MicroRNA-Regulated Ubiquitination and Lipid Metabolism Networks are Associated with Chemotherapy Response in Ovarian Cancer

High-grade serous ovarian cancer (HGSOC) is a highly lethal gynecologic cancer, in part due to resistance to platinum-based chemotherapy reported among 20% of patients. This study aims to characterize the biological mechanisms underlying chemotherapy resistance, which remain poorly understood.Analysis of mRNA/microRNA sequencing data from HGSOC patients of The Cancer Genome Atlas identied 196 differentially expressed mRNAs enriched for adaptive immunity and translation, and 21 differentially expressed microRNAs associated with angiogenesis. Co-expression network analysis identied two mRNA networks associated with chemotherapy response, which were enriched for ubiquitination and lipid metabolism, as well as three associated microRNA networks enriched for lipoprotein transport and oncogenic pathways. These network modules replicated in two independent ovarian cancer cohorts. Moreover, integrative analyses of the mRNA, microRNA and genomics (i.e. single nucleotide polymorphisms) datasets revealed potential regulation of the mRNA networks by the associated microRNAs and SNPs (i.e. expression quantitative trait loci).Thus, we report novel transcriptional networks and biological pathways associated with resistance to platinum-based chemotherapy among HGSOC patients. These results improve our understanding of the effector networks and regulators of chemotherapy response, which will help to elucidate novel therapeutic targets for ovarian cancer.


Introduction
High-grade serous ovarian cancer (HGSOC) is a highly lethal gynecologic cancer, in part due to resistance to rstline, platinum-based chemotherapy treatment among 20% of patients 1 . Chemotherapy resistant patients have a signi cantly shorter overall survival (OS) than sensitive patients, and many experience tumor recurrence within six months of completing chemotherapy 2 . There is currently no strategy for predicting response to platinum-based chemotherapy, which re ects our limited understanding of the underlying molecular mechanisms of chemoresistance 3 .
The majority of earlier studies reporting gene expression signatures associated with platinum-based chemotherapy resistance in ovarian cancer patients had used univariate analysis methods on transcriptomic data alone 4 . These methods assume that chemotherapy response is driven by a single gene. However, it is well established that chemotherapy response, like other drug response outcomes, is a complex multifactorial trait modulated by multiple genes contributing to common biological pathways 5,6 . To date, few studies have investigated chemotherapy response in ovarian cancer using multivariate methods to identify underlying gene networks or pathways [7][8][9][10][11] . Fewer still incorporated additional types of 'omics data such as microRNA (miRNA) expression and genomics data to study regulation of the gene networks 12,13 . Finally, earlier multivariate transcriptomics studies used RNA microarray data, which do not allow for discovery of novel transcript isoforms 14,15 . In contrast, RNA-sequencing data using next-generation technology include gene transcripts that may have been missed by traditional microarray pro ling.
In this study, we apply both univariate and multivariate analysis methods to high-throughput RNA sequencing data from tumors, as well as integrative analysis of germline polymorphisms, in order to identify novel biological pathways and networks associated with chemotherapy response in HGSOC patients. We further determine miRNAs and polymorphisms (i.e. expression quantitative trait loci, eQTLs) correlated with the expression of the associated gene transcripts. These ndings are validated using two independent ovarian cancer cohorts and improve our understanding of the biological mechanisms underlying resistance to platinum-based chemotherapy in HGSOC patients.

Chemotherapy response classi cation
Sequencing of mRNA and miRNA was derived from chemotherapy-naïve tumors of 191 and 205 HGSOC patients of TCGA, respectively 16 . Patients who received platinum-based adjuvant chemotherapy were selected and classi ed for chemotherapy response based on their platinum-free interval. Sensitive patients remained cancerfree for at least 12 months after chemotherapy completion, whereas resistant patients experienced cancer recurrence within 6 months (Table 1; Supplementary Methods).

