The Infectiousness of Tuberculosis Patients Coinfected with HIV

Background The current understanding of airborne tuberculosis (TB) transmission is based on classic 1950s studies in which guinea pigs were exposed to air from a tuberculosis ward. Recently we recreated this model in Lima, Perú, and in this paper we report the use of molecular fingerprinting to investigate patient infectiousness in the current era of HIV infection and multidrug-resistant (MDR) TB. Methods and Findings All air from a mechanically ventilated negative-pressure HIV-TB ward was exhausted over guinea pigs housed in an airborne transmission study facility on the roof. Animals had monthly tuberculin skin tests, and positive reactors were removed for autopsy and organ culture for M. tuberculosis. Temporal exposure patterns, drug susceptibility testing, and DNA fingerprinting of patient and animal TB strains defined infectious TB patients. Relative patient infectiousness was calculated using the Wells-Riley model of airborne infection. Over 505 study days there were 118 ward admissions of 97 HIV-positive pulmonary TB patients. Of 292 exposed guinea pigs, 144 had evidence of TB disease; a further 30 were tuberculin skin test positive only. There was marked variability in patient infectiousness; only 8.5% of 118 ward admissions by TB patients were shown by DNA fingerprinting to have caused 98% of the 125 characterised cases of secondary animal TB. 90% of TB transmission occurred from inadequately treated MDR TB patients. Three highly infectious MDR TB patients produced 226, 52, and 40 airborne infectious units (quanta) per hour. Conclusions A small number of inadequately treated MDR TB patients coinfected with HIV were responsible for almost all TB transmission, and some patients were highly infectious. This result highlights the importance of rapid TB drug-susceptibility testing to allow prompt initiation of effective treatment, and environmental control measures to reduce ongoing TB transmission in crowded health care settings. TB infection control must be prioritized in order to prevent health care facilities from disseminating the drug-resistant TB that they are attempting to treat.


A B S T R A C T Background
The current understanding of airborne tuberculosis (TB) transmission is based on classic 1950s studies in which guinea pigs were exposed to air from a tuberculosis ward. Recently we recreated this model in Lima, Perú , and in this paper we report the use of molecular fingerprinting to investigate patient infectiousness in the current era of HIV infection and multidrug-resistant (MDR) TB.

Methods and Findings
All air from a mechanically ventilated negative-pressure HIV-TB ward was exhausted over guinea pigs housed in an airborne transmission study facility on the roof. Animals had monthly tuberculin skin tests, and positive reactors were removed for autopsy and organ culture for M. tuberculosis. Temporal exposure patterns, drug susceptibility testing, and DNA fingerprinting of patient and animal TB strains defined infectious TB patients. Relative patient infectiousness was calculated using the Wells-Riley model of airborne infection. Over 505 study days there were 118 ward admissions of 97 HIV-positive pulmonary TB patients. Of 292 exposed guinea pigs, 144 had evidence of TB disease; a further 30 were tuberculin skin test positive only. There was marked variability in patient infectiousness; only 8.5% of 118 ward admissions by TB patients were shown by DNA fingerprinting to have caused 98% of the 125 characterised cases of secondary animal TB. 90% of TB transmission occurred from inadequately treated MDR TB patients. Three highly infectious MDR TB patients produced 226, 52, and 40 airborne infectious units (quanta) per hour.

Conclusions
A small number of inadequately treated MDR TB patients coinfected with HIV were responsible for almost all TB transmission, and some patients were highly infectious. This result highlights the importance of rapid TB drug-susceptibility testing to allow prompt initiation of effective treatment, and environmental control measures to reduce ongoing TB transmission in crowded health care settings. TB infection control must be prioritized in order to prevent health care facilities from disseminating the drug-resistant TB that they are attempting to treat.

