A combination of baseline plasma immune markers can predict therapeutic response in multidrug resistant tuberculosis

Objective To identify plasma markers predictive of therapeutic response in patients with multidrug resistant tuberculosis (MDR-TB). Methods Fifty HIV-negative patients with active pulmonary MDR-TB were analysed for six soluble analytes in plasma at the time of initiating treatment (baseline) and over six months thereafter. Patients were identified as sputum culture positive or negative at baseline. Culture positive patients were further stratified by the median time to sputum culture conversion (SCC) as fast responders (< 76 days) or slow responders (≥ 76 days). Chest X-ray scores, body mass index, and sputum smear microscopy results were obtained at baseline. Results Unsupervised hierarchical clustering revealed that baseline plasma levels of IP-10/CXCL10, VEGF-A, SAA and CRP could distinguish sputum culture and cavitation status of patients. Among patients who were culture positive at baseline, there were significant positive correlations between plasma levels of CRP, SAA, VEGF-A, sIL-2Rα/CD40, and IP-10 and delayed SCC. Using linear discriminant analysis (LDA) and Receiver Operating Curves (ROC), we showed that a combination of MCP-1/CCL2, IP-10, sIL-2Rα, SAA, CRP and AFB smear could distinguish fast from slow responders and were predictive of delayed SCC with high sensitivity and specificity. Conclusion Plasma levels of specific chemokines and inflammatory markers measured before MDR-TB treatment are candidate predictive markers of delayed SCC. These findings require validation in a larger study.


Methods
Fifty HIV-negative patients with active pulmonary MDR-TB were analysed for six soluble analytes in plasma at the time of initiating treatment (baseline) and over six months thereafter. Patients were identified as sputum culture positive or negative at baseline. Culture positive patients were further stratified by the median time to sputum culture conversion (SCC) as fast responders (< 76 days) or slow responders (! 76 days). Chest X-ray scores, body mass index, and sputum smear microscopy results were obtained at baseline.

Results
Unsupervised hierarchical clustering revealed that baseline plasma levels of IP-10/ CXCL10, VEGF-A, SAA and CRP could distinguish sputum culture and cavitation status of patients. Among patients who were culture positive at baseline, there were significant positive correlations between plasma levels of CRP, SAA, VEGF-A, sIL-2Rα/CD40, and IP-10 and delayed SCC. Using linear discriminant analysis (LDA) and Receiver Operating Curves (ROC), we showed that a combination of MCP-1/CCL2, IP-10, sIL-2Rα, SAA, CRP and AFB smear could distinguish fast from slow responders and were predictive of delayed SCC with high sensitivity and specificity. PLOS  Introduction promising candidates that were predictive of microbiological outcome in 50 HIV-negative MDR-TB patients. The following markers were analyzed as potential biomarkers of response to treatment, based on previously reported associations in DS-TB patients: interferon gammainduced protein 10 (IP-10)/CXCL10, vascular endothelial growth factor A (VEGF-A), soluble interleukin-2 receptor alpha (sIL-2Rα)/CD25, C-reactive protein (CRP), serum amyloid A (SAA) and monocyte chemoattractant protein 1 (MCP-1)/CCL2 [11, [20][21][22][23]. Drug susceptibility testing and bacteriological response to treatment Sputum culture and AFB smear microscopy were performed at the time of admission and monthly thereafter, as part of routine monitoring. Sputum culture conversion (SCC) was defined as two consecutive TB culture negative results, separated by at least 30 days, with no subsequent culture positive results. Time to SCC was defined as the interval between the date of treatment initiation and the date of collection of the sputum specimen that yielded the first negative culture result. Patients who were sputum culture positive at enrollment were classified, based on the median number of days to achieve SCC in this cohort, as either slow responders (! 76 days) or fast responders (< 76 days) for analysis (Table 1). Drug susceptibility testing (DST) (Bactec MGIT 960) was performed for all culture positive MDR-TB patients within 30 days of starting treatment. DST results, obtained from patient records, were available for the following drugs: isoniazid, rifampin, ethambutol, ethionamide, streptomycin, kanamycin and ofloxacin. As pyrazinamide DST is not routinely performed, these results were only available from 33 patients (14/19 fast; 19/22 slow responders). DST was not done for any other drugs, including clofazimine. Effective and potentially effective drugs in individual patient regimens were defined as those with known susceptibility only and those with no known resistance, respectively [24]. DST results and TB drug regimens of individual patients are shown in S1 Table. obtained from patient records. Cavitary lesions were present in 30 (65%) of the 46 patients with available chest X-ray reviews, of whom 22 (73.3%) had bilateral lung involvement (Table 1). Of the 8 sputum culture negative patients with chest X-ray reviews available, none showed signs of cavitary disease.

