Performance of Purified Antigens for Serodiagnosis of Pulmonary Tuberculosis: a Meta-Analysis

CLINICAL AND VACCINE IMMUNOLOGY, Feb. 2009, p. 260 276 Vol. 16, No. 2 1556-6811/09/$08.00 0 doi:10.1128/cvi.00355-08 Copyright 2009, American Society for Microbiology. All Rights Reserved. Performance of Purified Antigens for Serodiagnosis of Pulmonary Tuberculosis: a Meta-Analysis Karen R. Steingart, 1 * Nandini Dendukuri, 2 Megan Henry, 3 Ian Schiller, 2 Payam Nahid, 4 Philip C. Hopewell, 1,4 Andrew Ramsay, 5 Madhukar Pai, 2 and Suman Laal 6,7,8 Francis J. Curry National Tuberculosis Center, University of California, San Francisco, California 1 ; Department of Epidemiology, Biostatistics & Occupational Health, McGill University, Montréal, Quebec, Canada 2 ; San Joaquin County Public Health Services, Stockton, California 3 ; Division of Pulmonary and Critical Care Medicine, San Francisco General Hospital, University of California, San Francisco, California 4 ; UNICEF/UNDP/World Bank/WHO Special Programme for Research and Training in Tropical Diseases, World Health Organization, Geneva, Switzerland 5 ; Departments of Pathology 6 and Microbiology, 7 New York University Langone Medical Center, New York, New York; and Veterans Affairs Medical Center, New York, New York 8 Received 26 September 2008/Returned for modification 4 November 2008/Accepted 24 November 2008 Serological antibody detection tests for tuberculosis may offer the potential to improve diagnosis. Recent metaanalyses have shown that commercially available tests have variable accuracies and a limited clinical role. We reviewed the immunodiagnostic potential of antigens evaluated in research laboratories (in-house) for the serodiagnosis of pulmonary tuberculosis and conducted a meta-analysis to evaluate the performance of comparable antigens. Selection criteria included the participation of at least 25 pulmonary tuberculosis patients and the use of purified antigens. Studies evaluating 38 kda, MPT51, malate synthase, culture filtrate protein 10, TbF6, antigen 85B, -crystallin, 2,3-diacyltrehalose, 2,3,6-triacyltrehalose, 2,3,6,6 -tetraacyltrehalose 2 -sulfate, cord factor, and TbF6 plus DPEP (multiple antigen) were included in the meta-analysis. The results demonstrated that (i) in sputum smear-positive patients, sensitivities significantly >50% were provided for recombinant malate synthase (73%; 95% confidence interval [CI], 58 to 85) and TbF6 plus DPEP (75%; 95% CI, 50 to 91); (ii) protein antigens achieved high specificities; (iii) among the lipid antigens, cord factor had the best overall performance (sensitivity, 69% [95% CI, 28 to 94]; specificity, 91% [95% CI, 78 to 97]); (iv) compared with the sensitivities achieved with single antigens (median sensitivity, 53%; range, 2% to 100%), multiple antigens yielded higher sensitivities (median sensitivity, 76%; range, 16% to 96%); (v) in human immunodeficiency virus (HIV)-infected patients who are sputum smear positive, antibodies to several single and multiple antigens were detected; and (vi) data on seroreactivity to antigens in sputum smear-negative or pediatric patients were insufficient. Potential candidate antigens for an antibody detection test for pulmonary tuberculosis in HIV-infected and -uninfected patients have been identified, although no antigen achieves sufficient sensitivity to replace sputum smear microscopy. Combinations of select antigens provide higher sensitivities than single antigens. The use of a case-control design with healthy controls for the majority of studies was a limitation of the review. Efforts are needed to improve the methodological quality of tuberculosis diagnostic studies. * Corresponding author. Mailing address: Francis J. Curry National Tuberculosis Center, University of California, San Francisco, 3180 18th Street, Suite 101, San Francisco, CA 94110-2028. Phone: (415) 502-4600. Fax: (415) 502-4620. E-mail: karenst@u.washington.edu. Supplemental material for this article may be found at http://cvi.asm.org/. Present address: California Department of Public Health, Sacramento, CA. Published ahead of print on 3 December 2008. The failure to diagnose tuberculosis (TB) accurately and rapidly is a key challenge in curbing the epidemic (45, 88, 116). Sputum microscopy, currently the sole diagnostic test in most areas where TB is endemic, has several limitations; in particular, the sensitivity compared with that of culture is variable (80, 97, 104, 116), multiple patient visits are required (56, 93, 114), considerable technical training is necessary, and the procedure is labor-intensive (45, 65). Antibody detection tests (serological tests) are used for the diagnosis of many infectious s and could potentially improve the means of diagnosis of TB. These tests measure the presence of specific antibodies (most often immunoglobulin G [IgG]) directed against immunodominant antigens of the pathogen in question. Compared with microscopy, antibody detection methods may enable the rapid diagnosis of TB, as these tests have the advantages of speed (results can be available within hours), technological simplicity, and minimal training requirements. In addition, these tests can be adapted to point-of-care formats that can be implemented at lower levels of health services in low- and middle-income countries (21, 22, 57, 65). Efforts to develop antibody detection tests for the diagnosis of TB have been under way for decades, and the performance of these tests has been well described (13, 17, 22, 32, 40, 47, 48, 52, 60, 64, 100, 107). Several systematic reviews of these tests have been published (discussed below) (28, 94, 95). First-generation antibody detection tests were based on crude mixtures of constituents and products of Mycobacterium tuberculosis, for example, culture filtrate proteins and purified protein derivative, the preparation used in the tuberculin skin test. Several of these early tests had low specificities, as the tests contained antigens shared among different bacterial species (1, 22, 48, 57). During the past two decades, an increased understanding of humoral immune responses to M. tuberculosis and the new tools of 260

