The prognostic model and immune landscape based on cancer-associated fibroblast features for patients with locally advanced rectal cancer

Abstract

Background

This study aimed to construct a nomogram based on CAF features to predict the cancer-specific survival (CSS) rates of locally advanced rectal cancer (LARC) patients.

Methods

The EPIC algorithm was employed to calculate the proportion of CAFs. based on the differentially expressed genes between the high and low CAF proportion subgroups, prognostic genes were identified via LASSO and Cox regression analyses. They were then used to construct a prognostic risk signature. Moreover, the GSE39582 and GGSE38832 datasets were used for external validation. Lastly, the level of immune infiltration was evaluated using ssGSEA, ESTIMATE, CIBERSORTx, and TIMER.

Results

A higher level of CAF infiltration was associated with a worse prognosis. Additionally, the number of metastasized lymph nodes and distant metastases, as well as the level of immune infiltration were higher in the high CAF proportion subgroup. Five prognostic genes (SMOC2, TUBAL3, C2CD4A, MAP1B, BMP8A) were identified and subsequently incorporated into the prognostic risk signature to predict the 1-, 3-, and 5-year CSS rates in the training and validation sets. Differences in survival rates were also determined in the external validation cohort. Furthermore, independent prognostic factors, including TNM stage and risk score, were combined to established a nomogram. Notably, our results revealed that the proportions of macrophages and neutrophils and the levels of cytokines secreted by M2 macrophages were higher in the high-risk subgroup. Finally, the prognostic genes were significantly associated with the level of immune cell infiltration.

Conclusion

Herein, a nomogram based on CAF features was developed to predict the CSS rate of LARC patients. The risk model was capable of reflecting differences in the level of immune cell infiltration.

1. Introduction

As is well documented, colorectal cancer is one of the leading causes of death worldwide and is characterized by high morbidity and cancer-related mortality rates [1,2]. The majority of rectal cancer patients present with locally advanced disease at diagnosis and are recommended to undergo neoadjuvant chemoradiotherapy (NCRT). Despite locally advanced rectal cancer (LARC) patients benefiting from neoadjuvant treatment and surgical procedures [[3], [4], [5]], the long-term prognosis varies [6,7]. Accumulating evidence suggests that the close relationship between rectal tumor cells and cellular components of the tumor microenvironment (TME) in LARC patients influences survival outcomes [[8], [9], [10]]. Each cellular component in the TME (e.g., stromal and immune cells) exerts distinct functions (e.g., paracrine effects and immune response) and participates in tumor progression. Consequently, different proportions of cellular components in the TME among LARC patients undergoing NCRT can result in varying clinical outcomes [11]. Thus, it is pivotal to expand our understanding of the TME in those patients.

Cancer-associated fibroblasts (CAFs) are predominant stromal components of the tumor microenvironment [12,13]. Earlier studies have documented an intimate association between CAFs and the progression of numerous tumors, including LARC [14,15]. Via the secretion of multiple growth factors, cytokines, and chemokines, CAFs promote cancer cell proliferation and invasion, remodel the extracellular matrix, modulate immune responses, and confer resistance to treatment, thereby affecting prognosis [[16], [17], [18]]. However, studies on the prognostic role of CAFs in LARC patients are scarce. Besides, most bioinformatics studies have focused on the prognostic assessment of colon cancer rather than rectal cancer. Given the correlation between CAFs and the prognosis of cancers, there is an urgent need to identify CAF-related biomarkers to effectively predict prognoses and explore their potential mechanisms in LARC patients.

In the current study, the prognosis, clinicopathologic features, and TME status of the high CAF proportion subgroup were compared with the low CAF proportion subgroup in LARC patients. Next, a prognostic risk signature and nomogram were developed and validated based on the differentially expressed genes (DEGs) between the high and low CAF proportion subgroups. Finally, the immune cell infiltration status was explored in the high- and low-risk subgroups.

2. Method

2.1. Data acquisition

The GSE87211 [19] dataset, including the information on LARC patients undergoing NCRT, was initially downloaded. Afterward, raw data were extracted from the Gene Expression Omnibus database (GEO, http://www.ncbi.nlm.nih.gov/geo) and inputted into the R software. The backgroundCorrect function (method = normexp) of the “limma” package was utilized for background correction, while raw data were normalized via the normalizeBetweenArrays function [20]. Probes were annotated, and the annotated data were used in the subsequent analyses. Data on clinicopathological factors, including age, gender, pT stage, pN stage, pM stage, survival time, and survival status, were collected. Patients with missing data on age, tumor stage, and survival information were excluded from this study. The primary endpoint of this study was the cancer-specific survival (CSS) rate. A total of 195 patients were included in this study. The training and validation sets were randomly generated using the “caret” R package. This study was approved by the Ethics Committee of the First Affiliated Hospital of Fujian Medical University.

