Background
Ovarian cancer, the most lethal gynecologic cancer, is expected to account for 14,700 deaths in the USA in 2018 [
1]. High grade serous carcinoma (HGSC), the most frequent histological type [
2], is molecularly characterized by a few recurrent mutations, including in
TP53 (almost all tumors) and
BRCA1 or
BRC2 genes. HGSCs present high genomic instability with copy number alterations (CNA) affecting a large fraction of the genome [
3]. Approximately 50% of these tumors are characterized by homologous recombination (HR) deficiency, which has been associated with
BRCA1 or
BRCA2 germline or somatic mutations (20 and 5% of cases, respectively),
BRCA1 promoter methylation (10% of cases), additional mutations in HR repair pathway genes and CNA in their regulators (
PTEN and
EMSY) [
3].
Patients carrying tumors with
BRCA mutation tend to have elevated sensitivity to platinum-based chemotherapy [
4] and PARP inhibitors [
5‐
8]. These cases have shown a better medium term prognosis [
9] even if the cure rate and long term prognosis is unaltered [
10]. Secondary mutations have been associated with resistance to PARP inhibitors [
11,
12].
Two strategies have been used to identify tumors with HR deficiency or alterations in genes involved in the DNA repair system other than
BRCA1 and
BRCA2 mutations. The first approach is to identify mutations in genes related to HR pathway [
13], based on next generation sequencing (NGS), which has become feasible with the fast development of sequencing technologies. The second is the identification of “genomic scars”, which are supposed to be a functional consequence of HR deficiency independently of its cause [
14‐
17]. Clinical trials revealed that even patients with no HR deficiency, evaluated by two different HR deficiency scores, can achieve response to PARP inhibitors [
8,
13]. These outcomes could be explained by other mechanisms of action than synthetic lethality of the HR pathway deficiency or failure in identifying the HR defect, or both.
In addition to the genomic scars of HR deficiency, gains or losses involving specific genes have also been associated with response to therapy in ovarian cancer [
18]. Cyclin E1 and RB1 are cell-cycle proteins associated with the G1-S phase cell-cycle transition.
CCNE1 copy number gain is described in about 20% of HGSC and is seemingly rare in
BRCA mutated tumors [
3]. Tumors with
CCNE1 copy number gains are more resistant to platinum therapy [
19] while
RB1 loss are associated with high sensitivity to platinum therapy [
20,
21].
Platinum resistant disease has a low chance to be responsive to platinum-based chemotherapy. The standard treatment is monotherapy using different drugs than platinum salts [
2]. In daily clinical practice and despite of the resistant profile, a set of patients is retreated with platinum therapy and some of them are responsive to platinum retreatment [
22].
In this study, we sought to evaluate the association of HR pathway mutations, HR deficiency scores and CCNE1 and RB1 CNA with response to platinum retreatment in ovarian cancer patients in the platinum-resistant setting.
Methods
Patients
From 2005 to 2014, 405 patients with ovarian carcinoma were treated at AC Camargo Cancer Center, São Paulo, Brazil. Thirty-five of them presented platinum resistant recurrence and were retreated with platinum therapy. Patients with unavailable data regarding the platinum retreatment were excluded (4 patients) and a retrospective review of the medical records was performed (Additional file
1). Based on the quantity and quality of tumor DNA, 15 of 31 cases were selected for SNP array (OncoScan® FFPE, Thermo Fisher Scientific, Waltham, MA, USA) analyses, and 11 of them were also evaluated by targeted-next generation sequencing. Nine patients had the primary tumor naive of treatment, five patients had the tumor sample collected at platinum resistant recurrence, and one had the tumor sample collected at platinum sensitive recurrence. The study was conducted in accordance with the Declaration of Helsinki ethical guidelines and approved by the institutional Ethics Committee (CEP# 1933/14).
Clinical data
Clinical features were retrieved from the medical records including age at diagnosis of platinum resistant recurrence, tumor histological subtype, family or personal history of ovarian and breast cancer, number of previous treatment lines, platinum free interval (PFI), and type of chemotherapy associated to platinum (Table
1).
