Cryptic Microheteroresistance Explains Mycobacterium tuberculosis Phenotypic Resistance

Abstract

Rationale: Minority drug-resistant Mycobacterium tuberculosis subpopulations can be associated with phenotypic resistance but are poorly detected by Sanger sequencing or commercial molecular diagnostic assays.

Objectives: To determine the role of targeted next-generation sequencing in resolving these minor variant subpopulations.

Methods: We used single molecule overlapping reads (SMOR), a targeted next-generation sequencing approach that dramatically reduces sequencing error, to analyze primary cultured isolates phenotypically resistant to rifampin, fluoroquinolones, or aminoglycosides, but for which Sanger sequencing found no resistance-associated variants (RAVs) within respective resistance-determining regions (study group). Isolates also underwent single-colony selection on antibiotic-containing agar, blinded to sequencing results. As a positive control, isolates with multiple colocalizing chromatogram peaks were also analyzed (control group).

Measurements and Main Results: Among 61 primary culture isolates (25 study group and 36 control group), SMOR described 66 (49%) and 45 (33%) of 135 total heteroresistant RAVs at frequencies less than 5% and less than 1% of the total mycobacterial population, respectively. In the study group, SMOR detected minor resistant variant subpopulations in 80% (n = 20/25) of isolates with no Sanger-identified RAVs (median subpopulation size, 1.0%; interquartile range, 0.2–3.9%). Single-colony selection on drug-containing media corroborated SMOR results for 90% (n = 18/20) of RAV-containing specimens, and the absence of RAVs in 60% (n = 3/5) of isolates. Overall, Sanger sequencing was concordant with SMOR for 77% (n = 53/69) of macroheteroresistant (5–95% total population), but only 5% of microheteroresistant (<5%) subpopulations (n = 3/66) across both groups.

Conclusions: Cryptic minor variant mycobacterial subpopulations exist below the resolving capability of current drug susceptibility testing methodologies, and may explain an important proportion of false-negative resistance determinations.

At a Glance Commentary

Scientific Knowledge on the Subject

Heteroresistance is a potential diagnostic, prognostic, and therapeutic problem for diseases caused by pathogenic bacteria, viruses, and parasites, as well as human malignancies. Microheteroresistant Mycobacterium tuberculosis subpopulations (<5% of the total sampled population) are poorly described by conventional Sanger sequencing or commercial molecular tuberculosis assays, and hitherto uninvestigated using ultradeep sequencing approaches.

What This Study Adds to the Field

Using a targeted next-generation sequencing approach that dramatically reduces sequencing error relative to conventional next-generation sequencing, we describe the presence of resistant variant subpopulations in a majority of phenotypically resistant isolates without Sanger-identified resistance-associated variants, and corroborate our findings via single-colony selection on drug-containing media. Our study provides further evidence for a rich diversity of minor variant subpopulations below the resolving capability of Sanger sequencing. These microheteroresistant M. tuberculosis subpopulations convey phenotypic resistance and may explain an important proportion of false-negative resistance determinations by conventional sequencing and molecular tuberculosis assays.

Multidrug-resistant tuberculosis (MDR TB), defined as resistance to the essential first-line agents isoniazid and rifampin (RIF), affects nearly half a million people each year and is a major obstacle to TB elimination (1). Programmatic surveys (2) and clinical trials (3) suggest that approximately one-half of MDR TB isolates demonstrate resistance to at least one second-line drug, and that acquired resistance to fluoroquinolones (FQ) or second-line injectable (SLI) medicines, the backbone of current standardized regimens, occurs during treatment in up to 15% of patients (4). Although treatment failure and amplified drug resistance impose major human and programmatic costs (5), including ongoing transmission of drug-resistant strains (6, 7), laboratory capacity for the diagnosis of drug-resistant TB in most high-burden settings remains limited.

Although several existing assays detect resistance to first-line drugs (8), the first commercial molecular diagnostic for second-line Mycobacterium tuberculosis drug susceptibility testing (DST) (the reverse line blot hybridization assay, GenoType MTBDRsl version 2; Hain Lifescience, Nehren, Germany) (9, 10) was only recently endorsed by the World Health Organization in May 2016, including for direct testing on sputum smear-positive and smear-negative specimens. Although MTBDRsl and other current molecular assays are transformative advances, suboptimal sensitivity has been commonly ascribed to incomplete characterization of drug resistance loci. Furthermore, no commercial assay is yet available for many components of the MDR TB short-course regimen (e.g., clofazimine, ethionamide, ethambutol, pyrazinamide) (11).

