Infection with Helicobacter pylori Is Associated with Protection against Tuberculosis

Background Helicobacter pylori, a lifelong and typically asymptomatic infection of the stomach, profoundly alters gastric immune responses, and may benefit the host in protection against other pathogens. We explored the hypothesis that H. pylori contributes to the control of infection with Mycobacterium tuberculosis. Methodology/Principal Findings We first examined M. tuberculosis-specific IFN-γ and H. pylori antibody responses in 339 healthy Northern Californians undergoing routine tuberculin skin testing. Of 97 subjects (29%) meeting criteria for latent tuberculosis (TB) infection (LTBI), 45 (46%) were H. pylori seropositive. Subjects with LTBI who were H. pylori-seropositive had 1.5-fold higher TB antigen-induced IFN-γ responses (p = 0.04, ANOVA), and a more Th-1 like cytokine profile in peripheral blood mononuclear cells, compared to those who were H. pylori seronegative. To explore an association between H. pylori infection and clinical outcome of TB exposure, we evaluated H. pylori seroprevalence in baseline samples from two high risk TB case-contact cohorts, and from cynomolgus macaques experimentally challenged with M. tuberculosis. Compared to 513 household contacts who did not progress to active disease during a median 24 months follow-up, 120 prevalent TB cases were significantly less likely to be H. pylori infected (AOR: 0.55, 95% CI 0.0.36–0.83, p = 0.005), though seroprevalence was not significantly different from non-progressors in 37 incident TB cases (AOR: 1.35 [95% CI 0.63–2.9] p = 0.44). Cynomolgus macaques with natural H. pylori infection were significantly less likely to progress to TB 6 to 8 months after M. tuberculosis challenge (RR: 0.31 [95% CI 0.12–0.80], p = 0.04). Conclusions/Significance H. pylori infection may induce bystander effects that modify the risk of active TB in humans and non-human primates. That immunity to TB may be enhanced by exposure to other microbial agents may have important implications for vaccine development and disease control.


Introduction
The microbes that colonize the human host are diverse and demonstrate geographic and temporal variability. This variability is exemplified by H. pylori infection, a gastric mucosal pathogen that comprises part of the ''normal flora'' in much of the developing world, but has receded over time in higher socioeconomic regions of the world. H. pylori has been colonizing humans for at least 50,000 years [1]. Why its prevalence varies so dramatically based on socioeconomic status is not known but may relate to antimicrobial use, improved household and environmental sanitation, and decreased crowding. Another hypothesis, however, is that H. pylori infection provides a survival benefit against challenges present disproportionately in poorer geographic regions. By boosting mucosal and systemic immunity, the organism may limit the consequences of other infectious exposures [2,3] and selectively promote survival of H. pylori infected hosts.
One third of the world's population is latently infected with the intracellular pathogen M. tuberculosis [4]. Following exposure, the organism elicits an IFN-c-driven, cellular immune response, causing infected macrophages to be sequestered within organized lung granulomas [5,6]. In approximately 90% of humans, the host immune response controls but does not eliminate the infection, preventing progression to active disease throughout the host's lifespan [7]. Those infected with M. tuberculosis but with no symptoms of disease are referred to as latently infected. Although risk of active tuberculosis is significantly elevated in immunocompromised hosts (e.g., those with HIV infection [8] or treated with immunosuppressants [9]), the great majority of individuals who develop active TB do so in the absence of known immunocompromise. The nature of protective immunity remains unknown.
M. tuberculosis and H. pylori are the most prevalent bacterial pathogens worldwide. In much of the world's population, these obligate human infections coexist throughout most of the life span, continuously interacting with the host immune system without causing disease. Almost nothing is known about the crosstalk of these infections and whether one infection affects the clinical manifestations of the other. The few studies examining an epidemiologic linkage between H. pylori and tuberculosis have yielded conflicting results [10,11,12]. While conducting a study of TB diagnostics in a population that had been tested for H. pylori, we fortuitously identified differences in interferon gamma (IFN-c) and other cytokine responses to M. tuberculosis antigens in H. pyloriinfected and uninfected hosts. To explore this relationship further, we compared rates of H. pylori seropositivity in blood samples from TB cases and household contacts recruited from TB case-contact studies carried out in The Gambia [13] and Karachi, Pakistan [14]. We also compared outcome of M. tuberculosis challenge [15] in macaques with and without naturally-acquired H. pylori infection. Our results support further investigations into the contribution of H. pylori infection to the protective immune response to TB infection.

