Scholarlink

<p>Early liver metastasis is a major factor contributing to the poor prognosis of pancreatic ductal adenocarcinoma (PDAC). Single-cell RNA sequencing (scRNA-seq) can analyze the heterogeneity between the primary tumor and metastatic lesions, but its wide clinical application is limited by costs, tissue requirements, and analytical complexity. In contrast, H&amp;E-stained sections are more commonly available. However, it is not clear whether the risk signals derived from images can truly reflect the biological characteristics related to metastasis. We integrated the single-cell RNA sequencing data (GSE154778) of primary and metastatic pancreatic ductal adenocarcinoma (PDAC) with TCGA transcriptome, clinical pathological, and H&amp;E image data. The copy number pattern based on InferCNV was used to distinguish malignant ductal cells with high copy numbers from ductal cells with low copy numbers. Differential expression and LASSO screening identified a transfer-related feature consisting of four genes (ARHGAP18, ASPH, EIF4EBP1, LY6D), and this feature was subsequently associated with image-derived features extracted through a dual-stream pathomics pipeline. The reproducibility of transcriptional levels in prognosis was evaluated in six independent GEO PDAC subgroups, and the locked pathological model was further tested on an external CPTAC subset using frozen cutoff values from TCGA. Pseudotime analysis suggested that a subset of metastatic malignant ductal cells occupied a progenitor-like transcriptional state. Cell–cell communication analysis indicated reduced antigen-presentation/prostaglandin-related signaling and relative enrichment of MIF- and laminin-associated pathways in metastases. The pathology model retained prognostic stratification in the internal TCGA validation split, although discrimination was lower than in training. Across six external GEO cohorts analyzed with cohort-specific optimal cutoffs, LY6D showed significant adverse survival associations in four cohorts, ARHGAP18 and ASPH in three cohorts each, and EIF4EBP1 in one cohort. In the external CPTAC cohort, the locked pathomics score also remained prognostic (HR 1.60, 95% CI 1.11–2.30; log-rank P = 0.011), with 12-, 24-, and 36-month time-dependent AUCs of 0.635, 0.617, and 0.639. This study presents an integrative genotype-to-phenotype workflow linking scRNA-seq-derived metastatic features to routine pathology images. External transcript-level validation and supportive CPTAC pathomics evaluation strengthen the findings, but larger independent validation studies and mechanistic experiments remain necessary before any clinical translation.</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 analysis and digital pathology for risk stratification in pancreatic cancer

  • 期刊论文

  • 发表日期：

    2026年5月26日

查看全文
  • Wenhao Bao,
  • Kaiming Leng,
  • Xiaodan Xia,
  • Jingyu Chang,
  • Pengchao Ying,
  • Shaohai Luan,
  • Qihua Yuan

摘要

Early liver metastasis is a major factor contributing to the poor prognosis of pancreatic ductal adenocarcinoma (PDAC). Single-cell RNA sequencing (scRNA-seq) can analyze the heterogeneity between the primary tumor and metastatic lesions, but its wide clinical application is limited by costs, tissue requirements, and analytical complexity. In contrast, H&E-stained sections are more commonly available. However, it is not clear whether the risk signals derived from images can truly reflect the biological characteristics related to metastasis. We integrated the single-cell RNA sequencing data (GSE154778) of primary and metastatic pancreatic ductal adenocarcinoma (PDAC) with TCGA transcriptome, clinical pathological, and H&E image data. The copy number pattern based on InferCNV was used to distinguish malignant ductal cells with high copy numbers from ductal cells with low copy numbers. Differential expression and LASSO screening identified a transfer-related feature consisting of four genes (ARHGAP18, ASPH, EIF4EBP1, LY6D), and this feature was subsequently associated with image-derived features extracted through a dual-stream pathomics pipeline. The reproducibility of transcriptional levels in prognosis was evaluated in six independent GEO PDAC subgroups, and the locked pathological model was further tested on an external CPTAC subset using frozen cutoff values from TCGA. Pseudotime analysis suggested that a subset of metastatic malignant ductal cells occupied a progenitor-like transcriptional state. Cell–cell communication analysis indicated reduced antigen-presentation/prostaglandin-related signaling and relative enrichment of MIF- and laminin-associated pathways in metastases. The pathology model retained prognostic stratification in the internal TCGA validation split, although discrimination was lower than in training. Across six external GEO cohorts analyzed with cohort-specific optimal cutoffs, LY6D showed significant adverse survival associations in four cohorts, ARHGAP18 and ASPH in three cohorts each, and EIF4EBP1 in one cohort. In the external CPTAC cohort, the locked pathomics score also remained prognostic (HR 1.60, 95% CI 1.11–2.30; log-rank P = 0.011), with 12-, 24-, and 36-month time-dependent AUCs of 0.635, 0.617, and 0.639. This study presents an integrative genotype-to-phenotype workflow linking scRNA-seq-derived metastatic features to routine pathology images. External transcript-level validation and supportive CPTAC pathomics evaluation strengthen the findings, but larger independent validation studies and mechanistic experiments remain necessary before any clinical translation.