Suboptimal sensitivity of commercial molecular DST assays might also be a function of their limited resolution for detection of heteroresistance (Xpert MTB/RIF, able to detect resistant subpopulations ≥65% of total sampled population [12, 13]; Xpert Ultra, 5–10% [S531L], 20–40% [other rpoB mutations] [14]; pyrosequencing, ≥10% [15]; line probe assay, ≥5–10% [16]). Heteroresistance indicates the coexistence of drug-resistant and drug-susceptible strains, or MDR strains with discrete haplotypes (i.e., combinations of resistance-associated variants (RAVs) at several loci that are transmitted together), and may occur among 5–38% of resistant isolates, depending on antibiogram, population sampling, and detection method (17–23). Genetic heterogeneity at drug resistance loci may indicate mixed infection/reinfection (with strains typically differing by 100s of single-nucleotide polymorphisms) (24, 25); reflect sampling bias (26); and/or characterize the interplay of various subclones responding to ongoing selection pressure (27–30; see Figure 6 in Reference 31). Microheteroresistance may be present at the time of diagnosis in patients with active disease, persist for years, contribute to relapse, and complicate successful treatment (32).

Although Sanger sequencing (33) has traditionally been the reference standard for pathogen RAV verification, it has been replaced over the last 5 years with next-generation sequencing (NGS) technologies. NGS provides rapid, accurate, cost- and process-efficient (34) translation of sequence data into actionable knowledge from clinical specimens (25, 35, 36). However, conventional library preparation processes and the intrinsic error rate have limited the depth of coverage of conventional NGS to hundreds of reads, which are inadequate to resolve rare M. tuberculosis subpopulations. We recently demonstrated a highly sensitive, targeted NGS technique based on complete overlap of forward and reverse paired-end reads from the same DNA molecule (single molecule overlapping reads [SMOR]), providing 10,000–100,000X coverage and able to demonstrate an M. tuberculosis mutant spectrum to a subpopulation resolution of 0.1% (37).

We examined the utility of these advancements in targeted sequencing in resolving isolates phenotypically resistant to RIF, FQs, or SLIs, but without known genotypic correlate (i.e., wild-type) following analysis with Sanger sequencing, therefore hypothesizing that isolates may harbor “microheteroresistant” M. tuberculosis subpopulations below the resolving capability of Sanger sequencing. As a positive control, we analyzed isolates demonstrating Sanger-presumptive heteroresistance according to the presence of multiple colocalizing chromatogram peaks within respective resistance-determining regions (RDRs) submitted during the same time period.

Methods

Specimen Selection

Between 2006 and 2008, in accordance with the National TB Control Program of South Africa, sputum specimens from patients previously treated for TB, failing first-line therapy, or in contact with a patient with drug-resistant TB were submitted to the South African National Health Laboratory Service for DST. RIF-monoresistant (1 μg/ml) isolates grown from sputum and, as previously described (38), isoniazid and RIF-resistant isolates demonstrating additional resistance to ofloxacin (2 μg/ml) and/or amikacin (AMK; 4 μg/ml) on Middlebrook 7H11 (Becton Dickinson, Sparks, MD) underwent genotyping and Sanger sequencing of RDRs (rpoB, gyrA, and rrs, respectively) at Stellenbosch University. Resistant isolates without Sanger sequencing–determined RAVs, or with multiple convergent chromatographic peaks, were selected for our study. Mixed infections were identified by spoligotyping and pncA gene sequencing (38, 39), and were excluded. We excluded mixed strain infection to focus specifically on microevolutionary processes of in situ–derived individual clonal founder strains.

Definitions

We defined heteroresistance as coexisting subpopulations of drug-resistant and drug-susceptible M. tuberculosis organisms, or two or more separate populations of drug-resistant strains within the same patient specimen, detected by conventional Sanger sequencing or targeted deep sequencing, and not in the setting of mixed infection. Microheteroresistance, macroheteroresistance, and full drug resistance were defined as drug-resistant subpopulations less than 5%, 5–95%, and greater than 95% of the total M. tuberculosis population, respectively.

Culturing and DNA Extraction

Decontaminated and liquefied sputum was cultured in the MGIT 960 (Becton Dickinson) system until positive (by acid-fast bacilli smear microscopy and M. tuberculosis speciation), after which DST was done on Middlebrook 7H11 slants (Becton Dickinson). We estimate that at least 1,000 CFUs were plated on the DST control slant following inoculation with 1 ml diluted (1/100) from the MGIT culture; this was then added to 400 μl Tween80 saline solution (0.001% Tween80 and 0.08M NaCl). We then generated a crude DNA lysate (200 μl) by incubation of the cells at 100°C for 30 minutes.

