Limited data suggest that measures to reduce tuberculosis transmission should be based on locations rather than on personal contacts. Molecular epidemiologic methods (analysis of IS6110 patterns, spoligotypes, variable numbers of tandem DNA repeats, and automated DNA sequence data) identified a cohort of 48 persons who were infected with progeny of the same Mycobacterium tuberculosis strain. Epidemiologic investigation documented that a large proportion of the patients were gay white human immunodeficiency virus-positive men. Most practiced barhopping, an activity that involved patronizing many bars in the same neighborhood each night. Few subjects were directly linked to more than 1 or 2 other persons by conventional investigation methods, which shows that the transmission dynamics were unusually complex compared with most previously described episodes of strain spread. The data support the concept that identification of locations where pathogen dissemination likely occurs may provide additional strategies for targeted tuberculosis control.
Molecular typing of Mycobacterium tuberculosis used in concert with traditional epidemiologic investigation techniques has dramatically improved our understanding of the patterns of strain transmission [1–7]. An important theme that has emerged from these studies is that despite diligent efforts, most case clusters are not recognized by traditional tuberculosis-control strategies such as contact investigation. A second important concept that has been suggested more recently on the basis of one study in central Los Angeles is that measures to reduce M. tuberculosis transmission should be based on locations rather than on personal contacts . This concept has widespread ramifications for the conduct of routine tuberculosis-control procedures if it is generally applicable.
We used several molecular characterization techniques to document the occurrence of a large cluster of 48 tuberculosis cases in Houston. The magnitude of this outbreak and relationships among the affected persons were not detected by conventional tuberculosis-control strategies.
Materials and Methods
This study was conducted as part of a comprehensive, population-based epidemiologic analysis of tuberculosis spread in Houston and Harris County, Texas. Since 1995, 93% of all reported persons with tuberculosis in Houston and Harris County have been enrolled in the study. All enrolled patients have been interviewed (see below), and all available M. tuberculosis isolates have been analyzed by IS6110 typing. During the present analysis, 1297 M. tuberculosis isolates were characterized by DNA typing strategies from 1584 culture-positive persons. There were 1994 cases of tuberculosis reported during the period covered. For simplicity, Houston and Harris County will be referred to as Houston, unless it is important to differentiate the two. The study population was composed of 48 patients with tuberculosis identified between September 1994 and May 1997. All 48 were culture positive for M. tuberculosis: 45 were reported as cases in Houston and 3 from neighboring counties.
Molecular microbiology studies
The 48 M tuberculosis isolates were characterized by several molecular microbiologic methods. The isolates were typed by an internationally standardized method with insertion sequence IS6110, an element that is highly polymorphic in copy number and position in the chromosome . A computerized database (Bioimage Whole Band Analysis, version 3.2; Bioimage, Ann Arbor, MI) was used to classify and retrieve the DNA patterns. The isolates were also assigned to 1 of 3 principal genetic groups of M. tuberculosis on the basis of nucleotide polymorphisms located in codon 463 of the gene (katG) encoding catalase-peroxidase and codon 95 of the gyrA gene encoding the A subunit of DNA gyrase . In addition, the spacer oligotype pattern (“spoligotype”) was determined for the organisms by the protocol described by Gronen et al. and used by several investigators [11–13]. Isolates were also studied for chromosomal polymorphism in loci containing variable numbers of tandem repeats (VNTR loci) . This recently described strategy is based on the observation that M. tuberculosis complex isolates contain ⩾11 VNTR loci. The combination of VNTR alleles, designated as a VNTR allele profile or multilocus chromosomal genotype, provides a polymerase chain reaction (PCR)-based strategy to subtype M. tuberculosis isolates. The VNTR allele profiles were determined for 5 VNTR loci, designated ETR-A through ETR-E. These 5 loci were used for analysis because they amplify reproducibly by PCR from a wide variety of isolates. Each digit in the allele profile represents the number of tandem repeats identified at a particular locus. Thus, the VNTR allele profile 3-2-3-3-3 indicates that 3 tandem repeat units occur at locus ETR-A, 2 at locus ETR-B, and so forth.
