Molecular Response to Neoadjuvant Chemotherapy in High-Grade Serous Ovarian Carcinoma

Abstract

While high-grade serous ovarian carcinoma (HGSOC) is the most common histological subtype of ovarian cancer, significant tumor heterogeneity exists. In addition, chemotherapy induces changes in gene expression and alters the mutational profile. To evaluate the notion that patients with HGSOC could be better classified for optimal treatment based on gene expression, we compared genetic variants (by DNA next-generation sequencing [NGS] using a 50 gene Ion Torrent panel) and gene expression (using the NanoString® PanCancer 770 gene Panel) in the tumor from 20 patients with HGSOC before and after neoadjuvant chemotherapy (NACT). NGS was performed on plasma cell free DNA (cfDNA) on a select group of patients (n=14) in order to assess the utility of using cfDNA to monitor these changes. A total of 86 genes had significant changes in RNA expression after NACT. Thirty-eight genetic variants (including SNPs) from 6 genes were identified in tumors pre-NACT, while 59 variants from 19 genes were detected in the cfDNA. The number of DNA variants were similar after NACT. Of the 59 variants in the plasma pre-NACT, only 6 persisted; whereas 33 of 38 specific variants in the tumor DNA remained unchanged. Pathway analysis showed the most significant alterations in the cell cycle and DNA damage pathways.

Introduction

While the traditional treatment paradigm for high grade serous ovarian cancer (HGSOC) is surgical debulking followed by adjuvant chemotherapy, there has been a recent shift towards increased use of neoadjuvant chemotherapy (NACT), which aims to reduce disease burden prior to surgical cytoreduction (1, 2). NACT has been associated with increased rates of optimal debulking and debulking to no residual disease, as well as decreased surgical morbidity and mortality (3). It has been demonstrated that chemotherapy affects gene expression and causes molecular derangement over time (4, 5). The recent shift towards NACT in ovarian cancer provides an opportunity to evaluate molecular response to therapy within individual patients.

Although HGSOC is the most common histologic subtype of epithelial ovarian cancer, numerous studies have shown that there is significant tumor heterogeneity; however, practitioners continue to treat it as a single entity in the upfront setting (6). Only recently have PARP inhibitors been FDA approved in the recurrent setting based on germline BRCA status, somatic BRCA mutations, or platinum sensitivity (7, 8). Recent studies have attempted to use gene expression and mutational profiles to characterize distinct molecular subtypes of HGSOC for prognostication and identification of long-term responders to PARP inhibitors (9, 10). Unfortunately, these studies have not evaluated tumor heterogeneity and clonal evolution. The Cancer Genome Atlas (TCGA) analyzed 489 HGSOC tumors and found somatic mutations in nearly every chromosome; however, TP53 and BRCA1/2 were the only genes that was consistently mutated in a large percentage of patients (11). With such a high degree of heterogeneity and low rates of actionable mutations overall, it is highly unlikely that one treatment regimen will be effective for all patients with HGSOC. We need to improve upon our selection criteria of frontline therapy and find more successful treatments for chemoresistant cases. To accomplish this, stronger links between molecular profiles and drug sensitivity are needed. Additionally, because it is usually not feasible to perform serial tissue biopsies on patients, we need better strategies to evaluate genomic changes; thus, using peripheral blood to monitor clonal evolution could impact treatment decisions in the future.

Cell free DNA (cfDNA) is composed of short fragments of nucleic acids within the plasma that contains mutations generated via apoptosis of cancer cells, germline mutations, and somatic mosaic mutations in non-neoplastic DNA. Although Less that 1% of circulating cfDNA comes from cancer cells, patients with advanced disease exhibit increased amounts of cfDNA (12). Studies in breast and other cancers have suggested that cfDNA can be used for screening, therapeutic decision-making, prognostication, and to predict mechanisms of therapy resistance (13–18). Otsuka et al found the detection of TP53 mutations in the plasma of patients with ovarian masses could indicate the presence of malignancy, and there was a relationship between the reappearance of TP53 cfDNA mutations and recurrence (19).

The goal of this study was to compare the molecular landscape of HGSOC before and after NACT using the NanoString® PanCancer 770 gene Pathway Panel to evaluate changes in RNA expression and next generation sequencing (NGS) to evaluate changes in DNA. In addition, we assessed the utility of cfDNA to monitor genetic changes in patients pre- and post-NACT. We compared the DNA mutations identified by NGS in the tumor to those detected in the cfDNA in order to evaluate the potential ability of using cfDNA as a “liquid biopsy” in the management of HGSOC.

