Dietary sulfur amino acids enhance anti-tumor immunity in colon cancer via an NKT cell-XCL1-cDC1 circuit

SUMMARY

The efficacy of immune checkpoint inhibitors (ICIs) in colon cancer (CC) is limited by the tolerogenic immune landscape of both the tumor and the intestine. Here, we investigated the impact of microbial metabolic pathways associated with ICI responsiveness on anti-tumor immunity in CC. Sulfur amino acid (Saa) metabolic pathways were enriched in ICI responders and in individuals without CC. In murine models, high-Saa diets restricted CC progression, increased colonic mucus thickness, and promoted the expansion of mucus-dwelling Mucispirillum schaedleri (M. schaedleri). Dietary Saa supplementation increased the frequency of activated intratumoral cytotoxic CD8⁺ T cells. The protective effect of high-Saa diets required type 1 conventional dendritic cells (cDC1s) in tumor-draining lymph nodes, where M. schaedleri was sufficient to drive cDC1 expansion and activation. Mechanistically, M. schaedleri stimulated natural killer T (NKT) cell expansion and secretion of the chemokine XCL1, which was essential for cDC1 recruitment and anti-tumor immunity. Dietary modulation of this NKT-XCL1-cDC1 axis in combination with ICIs may open avenues for improved treatments of CC.

Graphical Abstract

In brief

Lobel, Fonseca-Pereira, et al. demonstrate that dietary sulfur amino acids (Saa) enhance anti-tumor immunity in colon cancer. A high-Saa diet expands the bacterium Mucispirillum schaedleri, which activates a natural killer T cell-dendritic cell axis to restrict tumor growth, illustrating how nutritional interventions can be leveraged for cancer immunotherapy.

INTRODUCTION

Colon cancer (CC) is a leading cause of cancer deaths, and its incidence is rising.¹ While immune checkpoint inhibitors (ICIs) have been a breakthrough for many cancers,² most CC does not respond to ICIs, due to proficient DNA mismatch repair (MMR)³ and the colon’s tolerogenic immune landscape.⁴ Beyond tumor cell-intrinsic factors,⁵ the gut microbiota also influences responsiveness to immunotherapy treatments and CC development,^6–9 and several specific bacterial species have been implicated.^10–27

Gut microbiota composition is influenced by lifestyle factors such as diet, and dietary interventions have been considered in personalized medicine approaches to cancer treatment.^28–34 Additionally, both the microbiota and dietary factors are associated with the etiopathogenesis of CC.^35,36 As microbiomes have broader effects on immunity beyond ICI treatment outcome, we hypothesized that examining microbial metabolic capabilities in ICI responders that can be targeted by dietary interventions may unveil opportunities to improve anti-tumor immunity in ICI-resistant tumors such as CC.

We performed a meta-analysis of publicly available human gut microbiome data and observed that gut microbial sulfur amino acid (Saa) metabolism was associated with ICI responsiveness (n = 17 studies)^10–26 and with the absence of CC (n = 12).^37–48 By employing two Saa diets that mimic either high (hi) Saa consumption (but poses no homocysteinemia risk) or low (lo) Saa consumption (but is not methionine restricted) in mice or humans, we found that a diet enriched in Saa blunted tumor growth and neoplastic progression in two CC mouse models: a heterotopic MMR-deficient CC model and a genetically engineered MMR-proficient and ICI-resistant CC model. A hi Saa diet induced colonic mucus thickening and a bloom of Mucispirillum schaedleri (M. schaedleri), a mucus-dwelling microbe present in the colon.^49–52 M. schaedleri stimulated X-C motif ligand 1 (XCL1) secretion by both mouse and human natural killer T (NKT) cells and promoted expansion of type 1 conventional dendritic cells (cDC1s) in mesenteric and tumor-draining lymph nodes, culminating in increased intratumoral, activated cytotoxic CD8⁺ T cells, a hallmark of effective anti-tumor immunity.

RESULTS

Saa metabolic pathways are enriched in ICI R and decreased in CC patients

Distinct gut microbiome signatures and microbial products are associated with ICI response.^10–27,53 However, specific microbe-ICI associations have not been consistent across studies.¹⁸ We investigated if microbial bioactivities shared across taxa associated with ICI responsiveness would reveal opportunities for improving CC anti-tumor immunity. We performed a meta-analysis focused on MetaCyc pathways⁵⁴ from the microbiomes of responders (R) vs. non-responders (NR) with melanoma, renal cell carcinoma, and non-small cell lung cancer from 17 ICI-microbiome studies^10–26 (R, n = 837; NR, n = 632). These studies were chosen because of their accessibility, data quality, and publicly available metadata. Among the top 10% of pathways enriched in ICI R vs. NR (ranked by effect size, Cohen’s D test), we found four pathways related to Saa metabolism (Figure 1A; Table S1). We also found that pathways related to inosine, reported to enhance ICI therapy responses in preclinical cancer models,²⁷ and histidine, reported to be higher in R and associated with increased CD8⁺ T cell activation and improved ICI response,^53,55 were enriched in the top 10% of pathways in ICI R vs. NR (Table S1). The ICI-microbiome cohorts we analyzed provided common co-variates such as age, sex, and cancer stage, which we controlled for during these analyses, although not all provided survival or dietary pattern data.

Figure 1. Dietary Saa reduce tumor progression in mouse CC models.

(A) MetaCyc pathways ranked by meta-analysis effect size from a linear mixed-effects model for R (n = 837) and NR gut microbiomes (n = 632) from 17 ICI studies.

(B) MetaCyc pathways summarized by correlating meta-analysis effect size from ICI cohort with CC gut microbiome cohort (12 CC studies) (control, n = 934; CC, n = 903). Dashed lines represent a false discovery rate (FDR) threshold of 0.25.

(C–E) Experimental scheme (C), tumor volume over time, representative tumor photographs (D), and tumor weights (E) in WT mice fed lo or hi Saa diets and flank-injected with MC38 cells.

(F) Representative colon photographs from Cdx2^Cre Apc^flox/+ (cAPC) mice fed lo or hi Saa diets with tumors highlighted by black circles (top) and data on dysplastic and neoplastic lesion grade and numbers (bottom). ACF, aberrant crypt foci; AdenoCA, adenocarcinoma. Each column represents data from a single mouse.

(G) Neoplastic lesions (Adenoma + AdenoCa) diameters measured in H&E-stained photomicrographs from cAPC mice fed lo or hi diets shown in (F).

(H) Colon tumor weights from cAPC mice fed lo or hi diets. Each symbol represents data from an individual mouse in (E), (G), and (H). Symbol and error bars represent mean ± SEM in (D). Data represent four independent experiments (D–H). p values for two-way ANOVA with Šídák’s multiple comparisons test (D), Mann-Whitney U test (E, G, and H) or chi-squared contingency test (G).

See also Figure S1.

Saa biosynthesis pathway enrichment across ICI R with different cancer types suggested a potential role in anti-tumor immunity and an opportunity for enhancing anti-tumor immunity in ICI-unresponsive tumors, like MMR-proficient CC. Unfortunately, there are no publicly available CC ICI-microbiome profiling studies to date. Thus, we examined metagenomic data from 12 CC cohorts (903 CC patients and 934 healthy controls)^37–48 and correlated the MetaCyc pathways that associated with CC patients or healthy controls with pathways that independently discriminated ICI R vs. NR (Table S1). We identified Saa metabolism pathways as correlated with both ICI responsiveness and a CC-free status in CC studies (Figure 1B). Our meta-analysis findings led us to consider that targeting Saa metabolic pathways may provide therapeutic benefits for CC.

Dietary Saa modulate neoplastic progression in CC mouse models

Dietary patterns regulate the microbiome by changing nutritional niche availability,^56,57 and microbiome functional modulation through diet is garnering interest for its immunomodulatory effects in diseases such as inflammatory bowel diseases, obesity, and cancer.^32,33,58,59 Given our findings in (Figures 1A and 1B), we hypothesized that modulating dietary Saa may influence CC anti-tumor immunity. We formulated two isocaloric mouse diets with lo vs. hi amounts of methionine and cysteine to represent edge cases of human Saa consumption (Table S2). Methionine restriction was previously shown to impair anti-tumor immunity.⁶⁰ Therefore, we formulated the Saa diets ensuring that the lo Saa diet (~0.1× the amount of cysteine and 0.5× the amount of methionine of standard mouse chow) was not methionine restricted and that the hi diet (~2× the amount of cysteine and 2.5× the amount of methionine of standard mouse chow) (Table S2) was well below the hyperhomocysteinemia threshold,^61,62 both representing physiologically relevant human Saa consumption achievable through dietary modification or supplementation.

We employed this dietary intervention in the MC38 cell-line heterotopic CC model, a MMR-deficient tumor model widely used in tumor immunology research.^63,64 Wild-type (WT) C57BL/6 mice raised in our vivarium were placed on a lo vs. hi Saa diet 2 weeks prior to MC38 cell flank injection and monitored for 12 days (Figure 1C). Mice fed the hi Saa diet had slower tumor growth with a ~50% reduction in tumor volume and 30% reduction in tumor weight (Figures 1D and 1E), suggesting that dietary Saa was a beneficial anti-CC intervention.

While we had promising results using this MMR-deficient CC model, 85% of CCs are MMR-proficient, often have mutations in the adenomatous polyposis coli (Apc) gene, and are resistant to ICIs.⁶⁵ Next, we investigated if the Saa diets impacted tumor progression in a more clinically relevant model and utilized conditional transgenics with colon-specific Apc heterozygous deletion, Cdx2^Cre × Apc^flox/+ (cAPC) mice.⁶⁶ In contrast to other Apc-based mice in our vivarium, cAPC mice develop tumors in the colon and thus better mimic human CC.⁶⁶ Thus, our studies were not survival studies, in contrast with other Apc models,⁶⁷ and these cAPC mice lived with tumors during our experiments for 20 weeks postnatally.

Histological analysis of aberrant crypt foci (ACF), adenomas, and adenocarcinomas (AdenoCA) (examples in Figure S1A) showed that cAPC mice fed the hi Saa diet had a >4-fold decrease in advanced neoplastic lesion occurrence (i.e., AdenoCA) (Figures 1F, S1B, and S1C). Colonic lesion size (adenomas and AdenoCAs) and tumor weights were also reduced by half in hi Saa-fed mice (Figures 1G, 1H, and S1D). Notably, cAPC mice on the regular diet used in our vivarium have a similar AdenoCA incidence to cAPC mice fed the lo Saa diet (Figures S1E–S1G). These data support that dietary Saa supplementation impair tumor progression in an ICI-refractory CC model.

Dietary Saa regulates mucus layer thickness and the abundance of M. schaedleri

CC development in mice is influenced by microbiome composition, which can be modulated by dietary interventions.^56,57,68,69 We confirmed that colonic tumors did not develop in our cAPC mice under germ-free (GF) conditions over an experimental time course of ~5 months (Figure S2A). We then performed a 16S rRNA gene amplicon survey on cecal contents from cAPC mice fed the lo or hi Saa diets to investigate the microbiome effects of Saa diets. Surprisingly, we found that after our subtle Saa dietary modification, only one taxon, M. schaedleri, was significantly increased and enriched in cAPC mice fed the hi Saa diet (Figure 2A; Table S1). M. schaedleri is a Gram-negative anaerobic Deferribacteraceae member that inhabits the mouse colonic mucus and has been detected in up to 42% of human colonic biopsies.^49–52,70 Quantitative PCR analysis of cecal DNA from cAPC mice confirmed that M. schaedleri abundance increased in mice fed the hi vs. lo Saa diet (Figure 2B). The hi Saa diet also promoted an increase in M. schaedleri abundance in the specific pathogen-free (SPF) mice used with the MC38 CC model and in gnotobiotically housed altered Schaedler flora (ASF) mice, which harbor M. schaedleri as part of their microbiota (Figure 2C).⁷¹

Figure 2. Hi Saa diet-fed mice have higher M. schaedleri abundance and thicker mucus layers.

(A) Volcano plot showing enrichment of microbial taxa from 16S rRNA amplicon profiling from cAPC mouse cecal contents fed lo vs. hi Saa diet using Microbiome Multivariable Associations with Linear Models 2 (MaAsLin 2).

(B) Quantitative (q) PCR analysis: M. schaedleri 16S rRNA gene, cAPC mouse cecal content DNA.

(C) qPCR analysis: M. schaedleri 16S rRNA gene, ASF, and SPF flank-injected with MC38 cells cecal content DNA.

(D) Relative abundance of M. schaedleri-specific reads in 16S rRNA amplicon sequence data from tissue biopsies of healthy controls and patients with colonic adenoma or AdenoCA.

