Department of Urology, Ruijin Hospital, School of Medicine, Shanghai Jiaotong University, Shanghai, China.
Department of Medical Oncology, Gaozhou People's Hospital, Gaozhou 525200, China.
Brief Bioinform. 2021 May 20;22(3). doi: 10.1093/bib/bbaa211.
Prostate cancer stemness (PCS) cells have been reported to drive tumor progression, recurrence and drug resistance. However, there is lacking systematical assessment of stemlike indices and associations with immunological properties in prostate adenocarcinoma (PRAD). We thus collected 7 PRAD cohorts with 1465 men and calculated the stemlike indices for each sample using one-class logistic regression machine learning algorithm. We selected the mRNAsi to quantify the stemlike indices that correlated significantly with prognosis and accordingly identified 21 PCS-related CpG loci and 13 pivotal signature. The 13-gene based PCS model possessed high predictive significance for progression-free survival (PFS) that was trained and validated in 7 independent cohorts. Meanwhile, we conducted consensus clustering and classified the total cohorts into 5 PCS clusters with distinct outcomes. Samples in PCScluster5 possessed the highest stemness fractions and suffered from the worst prognosis. Additionally, we implemented the CIBERSORT algorithm to infer the differential abundance across 5 PCS clusters. The activated immune cells (CD8+ T cell and dendritic cells) infiltrated significantly less in PCScluster5 than other clusters, supporting the negative regulations between stemlike indices and anticancer immunity. High mRNAsi was also found to be associated with up-regulation of immunosuppressive checkpoints, like PDL1. Lastly, we used the Connectivity Map (CMap) resource to screen potential compounds for targeting PRAD stemness, including the top hits of cell cycle inhibitor and FOXM1 inhibitor. Taken together, our study comprehensively evaluated the PRAD stemlike indices based on large cohorts and established a 13-gene based classifier for predicting prognosis or potential strategies for stemness treatment.
前列腺癌干细胞（PCS）已被报道可驱动肿瘤的进展、复发和耐药。然而，在前列腺腺癌（PRAD）中，缺乏对类干细胞指标的系统评估以及与免疫特性的关联。因此，我们收集了 7 个 PRAD 队列，共 1465 名男性，并使用单类逻辑回归机器学习算法计算每个样本的类干细胞指标。我们选择了 mRNAsi 来量化与预后显著相关的类干细胞指标，并据此确定了 21 个与 PCS 相关的 CpG 位点和 13 个关键特征。基于 13 个基因的 PCS 模型在 7 个独立队列中对无进展生存期（PFS）具有高预测意义。同时，我们进行了共识聚类，并将总队列分为 5 个具有不同结局的 PCS 聚类。PCScluster5 中的样本具有最高的干细胞分数，预后最差。此外，我们实施了 CIBERSORT 算法来推断 5 个 PCS 聚类之间的差异丰度。与其他聚类相比，PCScluster5 中激活的免疫细胞（CD8+T 细胞和树突状细胞）浸润显著减少，支持类干细胞指标与抗癌免疫之间的负调控。高 mRNAsi 也与免疫抑制检查点（如 PDL1）的上调有关。最后，我们使用 Connectivity Map（CMap）资源筛选针对 PRAD 干细胞的潜在化合物，包括细胞周期抑制剂和 FOXM1 抑制剂的首选靶点。总之，我们的研究基于大样本全面评估了 PRAD 的类干细胞指标，并建立了基于 13 个基因的分类器，用于预测预后或潜在的干细胞治疗策略。
