Bacterial-Driven Inflammation and Mutant BRAF Expression Combine to Promote Murine Colon Tumorigenesis That Is Sensitive to Immune Checkpoint Therapy

In the United States, colorectal cancer is the third leading cause of cancer death for both men and women (1). Classification of colorectal cancer tumors into subtypes using features such as location along the axis of the colon, differing genetic mutations, genomic instability [e.g., chromosomal instability or microsatellite instability (MSI)], pathology, and epigenetic biomarkers helps define predictive factors for treatment and patient survival. Furthermore, recent advances have implicated the microbiota as a key contributing factor of disease (2–4).

Characterization of both healthy and disease-state gut microbiomes has revealed complex, bidirectional relationships between microbes and host in which microbes play important roles in modifying the host immune response, metabolism, and protecting the host against pathogen invasion (5, 6). Colonization with enterotoxigenic Bacteroides fragilis (ETBF) is associated with patients with colorectal cancer and tumors (7–9). To study the role of microbes and the host immune response in colorectal cancer, we previously developed a murine model wherein ETBF colonization induces Wnt- and IL17-dependent tumorigenesis in the distal colon of multiple intestinal neoplasia (Min^ApcΔ716/+) mice that are heterozygous for the adenomatous polyposis coli (Apc) gene (10). However, how or if mutations other than Apc^+/− modify ETBF-driven tumorigenesis is unknown. To determine if other mutations commonly found in colorectal cancer have specific interactions with ETBF, we introduced into the Min ETBF model the BRAF^V600E activating mutation of the MAPK pathway, which comprises approximately 90% of all BRAF mutations in human colorectal cancer (11). The mouse obtained to carry out these experiments expresses human BRAF^V600E at the mouse Braf locus.

Herein, we demonstrate that, compared with the Min ETBF model of distal tumorigenesis, ETBF colonization of BRAF^F-V600ELgr5^{tm1(Cre/ERT2)Cle}Min^ApcΔ716/+ (BLM) mice results in a new more proximal locus of colon tumors, reminiscent of the predominant right-sided location of BRAF mutant tumors inpatients. ETBF-induced tumorigenesis in mice with a different MAPK-activating mutation, Kras^G12DLgr5^CreMin^ApcΔ716/+, maintained only distal tumorigenesis, which is consistent with the preponderance of KRAS-mutant colorectal cancer being left-sided in humans. In addition, BLM colon tumors display a serrated-like histopathology, changes in the mucus bilayer, and increased mucus production. We also investigated changes in DNA methylation in the colon tissues of BLM mice compared with Min mice, because in humans the BRAF^V600E mutation is associated with the CpG methylator phenotype (CIMP), and found marked changes in DNA methylation in the BLM mice. Furthermore, ETBF induced, in BLM mice, a colitis with an IFNγ-driven immune signature associated with a robust recruitment of CD274 (PD-L1)–expressing myeloid-derived suppressor cells (MDSC). Notably, strong infiltration of BLM tumors with CD8⁺ cells coincided with efficacy of PD-L1 blockade in significantly reducing colon tumor numbers, suggesting an adaptive immunosuppression mechanism induced by IFNγ producing CD8⁺ tumor-infiltrating T cells. Together, ETBF colonization of the BLM mouse model provides novel insight into oncogene–microbiota interactions potentially relevant to the pathogenesis of human BRAF^V600E-mutated colorectal cancer.

In wild-type (WT) C57BL/6 mice, ETBF induces a rapid onset, predominantly distal colitis despite uniform colonization of ETBF along the axis of the colon (12). In initial experiments, we compared the colonic response of BRAF^V600ELgr5^Cre (BL mice; see Methods) and WT mice with ETBF colonization. Upon ETBF colonization, the midproximal colon regions (defined in Supplementary Fig. S1A) of BL mice exhibit enhanced colitis and colon epithelial cell shedding (Fig. 1A). ETBF fecal colonization was similar between WT and BL mice (Supplementary Fig. S1B). ETBF colonization of BL mice also led to a disruption of the ordered mucus bilayer in both the distal and midproximal regions of BL mice compared with sham mice and compared with the unperturbed midproximal region of WT colons colonized by ETBF at the same time point (Supplementary Fig. S1C). After long-term ETBF colonization (15–43 weeks), we noted increases in spleen weight as well as splenic and liver inflammation in ETBF-colonized BL versus WT mice (Supplementary Fig. S1D and S1E). Long-term ETBF colonization (15–43 weeks) of BL mice did not result in tumor formation in the colon (WT, n = 31; BL, n = 31; 1–5 mice per experiment for >10 experiments). Combined, these results suggest mice with an activating BRAF^V600E mutation colonized by ETBF exhibit increased colon mucosal and systemic inflammation.

To further examine the impact of BRAF^V600E on colon inflammation in ETBF-colonized mice, we performed flow cytometry analysis of midproximal and distal colon mucosal lamina propria leukocytes isolated from BL and WT mice 7 days after ETBF colonization, when acute colitis peaks in ETBF-induced inflammation in WT mice (12, 13), and also at a later time point, 2 weeks after ETBF colonization. Notably, ETBF-colonized BL mice displayed significantly higher levels of CD45⁺CD11b⁺Ly6G⁻Ly6C⁻F4/80^lowI-A/E^low cells, denoted F4/80^lowI-A/E^low, in the midproximal region compared with WT mice (Fig. 1B). This increased influx of F4/80^lowI-A/E^low cells was not observed in the distal colon of either BL or WT mice at 7 days after ETBF colonization or in the sham controls, regardless of region (Supplementary Fig. S1F). There were no significant differences for other leukocyte populations examined for any of these colon regions (Supplementary Fig. S1G). Cytokines previously identified as crucial for ETBF-driven tumorigenesis, Il6, Il17a, Cxcl2, and Cxcl5, were examined by qPCR and no differences were found between distal and midproximal regions in WT versus BL mice 7 days after colonization (Supplementary Fig. S1H).

