Ketogenic diet suppresses colorectal cancer through the gut microbiome long chain fatty acid stearate

Abstract

Colorectal cancer (CRC) patients have been shown to possess an altered gut microbiome. Diet is a well-established modulator of the microbiome, and thus, dietary interventions might have a beneficial effect on CRC. An attenuating effect of the ketogenic diet (KD) on CRC cell growth has been previously observed, however the role of the gut microbiome in driving this effect remains unknown. Here, we describe a reduced colonic tumor burden upon KD consumption in a CRC mouse model with a humanized microbiome. Importantly, we demonstrate a causal relationship through microbiome transplantation into germ-free mice, whereby alterations in the gut microbiota were maintained in the absence of continued selective pressure from the KD. Specifically, we identify a shift toward bacterial species that produce stearic acid in ketogenic conditions, whereas consumers were depleted, resulting in elevated levels of free stearate in the gut lumen. This microbial product demonstrates tumor-suppressing properties by inducing apoptosis in cancer cells and decreasing colonic Th17 immune cell populations. Taken together, the beneficial effects of the KD are mediated through alterations in the gut microbiome, including, among others, increased stearic acid production, which in turn significantly reduces intestinal tumor growth.

Fig. 1. Ketogenic diet leads to lower colonic tumor incidence and smaller tumor size in an inflammatory model of CRC.

a Schematic representation of the dietary AOM/DSS experimental setup in the GF (top) and SPF (bottom) facilities (Created in BioRender. Rodriguez, F. (2024) BioRender.com/w69g496). b Compositions of the KD and SD, with % of calories from the indicated source. c Colonic tumor burden in GF KD- and SD-fed mice. d–f Colonic tumor burden (d), normalized average colonic tumor surface area per mouse (e) and tumor surface area distribution (small=first tertile, medium=second tertile, big=third tertile; one-sided right-tailed Chi-squared test χ2 (2, N = 87) = 3.4, p = 0.184; f) in SPF KD- and SD-fed mice. g Representative images of colons from one KD- and one SD-fed mouse in the SPF facility. Scale bar = 1 mm. Data is shown as mean ± SD in (c), (d) and (e). Data in (c) shows n = 6 mice per condition from one experiment. Data in (d), (e) and (f) shows n = 8 and 9 mice in the KD-fed condition and the SD-fed conditions respectively, pooled from two independent experiments in the SPF facility. Analysis of deviance of a negative-binomial generalized linear model regressing the number of tumors against treatment in (c), (d). Two-sided Mann-Whitney U test in (e). Source data are provided as a Source Data file.

Fig. 2. The tumor-inhibitory effect of ketogenic diet is mediated by the gut microbiome.

a Schematic representation of the cecal microbial transfer experimental setup in the GF facility (Created in BioRender. Rodriguez, F. (2024) BioRender.com/w69g496). b–e Colonic tumor burden (b), average colonic tumor surface area per mouse (in pixels) (c) and tumor surface area distribution (small=first tertile, medium=second tertile, big=third tertile; one-sided right-tailed Chi-squared test χ2 (2, N = 384) = 10.3, p = 0.00586; d in KC and SC recipient mice. n = 16 KC recipient mice and n = 13 SC recipient mice. Cecal matter from n = 8 different donor mice per condition was used for CMT. Data shown as mean ± SD in (b) and (c). Analysis of deviance of a negative-binomial generalized linear model regressing the number of tumors against treatment in (b). Two-sided Mann Whitney U test in (c). e Representative images of the distal portion of the colon from one KC and SC recipient mouse. Scale bar = 1 mm. Source data are provided as a Source Data file.

