Rapid and economical drug resistance profiling with Nanopore MinION for clinical specimens with low bacillary burden of Mycobacterium tuberculosis

Objective We designed and tested a Nanopore sequencing panel for direct tuberculosis drug resistance profiling. The panel targeted 10 resistance-associated loci. We assessed the feasibility of amplifying and sequencing these loci from 23 clinical specimens with low bacillary burden. Results At least 8 loci were successfully amplified from the majority for predicting first- and second-line drug resistance (14/23, 60.87%), and the 12 specimens yielding all 10 targets were sequenced with Nanopore MinION and Illumina MiSeq. MinION sequencing data was corrected by Nanopolish and recurrent variants were filtered. A total of 67,082 bases across all consensus sequences were analyzed, with 67,019 bases called by both MinION and MiSeq as wildtype. For the 41 single nucleotide variants (SNVs) called by MiSeq with 100% variant allelic frequency (VAF), 39 (95.1%) were called by MinION. For the 22 mixed bases called by MiSeq, a SNV with the highest VAF (70%) was called by MinION. With short assay time, reasonable reagent cost as well as continuously improving sequencing chemistry and signal correction pipelines, this Nanopore method can be a viable option for direct tuberculosis drug resistance profiling in the near future.


Introduction
The increasing threat of tuberculosis (TB) drug resistance highlights the importance of prompt drug susceptibility test (DST) results for better patient care and infection control [1,2]. Nevertheless, culture-dependent methods cannot provide a quick answer due to the fastidious nature of Mycobacterium tuberculosis (MTB). From literature, 24-61% of pulmonary TB cases were acid-fast bacilli (AFB) smear-negative [3,4], with smearnegative, culture-positive TB accounting for 13% of TB transmission [5]. Novel diagnostic tools are needed for rapid detection of drug resistance from the technically demanding smear-negative specimens.
Recent advent of next-generation sequencing (NGS) has facilitated comprehensive evaluation of MTB genome for drug resistance prediction [6]. Among various options in NGS market, Nanopore sequencers are ideal for infectious disease diagnosis, which requires short sampleto-answer time (Fig. 1). Despite its inferior sequencing accuracy [7], several groups utilized Nanopore MinION and successfully identified single nucleotide variants Open Access BMC Research Notes *Correspondence: bsftang@gmail.com 1 Department of Pathology, Hong Kong Sanatorium & Hospital, Hong Kong, China Full list of author information is available at the end of the article (SNVs) in Plasmodium falciparum [8], dengue virus [9] and chronic lymphocytic leukemia [10,11].
The goal of this study was to design and test a Nanopore targeted panel for direct TB drug resistance profiling. First, we attempted to amplify 10 resistance-associated loci from clinical specimens with low MTB burden. Second, we sequenced these amplicons with both Nanopore MinION and Illumina MiSeq, and the sequencing data was collated.

Clinical specimens
Twenty-three specimens were collected between August 2016 and May 2017 (Table 1). They were AFB smearnegative or classified as MTB detected low/very low by Xpert MTB/RIF assay (Cepheid, Sunnyvale, CA, USA). TB culture and DST were performed by 2 local reference laboratories, with ethambutol, isoniazid, rifampicin and streptomycin for first-line drug testing.

DNA extraction
Standard laboratory practice was applied to minimize the risk of infection and contamination. NucliSENS easyMAG automated system (bioMérieux, Marcy, I'Etoile, France) was used for DNA extraction. A maximum of 1-mL pretreated specimen was homogenized in lysis buffer and incubated at 80 °C for 20 min. DNA extraction was performed according to manufacturer's recommendations, with an elution volume of 25 μL.

Sequencing by MiSeq
For each specimen, 2-μL aliquots of each amplicon were pooled for Nextera XT DNA library preparation

Table 1 Details of clinical specimens, routine test results and amplification of genomic regions associated with TB drug resistance
AFB acid-fast bacilli, BA bronchial aspirate, Bx biopsy, eis P eis promoter, FNA fine needle aspirate, Ind indeterminate, INH isoniazid, inhA P : inhA promoter, LN lymph node, MTB M. tuberculosis, N/A not available, ND not detected, Neg negative, Pos positive, RIF rifampicin resistance, SM streptomycin, TAT turnaround time a Four first-line antibiotics were tested by local reference laboratory, including ethambutol, isoniazid, rifampicin and streptomycin b The 10 loci were eis promoter, embB, gyrA, inhA promoter, katG, pncA, rpoB, rpsL, rrs and tlyA genes c Negative by first PCR, positive by second PCR d eis promoter and rrs: negative by first PCR, positive by second PCR  [20]. IGV was used for manual inspection of BAM datasets. Nanopolish variants (Galaxy version 0.1.0) was used for signal-level variant calling [21].

