Differential peripheral immune signatures elicited by vegan versus ketogenic diets in humans

Abstract

Nutrition has broad impacts on all physiological processes. However, how nutrition affects human immunity remains largely unknown. Here we explored the impact of a dietary intervention on both immunity and the microbiota by performing a post hoc analysis of a clinical trial in which each of the 20 participants sequentially consumed vegan or ketogenic diets for 2 weeks ( NCT03878108 ). Using a multiomics approach including multidimensional flow cytometry, transcriptomic, proteomic, metabolomic and metagenomic datasets, we assessed the impact of each diet, and dietary switch, on host immunity and the microbiota. Our data revealed that overall, a ketogenic diet was associated with a significant upregulation of pathways and enrichment in cells associated with the adaptive immune system. In contrast, a vegan diet had a significant impact on the innate immune system, including upregulation of pathways associated with antiviral immunity. Both diets significantly and differentially impacted the microbiome and host-associated amino acid metabolism, with a strong downregulation of most microbial pathways following ketogenic diet compared with baseline and vegan diet. Despite the diversity of participants, we also observed a tightly connected network between datasets driven by compounds associated with amino acids, lipids and the immune system. Collectively, this work demonstrates that in diverse participants 2 weeks of controlled dietary intervention is sufficient to significantly and divergently impact host immunity, which could have implications for precision nutritional interventions. ClinicalTrials.gov registration: NCT03878108 .

Fig. 1. NK and T cells are significantly affected by change in diet.

a, Schematic of experimental setup. Twenty participants were split into two groups (first group: 4 females (pink), 6 males (blue); second group: 5 females, 5 males), with one group starting on vegan diet for 2 weeks and then immediately changing to ketogenic diet (Group A), whereas the other group started with ketogenic diet and changed to vegan diet (Group B). Data (indicated on the bottom) were collected directly before first diet as baseline, and at the end of the first and second diets. For microbiome samples, data were collected on different days (refer to Extended Data Fig. 1k for more details). b, Frequency of main cell types (as frequency of all CD45⁺ live cells) measured by flow cytometry for baseline, ketogenic and vegan diets shown for each participant. For the gating strategy for flow cytometry, see Extended Data Table 1 and Extended Data Fig. 2. Order of diet listed in this panel is the same for all participants independent of their first diet. Color of individual on top of the plot denotes starting diet (orange, ketogenic diet; blue, vegan diet). c, Fold changes of cell populations whose frequency significantly changed between ketogenic/vegan diet and baseline diet (P value < 0.01) (purple, upregulated in vegan/ketogenic diet; green, upregulated in baseline diet). Dots are scaled by −log₁₀(P value). Significance was calculated by two-sided paired t-test. d, Fold change of cell populations whose frequency significantly changed between ketogenic and vegan diets (P value < 0.01) (purple, upregulated in ketogenic diet; green, upregulated in vegan diet). Dots are scaled by −log₁₀(P value). Significance was calculated by two-sided paired t-test. For gating strategy refer to Extended Data Fig. 2 and ref. . Regulatory T (T_reg) cells, CD127^lowCD25^highCCR4⁺HLA-DR⁺; CD16⁺ NK cells, CD3⁻CD19⁻CD14⁻HLA-DR⁻CD123⁻CD56⁺CD16⁺; activated T helper (T_H) cells, CD3⁺CD19⁻CD4⁺CD8⁻HLA-DR⁺CD38⁺; activated NK cells, CD3⁻CD19⁻CD14⁻HLA-DR⁻CD123⁻CD56⁺CD16^lowCD57^high. BA, baseline; DC, dendritic cells; DN, double negative; Gr, granulocytes; pp, per person; S, sample.

