Whole-genome sequencing of rifampicin-resistant Mycobacterium tuberculosis strains identifies compensatory mutations in RNA polymerase genes.

Epidemics of drug-resistant bacteria emerge worldwide, even as resistant strains frequently have reduced fitness compared to their drug-susceptible counterparts. Data from model systems suggest that the fitness cost of antimicrobial resistance can be reduced by compensatory mutations; however, there is limited evidence that compensatory evolution has any significant role in the success of drug-resistant bacteria in human populations. Here we describe a set of compensatory mutations in the RNA polymerase genes of rifampicin-resistant M. tuberculosis, the etiologic agent of human tuberculosis (TB). M. tuberculosis strains harboring these compensatory mutations showed a high competitive fitness in vitro. Moreover, these mutations were associated with high fitness in vivo, as determined by examining their relative clinical frequency across patient populations. Of note, in countries with the world's highest incidence of multidrug-resistant (MDR) TB, more than 30% of MDR clinical isolates had this form of mutation. Our findings support a role for compensatory evolution in the global epidemics of MDR TB.

The worldwide emergence of multidrug-resistant (MDR) strains of Mycobacterium tuberculosis is threatening to make one of humankind’s most important infectious diseases incurable[8]. MDR strains of M. tuberculosis are resistant to isoniazid and rifampicin, the two most important anti-TB drugs. Theory predicts that one of the key factors driving the current epidemics of MDR-TB is the relative fitness of drug-resistant strains compared to drug-susceptible forms[9,10]. Experimental work has shown that drug resistance in bacteria is often associated with a fitness deficit[1,2,11-13], but some drug resistance-conferring mutations cause little- or no loss of fitness[12,14,15]. Furthermore, fitness cost linked to drug resistance can be reduced by compensatory evolution[2,14]. However, little data exist on the clinical relevance of this phenomenon[3,6]. In M. tuberculosis, compensatory mechanisms were identified for fitness defects related to isoniazid and aminoglycoside resistance[16,17]. However, the corresponding compensatory mutations are rare in clinical strains[18], suggesting they play a minor role in the epidemiology of MDR-TB. While some work on the compensatory evolution in resistance to rifampicin has been reported for Escherichia coli[14], nothing is known with respect to compensatory evolution in rifampicin-resistant M. tuberculosis.

Rifampicin binds to the beta-subunit of the RNA polymerase encoded by rpoB and inhibits transcription. More than 95% of M. tuberculosis clinical strains resistant to rifampicin harbour a mutation in an 81-basepair region of rpoB known as the rifampicin resistance determining region (RRDR)[18]. These mutations convey high-level resistance. We have previously shown that all laboratory-generated mutants of M. tuberculosis harbouring a rifampicin resistance-conferring mutation in RRDR suffer a significant fitness defect compared to their drug-susceptible ancestors when grown in the absence of rifampicin[12]. By contrast, some M. tuberculosis clinical strains isolated from TB patients who developed rifampicin resistance during treatment exhibited no fitness cost compared to their rifampicin-susceptible counterparts, despite carrying the same rpoB mutation as some of the laboratory-derived strains[12]. At the time, we hypothesized that these clinical strains might have acquired compensatory mutations during patient treatment.

Here we tested this hypothesis by comparing the genome sequences of 10 paired clinical rifampicin-resistant isolates to the genomes of the corresponding rifampicin-susceptible isolates recovered from the same patient (Supplementary Table 1)[12]. We extracted all non-synonymous and intergenic mutations found only in the rifampicin-resistant genomes (Supplementary Table 2). In addition, we experimentally evolved six laboratory-derived rifampicin-resistant mutants and their rifampicin-susceptible ancestors[12] during 45 weeks of serial sub-culture in the absence of rifampicin (Supplementary Table 3). Comparison of the whole-genome sequences of these in vitro evolved strains to their respective un-evolved rifampicin-susceptible ancestors allowed us to identify putative compensatory mutations, as well as mutations likely to represent adaptations to growth in the laboratory (Supplementary Table 4). Of note, all of the in vitro evolved rifampicin-resistant strains maintained their original rpoB mutation, which is consistent with a higher number of mutational targets for compensation compared to reversion[2,19].

