Establishing Molecular Subgroups of CD8+ T Cell-Associated Genes in the Ovarian Cancer Tumour Microenvironment and Predicting the Immunotherapy Response

Abstract

Background: The mechanism by which infiltrating CD8+ T lymphocytes in the tumour microenvironment influence the survival of patients with ovarian cancer (OC) remains unclear.

Methods: To identify biomarkers to optimise OC treatment, 13 immune-cell-line-associated datasets, RNA sequencing data, and clinical data from the GEO, TCGA, and the ICGC were collected. Gene expression in OC was assessed using quantitative reverse transcription polymerase chain reaction (qRT-PCR) and immunohistochemistry (IHC) staining.

Results: We identified 520 genes and three immunological clusters (IC1, IC2, and IC3) associated with CD8+ T cells. Higher IFN scores, immune T cell lytic activity, and immune cell infiltration and upregulated expression of immune-checkpoint-related genes indicated that IC3 is more responsive to immunotherapy, whereas IC1 and IC2 have a poorer prognosis. A 10-gene signature, including SEMA4F, CX3CR1, STX7, PASK, AKIRIN2, HEMGN, GBP5, NSG1, and CXorf65, was constructed, and a multivariate Cox regression analysis revealed a significant association between the 10-gene signature-based risk model and overall survival (p < 0.001). A nomogram was constructed with age and the 10-gene signature. Consistent with the bioinformatics analysis, IHC and qRT-PCR confirmed the accuracy of the signatures in OC tissue samples. The predictive ability of the risk model was demonstrated using the Imvigor210 immunotherapy dataset.

Conclusions: The development of a novel gene signature associated with CD8+ T cells could facilitate more accurate prognostics and prediction of the immunotherapeutic response of patients with OC.

Keywords: biomarkers; immunotherapy; ovarian cancer; risk model; tumour microenvironment.

Figure 1

Relationship between TMB and molecular subtypes. (A) Distribution of TMB for subtype samples. (B) Distribution of the number of mutations for subtype samples. (C) Mutation features of significantly mutated genes in samples of each subtype. * p < 0.05. ns: no significance.

Figure 2

Differences in immune molecule expression and function between molecular subtypes in the RNA-seq cohort for (A) chemokines, (B) chemokine receptors, (C) IFNγ, (D) immune T-cell lysis activity, (E) angiogenesis scores, and (F) immune checkpoint genes. Significance was determined using an ANOVA, with * p < 0.05; ** p < 0.01, *** p < 0.001, and **** p < 0.0001. ns: no significance.

Figure 3

Immunological features and pathway characteristics of molecular subtypes. (A) Proportions of 22 immune cell types in subtype samples. (B) Differences in immune cell scores of 22 immune cell components between subtype samples. (C) Differences in enrichment scores of ten pathways associated with tumour abnormalities between subtypes. (D) Distribution of immune infiltration scores between subtype samples. (E,F) Comparison of the molecular subtypes with six previously identified pan-cancer immunophenotypes. * p < 0.05; ** p < 0.01, *** p < 0.001, and **** p < 0.0001. ns: no significance.

Figure 4

(A) TIDE scores in the RNA-seq dataset samples. (B) T cell dysfunction scores in the RNA-seq dataset samples. (C) T cell rejection scores in the RNA-seq dataset samples. (D) TIDE scores in the GSE-OV dataset samples. (E) T cell dysfunction scores in the GSE-OV dataset samples. (F) T cell rejection scores in the GSE-OV dataset samples. (G) RNA-seq submap analysis showing that IC1 may be insensitive to anti-PD-1 (Bonferroni corrected, p < 0.05). (H) The response of different immune clusters in the RNA-seq dataset to traditional chemotherapy drugs. (I) GSE submap analysis showing that IC1 may be insensitive to PD-1 inhibitors (Bonferroni corrected, p < 0.05). (J) The response of different immune clusters in the GSE-OV dataset to traditional chemotherapy drugs. ** p < 0.01, *** p < 0.001, and **** p < 0.0001. ns: no significance.

Figure 5

Nomogram and forest plot constructed with the RiskScore and clinical features using the TCGA dataset. (A) Nomogram. (B) Calibration curves showing the observed OS versus predicted probability of 1-, 3-, and 5-year survival of the nomogram. (C) Decision curve analysis plot. * p < 0.05, *** p < 0.001.

Figure 6

(A) qPCR and (B) IHC results showing low HEMGN and TXK and high SEMA4F and STX7 expression in OC tissues. * p < 0.05.