The study collected 75 M. tuberculosis strains, including 35 fluoroquinolone-resistant and 40 fluoroquinolone-sensitive strains, which were collected from the Urumqi Municipal Centre for Disease Control and Prevention. The study was approved by the Ethics Committees of Urumqi Municipal Centre for Disease Control and Prevention. Subjects from Urumqi Municipal Centre for Disease Control and Prevention provided written consent to participate in the study, including consent for the storage of M. tuberculosis strains isolated from sputum samples for future testing.

The pC013-Twinstrep-SUMO-huLwCas13a plasmid was a gift from Feng Zhang (Addgene plasmid # 90097).¹ Briefly, the pC013-Twinstrep-SUMO-huLwCas13a plasmids were transformed into Escherichia. coli BL21 (DE3), protein expression was induced with 0.5 mM IPTG (Sangon Biotech), at 18°C for 16 h. Cell pellet was re-suspended in lysis buffer (20 mM Tris-HCl, 500 mM NaCl, 1 mM DTT, pH 8.0) supplemented with protease inhibitors (Ultra EDTA-free tablets), lysozyme, and benzonase followed by sonication. Lysate was cleared by centrifugation and filtered through a Stericup 0.22 μm filter (Millipore). Filtered supernatant was applied to StrepTactin Sepharose (IBA) and incubated with rotation for 2 h followed by washing of the protein-bound StrepTactin resin three times in lysis buffer. The resin was re-suspended in SUMO digest buffer (30 mM Tris-HCl, 500 mM NaCl 1 mM DTT, 0.15% Igepal (NP-40), pH 8.0) along with 100 μL (10 U/μL) of SUMO protease (Hai gene) and incubated overnight at 4°C with rotation. The protein eluate was isolated by spinning the resin down. The supernatant was concentrated via a Centrifugal Filter Unit to 1 mL in S200 buffer (10 mM HEPES, 1M NaCl, 5 mM MgCl₂, 2 mM DTT, pH 7.0), and was further loaded onto a gel filtration column (Superdex^® 200 Increase 10/300 GL, GE Healthcare Life Sciences) via fast protein liquid chromatography (FPLC). The resulting fractions from gel filtration were analyzed by SDS-PAGE and fractions containing LwCas13a were pooled and buffer exchanged into Storage Buffer (600 mM NaCl, 50 mM Tris-HCl, 5% glycerol, 2 mM DTT, pH 7.5) and frozen at −80°C for storage.

Partial sequence of quinolone resistance-determining region (QRDR) in the gyrA gene of M. tuberculosis H37Rv was obtained by synthesis (Tsingke Biotechnology) (sequence listed in Supplementary Table 1). A variety of QRDR plasmids containing fluoroquinolone resistance related mutations were obtained using Fast Site-Directed Mutagenesis Kit (TIANGEN, KM101). Nucleic acid targets were amplified with 2 × Taq DNA Polymerase (Invitrogen) via PCR with primers appended T7 promoter. Purified dsDNAs was incubated with T7 polymerase overnight at 37°C using the HiScribe T7 High Yield RNA Synthesis kit (New England Biolabs). Transcription reactions were quenched to reduce template DNA via treatment with DNase I (TAKARA) at 37°C for 15 min and then ssRNAs subsequently purified with the HiPure RNA Pure Micro Kit (Magen).

For preparation of crRNAs, ssDNA (Tsingke Biotechnology) (sequence listed in Supplementary Table 2) complementary to the crRNA sequence was ordered with an appended T7 promoter sequence. crRNA DNA was annealed to a short T7 primer (final concentrations 10 μM) to produce a partial dsDNA to increase T7 RNA polymerase transcription efficiency. The process of partial dsDNA as template transcribing into RNA was same as mentioned above.

Detection assays were performed with 450 nM purified LwCas13a, 900 nM crRNA, 120 nM quenched fluorescent RNA reporter (RNAse Alert v2, Thermo Scientific), 0.5 μL murine RNase inhibitor (New England Biolabs), 100 ng of background total human RNA (purified from HEK293FT culture), and 1,800 nM of input nucleic acid target in nuclease assay buffer (40 mM Tris-HCl, 60 mM NaCl, 6 mM MgCl₂, pH 7.3) to 50 μL final volume. Reactions were incubated in a fluorescence plate reader (BioTek) for up to 60 min at 37°C with fluorescence measurements taken every 1 min (λex: 490 nm; λem: 520 nm).