To better understand the lineage and functional natures of the F4/80^lowI-A/E^low cell subset, we performed RNAseq analysis of cell-sorted CD45⁺CD11b⁺Ly6G⁻Ly6C⁻F4/80^lowI-A/E^low (F4/80^lowI-A/E^low) versus CD45⁺CD11b⁺Ly6G⁻Ly6C⁻F4/80^highI-A/E^high (F4/80^highI-A/E^high) cells from the midproximal colons of 2-week ETBF-colonized BL and WT mice. Analysis of the absolute numbers of both cell subsets recovered by cell sorting confirmed the striking accumulation of F4/80^lowI-A/E^low compared with F4/80^highI-A/E^high in the midproximal colon of BL mice compared with WT mice (Fig. 1C). In BL mice, the F4/80^lowI-A/E^low cells have a predominant MDSC gene expression signature compared with F4/80^highI-A/E^high, including S100A8, S100A9, Arg1, and Slc7a11 (coding xCT; refs. 14–16), as well as the transcription factor Slfn4 (ref. 17; Fig. 1D; Supplementary Table S1). Metabolism of L-arginine by arginase 1 and transport of cysteine by x_c⁻ cystine/glutamate antiporter are fundamental hallmarks of the immunosuppressive functions of MDSCs (18). T cells that do not express the xCT chain of the x_c⁻ cystine/glutamate antiporter are dependent on antigen-presenting cells producing cysteine from the reduction of cystine in the microenvironment. However, MDSCs that do not express the ASC transporter do not, in turn, export cysteine, leading to the sequestration of this essential amino acid for T-cell activation (19). Deprivation of the tumor immune microenvironment (TiME) of L-arginine and cysteine by MDSC sequestration suppresses T-cell functions. Furthermore, this MDSC signature is combined with expression of neutrophil-associated genes, including Cxcr2, Ccr3, Cd177, Hdc, Ngp, Csf3r, Asprv1, and Padi4 (20, 21). Interestingly, this neutrophil-like cell subset remained Ly6G negative and F4/80^low, resembling the macrophage-like neutrophils previously described in chronic inflammation (20). These gene expression profiles established that F4/80^lowI-A/E^low cells represent polymorphonuclear MDSCs (PMN-MDSC) and that their early accumulation in the midproximal region of BL mice colon occurs in the same region that later develops midproximal tumors. Moreover, these PMN-MDSCs were characterized by an IFNγ gene signature (Fig. 1D), suggesting that PMN-MDSCs were activated and recruited by proinflammatory signals, such as IFNγ (22). One of the IFNγ target genes significantly upregulated in F4/80^lowI-A/E^low compared with F4/80^highI-A/E^high cells is Cd274 (coding PD-L1), which is associated with MDSC immunosuppressive function by interacting with the T-cell checkpoint PD-1. These PMN-MDSC cells also express antibacterial enzymes and peptides such as Retnlg, Pglyrp1, Reg3b, Reg3g, and Serpine1, highlighting that they may act in defense against ETBF and/or other microbiota members that induced their recruitment. F4/80^lowI-A/E^low cells in BL mice were also enriched for MDSCs and IFNγ gene signatures relative to F4/80^low I-A/E^low in WT mice, in which less IFNγ production was observed (Supplementary Fig. S1I). BL F4/80^highI-A/E^high demonstrated a typical gene expression signature of proregenerative macrophages (Cd4, Mrc1, Mgl2, Il10, Cd163, Mmp12, and Mmp13) expressing higher levels of MHC-II associated genes (H2-M2, H2-Eb1, H2-Aa, H2-Ab1, H2-Dmb1, CIITA, and Cd74) and costimulatory molecules (Cd86; Fig. 1D).

BL F4/80^low I-A/E^low cells were also enriched for the hallmark reactive oxygen species pathway gene set (Fig. 1D), and we indeed found that BL mice had more γH2AX foci, a marker of DNA damage, in the midproximal colon 7 days after ETBF colonization compared with WT mice (Fig. 1E). Of interest, despite the altered mucosal morphology, as well as increased inflammation and colon epithelial DNA damage, mice with the BRAF^V600E mutation survived the acute inflammatory response to ETBF colonization significantly better than WT mice (Fig. 1F).

At age 9 to 16 weeks, baseline (sham) colon tumorigenesis in mice displaying Apc^Δ716/+ only (Min) or both Apc^Δ716/+ and BRAF^V600ELgr5^Cre mutations (BLM mice; see Methods) was generally rare and restricted to the distal 0 to 3 cm of the colon (n = 11 mice for each genotype; median = 0 mice for each genotype; tumor range, 0–2 tumors in Min mice vs. 0–3 tumors in BLM). Similarly, no increase in microadenoma numbers was observed in sham BLM colons compared with Min colons [Min mice: n = 10, median = 0 tumors (range, 0–2); BLM mice: n = 8, median = 0 tumors (range, 0–1)]. Consistent with our previous work, tumor formation in Min mice colonized long term with ETBF (6–13 weeks) occurred predominantly in the distal 3 cm of the colon (Fig. 2A; ref. 10). However, BLM mice with long-term ETBF colonization had significantly more tumors due to the emergence of a new locus of tumor formation in the midproximal colon (Fig. 2A; Supplementary Fig. S2A). Although BLM mice had significant increases in midproximal tumors relative to Min mice, there was no significant difference in distal colon tumor formation. However, total tumors for BLM mice were significantly elevated at the end of long-term ETBF experiments [Min median tumors = 10 (range, 1–45); BLM median tumors = 21 (6–54); P < 0.0001, Mann–Whitney U test]. Total colon microscopic adenomas or microadenomas were also significantly elevated in BLM mice compared with Min mice in these long-term ETBF colonization experiments (Fig. 2B). Of note, the altered midproximal colon tumorigenesis pattern was unique to BLM mice as ETBF colonization of mice with Kras^G12D, a different MAPK-activating mutation common in human colorectal cancer (i.e., Kras^G12DLgr5^CreMin^ApcΔ716/+ mice), displayed tumorigenesis restricted to the most distal 0 to 3 cm of the colon, similar to ETBF colonization of Min mice (SupplementaryFig. S2B). No difference in fecal or mucosal ETBF colonization was detected between Min and BLM mice (Supplementary Fig. S2C and S2D).

