Migration dynamics of ovarian epithelial cells on micro-fabricated image-based models of normal and malignant stroma

Abstract

A profound remodeling of the collagen in the extracellular matrix (ECM) occurs in human ovarian cancer but it is unknown how this affects migration dynamics and ultimately tumor growth. Here, we investigate influence of collagen morphology on ovarian cell migration through the use of second harmonic generation (SHG) image-based models of ovarian tumors. The scaffolds are fabricated by multiphoton excited (MPE) polymerization, where the process is akin to 3D printing except it achieves much greater resolution (~0.5 microns) and utilizes collagen and collagen analogs. We used this technique to create scaffolds with complex 3D submicron features representing the collagen fiber morphology in normal stroma, high risk stroma, benign tumors, and high grade ovarian tumors. We found the highly aligned malignant stromal structure promoted enhanced motility and also increased cell and f-actin alignment relative to the other tissues. However, using models based on fiber crimping characteristics, we found cells seeded on linear fibers based on normal stromal models yielded the highest degree of alignment but least motility. These results show that both the fiber properties themselves and as well as their overall alignment govern the resulting migration dynamics. These models cannot be synthesized by other conventional fabrication methods and we suggest the MPE image-based fabrication method will enable a variety of studies in cancer biology.

Graphical Abstract

1. Introduction

Ovarian cancer is the leading cause of deaths from gynecological cancers, where the 5 year survival rate of high grade disease is about 25%. It is generally accepted that metastasis of ovarian cancer often occurs from exfoliation from the ovarian surface into ascites, and then by reattachment in the intraperitoneal cavity (IP), interacting with mesothelial cells, and then by invasion into the stroma.[1] These dynamics are generally regulated by cell binding integrins (e.g. β₁), matrix degrading proteinases (MMPs), and kinases (for example, focal adhesion kinase).[2–5] Better insight into these processes could lead to targeted therapeutics, however, treatments directed at molecular targets that govern migration have been investigated but have not yet led to positive clinical outcomes.[6–9] Single genes have not typically been associated with enhanced migration and it has been speculated that different species act in concert.[4, 10–12] In addition to gene expression, the resulting behavior will also likely depend on the on the local 3D composition and architecture in the tumor microenvironment (TME)[10]. Specifically, collagen is altered in the extracellular matrix (ECM) and is profoundly remodeled in human ovarian cancer.[13–17]

We have catalogued the collagen alterations in a series of human ovarian tumors through the use of the collagen specific Second Harmonic Generation (SHG) microscopy. In these efforts we have probed the supramolecular, fibril, and fiber structures through several SHG metrics and shown that quantitative discrimination can be achieved between normal structure, stroma from high risk (BRCA mutations) patients, benign tumors, low grade tumors and high grade serous tumors.[13, 15, 18, 19] Most importantly for this current study, we have shown these tissue classes can be quantitatively differentiated using machine learning techniques based on the respective morphology visualized in the SHG microscope.[15, 18] The most striking aspect of the morphological differences between the tissues is the increased fiber alignment and repeating fiber shapes in the high grade tissues (to be shown in Section 3.1).

This compelling data leads to the significant question of the biological significance of the remodeled collagen in the development and progression of ovarian cancer, especially as high grade disease can metastasize while tumors are still below the detection of conventional clinical modalities,[20–24] Migration dynamics are particularly important as the primary form of metastasis in ovarian cancer is exfoliation from the surface of the ovary to the peritoneum, as opposed to angiogenesis typical of many other cancers. Thus the motility on the surface, as well as stress fiber and focal adhesion assembly are expected to be critical players in the metastasis.

Such studies are best addressed through the use of in vitro biomimetic models where cellular and ECM features can be controlled and afford hypothesis testing of their relative roles on function. This is especially true for ovarian cancer as animal models remain difficult and are largely limited to xenografts.[25] Considerable insight into ovarian cell migration dynamics has been obtained in standard 2D culture where several operative pathways in metastasis involving proteases and kinases have been obtained.[26–29] Further details (e.g. operative integrins) on adhesion and migration have been achieved using flow chambers, which count numbers of migrating and invading cells across a membrane.[30–32] Microfluidic models have recently been used to model extravasation.[33–35]

