Abstract
DNA methylation aberrations have been implicated in acquired resistance to platinum drugs in ovarian cancer. In this study, we elucidated an epigenetic signature associated with platinum drug resensitization that may offer utility in predicting the outcomes of patients who are coadministered a DNA methyltransferase inhibitor. The ovarian cancer specimens we analyzed were derived from a recent clinical trial that compared the responses of patients with recurrent platinum-resistant ovarian cancer who received carboplatin plus the DNA methyltransferase inhibitor guadecitabine or a standard-of-care chemotherapy regimen selected by the treating physician. Tumor biopsies or malignant ascites were collected from patients before treatment (day 1, cycle 1) or after treatment (after 2 cycles) for epigenomic and transcriptomic profiling using the Infinium HumanMethylation450 BeadChip (HM450). We defined 94 gene promoters that were hypomethylated significantly by guadecitabine, with 1,659 genes differentially expressed in pretreatment versus posttreatment tumors. Pathway analysis revealed that the experimental regimen significantly altered immune reactivation and DNA repair pathways. Progression-free survival correlated with baseline expression levels of 1,155 genes involved in 25 networks. In functional investigations in ovarian cancer cells, engineered upregulation of certain signature genes silenced by promoter methylation (DOK2, miR-193a, and others) restored platinum drug sensitivity. Overall, our findings illuminate how inhibiting DNA methylation can sensitize ovarian cancer cells to platinum drugs, in large part by altering gene expression patterns related to DNA repair and immune activation, with implications for improving the personalized care and survival outcomes of ovarian cancer patients.
Significance: Epigenomic targeting may improve therapeutic outcomes in platinum-resistant and recurrent ovarian cancer in part by effects on DNA repair and antitumor immune responses. Cancer Res; 78(3); 631–44. ©2017 AACR.
Introduction
Advanced stage ovarian cancer has poor outcome, five-year survival rates remaining less than 25% despite advances in therapy (1, 2). Although most women with ovarian cancer initially respond to platinum-based chemotherapy, relapses occur in most patients, leading to the development of platinum-resistance, which is uniformly fatal (3, 4). It has been postulated that ovarian cancer progression to a platinum-resistant state is intimately linked to accumulated epigenomic alterations, including increased DNA methylation and modifications of histone marks (5–7). Such changes cause transcriptional repression of tumor suppressor (TSG) and of other genes associated with apoptotic responses to chemotherapy (8, 9). For example, promoter methylation has been involved in silencing of TSGs (e.g., BRCA1, MLH1, RASSF1A, DAPK, DOK2, OPCML) and of differentiation-associated transcription factors like HOXA10 and HOXA11 (5, 10, 11) and was linked to ovarian cancer initiation and chemotherapy resistance (8, 9).
On the basis of this rationale, over the past decade, others and we have conducted bench-to-clinic therapeutic interventions targeting aberrant DNA methylation to resensitize ovarian tumors to platinum (12–18). In phase I clinical trials, combinations of platinum and hypomethylating agents were found to be tolerable and biologically active, as measured through global and gene-specific DNA methylation assays in peripheral blood mononuclear cells (PBMC) and tumors. Two subsequent phase I/II clinical trials demonstrated significant clinical efficacy of combination regimens using DNA methyltransferase inhibitors (DNMTI, decitabine or 5-azacitadine) and platinum, including long progression-free survival (PFS) and high response rates in platinum-resistant ovarian cancer (15, 19). These observations supported a randomized clinical trial testing a next-generation DNMTI inhibitor guadecitabine (formerly known as SGI-110, Astex Pharmaceuticals, Inc.), and carboplatin against physician's choice chemotherapy (NCT01696032), which was recently completed (20). Guadecitabine administered daily for 5 days had been found to be tolerable, biologically, and clinically active in a previous trial in patients with myelodysplastic syndrome and acute myelogenous leukemia (21). Tumor biopsies or cancer cells harvested from malignant ascites were collected at baseline and after 2 cycles of treatment. Methylome and transcriptome analyses determined guadecitabine-induced changes in paired tumor biopsies and correlations with PFS. The randomized study was preceded and supported by a phase I component that demonstrated tolerability of guadecitabine at a dose of 30 mg/m2 daily for 5 days preceding carboplatin administration on day 8 in ovarian cancer patients (18). Pharmacodynamic analyses conducted using PBMCs showed that the regimen was biologically active, as evidenced by approximately 20% decrease in global (L1RE1 LINE1) DNA methylation (18). Together with the observed clinical activity, these data provided strong rationale for pursuing the randomized phase II trial. Clinical results of the study will be reported separately.
