Loss of Claudin-4 reduces DNA damage repair and increases sensitivity to PARP inhibitors

Abstract

High-grade serous ovarian cancer is the deadliest gynecologic malignancy due to progression to resistant disease. Claudin-4 is classically defined as a tight junction protein and is often associated with epithelial cancers. Claudin-4 is aberrantly expressed in nearly 70% of all ovarian cancer tumors and conveys a worse overall prognosis. Elevated claudin-4 expression correlates to increased DNA repair activity and resistance to DNA damaging agents. Poly(ADP) polymerase (PARP) inhibitors are emerging as an effective therapeutic option for patients with ovarian cancer and function by promoting DNA damage. The study examines the relationship between claudin-4 expression and the response to PARP inhibitors using both genetic and pharmacologic inhibition of claudin-4 in in vitro and ex vivo models of ovarian cancer to examine DNA repair markers and functional activity. Genetic inhibition of claudin-4 results in the downregulation of several DNA damage repair effectors, including 53BP1 and XRCC1. Claudin-4 knockdown did not change homology-directed repair but inhibited non-homologous end-joining and reduced 53BP1 foci formation. In 15 primary ovarian cancer tumors, higher claudin-4 expression significantly correlated to a dampened PARP inhibitor-mediated anti-proliferation response. Further, claudin-4 inhibition in high claudin-4 tumors sensitized tumor sections to PARP inhibition. These data highlight that claudin-4 expression in ovarian cancer tumors could serve as both a marker of PARP inhibitor response and a therapeutic target to improve PARP inhibitor response.

INTRODUCTION

High grade serous ovarian carcinoma (HGSOC) is the most common epithelial derived ovarian cancer and is the deadliest. Several factors contribute to HGSOC-related deaths including inadequate early detection, diagnosis at advanced stage disease, and the development of acquired therapy resistance. In the last six years, the development of poly(ADP) ribose polymerase (PARP) inhibitors (PARPi) has led to a significant shift in the clinical care of patients with ovarian cancer.

PARPi exploit deficiencies in DNA damage repair pathways. Approximately, 50% of ovarian cancers have some degree of Homologous Recombination (HR) DNA repair deficiency (HRD); these deficiencies are most common in tumors with mutations in BRCA1/2 [1]. BRCA1/2 proteins are directly involved in HR DNA repair and DNA damage repair capacity is an important biomarker of PARPi responsiveness. PARPi prevent single-strand DNA break repair and in the context of HR repair deficiencies, PARPi facilitate the accumulation of replication induced DNA double strand breaks (DSBs). These effects culminate in catastrophic mitotic failure and cell death. Clinical trials for olaparib and niraparib have demonstrated that the PARPi response can occur in tumors without mutations in BRCA1/2 genes [2, 3]. For instance, niraparib significantly extended progression-free survival in patients without germline BRCA1/2 mutations compared to placebo, 9.3 versus 3.9 months, respectively (hazard ratio, 0.45; 95% confidence interval, 0.34 to 0.61) [4]. In patients with HGSOC, the major clinical challenges in the use of PARPi include: (1) identification of patients likely to respond, (2) sensitizing non-BRCA1/2 mutated HGSOC tumors to PARPi, and (3) addressing resistance to PARPi.

Claudin-4 (CLDN4) is expressed in approximately 70% of all human HGSOC tumors. It is highly expressed in ovarian tumor cells that: 1) are more resistant to cisplatin/carboplatin and paclitaxel [5, 6] 2) have increased migratory potential [7, 8] and 3) display stem-like properties [6, 9]. Previously, we reported that the loss of claudin-4 expression led to aberrant cell cycle progression and G2/M arrest [10]. We previously designed a small claudin mimic peptide (CMP) that interferes with the DFYNP sequence in the second extracellular loop of claudin-4 [11]. CMP is used in the D-amino acid conformation to increase peptide stability. The peptide disrupts claudin-4 activity; in addition CMP increased tumor cell response to a potent apoptotic inducer, staurosporine, and inhibited tumor cell migration [7].

Patients with HGSOC tumors expressing claudin-4 protein have poorer survival compared to patients with tumors lacking claudin-4 [12]. Analyses of HGSOC tumors in The Cancer Genome Atlas (TCGA) revealed a tendency toward mutual exclusion of CLDN4 amplification/overexpression and BRCA1/2 mutation/downregulation [1], suggesting that claudin-4 may serve as potential marker of PARP inhibitor response. Notably, in breast cancer, the “claudin-low” subtype is often more responsive to PARP inhibitors [13]. For this reason, we have assessed the role of claudin-4 in PARP inhibitor response in models of HGSOC.

In particular, here we have extended our previous claudin-4 experiments to examine its effect on the PARPi response as well as to define claudin-4-mediated mechanisms of DNA repair. We discovered that in HGSOC cells without BRCA-mutations or HRD loss that knocking down claudin-4 expression increased PARPi sensitivity. Comparing the transcriptome of claudin-4 high and low expressing tumors, in claudin-4 high tumors we found enrichments for DNA repair, DNA replication, and cell cycle regulatory pathways. Examination of Dependency Mapping [14] across multiple cancer types revealed that increased CLDN4 expression correlated with a blunted response to three different PARPi (olaparib, rucaparib, niraparib). Examining a reverse phase protein array of claudin-4 knockdown HGSOC cells, we observed significant down regulation of several DNA repair proteins (e.g., XRCC1 and 53BP1). Further, claudin-4 knockdown did not inhibit HR repair, but did significantly attenuate non-homologous end joining (NHEJ) DNA repair. In ex vivo models of primary ovarian cancer tumors, combined claudin-4 inhibition via CMP and PARP inhibition promoted an anti-proliferation and apoptotic response. Thus, targeting claudin-4 expression in combination with PARPi in HGSOC tumors represents a novel therapeutic strategy.

