A cellular senescence-related gene prognostic index for biochemical recurrence and drug resistance in patients with prostate cancer

Abstract

In this study, we aimed to establish a novel cellular senescence-related gene prognostic index (CSG PI) to predict biochemical recurrence (BCR) and drug resistance in patients with prostate cancer (PCa) undergoing radical radiotherapy or prostatectomy. We performed all analyses using R version 3.6.3 and its suitable packages. Cytoscape 3.8.2 was used to establish a network of transcription factors and competing endogenous RNAs. Three cellular senescence-related genes were used to establish the CSGPI. We observed that CSGPI was an independent risk factor for BCR in PCa patients (HR: 2.62; 95% CI: 1.55-4.44), consistent with the results of external validation (HR: 1.88; 95% CI: 1.12-3.14). The CSGPI had a moderate diagnostic effect on drug resistance (AUC: 0.812, 95% CI: 0.586-1.000). The lncRNA PART1 was significantly associated with BCR (HR: 0.46; 95% CI: 0.27-0.77), and might modulate the mRNA expression of definitive genes through interactions with 57 miRNAs. Gene set enrichment analysis indicated that CSGPI was closely related to ECM receptor interaction, focal adhesion, TGF beta signaling pathway, pathway in cancer, regulation of actin cytoskeleton, and so on. Immune checkpoint analysis showed that PDCD1LG2 and CD96 were significantly higher in the BCR group compared to non-BCR group, and patients with higher expression of CD96 were more prone to BCR than their counterparts (HR: 1.79; 95% CI: 1.06-3.03). In addition, the CSGPI score was significantly associated with the mRNA expression of HAVCR2, CD96, and CD47. Analysis of mismatch repair and methyltransferase genes showed that DNMT3B was more highly expressed in the BCR group and that patients with higher expression of DNMT3B experienced a higher risk of BCR (HR: 2.08; 95% CI: 1.23-3.52). We observed that M1 macrophage, CD8+ T cells, stromal score, immune score, and ESTIMATE score were higher in the BCR group. In contrast, tumor purity was less scored in the BCR group. Spearman analysis revealed a positive relationship between CSGPI and M1 macrophages, CD4+ T cells, dendritic cells, stromal score, immune score, and ESTIMATE score. In conclusion, we found that the CSGPI might serve as a biomarker to predict BCR and drug resistance in PCa patients. Moreover, CD96 and DNMT3B might be potential treatment targets, and immune evasion might contribute to the BCR process of PCa.

Introduction

Prostate cancer (PCa) is the second most common cancer and the sixth leading cause of cancer death worldwide [1]. Since the introduction of prostate specific antigen (PSA), 81% of new cases have been localized, and radical radiotherapy (RRT) and prostatectomy are two preferred treatments for these patients [2,3]. Although the natural course of PCa is slow, among patients after radical therapy, the 15-year survival rate of those who suffer biochemical recurrence (BCR) within 3 years is 41% [4]. The impact of BCR on survival is believed to be limited to a subgroup of patients with specific clinical risk factors [5]. Nevertheless, BCR can promote the development of castration resistant prostate cancer (CRPC), and lead to an increased risk of long-term metastasis [5,6]. In this case, clinicians are more concerned about how to predict high-risk groups of BCR and avoid overtreatment at the same time.

Senescence is a stable cell cycle arrest that occurs in both primary cells and cancer cells [7,8]. Cell senescence may be a suboptimal response to anticancer therapies [9]. It is easy to understand that senescent cells can activate the immune system and promote the elimination of tumor cells, but it is worth noting that this activation is highly dependent on the tumor p53 status [10]. In prostate cancer, PTEN-deficient senescent tumors trigger highly immunosuppressive senescence-associated secretory phenotype (SASP) associated with increased infiltration of myeloid-derived suppressor cells [11]. Multiple studies have shown that senescent cells limit tumorigenesis and induce tumor progression, recurrence and metastasis of PCa at the late phase [12-15]. Furthermore, this two-sidedness of cellular senescence for cancer can be explained by the SASP of tumor cells, which refers to the model for explaining how senescent cells most likely promote senescence: the increased expression and secretion of inflammatory cytokines, chemokines, growth factors, and proteases [16]. However, the molecular and cellular mechanisms underlying cellular senescence and PCa are still poorly understood. In this study, we developed and validated a novel cellular senescence-related gene prognostic index (CSGPI) to predict biochemical recurrence (BCR) and drug resistance for patients with prostate cancer (PCa) undergoing radical radiotherapy or prostatectomy.

