Tuberculosis (TB) caused by Mycobacterium tuberculosis (MTB) accounts for an estimated 10 million new cases and 1.5 million TB-related deaths annually (World Health Organization, 2021). This situation was further complicated with drug-resistant TB emergence. Globally, around 3.4% of new TB cases and 18% of previously treated cases were classified as multidrug-resistant TB (MDR-TB) with resistance to both rifampicin (RIF) and isoniazid (INH), or RIF mono-resistant (RR-TB) strains. Approximately 6.2% of global MDR-TB/RR-TB cases could be further classified as extensively drug-resistant (XDR-TB) strains with additional resistance to fluoroquinolones and at least one additional Group A anti-TB drug (bedaquiline or linezolid; World Health Organization, 2021). According to a national survey back in 2007, approximately 10% of TB cases in China were classified as MDR-TB (Zhao et al., 2012). To ensure a better control of the global pandemic, a rapid, accurate and cost-effective diagnostic platform for TB detection and subsequent drug resistance identification is required.

Previous studies have explored the use of next-generation sequencing (NGS) for TB drug resistance determination, and the platform has proved its capability as a diagnostic tool for TB outbreak control and drug susceptibility predictions in clinical isolates (Walker et al., 2017; Satta et al., 2018; Comin et al., 2020). While these studies have demonstrated the feasibility of NGS in TB drug resistance diagnosis, the sequencing platforms are either obsoleted (Colman et al., 2016), or the study only focused on AFB positive specimens with limited variability in drug resistance patterns (Smith et al., 2020). More importantly, as most platform focused on whole genome sequencing data, high-quality genomic DNA prepared from MTB isolates are usually required (Lam et al., 2021; Olawoye et al., 2021). The long turnaround time of MTB culture would not be in favor of rapid MTB resistance analysis. Recently, several retrospective studies have demonstrated the efficacy of deep amplicon sequencing for culture-free MTB drug resistance detection using Deeplex Myc-TB platform with excellent performance among selected AFB-smear-positive clinical specimens (Kayomo et al., 2020; Bonnet et al., 2021; Jouet et al., 2021). However, with the low sensitivity of AFB smear test, a large proportion of MTB positive specimens with lower bacterial load could be missed, thus reducing the feasibility for rapid diagnosis of most drug-resistant specimens in actual clinical settings. A robust platform which is suitable for both AFB smear-positive and negative respiratory specimens is therefore in need to implementing the platform into routine clinical practice.

Our previous study introduced a highly sensitive and specific in-house IS6110-qPCR protocol for routine TB diagnosis with comparable performance to commercial platforms (Leung et al., 2018). In our practice, DNA extracts from IS6110-qPCR positive sputum were further tested for drug resistance-related mutation by Genotype MTBDRplus and MTBDRsl v2 (Hain Lifescience, Germany; MTBDR assay) for resistance-related mutation detection against RIF, INH, fluoroquinolones, and aminoglycosides. Sanger sequencing was performed on indeterminate results, such as specimens with absence of both wild-type probes and mutant probes (Lau et al., 2011; Siu et al., 2011a,b; Tam et al., 2019). Due to the high MDR-TB burden in our locality, false-resistant and false-susceptible cases could periodically be reported by line probe assay as a result of the vast variety of mutations in drug resistance-related genes (Javed et al., 2018). Although the discrepancies could be resolved by Sanger sequencing, the process is usually slow and labor intensive. To circumvent the problem, we recently developed a target amplicon sequencing protocol (TB-NGS) specifically targeting drug resistance-related genes in MTB isolates, which demonstrated complete concordance to conventional drug susceptibility testing (Tafess et al., 2020).

In order to reduce the turnaround time for drug-resistant MTB diagnosis, we extended TB-NGS into our rapid diagnostic platform to respiratory specimens. A total of 17 genes covering drug-resistant-related mutations of 8 classes of commonly used anti-TB drugs were selected, amplified and sequenced. As whole-genome data was not required by TB-NGS, data per sample could be greatly reduced and multiple specimens could be simultaneously analyzed in a single flow cell. The overall cost and turnaround time could be significantly reduced with the capability of direct application on respiratory specimen. In addition, we combined TB-NGS with our previously published IS6110-qPCR MTB diagnostic assay (Leung et al., 2018). The new routine diagnostic pipeline allows medium- to large-scale sample screening in daily clinical practice. For complicated TB cases such as retreatment cases and patients refractory to first-line treatment, TB-NGS could be performed to provide extra guidance.