Materials and Methods

Study population and Sample Collection

Under an IRB approved protocol at the University of Alabama at Birmingham, patients presenting to the Division of Gynecologic Oncology with suspected ovarian cancer and planned treatment with NACT between September 2013 and January 2015 were prospectively enrolled after providing written informed consent for tumor and blood collection. Recruitment was stopped when pre- and post-NACT tissue was collected on 20 patients with HGSOC (grade 2 or 3 serous carcinoma of the peritoneum, ovary, or fallopian tube). At the time of diagnostic laparoscopy, blood and an additional tumor biopsy from the omentum was obtained freshly from surgery for research purposes. Post-NACT, at the time of interval debulking, a biopsy of residual tumor from the omentum (if present) and blood was collected for research. All tumor samples were removed adjacent to tumor that was sent for pathological confirmation of tumor histology and were immediately snap frozen in liquid nitrogen and stored at −80˚C. Blood was centrifuged within 30 minutes of collection. Immediately following centrifugation, the plasma layer was pipetted into tubes with close attention to avoid any contamination from the buffy coat or red blood cells. Baseline demographic information was obtained at enrollment. Patient studies were conducted in accordance with the Declaration of Helsinki.

Details of the surgery were collected from the operative note. Optimal cytoreduction was defined as <1cm of residual disease at the time of interval debulking. Patients were followed prospectively at the University of Alabama or through communication with their treating oncologist for evaluation of response to primary treatment, progression-free survival, overall survival, and mean follow up time. Accession numbers for the data are SRP131968 and GSE109934.

Tissue Selection and Pathology Assessment

Sections of tumor adjacent to the tissue removed for research were stained with hematoxylin and eosin to determine neoplastic content. Pre- and post-NACT biopsies were assessed for the percent of neoplastic nuclei, psammoma bodies, and lymphovascular invasion. Post-NACT samples taken during debulking were assessed for size of tumor (cm), inflammation, necrosis, desmoplasia, and cellular atypia. A board certified gynecological pathologist determined neoplastic cellularity which is expressed as the percentage of neoplastic nuclei to non-neoplastic nuclei for the tumor that was removed adjacent to the tissue used for genomic analysis (20). We observed that 90% of the samples had at least 50% tumor cells. PCA plot of samples with low percentage tumor cells showed minimal differences, so we included them in the analysis (Supplementary Figure 1). Histopathologic assessment of response to NACT was performed using a validated chemotherapy response score (21).

NanoString nCounter Analysis

Snap frozen tumor was lysed and manually homogenized and passaged through an 18 gauge syringe needle. RNA was harvested using a RNA Isolation Kit (Roche, Basel, Switzerland) as per the manufacturer’s instructions. All RNA was quantified used the DeNovix DS-11 Spectrophotometer. RNA samples had concentrations ≥12.5ng/μL and an A_260/280 ratio between 1.7 and 2.3. mRNA was analyzed by the UAB NanoString Library (https://www.uab.edu/medicine/radonc/en/nanostring). Samples were processed for analysis on the NanoString nCounter Flex system using the 770 gene PanCancer Pathways Plus panel (606 critical genes from 13 canonical cancer pathways, 124 cancer driver genes, and 40 reference genes) per the manufacturer’s instructions (NanoString Technologies, Seattle, WA). One of the 20 samples had low RNA integrity and did not meet the threshold for further analysis and the matched pair was also excluded from analysis.

RNA Expression Analysis

Resource complier (RCC) data files were imported into NanoString nSolver 3.0 and further analyzed using the PanCancer Pathways Advanced Analysis Module, which normalizes gene expression to a set of positive and negative controls genes built into the platform. Using the nCounter Analysis software, we identified a list of genes with significantly altered expression before and after NACT. The fold change and p-values were calculated using nCounter default settings. As recommended, genes whose expression levels were at or below the level of the negative controls were removed from analysis. With the remaining list of genes on the PanCancer panel, a filter cutoff of fold change ≥ ±1.5 or ≥ ± 2 and p-value < 0.05 were used to identify the significant gene expression changes based on the nCounter analysis. A pathway score was calculated using nSolver Advanced Analysis from the expression levels of the relevant genes in 13 canonical pathways using measurements of pathway activity values derived from singular value decompositions (22). This method uses metagenes to represent pathway activity and aims to capture not only over-represented significantly altered genes but also smaller but cumulatively impactful changes within a pathway.

Ingenuity® Pathway Analysis (IPA®) and DAVID analyses were used to explore the pathways most affected by NACT (23, 24). IPA uses right-tailed Fisher’s exact test to calculate a p-value determining the probability that each biological function and/or disease assigned to that data set is due to chance alone. All pathway analysis used the full 770 gene set as background to calculate enrichment.