Luminex multiplex immunoassay
Blood samples were collected on the day after patients started MDR-TB therapy and were enrolled into the study and thereafter at 2, 4, and 6 months of treatment. Peripheral blood was obtained in sodium heparin Vacutainer tubes (BD Biosciences, San Diego, CA); plasma was prepared within 45 minutes of collection and stored at −80˚C for later analysis. Stored plasma samples were rapidly thawed and analyzed for the following markers: sIL-2Rα/CD25, VEGF-

Statistical analysis
Distribution of variables was assessed by Shapiro-Wilk and D'Agostino & Pearson tests. Heat map and clustering of significant differences were analyzed using Qlucore Omics Explorer 3.1 with interface to R (Qlucore AB, Lund, Sweden). Analyte levels were log 2 converted and standardized by subtraction of its mean value and division by its standard deviation across all samples. For concentrations that were outside the range of detection, the lowest detected level (LDL) or highest detected level (HDL) values were used to replace the out-of-range values prior to log 2 conversion. To identify markers with expression levels that differed at baseline and over time between unpaired groups, the Mann-Whitney U test was used. All values were corrected for false discovery rate (FDR) using a q-value of 0.05. To analyze longitudinal changes in parameters during follow-up, the Wilcoxon signed rank test was performed. The relationship between the concentration of baseline plasma analytes and time to SCC was assessed by Pearson or Spearman rank correlations. Receiver Operating Characteristic (ROC) analysis was used to assess the predictive nature of each analyte with time to SCC. Results were considered significant if the 95% confidence interval (CI) of the area under the curve (AUC) exceeded 0.70. All possible combinations of host analytes were evaluated using linear discriminant analysis (LDA) to find the subset of analytes which best predicted fast from slow responders. Fisher's exact test was used to compare drug resistance profiles and treatment regimens between the slow and fast responder groups.

Associations between clinical characteristics and microbiology
When we examined the associations between clinical and microbiological data (Table 1), a strong association existed between the extent of lung pathology, based on chest X-ray scores, and sputum smear AFB at baseline (Spearman's correlation coefficient r = 0.4811, p = 0.0008), where advanced radiological findings were positively correlated with higher smear grades. Likewise, patients who were culture positive at baseline were more likely to have pulmonary cavities than patients who were culture negative (p = 0.0001). Baseline chest X-ray scores also positively correlated with the time to SCC (Spearman's correlation coefficient r = 0.5383, p = 0.0002), and there was a strong association between time to SCC and baseline smear microscopy grades (Spearman's correlation coefficient r = 0.7293, p<0.0001). Collectively, these data show that disease severity (as indicated by chest X-ray scores) was directly associated with bacillary load.

Plasma levels of analytes reflect sputum culture status
We first interrogated how levels of plasma markers clustered with baseline sputum AFB smear and culture status. Fig 1A shows unsupervised hierarchical clustering of the 6 analytes at baseline for both sputum smear and culture. After correcting for FDR (q), we found significantly higher levels of CRP (q = 5.8355x10 -5 ), SAA (q = 0.0001), IP-10 (q = 0.0038) and VEGF-A (q = 0.0224) in patients who were smear positive compared to smear negative. The inverse was found for MCP-1 levels, which were lower in smear positive cases (q = 0.0153). No significant differences were found for sIL-2Rα levels, although there was a trend towards lower levels in smear positive compared to smear negative cases (q = 0.2921). The distinct grouping of these plasma analytes with sputum smear status is reflected in the principal component analysis (PCA) plot (Fig 1B). Similarly, higher levels of CRP (q = 0.0005), SAA (q = 0.0005), IP-10 (q = 0.0086), and VEGF-A (q = 0.0368) and lower levels of MCP-1 (q = 0.0406) were found in plasma from TB culture positive compared to culture negative patients (Fig 1A), and these two patient groups were distinguished by the markers (Fig 1C).
As sputum culture positivity was strongly associated with cavitation (Fig 1A), it was not surprising to find a similar pattern of soluble markers co-clustered with both cavitation and baseline culture status. Higher levels of CRP (q = 0.0003), SAA (q = 0.0007), IP-10 (q = 0.0028) and VEGF-A (q = 0.0203) were associated with lung cavitation, while MCP-1 and sIL-2Rα showed no significant association. Fig 1D shows the PCA plot reflecting this distinct grouping between soluble markers and cavitation. Thus, levels of plasma IP-10, VEGF-A, SAA and CRP were able to discriminate between sputum culture positive and negative cases and between the presence and absence of lung cavitation at baseline.