In this study, we focused on the clinical utility of incorporating TB-NGS platform into routine settings. We performed a large-scale evaluation of TB-NGS workflow among respiratory specimens. Using respiratory specimens collected in Hong Kong and Guangzhou, the diagnostic performance of TB-NGS was assessed and compared with reference to the best available reference standard in our study cohort.

Conventional TB diagnosis is confirmed fundamentally on AFB culture and subsequent phenotypic DST. However, the entire procedures from sample collection to phenotypic DST require at least 5–12 weeks. Due to the long turnaround time of MTB culture, results are usually unavailable to clinicians for treatment guidance when starting empirical treatments. The long turnaround time is more evident among drug-resistant specimens due to the even slower growth of resistant MTB strains. Molecular approaches such as line-probe assays are timely and reliable genotypic DST methods, but increasing number of studies showed that false resistance could sometimes be reported due to the presence of naturally occurring polymorphisms (Coll et al., 2015; Hu et al., 2019). In addition, the presence of uncommon mutation might also result in discordant genotypic and phenotypic drug susceptibility results. For instance, MTB strains with rpoB L533P mutation might exhibit slightly elevated RIF Minimum Inhibitory concentration compared to fully susceptible strains but remain phenotypically susceptible when RIF critical concentration at 1 μg/ml was performed in MGIT SIRE as recommended by WHO guidelines (Tam et al., 2017; Shea et al., 2021). Similarly, while mutation at promoter region of inhA is generally believed to cause low-level INH resistance, concomitant mutation at inhA I194T could drastically increase INH resistance, rendering the drug useless even when a higher INH dosage was used (Machado et al., 2013). A rapid and reliable diagnostic method that reveals the identity of drug resistance-related mutation is thus urgently needed to fully explore the identity of the mutation.

The basis of TB-NGS was developed according to our previous study, which has demonstrated excellent reliability in identifying drug resistance-related mutations among culture isolates (Tafess et al., 2020). To extend the test to respiratory specimens, several modifications were implemented. First of all, instead of multiplex PCR of 17 gene regions in a single-tube, target amplification was separated into five multiplex PCR with each tube amplifying 4–5 regions of similar sizes to improve the amplification efficiency. To improve the sensitivity of TB-NGS especially on RIF, an additional target at rpoB covering RRDR and codon I572 was included. With these modifications, TB-NGS successfully sequenced 83/92 (90.2%) of the specimens enrolled in this study with 100% concordance when compared with phenotypic DST. For the remaining 9 specimens with unsuccessful TB-NGS results, they were all AFB smear-negative specimens with a Ct value ranged from 19.57–24.1. Meanwhile, most of the sequenced specimens have IS6110 Ct value of less than 18, indicating that the detection limit of TB-NGS would be equivalent to approximately 250 CFU/ mL according to the LOD analysis performed in our previous study (Leung et al., 2018). The higher demand in bacterial load might also explain the nine specimens with positive IS6110-qPCR but indeterminate TB-NGS results in this study.

Based on the TB-NGS result several uncommon mutation patterns were identified. One specimen (WC14) exhibited concurrent INH-resistance-related mutations at both inhA regulatory region c-15 t and structural region I194T with mutation frequency of 99.39 and 98.50%, respectively. Phenotypic DST revealed that the specimen was resistant to INH at 0.4 μg/ml, indicating high-level INH resistance instead of the low-level resistance commonly reflected by single mutation at inhA regulatory site. In addition, two MDR-TB specimens from the same patient (WC11and WC12) exhibited concurrent rpoB H526N and I572L mutations with mutation frequency of >95%. Mutation at rpoB H526N is normally regarded as a natural polymorphism (Jamieson et al., 2014), while rpoB I572L is a RIF resistant-related mutation located outside RRDR and is commonly not recognized by GeneXpert MTB/RIF Ultra and MTBDRplus assay. Similarly, one INH mono-resistant specimen defined by MTBDR assay was confirmed as MDR-TB by TB-NGS with rpoB I572F mutation. For PZA resistance detection, two PZA resistant specimen from one patient were identified in our study cohort. TB-NGS showed that the strain carried pncA V139G mutation which is a PZA resistance-related mutation commonly found in our locality as previously reported (Tam et al., 2019). The resistance profiles identified by TB-NGS in this study highlighted the strength of our diagnostic platform. Being an unbiased platform for identifying drug resistance mutations, TB-NGS could provide a much more comprehensive analysis on the resistance pattern of each specimen with a much shorter turnaround time. This would facilitate a judicious use of antibiotics especially in the treatment of MDR-TB.

