Introduction

Tuberculosis (TB) remains a deadly disease with poorly controlled transmission in many impoverished populations1. Methods of detection that are more sensitive than usual tests have revealed an unexpectedly high prevalence of Mycobacterium tuberculosis (Mtb) bacilli in the respiratory bioaerosol of individuals spanning the TB disease spectrum2,3. Approximately two-thirds of apparently healthy individuals living in an informal settlement in South Africa were found to have viable Mtb detectable in collected respiratory bioaerosol samples. This is a novel finding resulting from the application of a customized bioaerosol sampling procedure4 and Mtb identification methodology5. Individuals with presumptive TB were found to have prevalence of 90% Mtb bioaerosol positivity regardless of whether a TB diagnosis was made microbiologically or clinically. In contrast to sputum, use of bioaerosol as a microbiological specimen allows access to the respiratory lining fluid in the lung periphery in disease-free states6. Cases of bioaerosol Mtb positivity amongst healthy, asymptomatic individuals were found both in those with and those without a positive interferon gamma release assay (IGRA). A significant proportion of individuals with an IGRA-negative/bioaerosol Mtb-positive status suggests false positive IGRA is not the explanation. Rather this independence from immunoreactivity, conventionally used to define TB infection, is suggestive of a state of Mtb carriage in the respiratory tract. This may be simple airway colonisation without mucosal invasion or co-existent airway colonisation with immunoreactivity and/or TB disease.

Longitudinal bioaerosol sampling indicates that this is a transient state with a rapid rate of loss of detectable Mtb bacilli from the respiratory lining fluid in the absence of treatment: the half-life was found to be 83 days as reported in our recent study2. We interpret this as a natural process of reversion from a state of Mtb carriage with a gradual reduction in bacillary load leading to clearance from the respiratory compartment that occurs frequently across the healthy population. This reversion rate combined with the prevalence of bioaerosol positivity in the same community allows estimation of a rate of exposure.

Immunological studies based on age-prevalence profiles of IgG antibody seropositivity have been used to study the epidemiological dynamics of viral, bacterial and parasitic infections7,8. Catalytic models can generate estimates of the force of infection (or sero-conversion) within the susceptible population using age-specific seroprevalence. In diseases where humoral responses wane over time, these models can be adapted to incorporate the rate of sero-reversion of IgG positivity. Here we apply a reverse catalytic model using a Bayesian approach to fit age-specific Mtb bioaerosol positivity data. The aim is to make an exploratory estimate of the rate of exposure to Mtb based on these prevalence data combined with reversion rates established from longitudinal sampling. This parameter is specifically related to exposure to aerosolized Mtb and distinct from typical force of infection measures based on IGRA. We challenged the strength of this estimate with sensitivity analyses of all the measured parameters.

Methods

Catalytic model

Catalytic models characterise the population dynamics of exposure to infection from birth and throughout life from cross-sectional prevalence data9. The annual force of infection, the rate at which susceptible individuals acquire an infection or reinfection, can be estimated from these models. By convention, TB ‘infection’ implies immunoreactivity (interferon gamma release assay [IGRA] or tuberculin skin test [TST] positivity) and so for clarity the term ‘rate of exposure’ will be used hereafter to refer to the rate at which the collected bioaerosol sample from an individual converts from undetectable to detectable Mtb.

Using a reverse catalytic model, we derived estimates of the rate of exposure based on empirically measured age-specific prevalence in healthy individuals and informed by empirical data indicating reversion rate. The model relates the prevalence, P, at age, a, to rate of exposure, λ, and rate of reversion, ρ (see Eq. 1 & Fig. 1). We assumed that respiratory bioaerosol from individuals at birth have no detectable Mtb, that all individuals are susceptible and that the rate of exposure remains constant through calendar time. We interpret loss of Mtb detection in exhaled aerosol samples as clearance of Mtb from the respiratory airway compartment. Also, conversion to aerosol Mtb positivity is assumed to be due to external exposure to airborne Mtb and not a fluctuating positivity based on cyclical underlying TB disease (unlikely for the vast majority of randomly selected individuals). For the primary analysis, the rate of exposure, λ, and rate of reversion, ρ are both assumed to be constant throughout life and homogenous across the community and consequently that the population reaches a steady prevalence.

