Genetic and microbial determinants of azoxymethane-induced colorectal tumor susceptibility in Collaborative Cross mice and their implication in human cancer

Figures & data

Figure 1. Identification of genetic variations and candidate genes associated with colorectal tumor susceptibility in CC mice.

(a) Illustrative graphics for study design. (b) AOM-induced colorectal tumor incidence in 27 CC strains. CC strains were divided into high susceptibility and low susceptibility groups based on their incidence. (c) Manhattan plot for the genetic association analysis of AOM-induced colorectal tumor in CC mice. The – log₁₀(p-value) is shown for SNPs ordered based on genomic position. Representative candidate genes located in each QTL are listed. (d) Network plot for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of candidate susceptible genes.

Figure 2. Subtyping human CRCs based on 334 CTS candidate genes.

(a) Consensus clustering model for CRC patient subtypes discovery and inference. (b, c) Subtype specific patients in TCGA-COAD (b) and GSE39582 (c) cohort show significant difference in overall survival.

Figure 3. Differences in gut microbiome between high and low CTS group.

(a) Alpha diversity boxplot between high and low CTS group. (b) PCoA using Bray–Curtis metric distances of beta diversity. The p value was obtained from PERMANOVA test of significant difference between low and high CTS groups based on Bray–Curtis dissimilarity. (c) Difference in abundance at order level between high and low CTS group. (d) Volcano plot showing genera enriched in high and low CTS group. Each dot represents a single genus. Genera significantly enriched in high and low CTS group are indicated by red and green dot, respectively (False discovery rate < 0.05 and | log₂(FC) |>1.0). (e) Histogram representation of differentially abundant genus between high and low CTS group identified by Linear Discriminant Analysis (LDA) effect size (LEfSe). 26 genera were significantly enriched for their respective groups (Kruskal-Wallis test, p < 0.05, LDA score > 2). (f) Cladogram representation of 26 genera from Figure 3(e). The green highlighted the genera enriched in low CTS group; the red highlighted the genera enriched in high CTS group.

Figure 4. The core genera for predicting CTS and their related function analysis.

(a) Random-forest analysis identified bacterial taxa that accurately predict high and low CTS. Biomarker taxa are ranked in descending order of importance to the accuracy of the model. The inset represented ten-fold cross-validation error as a function of the number of input genera used to differentiate high and low CTS. The dashed red line indicated the optimal cutoff for number of genera selected in the random forest predictive model. (b) ROC of the random forest model constructed using the 16 genera. The 95% confidence intervals were shown as shaded areas. (c) Microbial co-occurrence network was constructed based on samples of the low (n = 180, left panel) or high (n = 162, right panel) CTS group. The network connections were based on correlation test (p < 0.05 and Pearson’s correlation coefficient > 0.5). Genera nodes were colored according to families. (d) TaxFun2 analysis defines the difference of KEGG pathways between high and low CTS group.

Figure 5. Gut microbiome partially mediates the effects of genetics on CTS.

(a) The Venn diagram showed the microbiome-GWAS profile of 16 selected genera. Each compartment indicated the included selected genera and the number of microbiome-related SNPs in it. A SNP is considered related to microbiome when p < 10⁻⁶. All number values were included. (b) Circular visualization of chromosomal positions for all SNPs related to selected only genetic variants (SNPs) significantly associated with both the abundance of the five and above of 16 genera and colorectal tumorigenesis susceptibility. The outer layer shows chromosome location and candidate genes within each genetic locus are listed on the outside of the outer ring. The second layer (blue) shows SNPs related to colorectal tumorigenesis susceptibility (Mann-Whitney U test, p < 10⁻⁶). The inner layer (red) shows the microbiome related SNPs for ≥ 5 genera. (c) Boxplot showing Duox2 (UNC3869242)-specific association with Bacteroides, Allobaculum, Sutterella, Ruminococcus, Akkermansia, and Bifidobacterium abundance according to its genotype. (d) Venn diagram showing 47 SNPs related to selected only genetic variants (SNPs) significantly associated with both the abundance of the five and above of 16 genera and colorectal tumorigenesis susceptibility. (e) Microbial genera mediate the effect of host genetics on colorectal tumorigenesis susceptibility. (f) Six microbial genera were identified as mediators between genetic variants and colorectal tumorigenesis susceptibility. The estimate score and 95% CI for microbial genera associated with colorectal tumorigenesis susceptibility was calculated by mediation analysis.