2.2. CAFs infiltration and TME assessment

Estimate the Proportion of Immune and Cancer cells (EPIC, https://epic.gfellerlab.org), based on a specific set of RNA-seq reference gene expression datasets, is an effective deconvolution tool for assessing the proportion of various cellular components within the TME [21]. In the GSE87211 dataset, the EPIC algorithm was used to determine to compute the abundance of CAFs in the TME of LARC patients. The EPIC score was calculated for each patient, and the high and low CAF proportion subgroups were defined based on the median of the proportion of CAFs.

The Estimation of STromal and Immune cells in MAlignant Tumor tissues using the Expression (ESTIMATE) algorithm was used to evaluate the levels of stromal cell and immune cell infiltration in the TME [22]. Thereafter, the “estimate” R package was used to execute the ESTIMATE algorithm on expression profiles to determine the stromal score, immune score, and estimate score. These scores were subsequently compared between the high and low CAF proportion subgroups to explore differences in cell infiltration.

2.3. Prognostic risk signature and nomogram

Differentially expressed genes (DEGs) between the high and low CAF proportion subgroups were identified according to the following criteria: absolute log2 (fold change) > 0.5 and the false discovery rate (FDR) < 0.05. Univariate Cox regression was employed to identify prognostic genes with a significance threshold of P < 0.05. Furthermore, independent prognostic genes were determined via least absolute shrinkage and selection operator (LASSO) regression using the “glmnet” R package and multivariate Cox survival analyses. Following this, the prognostic risk signature was constructed based on independent prognostic genes and their corresponding coefficient. The risk score was calculated using the following formula: risk score = β_{gene 1} × Expression _{gene 1} +β_{gene 2} × Expression _{gene 2} +β_{gene 3} × Expression _{gene 3} + … +β_{gene n} × Expression _{gene n.} In other words, a risk score was calculated for each case. Patients were assigned to either the high-risk subgroup or the low-risk subgroup based on the median risk score. Then, differences in clinical outcomes were analyzed between the two subgroups. Moreover, the area under the receiver-operating characteristic (AUROC) curve was calculated to evaluate the accuracy of the prognostic risk signature. Of note, higher AUC values correspond to superior predictive values. A nomogram was generated based on independent prognostic factors using the R package “rms”. Individualized 1-, 3-, and 5-year CSS rates were predicted using this model. The performance of the nomogram was evaluated using the Harrell's concordance index (C-index), and the calibration curve was applied to assess the accuracy of the model-predicted survival compared to actual 1-, 3-, and 5-year CSS rates.

2.4. Validation datasets

Regarding the validation of the prognostic risk signature, external validation cohorts were collected from the GEO database. The GSE39582 dataset comprised a total of 585 samples, excluding 19 non-tumoral samples and 9 patients with unknown relapse-free survival (RFS) information. Finally, 557 patients were included in the analysis. Additionally, 122 patients with colorectal cancer from the GSE38832 dataset were also included.

2.5. Functional enrichment analysis

Regarding the assessment of differences in potential biological mechanisms between the high- and low-risk subgroups, gene set enrichment analysis (GSEA) was performed using hallmark gene sets (v7.2) from the Molecular Signatures Database (MSigDB, https://software.broadinstitute.org/gsea/msigdb/) [23]. An FDR q-value <0.25 was considered statistically significant.

2.6. Immune landscape

The correlation between risk scores and biomarkers of CAFs was initially determined. Given that CAFs interact with other immune cells in a paracrine manner, the relationship between risk scores and CAF-secreted cytokines was investigated. To further examine the level of immune cell infiltration between the two subgroups, single sample gene set enrichment analysis (ssGSEA) was conducted using the R package “gsva” [24]. Then, the enrichment scores of various immune cell components were determined, reflecting the relative abundances of immune cells in TME. The correlation between gene expression and immune cells was assessed using Tumor Immune Estimation Resource (TIMER) [25].

