Circulating metabolites associated with protection against Mycobacterium tuberculosis infection and inhibition of mycobacterial growth Authors: Todia P Setiabudiawan, Lika Apriani, Julian Avila-Pacheco, Edwin Ardiansyah, Vinod Kumar, Bachti Alisjahbana, Ayesha J Verrall, Andrew R DiNardo, Agnes R Indrati, Vera P Mourits, L Charlotte J de Bree, Simone J C F M Moorlag, Clary B Clish, Leo A B Joosten, Arjan van Laarhoven, Krista E van Meijgaarden, Mihai G Netea, Simone A Joosten, Valerie A C M Koeken, Philip C Hill, Reinout van Crevel Published: 2026 ABSTRACT AbstractA significant proportion of individuals who are heavily exposed to infectious tuberculosis patients do not acquire Mycobacterium tuberculosis (Mtb) infection, as detected by an interferon gamma release assay (IGRA). We examined circulating metabolite profiles and metabolic genotypes in 199 heavily exposed IGRA-negative tuberculosis household contacts in Indonesia. Based on differentially abundant metabolites, activity of several pathways including arachidonic acid, arginine and proline, glutathione, and tryptophan metabolism, correlated with a negative IGRA at three months. SNPs near PRODH (involved in arginine and proline metabolism) were associated with circulating proline concentrations and persistently negative IGRA results, while SNPstudys near AFMID (involved in tryptophan metabolism) were associated with IGRA conversion. For further validation, plasma metabolomic profiles were correlated with mycobacterial growth inhibition by peripheral blood mononuclear cells from individuals in a low-incidence setting. Lower circulating concentrations of six metabolites, including leukotriene B4, proline, glycine, and tryptophan correlated with better growth control in non-exposed healthy individuals, as well as with stronger protection against IGRA-conversion in the Indonesian tuberculosis household contacts. Collectively, these data support the notion that circulating metabolites may impact innate host defense against Mtb infection, and that metabolic interventions may prevent tuberculosis infection and disease. Subject terms: Metabolomics, Infection, Gene regulation, Tuberculosis Introduction A significant minority of individuals who are heavily exposed to an infectious tuberculosis patient do not develop evidence of an antigen-specific T-cell response, as measured with a tuberculin skin test (TST) or interferon-gamma release assay (IGRA)1–4. These individuals who remain IGRA-negative despite heavy exposure might have a protective innate host response, which clears inhaled Mycobacterium tuberculosis (Mtb) before adaptive immunity is engaged5. We have studied this phenotype, called “early clearance”6, in the context of a relatively well-defined exposure in the household of an infectious tuberculosis followed for 3 months, and identified clinical-microbiological4,7 and innate immune correlates8. As cellular metabolism is important for immune cell function and protection against infectious diseases, metabolic studies may increase our understanding of protection against Mtb infection. In the context of tuberculosis, different metabolic processes influence the outcome of the interaction between the host and Mtb9. For instance, glycolysis is upregulated in macrophages and dendritic cells upon Mtb infection and is necessary for protective immune responses, while downregulation of the tricarboxylic acid (TCA) cycle promotes clearance of Mtb and modulates inflammation10,11. Interestingly, Mtb downregulates glycolysis in macrophages12. Also, changes in lipid metabolism of Mtb-infected macrophages lead to the formation of lipid droplets, which promote intracellular Mtb growth13,14. Other examples include the effects of prostaglandin, eicosanoid and tryptophan metabolism linked to Mtb killing, inflammation and host immunity15–17. Besides intracellular metabolism, circulating metabolites derived from the gut microbiome, diet, tissues, and non-immune cells, such as hepatocytes, adipocytes, and myocytes, can also influence immune cell function18,19 and affect the outcome of host–pathogen interactions20–22. While most mechanistic studies have focused on intracellular metabolic changes in immune cells, the role of circulating metabolites as modulators of immune responses against Mtb remains largely unexplored. While studies have examined metabolite profiles in TB progressors versus non-progressors23, only two small studies have examined associations of metabolomic profiles with TST/IGRA conversion among tuberculosis household contacts. One study compared 10 household contacts with and without IGRA conversion in India and reported differences in plasma metabolomic profiles, some of which correlated with in vitro antimicrobial responses24. The other one, with 38 TST-converters and 40 TST-nonconverters in the Gambia, showed a differential abundance of circulating metabolites related to inflammation, fatty acid metabolism, and bile acid metabolism25. To our knowledge, no other (larger) metabolomic studies have been conducted in tuberculosis household contacts. Also, variants in genes encoding metabolic enzymes may affect metabolism and host immunity, but no studies have specifically correlated variation in metabolic genes with the risk of infection in tuberculosis contacts. Therefore, in the present study, we characterized circulating metabolites and genotypes associated with protection against Mtb infection in a large prospective and well-characterized cohort of household contacts of people affected by tuberculosis in Indonesia. For validation, we correlated circulating metabolites with mycobacterial growth inhibition by peripheral blood mononuclear cells from healthy individuals in a low-incidence setting. Results Tuberculosis household study design and subjects We followed tuberculosis household contacts for 14 weeks to study IGRA conversion in the context of a relatively well-defined exposure to an infectious patient in that household6 (Fig. 1A). Among 1347 household contacts of adults with sputum smear-positive tuberculosis, after exclusion of individuals with active tuberculosis, 433 had a negative IGRA result at baseline. Of these, 317 still had a negative IGRA result after 3 months, while 117 showed IGRA conversion. For the metabolomics study, we randomly selected 199 baseline IGRA-negative household contacts, including 71 IGRA converters (35%) and 128 (65% with repeatedly negative IGRA result (Fig. 1A). IGRA conversion was associated with higher exposure (e.g., sputum smear grade of the index patient, Table 1) in this subgroup. The presence of a BCG scar correlated with reduced risk of IGRA conversion (aRR 0.56 [95% CI, 0.40–0.77]; P < 0.001) in the whole group4, but not significantly in the subgroup selected