Transcriptomic correlates of cell cycle checkpoints with distinct prognosis, molecular characteristics, immunological regulation, and therapeutic response in colorectal adenocarcinoma

Backgrounds Colorectal adenocarcinoma (COAD), accounting for the most common subtype of colorectal cancer (CRC), is a kind of malignant digestive tumor. Some cell cycle checkpoints (CCCs) have been found to contribute to CRC progression, whereas the functional roles of a lot of CCCs, especially the integrated role of checkpoint mechanism in the cell cycle, remain unclear. Materials and methods The Genomic Data Commons (GDC) The Cancer Genome Atlas (TCGA) COAD cohort was retrieved as the training dataset, and GSE24551 and GSE29623 were downloaded from Gene Expression Omnibus (GEO) as the validation datasets. A total of 209 CCC-related genes were derived from the Gene Ontology Consortium and were subsequently enrolled in the univariate, multivariate, and least absolute shrinkage and selection operator (LASSO) Cox regression analyses, finally defining a CCC signature. Cell proliferation and Transwell assay analyses were utilized to evaluate the functional roles of signature-related CCCs. The underlying CCC signature, molecular characteristics, immune-related features, and therapeutic response were finally estimated. The Genomics of Drug Sensitivity in Cancer (GDSC) database was employed for the evaluation of chemotherapeutic responses. Results The aberrant gene expression of CCCs greatly contributed to COAD development and progression. Univariate Cox regression analysis identified 27 CCC-related genes significantly affecting the overall survival (OS) of COAD patients; subsequently, LASSO analysis determined a novel CCC signature. Noticeably, CDK5RAP2, MAD1L1, NBN, RGCC, and ZNF207 were first identified to be correlated with the prognosis of COAD, and it was proven that all of them were significantly correlated with the proliferation and invasion of HCT116 and SW480 cells. In TCGA COAD cohort, CCC signature robustly stratified COAD patients into high and low CCC score groups (median OS: 57.24 months vs. unreached, p< 0.0001), simultaneously, with the good AUC values for OS prediction at 1, 2, and 3 years were 0.74, 0.78, and 0.77. Furthermore, the prognostic capacity of the CCC signature was verified in the GSE24551 and GSE29623 datasets, and the CCC signature was independent of clinical features. Moreover, a higher CCC score always indicated worse OS, regardless of clinical features, histological subtypes, or molecular subgroups. Intriguingly, functional enrichment analysis confirmed the CCC score was markedly associated with extracellular, matrix and immune (chemokine)-related signaling, cell cycle-related signaling, and metabolisms. Impressively, a higher CCC score was positively correlated with a majority of chemokines, receptors, immunostimulators, and anticancer immunity, indicating a relatively immune-promoting microenvironment. In addition, GSE173839, GSE25066, GSE41998, and GSE194040 dataset analyses of the underlying CCC signature suggested that durvalumab with olaparib and paclitaxel, taxane-anthracycline chemotherapy, neoadjuvant cyclophosphamide/doxorubicin with ixabepilone or paclitaxel, and immunotherapeutic strategies might be suitable for COAD patients with higher CCC score. Eventually, the GDSC database analysis showed that lower CCC scores were likely to be more sensitive to 5-fluorouracil, bosutinib, gemcitabine, gefitinib, methotrexate, mitomycin C, and temozolomide, while patients with higher CCC score seemed to have a higher level of sensitivity to bortezomib and elesclomol. Conclusion The novel CCC signature exhibited a good ability for prognosis prediction for COAD patients, and the CCC score was found to be highly correlated with molecular features, immune-related characteristics, and therapeutic responses, which would greatly promote clinical management and precision medicine for COAD.


