Immune and metabolic effects of African heritage diets versus Western diets in men: a randomized controlled trial

Participant characteristics and dietary recall

Seventy-seven healthy adult men (median age = 25.6 years; interquartile range (IQR) = 21–27.2) participated in an open-label, short dietary intervention with 4-week follow-up to assess its longer-term immunometabolic effects. Participants were recruited from 22 April 2021 to 3 July 2021 with the final sampling on 9 August 2021. Baseline characteristics are summarized in Extended Data Table 1. Participants with a body mass index (BMI) of 18–25.9 were eligible for inclusion; however, four participants with a BMI > 26 were inadvertently included. Dietary habits were assessed using three nonconsecutive 24-h recalls, including a festival or weekend day, and are summarized in Supplementary Table 1.

The trial, depicted in Fig. 1a, consisted of three arms: (1) men living in rural areas (n = 23) who habitually consumed a Kilimanjaro heritage-style diet and switched to a Western-style diet for 2 weeks; (2) men living in urban areas (n = 22) who habitually consumed a Western-style diet and switched to a heritage-style diet for 2 weeks; and (3) men who habitually consumed a Western-style diet (n = 22) and who consumed Mbege, a traditional fermented banana beverage, for 1 week. Five participants who maintained their habitual diets were included in the first two arms as controls to assess diet-independent effects (Supplementary Tables 2 and 3). Blood samples were collected at three time points: baseline; after the intervention (immediately after the 2-week dietary intervention or 1-week fermented beverage intervention); and at the follow-up (4 weeks after the intervention).

a, The study was conducted at the Moshi district, Kilimanjaro region, Tanzania. The illustration describes the dietary interventions carried out in this study. After one participant withdrew from the study before the end of the intervention, the first group included young rural-living men (n = 22) who entered the intervention, whose habitual diet mainly consisted of a Kilimanjaro AHD and who changed for 2 weeks to a WD (top). The second group consisted of urban-living young men (n = 22) whose habitual diet mainly consisted of a WD, and who were switched for 2 weeks to a plant-based, fiber-rich heritage-style diet (middle). In both groups, five controls who remained on their habitual diet were also enrolled. The third group consisted of young men (n = 22) whose diet remained unchanged from their habitual WD, but who were supplemented with consumption of a locally brewed fermented banana beverage (Mbege), approximately 1 l per day for 1 week. Below each group, the number of participants sampled at each of the time points (baseline, t0; after the intervention, t1; and at the follow-up, t2), are presented in green, orange and purple filled squares, respectively. b, In all study arms, samples for hemocytometry, whole-blood cytokine responses and RNA sequencing (RNA-seq), as well as plasma proteome and metabolome, were collected at three time points: baseline (t0); on completion of the intervention (after the intervention) (t1) (at week 2 for the dietary intervention and week 1 for the banana beverage intervention); and 4 weeks after completion of the intervention (t2) (‘follow-up’). c, Images of example foods and the fermented beverage provided in the study arms. Schematic in a created using BioRender.com.

In the group of participants switching from AHD to WD, one participant withdrew before the start of the intervention and was excluded from the analyses, while two participants in this arm and one control lacked follow-up samples (CONSORT diagram in Extended Data Fig. 1). Outcomes were assessed based on sample availability and quality control (Fig. 1a,b and Supplementary Table 4), and were adjusted according to age, BMI and activity level (Supplementary Table 5).

Meals were provided three times daily in the Uru Shimbwe Juu village (AHD to WD switch) or Moshi town (WD to AHD switch). Examples and a detailed description of switched diets are shown in Fig. 1c and Extended Data Table 2. Weight monitoring showed a notable increase in the group who switched to a WD (median = 2.6 kg, IQR = 2.0–3.4 kg, P < 0.001; Mann–Whitney U-test).