Routine test results
Nineteen specimens were AFB smear-negative, and the only smear-positive specimen was 'MTB detected low' (Patient 19) (
The MTB isolate from Specimen 14 was resistant to streptomycin, with wildtype rpsL and rrs genes. From literature, 25-52% of streptomycin-resistant MTB isolates harbored wildtype rpsL and rrs genes [23][24][25]. Other molecular mechanisms, such as reduced cell permeability to aminoglycosides or presence of drug-modifying enzymes, might contribute to its streptomycin resistance. On the other hand, 281A > G was called in gyrA gene with variant allelic frequency (VAF) of 11%. The resulting D94G substitution is frequently found in fluoroquinolone-resistant strains [26], yet we could not confirm the resistance phenotype as second-line drug susceptibility data was not available.
For Specimen 16, 128A > G was called in rpsL gene. We did not have any streptomycin susceptibility data for interpretation as TB culture was not performed.

General features
Twelve pools of amplicons were sequenced separately on flow cells with 84-624 active pores (Additional file 1: Table S3). First FASTQ files were generated with an elapsed time of 5-86 min, 19.8 min in average, yielding 5.2-5.8 Mb of data and 2866-4685 'pass' reads. Mean coverage breadth and depth across all targets were 100% and 282.8, respectively.

Sequencing error analysis
General alignment error rate was 10.47-15.12%, and mean insertion and deletion error rate (per 100 aligned bases) were 2.09% and 2.83%, respectively. The raw data error rate was comparable to previous studies for chronic lymphocytic leukemia [10,11].

Comparison with MiSeq sequencing data
Nanopolish was used to improve the accuracy of Min-ION sequencing data from signal level. For rpsL amplicons of Patient 3 and 21, second and third FASTQ files were also included for data analysis to meet the default minimum depth requirement.

Filtering recurrent variants
From 'nanopolished' data, recurrent variants were called across different samples, which was also observed in other studies using Nanopolish [10], nanocorrect and Amplicon Long-read Error Correction (ALEC) python script [11] for data processing. In both studies, the authors filtered recurrent variants to improve sequencing accuracy. By applying this strategy, the concordance for wildtype bases increased from 99.9% (66,935/67,019) to 100%, and insertions and deletions were reduced from 32 to 8 and 44 to 4, respectively (Additional file 1: Table S5).

Conclusions
We developed a Nanopore targeted panel for direct detection of TB drug resistance, with full-set sequence data retrieved from about half of the specimens with low MTB burden. The assay time was 6-9 h, which improved an average of 69.5 days by DST, and was comparable to commercial methods like line probe assay (Fig. 1). The method was merited by reasonable reagent cost (64 USD per sample for 24-plex workflow), which could be further lowered with the 96-barcode option and Flongle flow cells [28,29]. With continuously improving sequencing chemistry, more sophisticated signal correction pipelines and the above-mentioned merits, we envision that Nanopore sequencing can be a viable option for 'end TB' in the near future.

Limitations
1. An increased sample size may better estimate the PCR success rate. 2. We could not correlate the data with second-line DST, which was not routinely performed for nonmultidrug-resistant MTB isolates. 3. As the MinION flow cells possessed suboptimal number of active pores, the sequencing time might be overestimated. 4. Nanopore sequencing accuracy might be improved with '1D 2 ′ chemistry, at the expense of lower sequencing depth [30]. 5. Nanopolish could not improve the detection of minor variants, which might be caused by high error rate of Nanopore raw data. Sequencing error might originate from inhomogeneous translocation speed of DNA, low signal-to-noise ratio and simultaneous passage of multiple nucleotides through nanopores [31]. Nevertheless, substitution-type miscalls might also be present in MiSeq data, which might arise from similar emission spectra of fluorophores, phasing and pre-phasing phenomena, and index PCR of library preparation [32,33]. In addition, polymerase error accumulated in target amplification could affect the accuracy of variant calling by both MiSeq and MinION. 6. Therefore, we should be vigilant that the variants called by MiSeq could be false whereas the variants called by MinION could be true in some occasions. It is advisable to confirm the presence of these variants by repeating the target amplification step followed by Sanger sequencing.