A phase 2 study of dasatinib in recurrent clear cell carcinoma of the ovary, fallopian tube, peritoneum or endometrium: NRG oncology/gynecologic oncology group study 0283

For both BAF250a cohorts, the study closed to accrual at the end of Stage 1 for futility of the dasatinib regimen. A total of 35 patients enrolled between February 03, 2014, and August 10, 2016. Four patients were considered ineligible, and three patients were never treated ( Figure 2 ). Of 28 evaluable patients, 15 had retained BAF250a and 13 patients had loss of BAF250a. Patient characteristics are summarized in Table 1 . The median age at study entry was 59.9 years (range: 35–86.8). The median age in patients with retained BAF250a status was 62.1 years (range: 43.8–79.1) and in those with loss of BAF250a was 56.9 years (range: 35–86.8). The majority of patients had cancer of the ovary, fallopian tube, or peritoneum (82.2%); the remaining 17.9% presented with endometrial cancer. The proportion of patients with ovary, fallopian tube, or peritoneum as the primary disease site versus the endometrium was lower in the BAF250a retained versus loss status (66.7% vs 100%). Descriptive statistics for treatment emergent adverse events are shown in Table 2 . The most common AEs were grades 1 and 2 including: nausea (n=17, 60.7%), fatigue (n=13, 46.4%), anemia (n=11, 39.2%), diarrhea (n=10, 35.7%), abdominal pain (n=10, 35.7%), and vomiting (n=7, 25%). Twenty (71%) patients had grade 3 AEs, most commonly anemia (n=8, 28.6%), fatigue (n=7, 25%), dyspnea (n=5, 17.9%), hyponatremia (n=3, 10.7%), pleural effusion (n=3, 10.7%) and vomiting (n=3, 10.7%). Two (7.1%) patients had grade 4 AE anemia. No treatment-related deaths were reported. The three deaths occurring on protocol therapy were attributed to disease progression. No new safety signals of dasatinib were identified. Twenty-four (85.7%) patients had a cumulative total of 33 dose modifications. There were 14 dose reductions in 11 patients. Of the 12 drug discontinuations, three were attributed to disease progression, one patient refused further treatment, and eight were attributed to adverse events including diarrhea (n=3), fatigue (n=2), nausea (n=1), anemia (n=1), GI bleed (n=1), GI toxicity (n=1), and abdominal pain (n=1). The ORR in both cohorts (BAF250a-loss and -retained) combined was 3.6% (one of the 28 patients; 1-sided binomial p value =0.95). The responding patient was in the BAF250a-loss cohort and had a partial response with duration of 5.65 months. No patients with retained BAF250a had an objective response. In the best overall response analysis 8 (28%) patients (4 for each cohort) had stable disease. The median duration of stable disease was 5.67 months (range 3.55–20.73 months) and 4.65 months (range 1.58–17.84 months) for loss of BAF250a and retained BAF250a, respectively. Fifteen (53.6%) patients had disease progression within the first 2 cycles (56 days) on protocol therapy (7 and 8 patients, in the BAF250a-loss and - retained cohorts, respectively). PFS and OS are shown in Figure 3 . The hazard ratios (HR, 95% CI) for BAF250a-loss (reference: BAF250a-retained) was 0.8 (95% CI: 0.37–1.73) and 0.48 (95% CI: 0.21–1.10) in PFS and OS respectively. Results for the evaluation of BAF250a IHC and ARID1A mutation status are shown in Table 3 . All 35 patients had consented to biospecimen analysis, whereby whole blood samples for matched normal and FFPE tumors were collected. Next generation sequencing was successful in 26 of the 35 patients, of which 11 (42.3%) had loss of BAF250a status and 15 (57.7%) had retained BAF250a as determined by IHC. An ARID1A mutant genotype was found in 11 (91.7%) of the loss of BAF250a samples and in 2 (14.2%) of the BAF250a retained samples. Of the 28 patients who were treated and evaluable for the study primary endpoint ( supplemental Table 2 ) five patients were missing ARID1A mutation status due to a central sequencing failure. Of the remaining 23/28 patients, 11 (47.8%) had loss of BAF250a status and 12 (52.1%) had retained BAF250a. An ARID1A mutant genotype was found in 9/11 of loss of BAF250a samples and in 3/12 of retained samples. Next generation sequencing highlighted other common cancer-associated mutations ( Figure 4 ). PIK3CA mutations were more prevalent in ARID1A mutant tumors, while TP53 mutations were more prevalent in ARID1A wild-type tumors.

