Associations between TMPRSS2 expression and immune signatures in LUAD
We found that TMPRSS2 had a significant negative expression correlation with the infiltration levels of CD8 + T cells, which represent the adaptive antitumor immune response, in three of the five LUAD cohorts (Spearman correlation, p < 0.05) (Fig. 1A). TMPRSS2 expression levels were also significantly and negatively correlated with the infiltration levels of NK cells, which represent the innate antitumor immune response, in two LUAD cohorts (p < 0.05) (Fig. 1A). Moreover, TMPRSS2 expression levels were negatively correlated with immune cytolytic activity, a marker for underlying immunity [18], in all the five LUAD cohorts. Meanwhile, TMPRSS2 had a significant negative expression correlation with PD-L1 in the five LUAD cohorts (Fig. 1A). TMPRSS2 expression levels were negatively correlated with the infiltration levels of CD4 + regulatory T cells and MDSCs in four LUAD cohorts, which represent tumor immunosuppressive signatures (Fig. 1A). Taken together, these results suggest a significant negative association between TMPRSS2 abundance and immune infiltration levels in LUAD. Interestingly, TMPRSS2 expression levels showed a significant positive correlation with the ratios of immune-stimulatory/immune-inhibitory signatures (CD8 + T cells/PD-L1) consistently in the five LUAD cohorts (Pearson correlation, p < 0.05) (Fig. 1B). It indicated that TMPRSS2 levels had a stronger negative correlation with immune-inhibitory signatures than with immune-stimulatory signatures. Furthermore, we found that the ratios of immune-stimulatory/immune-inhibitory signatures were positively correlated with DFS in the TCGA-LUAD cohort (log-rank test, p = 0.01) (Fig. 1C).
Associations between TMPRSS2 expression and oncogenic pathways, tumor phenotypes and prognosis in LUAD
We found that TMPRSS2 expression levels were inversely correlated with the activities of the cell cycle, mismatch repair, and p53 signaling pathways in the five LUAD cohorts (Spearman correlation, p < 0.001) (Fig. 2A). Moreover, TMPRSS2 showed a negative expression correlation with MKI67, a tumor proliferation marker, in the five LUAD cohorts (Pearson correlation, p < 0.001) (Fig. 2B). Tumor stemness indicates a stem cell-like tumor phenotype representing an unfavorable prognosis in cancer [19]. We observed that TMPRSS2 expression levels were inversely correlated with tumor stemness scores in these LUAD cohorts (Spearman correlation, p < 0.001) (Fig. 2C).
We detected that TMPRSS2 expression levels significantly decreased with tumor advancement in LUAD (Fig. 2D). For example, in the TCGA-LUAD cohort, TMPRSS2 expression levels were significantly lower in late-stage (Stage III-IV) than in early-stage (Stage I-II) LUADs (Student’s t test, p < 0.001; fold change (FC) = 1.6), in large-size (T3-4) than in small-size (T1-2) LUADs (p = 0.007; FC = 1.5), in LUADs with lymph nodes (N1-3) than in those without regional lymph nodes (N0) (p = 0.02; FC = 1.3), and in LUADs with metastasis (M1) than in those without metastasis (M0) (p = 0.07; FC = 1.6). In other two LUAD cohorts (GSE30219 and GSE50081) with tumor size and lymph nodes data available, TMPRSS2 expression levels were also significantly lower in large-size than in small-size LUADs (p < 0.001; FC = 6.4) in GSE30219 and were significantly lower in N1-3 than in N0 LUADs in both GSE30219 (p = 0.02; FC = 2.83) and GSE50081 (p = 0.02; FC = 1.6) (Fig. 2D). Furthermore, the lung cancer data from Jiangsu Cancer Hospital supported that TMPRSS2 expression levels were reduced in late-stage (Stage IV) than in early-stage (Stage I-II) LUADs (p < 0.001; FC = 1.6) (Fig. 2E). Survival analyses showed that TMPRSS2 downregulation was correlated with worse OS and/or DFS in these LUAD cohorts (log-rank test, p < 0.05) (Fig. 2F).
It has been shown that EGFR-mutated LUADs have a better prognosis than EGFR-wildtype LUADs [20]. We found that TMPRSS2 was more lowly expressed in EGFR-wildtype than in EGFR-mutated LUADs (p = 0.006; FC = 1.5) (Fig. 2G). Besides, LUAD harbors three transcriptional subtypes: terminal respiratory unit (TRU), proximal-inflammatory (PI), and proximal-proliferative (PP), of which TRU has the best prognosis [21]. We found that TMPRSS2 expression levels were the highest in TRU (TRU versus PP: p = 8.68 × 10− 14, FC = 2.98; TRU versus PI: p = 1.07 × 10− 11, FC = 3.16) (Fig. 2G).