The integrated reaction combining T7 RNA polymerase transcription and LwCas13a collateral detection consisted of 450 nM LwCas13a, 900 nM crRNA, 120 nM quenched fluorescent RNA reporter (RNAse Alert v2, Thermo Scientific), 0.5 μL murine RNase inhibitor (New England Biolabs), 100 ng of background total human RNA (purified from HEK293FT culture), and increasing amounts of DNA input, 0.5 μL T7 RNA polymerase Mix (New England Biolabs), 2.5 μL NTP Buffer Mix (New England Biolabs, E2050S) in nuclease assay buffer (40 mM Tris-HCl, 60 mM NaCl, 6 mM MgCl₂, pH 7.3) to 50 μL final volume. Reactions were allowed to proceed for 80 min at 37°C on a fluorescent plate reader (BioTek) with fluorescent kinetics measured every 1 min (λex: 490 nm; λem: 520 nm).

Indirect drug-susceptibility testing was performed from the positive M. tuberculosis culture with the use of the proportion method. The critical concentrations used for each drug were 4 μg per milliliter for ofloxacin, 1 μg per milliliter for moxifloxacin, 2 μg per milliliter for levofloxacin. Briefly, prepare 1 mg/ml bacillary suspension (McFarland No. 1), further diluted in sterile distilled water to give 10^–2 and 10^–4 dilution. Drug containing media and plain control median were inoculated with 1 loop-full bacillary suspension in indicated dilution. The slopes were incubated at 37^°C for 4 weeks. Resistance proportion (clones in drug containing medium/clones in control medium) > 1% were considered as resistance to certain drug. Para-nitrobenzoic acid (PNB) and thiophen-2-carboxylic acid hydrazide (TCH) growth tests were further done to rule out the non-tuberculous mycobacteria (NTM). Drug containing medium were provided by BASO diagnostics, Inc., Zhuhai, China.

Partial sequence of quinolone resistance-determining region (QRDR) in the gyrA of Mycobacterium tuberculosis H37Rv was 300 nt (7,431–7,730) and synthesized by Tsingke Biotechnology. Primers (T7F and T7R) for amplification of plasmids or clinical specimens including this region used in the study were shown in Supplementary Table 1. Primers for constructing template plasmids related to fluoroquinolone tolerance of M. tuberculosis were synthesized by Tsingke Biotechnology (sequence listed in Supplementary Table 1). ssDNA templates for T7 RNA polymerase in vitro transcription to produce crRNA and all primers in this study were with PAGE purification.

Fluorescence intensity and signal ratio were described as mean ± SD. Significant differences between mutation and wild type were evaluated by t-test. P-value less than 0.05 was considered statistically significant, P-values less than 0.001, 0.01, and 0.05 were shown as ^***, ^**, and *, respectively. The bar plots show the means ± SD. Receiver Operator Characteristic (ROC) curves was generated by pROC package, and the area under the curve (AUC) was estimated for quantifying accuracy of the detection of fluoroquinolones resistance mutation in clinical samples. All statistical analyses were performed using R software, version 3.6.0.

Based on our previous epidemiological survey of fluoroquinolone resistance in M. tuberculosis from China and other tuberculosis drug resistance mutation databases (Sandgren et al., 2009; Long et al., 2012; Rosenthal et al., 2017; Dai et al., 2019), we summarized the high-frequency mutations in gyrA that were related to fluoroquinolone resistance and designated these mutations as targets of CRISPR-Cas13a crRNA screening.

