Immune response drives outcomes in prostate cancer: implications for immunotherapy

Abstract

The heterogeneity of the immune microenvironment leads to different responses in immune checkpoint blockade therapy. We aimed to propose a robust molecular classification system to investigate the relevance of the immune microenvironment subtype and prognosis of prostate cancer patients, as well as the therapeutic response to immune checkpoint blockade therapy. A total of 1,557 prostate cancer patients were enrolled, including 69 real-world samples from our institute (titled the AHMU-PC cohort). The non-negative matrix factorization algorithm was employed to virtually microdissect patients. The immune enrichment was characterized by a high enrichment of T cell-, B cell-, NK cell-, and macrophage-associated signatures, by which patients were subclassified into nonimmune and immune classes. Subsequently, the immune class was dichotomized into immune-activated and immune-suppressed subtypes based on the stromal signature, represented by the activation of WNT/TGF-β, TGF-β1, and C-ECM signatures. Approximately 14.9% to 24.3% of patients belonged to the immune-activated subtype, which was associated with favorable recurrence-free survival outcomes. In addition, patients in the immune-activated subtype were predicted to benefit more from anti-PD-1/PD-L1 therapy. In conclusion, our study identifies a novel immune molecular classifier that is closely related to clinical prognosis and provides novel insights into immunotherapeutic strategies for prostate cancer patients.

Keywords: immune checkpoint blockade therapy; immune molecular subclassification system; immunotherapy; non-negative matrix factorization; prostate cancer.

Fig. 1

Flow chart of the current study. A total of 1,557 prostate cancer patients were analyzed, and the immunophenotypes were established based on 495 patients from the TCGA‐PRAD cohort and validated in the GSE70770, GSE116918, GSE79021, MSKCC, and AHMU‐PC cohorts. TCGA‐PRAD, The Cancer Genome Atlas‐prostate adenocarcinoma; MSKCC, Memorial Sloan‐Kettering Cancer Center.

Fig. 2

Identification of the immune‐related clustering factor by non‐negative matrix factorization (NMF) analysis. (A) 11 clustering factors obtained from NMF analysis, with the second factor enriched the most patients with high immune enrichment scores. (B) The immune and nonimmune classes were adjusted by the multidimensional scaling (MDS) random forest analysis, via the expression matrix of the top 150 exemplar genes. (C) Heatmap showing the distribution of patients in different NMF factors, immune factor weight, exemplar genes‐based clustering, immune enrichment score, and final immune classes. (D) Gene Set Enrichment Analysis (GSEA) results showing the activated signaling pathways in the immune class.

Fig. 3

Identification of the immunophenotypes among the TCGA‐PRAD cohort, and comparing their differences at tumor‐infiltrating lymphocytes, copy number alterations, gene mutations, neoantigens, tumor stemness, and PD‐L1 expression levels. (A) Consensus‐clustered heatmap by the exemplar genes of NMF selected immune factor and refined by multidimensional scaling random forest to define the immune class (200/495, 40.4%, sky‐blue bar); nearest template prediction (NTP) using a signature capturing activated stroma identified immune‐suppressed (126/495, 25.5%; light‐green bar) and immune‐activated (74/495, 14.9%; red bar) classes; in the heat map, high and low single‐sample gene set enrichment scores are represented in red and blue, respectively. Positive prediction of activated stroma signature as per NTP is indicated in brown and its absence is in gray; (B) different recurrence‐free survival in three immunophenotypes among patients less or equal to 60 years old in TCGA‐PRAD cohort; (C) subclass mapping analysis manifested that patients with immune‐activated subtype were more likely to respond to anti‐PD‐1 treatment (Bonferroni‐corrected P‐value = 0.0079); (D) arm‐level copy number amplification and deletion; (E) focal‐level copy number amplification and deletion; (F) tumor mutant burden difference; (G) differentially mutated genes among three immune subgroups (some patients in nonimmune class without gene mutations hided); (H) neoantigens difference; (I) tumor‐infiltrating lymphocytes difference; (J) PD‐L1 expression difference; (K) tumor stemness difference represented by the mRNAsi. The comparison between two groups was conducted by Student’s t‐test. TCGA‐PRAD, The Cancer Genome Atlas‐prostate adenocarcinoma; CYT, cytolytic activity score; TITR, tumor‐infiltrating Tregs; MDSC, myeloid‐derived suppressor cell; TLS, tertiary lymphoid structure; C‐ECM, cancer‐associated extracellular matrix. t‐Test, Student’s t‐test, K‐W test, Kruskal–Wallis test.

Fig. 4

Successful validation of the immunophenotypes in the AHMU‐PC cohort. (A) Heatmap showing the different enrichment of characteristic signatures among immune‐activated, immune‐suppressed, and nonimmune groups; (B) Kaplan–Meier plot showing the recurrence‐free survival outcome in three immunophenotypes; (C) subclass mapping analysis manifested that patients with immune‐activated subtype were more likely to respond to anti‐PD‐1/PD‐L1 treatment (Bonferroni‐corrected P‐value = 0.0399); immunohistochemistry staining and quantification of CD163 (D) and α‐SMA (E) in prostate cancer patients with different immune status (nonimmune, immune‐suppressed, and immune‐activated classes) from AHMU‐PC cohort, Scale bar, 200 μm, 100 μm. t‐Test, Student’s t‐test, K‐W test, Kruskal–Wallis test.

Fig. 5

Immunophenotypes associated with the different recurrence‐free survival outcomes of prostate cancer patients. (A–C) Consensus‐clustered heatmap by the exemplar genes of NMF selected immune factor and refined by multidimensional scaling random forest to define the immune class; nearest template prediction (NTP) using a signature capturing activated stroma identified two distinct immune response subtypes: immune‐suppressed and immune‐activated classes; in the heat map, high and low single‐sample gene set enrichment scores are represented in red and blue, respectively. Positive prediction of activated stroma signature as per NTP is indicated in brown and its absence is in gray; (D) different recurrence‐free survival of three immunophenotypes in GSE70770 cohort; (E) different recurrence‐free survival of three immunophenotypes in GSE116918 cohort; (F) different recurrence‐free survival of three immunophenotypes in MSKCC cohort. MSKCC, Memorial Sloan‐Kettering Cancer Center. t‐Test, Student t‐test, K‐W test, Kruskal–Wallis test.