Multiomics overview of immunometabolic changes

To assess the impact of the intervention on the variance in plasma proteome, cytokine production, plasma metabolome and whole-blood transcriptome, which are primary outcomes of the trial, we conducted variance partition analysis, using the intervention time points as the primary variable, with age, BMI and physical activity as covariates, and participant ID as a random effect (Extended Data Fig. 2). Each intervention induced distinct immune and metabolic adaptations both in the short (after the intervention versus baseline) and long (follow-up versus baseline) term. Switching from an AHD to a WD accounted for over 5% of variance in plasma proteome (25.9%), cytokine production (40%), plasma metabolome (41.3%) and whole-blood transcriptome (19.1%) features, respectively. These effects largely persisted at the follow-up except for plasma proteome variance, which reduced to 9.9%. Conversely, switching from a WD to an AHD mainly influenced plasma proteome and metabolome, explaining over 5% of variance in 27% and 42% of features, respectively. Fermented beverage consumption explained over 5% of variance in plasma proteome (21%), cytokine production (23.3%), plasma metabolome (35.6%) and whole-blood transcriptome (14.5%) features, with sustained variation primarily in cytokine production and whole-blood transcriptome.

Plasma inflammatory and cardiometabolic proteomes

Next, we analyzed changes in circulating proteins related to inflammation and metabolism, a prespecified primary outcome, using the 92-plex Olink inflammatory and cardiometabolic panels (proximity extension assays, Olink Proteomics AB)¹⁵. Relative protein concentrations were analyzed for 162 proteins (89 cardiometabolic, 73 inflammatory; Supplementary Table 6) after excluding proteins with values below the detection limits in more than 25% of samples. Differentially abundant proteins (DAPs) were identified using a linear mixed model (LMM) analysis and annotated to biological processes according to the Human Protein Atlas (v.23.0). Analysis focused on proteins that the time point variable explained over 5% of the variance (Methods and Supplementary Table 6).

Principal component analysis (PCA) of participant-corrected values revealed distinct shifts, which were more pronounced in participants switching from a WD to an AHD (Extended Data Fig. 3a). Participants switching from an AHD to a WD exhibited a significant increase in 26 proteins (19 cardiometabolic and seven inflammation panel) and a decrease in one protein after the intervention compared to baseline (Fig. 2a,b, Supplementary Fig. 1 and Supplementary Table 7). Among the most upregulated proteins were CNDP1, THBS4 (implicated in atherogenesis¹⁶), ANGPTL3 and TWEAK (Fig. 2a,b). Key biological processes linked to DAPs included ‘cell adhesion’ (for example, THBS2, NID1, CDH1), ‘hydrolase activity’ (for example, CNDP1, PAM), ‘apoptosis’ (for example, TWEAK, TRAIL, TGFB1) and ‘protease activity’ (for example, ANGPTL3, uPA, F7) (Extended Data Fig. 3b), suggesting early pathological adaptations to a WD, characterized by systemic inflammation and metabolic dysregulation.

Targeted plasma proteomics was conducted using the inflammatory and cardiometabolic Olink panel. Plasma CRP concentrations were measured using ELISA. Samples were obtained at baseline (t0), after the intervention (t1, week 2 for the dietary intervention and 1 week for the fermented beverage intervention) and 4 weeks later (t2, follow-up). Statistical analysis was carried out for participants who completed the full study, that is n = 20, 22 and 22 in the WD, AHD and fermented beverage intervention groups, respectively. Statistical analysis was carried out using an LMM for each comparison with a two-sided hypothesis test. Fixed effects included time point, baseline BMI, age and physical activity level; participant ID was included as a random effect. a, Triangle plots present DAPs in the (I) WD, (II) AHD and (III) fermented beverage intervention groups. The direction of the arrow (up or down) indicates the direction of change, comparing late versus early time points in each comparison as listed at the bottom of the plot, with the color fill depicting the log₂ paired fold change value for each comparison. b–e, Combined box and line plots with the group trend line indicating the levels of measured parameters at the three time points. Lines are grouped per participant. Protein concentrations are presented as Normalized Protein eXpression (NPX) values for Olink measures (switch to WD (b), switch to AHD (c) and fermented beverage (d)) or log₂-transformed CRP levels (mg ml⁻¹, e). In all box plots, the line defines the median level, the hinges depict the 25th and 75th centiles and the whiskers extend to ±1.5 times the IQR. The asterisks indicate the significance level based on Benjamini–Hochberg-adjusted P values (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

