Fluoroquinolone Resistance Mutation Detection Is Equivalent to Culture-Based Drug Sensitivity Testing for Predicting Multidrug-Resistant Tuberculosis Treatment Outcome: A Retrospective Cohort Study Free

Abstract

Background

Molecular diagnostics that rapidly and accurately predict fluoroquinolone (FQ) resistance promise to improve treatment outcomes for individuals with multidrug-resistant (MDR) tuberculosis (TB). Mutations in the gyr genes, though, can cause variable levels of in vitro FQ resistance, and some in vitro resistance remains unexplained by gyr mutations alone, but the implications of these discrepancies for treatment outcome are unknown.

Methods

We performed a retrospective cohort study of 172 subjects with MDR/extensively drug-resistant TB subjects and sequenced the full gyrA and gyrB open reading frames in their respective sputum TB isolates. The gyr mutations were classified into 2 categories: a set of mutations that encode high-level FQ resistance and a second set that encodes intermediate resistance levels. We constructed a Cox proportional model to assess the effect of the gyr mutation type on the time to death or treatment failure and compared this with in vitro FQ resistance, controlling for host and treatment factors.

Results

Controlling for other host and treatment factors and compared with patients with isolates without gyr resistance mutations, “high-level” gyr mutations significantly predict poor treatment outcomes with a hazard ratio of 2.6 (1.2–5.6). We observed a hazard of death and treatment failure with “intermediate-level” gyr mutations of 1.3 (0.6–3.1), which did not reach statistical significance. The gyr mutations were not different than culture-based FQ drug susceptibility testing in predicting the hazard of death or treatment failure and may be superior.

Conclusions

FQ molecular-based diagnostic tests may better predict treatment response than traditional drug susceptibility testing and open avenues for personalizing TB therapy.

The use of molecular-based diagnostics for the detection of tuberculosis (TB) drug resistance is gaining traction globally [1]. Molecular diagnostics detect drug resistance genetic mutations in 1 or more Mycobacterium tuberculosis (MTB) genes [2, 3]. Although the interpretation of the detected mutations in a particular MTB isolate is presently focused on a binary “resistant” or “sensitive” readout, it is well recognized that several genetic mutations can cause variable levels of drug resistance that may not be consistently classified as either resistant or sensitive [4, 5]. The clinical implications of such “intermediate” mutations are unknown and necessitate a closer look at treatment outcomes in patients whose isolates harbor these mutations. As we move beyond molecular diagnostics that focus on a handful of genes to whole-genome sequencing to detect resistance [6], the number of detected variants will grow, requiring an increasingly sophisticated and nuanced understanding of the impact of genotypic drug resistance on treatment response.

Despite the wide use of molecular diagnostics, several open questions remain that limit their ability to fully replace culture-based drug susceptibility testing (DST), such as a sensitivity gap that is consistently found when molecular diagnostics are compared with culture-based tests for several second-line MTB drugs [7, 8]. For the fluoroquinolones (FQs), as many as 31% of moxifloxacin-resistant isolates have been reported to not harbor any resistance mutations [4, 7]. Culture-based DST is widely considered the gold standard diagnostic for MTB drug resistance, although prospective data are limited on how well they predict clinical effectiveness [9]. In addition, there are well-known issues with accuracy and reproducibility [10]. The ultimate goal of DST is to distinguish between MTB isolates that will or will not respond to a drug used in treatment. The determination of resistance, when reliable, should lead to the choice of an alternative drug. Here we compare molecular diagnostics with culture-based DST in predicting overall treatment outcome.

FQs are considered critical to effective multidrug-resistant (MDR) and extensively drug-resistant (XDR) TB treatment [11, 12]. FQs exert their antitubercular activity by inhibiting DNA gyrase, limiting DNA replication and transcription. MTB gyrase consists of 2 subunits encoded by the genes gyrA and gyrB [2, 13]; mutations in these genes are the main cause of FQ resistance in MTB [14]. Here we performed gyr sequencing and analysis on strains from patients from a previously described cohort study of MDR-/XDR-TB treatment outcome [11] to determine (1) the significance of “intermediate” gyr mutations on treatment response and compare this with treatment response for “high” resistance level gyr mutations; and (2) how gyr mutation detection compares with FQ culture-based DST in predicting treatment response controlling for host and other drug resistance variables.