Methods

Data preparation

The training datasets were obtained from GSE46602 [17], GSE32571 [18], GSE62872 [19], and GSE116918 [20] after eliminating batch effects and the detailed process can be seen in our previous study [21]. We acquired the genes related to cellular senescence from GeneCards [22]. We used the TCGA database as the validation dataset. In addition, GSE42913 [23] was used to explore the diagnostic efficacy of CSGPI for drug resistance. Tumor related genes were considered as |r| ≥ 0.3 and p.adj. < 0.0001 in weighted gene coexpression network analysis, and differentially expressed genes were regarded as |logFC| ≥ 0.4 and p.adj. < 0.05.

Gene interaction, drug and cell line analysis

We analyzed the potential genes that might interact with definitive genes (ACACA, CTSB, and SERPINB5) using GeneMANIA [24]. We screened long noncoding RNAs (lncRNAs) associated with BCR-free survival and differentially expressed them between tumor and normal samples. Subsequently, we constructed a network of transcription factors (TFs) and competing endogenous RNAs (ceRNAs) using TRUST [25], lncBase [26], and miWalk [27]. We analyzed the drug sensitivity of definitive genes through GSCALite which included the data of the Cancer Therapeutics Response Portal (CTRP) [28], and the corporate cell lines of definitive genes was analyzed using canSAR [29].

Functional enrichment analysis

Gene Ontology (GO), including biological process (BP), cell composition (CC) and molecular function (MF), and Kyoto Encyclopedia of Genes and Genome (KEGG) analyses were conducted to explore the possible bioactivities and signaling pathways. We divided the 248 tumor patients undergoing RRT in GSE116918 [20] into high- and low-risk groups. We further conducted gene set enrichment analysis (GSEA) with “c2.cp.kegg.v7.4.symbols.gmt” and “h.all.v7.4.symbols.gmt” from the molecular signatures database [30]. We considered p.adj. < 0.05 and false discovery rate ≤ 0.25 were considered statistically significant.

Tumor immune microenvironment (TME) analysis

We explored the relationship between CSGPI and DNA mismatch repair (MMR) genes and methyltransferases using Spearman analysis [31]. The relationship between CSGPI and 20 common immune checkpoints was examined, as well as the differential expression between the BCR and non-BCR groups. We utilized the quanTiseq and ESTIMATE algorithms to score TME components [32-34]. Moreover, we conducted an analysis of differential expression, prognosis, and correlation for the above TME parameters and CSGPI score. Figure 1 shows an overview of the procedures in this study.

Figure 1.

The detailed flowchart in this study. WGCNA = weighted gene coexpression network analysis; GO = gene ontology; KEGG = Kyoto Encyclopedia of Genes and Genome; GSEA = gene set enrichment analysis; TF = transcription factor; CSGPI = cellular senescence-related gene prognostic index; mRNA = message RNA; long noncoding RNA = lncRNA.

Statistical analysis

We conducted all analyses using R software (version 3.6.3) and its suitable packages. Cytoscape 3.8.2 [35] was used to establish the TF-ceRNA network. We used the Wilcoxon test if the data did not satisfy a normal distribution. Variables were enrolled in the multivariate Cox regression analysis if the p value < 0.1 in the univariable Cox regression analysis. Statistical significance was set as two-sided P < 0.05. Significance was marked as follows: ns, P ≥ 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Results

