HIV-1-related pleural tuberculosis: elevated production of IFN-[gamma], but failure of immunity to Mycobacterium tuberculosis

Background: Pleural tuberculosis can resolve spontaneously, suggesting that the inflammatory process may represent a protective immune response. However, pleural tuberculosis is strongly associated with HIV infection. It has been suggested that cell-mediated immune responses may be reduced, and direct bacterial invasion may have a role in pathogenesis, in HIV-positive cases. To test this hypothesis, we compared production of the pro-inflammatory cytokines, interferon (IFN)-γ and tumour necrosis factor(TNF)-α, production of the immunosuppressive cytokine, interleukin (IL)-10, and mycobacterial culture positivity, in HIV-negative and HIV-positive patients with pleural tuberculosis.

Cited Here...: Cytokine levels were measured in serum and pleural fluid, and in supernatants of blood and pleural fluid stimulated in vitro using mycobacterial antigens. Intracellular IFN-γ and TNF-α production was measured after stimulation with phorbol myristate acetate and ionomycin in vitro.

Cited Here...: IFN-γ was strikingly elevated in serum and pleural fluid in HIV-positive, compared to HIV-negative subjects (P ≤ 0.02). TNF-α was elevated, but this was not statistically significant. IL-10 levels were higher in serum (P < 0.001), but similar in pleural fluid. IFN-γ responses to soluble mycobacterial antigen in vitro were reduced in peripheral blood (P = 0.006), but not pleural fluid, of HIV-positive subjects. Intracellular cytokine staining suggested that CD8+ T cells were a major source of IFN-γ in HIV-positive subjects. The proportion of subjects with a positive culture for Mycobacterium tuberculosis from pleural fluid was higher in the HIV-positive group.

Conclusions: HIV-positive patients with pleural tuberculosis show elevated production of IFN-γ, for which CD8+ T cells may be a major source. Mycobacterium tuberculosis can proliferate despite high levels of pro-inflammatory cytokines.

Introduction

Pleural tuberculosis in HIV-negative subjects has been taken to represent a protective immune response to Mycobacterium tuberculosis because a high proportion of individuals can recover without antibiotic therapy ^[1]. The development of a tuberculous pleural effusion in HIV-negative subjects has been shown to be associated with an intense cell-mediated immune response, with infiltration of CD4+ T cells ^[2] and production of high levels of pro-inflammatory cytokines, such as gamma interferon (IFN-γ) and tumour necrosis factor alpha (TNF-α) ^[3]. Typically, mycobacterial culture is positive in less than 30% of HIV-negative patients with pleural tuberculosis ^[4]. These observations have been taken to indicate a key role for these responses in protective immunity against tuberculosis in humans, in keeping with evidence that IFN-γ and TNF-α are necessary for protective immunity to tuberculosis in animals ^[5-7], and that abnormalities of the IFN-γ receptor are associated with susceptibility to mycobacterial disease in humans ^[8,9].

By contrast, HIV infection is characterized by depletion of CD4+ T cells, reduced proliferative and IFN-γ responses to mycobacterial antigens in peripheral blood ^[10-12], marked susceptibility to tuberculosis and a high mycobacterial burden in affected tissues ^[13,14]. It is therefore surprising to find that HIV infection is even more strongly associated with pleural tuberculosis than with pulmonary tuberculosis ^[15]. One suggested explanation for this paradoxical finding is that the pathogenesis of HIV-associated pleural tuberculosis might involve direct, bacterial invasion of the pleural space, rather than an active cell-mediated immune response ^[14]. However, although antigen-specific, lymphocyte responses such as IFN-γ production may be reduced as HIV disease progresses, production of other cytokines, such as TNF-α and interleukin (IL)-10, have been shown to increase, and production of TNF-α is especially high in HIV-positive individuals with active tuberculosis ^[16-18]. These responses may be mediated, to some extent, by monocytes and macrophages, monocytes being present in relatively high proportions in peripheral blood as lymphocyte depletion progresses during HIV infection ^[19]. Thus, HIV-associated pleural tuberculosis might be associated with reduction in antigen specific, CD4+ T-cell-mediated responses, but sustained non-specific production of pro-inflammatory cytokines such TNF-α, or of immunosuppressive cytokines such as IL-10. We therefore set out to investigate the effect of HIV-infection on the presence of M. tuberculosis in the pleural space, and on the production of the key cytokines IFN-γ, TNF-α and IL-10. We here describe the unexpected association of HIV infection with marked elevation of IFN-γ production in pleural tuberculosis, while levels of TNF-α and IL-10 were similar in HIV-negative and HIV-positive cases.

