Survival Following Chemotherapy in Ovarian Clear Cell Carcinoma Is Not Associated with Pathological Misclassification of Tumor Histotype

Selection of patients with OCCC with evaluable disease

Ovarian cancer, including OCCC, is typically treated with debulking surgery followed by adjuvant chemotherapy. To evaluate survival outcomes following platinum-based chemotherapy in patients with OCCC, we first excluded those with tumors who may have been cured by surgery alone and selected patients with residual disease (stages II–IV) following primary debulking surgery and who subsequently received adjuvant chemotherapy. We also included patients who were diagnosed with disease recurrence on/after adjuvant chemotherapy despite having no macroscopic residual disease after primary surgery, irrespective of their initial stage (Fig. 1; Supplementary Fig. S3). Disease relapse in these patients suggested that these cancers were inherently chemoresistant.

Given the stringent selection criteria and that OCCC is an uncommon histotype, we chose to ascertain patients from both a large national cohort study of Australian ovarian cancer patients (AOCS) and a high volume Japanese university clinical network (JIKEI), where the incidence of OCCC is higher than in Caucasian populations. One hundred and thirty-three patients (32%) from an initial cohort of 413 met our selection criteria. We further filtered based on the availability of material for pathologic, IHC, and molecular analyses, resulting in a final cohort of 68 patients. Despite differences in ethnicity and location, there were no significant clinical differences between the AOCS and JIKEI cohorts, although there was a trend towards younger patients among those from Japan (Table 1). Seven of 68 patients were mixed OCCC cases. Six of seven cases had clear cell and serous components, and one case, A-2486, consisted of three histologic regions: clear cell, serous, and endometrioid (Supplementary Table S3). Clinicopathologic characteristics are summarized in Supplementary Fig. S3. Although most known prognostic factors including age of diagnosis, stage, and extent of residual tumor were not significantly associated with survival in this cohort, only stage was correlated with PFS but not OS (Supplementary Fig. S4).

IHC reclassification of mixed OCCC as likely HGSOC

Morphological classification of OCCC recognizes tumors with uniform clear cell histologic features (pure) and those with components of clear cell and other histotypes (mixed), usually serous or endometrioid (18, 19). Our cohort of 68 patients was identified following initial diagnostic and subsequent expert morphologic review of diagnostic samples. Recently, probabilistic algorithms applied to panels of IHC markers have been used to more effectively subtype ovarian cancer in large patient cohorts (16, 22). We therefore applied this validated algorithm, using staining of WT1, HNF1B, Napsin A, p53, and ER to confirm or refute histotype. A majority of tumor samples showed a pattern of staining consistent with OCCC and all but one of these were of pure morphology (Fig. 2A; Supplementary Fig. S5). Nine samples (13.2%) had a staining pattern consistent with HGSOC, including aberrant p53 staining, positivity for WT1 and an absence of OCCC markers HNF1B and Napsin A. While most were of mixed histology, one third were classed as pure. Therefore, although mixed histology is highly suggestive of a misclassified HGSOC tumor, pure morphology was not an entirely reliable indicator of true OCCC status.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

IHC profiling the study cohort. A, Cases were reclassified as confirmed OCCC, Likely HGSC, or Unclassified using five IHC markers (red) and standardized scoring. Additional IHC stains are shown in black. Antibodies to MMR proteins included those to MLH1, MSH2, MSH6, and PMS2. B, Lack of association of pathologic and IHC reclassification with patient outcome. C, Significant correlation of progression free and OS with the level of expression in HNF1B.

For another five cases without Napsin A and HNF1B staining the possibility of endometrioid carcinomas could not entirely be excluded and those cases were regarded as uncertain/unclassified. The re-classification of all three groups (clear cell, likely HGSOC, uncertain) based on the algorithmic markers was consistent with observed immunoreactivity to ARID1A, which is typically lost in OCCC rather than HGSOC (Fig. 2A; Table 2; refs. 9, 10).

