Macrophages have received significant attention regarding the prognosis of several cancer types [33]. Particularly, the presence of TAMs infiltration in the tumor microenvironment was associated with disease progression following ADT, and preclinical studies also recommended that TAMs promote PCa cell proliferation and migration [34–36]. The heterogeneity of PCa is being revisited with the advent of single-cell technologies. Yu et al. discovered FMO2 as a biomarker of macrophage infiltration and prognosis in epithelial ovarian cancer [37]. Additionally, MS4A6A was found to be a new prognostic biomarker produced by macrophages in glioma patients [38]. However, the underlying mechanism of TAMs-induced remains unclear, and molecular stratification of TAMs based on predictive biomarkers to guide PCa treatment selection has not been implemented in the clinic yet. Therefore, broadly exploring the prognostic markers of PCa can guide future clinical management.
In the present study, we initially identified macrophage marker genes from PRAD tissue by scRNA-seq analysis. Further, we employed univariate Cox regression analyses and Lasso to screen 40 candidate genes that were highly correlated with the RFS of patients. We further established an integrative method to generate a consensus MRS with the expression profiles of these genes. One advantage of a complex machine learning algorithm is the capacity to develop better statistical models to forecast RFS across all cohorts to enhance classification performance. A total of nine models were fitted to the TCGA-PRAD database through the LOOCV framework. Subsequently, validation results in five independent cohorts obtained from the GEO dataset suggested that RSF was the best model with the capability of stratifying PCa patients into two MRS groups. Patients with higher MRS exhibited worse RFS in the training dataset. The survival curve in five external cohorts also confirmed the good reproducibility and robustness of the MRS in predicting patients’ RFS. Furthermore, the AUC value in all cohorts presented satisfying molecular subtyping accuracy. Previous studies indicated that various clinical predictors were widely used for prognostication and risk assessment of PCa, such as Gleason score, PSA, and stage [39–41]. Other parameters, such as TMB and MSI status, might potentially have an impact on therapy response and prognosis in addition to the clinicopathological features of the patient [42]. The C-index assessment suggested that the MRS signature has a significant advantage in predicting RFS relative to these factors. However, the difference showed no statistical significance in the MSKCC dataset. Notably, the decision curve, which synthesized the MRS with clinical characteristics, added a net benefit rate to current clinical values.
Furthermore, these DEGs within subgroups were also correlated with pathways involved in inflammation, such as the IL-17 signaling pathway, TNF signaling pathway, MAPK signaling pathway, PI3K-Akt signaling pathway, Cytokine-cytokine receptor interaction, peroxisome and cell cycle. Recent studies have shown that these variables and pathways are essential for metabolic regulation and macrophage polarization [43, 44]. Next, we discovered that PCa patients with high MRS were associated with an increased risk of biochemical recurrence and had a higher immune infiltration (like Tregs, activated CD8 T cells, activated NK cells, and M1 and M2 macrophages). These observations were in agreement with the previous studies [45, 46]. We observed that higher MRS scores correlated with higher expressions of immune checkpoints (CTLA4, HAVCR2, and CD86), which, in survival studies, was shown to have a mutually beneficial influence on patient prognosis. With the increase in MRS score, the expressions of checkpoints such as CXCL9 and CD80 also increased. This might be the result of the mechanism by which M1 hot TAMs may recruit T cells through CXCL9 expression [47]. As per recent studies, overexpression of HAVCR2 in T cells can lead to dysfunction of PSA-specific CD8 + T cells, further leading to a poor prognosis for PCa patients [48, 49]. Additionally, CD80 binds to CD28 or CTLA4 to activate T cell co-stimulation or initiate T cell co-inhibition, respectively [50]. Targeting immune checkpoints was considered to improve the therapeutic efficacy of immunotherapy modalities in solid tumors [51]. This study showed that patients with low MRS benefit more from ICB treatment, but those with high TIDE scores may react poorly to immunotherapy. Therefore, the identification of MRS classifiers that predict response to immunotherapies is critical for enhancing their use in treating patients. Therefore, the mechanism about how MRS affects RFS and ICB treatment of PCa patients may require more evidences and discussions.
