Investigation of the gut microbiome, bile acid composition and host immunoinflammatory response in a model of azoxymethane-induced colon cancer at discrete timepoints

Abstract

Background

Distinct sets of microbes contribute to colorectal cancer (CRC) initiation and progression. Some occur due to the evolving intestinal environment but may not contribute to disease. In contrast, others may play an important role at particular times during the tumorigenic process. Here, we describe changes in the microbiota and host over the course of azoxymethane (AOM)-induced tumorigenesis.

Methods

Mice were administered AOM or PBS and were euthanised 8, 12, 24 and 48 weeks later. Samples were analysed using 16S rRNA gene sequencing, UPLC-MS and qRT-PCR.

Results

The microbiota and bile acid profile showed distinct changes at each timepoint. The inflammatory response became apparent at weeks 12 and 24. Moreover, significant correlations between individual taxa, cytokines and bile acids were detected. One co-abundance group (CAG) differed significantly between PBS- and AOM-treated mice at week 24. Correlation analysis also revealed significant associations between CAGs, bile acids and the bile acid transporter, ASBT. Aberrant crypt foci and adenomas were first detectable at weeks 24 and 48, respectively.

Conclusion

The observed changes precede host hyperplastic transformation and may represent early therapeutic targets for the prevention or management of CRC at specific timepoints in the tumorigenic process.

Introduction

Colorectal cancer (CRC) is the third most commonly diagnosed malignancy and the second leading cause of cancer-related death worldwide [1]. Most cases are sporadic in nature (~75%) and occur in people without a genetic predisposition or a family history of CRC [2]. Recent studies have implicated the intestinal microbiome in the pathogenesis of CRC [3–5]. In healthy subjects, the gut is primarily populated by a core microbiota composed of obligate anaerobes belonging mainly to the phyla Firmicutes and Bacteroidetes, and to a lesser extent to Actinobacteria, Proteobacteria and Verrucomicrobia [6]. Analysis of community diversity and richness indices based on 16S rRNA gene sequencing has shown significant alterations in microbial diversity both at the site of the primary tumour and in faecal samples from CRC patients [3, 7]. Understanding the role of the human gut microbiota in colon cancer, however, has largely depended on examining patients already presenting with CRC. To determine temporal changes in the gut microbiota at different stages of human colon cancer development, some studies have examined the microbiota profile in patients with intestinal polyps [8], with others examining the microbiota at different stages of the tumorigenic process [3, 5, 9]. Aberrant crypt foci (ACF) are thought to be the earliest identifiable lesion in the colon carcinogenic process [10], with microbiome changes associated with ACF recently identified in a human study [11]. The role of the gut microbiota in the progression from healthy to adenoma to CRC, however, is undoubtedly multifactorial and can affect the various stages of the tumorigenic process. This represents a significant challenge for human-based studies. Further research in experimental animal models is necessary to better understand the mechanisms that underlie the association between the gut microorganisms and CRC.

One mechanism by which the gut microbiota may affect colon carcinogenesis is the production or modification of metabolites such as bile acids [12]. Bile acids are endogenous, amphipathic molecules, which facilitate uptake of dietary fats, and have been implicated in colon carcinogenesis [13]. For example, administration of cholic acid (CA), a primary bile acid, increased the incidence of colonic tumours in rats treated with genotoxic azoxymethane (AOM) [14]. In contrast, ursodeoxycholic acid (UDCA) reduced tumour burden [14], suggesting a dual role for bile acids in the tumorigenic process. Therefore, understanding the bile acid-gut microbiome axis in the development of colon cancer may reveal a dynamic mechanism by which the gut microbiota could influence cancer risk.

CRC-associated microbial communities also differentially correlate with the expression of host immunoinflammatory response genes [3]. Inflammation is a well characterised risk factor for CRC and a controlled inflammatory response is critical for immune protection against cancer [15]. Evidence suggests that the microbiota can influence colonic inflammation, and the microbiota is, in-turn, influenced by host inflammatory processes, resulting in complex reciprocal interactions [16, 17]. This is further supported by observations of microbial regulation of cytokines and chemokines in mouse models of CRC [18, 19]. Several cytokines, including interleukin 1β (IL-1β), interleukin 6 (IL-6) and tumour necrosis factor α (TNFα), have been shown to protect against cancer development in some circumstances, while contributing to tumour initiation and progression in others [20–22]. This highlights the importance of understanding the temporal expression profiles of these cytokines.

Here, we performed a time-course study in a C57BL/6 J mouse strain with a prolonged period of disease onset following administration of AOM to establish the temporal sequence of events during tumorigenesis involving the microbiota, bile acid metabolism and expression of host immunoinflammatory genes. We identified distinct changes in both the immune and bile acid profiles, as well as particular microbial signatures that varied from the initial genotoxic insult to the appearance of pre-malignant disease and observed significant interactions between these factors.

Materials and methods

Reagents

AOM (Merck, Darmstadt, Germany); RNAlater (Merck); Tetro cDNA synthesis kit (Bioline, Nottingham, UK); SensiFAST No-ROX kit (Bioline); QIAamp Fast DNA Stool Kit (Qiagen, Manchester, UK); nucleic acid probes from Roche Universal Probe Library; Custom oligo qPCR primers (Eurofins Genomics, Ebersberg, Germany).

Animals and study design

Animal experiments were conducted in accordance with the regulations and guidelines of the Irish Department of Health following approval by the University College Cork Animal Experimentation Ethics Committee (2011/023).

In this study, a total of 64 female C57BL/6JOlaHsd mice (6–8 weeks of age; Envigo, Blackthorne, UK) were housed in a specific pathogen-free facility on a 12-h light/dark cycle at 22 °C with access to water and chow ad libitum. After acclimatisation, equal numbers of mice were randomly assigned to two groups and were administered an intraperitoneal (i.p.) injection of 10 mg/kg AOM to induce tumorigenesis (n = 32) each week for five consecutive weeks while control mice received phosphate buffer saline (PBS; n = 32). Following necropsy, each sample was allocated a random number, to which the subsequent investigators were blinded. To help to prevent horizontal microbiome transmission between co-housed mice, mice were group housed in two cages per treatment group (PBS or AOM) per cull time. Timed culls were performed at 8, 12, 24 and 48 weeks (n = 8 per group at each timepoint) following AOM or PBS administration and mice were euthanised by cervical dislocation. The experimental unit was considered a single animal. Sample size was calculated using g*power, with the standard deviation and magnitude of difference calculated from previous studies quantifying medium to large ACF development in response to AOM as an endpoint.

Faecal 16S rRNA sequencing

DNA was extracted from faeces using the QIAamp Fast DNA Stool Kit as per the manufacturer’s instructions with the addition of a bead-beating step.

The V3-V4 variable region of the 16S rRNA gene was amplified from each extracted DNA sample according to the 16S metagenomic sequencing library protocol (Illumina, San Diego, CA, USA) and sequenced on an Illumina MiSeq. See Methods in Supplementary Material.

Generating co-abundance groups

To identify patterns in the variation of the microbiota, a set of co-abundance groups (CAGs) were determined by clustering operational taxonomic units (OTUs) by the correlation of their relative abundances. Initially, OTUs were trimmed to remove taxa present in less than 20% of samples and all non-prokaryotic taxa. A matrix of Kendall’s Tau values was then generated for each pair of OTUs, and these values were clustered by Ward-linkage according to their Pearson’s correlation coefficient and visualised using the Made4 package in R. Each cluster of taxa was then assigned to a CAG [23].

Ultra-performance liquid chromatography—mass spectrometry (UPLS-MS)

Faecal samples were used for analysis of bile acids. UPLC-MS was performed as described [24]. Briefly, five microliters of extracted bile acids were injected onto a 50-mm T3 Acquity column and analysed in negative electrospray mode by an LCT Premier mass spectrometer (Waters, Dublin, Ireland). Each analyte was identified according to its mass and retention time. Standard curves were performed using known bile acids, and each analyte was quantified according to the standard curve and normalised according to the deuterated internal standards.

Cell line maintenance

HT29 cells were obtained from ATCC and maintained in DMEM supplemented with 10% foetal calf serum (FCS), and 1% penicillin/streptomycin solution in a 37 °C, 5% CO₂ humidified incubator. Cells are routinely tested for mycoplasma contamination.

Faecal water preparation

Faecal water was prepared by suspending 0.3 g faeces in 1 mL PBS and subjected to bead beating for 15 s before centrifugation for 10 min. The supernatant was collected and stored at −20 °C before use. Faeces were pooled based on treatment and timepoint to generate sufficient material. HT29 cells were serum starved (0.5% FCS) overnight and then exposed to faecal water (1:10 dilution) for 4 h.

Quantitative real-time PCR

RNA extraction was performed using the GenElute Mammalian Total RNA kit (Merck) as per the manufacturer’s instructions and converted to cDNA using the Tetro cDNA Synthesis Kit (Bioline). Genes were amplified using primers matched to the appropriate hydrolysis probe from the Roche probe library (Supplemental Table 2) in a LightCycler 480 for 40 PCR cycles. Relative transcription was calculated using the 2^–ΔΔCT method standardised to the average of the control group ΔCT [25]. Human CXCL1 was amplified using a primer-probe combination from Integrated DNA Technologies (Iowa, USA).

Statistics

Statistics were performed in SPSS Version 24 (Chicago, IL, USA), GraphPad version 9 (San Diego, CA, USA) and R Version 3.5.0 using the Made4, vegan, pairwiseAdonis, compareGroups, phyloseq and ggplot packages. Statistical significance was set to p < 0.05. Benjamini–Hochberg FDR adjustment for multiple comparisons was applied where noted, with a false discovery rate set to 5%. Outliers were detected using Grubbs’ test. Normality was determined by the Shapiro–Wilk test. Groups were compared by two-tailed student’s t-test or MWU-test. Where the F-value was statistically significant, data were analysed using the Welch t-test. For HT29 cell analysis, a one-tailed student’s t-test was performed. Permutational ANOVA (PERMANOVA) was used to compare β-diversity and CAGs, using unweighted Unifrac and Euclidean-squared distance matrices, respectively. Correlations were examined using Pearson’s R² and Spearman’s R correlation coefficients. Throughout, asterisks denote significance where * represents p < 0.05, **p < 0.01 and ***p < 0.001.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Results

Macroscopic and microscopic changes in response to the carcinogen, AOM

Given that CRCs occur sporadically in most cases [2], we chose AOM alone to mimic human sporadic CRC development [26]. Moreover, as different mouse strains have been shown to exhibit differential sensitivity to AOM-induced tumorigenesis [26], we chose C57BL/6J mice, which display a lower sensitivity to AOM, to improve the temporal resolution of our study. Consistent with this approach, no signs of hyperplastic or neoplastic transformations were observed in mice at either 8- or 12-weeks post-AOM administration. Moreover, AOM-treated mice gained less weight than PBS-treated mice over the course of the study, and this difference in bodyweight-gain was significant at week 48 with AOM-treated mice weighing on average 3.4 g less than PBS-treated mice (t-test, p < 0.05). Faecal occult blood (FOB) and ACF were first observed at week 24 (Table 1). FOB was also apparent in the faeces in three out of eight AOM-treated mice prior to week 48. Two of these mice harboured at least one colonic adenoma (Table 1).

Table 1.

Incidence of faecal occult blood (FOB), aberrant crypt foci (ACF) and adenomas.

		Week 8	Week 12	Week 24	Week 48
PBS	FOB	0/8	0/8	0/7	0/8
	ACF	0/8	0/8	0/7	0/8
	Adenoma	0/8	0/8	0/7	0/8
AOM	FOB	0/8	0/7	1/8	3/8
	ACF	0/8	0/7	1/8	0/8
	Adenoma	0/8	0/7	0/8	2/8

Temporal microbiota changes in response to AOM

Shannon index (Fig. 1a) was used to assess the α-diversity and evenness of the gut microbiota in faecal samples from each experimental group. The observed species (OS; Fig. 1a) index was used to estimate microbial richness, and the phylogenetic diversity (PD; Fig. 1a) was also determined at each timepoint. Analysis of AOM- versus PBS-treated mice revealed that the α-diversity of the microbiota was altered very early in the tumorigenic process. At week 8, OS and PD were significantly increased in AOM-treated mice (Fig. 1a; MWU-test p < 0.05 after FDR adjustment), suggesting that there is an increase in diversity within the AOM group at this time. Beta (β)-diversity (Fig. 1b), which compares samples based on overall bacterial community composition across groups, also differed significantly between AOM- and PBS-treated mice at week 8 (Fig. 1a; PERMANOVA of unweighted Unifrac, p = 0.008, R² = 0.153). The dominant phyla in both AOM- and PBS-treated mice were Bacteroidetes and Firmicutes. Although the abundance of these phyla did not change significantly, we did observe a significant increase in Proteobacteria in AOM-treated mice (Fig. 2a).

Fig. 1. Alpha (α) and beta (β) diversity are altered across time in mice treated with either PBS or AOM.

Faecal samples were collected throughout the experiment and analysed by 16S rRNA gene sequencing 8-, 12-, 24- and 48-weeks following administration of either PBS or AOM. α-diversity was measured using Shannon, Observed Species and Phylogenetic Diversity (PD) *p < 0.05 after FDR correction. Data are presented as median (IQR) (a). PERMANOVA of unweighted Unifrac distances were used to examine the β-diversity visualised by principal coordinate analysis (b). n = 7–8 per group.

Fig. 2. Histograms of the community composition of gut microbiota at the phylum level and co-abundance groups (CAGs).

The impact of AOM on the major phyla in faecal samples of mice at 8-, 12-, 24- and 48-weeks following administration of either PBS or AOM (a). Each bar chart represents the average reads of the group (n = 7–8). Each phylum is expressed as a percentage of the total number of reads for the particular group. Species with a relative abundance of less than 1% were classified as unassigned. Bar charts showing the proportion of specific CAGs detected in AOM and PBS-treated groups (b). Seven CAGs were identified and PERMANOVA determined that all CAGs were significantly different (p < 0.05 after FDR adjustment). n = 7–8.

Of the significantly changed genera, the majority were among taxa corresponding to Firmicutes, with changes also observed within Proteobacteria and Actinobacteria. At week 8, there was a significant suppression of Lactobacillus and an increase in Olsenella in AOM-treated mice (Fig. 3a; MWU-test, p < 0.05 after FDR adjustment). Alterations in these taxa have previously been correlated with colon cancer [27, 28]. This decrease in Lactobacillus could account for some of the changes observed in the bile acid pool at this timepoint due to its role in bile acid metabolism, although we did not observe any correlation between bile acids and these bacterial genera. Since community structure can be more informative than abundance differences of individual taxa, we next analysed the microbiota by determining CAGs. The taxon composition of each CAG can be found in Supplemental Table 3. However, we observed no significant difference between CAGs in either PBS- or AOM-treated groups at week 8 (Fig. 2b; t-test and MWU-test, p > 0.05 after FDR adjustment).

Fig. 3. Taxa which differed significantly in their abundance between groups.

From the data acquired by 16S RNA gene sequencing, operational taxonomic units were clustered based on 97% sequence similarity and taxonomy was assigned using BLAST against the SILVA SSURef data base. Only significantly different taxa are presented present in AOM- and PBS-treated mice at weeks 8, 12, 24 and 48 here (a–d). Data are presented as the z-scores of the abundances scaled by row. Taxa highlighted in red font represent taxa that were altered at two or more timepoints. n = 7–8 per group. *p < 0.05, **p < 0.01, ***p < 0.001 after FDR adjustment.

At Week 12, there were no changes in α-diversity but β-diversity differed significantly between treatments (Fig. 1b; PERMANOVA of unweighted Unifrac distances, p = 0.017, R² = 0.149). At the phylum level, no alterations were detected (Fig. 2a). There were only nine individual genera that differed significantly between PBS- and AOM-treated mice (Fig. 3b). Moreover, as observed at week 8, there were no significant differences between CAGs (Fig. 2b; t-test and MWU-test, p > 0.05 after FDR adjustment).

By week 24, no additional changes in α-diversity were observed (Fig. 1a). However, β-diversity differed significantly between AOM- and PBS-treated mice at weeks 24 and 48 (Fig. 1b; PERMANOVA of unweighted Unifrac distances, p = 0.001 and p = 0.01, respectively). While there were no significant changes at the phylum level at week 24 (Fig. 2a), changes were observed in specific members of Firmicutes, Bacteroidetes, Proteobacteria and Tenericutes. Of the genera enriched in AOM-treated mice, Oscillibacter has previously been associated with increased cancer risk [3] (Fig. 3c). At this timepoint we observed the first significant differences in the CAGs, with CAG5 significantly decreased in AOM-treated mice (Fig. 2b; t-test, p < 0.001 after FDR adjustment). This CAG is dominated by Bacteroidetes, which comprise >90% of its abundance.

At week 48, there was a significant increase in Actinobacteria and a significant reduction in Verrucomicrobia in AOM-treated mice (Fig. 2a). Of the significantly changed genera at this timepoint, Akkermansia was the only member of the Verrucomicrobia phylum that was decreased (Fig. 3d). Similarly, for Actinobacteria, Bifidobacterium was the only member of this phylum that was increased. Moreover, at this timepoint we observed the greatest number of significantly altered genera (Fig. 3d). These were predominantly Firmicutes (7/21), Proteobacteria (5/21) and Tenericutes (5/21).

Bile acid metabolism following AOM administration

Dysregulation of bile acids has been implicated in tumorigenesis [13]. Bile that is not re-absorbed in the small intestine is subjected to microbial transformation. First, bile salts are deconjugated from taurine and glycine by the bacterial enzyme bile salt hydrolase (bsh) to form free bile acids. Unconjugated primary bile acids (mainly cholic acid (CA), chenodeoxycholic acid (CDCA) and muricholic acid (MCA) in mice) are converted into secondary bile acids, such as deoxycholic acid (DCA), lithocholic acid (LCA), and ursodeoxycholic acid (UDCA) by bacterial 7α-dehydroxylase. Moreover, the apical sodium-dependent bile acid transporter (ASBT) is expressed on the apical membrane of enterocytes and mediates the reabsorption of bile acids from the intestine. With respect to the faecal bile acid analysis, the concentration of total bile acids was significantly reduced in AOM-treated mice at week 8 (Fig. 4a; t-test, p < 0.01). Both unconjugated primary (CDCA) and conjugated and unconjugated secondary bile acids (DCA, LCA, T-LCA) were also reduced at this time (Fig. 4c). Interestingly, the proportion of the hydrophobic cytotoxic bile acids, DCA and LCA (Fig. 4c; t-test, p < 0.001 after FDR adjustment), were significantly decreased in the faeces of AOM-treated mice at week 8.

Fig. 4. Alterations in faecal bile acid profiles and transporter gene expression between AOM and PBS-treated mice.

Bile acids (a, c–f) were measured by UPLC-MS and ASBT (b) was measured by qRT-PCR 8-, 12-, 24- and 48-weeks following administration of either PBS or AOM. Faecal bile acid levels are presented as absolute values. Data are presented as mean ± SEM. n = 7–8 per group. *p < 0.05, **p < 0.01, ***p < 0.001.

The total amount of bile acids in the faeces, as well as the levels of DCA and LCA remained significantly reduced at week 12 (Fig. 4a, d; t-test, p < 0.001 after FDR adjustment). This agrees with previous work that demonstrated a similar pattern for LCA and DCA in a colitis-associated model of colon cancer [29]. Moreover, faecal waters from patients with colon cancer had decreased levels of DCA, LCA and cholate relative to healthy controls [30].

The concentrations of primary unconjugated (CDCA, β-MCA) and conjugated (T-CA, T-β-MCA) faecal bile acids were increased in AOM-treated mice at week 24, of which β-MCA was the most abundant (Fig. 4e; t-test, p < 0.05 after FDR adjustment). There was also an increase in both taurine-conjugated primary and secondary bile acids. In contrast, LCA was significantly reduced (Fig. 4e; t-test, p < 0.05 after FDR adjustment). Moreover, expression of ASBT was also significantly reduced (t-test, p < 0.01). By week 48, there was no significant change in the bile acid profile (Fig. 4a, f).

Colonic inflammatory response to AOM

Given that inflammation is a risk factor for colon cancer and can influence colon carcinogenesis, we measured the transcription of several cytokines and chemokines in the distal large intestine (Fig. 5a). Significant increases in cytokine and chemokine expression patterns become apparent at week 12, with the expression of TNFα, IL-1β, IL-12, CXCL1, CXCL2 and CXCL5 significantly up-regulated in AOM-treated mice (Fig. 5; t-test, p < 0.05, p < 0.01, p < 0.001 after FDR correction). TNFα remained elevated at week 24, together with an increase in gene expression of IL-6, IL-1β and IL-12 (Fig. 5a; t-test, p < 0.05 after FDR correction). However, these alterations in cytokine and chemokine gene transcription were absent by week 48. Consistent with these findings, human colonic tumour cells display a similar temporal response following stimulation with faecal waters isolated from AOM-treated mice relative to PBS-treated mice at the same timepoints. At weeks 12 and 24, CXCL2 and CXCL1 were significantly increased in the HT29 cells, respectively. In this human cell line, the changes in immunoregulatory gene expression were also increased following stimulation with faecal water prepared from week 48, but these changes were not significant.

Fig. 5. Alterations in immunoregulatory gene expression between AOM and PBS-treated mice.

Cytokines and chemokines were measured by qRT-PCR from mice at 8-, 12-, 24- and 48-weeks following administration of either PBS or AOM. n = 7–8 except for samples where the values were below the detection threshold (a). HT29 cells were stimulated for four hours with faecal waters derived from PBS- or AOM-treated mice (1:10 dilution) and changes in gene expression measured by qRT-PCR (n = 3) (b). The heat maps depict fold-change in gene expression. *p < 0.05, **p < 0.01, ***p < 0.001 after FDR correction. White box in panel (a) depicts gene expression that was greater than the fold-change range for the other genes but was not significantly different between treatment groups.

Correlation analyses between the gut microbiota and immunoinflammatory cytokines and bile acid composition

Correlation analyses were performed to identify any relationships between the microbiome, cytokine transcription and the bile acid pool. At week 8, Allobaculum negatively correlated with IL-12 (Spearman, p < 0.001, R = −1), Coriobacteriaceae_uncultured negatively correlated with CXCL1 and IL-6 (p < 0.001, R = −0.97; p < 0.001, R = −0.95, respectively) and Defluviitaleaceae_uncultured negatively correlated with IL-1β (Spearman, p < 0.001, R = −0.95). No additional correlations between individual taxa and cytokines were observed at any other timepoint in AOM-treated mice. At weeks 12 and 24, we observed significant correlations between bile acids and individual taxa (Supplemental Table 4). With the exception of Parasutterella, which positively correlated with T-UDCA (Spearman, p < 0.001, R = 1), all other correlations were negative (Supplemental Table 4). We also investigated whether any correlations existed between CAGs, inflammatory genes, bile acids, and ASBT expression. CAG2 negatively correlated with DCA at week 24 (Spearman, p < 0.001, R = −1). At week 48, CAG4 (Spearman, p < 0.001, R = 0.952) and CAG5 (Spearman, p < 0.01, R = 0.81) positively correlated with ASBT expression, while CAG8 negatively correlated (Spearman, p < 0.05, R = −0.857).

Discussion

Decreased α-diversity in the faecal microbiome has been described as a characteristic feature of CRC [7, 31]. However, studies have also reported an increase in α-diversity in patients with colon cancer [32, 33]. This divergence may be associated with the stage of the disease [31]. In support of this, our data suggest that carcinogenesis begins with an increase in α-diversity as characterised by an increase in PD and observed species in AOM-treated mice at week 8. This suggests that there are more OTUs present in the AOM group, which are further away from each other on the phylogenetic tree. The presence of a small number of highly divergent OTUs would greatly impact PD without having a major impact on richness. When considering which taxa might account for the increase in PD but not richness, we identified multiple taxa, which were either present in one treatment group and absent in the other or vice versa at week 8. Most of these OTUs fell within the phyla Bacteroidetes (uncultured), Firmicutes (e.g., Erysipelotrichaceae, Roseburia, Blautia) or Tenericutes (Mollicutes). From week 12 onward, this change in α-diversity was no longer apparent. However, by week 12, a significant upregulation of pro-inflammatory signalling was detected in mice administered AOM. At this timepoint, the microbial communities became more similar, and one possible explanation may be that the increase in α-diversity observed early in the tumorigenic process is due to the growth of opportunistic pathogens and is reversed later once the immune response becomes more active.

In this context, the opportunistic pathogen Clostridium sensu stricto, which was decreased at week 8 in our study, and then increased at week 12, occurred at a time where the inflammatory response also significantly changed in our mouse model of colon carcinogenesis. A similar positive relationship between this taxon and inflammation was shown in a mouse model of inflammatory bowel disease (IBD) [34]. Other taxa that changed over time included Turicibacter, which was decreased at week 8 but increased at week 48 in our model. While little is known about a direct role for this genus in inflammation and colorectal cancer, its abundance was previously found to be decreased in TNFα-deficient mice compared to wildtype mice [35]. Other studies have also suggested that this genus benefits from a pro-inflammatory environment, which is consistent with the absence of an overt inflammatory response in our model at week 8 [36]. Finally, Marvinbryantia, with the exception of week 12, was significantly decreased over time. This genus has also been shown to be reduced in CRC [37]. Moreover, in a rat model of colitis, Marvinbryantia was significantly increased in response to feeding with resistant starch. This was associated with decreased tumour multiplicity, increased short chain fatty acid production and reduced proliferation and inflammation, suggesting that it may have anti-inflammatory and protective properties [38]. These data reflect the dynamic interplay between inflammation and the microbiome and suggest that temporal changes in the abundance of specific genera may be dependent on the host inflammatory response at particular timepoints during the course of tumorigenesis.

Turicibacter, the second most abundant member of CAG5, correlates with the activity of Slc10a2, which encodes ASBT and helps recycle bile acids from the small intestine back to the liver [39]. Moreover, expression of ASBT is sensitive to changes in the microbiome [40]. We observed that ASBT expression tended to follow a similar dynamic pattern over time as total bile acids, but there was no significant correlation after FDR correction between the expression of this transporter and the bile acids measured. However, of the 12 faecal bile acids measured, eight were significantly increased when expression of ASBT was significantly decreased, in particular conjugated bile acids that have high substrate specificity for this transporter [41]. This in turn could potentially lead to increased luminal concentrations of the farnesoid X receptor (FXR) antagonist, T-βMCA, which is normally transported by ASBT, and this, coupled with an increase in the ASBT-independent transport of the FXR agonist, CDCA, may potentially influence tumorigenesis through alterations in FXR signalling.

The presence of high levels of bile acids also suppresses the activity of Turicibacter [39]. Moreover, studies examining the effect of Turicibacter sanguinis on bile acid metabolism and transformation suggest that this human isolate can deconjugate T-CA and transform CA and CDCA by the action of the microbial bile acid-metabolising enzyme 7α-hydroxysteroid dehydrogenase [39]. Therefore, fluctuations in the abundance of this taxon over time might not only affect bile acid metabolism, but the bile acid profile at any given timepoint could also influence the abundance of Turicibacter. The next most abundant members of CAG5 are Parasutterella and Bifidobacterium. Of these, Bifidobacterium display sensitivity to fluctuations in bile acids, in particular to the toxic effects of the secondary bile acids such as DCA [42]. Moreover, colonisation with Parasutterella modified bile acid metabolites, thereby impacting bile acid transport and synthesis [43]. Taken together, these findings suggest that significant changes in the relative abundance of individual taxa responsible for bile acid metabolism and modification may not necessarily be reflected in the expected faecal bile acid profile, or vice versa, particularly in the context of a more complex ecosystem in which members not only metabolise bile acids but are also sensitive to their growth inhibitory effects.

