INTRODUCTION
Despite the relative genetic conservation among
Mycobacterium tuberculosis strains compared with many other bacterial species (
91), numerous approaches for distinguishing mycobacterial isolates have been successfully developed. Early methods based on phage typing were capable of making only crude distinctions between mycobacterial strains. More recent developments (e.g., IS
6110 restriction fragment polymorphism [RFLP] analysis, spoligotyping, and mycobacterial interspersed repetitive unit–variable-number tandem repeat [MIRU-VNTR] analysis), each based on the detection of differences in the numbers or locations of particular genetic sequences or signatures, permit differentiation with much greater resolution (
63). In addition to the obvious public health applications of these tools for the identification of clusters of recent transmission, studies using molecular epidemiological methods have improved our understanding of the natural history of disease and how tuberculosis (TB) spreads in communities (
72,
105).
Early molecular epidemiological studies conducted in settings where the incidence of tuberculosis was low found an unexpectedly high proportion of disease that was attributable to recent transmission (
1,
89). These findings implied that control efforts that focus on reducing the transmission of disease might have a more important role than previously appreciated (
93), even in settings where the incidence of tuberculosis is low. Similarly, studies in settings where the incidence of tuberculosis is high documented that the majority of cases occur as a consequence of recent transmission (
103) and that many of these transmission events are taking place outside households (
108). Similar studies have also illuminated the importance of nosocomial (
34,
39,
69,
79,
81,
83), penitentiary (
17,
102), and casual (
54,
71) social settings in disease spread. These findings further underscore the epidemiological importance of diagnostic delays (due to both patient health-seeking and health care provider behaviors) and suggest that more rapid diagnosis and initiation of appropriate drug treatment can play an enormous role in interrupting transmission.
Molecular epidemiological studies also identified individuals with repeated episodes of disease due to reinfection, thus revealing the incomplete immunizing effect of previous
M. tuberculosis infection (
18,
107). These data confirm previously reported observational data (
15,
45) and statistical arguments (
4,
13,
96,
109) suggesting that previous exposure or disease does not confer complete protection against subsequent reinfection and have resulted in a paradigm shift in our understanding of the epidemiology of tuberculosis (
62).
While reinfection is expected to occur most frequently in settings where the incidence of TB is high (since multiple exposures are most likely) (
80,
101,
111), studies in some moderate- and low-incidence settings among individuals with repeated episodes of disease have found that reinfection also occurs at unexpectedly high rates (
14). Several explanations may account for the reinfection documented among those with repeated episodes of tuberculosis disease in low-incidence areas. Previously diseased individuals may have particular biological characteristics (genetic or acquired immunodeficiency) or acquired or environmental risk factors (alcoholism or nutritional deficiencies) that render them more prone to the development of disease after (re)infection. It is also possible that in low-incidence settings, the risk of exposure is concentrated within “sociospatial pockets,” such that individuals infected once are likely to be infected multiple times (
19). In other models that account for superinfection, because each infection is not immunizing, reinfection plays a dominant role in disease dynamics when the disease prevalence increases past a threshold (
44).
The fact that previous exposure provides incomplete protection from subsequent infection may have important implications for the implementation of interventions (e.g., treatment of latent infection) (
65) and for the prospect of developing effective preventive or therapeutic vaccines (
44). However, the overall importance of reinfection within the context of the tuberculosis epidemic has been questioned (
56). Ongoing doubt concerning the epidemiological importance of reinfection may reflect the fact that clear documentation of this phenomenon usually requires the collection of different strains from two disease episodes within an individual. This approach underestimates the frequency of reinfection since it cannot detect individuals who were repeatedly infected prior to developing a single episode of disease and will also misclassify reinfections as relapses when individuals are repeatedly infected with similar strains that are predominant in their community or social circle. Previously reported models suggested that a substantial fraction of first-time disease episodes is likely to be the result of reinfection and that this fraction is dependent on both the local prevalence of disease, which determines the force of infection that each individual experiences (
19,
65), as well as the degree of protection afforded by a previous infection (
44).