CSGPI score and prognostic values

We eventually determined ACACA, CTSB, and SERPINB5 to be definitive genes after intersection of tumor-related genes, differentially expressed genes, and genes associated with BCR free survival, and Lasso and COX regression analysis (Figure 2A-F). These genes could discriminate high-risk patients from low-risk patients (Figure 2G). The risk score based on ACACA, CTSB, and SERPINB5 was CSGPI score =-0.88714*ACACA + 0.97560*CTSB - 1.79755*SERPINB5. The CSGPI was highly positively correlated with PSA (r: 0.652, P=0.029; Figure 2H). We observed moderate diagnostic accuracy of the CSGPI score distinguishing BCR from no BCR stably (AUCs were 0.766, 0.714, and 0.635 for 1 year, 2 years, and 3 years, respectively; Figure 2I). In GSE116918 [20], patients in the high-risk group had higher risk of BCR (HR: 2.62, 95% CI: 1.55-4.44, P=0.001) and metastasis (HR: 3.86, 95% CI: 1.67-8.92, p=0.004) than those in the low-risk group (Figure 2J, 2K). We observed similar results in the TCGA dataset for BCR-free survival (HR: 1.88, 95% CI: 1.12-3.14, P=0.018; Figure 2L) and diagnostic accuracy (AUCs were 0.613, 0.627, and 0.575 for 1 year, 2 years, and 3 years, respectively; Figure 2M). For drug resistance, the diagnostic accuracy was 0.812 (95% CI: 0.586-1.000; Figure 2N). The possible genes that interacted with ACACA, CTSB, and SERPINB5 included CCT5, CSTA, PPP6C, ACLY, IRF6, UNC93B1, TP63, ERCC6L, MLX, HLCS, ADAMTSL4, TRIM29, PROCR, FASN, USP29, IDH2, TLR8, TLR7, IDH1, and CTSZ (Figure 2O). We detected that patients expressing higher lncRNA PART1 had a lower risk of BCR than their counterpart (HR: 0.46, 95% CI: 0.27-0.77, P=0.004; not shown). We subsequently constructed the ceRNA network. PART1 might regulate the expression of ACACA, CTSB, and SERPINB5 through 57 common miRNAs (Figure 2P). In addition, TFs including SPDEF, E2F1, SP1, CREB1, RELA, and NFKB1, could activate the expression of ACACA, CTSB, and SERPINB5, while AR could repress the expression of SERPINB5 (Figure 2P). Overall, Figure 2 shows the process of screening definitive genes, clinical values of CSGPI scores, possibly interacting genes and the regulatory network of definitive genes.

Figure 2.

Process of screening definitive genes and clinical values. A. Volcano plot showing the mRNA expression of definitive genes between tumor and normal tissues; B. Modules and phenotype showing the tumor-related modules; C. Venn plot showing DEGs associated with tumor and cellular senescence; D. Gene screening through Lasso regression; E. Genes associated with BCR-free survival in PCa using univariate and multivariate COX analysis after Lasso regression; F. Examining the clinical values of CSGPI score using univariate and multivariate COX analysis for BCR free survival; G. Plot of risk factor showing the distribution of high- and low-risk patients; H. Correlation between CSGPI score and PSA; I. Time dependent ROC curve of CSGPI score discriminating BCR from no BCR; J. Kaplan-Meier curve showing survival differences between high- and low-risk patients for BCR free survival; K. Kaplan-Meier curve showing survival differences between high- and low-risk patients for metastasis free survival; L. External validation of CSGPI score through Kaplan-Meier curve showing survival differences between high- and low-risk patients for BCR free survival in TCGA dataset; M. Time dependent ROC curve of CSGPI score discriminating BCR from no BCR in TCGA dataset; N. ROC curve showing the diagnostic ability of CSGPI for drug chemoresistance; O. Protein-protein network of ACACA, CTSB, and SERPINB5; P. TF-ceRNA network of ACACA, CTSB, and SERPINB5. CSGPI = cellular senescence-related gene prognostic index; ROC = receiver operating characteristic; BCR = biochemical recurrence; PSA = prostate specific antigen; TF = transcription factor; ceRNA = competing endogenous RNA; PCa = prostate cancer.

