Meta-Analysis Identification of Highly Robust and Differential Immune-Metabolic Signatures of Systemic Host Response to Acute and Latent Tuberculosis in Children and Adults

Abstract

Background: Whole blood expression profiling is a mainstay for delineating differential diagnostic signatures of infection yet is subject to high variability that reduces power and complicates clinical usefulness. To date, confirmatory high confidence expression profiling signatures for clinical use remain uncertain. Here we have sought to evaluate the reproducibility and confirmatory nature of differential expression signatures, comprising molecular and cellular pathways, across multiple international clinical observational studies investigating children and adult whole blood transcriptome responses to tuberculosis (TB). Methods and findings: A systematic search and quality control assessment of gene expression repositories for human TB using whole blood resulted in 11 datasets with a total of 1073 patients from Africa, Europe, and South America. A non-parametric estimation of percentage of false prediction was used for meta-analysis of high confidence differential expression analysis. Deconvolution analysis was applied to infer changes in immune cell proportions and enrichment tests applied using pathway database resources. Meta-analysis identified high confidence differentially expressed genes, comprising 372 in adult active-TB versus latent-TB (LTBI), 332 in adult active-TB versus controls (CON), five in LTBI versus CON, and 415 in childhood active-TB versus LTBI. Notably, these confirmatory markers have low representation in published signatures for diagnosing TB. Pathway biology analysis of high confidence gene sets revealed dominant metabolic and innate-immune pathway signatures while suppressed signatures were enriched with adaptive signaling pathways and reduced proportions of T and B cells. Childhood TB showed uniquely strong inflammasome antagonist signature (IL1RN and ILR2), while adult TB patients exhibit a significant preponderance type I and type II IFN markers. Key limitations of the study include the paucity of data on potential confounders. Conclusion: Meta-analysis identified high confidence confirmatory immune-metabolic and cellular expression signatures across studies regardless of the population resource setting, HIV status and circulating endemic pathogens. Notably, previously identified diagnostic signature markers for TB show limited concordance with the confirmatory meta-analysis. Overall, our results support the use of the confirmatory expression signatures for guiding optimized diagnostic, prognostic, and therapeutic monitoring modalities in TB.

Keywords: bioinformatics; immunity; meta-analysis; microarray; systemic responses; tuberculosis.

FIGURE 1

Analysis workflow and exploratory analysis. (A) Summary of the analysis workflow from data acquisition to pathway analysis; a detailed analysis workflow can be found in the Supplementary Material. (B) Hierarchical clustering of the top 1000 most variable gene probes largely clustering active TB from controls (uninfected controls and latently infected individuals). Pairwise comparison of FCs of different datasets can be found in Supplementary Figure S2 indicating the outlier dataset (GSE34608).

FIGURE 2

Meta-analysis identified highly robust confidence genes. (A) Heatmaps showing fold changes of differential expression obtained from meta-analysis using the RankProd and individual dataset specific analysis. (B) Venn diagrams comparing genes obtained from meta-analysis and individual dataset specific analysis to identify high confidence genes. HC, high confidence genes, which are consistently identified by both the meta-analysis and individual dataset specific approaches. TB, tuberculosis; LTBI, latent TB infection; CON, uninfected controls. Fold change cut-off 1.5, p-value < 0.05.

FIGURE 3

Comparison of meta-analysis to identified gene signatures. Genes significantly differentially regulated (fold change > 1.5) from the meta-analysis were compared to gene signatures identified by different studies as shown. (A–C) Comparison of individual diagnostics signature markers to genes from meta-analysis of adult active TB versus LTBI. (D) Comparison of the childhood TB signatures to childhood active TB versus LTBI.

FIGURE 4

Biological pathways enriched in both adult and childhood tuberculosis (active TB versus LTBI). High confidence genes differentially expressed (1.5-fold change and adjusted p-value < 0.05) in both adults and children were analyzed with InnateDB to identify enriched pathways. (A) Pathways downregulated and (B) pathways upregulated in active tuberculosis.

FIGURE 5

Comparisons of adult active TB versus LTBI and controls and childhood versus adult TB. (A) Correlation between adult active TB versus latent TB and active TB versus uninfected controls. (B) Number of genes common between active TB versus latent and active TB versus uninfected controls. (C) Correlation of fold changes obtained in active versus latent TB in children and adults. (D) Overlap of differentially expressed genes in adult and childhood TB.

FIGURE 6

Pathways associated with gene specific to (A) active TB versus latent TB and (B) active TB versus control. Long fold change cut-off at 1.5 and p-value of <0.05.

FIGURE 7

Difference in Immune cell proportions in adult tuberculosis. Gene expression signals were used to deconvolute cell proprtions in whole blood collected from TB and controls using the CellMix package. The median for each cell component per clinical phenotype was determined. Median cell proportions from controls were substracted from medians from active TB. (A) Immune cell proportion difference between active TB and latent TB. (B) Difference in active TB and uninfected controls. Red asterisks indicate those that are siginificant (p < 0.05) based on a Student’s t-test.

FIGURE 8

Immune cell proportions in childhood TB. Immune cell proportions were deconvoluted from whole blood based on gene expression using CellMix. Median cell proportions of latent TB patients were subtracted from the median proportions from active TB for each cell in each study. Red asterisks indicate significant differences based on Student’s t-test.

FIGURE 9

Immune cell signaling molecules modulated in tuberculosis. Significantly regulated immune cells stimulatory and inhibitory molecules obtained from tuberculosis in the meta-analysis (fold change > 1.5 adjusted p-value < 0.05). (A) Molecules modulated in both adults and childhood tuberculosis. (B) Molecules modulated only in childhood tuberculosis. (C) Molecules modulated only in adult tuberculosis.