Fig. 2. Ketogenic diet is associated with heightened adaptive immunity and vegan diet with heightened innate immunity.

a, BTM analysis showing enriched pathways for all comparisons noted on the bottom. Dots are scaled by −log₁₀(P value) and colored by network enrichment score (NES). Category names were shortened. Refer to Extended Data Fig. 3i for full names. Significance was calculated and multiple-testing corrected with the fgsea pathway package. b, Hallmark analysis showing enriched pathways for all comparisons noted on the bottom. Dots are scaled by −log₁₀(P value) and colored by NES. Category names were shortened. Refer to Extended Data Fig. 3j for full names. Significance was calculated and multiple-testing corrected with the fgsea pathway package. c, Bar graph showing results from IPA disease term analysis. The x axis shows activation Z-score (comparing ketogenic versus vegan diet). Bars are colored by −log₁₀(P value). Positive Z-score values show disease terms enriched in ketogenic diet, whereas negative Z-scores show disease terms enriched in vegan diet. Significance was calculated using Fisher’s exact test. d, Heat map of gene expression (as row Z-score) from sorted populations from the blood downloaded from the Human Protein Atlas. Depicted genes are members of pathways significantly differentially enriched between ketogenic and vegan diets in BTM analysis. B, baseline; K, ketogenic; V, vegan; P, previous diet.

Fig. 3. Proteomics data show upregulation of adaptive immunity following ketogenic diet.

a, Volcano plot for protein abundance of comparisons noted on top. Proteins that are significantly different (fold change greater than 2, false discovery rate (FDR) < 0.01) are colored purple. Significance was calculated with a paired Wilcoxon signed-rank test with multiple-testing correction. b, Dot plot showing tissue origin of differentially abundant proteins. Dots are scaled by number of proteins. c, Bar graph showing functional enrichment analysis using STRING. Analysis was performed on fold changes of all proteins between ketogenic and vegan (left) and ketogenic and baseline diets (right). Orange bars denote upregulation in ketogenic diet, blue in vegan, gray in baseline diet. d, PCA of proteome data colored by sex. e, Box-and-whisker plot showing Euclidean distance from PCA separated by sex. The lower and upper hinges of the box correspond to the first and third quartiles (the 25th and 75th percentiles); the line in the box indicates the median. The upper whisker extends from the hinge to the largest value no further than 1.5 × interquartile range (IQR) from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR from the hinge (n = 20, 11 males/9 females). Significance was calculated by a paired two-sided t-test. **P < 0.01. GO, gene ontology; FDR, false discovery rate.

Fig. 4. Ketogenic diet significantly alters composition and function of microbiome.

a, PCoA of microbiome data with 95% confidence interval. The data split into two clusters. b, Centroid analysis for beta dispersion plot. c, Beta diversity plot for each individual from PCoA analysis with connection between ketogenic and vegan diets for each participant. Connection lines are colored by starting diet. Significance was calculated with a PERMANOVA test using a marginal model ~Diet + SubjectID. d, Stacked bar graph showing distribution of abundant phyla (>1%) for all individuals following baseline, ketogenic and vegan diets (left). Individuals on top are colored based on the cluster membership of panel a. Significance was calculated using Maaslin2 and P values were adjusted with the qvalue R package. Dot plot shows significance of changes in phylum between diet comparisons (right). Purple dots show significant changes, whereas green dots denote no significance. e, Box-and-whisker plot of fold change of significantly differentially abundant species between ketogenic and vegan diets for all significant taxa (Q value < 0.2). The lower and upper hinges of the box correspond to the first and third quartiles (the 25th and 75th percentiles); the line in the box indicates the median. The upper whisker extends from the hinge to the largest value no further than 1.5 × IQR from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR from the hinge (n = 10). f, Volcano plot showing fold change of abundance of EC numbers for ketogenic diet versus baseline diet (left) and ketogenic diet versus vegan diet (right). Purple dots show enzymes from significantly differently abundant pathways for each comparison. Significance was calculated using Maaslin2 and P values were Bonferroni–Hochberg corrected. g, Lollipop plot showing number of significantly changed pathways from MetaCyc enrichment analysis (left) and changed subpathways for amino acids and vitamin biosynthesis (right) for ketogenic diet versus baseline diet and ketogenic diet versus vegan diet. h, Stacked bar graph showing which genera contribute to pool of all enzymes (left) and all enzymes from significantly differently enriched pathways between ketogenic diet and vegan diet (right). PAMP, pathogen-associated molecular patterns.