Next Generation Sequencing and Preparation of tumor and cell free DNA from plasma

Tumor DNA was extracted using the AllPrep DNA/RNA kit (Qiagen Inc., Germantown, MD). Circulating cfDNA was isolated from plasma using CirculoGene’s cfDNA enrichment and recovery technology CirculoGene Technique (CGT) method. Quantification of cfDNA was performed using the Qubit 2.0 Fluorometer with dsDNA BR and HS assay kits (Life Technologies, Carlsbad, CA).

Ultra-deep targeted sequencing of cfDNA and tumor DNA was performed using the Ion Torrent NGS. The targeted sequencing libraries were generated using the Ion AmpliSeq Library kit 2.0 and Cancer Hotspot Panel v2 according to the manufacturer’s instructions (Life Technologies, Carlsbad, CA). The starting material consisted of 1–10ng of cfDNA or 1–10ng of tumor DNA and each sample was analyzed using a CLIA-certified, CAP proficiency-supported clinical test. A selected 50-gene panel of 3,000 mutations was used [ABL1, AKT1, ALK, APC, ATM, BRAF, CDH1, CDKN2A, CSF1, CTNNB, EGFR, ERBB2, ERBB4, EZH2, FBXW7, FGFR1, FGFR2, FGFR3, FLT3, GNA11, GNAQ, GNAS, HNF1A, HRAS, IDH1, IDH2, JAK2, JAK3, KDR, KIT, KRAS, MET, MLH1, MPL, NOTCH1, NPM1, NRAS, PDGFRA, PIK3CA, PTEN, PTPN11, RB1, RET, SMAD4, SMARCB1, SMO, SRC, STK11, TP53, VHL]. After amplification, primers were partially digested by the Fupa enzyme (Lifetech), then ligated to Lifetech ION XPRESS barcodes and purified using Ampure Beads.

The quality of the libraries was assessed using quantitative PCR. The Ion Chef system was used for emulsion PCR to clonally amplify sequencing templates. Deep sequencing was performed on Ion Torrent Proton with coverage ranges of 3000–8000X. Two tiers of software validation (built-in VariantCaller 4.2 and built-out Station-X) were used for mutation validation. Sequencing data was analyzed by the VariantCaller 4.2 software using the somatic high stringency parameters and the targeted and hotspot pipelines. All the variants identified were further confirmed by analyzing the data through GenePool (Station-X, San Francisco, CA). Filter criteria was based on somatic and germline databases: COSMIC (25), dbGAP (26, 27), and 1000 genome (28), EXaC (29), and GNOMaD (http://gnomad.broadinstitute.org/). Libraries were also generated from blank and control samples to ensure proper construction. Two cell line controls (sw480 and na19240) and a “process control” with a true negative (from normal individuals who tested negative previously) were used to rule out false positives. The quality score threshold for reporting a mutation was 10, although the majority fall between 500–1000. Any mutation allele frequency under 1% (below NGS platform detection limit) was filtered out.

TCGA Data Analysis

We downloaded all serious ovarian cancer RNA-sequencing and clinical data available as of November 20, 2017 from https://firebrowse.org. There was a total of 295 samples with RNA-seq data. Of those 151 also had platinum resistance data and 292 had overall censored survival data. We used DESeq2 on raw counts to calculate differentially expressed genes based on platinum resistance. We used default settings and did not include covariates in the model. Seventy-seven genes reached an FDR less than 0.05. We also identified genes associated with survival using a similar approach. We defined poor prognosis as patients who succumbed to disease within two years and those with “better” outcomes as surviving at least 5 years beyond diagnosis. All patients were stage III-IV, so no adjustments were made for stage. Again, we used DESeq2 to identify 207 differentially expressed genes using an FDR cutoff of 0.05.

Statistical Analysis

NanoString data and pathway analysis is described above. Differential expression based on mutational status was done using the R statistical package version 3.2.1. The DESeq2 package was used to determine differential expression with a wald test and default parameters (30). All p-values were adjusted for multiple hypothesis testing and significant genes were considered to have an adjusted p-value of <0.05. The concordance between NGS assays was evaluated by calculating the gene-level sensitivity and precision of cfDNA NGS assay in detecting mutations present in the tumor tissue as described by Pishvaian et al (31). A gene variant was considered concordant if the exact nucleotide change was present in the cfDNA and tumor from the same time point (either pre- or post-NACT).