(E) Representative images of Muc2 immunofluorescence staining and bacterial FISH staining in colon sections from ASF mice fed Saa diets (pink: M. schaedleri probe; blue: universal bacterial probe; green: Muc2; gray: DAPI).

(F) Mucosal M. schaedleri abundance measurements from M. schaedleri probe intensity. M. schaedleri probe intensity was normalized by its area fraction of the mucus layer.

(G) Mucus layer thickness measurements from Muc2 intensity. Each symbol represents data from an individual mouse (average of 5 fields per mouse).

(H–J) Representative micrographs of Alcian blue staining of mucus, intestinal area fraction occupied by goblet cells, and mucus layer thickness measurements from WT GF (H), ASF (I), and SPF mice (J) colon. Red brackets represent mucus thickness. Each symbol represents data from an individual mouse (B, C, and F–J). Box and whiskers graphs for mucus layer thickness measurements in (H)–(J) show all points in the 90% CI. p values for Mann-Whitney U test.

See also Figure S2.

To assess the prevalence of M. schaedleri in humans, we searched for M. schaedleri-specific reads in human microbiome datasets but did not detect reliable M. schaedleri signals in human stool metagenomes. Aligned with its mucus-dwelling lifestyle and prior reports,^50,51 analysis of microbiome data from normal colonic tissue, adenoma, and AdenoCA (stage 1–4) biopsies^72–74 revealed that M. schaedleri was detectable and its abundance negatively correlated with tumorigenesis, showing higher abundance in normal colonic biopsies and absence in AdenoCA samples (Figure 2D). Unfortunately, such samples and profiling are not routinely done from ICI-responsive patients other than from those experiencing ICI-induced colitis.

We examined M. schaedleri’s localization in the mouse colon using bacterial-directed fluorescence in situ hybridization (FISH) and Muc2 indirect immunofluorescence microscopy. Confirming its mucus-dwelling location,^49–52 M. schaedleri bloomed in the colonic mucus in mice fed the hi Saa diet (Figures 2E and 2F). Analysis of Muc2 intensity in the mucus also showed increased mucus thickness in mice fed the hi Saa diet (Figure 2G), so we investigated if Saa diets alter mucus layer properties to promote M. schaedleri blooms. Mucus glycosylation, sulfation, and production can be altered by dietary interventions with significant effects on colonic inflammation and carcinogenesis.^75–79 The percentage of sulfated mucins in WT mice was unaffected by the Saa diet feeding (Figures S1B and S1C), so we examined if Saa diets altered mucin gene expression. There were no significant differences in Muc1, Muc2, Muc3, or Muc4 expression in the colonic epithelium of mice fed the lo and hi Saa diets (Figure S2D). We then tested if M. schaedleri binding to mucus was affected by Saa diets. By incubating fluorescein isothiocyanate (FITC)-labeled M. schaedleri with mucus from mice fed lo and hi Saa diet bound to fluorescently labeled wheat germ agglutinin (WGA), we observed that M. schaedleri had a similar capacity to bind mucus from both diets (Figure S1E).

Dietary amino acids such as cysteine promote mucin synthesis, while Saa dietary deficiency lowers goblet cell numbers.^75,80 To assess the direct effect of dietary Saa on the mucus, we fed the Saa diets to GF, ASF, and SPF mice and performed Alcian blue staining of colonic tissue. While we found no differences in goblet cell development in the proportion of goblet cells observed in the colon, we found that the hi Saa diet significantly increased the mucus layer thickness across all the conditions analyzed (Figures 2H–2J). Collectively, these data suggest that dietary Saa levels regulate the abundance of M. schaedleri by affecting colonic mucus and its habitability for this bacterium.

Hi Saa diet increased numbers of CD8⁺ T cells in cAPC tumors

To unravel the mechanistic links among dietary Saa, microbial activities, and tumorigenesis, we analyzed anti-tumor immune responses in cAPC mice fed the Saa diets. We began by investigating cytotoxic CD8⁺ T cells, critical players in anti-tumor immunity,^81–83 and found higher frequencies and numbers of intratumoral CD8⁺ T cells in tumors from hi Saa diet-fed cAPC mice (Figures 3A and S3A–S3C). We confirmed these flow cytometry results using immunofluorescence microscopy and found a 2-fold increase in CD8⁺ T cells in tumors from hi Saa diet-fed cAPC mice (Figures 3B and 3C). The CD8⁺ T cell increase was specific to colonic tumors, as CD8⁺ T cells in the mesenteric lymph nodes (mLNs) and colon remained unchanged (Figure S4). As effective anti-tumor CD8⁺ T cell responses require the production of cytotoxic effectors and controlled expression of immune co-inhibitory receptors, which in excess lead to T cell exhaustion,^81–83 we profiled the expression of several co-inhibitory receptors and found that proportions of PD-1⁺ and LAG-3⁺ CD8⁺ T cells were significantly decreased in tumors from hi Saa diet-fed mice (Figure 3D). We observed a tendency for increased interferon (IFN)-γ and a significant increase in granzyme B (GZMB) in intratumoral CD8⁺ T cells in hi Saa diet-fed mouse tumors (Figures 3E and 3F). The hi Saa diet increased CD8⁺ T cells well-armed for anti-tumor immunity, likely lowering colonic tumor incidence.

Figure 3. A hi Saa diet expands tumor cytotoxic CD8⁺ T cells.

(A) Frequency and numbers of CD8⁺ T cells in cAPC tumors from mice fed lo or hi Saa diets by flow cytometry.

(B) Representative immunofluorescence micrographs of colon tumors sections from cAPC mice fed lo or hi Saa diet. Right-side images are higher magnifications of regions indicated by white rectangles at left.

(C) Quantification of CD3⁺CD8⁺/CD3⁺ ratio in immunofluorescence micrographs. Each symbol represents data from an individual mouse (average of 4 fields per mouse).

(D) Expression of inhibitory receptors in CD8⁺ T cells from tumors from cAPC mice fed the Saa diets, flow cytometry.

(E) Frequency of IFN-γ⁺ CD8⁺ T cells from the tumors of cAPC mice.

(F) Frequency of GZMB⁺ CD8⁺ T cells from the tumors of cAPC mice. Each symbol represents data from an individual mouse. Bar graphs indicate mean ± SEM. Data represent three independent experiments. p values for Mann-Whitney U test.

See also Figures S3 and S4.

Anti-tumor effect of hi Saa diet requires cDC1 in the mLNs

Dendritic cells (DCs), particularly CD103⁺CD11b⁻ cDC1s, are important orchestrators of anti-tumor immunity as they cross-present tumor-derived antigens to elicit CD8⁺ T cell anti-tumor responses.^84–88 While we found that DC populations are unchanged in the tumor microenvironment and adjacent colon of cAPC mice fed the Saa diets (Figures S5A and S5B), cDC1s increased in frequency and number in the mLNs of hi Saa diet-fed cAPC mice (Figures 4A, 4B, gating strategy; Figure S3D). Tumor-draining lymph nodes are pivotal immunoregulatory sites where cDC1 cross-presentation events critical for anti-tumor CD8⁺ T cells occur.⁸⁵

Figure 4. cDC1s increase with hi Saa diet or M. schaedleri administration and are required for hi Saa diet anti-tumor effects.

(A) Representative flow cytometry analysis of CD11c⁺MHCII^high DCs from the mLNs of cAPC mice fed lo or hi Saa diet.

(B) Frequency and numbers of cDC1 (CD103⁺Cd11b⁻) and CD103^+/− cDC2 in mLNs of cAPC mice fed lo or hi Saa diet.

(C) Experimental scheme and data on dysplastic and neoplastic lesion grade and numbers from cAPC Batf3^−/− mice fed the lo or hi Saa diet. Each column represents data from a single mouse.

(D) Frequency and numbers of cDC1 (CD103⁺Cd11b⁻) and CD103^+/− cDC2 in the mLN of cAPC Batf3^−/− mice fed lo or hi Saa diet.

(E) Experimental scheme and data on dysplastic and neoplastic lesion grade and numbers from Zbtb46-DTR cAPC mice fed the hi Saa diet and injected i.p. with PBS or DT. Each column represents data from a single mouse.

(F) Frequency and numbers of cDC1 (CD103⁺Cd11b⁻) and CD103^+/− cDC2 in the mLNs of Zbtb46-DTR cAPC mice fed the hi Saa diet and injected with PBS or DT.

(G) Correlation plot of cDC1 frequencies and M. schaedleri abundance in lo Saa-fed and hi Saa-fed mice cecal content DNA. Solid line represents linear regression line, and dashed lines represent 95% CI.

(H) Frequency and numbers of cDC1 (CD103⁺Cd11b⁻) and CD103^+/−cDC2 in the mLNs of GF or M. schaedleri-monocolonized mice fed the hi Saa diet.

(I) Frequency of cDC1 in the mLNs of WT mice fed the lo Saa diet and gavaged with mBHI or M. schaedleri CM thrice weekly for 2 weeks. Each symbol represents data from an individual mouse. Bar graphs indicate mean ± SEM. Data represent five independent experiments (C), three independent experiments (B and D–F), and two independent experiments (H and I). p values for Mann-Whitney U test (B, D, F, H, and I) or chi-squared contingency test (E) are shown. Spearman correlation ρ and p value (G).

See also Figures S3 and S5–S8.

Thus, we considered the pivotal role of cDC1s in CD8⁺ T cell-mediated anti-tumor immunity, which has been formally tested recently only in the MC38 CC model.^89,90 To investigate if cDC1s mediate the hi Saa diet-dependent effects on anti-tumor immunity and tumor growth we observed, we leveraged two approaches to deplete cDC1 in cAPC mice. First, we generated cAPC Batf3^−/− mice and fed these mice the lo or hi Saa diets (Figure 4C). Basic leucine zipper ATF-like transcription factor 3 (BATF3) is a transcription factor crucial for CD8α/XCR1-expressing cDC1 development, and Batf3-deficient mice are used to interrogate cDC1 function.^91–93 mLN cDC1s were drastically reduced in number in cAPC Batf3^−/− mice in both dietary groups (Figure 4D). Batf3 deficiency led to loss of the protective effect of the hi Saa diet as there was no difference in tumor numbers or neoplastic progression between the two diet groups (Figures 4C and S6A–S6C). While BATF3 is a cDC1-required transcription factor, Batf3-deficient mice develop small numbers of cDC1-like cells and DCs with mixed phenotypes through compensatory pathways and have impaired development of memory CD8 T cells.^94–96 Thus, as a second approach for depleting cDC1 in cAPC mice, we generated bone marrow chimeric mice with hematopoietic cells from Zbtb46-DTR mice, which express the diphtheria toxin receptor (DTR) under the regulation of the conventionalDC-specific promoter upstream of the zinc-finger transcription factor Zbtb46.⁹⁷ While this transcription factor is expressed in all conventional DCs, published work demonstrated that DTR-mediated depletion preferentially targets cDC1s, which are almost completely ablated, while cDC2s populations are only reduced by 50%.⁹⁸ These cAPC Zbtb46-DTR mice were fed the hi Saa diet (which drives mLN cDC1 increases) and received diphtheria toxin (DT) injections starting 4 weeks post-irradiation and bone marrow reconstitution to deplete cDC1 or phosphate-buffered saline (PBS) injections as a control (Figure 4E). We observed a significant deletion of mLN cDC1s and a corresponding increase in the frequency of neoplastic lesions even in the presence of the hi Saa diet (Figures 4E, 4F, and S6D–S6F).^91–93 In aggregate, these complementary approaches supported the role of cDC1s in the immunoregulation of tumor growth in cAPC mice and link diet-microbial interactions and anti-tumor immunity.

M. schaedleri abundance correlates with cross-presenting cDC1 with an activated transcriptional signature with hi Saa diet

We previously observed that Saa diets affected the colonic mucus layer and M. schaedleri abundance in WT mice. To determine if the increase in mLN cDC1s required a tumorigenic event and was dependent on microbial factors, we fed the Saa diets to WT and GF mice. The hi Saa diet specifically increased cDC1 in the mLN of WT SPF mice but not in the colon (Figures S7A and S7B), and this effect was microbiota-dependent, as it was absent in GF mice (Figures S7C and S7D). M. schaedleri was the only microbiota member whose abundance significantly changed with the Saa diet regimen, and utilizing data from conventionally reared mice (WT and cAPC), we found a significant positive correlation (Spearman ρ = 0.585, p = 0.0002) between M. schaedleri cecal abundance and mLN cDC1 frequency (Figure 4G).