$$P\left( a \right) = \frac{\lambda }{\lambda + \rho } \left[ {1 - e^{{ - \left( {\lambda + \rho } \right)a}} } \right]$$
(1)
Fig. 1
figure 1

Reverse catalytic model fitted to age-specific bioaerosol positivity prevalence data, P(a) with \(\lambda\) as the rate of exposure and \(\rho\) as the rate of reversion. The pre-factor \(\frac{\lambda }{\lambda +\rho }\) gives the saturating prevalence.

Full size image

The average age at which Mtb is first identified in an individual’s bioaerosol is given by 1/ \(\lambda\) and the average duration of Mtb persistence is 1/ \(\rho\).

Data sources

TB aerobiology study and sample collection

The Aerobiology and TB Research Unit in Cape Town, South Africa has recently published studies demonstrating the detection of Mtb in captured respiratory bioaerosol from a broad range of individuals spanning the TB disease spectrum2,3. In brief, the sampling methodology involved a modified respiratory aerosol sampling chamber (RASC) which consists of a HEPA-filtered enclosure accommodating a single individual (see Supplementary information). The seated participant places their head into a metallic elliptical cone through which a unidirectional airflow is created by a high-flow (300L per minute) cyclone collector connected at the cone apex. Respiratory bioaerosol is extracted into sterile phosphate-buffered saline (PBS) through inertial impaction. The high velocity airflow (12 ms−1) enables efficient collection even of aerosols generated during explosive respiratory activities such as coughing. A study nurse directs the participant to complete a 15-min sampling protocol comprising 5-min sampling for each of 15 forced vital capacity manoeuvers (FVCs), tidal breathing, and 15 voluntary coughs. Each sample collection was processed in the on-site laboratory. After centrifugation the pellet was resuspended in 200μL Middlebrook 7H9 medium and incubated overnight with the DMN-trehalose probe10. Washed samples were added to a nanowell device5 which was sealed and centrifuged before visualization by fluorescence microscopy. Experienced microscopists determined the presence of Mtb based on the morphology and fluorescence staining characteristics of the bacilli5. Each sample was assessed by two microscopists who were blinded to participant or empty chamber controls. Counts of visualized bacilli were made, however, for the purposes of this study results are interpreted in binary fashion i.e. any Mtb bacillus detected or none detected.

Respiratory bioaerosol Mtb prevalence data

Age-specific prevalence profiles of Mtb bioaerosol positivity were constructed from cross-sectional sampling data of healthy individuals from a highly TB burdened South African peri-urban community 40 km south-west of Cape Town collected between February 2022 and December 2022. Erfs (land parcels) were randomly selected from the community and all individuals living on the erf were offered entry into the study. 135 individuals underwent sampling with ages ranging from 14–72 years (mean 33 and standard deviation 13) and 64% female.

Respiratory bioaerosol Mtb clearance rate

In conjunction, data from a recently published study2 with longitudinal bioaerosol sampling of TB clinic attendees, collected between May 2020 and May 2022, at baseline and approximately 2-weeks, 2-months and 6-months from the same community were evaluated to determine rates of bioaerosol clearance. Those in the non-TB diagnosed group (“Group C” as described in the aforementioned study2) who did not receive treatment (n = 30) were aged between 15–67 (mean age 40 standard deviation 12) and 63% male. The clearance rate was indistinguishable from those on treatment and so the combined result was used to establish a reversion rate.