Introduction
Colorectal cancer (CRC), one of the most common digestive malignancies, is still contributing a heavy burden to human health with a high mortality rate, with almost a million new cases annually all over the world (1).Colorectal adenocarcinoma (COAD) is the most prevalent subtype of CRC, and most patients are diagnosed in advanced stages.In the past decades, colonoscopy, flexible sigmoidoscopy, and computed tomography colonography have improved CRC screening and prevention (2).Meanwhile, the classification of tumor node metastasis stage (T, N, M stage) and clinical stage (I, II, III, IV stage) established by the Union for International Cancer Control and American Joint Commission on Cancer (UICC and AJCC) provided guidelines for the selection of the therapeutic regimens in CRC (3,4).However, within CRC patients with the same clinical stage, their prognosis was very distinct, which caused tremendous difficulties in CRC management (5).Classic criteria, such as TNM and clinical staging system, have limitations in CRC management.Currently, we need to develop more robust risk stratification tools to guide clinicians to tailor treatment interventions.
With the rapid development of next-generation sequencing (NGS), comprehensive genomic profiling of CRC has revealed that genetic (KRAS, TP53, PIK3CA, and APC) and epigenetic (hypomethylated CEP55, FOXD3, FOXF2, GNAO1, GRIA4, and KCNA5) alterations potentially contribute to the colorectal carcinogenesis (6,7).However, the genomic landscape is insufficient to demonstrate the suffused changes in gene expressions and alterations of cellular function (8,9).In tumor tissues, genetic and epigenetic alterations could directly cause the disturbed gene expression, possibly resulting in tumor cell invasiveness.Integrated omics studies confirmed that gene regulatory networks or protein-protein interaction networks representing biological processes have a better ability to predict CRC tumorigenesis and progression (10) than a single gene or protein.Quantified risk scores based on multigene expression have become promising diagnostic and prognostic biomarkers with great clinical value, which have significantly high accuracy and efficiency in predicting the diagnosis and prognosis of CRC patients, even response to therapeutics and risk of recurrence, etc.The mRNAbased signatures, such as Veridex (11), OncotypeDX (12), Coloprint (13), and GeneFx Colon (14) demonstrate their applicability in predicting survival outcomes of CRC patients, and these prognostic models have been validated in a separate study (15).Furthermore, the consensus molecular subtypes (CMSs: CMS1, CMS2, CMS3, and CMS4) classifier based on comprehensive transcriptional profiling could divide CRC patients into four CMSs with distinguishing clinical features (16), and this model has been shown to be extremely effective in predicting therapeutic responses that help clinical decision making in CRC (17).While the stroma plays an important role in patient stratification and prognosis, the accuracy of prognosis prediction via CMS is negatively affected due to stromal-derived intratumor heterogeneity (18).Compared to the stromal-dependent model CMS, CRC intrinsic gene expression signatures avoid misclassification of CRC patients (19).However, these biomarkers are predominantly based on TNM and/or microsatellite instability (MSI) stratification systems, gradually reducing their values to risk evaluation in clinical practices (20).Consequently, novel and robust predictive biomarkers are urgently needed to assist clinicians in better managing CRC patients and proposing personalized medicine.
Cancer cells, a group of cells dividing continuously and excessively, undergo aberrant cell cycle progression.Recent works have shown strong evidence that the disruption of cell cycle control plays an irreplaceable role in continuous cell division (21).During different cell cycle phases in eukaryotic cells, checkpoints, including DNA damage checkpoints, DNA replication stress checkpoints, and mitotic checkpoints, function to prevent DNA damage and chromosome missegregation (22).In brief, cell cycle checkpoints (CCCs) are responsible for inspecting the cell cycle to ensure the quality of DNA replication and chromosome distribution.For example, defects in checkpoint regulatory components, such as ataxia telangiectasia mutated (ATM), cause decreased efficiency of DNA repair and increased genomic instability widely in tumors (23).DNA damage checkpoint kinase CHK1 regulates DNA replication, phase transition, and mitosis, and the elevated expression of CHK1 is found to be significantly associated with prognosis, recurrence, and even therapy resistance in acute myeloid leukemia, hepatocellular carcinoma, breast carcinoma, and colorectal carcinoma (24).The dysregulated spindle checkpoint gene MAD2 is involved in colorectal carcinogenesis, and the overexpressed level of MAD2 protein is more frequently observed in CRC patients with lymph node metastasis (25,26).Moreover, the overexpression of spindle assembly checkpoints, including BUBR1, MAD1, MAD2, MPS1, and CDC20, has been reported in several cancer types.While the overexpression of MAD2 can lead to the hyperactivation of checkpoint (27), the expression of MAD1 has the opposite effect (28).Although the role of some cell cycle checkpoints in cancer progression has been described in detail, current studies focus on single checkpoints.Nevertheless, the prognostic significance of integrative transcriptional characterization of CCCs in CRC remains elusive.Moreover, the functional role of a lot of CCCs in CRC remains to be further investigated.
In the present study, comprehensive gene expression profiles of COAD patients with clinical information were derived from TCGA, which was used to explore the association between the expression of checkpoint mechanisms in the cell cycle and the OS of COAD patients.The Least absolute shrinkage and selection operator (LASSO) Cox regression analysis was conducted to develop a CCC gene-based signature and a related CCC score system, of which the prognostic ability was estimated by the AUC value and further verified in additional independent datasets from the GEO database.Subsequently, the molecular characteristics and tumor microenvironmental status were further investigated to disclose the distinct molecular mechanisms and immunological regulation associated with the CCC signature.Ultimately, the CCC signature was employed to predict the (chemo)therapeutic response, including chemotherapy, neoadjuvant therapy, frontline chemotherapy, immunotherapy, etc.The findings in the present study provided deep insights into the functional roles of cell cycle checkpoints affecting the prognosis, molecular mechanisms, and immunological regulation of COAD, which would help accelerate the development of precision medicine.

Development, evaluation, and validation of the CCC signature
The CCC-derived genes were analyzed via univariate Cox regression analysis.The LASSO Cox regression analysis was then utilized with the package "glmnet", and the optimal penalty regularization index l was selected via the 10-fold crossvalidation to prevent the overfitting effects.Eventually, OS-related CCCs were determined to construct the risk signature, and the CCC score (based on the CCC signature) of each COAD patient was calculated as the following formula: Where n is the gene number in the CCC signature, Expression j is the expression level of gene j, and Coefficient j is the coefficient value of gene j generated via multivariate Cox regression analysis.The COAD patients from TCGA cohort were divided into high and low CCC score groups according to the median CCC score, as the cutoff value.
The Kaplan-Meier (KM) curve analysis, by using the packages "survival" and "survminer", was conducted to compare the difference in patients' OS between two subgroups in training and validation cohorts.Moreover, the receiver operating characteristic (ROC) curve was performed by using the package "timeROC", and the value of area under ROC (AUC) was analyzed to evaluate the performance of CCC signature in prognosis prediction.Additionally, the correlation analysis was used to assess the relationship between CCC signature and clinical features.
Subsequently, the prognostic significance of the CCC signature was further verified in two independent validation datasets.The KM curve, the value of the AUC, and Cox regression analyses were also conducted in these additional COAD cohorts.The multivariate Cox proportional hazards regression model was utilized to confirm the independence of CCC signature and clinical features, including diagnosis age, gender, TNM stage, and clinical stage, by using the packages "rms" and "survminer".
To construct a gene overexpression system, the pCMV plasmid was transfected using Lipofect8000 (Thermo Fisher, CA, USA) to induce gene overexpression.pCMV-ZNF207-GFP and pCMV-NBN-GFP vectors were obtained from COBIOER Inc. (Nanjing, China).cDNAs for human genes ZNF207 and NBN were synthesized by Sino Biological Inc. (Beijing, China).

qRT-PCR and Western blotting analysis
The silencing efficiency of siRNA was determined by qRT-PCR analysis.Total RNA was extracted from HCT116 and SW480 cells and then reversely transcribed into cDNA with a reverse transcription kit (Applied Biosystems Inc., OK, USA).Quantified PCR was performed by applying the SYBR Green Real-Time PCR Master Mixes (Thermo Fisher, CA, USA) on an ABI7500 instrument (Applied Biosystems Inc., CA, USA).GAPDH was used as an internal control.The following primers were used for qR T -P C R an a ly s i s : C D K 5 R A P 2 f o r w a rd p r i m e r : 5′ -GGCCCCACTGAACATATCTACA-3′; CDK5RAP2 reverse primer: 5′-CACCTTCTTTCGAGCATCTTCTT-3′; MAD1L1 forward primer: 5′-TGGACTGGATATTTCTACCTCGG-3′; MAD1L1 reverse primer: 5′-CCTCACGCTCGTAGTTCCTG-3′; RGCC forward primer: 5′-CGCCACTTCCACTACGAGG-3′; RGCC reverse primer: 5′-CAGCAATGAAGGCTTCTAGCTC-3′.
The gene ZNF207-and NBN-overexpressed cell lines were testified by Western blotting analysis.The cells were lysed by using RIPA (50 mM Tris, 150 mM NaCl, 1% NPNP-40, 1% sodium deoxycholate, 0.05% SDS, pH = 7.4).The quantification of proteins was performed by using the BCA Kit (Bio-Rad Inc., CA, USA).Subsequently, SDS-polyacrylamide gel electrophoresis was performed and then transferred onto polyvinylidene difluoride, followed by blocking with 5% BSA, and incubated overnight at 4°w ith the first antibody (anti-GFP, 1:5,000).The next day, protein samples were incubated with a secondary antibody for 1 h, followed by visualization.