Sanger Sequencing

We sequenced amplification products using an ABI 3130XL genetic analyzer (Applied Biosystems, Foster City, CA), and the resulting chromatograms were analyzed using Chromas software (Technelysium Pty Ltd, South Brisbane, Australia). To identify mutations conferring resistance to RIF, the rifampicin RDR (RRDR) of rpoB (amplification product nucleotides 1,016–1,452) was subjected to DNA sequencing. The presence of more than one nucleotide at a defined sequence position was assigned if the peak height of the underlying nucleotide was greater than or equal to two times the height of the highest background peak. Mutations conferring ofloxacin resistance were determined by DNA sequencing of the quinolone RDR (QRDR) of the gyrA gene and flanking sequences (amplification product codons 18–132), and mutations conferring AMK resistance were determined by DNA sequencing of the region encompassing nucleotides 1,401 and 1,484 of the rrs gene (amplification product nucleotides 1,339–1,528).

Single-Colony Selection on Antibiotic-Containing Agar

Isolates that demonstrated phenotypic resistance in the absence of RAVs detected by Sanger sequencing were subcultured onto Middlebrook 7H10 medium with and without ofloxacin (2 μg/ml), AMK (4 μg/ml), or RIF (1 μg/ml) according to World Health Organization recommendations (40) for 3–4 weeks at 37°C. Thereafter, 8–10 individual CFUs were picked from the drug-containing plate and 1–2 CFUs from the control plate, suspended in 1 ml of enriched 7H9 medium (supplemented with oleic albumin dextrose catalase growth supplement), and incubated at 37°C for 4 days. Thereafter, a 500-μl aliquot was stored at −80°C, whereas the remaining aliquot was heat inactivated at 100°C to generate a crude DNA lysate for Sanger sequencing, as described previously.

Targeted Deep Sequencing

Isolates for which crude M. tuberculosis DNA extracted from the original DST 7H11 agar was available and viable were selected for targeted deep sequencing. DNA specimens were coded, blinded, amplified, and prepared for targeted SMOR sequencing, as described previously (37), with the following modifications. Following the gene-specific multiplex polymerase chain reaction, primer-dimer artifacts were removed using a single 0.8X, Agencourt AMPure XP bead (Beckman Coulter, Brea, CA) cleanup, instead of two sequential bead cleanups, eluting the amplicons in 15 μl of a 10 mM Tris-HCl 0.05% Tween20 solution. The SMOR assay’s gene-specific multiplex polymerase chain reaction contains gene regions critical for detecting mutations associated with the extensively drug resistant phenotype: rpoB to characterize RIF resistance; gyrA to characterize FQ resistance; rrs to characterize AMK resistance; and the eis promoter and rrs to characterize kanamycin resistance (37). All RAVs were covered with 10 or more SMOR reads (i.e., ≥20 standard reads, a pair of reads for each sequenced amplicon molecule), and 58% were covered with 100 or more SMOR reads (≥200 standard reads). Numerous no-template controls were used throughout the preparation process to ensure lack of well-to-well sample or amplicon contamination. DNA from a confirmed pan-susceptible M. tuberculosis H37Rv strain was used as a sequencing error control throughout the SMOR assay, as described previously (36). All sequencing read files were deposited to the Sequence Read Archive under accession numbers SRR5488531-SRR5488590.

Targeted Deep Sequencing Analysis

The previously published SMOR analysis tool (37) was incorporated into the TB Amplicon Sequencing Analysis Pipeline software (35). Briefly, this software automates the process of quantifying the alleles of interest within gene regions of interest, for every overlapping read pair. Paired reads from the same DNA molecule that disagree invariably indicate sequencing error, and were excluded. Therefore, the use of overlapping reads allows for high confidence of low-level subpopulation (>0.1%) detection, well below standard sequencing error rates (37). Targeted sequencing additionally allows for the detection of multiple RDR-associated RAVs within individual amplicons (i.e., haplotype analysis). The Amplicon Sequencing Analysis Pipeline software detects and quantifies the presences of multiple RAV haplotypes among the amplicons to further analyze the nature of heteroresistance within resistant subpopulations.

Results

We assessed 73 primary cultured isolates, and analyzed 61 total primary isolates with both Sanger and targeted deep sequencing. These included 25 phenotypically resistant isolates demonstrating absence of RAVs within respective target RDRs (Study Group; RRDR, n = 12; QRDR, n = 7; rrs, n = 6), and 36 phenotypically resistant primary isolate controls in which Sanger sequencing demonstrated colocalizing chromatogram peaks (Control Group; RRDR, n = 5; QRDR, n = 24; rrs, n = 10) (Figure 1). Clinical records from 24 (39%) patients were available and showed that 13 (54%) received or were receiving treatment at the time of sputum collection (Tables 1 and 2).

