Tuberculosis Vaccines and Prevention of Infection

Natural history studies: animal models.

A natural TB infection animal model offers the potential to address several difficult questions about the transmissibility of M. tuberculosis as well as mechanisms of resistance to infection and/or progression to active disease. The guinea pig is highly and uniformly susceptible to M. tuberculosis infection and disease under laboratory conditions of infection (89, 90). In contrast, under natural exposure conditions, guinea pigs demonstrate a range of susceptibility that is similar to that of humans. Riley et al. designed the original natural exposure experiments in the 1950s with shunting of air from hospital rooms of humans with pulmonary TB into separate rooms where guinea pigs were housed (91–93). Guinea pigs had serial TST skin tests during the experiment with necropsy at the conclusion and culturing tissue for viable M. tuberculosis. Some infected animals with positive TSTs reverted their skin test and had no evidence of active TB disease at necropsy (91, 92). These data suggested that reversion was associated with sterilizing immunity. An alternative interpretation of these data is that the initial TST conversion was due to exposure to dead bacilli that did not produce a true infection. Subsequent studies assessed this possibility by exposing the ward air to UV light and demonstrating that killed M. tuberculosis did not induce conversion to a positive TST (94).

More recently, Dharmadhikari et al. extended these studies and examined transmissibility and disease progression in guinea pigs naturally exposed to multidrug-resistant (MDR) M. tuberculosis (93). Serial TST skin testing was done at baseline and throughout 20 weeks of exposure to assess whether TST magnitude and kinetics were associated with disease progression. Among 362 guinea pigs exposed for 4 months to patients with MDR TB, 75% of animals had positive TSTs (>6 mm) (93). Interestingly, only 12% of the animals with a positive TST had active TB when assessed at necropsy by histology and culture of lung, spleen, and lymph node. To determine whether early or late TST conversions were predictive of different outcomes, animals with large TSTs (>14 mm) at 20 weeks were analyzed further. Those with early conversions had lower disease severity than those with later conversions. To further assess the kinetics of TST conversion, those with positive tests in the 6- to 13-mm range were analyzed. Twenty-two percent (n = 86) of these animals had a reversion of their test to negative, and only 2/86 had active disease. In contrast, 47% of animals with nonreverting skin tests had active disease. Even after steroid administration to a subset of animals, there was no evidence of active disease in the animals exhibiting TST reversion. A potential confounding variable is that serial skin testing could induce or boost cell-mediated immunity. Although this is possible, other investigators reported only rare false-positive conversions in control animals without exposure to M. tuberculosis that received serial skin tests (95). Together, these experiments demonstrate that natural exposure causes a more diverse set of outcomes (resistance, reversion, and range of disease severity) than laboratory conditions. Some of these differences may be due to the metabolic state of the bacillus or the magnitude of the exposure. In addition, these experiments suggest that a guinea pig can completely clear M. tuberculosis infection and that reversion of a TST was correlated with this possible sterilizing immune response. These studies are promising and provide an impetus to understand whether immune responses that can clear an established infection exist in humans. In addition, the difference in outcomes between natural and laboratory-controlled exposure suggests that vaccine trials in animal models may need to incorporate such conditions in their design to avoid artifacts of laboratory testing conditions.

Natural history studies: humans with resistance to TB infection.

