Streitz M, Fuhrmann S, Thomas D, Cheek E, Nomura L, Maecker H, Martus P, Aghaeepour N, Brinkman RR, Volk H-D, Kern F. The phenotypic distribution and functional profile of tuberculin-specific CD4 T-cells characterizes different stages of TB infection. Cytometry Part B 2012; 82B: 360–368.
The phenotypic distribution and functional profile of tuberculin-specific CD4 T-cells characterizes different stages of TB infection†
Article first published online: 7 SEP 2012
Copyright © 2012 International Clinical Cytometry Society
Cytometry Part B: Clinical Cytometry
Volume 82B, Issue 6, pages 360–368, November 2012
How to Cite
Streitz, M., Fuhrmann, S., Thomas, D., Cheek, E., Nomura, L., Maecker, H., Martus, P., Aghaeepour, N., Brinkman, R. R., Volk, H.-D. and Kern, F. (2012), The phenotypic distribution and functional profile of tuberculin-specific CD4 T-cells characterizes different stages of TB infection. Cytometry, 82B: 360–368. doi: 10.1002/cyto.b.21041
- Issue published online: 22 OCT 2012
- Article first published online: 7 SEP 2012
- Manuscript Accepted: 31 JUL 2012
- Manuscript Revised: 3 JUL 2012
- Manuscript Received: 11 FEB 2012
- NIH. Grant Number: 1R01EB008400
- Charite – Universitätsmedizin Berlin and Brighton and Sussex Medical School, the Michael Smith Foundation for Health Research
- The Terry Fox Research Institute
- The Terry Fox Foundation, a UBC4YF scholarship
- clinically active tuberculosis;
- latent tuberculosis infection;
Recent publications have suggested that altered proportions of functional CD4 T-cell subsets correlate with active pulmonary TB. Also, CD27-expression on tuberculin-activated IFN-γ+ CD4 T-cells is known to differ significantly between patients with active pulmonary TB and healthy TB-unexposed BCG vaccinees. Here, we explore links between CD4 T-cell phenotype, multiple functional subsets, and control of TB.
We examined ex-vivo overnight tuberculin activated CD4 T-cells in regards to CD27-expression and the activation markers, CD154 upregulation, IFN-γ, TNF-α, IL-2, and degranulation in 44 individuals, including cases of clinically active pulmonary TB, and hospital staff with prolonged TB exposure, some of whom had latent TB.
Active pulmonary TB generally showed an excess of TNF-α+ subsets over IFN-γ+ subsets, paralleled by decreased CD27 expression on activated IFN-γ+ or CD154+ CD4 T-cells. The single subset distinguishing best between active pulmonary TB and high TB exposure was CD154+/TNF-α+/ IFN-γ-/IL-2-/degranulation- (AUROC 0.90). The ratio between the frequencies of TNF-α+/IFN-γ+ CD4 T-cells was an effective alternative parameter (AUROC 0.87).
Functional subsets and phenotype of tuberculin induced CD4 T-cells differ between stages of TB infection. Predominance of TNF-α+ CD4 T-cells in active infection suggests an increased effort of the immune system to contain disease. © 2012 International Clinical Cytometry Society
Primary infection with Mycobacterium tuberculosis (Mtb) usually results in latent tuberculosis infection (LTBI) which is estimated to affect one third of the world's population (1). The successful control of LTBI by the immune systems is the single most important factor limiting the spread of TB worldwide. In otherwise healthy people, LTBI converts to active tuberculosis (TB), which can be pulmonary or extra-pulmonary, at a rate of 10% in a lifetime, compared with 10% per annum in the context of HIV infection (1). Interferon-γ release assays (‘IGRA’s) are used to detect TB infection in the population and are considered the gold standard for detecting LTBI by many authors (2–7), however they cannot efficiently distinguish LTBI from active pulmonary TB (apTB) (2, 6, 8, 9).
CD4 T-cells play a major role in the control of Mtb (10), however our understanding of what constitutes a successful immune response to TB is incomplete. T-cell polyfunctionality, often used synonymously for the simultaneous production of IFN-γ, TNF-α, and IL-2 by the same T-cell, has been postulated to be critical in the defence against infectious agents; this observation is originally based on vaccination experiments in mice (11), but soon developed into a new paradigm quickly adopted by other fields including HIV and TB (12, 13). However, a number of recent publications seem to cast doubt on whether polyfunctional CD4 T-cells are truly related to the control of TB (14, 15). One recent publication found that the size of a CD4 T-cell subset producing TNF-α but not IFN-γ or IL-2 is highly increased in apTB and a good discriminator between active and latent TB (15).
