The DNA Cytosine Deaminase APOBEC3B is a Molecular Determinant of Platinum Responsiveness in Clear Cell Ovarian Cancer

Clear cell ovarian carcinoma (CCOC) is a histopathologic subtype of epithelial ovarian carcinoma (EOC) with significant treatment challenges (1, 2). CCOC is the only EOC subtype with increasing incidence rates among North American Black, Hispanic, and Asian women (3). Early-stage disease, which comprises 60%–70% of all cases, has a better relative prognosis compared with stage III or IV disease, which has the poorest survival outcomes of all EOCs (4). CCOC is treated with surgical resection, followed by platinum and taxane chemotherapy (1, 2). A subset of CCOC is overtly nonresponsive to platinum-based drugs, suggesting a preexisting resistance mechanism(s), and another subset often develops resistance during therapy suggesting an acquired resistance mechanism(s). Numerous chemo-resistance mechanisms have been proposed for CCOC, including increased drug efflux, altered DNA repair, and lower cell proliferation (5). Nonetheless, platinum resistance remains a serious problem for patients with CCOC and is an important prognostic factor.

Genomic DNA sequences have helped distinguish EOC histopathologic subtypes. For instance, the most common mutation signature in high-grade serous ovarian carcinoma (HGSOC) is homologous recombination deficiency, which is attributable to loss-of-function mutations in BRCA1, BRCA2, and related recombination genes (6, 7). In contrast, the top mutation signatures in CCOC are attributable to two related processes, AGEING and APOBEC (7, 8). The former is due to spontaneous deamination of methylated cytosines to thymines in 5′-meCG motifs, which leads to C-to-T mutations following DNA replication or misrepair. The latter is also due to water-mediated deamination but occurs preferentially within 5′-TCA and 5′-TCT trinucleotide motifs and is catalyzed by one or more members of the APOBEC3 family of DNA cytosine deaminases. APOBEC3-catalyzed genomic uracils lead to C-to-T and C-to-G mutations by DNA replication or misrepair of intermediates in the uracil excision repair pathway (uracil and abasic sites, respectively; reviewed in ref. 9). Three enzymes, APOBEC3A, APOBEC3B, and APOBEC3H haplotype I, have been implicated to varying degrees in causing the APOBEC mutation signature in multiple cancer types (APOBEC3B foremost; refs. 10–15) but role(s) for these enzymes in CCOC have yet to be addressed.

In multiple cancer types, high levels of APOBEC3B have been associated with poor clinical outcomes including drug resistance (16–19). Accordingly, the studies here were initiated to determine the APOBEC enzyme(s) expressed in CCOC and investigate associations with clinical outcomes. The results of two orthologous methods, IHC using formalin-fixed, paraffin embedded (FFPE) tumor specimens and qRT-PCR using matched fresh frozen (FF) tumor tissues, demonstrated APOBEC3B overexpression in a pilot cohort of primary CCOC. IHC was then used to interrogate a larger cohort and assess associations with clinical outcomes. Surprisingly, high APOBEC3B levels in CCOC associated with longer (not shorter) durations of progression-free (PFS) and overall survival (OS). This potential paradox was explained by recognizing that all patients had been treated with a platinum-based chemotherapy and performing a series of cell-based studies to demonstrate that APOBEC3B and uracil excision activities combine to confer sensitivity to cisplatin and carboplatin. A similar relationship between APOBEC3B and platinum sensitivity was also observed in CCOC patient-derived xenografts (PDX). Collectively, these studies show that APOBEC3B may serve as a molecular determinant of platinum therapy responses in CCOC.

Clinical biospecimens were collected from the University of Minnesota BioNet Repository [n = 17; Institutional Review Board (IRB) 1792] and the Mayo Clinic Biorepository for Ovarian Cancer Research (n = 48; IRB 08-005749). Both repositories are operated in accord with the U.S. Common Rule, with samples collected from patients who provided informed consent after local IRB approval. BioNet provided nine CCOC cases with FF tumor tissues and seven of nine had matched FFPE blocks. Among the 182 CCOC cases registered in the Mayo Clinic Biorepository, FFPE blocks containing tumor were available from 48 cases who had surgical debulking and platinum/paclitaxel therapy. Patient outcome was assessed by semi-annual review of clinic notes during the first 2 years after diagnosis and annual reviews thereafter, supplemented with annual questionnaires for patients not seen at the Mayo Clinic (Rochester, MN) within the previous year. Histology was assessed independently by pathologists from each institution and confirmed by our own hematoxylin and eosin (H&E) analysis of each case. Clinical specimens are summarized in Supplementary Tables S1–S3. Correlations between H-score and mRNA expression were determined using Pearson test and linear regression. Log-rank and Cox regression (unadjusted and stage-adjusted) were used to determine associations between groups by the median APOBEC3B (low/high) and continuous APOBEC3B scores, respectively.