The study enrolled patients with recurrent or persistent ovarian, fallopian tube, peritoneal, and endometrial CCC, and measurable disease. Eligibility criteria required clear cell histomorphology of at least 50% in the primary tumor or a histologically documented recurrence. Tumors had to be negative for WT-1 antigen, and estrogen receptor (ER) antigen by IHC. All patients must have received at least one platinum-based chemotherapeutic regimen prior to enrollment and were at least 3 weeks from receipt of last chemotherapy or radiation. Adequate organ function was required on the following parameters: bone marrow (leukocytes ≥ 3000/mcl; absolute neutrophil count ≥1500/mcl; platelets ≥100000/mcl); renal (serum creatinine ≤ 1.5X the institutional upper limit of normal [ULN] or creatinine clearance ≥ 60 ml/min/1.73m 2 ); and hepatic (bilirubin ≤ 1.5 ULN and AST and ALT ≤3X ULN); cardiac (QTc interval on ECG ≤480 ms [Fridericia correction]). A GOG Performance Status of 0–2 and 0–1 for patients who had received one prior or ≥ 2 prior regimens, respectively, were required BAF250a IHC screening results were also required prior to enrollment. Patients with prior therapy including dasatinib, imatinib, or nilotinib were excluded, along with pregnant patients and those under the age of 18. Patients with a history of cardiac disease including uncontrolled angina, congestive heart failure, or myocardial infarction within 6 months prior to study entry, congenital long QT syndrome, and clinically significant ventricular arrhythmias were ineligible. As dasatinib can cause fluid shifts, patients with symptomatic effusions (pleural, pericardial, or peritoneal), or requiring a procedure for symptomatic effusions within 4 weeks of start of dasatinib were ineligible. Patients taking concomitant medications that were strong inducers or inhibitors of the CYP3A4 enzyme were required to stop 2 weeks prior to first dose of dasatinib, following confirmation of compliance with all other eligibility criteria. Patients who could not tolerate discontinuation of H2 blockers were ineligible. All enrolled patients signed an informed consent form. The trial was done in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. Patients received dasatinib 140 mg by mouth once daily on a continuous basis, in a 28-day cycle. Dasatinib tablets were to be swallowed whole and taken with at least 8 ounces of water. Treatment continued until progression of disease was demonstrated, the patient experienced unacceptable treatment-related toxicity or withdrawal of consent. Patients were monitored for adverse events for 30 days following treatment discontinuation. Treatment interruptions of up to 4 weeks were permitted for resolution treatment-related toxicities. The first dose level reduction was to 100mg once daily and the second to 70mg once daily. Patients requiring more than two dose level reductions due to drug-related adverse events were removed from study. Five years of follow-up was planned to support the overall survival endpoint. Adverse events were graded using Common Terminology for Adverse Events (CTCAE) v.4.0. Measurable disease, objective response and disease progression were evaluated by local investigators using by RECIST v 1.1 [ 26 ]. Patients who had a tumor evaluation following the first treatment cycle were considered evaluable for objective response. Confirmation was required for complete and partial response (CR/ PR) at least 4 weeks after initial documentation of best overall response. Patients without confirmed CR/PR were counted as treatment failures. The primary endpoint was objective response rate (ORR), defined as the proportion of evaluable patients with confirmed CR or PR. Progression free survival (PFS) was defined as the number of months from enrollment to documentation