Table 1
Clinical features of 31 patients with ovarian cancer who had previous platinum resistant relapse and were retreated with platinum
Age (years old) |
< 65 | 21 (67.7) |
> 65 | 10 (32.3) |
Histology |
High grade serous carcinoma | 21 (67.7) |
Endometrioid | 1 (3.2) |
Clear cell carcinoma | 1 (3.2) |
Undifferentiated carcinoma | 3 (9.7) |
Carcinosarcoma | 1 (3.2) |
Mixed | 1 (3.2) |
Family history of ovarian or breast cancer |
No | 19 (61.3) |
Yes | 9 (29.0) |
Number of previous treatment lines |
2 | 10 (32.3) |
3 | 7 (22.6) |
4 | 4 (12.9) |
5 | 6 (19.4) |
6 | 3 (9.7) |
8 | 1 (3.2) |
Platinum free interval (months) |
< 12 | 21 (67.7) |
> 12 | 10 (32.3) |
Primary platinum resistance* |
No | 12 (38.7) |
Yes | 19 (61.3) |
Chemotherapy with platinum rechallenge |
Platinum + taxane | 15 (48.4) |
Platinum + gemcitabina | 13 (41.9) |
Platinum + doxorubicin | 1 (3.2) |
Platinum + ifosfamide | 1 (3.2) |
Monotherapy | 1 (3.2) |
Recurrence was defined according to the GCIG (Gynecological Cancer Intergroup) criteria after the analysis of RECIST (Response Evaluation Criteria in Solid Tumors) [
23,
24]. The CA125 levels were extracted from the medical records. The date of the earliest event was considered for progression. The recurrence detected within 6 months after the last platinum infusion was defined as platinum resistant recurrence. All recurrences that followed this first platinum resistant recurrence were also considered platinum resistant. Progression-free survival (PFS) was defined as the interval between the date of the beginning of the platinum retreatment and disease progression or death by any cause. Overall survival (OS) was defined as the interval between the dates of the beginning of the platinum retreatment and death by any cause. The interval between the date of the last platinum compound infusion and the disease progression that preceded platinum retreatment was used to define the platinum-free interval (PFI).
The CA125 expression levels and the response to platinum retreatment data were retrieved from the medical records. The image reports were also collected. GCIG criteria were used to evaluate RECIST and CA125 data [
23,
24]. In accordance, each case was categorized as having “response” (complete or partial response) or “no response” (stable disease or disease progression).
Ten μm paraffin embedded tissue sections were deparaffinized with xylene, washed with descending concentrations of ethanol and water ultra-pure sterile. DNA was extracted using QIAamp DNA FFPE Tissue Kit (Qiagen, Valencia, CA) according to the manufacturer’s instructions. DNA integrity was evaluated using Agilent Genomic DNA ScreenTape (Agilent Techologies, Santa Clara, USA) and quantified using a Qubit Fluorometer (Life Technologies, Carlsbad, CA).
OncoScan assay
The OncoScan® FFPE platform (Thermo Fisher Scientific) allows the identification of over 200,000 SNPs and the detection of 74 somatic mutations in nine genes (BRAF, KRAS, EGFR, IDH1, IDH2, PTEN, PIK3CA, NRAS and TP53). However, we focused only in CNAs. The SNP array assay was performed according to the recommended protocol using 80 ng of genomic DNA. After the hybridization (18 h), the arrays were stained and washed (GeneChip® Fluidics Station 450) and loaded into the GeneChip® Scanner 3000 7G (Thermo Fisher Scientific/Affymetrix). The CEL files were generated by Affymetrix Gene Chip Comand Console® software (v. 4.0) and processed by OncoScan Console software (v. 1.3) resulting in OSCHP files and QC metrics.
Data generated from the SNP array was used to calculate previously defined scores of homologous recombination deficiency: loss of heterozygosity (LOH) [
16], telomeric allelic imbalance (tAI) [
14], large scale transition (LST) [
15], and Composite Score (CS) (the sum of LOH, tAI and LST scores) [
25]. LOH was calculated as the number of LOH regions spanning at least 15 MB but not involving the entire chromosome. The number of regions with allelic imbalance extending to one of the telomeres but not crossing the centromere, after filtering regions shorter than 11 MB or spanning less than 500 probes, was defined as tAI. The LST score was defined as the number of breakpoints between regions spanning at least 10 MB within a distance of maximum 3 MB. HRD was computed as LOH + tAI + LSTm (adjusted LST score). According to Timms et al. [
26], the HRD score introduced by Telli et al. [
25] increases with ploidy in both intact and deficient samples. The adjusted LST score is calculated as [LSTm = LST-15.5P], in which P is the tumor ploidy. Using logistic regression analysis, the constant 15.5 was derived to provide the best separation between intact and deficient samples. ASCAT [
27] was used for inferring tumor ploidy, calculating allele-specific copy numbers and segmentation. The cut-off values were previously established as HRD markers: > 10 for LOH [
16], > 42 for CS [
25] and > 15 for near-diploid tumors or > 20 for near-tetraploid tumors for LST scores [
15]. The tAI median value was used as cut-off to consider it as HRD marker.
CCNE1 and RB1 CNAs were evaluated using the SNP array with a resolution of 1 probe per 50Kb for both genes. The average copy number for all probes covering each gene > 2.0 and < 2.0 was defined as gain or loss, respectively.