Serous adenocarcinoma cell lines, PEO1 and PEO4 (from Fergus Couch, Mayo Clinic) were grown in RPMI1640 (GE Healthcare) containing penicillin/streptomycin (Thermo Fisher Scientific) and 10% FBS (Thermo Fisher Scientific). Lines were validated by short tandem repeat profiling and BRCA2 mutation sequencing. The CCOC lines OVMANA and OVISE (both from David Huntsman, University of British Columbia, Canada), OVTOKO, JHOC9, and JHOC5 (all three from Tian-Li Wang, Johns Hopkins University), and ES2 and TOV21G (both from ATCC) were grown in RPMI1640 containing penicillin/streptomycin and 10% FBS. JHOC5 and OVTOKO were confirmed Mycoplasma-free prior to downstream experiments. Genomic profiles of CCOC cells were compiled from previous studies (20–22) and are summarized in Supplementary Table S4.

The APOBEC3B-specific (shA3B-1) and control shRNA constructs have been described previously (18). The second APOBEC3B-specific (shA3B-2) construct was purchased from Open Biosystems (5′-TGGTACAAATTCGATGAAA). UGI and empty lentiviral expression vectors were also described previously (23) and transductions were done with established protocols (24).

Single base substitution data generated from previous whole-exome sequencing through the Wellcome Trust Sanger Institute, were downloaded from the COSMIC Cell Line Project online database (http://cancer.sanger.ac.uk/cell_lines). APOBEC mutation enrichment scores were calculated using the hg19 reference genome and published methods (13). Enrichment score significance was assessed using a Fisher exact test with Benjamini–Hochberg FDR correction. Cell lines with q < 0.05 were considered enriched for APOBEC mutation.

Cells were collected, washed with PBS, and resuspended in 100-μL reducing sample buffer per one million cells (0.5 mol/L Tris-HCl pH 6.8, 1% 2-mercaptoethanol, 10% SDS, 50% glycerol). Samples were boiled for 20 minutes and fractionated by SDS-PAGE. Proteins were transferred to a PVDF-FL Membrane (Millipore Sigma) and blocked in 5% milk in PBS + 0.1% Tween 20 (PBST). Antibodies were incubated in blocking buffer, except fluorescent secondaries which were incubated in blocking buffer + 0.2% SDS. Membranes were imaged using a LI-COR Odyssey instrument or LI-COR Odyssey-Fc (LI-COR Biosciences). Antibodies used were: anti-A3B (5210-87-13, 1:1,000; ref. 25), anti-HSP90 (Millipore Sigma, 1:3,000), anti-mouse IgG-HRP (Cell Signaling Technology, 1:3,000), and anti-rabbit IgG-HRP (Cell Signaling Technology, 1:3,000).

RNA was extracted using a High Pure RNA Isolation Kit (Roche). Quantification was done using described primer/probe combinations for the APOBEC3 family members and the housekeeping gene TBP (10, 26). All qPCR was done according to the manufacturer-recommended protocols using a LightCycler 480 (Roche). GraphPad Prism 6 was used for statistical analyses.

DNA deamination activity was measured by harvesting whole-cell lysate and lysing in 150-μL lysis buffer per one million cells (25 mmol/L HEPES, 5 mmol/L EDTA, 10% glycerol, 1 mmol/L DTT, and 1 protease inhibitor tablet). Samples were sonicated three times for 3 seconds to ensure complete lysis. Lysates were incubated at 37°C for 3 hours with a DNA oligo containing a single TCA motif (5′-ATTATTATTATTCAAATGGATTTATTTATTTATTTATTTATTT-FAM) following established protocols (10). Reactions were fractionated using a 20% acrylamide-urea gel and quantified using a Typhoon FLA 7000 with ImageQuant Software (GE Healthcare). For uracil excision activity, a modified version of the protocol with no recombinant uracil DNA glycosylase (UNG) was done for 30 minutes at 37°C using a DNA oligo containing a single uracil (5′-AGCAGTATTUGTTGTCACGA-FAM).

