Identification of biomarkers for Mycobacterium tuberculosis infection ...

Genes and Immunity (2013) 14, 356–364 & 2013 Macmillan Publishers Limited All rights reserved 1466-4879/13 www.nature.com/gene

ORIGINAL ARTICLE

Identification of biomarkers for Mycobacterium tuberculosis infection and disease in BCG-vaccinated young children in Southern India S Dhanasekaran1, S Jenum2, R Stavrum1, C Ritz3, D Faurholt-Jepsen3, J Kenneth4, M Vaz5, HMS Grewal1,6, TM Doherty1,7 and TB Trials Study Group8 Pediatric tuberculosis (TB) often goes undiagnosed because of the lack of reliable diagnostic methods. With the aim of assessing biomarker(s) that can aid in the diagnosis of TB infection and disease, we investigated 746 Indian children with suspected TB. Whole-blood mRNA from 210 children was examined by dual-color Reverse-Transcriptase Multiple Ligation-dependent ProbeAmplification for the expression of 45 genes and a Bio-Plex assay for the expression of cytokines/chemokines in QuantiFERON supernatants. The study shows that transcription of SEC14L1, GUSB, BPI, CCR7 and TGFb-1 (all Pp0.05) was downregulated in TB disease compared with uninfected controls, while transcription of RAB33A was downregulated in TB disease compared with both latent TB (Po0.05) and controls (Po0.01). The transcription of CD4, TGFb-1 (Po0.01) and the expression of IL-2 (Po0.01) and IL-13 (Po0.05) was upregulated in latent TB compared with that in controls. Using the Least Absolute Shrinkage and Selection Operator (lasso) model, RAB33A alone discriminated between TB disease and latent TB (area under the curve (AUC) 77.5%), whereas a combination of RAB33A, CXCL10, SEC14L1, FOXP3 and TNFRSF1A was effective in discriminating between TB disease and controls (AUC 91.7%). A combination of 11 biomarkers predicted latent TB with moderate discriminatory power (AUC 72.2%). In conclusion, RAB33A is a potential biomarker for TB disease, whereas CD4, TGFb-1 and IL-2, IL-13 may identify latent TB in children. Genes and Immunity (2013) 14, 356–364; doi:10.1038/gene.2013.26; published online 16 May 2013 Keywords: TB infection; TB disease; dcRT-MLPA; Bio-plex assay; biomarkers; cytokines/chemokines

INTRODUCTION Tuberculosis (TB) constitutes a serious global health problem. In 2011, there were an estimated 8.7 million incident cases and 1.0 million deaths and India had the largest number (2.0–2.5 million) of incident cases reported globally.1 The recent WHO global TB report estimated 0.5 million cases and 64 000 deaths among (HIVnegative) children o15 years, contributing to B6% of the disease burden globally (in 2011).1 The immune system in infants is immature and gradually evolves throughout childhood,2–4 affecting the ability to cope with infections. Thus, children aged o5 years have an increased risk of progression to TB disease following Mycobacterium tuberculosis infection and develop more severe disease.5–7 The diagnosis of TB in children is traditionally based on chest radiography, tuberculin skin testing (TST) and mycobacterial staining/culture, even if these investigations may not always be positive in children with TB.8 Commercially available interferongamma release assays (IGRAs) are being used in the diagnosis of M. tuberculosis infection in adults and children. The sensitivity of IGRAs and TST for TB disease is similar, and a reduced IGRA sensitivity has been found in high-burden compared with that in low-burden TB settings.9 However, IGRAs have increased specificity compared with TST as they are not compromised by 1

exposure to non-tuberculous mycobacteria (NTM) or Bacillus Calmette-Guerin (BCG) vaccination. However, WHO does not support the use of IGRAs in children because of the poor performance of the tests in low- and middle-income countries.10 Moreover, the 2012 WHO global TB report states that neither TST nor IGRAs can discriminate between latent TB infection and TB disease.1 Therefore, development of tests that accurately distinguish between M. tuberculosis infection and disease are essential.11 One of the reasons for the intensive search for diagnostic and predictive biomarkers for TB is the lack of a gold standard for the diagnosis of M. tuberculosis infection and disease in children.12 In attempts to discriminate between latent TB infection and TB disease, studies looking for potential biomarker(s) for TB in different populations, as well as studies looking at interactions between M. tuberculosis and its host, have been conducted. However, few studies have been conducted in young children.13–15 The present study is nested within a prospective childhood cohort study, which was set up by the TB Trials Study Group in a typical rural population in Southern India. A childhood cohort study was designed to prepare a field site for future vaccine trials and establish the true incidence of TB disease among children followed up within 15 days of birth to 2 years of age. As part of

Department of Clinical Science, Faculty of Medicine and Dentistry, University of Bergen, Bergen, Norway; 2Center for Immune Regulation, Rikshospitalet–Radium Hospitalet Medical Centre, University of Oslo, Oslo, Norway; 3Department of Nutrition, Exercise and Sports, University of Copenhagen, Copenhagen, Denmark; 4Division of Infectious Diseases, St John’s Research Institute, Bangalore, India; 5Division of Health and Humanities, St John’s Research Institute, Bangalore, India; 6Department of Microbiology, Haukeland University Hospital, University of Bergen, Bergen, Norway; 7GlaxoSmithKline Pharma, Vaccines, Copenhagen, Denmark and 8The members of the TB Trials Study Group who participated are listed at the end of the article. Correspondence: Professor HMS Grewal, Department of Clinical Science, Faculty of Medicine and Dentistry, University of Bergen, Bergen 5021, Norway. E-mail: [email protected] Received 19 December 2012; revised 4 March 2013; accepted 1 April 2013; published online 16 May 2013

Biomarkers for TB infection and disease S Dhanasekaran et al

357 the childhood cohort study, the present study was set up to look for potential biomarkers for M. tuberculosis infection and disease in young children referred from the cohort for a diagnostic work up for TB during the follow-up period. Accordingly, in appropriately selected samples from these children, we have assessed the transcriptional levels of a predefined panel of genes of interest using novel high-throughput technology (dual-color ReverseTranscriptase Multiple Ligation-dependent Probe-Amplification; dcRT-MLPA).16 Furthermore, the expression of selected chemokines/cytokines, at the protein level, was assessed using a multiplex bead assay.