Are any humans resistant to TB infection? Although many individuals become infected with M. tuberculosis after sustained exposure, only a small fraction resist infection as demonstrated by persistently negative TSTs or IFN-γ release assays (IGRAs) (96, 97). These individuals may have an innate macrophage response to M. tuberculosis that resists initial infection or rapidly clears the bacillus before a T-cell response develops. If such innate resistance did not exist, one might expect LTBI rates in regions where TB is endemic to approach 100%. Resistance is likely a threshold phenomenon that is related to the degree of exposure. Surprisingly, approximately 5 to 10% of individuals remain tuberculin skin test negative despite sustained exposure to an infectious TB case (96). Addressing this question experimentally requires a longitudinal study design with documented events of exposure to individuals with pulmonary TB, ideally with high bacillary loads in a household where there is extensive exposure with contacts. Since 1995, the National Institute of Allergy and Infectious Disease (NIAID)-supported Tuberculosis Research Unit (TBRU) conducted a household contact (HHC) study in Kampala, Uganda (98–105). Index cases were identified at the Uganda National Tuberculosis and Leprosy Program treatment center at Old Mulago Hospital in Kampala, Uganda. Participants were enrolled if they were 18 years or older, had sputum smear-positive and/or culture-positive pulmonary TB, and had at least one HHC living with them. Individuals whose TSTs were negative at all follow-up visits were considered to be resisters (RSTR), and individuals who developed a positive TST during study follow-up were considered TST converters (96, 97). All of the individuals in the HHC study live with the index case and have many risk factors for close proximity to the index case. Extensive epidemiological analysis has not revealed any exposure variables that distinguish those who convert their TST from those who do not. A potential confounder of the study design is that M. tuberculosis-specific T cells may be present in other tissues (e.g., skin or lung) and not detected by skin testing or peripheral blood assays. A whole-genome linkage scan and candidate gene study was performed to identify human genetic risk factors for resistance to TB infection (RSTR) and active TB disease. In the genome scan, regions linked to active TB differed from those linked to RSTR. Linkage results for RSTR included a chromosome 5 region that has since been replicated by an independent study (106). Although not conclusive, these data suggest that some humans are naturally resistant to TB infection and that genetic factors may regulate this difference. Identification of the immunogenetic factors associated with resistance may lead to novel strategies for immunomodulatory therapy. These strategies might include the use of small-molecule drugs that target host macrophage pathways or vaccines that induce T-cell responses which activate macrophages to kill M. tuberculosis (107).

Natural history studies in humans with preexisting TB infection: is there evidence of protection?

Is there evidence of acquired immunity to TB from natural history studies? This question has been partially addressed by studies in regions of high endemicity that compare active TB incidence rates in those with or without TB infection. A high-endemicity setting is required to assess this question since there is an assumption of ongoing exposure to M. tuberculosis during the study period. The classic study to address this question was performed by Johannes Heimbeck in Norway, who administered TSTs to nursing students who entered school between 1924 and 1936 and worked in a hospital with a large number of TB patients (108). Among 1,453 students, 45.3% were TST positive at study entry and 54.7% were negative. Within the TST-negative group who did not receive BCG vaccination (n = 284), 34.2% developed active TB disease and 10 died. In contrast, 3.3% (n = 22) of the TST-positive nurses developed disease, with no deaths. Multiple studies subsequently examined this question with prospective cohort designs in the prechemotherapy era. In a meta-analysis of 18 studies with an aggregate sample size of 19,886 individuals, Andrews et al. estimated that individuals with latent tuberculosis had a 79% lower risk of developing active TB disease after reinfection than uninfected individuals (109). In a more recent study in a region in South Africa with a high TB burden, 6,363 adolescents 12 to 18 years old were followed for 2 years with serial TST and QuantiFERON testing (QFT) (110, 111). Among those who had a positive baseline QFT, the incidence rate was 0.64 active TB cases/100 person years. In contrast, the rate among those who converted during the 2-year observation period was 1.46 cases/100 person years. In another recent study of 764 households in Peru in an area of high BCG vaccination rates, modeling of cross-sectional data estimated that previous infection reduced the risk of reinfection by 35% compared to that for uninfected individuals (112). Together, these studies suggest that individuals with established M. tuberculosis infection have decreased susceptibility to active TB disease after reinfection in comparison to those who are uninfected. These data are consistent with a concept that humans can acquire immunity to TB from natural exposure.

Natural history studies: spontaneous human reverters.

