Pathology

CTHRC1+ CAFs identify poor-prognosis, immunotherapy-resistant colorectal cancers across MSI and MSS

Across >3000 colorectal cancers, a CAF subset expressing CTHRC1 was strongly associated with increased TGF-β signaling and worse outcomes. These CTHRC1+ CAFs stratified both dMMR/MSI and pMMR/MSS tumors into immune-inflamed versus poorly immunogenic phenotypes. Retrospective analyses of multiple immunotherapy trials linked CTHRC1+ CAF–rich tumors with resistance to immune checkpoint inhibitors in MSI and MSS disease. The authors propose assessing CTHRC1+ CAFs as a stroma-based biomarker within routine histopathology workflows to guide immunotherapy and TGF-β–targeted combinations. Prospective validation will be required before using CTHRC1 status for treatment selection in stroma-rich, treatment-resistant colorectal cancers. 1

Deep learning on WSIs predicts tumor primary site, addressing cancer of unknown primary workups

This study presents a deep learning framework that predicts primary tumor site directly from cytology and histology whole-slide images. The model was trained and evaluated on three datasets, including 1,196 histopathology WSIs from 69 primary sites and separate cytology cohorts. It achieved high accuracy and outperformed four state-of-the-art methods for identifying common metastatic origins. The approach is fast and low-cost, suggesting potential utility as a decision-support tool in cancer-of-unknown-primary evaluations and metastatic workups. 2

Few-shot prompt-tuned foundation models improve rare cancer subtyping on WSIs

PathPT uses prompt-tuning of pathology vision-language foundation models to improve rare cancer subtyping with limited training data. The framework converts slide-level labels into tile-level supervision, improving localization of cancerous regions and cross-modal reasoning. Across 11 datasets covering 56 subtypes and 3958 WSIs, PathPT consistently outperformed state-of-the-art methods in data-scarce settings. It improved both subtyping accuracy and grounding of relevant regions, offering a scalable tool where subspecialty expertise for rare entities is limited. 3

Baseline methylation markers may predict CIN2 lesions unlikely to regress under surveillance

This epigenome-wide association study analyzed serial liquid-based cytology samples from 58 young women with CIN2 managed by active surveillance. Differential methylation at baseline distinguished lesions that regressed from those that persisted or progressed at 24 months. Higher methylation at cg12754953 (ALDH9A1) and increasing cg13556949 (TULP2) methylation over time were linked to non-regression, while lower cg18887759 (MED25) methylation predicted imminent regression. The authors propose that targeted methylation assays could guide decisions between continued surveillance and excision in CIN2, pending validation in larger cohorts. 4