While M. schaedleri’s mucus localization increases its potential for interactions with host cells, the immunomodulatory effects and underlying mechanisms of M. schaedleri remain less well characterized. M. schaedleri induces a T_H1-type colitis in Nod2^−/− × Cybb^−/− mice^99,100 and functions in peripheral regulatory T cell development.¹⁰¹ To investigate potential diet-M. schaedleri-cDC1 interactions, we measured M. schaedleri levels in the cDC1 depletion models. cDC1 depletion did not affect hi Saa-dependent M. schaedleri increases (Figures S6G and S6H). To assess if M. schaedleri contributed to cDC1 increases, we monocolonized hi Saa diet-fed GF mice with M. schaedleri, which was sufficient to drive the hi Saa-dependent increase in mLN cDC1 frequency and number (Figure 4H); colonic cDC1s were not affected (Figure S7E).

Many microbes influence immune responses and CC development through their metabolites.^{23,27,102–105} We tested if abiotic M. schaedleri conditioned medium (CM) affected mLN cDC1 frequencies, by gavaging CM to WT mice fed the lo Saa diet, to avoid an M. schaedleri bloom. M. schaedleri CM gavage over a 2-week period increased cDC1 frequency in mLNs, while the abiotic medium control had no effect (Figure 4I). M. schaedleri and its specific secreted factors were sufficient for the hi Saa diet-dependent increase in cDC1s, supporting cDC1’s role as a mediator of anti-tumor immunity via the effects of a hi Saa diet-M. schaedleri bloom.

To elucidate the dietary Saa-M. schaedleri effects on cDC1, we performed single-cell RNA sequencing (scRNA-seq) on DC-enriched cells from the mLNs of cAPC mice fed the lo vs. hi Saa diets. Sequencing of ~12,500 CD11c⁺MHCII⁺CD64⁻-enriched cells revealed substantial cellular heterogeneity (Figure S8A; Table S1).^106,107 Deeper analysis of clusters representing cDCs and other mononuclear myeloid cells with manual curation of known DC markers revealed a DC progenitor (pre-DC) and six DC clusters, with the largest cluster formed by CD11b⁺ cells, which include cDC2s, macrophages, and monocytes, followed by clusters of migratory DCs (expressing CCR7), DC-SIGN⁺(CD209) DCs, and a cDC1 cluster identified by the expression of Xcr1, Clec9a, and Irf8 genes (Figure S8A; Table S1). Fast gene set enrichment analysis of the hallmark gene set pathways¹⁰⁸ revealed enrichment of the inflammatory response pathway (M5932) in cDC1s from hi Saa diet-M. schaedleri-exposed mice (HSME) (Figure S8B), while cell-cycle-related pathways were enriched in cDC1s from lo Saa diet-fed mice (Figure S8B).

Differentially expressed genes (DEGs) that were higher in cDC1s from the hi Saa diet include Cxcl9 and Ccl4 (Figure S8C; Table S1), both of which play critical roles in anti-tumor immune responses by recruiting T cells.^92,109,110 Expression of Tmem176b, a gene that encodes a cation channel regulating phagosomal acidification and functions in antigen cross-presentation,¹¹¹ was also higher in cDC1s from hi Saa diet-fed mice. Other transcripts expressed at higher levels in cDC1s from hi Saa-fed mice included genes encoding the calcium binding proteins S100A6 and S100A4, annexin A2, and IFN-inducible transmembrane protein 3 (Ifitm3), which function in endosomal, phagosomal, and vesicular trafficking; effective DC antigen presentation; co-stimulatory molecule expression; and T cell activation.^112–114 Genes with statistically lower expression in cDC1s from the hi Saa diet and higher expression in cDC1s from lo Saa diet-fed mice included the thioesterase Ppt1, which balances viral resistance vs. T cell cross-priming,¹¹⁵ and decreased Ppt1 expression enhances tumor clearance in mice.¹¹⁶ Overall, our scRNA-seq analysis showed that cDC1s from HSME mice displayed higher transcription of activation and immunostimulatory genes that facilitate cross-presentation, consistent with our findings that the hi Saa diet restricted tumor growth and reduced neoplastic progression.

To attempt to come full circle on our reverse translation approach, we generated cDC1 gene expression signatures from HSME and lo Saa diet-M. schaedleri-depleted (LSMD) mice (Table S1), using the DEGs specific to both cDC1 states, and examined their association with survival data from the Tissue Cancer Gene Atlas (TCGA) colon AdenoCA (COAD) and rectal AdenoCA (READ) (TCGA COADREAD, N = 359) bulk RNA-seq transcriptomic datasets. People whose tumors had an HSME cDC1 transcriptional (HSME-DC) signature had a longer survival time (Cox model, p = 0.005 for univariate analysis and p = 0.017 for disease stage-adjusted multivariate analysis), compared with people whose tumors had an LSMD cDC1 transcriptional (LSMD-DC) signature (Figure S8D). Individuals with so-called LSMD-DC tumors had an increased hazard ratio (HR) of 1.611 (1.152–2.253, 95% confidence interval [CI]) in univariate analysis and an HR of 1.521 (1.077–2.147, 95% CI) in multivariate analysis including stage. This analysis revealed that hi Saa diet-fed mice have cDC1s with a gene signature that correlated with improved survival in CC patients. Future prospective studies in individuals with CC should include scRNA-seq and/or bulk RNA-seq from tumor-draining lymph nodes and data collection on overall and disease-free survival, which will enable robust testing of the HSME-DC and LSMD-DC scores.

As our data linked activation and expansion of cDC1, Saa diet, and M. schaedleri to anti-tumor immunity, we tested if M. schaedleri microbial products acted directly on cDC1s. We examined abiotic M. schaedleri CM capacity to directly enhance cDC1 activation by inducing expression of co-stimulatory molecules CD80 and CD86 and to promote cDC1-mediated CD8⁺ T cell proliferation and IFN-γ production after antigen cross-presentation. Contrary to cDC1 expansion in vivo after M. schaedleri colonization or CM administration (Figures 4H and 4I), our in vitro testing showed no direct effect of M. schaedleri CM (Figures S8E–S8G). These data suggested that a dietary Saa-M. schaedleri-cDC1 axis was crucial for CC anti-tumor immunity, and unidentified factors relayed the Saa-M. schaedleri signals critical for cDC1 activation.

NKT cell-produced XCL1 is essential for the anti-tumor effect of dietary Saa

To determine what may drive cDC1 accumulation following hi Saa diet or M. schaedleri administration, we interrogated the role of the chemokine XCL1, a critical chemoattractant for cDC1s, which are characterized by hi expression of its cognate receptor XCR1 (Table S1).^117,118 We administered an α-XCL1 neutralizing antibody to hi Saa diet-fed cAPC mice to reduce the levels of this chemokine.¹¹⁰ Akin to cDC1 depletion, reduction of XCL1 levels led to a significant increase in AdenoCA frequency and a decrease in mLN cDC1 frequency in hi Saa diet-fed cAPC mice (Figures 5A, 5B, and S9A–S9C), even though the α-XCL1 Ab only reduced serum XCL1 levels by 2-fold on average (Figure 5C).

Figure 5. Hi Saa diet or M. schaedleri promote anti-tumor immunity through induction of NKT cells and XCL1 secretion.

(A) Experimental scheme and data on dysplastic and neoplastic lesion grade and numbers from cAPC mice fed hi Saa diet and treated with either α-XCL1 antibody or isotype control.

(B) Frequency of cDC1 (CD103⁺Cd11b⁻) and CD103^+/− cDC2 in the mLNs of cAPC mice fed hi Saa diet and treated with α-XCL1 antibody or isotype control.

(C) Serum concentration of XCL1 in cAPC mice fed hi Saa diet and injected with α-XCL1 or isotype.

(D) Ex vivo secretion of XCL1 from NK.1.1⁺ NK/ILC1 cells, γδ T cells, CD4⁺, CD8⁺ CD4⁻CD8⁻ T cells, and NKT cells sorted from mLNs from WT mice fed the Saa diets.

(E) Frequencies and numbers of NKTs from mLNs of mice fed lo or hi Saa diet.

(F) Secretion of XCL1 from primary murine NKT cell line (GW1) following overnight stimulation with M. schaedleri CM or controls.

(G) Secretion of XCL1 from human peripheral blood-derived NKT cells following overnight stimulation with M. schaedleri CM or controls.

(H) Experimental scheme and data on dysplastic and neoplastic lesion grade and numbers from cAPC mice fed lo or hi Saa diet and treated with either isotype control or the NKT-depleting antibody NKT14. Each column represents data from a single mouse, and the number of each lesion type is shown. Each symbol represents data from an individual mouse (B and D) or from an individual culture well (C, E, and F). Bar graphs indicate mean ± SEM. Data of a representative experiment (C), two independent experiments (H), and at least three independent experiments (A, B, and D–G). p values for Mann-Whitney U test (B, C, and E), Brown-Forsythe ANOVA test with unpaired t with Welch’s correction (D, F, and G) or chi-squared contingency test (A and G) are shown. Full statistical analyses for (F) and (G) are in Table S3.

See also Figures S9 and S10.

Multiple lymphocyte populations, particularly tissue-resident CD8⁺ T cells, NK cells, and NKT cells, secrete XCL1 and thereby direct the critical role of cDC1s in orchestrating CD8⁺ T cell-mediated anti-tumor immunity.^118–120 To determine if XCL1 secretion by these lymphocytes was affected by dietary Saa, we sorted lymphocytes from the mLNs of mice fed the two Saa diets. On a per-cell basis, NKT cells from hi Saa diet-fed mice secreted higher XCL1 levels when cultured ex vivo, as compared with NKT cells from lo Saa diet-fed mice or other lymphocytes (NK/ILC1, γδT, CD4⁺T, CD8⁺T, CD4⁻8⁻T cells) (Figure 5D). mLN NKT but not NK1.1⁺CD3⁻ (NK/ILC1) cell frequency and number increased in hi Saa diet-fed vs. lo Saa-diet-fed cAPC mice (Figures 5E and S10A), suggesting that hi Saa diet induced both secretion of XCL1 from NKT cells and their expansion in mLNs.

To test if M. schaedleri directly affected NKT cell XCL1 production, we employed an in vitro system using a primary murine NKT cell line (GW1)¹²¹ and abiotic CM. Compared with untreated cells and cells treated with fresh bacterial medium (modified brain-heart infusion [mBHI]), M. schaedleri CM stimulated XCL1 production at a magnitude comparable to interleukin (IL)-12 treatment (Figure 5F). The M. schaedleri-derived products that stimulated NKT cell XCL1 production were heat-labile and primarily found in the organic phase of the supernatant (Figure 5F). These results led us to interrogate the potential of M. schaedleri CM in stimulating human NKT cells. We expanded NKT cells from healthy donor peripheral blood mononuclear cells before treating them with M. schaedleri CM. There was a ~3-fold increase in XCL1 production from CM-treated NKT cells, compared with mBHI-treated or untreated cells (Figure 5G).

NKT cells, particularly invariant NKT cells, have specific T cell receptors (TCRs) that respond to lipid antigens presented by CD1 molecules.^122–124 We tested if M. schaedleri CM products were antigens that activated NKT cells through TCR stimulation by using two different assays. First, we incubated CD1d-expressing antigen-presenting cells (APCs) (A20-CD1d APC cell line) with the CM overnight, removed the CM, and added the NKT cell hybridoma (DN3A4–1.2) that produces IL-2 upon TCR stimulation. Additionally, we used an APC-free assay employing CD1d protein-coated wells. α-galactosylceramide (α-GalCer) was used as positive control. Irrespective of the assay used, we found no evidence that M. schaedleri CM products bind to CD1d and stimulate NKT cells through TCR engagement (Figures S10C and S10D). These data suggest that NKT cell expansion and XCL1 production link dietary Saa, M. schaedleri, cDC1s, and anti-tumor immunity and that M. schaedleri’s bioactivity cannot be explained by furnishing CD1d ligands.