Fig. 5. Diets significantly affect host amino acid metabolism.

a, Volcano plot for all metabolites between ketogenic and vegan diets. Metabolites that are significantly different (FDR < 0.01) are colored purple. Significance was calculated by paired two-sided t-test and multiple-testing corrected. b, Bar graph showing all significantly differently enriched MetaboAnalyst pathway results upregulated following ketogenic diet (left) and following vegan diet (right). c, Abundance of all amino acids in ketogenic and vegan diets. The lower and upper hinges of the box correspond to the first and third quartiles (the 25th and 75th percentiles); the line in the box indicates the median. The upper whisker extends from the hinge to the largest value no further than 1.5 × IQR from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR from the hinge (n = 20). Significance was calculated by paired two-sided t-test and multiple-testing corrected. d, Comparison of pathway enrichment between plasma (x axis) and 24-h urine (y axis) samples. Only pathways enriched in both samples are labeled. e, Heat map showing quantity of all significantly differentially abundant lipids per participant (column), with color legend depicting if lipids contain saturated or unsaturated fatty acids. P values: *P < 0.05; **P < 0.01; ***P < 0.001.

Fig. 6. Highly interconnected network between data is driven by immunity, amino acids and lipids.

a, Comparison of pathways significantly differently enriched between ketogenic diet and vegan diet for metabolomics (green) and microbiome data (purple). b,c, Interconnected network from all microbial enzymes, metabolites and proteins with more than ten connections colored by immune category (b) and diet (c). d, Graphical summary of main findings.

Extended Data Fig. 1. Background information about study cohort and diets.

(a–d) Distribution of different characteristics in the study population. Plots showing race (a), gender (b), body mass index (BMI) (c), and age (d). (e) Amount of dietary fiber and dietary sugar in ketogenic and vegan diet in grams per 1000 kcal. (f, g) Box-whisker plot showing amounts of consumed components of diets in percentage (f) and total grams (g). (h) Amount of consumed types of fatty acid in diets in grams. (i) Amount of consumed amino acids in diets in grams. Significance was calculated by two-sided paired t-test with multiple testing correction. (j) Percentage of energy intake from different nutrients at baseline based on food questionnaire analysis. (k) Schematic showing when microbiome data was collected. (l) Frequency of CD45⁺ live cells in flow cytometry data. (m) Bar graph of frequencies of all cell types significantly different between baseline and diet (from Fig. 1c). Each dot represents one individual. Dots are colored by starting diet (blue: vegan, orange: ketogenic). (n) Bar graph of frequencies of all cell types significantly different between ketogenic and vegan diet (from Fig. 1d). Each dot represents one individual. Dots are colored by starting diet (blue: vegan, orange: ketogenic). Data is represented as box-whisker plots in f-j. The lower and upper hinges of the box correspond to the first and third quartiles (the 25th and 75th percentiles), the line in the box indicate the median. The upper whisker extends from the hinge to the largest value no further than 1.5 × interquartile range (IQR) from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR of the hinge mean (n = 20). Data is represented as bar graphs in m,n. Error bars show standard deviation (n = 7). P-values: * < 0.05; ** < 0.01; *** < 0.001; **** < 0.0001.

Extended Data Fig. 2

Gating strategy for high-dimensional flow cytometry data.

Extended Data Fig. 3. Supporting data for RNA-seq analysis.

(a) Heat map of all expressed genes (TPM > 32) in all participants and diets. (b) Principal component analysis (PCA) of RNA-seq data. (c) Heat map showing log 2 foldchange of all genes from significant innate immunity pathways (BTM analysis). Participant number is denoted on top of heat map and colored by first diet (blue: vegan, orange: ketogenic). (d) Heat map showing log 2 foldchange of all genes from significant adaptive immunity pathways (BTM analysis). Participant number is denoted on top of heat map and colored by first diet (blue: vegan, orange: ketogenic). (e) Heat map of all expressed endogenous retroviruses (ERVs) (TPM > 4). (f) Mean total daily iron intake in grams in ketogenic and vegan diet. Data is represented as box-whisker plot. The lower and upper hinges of the box correspond to the first and third quartiles (the 25th and 75th percentiles), the line in the box indicate the median. The upper whisker extends from the hinge to the largest value no further than 1.5 × interquartile range (IQR) from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR of the hinge (n = 20). Significance was calculated by a two-sided paired t-test. (g) Bar graph showing the number of pathways significantly enriched in IPA disease term analysis comparing ketogenic diet with vegan diet. Pie chart shows proportion of cancer pathways with stronger activation following ketogenic diet (orange) and vegan diet (blue). (h) Heat map of gene expression (as row Z-score) from sorted populations from the blood downloaded from the Human Protein Atlas. Depicted genes are members of pathways significantly differentially enriched between ketogenic and vegan diet in HALLMARK analysis. (i) Table of BTM long names – referred to Fig. 2a (j) Table of HALLMARK long names – referred to Fig. 2b. P-values: *** < 0.001.

