Introduction

Colorectal cancer (CRC) is the leading cause of death worldwide, with its incidence and mortality rates ranking third and second among malignant tumors, respectively [1]. In recent years, the incidence and mortality rates of CRC have increased significantly [2]. Surgical treatment, radiation therapy, and systemic chemotherapy are the main therapeutic approaches for treating CRC [3]. Surgery is usually the treatment of choice for early-stage CRC. Chemotherapy drugs such as 5-fluorouracil (5-FU), fluorouracil (capecitabine), and calcium oxide (oxaliplatin) can be used as adjuvant treatments to reduce tumor volume, control postoperative recurrence, or provide remission therapy in advanced stages [4,5,6].

Since the 1990s, fluorouracil (FU)-based adjuvant chemotherapy has been an essential option for treating advanced CRC [7]. The use of adjuvant fluorouracil-based chemotherapy in patients with stage III colon cancer is thought to be standard care; however, it is not routinely recommended for patients with stage II colon cancer [8, 9]. Some patients develop resistance to chemotherapy drugs, which is one of the main causes of tumor treatment failure [10]. Once chemoresistance emerges, tumors tend to relapse and metastasize, causing the death of 70 to 80% of cancer patients; thus, chemoresistance is one of the greatest challenges in the long-term management of incurable metastatic disease [11]. 5-FU, an intravenous synthetic fluorouracil analog, is currently the most important chemical for treating CRC [12]. Using 5-FU can effectively reduce tumor recurrence and metastasis and improve the survival rates. However, cancer cells gradually develop resistance during chemotherapy, leading to the failure of chemotherapy drugs [13]. Resistance to 5-FU can result from various factors, including metabolic enzymes and cancer stemness. Some studies have suggested that mutations in genes and changes in the expression levels of genes involved in metabolic pathways associated with 5-FU may contribute to the development of resistance [14]. Thymidylate synthase polymorphism is now an emerging focus of interest responsible for 5-FU resistance [15,16,17,18]. In addition, cancer stem cells have long been associated with chemotherapy resistance [19]. However, because of the complexity of the tumor microenvironment, the underlying mechanism leading to chemotherapy resistance remains unclear.

A tremendous benefit has been achieved with immunotherapy in cancer treatment in recent years. The FDA approved immune checkpoint regimens in 2017 for CRC patients with defective mismatch repair (dMMR) or high microsatellite instability (MSI-H) levels. However, immunotherapy is inefficient for tumors that are proficient in mismatch repair (pMMR), microsatellite stable (MSS), or have low levels of microsatellite instability (MSI-L), which account for a large proportion of CRCs [20]. Although chemotherapy and immunotherapy have achieved unexpected efficacy in treating CRC, with the development of precision therapy, the limitations of monotherapy, especially chemotherapy resistance and a low rate of immunotherapy response, have gradually emerged. Recent research suggests that there are complex interactions between the immune system and chemotherapy. While immunosuppressive effects of chemotherapeutic agents have been reported [21, 22], studies have shown that chemotherapy can enhance the immunogenicity of tumor cells, activate immune effectors, and alleviate tumor-induced immunosuppression [23]. However, there are currently no effective biomarkers for determining the prognosis of patients with CRC or for predicting their response to chemotherapy and immunotherapy.

In this study, we first identified genes associated with 5-FU resistance in CRC patients and constructed a novel 5-FU resistance-related signature (5-FRSig) according to these genes. We systematically investigated and validated the prognostic value, chemotherapeutic response, immune landscape, and immunotherapy predictive power of the signature. Our study demonstrated that the 5-FRSig can be used as an independent prognostic factor to predict the response to chemotherapy and immunotherapy in CRC patients. This study is expected to lead to more accurate and effective treatment strategies for patients with CRC, including chemotherapy, immunotherapy, targeted therapy, and combination therapy, providing guidance strategies for the precise diagnosis and treatment of CRC.

Materials and methods

Data sources and processing

The transcriptomic data of parental and 5-FU-resistant cells were obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE196900 [24]. Differentially expressed genes (DEGs) between 5-FU-resistant cells and parental cells in the HCT116 and SW480 cell lines were analyzed with the R package “DESeq2” [25]. Genes with an absolute value of log2 (fold change (FC)) > 2 and adjusted P value < 0.05 were considered DEGs. DEGs with consistent trends in both cell lines were considered 5-FU resistance-related candidate genes. The transcriptomic and clinical data of The Cancer Genome Atlas (TCGA) colon adenocarcinoma and rectum adenocarcinoma datasets were downloaded from the GDC data portal (https://portal.gdc.cancer.gov/). A total of 597 CRC samples with accessible clinical and survival data were enrolled in the training cohort. External validation was performed using the GSE37892 [26], GSE17537 [27], GSE192667 [28], and GSE29621 [29] datasets, all of which were downloaded from the GEO database. This study used consensus molecular subtypes (CMSs) obtained from the Colorectal Cancer Subtyping Consortium Synapse [30].

