This prospective multicenter, cross-sectional study enrolled six hospitals located in the Guangxi Zhuang Autonomous Region, demonstrating a high TB incidence in China (Yang et al., 2022). The six enrolled hospitals were as follows: the Guangxi Chest Hospital, the Fourth People’s Hospital of Nanning, the Third People’s Hospital of Guilin, the First People’s Hospital of Fangchenggang, the Third People’s Hospital of Wuzhou, and Baise People’s Hospital.

In total, 253 patients were screened in this study from those hospitals between January 2022 and July 2023. NanoTNGS assays were performed directly from the clinical specimens among all the participants. Other eligibility criteria include a positive culture result or a positive result from Xpert MTB/RIF.

We collected 253 specimens in sterile specimen collection tubes from patients with suspected DR-TB for laboratory testing at the participating hospitals. Several laboratory methods were employed, including the acid-fast bacillus (AFB) smear, Xpert MTB/RIF, NanoTNGS, culture-based Mycobacterium identification, and pDST, depending on specific laboratory tests used in different hospitals. A total of 184 samples from the screened population met the eligibility criteria. Patient information, including gender, age, body mass index (BMI), nationality, previous therapeutic regimen of anti-TB drugs, smoking history, and symptoms, were also collected. The results of NanoTNGS were analyzed by individuals who were blind to the pDST and Xpert MTB/RIF results. The laboratory test results were confirmed with the electronic medical record in each hospital at the end of the study.

Sputum samples were decontaminated, liquefied, and lysed using N-acetyl-L-cysteine (NALC)-NaOH at a final concentration of 2% (Leung et al., 2022). The bronchoalveolar lavage fluid (BALF) samples were then centrifuged at low speed (3,000 rpm for 10 min) to enrich the cells. The enriched cells were then adequately ground and digested with lysozyme according to standard laboratory procedures. After sample pre-treatment, DNA extraction was performed using the QIAamp DNA Microbiome Kit (Cat. No. 51707, Qiagen, Hilden, Germany) according to the manufacturer’s protocol. Blank EB buffer was used as a negative control for nucleic acid extraction. The concentration of extracted DNA was measured using the Invitrogen Qubit 4 Fluorometer and then stored at −20°C for subsequent library preparation.

The library preparation for the NanoTNGS assay involved a two-step polymerase chain reaction (PCR) process. A total volume of 30 μL of PCR mixture included 50 ng DNA extract, 3 μL of primers with a concentration of 0.2 pmol/μL, and 15 μL of Multiplex PCR Master Mix (Cat. No. BR0200801, Biotechrabbit GmbH, Berlin, Germany) during the first PCR step. The primers were designed to target 16S rDNA and hsp65 for the identification of Mycobacterial species and rpoB, katG, thyA, ahpC, alr, eis, embA, embB, ethA, folC, gibB, gyrA, gyrB, inhA, pncA, rplC, rpsL, rrs, rv0678, atpE, and tlyA for the prediction of resistance genes. The types of anti-TB drugs that correspond to the resistant gene regions are shown in Supplementary Table 1. The amplifying reaction was carried out as follows: pre-denaturation at 95°C for 3 min, followed by 35 cycles of denaturation at 95°C for 15 s, annealing at 60°C for 45 s, extension at 72°C for 15 s, and a final extension at 72°C for 3 min.

The products obtained from the first amplification step were then subjected to Nanopore PCR Barcode Expansion using the same Multiplex PCR Master Mix. The reaction conditions for this step were pre-denaturation at 95°C for 3 min, followed by 35 cycles of denaturation at 95°C for 30 s, annealing at 64°C for 30 s, extension at 72°C for 1 min, and a final extension at 72°C for another 3 min.

To assess the concentration of the purified library after this process, we used the Equalbit 1 × dsDNA HS Assay Kit (Cat. No. EQ121-01, Nanjing Vazyme Biotech, Nanjing, CN). An end-tail repair was performed using the End Prep Mix 4 from the VAHTS Universal End Preparation Module for Illumina V2 (Cat. No. N203, Nanjing Vazyme Biotech, Nanjing, CN). Adapter ligation was carried out using the Rapid DNA Ligase and Rapid Ligation Buffer from the VAHTS Universal DNA Library Prep Kit for Illumina V3 (Cat. No. ND607, Nanjing Vazyme Biotech, Nanjing, CN). Ligation Sequencing Kit V14 (SQK-LSK114, Oxford Nanopore Technologies, Oxford, United Kingdom) was employed to construct the pooled libraries. Following each amplification step, the VAHTS DNA Clean Beads (Cat. No. N411-01, Nanjing Vazyme Biotech, Nanjing, CN) were utilized for purification.