for this study (Table 1). To strengthen the phenotypes, we used stricter IGRA cut-offs than the manufacturer’s (0.35 IU/mL) based on previous studies26,27; <0.15 IU/mL at baseline and follow-up for a persistently negative IGRA status, and from <0.15 IU/mL at baseline to >0.7 IU/mL for IGRA conversion at 14 weeks (Fig. 1B). Using these IGRA cut-offs, differences between IGRA converters (N = 34) and persistently IGRA-negative individuals (N = 95) in the level of exposure to the index patient (e.g. the sputum bacterial load), and in the proportion of individuals with a BCG scar were more pronounced (Table 1, right column). Study overview. A Graphical representation of the Indonesian household contact study. Household contacts with a negative IGRA at baseline (grey) were classified as either IGRA converters (red) or persistently IGRA-negative individuals (blue) based on a repeat IGRAs at 14 weeks, with stricter IGRA cut-offs (B) to strengthen the phenotypes. C We measured circulating metabolites at baseline and identified metabolites and metabolic pathways associated with IGRA status at follow-up. Next, we searched for variation in metabolic genes of pathways associated with IGRA status at follow-up. Then, we correlated circulating metabolite levels among nonexposed Dutch adults with mycobacterial growth inhibition in PBMCs from these individuals, and searched for metabolites that both correlated with IGRA status in Indonesian household contacts and with mycobacterial growth inhibition in nonexposed Dutch adults. Table 1Clinical characteristics of Indonesian tuberculosis household contacts with metabolomics dataAllStrict IGRA criteria*IGRA convertersPersistently IGRA-negative individualsP valueIGRA convertersPersistently IGRA-negative individualsP valueN711283495Case contact characteristics Sex (male %)35 (49.3)54 (42.2)0.4114 (41.2)41 (43.2)1 Age (median [IQR])21 [12–37]22 [12–42]0.8819 [10–33]20 [11–35]0.68 Presence of BCG scar (%)52 (73.2)100 (78.7)0.4823 (67.6)77 (81.9)0.4 Smoking (%)0.260.11 Never46 (65.7)98 (76.6)23 (67.6)78 (82.1) Quit >6 months ago3 (4.3)4 (3.1)2 (5.9)1 (1.1) Current smoker21 (30.0)26 (20.3)9 (26.5)16 (16.8) BMI (median [IQR])20 [17–24]19 [16–24]0.3519 [16–22]19 [16–23]0.89 Diabetes mellitus (%)4 (5.6)2 (1.6)0.241 (2.9)1 (1.1)1Exposure to the index case Exposure scorea (median [IQR])0.32 [0.25– 0.36]0.24 [0.18–0.32]<0.0010.32 [0.27– 0.38]0.24 [0.18–0.30]<0.001 Sleep in the same room as the index case (%)16 (22.5)31 (24.2)0.936 (17.6)20 (21.1)0.86 Hours spent with index case (median [IQR])6 [3–10]4 [1–10]0.195 [3–10]4 [2–9]0.35 Index case highest smear grade (%)0.040.005 Scanty1 (1.4)11 (8.6)0 (0.0)7 (7.4) +118 (25.4)42 (32.8)5 (14.7)33 (34.7) +215 (21.1)32 (25.0)7 (20.6)25 (26.3) +337 (52.1)43 (33.6)22 (64.7)30 (31.6) Presence of cavities on chest X-ray (%)33 (46.5)40 (31.2)0.0515 (44.1)26 (27.4)0.11 Extent X-ray abnormalities (median [IQR])0.40 [0.25– 0.50]0.35 [0.20–0.51]0.460.35 [0.25– 0.45]0.35 [0.25–0.50]0.72 Mtb strain of index caseb (%)0.570.17 L2 East-Asian (Beijing)23 (34.3)31 (29.0)14 (42.4)21 (26.9) Other genotypes44 (65.6)76 (71.0)19 (57.6)57 (73.1)Blood count parameters at baseline Hemoglobin g/dL (median [IQR])14.0 [13.0– 15.3]13.3 [12.7–14.6]0.0313.7 [12.8– 14.7]13.3 [12.8–14.6]0.47 Leukocytes 1000/mm3 (median [IQR])7.6 [6.5– 8.5]7.4 [6.1–8.7]0.857.7 [6.6– 8.5]7.4 [6.1–8.7]0.80 Platelets 1000/mm3 (median [IQR])308 [260 – 364]311 [265–374]0.80311 [276– 377]315 [275–380]0.99 Neutrophils 1000/µL (median [IQR])4.2 [3.2– 5.0]3.9 [3.2–5.0]0.864.1 [3.4– 4.9]4.1 [3.3–5.1]0.68 Lymphocytes 1000/µL (median [IQR])2.6 [2.1– 3.2]2.6 [2.0–3.2]0.462.6 [2.2– 3.3]2.56 [2.0–3.1]0.20*Strict IGRA cut-offs: baseline IFNγ (TBAg–nil tube) <0.15 IU/mL, and follow-up IGRA either <0.15 IU/mL (persistently IGRA-negative) or >0.7 IU/mL (IGRA converters).BCG bacillus Calmette-Guerin, BMI body mass index, IGRA interferon-gamma release assay, IQR interquartile range.Bold values are P < 0.05.aExposure scores were derived from a logistic regression model of M. tuberculosis exposure variables (index case: sputum smear grade, cavities, extent of X-ray disease; contacts: hours spent with and sleeping proximity to the case) with IGRA results at 14 weeks as the dependent variable.bThe M. tuberculosis genotype was only known for 179 index patients in the full set and 111 in the subset. Clinical characteristics of Indonesian tuberculosis household contacts with metabolomics data *Strict IGRA cut-offs: baseline IFNγ (TBAg–nil tube) <0.15 IU/mL, and follow-up IGRA either <0.15 IU/mL (persistently IGRA-negative) or >0.7 IU/mL (IGRA converters). BCG bacillus Calmette-Guerin, BMI body mass index, IGRA interferon-gamma release assay, IQR interquartile range. Bold values are P < 0.05. aExposure scores were derived from a logistic regression model of M. tuberculosis exposure variables (index case: sputum smear grade, cavities, extent of X-ray disease; contacts: hours spent with and sleeping proximity to the case) with IGRA results at 14 weeks as the dependent variable. bThe M. tuberculosis genotype was only known for 179 index patients in the full set and 111 in the subset. Circulating metabolites in tuberculosis household contacts Using liquid chromatography-mass spectrometry, we measured a broad range of untargeted circulating metabolites, including polar metabolites (e.g., organic acids), lipids (e.g., triglycerides), free fatty acids, and bile acids in 199 household contacts. After quality control, 528 annotated circulating serum metabolites were included in further analysis. A number of metabolites were significantly associated with contacts’ age (37% of 528 metabolites), sex (14%) and BMI (26%), as well as hemoglobin (19%), but very few were associated with exposure or with the presence of a BCG scar, which had shown a strong association with IGRA status at follow-up in the complete cohort4 (0.004%; false discovery rate [FDR] < 0.05 for all variables, Fig. 2A). Baseline circulating metabolites associated with Indonesian household contacts characteristics and IGRA-status at 14 weeks. A Annotated metabolites were correlated with household contacts characteristics using Spearman's corrected for multiple testing correction (FDR, false discovery rate). The heatmap visualizes significant correlations (FDR < 0.05) with positive (green) and negative (pink) correlations. B Volcano plot of metabolites predicting IGRA status at follow-up based on logistic regression while adjusting for age, sex, BMI, presence of BCG scar and exposure risk score. Strict IGRA cut-offs were used (see Fig. 1). P values shown on the y-axis are not corrected for multiple testing. C Heatmap of coefficient estimates of all significant metabolites (P < 0.05) from all contacts (N = 199) and the