To determine if NKT cells were essential to the anti-tumor effect of dietary Saa, we used an antibody, NKT14, that depletes NKT cells in vivo.¹²⁵ We administered NKT14 to cAPC mice fed the hi or lo Saa diets. Depletion of NKT cells abrogated the anti-tumor effect of the hi Saa diet and led to a significant increase in AdenoCA s in cAPC mice (Figures 5H and S9D–S9F). These results supported that NKT cells integrated hi Saa diet-M. schaedleri cues, as NKT depletion in lo Saa diet-fed mice did not affect CC progression. Having observed signals of the Saa diet DC signatures in human CC survival outcomes (Figure S8D) and having established the import of NKT cells for the anti-tumor dietary Saa-microbiome effects in a CC preclinical model (Figure 5H), we probed if there were connections between clinical microbiome interventions, cDC1, and NKT cells and improvement in anti-tumor immunity in patients, using publicly available data. As CCs are largely MMR-proficient and rarely receive ICIs and rectal cancers treated with ICIs are often also treated with neoadjuvant chemotherapy or chemoradiotherapy, which can perturb the microbiome and alter the tumor microenvironment, we were constrained in the clinical cohort data available. Additionally, we required RNA-seq tumor data both before and after microbiome intervention. One study included in our analyses (Figures 1A and 1B) that focused on melanoma, Baruch et al.,²⁶ fit these requirements. We examined if there were any associations between the HSME-DC and LSMD-DC signatures (Figure S8D) and tumor gene expression data from the Baruch et al. study pre- and post-fecal microbiota transplantation (FMT) stratified by clinical response (NR vs. R). The HMSE-DC signature expression increased post-FMT in all R patients (Figures S11A and S11B). We also generated an NKT gene signature that included 18 marker genes (Table S1), based on human NKT cells scRNA-seq data from Zhou et al.¹²⁶ All R patient samples had an increase of the NKT gene signature (Figure S11C–S11E). These observations stir interest in the NKT-cDC1 axis and raise curiosity and enthusiasm about ways to boost cDC1- and NKT-mediated immunity in cancer.

DISCUSSION

We identified the enrichment of Saa pathways in the gut microbiome as a shared feature of ICI responsiveness across cohorts. Targeting these pathways via dietary Saa supplementation slowed tumor progression in both MMR-deficient and -proficient CC mouse models. Here, we identified a dietary Saa-M. schaedleri-NKT cell axis that promoted cDC1 activity and anti-tumor immunity.

There is tremendous interest in dietary interventions for the prevention and treatment of cancer.^28–33 We acknowledge that these approaches might display pleiotropic effects and hormetic responses. The effect of severe restriction of the Saa methionine has been investigated in animal models with conflicting results. Methionine-dependence has been reported in CC tumor cells, and methionine restriction limits tumor growth in mouse models.^127–130 However, a different study showed that methionine restriction impairs anti-tumor immunity by altering the gut microbiota.⁶⁰ These seemingly contradictory findings should be interpreted from a dosage perspective—a very lo Saa diet can restrict cancer cell growth due to tumor cell metabolic dependency on exogenous methionine, while a moderate increase in dietary Saa may enhance anti-tumor immunity. In our study, a lo Saa diet did not affect tumor growth, likely because it was not sufficiently methionine restrictive.

Our intervention was distinct from emerging data from our group on total dietary sulfur intake and its correlation with human colonic adenomas and CC.^131,132 Work from our group and others has also shown that dietary Saa modulation has pleiotropic effects on the host and microbiome, from altering redox potential to protein posttranslational modifications, all within a dose range that is non-toxic for the host.^61,133 Given the anti-tumor immune effects observed herein, Saa supplementation may warrant clinical investigation in CC patient cohorts. In addition, further investigation in ICI-refractory patients should also be considered, given that this pathway was enriched in the microbiota of ICI NR melanoma patients who received a fecal microbiota transplant and became ICI responsive.^25,26

Diet-microbe interactions in patients are likely more complex than the single microbe change observed in our mouse model studies; nevertheless, M. schaedleri provided us with a useful model to understand that mucus may be a key link connecting diet, the microbiota, and immune function. Here, we found that M. schaedleri’s immunomodulatory effects on NKT cells were an under-explored link in the chain connecting diet, gut microbial factors, and enhanced cDC1 cell function, converging on anti-tumoral CD8⁺ T cells. Our findings complemented and extended prior data on cDC1s and their intriguing role in anti-tumor immunity and added indirect microbial signals as a rheostat of their function.

Collectively, this study highlighted the utility of mouse models in exploring diet-microbe interactions not only for identifying biogeographies such as the mucus layer or tumor-draining lymph nodes that warrant further investigation but also for identifying CC prognostic signatures that lie at the interface of diet, the microbiome, and host immunity. Overall, our study provided a proof of concept for a reverse translation framework to examine patient stool metagenomic profiles from ICI-responsive patients to unravel mechanisms by which microbiota-diet interactions can enhance anti-tumor immunity.

Study limitations

We acknowledge limitations of our study regarding the in vivo cDC1 genetic depletion models employed. Zbtb46-DTR-dependent DC depletion is not completely specific for cDC1 and decreases total DCs. DC-depleted mice in this model had increased early neoplastic events, when compared with the later stage differences observed in cAPC mice fed the Saa diets or in cAPC mice fed the hi Saa diet and depleted of XCL1. This can be explained by a delay in the dynamics of tumor progression caused by the total body lethal irradiation required to generate these bone marrow chimeric mice. The other approach we employed for cDC1 perturbation was crossing the cAPC line to Batf3^−/− mice. Batf3 deficiency can impact other immune cells, and in Batf3^−/− mice, a small number of cDC1-like cells can develop through compensatory or alternative pathways.^94–96 However, these complementary cDC1 depletion approaches in conjunction with the data on depleting XCL1—a critical cDC1 chemoattractant and the ligand for the XCR1 receptor that characterizes cDC1s^117,118—support a central role of cDC1 in explicating the links we observed connecting Saa, M. schaedleri, and enhanced immunity against CC. Future studies expanding on the production of immune effectors by NKT cells triggered by M. schaedleri identified metabolites and on immune pathways activated in cDC1s indirectly and directly by microbial metabolites will provide further mechanistic insights on the anti-tumor effects of high-Saa feeding and M. schaedleri.

While our studies found that M. schaedleri and its derived soluble compounds induced NKT cell production of XCL1, we came short of identifying the exact metabolite or metabolites responsible for this effect. Comparative analyses of untargeted metabolomics from fractionated serum and colonic contents from M. schaedleri-monocolonized mice and hi vs. lo Saa diet-fed mice as well as from M. schaedleri itself are useful next steps in identifying M. schaedleri immunomodulatory bioactives.

We acknowledge the limitations of using a study of nine ICI-refractory melanoma patients who received FMT prior to anti-PD1 monotherapy re-challenge to test diet-microbe-immune gene signatures. Currently, no suitable clinical cohort data are available to validate a cDC1-NKT cell axis correlation with improved anti-tumor immunity, following a diet and microbiota perturbation in CC. Future studies interrogating the microbiome-immune axis in anti-tumor immunity should possess a direct microbiota intervention, RNA-seq immunological readouts from tumor and lymph node tissues, and clinical outcome data. In particular, future FMT-ICI studies with healthy FMT donors that harbor colonic M. schaedleri, cancer clinical trials with FMT donors with or without M. schaedleri, or studies where FMT is combined with or without cysteine and methionine supplementation would better test the translation of our observations on diet, M. schaedleri, cDC1, and NKT cells from preclinical models to cancer-directed patient care.

RESOURCE AVAILABILITY

Lead contact

Further information and requests should be directed to the lead contact, Wendy S. Garrett (wgarrett@hsph.harvard.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

All sequencing data were deposited in the NCBI BioProject database (accession number PRJNA688219) and are publicly available as of the date of publication. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mouse models

C57BL/6, C57BL/6 CD45.1, Cdx2^Cre mice to Apc^flox/+ (cAPC), Zbtb46-DTR, cAPC Batf3^−/− and C57BL/6-Tg(TcraTcrb)1100Mjb/J (OT-I) mice were bred and housed at the Harvard T.H. Chan School of Public Health (HSPH) barrier facility with constant ambient temperature of 24°C and 12 h of day/night cycles. All mouse strains were initially purchased from Jackson Laboratory. Mice of both sexes were used for experiments, as we do not observe sex differences in neoplastic progression in our vivarium. As cAPC mice rarely show poor body condition in our vivarium, no mice were excluded for downstream analysis due to weight loss or diarrhea. In Zbtb46-DTR cAPC chimeric mice experiments, there was a significant mortality of cAPC mice shortly after irradiation, thus mice that died in the first four weeks after irradiation and before diphtheria toxin administration were excluded from the analysis. For gnotobiotic experiments, mice were housed at the Harvard T.H. Chan Gnotobiotics Center for Mechanistic Microbiome Studies in semi-rigid isolators (Plastic Concepts Inc.) and experiments were conducted in individual ventilated IsocageP system (Tecniplast). Routine surveillance, including 16S rRNA gene amplicon sequencing and qPCR analyses (using universal 16S rDNA primers) and Sanger sequencing, were performed on fecal samples and cage swabs to validate the gnotobiotic status (e.g., germ-free, monocolonized, or ASF) of the mice. For re-derivation of cAPC mice, pregnant female mice were euthanized, and their uteri were removed under aseptic conditions, using chemical sterilant (MB-10, Quip Labs) in a semi-rigid isolator. The sterilized pups were introduced to germ-free foster dams. After 3 weeks, the germ-free status of the mice and their genotypes were evaluated using PCR. Animal studies and experiments were approved and carried out in accordance with Harvard Medical School’s Standing Committee on Animals and the National Institutes of Health guidelines for animal use and care.

Mucispirillum schaedleri culture

Mucispirillum schaedleri was cultured in modified brain-heart infusion (mBHI) medium composed of 37 g BHI, 5 g yeast extract, 2 mg vitamin K, 5 mg hemin, 0.5 g L-cysteine and 150 mL fetal bovine serum in 1 L tap water. The medium was pH adjusted to 7.2 and filter sterilized using a 0.2 μm bottle-top sterile filter unit. For agar plates, the same medium was used with the addition of 12 g of agar per 1 L. M. schaedleri was grown in mBHI under anaerobic conditions (Coy anaerobic chamber with 80% N₂, 10% CO₂ and 10% H₂ gas mix atmosphere) at 37°C for 5 days, at which point a bacterial pellet was observed.

Culture of cell lines

MC38 mouse colon adenocarcinoma cells¹⁶⁵ (Kerafast) were grown in RPMI GlutaMAX medium supplemented with 10% fetal bovine serum (FBS), 500 U/mL penicillin/streptomycin, 1 mM sodium pyruvate and 50 μM β-mercaptoethanol. The B cell lymphoma A20 cells (ATCC) (A20-CD1d antigen-presenting cell-line, APCs) were grown in RPMI GlutaMAX medium supplemented with 10% FBS, 500 U/mL penicillin/streptomycin. DN3A4-1-2 hybridoma,¹³⁴ derived from C57BL/6 NK1.1⁺ thymocytes, was grown in RPMI GlutaMAX medium supplemented with 10% FBS, 500 U/mL penicillin/streptomycin supplemented with 104 U/mL of rIL-2 and 6 × 10³ ng/mL of IL-7 (BD Pharmingen). All cells were grown at 37°C, 5% CO_2, in a humidified incubator.

METHOD DETAILS

Sulfur amino acid (Saa) diets formulation and dietary interventions

Sulfur amino acid (Saa) diets were formulated based on the published literature^166,167 and represent edge cases of Saa consumption, with the following considerations. Human dietary cysteine and methionine consumption typically ranges between 0.03–0.06 g/Kg body weight/day.^57,168,169 Across human diets with a range in their protein consumption (44 g-140 g/day), lo levels of cysteine are between 0.01–0.04 g/Kg/d.¹⁷⁰ Very hi levels of methionine or cysteine are in the range >=6 g/Kg/day; such levels raise concerns for contributing to homocysteinemia in humans. Given these data and with veterinary approval at our university, we formulated the two isocaloric diets employed in these studies (see Table S2 for the diet formulations) and had them manufactured by Research Diets, Inc. For gnotobiotic experiments, the same formulations, irradiated, were ordered from Test Diet.

At 6–8 weeks of age, WT, cAPC, cAPC Batf3^−/− or cAPC Zbtb46-DTR mice were transitioned to Saa diets. After 12 weeks on the Saa diets, mice were sacrificed and tissues (normal, neoplastic, and adjacent normal) were collected either for histology, flow cytometry or immunofluorescence, and cecal contents were frozen for microbial analysis. For gnotobiotic experiments, GF, ASF and M. schaedleri-monocolonized mice were transferred to Saa diets at 6–8 weeks of age. After 4 weeks, mice were sacrificed, and tissues were analyzed by flow-cytometry. GF cAPC mice transitioned to the Saa diets on weaning and were maintained on the diets for 12 weeks. In bacterial CM feeding experiments, WT SPF mice fed lo Saa-diet were gavaged every two days with 100 μL of 5 to 7-day-old filtered culture supernatant of M. schaedleri or sterile mBHI medium.

cAPC Zbtb46-DTR bone marrow chimeras and DT depletion

To generate cAPC Zbtb46-DTR mice, cAPC mice were lethally irradiated with one dose of 10 Gy (1000 rad) at 10 weeks of age and then injected with 10⁶ bone-marrow cells from Zbtb46-DTR mice as described.⁹⁸ To deplete Zbtb46-expressing cells in cAPC Zbtb46-DTR bone chimera mice, mice were injected with 400 ng (~20 ng/g body weight) of DT 4 weeks after irradiation, followed by twice weekly injections of 100 ng (~4 ng/g body weight) to maintain the depletion of Zbtb46-expressing cells for the rest of the experiment. We used C57BL/6 CD45.1 bone-marrow cells to confirm the efficiency of the transplantation protocol.