The proportion of likely HGSOC within the cohort was similar to reported rates of response to chemotherapy in OCCC (2, 5), suggesting that misclassified samples could account for occasional responses seen in some patients. We first considered possible confounders of age, stage, and extent of residual disease but found no statistically significant difference between the immunohistochemically-defined subtypes (Table 2). We then performed Kaplan–Meier survival analyses, but surprisingly found no difference in PFS or OS between patients with tumors of pure or mixed morphologies, or tumor classifications assigned immunohistochemically (Fig. 2B). Importantly, survival benefit was not confined to those likely HGSOC that were originally misclassified as OCCC.

HNF1B loss was associated with longer PFS and OS in OCCC

A number of IHC markers, including those used in this study, have been associated with clinical outcome in OCCC. However, most prior studies have not focused on those with an assessable tumor burden to evaluate clinical outcomes following chemotherapy. We therefore assessed known prognostic factors (age of diagnosis, stage, extent of residual tumor) and expression patterns of a panel of IHC markers (HNF1B, ARID1A, Ki67, FOLR1, IGF2BP3) for associations with clinical characteristics and survival in the 54 confirmed patients with OCCC (Supplementary Table S4; Supplementary Figs. S6 and S7). All known prognostic factors were not significantly associated with survival in 54 confirmed OCCC cases (Supplementary Fig. S6). Loss of ARID1A expression was frequently observed in cases with advanced disease and residual tumor compared with early stage and completely resected cases (P = 0.035 and 0.027, respectively; Supplementary Table S4). Interestingly, Ki67 scores in JIKEI cases were significantly higher than those in AOCS cases (P = 0.020), but we are uncertain whether this is related to differences in ethnicity between patient populations or potential differences in tissue processing (Supplementary Table S4). Interestingly, loss of HNF1B expression was significantly associated with longer PFS (P = 0.0194) and OS (P = 0.0395; Fig. 2C). To our knowledge, there has been no previous study showing an association between loss of HNF1B expression and favorable outcomes in OCCC.

We also investigated the correlation of IHC markers with outcomes in some clinical subgroups. In 24 OCCCs with residual tumor at primary surgery, only loss of HNF1B showed significant correlation with longer PFS (P = 0.0152) and a trend towards longer OS (P = 0.0948; Supplementary Fig. S8). In 30 OCCCs with no residual tumor at primary surgery, loss of ARID1A and higher Ki67 expression were associated with shorter PFS (P = 0.0378 and 0.0257, respectively), but not significantly with OS (P = 0.0727 and 0.1351, respectively; Supplementary Fig. S9).

Patients with OCCC with stage I tumors generally have a favorable prognosis, however a relatively small proportion of stage I patients experience recurrence on or after first-line chemotherapy. Therefore, we investigated the relationship of PFS with expression of IHC markers in 13 patients with stage I disease. A high level of Ki67 and loss of ARID1A were associated with shorter PFS (P = 0.0029 and 0.0716, respectively) and OS (P = 0.0216 and 0.0428, respectively; Supplementary Fig. S10).

Concordant IHC and mutational profiles in OCCC samples

OCCC is characterized by frequent mutations in ARID1A and PIK3CA, and a low frequency of TP53 mutations (7, 9, 10). Given that some previous studies may have erroneously included non-OCCC samples in their analysis, we used our immunopathologically-characterized cohort to better define the frequency of mutations in genes commonly altered in OCCC. We performed targeted sequencing using a custom capture panel of 103 genes that have previously been reported as mutated in OCCC or HGSOC (Supplementary Table S2).