subset with stricter IGRA criteria cut-offs (N = 129). After correction for multiple testing (FDR < 0.05), there were no metabolites for which concentrations were significantly different between IGRA converters (N = 71) and persistently IGRA-negative individuals (N = 128). However, exploratory analysis using an uncorrected P < 0.05 threshold identified 36 metabolites with differential abundance, providing hypothesis-generating candidates for pathway analysis (Fig. S1). After application of the stricter IGRA cut-offs (leaving 129 household contacts), again, no metabolites were different between the two groups after correction for multiple testing (FDR < 0.05). However, exploratory analysis using uncorrected P < 0.05 showed that 64 metabolites were differentially abundant (Fig. 2B). The directionality of effects was similar, but effects were larger when stricter IGRA cut-offs were applied (Fig. 2C). Almost all (94%) differentially abundant metabolites also significantly correlated with quantitative IGRA results at follow-up (Fig. S2). Importantly, logistic regression models included the level of exposure as a covariate to control for confounding. However, separate Spearman correlation analysis (Fig. 2A) revealed that none of the metabolites for which plasma concentration were significantly associated with IGRA status also showed a significant association with exposure level (FDR > 0.05), suggesting that these metabolites contribute to immune protection independent of exposure intensity. Metabolic pathways associated with IGRA conversion Visualization of the differentially abundant metabolites identified one large and connected network including: arginine, proline, aspartate, and glutamate metabolism; tryptophan metabolism; valine, leucine, and isoleucine degradation; glycine, serine, and threonine metabolism; and glycerolipid metabolism (Fig. 3A). Persistently IGRA-negative individuals showed decreased concentrations of upstream metabolites (Fig. 3A, red nodes, e.g., glycine, glutamate, serine, proline, and tryptophan) and increased concentrations of downstream metabolites (blue nodes, e.g., phosphatidylcholine, cholesterol ester, serotonin, and indoleacetic acid) suggesting increased flux through these pathways in this group. There was also one small unconnected network of arachidonic acid metabolism in which the metabolites were higher in IGRA converters. Similar results were found when we used metabolic overrepresentation analysis as an alternative method (Fig. 3B). Three metabolic pathways were significantly associated with IGRA status at follow-up: arachidonic acid metabolism (FDR = 0.005); glutathione metabolism (FDR = 0.017); arginine and proline metabolism (FDR = 0.017); while the association for tryptophan metabolism almost reached significance (FDR = 0.087). The metabolites’ position depicting the flux in each of these pathways is shown in Fig. S3. Metabolic pathways associated with Indonesian household contacts IGRA-status at 14 weeks. A KEGG metabolic pathway network analysis based on metabolite concentration at baseline, which were either more abundant in IGRA converters (red) or persistently IGRA-negative individuals (blue). B Overrepresentation enrichment analysis based on differentially abundant metabolites, whereby enrichment ratio = [number of differentially abundant metabolites (“hits”)/expected number of hits]. Genetic variations in metabolic genes associated with IGRA conversion Genetic variants located in or near genes encoding metabolic enzymes can affect both enzyme activity and the function of associated metabolic pathways28,29. To examine whether genetic variation in metabolic pathway genes may contribute to IGRA conversion, we performed a targeted genetic association study in 367 household contacts, including 95 with IGRA conversion and 272 with repeatedly negative IGRAs. We extracted available SNPs for the four metabolic pathways showing differential activity (arachidonic acid metabolism, glutathione metabolism, arginine and proline metabolism, and tryptophan metabolism) with a ±50 kb window around gene boundaries to capture potential regulatory variants. After LD-based clumping (r² < 0.1, East Asian reference panel) to identify independent signals, we obtained 120 lead SNPs with an independent association with IGRA conversion (P < 0.05). We then examined if these GWAS-identified SNPs have shown to affect gene expression. Using the open targets platform molQTL database we found that 14 out of 120 GWAS-identified lead SNPs were associated with metabolic gene expression. Two SNPs showed a particularly strong association with IGRA conversion among our household contacts as well as with gene expression in the open targets platform. The variant rs3642 (P = 1.8 × 10−⁵ for IGRA status at follow-up, Fig. 4A) lies 21 kb upstream of AFMID, a gene involved in tryptophan metabolism (Fig. 4C). Among household contacts, carriers of the alternative A allele (GA genotype) more often showed IGRA conversion compared to the reference GG genotype (Fig. 4E), with the homozygous AA genotype being absent. Rs3642 is part of a 22-variant credible set associated with reduced AFMID expression in cerebellum, with a posterior inclusion probability (PIP; a statistical measure used to determine the likelihood that a specific genetic variant is the true causal variant for a particular trait) of 0.049. The lead variant shows a strong negative effect on AFMID expression (β = −0.52, P = 3.6 × 10−⁸; Fig. 4B), suggesting that rs3642 contributes to lower AFMID expression, which may reduce tryptophan catabolism and increase circulating tryptophan levels. Genetic variants of selected metabolic genes associated with IGRA-status in Indonesian household contacts at 14 weeks. A Volcano plot of genetic associations between 68 lead SNPs (in 214 metabolic genes ±50 kb, after LD pruning at r² < 0.1) and IGRA status at 14 weeks. Two SNPs exceeded the significance threshold: rs3642 near AFMID and rs13056032 near PRODH. B eQTL summary for significant SNPs from the open targets platform, showing credible set information, tissue-specific expression effects, and posterior inclusion probabilities from SuSiE fine-mapping. Regional association plots for the AFMID locus on chromosome 17 (C) and PRODH locus on chromosome 22 (D), with SNPs colored by LD (r²) with lead variants rs3642 and rs13056032, respectively. Genotype distribution according to IGRA status for the top SNPs in AFMID (E) and PRODH (F). Serum tryptophan levels by rs3642 genotype (G) and proline levels by rs13056032 genotype (H). No significant difference was observed for tryptophan (P = 0.21), while TT carriers had significantly higher proline concentrations compared to C allele carriers (P = 0.017). The second variant