Basic characteristics
The demographics and haematological data of the study participants with their significance values are described in detail in our previous paper28. The median age was 35 years (range 18–63) for DR-TB, 28 years (range 14–49) for DS-TB, 27 years (21–50) for LTB, and 27.5 years (range 18–50) for HCs were not significantly different between recruited individuals.
Drug-resistant tuberculosis is associated with increased levels of chemokines
We wanted to determine the dynamics of chemokines at the different spectra of TB disease and/or infection may therefore be useful as potential biomarker targets for diagnosis. We examined an array of CC and CXC chemokines using multiplex assay profiles in plasma of drug-resistant (DR-TB), drug-sensitive (DS-TB), and LTB and compared them with healthy controls. Chemokine concentration was shown as median and IQR in Table 1. As shown in Fig. 1, DR-TB exhibited significantly increased CC chemokines CCL2 (p = 0.0492), CXC chemokines CXCL9 (p = 0.376) and CXCL10 (p = 0.0317) in comparison to DS-TB.
Further, DR-TB patients exhibited significantly increased levels of CC chemokines CCL1 (p < 0.0001), CCL2 (p = 0.0012), CCL3 (p < 0.0001), CXC chemokines CXCL1 (p < 0.0001), CXCL9 (p < 0.0001) and CXCL10 (p < 0.0001) in comparison to LTB individuals. DR-TB patients exhibited significantly higher levels of CCL1 (p < 0.0001), CCL2 (p < 0.0001), CCL3 (p = 0.0002), CXC chemokines CXCL1 (p < 0.0001), CXCL9 (p < 0.0001), CXCL10 (p < 0.0001) and CXCL11 (p < 0.0001) in comparison to the control group of individuals. DS-TB exhibited significantly higher levels of CCL1 (p = 0.0036), CCL3 (p = 0.0486), CXCL1 (p < 0.0001), CXCL9 (p < 0.0001) and CXCL10 (p < 0.0001) in comparison to individuals with LTB. DS-TB exhibited significantly higher levels of CCL1 (p < 0.0001), CCL2 (p = 0.0054), CXCL1 (p < 0.0001), CXCL9 (p < 0.0001), CXCL10 (p < 0.0001) and CXCL11 (p = 0.0002) compared to the control group of individuals. LTB individuals exhibited significantly increased levels of CXCL9 (p = 0.0130) and CXCL11 (p = 0.0141) in comparison to the control group of individuals. Thus, the clinical spectrum of TB disease/ infection is associated with increased levels of chemokines.
Heatmaps divulge tendencies in the chemokine milieu in DR-TB, DS-TB, LTB, and HC
The trends in the chemokine expression profile were assessed by hierarchical clustering of chemokines using normalized values. For this, the raw individual chemokine expression counts were transformed to log 2 values and normalized with group mean value of HC for respective chemokine across all the samples. The normalized counts were shown in Fig. 2, and the color panel of the heat map reveals the serial increase of chemokines (both numbers and levels) from latency (black or blue) to drug-sensitive (blue, green, and red) and to drug-resistant TB (blue, green, orange, and red). Before infection, the latent condition presented the chemokine panel with near-high levels of CXCL11 and mild or moderate levels of CXCL2, CCL4, CCL2, CCL1, and CXCL9. In the diseased state, DS-TB individuals presented differential chemokine expression with high levels of CXCL1 and CXCL10; near high levels of CCL11, CCL2 and CXCL11; moderate levels of CCL1 and CXCL9 and mild levels of CXCL1, CCL4 and CCL3. DR-TB individuals are associated with abundant chemokine expression with high levels of CXCL1 and CXCL10; near high levels of CXCL2, CCL11, CCL3, CCL2, CCL1, CXCL9 and CXCL11 and with moderate expression of CCL4. Thus, these analyses help to reveal the power of chemokines to demarcate the spectrum of TB disease/infection (DR-TB, DS-TB, and LTB) from HC.
Heatmaps representing the measured chemokines and their hierarchical clustering across the TB disease spectrum by log 2 conversion and HC group mean normalization.
Diagnostic performance of the top chemokines for the bifurcation of DR-TB, DS-TB, LTB and HC
We conducted a ROC analysis of single variables to determine the diagnostic capabilities of each chemokine to distinguish between the study groups. Representative curves showing the chemokines with the best diagnostic precision between and among these groups are shown in Fig. 3, CXCL9 (AUC = 0.82, p < 0.0001) and CXCL10 (AUC = 0.84, p < 0.0001) discriminate DR-TB from DS-TB. Chemokines such as CXCL1 (AUC = 0.80, p < 0.0001), CXCL9 (AUC = 0.98, p < 0.0001) and CXCL10 (AUC = 0.98, p < 0.0001) discriminate DR-TB from LTB. Additionally, the chemokines CCL1 (AUC = 0.88, p < 0.0001), CCL2 (AUC = 0.88, p < 0.0001), CXCL1 (AUC = 0.87, p < 0.0001), CXCL9 (AUC = 1, p < 0.0001), CXCL10 (AUC = 1, p < 0.0001) and CXCL11 (AUC = 0.82, p < 0.0001) discriminate DR-TB from HC group. CXCL1 (AUC = 0.85, p < 0.0001), CXCL9 (AUC = 0.92, p < 0.0001) and CXCL10 (AUC = 0.94, p < 0.0001) discriminate DS-TB from individuals with LTB. CCL1 (AUC = 0.85, p < 0.0001), CXCL1 (AUC = 0.91, p < 0.0001), CXCL9 (AUC = 1, p < 0.0001) and CXCL10 (AUC = 0.99, p < 0.0001) discriminate DS-TB from the HC group of individuals. CXCL9 (AUC = 0.8503, p < 0.0001) discriminate the LTB individuals from the HC group of individuals. However, other chemokines CCL3, CCL4, CCL11 and CXCL2 could not significantly discriminate DR-TB from DS-TB, LTB, and the control group.
