Integration of the Microbiome, Metabolome and Transcriptomics Data Identified Novel Metabolic Pathway Regulation in Colorectal Cancer

Abstract

Integrative multiomics data analysis provides a unique opportunity for the mechanistic understanding of colorectal cancer (CRC) in addition to the identification of potential novel therapeutic targets. In this study, we used public omics data sets to investigate potential associations between microbiome, metabolome, bulk transcriptomics and single cell RNA sequencing datasets. We identified multiple potential interactions, for example 5-aminovalerate interacting with Adlercreutzia; cholesteryl ester interacting with bacterial genera Staphylococcus, Blautia and Roseburia. Using public single cell and bulk RNA sequencing, we identified 17 overlapping genes involved in epithelial cell pathways, with particular significance of the oxidative phosphorylation pathway and the ACAT1 gene that indirectly regulates the esterification of cholesterol. These findings demonstrate that the integration of multiomics data sets from diverse populations can help us in untangling the colorectal cancer pathogenesis as well as postulate the disease pathology mechanisms and therapeutic targets.

Keywords: biomarkers; colorectal neoplasms; metabolomics; microbiota; omics integration; transcriptome.

Figure 1

(a) Results from Kim et al. microbiome and metabolome analysis using BART methods; (a) Bayesian Additive Regression Trees (BART) model partial dependence plots for harmane, plotted at various quantiles (x-axis). y-axis shows the probits, a value of 0 indicates that CRC and normal are equally likely; values above 0 indicate that CRC is more likely and values below 0 indicate CRC is less likely. The shaded area shows the 0.95 Bayesian credible intervals of the probits; (b) BART model partial dependence plots for 5-aminovalerate, plotted at various quantiles (x-axis). y-axis shows the probits, a value of 0 indicates that CRC and normal are equally likely; values above 0 indicate that CRC is more likely and values below 0 indicate CRC is less likely. The shaded area shows the 0.95 Bayesian credible intervals of the probits; (c) matrix of counts of pairwise interactions from the BART mode, shown in the heatmap. The genera are Veillonella (OTU57), Eubacterium (OTU37), Haemophilus (OTU65), Adlercreutzia (OTU2), Anaerotruncus (OTU23), and SMB53 (OTU53). The metabolic and microbiome features are separated.

Figure 2

Partial dependence on CRC for representative pairs of microbial/metabolomic features, shown as contour plots, coloured by log-odds of CRC. Higher values (yellow) indicate increased risk of CRC; lower values (blue) indicate decreased risk of CRC; a value of 0 indicates the risks are equal. Plots for more feature pairs can be found in Supplementary Figure S2. Top to bottom: (5-aminovalerate, harmane), (5-aminovalerate, 1,2-Dilinoleoyl-GPC), (5-aminovalerate, 1-oleoyl-2-linoleoyl-GPC), (guanosine, 5-aminovalerate).

Figure 3

Partial results for Clos-Garcia microbiome and metabolome analysis using BART methods; (a) partial dependence plot of ChoE(20:4); (b) contour plots for microbiome–metabolome interactions between (1) Staphylococcus (OTU7517) and ChoE(20:4), (2) Blautia (OTU14967) and ChoE(20:4), (3) Roseburia (OTU9785) and ChoE(20:4); microbial interactions between (1) Roseburia (OTU9785) and Staphylococcus (OTU7517), (2) Blautia (OTU5962) and an unknown genus from family Lachnospiraceae (OTU17213), (3) OTU17213 and Blautia (OTU14967). The z-value (level) is the log-odds of CRC, interpreted in the same way as Figure 2.

Figure 4

(a) Volcano plot for 6066 genes selected after preprocessing with p adjusted cut-off 10e-100 shown. Significantly up- or downregulated genes are shown in red; (b) enrichment analysis of the 76 genes participating in pathways that are presented in barplot. (c) Venn diagram representation of the common genes between bulk RNAseq analysis and single-cell analysis. In total 17 genes were found common between the RNA and single cell datasets.

Figure 4

(a) Volcano plot for 6066 genes selected after preprocessing with p adjusted cut-off 10e-100 shown. Significantly up- or downregulated genes are shown in red; (b) enrichment analysis of the 76 genes participating in pathways that are presented in barplot. (c) Venn diagram representation of the common genes between bulk RNAseq analysis and single-cell analysis. In total 17 genes were found common between the RNA and single cell datasets.

Figure 5

(a) Shows the heatmaps for the 17 genes in the bulk RNA sequencing dataset; (b) heatmap for the 17 genes (single cell RNA sequencing). Red denotes upregulation, white indicates no expression or zero expression and blue depicts low expression; both heatmaps were performed using hierarchical clustering on the rows. The three clusters are coloured as red, black and green; (c) the network was generated using the 17 genes that were found to be differentially expressed in the network. The seed genes are labelled and shown in red, while the proteins connected to these genes are shown in green. The interactions between the genes and the proteins are shown in red. The blue coloured proteins or seed genes are involved in the metabolic pathways from KEGG.

Figure 6

ABCG2 and AQP8 genes are responsible for bile secretion and bile is converted to cholesterol by the gut enzymes [63]. Cholesterol is further converted to coprostanol by members of the Lachnospiraceae family [64].

Figure 7

Diagrammatic representation of the qualitative integration of different methods and omics data sets to identify markers for colorectal cancer using metabolomics, transcriptomics (single-cell RNA and bulk RNA) and microbiome data sets. Gene set enrichment was used to identify associations between the genes and metabolites.