Moreover, Cell-type Identification By Estimating Relative Subsets of RNA Transcripts (CIBERSORTx, https://cibersortx.stanford.edu/) was used to evaluate the immune status [26]. The relationship between risk score and macrophage subtypes (M0, M1, and M2) was also assessed. Lastly, the levels of cytokines secreted by M2 macrophages across different risk scores were determined [27,28].

2.7. Statistical analysis

Statistical analyses were performed using R software (version 3.6.3) and GraphPad Prism 8. Between-group differences were assessed using the Wilcoxon Rank-sum test. The “pheatmap” R package was employed for generating heatmaps. Survival analysis and Kaplan-Meier (KM) curves were performed using the R package “survival”. Principal component analysis (PCA) and t-distributed Stochastic Neighbor Embedding (tSNE) were performed for dimension reduction, followed by data visualization using the R package “ggplot2”. The Spearman correlation test was used for correlation analysis. P < 0.05 is deemed statistically significant.

3. Results

3.1. CAF proportions in LARC patients

The flow diagram of this study is depicted in Fig. 1. Based on the median of CAF proportion acquired from the EPIC algorithm, 92 and 103 patients were included in the high and low CAF proportion subgroups, respectively. Their clinicopathologic features are listed in Supplementary Table 1. A significant difference was observed in CSS between the two subgroups, with patients with high CAF infiltration exhibiting worse clinical outcomes (Fig. 2A). Next, the relationship between clinicopathologic features and CAF proportion was evaluated (Fig. 2B–F). The analysis exposed that high CAF proportion was significantly associated with lymph node metastasis and distant metastasis (all P < 0.05), whereas age, gender, and pT stage were comparable between the two subgroups. Considering the close intracellular communication between CAFs and cellular components, differences in stromal and immune cell infiltration were further compared between the subgroups. As anticipated, the ESTIMATE score, stromal score, and immune score of the high CAF proportion subgroup were significantly higher than those of the low CAF proportion subgroup (Fig. 2G–I). These results collectively indicated that CAFs promote LARC progression and affect survival and that the prognostic risk signature based on CAF-related characteristics holds potential for clinical application.

Fig. 1.

Study flow diagram.

Fig. 2.

Kaplan–Meier survival curves depicting the prognostic value of CAFs in LARC patients undergoing NCRT (A), differences in clinicopathological features and immune infiltration levels between the high and low CAF proportion subgroups: age (B), gender (C), pT stage (D), pN stage (E), pM stage (F), ESTIMATE score (G), stromal score (H), immune score (I). CAF: cancer-associated fibroblasts; LARC: locally advanced rectal cancer; NCRT: neoadjuvant chemoradiotherapy.

3.2. Development and validation of the prognostic risk signature

1022 DEGs were identified between the high and low CAF proportion subgroups. Then, patients were randomized into a training set and a validation set (Table 1). In the training set, 19 genes were found to be positively correlated with prognosis. Meanwhile, LASSO regression (Fig. 3A–B) and multivariate Cox regression (Fig. 3C) analyses yielded 5 independent prognostic genes (SMOC2, TUBAL3, C2CD4A, MAP1B, and BMP8A). Next, the prognostic risk signature was constructed based on the expression levels of independent prognostic genes and their corresponding coefficient: 0.633 × SMOC2+1.030 × TUBAL3—0.952 × C2CD4A+1.252 × MAP1B+1.252 × BMP8A. Afterward, the risk score was calculated for each individual. Patients were further stratified into high- and low-risk subgroups based on the median risk score for the ensuing analyses. In addition, the expression levels of independent prognostic genes in the training and validation sets were detected. As illustrated in Fig. 3D–E, the expression level of SMOC2, TUBAL3, MAP1B, and BMP8A was higher in the two sets, whereas that of CD4A was low. At the same time, survival analyses determined that higher scores were significantly correlated with worse clinical outcomes in the training set (Fig. 4A–B), consistent with the findings in the validation set (Fig. 4D–E). Interestingly, the AUC values for the 1-, 3-, and 5-year CSS rates were 0.864, 0.946, and 0.947 in the training set (Fig. 4C), 0.716 and 0.696, 0.701 in the validation set, respectively (Fig. 4F). Finally, the ROC curves displayed that the prognostic risk signature had satisfactory predictive strength in both sets.

Table 1.

Clinicopathological features of patients in training set and validation set.

