Microbiota-derived tryptophan catabolites mediate the chemopreventive effects of statins on colorectal cancer

Epidemiological studies have indicated an association between statin use and reduced incidence of colorectal cancer (CRC), and work in preclinical models has demonstrated a potential chemopreventive effect. Statins are also associated with reduced dysbiosis in the gut microbiome, yet the role of the gut microbiome in the protective effect of statins in CRC is unclear. Here we validated the chemopreventive role of statins by retrospectively analysing a cohort of patients who underwent colonoscopies. This was confirmed in preclinical models and patient cohorts, and we found that reduced tumour burden was partly due to statin modulation of the gut microbiota. Specifically, the gut commensal Lactobacillus reuteri was increased as a result of increased microbial tryptophan availability in the gut after atorvastatin treatment. Our in vivo studies further revealed that L. reuteri administration suppressed colorectal tumorigenesis via the tryptophan catabolite, indole-3-lactic acid (ILA). ILA exerted anti-tumorigenic effects by downregulating the IL-17 signalling pathway. This microbial metabolite inhibited T helper 17 cell differentiation by targeting the nuclear receptor, RAR-related orphan receptor γt (RORγt). Together, our study provides insights into an anti-cancer mechanism driven by statin use and suggests that interventions with L. reuteri or ILA could complement chemoprevention strategies for CRC. Statins have anti-cancer effects that are modulated by the gut commensal Lactobacillus reuteri and the tryptophan metabolite, indole-3-lactic acid, in mice and humans.

Epidemiological studies have indicated an association between statin use and reduced incidence of colorectal cancer (CRC), and work in preclinical models has demonstrated a potential chemopreventive effect. Statins are also associated with reduced dysbiosis in the gut microbiome, yet the role of the gut microbiome in the protective effect of statins in CRC is unclear.
Here we validated the chemopreventive role of statins by retrospectively analysing a cohort of patients who underwent colonoscopies. This was confirmed in preclinical models and patient cohorts, and we found that reduced tumour burden was partly due to statin modulation of the gut microbiota. Specifically, the gut commensal Lactobacillus reuteri was increased as a result of increased microbial tryptophan availability in the gut after atorvastatin treatment. Our in vivo studies further revealed that L. reuteri administration suppressed colorectal tumorigenesis via the tryptophan catabolite, indole-3-lactic acid (ILA). ILA exerted anti-tumorigenic effects by downregulating the IL-17 signalling pathway. This microbial metabolite inhibited T helper 17 cell differentiation by targeting the nuclear receptor, RAR-related orphan receptor γt (RORγt). Together, our study provides insights into an anti-cancer mechanism driven by statin use and suggests that interventions with L. reuteri or ILA could c om pl em ent c he moprevention strategies for CRC.
The adenoma-carcinoma sequence accounts for 85-90% of sporadic colorectal cancer (CRC) 1 . Screening colonoscopy is vital for reducing CRC incidence by removing colorectal adenoma (CRA) before the malignant transformation. However, current screening programmes are insufficient to counter the rise in prevalence, and CRC remains the third most commonly diagnosed cancer worldwide 2 . Therefore, there is an unmet clinical need for safe and efficacious chemoprevention against CRC.
Nature Microbiology | Volume 8 | May 2023 | 919-933 920 Article https://doi.org/10.1038/s41564-023-01363-5 exposure independently suppressed CRC tumorigenesis, which had also been observed in previous studies [21][22][23] . This effect made it difficult to conclude on the role of gut microbiota because both antibiotic-treated groups have low tumour development similar to mice receiving atorvastatin alone. To better demonstrate the role of gut microbiota, we administered faecal microbiota from mice receiving atorvastatin or vehicle control to microbiota-depleted Apc Min mice (Extended Data Fig. 1p). Reduced tumour burden was observed in Apc Min mice receiving faecal microbiota from atorvastatin-treated mice (Fig. 1h-j and Extended Data Fig. 1q). These data indicate that the anti-cancer effect of atorvastatin was dependent on the microbiota.

