Hydrogel Microwell Arrays Allow the Assessment of Protease-Associated Enhancement of Cancer Cell Aggregation and Survival

Current routine cell culture techniques are only poorly suited to capture the physiological complexity of tumor microenvironments, wherein tumor cell function is affected by intricate three-dimensional (3D), integrin-dependent cell-cell and cell-extracellular matrix (ECM) interactions. 3D cell cultures allow the investigation of cancer-associated proteases like kallikreins as they degrade ECM proteins and alter integrin signaling, promoting malignant cell behaviors. Here, we employed a hydrogel microwell array platform to probe using a high-throughput mode how ovarian cancer cell aggregates of defined size form and survive in response to the expression of kallikreins and treatment with paclitaxel, by performing microscopic, quantitative image, gene and protein analyses dependent on the varying microwell and aggregate sizes. Paclitaxel treatment increased aggregate formation and survival of kallikrein-expressing cancer cells and levels of integrins and integrin-related factors. Cancer cell aggregate formation was improved with increasing aggregate size, thereby reducing cell death and enhancing integrin expression upon paclitaxel treatment. Therefore, hydrogel microwell arrays are a powerful tool to screen the viability of cancer cell aggregates upon modulation of protease expression, integrin engagement and anti-cancer treatment providing a micro-scaled yet high-throughput technique to assess malignant progression and drug-resistance.

scanning microscopy as well as quantitative image, gene and protein analyses dependent on varying microwell and aggregate size.

Time-Lapse Microscopy.
Time-lapse microscopy of hydrogel microwell arrays of varying size was performed to live image cell aggregation and survival as reported previously [36]. Samples were imaged 24 h after seeding using an inverted microscope (Zeiss Axio Observer.Z1 and Zeiss Axiovert) equipped with a motorized scanning stage under sterile humidified atmosphere at 37 °C/5% (v/v) CO 2 over 96 h, with images taken every 6 h using a 10× air objective (Figure 1(B); Supplementary file). The resulting phase contrast images were then automatically compiled into a stack using Metamorph (Molecular Devices, Sunnyvale, CA, USA). To identify dead cells, propidium iodide (PI; 1:1,000; Invitrogen) was added to the media and fluorescently imaged at the end of each experiment. Cell aggregates were grown within different sized microwells and visualized at up to 20 different positions per condition. Confocal microscopy of four cell aggregates grown over 96 h within microwells (100 × 50 µm) ± paclitaxel treatment (100 nM); nuclei stained blue with DAPI; dead cells stained red with PI (bottom panel). Scale bars, 30 µm. (B) Time-lapse microscopy of cell aggregation ± paclitaxel treatment was performed over 96 h within microwells. Scale bars, 100 µm. (C) Confocal microscopy of cell aggregates grown over 96 h within microwells ± paclitaxel treatment and 3D reconstructions using Imaris; F-actin filaments stained green with Alexa488-conjugated phalloidin; nuclei stained blue with DAPI. Scale bars, 10 µm. (D) Confocal microscopy of the morphological marker N-cadherin in cell aggregates ± paclitaxel treatment; N-cadherin stained red using a respective primary and secondary Alexa555-conjugated IgG; nuclei stained blue with DAPI. Scale bars, 10 µm.

Calculation of Cell Aggregate Area and Number.
For cell aggregate number and area calculations, the integrated morphometry analysis tool in Metamorph or ImageJ ( [41]) was applied to trace the aggregation number and area using either stacked bright field images or fluorescently labelled aggregates. Maximal projections using separate channels of bright field or fluorescent images were arithmetic processed, set to auto-threshold and gray levels binarized. An integrated morphometry analysis was performed to graphically identify the aggregate area. The aggregate number per microwell indicates the ratio of the number of aggregates per microwell to the number of microwells (seeding efficacy of 75-82%) counted per condition. Averages and standard errors were calculated using Excel (Microsoft, Redmond, WA, USA). For each experiment, 20 different positions per condition containing 60-960 aggregates were analyzed. For calculation of cell aggregation after paclitaxel treatment, only intact, non-lyzed cells (without the appearance of apoptotic bodies) were taken into account. Data are expressed as -relative aggregation (%)‖, describing the ratio of the number of aggregates to the number of microwells analyzed per condition, and -relative cell death (%)‖, referring to the ratio of the number of aggregates containing death cells (as indicated by PI staining) to the number of viable aggregates (no PI staining).