Here, we show that guadecitabine delivered daily over 5 days in low dose alters the epigenome and transcriptome of ovarian tumors, particularly impacting the expression of pathways associated with immune reactivation and DNA repair. Expression of 293 genes on day 1 cycle 1 organized in networks related to immune response processes were associated with PFS, supporting the hypothesis that clinical endpoints are impacted by the biological activity of the DNMTI. Furthermore, we identified several TSGs (e.g., DOK2, miR193a) whose regulation was directly linked to the development of platinum resistance in ovarian cancer cell lines. However, we believe that alteration of functional pathways, rather than of specific transcripts, are responsible for the clinical activity induced by the combination regimen. A key pathway impacted by treatment with the DNMTI and associated with longer PFS relates to T-cell–mediated immune response, suggesting that alterations in the tumor microenvironment induced by guadecitabine contribute to the clinical effects of this new resensitization regimen.
Patients and Methods
Patients
Patients with platinum-resistant recurrent high-grade serous, endometrioid, mixed cell, or clear cell epithelial ovarian cancer (grade 2 or 3), primary peritoneal carcinomatosis (PPC), or fallopian tube cancer were eligible for treatment. Eligible patients had acceptable organ function based on laboratory data, Eastern Cooperative Oncology Group performance status (ECOG PS) of 0 or 1, were at least 3 weeks from the last therapy. The protocol and informed consent form were reviewed and approved by an Institutional Review Board/Independent Ethics Committee at each study center prior to implementation. Patients provided written informed consent before enrollment. The study is registered on ClinicalTrials.gov, number NCT01696032 and was conducted in accordance with the International Council for Harmonisation Good Clinical Practice guidelines, applicable local regulatory requirements, and the principles enunciated in the Declaration of Helsinki.
Study design
This study had two stages: A safety lead-in stage (stage 1) followed by a randomized, controlled, open-label stage (stage 2). Study design is detailed in Supplementary Material. Tumor biopsies or malignant ascites/or pleural fluid were obtained from consenting patients at baseline (cycle 1, day 1) and after 2 cycles of guadecitabine + carboplatin (cycle 2, day 8, prior to the carboplatin infusion) by imaging-guided core biopsies. Two 18-gauge cores were obtained for each time point and material was immediately snap frozen (∼25–50 mg/specimen). Ascites or pleural fluid was centrifuged and fluid and cell pellets were separated prior to cryopreservation.
Cell lines and culture
The ovarian cancer cell lines OVCAR3 and SKOV3 were obtained from the ATCC and the OVCAR5 cell line from the Developmental Therapeutics Program at the National Cancer Institute. Cell culture conditions and transfection are described in Supplementary Materials. The SKOV3 and OVCAR3 cell lines were authenticated by the ATCC in 2012 and 2015, respectively. The OVCAR 5 cell line was authenticated by IDEXX BioResearch in 2015.
DNA and RNA extraction
DNA and total RNA were extracted from approximately 25 mg tumor tissue or 200 μL ascitic fluid using AllPrep DNA/RNA/Protein Mini kit (Qiagen) following the manufacturer's protocol. RNA used for real-time RT-PCR was extracted from cultured cells with TRI-Reagent (Sigma-Aldrich) or a miRNeasy kit (Qiagen). DNA and RNA concentrations were determined using the absorbance at 260 nm, and purity was assessed on the basis of the 260/280 nm absorbance ratio.