Overview
Three distinct studies were undertaken sequentially. Studies involving human subjects or samples were conducted in accordance with principles expressed in the Declaration of Helsinki. Non-human primate studies were conducted in accordance with the United States Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals of the Institute for Laboratory Animal Research, National Academies of Science. Each study was approved by the appropriate Institutional Review Boards as described.
(1) IFN-c responses to TB antigens in Northern (2) H. pylori sero-prevalence in human tuberculosis casecontact cohorts. De-identified plasma samples obtained at a baseline screening visit were recruited from the specimen banks of tuberculosis case-contact studies conducted by the Medical Research Council, The Gambia, West Africa [16] and the Aga Khan University, Karachi, Pakistan [14], respectively. Each study enrolled households based on an index case of active tuberculosis, and assessed participants clinically for at least 24 months from baseline, with overall rates of activation 1.1% (The Gambia [13]) and 6.4% (Karachi [14]) previously reported. Active TB was ascertained by symptoms, chest X-ray and AFB smear and culture in The Gambia [16], and by symptoms, chest X-ray and AFB smear in Pakistan [14]. Baseline TB infection was determined by positive ($10 mm) TST and/or ELISPOT in The Gambia [17] and by positive TST ($10 mm in Pakistan [14]. Blood samples recruited for the present studies were from HIV-negative subjects with blood drawn before the onset of treatment of a TB prevalent or incident case. Sampling proportions and baseline characteristics of subjects whose samples were selected for analysis are shown in Table S4. From The Gambia, 549 samples were randomly selected in 2 phases from a pool of 2626 eligible samples ( Figure 1A). The first group consisted of samples from 100/295 eligibleTB cases (household index cases) and 100/2271 eligible age-and sexmatched household contacts who were TB infected at baseline and known to have remained disease-free for at least 2 years of followup (nonprogressors). Samples from Gambian TB cases were proportionately representative of Gambian TB isolates (approximately 40% M. africanum and 60% M. tuberculosis [18]). The second sampling phase included samples from 29/32 subjects who developed active TB 3-47 months from baseline, and 320 ageand sex-matched nonprogressors. The study was approved by the Institutional Review Boards of Stanford University and the MRC Ethics Committee. A complete sample was obtained from the Pakistan cohort ( Figure 1B), including 20 index TB cases, 8 incident TB cases ascertained 3-49 months from baseline, and 93 (3) TB challenge in cynomolgus macaques with and without naturally acquired H. pylori infection. Stored baseline sera, frozen gastric necropsy samples, and clinical necropsy reports were obtained from a convenience sample of 45 cynomolgus macaques (Macaca fasicularis) (.4 years of age) used in ongoing or completed low-dose TB challenge studies. All monkeys included in our study had been challenged endotracheally with 25 CFU of the Erdman strain of M. tuberculosis. Monkeys were assessed clinically for at least eight months post-challenge, and were classified as having active TB or latent infection based on previously defined criteria [19]. Specifically, monkeys were classified as having active TB if they had any clinical signs of disease, including weight loss, appetite loss, cough, radiographic findings, elevated erythrocyte sedimentation rate (ESR), or BAL or gastric aspirate that grew M. tuberculosis after the first two months of infection. Monkeys were classified as latent if they exhibited no signs of active disease (by the measures described above) and had culture negative BAL and gastric aspirate samples after 2 months of infection. These criteria have been validated at necropsy in a large group of monkeys [20]. The Pittsburgh model has previously yielded approximately 50% active and 50% latent disease post-challenge [19,20]. The outcome is independent of gender, weight or age (between 4-10 years of age) in this model [20]. The animals were housed and maintained in accordance with standards established in the Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals. All procedures and protocols were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Pittsburgh.

Laboratory Methods
H. pylori infection in humans. The diagnosis of H. pylori infection was ascertained by enzyme-linked immunosorbent assay (ELISA). Per protocol, all samples were run in triplicate by technicians blinded to the TB infection or outcome classification of samples, and samples with different infection classifications were intermixed on the same plates and run on the same days. Final serologic interpretations were completed before samples were linked to case records.