rs13056032 (P = 9.7 × 10−⁵) is located near PRODH, involved in arginine and proline metabolism (Fig. 4D). Carriers of the alternative C allele (TC + CC) were less likely to show IGRA conversion compared to the reference TT genotype (Fig. 4F). Rs13056032 appears in multiple credible sets for reduced PRODH expression, including tibial nerve (lead β = −0.34, P = 3.9 × 10−¹¹, PIP = 0.041) and skeletal muscle, where it is the lead variant with higher support (β = −0.28, P = 2.6 × 10⁻⁹, PIP = 0.244; Fig. 4B). Reduced PRODH expression may therefore result in higher proline levels. We next examined the correlation between AFMID and PRODH genotype and levels of metabolites upstream of the encoded enzyme. For rs3642 in AFMID, among 151 household contacts with genotype and metabolite data, compared to individuals with the GG genotype, there were no differences with the GA genotype (P = 0.21; Fig. 4G). For rs13056032 at PRODH, the homozygous reference genotype TT, which was associated with lower PRODH gene expression compared to the alternative allele, was also associated with higher circulating proline levels compared to TC (P = 0.031), and TC + CC genotypes (P = 0.017; Fig. 4H). As shown in Fig. 2B, C, higher circulating concentrations of both tryptophan and proline were associated with IGRA conversion. Collectively, our findings on the genetic regulation of metabolic enzymes and concentrations of circulating metabolites associated with IGRA results support the notion that metabolic activity contributes to protection against Mtb infection. Circulating metabolites associated with ex vivo mycobacterial growth To strengthen the link between circulating metabolites and prevention of Mtb infection, we next examined if concentrations of circulating metabolites correlated with mycobacterial growth inhibition ex vivo (Fig. 1C)30. We purposely used controls from the Netherlands as a low-incidence setting to avoid a possible bias due to pre-existing immunity to Mtb. No IGRAs were performed in these individuals (n = 42, median age 23 years, 48% female), but TST- or IGRA-positivity is extremely low in the Netherlands (<1% for individuals born after 1970; Dutch guideline for contact investigation [Tuberculose bron-en contactonderzoek] 2024, appendix II). Also, BCG vaccination is not routinely done in the Netherlands, and none of the study subjects had a visible BCG scar. Paired PBMCs and plasma samples were collected from healthy Dutch donors enrolled at the time of entry in the 300BCG study, before their actual BCG vaccination. Peripheral blood mononuclear cells (PBMCs) from 11 out of 42 participants (26%) exhibited mycobacterial killing capacity in the mycobacterial growth inhibition assay (MGIA), defined as ≥ 50% reduction of M. bovis BCG Pasteur colony-forming units (CFU) compared to the initial inoculum, while 31 participants did not (Fig. 5A)30. We correlated plasma concentrations of 1334 annotated metabolites with CFUs as a reflection of mycobacterial growth control in the MGIA. We used linear regression to predict whether circulating metabolite levels correlated with the CFU from MGIA without adjustment for covariates, as no other variables both correlated with metabolites and CFU of the MGIA (Fig. S4). The abundance (peak intensity) of 329 circulating metabolites (24.6%) correlated with CFU in MGIA using a P < 0.1 to identify a sufficient number of metabolites for pathway analysis (Fig. 5B). Enrichment analysis of metabolites associated with MGIA results identified similar pathways (Fig. 5C) as we had found to be associated with IGRA conversion in the Indonesian tuberculosis household contacts (Figs. 3B and 4B). Correlation of circulating metabolites with mycobacterial growth inhibition in healthy control subjects. A Histogram showing mycobacterial growth inhibition of PBMCs of Dutch healthy individuals (n = 42), divided in those with control (n = 11, blue, defined as ≥50% CFU reduction of the inoculum of 2.38 log CFU M. bovis BCG [dashed line] after 4 days), and those without growth control (n = 31, pink, <50% CFU reduction). B Volcano plot of circulating metabolite concentrations significantly correlated (blue) or inversely correlated with growth inhibition (MGIA) of M. bovis BCG based on linear regression (P < 0.1). C Overrepresentation enrichment analysis based on metabolites that significantly correlated with MGIA results (P-values not corrected for multiple testing). Concentrations of seven metabolites (labeled in 5B), shown in peak ion intensity (log MFI), correlated with colony-forming units (CFU) in MGIA (D), and IGRA status at follow-up among household contacts (E). We then examined the overlap between individual metabolites associated with MGIA results in healthy Dutch adults and metabolites associated with IGRA status in Indonesian tuberculosis household contacts. Out of 125 metabolites available in both datasets, seven both correlated with MGIA and IGRA results (labeled in Fig. 5B). These seven metabolites were prostaglandin F2 alpha and leukotriene B4 (included in arachidonic acid metabolism), 13-HODE (linoleic acid metabolism), tryptophan and indoleacetic acid (tryptophan metabolism), glycine (glycine and serine metabolism) and proline (arginine and proline metabolism, Fig. S5). Lower concentrations of all seven correlated with lower CFUs (better mycobacterial growth inhibition) in MGIA (Fig. 5B, D). In line with these findings, lower circulating concentrations of six of these seven metabolites (all except indoleacetic acid) were found in household contacts with a negative follow-up IGRA result (reflecting protection against Mtb infection; Fig. 5E). For instance, lower circulating concentrations of tryptophan and proline both correlated with lower CFU in MGIA (β = 3.6, P = 0.05; β = 3.60, P = 0.04), reflecting better mycobacterial growth control, as well as with a persistently negative IGRA result (OR = 0.10, P = 0.003; OR = 0.25, P = 0.02; adjusted for age, sex, BMI, BCG scar, and exposure risk score). Discussion We examined circulating metabolites and genetic traits among heavily exposed tuberculosis household contacts in Indonesia, of whom approximately one-fourth still had a negative IGRA result 3 months after diagnosis of tuberculosis in the index case. Based on differentially abundant baseline metabolites, lack of IGRA conversion during follow-up correlated with activity of several metabolic pathways, including arachidonic acid, glutathione, tryptophan, and arginine and proline metabolism. Also, SNPs near genes encoding for enzymes in tryptophan, arginine and proline metabolism both correlated with enzyme expression at the tissue level in a publicly available dataset, as well as with lack of IGRA conversion and circulating proline concentrations in