Figure 1.

Flow diagram of specimen selection and analysis. DST = drug susceptibility testing; RAV = resistance-associated variant; RDR = resistance-determining region; RIF = rifampin; SA NHLS = South African National Health Laboratory Service; w.t. = wild-type; XDR = extensively drug resistant. *Three isolates in the control group were analyzed for both gyrA and rrs.

Table 1.

Targeted Deep Sequencing of Resistance-Determining Regions Wild-Type by Sanger Sequencing in the Setting of Phenotypic Resistance (Study Group)

Strain Identifier	Sanger	SMOR				Plating on Drug-Containing Agar Confirms SMOR*
		Gene/Position	Codon	RAV Population Size per SMOR (%)	Other Heteroresistance Detected within RDR?
R_1941†	gyrA w.t.	gyrA 94	GGC	13	Micro (multiple)	Yes
X_7‡	gyrA w.t.	gyrA 94	GGC	4.2	Micro (multiple)	Yes
X_62†	gyrA w.t.	gyrA 94	GGC	14	Micro (multiple)	Yes
X_112‡	gyrA w.t.	gyrA 90, 94	GTG, GGC	1.2; 3.2	Micro (multiple)	Yes
X_139‡	gyrA w.t.	—	—	None; w.t.	—	Yes
X_142‡	gyrA w.t.	gyrA 94	GGC	0.3	Micro (multiple)	Yes
X_173	gyrA w.t.	—	—	None; w.t.	—	Yes
R_3271	rrs w.t.	rrs 1401, 1484	G, T	3.9; 1.2	No	Yes
R_3315	rrs w.t.	rrs 1401	G	5.1	No	Yes
R_3720	rrs w.t.	rrs 1401	G	6.4	No	Yes
X_96	rrs w.t.	rrs 1401	G	99	No	Yes
X_100	rrs w.t.	rrs 1401	G	8.6	No	Yes
X_130	rrs w.t.	—	—	None; w.t.§	eis -12T, 100%	No (1401G)
R_2362	rpoB w.t.	rpoB		Unknown (del516_525)	—	Yes
R_3119	rpoB w.t.	rpoB 533	CCG	4.4	Micro (multiple)	No (del516_531)
R_3486	rpoB w.t.	rpoB 516	GTC, TAC	43; 39	Micro (multiple)	Yes
R_4093	rpoB w.t.	rpoB 531	TTG	2.7	No	Yes
R_4271	rpoB w.t.	rpoB 531	TTG	0.1	No	Yes
R_4370	rpoB w.t.	—	—	None; w.t.	—	No (531TTG)
R_4485	rpoB w.t.	rpoB 526	TAC	2.4	No	Yes
R_4927	rpoB w.t.	rpoB 526	TAC	6.5	No	Yes
R_4956	rpoB w.t.	rpoB 526, 531	AAC, TTG	0.1; 0.1	No	No (w.t.)
R_5491	rpoB w.t.	—	—	None; w.t.	—	Yes
R_5603	rpoB w.t.	rpoB 531	TTG	0.2	No	Yes
R_5650	rpoB w.t.	rpoB 526	GAC	0.25	Micro (multiple)	Yes

Definition of abbreviations: RAV = resistance-associated variant; RDR = resistance-determining region; SMOR = single molecule overlapping reads; w.t. = wild-type.

^*Sanger sequencing of single colonies selected following plating on drug-containing agar consistent with largest determined SMOR-resistant subpopulation.

^†Participant had no history of treatment and was not taking fluoroquinolones at time of sampling.

^‡Participant was on treatment with fluoroquinolones at time of sampling.

^§Rare minor variants (0.4% 1401G) were detected in X_130 but did not meet criteria for minimum number of reads.

Table 2.

Targeted Deep Sequencing of Resistance-Determining Regions with Colocalizing Chromatogram Peaks by Sanger Sequencing in the Setting of Phenotypic Resistance (Control Group)