In contrast, participants switching from a WD to an AHD exhibited a significant reduction in 28 proteins (22 inflammation, 6 cardiometabolic) and an increase in 6 proteins after the intervention compared to baseline (Fig. 2a,c, Supplementary Fig. 2 and Supplementary Table 7). These DAPs were linked to ‘cytokines,’ ‘inflammatory response,’ ‘chemotaxis’ and ‘host–virus interactions’ (Extended Data Fig. 3b). At the follow-up, the changes had partially reversed; however, several proteins, including chemokines (CXCL1, CXCL5, CXCL6, CXCL8, CXCL10, CXCL11, CCL4, CCL5, MCP2, MCP4), cytokines (IL-6, OSM, IL-17C, TNFSF14) and others (CASP8, MMP1, CA1, CA3, STAMBP, ST1A1, 4E-BP), remained significantly decreased compared to baseline. CASP8 and ST1A1 exhibited changes opposite to those in the AHD to WD group. In the control group, most DAPs remained unchanged, except for reduced values of CA1 (follow-up versus after the intervention) and CCL5 (follow-up versus baseline) (Extended Data Fig. 3c and Supplementary Table 7).

Consumption of the fermented beverage resulted in a more diverse pattern of protein changes, with plasma concentrations of 29 proteins (27 cardiometabolic, two inflammatory) increasing and three inflammatory proteins decreasing after the intervention compared to baseline (Fig. 2a,d, Supplementary Fig. 3 and Supplementary Table 7). At the follow-up time point, cardiometabolic proteins returned to baseline but inflammatory markers, including CC chemokines (CCL5, CCL4, CCL28), MMP1, IL-17A and OSM, remained lower than baseline. This is consistent with Wastyk et al.⁵, who observed reductions in inflammatory markers with a highly fermented food diet. In the control group, most DAPs remained unchanged, except for a decrease in CCL5 (follow-up versus baseline; Extended Data Fig. 3c and Supplementary Table 7).

To confirm the inflammatory effects, plasma C-reactive protein (CRP) concentrations were measured using a high-sensitivity enzyme-linked immunosorbent assay (ELISA). Samples with values below the detection threshold (24.4% of samples) were assigned the limit of the detection value (0.2 mg l⁻¹; Fig. 2e). Using an LMM analysis as for the Olink data, a significant increase in plasma CRP concentrations was observed in participants switching to a WD (after the intervention versus baseline, P = 0.020), while a trend toward reduced CRP concentrations was observed in participants switching to an AHD (follow-up versus after the intervention, P = 0.059), with no significant changes in the fermented beverage group.

Overall, switching to a Kilimanjaro heritage-style diet resulted in sustained reductions in inflammatory and metabolic plasma proteins. Conversely, switching to a WD increased metabolic proteins and, to a lesser extent, inflammatory proteins, with minimal lasting effects. Fermented beverage consumption induced a rapid shift in cardiometabolic proteins and a sustained reduction in inflammatory proteins.

Cytokine production capacity

Subsequently, we analyzed the effect of the interventions on cytokine production capacity, a prespecified primary outcome. Whole blood was incubated with Candida albicans, Escherichia coli lipopolysaccharide (LPS), Mycobacterium tuberculosis (MTB), Salmonella enteritidis, Staphylococcus aureus and the TLR3 agonist polyinosinic:polycytidylic acid (poly(I:C)) for 48 h. Cytokines (tumor necrosis factor (TNF), interleukin-6 (IL-6), IL-1β, IL-10, interferon-y (IFNγ)) were measured in supernatants using ELISAs, producing 30 cytokine response measures.