Introduction
Seminal experiments demonstrating airborne tuberculosis (TB) transmission by droplet nuclei were performed by Riley and coworkers in the 1950s-1960s [1,2]. Guinea pigs acquired TB by breathing exhaust air from a TB ward. The studies demonstrated TB transmission from a minority of patients, marked variability in patient infectiousness, and reduced infectiousness following initiation of effective chemotherapy [1][2][3]. These classic studies were recently recreated in Lima, Perú , in the modern era of HIV infection and multidrugresistant (MDR) TB, and again showed striking variability in patient infectiousness [4].
The strongest predictor of TB patient infectiousness is sputum smear status [5][6][7]. However, considerable variation exists in TB prevalence amongst contacts of smear-positive patients [5], and the importance of smear-negative transmission has also been demonstrated [8]. Other determinants of infectiousness include lung cavitation and cough frequency [9], and cough-inducing procedures have been associated with extensive transmission [10][11][12]. Additional factors are likely and may include sputum volume or consistency, and TB strain variables such as ability to survive in the airborne state. Recent studies detected culture-positive cough-generated aerosols in only 25% of new TB patients [13].
The relative infectiousness of MDR versus drug-susceptible TB remains controversial. The patients with drug-susceptible TB studied by Riley were four to eight times more likely to infect guinea pigs than were those with drug-resistant disease [1][2][3]. Epidemiological studies of contacts of both drugresistant and drug-susceptible TB infected patients have shown no difference in relative transmissibility [14][15][16], whilst studies utilising molecular epidemiology have had conflicting results [17][18][19][20]. Similarly, the effect of HIV infection on TB infectiousness remains disputed [21]. The beneficial effect of anti-TB chemotherapy on reducing TB infectiousness is well known, but this presupposes that the treatment administered is effective. The inadequate treatment of MDR TB and the emergence of extensively drug-resistant (XDR) TB therefore have important consequences for hospital TB infection control policies.
In this paper we report the results of molecular fingerprinting to determine which patients had infected which guinea pigs in our in vivo air sampling model above a TB ward. We investigate factors associated with TB transmission from this group of HIV-coinfected TB patients with a high prevalence of MDR TB.

Methods Setting
An airborne infection study facility was constructed on the roof of a TB-HIV ward at Hospital Nacional Dos de Mayo, Lima, Perú , as previously described [4]. All air from two 4bedded negative-pressure patient rooms passed over guinea pigs in the study facility before being exhausted into the atmosphere. Airflow from the ward and into the animal facility was measured using a capture hood (Alnor, Shoreview, United States) at air injection and extraction vents.

Patients
The ward operated as the hospital TB-HIV service, and patient admission, management, and duration of stay were not influenced by the study. All patients were invited to join the study through written consent. An admission questionnaire recorded daily symptoms including cough, haemoptysis, and fever. Twenty-four-hour sputum collections were made daily for auramine microscopy [22] and TB culture using MODS [23]. For nonconsenting patients, anonymised unlinked information routinely available for ward infection control, including sputum microscopy and treatment, was recorded.

Experimental Protocol
Outbred Peruvian guinea pigs were maintained in quarantine for 1-3 mo. Animals were skin tested monthly as described [4] and induration measured 48 h later. Animals were transferred to the hospital following at least two negative skin tests to ensure freedom from TB. In May 2002, 144 animals began ward air exposure, and 6 mo later, 148 animals were added. Monthly skin tests continued throughout 505 d exposure to ward air, and positive reactors were removed for humane killing and autopsy. Evidence of TB infection was sought in lungs, mediastinal lymph nodes, spleen, and liver. Tissues were homogenised and cultured for TB as described [4]. Forty negative control guinea pigs were maintained separately, breathing fresh air.

Determination of Patient Infectiousness
TB drug-susceptibility testing of animal and patient strains for susceptibility to isoniazid, rifampicin, streptomycin, ethambutol, capreomycin, and ciprofloxacin was performed using the tetrazolium microplate assay [24] and DNA fingerprinting performed using spoligotyping [25]. Linkage of source patient with infected guinea pigs relied upon genotypic (identical spoligotype) and phenotypic (drugsusceptibility pattern) concordance and patient ward occupancy 3-9 wk prior to animal PPD conversion, reflecting TB incubation in these guinea pigs [26]. Individual patient infectiousness was determined using the Wells-Riley airborne infection equation (see Text S1) [27]. Determinants of patient infectiousness were assessed by univariate and multiple logistic regression in which nonsignificant factors were sequentially dropped from the model using SPSS and SAS statistical software [28].