Strain Identifier	Sanger Sequencing		SMOR			Prior Treatment	Current Treatment
	Gene/Position	Codon	Gene/Position	Codon	RAV Population Size per SMOR (%)
R_2335	gyrA 90	GTG	gyrA 90	GTG	21	Yes	Yes
R_2335	—	—	gyrA 91	CCG	11	Yes	Yes
R_2335	gyrA 94	GGC	gyrA 94	GGC	17	Yes	Yes
R_2335	gyrA 94	TAC	gyrA 94	TAC	39	Yes	Yes
R_2335	gyrA 94	AAC	gyrA 94	AAC	9	Yes	Yes
R_2335	—	—	gyrA	mult.	<1	Yes	Yes
R_2652	gyrA 90	GTG	gyrA 90	GTG	74	No	No
R_2652	—	—	gyrA 94	mult.	<1	No	No
R_2658	gyrA 94	GGC	gyrA 94	GGC	24	Yes	Yes
R_2658	gyrA 94	GCC	gyrA 94	GCC	75	Yes	Yes
R_2934	gyrA 94	GGC	gyrA 94	GGC	61	No	No
R_2934	gyrA 94	GCC	gyrA 94	GCC	39	No	No
R_2934	—	—	gyrA	mult.	<1	No	No
R_3034	gyrA 94	AAC	gyrA 94	AAC	69	—	—
R_3034	gyrA 94	GGC	gyrA 94	GGC	31	—	—
R_3034	—	—	gyrA 94	TAC	0.1	—	—
R_3206	gyrA 90	GTG	gyrA 90	GTG	25	No	No
R_3206	gyrA 94	GGC	gyrA 94	GGC	63	No	No
R_3206	gyrA 94	AAC	gyrA 94	AAC	9	No	No
R_3206	—	—	gyrA	mult.	<1	No	No
R_3428	gyrA 88	TGC	gyrA 88	TGC	6	Yes	Yes
R_3428	gyrA 94	AAC	gyrA 94	AAC	91	Yes	Yes
R_3658	gyrA 90	GTG	gyrA 90	GTG	8	—	—
R_3658	gyrA 94	GGC	gyrA 94	GGC	91	—	—
R_3658	—	—	gyrA 94	AAC	0.3	—	—
R_3731	gyrA 94	AAC	gyrA 94	AAC	48	—	—
R_3731	gyrA 94	GGC	gyrA 94	GGC	21	—	—
R_3731	gyrA 94	TAC	gyrA 94	TAC	30	—	—
R_3731	—	—	gyrA 94	CAC	0.1	—	—
R_3731	—	—	gyrA 88	TGC	0.4	—	—
R_4007	gyrA 90	GTG	gyrA 90	GTG	50	—	—
R_4007	gyrA 94	GGC	gyrA 94	GGC	43	—	—
R_4007	gyrA 94	AAC	gyrA 94	AAC	5.7	—	—
R_4007	—	—	gyrA	mult.	<1	—	—
X_1	gyrA 90	GTG	gyrA 90	GTG	13	No	No
X_1	—	—	gyrA	mult.	<5	No	No
X_6	gyrA 90	GTG	gyrA 90	GTG	6	Yes	Yes
X_6	gyrA 94	GGC	gyrA 94	GGC	32	Yes	Yes
X_6	gyrA 94	GCC	gyrA 94	GCC	27	Yes	Yes
X_6	—	—	gyrA 91	CCG	20	Yes	Yes
X_6	—	—	gyrA 94	TAC	14	Yes	Yes
X_8	gyrA 91	CCG	gyrA 91	CCG	44	No	No
X_12	gyrA 90	GTG	gyrA 90	GTG	55	—	—
X_12	—	—	gyrA 91	CCG	15	—	—
X_14	gyrA 94	GGC	gyrA 94	GGC	56	—	—
X_14	gyrA 94	AAC	gyrA 94	AAC	24	—	—
X_14	—	—	gyrA 94	TAC	12	—	—
X_14	—	—	gyrA	mult.	<5	—	—
X_25	gyrA 94	GGC	gyrA 94	GGC	77	Yes	No
X_25	gyrA 94	AAC	gyrA 94	AAC	23	Yes	No
X_25	—	—	gyrA 94	TAC	0.2	Yes	No
X_25	—	—	gyrA	mult.	<1	Yes	No
X_58	gyrA 94	GCC	gyrA 94	GCC	2.6	Yes	Yes
X_58	—	—	gyrA	mult.	<1	Yes	Yes
X_101	gyrA 94	GGC	—	—	—	Yes	No
X_101	gyrA 94	TAC	—	—	—	Yes	No
X_101	—	—	gyrA 91	CCG	13	Yes	No
X_116	gyrA 90	GTG	gyrA 90	GTG	58	Yes	Yes
X_116	gyrA 94	GGC	gyrA 94	GGC	34	Yes	Yes
X_116	—	—	gyrA	mult.	<5	Yes	Yes
X_125	gyrA 94	GGC	gyrA 94	GGC	99.7	No	No
X_126	gyrA 94	GGC	gyrA 94	GGC	80	—	—
X_126	—	—	gyrA	mult.	<1	—	—
X_134	gyrA 94	GGC	gyrA 94	GGC	93	Yes	No
X_134	gyrA 94	AAC	gyrA 94	AAC	6	Yes	No
X_160	gyrA 90	GTG	gyrA 90	GTG	90	—	—
X_160	gyrA 94	GGC	gyrA 94	GGC	10	—	—
X_160	—	—	gyrA	mult.	<1	—	—
X_184	gyrA 90	GTG	gyrA 90	GTG	76	No	No
X_184	gyrA 94	GGC	gyrA 94	GGC	23	No	No
R_2652	rrs 1401	G	rrs 1401	G	18	—	—
R_2654	rrs 1401	G	rrs 1401	G	35	—	—
R_2934	rrs 1401	G	rrs 1401	G	71	No	No
R_3275	rrs 1401	G	rrs 1401	G	36	—	—
X_4	rrs 1401	G	rrs 1401	G	100	—	—
X_105	rrs 1401	G	rrs 1401	G	100	—	—
X_113	rrs 1401	G	rrs 1401	G	92	—	—
X_124	rrs 1401	G	rrs 1401	G	98	No	No
X_125	rrs 1401	G	rrs 1401	G	100	—	—
X_148	rrs 1401	G	rrs 1401	G	100	—	—
R_2115	rpoB 531	TTG	rpoB 531	TTG	62	—	—
R_2115	rpoB 533	CCG	rpoB 533	CCG	32	—	—
R_2115	—	—	rpoB 516	GTC	7	—	—
R_4024	rpoB 531	TTG	rpoB 531	TTG	4.1	—	—
R_4387	rpoB 531	TGG	rpoB 531	TGG	63	—	—
R_4387	rpoB 531	TTG	rpoB 531	TTG	20	—	—
R_4387	—	—	rpoB 526	AAC	0.3	—	—
R_4664	rpoB 531	TTG	rpoB 531	TTG	3	—	—
R_4664	—	—	rpoB 533	CCG	14	—	—
R_4849	rpoB 531	TGG	rpoB 531	TGG	16	—	—
R_4849	—	—	rpoB 531	TTG	2.4	—	—