Characteristics	Training set	Validation set
	N (%）	N (%）
Age, years
<65	58(59.2)	56(57.7)
>65	40(40.8)	41(42.3)
Sex
Female	32(32.7)	28(28.9)
Male	66(67.3)	69(71.1)
pT stage
T0	18(18.4)	15(15.5)
T1	10(10.2)	11(11.3)
T2	20(20.4)	23(23.7)
T3	43(43.9)	44(45.4)
T4	7(7.1)	4(4.1)
pN stage
N0	66(67.3)	72(74.2)
N1	24(24.5)	18(18.6)
N2	8(8.2)	7(7.2)
pM stage
M0	90(91.8)	90(92.8)
M1	8(8.2)	7(7.2)
TNM stage
pCR	18(18.4)	15(15.5)
Stage I	23(23.5)	30(30.9)
Stage II	21(21.4)	26(26.8)
Stage III	28(28.6)	19(19.6)
Stage IV	8(8.2)	7(7.2)

TNM: tumor-node-metastasis, pCR: pathologic complete response.

Fig. 3.

Identification of independent prognostic CAF-related genes by LASSO regression (A–B) and multivariable Cox regression (C) in the training set. Differential expression of five CAF-related genes in the training set (D) and validation set (E). CAF: cancer-associated fibroblasts; LASSO: least absolute shrinkage and selection operator.

Fig. 4.

The predictive capability of the prognostic risk signature. Kaplan–Meier survival curves between the high- and low-risk subgroups in the training set (A) and validation set (D), the risk score distribution plot in the training set and (B) validation set (E), ROC curve for predicting 1-, 3-, and 5-year CSS rates in training the set and (C) validation set (F). ROC: receiver-operating characteristic, CSS: cancer-specific survival.

3.3. Construction of the nomogram

There was a significant difference in survival rates between the high- and low-risk subgroups in the whole set (P < 0.001, Fig. 5A). As is well documented, the combination of independent prognostic factors further enhances predictive abilities. Consequently, univariate and multivariate Cox analyses were performed to identify independent prognostic factors. As delineated in Fig. 5B–C, TN and risk scores were significantly and independently correlated with CSS. The nomogram was subsequently constructed by incorporating significant variables to predict the 1-, 3-, and 5-year CSS rates for LARC patients undergoing NCRT (Fig. 5D). The nomogram demonstrated accurate prognostic accuracy, with a C-index of 0.827 (95% confidence interval: 0.745–0.909). Besides, the calibration curves exhibited excellent agreement between the predictive- and actual 1-, 3-, and 5-year CSS rates (Fig. 5E).

Fig. 5.

The construction of the nomogram to predict CSS rates for LARC patients undergoing NCRT. Kaplan–Meier survival curves between the high- and low-risk subgroups (A), identification of independent prognostic factors via univariate (B) and multivariate Cox analyses (C), development of nomogram to predict CSS rates (D), and model assessment via the calibration curve (E). CSS: cancer-specific survival, LARC: locally advanced rectal cancer; NCRT: neoadjuvant chemoradiotherapy.

3.4. External validation of the prognostic risk signature

Based on the gene expression of the signature, the risk score was calculated in the external validation cohort. As illustrated in Fig. 6A, the difference in RFS was significant, with the high-risk group exhibiting worse clinical outcomes (P = 0.024) in the GSE39582 cohort. Likewise, the high-risk group was also linked to shorter CSS rates in the GSE38832 cohort (P = 0.024, Fig. 6B). Furthermore, the ROC curves (Fig. 6C–5D) indicated that the risk score could efficiently predict survival in both cohorts (GSE39582: AUC = 0.576, P = 0.004; GSE38832: AUC = 0.698, P = 0.002).

Fig. 6.

External validation of the prognostic risk signature. Kaplan–Meier survival curves between the high- and low-risk subgroups in the GSE39582 (A) and GSE38832 datasets (B) and the receiver-operating characteristic curves based on risk scores to predict survival rates in the GSE39582 (C) and GSE38832 datasets (D).

3.5. Subgroup analysis of clinicopathologic features

To further validate the predictive capability of the prognostic risk signature, subgroup analyses were performed on clinicopathologic features such as age, gender, pT stage, pN stage, and pM stage (Fig. 7A–J). The analysis uncovered significant differences between the high- and low-risk subgroups in patients aged <65 years, gender, pT3-4 stage, pN stage, and pM0 stage (all P < 0.05), demonstrating the outstanding predictive value of the risk signature.