Ethical Approval
The study was approved by the Institutional Review Boards at Hospital Dos de Mayo, Asociació n Bené fica PRISMA, Perú , and Imperial College London, Hammersmith Hospital Campus, UK. Animal ethical approval was obtained from the Veterinary Medicine Faculty, Universidad Nacional Mayor San Marcos, Lima, Perú , which supervised all animal work. more patient bed days were accounted for by sputum culture-positive patients than by sputum culture-negative patients; in contrast, for non-MDR TB patients, significantly more patient bed days were accounted for by sputum culturenegative patients than by sputum culture-positive patients (both p , 0.0001; Figure 2).

Patient Characteristics
The treatment status of pulmonary TB patients is shown in Table 1. ''Optimal'' signifies an antibiotic regimen suitable for the TB strain drug resistance pattern, whilst ''suboptimal'' signifies a regimen less likely to result in cure. The suboptimal treatment category included standard first-line therapy (4 mo of isoniazid, rifampicin, pyrazinamide, and ethambutol followed by 2 mo of rifampicin and isoniazid [29]) administered to MDR TB cases whilst indirect drug susceptibility results were awaited, a process commonly taking 2-3 mo. The suboptimal treatment group also included patients with drugresistant strains given the now abandoned second-line regimen [29] containing standard first-line drugs with the addition of a single drug, streptomycin, and any first-line or second-line therapy that was delayed because of adverse drug reactions or drug shortages.
Spoligotype results were available for 49 (42%) pulmonary TB admissions, corresponding to 39 (40%) of 97 patients. The remaining admissions were predominantly patients with culture-negative sputum, or those unable to produce sputum. Sixteen culture-positive patients, of whom four were smear positive, had no spoligotyping result, due to laboratory contamination or unwillingness to consent. For the 49 admissions with spoligotyping results, 25 different patterns were observed, none of which corresponded to Beijing strains of M. tuberculosis. Figure 1B shows the temporal distribution of patient days accounted for by patients with different spoligotype patterns.

Detection of TB Transmission to Guinea Pigs
No positive PPD skin tests were seen in quarantine (760 tests) or unexposed negative control animals (287 tests).
During 505 study days, 292 animals were exposed to ward air, an average of 92 each month. A total of 159 animals developed positive tests (! 7.5 mm), 124 were TB culture positive, and a further five had autopsy evidence of pulmonary TB [4]. Nine ward air-exposed PPD-negative animals had evidence of TB, as did six of 25 intercurrent deaths between skin tests [4]. In total, cultures were positive in 135 (46%) of 292 animals. No evidence of TB was found in unexposed negative controls.
Of 135 culture-positive guinea pigs, drug susceptibility and spoligotyping results were available for 125. Of this group 121 (97%) were multidrug resistant, one was isoniazid monoresistant, and three were fully drug susceptible. Eight different spoligotype patterns were observed. Figure 1C shows the distribution of guinea pig TB infections and corresponding spoligotype patterns according to the monthly skin test when TB was diagnosed. Data from intercurrent deaths are included in the subsequent skin test. There was a large monoclonal TB outbreak in month 10 (spoligotype pattern #7), which continued into months 11 and 12. This strain reappeared in months 14 and 15, but was not seen in month 13.
In ten guinea pigs, spoligotyping was performed on positive cultures from three separately dissected tissues (lymph nodes, lung, spleen). Spoligotype patterns were identical in all tissues. In nine animals with multiple pulmonary primary foci, two foci were dissected individually from different lung lobes with separate sets of instruments. All primary foci yielded identical spoligotype patterns. Distribution of infection according to animal cage location was random, consistent with airborne transmission from the ward, not horizontal spread between animals.

