key: cord-0286043-a8kl9hm0
authors: Son, S.; Lyden, A.; Shu, J.; Stephens, S. I.; Fozouni, P.; Knott, G. J.; Smock, D. C. J.; Liu, T. Y.; Boehm, D.; Simoneau, C.; Kumar, R.; Doudna, J. A.; Ott, M.; Fletcher, D. A.
title: Sensitive and multiplexed RNA detection with Cas13 droplets and kinetic barcoding
date: 2021-08-05
journal: nan
DOI: 10.1101/2021.08.02.21261509
sha: 156ba78d5eb1a29abc0066a8e94b9605bb32336e
doc_id: 286043
cord_uid: a8kl9hm0

Rapid and sensitive quantification of RNA is critical for detecting infectious diseases and identifying disease biomarkers. Recent direct detection assays based on CRISPR-Cas13a avoid reverse transcription and DNA amplification required of gold-standard PCR assays, but these assays have not yet achieved the sensitivity of PCR and are not easily multiplexed to detect multiple viruses or variants. Here we show that Cas13a acting on single target RNAs loaded into droplets exhibits stochastic nuclease activity that can be used to enable sensitive, rapid, and multiplexed virus quantification. Using SARS-CoV-2 RNA as the target and combinations of CRISPR RNA (crRNA) that recognize different parts of the viral genome, we demonstrate that reactions confined to small volumes can rapidly achieve PCR-level sensitivity. By tracking nuclease activity within individual droplets over time, we find that Cas13a exhibits rich kinetic behavior that depends on both the target RNA and crRNA. We demonstrate that these kinetic signatures can be harnessed to differentiate between different human coronavirus species as well as SARS-CoV-2 variants within a single droplet. The combination of high sensitivity, short reaction times, and multiplexing makes this droplet-based Cas13a assay with kinetic barcoding a promising strategy for direct RNA identification and quantification.

3 combining multiple crRNAs recognizing different regions of the target RNA. For the SARS-CoV-2 genome, LbuCas13a directly measured as low as ~200 copies/µL in 30 minutes with 3 crRNAs in a mobile phone detector 1 and ~63 copies/µL in 2 hours with 8 crRNAs in a standard plate reader 2 . However, PCR-level sensitivity has not yet been achieved with direct Cas13 detection.

In addition, approaches for identifying which of the multiple viral variants are present in a single sample are limited 10 .

Here we show that RNA detection with high sensitivity and multiplexed specificity can be achieved in short times by encapsulating the Cas13 reaction in droplets and monitoring enzyme kinetics using fluorescence. Like droplet digital PCR (ddPCR) 11 , our assay enables quantification of the absolute amount of target RNA based on the number of positive droplets. Unlike commercially-available ddPCR, the small droplet volume in our assay accelerates signal accumulation of the direct Cas13 reaction. When a single target RNA is encapsulated in a ~10 pL droplet, the Cas13 signal accumulation rate is equivalent to that of a bulk reaction containing 10 5 copies/µL target (Fig. 1A) . To rapidly generate ~10 pL droplets, we emulsified the reaction mix containing LbuCas13a for 2 min in an excess volume of an oil/surfactant/detergent mixture using an automatic multi-channel pipettor (Fig. 1B, Supplementary Fig. 1A and 1B, Methods). We imaged the resulting emulsion on a fluorescence microscope, which revealed the formation of millions of droplets, with diameters ranging from 10 to 40 µm (Fig. 1C, Supplementary Fig. 1A ).

Imaging the droplets allowed us to normalize the fluorescence signal by droplet size 12 and avoid the need for slower and more complex systems to generate uniform droplet sizes.