While important tools, these and other standard methods do not recapitulate the fibrillar structure of collagen in the TME. For example, lithography has been successfully used to demonstrate that substrate topography (e.g. microgrooves or other patterns) can lead to cell contact guidance, resulting in alignment of the cell body and focal adhesions [36–39] However, the use of masks, masters and stamps limits the replication of complex biomimetic 3D structures with spatially varying morphologies and concentration. Moreover, 2D culture does not also reflect the migration dynamics in a 3D environment of the native ECM. For example, Yamada and coworkers elegantly showed that even simple 1D lithographic patterns better emulated 3D behavior than standard 2D culture. [40] More powerful models would reproduce the complex collagen morphology in normal and malignant tumors.

To address these limitations, we developed a fabrication scheme that uses multiphoton excited (MPE) photochemistry [41–45]. The method is akin to 3D printing but produces smaller feature sizes (submicron) and can utilize collagen and collagen analogs to reproduce the complex collagen architecture of the ovarian ECM.[46] We have previously shown that simple MPE fabricated ECM patterns (e.g. collagen, laminin, and fibronectin), and concentration gradients thereof, govern cell migration speed and directionality and cytoskeletal alignment of several cell types.[47–49] In this report, we fabricated scaffolds directly based on SHG images of human ovarian tissues characterized as normal, high risk, benign tumors, and high grade tumors. These image-based scaffolds are then used to study migration dynamics of an ovarian cell line to better understand the specific effects of the varying stroma morphology. We further exploited this approach to create “synthetic patterns” based on characteristic fiber features (e.g. periodicity) in the image to decouple the influence of collagen fiber alignment from fiber morphology. The combination of image-based and synthetic models affords the hypothesis testing of all these factors on cell migration dynamics.

2. Methods and materials

2.1. Microscope and photochemistry

The fabrication microscope system has been described in detail previously [50] and is only described briefly here. A mode-locked titanium sapphire femtosecond laser (Mira; Coherent, Santa Barbara, California) is coupled to an upright microscope stand (Axioskop 2, Zeiss, Thornwood, NY) and scanning is performed by a combination of galvos (Cambridge Technologies, Bedford, MA) and a motorized stage (x-y-z, Ludl Electronic Products Ltd, Hawthorne, NY). A wavelength of 780 nm is used for two-photon excited fabrication and the laser power is controlled through a 10 kHz electro-optic modulator (EOM, Conoptics, Danbury, CT). In addition, the laser is rapidly shuttered by a second higher speed EOM (maximum 100 MHz, Conoptics). The “open” fraction in each pixel is mapped to the gray scale level (0–8 bits) of the corresponding pixel of the original SHG image. Thus, increased laser exposure linearly corresponds in increased concentration [50]. The minimum feature sizes for crosslinked protein agree with the theoretical two-photon excited resolution. For instance, using 0.75 NA and 780 nm two-photon excitation, the respective lateral and axial resolutions are ~600 nm and 1.8 microns [46].

Sodium 4-[2-(4-morpholino)benzoyl-2 dimethylamino]butylbenzenesulfonate (MBS) was chosen as the photoinitiator because it is non-cytotoxic and water-soluble, with comparable efficiency to vinyl photoinitiators soluble in organic solvents. MBS was synthesized in house, using published protocols [51]. Two-photon excitation of MBS at 780 nm drives a photochemical reaction that makes benzoyl and alpha amino alkyl radicals. At the focal volume, reactive radicals interact with the surrounding proteins targeting oxidizable amino acid residues. The active radicals add to the monomer, inducing a chain growth crosslinking and eventually termination [52].

2.2. Sample preparation

The scaffolds are created from a mixture of 75% gelatin methacrylate (GelMA)[53] and 25% rat tail collagen I (v/v). GelMA was prepared from well-established protocols without further modification [53]. We chose this combination for several reasons: (i) due to pH issues and resulting solubility, 100% collagen is not compatible with most available photochemistries, (ii) while the single stranded GelMA presents RGD cues, GFOGER is the relevant binding site presented in the triple helical collagen[54, 55] and its incorporation even at low levels increases the biomimicry, (iii) GelMA is biocompatible is widely used in collagen scaffolds [56–58], and (iv) GelMA has advantages over gelatin in terms of ease of photochemistry and the ability to form stiffer structures.

