Loss of ARID1A in Tumor Cells Renders Selective Vulnerability to Combined Ionizing Radiation and PARP Inhibitor Therapy

Altered chromatin structure, due to somatic mutations or epigenetic alterations of genes involved in chromatin remodeling, is a major contributor to tumor development (1). SWI/SNF chromatin remodeling complexes are an important class of epigenetic modulators, which regulate a plethora of basic biological functions including DNA replication, transcription, and DNA repair by remodeling chromatin configuration. Loss-of-function mutations in SWI/SNF genes have been identified in many types of human cancer, and an important example is the AT-rich interactive domain 1A (ARID1A) gene, which is the most frequently mutated subunit in the SWI/SNF chromatin remodeling complex (2–5). ARID1A mutations are frequently detected in endometrium-derived cancers including approximately 50% of ovarian clear cell carcinomas, approximately 35% of uterine endometrioid carcinomas, and approximately 30% of ovarian endometrioid carcinomas (2, 5–7). Mutations in ARID1A have also been reported in stomach, bladder, pancreas, and hepatocellular carcinomas among others (8–12). ARID1A is a tumor suppressor, and functions as a cell-cycle checkpoint protein (3). Genetically engineered mice with ARID1A deletions in their ovarian surface epithelium are known to develop ovarian carcinomas in a background of PTEN inactivation or PIK3CA activation (13, 14).

The finding of frequent ARID1A mutations in endometrium-derived cancers suggests that targeting the ARID1A pathway as an anticancer intervention has translational potential. However, it has become apparent that targeting tumor suppressor genes such as TP53 using a standard pharmacologic approach is highly challenging; in contrast, targeting gain-of-function oncogenes by pharmacologic and antibody interventions has proven to be more promising. Nonetheless, loss-of-function mutations in tumor suppressors may result in acquisition of dependence of cancer cells on alternative compensatory pathway(s) or downstream molecular effector(s). This unique feature of loss-of-function mutations in tumor suppressor genes offers opportunities for targeting cancer cells by disrupting compensatory or alternative pathway(s) (15). A well-known example is the sensitivity of tumor cells with BRCA-inactivation to PARP inhibition, a consequence of their deficiency on homologous repair (HR; refs. 16, 17).

To extend this strategy to ARID1A^−/− tumors, we sought first to understand the impact of inactivation of ARID1A or other SWI/SNF chromatin remodeling proteins in DNA damage repair pathways (18–20). In mammalian cells, DNA double-strand breaks (DSB) are predominantly repaired by the NHEJ and HR pathways, each of which harness a unique set of molecular players. The balance between both pathways is essential for genome stability, and disturbance of the balance often leads to disease, including cancer. SWI/SNF chromatin remodeling has been reported to participate in the early phase (before strand intrusion phase) of DSB repair through rapid localization to the DSB sites, clearing local nucleosome occupancy, and physically facilitating recruitment of DNA repair enzymes and other modulators to the vicinity of DSBs (18–20). Thus, loss of ARID1A may disturb the balance of HR/NHEJ DNA repair efficacy and may render cells susceptible to specific genotoxic treatment. Indeed, two studies reported the involvement of ARID1A in NHEJ- and HR-mediated DSB repair, respectively (21, 22). However, ARID1A inactivation–induced negative regulation of both NHEJ and HR repair pathways remains of interest, and has yet to be fully elucidated (23). It also remains to be determined which DNA repair mechanism or molecular pathway is employed by ARID1A-deficient tumors for survival and maintenance of DNA integrity in the face of endogenous stress and environmental challenges that result in DNA damage.

In this study, we first established that ARID1A deficiency led to a functional compromise in NHEJ repair and, to a lesser degree, in HR or alt-NHEJ. Similar to cells with NHEJ deficiency, ARID1A-deficient cells were partially sensitive to radiation-induced DNA damage, likely due to sufficient HR DNA repair activity. Using a synthetic lethal screen to identify drugs affecting DNA repair that might act in concert with irradiation in ARID1A-deficient tumor cells, we identified PARP inhibitor as a strong candidate. Applying this treatment approach in animal models, we were able to induce long-term remission in ARID1A-deficient tumors, which persisted after completion of the treatment, whereas the same treatment was not effective in tumors with intact ARID1A function. Our findings indicate that disturbance of the DNA repair balance associated with ARID1A deficiency can be exploited to develop highly specific and potent anticancer treatments.