Baseline soluble plasma analytes as predictors of delayed sputum culture conversion (SCC)
A wide distribution in days to achieve SCC was noted among the MDR-TB patients who were sputum culture positive at baseline (Fig 2A). The median plasma levels of all markers, measured at baseline, were significantly different between fast and slow responders (Table 2). Moreover, there were significant positive correlations between delayed SCC and baseline levels of SAA, CRP, VEGF-A, sIL-2Rα and IP-10 ( Fig 2B, 2C, 2D, 2E and 2F, respectively), while the combination of these markers showed a trend towards distinguishing fast and slow responder groups (Fig 2G). There was no significant association of MCP-1 with SCC (not shown).
We next asked whether any of the analytes could be useful predictors of delayed SCC. Baseline levels of single markers showed significant area under the curve (AUC), although with variably low sensitivity or specificity (S1 Table). However, when plasma levels of MCP-1 (CCL2), sIL-2Rα and SAA were combined, there was a significant area under the curve (AUC) of 83% (p = 0.0003) with 82% sensitivity and 79% specificity, and an optimal misclassification error rate of 19.5% (Fig 3A). A combination of MCP-1 (CCL2), IP-10 (CXCL10), sIL-2Rα, SAA and CRP, together with baseline smear status, yielded an AUC of 86% (p = 0.0001) with 86% sensitivity and 83% specificity, and an optimal misclassification error rate of 15.3% (Fig 3C). Moreover, slow responders had significantly greater LDA scores for both the 3-variable and the 6-variable combinations (Fig 3B and 3D), indicating that higher levels of these markers at the start of treatment were associated with delayed time to bacillary clearance.
When we examined the impact of treatment on these plasma analytes over 6 months of chemotherapy, both fast and slow responders showed significant reductions in levels of IP-10, SAA, CRP and sIL-2Rα at month 2, with a further decline of SAA in slow responders to month 4 (Fig 4). There was a downward trend in plasma levels of these four markers over time in slow responders, but in fast responders only SAA and CRP showed a decline over 6 months. Plasma levels of sIL-2Rα remained significantly higher in slow, compared to fast, responders until month 6. VEGF-A levels were higher in slow responders over the first two months and showed a significant drop from baseline levels to month 6 (p = 0.0266). Thus, effective treatment of MDR-TB resulted in reduced levels of the markers over time, with some mediators responding at different rates between fast and slow responders. As a result, the predictive nature of these analytes to time to SCC was lost by month 2 of chemotherapy, most notably for SAA and CRP (data not shown).  MDR-TB regimens including a greater number of effective drugs (i.e. no known resistance) have been correlated with improved response to therapy [24], while fluoroquinolones (FQN), pyrazinamide (PZA) and clofazimine (CFZ) are known to accelerate bacillary clearance. To examine the possibility that treatment differences between the two patient groups may have been responsible for the observed differential SCC, we compared drug regimens for fast and slow responders (Table 1). There were no significant differences in the average number of effective or potentially effective drugs (p = 0.1335; p = 0.3724, respectively) nor in the cumulative months of drugs known to accelerate bacillary clearance, pyrazinamide, clofazimine and fluoroquinolone (p = 0.4778), included in the regimens between the fast and slow responders. Plasma markers of treatment response in MDR-TB Whatever the mechanism accounting for the rapidity of bacterial clearance, baseline plasma analytes were able to predict the rate of this occurring.