Finally, 12 μL of the pooled library was loaded into a nanopore flow cell (R9.4.1) for sequencing on the GridION sequencing platform. Base calling data were generated using MinKNOW version 2.0, and reads of small size and/or low quality were removed using NanoFilt. Mycobacterium tuberculosis was considered positive when at least one read was mapped to either the species or genus level (Miao et al., 2018). Antibiotic-resistance genes detected by targeted nanopore sequencing were predicted using the Comprehensive Antibiotic Resistance Database (CARD), the ResFinder, or the TBProfiler database with default alignment settings (≥80% identity over ≥60% of the length of the target gene; Yonkus et al., 2022). Thresholds for variant calling were set at allele frequency greater than 10%, depth greater than 10, and allele depth greater than 5 (CRyPTIC Consortium and the 100,000 Genomes Project et al., 2018).

The sputum and BALF were pre-treated with NALC-NaOH and subjected to vortex oscillation. The suspension was adjusted to a final concentration of 2%. A 100 μL of suspension was then incubated on the upper and middle parts of the Löwenstein–Jensen agar slant for up to 8 weeks. Slides prepared for direct microscopy were stained with the Ziehl–Neelsen method after heat fixation and were evaluated semi-quantitatively as negative, +, ++, and +++ (Pfyffer, 2015). Heparin was added to the hydrothorax and cerebrospinal fluid for anticoagulation, but no other pre-treatment was performed prior to DNA extraction, culture, and microscopy.

Suspensions of isolates from positive samples were incubated on 7H10 agar using an agar proportion assay, following the methods of the Clinical and Laboratory Standards Institute (CLSI). The pDST was carried out on drugs including RIF, isoniazid, ethambutol (EMB), streptomycin, kanamycin, amikacin, capreomycin, moxifloxacin, or levofloxacin. In cases where the laboratory tests yielded negative cultures, a recent, preceding DST was used for comparison with the NanoTNGS results.

The sensitivity of each drug was determined by computing the ratio of samples identified as resistant by the pDST, and the samples were also identified as resistant by the Xpert MTB/XDR or NanoTNGS assay (Penn-Nicholson et al., 2022). The specificity was computed by the ratio of samples identified as susceptible according to both the pDST and Xpert MTB/XDR results. Continuous data without normal distribution were presented using medians and interquartile ranges (IQRs), while categorical variables were described using proportions. All analyses were conducted using IBM SPSS Statistics 24.

During the period between January 2022 and July 2023, a total of 253 suspected DR-TB patients were screened for eligibility. A total of 69 participants were excluded from the downstream analysis mainly due to negative culture tests, missing culture and Xpert MTB/RIF results or low quality of genomic DNA extraction (Figure 1). Only samples with positive culture results will be subjected to subsequent phenotypic drug susceptibility tests to compare with the prediction of resistance by NanoTNGS. Therefore, we excluded 48 cases of culture-negative or result-missing results. Finally, 184 patients were enrolled in the study. The samples included 90 (48.91%) BALF, 89 (48.37%) sputum, 2 mixed BLAF and sputum (0.59%), and 3 (1.63%) other sample types (hydrothorax, ascites, and pus).

The median age of the patients was 48 years, with 77.13% being male individuals and 15.21% being current smokers. The Han population accounted for the majority at 65.76%, followed by Zhuang people at 29.35%, and other ethnic minorities, including Miao, Dong, Yao, Maonan, and Tibetan, at a combined percentage of 4.89%. Among the enrolled participants, 46.20% (85 out of 184) were underweight with a BMI below <18.5, indicating increased nutrient consumption needs. The underweight population, as defined by BMI, was associated with a higher risk of TB incidence (Cho et al., 2022). These participants exhibited typical symptoms associated with TB, such as cough (79.35%), expectoration (58.15%), fever (9.24%), and hemoptysis (7.61%). All the patients presented imaging evidence of TB infection: cavity findings in 52.17% cases, calcification in 15.76%, pleural thickening and adhesion in 27.17%, fibrosis in 26.63%, and patchy, strip-like shadows in 2.17%. The results of the erythrocyte sedimentation rate and C-reactive protein demonstrated a possible inflammation reaction (Table 1).

Of these enrolled participants, 130 (86.67%) out of 150 were smear positive for TB, and 180 (97.83%) out of 184 were Mycobacterium tuberculosis complex (MTBC) culture positive (Figure 2A). There were four samples that did not grow in the culture medium but were detected as positives via AFB, NanoTNGS, and Xpert MTB/RIF. Two of four patients have a history of anti-TB treatment, which may lead to false-negative culture results. Other factors contributing to Mycobacterium tuberculosis culture-negative results include low bacterial load, delays in specimen processing, and decontamination with NaOH (Traore et al., 2006; Madico et al., 2019). NanoTNGS identified 183 (99.46%) MTBC-positive cases from 184 samples. The positive number detected by Xpert MTB/RIF was 132 (88.00%, 132/150). For the one NanoTNGS missed sample, MTBC cases were identified via culture and AFB smear microscopy. Overall, NanoTNGS assays detected 179 MTBC positive cases among 180 samples (99.44%) confirmed by the MTBC culture, demonstrating a high level of concordance between the two methods.