VOL. 16, 2009 META-ANALYSIS OF SERODIAGNOSIS OF PULMONARY TB 261 FIG. 1. Flow diagram for selection of subgroups, IgG and/or IgA antibody detection: an example with MPT51. The same sequence of steps was repeated for each antigen. Having at least four studies available was a condition for inclusion in the meta-analysis., other antibody combinations included IgG and IgM (n 3 studies); IgM and IgA (n 0); IgG, IgA, and IgM (n 7); and not reported (n 10);, other recombinant antigens included 38 kda (n 13 studies), CFP-10 (n 9), malate synthase (n 8), TbF6 (n 4), -cystallin (n 4), Mtb48 (n 3), Ag85C (n 2), DPEP (n 2), ESAT6 (n 2), and other antigens (n 16) that appeared in only one study each. Downloaded from http://cvi.asm.org/ genomics and proteomics have led to the discovery of new antigens reported to provide improved sensitivities and specificities for the diagnosis of TB compared with those achieved with the antigens in the first-generation tests (48). We reviewed the immunodiagnostic potential of different antigens evaluated in research laboratories (in-house) for the serodiagnosis of pulmonary TB and carried out a meta-analysis to evaluate the performance of various antigens singly and in combination. Previous meta-analyses have shown that commercially available serological tests for both pulmonary TB (94) and extrapulmonary TB (95) have variable accuracies and, consequently, a limited clinical role. Another systematic review (searches through 2003) limited studies to the cohort or case series type of design and included only nine studies relating to in-house anti-tb antibody serological tests (28). A recently published expert review (1) did not include a meta-analysis. We are unaware of other systematic reviews on this topic. The current review addresses the following questions. (i) What is the performance of different antigens in the serodiagnosis of pulmonary TB in sputum smear-positive and smearnegative patients? (ii) What is the performance of these antigens in the serodiagnosis of pulmonary TB in patients with human immunodeficiency virus (HIV) infection? MATERIALS AND METHODS Standard guidelines and methods for systematic reviews and meta-analyses of diagnostic tests were followed (25, 31, 61). The following electronic databases (1990 to November 7, 2007) were queried for primary studies in the English language: PubMed, EMBASE, Biosis, and Web of Science. The search terms included tuberculosis, Mycobacterium tuberculosis, immunological tests, serological tests, antibody detection, antigen detection, ELISA (enzyme-linked immunosorbent assay), Western blot, and sensitivity and specificity. Additional studies were identified by contacting experts and searching the reference lists of primary studies and review articles. The criteria for including studies for the review were as follows. Cross-sectional and case-control study designs were eligible. The sample size had to be at least 25 patients with sputum smear-positive or smear-negative pulmonary TB who provided sera before or within 14 days of receiving antituberculous treatment. For comparison with TB patients, we selected only one group for each study, preferentially, patients in whom pulmonary TB was initially suspected but was later ruled out, as opposed to healthy participants. The index test (serological antibody detection) had to be evaluated in-house with purified antigens; studies that used purified protein derivative, culture filtrates, or sonicated antigens were not included. The reference standard was either the isolation of M. tuberculosis on sputum culture or, for studies conducted in countries where TB is endemic ( 20 cases per 100,000 population in 2005) where cultures are not routinely performed, the presence of acid-fast bacilli detected by sputum smear microscopy (16, 115). For the determination of outcome measures, there had to be sufficient data to construct a two-by-two table for calculations of sensitivity, specificity, and likelihood ratios. The following studies were excluded: (i) studies whose results were published before 1990, for the reason that many studies used crude antigen extracts or obsolete methods; (ii) studies of latent M. tuberculosis infection; (iii) studies of nontuberculous mycobacteria; (iv) studies describing nonimmunologic methods for the detection of antibodies; (v) studies in the basic science literature concerning cloning of new antigens or their immunologic properties (e.g., epitope mapping); and (vi) case reports and reviews. Study selection. Initially, two reviewers independently screened citations retrieved from all sources for relevance. Screening of full-text articles by using prespecified inclusion criteria was carried out by two reviewers, and the articles on October 19, 2018 by guest

262 STEINGART ET AL. CLIN. VACCINE IMMUNOL. Downloaded from http://cvi.asm.org/ FIG. 2. Flow of studies through the review process. PTB, pulmonary tuberculosis. included were checked by a third reviewer. Disagreements were resolved by consensus. Data extraction. A data extraction form was created and pilot tested with a subset of eligible studies and then finalized. Two reviewers (each of whom was responsible for approximately 50% of the studies) extracted data from all included studies with the standardized form. To verify reproducibility, a third reviewer independently performed data extraction on all studies. Differences among reviewers were resolved by consensus. When necessary, authors were contacted for additional information. Assessment of study quality. The quality of studies was appraised by using a subset of criteria from QUADAS, a validated tool for diagnostic studies (see Table S1 in the supplemental material) (110). Antigen classification. Antigens were classified into five categories according to the type of compound: (i) recombinant proteins, (ii) native proteins, (iii) lipids, (iv) multiple antigens (protein-protein or lipid-lipid additive reactivity), and (v) protein-lipid antigens. Several investigators evaluated antibody responses to multiple antigens in the same patient population to enhance sensitivity. These studies have taken two approaches. In some cases, different antigens (or portions thereof) have been cloned as single protein entities (polyproteins) and tested for their reactivities with sera. In other cases, multiple antigens have been tested as single entities and cumulative results (additive reactivity) were calculated. In the former case, we considered polyproteins to be single antigens; in the latter case, we classified the entities as multiple antigens. Data analysis. Estimates of sensitivity and specificity from individual studies and their exact 95% confidence intervals (CIs) were obtained by using Meta- DiSc (version 1.4) software (117). Sensitivity refers to the proportion of TB patients with positive test results; specificity refers to the proportion of participants without TB with negative test results. For sensitivity, we included studies that used sputum smear as the reference standard along with studies that used culture. For specificity, we noted the type of comparison group, e.g., healthy participants or patients with nontuberculous. Likelihood ratio positive was calculated as sensitivity/(1 