ETBF-colonized BLM mice also displayed a marked change in tumor histology. In contrast to the tubular adenoma morphology of colon tumors in ETBF-colonized Min mice, the tumors of ETBF-colonized BLM mice exhibited a mixture of villous and tubular adenoma histology and pink eosinophilic dysplasia, abundant cytoplasm, and enlarged nuclei, similar to the histopathology of sessile serrated tumors in humans with BRAF^V600E colorectal cancer (Fig. 2C; Supplementary Fig. S2E). As with BL mice, BLM mice display increased mucus production and a disordered bilayer after ETBF colonization when intact colons are fixed in methacarn solution and observed by periodic acid-Schiff (PAS) or MUC2 IHC (Fig. 2C; Supplementary Fig. S2F). Also similar to BL mice, ETBF-colonized BLM mice exhibited increased survival compared with ETBF-colonized Min mice within the first 10 days of colonization (Fig. 1F; BLM, 100% survival, n = 32; Min, 60% survival; n = 35; P < 0.0001, Fisher exact test). Altogether, these data indicate that the addition of BRAF^V600E to the ETBF-colonized Min murine model results in a new locus of serrated-like, midproximal tumors.

Changes in epithelial DNA methylation have been associated with both ETBF and the expression of BRAF^V600E (23–25). Thus, we next used methylation-binding domain sequencing (MBD-seq) to examine the impact of ETBF on CpG island DNA hypermethylation in BLM and Min mice.

Analysis of all DNA-hypermethylated regions in any ETBF sample (BLM or Min, epithelium or tumor) relative to sham epithelium indicated that the BLM and Min genotypes were associated with distinct DNA methylation patterns (Fig. 3A). After ETBF colonization, BLM midproximal epithelium had a higher number of CpG islands displaying DNA hypermethylation relative to sham BLM midproximal epithelium than Min distal epithelium relative to sham Min distal epithelium (Fig. 3B; Supplementary Table S2). ETBF-induced BLM tumors also had more CpG islands with DNA hypermethylation than Min tumors when compared with their respective controls (Fig. 3C). Unsupervised hierarchical clustering of the distinct regions harboring significant DNA hypermethylation in either ETBF-colonized epithelium or ETBF-induced tumors compared with sham epithelium demonstrated that ETBF-associated DNA hypermethylation was similar in BLM epithelium and tumor samples (Fig. 3A; Supplementary Fig. S3A). In contrast, ETBF colonization had more of an effect on DNA hypermethylation in Min tumors than epithelium, as demonstrated by ETBF-induced Min tumors clustering separately from the ETBF-colonized Min epithelium samples (Supplementary Fig. S3B). CIMP-related genes, Cdkn2a and Gata5, as well as Polg, Fut4, and Ano1, all had relative DNA methylation levels that were consistent with the MBD-seq data when validated by quantitative methylation-specific PCR (Supplementary Fig. S3C).

To further define the differences in ETBF-driven colon tumorigenesis between BLM and Min mice, we performed RNA sequencing on mRNA extracted from ETBF-induced BLM midproximal and Min distal colon tumors as well as BLM midproximal and Min distal normal epithelium from sham mice. Initial principal components analysis using all expressed genes revealed clustering by genotype and ETBF status, indicating that genotype as well as ETBF colonization and tumor phenotype lead to distinct gene expression signatures (Fig. 4A; Supplementary Table S3).

By performing gene set enrichment analysis (GSEA) using gene sets derived from hallmark pathways (26), we detected a significant change in the baseline mucosal immunity gene expression between BLM and Min sham epithelium. GSEA indicated that BLM sham epithelium is significantly enriched for the hallmark IFNγ and IFNα gene sets (Fig. 4B; Supplementary Fig. S4A). We also used CIBERSORT, a computational method for quantifying immune subsets from bulk tissue gene profiles, to survey potential infiltrating leukocytes and other immune components previously shown to correlate with both prognosis and therapeutic response (27). Consistent with the GSEA, CIBERSORT analysis demonstrated that the BLM sham epithelium is enriched in a Th1 immune signature while Min sham epithelium is enriched in regulatory T-cell (Treg) and monocyte gene profiles (Fig. 4C). These data suggest that the mucosa of BLM mice is poised for Th1 polarization prior to ETBF colonization.

Using gene sets derived from BioCarta pathways (28), GSEA analysis comparing ETBF-induced BLM midproximal and Min distal tumor profiles demonstrated significant enrichment in Th1-related expression gene sets in ETBF-induced BLM tumors compared with Min tumors, including genes associated with IL12 and Stat4-dependent signaling in Th1 development (Fig. 4D). Relative to Min tumors, BLM tumors displayed decreased enrichment of pathways involved in cell motility (RHO, MET, CDC42RAC, and integrin) and activation of calcium signaling and protein kinase C (PYK2, MEF2D, AT1R, and BCR) (Fig. 4D). CIBERSORT analysis also found a predominance of Th17-related gene expression in ETBF-induced Min tumors compared with BLM tumors (Fig. 4E). BLM tumors have increased expression of genes related to mast cells that can be activated by microbial pathogens (29), are proinflammatory, and have been shown to play a role in polyp formation in Min mice and colorectal cancer in humans (30). Quantitative RT-PCR confirmed that ETBF-induced Min tumors had a higher expression of Th17-associated genes such as Il6, Il1b, and Tgfb relative to ETBF-induced BLM tumors, as well as IL17 target genes Cxcl1, 2, and 5 (Supplementary Fig. S4B). ETBF-induced Min tumors were also enriched for genes associated with monocytic immature myeloid cell (MO-IMC) function and recruitment, including Arg1, Nos2 [which encodes inducible nitric oxide synthase (iNOS)], Ccl2, S100A8/9, and Mmp9 compared with ETBF-induced BLM tumors (Supplementary Fig. S4B). In contrast, relative to ETBF-induced Min tumors, ETBF-induced BLM tumors are enriched for the antimicrobial genes Reg3b, Reg3g (bactericidal peptides), and Ccl28 (homing of IgA-producing plasma cells at mucosal surface). Furthermore, Muc2 and Cldn15 (claudin 15, barrier function) expression is enriched in both sham BLM epithelium and ETBF-induced BLM tumors relative to their respective Min controls. Enhanced expression of Muc2 and barrier function genes in the sham BLM midproximal epithelium is lost in ETBF-induced BLM midproximal tumors, which are more similar to ETBF-induced Min distal tumors for these features (Supplementary Fig. S4B).