More recently, molecular studies that documented the phenomenon of polyclonal or “mixed” infections, in which multiple strains of
M. tuberculosis are retrieved from an individual host at a single point in time, underscored the potential importance of reinfection in tuberculosis epidemics (
Table 1). This phenomenon of mixed infections is not unique to
M. tuberculosis and appears to occur with pathogens of all types (
6). At present, accounts of mixed
M. tuberculosis infections are accumulating for various geographic settings, but there have been few efforts to examine the public health implications of the fact that a single host can simultaneously harbor more than one distinct strain of
M. tuberculosis. In this paper, we review existing evidence of mixed
M. tuberculosis infections collected from many different epidemiological settings. We highlight why these studies, while very important for documenting that these types of infections occur in all settings, may greatly underestimate the frequency of mixed infections due to the limited sensitivity of current approaches for the detection of the presence of mixed infections. We review evidence of the importance of mixed infections for treatment outcomes of tuberculosis patients receiving standard regimens of combination therapy and discuss how mathematical models reveal the importance of mixed infections to the projected trajectory of tuberculosis epidemics and the effects of interventions. Throughout our review, we aim to identify areas where additional research is needed to clarify how mixed
M. tuberculosis infections modify disease dynamics and control.
CONCLUSIONS
Previous reviewers have questioned the overall importance of mixed infections for individuals with tuberculosis or for the dynamics of tuberculosis within communities (
10). In this paper, we present an overview of studies that documented mixed-strain infections, enumerate challenges for the detection of mixed infections, and discuss the individual-level and potential population-level implications of this phenomenon.
The low sensitivity of most current approaches for the detection of mixed infections constitutes a serious obstacle to our ability to understand their importance. Our inability to reliably document existing mixed infections stems from challenges with specimen collection (we may not retrieve all strains when obtaining our clinical sample), specimen handling (minority variants present in the clinical sample may be lost during transit, decontamination, culture, or sampling), and typing (the strain typing method may not be sensitive enough to pick up rare variants or may not have adequate discriminatory power).
Despite these serious limitations, investigators have documented that a substantial minority of tuberculosis patients (10 to 20%) in certain settings are infected with multiple strains; this strongly suggests that mixed infections are very common. Given existing, though admittedly limited, evidence of the effect of mixed infections on treatment outcomes (particularly when strains with different drug resistance phenotypes are involved) and the model-based arguments for the importance of mixed infections in tuberculosis dynamics and control, we believe that mixed infections are an important subject for future research.
Proximal practical questions of interest include assessing which methods of specimen collection and processing are most sensitive for documenting mixed infections. Extensions of current collection methods, including the collection of 24-h sputum samples or the collection of strains from multiple anatomical sites, may result in clinical samples that better reflect the within-host population structure of mycobacteria. Decontamination and culture steps, while necessary prerequisites for many of the current typing methods, may also erode test sensitivity, and new methods that can type strains directly from sputum samples will be valuable. Deep sequencing, which we expect to be used routinely for the detection of mixed infections in the near future, is likely to displace many other methods and should generate better data to understand many aspects of within-host diversity (
37). New animal studies of mixed infection can identify whether certain strains are more capable of superinfection than others and clarify whether previous infection shapes the host immune response in such a way that makes reinfection easier (
47). The relevant scale of strain interactions for both individual- and population-level consequences of mixed infections is not known; direct competition between individual organisms likely arises by local interactions within a single granuloma (
24), whereas more indirect forms of competition may occur as a result of interactions between strains confined to different granulomas or organs (
33). Studies of these types of interactions would allow for a more complete understanding of strain competition and even the potential for strain synergy.
Despite the limitations of current methods of detection and unanswered questions about effects, the current data suggest that mixed infections are likely to have an important impact on accurate disease diagnosis, the effective treatment of individuals, and the control of tuberculosis in populations. New studies aiming to measure the prevalence of mixed infections that use methods of specimen collection and handling that are likely to sample and protect minority strains in combination with the most sensitive typing methods are needed to provide better estimates of how common these mixed infections are. Laboratory, animal model, and epidemiological studies that focus on the within-host complexity of
M. tuberculosis infection and disease can contribute complementary information that improves our ability to effectively combat this pathogen. For example, an understanding of the strengths and mechanisms of within-host strain interactions will allow us to better project the trajectories of tuberculosis epidemics and the expected diversity of pathogen populations (
8,
23,
85), understand the effects of the scaling up of existing interventions (e.g., IPT) (
21), and evaluate the potential effectiveness of novel vaccines (
20,
64).