METHODS

Cell culture:

Human-derived OVCAR3 (RRID:CVCL_0465), OVCAR4 (RRID:CVCL_1627), OVCAR8 (RRID:CVCL_1629), OVCAR10 (RRID:CVCL_4377), OVCA432 (RRID:CVCL_3769), OVCAR2 (RRID:CVCL_3941), OV167 (RRID:CVCL_A402), CaOV3 (RRID:CVCL_0201), OVCA429 (RRID:CVCL_3936), DOV13 (RRID:CVCL_6774), IGROV (RRID:CVCL_1304), 41M (RRID:CVCL_Y037), TOV-112D (RRID:CVCL_3612), TYKNU (RRID:CVCL_1776), and SNU251 (RRID:CVCL_5040) were obtained from The Gynecologic Tumor and Fluid Bank at the University of Colorado. UWB1.289 (RRID:CVCL_B079) were obtained from American Type Culture Collection. Cells were cultured in RPMI-1640 medium (Gibco, Thermo Fisher Scientific, Cat. #11875) plus 10% heat-inactivated fetal bovine serum (Phoenix Scientific, Cat. # PS-100, Lot # 20055-01-01) and 1% penicillin/streptomycin (Corning, Cat. #30-002-CI) at 37°C and 5% CO₂. UWB1.289 were cultured in 50% RPMI, 50% MEGM + supplement (Lonza, Cat # Catalog #: CC-4136), 3% heat-inactivated fetal bovine serum, 1% penicillin/streptomycin. Cell lines were cultured for a maximum of two months or 20 passages, which ever came first. Claudin-4 expression and knockdown were confirmed upon thawing a fresh vial of cells. 293FT (RRID:CVCL_6911) cells were obtained by the University of Colorado Functional Genomic shared resource. All cell lines were authenticated at the beginning of this study by short tandem repeat profiling at the University of Arizona. Cells are monthly tested for mycoplasma (LookOut, Sigma # MP0035-1KT) and were tested as recently as July 20^th, 2021.

Immunoblot analysis:

To analyze levels of claudin-4 protein expression, tumor cells were scraped from culture plates in the presence of lysis buffer (30 mM Tris HCl pH7.4, 150 mM NaCl, 1% TritonX-100, 10% glycerol, 2 mM EDTA, 0.57 mM PMSF, 1X cOmplete™ Protease Inhibitor Cocktail), placed on a shaker for 10 minutes and spun at 13,000 rpm for 10 minutes. Protein was separated by SDS-PAGE and transferred to PVDF membrane using the TransBlot Turbo (BioRad). Membranes were blocked with Intercept Blocking Buffer (LI-COR, #927-60001) for 1 hour at room temperature. Mouse anti-human claudin-4 (Thermo Fisher Scientific Cat# 32-9400, RRID:AB_2533096, 1:500), rabbit anti-human β-actin (Abcam Cat# ab6276, RRID:AB_2223210,1:10,000) , or mouse anti-GAPDH (Thermo Fisher Scientific Cat# MA5-15738, RRID:AB_10977387,1:2500) primary antibody incubation was performed overnight at 4 °C. Membranes were washed 3 times for 5 minutes each in TBST (50 mM Tris pH 7.5, 150 mM NaCl, 0.1% Tween-20), followed by secondary antibodies for one hour at room temperature. Membranes were washed again 3 times for 5 minutes each in TBST. For fluorescent detection, bands were visualized using the LI-COR Odyssey Imaging System.

Reverse-transcriptase quantitative PCR (RT-qPCR):

RNA was isolated from cells using the RNeasy Plus Mini Kit (Qiagen). mRNA expression was determined using SYBR green Luna One Step RT-qPCR Kit (New England BioLabs) on a C1000 Touch (Bio-Rad) or QuantStudio 6 (Applied Biosystems) thermocycler. Expression was quantified by the ΔΔCt method using target-specific and control primers. β-2-microglobulin (B2M) used as an internal control. Primer sequences: B2M Fwd 5’-GGCATTCCTGAAGCTGACA-3, Rev 5’-CTTCAATGTCGGATGGATGAAAC-3’, CLDN4 Fwd 5’-GCCTTACTCCGCCAAGTATT-3’, Rev 5- AGGGAAGAACAAAGCAGAGAG-3’ [15].