Cell proliferation and Transwell assay
HCT116 or SW480 cells were pretreated with siRNA or vectors.Cell proliferation level was estimated by using the Cell Counting Kit-8, namely CCK-8 assay (Beyotime, Shanghai, China), according to the manufacturer's instructions.Subsequently, cell viability was performed using a microplate reader at 24, 48, and 72 h with the absorbance at 450 nm.In this experiment, HCT116 and SW480 cells were analyzed and repeated independently three times.A Transwell assay was conducted to evaluate the invasion rate of HCT116 and SW480 cells; 5 × 10 4 cells were seeded in a Transwell chamber (8 mm, Thermo Fisher, CA, USA) containing 300 mL of culture medium (10% FBS).Subsequently, 1 mL of FBS-free culture medium was added to the lower chamber (a 24-well plate); 24 h later, the chamber was fixed with paraformaldehyde and stained with crystal violet.Cells were visualized, and cell numbers were counted under the microscope (Leica, Wetzlar, Germany).

Exploration of CCC signature in distinct molecular subtypes
Compared to wild types, it was generally believed that KRAS, NRAS, or BRAF (KRAS/NARS/BRAF) had a great influence on prognosis in CRC patients with mutated APC or TP53 (41).Therefore, according to the status of these genes, in this study, COAD patients could be assigned to the various mutant subtypes.Additionally, the MSI (highly associated with deficient mismatch repair) stratification system, defined by the MSI sensor score in this study, could divide COAD patients into MSI-H (MSI sensor score ≥ 10), MSI-L, and MSS (MSI sensor score< 10) subgroups.In addition, as described before, CMS subtyping is now widely accepted and reliable for CRC.The CMS classifier was defined and calculated with reference to the previously reported (16).

Molecular characteristics associated with CCC signature
The genomic alteration profiles for each COAD patient from the TCGA cohort, shown in the oncoprint plots, between high and low CCC score groups were investigated by using the package "maftools" to illustrate a variety of alteration events (substitution, frameshift, multi-hit, etc.) of the top 20 prevalently altered genes.The Kyoto Encyclopedia of Genes and Genomes (KEGG, www.kegg.jp/(42),) pathway enrichment analysis was conducted using the "clusterProfiler" package, and the significantly enriched signaling pathways were annotated in KEGG and visualized by utilizing the "ggplot2" package.HALLMARK gene set enrichment analysis (GSEA), based on GSEA v.3.0, was also performed to explore the enriched pathways correlated with the CCC signature, and the enrichment score (ES) would be further normalized for each involved gene set.

Immune characteristics and anti-cancer immunity analysis
To explore the tumor microenvironment (TME) associated with CCC score, the previously described ESTIMATE algorithm (43) was employed to assess the stromal and immune microenvironments, and CIBERSORT algorithm analysis (44) deconvoluting the expression profile (containing 547 genes) of immune-related cells was conducted to infer the difference in the proportion of 22 tumor-infiltrated lymphocytes between high and low CCC score groups, respectively.Furthermore, several indexes, including immunomodulator gene expression, tumor immune cell infiltration, cancer immunity cycle, and inhibitory immune checkpoint-related gene expression, were employed for the evaluation of the immune-related features, anticancer immunity, and microenvironmental status in COAD.A total of 122 immunomodulators (chemokines, paired receptors, MHC molecules, and immunostimulators), which had been reported previously, were collected to estimate the immunomodulation of TME (45).The cancer immunity cycle was reviewed to represent anticancer immunity, and it contained seven steps: step 1: release of cancer cell antigens; step 2: cancer antigen presentation; step 3: priming and activation; step 4: trafficking of immune cells to cancer cells; step 5: immune cell infiltration into the tumor; step 6: recognition of cancer cells by T cell; and step 7: killing of cancer cell (46).The activities of these steps were performed using a single sample gene set enrichment analysis (ssGSEA), based on the transcriptomic data of each patient sample (47).To decrease the errors, more algorithms were performed to assess the infiltrating levels of tumor immune cells (mainly CD8+, macrophage, dendritic cells, natural killer (NK) cells, and Th1 cells), including the TIMER, CIBERSORT, QUANTISEQ, MCPCOUNTER, XCELL, and EPIC algorithms.The profile of inhibitory immune checkpoints was obtained from the study of Auslander (48).

Evaluation of therapeutic response underlying CCC signature
Initially, the GSE153412 dataset about the therapeutic response evaluation of COAD cancer cell lines (HCT116 and HT29) and the GSE196263 dataset of CMS4 subtype colon cancer patients receiving imatinib treatment were downloaded to explore the potential predictive role of CCC score in the therapeutic response for COAD patients.Moreover, the GSE173839 (HER2-negative stage II/III breast cancer patients on the durvalumab with olaparib and paclitaxel arm: 29 responders vs. 42 nonresponders), GSE25066 (breast cancer patients treated with taxane-anthracycline chemotherapy: 99 patients with pathological complete response (pCR) vs. 389 patients with residual disease (RD)), and GSE41998 (early-stage breast cancer patients receiving neoadjuvant cyclophosphamide/doxorubicin with ixabepilone or paclitaxel: 69 responders vs. 184 responders) datasets were also downloaded to further evaluate the predictive ability of the CCC signature in the therapeutic response.The dataset GSE194040 (the I-SPY2-990 mRNA/RPPA Data Resource) contained gene expression and clinical data for 987 patients from 10 arms of the neoadjuvant I-SPY2 TRIAL for aggressive early-stage breast cancer.The GSE48276 dataset included 128 muscle-invasive bladder cancer patients receiving frontline chemotherapy.In addition, the Genomics of Drug Sensitivity in Cancer (GDSC, https:// www.cancerrxgene.org/)database was usually employed to predict chemotherapeutic response; therefore, the GDSC database was further utilized by the "pRRophetic" package.The value of half maximal inhibitory concentration (IC50) was used to assess the chemotherapeutic response.