Furthermore, genes associated with DNA hypermethylation and altered gene expression in ETBF-induced Min distal tumors relative to Min distal sham epithelium were enriched for gene ontology pathways associated with regulation of α-β T-cell activation and leukocyte differentiation. Many of these genes had significantly lower DNA methylation levels in BLM epithelium and/or tumors than the respective Min samples (Gata2, Gata3, Hlx, Runx3, and Runx1) and/or were not DNA hypermethylated in ETBF-induced BLM tumors relative to BLM epithelium (Cd80, Runx3, Runx1, and Zfpm1; Fig. 4F). These DNA methylation differences were associated with gene expression differences between the Min and BLM sample types for some genes (Cd80, Gata3, Hlx, Runx3, and Zfpm1; Fig. 4G). Although overall DNA hypermethylation in ETBF-induced BLM tumors was not strongly associated with genes involved in immune cell regulation, some immune genes with altered expression also had changes in DNA methylation. For example, Il1rl2 (encodes IL36R), a gene involved in IL17-mediated fibrosis (31), was significantly DNA hypermethylated in ETBF-induced BLM tumors relative to sham BLM epithelium (Fig. 4F). The DNA methylation change was associated with significantly altered Il1rl2 expression in ETBF-induced BLM tumors relative to sham BLM epithelium and Min distal tumors (Fig. 4G). Expression of Lcn2 (Lipocalin2), a target gene of IL36R signaling and mediator of epithelial protection to microbial inflammation (32), was also reduced in ETBF-induced BLM midproximal tumors relative to ETBF-induced Min distal tumors.

Last, similar to ETBF-induced Min tumors, ETBF-induced BLM tumors were found to be microsatellite stable (MSS) and distinct from tumors from mismatch repair–deficient mice, which are MSI⁺ (Supplementary Table S4). Altogether, the DNA methylation data demonstrate that when the BRAF^V600E mutation is present, ETBF colonization results in a greater number of CpG islands with DNA hypermethylation than in Min-only tissue. Furthermore, there are differences in DNA hypermethylation of genes related to the regulation of T cells between ETBF-induced Min and BLM tumors. Altogether, our findings suggest that DNA methylation changes associated with BRAF^V600E mutation contribute to the altered inflammatory response to ETBF in BLM mice.

Because BRAF^V600E affects the methylation and expression of some immune-related genes, we sought to characterize the nature of the tumor-infiltrating leukocytes in BLM and Min mice. We performed intracellular cytokine staining and flow cytometry analysis on enzymatically digested tumors. ETBF-induced BLM tumors had a significant increase in CD3⁺CD8⁺ T cells and IFNγ-producing cells compared with ETBF-induced Min tumors (Fig. 5A). The MO-IMC (CD11b⁺Ly6C^hiLy6G⁻) signature characteristic of ETBF-induced Min tumors (33) was significantly decreased in BLM tumors (Supplementary Fig. S5A). There was no difference in IL17-producing cells or other immune cell types examined between the different genotypes. IHC staining of tumor sections confirmed the accumulation of the CD8⁺ cells (Fig. 5B) and demonstrated the overall accumulation of F4/80⁺ myeloid cells (Supplementary Fig. S5B) in ETBF-induced BLM tumors. Altogether, these results indicate BLM tissues with ETBF treatment have a type 1 immune microenvironment characterized by IFNγ production and prevalent CD8⁺ cell infiltration.

ETBF-induced tumorigenesis in the midproximal region of BLM mice, despite increased tissue infiltration with CD8⁺ T cells and an IFNγ signature, is reminiscent of adaptive immunosuppression described in human cancer, whereby adaptive IFNγ-induced upregulation of T-cell checkpoint ligands in the tumor microenvironment suppresses the function of tumor-infiltrating effector T cells (34). Because our gene expression data confirmed higher expression of an IFNγ-signaling pathway (Stat1, Jak2, and Irf1) and IFNγ-target T-cell checkpoint genes (Ido1 and Cd274/Pdl1) in ETBF-induced tumors in BLM mice relative to Min mice, and Cd274 gene expression was upregulated in accumulating PMN-MDSCs compared with tissue residential macrophages in the early phase of ETBF colitis (Fig. 1D), we next tested the hypothesis that checkpoint blockade using anti–PD-L1 antibodies would inhibit tumor development in ETBF-colonized BLM mice. ETBF-colonized BLM and Min mice were treated with anti–PD-L1 or isotype (control) antibodies (see Methods), and tumor numbers were evaluated 6 weeks after onset of ETBF colonization. We found that anti–PD-L1 therapy significantly reduced the total number of ETBF-induced colon tumors in BLM mice, whereas anti–PD-L1 had no effect in Min mice (Fig. 6A). This reduction was seen in both the distal (0–3 cm) and midproximal (4–7 cm) region of the colon in BLM ETBF mice treated with anti–PD-L1 (Fig. 6B).

To further characterize the BLM mouse response to anti–PD-L1 therapy, we performed RNA sequencing on mRNA extracted from ETBF-induced BLM midproximal and distal and Min distal colon tumors following isotype or anti–PD-L1 therapy. Analysis of the RNA-seq data by t-distributed stochastic neighbor embedding revealed that tumors from BLM mice treated with anti–PD-L1 (BLM_PDL1) predominantly clustered separately from tumors isolated from isotype-treated BLM mice (BLM_Iso), whereas Min_Iso and Min_PDL1 tumors clustered more closely together (Fig. 6C). Furthermore, BLM_PDL1 tumors had 334 and 510 genes significantly (adjusted P < 0.01) up- and downregulated, respectively, compared with BLM_Iso tumors. In contrast, Min_PDL1 had no genes with significantly altered expression relative to Min_Iso tumors (Fig. 6D; Supplementary Table S5). A heatmap of all differentially expressed genes (adjusted P < 0.05 for any pairwise group comparison) indicated the midproximal and distal tumors from BLM mice responded to anti–PD-L1 therapy in a similar manner, and the similarity between distal and midproximal tumors in the BLM_Iso group also suggests that the addition of the BRAF^V600E mutation results in similar tumors regardless of region (Supplementary Fig. S6A). GSEA with gene sets derived from hallmark pathways demonstrated that BLM_PDL1 tumors are significantly depleted for pathways related to cell proliferation (E2F targets, G2M checkpoint, and Myc targets) relative to BLM_Iso tumors. BLM_PDL1 tumors were enriched for epithelial–mesenchymal transition (EMT), TGFβ signaling, and angiogenesis pathways, suggesting potential mechanisms connected to anti–PD-L1 resistance (Fig. 6E). The potential resistant state of these tumors is further confirmed by a lack of increase in CD8⁺ cells (Supplementary Fig. S6B) and a reduction in expression of IFNγ pathway genes (Supplementary Fig. S6C) in BLM_PDL1 versus BLM_Iso tumors.

