Here we hypothesized that we could enhance the sensitivity of isothermal one-pot recombinase polymerase amplification (RPA)-CRISPR assays by using gRNAs that favor Cas12a trans- vs. cis-cleavage activity, as this should favor target accumulation that drives reporter trans-cleavage, and that sensitivity could be further enhanced by the use of Cas12a RNPs targeting distinct sites on their target amplicon. We therefore developed a one-pot asymmetric cis/trans CRISPR cleavage assay for TB (ActCRISPR-TB) that employed multiple gRNAs favoring trans- vs. cis-cleavage activity upon recognition of their targets within an Mtb-specific insertion sequence. In an analysis of 603 clinical specimens from 479 individuals, ActCRISPR-TB diagnostic performance was comparable to reference molecular diagnostics when analyzing respiratory samples, but was greater than these tests when applied to non-respiratory specimens, and markedly enhanced Mtb DNA detection in tongue swabs. No loss in analytical or diagnostic sensitivity was observed when this assay was adapted to a lateral flow assay (LFA) format that analyzed self-collected tongue swab specimens suitable for use in streamlined active TB disease finding and screening efforts.
We have previously described a two-step CRISPR-based TB assay that detects cell-free Mtb DNA in serum/plasma of TB patients using a gRNA (gRNA-0, Table S1) that recognizes a site in the multi-copy Mtb complex-specific IS6110 insertion element with a canonical PAM (TTTV). However, an integrated one-pot assay derived from this assay exhibited delayed and attenuated signal, likely due to a competitive cis-cleavage of the RPA amplicon by its Cas12a RNP (Fig. S1). Since gRNAs targeting sequences that lack canonical PAM sites (non-canonical gRNAs) can yield differential cis-cleavage vs. trans-cleavage activity, we tiled the IS6110 amplicon sequence to identify non-canonical gRNA candidates that had stable secondary structures (Fig. 1a and S2-S3; Table S1). All gRNAs had strong target specificity but variable cis- and trans-cleavage activities as determined by monitoring the activity of the RNPs to deplete their dsDNA substrate and ssDNA reporter (Fig. 1b-d, S4-S6; Table S2-S3). All non-canonical gRNAs produced weaker cis-cleavage activity than gRNA-0, although trans-cleavage activities detected with gRNA-2, -5, and -0 were comparable, and substantially greater than the those of the other five gRNAs (12-42% of gRNA-0 activity) (Fig. 1d).
These results agreed with reports that the cis- and trans-cleavage activities of Cas12a are functionally independent, while reanalysis of publicly available datasets from a recent study identified another non-canonical gRNA yielding asymmetric Cas12a cleavage activity (Fig. S7). Structural modeling of the Cas12a RNPs employed in our study detected similar Cas12a and gRNA-template heteroduplex structures but divergence in the relative positions of the non-template strands in proximity to Cas12a catalytic site (Fig. S8), which could potentially influence their asymmetric cleavage activity.
Normalized activity comparisons (Table S4) revealed that gRNA-4, gRNA-5, and gRNA-6 differentially promoted trans- vs. cis-cleavage, and one-pot assays performed with these three gRNAs revealed better reaction kinetics than the four gRNAs that had more balanced cis- and trans-cleavage activities (Fig. 1e), suggesting asymmetric activity reduced amplicon cleavage to improve amplification efficiency, Cas12a trans-cleavage kinetics, and assay signal (Fig. 1c). However, assay signal generated by gRNA-6 was template independent and apparently derived from a 10-nucleotide sequence overlap with the RPA forward primer (Fig. 1a, S2 and S10). This gRNA was thus excluded from all subsequent analyses.
Since gRNA-5 revealed the best reaction kinetics, the RPA and CRISPR reaction conditions of the gRNA-5-based one-pot ActCRISPR-TB assay were optimized to maximize signal (Fig. S11) and minimize the half-maximum signal time (Fig. S12). Subsequent analyses were performed using these conditions (500 nM primers, 16.8 nM Mg, and 40 nM RNP), as alternate parameters yielded less favorable kinetics. Primer increases tended to reduce RPA efficiency, Mg increases reduced assay signal, and reporter increases had no prominent effect on assay kinetics so that a 600 nM concentration was selected to minimize background and assay expense. Comparable ActCRISPR-TB end results were obtained across a 36-40 °C temperature range and with RPA reagents from different manufacturers.