Statistical analysis
The statistical data analyzed in this study were all performed in R Studio (https://rstudio.com/).The univariate and multivariate analyses investigating the prognosis-related risk factor(s) were conducted via the Cox proportional hazards regression model.The exploration of the survival difference of OS via the KM curves was based on a log-rank test.The Chi-square test, Fisher's test, Wilcoxon-rank test, and Kruskal-Wallis H test were utilized to conduct the correlation analysis between the CCC signature and clinical features.The (adjusted) p-value< 0.05 was considered statistically significant.

Dysregulation of CCCs in the development and progression of COAD
Initially, the DEGs analysis between COAD and normal samples showed that there were 3,890 DEGs in total, including 1,871 upregulated and 2,019 downregulated genes in COAD samples, with a statistically significant difference (|log 2 (FC)|>1, adjust p-value< 0.05, Figure 1A).The functional enrichment analysis (Figures 1B, C) suggested that those DEGs were mainly enriched in the signaling pathway of the cell cycle, illustrating that cell cycle-related dysregulation had a great influence on the development of COAD.Based on the expression profiling of 209 CCCs, COAD patients were divided into two clusters via the unsupervised hierarchy clustering analysis (Figure 1D).Moreover, the principal component analysis (PCA) confirmed that the expression profiling of CCCs apparently distinguished these two clusters (Figure 1E).Intriguingly, these two clusters had the distinct OS (median OS of cluster 1 vs. cluster 2: 61.84 months vs. unreached, p = 0.011, Figure 1F), suggesting that dysregulation of CCC expression was markedly associated with the prognosis of COAD patients.

Establishment of CCC gene-based signature in COAD
The univariate Cox regression analysis was conducted to classify the prognostic association between the expression of CCCs and the OS of COAD patients (Table 1), and eventually, 27 CCC genes were identified to be significantly correlated with COAD patients' OS (p< 0.05, Figure 2A).Subsequently, these survivalrelated genes were enrolled into the LASSO Cox regression analysis (Figure 2B), and a 20-CCC gene signature was constructed (Supplementary Table S1).According to the CCC signature scoring system, COAD patients from TCGA cohort were stratified into a high CCC score group and a low CCC score group based on the median CCC score (Figure 2C).Patients with higher CCC scores had inferior OS compared to those cases from the low CCC score group (median OS: 57.24 months vs. unreached, p< 0.0001, Figure 2D).Moreover, the AUC values of CCC signature at 1, 2, and 3 years were 0.74, 0.78, and 0.77, respectively (Figure 2E).The decision curve analyses (DCA) further revealed that CCC signature had a better net benefit for OS than TNM staging or clinical staging (Figures 2F-H), suggesting that CCC signature was a robust prognosis prediction model.

CCC signature-related genes correlated with clinical features, immune characteristics, and their IHC staining
Obviously, it was observed that BRSK1, CDK5RAP2, RGCC, and MAD1L1 expression was negatively correlated with TNM stages and clinical stages; in reverse, BRCA1, ZNF207, PLK1, ZW10, CDKN1B, CLOCK, CCNB1, BUB1, CNOT6, CNOT6L, NBN, SETMAR, ORC1, MAD2L1, CNOT7, and CDC25C positively associated with advanced stages in COAD (Figure 3A).Of note, the expression of a majority of CCC signature-related genes was closely correlated with stromal and immune microenvironment characteristics; meanwhile, the expression of some certain genes was significantly correlated with immune profiles, including TMB, TIDE score, and 22 infiltrating immune cell types (Figure 3B).The representative IHC staining of BRCA1, BRSK1, CCNB1, CDC25C, CDK5RAP2, CDKN1B, CLOCK, CNOT6, CNOT6L, MAD1L1, ZW10, MAD2L1, NBN, ORC1, PLK1, RGCC, ZNF207, and ZW10 was shown in Figure 3C, while BUB1, CNOT7, and SETMAR protein expression in COAD was not analyzed via the investigation of the HPA database.By further statistical analysis, it was identified that the IHC staining of NBN and ZNF207 protein expression was particularly strong (Figure 3D).Except for BRSK1, CNOT6, and CNOT6L, the IHC staining of most of the CCC signature-related genes was moderate.

Functional verification of signature-related genes in COAD-derived cell lines
In the foundation of the literature review, five genes involved in the CCC signature were selected, including the identified risk factors of CDK5RAP2, MAD1L1, and RGCC and the protective factors of NBN and ZNF207, whose functions were scarcely known in the COAD-derived cell lines.Initially, qRT-PCR analysis confirmed the RNA interference by siRNA targeting CDK5RAP2, MAD1L1, and RGCC (p< 0.05, Figures 4A-C) in the COAD-derived cancer cell lines of HCT116 and SW480.Meanwhile, Western blotting analysis validated the success of the construction of NBN-and ZNF207-overexpressed pCMV plasmid systems in the HCT116 and SW480 cell lines (Figures 4D, E).Impressively, Transwell assay analysis demonstrated that the downregulation of CDK5RAP2, MAD1L1, and RGCC reduced the cell migration rates of both HCT116 and SW480 cell lines (p< 0.05, Figure 4F); moreover, NBN-and ZNF207-overexpressed HCT116 and SW480 cell lines also exhibited the decline of cell migration rates (p< 0.05, Figure 4G).In addition, it was further observed that the downregulation of CDK5RAP2, MAD1L1, and RGCC decreased cell proliferation levels of both HCT116 and SW480 cell lines (p< 0.05, Figure 4H).Furthermore, cell proliferation levels were inhibited in the NBN-and ZNF207-overexpressed HCT116 and SW480 cell lines (p< 0.01, Figure 4I).