Definition of abbreviations: mult. = multiple; RAV = resistance-associated variant; SMOR = single molecule overlapping reads.

Prior treatment refers to ≥30 days of ofloxacin or kanamycin for gyrA- and rrs-associated specimens, respectively.

Mutation Detection within Sanger-Determined Wild-Type Gene Regions (Study Group)

SMOR detected minor resistant variant subpopulations in 80% (n = 20/25) of isolates with no Sanger-identified RAVs, at a median population size of 1.0% (interquartile range, 0.2–3.9%) within respective RDRs. Single-colony selection on drug-containing media corroborated SMOR results for 90% (n = 18/20) of RAV-containing specimens, and absence of RAVs in 60% (n = 3/5) of SMOR-determined wild-type RDRs (Table 1). Relative to phenotypic DST, the sensitivity of SMOR in this group was 80% (95% confidence interval, 59–93%). Additional coexisting microheteroresistant subpopulations were noted within the target regions of half (50%; n = 10/20) of non-wild-type specimens (Figure 2A).

Figure 2.

Distribution of single molecule overlapping reads–determined microheteroresistant subpopulations. The heat maps indicate targeted deep sequencing–determined resistant Mycobacterium tuberculosis populations as follows: dark blue, minor resistant subpopulation, 1% of the total M. tuberculosis population; light blue, minor resistant subpopulation, between 1% and 5% of the total M. tuberculosis population; light red, macroheteroresistant subpopulation, 5–95% of the total M. tuberculosis population; red, fixed resistance mutations, >95% total M. tuberculosis population. (A) Study group. The study group consisted of isolates with phenotypic drug resistance (subheading, y-axis) without Sanger sequencing–determined genotypic resistance within respective resistance-determining regions (black outlined boxes). Note that microheteroresistant subpopulations were often detected within rpoB and gyrA, consistent with phenotypic resistance, despite lack of Sanger-identified resistance-associated variant. (B) Control group. The control group was selected on the basis of multiple chromatographic peaks within resistance-determining regions corresponding to phenotypic drug resistance (subheading, y-axis) for each analyzed drug (black outlined boxes). Open circles indicate resistance-associated variants also detected by Sanger sequencing. *Most reads for sample R_2362 identified the 9-bp deletion del516_525. ^†According to national tuberculosis control program policy at the time, second-line DST was not performed for the RIF-monoresistant group. DST = drug susceptibility testing; FQ = fluoroquinolone; INH = isoniazid; PXDR = pre–extensively drug resistant (RIF and INH resistance, with additional resistance to either FQ or SLI); RIF = rifampin; RIF-R = rifampin monoresistance; SLI = second-line injectable medication; XDR = extensively drug resistant (RIF and INH resistance, with additional resistance to FQ and SLI).