Taken together, these results suggest that TMPRSS2 downregulation is associated with worse outcomes in LUAD.
Association between TMPRSS2 expression and genomic instability in LUAD
Genomic instability plays prominent roles in cancer initiation, progression, and immune invasion [22] by increasing TMB [23] and aneuploidy or somatic copy number alterations [24]. In the TCGA-LUAD cohort, TMPRSS2 expression levels had a negative correlation with TMB (Spearman correlation, ρ = -0.31; p = 2.58 × 10–12) (Fig. 3A). Homologous recombination deficiency (HRD) may promote chromosomal instability and aneuploidy levels in cancer [25]. We found that TMPRSS2 expression levels were inversely correlated with HRD scores [25] in LUAD (ρ = -0.27; p = 5.76 × 10–10) (Fig. 3B). DNA damage repair (DDR) deficiency can lead to genomic instability [26]. Knijnenburg et al. [25] identified deleterious gene mutations for nine DDR pathways in TCGA cancers. We divided LUAD into pathway-wildtype and pathway-mutated subtypes for each of the nine DDR pathways. The pathway-wildtype indicates no deleterious mutations in any pathway genes, and the pathway-mutated indicates at least a deleterious mutation in pathway genes. Interestingly, we found that TMPRSS2 expression levels were significantly lower in the pathway-mutated subtype than in the pathway-wildtype subtype for seven DDR pathways (p < 0.05; FC > 1.5) (Fig. 3C). The seven pathways included base excision repair, Fanconi anemia, homologous recombination, mismatch repair, nucleotide excision repair, translesion DNA synthesis, and damage sensor. These results suggest a correlation between TMPRSS2 downregulation and DDR deficiency.
TP53 mutations often leads to genomic instability because of the important role of p53 in maintaining genomic stability [27]. We found that TMPRSS2 displayed significantly lower expression levels in TP53-mutated than in TP53-wildtype LUADs (p = 0.006; FC = 1.5) (Fig. 3D). Moreover, we found numerous DDR-associated genes having significant negative expression correlations with TMPRSS2 in these LUAD cohorts (Pearson correlation, p < 0.05), including MSH2, MSH6, POLE, PCNA, and RAD51 (Fig. 3E). Furthermore, we observed significant negative expression correlations between TMPRSS2 and DNA mismatch repair proteins MSH6 (Pearson correlation, r = -0.30; p = 6.6 × 10− 9) and PCNA (r = -0.25; p = 1.5 × 10− 6) in the TCGA-LUAD cohort (Fig. 3F). These results indicated an association between TMPRSS2 downregulation and the upregulation of DDR molecules, the signature of increased genomic instability.
Genomic instability can promote tumor heterogeneity, which is associated with tumor progression, immune evasion, and drug resistance [28]. We used the DEPTH algorithm [29] to score ITH for each TCGA-LUAD sample and found a significant negative correlation between TMPRSS2 expression levels and ITH scores in LUAD (ρ = -0.55; p < 0.001) (Fig. 3G). It indicates a significant association between TMPRSS2 downregulation and increased ITH in LUAD.
Taken together, these results suggest that TMPRSS2 downregulation is associated with enhanced genomic instability in LUAD.
Co-expression networks of TMPRSS2 in LUAD
We found 150 and 135 genes having strong positive and negative expression correlations with TMPRSS2 in the TCGA-LUAD cohort, respectively (Pearson correlation, |r| > 0.5) (Fig. 4A; Supplementary Table S3). GSEA [14] revealed that the cell cycle, p53 signaling, mismatch repair, and homologous recombination pathways were significantly associated with the 135 genes with strong negative expression correlations with TMPRSS2. This conforms to the previous findings that TMPRSS2 downregulation was correlated with increased activities of these pathways.
WGCNA [17] identified six gene modules (indicated in blue, turquoise, brown, magenta, purple, and pink color, respectively) highly enriched in the high-TMPRSS2-expression-level LUADs. The representative GO terms associated with these modules included cell projection, chromosome segregation, response to endogenous stimulus, cell adhesion, cellular response to lipopolysaccharide, and micro-ribonucleoprotein complex. In contrast, three gene modules (indicated in green, black, and green-yellow color, respectively) were highly enriched in the low-TMPRSS2-expression-level LUADs (Fig. 4B). The representative GO terms for these modules included extracellular matrix (ECM), small molecule metabolic process, and postsynapse (Fig. 4B). The ECM signature plays a crucial role in driving cancer progression [30]. Its upregulation in the low-TMPRSS2-expression-level LUADs is in accordance with the correlation between TMPRSS2 downregulation and LUAD progression.