Discussion
The goal of this study was to identify plasma markers that are predictive of response to treatment in MDR-TB patients. Although similar studies have been carried out in patients initiating conventional treatment for DS-TB [20,21,25], there have been few reports on early markers of response to treatment among MDR-TB patients. In this study, we examined the potential of six plasma markers, measured at the start of second-line drug treatment to predict the time to sputum culture conversion (SCC) in MDR-TB patients.
CRP and SAA are acute phase proteins produced by the liver and released during inflammation, functioning as opsonins and in recruitment of cells to inflammatory sites, respectively [26]. Serum CRP has demonstrated value in diagnosis of active TB in both HIV-positive [27][28][29] and HIV-negative patients [30,31]. Positive correlations have been noted between circulating CRP levels and cavitation/impaired lung function [32][33][34][35] and sputum culture status [36]. High CRP levels have been associated with delayed SCC [10, 37], and a decline in CRP was correlated with clinical response to treatment, although CRP alone is a poor predictor of Plasma markers of treatment response in MDR-TB outcome due to its low specificity [38]. Serum levels of SAA were similarly elevated in TB patients compared to healthy controls [30,31] and have been associated with sputum culture status and response to TB treatment [36,39]. Consistent with these reports, baseline levels of CRP and SAA correlated with both cavitation and sputum culture status in our study cohort, and both acute phase markers declined with treatment.
VEGF is a key mediator of angiogenesis and lymphangiogenesis in granulomas and plays an important role in TB pathogenesis [40]. In the present study, we found a significant positive correlation between baseline levels of VEGF-A and delayed response to MDR-TB treatment. In addition, VEGF-A levels were positively associated with both cavitation and positive TB culture, suggesting that this marker may be a surrogate measure of disease severity. Elevated levels of VEGF have been found in serum and saliva from patients with DS-TB compared to healthy controls or individuals with latent TB infection (LTBI) [30,[41][42][43] and have been positively correlated with bilateral and/or cavitary TB and sputum AFB smear grade, as a measure of bacillary burden [40], similar to our observations. In a recent study, treatment with anti-VEGF antibody was found to normalize vascularization, decrease hypoxia and reduce granuloma size in a rabbit model of pulmonary TB, supporting a functional role for this molecule in TB pathogenesis [44]. In our MDR cohort, plasma levels of VEGF-A did not decline in fast responders over the first 6 months of treatment, while slow responders showed a significant drop in levels only after 5 months of chemotherapy. In contrast, several studies have reported a decrease in circulating VEGF levels in association with successful treatment of DS-TB [11,40,41,45] and tubercular meningitis [46]. Persistently elevated VEGF-A in our MDR-TB cohort may be an indication of delayed pulmonary recovery despite microbiological response to treatment. However, follow up chest X-rays were not available, so we are not able to confirm this possibility.
Soluble IL-2Rα is involved in T cell homeostasis, and elevated levels in blood are considered to be an indication of sustained immune activation [47]. In our cohort, higher levels of sIL-2Rα in plasma at baseline were significantly associated with delayed SCC. Similar results were described from a recent study of patients with DS-TB, although the trend of higher sIL-2Rα levels in slow versus fast responders did not reach significance [21]. Other studies have reported elevated serum levels of sIL-2Rα in TB patients compared to healthy controls, which declined with successful treatment [48][49][50][51][52][53]. In contrast, the patients in our MDR cohort showed persistently high plasma levels of sIL-2Rα over 6 months of therapy, which may reflect continuing immune activation despite bacillary clearance. Interestingly, sIL-2Rα has been shown to impair T cell proliferation and to induce differentiation of Treg cells in vitro [47]. Although we did not examine T cell subsets in this study, it is possible that the persistently high sIL-2Rα levels in blood observed may have contributed to elevated Treg cells in these MDR-TB patients, as has been previously reported [15,16].
Serum levels of the chemokine IP-10 (CXCL10) have demonstrated diagnostic value in differentiating between active TB and LTBI and to decline with treatment [11,20,54,55]. In contrast to these reports, despite significant differences in IP-10 levels between fast and slow responders in our study, baseline levels of IP-10 alone showed relatively low sensitivity as a predictor of response to treatment. While it is difficult to make comparisons between absolute levels of plasma markers in different studies, it is interesting to note that IP-10 levels in plasma from our MDR patients described here were considerably lower than those reported in similar studies conducted in DS-TB patients [11,20,30,54]. One possible explanation for these discrepancies may be related to the suppression of Th1 responses previously reported in association with MDR-TB [12,13,56,57]. A large proportion (92%) of the patients in our study had a previous history of treatment for TB. Repeated exposure to TB antigens in re-treatment cases may be responsible for the characteristics of immune suppression seen in MDR-TB patients, who often have a prior history of disease. This idea is supported by a recent study of DS patients showing reduced responsiveness to TB antigens of PBMCs from re-treatment as compared to new cases, including lower production of IP-10 [58]. Collectively, the differential kinetics of decline in these plasma markers between slow and fast response to drug regimens resulted in loss of predicting time to SCC when analyzed after 2 months of treatment.
It is important to note that the present study is a pilot involving a small number of patients, which requires validation through a larger cohort study. Other limitations include the lack of DST results for clofazimine and pyrazinamide.
In conclusion, this study demonstrates that a combination of chemokines and markers of acute inflammation can be used as predictors of microbiological outcome in MDR-TB patients and have utility as potential early indicators in clinical trials and in guiding therapy for patients who may be at elevated risk of TB treatment failure.
Supporting information S1 Table. Fast and slow responder DST results and treatment regimens.