Functional enrichment analysis

Figure 3 presents the GO functions of the candidate genes. BP analysis indicated that candidate genes were mainly involved in cell junction assembly and organization, reproductive structure and system development, gland development, and regulation of epithelial cell proliferation (Figure 3A). CC analysis showed that candidate genes were mainly involved in collagen-containing extracellular matrix (ECM), contractile fiber, I band, myofibril, sarcomere, Z disc, and focal adhesion (Figure 3B). MF analysis showed that candidate genes mainly participated in actin binding, extracellular matrix binding, cell adhesion mediator activity, structural constituent of cytoskeleton, cadherin binding involved in cell-cell adhesion, and integrin binding (Figure 3C). KEGG analysis indicated that candidate genes were mainly involved in focal adhesion, proteoglycans in cancer, glutathione metabolism, TGF-beta and Wnt and MAPK signaling pathways, platinum drug resistance, and pyruvate metabolism (Figure 3D). Figure 4 shows the GSEA results of high- and low-risk patients. GSEA showed that high-risk patients were enriched in ECM receptor interaction, focal adhesion, TGF-beta signaling pathway, regulation of actin cytoskeleton, NOD like receptor signaling pathway, FC gamma-R mediated phagocytosis, chemokine signaling pathway, apoptosis, complement and coagulation cascades, intestinal immune network for IGA production, lysosome, Wnt signaling pathway, adhesion junction, P53 signaling pathway, GAP junction, and cytokine-cytokine receptor interaction (Figure 4A).

Figure 3.

Gene ontology analysis of candidate genes. A. BP analysis; B. CC analysis; C. MF analysis; D. KEGG analysis; KEGG = Kyoto Encyclopedia of Genes and Genome; BP = biological process; CC = cell composition; MF = molecular function.

Figure 4.

GSEA analysis of high- and low-risk patients with prostate cancer. A. GSEA C2 analysis; B. GSEA hallmark analysis; GSEA = gene set enrichment analysis. Prostate cancer patients were divided into high- and low-risk groups according to the median value of the cellular senescence-related gene prognostic index.

TME, drug, and cell line analysis

We observed that PDCD1LG2 (P=0.038) and CD96 (p=0.013) were expressed at higher levels in the BCR group than in the non-BCR group (Figure 5A), and higher expression of CD96 was associated with a higher risk of BCR (HR: 1.79, 95% CI: 1.06-3.03, P=0.032; Figure 5B). Spearman analysis showed that the CSGPI score was significantly associated with the mRNA expression of HAVCR2 (r: 0.29, P < 0.001), CD96 (r: 0.15, P=0.016), CD47 (r: 0.16, P=0.012), and LAG3 (r: -0.15, P=0.018) (Figure 5C). DNMT3B was expressed at higher levels in the BCR group (Figure 5D) and was closely associated with BCR free survival (HR: 2.08, 95% CI: 1.23-3.52, P=0.008; Figure 5E). Moreover, we observed that M1 macrophages (P=0.022), CD8+ T cells (P=0.016), stromal score (P=0.003), immune score (P=0.012), and estimate score (P=0.003) had higher scores in the BCR group than in the non-BCR group (Figure 5F, 5G). However, the tumor purity showed a opposite difference (P=0.003, Figure 5G). We found that the CSGPI score was significantly related to M1 macrophages (r: 0.35), M2 macrophages (r: -0.31), neutrophils (r: -0.13), CD4+ T cells (r: 0.2), dendritic cells (r: 0.26), stromal score (r: 0.49), immune score (r: 0.45), estimate score (r: 0.49), and tumor purity (r: -0.49) (Figure 5H). For drug analysis, we found that 24 drugs might be sensitive to ACACA, CTSB, and SERPINB5 (Figure 5I), among which the top 10 drugs were 1S, 3R-RSL-3, CIL70, ML162, ML210, PI-103, PYR-41, UNC0638, bendamustine, manumycin A, and sunitinib (Figure 5J). The cell line analysis indicated that PRECLH, DU145, PC3, MDAPCA2B, 22RV1, NCIH660, and VCAP were potential cell lines to investigate CTSB, ACACA, and SERPINB5 in PCa (Figure 5K). Overall, Figure 5 shows the analyses of TME, MMR, methyltransferase genes, drug and potential cell lines in PCa patients.