Infectiousness of Patients
Of 125 guinea pigs with spoligotype results, 122 (98%) were matched with a patient with an identical TB drug-susceptibility and spoligotype pattern who had resided on the ward 3-9 wk prior to the guinea pig skin test conversion. Ten different infectious patients were identified, who had seven spoligotype patterns. Of these, one was unique to one patient, and the remainder were seen in multiple patients. However, in the 3-9 wk preceding the infection of each guinea pig cluster, only one patient was ever identified with such a pattern. The characteristics of the ten identified infectious patients are shown in Table 2. Six had MDR, one had monoresistant, and three had drug-susceptible TB. Amongst the six patients with MDR TB strains, five strains were additionally resistant to ethambutol, of which two strains were also resistant to streptomycin. Two of the three patients with drug-susceptible TB were newly started on treatment and hence they had spent time on the ward untreated, and the third had stopped treatment. All identified patients with infectious drug-resistant TB had been suboptimally treated: six were on inadequate treatment regimens, and one had had the initiation of correct treatment delayed for 11 d.
There were at least two further infectious patients. One infected two animals with an MDR TB strain not seen amongst patients' spoligotyping results. During the 3-9 wk prior to the animals' infection there were three sputum culture-positive MDR TB (one smear-positive) patients without spoligotyping results on the ward, any of whom could have infected these animals. A second unidentified infectious patient (potentially one of the three above) resulted in a guinea pig PPD conversion seen in month 5, with an MDR TB strain with spoligotype pattern #1. Table 2 also shows patient infectiousness, q, the number of infectious quanta produced per hour. For absolute ventilation in the Wells-Riley equation, mean total air entering the animal enclosure was used: 28 m 3 /min (standard deviation [SD] 0.9 m 3 /min). This comprised ward air (21 m 3 /min), air from a small room used in the mornings for non-TB patients (3.4 m 3 /min), and outside air infiltration into ducts between ward and animal house (3.4 m 3 /min). The calculation methodology and a worked example for infectious patient number 6 are provided in Text S1. Average patient infectiousness over the entire study was 8.2 infectious quanta/h (using 174 total animal infections) or 6.7 (using 144 total animal infections with evidence of disease, excluding PPD-positive animals without autopsy or culture evidence of TB).

Determinants of Infectiousness
Regression analysis was performed for putative determinants of infectiousness for pulmonary TB admissions with spoligotype results. Admissions rather than patients were analysed, because patients with multiple admissions changed sputum smear, drug-susceptibility, or treatment status. Fortynine admissions (39 patients) were included. Ten were classed as infectious resulting in TB transmission to guinea pigs, and 39 as noninfectious. In univariate analysis, the same patient characteristics were significantly associated with both whether or not TB transmission occurred (Table 3) and with the degree of patient infectiousness (q, see Text S1; Table 4): sputum smear-positivity, MDR TB versus non-MDR TB, and suboptimal treatment. Multiple regression analysis of whether or not a patient admission caused TB transmission to guinea pig(s) showed an independent statistically significant association only with MDR status (p ¼ 0.02); TB transmission to guinea pigs was also observed more frequently in admissions of sputum smear-positive (versus smear-negative) patients, but this association was not statistically significant (p ¼ 0.08; Table 3). Predicted probabilities were estimated from an exact multiple logistic regression: the predicted probability of infectiousness was 3% for smearnegative non-MDR TB patients; 18% for patients who were either smear positive or had MDR TB; and 59% for smearpositive MDR TB patients. Further univariate analyses were performed on a subset of 33 consenting patient admissions for which symptom information such as cough frequency was available, but no variable approached significance (all p . 0.6), which may reflect low power due to the small sample.
Multiple regression analysis of data from all patients identified as having caused TB transmission demonstrated that the degree of patient infectiousness was independently significantly associated with MDR TB; patient infectiousness appeared possibly to increase with sputum-smear positivity and days on the ward without treatment, but these associations were not found to be statistically significant (Table 4). This statistical model allowed patient character-  (B) Number of bed days in each study month resulting from pulmonary TB patients for whom a TB strain spoligotype pattern was available. Each block of colour corresponds to one patient, and each colour to one of the eight spoligotype patterns observed in the guinea pigs. Pale yellow represents the remaining 19 patterns observed in patients whose TB was not seen in the guinea pigs. If a patient resided on the ward for a period spanning more than one study month, that patient is included in the month where they accounted for more smear positive patient bed days. The coloured blocks containing numbers correspond to the ten identified infectious patients, numbered 1 to 10 in Table 2.
(C) Percentage of animal colony skin tested each study month that were PPD positive. Each colour represents one spoligotype pattern, except for pale blue, which represents ten guinea pigs culture positive for TB but for which spoligotype patterns were unavailable. White represents animals that were PPD positive but TB culture negative. Animals culture positive for TB that were PPD false negatives or that died between skin tests were included in the subsequent month's skin test. doi:10.1371/journal.pmed.0050188.g001 istics to be used to predict the number of infectious quanta produced per hour (q). The regression model (Table 4) indicated that: non-MDR TB, smear-negative patients with undelayed treatment had a predicted infectiousness q ¼ 0.3; patients with MDR TB or sputum-smear positivity without treatment delay had a predicted q ¼ 3.9; more than 2 d of treatment delay resulted in an increase in q of 1.7; sputumsmear positive MDR TB had a predicted q ¼ 24; MDR TB and (more than 2 d of delayed treatment or sputum-smear positivity) had predicted q ¼ 14; and sputum-smear positive MDR TB with more than 2 d of delayed treatment had a predicted q ¼ 54.