Results

Clinical Characteristics of Study Cohort

Of 37 consented patients, 20 patients met all inclusion criteria: high-grade papillary serous ovarian cancer who underwent interval debulking surgery after 2–6 cycles of neoadjuvant chemotherapy without a complete pathological response. Nineteen of the 20 patients had quality RNA extracted from tumor for analysis. Fourteen of these 19 patients had paired plasma samples. Additionally, plasma from 4 of the 14 patients was collected at the time of recurrence (Supplementary Figure 2).

Notable features of the cohort relevant for this study are that patients were all late stage and ranged in age from 53–85 (Table 1). Each received 2–6 cycles of chemotherapy before surgery and 8 of these were classified as platinum sensitive. Platinum sensitivity was defined by a complete response during adjuvant chemotherapy and clinical remission for at least 6 months after the completion of chemotherapy. Platinum resistance was defined as progressive or persistent disease or progression within 6 months of completing platinum therapy.

Table 1.

Patient Demographics

Characteristics	NanoString (n=19)	NGS cfDNA and Tumor (n=14)
Age (range)	Mean 72, Median 74 (53–85)	Mean 73, Median 74 (61–85)
BMI (range)	Mean 26.1, Median 25.8 (18.6–31.7)	Mean 26.9, Median 26.2 (22.1–31.2)
Stage
IIIC	13	10
IV	6	4
CA-125
Pre NACT	Mean 1744.5, Median 540.0 (33.3–13818.0)	Mean 1038.9, Median 463.3, (79.3–5503.0)
Post NACT	Mean 226.5, Median 37.9 (4.7–2802.0)	Mean 277.6, Median 33.5, (4.7–2802.0)
Debulking Status
NRD	7	5
OPT	10	7
SO	2	2
Number of NACT Cycles	Mean 4, Median 3 (2–6)	Mean 4, Median 4.0 (2,6)
Platinum Status
Resistant	9	5
Sensitive	8	7
N/A*	2	2
Status after Adjuvant Therapy
Persistence	5	3
Complete Response	12	9
N/A*	2	2
Overall Survival (mos)	Mean 23.30, Median 27.30 (4.33–44.03)	Mean 25.30, Median 27.93 (4.33–40.10)
Progression Free Survival (mos)	Mean 16.55, Median 17.13 (8.03–26.10)	Mean 17.54, Median 17.27 (10.27–26.10)
Mean Follow-up (mos)	Mean 26.85, Median 28.93 (5.80–32.33)	Mean 31.06, Median 30.07 (27.30–32.33)

^*Died prior to completion of therapy

Abbreviations: NRD, No residual disease; OPT, optimal; SO, suboptimal

Chemotherapy Response Score

A three-tiered histopathological scoring system (chemotherapy response score) that has been validated in stage IIIC and IV HGSOC patients at the time of interval debulking surgery was used to measure the response to therapy in post-treatment tissue (Supplementary Table 1) (21). Two cases had a grade 1 response (no or minimal tumor) post-NACT. Sixteen cases had a grade 2 response (regression-associated fibro-inflammatory changes). Grade 3 represents complete response (excluded) with no residual tumor or tumor less than 2.0 mm in size (2 cases). In these two cases, a microscopic tumor was identified. All cases with grade 2 and 3 showed desmoplastic changes. In both cases with minimal tumor response (grade 1), desmoplastic changes were not identified. Histologic comparison of pre- and post- NACT samples are shown in Figure 1A. Post-NACT samples had more inflammatory cells and extensive fibrosis. In 6 cases, significant post-treatment cytological atypia was present. Psammoma bodies were more commonly present in post-NACT tissue. Fourteen of the post-treatment cases showed inflammation. No necrosis was seen in any of the post-treatment sections (Supplementary Table 1).

Figure 1.

Genes pre- and post-NACT. A, Histologic comparison of pre and post NACT samples. 1) Pre-NACT sample biopsy. 2) Post-NACT samples had significant reduction of adenocarcinoma in omentum. B, Heirarchical cluster analysis of all 770 genes. C, Heirarchical clustered analysis of 86 significantly changed genes. Cluster 1 (green) was enriched for pre-NACT; cluster 2 (red) was enriched with post-NACT. D, Volcano plot showing genes expression changes post-NACT by Log2 fold change (x-axis) and minuslog₁₀ p-value (y-axis). The top 10 genes with increased and decreased expression, fold change (post vs. pre) and p-value are shown on the right.