Hydrogel Microwell Arrays Allow the Aggregation of Ovarian Cancer Cells
We sought to apply high-throughput assays-to our knowledge for the first time-to allow defined aggregation of ovarian cancer cells and monitored this cellular process by confocal laser scanning microscopy ( Figure 1(A,C,D)) and live cell microscopy over 96 h (Figure 1(B)) to establish their suitability as a drug screening tool using the clinically applied anti-cancer drug paclitaxel.
Cancer cells cultured as single cell suspension (1 × 10 4 cells/mL) did not form aggregates on top of 3D cultures within microwells, and underwent only one cell division within the first 36 h after seeding (data not shown). Microwells coated with laminin or type I collagen did not increase the cell survival rates of single cell suspensions over 96 h (data not shown). As ovarian cancer cells aggregate in the tumor fluid (ascites) accumulated within the abdominal cavity of patients with advanced disease [10], we increased the number from single cancer cells per microwell (100 × 50 µm) to 5 × 10 4 cells/mL. Time-lapse and confocal laser scanning microscopy revealed compact aggregate formation after 96 h of 3D culture with negligible cell death as indicated by minor propidium iodide (PI) staining. Upon paclitaxel treatment (100 nM), cell aggregation was dramatically reduced and scattered and cell death increased as indicated by a positive PI staining ( Figure 1(A,B)). 3D reconstructions and immunostaining of the morphological marker N-cadherin confirmed compact aggregation without treatment and scattered aggregation upon paclitaxel treatment with the appearance of apoptotic nuclei (Figure 1(C,D)). These results suggest that hydrogel microwell arrays allow cancer cell aggregation.
The multi-cellular aggregate population in human ovarian tumor fluid (ascites) is thought to be a critical source for intra-abdominal metastases, and thereby, represents a key target for anti-metastatic interventions. Currently, most chemotherapies are ineffective in preventing aggregate dissemination, and the biological mechanisms leading to their formation remain poorly understood [9,10,43]. To improve our understanding of ovarian cancer biology, controlled in vitro models are needed to accurately mimic the in vivo conditions seen in patients [44]. Ill-advisedly, the terms aggregate and spheroid are inconsistently used throughout the literature, and yet, this definition is critical to the rationale of experimental 3D model approaches. The term aggregate is primarily but not always used to describe and eventually to discriminate loose packages of cells from compact spherical cultures. Aggregates with a size smaller than 150 µm may exhibit cell-cell and cell-matrix interactions. Spheroids comprise a defined cell mass of uniform geometry and physiological gradients at diameters ranging from 200-500 µm that can be manipulated and suited for large scale approaches in preclinical drug testing routines [45]. Both aggregate and spheroid cultures are well suited for developing high-throughput screening technologies [38,45], and their gene expression profiles are more truly indicative of clinical expression profiles than those detected in flat cell cultures [38,45,46]. Flat cell cultures fail to reproduce crucial aspects of carcinogenesis, such as 3D growth and architecture, cell-cell associations and cellular heterogeneity of in vivo samples. In this study, we have provided proof that bioengineered arrays represent a high-throughput platform reflecting 3D growth conditions of ovarian cancer cells and validated their responses by applying a clinically used therapeutic concept in vitro. Ovarian cancer cells grew as anchorage-independent as multi-cellular aggregates. Immunostaining of structural components indicated cell-cell interactions within aggregates promoting cell survival. This microarray platform has also been used to re-create biophysical and biochemical microenvironmental cues that control stem cell fate [38], further underlining the suitability of this in vitro assay as a powerful 3D culture model.