Atorvastatin expands the gut commensal L. reuteri
We further analysed the gut microbiota dynamics after atorvastatin treatment by using 16S ribosomal RNA (16S rRNA) sequencing (Extended Data Fig. 2a). Principal coordinate analysis (PCoA) showed that gut microbiota was reshaped by atorvastatin (Fig. 2a). The alpha diversity of gut microbiota was increased after atorvastatin treatment (Fig. 2b,c). The bacterial composition was further analysed by a linear discriminant analysis effect size (LEfSe) algorithm (Fig. 2d) 24 . To explore the clinical importance of these altered genera, we overlapped them with our previously identified genera that changed in patients during CRC development 25 . Lactobacillus, the most enriched genus in Apc Min mice receiving atorvastatin, was also enriched in healthy individuals and significantly decreased during CRC development, indicating a beneficial role of this genus ( Fig. 2d and Extended Data Fig. 2b). In contrast, no common genus was found between those enriched in Apc Min mice receiving vehicle control and those that increased during CRC development ( Fig. 2d and Extended Data Fig. 2b). Thus, we decided to investigate the role of Lactobacillus in the chemopreventive effect of atorvastatin. We quantified the changes of 9 common Lactobacillus species after atorvastatin treatment by real-time polymerase chain reaction (PCR). L. reuteri was the only species that stably increased ( Fig. 2e and Extended Data Fig. 2c). Additionally, we found that the increased L. reuteri abundance could be transferred during microbiota transplantation (Extended Data Fig. 2d).
To corroborate our findings in humans, we conducted a 5 d exploratory trial in 20 healthy volunteers (Supplementary Table 2). Participants were given atorvastatin and were monitored closely for adverse events (Extended Data Fig. 2e). No adverse event was observed throughout the trial. The abundance of L. reuteri was significantly increased after atorvastatin treatment (Fig. 2f).
To determine why L. reuteri abundance was increased after atorvastatin intervention, we first observed the direct effect of atorvastatin on L. reuteri growth. However, the growth of L. reuteri in vitro was not promoted by atorvastatin (Extended Data Fig. 2f), implying that the increased L. reuteri abundance in vivo required a complex intestinal microenvironment. Next, we performed RNA-sequencing on colonic epithelium and overlapped our results with the expression profile of GSE81375. The overlapping analysis found 41 commonly downregulated genes and 15 commonly upregulated genes after atorvastatin treatment (Fig. 2g). Among these genes (Supplementary Table 3), we noticed Ido1 which had been reported to be associated with the abundance of L. reuteri 16,26 . IDO1 metabolizes 99% of dietary tryptophan that is not used for protein synthesis in mammals 16 . Therefore, host IDO1 activity determines the tryptophan availability in the gut and the abundance of bacteria able to adapt tryptophan as an energy source, such as L. reuteri 16 . Real-time PCR analysis confirmed the reduced Ido1 expression in intestinal epithelium after atorvastatin treatment ( Fig. 2h and Extended Data Fig. 2g). Consistently, we observed that atorvastatin treatment reduced Ido1 expression in mice colon organoids (Extended Data Fig. 2h). To further validate the role of Ido1 in the expansion of L. reuteri, we gave Apc Min mice epacadostat (an IDO1 inhibitor) for 5 d and assessed faecal L. reuteri abundance. Interestingly, the abundance of L. reuteri was significantly increased in mice receiving Aspirin has been proposed as a promising chemopreventive agent [3][4][5] . Nevertheless, the side effects (especially gastrointestinal bleeding) have triggered concerns 6 . Epidemiological and preclinical studies have revealed an association between statin use and reduced incidence of CRC [7][8][9][10][11][12] . However, the underlying mechanism needs to be better understood.
Studies have demonstrated that microbiota can mediate the clinical benefits of medications 13 . A recent study using metagenomics analysis identified statin therapy as correlated with an improved microbiota profile 14 . These findings suggest that microbiota might partly mediate the protective effect of statins in CRC.
The gut microbiota produces a diverse repertoire of bioactive molecules, including short-chain fatty acids, microbially transformed bile acids and tryptophan catabolites 15 . Dietary tryptophan can follow host routes (the kynurenine and serotonin pathways) or microbial routes 15 . In the microbial routes, tryptophan can be transformed into indole derivatives by members of the Lactobacillus species 16,17 . Many of these indole derivatives exert important regulatory functions on the host 18,19 .
In this work, we validated the chemopreventive role of statins in CRC by showing that statin users carry a reduced risk of CRA recurrence, and that gavage with statins reduced tumour burden in Apc Min mice. We found that the chemoprevention was attributed to an increased abundance of Lactobacillus reuteri and its tryptophan catabolite indole-3-lactic acid (ILA). ILA inhibited the differentiation of TH17 cells in the colon to suppress tumorigenesis. Our results identified a novel mechanism underlying the chemopreventive effects of statins and provided new targets for CRC prophylaxis.

Statin-modulated microbiota mediates chemopreventive effects
To verify the protective effects of statins in patients, a retrospective study was conducted to analyse the CRA recurrence rate within 3 years after adenoma removal using a cohort (Cohort 1) in our hospital. This cohort comprised 337,035 people who underwent colonoscopies between 2004 and 2021. After the screening, 5,703 individuals were determined to be eligible for analysis ( Supplementary Fig. 1a). Of these, the age distribution was significantly different between statin users and non-users (Supplementary Table 1), and this bias was balanced by propensity score (PS) matching. Statin users had a lower risk of CRA recurrence than non-users in our 3,646 matched participants (41.8% versus 52.4%, P = 0.0104) (Fig. 1a). Our results indicated statin use to be beneficial for reducing CRA recurrence and, thus, its role in lowering the CRC risk.
We then demonstrated the anti-cancer effects of statins in Apc Min mice (Extended Data Fig. 1a). Atorvastatin treatment caused no significant changes in body weight during the experiment (Extended Data Fig. 1b). Compared with the vehicle control, atorvastatin administration resulted in a significantly reduced tumour burden (Fig. 1b,c and Extended Data Fig. 1c) and improved histopathological status (Fig. 1d). Gavage with lovastatin recapitulated these phenotypes (Extended Data Fig. 1d-h), indicating that the anti-cancer effects might be common to statins. Thus, we chose atorvastatin as the representative in subsequent experiments because it was the most commonly prescribed statin in the United States in 2019 (ref. 20). Next, we explored whether the anti-cancer effects of statins were dependent on their lipid-lowering effects. For this purpose, fenofibrate was administered to Apc Min mice (Extended Data Fig. 1i). Fenofibrate failed to confer anti-cancer activity (Extended Data Fig. 1j-m). Thus, statins may prevent tumour formation in a manner that is independent of lipid-lowering effects.
To evaluate the potential involvement of gut microbiota, we gave an antibiotic cocktail to Apc Min mice and administered atorvastatin or vehicle control (Extended Data Fig. 1n). The discrepancy in tumour development was eliminated by antibiotic treatment (Fig. 1e-g and Extended Data Fig. 1o). However, we noticed that sustained anti biotic Nature Microbiology | Volume 8 | May 2023 | 919-933 921 Article https://doi.org/10.1038/s41564-023-01363-5 epacadostat (Fig. 2i). These data suggest that atorvastatin expands the commensal L. reuteri by increasing microbial tryptophan availability in the gut lumen after inhibiting epithelial Ido1 expression.