The slides were prepared first with a rubber hybridization chamber secured to a silanized microscope slide.[59] A monolayer of 30mg/mL bovine serum albumin (BSA) formed on the slide before the final collagen solution was applied. The collagen solution used was a 1:3 ratio (25μL Collagen and 75μL Gel MA). The Gel MA and collagen was kept strictly below 40°C. Scaffolds were kept in phosphate buffer solution (PBS) until cell seeding.

2.3. Cell seeding and time-lapse imaging

Immortalized Ovarian Surface Epithelium cells (IOSE; from Dr. Susan Huang, MD Anderson Cancer Center) [60] were cultured at 37°C and 5% CO2 in DMEM/F12 medium base (LifeTechnologies 11330) supplemented with 10% FBS (LifeTechnologies 10082).

Prior to cell seeding, the fabricated scaffolds were sterilized with 1X PBS containing 100 U/mL Penicillin-Streptomycin (Invitrogen 15140–122). Cells were seeded at a density of 50K cell/mL and incubated overnight. Time-lapse imaging of migration was then performed by phase contrast imaging (10× 0.25NA objective; Nikon Ti-Eclipse microscope with Pathology Devices, Inc. - LiveCellTM incubator system). Phase-contrast images of each seeded scaffold were collected at 30 minutes intervals over 72 hours. Cells became too densely populated at longer times to isolate the cell-matrix interactions. At least 60–80 cells were used for each cell/scaffold combination.

2.4. Cell tracking

Cell tracking was performed with Imaris (v7.6.5 - Bitplane AG). For each fabricated scaffold, at least 20 cells were tracked for statistical significance. Data representing tracked cells were exported to an Excel document for further statistical analysis using MATLAB (Mathworks – Natick MA). The tracked cell data was analyzed with self-written code in MATLAB. The main features of this code include: calculations of cell position, the instantaneous speeds, direction of the migration and mean square displacement (MSD) measurements. Motility coefficients (μ) were isolated by applying non-linear least square regression modeling of the MSD measurement to the following equation [61]:

$$\langle d^2\left(t\right)\rangle =2n_d\mu \left[t-P\left(1-e^{-\frac{t}{P}}\right)\right]$$ (1)

2.5. F-actin and focal adhesion staining

Ovarian cells were grown on fabricated structures between 16 to 24 hours prior to staining. These cells were fixed in with 4% paraformaldahyde in PBS for 15 minutes. Following two washes with 1x PBS, the cells were permeabilized with 0.3% Triton X-100 for 10 minutes and stained with Texas Red conjugated phalloidin for 30 minutes. At least 8–12 cells were analyzed for each cell/scaffold combination. CurveAlign was used to quantify the angular distribution of f-Actin fibers for cells in a given pattern as well as the overall collagen alignment from the SHG images [62]. To stain for focal adhesions, the cells were incubated with an anti-vinculin primary antibody (at 1:200 dilution; EMD Millipore) overnight at 4°C followed by incubation with a fluorescent secondary antibody, IgG Alexa488 (1:1000 dilution; Invitrogen) for 1 hour at room temperature. The number of focal adhesions per cell and integrated areas were determined in ImageJ.

2.5. Statistical analyses

The image stacks from each time-lapse were analyzed with Imaris software (Bitplane, Switzerland) to determine instantaneous velocities and motility attributes. The stacks were first processed with ImageJ using the Enhance Local Contrast (CLAHE) macro before importing to Imaris software. The cells were manually tracked to plot cell migration distance, speed, and directionality. Cell shape characteristics were determined with ImageJ software. Analysis of this data was run via Origin 2015 (OriginLab) to determine statistical significance with ANOVA and two-sample t-test analysis. Origin 2015 was also used to perform ANOVA and two-sample t-tests for the cell shape analyses.

Watson’s U² tests were performed on f-Actin and collagen fiber distributions using Oriana (Kovach Computing Services) to calculate directional statistics of the distribution and mean direction. Pearson correlation coefficients between these distributions were also calculated to measure correlation of the stress fibers and the collagen fibers in the stromal models.

