Variability of Airborne Transmission of Mycobacterium tuberculosis: Implications for Control of Tuberculosis in the HIV Era

This is the authors' abstract. We don't add key points for this paper.

The recent collision of extensively drug-resistant (XDR) tuberculosis with HIV infection in South Africa [1] created an outbreak of tuberculosis with a 98% mortality rate and raised the specter of untreatable tuberculosis. Most of these cases of XDR tuberculosis were probably a result of nosocomial transmission, because two-thirds of the patients had been recently hospitalized prior to the onset of tuberculosis, and at least 2 health care workers died. A World Health Organization Global Task Force on XDR tuberculosis recommended several interventions, including accelerated implementation of infection-control measures to reduce transmission and strategies that promote prevention and control of XDR tuberculosis [2]. In this context, the article by Escombe et al. [3] in this issue of Clinical Infectious Diseases could not be more relevant. At first glance, it might seem that studies of guinea pigs regarding airborne transmission of pulmonary tuberculosis could offer nothing new. Less than 2 decades after Koch's elegant demonstration of Mycobacterium tuberculosis as the etiologic agent of tuberculosis in 1882 [4], Flugge [5] infected guinea pigs by having patients with tuberculosis cough on them. His logical but erroneous interpretation was that tuberculosis was transmitted by large respiratory droplets. Less than 50 years ago, Wells, his protégé Richard Riley, and their coinvestigators convinced the world that M. tuberculosis is transmitted by the airborne route. To do so, they developed a model of guinea pigs in a penthouse exposed to the air exhausted from an experimental tuberculosis ward [6], which was reproduced by Escombe et al. [3]. When I had the pleasure of meeting Richard Riley several years before his death, he thought that his work would probably never be reproduced because of its complexity and expense. Escombe and his team are to be congratulated for proving Riley wrong about this. What is new in this report by Escombe et al. [3]? Most obviously, they have reproduced this method in a resource-limited setting and applied it to the study of tuberculosis and HIV coinfection. Most importantly, they have confirmed the considerable variability of infectiousness of patients with tuberculosis and have done so for the first time among HIV-infected patients. This important concept remains generally unappreciated. Riley et al. [7] initially reported that only 3 (4%) of 77 (HIV-negative) patients produced >73% of the infections in the guinea pigs. In a second study [8], they found that only 8 (13%) of 61 patients transmitted all of the infections to the guinea pigs, with 50% of infections due to 1 patient with tuberculosis laryngitis. Epidemiological data from contact tracing of patients with tuberculosis from Rotterdam during 1967–1969 [9] are congruent with these animal data. Among the index case patients with positive smear results for tuberculosis, only 28% transmitted infection to their contacts, according to tuberculin skin test results [9]. Furthermore, there was variability based on age. Of the patients aged <40 years, 57% transmitted infection, whereas only 9% of those aged >60 years did so [9]. A study in the United States of 796 cases of tuberculosis had similar results, suggesting that younger patients were more infectious [10]. Overall, 491 (36.3%) of the 1352 contacts were infected, but there was a wide range of variability in transmission, with a strong tendency for either all or none of the household contacts to be infected [10]. A smaller US study [11] reported a more uniform range of variability of infection among household contacts, again ranging from 0% to 100%. Molecular epidemiology in New York City and San Francisco has suggested that ∼10% of patients transmitted most disease, although this may be controversial, depending on how clustering of DNA fingerprinting is interpreted [12]. In a molecular epidemiology study matching source cases to subsequent cases in The Netherlands, transmission of tuberculosis was again higher from younger source patients and also from male patients than from female patients [13]. Whether HIV-infected patients with tuberculosis are more or less infectious than HIV-uninfected patients with tuberculosis has been debated for years, with studies supporting both more and less transmission from HIV-infected patients, compared with HIV-uninfected patients [14]. Escombe et al. [3] suggest that the HIV-infected patients in their cohort were more infectious for tuberculosis than were the HIV-uninfected patients studied by Riley et al. [6–8, 15] during the 1950s. This conclusion by Escombe et al. [3] is based on their calculation of a mean of 8.3 airborne infectious units (termed “quanta” by Wells and Riley) per h for their HIV-infected cohort, whereas Riley et al. [6–8, 15] found a mean of 1.25 quanta per h. This conclusion should be viewed with caution, because methodological differences may account for these discrepant numbers, and no statistical analyses were performed. These results are especially surprising given that only 41 (42%) of the 97 patients with pulmonary tuberculosis in the study by Escombe at al. [3] had sputum specimens with positive results on microscopy for acid-fast bacilli, whereas all of the patients in the studies by Riley et al. [6–8, 15] had positive sputum smear results for acid-fast bacilli. A key difference in methods between the studies by Riley et al. [6–8, 15] and Escombe et al. [3] is the tuberculin skin test criterion to define infection. Riley et al. [6–8, 15] aimed to provide convincing data that tuberculosis is transmitted by the airborne route in an era when transmission was attributed to large respiratory droplets or contact. Thus, they sought a very specific criterion for infection in the guinea pigs to minimize false-positive test results; their criterion was a tuberculin skin test result of >10 mm of induration, as well as evidence of necrosis. The cutoff used by Escombe et al. [3] was ⩾7.5 mm of induration, with no mention of necrosis. The studies by Riley et al. [6–8, 15] were also performed before the standardization of international units of purified protein derivative; thus, it is difficult to know whether