Mutation Detection within Sanger-Determined Heteroresistant Gene Regions (Control Group)

Sanger sequencing identified multiple chromatogram peaks at 64 individual loci (RAVs) among 36 isolates. SMOR determined six isolates (17%) to be unexpectedly fully resistant (most [n = 5/6] within rrs) and two (6%) to be unexpectedly wild-type. The remaining isolates (78%; n = 28/36) were correspondingly heteroresistant by Sanger sequencing and SMOR at 56 individual loci (seven within RRDR, 44 within QRDR, and five within rrs). Subpopulation median size of these 56 heteroresistant RAVs was 33% (interquartile range, 18–62%) of total M. tuberculosis population (Figure 2B), consistent with the known limited resolution of Sanger sequencing. In addition to those RAVs identified by Sanger, SMOR detected 38 additional RAVs across eight loci; subpopulations detected by SMOR only were significantly smaller than those detected by both Sanger and SMOR (P < 0.01) (Figure 3).

Figure 3.

Comparison of Sanger sequencing and single molecule overlapping reads (SMOR) for detection of Mycobacterium tuberculosis heteroresistant subpopulations. Each circle represents 1 of 135 total heteroresistant resistance-associated variants detected by next-generation sequencing (NGS) within the resistance-determining regions of interest, stratified by study group (Sanger-identified wild-type resistance-determining regions) or control group (Sanger-identified multiple colocalizing chromatogram peaks within resistance-determining regions), and whether they were detected by SMOR but not by Sanger (NGS Only), or by both SMOR and Sanger sequencing (NGS and Sanger). The dashed line represents the 5% resistant subpopulation cut point defining microheteroresistance. The blue box plots represent the first quartile, median, and third quartile of the heteroresistant subpopulation distribution in each category. Note that two resistance-associated variants were detected by Sanger sequencing but not by SMOR; these were omitted for clarity.

Spectrum of M. tuberculosis Heteroresistant Variants (Combined Groups)

Based on an average sequencing depth ranging from 63,000 to 144,000X, and excluding six fully resistant RAVs, SMOR described 66 (49%) and 45 (33%) of 135 total heteroresistant RAVs within Sanger-analyzed RDRs at frequencies less than 5% and less than 1%, respectively (Figures 2A and 2B). Sanger sequencing was concordant with SMOR for 77% (n = 53/69) of macroheteroresistant (5–95% total population) subpopulations, but only 5% of microheteroresistant (<5%) subpopulations (n = 3/66) across both groups (Figure 3). Individual mutations within each gene exhibited a spectrum of existence as subphenotypic (<1%), microheteroresistant (<5%), or macroheteroresistant variants (Figure 4). Microheteroresistant subpopulations occurred more commonly within rpoB or gyrA than rrs (P = 0.03). The median number of microheteroresistant subpopulations was two per RDR among those with RDR-specific treatment history, and one per RDR among those with no history of treatment (P = 0.6, by Wilcoxon rank sum test). Haplotype mixtures (i.e., multiple RDR-associated RAVS within individual amplicons), however, were most commonly found within gyrA, including up to six separate heteroresistant subpopulations within a single isolate (see Figure E1 in the online supplement).

Figure 4.

Proportional subpopulation size stratified by resistance-associated variant. The frequencies and relative proportions of subphenotypic (dark blue), microheteroresistant (light blue), and macroheteroresistant (light red) subpopulations are presented for each resistance-associated variant analyzed.

Discussion

NGS has an emerging role in rapidly informing treatment of drug-resistant pathogens because of unprecedented resolution, high efficiency, and increasing portability. We analyzed phenotypically RIF-, FQ-, and SLI-resistant isolates without genotypic correlate by Sanger sequencing. Our targeted deep sequencing assay revealed complex microheteroresistant subpopulation structures, often within baseline, pretreatment samples and among patients without prior treatment history, and was validated by growth selection on drug-containing media. These results provide further evidence that discrete bacterial subpopulations may independently contribute to suboptimal sensitivity of commercial molecular TB assays, and that such sensitivity can be improved with greater sequencing depth.

We used a targeted NGS approach combining ultradeep sequencing with validated control of intrinsic sequencing error (35) to uncover a rich diversity of rare variant resistant subpopulations, often coexisting as haplotypes (multiple mutations on a single sequencing read) at presumably subphenotypic (<1%) levels. Our findings were corroborated down to 0.1% of the total sampled population through single-colony analysis following enrichment on drug-containing agar in 90% of cases, and are consistent with M. tuberculosis heteroresistance described in recent reports using deep sequencing (22, 30, 32, 41). As with other pathogenic bacteria (42, 43), malaria (44), HIV (45), and human malignancies (46, 47), M. tuberculosis exhibits dynamic microvariation within genes whose products interface directly with selective pressure, the clinical outcome of which is mediated by host immunity, extent of disease, drug exposure, and fitness cost.

The gradual emergence of drug-resistant M. tuberculosis from very low pretreatment frequencies (∼1 × 10⁻⁵) has been known since studies of streptomycin monotherapy in the 1940s (48, 49), a process increasingly appreciated as biologically complex (50). Contemporary genomics studies using minimally invasive autopsies (26), careful characterization of serial within-host diversity (31), and physiologic imaging (32) now collectively describe substantially greater intrahost genetic diversification than previously appreciated, and confirm positive selection as a driving force during periods of ineffective treatment. Our findings complement this literature and underscore a depth of discrete resistant variant subpopulations, an abundant “playing field” for clonal interference. Although appropriately controlled longitudinal studies are required to characterize the fate of individual RAVs, consideration of the genetic heterogeneity we describe vis-a-vis the most common fixed mutations in resistant clinical strains may be instructive. For example, in our study the gyrA 94GAC->TAC (D94Y) mutation occurred commonly, although often as a rare (<1%) variant, and never as a fixed mutation. Although considered a high-confidence mutation in association studies (51), gyrA D94Y is described relatively less often in multinational cohorts of FQ-resistant strains (22, 41, 52), suggesting that fitness cost associated with this mutation (well-demonstrated for gyrA among other human pathogens [53–55], despite long-standing selective pressure) (56) may be difficult to overcome with epistatic mechanisms. In contrast, the gyrA 94GAC->GGC (D94G) mutation, associated with high-level resistance to newer generation FQ (57), was the most prevalent fixed gyrA mutation in our cohort and several others (22, 41, 52); it occurred proportionally far less often as a rare variant. Correspondingly, within rpoB, the most prevalent mutation in our study and globally (531TCG->TTG [S531L]), is known to confer little to no fitness cost (58), and occurred in approximately equal proportions as a subphenotypic and larger variant.

Consistent with prior comparisons with pyrosequencing (35), we found that SMOR was superior to Sanger (or first-generation) sequencing for detection of low-level resistant variants. Sanger sequencing has been used to systematically confirm single-nucleotide polymorphisms calls in whole genome sequencing studies, and as a reference standard for assessments of novel molecular TB assays (59). However, because of off-target wild-type amplification, it has a well-described 10–20% detection threshold for minor components in a mixed sample that relies on subjective interpretation of chromatogram peaks (33, 60). This detection threshold is largely congruent with our findings, where Sanger detected only one in five heteroresistant subpopulations below a 10% threshold. Furthermore, Sanger sequencing is low-throughput and unable to discern the components of M. tuberculosis haplotype mixtures, an aspect of subpopulation complexity that may be clinically relevant given that accumulated mutations are known to increase minimum inhibitory concentration (57).

There are some limitations to our study. First, there is no accepted gold standard for determination of M. tuberculosis biologic variability at subphenotypic levels. Use of completely overlapping reads in SMOR reduces sequencing error by orders of magnitude; has been established in vitro through use of contrived control mixtures (37); and is validated in the current study through demonstration of minor variants following selection on drug-containing media, a process known to increase the sensitivity for detection of low-frequency events by several logs (61). Second, although a strength of our study is that patients were diagnosed and treated under programmatic conditions, increasing relevance, an accompanying consequence is that treatment data are limited, and treatment outcomes along with important comorbidities including HIV coinfection are unavailable. Relatedly, our study revealed some instances of false-negative phenotypic DST reporting (in particular for SLIs), although this is not unusual for ultrahigh-throughput laboratories. Third, because deep whole genome sequencing was not performed in parallel, we cannot comment on other areas of the genome determinative for drug resistance or its adaptation (62, 63). Lastly, microbiology laboratory procedures (e.g., M. tuberculosis expansion in subculture media, drug concentrations, bacterial dilutions) likely influence the spectrum of microheteroresistance detectable by deep sequencing (64), although the relative impact of each factor remains poorly characterized.

The incorporation of novel targeted sequencing technologies into the care of patients with drug-resistant TB will contribute to an evidence base around the clinical impact of microheteroresistance and, most importantly, allow early, comprehensive, and personalized profiling of drug resistance. Accordingly, a user-friendly kit-based version of our assay (35) is currently under development within a tabletop NGS platform for rapid DST in limited resource settings (65). Large, prospective studies with well-characterized drug exposure, comprehensive clinical annotation including surrogates of host immune response, and serial deep sequencing at clinically relevant time points will contribute substantially to characterization of the drivers of M. tuberculosis microheteroresistance and its clinical significance.