our household contacts. Finally, concentrations of six circulating metabolites both correlated with better ex vivo control of mycobacterial growth among healthy individuals in a low-incidence setting as well as with lack of IGRA conversion in the Indonesian tuberculosis household contacts. Collectively, these data support the notion that circulating metabolites may impact innate host defense against Mtb infection. Several studies have examined the metabolism of immune cells infected with Mtb, showing how glycolysis31–33 and host lipid metabolism34 control Mtb growth. However, circulating metabolites, derived from the gut microbiome, diet, liver and other tissues, may also impact immune cell function18,19. In line with our findings, plasma eicosanoid 9-HETE, 5-HETE, leukotriene B4, and 12-HETE were higher in Gambian contacts who experienced TST conversion within three months of diagnosis of the index patient (N = 38) compared to those who did not (N = 40)25. A study from India showed surprisingly large differences in baseline plasma metabolite profiles of 10 IGRA-negative household contacts who either showed IGRA conversion or not, but the directionality of some highlighted metabolites (e.g., plasma N-formylmethionine) was opposite to ours24. Looking for further support for the notion that circulating metabolites may contribute to protection against Mtb infection, we examined eQTLs of genes encoding for enzymes of metabolic pathways. From our identified pathways, eQTLs of AFMID and PRODH correlated with IGRA status at follow-up. For PRODH, eQTL SNPs correlated with levels of the upstream metabolite proline in our cohort. This suggests that these genetic variants affect enzyme activity and systemic metabolite concentrations, which in turn contribute to protection against Mtb infection. This finding is reminiscent of other genotype-phenotype relationships in tuberculosis. For instance, a SNP in LTA4H which encodes for leukotriene A4 hydrolase (the final step in leukotriene B4 biosynthesis which regulates eicosanoid metabolism) was shown to impact inflammation and mycobacterial growth in a zebra fish model35, and to predict survival of tuberculous meningitis patients36. For further functional validation, we examined if circulating metabolite profiles correlated with in vitro control of mycobacterial growth by PBMCs. We purposely selected healthy individuals from a low-incidence setting to minimize possible confounding by prior Mtb exposure or pre-existing immunity from BCG vaccination. In line with our data from Indonesian tuberculosis household contacts, metabolic pathways including arachidonic acid, glutathione, tryptophan, arginine, and proline metabolism correlated with control of M. bovis BCG growth. Notably, six key metabolites, prostaglandin F2α, leukotriene B4, 13-HODE, tryptophan, glycine, and proline, both correlated with mycobacterial growth control in vitro in the Dutch cohort and lack of IGRA conversion in Indonesian household contacts. How should we interpret these findings? Arachidonic acid metabolism, through its balance of various mediators, including prostaglandins and leukotrienes, plays a role in orchestrating inflammatory responses and macrophage function during interaction with Mtb37–40. Higher levels of pro-inflammatory arachidonic acid-derived metabolites correlated with IGRA conversion in our cohort, in alignment with findings in Gambian TST-converters25. Similarly, our findings on higher circulating tryptophan levels in IGRA-converters concur with the role of tryptophan metabolism in the antimicrobial and inflammatory responses to Mtb15,41,42. Interestingly, we have previously shown that elevated cerebrospinal fluid tryptophan is associated with higher mortality in tuberculous meningitis patients43. Finally, for proline, higher concentrations were associated with IGRA conversion and lack of ex vivo mycobacterial growth control. This aligns with the fact that local cellular proline catabolism by PRODH generates reactive oxygen species (ROS)44 and induces autophagy45, which contributes to control of Mtb in mice46. Our findings may help develop preventive strategies through dietary or pharmacological interventions. A study in Tanzania has shown that an urban diet leads to increased inflammation with higher circulating levels of arachidonic acid metabolites, such as leukotriene B4 and prostaglandin F2α, while apigenin, the most abundant flavanone in the rural diet, reversed pro-inflammatory monocyte reprogramming19. Specifically for tuberculosis, vitamin supplementation has shown to prevent incident disease47,48, with a possible role for tryptophan metabolism suggested from experimental studies31,49 Our study has several limitations. We had no second cohort of tuberculosis contacts for validation, but our cohort was large, and our phenotype is strong, comparing heavily exposed individuals whose IGRA either remained negative or converted in the context of a well-defined exposure to an infectious index patient with a known Mtb isolate4,6,7. Moreover, we further strengthened our phenotypes by using stricter IGRA cut-offs to improve the signal-to-noise ratio, as it has been previously shown that individuals with quantitative IGRA values between 0.15 and 0.35 IU/mL may have Mtb infection26, and that IGRA results around 0.2–0.7 IU/ml are often explained by test variability27. Also, we did not examine possible IGRA conversion or reversion beyond the 3-month time point. However, our early clearance phenotype of heavily exposed household contacts with a repeatedly negative IGRA is a different and more frequent phenotype than the “TB resisters”, individuals who remain TST/IGRA negative despite years of exposure6. By repeated IGRA testing over a relatively short period, we look at a defined exposure within that household. Using this approach, we have shown that the presence of a BCG scar is strongly associated with lack of IGRA conversion4, but not if the index patient is infected with an L2 (Beijing) Mtb strain7, and less so with a higher bacterial load or higher exposure (e.g., close sleeping proximity)4. Our phenotype likely reflects stronger innate immune protection, and we have indeed identified innate immune cell phenotype and function to be associated with the phenotype8. Finally, one might also wonder if our findings on the differential abundance of metabolites in persistently IGRA-negative contact reflect a lack of exposure rather than protection against Mtb infection. However, contacts with a BCG scar were strongly protected against IGRA conversion, unless they were exposed to an L2 Mtb isolate7, and neither BCG scar nor index Mtb genotype would logically depend on sleeping proximity to the index patient or other markers of exposure. Experimental