If humans could eradicate M. tuberculosis after infection, it would support the rationale that such protective immunity could be induced in a vaccine. Although this immunity would not prevent infection, it could possibly eliminate M. tuberculosis in the early stages before it establishes a chronic infection. Is there evidence from natural history studies that such “natural immunity” exists? Although the majority of individuals with a positive TST will remain positive for their lifetime, the phenomenon of reversion has been recognized for over a century (113). Spontaneous reversion (in the absence of INH treatment) may represent a successful host response, and these individuals can be studied to identify correlates of protective immunity. Alternatively, reverters may have suppressed or defective immune responses and still harbor active bacilli that are fully capable of causing active TB disease. Does any epidemiological evidence favor one of these models? Early studies demonstrated that many individuals spontaneously revert from a positive to a negative TST. This was first published in 1913 by Gelien and Hamman in their study of 1,000 individuals in Baltimore, MD, who received a TST in ∼1908 and a repeat test 1 to 3 years later (114) (Table 2). Nearly 50% of those subjects reverted their positive test. Many additional studies had similar findings over the next 100 years and explored factors associated with reversion. Consistent findings across multiple studies suggest that reversion is more common in children (115–117), less common in those with a higher-magnitude initial conversion (113, 118), and possibly more common in females (115, 119). Rates of spontaneous reversion may depend on several variables, including BCG vaccination, exposure to nontuberculous mycobacteria (NTM), a booster phenomenon from prior TSTs, duration of positivity before reversion, age, and magnitude of the conversion and reversion (118, 120). To control for some of these variables, recent studies have used M. tuberculosis antigen-specific tests and found high rates of reversion for recent converters in analyses that were adjusted for these variables (110, 117, 119). Hill et al. used a CFP10/ESAT-6-based enzyme-linked immunosorbent spot (ELISPOT) assay in the Gambia and reported reversion rates of up to 36% (117). Using the IGRA QuantiFERON-TB Gold (QFT-TB Gold), Shah et al. found a reverter rate of 15% in children in Soweto, South Africa, and Machingaidze et al. found a rate of 9.2% in Worcester, South Africa (110, 119).

Several reasonable concerns have been raised regarding the interpretation of the significance of reversion. First, some reversions likely represent borderline positive results with fluctuations slightly above and below the positivity threshold. Although this is the case for some reverters, multiple studies have demonstrated that the majority of reverters have large and often complete reversions, with a second test of 0-mm induration after an initial conversion magnitude of 10 to 15 mm. Second, reverters may have transient exposure to nontuberculous mycobacteria rather than true infection with M. tuberculosis. More recently, this concern has been fully addressed with three studies that used IGRAs (QFT-TB Gold In-Tube [QFT-GIT] or ELISPOT) and found reversion rates of 8.9 to 15% (110, 117, 119). Third, reversions are a phenomenon found only in low-incidence areas and may represent false-positive tests and/or tests compounded by laboratory errors. This concern is unlikely for multiple reasons, including that the majority of the studies are from countries that currently have high TB endemicity or those that previously had high rates (e.g., the United States in 1900 to 1950). The consistency, magnitude, and frequency of the reverter phenomenon over multiple studies suggest a durable finding.

What does reversion represent immunologically? The reversion of M. tuberculosis-specific immune responses could be caused by immune suppression, egress of M. tuberculosis-specific T cells from the blood, clearance of TB infection, or lowering of the M. tuberculosis bacillary load. Although the TST conversion and reversion event likely includes CD4⁺ T-cell IFN-γ, there has been no systematic study of the details of the cellular source of IFN-γ or the breadth of other immunological responses and whether they also revert. For example, other immunological responses that may revert include other cytokines (e.g., IL-2, TNF, and other T-cell cytokines), different T-cell subsets (e.g., CD4⁺ versus CD8⁺ versus CD1), different CD4⁺ T-cell phenotypes (e.g., TH1 versus TH2 versus TH17 and central versus effector memory), and responses to other M. tuberculosis antigens (beyond CFP10, ESAT-6, and TB7.7 in the IGRA). Reversion of either global or restricted M. tuberculosis-specific immune responses could represent immunologically useful signals to dissect protective immune responses.