For the Northern California subjects, a high molecular weight whole cell lysate ELISA antigen was derived from 2 Asian, 2 Mexican, and 2 US isolates of H. pylori with results interpreted according to validated cut-offs, as previously described [2,21]. Because antibodies to H. pylori's immunodominant CagA protein, present in approximately 65% of H. pylori isolates, may be more reliable in samples from developing countries [22,23], Pakistan and Gambian samples were tested for IgG antibody response to the orv220 fragment of H. pylori CagA (courtesy H. Kleanthous) as previously described [24]. A sample was considered positive if . Each study followed household members exposed to an active (index) case of tuberculosis for at least 2 years. Eligible samples were from HIV-negative participants. Samples from Gambia were randomly selected in 2 groups to match household contacts by age and sex to index and incident TB cases, respectively. LTBI: latent tuberculosis infection at baseline, defined as TST $10mm or ELISPOT $8 SFU (Gambia), or as TST $10mm (Pakistan). Baseline characteristics are shown in Table S4. doi:10.1371/journal.pone.0008804.g001 optical density reading exceeded the mean + 3SD of 15 CagA negative controls. Indeterminate samples (n = 24) were classified as negative for analysis.
Measles antibodies. To establish baseline humoral response, quantitative IgG antibody response to measles was compared in a subset of samples from 100 culture-confirmed Gambian TB patients and 100 age and sex-matched household contacts with latent infection, using the Bioquant IgG ELISA (San Diego, CA) with qualitative results interpreted according to manufacturer's criteria.
TB antigen-induced cytokine studies. We used the QuantiFERON-TB-GOLDH (Cellestis, Ltd, Carnegie, Victoria, Aus) assay to quantify stimulated PBMC IFN-c responses to specific M. tuberculosis peptides as previously described [25]. Quantitative results, expressed as IU/ml, representing the difference in IFN-c response to M. tuberculosis peptides or PHA and a saline control for each subject, were transformed for statistical analysis using the natural log base. Residual supernatants from stimulated PBMCs were stored at 280uC and subsequently, 65 samples from adults were further assayed in a Luminex xMAP 200 platform using customized bead preparations (Beadlyte Human 26-plex Multi-Cytokine, Millipore, Billerica, MA). Concentrations for a standard reference curve were included on each plate per manufacturer's instructions. For each subject cytokine concentrations in the unstimulated control well were subtracted from TB specific responses, expressed as pg/ml.
H. pylori diagnostic procedures in monkeys. Frozen serum samples obtained before M. tuberculosis challenge (N = 44) or soon thereafter (1 sample obtained 33 days post infection) were sent to Stanford University for ELISA. A rhesus-derived H. pylori strain was employed for the ELISA antigen. In prior studies with the rhesus model, this assay had an estimated sensitivity and specificity against endoscopy or breath test of 96% and 88% respectively [26]. Gastric tissue obtained at necropsy (n = 28 monkeys) and frozen at 280C for between 1 week and 22 months (median 13.8 months), was shipped to U.C. Davis, fixed, and processed in the laboratory of J. Solnick for H. pylori culture and histopathology respectively as previously described [15].

Statistical Analysis
Statistical analysis was conducted using SAS v. 9.3. All statistical tests were carried out at alpha set to 0.05 for a two-sided test. The following analyses were conducted: (1) IFN-c responses to TB antigens in Northern Californians with and without H. pylori infection. To evaluate whether concurrent H. pylori infection affects the immune response to M. tuberculosis antigens, we compared quantitative TB specific IFN-c responses stratified by H. pyloriand latent M. tuberculosis infection. H. pylori infection was defined by results of the whole cell lysate ELISA. We defined latent TB infection (LTBI) presumptively as either M. tuberculosis specific IFN-c response $0.35 IU/ml or TST result $10 mm, with no signs or symptoms of active TB. A Generalized Linear Model was used to compare group means accounting for the interaction of H. pylori and LTBI as well as to adjust for potential confounders (age, sex, and country of origin); p-values are based on the F-test (Fisher's LSD) for pre-planned contrasts (e.g. H. pylori effect within LTBI). For multiplex-cytokine analysis, Principal Components Analysis [27] was performed with the correlation matrix composed of six variables found to discriminate by Wilcoxon's test between 40 latently infected and 25 controls (Tables S1 and S2), and components having Eigenvalue .1 were selected to compute summary component scores for each subject using eigenvector coefficients.