We validated the Cas13 droplet assay by first forming droplets containing 10,000 copies/µL of SARS-CoV-2 RNA, LbuCas13a, crRNA targeting the SARS-CoV-2 N gene (crRNA 4) and a fluorophore-quencher pair tethered by pentauridine (5U) RNA (reporter) and then monitoring the reaction in droplets over time (Fig. 1D ). At this target concentration, ~7% of the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 4 droplets should contain the target RNA, with the vast majority of those containing only a single copy. As expected, we found that the signal accumulation rate in positive droplets was inversely proportional to droplet size ( Supplementary Fig. 1C) , with smaller droplets increasing faster than larger droplets (9-fold increase for a 23 µm droplet vs. 3-fold increase for a 42 µm droplet) (Fig.   1E ). Our measurements show that a single LbuCas13a can cleave 471 ± 47 copies of reporter every second in the presence of 400nM reporter, or a Kcat/KM of 1.2 x 10 9 M -1 s -1 , which is two orders-of-magnitude higher than that measured for LbCas12a 13 and consistent with that measured for LbuCas13a based on a bulk assay 14 . Notably, the absolute trans-cleavage rate of a single LbuCas13a remains consistent regardless of droplet size (Fig. 1F, Supplementary Fig.   1D ). Longer incubation times resulted in a linear increase in average signal per droplet (Fig. 1G ), but all the positive reactions could be correctly identified as early as 5 minutes with a 20X/0.95NA objective (Fig. 1H ) and 15 minutes with a 4X/0.20NA objective (Supplementary Fig. 1E and 1F ), while the "no target" condition rarely showed positive droplets (Fig. 2) .

We next tested whether crRNA combinations, which can generate more signal per target RNA 1 , could further increase the speed of the Cas13a droplet assay. We loaded in vitro transcribed (IVT) target RNA corresponding to the N gene of SARS-CoV-2 to droplets containing either crRNA 2, crRNA 4, or both crRNAs, which target different regions of the N gene ( Fig. 2A) , and then quantified the number and signal of the positive droplets. Surprisingly, while crRNAs 2 and 4 generate similar signals when used individually (Supplementary Fig. 2A ) and might be expected to double the signal when combined, we found that the signal per droplet was not significantly different in any of the positive droplets (Fig. 2B) . Rather, the number of positive droplets almost doubled when droplets contain both crRNA 2 and crRNA 4 compared to just one of the crRNAs (Fig. 2C ). This suggests that the N gene IVT was fragmented through cis-cleavage upon initiation of the reaction prior to droplet formation, causing the regions targeted by different . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 5 crRNAs to be loaded into separate droplets (Fig. 1B, Fig. 2A ). The resulting fragmented loading we observe is consistent with a recent study of microwell-based Cas13a reaction 3 and suggests that multiple crRNAs activate independent Cas13a reactions whether in bulk solution or in droplets.

We then tested whether increasing crRNA combinations would further increase the speed of the Cas13a droplet assay. We identified 26 crRNAs targeting various regions of the SARS-CoV-2 genome that individually produce a strong signal in Cas13 detection reactions (Supplementary Table 1 ). Adding more crRNAs while keeping the total RNP concentration constant led to an increasing number of positive droplets (Fig. 2D ). This was possible because the activity of Cas13a remains constant even when only a small fraction (1/50 or less) of total RNPs in a droplet contains crRNAs matching the target ( Supplementary Fig. 2B ). Overall, this suggests that a large number of crRNAs can be combined to maximize the number of independent Cas13a reactions from the same target RNA. Given that the sensitivity of droplet-based assays is fundamentally limited by the false-positive rate observed in the absence of target RNA, the generation of multiple positive droplets per target RNA has the potential to increase the signal-tobackground, and thus, sensitivity of the assay.

To test the sensitivity of our Cas13a droplet assay with crRNA combinations, we carried out serial dilutions of precisely quantified SARS-CoV-2 genomic RNA obtained from the Biodefense and Emerging Infections Research Resources Repository (BEI Resources). We quantified the number of positive droplets in each dilution using either a single crRNA or all 26 crRNAs, using 36 images per condition (~160,000 droplets) after 15 minutes of reaction incubation (Fig. 2E) . For the single crRNA (crRNA 4), the number of positive droplets was significantly higher than the no-target control for the samples containing 20 target copies/µL or more. For the combination of 26 crRNAs, the limit of direct detection was 1 copy/µL target, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 6 comparable to the sensitivity of PCR. This limit of detection was not improved if we incubated the reaction for 30 minutes instead of 15 minutes ( Supplementary Fig. 2C ).