Figure 5.

TME, drug, and cell line analysis. A. Comparison between BCR and no BCR group for immune checkpoints; B. Kaplan-Meier curve showing survival differences of high- and low-expression of CD96 for BCR free survival; C. Radar plot showing correlation between immune checkpoints and CSGPI score; D. Comparison between BCR and no BCR group for mismatch repair and methyltransferase genes; E. Kaplan-Meier curve showing survival differences of high- and low-expression of DNMT3B for BCR free survival; F. Comparison between BCR and no BCR group for TME cells; G. Comparison between BCR and no BCR group for TME score; H. Radar plot showing correlation TME parameters and CSGPI score; I. Venn plot showing common sensitive drugs of ACACA, CTSB, and SERPINB5 through the CTRP database; J. Plot showing the top 30 potential drugs for ACACA, CTSB, and SERPINB5 through the CTRP database; K. Venn plot showing common cell lines of ACACA, CTSB, and SERPINB5 in prostate cancer. TME = tumor immune microenvironment; CSGPI = cellular senescence-related gene prognostic index; BCR = biochemical recurrence; CTRP = the cancer therapeutics response portal.

Discussion

For low- and intermediate-risk localized prostate cancer, RRT is as effective as radical prostatectomy (RRP) [36]. However, for the high-risk subgroup, the risk of recurrence after RRT is increased [37]. After treatment and cure of PCa, some patients may have disease recurrence confirmed by PSA blood tests, namely BCR. However, not everyone who experiences BCR will develop a progressive disease [38]. Despite the development of the diagnosis and treatment of PCa during the past few decades, the survival rate of patients has often improved by meager months [39]. The development of models that predict BCR can help optimize decision-making strategies for PCa management.

Cellular senescence is a complex stress response, accompanied by a large number of changes in gene expression [40]. Senescence can be induced by cancer chemotherapy drugs and radiation, known as therapy-induced senescence (TIS) [9,41]. What needs to be clear is that tumor cells in vitro and in vivo have been found to escape from TIS, accompanied by a reduction in the expression of select SASP components [42,43]. In other words, senescent tumor cells can actually re-enter the cell cycle after senescence, and these cells acquire stem cell-like characteristics, which may represent a possibility of recurrence [44-46]. Demaria et al. found that senescent nontumor cells are conducive to cancer recurrence and metastasis after chemotherapy in a murine model [14]. Milanovic et al. found that senescence-associated reprogrammed cells, with stemness, were found to have much higher tumor initiation potential than virtually identical cells and is enriched in relapsed tumors, which may have a long-term impact on tumor aggressiveness and prognosis for leukemia [16]. The elimination of senescent cells after doxorubicin treatment can improve inflammation and tumor recurrence through cell-autonomous mechanisms as well as paracrine signaling through the SASP. SASP factors are involved in the recruitment of natural killer cells (NK) and macrophages and the “reprogramming” of macrophages to the tumor-inhibiting M1 phenotype [47,48].

Moreover, SASP has been shown to induce epithelial-mesenchymal transition (EMT), thereby increasing invasiveness [49-52]. However, for PCa, the presence of senescent cells only increased the proliferation of cocultured cells in vitro but did not significantly change tumor growth in vivo, which indicates negligible proliferative bystander effects of senescent PCa cells that depend on the expression of SASP components in the TME [47,53]. In fact, SASP is now divided into those derived from acute senescent cells (A-SASP) and chronic senescent cells (C-SASP), among which A-SASP is more effective in inducing the senescence of immortalized prostate cells [54,55]. Recent studies have shown that SASP induced senescence of immortalized prostate cells but not of metastatic PCa cells in vitro, suggesting that acute senescent cells only act to resist tumorigenesis rather than directly fight against malignant cells [56]. Borrowing this theory can partly explain how senescent cells mediate two opposite effects of tumor suppression and promotion [57,58]. Recently, Tonnessen-Murray et al. found that chemotherapy-induced senescent cancer cells often engulf adjacent senescent or nonsenescent tumor cells at a significant frequency to gain a survival advantage, leading to breast cancer recurrence and poor prognosis [59]. We speculated that a similar mechanism might exist for PCa cells after receiving RRT, since adherent senescent-like cells expressing common senescence-associated markers resulted in generation among several prostate cancer cell lines after ionizing radiation [60]. The mechanism may involve miR-106a, which can confer radiation resistance by reducing senescence [61].