Data collection and questionnaire
Each patient who agreed to participate in an ongoing population-based survey of tuberculosis spread in Houston was interviewed. A questionnaire designed to acquire demographic and socioeconomic information was used. The interviewer also inquired about risk factors that may contribute to acquisition and spread of M. tuberculosis, such as place of employment, incarceration history, living situation, transportation habits, travel history, and social contacts. Questions were asked about the possible location and time of exposure to M. tuberculosis, alcohol and drug use, smoking history, and sexual orientation and behavior. Human immunodeficiency virus (HIV) status was initially determined by self-report and confirmed by medical record review for all patients. To avoid bias, the questionnaire and strain typing data were obtained in parallel. Participation in the study was voluntary, and the patients could refuse to answer any question or terminate the interview at their discretion.
Data were entered into a database and analyzed by computer (Epi Info version 6.04, CDC, Atlanta; version 6.12, SAS Institute, Cary, NC). Proportions were compared by x2 or two-tailed Fisher's exact tests as appropriate. Continuous variables were compared by one-way analysis of variance (ANOVA) or Krus-kal-Wallis ANOVA by ranks.
Molecular strain typing data
As of June 1997, DNA typing had been conducted by IS6110 profiling of 1660 M. tuberculosis isolates cultured from persons in Houston. One DNA profile group consisted of isolates from 46 subjects and was arbitrarily designated profile 004. In addition, 2 isolates had a similar DNA molecular type (designated profile 047) that differed from profile 004 by the addition of 1 hybridizing band (figure 1). Study of the katG codon 463 and gyr A codon 95 polymorphisms revealed that all profile 004 and 047 organisms were members of M. tuberculosis principal genetic group 2 .
To determine whether the strains from these persons also had the same VNTR allele profiles, 83 M. tuberculosis chromosomal samples were analyzed in a blinded fashion. These islates included a random sample of 23 IS6110 profile 004 organisms and 60 isolates chosen to represent a diverse array of IS6110 subtypes present in Houston patients. Twenty-two VNTR allele profiles were identified among the 83 samples analyzed. Virtually all (22/23) case-cluster organisms had the same VNTR allele profile designated 3-2-3-3-3. This allele profile is the most common one identified in 437 M. tuberculosis isolates studied to date, accounting for 16% of all organisms. The 3-2-3-3-3 VNTR allele profile is distributed worldwide (unpublished data). Hence, although the VNTR data were consistent with the idea that the 46 strains with IS6110 pattern 004 and 2 strains with pattern 047 shared a recent ancestor, this VNTR pattern was clearly not specific for the strains with these 2IS6110 types.
The spoligotyping data also confirmed that the profile 004 and 047 isolates were related. Among a random sample of 16 IS6110 profile 004 and profile 047 organisms analyzed, 15 had the same spoligotype, arbitrarily designated S3. One isolate had spoligotype S5, which differs from the S3 pattern by 3 hybridization signals (figure 1). However, similar to the VNTR typing data, the S3 spoligotype pattern was identified in organisms with 29 distinct IS6110 profiles that were not profile 004 or profile 047.
Demographic and epidemiologic characteristics of the patient population are shown in table 1. Forty-three patients were male (89.6%); 72.9% were white (n = 35). Thirty-five (72.9%) were unemployed when identified as a case, and all 48 resided in a house or an apartment. Strikingly, more than two-thirds of the patients were HIV positive, including 31 men and 2 women. Twenty-eight (65.1%) of the male patients reported exclusively homosexual behavior, 7 (16.3%) bisexual behavior, and 8 (18.6%) exclusively heterosexual behavior. Five patients were female (3 heterosexual, 2 bisexual, on the basis of self-reported information).
