Antibiotic treatment and selection for glpK mutations in patients with active tuberculosis disease

We read with great interest the paper by Safi et al. (1) describing frameshifts in glpK’s homopolymeric tract (HT) of seven cytosines (7C) as a potential cause of antibiotic tolerance. These results have implications for tuberculosis (TB) treatment, but other forces than antibiotic pressure may be responsible for the emergence of glpK mutations (1, 2). We raise the possibilities of selection in vitro, selection in vivo from nonantibiotic host factors, and population bottlenecks in determining the fate of glpK frameshifts.

In a recent study (3), we observed the landscape of Mycobacterium tuberculosis mutations occurring longitudinally in-host for 200 subjects with active TB disease. We tracked mutations arising in-host between two sputum samples taken at different time points during and/or after treatment. To approximate the experimental noise, we analyzed mutations arising between 62 technical replicate pairs of isolates with interval passaging in vitro and/or freezing without antibiotic exposure. Mutations between replicates are thus unrelated to any in-host selection and may result from genetic drift or in vitro selection.

We detected four glpK nonsynonymous mutations (nSNPs) from 200 longitudinal pairs and five nSNPs from 62 replicate pairs (Table 1). Additionally, we observed frameshifts in the 7C HT of glpK (1, 2) to occur commonly between both longitudinal and replicate isolate pairs. GlpK frameshifts were detected more commonly among technical replicates (13/62 vs. 14/200 OR = 2.9, P = 0.003 Fisher’s exact test). We assessed whether glpK frameshifts were more abundant as a function of time between sputum samplings (Fig. 1). Isolates with glpK frameshifts were sampled at similar intervals to those without, indicating that chronic infection or longer duration of antibiotic exposure did not associate with a higher prevalence of these variants.

Table 1.

Mutations detected within the coding sequence for glpK between 200 longitudinal and 62 replicate isolate pairs

H37Rv reference position	H37Rv reference allele	Alternate allele	Change in allele frequency	H37Rv locus tag	Isolate pair type	Gene position	Variant type	Amino acid change/indel type	Alternate allele frequency A	Alternate allele frequency B
4138945	A	G	0.41	Rv3696c	Longitudinal	811	nSNP	C271R	0.41	0
4138645	G	A	0.93	Rv3696c	Longitudinal	1111	nSNP	P371S	0.93	0
4138599	C	A	0.43	Rv3696c	Longitudinal	1157	nSNP	R386L	0.43	0
4138398	G	A	0.79	Rv3696c	Longitudinal	1358	nSNP	A453V	0.79	0
4139183	A	AC	0.386781609	Rv3696c	Longitudinal	573	Indel	Insertion	0.903448276	0.516666667
4139183	A	AC	0.573228751	Rv3696c	Longitudinal	573	Indel	Insertion	0.910891089	0.337662338
4139183	A	AC	0.941666667	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.941666667
4139183	A	AC	0.633663366	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.633663366
4139183	A	AC	0.894736842	Rv3696c	Longitudinal	573	Indel	Insertion	0.894736842	0
4139183	A	AC	0.882352941	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.882352941
4139183	A	AC	0.534482759	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.534482759
4139183	A	AC	0.95323741	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.95323741
4139183	A	AC	0.326530612	Rv3696c	Longitudinal	573	Indel	Insertion	0.326530612	0
4139183	A	AC	0.827160494	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.827160494
4139183	A	AC	0.818181818	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.818181818
4139183	A	AC	0.873417722	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.873417722
4139183	A	AC	0.627118644	Rv3696c	Longitudinal	573	Indel	Insertion	0	0.627118644
4139183	AC	A	0.495798319	Rv3696c	Longitudinal	573	Indel	Deletion	0	0.495798319
4138684	CG	C	0.343065693	Rv3696c	Longitudinal	1072	Indel	Deletion	0.343065693	0
4138376	C	CA	0.507246377	Rv3696c	Longitudinal	1380	Indel	Insertion	0	0.507246377
4139424	G	A	0.28	Rv3696c	Replicate	332	nSNP	A111V	0	0.28
4139083	G	A	0.38	Rv3696c	Replicate	673	nSNP	R225W	0	0.38
4138995	G	A	0.29	Rv3696c	Replicate	761	nSNP	P254L	0	0.29
4138398	G	A	0.81	Rv3696c	Replicate	1358	nSNP	A453V	0.02	0.83
4138326	A	G	0.47	Rv3696c	Replicate	1430	nSNP	L477P	0	0.47
4139183	A	AC	0.521403509	Rv3696c	Replicate	573	Indel	Insertion	0.894736842	0.373333333
4139183	A	AC	0.470588235	Rv3696c	Replicate	573	Indel	Insertion	0.5	0.970588235
4139183	A	AC	0.85	Rv3696c	Replicate	573	Indel	Insertion	0.85	0
4139183	A	AC	0.25	Rv3696c	Replicate	573	Indel	Insertion	0.25	0
4139183	A	AC	0.447058824	Rv3696c	Replicate	573	Indel	Insertion	0.447058824	0
4139183	A	AC	0.670807453	Rv3696c	Replicate	573	Indel	Insertion	0.670807453	0
4139183	AC	A	0.237288136	Rv3696c	Replicate	573	Indel	Deletion	0.237288136	0
4139183	A	AC	0.534482759	Rv3696c	Replicate	573	Indel	Insertion	0.534482759	0
4139183	A	AC	0.513513514	Rv3696c	Replicate	573	Indel	Insertion	0.513513514	0
4139183	A	AC	0.573863636	Rv3696c	Replicate	573	Indel	Insertion	0.573863636	0
4139183	A	AC	0.561643836	Rv3696c	Replicate	573	Indel	Insertion	0.561643836	0
4139183	A	AC	0.348314607	Rv3696c	Replicate	573	Indel	Insertion	0.348314607	0
4139183	A	AC	0.868686869	Rv3696c	Replicate	573	Indel	Insertion	0.868686869	0

