IMetaboMap - Immnune responses and circular Metabolic interactions Map


IMetaboMap, is a pioneering resource that catalogs interactions between plasma metabolites and cytokine responses to various stimuli in humans. This tool is instrumental in examining how these interactions differ among various populations, between sexes, and across different cell systems. Our findings in Cohort_EU1 revealed a total of 125,307 metabolite-cytokine associations. In contrast, Cohort_EU2 yielded 28,833 associations, while Cohort_AF added another 80,550, bringing the comprehensive tally to 234,690 unique associations cataloged in IMetaboMap. This tool stands as a testament to our comprehensive approach, enabling detailed exploration of the dynamic relationships between metabolites and cytokines, and shedding light on specific associations that may vary by sex and population.



Cohorts’ description

The first cohort, designated as Cohort_EU1, is part of the Human Functional Genomics Project's 500FG cohort and comprises 534 healthy Caucasian individuals, with ages spanning from 18 to 75 years. This cohort was carefully selected to exclude individuals with mixed genetic backgrounds or chronic diseases. Key measurements within Cohort_EU1 included cytokine production in response to various stimulations and comprehensive metabolomic profiling. More detailed information can be found in previous publications (Li et al., 2016). The second cohort from Western Europe, Cohort_EU2, included 324 healthy volunteers of Western European descent, aged 18 to 71 years. These participants were enrolled in the 300BCG cohort from April 2017 to June 2018, with further details documented in previous publication (Koeken et al., 2020). The third cohort, Cohort_AF, encompasses 323 healthy Tanzanians between 18 to 65 years old from the Kilimanjaro region, who were recruited through the Kilimanjaro Christian Medical Center and Lucy Lameck Research Center from March to December 2017, as previously described (Temba et al., 2021). Plasma samples of these donors were obtained and used to measure their circulating metabolites profiles. For the metabolomics analysis, untargeted measurements from plasma samples were conducted using high-throughput flow injection-time-of-flight mass spectrometry (Fuhrer et al., 2011). Blood samples of these donors were also collected and exposed to different stimulators for cytokine profile measurement.


Methods Overview

Confounder Adjustment and Metabolite–Cytokine Association Analysis

We first assessed the extent to which age, sex, and BMI confounded raw metabolite–cytokine correlations. In three cohorts (Cohort_AF, Cohort_EU1 and Cohort_EU2), we performed Spearman rank correlations between metabolite features and each demographic variable (sex, age, BMI), applying Benjamini–Hochberg FDR correction (FDR < 0.05). Before adjustment, the proportions of metabolites significantly associated with each covariate were as follows (Appendix Figure S6): Age: 50.7% (Cohort_AF), 32.6% (Cohort_EU1), 22.1% (Cohort_EU2); Sex: 43.9% (Cohort_AF), 53.1% (Cohort_EU1), 39.9% (Cohort_EU2); BMI: 20.6% (Cohort_AF), 18.4% (Cohort_EU1), 0% (Cohort_EU2). To remove these effects, we fitted a multivariable linear regression model for each metabolite in R (v4.x). In Cohort_AF and Cohort_EU1—where ≥ 15% of metabolites were BMI-associated—we included sex, age and BMI as covariates. In EU2, where no metabolites showed BMI associations, we included only sex and age to avoid over-adjustment. After model fitting, we extracted the residuals and re-tested their Spearman correlations with sex, age and BMI (FDR < 0.05). Post-adjustment, fewer than 1% of metabolites remained significantly correlated with any covariate in all cohorts (Cohort_AF: 0.7% age, 1.2% sex, 0.7% BMI; Cohort_EU1: 0.1% age, 0.1% sex, 0% BMI; Cohort_EU2: 0% age, 0% sex, 0% BMI; Appendix Figure S6), demonstrating effective confounder removal. For downstream metabolite–cytokine association testing, all analyses were conducted on these de-confounded residuals. We again performed Spearman correlations against each immune cytokine and retained only those metabolite features with FDR < 0.05 in at least one cytokine comparison for further analysis. All model fitting and residual extraction used R’s lm() function, correlations were computed with corr.test(), and multiple-testing correction was applied via p.adjust(method="fdr").