3. Results

3.1. SHG image-based blueprints for fabrication

Previously acquired SHG images of stroma sectioned from the cortex below the surface epithelium from four different types of ovarian tissues (normal, high risk, benign tumors and high grade malignant tumors (stage III or IV) were used as blueprints to fabricate image-based scaffolds. For statistical relevance, images from four patients of each class were chosen in this study. We have previously shown by texture analysis that the collagen fiber topography in each tissue type is different across these classes, where we classified over one hundred single optical sections in each respective group. [18]. The images chosen for the blueprints were randomly picked from those that were classified correctly.

The top row of Figure 1 (Fig. 1A) shows representative SHG images of collagen topographies across the four classes. These images were taken at 40x. 0.8 NA with a resolution of about 600 nm and 2.5 microns in the lateral and axial dimensions, respectively with a field size of about 200 ×200 microns. The fabricated structures were created at these same conditions. Image processing techniques available in ImageJ, such as filters (e.g. Hessian filter, and threshold) and enhancements (e.g. tubeness) were used to discretize fiber structures from the SHG images and generate the templates.

Figure 1.

Ovarian stromal collagen and corresponding fabricated structures. (A) SHG optical sections of collagen from the four categories of ovarian tissues, where the field of view is 200 × 200 μm (B) 3D reconstructed multiphoton excited fluorescence images of the fabricated structures from Figure 1 A. Each pattern is 200 × 200 μm in size with ~10 μm in height. Scale bar=50 microns. (C) Antibody staining for Col for GelMA and GelMA/Col I, showing the incorporation of collagen into the fabricated structure.

Figure 1B shows the 3D reconstructed fluorescence images of the fabricated structures models, representing the respective ovarian stroma from Figure 1A. For imaging, the scaffolds were stained with rhodamine B and imaged via two-photon excited fluorescence. Fidelity of the fabricated structures was 95% or higher to their respective template, where this is obtained by co-localization of both the spatial pixel overlap and the respective gray scale intensities between the model of the image data and the fabricated structure [50]. We define fidelity by the correlation between the model used for fabrication (rather than the raw SHG image) and the fabricated construct. This is because there is often some overlap of fibers in both the lateral and axial planes, in which case, we need to discretize the fibers to aid the fabrication process. We used immunofluorescence to confirm that Col I molecules were incorporated in the GelMA + Col I scaffolds. Here GelMA and GelMA+collagen structures were treated with a Col I primary antibody (Abcam) and secondary IGG-Alexa488 antibody. The relative fluorescence in shown in Fig 1C, which shows a significant increase in fluorescence intensity for the GelMA/collagen structure, indicative of specific binding to collagen over the non-specific GelMA alone.

3.2. Development of synthetic models based on fiber attributes

Most reports in cancer have focused on collagen alignment [63, 64] however, the role of fiber morphology itself has not been as well addressed. The MPE fabrication technique affords the investigation of both aspects. To this end, based on the SHG image data, collagen fibers were modeled as sine waves (Fig. 2A), where we define the peak-to-peak distance of the collagen fibers as the “spatial frequency” (F) and the height as the “spatial amplitude” (A), (red and blue arrows respectively). These fiber aspects were determined for each tissue type using the curvelet transform [62].

Figure 2.

Modelling collagen fibers as sinusoidal waves. (A) Definition of fiber properties as amplitude and frequency. (B) 3D plot of overall collagen alignment, spatial amplitude and frequency from normal (blue), benign (green), high risk (magenta) and high grade (red) tissues. (C) Graph showing only distribution of spatial amplitude and frequency measurements. (D) Synthetic patterns matching representative chosen points (plot C) with a single direction of alignment. (E) fluorescence images of the fabricated models.

Figure 2B shows a 3D plot of the fiber alignment, spatial amplitude and frequency across the four classes of tissues. Normal ovarian tissue samples are characterized by nearly straight collagen fibers, with low frequency and overall low alignment. In contrast, high grade cancer stroma images present overall high alignment with a characteristic high fiber frequency. This is also observed in the benign tumor images, but these fibers have higher amplitude and overall lower alignment. High risk tissue images also have low collagen fiber alignment, where the fiber morphology is characterized by high frequency and low amplitude. Figure 2C reduces the parameter space to the fiber morphology (amplitude and frequency), showing that tissue types can also be separable by these two characteristics. We define scaffolds comprised of repeating aligned fibers based on these attributes as synthetic models.