specificity); likelihood ratio negative was calculated as (1 sensitivity)/specificity. Selection of subgroups for meta-analysis. We recognized that studies were heterogeneous in many respects, particularly concerning antigen characteristics and antibody class. Therefore, in order to address heterogeneity and combine study results, subgroups of comparable antigens were determined (Fig. 1). Initially, studies were grouped by the class of antibody detected by the test: (i) IgG and/or IgA, (ii) IgM, and (iii) other IgM-containing combinations (IgM-IgG, IgM-IgA, and IgM-IgG-IgA). This division was based on the understanding that IgM antibodies are expressed transiently and earlier in infection than other antibodies. Next, studies were stratified by antigen number (single or multiple antigens); the type of compound (protein or lipid); and, for proteins, the source of the compound (recombinant or native). Finally, for each distinct single antigen or multiple-antigen combination, studies were stratified by patient sputum smear status and HIV status. At least four studies were required to be available for inclusion in a subgroup in order to strengthen the results and reduce the possibility of finding a significant result by chance. In this way, we identified 16 subgroups. To summarize sensitivity and specificity within each subgroup, separate metaanalyses were performed by using the hierarchical summary receiver operating characteristic curve model (72). The advantages of the hierarchical summary receiver operating characteristic are that it jointly models both sensitivity and specificity, weights studies according to the number of participants, and takes into account unmeasured heterogeneity between studies by using random effects (31). The model was estimated by using a Bayesian approach with noninformative prior distributions and was implemented with WinBUGS (version 1.4.1) software program (91). The average sensitivity, specificity, and likelihood ratios from each metaanalysis were estimated. From the posterior distribution of each parameter of on October 19, 2018 by guest

VOL. 16, 2009 META-ANALYSIS OF SERODIAGNOSIS OF PULMONARY TB 263 TABLE 1. Guide to tables and figures in the review Table or Category or antigen Figure Antigen names...table 2 Characteristics of study quality...table 3 Meta-analysis...Table 4 Recombinant malate synthase...table 5 Recombinant CFP-10...Table 6 DAT...Table 7 Specificity estimates by control group...table 8 Antigen performance by Ig class...table 9 Recombinant 38 kda...table A1 Recombinant MPT51...Table A2 Native 38 kda...table A3 Native Ag85B...Table A4 Questions for quality assessment...table S1 Recombinant protein antigens...table S2 Native protein antigens...table S3 Lipid antigens...table S4 Multiple antigens...table S5 Protein/lipid antigens...table S6 Flow diagram for selection of subgroups...figure 1 Flow of studies in the review...figure 2 SROC curve for recombinant proteins...figure 3A SROC curve for native proteins...figure 3B SROC curve for lipid antigens...figure 3C Sensitivity, TB-HIV coinfection...figure 4A Specificity, TB-HIV coinfection...figure 4B interest, we extracted the mean and the 95% credible interval (the Bayesian equivalent of the classical confidence interval) on the basis of the 2.5% and 97.5% quartiles. When feasible, specificity estimates were stratified by type of comparison group. Finally, a summary receiver operating characteristic (SROC) curve from each meta-analysis was obtained. The SROC curve plots sensitivity versus 1 specificity for the range of specificity values observed for each study, as extrapolation beyond this range is not advisable (42). The SROC curve gives an idea of the overall performance of a test across different thresholds (54, 61). The closer that the curve is to the upper-left-hand corner of the plot (sensitivity and specificity are both 100%), the better the performance of the test is (42). The plots were made by using the R (version 2.6.1) software program (70). Descriptive analysis. Descriptive analyses were performed by using SPSS (version 14.0.1.366) software (92). Forest plots were made by using Meta-DiSc (version 1.4) software (117). RESULTS Description of studies included. The literature searches identified over 5,000 citations, of which 49 publications (254 studies) were included (Fig. 2) (2 4, 6, 8, 10, 12, 15, 18 20, 23, 24, 26, 27, 29, 33 37, 39, 43, 44, 55, 58, 59, 66 69, 73, 76, 77, 81 87, 99, 101 103, 105, 108, 109, 118). Mycobacterial culture was used as the reference standard in 199 (78%) studies, sputum smear was used as the reference standard in 29 (11%) studies, and sputum smear and/or culture was used as the reference standard in 26 (10%) studies. Two hundred thirteen (84%) studies involved smear-positive patients, and 41 (16%) involved culture-confirmed smear-negative patients. Four studies involved children younger than 15 years of age, and 30 (12%) studies involved HIV-infected persons. The vast majority (96%) of studies performed antibody tests by ELISA. The median number of participants with TB was 51 (interquartile range, 39 to 105); the median number of participants without TB was 57 (interquartile range, 35 to 83). Two hundred fifty-four studies evaluating 51 distinct single antigens (9 native proteins, 27 recombinant proteins, and 15 TABLE 2. Antigens evaluated for serodiagnosis of pulmonary TB Name(s) of antigen(s) a Protein Rv designation Reference(s) Ag85C, 32.5 kda, FbpC2, MPT-45 0129c 76, 77 38 kda, Ag 5, PstS1, PhoS, PhoS1 0934 3, 8, 15, 18, 20, 27, 33, 35, 37, 55, 59, 69, 76, 77, 81, 102 Mtb 81/88 kda, malate synthase, GlcB 1837c 20, 35, 37, 76, 85, 86, 108 DPEP, MPT32; 45/47 kda, Apa, 1860 19, 26, 76 ModD Ag85B 1886c 23, 67, 77, 103 16 kda; -crystallin; 14 kda, 2031c 33, 39, 66, 103 HspX, Acr 27 kda; MPT51, FbpC1, MPB 51 3803c 10, 69, 85, 108 CFP-10, MTSA-10, EsxB; Lhp, 3874 27, 33, 59, 118 Mtb11 ESAT-6 3875 33, 118 Mtb48 3881c 55 DAT 43, 44, 83, 105 TAT 43, 44 SL-I, sulfolipid I 43, 44 Cord factor 43, 44, 101 a Boldface indicates the name of the antigen in common use. lipids) and 30 distinct multiple-antigen combinations were identified. Many of these antigens were evaluated in only one study. In order to accommodate the large number of antigens identified in the review, only those antigens appearing in two or more studies are included in Tables S2 though S6 in the supplemental material. A guide to the tables and figures is presented in Table 1. The antigens and their alternative names are listed in Table 2. The most frequently evaluated antigens are described below and in additional tables and figures, as noted. Assessment of study quality. The majority of studies used a case-control study design. Only 65 (26%) studies reported blinded interpretation of index test results. Almost all studies provided sufficient detail describing the execution of the index test (Table 3). Meta-analysis (Table 4 and Fig. 3). (i) Recombinant proteins. (a) Recombinant 38 kda (Rv0934) (Table A1). 