models may explain how our identified metabolic pathways correlate with innate immune responses to Mtb. We identified candidate SNPs based on proximity to metabolic pathway genes and LD structure. While these SNPs showed associations with gene expression in public databases, formal colocalization analysis would be needed to definitively establish that the same causal variants drive both IGRA conversion and altered gene expression. Additionally, the tissue-specific nature of eQTL effects means that the relevant tissues for TB protection may differ from those available in current databases. Taken together, we identified metabolites, metabolic pathways and functional genetic loci that correlated with lack of IGRA conversion after heavy Mtb exposure and with ex vivo mycobacterial growth control. As such, our data support the notion that circulating metabolites may affect outcome after exposure to Mtb. Future studies should evaluate if metabolic modulation, dietary interventions or supplementation can help prevent Mtb infection. Materials and methods Tuberculosis household contacts This study was embedded within a large household contact study (INFECT), which was conducted in Bandung, Indonesia, between 2014 and 2018. A detailed description of the study is found elsewhere4. In short, household contacts of sputum smear-positive tuberculosis patients were eligible if they were over five years of age, had no history of tuberculosis, and had lived with the adult with tuberculosis for more than 5 h per week in the month prior to enrollment. They were enrolled within one week after diagnosis of the index patient and screened for tuberculosis disease using symptoms screen, chest X-ray and sputum microscopy and culture. Sociodemographic data and risk factors for Mtb infection were collected, including the level of exposure4, as measured by sleeping proximity, time spent with the index patient, and presence of cavities, and sputum mycobacterial load in the index patient. Given the low prevalence of HIV among index patients (0.5%) and the general population at the time of the study (0.2%), contacts were not tested for HIV. Mtb infection status of contacts was assessed by QuantiFERON-TB gold in-tube (QFT-GIT) IGRA, which was repeated at 14 weeks in those who were initially IGRA-negative. Based on QFT-GIT results, contacts were first classified as persistently IGRA-negative individuals and IGRA converters using the manufacturer’s cut-off value for the TBAg tube (0.35 IU/mL). Similar to a previous study8, to improve the strength of the phenotypes, instead of the manufacturer’s cut-off (0.35 IU/mL), we applied stricter cut-offs, only including individuals whose baseline IFN-γ result (TBAg–nil tube) was <0.15 IU/mL, and whose follow-up IGRA (TBAg–nil) was either <0.15 IU/mL (persistently IGRA-negative individuals) or >0.7 IU/mL (IGRA converters)50,51. Circulating metabolites and pathways analysis in household contacts Among Indonesian tuberculosis household contacts, serum was collected after centrifugation of peripheral blood at 1500 × g for 15 min. Serum samples from 199 study participants from the INFECT cohort who were IGRA-negative at baseline were selected for metabolomics analysis. Four liquid chromatography tandem mass spectrometry (LC-MS) methods were used to detect a broad range of metabolite classes, including polar metabolites (e.g., organic acids), lipids (e.g., triglycerides), free fatty acids, and bile acids. More detailed description of the methodology is available elsewhere52. A total of 537 annotated metabolites were measured before quality control. For metabolites significantly associated with the phenotype, metabolic pathway enrichment analysis was performed (see “Statistics and reproducibility”). Correlation of variants in metabolic genes with IGRA status DNA samples were genotyped using a commercially available single-nucleotide polymorphism (SNP) chip (HumanOmniExpressExome-8 v1.0; Illumina; San Diego, CA, USA). Genotype calling was performed using Opticall 0.7.0 using the default settings. Samples with a call rate less than 0.99 were excluded, as were variants with a Hardy–Weinberg equilibrium p < 0.0001, and minor allele frequency less than 0.001. No genetic outliers were detected. We imputed the samples using IMPUTE2 and used a combination of Whole Genome Sequencing (WGS) data from 227 West Javanese individuals with East Asian data from the 1000 Genomes Project (EASp+INDp) as reference panel for better imputation accuracy53, and then selected variants with an R² of at least 0.8. After quality control, 4,407,262 genetic variants were identified. Circulating metabolites and mycobacterial growth inhibition For correlation of circulating metabolites with mycobacterial growth inhibition, healthy Dutch adults were recruited between April 2017 and June 2018 as part of the 300BCG21 (NL58553.091.16, https://onderzoekmetmensen.nl/en/trial/45603) study. Blood for the separation of plasma and isolation of PBMCs was collected before BCG vaccination. Plasma metabolite levels were measured before BCG vaccination was administered. The metabolic features were measured and annotated by the General Metabolics (Zurich, Switzerland) using flow injection time-of-flight mass (flow-injection TOF-M) spectrometry, where it was performed on a platform consisting of an Agilent Series 1100 LC pump coupled to a Gerstel MPS2 autosampler and an Agilent 6520 Series Quadrupole Time-of-flight mass spectrometer (Agilent, Santa Clara, CA) equipped with an electrospray source operated in negative and positive mode54. For the hydrophilic interaction chromatography negative ion method, serum samples were precipitated in 80% methanol containing inosine-15N4, thymine-d4, and glycocholate-d4 internal standards (Cambridge Isotope Laboratories; Andover, MA, USA). Samples were cleared by centrifugation, and supernatants injected onto a 150 mm × 2 mm Luna NH2 column (Phenomenex; Torrance, CA, USA). For the hydrophilic interaction chromatography positive ion method, samples were precipitated with nine volumes of 74.9% acetonitrile, 24.9% methanol, and 0.2% formic acid, by volume, containing stable isotope-labeled internal standards (valine-d8; Isotec, Sigma-Aldrich; St Louis, MI, USA; and phenylalanine-d; Cambridge Isotope Laboratories). After centrifugation, supernatants were injected onto a 150 mm × 2 mm Atlantis hydrophilic interaction chromatography column (Waters; Milford, MA, USA). For measuring compounds of intermediate polarity, serum and plasma samples were extracted in methanol containing PGE2-d4 as an internal standard (Cayman Chemical Company, Ann Arbor, MI, USA), and supernatants of samples were cleared by centrifugation. Extracted samples underwent solid phase extraction