Antibody depletion of XLC1 and NKT cells

For XCL1 depletion, cAPC mice were fed the hi Saa diet for 8 weeks and then injected i.p. with 300 μg of anti-XCL1 antibody¹¹⁰ or InVivoMAb rat IgG1 Isotype control, anti-trinitrophenol (BioXcell) every 2–3 days for 4 weeks. For NKT depletion, 6–8 weeks old cAPC mice were fed hi or lo Saa diet for 2 weeks before i.p. injections of 75 μg of NKT14 antibody¹²⁵ or InVivoMAb mouse IgG2a Isotype control (BioXcell) every 2 weeks for 12 weeks.

Meta-analysis of microbial metabolic pathways in ICI and CC microbiomes

Fecal metagenomic shotgun reads for the ICI^10–26 and CC^37–48 datasets were obtained from the sequence read archive of the National Center for Biotechnology Information. Metagenomes with known use of antibiotics as described in available clinical metadata were excluded from meta-analysis. Reads were quality trimmed using Trimmomatic (v0.39),¹⁴⁶ which was configured to perform sliding window scan with the following parameters: “illumina_adapters:2:36:7:1:TRUE leading:3 trailing:3 slidingwindow:4:15 minlen:36 tophred33”. Reads mapping to human chromosomes (CHM13v2) and sequencing vector library (UniVec and EmVec) were removed using BWA-MEM (v.0.7.17-r1188),¹⁴⁷ Samtools (v.1.17),¹⁴⁸ and Picard (v.2.27.5).¹⁴⁹ Relative abundance profiles of MetaCyc pathways were then generated using the HUMAnN 3 pipeline (v3.8),¹⁵⁰ and these data were used to fit linear mixed-effects models (LMM) with study effect as random covariate as implemented in the lmerTest R/CRAN package.¹⁵¹ Effect sizes (Cohen’s d or standardized mean differences) were derived from fitted LMMs for MetaCyc pathways using the emmeans R/CRAN package.¹⁵²

MC38 colon adenocarcinoma flank tumor model

MC38 cells at ~80% confluence were harvested, washed in PBS and resuspended on ice in a 1:1 solution of Cultrex reduced growth factor basement membrane extract (R&D systems) and PBS at a concentration of 3.3×10⁶ cells/mL. WT C57BL/6J mice were put on lo or hi Saa diets two weeks prior to MC38 engraftment. On the day of engraftment, the mice were shaved on their left flank and injected subcutaneously with 150 μL (0.5×10⁶ cells) of the MC38 cell solution. Mice were monitored every other day and tumor volume was calculated using the Volume = (Width² x Length)/2 formula.¹⁷¹ 12 days post-engraftment, mice were sacrificed, tumors excised and weighed.

Histopathology

After sacrifice, colons were opened using blunt scissors and the luminal contents were removed. Tissues were fixed in 4% paraformaldehyde, processed, and paraffin-embedded using standard protocols by the rodent histopathology core at HMS. Five hematoxylin and eosin (H&E) stained slides (five level sections obtained 50 μm apart) for each sample were blindly evaluated by Dr. Jonathan Glickman (a board-certified GI pathologist) for neoplastic lesions. Neoplastic lesions were graded as: Aberrant crypt foci – microscopic cluster of abnormal crypts (5 or fewer) with cytologic dysplasia; Adenoma – discrete macroscopic lesion containing dysplastic epithelium (both low-grade and high-grade dysplasia were considered) without invasion; Adenocarcinoma – discrete macroscopic lesion containing dysplastic epithelium and invasion of lamina propria, submucosa or deeper structures.

Alcian Blue staining

Unopened colons were fixed overnight in methanol-Carnoy’s solution (60% methanol, 30% chloroform, 10% glacial acetic acid) followed by routine paraffin embedding and sectioning. Sections were deparaffinized, hydrated and stained with 1% Alcian Blue solution to visualize the mucus layer, then counterstained with Nuclear Fast Red (Amresco 1B1369). For quantification of sulfated mucins, colons were opened, luminal contents removed and fixed in 4% paraformaldehyde followed by routine paraffin embedding and sectioning. Consecutive tissue sections (“twin” sections) were stained with regular 1% Alcian Blue solution (pH2.5) or with Alcian Blue solution at pH1 (stains sulfated mucins). Approximately ~6 fields per mouse were imaged at the exact same areas on each “twin” section. Stained areas in each “twin” sections were used to calculate proportion of sulfated mucins. Histomount (National Diagnostic HS-103) was used as a mounting medium and images were acquired and quantified on Nikon Eclipse NI-U microscope. Images analysis was performed using NIS Elements (v4.11) or Fiji (v1.5).¹³⁸