We hypothesized that assays employing distinct Cas12a RNPs could increase signal (Fig. 1f), as previously reported for an amplification-free CRISPR system, and found that adding other gRNAs to an optimized ActCRISPR-TB gRNA-5 assay with a constant total gRNA concentration differentially increased (gRNA-2 > gRNA-3 > gRNA-4) or decreased (gRNA-0 >> gRNA-1) its kinetics (Fig. 1g). Signal increases induced by gRNA additions were influenced by the relative gRNA ratios: slightly shifting the gRNA-5 to gRNA-2 ratio to favor gRNA-5 (30:10) modestly enhanced signal, whereas a corresponding decrease (10:30) markedly attenuated assay kinetics and signal (Fig. S13), likely due to the greater cis-cleavage activity associated with gRNA-2. Adding gRNA-3 or gRNA-4 to a gRNA-2/gRNA-5 assay modestly increased its kinetics, while adding gRNA-1 or gRNA-0 had the opposite effect (Fig. 1h and S14). Multi-gRNA ActCRISPR-TB assays that used gRNA-2, -3, and -5 achieved a limit of detection (LoD) of 5 copies/μL (Fig. 1i, j) -- 20 times lower than a one-pot assay using canonical gRNA-0 (Fig. S15) -- while retaining specificity for Mtb complex species (Fig. 1k).
This ActCRISPR-TB assay was next systematically evaluated for its diagnostic performance when used to analyze different types of cryopreserved specimens (Figs. 2a and 3a). Sputum samples with high to very low Xpert grades, as assigned by their Ct values, were DNA-extracted and analyzed to determine the minimum time required for sensitive detection of positive samples (Fig. 3b). Most Xpert-positive sputum DNA isolates (85%; 17 of 20) revealed at least weak ActCRISPR-TB signal by 15 min, with false-negative results detected only in samples with low or very low Xpert grades. Maximum diagnostic sensitivity (95%; 19 of 20 samples) was achieved by 45 min, and the remaining false-negatives did not differ from Xpert-negative specimens at 60-min. A 45-min read time was thus selected to optimize assay time and performance. A subsequent ActCRISPR-TB validation study performed with 56 Xpert-positive and 56 Xpert-negative sputum specimens from adults with presumed TB (Table S5) yielded comparable performance, misclassifying only four samples with low or very low Xpert grades (Fig. 3c and S16) to achieve 93% (95% CI: 83-98%) sensitivity and 100% (95% CI: 94-100%) specificity.
We next evaluated ActCRISPR-TB performance using sputum samples obtained from a cohort of 28 HIV-positive adults with low CD4 counts (median 72 cells/mL; IQR 13.5-158.5) and with and without evidence of TB, since it is frequently more difficult to diagnose TB in this patient population due to the paucibacillary nature of their sputum specimens. These individuals were post-hoc classified as confirmed, unconfirmed, unlikely, or non-TB cases based on their clinical, laboratory, and treatment data (Table S6). ActCRISPR-TB results demonstrated 80% (95% CI: 44-97%) sensitivity for the confirmed and unconfirmed TB cases, and 89% (95% CI: 65-99%) specificity with unlikely and non-TB cases (Fig. 3d), whereas Xpert MTB/RIF results had 40% sensitivity overall as it detected none of the unconfirmed TB cases. Notably, these results may underestimate ActCRISPR-TB specificity, since the two unlikely TB cases with false-positive results were both culture positive but judged to have non-tuberculous mycobacterial (NTM) infections based on additional findings. However, given the species specificity of the assay, the observation that the remaining four unlikely TB cases with NTM diagnoses all had negative ActCRISPR-TB negative results, and the elevated TB risk in this severely immunocompromised patient cohort, we cannot exclude the possibility that these two unlikely TB patients had both NTM infections and TB.