Discussion
This study provides novel characterization of the heterogeneity and determinants of infectiousness of HIV-positive TB patients by applying molecular strain characterization to track airborne TB transmission to guinea pigs. This research has for the first time (to our knowledge) demonstrated that amongst HIV-positive patients TB infectiousness is extremely variable, that a few HIV-positive patients were highly infectious, and that inadequately treated MDR TB patients accounted for the great majority of TB transmission. In contrast to seminal studies of TB transmission using a similar guinea-pig method of detection half a century ago, this study was conducted in a real-life busy ward in a low-resource setting with unselected patients, composed of a heterogeneous mix of new and established diagnoses of drugsusceptible and drug-resistant TB, with varying treatment regimens. These results therefore have important implications for TB infection control, especially in the era of increasingly integrated TB and HIV care and the emergence of XDR TB strains.
Average patient infectiousness over the whole study period for these HIV-positive TB patients with high rates of MDR TB was up to six times greater than that calculated for the heterogeneous mix of patients in the 1950s studies (q ¼ 1.25) [4,27]. However, this average masks considerable variability in infectiousness between patients. Three highly infectious patients were observed, all with MDR TB, with q-values of 40, 52, and 226. It should be noted that these q-values reflect TB transmission from humans to guinea pigs. The infectious dose of M. tuberculosis for humans is unknown; hence the concept of infectious quanta used in airborne infection models [27]. For fully virulent M. tuberculosis strains just one droplet nucleus can establish infection and disease in guinea pigs, but for strains of reduced virulence for guinea pigs, up to four aerosolised colony-forming units may be required to establish a single pulmonary primary focus [30]. Thus some caution is needed in comparing the infectiousness of patients in this study with published q-values calculated for human-tohuman transmission, such as q ¼ 13 for an untreated office worker who infected 27 coworkers over 4 wk prior to diagnosis [31], and q ¼ 250 for an outbreak associated with intubation and bronchoscopy of a TB patient [11,27]. However, comparisons can be drawn with q-values calculated for Riley's study: q ¼ 1.25 average for all patients; q ¼ 60 for the most infectious case, with laryngeal TB [27]. Direct comparisons should, however, be made cautiously owing to methodological differences between the studies such as the type of guinea pig used and the cutoff for a positive skin test [32]. This model for identifying infectious patients required TB strains to be not only transmissible to guinea pigs, but also sufficiently virulent to cause disease from which a strain could be recovered with corresponding spoligotype. Animals with positive PPD skin test conversions but culture negative for TB were observed throughout the study. Due to the continuous exposure to ward air, mixed infections might be expected in the animals, but in fact were not observed despite the culture of separately dissected lung foci. Because they are relatively uncommon, mixed infections were not specifically sought in patients [33]. The relative infectiousness of patients in this study was highly variable, and a patient with MDR TB was found to be highly infectious, producing 226 infectious quanta/h. The TB strain responsible, not a Beijing strain, was also seen in a second highly infectious MDR TB patient, producing 52 infectious quanta/h. This observation suggests a potentially strain-related factor involved in transmissibility, perhaps an enhanced ability to survive aerosolisation and the physical stresses of being airborne. An alternative explanation might be an effect on disease phenotype. It is interesting to note that both patients had fever and cough and produced large volumes of sputum, between 10 and 90 ml daily. Neither patient had cavitation on chest X-ray. Although no formal ear-nose-throat assessment was documented, neither was diagnosed with laryngeal TB. Both patients also spent time on the ward untreated. The first was untreated for the first 11 d of a 32 d admission before second-line drugs were commenced, due to difficulties in access to medications. The second patient had recently been commenced on suboptimal treatment (standard first-line therapy plus streptomycin) and this was suspended for ten of 26 ward days because of adverse effects.