RESULTS Nested case–control study group characteristics Baseline characteristics for the study groups are provided in Table 1. Age (in months; P ¼ 0.011), proportion of children with failure to thrive Po0.001) and malnutrition (weight-for-height Zscore (WHZ); Po0.001 and weight-for-age Z-score (WAZ); P ¼ 0.017) were differently distributed between the clinical groups. The proportion of TB disease cases was found to be higher in the 0–12 (30.8%) and 13–24 (53.8%) months age categories (Table 1). Children classified with TB disease and latent TB were more likely to have persistent cough and a history of TB contact compared with uninfected controls (both Po0.001). The proportion of NTMpositive cases was higher in the uninfected controls versus those with TB disease and the latent TB group (P ¼ 0.029). Unexpectedly, the majority of infants classified with TB disease did not have a positive TST or QuantiFERON-TB Gold In-tube (QFT-GIT; Cellestis Inc, Valencia, CA, USA) (Table 1). Identification of gene signatures We analyzed the dcRT-MLPA data using the global test for a pairwise comparison between TB disease versus latent TB, TB disease versus uninfected controls and latent TB versus uninfected controls. The binary outcome from the global test showed that none of the genes were associated with the three clinical groups (Supplementary Figure 1). Differences in baseline characteristics were identified by the Pearson’s w2 test (with Yates continuity correction) or Fisher’s exact test where appropriate, and used for adjusting the confounders. As cough, history of exposure to TB, failure to thrive, chest X-ray (CXR) changes, TST and QFT-GIT were used as selection criteria for referring children to the TB case verification ward (CVW)/classification to clinical groups, binary logistic regression analysis was performed with and without adjusting for age (months), nutritional status (WHZ and WAZ) and culture positivity for NTM. The unadjusted and adjusted results are shown in Table 2. The results from the adjusted analysis were plotted and shown in Figure 1. The transcription of RAB33A was downregulated in children with TB disease compared with that in latent TB (P ¼ 0.015) and uninfected controls (P ¼ 0.002) (Figure 1a). Conversely, transcription of TGF-b1 was upregulated in children with TB disease (P ¼ 0.015) and latent TB (P ¼ 0.003), compared with that in uninfected controls (Figure 1b). Similarly, the transcription of CD4 (P ¼ 0.006) was upregulated in the latent TB group compared with that in uninfected controls (Figure 1c). Children with TB disease had reduced transcription for CCR7 (P ¼ 0.038), BPI (P ¼ 0.046) SEC14L1 (P ¼ 0.05) and GUSB (P ¼ 0.05) compared with uninfected controls (Figures 1d-g). In unadjusted analysis, the transcription of FOXP3 was downregulated in TB disease compared with latent TB (P ¼ 0.025), whereas the transcription of TNFRSF1A (P ¼ 0.047) was upregulated in the latent TB compared with uninfected controls (Table 1); however, these markers failed to reach statistical significance in adjusted analysis (Figures 1h and i, respectively). & 2013 Macmillan Publishers Limited

The least absolute shrinkage and selection operator (lasso) regression analysis was performed on dcRT-MLPA data with and without adjusting for age (months), nutritional status (WHZ and WAZ) and culture positivity for NTM. For TB disease versus latent TB group, RAB33A alone gave an area under the curve (AUC) of 85.0 and 77.5% for unadjusted and adjusted analysis (Figures 2a and d, respectively). For TB disease versus uninfected controls, the combination of SEC14L1 and RAB33A gave an AUC of 81.3% (unadjusted; Figure 2b) and the combination of CXCL10, SEC14L1, FOXP3, RAB33A and TNFRSF1A presented an AUC of 91.7% (adjusted; Figure 2e). The biomarkers proposed by both models were RAB33A and SEC14L1. Similarly, for the latent TB versus uninfected controls, the combination of CD4, SEC14L1, TGFb1 and TNFRSF1A provided an AUC of 60.1% (unadjusted; Figure 2c) and a combination of 11 biomarkers: CXCL10, CD4, SEC14L1, IL4, RAB33A, TGFBR2, BCL2, TGFb1, TNFRSF1B, RAB13 and CD19 gave an AUC of 72.2% (adjusted; Figure 2f). Again, the biomarkers proposed by both models were CD4, SEC14L1 and TGFb1 (Figure 2). Identification of cytokine/chemokine profiles The assessment of the association of the selected cytokine/ chemokine biomarkers in the 10-plex assay, between the clinical categories (pair-wise comparison) was undertaken by the global test. Global test analysis showed that none of the selected cytokines/chemokines (from TB-specific antigen-stimulated tubes) could differentiate between the three clinical groups (Supplementary Figure 2). The results from the binary logistic regression analyses (unadjusted and adjusted for age in months, WHZ, WAZ and culture positivity for NTM) are shown in Table 3. As shown in Figure 3, compared with uninfected controls, the expression of IL-2 (P ¼ 0.009) and IL-13 (P ¼ 0.047) was higher in children with latent TB (Figures 3a and b). The expression of MCP1 and IP-10, although not approaching statistical significance (P ¼ 0.07), was somewhat higher in latent TB compared with that in uninfected controls (Figures 3c & d). The lasso regression models (unadjusted and adjusted for age in months, nutritional status (WHZ and WAZ) and culture positivity for NTM) proposed a combination of biomarker signatures for the three clinical groups (TB disease vs. uninfected controls; TB disease vs latent TB and latent TB vs uninfected controls), as shown in Supplementary Figure 3. Lasso regression analysis identified a combination of IL-2 and IL-8 that could differentiate with moderate power (AUCB70.0%) between TB disease and latent TB, but was unable to identify a biomarker signature with sufficient discriminatory power for the other clinical categories (Supplementary Figure 3).

DISCUSSION TB in children is a devastating problem globally; young children are at increased risk of developing disseminated TB disease and death.14,17,18 There is a need for potential TB biomarker(s) that can differentiate between subjects with TB disease and latent TB infection, indicate the reactivation risk for latent TB subjects, predict treatment success and provide immune correlates of protection against TB by new vaccines.19 In the present study, conducted in a setting where TB is prevalent, we screened for biomarkers with the potential to discriminate between TB disease, latent TB infection and uninfected controls in children suspected of having TB. A predefined panel of 45 genes was assessed in unstimulated whole-blood samples from these children, using novel high-throughput technology (dcRT-MLPA), which is inexpensive and equally robust as qPCR.16 Further, the concentration of cytokines/chemokines in the supernatants from TB-specific antigen-stimulated and unstimulated whole blood were analyzed using a Bio-plex assay. Genes and Immunity (2013) 356 – 364


358 Table 1.

Baseline characteristics of 210 TB suspect children

Baseline characteristics Surveillance group Active (n ¼ 142) Passive (n ¼ 68) Age (months) 0–12 (n ¼ 43) 13–24 (n ¼ 135) 25–35 (n ¼ 32) Gender Male (n ¼ 125) Female (n ¼ 85) Birth weight Low (n ¼ 67) Normal (n ¼ 118) Not recorded (n ¼ 25) CXR Abnormal-TB (n ¼ 16) Normal (n ¼ 194)