The high MRS subgroup presented a higher mutation frequency of TP53 (31%), and the low MRS subgroup tended to have a higher proportion of SPOP mutations (12%). Studies have reported a mutually exclusive relationship between TP53 and SPOP mutations, both of which are independent prognostic markers for CRPC [52, 53]. For example, alterations in TP53 were associated with a decreased dependence of the tumor on AR signaling, which was associated with ARSI, whereas the highest levels of AR activity were found in the SPOP mutant subtype, indicating a good response to androgen receptor signaling inhibitor (ARSI) [54]. In addition to modifying CSF1 signaling, the combination of PTEN and p53 loss increased CXCL17 secretion, further influencing macrophage recruitment and function [55, 56]. ABCA13 was highly mutated in the high MRS group. It is reported that overexpression of ABCA13 is associated with decreased progression-free survival and reduced sensitivity to temozolomide in glioblastoma [57]. Patients with microsatellite instability-high (MSI-H) responded well to immunological response [58]. This study showed a better immune response in patients with low MRS, which is in line with both the findings from the TIDE approach. However, as confirmed by various studies, the high MRS group had a higher TMB score, suggesting that TMB is unlikely to be a major determinant of response to immunotherapy for this disease [59–61]. Considering the present and previous findings, it is concluded that the MRS score could be a credible biomarker that assists with filtering the dominant population of immunotherapy patients.
Neoadjuvant leuprolide enhances radiation-mediated apoptosis of PCa cells with well-documented effects in reducing tumor bulk and increasing progression-free survival [62]. According to this data, patients with low MRS were more inclined to respond to leuprolide-based ACT. ROC analysis suggested that MRS provided better accuracy in predicting leuprolide-based ACT. In addition, potential drug targets and corresponding compounds for high MRS PCa patients were screened as per the established MRS model, and several promising compounds were selected from three drug response databases, like paclitaxel, vorinostat, cabazitaxel, and fludarabine. It was discovered that Zn promotes the chemosensitivity of PCa cells to paclitaxel by inhibiting epithelial-mesenchymal transition and inducing apoptosis, which increases life expectancy in PCa patients [63]. Currently, Poly (ADP-ribose) polymerase inhibitors (vorinostat) can be used clinically to activate genotoxic and proteotoxic stress response pathways in human PCa [64]. Clinical trial data revealed that cabazitaxel enhanced survival with manageable side effects in patients with metastatic CRPC [65]. Furthermore, increased reactive oxygen species production by fludarabine phosphate may represent an effective treatment option for patients with N-MYC overexpressing NEPC tumors [66]. As a result, MRS might be a powerful tool for making choices for targeted drug development and immunotherapy combination for PCa patients.
Considering the impact of the tumor microenvironment on drug resistance, a differential analysis of critical genes in the external cohorts of PCa cells was performed. The GSE36135 and GSE33455 cohorts demonstrated high expression of ATF3 in docetaxel-resistant DU-145 cells compared to sensitive or original cells; however, the outcomes were opposite in the PC-3 cell line. Furthermore, GSE158494 showed that ATF3 expression was upregulated in cabazitaxel resistant cells compared to original and docetaxel resistant cells. As an inducer of oxidative stress and inflammation, ATF3 may regulate macrophage-associated host defense [67]. For example, in response to chemotherapy, wild-type macrophages exhibit pro-oncogenic activity, whereas ATF3 knockout macrophages exhibit anti-cancer activity [68]. ATF3 has been identified as a tumor suppressor for a major subset of PCa with dysfunctional PTEN and has also been shown to inhibit hormone-induced prostate carcinogenesis in mice [69, 70]. Our study further shows that the high expression of ATF3 was associated significantly with better prognosis and low MRS score. Therefore, ATF3 can be used as a key indicator of apoptosis and drug resistance in PCa cells. Although external cohorts were validated in this study, its predictive power needs further validation in prospective multicenter cohorts. Furthermore, given the complexity of the tumor microenvironment, which is influenced by multiple factors, the interaction between tumor cells and macrophage-associated genes requires more exploration and evidence. Potential drug combination interventions for MRS subgroups are expected to be further explored in clinical trials.