Immunofluorescence:

Cells grown on Type I collagen (Sigma, Cat. # C3867) coated glass slides were fixed with 10% phosphate buffered formalin (Sigma, Cat. # HT501128) at room temperature (RT) for 15 min. Cells were washed twice with phosphate buffered saline (PBS), permeabilized with 0.5% Triton X-100 (IBI Scientific, Cat. #IB07100) for 5 minutes at RT, washed twice with PBS, and blocked with 2% bovine serum albumin (Sigma, Cat. # A2153) for 1 hour at RT before application of primary antibody to claudin-4 (Thermo Fisher Scientific Cat# 32-9400, RRID:AB_2533096, 1:100), XRCC1 (Cell Signaling Technology Cat# 2735, RRID:AB_2218471, 1:100) 53BP1 (Cell Signaling Technology Cat# 4937, RRID:AB_10694558, 1:100) and γH2ax (Millipore Cat# 05-636, RRID:AB_309864, 1:400) overnight at 4°C. Cells were then washed with PBS five times before application of secondary antibody Alexa Fluor 488-conjugated donkey anti-mouse secondary antibody (Thermo Fisher Scientific Cat# A-21202, RRID:AB_141607, 1:1000) or CY3 conjugated donkey anti-rabbit secondary antibody (Jackson ImmunoResearch Labs Cat# 711-165-152, RRID:AB_2307443,1:100); 5 μg/ml 4’,6-diamidina-2-phenylindole (DAPI; MilliporeSigma, Cat #D8417) was applied for 45 minutes at room temperature followed by five washings with PBS. After removal of PBS, o-phenylenediamine dihydrochloride (20 mg/ml; OPDA) in 1M Tris, pH 8.5 was added to the slides to preserve fluorescence and coverslip mounted. Imaging was performed in the Advanced Light Microscopy Core in the NeuroTechnology Center at University of Colorado Anschutz Medical Campus using the 3I Marianas inverted spinning disk confocal microscope. Images were analyzed using SlideBook software (Intelligent Imaging Innovations, Inc., Denver, CO, USA) [10].

Lentiviral shRNA knockdown:

Lentiviral knockdown was completed as described [10]. Briefly, lentivirus was packaged as previously described in [16]. Using 293FT cells using 3^rd generation packaging plasmids (Virapower, LifeTechnologies, Cat. #K4950) with polyethylenimine (PEI, MilliporeSigma, Cat. #408727) transfection in a 1:3 DNA:PEI ratio. Culture supernatant was harvested at 48-72 h post-transfection and processed through 0.45 μM filters. Viruses encoded a puromycin resistance gene. Transduced OVCAR3 and OVCA429 cells were selected in 1 μg/mL puromycin. Claudin-4 shRNA (TRC#: TRCN0000116627 or TRCN0000116628) or control shRNA (SHC001, pLKO.1-puro Empty Vector; RRID:Addgene_8453) lentiviral suspension (Sigma-Aldrich MISSION® shRNA, University of Colorado Functional Genomics Facility, Aurora, CO, USA) was added to the cells and incubated overnight at 37°C. Fresh medium was added to remove lentivirus and cells were allowed to recover for 24 hours before being treated with 0.5 μg/ml puromycin for selection and expansion of transduced cells. Western blot analysis was performed to confirm loss of claudin-4 expression.

Colony formation assay:

As previously described in [15]. Olaparib (#S1060) and Rucaparib (#S4948) were obtained from SelleckChem. Cell lines were seeded in 24-well plates and treated with increasing doses of olaparib. Media and PARP inhibitor were changed every three days for 12 days or until control wells were confluent, whichever occurred first. Colonies were washed twice with PBS, then incubated in fixative (10% methanol/10% acetic acid in PBS). Fixed colonies were stained with 0.4% crystal violet in PBS. After imaging, crystal violet was dissolved in fixative and absorbance was measured at 570 nm using a Molecular Devices SpectraMax M2e plate reader.

Ovarian cancer dataset analysis:

Publicly available cancer databases: Ovarian cancer single cell RNA-sequencing [17] accessed April 9^th, 2021 via http://blueprint.lambrechtslab.org/; Claudin-4 expression correlation with cisplatin, olaparib, rucaparib, and niraparib sensitivity was accessed on March 22^nd 2021 via https://depmap.org/portal/ [14, 18]. Claudin-4 low and high (based on median RNA expression) RNA-sequencing and mutational count analyses from The Cancer Genome Atlas Ovarian Serous Cystadenocarcinoma Firehose Legacy accessed on Dec. 20^th 2020 via https://cbioportal.org [1, 19, 20]. Claudin-4 Molecular Score from CanSar Black [21] accessed May 5^th, 2021 via https://cansarblack.icr.ac.uk/; Mass spectrometry data of ovarian cancer cells lines derived from [22], PrognoScan [23] of CLDN4 assessing overall survival accessed on May 5^th, 2021 via http://dna00.bio.kyutech.ac.jp/PrognoScan/. Dependency Mapping to assess claudin-4 mRNA expression and drug sensitivity (Area under the curve from Genomics of Drug Sensitivity in Cancer version 2) accessed on Dec. 9th, 2021 [14].

FDA-approved oncology drug screen:

OVCAR3 shCtrl and OVCAR3 shCLDN4#1 cells were plated (2000 cells/well) into 96 well plates. Following 24 hrs, FDA-approved oncology drugs panel were transferred the 96 well plate using a Scinomix disposable pin tool to final concentrations of either 1 μM or 10 μM final concentrations. After 72 hrs incubation with drugs, plates were fixed and analyzed per the Colony Formation Assay.

Two-plasmid functional DNA repair assay:

Two-plasmid functional assays were performed to assess distal non-homologous end joining (NHEJ) and homology directed repair (HDR) [24]. Cells were transfected with pimEJ5GFP (NHEJ, RRID:Addgene_44026) or pDRGFP (HDR, RRID:Addgene_26475). Cells were transfected with I-SceI (RRID:Addgene_26477), which introduces DNA double-strand breaks in the plasmids. After 72 hours, cells were collected and examined via a flow cytometer to quantify GFP positive cells. pimEJ5GFP was a gift from Jeremy Stark (Addgene # 44026). pDRGFP and pCBASceI were gifts from Maria Jasin (Addgene # 26475/26477).