The expression of lncRNA PART1 was found to be related to the poor prognosis and tumor recurrence of stage I-III non-small cell lung cancer and hepatocellular carcinoma [62-64]. SERPINB5, as a gene related to cancer cell motility, is believed to contribute to tumor invasion, migration and final metastasis [65]. Zhang et al. found that the ubiquitination of guanine monophosphate synthase (GMPS) mediated by SERPINB5 promotes TP53 inhibition, resulting in radiation resistance in nasopharyngeal carcinoma cells, which may help us understand that SERPINB5 may also be involved in the survival promotion and recurrence of PCa cells after RRT [66]. Here, we demonstrated for the first time the ability of the expression of lncRNA PART1 and its regulated mRNA to predict BCR and drug resistance after RRT in PCa. In addition, the CSGPI score based on in this study could predict BCR free survival for PCa patients undergoing RRP.

The senescence bystander effect mentioned above refers to the phenomenon that senescent cells cause the development of senescent phenotype in nearby cells [67]. This effect was found to be related to thrombospondin-1-dependent activation of the TGF-β1 signaling pathway through ROS and their downstream effector, p38 MAPK [68,69]. The TGF-β1 signaling pathway is related to premature senescence of human diploid fibroblasts (HDFs) [70,71]. Furthermore, inhibition of the TGF-β1 signaling pathway was found to prevent mouse primary prostate fibroblasts from radiation-induced damage, which means, from another perspective, radiation resistance and the subsequent recurrence of PCa [72]. This is consistent with the results of our gene set enrichment analysis. The possible speculation is that the oxidative stress and the subsequent cell senescence caused by the TGF beta signaling pathway led to the bystander effect of the tumor and its surrounding counterparts in PCa tissue thus promoting the occurrence of BCR. ECM receptor interaction and focal adhesion are two other mechanisms that may link cell senescence with BCR. The ECM is a component of the TME that affects the biological behavior of PCa and mediates cell differentiation, migration and invasion [73,74]. Lichner et al. found that miR-29c, miR-34a and miR-141 are differentially expressed in different Gleason grades, and their main biological processes include ECM-mediated signaling and focal adhesion kinase- and mitogen-activated kinase pathways, with miR-29c and miR-34a influencing downstream pathways that affect actin cytoskeleton organization [75]. Furthermore, miR-29c, miR-34a, miR-141 and miR-148a showed inverse correlations with BCR [75].

The high expression level of PDCD1LG2 has been found to be associated with a worse BCR free survival for PCa [76]. PDCD1LG2 was found to be related to immunomodulatory and radiation response pathways, suggesting its role in predicting prognosis and response to treatment as a promising immune checkpoint target [76]. In our study, we found higher expression of PDCD1LG2 in the BCR group, but an association with BCR-free survival was not observed. The DNA methyltransferase DNMT3B is highly abundant in several prostate cancer cell lines. By targeting RAD9 to methylate, it regulates tumorigenicity, castration resistance, androgen-independent growth and metastasis of PCa [77,78]. Moreover, DNMT3B mRNA expression is associated with an increased cancer aggressiveness and risk of lethal PCa [79]. Combined with our research, it is also suggested that DNA methylation may be related to the occurrence of BCR.