Fig. 7.

Subgroup analysis according to clinicopathological features. Differences in survival rates between the high- and low-risk subgroups in patients aged <65 years (A), aged >65 years (B), female (C), male (D), pT0-T2 stage (E), pT3-T4 stage (F), pN0 stage (G), pN + stage (H), pM0 stage (I), pM1 stage (J). pCR: pathologic complete response, TNM: tumor-node-metastasis.

3.6. Visualization and potential mechanism of high- and low-risk subgroup

The benefits of the prognostic risk signature were further supported by the PCA and tSNE plots. As displayed in Fig. 8A–B, the risk subgroups were clearly discriminated. Furthermore, GSEA was performed to discover potential biological pathways in the high- and low-risk subgroups. Noteworthily, the high-risk subgroup was enriched in pathways related to angiogenesis, epithelial-mesenchymal transition (EMT), hypoxia, inflammatory response, apoptosis, complement, Notch signaling, Wnt/β-catenin signaling, and TGF-βsignaling pathway (Fig. 8C). These results can contribute to a better understanding of malignant biological behaviors and warrant further investigation.

Fig. 8.

The ability of the prognostic risk signature to distinguish between the high- and low-risk subgroups. PCA plot (A), tSNE plot (B), and GSEA for the high-risk subgroup (C). PCA: principal component analysis, tSNE: t-distributed Stochastic Neighbor Embedding, GSEA: gene set enrichment analysis.

3.7. Analysis of immune infiltration

The expression of CAF-related biomarkers was detected across different risk scores. A significant correlation was observed between the risk score and CAF-associated biomarkers (e.g., ACTA2, FAP, FSP1, S100A4) (Fig. 9A), implying the prognostic risk signature was characterized by CAFs and that CAFs exert biological functions through the secretion of cytokines. Consequently, the correlation between risk scores and cytokines secreted by CAFs was further investigated. The correlation heatmap (Fig. 9B) validated that the levels of cytokines (e.g., IL-6, CXCL-12, FGF2, CCL2) were higher in the high-risk subgroup.

Fig. 9.

Features of the prognostic risk signature and immune landscape. The correlation between risk scores and CAF-related biomarkers (A) and cytokines (B), the difference in stromal scores (C), immune scores (D), and infiltration of immune cells (E–F) between the high- and low-risk subgroups, The correlation between risk scores and macrophage subtypes (M0, M1, M2, G-I), The correlation between risk scores and cytokine related to M2 macrophages (J). CAF: cancer-associated fibroblasts, *: P < 0.05, **: P < 0.01.

As portrayed in Fig. 9C–D, the stromal scores and immune scores were significantly higher in the high-risk subgroup. CAFs have been reported to interact with other cellular components in the TME and synergistically influence tumor progression. Hence, the level of immune infiltration in the two risk subgroups was investigated by conducting ssGSEA. The results unveiled that the number of macrophages was higher in the high-risk subgroup (Fig. 9E–F). Macrophages typically display two polarization states, namely activated macrophages (M1) and alternatively activated macrophages (M2). Thus, the CIBERSORTx algorithm was adopted to explore differences in the expression of M0 (unpolarized), M1, and M2 macrophages. A positive correlation was noted between risk score and M0 and M2 macrophages, which are pro-tumorigenic subtypes (Fig. 9G and I). In contrast, no significant differences were detected in M1 macrophages, the anti-tumor subtype (Fig. 9H). M2 macrophages release cytokines that participate in tumor proliferation and TME remodeling, and as shown in Fig. 9J, M2 macrophages released high levels of cytokines (e.g., CCL2, CCL14, IL-4, TGFβ-1) in the high-risk subgroup. Conversely, a higher degree of natural killer (NK) cell infiltration was detected in the low-risk subgroup. Similarly, the infiltration levels of B cells, neutrophils, and regulatory T (T-reg) cells were markedly different between the two subgroups (Fig. 9E–F).

In addition, correlations between the expression levels of the five prognostic genes and immune cells were determined using the TIMER database (Fig. 10A–E). The results revealed a positive correlation between BMP8A and B cells (P = 0.009), CD4⁺ T cells (P < 0.001), macrophages (P = 0.003), neutrophils (P = 0.024), and dendritic cells (P < 0.001).

Fig. 10.