All animal-related procedures were approved by the Johns Hopkins University Animal Care and Use Committee. PAX8-rtTA/TetO-Cre mouse strains were acquired from Dr. Ronny Drapkin (13, 24). Arid1a^flox/flox mice were generated as described previously (13, 24). For xenograft assays, 2 × 10⁶ cells were injected subcutaneously with Matrigel (v/v; BD Biosciences) into flanks of athymic nu/nu mice. Once the tumor volume reached approximately 100 mm³ (approximately 10–14 days), mice were randomized into four groups for further analysis.

Archived formalin-fixed and paraffin-embedded human tissues were provided by Seirei Mikatahara General Hospital (Shizuoka, Japan). All specimens were deidentified and the study was approved by local ethical committee (IRB No.14–46; 2014/12/15) for a waiver of informed consent of the subjects.

The immortalized normal human endometrial cell line, hEM3, previously established in our team was used for this study (25). Knockout of ARID1A was performed as described previously (25). Two pairs of isogenic cell lines, HCT116 and MCF10A (ARID1A^+/+, ARID1A^−/−), were purchased from Horizon Discovery Ltd. We used a previously established ARID1A-Tet-on OVISE cell line, which originally lacks ARID1A expression due to inactivating mutations and reexpresses ARID1A upon tetracycline induction (3). Other cancer cell lines (TOV21G1, RMG1, HEC151, and ECC1) were grown as described previously (25).

Cells were seeded in a Nunc glass-bottom dish (Thermo Fisher Scientific). After 5-minute incubation with 2 μmol/L Hoechst (Thermo Fisher Scientific), cells were mounted on a preheated (37°C) stage of a Zeiss LSM 710 confocal microscope equipped with a 405-nm laser source. To induce localized DSBs, the laser setting was set to 100% power output with four laser iterations. Image analysis was performed using Zeiss Zen 2010 software.

Mann–Whitney test (two-tailed) was used to calculate the statistical significance between two experimental groups. A combination index was calculated as described previously (26), and synergistic drug interactions were analyzed by the method developed by Chou and Talalay (27). Two-way ANOVA with Bonferroni correction was used to calculate the statistical significance in experiments (Fig. 2C–E; Supplementary Fig. S4). Statistical analyses were performed using GraphPad Prism 7 software. All results are expressed as means ± SEM. Two-sided P ≤ 0.05 was considered significant.

Other methods can be found in Supplementary Materials.

To study the role of ARID1A in DNA damage, we first quantified DNA damage and the damage response in complex atypical hyperplasia (CAH) of endometrium, which exhibited focal and heterogeneous loss of ARID1A in the same tissue sections (left, Fig. 1A). IHC analysis showed more intense phospho-H2AX (S139, γH2AX) immunoreactivity in ARID1A-negative tumor areas, compared with adjacent areas with ARID1A expression (middle, Fig. 1A; Supplementary Fig. S1). To confirm that increased γH2AX staining reflected DSBs, we also stained tissues with phosphorylated ATM (S1981), a marker of an activated DNA damage response. We observed concordance between pATM and γH2AX staining (right, Fig. 1A), implying that unrepaired DSBs were more prevalent in ARID1A-deficient epithelial cells.

Although previous studies implicate chromatin remodeling proteins such as ARID1A in DNA repair, such findings have rarely been validated in vivo. To functionally establish the in vivo role of ARID1A in DSB repair, we assessed the status of DNA damage using a genetically engineered mouse model with conditional deletion of ARID1A. This model was established by crossing ARID1A^flox/flox mice with PAX8-Cre mice carrying doxycycline-inducible Cre recombinase under control of the Pax8 promoter (Fig. 1B). Pax8 is a transcription factor expressed in Mullerian epithelial cells lining the gynecologic tract, including endometrial epithelial cells; thus, doxycycline administration caused Pax8-driven, endometrial epithelium–specific deletion of ARID1A. In early stages of doxycycline administration, we observed a heterogeneous loss of ARID1A in murine uterine epithelium, shown by stretches of endometrial epithelium with ARID1A loss (stars in left panels, Fig. 1C) alternating with stretches of ARID1A-expressing epithelium (Fig. 1C). This provided an opportunity to directly compare DNA damage in ARID1A-expressing and nonexpressing epithelium from the same animals. Whole-animal irradiation (2.5 Gy) was used to induce DSBs. IHC staining revealed more prominent γH2AX punctate foci in ARID1A^−/− cells at 2 hours and 8 hours postirradiation compared with ARID1A-intact cells (right, Fig. 1C). Quantification of γH2AX foci per cell indicated that the increase in punctate foci in ARID1A^−/− cells was significant (Fig. 1D).