The TME is a collection of tumor cells and quiet nontumor cells [80]. A large number of studies have shown that the TME is involved in tumor progression and response to treatment by nourishing the tumor parenchyma [81,82]. In the TME, macrophages play an important regulatory function [83]. Classically activated macrophages in TME (M1 macrophages, CD14++ CD16-) show antitumor activity, while alternatively activated macrophages (M2 macrophages, CD14+ CD16++) possess anti-inflammatory functions and promote wound healing, angiogenesis and tissue remodeling, thereby supporting tumor progression and metastasis [84]. The polarization of macrophages in particular depends on SASP in the TME [85]. As mentioned above, p53-dependent senescent hepatic stellate cells (HSCs) release specific SASP factors including IFN-γ and IL-6, which bias the polarization of macrophages in favor of the M1 state, while proliferating p53-deficient HSCs promote the conversion of macrophages to the M2 phenotype through IL-4 [10]. Di Mitri et al. found that PTEN-null PCa tissue, which is vulnerable to TNF-α-induced senescence, was strongly infiltrated by macrophages and promoted the polarization of macrophages to the TNF-α-secreting M1 phenotype through CXCR2 [86]. Senescence establishes the antitumor TME through SASP, which regulates the function of macrophages, and inhibits the tumorigenesis of neighboring cells in a noncell-autonomous manner [48]. Most tumor-associated macrophages (TAMs) have the M1 phenotype, and TAMs can induce senescence and tumor inhibition [86,87]. Combined with the results of our analysis, it is further confirmed that senescent PCa cells, such as HSCs, can promote the transformation of macrophages from the M2 to the antitumor M1 phenotype through SASP. Here we suggest the therapeutic effect of macrophage-targeting therapy in PCa, such as the application of an α-CSF-1R monoclonal antibody for colorectal adenocarcinoma and fibrosarcoma [88].

Senescent cells undergo immune surveillance from T cells through adaptive immunity as well. Various SASP factors including CCL27, CCL2, CXCL11 and IL-1α are related to the mobilization, activation and differentiation of T cells [89,90]. The potential recruitment of activated T lymphocyte subsets to sites occupied by senescent cells has been witnessed [49,90-92]. The activation, differentiation and functional specialization of T cells are finely regulated [93]. The CD4+ T-cell response as a Th1 type response, rather than direct T cell cytotoxicity, effectively kills precancerous senescent hepatocytes [92]. The percentage of stromal cells in the TME represents the stromal score [94]. In all solid tumors, abundant matrix often represents a worse prognosis, with deeper invasion depth and lymph node metastasis probability [95]. The prostate stroma is an important component for normal prostate growth and differentiation, compared with PCa, where the increase in collagen fibers and carcinoma-associated fibroblasts (CAFs) accompanied a decrease in smooth muscle cells as cancer progresses [96]. This change in stroma is similar to wound healing and is called ‘reactive stroma’ [97-99]. Reactive matrix grading (RSG) has been a tool to assess PCa-specific mortality in diagnostic prostate needle biopsies [100]. In univariate analysis, level 3 RSG can predict the time and risk of biochemical recurrence after radical prostatectomy [101]. Moreover, CD96 was expressed on T cells and NK cells together with CD226 and TIGIT, contributing to tumor escape from the immune system [102]. Thus, we proposed that immune evasion played a vital role in the BCR process of PCa.

Here, we identified genes and pathways related to cellular senescence and BCR in PCa, and built a CSGPI risk score to predict BCR and drug resistance. Behind this score, we also revealed that the occurrence of BCR may be related to various immune cells and SASP in the TME. In summary, senescence is a specific response of cancer cells to antitumor treatments including RRT. Combined with research results in other cancers, we speculated that senescent cells might play a major role in promoting BCR after RRT in the short term through the reversible process of TIS and SASP in the TME. However, for the subsequent treatment of prostate cancer, senescence induction remains a potential treatment method through activation of the immune system. Indeed, we have to admit that most of the findings, such as the ceRNA network and the potential targets identified in this study, warrant further investigation.

Conclusion

We found that the CSGPI might serve as a biomarker to predict BCR and drug resistance in PCa patients. Moreover, CD96 and DNMT3B might be potential treatment targets, and immune evasion might contribute to the BCR process of PCa.

Disclosure of conflict of interest

None.