38 kda, a major protein present in culture filtrates of M. tuberculosis, has been studied extensively (1, 17, 22). Several studies have shown an association between the presence of anti-38 kda antibodies and advanced cavitary TB (14, 22, 75). In smear-positive patients, recombinant 38 kda yielded a sensitivity of 47% (95% CI, 39 to 55) and a specificity of 94% (95% CI, 86 to 98) (8, 27, 33, 35, 37, 55, 77, 81). (b) Recombinant malate synthase (Rv1837c) (Table 5). Malate synthase (81 kda), present in M. tuberculosis culture filtrates, the cell wall, and cytoplasmic subcellular fractions, is an enzyme of the glyoxylate pathway used by M. tuberculosis during intracellular replication in macrophages (90) and has adapted to function as an adhesin that enhances bacterial adherence to host cells (46). In sputum smear-positive patients, malate synthase achieved a sensitivity of 73% (95% CI, 58 to 85) and a specificity of 98% (95% CI, 95 to 100) (35, 37, 85, 86, 108). The likelihood ratio positive (40.78) was considerably higher for malate synthase than for other antigens. Earlier studies have

264 STEINGART ET AL. CLIN. VACCINE IMMUNOL. TABLE 3. Characteristics of study quality Characteristic No. (%) of studies Study design Cross-sectional... 39 (15)...208 (82) Nested within observational study... 7 (3) Recruitment of participants Consecutive or random... 20 (8) Convenience or not reported...234 (92) Selection criteria clearly described...141 (56) Complete verification by use of the reference standard...107 (42) Execution of test described in sufficient detail...253 (100) a Index test results blinded to reference standard? Yes... 65 (26) No... 1 (0) Not reported...188 (74) a The description of the test execution was deemed insufficient in one study. demonstrated that whereas antibodies to the 38-kDa antigen are present in patients with extensive cavitary lesions, antimalate synthase antibodies are elicited earlier during the progression of TB, being present in patients who have not yet developed cavities (74). This is reflected in the higher sensitivity of TB diagnosis provided by malate synthase. (c) Recombinant MPT51 (Rv3803c) (Table A2). The 27-kDa protein MPT51, a culture filtrate protein, is closely related to the antigen 85 (Ag85) complex, which comprises Ag85A, Ag85B, and Ag85C. MPT51 is an adhesin (108) reported to be a fibronectin-binding protein of M. tuberculosis (113). A sufficient number of studies evaluating the performance of MPT51 were available to stratify the results by HIV infection status. In sputum smear-positive patients, MPT51 provided equivalent sensitivities in both HIV-negative TB patients (59%; 95% CI, 38 to 76) and HIV-positive TB patients (58%; 95% CI, 30 to 82); the specificities were 94% and 97%, respectively (10, 69, 85, 108). (d) Recombinant CFP-10 (Rv3874) (Table 6). Culture filtrate protein 10 (CFP-10), a culture filtrate and cell wall protein, has been identified as one of the earliest proteins expressed by M. tuberculosis during culture in bacteriological media (9). In sputum smear-positive patients, CFP-10 provided a sensitivity of 48% (95% CI, 29 to 68) and a specificity of 96% (95% CI, 83 to 99) (27, 33, 59, 118). (e) Recombinant TbF6 (see Table S2 in the supplemental material). TbF6 is a single antigen combining four distinct antigens (CFP-10, MTB8, MTB48, and 38 kda) as a genetically fused polyprotein (37). In sputum-smear positive patients, TbF6 achieved a sensitivity of 70% (95% CI, 37 to 90) and a specificity of 93% (95% CI, 69 to 99) (6, 37). The high sensitivity obtained with TbF6 is likely due to the fact that it comprises immunogenic domains from multiple antigens. (ii) Native proteins. (a) Native 38 kda (Rv0934) (Table A3). In sputum smear-positive patients, native 38 kda provided a sensitivity of 49% (95% CI, 37 to 61). In sputum smear-negative patients, the sensitivity reported was lower (31%; 95% CI, 15 to 52). Specificities were 97% in both subgroups (15, 18, 59, 69, 76, 77, 102). (b) Native Ag85B (Rv1886c) (Table A4). Ag85B, present in M. tuberculosis culture filtrates and cell walls, is a major component of the Ag85 complex (112). Like MPT51, Ag85B is a fibronectin-binding protein (113). In HIV-negative TB patients, native Ag85B yielded a sensitivity of 53% (95% CI, 20 to 83), and in HIV-positive TB patients, a it yielded a sensitivity of 62% (95% CI, 19 to 92). The specificities were 95% (23, 67, 77, 103). (c) Native -crystallin (2031c) (see Table S3 in the supplemental material). -Crystallin is a 14/16-kDa cell wall protein (106) shown to be induced in bacteria under hypoxia (78). In sputum TABLE 4. Overall sensitivities, specificities, and likelihood ratios for antigens evaluated for serodiagnosis of pulmonary TB with assays detecting IgG and/or IgA antibodies Type of compound Antigen name Rv designation studies Smear status HIV status Sensitivity Specificity Likelihood ratio (%) a (%) a positive a Likelihood ratio negative a Recombinant 38 kda 0934 12 Positive / 47 (39 55) 94 (86 98) 8.22 (3.41 24.85) 0.56 (0.48 0.65) Malate synthase 1837c 8 Positive / 73 (58 85) 98 (95 100) 40.78 (14.43 155.7) 0.27 (0.16 0.42) MPT51 3803c 5 Positive 59 (38 76) 94 (77 99) 10.50 (2.70 69.69) 0.44 (0.26 0.67) MPT51 3803c 4 Positive 58 (30 82) 97 (84 100) 19.03 (3.73 172.3) 0.44 (0.19 0.73) CFP-10 3874 6 Positive / 48 (29 68) 96 (83 99) 12.11 (3.20 64.63) 0.55 (0.35 0.73) TbF6 b 4 Positive 70 (37 90) 93 (69 99) 9.61 (2.23 53.99) 0.33 (0.13 0.66) TbF6, DPEP c 4 Positive 75 (50 91) 95 (86 99) 14.97 (5.43 56.66) 0.26 (0.10 0.53) Native protein 38 kda 0934 13 Positive / 49 (37 61) 97 (94 99) 15.73 (8.84 31.55) 0.53 (0.41 0.65) 38 kda 0934 7 Negative 31 (15 52) 97 (92 99) 9.13 (3.88 24.05) 0.72 (0.51 0.87) Ag 85B 1886c 4 Positive 53 (20 83) 95 (78 99) 9.36 (2.52 53.81) 0.51 (0.20 0.84) Ag 85B 1886c 4 Positive 62 (19 92) 97 (89 99) 17.83 (4.04 62.32) 0.39 (0.08 0.84) -Crystallin 2031c 6 Positive / 54 (32 75) 96 (83 99) 13.23 (3.52 66.61) 0.48 (0.28 0.71) Lipid DAT 7 Positive / 63 (45 78) 81 (50 96) 3.32 (1.32 13.35) 0.47 (0.30 0.74) TAT 4 Positive / 81 (21 99) 44 (24 67) 1.44 (0.42 2.31) 0.42 (0.03 1.71) SL-I 4 Positive / 80 (56 93) 59 (8 96) 1.94 (0.89 20.90) 0.34 (0.14 2.22) Cord factor 5 Positive / 69 (28 94) 91 (78 97) 7.03 (2.44 20.65) 0.35 (0.06 0.80) a The data represent the posterior means (95% credible intervals). b Polyprotein. c Multiple antigen (additive reactivity).