Metabolically active L. reuteri suppresses CRC development
To investigate the biological function of L. reuteri in CRC development, we employed a cohort of faeces samples from healthy controls and CRC patients (Cohort 2). Faecal L. reuteri abundance was significantly reduced in CRC patients (Fig. 3a). This reduction was also observed in CRC tissues compared with adjacent normal tissues in another cohort (Cohort 3) (Fig. 3b). These data imply that L. reuteri is clinically associated with CRC development.
To further explore the anti-cancer component of L. reuteri, we administered phosphate-buffered saline (PBS), live L. reuteri or heat-killed L. reuteri to microbiota-depleted Apc Min mice (Extended Data Fig. 3a). The tumour-suppressive effect was only observed with  Only taxa with LDA score greater than 3.5 are shown. e, Real-time PCR was performed to detect the abundance of L. reuteri in Apc Min mice receiving vehicle control or atorvastatin (n = 8 or 9 mice, respectively). f, Real-time PCR was performed to detect the abundance of L. reuteri in healthy volunteers before and after atorvastatin intervention (n = 20). g, Venn diagrams showing the genes upregulated vs downregulated by atorvastatin treatment in our study (66 vs 219), in GSE81375 (2,005 vs 2,208) and in both (15 vs 41). h, Real-time PCR was performed to detect the mRNA expression of Ido1 in colonic epithelium from Apc Min mice receiving vehicle control or atorvastatin (n = 5 mice per group). i, Real-time PCR was performed to detect the abundance of L. reuteri in Apc Min mice receiving epacadostat (an IDO1 inhibitor) (n = 5 mice per group). OTU, operational taxonomic unit. Data with error bars represent mean ± s.d. Data were analysed using two-tailed unpaired Student's t-test (c, h), two-tailed Welch's t-test (b, e, i) and two-tailed paired Student's t-test (f).
Article https://doi.org/10.1038/s41564-023-01363-5 chromatography-mass spectrometry (LC-MS) was carried out on faeces to analyse differences in tryptophan metabolism between Apc Min mice receiving atorvastatin or vehicle control. Orthogonal partial least squares discriminant analysis (OPLS-DA) showed the distinct metabolite profiles of two groups (Fig. 3g). The ILA concentration exhibited the most striking increase in the atorvastatin group (Fig. 3h,i and Supplementary Table 4). Combined with 16S rRNA-sequencing data, we found that Lactobacillus abundance was positively correlated with the concentration of ILA (Extended Data Fig. 3f). Consistently, faecal ILA concentration was also increased in healthy volunteers after atorvastatin intervention (Fig. 3j). Together, our data show that L. reuteri protects Apc Min mice from tumorigenesis through a metabolite(s) and that ILA is enriched in faecal samples after atorvastatin intervention.

ILA is essential to the anti-cancer effects of L. reuteri
We postulated that ILA might be one of the key metabolites responsible for the anti-cancer effects of L. reuteri. To test this hypothesis, ILA and vehicle control were administered to Apc Min mice (Extended Data Fig. 4a). ILA gavage brought faecal ILA concentration to levels similar to those of atorvastatin gavage (Extended Data Fig. 4b and Fig. 3i). Significantly, ILA treatment led to reduced tumour burden (Fig. 4a,b and Extended Data Fig. 4c) and alleviated histological severity (Fig. 4c).  ILA production by L. reuteri depends on the enzyme aromatic amino acid aminotransferase (ArAT) (Fig. 4d). We next addressed whether the deprivation of ArAT activity would abrogate the tumour-suppressive effect of L. reuteri by generating a mutant strain lacking a functional ArAT (L. reuteri ΔArAT) (Extended Data Fig. 4d-f). To functionally validate the compromised ILA production by L. reuteri ΔArAT, resting cell studies were performed and supernatants were analysed by LC-MS. OPLS-DA showed the distinct metabolite profiles of wild-type (WT) and mutant L. reuteri (Fig. 4e). Deprivation of ArAT significantly reduced the ILA concentration in the supernatant (Fig. 4f). L. reuteri ΔArAT was then gavaged to microbiota-depleted Apc Min mice to evaluate its influence on colorectal tumorigenesis (Extended Data Fig. 4g). Indeed, the anti-cancer effect of L. reuteri was largely compromised without ArAT activity (Fig. 4g,h and Extended Data Fig. 4h). Collectively, these results show that ArAT and its product, ILA, are indispensable for the tumour-suppressive effect of L. reuteri.

Statins induce a microenvironment with low TH17 responses
To gain molecular insights into the anti-cancer effect of atorva statin, we performed functional annotation and enrichment analysis on our epithelial RNA-sequencing results. We noticed that several of the top 10 KEGG pathways were related to immune regulation (Extended Data Fig. 5a). Considering that L. reuteri and ILA had been widely implicated in regulating host immune responses 17,18,28 , we then conducted RNA-sequencing on lamina propria cells. Interestingly, both compartments demonstrated transcriptional changes in TH17 cells and their cytokine signalling pathway after atorvastatin treatment (Extended Data Fig. 5a-d). TH17 cell-derived IL-17A has been shown to promote spontaneous intestinal tumorigenesis in Apc Min mice [29][30][31] . To investigate the clinical importance of the IL-17 signalling pathway in CRC initiation, we compared IL-17A expression in malignant tissues and adjacent normal tissues from CRC patients. We found significantly higher IL-17A expression in cancerous tissues (Fig. 5a,b). Statistical analyses from The Cancer Genome Atlas (TCGA) colon and rectal cancer dataset confirmed our observation (Fig. 5c). Next, we investigated the relationship between atorvastatinshaped microbiota and TH17-associated cytokines. Interestingly, bacteria enriched after atorvastatin gavage were negatively correlated with TH17-associated cytokines (Fig. 5d). To validate this in humans, the L. reuteri abundance and IL-17A expression were evaluated in  samples from Cohort 3, and a negative correlation was found (Extended Data Fig. 5e). We then investigated the dynamics of TH17 cells and the IL-17 signalling pathway after atorvastatin treatment using real-time PCR and flow cytometry. Real-time PCR analysis on colonic lamina propria cells showed significant downregulation of Il-17a expression after atorvastatin treatment (Fig. 5e), whereas RAR-related orphan receptor C (Rorc) expression was not altered (Extended Data Fig. 5f). Flow cytometry revealed a significant reduction in TH17 cells after atorvastatin treatment ( Fig. 5f and Extended Data Fig. 5g). In contrast, no significant difference in TH1 or Treg cells was observed (Extended Data Fig. 5h-k). Similarly, gavage with live L. reuteri significantly downregulated Il-17a expression (Fig. 5g). Flow cytometry showed that live L. reuteri led to a significantly lower frequency of IL-17A-producing CD4 + lymphocytes, which was not observed with heat-killed L. reuteri ( Fig. 5h and Extended Data Fig. 5l). Finally, we determined whether ILA was sufficient for TH17 inhibition. ILA administration resulted in significantly lower expression of Il-17a and reduced frequency of CD4 + IL-17A + T cells (Fig. 5i,j and Extended Data Fig. 5m). Together, these findings illustrate that, by modulating microbiota, atorvastatin induces a microenvironment with low TH17 responses, thereby suppressing colorectal tumorigenesis.

ILA suppresses TH17 differentiation by targeting the RORγt
To elaborate on how ILA inhibits TH17 responses, naïve CD4 + T cells were purified from mouse spleens and polarized in vitro (Extended Data Fig. 6a). Compared with vehicle control, ILA substantially reduced IL-17A-secreting cells in a dose-dependent manner (Fig. 6a and Extended Data Fig. 6b), in line with a previous report 28 . Given that RORγt has been recognized as the 'master' transcription factor of TH17 cells, we questioned whether RORγt could be a target of ILA. We first compared Article https://doi.org/10.1038/s41564-023-01363-5 the inhibitory efficacy of ILA and GSK805, a RORγt antagonist, on the differentiation of TH17 cells 32 . Although ILA was not as efficient as GSK805, the magnitude of inhibition of TH17 cells under the combination of two reagents was comparable with that under GSK805 treatment (Fig. 6b), indicating that RORγt could be a potential target downstream of ILA. Thus, we undertook RNA-sequencing on TH17 cells polarized in the presence of ILA, GSK805 and a combination of two reagents to compare global changes in gene expression. Expression of a set of TH17 genes, including Il-17a, Il-17f, Il-23r and Ccr6, was commonly downregulated by ILA and GSK805 (Fig. 6c). Our data demonstrate that ILA partly recapitulates the transcriptional effect of RORγt inhibition. Next, we questioned how ILA influenced the transcriptional activity of RORγt. We first measured Rorc expression between the ILA and the control group; however, no significant difference was found (Extended Data Fig. 6c). To test whether ILA impacts the occupancy of RORγt at its target genomic elements, chromatin immunoprecipitation-qPCR (ChIP-qPCR) was performed to evaluate the binding of RORγt to sites in two TH17 relevant loci, Il-17a and Il-23r. ILA decreased the binding of RORγt to these targets (Fig. 6d). Additionally, luciferase reporter assays using a fusion protein of RORγt ligand-binding domain (LBD) and GAL4 DNA-binding domain revealed decreased activity of the RORγt reporter upon ILA treatment (Fig. 6e). We then determined whether ILA physically interacted with the RORγt in vitro by microscale thermophoresis (MST) (Extended Data Fig. 6d) 33 . SR2211, a potent synthetic RORγt-selective modulator 34 , exhibited a robust physical interaction with recombinant RORγt LBD at a dissociation constant (K d ) of ~3.58 μM (Fig. 6f). Interestingly, we observed direct binding of ILA to RORγt LBD, albeit weaker than that of SR2211, with a calculated K d of ~14.37 μM (Fig. 6f). Therefore, ILA probably inhibits the transcriptional activity of RORγt by physically interacting with it and reducing its occupancy at targets such as Il-17a and Il-23r.
The aryl hydrocarbon receptor (AHR) is another transcription factor associated with TH17 responses that can be activated by indole derivatives 27,35 . To address whether AHR plays a part in inhibiting TH17 cells by ILA, we analysed the inhibitory effect of ILA in the absence of the AHR signalling pathway. CH-223191, an AHR antagonist, did not abrogate the inhibitory effect of ILA, indicating that AHR was not required for the inhibition of TH17 cells by ILA (Extended Data Fig. 6e). Because activation of the AHR in the epithelium has benefits, including improved intestinal barrier integrity, we examined whether the AHR contributed to the global anti-cancer effect of ILA in our model 19,36 . Real-time PCR showed AHR activation in the colonic epithelium of mice given atorvastatin or ILA (Extended Data Fig. 6f,g). Immunofluorescence staining of Zonula occludens-1 (ZO-1), a tight junction protein, showed improved gut barrier structure after treatment with atorvastatin or ILA (Extended Data Fig. 6h,i). Our in vivo studies showed that while gavage with ILA continued to protect mice from tumorigenesis, the blockade of the AHR pathway by CH-223191 weakened the protective effect (Extended Data Fig. 6j-m). Collectively, our results illustrate that ILA inhibits TH17 responses by targeting RORγt and improves barrier integrity by activating the AHR in the epithelium.