Methods

Patients

The study was conducted at the National Tuberculosis Treatment Centre at Mulago Hospital, Kampala, Uganda between August 1998 and December 1999. All patients with clinical and radiological evidence of pleural tuberculosis were invited to participate in the study if they were adults, aged 18 years or above, with no history of previous treatment for tuberculosis, and no treatment with glucocorticoids within the last month. Women who were pregnant were excluded. After giving written informed consent, participants underwent clinical examination, thoracocentesis and pleural biopsy. Blood was obtained for clinical and immunological analysis on the same day as the thoracocentesis. All specimens were obtained before the participants received any treatment for tuberculosis. The study was approved by the Uganda National Council for Science and Technology, the Science and Ethics Committee of the Uganda Virus Research Institute and the Ethics Committee of the London School of Hygiene & Tropical Medicine.

Diagnostic investigations

HIV-1 infection was diagnosed using a rapid test (Determine HIV-1/2, Abbott Laboratories, Tokyo, Japan) and confirmed by enzyme-linked immunosorbent assay (ELISA; Vironostika HIV-1 Microelisa System, Organon Teknika Corporation, Durham, North Carolina, USA). Viral load was measured in plasma and pleural fluid for the first 20 HIV-positive subjects enrolled (Amplicor HIV-1 monitor test, Roche Diagnostic Systems, Branchburg, New Jersey, USA). Pleural fluid specimens were examined for tuberculosis by culture in BACTEC 13A radiometric culture vials containing Middlebrook 7H-13 broth (Becton Bickinson, Sparks, Maryland, USA) and, after centrifugation, on Lowenstein Jensen slopes. Pleural biopsy specimens were cultured in BACTEC and examined by histology. Sputum samples were also examined by smear and by culture on Lowenstein Jensen medium. The diagnosis of tuberculosis was categorized as follows:

1.Confirmed tuberculosis: any culture was positive for Mycobacterium tuberculosis, or histology was characteristic of tuberculosis.

2.Probable tuberculosis: the laboratory investigations were negative but there was a good clinical response to tuberculosis treatment (at least two of three criteria: weight gain of 2 kg or more, improvement in symptoms, and radiological improvement).

3.Uncertain: the presenting clinical features suggested tuberculosis but no confirmation could be obtained. These individuals were excluded from the analysis.

Immunological investigations

Direct measurement of cytokines in serum and pleural fluid was carried out on samples from every participant. More detailed immunological investigations were conducted sequentially, on samples given by separate subgroups of participants, since the available material was insufficient to allow all assays to be conducted on the same individuals.

Subgroup 1:cytokine production following stimulation by mycobacterial antigens in vitro (21 consecutive HIV-positive subjects and 15 HIV-negative subjects enrolled concurrently).

Subgroup 2:examination of cell types by flow cytometry (29 HIV-positive subjects and nine HIV-negative subjects enrolled concurrently).

Subgroup 3:intracellular cytokine staining was conducted using samples from eight participants.

Cytokine production in vivo

Clotted samples of blood and pleural fluid were centrifuged at 1750 g. The supernatant was removed and stored at -80°C until analysis. Cytokine levels were measured by ELISA using optEIA reagents (PharMingen, San Diego, California, USA). The sensitivity of the assays was 8 pg/ml. Analysis of samples giving values greater than the highest standard was repeated using dilutions as necessary to obtain a value within the standard range.