CCC signature in different clinical stages and histologic subtypes
The prognostic value of the CCC signature in different UICC/ AJCC clinical stages was further investigated.Except for COAD patients in clinical stage I (median OS in the high CCC score group vs. low CCC score group: unreached vs. unreached, p = 0.35, Figure 5A), in clinical stages II, III, and IV patients in high CCC score group had worse OS (median OS in stage II: 70.16 months vs. unreached, p = 0.0016; in stage III: 31.96months vs. unreached, p = 0.0015; in stage IV: 5.88 months vs. 61.84months, p< 0.0001, Figures 5B-D).In parallel, the above results suggested that the prognostic value of the CCC signature was more robust for patients with advanced stages, and it was further validated by the ROC curve analysis (Figures 5E-H).It was revealed that CCC signature had good prognosis prediction for COAD patients in clinical stage II (AUC at 1, 2, and 3 years were 0.72, 0.70, and 0.70, respectively, Figure 5F), as well as for patients in clinical stage III (AUC at 1, 2, and 3 years were 0.63, 0.73, and 0.76, respectively, Figure 5G).Of note, CCC signature had the best prognosis prediction for COAD patients in clinical stage IV, with the AUC values for OS at 1, 2, and 3 years were 0.85, 0.82, and 0.76, respectively (Figure 5H).
Generally, COAD patients from TCGA cohort can be divided into two main types: colon and rectum.As expected, high CCC score group had worse OS both in the colon (median OS: 60.79 months vs. unreached, p< 0.0001, Figure 5I) and rectum cancer patients (median OS: 57.24 months vs. unreached, p< 0.001, Figure 5J).According to the colon classification of anatomic sites, the left-side colon part (descending colon, rectosigmoid junction, sigmoid colon, splenic flexure) and right-side colon part (ascending colon, cecum, hepatic flexure, transverse colon) were included in TCGA cohort.Notably, regardless of patients with left-or right-side colon cancer, higher CCC score still indicated a significantly inferior OS (median OS for left-side colon cancer: 56.25 months vs. unreached, p< 0.001, Figure 5K; for right-side colon cancer: 26.47 months vs. unreached, p< 0.001, Figure 5L).The above results showed the good performance of CCC signature predicting prognosis in different histological or anatomical subtypes.

Validation of CCC signature, independent from clinical features
CCC signature was further assessed in two independent cohorts from the GEO database to validate its robustness.In the GSE24551 dataset, the KM curve showed that patients with higher CCC scores had a worse OS (median OS: 57.00 months vs. unreached, p< 0.01, Figure 6A), and the AUC values of CCC signature for OS at 1, 2, and 3 years were 0.70, 0.68, and 0.62, respectively (Figure 6B).Likewise, in the GSE29623 dataset, there was a lower OS probability of patients from the high CCC score group (median OS: 54.90 months vs. unreached, p< 0.05, Figure 6C), and the AUC values of CCC signature for OS at 1, 2, and 3 years were 0.72, 0.67, and 0.60, respectively (Figure 6D).In addition, the prognostic capability of CCC signature for advanced COAD patients was also verified in these two validation cohorts.Regarding COAD patients in clinical stage III from the GSE24551 dataset, it could be obviously seen that patients in the high CCC score group also had a shorter OS (median OS: 13.02 months vs. unreached, p< 0.0001, Figure 6E), and the AUC values for OS at 1, 2, and 3 years were 0.66, 0.68, and 0.65, respectively (Figure 6F).Similarly, for patients in clinical stage III from GSE29623 dataset, higher CCC score indicated poorer OS (median OS: 38.72 months vs. unreached, p< 0.05, Figure 6G), and the AUC values for OS at 1, 2, and 3 years were 0.69, 0.79, and 0.79, respectively (Figure 6H).More importantly, the univariate and multivariate Cox proportional hazards regression analyses (Table 2) further identified that the CCC signature was independent of clinical features (p< 0.001) in predicting prognosis for COAD patients.

Prognostic value of CCC signature in distinct molecular subtypes
Subsequently, the prognostic value of the CCC signature in different mutant subgroups was explored.In APC-mutant, TP53mutant, KRAS/NARS/BRAF-mutant subgroups and along with corresponding wild subgroups, respectively, it was also found that higher CCC score was always correlated with the poorer clinical outcomes (median OS in APC-wild subgroup: 67.29 months vs. 70 S3G, H).Regarding the CMS molecular subtypes (CMS1, N = 29; CMS2, N = 68; CMS3, N = 58; CMS4, N = 114), it was initially found that patients in the CMS3 subgroup had the best OS (median OS of CMS4 vs. CMS3 vs. CMS2 vs. CMS1: 65.85 months vs. unreached vs. 49.41months, p< 0.05, Supplementary Figure S4A).Accordingly, a higher proportion of CMS3 subtype patients c o u l d b e a s s i g n e d i n t o t h e l o w C C C s c o r e g r o u p (Supplementary Figure S4B), and impressively, CMS3 subgroup had the significantly lowest CCC score than those of other subgroups (p< 0.0001, Supplementary Figure S4C).Even on the CMS stratification, CCC signature still performed well in the prognosis prediction, and the KM curve analysis demonstrated that the high CCC score group always had the worse OS in each CMS subgroup (median OS in CMS1: 38.07 months vs. unreached, p< 0.01; in CMS2: 41.36 months vs. unreached, p< 0.05; in CMS3: unreached vs. unreached, p = 0.097; in CMS4: 22.42 months vs. 65.85months, p< 0.001, Supplementary Figures S4D-G).

Molecular characteristics of different CCC score subgroups
Subsequently, genomic alteration profiling was conducted to uncover the genomic difference between high and low CCC score groups in TCGA cohort.As shown, nearly all COAD patients had at least one genomic mutation event, consistent with previous studies, no matter whether in a high CCC score group or a low CCC score group (total mutation frequency in high CCC score group vs. in low CCC score group: 99.42% vs. 100%).The genomic mutation profile of the top 20 altered genes demonstrated that APC, TP53, TTN, and KRAS were the most prevalently altered genes in both groups (Figures 7A, B).By further statistical analysis, it was identified that only TP53 was prevalently altered in the high CCC score group (p< 0.05, Figure 7C); whereas, FAT3, NBEA, ATM, ANK3, AHNAK2, TENM4, ARID1A, CSMD2, and LRRK2 were more frequently altered in the low CCC score group (p< 0.05, Figure 7C).Furthermore, the lollipop plot further displayed that the altered site p.R273C was more prevalent in the high CCC score group (p< 0.05, Figure 7D).Additionally, HALLMARK and KEGG pathway enrichment analyses based on the DEGs between high- and low-risk groups (Figure 8) revealed that extracellular matrix (ECM)-related signaling, angiogenesis, cell cycle-related signaling pathways, immune (chemokine)-related signaling pathways, valine leucine and isoleucine degradation, citrate cycle_TCA cycle, butanoate and propanoate metabolisms were significantly enriched.

