Development and validation of a cancer-associated fibroblast-derived lncRNA signature for predicting clinical outcomes in colorectal cancer

Abstract

Cancer-associated fibroblasts (CAFs) are actively involved in cancer progression through generating extracellular matrix and orchestrating the crosstalk within the tumor microenvironment (TME). This study aimed to develop and validate a CAF-derived lncRNA (long non-coding RNA) (CAFDL) signature for predicting clinical outcomes in colorectal cancer (CRC). Clinical data and transcriptomic profiles of 2,320 patients with CRC from The Cancer Genome Atlas (TCGA)-COAD and TCGA-READ datasets and 16 Gene Expression Omnibus datasets were included in this study. CAFDLs were identified using weighted gene co-expression network analysis. The CAFDL signature was constructed using the least absolute shrinkage and selection operator analysis in the TCGA-CRC training set. Multiple CRC cohorts and pan-cancer cohorts were used to validated the CAFDL signature. Patients with high CAFDL scores had significantly worse overall survival and disease-free survival than patients with low CAFDL scores in all CRC cohorts. In addition, non-responders to fluorouracil, leucovorin, and oxaliplatin (FOLFOX)/fluorouracil, leucovorin, and irinotecan (FOLFIRI) chemotherapy, chemoradiotherapy, bevacizumab, and immune checkpoint inhibitors had significantly higher CAFDL scores compared with responders. Pan-cancer analysis showed that CAFDL had prognostic predictive power in multiple cancers such as lung adenocarcinoma, breast invasive carcinoma, stomach adenocarcinoma, and thyroid carcinoma. The CAFDL signature was positively correlated with transforming growth factor-beta (TGF-β) signaling, epithelial-mesenchymal transition, and angiogenesis pathways but negatively correlated with the expression of immune checkpoints such as PDCD1, CD274, and CTLA4. The CAFDL signature reflects CAF properties from a lncRNA perspective and effectively predicts clinical outcomes in CRC and across pan-cancer. The CAFDL signature can serve as a useful tool for risk stratification and provide new insights into the underlying mechanisms of CAFs in cancer immunity.

Keywords: cancer-associated fibroblasts; colorectal cancer; long non-coding RNAs; pan-cancer; prognosis; signature.

Figure 1

Flow chart of this study.

Figure 2

WGCNA identified CAFDL and LASSO analysis. (A) WGCNA identified modules associated with CAF infiltration calculated by EPIC, MCPcounter, and xCell in TCGA-CRC and meta-GEO cohorts. (B) Correlation between gene significance for CAF infiltration and module membership in turquoise module in TCGA-CRC cohort. (C) Correlation between gene significance for CAF infiltration and module membership in green module in meta-GEO cohort. (D) A Venn diagram showing the number of lncRNAs in the turquoise module in the TCGA-CRC cohort and the green module in the meta-GEO cohort. (E) LASSO analysis identifies 21 CAF-derived lncRNAs. (F) Multivariate Cox analysis calculated the coefficient for each lncRNA in the CAFDL signature. (G) Expression of 21 CAF-derived lncRNAs in CRC and normal tissues. **P < 0.01, ***P < 0.001, NS non-significant.

Figure 3

CAFDL signature can effectively predict the prognosis of patients with CRC. (A) Patients with high CAFDL scores have significantly worse overall survival than those with low CAFDL scores in TCGA-CRC, TCGA-COAD, TCGA-READ, GSE17536, GSE17537, GSE29621, GSE38832, GSE39582, GSE72970, and total CRC cohorts. (B) Patients with high CAFDL scores have significantly worse disease-free survival than those with low CAFDL scores in TCGA-COAD, TCGA-READ, GSE17536, GSE17537, GSE29621, GSE31595, GSE33113, GSE37892, GSE38832, GSE39582, GSE92921, and GSE143985 cohorts.

Figure 4

CAFDL signature can effectively predict the response to mainstay treatments of CRC. (A–E) Non-responders to FOLFOX (A, B) and FOLFIRI (C) chemotherapy, chemoradiotherapy (D), and bevacizumab targeted therapy (E) had significantly higher CAFDL scores compared with responders (left panels). ROC curves demonstrate the predictive power of the CAFDL signature for response to these treatments (right panels). (F–H) Non-responders to ipilimumab/nivolumab (F), pembrolizumab (G), and nivolumab (H) had significantly higher CAFDL scores compared with responders (left panels). ROC curves demonstrate the predictive power of the CAFDL signature for response to these treatments (right panels). *P < 0.05, **P < 0.01, and ****P < 0.0001.

Figure 5

CAFDL signature predicts clinical outcomes in immunotherapy cohorts and pan-cancer cohorts. (A) Patients with high CAFDL scores have significantly worse overall survival than those with low CAFDL scores in Braun’s, Gide’s, IMvigor210, Liu’s, and Nathanson’s cohorts. (B, C) In the IMvigor210 cohort, patients in the low CAFDL score group had significantly higher PD-L1 protein expression levels in immune cells (B) and tumor cells (C). (D)The high CAFDL score group had higher proportion of immune desert phenotype and lower proportion of immune-inflamed phenotype. (E) The high CAFDL score group had significantly lower CD8⁺ T effector infiltration. (F) In addition to COAD and READ, patients with high CAFDL scores have significantly worse overall survival than those with low CAFDL scores in 12 TCGA datasets: LUAD, BRCA, STAD, THCA, KICH, KIRC, ACC, SARC, BLCA, CESC, THYM, and UCEC. **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Figure 6

Pan-cancer immune correlates of CAFDL signature. (A) Correlation of CAFDL signature with immune cell infiltration evaluated using seven algorithms: TIMER, EPIC, xCELL, CIBERSORT, QUANTISEQ, MCPcounter, and TIDE across pan-cancer. (B–D) Correlation of CAFDL signature with stromal score (B), immune score (C), and ESTIMATE score (D) across pan-cancer.

Figure 7

Molecular features of CAFDL signature. (A) Correlation of CAFDL signature with cancer hallmark gene sets across pan-cancer. (B) Correlation of CAFDL signature with common immune regulators across pan-cancer. (C–E) Correlation of CAFDL signature with immune exclusion score (C), immune dysfunction score (D), and TIDE score (E) across pan-cancer.