Tumor FAK supports fibrillar collagen expression in ovarian cancer ascites

Summary

High-grade serous ovarian cancer (HGSOC) is a lethal malignancy where matrix proteins enhance ascites-associated recurrent tumor growth. Elevated focal adhesion kinase (FAK/PTK2) and collagen protein expression support tumor survival, chemoresistance, and immune evasion. However, the sources and regulation of collagen production within HGSOC tumorspheres is unclear. Using CRISPR-mediated FAK knockout and FAK re-expression in two syngeneic HGSOC models (MOVCAR and KMF), FAK loss impaired tumorsphere growth in vitro and in vivo while reducing collagen expression, as determined by RNA sequencing, quantitative RT-PCR, and protein mass spectrometry. FAK re-expression promoted spontaneous metastatic ovarian tumor dissemination. Integrated bioinformatic analyses of MOVCAR, KMF, and HGSOC patient datasets identified 56 common mRNAs co-elevated with PTK2, including COL3A1 and COL5A2. Single-cell RNA sequencing of murine KMF ascites revealed a subset of tumor cells expressing Col3a1 and Col5a2 that exhibited epithelial-to-mesenchymal pathway enrichment that was diminished by genetic or pharmacologic FAK inhibition. In primary HGSOC ascites samples, FAK inhibitor (IN10018 or VS-4718) or FAK protein degrader (PROTAC) treatment selectively reduced COL1A1, COL3A1, and COL5A2 mRNA levels in tumor cells. Further, human OVCAR8 spheroid COL1A1, COL3A1, and COL5A2 mRNA, COL3A1 protein levels, and spheroid structure were significantly impacted by IN10018, FAK PROTAC, and blocking anti-α1 or anti-β1 integrin, but not by control IgG, anti-α3 or anti-αv integrin antibodies. As elevated PTK2 and COL3A1 expression correlate with poor ovarian cancer patient survival, and α1β1 integrin binding to collagen activates FAK, our findings support a feed-forward FAK-collagen signaling axis promoting HGSOC tumorsphere survival and metastatic spread.

Introduction

High-grade serous ovarian cancer (HGSOC) remains the most lethal gynecologic malignancy in the United States, with many patients initially responding to platinum-based chemotherapy but ultimately developing recurrent, therapy-resistant disease [1, 2]. Unlike many solid tumors that metastasize through the bloodstream, HGSOC commonly disseminates throughout the peritoneal cavity, where detached tumor cells survive as multicellular spheroids suspended within malignant ascites [3]. These anchorage-independent tumorspheres attach to mesothelial surfaces, colonize the omentum and peritoneum, and ultimately give rise to recurrent metastatic lesions [4]. The biological mechanisms that enable ovarian tumorspheres to survive under non-adherent conditions and resist therapy remain incompletely understood.

The ascites microenvironment functions as an active participant in ovarian cancer progression rather than simply a byproduct of advanced disease [5, 6]. Ascites contains a complex mixture of growth factors, cytokines, lipids, and extracellular matrix (ECM) proteins that collectively support tumor cell survival and metastatic outgrowth [7–9]. Matrix remodeling has emerged as a hallmark of aggressive ovarian cancer [10, 11]. Elevated expression of fibrillar collagens and other matrisome components are associated with poor patient survival, increased metastatic burden, and resistance to both chemotherapy and immunotherapy [12–14]. Dense collagen matrices not only provide structural support for tumor growth but also activate integrin-dependent signaling pathways, enhance mechano-transduction, limit drug penetration, and establish an immune-excluded tumor microenvironment [15–19]. Despite the recognized importance of collagen, and the integrin receptors on tumor cells that bind collagen in ovarian cancer progression [20, 21], the cellular sources and regulatory mechanisms controlling collagen production within malignant ascites remain poorly defined.

Focal adhesion kinase (FAK), encoded by the PTK2 gene, is a non-receptor tyrosine kinase that integrates signals from integrins and growth factor receptors to regulate cell survival, migration, proliferation, and mechanosensing [22]. PTK2 amplification and elevated FAK activation occur frequently in HGSOC and are associated with a poor clinical outcome [23]. Experimental studies have shown that FAK signaling promotes ovarian cancer stem-like properties, tumorsphere growth, metastatic dissemination, and resistance to platinum chemotherapy [24]. In addition to these tumor-intrinsic functions, FAK has emerged as an important regulator of the tumor microenvironment by controlling inflammatory transcriptional programs, stromal remodeling, and anti-tumor immunity [25–27]. These observations have stimulated the clinical development of selective FAK inhibitors, several of which are currently being evaluated in combination with chemo- and immuno-therapies [28].

An important but unresolved question is whether tumor FAK directly regulates extracellular matrix production during ovarian cancer progression. FAK is activated downstream of collagen-β1 integrin interactions [29], and matrix stiffness further enhances FAK signaling, suggesting the existence of a reciprocal signaling relationship between FAK and the extracellular matrix [30]. Although FAK signaling in stromal fibroblasts is an established source of collagen in many solid tumors [31, 32], recent single-cell studies have also identified epithelial-to-mesenchymal transitioning (EMT) tumor cells capable of producing extracellular matrix proteins [33]. Whether similar collagen-producing tumor cell populations exist within ovarian cancer ascites and whether their activity depends upon FAK signaling are not known.

Herein, we find that tumor-intrinsic FAK regulates a collagen-producing mesenchymal program associated with anchorage-independent ovarian cancer growth and metastatic dissemination. Using complementary genetic and pharmacologic approaches in two syngeneic ovarian cancer models together with transcriptomic, proteomic, and single-cell RNA sequencing analyses, we identify a FAK-dependent population of collagen-expressing tumor cells enriched for EMT-associated pathways. Integration with human ovarian cancer datasets reveals a conserved FAK-collagen transcriptional program associated with PTK2 amplification and poor patient outcome. Analyses of primary patient ascites specimens demonstrate that clinically relevant FAK inhibitors and FAK PROTAC degraders suppress fibrillar COL1A1, COL3A1, and COL5A2 expression in recurrent HGSOC tumor cells. Further, as anti-α1 or -β1 integrin blocking antibodies suppressed collagen expression in OVCAR8 ovarian spheroids, our results support the existence of a feed-forward integrin-FAK-collagen signaling axis in ovarian cancer ascites that promotes tumorsphere survival and metastatic dissemination.

Results

FAK promotes anchorage-independent ovarian tumorsphere growth and collagen expression

To investigate the role of FAK in matrix production during ovarian cancer progression, we isolated tumor ascites of T-antigen transformed FAK^fl/fl mice at 17 weeks and expanded FAK^fl/fl mouse ovarian carcinoma (MOVCAR) tumor cells ex vivo as tumorspheres (Fig. 1A). To not influence transformation in vivo, adenoviral Cre recombinase was used to inactivate FAK expression in isolated FAK^fl/fl MOVCARs ex vivo as determined by immunoblotting (Fig. 1B). A green fluorescent FAK fusion protein (GFP-FAK) was re-expressed in FAK^−/− MOVCARs and was active as visualized by Tyr-397 (FAK pY397) phosphorylation (Fig. 1B). No changes in FAK related PYK2 protein expression were detected in FAK-deleted and reconstituted MOVCARs.

Figure 1. Genetic FAK inactivation limits tumorsphere growth and collagen expression in a mouse ovarian carcinoma (MOVCAR) model.

(A) Representative image of a SV40 TAg-driven ovarian tumor from FAK^fl/fl mice with ascites harvested at 17 weeks (left) and tumorspheres expanded from ascites-derived cells (right).