To further assess the mechanistic role of ARID1A in DSB repair, we deleted ARID1A in an immortalized endometrial epithelial cell line, hEM3, using CRISPR/Cas9. Consistent with in vivo observations, irradiation resulted in higher levels of γH2AX in hEM3-ARID1A^−/− than in hEM3-ARID1A^+/+ cells at all timepoints examined (top, Supplementary Fig. S2A). We verified these results in parental hEM3 cells following siRNA-mediated silencing of ARID1A (bottom, Supplementary Fig. S2A). Introduction of ARID1A in Tet-on OVISE diminished γH2AX levels (Supplementary Fig. S2B). Results of a comet assay showed that ionizing radiation–induced DNA DSBs were exacerbated by siRNA-mediated ARID1A silencing in parental hEM3 cells, and were rescued by ARID1A reexpression in ARID1A-mutated OVISE cells (Supplementary Fig. S2C and S2D).

We next performed a clonogenic assay to determine whether ARID1A-deficient cells were more sensitive to radiation-induced DNA damage. After irradiation, we observed a modest decrease in survival of HCT116-ARID1A^−/− cells; however, we could not observe a different clonogenic survival after irradiation in hEM-ARID1A^+/+ and ARID1A^−/− cells (Fig. 1E). These results suggest that increased DNA damage in ARID1A-deficient cells could be rescued by a compensatory survival pathway, and suggest that an additional treatment combined with irradiation would be required for killing ARID1A-deficient cells.

To explore the underlying mechanism of increased DNA damage signals in ARID1A-depleted cells, we performed global in vivo serine/threonine phosphoproteomic profiling as an unbiased approach to interrogate differential cellular responses to irradiation between ARID1A^−/− and ARID1A^+/+ tumor xenografts (Supplementary Table S1). Network analysis showed enrichment of proteins with altered phosphorylation involved in DNA replication and repair, including ATM, 53BP1, TOP2A, and TOP2B, in ARID1A^−/− tumors (Supplementary Fig. S3A). The finding of enriched pATM expression by phosphoproteomic profiling orthogonally validated our immunostaining data. Using ingenuity pathway analysis (IPA), we found enriched pathways including NHEJ and base excision repair (BER) DNA repair pathways involving XRCC1, XRCC4, ATM, and PRIM1, as well as cell-cycle control of chromosomal replication pathway involving TOP2A, TOP2B, and CDC7 (Supplementary Fig. S3B). Collectively, our data provide new evidence of ARID1A in DSB, replication, and repair regulation. The data also support a view that defects in these pathways resulting from ARID1A loss led to a prolonged DNA damage response and persistence of DSBs after DNA damage insults.

To determine whether ARID1A was required for NHEJ and/or HR, we monitored both repair mechanisms simultaneously (Fig. 2A; ref. 28). A significant decrease in NHEJ activity was observed in ARID1A^−/− cells. HR activity was also slightly decreased in ARID1A^−/− cells, although precise quantitation of HR was difficult due to the low level of HR even in the setting of ARID1A-intact cells. The impact of ARID1A loss on NHEJ appears to be of biological significance, as a high proportion of cells rely on this pathway. We further characterized the impact of ARID1A in alt-NHEJ, a NHEJ subpathway (29, 30). Knockdown of CtIP (effector) and 53BP1 (suppressor) served as controls. We found that alt-NHEJ was also decreased in ARID1A-depleted cells (Fig. 2B). Thus, the data indicate that ARID1A deficiency causes impairment of DSB repairs, with NHEJ repair pathway being most significantly compromised.

Because the activity of NHEJ and HR DNA repair pathways is cell-cycle dependent (31), we asked whether ARID1A-deletion caused cell-cycle alterations, which may indirectly affect DNA repair. Cell-cycle analysis of hEM3 isogenic cell lines indicated that ARID1A^−/− cells displayed reduced S and G₂–M fractions compared with ARID1A^+/+ cells (Supplementary Fig. S4). This result is consistent with that observed in a pair of isogeneic HCT116 cell lines (22). Because HR occurs primarily during S and G₂–M phases, we postulate that reduced HR observed in ARID1A-deficient cells can likely be attributed to attenuation in S and G₂–M phases.