VOL. 16, 2009 META-ANALYSIS OF SERODIAGNOSIS OF PULMONARY TB 265 FIG. 3. SROC curves of antigen performance for serodiagnosis of pulmonary TB. (A) Recombinant proteins; (B) native proteins; (C) lipids. smear-positive patients, -crystallin provided a sensitivity of 48% (95% CI, 29 to 68) and a specificity of 96% (95% CI, 83 to 99) (66, 103). (iii) Lipids. A variety of lipid-containing antigens are common to mycobacterial species (71). Several lipid moieties have been purified and intensely studied for their serological potential for TB diagnosis (44). Four lipid antigens, all acylated trehaloses (107), were evaluated in smear-positive patients and included in the meta-analysis. (a) DAT (Table 7). 2,3-Diacyltrehalose (DAT), a component of the M. tuberculosis cell wall, has been postulated to play a role in modulating host immune responses (50). DAT yielded a sensitivity of 63% (95% CI, 45 to 78) and a specificity of 81% (95% CI, 50 to 96) (43, 44, 83, 105). (b) TAT (see Table S4 in the supplemental material). 2,3,6- Triacyltrehalose (TAT) is an antigenic glycolipid compound found in the M. tuberculosis cell wall (41, 43). TAT provided a sensitivity of 81% (95% CI, 21 to 99) and a specificity of 44% (95 CI, 24 to 67) (43, 44). (c) SL-I (see Table S4 in the supplemental material). 2,3,6,6 - Tetraacyltrehalose 2 -sulfate (sulfolipid I [SL-I]), a compound found abundantly in the M. tuberculosis cell wall, may affect the human immune system and play a role in the pathogenesis of TB (51). SL-I yielded a sensitivity of 80% (95% CI, 56 to 93) and a specificity of 59% (95% CI, 8 to 96) (43, 44). (d) Cord factor (see Table S4 in the supplemental material). Cord factor (trehalose 6,6 -dimycolate), a major component of M. tuberculosis cell walls, is named for its central role in aggregating mycobacteria into cord structures (7, 30). Cord factor may contribute to the virulence of M. tuberculosis by facilitating cavity formation (38). Cord factor achieved a sensitivity of 69% (95% CI, 28 to 94) and a specificity of 91% (95% CI, 78 to 97) (43, 44, 101). (e) TbF6 plus DPEP (see Table S5 in the supplemental material). TbF6 polyprotein plus DPEP was the multiple-antigen combination most frequently evaluated; four studies involved HIV-uninfected individuals, and one study involved HIV-infected individuals. As described above, TbF6 is a polyprotein. DPEP, also known as MPT32, is a proline-rich 45/47-kDa antigen suggested to have a role in the cross-linking of molecules produced by or bordering M. tuberculosis (49). Earlier studies with native MPT32 have demonstrated that it is a highly immunogenic protein that provided higher sensitivity than the 38-kDa protein when it was tested in the same patient cohort (74). In HIV-negative TB patients, TbF6 plus DPEP achieved a sensitivity of 75% (95% CI, 50 to 91) and a specificity of 94% (95% CI, 86 to 99) (6, 37). The single study evaluating the serodiagnostic potential of TbF6 plus DPEP in HIV-infected individuals is described below. Assessment of specificity in healthy volunteers compared with assessment of specificity in patients with nontuberculous s (Table 8). Sufficient numbers of studies evaluated five antigens, four proteins (recombinant 38 kda, native 38 kda, malate synthase, and CFP-10) and one lipid (DAT) for comparison of the specificities for healthy and d controls. For the four proteins, both subsets showed similar specificity values. For DAT, studies involving patients with nontuberculous yielded a significantly higher specificity, 57% (95% CI, 30 to 76), than studies with healthy volunteers, 97% (95% CI, 88 to 100).

