Prediction of clinical outcomes in ovarian cancer using genomic variation

Katharine Linder; Sofia Gabrilovich; Keely Ulmer; David Bender; Michael Goodheart; Jesus Gonzalez Bosquet

doi:10.1016/j.ygyno.2024.07.209

Objectives Single nucleotide variants (SNVs) and copy number variants (CNV) contribute to cancer heterogeneity, with thousands of variations found in each ovarian cancer. Their clinical significance and whether they can be used to predict clinical outcomes is yet to be elucidated. The objective of this study was to evaluate the association of SNV and CNVs to clinical outcomes, including survival, surgical outcomes, and chemotherapy response using multivariate prediction models. Methods We extracted RNA from tumor samples of patients with high-grade serous ovarian cancer (HGSOC; n = 112) and normal fallopian tubes (FT; n = 12) of women without personal or familial history of ovarian cancer (controls). Clinical data was collected from medical records. RNA was sequenced with the Illumina HiSeq 4000 platform. We used SAMtools and VarScan to create VCF files from the alignment and detect SNVs, as recommended by the best practices of genome sequencing. SuperFreq was used to identify significant CNVs. Survival analyses of genes with and without SNVs were performed with Cox proportional hazard ratios (HR). Association of SNV and CNV with surgical outcomes and response to chemotherapy were performed with univariate ANOVA regression analyses. P-values were adjusted with a false discovery rate (FDR) to account for multiple comparisons. Significant genomic variation from the univariate analysis was introduced in multivariate lasso regression prediction models. Results There were 4,110 significant gene-based CNV (out of 23,443) shared between cases (HGSOC) and controls; 593 (out of 15,394) unique genes had differences in SNV counts between cases and controls. There were differences in survival between the genes with and without SNVs: COL4A1, PLEKHG7, ARFGEF2, CCDC17, CEP126, CDH11, ZMYM2, and PDPN (Fig. 1). Multivariate prediction modeling for complete cytoreduction with SNVs yielded an AUC of 0.75 (95 % CI: 0.62–0.88). Lasso regression prediction modeling for response to chemotherapy with SNVs yielded an AUC of 0.79 (95 % CI: 0.71–0.88). The performance of prediction models of clinical outcomes with significant CNVs was inferior to models with SNVs. Conclusions In HGSOC, genes with somatic SNVs are associated with cancer survival. Additionally, somatic SNV and CNV in HGSOC were associated with surgical outcomes and chemotherapy response. However, prediction performance of surgical outcomes and chemotherapy response based on genomic variation was moderate. Further analyses integrating diverse types of genomic variation may be needed to improve the prediction performance of clinical outcomes.

Prediction of clinical outcomes in ovarian cancer using genomic variation

View Online

Abstract

Details

Metrics