Understanding the interaction between genetics, microbes, and inflammation during cancer pathogenesis will aid in predicting and treating cancers. Mouse models of colorectal cancer often fail to recapitulate many physiologic aspects of human disease. For example, the APC gene is mutated in 85% of sporadic human colorectal cancers, yet in so-called Min mice, Apc loss of heterozygosity leads predominantly to small bowel tumors. However, our group has found that upon infecting Min mice with ETBF, the combination of the B. fragilis toxin (BFT) and a Th17 inflammatory response induces the formation of numerous distal colon tumors. ETBF is detected in patients with colorectal cancer significantly more often compared with healthy individuals (7, 8). ETBF may be an “oncogenic” bacterium candidate triggering in vivo tumorigenesis via inflammation, the production of BFT, and epigenetic modulation of epithelial gene expression [reviewed (2)]. Thus, we hypothesize that the multifactorial cause of human tumors may require similar such components in mouse models. To assist in testing specifically our hypothesis regarding a gene–microbe–inflammation interaction, genotypes in our experiments were cohoused to diminish, if not eliminate, the impact of the broader microbiome members on our results.

BRAF^V600E mutation is found in an estimated 90% of BRAF-mutated tumors. BRAF mutation in patients with MSS colorectal cancer has been associated with a serrated, poorly differentiated, mucinous tumor phenotype and a poor prognosis compared with other subtypes of colorectal cancer, including the more common BRAF^mut MSI colorectal cancers (35). Herein we tested the impact of the BRAF^V600E mutation on ETBF colon tumorigenesis to better understand the pathogenesis of BRAF-mutated colorectal cancer. In other murine studies, introduction of the BRAF^V600E mutation in all intestinal epithelial cells during gestation resulted in tumor formation predominantly in the small intestine (36). Other studies in which activation of mutant BRAF occurred in adult intestinal cells demonstrated that mutant BRAF inefficiently induced intestinal tumorigenesis without the expression of additional genes (37, 38). Our group has previously demonstrated that in the presence of induced BRAF^V600E mutation, xenografts derived from aging transverse colon organoids (5 months in vitro) selected for CIMP-like promoter DNA hypermethylation and activation of the Wnt pathway demonstrate cell transformation with right-sided colorectal cancer characteristics (25). These events accelerated with advancing age of the initiating organoids. Thus, although the BRAF mutation likely alters the colon epithelium, key microenvironmental changes, such as ETBF-induced inflammation and additional genetic abnormalities, like loss of APC function, are required for full tumorigenesis. Together, this work suggests that, in vivo, environmental stress and/or aging is needed for changes in the methylation phenotype in cells bearing the BRAF^V600E mutation. In this regard, we confirmed here that the BRAF^V600E mutation alone does not result in increased colonic tumors in Min mice, yet upon ETBF colonization, introduction of BRAF^V600E drove right-sided colon tumorigenesis with a serrated-like histology, similar to BRAF mutant tumors in humans. Interestingly, we also demonstrated that BRAF^V600E mutation and ETBF colonization result in disruption of the mucus layer and significant changes in myeloid populations that persist through tumorigenesis. Overall, the immune and epithelial results did not differ between the midproximal and distal regions of BLM mice, suggesting that the main drivers of these changes are BRAF^V600E mutation combined with ETBF colonization. Disruption of the protective inner layer of mucus that is typically devoid of bacteria likely causes altered interactions between the microbiota and epithelial and immune cells, illustrated herein by the recruitment of PMN-MDSCs and their activation by local proinflammatory cytokines. Interestingly, we have recently demonstrated that biofilms, dense communities of inner mucus-invasive bacteria encased in complex matrices, predominantly form on right-sided colon cancers in humans (39), which further underscores regional and genetic differences in the epithelial response to gut flora. These findings suggest that microbiota can play a critical role in the formation of BRAF mutant colorectal cancer.

BRAF^V600E is associated with CIMP in human colorectal cancer. We observed that the BLM epithelium of uninfected mice had low baseline levels of DNA methylation, demonstrating that the presence of the BRAF^V600E mutation is not sufficient to cause DNA hypermethylation, as has been suggested previously (25). However, exposure to ETBF induced significant CpG island DNA hypermethylation in both the BLM midproximal epithelium and tumors compared with the sham midproximal epithelium baseline. Importantly, this degree of DNA hypermethylation is not present in ETBF-colonized Min distal epithelium, suggesting that the presence of the BRAF^V600E mutation and ETBF uniquely induces high levels of CpG island DNA hypermethylation outside of tumorigenesis. BRAF^V600E activates constitutive MAPK signaling and can contribute to cell proliferation and survival. We noted that there was both an increase in survival of the acute ETBF infection of mice with BRAF^V600E mutation 10 days after ETBF colonization (BL vs. WT or BLM vs. Min), as well as an increase in γH2AX foci detected by IHC. Previously, we have observed that oxidative damage to chromatin during inflammation leads to epigenetic silencing (23, 40). We hypothesize that the increased survival of ETBF-colonized BRAF^V600E mice compared with WT or Min controls may indicate that these mice have more intestinal epithelial cells that have undergone and survived oxidative DNA damages, which may, in turn, result in a greater impact of ETBF on epigenetic changes such as DNA methylation. Furthermore, we speculate that enhanced survival fosters mucosal conditions favoring the emergence of a new locus of tumorigenesis in parallel with sensitivity to PD-L1 blockade.