of disease progression or death, whichever occurred first. Overall survival (OS) was defined as the number of months from enrollment to death from any cause. For both PFS and OS, patients without document progression or death events were censored on the date of last follow-up. The study schema is shown in Figure 1 . The guiding principle for this study was to limit the number of patients treated with clinically ineffective therapy and to estimate the efficacy with reasonable precision for a clinically active agent. After enrollment, patients were stratified into cohorts with retained BAF250a or loss of BAF250a as determined by IHC in a central laboratory. Within each BAF250a subgroup, the primary objective was to assess the clinical activity of dasatinib. The primary endpoint was ORR. As designed, each cohort study had 90% power to detect an ORR of 30% or greater, assuming 10% ORR among historical controls, using one-sided α=0.1 level binomial test. Because no historical response rate data were available for these patient populations, we tested whether the response rate was indicative of treatment activity for dasatinib. The cohorts were tested separately. No cohort comparisons and no type 1 error adjustments for multiple testing were planned. In each BAF250a treatment efficacy was assessed using identical single-arm, two-stage, historically controlled optimal-flexible designs, as developed by Chen and Ng [ 27 ]. The designs included early stopping for futility of the dasatinib regimen. Stage 1 and Stage 2 target accruals and critical accrual values for number of patients and responders (CR/PR) are shown in Table 1 . If the requisite number of responders was not met in Stage 1, that cohort would close and considered not worthy of further evaluation for that cohort. Otherwise, accrual would open to Stage 2. Dasatinib was expected to be active among patients with loss of BAF250a, so the standard decision rules were modified to allow Stage 2 accrual to the BAF250a-loss cohort if the BAF250a-retained cohort opened to Stage 2 accrual. A total of 62 patients were expected to enroll, with about 40% having BAF250a-loss. ORR confidence intervals (2-sided 95%) were estimated using Jeffreys method [ 28 ] . PFS and OS distributions were summarized using Kaplan-Meier methods. Hazard-ratios estimates comparing the BAF250a-loss (vs -retained) cohorts were obtained from a proportional hazards model with covariate effects for BAF250a group. Formalin-fixed paraffin-embedded (FFPE) tumor tissue slides were submitted to a Clinical Laboratory Improvement Amendments (CLIA) certified immune-pathology laboratory at Memorial Sloan Kettering Cancer Center. Only one specimen was required for this test, with the more advanced and most recently obtained specimen used. Clear cell carcinoma and normal tissue were required to be present on the slide. BAF250a status was determined by IHC. Slides were prepared and stained with BAF250a antibody (Sigma, St Louis, US), as previously described. [ 15 , 29 ]. Immunoreactivity in any nuclei was deemed positive (or retained). Loss of BAF250a expression was defined as diffuse loss of expression throughout the tumor (null pattern) or a null pattern in a distinct geographic focus in a background of retained expression. Hematoxylin and Eosin staining was used to confirm presence of normal tissue on the slide to serve as a normal control. BAF250a nuclear immunoreactivity is seen in non-neoplastic cells, such as endothelial cells and fibroblasts, serving as a positive internal control. To examine the agreement between BAF250a IHC and ARID1A status, next generation exon-capturing sequencing was used in whole blood samples and FFPEs. Methods are described in the MSK-IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets) protocol [ 30 ].