Ex vivo culture.

Primary ovarian cancer tumor samples were obtained from the Gynecologic and Tumor Fluid Bank at the University of Colorado (COMIRB #07-935). The protocol was approved and conducted in accordance with the University of Colorado (Aurora, CO) ethical policy. Tumors were sectioned at 300 micron thickness using a Krumdieck Tissue Slicer. Claudin-mimetic peptide CMP (NH2-GDGYNPG-OH, D-amino acid conformation) was obtained from the University of Colorado Protein and Peptide Chemistry Core [11]. Tumor sections were cultured with vehicle control, olaparib (1 μM), CMP (400 μM) or in combination olaparib/CMP (1 μM/400 μM, respectively) for 72 hours. Following a 72-hour incubation, tissue sections were fixed in 10% buffered formalin, paraffin embedded and sectioned. Tumor sections were analyzed by immunohistochemistry.

Immunohistochemistry:

Paraffin embedded sections were dewaxed in xylene twice for 5 min and rehydrated in graded alcohols at 100%, 90%, 70%, and 30% for 3 min each and then in PBS for 5 min. Antigen retrieval was performed using Antigen Unmasking Solution #H-3300, Vector Laboratories, Burlingame, CA, placing the slides in the solution for 10 rounds of heating to boiling for 10 seconds followed by cooling for 45 seconds. Sections were treated with 0.2% glycine in PBS for 30 min, rinsed twice with PBS, and then blocked with 10% normal donkey serum and 0.1 mg/ml saponin in PBS for 60 min. Sections were incubated at 4 C overnight with antibody to claudin-4 (Thermo Fisher Scientific Cat# 32-9400, RRID:AB_2533096, 1:100). Sections were rinsed five times in PBS and then incubated with Alexa Fluor 488 conjugated donkey anti mouse secondary antibody (Jackson ImmunoResearch Labs Cat# 715-545-150, RRID:AB_2340846) and 5 μg/ml 4’,6-diamidina-2-phenylindole (DAPI, Sigma) for 45 minutes at RT followed by five washes with PBS. OPDA (20 mg/ml, o-phenylenediamine dihydrochloride (Sigma) in 1M Tris, pH 8.5 and 10% glycerol) was applied to slides for preservation of fluorescence and coverslip mounted. Ki67 (Thermo Fisher Scientific Cat# RM-9106, RRID:AB_2341197) and cleaved caspase 3 (Cell Signaling Technology Cat# 9661, RRID:AB_2341188) were completed by the Histopathology Shared Resource as described in [25]. Tumor sections were also stained for the Mullerïan marker, PAX8 (Proteintech Cat# 10336-1-AP, RRID:AB_2236705, 1:100). Treatment groups were de-identified and an H-scores were calculated for each tumor section as previously described [26]. Imaging was performed in the Advanced Light Microscopy Core part of NeuroTechnology Center at University of Colorado Anschutz Medical Campus using the 3I Marianas inverted spinning disk confocal microscope. Images were analyzed using SlideBook software (Intelligent Imaging Innovations Inc., Denver, CO, USA)

Statistics:

Graphs and statistical analysis were completed in Prism GraphPad (v9). Data are presented as mean ± standard error of the mean (s.e.m.). An unpaired Student t test was used for statistical comparison between control and treatment groups. Dose response curves were calculated using non-linear regression via the log(inhibitor) vs. response equation. A one-way ANOVA was used to determine variance among multiple gestational groups, with a Benjamini-Hochberg Multiple Comparison post-test to determine significance between individual groups. A p value of < 0.05 was considered significant.

Data Availability Statement:

Data are available upon request from the corresponding author

RESULTS

Claudin-4 expression is correlated with cell cycle regulation and DNA repair in high grade serous epithelial ovarian tumors.

Based on the combined molecular score of claudin-4/CLDN4 across a variety of epithelial cancers, the molecule is often associated with female cancers, specifically ovarian cancer (Figure S1A). Further, elevated claudin-4 expression portends poor prognosis in three of six primary ovarian cancer tumor datasets (Figure S1B). For this reason, to begin to identify the role of claudin-4 in HGSOC tumors, TCGA tumors were stratified to “high” or “low” claudin-4 expression based on median expression (Figure 1A). Comparing the transcriptome of the tumors based on claudin-4 expression we identified 1,582 differentially expressed transcripts (FDR<0.05, Figure 1B and Table S1). Comparing these 1,582 genes to the Reactome [27] gene sets, there was a significant enrichment of genes involved in cell cycle, DNA repair and DNA replication pathways (Red dots, Figure 1C), including RNA Polymerase II Subunit J (POLR2J), BRCA1 Interacting Protein C-Terminal Helicase 1 (BRIP1), Nucleoporin 160 (NUP160) and Protein Kinase, DNA-Activated, Catalytic Subunit (PRKDC) (Figure 1D). These data are consistent with previous reports that loss of claudin-4 expression promotes cell cycle arrest [10]; however, a claudin-4 connection to DNA damage response had not been established.

Figure 1: Elevated claudin-4 expression correlates to DNA repair and response to DNA damaging agents.