Table 1. P-value

TB disease, n ¼ 13 (%)

Latent TB, n ¼ 90 (%)

Uninfected controls, n ¼ 107 (%)

10 (76.9)

58 (64.4)

74 (69.2)

0.597

3 (13.1)

32 (35.6)

33 (30.8)

—

4 (30.8) 7 (53.8)

17 (18.9) 48 (53.3)

22 (20.6) 80 (74.8)

2 (15.4)

25 (27.8)

5 (4.6)

7 (53.8) 6 (46.2)

53 (58.8) 37 (41.2)

65 (60.7) 42 (39.3)

0.011 — — 0.882 —

Baseline characteristics

TST Positive (n ¼ 75) Negative (n ¼ 135) QFT-GIT Positive (n ¼ 40) Negative (n ¼ 166) Indeterminate (n ¼ 4)

TB disease, n ¼ 13 (%)

Latent TB, n ¼ 90 (%)

Uninfected controls, n ¼ 107 (%)

P-value

4 (30.8)

71 (78.9)

0 (0.0)

o0.001

9 (69.2)

19 (21.1)

107 (100.0)

—

3 (23.1)

37 (41.1)

0 (0.0)

o0.001

8 (61.5)

53 (58.9)

105 (98.1)

—

2 (15.4)

0 (0.0)

2 (1.9)

—

Differences in baseline characteristics were identified by the Pearson’s Chisquared test (with Yates continuity correction) or Fisher’s exact test, where appropriate.

4 (30.8) 6 (46.2)

27 (30.0) 51 (56.7)

36 (33.6) 61 (57.0)

0.891 —

3 (23.0)

12 (13.3)

10 (9.4)

—

11 (84.6)

0 (0.0)

0 (0.0)

o0.001

2 (15.4)

90 (100.0)

107 (100.0)

—

Cough more than 2 weeks Yes (n ¼ 18) 4 (30.8) No (n ¼ 192) 9 (69.2)

14 (15.6) 76 (84.4)

0 (0.0) 107 (100.0)

o0.001 —

Failure to thrive Yes (n ¼ 154) No (n ¼ 56)

44 (48.9) 46 (51.1)

102 (95.3) 5 (4.7)

o0.001 —

24 (26.7)

57 (53.3)

o0.001

66 (73.3)

50 (46.7)

—

32 (35.6)

39 (36.4)

0.761

58 (64.4)

68 (63.6)

—

40 (44.4)

69 (64.5)

0.017

50 (55.6)

38 (35.5)

—

10 (11.1) 80 (88.9)

0 (0.0) 107 (100.0)

0.002 —

19 (21.1)

34 (31.8)

0.029

71 (78.9)

73 (68.2)

—

8 (61.5) 5 (38.5)

Weight for height (WHZ) o2 7 (53.8) (wasting) (n ¼ 88) X2 (normal) 6 (46.2) (n ¼ 122) Height for age (HAZ) o2 6 (46.2) (stunted) (n ¼ 77) X2 (normal) 7 (53.8) (n ¼ 133) Weight for age (WAZ) 8 (61.5) o2 (underweight) (n ¼ 117) X2 (normal) 5 (38.5) (n ¼ 93) History of contact with TB Yes (n ¼ 11) 1 (7.7) No (n ¼ 199) 12 (92.3) Culture for NTM Positive (n ¼ 52) Negative (n ¼ 158)

(Continued )

0 (0) 13 (100.0)

Genes and Immunity (2013) 356 – 364

The transcription of RAB33A was downregulated in children with TB disease compared with children with latent TB and uninfected controls in both unadjusted and adjusted analysis. The transcription of FOXP3 was downregulated in children with TB disease compared with latent TB in unadjusted analysis (Po0.05), although it failed to reach statistical significance in adjusted analysis (P ¼ 0.08). Unadjusted and adjusted (for age in months, nutritional status (WHZ and WAZ) and culture positivity for NTM) lasso regression models identified RAB33A that could differentiate between TB disease and latent TB with a good discriminatory power (AUC: 85.0%, 77.5%, respectively). RAB33A is a small guanosine triphosphatase (GTPase) suggested to be involved in vesicle transport and fusion.20 Dysregulation of GTPases have been shown to have a role in blocking phagosome maturation,21 which is a major survival strategy for M. tuberculosis.22,23 Earlier studies from Caucasian and South African adults have shown a downregulation of the expression of RAB33A, CD64 (Fcg receptor 1A) and LTF (lactoferrin) in unstimulated peripheral blood mononuclear cells (PBMCs) from patients with TB disease.24–26 To our knowledge, this is the first study to identify RAB33A as a potential biomarker for TB disease in young Indian children. FOXP3 is a crucial regulator of the development and function of regulatory T-cells.27 Evidence suggests that FOXp3 þ regulatory T-cells can have both protective and detrimental roles in host defense against infection.28 Previous studies have shown that the expression of FOXP3 is increased in adults with active TB.29–31 Furthermore, Burl et al.32 reported that during early TB infection, the expression of FOXP3 is reduced and subsequently increased during TB disease. In the present study, we found that in unadjusted binary logistic regression analysis, the expression of FOXP3 was significantly downregulated in the TB disease group, as compared with the latent TB group (P ¼ 0.025), whereas in adjusted analysis it did not maintain statistical significance (P ¼ 0.08). Compared with that in uninfected controls, the transcription of TGFb1 was upregulated in children with TB infection (Po0.01) and TB disease (Po0.05). Similarly, CD4 was also upregulated in latent TB infection (Po0.01), but not in TB disease. Reports have shown that the expression of TGF-b is increased in adult with TB disease.29,33,34 TGF-b1 performs many cellular functions including control of cell growth, proliferation and differentiation. Furthermore, it is an anti-inflammatory marker, and involved in & 2013 Macmillan Publishers Limited


359 Table 2.

dcRT-MLPA based gene comparisons for the TB disease, latent TB and uninfected controls

Genes

CD8A CXCL10 CD4 IL4d2 BLR1 SEC14L1 GUSB TIMP2 CCL19 FPR1 IL4 NCAM1 FOXP3 CTLA4 TNF RAB33A ABR TGFBR2 IL7R IL10 BCL2 CASP8 TNFRSF18 BPI CCR7 SPP1 CCL22 MMP9 RAB24 CD163 TGFB1 TNFRSF1B FCGR1A TNFRSF1A RAB13 CD3E CD19

TB disease vs latent TB (P-value)

TB disease vs uninfected controls (P-value)

Latent TB vs uninfected controls (P-value)

