Scholarlink

Background <p>Pancreatic adenocarcinoma (PAAD) has an extremely poor prognosis, and existing prognostic markers fail to fully capture the complex heterogeneity of the tumor microenvironment. This study aimed to integrate ligand–receptor (L–R) interactions, multi-omics data, and deep learning-based pathological images to construct an interpretable multimodal prognostic model and to elucidate the mechanisms underlying the cancer-associated fibroblast (CAF) microenvironment.</p> Methods <p>Significant L–R interactions were identified using <i>BulkSignalR</i>, followed by sequential Cox, least absolute shrinkage and selection operator (LASSO)–Cox, and random survival forest analyses to construct a prognostic model. Multi-omics profiling characterized molecular distinctions between risk groups. Key L–R pairs were evaluated with single-cell and spatial transcriptomics, validated in 39 paired clinical specimens via immunofluorescence, and linked to histopathological features through deep learning on hematoxylin and eosin-stained whole-slide images.</p> Results <p>We identified 236 significant L–R pairs, with 47 associated with prognosis. Integration of LASSO–Cox and random survival forest analyses yielded five key pairs: <i>IL16_KCND1</i>,<i> PLAU_ITGA5</i>,<i> FN1_ITGB3</i>,<i> GNAS_ADCY1</i>, and <i>CALM1_PDE1B</i>. The resulting risk model effectively stratified overall survival. The high-risk group showed higher tumor mutational burden, more frequent <i>KRAS</i> and <i>TP53</i> mutations, and enrichment of extracellular matrix remodeling, transforming growth factor‑<i>β</i> signaling, and glycolysis pathways. Single-cell and spatial analyses revealed preferential enrichment of <i>PLAU_ITGA5</i> and <i>FN1_ITGB3</i> in fibroblast-related compartments. Immunofluorescence confirmed upregulation of these pairs in tumor tissues, and deep learning identified fibroblast-associated histopathological features with strong concordance to the risk axes.</p> Conclusions <p>This study established the first multimodal prognostic framework integrating L–R interactions and histopathological features, revealing the central role of CAF-mediated L–R signaling in remodeling the PAAD microenvironment and providing a novel strategy for precise prognostic stratification and targeted microenvironmental therapy.</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

Improving pancreatic adenocarcinoma prognosis models through ligand receptor interactions and histopathological integration

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月3日

查看全文
  • Kejun Liu,
  • Zhenyao Tan,
  • Yezhen Tang,
  • Yongxue Lv,
  • Yang Bu

摘要

Background

Pancreatic adenocarcinoma (PAAD) has an extremely poor prognosis, and existing prognostic markers fail to fully capture the complex heterogeneity of the tumor microenvironment. This study aimed to integrate ligand–receptor (L–R) interactions, multi-omics data, and deep learning-based pathological images to construct an interpretable multimodal prognostic model and to elucidate the mechanisms underlying the cancer-associated fibroblast (CAF) microenvironment.

Methods

Significant L–R interactions were identified using BulkSignalR, followed by sequential Cox, least absolute shrinkage and selection operator (LASSO)–Cox, and random survival forest analyses to construct a prognostic model. Multi-omics profiling characterized molecular distinctions between risk groups. Key L–R pairs were evaluated with single-cell and spatial transcriptomics, validated in 39 paired clinical specimens via immunofluorescence, and linked to histopathological features through deep learning on hematoxylin and eosin-stained whole-slide images.

Results

We identified 236 significant L–R pairs, with 47 associated with prognosis. Integration of LASSO–Cox and random survival forest analyses yielded five key pairs: IL16_KCND1, PLAU_ITGA5, FN1_ITGB3, GNAS_ADCY1, and CALM1_PDE1B. The resulting risk model effectively stratified overall survival. The high-risk group showed higher tumor mutational burden, more frequent KRAS and TP53 mutations, and enrichment of extracellular matrix remodeling, transforming growth factor‑β signaling, and glycolysis pathways. Single-cell and spatial analyses revealed preferential enrichment of PLAU_ITGA5 and FN1_ITGB3 in fibroblast-related compartments. Immunofluorescence confirmed upregulation of these pairs in tumor tissues, and deep learning identified fibroblast-associated histopathological features with strong concordance to the risk axes.

Conclusions

This study established the first multimodal prognostic framework integrating L–R interactions and histopathological features, revealing the central role of CAF-mediated L–R signaling in remodeling the PAAD microenvironment and providing a novel strategy for precise prognostic stratification and targeted microenvironmental therapy.