OVTOKO (250 and 500 cells per well) and JHOC5 (200 and 400 cells per well) were seeded in 6-well plates. Cells were treated with the indicated drug concentrations (Selleckchem) every 48 hours for a minimum of seven treatments (14 days total). Once colonies formed in DMSO control conditions (14–16 days), plates were washed with PBS and stained with 50% methanol/0.5% crystal violet. Colonies were counted and colony formation efficiency (number of colonies relative to number of cells plated) was used to measure survival between wells with different seeding densities. Two-way ANOVA tests were used for dose/response curves and statistics were calculated with GraphPad Prism 6.

Immunostaining was done following established protocols (27). Cells were grown on coverslips in 6-well plates and treated with cisplatin or vehicle (DMF) for 24 hours. Cells were then fixed with 4% paraformaldehyde in PBS for 15 minutes at room temperature. Slides were blocked in 5% goat serum, 4% BSA in PBS for 1 hour at room temperature. Antibodies were diluted in blocking buffer. Antibodies used were: rabbit anti-RAD51 (Abcam, 1:1,000) and anti-rabbit IgG-Cy5 (Abcam, 1:1,000). All slides were treated with 0.1% DAPI to stain nuclei and mounted with 50% glycerol. Slides were imaged on an EVOS Microscope (Thermo Fisher Scientific). Nuclear foci formation was analyzed using ImageJ-FIJI software. One-way ANOVA statistical comparisons were done using GraphPad Prism 6.

FFPE tissue preparation and immunostaining was done following established protocols (25). Nuclear APOBEC3B immunoreactivity was visualized using the Aperio ScanScope XT at 40× magnification and quantified using the Aperio Nuclear Algorithm (Leica Biosystems). This program identifies cell nuclei in selected areas and categorizes immunostaining as negative, weak positive, moderate positive, or strong positive. APOBEC3B histoscore (H-score) was calculated as described using a linear formula: H-score = 1 × (%weak positive cells) + 2 × (%moderate positive cells) + 3 × (%strong positive cells; refs. 25, 28, 29).

PDX models were generated as described previously (30) and patient/PDX baseline characteristics were updated here according to recommended standards for PDX reporting (31). Primary (nonmetastatic) fresh tissues from two consenting patients with FIGO stage IC clear cell carcinomas (PH450 age 59 and PH578 age 63) were collected at the time of primary debulking surgery at Mayo Clinic (Rochester, MN, IRB 09-008768). All tissues were coded with a patient heterotransplant (PH) number to protect patient identity in accordance with the Mayo Clinic IRB and the Health Insurance Portability and Accountability Act. PDXs were developed by intraperitoneal injection of the patient tumor into female SCID beige mice (C.B.-17/IcrHsd-Prkdc^scid Lyst^bg; ENVIGO), in accordance with Mayo Clinic Institutional Animal Care and Use Committee established protocols (32). No enzymatic or mechanical tumor dissociation was performed. Mice were monitored by routine palpation for engraftment and tumors were harvested when moribund. Quality control included qPCR and IHC methods to rule out spontaneous murine tumors and human lymphoproliferative neoplasia as described previously (32).

To determine the in vivo platinum sensitivity of PH450 and PH578, tumors were revived from cryogenic storage into SCID-bg mice (30). Animals were observed for reengraftment and at a tumor size threshold of 0.3–0.5 cm² cross-sectional area by ultrasound (SonoSite S-Series with SLAx 13-6 MHz linear transducer), experiments were initiated with seven and three animals in the control groups and 24 and 10 animals in the chemotherapy groups (PH450 and PH578, respectively). Animals were treated with intraperitoneal carboplatin (25 mg/kg) and paclitaxel (7.5 mg/kg) weekly ×2, followed by 2 weeks of observation. Resistance to chemotherapy was defined as tumor growth similar to controls. In contrast, a chemotherapy-responsive tumor was defined by a growth trajectory that was significantly attenuated from controls. Statistical comparison of growth trajectories is as described in Supplementary Materials and Methods. All animals were sacrificed at the study endpoint or when moribund, whichever came first. Criteria for humane endpoints included a tumor size greater than 10% of animal body weight as estimated by ultrasound, weight loss greater than 20%, or body condition score ≤ 5 (33). For each PDX model, the corresponding patient's clinical chemotherapy response was defined as platinum refractory if the tumor grew on chemotherapy or platinum sensitive if there was no clinical, radiographic, or biochemical evidence for recurrent disease greater than 6 months following the completion of chemotherapy.