Do reverters have a lower incidence of TB disease than converters? The most comprehensive assessment of this question was performed by Arthur Dahlstrom with an HHC study of TB index cases in Philadelphia in the 1920s (116). A total of 3,919 individuals from 513 households were followed for at least 5 years with TSTs, chest X rays (CXRs), and clinical assessments. Among 2,828 individuals for whom at least 2 TSTs were performed, 276 (11.1%) reverted their test to negative (<5 mm). Among the 913 (23.3%) of individuals with active TB in this study, none were TST reverters. Among 274 reverters with CXRs obtained at the time of their reversion, 2 had evidence of active childhood TB (presumably without symptoms), while the remainder showed either no TB (n = 256) or inactive TB (n = 16). Together, these data suggest that reverters have a minimal risk of developing active TB over 5 years (0.72% versus 23.3% of the entire cohort). In addition, the same study as well as others demonstrated that some reverters develop evidence of substantial TB disease that subsequently heals with calcified granulomas (113, 116). Together, these studies support the concept that reverters are individuals who possess protective immune responses to M. tuberculosis that could be immunologically interrogated to discover correlates of protective immunity. To our knowledge, such studies have not yet been performed. Although suggestive, these studies were performed without current standards of study design and statistical analysis, and interpretation of some of the data is uncertain. Repeating a longitudinal study of reversion and its clinical outcomes along with immunological interrogation of the response is needed to determine the significance of reversion.

BCG animal studies.

Since BCG was first developed, its effects as a vaccine have been investigated in countless experiments in different animals from rodents to rudiments to primates, each with intrinsic limitations and differences. Generally, the majority of these studies employed aerosol routes to deliver M. tuberculosis challenge doses that are as low as possible and yet still ensure uniform infection and disease outcome for all animals. Thus, infection doses in experimental animal studies tend to be significantly higher (50 to 3,000 CFU per animal, commonly 10 to 20 for guinea pigs and 100 for mice) than the relatively few bacilli thought to be transmitted during natural exposure (though the precise number has never been measured) (91, 121–123). In addition, the bacillus is prepared under laboratory conditions with a metabolic state and physical conditions (sputum versus broth culture) that likely differ from those in natural transmission settings (93). In this setting, BCG does not promote sterile clearance of M. tuberculosis but rather controls M. tuberculosis replication to different degrees. In the best-studied murine, macaque, guinea pig, and rabbit models, BCG recall responses have been shown to arrest M. tuberculosis growth several days earlier than in primary infection (89, 90, 124–127). Consequently, bacterial burdens and disease severity are reduced and time to death is increased.

BCG human studies.

In addition to animal data, there are human studies that suggest a possible protective effect of BCG. Prior to the development of IGRAs, this question was confounded by cross-reactivity between purified protein derivative (PPD) and BCG due to common antigens. The antigens used in the T-SPOT ELISPOT assay (CFP10/ESAT-6) and QFT-TB Gold In-Tube assay (CFP10/ESAT-6/TB7.7) are not present in BCG and thus can be used to assess whether BCG vaccination is associated with protection from TB infection. A recent meta-analysis of 14 retrospective case-control studies suggests that BCG is associated with protection from M. tuberculosis infection (n = 3,855; overall risk ratio, 0.81; 95% confidence interval [CI], 0.71 to 0.92) (156). Four additional studies were not included in the meta-analysis. Three of those four studies also showed a protective effect (Table 3) (117, 157, 158). One limitation of these studies is that they were retrospective and most relied on the presence of a BCG scar to document vaccination status (except the study from Greenland, which had birth records of BCG vaccination). Several studies that demonstrated a protective association were performed in countries with low or medium incidence (United Kingdom, Europe, Turkey, and Greenland). These data suggest that BCG protection could be dependent on the level of exposure, with protection waning in high-exposure settings. Differences in efficacy that correlate with geography also suggest possible effects from nontuberculous mycobacteria (NTM) which could potentially modulate vaccine responses through heterologous priming from cross-reactive mycobacterial antigens. NTM exposures can vary substantially in different geographic regions, which might explain the disparate outcomes of BCG vaccination in large trials conducted in northern latitudes (e.g., British Medical Research Council trials with 80% efficacy) and equatorial regions (Indian Council of Medical Research Trial in India and Karonga Prevention Trial in Malawi with 0% efficacy) (159–161). Although some immunological data support this hypothesis, the nature, magnitude, and mechanism of this potential modulatory effect are unknown (162, 163). These issues could affect the choice of where to conduct a prevention-of-infection trial. Although these studies are not randomized, prospective, consistent, or conclusive, they do suggest that BCG may protect against TB infection. Based on the animal and human studies published to date, an important next step for the field would be to conduct a randomized clinical trial of BCG vaccination for prevention of TB infection.

Model for intensity and magnitude of M. tuberculosis exposure.