KLK-Expressing Cells Increase Aggregation and Viability upon Paclitaxel Treatment
As cancer-associated proteases like kallikrein-related (KLK) peptidases have been attributed to chemoresistance-in particular to taxane-based drugs-in ovarian cancer [23,26,27], we further sought to investigate the effect of paclitaxel using gradually increasing doses (0-100 nM) on cell aggregation. Confocal micrographs represented the aggregate morphology with and without paclitaxel treatment (100 nM): large and compact aggregates were formed in non-treated conditions, whereas paclitaxel exposure caused smaller and scattered aggregates and the presence of apoptotic bodies. Paclitaxel treatment was correlated with a positive PI staining, indicating an increased cell death (Figure 2(A)). Both OV-Vector/OV-KLK cells formed significantly fewer aggregates at higher paclitaxel concentrations (10 nM: OV-Vector 47 ± 8%/OV-KLK 54 ± 11%; 100 nM: OV-Vector 44 ± 8%/OV-KLK 60 ± 6%) compared to a lower dose (1 nM; OV-Vector 56 ± 14%/OV-KLK 56 ± 7%) and non-treated controls (0 nM: OV-Vector 57 ± 10%/OV-KLK 61 ± 8%). Strikingly, OV-KLK cells grew significantly more aggregates at higher paclitaxel concentrations (10, 100 nM) than OV-Vector cells (Figure 2(B), top panel). Cell death in both OV-Vector/OV-KLK cell aggregates was significantly increased at higher paclitaxel concentrations (10 nM: OV-Vector 35 ± 1%/OV-KLK 23 ± 7%; 100 nM: OV-Vector 53 ± 3%/OV-KLK 38 ± 4%) compared to a lower dose (1 nM: OV-Vector 10 ± 1%/OV-KLK 9 ± 3%) and non-treated controls (0 nM: OV-Vector 6 ± 1%/OV-KLK 4 ± 2%). Interestingly, OV-KLK cell aggregates showed significantly less cell death at higher paclitaxel concentrations (10, 100 nM) than OV-Vector cells, indicating an increased cell survival (Figure 2(B), bottom panel). Over the monitored time frame of 96 h no release of trapped cells and uniform aggregation of the trapped cells ± paclitaxel treatment were evident as indicated by time-lapse microscopy (Supplementary file). As integrins are associated with cell survival and chemoresistance [30,31,47], we analyzed the expression levels of β1 integrin (ITGB1) and focal adhesion kinase (FAK), an integrin-related factor, after paclitaxel administration. In both OV-Vector/OV-KLK cell aggregates, ITGB1 and FAK levels were increased upon paclitaxel treatment, with a further upregulation in OV-KLK cell aggregates (Figure 2(C)). These results suggest that hydrogel microwell arrays increase cell aggregation and viability of KLK-expressing cells upon paclitaxel treatment. These findings are in line with our previously reported data, showing that KLK4 and KLK7 promote paclitaxel-induced resistance of ovarian cancer cell aggregates that were formed in a tumor fluid (ascites) mimicking microenvironment [25,26]. It was shown that multi-cellular aggregates, harboring a 3D architecture, are more resistant compared to flat cell cultures [48], and compact aggregates are less responsive to different therapeutic regimes, such as chemotherapies, than scattered aggregates [49]. We have also reported that combined expression of KLK4, KLK5, KLK6, and KLK7 in ovarian cancer cells (OV-KLK) mediates resistance to paclitaxel at higher doses (10, 100 nM) compared to control cells (OV-Vector) when grown as flat cell cultures [28]. When the same cells were grown as aggregates in this study, we observed a similar cell survival effect upon KLK expression and paclitaxel treatment. Interestingly, the expression of β1 integrin was decreased upon KLK expression [28], but upon paclitaxel treatment increased in both KLK-expressing and KLK-deficient aggregates, suggesting a critical function of this integrin in paclitaxel-related resistance, only partially induced by these four KLKs.
