A novel anoikis related gene prognostic model for colorectal cancer based on single cell sequencing and bulk transcriptome analyses

Abstract

Colorectal cancer (CRC) is a most deadly cancer, and effective prognostic biomarkers are urgently needed. Although anoikis has diverse regulatory roles in tumor progression, the impact of anoikis-related genes (ANRG) by single-cell and bulk transcriptome analyses on the prognostic value for CRC have not been studied. Differentially expressed genes (DEGs) associated with anoikis were obtained by performing single-cell RNA-sequencing (scRNA-seq) analysis in cells with high and low ANRG expression and weighted correlation network analysis (WGCNA) in a bulk RNA sequencing dataset. Key prognostic genes were selected from anoikis associated DEGs by least absolute shrinkage and selection operator (LASSO)-Cox regression analysis, and a prognostic model was established based on the risk score calculated from the expression levels of the identified key prognostic genes. A 10 anoikis-related-gene prognostic model (MGP, TPM2, CRIP2, TUBB6, C1orf54, NOTCH3, LTBP1, CSRP2, FSTL3, and VIM) was developed and the area under the curve (AUC) values of the model in predicting 1-, 3- and 5-year survival probabilities reached 0.744, 0.797, and 0.755, respectively. In conclusion, anoikis related genes could be promising prognostic factors for risk stratification of CRC patients.

Keywords: Anoikis; Colorectal cancer; Prognosis; Single cell sequencing.

Fig. 1

Single-cell RNA sequencing (scRNA-seq) data analyses. a A t-SNE plot showing the division of cells into 31 clusters. b A t-SNE plot presenting the annotated 6 cell types from the 31 cell clusters. c A t-SNE plot of cellular compositions in the normal and tumor samples. d A heat map showing the marker genes used to annotate the 6 cell types. e A t-SNE plot illustrating the distribution of cells by the expression levels of anoikis related genes (ANRGs). f A t-SNE plot presenting the distribution of cells by ANRG expression in the normal and tumor samples.

Fig. 2

Functional enrichment analysis of the differentially expressed genes between the two groups with high and low expression of ANRGs. a The top 5 kyoto encyclopedia of genes and genomes pathways enriched in the ANRG-high and ANRG-low groups. b The top 5 gene ontology biological processes enriched in the ANRG-high and ANRG-low groups. c The top 5 gene ontology cellular components enriched in the ANRG-high and ANRG-low groups. d The top 5 gene ontology molecular functions enriched in the ANRG-high and ANRG-low groups.

Fig. 3

Weighted correlation network analysis and module-trait analyses. a Selection of the optimal soft threshold power. b Clustering dendrogram. c A heat map showing the correlations between modules and anoikis and cytotoxic T lymphocytes. d A venn plot showing the intersection of anoikis associated DEGs derived from scRNA-seq analysis and anoikis associated genes obtained from WGCNA.

Fig. 4

Filtering of prognostic genes by univariate cox regression analysis. a A forest plot showing genes significantly associated with the prognosis of patients identified through univariate cox regression analysis. b A forest plot showing genes significantly associated with the prognosis of patients identified through multivariate cox regression analysis. c Expression levels of the identified prognostic genes in TCGA dataset.

Fig. 5

Construction and evaluation of a prognostic model. a Lasso coefficient profiles of the most relevant prognostic genes (left panel) and ten-fold cross validation for tuning parameter selection in the Lasso model (right panel). b Survival curves showing the association between the risk score and the prognosis in TCGA dataset. c Receiver operating characteristic curves showing the predictive accuracy of the model in predicting 1-, 3- and 5-year survival rates in TCGA dataset. d Survival curves showing the association between the risk score and the prognosis in the GSE17536 dataset. e Receiver operating characteristic curves showing the predictive accuracy of the model in predicting 1-, 3- and 5-year survival rates in the GSE17536 dataset.

Fig. 6

Identification of independent prognostic factors and establishing a nomogram. a A forest plot showing independent prognostic factors identified in univariate cox regression analysis. b A nomogram for predicting 5- and 3-year survival rates. The total point of a subject could be calculated by adding all the points assigned to each factor, and the subject’s risk could be determined by drawing a perpendicular line from the position of the total point to the probability axis.

Fig. 7

Infiltrated immune cells and biological pathways enriched in samples with high FSTL3 expression. a A box plot showing the association between FSTL3 expression and tumor stage in TCGA dataset. b A box plot showing the association between FSTL3 expression and tumor stage in the GSE41258 dataset. c KEGG pathways enriched in samples with high FSTL3 expression. d Hallmark pathways enriched in samples with high FSTL3 expression. e The most significantly enriched Hallmark pathway in samples with high FSTL3 expression.