Experimental validation of the bioinformatics findings
To validate the findings from the bioinformatics analysis, we performed in vitro experiments with the human LUAD cell line A549 and in vivo experiments with mouse tumor models. We found that TMPRSS2 knockdown markedly promoted proliferation and invasion potential in A549 cells (Fig. 5A) and increased tumor volume and progression in Lewis tumor mouse models (Fig. 5B). This is consistent with the previous results showing that TMPRSS2 downregulation is associated with tumor progression and unfavorable prognosis in LUAD. Furthermore, in vitro experiments showed that MSH6 expression was upregulated in TMPRSS2-knockdown versus TMPRSS2-wildtype A549 cells (Fig. 5C). This is in agreement with the previous finding of the significant negative correlation between TMPRSS2 expression levels and MSH6 abundance in LUAD.
Our bioinformatics analysis revealed a significant inverse correlation between TMPRSS2 abundance and immune infiltration levels in LUAD. Consistently, the MHC class I genes (HLA-A, HLA-B, and HLA-C) showed significantly higher expression levels in TMPRSS2-knockdown than in TMPRSS2-wildtype A549 cells, demonstrated by real-time qPCR (Fig. 5D). NK cells co-cultured with TMPRSS2-knockdown A549 cells displayed significantly stronger proliferation ability than NK cells co-cultured with TMPRSS2-wildtype A549 cells, evident by the EdU proliferation assay (Fig. 5E). Furthermore, in vivo experiments showed that infiltration of CD8 + T cells and NK cells significantly increased in TMPRSS2-knockdown tumors (Fig. 5F). Moreover, on CD8 + T cells from TILs in TMPRSS2-knockdown tumors, the expression of TNF-α and IFN-γ were significantly upregulated (Fig. 5G, H), indicating that TMPRSS2 knockdown can enhance the activity of CD8 + TILs. Meanwhile, the expression of PD-1 and LAG3 also significantly increased on CD8 + TILs in TMPRSS2-knockdown tumors (Fig. 5I, J), indicating that TMPRSS2 deficiency can also promote the exhaustion of CD8 + TILs.
Our bioinformatics analysis revealed a significant negative correlation between TMPRSS2 and PD-L1 expression levels. This result was confirmed by both in vitro and in vivo experiments; knockdown of TMPRSS2 increased PD-L1 expression in A549 cells, as evidenced by Western blotting (Fig. 5C); TMPRSS2-knockdown tumors had significantly enhanced PD-L1 expression (Fig. 5F). Furthermore, bioinformatics analysis revealed a significant positive correlation between TMPRSS2 expression levels and the ratios of CD8 + T cells/PD-L1. This was confirmed by that TMPRSS2-knockdown tumors displayed a higher level of increases in CD8 + T cell infiltration than in PD-L1 abundance (Fig. 5F). Because PD-L1 expression is a predictive biomarker of response to immune checkpoint inhibitors (ICIs) in cancer [31], we anticipated that knockdown of TMPRSS2 would promote the response to ICIs in LUAD. As expected, the volume of the TMPRSS2-knockdown tumors had a significantly higher level of decreases than that of TMPRSS2-wildtype tumors after treatment with BMS-1, an inhibitor of PD-1/PD-L1 (Fig. 5K); this result supports that knockdown of TMPRSS2 can enhance the sensitivity of LUAD to the PD-1/PD-L1 inhibitor. Furthermore, the activities of CD8 + TILs and NK TILs markedly increased in TMPRSS2-knockdown tumors after treatment with BMS-1; they were significantly higher in TMPRSS2-knockdown than in TMPRSS2-wildtype tumors after treatment with BMS-1 (Fig. 5L, M). These results support that the PD-1/PD-L1 inhibitor promotes immune elimination of tumor cells by inhibiting the exhaustion of CD8 + TILs and NK TILs in TMPRSS2-depleted LUAD.
To summarize, bioinformatics analysis revealed a negative correlation between TMPRSS2 abundance and immune infiltration levels in LUAD. Experimental results demonstrated that this relationship was a causal relationship. That is, reduced TMPRSS2 abundance can boost immune infiltration for LUAD.