The aim of this work was to investigate the relationship between the presence of certain TB specific CD4 T-cell subsets and the ability to control infection in a cohort of individuals with apTB and high TB exposure (hE), including latent infection. We used five common activation readouts, IFN-γ, TNF-α, IL-2 production, CD154 upregulation (16), and degranulation (CD107 mobilization) (17). CD154 (CD40 ligand) was of particular interest in this context as it is a very inclusive activation marker expressed on activated CD4 T-cells. Degranulation on the other hand has been described as a function displayed by terminally activated CD4 T-cells (18). This marker combination allowed us to dissect the CD4 T-cell response to TB into 31 functional subsets that were analyzed individually or in various combinations and compared between apTB and hE. We hypothesized that the percentage of CD27 negative tuberculin activated IFN-γ+ CD4 T-cells would discriminate between these groups, since we had previously observed successful discrimination of apTB from healthy TB unexposed BCG-vaccinees by this marker combination (19). The co-stimulatory receptor, CD27, is central to models of T-cell differentiation (18, 20–22). Changes in its expression reflect changes in T-cell function, circulation pathways and co-stimulation requirements (23). On CD4 T-cells CD27 is lost as cells proceed from a moderately (“central memory”) to a more terminally differentiated stage (‘effector memory’) (18, 20, 23–26). Lung CD4 T-cells are predominantly CD27-, and loss of CD27 is associated with lung homing (24, 25). In light of the findings by Harari et al. it was of particular interest to investigate the relationship between the TNF-α+/ IFN-γ -/IL2- subset and the expression of CD27 on tuberculin induced CD4 T-cells. As a result, our study investigated the changes of CD27-surface expression on activated CD4 T-cells in relation to the concomitant changes in the functional subset composition of the overall tuberculin specific population and identified several T-cell parameters that were significantly different in apTB and hE. The functional composition of tuberculin specific T-cells changes dramatically as the expression of CD27 on CD154+ and IFN-γ+ tuberculin-specific CD4 T-cells is lost, shedding light on what are likely to be mechanisms of the successful control of Mtb by the immune system.
MATERIALS AND METHODS
Patients and Patient Materials
Twenty-seven TB patients were recruited from Lungenklinik Lostau, Lostau/Germany (LL). Twenty patients with positive Mtb sputum and/or bronchio-alveolar lavage fluid (BAL) culture (gold standard) were designated “active pulmonary tuberculosis” (apTB) according to the criteria of the American Thoracic Society (27), including 9 females and 11 males (48 ± 24 years, mean ± STD). They received up to 6 weeks of therapy (19.7 ± 9.4 days, mean ± STD), none of them were previously treated for TB. Seven additional patients were treated as a separate group, because they had previously been treated for TB (n = 5), had currently been on treatment for longer (43–274 days), had extensive extrapulmonary manifestations (n = 5) including multiorgan TB (n = 2), or had no positive Mtb culture result but positive histology/microscopy (n = 2). However, all had been clinically diagnosed to have active TB. None of the patients were HIV-positive, and the majority were Bacille-Calmette Guerin (BCG) vaccinated. TB specific therapy included isoniazid, ethambutol, and rifampicin; in some cases also pyrazinamide. Seventeen BCG vaccinated, hospital staff with long-term exposure to apTB were recruited at LL and Charité Campus Mitte, including 12 females and 5 males (age ± STD = 42 ± 11 years). They were considered subject to high exposure (hE). Among these, seven donors were confirmed to have latent TB infection by the “Quantiferon Gold in-tube” (QFT) test (14 of 17 were tested, 3 were unavailable for the test), none of them were known to be HIV-positive. Additionally, seven TB unexposed individuals were recruited from University staff and students. There were no significant ethnic differences between the groups. Written informed consent was obtained from all participants. The study was approved by the Charité Ethics Committee.
Reagents for Stimulation
Tuberculin (Statens Serum Institute, Copenhagen, Denmark) and staphylococcus enterotoxin B (SEB, Sigma, CA) were dissolved in dimethyl-sulfoxide (DMSO) (Pierce, USA) prior to lyophilisation. Overlapping peptide pools covering the “culture-filtrate-protein-10” (CFP-10) and the ”early secretory antigen 6” (ESAT-6) consisted in 15-amino-acid peptides with 11 overlaps (minimum 75% purity by HPLC) were synthesized to order by JPT Peptide Technologies Berlin, Germany.