(B) Immunoblot analysis confirming FAK deletion, following Cre recombinase treatment, and stable GFP-FAK re-expression as visualized with antibodies to FAK, activated FAK pY397, PYK2, and actin loading control.

(C) Real-time XCelligence analysis showing that FAK loss does not significantly impair two-dimensional adherent cell proliferation compared to parental and GFP-FAK re-expression. Dats is representative of 3 independent experiments.

(D) Quantification of MOVCAR three-dimensional growth over 7 days of FAK^−/− cells and enhancement by GFP-FAK expression. Dotted line represents starting cell number. Circles are experimental points and values are means (± SEM, **** P<0.0001 t-test).

(E) RNA sequencing volcano plot showing differential RNA expression in FAK^−/− and FAK^−/− + GFP-FAK MOVCARs grown as 3D tumorspheres identifying differentially expressed genes enriched for collagens and secreted factors including Col3a1, Col18a1, Tgfb2, and Igf1.

(F) Reactome pathway enrichment analysis of RNAs reduced in FAK^−/− tumorspheres reveals significant suppression of extracellular matrix organization, collagen biosynthesis, post-translational protein phosphorylation, and regulation of Insulin-like growth.

(G) Quantitative RT-PCR validation of selected FAK-regulated targets finds increased expression of Col3a1, Col5a1, Col5a2, Col1a1, Lyve1, and Ndrg1 following GFP-FAK re-expression. Values are means of two independent points and FAK^−/− values set to one.

(H) Mass spectrometry analysis of tumorspheres confirms reduced Col1a1, Col1a2, Col3a1, Col5a2 protein expression in FAK^−/− MOVCARs with limited increases in Col4a6 and Col4a5. Values are expressed as Log2-fold change between FAK^−/− and GFP-FAK MOVCARs.

MOVCAR FAK deletion did not alter two-dimensional adherent proliferation as measured by real-time impedance analysis, indicating that FAK was not essential for conventional monolayer growth (Fig. 1C). In contrast, FAK was required for anchorage-independent tumorsphere expansion. FAK-deficient MOVCAR cells exhibited markedly impaired three-dimensional growth over five days, whereas GFP-FAK re-expression restored and enhanced tumorsphere formation (Fig. 1D). Independent FAK knockout clones showed a similar reduction in tumorsphere growth, confirming that this phenotype was not clone-specific (Supplemental Fig. 1).

To identify molecular pathways regulated by FAK during tumorsphere growth, RNA sequencing was performed on FAK^−/− and GFP-FAK reconstituted MOVCAR tumorspheres. Differential expression analysis identified 701 transcripts reduced and 464 increased following FAK loss, with prominent suppression of extracellular matrix-associated genes including Col3a1, Col18a1, Col12a1 and secreted factors such as Ccl5, Angptl7, and Crlf1 by FAK loss (Fig. 1E, Table 1, and Supplemental Data 1). Reactome pathway analysis revealed significant enrichment for extracellular matrix organization, collagen formation, collagen biosynthesis, assembly of collagen fibrils, integrin-mediated interactions, and regulation of insulin-like growth factor signaling among FAK-dependent transcripts (Fig. 1F). Quantitative real-time PCR (RT-PCR) validated FAK-dependent induction of multiple extracellular matrix genes, including Col1a1, Col3a1, Col5a1, and Col5a2, together with Lyve1 and Ndrg1 (Fig. 1G). Mass spectrometry analysis of extracellular matrix-enriched tumorsphere preparations confirmed reduced collagen protein abundance after FAK loss, with Col1a1, Col1a2, Col3a1, and Col5a2 among the most strongly decreased proteins (Fig. 1H and Supplemental Data 2). Moreover, FAK expression was associated with elevated levels of collagens, ECM-affiliated proteins, ECM glycoproteins, ECM regulators, proteoglycans, and secreted factors (Supplemental Fig. 2). Taken together, these findings show that FAK supports collagen production during anchorage-independent ovarian cell survival.

Table 1.

Top-10 in vitro differentially expressed matrisome-associated genes by RNA sequencing of MOVCAR FAK^−/− and MOVCAR FAK^−/− + GFP-FAK cells grown as 3D tumorspheres over 5 days.

Category	Subcategory	Gene Name	log2(FC)	padj
Core matrisome	Collagens	Col12a1	−3.802632075	1.37E-06
Core matrisome	ECM Glycoproteins	Fras1	−3.589483106	2.75E-21
Matrisome-associated	Secreted Factors	Ccl5	−3.5733191	4.11E-06
Matrisome-associated	ECM Regulators	Serpina11	−3.366847812	0.048941157
Core matrisome	Collagens	Col18a1	−3.239912064	1.10E-17
Matrisome-associated	Secreted Factors	Angptl7	−3.1239101	2.29E-14
Matrisome-associated	Secreted Factors	S100a4	−3.1061177	0.03608576
Core matrisome	ECM Glycoproteins	Spon2	−3.03349759	1.92E-09
Matrisome-associated	Secreted Factors	Crlf1	−2.950932938	2.88E-05

FAK promotes spontaneous ovarian tumor dissemination and in vivo extracellular matrix remodeling

To determine whether FAK regulation of tumorsphere survival and changes in collagen production may impact growth as tumors in mice, FAK^−/− and GFP-FAK reconstituted MOVCAR cells were orthotopically implanted into the ovarian bursa of syngeneic mice (Fig. 2). At 12 weeks, the mean ovary mass was greater in mice injected with GFP-FAK compared to FAK^−/− MOVCARs (Fig. 2A) and peritoneal wash of tumor-bearing mice revealed more extensive hemorrhagic ascites and significantly higher numbers of viable cells from GFP-FAK compared to FAK^−/− MOVCARs (Figs. 2B and C). Restored growth of FAK^−/− MOVCARs by GFP-FAK expression equivalent to parental FAK^fl/fl MOVCARs was also observed upon intraperitoneal injection and growth of ascites tumors (Supplemental Fig. 3).

Figure 2. MOVCAR FAK promotes tumor growth, spontaneous metastasis, and extracellular matrix organization.

(A) Ovary weight 12 weeks after bursa-injection of FAK^−/− or FAK^−/− +GFP-FAK MOVCARs. Each symbol represents an individual mouse, and bars indicate mean ± SEM. Differences are not significant.

(B) Representative images of peritoneal lavage fluid collected at the experimental endpoint demonstrating increased hemorrhagic ascites associated with GFP-FAK re-expression.

(C) Significantly increased viable ascites cells recovered from mice with GFP-FAK compared to FAK^−/− MOVCARs. Symbols are individual mice and values means (± SEM. ***P < 0.001, t-test).

(D) Representative Masson’s trichrome staining of spontaneous omental metastases at 12 weeks. GFP-FAK tumors exhibit extensive metastatic colonization associated with abundant collagen-rich stromal deposition (blue staining), whereas FAK^−/− tumors display markedly reduced omental invasion and extracellular matrix accumulation. Enlarged regions are indicated by red boxes. Scale is 200 μm.

(E) Representative Masson’s trichrome stained tumorspheres harvested by cytospin of ascites. GFP-FAK promotes large multicellular tumorspheres with prominent collagen deposition, whereas FAK^−/− tumors form smaller and less abundant spheroids. Insets show higher magnification views of the boxed regions. Scale is 200 μm.

(F) Reactome pathway enrichment analysis of RNA sequencing data from bursa-derived tumors. Differentially expressed genes associated with FAK expression were significantly enriched for extracellular matrix organization, transmission across chemical synapses, neuronal systems, and biological oxidations.