Figure 2D shows images of the computer design for each of the four synthetic patterns: (i) Synthetic normal (sNormal) represented by relatively straight fibers (low amplitude and frequency); (ii) Synthetic benign (sBenign: A=10μm, F= 20μm) represented by fibers with large amplitude and moderate frequency; (iii) Synthetic high grade (sHighGrade: A=4μm, F=20μm) with overall low amplitude and moderate frequency; and (iv) Synthetic High Risk (sHighRisk: A=4μm, F=10μm) modeled as fibers with overall low amplitude and high frequency. Figure 2E shows the fluorescence image of the structures.

3.3. Migration dynamics using image-based and synthetic models

We investigated how the highly different collagen topographic patterns in the four tissue classes influence the IOSE cell migration dynamics, using both image-based and synthetic patterns. The scaffolds are comprised of 3×3 repeats of the same 200× 200 micron pattern (10 microns in thickness), for an overall size of 600 × 600 microns to yield sufficient area to simultaneously monitor multiple cells migrating on the surface of the scaffold. We have previously shown that the migration on fabricated substrates arises from the ECM morphology and not purely form the serum in the media. Specifically cells on patterned BSA exhibit only random migration.[49, 65]

(a). Cell migration on image-based scaffolds

Figure 3 shows representative phase contrast images of IOSE cells on high grade image-based (Fig. 3A) and high grade synthetic (Fig. 3B) scaffolds. Migration on the scaffolds is measured for 72 hrs. As representative data, Fig. 4A and 4B show the migration trajectories for 20 IOSE cells on a normal and high grade stroma model, respectively. We observed significance difference in migration trajectories from localized paths (over ~100 μm) on the normal stroma to longer paths (~300 μm range) on high grade stroma. These results indicate highly aligned fibers promote cell migration to a larger extent than a more random structure.

Figure 3.

Phase-contrast images of IOSE cells on high grade cancer image-based (A) and high grades synthetic (B) scaffolds. Scale bar = 100 μm.

Figure 4.

Migration dynamics of IOSE cells on image-based stromal models. (A&B) Trajectories of IOSE migration on normal and high grade stroma, respectively. (C) Instantaneous cell migration speeds. (D) Motility coefficients (*p < 0.05; **p < 0.005).

To quantify these data, we determined the instantaneous velocity and motility coefficients on each model. Figure 4C shows the average velocity values for IOSE cells on the four scaffolds. First, we note that that instantaneous speed did not change significantly for each of the cell lines on the four scaffold types.

We next examined the role of motility, i.e. the ability to migrate in one direction before changing direction, where this is determined by measuring the mean square displacement (MSD) in conjunction with Eq. 1 (see Methods section 2.4). The averaged motility results for the IOSE cells on the four structures are shown Fig. 4D. First, we can compare the cell behavior on the different morphologies. When comparing the role of the scaffolds we find the lowest motilities on the normal stromal model, which has a fairly random alignment. In contrast, the more aligned fibers of the benign and high grade tumor models resulted in higher motility. In sum, these measurements show that fiber architecture is an important factor in the resulting migration.

(b). Cell migration on synthetic models

Similarly, we analyzed the migration dynamics of IOSE cells on the synthetic patterns (Figure 5, where 5A and 5B show representative trajectories for cells on a normal and high grade cancer synthetic model, respectively, and 5C and 5D are the resulting respective instantaneous velocities and motility coefficients. The results are mostly analogous to those on image-based patterns. We find that both the velocity and motility for the IOSE cells are higher on a serous cancer model relative to the normal stroma, indicating that the frequency and amplitude of the collagen fiber crimping pattern enhances migration of these large, nonpolar cells. Therefore, these results reinforce the hypothesis that both collagen alignment and fiber morphology increase cell migration velocity and motility.

Figure 5.

Migration dynamics of IOSE cells on synthetic stromal models. (A&B) Trajectories of IOSE migration on synthetic normal and synthetic high grade, respectively. (C) Instantaneous cell migration speed. (D) Motility coefficients. (*p < 0.05; **p < 0.005).