MATERIALS AND METHODS

Study Population

We studied a subset of patients from a retrospective cohort study (N = 669) of individualized MDR-/XDR-TB outpatient treatment in Lima, Peru, between 1 February 1997 and 31 July 2003 [11, 15–18]. We excluded subjects with incomplete treatment regimen composition data, and those not treated with FQs. Baseline DST was performed at the Massachusetts State Laboratory using the indirect proportion method on 7H10 media for all drugs with the exception of pyrazinamide, which was tested using BACTEC, as described previously [18]. DST to the first-line drugs (isoniazid, rifampin, pyrazinamide, ethambutol, and streptomycin) was routinely performed. In >75% of subjects, DST was also performed to the following second-line drugs: amikacin, capreomycin, cycloserine, ethionamide, kanamycin, and para-aminosalicylic acid; ciprofloxacin or ofloxacin; and either gatifloxacin, levofloxacin, or moxifloxacin. Less than 25% of subjects had isolates tested to any of the following agents: amoxicillin-clavulanic acid, clofazimine, clarithromycin, or rifabutin. Regimens were constructed based on DST results and prior treatment according to previously described principles [19]. Data were collected and recorded in a Web-based electronic medical record during treatment [20]. A standardized paper records abstraction was conducted to complete the dataset.

Strain Selection, Sequencing, and Minimum Inhibitory Concentration Testing

We selected MTB isolates with DST results for an FQ and an injectable (amikacin, capreomycin, or kanamycin) yielding confirmed XDR-TB, pre–XDR-TB, or MDR-TB with additional resistance to at least 1 tested agent (n = 176). We excluded subjects if the time from when an available archived TB isolate was collected to the start of MDR-TB treatment exceeded 18 months (n = 5). If multiple TB isolates were archived for a given subject, we chose the isolate obtained at the time closest to the MDR-TB treatment start, except when the isolates differed in their FQ resistance phenotype, in which case we chose the more highly resistant isolate. The 171 isolates underwent gyrA and gyrB gene sequencing using molecular inversion probes (MIPs) as part of a separate study [21]. Of the 171 archived isolates, 112 also underwent moxifloxacin minimum inhibitory concentration (MIC) testing at the Massachusetts State Laboratory between 2012 and 2013 using the indirect proportion method on 7H10 agar. Concentrations tested were 0.125, 0.25, 0.5, 1.0, 4.0, and 8.0 mg/L; results were quality controlled as previously described [4].

Patient Variables and Outcomes

Monthly FQ exposure was classified as late-generation FQ administration (levofloxacin, gatifloxacin, or moxifloxacin) for at least 75% of days in a month. Reference exposure was that to an early-generation FQ (ciprofloxacin, ofloxacin, or sparfloxacin).

Exposure to an “otherwise adequate” regimen in the intensive phase was defined as administration of at least 4 agents that were likely efficacious—including 1 of the injectable agents (kanamycin, capreomycin, amikacin), for at least 6 months after culture conversion. This distinguished likely activity of the companion drugs from that of the FQs; the latter, as represented by the presence of FQ resistance mutations, was the exposure of interest. The continuation phase of an otherwise adequate regimen contained at least 3 likely efficacious oral drugs [19]. An agent (excluding FQs) was considered likely efficacious if either (1) all in vitro DST prior to the start of this regimen confirmed susceptibility to the agent used; or, (2) in vitro DST to the agent was not available and the subject had not received the agent for > 1 month prior to individualized treatment. If at least 75% of regimen days in a month met the adequate definition, that month was coded as “exposed.”

We also created a variable reflecting previous treatment. This was abstracted from the record and dichotomized into (1) less prior treatment, defined as ≤2 prior regimens, not including the standardized regimen for MDR-TB; and (2) more prior treatment, defined as >2 regimens or prior treatment with the standardized regimen for MDR-TB.