After combining our clinical and in vitro data, and excluding mutations representing laboratory adaptations or phylogenetic markers (Supplementary Tables 4 and 5), we identified 54 putative compensatory mutations in 38 genes and 10 intergenic regions (Supplementary Table 6). RpoA and rpoC stood out by harbouring multiple mutations in the evolved strains (1 strain) or the paired clinical strains (4 strains; Table 1, Fig. 1). These genes encode the α- and β′-subunits of the RNA polymerase, respectively. Based on the known interactions between the RpoA, RpoB, and RpoC subunits[20], we reasoned that non-synonymous changes in rpoA and rpoC occurring only in rifampicin-resistant genomes were likely compensatory. Mapping these mutations onto the 3D-structure of the E. coli RNA polymerase[20] showed them localized at the interface between the α- and β′-subunits (Fig. 2), indicating a potential impact on the interaction between these subunits.

Table 1

Putative compensatory mutations in rifampicin-resistant Mycobacterium tuberculosis.

Gene	Amino acidchange	Experimentallyevolved strains	Clinicalpairedisolates	GlobalMDRstrains	HighBurdenMDRstrains	Totalstrains	High-probabilitycompensatorymutation	SIFT score
rpoA	R182L	0	0	1	0	1	No	0.00
rpoA	R186C	0	1	0	0	0	No	0.00
rpoA	T187P	0	1	0	0	0	Yes	0.00
rpoA	T181I	0	0	0	1	1	No	0.00
rpoA	T187A	0	1	5	10	15	Yes	0.00
rpoA	V183A	0	0	1	0	1	No	0.00
rpoA	V183G	0	0	1	0	1	No	0.00
rpoC	A521D	0	0	1	0	1	No	0.00
rpoC	D485H	0	0	0	1	1	Yes	0.00
rpoC	D485N	0	1	0	3	3	Yes	0.00
rpoC	E702K	0	0	0	1	1	No	0.00
rpoC	G433S	0	0	0	1	1	No	0.00
rpoC	G519D	0	0	0	1	1	No	0.00
rpoC	H525N	0	0	0	1	1	No	0.00
rpoC	I491T	0	0	0	1	1	Yes	0.00
rpoC	I491V	0	0	0	3	3	Yes	0.00
rpoC	L449V	0	0	0	5	5	No	0.00
rpoC	L516P	1	0	0	2	2	No	0.00
rpoC	N698H	0	0	0	1	1	Yes	0.00
rpoC	N698K	0	0	0	1	1	Yes	0.00
rpoC	N698S	0	0	1	7	8	Yes	0.00
rpoC	P434A	0	0	0	2	2	No	0.00
rpoC	P434R	0	0	1	0	1	Yes	0.00
rpoC	Q435H	0	0	0	1	1	No	0.00
rpoC	Q523E	0	0	0	1	1	No	0.00
rpoC	Q523K	0	0	0	1	1	No	0.00
rpoC	G433S	0	0	1	0	1	No	0.00
rpoC	V432A	0	0	1	0	1	No	0.00
rpoC	V483A	0	0	1	5	6	Yes	0.00
rpoC	V483G	0	0	6	11	17	Yes	0.00
rpoC	V517L	0	0	1	3	4	No	0.00
rpoC	L527V	0	0	1	0	1	No	0.00
rpoC	W484G	0	0	1	3	4	No	0.00

Figure 1

Putative compensatory mutations in a) rpoA and b) rpoC of Mycobacterium tuberculosis

Mutations identified after genome sequencing of experimentally evolved strains or paired clinical isolates are indicated above the gene by a circle and a triangle, respectively. Mutations identified by screening a global and a high-burden collection of MDR strains are indicated by stars below the gene. Colours indicate the respective strain lineage (blue - lineage 2, red - lineage 4, brown - lineage 5, pink - lineage 1). Some of these mutations occurred in multiple lineages and/or affect the same codon position.

Figure 2

Putative compensatory mutations in rpoA and rpoC fall at the interface of RNA polymerase subunits

Mutations identified in rifampicin-resistant experimentally evolved isolates and paired clinical isolates were mapped onto the RNA polymerase of Escherichia coli. The mutations fall in predicted interacting residues of RpoA (light blue) and RpoC (orange) of the E. coli RNA polymerase. Residue numbers are indicated according to M. tuberculosis coordinates. Colour code: RpoA (α-subunit) - blue, RpoB (β-subunit) - red, RpoC (β'-subunit) - yellow, RpoD (σ-subunit) - green.