Weighted correlation network analysis
The weighted correlation network analysis (WGCNA) R package 18

Replication cohort and analysis
Our results were replicated using two independent ovarian cancer cohorts. First, mRNA results were replicated in the Australian Ovarian Cancer Study cohort (AOCS; GSE9891; n = 285) 22

Results
Differentially expressed mRNAs and miRNAs between sensitive and resistant patients Differential expression analysis identi ed 196 mRNAs associated with chemotherapy response (adjusted p < 0.05) that map to 190 unique genes (Fig. 1A, Supplementary Table S1). Pathway enrichment analysis of these associated transcripts indicated enrichment of 41 annotation terms, including B-cell receptor regulation, complement activation, and peptide chain elongation (Supplementary phenotypes from published genome-wide association studies. The most common phenotypes are related to triglycerides, high-density lipoprotein (HDL), and low-density lipoprotein (LDL) cholesterol.
Network integration reveals miRNA-mediated regulation of chemotherapy response mechanisms Integration of the associated mRNA and miRNA networks determine that the plum miRNA network signi cantly correlates with the lavenderblush3 mRNA network (Spearman's ρ = -0.26, p < 0.001), and the ivory miRNA network signi cantly correlates with the darkolivegreen mRNA network (Spearman's ρ = -0.17, p = 0.023). Annotations using miRNet and miRGate determined that 20 of these mRNA-miRNA interactions are experimentally validated, while 15 others are supported by in silico predictions ( Table 2, Supplementary Tables S9-S10). Combined with the potential cis-eQTL regulation of mRNAs and miRNAs in these networks, these results reveal an integrative, multiomics view of transcriptional networks associated with chemotherapy response in ovarian cancer (Fig. 3).  Table S11). The lavenderblush3 and darkolivegreen mRNA network modules replicated in the AOCS cohort (p = 1.6e-10, log HR = -1.27 and p = 1.3e-10, log HR = -1.27, respectively) (Fig. 4A, B).