Immune features and anticancer immunity underlying CCC signature
Subsequently, the immunological regulation underlying the CCC signature was deeply investigated in TCGA COAD cohort.As demonstrated, the expression of a majority of chemokines and paired receptors (CCL2, CCL14, CCL16, CCL18, CCL21, CCL23, CCL26, CCL28, CXCL12, CX3CL1, CCR6, CCR10, and CXCR5) was significantly higher in the high CCC score group when compared to the low CCC score group (p< 0.05, Figure 9A).Most immunostimulators (TNFSF4, TNFSF13, TNFRSF4, TNFRSF8, TNFRSF13, TNFRSF17, TNFRSF18, TNFRSF25, TNFRSF13C, IL6, ENTPD1, KLRK1, CD40, and CD276) were also markedly upregulated in the high CCC score group.Whereas, it was identified that only TAPBP (MHC-associated gene) expression was significantly elevated in the high CCC score group.In short, COAD patients in the high CCC score group might have an immune-promoting microenvironment compared to those in the low CCC score group.Correspondingly, the high CCC score group had a higher level of most of the anticancer immunity cycle activities, including step 1 of cancer cell antigen release, step 3 of priming and activation, step 4 (trafficking to tumors) of CD4 T cell, dendritic cell, eosinophil, macrophage, and Th17 cell recruiting, and step 5 of infiltration of immune cells into tumors (p< 0.05, Figure 9B).By different algorithms, including TIMER, CIBERSORT, QUANTISEQ, XCELL, TISIDB, TIP, and MCPCOUNTER (Figure 9C), it was found that there was nearly no correlation between the CCC score and the infiltrating levels of macrophages, dendritic cells, NK cells, and CD8+ T cells.However, some effector genes (such as CD8 T cell: FLT3LG; dendritic cell: SIGLEC1, SLAMF8, and SLC15A3; macrophage: C1QA, MARCO, and MMP8; natural killing cell: SPON2) expression was significantly upregulated in the high CCC score group (Figure 9D).In addition, it was further identified that the CCC score was not closely associated with PD-L1, PD-1, CTLA-4, LAG-3, TIM-3, TIGHT, IDO1, and other inhibitory immune checkpoints at the transcriptomic level (Figure 9E).

CCC signature in the prediction of therapeutic response
Eventually, the predictive role of the CCC signature in the therapeutic response was investigated.In the GSE153412 dataset, including HCT116 & HT29 cell lines treated by the chemotherapy or radiochemotherapy, it was identified that the low CCC score group seemed to be likely more sensitive to 5-fluorouracil (5-FU), uracil, or radiochemotherapy of 5-FU and irradiation (p< 0.01, Supplementary Figure S5A).In the GSE196263 dataset, five colon cancer patients belonging to the CMS4 subtype had a significantly reduced CCC score after imatinib treatment (p< 0.01, Supplementary Figure S5B).In the GSE173839 dataset of HER2negative stage II/III breast cancer patients on the durvalumab with olaparib and paclitaxel arm, it seemed that higher CCC score was correlated with pCR (p< 0.01, Figure 10A).Impressively, the AUC value of the CCC score for response prediction was 0.66 (Figure 10B).In another breast cancer dataset of GSE25066, a higher CCC score was also correlated with the pCR of patients receiving taxane-anthracycline chemotherapy (p< 0.01, Figure 10C), and the AUC value for response prediction was 0.60 (Figure 10D).Moreover, it was further found that the CCC score was positively correlated with the pCR (p< 0.05, Figure 10E) of neoadjuvant cyclophosphamide/doxorubicin with ixabepilone or paclitaxel for early-stage breast cancer patients in the GSE41998 dataset, and the AUC value for response prediction was 0.61 (Figure 10F).A recent study of the I-SPY2-990 mRNA/RPPA Data Resource containing gene expression and clinical data for 987 patients from 10 arms of the neoadjuvant I-SPY2 TRIAL for aggressive early-stage breast cancer (GSE194040) found that, interestingly, a higher CCC score was positively correlated with the pCR of neoadjuvant chemotherapy (p< 0.0001, Figure 10G), and the AUC value for response prediction was 0.60 (Figure 10H).Whereas, in the GSE48276 dataset of bladder cancer patients, those cases had a significantly decreased CCC score after receiving the frontline chemotherapy (Figure 10I), and the AUC value was 0.74 (Figure 10J).

Evaluation of chemotherapeutic response underlying CCC signature
Through the analysis of the GDSC database, the predictive value of the CCC signature for chemotherapy response was estimated.Comparison analysis of the IC50 value revealed that patients with a lower CCC score were likely to be more sensitive to 5-fluorouracil, bosutinib, gemcitabine, gefitinib, methotrexate, mitomycin C, and temozolomide (p< 0.01, Figure 11).Whereas, patients with higher CCC score seemed to have the higher level of sensitivity to bortezomib and elesclomol (p< 0.001, Figure 11).