The finding that sputum smear positivity was associated with TB transmission concurs with previous studies [5][6][7]. The effect of treatment on the infectiousness of TB patients is also well known [2,5], with numbers of viable bacteria falling precipitously following initiation of effective chemotherapy [34,35]. Whilst some data suggest that apparent cure of MDR TB may be achieved with first-line drugs [36,37], other studies have shown poor outcomes for such patients [38]. The current study shows how suboptimal treatment of MDR TB patients is likely to facilitate ongoing TB transmission. There is conflicting evidence concerning the relative transmissibility of MDR versus drug-susceptible TB strains [14][15][16][17][18][19][20]. In this study, patients with MDR TB were significantly more likely than those without MDR TB to transmit TB to guinea pigs. However, this finding should be interpreted with caution because of colinearity with suboptimal treatment (5 of 6 identified infectious MDR TB patients were on suboptimal regimens, and the other had treatment initiation delayed for 11 days whilst suitable medications were acquired). Regardless, the high relative infectiousness of inadequately treated MDR TB patients demonstrated in this study underscores the importance of prompt specific treatment guided by rapid drug-susceptibility testing, rather than restricting MDR TB testing and specific therapy to patients who survive failing empiric first-line therapy, as currently happens in most lowresource settings. It also has important implications for hospital policies that allow suboptimally treated MDR TB cases to be cared for in multi-bedded rooms.
The highly infectious nature of some of the HIV-positive MDR TB patients identified in this study has important implications for TB infection control. Administrative control measures that facilitate the rapid diagnosis, isolation, and prompt treatment of such patients are paramount. With increasing congregation of infectious and susceptible individuals not only in hospitals but also in such settings as antiretroviral therapy roll-out, HIV antenatal care, and voluntary counselling and testing facilities [39], environmental control measures are also of great importance. As the infectiousness of a TB source increases, the relative protection provided by dilutional room ventilation decreases [31]   and may become inadequate at the relatively low levels of air exchange usually provided by mechanical ventilation. High rates of ventilation would be required to provide protection from the extremely infectious newly diagnosed MDR TB case observed in this study. Achieving this through mechanical means is an expensive solution for much of the world where TB is most prevalent. In contrast, well-designed natural ventilation [40] provides high ventilation rates for little cost, and furthermore is highly applicable to areas such as crowded waiting rooms where infectious, untreated TB patients are most likely to be encountered. TB infection control must be a priority in the current roll-out of enhanced HIV care, and should be carefully considered in the design and construction of any new infrastructure for such programmes. The need for strengthened TB infection control is also highlighted by the recent outbreak of XDR TB amongst HIVcoinfected patients in South Africa; this outbreak was predominantly nosocomial and resulted in extremely high mortality [41]. The variability of infectiousness of patients demonstrated in this study highlights the usefulness of a potential test for TB infectiousness that would allow targeted isolation of the most infectious patients in the settings where isolation facilities are sparse, as is unfortunately the case in much of the world where TB is most prevalent. One patient in our study, with MDR TB, infected over half of the guinea pig colony. The development of tests that allow early identification and isolation of such patients in a clinical setting is a research priority.
There are some limitations to this study. The first is the incomplete set of spoligotyping data, with results in only 49 of 118 pulmonary TB admissions. Despite this deficiency, ten of at least 12 infectious patients were identified. It is possible that transmission occurred from other patients for whom spoligotyping was unavailable, but certain factors suggest that this is not the case. Most patients without spoligotype results had negative sputum cultures, and whilst smear-negative TB transmission occurs [8], smear-positive patients account for the majority [5][6][7]. In this study, 16 culture-positive patients, of whom four were smear positive, had no spoligotype result. Fortuitously, these patients were either temporally or