Statistical analysis

A Bayesian inference approach was used to fit the reverse catalytic model to the bioaerosol sampling data of healthy individuals using Markov chain Monte Carlo (MCMC) with the Gibbs sampling algorithm11. Estimates of parameter posterior distributions were informed by uniform prior distributions for \(\lambda\) and \(\rho\). Observational data from symptomatic individuals2 informed the prior for the rate of reversion, \(\rho\), given by:

$$\rho = {\text{ln }}\left( {2} \right)/{\text{half - life}}_{{({\text{years}})}}$$
(2)

Therefore, a clearance half-life of 83 days with a 95% confidence interval (CI) of 63–167 days gives a reversion rate of 3.0/year (95% CI 1.5–4.0). The prior for \(\rho\) was therefore chosen as a uniform distribution between 1.5 and 4.0. A relatively non-informative prior with a uniform distribution between 0 and 10 was used for \(\lambda .\) All analyses were performed using R version 4.3.1 and the models were implemented using the RJags package (version 4–14)12.

MCMC convergence was evaluated by the Gelman-Rubin statistic with a threshold of < 1.1 and the effective sample size (ESS), which is the estimated number of independent samples discounting autocorrelation generated by the MCMC run, with an ESS of > 200 used.

Sensitivity analyses

Bioaerosol prevalence

The data for this study are limited to a single community in Cape Town, South Africa and therefore we sought to examine the strength of our findings by conducting sensitivity analyses of all the key parameters. The South African TB incidence is estimated at 615 per 100,0007 and the population sampled is from a highly burdened region within South Africa. The main sensitivity analysis was therefore performed to account for settings where a lower bioaerosol positivity prevalence would be expected with data simulated to give mean prevalence of 0.1, 0.2, 0.3 and 0.4. These data were then fitted to the reverse catalytic model.

Age-varying rate of exposure

The baseline reverse catalytic model assumed a time-constant \(\lambda\) without variation by age. However, studies using cross-sectional surveys of TST from the same setting have shown an age-variable force of infection with a peak in the mid-teens and significantly lower in infancy13. It is plausible therefore that exposure leading to detectable Mtb in respiratory bioaerosol follows a similar pattern. We modelled the impact on estimates of \(\lambda\) by assuming a \(\lambda\) of zero from ages 1 to 10 and then a uniform prior between 0 and 10 for all ages above 10.

Long-term persistence of Mtb aerosol positivity

The rate of reversion was taken from sampling data from individuals who were symptomatic, it is conceivable therefore that loss of Mtb aerosol positivity is far less frequent or negligible in healthy (untreated) individuals. To examine this possibility a model without reversion (see Eq. 3) was fitted to the prevalence data with comparison to the primary analysis through visual inspection (comparison of model output with superimposed cross-sectional prevalence data) and the Watanabe-Akaike Information Criterion (WAIC)—a measure which balances model complexity with goodness of fit.

$$P\left( a \right) = 1 - e^{ - \lambda \cdot a}$$
(3)

Ethics statement

The Human Research Ethics Committee (HREC/REF: 529/2019) of the University of Cape Town approved the studies which generated the data that informed this study.

Results

The reverse catalytic model was fitted to healthy individual sampling data as shown in Table 1 and the Bayesian analysis generated posterior distributions with parameter estimates shown in Table 2. The rate of exposure was found to be a median of 5.1 (CrI: 2.5 to 8.5) TB exposures per year for every healthy individual throughout life. Assuming constancy throughout life including in infancy this implies the first exposure (1/ \(\lambda\)) occurs at 2.3 months (CrI 1.4 to 4.7) of age on average. The point estimate of the reversion rate from the posterior distribution was 3.0/year (CrI: 1.6 to 4.0) which is very similar to the highly informative prior, based on longitudinal observations. Since the reciprocal of the reversion rate can be interpreted as the duration of aerosol positivity, these data indicate that detectable Mtb persists in the bioaerosol 4.0 months (CrI: 3.0–7.4). Figure 2 shows the age-prevalence profile generated from these model outputs with empiric cross-sectional prevalence data superimposed.