Unadjusted

Adjusteda

Unadjusted

Adjusteda

Unadjusted

Adjusteda

0.718 0.560 0.682 0.976 0.141 0.218 0.442 0.491 0.349 0.305 0.356 0.257 0.025(*) 0.095 0.660 0.002(**) 0.743 0.712 0.306 0.209 0.864 0.980 0.940 0.398 0.163 0.775 0.283 0.183 0.560 0.233 0.192 0.406 0.354 0.729 0.545 0.602 0.465

0.812 0.492 0.937 0.675 0.169 0.427 0.562 0.232 0.220 0.060 0.451 0.915 0.080 0.196 0.372 0.015(*) 0.643 0.563 0.249 0.527 0.925 0.765 0.366 0.597 0.413 0.878 0.634 0.361 0.597 0.818 0.352 0.080 0.232 0.150 0.234 0.372 0.480

0.960 0.529 0.265 0.681 0.176 0.039(*) 0.164 0.169 0.485 0.084 0.429 0.151 0.167 0.352 0.436 0.004(**) 0.278 0.699 0.274 0.161 0.882 0.934 0.889 0.237 0.076 0.414 0.148 0.109 0.826 0.175 0.010(*) 0.170 0.342 0.184 0.151 0.207 0.820

0.552 0.555 0.246 0.401 0.139 0.050 0.050 0.229 0.444 0.068 0.185 0.094 0.182 0.387 0.459 0.002(**) 0.396 0.949 0.201 0.094 0.983 0.889 0.974 0.046(*) 0.038(*) 0.204 0.061 0.241 0.670 0.113 0.015(*) 0.258 0.434 0.148 0.109 0.181 0.688

0.632 0.682 0.002(**) 0.339 0.777 0.124 0.109 0.114 0.479 0.130 0.925 0.807 0.194 0.455 0.318 0.581 0.076 0.962 0.967 0.547 0.523 0.829 0.652 0.508 0.337 0.252 0.240 0.437 0.469 0.790 0.012(*) 0.215 0.926 0.047(*) 0.081 0.222 0.315

0.796 0.549 0.006(**) 0.387 0.812 0.070 0.116 0.146 0.459 0.152 0.348 0.603 0.380 0.471 0.318 0.114 0.070 0.818 0.843 0.424 0.706 0.411 0.674 0.303 0.172 0.164 0.113 0.524 0.637 0.520 0.003(**) 0.268 0.954 0.085 0.060 0.087 0.412

P-values o0.05 (*) and o0.01(**) were considered to be significant. The significant P-values are in bold. aAdjusted for age in months, nutritional status (WHZ and WAZ) and culture positivity for NTM.

the wound-healing process of granulomatous lesions in patient with TB disease.34 CD4 has a major role in adaptive immunity to TB.16 Thus, the differential expression in young children of TGFb1 and CD4, both previously shown to have a role in the host responses to TB, is not unexpected. The combination of 11 biomarkers (CXCL10, CD4, SEC14L1, IL4, RAB33A, TGFBR2, BCL2, TGFb1, TNFRSF1B, RAB13 and CD19) identified by the lasso regression model gave only moderate discriminatory power (AUC 72.2%) to differentiate between latent TB and uninfected controls (Figure 2f). The transcription of SEC14L1, RAB33A, GUSB, CCR7, BPI, and TGFb1 was differentially expressed between children with TB disease and uninfected controls. For TB disease, RAB33A, SEC14L1, TGFb1 and for latent TB, CD4 and TGFb1 were identified as potential biomarkers in both unadjusted and adjusted binary logistic regression analysis, whereas CCR7, BPI and GUSB were only identified in the TB disease group by adjusted regression analysis. CCR7 is a chemokine receptor that controls the migration of antigen-presenting cells to the lymph node (ligands CCL19, CCL21) where the adaptive immune response is initiated. CCR7 has specifically been shown to mediate trafficking of dendritic cells from the lungs to the mediastinal lymph node during TB disease.35 SEC14-like protein 1 belongs to the SEC14 cytosolic & 2013 Macmillan Publishers Limited

factor family, which has a role in the intracellular transport system,36 whereas BPI encodes a lipopolysaccharide-binding protein associated with human neutrophil granules. Earlier findings have shown that the serum BPI concentration is elevated in TB disease.37 In the mouse model, Vishwanath et al.38 have shown an increased b-glucuronidase (GUSB) activity during phagocytosis of M. tuberculosis by peritoneal macrophages. In contrast, the findings from the current study in young children suggest that the expression of BPI and GUSB is reduced in TB disease, perhaps due to the pathogen interference with intracellular trafficking of phagocytes. Adjusted lasso regression analysis identified a combination of CXCL10, SEC14L1, FOXP3, RAB33A and TNFRSF1A that could differentiate with excellent discriminatory power (AUC 91.7%) between TB disease and uninfected controls. Young children recruited in the childhood cohort study study conducted in Palamaner taluk, Southern India, were referred to the CVW if suspected of having TB. As described, all referred children had either symptoms suggestive of TB, contact with a TB case and/or failure to thrive. Cytokine/chemokine expression in supernatants from stimulated and unstimulated tubes (TB-ag and Nil-, respectively) in the QFT-GIT were analyzed separately. Unstimulated whole-blood supernatants did not identify a Genes and Immunity (2013) 356 – 364


360

Figure 1. Dot-plot graph depicting genes that are differentially expressed between the three clinical groups: TB disease, latent TB and uninfected controls. Data were adjusted for age (in months), nutritional status (WHZ, WAZ) and culture positivity for NTM. Median with interquartile range was used. The figure illustrates the relative gene expression of: (a) RAB33A; (b) TGFB1; (c) CD4; (d) CCR7; (e) BPI; (f) SEC14L1; (g) GUSB; (h) FOXP3; and (i) TNFRSF1A. *P-value o0.05 in unadjusted binary logistic regression analysis.

differential cytokine/chemokine expression between the three clinical groups (data not shown). The TB-specific antigenstimulated samples demonstrated significant differences in cytokine/chemokine concentrations for IL-2 (P ¼ 0.009) and IL-13 (P ¼ 0.047) in the latent TB group compared with the uninfected controls (Figure 3). In a previous study, IP-10 has been shown to be a marker of M. tuberculosis infection and disease in children, but could not discriminate between the two.13 A similar finding was observed in the present study; however, the values failed to reach statistical significance. Previous studies have shown that the expression of IL-2, MCP-1 and IL-13 is increased in adults with TB compared with that in controls.39–42 However, the present study shows that compared with that in uninfected controls, the expression of IL-2 and IL-13 is increased in the latent TB group only, whereas the expression of MCP-1 was upregulated in children with latent TB infection; albeit, it did not reach statistical significance. Furthermore, a recent study showed that IFN-g, TNF-a (type 1 cytokines) and IL-13 (type II cytokines) had a reduced expression in children with TB disease compared with that in healthy controls; however, these findings could not be confirmed in the present study.14 Malnutrition has an impact on the age-dependent maturation of the immune system in children.43 In the present study, in a rural Genes and Immunity (2013) 356 – 364