To determine whether an ARID1A-dependent process is required for recruitment of NHEJ repair proteins, we monitored recruitment of 53BP1 to DNA damage lesions induced by laser microirradiation. We found a significant delay in the recruitment of this key NHEJ repair factor to sites of laser-induced DNA damage in ARID1A-deficient cells (Fig. 2C). Restoration of ARID1A expression in ARID1A-deficient cells partially restored 53BP1 recruitment efficiency (Fig. 2D). Similar results were observed for RIF1, another crucial component of the NHEJ repair pathway acting downstream of 53BP1 (Fig. 2E). There was no evidence of reciprocal regulation of ARID1A recruitment to DSBs by 53BP1. Knockdown or knockout of ARID1A in mammalian cells did not affect 53BP1 recruitment to DSBs induced by microirradiation (Supplementary Fig. S5). Collectively, our results show that ARID1A deficiency compromises NHEJ repair by preventing the recruitment of key NHEJ proteins, including 53BP1 and RIF1, to DSB sites.

Because SWI/SNF complexes function as master regulators of chromatin structure, we hypothesized that regulation of the chromatin landscape by ARID1A is a key step in response to DSBs. To assess genome-wide chromatin configuration (openness vs. compactness), we performed ATAC-seq in ARID1A^+/+ and ARID1A^−/− cells before, and 1 hour after, irradiation (Fig. 2F). Enrichment of 146 bp DNA fragments indicates an increased open chromatin state because these fragments correspond to mononucleosome occupancy. As compared with ARID1A^−/− cells, a larger 146 bp DNA peak was observed in ARID1A^+/+ cells after irradiation, implicating a functional role of ARID1A in modulating the global chromatin state in response to ionizing radiation. We introduced site-specific DSBs near the AAVS1 locus (chromosome 19) and the GAPDH locus (chromosome 12) in ARID1A^+/+ and ARID1A^−/− cells using CRISPR/Cas9, and evaluated chromatin accessibility near the introduced DSBs (described in Supplementary Materials). As expected, we observed increased chromatin accessibility at genomic regions proximal to the AAVS1 locus in ARID1A^+/+ cells compared with ARID1A^−/− cells (Fig. 2G). We did not observe significant differences in chromatin accessibility at genomic regions near GAPDH in ARID1A^+/+ cells compared with ARID1A^−/− cells (Fig. 2G). On the basis of these data, we conclude that ARID1A-dependent alteration in chromatin configuration, which facilitates the recruitment of key effectors of the DNA repair machinery to the damaged sites, is a prerequisite step for initiating DSB repair.

Because ARID1A-deficient cells were only marginally more radiosensitive than ARID1A-intact cells, we sought to identify an approach to enhance the therapeutic index. Because platinum-based chemotherapy (for example, carboplatin), is routinely used for treatment of ovarian and endometrial carcinomas, and has been shown to increase sensitivity to radiotherapy (32, 33), we tested the sensitivity of ARID1A-deficient cells to carboplatin, either as a single agent or in combination with radiation. Clonogenic assays were performed on isogenic pairs of ARID1A^+/+ and ARID1A^−/− cells derived from hEM3 and HCT116 parental cell lines. Although hEM3-ARID1A^−/− cells were slightly more sensitive to carboplatin than ARID1A^+/+ cells, differences were less apparent in the HCT116 isogenic model. For both cell types, carboplatin combined with irradiation did not result in appreciable improvements over single-agent treatment (Supplementary Fig. S6A and S6B). This negative result prompted us to screen a panel of additional chemotherapeutic drugs and inhibitors of DNA repair and of epigenetic regulators (Fig. 3A; Supplementary Fig. S7). Among these drugs screened, the PARP inhibitor, olaparib, emerged as the most potent and specific radiosensitizer for ARID1A^−/− cells (Fig. 3A). ATR inhibitor, AZD-6783, also enhanced potency of radiotherapy for ARID1A^−/− cells. PIK-587, a dual PI3K and mTOR kinase inhibitor, potently inhibited cell growth irrespective of ARID1A deletion status (Supplementary Fig. S7). Next, we tested the efficacy of combining PARP inhibitor and irradiation in a 3D culture system on both HCT116 and hEM3 pairs of isogenic cell lines. In these experiments, we observed synergistic cytotoxicity in ARID1A^−/− cells (Fig. 3B). The increased sensitivity to ionizing radiation plus PARP inhibitor was further verified in clonogenic cell survival analysis in both hEM3 and HCT116 isogenic cell lines (Fig. 1E).