phenotypically (drug susceptibility pattern) unrelated to the ten guinea pig clusters linked with infectious patients,  For categorical independent variables, the number of cases (% of total in parentheses) is shown, and for continuous independent variables, the median (IQR in parentheses) is shown. The primary analysis treated the independent variables ''Days on Ward'' and ''Days on Ward without Treatment'' and ''Age'' as raw, continuous variables. These primary analyses are complemented by secondary analyses (in parentheses) of each of these variables divided into categories around cutoff points chosen to maximize the likelihood of the outcome. Note that the analysis of these independent variables as continuous or categorical data gave similar results. b Regression analysis was performed on putative determinants of infectiousness for the 49 ward admissions by pulmonary TB patients for which spoligotype patterns were available. Four patients accounted for the ten multiple admissions in this group. The first had MDR TB and was infectious on one admission when inadequately treated, and non-infectious on a subsequent admission when adequately treated. The second had four admissions: three were noninfectious with drug-susceptible TB that was first smear-positive and then smearnegative. The fourth admission was infectious, with acquired MDR-TB. The third patient had two noninfectious admissions separated by 4 mo with smear-positive inadequately treated monoresistant TB. The fourth patient had two noninfectious admissions separated by two months with smear-negative drug-susceptible TB. excluding them as coinfectors. Indeed the two clusters of MDR TB guinea pigs with unidentified infectious sources became infected at times corresponding to the ward residency of smear-positive MDR TB patients without spoligotype results. We cannot exclude the possibility that guinea pig infections occurred from staff or visitors with TB, however all staff and visitors wore particulate respirators. It is possible that the large monoclonal outbreak observed in the guinea pigs was in fact made up of more than one strain. However, the concordant drug susceptibility data and epidemiological match with a patient on the ward at an appropriate time prior to the infections obviates the need for secondary typing of strains. The Wells-Riley model has inherent limitations [42], but these do not influence evaluations of relative infectiousness, and it allows comparison with published values of TB infectiousness calculated using the same model. The design of this study did not permit determination of the duration of patient infectiousness because of the interval of one month between skin tests, and the variability in the period required for these guinea pigs to become PPD positive following TB infection. It is possible that values for patient infectiousness are underestimates, because the entire period of a patient's hospital admission was used for the exposure duration variable in calculations, and it would normally be expected for patient infectiousness to tail off once treatment was initiated, although this would not be the case with suboptimal treatment. However, in univariate analyses patient days on the ward was not significantly associated with TB transmission. The relatively narrow age range and the small number of women amongst the patients is a further limitation of this study, because both young age and male sex have been associated with TB transmission to contacts [15,43]. Because all patients were HIV positive, our study was unable to yield evidence concerning the infectiousness of HIV-positive versus HIV-negative MDR TB patients, and this could be a future area of study using our airborne infection facility.
In conclusion, this study has demonstrated the potential of HIV-positive patients with MDR TB to be highly infectious. With the great majority of TB transmission in this study occurring from inadequately treated MDR TB patients, these results identify the importance of early drug susceptibility testing and initiation of effective chemotherapy for drugresistant TB to prevent ongoing transmission and facilitate TB control. Furthermore, this study highlights the importance of environmental control measures to prevent airborne TB transmission in crowded health care settings, especially in areas with a high prevalence of HIV infection and drugresistant TB, and in today's era of emerging XDR TB. HIVpositive patients with unrecognised or inadequately treated MDR TB coinfection may be highly infectious, and effective TB infection control measures are essential to prevent health care facilities from disseminating drug-resistant TB.