Other covariates collected included demographics (age, sex, location of residence, treatment time period). Indicators of disease severity were collected including: hematocrit, nutritional status (body mass index [BMI] and clinical diagnosis of malnutrition), presence of extrapulmonary TB, respiratory difficulty (dyspnea or resting respiratory rate >26 breaths/minute), tachycardia (heart rate >100 beats/minute), and cavitary and bilateral disease on chest radiography. Comorbidities and risk factors—substance and tobacco use, human immunodeficiency virus infection, diabetes mellitus, cardiovascular and renal disease, and psychiatric and seizure disorders—were also recorded.

Treatment outcomes (cure, completion, failure, death) were defined according to definitions in place at the time and as previously described [22]. There was no systematic follow-up after treatment completion to assess relapse. The primary study endpoint was time from initiation of the individualized regimen to death from any cause or treatment failure. Data were censored when an outcome other than death or treatment failure was recorded.

Sequencing and gyr Mutation Variables

Sequencing was performed between 2010 and 2012 using archived patient samples. The information from the sequencing was not used to guide clinical care. In brief, the gyrA and gyrB gene sequences were captured using MIPs [21] designed to cover both DNA strands of the open reading frames, promoter regions, and 100 flanking bases on either side of the selected genes. We called mutations using Bowtie [23] 0.12.7/ SAMtools [24] 0.1.18. We considered any synonymous mutations and the following polymorphisms in gyrA (E21Q, T80A, S95T, G247S, and G668D) neutral variants, as they are recognized to be neutral variants or are known markers of genetic lineage [6, 7, 25]. Other gyr mutations detected (gyrA: A288D and R128K; gyrB: S486Y) were also considered to be neutral variants as they occurred outside of the gene resistance determining region [4].

We coded the gyr resistance mutation variable into 2 dichotomous indicator variables: The first indicated the presence of the gyrA mutations A90V and D94A, which are associated with intermediate MICs to moxifloxacin; and the second indicated gyr mutations that were associated with high MICs to moxifloxacin (gyrA: D94G, D94N, D94Y, D89N, and S91P; gyrB: T539A, N538D, and E540D) [4, 26].

Analysis

We modeled the association between the gyr mutations and the hazard of death or treatment failure using Cox proportional hazards analysis. In addition to treatment variables, possible confounders of treatment response (including prior treatment, sex, age, and extent of disease on chest radiography identified as such based on previous evidence [11]) were evaluated for an association with hazard of death or treatment failure. Those variables that predicted the outcome at a P value ≤.20 were considered candidates for the multivariate model and retained if inclusion of that variable changed the effect estimate of gyr mutation in the model by ≥10%. In addition, we tested for interaction between the gyr variables and the late-generation FQ treatment variable (2 interaction terms). Finally, we compared the strength of the association with clinical outcomes across the indicators of resistance. First, we compared the coefficients for high and intermediate gyr mutations. We constructed a model containing indicator variables for both and then nested models containing both indicators but assuming equal coefficients. Similarly, to evaluate the clinical utility of FQ phenotypic DST vs gyr sequencing, we compared a model in which resistance was defined by presence of a high-level gyr mutation or by FQ DST, to 2 nested models: (1) in which resistance was defined by the presence of a high-level gyr mutation, and (2) in which resistance was defined by DST. In all cases, we used the likelihood ratio test (LRT) to assess the predictive power of the nested models. The comparisons were repeated for the different FQ exposures (ciprofloxacin, for which we had complete DST; and levofloxacin and moxifloxacin, which are considered to be more efficacious in treating MDR-TB).

The proportional hazards assumption was tested by examining the interaction between the gyr mutation variable and the treatment semester. Informative censoring was assessed by evaluating the association between default and the gyr variables. Missing values were multiply imputed using additive regression, bootstrapping, or predictive mean matching as implemented in the R “Hmisc” package version 4.0-1. All statistical tests were 2-sided. Analyses were conducted using R version 3.3.2 and the “survival” package 2.40-1. Throughout, we used a P value cutoff of .05 to assess statistical significance.

Ethics Review

This retrospective study was approved by Harvard Medical School’s Committee on Human Studies, Partners Healthcare, and the Ministry of Health of Peru. These ethics boards did not require informed consent.