The only correlation between CAGs and bile acids was between the secondary bile acid, DCA and CAG2 and this occurred at week 24. DCA also significantly correlated with Bacteriodes at this timepoint. DCA has been shown to be toxic and inhibit the growth of Bacteriodes [44]. However, we observed no significant difference in either this taxon or in the faecal levels of DCA at this timepoint. This taxon is also the most abundant member of CAG2, which, in turn, was also negatively associated with DCA. While the relative abundance of the taxon within the CAG did not change, another member of this CAG, Bacteroidales S24-7, showed a fivefold reduction in AOM-treated mice relative to control mice. Moreover, this taxon was significantly less abundant in AOM-treated mice at this timepoint (see Fig. 3). Bacteroidales S24-7 has previously been shown to positively correlate with caecal levels of T-DCA in a mouse model of liver regeneration [45]. Of note, levels of Bacteroidales have been shown to be significantly reduced in patients with CRC relative to healthy controls [46].

Expression of the bile acid transporter ASBT positively correlated with CAG5 at week 48. At this timepoint the abundance of Bifidobacterium also significantly increased in AOM-treated mice (see Fig. 3). Moreover, within CAG5, the abundance of this taxon increased threefold. Bifidobacteria have bile acid-deconjugating activity, which is consistent with the normalisation of conjugated bile acids at week 48, relative to week 24, when the abundance of this species significantly increased. Notably, expression of the bile acid receptor FXR is downregulated in human colorectal tumours and colon cancer cell lines [47]. Moreover, administration of tauro-conjugated β-MCA, which is an FXR antagonist, accelerated tumour growth and increased the serum levels of pro-inflammatory cytokines in APC^MIN mice [48]. Given the ability of bifidobacteria, which are significantly more abundant at week 48, to deconjugate the endogenous FXR antagonist T-β-MCA and relieve its FXR antagonism in mice, this is consistent with a possibly pro-tumorigenic effect of T-β-MCA and FXR 24 weeks after AOM administration.

CAG4 also positively correlated with ASBT expression at week 48. Although the most abundant taxon within CAG4 was Bacteroidales S24- 7, this was comparable within the CAG for PBS- and AOM-treated mice. However, Akkermansia, the second most abundant taxon within this CAG, differed between treatments both at the taxon level (see Fig. 3) and within CAG4 (approx. 2.5-fold reduction in AOM-treated mice). A. muciniphila is one of the most studied species of this genus and displays sensitivity to several bile salts [49], further highlighting the complexity and inter-relationship between bacterial metabolites and the composition of the microbiome. We also observed a negative correlation between the phylum to which this genus belongs, Verrucomicrobia and the genus itself at weeks 12 and 24 with UDCA, and T-CA and T-UDCA, respectively. Studies have demonstrated that these bile acids do not affect the growth of A. muciniphila [49], suggesting that this relationship may be driven by other members of this phylum.

In contrast, CAG8 negatively correlated with ASBT expression at week 48. The most abundant members of CAG8 are Blautia, Ruminococcaceae_uncultured and Lachnospiraceae, which have all been associated with secondary bile acids and bile acid deconjugation [50, 51]. Of these three taxa, only the abundance of Lachnospiraceae in CAG8 was differentially altered between PBS- and AOM-treated mice (1.5-fold reduction). Moreover, members of the Lachnospiraceae family negatively correlated with CDCA in IBD [52]. However, in a model of liver regeneration no relationship between either individual caecal bile acids or ASBT was observed with Lachnospiraceae [45]. Despite the bile acid-metabolising activity of this taxon, we did not observe any significant relationship between Lachnospiraceae and individual bile acids. Moreover, the decrease in abundance is consistent with observations that this taxon is significantly reduced in the gut of individuals with CRC [46]. Given that ASBT expression and faecal bile acid profile had normalised at week 48 in our study, the role of this CAG in bile acid transport and metabolism, respectively, is unclear.

Of the genera that correlated with inflammatory gene expression, these occurred at week 8 when there was no obvious change in inflammation. Of note, Allobaculum negatively correlated with the expression of IL-12. Little is known about Allobaculum, although studies have shown that it is increased in IBD [53] and with Th 17 cell activity [54], suggesting that this taxon may be pro-inflammatory. IL-6, IL-1β and CXCL1 have all been linked to colon carcinogenesis and at week 8, both IL-6 and CXCL1 negatively correlated with Coriobacteriaceae, while Defluviitaleaceae negatively correlated with IL-1β. Despite these correlations, however, the only significant alterations in inflammatory gene expression occurred at weeks 12 and 24. Previous studies have proposed that particular bacterial clusters or CAGs may be more important in colon tumorigenesis than individual taxa [3]. It could be argued in our study that CAG1 is pathogenic, given that it contains Citrobacter [55], Hydrogenoanaerobacterium [56] and Anaeroplasma [57], which have been associated with colon cancer. However, the most abundant member of this CAG belongs to the uncultured Clostridium vadinBB60 group, and therefore we could not definitively classify this CAG as pathogenic or pro-inflammatory.