In this study dasatinib was not effective as a single-agent treatment for recurrent or persistent ovarian, fallopian tube, primary peritoneal, and endometrial CCC with or without loss of BAF250a. Only one of the 28 patients enrolled in Stage 1 achieved a partial response. Both cohort studies closed early to futility of the experimental regimen. The translational component of this trial examined the concordance between BAF250a expression and ARID1A mutational status. Our rationale was to develop a reliable method for tumor selection that could overcome some of the challenges of next generation sequencing. We indicated the utility of BAF250a immunochemistry in selecting for ARID1A mutant cancers. In addition, next generation sequencing showed a high rate of ARID1A and PIK3CA co-mutations in ovarian clear cell carcinoma (OCCC) tumors. Coexistence of these two mutations has been associated with tumorigenesis in OCCC animal models and shown to be prevalent in endometrial CCC patient-derived tumors. Dasatinib is a second-generation competitive inhibitor of tyrosine kinases including the Bcr/Abl fusion, cKIT, PDGFR, erythropoietin producing hepatocellular (Eph) receptors, and Src and Btk tyrosine families. In preclinical in vitro models, Miller et al ., demonstrated the sensitivity of ARID1A -mutant cells to dasatinib when compared to wild-type ovarian cancer cells [ 17 ]. Furthermore, mice treated with dasatinib had superior OS compared to vehicle-treated animals and significant tumor shrinkage was observed. The proposed mechanism that made these models sensitive to the drug was addiction to Src family tyrosine kinase,YES1, which was previously shown to be sensitive to dasatinib inhibition [ 31 , 32 ], including in ovarian cancer cell lines [ 33 ]. Enrichment of YES1 was observed in the mutant in vitro models. Activation of YES1 was reduced in cells exposed to dasatinib, although no change in the activity of ARID1A was observed when compared to wild-type cells. Although the results from the aforementioned Miller study were promising and warranted further investigation of datasinib in OCCC and endometrial CCC patients with loss of ARID1A gene expression, we failed to translate those results into efficacy in our patient population [ 17 ]. This may be due to additional molecular mechanisms or genetic events that were not tested in animal models. For example, at the time of trial activation, the involvement of YES1 as a potential mechanism for dasatinib sensitization was unknown. When overexpressed, YES1 has been shown to be a driver for chemotherapy resistance in colon and prostate cell models [ 34 , 35 ]. In addition to YES1, high expression of Eph2A, a member of the Eph receptor tyrosine kinase family, is also associated with poor clinical outcomes in ovarian cancer and exhibits in vitro sensitivity to dasatinib [ 33 , 36 ]. As we did not test for the status of YES1 or Eph2A tumor expression prior to enrollment, a potential disturbance in YES1 or Eph2A pathways could have conferred resistance to dasatinib. It is also possible that YES1 and/or Eph2A are enriched in mutant ARID1A chemoresistant tumors. Finally, although OCCC and endometrial CCC have the highest age-standardized incidence rate among Asians, only 11% of the patients enrolled on our study were Asian, while the vast majority (82%) were White [ 1 , 6 ]. This highlights the need for more concerted efforts to ensure accurate representation of the patient population most affected by the disease [ 37 ]. Many trials evaluating dasatinib as a single agent in solid tumors also failed to validate the drug as an effective therapy [ 38 – 44 ]. Specifically, a phase II trial of dasatinib for epithelial ovarian cancer did not show efficacy in this patient population [ 24 ]. Herein, we offer a potential explanation for the failure of dasatinib as a therapeutic agent in patients with OCCC and endometrial CCC based on the limited literature aiming to elucidate mechanistic drivers of these malignancies. There is still a significant unmet need to understand the biologic role of ARID1A and other molecular events associated with this mutation. We increasingly recognize the importance of treating OCCC and endometrial CCC as biologically distinct diseases from other epithelial tumors. As such, targeted therapies effective in other subtypes may not be successful in these tumors. More studies are needed that specifically target these rare gynecologic variants. Our study shows that biomarker driven studies in such rare populations are possible through the NCI mechanism.

