Related Experiment Video
Updated: Sep 10, 2025
Author Spotlight: Investigating Immune Cell Dynamics in the Tumor Microenvironment — Challenges and Innovations in Cancer Prognosis
Published on: April 12, 2024
Pathomics predict CD40LG expression and clinical prognosis in lung adenocarcinoma based on haematoxylin-eosin-stained
Jianwei Shi1, Linchuan Liang2, Xiangzhi Meng2
1Department of Thoracic Surgery, National Cancer Centre/National Clinical Research Centre for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China; Nuffield Department of Primary Care Health Sciences, University of Oxford, Oxford OX2 6GG, United Kingdom.
Aim:
To develop and validate a pathomics model that non-invasively predicts CD40LG expression from routine haematoxylin-eosin (HE) slides and clarifies its prognostic value in lung adenocarcinoma (LUAD).
Methods:
HE whole-slide images from 327 TCGA-LUAD cases were randomly split into training (70 %) and internal-validation (30 %) sets; an external cohort of 89 patients from the Cancer Hospital Chinese Academy of Medical Sciences provided independent validation. From 1488 quantitative pathomic features, maximum-relevance minimum-redundancy and recursive feature elimination identified the most informative variables. A gradient-boosting machine (GBM) classified tumours as CD40LG-high or -low. Model performance was assessed with ROC curves, area under the curve (AUC), calibration plots and decision-curve analysis. Propensity-score matching (PSM) balanced baseline clinicopathologic factors in the TCGA cohort. Immune-cell deconvolution (CIBERSORTx) and gene-set variation analysis explored biological correlates of the pathomics score (PS).
Results:
After PSM, high CD40LG expression remained an independent protective factor for overall survival (OS) in the TCGA cohort (HR = 0.601, p = 0.006) and in the external cohort. The GBM model achieved AUCs of 0.809 (training), 0.736 (internal validation) and 0.725 (external validation). The derived PS independently predicted improved OS and correlated with greater infiltration of naïve B cells and CD8⁺ T cells. Genes linked to epithelial-to-mesenchymal transition-including IL32 and TGM2-were up-regulated in tumours with high PS.
Conclusions:
This pathomics model accurately infers CD40LG expression from standard histology and stratifies LUAD patients by prognosis, offering a practical, low-cost tool for precision oncology while providing insight into immune-mediated disease mechanisms.