ActCRISPR-TB performance was next assessed using non-sputum samples, since 25% of symptomatic and >90% of asymptomatic TB cases cannot spontaneously produce sputum. Lower respiratory tract specimens, including bronchoalveolar lavage fluid (BALF), are often collected to improve diagnosis of such patients. BALF specimens analyzed in this study were obtained from 47 bacteriologically confirmed TB cases, 19 clinically diagnosed TB cases, and 19 non-TB cases (Fig. 2a; Table S7). Most non-TB cases (66%; 56 of 85) had clinical features of TB but were not diagnosed with TB due to negative sputum test outcomes or missing sputum specimens (Fig. S16). BALF ActCRISPR-TB results identified 96% (45 of 47) of the bacteriologically confirmed TB cases, 21% (4 of 19) of the clinically diagnosed TB cases, and all (19 of 19) non-TB cases (Fig. 3e). Notably, there was a strong correlation between Xpert grade and ActCRISPR-TB signal and positivity for both sputum and BALF specimens (Fig. S17). BALF ActCRISPR-TB results yielded false-negatives for two individuals with "very low" positive Xpert sputum results and identified eight TB cases with false-negative BALF Xpert results (four with microbiologic confirmation and four with clinical diagnosed TB). Overall, ActCRISPR-TB showed high concordance (93%; 95% CI, 89-96%) with Xpert in respiratory specimens, with a pooled sensitivity of 94% (95% CI, 88-98%) and specificity of 92% (95% CI, 85-96%) (Fig. 3f).
ActCRISPR-TB performance was next assessed using stool specimens from a cohort of children with suspected TB, since children are often unable to expectorate sputum and invasive sampling is often not feasible, leading the WHO to recommend testing stool specimens. We analyzed archived stool samples from 47 children with presumed TB (Table S8). Only 15 of these children had sputum samples, and just five had positive Xpert Ultra sputum results. However, Xpert Ultra detected Mtb DNA in 23 gastric aspirate and 24 stool specimens, whereas ActCRISPR-TB detected 26 positive stool samples, yielding 83% (95% CI: 65-94%) sensitivity and 94% (95% CI: 71-100%) specificity (Figs. 3g and S18).
Finally, ActCRISPR-TB was evaluated for its ability to diagnose extrapulmonary tuberculosis (EPTB) since Mtb dissemination increases the risk of poor outcomes and death, but direct EPTB diagnosis requires the analysis of paucibacillary samples obtained from suspected infection sites. Since tuberculous TB meningitis (TBM) cases require prompt diagnosis and treatment to avoid poor ourtcomes, we analyzed cerebrospinal fluid (CSF) specimens from a small cohort of adults with suspected TBM (Fig. 2a; Table S9). ActCRISPR-TB detected 93% (14 of 15) of microbiologically confirmed and 64% (7 of 11) of clinically diagnosed TBM cases (Fig. 3h), yielding 81% overall sensitivity vs. the 35% (9 of 26) overall sensitivity of Xpert. Specificity was not estimated with the non-TBM group since several individuals had pulmonary TB diagnoses and missing CSF clinical test results, which could have yielded false-negative TBM diagnoses. Notably, three of the four non-TBM patients with strong ActCRISPR-TB false-positives had pulmonary TB, and their positives could have reflected Mtb-DNA dissemination into the CSF from the circulation (valid false-positives) or early-stage of TBM undetected by other methods (missed true-positives).
New TB diagnostics that rapidly and accurately analyze accessible specimen types are required to improve access to TB testing and achieve the End TB program's goal. Tongue swabs can simplify specimen collection for pulmonary TB diagnosis since Mtb bacilli expelled from the lungs accumulate on the tongue to provide direct evidence of TB disease. However, sensitive and streamlined assays that can detect trace Mtb DNA concentrations in these specimens are required for large-scale active TB screening efforts. We therefore compared ActCRISPR-TB and GeneXpert MTB/RIF Ultra (Xpert Ultra) performance with 205 tongue swab specimens obtained from 134 individuals with microbiologically confirmed TB or non-TB diagnoses, with further analyses performed to examine the effect of swab type and storage conditions (Figs. 2b and 4a, and Table S10). Positive ActCRISPR-TB and Xpert Ultra results were detected for 60 and 45 of these TB cases [74% (95% CI: 63-83%) vs. 56% (95% CI, 44-67%) sensitivity], but not for specimens of the 53 non-TB cases (100% specificity; 95% CI, 93-100%) (Fig. 4b and S19).