LMM analysis revealed that cytokine responses in participants switching from an AHD to a WD declined in 12 and 14 measures after the intervention and at the follow-up, compared to baseline, respectively, while four measures increased after the intervention compared to baseline. Despite elevated plasma inflammatory proteins (Fig. 2a), TNF and IFNγ responses to all stimuli, except poly(I:C), decreased after the intervention (Fig. 3a, Supplementary Fig. 4a and Supplementary Table 8), which is consistent with reduced cytokine production commonly observed in inflammatory conditions such as aging¹⁷, obesity¹⁸, endotoxemia¹⁹ or sepsis²⁰. The largest decline was observed for C. albicans responses, especially in IFNγ, but also in other cytokines (Fig. 3b). IL-1β and IFNγ responses to poly(I:C) initially increased after the intervention, but this effect was not sustained at the follow-up (Fig. 3a and Supplementary Fig. 4a). IL-6 responses to LPS, S. aureus, C. albicans, MTB and S. enteritidis decreased at the follow-up compared to baseline (Fig. 3a).

Whole blood collected at baseline (t0), after the intervention (t1: week 2 for the dietary intervention group and week 1 for the banana beverage intervention group) and 4 weeks later (t2, follow-up), was subjected to 48-h ex vivo stimulation using the following stimuli: C. albicans hyphae (1 × 10⁷ ml⁻¹), E. coli LPS (10 ng ml⁻¹), MTB (5 µg ml⁻¹), S. enteritidis (1 × 10⁶ ml⁻¹), S. aureus (1 × 10⁶ ml⁻¹) and the TLR3 agonist poly(I:C) (50 µg ml⁻¹). Cytokine concentrations were measured in the supernatants at three time points for participants in the WD (n = 21), AHD (n = 22) and fermented beverage (n = 22) study arms using ELISA. Statistical analysis was carried out using an LMM for each comparison with a two-sided hypothesis test. Fixed effects included time point, baseline BMI, age and physical activity level, while participant ID was included as a random effect. a,c,e, Arrow plots depicting statistical comparisons between time points for switch to WD (a), switch to AHD (c) and fermented beverage (e): after the intervention (t1) versus baseline (t0), follow-up (t2) versus after the intervention (t1), and follow-up (t2) versus baseline (t0). The arrows indicate the direction of change compared to the earlier time point in each comparison (up or down); the fill color represents the log₂ paired fold change, indicating the size and direction of change (red to blue, with up or down changes shown by the colors). The asterisks indicate the significance level of the Benjamini–Hochberg-adjusted P value (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). b,d,f, Combined box and line plots with the group trend line indicating concentrations of measured parameters at the three time points (switch to WD (b), switch to AHD (d) and fermented beverage (f)). Lines are grouped per participant. Cytokine production is presented as the log₂ concentrations of IFNγ, IL-10, IL-1β, IL-6 and TNF in the supernatant. In all box plots, the box line defines the median level, the hinges depict the 25th and 75th centiles and the whiskers extend to ±1.5 times the IQR. ^aChange in control group.

Conversely, switching from a WD to an AHD resulted in only few changes (Fig. 3c, Supplementary Fig. 4b and Supplementary Table 8). At the follow-up, TNF, IL-1β, and IFNγ responses to S. aureus and TNF responses to S. enteritidis were decreased compared to both baseline and after the intervention. In contrast, IL-1β responses to MTB showed an increase at the follow-up compared to baseline and after the intervention (Fig. 3c,d). Finally, the fermented beverage group showed increased responses of the anti-inflammatory cytokine IL-10 after the intervention and at the follow-up compared to the baseline, alongside increased TNF, IL-1β, IL-6 and IL-10 responses to C. albicans (Fig. 3e,f, Supplementary Fig. 4c and Supplementary Table 8). In contrast, the transitioning to a WD led to a marked decline in C. albicans responses.

In summary, the interventions had notable effects on cytokine production. Switching to a WD reduced TNF and IFNγ production, while the fermented beverage increased anti-inflammatory IL-10 responses and enhanced C. albicans cytokine responses.