The correlation between the expression level of SMOC2 (A), TUBAL3 (B), C2CD4A (C), MAP1B (D), and BMP8A (E) and immune cells (including B cells, CD8⁺ T cells, CD4⁺ T cells, macrophages, neutrophils, and dendritic cells).

4. Discussion

At present, NCRT followed by total mesorectal excision is the gold standard for the treatment of LARC, demonstrating superiority in shrinking tumor size and stage, preserving the anal sphincter, and improving local tumor control [3,4]. This approach yields a pathologic complete response rate of approximately 10%–20% [[29], [30], [31]]. However, some LARC patients are unresponsive to NCRT due to differential treatment responses, which can result in varying survival outcomes [32]. More importantly, survival rates among patients with similar clinicopathologic features differed as well. Thus, pioneering an individualized prognostic model is warranted for LARC patients undergoing NCRT. In recent years, mounting evidence indicated that TME plays a crucial role in tumor progression and survival. As a principal cellular component in the TME, CAFs display a high degree of aggressive behaviors and enhanced secretory functions [33]. According to recent studies, CAFs confer treatment resistance by secreting cytokines and the delivery of exosomal vesicles in colorectal cancer [[34], [35], [36], [37]]. Saigusa et al. reported that the up-regulated expression of biomarkers of CAFs, encompassing fibroblast activation protein-α (FAP-α) and stromal cell-derived factor-1, was correlated with a poor prognosis in LARC patients undergoing NCRT [8]. Furthermore, previous studies also identified FAP as a poor predictive factor in colorectal cancer [38]. It is evident that CAF infiltration is correlated with a poor prognosis in colorectal cancer patients. However, to the best of our knowledge, studies focusing on prognostic risk signatures based on CAF-related features for LARC patients undergoing NCRT are limited.

Herein, the level of CAF infiltration was calculated in LARC patients undergoing NCRT using the EPIC algorithm. Significant differences in survival rates were observed between the two subgroups, with the clinical outcomes of patients with a high level of CAF infiltration being worse. Noteworthily, lymph node metastases and distant metastases were more prevalent in the high CAF proportion subgroup, and the degree of immune cell and stromal cell infiltration was higher. Taken together, these findings demonstrate that CAFs impact the progression and prognosis of LARC patients undergoing NCRT, with the immunologic microenvironment potentially playing a synergistic role. Thus, we postulate that the CAF-related prognostic signature holds considerable application potential. Moreover, in the training set, independent prognostic-related genes (SMOC2, TUBAL3, C2CD4A, MAP1B, and BMP8A) were screened and identified from DEGs between the high and low CAF proportion subgroups. Following this, an individualized prognostic risk signature was constructed based on the expression of prognostic genes and their corresponding coefficients. Notably, this signature was accurate in predicting the 1-, 3-, and 5-year CSS rates in both the training and validation sets. Furthermore, a nomogram was constructed by combining the signature with independent prognostic factors in order to predict survival. This nomogram may contribute to the development of personalized treatment strategies and cancer surveillance. The pro-tumorigenic functions of C2CD4A are well documented [39]. Nevertheless, the role of SMOC2 in the progression of rectal cancer remains controversial [40,41]. Furthermore, the biological functions of TUBAL3, MAP1B, and BMP8A in rectal cancer remain to be elucidated.

In addition, the potential mechanism was also explored in the risk subgroups via GSEA analysis, which determined that the high-risk subgroup was significantly enriched in cancer-associated pathways, such as angiogenesis, EMT, hypoxia, inflammatory response, apoptosis, and TGF-βsignaling pathway. A recent study undertaken by Unterleuthner et al. evinced that CAFs secrete WNT2 that promotes tumor angiogenesis in colon cancer [42]. One study indicated that CAFs promote the EMT of colorectal cancer cells by exosome-mediated intercellular communication [34]. Similar biologic behavior has also been described in gastric cancer [43], prostate cancer [44], and oral cancer [45]. Besides, the hypoxic microenvironment enhances the activity of CAFs, which in turn promotes tumor progression [[46], [47], [48]]. Exosomes derived from CAFs enhanced colorectal cancer cell stemness and promoted resistance to radiotherapy by activating the TGF-βpathway [37]. These results conjointly suggested the potential mechanisms in the high-risk subgroup.