Discussion
Chemoprevention refers to the use of a synthetic or natural substance to suppress cancer initiation or to suppress malignant transformation of premalignant lesions 37,38 . CRC is one of the most preventable cancers due to the slow progression of the adenoma-carcinoma sequence 39 . The chemopreventive effect of statins in CRC is a focus of clinical research 8,9,[40][41][42] . Here we validate the chemopreventive effects of statins in patients and Apc Min mice, and demonstrated that a statin-shaped microenvironment plays a vital role.
The fact that microbiota mediates the efficacy of medications has been demonstrated previously 13,21,43 . Our study further shows that microbiota mediates the chemoprevention of atorvastatin. Through in vivo and in vitro studies, we demonstrate that the reduced host IDO1 activity and the consequently increased microbial tryptophan availability in the gut lumen contribute to the expansion of L. reuteri after atorvastatin treatment. Such host-microbiota interactions have also been observed in other studies where researchers found that, under a high-fat diet, the impaired ability of the colonic epithelium to utilize oxygen and nitrate increased the choline catabolism of Escherichia coli in the gut 44 .
L. reuteri is a well-studied natural colonizer of the human gut 45 . Several randomized clinical trials have examined the potential application of this commensal, and a well-tolerated safety profile has been documented [46][47][48] . The beneficial effects of L. reuteri can be attributed to its antimicrobial activity and immunoregulatory functions. A previous study revealed that reuterin, one of the antimicrobial molecules synthesized by L. reuteri, reduced the growth of CRC cells by inducing protein oxidation and inhibiting ribosomal biogenesis 49 .
The current study shows that ILA is another key molecule produced by L. reuteri that suppresses CRC initiation. Interestingly, a previous study showed that ILA produced by Lactobacillus gallinarum promoted the apoptosis of CRC cell lines 50 . Distinct from the protective effects in CRC initiation, another study revealed indoles signalled through AHR in tumour-associated macrophages to decrease inflammatory T cell infiltration and suppress the efficacy of immunotherapy in pancreatic cancer 51 . The observed diversification hints that the role of indole derivatives may be context dependent. Alleviated inflammation can be beneficial during cancer initiation but leads to worse outcomes in immunotherapy. Different anatomical locations can also provide discrepancy because some colonically localized benefits may not affect distal organs.
The contribution of IL-17A to intestinal tumour growth has been well documented and IL-17A ablation significantly decreases the tumour burden in Apc Min mice [29][30][31]52,53 . Previous studies showed that oncomicrobes activate the IL-17 signalling pathway to promote CRC development [54][55][56] . Here we show that L. reuteri suppresses CRC tumorigenesis by inhibiting the IL-17 signalling pathway. The AHR functions downstream of indole derivatives to assist in cellular adaptation to environmental challenges 19 . Our data demonstrate that activation of AHR in epithelium improves barrier integrity in Apc Min mice, which also partly contributes to the global anti-cancer effect of ILA. Together, we show that atorvastatin 'orchestrates' the L. reuteri-ILA-IL-17 signalling/AHR network to suppress CRC initiation (Extended Data Fig. 6n).
Our study has the following limitations. First, the assumption that the increased L. reuteri abundance after atorvastatin treatment is due to decreased Ido1 expression remains to be verified. More work is required to illustrate how atorvastatin reduces Ido1 expression and whether other genes are involved in L. reuteri expansion. Second, the improved barrier integrity is insufficient to fully elucidate the role of AHR in our model. AHR affects CRC development in multiple ways 57,58 , which includes protecting stem cells from genotoxic stress 59 , balancing controlled regeneration and malignant transformation 60 , promoting degradation of β-catenin 61 and repressing tumorigenic inflammation 62,63 . Further research is warranted to explore the potential involvement of pathways beyond barrier integrity.

Ethical approval of clinical studies
The Renji Hospital Ethics Committee (Shanghai Jiao Tong University School of Medicine, Shanghai, China) approved the protocol for the clinical studies. Written informed consent was obtained from all participants. The research was undertaken in compliance with the provisions of the Helsinki Declaration of 1975 and its later amendments.

Cohort study and clinical specimens
Cohorts of CRA patients, CRC patients and healthy controls were patients from Renji Hospital (Shanghai Jiao Tong University School Article https://doi.org/10.1038/s41564-023-01363-5 of Medicine, Shanghai, China) between 2004 and 2021. CRA or CRC was confirmed by pathology.
Cohort 1 (colonoscopy cohort) comprised 337,035 patients who underwent colonoscopies at Renji Hospital between 2004 and 2021. We undertook a retrospective study on Cohort 1 to explore whether statins had a protective role against CRA recurrence. CRA recurrence was defined as pathologically confirmed adenomas diagnosed between 6 and 36 months after index colonoscopy 8 . Statin user was defined as a patient with statin administration at least 5 times a week for more than 3 months per year during colonoscopy intervals 64,65 . The study patient selection diagram is displayed in Supplementary Fig. 1a. After the screening, 5,703 individuals were found eligible for analysis, including 3,418 males and 2,285 females. The age range is 19-89 yr. The prevalence of CRA recurrence was calculated for statin and non-statin users.
Cohort 2 comprised 88 healthy controls (48 males and 40 females, age range of 39-75 yr) and 85 CRC patients (51 males and 34 females, age range of 41-76 yr) enrolled between 2018 and 2019. Cohort 3 included 60 CRC patients (38 males and 22 females, age range of 42-74 yr) enrolled between 2019 and 2020. There were 173 frozen stool samples in Cohort 2, 60 paired frozen biopsies and 60 paired formalin-fixed paraffin-embedded (FFPE) tissues in Cohort 3. We used stool samples from Cohort 2 to explore whether L. reuteri abundance was clinically relevant to CRC. We used frozen biopsies from Cohort 3 to validate the conclusion we made for Cohort 2 and to determine the correlation between L. reuteri abundance and IL-17A expression. We used FFPE tissues in Cohort 3 to visualize IL-17A expression in normal tissues and cancer tissues.