Construction of the 5-FRSig

Univariate Cox regression was conducted for 5-FU resistance-related candidate genes in the TCGA cohort. Subsequently, multivariate stepwise regression was conducted using the R package “MASS.” An optimized risk model associated with 5-FU resistance was then established, including thirteen 5-FU resistance-related genes (5-FRGs). For each patient, the risk score was calculated as follows: \(\text{R}\text{i}\text{s}\text{k} \,\text{s}\text{c}\text{o}\text{r}\text{e}={\sum }_{i=1}^{13}\text{E}\text{x}\text{p}\text{r}\text{e}\text{s}\text{s}\text{i}\text{o}\text{n} \left({\text{m}\text{R}\text{N}\text{A}}_{i}\right)\times \text{C}\text{o}\text{e}\text{f}\text{f}\text{i}\text{c}\text{i}\text{e}\text{n}\text{t}\left({\text{m}\text{R}\text{N}\text{A}}_{i}\right)\).

The correlation between the risk score and the expression level of the thirteen genes was analyzed using Spearman’s correlation. The R package “ComplexHeatmap” was applied to depict the results [31].

Prognostic analysis and construction of the nomogram

Patients were divided into high- and low-risk groups according to the median value of the risk score. The prognostic value of the 5-FRSig was evaluated by Kaplan‒Meier (K‒M) survival analysis, multivariate Cox regression analysis, and time-dependent receiver operating characteristic (ROC) curve analysis. A nomogram model integrating all independent prognostic factors, including risk factors, was established with the R package “rms” to further improve the prediction power. Moreover, calibration curves were generated. Overall survival (OS) and progression-free survival (PFS) rates were analyzed in the TCGA cohort. In addition, associations between the risk score, stage, and CMS were analyzed.

Functional enrichment analysis

DEGs between the high- and low-risk groups were identified using the R package “DESeq2” with a threshold of adjusted P < 0.05 and an absolute value of log2FC > 0.5. Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses, along with gene set enrichment analysis (GSEA), were conducted using the R package “clusterProfiler” [32].

Tumor immune microenvironment analysis

First, we assessed the immune scores of risk groups using the Estimation of STromal and Immune cells in MAlignant Tumors using Expression data (ESTIMATE) algorithm, including estimate, immune, and stromal scores [33]. In addition, multiple deconvolution algorithms were used to evaluate immune subsets, including ESTIMATE [33], Tumor Immune Estimation Resource (TIMER) [34], Cell-type Identification By Estimating Relative Subsets of RNA Transcripts (CIBERSORT) [35], Estimating the Proportion of Immune and Cancer cells (EPIC) [36], xCELL [37], Microenvironment Cell Populations-counter (MCP-counter) [38], Quantification of the Tumor Immune Contexture from Human RNA-seq Data (quanTIseq) [39] and gene set variation analysis (GSVA) [40]. In addition, the expression levels of immune-related genes were extracted and compared between the high- and low-risk groups.

Prediction of the response to immunotherapy and chemotherapy

The correlation between the gene signature and somatic mutations and neoantigens was analyzed. The mutation annotation format (MAF) for the TCGA cohort was obtained from the TCGA data portal (https://portal.gdc.cancer.gov) and analyzed with the R package “maftools.” The numbers of somatic mutations and neoantigens were retrieved from The Cancer Immunome Atlas (TCIA) (https://tcia.at) [40]. Patient responses to immune checkpoint inhibitors were predicted using tumor immune dysfunction and exclusion (TIDE, http://tide.dfci.harvard.edu) [41], a well-developed and accurate method for predicting the efficacy of immunotherapy. Four independent cohorts with immunotherapy information, namely, the IMvigor210 (n = 298) [42], CheckMate 025 (n = 281) [43], GSE176307 (n = 88) [44], and GSE78220 (n = 27) [45] cohorts, were retrieved to further validate the results in samples with treatment response data. Receiver operating characteristic (ROC) curves were used to assess the ability of the 5-FRSig to predict the response to 5-FU. The data of three GEO cohorts, GSE39582 [46], GSE106584 [47], and GSE103479 [48], were retrieved, and only samples from patients who received adjuvant 5-FU treatment were included in the analysis. For GSE39582, GSE103479, and GSE106584, 82, 66, and 35 samples were retained, respectively. Patients who experienced relapse were considered nonresponders to 5-FU treatment. The R package “oncoPredict” [49] was used to analyze the sensitivity of patients to commonly used drugs. This analysis was conducted using transcriptomic data and drug sensitivity data downloaded from Genomics of Drug Sensitivity in Cancer (GDSC, https://www.cancerrxgene.org/) [50], The Cancer Therapeutics Response Portal (CTRP, https://portals.broadinstitute.org/ctrp.v2.1/) [51], and Profiling Relative Inhibition Simultaneously in Mixtures (PRISM, https://www.theprismlab.org/) [52]. The half-maximal inhibitory concentration (IC50) in the GDSC and the area under the dose‒response curve (AUC) in the CTRP and PRISM cohorts were negatively correlated with drug sensitivity.