Histological analyses demonstrated that FAK promoted spontaneous metastatic dissemination to omental tissue with invasion and collagen production as detected by Masson’s trichrome staining (Fig. 2D). Few FAK^−/− MOVCAR omental metastases were detected, and when detected, these were small clusters of non-invasive and low collagen producing tumor cells (Fig. 2D). Similarly, ascites-derived tumorspheres from GFP-FAK tumors were larger and contained abundant collagen-rich matrix compared with the smaller and dense spheroids formed by FAK-deficient cells (Fig. 2E).

Transcriptomic analysis of orthotopic tumors further supported these observations. Reactome pathway analysis identified extracellular matrix organization as one of the most significantly enriched FAK-associated biological programs in vivo (Fig. 2F and Table 2). Consistent with these findings, matrisome-focused analysis identified Col3a1 among the most significantly FAK-regulated genes in orthotopic tumors. Together, these results show that tumor-mediated regulation of collagen by FAK occurs in vitro and in vivo.

Table 2.

Top-10 in vivo matrisome-associated genes by RNA sequencing of orthotopic ovarian bursa-injected MOVCAR FAK^−/− and MOVCAR FAK^−/− + GFP-FAK tumors after 12 weeks.

Category	Subcategory	Gene Name	log2(FC)	padj
Matrisome-associated	ECM Regulators	Adamts15	−2.143947158	2.16E-05
Matrisome-associated	Secreted Factors	Bmp6	−1.887632802	0.001389797
Matrisome-associated	Secreted Factors	Nrg2	−1.883973493	0.043503187
Matrisome-associated	ECM Regulators	Mmp17	−1.836241877	0.00381128
Core matrisome	Collagens	Col3a1	−1.804364978	0.000389751
Matrisome-associated	Secreted Factors	Igf1	−1.591000241	0.000513534
Core matrisome	ECM Glycoproteins	Papln	−1.501089718	6.58E-05
Core matrisome	ECM Glycoproteins	Nid2	−1.438598162	0.009598653
Matrisome-associated	ECM Regulators	Prss12	−1.43326125	0.023145891
Matrisome-associated	Secreted Factors	Megf6	−1.429637425	9.77E-05

Conserved FAK-dependent collagen programs are shared between murine models and human HGSOC

To determine whether FAK-regulated extracellular matrix pathways identified in mouse models were conserved in human disease, we integrated transcriptomic datasets from MOVCAR, KMF, and HGSOC from The Cancer Genome Atlas (TCGA) (Fig. 3). Stratification of TCGA tumors according to PTK2 expression identified a distinct PTK2-high transcriptional state (Fig. 3A). Comparison of FAK-regulated genes from both murine models with PTK2-associated transcripts in TCGA revealed a common set of fifty-six conserved genes (Fig. 3B). Ranking of these genes identified COL5A2, COL3A1, and COL5A1 among the strongest conserved FAK-associated targets, and several of these were independently validated at the protein level by mass spectrometry (Fig. 3C). These data support the existence of an evolutionarily conserved FAK-dependent fibrillar collagen transcriptional program operating across murine ovarian cancer models and human HGSOC.

Figure 3. Conserved FAK-dependent association with collagen expression in mouse ovarian cancer models and human HGSOC.

(A) TCGA GDC 2025 ovarian cancer dataset clustering of RNAs associated with low- or high-PTK2 expression. Transcript abundance is presented as fragments per kilobase of transcript per million mapped reads (FPKM). Violin plots show data points and quartiles (****P < 0.0001, two-sided Welch’s t-tests).

(B) Venn diagram showing the overlap between genes associated with elevated PTK2 expression in HGSOC and genes regulated by FAK expression in two independent murine ovarian cancer models (MOVCAR and KMF). Differentially expressed genes identified following GFP-FAK re-expression in FAK-deficient MOVCAR and KMF cells were compared with the TCGA PTK2-associated gene signature, revealing a conserved group of fifty-six targets.

(C) Top eight FAK-associated genes and those also identified by MOVCAR tumor mass spectrometry (designated with *). Order established by baseMean (counts) and fold-change. COL5A2, COL3A1, and Col5A1 were significantly associated with elevated PTK2.

Single-cell RNA sequencing identifies a FAK-regulated mesenchymal tumor cell population expressing Col3a1 and Col5a2

We next sought to identify the cellular source of collagen production within the ascites ovarian cancer microenvironment. Single-cell RNA sequencing was performed on total ascites cells collected from mice bearing parental KMF tumors, FAK-knockout tumors, or tumors treated with the FAK selective inhibitor ifebemtinib (IN10018) (Fig. 4A). UMAP analysis demonstrated that Col3a1 and Col5a2 expression localized to a discrete subset of tumor cells rather than broadly across the tumor population (Fig. 4B). Both pharmacologic FAK inhibition and genetic FAK deletion significantly reduced Col3a1 and Col5a2 expression within KMF ascites tumor cells (Fig. 4C). Pathway enrichment analysis of shared Col3a1/Col5a2-associated gene programs identified MET-PTK2 signaling, extracellular matrix organization, collagen biosynthesis, collagen fibril assembly, and integrin-mediated pathways as associated biological processes (Fig. 4D).

Figure 4. Single-cell transcriptomic analyses identify FAK activity-dependent regulation of Col3a1 and Col5a2 associated with epithelial to mesenchymal (EMT) pathway activity.

(A) Experimental design. C57BL/6 mice were injected intraperitoneally with KMF FAK-knockout (FAK-KO), KMF FAK-KO cells re-expressing GFP-FAK, or parental KMF tumor-bearing mice were treated with FAK inhibitor (IN10018, 25 mg/kg, once daily). Ascites cells were harvested on day 25 and subjected to single-cell RNA sequencing (scRNA-seq) as previously described [26].

(B) UMAP projection of tumor cells identified by scRNA-seq with feature plots demonstrating expression of Col3a1 and Col5a2 enriched in distinct tumor cell populations (tumor cells are blue with myeloid cells in green).

(C) Violin plots show log-normalized Col3a1 and Col5a2 expression in gene-positive tumor cells across WT (cyan), FAKi (magenta), and KO (orange) conditions. Pharmacologic inhibition or genetic ablation of FAK significantly reduced expression of both collagen genes. Pairwise differences in expression were assessed using Mann–Whitney U tests with Benjamini–Hochberg correction following a Kruskal–Wallis omnibus test. (***FDR < 0.001).

(D) mouse transcriptomic Reactome pathway enrichment analysis of shared Col3a1/Col5a2 programs. Significantly enriched pathways included MET-PTK2 signaling, extracellular matrix organization, extracellular matrix degradation, collagen biosynthesis, collagen fibril assembly, and integrin-mediated interactions (FDR ≤ 0.05).

(E) Schematic illustrating the analytical workflow used to identify a Col3a1/Col5a2-positive tumor cell population, integrate these data with human ovarian cancer pathway analyses, and define the top 100 conserved marker genes.

(F) Gene set variation analysis (GSVA) demonstrating a positive correlation (Pearson = 0.44) between the Col3a1-specific transcriptional program and EMT pathway activity.

(G) Heatmap of pathways associated with the Col3a1-specific gene signature.

(H-I) GSVA analyses and pathway heatmap for Col5a2-specific transcriptional program with EMT and Ras/MAPK pathway activation associations.

(J) Shared Col3a1/Col5a2 gene programs positively correlate with both EMT and Ras/MAPK pathway activity, supporting coordinated regulation of extracellular matrix remodeling and oncogenic signaling.

(K) Heatmap of genes comprising the shared Col3a1/Col5a2 program and their associations with EMT pathway activation with cell cycle and DNA damage pathway inhibition.