Exploratory trial
We conducted an open-label, non-randomized exploratory trial to investigate whether and how statins shaped human gut microbiota. The trial was preregistered on the Chinese Clinical Trial Registry (http:// www.chictr.org.cn: ChiCTR2200059241). The study was conducted at Renji Hospital, Shanghai Jiao Tong University School of Medicine, under the ethical standards of the institutional review board. Participant compensation was metered on average wages and time spent on tasks.
The inclusion criteria were: (1) healthy 23-33-yr-old male or female and (2) normal blood lipids and normal liver function. The exclusion criteria were: (1) history of colorectal disease or familial history of colorectal-related disease; (2) history of liver disease; (3) regular use of aspirin, non-aspirin non-steroidal anti-inflammatory drugs, metformin, or statins; (4) use of antibiotics or probiotics 4 weeks before the study; (5) adverse reaction to statins in the past; (6) abuse of alcohol or drugs; (7) pregnancy or breastfeeding; and (8) participants that the researcher believes are not suitable for inclusion. The flow diagram for the exploratory trial is displayed in Supplementary Fig. 1b.
Participants were told not to change their dietary habits and not to take probiotics, antibiotics or any other medications during the study. Twenty healthy volunteers (10 males, 10 females) were recruited, and each was given one tablet of Lipitor (atorvastatin calcium, 20 mg per tablet) per day for 5 d. The detailed characteristics of participants (for example, age, body weight, lipid analysis and liver function) are displayed in Supplementary Table 2. Samples of serum and stools were collected at baseline and on day 5 of the study. Serum samples were used for lipid analysis and the liver function test. Stool samples were snap-frozen in dry ice and then stored at −80 °C for further bacteria and LC-MS analysis. No adverse event was observed during the study. The full trial protocol is provided in Supplementary Information.

Ethical approval of animal studies
All animal procedures were conducted in compliance with the guidelines of the Shanghai Model Organisms Center, Inc. Institutional Animal Care and Use Committee (SMOC IACUC). Protocols for animal studies were approved by the SMOC IACUC (No. 2018-0033).

Mice
Male Apc Min mice ( Jackson Laboratory, 002020) were housed (no more than five per cage) in a specific pathogen-free animal facility at 20-26 °C and 40%-70% humidity under a 12 h light/12 h dark photoperiod, with free access to food and water. Mice were fed chow diet (Xie Tong Biotechnology, SFS9112). Apc Min mice were euthanized when they started to present with haematochezia (the indication of tumour development), usually at 15 weeks old. The group sizes for macroscopic tumour assessment or mechanistic exploration were 10-13 or 4-6 mice per condition, based on previous studies 54, 66 .
For microbiota depletion in some experiments, a cocktail containing vancomycin (0.25 g l −1 , Sigma-Aldrich V2002), neomycin (0.5 g l −1 , Sigma-Aldrich N6386), metronidazole (0.5 g l −1 , Sigma-Aldrich M1547) and ampicillin (0.5 g l −1 , Sigma-Aldrich A9518) was provided in autoclaved drinking water starting at 2 weeks before oral gavage. Adult male Apc Min mice (5-6 weeks) were administered atorvastatin (20 mg kg −1 , LKTlabs A7658), lovastatin (20 mg kg −1 , Sigma-Aldrich PHR1285) or fenofibrate (100 mg kg −1 , Sigma-Aldrich F2050) by oral gavage every other day. Faecal microbiota transplantation was performed according to previous studies with some modifications, and details are described below [67][68][69] . Fresh stool samples from the same group were weighed, pooled and suspended in pre-reduced 30% glycerin/PBS at a final concentration of 100 mg ml −1 , followed by filtration through a 100 μm nylon filter to remove large particulate and fibrous matter in an anaerobic chamber. Faecal slurries were aliquoted and stored at −80 °C until use. The faecal slurries were administered 3 times per week for 4 weeks at a dosage of 200 μl for each microbiota-depleted Apc Min mouse. For donor and recipient mice, mice with the same intervention were co-housed in groups of 3 or 4 per cage, and 13 mice were used for each intervention. For oral gavage with L. reuteri, microbiota-depleted Apc Min mice were administered 10 8 colony-forming unit live, heat-killed, or mutant L. reuteri resuspended in 100 μl PBS every other day. Heat-killed L. reuteri was prepared by incubation at 65 °C for 15 min. For oral gavage with culture supernatants, L. reuteri cultured in MRS broth was centrifuged for 15 min at 6,000 × g at 4 °C, and cell pellets were discarded. In some cases, proteinase K (Thermo Fisher EO0491) was added to culture supernatants at 100 μg ml −1 , culture was incubated at 50 °C for 1 h and then inactivated at 90 °C for 5 min. For oral gavage with ILA (Aladdin I157602), adult male Apc Min mice (5-6 weeks) were administered ILA (0.1 mg kg −1 ) every other day. For the blockade of the AHR pathway, CH-223191 (MCE HY-12684) (10 mg kg −1 , intraperitoneal injection) was administered. For the blockade of the host IDO1 activity, mice were administered epacadostat (MCE HY-15689) (50 mg kg −1 ) by oral gavage.
Tumour numbers in colons were observed and recorded using a stereomicroscope (Olympus). The size of each tumour was measured using a previously published formula 70 .

Organoids
Colon organoids were isolated from male C57BL/6J mice ( Jackson Laboratory 000664) at 10 weeks of age as previously described 71  protocol. Cells were cultured in DMEM (Gibco 11995065) supplemented with 10% fetal bovine serum (FBS; Gemini 900-108) and placed in an incubator at 37 °C in an atmosphere of 5% CO 2 .