266 STEINGART ET AL. CLIN. VACCINE IMMUNOL. TABLE 5. Studies evaluating recombinant malate synthase (Rv1837c) for serodiagnosis of pulmonary TB Author, yr (reference) Study design Reference standard (smear status a ) Patient (comparison) country Status of individuals used for comparison HIV status of patient (comparator) Ig class participants b Sensitivity Specificity (%) c (%) c Chaudhary et al., 2005 (20) Hendrickson et al., 2000 (35) Hendrickson et al., 2000 (35) Houghton et al., 2002 (37) Houghton et al., 2002 (37) Singh et al., 2003 (86) Singh et al. 2005 (85) Wanchu et al., 2008 (108) Wanchu et al., 2008 (108) Smear (SP) India (India) Healthy NR d (NR) IgG, IgM 44/105 32 (19 48) 99 (93 100) Culture (SP) South Africa and Uganda (China) Culture (SP) South Africa and Uganda (China) Sub-Saharan Africa (China) Sub-Saharan Africa (China) Overall performance of antigens (Fig. 3). Among recombinant proteins, malate synthase and TbF6 plus DPEP (multiple antigen) provided the highest sensitivities for the specificities reported (Fig. 3A). For native proteins, Ag85B in HIV-infected TB patients achieved better performance than other native antigens (Fig. 3B). Among lipid antigens, cord factor had the best performance (Fig. 3C). Nontuberculous (NR) IgG 52/31 60 (45 73) 97 (83 100) Nontuberculous (NR) IgG 25/31 92 (74 98) 97 (83 100) Nontuberculous (NR) IgG 59/31 78 (65 88) 97 (83 100) Nontuberculous (NR) IgG 66/31 58 (45 70) 97 (83 100) Healthy (NR) IgG 43/25 77 (61 88) 100 (86 100) Smear (SP) India (area of endemicity) Smear (SP) India (area of Healthy (NR) IgG, IgA 40/29 75 (67 82) 100 (91 100) endemicity) Smear (SP) India (India) Healthy ( ) IgG, IgA 138/38 73 (52 88) 100 (91 100) Smear (SP) India (United States) a SP, smear positive. b Number of participants with TB/number of participants without TB. c 95% CIs are given in parentheses. d NR, not reported. Author, yr (reference) Dillon et al., 2000 (27) Greenaway et al., 2005 (33) Zhang et al., 2007 (118) Study design Nested within observational study Healthy ( ) IgG, IgA 26/65 73 (52 88) 99 (92 100) Descriptive analysis. (i) Performance of antigens in pulmonary TB patients with HIV infection (Fig. 4). Thirty studies evaluating antigens (all proteins) in HIV-infected TB patients were identified; all studies involved sputum smear-positive patients. Of the total, 23 (77%) studies evaluated assays for the detection of IgG and/or IgA antibodies (23, 35, 37, 69, 103, 108), four studies evaluated assays for the detection of IgM TABLE 6. Studies evaluating recombinant CFP-10 (Rv3874) for serodiagnosis of pulmonary TB Reference standard (smear status a ) Patient (comparison) country Brazil (United States and areas of endemicity) Status of individuals used for comparison HIV status of patient (comparator) Culture (SP) Gambia (Gambia) Healthy and ( and ) Culture (SP) India (India) Nontuberculous Culture (SP) India (India) Nontuberculous Culture (SP) India (India) Nontuberculous Ig class participants b Sensitivity Specificity (%) c (%) c Healthy ( ) IgG 250/57 28 (22 34) 97 (88 100) IgG 100/100 63 (53 72) 55 (45 65) ( ) IgG 262/76 42 (36 49) 99 (93 100) ( ) IgA 262/76 25 (20 31) 99 (93 100) ( ) IgG, IgA 262/76 57 (51 63) 97 (91 100) Smear (SP) China (China) Healthy ( ) IgG 50/28 78 (64 89) 96 (82 100) Culture (SN) India (India) Nontuberculous Culture (SN) India (India) Nontuberculous Culture (SN) India (India) Nontuberculous ( ) IgG 60/76 10 (4 21) 99 (93 100) ( ) IgA 60/76 43 (31 57) 99 (93 100) ( ) IgG, IgA 60/76 50 (37 63) 97 (91 100) a SP, smear positive; SN, smear negative. b Number of participants with TB/number of participants without TB. c 95% CIs are given in parentheses.

VOL. 16, 2009 META-ANALYSIS OF SERODIAGNOSIS OF PULMONARY TB 267 TABLE 7. Studies evaluating DAT for serodiagnosis of pulmonary TB Author, yr (reference) Julian et al., 2002 (44) Julian et al., 2002 (44) Julian et al., 2002 (44) Julian et al., 2004 (43) Julian et al., 2004 (43) Julian et al., 2004 (43) Study design Reference standard (smear status a ) Patient (comparison) country Status of individuals used for comparison Cross-sectional Culture (SP) Spain (Spain) Nontuberculous Cross-sectional Culture (SP) Spain (Spain) Nontuberculous Cross-sectional Culture (SP) Spain (Spain) Nontuberculous Cross-sectional Culture (SP) Spain (Spain) Nontuberculous Cross-sectional Culture (SP) Spain (Spain) Nontuberculous Cross-sectional Culture (SP) Spain (Spain) Nontuberculous Simonney et al., 1997 (83) Cross-sectional Culture (SP) France (France) Vera-Cabrera et Culture (SP) Mexico al., 1999 (105) e (Mexico) Vera-Cabrera et Culture (SP) Mexico al., 1999 (105) (Mexico) Simonney et al., Cross-sectional Culture France 1997 (83) (SN) (France) a SP, smear positive; SN, smear negative. b Number of participants with TB/number of participants without TB. c 95% CIs are given in parentheses. d NR, not reported. e Dot immunoassay was used; all other studies performed ELISA. HIV status patient (comparator) Ig class participants b Sensitivity Specificity (%) c (%) c and ( ) IgG 42/48 60 (43 74) 58 (43 72) and ( ) IgA 42/48 79 (63 90) 50 (35 65) and ( ) IgM 42/48 10 (3 23) 100 (93 100) ( ) IgG 29/35 52 (33 71) 57 (39 74) ( ) IgA 29/35 79 (60 92) 51 (34 69) ( ) IgM 29/35 7 (1 23) 100 (90 100) Healthy and IgG 31/50 32 (17 51) 96 (86 100) (NR d ) Healthy NR (NR) IgG 39/35 49 (32 65) 97 (85 100) Healthy NR (NR) IgG 39/35 80 (64 91) 97 (85 100) Healthy and (NR) IgG 29/50 21 (8 40) 96 (86 100) Downloaded from http://cvi.asm.org/ TABLE 8. Specificity estimates by type of comparison Antigen name Patients with nontuberculous Specificity (%) a Healthy subjects Recombinant 38 kda 97 (90 99) (6) 90 (57 99) (6) Recombinant malate synthase 97 (91 100) (4) 99 (81 100) (4) Recombinant CFP-10 99 (92 100) (3) 90 (43 99) (3) Native 38 kda 96 (90 99) (6) 98 (92 100) (4) DAT 55 (30 76) (4) 97 (88 100) (3) a The data represent the posterior means (95% credible intervals) (number of studies). antibodies (69, 103), and three studies evaluated assays for the detection of IgG plus IgA plus IgM antibodies (69, 103). Four studies based on multiple-antigen combinations are described in more detail below (35, 37, 108). Antibodies to all five single antigens (38 kda, malate synthase, Ag85B, -crystallin, and recombinant MPT51) evaluated in these studies were detected. As discussed, only MPT51 and Ag85B were investigated in a sufficient number of studies for inclusion in the meta-analysis (Table 4). With assays for the detection of IgG and/or IgA antibodies, the sensitivities reported for 38 kda (range, 35% to 68%) and -crystallin (range, 15% to 58%) were similar to those provided for MPT51 and Ag85B. Malate synthase achieved higher sensitivities (range, 73% to 92%). Compared with tests for the detection of only IgG and/or IgA antibodies, tests for the detection of IgM antibodies provided considerably lower sensitivities (range, 4% to 5%). The inclusion of tests for the detection of IgM (IgG plus IgA plus IgM) did not appreciably increase the sensitivity. The specificities provided by all of the above antigens were high (range, 89% to 100%). However, only 6 (20%) studies involved controls with nontuberculous (23, 35, 37), while 14 (47%) studies involved either healthy volunteers or HIV-infected individuals without TB (35, 37, 103, 108). In 10 (33%) studies, the control group involved HIV-infected individuals whose clinical status ranged from to symptomatic with opportunistic infections other than TB (69). (ii) Performance of tests with multiple antigens (see Table S5 in the supplemental material). Assays based on multiple antigens provided higher sensitivities (median, 76%; range, 16% to 96% [57 studies]) than assays based on single antigens (median, 53%; range, 2% to 100% [197 studies]), while they maintained high specificities (median, 96%; range, 79% to 100%) (data not shown). The combination of malate synthase plus MPT51 was evaluated in three studies, two studies involving HIV-negative TB patients (2, 108) and one study involving HIV-infected TB patients (108). The sensitivities provided by malate synthase plus MPT51 were similar with HIV-uninfected and -infected TB patients from India: 80% (95% CI, 73 to 87) and 77% (95% CI, 56 to 91), respectively. The specificities were equivalent (97%) whether the comparison group involved on October 19, 2018 by guest

268 STEINGART ET AL. CLIN. VACCINE IMMUNOL. FIG. 4. Performance of antigens for the serodiagnosis of pulmonary TB in HIV-infected patients. (A) Sensitivity; (B) specificity. The circles and the lines represent the point estimates and the 95% CIs, respectively. The size of the circle indicates the study size. MS, malate synthase; n, native; r, recombinant; 1, IgG; 2, IgM; 3, IgA; 4, IgG/A; 5, IgG/IgA/IgM.

