HIV Coinfection Is Associated with Low-Fitness rpoB Variants in Rifampicin-Resistant Mycobacterium tuberculosis

We analyzed 312 drug-resistant genomes of Mycobacterium tuberculosis isolates collected from HIV-coinfected and HIV-negative TB patients from nine countries with a high tuberculosis burden. We found that rifampicin-resistant M. tuberculosis strains isolated from HIV-coinfected patients carried disproportionally more resistance-conferring mutations in rpoB that are associated with a low fitness in the absence of the drug, suggesting these low-fitness rpoB variants can thrive in the context of reduced host immunity.

spread of drug-resistant M. tuberculosis strains from immunocompromised patients (3)(4)(5). The propensity of drug-resistant M. tuberculosis strains to spread is influenced by the fitness cost associated with drug resistance determinants (6). Specifically, bacterial strains that have acquired drug resistance-conferring mutations may be less transmissible than their susceptible counterparts, although this fitness cost can be ameliorated by compensatory mutations (7)(8)(9)(10). Moreover, the effect of different resistanceconferring mutations on fitness can be heterogeneous (11). In the clinical setting, there is a selection for high-fitness and/or compensated drug-resistant M. tuberculosis strains in TB patients (12). However, in immunocompromised hosts, such as HIV-coinfected patients, even strains with low-fitness resistance mutations might propagate efficiently (13)(14)(15), which could partially explain why drug-resistant TB has been associated with HIV coinfection (16,17). However, to date, no evidence directly supports the notion that the immunological environment created by HIV coinfection modifies the fitness of drug-resistant M. tuberculosis (5,18,19).
In this study, we tested the hypothesis that resistance-conferring mutations with low fitness in M. tuberculosis are overrepresented among HIV-coinfected TB patients. We focused our analysis on isoniazid and rifampicin, the two most important first-line anti-TB drugs, for which resistance-conferring mutations have been shown to differ in their fitness effects when measured in the laboratory (11). In addition, the frequency of the resistance alleles found in a clinical setting correlates well with the in vitro fitness of strains (12,20). To explore the association between HIV coinfection and the fitness effect of different drug resistance-conferring mutations in M. tuberculosis, we compiled a collection of drug-resistant strains using the global International Epidemiology Databases to Evaluate AIDS (IeDEA, http://www.iedea.org) consortium (21,22) as a platform. For this study, 312 strains were collected from HIV-coinfected and HIV uninfected TB patients originating from nine countries on three continents: Peru, Thailand, South Africa, Kenya, Côte d'Ivoire, Botswana, Democratic Republic of the Congo, Nigeria, and Tanzania ( Fig. 1; see also Table S1 in the supplemental material). The association between the fitness of isoniazid resistance-conferring mutations and HIV coinfection was tested in a univariate analysis (Fig. S1). Isoniazid resistance-conferring mutations were divided into three groups, as previously described (23): katG S315T mutation, katG mutations other than S315T, and inhA promoter mutations only. The S315T substitution in katG causes high-level isoniazid resistance while retaining some catalase/peroxidase functions (24). Conversely, the inhA promoter mutation does not affect KatG activity. Other substitutions/deletions in katG have been associated with a lower fitness in the laboratory and are observed only rarely among clinical isolates (23,25,26). In the case of rifampicin, the association between the fitness of rpoB variants and HIV coinfection was tested in both a univariate and multivariate analysis (Table 1). Resistanceconferring variants in rpoB were classified into two groups based on their fitness effects documented previously (11,20,27). The mutation rpoB S450L was considered high fitness, since this mutation was previously shown to confer a low fitness cost in the laboratory (11) and is generally the most common in clinical strains (28). Any other resistance-conferring variant affecting rpoB was considered low fitness (11). The multivariable logistic regression model with outcome of low-fitness rpoB variants was adjusted for host-related factors (history of TB, country of isolation, sex, and age) (29) and bacterial factors (M. tuberculosis lineage, presence of an rpoA-C compensatory mutation, clustering of the genome inferred by genetic relatedness). Seventy-six patients from Tanzania and Botswana were excluded from the model due to missing or unknown clinical data (see the supplemental methods file).