There were 171, 141, and 119 drug-resistant samples detected by NanoTNGS, pDST, and Xpert MTB/RIF, respectively (Figure 2B). In four cases, drug resistance was confirmed only through pDST, while NanoTNGS did not detect drug-resistance mutations in these cases. This aspect might be because of the low level of drug-resistant gene variation. The consumable cost of NanoTNGS was approximately RMB¥ 800–900 per sample. The turnaround time for NanoTNGS was 12 h from DNA extraction to test result delivery, at least 6 weeks earlier than pDST. The workflow of NanoTNGS is shown in Supplementary Figure 1.

The drug-related resistant gene mutations were detected by NanoTNGS directly from clinical specimens. Resistance genes were found in 171 patients for the following drugs: RIF (163 cases), isoniazid (120 cases), EMB (65 cases), streptomycin (56 cases), fluoroquinolones (43 cases), pyrazinamide (32 cases), aminoglycosides (kanamycin, amikacin, capreomycin, 5 cases), and bedaquiline (1 case). The average coverage depth of resistance genes ranged from 295× to 1,777× across 10 common gene regions detected by NanoTNGS (Supplementary Figure 2). The coverage depth of each sample is shown in Supplementary Table 2.

In total, 176 rpoB gene mutations associated with potential RIF resistance were identified from the 163 samples by NanoTNGS. Among those mutations, 18 samples had multiple rpoB gene mutations, suggesting that different drug-resistance mechanisms may exist. The most frequent mutation was rpoB S531L, which accounted for 47.7% (84/176) gene mutations (Figure 3A). Other frequently observed mutations included L511P (9%, 16/176), D516V (4%, 7/176), H526N (4%, 7/176), and H526D (4.5%, 8/176).

The previous study showed that the katG gene at position p.S315T was associated with high-level isoniazid resistance (Finci et al., 2022). In the present study, we identified 134 gene mutations related to isoniazid resistance from 120 samples. The most common mutation was katG S315T, which was observed in 84 of 134 samples (Figure 3B). Mutations located upstream of the inhA and ahpC promoters were also detected. The most common upstream mutations (c-15 t and t-8c) in the inhA promoter region were observed in 21 of the 120 resistant isoniazid samples.

Relatively fewer resistance mutations were detected for EMB (n = 65), fluoroquinolones (n = 49), and streptomycin (n = 56). The most common mutations in the EMB-resistance gene were embB M306V (26/65) and M306I (19/65; Figure 3C). Meanwhile, 32 patients carried 35 pyrazinamide resistance-conferring pncA gene mutations. The mutations were dispersed in the pncA gene regions. The detailed gene mutation locus is shown in Figure 3D. The most common resistance gene mutation related to streptomycin was rpsL K43R (Figure 3E). For fluoroquinolones, the most frequent mutations were gyrA D94G (19/46) and D94A (8/46; Figure 3F). There were two gyrB T500N mutations and one gyrB A504T mutation potentially informing fluoroquinolone resistance. The depth, allele depth, and allele frequency of the top three variants in resistance genes detected via NanoTNGS are shown in Supplementary Table 3. The minimum average allele depth was 75.5×, which is greater than the 5× cutoff for variant call (Supplementary Figure 3). In addition, there were five rrs gene mutations that were associated with amikacin, kanamycin, and capreomycin resistance in five samples. One Rv0678 D94Q mutation was detected by the NanoTNGS assay, indicating potential resistance to bedaquiline (Supplementary Table 4). Identification of resistance gene mutations by NanoTNGS allows comprehensive profiling of the resistance mechanism and interpretation of genotypic DST results.

RIF-resistance prediction achieved a sensitivity of 93.53% compared to the composite reference standard (CRS), while Xpert MTB/RIF-resistance prediction had a lower sensitivity of 83.80%. The sensitivity of NanoTNGS for predicting resistance was 89.72% for isoniazid, 85.45% for EMB, 75.22% for streptomycin, and 88.89% for fluoroquinolones, respectively, compared to pDST (Table 2).

The specificity for NanoTNGS-resistance detection was high for streptomycin (92.59%), fluoroquinolones (91.67%), EMB (89.42%), and for isoniazid (84.91%). For RIF, NanoTNGS achieved a specificity of 100% compared with CRS. Overall, the NanoTNGS achieved high specificity when detecting resistance genes.