RESULTS

Of the 171 patients, 40% were female and the mean age was 30 years (Table 1). Overall, the observed TB disease severity was high: 56% of the patients had cavitary or bilateral disease, 43% were malnourished or had a low BMI, and 74% manifested respiratory distress. On average, the subjects’ TB isolates had phenotypic resistance to 6.4 drugs with 91 being XDR (53%) and 22 (13%) MDR with additional resistance to a second-line injectable (capreomycin, kanamycin, or amikacin) or an FQ—that is, pre-XDR. Overall, more than half (55%) were resistant to either ciprofloxacin or levofloxacin. On sequence analysis, 12 isolates (7%) harbored the gyrA A90V mutation, and 8 (5%) the gyrA D94A mutation; another 24 isolates carried other gyr mutations that were previously associated with high-level moxifloxacin resistance (Table 1). There were significant discrepancies between gyr sequencing and each FQ DST, with 14%–53% of isolates found to be resistant in culture lacking specific gyr resistance mutations (Table 2). Nearly all patients with genotypic (42/44 [95%]) or phenotypic FQ resistance (90% [84/93] with ciprofloxacin resistance, 85% [40/47] with levofloxacin resistance, and 93% [38/41] with moxifloxacin MIC >0.5 μg/mL) were treated with moxifloxacin or levofloxacin.

There were 40 deaths (23%) and 11 treatment failures (6%). Cure or completion was reported in 107 (62%) patients. Thirteen subjects (8%) were lost to follow-up or transferred out of care.

In univariate analysis, the administration of a late-generation vs early-generation FQ was not significantly associated with death or treatment failure; the administration of an otherwise adequate treatment regimen was highly protective (Table 3).High-level gyr mutations predicted poor outcome (hazard ratio [HR], 2.04; 95% confidence interval [CI], 1.08–3.84; P = .03), but the intermediate gyr mutations did not (HR, 1.48; 95% CI, .69–3.16; P = .31). Upon adjustment for other covariates, including interactions with FQ treatment that were not significant, the hazard associated with the presence of high-level gyr mutations increased to 2.60 (95% CI, 1.21–5.55; P = .01) (Table 4).

There was no difference in the relative strength of the association between death and failure and the 2 gyr mutation categories, high-level and intermediate (Table 5). The unadjusted rate of poor outcomes was higher for isolates sensitive by gyr sequencing but resistant by DST, than for those resistant by both measures (Table 6). We assessed if resistance determined by culture-based FQ DST was a better predictor of treatment outcome than gyr mutations. High-level gyr mutations were superior to ciprofloxacin DST alone, and were comparable to levofloxacin or moxifloxacin DST (Table 5). Resistance defined by the detection of any gyr resistance mutation (ie, high-level or intermediate) was also superior to resistance by ciprofloxacin DST alone (LRT χ² = 4.7; P = .03) in predicting poor treatment outcome.

DISCUSSION

In this well-characterized treatment cohort in Peru with high levels of drug resistance, we found high-level gyr resistance mutations to be strongly associated with death or failure, controlling for severity of illness, comorbidities, and other tre‑atment variables. Although patients with isolates harboring these mutations were disproportionately treated with late-generation FQs, these findings suggest that the regimens remain compromised and should be reinforced in such patients. Isolates with the intermediate mutations gyrA A90V and D94A have been consistently observed to have lower levels of in vitro FQ resistance [27]. These intermediate gyr mutations, occurring at similar frequency to high-level gyr mutations, were not significantly associated with treatment outcome, suggesting that patients with isolates harboring only these intermediate mutations may still respond to treatment with late-generation FQs. Both results are consistent with findings by Rigouts et al [26], which discriminated between high-level and intermediate gyr mutations in their association with treatment failure (excluding death and default) of a 9-month standardized regimen (the “Bangladesh regimen”) containing high-dose, late-generation FQs. Rigouts et al did exclude MDR patients with resistance to second-line injectables, and did include additional gyr mutations in their intermediate category, notably all nonglycine mutations at codon 94 and all mutations outside of‑ codon 94 [26]. Although the association between our category of intermediate gyr mutations and poor treatment outcome was not significant, the value of intermediate gyr mutations as a predictor of poor outcomes was not statistically distinguishable from that of high-level gyr mutations. Thus, the measurement of unfavorable outcomes in larger cohorts with intermediate mutations is necessary before any recommendation against treatment reinforcement can be made in this group.

The clinical significance of resistance classified by gyr mutations was not enhanced by the information on resistance classification from culture-based DST. Rather, we observed improved prediction through the gyr mutation detection compared to ciprofloxacin DST, if not for the later-generation agents. This is consistent with prior reports that have described that DST to early-generation FQs such as ofloxacin overcall drug resistance relative to DST to later-generation FQs [28, 29]. As the observed HRs of poor treatment outcome for all 3 FQ drug susceptibility tests were not statistically significant, our comparison with culture-based DST is the first to suggest that gyr mutation detection is as good, if not better, in identifying patients unlikely to respond to FQ treatment. We observed this to be the case even with the relatively small number of gyr mutations that we defined as relevant to FQ resistance. All gyr mutations that we used to define resistance are detectable by the HAIN MTBDRsl version 2.0, with the 2 intermediate mutations detected by the probes MUT1 (A90V) and MUT3A (D94A) [8, 30]. These results argue against focusing efforts on identifying additional mutations to close the sensitivity gap of FQ molecular diagnostics using culture-based DST as a gold standard. Instead, they support a closer look at how isolates with different gyr resistance mutations respond to therapy in the host.

There are several possible limitations. First, there may be unmeasured confounding in this retrospective study. Differences in treatment effectiveness, baseline disease severity, treatment-related adverse events, or comorbidities all have the potential to confound the association between FQ resistance and the patient outcome. Although we attempted to capture these effects using the covariates described, we had no measurement of adverse events. Second, although we specifically selected patients with more advanced levels of resistance, the number of observed mutations was small, with only 45 of 172 isolates harboring gyr resistance mutation. This may have contributed to the limited power to detect a statistically significant difference between the HRs of the high and intermediate mutations.

This study represents the first attempt to compare molecular and culture-based FQ resistance in patients with pre–XDR-TB or XDR-TB on individualized therapy. This is an increasingly relevant question as TB programs roll out molecular diagnostics and grapple with interpreting discrepancies between them and traditional methods. Furthermore, it provides strong evidence that treatment reinforcement is needed in the case of high-level gyr mutations. For the first time, viable treatment alternatives exist for patients in whom regimens may be compromised by FQ resistance. Using the results of the validated, rapid molecular test for FQ resistance is a way to rule out use of the shortened regimen and rule in the use of bedaquiline or delamanid, per World Health Organization guidelines [31]. We present evidence that patients with resistance detected by phenotypic testing alone without gyr mutations or with only intermediate gyr mutations may still respond to conventional treatment with late-generation FQs, providing evidence that not all resistance mutations are equal in terms of treatment response. We anticipate that results from the large ongoing clinical trials on MDR-TB treatment will help validate these observations in the near future.

Notes

Author contributions. M. R. F. and C. D. M. designed the study with key contributions from K. R. J. and M. M. D. K. maintained the strain bank and performed the MIC measurements. M. R. F. performed the data analysis with key input from M. F. F. and wrote the first draft of the manuscript. M. M., C. D. M., M. F. F., and K. R. J. provided key edits to the manuscript.

Acknowledgments. We thank the Peruvian team for their patient care and for providing the clinical isolates that made this study possible.

Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health (NIH).

Financial support. This work was supported by the Harvard Center for AIDS Research, an NIH-funded program (grant number P30 AI060354), which is supported by the following NIH co-funded and participating institutes and centers: National Institute of Allergy and Infectious Diseases (NIAID); National Cancer Institute; Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Heart, Lung, and Blood Institute; National Institute on Drug Abuse; National Institute of Mental Health; National Institute on Aging; National Institute of Diabetes and Digestive and Kidney Diseases; National Institute of General Medical Sciences; Fogarty International Center; and Office for AIDS Research. This work was also supported by the NIH BD2K initiative (grant number K01 ES026835); the Parker B. Francis Fellowship (to M. R. F.); and the NIH/NIAID (grant numbers R01 AI119037 to K. R. J. and U19-AI109755 to M. M.).

Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

References