the purified protein derivatives used by the 2 groups were similar. A major strength of the work by Escombe et al. [3] was the rigor with which they evaluated this criterion, including the receiver operating characteristic curve analysis. Thus, their more sensitive criterion is probably more appropriate for defining infection, especially in the HIV era. Another potential difference is that Escombe et al. [3] used native guinea pigs from Peru, whereas those in the studies by Riley et al. [6–8, 15] were obtained from commercial laboratories. The most important limitation of the study by Escombe et al. [3] is the lack of data on individual patients with tuberculosis. We are unable to ascertain whether the increased numbers of infections in guinea pigs during months with very infectious air (e.g., month 10) were due to 1 very infectious patient or >1 patient. What was the variability of cough strength or cough frequency among the patients? Which patients had cavitary disease or multidrug-resistant tuberculosis? How long did they undergo appropriate antituberculous chemotherapy? Were the more infectious periods associated with patients infected with more transmissible or virulent bacillary strains? The authors suggest that genotyping of the strains in the guinea pigs and in the source cases may allow further investigations, but they do not explicitly state that such research is in progress. Another limitation is the lack of data on calibration or validation of this experimental system, as was performed by Riley et al. [15] using surrogate bioaerosols of Escherichia coli or Mycobacterium bovis or as could now be performed with a nonbiological aerosol or tracer gas. Such data may be inappropriately technical and voluminous for this journal. However, the publication of such data in an engineering journal may help to resolve discrepancies between the studies by Escombe et al. [3] and those by Riley et al. [6–8, 15] or studies planned for the Airborne Infection Research Facility in Witbank, South Africa (E. A. Nardell, personal communication). A general limitation to the use of the guinea pig model in the studies by Riley et al. [6–8, 15] is the need for a sophisticated ventilation system and laboratory, limiting such studies to specialized centers. Another limitation is the relatively poor resolution in time, because it would be very expensive and impractical to perform skin tests on and kill guinea pigs over brief periods of time, such as hours. This approach is also not designed to assess the spatial variability of exposures to infectious aerosols. Data from this model have been used to support the concept that humans are infected with only 1 infectious unit, but this may be an artifact of the design of the system that necessarily produces dilute aerosol exposures of the guinea pigs. We developed a cough aerosol sampling system using mechanical air sampling to quantify and size the culturable aerosols generated by patients coughing into a chamber that prevents dispersion [16]. The particle size distributions of these aerosols of M. tuberculosis are consistent with them being immediately respirable, as predicted by Wells' calculation that most infectious droplet nuclei desiccate in <1 s [17]. This suggests that persons close to a coughing patients with tuberculosis are probably exposed to larger aerosol concentrations than are persons across the room from them. This would increase the chance of inhalation and deposition not only of 1 infectious quantum (or unit), but also of multiple quanta. Our data have also shown variability in viable aerosol production that is consistent with the data on variability of infectiousness described above and by Escombe et al. [3]. However, a major advantage of the guinea pig model is the elimination of contamination by airborne molds that inevitably occurs with sampling air over longer periods of time [16]. Thus, it appears that our cough aerosol sampling system and the guinea pig model are complementary in several ways. The major advantages of the cough aerosol sampling system are short-term assessments of infectious aerosol production or “source strength” and of changes with interventions directed at individual patients. The major advantages of the guinea pig model are long-term assessments of infectious aerosol production and of changes with interventions directed at the environment. Despite the limitations mentioned above, the work by Escombe et al. [3] is an important contribution to the literature and to infection control of M. tuberculosis. Despite the laudable calls to action against XDR tuberculosis from the World Health Organization and others, there has been little attention paid to the need for more research on prevention of transmission. Airborne infection research requires intense multidisciplinary efforts, but it can substantially increase our understanding of the variability and control of airborne transmission of tuberculosis through the simultaneous study of source patients, the environment, bacillary strains, and susceptible hosts. The research infrastructure of Escombe et al. [3] could be used to study the efficacy of interventions proposed to decrease production of infectious aerosols from individual patients [18, 19] or the efficacy of environmental measures, such as UV germicidal irradiation or air ionization, as their preliminary data suggest are effective [20]. Escombe et al. [3, 20] could resolve the questions of differential infectiousness associated with HIV infection or with multidrug-resistant or XDR tuberculosis by using the same laboratory system to study subjects with and without these characteristics. Improved assessment of infectiousness could also improve the design and cost-effectiveness of vaccine clinical trials by reducing exposure misclassification biases [21, 22]. If only one-third of patients with tuberculosis are truly infectious, as the above data suggest, then improved identification of infectious patients could allow for more cost-effective allocation of resources, such as respiratory isolation in health care facilities or contact investigations in the community. Most importantly, these prevention measures can help to break the cycle of transmission, thus preventing others from contracting tuberculosis and moving the world toward elimination of this resilient foe. Potential conflicts of interest.K.P.F.: no conflicts.