The purity of the LwCas13a protein was analyzed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (PAGE) and Coomassie blue staining. Then, the crRNA-guided target-activated trans-endonuclease activity of LwCas13a was determined using crRNA containing a 28 nt spacer and target RNA by fluorescence assay (Figure 1). ssRNA transcribed from a plasmid containing a gyrA fragment (188 nt 7,462–7,649) was chosen as the model target, and crRNA with a 28 nt spacer that was completely complementary to gyrA (7,462–7,649) was used to verify the specificity of the CRISPR-Cas13a/crRNA system. Another 188 nt ssRNA transcript was used as the non-target control (sequences listed in Supplementary Table 1). As shown in Figure 1, the fluorescence intensity of the sample with the Cas13a/crRNA complex and gyrA target ssRNA increased rapidly and reached saturation within 40 min. However, the fluorescent signal barely changed without crRNA or target ssRNA. The non-specific ssRNA also could not induce fluorescence enhancement, confirming that CRISPR-Cas13a crRNA systems can only be activated by specific ssRNA.

Previous studies showed that LwCas13a target cleavage was reduced when there were two or more mismatches in the target crRNA duplex but was relatively unaffected by single mismatches (Gootenberg et al., 2017). On a target with a mutation at the third position, LwCas13a shows maximal specificity when the synthetic mismatch is in position 5 of the spacer and the spacer length is 28 nt (designated 3–5crRNA hereafter) (Gootenberg et al., 2017). Based on this approach, taking GyrA G88A as the target mutation, we also shifted the target mutation across positions 3–6 and tiled synthetic mismatches in the spacer around the mutation (Figure 2A). In our data, the target mutation at position 4 of the spacer and synthetic mismatch at position 6 (4–6 crRNA) showed the highest specificity ratio (38.22 ± 0.91) with a comparable high fluorescence intensity (9891.34 ± 2095.38) (Figures 2B,C). The 3–5 crRNA design strategy is the suboptimal strategy in G88A SNP recognition with a specificity ratio of 24.45 ± 3.09.

Based on the performance of the 4–6 and 3–5 crRNA strategies in recognizing SNPs leading to the G88A, we evaluated the 4–6 and 3–5 crRNA design strategies in recognizing SNPs leading to A90V, S91P, and D94N (H)(T)(G). The 3–5 crRNA can effectively distinguish SNPs leading to aa substitutions in A90V, with a specificity ratio of 8.29 ± 1.51 (Figures 3A,B). The function of crRNAs based on this rational design strategy in the recognition of SNPs leading to S91P or D94N (H)(T)(G) is undesirability (S91P substitution, 3–5 crRNA:1.50 ± 0.02, 4–6 crRNA:1.57 ± 0.08, D94N, 3–5 crRNA:0.96 ± 0.01, 4–6 crRNA: 1.22 ± 0.31, D94H, 3–5 crRNA:1.04 ± 0.01, 4–6 crRNA: 2.69 ± 0.3, D94T, 3–5 crRNA:1.27 ± 0.1, 4–6 crRNA:2.36 ± 0.05, D94G 3–5 crRNA:0.78 ± 0.11, 4–6 crRNA:1.01 ± 0.04). These data demonstrated that the target mutation at the 5’ region (position 3–6) combined with synthetic mismatches around the mutation does not fit all strategies in the design of crRNAs that have single-base specificity.