Out of 312 patients, 113 (36.2%) were HIV coinfected, 120 (38.5%) were women, 115 (36.9%) were newly diagnosed TB cases (therefore, treatment naive), 276 (88.5%) harbored isoniazid resistance-conferring mutations, with or without additional resistance, and 282 (90.4%) harbored rifampicin resistance-conferring mutations, with or without additional resistance. In total, 78.8% (n ϭ 246) of the strains were classified as being at least multidrug resistant, defined as resistance to isoniazid and rifampicin with or without additional resistance to second-line drugs. Among the 113 HIV-coinfected individuals, 34 (30%) were on antiretroviral therapy (ART), 26 (23%) were not, and 53 (47%) had an unknown ART start date. Four of the eight known M. tuberculosis lineages were represented in the following proportions: 11 L1 (3.5%), 57 L2 (18.3%), 38 L3 (12.2%), and 206 L4 (66.0%). After dividing a total of 276 isoniazid-resistant strains into the three groups of isoniazid resistance-conferring mutations defined above, we found similar proportions in HIV-coinfected and HIV-uninfected patients (chi-square test, P ϭ 0.54; Fig. S1), and, as expected, the katG S315T mutation was the most frequent mutation in both categories (overall, found in 80% of isoniazid-resistant strains). In the case of rifampicin resistance, a univariate and multivariate analysis of 203 strains with complete clinical records indicated that HIV-coinfected TB patients carried a higher proportion of low-fitness rpoB resistance variants than HIV-negative patients (72.3%   (30,31). The resulting smaller M. tuberculosis population size would lead to fewer replication events, possibly reducing the number of mutations available for selection to act upon. In other words, low-fitness variants and high-fitness variants would co-occur less often in an HIV-coinfected patient, such that competition between them would be less likely. This scenario would be relevant for de novo acquisition of low-fitness drug-resistant variants within an HIV-coinfected patient. Second, following the transmission of a drug-resistant strain with low fitness to a host with reduced immunity, weaker immune pressure acting on this strain might lead to better bacterial survival. The association between low-fitness rpoB variants and HIV coinfection remained significant even after adjusting for the different epidemiologic settings (i.e., countries) and the strain genetic background (i.e., M. tuberculosis lineages). We also observed that strains carrying the rpoB S450L resistance-conferring mutation were more likely to also carry a compensatory mutation in rpoA-C (97.4% versus 2.6%) ( Table 1). Even though this phenomenon seems counterintuitive, it has been described multiple times (7,9,(32)(33)(34) and, thus, might point to different mechanisms of compensation in strains carrying resistance mutations other than rpoB S450L. In addition, in our study, L4 strains were associated with low-fitness rpoB variants compared to L2 (odds ratio, 3.10 [95% confidence interval, 0.94, 10.21], P ϭ 0.06) ( Table 1), indicating that the strain genetic background plays a role in shaping the cost of resistance, as was previously shown for other bacterial species (35) and for other drugs (36). In the regression analysis, we had several categorical variables with only a few observations. Therefore, statistical power, especially for country of isolation, was low, and the results should be interpreted with care.
HIV-coinfected TB patients are generally thought to have a reduced potential for TB transmission (30,37), because these patients have reduced formation of lung cavities, more extrapulmonary disease, and a shorter period of infectiousness due to earlier diagnosis or higher mortality, especially in the absence of antiretroviral treatment and if antibiotic resistance is already present (4). Based on the overrepresentation of low-fitness rpoB mutations in the context of HIV coinfection, one would expect a further reduction of the transmission potential of drug-resistant TB in this context. However, outbreaks of drug-resistant TB in HIV-coinfected patients have been reported (3). Such outbreaks might be explained by (i) a higher risk of M. tuberculosis infection and reinfection due to diminished host immunity, (ii) on-going transmission of drugresistant M. tuberculosis from a larger pool of immunocompetent TB patients to immunocompromised patients, (iii) transmission occurring in conducive environments, such as health care settings, where both HIV-coinfected individuals and drug-resistant TB patients are more likely to coexist, and (iv) M. tuberculosis strains carrying highfitness drug resistance mutations.
In summary, using a global sample of drug-resistant M. tuberculosis clinical strains from HIV-coinfected and HIV-negative TB patients, we showed that low-fitness rpoB variants were overrepresented in HIV-coinfected patients, and that this association was independent from other potential confounding factors. Taken together, our results provide new insights into how HIV coinfection can impact the fitness of drug-resistant M. tuberculosis.
Data availability. The M. tuberculosis whole-genome sequences from the patients are available on NCBI under several project identifiers. The accession number for each genome is indicated in Supplemental Table S1.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 0.1 MB. SUPPLEMENTAL FILE 2, XLSX file, 0.03 MB.