RNA expression and pathway analysis of pre- and post-NACT tumor samples

Gene expression was measured from a panel of 770 cancer related genes (NanoString Pan Cancer panel). Eighty-six genes showed significantly altered expression (p<0.05, fold change >2) following NACT. Initially, we used an unsupervised clustering analysis of all 770 genes to group all pre- and post-NACT samples by gene expression profiles, which did not show evidence of the pre-NACT clustering together separate from the post-NACT (Figure 1B). Hierarchical clustering of only the 86 genes that had a significant change with NACT, two groups were established (one group enriched with the pre-NACT samples and the other group enriched for the and post-NACT samples) (Figure 1C). Figure 1D highlights the most significantly altered genes in the panel. The 10 genes with the greatest increase in expression are: NR4A3, NR4A1, THBS4, SFRP2, RASGRF2, OSM, FOS, NTRK2, SFRP4, and GATA2. The top 10 genes that were the most significantly reduced following chemotherapy were E2F1, HIST1H3G, BRCA2, TTK, CCNA2, CCNB1, HIST1H3B, BRIP1, CDC7, CHEK1. NR4A3 and NR4A1, which both increased after NACT, are Nuclear Receptor Subfamily (NR4A) members and play a critical role in controlling proliferation and apoptosis of cancer cells. Additionally, the Secreted Frizzled Related Protein (SFRP2) which modulates WNT signaling and is involved in chemoresistance and regulation of cancer stem cells had greater than a 5-fold increase in expression. E2F1, which was significantly decreased, is a transcription factor associated with DNA repair; it regulates BRCA2, BRIP1, and CHEK1 – all genes in the homologous recombination deficiency (HRD) pathway. Comparing our list of significant genes to published RNA-sequencing data generated by the Cancer Genome Atlas (all generated from pre-treatment patients) showed that some genes associated with survival and platinum response were common. We found six genes that were associated with response and platinum sensitivity (CCND2, FGF11, COL2A1, COL4A5, PRKAR1B and PRKCG) and two genes associated with treatment and survival (MAPT and SHC2). While our 86 genes are not enriched (p=0.31, Chi-square test) for genes associated with patient prognosis, the overlapping genes may represent a particularly impactful subset. A list of all 770 genes with fold changes and p-values are provided in Supplementary Table 2.

To characterize the effect of altered gene expression on critical cancer pathways, we used nSolver Advanced Analysis Module’s pathway analysis tool, which condenses each sample’s gene expression profile to calculate a pathway score using the first principal component analysis. We calculated pathway scores pre- and post-NACT for 13 canonical pathways. This showed that DNA damage response is upregulated and the remaining pathways were downregulated. (Figure 2A). The most significantly affected pathways were MAPK signaling, Cell-cycle apoptosis, DNA Damage – Repair, Transcriptional regulation and PI3K signaling (Figure 2B). Supplementary Figure 3A–F shows the heatmaps of the top 5 pathways effected by NACT. The individual genes in the DNA Damage Repair and the Cell Cycle Pathways are highlighted in Supplementary Table 2. It is noteworthy that the one patient with a known germline BRCA2 mutation (patient 32) had the lowest DNA damage pathway score post-NACT (Figure 2C). Of particular interest, we noted that genes involved in Hereditary Ovarian Cancer Signaling showed decreased expression (genes highlighted in green) in post-versus pre-NACT (Figure 2D).

Figure 2.

Pathway changes pre-and post-NACT. Pathway scores condense each sample’s gene expression profile using the first principal component analysis. A, Individual pathway scores pre- and post-NACT. B, Box plots for individual pathway scores. Cell cycle, DNA damage, and transcriptional misregulation pathways were statistically different in pre- vs post-NACT (p <0.05). P-values are calculated using paired T-test. C, Individual patient DNA damage pathway scores pre- and post-NACT. D, Individual pathway and gene changes involved in Hereditary Ovarian Cancer Signaling: BRCA1, BRCA2, p53, PTEN, DNA repair, DNA damage response and checkpoint control, protein ubiquitination, transcriptional control and chromatin remodeling.

To supplement the nSolver Advanced Analysis, pathway analysis was also conducted using Ingenuity Pathway Analysis (IPA) and DAVID. The top 3 canonical pathways identified by IPA that were affected by NACT were Cell Cycle Control, GADD45 Signaling and ATM Signaling (Supplementary Table 3, Supplementary Figure 4A–C). The networks with the most significant changes were the Cell Cycle/DNA Replication/Recombination/Repair and Cellular Growth/Proliferation Networks (Supplementary Table 3, Supplementary Figure 4D–E). When the IPA results were compared to pathway changes using the DAVID analysis tool, the results were similar.