To determine whether these collagen-expressing tumor cells exhibited distinct phenotypic markers, we integrated the murine single-cell signatures and identified the top 100 shared markers with human ovarian cancer (Fig. 4E). Differential expression analysis within the Col3a1 and Col5a2 tumor fraction yielded 100 top-ranked marker genes for each subpopulation, of which 83 were shared, pointing to a common transcriptional program. The remaining 17 targets were unique to each subpopulation, likely reflecting functionally distinct collagen-expressing states (Supplemental Data 3). Pathway enrichment analysis of these shared markers in the TCGA ovarian cancer cohort (n = 309) revealed significant activation of EMT pathways in Col3a1⁺ and Col5a2⁺-associated gene programs (FDR ≤ 0.05). (Figs. 4F and G). Col5a2-associated programs were similarly linked to EMT and Ras/MAPK pathway activation (Figs. 4H and I). The shared Col3a1/Col5a2 gene signature exhibited the strongest positive associations with both EMT and Ras/MAPK signaling while inversely correlating with cell cycle and DNA damage response pathways (Figs. 4J and K). Together, these findings demonstrate that Col3a1 and Col5a2 mark a discrete, FAK-regulated EMT-transitioning tumor cell state within the HGSOC peritoneal niche, rather than reflecting contaminating stromal populations.

FAK regulates collagen mRNAs in primary human ovarian cancer ascites tumor cells

To evaluate the clinical relevance of murine model findings, primary ascites specimens from patients with recurrent HGSOC were analyzed ex vivo (Fig. 5 and Supplemental Table 1). Ascites-derived cells were maintained under anchorage-independent conditions and treated with selective FAK inhibitors, dual FAK/PYK2 inhibition, or FAK and or FAK/PYK2-targeting PROTAC degraders (Fig. 5A). Across independent patient samples, FAK inhibition or FAK protein degradation consistently reduced COL1A1 and COL3A1 mRNA levels (Figs. 5B and C). In contrast, COL5A1 and COL5A2 responses were more heterogeneous among individual patients (Figs. 5D and E). Analysis of density gradient-purified tumor cells confirmed that FAK inhibition directly suppressed tumor cell COL1A1 and COL3A1 mRNA levels (Fig. 5F). Moreover, analysis of an independent TCGA cohort demonstrated that tumors harboring PTK2 amplification expressed significantly higher levels of both COL1A1 and COL3A1 than PTK2-diploid tumors (Fig. 5I).

Figure 5. FAK expression and activity regulate collagen gene expression in human ovarian cancer ascites tumor cells.

(A) Experimental workflow for ex vivo analyses of recurrent HGSOC specimens. Ascites was collected by paracentesis and either cultured as unfractionated cell suspensions or subjected to density gradient centrifugation to isolate tumor cells. Samples were treated with vehicle (DMSO), the selective FAK inhibitor IN10018, the dual FAK/PYK2 inhibitor VS4718, a FAK-specific PROTAC (FAK-PROTAC-1), or a FAK-PYK2-targeting (PROTAC FC-11) degrader (1 μM, 48 h) prior to molecular analyses.

(B-E) Quantitative real-time PCR (qRT-PCR) heatmaps showing relative collagen isoform mRNA levels in primary ovarian cancer samples following FAK pathway inhibition. COL1A1 (B) and COL3A1 (C) expression was broadly reduced by pharmacologic inhibition or degradation of FAK signaling components, whereas COL5A1 (D) and COL5A2 (E) exhibited patient-dependent responses. Gene expression is presented relative to DMSO-treated controls. PAS-3 represents a low-grade serous ovarian cancer (LGSOC) sample, whereas all other specimens are recurrent HGSOC.

(F) Quantitative RT-PCR analysis of gradient-purified PAS-1 tumor cells (4/25) demonstrating reduced COL1A1 and COL3A1 expression following FAK inhibition. Data are normalized to vehicle-treated controls.

(G) TCGA ovarian cancer dataset analysis showing that tumors harboring PTK2 amplification express significantly higher levels of COL1A1 and COL3A1 mRNA compared with PTK2-diploid tumors. Each symbol represents an individual patient sample. (*P < 0.05, two-sided Welch’s t-tests).

(H) Phase contrast images of PAS-9 ascites cells 48 h after DMSO, IN10018, or VS4718 treatments. FAK inhibition triggered cell dispersal. Scale is 200 μm.

(I) Protein Immunoblot analysis in PAS-9 samples confirming FAK activation (pY397 FAK) in DMSO control ascites, FAK inhibition by IN10018 or VS4718, and the reduction of both FAK and PYK2 by FC-11 PROTAC degradation. Actin is a loading control.

(J) No change in Live/Dead cell staining by flow cytometry after treatment with 1 μM IN10018 or FAK-PYK2 PROTAC for 48 h. Values are means of 3 independent experimental points.

Phase contrast imaging of ascites cells treated with IN10018 or VS4718 small molecule FAK inhibitors revealed a dispersion of multi-cell clusters after 48 h (Fig. 5H). Immunoblot analyses verified effective inhibition of FAK autophosphorylation by IN10018 and VS4718 and demonstrated efficient depletion of FAK and PYK2 following PROTAC treatment (Fig. 5I). These interventions did not significantly alter cell viability, indicating that reduced collagen expression was not secondary to acute cytotoxicity (Fig. 5J). Kaplan-Meier analyses reveal that elevated PTK2, COL1A1, COL3A1, and COL5A2 correlate with poor overall survival (Supplemental Fig. 4), these data support a model whereby FAK activation, potentially by collagen binding integrin within ascites tumor cells supports fibrillar collagen production and a possible feed-forward regulatory circuit supporting tumorsphere survival.

FAK and α1β1 integrins regulate human OVCAR8 fibrillar collagen expression and spheroid morphology

FAK activity is regulated by β1 integrin clustering in both normal and tumor cells [22]. Fibrillar collagens bind to β1 integrin heterodimeric receptors formed with α1, α2, α10, and α11 integrins [34]. To determine whether integrin-dependent FAK signaling contributes to fibrillar collagen expression, human OVCAR8 cells were cultured in serum-free conditions as cell number defined spheroids in agarose microwells (Fig. 6A) and treated with IN10018 FAK inhibitor, FAK PROTAC (FC-11) or with validated α1, α3, αv, and β1 anti-integrin matrix blocking antibodies. Within 24 h, changes in spheroid morphology were detected in FAKi, FC-11, anti-α1, and anti-β1 integrin treated conditions (Fig. 6B). By 48 h, these spheroids exhibited significantly enlarged cross-sectional area compared to DMSO, IgG control, α3, and αv integrin antibody treated spheroids (Fig. 6C). Analysis of fibrillar collagen expression revealed a striking dependence on FAK and α1β1 integrin signaling. FAK inhibition and FAK degradation markedly reduced OVCAR8 COL1A1, COL3A1, and COL5A2 mRNA expression compared with controls (Figs. 6D–F) whereas α3 or αv integrin blockade did not suppress collagen expression. Notably, integrin α1 mRNA is positively associated with COL3A1 expression (Spearman 0.66 and Pearson 0.61) in human ovarian cancer (Supplemental Fig. 5). Lastly, consistent with transcriptional changes by FAK or integrin inhibition, COL3A1 protein levels were evaluated in OVCAR8 spheroids following FAK inhibition or integrin blockade (Fig. 6H).

Figure 6. FAK activity and α1β1 integrin signaling promote fibrillar collagen expression and compact spheroid morphology in OVCAR8 ovarian cancer cells.

(A) Schematic of the agar-mold casting method used to generate tumor spheroids in agarose microwells (adapted from www.microtissues.com).

(B) Representative phase-contrast images of OVCAR8 spheroids after 24 h treatment with vehicle (DMSO), control IgG, FAK inhibitor IN10018 or FAK PROTAC FC-11, or blocking antibodies against β1, α1, αv, or α3 integrins. Scale is 200 μm.