T-Cell Activation and Flow-Cytometry
Peripheral blood mononuclear cells (PBMC) were prepared from anticoagulated blood (sodium-citrate) by gradient centrifugation within four hours of blood collection, 200 μl aliquots (5 x 106 cells/ml) in complete culture media (RPMI, Gibco/Invitrogen, Paisley, UK) were transferred to 8-well-strips (removed from 96-well strip-well plates) containing the lyophilized stimulants (DMSO alone for negative control, SEB for positive control, ESAT-6 peptide pool, CFP-10 peptide pool, tuberculin), anti-CD107a (17), and Monensin (Sigma). All wells except tuberculin also contained Brefeldin A (BFA, Sigma). All wells were reconstituted by adding 25 μl of PBS (Sigma) with careful mixing prior to PBMC addition. Totally, 25 μl of complete culture media was added to all wells after 2 h, BFA was added to the tuberculin wells at that time (allowing tuberculin processing by antigen presenting cells during the first 2 h). Unstimulated controls were run with respect to all stimulation conditions. The final volume was 250 μl in all wells, final concentrations of Monensin and BFA were 5 mg/ml each, the final concentrations of tuberculin and SEB were 10 mg/ml and 1 mg/ml, respectively. Following incubation for 16 h (37°C, 5% CO2) 20 μl of 20 mM EDTA (Sigma) in PBS/0.1% sodium azide (Sigma) was added to each well; extra- and intracellular staining was performed according to a published protocol for lyoplates (28).
A minimum of 200,000 lymphocytes was acquired per tube (LSR II flow-cytometer, BD Biosciences, CA). Frequencies of 0.01% positive CD4 T-cells or more (with respect to each activation marker) were considered positive responses (19).
Monoclonal Antibodies and Beads
Antibodies and fluorochrome labels are listed in Table 1. Compensation Particles binding Anti-Mouse Ig, κ/Negative Control (FBS) (BD Biosciences) were used for preparing single color staining tubes required for compensation.
FlowJo™ software, version 8 (Treestar, OR) was used. As previously described, only response subsets including at least 50 activated events were evaluated in regards of CD27-expression (i.e., CD27 pos. and CD27 neg.) (19). The threshold between CD27+ and CD27− events was set according to CD27-expression of CD3 negative lymphocytes (mostly natural killer cells) (Figs. 1A and 1B). These are almost exclusively CD27− and easy to gate. Optimization experiments with 8-peak calibration beads (Spherotec) indicated that the limit between CD27 positive and negative cells is objectively reproducible. Boolean gating created 31 nonoverlapping subsets covering all possible activation marker combinations. Activated cells were considered those displaying at least one of the activation markers. A background (unstimulated sample) was subtracted per nonoverlapping subset. Only data sets containing 250 or more events expressing at least one of the activation markers were analyzed in this way.
Interferon-γ Release Assay
The QFT test (Cellestis, Darmstadt, Germany) was performed according to the manufacturers' instructions.
Statistical analysis was carried out on up to 62 (nonoverlapping) cellular subsets. The Mann-Whitney test was used to test differences in subset proportions between groups, the Wilcoxon test was used to test differences between different subsets. ROC analysis was used to explore the ability of parameters to discriminate between groups and determine threshold levels for optimum sensitivity and specificity. The level of significance was 0.05 (two-sided) for statistical tests. Where comparisons were made for 31 or 62 subsets, a Bonferroni correction with a factor of 31 or 62, respectively, was applied. Thus, only P-values <= 0.0016 or < 0.0008 were considered significant. PASW 18.0 (IBM, USA) software was used for all statistical calculations. In some cases outliers were removed to analyze the correlation between parameters. Values outside the upper quartile + 1.5 interquartile ranges were considered outliers (29).
To avoid bias, clinical details were not consulted until after completion of the flow-cytometric sample analysis. With the clinical information in hand we initially focused on apTB (n = 19) and hE (n = 17) as these were the most clearly defined and homogeneous groups. In agreement with published results (19, 30) tuberculin-induced CD4 T-cell responses were generally higher than Mtb subunit antigen-induced responses (CFP-10, ESAT-6), in particular in hE. The figure also includes the 7 individuals with no known exposure to TB (Fig. 2A). While it might be possible to identify LTBI based on T-cell responses to CFP-10 and ESAT-6 measured by flow-cytometry, for the purpose of this study we considered LTBI to be present only when this was supported by a positive Quantiferon test.