A) Patient tumors within the TCGA (Firehose Legacy) were stratified based on the median expression of CLDN4 expression (Low, N=154 and High, N=153). B) Volcano plot of 1582 differentially regulated genes between low CLDN4 expressing tumors and high CLDN4 expressing tumor. C) Reactome pathway analysis of 1582 genes. D) Representative genes from Reactome DNA repair, replication and cycle pathways. E) Drug sensitivity of DNA damaging agents based on CLDN4 expression. Error bars, SEM. Statistical test, unpaired t-test with Benjamini-Hochberg multicomparison correction, ***p<0.001, ****p<0.0001.

DNA damaging agents are standard of care treatments for HGSOC, including platinum--based chemotherapies and Poly(ADP) Ribose Polymerase inhibitors (PARPi). Using Dependency Mapping (DepMap) [18], we examined the correlations of claudin-4 mRNA expression with sensitivity to cisplatin and three FDA-approved PARPi: olaparib, rucaparib, and niraparib. Examining over 400 different cell lines there was a significant positive correlation between claudin-4 mRNA expression and decreased sensitivity measured as Area Under the Curve (AUC) to cisplatin, olaparib, niraparib, and rucaparib (Figure 1E). Specifically, the higher expression of claudin-4 mRNA significantly correlated to drug sensitivity AUC. Thus, cells with low claudin-4 expression have increased sensitivity to DNA damaging agents. These data are consistent with the TCGA RNA-seq analysis and suggest a novel and previously undefined claudin-4/DNA repair axis.

Loss of claudin-4 expression increases sensitivity to PARPi

To more directly establish the role of claudin-4 in conveying sensitivity to anti-cancer therapies we examined claudin-4 protein expression in 13 HGSOC cell lines (Figure 2A and S2A). We detailed the BRCA-status of each of the cell lines (Table S2). Appreciating a need to improve options for patients without BRCA-mutations the cell lines examined are BRCA1/2-wildtype, essential in considering sensitivity to PARPi. In the four HGSOC cell lines shown in Fig 2A, we measured PARPi sensitivity using either olaparib or rucaparib (Figure 2B–D and S2B). Olaparib and rucaparib are both PARPi and trap PARP onto DNA at similar rates, but they have several distinct features including, rucaparib’s ability to more broadly target the family of PARP enzymes [28]. We observed that the 50% inhibitory concentration (IC50) of olaparib did not correlate with claudin-4 expression. In contrast, the IC50 of rucaparib aligned with the DepMap data showing the lower claudin-4 expressing cells had increased sensitivity. As nearly 70% of HGSOC tumors express claudin-4, we next wanted to determine if loss of claudin-4 expression improved response to PARPi. Using a control shRNA or two independent small-hairpin RNAs (shRNA) specific for CLDN4, we knocked down both claudin-4 mRNA and protein expression in both OVCAR3 and OVCA429 cell lines (Figure 2E and S2C). Next, the shControl and claudin-4 knockdown cells were examined for olaparib and rucaparib response. The loss of claudin-4 significantly reduced the IC50 of both olaparib and rucaparib compared to the shControl cells (Figure 2F–G and S2D–E). As olaparib is more widely used in the clinic compared to rucaparib, we used olaparib for the remaining experiments. These data directly demonstrate that reducing claudin-4 expression increases sensitivity to PARPi.

Figure 2: PARP inhibitor response in HGSOC cell lines with varying Claudin-4 expression.

A) Claudin-4 protein expression was measured via immunoblot in a panel of HGSOC cell lines. Loading control, β-actin. Olaparib and rucaparib dose response measured via colony formation in OVCAR8 (B), OVCAR10 (C), OVCAR4 (D). E) OVCAR3 cells transduced with shControl (shCtrl) or shCLDN4 (#1 and #2). Claudin-4 mRNA and protein levels were measured via qRT-PCR (Internal control, B2M) and immunoblot (Loading control, β-actin). F) Olaparib dose response in shCtrl and shCLDN4 cells. G) Rucaparib dose response in shCtrl and shCLDN4 cells. Experiments performed at least in triplicate. CI, confidence intervals. Error bars, SEM. Statistical test, one-way ANOVA with multicomparison correction, ***p<0.001, ****p<0.0001.