To begin to define the APOBEC3 proteins in CCOC, nine tumors from a deidentified retrospective cohort at the University of Minnesota (Minneapolis, MN) were used to determine the repertoire of expressed family member(s). Anti-APOBEC3B IHC was performed to analyze protein levels in FFPE tumor tissues (n = 7/9), and qRT-PCR assays were used to quantify mRNA levels of the full 7-gene APOBEC3 repertoire in corresponding FF tumor tissues (n = 9/9). Normal FF tissues from five of the same patients with CCOC were analyzed in parallel for comparison. Normal ovarian and fallopian FFPE tissues from three unrelated patients were analyzed as additional controls. Supplementary Table S1 summarizes information for these specimens.

All CCOC tumors showed characteristic cytoplasmic clearing in the majority of neoplastic cells, which were cuboidal, ovoid, and, less frequently, hobnail in shape (Fig. 1A). Nuclear pleomorphism and prominent nucleoli were present in some cancer cells in a subset of cases. All architectural patterns were observed including papillary, tubulocystic, solid, and mixtures. A rabbit anti-human APOBEC3B mAb (25) was used to stain FFPE tissues (n = 7). Strong immunopositivity was detected in the nuclei of a substantial proportion of cells in each tumor (Fig. 1A), consistent with the established nuclear localization of APOBEC3B (34, 35). The percentage of nuclear-positive cells varied three-fold among different CCOC tumors (H-scores from 19.4 to 57.9; Fig. 1B; Supplementary Table S1). In contrast, normal ovarian and fallopian epithelial tissues obtained from unrelated donors stained much lower (H-scores from 3.7 to 4.1; Fig. 1B; Supplementary Table S1). Although the rabbit anti-human APOBEC3B antibody used here has the potential to cross-react with APOBEC3A and APOBEC3G (25), the characteristic cytoplasmic localization of these other family members (35, 36) was not seen in the vast majority of CCOC tumor cells, further confirming that APOBEC3B is the dominant protein identified. Two cases had to be excluded from this and subsequent analyses because of limited neoplastic tissue (<20%, T022732) and marked lymphocytic infiltration (T100332). APOBEC3G is expressed in the cytoplasm of T lymphocytes (37–39), which can preclude APOBEC3B quantification in tumor cells. Even without two specimens, these results indicated that APOBEC3B may be the only DNA deaminase overexpressed in the nucleus of CCOC tumor cells.

As an orthologous and more comprehensive approach to determine the subset of human APOBEC3s expressed in CCOC, an established panel of qRT-PCR assays was used to quantify APOBEC3 family mRNA levels (37, 39, 40). All 9 patients with CCOC at University of Minnesota had FF tumor tissues for this analysis and five of nine also had FF normal tissues (including one fallopian tube). Consistent with IHC results above, APOBEC3B mRNA levels were uniformly low in normal tissues and significantly elevated in CCOC tissues, ranging from 0.12 to 0.91 relative to the housekeeper gene TBP (Fig. 1C; P < 0.05 by paired t test for the five tumors with corresponding normal tissue). Importantly, a strong positive correlation was evident between APOBEC3B H-score (above) and APOBEC3B mRNA levels (n = 5; Pearson = 0.9; P = 0.04), confirming IHC and qRT-PCR assays specificity (Fig. 1D; Supplementary Fig. S1).

Prior studies in breast, lung, and clear cell renal carcinoma have associated high APOBEC3B with poor outcomes, including OS and drug resistance (16–19). To ask whether this may also be the case for CCOC, a Mayo Clinic cohort was scored by IHC for APOBEC3B protein levels (n = 48; Supplementary Tables S2 and S3). As in the pilot cohort above, a range of nuclear APOBEC3B intensities was observed both in intra- and intertumoral samples (H-score range = 15.8–128; median = 66.2). Figure 2A shows representative photomicrographs of CCOC tumors with low, medium, and high APOBEC3B levels and Fig. 2B reports quantification of the full dataset. Unlike the smaller cohort with one tumor infiltrated by T lymphocytes, none of these tumors had significant T-cell infiltration.