A Number of Functional Tuberculin-Specific CD4 T-Cell Subsets Differ Significantly Between hE and apTB
Recently, Harari et al. (15) reported that the proportion of CD4 T-cells producing TNF-α but not IFN-γ or IL-2 (“TNF-α single producers”) among all ESAT6 and/or CFP10 activated CD4 T-cells discriminates LTBI from active TB. Therefore, we explored if we could confirm a similar group difference based on tuberculin stimulation. However, in our experiments the responses to ESAT-6 and CFP-10 appeared too infrequent and too small to be subjected to multiple subset analysis, so that we relied on tuberculin activated CD4 T-cells. Of note, the number of nonoverlapping activated Boolean subsets arising from three activation markers is seven, but thirty-one when five activation markers are used (Fig. 2B). The proportions of these subsets appeared quite similar in hE and apTB; only three subsets were significantly different (subsets 2, 6, and 8; significance level P = 0.0016, corrected for multiple testing), all of which were TNF-α positive but IFN-γ negative (Fig. 2B). ROC confirmed the usefulness of these subsets in discriminating hE from apTB (Table 2, single subsets) with subsets 2 and 6 showing the biggest AUROC. Meanwhile, subsets simultaneously producing IFN-γ, IL-2, and TNF-α were very small except one; none of them were significantly different between the groups (Fig. 2B).
|Subsets (proportions of all activated CD4 T-cells)|
|Subset 2(TNF-α+/IFN-γ−/ IL-2+/CD154+/CD107+)||0.84||0.001|
|Subset 6(TNF-α+/IFN-γ−/ IL-2−/CD154+/CD107−)||0.90||0.000|
|Subset 8(TNF-α+/IFN-γ−/ IL-2−/CD154−/CD107−)||0.82||0.002|
|Functional components (based on subset proportions)|
|Ratio TNF-α/IFN-γ (components)||0.87||0.000|
|TNF-α − IFN-γ (components)||0.88||0.000|
|Directly measured frequencies|
|Ratio TNF-α/IFN-γ (frequencies)||0.87||0.000|
|TNF-α − IFN-γ (frequencies)||0.83||0.001|
“TNF-α single-producers” based on 5 activation markers in our study differ from those based on 3 activation markers as in Harari's work (15). For comparing our results with theirs in regards of apTB and hE, we recalculated the proportions of single TNF-α producers in terms of their study as the sum of all subsets expressing TNF-α but not IFN-γ or IL-2 (ignoring the other functions) and expressed those as proportion of all cells producing at least one of these three markers (leaving out subsets 29, 30, and 31 shown in Fig. 2B). This recalculated proportion discriminated the groups well (AUROC 0.85, P = 0.000, not shown) confirming the potentially important functional role of this subset. TNF-α single producers in terms of our study (subset 8, Fig. 2B and Table 2) discriminated the groups well, for IFN-γ single producers (subset 24 in Fig. 2B) the difference between the groups was not significant (P = 0.006) given the significance level of P = 0.0016, see above. More TNF-α combined with fewer IFN-γ producers in apTB (and vice versa in hE) represented a general functional difference between the groups. We, therefore, also explored the use of summed proportions of all TNF-α-producing subsets (“TNF-α component”) or IFN-γ-producing subsets (′IFN-γ component) as “compound” measures. These refer to the sums of the subsets 1–16 and 9–24 shown in Figure 2B, respectively. Both the difference and the ratio of these two compound measures discriminated the groups well (Table 2, components) and so confirmed an “excess” of TNF-α producers in apTB, and of IFN-γ producers in hE. This was also true when the measured net frequencies of IFN-γ and TNF-α producers (in % of all CD4 T-cells) were used instead (Figs. 3A–3C, Table 2). These net frequencies were determined by subtracting the frequencies of TNF-α or IFN-γ positive events measured in unstimulated samples (i.e., quadrants Q1-1 plus Q2-2 and Q2-2 plus Q4-2 in the lower panels in Fig. 1A, respectively) from the TNF-α or IFN-γ positive events (corresponding quadrants in the upper panels in Fig. 1A). In hE the ratio of TNF-α/IFN-γ (frequencies of CD4 T-cells producing each cytokine) was typically below 1.2 (Fig. 3C).