Methylome analysis
DNA extracted from patient samples was bisulfite converted and DNA methylation was assayed by using the Infinium Human Methylation 450 BeadChip (HM450; Illumina) at the University of Chicago Genomics Facility. Data preprocessing and analyses were conducted in the statistical programming environment R v3.1.2 with the package RnBeads v0.99 (16, 22, 23). Normalization and background correction were applied to methylation data with manufacturer-recommended algorithms implemented in methylumi package (see Supplementary Materials; refs. 24, 25). Methylation levels were averaged for the replicates for each biopsy after normalization and the difference in methylation β-value between two groups or the mean of the pairwise difference for paired samples was calculated. To correct P values for multiple hypothesis testing, false discovery rates (FDR) was calculated by using an improved Benjamini–Hochberg procedure (26), and the methylation changes in CpG sites/regions with FDR < 0.05 were considered statistically significant. Ingenuity Pathway Analysis (IPA) was used to identify functional interactions of genes differentially methylated between groups. Average methylation signals on the CpG sites within each CpG island and/or promoter region were hierarchically clustered with Pearson dissimilarity and average linkage as clustering parameters.
In the analysis of the genomic context distribution of CpG sites, individual sites were annotated in regards to positional context relative to nearby CpG island and transcriptional starting site (TSS) based on the Illumina HumanMethylation450 BeadChip manifest. CpG islands were defined as DNA regions longer than 500 bases containing more than 55% GC content and with an observed-to-expected CpG ratio greater than 40%. CpG shores were defined as 2-kb regions immediately upstream and downstream of CpG islands and CpG shelves were defined as 2–4 kb regions from the islands. CpGs unrelated to a CpG islands were classified as “open sea.” CpG sites were considered as being linked to a transcript if they mapped to the 200 to 1,500 bases region upstream of TSS (TSS1500), within 200 bases upstream (TSS200), 5′UTR, 1st exon, gene body, or 3′UTR. All other CpG sites not mapping to the abovementioned regions relative to a transcript were defined as “not linked to gene.” The Illumina HumanMethylation450 BeadChip results are available for download at Gene Expression Omnibus data repository at the National Center for Biotechnology Information under the accession number GSE102119.
Transcriptome analysis
RNA sequencing was performed essentially as we have described previously (27, 28). See Supplementary Materials for detailed description. The resulting reads were mapped against GRCh38.p5 using TopHat2 version 2.1.1 (29). TopHat uses Bowtie, which is based on the Burrows–Wheeler transform algorithm, for sequence alignment and allows for mapping across exon junctions (25). Read counts for each gene were created using HTSeq-Count from the HTSeq package version 0.6.1p1 and Gencode v23 as the annotation (30, 31). Custom perl scripts were used for estimation of transcript abundances based on fragments per kilobase of exon per million fragments mapped (FPKM). The differential expression analysis was carried out using the DESeq2 package (version 1.12.3) in R/Bioconductor (R version 3.3.1; ref. 32). DESeq2 fits a generalized linear model (GLM) of the negative binomial family, and a Wald test is performed to evaluate the significance of the coefficients. To define broad pathway responses, gene expression changes in the paired biopsies were assessed using Kyoto Encyclopedia of Genes and Genomes, Gene Ontology, and IPA and GeneGo. The RNA-sequencing results are available for download at Gene Expression Omnibus data repository at the National Center for Biotechnology Information under the accession number GSE102118.
Real-time RT-PCR
For validation of the RNA sequencing data, qRT-PCR for selected targets was done as we have described previously (see Supplementary Materials; ref. 16).
Cell proliferation
Cell survival was measured using a Cell Counting Kit-8 (Dojindo Molecular Technologies, Inc.) according to the protocol provided by the manufacturer. Absorbance at 450 nm was determined with an EL800 microplate reader (BioTek Instruments, Inc.). Student t tests were used to compare experimental groups. P < 0.05 was considered significant.
Clonogenic assay
Cells were seeded on 6-well plates at the density of 100 cells per well and cultured for 10 days. Colonies were fixed with 4% paraformaldehyde, stained with 1% crystal violet in 2% ethanol, and then counted. Student t tests were used to compare experimental groups. P < 0.05 was considered significant.
