Prognostic and therapeutic relevance of molecular subtypes in high-grade serous ovarian cancer

Abstract

Molecular classification of high-grade serous ovarian cancer (HGSOC) using transcriptional profiling has proven to be complex and difficult to validate across studies. We determined gene expression profiles of 174 well-annotated HGSOCs and demonstrate prognostic significance of the prespecified TCGA Network gene signatures. Furthermore, we confirm the presence of four HGSOC transcriptional subtypes using a de novo classification. Survival differed statistically significantly between de novo subtypes (log rank, P = .006) and was the best for the immunoreactive-like subtype, but statistically significantly worse for the proliferative- or mesenchymal-like subtypes (adjusted hazard ratio = 1.89, 95% confidence interval = 1.18 to 3.02, P = .008, and adjusted hazard ratio = 2.45, 95% confidence interval = 1.43 to 4.18, P = .001, respectively). More prognostic information was provided by the de novo than the TCGA classification (Likelihood Ratio tests, P = .003 and P = .04, respectively). All statistical tests were two-sided. These findings were replicated in an external data set of 185 HGSOCs and confirm the presence of four prognostically relevant molecular subtypes that have the potential to guide therapy decisions.

Figure 1.

Survival per molecular subtype. Kaplan Meier curves for (A) TCGA classification in the original cohort of 489 patients with high-grade serous ovarian cancer and (B) for the TCGA classification applied to 174 patients with high-grade serous ovarian cancer from Mayo Clinic with long-term clinical follow-up. For the original TCGA data set, survival was capped at 60 months. All statistical tests were two-sided.

Figure 2.

Subclasses were computed by applying a consensus NMF clustering method, which confirmed the presence of four stable clusters. A) Consensus matrices and sample correlation matrices are shown for k = 2 to k = 8, using the 1850 genes (2040 probe sets) with the highest variability across patients using median absolute deviation. B) Differentially expressed marker genes were determined by significance analysis of microarrays, and subtype names were chosen based on prior nomenclature and the expression of signature genes: immunoreactive-like, differentiated-like, proliferative-like, and mesenchymal-like. Using the gene class signatures, 174 samples and the 1000 most differentially expressed genes were ordered based on subtype assignments. Each column represents a sample; each row represents a gene set. C) Kaplan Meier curves for molecular subtypes among 174 Mayo Clinic high-grade serous ovarian cancers for overall survival is shown. The P value was calculated using a two-sided log rank test. Differentially expressed marker genes were determined by significance analysis of microarrays, and subtype names were chosen based on prior nomenclature and the expression of signature genes: immunoreactive-like, differentiated-like, proliferative-like, and mesenchymal-like.

Figure 3.

Kaplan Meier curves for (A) the de novo Mayo Clinic molecular classification and (B) for the TCGA classification applied to 185 high-grade serous ovarian cancers (Bonome cohort) and overall survival (OS) for each subtype. P values were calculated using the two-sided log-rank test.

Figure 4.

HGSOC signature ssGSEA scores. ssGSEA scores for 174 HGSOC tumor samples were generated using described gene expression signatures: differentiated-like, immunoreactive-like, mesenchymal-like, and proliferative-like. A) Raw gene set activation scores. Each column represents one sample; each row represents one gene signature. B) Binary scores indicating whether a tumor sample activates the gene signature. Each column represents one sample; each row represents one gene signature. Red = activated; black = not activated.