APOBEC3B protein levels were not associated with clinical stage, suggesting its induction occurs early and possibly by multiple mechanisms (early vs late stage; P = 0.34 by Wilcoxon rank-sum test; Supplementary Fig. S2). We then explored associations between APOBEC3B protein levels and clinical outcomes. Surprisingly, in contrast to expectations from prior studies with other cancer types, higher APOBEC3B protein levels associated with better (not worse) PFS(Fig. 2C, left; log-rank test P = 0.085 split based on the median for illustration purposes; stage-adjusted Cox regression using continuous APOBEC3B H-scores, P = 0.026). Higher APOBEC3B levels were also associated moderately with improved OS (Fig. 2C, right; log-rank test P = 0.063; stage-adjusted Cox regression P = 0.057).

Although it is difficult to compare APOBEC3B levels and clinical associations between different cancer types treated with different therapies, a distinguishing feature of the Mayo Clinic cohort is that all patients with CCOC had received a platinum-based therapy following surgery. Together with prior studies showing that APOBEC3B-catalyzed DNA damage sensitizes cells to DNA repair inhibition (23, 41–43), a plausible molecular explanation for the unexpected clinical associations here with CCOC is that genotoxic stress imposed by APOBEC3B may sensitize cells to DNA cross-links caused by cisplatin and carboplatin. To test this idea, a panel of CCOC cell lines was assembled and characterized (Supplementary Table S4).

First, qRT-PCR was used to quantify mRNA levels of all seven APOBEC3 family members. Two serous ovarian adenocarcinoma cell lines, PEO1 and PEO4, previously shown to express little APOBEC3B (40), were included as negative controls (Fig. 3A–C; Supplementary Table S4). As for CCOC tumor specimens described above, APOBEC3A expression was undetectable or very low, and all other APOBEC family members, including APOBEC3B, were expressed at appreciable levels (except for APOBEC3H in ES2; Supplementary Fig. S3). APOBEC3B mRNA ranged from low in OVMANA to high in JHOC5 (0.31–3.1 units relative to TBP; Fig. 3A). APOBEC3B mRNA levels were largely proportional to protein and DNA cytosine deaminase activity levels (Fig. 3B–C). APOBEC3A was undetectable by immunoblot and APOBEC3G was expressed in several lines (Fig. 3B). Nevertheless, only APOBEC3B mRNA and protein levels correlated with DNA deaminase activity in whole-cell extracts (Fig. 3C).

Second, whole-exome sequencing data were analyzed to determine the subset of CCOC cell lines with enrichments of APOBEC signature mutations. APOBEC enrichment is defined as the proportion of C-to-T and C-to-G mutations in the APOBEC signature trinucleotide context (TCA or TCT) relative to the total number of cytosine mutations (13). A ratio > 1 indicates enrichment. Whole-exome sequence information was available for OVISE (470 total base substitutions) and OVTOKO (517 total base substitutions) and both showed significant APOBEC enrichment scores (1.47 and 1.64).

Third, to further inform studies with platinum-based drugs, additional genomic information was considered (Supplementary Table S4). Mutations in BRCA1 and BRCA2 have been associated with sensitivity to platinum-based drugs, however, six of seven cell lines were wild-type for BRCA1 and BRCA2 (44). The lone exception was OVMANA, which harbors a BRCA2 variant (L759V) with unknown significance (22). Pathogenic mutations occurred in ARID1A (5/7), PIK3CA (4/7), and TP53 (2/7) consistent with CCOC tumor genome–sequencing studies (20–22). ES2 and TOV21G were not considered for further experimentation because the former line was reported to have a mixed histotype (45) and the latter to be mismatch repair deficient (21). Thus, OVTOKO, which exhibits a clear enrichment for APOBEC signature mutations, and JHOC5, which has the highest APOBEC3B expression, were chosen for mechanistic studies.