CD27-Expression on Activated Subsets Is Significantly Different Between hE and apTB
Proportions of CD27- events with respect to each activation marker were determined if at least 50 activated events were present, as calculated proportions of smaller populations are very unreliable (19). Unfortunately, after ESAT-6 and CFP-10 stimulation the activated populations were generally too small for this analysis. The performance of CD27-expression on tuberculin activated events in discriminating apTB and hE was very good when using CD154+ (AUROC 0.87, P = 0.000) and IFN-γ+ (AUROC 0.80, P = 0.004), less good with IL-2+ (0.70, P = 0.084), and not good with TNF-α+ (AUROC 0.40, P = 0.345), or degranulation+ (0.42, P = 0.522). Interestingly, the percentages of CD27−/CD27+ events on TNF-α+/ IFN-γ− events in each group were very similar (not shown).
Figures 4A and 4B show the distribution of the measurements for CD27− events in more detail with respect to CD154+ and IFN-γ+ tuberculin induced CD4 T-cells (percentages correspond to the lower right quadrants for CD154+ and IFN-γ+ in Fig. 1 A). As expected, the proportions of CD27− events among the IFN-γ+ and CD154+ tuberculin induced CD4 T-cells were highly, but not perfectly correlated (Fig. 4C), meanwhile the correlation between the frequencies of IFN-γ+ and CD154+ events was a little weaker (Spearman's Rs = 0.608, P = 0.000, not shown). The greater the response to one or both of the Mtb. subunit antigens, ESAT-6 and CFP-10, the greater was the number of CD27− neg. events among the tuberculin activated CD4 T-cells, both when CD154 (Fig. 4D) and IFN-γ were analyzed (not shown).
Considering CD27-Expression for Each Boolean Subset Increased the Number of Subsets to 62 But Did Not Improve Discrimination Between Groups
Next we analyzed differences between the groups in regards to the combination of CD27-expression (positive or negative) with each of the 31 functional subsets (i.e., proportions of all activated CD4 T-cells).This resulted in 62 subsets (i.e., end-points), but the discrimination did not improve. Only subsets CD27+/12, subset CD27−/2, CD27−/6, and CD27−/12 were significantly different between apTB and hE at the P = 0.0008 level but the AUROC was not improved.
At this stage we also included the additional seven heterogeneous TB cases, to see if they would be correctly assigned to the groups (Figs. 5A and 5B). The ratio of TNF-α+/IFN-γ+ events combined with the proportion of CD27−CD154+ CD4 T-cells misclassified only two hE individuals as active TB (false positive) when the previously established thresholds of 1.2 (ratio) and 50% (CD27− in %) were used, respectively (Fig. 5A). This would translate to 100% sensitivity and >94% specificity. The TNF-α+/ IFN-γ+ CD4 T-cell ratio combined with the proportion of CD27−IFN-γ+ CD4 T-cells misclassified four hE individuals as active TB (i.e., false positive) but no TB case was misclassified when using the previously established thresholds of 1.2 (ratio) and 49%, respectively. However, if the latter was set to 65%, only two hE cases were false positive, and one TB case was false negative, which would translate into a sensitivity of >95% and specificity >88% (Fig. 5B).