In addition to these plausible effects on RNA polymerase structure, we expect compensatory mutations in rifampicin-resistant M. tuberculosis i) to occur frequently in MDR clinical isolates, ii) not to occur in rifampicin-susceptible isolates, iii) to be associated with mutations in RRDR, and iv) not to occur in rifampicin-resistant strains without rpoB mutations. Because M. tuberculosis is genetically monomorphic[21], with no ongoing horizontal gene transfer[22,23], rates of convergent evolution in this microbe are extremely low[24,25]. By contrast, drug resistance-conferring mutations exhibit convergent evolution, as drug pressure selects for the same mutations across the different phylogenetic lineages of M. tuberculosis[26]. Following this rational, we expect compensatory mutations also to show convergent evolution. Furthermore, because M. tuberculosis is genetically homogeneous, more than two amino acid variants at the same codon position are rare[27], except in the context of drug resistance[18]. Thus, we also expect particular codon positions involved in compensation to harbour multiple alleles, as several alternative amino acid substitutions might have similar compensatory effects.

To test these predictions, we screened four complementary panels of clinical M. tuberculosis strains for non-synonymous changes in rpoA and rpoC. The first panel comprised 117 MDR strains from global sources, representing five of the six major lineages of human-adapted M. tuberculosis[28] (Supplementary Table 7). The second panel served as a control and included 131 rifampicin-susceptible strains representing the global diversity of M. tuberculosis[29,30]. The third panel consisted of 212 MDR clinical isolates from Abkhazia/Georgia, Uzbekistan and Kazakhstan (Supplementary Table 7)[31-33]; these countries are among the regions with the highest MDR-TB incidence in the world[8]. The fourth panel comprised 40 pan-susceptible isolates from Uzbekistan (Supplementary Table 8). All of the 329 MDR strains included in panel 1 and 3 had phenotypically confirmed rifampicin resistance, and 321/332 (99.7%) of them harboured at least one non-synonymous mutation in RRDR.

Our results showed that after exclusion of phylogenetic markers and mutations likely due to laboratory adaptation, 89/329 (27.1%) of all MDR strains harboured a non-synonymous mutation in rpoA or rpoC. In addition to the mutations already observed in our clinically paired or experimentally evolved strains, we found 28 additional non-synonymous changes in these genes (Table 1). By contrast, none of the 171 rifampicin-susceptible control strains harboured any of these mutations. Furthermore, all MDR strains harbouring an rpoA or rpoC mutation also had a mutation in RRDR, whereas none of the 11 rifampicin-resistant strains without rpoB mutation had any mutation in rpoA or rpoC.

When combining all our data, we found that 11 codon positions in rpoA or rpoC had the same putative compensatory mutations in more than one phylogenetic lineage of M. tuberculosis, and 8 codon positions were found to have more than one amino acid change (Fig. 1). As discussed above, these phenomena are rarely observed in M. tuberculosis outside of drug resistance, and positions exhibiting both phenomena by chance are particularly unlikely. Hence, we focused the remaining of our investigation on the mutations falling in codon positions that satisfied both of these criteria (Table 1).

We computationally predicted the effect of these high-probability compensatory mutations (HCMs) on protein function, by comparing the degree of evolutionary conservation of the orthologous protein positions in other bacteria using SIFT scores [34]. As a comparison, we used 13 publicly available mycobacterial genomes not belonging to the M. tuberculosis complex. As a proof of concept, we first tested whether we could correctly predict that mutations in rpoB known to convey rifampicin resistance (Supplementary Table 7) were more likely to be functional than phylogenetic markers in the same gene (Supplementary Table 5); we found this to be the case (Mann-Whitney U Test p < 0.01). When testing the HCMs in rpoA and rpoC (Table 1), we found that these mutations were also predicted to be more functional compared to phylogenetic markers found in the same genes (Supplementary Table 5, Mann-Whitney U Test p < 0.01).

To test whether the predicted functional effects of HCMs correlated with strain fitness, we combined our new data on the occurrence of these mutations with our older data on the relative fitness of rifampicin-resistant M. tuberculosis[12]. We found that three out of four clinical MDR strains with no competitive fitness defect harboured an HCM (Fig. 3A). By contrast, none of the six clinical strains with a statistically significantly reduced fitness harboured any HCM (Fischer’s exact test p < 0.05). Furthermore, when calculating the difference in fitness between the laboratory-derived rifampicin-resistant strains and the clinical strains harbouring the same rifampicin resistance-conferring mutation and belonging to the same phylogenetic lineage, we found that this difference was always towards increased fitness in the clinical strains harbouring an HCM, and larger compared to the median difference in fitness among the other clinical strains (Fig. 3B; Mann-Whitney U test p < 0.05). Finally, we found that the median time between the isolation of the susceptible patient isolate and the follow-up resistant isolate was longer for strains harbouring an HCM than for strains carrying only an RRDR mutation (20 months versus 6 months; Fig. 3C; Mann-Whitney U test p < 0.05). Taken together, these data show that HCMs in rpoA and rpoC are associated with a high in vitro fitness of MDR clinical strains of M. tuberculosis, and that the emergence of HCMs is time-dependent.