To test whether APOBEC3B is required for the responsiveness of CCOC to platinum-based drugs, APOBEC3B was depleted in OVTOKO and JHOC5 by transduction with lentiviral constructs expressing short-hairpin RNA (shA3B-1 or shA3B-2). As a control, cells were transduced in parallel with a nontargeting shRNA (shCTRL). Robust APOBEC3B knockdown by shA3B-1 and shA3B-2 was confirmed by qRT-PCR, immunoblot, and DNA deaminase activity assays (Fig. 4A and B). APOBEC3B depletion by shA3B-1 was nearly complete, whereas residual mRNA and deaminase activity were still observed in cells expressing shA3B-2 despite undetectable protein by immunoblots. APOBEC3B knockdown alone did not affect growth rates or viability of either CCOC line as quantified by colony formation assays (representative images in Fig. 4C and quantification vs vehicle control in Fig. 4D and E). In contrast, endogenous APOBEC3B depletion compromised the sensitivity of each cell line to cisplatin and carboplatin (representative images in Fig. 4C and dose-response quantification in Fig. 4D and E). OVTOKO and JHOC5 showed five-fold and twofold higher IC₅₀s for cisplatin following APOBEC3B knockdown and even larger differences for carboplatin (ninefold and threefold higher, respectively). It is also noteworthy that lower drug concentrations caused significant differences in survival between shCTRL and shA3B cells (eg, 10 nmol/L cisplatin, 50% survival of JHOC5-shCTRL vs 100% survival of JHOC5-shA3B; and 60% survival of OVTOKO-shCTRL vs 90% survival of OVTOKO-shA3B; Fig. 4D and E).

Additional experiments were performed to identify the DNA lesions responsible for the APOBEC3B-dependent sensitivity of CCOC lines to platinum agents (i.e., APOBEC3B-catalyzed DNA uracils and/or downstream DNA breaks). The rationale for these experiments was based in part on our recent work showing that APOBEC3B-induced uracil lesions in chromosomal DNA are processed by UNG-mediated base excision repair (BER) and generate a strong DNA damage response evidenced by elevated DNA breakage (COMETs), γ-H2AX, and phosphorylated CHK1 (23). Additional justification was provided by studies showing that deamination events within 5′-CG interstrand cross-links cause an UNG-dependent sensitivity to cisplatin (46).

Therefore, to determine whether UNG-BER is required for APOBEC3B-dependent sensitivity to cisplatin, the universal uracil DNA glycosylase inhibitor (UGI) or an empty vector control was expressed in JHOC5-shCTRL and JHOC5-shA3B-1 cells. Potent UNG inhibition was confirmed by testing whole-cell extracts from each cell line in uracil excision activity assays (Fig. 5A). These engineered lines were then subjected to colony formation assays as above. In these experiments, APOBEC3B depletion increased the resistance of JHOC5 cells to cisplatin by five-fold (shA3B vs shCTRL in Fig. 5B). In comparison, UNG inhibition by UGI caused a similar level of resistance (shCTRL vs shCTRL+UGI in Fig. 5B). Moreover, simultaneous APOBEC3B depletion and UNG inhibition failed to cause further resistance. This epistatic relationship demonstrated that UNG-BER is required for APOBEC3B-mediated cisplatin sensitivity.

Finally, because cisplatin is known to induce DNA double-stranded breaks and RAD51 foci (47), the same panel of JHOC5 cells was used to quantify this DNA damage response marker. Cisplatin treatment caused significant increases in RAD51 foci in all conditions (Fig. 5C and D). However, the mean number of foci per cell was reduced significantly in both shA3B and UGI cells in comparison with shCTRL cells (respectively, 14 foci vs 9.2 and 7.7 foci per cell; Fig. 5D). The combination of APOBEC3B knockdown and UNG inhibition did not cause a further reduction in RAD51 focus formation indicating that the APOBEC3B/UNG platinum sensitivity pathway can be blocked by either manipulation (and that platinum agents induce RAD51 foci by at least one additional mechanism).