3.4. Collagen topography drives cell morphology

We next examined cell shape of IOSE cells on both the image-based and synthetic structures to determine how the overall architecture and fiber characteristics affect the resulting phenotype. As an example, Scanning Electron Microscope (SEM) images of IOSE cells seeded on the image-based and synthetic scaffolds were collected and several are shown in Fig. 6. We first note that the cells on synthetic benign and high risk structures resulted in a more rounded cell shape with many short protrusions in multiple directions. In contrast, IOSE cells on synthetic normal and high grade structures were more aligned with the fiber axis, and the resulting protrusions are sparser but longer and oriented along the direction of the fabricated structures. These observations suggest that even for the same cell type, higher frequency/amplitude patterns result in less cell alignment relative to the fibers. Similar behavior was observed for the image-based structures, where the higher frequency patterns resulted in protrusions in multiple directions.

Figure 6.

Scanning Electron Microscopy (SEM) images of IOSE cells on selected image-based and synthetic patterns. Scale bar = 20 μm.

To expand these observations, we quantified changes in cell morphology on all the fabricated structures (image-based and synthetic) using cell circularity measurements. This metric is given by $\frac{4\pi A}{2p^2}$, where A and p are the area and perimeter, respectively. Higher and lower values correspond to greater circular (less aligned) or elliptical shapes (more aligned), respectively. Figure 7A gives the circularity for IOSE cells on image-based structures. For the highly aligned high grade model (Fig. 2) the circularity values are significantly lower than on the normal and benign, patterns, indicating that cells are more elongated and aligned along fibers (as also shown in Fig. 6). The cells on the normal stromal model have the highest circularity, which is expected as this architecture has the lowest alignment (Fig 2).

Figure 7.

Cell circularity for IOSE cells on (A) image-based and (B) synthetic patterns (*p < 0.05).

Similarly, we quantified the IOSE circularity on synthetic patterns (Fig. 7 B) and observed that overall, circularity is significantly lower for cells on synthetic high grade and normal patterns compared to the synthetic benign. This likely because the high grade and normal models have lower frequencies and amplitude, i.e. are more linear, and the cells efficiently align along the fiber axis. In contrast, the higher frequency, wavier patterns do not constrain the cells as strongly along the main axis.

3.5. F-actin alignment in response to stroma architecture

We determined the spatial distribution of f-Actin fibers of cells on image-based and synthetic stromal architectures as means of understanding cytoskeleton alignment relative to the collagen morphology. Representative images of fluorescence images (phalloidin) for IOSE cells on image-based and synthetic patterns are shown in Figure 8. The top row of Figure 9 shows representative SHG images used as the image templates to determine the correlation of the radial distribution of stromal and f-Actin fibers. The next row is the radial distribution of the stromal fibers, followed by the radial distribution of the stress fibers. We first point out that cells on the normal and benign tumor stromal patterns resulted in a broader distribution of f-Actin fibers. In contrast, high risk and high grade patterns resulted in higher f-actin alignment along the orientation of the stromal fibers. These relationships are quantified by the Pearson correlation coefficients between the radial distributions of the stromal and f-Actin fibers in the bottom row in Fig 9.

Figure 8.

2-photon excited Immunofluorescence of f-Actin filaments of IOSE cells stained with Texas Red phalloidin on image-based (top row) and synthetic patterns (bottom row). Scale bar = 20 μm.

Figure 9.

Top row is a representative SHG image with the radial distribution of collagen fibers in the next row. The bottom row is the angular distributions of f-Actin for IOSE cells on the four image-based structures. (A) Normal. (B) High risk. (C) Benign. (D) High grade. **p < 0.001 - Watson’s U² test. Scale bar= 50 microns. The bottom row gives Pearson correlation coefficients between f-Actin and stromal collagen fiber distributions for image-based patterns.

IOSE cells on normal and benign structures resulted in a low degree of correlation with the stromal fibers (0.37 and 0.32, respectively), whereas those on high risk and high grade scaffolds resulted in highly correlated alignments of 0.83 and 0.97, respectively. Therefore, the highly aligned stromal architectures promote stress fiber alignment. These results are consistent with the migration and cell shape data, where IOSE cells displayed higher motility on the high grade models, as well as the lowest values of circularity.

These results using image-based models present coupled cues of stroma alignment and individual fiber morphology in the f-actin alignment. In order to decouple these factors, the same experiments were performed on synthetic patterns maintaining a single direction of alignment (Fig. 10), in analogy with the cell migration (Fig. 5) and shape (Fig. 7B) determinations.

Figure 10.