Cytokine production after stimulation in vitro

Cytokine responses to stimulation with whole antigens in vitro were examined using a whole blood method for peripheral blood and a similar method adapted for pleural fluid. The whole blood assay was conducted as previously described ^[11]. Briefly, unseparated, heparinized blood was diluted to a final concentration of 1 in 5 with serum-free medium [RPMI, GibcoBRL (Life Technologies, Paisley, Scotland)] supplemented with penicillin and streptomycin plus 2 mm glutamine), plated in 96-well plates, and stimulated with antigen or mitogen at a final concentration of 10 mg/ml, or left unstimulated. To create a similar 'whole pleural fluid' assay, unseparated, heparinized pleural fluid was used. Cells were counted, spun down and resuspended in pleural fluid with serum-free medium, to produce a final concentration of pleural fluid to medium of 1 in 5, with 10⁶ cells/ml. The cells were plated and stimulated as above. Supernatants were harvested on day 1 (approximately 18 h incubation) for TNF-α and IL-10, and on day 6 for IFN-γ, and were frozen at -80°C until analysed. Antigens were purified protein derivative of M. tuberculosis (batch RT-49) (PPD) (Statens Seruminstitut, Copenhagen, Denmark) and crude culture filtrate proteins of M. tuberculosis H37Rv (CFP) and CFP with lipoarabinomannan removed (CFP-LAM), kindly provided by Dr J. T. Belisle, Colorado State University, Fort Collins, Colorado, USA through the National Institutes of Health contract NO1-A1-25147. Cytokine concentrations were measured by optEIA ELISA, as above. Low level production of cytokine in unstimulated wells was subtracted from the concentration produced in response to stimulation for analysis.

Examination of cell types by flow cytometry

Monoclonal antibodies for cell surface markers and lysing solution were obtained commercially (PharMingen). Aliquots of whole blood were stained according to the manufacturers instructions using fluorescein isothiocyanate (FITC) anti-CD3 for T lymphocytes; with 'Cy-Chrome' (CY) anti-CD56 (for CD3 negative, CD56 positive natural killer (NK) cells) and with r-phycoerythrin (PE) anti-CD14 for monocytes; FITC anti-CD3, PE anti-CD4 and CY anti-CD8, or FITC anti-CD3 and PE anti-γδ T-cell receptor for lymphocyte sub-types; and appropriate isotype controls. Briefly, red cells were lysed and the remaining cells were washed, stained, fixed using 2% paraformaldehyde and kept at 4°C until analysis by flow cytometry. Pleural fluid cells were counted, and aliquots containing approximately 500 000 cells were processed in the same manner as whole blood, without separation. Flow cytometry was conducted using a Coulter Epics Profile II flow cytometer and software (Coulter, Hialeh, Florida, USA). Approximately 10 000 events were accumulated for each sample.

Intracellular cytokine staining

Cells were separated from blood and pleural fluid using Histopaque (Sigma, Irvine, UK). They were resuspended in RPMI supplemented with 10% human serum, stimulated using phorbol myristate acetate (PMA) and ionomycin in the presence of monensin (`Golgistop', PharMingen) for 4 h at 37°C in 5% CO₂. Cells were harvested, stained for surface CD4 and CD8 as above, and fixed and permeablized overnight using 'Cytofix/ Cytoperm' (PharMingen). They were then stained for intracellular IFN-γ or TNF-α using FITC monoclonal antibodies or appropriate isotype controls (PharMingen). Flow cytometry was conducted as above. Lymphocytes were gated in forward and side scatter, and approximately 10 000 to 30 000 gated events were further analysed.

Statistical analysis.

Simple comparisons were made using χ² tests for two by two tables and the t test for means. Parameters showing markedly skewed distributions (cytokine concentrations, lymphocyte counts) were compared using the Wilcoxon rank sum test for unmatched samples, and the Wilcoxon signed rank test for matched pairs. Spearman's rank correlation coefficients were estimated.

Results

Clinical characteristics

In all, 101 sequential, untreated patients with probable pleural tuberculosis were enrolled in this study. Six were excluded because the final diagnosis was other than tuberculosis (four subjects), or remained uncertain (two subjects), leaving a total of 95 subjects for analysis. Of these 29 (31%) were HIV-negative and 66 (69%) were HIV-1-positive. Among the HIV-negative participants the diagnosis of tuberculosis was confirmed in 79% and probable in 21%. Among HIV-positive participants the diagnosis of tuberculosis was confirmed in 97% and probable in 3%. The clinical characteristics of HIV-negative and HIV-positive subjects are described in Table 1. Comparison of individuals included in each of the immunological subgroup studies, examined separately, showed similar effects to those presented for all subjects. HIV-positive participants were sicker, with lower Karnofsky score and body mass index, more severe anaemia, and reduced CD4+ cell counts. Total pleural fluid cell counts were available for 87 of the 95 subjects. Mononuclear cells were predominant in all but two cases. The total white cell count in pleural fluid was lower in HIV-positive subjects (955 compared with 1218 × 10⁶ cells/l, P = 0.05), as was the total mononuclear cell count in pleural fluid (850 compared with 1121 × 10⁶ cells/l, P = 0.05). There were no statistically significant differences in the X-ray findings between HIV-negative and HIV-positive subjects, although a slightly higher proportion of HIV-positive subjects were noted to have upper lobe infiltrates, cavities or adenopathy. The radiological extent of the effusions was similar in the two groups.

