5-Hydroxymethylcytosine signals in serum are a predictor of chemoresistance in high-grade serous ovarian cancer

Abstract

Objective: The genome-wide profiling of 5-hydroxymethylcytosines (5hmC) on circulating cell-free DNA (cfDNA) has revealed promising biomarkers for various diseases. The purpose of this study was to investigate 5hmC signals in serum cfDNA and identify novel predictive biomarkers for the development of chemoresistance in high-grade serous ovarian cancer (HGSOC). We hypothesized that 5hmC profiles in cfDNA reflect the development of chemoresistance and elucidate pathways that may drive chemoresistance in HGSOC. Moreover, we sought to identify predictors that would better stratify outcomes for women with intermediate-sensitive HGSOC.

Methods: Women diagnosed with HGSOC and known platinum sensitivity status were selected for this study. Nano-hmC-Seal was performed on cfDNA isolated from archived serum samples, and differential 5hmC features were identified using DESeq2 to establish a model predictive of chemoresistance.

Results: A multivariate model consisting of three features (preoperative CA-125, largest residual implant after surgery, 5hmC level of OSGEPL), stratified samples from intermediate sensitive, chemo-naive women diagnosed with HGSOC into chemotherapy-resistant- and sensitive-like strata with a significant difference in overall survival (OS). Independent analysis of The Cancer Genome Atlas data further confirmed that high OSGEPL1 expression is a favorable prognostic factor for HGSOC.

Conclusions: We have developed a novel multivariate model based on clinico-pathologic data and a cfDNA-derived 5hmC modified gene, OSGEPL1, that predicted response to platinum-based chemotherapy in intermediate-sensitive HGSOC. Our multivariate model applies to chemo-naïve samples regardless if the patint was treated with adjuvant or neoadjuvant chemotherapy. These results merit further investigation of the predictive capability of our model in larger cohorts.

Keywords: Biomarkers; Cell-free DNA, liquid biopsies; Chemotherapy resistance; High-grade serous ovarian cancer.

Fig. 1.

Serum-derived cfDNA as an analyte for nano-hmC-Seal. (A) Normalized cfDNA yield from patient-matched serum and plasma samples. (B) Spearman correlation coefficient matrix of serum-, plasma- and gDNA (WBC)-derived 5hmC annotated reads using a 10 kb sliding window.

Fig. 2.

Clinical cohort characteristics and serum-based 5hmC profiling in women with HGSOC. (A) Flowchart illustrating the sample selection and inclusion criteria for the final cohort and (B) Heatmap of spearman correlation matrix for factors influencing cfDNA concentration. Spearman rank correlation coefficient (r) for each comparison is shown in individual cells that have a p-value of <0.05. (C) Principal component analysis (PCA) plot of global differentially hydroxymethylated genes (DhMGs) by platinum sensitivity. (D) 5hmC distribution enrichment by genomic features for all samples. LINE: Long-interspersed nuclear element, SINE: Short-interspersed nuclear element, LTR: Long terminal repeat, TTS: Transcription termination site, 3’UTR: 3’ prime untranslated region, CpG-Island: palindromic cytosine-phosphate-linked-guanine islands, ncRNA: Non-coding RNA and 5’UTR: 5’ prime untranslated regions.

Fig. 3.

5hmC patterns during treatment in HGSOC.A, B) Hierarchical clustering based on 129 DhMGs (A) and functional enrichment analysis (B) of all samples by chemotherapy-treatment.

Fig. 4.

DhMGs can separate platinum sensitive from -resistant women. A, B) Hierarchical clustering based on 29 DhMGs (A) and functional enrichment analysis of chemo-naïve samples from women with platinum-sensitive and –resistant HGSOC. C, D) Hierarchical clustering based on 32 DhMGs (C) and functional enrichment analysis of 5hmC modified genes (D) in women with prior chemotherapy treatment. A, C) Rows indicate DhMGs.

Fig. 5.

DhMGs can be used as biomarkers to predict survival. A, D) Overview of the steps involved in model building and validation. B, E) Receiver operating curves optimized for the area under the curve for the AC- (B) and NACT-model (E). C, F) Survival curve analysis of AC-treated intermediate sensitive (C) and NACT-treated intermediate sensitive women (F) as classified by the model.