Top row is the design of the synthetic scaffold and the bottom row gives the angular distributions of f-Actin for the IOSE cells. (A) sNormal. (B) sHigh risk. (C) sBenign. (D) sHigh grade. **p < 0.001 - Watson’s U² test.

We first note that increased frequency resulted in a broader distribution of f-Actin fibers. For example, the lower frequency of normal stromal fibers is nearly linear and resulted in highly aligned stress fibers. In contrast, the higher frequency high risk and benign fiber models resulted in broader distributions. In addition, the amplitude of the sine wave also plays an important role. For example, while the high grade and benign tumor models have the same frequency, the benign architecture with higher amplitude led to a broader distribution of stress fibers, indicating that larger crimp patterns decrease f-actin alignment. In sum, small differences on the crimping pattern (frequency and amplitude) of stromal fibers can strongly influence the stress fiber alignment, where this is consistent with the cell morphology analysis, where aligned fibers led to decreased circularity.

3.6. Focal adhesion expression in response to stromal architecture

Using vinculin staining, we have performed an assessment of the relative focal adhesion expression. Representative images for IOSE cells on the image-based and synthetic models are presented in Figure 11 (A) and (D), respectively. The vinculin stain was quantified in terms of numbers of focal adhesions per cell and the corresponding integrated area. We found that the highly aligned high grade structure had the highest density and integrated area of the four image-based structures, and the least aligned normal structure had the least, where benign and high risk tissues had intermediate behavior (Fig 11 (B) and (C)). This indicates the high frequency, highly aligned fiber pattern of the high grade tumor best promotes focal adhesion expression, and is similar to the motility findings shown in Figure 4D. We further performed this analysis for IOSE cell son the synthetic patterns, (Fig 11 (E) and (F). Most of the data using the synthetic structures was not statically different, but the overall trend was that the higher frequency structures had slightly higher number of focal adhesion per cell. Taken together, the data from the image-based and synthetic models suggests that alignment has a larger role on focal adhesion expression than fiber characteristics.

Figure 11.

Anti-vinculin staining of focal adhesion expression on image-based and synthetic models. Representative two-photon excited fluorescence images of IOSE cells on image-based structures (A) and synthetic structures (D), where punctate fluorescence near the membrane is pseudo-colored in violet. Resulting analysis of numbers of focal adhesions and their integrated area for image-based structures (B) and (C) and for synthetic models (E) and (F).

4. Discussion

A better understanding of the cell interactions with the stromal microenvironment in ovarian cancer could lead to better diagnostics as well as assessments of treatment efficacy. However, there is still a lack of efficient models to understand these multi-faceted interactions between cell migration and ECM morphology [66]. While image-based MPE fabrication based cannot recapitulate the entire ECM structure, we believe it is the optimal method for reproducing the collagen morphology. For commonly used lithography and hydrogel approaches are limited in terms of materials and/or patterning. It would be advantageous to create scaffolds from collagen alone, however this not practical due to solubility issues, Still, the GelMA presents RGD binding cues to that promote migration. Moreover, we have previously shown that fibroblasts and cancer cells migrate similarly on different ECM proteins that use different integrins, where we concluded that the combination of morphology and relative concentration of ECM cues governed the response, rather than the specific integrin.[49, 65, 67] Thus, the scaffolds reported here have good biomimetic attributes. While these scaffolds are only 10 microns in height, they do possess 3D morphology. Moreover, unlike in other tissues, ovarian epithelial cells are exposed to the peritoneal cavity and do not naturally dwell in or adjacent to 3D environments. Thus, examination of migration on the surface represents some aspects of the natural behavior of these epithelial cells.

We also point out that current minimum feature sizes of about 600 nm (both SHG imaging and fabrication) will capture most but not all of the collagen fibers in the matrix. At higher resolution (e.g. 1.4 NA), we could achieve about 300 nm features, which is about the diameter of small collagen fibers. This is also approximately the width of small focal adhesion and likely would be suitable as models for most collagen fibers. However, there are optical considerations in both the SHG imaging and fabrication that makes this higher resolution difficult in practice and we therefore compromise on the resolution used here.