Indian setting, 440% of the children were malnourished (as measured by wasting, stunting and/or underweight). A high proportion of young children that are malnourished could potentially be a challenge in studies that evaluate TB biomarker(s). Longitudinal studies in similar settings together with the inclusion of additional promising transcriptional biomarkers (for example, IFN-inducible genes as identified in the recent study by Berry et al.44) are crucial to the development of improved diagnostic and treatment modalities for TB. The strength of this study lies in its prospective design and that it is conducted in a TB-endemic population with low prevalence of HIV infection. The study setting resembles a ‘real-life’ context and, as a result, the conclusions are clinically relevant. Malnutrition and exposure to NTM are common in children in TB-endemic countries and may limit the generalizability to other populations. In conclusion, this study in children aged o3 years in a rural Southern India setting, strengthens the evidence for RAB33A as a potential biomarker in children that distinguishes between TB disease and latent TB. Furthermore, CD4 and TGF-b1 at the mRNA level, and IL-2 and IL-13 at the protein level demonstrated some potential to discriminate latent TB from uninfected children. Additional studies in children, together with the inclusion of other candidate biomarkers are required. Such studies are critical for the & 2013 Macmillan Publishers Limited


361

Figure 2. Results from Lasso analysis of dcRT-MLPA data. The ability of biomarker signatures to predict clinical outcomes (TB disease, latent TB and uninfected controls) was identified following unadjusted and adjusted (age in months, nutritional status (WHZ, WAZ) and culture positivity for NTM) Lasso regression analysis. The predicted probability of the identified biomarker signatures to discriminate between TB disease, latent TB and uninfected controls is shown by: receiver operator characteristics curves (ROCs), area under the curve (AUC) and box-and-whisker plots (5th-95th percentiles). Panels a, b and c display the results from unadjusted Lasso regression analysis and panels d, e and f display the results from adjusted Lasso regression analysis.

Table 3.

Bio-plex (10-plex) results based on cytokines/chemokines concentrations from the QFT-GIT supernatant (TB-ag tube)

Cytokines/ Chemokines

TB disease vs latent TB (P-value)

TB disease vs uninfected controls (P-value)

Latent TB vs uninfected controls (P-value)

Unadjusted

Adjusteda

Unadjusted

Adjusteda

Unadjusted

Adjusteda

0.275 0.362 0.784 0.094 0.295 0.426 0.872 0.519 0.585 0.959

0.111 0.343 0.759 0.100 0.218 0.201 0.761 0.436 0.826 0.835

0.105 0.662 0.694 0.209 0.323 0.141 0.888 0.087 0.292 0.961

0.188 0.526 0.862 0.203 0.347 0.337 0.645 0.090 0.607 0.572

0.018(*) 0.701 0.731 0.183 0.960 0.093 0.960 0.080 0.188 0.998

0.009(**) 0.328 0.588 0.090 0.640 0.047(*) 0.583 0.070 0.070 0.554

IL-2 IL-5 IL-6 IL-8 IL-10 IL-13 IFN-g IP-10 MCP-1 TNF-a

P-values o0.05 (*) and o0.01 (**) were considered to be significant. The significant P-values are in bold. aAdjusted for age in months, nutritional status (WHZ and WAZ) and culture positivity for NTM.

development of improved TB diagnostic tests for children, particularly in developing countries. MATERIALS AND METHODS Study details The case–control study described here was nested within a prospective longitudinal study of 4382 neonates in Palamaner taluk, Southern India. The neonates included in the study were all BCG-vaccinated by the study & 2013 Macmillan Publishers Limited

team within 72 h of birth and were enrolled within 2 weeks from the date of birth with parental consent. Parents/guardians who did not provide informed consent or who planned to move out from the study area were excluded from the study. The recruited children were randomly (based on the population units where they were born) assigned to active (visited bimonthly; to check for recent TB contact, symptoms and anthropometry; N ¼ 2215) and passive (TB education given to parents/guardian, but with no scheduled home visits; N ¼ 2167) surveillance arms, and monitored at fixed time points, as outlined in the study protocol, for 2 consecutive years. Genes and Immunity (2013) 356 – 364


362

Figure 3. Scatter-plot graph depicting median cytokine concentrations (pg ml 1) from the QFT-GIT supernatants (TB-ag tube) by the Bio-plex assay. The data were adjusted for age (in months), nutritional status (WHZ, WAZ) and culture positivity for NTM. The figure illustrates the median cytokine concentrations in pg ml 1 of: (a) IL-2; (b) IL-13; (c) MCP-1; and (d) IP-10.

Parents/guardians were provided with a study identification card with telephonic contacts of the study team physician to ensure follow-up if they visited facilities other than the study diagnostic centre. Out of 4382 children included in the study, 746 children with suggestive symptoms of TB (coughX2 weeks), failure to thrive (unexplained weight loss/delay in a child’s growth according to weight for age growth chart, persistent tracking below the 3rd percentile of weight for age) or history of contact with TB patients were sent to the TB CVW at the Emmaus Swiss Hospital, Palamaner, India. At the CVW, the children underwent an anthropometric assessment (age, length and weight) and hemoglobin (Hgb) (heel prick) estimations. Approximately 5 ml blood was drawn for further analyses; 3 ml was collected for the QuantiFERON-TB Gold In-tube (QFT-GIT) assay and the supernatants (plasma) from the Nil, TB antigen (ag) and Mitogen tubes were collected in two microfuge tubes and stored at 20 1C. One microfuge tube was used for the QFT-GIT assay and the other tube was stored for Bio-Plex assay. Approximately 1.0–2.5 ml blood was collected in PAXgene blood collection tubes (PreAnalytiX (a Qiagen/BD company), Hombrechtikon, Switzerland) and stored at 80 1C until RNA extraction was performed. Following the collection of the blood sample, a CXR (anteroposterior view) was taken and the TST was administered with 2 TU per 0.1 ml of purified protein derivative RT-23 (Span Diagnostics Ltd., Bangalore, India), and read after 48 h. A cutoff X10 mm was used. In addition, induced sputum and gastric aspirates were collected on each of two consecutive mornings for smear microscopy and culture by MGIT and LJ by standard methods.45 Positive cultures were confirmed by using the GenoType MTBC kit (Hain Life science GmBH, Nehren, Germany). Direct PCR by COBAS Taqman MTB test (Roche Diagnostics Ltd, Rotkreuz, Switzerland) was undertaken on culture-negative specimen for infants with CXR finding suggestive of TB.