ActCRISPR-TB signal intensities for tongue swabs from TB cases with negative and positive sputum culture results markedly differed [median (IQR) values of 509 (447, 1802) vs. 2119 (503, 2119)], and had positive signal frequencies of 38% and 66%, respectively (Fig. 4c). Swab ActCRISPR-TB signal also distinguished individuals with positive vs. negative sputum smear results but not individuals with differing smear positivity grades, despite a trend for increased swab positivity (33 to 96% positive) with increasing smear grade (Fig. 4d). Similarly, swab ActCRISPR-TB signal distinguished TB cases with positive vs. negative Xpert results and very low vs. median and high sputum Xpert grades, with positive results tending to increase with increasing Xpert grade (from 14 to 75% positive) (Fig. 4e); and a similar trend was detected when comparing Xpert Ultra swab and sputum results (Fig. S20). Notably, ActCRISPR-TB swab results detected 33% (31/82) and 14% (3/22) of the TB cases missed by sputum smear and Xpert results.
Subsequent analysis of 52 paired tongue swabs from hospitalized TB cases did not detect ActCRISPR-TB signal differences among the tested swab types (Fig. 4f). However, swabs collected in the morning had higher ActCRISPR-TB signal and positivity rates than those collected at night [median (IQR): 1802 (503, 5128) vs. 1105 (438, 4026); 68% vs. 55% positive] (Fig. 4g), and similar trends were observed for Xpert Ultra results (Fig. S21). Swabs preserved in Tris/EDTA buffer yielded stable signal and positivity when stored at -20 °C for 1 week (Fig. 4h). Swab signal but not positivity decreased after 5 days storage at 4 °C, and signal and positivity decreased after 3 and 5 days at room temperature (25 °C). Swabs can thus provide useful results after short-term swab storage at conditions encountered in both well-equipped and resource-limited settings.
ActCRISPR-TB assays analyze NA extracts and are read by benchtop instruments, like other reported CRISPR-based TB assays (Table S11), and are thus not practical as point-of-care assays, which should minimize operator and equipment requirements. We therefore refined the ActCRISPR-TB workflow, employing a simple thermal/mechanical lysis procedure to generate sample lysates that could be directly analyzed by the assay, yielding signal that could be visualized using a LFA (Fig. 5a). This streamlined ActCRISPR-TB LFA workflow employed a 10-min thermal/mechanical lysis step to release Mtb DNA, a 45-min ActCRISPR-TB reaction, and a 2-min LFA incubation step, followed by visual or smartphone detection approaches that respectively produced qualitative or semi-quantitative results (Fig. 5b and S22-24). Sample classification results obtained with thermal/mechanical lysates were equivalent to those determined with standard NA extracts despite yielding slightly weaker signal, and significantly reduced workflow complexity, time, and cost (Fig. S25; Table S12).
Sensitivity and specificity results for the optimized ActCRISPR-TB LFA and benchtop ActCRISPR-TB assay were comparable (Fig. 5c, d and S26). Weakly and moderately positive swabs spiked with 10 and 100 target copies/μL did not exhibit signal variability detectable by visual inspection, but smartphone readouts detected 2.3% to 15.6% intra-assay and 9.9% to 14.5% inter-assay coefficients of variation (Fig. 5e). Food residue attenuated signal intensity but not positive signal identification rates in weakly positive spiked swabs samples (Fig. 5f).
ActCRISPR-TB benchtop and LFA results for self-collected tongue swabs detected 17 of 24 TB cases and 0 of 11 non-TB controls (Fig. 5g; Table S13). Notably, three weakly positive ActCRISPR-TB LFA results detected at 45 min were readily detected when read by smartphone (Fig. S27) and had strongly positive visual results when read at 90 min (Fig. S28). Xpert Ultra swab results detected 16 TB cases, including 15 with positive ActCRISPR-TB benchtop/LFA results, with the former and latter assays uniquely detecting one and two additional TB cases and missing six TB cases (Fig. 5e and S29a), despite Xpert Ultra using 40-fold more lysate. Sputum Xpert results detected more true-positives than ActCRISPR-TB LFA swab results, but missed one case identified by ActCRISPR-TB, and both returned false-negative results for two TB cases (Fig. S29b).