Integrins and integrin-related factors are required for the responsiveness to anti-cancer drugs that bind to microtubules [50]. Although integrins lack kinase activity, by clustering they recruit and activate kinases, such as FAK. FAK is overexpressed in most ovarian cancers, associated with poor clinical outcome and plays a role in regulating invasion and metastasis [51,52]. Paclitaxel treatment stabilizes microtubule dynamics, thereby inhibiting mitosis [40], and FAK is required for integrin-dependent microtubules stabilization and paclitaxel responsiveness [53]. It was shown that FAK regulates the efficacy of taxane-based drugs in both treatment-sensitive and treatment-resistant cells [54]. We detected increased mRNA levels of FAK in aggregates after paclitaxel treatment, further indicating that FAK is an important cell survival factor in ovarian cancer cells. These findings imply the potential of combinatorial therapeutic approaches including the inhibition of KLKs, integrin and integrin-related factors with cytotoxic drugs for the treatment of ovarian cancer patients, especially those with high KLK levels in their tumors.

Paclitaxel Treatment Alters Integrin Expression of Tailor-Made KLK-Expressing Cell Aggregates
Ovarian cancer cell aggregates derived from the tumor fluid (ascites) of patients with late-stage stage disease range in number (from two to more than 20) and size (from 30-200 µm, even up to 750 µm in diameter) and contain up to 100 cells, suggesting a high patient to patient variability [35,[55][56][57]. This high variability in aggregate size is also reflected in in vitro aggregate cultures applying the liquid overlay technique [35,48,57,58] or hanging droplet method [46,59] using different ovarian cancer cell lines [34]. In order to control the cellular microenvironment of hydrogel microwell arrays, photolithography was used to fabricate microwells of varying sizes (50 × 50, 100 × 100, 150 × 150, 200 × 200 µm) to generate aggregates of different sizes (Figure 3  As paclitaxel is subject to multi-cellular-mediated resistance for ovarian cancer cells [48], we treated tailor-made aggregates with paclitaxel (100 nM), and analyzed the aggregation and death of both OV-Vector/OV-KLK cells relative to the total numbers (Figure 3(B)). While in the larger microwells (100-200 × 100-200 µm) complete cell aggregation (98-100%) and no cell death (0-2%) was detected, the smallest microwells (50 × 50 µm) caused incomplete cell aggregration (73-92%) and cell death (8-27%). Upon paclitaxel treatment both OV-Vector/OV-KLK cells aggregated less (42-96%) in the medium sized (100-150 × 100-150 µm) and smallest (50 × 50 µm) microwells and showed higher cell death (4-58%) rates (Figure 3(B)). Then, we analyzed the aggregate number and area in each microwell size performing time-lapse microscopy ( Figure 4(A)). Both OV-Vector/OV-KLK cells formed one aggregate (1.27-1.34 × 10 3 cm 2 )/well in the smallest (50 × 50 µm) microwells, whereas in the next larger (100 × 100 µm) microwells, two aggregates (4.33-5.34 × 10 3 cm 2 )/well were formed. In the medium sized (150 × 150 µm) and largest (200 × 200 µm) microwells, three aggregates (10.68-19.14 × 10 3 cm 2 )/well were formed. Upon paclitaxel treatment both OV-Vector/OV-KLK cells formed one aggregate (0.72-0.80 × 10 3 cm 2 )/well in the smallest (50 × 50 µm) microwells, while in the next larger (100 × 100 µm) microwells, two aggregates (2.31-3.78 × 10 3 cm 2 )/well were detected. In the medium sized (150 × 150 µm) and largest (200 × 200 µm) microwells, three aggregates (6.98-9.68 × 10 3 cm 2 )/well were formed. OV-KLK cells formed larger aggregates in the second smallest (100 × 100 µm) microwells after paclitaxel treatment compared to OV-Vector cells (Figure 4(B)). These results suggest that OV-KLK cells had a higher ability to aggregate and survive with and without paclitaxel in all microwell sizes compared to OV-Vector cells. The administration of paclitaxel reduced aggregate area but not numbers compared to non-treated conditions. Within these bioengineered microwells, the