with an Oasis HLB Cartridges column (Waters) before injection onto a 150 mm × 2 mm ACQUITY T3 column (Waters). Finally, polar and nonpolar lipids were obtained by extracting samples in isopropanol containing 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphocholine as an internal standard (Avanti Polar Lipids; Alabaster, AL, USA). After centrifugation, supernatants were injected directly onto a 100 mm × 2.1 mm ACQUITY BEH C8 column (1.7 μm; Waters). Non-targeted metabolites were annotated according to human metabolites database (HMDB). All metabolomic measurements were performed in duplicate, and average values were calculated for each sample per feature. The mycobacterial growth inhibition assay used PBMC30. In short, cryopreserved PBMCs were thawed and rested in “R10 medium”, consisting of RPMI supplemented with glutamax (Gibco) and 10% FBS (Hyclone, Thermo Fisher Scientific Inc.), at a concentration of 2 × 106 cells/mL for 2 h in the presence of benzonase (20 U/ml, Merck, Amsterdam, the Netherlands). After resting, cells were washed with R10 medium and counted with a Casy Cellcounter (Roche, Woerden, the Netherlands). A total of 1 × 106 PBMCs were co-cultured for 4 days in RPMI supplemented with glutamax and 10% autologous human serum with 2.38 log CFU (±0.17 SD) of the M. bovis BCG Pasteur (P3) strain on a rotator in a 37 °C humidified CO2 incubator in a final volume of 600 μl. All samples were run in duplicates. After 4 days, 100 μL supernatant was harvested and stored for future analysis, and the remaining 500 μl per sample were transferred to a PANTA/Enrichment supplemented MGIT tube (Becton Dickinson) and placed in a BACTEC MGIT 960 system (BD) until time to positivity (TTP) was reached. All tubes included in the analysis were checked visually for possible contamination. Samples reaching positivity within 100 h were considered contaminated and thus deleted as false positive, since the inoculum of 2.38 log CFU BCG is expected to reach TTP after more than 250 h. As a control for the BCG inoculum all experiments included a serial dilution (107–102) of the BCG stock for time to positivity in PANTA/Enrichment supplemented MGIT tubes and plating on Middlebrook 7H10 agar plates, supplemented with 10% OADC (BD) for CFU determination. When all dilutions showed colonies on the 7H10 plates, the plates were scanned on a Canon Scanner 9000F, and colonies were counted using ImageJ software. CFUs were converted to log CFU and plotted against TTP. To identify metabolites associated with mycobacterial growth control, we performed linear regression analysis between plasma metabolite levels and log CFU without covariate adjustment due to the limited sample size (N = 42). Given this cohort’s role as a validation dataset and to allow for pathway analysis with a sufficient number of metabolites, we used a more lenient significance threshold (uncorrected P < 0.1). For metabolites significantly associated with CFU counts, metabolic pathway enrichment analysis was performed using MetaboAnalyst, with all detected metabolites serving as the background reference set. Only KEGG metabolic pathways were considered in the enrichment analysis. To validate the biological relevance of these findings, we intersected metabolites significantly associated with mycobacterial growth inhibition in with metabolites previously identified as significantly associated with lack of IGRA conversion in the Indonesian INFECT cohort. Statistics and reproducibility All statistical analyses were done in R version 4.4.1 with Rstudio integrated development environment55,56. Figures were generated using the R package “ggplot2” and “ggpubr”, unless stated otherwise57,58. Participant characteristics were compared using Mann–Whitney nonparametric tests. The tables were made using the R package “tableone”59. For metabolomics analysis, peak intensity liquid chromatography-mass spectrometry data were first log2-transformed before analysis, values below the limit of detection were imputed by half of the minimally detected value, and metabolites with missing values of more than 25% of the samples as well as those with a coefficient of variation >30% were omitted. For correlation with host factors, Spearman correlations were calculated, and the Benjamini–Hochberg adjustment made for multiple testing. A heatmap of the correlation matrix was created using “ComplexHeatmap”60. For identification of differentially abundant metabolites in the metabolomics substudy (N = 199 household contacts), logistic regression models were used, adjusted for age, sex, BMI, presence of BCG scar, and exposure score. Logistic regression implemented using the glm() function in R with the binomial family. The exposure risk scores were predicted from a logistic regression of Mtb exposure variables (index case: sputum smear grade, cavities, extent of radiographic disease; contacts: hours spent with, and sleeping proximity to the case) against IGRA results4. An uncorrected P < 0.05 was considered statistically significant in light of the discovery nature of the study, and volcano plots were created using “EnhancedVolcano”61. Two separate approaches were used for pathway analysis: visualization of metabolomics results using Metscape 3.10 app for Cytoscape and overrepresentation analysis based on KEGG pathways using MetaboAnalyst 5.0 with all measurable annotated metabolites in our dataset as the metabolome reference set (N = 537)62–64. For the metabolic genes analysis, unique genes from four KEGG metabolism pathways (arachidonic acid metabolism, arginine and proline metabolism, glutathione metabolism, and tryptophan metabolism) were selected for genetic validation (N = 214). We used GEMMA65 to find genetic variants associated with IGRA result at follow-up, by adjusting for age, sex, BMI, exposure risk score, and family relatedness by incorporating the genetic relatedness matrix into the analysis. We filtered SNPs located ±50 kb of the coding region of genes in the aforementioned metabolic pathways for further analysis. To identify independent association signals, we performed linkage disequilibrium (LD) pruning using the LDlinkR package with an r² threshold of 0.1 and MAF > 0.01, using combination of East Asian reference (combination of Chinese Han Beijing, Japanese in Tokyo, Southern Han Chinese, Dai Chinese, and Kinh people Vietnam) from the 1000 Genomes Project66. To investigate whether these GWAS-identified lead SNPs affect gene expression, we queried the open targets platform using the otargen R package67. We identified which of our lead SNPs were members of fine-mapped expression quantitative trait loci (eQTL) credible sets. For SNPs within credible sets, we extracted: (1) the lead variant of each credible set and its association