The fast Cas13a kinetics we observe in droplets critically depends on the choice of the crRNA and its target. For example, two other crRNAs targeting the SARS-CoV-2 N gene-crRNA 11 and crRNA 12-exhibit significantly slower rates in bulk reactions than crRNA 2 or 4 ( Fig. 3A,   Supplementary Fig. 2A ). For this reason, the selection of "good" crRNAs that support efficient Cas13 activity is critical for bulk Cas13-based molecular diagnostics 15 , though how different crRNAs affect the activity of Cas13 is not well understood 16 . We therefore tested crRNAs 11 and 12 in our droplet assay and compared it to the activity of crRNA 4. As expected, the number of positive droplets was reduced for crRNAs 11 and 12, but the difference to crRNA 4 was less than observed in bulk reactions (Fig. 3B ). On the other hand, the signal in each positive droplet was significantly reduced for crRNA 11 and 12 compared to crRNA 4 (Fig. 3C ). To better understand how different crRNAs affect Cas13a activity, we examined the individual reaction trajectories within positive droplets, which report the change in fluorescence between each 30 second measurement time point (Fig. 3D-F) . For all three crRNAs, most positive droplets exhibited slopes significantly higher than those recorded in RNP-only droplets lacking the target RNA (Fig. 3G ).

However, the three crRNAs induced distinct kinetic behaviors in droplets, with the slope, shape, and x-intercept of the individual trajectories varying widely depending on the individual crRNA.

This ultimately resulted in different endpoints for each crRNA. In some cases, we found a strikingly stochastic behavior in which trajectories exhibited periods of no signal increase followed by periods of rapid signal increase ("rugged" vs smooth trajectory) ( Fig. 3H -I). Since each droplet contains, on average, a single target, these results indicate that specific crRNA/target combinations modulate Cas13a enzymatic activity at a single-molecule level.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint

To quantify these kinetic differences in droplets, we characterized individual trajectories by their average slope, root-mean-square-deviation (RMSD), and time from target addition to initiation of enzyme activity (Tinit) (Fig. 3H) . By calculating the instantaneous slopes at each point in the trajectory and fitting their distribution to a Gaussian, we found that the trajectories exhibited two different slopes, one "fast" when fluorescence was increasing and one "slow" when fluorescence was not increasing (Fig. 3I) . Interestingly, even though the average slopes differed significantly for the different crRNAs, the instantaneous "fast" slopes were constant across all three crRNAs (Fig. 3J) . Consistent with this, crRNA 12, which exhibited the lowest average slope of the three crRNAs, showed extended "slow" periods ( 3C ). In contrast, when we changed the target from genomic SARS-CoV-2 RNA to a 20nucleotide fragment complementary to crRNA 12's spacer sequence, the stochastic behavior of the reaction was no longer observed and Tinit was significantly shortened (Supplementary Fig. 4 ).

This suggests that RNA regions outside of the targeted spacer sequence affect Cas13a activity.

We hypothesized that the distinct kinetic signatures of the different crRNAs could be used to identify specific crRNA-target pairs in one droplet, thus providing a method for multiplexed detection of different RNA viruses (Fig. 4A) or variants of the same virus (Fig. 4B ). To test this idea, which we termed 'kinetic barcoding', we combined a crRNA targeting a common cold virus . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 8 HCoV-NL-63 (HCoV crRNA 6) and a second crRNA targeting SARS-CoV-2 (crRNA 12); both crRNAs were chosen because they individually exhibit different kinetic signatures on their respective targets (Fig. 4C) . We collected 30-minute trajectories from hundreds of droplets containing either HCoV-NL-63 or SARS-CoV-2 viral RNA along with Cas13a and both crRNAs together. The two groups of trajectories associated with the two different crRNAs were clearly distinguishable based on their average slope and RMSD (Fig. 4D ). To determine how clearly those two groups can be distinguished, we randomly sampled a subset of trajectories and compared their difference by performing Student's t-test on their binary classification result ( Similar experiments were performed to differentiate the original WA1 SARS-CoV-2 strain from common variants. First, we used a crRNA (crRNA 6) spanning the wild-type D614 sequence in the SARS-CoV-2 Spike (S) gene and compared the signal trajectories generated from IVT wildtype S gene with the trajectories from the S gene encoding a D614G mutation now shared by most SARS-CoV-2 variants since June 2020 18 (Fig. 4F ). Both crRNAs yielded predominantly "smooth" trajectories (as indicated by the low RMSD), but the average slopes obtained with the mutant target were significantly lower than those obtained with the matching wild-type target (Fig.   4G ). Based on this difference in slopes, we distinguished the D614G mutant from the wild-type target within 5 minutes using a minimum of 30 trajectories ( Supplementary Fig. 5C ).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint Second, we focused on a specific California variant of concern (B.1.427/B.1.429; Epsilon), which among others harbors a unique S13I mutation in the S protein; this variant exhibits ~20% increased transmissibility and a 4-to 6.7-fold and 2-fold reduced neutralization by convalescent and post-vaccination sera, respectively 19 . We used a crRNA matching the mutant sequence (crRNA 45) and measured RNA extracted from cultured wild-type or mutant viruses as well as nasal/oral swab samples from people known to be infected with wild-type virus or the variant based on sequencing (Supplementary Table 2 We found that, regardless of the choice of trajectories, the average slope distribution was clearly distinguishable between wild-type virus and the B.1.427 variant, providing a detection accuracy of >98% for both cases (Fig. 4J ).