BRAF mutation in MSS colorectal cancer is associated with a particularly poor prognosis, but our findings suggest that this combination may delineate a population of patients with colorectal cancer who could benefit from immune checkpoint blockade. The IFNγ - and Th1-associated expression signatures found in the colon tissues of mice with the BRAF^V600E mutation suggest that they are primed to have a predominantly Th1 immune environment. Sham mice bearing BRAF^V600E do not show a significant increase of specific IFNγ-producing cell subsets, including CD4⁺, CD8⁺, ILC, or γδ-T cells, suggesting that the overall increase in IFNγ-producing cells stimulated by ETBF results from a general type 1 polarization rather the induction of a specific cell type. Thus, the Th1-related gene expression patterns correlate with priming the BLM tissues to form tumors with high levels of CD8⁺ T cells and IFNγ-activated, macrophage-like PMN-MDSCs (Ly6G^negF4/80^lowI-A/E^low) (20), as well as a decrease in MO-IMC cells typically detected in Min ETBF tumors. Importantly, the IFNγ activation of these PMN-MDSCs in BL mice was associated with the upregulation of the immunosuppressive ligand CD274 (aka PD-L1). These findings are consistent with the association between BRAF^V600E, CIMP, and the presence of tumor-infiltrating lymphocytes in human colorectal cancer (41). Importantly, these characteristics of a type 1 TiME correlated with BLM tumors being responsive to anti–PD-L1 therapy. Similar observations were recently made in patients with BRAF-mutated colorectal cancer and non–small cell lung cancer who responded to anti–PD-1 treatment even if they lacked traditional biomarkers such as high tumor mutational burden or PD-L1 expression that are normally associated with a positive response to checkpoint therapy (42). Thus, we posit that the BRAF^V600E mutation conditions the epithelium to a distinct tumor pathogenesis and TiME during ETBF-driven tumorigenesis that gains responsiveness to checkpoint inhibition. Tumors that persisted in BLM mice after anti–PD-L1 therapy are depleted in gene signatures related to cell proliferation, which is consistent with findings in melanomas from patients who responded to anti–PD-1 checkpoint therapy (43). Responding melanomas were also found to have upregulation of IFNγ gene signatures. However, persisting BLM_PDL1 tumors were depleted for IFNγ and enriched for cancer-promoting pathways, including EMT and angiogenesis. Interestingly, EMT and angiogenesis pathways were enriched in melanomas and other tumor types, including colon cancer, with innate anti–PD-1 resistance (44). The persistence of strong signatures associated with nonresponse in persisting BLM_PDL1 tumors raises important questions. For example, although we saw a response to anti–PD-L1 therapy in BLM mice, the signatures we observed in the remaining tumors may herald a relapse, which was not tested in our experiments. Furthermore, in our model, the microbiome driver of tumorigenesis is still present, and perhaps eliminating ETBF in parallel with anti–PD-L1 therapy would change the outcome and/or durability of the immune checkpoint therapy response. Our in situ model of tumorigenesis provides the ideal system to test these hypotheses further. Lastly, despite some similarity between responses in our model and those reported in melanoma, immune checkpoint therapy responses may be tumor specific (42). Currently, studies on responses in cancer types other than melanoma are limited.

Our interest in how microbe-driven inflammation in a BRAF- and Apc-mutated background may change the landscape of the colonic epithelium, mucosal immunology, and tumorigenesis potential shows parallels to human colorectal cancer development that may help us understand these pathways and better develop therapies and biomarkers for selecting therapies for patients. These observations support the hypothesis that specific microbe–oncogene interactions are crucial to colorectal cancer pathogenesis and to guide targeted therapies.

C57BL/6J (originally from The Jackson Laboratory) and Min^ApcΔ716/+ mice (from Drs. David Huso and Bert Vogelstein, Johns Hopkins University, Baltimore, Maryland) were handled and inoculated with enterotoxigenic B. fragilis strain 86–5443–2-2 (ETBF), as in Wu and colleagues (10). Loxp-flanked BRAF^F-V600E were obtained from The Jackson Laboratory (RRID:IMSR_JAX:017837), as were leucine rich repeat containing G protein-coupled receptor 5 (Lgr5) CreERT2 knock-in mice (Lgr5^{tm1(Cre/ERT2)Cle}; RRID:IMSR_JAX:008875), which were then crossed with the Min^ApcΔ716/+ mice to produce BRAF^F-V600E Lgr5^{tm1(Cre/ERT2)Cle}Min^ApcΔ716/+ (BLM) mice. Loxp-flanked Kras^G12D mice (also known as B6.129S4-^Krastm4Tyj/J; RRID:IMSR_JAX:008179) were also obtained from The Jackson Laboratory. Recombination in mice bearing Lgr5^Cre was induced with tamoxifen, as in Barker and colleagues (45), at 4 weeks of age. All mice were bred and maintained in a specific pathogen-free barrier facility in accordance with the Association for the Assessment and Accreditation of Laboratory Animal Care International. For all experiments, both male and female mice were used. Mice of different genotypes were cohoused upon weaning and for most of the experiments to maintain a uniform microbiome. The Johns Hopkins University Animal Care and Use Committee approved all experimental protocols.

Colon regions were defined as distal (7-day experiments, 0–2 cm measured from the rectum; longer time points, 0–3 cm) and midproximal (the 2-cm region distal to the portion of the mouse colon with transverse folds commonly referred to as the “proximal” colon). Colons were flushed with PBS, measured, and either snap-frozen and stored at −80°C or processed immediately as appropriate for assay. Tumors were counted and mapped along the length of the colon using a dissecting scope, and for some assays, individual tumors were removed for snap-freezing or processing at the time of harvest. Samples specifically described as “epithelial” were collected by scraping the debris-free mucosal surface of the dissected colon (tissues were verified to have no gross tumors under the dissecting scope), snap-frozen, and stored at −80°C. Others have demonstrated that this method is effective for obtaining intestinal epithelial cells (46). Tissues for immune studies or profiling were snap-frozen whole or immediately processed as appropriate. Tissues were washed three times in PBS before using in the indicated protocol.