Supporting Information
Text S1.
The infectiousness of individual patients (q, see Text S1) was tested for associations with the independent variables listed in the table. The p-value refers to the univariate associations assessed with the Mann-Whitney U test for categorical independent variables and the nonparametric Spearman correlation coefficient for continuous independent variables. The p-value is also shown for equivalent exact statistical testing. The Tobit (censored) regression analysis allows for the censoring of data from patients who did not infect any guinea pigs and who therefore had an undetectable degree of infectiousness (see Text S1). After an initial model fit for all covariates individually, the variable that entered the model based on the highest pseudo-R 2 was MDR TB. Each additional variable added would be based on the lowest p-value derived from likelihood ratio statistic provided that the p-value was , 0. The answer to this question might help to optimize the control of tuberculosis infection, particularly in hospitals. In this study, the researchers investigate current patterns of tuberculosis infectiousness among HIV-positive patients by recreating the 1950s guinea pig model for tuberculosis transmission in a hospital in Lima, Perú .
What Did the Researchers Do and Find? The researchers passed all the air from an HIV-tuberculosis ward over guinea pigs housed in an animal facility on the hospital's roof. The guinea pigs were tested monthly with tuberculin skin tests, and tissues from positive animals were examined for infection with M. tuberculosis. Sputum was also collected daily from the patients on the ward. The researchers then used the timing of patient admissions and guinea pig infections, together with the drug susceptibility patterns and DNA fingerprints of the M. tuberculosis strains isolated from the animals and the patients, to identify which patients had infected which guinea pigs. Finally, they used a mathematical equation to calculate the relative infectiousness of each patient in airborne infectious units (''quanta'') per hour. During the 505 study days, although 97 HIV-positive patients with tuberculosis were admitted to the ward, just ten patients were responsible for virtually all the characterized cases of tuberculosis among the guinea pigs. Six of these patients had MDR tuberculosis that had been suboptimally treated. The average patient infectiousness over the entire study period was 8.2 quanta per hour-six times greater than the average infectiousness recorded in the 1950s. Finally, the three most infectious patients (all of whom had suboptimally treated MDR tuberculosis) produced 226, 52, and 40 quanta per hour.
What Do These Findings Mean? These findings show that a few inadequately treated HIV-positive patients with MDR tuberculosis caused nearly all the tuberculosis transmission to guinea pigs during this study. They also show for the first time that tuberculosis infectiousness among HIV-positive patients is very variable. The increase in the average patient infectiousness in this study compared to that seen in the 1950s hints at the possibility that HIV infection might increase tuberculosis infectiousness. However, studies that directly compare the tuberculosis infectiousness of HIV-positive and HIV-negative patients are needed to test this possibility. More importantly, this study demonstrates the potentially high infectiousness of inadequately treated MDR TB patients and their importance in ongoing TB transmission. These findings suggest that rapid, routine testing of antibiotic susceptibility should improve tuberculosis control by ensuring that patients with MDR TB are identified and treated effectively and quickly. Finally, they re-emphasize the importance of implementing environmental control measures (for example, adequate natural or mechanical ventilation of tuberculosis wards, or crowded waiting rooms or emergency departments where tuberculosis patients may be found) to prevent airborne tuberculosis transmission in health-care facilities, particularly in areas where many patients are HIV positive and/or where MDR tuberculosis is common.
Additional Information. Please access these Web sites via the online version of this summary at http://dx.doi.org/10.1371/journal.pmed. 0050188.
The US National Institute of Allergy and Infectious Diseases provides information on all aspects of tuberculosis, including multidrugresistance tuberculosis, and on tuberculosis and HIV The US Centers for Disease Control and Prevention provide several fact sheets and other information resources about all aspects of tuberculosis (in English and Spanish) The World Health Organization's 2008 report on global tuberculosis control-surveillance, planning, financing provides a snapshot of the current state of the global tuberculosis epidemic and links to information about all aspects of tuberculosis and its control (in several languages) HIVInsite provides detailed information about coinfection with HIV and tuberculosis Avert, an international AIDS charity, also provides information about the interaction between HIV and tuberculosis Tuberculosis Infection-Control in the Era of Expanding HIV Care and Treatment is a report from the World Health Organization