statistics (P value, beta coefficient), (2) the posterior inclusion probability (PIP) for our SNP, indicating its likelihood of being the causal variant within the credible set, (3) the target gene and tissue type, and (4) the credible set size. All credible sets in open targets were fine-mapped using the SuSiE (sum of single effects) method with in-sample LD. Bonferroni’s correction were used to adjust the Wald’s P value from GEMMA after targeting the filtered variants. For genotype-metabolite associations, we compared metabolite levels between genotypes using Mann–Whitney’s nonparametric test. Study approval The Indonesian household contact study was approved by the Health Research Ethics Committee Universitas Padjadjaran (14/UN6.C2.1.2/KEPK/PN/2014) and the Southern Health and Disability Ethics Committee, New Zealand (13/STH/132). All participants gave their written informed consent before any study procedures were performed. The 300BCG21 (NL58553.091.16, https://onderzoekmetmensen.nl/en/trial/45603) study was approved by the Arnhem-Nijmegen Medical Ethical Committee and registered in The Overview of Medical Research in the Netherlands (OMON), formerly the Dutch Trial Register managed by Central Committee on Research Involving Human Subjects (CCMO). Informed consent was obtained from all subjects and/or their legal guardian(s). All ethical regulations relevant to human research participants were followed. Supplementary information Supplementary Information Supplementary Data 1 Acknowledgements We acknowledge the dedicated field, laboratory, and data management team that recruited the cohort, including Andini Cahya Nurani, Novianti, Deni, Wiwik Pratiwi, Dody Taufik Akbar, Emira Diandini, Dwi Febni Ratnaningsih, Inas Kathina, Yusak Sastra Atmaja, Nuni Haeruni, Anbarunik Puteri Danthin, Harold Eka Atmaja, Alif Al Birru, Nopi Susilawati, and Runi Rahmawati. Also, Rachel F. Hannaway, who works on INFECT from Otago University. Special thanks are also due to the TANDEM study team, especially coordinator Raspati C. Koesoemadinata, Lidya Chaidir, Jessi Anisa, and Ria Windyani, for their cooperation. We would also like to thank all volunteers from the 300BCG cohort for their participation in the study. A.J.V. was supported by a New Zealand Health Research Council Clinical Training Research Fellowship. The INFECT cohort recruitment was funded by the University of Otago and Mercy Hospital (through an endowment fund and directly), Dunedin, New Zealand. Index case recruitment and investigation was part of the TANDEM project (www.tandem-fp7.eu), which is supported by the European Union’s Seventh Framework Programme (FP7/2007–2013) under grant agreement number 305279. The IGRA (QuantiFERON) was donated by Qiagen. R.v.C. was supported by the Royal Netherlands Academy of Arts and Sciences (09-PD-14) and the VIDI grant 017.106.310 of the Netherlands Organization for Scientific Research. M.G.N. was supported by an ERC advanced grant (833247) and a Spinoza grant of the Netherlands Organization for Scientific Research. L.C.J.D.B. was partly funded by a grant to the Research Center for Vitamins and Vaccines (CVIVA) from the Danish National Research Foundation (DNRF108). Author contributions Conceptualization: T.P.S., A.J.V., P.C.H., V.A.C.M.K., and R.v.C.; methodology: T.P.S., A.J.V., P.C.H., J.A.P., E.A., V.K., V.A.C.M.K., and R.v.C.; data curation: T.P.S.; formal analysis: T.P.S., K.v.M., and S.J.; investigation: T.P.S., L.A., A.J.V., P.C.H., J.A.P., V.A.C.M.K., S.J.C.F.M.M., L.C.J.D.B., V.P.M., L.A.B.J., K.v.M., and S.J.; resources: A.J.V., P.C.H., A.R.I., B.A., C.C., and R.v.C.; writing-original draft: T.P.S., R.v.C.; writing-review and editing: T.P.S., V.K., A.R.D., K.v.M., A.v.L., L.A.B.J., P.C.H., M.G.N., S.J., V.A.C.M.K., and R.v.C.; visualization: T.P.S.; supervision: V.A.C.M.K., R.v.C., and P.C.H.; project administration: P.C.H.; funding acquisition: A.J.V., P.C.H., and R.v.C. Peer review Peer review information Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary handling editor: Johannes Stortz. Data availability All data are available in the main text or the Supplementary data 1. All Supplementary Figs. are available in the “Supplementary information” file. The early clearance GWAS summary statistic with accession number GCST90824109 has been deposited in the NHGRI-EBI GWAS Catalog. Supplementary data are provided as a “Supplementary data 1” file with this paper and available in Figshare 10.6084/m9.figshare.31769128. The metabolomics raw data of the INFECT household contacts in Indonesia deposited in the NIH Common Fund's National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench, https://www.metabolomicsworkbench.org where it has been assigned Study ID ST00479568. The data can be accessed directly via its Project 10.21228/M81Z8W. Metabolomics Workbench is supported by NIH grant U2C-DK119886 and OT2-OD030544 grants. The metabolomics data of the 300BCG Dutch cohort can be accessed in Gitlab page https://gitlab.com/xavier-lab-computation/public/bcg300. Code availability Details of publicly available software used in the study are given in the “Methods”. All the R scripts used to do the analysis related to this manuscript can be accessed in this Github page https://github.com/RC3ID-Data-Science/INFECT-metabolomics69. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains Supplementary material available at 10.1038/s42003-026-10241-9. References Associated Data Supplementary Materials Supplementary Information Supplementary Data 1 Data Availability Statement All data are available in the main text or the Supplementary data 1. All Supplementary Figs. are available in the “Supplementary information” file. The early clearance GWAS summary statistic with accession number GCST90824109 has been deposited in the NHGRI-EBI GWAS Catalog. Supplementary data are provided as a “Supplementary data 1” file with this paper and available in Figshare 10.6084/m9.figshare.31769128. The metabolomics raw data of the INFECT household contacts in Indonesia deposited in the NIH Common Fund's National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench, https://www.metabolomicsworkbench.org where it has been assigned Study ID ST00479568. The data can be accessed directly via its Project 10.21228/M81Z8W. Metabolomics Workbench is supported by NIH grant U2C-DK119886 and OT2-OD030544 grants. The metabolomics data of the 300BCG Dutch cohort can be accessed in Gitlab page https://gitlab.com/xavier-lab-computation/public/bcg300. Details of publicly available software used in the study are given in the “Methods”. All the R scripts used to do the analysis related to this manuscript can be accessed in this Github page https://github.com/RC3ID-Data-Science/INFECT-metabolomics69.