Consensus clustering analysis

Sixty-five prognostic genes were identified in the univariate Cox regression analysis. Unsupervised consensus clustering was conducted using the R package “ConsensusClusterPlus” with 50 iterations and a resampling rate of 80% [53]. K‒M survival, tumor immune microenvironment, DEG, and functional enrichment analyses were conducted between subclusters to explore the biological properties. In addition, the correlation between the gene signature and clusters was analyzed.

Real-time quantitative PCR validation

Forty-five pairs of CRC and adjacent normal tissue samples were collected from the Affiliated Hospital of Qingdao University for RT–qPCR validation. The ethical considerations and the criteria for inclusion and exclusion were used, as previously described [54]. Total RNA was extracted using an RNeasy kit (Beyotime, Shanghai, China, R0027) according to the manufacturer´s instructions. Then, 1 µg of total RNA was reverse transcribed with SuperScript II reverse transcriptase (Takara, Japan, RR047). Quantitative PCR analysis was performed with SYBR Green Mix (Takara, Japan, RR820) using an ABI 7900 HT Real-Time PCR System. GAPDH was used as an internal control. The primers used in this study are listed in Supplementary Table 1.

Identification and verification of candidate small molecules to reverse 5-FU resistance

To improve the clinical application of our signature, we used a Connectivity Map (CMap) to predict candidate small molecules that might reverse 5-FU resistance. CMap is a public resource comprising a comprehensive catalog of cellular signatures representing systematic perturbation with genetic and pharmacologic interference. The connectivity score, as calculated by CMap, indicates that the molecule could enhance a biological property if positive and reverse the biological property if negative [55]. In this study, the hub genes were divided into upregulated and downregulated groups and imported into the CMap database. Then, a list of small molecules was obtained, and these molecules were ranked by the connectivity score. Small molecules with negative connectivity scores, an FDR q value < 0.05, a specific mechanism of action (MoA), and targets were considered to have reliable potential to reverse 5FU resistance. Molecular docking was used to verify the reliability of these small molecules in reversing 5FU resistance. First, the 3D structures of the target proteins were downloaded from the RCSB Protein Database Bank (PDB, http://www.rcsb.org/) or the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk/). The proteins were dehydrated and/or ligand-removed with PyMOL 2.5 software and saved in PDB format. The processed target protein was then imported into AutoDock Tools 1.5.6 software for hydrogenation and charge calculations and stored in PDBQT format. Second, the 3D structures of small molecules in SDF format were downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) and saved in mol2 format with Open Babel 2.3.1 software. Mol2 files of small molecule drugs were imported into AutoDock Tools, the total charge was detected, the charge was assigned, and flexible rotatable bonds were viewed and saved in PDBQT format. The grid box was obtained by using the GetBox Plugin in PyMOL. Finally, AutoDock Vina 1.1.2 was used to conduct molecular docking for 10 potential molecules and thirteen target proteins. The binding strength was evaluated according to the docking binding energy. The results were visualized with PyMOL 2.5 software.

Single-cell RNA sequencing analysis

We used the scRNA-seq cohort GSE178318 [56], which contains single-cell expression profiling of CRC liver metastases from treated patients and untreated patients. The R packages “Seurat” [57] and “Harmony” [58] were used to read sample data and remove batch effects between the samples. We then used the t-SNE method for dimension reduction processing to obtain the clusters and performed cell type annotation through the R package “SingleR” [59]. The score and distribution of the 5-FRSig in the single-cell samples were calculated using two methods, namely, Ucell and singscore. The Ucell algorithm uses the Mann–Whitney U statistic to calculate the gene set enrichment score for a single sample based on the gene expression ranking, while the singscore algorithm employs a gene enrichment score that is calculated based on the gene expression ranking of a single sample. This score assesses the distance of the gene set from the center. Both of these algorithms can be implemented through the R package “irGSEA.”

Statistical analysis

R software (version 4.1.3) and GraphPad Prism 8.0 (GraphPad Software Inc., San Diego, CA, United States) were used for data analysis and visualization. The continuous variables were analyzed using the Wilcoxon or Kruskal‒Wallis tests. Categorical variables were analyzed using the chi-square test (χ2) or Fisher’s exact test. Relationships between the risk scores and the expression levels of different genes were examined by Spearman’s correlation analysis. A P value < 0.05 was considered to indicate significance.

