Identification of an immune classification for cervical cancer and integrative analysis of multiomics data

Abstract

Background: To understand the molecular mechanisms of the antitumour response, we analysed the immune landscape of cervical cancer to identify novel immune molecular classes.

Methods: We established a stable immune molecular classification using a nonnegative matrix factorization algorithm and validated the correlation in two validation sets of 249 samples.

Results: Approximately 78% of cervical cancers (CCs) (228/293) were identified to show significant enrichment in immune cells (e.g., CD8 T cells and macrophages), a type I IFN response, enhanced cytolytic activity and high PDCD1, and these CCs were referred to as the "immune class". We further identified two subtypes of the immune class: active immune and exhausted subtypes. Although the active immune subtype was characterized by enrichment of IFN signatures and better survival, the exhausted subtype expressed activated stroma, a wound healing signature, enhanced M2 macrophages and absence of CD8 T cells and the TGF-β response signature. Integrative analysis of multiomics data identified EGFR, JUN, MYC, FN1 and SERPINE1 as key modulators of the tumour immune microenvironment and potential targets for combination therapies which was validated in two validation sets.

Conclusions: Our study introduces a novel immune classification that might predict ideal candidates to receive immunotherapy or specific combination therapies.

Keywords: Cervical cancer; Immune classification; Multiomics.

Fig. 1

Identification of the immune class of CC and the molecular characterization of the subgroups. a NMF analysis of whole-tumour gene expression data using a molecular signature able to identify the immune class of CC. In the heatmap, high and low gene set enrichment scores are represented in red and blue, respectively; the same representation is used for high and low gene expression. b An integrated analysis of these immune molecular subgroups with the three published molecular classes. c The five modules of the immune subgroups are indicated by the heatmap. High and low scores are represented in red and blue, respectively. d Correlation of key immune characteristics with immune subgroups. e Infiltration of immune cells by immune subgroups. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 2

Validation of the immune class in independent publicly available datasets. Presence and molecular characteristics of the immune classes were successfully validated in 2 additional independent datasets. Results in validation set 1 (a) and validation set 2 (b) are here reported. High and low gene set enrichment scores are represented in red and blue, respectively

Fig. 3

Differences in the mutational landscape according to immune classes. a Correlation of DNA damage (rows) with immune subgroups. b Differential CNVs between the immune class and non-immune class. Circle: Differential CNV-associated genes in samples according to their chromosomal location. Genes that were gained are labelled in black, and genes that were deleted are labelled in blue. c Correlation of immunological parameters with mutational signatures. d Correlation between immune classes and MutSigs 2 and 13. e OncoPrint of the distribution of mutations in genes between patients of the immune class and non-immune class. f OncoPrint of the distribution of mutations in genes between patients of the exhausted class and active immune class

Fig. 4

Differences in lncRNA expression and protein expression according to immune classes. a The network summarizes complex connections between differentially expressed lncRNAs (pink dots), miRNAs targeted by lncRNAs (green dots), and DEGs (yellow dots, log2 FC > 1 & FDR < 0.05). b The network of protein–protein interactions according to the STRING database with highlighted key nodes and key pathways

Fig. 5

An integrative analysis of multiomics analyses and prognostic impacts according to immune classes. a Venn diagrams show different genes between the immune class and the non-immune class (left) or the exhausted class and the active immune class (right) affected by at least one of the indicated CNV, DEG, lncRNA, miRNA, protein or SNP events. The KEGG pathway network was constructed using the most significantly enriched pathways (p < 0.05) in the immune class (b), non-immune class (c), and exhausted and active immune subgroups (d). e, f The network of protein–protein interactions was constructed using the key genes in the KEGG pathway network in the immune class and non-immune class. g The Kaplan–Meier survival curves, including all the key nodes within the protein–protein interaction network, showed that 5 mRNAs were significant for predicting patient overall survival in the immune class, non-immune class and exhausted class. h 5 mRNAs expression was evident in 2 validation datasets. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 5

An integrative analysis of multiomics analyses and prognostic impacts according to immune classes. a Venn diagrams show different genes between the immune class and the non-immune class (left) or the exhausted class and the active immune class (right) affected by at least one of the indicated CNV, DEG, lncRNA, miRNA, protein or SNP events. The KEGG pathway network was constructed using the most significantly enriched pathways (p < 0.05) in the immune class (b), non-immune class (c), and exhausted and active immune subgroups (d). e, f The network of protein–protein interactions was constructed using the key genes in the KEGG pathway network in the immune class and non-immune class. g The Kaplan–Meier survival curves, including all the key nodes within the protein–protein interaction network, showed that 5 mRNAs were significant for predicting patient overall survival in the immune class, non-immune class and exhausted class. h 5 mRNAs expression was evident in 2 validation datasets. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 6

Prognosis and therapeutic strategies according to immune classes. a Kaplan–Meier plots of overall survival according to the immune and non-immune classes. b Kaplan–Meier plots of overall survival according to the active immune, exhausted, and non-immune classes. c Kaplan–Meier plots of overall survival according to the exhausted and non-exhausted classes before and after adjustment for risk factors. d The expression of PDCD1 and CTLA4 was significantly upregulated in the active immune class compared with the other classes. e Immunophenoscore (IPS) of patients under anti-CTLA4 treatment in the active immune, exhausted, and non-immune classes. f Chemosensitivity according to the active immune, exhausted, and non-immune classes. The drugs with red box are used for First- and second-line treatment in CCs. g The expression of EGFR was significantly upregulated in the exhausted class compared with the other classes. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 7

Characterization of the immune class in cervical cancer