Scholarlink

Background <p>Cancer-associated fibroblasts (CAFs) are key players in the tumor microenvironment (TME), but their roles in prognosis and immunotherapy response in mismatch repair-deficient endometrial cancer (dMMR EC) remain unclear. This study used single-cell RNA sequencing (scRNA-seq) and bulk RNA sequencing to identify Wnt-related CAF subclusters and applied multi-algorithm machine learning to build a risk signature for predicting clinical outcomes and immunotherapy response.</p> Methods <p>We obtained scRNA-seq data for dMMR EC and bulk RNA-seq data from public databases. The Seurat R package was used to define Wnt-related CAFs. We identified tumor and normal cells through copy number variation (CNV) and screened differentially expressed genes with the limma package, determining their correlation with CAF clusters via Pearson correlation analysis. Prognostic genes related to CAFs were first filtered using univariate Cox regression, followed by a machine learning framework comprising ten algorithms and 101 combinations, evaluated by the concordance index (C-index) to identify the optimal gene signature. A nomogram model was developed, combining clinicopathological features and risk score. We conducted SNV mutation risk analysis, immune landscape analysis, and validated the response to immune checkpoint modules.</p> Results <p>By using scRNA-seq data, we identified six CAF clusters in dMMR EC, five of which were related to prognosis. We distinguished 9,682 tumor cells and 5,476 normal cells using copykat methods. Gene Set Variation Analysis (GSVA) revealed significantly higher tumor-related pathway scores in the tumor group, with 1,329 upregulated and 2,726 downregulated DEGs, among which 1,319 were significantly related to prognosis. Based on the highest C-index and minimal gene number, the CoxBoost + Enet [alpha = 0.6] model was selected to construct the prognostic signature. Six significant genes were identified: HAPLN1, CIT, CDK16 (risk gene), RARRES2, LRRN4CL, and LTB (protective gene). Ten pathways were significantly associated with these genes, including cell cycle, focal adhesion, and vascular smooth muscle contraction. Kaplan-Meier analysis showed that high-risk patients had poorer survival. Protective genes correlated positively with immune infiltration, while risk genes showed a significant negative correlation. Mutation analysis found that LTB was positively correlated with Aneuploidy Score, whereas LRRN4CL was negatively correlated with the Number of Segments. The nomogram model, incorporating clinical characteristics (Stage, Age) and risk genes, identified the risk score as an independent prognostic factor for EC. TimeROC analysis highlighted the nomogram’s superior predictive performance, and survival analysis confirmed the Risk Score’s response to immune checkpoint modules.</p> Conclusion <p>By systematically integrating 101 machine learning models derived from 10 algorithms, we constructed a Wnt-CAF-driven risk signature. Combined with clinicopathological features in a nomogram, this model effectively predicted prognosis and potential immunotherapy response in dMMR EC. These findings deepen the understanding of the EC microenvironment and provide valuable guidance for personalized treatment strategies and prognostic assessment in clinical practice.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating single-cell and bulk transcriptomes with machine learning reveals a CAF signature for immunotherapy response in dMMR endometrial cancer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月14日

查看全文
  • Lu Zhang,
  • Quan Zhang,
  • Mengjie Yang,
  • Mengyao Xu,
  • Qiyuan Li,
  • Qionghua Chen

摘要

Background

Cancer-associated fibroblasts (CAFs) are key players in the tumor microenvironment (TME), but their roles in prognosis and immunotherapy response in mismatch repair-deficient endometrial cancer (dMMR EC) remain unclear. This study used single-cell RNA sequencing (scRNA-seq) and bulk RNA sequencing to identify Wnt-related CAF subclusters and applied multi-algorithm machine learning to build a risk signature for predicting clinical outcomes and immunotherapy response.

Methods

We obtained scRNA-seq data for dMMR EC and bulk RNA-seq data from public databases. The Seurat R package was used to define Wnt-related CAFs. We identified tumor and normal cells through copy number variation (CNV) and screened differentially expressed genes with the limma package, determining their correlation with CAF clusters via Pearson correlation analysis. Prognostic genes related to CAFs were first filtered using univariate Cox regression, followed by a machine learning framework comprising ten algorithms and 101 combinations, evaluated by the concordance index (C-index) to identify the optimal gene signature. A nomogram model was developed, combining clinicopathological features and risk score. We conducted SNV mutation risk analysis, immune landscape analysis, and validated the response to immune checkpoint modules.

Results

By using scRNA-seq data, we identified six CAF clusters in dMMR EC, five of which were related to prognosis. We distinguished 9,682 tumor cells and 5,476 normal cells using copykat methods. Gene Set Variation Analysis (GSVA) revealed significantly higher tumor-related pathway scores in the tumor group, with 1,329 upregulated and 2,726 downregulated DEGs, among which 1,319 were significantly related to prognosis. Based on the highest C-index and minimal gene number, the CoxBoost + Enet [alpha = 0.6] model was selected to construct the prognostic signature. Six significant genes were identified: HAPLN1, CIT, CDK16 (risk gene), RARRES2, LRRN4CL, and LTB (protective gene). Ten pathways were significantly associated with these genes, including cell cycle, focal adhesion, and vascular smooth muscle contraction. Kaplan-Meier analysis showed that high-risk patients had poorer survival. Protective genes correlated positively with immune infiltration, while risk genes showed a significant negative correlation. Mutation analysis found that LTB was positively correlated with Aneuploidy Score, whereas LRRN4CL was negatively correlated with the Number of Segments. The nomogram model, incorporating clinical characteristics (Stage, Age) and risk genes, identified the risk score as an independent prognostic factor for EC. TimeROC analysis highlighted the nomogram’s superior predictive performance, and survival analysis confirmed the Risk Score’s response to immune checkpoint modules.

Conclusion

By systematically integrating 101 machine learning models derived from 10 algorithms, we constructed a Wnt-CAF-driven risk signature. Combined with clinicopathological features in a nomogram, this model effectively predicted prognosis and potential immunotherapy response in dMMR EC. These findings deepen the understanding of the EC microenvironment and provide valuable guidance for personalized treatment strategies and prognostic assessment in clinical practice.