Gynecologic cancers have historically been managed according to their presumed site of origin, without regard to the underlying tissue histology [ 1 ]. It is well understood that the initial presentation, natural history, and response to therapy can vary widely according to histologic subtype [ 2 ]. Rigorous pathologic investigation, and unbiased genomic profiling of these cancers, are elucidating the underlying biology of these disparate malignancies. Clear cell carcinomas (CCC) of the gynecologic tract are one such rare histologic subtype. Clear cell histology is associated with chemotherapy refractoriness and poorer survival compared to other more common subtypes [ 3 , 4 ]. Epidemiologically, CCCs are associated with a history of endometriosis and demonstrate similar tumor genomic profiles regardless of their site of origin [ 5 – 7 ]. Genomic studies of CCCs, as well as endometriosis-associated endometrioid carcinomas, identified a high incidence of somatic mutations in the tumor suppressor gene AT-rich Interactive Domain-containing protein 1A ( ARID1A ) [ 6 ]. The ARID1A gene encodes BAF250a (ARID1A), a protein that forms a subunit of several different ATP-dependent chromatin remodeling SWItch/Sucrose Non-Fermentable (SWI/SNF) protein complexes [ 8 ]. The SWI/SNF complex is an epigenetic regulator that plays a key role in nucleosome repositioning on DNA during early embryonic development and tissue differentiation [ 9 , 10 ]. ARID1A mutations occur in a broad array of cancers, endometriosis-associated cancers having the highest estimated frequency of 30–50% depending on histologic subset [ 11 – 13 ]. Loss of ARID1A seems to occur early during the endometriosis-to-cancer progression [ 14 , 15 ]. Specifically, studies have identified ARID1A mutations in atypical endometriotic cysts found in patients with these carcinomas. These findings further support that ARID1A loss is an early event in oncogenesis [ 12 ]. Several mechanisms by which loss of ARID1A drives cancer transformation have been proposed. Loss of integrity of the SWI/SNF nucleosome remodeling complex leads to defective control of distal regulatory elements. These “enhancers” have emerged as early drivers of cell transformation in several cancer types, including in ovarian clear cell cancer cell models [ 16 ]. Cancer-associated somatic ARID1A mutations in ovarian clear cell carcinoma (OCCC) are often characterized by frame-shift insertion or deletion mutations or nonsense mutations that prematurely truncate the protein [ 12 , 17 ]. Mouse models show that loss of ARID1A leading to altered phenotype when coupled with loss of PTEN. However, PTEN alone was not a driver of altered phenotypes in ovarian tissue. This finding suggests collaboration between ARID1A and the PI3K/AKT pathway in tumorigenesis [ 18 ]. This theory is further supported by studies that show aberrant PI3K pathway activation in approximately half of OCCCs [ 6 , 15 , 19 ]. Clinically, both clear cell histology overall and loss of ARID1A are associated with chemoresistance and inferior disease-free survival [ 4 , 6 , 20 ]. Loss of BAF250a expression as detected by immunohistochemistry (IHC) correlates well with the presence of an ARID1A mutation with estimated sensitivity and specificity of 73% and 89%, respectively [ 12 , 15 ]. As such, considerable effort has been dedicated to defining potential mechanisms of synthetic lethality associated with loss of ARID1A function and loss of BAF20 expression in these cancers [ 21 ]. Preclinical studies proposed inhibition of EZH2, PARP, ATR and cell cycle modulators as potentially effective therapies. High-throughput cell-based drug screening was conducted by Miller et al . to identify drugs that would be predicted to deliver an ARID1A -mutant-selective effect through estimation of tumor sensitivity to 68 drugs in ARID1A -mutant and parallel wild-type ovarian clear cell cancer cohorts [ 17 ]. Synthetic lethality was demonstrated between the tyrosine kinase inhibitor (TKI), dasatinib, and ARID1A mutations in OCCC cell lines [ 17 ]. The subsequent in vivo studies endorsed the therapeutic activity of the TKI in ARID1A-aberant OCCC xenografted into immunocompromised mice. Dasatinib, a potent synthetic oral dual BCR-ABL and SRC family TKI, is FDA approved for the treatment of Philadelphia positive (Ph+) chronic myeloid leukemia and Ph+ acute lymphoblastic leukemia [ 22 , 23 ]. Dasatinib showed minimal activity in a phase II study of 34 patients with recurrent ovarian cancer, but only two of these patients had clear cell histology [ 24 ]. Dasatinib was also explored in combination with carboplatin and paclitaxel in a phase I study in advanced or recurrent ovarian cancer (no patients with clear cell histology) with a response rate of 40% for the triplet [ 25 ]. The authors, Secord et al ., concluded that discovery of biomarkers would be critical to identifying patients most likely to benefit from dasatinib. To explore its activity in ARID1A -mutant CCC we conducted a single-arm phase 2 clinical trial of dasatinib in patients with recurrent or persistent ovarian, fallopian tube, primary peritoneal, and endometrial CCC. After enrollment, patients were assigned to biomarker-defined subgroups, based on retained or lost expression of BAF250a determined by IHC. Translational endpoints included broad next generation sequencing to assess concordance of protein expression with underlying genetics and treatment outcomes.