Same-Day Diagnostic and Surveillance Data for Tuberculosis via Whole-Genome Sequencing of Direct Respiratory Samples

Abstract

Routine full characterization of Mycobacterium tuberculosis is culture based, taking many weeks. Whole-genome sequencing (WGS) can generate antibiotic susceptibility profiles to inform treatment, augmented with strain information for global surveillance; such data could be transformative if provided at or near the point of care. We demonstrate a low-cost method of DNA extraction directly from patient samples for M. tuberculosis WGS. We initially evaluated the method by using the Illumina MiSeq sequencer (40 smear-positive respiratory samples obtained after routine clinical testing and 27 matched liquid cultures). M. tuberculosis was identified in all 39 samples from which DNA was successfully extracted. Sufficient data for antibiotic susceptibility prediction were obtained from 24 (62%) samples; all results were concordant with reference laboratory phenotypes. Phylogenetic placement was concordant between direct and cultured samples. With Illumina MiSeq/MiniSeq, the workflow from patient sample to results can be completed in 44/16 h at a reagent cost of £96/£198 per sample. We then employed a nonspecific PCR-based library preparation method for sequencing on an Oxford Nanopore Technologies MinION sequencer. We applied this to cultured Mycobacterium bovis strain BCG DNA and to combined culture-negative sputum DNA and BCG DNA. For flow cell version R9.4, the estimated turnaround time from patient to identification of BCG, detection of pyrazinamide resistance, and phylogenetic placement was 7.5 h, with full susceptibility results 5 h later. Antibiotic susceptibility predictions were fully concordant. A critical advantage of MinION is the ability to continue sequencing until sufficient coverage is obtained, providing a potential solution to the problem of variable amounts of M. tuberculosis DNA in direct samples.

Keywords: DNA sequencing; Mycobacterium tuberculosis; antibiotic resistance; diagnostics.

FIG 1

Concentrations of DNA extracted from MGIT cultures and direct clinical samples. Each dot represents a single extraction. The horizontal line at 0.2 ng/μl represents the DNA concentration theoretically required for MiSeq library preparation. The horizontal line at 0.05 ng/μl represents the minimum DNA concentration used for MiSeq library preparation from direct samples in this study. One sample not shown as DNA was below the limit of detection.

FIG 2

Proportions of reads assigned to various species categories in each sample sorted by increasing total counts of M. tuberculosis complex (MTBC) reads. (a) Direct samples show that removal of human DNA (red) has been broadly successful, but removal of NPF (green) and other bacteria (purple) had more variable success. (b) MGIT samples show much more uniform dominance of M. tuberculosis reads, as expected after 2 weeks of culture designed to favor mycobacterial growth.

FIG 3

Recovery of M. tuberculosis genome in direct samples and robustness to contamination. (a) Depth versus proportion of the M. tuberculosis reference recovered (at >5× depth). The vertical dotted line at 3× depth is the threshold used for resistance prediction in this study. (b) Proportion of contamination (reads not mapping to M. tuberculosis reference) versus proportion of genome recovered. Samples with <95% of the M. tuberculosis genome recovered all have >75% contaminated reads.

FIG 4

Genotypic concordance between direct and paired MGIT samples. (a) Histogram of genetic (SNP) differences, excluding the one pair that differ by 1,106 SNPs; the median (and modal) difference is 1, and thus, direct sequencing is identifying the same strain of M. tuberculosis that culture-based sequencing would. (b) Placing direct/MGIT pairs on a phylogenetic tree of 3,480 samples shows the distribution of samples across world diversity. A circle indicates the sequence from the direct sample, and an X indicates the sequence from the corresponding MGIT sample. For the 1 pair (of 17) with 1,106 differences (turquoise), the MGIT sample is placed very close to the other samples (zero SNP differences from one [MGIT] sample and five SNP differences from the others) and so is possibly due to a labeling error.

FIG 5

Timelines. We compared the method of Brown et al. with the results of this study obtained with Illumina MiSeq and MiniSeq and ONT MinION. We assumed that no step of the process can be initiated after 6 p.m. or before 8 a.m. The method of Brown et al. has a rapid extraction step but also a 20-h overnight enrichment step, resulting in a 50-h turnaround time. In our study, we did 27-h MiSeq runs (paired 150-bp reads), but since Mykrobe is k-mer based, a 16-h run (paired 75-bp reads) would give equivalent results; we therefore display that potential timeline here. The DNA extraction process was updated for the MiniSeq and MinION experiments, removing the ethanol precipitation step. In normal use, this would take 3 h. The 1.5-h orange rectangle on the MinION time lines includes both PCR and the 10-min sample preparation step. In this experiment, since we used spiked BCG DNA in sputum, we did not use a human depletion step, thus taking only 2 h. This image is intended to show comparable real-use time lines, and so the MiniSeq/MinION time lines are shown with 3-h extraction steps. MiniSeq enables a 16-h turnaround time by sequencing for only 7 h. R9 MinION also delivers sub-24-h results but requires one flow cell per sample. R9.4 MinION gives a 12.5-h turnaround time (6 h of sequencing with real-time [i.e., simultaneous] base calling when used on a single sample).