Mechanism by which loss of Claudin-4 expression increases sensitivity to PARPi

To define the mechanism of claudin-4 mediated PARPi response, we examined via RPPA the signaling pathways effected by claudin-4 downregulation. Using the OVCAR3 shControl compared to the shCLDN4 cells, we observed differential expression of multiple proteins encompassing the pathways, DNA damage response, receptor tyrosine kinase signaling, and transcription factor expression (Figure 3A). Specifically, we observed loss of claudin-4 with two different shRNA constructs significantly reduced the protein expression of XRCC1 and 53BP1 (Figure 3B–C). X-Ray Repair Cross Complementing 1 (XRCC1) directly interacts with PARP to promote single-strand DNA damage repair and XRCC1 knockdown increases sensitivity to olaparib [29]. Tumor Protein P53 Binding Protein (TP53BP1/53BP1) is recruited to site of DNA damage via histone modifications (e.g., phosphorylation, methylation, ubiquitination) and promotes non-homologous end-joining (NHEJ) and antagonizes HR repair. Using the DepMap database of ovarian cancer cell lines, while claudin-4 mRNA expression did not correlate with 53BP1 protein expression in HGSOC cell lines, it did with XRCC1 protein expression (Figure 3D–E). Interestingly, using untargeted mass spectrometry data in 12 ovarian cancers [22], when we examined the correlation between claudin-4 and XRCC1 or 53BP1 protein expression, there was only a positive correlation between claudin-4 and 53BP1 (Figure S3A–B). Consistently, we performed an unbiased screen of 126 FDA approved anti-cancer therapies (NCI Approved Oncology Drug Panel). In shCLND4 compared to shCtrl cells, most of the agents that demonstrated increased sensitivity in shCLND4 cells were DNA damaging agents measured by colony formation (Red spots, Figure 3F and Table S3). In contrast, the loss of claudin-4 expression promoted reduced sensitivity to a topoisomerase II inhibitor, teniposide (gray dot above the dotted line, Figure 3F and Table S3). Using Dependency Mapping’s custom analysis tool [14], claudin-4 expression was correlated to the response (Area under the Curve, AUC) to 174 cancer therapeutics. Consistent with the screen data from the shCLND4 cells, higher claudin-4 expression correlated to decreased sensitivity to tubulin targeting agents, DNA damaging (Figure S3C; PARPi are called out) agents, and inhibitors of DNA replication (Figure S3C and Table S3). These findings are consistent with previous reports demonstrating claudin-4 correlates to reduced sensitivity to DNA damaging agents such as cisplatin [12]. However, little is known regarding claudin-4 and the response to PARPi, although they are currently approved for the HGSOC upfront maintenance therapy. Single-cell sequencing of HGSOC tumors [17] showed that claudin-4, TP53BP1, and XRCC1 expression were largely restricted to PAX8-positive tumor cells (Figure 3G), highlighting that loss of claudin-4 would likely impact the DNA repair capacity of the tumor compartment. While there is variability between the claudin-4 knockdown, these data demonstrate that modulation of claudin-4 expression in tumor cells is correlated to DNA damage response effectors, including XRCC1 and 53BP1.

Figure 3: Loss of claudin-4 downregulates DNA repair response effectors.

A) Protein isolated from OVCAR3 shCtrl (n=3) and shCLDN4#1 (n=3) cells and used for Reverse Phase Protein Array. Heatmap of the protein differentially regulated (p<0.05, FDR<15%). Arrows – DNA repair effectors. B) Protein expression of 53BP1 in OVCAR3 shCtrl, shCLDN4#1, and shCLDN4#2 cells. C) Protein expression of XRCC1 in OVCAR3 shCtrl, shCLDN4#1, and shCLDN4#2 cells. D) Claudin-4 mRNA expression (x-axis) and 53BP1 protein expression (y-axis) of 46 ovarian cancer cell lines. E) Claudin-4 mRNA expression (x-axis) and XRCC1 protein expression (y-axis) of 46 ovarian cancer cell lines. F) Cell viability screen of FDA-approved oncology drugs comparing OVCAR3 shCtrl (x-axis) and shCLDN4 (y-axis). Dotted lines represent – 2 standard deviations. Compound class is indicated by the color. G) Single cell sequencing from 5 HGSOC tumors, PAX8 (black, tumor cells), claudin-4 (red), 53BP1 (green), and XRCC1 (blue). Statistical test, one way ANOVA with multicomparison correction. *p<0.05, **p<0.01, ***p<0.001.

Loss of claudin-4 attenuates XRCC1and 53BP1 foci formation and blunts the DNA damage repair response

Knockdown of claudin-4 resulted in the downregulation of DNA repair effectors, including 53BP1 and XRCC1. Thus, using the I-SceI-based functional DNA repair assays [24] we measured the DNA repair activity in cells that have claudin-4 knocked down. While shCLDN4#1 did not show altered homology-directed DNA repair compared to shCtrl, shCLDN4#2 did reduce repair activity, although not significantly (Figure 4A). In contrast, in both shCLDN4 #1 and #2 cells we observed a significant 2-fold decrease in GFP positive cells following DNA damage induction via I-SceI transfection indicating that loss of claudin-4 inhibited canonical NHEJ-mediated DNA repair (Figure 4B). This finding is consistent with the loss of 53BP1 expression in claudin-4 knockdown cells.

Figure 4. Claudin-4 knockdown inhibits non-homologous end joining and reduces 53BP1 foci formation.

A) Homology-directed DNA repair activity measured in OVCAR3 shCtrl and shCLDN4 (#1 and #2) cells transfected without I-SceI and with I-SceI. B) Non-homologous end-joining DNA repair activity measured in OVCAR3 shCtrl and shCLDN4 (#1 and #2) cells transfected without I-SceI and with I-SceI. C) OVCAR3 shCtrl and shCLDN4 (#1 and #2) cells treated with 375 nM olaparib used for immunofluorescence against γH2ax (green) and XRCC1 (red). Nuclei stained using DAPI (blue). D) OVCAR3 shCtrl and shCLDN4 (#1 and #2) cells treated with vehicle control or 375 nM olaparib used for immunofluorescence against γH2ax (green) and 53BP1 (red). Nuclei stained using DAPI (blue). E) Same as D, quantification of 53BP1 foci per nuclei from five field of views. F) Low (2-69, n=95) and high (70-1899, n=107) mutation count from TCGA ovarian cancer tumors (x-axis) compared to claudin-4 mRNA expression (y-axis). Claudin-4 protein expression (CPTAC) versus Mutation count (y-axis) from 82 HGSOC TCGA tumors. Experiments performed at minimum in triplicate. Error bars, SEM. Statistical Test, one way ANOVA with multicomparison correction (A, B, E, F) or unpaired t-test (F) or Spearman Correlation (F). *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