Target enrichment next generation sequencing (tNGS)
The HaloPlexHS target enrichment technology (Illumina 100 custom design with a 229,506 bp region of interest) (Agilent Technologies, 2016) was used to investigate mutations in 24 genes (BRCA1, BRCA2, TP53, BRIP1, CDH1, PALB2, RAD51C, RAD51D, XRCC2, MLH1, MSH2, MSH6, PMS2, EPCAM, APC, MUTYH, BMPR1A, SMAD4, STK11, PTEN, POLD1, POLE, NTHL1 and VHL). The fraction of bases in the region of interest covered 98.76% of the target region. HaloPlexHS libraries were constructed according to manufacturer’s protocol v.B0 (Agilent Technologies, June 2015). Indexes were incorporated for each sample during enrichment, allowing samples to be multiplexed before sequencing. A total of 11 HaloPlexHS libraries were validated on a TapeStation using High Sensitivity screentape (Agilent Technologies). After enrichment, HaloPlexHS libraries were diluted to 10 nM, pooled, denatured and subjected to paired-end (2× 150 bp), single index (8 bp) reversible terminator-based DNA sequencing on a NextSeq550 (Illumina) using a mid-output kit and loading 1.8 pM of denatured library pool.
For each sequenced sample, the raw FastQ files were trimmed with TrimGalore (v. 0.4.2), subsequently mapped to the GRCh37/hg19 human reference genome using MOSAIK (v. 2.2.26) and converted to BAM using Sambamba (v. 0.6.3). The BAM file for each sample was preprocessed with Genome Analysis Toolkit (GATK v. 3.6; local realignment around indels and base quality score recalibration), prior to variant calling. General alignment statistics (e.g. number of aligned reads etc.) was generated with BAMtools (v.2.3.0). Target specific alignment statistics (i.e. per base−/region−/gene−/sample-coverage and coverage percentage of ROIs), were obtained using GATK DepthOfCoverage. Variant calling was performed using the variantcallers: GATK HaplotypeCaller in genomic VCF mode, GATK UnifiedGenotyper (GATK v. 3.6), FreeBayes (v. 1.1.0) and finally PLATYPUS (v. 0.8.1). A single multisample VCF file comprising all analyzed samples was generated for each variantcaller. Four multisample VCF files were subsequently merged to produce a single multisample VCF file.
The variants were annotated with Ingenuity Variant Analysis (Qiagen), SnpEff (v 4.1c), ANNOVAR (v July 2017) and VariantTools (v.2.3), using build-in and custom annotation tracks. Additional file
2 summarizes the pipeline used to evaluate the generated variants.
Statistical analysis
Statistical analyzes were performed using the SPSS (v. 21.0; SPSS, Chicago, IL, USA) software, adopting a two-tailed P < 0.05 value as significant. The association among response rate to platinum retreatment in patients with platinum resistant ovarian cancer and the HRD scores, all mutations, CCNE1 CN gains and RB1 CN losses, were investigated using Mann-Whitney and Fischer’s Exact tests. Correlation analysis between the HRD scores was tested using Pearson’s coefficient and linear regression. Overall survival and progression free survival analyses were performed using Kaplan-Meier and log-rank test.
Discussion
In this study, we evaluated markers of response to platinum retreatment in a selected group of patients with heavily pretreated platinum resistant ovarian cancer. The rationale was to identify patients that are still sensitive even after a platinum free interval shorter than 6 months and numerous previous treatment lines.
Nine of 15 cases (60%) of our cohort presented HR deficiency, which is in accordance with the 50% described in the TCGA dataset [
3]. Patients with CS score higher than 42 had an ORR of 55.6% versus 33.3% observed in those with lower scores. In addition, the HRD median values of each score were higher among responders than in non-responders. However, the differences were not statistically significant probably due to the small number of cases evaluated.
Two scores based on the patterns of CNA and LOH were used in the phase III clinical trials of PARP inhibitors showing their ability to identify patients with
BRCA mutation [
7,
8,
13]. These studies also described a second group of patients negative for
BRCA mutations but with high sensitivity to PARP inhibitors. Our findings give additional evidence that the scores are able to identify patients with high sensitivity to platinum agents. The absence of statistical significance in our study may be due to the small number of cases or the accuracy of these scores. Interestingly, the
BRCA non-mutated cases and those with low HRD (based on the scores) benefited from treatment with PARP inhibitors [
7,
8,
13]. The definition of all scores, except tAI [
14], were based on the
BRCA mutation as a gold standard to define HRD [
15‐
17]. Therefore, the CNA and LOH pattern calculated by the scores are similar to the ones promoted by
BRCA mutations, which may be true for most but not necessarily all causes of the HRD.