Extended Data Fig. 4. Supporting data for proteomics data.

(a-c) Heat map of log2 fold change with participants colored by starting diet (blue: vegan, orange: ketogenic) for all significant differentially abundant proteins for vegan diet versus baseline diet (a), ketogenic diet versus baseline diet (b), and ketogenic diet versus vegan diet (c). (d) Venn diagram showing overlap of significantly differentially abundant proteins from complete data set, data from group A (starting diet: vegan), and group B (starting diet: ketogenic) for ketogenic diet versus baseline diet (left) and ketogenic diet versus vegan diet (right). There were no significantly differentially abundant proteins for baseline diet versus vegan diet in group A and group B. (e) Proteins that are significantly differentially abundant between female and male participants. Significance was calculated by Student’s t-test with multiple testing correction.

Extended Data Fig. 5. Supporting data for microbiome data.

(a) Stacked bar graph showing differences between clusters from Fig. 4a which is driven by differences in abundance of Prevotella. Color on top denotes which cluster the data was from in Fig. 4a. S: Sample (b) Alpha diversity measured by Chao richness (left) and Shannon diversity (right). (c) Alpha diversity measured by Chao richness (left) and Shannon diversity (right) divided by group per diet. (d) Box-whisker plots showing overall health index (HEI) score and intake of fiber at baseline diet based on food questionnaire. (e) Volcano plot of microbial enzymes colored by source of polysaccharide degradation. Data is represented as box-whisker plots in b - d. The lower and upper hinges of the box correspond to the first and third quartiles (the 25^th and 75^th percentiles), the line in the box indicate the median. The upper whisker extends from the hinge to the largest value no further than 1.5 × interquartile range (IQR) from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR of the hinge (n = 10 for b, c; in d: n = 8 for Prevotellaceae low, n = 2 for Prevotellaceae high).

Extended Data Fig. 6. Supporting data for metabolomics data.

(a, b) Principal component analysis (PCA) for plasma metabolomics data showing PC1 (explaining 10.1% of variation) and PC2 (explaining 7.7% of variation) colored by diet (a) and sex (b). (c) Box-whisker plot showing Euclidean distance from PCA separated by sex. The lower and upper hinges of the box correspond to the first and third quartiles (the 25^th and 75^th percentiles), the line in the box indicate the median. The upper whisker extends from the hinge to the largest value no further than 1.5 × interquartile range (IQR) from the hinge and the lower whisker extends from the hinge to the smallest value at most 1.5 × IQR of the hinge (n = 20 – 11 males/9 females). (d) Volcano plot for metabolites comparing vegan diet versus baseline diet (left) and ketogenic diet versus baseline diet (right). Significance was calculated with paired two-sided t-test with multiple testing correction. (e) Bar graph showing number of total metabolites per category (left) and percentage of significantly changed metabolites per category (right). (f) Venn diagram showing overlap of differentially abundant metabolites for participants for group A (start diet: vegan), group B (start diet: ketogenic) and complete data set. (g) MetaboAnalyst analysis for 24 h urine samples.

Extended Data Fig. 7. Supporting data for network analysis.

(a–c) Correlation heat maps showing correlations of all metabolites and microbial enzymes (a), metabolites and proteins (b), and microbial enzymes and proteins (c). (d–f) Histogram of number of significant correlations for all metabolites and microbial enzymes (d), metabolites and proteins (e), and microbial enzymes and proteins (f). (g, h) Network of highly connected data points colored by significance (g) and categories (h).