To confirm that the loss of claudin-4 blunted 53BP1 and NHEJ DNA damage repair, we next used immunofluorescence to evaluate XRCC1, 53BP1, and γH2ax foci formation in claudin-4 knockdown HGSOC cells following PARPi treatment. Consistent with the RPPA data we observed a decrease in XRCC1 protein expression (Figure 4C). Olaparib treatment induced variable γH2ax foci formation in shCtrl and shCLDN4 cells (Figure 4C). Consistent with the functional DNA repair assays, olaparib induced 53BP1 foci formation in shCtrl cells, but 53BP1 foci formation was significantly blunted following claudin-4 knockdown (Figure 4D–E), suggesting a decrease in NHEJ DNA repair. These data clearly demonstrate a relationship between claudin-4 expression and DNA repair capacity. To further establish this relationship, we examined mutational burden in ovarian cancer tumors. In ovarian cancer tumors profiled in TCGA, lower claudin-4 mRNA and protein expression significantly correlated to elevated mutational burden (Figure 4F). These data suggest that down regulation of claudin-4 directly inhibits DNA repair and claudin-4 loss appears to be more detrimental to NHEJ activity.

Treating ovarian cancer tumors in combination with claudin-4 inhibitory peptide and olaparib

We next assessed the apoptotic and proliferation capacity of primary ovarian cancer tumors treated with CMP and/or olaparib. Primary ovarian cancer tumors were uniformly sectioned and treated for 72 hrs (Figure 5A). Tumor sections were analyzed for Ki67 (proliferation), cleaved caspase 3 (CC3), claudin-4, and PAX8. Through the collaboration with the Gynecologic Tumor and Fluid Bank, we sectioned and examined the response of CMP and/or olaparib in 15 primary ovarian cancer tumors (10 HGSOC, 2 high grade carcinosarcoma [HGCS], 1 low grade SOC, 1 mucinous, and 1 granulosa cell tumor) (Table S4). In the non-HGSOC tumors, we observed a varied effect of CMP and olaparib treatment on proliferation and apoptosis (Figure S4A). Expression of the Mullerïan marker PAX8, an established HGSOC cell marker [30], was determined to ensure histological analysis within the HGSOC tumor compartments (Figure S4B). In the HGSOC tumors, we observed three HRD positive, one HRD negative and eight HRD unknown tumors. There were six BRCA-wildtype and six tumors with unknown BRCA-status (Table S4). In the serous and carcinosarcoma tumors, we determined claudin-4 expression via immunofluorescence (Figure 5B–C and S4B). We noted varied apoptotic effects within CMP, olaparib, and CMP/olaparib (Figure S4C–D). In contrast, within the HGSOC tumors, we consistently observed that olaparib and/or CMP significantly inhibited proliferation as shown by a significant decrease in Ki67 (Figure 5B–D and S4E–F). Consistent with the in vitro data, claudin-4 null/low tumors had consistently and significantly lower Ki67 compared to control (p<0.0001). In contrast, claudin-4 high tumors exhibited a variable and non-significant anti-proliferative response to olaparib compared to the control tumors (Figure 5E). However, CMP/olaparib combination significantly inhibited proliferation independent of claudin-4 expression, suggesting claudin-4 inhibition may be promoting a “BRCAness” state in improving olaparib response. These data generated from primary ovarian cancer tumors demonstrate that low claudin-4 expression correlates to olaparib sensitivity. Also, combining claudin-4 inhibition and olaparib is potentially a novel combinatory approach to treating patients that are insensitive to olaparib.

Figure 5: Olaparib and inhibition of claudin-4 inhibit proliferation of primary ovarian cancer tumors.

A) Workflow of the ex vivo experiments using primary tumor tissue. Primary tumors were classified as claudin-4 null/low or claudin-4 high based on immunofluorescence (IF, claudin-4, green and nuclei, DAPI). IHC against Ki67 was completed in all the tumors, tumors were de-identified, and a histology score (H-score) was calculated based on intensity of Ki67 signal and percentage of tumor cells from at least three fields of view. B) Representative H-score and claudin-4 IF for claudin-4 null/low expressing ovarian cancer tumors. C) Representative H-score and claudin-4 IF for claudin-4 high expressing ovarian cancer tumors. D) Representative Ki67 images of claudin-4 null/low and high expressing ovarian cancer tumors (Scale bar, 50 microns). E) Normalized Ki67 H-score based control (Ctrl) treated tumors for all the claudin-4 null/low (N=5) and high (N=7) expressing serous ovarian tumors. Three fields from each tumor were analyzed and each tumor is represented by three matched colored spots. Error bars, SEM. Statistical test, one way ANOVA with multicomparison correction. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

DISCUSSION

Beyond BRCA and HRD-status there are limited molecular markers that correspond to response to standard of care DNA damaging agents, such as platinum-based chemotherapies and PARP inhibitors. Claudin-4 is classically depicted as a tight junction protein, is expressed in 70% of all ovarian cancers, and in a recent review of 17 cancer types, CLDN4 is upregulated in the cancer tissue versus normal tissue in 16 of the 17 cancers [31]. Thus, having a broad applicability claudin-4 may serve as additional predictive marker for certain therapies . Claudin-4 has been linked to the response to both taxane-based and platinum-based chemotherapeutics [10, 32]. In this report, we observed that lower claudin-4 expression correlated to improved response to PARP inhibitors.

