Deep amplicon sequencing for culture-free prediction of susceptibility or resistance to 13 anti-tuberculous drugs

Abstract

Conventional molecular tests for detecting Mycobacterium tuberculosis complex (MTBC) drug resistance on clinical samples cover a limited set of mutations. Whole-genome sequencing (WGS) typically requires culture.Here, we evaluated the Deeplex Myc-TB targeted deep-sequencing assay for prediction of resistance to 13 anti-tuberculous drugs/drug classes, directly applicable on sputum.With MTBC DNA tests, the limit of detection was 100-1000 genome copies for fixed resistance mutations. Deeplex Myc-TB captured in silico 97.1-99.3% of resistance phenotypes correctly predicted by WGS from 3651 MTBC genomes. On 429 isolates, the assay predicted 92.2% of 2369 first- and second-line phenotypes, with a sensitivity of 95.3% and a specificity of 97.4%. 56 out of 69 (81.2%) residual discrepancies with phenotypic results involved pyrazinamide, ethambutol and ethionamide, and low-level rifampicin or isoniazid resistance mutations, all notoriously prone to phenotypic testing variability. Only two out of 91 (2.2%) resistance phenotypes undetected by Deeplex Myc-TB had known resistance-associated mutations by WGS analysis outside Deeplex Myc-TB targets. Phenotype predictions from Deeplex Myc-TB analysis directly on 109 sputa from a Djibouti survey matched those of MTBSeq/PhyResSE/Mykrobe, fed with WGS data from subsequent cultures, with a sensitivity of 93.5/98.5/93.1% and a specificity of 98.5/97.2/95.3%, respectively. Most residual discordances involved gene deletions/indels and 3-12% heteroresistant calls undetected by WGS analysis or natural pyrazinamide resistance of globally rare "Mycobacterium canettii" strains then unreported by Deeplex Myc-TB. On 1494 arduous sputa from a Democratic Republic of the Congo survey, 14 902 out of 19 422 (76.7%) possible susceptible or resistance phenotypes could be predicted culture-free.Deeplex Myc-TB may enable fast, tailored tuberculosis treatment.

FIGURE 1

Deeplex Myc-TB results identifying a pre-extensively drug-resistant Mycobacterium tuberculosis complex (MTBC) strain in a sputum DNA sample collected in a tuberculosis drug resistance survey conducted in the Democratic Republic of the Congo. RIF: rifampicin; INH: isoniazid; PZA: pyrazinamide; EMB: ethambutol; SM: streptomycin; FQ: fluoroquinolones; KAN: kanamycin; AMI: amikacin; CAP: capreomycin; ETH: ethionamide; LIN: linezolid; BDQ: bedaquiline; CFZ: clofazimine; NA: not applicable; SIT: spoligotype international type; SNP: single nucleotide polymorphism; LOD: limit of detection. Information on hsp65 best-match-based identification, spoligotype (in this case, not yet known to the SITVIT database) and phylogenetic SNP-based identification of MTBC lineage is shown in the centre of circle. Information on drug susceptibility and drug resistance predictions for 13 anti-tuberculous drugs/drug classes is as follows. Target gene regions are grouped within sectors in a circular map according to the anti-tuberculous drug resistance with which they are associated. Sectors in red and green indicate targets in which resistance-associated mutations or either no mutation or only mutations not associated with resistance (shown in grey) are detected, resulting in predictions of resistant or susceptible phenotypes, respectively. Blue sectors refer to regions where as-yet uncharacterised mutations are detected. Green lines above/below gene names represent the reference sequences with coverage breadth >95%. LOD of heteroresistance (reflected by subpopulations of reads bearing a mutation) depends on the read depth at mutation position and is shown either as grey (LOD 3%) or orange zones (LOD >3–80%) above/below reference sequences. Here, LOD is >3% at the end of a few targets only and over two rrs regions with usual lower coverage.

FIGURE 2

Limit of detection (LOD) of Deeplex Myc-TB for resistance variant detection. a) Read depth at resistance-associated Deeplex Myc-TB targets versus the number of input genomes. Box and whisker plots show median with interquartile range (IQR) and minimum–maximum range with a maximum of 1.5 IQR, respectively; outliers are indicated. b) For each dilution level with 10¹, 10², 10³ and 10⁴ genome copies, LOD was measured as the fraction of detected or undetected resistance variants in total sets of 36 (near-)fixed (95–100% frequency) and 16 minority (5% frequency) mutations, spread across four independent replicated tests of four different Mycobacterium tuberculosis complex genomic DNA extracts.

FIGURE 3

Venn diagram representing the agreement between resistant phenotypes identified by four Mycobacterium tuberculosis resistance and susceptibility prediction tools: Deeplex Myc-TB, MTBSeq, Mykrobe and PhyResSE. WGS: whole-genome sequencing. The numbers of resistant phenotypes predicted by Deeplex Myc-TB analysis on 109 sputum samples from Djibouti and other analysis tools fed with WGS data from corresponding cultures are shown. ^#: two rifampicin resistance phenotypes predicted by MTBSeq and PhyResSE and/or Mykrobe based on rpoB S431T and D435V, reflecting probable WGS or culture contaminations (see text); ^¶: seven pyrazinamide resistance phenotypes predicted for “Mycobacterium canettii”-containing cultures by MTBSeq and Mykrobe based on panD M117T and pncA A46A, respectively; ⁺: one pyrazinamide resistance phenotype predicted by MTBSeq based on pncA D136G; ^§: 11 resistant phenotypes predicted by Deeplex Myc-TB based on 10 minority variants (3–12%) and one ethA frameshift-causing indel; ^ƒ: two streptomycin resistance phenotypes predicted by Deeplex Myc-TB and Mykrobe based on gidB G69D; ^##: two ethambutol resistance phenotypes predicted by Deeplex Myc-TB and PhyResSE based on embB S297A and Y319S.

FIGURE 4

Log read depth obtained by direct Deeplex Myc-TB testing of DNA extracted from clinical specimens collected in a tuberculosis drug resistance survey conducted in the Democratic Republic of the Congo: a) log read depth at each drug resistance-associated Deeplex Myc-TB target on the total set of 1494 sputum samples and b) log read depth at Deeplex Myc-TB targets according to acid-fast bacilli (AFB) microscopy grading of 1143 sputum samples with available microscopic examination data. Box and whisker plots show median with interquartile range (IQR) and minimum–maximum range with a maximum of 1.5 IQR, respectively .