Table 1 Age-specific cross-sectional data for prevalence of Mtb bacillary positivity in sampled respiratory bioaerosol of individuals living in a high TB burden setting.
Full size table
Table 2 Parameter estimates extracted from the posterior distribution of the rate of exposure and rate of reversion from the reverse catalytic model.
Full size table
Fig. 2
figure 2

Reverse catalytic model showing the proportion of bioaerosol positive healthy individuals by age. The dark blue line is the median prevalence estimate from the model and the blue shading shows model uncertainty with 95% credible intervals. The overlaid black points represent age-specific bioaerosol positive proportions with error bars derived from a binomial distribution using the exact method of Clopper and Pearson. Median point estimates of lambda (\(\lambda\)) and rho (\(\rho\)) are shown in the top-right corner.

Full size image

Sensitivity analyses

The above primary analysis estimated a high rate of exposure to account for a population bioaerosol Mtb positive prevalence plateauing at 0.65 in the context of the observed high rate of reversion. In this sensitivity analysis, we found point estimates for \(\lambda\) of 0.31/year (CrI: 0.13 to 0.61); 0.70/year (CrI: 0.33 to 1.2); 1.2/year (CrI: 0.57 to 2.0) and 2.0/year (CrI: 0.99 to 3.4) for simulated data with prevalence plateauing at 0.1, 0.2, 0.3, 0.4 respectively (see Fig. 3A–D). The same highly informative prior for \(\rho\) gave a posterior distribution with a median point estimate of 3.0/year for the reversion rate for all four simulations.

Fig. 3
figure 3

Age-prevalence profiles are shown for all panels with the dark blue line giving the prevalence estimate from the model and the shading showing the model uncertainty with 95% credible intervals. The overlaid black points represent age-specific bioaerosol positive proportions taken from the data with error bars (panels E & F) derived from a binomial distribution using the exact method of Clopper and Pearson. For each analysis median point estimates of lambda (\(\lambda\)) and rho (\(\rho\)) are shown in the top-right corner. Results for the MCMC model using simulated data with plateaux prevalence of 0.1, 0.2, 0.3 and 0.4 are shown in panels (A–D) respectively. Panel (E) shows age-variable rate of exposure, with \(\lambda\) set at 0 for ages 0–10. Panel (F) shows a catalytic model without reversion.

Full size image

A further sensitivity analysis assessed the impact of a reduction in rate of exposure in childhood by assuming a \(\lambda\) of zero until the age of 10. This led to no significant change in the estimate for rate of exposure with a median estimate of 5.1/year (CrI 2.5 to 8.5) (see Fig. 3E). Finally, the catalytic model without reversion is shown in Fig. 3F and gave a point estimate of the rate of exposure of 0.030/year (CrI 0.023 to 0.037). Visual inspection of this model shows that the empiric prevalence data is poorly predicted by this non-reversible catalytic model and, as shown in Table 3, removing the reversion worsens the model fit compared with the primary analysis based on the reversible catalytic model.

Table 3 The WAIC values for the catalytic and reverse catalytic models.
Full size table

Discussion

In this exploratory modelling study we show that for individuals living in a TB endemic community the rate of TB exposure leading to Mtb positivity of exhaled respiratory bioaerosol may be extremely high. The base model indicates that at least 5 exposures per year occur informed by data suggesting reversion at a frequency of 3 per year. Thus a cyclical pattern of aerosol positivity and clearance is suggested with a period of only a few months. The sensitivity analyses highlight that this estimate is highly sensitive to the community bioaerosol Mtb positive prevalence which is likely to be considerably lower in most parts of the world outside of the highly burdened countries. In other settings, where we envisage the bioaerosol Mtb positive prevalence is below 20%, it is therefore likely that the rate of exposure is much lower i.e. < 1 episode per year. The impact of varying the rate of exposure with age was also examined, in particular the probable scenario where infants and small children have a much lower exposure rate. Lowering the rate of exposure to zero before the age of 10 produced an overall estimate with the same high exposure rate.