Colon tissues were enzymatically processed using 400 U/mL Liberase (cat. 5401127001; Sigma-Aldrich) and 0.1 mg/mL DNAse1 (cat. 10104159001; Sigma-Aldrich). Splenocytes were isolated from Liberase-treated spleen samples using Lymphoprep density gradient (Accurate Chemical and Scientific Corporation). Leukocytes were isolated from single-cell suspensions using 80/40/20 Percoll density gradient centrifugation (GE Healthcare Life Science) as described previously (10). Single-cell suspensions were stained and myeloid and lymphoid populations characterized by flow cytometry as in Chung and colleagues (13) and Thiele Orberg and colleagues (33). Briefly, MO-IMCs were characterized as CD45⁺CD3⁻CD11b^hiLy6C^hiLy6G⁻F4/80^−/lowI−A/E^−/low, polymorphonuclear immature myeloid cells were CD45⁺CD3⁻CD11b^hi Ly6C^intLy6G⁺F4/80⁻I-A/E⁻, and myeloid CD45⁺CD3⁻CD11b^intLy6C⁻Ly6G⁻F4/80⁺ cells were further subsetted into F4/80^highI-E/A^high (CD45⁺CD11b⁺Ly6G⁻Ly6C⁻F4/80^highI-A/E^high) and F4/80^lowI-A/E^low (CD45⁺CD11b⁺Ly6G⁻Ly6C⁻F4/80^lowI-A/E^low).

Intracellular cytokine staining was performed following 4.5-hour stimulation with phorbol 12-myristate 13-acetate and ionomycin stimulation cocktail (eBioscience) in the presence of protein inhibitor cocktail (eBioscience). Cells were subsequently stained as referenced (13, 24) with cell surface markers followed by fixation/permeabilization (Cytofix/Cytoperm; BD), enabling intracellular staining. Flow cytometry was performed using an LSR-FORTESSA cytometer (BD), and data were analyzed with FACSDiVa 6.1.3 software (BD FACSDiva Software, RRID:SCR_001456; BD).

In the case of tumor tissues, for each mouse, all colon tumors were pooled, minced in 5% FBS RPMI, and enzymatically processed (400 U/mL Liberase and 0.1 mg/mL DNAse1; Sigma-Aldrich) for 15 minutes in a 37°C 5% CO₂ incubator. Tissues pieces were then passed through a 70-μm cell strainer to create a single-cell suspension. Cells were then washed and counted before being stained for flow cytometry analysis.

Single-cell suspension of midproximal portions of the colons were obtained by enzymatic digestion as described above and stained with CD45, Ly6C, Ly6G, F4/80, and I-A/E antibodies (13, 33). CD45⁺CD11b⁺Ly6G⁻Ly6C⁻F4/80^lowI-A/E^low and CD45⁺CD11b⁺Ly6G⁻Ly6C⁻F4/80^highI-A/E^high were cell sorted using a BD FACSAria FUSION (BD) cell sorter. Cells were directly collected in RLT buffer of the RNeasy Micro Kit (cat. 74004; Qiagen) for RNA extraction as described below.

Colons for histology were flushed, bisected longitudinally, pinned flat, and fixed using 10% buffered formalin for 48 hours at room temperature. Colons were then rolled into cassettes for standard paraffin embedding. For samples in which the maintenance of mucus and/or mucosal/epithelial contact was desired, intact colons were fixed directly in methacarn (60% methanol, 30% chloroform, and 10% acetic acid) for 48 hours at 4°C. PAS staining of these tissues was carried out according to kit instructions (cat. 101646; Millipore). Other tissues such as spleen, liver, and mesenteric lymph nodes were fixed in formalin for 48 hours.

Inflammation scores for colons 7 days post-ETBF colonization were graded blindly by a board-certified pathologist as in Rhee and colleagues (12). Briefly, scores for inflammation were as follows: 0, normal; 1, mild increase in inflammatory cells and no mucosal epithelial changes (no proliferation or loss of crypt structure); 2, moderate increase in inflammatory cells and mild scattered mucosal epithelial proliferation with or without focal loss of crypt architecture; 3, moderate increase in inflammatory cells, diffuse or nearly diffuse (more than two sites) mucosal epithelial proliferation, and edema with or without focal loss of crypt architecture; and 4, severe increase in inflammatory cells and marked consistent proliferation with extensive loss (more than two sites) of crypt architecture.

Immunostaining of colon sections was performed with the PowerVision kit according to the manufacturer's protocol (Leica Biosystems). Briefly, slides were heated at 60°C for 10 minutes, deparaffinized, and hydrated through xylene, graded ethyl alcohols, dH₂O, and dH₂O with 20% Tween 20 (cat. P-7949; Sigma-Aldrich). After antigen retrieval and 25 minutes of steaming in Target Retrieval Solution (cat. S170084–2; Dako) using the Black and Decker Handy Steamer Plus, sections were treated for 5 minutes with Dual Blocking Solution (cat. S2003; Dako). Primary antibodies γH2AX (cat. 9718, 1:800; Cell Signaling Technology) and MUC2 (cat. sc-15334, RRID:AB_2146667, 1:100; Santa Cruz Biotechnology) were incubated at room temperature for 45 minutes. Slides for CD8 and F4/80 staining were preblocked in DakoCytomation Biotin Blocking System (cat. X0590; Dako). Antibodies for CD8 (cat. 14–0808, 1:800; eBiosciences) and F4/80 [cat. MCAP497, RRID:AB_2335598 (formerly Serotec, MCAP497), 1:1,000; Bio-Rad] were incubated at room temperature for 45 minutes, soaked an additional 45 minutes in PBS-Tween, and followed by mouse-adsorbed biotinylated anti-Rat IgG (cat. BA-4001, RRID:AB_10015300, 1:500; Vector Laboratories). For all, the secondary antibody used was the anti-rabbit IgG reagent provided in the Powervision + PolyDAB kit (cat. PV6119; Leica Biosystems). DAB chromogen (cat. D4293; Sigma-Aldrich) was used for immunostaining visualization, and sections were counterstained with Mayer's hematoxylin.

DNA was isolated from tumors or epithelium using the QIAamp DNA Mini Kit (cat. 51306; Qiagen) following the manufacturer's protocol. To identify differentially methylated regions, MBD enrichment was performed on DNA from epithelium or individual tumors using Diagenode's MethylCap kit. Libraries were prepared using Bio Scientific's Methyl Sequencing kit. Single-end 75-bp sequencing was performed using an Illumina Nextseq. For this study, MBD-seq was performed on Min sham epithelium (n = 1), Min ETBF tumor (n = 1), BLM sham epithelium (n = 3), BLM ETBF epithelium (n = 3), and BLM ETBF tumors (n = 5). Previous MBD-seq data from Maiuri and colleagues (14) were also included (n = 3 Min sham epithelium, n = 3 Min ETBF epithelium, n = 5 Min ETBF tumors). The z scores were calculated using a 500-bp fixed-sized bin spanning CpG islands based on the distribution of coverage from uniquely mapped reads. The z ratios were derived from the comparison of z scores for the different sample types for the 500-bp-sized regions. Additional details are in Maiuri and colleagues (14).

