Blood transcriptomic markers of Mycobacterium tuberculosis load in sputum

Abstract

Background: Peripheral blood transcriptome signatures that distinguish active pulmonary tuberculosis (TB) from control groups have been reported, but correlations of these signatures with sputum mycobacterial load are incompletely defined.

Methods: We assessed the performance of published TB transcriptomic signatures in Haiti, and identified transcriptomic biomarkers of TB bacterial load in sputum as measured by Xpert® MTB/RIF molecular testing. People in Port au Prince, Haiti, with untreated pulmonary TB (n = 51) formed the study cohort: 19 people with low and 32 with high sputum Mycobacterium tuberculosis load. Peripheral whole blood transcriptomes were generated using RNA sequencing.

Results: Twenty of the differentially expressed transcripts in TB vs. no TB were differentially expressed in people with low vs. high sputum mycobacterial loads. The difference between low and high bacterial load groups was independent of radiographic severity. In a published data set of transcriptomic response to anti-tuberculosis treatment, this 20-gene subset was more treatment-responsive at 6 months than the full active TB signature.

Conclusion: We identified genes whose transcript levels in the blood distinguish active TB with high vs. low M. tuberculosis loads in the sputum. These transcripts may reveal mechanisms of mycobacterial control of M. tuberculosis during active infection, as well as identifying potential biomarkers for bacterial response to anti-tuberculosis treatment.

Figure 1.

Performance of two published active TB transcriptomic signatures in a Haitian population. A) The 86-gene signature reported in Berry et al was applied to the cohort of active TB cases (grey) or controls (blue) by unsupervised hierarchical clustering. B) The active TB signature reported by Zak et al , was applied to the active TB cohort and controls.

Figure 2.

Discriminatory power of the TB three gene score reported by Sweeney et al in people with TB in Haiti. A) Unsupervised hierarchical clustering of expression values for each of the three genes in the score for TB cases (grey) and controls (blue). B) Violin plots of the calculated three gene score in control and active TB cases. The internal box is the interquartile range, with the white dot being the median. Whiskers extend to 1.5 times the interquartile range or the maximum/minimum, whichever is closer to the median. The p-value for the difference in scores is 6.6×10^-16. KLF2 = kruppel like factor 2, DUSP3 = dual specificity phosphatase 3, GBP5= guanylate binding protein 5

Figure 3.

Discriminatory power of the TB four gene score reported by Maertzdorf et al in people with TB in Haiti. A) Unsupervised hierarchical clustering of expression values for each of the four genes in the score for TB cases (grey) and controls (blue). B) Violin plots of the calculated four gene score in control and active TB cases. The internal box is the interquartile range, with the white dot being the median. Whiskers extend to 1.5 times the interquartile range or the maximum/minimum, whichever is closer to the median. The p-value for the difference in scores is 5.6×10^-12. P2RY14=purinergic receptor P2Y14, GBP1 = guanylase binding protein 1, IFITM3 = interferon-induced transmembrane protein 3, ID3 = inhibitor of DNA binding 3.

Figure 4.

Differences in transcript counts in peripheral blood for three representative genes found to be related to sputum M. tuberculosis load as determined by GeneXpert testing. No TB, low M. tb, and high M. tb groups were compared using ANOVA with adjustment for multiple comparisons (Figure 4A). P-values for the differences in groups for Suppressor of cytokine signaling 3 (SOCS3) (p=2.5×10⁻¹¹), complement C3b/C4b receptor 1 (CR1) (p=1.99×10⁻¹¹), and oncostatin M (OSM) (p=2.41×10⁻¹¹) were statistically significant. There were no significant differences in low sputum mycobacterial load vs. high sputum mycobacterial load for the Maertzdorf et al 4-gene signature score (Figure 4B, p=0.1087) or the Sweeney et al 3-gene signature score (Figure 4C, p=0.1662).

Figure 5.

Boxplots showing the post-treatment change in gene expression for the Berry (light grey) and the bacterial load 20-gene signature (dark grey) at each time point after starting TB treatment. There was a significant difference in the expression change from baseline at the 2 month, 6 month, and 12 month time points. The box is the interquartile range, with the horizontal line being the median. Whiskers extend to 1.5 times the interquartile range or the maximum/minimum, whichever is closer to the median.