formation of cell aggregates was achieved in sizes ranging from 50-200 µm. Similar aggregate sizes are described in experimental and clinical samples [35,[55][56][57] showing a high cell viability in combination with KLK expression, and the results presented in this study are in line with our former reports [25,26]. Paclitaxel treatment revealed that the aggregate area but not aggregate number was reduced, further corroborating the existence of survival-promoting factors, such as integrins, and multi-cellular-mediated drug resistance mechanisms in ovarian cancer cells [48]. A similar bioengineered approach to the one described here has been used to control the size and shape of embryonic bodies employing microwells of varying diameters ranging from 40-150 µm and heights of 20-35 µm and has proven its potential to investigate differentiation of embryonic stem cells [60]. These findings indicate that hydrogel microwell arrays can be used to control cell aggregation, aggregate size and viability, to study factors involved in the responsiveness of different sized aggregates to anti-cancer drugs and the contribution of KLKs.
Integrins are integral in mediating cell survival and chemoresistance, in particular α5/β1 integrins [30,31,47]. Hence, we sought to determine α5/β1 integrin mRNA and protein levels in aggregates of varying size upon paclitaxel treatment (100 nM). While no difference in both OV-Vector/ OV-KLK cell aggregates without treatment was found, after paclitaxel treatment ITGA5 was increased in aggregates grown in the largest (150-200 × 150-200 µm) microwells, and ITGB1 was enhanced in aggregates, with highest expression levels in OV-KLK cell aggregates grown in the smallest (50-100 × 50-100 µm) microwells ( Figure 5(A)). Western blot and densitometrical analyses showed that α5 and β1 integrins were enhanced after paclitaxel treatment in OV-KLK cell aggregates compared to OV-Vector cells, which only had increased α5 integrin in the smallest (50 × 50 µm) and medium sized (150 × 150 µm) microwells ( Figure 5(B)). Interestingly, the biggest (200 × 200 µm) microwells resulted in multiple smaller aggregates per microwell (34%), which have the same integrin expression pattern as the aggregates formed in the smallest (50 × 50 µm) microwells. These results suggest that integrin expression is upregulated upon paclitaxel treatment depending on the aggregate size and partially on KLK expression, especially in smaller (50 µm) and larger (150-200 µm) aggregates.
Caspases play an important role in apoptosis induced by anti-cancer drugs [61]. In both OV-Vector/OV-KLK cell aggregates, caspase8 expression followed β1 integrin levels in the smallest (50 × 50 µm) and largest (200 × 200 µm) microwells. OV-Vector cell aggregates showed a downregulation of capsase8 in medium sized (100-150 × 100-150 µm) microwells upon paclitaxel treatment ( Figure S1). These results imply an involvement of integrins in paclitaxel-induced apoptosis. However, our findings suggest a bi-functional effect of drug treatment: (i) upregulation of integrins to promote cell aggregate survival, and (ii) upregulation of caspase-8 to mediate cell death, further underlining the fine-tuned balance between drug sensitivity and drug resistance.
It was shown that the membrane type 1 matrix metalloproteinase (MT1-MMP) regulates ovarian cancer cell aggregation and disaggregation, and its expression level is increased in aggregates relative to flat cell cultures [43]. Ovarian cancer cell aggregates grown within microwells of varying sizes showed MT1-MMP expression in all aggregate sizes independent of KLK expression and paclitaxel treatment ( Figure S1). MT1-MMP can be regulated by integrin clustering which was shown to be stimulated by a 3D collagen type I microenvironment [62]. In addition to MT1-MMP activity [43], other factors, such as contractile forces [59], promote cell aggregation. The simultaneous presence of MT1-MMP and integrins in aggregates grown within hydrogel microwell arrays further indicates their interactive relationship within this microarray platform.