For cell-sorted MDSCs and macrophages (midproximal regions from mice; WT, n = 3; BL, n = 3), RNA was isolated and DNase treated using the RNeasy Micro Kit (cat. 74004; Qiagen). Libraries were prepared with Nugen Ovation RNAseq System v2. Paired-end 150-bp sequencing was performed using an Illumina HiSeqX. Illumina's CASAVA 1.8.4 was used to convert BCL files to FASTQ files using default parameters. Data were aligned to the mm10 reference genome.

For BLM and Min tumors and epithelium, RNA was isolated from epithelium (four each sham Min distal and sham BLM midproximal) and tumors (four ETBF Min distal and five ETBF BLM midproximal) using the RNeasy Mini Kit (cat. 74104; Qiagen) and RNeasy Micro Kit (cat. 74004; Qiagen), respectively, and DNase-treated. Libraries were prepared with Illumina's TruSeq Stranded mRNA library kit and poly-A tail selected. Paired-end 75-bp sequencing was performed using an Illumina Nextseq. Raw fastq files reflecting single-end TruSeq Stranded RNA libraries were submitted to RSEM for stranded alignment (Bowtie2) using the GRCm38 reference genome masked for miscellaneous RNA (e.g., rRNA, lincRNA, and snRNA), mitochondrial genes, and ribosomal proteins.

For tumors from the PD-L1 checkpoint experiment, RNA was isolated from tumors (BLM isotype: distal n = 3, midproximal n = 4; BLM_PDL1: distal n = 3, midproximal n = 3; Min isotype: distal n = 4; Min_PDL1: distal n = 4) and then DNase treated using the RNeasy Micro Kit. Libraries were generated using NEB Next Ultra RNA Library Prep Kit for Illumina (NEB). Paired-end 150-bp sequencing was performed on an Illumina Novaseq 6000. Data were aligned to the GRCm38 reference genome.

RSEM quantified counts were submitted to DESeq2 (RRID:SCR_000154) for downstream normalization and differential expression analysis. P values were calculated using the Wald test implemented in DESeq2, and adjusted P values reflect an FDR correction for multiple hypothesis testing. CIBERSORT (RRID:SCR_016955) was run on RNA-seq profiles with default parameters and using a mouse-specific immune cell gene signature matrix [ImmuCC; (47)]. GSEA was performed on RNA-seq data using the GSEA software SeqGSEA (RRID:SCR_005724) and Hallmark, BioCarta Pathways (RRID:SCR_006917), and mouse MDSC gene sets (48, 49).

Antibodies from Bristol Myers Squibb were isotype control (anti–DT1-D12 mIgG1 clone 4F7_RAS_Ab U2017010. AB107918.13 1469/Diptheria) and anti–PD-L1 (PDL1 14D8 Chimeric_RAS_Ab_05 U20161214. AB294451.01 1105/B7-H1).

Anti-mouse PD-L1 and isotype controls were each given i.p. at 5-mg/kg dosages. Mice began treatment 7 days after ETBF colonization, and antibodies were administered twice a week for a 6-week tumorigenesis terminal time point.

Raw next-generation sequencing data used in these studies are available through NCBI under the following accession numbers: PRJNA687231, PRJNA687232, PRJNA687233, PRJNA687236, and SRP105286.

J.R. White reports personal fees from Resphera Biosciences, LLC outside the submitted work. J. Michel reports grants from the NIH outside the submitted work. R.A. Anders reports grants and personal fees from Bristol Myers Squibb, personal fees from Merck SD and Incyte, and grants and personal fees from RAPT Biotech outside the submitted work. H.M. O'Hagan reports grants from the NIH and the Indiana Clinical and Translational Sciences Institute during the conduct of the study. F. Housseau reports grants from the NIH/NCI and Swim Across America during the conduct of the study. C.L. Sears reports grants from Bloomberg Philanthropies and Cancer Research UK Grand Challenge during the conduct of the study, as well as grants from Janssen and Bristol Myers Squibb outside the submitted work. No disclosures were reported by the other authors.

C.E. DeStefano Shields: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J.R. White: Formal analysis, writing–review and editing. L. Chung: Investigation. A. Wenzel: Investigation. J.L. Hicks: Investigation, methodology. A.J. Tam: Investigation. J.L. Chan: Investigation. C.M. Dejea: Investigation. H. Fan: Investigation. J. Michel: Investigation. A.R. Maiuri: Investigation. S. Sriramkumar: Investigation. R. Podicheti: Investigation. D.B. Rusch: Investigation.H. Wang: Formal analysis. A.M. De Marzo: Resources, methodology. S. Besharati: Investigation. R.A. Anders: Supervision, investigation. S.B. Baylin: Conceptualization, resources, supervision, funding acquisition, investigation, writing–review and editing.H.M. O'Hagan: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. F. Housseau: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. C.L. Sears: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing.

We thank Dr. David L. Huso for contributions in evaluating the pathology of this model, the Indiana University Center for Genomics and Bioinformatics and the Indiana Molecular Biology Institute for their assistance, members of the Sidney Kimmel Comprehensive Cancer Center Experimental and Computational Genomics Core (supported by NCI grant P30CA006973) for their support with RNA-seq, and the Johns Hopkins University Oncology Tissue Services for tissue processing and slides (supported by NCI grant P30CA006973).

Support for this research was provided from the following grants: National Institute of Environmental Health Sciences, National Institutes of Health grants R01 ES023183 (to H.M. O'Hagan) and R01 ES011858 (to S.B. Baylin); Bloomberg Philanthropies, Cancer Research UK grant C10674/A27140 and institutional resources from JHU SOM and DOM (to C.L. Sears); Swim Across America and National Cancer Institute, National Institutes of Health grant R01 CA203891 (to F. Housseau).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.