Gene expression associated with Platinum Sensitivity

In order to investigate how gene expression and signaling pathway changes were affected by platinum sensitivity, we compared our platinum sensitive patients (n=8) to those that were platinum resistant (n=9). Analysis of pre-NACT tumor samples identified only one gene (IL12A) that was significantly altered in platinum resistant patients compared to platinum sensitive patients (p=0.0099) (Figure 3A). Using a less stringent cut-off of an unadjusted p-value of less than 0.05 (40 genes) we used DAVID pathway which revealed enrichment of a variety of cancer-related pathways including MAPK signaling (10 genes), cell cycle (5 genes), Wnt signaling (4 genes), and epithelial-mesenchymal transition (EMT, 3 genes). Genes that were differentially altered by NACT in platinum resistant versus sensitive patients were also identified. The small sample size limits the statistical power, but the change in expression level from pre- to post-NACT samples for 5 genes (IL6, ITGB3, MAML2, NTF3, PL2G2A,) had unadjusted p-values of <0.01 between the platinum resistant and platinum sensitive patients indicating a potential interaction between expression change with treatment and platinum resistance. Among these, MAML2 and ITGB3 show increased expression following treatment in the sensitive patients with no change or reduced expression in the resistant patients (Figure 3B). The number of genes associated with platinum response is limited in this analysis, but there is no significant overlap between genes predicting platinum response and those changing following response. This could be explained by the fact that resistance may result from a variety of mechanisms including pathways that are not represented on this platform.

Figure 3.

Platinum sensitivity gene expression analysis. A, Box plot of NTF3 expression in platinum vs resistant patients pre-NACT. B, Box plots of individual gene expression in chemosensitive (orange) vs chemoresistant (red) pre- and post-NACT samples.

Targeted Next Generation Sequencing of Tumor Samples pre-NACT and post-NACT

A pre-identified 50 cancer gene targeted NGS panel was performed on 14 of the 19 patients both pre-NACT and post-NACT. The frequency of the most common mutations found in this study cohort (pre-NACT) was similar to those described in the Australian Ovarian Cancer Study (AOCS) (Figure 4) (32). Figure 5A shows the specific tumor variants found pre- and post- NACT. We identified 38 variants across 6 genes (TP53, KDR, KIT, PIK3CA, KRAS, and PTEN). Nineteen of these variants were found in only one patient in our cohort, and four represent common polymorphisms found at a frequency of greater than 5% minor allele frequency in the general population. Filtering common alleles from our analysis, we identified 15 variants likely to have arisen in the tumors, all of which are predicted to be damaging mutations by SIFT (33). These variants are in KRAS (1), PTEN (1), and TP53 (13) (Supplementary Table 4). Pre-NACT, 11/14 (86%) patients had TP53 mutations, 1/14 (7%) had KRAS, and 1/14 (7%) had a PTEN mutation. Post-NACT, the mutation profiles were very similar. In one case without TP53 mutations pre-NACT, a TP53 mutation was detected after NACT. In another case, a TP53 mutation was seen pre-NACT but not observed post-NACT. The majority of mutations that were predicted to be damaging somatic mutations (9/15) were detected both before and after treatment. Of the variants that were suspected to be germline polymorphisms, 21 of 25 were consistent pre- and post-NACT.

Figure 4.

Arend Variants Compared to AOCS Ovarian Cancer Database. Number of protein-coding mutations found in AOCS 93 patients compared to our 14 patients samples in the Pre-NACT setting.

Figure 5. Tumor mutations.

A, Pie-charts of specific tumor variants in pre- and post-NACT. Variants in red are found in both pre- and post-NACT samples. B, Tumors with KIT mutations showed increased (≥ 2 fold) expression in LEFTY2, FZD2, and FGF19; although not statistically significant, because overall levels of expression were low.

Integrating DNA mutations with RNA expression

One potential consequence of somatic DNA mutations is altered transcriptional profiles. We asked whether variants in any of the six most commonly mutated genes showed evidence for altering RNA transcription. We included variants that were potentially germline to allow for the possibility that germline mutations could contribute to therapeutic response or tumor progression. We used differential expression analysis to determine whether any genes were differentially expressed among the patients with variants in those genes. Of the 5 patients that had non-reference KIT sequences, all of these had the relatively common polymorphism M541L which is predicted to be likely benign; although, studies have suggested a role for this variant in therapeutic response (34, 35). Taken as a group, these 5 patients showed an approximately 2-fold increase in expression of 3 genes: LEFTY2 (padj=0.081, DESeq2), FGF19 (padj=0.081) and FZD2 (padj=0.081), but overall levels of expression were low (Figure 5B). While the sample size did not provide strong statistical power, given the link between FZD2 expression (Wnt pathway) and LEFTY2 (TGF-beta pathway) and platinum resistance, this finding could support the conclusion that patients with KIT variants are more resistant to platinum therapy. Previous studies have shown that some protein altering mutations in KIT affect platinum response (36, 37).