In this study, three quasispecies were identified among wildtype specimens, of which two were rpoB mutations that could be correlated to phenotypic RIF resistance. These minor variants were consistent in multiple technical replicates, suggesting that these minor variants were unlikely to be sequencing artifacts. Phenotypic DST on these three quasispecies revealed pan-susceptible phenotypes, and the patients fully recovered after 6-months DOTS follow-up without complications. With a mutation frequency of less than 35%, these minor variants could be well distinguished from dominant resistant mutations which usually carried a mutation frequency of >60%. Therefore, these minor variants would not complicate empirical clinical judgment. Nevertheless, to fully elucidate the effect of these minor variants, a large-scale longitudinal study could be required to determine the progression of minor variant resistance developments.

TB-NGS generally requires a minimum turnaround time of 6 working days. Depending on the Illumina flow cell being used, TB-NGS can accommodate 12–96 samples per single run, which would fit low to high throughput analysis. Furthermore, with Bacteriochek having CE-IVD systems, the assay would be well suited in most clinical laboratories. While the platform is suitable for routine surveillance in low endemic regions in developed countries, a cost-effective protocol is highly recommended for high endemic regions as demonstrated in our study (Figure 1). Preliminary screening can first be performed for MTBC detection, followed by molecular genotypic resistance detection. For cases with resistant mutation or upon clinical suspicions, such as patients unresponsive to treatment or requiring additional support among treatment defaulters and retreatment cases, TB-NGS can be performed to provide a comprehensive drug resistance profile. Applying this algorithm to our study, only 25 out of the 4,000 specimens would require additional aid from TB-NGS. More importantly, our diagnostic algorithm missed no drug-resistant case as indicated by phenotypic DST results on all MTBC culture-positive specimens. The implementation of our proposed algorithm would therefore greatly reduce the workload and financial burden in clinical microbiology laboratories, especially in high endemic regions with limited resources.

Despite the potential of TB-NGS for better MTB management support, the assay has its limitations. The current platform is based on the knowledge of resistance-related mutation. For novel mutations located outside targeted sites they will not be detected. Nevertheless, the current test has included all well-established resistance-conferring mutations which would have been suitable for normal routine clinical practice. On the other hand, while the guideline updated in late 2020 included bedaquiline as one of the medications for XDR-TB definition, our current TB-NGS protocol did not include bedaquiline genotypic resistance test due to several reasons. As a new anti-TB drug recommended by WHO in 2016, the exact mechanism for bedaquiline resistance pattern in clinical isolates are still far from fully elucidated (Nguyen et al., 2018; Nieto Ramirez et al., 2020; Gomez-Gonzalez et al., 2021). In addition, the occurrence of bedaquiline resistance TB strains in our locality is still extremely rare. Among 4,043 specimens collected in this study, only 6 pre-XDR-TB specimens were identified while still susceptible to PZA, LZD and aminoglycosides. Genotypic resistance test would still be necessary in further development of TB-NGS platform due to increasing use of bedaquiline in complicated TB cases as well as the emergence of bedaquiline resistant strains (Chesov et al., 2021). Lastly, while other sequencing platforms such as Nanopore might be a more rapid and budget-friendly approach, we chose Illumina as the primary sequencing platform due to its high single read accuracy for the investigation of minor variants in our study cohort. Evaluation of TB-NGS by Nanopore in a multi-center analysis is therefore warranted to further extend the utility of TB-NGS for more flexible and cost-effective routine service.

To conclude, TB-NGS is an accurate platform for TB drug resistance detection. When accompany with our proposed diagnostic algorithm, the workflow would provide a cost-effective diagnostic solution for complicated TB cases and allow early detection of specimens with potential drug resistance developments.

The data presented in this study are deposit in NCBI database, Accession Number SAMN30429057-SAMN30429139.

This study has been approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (Ref. number: UW 12-309).

KL, KT, TN, and GS were responsible for designing the study, performing the experiments and writing the manuscript. KL was responsible for bioinformatic and statistical analysis of this study. H-YL and RS were responsible for doing the experiments. OM, S-HY, J-XC, and QH were responsible for providing processing specimens from mainland China. W-CY is the corresponding author and was responsible of supervising the study. All authors contributed to the article and approved the submitted version.

This study is supported by the Science and Technology Planning Project of Guangdong Province, China (grant number: 2019B121205010).

OM, S-HY, J-XC and QH were employed by KingMed Diagnostics.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.