VOL. 16, 2009 META-ANALYSIS OF SERODIAGNOSIS OF PULMONARY TB 269 TABLE 9. Sensitivity and specificity of in-house antibody detection tests by Ig class Ig class studies HIV-negative healthy volunteers from India or HIV-infected (tuberculin skin test-positive and -negative) individuals from the United States (108). However, this antigen combination yielded a sensitivity of only 55% (95% CI, 36 to 72) with TB patients from the United States (2). The combination of TbF6 plus DPEP plus malate synthase achieved a sensitivity of approximately 85% with both HIVnegative and HIV-infected TB patients (37). The specificities were high (97%; 95% CI, 83 to 100), even when this antigen combination was evaluated with patients with nontuberculous. In studies in which 38 kda plus malate synthase was evaluated, the sensitivities reported were 71% (95% CI, 57 to 83) for HIV-negative TB patients and 96% (95% CI, 80 to 100) for HIV-infected TB patients; the specificity was 89% (95% CI, 78 to 96) when it was assessed with healthy volunteers (35). With HIV-negative, sputum smearpositive patients, the combination of 38 kda plus Ag85B and -crystallin achieved a sensitivity of 89% (95% CI, 84 to 93) and, with the addition of MPT51, a sensitivity of 91% (95% CI, 86 to 95) (68). With non-hiv-infected, sputum smear-negative patients, the two combinations provided sensitivities of 73% (95% CI, 57 to 86) and 78% (95% CI, 62 to 89), respectively, and a specificity of 87% (95% CI, 75 to 94) when they were assessed with patients with nontuberculous s (68). Only two studies with multiple lipid antigens were identified (see Table S6 in the supplemental material). (iii) Test performance by Ig class (Table 9). Stratification by Ig class showed that in comparison with the results of studies that detected antibodies to IgG (median sensitivity, 61%; range, 8% to 100%) or IgA (median sensitivity, 40%; range, 10% to 90%), studies that detected antibodies to IgM had considerably lower sensitivities (median, 11%; range, 2% to 71%). The median specificities were similar: 96%, 96%, and 98%, respectively. In addition, compared with the results of tests that detected only anti-igg or anti-iga antibodies, tests that detected IgG plus IgA showed higher sensitivities (median, 71%; range, 43% to 97%). The inclusion of IgM (IgG plus IgA plus IgM) did not further enhance the sensitivity (median, 71%; range, 60% to 83%). DISCUSSION Sensitivity Median (range) % Specificity IgG 151 61 (8 100) 96 (26 100) IgA 25 40 (10 90) 96 (48 100) IgM 24 11 (2 71) 98 (89 100) IgG and IgA 34 71 (43 97) 97 (85 100) IgG and IgM 3 36 (32 52) 98 (98 99) IgM and IgA 0 IgG, IgA, and IgM 7 71 (60 83) 93 (90 93) Not reported 10 41 (16 74) 96 (89 99) All 254 58 (2 100) 96 (26 100) Principal findings. This systematic review yielded 254 studies evaluating 51 distinct single antigens and 30 multiple-antigen combinations. The performance of these antigens was examined in in-house tests for the serodiagnosis of pulmonary TB. Studies evaluating 13 distinct antigens (recombinant 38 kda, native 38 kda, MPT51, malate synthase, CFP-10, TbF6 polyprotein, Ag85B, -crystallin, DAT, TAT, SL-I, cord factor, and TbF6 plus DPEP [multiple antigen]) were included in the meta-analysis. The results demonstrate that (i) in sputum smear-positive patients, only recombinant malate synthase (sensitivity, 73%; 95% CI, 58 to 85) and TbF6 plus DPEP (sensitivity, 75%; 95% CI, 50 to 91) provided sensitivities significantly 50%; (ii) all protein antigens achieved high specificities; (iii) among the lipid antigens, cord factor had the best overall performance (sensitivity, 69% [95% CI, 28 to 94]; specificity, 91% [95% CI, 78 to 97]); (iv) compared with single antigens (median sensitivity, 53%; range, 2% to 100%), multiple antigens yielded higher sensitivities (median sensitivity, 76%; range, 16% to 96%); (v) in HIV-infected patients who are sputum smear positive, antibodies to several single and multiple antigens were detected; and (vi) data on seroreactivity to specific antigens in sputum smear-negative or pediatric patients were insufficient. These results demonstrate that no single antigen provides a sensitivity that is sufficient for a single antigen to be used to devise a serodiagnostic test for TB and that it is unlikely that a single antigen-based serodiagnostic test can be devised. This is not surprising, since the titers of antibodies to each antigen would differ in individuals and the detection of low titers of antibodies would be occluded due to the formation of immune complexes. It is also interesting that while both DPEP and malate synthase are conserved in the M. tuberculosis complex species and in all clinical isolates of M. tuberculosis whose genomes have been sequenced, antibodies to these antigens are not detected in a vast majority of tuberculin skin test-positive individuals with likely latent infection. Proteins that are approximately 50 to 60% homologous to these antigens are present in some other mycobacteria and whether cross-reactive antibodies exist in nontuberculous mycobacterial s remains to be reported. Stratification by Ig class demonstrated that assays for the detection of IgG and/or IgA antibodies provided higher sensitivities than assays for the detection of IgM antibodies. This is not surprising, since IgM antibodies are likely to be expressed early during the onset of infection, with the levels quickly decreasing after this period. By the time that bacteriologically detectable TB manifests, whether it is during primary infection or