Cyclin E1 overexpression promotes cell cycle progression abrogating DNA repair during the G1 phase. In addition,
CCNE1 amplification has been associated with an apparent synthetic lethality in cases with HR deficiency [
28]. Nine of our 15 tumors (60%) showed
CCNE1 copy number gains while data from TCGA (509 high grade serous ovarian carcinomas) presented
CCNE1 gains in 32.4% or focal amplification in 20.8% [
3]. We observed that cases showing
CCNE1 gains had lower ORR, and shorter PFS and OS compared with those not presenting gains. Previous studies in ovarian cancer patients described increased expression levels of cyclin E1 and an association with worse survival [
29]. At the genomic level, two studies described high frequency of
CCNE1 gains in patients with primary platinum resistant disease [
19,
30]. The worse survival observed in the TCGA cohort for patients with
CCNE1 amplification was attributed to its negative association with
BRCA mutations [
3]. In our study, two of four patients harboring
BRCA mutations also presented
CCNE1 gains. This finding suggests that the correlation between
CCNE1 gain and outcome is not exclusively due to its negative association with
BRCA mutations. Ten of our 15 patients had primary platinum resistant disease and 7 of 9 patients with
CCNE1 gains presented primary resistant disease. This finding supports the association of
CCNE1 aberrations and resistance to platinum therapy and may explain the higher than expected frequency of
CCNE1 gains in our study. Previous studies showed
CCNE1 gains and
BRCA mutation or homologous recombination deficiency as mutually exclusive [
29]. However, the authors showed that complete mutually exclusive alterations were not observed between low levels of
CCNE1 gains and
BRCA mutations. Our two patients with co-occurrence of
BRCA mutation and
CCNE1 gain presented low
CCNE1 copy number gains (2.2 and 2.3).
The RB1 protein is involved in the S-phase checkpoint to repair DNA breaks and in the regulation of DNA replication. The loss of the tumor suppressor
RB1 leads to cell cycle progression and replication fork progression leading to replication stress and DNA damage, which could be repaired by HR machinery [
31]. We found an association of
RB1 loss with PFS and OS. To our knowledge, two previous studies addressed the impact of
RB1 loss in ovarian carcinoma. In 2013, Milea et al. showed loss of RB1 protein expression associated with longer OS [
20]. Recently, Garsed et al. reported the association of
RB1 loss and HR gene mutations with extremely long PFS and OS [
21]. Taken together, these findings suggest that
RB1 loss is a biomarker with the potential to identify sensitive patients to platinum treatment. In addition, this alteration could be used in the clinical practice and potentially select
BRCA mutated patients with higher chance to be PARP inhibitors responsive.
Four of 15 tumors (26%) presented
BRCA mutations, an expected frequency in high grade serous carcinomas [
32]. No differences in the response of the treatment were found between mutated and non-mutated tumors. Although,
BRCA mutations are well known markers of response to platinum and PARP inhibitors therapy, other studies also failed to show higher frequency of
BRCA mutations in long term responders [
21]. In addition to
BRCA, two non-responder patients presented tumors with low HR deficiency scores and
XRCC2 mutations.
XRCC2 is involved in the repair of DNA double-strand breaks by HR pathway. This finding highlights the value of an investigation using a panel of genes in ovarian cancer. Furthermore, conclusions regarding HR deficiency based solely on the presence of certain mutations may not be precise, as was found in the ARIEL2 trial [
13].
Our study has several limitations, mostly due to the small sample size and its retrospective nature. For example, we found the ORR higher than expected for platinum resistant patients. High ORR may be due to patient over-selection including those who have received several previous treatment lines. Despite the limitation of the ORR evaluation, CCNE1 gain and RB1 loss were both associated to OS, which is an objective endpoint even in retrospective studies. No association between HR deficiency scores and response to therapy was found. The small number of patients limits conclusions regarding the low accuracy of these scores, even if previous literature data also pointed out to the limitation of the scores. However, the strength of our study was the evaluation of well selected individuals for whom it would be expected a low response rate with platinum retreatment. Unexpectedly, a high response rate to platinum retreatment was found suggesting that this selected cohort might be enriched for extremely sensitive tumors. HR deficiency scores were not able to show a strong association with therapy response. Interestingly, CCNE1 copy number gain was a negative prediction marker of platinum sensitivity, and RB1 copy number loss identified patients with sensitive disease.
In this study we explored the mutational profile and HR deficiency score in ovarian cancer patients to better understand the platinum resistant recurrence as defined by the platinum free interval. We demonstrated that HR deficiency scores, CCNE1 gains and RB1 losses could be used to distinguish patients who are still sensitive to platinum retreatment from those resistant to platinum therapy. Considering similar mechanisms of sensitivity to platinum salts and PARP inhibitors, these markers could be useful to better select the patients for PARP inhibitors therapy in the platinum resistant relapse.