Results

Construction of a prognostic 5-FRSig

The flowchart provides an overview of the primary design of the current investigation (Fig. 1). We selected DEGs between 5-FU-resistant cells and parental cells in the HCT116 and SW480 cell lines from GSE196900. A total of 565 DEGs with a consistent trend in the two cell lines were identified; of these 5-FRGs, 513 were identified in the TCGA cohort (Fig. 2A). Then, we used univariate Cox regression analysis and obtained 65 of the 513 5-FRGs associated with prognosis (P < 0.05). Multivariate Cox analysis was subsequently applied, and 13 5-FRGs with independent prognostic value were identified. Figure 2B shows the results of the univariate Cox regression analysis of 13 5-FRGs. Then, an optimized risk model with 13 5-FRGS was established stepwise, including ALPK3 (alpha kinase 3), CPA4 (carboxypeptidase A4), DNAH7 (dynein axonemal heavy chain 7), FGF2 (fibroblast growth factor 2), HOXD13 (homeobox D13), NRG1 (neuregulin 1), PPP1R3F (protein phosphatase 1 regulatory subunit 3 F), SIX2 (SIX homeobox 2), SLC39A8 (solute carrier family 39 member 8), TMEM139 (transmembrane protein 139), TNFRSF19 (TNF receptor superfamily member 19), ZDHHC2 (zinc finger DHHC-type palmitoyltransferase 2), and ZNF607 (zinc finger protein 607). The patients were divided into high- and low-risk groups using the median risk score. Figure 2C displays the distribution of risk scores among TCGA-CRC patients. Additionally, the expression levels of twelve of the thirteen genes were significantly correlated with the risk score. Notably, SLC39A8 showed the strongest negative correlation (r = − 0.485, Fig. 2D). As shown in Fig. 2E, HOXD13, NRG1, SIX2, DNAH7, CAP4, ALPK3, and FGF2 were significantly upregulated in 5-FU-resistant cells, and TNFRSF19, TMEM139, PPP1R3F, SLC39A8, ZNF607, and ZDHHC were significantly downregulated. In addition, we analyzed the correlation between the risk score and clinicopathological features; the risk score was significantly correlated with status, T stage, N stage, and tumor stage (Fig. 2F). We then used three external datasets for validation and found that the risk scores of the GSE17537, GSE39582, and GSE37892 datasets differed significantly at various stages (Fig. 2G).

Fig. 1
figure 1

Flowchart of the entire study

Full size image
Fig. 2
figure 2

Construction and validation of the 5-FU resistance-related signature. A. The DEGs of two cell lines (HCT116 and SW480) in the GSE196900 dataset. B. Univariate Cox analysis of TCGA-OS data for the 13 5-FU resistance-related genes used to construct the signature. C. Distribution of risk score (high and low) and status (dead and alive) in the TCGA-CRC cohort; D. The correlation between risk score and thirteen genes; E. Expression profiles of the genes in the normal and resistant groups in GSE196900. F. Correlation heatmap between risk groups and clinical characteristics. G. Validation of the signature in the GSE17537 and GSE37892 datasets. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001

Full size image

Prognosis analysis of the 5-FRSig and construction of a nomogram

We then evaluated the prognostic capacity of the signature using the TCGA cohort. According to K‒M survival curves, patients in the high-risk group had significantly worse OS (P < 0.0001) and PFS (P < 0.001, Fig. 3A and B) rates. The risk score was confirmed to be an independent prognostic factor for both OS (HR = 3.136, 95% CI = 1.997–4.924, P < 0.001) and PFS (HR = 1.418, 95% CI = 1.002–2.007, P = 0.049). This result was validated in four external datasets, namely, GSE192667, GSE29621, GSE17537, and GSE37892 (Fig. 3C–F). To improve discrimination and make the model more applicable, we established a prognostic nomogram integrating the signature and independent risk factors (Fig. 4A–C). To confirm the superiority of the nomogram, calibration curves and ROC analysis were used to validate the nomogram’s prognostic accuracy and specificity. The results indicated that the 5-FRSig score and nomogram were superior to the stage in predicting OS or PFS outcomes in the TCGA cohort and GSE39582 cohort (Fig. 4D–E). These results indicate that our risk signature is an independent prognostic factor reliable for predicting survival probability. Furthermore, the nomogram integrating the risk score and clinicopathological characteristics was more reliable and accurate in predicting survival outcomes.

Fig. 3
figure 3

Prognostic analysis and validation of the 5-FU resistance-related signature. K‒M survival curve and multivariate analysis of the TCGA-OS (A), TCGA-PFS (B), GSE192667-OS (C), GSE29621-OS (D), GSE17537-OS (E), and GSE37892 (F) cohorts

Full size image
Fig. 4
figure 4

Construction and validation of a nomogram. Nomograms constructed with three independent prognostic factors for 1-, 3-, and 5-year OS rates in the TCGA cohort (A), PFS rate in the TCGA cohort (B), and OS rate in the GSE39582 cohort (C). Calibration curves and receiver operating characteristic (ROC) curves showing the predictive accuracy of the risk score and nomogram in the TCGA-OS (D), TCGA-PFS (E), and GSE39582-OS (F) cohorts

Full size image

The 5-FRSig predicts patient response to 5-FU therapy

Given that this model was constructed according to 5-FU resistance-related genes, we wondered whether it could discriminate patients’ responses to 5-FU treatment. We validated this idea using CRC samples treated with 5-FU. The risk score was calculated, and the association between the risk score and patient response to 5-FU treatment was analyzed. The results revealed that the recurrence rates were significantly greater in the high-risk group, and patients who experienced relapse had significantly greater risk scores. In addition, the AUCs of the ROC curves were 0.782, 0.755, and 0.944, all of which were greater than 0.75, demonstrating the signature’s good discriminative ability (Fig. 5A–C).

