Prognosis prediction of stage IV colorectal cancer patients by mRNA transcriptional profile

Abstract

Background: Stage IV colorectal cancer patients with liver metastasis represent a special group of CRC patients with poor prognosis. The prognostic factors have not been investigated for stage IV CRC patients undergoing primary cancer resection but not candidates for metastasis resection.

Methods: Ninety-nine stage IV CRC patients who underwent primary cancer resection without metastasis resection were retrospectively recruited. Both whole-exome sequencing (WES) and RNA-seq were performed with frozen primary cancer tissues, using para-cancerous normal tissues as the control. Valid data were obtained from 78 patients for WES and 84 patients for RNA-seq. Univariate, multivariate Cox analyses were performed and Nomogram model was established to predict patient prognosis.

Results: The correlation between patient prognosis and clinicopathological factors, mutational status, or mRNA level changes was examined. Univariate (p = 0.0007) and subsequent multivariate analyses on clinicopathological factors showed that location (left or right) was the only independent risk factor for patient prognosis (HR = 3.63; 95% CI: 1.56-8.40, p = 0.003), while T, N, M staging, gender, race, location (rectum or colon), and pathological types were not stratifying factors. The mutational status of APC, TP53, KRAS, TTN, SYNE1, SMAD4, PIK3CA, RYR2, and BRAF did not show significant stratification in patient prognosis. RNA-seq showed that genes related to membrane function, ion channels, transporters, or receptors were among those with significant mRNA level alterations. Univariate analysis identified 97 genes with significantly altered mRNA levels, while NEUROD1, FGF18, SFTA2, PLAC1, SAA2, DSCAML1, and OTOP3 were significant in multivariate analysis. A risk model was established to stratify the prognosis of stage IV CRC patients. A Nomogram model was established with these genes to predict individual patient prognosis.

Conclusions: A panel of eight genes with significant mRNA level alterations was capable of predicting the prognosis and risk of the specific patient group. Future prospective study is needed to validate the model.

Keywords: RNA-seq; WES; colorectal cancer; expression; mutation; prognosis; transcription.

FIGURE 1

Flowchart for data analysis and model establishment in this study. Both sequencing data in WES, RNA‐seq, and clinicopathological data from patients were collected. Location (left/right) was found to be the only significant risk factor among all clinicopathological factors. Univariate and multivariate Cox analyses were performed with mutational data, along with the prognosis data, while the results revealed no significant mutations for prognosis stratification. In contrast, RNA‐seq data reveal several significant genes in multivariate Cox analysis and a prognostic model (Nomogram model) was established based on the findings.

FIGURE 2

Results for multivariate analysis of clinicopathological factors in this study. Location (left/right) was found to be the independent risk factors for patient prognosis (p = 0.003).

FIGURE 3

Mutational landscape and characteristics of the primary cancer tissue of stage IV CRC patients in this study. (A) The mutational landscape of 78 patients with valid sequencing data for analysis. APC, TP53, and KRAS were among the genes with highest mutational rate. Mutation types and pathological stages are labeled by colors as indicated. (B) The mutational characteristics, including the variant classification, types, base changes, variants per sample, and top mutated genes are shown in individual panels.

FIGURE 4

Kaplan–Meier analysis by mutational status. The patient survival was stratified by mutational status (wild type or mutation) of top mutated genes, as indicated in each panel. No significant stratification of prognosis was found among these genes, as indicated by the P values, although the stratification by KRAS was close to significant (p = 0.095).

FIGURE 5

Results for differential mRNA levels between primary cancer tissue and adjacent normal tissues in stage IV cancer patients in this study. (A) heatmap shows the mRNA level difference of the top 100 differential expressed mRNA genes. Blue bar indicates normal control while red bar indicates cancer tissue. (B) Volcano plot shows the significant upregulated and downregulated genes with mRNA level change. p < 0.05 and |logFC| > 2.5 were used as thresholds for interpreting significant mRNA level change. (C, D, and E) the results for GO, KEGG, and Reactome enrichment analysis for differentially expressed genes.

FIGURE 6

Multivariate analysis and establishment of risk stratification models for the prognosis of stage IV CRC patients. (A) The results of multivariate analysis of genes with differential mRNA levels. Seven genes were found to be significant in the analysis, while the other two also had significant influence on the final risk model and there were also included. (B) Risk stratification and Kaplan–Meier analysis based on results from multivariate analyses. Significant stratification of patient prognosis was found between high‐ and low‐risk patients (p = 0.012). (C) The diagnostic model (ROC curves) is shown for the risk stratification model at different time points. Maximal AUC of 0.89 was achieved at 18 months, as indicated.

FIGURE 7

Nomogram model for prediction of the individual risk of stage IV CRC. Genes from the multivariate analysis were used to establish the model. Scores can be calculate for each gene based on mRNA profile, and risk and can be calculated based on the formulation in the result section. Risk can then be quantified by the overall risk score (Linear predictor). The corresponding survival probability at each time point can then be calculated.