Comparison of H. pylori prevalence in human tuberculosis case-contact cohorts. We compared rates of H. pylori CagA infection in index TB cases, incident TB cases, and household contacts known not to have progressed to active disease during follow-up (nonprogressors). Logistic regression was used to evaluate factors associated with H. pylori infection in each cohort and in the cohorts combined. Odds ratios and 95% confidence intervals of H. pylori infection in nonprogressors compared to prevalent or incident TB cases were computed with adjustment for age, gender, latent M. tuberculosis infection at baseline, and cohort location (Pakistan or The Gambia). Goodness of fit was evaluated using the Hosmer-Lemeshow test [28]. A generalized nonlinear mixed model with random intercept (SAS GLIMMIX) was used to examine results with adjustment for household membership.
M. tuberculosis challenge in cynomolgus macaques with and without naturally acquired H. pylori infection.
Monkeys were categorized clinically by 6-8 months after aerosol infection as having active tuberculosis or latent infection. H. pylori infection in monkeys was defined as a positive serologic test for H. pylori prior to M. tuberculosis infection. Because gastric tissue had not been obtained or conserved for the purposes of H. pylori histology or culture, negative results to these two tests were not considered reliable for determining presence or absence of infection. Relative risk and 95% confidence intervals of active vs. latent tuberculosis in H. pylori infected versus uninfected monkeys was computed.
Compared to 52 latently infected subjects without H. pylori infection, mean IFN-c responses to specific TB antigens were approximately 1.5 fold greater in 46 subjects with concurrent H. pylori infection (Figure 2A), and this difference remained significant after adjusting for age, sex and country of birth (p = 0.04, ANOVA). In contrast, IFN-c responses did not vary significantly with respect to H. pylori infection among 242 LTBInegatives (p = 0.40), indicating an interaction effect between H. pylori and LTBI (p = 0.03, ANOVA). These results were similar when restricting the definition of LTBI to TST positivity only (data not shown).
The effect on IFN-c responses to M. tuberculosis antigens of past exposure to Hepatitis A virus, a picornavirus infection with epidemiologic patterns similar to those of H. pylori [29] was similar to that for H. pylori ( Figure 2B). In fact, in a 3-way interaction model, having H. pylori infection, HAV and LTBI was associated with greater IFN-c responses to TB than having either infection singly with LTBI ( Figure 2C).

Comparison of H. pylori Sero-Prevalence in Human Tuberculosis Case-Contact Cohorts
Among a total 670 samples tested, 425 (63%) were H. pylori seropositive according to the CagA assay, including 59/121 (49%) of Karachi, Pakistan and 366/549 (67%) of Gambian household contacts (p = 0.0002, supplemental Table S5). H. pylori seroprevalence varied significantly by TB outcome classification ( Table 2, and Table S6 for results by cohort). Compared to 513 household contacts who remained disease-free throughout follow-up, the 120 TB cases were about K as likely to be H.
To explore whether the diminished antibody responses to H. pylori in prevalent TB patients might result from immune deficits associated with having active disease, we evaluated measles antibody responses in 100 Gambian culture-confirmed TB cases and 100 age-and sex-matched latently infected nonprogressors. Equal numbers of cases and contacts responded to a measles IgG antibody test (90% and 88% respectively) with virtually identical mean optical density values ( Figure S1), suggesting that active TB disease did not cause a diminution of humoral responses.   Although, as expected, culture yields in seropositive monkeys were low due to tissue autolysis and storage, all 6 H. pylori culturepositive monkeys were positive by serology, as were 4 of 6 H. pylori histology positive monkeys. Of the 30 monkeys with positive H. pylori serology (Table 3), only 5 (16.7%) developed active tuberculosis whereas 6 (54.6%) of the 11 negative monkeys developed active tuberculosis (relative risk 0.31, 95% CI 0.12-0.80, p = 0.04, Fisher's Exact test).

Discussion
Humans are colonized by complex, site-specific microbial communities that are increasingly implicated in human health and disease in unexpected ways. In some cases, microbial communities are thought to modulate non-infectious diseases, such as obesity [30], asthma and autoimmune diseases [31]. Viewed broadly, the human microbiota may be a major regulator of the immune system modulating not only inflammatory disorders (the ''hygiene hypothesis''), but responses to infectious challenges as well [32]. Because they continuously stimulate non-specific responses in the host, chronic mucosal infections may be particularly important in this regard.
