Epidemiologic Correlates of Pyrazinamide-Resistant Mycobacterium tuberculosis in New York City

Abstract

Pyrazinamide (PZA) has important sterilizing activity in tuberculosis (TB) chemotherapy. We describe trends, risk factors, and molecular epidemiology associated with PZA-resistant (PZA^r) Mycobacterium tuberculosis in New York City (NYC). From 2001 to 2008, all incident culture-positive TB cases reported by the NYC Department of Health and Mental Hygiene (DOHMH) were genotyped by IS6110-based restriction fragment length polymorphism and spoligotype. Multidrug-resistant (MDR) isolates underwent DNA sequencing of resistance-determining regions of pncA, rpoB, katG, and fabG1. Demographic and clinical information were extracted from the NYC DOHMH TB registry. During this period, PZA^r doubled (1.6% to 3.6%) overall, accounting for 44% (70/159) of the MDR population and 1.4% (75/5511) of the non-MDR population. Molecular genotyping revealed strong microbial phylogenetic associations with PZA^r. Clustered isolates and those from acid-fast bacillus (AFB) smear-positive cases had 2.7 (95% confidence interval [CI] = 1.71 to 4.36) and 2.0 (95% CI = 1.19 to 3.43) times higher odds of being PZA^r, respectively, indicating a strong likelihood of recent transmission. Among the MDR population, PZA^r was acquired somewhat more frequently via primary transmission than by independent pathways. Our molecular analysis also revealed that several historic M. tuberculosis strains responsible for MDR TB outbreaks in the early 1990s were continuing to circulate in NYC. We conclude that the increasing incidence of PZA^r, with clear microbial risk factors, underscores the importance of routine PZA drug susceptibility testing and M. tuberculosis genotyping for the identification, control, and prevention of increasingly resistant organisms.

INTRODUCTION

Pyrazinamide (PZA) is a first-line antituberculosis (anti-TB) drug and the cornerstone of modern short-course chemotherapy. PZA acts synergistically with other TB drugs to accelerate culture conversion and reduce the risk of relapse among patients with drug-susceptible TB (1). In patients with multidrug-resistant (MDR) TB, defined as resistance to at least isoniazid (INH) and rifampin (RIF), inclusion of PZA is recommended to reduce the treatment duration (2, 3), while optimizing MDR TB treatment regimens based on PZA susceptibility may improve clinical outcomes (4). Due to its well-documented sterilizing capability, PZA has been included in several new TB drug regimens (5–7).

PZA is a prodrug that requires activation to pyrazinoic acid by the pyrazidamidase of M. tuberculosis under acidic conditions (8). While at least one other M. tuberculosis gene has been associated with PZA resistance (PZA^r) (9), mutations in pncA, which encodes the pyrazidamidase, account for the majority of PZA^r in vitro (10, 11). Most genes associated with drug resistance in M. tuberculosis, such as katG (INH) and rpoB (RIF), have clear mutational hot spot regions 7 to 66 nucleotides (nt) in length. In contrast, mutations observed in pncA span a region of ∼600 nt, comprising the entire gene and the putative promoter region (10, 12, 13). Growth-based assays of M. tuberculosis PZA susceptibility are the standard but are not always performed routinely, except in large referral laboratories, because they are technically challenging (14, 15). However, pncA mutations in clinical M. tuberculosis isolates have generally been found to correlate with phenotypic PZA drug susceptibility testing (DST) results, supporting the value of pncA sequence analysis as an alternative means to establish PZA resistance (13, 16).

Genetic markers have been used to confirm or refute TB outbreaks and to estimate the proportion of recent transmission in a population (17). For instance, due to the wide diversity of pncA mutations, these sequences can provide a genetic marker to confirm or resolve genotypic clusters, where the presence of identical pncA mutations in genotypically clustered strains is supportive of primary transmission while genotypic clusters with diverse pncA mutations suggest acquired (de novo) PZA^r, indicating these cases may not be epidemiologically related (17).

Despite the unique role PZA plays in modern TB chemotherapy, few studies have investigated the epidemiology of PZA^r in both general and MDR TB populations (18). Using TB case records and surveillance data provided by the New York City (NYC) Department of Health and Mental Hygiene (DOHMH) from 2001 to 2008, we conducted a population-based study of M. tuberculosis isolates from 6,260 culture-positive TB cases to examine PZA^r in terms of clinical, microbial, and demographic risk factors. Due to the clinical and epidemiologic importance of MDR TB, we also performed a case-control study to identify PZA^r risk factors and examine PZA^r acquisition and clustering among the MDR population.

MATERIALS AND METHODS

All incident culture-positive TB cases reported and verified by the NYC DOHMH between January 2001 and December 2008 (n = 6,260) were included in the study (Fig. 1). Routine genotyping was performed by the Public Health Research Institute (PHRI) Tuberculosis Center at Rutgers University (IS6110-restriction fragment length polymorphism [RFLP]) and New York State Wadsworth Center (spoligotype) (19). All strains with genotyping results available (n = 5,877) were assigned a molecular lineage using the taxonomic designation previously described by Gagneux and Small (20) and a strain code following a nomenclature system of the PHRI TB Center that has been described previously (21, 22). Strains identified as Mycobacterium bovis (n = 68) or Mycobacterium africanum (n = 78) (23) were omitted from our analysis. Clusters were defined as two or more strains sharing identical IS6110-RFLPs and spoligotypes. DST was performed at the NYC DOHMH Public Health Laboratories and the New York State Wadsworth Center (24) reference laboratories, which utilized the Bactec 460TB system (Becton, Dickinson and Company, Franklin Lakes, NJ) until 2003 and the BD Bactec MGIT 960 mycobacterial detection system thereafter. DST results were available for culture-positive TB cases for all first-line anti-TB agents: INH, RIF, ethambutol (EMB), and PZA. DST results for second-line drugs (SLD), including kanamycin, capreomycin, amikacin, fluoroquinolones, and ethionamide, were available for all cases with reduced susceptibility to any first-line agent. M. tuberculosis isolates were classified as MDR if they were resistant to at least INH and RIF (3). Isolates were classified as poly-PZA^r if they were resistant to PZA and at least one other drug, excluding MDR isolates (Fig. 1).

FIG 1.

Study schema. MDR, resistant to at least isoniazid and rifampin; PZAR, PZA resistant; PZAS, PZA susceptible; mono-PZAR, resistant only to PZA and no other drug; MDR-PZAR, resistant to at least isoniazid, rifampin, and PZA; poly-PZAR, resistant to PZA and at least one other drug, excluding MDR-PZAR.

Data collection.

Demographic and clinical information was provided by the NYC DOHMH TB Registry, which contains information for each reported TB patient obtained by interview and medical-record abstraction performed by trained Bureau of Tuberculosis Control (BTBC) staff, using standard data collection forms. Demographic variables included age at TB diagnosis, sex, birthplace (United States or foreign born with country of birth), number of years since arrival in the United States for foreign-born patients, and race/ethnicity. Sociodemographic variables included reported homelessness; substance use (injection drug use), noninjection crack cocaine use, or noninjection drug use (consolidated into yes or no); alcohol abuse; and history of TB treatment at Rikers Island Prison Complex. Clinical variables included initial chest radiography results (normal/abnormal and absence/presence of cavities), the anatomical site of TB disease, respiratory acid-fast bacillus (AFB) smear status, final culture conversion (final conversion from positive to negative culture), HIV status (infected, uninfected, or unknown); and death from any cause (yes or no, excluding patients who refused treatment or were lost to follow-up).

All MDR isolates (n = 159) were subjected to PCR amplification (primers, 5′-ATGCGGCGTTGATCATCG-3′ and 5′-CAGGAGCTGCAAACCAACTCG-3′), followed by standard capillary sequencing of pncA promoter and coding DNA sequence (CDS), as previously described (12). Mutations were identified by alignment of nucleotide sequences to the M. tuberculosis H37Rv reference strain (NCBI accession number AL123456) (25) using ClustalW2 (26). All MDR isolates were also analyzed for the presence of known drug resistance-conferring mutations in the gene targets katG, fabG1 (promoter), and rpoB, using the IBIS platform (Ibis Biosciences Carlsbad, CA), which is based on PCR followed by electrospray ionization mass spectrometry, as described previously (27).

Statistical analysis.

All statistical analysis was performed using SAS (Cary, NC) 9.3. To examine aggregate patient risk factors associated with PZA^r, unadjusted odds ratios (OR) were estimated using standard two-by-two contingency table univariate analysis with Fisher's exact test at a 0.05 significance level. We compared demographic and clinical characteristics of TB patients infected with PZA^r M. tuberculosis to those of patients infected with PZA-sensitive (PZA^s) M. tuberculosis, together with the microbial features (phylogenetic lineage) of the infecting isolates. Trend analysis was performed using the least-squares method, and R² values are reported. The analysis of PZA susceptibility was stratified by MDR. In the case-control study, we compared the characteristics of MDR-PZA^r cases to those of MDR-PZA^s controls. A multivariate analysis by logistic regression was performed, using a priori variables reported in the literature to be associated with drug resistance, including age and HIV status. Variables with P values of ≤0.2 in the univariate analysis were included in the multivariate models. We further examined genotypic clustering of MDR-PZA^r M. tuberculosis strains using pncA sequence data to support primary PZA^r transmission events (defined by identical pncA mutations in genotypically clustered strains) and to distinguish independently acquired PZA^r (IS6110-RFLP/spoligotype-based clusters showing diverse pncA mutations) within the MDR population (17). Finally, we evaluated the overall agreement between PZA susceptibility defined by phenotype (DST) and genotype (pncA sequence) among MDR isolates using a kappa statistic.

Ethical approval for this study was obtained from the Institutional Review Boards of the NYC DOHMH and Rutgers University (Newark, NJ).

RESULTS

PZA resistance trends and associated risk factors among NYC TB cases.

Culture-positive TB cases in New York City steadily declined from 950 cases in 2001 to 649 in 2008 (R² = 0.9). During this time, a total of 6,260 culture-positive TB cases were reported, 5,670 (91%) of which were due to M. tuberculosis infection and had genotype and DST results available (Fig. 1). Of these, 145 (2.6%) cases involved infection with PZA^r M. tuberculosis. The annual M. tuberculosis PZA^r prevalence was between 1.6% and 3.6%, resulting in an average of 18 cases per year. The burden of PZA^r fluctuated considerably, peaking in 2002, 2005, and 2008 (Fig. 2). From 2001 to 2005, MDR-PZA^r (resistance to PZA, INH, and RIF only) accounted for an average of 60% of all PZA^r in NYC, whereas mono-PZA^r accounted for an average of 24%. Between 2006 and 2008, there was a shift in the relative proportions of mono-PZA^r and MDR-PZA^r. While the overall proportion of PZA^r remained relatively constant during the study period (R² = 0.2), by 2008, mono-PZA^r accounted for 60% of PZA^r in NYC, whereas MDR-PZA^r accounted for less than 30% of PZA^r.

FIG 2.

Distribution of PZA^r, mono-PZA^r, and MDR-PZA^r TB in NYC from 2001 to 2008.

The univariate analysis of epidemiologic risk factors associated with PZA^r is presented in Table 1. Predictors of PZA^r included AFB smear positivity, history of TB, death, and strain clustering. Among drug-resistant strains (n = 915), MDR was strongly associated with PZA^r (OR = 7.25; 95% confidence interval [CI] = 4.73 to 11.13) compared to any monoresistance. In the multivariate analysis (Table 2), AFB smear positivity (OR = 2.02; 95% CI = 1.19 to 3.43) and clustering (OR = 2.73; 95% CI = 1.71 to 4.36) maintained significant associations with PZA^r TB caused by PZA^r M. tuberculosis.

TABLE 1.

Epidemiologic characteristics associated with pyrazinamide resistance

Characteristic	PZAr (n = 145)	PZAs (n = 5,525)	Unadjusted OR	95% CI	P valuea
Median age, yr (IQR)	40 (30–50)	42 (30–57)	0.99	0.98–1.00	0.0118
Sexb
M	89 (61.38%)	3,469 (62.79%)	0.94	0.67–1.32	0.7292
F	56 (38.62%)	2,056 (37.21%)
U.S. bornc
Yes	47 (32.64%)	1,598 (29.03%)	1.18	0.83–1.69	0.3472
No	97 (67.36%)	3,906 (70.97%)
History of TB
Yes	8 (5.52%)	139 (2.52%)	2.26	1.09–4.71	0.0248
No	137 (94.48%)	5,386 (97.48%)
History of LTBI
Yes	5 (3.45%)	320 (5.79%)	0.58	0.24–1.43	0.2308
No	140 (96.55%)	5,205 (94.21%)
History of homelessness
Yes	18 (12.41%)	518 (9.38%)	1.37	0.83–2.26	0.2171
No	127 (87.59%)	5,007 (90.62%)
History of substance abuse
Yes	19 (13.29%)	494 (9.15%)	1.52	0.93–2.49	0.0924
No	124 (86.71%)	4,903 (90.85%)
History of alcohol abuse
Yes	24 (16.78%)	875 (16.20%)	1.04	0.67–1.63	0.8513
No	119 (83.22%)	4,527 (83.80%)
History of Rikers treatment
Yes	3 (2.07%)	106 (1.92%)	1.08	0.34–3.44	0.8964
No	142 (97.93%)	5,419 (98.08%)
HIV serostatus
Positive	31 (21.38%)	825 (14.93%)	1.39	0.92–2.11	0.1226
Negative	85 (58.62%)	3,145 (56.92%)		Reference
Unknown	29 (20.00%)	1,555 (28.14%)	0.69	0.45–1.06	0.0877
TB infection site
Pulmonary	114 (78.62%)	3,882 (70.26%)		Reference
Extrapulmonary	14 (9.66%)	986 (17.85%)	0.48	0.28–0.85	0.0109
Both	17 (11.72%)	657 (11.89%)	0.88	0.53–1.48	0.6309
Respiratory AFB smear status
Positive	104 (74.82%)	2,967 (57.77%)	2.17	1.48–3.2	<0.0001
Negative	35 (25.18%)	2,169 (42.23%)
Abnormal chest X-ray
Yes	59 (93.65%)	2,091 (88.87%)	1.85	0.67–5.13	0.2311
No	4 (6.35%)	262 (11.13%)
Final culture conversion
Yes	106 (86.18%)	3,563 (85.92%)	1.02	0.61–1.72	0.9345
No	17 (13.82%)	584 (14.08%)
Any cavitation
Yes	32 (25.00%)	1,010 (21.62%)	1.21	0.81–1.81	0.3606
No	96 (75.00%)	3,661 (78.38%)
Any deathd
Yes	23 (17.83%)	582 (11.18%)	1.72	1.09–2.70	0.0187
No	106 (82.17%)	4,623 (88.82%)
Drug resistancee
Mono	50 (34.48%)	461 (59.87%)		Reference
Poly	25 (17.24%)	220 (28.57%)	1.05	0.63–1.74	0.8568
MDR	70 (48.28%)	89 (11.56%)	7.25	4.73–11.13	<0.0001
Molecular epidemiology
Clustered
Yes	83 (57.24%)	2,414 (43.69%)	1.73	1.24–2.41	0.0012
No	62 (42.76%)	3,111 (56.31%)
Phylogenetic lineage
1	18 (12.50%)	490 (8.90%)	1.91	1.13–3.22	0.0159
2	44 (30.56%)	878 (15.95%)	2.60	1.78–3.81	<0.0001
3	9 (6.25%)	347 (6.31%)	0.67	0.67–2.71	0.4063
4	73 (50.69%)	3,788 (68.84%)		Reference

^aCalculated using chi-square.

^bM, male; F, female.

^cIncludes Puerto Rico.

^dExcludes patients who refused treatment or who were lost to follow-up.

^eMono, resistance to any 1 drug only (isoniazid, rifampin, ethambutol, or second-line [fluoroquinolone, kanamycin, capreomycin, amikacin, or ethionamide)]; Poly, resistance to any combination of drugs excluding MDR; MDR, multidrug resistant only.

TABLE 2.

Adjusted^a epidemiologic characteristics of pyrazinamide resistance

Characteristic	OR	95% CI	P value
Age (yr)	0.99	0.98–1.01	0.2007
HIV positivity	1.39	0.80–2.43	0.2467
TB infection site
Extrapulmonary vs pulmonary	0.84	0.33–2.14	0.7192
Both vs pulmonary	0.81	0.42–1.55	0.5272
AFB smear positivity	2.02	1.19–3.43	0.0094
History of TB	0.93	0.28–3.08	0.9032
History of substance abuse	1.29	0.70–2.40	0.4164
Any death	1.77	0.89–3.52	0.1033
Clustered	2.73	1.71–4.36	<0.0001
Lineage 1 vs 4	3.45	1.72–6.95	0.0005
Lineage 2 vs 4	5.01	3.13–8.03	<0.0001
Lineage 3 vs 4	3.30	1.33–8.19	0.0101

^aAdjusted for known TB risk factors, including age, HIV, and any univariate variable with a P value of <0.2.

PZA resistance trends and risk factors among the NYC MDR population.

PZA resistance was high among patients with MDR TB, accounting for 44% (70/159) of all MDR cases, while only 1.4% (75/5511) of the non-MDR population was PZA^r. Table 3 shows the odds ratios for PZA^r according to clinical and demographic characteristics, stratified by patients with MDR and non-MDR TB. Factors independently associated with PZA^r among patients with MDR TB in the stratified analysis were EMB-SLD resistance and AFB smear positivity. AFB smear positivity was the only significant characteristic among patients with non-MDR TB (Table 3). In the multivariate MDR TB analysis, only EMB-SLD resistance (OR = 3.48; 95% CI = 1.57 to 7.69) maintained significance (Table 4), while a history of latent M. tuberculosis infection (LTBI) trended toward significance (OR = 0.12; 95% CI = 0.01 to 1.07).

TABLE 3.

Epidemiologic characteristics associated with pyrazinamide resistance stratified by MDR

Characteristic	Non-MDR TB Cases					MDR TB cases					Breslow-Day test for homogeneity
	PZAr (n = 75)	PZAs (n = 5,436)	OR	95% CI	P value	PZAr (n = 70)	PZAs (n = 89)	OR	95% CI	P value	χ2	P value
Median age, yr (IQR)	40 (31–51)	42 (30–57)	0.99	0.98–1.01	0.262	40 (28–49)	42 (29–53)	0.98	0.96–1.01	0.120	NAa
Sexb
M	49 (65.33%)	3,425 (63.01%)	1.11	0.69–1.79	0.678	40 (57.14%)	44 (49.44%)	1.36	0.73–2.56	0.334	0.2679	0.6047
F	26 (34.67%)	2,011 (36.99%)				30 (42.86%)	45 (50.56%)
U.S. born
Yes	17 (22.67%)	1,563 (28.86%)	0.72	0.42–1.24	0.239	30 (43.48%)	35 (39.33%)	1.19	0.63–2.25	0.599	1.3523	0.2449
No	58 (77.33%)	3,852 (71.14%)				39 (56.52%)	54 (60.67%)
History of TB
Yes	1 (1.33%)	132 (2.43%)	0.54	0.07–3.94	0.539	7 (10.00%)	7 (7.87%)	1.30	0.43–3.9	0.637	0.5967	0.4399
No	74 (98.67%)	5,304 (97.57%)				63 (90.00%)	82 (92.13%)
History of LTBI
Yes	3 (4.00%)	313 (5.76%)	0.68	0.21–2.18	0.516	2 (2.86%)	7 (7.87%)	0.34	0.07–1.71	0.175	0.4622	0.4966
No	72 (96.00%)	5,123 (94.24%)				68 (97.14%)	82 (92.13%)
History of homelessness
Yes	6 (8.00%)	506 (9.31%)	0.85	0.37–1.96	0.698	12 (17.14%)	12 (13.48%)	1.33	0.56–3.17	0.522	0.533	0.4654
No	69 (92.00%)	4,930 (90.69%)				58 (82.86%)	77 (86.52%)
History of substance abuse
Yes	5 (6.67%)	481 (9.06%)	0.72	0.29–1.79	0.473	14 (20.59%)	13 (15.12%)	1.46	0.63–3.35	0.375	1.2787	0.2581
No	70 (93.33%)	4,830 (90.94%)				54 (79.41%)	73 (84.88%)
History of alcohol abuse
Yes	11 (14.67%)	853 (16.05%)	0.90	0.47–1.71	0.747	13 (19.12%)	22 (25.58%)	0.69	0.32–1.49	0.273	0.6013	0.737
No	64 (85.33%)	4,463 (83.95%)				55 (80.88%)	64 (74.42%)
History of Rikers treatment
Yes	0 (0.00%)	104 (1.91%)				3 (4.29%)	2 (2.25%)	1.95	0.32–11.99	0.465	2.2788	0.1312
No	75 (100.00%)	5,332 (98.09%)				67 (95.71%)	87 (97.75%)
HIV serostatus
Positive	11 (14.67%)	803 (14.77%)	0.94	0.49–1.83	0.859	20 (28.57%)	22 (24.72%)	0.18	0.57–2.46	0.655	0.4642c	0.4957c
Negative	45 (60.00%)	3,093 (56.90%)		Reference		40 (57.14%)	52 (58.43%)		Reference
Unknown	19 (25.33%)	1,540 (28.33%)	0.85	0.49–1.46	0.549	10 (14.29%)	15 (16.85%)	0.87	0.35–2.13	0.755
TB infection site
Pulmonary	55 (73.33%)	3,816 (70.20%)		Reference		59 (84.29%)	66 (74.16%)		Reference		1.4359c	0.2308c
Extrapulmonary	8 (10.67%)	976 (17.95%)	0.57	0.27–1.20	0.138	5 (7.14%)	10 (11.24%)	0.67	0.23–1.96	0.466
Both	12 (16.00%)	644 (11.85%)	1.29	0.69–2.43	0.424	5 (7.04%)	13 (14.61%)	0.43	0.15–1.28	0.129
Respiratory AFB smear status
Positive	50 (70.42%)	2914 (57.74%)	1.74	1.04–2.91	0.032	54 (79.41%)	53 (59.55%)	2.62	1.27–5.41	0.008	0.8126	0.3673
Negative	21 (29.58%)	2,133 (42.26%)				14 (20.59%)	36 (40.45%)
Abnormal chest X-ray
Yes	31 (91.18%)	2,056 (88.81%)	1.30	0.4–4.29	0.664	28 (96.55%)	35 (92.11%)	2.40	0.24–24.35	0.447	0.2147	0.6431
No	3 (8.82%)	259 (11.19%)				1 (3.45%)	3 (7.89%)
Final culture conversion
Yes	54 (88.52%)	3,498 (85.90%)	1.27	0.57–2.8	0.559	52 (83.87%)	65 (86.67%)	0.80	0.31–2.07	0.645	0.532	0.4658
No	7 (11.48%)	574 (14.10%)				10 (16.13%)	10 (13.33%)
Any cavitation
Yes	13 (19.70%)	992 (21.60%)	0.89	0.48–1.64	0.709	19 (30.65%)	18 (23.08%)	1.47	0.69–3.13	0.313	1.0368	0.3086
No	53 (80.30%)	3,601 (78.40%)				43 (69.35%)	60 (76.92%)
Any death
Yes	9 (12.50%)	568 (11.07%)	1.15	0.57–2.32	0.701	14 (24.56%)	14 (19.18%)	1.37	0.60–3.15	0.459	0.1021	0.7493
No	63 (87.50%)	4,564 (88.93%)				43 (75.44%)	59 (80.82%)
Drug resistanced
INH
Yes	12 (40.00%)	345 (57.02%)	0.50	0.24–1.06	0.089
No	18 (60.00%)	260 (42.98%)
INH-EMB
Yes	3 (10.00%)	28 (4.63%)	2.29	0.65–8.00	0.175
No	27 (90.00%)	577 (95.37%)
INH-EMB-SLD
Yes	1 (3.33%)	6 (0.99%)	3.44	0.40–29.54	0.289
No	29 (96.67%)	599 (99.01%)
EMB
Yes						22 (32.35%)	22 (25.00%)	1.43	0.71–2.89	0.371
No						46 (67.65%)	66 (75.00%)
EMB-SLD
Yes						36 (52.94%)	23 (26.14%)	3.18	1.62–6.23	<0.0001
No						32 (47.06%)	65 (73.86%)
Molecular epidemiology
Clustered
Yes	38 (50.67%)	2,363 (43.47%)	1.34	0.85–2.11	0.212	45 (64.29%)	51 (57.30%)	1.34	0.7–2.56	0.372	0.0001	0.9918
No	37 (49.33%)	3,073 (56.53%)				25 (35.71%)	38 (42.70%)
Phylogenetic lineage
1	17 (22.97%)	488 (9.01%)	3.10	1.75–5.49	<0.0001	1 (1.43%)	2 (2.25%)	0.79	0.07–9.09	0.850	10.8398c	0.0010c
2	9 (12.16%)	846 (15.63%)	0.95	0.46–1.95	0.883	35 (50.00%)	32 (35.96%)	1.73	0.90–3.34	0.103
3	6 (8.11%)	341 (6.30%)	1.57	0.67–3.71	0.308	3 (4.29%)	6 (6.74%)	0.79	0.18–3.39	0.7515
4	42 (56.76%)	3,739 (69.06%)		Reference		31 (44.29%)	49 (55.06%)		Reference

^aNA, not applicable.

^bM, male; F, female.

^cCochran-Mantel-Haenszel nonzero correlation statistic.

^dDrug resistance analysis excludes drug-susceptible M. tuberculosis strains and is independent of PZA^r. The INH category excludes strains with additional RIF^r, EMB^r, or SLD^r. INH-EMB excludes strains with additional RIF^r or SLD^r. INH-EMB-SLD excludes strains with additional RIF^r. The EMB category excludes strains with additional SLD^r.

TABLE 4.

Adjusted^a epidemiologic characteristics of pyrazinamide resistance among MDR and non-MDR TB cases

Characteristic	OR	95% CI	P value
MDR
Age (yr%)	0.98	0.95–1.01	0.1518
HIV positivity	1.38	0.57–3.37	0.4724
TB infection site
Extrapulmonary vs pulmonary	0.83	0.16–4.23	0.8211
Both vs pulmonary	0.44	0.12–1.60	0.2149
AFB smear	1.79	0.66–4.87	0.2556
History of LTBI	0.12	0.01–1.07	0.0572
EMB-SLD	3.48	1.57–7.69	0.0021
Non-MDR
Age (yr)	0.99	0.98–1.01	0.5188
HIV positivity	0.95	0.46–1.98	0.8891
TB infection site
Extrapulmonary vs pulmonary	0.89	0.28–2.81	0.8379
Both vs pulmonary	1.37	0.63–2.97	0.4256
AFB smear	1.82	0.93–3.58	0.0820
Lineage 1	3.27	1.64–6.51	0.0008

^aAdjusted for known TB risk factors, including age, HIV, and any univariate variable with a P value of <0.2.

Molecular epidemiology of PZA resistance.

Overall, PZA^r isolates accounted for 3.5% of lineage 1 (18/508), 4.8% of lineage 2 (44/922), 2.5% of lineage 3 (9/356), and 1.9% of lineage 4 (73/3861). Significant lineage-specific PZA^r associations were identified in the univariate analysis, where lineage 1 and lineage 2 M. tuberculosis isolates were approximately twice as likely to be PZA^r as lineage 4 isolates (Table 1). These associations maintained significance in the multivariate model and indicate a phylogenetic M. tuberculosis lineage association with PZA^r: lineage 1 (OR = 3.45; 95% CI = 1.72 to 6.95), lineage 2 (OR = 5.01; 95% CI = 3.13 to 8.03), and lineage 3 (OR = 3.30; 95% CI = 1.33 to 8.19) compared to lineage 4 (Table 2). Among the non-MDR population, phylogenetic lineage 1 was the only PZA^r predictor that maintained significance in the multivariate model (OR = 3.27; 95% CI = 1.64 to 6.51) (Table 4).

MDR TB case-control study.

To examine PZA^r risk factors among the 159 MDR cases, we performed a case-control study. Sixty percent of MDR isolates (96/159) were considered genotypically clustered (identical IS6110-RFLPs and spoligotypes), with clusters ranging in size from 2 to 21 members. In contrast, only 43.6% (2,401/5,511) of the non-MDR strains were genotypically clustered (Table 3). Forty-four percent of MDR isolates (70/159) were PZA^r based on DST, 64% (45/70) of which belonged to a genotypic cluster. The genetic markers pncA, katG, and rpoB were used to further resolve clustering within the MDR population. Sequence data for the fabG1 promoter did not provide additional cluster resolution and are not shown.

Complete sequence data (katG, rpoB, and pncA) were available for 88% (140/159) of MDR M. tuberculosis isolates (Fig. 1), while pncA sequence data were available for 143 isolates. Analysis of the pncA CDS-promoter region of MDR isolates identified 83 (58%) mutants. In total, there were 37 unique pncA mutations, which included insertions, deletions, and nonsynonymous substitutions. Some discordances between PZA^r determined by DST and the presence of pncA mutations (PZA^r-pncA) were observed (109/143; kappa = 0.54). However, the majority of these were accounted for by a single family of Beijing strains (termed W; n = 54) (29/54; kappa = 0.17), many of which carried a specific pncA mutation known to exhibit discordance with phenotypic PZA^r (16, 28), as discussed below. Concordance between the pncA sequence and PZA DST results was significantly higher when W strains were excluded and the remaining 89 strains were evaluated (80/89; kappa = 0.79).

Table 5 shows the resolution of genotypic clusters, using DNA sequence data to distinguish primary transmission from independent acquisition of MDR TB. Identical sequences for katG, rpoB, and pncA confirmed the genotyping assignment of seven clusters (n = 38). Conversely, diverse pncA mutations were identified within four genotypic clusters possessing identical IS6110-RFLPs, spoligotypes, and mutations in katG and rpoB, while a fifth cluster with identical katG and pncA sequences (P strain; spoligotype S00086) was resolved by different rpoB mutations.

TABLE 5.

MDR cluster analysis using katG, rpoB, and pncA sequence data

Infection	Genotypic cluster			katG-rpoB		pncA		Reference
	RFLP (spoligotype)	Lineage (spoligotype)	Count	Cluster	Count	Cluster	Count
Primary transmission	W (S00034)	2 (Beijing, ST523, ST623)	22	S315T_H526Y	21a	ACC(T)47GCC(A)	21a	28, 37
	W665 (S00034)	2 (Beijing, ST523, ST623)	3	S315T_H526Y	3	ACC(T)47GCC(A)	3
	P1 (S00086)	4 (X, Harlem, LAM, Uganda)	2	S315T_H526Y	2	CTG(L)85CCG(P)	2	40, 41
	BW900 (S00005)	4 (X, Harlem, LAM, Uganda)	5	WT_H526D	5	WTb	5
	C (S00030)	4 (X, Harlem, LAM, Uganda)	2	WT_S531L	2	WT	2	38
	DK22 (S00245)	4 (X, Harlem, LAM, Uganda)	2	S315T_L511P	2	WT	2
	BW230 (S00241)	4 (X, Harlem, LAM, Uganda)	2	S315T_S531L	2	WT	2	39
Independent acquisition	W283 (S00034)	2 (Beijing, ST523, ST623)	3	S315T_S531L	2	290 Δ G	2
				S315_WT	1	AGG(R)154GGG(G)	1
	W148 (S00034)	2 (Beijing, ST523, ST623)	3	S315T_S531L	3	WT	1	54
						CAT(H)71CGT(R)	1
						GGA(G)108CGA(R)	1
	W1 (S00034)	2 (Beijing, ST523, ST623)	2	S315T_H526Y	2	ACC(T)47GCC(A)	1	28, 43
						ACC(T)47AGC(S)	1
	W33 (S00034)	2 (Beijing, ST523, ST623)	2	S315T_H526Y	2	ACC(T)47GCC(A)	1
						ACC(T)47AGC(S)	1
	P (S00086)	4 (X, Harlem, LAM, Uganda)	3	S315T_D516V	1	CTG(L)85CCG(P)	3	40, 41
				S315T_H526Y	1
				S315T_S531L	1
Both	H (S00009)	4 (X, Harlem, LAM, Uganda)	11	S315T_S531L	9	70 Δ G	10	40, 41
				WT_WT	1
				WT_L511P	1	WT	1
Unresolved	W738 (S00034)	2 (Beijing, ST523, ST623)	3	S315T_WT	2	TCG(S)67CCG(P)	1a
				S315T_S531L	1	WT	1
	AB (S00145)	4 (X, Harlem, LAM, Uganda)	2	S315T_V146F	2	WT	1a	40
	W269 (S00034)	2 (Beijing, ST523, ST623)	2	WT_S531L	1a	ACG(T)142ATG(M)	1a
	GD265 (S00474)	3 (CAS)	2	S315T_S531L	1a	CAG(Q)141CCG(P)c	2

^aOne bad sequence.

^bWT, wild type.

^cAlso contains a lineage-specific synonymous mutation (TCC[S]65TCT[S]).

DISCUSSION

The prevalence of PZA^r has been reported to range from 0.8 to 10% among patients with non-MDR TB and from 10 to 85% among patients with MDR TB worldwide (4, 29, 30). During the study period, TB caused by PZA^r M. tuberculosis in NYC was 50% higher than the national average and 20% higher among patients with MDR TB (31). Moreover, the MDR TB burden in NYC was 2-fold higher than national estimates. Thus, the high prevalence of MDR TB could be an explanation for the high proportion of PZA^r we observed. However, the MDR TB incidence in NYC declined over this period (32), while the incidence of PZA^r TB cases increased, indicating that additional factors were contributing to PZA^r in this population. To explain the national increase in PZA^r, Kurbatova and colleagues proposed that a higher proportion of foreign-born TB patients may reflect international programmatic variation in TB treatment protocols, leading to increased levels of PZA^r TB imported to the United States (31). While we observed high proportions of foreign-born TB patients annually (70 to 80%), we did not find being born outside the United States to be a statistically significant risk factor for PZA^r in NYC. In contrast to our findings in NYC, a recent analysis of PZA^r TB trends over 2 decades in San Francisco found rates similar to the national average and no significant association between PZA^r and MDR TB (33).