Discussions
Substantial evidence shows that CRC is a kind of heterogeneous cancer, for which the prognosis and the efficacy of treatment strategies vary.Especially for most of the CRC patients with advanced/metastatic disease, chemotherapy, radiation therapy, and targeted therapy become less effective and the 5-year survival rate is notably lower (49).Molecular profiling, such as genomic profiling, transcriptomic profiling, and proteomic profiling, is now a major objective to classify heterogeneous CRC patients and guide personalized treatment (50).In our study, the mRNA expression profiling of a total of 209 CCCs was investigated, and finally, a novel CCC signature based on 20 CCC-related genes was constructed along with a scoring system that performed robustly in the prognosis prediction for COAD patients, especially for advanced patients.Also, the CCC signature also had good risk stratification in different subgroups with distinct clinical features or molecular characteristics.The differentiation analysis of biological functions and TMEs between high and low CCC score groups further revealed the underlying molecular characterization and immunological regulation in COAD, which could guide clinical decision-making.Moreover, the CCC signature was also of considerable guiding Validation of cell cycle checkpoint (CCC) signature in two additional independent datasets from the Gene Expression Omnibus (GEO) database.proliferation and Transwell assay analysis.From another perspective, it could be inferred that inhibition of CDK5RAP2, MAD1L1, and RGCC expression could suppress the proliferation, migration, and invasion of COAD cells, which might be an effective treatment strategy for COAD patients.Furthermore, according to the CCC signature scoring system, the integrative role of cell cycle checkpoint mechanisms as the risk factor in COAD was quantified, and the constructed CCC signature had a good ability of stratifying COAD patients into different risk groups in various TNM/clinical stages, histological subtypes, or molecular subtypes, allowing the molecular characterization of COAD progression.
Considerable studies have revealed that continuous cancer cells are generally driven by mutations, not only compromising cell cycle exit but also preventing apoptosis, rather than promoting boundless cell cycle progression.During processing, the checkpoint mechanisms strictly control cell cycle entry and exit and regulate cell cycle progression (22).As for reversible or irreversible DNA damage, the checkpoint mechanisms could function to initiate quiescence or induce apoptosis or senescence, respectively (65).Nevertheless, in cancer cells, compromised checkpoints could tolerate chromosome instability and aneuploidy; however, too much is harmful that could suppress cancer cell growth and promote cell death instead (66,67), and high level of genomic alterations has been identified to be correlated with the improved prognosis of cancer patients (68).Generally, DNA damage checkpoints function in response to irreversible DNA damage mostly by TP53-dependent pathways, but it should be highlighted that p53 inactivation prominently led to the suppression of cell cycle arrest and apoptosis (65).Therefore, the aberrant regulation of cell cycle checkpoints (DNA damage checkpoints) would promote uncontrolled cell cycle progression.Moreover, this study simultaneously used two cell lines, SW480 (a p53 mutated cell line) and HCT116 (a p53 wildtype cell line), to try to prove that the aberrant regulation of CCC signature-related gene(s) expression was relatively independently correlated with the proliferation, progression, and invasion of colorectal cancer cells.Unsurprisingly, our finding also revealed the higher prevalence of the TP53 mutation (72% vs. 59%) in the high CCC score group.The characterization of genomic alterations in CRC by TCGA have already revealed that APC, TP53, SMAD4, PIK3CA, and KRAS mutations were most prevalent (69), in which APC, TP53, and KRAS/NARS/BRAF mutations have been found to be highly predictive for outcomes in clinical practice (41,70).Therefore, the prognostic role of the CCC signature in different mutant subgroups was investigated, and remarkably, a lower CCC score always indicated a superior OS.Functional enrichment of the KEGG pathway underlying the CCC signature revealed that ECM-related signaling pathways, angiogenesis, cell cycle-related signaling pathways, immune (chemokine)-related signaling pathways, valine leucine and isoleucine degradation, citrate cycle_TCA cycle, butanoate, and propanoate metabolisms were enriched.Previous study has found that the downregulation of butanoate and propanoate metabolism and valine, leucine, and isoleucine degradation happened in the metastatic tissues from all sites in metastatic colon patients, and metastatic tissues from the liver and omentum had the decreased regulation of oxidative phosphorylation (71).The increased intake of branched-chain amino acids, such as valine, leucine, and isoleucine, was positively correlated with the mortality rate of CRC patients (72).Moreover, in the mouse model, it was found that mirabilite exerted a therapeutic effect on CRC, and the significantly enriched propanoate metabolism was one of the therapeutic targets (73).Furthermore, the high-throughput metabolomic analysis disclosed that an active polyphenolic honokiol (HNK) had anticancer activities mainly by regulating the citrate cycle_TCA cycle, tryptophan metabolism, and pentose phosphate pathway in the mouse model (74).Except as a diagnostic biomarker in CRC (75), the role of butanoate and related metabolism in CRC progression is unknown until now.Overall, CCC inhibitors, together with the controlled level of valine, leucine, and isoleucine, and in combination with the targeted therapeutics for propanoate and butanoate metabolisms as well as citrate cycle_TCA cycle, might help improve the prognosis of COAD patients with specific subtypes.However, the underlying mechanisms of cell cycle checkpoints regulating metabolisms are scarcely known and need to be further verified.
Frequently, immunomodulators, cancer immunity, tumor immune cell infiltration, and inhibitory immune checkpoints were used to evaluate the immunological status of TME (76).A majority of chemokines and paired receptors could stimulate the recruitment of CD8+ T cells in human cancers (45,(77)(78)(79).As shown, these downregulated chemokine/receptors among patients with lower CCC scores would reduce the activity of anticancer immunity.Impressively, patients with higher CCC scores seemed to have an immunepromoting microenvironment, owing to the significantly upregulated expression of a majority of chemokines, receptors, and immunostimulators.Comprehensively, the complex functions and interactions of these immunomodulators could be integrated and reflected by seven major steps, directly representing the anticancer immunity of tumor cells (46).As further confirmed, patients with higher CCC scores had a higher level of most of the anticancer immunity cycle activities, including step 1 of cancer cell antigen release, step 3 of priming and activation, step 4 (trafficking to tumors) of CD4 T cell, dendritic cell, eosinophil, macrophage, and Th17 cell recruiting, and step 5 of infiltration of immune cells into tumors.Indeed, an immune-promoting microenvironment should have improved the prognosis of COAD patients; however, the anticancer immunity cycle of step 6 of recognition of cancer cells by T cells and step 7 of the killing of cancer cells between high and low CCC score groups were nearly equivalent.Thus, it could be inferred that this antitumor effect in the high CCC score group might be restrained by tumor immune escape, owing to the higher level of PD-1 and/or PD-L1.In brief, the CCC score could be of guiding significance to clarifying COAD patients with an immunosuppressive or immunepromoting microenvironment, which would help predict the response of immunotherapy in COAD.
Ultimately, this study further investigated the role of CCC signature in the prediction of therapeutic response.Regarding the analysis of the GSE153412 dataset of HCT116 and HT29 cell lines treated with chemotherapy or radiochemotherapy, it was suggested that the CCC signature might distinguish a patient subgroup, with the low CCC score being likely more sensitive to chemotherapy of 5-FU/ uracil or radiochemotherapy.Notably, it was identified in the present study that COAD patients belonging to the CMS4 subtype might have improved OS after imatinib treatment because the CCC score was markedly reduced.Imatinib, as a kind of inhibitor-targeting tyrosine kinase, already exhibited its effects of inhibiting the proliferation of stromal fibroblasts and preventing colorectal metastases (80).In spite of that, more clinical trials were needed to validate the efficacy of imatinib treatment for CMS4 subtype COAD patients, while the CCC signature seemed to be a good biomarker to predict the treatment efficacy of imatinib.Furthermore, the analysis results of GSE173839, GSE25066, GSE41998, and GSE194040 datasets underlying CCC signature suggested that durvalumab with olaparib and paclitaxel, taxane-anthracycline chemotherapy, neoadjuvant cyclophosphamide/ doxorubicin with ixabepilone or paclitaxel, and immunotherapeutic strategies might also be suitable for COAD patients with a higher CCC score.Altogether, the present study recommended more potential treatment selections for COAD patients with distinct molecular characteristics.Whereas, more experimental explorations and clinical trials are urgently needed to verify the related analysis results in this research.