Correlation of C1D1 levels with clinical outcome
Correlations of C1D1 promoter methylation and gene expression profiles with clinical outcomes were conducted using both logistic and Cox regression. For the logistic regression models, the outcome was PFS ≤ 91 days (the median time to PFS) versus PFS > 91 days. For Cox regression models, the outcome was PFS. Promoter methylation and gene expression levels were tested separately with the FDR controlled at the 0.05 level. For the DNA methylation data, M values were adapted to correlate with clinical outcomes. Potential covariates considered included age, race (white, black, Asian, other), ECOG score (0 or 1), ovarian cancer (yes/no), fallopian tube cancer (yes/no), crossover from treatment choice (TC; yes/no), and number of treatment regimens > 3 (yes/no). A step-wise selection method was used to determine final covariate models, and P < 0.05 was used as the threshold for inclusion of a covariate. Covariates in the final models included ECOG, crossover from TC, race, and number of treatment regimens > 3. Models were fit using R v3.1.2 (glm and coxph functions).
Results
Study population
Figure 1A illustrates the study design. A total of 124 subjects were enrolled in the study and 120 were treated (see Supplementary Materials). A total of 98 subjects received guadecitabine + carboplatin (20 subjects in stage 1, 51 subjects in stage 2, and 27 TC subjects who were crossed over to guadecitabine + carboplatin after disease progression in stage 2). Figure 1B illustrates treatment administration schema and timing of tumor biopsies. “Baseline” biopsies were collected preguadecitabine on cycle 1 day 1 (C1D), and posttreatment were collected after two full cycles of daily × 5 doses of guadecitabine on day 8 (C2D8). Pre-guadecitabine (C1D1) tumor biopsies were obtained from 42 patients enrolled on this trial, of which 40 yielded high quality nucleic acids, whereas post-guadecitabine (C2D8) biopsies were obtained from 11 patients, of which 9 yielded high-quality nucleic acids. Malignant fluid collection on C1D1 and C2D8 were obtained from 23 and 8 patients, respectively (Fig. 1C). Clinical and DNA methylation data at C1D1 were available for 35 patients. Clinical and RNA-sequencing data at C1D1 were available for 33 patients.
Genome-wide effects of guadecitabine on DNA methylation
Epigenetic deregulation contributes to both the onset and maintenance of chemoresistance. To evaluate biological effects of the regimen guadecitabine + carboplatin, demethylating activity in vivo was assessed. Global DNA methylation was assessed by using the Infinium HumanMethylation450 BeadChip (HM450) arrays on tumor biopsies collected at baseline on C1D1 (n = 40) and posttreatment collected on C2D8 (n = 9). Differential methylation was conducted in both CpG site level and region level. P value after multiple comparison correction FDR = 0.05 was considered as significant; 18,644 sites and 191 promoters were found significantly differentially methylated in tumor samples after treatment. Results of the differential methylation between treatment and baseline in paired tumor and paired ascites are shown as volcano plots in Fig. 2. Methylation changes were defined on the basis of the β-value for each CpG site with a Δβ > 0.2 for patients within a group for gene hypermethylation and Δβ < 0.2 for hypomethylation (i.e., “demethylation”). Setting a significance threshold of P < 0.01 and displaying CpG site differential methylation (post- vs. pretreatment) as volcano plots, we describe 94 genes with significantly altered CpGI methylation levels between paired tumor biopsies (pre- vs. posttreatment; Fig. 2A, left, indicated by the shift to the left, red dots are differentially methylated CpG sites with FDR < 0.05), and 17 genes had significantly altered CpGI methylation levels between paired ascites (Fig. 2A, right).
As the lists of differentially methylated genes (Supplementary Table S1) in tumors (94 genes) and ascites (17 genes) by themselves do not provide immediate insights into the biological mechanisms of response to the combined therapy, we examined the pathways enriched by genes hypomethylated after guadecitabine treatment by using IPA. Significantly enriched pathways in the paired tumor biopsies were related to altered metabolism and signaling (Fig. 2B, left), whereas the most enriched pathways in the ascites were related to DNA damage and repair and epithelial-to-mesenchymal transition (EMT; Fig. 2B, right). One of the pathways enriched by guadecitabine-induced hypomethylation in tumors was 14-3-3σ, a pathway involved in DNA repair and chemoresistance (33). To define biologically meaningful molecular interactions, we constructed gene–gene interaction networks for the two top networks: Immuno-Response-Related and DNA Replication and Repair for tumors (Fig. 2C, left) and ascites (Fig. 2C, right). These results demonstrate significant guadecitabine-induced gene demethylation in tumors and ascites, affecting interacting gene pathways and networks that potentially contribute to platinum resistance and can be involved in reversal of resistance by the hypomethylating agent.