Although we observed significant changes in both inflammatory factors and bile acids very early on in the tumorigenic process, few correlations were detected between these factors. However, there are other microbial drivers that could influence colonic tumorigenesis. For instance, gut microbiota-derived metabolites such as hydrogen sulphide and N-nitroso compounds have been implicated in CRC [58]. Moreover, oxidative stress and reactive oxygen species (ROS) have also been linked to CRC [59]. Indeed, continuous exposure of intestinal epithelial cells to high concentrations of secondary bile acids has been shown to induce the production of ROS and active nitrogen species [60]. However, given that we observed a significant decrease in DCA and LCA early in response to AOM, we can only speculate on the contribution of bile acid-induced ROS generation early in the tumorigenic process.

Early changes in the microbiome or microbiome-associated metabolites could potentially represent early biomarkers for CRC development. A recent human study has examined the microbiome of ACF and ACF with synchronous polyps, which likely reflects some of the earliest changes in the microbiome [11]. When these human samples were stratified by the presence of ACF alone versus those with ACF and polyp, two distinct microbial clusters were apparent, with compositional changes in Firmicutes predominating [11]. Of the significantly changed phyla at weeks 24 and 48 in our study, the majority also occurred within the Firmicutes phylum. Furthermore, in taxon-based analysis, the microbiota profile from patients with conventional adenomas was depleted in a network of Clostridia OTUs from families Ruminococcaceae, Clostridiaceae and Lachnospiraceae [61]. While we also saw significant changes in Clostridia OTUs from these families, these were divergent, with increases and decreases in abundance seen. Our study identifies early changes in the microbiome prior to tumour development and likely reflects the equivalent pre-malignant lesions (ACF and adenoma) in human studies.

Most of the significant changes in phyla occurred at week 48 and was characterised by a significant reduction in Verrucomicrobia and an increase in Actinobacteria. This reduction in Verrucomicrobia, if sustained, could be associated with improved outcome as a decrease in this phylum is associated with a reduction in tumour development, invasiveness, and inflammation in a mouse model of colon cancer [62, 63]. However, the alterations in Actinobacteria observed in our study do not appear to be reflected in human disease and therefore Actinobacteria is unlikely to reflect a pre-cancerous biomarker for CRC in human studies [64, 65]. Moreover, bile acids were also investigated for their potential as microbiome-associated biomarkers for the development of CRC [66], but the findings were unclear. Discrete taxonomic changes were also observed in our study, which may reflect a potentially beneficial pre-malignant microbiome biomarker. The composition of the microbiome has also been implicated in treatment response. For example, Vibrio and Psychrobacter, both of which are Gammaproteobacteria, were significantly reduced at all four timepoints in our study. Notably the presence of intra-tumoral Gammaproteobacteria in pancreatic cancer resulted in treatment resistance to the chemotherapeutic, gemcitabine [67]. Whether the changes in the faecal microbiome represent prognostic or predictive biomarkers for disease or treatment response warrants further investigation in CRC.

In conclusion, the first changes we observed in response to AOM treatment were microbial in nature, potentially pro-tumorigenic, and preceded inflammatory changes in the host. Concurrent alterations in the bile acid pool, possibly reflecting a reduction in microbial bile acid metabolism, were also significant in the earlier phase following AOM treatment. While a significant cytokine response ensued, this was largely ameliorated by week 48 when macroscopic adenomas appeared. Our study highlights the complexity of microbe–host interactions in the pathogenesis of colon cancer and the discrete events, which occur following a genotoxic insult. Improved understanding of these interactions could lead to better interventional strategies to suppress the development of colon cancer at key stages in the tumorigenic process.

Author contributions

JMK acquired data and played an important role in interpreting the results and drafted the manuscript. CJW, PC and KB acquired data. SM helped to design the work that led to the submission. PDC helped draft the manuscript, acquired data, and/or played an important role in interpreting the results. SAJ, CGMG, AH and NPH conceived and designed the work that led to the submission, played an important role in interpreting the results and drafted the manuscript. All authors approved the final version and agreed to be accountable for all aspects of the work.

Funding

This work was supported by the APC Innovation Platform. APC Microbiome Ireland is a research institute funded by Science Foundation Ireland (SFI) through the Irish Governments National Development Plan (Grant SFI/12/RC/2273).

Data availability

The datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors are not aware of any competing interests that might be perceived as affecting the findings of this study.

Ethics approval and consent to participate

Animal experiments were conducted in accordance with the regulations and guidelines of the Irish Department of Health following approval by the University College Cork Animal Experimentation Ethics Committee (2011/023).

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-022-02062-4.