SNPs were reported for a pair only if the alternate allele frequency between both isolates changed by at least 25%. Indels were reported if the alternate allele was detected at any allele frequency in at least one isolate for each pair. In a majority of cases, the mutant allele was detectable only in one of the isolates in each pair. The values in columns alternate allele frequency A and alternate allele frequency B are ordered according to sample collection dates for the longitudinal pairs. This ordering is arbitrary for replicate pairs since we did not have sample collection dates for the replicate isolates.

Fig. 1.

Emergence of glpK frameshift mutations in-host. Isolate collection times were available for 195/200 longitudinal pairs, and glpK frameshift mutations were detected in 14/195 longitudinal pairs. Boxplots display the number of days elapsed between isolate collections from subjects with active TB disease in which no glpK frameshift mutations were called (median 291 d) and in which a glpK frameshift mutation was called (median 167 d).

Our data confirm that nSNPs and reversible frameshifts are common within glpK but the frequency varies with several factors that include experimental conditions, genetic drift following a sampling bottleneck, and/or selection in culture on glycerol-based media. Pethe et al. (4) showed that M. tuberculosis glycerol dissimilation was dispensable in a mouse model of TB, suggesting that glpK is dispensable for growth in humans. We hypothesize parsimoniously that inactivating glpK mutations may arise in-host due to the cost of protein production for a potentially dispensable enzyme rather than selection pressure from antibiotic treatment. Adding antibiotic pressure results in a population bottleneck or may increase the fitness differential between strains with and without glpK frameshifts. Once samples are isolated and cultured in the presence of glycerol, bacteria can rapidly revert to wild type since glpK function is necessary for glycerol catabolism (2) as demonstrated by Safi et al. (1) in their in vitro passaging experiments.

Still, the evidence presented makes a strong case that glpK mutations alter drug tolerance in glycerol-containing culture media (1). Further research including the sequencing of M. tuberculosis directly from sputum can overcome the problems of selection and bottlenecks related to in vitro culture (5). Clinical studies assessing treatment outcomes for TB with and without glpK variants are also needed to assess the potential future use of glpK inhibitors for patient treatment.