We identified 145 likely pathogenic variants in 28 genes, present in 63 of 68 cases (92.6%; Fig. 3A; Supplementary Table S5). Mutations in ARID1A, PIK3CA, PPP2R1A, RAS pathway members (KRAS and BRAF), and TP53 occurred most frequently. To further validate findings, we examined the consistency between IHC and mutational results using informative antibodies for ARID1A and p53. Abnormal IHC staining patterns and presence of mutations in TP53 and ARID1A were strongly associated (P < 0.0001 in both TP53 and ARID1A; Supplementary Fig. S11). Sensitivity and specificity were 0.82 and 1.00 for TP53, and 0.88 and 0.91 for ARID1A, respectively.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Mutational profiling of OCCC. A, Likely pathogenic alterations detected by custom capture sequencing of tumor DNA from the study cohort. Colored boxes indicate types of mutations detected. Cross-hatching indicates existence of other variants. Vertical bars indicate number of mutated genes, and horizontal bars show percentages of cases with mutation. B, Association of PFS with mutational status of TP53, ARID1A, and PIK3CA. C, Frequently altered pathways in 54 CCCs. PI3K pathway alterations include PIK3CA, PTEN, PIK3R1, and PIK3R2. RTK mutations include mutations in ERBB2, ERBB3, ERBB4, EGFR, MET, FGFR2, PDGFRB, and FLT4. RAS pathway mutations have been grouped as KRAS and BRAF. SMI/SNF pathway alterations include ARID1A, ARID1B, and ARID5B.

ARID1A, PIK3CA, PPP2R1A, and BRAF mutations were restricted to tumors that had been classified immunohistochemically as OCCC (Table 2). Mutations in KRAS were observed in OCCC tumors, but also in one likely HGSOC and three unclassified cases. Mutations in TP53 were observed in all likely HGSOC cases (100%), consistent with almost invariant TP53 mutation in this histotype (6, 7). By contrast, TP53 mutation was detected in only five of 54 OCCC cases (9.3%). We tested whether the small proportion of TP53 mutation-positive OCCC were associated with differential PFS. Kaplan–Meier analysis showed a nonsignificant trend to longer PFS (Fig. 3B). ARID1A mutations tended to be associated with shorter PFS, although this did not reach statistical significance, possibly due to limited sample size. ARID1A mutations have been reported to be associated with poorer prognosis and chemo-resistance in some studies, whereas others have failed to confirm this (13–15). PPP2R1A mutations were only seen in tumors from patients who were optimally debulked (Supplementary Table S4). No other frequent mutations, including PIK3CA, KRAS, and PPP2R1A, were associated with differences in PFS (Fig. 3B; Supplementary Fig. S12A).

Components of the SWI/SNF complex, including ARID1A, ARID1B, and ARID5B, were the most commonly altered signaling pathway, with 63% of tumors tested having at least one mutation in a pathway component (Fig. 3C). Mutations or amplification in PI3K family members, RTK and RAS pathways were common, detected in 51.9%, 22.2%, and 16.7% of tumors, respectively. RTK and RAS mutations were largely mutually exclusive. Overall, we did not observe differences between OCCC patients with and without clearly discernable pathway aberration, for either PFS or OS (Supplementary Fig. S12B).

We next evaluated the association between frequent mutations and outcomes in clinical subgroups. In 24 cases with residual tumor at primary surgery, no frequent mutations showed significant differences in PFS, but ARID1A mutation tended to be associated longer survival contrary to previous reports (Supplementary Fig. S13). By contrast, in 30 cases with optimal debulking, ARID1A mutations were associated with shorter PFS, supporting the observation that loss of ARID1A expression is associated with shorter PFS (Supplementary Fig. S14). In the subset of patients with stage I disease, we did not identify any associations between mutation status and PFS (Supplementary Fig. S15).

To further support the conclusion that mixed tumors were mostly misclassified HGSOC rather than two distinct coexisting tumors, we evaluated the mutational profile of three cases in which we performed targeted sequencing on separate regions of clear cell or serous histology. In all patients, TP53 mutations were identical in the matching regions showing that they had shared origins rather than being collision tumors (Supplementary Fig. S16).

HGSOC and OCCC display distinct DNA copy number profiles (23–25). To examine the correlation between copy number alterations and IHC subtype, we compared DNA gains and losses in 54 OCCCs and 9 likely HGSOC by making use of read depth from the targeted sequencing data. Although a relatively low-resolution method given the number of genes assayed, unsupervised hierarchical clustering cleanly separated the groups as defined by IHC analysis (Supplementary Fig. S17A). OCCC samples segregated into three groups but these were not associated with differential clinical outcome (Supplementary Fig. S17B).