This study aimed to better understand the immunological differences between cases of active tuberculosis and individuals resisting or controlling infection. Analysis of the functional subset composition of all tuberculin specific CD4 T-cells revealed that the proportions of several subsets (but also simply the ratio of TNF-α+/ IFN-γ+ producing subsets) were significantly different between these groups. We also established that CD27-expression on IFN-g+ or CD154+ tuberculin-activated CD4 T-cells is able to discriminate apTB from highly exposed/latently infected individuals, which extends our previous findings when comparing healthy BCG vaccinees to cases of apTB (19). However, interestingly, there was no significant direct correlation between the proportion of any functional subset and the CD27 expression on IFN-γ+ and CD154+ CD4 T-cells. Meanwhile, the general excess of TNF-α+/IFN-γ- CD4 T-cells in active TB paralleled the increase in CD27-CD4 T-cells among CD154+ or IFN-γ+ tuberculin-activated CD4 T-cells. This explains why the ratio of TNF-α+/IFN-γ+ positive CD4 T-cells, which can be directly and easily measured by flow-cytometry (Figs. 5A and 5B), correlated loosely (R about 0.30) but significantly with the percentage CD27− events among both CD154+ CD4 T-cells and IFN-γ+ CD4 T-cells. As a consequence of this, the combination of the different tests is more powerful than each one alone. Of note, the presence of “polyfunctional” T-cells producing TNF-α, IFN-γ, and IL-2 at the same time was not specifically associated with hE or apTB. This is in agreement with Harari et al. (15). We propose that the excess of TNF-α+IFN-γ− CD4 T-cells represents an increased effort of the immune system to contain Mtb infection by activating and recruiting immune cells to the infectious site. Recent work in a macaque TB model has demonstrated that TNF-α is critical for macrophage activation, chemokine, and adhesion molecule upregulation rather than just stabilizing granuloma architecture as often claimed. This appeared unchanged in the presence of TNF-α-antagonists but infection was disseminated (31). As a result, TNF-α is likely to help attracting immune cells to the infection site. It is well known that IFN-γ produced by CD4 T-cells activates macrophages to kill intracellular mycobacteria, and individuals with defects in signaling pathways related to IFN-γ production experience severe manifestations of mycobacterial infections (32, 33). However, TNF-α-facilitated accumulation of immune cells at the site of infection is probably necessary for IFN-γ to achieve its local effects. Therefore, in a situation where Mtb is not or no longer contained, TNF-α production by CD4 T-cells may take priority until containment is re-established by additional immune cells.
In regards to CD27 expression, it appears from our data that CD27-negative tuberculin activated CD4 T-cells have a tendency to express relatively more TNF-α and less IFN-γ. In terms of T-cell differentiation, loss of CD27 occurring in the lung (24, 25) would suggest that its ligand, CD70, is not only present on APCs in intestinal mucosa (34) but also expressed in lung mucosa or bronchus associated lymphatic tissue (BALT). Lyadova et al. reported that IFN-γ producing CD4 T-cells in the lung of TB infected mice are CD27− (25) but they did not analyze TNF-α production in this study. Kapina et al. showed that the number of CD27- CD4 T-cells in the lungs of infected mice correlates with their level of protection (24) and in light of our data it may be suspected that this is related to TNF-α. Of interest, Mack et al. reported that human lung CD27- CD4 T-cells are indeed effector cells (35). Jiang et al. recently confirmed that the presence of this subset correlates with persistence of active TB infection (HIV uninfected) (36) and more recently Schuetz et al. showed that this is also true in the context of HIV infection (37). Here, we show that CD27-expression on IFN-γ or CD154+ cells reliably identifies cases of active TB even against a background of high TB exposure and latent infection. CD154 as a marker is of particular interest, not only because it performed better than IFN-γ, but also, because it may be stained on the surface without the need for intracellular staining (16). If such methodology will be equivalent for this test, however, needs to be addressed.
In summary, there is a gradual change in the immune response from high TB exposure via LTBI to active TB by three alterations. These are (i) the change of CD27-expression of activated T-cells, (ii) the changes in some specific functional subsets, and (iii) by changes in the cumulative proportions of all IFN-γ and/or TNF-α producing subsets. The balance between IFN-γ and TNF-α producing CD4 T-cell subsets is tilted towards TNF-α in active TB (compared with LTBI or high exposure) which is likely to indicate an increased effort to contain infection. In the future it will be important to investigate which of these diagnostic signs, if any, are retained in HIV/TB co-infection. There is no doubt in light of recent publications and the findings presented here that flow-cytometry should be seen as an efficient way forward in the diagnosis of active TB. The best final format of such a standard clinical diagnostic test will have to be established in further studies.
The authors would like to thank Dr. Ali Quassem (Lungenklinik Lostau) for providing patient materials and clinical expertise.
- 5Guidelines for using the QuantiFERON-TB Gold test for detecting Mycobacterium tuberculosis infection, United States. MMWR Recomm Rep 2005; 54: 49–55., , , , , .
- 27Diagnostic Standards and Classification of Tuberculosis in Adults and Children. This official statement of the American Thoracic Society and the Centers for Disease Control and Prevention was adopted by the ATS Board of Directors, July 1999. This statement was endorsed by the Council of the Infectious Disease Society of America, September 1999. Am J Respir Crit Care Med 2000; 161: 1376–95.
- 29Introduction to the Practice of Statistics. New York: W.H. Freeman; 1999., .