Sample selection for biomarker study Drawing on participants (n ¼ 746) that were referred to the TB CVW, children were classified into four groups according to a generally accepted TB pediatric diagnostic algorithm,46 which was modified to also include the TST and QFT-GIT results. These were: (a) definite TB—positive for M. tuberculosis by smear/culture, and/or standard PCR (n ¼ 4), (b) probable TB—abnormal changes in CXR (2/3 independent readers reported, an abnormal CXR abnormal) suggestive of pulmonary TB but culture-negative for M. tuberculosis (n ¼ 9). (c) Latent TB—these children were smear-/ culture-negative for M. tuberculosis, had normal CXR and were either TSTGenes and Immunity (2013) 356 – 364

positive, QFT-GIT-positive or positive in both tests (n ¼ 90). (d) Uninfected controls—these children were smear-/culture-negative for M. tuberculosis, TST/QFT-GIT negative, and had normal CXR (n ¼ 107). The combination of definite and probable TB was designated as the TB disease group (n ¼ 13). The uninfected controls were matched on sex to the latent and TB disease groups for use as controls.

RNA extraction Total RNA was extracted from the PAXgene blood collection tubes using the ‘PAXgene Blood RNA kit’ with RNase free DNase on-column digestion (PreAnalytiX, Hilden, Germany) according to the manufacturer’s instructions. The total RNA concentration and purity (A260/280 nm ratio) were measured using a Nanodrop spectrophotometer (Thermoscientific, Wilmington, DE, USA) and ranged between 0.4–24.5 mg (average 6.6±4.85 mg). The RNA quality was further assessed by agarose gel electrophoresis.

Dual-color Reverse-Transcriptase Multiplex Ligation-dependent Probe-Amplification (dcRT-MLPA) As the blood sample available from infants is limited, we used a novel high-throughput technique, which requires only 130–150 ng of total RNA for analyzing a predefined panel of genes of interest. The dcRT-MLPA experiment protocol has been previously described in detail,16 and is therefore only briefly discussed here. dcRT-MLPA probes and primers (reverse transcription gene target-specific primers, right hand and left hand half MLPA probes, FAM-labeled MLPA primers, HEX-labeled MAPH primers) were obtained from the Department of Infectious Diseases, Leiden Medical University, Leiden, The Netherlands. Samples with a concentration o50 ng ml 1 were concentrated at 45 1C using a speed vacuum concentrator (Eppendorf AG, Hamburg, Germany). A positive control (using synthetic template oligonucleotides as hybridization templates) and a commercial Human Universal Reference RNA were included on each plate. All samples (n ¼ 210) were run in duplicate. The amplified PCR products were diluted 1:10 with nuclease-free H2O and added to a mixture of Hi-Di-Formamide with 400HD ROX size standard. The PCR products were denatured at 95 1C for 5 min and then immediately cooled on ice. Subsequently, the PCR fragments were analyzed on a 3730 capillary sequencer in Gene scan mode (Life Technologies, Carlsbad, CA, USA). & 2013 Macmillan Publishers Limited


363 Processing of dcRT-MLPA data Data were analyzed using GeneMapper software version 4.0 (Life Technologies) according to the GeneMapper version 4.0 manual. The default peak detection settings were inspected and adjusted if necessary. The peak area (in arbitrary units) of replicates were averaged, then normalized against GAPDH using Microsoft Excel spreadsheet software and log2-transformed as described by Joosten et al.16 Of the 45 genes, 7 genes had expression levels below the cutoff value of 7.64 (peak areao200 arbitrary units) and one gene CD14, colocalized with a primer–dimer peak and was therefore omitted from further analysis.

contracted by Aeras. It was also approved by the Ministry of Health Screening Committee of the Government of India (No. 5/8/9/60/ 20006-ECD-I).

CONTRIBUTORS Members of the TB Trials Study Group: Doherty M, Grewal HMS, Hesseling AC, Jacob A, Jahnsen F, Kenneth J, Kurpad AV, Lindtjorn B, Macaden R, Nelson J, Sumithra S, Vaz M, Walker R.

Multiplex bead array-Bio-Plex assay For the Bio-Plex assay, we used supernatants from the QFT-GIT assay (Nilag, TB-ag and Mitogen tubes). A pilot study was conducted using 52 randomly selected samples from the biomarker study samples (TB disease (n ¼ 11), latent TB (n ¼ 29) and uninfected controls (n ¼ 12)). The selected 52 samples were tested using the ‘Human cytokines 27-plex’ kit (Bio-Rad Laboratories Inc., Hercules, CA, USA), according to the manufacturer’s instructions. Five of 27 biomarkers (IL-5, IL-13, IP-10, IFN-g and TNF-a) were differentially expressed between the three clinical groups (data not shown) and 5 (IL-2, IL-6, IL-8, IL-10 and MCP-1 (MCAF)) were selected based on previous literature on TB biomarkers,41,47,48 in a customized 10-plex kit. The study samples (n ¼ 210) were then tested on the 10-plex kit. For data analysis, the cytokine/chemokine concentrations (pg ml 1) in the Nil and TB-ag tubes were used. The cytokine/chemokine concentrations in the unstimulated (Nil) and stimulated (TB-ag) tubes were analyzed individually.

CONFLICT OF INTEREST

Statistical analysis

REFERENCES 2

Differences in baseline characteristics were identified by the Pearson’s w test (with Yates continuity correction) or Fisher’s exact test. A global test was performed on both transcriptional and translational data, evaluating the null hypothesis that none of the biomarkers were associated with the binary outcome, using the locally most powerful test.49 Specifically, single biomarkers from both dcRT-MLPA and the Bio-Plex assay were evaluated to determine if the biomarker predicted the TB status considered. These analyses were performed with and without adjusting for age in months, WHZ, WAZ and culture positivity for NTM. No correction for multiple testing was undertaken. The ‘lasso’ is a regression model that we used to analyze the transcriptional and translational data sets. This method penalizes the absolute size of the regression coefficients, shrinking them towards zero. This can be an appropriate approach when dealing with highly correlated or interdependent predictors, where standard regression may generate overly generous regression coefficients.50 It is an alternative to backwards step-wise elimination of non-significant terms, not relying on statistical significance tests. As a result, lasso is useful for establishing a prediction model at an individual level (that is, identifying important predictors), but not for summarizing average trends in the data.50 The optimal lasso model (that is finding the optimal lambda value or penalty) was determined using leave-one out cross validation. The lasso model corresponding to the lambda value resulting in the smallest prediction error was considered the optimal model. The cross-validation approach replaces the classical adjustment for multiple testing.50 The prediction performance reported in terms of probabilities and receiveroperator characteristics curves was based on evaluating the optimal lasso model on the test data. The non-zero estimated coefficients were also reported. A P-value was calculated using two-tailed tests and the P-valueo0.05 was considered as significant. Dot plots were created using GraphPad Prism 5 (GraphPad Software, Inc., La Jolla, CA, USA). SPSS software version 19.0 was used for Pearson’s w2 or Fischer’s exact testing and binary logistic regression. The statistical environment R was used for the global test51 to assess the association of the three binary outcomes, that is, TB disease versus latent TB, TB disease versus uninfected controls, and latent TB versus uninfected controls, to establish if all biomarkers had an effect on the binary outcome. For the lasso model the add-on package ‘glmnet’ was used52 in R program.

