Cancer-associated fibroblast-derived gene signatures determine prognosis in colon cancer patients

A “CAF signature”, involving 596 protein-coding genes, was identified when gene expression data from colon tumor CAFs and NFs were compared (Additional files 1 and 2). Functional gene-set analysis identified catabolic process, intracellular transport and dimerization and binding of proteins, among others, as enriched biological terms (Additional file 3). The prognostic value and clinical relevance of the new “CAF signature” were evaluated in a meta-dataset of 1235 colon cancer patients. Figure 2a shows the results of the risk prediction for each of the 1235 colon tumors and the best cut-off to divide patients into lower/higher “signature gene score”, based on expression levels of selected genes. Interestingly, the survival analysis showed that a high “signature gene score” subset of patients had shorter overall survival (OS) than the low “signature gene score” patients (Fig. 2b). In line with these data, the prognostic value of our previously reported “pro-migratory CAF signature” [8] was also confirmed in this cohort of 1235 patients (data not shown).

The ncRNA consists of functional RNA molecules that act by means of multiple mechanisms in the regulation of diverse cell functions [11]. In this study, we analyzed a second signature, “CAF-derived exosomes signature”, containing 67 predicted target genes of over-distributed ncRNA genes in CAF-derived exosomes previously described by our group [10] (Additional file 4). Gene set functional enrichment analysis identified important pathways related to cancer progression and microenvironment regulation, such as proliferation, positive regulation of signaling, regulation of multicellular organismal process and DNA repair processes, among others (Additional file 5). The risk cut-off point (Fig. 2c) of the “CAF-derived exosomes signature” in colon cancer was also evaluated in the same large meta-dataset (N = 1235), confirming the signature’s prognostic value (Fig. 2d).

Our first signature is related to CAF activation or status, while the second is associated with the crosstalk between fibroblasts and stromal/cancer cells orchestrated by CAF-derived exosomes. Both CAF-related signatures clearly overlapped in the classification of patients as at high or low risk of death (82.6% of concordance). To achieve a better and more balanced comparison of the behavior of the two CAF-related signatures, we undertook a new analysis of only the top 50 genes of each signature (Additional file 6), which confirmed their prognostic value (Fig. 2e and f). The combination of the two signatures (50 + 50 genes) did not greatly improve the separation of the survival curves of low- and high-“signature gene score” patients (Fig. 2g), indicating that there was no overlapping gene between the two gene signatures. Therefore, they include features that are closely related or measure similar characteristics in the colon tumors, though from two perspectives: CAF gene deregulation and fibroblasts/other cells’ crosstalk.

Initial analysis of survival of our cohort of patients confirmed the prognostic value of tumor stage (Additional file 7A). Interestingly, when we calculated the risk of each individual by splitting the population into stages (Fig. 3a), we found that all CAF-related signatures predicted higher risk at advanced stages than at early stages. In parallel, the prognostic value of CAF signatures was stronger in the stage III-IV group of patients than in the group of patients at early tumor stages (Fig. 3b-c and Additional file 8A). Our previous studies corroborated these data and showed the greater impact of microenvironment activation in patients’ outcome at advanced CC stages [8, 12]. Since these patients are usually those with worse outcomes, new therapies targeting microenvironment components would improve their clinical management. Specifically, stage III colon cancer patients are likely to develop recurrence. However, a subgroup of stage III patients would have low risk of recurrence and better outcome, with the advantage of shorter chemotherapy treatments that would avoid oxaliplatin cumulative neurotoxicity [13]. However, in daily clinical practice there is not enough evidence for better risk stratification in high/low-risk stage III patients. Our defined CAF-derived signatures with a clear prognostic value might improve patients’ classification and thus cancer patients’ management.

In addition, we confirmed that CMS4 patients had lower survival than CMS1, 2 or 3 ones (Additional file 7B). The Wilcoxon test revealed a higher “signature gene score” predicted by any of the 3 CAF-related signatures in CMS4-type tumors than in CMS1–3 ones (Fig. 3d). Although the Kaplan-Meier survival analysis showed that our CAF-related signatures marked differences in both CMS1–3 and CMS4 tumors (Fig. 3e-f and Additional file 8B), the comparison of the “signature gene score” distributions clearly indicated that in CMS4 risks were higher. CMS4-like subtypes and treatment associations have been studied in some clinical and preclinical studies. However, the clinical translation of the CMS subtypes into treatment decision-making is not widespread [14]. As our defined CAF-derived signatures had a higher prognostic value in those patients in the CMS4-like subtype, this might provide a new framework for patient classification in clinical trials and for the adjustment of suitable treatments. Therapies with the tumor microenvironment as the main target are currently an area of intensive research [15].

In summary, clinical decisions are based on the usual histological and clinical parameters that do not accurately predict the biological behavior of histologically equal tumors. The risk of recurrence of CC cancer is a critical factor in correctly setting up therapeutic guidelines, which highlights the importance of studying prognosis and predictive biomarkers. These new biomarkers classify patients on the basis of tumor microenvironment differences and define those pathological events at the molecular and cell level associated with patients’ survival. Our results emphasize ongoing efforts to decipher CAF biomarkers in order to advance prognosis and chemotherapeutic responses’ heterogeneity and to improve patients’ survival by tailoring therapies based on specific biological characteristics of CC tumors.