An important observation obtained from analysis of the questionnaire data was that virtually all men (40 [93%] of 43) and 3 of 5 women were regular customers of 17 small bars located in one postal code area. Further investigation found that 15 (88%) of these 17 bars had predominantly gay patrons. These 17 bars are located within walking distance of each other and functioned as social gathering points. The close proximity of the 17 bars facilitated the ability of the patrons to regularly barhop among many of the bars on the same day. For example, the 43 regular bar customers reported patronizing 3.5 bars, on average (median, 3.0 bars). Most of these patients spent an average of several hours daily in the bars and some persons patronized the bars more than 10 hours several days each week. One patient who lived ∼160 km from Houston spent the entire day at one bar 3 or 4 days each week. Most patients reported that barhopping was used predominantly to socialize with friends, meet new people, and enjoy live entertainment. Some patients also reported that they exchanged sexual favors and engaged in commercial sex activities in the bars. In addition, the investigation revealed that 15 of the 48 patients had been employed by 1 or more of the 17 bars as a bartender, bartender assistant, bar stocker, or entertainer.
Patients not directly linked to the bars
Five of the 48 patients infected with the clonally related strains did not report a history of barhopping. One was a nurse at a local hospital who cared for HIV-positive patients with M. tuberculosis infection. We were unable to link this patient directly to any of the other 47 tuberculosis patients in the study cohort, but an exhaustive search of hospital records was not conducted. One patient denied patronizing any of the 17 bars and believed he was exposed to M. tuberculosis while in a Houston jail. It is noteworthy that 34 (72%) of 47 other patients reported a history of incarceration, including all the women and 67% of the men. One gay man frequently went to bars, but not in the same area of Houston as the 17 bars where the other patients in the cluster were customers. One patient was a delivery person who consistently was a customer at a large number of these bars, which were located on his delivery route. The fifth patient not directly linked to the core group of bar patrons was an HIV-positive sex worker in the neighborhood where the 17 bars were located. This patient used illegal drugs extensively and exchanged sex for drugs. Thirty percent of the patients in this cohort also acknowledged regular exchange of sex for money. Of note, 2 of the patients not directly linked to the barhopping group obtained and used drugs at the same house. These 2 persons reported that many other people also used this residence for drug activity.
Another important epidemiologic characteristic of this large cohort was substantial alcohol and drug use: 26 (54%) of the 48 patients consumed >12 alcoholic drinks each week. Moreover, 36 of the 48 used illegal drugs, and 18 of these 36 used injection drugs (table 1).
To facilitate the descriptive analysis of the data, we conducted a matched case-control study using patients from the DNA profile 004 cohort and tuberculosis patients randomly drawn from the general Houston population. The comparison population consisted of persons with culture-proven M. tuberculosis matched for age and sex at a ratio of 3 : 1. The analysis identified a distinct subpopulation of persons who were patrons of these 17 bars, who were HIV positive, white, and had a history of drug use, including injection drugs (table 2).
Drug resistance acquisition
One of the important themes in some tuberculosis case clusters is the acquisition of drug resistance by ⩾1 of the linked strains, generally due to non-adherence to antimicrobial therapy regimens. Five of the 48 strains cultured from the patients were resistant to ⩾l anti-tuberculosis medications [15, 16]. Three patients had rifampin-resistant organisms. Sequence analysis of the rifampin-resistance-determining region of the rpo B gene encoding the β subunit of RNA polymerase revealed that each organism had a distinct mutation. This means that rifampin resistance had evolved independently in these 3 isolates. Similarly, 2 patients had isoniazid-resistant organisms. Sequence analysis of the kat G gene encoding the catalase-peroxidase enzyme involved in activation of isoniazid identified distinct mutations in each isolate. This result indicates that isoniazid resistance was also acquired independently by the 2 organisms. Three of the 5 persons with drug-resistant organisms were known to be non-adherent with their drug therapy regimen. The exact timing of culture acquisition relative to drug therapy was not known.
Geographic spread of the case clone
A second theme that has emerged in studies of tuberculosis transmission is geographic spread of the case clone, sometimes over large distances, including interstate and intercontinental dissemination [17–19]. After the analysis of the 48-patient cohort was complete, we learned through our ongoing comprehensive epidemiologic investigation that a tuberculosis patient in California possibly was linked to this case cluster. Acquisition and subsequent molecular analysis of the organism cultured from the California patient revealed that the isolate also had IS6110 profile 004 and spoligotype S3. It is unknown whether this patient spread tuberculosis to others in California. Similarly, a patient in a north Texas city was linked to the Houston case cluster by molecular and conventional epidemiologic studies. In addition, 1 patient moved to Florida and 1 to Mississippi, but it is not known whether these persons subsequently transmitted M. tuberculosis to others.