We next studied the chemotherapeutic responses of CCOC tumors in a PDX system. CCOC tumors were first expanded in immunodeficient mice and tested by IHC for APOBEC3B. As above for patient CCOC tumors and CCOC cell lines, a range of APOBEC3B protein levels was evident (eg, PH450, H-score = 40.5; PH578, H-score = 82.6; Fig. 6A and B). APOBEC3B-low and APOBEC3B-high tumors were minced and used for intraperitoneal inoculation of additional immunodeficient animals. Upon reaching 0.3–0.5 cm² by ultrasound, animals were assigned randomly to a control group or to an experimental group for carboplatin/paclitaxel chemotherapy. Tumor growth was monitored weekly by ultrasound. The untreated control groups for both the APOBEC3B-low and APOBEC3B-high PDXs nearly doubled in area within 3 weeks (Fig. 6C and D; black lines). While the APOBEC3B-low PDX PH450 grew slightly slower on chemotherapy than those in the untreated control group (Fig. 6C; red lines; 2 degree of freedom test of coincident growth curves by mixed effects ANOVA, P = 0.06); both arms continued to grow, doubling in area in less than 2 weeks. In contrast, the APOBEC3B-high PDX PH578 was responsive to chemotherapy and grew less than 25% on carboplatin/paclitaxel in the same timeframe (Fig. 6D; red lines; 2 degree of freedom test of coincident growth curves by mixed effects ANOVA, P = 0.03); area doubling times estimated from the mixed effects ANOVA were 24 and 110 days for the control and chemotherapy groups, respectively. In addition, these PDX responses to carboplatin/paclitaxel were consistent with the matched patients' clinical outcomes. Patient PH450 (APOBEC3B low) had primary platinum refractory disease with new lymphadenopathy and metastatic liver lesions at the completion of chemotherapy. In contrast, patient PH578 (APOBEC3B-high) had platinum-sensitive cancer and remained disease free more than 3 years after completing chemotherapy. Because both patients were stage IC at diagnosis and received six cycles of the same chemotherapy, this difference in clinical outcomes was likely due to inherent differences in chemotherapy sensitivity.

We show that APOBEC3B is overexpressed in CCOC primary tumors relative to normal tissues, and that CCOC cell lines express APOBEC3B at similarly high levels. IHC analysis was used to rank APOBEC3B protein expression in tumors, and then associate expression with clinical characteristics. These analyses revealed a surprising association between high intranuclear APOBEC3B protein levels and improved PFS and OS for patients with CCOC. These results were unexpected because previous findings in breast, lung, and kidney carcinoma supported an association between high APOBEC3B levels and poor clinical outcomes (16–19). However, a major feature distinguishing CCOC and these other cancer types is that all patients with CCOC received a platinum-based combination therapy. Together with prior work on synergy between APOBEC and inhibitors of the DNA damage response (23, 41–43), these observations suggested that the improved responses of patients with CCOC to platinum therapy may be due at least in part to genotoxic damage inflicted by APOBEC3B. This idea was tested with two different CCOC cell lines and two different platinum agents. In all cases, APOBEC3B depletion conferred resistance to cisplatin and carboplatin. Moreover, this resistance phenotype was recapitulated by UNG inhibition, demonstrating that the first step in uracil excision repair is also essential for an APOBEC3B-dependent therapeutic response. This mechanistic association was extended to CCOC PDXs, where high APOPEC3B associated with a stronger response to carboplatin combination therapy. Taken together, we conclude that APOBEC3B and UNG are molecular determinants of platinum sensitivity in CCOC. A schematic of interaction is shown in Supplementary Fig. S4.

Our studies are the first to link APOBEC3B expression to CCOC. Previous work reported elevated APOBEC3B mRNA in HGSOC but found no significant clinical associations (39, 40). The same reports also found no evidence for an APOBEC mutation signature in HGSOC. It is therefore possible that APOBEC3B expression alone may be insufficient to inflict an APOBEC mutation signature and that at least one other factor is required for APOBEC3B-catalyzed deamination events to become immortalized as mutations. CCOC may be a good system to investigate this possibility. A different study reported anti-APOBEC3B IHC for 55 CCOC tumors (48). However, this work relied on a commercial anti-APOBEC3B antibody that is unlikely specific due to uncharacteristic cytoplasmic staining and positive signal with rat tissues (despite the facts that APOBEC3B is nuclear and rodents lack APOBEC3B). Thus, the studies here are the first to report both APOBEC3B mRNA and protein expression in CCOC, as well as demonstrate that APOBEC3B activity is a key determinant in platinum responsiveness. Indistinguishable platinum resistance phenotypes caused by APOBEC3B knockdown and UNG inhibition revealed an additional layer of the underlying molecular mechanism.