Blocking of Integrin Function Does Not Affect Cell Aggregation
It was shown that β1 integrin regulates the formation of ovarian cancer cell aggregates that were generated using the liquid overlay technique [35,55,56]. Hence, we sought to test whether the formation OV-Vector/OV-KLK cell aggregates produced in hydrogel microwell arrays is dependent on β1 integrin by using a functionally blocking antibody (10 µg/mL). Surprisingly, both OV-Vector/OV-KLK cell aggregate number and area was enhanced with increasing microwell size (150-200 × 150-200 µm) upon integrin inhibition, with more (up to three aggregates/well) and larger (12.84-18.18 × 10 3 cm 2 ) aggregates being formed compared to non-treated conditions. With decreasing microwell size (50-100 × 50-100 µm), only one to two aggregates/well and smaller aggregates (1.30-5.35 × 10 3 cm 2 ) were formed. In the medium sized microwells (150 × 150 µm), OV-KLK cells formed significantly larger aggregates after integrin inhibition compared to non-treated conditions ( Figure 6(A,B)). Different to the study by Casey et al. [35], which reported the inhibition of aggregation using the same blocking β1 integrin antibody after 8 h and 24 h in serum-free media, is that we documented the integrin inhibition over 96 h in serum-containing media. Casey et al. [35] showed that after 8 h aggregate formation was inhibited by the blocking β1 integrin antibody, resulting in none or small aggregates. At 24 h, β1 integrin inhibition continued to partially block aggregate formation, resulting in medium to large aggregates. The incomplete inhibition of the β1 integrin at the 24 h time point suggests that if this integrin is inactivated, ovarian cancer cells might possess a compensatory mechanism to facilitate aggregation. However, the antibody might have been internalized over 24 h and 96 h, eventually enabling ovarian cancer cells to aggregate. Moreover, the presence of the serum-containing media allows the continuous proliferation of cells over a longer period of time. It was suggested that β1 integrin mediates the initial formation of cell aggregates and that multiple integrin-ECM interactions, such as αv integrin/vitronectin [57], are involved in this process.
Contrary to Casey et al. [35] we hypothesized that aggregation time and technique are important parameters. Casey et al. [35] demonstrated that NIH:OVCAR5 cells formed stable aggregates within 48 h using the liquid overlay method, whereas the OV-MZ-6 cells used in our study formed compact aggregates for up the 120 h within hydrogel microwell arrays. In our previously published work, we demonstrated that OV-MZ-6 cell spheroids proliferated for up to 28 days [8], underlining the robustness of this cell line when combined with a biomimetic hydrogel in a high-throughput system. Although the capacity to form compact aggregates differs between ovarian cancer cell lines [8,35,57], the aggregates formed in bioengineered microenvironment emerge to be similar to those present in the tumor fluid (ascites) of patients.

Conclusions
When entering the third dimension, investigators need to consider the design of microenvironments for supporting the cell architecture and the capability to conduct such a system in high-throughput. We provide evidence that hydrogel microwell arrays can be engineered to replicate intricate biological functions the tumor microenvironment by allowing aggregation of ovarian cancer cells, and thus, are well suited to decipher the function of cancer-associated proteases and integrins in disease progression and therapy-resistance. Tailor-made hydrogel microwells increase cell aggregation and insensitivity to paclitaxel treatment, in particular in KLK-expressing cancer cells, and thus, representing events seen in patients with metastatic outgrowth. KLK expression in cancer cell aggregates was accompanied with altered integrin levels and integrin-related factors upon paclitaxel treatment. However, blocking of integrin function did not affect cancer cell aggregation, suggesting that the involvement of other cell surface molecules and/or receptors play an important role. In conclusion, the technology platform presented in this study has the potential to provide an alternative screening tool for the efficacy of novel therapeutics specifically targeting multi-cellular aggregates for intra-abdominal intervention of late-stage disease.