It is worthwhile emphasizing that the level of macrophage infiltration was significantly different between the high- and low-risk subgroups. Specifically, the degree of M2 macrophage recruitment was higher in the high-risk subgroup. Indeed, macrophages constitute tumor-infiltrating immune cell components in the TME and exhibit two polarization states, namely classically activated M1 subtype and alternatively activated M2 subtype [49]. The former possesses a tumoricidal activity phenotype, whereas the latter stimulates tumor progression. The crosstalk between CAFs and macrophages has been described in recent studies. In colorectal cancer, the expression of biomarkers of M2 macrophages (CD206, CD163) has been reported to be up-regulated following co-culture with CAFs and macrophages [50]. In addition, macrophage colony-stimulating factor was secreted by CAFs to shift macrophages to the M2 polarization state and promote pancreatic ductal adenocarcinoma cell proliferation, migration, and invasion [51]. CAFs promote tumor metastasis and angiogenesis in prostate carcinoma in conjunction with M2 macrophages [52]. Moreover, the killing capacity of NK cells was inhibited after the synergistic interaction between CAFs and macrophages, protecting colorectal tumor cells against NK cells [50]. In melanoma, CAFs induce a decrease in the killing capacity of NK cells [53]. On the other hand, CAFs were found to be implicated in the regulation of T-reg cells, and a high level of T-reg cells was associated with a poor prognosis in lung adenocarcinoma [54]. Overall, these findings highlight that CAFs closely interact with immune cells to synergistically affect tumor characteristics, warranting further investigation into their role in colorectal cancer.

Nonetheless, some limitations of this study merit acknowledgment. To begin, the biological function of the five CAF-related genes necessitates further exploration. Secondly, the detailed mechanism underlying intercellular communications between CAFs and immune cells requires further studies.

5. Conclusion

A higher CAF proportion was correlated with worse survival in LARC patients undergoing NCRT. Herein, a prognosis risk signature was established based on DEGs between the high and low CAF proportion subgroup. Furthermore, a nomogram was developed to predict the 1-, 3-, and 5-year CSS rates by incorporating independent prognostic factors into the model. Lastly, the risk model was also capable of reflecting differences in the level of immune cell infiltration.

Funding

This study was supported by National Natural Science Foundation of China (No. 82172800), Special Financial Foundation of Fujian Provincial (No. 2020B019), Joint Funds for the Innovation of Science and Technology, Fujian Province (No. 2020Y9125), Talent programs granted from The First Affiliated Hospital of Fujian Medical University (YJRC3600), and Startup Fund for Scientific Research of Fujian Medical University (No.2020QH2036).

Ethics statement

This study was reviewed and approved by the Ethics Committee of the First Affiliated Hospital of Fujian Medical University, with the approval number:2021323.

Data availability statement

Data is available from the GEO database (https://www.ncbi.nlm.nih.gov/geo, GSE87211, GSE39582, and GSE38832).

CRediT authorship contribution statement

Huajun Cai: Writing – original draft, Methodology, Funding acquisition, Formal analysis, Data curation, Conceptualization. Yijuan Lin: Methodology, Investigation. Yong Wu: Validation, Software. Ye Wang: Supervision. Shoufeng Li: Data curation. Yiyi Zhang: Methodology. Jinfu Zhuang: Supervision. Xing Liu: Supervision. Guoxian Guan: Writing – review & editing, Methodology, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Abbreviations

LARC

locally advanced rectal cancer

CAFs

cancer-associated fibroblasts

CSS

cancer-specific survival

EPIC

Estimate the Proportion of Immune and Cancer cells

LASSO

least absolute shrinkage and selection operator

ssGSEA

single sample gene set enrichment analysis

ESTIMATE

Estimation of Stromal and Immune cells in Malignant Tumor tissues using Expression

CIBERSORTx

Cell-type Identification by Estimating Relative Subsets of RNA Transcripts

TIMER

Tumor Immune Estimation Resource

TNM

Tumor-node-metastasis

NCRT

neoadjuvant chemoradiotherapy

TME

tumor microenvironment

DEGs

differentially expressed genes

GEO

Gene Expression Omnibus

FDR

false discovery rate

AUC

area under the curve

ROC

receiver-operating characteristic

C-index

concordance index

RFS

relapse-free survival

PCA

principal component analysis

tSNE

t-distributed Stochastic Neighbor Embedding

T-reg

regulatory T

NK

natural killer

FAP-α

fibroblast activation protein-α

Appendix A. Supplementary data

The following is the Supplementary data to this article: