Does Resistance to Pyrazinamide Accurately Indicate the Presence of Mycobacterium bovis?

Mycobacterium bovis is intrinsically resistant to pyrazinamide (PZA), and the prevalence of clinical infection with M. bovis is low in countries with good bovine tuberculosis control programs. This finding has supported a strategy of using PZA monoresistance as an initial screening tool for M. bovis, a strategy that risks missing cases of infection with M. bovis strains that have broader resistance. A previous study found that screening for M. bovis by using PZA monoresistance had a poor positive predictive value (3), but the study was not able to assess the sensitivity or specificity of PZA resistance screening for the detection of M. bovis as it did not include data on the denominator. We therefore sought to assess the efficacy of using either PZA monoresistance or any PZA resistance to identify M. bovis in a population-based study. At the same time, we estimated the prevalence of M. bovis in San Francisco.

Mycobacterial isolates in all diagnosed cases of tuberculosis in San Francisco were collected in a prospective study (4). The isolates underwent biochemical testing for niacin and nitrate production for identification of species within the Mycobacterium tuberculosis complex, as the vast majority of isolates of M. bovis are niacin and nitrate negative (6, 8). Susceptibility testing for PZA and other drugs was done with the Bactec method (S. Siddiqi, Bactec 460TB system product and procedure manual, Becton Dickinson and Co., Sparks, MD). Isolates were sent to Stanford University for standard molecular typing using IS6110 and polymorphic guanine-cytosine-rich sequence (PGRS) restriction fragment length polymorphism (2, 9, 10).

We performed a retrospective cohort study, which included all available isolates from San Francisco from 1991 to 1999 identified as M. bovis by nitrate and niacin testing and all isolates that were PZA resistant. Isolates without a PZA susceptibility test result were excluded. For the identification of M. bovis, isolates were screened for the absence of the region of difference 4 (RD4) and RD9 (1). All M. bovis isolates were screened for the RD1 deletion, which is specific to M. bovis bacillus Calmette-Guérin (BCG) (1). Spoligotyping was also performed (5); M. bovis has a characteristic pattern consisting of deleted direct repeat spacers 39 to 43, and BCG has a characteristic pattern with three additional deletions. Isolates without available DNA were classified based on the biochemical test results only.

Statistical analysis was done using Stata version 8 (Stata Corporation, College Park, TX). We used the chi-square test of proportions or Fisher's exact test to compare the characteristics of cases of M. bovis infection with those of cases of M. tuberculosis infection.

Of 2,476 cases of tuberculosis diagnosed and reported in San Francisco from 1991 to 1999, 2,115 (85.4%) were culture positive and 1,526 (72.2%) had a PZA susceptibility test result. The rate of PZA susceptibility testing varied by year, with <20% of isolates tested from 1993 to 1994, 70 to 80% in 1992 and 1995, and >90% of isolates from other years. PZA testing was random, and isolates of M. bovis as identified by biochemical testing were not more likely to have been tested than non-M. bovis isolates (P = 0.54). Our study identified 30 PZA-resistant isolates; 25 had a viable culture and/or DNA available. PZA monoresistance was present in 18 of 30 isolates (Table 1). One additional isolate was pansensitive and was identified as M. bovis by biochemical testing but as M. tuberculosis by spoligotyping and genomic deletion analysis. Conversely, two PZA-resistant isolates were identified as M. tuberculosis by biochemical testing and as M. bovis by spoligotyping and genomic deletion analysis (Table 2). The sensitivity for detecting M. bovis increased from 81.8% when PZA monoresistance was used to 100% when any PZA resistance was used. Two-thirds of PZA-resistant isolates and half of PZA-monoresistant isolates were M. tuberculosis. The positive predictive values of PZA monoresistance versus any PZA resistance were 50.0 and 36.7%, respectively. The specificity and negative predictive value were >98% with both strategies (Table 3).

After excluding two cases of M. bovis BCG administration for bladder cancer that resulted in genitourinary disease, we identified nine cases of M. bovis infection in San Francisco between 1991 and 1999. These represented 0.6% of all culture-positive cases of tuberculosis with PZA susceptibility test results. One of the nine isolates was identified as M. bovis BCG and was resistant to isoniazid (INH) and PZA. This strain was isolated from a 40-year-old Chinese woman with noncavitary pulmonary disease who had not undergone human immunodeficiency virus testing. Patients infected with M. bovis versus M. tuberculosis were more likely to have been born in Mexico (odds ratio [OR], 33; P < 0.005) and were younger (median ages, 32 and 45 years old, respectively; P = 0.024).

Among the nine M. bovis isolates with IS6110 and PGRS restriction fragment length polymorphism data available, eight had a single IS6110 band. There were two clusters corresponding to two persons with identical genotyping patterns, but there were no epidemiological links between them.

Our results suggest that definitive testing to distinguish M. bovis from M. tuberculosis should include all PZA-resistant isolates, not only those with PZA monoresistance.

The prevalence of M. bovis in San Francisco was 0.6%, similar to the prevalence in other countries with good bovine tuberculosis control programs but lower than the reported 6.6% prevalence in San Diego (7). The association between birth in Mexico and M. bovis infection was similar to that found in the study in San Diego. However, the lower prevalence of M. bovis in San Francisco likely reflects the greater distance of San Francisco from the Mexican border.

The lack of DNA for 5 of the 30 isolates classified by biochemical test results might have introduced misclassification of M. bovis and M. tuberculosis isolates. However, the biochemical test results correlated well with the results of spoligotyping and genomic deletion analysis. Although 28% of the isolates were not tested for PZA resistance, PZA testing was random and should not bias our estimates of the sensitivity and specificity. Given the small number of cases of M. bovis infection, our estimates of sensitivity and positive predictive value have wide confidence intervals. The positive predictive value of PZA resistance for the identification of M. bovis would likely be higher in areas where M. bovis is endemic in cattle. Since the techniques we used are not universally available, optimal testing strategies will vary depending on local resources.

• Copyright © 2005 American Society for Microbiology