Table 1

A significantly higher proportion of HIV-positive participants than HIV-negative participants had a positive culture for M. tuberculosis from pleural fluid. This finding was consistent when the analysis included all 95 subjects (Table 1) and when restricted to subjects in whom tuberculosis was confirmed by culture from any site or by histology (P = 0.001). Among HIV-positive participants, the HIV viral load was markedly higher in pleural fluid (median 1 313 000 copies/ml) than in plasma (207 255 copies/ml) (P < 0.001).

Cytokine production in vivo

Cytokine production in vivo was examined by direct measurement in serum and pleural fluid for all subjects (Table 2). Comparing HIV-negative and HIV-positive subjects, the most striking observation was the markedly higher level of IFN-γ in both serum and pleural fluid of HIV-positive subjects. Among HIV-positive patients the level of IFN-γ in serum showed a negative correlation with peripheral blood CD4 cell count (correlation coefficient -0.24, P = 0.06) and a weak positive correlation with plasma viral load (correlation coefficient 0.33, P = 0.17). The level of IFN-γ in pleural fluid showed no statistically significant correlation with cell counts, but a strong positive correlation with pleural fluid viral load (correlation coefficient 0.54, P = 0.02). The level of IFN-γ was also higher among HIV-positive patients with a positive culture for M. tuberculosis from pleural fluid (median 676 pg/ml) than those with negative mycobacterial culture (median 488 pg/ml), but this difference was not statistically significant. A higher proportion of HIV-positive than HIV-negative patients had detectable TNF-α in both serum and pleural fluid, although this difference was not statistically significant, and levels of TNF-α were below the limit of sensitivity of the assay for more than 50% of subjects in this study. Levels of IL-10 were higher in the serum of HIV-positive subjects, but similar in pleural fluid from HIV-negative and HIV-positive subjects. Comparing serum and pleural fluid, all three cytokines showed higher levels in pleural fluid for both HIV-negative and HIV-positive subjects (Wilcoxon sign rank test, P < 0.02 for all comparisons, except TNF-α in HIV-negative subjects, P = 0.24).

Table 2

Cytokine production in response to stimulation in vitro

Cytokine production following stimulation in vitro was studied for 15 HIV-negative and 21 HIV-positive participants. Three patients who did not have tuberculosis (two HIV-negative and one HIV-positive) were excluded from the analysis. The responses to stimulation with PPD are shown in Table 3. Comparing HIV-negative and HIV-positive subjects, IFN-γ production in response to stimulation was reduced in peripheral blood of HIV-positive subjects. IFN-γ production in pleural fluid was also slightly lower in HIV-positive subjects, although this difference was not statistically significant. There were no statistically significant differences between HIV-negative and HIV-positive subjects in the production of TNF-α or IL-10 in either blood or pleural fluid, although TNF-α responses were somewhat higher for HIV-positive subjects. Similar results were obtained following stimulation with CFP and CFP-LAM (data not shown).

Table 3

Cells types

Next, we examined the cell types in blood and pleural fluid by flow cytometry for nine HIV-negative and 29 HIV-positive subjects, of whom three were excluded from the analysis: one HIV-negative patient, who did not have tuberculosis, and two HIV-positive subjects (one for whom cells were insufficient, and one for whom the assay failed). HIV-negative and HIV-positive subjects showed no difference in the ratio of monocytes or NK cells to T lymphocytes in either blood or pleural fluid. In both HIV-negative and HIV-positive subjects, T lymphocytes were the predominant cell type in pleural fluid, and were enriched relative to monocytes and NK cells in pleural fluid compared with blood (data not shown).

The principal difference in cell types between HIV-negative and HIV-positive subjects was a relative reduction in CD4+ lymphocytes, and a relative increase in CD8+ lymphocytes in the HIV-positive subjects, in both peripheral blood and pleural fluid. There was no difference in the proportion of γδ T cells (Table 4). Comparing blood and pleural fluid, CD4+ cells were enriched in pleural fluid in both HIV-negative and HIV-positive subjects (Wilcoxon signed-rank test, P < 0.02), and γδ T cells were reduced (P = 0.18 for HIV-negative subjects, P = 0.001 for HIV-positive subjects).

