A myeloid IFN gamma response gene signature correlates with cancer prognosis

Abstract

Background: The IFN-γ cytokine plays a dual role in anti-tumor immunity, enhancing immune defense against cancer cells while promoting tumor survival and progression. Its influence on prognosis and therapeutic responses across cancer types remains unclear.

Objective: This study aimed to perform a pan-cancer analysis of IFN-γ response genes to determine their prognostic significance and evaluate their impact on clinical outcomes and anti-PD1 immunotherapy responses.

Methods: Using multiple datasets, 46 IFN-γ response genes were identified as prognostic for disease-specific survival, and their expression was used to construct the IFN-γ Response Gene Network Signature (IFGRNS) score. The prognostic and therapeutic relevance of the IFGRNS score was assessed across cancer types, considering tumor pathology, genomic alterations, tumor mutation burden, and microenvironment. Single-cell transcriptomic analysis identified cellular contributors, and a murine pancreatic cancer (PAN02) model was used to validate findings with anti-PD1 therapy.

Results: The IFGRNS score emerged as a robust prognostic indicator of survival, with higher scores correlating with worse outcomes in most cancer types. The prognostic significance of the score was influenced by factors such as cancer type, tumor pathology, and the tumor microenvironment. Single-cell analysis revealed that myeloid cells, particularly the M2 macrophage subtype, demonstrated high levels of IFGRNS expression, which was associated with tumor progression. A negative correlation was observed between the IFGRNS score and outcomes to anti-PD1 immunotherapy in urologic cancers, where patients with higher scores showed worse prognosis and lower response rates to therapy. Experimental validation in the PAN02 murine model confirmed that anti-PD1 therapy significantly reduced tumor size and IFGRNS expression in M2 macrophages, supporting the clinical findings.

Conclusions: The IFGRNS score is a novel prognostic indicator for survival and therapeutic responses in cancer. These findings underline the complexity of IFN-γ signaling and suggest potential applications for the IFGRNS score in cancer diagnosis, prognosis, and immunotherapy. Novelty & impact statements: IFN-γ response genes play a significant role in tumour biology, yet comprehensive analysis across various cancers is limited. This study identifies a novel prognostic biomarker, the IFGRNS score, which is elevated in myeloid lineage cells and correlates with survival across multiple cancers. The IFGRNS score is also associated with tumour pathology, immune microenvironment, and immunotherapy response, highlighting its diagnostic and therapeutic potential in cancer management.

Key points: IFN-γ cytokine plays a dual role in cancer, aiding immune defense but also promoting tumor progression. A novel IFGRNS score, based on 46 IFN-γ response genes, was identified as a strong prognostic marker for survival across cancer types. Higher IFGRNS scores correlate with worse prognosis and reduced response to anti-PD1 immunotherapy, particularly in urologic cancers. M2 macrophages were identified as key contributors to high IFGRNS scores, associated with tumor progression. Findings were validated in a murine cancer model, highlighting the potential of the IFGRNS score for cancer prognosis and therapy guidance.

Keywords: IFN‐γ response genes; LASSO; anti‐PD1 immunotherapy; pan‐cancer; prognosis; tumour microenvironment.

FIGURE 1

Construction of the IFN‐γ response gene‐based model that is prognostic for disease‐specific survival. (A) 70 genes were selected using LASSO Cox regression analysis for disease‐specific survival. The coefficient profiles of those genes were shown. (B) A forest plot showing the point estimates and confidence intervals of the HRs for each of those 46 genes derived from the result of the multivariate Cox regression analysis. The squares indicate the HRs and the whiskers indicate the confidence intervals of HRs. (C, D) Kaplan−Meier survival curves of patients with high IFGRNS scores and low IFGRNS scores, in the train dataset (C) and in the test dataset (D). (E) Hazard ratio (HR) and 95% confidence interval (CI) for disease‐specific survival (DSS) of the IFGRNS scores for the training sets (red) and testing sets (black) in the cohorts of TCGA data. The patients in each dataset were divided into high‐score and low‐score groups by the median of the scores. Hazard ratios were calculated with COX regression (high‐score group vs. low‐score group).

FIGURE 2

The prognostic value of the IFGRNS score in different types of cancer in external datasets. Kaplan−Meier survival curves of patients with high and low IFGRNS scores in the external validated datasets, including patients with glioma, breast cancer, pancreatic ductal adenocarcinoma, lung cancer, rectal cancer, and colorectal cancer. (A) Overall survival (OS) of patients stratified by the IFGRNS score. (B) Tumour relapse‐free survival (RFS) of patients stratified by the IFGRNS score. (C) Disease‐free survival (DFS) of patients stratified by the IFGRNS score. The patients in each dataset were divided into high‐score and low‐score groups based on the optimal cutoff value of the IFGRNS score, which was determined by the ‘cutp’ function in the ‘survMisc’ package for R. The size of population (percent of patients) for high‐score and low‐score groups are labelled in the graph. The optimal cutoff value and the adjusted p value for each dataset are shown above the graph. The adjusted p‐value was calculated by correcting for multiple testing across different cutoff values using the method of Contal and O'Quigley. NSCLC, non‐small‐cell lung cancer.

FIGURE 3

The correlations between the IFGRNS score and clinical characteristics in each cancer type. (A) The violin plot showing the IFGRNS scores across the cancer types. (B) A heatmap showing the correlations of IFGRNS score to the clinical characteristics and genomic alterations in each cancer type. The top panel consists of the categorical features and the middle panel consists of the numeric features. The bottom panel consists of the correlation with prognosis (OS, DSS, and PFI) in training and testing datasets, respectively. The size of the dot indicates the p‐values. The filled colour of the dot indicates the correlation coefficient or hazard ratio for survival. The white colour of the background indicates the significance (p < .05) of the correlation.