The probes interrogated on the HumanMethylation450 array are distributed between CpG islands (∼one-third), regions flanking the CGIs (within 4 kb of the nearest island) referred to as “shelves” and “shores” and representing another third of all sites, and sites unrelated to an island, generally occurring within the gene body and termed as “open sea” sites, which represent another third of all probes. The differentially methylated sites were primarily found in the open sea (46%), versus islands (18%), shores (17%), and shelves (18%; Fig. 2D, left). Of all guadecitabine-hypomethylated sites, 9% were within 1,500 bp of the TSS or in the first exon of a gene, whereas another 8% and 6%, respectively, were found in the 3′ UTR and 5′ UTR, respectively (Fig. 2D, right). A significant percentage of differentially methylated sites in response to guadecitabine resided in gene bodies (∼65%).
Guadecitabine-induced transcriptomic changes
To examine global gene expression changes induced by guadecitabine in tumors, we performed RNA-sequencing on tumor samples (C1D1/baseline, 39 patients; vs. C2D8/posttreatment, 8 patients). First, the tumor content of the biopsies was ascertained by using a Nonnegative Matrix Factorization (NMF) approach (see Supplementary Material). All specimens, except one (which was subsequently excluded from analyses), had tumor content ≥60%, exceeding the tumor purity of the TCGA ovarian cancer samples (See Supplementary Fig. S1). A total of 502 genes were significantly upregulated by guadecitabine (fold change > 2, FDR < 0.05; Supplementary Table S2). The heatmap in Fig. 3A and Supplementary Table S2 illustrates significant transcriptomic changes induced by guadecitabine treatment (C2D8, orange samples) compared with pretreatment (C1D1, green). Up- and downregulated genes are marked by red and blue, respectively. To further investigate the potential functional relevance of the differentially expressed genes before (C1D1) and after (C2D8) guadecitabine, IPA was performed. The most significant pathways altered in response to treatment were IL8 (−logP = 8.31; Fig. 3B) followed by signaling by Rho Family GTPases (−logP = 7.55; Fig. 3B). Of the top 45 pathways, the most highly enriched were predominantly immune-related (15/45) and cancer signaling (20/45) pathways (Fig. 3B; Supplementary Fig. S2). Also noteworthy was enrichment of the DNA double-strand break repair by homologous recombination pathway (−logP = 2.05; Supplementary Fig. S2), an important pathway involved in chemoresistance and chemoresensitization. The complete list of enriched pathways is included in Supplementary Fig. S2. Networks were also constructed by IPA among the top 1,659 differentially expressed genes (FC > 2, FDR < 0.05). The top networks predicted to be activated by guadecitabine were immune response (phagocytes, neutrophils, leukocytes), adhesion of immune cells and homing of leukocytes (Fig. 3C, center hubs; red and green circles represent up- or downregulated genes activating (or inhibiting) these immune networks shown on periphery), suggesting that the hypomethylating agent upregulated the immune signaling/response in vivo.
Integrated methylome–transcriptome analysis
DNA methylation plays an important role in gene regulation and by inducing demethylation, DNMTIs can cause gene upregulation. To examine the relationships between CpG island methylation and gene expression in tumor cells after treatment with guadecitabine, we performed an integrated transcriptome–methylome analysis on all C1D1 versus C2D8 tumor samples. As CpG island hypermethylation has been shown to directly silence genes, we focused on promoter CpG island-containing genes that were hypomethylated and upregulated after guadecitabine treatment and identified 77 genes meeting this criteria (Fig. 4A; Supplementary Table S3) and observed separate clustering of the C1D1 (green) from C2D8 (orange) samples. The network constructed from the 77 reactivated genes included changes in T lymphocyte chemotaxis, inflammatory response, migration of neutrophils, adhesion of immune cells, and homing of leukocytes (Fig. 4B), in total predictive of a significantly altered immune response after guadecitabine treatment. Overall, our integrated ‘omics analysis showed that guadecitabine induced a potent reprogramming of the ovarian cancer methylome and transcriptome to drive major immunomodulatory changes.