Cecal DNA extraction and quantitative PCR (qPCR) analysis

Mouse cecal contents were collected into 1.5 mL tubes and immediately frozen in liquid N₂. Thawed cecal contents were resuspended in 300 μL of Tris-EDTA solution (100 mM Tris and 15 mM EDTA) in 2 mL tubes with ~300 μL of zirconium beads (20 micron) and 500 μL of TE-saturated phenol (Sigma-Aldrich) were added. The tubes were placed in a bead-beater for 2 min followed by centrifugation at max speed for 15 min. The aqueous phase was moved to a clean tube and 1 volume of phenol:chloroform:isoamyl alcohol (25:24:1) was added. Tubes were vortexed for 1 min and then centrifuged for 2 min at max speed. The procedure was repeated 3 times. The aqueous phase was moved to a clean tube and 2 volumes of 100% ethanol (EtOH) and 1/10 volume of 3M sodium acetate pH 5.2 were added. After 1 hour at −20°C, the tubes were centrifuged at 14,000 rpm for 20 min at 4°C, the liquid was discarded and 1 mL of cold 70% EtOH was added to wash the pellet, followed by another 20 min of centrifugation at 4°C at 14,000 rpm. Finally, the liquid was aspirated off, and the pellet was air-dried for 10 min at room temperature, followed by resuspension in 100 μL of sterile, molecular biology-grade H₂O. DNA concentration was measured by spectrophotometry at 260 nm. For RT-PCR analysis, 50 ng of cecal DNA were mixed with 10 μL 2X SYBR Green (KAPPA SYBR FAST) and 0.29 μM of each forward/reverse primer set (key resources table; Table S2) in a 20 μL reaction. Real-time PCR reactions were performed on an Agilent Technologies Stratagene MX3005P machine. Relative M. schaedleri abundance was presented using the 2^−ΔΔCt method and using lo Saa or PBS conditions as reference samples.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
FITC anti-mouse CD3ε Antibody (clone 145-2C11)	Biolegend	Cat# 100305; RRID:AB_312670
Pacific Blue™ anti-mouse CD3ε Antibody (clone 145-2C11)	Biolegend	Cat# 100333; RRID:AB_2028473
APC/Cyanine7 anti-mouse CD4 Antibody (Clone GK1.5)	Biolegend	Cat# 100413; RRID:AB_312698
Pacific Blue™ anti-mouse CD8a Antibody (Clone 53-6.7)	Biolegend	Cat# 100728; RRID:AB_493426
APC anti-mouse CD45 Antibody (Clone 30-F11)	Biolegend	Cat# 103111; RRID:AB_312976
Pacific Blue™ anti-mouse CD45 Antibody (Clone 30-F11)	Biolegend	Cat# 103125; RRID:AB_493536
Pacific Blue™ anti-mouse CD45.2 Antibody (Clone 104)	Biolegend	Cat# 109819; RRID:AB_492873
APC anti-mouse CD45.1 Antibody (Clone A20)	Biolegend	Cat# 110714; RRID:AB_313503
CD103 (Integrin alpha E) Monoclonal Antibody (2E7), PE	ThermoFisher Scientific	Cat# 12-1031-81; RRID:AB_465798
FITC anti-mouse CD11c Antibody (Clone N418)	Biolegend	Cat# 117305; RRID:AB_313774
PerCP anti-mouse/human CD11b Antibody (Clone M1/70)	Biolegend	Cat# 101229; RRID:AB_2129375
NK1.1 Monoclonal Antibody (PK136), PE-Cyanine7	ThermoFisher Scientific	Cat# 25-5941-82; RRID:AB_469665
APC/Cyanine7 anti-mouse I-A/I-E Antibody (Clone M5/114.15.2)	Biolegend	Cat# 107627; RRID:AB_1659252
CD223 (LAG-3) Monoclonal Antibody (eBioC9B7W (C9B7W)), PerCP-eFluor™ 710	ThermoFisher Scientific	Cat# 46-2231-80; RRID:AB_11151148
FITC anti-mouse CD279 (PD-1) Antibody (Clone 29F.1A12)	Biolegend	Cat# 135213; RRID:AB_10689633
PE anti-mouse CD152 (CTLA-4) Antibody (Clone UC10-4B9)	Biolegend	Cat# 106305; RRID:AB_313254
PE/Cyanine7 anti-mouse CD366 (Tim-3) Antibody (Clone B8.2C12)	Biolegend	Cat# 134010; RRID:AB_2632734
PE anti-mouse CD64 (FcγRI) Antibody (Clone X54-5/7.1)	Biolegend	Cat#139303; RRID:AB_10613467
PE/Cyanine7 anti-mouse IFN-γ Antibody (Clone XMG1.2)	Biolegend	Cat# 505826; RRID:AB_2295770
Alexa Fluor® 647 anti-human/mouse Granzyme B Antibody (Clone GB11)	Biolegend	Cat# 515406; RRID:AB_2566333
APC anti-mouse CD80 Antibody (Clone 16-10A1)	Biolegend	Cat# 104713; RRID:AB_313134
APC/Cyanine7 anti-mouse CD86 Antibody (Clone GL-1)	Biolegend	Cat# 105029; RRID:AB_2074993
PE-labeled CD1d tetramer (unloaded)	NIH Tetramer Core Facility	N/A
PE-labeled CD1d tetramer (CD1d PBS-57)	NIH Tetramer Core Facility	N/A
Anti-CD3 epsilon antibody	Abcam	Cat# ab5690; RRID: AB_305055
CD8a Monoclonal Antibody (4SM15)	ThermoFisher Scientific	Cat# 14-0808-80; RRID: AB_2572860
Alexa Fluor® 594 AffiniPure® Donkey Anti-Rat IgG (H+L)	Jackson ImmunoResearch	Cat# 712-585-153; RRID: AB_2340689
Peroxidase AffiniPure® Donkey Anti-Rabbit IgG (H+L)	Jackson ImmunoResearch	Cat# 711-035-152; RRID: AB_10015282
anti-XCL1 antibody	Laboratory of Dr. Shigeyuki Kon	N/A
InVivoMAb rat IgG1 Isotype control, anti-trinitrophenol	BioXcell	Cat# BE0290; RRID: AB_2687813
NKT14 antibody	Laboratory of Dr. Mitchell Kronenberg	N/A
InVivoMAb mouse IgG2a isotype control, unknown specificity	BioXcell	Cat# BE0085; RRID: AB_1107771
anti-Muc2 antiserum	Laboratory of Dr. Gunnar C. Hansson	N/A
Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 488	ThermoFisher Scientific	Cat# A-21206; RRID: AB_2535792
Purified anti-mouse CD3ε Antibody (Clone 145-2C11)	Biolegend	Cat# 100301; RRID: AB_312666
Purified anti-mouse CD28 Antibody (Clone 37.51)	Biolegend	Cat# 102101; RRID: AB_312866
Ultra-LEAF™ Purified anti-mouse CD16/32 Antibody	BioLegend	Cat# 101340; RRID: AB_2616684
Purified anti-human CD28 Antibody (Clone CD28.2)	BioLegend	Cat# 302901; RRID: AB_314303
Purified anti-human CD3 Antibody (Clone HIT3a)	BioLegend	Cat# 300301; RRID: AB_314037
APC anti-human CD3 Antibody (Clone HIT3a)	BioLegend	Cat# 300311; RRID: AB_314047
PE anti-human TCR Vα24-Jα18 (iNKT cell) Antibody (Clone 6B11)	BioLegend	Cat# 342904; RRID: AB_2078494
Human TruStain FcX™ (Fc Receptor Blocking Solution)	BioLegend	Cat# 422302; RRID: AB_2818986
BD Pharmingen™ Purified Rat Anti-Mouse IL-2 (Clone JES6-1A12)	BD Biosciences	Cat# 554424; RRID: AB_395383
BD Pharmingen™ Biotin Rat Anti-Mouse IL-2 (Clone JES6-5H4)	BD Biosciences	Cat# 554426; RRID: AB_395384
Bacterial and virus strains
Mucispirillum schaedleri	This study (isolated from mouse colony)	N/A
Biological samples
buffy coats from human donors	Research Blood Components	N/A
Chemicals, peptides, and recombinant proteins
L-Amino Acid Rodent Diet With 0.8% L-Cystine and 1.5% Methionine (High Saa Diet)	Research Diets	Cat# A15121501
L-Amino Acid Rodent Diet With 0.03% L-Cystine and 0.3% Methionine (Low Saa Diet)	Research Diets	Cat# A15121502
Diphtheria Toxin from Corynebacterium diphtheriae	MilliporeSigma	Cat# D0564
OVA EndoFit™	InVivoGen	Cat# vac-pova
Mouse IL-7 Recombinant Protein, PeproTech®	ThermoFisher Scientific	Cat# 217-17-10UG
Mouse IL-2 Recombinant Protein, PeproTech®	ThermoFisher Scientific	Cat# 212-12-20UG
Mouse IL-12 p70 Recombinant Protein, PeproTech®	ThermoFisher Scientific	Cat# 210-12-10UG
α-Galactosylceramide	Enzo Life Sciences	Cat# BML-SL232-0100
Recombinant Human IL-2 (Cys145Ser) Protein, CF	R&D Systems	Cat# 1081-IL-020/CF
Human IL-2 Recombinant Protein, PeproTech®	ThermoFisher Scientific	Cat# 200-02-10UG
BD Pharmingen™ Recombinant Mouse IL-2	BD Biosciences	Cat# 550069
mouse CD1d, unbiotinylated monomer	NIH Tetramer Core Facility	N/A
RPMI 1640 Medium, GlutaMAX™ Supplement	Thermo Fisher Scientific	Cat# 61870036
HEPES	Thermo Fisher Scientific	Cat# 15630080
penicillin/streptomycin	Thermo Fisher Scientific	Cat# 15140122
sodium pyruvate	Thermo Fisher Scientific	Cat# 11360070
MEM Non-Essential Amino Acids Solution	Thermo Fisher Scientific	Cat# 11140050
β-mercaptoethanol	Thermo Fisher Scientific	Cat# 31350010
DNase I (Roche)	Millipore Sigma	Cat# 10104159001
collagenase D (Roche)	Millipore Sigma	Cat# 11088882001
Dispase (5 U/mL)	StemCell Technologies	Cat# 07913
Fetal Bovine Serum, Value	Thermo Fisher Scientific	Cat# A5256801
BD BACTO™ Brain Heart Infusion	BD	Cat# 237500
BD DIFCO™ Brain Heart Infusion Agar	BD	Cat# 241830
Bacto™ Yeast Extract	BD	Cat# 212750
Vitamin K1	MilliporeSigma	Cat# 95271
Hemin	MilliporeSigma	Cat# 51280
L-Cysteine	MilliporeSigma	Cat# 168149
Qiazol	Qiagen	Cat# 79306
Critical commercial assays
TSA Plus Fluorescein kit System	PerkinElmer	Cat# NEL741E001KT
ProLong™ Gold Antifade Mountant	ThermoFisher Scientific	Cat# P36934
DAPI	MilliporeSigma	Cat# 32670
LIVE/DEAD™ Fixable Yellow Dead Cell Stain Kit	ThermoFisher Scientific	Cat# L34959
LIVE/DEAD™ Fixable Near-IR Dead Cell Stain Kit	ThermoFisher Scientific	Cat# L10119
Alcian blue 1% in 3% acetic acid pH 2.5	Avantor	Cat# 101410-878
Nuclear Fast Red solution	Avantor	Cat# 101410-974
Histomount	National Diagnostics	Cat# hs-103
KAPA SYBR FAST Universal Master Mix	KAPA Biosystems	Cat# KK4618
Platinum™ Hot Start PCR Master Mixes (2X)	ThermoFisher Scientific	Cat# 13000014
AMPure XP PCR Purification Reagent	Beckman Coulter	Cat# A63880
Qubit dsDNA BR Assay Kit	ThermoFisher Scientific	Cat# Q32850
TrueBlack® Lipofuscin Autofluorescence Quencher	Biotium	Cat# 23007
Fluorescein-5-Isothiocyanate	ThermoFisher Scientific	Cat# F1906
Wheat Germ Agglutinin (WGA), Alexa Fluor™ 647	ThermoFisher Scientific	Cat# W32466
DNase I (Roche)	Millipore Sigma	Cat# 10104159001
collagenase D (Roche)	Millipore Sigma	Cat# 11088882001
Dispase (5 U/mL)	StemCell Technologies	Cat# 07913
Dead Cell Removal Kit	Miltenyi Biotec	Cat# 130-090-101
CD8a+ T Cell Isolation Kit, mouse	Miltenyi Biotec	Cat# 130-104-075
eBioscience™ Foxp3 / Transcription Factor Staining Buffer Set	ThermoFisher Scientific	Cat# 00-5523-00
Ebioscience™ Protein Transport Inhibitor Cocktail (500X)	ThermoFisher Scientific	Cat# 00-4980-93;
Pan Dendritic Cell Isolation Kit, mouse	Miltenyi Biotec	Cat# 130-100-875
BD Cytofix	BD Biosciences	Cat# 554655
CellTrace™ CFSE Cell Proliferation Kit, for flow cytometry	ThermoFisher Scientific	Cat# C34554
Mouse XCL1/Lymphotactin DuoSet ELISA	R&D Systems	Cat# DY486
Mouse XCL1 ELISA Kit (Lymphotactin)	Abcam	Cat# ab269540
CD14 MicroBeads, human	Miltenyi Biotec	Cat# 130-050-201
Anti-iNKT MicroBeads, human	Miltenyi Biotec	Cat# 130-094-842
Human XCL1 ELISA Kit	Abcam	Cat# ab264620
DNA-free™ DNA Removal Kit	ThermoFisher Scientific	Cat# AM1906
iScript™ cDNA Synthesis Kit	Bio-Rad	Cat# 1708891
Deposited data
16S rRNA sequencing	This paper	NCBI BioProject; PRJNA688219
Single-cell RNA sequencing	This paper	NCBI BioProject; PRJNA688219
RNA-sequencing from tumor biopsies from refractory metastatic melanoma patients before and after treatment with fecal microbiota transplantation (FMT) and anti-PD-1 re-induction	Baruch et al.26	NCBI BioProject ID GSE162436
Experimental models: Cell lines
MC-38 Cell Line	Kerafast	Cat# ENH204-FP
GW1 primary murine NKT cell line	Donado et al.121	N/A
A20 (A20-CD1d antigen-presenting cell-line)	ATCC	Cat# TIB-208
DN3A4-1.2 hybridoma	Brossay et al.134	N/A
Experimental models: Organisms/strains
C57BL/6J	The Jackson Laboratory	Strain#: 000664; RRID: IMSR_JAX:000664
B6.SJL-Ptprca Pepcb/BoyJ (C57BL/6 CD45.1)	The Jackson Laboratory	Strain#: 002014; RRID: IMSR_JAX:002014
C57BL/6-Tg(TcraTcrb)1100Mjb/J (OT-1)	The Jackson Laboratory	Strain#: 003831; RRID: IMSR_JAX:003831
B6.Cg-Tg(CDX2-cre)101Erf/J (Cdx2Cre)	The Jackson Laboratory	Strain#: 009350; RRID: IMSR_JAX:009350
B6.129P2-Apctm1Rsmi/RfoJ (Apcflox)	The Jackson Laboratory	Strain#: 029275; RRID: IMSR_JAX:029275
B6.129S(C)-Batf3tm1Kmm/J (Batf3−/−)	The Jackson Laboratory	Strain#: 013755; RRID: IMSR_JAX:013755
B6(Cg)-Zbtb46tm1(HBEGF)Mnz/J (Zbtb46-DTR)	The Jackson Laboratory	Strain#: 019506; RRID: IMSR_JAX:019506
Oligonucleotides
M.schaedleri_F: TCTCTTCGGGGATGATTAAAC	Gomes-Neto et al.135	N/A
M.schaedleri_R: AACTTTTCCTATATAAACATGCAC	Gomes-Neto et al.135	N/A
M. schaedleri specific probe MCS487: 5’-Cy5-GCCGGGGCTGCTTATACAGGT-3’	Berry et al.136	N/A
M. schaedleri specific probe MCS547: 5’-Cy5-CAGTCACTCCGAACAACGCT-3’	Berry et al.136	N/A
Universal 16S rRNA probe EUB338: 5’-Cy3-GCTGCCTCCCGTAGGAGT-3’	Amann et al.137	N/A
See Table S2 for list primers used mucin expression analysis and for 16S rDNA library construction		N/A
Software and algorithms
FlowJo (v10.10)	FlowJo	RRID: SCR_008520; https://www.flowjo.com
FlowLogic™ V.8	Inivai Technologies	https://www.inivai.com/flowlogic
GraphPad Prism (v10.4.1)	GraphPad Software	RRID: SCR_002798; https://www.graphpad.com
NIS Elements (v4.11)	Nikon Instruments Inc.	RRID: SCR_002776; https://www.microscope.healthcare.nikon.com
Fiji (v1.5)	Schindelin et al.138	RRID:SCR_002285; https://imagej.net/software/fiji/
FastQC (v0.11.5)	Andrews139	RRID:SCR_014583; http://www.bioinformatics.babraham.ac.uk/projects/fastqc/
PEAR	Zhang et al.140	RRID:SCR_003776; http://www.exelixis-lab.org/software.html
VSEARCH	Rognes et al.141	RRID:SCR_024494; https://github.com/torognes/vsearch
QIIME (v1.9)	Caporaso et al.142	RRID:SCR_008249; http://qiime.org/
SortMeRNA	Kopylova et al.143	RRID:SCR_014402; http://bioinfo.lifl.fr/RNA/sortmerna/
phyloseq (v1.30)	McMurdie et al.144	RRID:SCR_013080; http://www.bioconductor.org/packages/2.12/bioc/html/phyloseq.html
MaAsLin2	Mallick et al.145	RRID:SCR_023241; https://bioconductor.org/packages/release/bioc/html/Maaslin2.html
Trimmomatic (v0.39)	Bolger et al.146	RRID:SCR_011848; http://www.usadellab.org/cms/index.php?page=trimmomatic
BWA-MEM (v.0.7.17-r1188)	Li et al.147	RRID:SCR_010910; http://bio-bwa.sourceforge.net/
Samtools (v.1.17)	Li et al.148	RRID:SCR_002105; http://htslib.org/
Picard (v.2.27.5)	McKenna et al.149	RRID:SCR_006525; http://broadinstitute.github.io/picard/
HUMAnN 3 (v3.8)	Beghini et al.150	RRID:SCR_014620; http://huttenhower.sph.harvard.edu/humann
lmerTest R/CRAN package	Kuznetsova et al.151	RRID:SCR_015656; http://CRAN.R-project.org/package=lmerTest
emmeans R/CRAN package	Searle et al.152	RRID:SCR_018734; https://CRAN.R-project.org/package=emmeans
Cell Ranger 3.0.2	10x Genomics	RRID:SCR_017344; https://www.10xgenomics.com/
Seurat v3 R/CRAN package	Stuart et al.153	RRID:SCR_016341; https://satijalab.org/seurat/get_started.html
MAST	Finak et al.154	RRID:SCR_016340; https://bioconductor.org/packages/release/bioc/html/MAST.html
fgsea (v1.16.0) R/Bioconductor package	Korotkevich et al.155	RRID:SCR_020938; https://bioconductor.org/packages/fgsea/
UCSCXenaTools (v1.3.6) R/CRAN package	Wang et al.156	RRID:SCR_018938; http://xena.ucsc.edu/
metacoder R/CRAN package	Foster et al.157	https://github.com/grunwaldlab/metacoder
microbiome-helper	Comeau et al.158	https://github.com/LangilleLab/microbiome_helper
mothur	Schloss et al.159	RRID:SCR_011947; http://www.mothur.org/
AmpliconNoise	Quince at al.160	RRID:SCR_007814; https://mothur.org/wiki/shhh.flows/
NAST	DeSantis et al.161	http://greengenes.lbl.gov/NAST
Single-linkage precluster	Huse et al.162	https://mothur.org/wiki/pre.cluster/
RDP Classifier	Wang et al.163	RRID:SCR_022773; https://sourceforge.net/projects/rdp-classifier/
Prokka	Seeman164	RRID:SCR_014732; https://github.com/tseemann/prokka

Generation of Conditioned medium from Mucispirillum schaedleri cultures

For conditioned medium (CM) assays, the cultures were centrifuged at max rpm for 5 min and the supernatant was filtered through a 0.2 μm filter and kept at −20°C. To heat-treat supernatants of M. schaedleri, samples were incubated for 10 minutes at 100°C. To fractionate organic and aqueous fractions from bacterial supernatants, samples were processed using methanol:chloroform extraction. Briefly, 150 mL of aqueous sample (supernatant sample, or resuspended cell pellet) was mixed with 160 mL of methanol (Sigma-Aldrich), followed by addition of 320 mL of chloroform (Sigma-Aldrich) and a short vortex. Samples were then centrifuge at max rpm for 3 minutes and the organic and aqueous phases were separated into new 1.5 mL tubes. The fractions were dried using a Speed-Vac machine (Eppendorf) and resuspended in 150 mL of fresh culture medium. For mono-colonization experiments, a 5 mL M. schaedleri culture (O.D. 600nm ~ 0.8) was centrifuged at 4,000 rpm for 5 minutes and the bacterial pellet was resuspended in 1 mL mBHI in the anaerobic chamber. Thereafter, each germ-free mouse received 100 μL of the bacterial suspension or sterile mBHI medium.

