Distinguish active tuberculosis with an immune-related signature and molecule subtypes: a multi-cohort analysis

Abstract

Background: Distinguishing latent tuberculosis infection (LTBI) from active tuberculosis (ATB) is very important. This study aims to analyze cases from multiple cohorts and get the signature that can distinguish LTBI from ATB.

Methods: Thirteen datasets were downloaded from the gene expression omnibus (GEO) database. Three datasets were selected as discovery datasets, and the hub genes were discovered through WGCNA. In the training cohort, we use machine learning to establish the signature, verify the authentication ability of the signature in the remaining datasets, and compare it with other signatures. Cluster analysis was carried out on ATB cases, immune cell infiltration analysis, GSVA analysis, and drug sensitivity analysis were carried out on different clusters.

Results: In the discovery datasets, we discovered five hub genes. A signature (SLC26A8, ANKRD22, and FCGR1B) is obtained in the training cohort. In the total cohort, the three-gene signature can separate LTBI from ATB (the total area under ROC curve (AUC) is 0.801, 95% CI 0.771-0.830). Compared with other author's signatures, our signature shows good identification ability. Immunological analysis showed that SLC26A8, ANKRD22, and FCGR1B were closely related to the infiltration of immune cells. According to the expression of the three genes, ATB can be divided into two clusters, which are different in immune cell infiltration analysis, gene set variation, and drug sensitivity.

Conclusion: Our study produced an immune-related three-gene signature to distinguish LTBI from ATB, which may help us to manage and treat tuberculosis patients.

Keywords: Active tuberculosis; Immune infiltration; Latent tuberculosis infection; Molecule subtypes; Signature.

Fig. 1

Flow chart of finding immune-related signature through multi-cohort study.

Fig. 2

Evaluating the module-trait association through the correlation between the module characteristic genes and the sample traits in GSE37250 (A), GSE28623 (C), and GSE19141 (E) dataset. B. The gene scatter plot in the red module in the GSE37250 dataset. D. The gene scatter plot in the yellow module in the GSE28623 dataset. F. The gene scatter plot in the brown module in the GSE19141 dataset. G. The intersection gene of important genes in three datasets. H. The expression of hub genes in LTBI and ATB cases in the training cohort. I. The expression of hub genes in LTBI and ATB cases in testing cohort. J. ROC curve analyses for SLC26A8, CD274, FCGR1B, SERPING1, and ANKRD22 in the training cohort. K. Residual distribution of each machine learning model. L. ROC analysis of four machine learning models in the training cohort.

Fig. 3

To verify the diagnostic value of the three-gene signature by ROC analysis in the training set (A), testing cohort (B), and total cohort (C). D-O. In the total cohort, the ability of other signatures to distinguish LTBI from ATB.

Fig. 4

A. The AUC curve of the three-gene signature in HIV positive cases. B. Difference of immune cell infiltration between LTBI and ATB. C. Correlation analysis between SLC26A8, ANKRD22, FCGR1B, and infiltrated immune cells. D. The cluster number is most stable when the k value is set to 2. E. PCA showed significant differences between the two clusters. F. SLC26A8, ANKRD22, and FCGR1B expression in C1 and C2 cases in total cohort. G. Difference of immune cell infiltration between C1 and C2. H. The difference in biological function between C1 and C2 was analyzed by the GSVA method.

Fig. 5

Drug sensitivity analysis. A-E showed that the drug had a lower IC₅₀ value in the C2 group. F-J showed that the drug had a lower IC₅₀ value in the C1 group.