(C) Quantification of OVCAR8 spheroid size at 48 h following the indicated treatments.

(D–F) Relative COL1A1 (D), COL3A1 (E), and COL5A2 (F) mRNA expression following FAK or integrin inhibition, normalized to IgG control. FAK and α1 or β1 integrin inhibition significantly reduced fibrillar collagen mRNA expression, whereas α3 or αv integrin blockade had little or no inhibitory effect.

(G) Representative phase-contrast images of OVCAR8 spheroids at 48 h showing reduced spheroid compaction following FAK inhibition with IN10018 or FAK PROTAC degradation by FC-11 compared with DMSO control. Scale is 200 μm.

(H) Immunoblot analysis of COL3A1 from OVCAR8 spheroids following the indicated treatments; actin is a loading control.

Symbols are individual experimental points and values means (± SEM ***P<000.1, ****P < 0.0001 by ANOVA with Tukey post-hoc test).

These results support an α1β1 integrin–FAK signaling pathway that maintains fibrillar collagen expression in ovarian tumor ascites and promotes the compact organization of OVCAR8 ovarian tumor spheroids. Together, these findings support the existence of a feed-forward FAK-collagen signaling axis that promotes tumorsphere survival and metastatic dissemination.

Discussion

High-grade serous ovarian cancer (HGSOC) recurrence is characterized by the persistence of anchorage-independent tumor spheroids within the peritoneal cavity, where extracellular matrix (ECM) remodeling contributes to metastatic spread and therapeutic resistance. Here, we identify tumor-intrinsic focal adhesion kinase (FAK) as a central regulator of fibrillar collagen production during ovarian tumorsphere growth. Using complementary genetic, pharmacologic, proteomic, transcriptomic, and single-cell approaches, we show that FAK promotes expression of multiple collagen genes and proteins, enhances spontaneous metastatic dissemination, and supports the emergence of a mesenchymal-like tumor cell state specialized for collagen production. Importantly, our findings extend to primary ovarian cancer patient ascites specimens, where FAK inhibition or FAK degradation suppressed COL1A1 and COL3A1 expression independent of prior chemotherapy exposure. Moreover, we find that FAK and α1β1 integrins regulate human OVCAR8 fibrillar collagen expression and spheroid morphology. Together, these results support a model in which a FAK-collagen feed-forward circuit may act to sustain tumorsphere survival, promote chemo-resistance signaling, and possibly contribute physical barriers to immune cell infiltration (Fig. 7) [28].

Figure 7. Proposed FAK–collagen feed-forward signaling loop promotes ovarian tumor spheroid organization.

Integrin engagement of extracellular matrix (ECM) collagen activates FAK signaling, which promotes expression of fibrillar collagens COL3A1, COL1A1, and COL5A2 and supports formation of dense, collagen-rich tumor spheroids. Increased collagen deposition can further stimulate integrin-FAK signaling, establishing a feed-forward loop that promotes tumor growth and supports immune evasion. Conversely, FAK inhibition decreases collagen expression and disrupts spheroid organization, resulting in reduced cell-cell contacts and a lower-density collagen matrix. Blocking α1 and β1 integrins is proposed to interrupt collagen-dependent FAK activation and thereby suppress this feed-forward signaling circuit.

A major finding of this work is that collagen production in ascites is not simply a stromal phenomenon but can arise from a discrete population of tumor cells. Although cancer-associated fibroblasts are recognized as important sources of extracellular matrix proteins, single-cell sequencing studies have identified epithelial-to-mesenchymal transitioning (EMT) tumor cells with matrix-producing properties [33, 35, 36]. Our analyses reveal that Col3a1- and Col5a2-expressing murine tumor cells are enriched for EMT and Ras/MAPK signaling programs while being diminished by either genetic FAK deletion or pharmacologic FAK inhibition. These observations support the concept that epithelial plasticity and matrix production are tightly linked and that a subset of mesenchymal-like tumor cells may function as architects of the metastatic niche.

The identification of FAK as an upstream regulator of collagen production has important mechanistic implications [31]. FAK is activated downstream of β1-integrin engagement and collagen-rich matrices, and increased matrix stiffness further activates FAK signaling via mechano-transduction pathways [37]. This relationship may be reciprocal as FAK signaling promotes collagen production, which may subsequently reinforce integrin-mediated FAK activation. Such a feed-forward loop may provide a durable mechanism for sustaining tumorsphere survival under anchorage-independent conditions (Fig. 7). The observation that FAK deletion minimally affected two-dimensional proliferation while profoundly impairing three-dimensional growth further supports the idea that FAK-dependent matrix remodeling is particularly important in the unique biology of peritoneal dissemination.

The translational significance of these findings also extends beyond tumor growth. Increasing evidence indicates that fibrillar collagens are active regulators of immune suppression within the tumor microenvironment [18, 38]. Dense collagen matrices physically limit T-cell trafficking, can promote T-cell exhaustion, can influence macrophage polarization, and impair dendritic cell function; thereby collectively reduce anti-tumor immunity [12, 39–41]. In ovarian cancer, collagen-rich stroma has been associated with reduced cytotoxic lymphocyte infiltration and inferior patient outcome [11, 42]. Thus, FAK-dependent collagen production may contribute to mechanical tumorsphere signaling and to the establishment of an immunologically excluded microenvironment.

These observations suggest a broader model in which FAK functions as a central coordinator of both the physical and immunological components of the tumor microenvironment. In this model, FAK-dependent collagen production reinforces integrin signaling and tumorsphere survival while simultaneously establishing an immune-excluded environment. FAK inhibition would therefore be predicted to disrupt this pathogenic cycle at multiple levels: reducing collagen synthesis, diminishing mechano-transduction signaling, increasing accessibility of immune cells to tumor deposits, and promoting the development of CXCL13-driven tertiary lymphoid structures that support local B-cell and T-cell activation [26, 28]. The reciprocal relationship between matrix remodeling and immune organization may be particularly important in the peritoneal cavity, where metastatic spheroids exist near immune cells capable of orchestrating adaptive immune responses.

An important translational aspect of this study is the consistency of the findings across murine models, human genomic datasets, human OVCAR8 cells, and ex vivo patient samples. The identification of a conserved set of FAK-associated genes, including COL1A1, COL3A1, and COL5A2, together with the association between PTK2 amplification and elevated collagen expression, suggests that FAK-driven matrix programs are maintained across species and disease settings. Moreover, the observation that FAK inhibitor responses (in patient ascites) were independent of previous chemotherapy exposure indicates that recurrent tumors may remain dependent upon FAK despite extensive clinical treatment. This raises the possibility that changes in collagen biomarkers could serve as pharmacodynamic indicators of FAK pathway inhibition.

Several limitations should also be considered. Although our data strongly support tumor cells as a source of collagen within ascites, the relative contribution of fibroblasts, mesothelial cells, and other stromal populations during advanced disease remains to be fully defined. Likewise, the downstream signaling pathways by which FAK regulates collagen transcription requires further investigation. Future studies should also determine whether FAK-dependent matrix remodeling directly alters immune cell localization, tertiary lymphoid structure maturation, and anti-tumor immune function within the peritoneal cavity.