Fig. 5
figure 5

The risk signature predicts patient response to 5-FU therapy. The 5-FU resistance-related signature predicts the recurrence of patients who received 5-FU treatment and the relationship between the risk score and recurrence in the GSE39582 (A), GSE103479 (B), and GSE106584 (C) datasets. ∗∗∗P < 0.001

Full size image

Analysis of the underlying biological pathway of the 5-FRSig

We used GSEA and GO and KEGG analyses to explore the underlying molecular mechanism of the 5-FRSig. First, the DEGs between subgroups in the TCGA cohort were analyzed. The results are displayed in a volcano plot (Fig. 6A). A total of 1,057 upregulated genes and 664 downregulated genes were identified with an adjusted P value < 0.05 and abs(logFC) > 0.5. GO analysis revealed that the DEGs were mainly enriched in inflammation-related pathways, including acute inflammatory response and chemokine activity. In addition, cancer-related pathways, such as ERK1 and ERK2 cascade, and immune-related pathways, such as granulocyte migration, were also found to be enriched in the analysis (Fig. 6B). GSEA further revealed that cancer-related pathways, such as epithelial mesenchymal transition, apical junction, KRAS signaling, WNT beta-catenin signaling, Hedgehog signaling, angiogenesis and hypoxia, were enriched in the high-risk group (Fig. 6C). KEGG analysis revealed that the upregulated genes were enriched in cancer-related pathways, including the PI3K-Akt signaling pathway, MAPK signaling pathway, Wnt signaling pathway, Rap1 signaling pathway, gastric cancer pathway, and TGF-beta signaling pathway (Fig. 6D), while the downregulated genes were enriched in immune-related pathways, including cytokine‒cytokine receptor interaction, the IL-17 signaling pathway, viral protein interaction with cytokine and cytokine receptor, rheumatoid arthritis and the TNF signaling pathway (Fig. 6E). The enrichment analysis demonstrated that cancer-immunity interactions may explain the prognostic power of our signature.

Fig. 6
figure 6

Analysis of the underlying biological pathways of the 5-FU resistance-related signature. A. Volcano plot of differentially expressed genes in the high- and low-risk groups with a threshold of FDR < 0.05 and absolute log2(FC) > 0.5. B. GO enrichment of differentially expressed genes. C. GSEA enrichment of differentially expressed genes. KEGG enrichment of the high- (D) and low-risk groups (E)

Full size image

Immune profile analysis of the 5-FRSig

Multiple computational methods were used to determine the degree of immune cell infiltration in each sample to further investigate the relationship between the signature and the immune system. A heatmap of the cells that differed significantly between the subgroups is shown in Fig. 7A. According to the heatmap, the low-risk group exhibited increased levels of antitumor immune cells, such as CD4 + T cells (EPIC, xCELL), CD8 + T cells (EPIC, xCELL), neutrophils (EPIC, xCELL) and B cells (EPIC, xCELL), while the high-risk group exhibited increased levels of cancer-associated fibroblasts (MCPcounter, EPIC, xCELL) and macrophages (CIBERSORT, xCELL). We plotted radar charts to better visualize the differences between subgroups (Fig. 7B), demonstrating that the cytotoxicity score, immune score, number of neutrophils, number of CD4 + T cells, number of CD8 + T cells, number of B cells and number of NK cells were significantly greater in the low-risk group, while the stroma score and the numbers of cancer-associated fibroblasts and macrophages were significantly greater in the high-risk group. In addition, the analyses of immune-related genes demonstrated that the low-risk group had significantly greater immunostimulator levels and moderately greater cytotoxicity and immune inhibitor levels (Fig. 7C). Given these findings, we concluded that the low-risk group had more immune cell infiltration and greater antitumor activity, explaining the better outcomes in this group.

Fig. 7
figure 7

Immune profile analysis of the 5-FU resistance-related signature in the TCGA-CRC cohort. The heatmap (A) and radar map (B) show immune cell infiltration in the high- and low-risk groups calculated by multiple algorithms. C. Expression of immunomodulators in the high- and low-risk groups. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001

Full size image

Mutation and immunotherapy response analysis of the 5-FRSig

Since immune cells are a prerequisite for immunotherapy, we hypothesized that the low-risk group would have a greater response rate to immune checkpoint inhibitors. We validated this hypothesis from different perspectives. First, we investigated the mutational landscape. We identified the top 20 mutated genes in the TCGA cohort (Fig. 8A). Significantly greater mutation rates were detected in two genes in each of the high- and low-risk groups, while no differences were detected in other genes (Fig. 8B). Moreover, there was no significant difference in the number of somatic mutations between the subgroups (Fig. 8C). Thus, there were no significant differences in mutations between the subgroups. However, there were more neoantigens in the low-risk group. Then, we analyzed the expression of the immune checkpoints PD-1, PD-L1, PD-L2, and CTLA4. The results showed that the expression levels of CD274 and CTLA4 were significantly greater in the low-risk group (Fig. 8D). Subsequently, we applied the TIDE algorithm to predict patient response to immune checkpoint blockade therapy. We observed a significantly greater response rate in the low-risk group and a significantly lower risk score for responders. Consistent with this finding, a lower TIDE score/dysfunction score/exclusion score and higher MSI score were found in the low-risk group, all of which indicated a greater response rate. Immunosuppressive MDSCs and CAFs were significantly more abundant in the high-risk group (Fig. 8E). We then analyzed four datasets with immunotherapeutic information to validate the above results. Consistent results were obtained, including significantly greater response rates (CR/PR/SD) in the low-risk group and significantly lower risk scores for responders (Fig. 8F-I). Therefore, we directly demonstrated the association between a low risk score and a high response rate to immunotherapy. These results suggested that the 5-FRSig could predict immunotherapy efficacy.