Relationship of guadecitabine-upregulated genes and clinical responses
At C1D1, 1,155 genes were significantly associated with PFS in the multivariable Cox regression analysis after correcting for multiple comparisons (Fig. 5A; Supplementary Table S4). Pathways enriched by the 1,155 genes include DNA damage-induced 14-3-3σ signaling (Fig. 5B; also seen enriched in Fig. 2B). After guadecitabine treatment, of the 1,155 genes, 203 were significantly demethylated (P < 0.05, Supplementary Table S5; Fig. 5C). Among this group of demethylated/reexpressed genes with C1D1 levels that correlated with clinical outcomes, cancer metabolic pathways and the transcriptional regulatory network of embryonic stem cells were enriched (Fig. 5C). An additional 293 genes upregulated by guadecitabine (FC > 1.3, P < 0.05, Supplementary Table S6) had C1D1 levels that correlated with clinical outcome. Pathways enriched in the group of PFS-associated genes included molecular mechanisms of cancer, apoptosis signaling, cancer metabolism (e.g., amino acid degradation and biosynthesis), and netrin signaling, a tumor suppressor pathway known to be inactivated by methylation in other cancer types (Fig. 5D). No significant correlations were found between promoter methylation or gene expression and PFS by logistic regressions or between C1D1 methylation and PFS by Cox regressions after correcting for multiple comparisons.
Functional validation
We used in vitro cellular assays to validate and further investigate specific genes and pathways of interest, selected based on the RNA sequencing and DNA methylation analyses, our previous studies, and other reports describing their positive contributions to DNA damage–associated apoptosis and/or ovarian cancer prognosis/outcome. For validation, we selected two genes re-expressed in response to guadecitabine representing functional classes, TSGs, and cellular signaling (docking protein 2, DOK2 and miR193a; Fig. 6A). Both DOK2, an adapter protein downstream of tyrosine kinase, and miR-193a, have been previously reported to play tumor-suppressive roles in cancer and to modulate chemoresistance (34–36).
Treatment of ovarian cancer cell lines SKOV3, OVCAR3, and OVCAR5 with the DNMTI inhibitor decitabine significantly increased the expression of both DOK2 and miR193a (Fig. 6B and C). To determine the mechanism by which gene reactivation by guadecitabine resensitizes ovarian cancer cells to platinum therapy, we overexpressed DOK2 (Fig. 6D) or miR-193a (Fig. 6E) in ovarian cancer cells and examined functional changes using cell proliferation and clonogenic assays in vitro. DOK2 overexpression significantly lowered the concentration of cisplatin required to inhibit SKOV3 ovarian cancer cell proliferation compared with control (Fig. 6F). Furthermore, miR-193a overexpression decreased (P < 0.05) proliferation (Fig. 6E) and colony formation of SKOV3 (Fig. 6G) ovarian cancer cells, supporting a tumor suppressor function. These results demonstrate that DNA methylation inhibitors lead to upregulation of expression of genes silenced in ovarian cancer, induce cancer cell growth inhibition, improve chemosensitivity, likely contributing to the clinical benefit recorded in the trial.
Furthermore, we validated one of the pathways found to be enriched in hypomethylated genes in guadecitabine-treated tumors (Fig. 2B), which is involved in cell-cycle control and response to chemotherapy (SFN, also known as 14-3-3σ). Knock down of 14-3-3σ by shRNA (Fig. 7A) resulted in increased sensitivity to cisplatin (Fig. 7B), decreased colony formation ability (Fig. 7C), and cell proliferation (Fig. 7D). These results support the involvement of 14-3-3σ in platinum resensitization and tumor growth inhibition induced by the regimen.
Discussion
Here we present integrated methylomic and transcriptomic analyses of tumor specimens obtained at baseline and after two cycles of guadecitabine and carboplatin from women with recurrent, platinum-resistant ovarian cancer enrolled in a randomized phase II trial comparing this strategy against physician's choice of standard chemotherapy. These tumor analyses revealed profound gene expression changes induced by the hypomethylating agent in combination with platinum. Furthermore C1D1 values of these same genes were correlated with observed clinical effects of the regimen. Our study has several important implications.