Meta‐analysis of metabolite–cytokine correlations

We combined correlation results for four key cytokines (IL-1β, IL-6, TNF, IFN-γ) across three cohorts (AF, EU1 and EU2). First, we loaded each cohort’s metabolite–cytokine correlation table, renamed and standardized columns (Metabolite, Cytokine, r, cohort, n), and mapped all cytokine names to a common scheme. We then filtered to the four target cytokines and kept only metabolites present in all three cohorts. For each cohort–cytokine pair, we transformed Spearman’s r to Fisher’s Z, estimated its variance (1/(n–3)) and back‐transformed 95 % confidence bounds to r. We reshaped the data so that AF, EU1_WB, EU1_PBMC and EU2 Z‐scores and variances aligned by metabolite–cytokine combination. Wherever at least two cohorts contributed data, we ran fixed‐effect and REML random‐effects meta‐analyses (metafor::rma), extracting pooled Z (and corresponding r), 95 % CIs, p-values, Cochran’s Q and I². The full meta‐analytic table was saved as an Excel file, and individual forest plots for each metabolite–cytokine pair were generated and exported as PNG images. Metabolite–cytokine correlation meta‐analysis was performed across three cohorts (AF, EU1 and EU2) for IL-1β, IL-6, TNF and IFN-γ. Cohort‐specific tables were first harmonized by standardizing column names and mapping cytokine assay identifiers to a common scheme. Only metabolites quantified in all sources were retained. For each metabolite–cytokine pair, Spearman’s r was transformed to Fisher’s Z, its variance estimated as 1/(n – 3), and 95 % confidence bounds back‐transformed to r. Z‐scores and variances from any two or more sources were then meta‐analyzed using both fixed‐effect and REML random‐effects models (metafor::rma), yielding pooled Z (and corresponding r), 95 % CIs, p-values, Cochran’s Q and I².


Citing the article:

Jianbo Fu, Nienke van Unen, Andrei Sarlea, Nhan Nguyen, Martin Jaeger, Javier Botey-Bataller, Valerie A.C.M. Koeken, L. Charlotte de Bree, Vera P. Mourits, Simone J.C.F.M. Moorlag, Godfrey Temba, Vesla I. Kullaya, Quirijn de Mast, Leo A.B. Joosten, Cheng-Jian Xu, Mihai G. Netea, Yang Li.Deciphering Cross-Cohort Metabolic Signatures of Immune Responses and Their Implications in Disease Pathogenesis. https://www.embopress.org/doi/full/10.1038/s44320-025-00146-w

Waiting...
Legend:
Positive correlation
Negative correlation
Cytokine
Metabolite
Legend:
Positive correlation
Negative correlation
Cytokine
Metabolite
Legend:
Positive correlation
Negative correlation
Cytokine
Metabolite

Cohort_EU1


Cohort_EU2


Cohort_AF


1. Corellation Analysis

Cohort_EU1

Cohort_EU2

Cohort_AF

2. Fisher’s Z Test for Sex-Specific Metabolite–Cytokine Correlations

Cohort_EU1

Cohort_EU2

Cohort_AF

Summary of sex‐specific differences in metabolite–cytokine correlations across each cohort. For each metabolite–cytokine pair and cohort, r_male and p_male report Spearman’s ρ and two‐tailed p‐value in males; r_female and p_female report the same values in females. The z_diff column gives the difference between Fisher’s Z–transformed male and female correlation coefficients; p_diff is the p‐value from a two‐sample Z‐test comparing these transformed correlations, and q_diff is the Benjamini–Hochberg FDR‐adjusted p_diff. Values with p_diff < 0.05 are highlighted as suggestive of sex‐specific effects.


Meta‐analysis of metabolite–cytokine associations across four data sources (AF, EU1‐WB, EU1‐PBMC, EU2). For each metabolite–cytokine pair, the Metabolite and Cytokine columns list the names tested; r_AF, r_EU1_WB, r_EU1_PBMC, r_EU2 (with their corresponding CI_…_low and CI_…_high columns) report cohort‐specific Spearman’s correlation coefficients and 95 % confidence intervals. Fixed‐effect meta‐analysis results are given in fixed_r, fixed_CI_low, fixed_CI_high, fixed_p, while REML random‐effects results appear in random_r, random_CI_low, random_CI_high, random_p. Between‐study heterogeneity is summarized by Cochran’s Q (p-value Q_p) and the I² statistic.




1. Overview

Study overview and cohorts information.

2. Network

Step 1. Either entering the metabolite manually or selecting it from a dropdown box. The network graph shown on the right is based on your search, but of course you can also download the correlation data based on your search.

Step 2. Enter the cytokine yourself or select it from the drop down box.

Step 3. Enter the stimulus yourself or select it from the drop down box.

Note: If the steps are not followed above, for example, if only cytokines or stimuli are entered, the network graph on the right may take a long time to wait or even get stuck because of the amount of data, and needs to be turned off and re-entered.

3. Data

The results of associations analysis between plasma metabolites and cytokine responses to various stimuli. Users can search for their target metabolite or cytokine, or both, as explained below for the corresponding results of various combinations of searches. Please wait for a new table to be generated during the search.

3.1 Search by metabolite only

3.1.1 Results in multi-cohorts

2.1.2 Results in different sex

2.1.3 Metabolites’ information

3.2 Search by metabolite and cytokine

3.2.1 Results in multi-cohorts

3.2.2 Results in different sex

3.3 Search by metabolite, cytokine, and stimulus

3.3.1 Results in multi-cohorts

3.3.2 Results in different sex

5. About

Supplementary information concerning the contact references and data release platforms.