Bacterial 16S rRNA gene amplicon library generation and sequencing

The procedures in this section were performed in a biological safety cabinet to minimize potential contamination and were based on the Earth Microbiome Project protocol.¹⁷² 50 ng of cecal DNA was used in a PCR reaction with the ThermoFisher Platinum Hot Start PCR Master Mix (cat. no. 13000014) according to the reagent protocol. Forward (10 μM) and reverse (1.3 μM) primers were used (see Table S2 for primer sequences) to amplify the V4 region of the 16S rRNA gene. For each sample, the reverse primer contains a unique 12 bp Golay barcode.¹⁷³ Each sample was amplified in triplicate in a 25 μL reaction volume in 96-well plates. Sterile, molecular biology-grade water and E. coli genomic DNA served as negative and positive controls, respectively. The PCR reaction started with 3 min of 94°C, then 35 cycles of 45 seconds 94°C, 60 seconds 50°C and 90 seconds 72°C, followed by 10 min of 72°C. After amplification, the triplicate reactions were pooled, and amplicons were purified using AMPure magnetic beads. DNA concentration was determined by the dsDNA broad range assay kit (ThermoFisher) using a Qubit machine and a sample of several libraries was run on an agarose gel to visualize the specific amplicon. The libraries were pooled so that the final DNA concentration was 50 ng/μL and each library has an equal abundance. The pooled amplicon library was analyzed on an Agilent 4200 TapeStation system. DNA sequencing was performed on an Illumina MiSeq machine at the bio-polymer core of HMS using the MiSeq V2 kit with 250 bp paired-end reads.

Analysis of bacterial 16S rRNA gene amplicon surveys in mice

16S rRNA gene amplicon survey analysis was conducted according to the suggested standard operating protocol (SOP) by Comeau et al.¹⁵⁸ using the microbiome-helper wrapper. Briefly, fastq files were obtained for each library (median paired-end read count: 79,584 250-bps) and quality of reads was evaluated using FastQC (v0.11.5). Based on the sequence quality report, reads were trimmed using fastx-toolkit to keep only high-confidence base calls and stitched using PEAR.¹⁴⁰ Chimeric reads were filtered using VSEARCH.¹⁴¹ Operational taxonomic units (OTUs) were picked using QIIME (v1.9)¹⁴² with SortMeRNA filtering¹⁴³ and OTUs with fewer than 0.1% of the reads were excluded as low-confidence. Finally, OTU read count data were rarefied to the lowest library size. Analyses of α-diversity and β-diversity (weighted Unifrac PCoA) were performed and visualized using the phyloseq R/Bioconductor (v1.30)¹⁴⁴ and metacoder R/CRAN¹⁵⁷ packages. Differential abundances between the two diets were analyzed using MaAsLin 2¹⁴⁵ with cage as a random covariate. Data were deposited in the NCBI BioProject database under the accession number PRJNA688219.

Analysis of Mucispirillum abundance in human colon biopsies

For measuring the abundance of M. schaedleri in human mucosal-associated bacterial amplicon samples from three colon biopsies datasets,^72–74 16S rRNA gene sequences were analyzed using mothur software suite following 454 and Illumina MiSeq SOPs.¹⁵⁹ Flowgrams were trimmed to a minimum and maximum of 360 and 720 flows, respectively, and were corrected using the PyroNoise algorithm with default parameters.¹⁶⁰ Sequences that had homopolymers less than 8–10 nucleotide bases and no ambiguous bases were aligned against the SILVA database (version 138.1) using the NAST algorithm.^161,174 Illumina sequences that failed to align by study primer coordinates and were less than 100 base pairs in length were discarded. Pyrosequences were filtered by optimizing alignment end coordinates and minimum read lengths at 95% threshold. Sequences were then clustered with more abundant sequences by a maximal difference of three or four nucleotide bases¹⁷⁵ to reduce the effects of sequencing artifacts and screened for chimeras using the vsearch algorithm.¹⁴¹ Quality-filtered sequences were classified against the Greengenes database (version 13.8) using the naive Bayesian classifier with pseudo-bootstrap confidence scores of at least 80% averaged over 1,000 iterations¹⁶³; those that mapped to eukarya, archaea, mitochondria, chloroplasts or unknown kingdoms, were discarded. Final sequences were assigned to Mucispirillum taxa by consensus taxonomy at 80% confidence thresholds.

Immunofluorescence staining and FISH

To detect CD3 and CD8 positive T cells in mouse colon tumors, we performed immunofluorescence staining of mouse colon tissues on formalin-fixed paraffin embedded 5 micrometer sections. Colon sections were deparaffinized, pretreated with 3% hydrogen peroxide in methanol, then with 1 mM EDTA pH 8.0 (at >90°C for 15 min) for antigen retrieval. After blocking in Tris-buffered saline (TBS) 1% BSA with 10% donkey serum, the sections were stained O/N at 4°C with anti-CD3 (1:3400, Abcam ab5690) and anti-CD8 (1:2500, eBioscience 14-0808-80). After washing, we applied Donkey anti-Rat Alexa Fluor 594 IgG (Jackson ImmunoResearch 712-585-153) to detect CD8 and Donkey anti-rabbit-HRP IgG (Jackson ImmunoResearch 711-035-152) followed by TSA Fluorescein reagent (1:2200, NEL741E001KT, PerkinElmer) to reveal CD3 positive cells. We used DAPI (0.5 μg/mL) as nuclear counterstain and Prolong Gold antifade as the mounting medium (P36934, Life Technologies). Images were acquired on a Nikon Eclipse Ti laser scanning microscope with 20X and 60X objectives.

For Mucispirillum spp. detection, mucus layer and goblet cell visualization, unopened distal colons were fixed overnight in methanol-Carnoy’s fixative followed by routine paraffin embedding and sectioning. Sections were deparaffinized, pretreated with 3% hydrogen peroxide in methanol, and blocked with TrueBlack Lipofuscin Autofluorescence Quencher (Biotium 23007) according to manufacturer recommendations. Fluorescence in situ hybridization was performed at 50°C for 90 min in 5% formamide-0.1% SDS-TBS buffer with 2.5 ng/μL of each Mucispirillum genus specific probes MCS487 (5’-Cy5-GCCGGGGCTGCTTATACAGGT-3’) and MCS547 (5’-Cy5-CAGTCACTCCGAACAACGCT-3’),¹³⁶ and 5 ng/μL of a eubacterial 16S RNA sequence specific probe EUB338 (5′-Cy3-GCTGCCTCCCGTAGGAGT-3’). After washing, all the subsequent steps were performed at 4°C. The tissue sections were blocked with 3% Donkey serum in 1% BSA-TBS for 1 h, stained O/N using an anti-Muc2 antiserum (generously provided by Prof. Gunnar C. Hansson¹⁷⁶), and a Donkey anti-rabbit Alexa Fluor 488 IgG (Invitrogen A21206). DAPI was used as nuclear counterstain and ProLong Gold antifade as mounting medium. For accuracy, mucus layer thickness measured by Muc2 intensity was analyzed only in mice that had intact mucus layer around stool pellets. Images were acquired on a Nikon Eclipse NI-U equipped with a 40X objective. Images analysis was performed using NIS Elements (v4.11) or Fiji (v1.5).¹³⁸

Mucispirillum schaedleri binding to mucins

M. schaedleri and mucin labeling and binding was performed as previously described.^13,177 Briefly, colons open longitudinally, fecal matter gently removed, and mucus collected with a cell scraper and weighted. Mucus was dissolved in 50 mM Carb/Bicarb buffer (Na₂CO₃/NaHCO₃) and incubated with AF647-tagged wheat germ agglutinin (WGA) (Thermos Fisher Scientific, W324466) for 15 minutes at 25°C. After labelling, mucus was washed three time with PBS by centrifuging for 5 minutes at 8,000g. M. schaedleri was labeled with 0.1 mg/mL fluorescein isothiocyanate (FITC) (Thermos Fisher Scientific, F1906) for 30 minutes at room temperature in the dark and washed three times in PBS. Labelled mucins and M. schaedleri were incubated for 1h in a 37°C, washed and analyzed on an LSR-II (BD Biosciences) flow cytometer.

Cell suspension preparation and flow cytometry analysis

Tumors were removed and placed in phosphate-buffered solution (PBS) on ice. Tumor tissue was mechanically minced with scissors or blade before being resuspended in 5 mL of digestion medium (RPMI with GlutaMAX, 10% FBS, penicillin/streptomycin, 0.5 mg/mL of dispase enzyme (STEM Cell Technology), 1 mg/mL of collagenase D (Roche) and 50 μg/mL of DNase I (Roche). Tissue suspension was moved to a 50 mL conical tube and put on a rotating wheel for 1h at 37°C. Digested cell suspension was passed through a 40 μm cell-strainer to a new tube containing 5 mL PBS with 5 mM EDTA, centrifuged and resuspended in FACS buffer (PBS with 1 mM EDTA and 2% FBS). Colons were opened longitudinally, debris gently removed, and washed in PBS, 2% FBS (Gibco), 5 mM HEPES, 1 mM DTT (Sigma-Aldrich) for 10 minutes at 4°C. Cells in the epithelial layer were separated by incubating the washed tissue twice in pre-warmed PBS, 2% FBS, 5 mM HEPES, 5 mM EDTA (VWR) and rotated at 37°C for 15 and 10 minutes, respectively. Epithelial cell suspension was obtained by filtering with 100 μL filters. To isolate lamina propria cells, the remaining tissue was washed in PBS and cut into small pieces (1mm) or minced with scissors in digestion medium. Colons were then digested in 10 ml of digestion medium for 30 minutes at 37°C. Single cell suspension was obtained by filtering the digested tissues with 40μm filters and leftover tissue was subjected to a second round of digestion. Both single cell suspensions pooled for analysis. Mesenteric lymph nodes (mLN) were surgically removed, placed in 1 mL of RPMI medium and kept on ice, before being crushed through a 70 μm cell-strainer using a 1 mL syringe plunge.

For IFNγ and granzyme B (GZMB) measurements in CD8⁺ T cells, tumor cells were enriched for CD8⁺ T cells using magnetic assisted cell sorting (MACS) in a two-step process. First, dead cells were depleted using Dead Cell Removal kit (Miltenyi Biotech) and then CD8⁺ T cells were isolated using the mouse CD8α⁺ T cell Isolation kit (Miltenyi Biotech) and incubated overnight in a 96-well plate precoated with 2 μg/mL anti-CD3 antibody (BioLegend) and 5 μg/mL anti-CD28 (BioLegend) in RPMI with GlutaMAX, 10% FBS, penicillin-streptomycin (50 U/mL), 1 mM sodium pyruvate, 55 μM β-mercaptoethanol and 10 mM HEPES. Following overnight stimulation, the cells were treated with Ebioscience^™ Protein Transport Inhibitor Cocktail for 2 h and then fixed, stained for IFNγ and GZMB, and analyzed.