Table 4

Intracellular cytokine production

Finally, we examined intracellular IFN-γ and TNF-α production in CD4+ and CD8+ cells in blood and pleural fluid following stimulation with PMA and ionomycin (Table 5). These studies revealed that, in both HIV-negative and HIV-positive subjects, CD8+ T cells were more likely to produce IFN-γ than CD4+ T cells. This is illustrated by the scatterplot for IFN-γ production from pleural fluid cells for patient number 8, shown in Figure 1. Conversely, CD4+ T cells were more likely to produce TNF-α than CD8+ T cells. Comparing HIV-negative and HIV-positive subjects, the proportion of pleural fluid CD8+ T cells producing IFN-γ was significantly higher for HIV-positive cases (P = 0.02). Comparing blood and pleural fluid, increased responsiveness of both CD4+ and CD8+ lymphocytes, for both cytokines, was seen in the pleural fluid of both HIV-negative and HIV-positive subjects.

Fig. 1

Table 5

Experiments in which cells were not stimulated with PMA and ionomycin showed no intracellular cytokine staining. Control experiments using stimulated cells which were not permeablized, or in which surface staining was blocked using an unlabelled anti-IFN-γ antibody, indicated that the cytokine staining measured was intracellular.

Discussion

In this study the levels of IFN-γ were markedly elevated in serum and pleural fluid of HIV-1-positive subjects with pleural tuberculosis, compared with HIV-negative subjects. Levels of TNF-α in pleural fluid were also elevated but this difference was not statistically significant and was less marked than reported for pulmonary tuberculosis ^[18]. Levels of IL-10 were higher in serum of HIV-positive cases, but similar in pleural fluid of HIV-negative and HIV-positive cases.

The pathogenesis of pleural tuberculosis may differ between HIV-negative and HIV-positive patients. The proportion of HIV-positive subjects in whom pleural tuberculosis represents progressive primary infection or reactivation of latent infection is not known and is difficult to establish. The presence of pulmonary infiltrates has been taken to indicate reactivation in HIV-negative subjects ^[20]. Upper lobe infiltrates were present in a slightly higher proportion of HIV-positive subjects in this study although this finding was not statistically significant. This might suggest that reactivation of disease, accompanied by activation of memory T cells, might be more important in HIV-positive patients, but equally, may simply reflect a tendency to develop disease at multiple sites in HIV-positive patients ^[15].

Loss of immunity to tuberculosis in HIV infection has been thought to result from CD4+ T-cell depletion and reduction in antigen-specific cytokine responses ^[10-12]. In keeping with this view, it was found that HIV-positive subjects had reduced IFN-γ responses to crude whole antigens in peripheral blood. This result fits the hypothesis that this assay elicits IFN-γ production predominantly through major histocompatibility (MHC) class II antigen presentation ^[21], and that CD4+ T-cell depletion was responsible for the reduced response in this assay, but contrasts with the elevated serum levels of IFN-γ observed in vivo.

In pleural fluid, as in blood, CD4+ T cells were relatively deficient in HIV-positive subjects and production of IFN-γ in the stimulation assay was lower, although in pleural fluid this difference was not statistically significant. Enhanced levels of IFN-γ in pleural fluid of HIV-positive subjects could result from increased IFN-γ production per CD4+ T cell (perhaps in response to higher levels of mycobacterial antigen in pleural fluid, not reflected in vitro); from increased production by other cell types (natural killer cells, γδ or CD8+ T cells, or macrophages); or from failure of utilization or clearance of IFN-γ. Human natural killer cells and γδ T cells can produce IFN-γ in response to mycobacteria in a non-MHC-dependent manner ^[22-25]. However, neither cell type was elevated in HIV-positive subjects, and both were present at lower percentages in pleural fluid than in blood. Messenger RNA for IFN-γ has been found in macrophages at the site of disease in patients with active tuberculosis ^[26,27] and macrophages in the pleural tissue may have contributed to IFN-γ production in our patients, but again, monocyte/macrophage numbers in pleural fluid were low. Therefore, the most probable alternative source for IFN-γ production in HIV-positive subjects was CD8+ T cells, which were present in high numbers.

