Liquid-liquid phase separation-driven molecular subtyping and prognostic modeling in colorectal cancer

Hui Guan, Chengzi Tian, Jiawen Lin, Lihuan Zhang, Run Shi, Wenjing Wang, Li Shu-ping, Yuan Sui, Yanwen Lu, Tianjiao Cui, Duo ChenFrontiers in Immunology 2026 · DOI

TL;DR

This study identifies a novel prognostic model for colorectal cancer (CRC) based on genes involved in liquid-liquid phase separation (LLPS). This LLPS-driven model effectively stratifies patients into high- and low-risk groups, offering potential for more personalized treatment strategies and a deeper understanding of CRC heterogeneity.

What Problem Does This Paper Address?

Colorectal cancer (CRC) remains a significant global health challenge, with a substantial proportion of patients experiencing recurrence and metastasis despite current therapies. Understanding the intricate molecular mechanisms driving CRC progression is crucial for developing more effective diagnostic and therapeutic strategies. Liquid-liquid phase separation (LLPS) is a fundamental cellular process that organizes biomolecules into membraneless organelles, playing critical roles in various biological functions. While LLPS dysregulation is implicated in other cancers, its specific role and potential as a biomarker in CRC have not been fully elucidated. This paper aims to bridge this gap by exploring the LLPS landscape in CRC and developing an LLPS-based prognostic model.

Methods

The researchers employed a multi-faceted approach to identify and validate an LLPS-driven prognostic signature for CRC.

  • Differential Gene Expression Analysis: They began by identifying differentially expressed genes (DEGs) between CRC tumor samples and normal tissues using the "limma" package on the GSE39582 dataset.
  • LLPS Gene Integration: These DEGs were then intersected with a curated list of LLPS-related genes from the DrLLPS database to identify LLPS-linked DEGs specific to CRC.
  • Prognostic Signature Construction: A three-step statistical approach was used to build the prognostic signature:
  • Univariate Cox Regression: Identified genes significantly associated with overall survival (OS).
  • LASSO Regression: Further filtered these genes to reduce dimensionality and prevent overfitting.
  • Stepwise AIC Algorithm: Refined the selection to identify the most prognostically significant genes.
  • Multivariate Cox Regression: Integrated the selected genes into a final prognostic model, generating a risk score for each patient.
  • Validation: The prognostic model was rigorously validated using multiple independent datasets (GSE17536, COAD, READ) to assess its generalizability.
  • In-depth Analysis: Beyond prognosis, the study explored various aspects:
  • Tumor Microenvironment (TME) Evaluation: Assessed immune cell infiltration, immune checkpoint expression, and stromal/immune scores using ssGSEA, CIBERSORT, and ESTIMATE.
  • Stemness Signature Analysis: Evaluated stemness-associated signatures using StemChecker.
  • Somatic Variant Analysis: Analyzed mutation frequencies and copy number variations (CNVs) in the identified genes.
  • Single-cell and Spatial Transcriptomics: Investigated the expression and spatial distribution of prognostic genes at a single-cell level and within tissue microregions.
  • Functional Enrichment Analysis: Utilized GO, KEGG, GSEA, and GSVA to understand the biological pathways associated with LLPS-linked DEGs and risk groups.
  • In Vitro Experiments: Conducted cell-based assays (viability, apoptosis, proliferation, migration, invasion) to experimentally verify the functional role of the key gene HADH.
  • Drug Sensitivity Screening: Predicted the sensitivity of CRC cells to small-molecule compounds based on gene expression and risk scores.

Datasets

The study utilized several publicly available datasets for training and validation:

  • GSE39582: Primary dataset for DEG identification and prognostic signature construction.
  • GSE17536: Validation cohort for the prognostic model.
  • COAD (The Cancer Genome Atlas): Validation cohort for the prognostic model and somatic variant analysis.
  • READ (The Cancer Genome Atlas): Validation cohort for the prognostic model and somatic variant analysis.
  • GSE166555: Single-cell RNA sequencing dataset for detailed cell-type-specific expression analysis.
  • DrLLPS database: Source for LLPS-related genes.
  • GDSC2 database: Used for predicting drug sensitivity.