Fig. 8
figure 8

Mutation and immunotherapy response analysis of the 5-FU resistance-related signature. A. The top 20 mutated genes in the high- and low-risk groups. B. Comparison of the mutation ratio in the high- and low-risk groups. C. Comparison of mutations and neoantigens between the high- and low-risk groups. D. Expression of key immune checkpoint molecules in the high- and low-risk groups. E. The distribution of different responders and calculated scores in the high- and low-risk groups in the TCGA cohort, calculated by the TIDE algorithm. The distribution of different responders in the high- and low-risk groups in the IMvigor210 (F), CheckMate 025 (G), GSE176307 (H), and GSE78220 (I) datasets. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001

Full size image

Construction and prognosis analysis of 5-FU resistance-related clusters

The 65 5-FRGs associated with prognosis were subjected to unsupervised cluster analysis. The ideal number of clusters was found to be two using the consensus CDF curve. After unsupervised clustering, we identified two clusters within the TCGA cohort (Fig. 9A). Figure 9B displays the distribution of clusters and status. Subsequently, we investigated the prognostic value of 5-FU resistance-related clusters. The results of the K‒M survival curve and multivariate Cox regression analyses showed that patients in Cluster 1 had a better survival probability, and this cluster was a significant independent prognostic factor for CRC (Fig. 9C and D). As shown in Fig. 9E, the cluster was significantly correlated with status, MSI status, and CMS status. In addition, we conducted functional enrichment analyses of the clusters, including GSEA and KEGG and GO analyses. The DEGs in Clusters 1 and 2 are shown in Supplementary Fig. 1A. We found that the pathways enriched in the clusters were mainly inflammatory and immune-related pathways (Supplementary Fig. 1B-D). Similarly, we analyzed the immune landscape of the clusters, including immune cell infiltration and the expression of immunomodulators. We found that a greater proportion of inhibitory immune cells infiltrated Cluster 2, including CAFs, Tregs, and MDSCs (Supplementary Fig. 2A–C). Interestingly, in the low-risk group, the proportion of patients in Cluster 1 was greater than that in Cluster 2, and the risk score of patients in Cluster 1 was significantly lower than that in Cluster 2 (Fig. 9F and G). A Sankey diagram was constructed to show the connections among status, risk score, and cluster (Fig. 9H). In addition, we analyzed the ability of the clusters to predict immunotherapy efficacy. We found that the response rate of patients in Cluster 1 was slightly greater than that of patients in Cluster 2 (Fig. 9I), and the TIDE score of patients in Cluster 1 was significantly greater than that of patients in Cluster 2 (Fig. 9J), indicating that patients in Cluster 1 were more sensitive to immunotherapy.

Fig. 9
figure 9

Construction and prognosis analysis of 5-FU resistance-related clusters. A. Consensus CDF curve of unsupervised cluster analysis. B. Distribution of the cluster (Cluster 1 and Clusters) and status (dead and alive) in the TCGA-CRC cohort. C. Kaplan‒Meier survival curves of clusters in the TCGA cohort. D. Multivariable analysis of clusters in the TCGA cohort. E. Heatmap showing the correlations between clusters and clinical characteristics. F. The distribution of clusters in the high- and low-risk groups. G. The difference in the risk score between Cluster 1 and Cluster 2. H. Sankey diagram combining OS, cluster, and risk score data. I. The distribution of different responses in Cluster 1 and Cluster 2. J. The TIDE algorithm, including the TIDE score, dysfunction score, and CAF score, was used to predict patient response to immunotherapy in different clusters. ∗P < 0.05, ∗∗∗P < 0.001

Full size image

Clinical validation of the 5-FRSig

The expression levels of the 13 5-FRGs were measured by RT‒qPCR, and the risk score was calculated according to the formula for each patient in the external cohort. Spearman’s test was used to assess the correlation between the risk score and the 13 5-FRGs. As shown in Fig. 10A, the expression levels of FGF2 and CPA4 were positively correlated with the risk score, while the expression levels of ZNF607, SIX2, DNAH7, and TMEM139 were negatively correlated. The patients were divided into two subgroups according to the median risk score. The risk score significantly correlated with N stage. Patients with N0 stage tumors had significantly lower risk scores, and the low-risk group had a greater percentage of N0-stage tumors (Fig. 10B). Subsequently, we examined CD8A and CD8B expression levels in high-risk and low-risk samples, and we found no significant differences between the two groups of patients (Supplementary Fig. 3). However, the low-risk group had increased levels of cytolytic factors, including GZMA and GZMB. Moreover, the low-risk group had increased levels of immune stimulators, including HHLA2, CD28, and CD40LG (Fig. 10C). In particular, the low-risk group had significantly greater levels of CTLA4 (Fig. 10D). These data demonstrated that the low-risk group had significantly greater antitumor immune function, which was consistent with the better outcome and the predicted greater response rate to immunotherapy in this group.