In summary, we identify a FAK-dependent ascites-associated mesenchymal tumor cell population that contributes to fibrillar collagen production. We propose that reciprocal signaling between α1β1 integrin-mediated FAK activation and collagen production creates a feed-forward circuit that supports metastatic dissemination, chemoresistance, and immune evasion [25, 26]. These data support a unifying model in which FAK orchestrates three interconnected processes within the recurrent ovarian cancer niche: maintenance of mesenchymal tumorsphere survival, establishment of a collagen-rich extracellular matrix, and tumor immune evasion. Therapeutically, disruption of this FAK-collagen-immune signaling axis may convert the collagen-rich, immune-resistant metastatic niche into one that is permissive for CXCL13-driven tertiary lymphoid structure formation and durable responses to combined chemotherapy and immunotherapy.

Materials and Methods

Mice

All animal procedures were performed under protocols approved by the University of California San Diego Institutional Animal Care and Use Committee (Protocol S07331). Female C57BL/6 mice (8–10 weeks old) were obtained from Charles River Laboratories. MISIIR-TAg transgenic strains expressing SV40 large T antigen downstream of the Müllerian inhibitory substance type II receptor promoter [44], together with conditional FAK floxed mice [45], have been described.

Reagents

The dTomato with luciferase lentiviral vector, pUltra-Chili-Luc, was a gift from Malcolm Moore (Addgene #48688). Small molecule FAK inhibitors IN10018 was from InxMed Inc., VS-4718 was from SelleckChem (#S7653) or FAK PROTAC degraders FAK PROTAC-1 (HY-119932) or FC-11 (HY-139344) were from MedChemExpress.

Cells and culture

Murine KMF ovarian carcinoma cells, the CRISPR-derived FAK-null KT13 clone, and GFP-FAK reconstituted derivatives were maintained under standard culture conditions in DMEM containing 10% fetal bovine serum (FBS) as described [46, 47]. MOVCAR cells were established from ascites collected from tumor-bearing MISIIR-TAg⁺;FAK^fl/fl mice and expanded under non-adherent conditions as described [26]. Human OVCAR8 ovarian carcinoma cells (RRID:CVCL_1629) were obtained from D. Connolly (Fox Chase Cancer Center) and were maintained in RPMI (R&D Systems Optima, S12450), 1 μg/mL insulin (Sigma-Aldrich, I9278), 1% penicillin–streptomycin (Gibco, 15140-122), and 1% L-glutamine (Gibco, 25030-081). Cell identity was confirmed by sequencing and protein analyses, and cultures were routinely screened for mycoplasma contamination. Anchorage-independent growth assays were carried out in poly-HEMA-coated plates, after which viable cells were quantified with an automated cell counter (ViCell XR, Beckman).

Protein extraction and immunoblotting

Cells were lysed in ice-cold RIPA buffer, resolved by SDS-PAGE, and immunoblots performed as described [26]. Primary antibodies used were anti-FAK (clone 4.47, Millipore Sigma 05–537), anti-PYK2 (clone 5E2, Cell Signaling, 3480), anti-phospho-FAK (pTyr397) (Thermo, 44-625G), anti-COL3A1 (Santa Cruz Bio, SC-271249) and β-actin (Proteintech, 60008-1-Ig).

Incucyte 2D cell growth

For each condition, 2000 cells/ well were plated into 96-well plate as 3 technical replicates. The confluency of the wells was determined every 2 hours using Incucyte SX5 live cell analysis system (Essen Biosciences) during 72-hour period. Confluency percentage was determined by automated confluency mask detection from 3 different well areas, including standard deviation between technical replicates. Experiments were repeated three times.

Immunohistochemistry

Mouse tumors were divided into thirds and either processed for protein lysates, fixed in formalin, or extracted for RNA analysis. Paraffin embedded tumors were sectioned, mounted onto glass slides, deparaffinized, rehydrated, and stained with Hematoxylin and Eosin (H&E) or Masson’s Trichrome Stain (Thermo, 50-168-9473) by standard procedures. High resolution digital scans were acquired (Aperio CS2 scanner) using Image Scope software (Leica Biosystems).

Cytospin

Tumor and immune cells from MOVCAR IP injected mice were isolated by peritoneal lavage at the final experimental day and subjected to removal of red blood cells with RBC lysis buffer (BioLegend, 420302). Gentle tissue processing was followed to preserve cell-cell contacts and spheroid structure of floating cell clusters. Cells were fixed with 4% PFA in PBS for 15 minutes at RT, thereafter, collected by centrifugation and resuspended in molten Histogel (Thermo Scientific., HG-4000-012) and allowed to solidify in biopsy mold (Fisherbrand, 22363553). The mold was transferred to Tissue Cassettes (Epredia^™, B851738WH), fixed in 10% formalin for 24h, washed in 70% ethanol, paraffin embedded and processed as tissue for immunohistochemistry analysis.

Proteomics

Samples were resuspended in a lysis buffer containing 50 mM triethylammonium bicarbonate (TEAB), 5% SDS, pH 7.5, then heated first at 95 °C for 10 minutes, followed by incubation at 60 °C for 1 hour in a thermoshaker at 1400 rpm. Next, the samples were mixed with the 7.5 M urea solution prepared in the lysis buffer, reduced with 20 mM DTT, and reheated at 95 °C for 10 minutes. After cooling to room temperature, the reduced proteins were alkylated by adding 40 mM iodoacetamide (IAA) and placed in the dark for 30 minutes. Following alkylation, the samples were acidified with 1.2% aqueous phosphoric acid and mixed with 6X volume of S-trap binding buffer (90% methanol, 100 mM TEAB, pH 7.1). The samples were then loaded onto the S-Trap micro column (ProtiFi) and centrifuged at 4000×g for 30 seconds, followed by 10 wash cycles with the S-trap binding buffer. Trypsin/Lys-C Mix (Promega, USA) was added at an enzyme-to-protein ratio of 1:25 (w/w) to digest the proteins trapped on the column at 47°C for 1 hour. The digested peptides were sequentially eluted using three solutions in the order: 50 mM TEAB (pH 8.0), 0.2% aqueous formic acid, and 50% acetonitrile/0.2% aqueous formic acid. The eluted peptides were first lyophilized in SpeedVac, then desalted with C18 tips (Empore) and finally dissolved in 0.1% formic acid for analysis by liquid chromatography-mass spectrometry.

Mass spectrometry and data analysis

The peptide mixture was separated by reverse-phase chromatography on an Easy-nLC1000 nanoflow HPLC system connected to a Q Exactive HF Orbitrap mass spectrometer (Thermo Fisher Scientific). The peptide mixture was loaded into a 15 cm C18 column (75 μm × 15 cm, ReproSil-Pur 3 μm 200 Å C18-AQ, Dr. Maisch HPLC GmbH, Ammerbuch-Entringen, Germany) and separated using a 50-minute gradient. The gradient profile was: 8–35% B in 40 minutes, 35-100% B in 2 minutes and held at 100% B for 8 minutes. Mobile phase A consisted of 0.1% formic acid, and mobile phase B was acetonitrile/water (95:5 (v/v)) with 0.1% formic acid. For protein quantitation, the data-independent acquisition (DIA) method was employed. Each MS cycle consisted of a full scan covering 366-1007 m/z at a resolution of 120,000 with an AGC target of 3e6 and a maximum injection time of 50 ms. This was followed by 40 MS2 scans, each with a 17 m/z isolation window at a resolution of 30,000, and an AGC target of 1e6.

Spectronaut software (version 16.0.220606.53000, Biognosys) was used to process the DIA data via the directDIA workflow. The DIA data were searched against the FASTA database containing reviewed and unreviewed mouse sequences of UniprotKB release 2022_03. Fixed modification of carbamidomethyl (C) and dynamic modifications of acetyl (Protein N-term) and oxidation (M, P, K) were employed, allowing a maximum of 5 variable modifications. Two missed cleavages were permitted, and peptide lengths between 7 and 52 amino acids were accepted. Precursors were filtered based on the identified Q-value, and the protein LFQ method was set to automatic. The quantification was performed at MS2 level, utilizing the automatic cross-run normalization strategy. For downstream analysis, the MS2 intensity data were log2-transformed, followed by minimum value imputation and median normalization. Matrisome proteins were categorized according to the mouse matrisome protein list available at https://sites.google.com/uic.edu/matrisome [48].

