Impact of Intermediate Hyperglycemia and Diabetes on Immune Dysfunction in Tuberculosis

Abstract

Background: People with diabetes have an increased risk of developing active tuberculosis (TB) and are more likely to have poor TB-treatment outcomes, which may impact on control of TB as the prevalence of diabetes is increasing worldwide. Blood transcriptomes are altered in patients with active TB relative to healthy individuals. The effects of diabetes and intermediate hyperglycemia (IH) on this transcriptomic signature were investigated to enhance understanding of immunological susceptibility in diabetes-TB comorbidity.

Methods: Whole blood samples were collected from active TB patients with diabetes (glycated hemoglobin [HbA1c] ≥6.5%) or IH (HbA1c = 5.7% to <6.5%), TB-only patients, and healthy controls in 4 countries: South Africa, Romania, Indonesia, and Peru. Differential blood gene expression was determined by RNA-seq (n = 249).

Results: Diabetes increased the magnitude of gene expression change in the host transcriptome in TB, notably showing an increase in genes associated with innate inflammatory and decrease in adaptive immune responses. Strikingly, patients with IH and TB exhibited blood transcriptomes much more similar to patients with diabetes-TB than to patients with only TB. Both diabetes-TB and IH-TB patients had a decreased type I interferon response relative to TB-only patients.

Conclusions: Comorbidity in individuals with both TB and diabetes is associated with altered transcriptomes, with an expected enhanced inflammation in the presence of both conditions, but also reduced type I interferon responses in comorbid patients, suggesting an unexpected uncoupling of the TB transcriptome phenotype. These immunological dysfunctions are also present in individuals with IH, showing that altered immunity to TB may also be present in this group. The TB disease outcomes in individuals with IH diagnosed with TB should be investigated further.

Keywords: diabetes; hyperglycemia; inflammation; tuberculosis.

Figure 1.

Differential expression analysis of all the disease phenotypes in South Africa compared to healthy controls before the initiation of tuberculosis (TB) treatment. Gene expression profiles of TB only (n = 11, A), diabetes mellitus (DM) only (n = 33, B), DM-TB (n = 15, C), and intermediate hyperglycemia (IH)–TB (n = 20, D), each relative to healthy controls (n = 24). Genes that were deemed statistically significantly differentially expressed had an adjusted P < .05 after multiple testing correction (Benjamini-Hochberg). Black corresponds to the genes whose expression was significantly changed, and gray shows genes without significant expression change. Abbreviation: FDR, false discovery rate.

Figure 2.

Concordance and discordance of gene expression between the comparisons of each disease group and healthy controls in South Africa. Log fold change (FC) and P value between groups was calculated with R-package DESeq2. A disco.score was calculated for each pair of corresponding genes. The axes show log₂ FC between the conditions indicated by the labels. For example, on the top left plot the x-axis corresponds to the comparison between tuberculosis (TB) and healthy controls (HC), and the y-axis shows the log₂ FC between diabetes mellitus (DM)-TB and HC. Red dots show genes that are significantly different from the controls in the same direction (concordant genes), and blue dots show genes that are significantly different in both comparisons, but in opposite directions. Intensity of color indicates the strength of concordance/discordance as measured by the disco.score.

Figure 3.

Principal component analysis (PCA) of South African participants. The list of all genes that were significantly differentially expressed in any patient group comparison with healthy controls was used in a PCA of all the samples obtained from participants recruited in South Africa. Abbreviations: DM, diabetes mellitus; IH, intermediate hyperglycemia; PC, principal component; TB, tuberculosis.

Figure 4.

Transcriptional modules that were significantly differentially expressed in tuberculosis (TB)-only, diabetes mellitus (DM)-TB, intermediate hyperglycemia (IH)-TB, and DM only compared to healthy controls in South Africa before initiation of TB treatment. Transcripts were evaluated using a preexisting modular framework. Significantly up-regulated (red) and down-regulated (blue) modules are shown: the length of each bar corresponds to the effect size (magnitude of change) of that module, and the color saturation represents the adjusted P value (<.0001). The amount of color represents the proportion of genes within that module that were differentially expressed.

Figure 5.

Differential gene expression analysis of diabetes mellitus (DM)–tuberculosis (TB) and intermediate hyperglycemia (IH)–TB patients relative to TB-only patients, in the combined dataset from all 4 field sites. Samples collected in South Africa, Romania, Peru, and Indonesia from DM-TB (A) patients and IH-TB patients (B) were compared with patients with TB only in a combined analysis. Genes significantly differentially expressed after multiple testing correction are shown in black (P < .05). Genes in gray are not statistically significantly altered compared to patients with untreated TB only. Abbreviation: FDR, false discovery rate.

Figure 6.

Summary of modular analysis in all 4 field sites. The fold changes of the genes within the top significantly differentially expressed modules are shown (adjusted P < .05). Inside: intermediate hyperglycemia–tuberculosis (TB) compared to TB only. Outside: diabetes mellitus–TB compared to TB only. Up-regulated genes are shown in red, and down-regulated genes are in blue. The saturation of color represents the magnitude of differential expression.

Figure 7.

Predictive model of known signature in predictive model of known signature in TANDEM data. Receiver operating characteristic curves are based on a machine learning model generated from 2 different external datasets of transcriptome profiles of patients with tuberculosis and healthy controls (Kaforou et al [29] and Sweeney et al [30] training set). The random forest model was applied to the TANDEM cohort (test set), separately to individuals with and without diabetes mellitus. Abbreviations: AUC, area under the curve; CI, confidence interval; DM, diabetes mellitus; ROC, receiver operating characteristic.