Bacteria
L. reuteri (ATCC 23272) was purchased from ATCC. L. reuteri was cultured in MRS broth or agar as appropriate at 37 °C under anaerobic conditions (Don Whitley Scientific).
L. reuteri ΔArAT mutant was generated by an in-frame deletion strategy using the allelic replacement vector pJAM083 as described previously 17,72 . Briefly, A 1,407 bp PCR product was amplified from pCR2.3. Next, a 2,873 bp segment containing temperature-sensitive Gram-positive ori elements and an erythromycin resistance cassette was synthesized by Tsingke Biotechnology. These two segments were ligated to generate the pGCP213. The in-frame deletion allele was synthesized by Sangon Biotech. The pGCP213 and the in-frame deletion allele were ligated to generate pJAM083. The latter was used to electro-transform competent L. reuteri. Integrants were identified by culturing cells in erythromycin-containing MRS (10 μg ml −1 ; Sangon Biotech A600192) at 37 °C under anaerobic conditions. Then, integrants were moved to erythromycin-free MRS at 30 °C to promote double-crossover. Recombination was identified by plating the culture on both erythromycin-free and erythromycin-containing MRS agar. Mutants were further identified by PCR and Sanger sequencing. The primers 17 used to verify the mutant are listed in Supplementary Table 5.

Resting cell studies
Resting cell studies were performed as previously described with some modifications 73,74 . Wild-type and mutant strains of L. reuteri cultured in MRS medium (20 ml) to an optical density (OD) 600 of 1 were collected by centrifugation (15 min at 6,000 × g at 4 °C) and washed twice with pre-reduced PBS. The bacterial pellets were then resuspended in PBS and incubated at room temperature for 45 min to consume the residual substrates from the rich medium. After incubation, the cells were centrifuged, resuspended in 1 ml of reaction buffer and incubated at 37 °C. The reaction buffer was PBS containing a 2 mM tryptophan (Sigma-Aldrich T8941) and 10 mM α-ketoglutarate (Sigma-Aldrich K1128). After incubation for 2 h in an anaerobic chamber, the reaction mixtures were centrifuged, and supernatants were collected for subsequent LC-MS analyses.

Isolation of colonic epithelial cells
Colonic epithelial cells were isolated according to a previously described protocol with some modifications 75 . Mice were killed and colons were collected. The intestinal lumen was washed with ice-cold PBS. Next, the colon was opened longitudinally and cut into pieces of 1-2 cm length, followed by several washes with PBS. The colon pieces were then transferred into Hank's balanced salt solution (Gibco 14175095) supplemented with EDTA (2.5 mM; Beyotime ST066) and dithiothreitol (1 mM; Roche 10197777001), followed by incubation at 37 °C under slow rotation for 20 min. The suspensions were filtered, and colonic epithelial cells were collected by centrifugation (800 × g) and used for further analyses.

Isolation of colonic lamina propria cells and flow cytometry
Intestinal lamina propria cells were isolated as previously described with some modifications 76 . Briefly, colons were isolated and opened longitudinally. Faeces were washed, and fat tissues and Peyer's patches were removed with forceps. Colons were cut into 2-cm-long pieces and washed with PBS. Colon pieces were then transferred into a vessel containing HEPES (10 mM) supplemented with dithiothreitol (1 mM) and EDTA (30 mM), followed by incubation at 37 °C at 200 r.p.m. for 10 min. Suspensions were filtered, and the remaining tissues were transferred into a vessel containing HEPES (10 mM) supplemented with EDTA (30 mM), followed by incubation at 37 °C at 200 r.p.m. for 10 min.
Suspensions were filtered again, and the remaining tissues were cut into 0.5-cm-long pieces and digested with RPMI 1640 medium (Gibco 11875093) supplemented with 5% FBS, collagenase VIII (1 mg ml −1 ; Sigma-Aldrich C2139) and DNase I (0.5 mg ml −1 ; Roche 10104159001) in an incubator at 37 °C in an atmosphere of 5% CO 2 for 50 min. The digested solution was filtered, centrifuged and resuspended with Percoll (GE Healthcare GE17-0891-01). The tube was spun at 1,000 × g for 25 min at room temperature to enrich live mononuclear cells. The lamina propria lymphocytes were then collected from the interphase of 40:80 Percoll gradient and washed with PBS. The lamina propria cells were stimulated with Leukocyte Activation Cocktail with BD GolgiPlug (BD Pharmingen 550583) for 4 h, stained with fluorescent-conjugated antibodies and analysed by flow cytometry. For intracellular staining, cells were fixed and permeabilized with BD Cytofix/Cytoperm (BD 554714) or Transcription Factor buffer set (BD 562574) according to manufacturer instructions. Data were collected by FACSDiva. The FACS gating strategy to define TH17 cells is displayed in Supplementary Fig. 2a.

TH17 polarization of naïve CD4 + T cells from mice
Naïve CD4 + T cells were isolated from the spleens of male C57BL/6J mice ( Jackson Laboratory 000664) at 8 weeks of age. Briefly, spleens were ground in PBS and filtered through a 70 μm cell strainer. Red blood cells were lysed with red blood cell lysing buffer (Sigma-Aldrich R7757). Naïve CD4 + T cells were isolated by magnetic sorting using the Naive CD4 + T Cell Isolation kit (Miltenyi Biotec 130-104-453) following manufacturer guidelines. Cells were plated at 1 × 10 6 cells per cm 2 . For polarization of TH17 cells, the T Cell Activation/Expansion kit (Miltenyi Biotec 130-093-627) and CytoBox TH17 (Miltenyi Biotec 130-107-758) were used following manufacturer guidelines. Cells were stimulated with Leukocyte Activation Cocktail with BD GolgiPlug for 4 h on day 4, stained with fluorescent-conjugated antibodies and analysed by flow cytometry. The FACS gating strategy to define TH17 cells is displayed in Supplementary Fig. 2b.

DNA extraction
Tissue and faecal DNA were extracted using the AllPrep DNA/RNA mini kit (QIAGEN 80204) and QIAamp PowerFecal Pro DNA kit (QIAGEN 51804), respectively, following manufacturer guidelines.