Conclusion
The novel CCC signature had good sensitivity and specificity as a robust prognostic model, regardless of clinical features, histological subtypes, or molecular subgroups, which could help promote the clinical management of COAD.Furthermore, the CCC signature distinguished different groups with distinct molecular features, immune-related characteristics, and therapeutic responses, which would help advance precision medicine for COAD patients.

1
FIGURE 1 Dysregulation of cell cycle checkpoints (CCCs) in the development and progression of colorectal carcinoma (COAD).(A) The differentially expressed gene (DEG) analysis between normal and tumor tissues.(B) The functional enrichment analysis is based on the DEGs between normal and tumor tissues.(C) The gene set enrichment analysis is the foundation of the HALLMARK pathways.(D) The unsupervised hierarchy clustering analysis is based on the transcriptomic profile of CCCs in the TCGA COAD cohort.(E) The principal component analysis (PCA) of two separated clusters.(F) The Kaplan-Meier curve analysis of two divided clusters.

2
FIGURE 2 Transcriptomic correlates of cell cycle checkpoints (CCCs) with the overall survival (OS) of colorectal carcinoma (COAD) and construction of a novel CCC signature.(A) The univariate Cox regression analysis identified the OS-related CCCs in COAD.(B) The least absolute shrinkage and selection operator (LASSO) Cox regression analysis for the construction of a novel CCC signature.(C) COAD patients from TCGA COAD cohort were divided into the high and low CCC score groups.(D) The Kaplan-Meier curve analysis revealed the differential OS between high and low CCC score groups.(E) The receiver operating characteristic curve analysis.The decision curve analyses (DCA) of CCC signature in predicting 1-year (F), 2-year (G), and 3-year (H) OS for COAD patients.

3
FIGURE 3 Correlation between the cell cycle checkpoints (CCC) signature score (related genes) with clinical features, immune characteristics, and immunohistochemistry (IHC) staining analysis.(A) The heatmap illustrated the correlation between CCC score (related gene expression) and clinical features.(B) The correlation analysis between signature-related CCC gene expression and immune-related characteristics, including ESTIMATE score (consisting of the immune score and stromal score), TIDE score, tumor mutational burden (TMB) level, and the infiltration level of 22 immune cells.(C) The IHC analysis of signature-related CCC genes.(D) The statistical analysis of IHC-related results for signature-related CCC genes.

4
FIGURE 4 Transwell assay analysis and cell proliferation analysis of five screened (functional role was unknown in COAD) cell cycle checkpoints (CCC) signature-related genes in the COAD-derived cell lines.The qRT-PCR analysis for the verification of the RNA interference by siRNA targeting CDK5RAP2 (A), MAD1L1 (B), and RGCC (C) in the HCT116 and SW480 cell lines.The construction of NBN-(D) and ZNF207-overexpressed (E) pCMV plasmid systems in the HCT116 and SW480 cell lines.The Transwell assay analysis demonstrated the effects of the suppressed expression of CDK5RAP2, MAD1L1, and RGCC (F) and NBN and ZNF207 overexpression (G) on the cell migration of the HCT116 and SW480 cell lines.The effects of the suppressed expression of CDK5RAP2, MAD1L1, and RGCC (H) and NBN and ZNF207 overexpression (I) on the cell proliferation of HCT116 and SW480 cell lines.*, p<0.05; **, p<0.01.

5
FIGURE 5 Clinical application of cell cycle checkpoint (CCC) signature.Based on the optimal cutoff value of the CCC score, the Kaplan-Meier curve analysis of the high and low CCC score groups among colorectal carcinoma (COAD) patients in clinical stages I (A), II (B), III (C), and IV (D).The receiver operating characteristic curve analysis of CCC signature in predicting prognosis for COAD patients in clinical stages I (E), II (F), III (G), and IV (H).The application of the CCC signature among patients with colon cancer (I) or rectum cancer (J).Regarding colon cancer patients, the Kaplan-Meier curve analysis between the high and low CCC score groups among patients with left-side (K) or right-side (L) colon cancer.
(A) The Kaplan-Meier curve analysis between the high and low CCC score groups in the GSE24551 dataset.(B) The receiver operating characteristic curve analysis of the CCC signature in predicting prognosis for COAD patients in the GSE24551 dataset.(C) The Kaplan-Meier curve analysis between the high and low CCC score groups in the GSE29623 dataset.(D) The receiver operating characteristic curve analysis of CCC signature in predicting prognosis for COAD patients in the GSE29623 dataset.(E) The application of the CCC signature for clinical stage III COAD patients from the GSE24551 dataset.(F) The receiver operating characteristic curve analysis of the CCC signature for clinical stage III COAD patients from the GSE24551 dataset.(G) The application of the CCC signature for clinical stage III COAD patients from the GSE29623 dataset.(H) The receiver operating characteristic curve analysis of the CCC signature for clinical stage III COAD patients from the GSE29623 dataset.

7 8
FIGURE 7The genomic differences between the high and low cell cycle checkpoints (CCC) score groups.The oncoprint plots exhibited the top 20 altered genes in the high (A) and low (B) CCC score groups, respectively.(C) The forest plot deciphered the prevalent genes in the high and low CCC score groups.(D) The lollipop plot showed the prevalence of altered sites in the high and low CCC score groups.

9 11
FIGURE 9 Underlying cell cycle checkpoints (CCC) signature, immune features, and anti-cancer immunity analysis.(A) The differentially expressed genes analysis of chemokines, immunostimulators, MHC molecules, and receptors between the high and low CCC score groups.(B) The comparison analysis of anticancer immunity cycle activities between the high and low CCC score groups.(C) Multiple algorithms were employed to investigate the relationship between the CCC score and the infiltration level of immune-related cells.(D) Correlation analysis between CCC score and the effector gene expression of CD8+ T cells, dendritic cells, macrophages, NK cells, and Th1 cells.(E) Correlate of CCC score with the expression of inhibitory immune checkpoints.*, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001.

TABLE 1
Clinical features of 359 patients with COAD from TCGA.

TABLE 2
Univariate and multivariate Cox proportional hazard regression analyses based on CCC score and clinical features in TCGA COAD cohort.