Discussion
In this study, we analyzed mRNA and miRNA sequencing data from chemotherapy-naïve tumors of HGSOC patients to identify transcripts and networks associated with chemotherapy response. Our ndings implicate novel and known biological pathways that replicated in independent cancer cohorts. In addition, we identi ed potential interactions among miRNAs and mRNAs, as well as eQTLs that potentially regulate the associated transcripts. Thus, our results provide an integrative, multi-omics view of biological networks associated with chemotherapy response.
We identi ed one mRNA co-expression network (lavenderblush3) signi cantly upregulated in platinum sensitive patients, which replicated in the AOCS. This module consists of genes involved in ubiquitin-mediated proteolysis in the endoplasmic reticulum (ER). We also detected a signi cant downregulation of genes responsible for translation initiation in sensitive patients. These ndings suggest that the unfolded protein response (UPR), a cellular process responsible for resolving ER stress, may be increasingly activated in sensitive patients compared to resistant cases. The UPR alleviates ER stress through several pathways, including increased ER-associated protein degradation (ERAD) to remove misfolded proteins, and inhibition of translation to reduce protein load in the ER 26 . ER stress promotes cisplatin resistance in OC cell lines 27 and the upregulation of ERAD genes such as VCP in the lavenderblush3 module is associated with longer OS and platinum sensitivity in HGSOC cohorts 13,28 . Finally, the lavenderblush3 genes VCP, DNAJA1, and TOPORS are overexpressed in platinum-sensitive HGSOC patients as part of a cell cycle and damage response-associated network 12 .
We identi ed a second mRNA co-expression network associated with chemotherapy resistance in our HGSOC cohort that replicated in the AOCS. The darkolivegreen module included genes associated with fatty acid metabolism (SREBF1, ACAA1, ACADVL), and the protein kinase B oncogene (AKT1), which promotes de-novo lipid biosynthesis in cancer 29 . SREBF1 is a key enzyme for cholesterol and fatty acid synthesis, and an essential gene for OC tumor growth 30 . Speci cally, SREBF1 is activated by AKT1, promoting fatty acid synthesis 31 , which favors cell proliferation in OC 32 . Expression of ACADVL, involved in the β-oxidation of long-chain fatty acids, is linked to OC metastasis and cell survival 33 . Our ndings indicate the upregulation of these lipid metabolism genes among chemotherapy resistant patients. Lipid metabolism dysregulation activates the UPR, which triggers lipid metabolism-based adaptations in the cell through several pathways, including SREBF1 regulation 34 . The interaction of these pathways may present a link between our two gene co-expression modules and warrants further study.
Differential expression analysis identi ed a downregulation of mRNA transcripts involved in the adaptive immune system, which is associated with chemoresistance. Previous studies reported that a high tumor immune score is a strong predictor of chemosensitivity in HGSOC 35 . In addition, there are potential links between this immune response activation, UPR and lipid metabolism. ER stress can induce pro-in ammatory cytokine production and UPR activation in tumor cells 36 , which can disrupt dendritic cell function in the OC tumor microenvironment 37 .
Moreover, dendritic cell function can also be inhibited by increased lipid uptake in various cancers 38 .
The ivory miRNA co-expression network module, associated with chemotherapy sensitivity in HGSOC and BLCA, is involved in the negative regulation of lipid transport. This enrichment is mainly mediated by miR-128-1 and miR-128-2, which play a key role in cholesterol and lipid homeostasis through their suppression of the ABCA1 cholesterol e ux transporter and the low-density lipoprotein receptor (LDLR) 39,40 . MiR-148a is also a regulator of these key genes 39 , which is signi cantly downregulated in resistant patients. The overexpression of ABCA1 is associated with reduced survival in OC patients 41  Integrative analysis of mRNA-seq and miRNA-seq datasets identi ed potential interactions of the associated transcript co-expression modules. The overexpression of miR-221/222 in resistant patients may be inhibiting the chemosensitivity-associated lavenderblush3 mRNA network, revealing a novel potential mechanism of chemotherapy resistance. This nding, combined with the accumulating evidence of miR-221/miR-222 involvement in chemoresistance, may point to a promising avenue for therapeutic intervention. However, overexpression of miR-221/miR-222 promotes UPR-induced apoptosis in hepatocellular carcinoma (HCC) cells 47 . Additionally, ER stress suppresses miR-221/miR-222 in HCC, promoting resistance to apoptosis. The contribution of this mechanism to chemotherapy response in HGSOC is currently unclear and presents an area for future investigation.
We also identi ed potential regulation of the darkolivegreen mRNA module by the ivory miRNA network, which may inhibit lipid metabolism in chemotherapy sensitive patients. As increased lipid metabolism by cancer cells is a known mechanism of chemoresistance in HGSOC, this miRNA-mediated inhibition may present a novel mechanism of chemotherapy sensitivity.
Finally, cis-eQTL analysis identi ed known and novel genomic variants correlated with the expression of mRNAs and miRNAs, which are associated with lipid-related phenotypes. High HDL and triglyceride levels have been correlated with increased cancer stage at diagnosis in OC patients 48 . In addition, advanced-stage OC patients with high LDL levels have a shorter PFS than patients with normal levels 49 . Further investigation of these eQTLs is necessary to further elucidate their role in platinum-based chemotherapy resistance and HGSOC prognosis.

Conclusion
Our study provides novel insight of the underlying mechanisms modulating resistance to platinum-based chemotherapy in HGSOC. Speci cally, we conducted whole-transcriptome analysis of mRNA-seq and miRNA-seq data to generate novel mechanistic hypotheses using both univariate and network methods. Moreover, we integrated this data with miRNA-seq and genome-wide SNPs to determine potential regulation of the associated transcripts and networks. Our ndings implicate novel and known signaling pathways and networks associated with chemotherapy response in HGSOC as well as regulators, which could become novel drug targets. Further studies are needed to validate these ndings in other cancers.   Supplementary Files