RNA extraction and real-time PCR
Total RNA from mouse tissues, cell lines and organoids was extracted using RNAiso Plus (Takara 9108). Total RNA from human tissues was extracted using the AllPrep DNA/RNA mini kit following manufacturer guidelines. The PrimeScript RT reagent kit (Takara RR037A) was used to reverse-transcribe 1 μg of total RNA to detect the relative expression of messenger RNAs (mRNAs). Quantitative real-time PCR was performed in triplicate on the StepOnePlus Real-Time PCR system (Applied Biosystems). The cycle threshold (Ct) values gained from samples were compared using the 2 -ΔCt method. For the detection of host genes, β-actin served as the internal reference gene. For the detection of bacterial abundance, 16S rRNA served as the reference gene. For absolute quantification, genomic DNA from L. reuteri was used to generate the standard curve. All primers are listed in Supplementary Table 5.

MST assay
The binding affinity of SR2211 (Sigma-Aldrich 557353) and ILA with RORγt LBD was tested by MST as previously described 78 . Purified RORγt LBD was labelled with the protein labeling kit RED-NHS 2nd Generation (NanoTemper Technologies MO-L011). Serially diluted compounds were mixed with labelled RORγt LBD and loaded into Monolith Premium Capillaries (NanoTemper Technologies MO-K025). Binding affinity was measured by monitoring thermophoresis with automatic LED power and 40% MST power on a Monolith NT.115 instrument (NanoTemper Technologies) with the following time settings: 5 s Fluo, before; 20 s MST on; and 5 s Fluo, after. K d values were fitted using MO.Affinity Analysis software (NanoTemper Technologies).

ChIP-qPCR
ChIP was performed using the ChIP-IT PBMC kit following manufacturer guidelines (Active Motif 53042). Briefly, CD4 + T cells cultured under the TH17 cell-polarizing condition were crosslinked with 1% formaldehyde (Sigma-Aldrich F8775), washed and lysed. Chromatin was then sonicated into 200-1,000 bp lengths and immunoprecipitated with RORγt antibody (eBioscience 14-6988-82) or IgG (BioLegend 400501) overnight at 4 °C. The antibody-chromatin complexes were pulled down using Protein G agarose beads, and the immunoprecipitants were then washed and eluted. The protein-DNA crosslinks were reversed, and DNA was purified. The enrichment of immunoprecipitated materials was analysed by real-time PCR. The primers 77 used are listed in Supplementary Table 5.

16S rRNA-sequencing
The 16S rRNA-sequencing was carried out as previously described 79 . Data were analysed on the online platform Majorbio Cloud (www. majorbio.com).

RNA-sequencing
Cells were collected, and total RNA was extracted using RNAiso Plus. Enrichment, fragmentation, reverse transcription, library construction and sequencing of mRNA were achieved using Novaseq 6000 (Illumina). Clean reads were aligned to the mouse genome (mm10) using TopHat v2.0.117 (http://tophat.cbcb.umd.edu) with default options and a TopHat transcript index built from Ensembl_GRCm38 (ref. 80). The workflow used to analyse the data has been previously described 81 . Analyses of enrichment of signalling pathways were done using the Kyoto Encyclopedia of Genes and Genomes (KEGG; https:// www.genome.jp/kegg/) and the Database for Annotation, Visualization and Integrated Discovery (DAVID; https://david.ncifcrf.gov) 82,83 .

Quantification of tryptophan and its metabolites using LC-MS
Tryptophan and its metabolites were detected on the basis of the SCIEX QTRAP 6500 LC-MS/MS system. Briefly, samples were thawed on ice, extracted with methanol, and internal standard was added. The Waters ACQUITY UPLC HSS T3 Column (1.8 μm, 2.1 mm × 100 mm) was used for liquid chromatography. The mobile phase for liquid chromatography was water with 0.1% (v/v) formic acid (A) and acetonitrile with 0.1% (v/v) formic acid (B). Tryptophan and its metabolites were analysed using multiple reaction monitoring. MultiQuant software was used to process and quantify data. The ropls R package was used to implement the OPLS-DA analysis.

Statistics and reproducibility
No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications 54,66 . Data distribution was assumed to be normal, but this was not formally tested. The exploratory trial was not randomized because of the absence of different treatment groups. For in vivo experiments, animals were allocated randomly into each experimental group by using a randomization chart. For in vitro experiments, randomization was not used because of the homogeneous feature of cells. Data collection and analysis were not performed blind to the conditions of the experiments. Data were excluded in the case of technical failures. For in vivo experiments, animals were excluded if they died.
All replicates in the study are biological and when experiments were repeated in the same fashion, they yielded comparable results. Data are presented as means ± s.d. Statistical analyses were carried out using R (www.r-project.org) and GraphPad Prism. Unpaired Student's t-tests were performed to compare results when the data had the same variances. Welch's t-tests were performed if the data had different variances. Paired t-tests were performed to analyse differences between pairs. One-way analysis of variance (ANOVA) was performed to compare means among three groups and multiple comparisons were corrected using Holm-Sidak or Tukey's test. Chi-square (and Fisher's exact) tests were performed if the data are presented as contingency tables. Correlation analysis was performed to quantify the degree to which two variables were related. Statistical tests were two-tailed. Exact P values are shown for significant comparisons but not for non-significant comparisons. We treated P < 0.05 as significant.

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

Data availability
Raw reads of 16S rRNA-sequencing were uploaded to the National Center for Biotechnology Information (NCBI) Sequence Read Archive database (accession number: PRJNA831634; https://www.ncbi.nlm. nih.gov/bioproject/PRJNA831634). RNA-sequencing data have been deposited in the Gene Expression Omnibus (GEO; https://www.ncbi. nlm.nih.gov/geo/) of NCBI and are accessible through the GEO Series accession number GSE201453. Datasets analysed during the current study are the TCGA colon and rectal cancer datasets, which are available in the TCGA database, and GSE81375, which is available in the GEO database. Source data are provided with this paper.