We start by assuming that M represents the number of exposure events over a given year of follow-up. Note that we implicitly assume a model of discrete exposure events such as might arise via social contacts rather than a model of continuous exposure such as might arise via constant contact with an infected caregiver. We assume that the distribution of M has expected value μ_M and variance σ²Mμ_M. The parameter μ_M is used as an index of the intensity of exposure over time, and σ²M is an index of variability over individuals in exposure intensity. We specifically use an overdispersed Poisson (negative binomial) distribution for M to simplify our formal calculations.

Now suppose that for a given exposure event, the number of viable bacilli (or droplet micronuclei) deposited on the lung alveolar surface is represented by X. We assume that the distribution of X is also negative binomial with expected value μX and variance σ²Xμ_X and that the magnitudes of exposure realized over separate exposure events are independent. The parameter μX is used as an index of the magnitude of exposure over exposure events, and σ²N is an index of variability in magnitude of exposure events over time. Note that X can more generally be thought of as a surrogate measure of the infectious potential for a given exposure (e.g., due to M. tuberculosis strain variation).

The model linking exposure to infection is completed by assuming that given an exposure event of magnitude X, the probability of a persistent M. tuberculosis infection is given by the function P(X:τ), where τ is a parameter that controls the absolute probability of infection. For fixed values of τ, the function P(x;τ) should take value zero when x = 0 (no exposure = no chance of infection), and if x1 ≤ x2 then P(x1;τ) ≤ P(x2;τ) (an equal or greater magnitude of exposure should not decrease the probability of infection). The assumption of independent infection outcomes over multiple exposure events then leads to a simple expression for the probability of infection over a year follow-up period, given by Pr(M. tuberculosis infection) = 1 − E{E[1 − P(X;τ)]}^M, where E represents expectations taken over the assumed distributions for X and M. One specific form for P(X:τ) is given by 1 − (1 − τ)^X, which is motivated by assuming that each bacillus has an independent probability τ of forming a persistent infection at the site. Although this functional form is motivated by a specific biological model, it provides a broad range of shapes for P(X;τ) (Fig. 2A) and is reasonable to consider on that basis even without reference to the specific biological model. The impact of innate immunity, host genetics, vitamin D levels, or other factors that may influence the probability of an infectious quantum establishing a persistent focus of infection following deposition in the lung is contained within the assumptions of the model as an aggregate host risk factor. A more nuanced incorporation of such factors into the model is not feasible with currently available data.

In summary, our model linking exposure intensity and magnitude to infection posits M exposures per year with exposure intensities X, where the probability of infection from a given exposure depends on the intensity of that exposure X together with the parameter τ, which can be interpreted as the probability of infection given exposure to a single bacillus.

Suppose that the annual rate of M. tuberculosis infection in a given high-burden population is approximately 5%. Thus, for a given degree of intensity and magnitude of exposure, one can compute what the value of τ must be in order to match that known rate of infection. Simply put, if exposure intensity and/or magnitude is much higher than the observed infection rates, then the probability of infection per exposure to a single bacillus must be quite small. Conversely, if exposure is low relative to observed infection rates, almost every exposure must lead to infection. There is a threshold or exposure below which it is impossible to achieve a given infection rate even if exposure leads to infection with certainty. The result of this computation calibrates the model to a given annual infection rate given the assumed degree of exposure. Figure 2B displays contours of the value of τ calibrated to an annual infection rate of 5% across different combinations of intensity and magnitude of exposure (indexed values of μ_X and μ_M on the natural log scale). This figure shows one region in blue for which levels of exposure are too low to be consistent with an annual infection rate of 5%. The boundary of this region corresponds to the deterministic situation for which every exposure results in a persistent M. tuberculosis infection (τ = 1). Higher levels of exposure correspond to the stochastic situation for which each exposure does not inevitably result in persistent infection (τ < 1).

Modeling observed VE.