Screening a panel of FDA-approved oncology therapies, the loss of claudin-4 expression improved the response to several classes of agents, most notably DNA damaging agents. However, the screen did identify several agents with reduced sensitivity following claudin-4 knockdown, suggesting that using claudin-4 expression as a biomarker for pan-therapeutic response is not realistic. In addition, in a cohort of 140 primary ovarian cancer tumors with mixed histology, high claudin-4 expression did correlate to worse overall survival, but was not associated with sensitivity to another class of DNA damaging agents, platinum-based chemotherapy [33]. Consistently, while our study focused on claudin-4’s link to therapeutic response only in high grade serous ovarian cancer, when evaluating the proliferative and apoptotic response to olaparib and CMP in ex vivo cultures of other ovarian cancer histology (low grade serous and mucinous) the responses did not match the high grade serous tumor response (Figure S4). Thus, more so in high grade serous tumors, expression of claudin-4 may be more indicative of disease progression and therapeutic response.

PARP inhibitors represent a breakthrough in cancer care and while initially approved for ovarian cancer, they are now FDA-approved for metastatic breast, pancreatic, and castration-resistant prostate cancer [34]. Germline BRCA-mutations and homologous recombination deficiency (HRD) status remain the most widely used molecular markers that predict PARP inhibitor response. However, there remains a paucity of targeted therapeutics for patients without BRCA-mutations and/or are HR proficient. For this reason, there is significant interest in promoting a “BRCAness” like state in HR proficient tumors to improve PARP inhibitor response. Knocking down claudin-4 decreased functional non-homologous mediated DNA repair activity and modulated the expression and foci formation capacity of DNA repair effectors. Also, knockdown of claudin-4 led to the downregulation of several DNA repair factors, including 53BP1 and XRCC1 and inhibition of NHEJ repair. Previous reports have shown 53BP1 loss leads to PARP inhibitor resistance, but the resistant phenotype was dependent on the status of BRCA and HRD [35, 36]. With respect to XRCC1, in an ovarian cancer model loss of XRCC1 induced a synthetic lethality with olaparib [37]. While we report here a direct link between loss of claudin-4 expression and DNA damage repair activity, there is evidence of this link within the ovarian cancer TCGA dataset: low expressing claudin-4 tumors more often have an elevated mutational count. These studies suggest that a claudin-4 mediated decrease of both XRCC1 and 53BP1 could be promoting DNA repair-deficiency or BRCAness, enhancing the response to DNA damaging agents.

Previous examination of claudin-4 expression related to the response of taxane-based microtubule stabilizing chemotherapy (e.g., paclitaxel) found that elevated claudin-4 reduced paclitaxel induced apoptosis [10]. Mechanistically, claudin-4 was associated with tubulin and regulated tubulin polymerization. Moreover, loss of claudin-4 significantly altered the cell cycle, with cells accumulating in G2/M. PARPi also drove G2/M arrest and DNA damage pathways are dependent on the phase of the cell cycle [15, 38]. Thus, claudin-4’s function to regulate both tubulin polymerization and DNA damage repair pathways may likely converge on the cell cycle.

Leveraging ex vivo culture of primary human ovarian cancer tumors we demonstrated that tumors with high claudin-4 were more resistant to olaparib compared to tumors with low claudin-4. Further, in a panel of primary ovarian cancer tumors, low claudin-4 expressing tumors had an improved anti-proliferative response to olaparib and in high claudin-4 expressing tumors the combination of claudin-4 inhibition and olaparib promoted a robust anti-proliferative response. Excitingly, while the number of tumors examined was limiting, claudin-4 inhibition via CMP in ovarian cancer tumors with high claudin-4 increased the sensitivity to olaparib. Several studies have suggested the potential of targeting claudin-4 to inhibit ovarian cancer progression [7, 10, 39], and while CMP has both in vitro and in vivo efficacy [7, 10] the necessary dose is not physiologically possible. Claudin-4 expression is mainly confined to transformed epithelial cells [10, 39] and is described to be mainly localized to the plasma membrane. Notably, several studies in cancer cells detail non-membrane associated claudin-4, including cytosolic and nuclear accumulation, suggesting an alternative oncogenic function of claudin-4 that is independent of tight junctions [40, 41]. These studies combined with the results presented in this study highlight that targeting claudin-4 offers an approach to inhibiting cancer progression with minimal off-target effects. Thus, future work will investigate combining CMP’s claudin-4-specificity [7, 11] with the targeted protein degrader technology, proteolysis-targeting chimeras (PROTAC).

In conclusion, claudin-4 expression is directly related to the DNA damage response and conveys decreased sensitivity to the standard of care for ovarian cancer, including PARP inhibitors. In the context of BRCA-wildtype and HR-proficient ovarian cancer, claudin-4 expression may serve as predictor of therapeutic response and offer a molecular marker-based rationale for combining claudin-4 inhibition with DNA damaging agents.

Abbreviations:

PARP

Poly (ADP-ribose) polymerase

HGSOC

high grade serous ovarian carcinoma

CMP

claudin mimetic peptide

TCGA

The Cancer Genome Atlas