Plasma metabolome

Diet influences plasma metabolome composition²¹, with immunomodulatory effects from food-derived and microbiome-derived metabolites. We analyzed the plasma metabolome, a prespecified primary outcome, using untargeted metabolomics, identifying 1,266 ions and 1,339 molecular formulas, with 935 metabolites annotated in the Human Metabolome Database (HMDB) (Methods and Supplementary Table 9). PCA revealed altered metabolic profiles after the intervention across all arms compared to baseline (Extended Data Fig. 4a). An LMM analysis was applied to detect differentially abundant metabolites (DAMs) (absolute fold change greater than 1.2, P_adj < 0.05), focusing on metabolites where time points accounted for more than 5% of the variance (Supplementary Table 10). DAMs observed in control participants were reported in the downstream analysis (Supplementary Fig. 5 and Supplementary Table 10). Metabolites were classified into 14 chemical groups; pathway enrichment analysis was conducted using the Relational Database of Metabolomics Pathways (RaMP) ( focusing on WikiPathways terms (1,377 pathways, 3,966 metabolites and 14,191 metabolite–pathway associations) to identify metabolites within or influencing specific pathways (Supplementary Table 11)²².

Transitioning from the AHD to WD resulted in 309 DAMs. Among these, 63 metabolites increased and 58 decreased after the intervention compared to the baseline, while 13 increased and 51 decreased at the follow-up compared to the baseline (Fig. 4a). Furthermore, 43 metabolites increased and 81 decreased at the follow-up compared to after the intervention. Top DAMs and their chemical classes are shown in Extended Data Fig. 4b. Pathway enrichment analysis revealed increases in metabolites linked to ‘glucose homeostasis’, the urea cycle (for example, ‘biomarkers for urea cycle disorders’), ‘tRNA aminoacylation’ and ‘tryptophan catabolism’ after the intervention (Fig. 4a and Supplementary Table 11), presumably reflecting the higher animal protein content in the WD, also indicated by increases in metabolites as 1-methylhistidine. Conversely, metabolites linked to the ‘flavan-3-ol’ pathway, a flavonoid subclass, decreased. The ‘omega-3/omega-6 fatty acid synthesis’ pathway was enriched at the follow-up compared to after the intervention and baseline. The omega-3 polyunsaturated fatty acid eicosapentaenoic acid (EPA) significantly decreased after the intervention compared to the baseline, while docosahexaenoic acid (DHA) showed a trend toward decreasing (P = 0.09), with both increasing again at the follow-up (Fig. 4b,c and Supplementary Table 11). Arachidonic acid (ARA) and its pro-inflammatory metabolite leukotriene B4 transiently decreased after the intervention and significantly increased at the follow-up, alongside a rise in 20-carboxy leukotriene B4 after the intervention (Fig. 4b,c). Lastly, dopamine and dopamine sulfate transiently decreased after the intervention, which is consistent with studies linking high-fat diets to hypodopaminergic effects²³.

a, For each dietary arm and time comparison, the top bar plots present the number of DAMs at each time point. An LMM was used for each comparison with a two-sided hypothesis test. Fixed effects included time point, baseline BMI, age and baseline physical activity level; participant ID was included as a random effect. Significance level was set to Benjamini–Hochberg-adjusted P ≤ 0.05 and absolute paired fold change greater than 1.2. The lower dot plots present the top eight significantly enriched WikiPathway metabolic pathways for DAMs using a Fisher’s exact test via the RaMP database. The color indicates the direction of the DAM, up or down in red or blue, respectively. The size of the dots depicts the number of metabolites related to the pathway. Pathways are ordered alphabetically. b,d,f, Heatmaps showing examples of the enriched pathways discussed in the main text for switch to WD (b), switch to AHD (d) and fermented beverage (f), with the fill color indicating the relevant log₂ paired fold change. c,e,g, Respective associated DAMs for switch to WD (c), switch to AHD (e) and fermented beverage (g) are shown in combined box and line plots with the group trend line indicating concentrations of measured parameters at the three time points. Lines are grouped per participant. In all box plots, the line defines the median value, the hinges depict the 25th and 75th centiles and the whiskers extend to ±1.5 times the IQR. ADA, adrenic acid; ARA, arachidonic acid; DHA, docosahexaenoic acid; DPA, docosapentaenoic acid; EPA, eicosapentaenoic acid; LTB4, leukotriene B4; ALA, alpha-linolenic acid; Valerolactone, 5-(hydroxyphenyl)-gamma-valerolactone-O-sulphate.