Contact investigation and patient follow-up
Although there were 48 patients in the cohort, in aggregate, these persons identified only 215 total contacts (mean, 4.5). Nine persons (19%) did not identify a single contact and, importantly, only 4 patients identified another patient in the cohort during contact investigation. If the 5 patients with solely extrapulmonary tuberculosis are excluded, then 7 persons (16%) did not identify a contact. There was a mean of 5.7 contacts per case for the 43 patients with pulmonary or extrapulmonary plus pulmonary tuberculosis
On the basis of medical chart review, or interviews with 39 patients who could be located, we found that 17 (44%) had died as of November 1998, a mean follow-up of 23 months. Of the 17 who died, 16 were HIV positive, whereas only 13 of the 22 who remained alive were HIV positive (P = .02).
Our data add to studies documenting that molecular genetic analysis can identify clonally related strains constituting a potential tuberculosis outbreak in the absence of prior knowledge of patient linkage by conventional contact tracing. Importantly, the findings add to the theme that the magnitude of many large tuberculosis case clusters is not detected by standard tuberculosis-control contact-tracing practices, regardless of thoroughness. The analysis also adds strong circumstantial support to the concept that tuberculosis-control efforts based on locations where transmission probably occurs, rather than on contact tracing alone, may be a useful strategy to limit pathogen dissemination.
Several molecular strategies are available for subtyping M. tuberculosis isolates, including IS6710-based restriction fragment length polymorphism (RFLP) analysis, pTBN12-based RFLP analysis, spoligotyping, VNTR loci analysis, and automated DNA sequencing of polymorphic chromosomal genes and regions [9–11, 14, 20]. Because of the relatively large number of persons studied and the availability of detailed epidemiologic information linking the 48 patients, this case cluster provided an opportunity to examine the utility of spoligotyping and VNTR loci analysis to assess clonal relationships among strains. Virtually all strains studied with the IS6110 pattern 004 had the same spoligotype (S3). However, the S3 spoligotype was also identified among many other strains that have IS6110 types with no apparent close relationship to profile 004, indicating a lack of specificity in the spoligotype results. Similarly, the VNTR allele profile characterizing the case-clone isolates has been identified in many unrelated strains distributed globally, and hence is not specific for the case clone. Taken together, the data suggest that identification of clonally related strains should rely on results of IS6710-based RFLP analysis in combination with data from one or more additional subtyping methods. Our experience in this study and the results of other analyses [21–26] indicate that IS6110-based RFLP analysis and spoligotyping provide reasonably robust data to accurately infer epidemiologic relationships among patients. Inasmuch as ours is the first study to compare results obtained by all 3 methods, the limited data did not permit us to determine the relative discriminative ability of VNTR analysis and spoligotyping. Because polymorphism in VNTR loci and the direct repeat region accumulates independently, and perhaps at different rates, it will be important to study this issue in large samples of isolates.
Although our analysis showed that virtually all strains of M. tuberculosis in this case cluster were identical or nearly so, 2 types of molecular variation were detected among the organisms. Two organisms had an IS6110 RFLP profile that differed by 1 hybridizing band from the common 6-copy pattern identified in the other 46 isolates. Gain or loss of hybridizing copies of IS6110 among isolates recovered from patients in case clusters has been documented by several investigators [15, 27, 28]. Bifani et al.  showed a gain of 1 or 2 IS6110 hybridizing bands among several clonally related variants arising from the multidrug-resistant strain “W” responsible for roughly 300 cases of tuberculosis cases in New York City and elsewhere. A second type of molecular evolution that we detected among case-clone organisms was the independent acquisition of mutations conferring resistance to rifampin and isoniazid. Five patients had profile 004 strains resistant to one or both of these drugs. Automated DNA sequencing of the rpoB and kat G genes identified distinct resistance-conferring mutations in each strain, thereby verifying independent resistance acquisition, rather than the spread of a resistant organism between patients. Several patients with drug-resistant organisms were known to be nonadherent with their antituberculosis therapy, and 1 of the patients was institutionalized for treatment. Hence, although the organisms responsible for the case cluster are clonally derived, molecular variation accumulated in the progeny in the relatively brief time span studied. This observation stresses the importance of linking the molecular data with conventional epidemiologic information.