Ethics statement The study was conducted according to the Helsinki (4th revision) Declaration and approved by the institutional ethical review board of the St John’s Medical College and an independent ethics committee & 2013 Macmillan Publishers Limited

The authors declare no conflict of interest.

ACKNOWLEDGEMENTS We thank Harald G Wiker (University of Bergen) for constructive advice; Aud Eliassen at the sequencing laboratory at Haukeland University Hospital, Bergen, Norway. Diana Mahelai, Geojith George, Sumithra Selvan, Nelson Jesuraj and Naveen Kumar Kellengere at St John’s Research Institute, Bangalore; Lien M Diep (University of Oslo) for technical and statistical assistance. This study is conducted as part of the IndoNorway research program. This study was supported by the Research Council of Norway (grants 179342, 192534 and 196362), University of Bergen, Aeras USA, and St John’s Research Institute, India.

1 WHO. Global Tuberculosis Report 2012. 2 Delespesse G, Yang LP, Ohshima Y, Demeure C, Shu U, Byun DG et al. Maturation of human neonatal CD4 þ and CD8 þ T lymphocytes into Th1/Th2 effectors. Vaccine 1998; 16: 1415–1419. 3 Upham JW, Lee PT, Holt BJ, Heaton T, Prescott SL, Sharp MJ et al. Development of interleukin-12-producing capacity throughout childhood. Infect Immunity 2002; 70: 6583–6588. 4 Upham JW, Rate A, Rowe J, Kusel M, Sly PD, Holt PG. Dendritic cell immaturity during infancy restricts the capacity to express vaccine-specific T-cell memory. Infect Immunity 2006; 74: 1106–1112. 5 Seth V, Singhal PK, Semwal OP, Kabra SK, Jain Y. Childhood tuberculosis in a referral centre: clinical profile and risk factors. Indian Pediatr 1993; 30: 479–485. 6 Beyers N, Gie RP, Schaaf HS, Van Zyl S, Talent JM, Nel ED et al. A prospective evaluation of children under the age of 5 years living in the same household as adults with recently diagnosed pulmonary tuberculosis. Int J Tuberc Lung Dis 1997; 1: 38–43. 7 Marais BJ, Gie RP, Schaaf HS, Hesseling AC, Enarson DA, Beyers N. The spectrum of disease in children treated for tuberculosis in a highly endemic area. Int J Tuberc Lung Dis 2006; 10: 732–738. 8 Shingadia D, Novelli V. Diagnosis and treatment of tuberculosis in children. Lancet Infect Dis 2003; 3: 624–632. 9 Machingaidze S, Wiysonge CS, Gonzalez-Angulo Y, Hatherill M, Moyo S, Hanekom W et al. The utility of an interferon gamma release assay for diagnosis of latent tuberculosis infection and disease in children: a systematic review and meta-analysis. Pediatr Infect Dis J 2011; 30: 694–700. 10 WHO. Strategic and technical advisory group for tuberculosis (STAG-TB). Report of the Tenth Meeting 2010. 11 WHO. Joint TDR/EC expert consultation on biomarkers in tuberculosis 2008. 12 Lopez Avalos GG, Prado Montes de Oca E. Classic and new diagnostic approaches to childhood tuberculosis. J Tropical Med 2012; 2012: 818219. 13 Whittaker E, Gordon A, Kampmann B. Is IP-10 a better biomarker for active and latent tuberculosis in children than IFNgamma? PloS One 2008; 3: e3901. 14 Kumar NP, Anuradha R, Suresh R, Ganesh R, Shankar J, Kumaraswami V et al. Suppressed type 1, type 2, and type 17 cytokine responses in active tuberculosis in children. Clin Vaccine Immunol 2011; 18: 1856–1864. 15 Thomas T, Brighenti S, Andersson J, Sack D, Raqib R. A new potential biomarker for childhood tuberculosis. Thorax 2011; 66: 727–729. 16 Joosten SA, Goeman JJ, Sutherland JS, Opmeer L, de Boer KG, Jacobsen M et al. Identification of biomarkers for tuberculosis disease using a novel dual-color RTMLPA assay. Genes Immunity 2012; 13: 71–82. 17 Nelson LJ, Wells CD. Global epidemiology of childhood tuberculosis. Int J Tuberc Lung Dis 2004; 8: 636–647. 18 Swaminathan S, Rekha B. Pediatric tuberculosis: global overview and challenges. Clin Infect Dis 2010; 50(Suppl 3): S184–S194.