KMF mouse tumor growth

Luciferase/dTomato-labelled tumor cells were generated using pUltra-Chili-Luc vector (Addgene plasmid #48688) and sorted to normalize reporter expression before implantation. For intraperitoneal experiments, 5×10⁶ tumor cells were mixed with growth factor-reduced Matrigel and injected into female C57BL/6 mice. Animals were randomized five days later and treated with oral ifebemtinib (IN10018) or control vehicle according as described [26]. At study endpoints, peritoneal lavage was performed, erythrocytes were removed, and viable cells were processed for downstream flow cytometry or sequencing analyses. Orthotopic ovarian implantation experiments were performed by injecting MOVCAR cells into the ovarian bursa [49] followed by ultrasound monitoring (Vevo 2100, Visual Sonics) and tissue collection for histology and RNA analysis [26].

RNA sequencing

Total RNA from independent MOVCAR tumorsphere cultures (Corning, 3471) and orthotopic tumors was purified using a commercial RNA isolation kit (Invitrogen, 12183018A). Sequencing libraries were prepared for paired-end Illumina sequencing (Abclonal mRNA-seq Lib Prep Kit for Illumina^®). Quality-filtered reads were aligned to the mouse reference genome, and differential expression analysis was performed with standard RNA-seq workflows (NovaSeq X Plus platform, PE150). Functional enrichment analyses included pathway and gene set enrichment approaches, with statistical significance defined using multiple-testing correction as described [26].

Single-cell RNA sequencing and pathway enrichment

Tumor and immune cells were isolated by peritoneal lavage at experimental day 24 and subjected to single-cell 3′ RNA sequencing using the 10x Genomics Chromium Next GEM Single Cell 3′ HT v3.1 platform. Library preparation was performed according to the manufacturer’s protocol, and sequencing was conducted on an Illumina NovaSeq 6000 platform (~1 billion reads per sample). FASTQ files were processed using Cell Ranger (v6.1.2) for alignment to the mm10 mouse reference genome and UMI quantification as described [26]. Downstream analysis was performed in Python (v3.12.13) using Scanpy (v1.12.1). Cells were filtered based on standard quality control criteria (250–10,000 genes, 1,000–100,000 UMIs, <20% mitochondrial reads, <40% ribosomal reads), and doublets were removed using Scrublet. Data were normalized to 10,000 counts per cell, log-transformed, and highly variable genes (n = 3,000) were identified using the Seurat v3 method. Prior to PCA, total UMI counts and mitochondrial gene content were regressed out as covariates. PCA was followed by construction of a shared nearest-neighbor graph (15 neighbors, 30 PCs) and UMAP visualization.

Tumor cells were defined based on a composite epithelial score summing raw UMI counts across available markers (Epcam, Krt8, Krt18, Krt19, Pax8, Wt1, Msln). Cells in WT samples scoring above the 70th percentile of this distribution were classified as tumor-positive, a threshold selected based on enrichment of co-expression of ≥2 epithelial markers and concordant localization within epithelial UMAP clusters in WT cells. Tumor cells were further stratified by Col3a1 and Col5a2 expression using raw counts (≥1 UMI = positive). To assess stromal contamination of gene-positive populations, mean expression of fibroblast markers (Col1a1, Col1a2, Dcn, Pdgfra, Fap) and epithelial markers (Epcam, Krt8, Krt18, Krt19, Pax8) was compared between positive and negative tumor cell subsets. Differential expression between Col3a1⁺/⁻ and Col5a2⁺/⁻ tumor cells was performed using the Wilcoxon rank-sum test with Benjamini-Hochberg correction across all expressed genes. Cross-condition differences in gene expression levels were assessed using Kruskal-Wallis omnibus tests followed by pairwise Mann–Whitney U tests with Benjamini-Hochberg correction.

Reactome pathway enrichment analysis for Col3a1 and Col5a2-associated genes was performed using WebGestalt [50]. Validation was conducted using the ovarian serous cystadenocarcinoma TCGA cohort (n = 309) from the GSCA platform [51]. Pathway activity scores for key cancer-related pathways (EMT, RAS/MAPK, mTOR, RTK, PI3K/AKT, DNA damage, hormone AR/ER, cell cycle, and apoptosis) were computed as the difference between positive and negative regulatory components [52]. Samples were stratified by median gene expression, and pathway differences between groups were assessed using Student’s t-test with FDR correction (≤0.05). Genes were classified as activating or inhibitory based on the direction of pathway activity changes [53]. All analyses were performed using NumPy, pandas, and SciPy, and visualizations were generated using Matplotlib and seaborn and Chiplot https://www.chiplot.online/.

Murine-human transcriptomic comparisons

HGSOC patient RNA sequencing data was obtained from The Cancer Genome Atlas (TCGA) Genomic Data Commons (GDC) high grade ovarian serous cohort [54]. Gene expression matrices were downloaded from publicly available repositories and processed using Python (v3.8) and R (v4.0). Patients were stratified into PTK2-low and PTK2-high cohorts based on PTK2 mRNA expression, defined as the lower quartile (≤25th percentile) and upper quartile (≥75th percentile) of PTK2 transcript abundance, respectively. Differential expression analyses were subsequently performed to compare PTK2 mRNA levels between the two cohorts. Transcriptomic datasets from MOVCAR (herein) and KMF murine ovarian cancer models [47] were integrated with human transcriptomic analyses to identify mRNAs co-expressed with high levels of PTK2. Samples with PTK2 expression above the cohort median were classified as PTK2-high, whereas samples below the median were classified as PTK2-low. DESeq2 was used to calculate differential gene expression (DEG) gene-wise dispersion estimates and the negative binomial framework. DEG significance was determined using Wald statistics. Multiple-testing correction was performed using the Benjamini–Hochberg method. Genes with adjusted P values below the predefined significance threshold were retained for downstream analyses. DEGs were ranked according to adjusted P value and log2 fold change. Candidate genes were prioritized based on expression abundance and magnitude of differential expression. Genes were required to exhibit a normalized mean expression value (baseMean) greater than 100 and an absolute log2 fold change greater than 0.25. Genes satisfying all filtering criteria were designated as high-confidence PTK2-associated candidate dependencies. Expression values were transformed to gene-wise Z scores to facilitate comparison across samples. DEGs identified in the TCGA cohort were compared with independently generated gene sets derived from KMF and MOVCAR tumors and visualized by Venn diagram analysis.

Human samples

Ascites samples were obtained from consenting patients with newly diagnosed or recurrent ovarian cancer under a UCSD Institutional Review Board-approved protocol (110805). Following centrifugation, red blood cells were removed by lysis (BioLegend, 420302), and viable cells were recovered for molecular and cellular analyses as described [26].

Purification of ascites tumor cells

Tumor cells were enriched from malignant ascites by discontinuous Percoll density-gradient centrifugation. Cell suspensions were layered over sequential Percoll solutions (70%, 45%, and 25%), centrifuged (800 x g for 30 minutes) without braking, and the tumor-enriched interface was collected, washed, and pelleted before use in downstream experiments.

