Cancer-associated fibroblasts (CAFs) gene signatures predict outcomes in breast and prostate tumor patients

Over the last two decades, tumor-derived RNA expression signatures have been developed for the two most commonly diagnosed tumors worldwide, namely prostate and breast tumors, in order to improve both outcome prediction and treatment decision-making. In this context, molecular signatures gained by main components of the tumor microenvironment, such as cancer-associated fibroblasts (CAFs), have been explored as prognostic and therapeutic tools. Nevertheless, a deeper understanding of the significance of CAFs-related gene signatures in breast and prostate cancers still remains to be disclosed. RNA sequencing technology (RNA-seq) was employed to profile and compare the transcriptome of CAFs isolated from patients affected by breast and prostate tumors. The differentially expressed genes (DEGs) characterizing breast and prostate CAFs were intersected with data from public datasets derived from bulk RNA-seq profiles of breast and prostate tumor patients. Pathway enrichment analyses allowed us to appreciate the biological significance of the DEGs. K-means clustering was applied to construct CAFs-related gene signatures specific for breast and prostate cancer and to stratify independent cohorts of patients into high and low gene expression clusters. Kaplan-Meier survival curves and log-rank tests were employed to predict differences in the outcome parameters of the clusters of patients. Decision-tree analysis was used to validate the clustering results and boosting calculations were then employed to improve the results obtained by the decision-tree algorithm. Data obtained in breast CAFs allowed us to assess a signature that includes 8 genes (ITGA11, THBS1, FN1, EMP1, ITGA2, FYN, SPP1, and EMP2) belonging to pro-metastatic signaling routes, such as the focal adhesion pathway. Survival analyses indicated that the cluster of breast cancer patients showing a high expression of the aforementioned genes displays worse clinical outcomes. Next, we identified a prostate CAFs-related signature that includes 11 genes (IL13RA2, GDF7, IL33, CXCL1, TNFRSF19, CXCL6, LIFR, CXCL5, IL7, TSLP, and TNFSF15) associated with immune responses. A low expression of these genes was predictive of poor survival rates in prostate cancer patients. The results obtained were significantly validated through a two-step approach, based on unsupervised (clustering) and supervised (classification) learning techniques, showing a high prediction accuracy (≥ 90%) in independent RNA-seq cohorts. We identified a huge heterogeneity in the transcriptional profile of CAFs derived from breast and prostate tumors. Of note, the two novel CAFs-related gene signatures might be considered as reliable prognostic indicators and valuable biomarkers for a better management of breast and prostate cancer patients. Identification of CAFs-related gene signatures for breast and prostate cancer patients by integrating RNA-seq data from both primary CAFs and cancer patients datasets.

过去二十年间，为优化全球最常见的两类确诊肿瘤——前列腺癌与乳腺癌的预后预测及治疗决策，科研人员已开发出肿瘤来源的RNA表达特征（RNA expression signatures）。在此背景下，基于肿瘤微环境主要组分构建的癌相关成纤维细胞（cancer-associated fibroblasts, CAFs）分子特征，已被探索作为预后评估与治疗干预工具。然而，人们对乳腺癌与前列腺癌中CAFs相关基因特征的生物学意义仍缺乏深入认知，有待进一步阐明。
本研究采用RNA测序技术（RNA-seq）对乳腺癌与前列腺癌患者分离得到的CAFs转录组进行测序分析并开展比较。将乳腺癌与前列腺癌CAFs的差异表达基因（differentially expressed genes, DEGs）与公共数据集来源的批量RNA-seq（bulk RNA-seq）肿瘤患者转录组谱数据进行交集分析。通过通路富集分析（pathway enrichment analyses），明确了差异表达基因的生物学功能。采用K-means聚类（K-means clustering）构建乳腺癌与前列腺癌特异性CAFs相关基因特征，并将独立患者队列分层为高基因表达簇与低基因表达簇。借助Kaplan-Meier生存曲线（Kaplan-Meier survival curves）与log-rank检验（log-rank tests），预测不同患者簇的预后参数差异。采用决策树分析（decision-tree analysis）验证聚类结果，并通过提升算法（boosting calculations）优化决策树算法所得的分析结果。
针对乳腺癌CAFs的分析结果，我们鉴定出包含8个基因的特征集（ITGA11、THBS1、FN1、EMP1、ITGA2、FYN、SPP1及EMP2），这些基因参与促转移信号通路，如黏着斑通路（focal adhesion pathway）。生存分析显示，高表达上述基因的乳腺癌患者簇临床预后更差。
随后，我们鉴定出包含11个基因的前列腺癌CAFs相关特征集（IL13RA2、GDF7、IL33、CXCL1、TNFRSF19、CXCL6、LIFR、CXCL5、IL7、TSLP及TNFSF15），这些基因与免疫应答密切相关。低表达该特征集基因的前列腺癌患者生存率显著更低。
本研究结果通过两步法得到有效验证——基于无监督学习（unsupervised learning）与有监督学习（supervised learning）技术，在独立RNA-seq队列中展现出≥90%的高预测准确率。
本研究证实，乳腺癌与前列腺癌来源CAFs的转录组谱存在显著异质性。值得注意的是，这两种新型CAFs相关基因特征可作为可靠的预后指标与有价值的生物标志物，用于优化乳腺癌与前列腺癌患者的临床管理。
本研究通过整合原代CAFs与癌症患者数据集的RNA-seq数据，成功鉴定出适用于乳腺癌与前列腺癌患者的CAFs相关基因特征。