically depends on ECM molecules for migration of CXCR4+ cells.

Our technology for patterning hydrogels relies on a simple surface patterning technology that biases cells and ligands towards interactions with the hydrogel matrix rather than the rigid substrate. By controlling relative surface energies between oxidized and native PAA-functionalized PDMS, we create geometrically-defined hydrogel patterns while minimizing cell interactions with the solid substrate. This patterning technique is simpler than channel-based or laser lithography-based hydrogel patterning systems and provides more robust patterns than liquid-liquid patterning techniques, such as aqueous two-phase systems [32, 38]. Simple but robust hydrogel patterning with our system decreases the barrier for adoption of 3D assays and improves their flexibility for researchers. Moreover, given the biased interaction among patterned cells, their secreted factors, and the hydrogel ECM, our system facilitates studies of the role of matrix interactions in physiological settings.

Dysregulated composition and architecture of the extracellular matrix is a fundamental hallmark of cancer. Perturbations in the extracellular matrix increase stiffness of the environment, facilitate invasion of cancer cells along collagen fibrils, and enhance tumor growth [39]. Less recognized is that changes in ECM content in tumors augments the capacity of the microenvironment to sequester growth factors in a way that increases local concentrations and establishes chemotactic gradients in pancreatic [40] and breast cancers [41, 42]. For example, heparan sulfate proteoglycans in tumors binds CXCL12 and other soluble chemokines to create a chemotactic gradient that directs cell migration [43, 44]. Our device enables facile control of extracellular matrix environments to study how interactions between chemotactic molecules and extracellular matrix molecules regulates magnitude of chemokine gradients and resultant cell migration.

As a prime example of the interplay among migrating cells, cells secreting chemokines, and different ECM, we investigated migration of CXCR4+ cells towards gradients of different CXCL12-isoforms. We modulated the interaction between the CXCL12 and the ECM by incorporating matrigel and expressing CXCL12 isoforms with different affinities for matrigel. CXCL12-β, which has a higher affinity for matrigel than the α-isoform (Figure 5), drove substantially higher migration of CXCR4+ cells. Our 3D hydrogel system revealed another nuance of the source-sink model where CXCL12 gradient formation and sensing does not require CXCL12 scavenging by receptor CXCR7 as previously reported [4, 5, 18, 45]. We found that for some CXCL12-isoforms, binding to the ECM may effectively form a gradient instead of CXCR7 ligand scavenging cells. Our system demonstrates how chemotactic phenotypes evolve differently among cell types in hydrogels and may better recreate in vivo physiology.

To further investigate effects of different ligand-matrix interactions on gradient dynamics, we developed a simple computational model of ligand diffusion away from cells secreting CXCL12 isoforms and binding to the ECM. Owing to precisely patterned hydrogel matrix in 3D and the large capacity for CXCL12 binding in the ECM, our computation and experimental model indicates that ligand-matrix interactions replace the cell-scavenging sink required in the previous 2D assay. It should be noted that we cannot decouple the individual contribution of bound and free CXCL12 gradients, which is a limitation of this study. We assume that the combination of bound and unbound gradients of CXCL12 drive CXCR4+ migration (haptotaxis and chemotaxis, respectively). In the future work, bound gradients of CXCL12 can be generated in the hydrogel matrix to selectively assess CXCL12-dependent haptotaxis of CXCR4+ cells. This can be achieved through the use of 3D ECM engineering techniques to tether gradients of molecules to the matrix [46–49].

Our system’s flexibility extends to many different biological questions of how multicellular interactions in 3D environments determine movement and morphogenesis, such as in angiogenesis [50, 51], metastasis [52], and other processes [53] that require multiple patterns and cell populations. Here we use our system to analyze a tumor microenvironment to allow cell-autonomous gradient formation and sensing of chemotactic molecules. Our results support the idea that secreted signaling molecules interact with the ECM and direct cancer cell migration. We also note that our ECM patterning technique is amenable to multiplexing for studies of interactions between more than two cell-type or matrix combinations (Figure S6).

5. Conclusions

Patterns on a non-adhesive surface can serve as a template to create patterns of cell-laden hydrogels with different gel compositions, cell types and soluble signals present. In this work, spatially patterned hydrogel matrices revealed distinct chemotactic behavior of breast cancer cells in response to production of the alpha and beta isoforms of the soluble signaling factor CXCL12. Our computational models indicate that a steep gradient of CXCL12 forms as a result of ligand-matrix binding interactions in reconstituted ECM, and predicts a lag-time in gradient advancement between the CXCL12 isoforms. Our system establishes that autologous gradient formation arises as a result of this interaction, without the presence of cells that scavenge CXCL12. More broadly, our approach presents a simple alternative to generating microstructured biomaterials for advanced migratory and morphogenetic studies that can be easily adopted in a variety of wet-lab environments.