Flow cytometry

Ascites-derived cells were resuspended in flow cytometry buffer consisting of phosphate-buffered saline (PBS) supplemented with 0.5 mM EDTA and 2% BSA. Cells were filtered through a 70 μm strainer to remove aggregates and maintain single-cell suspensions. Prior to staining, a cell viability marker was used (LIVE/DEAD Aqua; Thermo, L34957) and then samples were processed with fluorochrome-conjugated antibodies as described [26]. Tumor cells were identified by exclusion of immune lineage cells, defined as CD45-negative populations using anti-human CD45 (BV605, BioLegend, 304041). Flow cytometric acquisition was performed on a Fortessa-X20 multicolor flow cytometer, compensation was applied using single-stained controls, a sequential gating strategy was to ensure high-confidence identification of viable single tumor cells, and acquisition data was analyzed using FlowJo software (version 10).

Ex vivo patient ascites analyses

Crude patient ascites cell suspensions, which contain both tumor and immune cells, were seeded (2 × 10⁶ cells per well) in poly-HEMA-coated plates (Corning, 3471) in RPMI 1640 (Gibco, 11875-093) supplemented with 10% FBS (Gibco, 16000-044), 1 μg/mL insulin (Sigma-Aldrich, I9278), 1% penicillin–streptomycin (Gibco, 15140-122), 1% L-glutamine (Gibco, 25030-081), 0.25 mM L-ascorbate (Sigma-Aldrich, A4544), and 0.45 mM L-ascorbate phosphate (Sigma-Aldrich, A8960). Cells were equilibrated for 24 h prior to treatment. Cells were then treated with 1 μM FAK inhibitor (IN10018 or VS-4718), or 1 μM FAK PROTAC degraders (FAK-PROTAC-1, or FC-11) or DMSO for 48 h under standard culture conditions. At endpoint, cells were collected using a scraper (Corning, No. 3010) and processed in parallel for RNA and protein analyses.

Quantitative RT-PCR

Total RNA was extracted using the Invitrogen PureLink RNA Mini Kit (Thermo, 12183018A) according to the manufacturer’s instructions. RNA concentration and purity were assessed prior to reverse transcription. Complementary DNA (cDNA) was synthesized using the iScript RT Supermix (Bio-Rad, 1708841). Quantitative PCR was performed using SsoAdvanced Universal SYBR Green Supermix (Bio-Rad, 1725271) with 100 ng cDNA per technical replicate and 250 nM forward and reverse primer concentrations. Relative gene expression was calculated using the ΔΔCT method, normalizing to β-actin or GAPDH and expressed as log2 fold change relative to DMSO-treated controls. Data presented is from at least 3 independent experiments. Primer sequences used are listed in Supplemental Table 2.

Kaplan Meier

Expression array data were evaluated using the Kaplan-Meier Plotter (www.kmplot.com/ovar). The PTK2 [208820_at], COL3A1 [211161_s_at], and COL5A2 [221729_at] probe sets were used. Query parameters were overall survival, auto-select best cutoff, and a follow up threshold of 10 years. Restriction analyses were all tumor stages, serous histology, all tumor grades, TP53 mutation, debulk (all), and no selection for surgery or chemotherapy treatments. A total of 1207 patient samples were analyzed.

Human PTK2 gene gains and collagen

Transcriptomic and copy-number data from an independent HGSOC cohort was obtained through the National Cancer Institute Genomic Data Commons (GDC; 2025 release), which provides uniformly processed sequencing datasets generated using standardized analytical pipelines. To independently validate PTK2-associated collagen expression, mRNA expression levels of COL1A1 and COL3A1 were evaluated in relation to PTK2 copy-number status, and patients were stratified according to PTK2 diploid or amplified tumors for downstream comparative analyses.

OVCAR8 tumor spheroids

Multicellular tumor spheroids were generated using agarose 3D Petri Dish^® micro-molds (MicroTissues Inc.). Micro-molds (#12-81) were sterilized by autoclaving, and 2% (w/v) agarose was prepared in sterile 0.9% NaCl, heated until completely dissolved, and cooled to approximately 60–70°C. Molten agarose was pipetted into the micro-molds and allowed to solidify. The resulting agarose 3D Petri Dishes were transferred to tissue-culture plates and equilibrated twice with OVCAR8 complete culture medium for ≥15 min per incubation. To set up spheroids, OVCAR8 cells were harvested by 0.25% trypsin-EDTA (Gibco, #25200-056) treatment for 5 minutes, inactivated by complete media, washed PBS, enumerated as viable single cells (ViCell XR), and then suspended in 5 ml of PromoCell 3D Tumorsphere Medium XF (C-28075) supplemented with Supplement Mix (C-39870, a serum-free, xeno-free medium) with 1% penicellin-streptomycin, 0.25 mmol/L L-ascorbate, and 0.45 mmol/L L-ascorbate phosphate. 650,00 cells in 190 μL were plated into MicroTissue 12 well pod (#12-81) wells forming 81 distinct spheroids of ~8000 cells/each within 2 ml to total media within a 6 well culture plate. After 24 h, DMSO, 1 μM IN10018, 1 μM FC-11, or 10 ug/mL control IgG1, anti-α1 (Invitrogen, clone TS2/7), anti-α3 (Millipore, clone P1B5), anti-αV (Abcam, clone 17E6), or anti-β1 (Abcam, clone P5D2) blocking antibodies. At 24 and 48 h, brightfield images (Echo Revolve) were acquired of spheroids within agar wells, and total spheroids were harvested by agar mold inversion, centrifugation, and either RNA or protein isolation as described above. Tumor spheroid size at 48 h was quantified from 2D brightfield images using FIJI-ImageJ software. Images were calibrated using the 200-μm scale bar, and the diameter of each spheroid was measured using the line measurement tool. The projected cross-sectional area (A) was calculated from the measured diameter assuming a circular spheroid profile, A = π (d/2)². Experiments were repeated three times.

Sex as a biological variable

Experiments were performed using female animals and female-derived biological specimens as appropriate for ovarian cancer models. Public genomic analyses used de-identified ovarian cancer datasets (https://www.cancer.gov/ccg/research/genome-sequencing/tcga).

Statistics

Experiments were performed at least two independent times. Statistical analyses were performed in Prism v10 (GraphPad Software). For experimental groups of three or more, statistical significance was calculated based on one-way ANOVA with Tukey’s multiple comparison test. Unpaired T test was used to determine statistical difference between the means from two different samples. P values <0.05 were considered significant. Differential expression analyses were performed using DESeq2. Significance was determined using two-sided Wald tests with Benjamini–Hochberg correction for multiple hypothesis testing. For comparisons between PTK2 diploid and amplified tumors, log-transformed expression values were analyzed using two-sided Welch’s t-tests. Adjusted P values were used for all transcriptomic analyses unless otherwise specified. Statistical analyses were performed using R (v4.0) and Python (v3.8).

Supplementary Material

Supplemental Data 3

Supplemental Data 3: Enrichment of Reactome categories identified by single cell sequencing associated with GFP-FAK WT re-expressing versus FAK^−/− KMF ascites-associated tumor cells in vivo.

Supplemental Data 1

Supplemental Data 1: Listing of matrisome mRNA categories and log2 fold-change from FAK^−/− and GFP-FAK WT MOVCAR RNA sequencing of 3D tumorspheres.

Supplemental Data 2

Supplemental Data 2: Listing of matrisome protein changes, matrisome categories, and log2 fold-change from FAK^−/− and GFP-FAK WT MOVCAR 3D tumorspheres by mass spectrometry.

Data Availability

RNA sequencing FASTQ files have been deposited to NCBI SRA database under the accession number PRJNA1479482.

The proteomics data based on mass spectrometry in this study have been deposited to the ProteomeXchange Consortium via the PRIDE [43] partner repository at the European Bioinformatics Institute (http://www.ebi.ac.uk/pride/) with the dataset identifier PXD079775.