364 19 Wallis RS, Pai M, Menzies D, Doherty TM, Walzl G, Perkins MD et al. Biomarkers and diagnostics for tuberculosis: progress, needs, and translation into practice. Lancet 2010; 375: 1920–1937. 20 Schimmoller F, Simon I, Pfeffer SR. Rab GTPases, directors of vesicle docking. J Biol Chem 1998; 273: 22161–22164. 21 Brumell JH, Scidmore MA. Manipulation of rab GTPase function by intracellular bacterial pathogens. Microbiol Mol Biol Rev 2007; 71: 636–652. 22 Doherty M, Wallis RS, Zumla A. Biomarkers for tuberculosis disease status and diagnosis. Curr Opin Pulm Med 2009; 15: 181–187. 23 Sturgill-Koszycki S, Schlesinger PH, Chakraborty P, Haddix PL, Collins HL, Fok AK et al. Lack of acidification in Mycobacterium phagosomes produced by exclusion of the vesicular proton-ATPase. Science 1994; 263: 678–681. 24 Jacobsen M, Repsilber D, Gutschmidt A, Neher A, Feldmann K, Mollenkopf HJ et al. Ras-associated small GTPase 33A, a novel T cell factor, is down-regulated in patients with tuberculosis. J Infect Dis 2005; 192: 1211–1218. 25 Jacobsen M, Repsilber D, Gutschmidt A, Neher A, Feldmann K, Mollenkopf HJ et al. Candidate biomarkers for discrimination between infection and disease caused by Mycobacterium tuberculosis. J Molecular Med 2007; 85: 613–621. 26 Maertzdorf J, Repsilber D, Parida SK, Stanley K, Roberts T, Black G et al. Human gene expression profiles of susceptibility and resistance in tuberculosis. Genes Immun 2011; 12: 15–22. 27 Hori S, Sakaguchi S. Foxp3: a critical regulator of the development and function of regulatory T cells. Microbes Infect 2004; 6: 745–751. 28 Johanns TM, Ertelt JM, Rowe JH, Way SS. Regulatory T cell suppressive potency dictates the balance between bacterial proliferation and clearance during persistent Salmonella infection. PLoS Pathog 2010; 6: e1001043. 29 Guyot-Revol V, Innes JA, Hackforth S, Hinks T, Lalvani A. Regulatory T cells are expanded in blood and disease sites in patients with tuberculosis. Am J Respir Crit Care Med 2006; 173: 803–810. 30 Wu B, Huang C, Kato-Maeda M, Hopewell PC, Daley CL, Krensky AM et al. Messenger RNA expression of IL-8, FOXP3, and IL-12beta differentiates latent tuberculosis infection from disease. J Immunol 2007; 178: 3688–3694. 31 He XY, Xiao L, Chen HB, Hao J, Li J, Wang YJ et al. T regulatory cells and Th1/Th2 cytokines in peripheral blood from tuberculosis patients. Eur J Clin Microbiol Infect Dis 2010; 29: 643–650. 32 Burl S, Hill PC, Jeffries DJ, Holland MJ, Fox A, Lugos MD et al. FOXP3 gene expression in a tuberculosis case contact study. Clin Exp Immunol 2007; 149: 117–122. 33 Roberts T, Beyers N, Aguirre A, Walzl G. Immunosuppression during active tuberculosis is characterized by decreased interferongamma production and CD25 expression with elevated forkhead box P3, transforming growth factorbeta, and interleukin-4 mRNA levels. J Infect Dis 2007; 195: 870–878. 34 Toossi Z, Gogate P, Shiratsuchi H, Young T, Ellner JJ. Enhanced production of TGFbeta by blood monocytes from patients with active tuberculosis and presence of TGF-beta in tuberculous granulomatous lung lesions. Journal of immunology 1995; 154: 465–473. 35 Olmos S, Stukes S, Ernst JD. Ectopic activation of Mycobacterium tuberculosisspecific CD4 þ T cells in lungs of CCR7 / mice. J Immunol 2010; 184: 895–901.

36 Bankaitis VA, Malehorn DE, Emr SD, Greene R. The Saccharomyces cerevisiae SEC14 gene encodes a cytosolic factor that is required for transport of secretory proteins from the yeast Golgi complex. J Cell Biol 1989; 108: 1271–1281. 37 Juffermans NP, Verbon A, van Deventer SJ, Buurman WA, van Deutekom H, Speelman P et al. Serum concentrations of lipopolysaccharide activity-modulating proteins during tuberculosis. J Infect Dis 1998; 178: 1839–1842. 38 Vishwanath V, Meera R, Puvanakrishnan R, Narayanan PR. Fate of Mycobacterium tuberculosis inside rat peritoneal macrophages in vitro. Mol Cell Biochem 1997; 175: 169–175. 39 Ruhwald M, Bjerregaard-Andersen M, Rabna P, Kofoed K, Eugen-Olsen J, Ravn P. CXCL10/IP-10 release is induced by incubation of whole blood from tuberculosis patients with ESAT-6, CFP10 and TB7.7. Microbes Infect 2007; 9: 806–812. 40 Ruhwald M, Bjerregaard-Andersen M, Rabna P, Eugen-Olsen J, Ravn P. IP-10, MCP1, MCP-2, MCP-3, and IL-1RA hold promise as biomarkers for infection with M. tuberculosis in a whole blood based T-cell assay. BMC Res Notes 2009; 2: 19. 41 Frahm M, Goswami ND, Owzar K, Hecker E, Mosher A, Cadogan E et al. Discriminating between latent and active tuberculosis with multiple biomarker responses. Tuberculosis 2011; 91: 250–256. 42 Kim SY, Park MS, Kim YS, Kim SK, Chang J, Lee HJ et al. The responses of multiple cytokines following incubation of whole blood from TB patients, latently infected individuals, and controls with the TB antigens ESAT-6, CFP-10, and TB7.7. Scand J Immunol 2012; 76: 580–586. 43 Chandra RK. Nutrition and the immune system: an introduction. Am J Clin Nutr 1997; 66: 460S–463S. 44 Berry MP, Graham CM, McNab FW, Xu Z, Bloch SA, Oni T et al. An interferoninducible neutrophil-driven blood transcriptional signature in human tuberculosis. Nature 2010; 466: 973–977. 45 Mengatto L, Chiani Y, Imaz MS. Evaluation of rapid alternative methods for drug susceptibility testing in clinical isolates of Mycobacterium tuberculosis. Mem Inst Oswaldo Cruz 2006; 101: 535–542. 46 Mathur HC, Saxena S, Bhardwaj RM. Evaluation of Kenneth Jones’ criteria for diagnosis of childhood tuberculosis. Indian J Pediatr 1974; 41: 349–355. 47 Hussain R, Kaleem A, Shahid F, Dojki M, Jamil B, Mehmood H et al. Cytokine profiles using whole-blood assays can discriminate between tuberculosis patients and healthy endemic controls in a BCG-vaccinated population. J Immunol Met 2002; 264: 95–108. 48 Pokkali S, Das SD. Augmented chemokine levels and chemokine receptor expression on immune cells during pulmonary tuberculosis. Hum Immunol 2009; 70: 110–115. 49 Goeman JJ, Van De Geer SA, Van Houwelingen HC. Testing against a high dimensional alternative. J R Stat Soc B 2006; 68: 477–493. 50 Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. 2nd EdnSpringer, 2009 68–78 51 Goeman JJ, van de Geer SA, de Kort F, van Houwelingen HC. A global test for groups of genes: testing association with a clinical outcome. Bioinformatics 2004; 20: 93–99. 52 Friedman J, Hastie T, Tibshirani R. Regularization Paths for Generalized Linear Models via Coordinate Descent. J Stat Software 2010; 33: 1–22.

Supplementary Information accompanies this paper on Genes and Immunity website (http://www.nature.com/gene)


& 2013 Macmillan Publishers Limited