Our studies raise the prospect of using APOBEC3B as a biomarker for CCOC. APOBEC3B may have diagnostic utility because it is often high in CCOC tumors and only detectable at low levels in normal ovarian and fallopian tube epithelial cells. APOBEC3B may also have predictive utility because it could help identify patients most likely to benefit from platinum-based therapies. Larger-scale studies will be needed to determine whether APOBEC3B-low CCOC tumors will be mostly or even fully resistant to platinum treatments, and whether APOBEC3B-high CCOC tumors may develop resistance mechanisms over time and, if so, establish underlying molecular mechanism(s). Importantly, the studies here may inform the design of clinical trials to consider APOBEC3B as a molecular determinant for responses to other types of therapy, particularly to targeted DNA damage response inhibitors that may synergize with APOBEC3B-catalyzed DNA damage in creating synthetic lethality (41, 43). An additional salient point is that CCOC tumors mostly have functional BRCA1, BRCA2, and related homologous recombination repair genes and that APOBEC3B provides an opportunity to create new synthetic lethal drug combinations with general chemotherapeutics such as the platinum agents described here, as well as targeted therapies that similarly sensitize tumor cells to APOBEC3B damage. Given the breadth of APOBEC3B overexpression in cancer (10, 11, 15, 49), these approaches have potential to benefit multiple tumor types.

R.S. Harris is an employee/paid consultant for and holds ownership interest (including patents) in ApoGen Biotechnologies. M.J. Maurer is an employee/paid consultant for Morphosys, Kite Pharma, and Pfizer, and reports receiving commercial research grants from Celgene and NanoString. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R.S. Harris, A.A. Serebrenik, P. Argyris, M.C. Jarvis, B.K. Erickson, X. Hou, S.H. Kaufmann

Development of methodology: A.A. Serebrenik, P. Argyris, M.C. Jarvis, B.K. Erickson, X. Hou, S.J. Weroha, E.P. Heinzen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.A. Serebrenik, P. Argyris, M.C. Jarvis, M. Bazzaro, B.K. Erickson, X. Hou, S.J. Weroha, S.H. Kaufmann

Analysis and interpretation of data (eg, statistical analysis, biostatistics, computational analysis): R.S. Harris, A.A. Serebrenik, P. Argyris, M.C. Jarvis, R.I. Vogel, B.K. Erickson, K.M. Goergen, M.J. Maurer, A.L. Oberg, X. Hou, S.J. Weroha, S.H. Kaufmann, Y. Huang, E.P. Heinzen

Writing, review, and/or revision of the manuscript: R.S. Harris, A.A. Serebrenik, P. Argyris, M.C. Jarvis, R.I. Vogel, B.K. Erickson, A.L. Oberg, X. Hou, S.H. Kaufmann, E.P. Heinzen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.S. Harris, A.A. Serebrenik, W.L. Brown, B.K. Erickson, S.-H. Lee, K.M. Goergen, X. Hou, S.J. Weroha, S.H. Kaufmann, Y. Huang, E.P. Heinzen

Study supervision: R.S. Harris, A.A. Serebrenik, X. Hou, S.J. Weroha, S.H. Kaufmann

We thank David Huntsman and Tian-Li Wang for sharing CCOC cell lines, Brian Dunnette for assistance with Aperio software, and Kathleen Gavin, Melissia Geller, Tim Starr, and Elizabeth Swisher for their thoughtful comments. We acknowledge the UMN Clinical and Translational Science Institute tissue procurement facility (BioNet) for biospecimen acquisition, which is supported by the NIH's National Center for Advancing Translational Sciences (UL1TR002494). These studies were supported in part by the Minnesota Ovarian Cancer Alliance (to S.J. Weroha, S.H. Kaufmann, and R.S. Harris), P01-CA234228 (to R.S. Harris), P50-CA136393 (to S.J. Weroha, A.L. Oberg, and S.H. Kaufmann), R01-GM130800 (to M. Bazzaro), and University of Minnesota College of Biological Sciences and Academic Health Center (to R.S. Harris). NIH training grants and career development awards provided salary support for MCJ (T32 CA009138) and BKE (K12 HD055887). R.S. Harris is the Margaret Harvey Schering Land Grant Chair for Cancer Research, a Distinguished McKnight University Professor, and an Investigator of the Howard Hughes Medical Institute.

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.