With the model calibrated to an annual infection rate of 5%, we can then explore how a reduction in the probability τ by prior vaccination would translate into observed vaccine efficacy (VE) and how variations in the intensity and magnitude of exposure might affect that vaccine efficacy parameter. We define the population-level vaccine efficacy as the percent reduction in the annual infection probability and compute it from the expression given above using the values of τ represented in Fig. 2A to calibrate the control group infection rate at 5% and using values of τ * RR to give the corresponding infection rate among vaccinees, where RR is the relative reduction in the probability τ due to prior vaccination. One minus RR might be thought of as “biological vaccine efficacy” at the level of a single exposing bacillus. Examples of the assumed shift in the function P(X;τ) due to prior vaccination (with 60% reduction in the probability τ) are shown in Fig. 2C.

Contours of values for vaccine efficacy (VE) plotted versus exposure intensity and magnitude are given in Fig. 2D for a 60% reduction in the probability τ (RR = 0.4). As expected, the value of population-level vaccine efficacy is attenuated relative to the “biological vaccine efficacy.” The magnitude of attenuation is greatest for lower levels of exposure at the threshold of a deterministic link between each exposure event and infection. Thus, if exposure is driving infection with little biological variation in risk subsequent to exposure (i.e., τ ≈ 1), then vaccine efficacy is more likely to be significantly attenuated. In addition, the degree of attenuation appears to be more sensitive to intensity rather than magnitude of exposure, with the combination of low intensity and high magnitude resulting in vaccine efficacy of nearly one-half the value of τ, while high intensity and low magnitude results in vaccine efficacy of only 22% less than the value of τ. The result of this simple modeling exercise supports the idea that variation in exposure would not dilute a biological effect of preexposure vaccination to a degree that population-level vaccine efficacy could not be reliably detected in a clinical trial.

Summary of modeling exercise and conclusions.

To recap in simple prose, suppose a vaccine has a “biological efficacy” of reducing the probability of infection per single unit exposure by 60%. The question is how much of a reduction in population rates of infection such a vaccine would produce given the multiplicity of exposures over time and variation in the magnitude of exposures (over different exposure events). For a given intensity and magnitude of exposure in a population, we can model the rate of infection as a function of infection probability from a single unit exposure. For a known infection rate in a specific population, e.g., 5%, we can then calibrate the model to compute the infection probability per single unit exposure that corresponds to the population infection rate under the specified intensity and magnitude of exposure. If the exposure is at very high levels relative to the population infection rate, then the infection probability per single unit exposure must be small. If exposures are at very low levels, then the infection probability per single unit exposure must be high, and at some point it must approach one, where every exposure results in infection. With this model calibrated to a population infection rate in an unvaccinated population, we can compute the population infection rate among vaccinees, assuming a given level of biological efficacy of the vaccine. From these population rates we can compute observed vaccine efficacy and compare it to biological vaccine efficacy to see how much of that biological efficacy is attenuated due to assumed exposure intensity and magnitude. What we find is that biological vaccine efficacy decreases when exposure magnitude and intensity decrease, with the lowest efficacy when approaching the threshold at which the probability of infection per single unit exposure approaches 1 (i.e., infection occurs after every exposure). However, provided that there is some stochasticity in infection per given exposure (i.e., exposure does not inevitably result in infection), then attenuation of biological vaccine efficacy is not that great and estimates of the observed vaccine efficacy in a vaccine trial may reasonably reflect the levels of biological efficacy of the vaccine.

Previous modeling studies suggest that the protective effects of a “leaky vaccine” are imperfect and are realized independently over multiple exposures (165) The overall protective effect of the vaccine declines, and measurable vaccine efficacy is much less than that realized in a population of individuals with few exposures over time. The contribution of the modeling work here is to calibrate the model to a fixed population-level rate of infection (e.g., 5%) and note that increased levels of exposure in this calibrated model naturally must be offset by a reduction in the absolute probability of infection per exposure. The novel finding of this work is that, once the model is calibrated, the attenuation of protective effects of a vaccine is increased when there is less exposure and the probability of infection per exposure is close to one. Thus, the general concern about protective effects of a TB vaccine being overwhelmed by multiple exposure events is perhaps misplaced. A second result of this modeling has to do with the relative impact of heterogeneity in magnitude of exposure compared to intensity of exposure over time. Again, fewer exposures that are each of a very high magnitude will erode the protective effects of a TB vaccine more than greater numbers of exposures that are each at a low magnitude.