For exclusion of dead cells, cells were incubated with LIVE/DEAD Fixable Yellow Dead Cell Stain Kit (Thermo Fisher) for 15 minutes at room temperature, before washing in FACS buffer. The cells were resuspended in 100 μL of FACS buffer containing 1 μL of anti-CD16/CD32 antibody (BioLegend, cat no 101302) to block FC receptors and placed in 4°C for 10 minutes. After washing with FACS buffer, 100 μL of antibody staining mix in FACS buffer (see key resources table for antibodies listing) were added to each well and incubated at 4°C for 30 minutes, washed in FACS buffer and resuspended in 250 μL of FACS buffer. For intracellular protein staining BioLegend FOXP3 fixation/permeabilization kit (cat #421403) was used according to the manufacturer’s instructions. The samples were analyzed on either an LSR-II or a FACSymphony (BD Biosciences) flow cytometer. Data were analyzed using FlowJo (Tree Star Inc.) and FlowLogic (Inivai Technologies Pty. Ltd.) software.

Analysis of mucin gene expression by RT-qPCR

Epithelial cell suspensions were homogenized in Qiazol (Qiagen) and extracted using Phase Lock Gel tubes (QuantaBio) following manufacturer’s instructions. DNA contamination was removed with DNA-free^™ DNA Removal Kit (ThermoFisher Scientific) and cDNA was synthesized with iScript cDNA synthesis kit (Bio-Rad). Quantitative real-time PCR was performed using KAPA SYBR FAST Universal Master Mix (KAPA Biosystems) and primers shown in Table S2 on a Stratagene Mx3005P machine (Agilent Technologies).

In vitro cDC1 immunological assays

Splenic cDC1 cells were isolated from WT C57BL6/J mice using magnetic associated cell separation (MACS) kit (Pan Dendritic Cell Isolation Kit, mouse, Miltenyi Biotech) according to the manufacturer instructions and plated in a 96-well plate, at a density of 5×10⁴ cells per well. The cells were incubated overnight with sterile mBHI medium or M. schaedleri CM. Cells were washed, stained with antibodies and analyzed using flow cytometry. For CD8⁺ T cell activation experiments, Naive CD8⁺ T cells (CD44⁻/CD62L⁺) were isolated from OT-I mouse spleens. For cross-presentation assays, splenic cDC1 were treated with M. schaedleri CM or sterile medium in the presence of 0.7 mg/mL of ovalbumin protein (OVA, EndoFit) for 12 hours and then washed with PBS, before the naive OT-I CD8 T cells were added at a 1:1 ratio of DC:T cells. For CD8 T cell activation assay, after overnight incubation with the loaded DCs, the cells were treated with Brefeldin A and monensin (both 1:1000, BioLegend) for 2 h, fixed (BD Cytofix) and IFNγ expression was measured using flow cytometry. For CD8 T cell proliferation assay, OT-I CD8 T cells were stained with CFSE (Celltrace, Thermo Fisher) and the co-culture was incubated for 3 days and then the cells were analyzed using flow cytometry to measure the number of cell divisions.

Single cell analysis of mLN cDC1 from cAPC mice

mLN were surgically removed from cAPC mice and enzymatically digested into a single-cell solution following Fletcher et al.¹⁷⁸ mLN CD45⁺CD11c⁺MHCII^highCD64⁻ cells were sorted by flow cytometry at the Broad Institute flow cytometry platform using a Sony SH800S sorter. Following sorting, the cells were spun down and approximately 9,000 single cells per sample were loaded to the Chromium Controller (10x Genomics). scRNA-Seq libraries were generated using the 10X Genomics Chromium Single Cell 3’ Kit v3 and the 10x Chromium Controller (10x Genomics) according to the standard v3 protocol.

The resulting 3’ scRNA-Seq libraries were pooled and sequenced on a HiSeq 3000 (Illumina, R2 read length 98 base pairs). Reads were mapped using Cell Ranger 3.0.2 (10x Genomics) to the mouse transcriptome reference dataset, and transcript-per-million (TPM) was calculated for each gene in each filtered sample. Cells with either <300 or more than 3,500 detected genes or >0.15 mitochondrial fraction were excluded from further analysis. Finally, the resulting expression matrix was filtered to remove genes detected in <3 cells. Cell expression was normalized followed by selection of highly variable features, data scaling and cell clustering. All the above steps were performed using the Seurat v3 R/CRAN package.¹⁵³ We identified differentially expressed genes (p value lower than 0.05 and a |log₂FC|>0.2) using the MAST¹⁵⁴ test implemented in Seurat v3. To evaluate enrichment of gene sets signatures in cDC1, we performed a fast gene set enrichment analysis (fgsea) using the package fgsa v1.16.0¹⁵⁵ R/Bioconductor with the Hallmark Gene Set Pathways¹⁰⁸ as the reference. Sequencing data were deposited in the NCBI BioProject database under the accession number PRJNA688219.

Survival analysis of cDC1 gene signature in human CC bulk transcriptomic datasets

We obtained normalized read count data for 29 DE-cDC1 genes (FPKM) from the COAD and READ projects under The Cancer Genome Atlas (TCGA) Research Network using the XenaHost (“gdcHub”) function implemented in the UCSCXenaTools (v1.3.6) R/CRAN¹⁵⁶ package. See Table S1 for genes and transcriptional signature scores of cDC1 from the mLN of cAPC mice fed lo vs. hi Saa diets. Survival metadata were curated using clinical and molecular phenotype variables available through the latest “gdcHub” filter and the legacy “tcgaHub” host. To evaluate the prognostic significance of cDC1 transcriptional signature found in mouse mLN for CC patients, we computed activation scores of DE-cDC1 genes for patient j as follows:

$${\text{log}}_{10}\left[\frac{\sum _{i\in H}{g}_{ij}}{\sum _{i\in L}{g}_{ij}}\right],$$

where g_ij is the normalized expression of scRNA-Seq cDC1 gene I from either hi Saa or lo Saa diet DEG set, which are denoted by H and L, respectively. These were used to dichotomize individuals into distinct survival outcomes by identifying a cut-point that maximized the hazard ratio through univariate Cox regression analysis. To assess whether risk stratification by transcriptional state score is an independent prognosticator, we tested a list of patient metadata variables—age, gender, MSI status, tumor mutational burden, and TNM clinical stage—by univariate Cox regression and included TNM clinical stage as a significant survival covariate in the multivariate Cox regression analysis for model adjustment.

Measurements of XCL1 secretion from mouse NK and NKT cells and mouse serum

NK and NKT cells were sorted from the mLN of WT mice fed Saa diets using a BD FACSAria IIu cell sorter. Sorted NK and NKT cells were plated in a 96-well plate and incubated overnight with IL-2 (10 ng/mL). The following day, cell media was collected and XCL1 was measured using R&D Systems Mouse XCL1/Lymphotactin DuoSet ELISA kit. For in vitro XCL1 secretion assays, 6×10⁵ cells per well of GW1 primary murine NKT cell line,¹²¹ generously provided by the laboratory of Dr. Michael B. Brenner (Brigham & Women’s Hospital), were plated in a 24-well plate (0.5 mL volume) and were stimulated overnight with 10% v/v of M. schaedleri CM or its fractions (see above). Sterile mBHI medium was used as control. 10 ng/mL of IL-12 (PeproTech) was used as positive control. Cell culture supernatants were collected and XCL1 concentration was determined using R&D Systems Mouse XCL1/Lymphotactin DuoSet ELISA kit or OneStep Mouse XCL1 ELISA kit (Abcam). For measurements of XCL1 in mouse serum, mouse blood was collected into serum separator tubes (BD) tubes, inverted 5 times and allowed to clot for 30min at room temperature. Then, samples were centrifuged for 15minutes at 1300g at 4°C and the serum layer was carefully removed into a new 1.5ml tube without disturbing the buffy coat layer. Serum XCL1 was measured using the OneStep Mouse XCL1 ELISA kit (Abcam).

Measurements of XCL1 secretion from human NKT cells

Human NKT cells were generated using a published protocol.¹⁷⁹ Briefly, buffy coats from human donors’ peripheral blood were obtained from Research Blood Components LLC and separated using differential centrifugation on a Ficoll density gradient. Magnetically Isolated CD14⁻ cells were cultured with 10 U/mL rhIL-2 for 5 days. Then, Vα24⁺ hNKT cells were magnetically isolated and incubated for 2 days with mitomycin C-treated human dendritic cells loaded with α-galactosylceramide (α-GalCer) + 20 U/mL rhIL-2. Then, the cells were harvested and further expanded for 10 days in complete RPMI medium + 20 U/mL rhIL-2, with medium replacement every 2 days. Finally, the purity of Vα24Jα18⁺ CD3⁺ human NKT cells was estimated using flow cytometry (~98%) and cells were frozen in complete 50% RPMI medium + 40% FBS+ 10% DMSO at 5 × 10⁶ cells per vial. For in vitro assays on the effect of M. schaedleri CM on human NKT cells, frozen cells were thawed and cultured in complete RPMI medium + rhIL-2 at a cell density of 2 × 10 per well in a 96-well plate. The cells were treated with either sterile mBHI or M. schaedleri CM at 20% v/v for 16 h and then the culture supernatant was collected. As positive control we used platebound anti-human CD3 (2 μg/mL) and CD28 (5 μg/mL) to stimulate the NKT cells. XCL1 concentration was measured using OneStep human XCL1 ELISA (Abcam) according to the manufacturer’s instructions.

Assessment of NKT cell TCR stimulation by Mucispirillum schaedleri CM

10⁵ A20-CD1d APCs were incubated with M. schaedleri CM or its fractions (see above) overnight at 37°C. Wells were washed and 5 × 10⁴ DN3A4–1.2 NKT cell hybridoma were added. DN3A4–1.2 hybridoma was chosen due to stable TCR expression and hi reactivity to canonical NKT antigen α-GalCer, which was used as positive control. After 24h of co-culture, supernatant was collected and IL-2 production upon TCR engagement was measured in a sandwich ELISA using a pair of rat anti-mouse IL-2 antibodies for capture (cat. #554424) and detection (cat. #554426) and a recombinant IL-2 standard (cat. #550069) (BD Biosciences). For APCs-free assay, 96-well flat bottom plate was coated with 10 μg/mL of soluble mouse CD1d protein for 1h at 37°C, washed 3 times with PBS and nonspecific binding blocked with 10% FBS in PBS for 1h at 37°C. Plate-bound CD1d plate was then incubated with M. schaedleri CM or its fractions (see above) overnight at 37°C. After washing, assay with 10⁴ DN3A4–1.2 NKT cell hybridoma and IL-2 measurement was performed as described above.

Analysis of cCD1 and NKT signatures in human melanomas following fecal microbiota transplantation

Baruch et al.²⁶ reported a clinical trial in which ten patients with anti-PD-1 refractory metastatic melanoma underwent fecal microbiota transplantation (FMT) followed by anti-PD-1 re-induction. FMT donors were patients with metastatic melanoma who achieved a complete response (CR) on anti-PD-1 monotherapy. Three out of the ten patients in the trial responded (R), while seven did not respond (NR) to anti-PD-1 therapy after FMT. Bulk RNA sequencing data of tumor biopsies was analyzed for nine patients (P1, P3-P10; coded as R1, R3-R10 in the original study). For gene count enumeration and normalization, raw fastq files were obtained from the NCBI’s Sequence Read Archive (SRA) and trimmed to remove adapters and low-quality tails using BBMap (v 38.84). Trimmed fastq files were aligned to the GRCh38.p14 human reference genome using STAR (2.7.11b) and gene counts were enumerated using featureCounts (v2.0.2) and normalized between-samples using the DESeq2 R package (v1.44.0). Gene signature scores for each patient were obtained by summing up the normalized counts for the gene list of each respective signature.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis was performed using GraphPad Prism (GraphPad) or R v4.0.2 (when indicated in specific methods section) (Table S1). Tumor growth was analyzed using a for 2-way ANOVA with Šídák’s multiple comparisons test. Comparisons between two groups use the Mann-Whitney U test, while comparisons of multiple groups use Brown-Forsythe ANOVA test with unpaired t test with Welch’s correction. Presence or absence of tumor lesions at each neoplastic stage were statistically evaluated using Chi-square contingency test. Paired analyses in Figures S11A–S11C were analyzed with paired t tests. Correlations in Figures 4G, S11D, and S11E were analyzed using Spearman’s rank correlation tests. Each symbol represents data from an individual data point (mouse or in vitro well). Full statistical analysis for in vitro experiments in Figures 5F, 5G, S10B, and S10C are shown in Table S3. Bar graphs indicate mean ± SEM. Box and whiskers graphs show Min to Max values after Grubbs outlier exclusion. Violin plots show median with full line and quartiles with dashed line.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.immuni.2026.05.001.

Highlights.

• Dietary sulfur amino acids restrict colon cancer progression in mouse models

• Sulfur amino acids thicken mucus and expand Mucispirillum schaedleri

• Mucispirillum schaedleri triggers natural killer T cell XCL1 secretion

• Dendritic cell recruitment by natural killer T cells enhances anti-tumor immunity