Molecular aberrations in clinical outliers

Recently, patients with unusually favorable or unfavorable responses to conventional and targeted therapies have provided unique insights into the molecular determinants of clinical outcomes (26, 27). Having carefully characterized the patient cohort, we then performed a more detailed molecular analysis on 10 patients who had either exceptionally favorable (Supplementary Fig. S18) or unexpectedly poor clinical outcomes (see Materials and Methods).

We performed whole-exome sequencing to investigate further genomic aberrations in the 10 outlier cases. One thousand one hundred ninety-four high confidence variants were identified overall (Supplementary Table S7). There was no difference in the median mutational rates in favorable or unfavorable groups (2.0: range 1.0–3.0 and 2.1: range 1.5–3.9 mutations per Mb, respectively; Fig. 4A). Seventy-nine likely pathogenic variants were identified in 65 genes among the 10 outlier patients. Reflective of the quality of the exome data, all but two of the variants detected by targeted sequencing were also detected by whole-exome sequencing (Fig. 4B; Supplementary Table S8), with a common hotspot mutation in PIK3CA, p.E542K, only found by targeted sequencing due to poor sequence coverage at this locus in the whole-exome data.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Mutational signatures of clinical outliers. A, Somatic mutation burden identified in tumors of clinical outliers by whole-exome sequence analysis. Survival category is shown under the sample names. B, TFI and OS for individual patients and predicted pathogenic mutations. See text for criteria defining given patients as having exceptional survival characteristics. C, Heat map showing differences of the contribution in mutational signatures. Survival category is shown to the left of sample names.

ARID1A, BRAF, and PIK3CA mutations occurred in both the exceptional and poor outcome groups, but no novel pathogenic mutations were common within either group. We noted a germline frameshift deletion mutation in POLE in one exceptional outlier. POLE somatic mutations have been associated with unusually favorable outcomes in other solid cancers, including glioblastoma and endometrial cancers (28, 29), and are characterized by very high mutational load with a distinct pattern of nucleotide change. However, we did not observe a pattern of genomic scarring in this patient consistent with functional inactivation of POLE. In addition to POLE, an in-frame deletion detected in ERBB3 by custom capture sequencing was also confirmed as a germline variant in one short survival patient.

We extended the analysis of DNA base changes to investigate signatures that are reflective of other mutational processes in cancer (30). Signatures 1 (age), 2/13 (APOBEC), 3 (homologous recombination), and 6/15/20/26 (mismatch repair defect) have been reported as common mutational signatures observed in OCCC in recent studies (31, 32). Similar to previous results, broadly similar patterns of mutational signatures associated with age, mismatch repair, and homologous recombination were observed in both the favorable and unfavorable outcome groups (Supplementary Fig. S19). We next compared differences of contribution scores across all samples. Interestingly, signatures 2 and 13 related with APOBEC were closely located by clustering analysis and tended to show high contribution with cases with unfavorable outcome compare to cases with favorable prognosis (Fig. 4C).

We investigated copy number alterations based on the ratio of read coverage between tumor and normal samples (Supplementary Materials and Methods). A similar pattern of copy number change was observed in the clinical outliers to that reported in previous studies (Supplementary Fig. S20A; refs. 23, 25, 33). Although the sample sizes were too small to be statistically meaningful, we determined whether specific regions of copy number change were associated with clinical groups. Wide copy number gain in whole chromosomal arm 8p was observed in two patients with the longest PFS (Supplementary Fig. S20B). Gain of 8p is rare in OCCC, however, amplification at 8p11-12 has been observed in 10% to 15% of human breast cancers (34). In contrast, long deletions in chromosome 9q, rarely observed in acute myeloid leukemia, were seen in three of seven resistant cases for first-line chemotherapy, but not detected in three sensitive cases (35).