Our study found AFB smear positivity, clustering, and death to be independently associated with PZA^r. Together, these risk factors suggest that patients infected with PZA^r strains were infectious, transmitting, and not responding well to treatment. A history of TB was also independently associated with PZA^r, perhaps suggesting that a proportion of patients experienced relapses of drug-resistant TB, though relapse data were not available for this analysis (34, 35). Given the clinical importance of MDR TB and the high proportion of PZA^r among these patients, we sought to determine PZA^r risk factors in the MDR TB population. Concurrent resistance to EMB and SLDs was the only PZA^r risk factor that maintained statistical significance in the adjusted case-control MDR model. While a strong association between MDR-PZA^r and EMB^r has been associated with the inappropriate use of standard short-course therapy in patients with MDR TB (36), we do not have sufficient evidence here to address this question. A larger proportion of MDR-PZA^r among clustered versus nonclustered strains suggests that primary transmission was responsible for more PZA^r than acquired resistance within the MDR TB population.

Based on our molecular examination of MDR strains, we were able to further refine our cluster analysis. In particular, we identified MDR TB clusters with IS6110-RFLP; spoligotype; and katG, rpoB, and pncA mutations identical to those previously described in a number of historic NYC drug-resistant outbreak strains (28, 37), e.g., strain C (S00030) (38), strain H (S00009), strain BW (S00241) (39), and strain P (S00086) (40, 41). The sequence-based confirmation of MDR genotypic clusters strongly suggests primary transmission and shows evidence that these historic strains were continuing to circulate and reactivate within NYC during the study period. For example, a 21-member cluster of the W (S00034; Beijing family) MDR-PZA^r strain contained genetic markers identical to those of a highly clonal strain that was responsible for a large NYC MDR outbreak in the early 1990s (28, 37). Smaller clusters of historic W variant strains previously described (28, 37, 42, 43) were also identified. Our data suggest that PZA^r also appeared independently in MDR TB (i.e., not as a result of primary transmission) (44). For example, the P strain was likely transmitted as a katG-pncA mutant prior to becoming MDR, which indicates PZA^r was at times acquired prior to MDR development, as seen in other strains (45, 46).

In addition to clustering, further microbial phylogenetic lineage effects were observed. In the general NYC TB population, PZA^r was associated with lineage 1 (East African/Indo-Oceanic), lineage 2 (East Asian/Beijing), and lineage 3 (East Africa/Central Asia) compared to lineage 4 (Euro-American). PZA^r associations with lineages 1 and 2 were consistent with lineage effects reported in the national study conducted by the CDC (31). Our data indicate that PZA^r among patients with MDR TB was associated with lineage 2, while PZA^r was associated with lineage 1 among patients with non-MDR TB. Phylogenetic M. tuberculosis lineages are strongly associated with geographic locations. Therefore, the observed phylogenetic associations in this study are curious and may be a proxy for the importation of PZA^r from specific non-U.S. locations. Additional studies would be needed to examine these microbial associations in light of social networks and neighborhood level effects.

Limitations of this study include moderate amounts of HIV data (70%), which is a well-established risk factor for drug-resistant TB. In addition, many demographic variables were self-reported, including a history of TB/LTBI, alcohol or substance abuse, and homelessness, which may have been subject to information bias and misclassification, though it is likely nondifferential. Furthermore, due to the technical complications of PZA-DST (8, 18), our findings may have been subject to some PZA^r misclassification. However, pncA sequence data have been shown to be a useful tool to confirm PZA^r-DST results (47). The most common pncA-PZA^r-DST inconsistency we observed was that of W strains with a pncA mutation in codon 47 (ACC [Thr]→GCC [Ala]). This specific mutation has been shown to correlate poorly with PZA-DST in several previous studies (12, 16, 28, 48, 49), suggesting that the mutation confers borderline resistance at PZA concentrations used routinely in PZA-DST (16). Also, the molecular data suggest that IS6110-RFLP analysis and spoligotyping may have slightly overestimated the extent of clustering and primary transmission.

Despite a steady decline in the total number of TB cases in NYC from 2001 to 2008, the incidence of PZA^r increased. Patients with PZA^r TB were more infectious and actively transmitting and had poor clinical outcomes compared to patients with TB caused by PZA^s M. tuberculosis. Among the MDR TB population, PZA^r was acquired somewhat more frequently via primary transmission than independently. In addition, concurrent EMB-SLD resistance was the only risk factor for PZA^r among patients with MDR TB. These observations have important clinical and public health implications for control of drug-resistant TB. Specifically, the strength of microbial risk factors for PZA^r highlights the importance of routine PZA-DST, genotyping, and confirmatory sequence analysis for ensuring appropriate drug therapy and disrupting transmission. Finally, as clinical trials of new regimens to shorten treatment of drug-susceptible and MDR TB are expanding, these results support the need to consider PZA^r in trial design (50–53).