reactivation, the infection has already progressed for months to years in immunocompetent individuals and weeks to months in immunocompromised patients. Thus, the detection of IgM antibodies may have a role in the identification of early infection, but its value for the serodiagnosis of active TB may be limited. Considering that the profile of antigens recognized by antibodies is altered with the progression of M. tuberculosis infection (74), the antigens used in a serodiagnostic test during contact tracing are likely to differ from those used in a test for the diagnosis of clinical TB. To our knowledge, no antigens that can be the basis for an accurate serodiagnostic test for contact tracing have been reported. The discovery, evaluation, and comparison of such tests with gamma interferon release assays need to be considered. This systematic review and meta-analysis had several strengths. Standard protocols for the conduct of the review (61) and assessment of the quality of the studies (110) were

270 STEINGART ET AL. CLIN. VACCINE IMMUNOL. TABLE A1. Studies evaluating recombinant 38 kda (Rv0934) for serodiagnosis of pulmonary TB Author, yr (reference) Study design Reference standard (smear status a ) Patient (comparison) country Status of individuals used for comparison HIV status of patient (comparator) Ig class participants b Sensitivity Specificity (%) c (%) c Amicosante et al., 1995 (3) Amicosante et al., 1995 (3) Ben Amor et al., 2005 (8) Chaudhary et al., 2005 (20) Dillon et al., 2000 (27) Greenaway et al., 2005 (33) Hendrickson et al., 2000 (35) Hendrickson et al., 2000 (35) Houghton et al., 2002 (37) Houghton et al., 2002 (37) Houghton et al., 2002 (37) Lodes et al., 2001 (55) Lodes et al., 2001 (55) Lodes et al., 2001 (55) Senthil Kumar et al., 2002 (77) Silva et al., 2003 (81) Culture (SP) Italy and United States (Italy and United States) Healthy NR (NR d ) IgM 41/30 54 (37 69) 100 (88 100) Culture (SN) Italy and United States Healthy NR (NR) IgM 29/30 55 (36 74) 100 (88 100) (Italy and United States) Smear (SP) Mexico (Mexico) Nontuberculous ( ) IgG 50/48 56 (41 70) 96 (86 100) Smear (SP) India (India) Healthy NR (NR) IgG, IgM 44/105 36 (22 52) 99 (95 100) Nested within observational study Brazil (United States and areas of endemicity) Culture (SP) Gambia (Gambia) Healthy and ( and ) Culture (SP) South Africa and Uganda (China) Culture (SP) South Africa and Uganda (China) Brazil (United States and areas of endemicity) Philippines (United States and areas of endemicity) Sub-Saharan Africa (United States and areas of endemicity) Brazil (China) South Africa (China) Philippines (China) followed. The application of a comprehensive search strategy with various overlapping approaches enabled the retrieval of relevant studies published since 1990. Screening and data extraction were performed independently among three reviewers, and authors were contacted to clarify points and obtain missing data. None of the studies in the review used the result from the antibody test as a reference to confirm TB (incorporation bias). When possible, patients with were selected, in preference to healthy controls, to evaluate the performance of antigens with persons in whom TB was initially suspected and subsequently ruled out. The meta-analysis was limited by the relatively small number of studies investigating the same antigens or antigen combinations. The small number of comparable antigens made it difficult to relate study quality to antigen performance. However, several important deficiencies in study design and quality were noted. Only 20 (7%) studies recruited participants in a random or consecutive manner. Therefore, most studies lacked a sound Healthy ( ) IgG 250/57 49 (43 55) 91 (81 97) IgG 100/100 49 (39 59) 50 (40 60) Nontuberculous (NR) IgG 52/31 46 (32 61) 97 (83 100) Nontuberculous (NR) IgG 25/31 68 (47 85) 97 (83 100) Healthy (NR) IgG 105/57 57 (47 67) 91 (81 97) Healthy (NR) IgG 40/57 35 (21 52) 91 (81 97) Healthy (NR) IgG 66/57 41 (29 54) 91 (81 97) Nontuberculous Nontuberculous Nontuberculous ( ) IgG 248/31 48 (42 55) 97 (83 100) ( ) IgG 51/31 29 (18 44) 97 (83 100) ( ) IgG 31/31 36 (19 55) 97 (83 100) Culture (SP) India (India) Healthy NR (NR) IgG 25/25 60 (39 79) 100 (86 100) Cross-sectional Culture (SN) Canada (Canada) Healthy NR (NR) IgG 33/54 39 (23 58) 89 (77 96) a SP, smear positive; SN, smear negative. b Number of participants with TB/number of participants without TB. c 95% CIs are given in parentheses. d NR, not reported. probabilistic sampling framework. The majority of studies used a case-control design with healthy controls. This design has been found to overestimate test sensitivity and specificity (53, 111), although for the four protein antigens in this review for which a comparison was feasible, the specificities were found to be similar with healthy and d controls. Few (26%) studies reported the use of the blinded interpretation of test results and a reference standard. This was not unexpected, since the primary aims of in-house studies are the discovery of novel antigens, evaluation of their diagnostic potential, and/or comparison of different antigens. Nonetheless, the lack of blinding and the dearth of data from cross-sectional studies are major shortcomings of the currently available literature and may have resulted in an overestimation of antigen performance (53). Both errors in design and deficiencies in reporting have been noted as concerns in TB diagnostic studies (62, 89). An additional limitation was the lack of information about