To gain molecular insight into the synergistic effects, we compared PARP activity between ARID1A^−/− and ARID1A^+/+ cells. We observed elevated basal levels of PolyADP-ribosylation (PARylation) in ARID1A^−/− cells (Fig. 3C). This result was confirmed by gene silencing using two different siRNAs (Fig. 3D). Because PARP-mediated PARylation occurs via consumption of NAD⁺, the relative level of NAD⁺ (NAD/NADH ratio) is another indicator of cellular PARP activity. We found that the NAD/NADH ratio was reduced in hEM3 cells after ARID1A knockdown, confirming increased PARP activity in ARID1A-deficient cells (Fig. 3D). The extent of DNA breaks resulting from single or combined irradiation and PARP inhibitor exposure was assessed by measuring the levels of γH2AX (Fig. 3E). We found that combining PARP inhibition with irradiation led to significant increases in γH2AX levels in ARID1A^−/− cells but not in isogenic ARID1A^+/+ cells (Fig. 3E).

PARP inhibitors are thought to compromise DNA damage repair through two major mechanisms. First, they cause “PARP-trapping” by preventing PARP dissociation from DNA. Another mechanism is through persistence of single-strand breaks, which ultimately progress to DSBs during replication. To determine which mechanisms are relevant to our model, we excluded a PARP-trapping effect by RNAi-mediated knockdown of PARP1 in hEM3-ARID1A^−/− and ARID1A^+/+ cells. When PARP1 was depleted, ARID1A^−/− cells were still more sensitive to irradiation than ARID1A^+/+ cells (Fig. 3F). These data are consistent with the notion that suppression of PARP1 catalytic activity is the primary mechanism underlying synergy between PARP inhibitor and irradiation.

In light of the observation that irradiation increases PARP inhibitor sensitivity in ARID1A^−/− cells, we evaluated this treatment strategy in mice using isogenic HCT116 cell lines that differed only in ARID1A status (right, Fig. 4A). Tumor volumes and animal weights did not differ between different groups before treatment (Supplementary Fig. S8A and S8B). Strikingly, irradiation combined with PARP inhibitor completely prevented the growth of ARID1A^−/− tumors, demonstrating superior efficacy over either treatment alone (Fig. 4B). In contrast, the addition of PARP inhibitor did not enhance the effect of irradiation on reducing growth of ARID1A^+/+ tumors, which continued to progress insidiously (Fig. 4B). Immunoblotting of tumor lysates confirmed ARID1A expression and depletion of PARylated proteins by olaparib (Supplementary Fig. S8C and S8D). To determine whether combination treatment led to long-lasting effects that persisted after treatment was terminated, we followed the animals that received combination therapy for 4 additional weeks after the final treatment. We found that ARID1A^+/+ tumors regrew quickly after cessation of therapy, whereas ARID1A^−/− tumors continued to shrink (Fig. 4C).

To determine the generalizability of these findings, we evaluated the same treatment regimen in additional xenograft models: ovarian clear cell carcinoma cell lines (RMG-1 and TOV21G) and uterine endometrioid cancer cell lines (ECC1 and HEC151). ECC1 and RMG-1 are ARID1A wild-type and express ARID1A protein, while HEC151 and TOV21G lose expression of ARID1A due to inactivation mutations (right, Fig. 4A). In mice bearing ARID1A-expressing tumors (ECC1 and RMG-1), PARP inhibition did not confer additional antitumor benefits over irradiation, whereas combination therapy was more effective than irradiation or PARP inhibitor monotherapy in mice bearing ARID1A-deficient tumors (HEC151 or TOV21G; Fig. 4D). IHC staining for γH2AX and cleaved caspase-3 revealed significantly higher levels of DSBs and cell death in ARID1A-deficient tumors than in ARID1A-expressing tumors following combination treatment (Fig. 4E). We continued monitoring the mice for another 3 weeks after treatment. While long-term remissions were achieved in mice bearing ARID1A-deficient tumors (HEC151 and TOV21G), early recurrence and tumor regrowth was observed in mice with ARID1A-expressing tumors (Supplementary Fig. S8E).

This investigation of mechanisms underlying the biological functions of ARID1A in NHEJ repair yielded novel biological and translational implications. Of particular importance, we discovered that combining radiation with PARP inhibitor therapy is highly effective in eradicating ARID1A-mutated tumors. The effect of PARP inhibition on ARID1A-mutated tumors is mechanistically distinct from effects on HR-deficient tumors, such as those harboring deleterious BRCA mutations, which are intrinsically sensitive to PARP inhibitor monotherapy due to their loss of ability to repair PARP inhibitor–induced replication fork collapse (34, 35). As to why PARP inhibitor monotherapy is less effective in ARID1A-deficient tumors, we speculate that the HR machinery in ARID1A-deficient cells remains intact, and the activity is sufficient to repair DSBs resulting from replication fork stalling, including those induced by PARP trapping.