Fig. 3. Ketogenic diet alters gut microbiome composition.

a-f β-diversity as assessed by PCoA (% confidence interval = 95) of Bray-Curtis distance in five individual healthy donor stool samples, in a mix of the five healthy donor stool samples (5HD mix) and in murine fecal samples three days after gavage (TG, a and b); β-diversity as assessed by PCoA (% confidence interval = 95) of Bray-Curtis distance in murine fecal samples collected over time after diet change (c and d); Abundance of selected genera which are differentially abundant at at least one time point (using a generalized linear model on CLR, e and f data is shown as mean ± SEM) in fecal samples collected over time after diet change from SD- and KD-fed mice in the GF (a, c and e) and SPF (b, d, and f) dietary experiments (T0 at diet change, T1-T4 weeks after diet change), as analyzed by 16S rRNA sequencing. g–i β-diversity as assessed by NMDS (% confidence interval = 95) based on an Aitchison distance matrix (g), relative abundance of bacterial families based on Kraken2 IMP output (h), and highly characteristic bacterial MAGs (detected in at least six mice in one condition but not more than three mice in the other (diet experiment); detected in at least four mice in one condition but not more than three mice in the other (CMT experiment)), detected in KD, SD, KC and SC murine fecal samples at endpoint, as analyzed by WGS (i). In (i) Rikenellaceae are highlighted in light blue and Muribaculaceae in dark blue. Data shows n = 1 5HD mix, n = 5 individual human donor samples in (a) and (b), n = 15 humanized mouse samples in (a) and n = 17 humanized mouse samples in (b), n = 6 mice per condition in (c) and (e), n = 8 SPF KD-fed mice and n = 9 SPF SD-fed mice, pooled from two independent experiments in (b), (d), (f), (g), (h) and (i), and n = 8 KC and n = 8 SC mice from one experiment in (g), (h) and (i). Source data are provided as a Source Data file.

Fig. 4. Ketogenic diet alters metabolic profiles, leading to elevated levels of fecal stearate.

a, b PCA (% confidence interval = 95) of identified metabolites from fecal samples from SPF KD- and SD-fed mice (a) and from KC and SC recipient mice (b), as analyzed by united GC-MS and LC-MS. c, d Scaled free stearic acid levels detected in fecal samples from SPF KD- and SD-fed mice (c) and from KC and SC recipient mice (d) by LC-MS (polar phase). e Correlation of colonic tumor burden with free fecal stearic acid levels in SPF KD- and SD-fed mice (dashed line) and in KC and SC recipient mice (dotted line). Two-sided Pearson correlation testing was used. Data shown in (a), (c), and (e, dashed line) are from n = 9 SPF SD-fed and n = 8 SPF KD-fed mice, pooled from two independent experiments. Data shown in (b), (d) and  (e, dotted line) are from n = 8 mice per condition. Data shown as mean ± SD, two-sided Mann-Whitney U test in (c) and (d). Source data are provided as a Source Data file.

Fig. 5. Stearate is a microbial metabolite.

a Schematic representation of type II bacterial fatty acid synthesis created in BioRender. Rodriguez, F. (2024) BioRender.com/w69g496. Core enzymes are shown with their corresponding KEGG orthologs (KOs). Abbreviations: acetyl-CoA carboxylase enzyme complex (AccABCD), β-ketoacyl-ACP synthase 1 (FabB), malonyl-CoA:ACP transacylase (FabD), β-ketoacyl-ACP synthase II (FabF), β-ketoacyl-ACP reductase (FabG), β-ketoacyl-ACP synthase III (FabH), enoyl-ACP reductase enzyme isoforms (FabI, FabK, FabL and FabV), β-hydroxyacyl-ACP dehydratase (FabZ), long-chain-fatty-acid-CoA ligase (FadD), long-chain fatty acid transport protein (FadL), fatty acid kinase (FakA, FakB), acyl-ACP thioesterase (FatA), iron transport protein (FatB), thioesterase 1/protease 1/lysophospholipase L1 (TesA), phosphate acyltransferase (PlsX), Coenzyme A (CoA), acyl carrier protein (ACP). Thioesterases of interest are bolded. b Contigs from metagenomic sequencing data carrying the KOs of the bacterial acyl-CoA thioesterases FatA and TesA were assigned to bacterial classes using Kraken2. Relative proportions (log10-transformed coverage of contigs) of relevant bacterial classes are displayed. Remaining bacterial classes were summarized as other. c Levels of stearic acid measured in bacterial cultures (in relation to sterile growth medium, shown as log2) of the Sonnenburg dataset published by Han et al. Families of the class Clostridia are listed in red, while families of the class Bacteroidia are listed in blue. d Relative abundance (centered-log transformed total read counts) of stearate-producing and -consuming families of interest (bolded in c) detected in murine fecal samples from KD-, SD-fed, KC and SC recipient mice at endpoint. Boxplots show medians with 1^st and 3^rd quantiles. The whiskers from the hinges to the smallest/largest values represent 1.5*inter-quartile range (IQR). Two-sided Wilcoxon signed-rank test with Bonferroni correction. Data shown in (d) are from n = 9 SPF SD-fed and n = 8 SPF KD-fed mice, pooled from two independent experiments (red and black) and from n = 8 mice per SC/KC condition (gray and orange). Source data are provided as a Source Data file.