First, the clinical trial from which tumor and ascites specimens were collected and analyzed represents the first randomized comparison between a novel epigenome targeting strategy and the standard chemotherapy approach for platinum-resistant ovarian cancer. The concept of methylome targeting used in our trial is rooted in accumulating evidence that chromatin changes associated with DNA methylation occur during cancer progression as a consequence of cellular stress, chronic inflammation, and DNA damage. DNMT inhibitors have been shown to be global chromatin remodelers and to exert untargeted effects on the epigenome. While these agents have shown clear activity in myelodysplasia and leukemia, where they are approved for clinical use, their development in solid tumors has lagged due to difficulties of designing tolerable combinations with chemotherapy (37) or biological therapy, and limitations in clinical trial design. In recent years, our group and others have shown that DNMT inhibitors can be safely combined with platinum when using low repetitive doses, which maintain drug biological activity (12, 15, 19). The clinical results of the current study are reported separately, but of note is that the median rate of PFS at 6 months was 37% for the experimental regimen compared with 13% for physician's choice treatment (P = 0.05), suggesting that the methylome-targeted regimen induces platinum resensitization in this setting, and is clinical active.
Second, we show here that guadecitabine induced significant hypomethylation in tumor biopsies and in cells recovered from ascites fluid. These observations are consistent with analyses of PBMCs from patients enrolled in this trial in which guadecitabine-induced demethylation of LINE1 elements was recorded (18). Differences between networks of genes hypomethylated in tumors versus ascites likely reflect differences in tumor milieu and the “metastatic” state of tumor cells in ascites, with EMT being one of the prominent pathways affected. Significant guadecitabine-induced hypomethylation was also noted in vivo by our group in preclinical studies using ovarian cancer xenografts treated with this agent (13, 16). Global hypomethylation promoted by other DNMTIs in clinical specimens was demonstrated in previous studies testing either decitabine or 5-azacitadine in patients with ovarian cancer or solid tumors (12, 15, 38). Our study is limited by the lack of histologic confirmation, due to scant material obtained through core biopsies and prioritization of nucleic acid extraction for genomic analyses. To mitigate this limitation, an NMF strategy deconvoluted the transcriptomic data and samples included in analyses had a predicted tumor content exceeding 60% (see Supplementary Material; Supplementary Table S7). CIBERSORT analysis confirmed that the observed significantly demethylated/methylated CpG sites were independent of infiltrating lymphocytes in pre-/posttreatment tissues (See Supplementary Material; Supplementary Table S8; Supplementary Fig. S3). In addition, changes in posttreatment tumor biopsies may reflect not only effects of HMA, but also cellular stress in response to platinum.
Interestingly, more than half of guadecitabine-induced differentially methylated sites occurred in gene bodies, consistent with prior observations, which identified such sites as becoming demethylated in response to DNMTIs and rapidly rebounding after the inhibition was removed (39). While the significance of gene body site methylation is not fully elucidated, emerging evidence points to such sites located near enhancer regions as potential regulators of gene expression (40). In addition, removal of gene body CG sites methylation has been shown to be associated with increased H2A.Z occupancy and enrichment in H3K27me3, contributing to regulation of gene transcription (39).
Integration of methylomic and transcriptomic signals induced by DNMTIs has been generally difficult to demonstrate in human specimens. This is partly explained by variable clinical responses in small datasets and limitations of using core biopsies, which generate small amounts of high quality nucleic acids. In addition, we acknowledge the nonspecific activity of hypomethylating agents, which exert genome-wide effects. It is likely that the resensitization effects induced by the regimen are not caused by a single transcript reexpression, but by genome-wide reprogramming and activation (or inhibition) of cancer-relevant pathways. Interestingly, among the hypomethylated and reexpressed pathways in tumors, we identified gene networks related to DNA repair, metabolism, and immune response, similar to our previous findings in the phase I/II clinical trial using decitabine in combination with carboplatin (15). A number of metabolic pathways were represented among genes found to be hypomethylated and associated with clinical outcomes. These results are consistent with findings in colon cancer cells exposed to 5-azacitadine, where gene body hypomethylation affected primarily metabolic processes regulated by c-MYC (39).