NHEJ is the primary repair mechanism for ionizing radiation–induced DSBs and the repair is active at all stages of cell cycle including G₁ (36). We hypothesize that due to the reduced NHEJ repair capacity in ARID1A-deficient cells, radiation-induced DNA breaks are not effectively repaired. As a result, ARID1A-deficient cells depend on other PARP-dependent DNA repair systems including HR, alt-NHEJ, and replication fork repairs to maintain DNA integrity. Accordingly, ionizing radiation–induced DNA breaks cause ARID1A-deficient cells to become profoundly dependent on these PARP-dependent repair pathways, which renders them highly sensitive to PARP inhibitors (Fig. 5; ref. 35). If this hypothesis is correct, DNA-damaging agents that induce similar types of DNA breaks should exhibit synergy with PARP inhibition in ARID1A-mutated tumors, including human tumors. Indeed, our unpublished data using the alkylating agent, temozolomide, supports this view.

A treatment regimen of combined ionizing radiation and PARP inhibition could, in principle, mediate effective killing of ARID1A-mutated tumor cells while sparing normal tissues, which retain ARID1A expression and functionality for efficient DNA repair. While the synthetic lethality of PARP inhibition in HR-defective BRCA-mutated ovarian carcinomas is well-documented, this study provides evidence in a new arena, demonstrating compromised DNA repair efficiency as an Achilles' heel in ARID1A-mutated malignancies.

Radiotherapy is often a consideration in the management of locally advanced and recurrent endometrial cancers, particularly for patients who are too frail for surgery (37, 38). The anatomic distribution of clear cell and endometrioid carcinomas, which are characteristically confined to the pelvis, but which may metastasize to lymph nodes, renders these tumor types amenable to radiotherapy. ARID1A loss, which occurs in a significant proportion of cases, also confers intrinsic sensitivity to radiation. Importantly, our preclinical data support enhancement of efficacy by the combination of fractionated radiation with PARP inhibition, and should be tested in late-stage, recurrent uterine endometrioid and clear cell carcinomas. With improvements in delivery of targeted radiation, there is renewed interest for applications of radiotherapy for ovarian cancer (39), and future studies should consider the addition of a PARP inhibitor to treat chemoresistant ovarian endometrioid and clear cell carcinomas.

Importantly, identifying molecular genetic alterations predicting clinical response to radiotherapy remains an unmet need. Future clinical trials to test the proposed combination treatment strategy might consider accruing patients based on ARID1A mutation status, with the intent to evaluate ARID1A mutation or inactivation as a potential biomarker for predicting treatment outcome. A relevant question to be addressed is whether tumors with mutations in other subunits of the ARID1A chromatin remodeling complex display a similar phenotype. Finally, efforts to optimize dosing schedules and safety monitoring will be necessary for translation of our preclinical findings to human trials.

S. Gaillard is a consultant/advisory board member for AstraZeneca, Tesaro, Merck, and Immunogen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Park, M.H. Chui, M.A. Bellani, V.A. Bohr, I.-M. Shih, T.-L. Wang

Development of methodology: Y. Park, Y.S. Rahmanto, R.A. Shamanna, M.M. Seidman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Park, Y.S. Rahmanto, Z.-C. Yu, R.A. Shamanna, M.A. Bellani, A. Ayhan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Park, M.H. Chui, Z.-C. Yu, R.A. Shamanna, A. Ayhan, S. Franco, V.A. Bohr, I.-M. Shih, T.-L. Wang

Writing, review, and/or revision of the manuscript: Y. Park, M.H. Chui, Z.-C. Yu, S. Gaillard, A. Ayhan, A. Viswanathan, M.M. Seidman, S. Franco, A. Leung, V.A. Bohr, I.-M. Shih, T.-L. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.S. Rahmanto, A. Ayhan, I.-M. Shih

Study supervision: A. Ayhan, S. Franco, A. Leung, T.-L. Wang

The authors thank Dr. Tae Magomi's contribution to the preliminary work. The study is supported by NIH/NCI R21CA165807, RO1CA129080, P50CA228991, RO1CA215483, Gray Foundation, and the Richard W. TeLinde Endowment Fund from the Department of Gynecology and Obstetrics, Johns Hopkins University (Baltimore, MD).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.