Switching from a WD to an AHD resulted in 334 DAMs. Among these, 93 metabolites increased and 41 decreased after the intervention, while 74 increased and 24 decreased after the intervention compared to the baseline (Fig. 4a). The top increased DAMs after the intervention compared to the baseline included the fatty acids dodecanoic acid (lauric acid) and cis-4-decenoic acid (Extended Data Fig. 4b). Dodecanoic acid was suggested to improve mitochondrial function and glucose and lipid metabolism²⁴. Additional DAMs that increased included benzoic acid, trans-piceid (a resveratrol glycoside with cardiovascular benefits)²⁵ and fumaric acid (a tricarboxylic acid cycle intermediate and immunomodulator)²⁶. In the controls, with the exception of benzoic acid, these metabolites were not differentially abundant or were only observed as differentially abundant at the follow-up compared to the baseline (Supplementary Fig. 5 and Supplementary Table 10). Among the top DAMs that decreased after the intervention were cresol, p-cresol sulfate and indoxyl sulfate, which returned to baseline levels at the follow-up. The latter two metabolites, produced via tyrosine and tryptophan fermentation by gut bacteria, are known to decrease with vegetarian diets²⁷. Pathway enrichment analysis linked increased DAMs, such as the omega-3 polyunsaturated fatty acids docosahexaenoic acid (DHA), docosapentaenoic acid (DPA) and EPA, as well as arachidonic acid, dodecanoid acid and fumaric acid, to ‘omega-9 fatty acid synthesis’, ‘omega-3/omega-6 fatty acid synthesis’, ‘neurotransmitter release cycle’ and ‘incretin synthesis, secretion and inactivation’ pathways, with varying roles depending on the metabolite (Fig. 4a,d,e and Supplementary Table 11).

Consumption of the fermented beverage resulted in 177 DAMs, with 79 metabolites increasing and seven decreasing after the intervention compared to the baseline, and 29 increasing and nine decreasing at the follow-up compared to the baseline (Fig. 4a and Supplementary Table 10). Notable increases were observed in metabolites of the ‘flavan-3-ol metabolic pathway’, including hippuric acid, hydrocinnamic acid, p-hydroxyphenylacetic acid, pyrocatechol and valerolactone-O-sulfate (Fig. 4f,g, Extended Data Fig. 4b and Supplementary Table 11). These metabolites, formed by gut microbial breakdown of dietary polyphenols, have been associated with vascular health and anti-inflammatory effects²⁸. The sustained decrease in cysteinylglycine (a glutathione breakdown product) and transient increase in tyrosol (a microbial metabolite of polyphenols), and organic sulfates like 4-ethylphenyl sulfate, may highlight microbial fermentation effects (Fig. 4g and Extended Data Fig. 4b)²⁹.

In summary, the dietary switch and fermented beverage both had distinct effects on metabolic pathways related to glucose, lipid, fatty acid and amino acid metabolism, and pathways associated with inflammation.

Whole-blood transcriptome

We also analyzed changes in the prespecified primary outcome of whole-blood transcriptome. We performed differential expression analysis (P_adj < 0.2, no fold change cutoff), considering both the individual and control group of each intervention arm (Methods). PCA of the transcriptome alterations revealed subtle shifts across the first (PC1) and second (PC2) PCs in the participants switching to a WD and those in the fermented beverage group intervention, both after the intervention and at the follow-up, compared to the baseline (values corrected for participant; Extended Data Fig. 5a).

The switch from the AHD to WD resulted in the most differentially expressed genes (DEGs), with 97 upregulated and 26 downregulated after the intervention compared to the baseline, and 410 upregulated and 163 downregulated at the follow-up compared to the baseline (Fig. 5a). In contrast, the switch from WD to AHD had minimal effects, with seven upregulated and seven downregulated genes after the intervention compared to the baseline, and no differences at the follow-up compared to the baseline. Fermented beverage consumption resulted in 26 upregulated and 104 downregulated DEGs after the intervention compared to the baseline and 15 upregulated and 58 downregulated genes at the follow-up compared to the baseline (Fig. 5a and Supplementary Tables 12–14).