Key Insights and Ideas

  • LLPS as a Driver of CRC Heterogeneity: The study strongly suggests that LLPS is not just a passive cellular process but actively contributes to the molecular heterogeneity observed in CRC, influencing distinct subtypes with different clinical outcomes.
  • A Novel Prognostic Paradigm: Moving beyond traditional molecular markers, this work introduces LLPS as a new lens through which to view CRC prognosis, offering a potentially more mechanistically grounded approach.
  • HADH as a Potential Tumor Suppressor: The experimental validation of HADH highlights its role in inhibiting CRC cell proliferation, migration, and invasion, suggesting it acts as a tumor suppressor and a potential therapeutic target. Its downregulation in tumor tissues further supports this.
  • TME Differences Dictate Prognosis: Significant disparities in immune cell infiltration and immune checkpoint expression between high- and low-risk groups underscore the critical role of the TME in dictating patient prognosis and response to immunotherapy.
  • Drug Sensitivity Stratification: The LLPS-based risk score can stratify patients based on their predicted sensitivity to certain small-molecule compounds, paving the way for more targeted drug selection.

Key Results

  • 430 LLPS-related DEGs identified: These genes showed significant differential expression between CRC and normal tissues.
  • Five-gene prognostic signature developed: This signature (HADH, AQP11, CCDC34, FBL, RAB15) demonstrated robust predictive performance across multiple independent cohorts (GSE39582, GSE17536, COAD, READ).
  • High-risk patients exhibit poorer OS: The LLPS-based risk score effectively stratified patients, with high-risk individuals showing significantly worse overall survival.
  • Nomogram model predicts survival: A nomogram integrating the risk score with clinical factors (age, M classification) showed strong predictive accuracy for 1-, 3-, and 5-year OS.
  • Distinct TME and mutation profiles: High-risk cohorts displayed altered immune cell infiltration, increased immune checkpoint expression, and different somatic mutation patterns compared to low-risk cohorts.
  • HADH overexpression inhibits CRC cell growth: In vitro experiments confirmed that HADH overexpression reduced cell viability, proliferation, migration, and invasion, while increasing apoptosis.

Limitations and Future Work

  • Retrospective Nature: The study primarily relies on retrospective analysis of public datasets, highlighting the need for prospective clinical cohort studies to further validate the findings.
  • Mechanistic Elucidation of HADH: While HADH's role is experimentally verified, the precise molecular mechanisms by which it influences CRC progression require further in-depth investigation.
  • Exploration of Other Prognostic Genes: The study focused on experimental validation of HADH. Future work should extend this to the other four prognostic genes (AQP11, CCDC34, FBL, RAB15) to fully understand their functional roles.
  • Broader LLPS Gene Investigation: While the study identified key LLPS-related genes, a more comprehensive exploration of the entire LLPS landscape in CRC could reveal additional insights.
  • Clinical Utility: Further research is needed to translate these findings into clinical practice, potentially through the development of diagnostic tools or targeted therapies based on the LLPS signature.

Summary updated

Abstract

An LLPS-related prognostic model was created, enabling the stratification of the CRC population according to risk and informing individualized therapeutic strategies.
BibTeX
@inproceedings{guan2026liquid,
  author = {Hui Guan and Chengzi Tian and Jiawen Lin and Lihuan Zhang and Run Shi and Wenjing Wang and LI Shu-ping and Yuan Sui and Yanwen Lu and Tianjiao Cui and Duo Chen},
  title = {{Liquid-liquid phase separation-driven molecular subtyping and prognostic modeling in colorectal cancer}},
  booktitle = {Frontiers in Immunology},
  year = {2026},
  doi = {10.3389/fimmu.2025.1741979},
}
Figure