An important observation from our study relates to reactivation of immune response pathways in ovarian tumors treated with guadecitabine. These results are consistent with our previous and others' recent findings that epigenome reprogrammers elicit immune stimulation in vivo (15, 41, 42). A proposed mechanism is DNMTI-induced tumor antigen reexpression (43), as previously shown for the cancer-testis antigen NY-ESO-1 and MAGE antigens, which are frequently repressed epigenetically in ovarian tumors (43, 44) and reexpressed as a consequence of treatment with hypomethylating agents (45, 46). A recent study using preclinical ovarian cancer models showed that Th responses are repressed epigenetically and that removal of this break using epigenetic modulators potentiates response to immunotherapy, including checkpoint inhibitors (47). In addition, two other studies reported that DNMTIs induce immune signaling in cancer cells by augmenting the viral response pathway and inducing reexpression of endogenous retroviral genes incorporated in the human genome (48, 49), suggesting that other immunogenic pathways are also regulated epigenetically. These observations suggest that epigenetic modulators, in particular DNMTIs and HDAC inhibitors may serve as potent primers for immune-directed therapy. Indeed, in a phase I study, decitabine potentiated the effects of a NY-ESO-1–directed vaccine, augmenting T-cell immune responses and inducing antitumor activity (46) and several new clinical trials exploring combinations of immunotherapy and DNMT inhibitors have recently been activated.
Finally, we demonstrate that guadecitabine and carboplatin promote reactivation of a number of genes and pathways with tumor suppressor function. Among genes reexpressed in tumor specimens after guadecitabine, we validated that DOK2 and miR193a are silenced epigenetically in ovarian cancer cell lines. Upon reexpression in response to guadecitabine, DOK2 resensitized ovarian cancer cells to platinum and inhibited cancer cell proliferation and clonogenicity, consistent with its previously described functions (50). Likewise, we show that miR193a has tumor suppressor function and is rapidly induced by hypomethylating agents. Finally, forced downregulation of 14-3-3σ, a pathway whose activation in response to guadecitabine associated with clinical outcomes, inhibited cancer cell proliferation and clonogencity, and resensitized ovarian cancer cells to platinum.
In conclusion, coupled with the clinical results of this study, our tissue-based analyses provide important insight into the mechanisms by which methylome-targeting strategies exert antitumor effects. We show that global DNA hypomethylation induced by guadecitabine affects metabolic and immune responses and reactivates TSGs, which contribute to response to platinum. Our data suggest that guadecitabine resensitizes ovarian tumors to platinum; however, antitumor effects exerted by the novel DNMTI either directly, or indirectly via immune reactivation, cannot be excluded and support future studies exploring hypomethylating agents with other biological interventions.
Disclosure of Potential Conflicts of Interest
D. Matei is a consultant/advisory board member for Astex Inc. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: Y. Liu, K.P. Nephew, D. Matei
Development of methodology: F. Fang, H. Huang, G. Jiang, Y. Liu, K.P. Nephew, D. Matei
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Fang, H. Cardenas, C. Zhang, K.P. Nephew, D. Matei
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Fang, H. Cardenas, H. Huang, G. Jiang, S.M. Perkins, C. Zhang, Y. Liu, K.P. Nephew, D. Matei
Writing, review, and/or revision of the manuscript: F. Fang, H. Cardenas, H. Huang, G. Jiang, S.M. Perkins, C. Zhang, H.N. Keer, Y. Liu, K.P. Nephew, D. Matei
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H.N. Keer, K.P. Nephew, D. Matei
Study supervision: K.P. Nephew, D. Matei
Other (performed experiments): H. Cardenas
Acknowledgments
We thank Dr. Doug Rusch, Aaron Buechlein, and Dave Miller (Center for Genomics and Bioinformatics, Indiana University, Indianapolis, IN) for help with data analysis and Dr. Pearlly Yan (The Ohio State University, Columbus, OH) for technical assistance. This work was funded by the National Cancer Institute Award CA182832-01 and the V- Foundation (to D. Matei and K.P. Nephew).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.