FIGURE 4

Construction of a comprehensive model that combined the IFGRNS score and clinical characteristics. (A) A forest plot showing the point estimates and confidence intervals of the HRs for each of risk factors derived from the result of the multivariate Cox regression analysis. The squares indicate the HRs and the whiskers indicate the confidence intervals of HRs. (B) A nomogram including IFGRNS score, person neoplasm cancer status, fraction genome altered, cancer type, pathology stage, age, and Buffa Hypoxia score, for 1‐, 3‐, 5‐ and 10‐year disease‐specific survival. (C) Decision curve analysis for the 3, 5, or 10‐year DSS outcomes, respectively. Black line: All patients had the events. Gray line: None patients had the events. Magenta line: the integrated model of nomogram. (D) Comparison of sensitivity and specificity between the integrated model and the individual models in external datasets. The concordance index (C‐index) is a measure of how well the model discriminates between patients with different survival outcomes. It is similar to the area under the ROC curve (AUC), but it can handle censored data. A higher C‐index means that the model can more accurately rank the survival times according to the risk scores.

FIGURE 5

The correlations of the IFGRNS score to the signatures of the tumour immune microenvironment. (A) The design of the lung cancer mouse model with an intranasal administration of AD‐Cre induced KRAS G12D mutation, workflow diagram illustrating the single‐cell sequencing process, from sample collection to data analysis, and images of collected tissue samples, showcasing the distinct histopathological features identified through immunohistochemistry staining. (B, C) 2D uniform manifold approximation and projection visualization of all CD45⁺ (left) and CD45⁻ (right) cells across lung or tumour tissue, coloured according to main cell type (B) and subtypes (C). (D) A heatmap showing the expression levels of IFGRNS across various cell subtypes in the lung cancer mouse model, comparing samples across the different stages. Each row represents a cell subtype, with expression values normalized and scaled for clear comparison. Adjacent to the heatmap, a bar summarizes the aggregate IFGRNS expression for each cell subtype across all samples.

FIGURE 6

The correlation of IFGRNS score to the anti‐PD1 therapy resistance. (A) A heatmap showing the correlations of IFGRNS score to the Ayers IFN‐γ signature, Mariathasan pan‐fibroblast TGF‐β signature, Chakravarthy TGF‐β‐associated ECM, Hugo IPRES, Riaz nivolumab responsive signature, Joseph ISG resistance signature (ISG.RS), and Peng TIDE signature in the TCGA cohorts. The size of the dot indicates the p values. The filled colour of the dot indicates the correlation coefficient. The white colour of the background indicates the significance (P < .05) of the correlation. (B) Boxplots showing the differences in IFGRNS scores across tumour regression response in the clinical trial cohorts with the anti‐PD1 therapy. The upper, middle, and lower hinges of the box plot are 75th, 50th, and 25th quartiles, and the whiskers extend to the range below and above, respectively. CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease. (C) Hazard ratio and 95% confidence interval (CI) for progression‐free survival of the signature scores in the cohorts with the anti‐PD1 therapy. (D) Hazard ratio and 95% confidence interval (CI) for overall survival of the signature scores in the cohorts with the anti‐PD1 therapy. (E, F) Comparison of C‐index for IFGRNS and other ICI‐Responsive signatures in predicting progression‐free survival (E) and overall survival (F) for patients treated with anti‐PD‐1 therapy. mUC, metastatic urothelial cancer; HCC, hepatocellular carcinoma; Atezo, atezolizumab; Pembro, pembrolizumab. Opdivo is also the nivolumab, and Keytruda is also the pembrolizumab.

FIGURE 7

Single‐cell transcriptome sequencing of murine pancreatic cancer model treated with anti‐PD‐1. (A) The design of the pancreatic cancer mouse model treated with anti‐PD‐1, sample collection and single‐cell sequencing. (B) The effect of anti‐PD‐1 therapy on the growth of the PAN02 allografts in C57 mice. Data shown are mean + SD. (C, D) 2D uniform manifold approximation and projection visualization of all cells across both PBMCs and tumour tissue, coloured according to main cell type (C) and subtypes (D). (E) A heatmap showing the expression levels of IFGRNS across various cell subtypes in the pancreatic cancer mouse model, comparing samples treated with anti‐PD1 to untreated controls. Each column represents a cell subtype, with expression values normalized and scaled for clear comparison. Adjacent to the heatmap (top), a bar summarizes the aggregate IFGRNS expression for each cell subtype across all samples.

FIGURE 8

Functional and pathway analysis of IFN‐γ response genes in different subsets. (A) Gene ontology (GO) enrichment analysis of the genes involved in the IFGRNS, Ayers IFN‐γ signature, and other remaining IFN‐γ response genes. (B) Pathway (KEGG and REACTOME terms) enrichment of the same gene subsets. (C) The gene interaction network of the IFGRNS genes belonging to the GO term of negative regulation of cytokine production. (D) The gene interaction network of the IFGRNS genes belonging to the KEGG pathway of natural killer cell‐mediated cytotoxicity is displayed as a graph. The nodes represent genes and the edges represent interactions. Gene sets of GO and pathways were obtained from the GSEA database. Gene interaction relationships were obtained from the STRING database.