Fig. 10
figure 10

Clinical validation of the 5-FU resistance-related signature. A. Correlations between the risk score and the expression of thirteen genes in tumor samples. B. The difference and distribution of risk scores in different clinical stages. C. Expression of immunomodulators in high- and low-risk groups in tumor samples. D. Expression of four key immune checkpoint molecules in tumor samples from the high- and low-risk groups. ∗P < 0.05, ∗∗P < 0.01

Full size image

scRNA-seq analysis of the 5-FRSig

The scRNA-seq cohort GSE178318 was used to conduct scRNA-seq analysis of the 5-FRSig. After t-SNE reduction and cell annotation, we obtained nine cell clusters, including B cells, CAFs, cancer cells, endothelial cells, mast cells, myeloid cells, NK cells, plasma cells, and T cells (Supplementary Fig. 4A). Supplementary Fig. 5A shows the proportions of different cells in each sample. The cell markers used for cell annotation are displayed in Supplementary Fig. 4B and Supplementary Fig. 5B. We also analyzed the distribution of cells in diverse types of tissues (Supplementary Fig. 4C and 4D). Next, we explored the cellular distribution of the 13 5-FRGs used to construct the signature (Supplementary Fig. 5C). Subsequently, we used two methods to analyze the distribution of risk scores in the cells and found high-risk cells mainly in the tumor cell and myeloid cell populations (Supplementary Fig. 4E). Therefore, we speculated that the poor prognosis of patients in the high-risk group was closely related to these cells.

Drug sensitivity analysis in the high- and low-risk groups

We next assessed the correlation between the signature score and drug sensitivity. The GDSC analysis showed that the low-risk group was more sensitive to chemotherapy, including camptothecin, cisplatin, fluorouracil, oxaliplatin, irinotecan and irinotecan. Similarly, the low-risk group was more sensitive to drugs targeting EGFR signaling, including gefitinib, afatinib, erlotinib, lapatinib, AZD3759 and osimertinib. The low-risk group was also more sensitive to drugs targeting ERK MAPK signaling and RTK signaling, with one or two exceptions (Fig. 11A). In addition, we screened patients in the high-risk group for sensitivity to drugs in the CTRP and PRISM databases (Fig. 11B and C). The above results demonstrate that our model has good discriminative ability for 5-FU adjuvant therapy and that the low-risk group is more sensitive to most chemotherapeutic and targeted drugs.

Fig. 11
figure 11

Drug sensitivity analysis in the high- and low-risk groups. A. Compounds in the GSDC database that target chemotherapy, EGFR signaling, ERK MAPK signaling, and RTK signaling. B. Compounds in the PRISM database. C. Compounds in the CTRP database. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001

Full size image

Candidate small molecules to reverse 5-FU resistance

The top 10 small molecules with negative connectivity with specific targets were screened out (Fig. 12A). These small molecules have the potential to reverse 5-FU resistance by inhibiting the upregulation of hub proteins. We verified the binding energy of these small molecules to the hub proteins, and the results are shown in Fig. 12B. Normally, a binding energy less than 0 indicates spontaneous binding, and the lower the binding energy is, the greater the possibility of interaction. As shown in Fig. 12B, the binding energies of the 10 potential molecules and the hub proteins were typically below − 5 kcal/mol, demonstrating their potential interaction. For each protein, the small molecule with the lowest binding energy was selected for docking visualization (Fig. 12C). In the Fig. 12C, the amino acids to which the small molecule binds are labeled, and the dotted lines show hydrogen bonds. We screened 10 small molecules that have the potential to reverse 5-FU resistance and validated the binding potential between the small molecules and the hub proteins.

Fig. 12
figure 12

Candidate small molecules that reverse 5-FU resistance. A. The structures of the top 10 compounds predicted by the CMap website. B. Heatmap displaying the binding energy of the small molecule drug and the hub protein. C. Small molecule drug docking targets with the lowest binding energy

Full size image

Discussion

5-FU is a classic chemotherapy drug that has long played a key role in the treatment of cancer, especially CRC and breast cancer [60]. Despite the remarkable achievements of 5-FU in cancer treatment, some patients develop resistance to the drug, likely due to individuals’ unique genetic and epigenetic makeup [61]. Therefore, there is an urgent need for molecular identification to guide clinical chemotherapy. In this study, we constructed a 5-FRSig model using 5-FU-resistant CRC cell lines to explore the related mechanisms and provide more evidence for the precise treatment of CRC patients.