Whole-blood transcriptome data were analyzed for participants in the AHD to WD switch group (n = 22), the WD to AHD switch group (n = 22) and the fermented beverage group (n = 21). Blood samples were collected at three time points: baseline (t0), after the intervention (t1) and the follow-up (t2), 4 weeks after the end of the intervention. Five control individuals for each dietary arm were included in the differential expression analysis, using DESeq2. The model design incorporated the individual (ind.n), intervention (diet versus control) and time point (baseline, after the intervention or at the follow-up) as follows: ~intervention + intervention:ind. n + intervention:time point. Differential expression analysis was performed between time points using the Benjamini–Hochberg method for P value adjustment and independent hypothesis weighting (IHW). A two-sided significance threshold was set at adjusted P value ≤ 0.2, with no fold change cutoff. a, A scheme presenting the number of DEGs for each comparison in each of the intervention arms are presented. b,e, Dot plots presenting the results of gene set overrepresentation analyses for GO terms performed with the DEGs from each comparison for the switch to WD (b) and fermented beverage (e). The size and color of the dots depict the gene ratio and Benjamini–Hochberg-adjusted P, respectively. c,d,f,g, Tile plots presenting the log₂ fold change of the genes related to selected enriched GO terms for switch to WD (after intervention (t1) versus baseline (t0) (c) and follow-up (t2) versus baseline (t0) (d)), and the fermented beverage (after intervention (t1) versus baseline (t0) (f) and follow-up (t2) versus baseline (t0) (g)).

In participants switching to a WD, upregulated genes after the intervention were enriched for Gene Ontology (GO) terms related to reactive oxygen species (ROS), including ‘response to ROS’, ‘cellular response to hydrogen peroxide’ and ‘positive regulation of ROS production’, and pathogen defense terms such as ‘response to fungus’ and ‘defense response to bacterium’ (Fig. 5b,c and Supplementary Table 15). Key genes included S100A8, MPO, MMP8 and DEFA3 and DEFA4, all strongly expressed in neutrophils³⁰. In contrast, downregulated DEGs at the follow-up compared to the baseline were related to adaptive immunity, showing enrichment in processes such as ‘immunoglobulin production’, ‘positive regulation of T cell activation’ and ‘MHC class II protein complex assembly’, including several human leukocyte antigen (HLA) genes (Fig. 5b,d and Supplementary Table 15). Upregulated genes at the follow-up compared to the baseline were associated with posttranscriptional modification terms, such as ‘production of miRNA involved in gene silencing’ and ‘histone modification’ (Fig. 5b and Extended Data Fig. 5c).

In the fermented beverage group, downregulated genes after the intervention compared to the baseline were linked to cellular migration processes, including ‘positive regulation of cell migration’ and ‘regulation of cell development’, including genes such as ACTN4, STAT3, DIAPH1, NOTCH1 and NOTCH2, LRP1, JAK1 and SEMA4D (Fig. 5e,f and Supplementary Table 16). Transcription factor enrichment analysis identified STAT3, STAT5B and SON DNA binding protein (an intracellular infection mediator in macrophages)³¹, and ELF4 (a known regulator of IFN induction)³², as key regulators of downregulated DEGs (Extended Data Fig. 5d and Supplementary Table 17). Moreover, aryl hydrocarbon receptor expression encodes a transcription factor responsive to metabolites such as indole-lactic acid and from fermented foods; its targets were increased (Extended Data Fig. 5d and Supplementary Table 17)³³. Comparing the follow-up with the baseline, a decrease in genes related to inflammation progression and resolution pathways was observed. This included pathways such as ‘cell activation involved in immune response’, ‘response to wounding’, ‘acute phase response’ and ‘positive regulation of cellular component movement’, with key genes such as IL6R, DOCK2, DOCK5, DOCK8, DIAPH1 and LRP1 (Fig. 5e,g and Supplementary Table 16).