The high prevalence of H. pylori in populations where TB and other lethal infections remain endemic suggests the host-pathogen interplay of these infections has co-evolved. In this report, we have presented evidence from three different study designs that H. pylori infection affects response to M. tuberculosis. First, we found in a cross-sectional human study that H. pylori is associated with enhanced IFN-c and Th1-like responses to specific TB antigens. Second, in baseline samples from two high risk human tuberculosis case-contact cohorts, we observed that infected household contacts who maintain latency over two years are significantly more likely to have concurrent H. pylori infection than are TB cases, although seroprevalence was not different in incident TB cases. Finally, in a retrospective cohort study of low-dose M. tuberculosis challenge monkeys, we observed that H. pylori-infected animals are significantly less likely to develop active disease than are uninfected animals. While further studies are indicated, taken together, these lines of investigation offer a heretofore unexplored role for infections like H. pylori to alter the outcome of M. tuberculosis infection.
We propose two potential models by which H. pylori could promote a protective immune response to TB infection. In the first model, an infection in early childhood could permanently differentiate immature T-cells to a Th1-like phenotype. Such a ''hygiene hypothesis,'' has been advanced for hepatitis A [33], and would favor past as well as ongoing infections contributing to a state of heightened immunity. Alternatively, H. pylori might induce a bystander effect with continuous inflammation and T-cell signaling enhancing the host's innate response to a spectrum of infectious challenges. Similar effects have been demonstrated with c-herpesvirus in mice challenged with Listeria monocytogenes and Yersinia pestis [34]. These authors speculate that non-specific induction of interferon-c (IFN-c) activates macrophages and primes innate defenses to other infections. We did not test for antibodies to HAV or herpesviridae in the human tuberculosis case-contact cohorts. Because these exposures are likely to be very common in populations where both H. pylori and M. tuberculosis infections are common, more systematic sampling designs may be needed to adequately explore interaction effects.
H. pylori naturally infects macaques. The fact that we were able to replicate a significant association of H. pylori infection with latent outcomes in the Pittsburgh cynomolgus monkey TB challenge model [19,35] offers new opportunities to explore mechanistic arguments in depth. Because the present work is based on a convenience sample from the Pittsburgh laboratory, repeated measures of immunologic responses and clinical progression were not investigated. More systematic experimental study designs, including prospective studies utilizing H. pylori challenge, are planned.
The accuracy of H. pylori serology in the developing world is suboptimal [22]. Although the CagA assay may be more reliable for comparing different populations, misclassification of H. pylori infection status still cannot be excluded. Typically, misclassification tends to yield a null result unless, for some reason, cases and controls respond differently to the assay. We speculated that TB cases might have weakened antibody responses, causing us to have a spurious result. However, the fact that measles antibody titers were robust and virtually the same in Gambian TB cases and age and sex-matched latently infected contacts argues against this explanation of results. That H. pylori seroprevalence did not differ between incident TB cases and nonprogressors in our cohort samples, while limiting our conclusions, can also reflect artifacts of the TBCC study model, including the low numbers of secondary cases, ascertainment differences or biases, as well as limitations of H. pylori serologic detection in this setting. In both cohorts, most cases of TB were identified at an initial screening visit, and M. tuberculosis infection at baseline was a relatively weak predictor of progression [14,36]. Thus, continued work with longitudinal cohorts may benefit from multi-site study designs incorporating additional H. pylori diagnostics as well as other prognostic biomarkers of immune response to TB exposure and infection.
M. tuberculosis specific antigen interferon-c release assays are important noninvasive tools for measuring immunogenicity of the heterologous prime boost T-cell based vaccines for TB [37,38]. Our results raise the possibility that H. pylori infection and other Th-1 modifying infections present in the host background can alter these profiles. As our results also do not exclude the possibility that M. tuberculosis modifies the host response to H. pylori infection, PBMC responses to TB antigens pre-and post-antibiotic treatment for H. pylori would help shed light on the specificity of our findings, including reversibility. In vitro studies examining possible mechanisms of T-cell cross reactivity, such as effect of H. pylori antigen on M. tuberculosis antigen presenting cells and MHC expression, are also needed.
Why only 10% of infected individuals succumb to tuberculosis remains one of the most vexing public health questions-one which the one-pathogen-one-disease paradigm is ill-equipped to answer. While preliminary, our work suggests that one factor contributing to the clinical outcome of TB infection may be a concurrent chronic infection. The hypothesis that the human microbiome has evolved to provide context-specific competitive risk advantages to the host [39] also raises the intriguing possibility that our microbiota can be manipulated to modulate disease risk from M. tuberculosis, as well as other common human pathogens [40].