Fig. 6. The microbial metabolite stearate exhibits anti-cancer effects.

a Schematic representation of the dietary stearic acid supplementation experimental setup (Created in BioRender. Rodriguez, F. (2024) BioRender.com/w69g496). b Compositions of the control (CD) and stearate-supplemented (SSD) diets, with % of calories from the indicated source. c, d Stearic acid concentrations in serum (µg/µL, c) and feces (µg/mg, d) from CD- and SSD-fed mice, as detected by LC-MS-based long chain fatty acid quantification. e Colonic tumor burden in CD and SSD mice. f Representative images of colons from one CD and SSD mouse. Scale bar = 1 mm. Data shows mean ± SD from n = 6 CD and n = 5 SSD mice in (c), from n = 8 CD and n = 8 SSD mice in (d). Two-sided Mann-Whitney U test in (c) and (d). Data shows mean ± SD from n = 8 CD and n = 8 SSD mice in (e). Analysis of deviance of a negative-binomial generalized linear model. g, h HT29 proliferation over time (g) or scaled confluence at 72 hours (h) after treatment with stearic acid or corresponding control. Data shows mean (g) or mean ± SD (h) from n = 3 independent experiments, indicated by different datapoint shapes (h), each with eight technical replicates per condition. i ATP measurement (CellTiterGlo) of HT-29 cells at 72 hours after fatty acid or control treatment (200 µM). Data shows mean ± SD of n = 3 independent experiments (indicated by different datapoint shapes), each with eight technical replicates. Two-way ANOVA with multiple comparisons in (h), (i). j Principal component analysis of fatty-acid-treated HT-29 cells (200 µM). k Hallmark gene sets in stearic acid-treated versus control cells. l Cell cycle and apoptosis related gene expression in control and fatty-acid-treated HT-29 cells and in tumor and matching normal tissue samples. m, n Caspase-3 activity in HT-29 (m) and HCT-116 (n) cells treated with stearic acid ± broad caspase inhibitor (zVad). Staurosporine 2uM was used as a positive control. Data shown in (j), (k), (l) is from n = 3 independent experiments. Data shows mean ± SD of n = 3 biological replicates (indicated by different datapoint shapes), in (m) and (n). Ordinary one-way ANOVA in (m) and (n), data passed Shapiro-Wilk normality test. Source data are provided as a Source Data file.

Fig. 7. Functional ketogenic-diet-induced changes in the gut microbiome are responsible for its anti-cancer properties.

Ketogenic diet consumption leads to lasting functional changes in the gut microbiome, which are sufficient to reduce colonic tumor burden and which impact the luminal long-chain fatty acid pool. The enriched microbial product stearate demonstrates tumor-suppressing properties by inducing apoptosis in cancer cells and decreasing colonic Th17 immune cell populations. Created in BioRender. Rodriguez, F. (2024) BioRender.com/w69g496.