In summary, these results support that a WD is associated with upregulation in genes of the innate immune system, whereas the fermented banana beverage induced a more anti-inflammatory transcriptional profile.

Safety and adverse events

Possible adverse events were monitored through spontaneous reporting because the risks of the short-term dietary interventions were considered minimal. Participants were instructed to report any adverse events to the study team, but none occurred.

Post-hoc analyses of circulating leukocyte activation

We conducted a post-hoc analysis to assess the intervention’s impact on circulating leukocyte subsets and activation status. Using a Sysmex XN-450 hematology analyzer with its extended inflammation parameters, we measured complete blood counts and activation parameters for lymphocytes (reactive lymphocytes (RE-LYMPHs)), monocytes (reactive monocytes (RE-MONOs)) and neutrophils (neutrophil reactivity intensity (NEUT-RI) and neutrophil granularity intensity (NEUT-GI))³⁴. LMM analysis revealed significant leukocyte changes, predominantly in participants switching to a WD, with smaller effects in participants consuming the fermented beverage (Fig. 6a and Supplementary Table 18). Switching to a WD increased white blood cells, neutrophils, immature granulocytes, NEUT-RI, monocytes, RE-MONOs and RE-LYMPHs from baseline to after the intervention (Fig. 6a,b). While white blood cell counts returned to baseline levels at the follow-up, activation markers (NEUT-RI, NEUT-GI, RE-MONOs, RE-LYMPHs) remained elevated, suggesting persistent immune cell activation. Conversely, switching to an AHD showed no significant leukocyte changes. The fermented beverage reduced NEUT-RI after the intervention, with effects persisting at the follow-up, alongside lower immature granulocytes and higher basophil percentages (Fig. 6a,c). No significant changes were observed in the control participants maintained on their habitual diet (Supplementary Table 18).

a, Triangle plot presenting the significantly different leukocyte parameters across the baseline (t0), after the intervention (t1, 2 weeks for diets and 1 week for fermented beverage) and at the follow-up (t2, 4 weeks after t1) time points (n = 22 from each diet). Parameters were measured using the XN-450 hematology analyzer (Sysmex). The direction of the arrows (up or down) indicates the direction of change, comparing late versus early time points in each comparison as listed at the bottom of the plot. An LMM was used for each comparison, with a two-sided hypothesis test. Fixed effects included time point, baseline BMI, age and baseline physical activity level, while participant ID was included as a random effect. The asterisks indicate the significance level based on Benjamini–Hochberg-adjusted P values (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). b,c, Combined violin, box and line plots depicting the values of the measured parameters (switch to WD (b) and fermented beverage (c)) across the three time points with participant-specific lines linking observations. In all box plots, the central line represents the median, the hinges indicate the 25th and 75th centiles, and the whiskers extend to ±1.5 times the IQR. NEUT-RI and NEUT-GI values are expressed in fluorescence intensity (FI) units and scatter intensity (SI) units, respectively. WBC, white blood cell count.

Overall, the intervention substantially influenced leukocyte numbers and phenotype, with the WD promoting myelopoiesis and immune cell activation, while the fermented beverage reduced neutrophil activation.

Independent validation via cross-sectional cohort

To validate the findings, we analyzed data from the 300 Tanzania functional genomics (TZFG) cohort, a cross-sectional study of urban- or rural-living Tanzanians (n = 295) from the same region. Participants were previously categorized into ‘Kilimanjaro heritage-style diet’ (n = 138) and ‘Western-style diet’ clusters (n = 157) based on food-derived plasma metabolome profiles (Extended Data Fig. 6a)⁹. Differential analysis using the Olink inflammation panel identified 18 proteins that were lower in the heritage-style diet cluster compared to the Western-style diet cluster (Supplementary Table 19). Among these, 83% (15 proteins) also decreased in the current WD to AHD arm (Extended Data Fig. 6b,c and Supplementary Table 19). These proteins included the chemokines CXCL1, CXCL5, CXCL6, CXCL11, MCP2 and MCP4.