Similarly, in the three patients with non-reference sequences in PIK3CA (I391L) showed altered CHEK2 expression. Finally, we looked at variants in TP53. 11 of 14 patients showed at least one variant. Comparing pre-NACT samples in TP53 mutant patients to those with no variants or only the likely benign P72R mutation, we found 8 genes significantly (FDR<0.05) associated with mutational status. Three of these genes, RASAL1, MAP3K5 and IL3RA, are associated with MAPK signaling. For 1 of the 3 patients without P53 mutations, a KRAS G12S mutation was observed which could suggest that the KRAS was the predominant driver.

Next Generation Sequencing of cfDNA from Plasma Samples

Circulating cfDNA was isolated from the plasma of 14 patients pre- and post-NACT and from 4 patients at the time of their recurrence. NGS using the same 50 targeted gene sequencing panel was performed. Pre-NACT, there were 59 variants in 19 genes detected in the cfDNA (Figure 6A, Supplementary Table 5). Of the 59 variants, 12 were also observed in the tumor pre-NACT, but all of these 12 sequences were likely germline polymorphisms (based on a minor allele frequency of greater than 5% in the EXAC database). Post-NACT, there were 54 specific variants in 21 genes detected in cfDNA. Of the 59 variants in the plasma pre-NACT, only 6 persisted after treatment, but all of these could also be attributed to likely germline variation.

Figure 6. Tumor and cfDNA mutations.

A, Pie-charts of specific variants in plasma pre- and post-NACT. Variants in red are found in both pre- and post-NACT samples. B, Comparison of genes mutations identified in tumor (green), plasma (red), or both (blue) in pre- and post-NACT samples. Variants that were likely germline are indicated by a star.

Among the 4 patients that had plasma collected at the time of recurrence, the only persistent non-reference sequences were those that could be attributed to common germline variation. Three of 4 patients had new variants identified in the recurrent blood samples that had not been seen in the cfDNA or tumor previously (Supplementary Table 6).

Comparison of DNA Mutations in Tumor and cfDNA

All NGS results from patients pre- and post-NACT are summarized in Figure 6B. Variants identified only in the tumor remained relatively consistent in the matched samples. Genes with variants found only in the plasma showed significant variation in the matched samples. In order to compare variants found in the cfDNA to those found in the tumor, we determined the precision (proportion of cfDNA variants detected in both) and the sensitivity (proportion of tumor DNA variants detected in both) of the cfDNA using the tumor as the reference at both pre- and post-NACT. Pre -NACT and post-NACT, precision and sensitivity could be evaluated for the 4 genes with variants in both the cfDNA and tumor (PIK3CA I391M, KIT M541L41, KDR Q472H and TP53 P72R). All four of these variants have minor allele frequencies in the EXAC database of greater than 10%. Sensitivity was 100% for PIK3CA pre- and post-NACT. Precision was 60% pre-NACT and 40 % post-NACT. KDR had 60% sensitivity pre-NACT and 33% post-NACT, while precision was 100% in both the pre- and post-NACT setting. KIT had 60% sensitivity pre- and 75% post, and precision was 50% pre-NACT and 43% post-NACT. The TP53 variant P72R had 27% sensitivity pre-NACT and 17% post-NACT with a precision of 29% and 24% in the pre- and post-NACT setting, respectively. All other genes had 0% sensitivity and precision because they were only observed in either the cfDNA or the tumor DNA, but not both (Supplementary Figure 5A and 5B). Supplementary Figure 5C compares the top 4 mutated genes on the gene level in the tumor vs plasma pre/post-NACT.

Discussion

This study describes the gene expression and mutation profiles of HGSOC patients pre- and post-NACT. Significant expression changes (118 with >= 1.5 and 86 with >= 2-fold change) were observed when comparing matched tumors (p<0.05). A total of 86 genes had significant changes in RNA expression after NACT. These genes were enriched for activities important for the Cell Cycle, GADD45 and ATM Signaling Pathways. There was no significant difference in the number of variants seen in pre-and post NACT, but of the 59 variants in the plasma pre-NACT, only 6 persisted; whereas 33 of 38 specific variants in the tumor DNA remained unchanged. Our NACT used Taxanes which are microtubule stabilizing agents and cause cell cycle arrest, which explains why the expression of genes involved in the Cell Cycle Control were altered. Platinum is known to cause DNA damage, accounting for the change in the DNA damage-inducible 45 (GADD45) signaling proteins, a group of proteins critical for signal transduction involved in cellular differentiation, cell survival, and apoptosis. The top 2 up-regulated genes where nuclear receptor subfamily 4 genes (NR4A3 and NR4A1). This is a family of orphan nuclear receptors that promote cell growth and survival by activating the transcription of downstream anti-apoptotic and pro-proliferative genes, suggesting that NACT could further contribute to chemoresistance (38).