Figure 6. Depletion of Duox2 alters gut microbiome structure in mice.

(a) The microbiota structure of Duox2 ^fl/fl (WT) and Duox2^fl/fl : Vilin-cre (CKO) mice. (b) Alpha diversity boxplot between WT and CKO group. (c) Principal component analysis and maximally collapsing metric learning of WT and CKO. (d, e) Differentially abundant taxa between low and high group analyzed by Linear Discriminant Analysis (LDA) effect size (LEfSe) were projected as histogram (d) and cladogram (e). (f) Microbial co-occurrence network constructed based on samples of WT and CKO group. The network connections are based on correlation test (p < .05 and Pearson’s correlation coefficient > 0.5). Edge connection between genera is shown in black lines. Genera nodes are colored according to families.

Figure 7. Interplay of Duox2 with gut microbiota and contributed to colorectal tumorigenesis susceptibility.

(a) Duox2^fl/fl (WT) and Duox2^fl/fl Villin-Cre (CKO) mice were challenged with AOM and DSS to establish colitis-associated tumorigenesis, which mimics the colitis-related tumorigenesis. (b) After AOM and DSS administration, there is similar colorectal tumorigenesis in CKO mice (n = 14) and WT littermates (n = 14) evidenced by similar the number, size and tumor load of macroscopic polyps between CKO mice and WT group. (c) Mice were treated with antibiotics (ABX) two weeks prior to induction with AOM and DSS. The efficacy of ABX was verified. (d) After AOM and DSS administration, less colorectal tumorigenesis susceptibility was seen in CKO mice (n = 15) than WT littermates (n = 16) via comparing the number, size, and tumor load of macroscopic polyps. (e) GSEA plot showed that conditional knockout Duox2 would lead to downregulate Wnt/β-Catenin pathway, inflammatory response and interferon-γ response pathway. (f) GSEA plot showed that conditional knockout Duox2 would lead to upregulate reactive oxygen species pathway and oxidative phosphorylation pathway.

Figure 8. Validation of core different abundant taxa and the function of DUOX2 according to public data.

(a) Compared with healthy donors, patients with colorectal cancer (CRC) had significantly lower levels of Bifidobacterium (p < .001), Ruminococcus (p < .001) and Akkermansia (p < .001), while the abundance of Allobaculum (p < 0.001), Bacteroides (p < .001), and Sutterella (p < .001) bacterial abundance significantly increased. (b) DUOX2 was significant negatively with Bacteroides abundance based on The Cancer Microbiome Atlas (TCMA) database. (c) 106 cases of colorectal cancer patients were divided into two groups with high expression (High) and low expression (Low) with the median of DUOX2 expression in colorectal cancer tissues. It was found that there was a significant difference in gut microbiome community between the two groups. Bray–Curtis distance for PCoA analysis and PERMAOVA for difference testing. The PCol interpretation rate was 33%, the PCo2 interpretation rate was 14%, and the PERMAOVA test (p = 0.001). (d) Boxplot showed that the abundance of Akkermansia and Prevotella was significantly higher in the high group than that in low group, while the abundance of Bacteroides and Parabacteroides were significantly lower in high group than that in low group.

Data availability statement

Mouse gut microbiome 16S rRNA gene sequencing data and RNA-Seq data from Doux2 knockout mice are are available in the National Center for Biotechnology Information (NCBI) BioProject Repository (https://www.ncbi.nlm.nih.gov/bioproject) under the BioProject “PRJNA802804”. Mouse gut microbiome 16S rRNA gene sequencing data from CC mouse cohort is available on OSF (https://osf.io/jbt5g/).³⁰ The mouse data has been included in Supplemental Table S1. The numeric counts table at genus level and corresponding taxonomic classifications have all been included as Supplemental Table S5.