The possibility that CD8+ T cells were a major source of IFN-γ was explored by intracellular cytokine staining and flow cytometry. The choice of antigen for stimulation in vitro was problematic. Whole protein antigens bias findings towards class II presentation and CD4+ T-cell responses. Live M. tuberculosis would be preferable, but antigen presentation in vitro might differ from that found in vivo where macrophages present in the pleura might participate. The choice of any mycobacterial antigen would fail to reveal activation in response to HIV antigens. Therefore, for this study, stimulation with PMA and ionomycin was used to characterize the capacity of both CD4+ and CD8+ T cells to produce cytokines. A more T-cell-specific stimulus would have been the combination of anti-CD3 and anti-CD28 antibodies. However, it is likely that the pattern of response would have been similar. Lymphocytes were the predominant cell type in pleural fluid, and were further selected in this study by gating in forward and side scatter, and studies of both murine and human cells have shown that PMA-ionomycin elicits a similar pattern of cytokine response, but a higher percentage of positive cells, when compared directly with anti-CD3 and anti-CD28 antibodies ^[28,29]. In this study, in response to PMA-ionomycin, a higher percentage of CD8+ T cells than CD4+ T cells produced IFN-γ and the proportion of CD8+ T cells capable of producing IFN-γ in pleural fluid of HIV-positive patients was significantly elevated compared with HIV-negative subjects.

The specificity of the CD8+ T-cell response is of great interest and is being examined in further studies. There is increasing evidence for a role for CD8+ T cells in the human immunity to M. tuberculosis^[30-32], and findings amongst HIV-negative participants in this study suggested that CD8+ T cells were activated by M. tuberculosis : CD8+ T cells were more likely to produce IFN-γ or TNF-α in the pleural fluid than in the blood of HIV-negative subjects. In these cases activation by HIV could not be the explanation. On the other hand, enhanced IFN-γ production by CD8+ T cells has been observed in blood of HIV-positive subjects ^[33]. Thus high IFN-γ levels in HIV-positive subjects might be attributed to responses to HIV, rather than to M. tuberculosis, and indeed we found a positive correlation between IFN-γ and viral load in the pleural fluid.

Our findings emphasize the complexity of protective immunity to tuberculosis in humans. IFN-γ clearly plays a role and administration of IFN-γ has some value in therapy of refractory human mycobacterial disease ^[34]. However, in this study, despite high levels of IFN-γ in pleural fluid of HIV-positive subjects, replication of M. tuberculosis was not controlled. Our findings may be analogous to an observation in CD4+ T-cell-deficient mice, where IFN-γ production was compensated by CD8+ T cells, but the time course of the CD8+ T-cell-mediated response was slow and bacterial replication was not controlled ^[35]. Other recent animal studies emphasize, however, that strong IFN-γ responses are not the sole requirement for protective immunity. Strategies which enhance IFN-γ responses do not necessarily enhance protective immunity against M. tuberculosis^[36,37], and over-production of IFN-γ may represent an attempt to compensate for failure of another immune effector mechanism ^[38]. In HIV-positive humans, several such explanations for failure of immunity in the presence of high levels of IFN-γ may apply. In addition, activation of immunosuppressive pathways may oppose the effects of IFN-γ and the higher numbers of mycobacteria in HIV-positive subjects may exacerbate immunosuppression by inhibiting the ability of macrophages to respond to IFN-γ^[39].

We conclude that HIV-positive individuals are capable of producing high levels of IFN-γ at the site of disease in pleural tuberculosis, and that CD8+ T cells may be a major source of IFN-γ in this setting. Despite high levels of IFN-γ, mycobacterial replication is not controlled. Investigation of the reasons for this failure could reveal critical components of the immune deficiency induced by HIV infection, and identify additional elements essential for protective immunity against M. tuberculosis in humans.

Acknowledgements

We thank the staff of the Uganda-Case Western Reserve University Research Collaboration on Tuberculosis, and of the British Medical Research Council Programme on AIDS in Uganda. We thank the patients who participated in the study, and Steven Kyaligonza and Safina Nakawagi for field work and nursing. We thank Jessica Nakiyingi for statistical support and management of the database. We thank Dr Virpi Nurmi, Professor James Whitworth, Dr Hazel Dockrell and Professor Jerrold Ellner for scientific advice and support. We thank the anonymous reviewers for their helpful suggestions regarding the discussion of our findings.