ROC curves of significant chemokines with AUC > 0.8 showing the diagnostic efficiency between the study groups, (a) HC vs LTB, (b) LTB vs DS-TB, (c) DS-TB vs DR-TB, (d) HC vs DS-TB, (e) HC vs DR-TB and (f) LTB vs DR-TB.
In addition, we performed a random forest (RF) analysis to understand the importance of these chemokines and their distinguishing ability toward the separation of study groups. According to the order of importance, RF plots of overall comparison (HC vs LTB vs DS-TB vs DR-TB) presented CXCL9, CXCL10, and CXCL1 as the top classifiers (Fig. 4A). This was in accordance with the ROC results where these chemokines displayed higher AUC values of above 0.8. Similarly in the subgroup comparisons, the same CXCL9 was obtained as the top classifier for HC vs LTB/DS-TB/DR-TB (Fig. 5A-1–A-3) whereas, CXCL10 for LTB vs DS-TB/DR-TB (Fig. 5A-4,A-5) and DS-TB vs DR-TB (Fig. 5A-6).
Random-forest analysis plot (A) and principal component analysis plot (B) of top 3 chemokines across the study groups (HC vs LTB vs DS-TB vs DR-TB).
Sub-group comparisons of top 3 chemokines by random-forest analysis (A1 HC vs LTB, A2 HC vs DS-TB, A3 HC vs DR-TB, A4 LTB vs DS-TB, A5 LTB vs DR-TB and A6 DS-TB vs DR-TB) and principal component analysis (B1 HC vs LTB, B2 HC vs DS-TB, B3 HC vs DR-TB, B4 LTB vs DS-TB, B5 LTB vs DR-TB and B6 DS-TB vs DR-TB).
All chemokine variables were then dimensionally reduced through the principal component analysis, resulting in a lower variation of the first two dimensions, and the ellipses of HC overlapped within LTB and DS-TB within DR-TB. To achieve better bifurcation with a minimum of 80% variance, the weaker chemokines from the RF plots were removed and the dimensionality reduction analysis was repeated. The PCA of the top 3 chemokines (CXCL9, CXCL10 and CXCL1) exhibited better separation of clusters (LTB, DS-TB, and DR-TB) with variances above 80% (Fig. 4B). However, HC overlapped completely within the LTB cluster. The discriminative accuracy and the ranges are as follows: 80% (73–85.9%) for HC vs LTB vs DS-TB vs DR-TB and for subgroup comparison with a descending accuracy order of 100% (95.5–100%) for HC vs DR-TB; 98.8% (93.2–100%) for LTB vs DR-TB and HC vs DS-TB; 95% (87.7–98.6%) for LTB vs DS-TB; 86.3% (76.7–92.9%) for DS-TB vs DR-TB and the least 76.3% (65.4–85.1%) for HC vs LTB (Fig. 5B-1–B-6).
Correlation between plasma chemokines and spectrum of TB infection/disease
For each pair of chemokines, Spearman’s rank correlation coefficient was calculated to determine the strength of association. There was a positive moderate correlation between the following pairs of chemokines in DR-TB: CXCL10 and CCL3 (r = 0.4, p = 0.0176), CXCL11 and CCL3 (r = 0.4, p = 0.0111), CCL2 and CCL1 (r = 0.3, p = 0.0388); DS-TB: CXCL1 and CCL1 (r = 0.5, p = 0.0024), CXCL10 and CXCL11 (r = 0.5, p = 0.0014), CXCL1 and CXCL2 (r = 0.4, p = 0.0112); LTB: CCL4 and CXCL2 (r = 0.5, p = 0.0027), CCL11 and CCL3 (r = 0.4, p = 0.0050); and HC: CCL4 and CXCL2 (r = 0.5, p = 0.0018), CXCL9 and CCL1 (r = 0.4, p = 0.0143), CXCL11 and CXCL1 (r = 0.4, p = 0.0195), CCL11 and CCL3 (r = 0.3, p = 0.0381). Whereas, there was a negative moderate correlation between the following pairs of chemokines in DR-TB: CXCL1 and CCL2 (r = − 0.4, p = 0.0136), CCL4 and CXCL2 (r = − 0.3, p = 0.0478); DS-TB: CXCL11 and CXCL1 (r = − 0.5, p = 0.0004), CXCL11 and CCL1 (r = − 0.4, p = 0.0092), CXCL9 and CCL3 (r = − 0.4, p = 0.0095), CCL11 and CCL2 (r = − 0.4, p = 0.0119); LTB: CXCL9 and CXCL10 (r = − 0.4, p = 0.0039), CXCL11 and CXCL10 (r = − 0.4, p = 0.0159), CCL4 and CXCL10 (r = − 0.4, p = 0.022), CCL4 and CCL2 (r = − 0.4, p = 0.0205); and HC: CXCL11 and CCL4 (r = − 0.6, p = 0.0001), CXCL2 and CCL11 (r = − 0.3, p = 0.0448) shown in Fig. 6a–d.
Correlation matrix using spearman rank correlation between the measured chemokines of the study groups, (a) DR-TB, (b) DS-TB, (c) LTB and (d) HC.