The final sensitivity analysis explores the rate of bioaerosol Mtb clearance. In the primary analysis, the reversion rate identified from longitudinal sampling of presumptive TB cases arriving at clinic but remaining off treatment was used. On first presentation ~ 90% of these individuals were bioaerosol Mtb positive reverting to ~ 20% over 6 months (reversion rate of 3 per year). Since this (initially) symptomatic cohort may not be representative of the asymptomatic majority, we explored the hypothesis that Mtb persists lifelong in the general population by applying a non-reversible catalytic model (reversion rate of zero). The primary model was found to better fit the cross-sectional prevalence data available and so we infer that there is an underlying clearance of Mtb from bioaerosol. The precise rate of clearance and whether it differs from that in the symptomatic cohort is unclear in the absence of longitudinal sampling of asymptomatic, healthy individuals but may be important to investigate further. Indeed, we have hypothesised that those in the symptomatic cohort are undergoing a process of immune-driven self-clearance since the bioaerosol Mtb positivity prevalence falls far below that in the general population by the end of monitoring2. Exploration of the underlying mechanism(s) of bioaerosol Mtb clearance may be critical for the development of new ‘latency’ drugs, vaccine strategies and host directed therapies14.

It is notable that the estimated rate of exposure is two orders of magnitude greater than the force of infection (0.03–0.07/year) calculated from TST results in the same community15. This implies, that exposure, leading to detectable Mtb in respiratory bioaerosol samples, only infrequently leads to infection as conventionally defined. This finding is in keeping with the lack of correlation between Mtb bioaerosol positivity and QuantiFERON positivity found in the healthy community member sampling study3 and also consistent with the detection of Mtb complex DNA in peripheral blood mononuclear cells from 79% of asymptomatic TB contacts from Ethiopia half of whom had negative QuantiFERON results16. The explanation for those cases where Mtb exposure occurs without TST or IGRA conversion may be immunological control via innate immunity in the absence of priming antigen specific T cells17 or this may represent airway colonisation of an immunologically privileged site.

The source for these exposures is unclear but evidence is emerging for the transmission potential across the TB spectrum. Linkage of sputum culture positive cases via molecular epidemiological methods demonstrate transmission from sputum smear negative cases albeit at a lower rate18. Furthermore, a recent analysis of TB prevalence and household contact studies has suggested a high rate of transmission leading to onward infection from subclinical cases which may even be equivalent to transmission from those with clinical TB19,20.

The key limitation of this work is the paucity of data from other centres and settings to inform the model. There is no standard approach to measuring aerosolized Mtb and different methodologies will likely have different sensitivities which is a critical determinant to reproduce these findings in such low biomass bioaerosol specimens21. Moreover, other groups conducting aerosol sampling studies22,23 have not focussed on asymptomatic individuals and there are few longitudinal data to assess rate of acquisition or loss of Mtb bioaerosol positivity over time. There are also likely to be significant variations with the communities investigated and therefore the extremely high rate of exposure estimated from this highly TB endemic community is not generalisable to other lower prevalence settings. Finally, an alternative explanation for observed Mtb bioaerosol positivity may be intermittent Mtb shedding from individuals experiencing cycles of clearance and recrudescence of the same Mtb strain over many years. However, this is only likely to be true for a small proportion of community members with (subclinical) TB disease and therefore the rate of exposure to account for the high population prevalence of Mtb bioaerosol positivity would remain high.

Conclusion

This modelling study explores the implications that can be drawn from recent sensitive aerosol sampling investigations that identified highly prevalent Mtb bacillary colonization in the respiratory bioaerosol of individuals from highly burdened communities often in the absence of immunoreactivity. In these settings individuals are likely to experience frequent, intense exposures throughout life and seemingly often independent from the development of symptoms, clinical TB disease or measurable Mtb infection. Importantly these findings also suggest that full characterisation of transmission cycles may not be achieved through traditional markers of infection as these can significantly underestimate the degree of exposure.