It is noteworthy that 3 episodes of laboratory cross-contamination of samples were discovered when patients presumed to be infected by profile 004 organisms were found to deviate significantly from the common epidemiologic profile shown by our analysis. This stimulated an investigation that revealed that the specimens had been processed on the same day as a smear-positive sputum sample collected from a person with pulmonary tuberculosis.
Although many molecular epidemiology studies have documented patient-to-patient transmission by IS6110 RFLP analysis and other molecular techniques, relatively little information has been published bearing on the natural history of a large cohort of tuberculosis patients infected with clonally related organisms [29, 30]. The high mortality rate that we identified in this cohort is probably attributable to the preponderance of HIV-infected persons in the group. However, it is also possible that the case clone is unusually virulent. In contrast to many other bacterial pathogens , little information is available bearing on biomedically relevant differences among strains. Sreevatsan et al.  analyzed 6000 M. tuberculosis isolates from persons in the New York City area and Houston and found that, on average, principal genetic group 1 and 2 organisms, but not group 3 organisms, were commonly associated with large case clusters. The investigators proposed that group 1 and 2 organisms had a greater propensity to spread than genetic group 3 M. tuberculosis isolates. Valway et al.  reported that a principal genetic group 2 strain (CSU 93 or CDC 1551) was extensively spread among factory coworkers by a person with cavitary pulmonary tuberculosis. Studies with C57BL/6 mice inoculated by the aerosol route indicated that CDC 1551 strain was unusually virulent, as indexed by enhanced rate and extent of growth in the lungs, compared with M. tuberculosis strain Erdman. Of interest, the mouse virulence of a randomly chosen profile 004 organism equaled or exceeded that of strain CDC 1551 (unpublished data). However, we currently favor the hypothesis that HIV status is a more important factor in determining mortality in the patient cohort than the relative virulence of this strain in mice.
Analysis of this large case cluster has made several contributions to the understanding of the spread of clonally related progeny of M. tuberculosis in a large metropolitan area. Traditional contact investigation alone detected epidemiologic links among only 4 of the 48 patients studied. Thus, the findings confirm the important theme that the extent of a large case cluster may go undetected by conventional tuberculosis-control contact-investigation practices. Barnes et al.  observed that traditional contact investigation is frequently unrewarding in urban areas because many of the patients are homeless, have unstable families or lifestyles, and cannot identify their contacts by name. As a consequence, it has been advocated that rather than conventional contact investigation alone, effective tuberculosis control in urban areas should consider location-based control efforts. A similar conclusion was reached recently by investigators who studied the dynamics of tuberculosis spread in Baltimore . Our findings mirror these reports from investigators working in other cities. Taken together, our data support the contention that under some circumstances, effective tuberculosis control may depend on implementation of non-traditional approaches such as location-based analyses, study of social networks, and enhanced modeling efforts [34–37]. Diligent contact investigation should remain an important mainstay of tuberculosis-control programs.
We thank the extremely dedicated staff of the Houston Tuberculosis Initiative, including Thanh Tung Bui, Pandora Davis, Olga Escobedo, Selina Gonzalez, Lorretta Jackson, Deborah Meyer, Tony Prejean, and Mary Rocha. Xi Pan, Subeeh Siddiqui, Alison J. Cobb, and Percy L. Strickland provided technical assistance; and Donna Zawada assisted in manuscript preparation. We thank Jeffrey Starke for ongoing intellectual stimulation; Peter F. Barnes for critical reading of the manuscript; and Kathy Penrose, John Langston, Cathy Barnett, and Hoa Quach for support and assistance.