TP53 mutations are known as one of the most common mutations in cancer and over 96% of patients with HGSOC have TP53 alterations (39). In normal cells p53 is responsible for inducing growth arrest by holding the cell cycle at the G1/S regulation checkpoint when DNA damage occurs. It also activates DNA repair proteins and initiates apoptosis when DNA damage is beyond repair, which explains why genes in the TP53 signaling pathway were affected by NACT. Somatic mutations in genes involved in the homologous recombination pathway have been identified in approximately 50% of HGSOC, with only a subset of these due to germline mutations (11). In looking at the DNA Damage-Repair Pathway Scores of individual patients pre-NACT and post-NACT, the one patient with a known BRCA mutation in our cohort had the lowest DNA Damage-Repair Pathway Score at the time of interval debulking. Given this and the fact that genes involved in Hereditary Ovarian Cancer Signaling showed decreased expression in post-versus pre-NACT, this could indicate that tumors might be more susceptible to targeting the BRCA pathway after NACT. This finding warrants further investigation to determine if RNA expression changes in the DNA Damage-Repair pathway after chemotherapy are indicative of homologous recombination deficiency. While every attempt was made to remove tissue from the same anatomical location (omentum) before and after chemotherapy, changes seen following chemotherapy could reflect variability in anatomical locations rather than an effect of the chemotherapy itself.

DNA mutations in the tumor determined by targeted NGS had minimal differences before and after chemotherapy. We identified increased level of variants in the cfDNA and minimal overlap between cfDNA and tumor DNA. The majority of mutations found in cfDNA were not present in tumor and likely represent a combination of technical artifact resulting from amplifying a small amount of cfDNA and heterogeneity in the tumor that is detected in cfDNA but not represented in the tumor tissue we obtained as multiple sites were not sampled (40). Non-neoplastic DNA makes up the majority of cfDNA and is known to dilute the proportion of circulating tumor cfDNA, especially in the setting of tissue-damaging therapy (14, 41). Given the consistency between DNA mutations detected by NGS in the tumor before and after NACT and the variation in the cfDNA, it is unlikely that all the mutations detected in the cfDNA were coming from the tumor. This finding is consistent with a recent study by Nair et al in which patients with no histopathologic evidence of endometrial cancer were found to have a high frequency of somatic driver mutations in cfDNA samples from uterine lavage (42).

While our cohort is relatively small and the unique gene signature of the patients that developed platinum resistance is not statistically significant, the data contributes to the hypothesis that patients could be better classified for optimal treatment based on gene expression. In particular there is interest in genes that are associated with response to platinum therapy based on large public data sets that are also differentially expressed following treatment because they represent potentially useful drug targets. In addition, evaluation of genomic mutations in the post-NACT tumor may help identify patients for targeted therapy. Serial tissue biopsies after the completion of chemotherapy require an invasive procedure, thus making it difficult to monitor clonal evolution. The ability to use “liquid biopsies” such as cfDNA to detect prognostic serum biomarkers would allow for faster translation of discoveries into the clinic and potentially help predict therapeutic response (43).

Personalized medicine initiatives will continue to sequence individual tumors using profiling panels that capture a variety of mutations and gene expression signatures. Of note, the ongoing, 30-arm NCI-MATCH Trial is evaluating three of the genes we have highlighted in this study, including: cKIT with sunitinib; PTEN, both mutation and loss, with GSK2636771; and PIK3CA with taselisib. Possible targeted agents that could be paired with the variants seen in our cohort are shown in Supplementary Table 7; although, currently not enough is known about the dynamic biology of circulating cfDNA for it to alter treatment planning in ovarian cancer (44). Additional research studies such as this one with matched tumor and cfDNA from multiple time points are needed to better understand the role of cfDNA in the management of HGSOC. In order for cfDNA to be useful as a liquid biopsy, further refinements in both wet lab and computational techniques are necessary. Utilizing cfDNA will hopefully in the future provide a non-invasive approach to identify mutations present at the time of recurrence that could help guide therapy.

Implications:

Gene expression profiles at the time of interval debulking provides additional genetic information that could help impact treatment decisions after NACT; although, continued collection and analysis of matched tumor and cfDNA from multiple time points are needed to determine the role of cfDNA in the management of HGSOC.