Previously, data showed that a target mutation at the 5′ region (position 3–6) combined with synthetic mismatches around the mutation did not give a robust ability to distinguish certain SNPs (Figures 3A,B). In further studies, the 28 nt spacer was divided into 4 regions (region A: 1–7 nt, region B: 8–14 nt, region C: 15–21 nt, and region D: 22–28 nt); target mutations were introduced into regions A, B, C and D; and synthetic mismatches were introduced into the middle of the other 3 regions (Figure 4A). The optimal target mutation was identified after the first round of screening, the optimal position combination of the target mutation and synthetic mismatch was settled, and the fine tuning of synthetic mismatches was further carried out in the second round of screening. Based on this screening strategy, we designed crRNAs that targeted SNPs leading to GyrA D94N substitution, and the target mutation was anchored in region B, while the synthetic mismatches introduced into regions A, C and D, with specificity ratios of 9.12 ± 1.37, 5.93 ± 0.63 and 2.35 ± 0.08, were obviously higher than those of other mismatch combinations (Figures 4B,C). During fine tuning, target mismatches were further set to positions 8, 10, and 12 in region B, while synthetic mismatches were introduced into positions 2, 4 and 6 in region A and positions 16, 18, and 20 in region C (Figures 4D–F). When the target mismatch was set to position 10 and the synthetic mismatch to position 16 (crRNA 10–16), the strongest specificity ratio (15.09 ± 0.64) was found, and that for the first round of screening was also higher (crRNA 10–4 9.12 ± 1.37). For SNPs leading to D94H and D94T substitutions, which were in the same position as D94N, crRNA 10–16 presented an excellent specificity ratio (D94H 36.90 ± 1.63, D94T 35.77 ± 4.78) (Figures 5A,B). Notably, the 10–16 design strategy could not obtain a crRNA with a desirable specificity ratio for discriminating the D94G aa substitution, even though the target mutation leading to D94G was just one nucleotide shift compared with D94N, H, and T (Table 1 and Figures 5A,B). Thus, the single-base specificity crRNA is highly sensitive to the target, and no general rule can be directly applied to other conditions. Based on this Target Mutation-Anchored Screening Strategy (TMAS) strategy, we further obtained crRNAs that can efficiently distinguish SNPs that lead to S91P aa and D94G aa substitutions from wild-type SNPs with signal ratios of 11.98 ± 0.21 and 9.29 ± 0.73, respectively (Supplementary Figures 1A–F, 2A–F).

Overall, based on the rational design and TMAS strategy, we screened a set of crRNAs with desirable fluoroquinolone resistance mutation-resolving ability (G88A, 4–6 crRNA, 38.22 ± 0.91; A90V, 3–5 crRNA, 8.29 ± 1.51; S91P, 10–17 crRNA, 11.98 ± 0.21; D94N, 10–16 crRNA, 15.09 ± 0.64; D94H, crRNA 10–16, 36.90 ± 1.63; D94T, 10–16 crRNA, 35.77 ± 4.78; and D94G, 17–10 crRNA, 9.29 ± 0.73).

The application ability of LwCas13a was investigated by detecting genomic DNA from different clinical MTB strains. Genomic DNA of M. tuberculosis H37Rv was selected as the control sample. Seventy-five M. tuberculosis strains, including 35 fluoroquinolone-resistant (resistant to at least ofloxacin or levofloxacin) and 40 fluoroquinolone-sensitive strains, were collected from the Urumqi Municipal Centre for Disease Control and Prevention (Supplementary Table 3). The mutations in gyrA and drug resistance information are shown in Figure 6. Genomic DNA was extracted and diluted to 5 ng/μL in a PCR amplification system, and the T7 RNA polymerase promoter sequence was appended ahead of the PCR primer. PCR products were further analyzed by fluoroquinolone resistance mutation detection integrated with the T7 RNA polymerase transcription system and LwCas13a trans-cleavage RNase activity fluorescence detection system. crRNAs designed for different fluoroquinolone resistance SNPs were added to LwCas13a systems. The limitation of the current PCR-Cas13a detection system to distinguish the mutations from wild-type temples varied from 1 × 10⁰ copies/μL (S91P) to 1 × 10² copies/μL (D94G), depending on the performance of different crRNAs (Supplementary Figure 3). The cut-off value of fluorescence to judge whether the sample contained certain mutations was set to the mean value plus 5 times the standard deviation based on M. tuberculosis H37Rv. In Figure 6, the sensitivity and specificity of the LwCas13a-based assay for the detection of fluoroquinolone phenotypic resistance were 91.43 and 100.00%, respectively. The positive prediction value in fluoroquinolone resistance mutation detection was 91.43% (95%CI, 76.94–98.20%) while the negative prediction value was 93.02% (95%CI, 80.94–98.54%) (Table 2). Three isolates that were found to be resistant by phenotypic drug susceptibility testing but found to be susceptible by the investigational assay were also found to be wild-type based on method of DNA sequencing. The sensitivity and specificity of the LwCas13a-based assay for detecting resistance mutations from DNA sequencing were 100 and 100%, respectively, in this small clinical sample cohort.