Figure 3

Experimental and clinical relevance of putative compensatory mutations

a) Experimental competitive fitness of 10 clinical isolates that acquired rifampicin resistance during treatment compared to their susceptible counterparts. High-probability compensatory mutations (HCM) are indicated in the pair in which they were identified. Bar colours indicate strain lineage (blue - lineage 2, red - lineage 4). b) Difference in relative fitness between 10 rifampicin-resistant paired clinical isolates compared to laboratory-generated mutants carrying the same rifampicin resistance-conferring mutation and genetic background as defined by strain lineage. Data are shown for clinical strains harbouring (or not) an HCM. c) Time in months between the first and the second isolate of each clinical pair. The horizontal line indicates the median time interval. d) Percentage of MDR strains carrying putative compensatory mutations in rpoA or rpoC. Grey bars refer to the percentage of strains carrying HCMs and black bars refer to strains carrying any putative compensatory mutation. Data for a global collection of strains and for the high MDR-TB burden regions of Abkhazia/Georgia, Uzbekistan and Kazakhstan are shown.

One could argue that because the three clinical strains harbouring HCMs have accumulated additional mutations (Supplementary Table 2), the high fitness of these strains cannot directly be attributed to HCMs. While not discarding possible alternative compensatory mechanisms, at least for clinical Pair 3 (Fig. 3a), several observations support a causal relationship between the HCM observed in rpoC and increased fitness. Specifically, this strain contains only one additional non-synonymous mutation compared to its rifampicin-susceptible ancestor (Supplementary Table 2). However, this additional mutation occurs in aroG, a gene which does not belong to the transcription functional class and for which no interactions with RNA polymerase subunits are known according to the latest version of the STRING database[35]. Moreover, when performing a SIFT analysis as outlined above[34], the aroG change M311T was predicted to have no functional consequence (SIFT score=1.00). Of note, several environmental mycobacteria show the same amino acid substitution, providing additional evidence against a putative role of this aroG mutation in compensatory evolution. In sum, the HCM appears solely responsible for the high fitness observed in this strain.

We and others have shown that for rifampicin resistance, in vitro competitive fitness in M. tuberculosis correlates with in vivo fitness as measured by the frequency of different rifampicin resistance-conferring mutations in clinical settings[12,13]. In other words, rifampicin resistance-conferring mutations associated with no- or low fitness cost in vitro are the most frequent in clinical strains. Hence, in the context of rifampicin resistance, clinical frequency of mutations can be used as a proxy for in vivo fitness of drug-resistant M. tuberculosis among different patient populations[4,15]. When we determined the proportion of HCMs in MDR clinical strains, we found that 12.0% of our global panel of MDR isolates carried such a mutation (Fig. 3D). This proportion increased to 21.3% in our strain panel from countries with a high MDR-TB burden (Chi2=4.5, p < 0.05). When we relaxed our selection criteria and repeated this analysis with all putative compensatory mutations in rpoA and rpoC (Table 1), we found that 19.7% of the global MDR strains carried such a mutation, compared to 31.3% of MDR strains from high MDR-TB burden countries (Chi2=5.14, p < 0.05). The high proportion of compensatory mutations in strains from Abkhazia/Georgia, Uzbekistan and Kazakhstan is consistent with the success of MDR strains in these countries, where up to 50% of TB patients are estimated to carry MDR strains, compared to a global average of only 3%[7].

In conclusion, our results suggest that the emergence over time of particular mutations in rpoA or rpoC of rifampicin-resistant M. tuberculosis lead to MDR strains with high fitness. Furthermore, our data show that these mutations occur at high frequencies in clinical settings, particularly in hotspot regions of MDR-TB[9]. Future studies will tell whether MDR strains of M. tuberculosis harbouring mutations in rpoA or rpoC are particularly transmissible, and how these mutations contribute to the success of these strains. Moreover, targeted genotyping will enable TB control programmes to focus on the most transmissible MDR strains. Our findings also suggest that mathematical models aiming at predicting the future of the global MDR-TB epidemic should consider the effects of compensatory mutations, as well as the time necessary for such mutations to emerge.