The 5-FRSig was constructed from thirteen 5-FRGs, including ALPK3, CPA4, DNAH7, FGF2, HOXD13, NRG1, PPP1R3F, SIX2, SLC39A8, TMEM139, TNFRSF19, ZDHHC2, and ZNF607. In a preliminary study, ALPK3 was shown to be associated with metastasis in osteosarcoma patients [62]. The prognostic value of CPA4 in non-small cell lung cancer (NSCLC), pancreatic cancer, and bladder cancer has also been reported [63,64,65]. In addition, mutations in DNAH7 were found to benefit CRC patients receiving immune checkpoint inhibition therapy [66]. However, FGF2 and SIX2 have been found to promote the development of NSCLC and breast cancer [67,68,69,70]. Similarly, HOXD13 has been found to promote the malignant progression of colon cancer [71], and the overexpression of NRG1 promotes the progression of gastric cancer [72]. Interestingly, studies have shown that PPP1R3F and TNFRSF19 are associated with the prognosis of CRC patients [73, 74]. Liu et al. reported that SLC39A8 suppressed the progression of clear cell renal cell carcinoma [75]. Zhang et al [76]. reported that TMEM139 prevents NSCLC metastasis by inhibiting lysosomal degradation of E-cadherin. Moreover, it has been reported that ZDHHC2 and ZNF607 expression levels are reduced and increased, respectively, in gastric adenocarcinoma patients [77, 78].

In this study, the 5-FRSig constructed from 5-FRGs was found to be an independent prognostic factor for CRC. Our signature also allowed better patient risk stratification. Using the ROC curve to predict patient survival, we can see that the AUC for risk score is higher than that for TNM staging in the TCGA cohort. Compared with traditional TNM staging to predict the prognosis of patients, our nomogram model based on the risk score has better predictive power. Furthermore, the ROC curves for predicting the therapeutic effect of 5-FU demonstrate that all the AUC values are above 0.7, indicating that the model exhibits high specificity and sensitivity in predicting the efficacy of 5-FU. In the clinical validation of the 5-FRSig, we further demonstrated that patients with high risk scores had worse outcomes. In addition, as the development of 5-FU resistance is an important cause of cancer treatment failure, our signature can predict the responsiveness of patients to 5-FU treatment and immunotherapy well, providing a basis for the precise treatment of patients with CRC.

It has been reported that 5-FU resistance involves various complex factors, including noncoding RNA regulation, tumor stem cells, tumor cell autophagy, epigenetics, and ATP-binding protein overexpression [79,80,81,82,83]. In this study, when we explored the mechanism underlying differences in the prognosis of patients in different risk groups through pathway enrichment, the results showed many pathways associated with tumor immune pathways. We speculated that the occurrence of 5-FU resistance might be related to changes in the patients’ immune microenvironment. Further analysis of immune cell infiltration and immune regulatory factor expression showed that patients in the high-risk group presented an immunosuppressive tumor microenvironment. The immunotherapy results also demonstrated that low-risk patients were more sensitive to immunotherapy. In addition, through scRNA-seq, we found that myeloid cells had greater risk scores than other immune cells, which could provide a new target for the future study of 5-FU resistance. Overall, our study links 5-FU resistance to the immune microenvironment, providing additional evidence for a combination of chemotherapy and immunotherapy in patients with CRC.

The results showed that our 5-FRSig has good potential for diagnosing and treating CRC. Nonetheless, our research still has certain limitations. This study provides a solid theoretical basis for subsequent research. We obtained many novel research results through in-depth data mining. Although we used our own clinical cohort for validation, we lack sufficient clinical data to validate the ability of the 5-FRsig to predict the response to chemotherapy and immunotherapy in patients with CRC, and further studies are needed. Furthermore, the TCGA database lacks patient treatment information, preventing direct inspection of the signature’s predictive ability for 5-FU efficacy using TCGA data. Instead, the efficacy of the signature for predicting 5-FU treatment outcomes was validated using the GEO dataset, which introduces a limitation to our results. Finally, the inclusion of an internal cohort, including a combination of chemotherapy and immunotherapy, to validate the hypotheses of this study will be the focus of future research. Although we conducted a comprehensive analysis of multiple independent cohorts and obtained some clinically promising conclusions, the molecular mechanisms of the 5-FRSig in CRC prognosis, chemotherapy, and immunotherapy need to be further validated in vivo and in vitro.

Conclusion

In this study, 5-FU resistance in CRC was comprehensively analyzed using various methods, and a novel 5-FRSig was successfully constructed. This signature could be used for risk stratification, prognosis prediction, 5-FU sensitivity prediction, and immunotherapy prediction in CRC patients. In addition, it was found that the underlying mechanism was related to tumor immune pathways. Finally, we used drug sensitivity analysis and molecular docking technology to explore 10 suitable drugs for CRC patients. This study provided a new perspective on 5-FU resistance in CRC patients and a theoretical basis for improvements in chemotherapy, immunotherapy, targeted therapy, combination therapy, and individualized antitumor therapy.