As the ovary can be either the primary site or first metastatic site (from the fallopian tubes) in this cancer [68], we focused on modelling the surface of the ovary to study migration dynamics. As altered migration is a hallmark in cancer,[69] we have focused the analyses on migration and migration related structural aspects (cell morphology and f-actin fiber alignment). Migration in microchannel models have shown that cancer cells utilize aligned stromal architecture for enhanced and persistent migration [70–73]. However, the role of the fiber aspects themselves, i.e. the crimping characteristics of frequency and amplitude have been less explored. The MPE fabrication approach is well matched to this task as the freeform nature affords creation of user defined spatial characteristics, allowing the decoupling of overall alignment of the fibers to the fiber shapes themselves. (Fig. 2).

A primary finding of the current study is that both increased overall fiber alignment, as well higher crimping frequency, together and separately, enhance motility. We also find that highly aligned stromal patterns promoted spatially correlated f-actin alignment. While this overall result may be somewhat anticipated, more details are found through using the synthetic patterns that explore the effects of fiber aspects directly. We find that the f-actin distribution is highly sensitive to the fiber shape, where both the crimping frequency and amplitude contribute to the resulting distribution. Moreover, small changes in geometry can result in a significant difference in stress fiber distribution. For example, the high grade and benign tumor models have similar fiber frequencies, but the latter has a larger crimp amplitude, which led to a broader radial distribution. Analogous trends were found for the cell shape on both the image-based and synthetic models and we found greater motility on aligned wavy fibers over aligned straight fibers. This may be due to greater area for cells to express focal adhesions (see Figure 11). This is consistent with cells having greater circularity (less aligned) and more f-actin protrusions for which to interact with the greater surface area of the wavy fibers. In sum, these results show the importance of the complex ECM structure in governing the migration dynamics.

Our intent in this study was to use a bottom up approach to study migration in ovarian cancer using the normal IOSE cells as a constant in which to systematically isolate the role of matrix alterations on the dynamics. However, it is also interesting to consider top-down approaches. For example, Balkwill and co-workers recently used this idea to deconstruct the stroma in an ovarian cancer tumor to identify common occurrences in ECM response to the cancer.[74] We suggest there is significant potential synergy between these overall approaches where each may inform the other for improved insight. For example, moving forward we will incorporate additional ECM and cell components and extend the studies to gene expression.

We stress that while the bottom up methodology can be applied to many tissues and diseased states, it is critical to consider the morphology of the specific tissues. For example we can draw comparisons to collagen alterations in breast cancer.[64, 75] It has been documented that changes in collagen architecture in breast cancer induce important signaling events that regulate cell function, affect acto-myosin organization and, consequently, regulate cell motility, where specifically aligned fibers enhanced the motility of cancer cells in vivo and ex vivo [76–81]. While increased alignment increases motility in both breast and ovarian cancer, normal breast stroma is characterized by wavy fibers that become straighter in cancer [75] whereas the shape in ovarian cancer is in the opposite direction. Additionally, the method can be applied to cancer cells to afford probing of the roles of both cell phenotype and matrix morphology.

5. Conclusions

Using multiphoton excited fabrication, we constructed image-based models of ovarian tissues as well as sine wave synthetic models based on fiber attributes. The technique is superior to other fabrication methods as the complex morphology of the collagen visualized by SHG microscopy can be recapitulated with high fidelity. Moreover, the approach affords hypothesis testing of alignment of fibers as well as the fiber characteristics themselves. A main finding is that both these factors govern the resulting migration dynamics. Interestingly, we find that aligned and high frequency fibers promote motility over straight fibers. Additionally, relatively small changes in fiber morphology result in significant changes in migration dynamics. We suggest the scaffolds can be used for further cell biological studies and as platforms for testing of drug efficacy.

Statement of Significance.

The extracellular matrix collagen in ovarian cancer is highly remodeled but the consequences on cell function remain unknown. It is important to understand the operative cell matrix interactions, as this could lead to better prognostics and better prediction of therapeutic efficacy. We probe migration dynamics using high resolution (~0.5 microns) multiphoton excited fabrication to synthesize scaffolds whose designs are derived directly from Second Harmonic Generation microscope images of the collagen in normal ovarian tissues as well as benign and malignant tumors. Collectively our results show the importance of the matrix morphology (fiber shape and alignment) on driving cell motility, cell shape and f-actin alignment. These collagen-based models have complex fiber morphology and cannot be created by conventional fabrication technologies.