In summary, we demonstrate that a droplet-based Cas13a direct detection assay can achieve PCR-level sensitivity and distinguish RNA targets based on reaction kinetics. As one crRNA can be diluted at least 50 times or more without compromising its performance ( Supplementary Fig. 2B ), we speculate that combining multiple different crRNAs can be used to further enhance detection sensitivity and bring the limit of detection below 1 copy/µL. At this sensitivity, a droplet-based Cas13a direct detection assay would be used to measure very low viral loads such as at the beginning or end of an infection, in environmental samples, or with latent viruses such as HIV, without the need for additional reverse transcription or amplification steps.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint

We found that LbuCas13a is an efficient, diffusion-limited enzyme, whose kinetics are controlled by the specific combination of the crRNA and the target 20,21 . Interestingly, the distribution of Cas13a RNP trajectories is largely homogenous for "good" crRNAs supporting high activities (Fig. 1F) , implying that the active conformation of Cas13a RNP is stable over time.

However, certain "bad" crRNAs were found to induce "rugged" trajectories and to switch off Cas13a activity for more than a minute. The observation that this kinetic feature is abolished when a short RNA fragment is present instead of the viral RNA ( Supplementary Fig. 4 ) suggests a role of local or global RNA structure or its chemical modifications in the kinetics of an individual Cas13a RNPfeatures that are lost during target amplificationrather than an effect of enzyme conformational switching 22 . On the other hand, we show that sequence mismatches between the crRNA and its target can reduce the slope of a reaction without introducing "rugged" stochastic activity switching ( Fig. 4G and H) , pointing to multifactorial mechanisms governing Cas13a kinetics. Taking advantage of these characteristic kinetic signatures, we are able to conclusively determine which virus or variant was present in a given droplet, demonstrating the diagnostic utility of kinetic barcoding and establishing kinetic variety as an additional feature to be considered during crRNA selection.

Digital assays enhance the sensitivity and quantitative performance in ddPCR 23,24 , protein detection 25 , and recently CRISPR-Cas-based nucleic acid detection approaches 3, 4, 26, 27 . However, commercially-available droplet technologies tailored for ddPCR assays were not suitable for this assay, as amplification-free Cas13a assays require smaller droplets (~10pL) than those typically used in ddPCR (~900pL 28 ) to achieve the desired signal accumulation. On the other hand, microwell-based assays often encapsulate reactions in femtoliter-scale volumes 3 ), reducing the sample volume and limiting the detection sensitivity. We show that for a compartmentalized assay with average Cas13a kinetics, a ~10pL reaction volume provides optimal speed and sensitivity.

Our current workflow based on bulk droplet formation followed by direct imaging was chosen to . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 11 maximize throughput at the same time as accounting for droplet heterogeneity, an approach that could be scaled up to process samples in parallel. As a laboratory assay, Cas13a reactions in droplets have potential advantages over PCR for RNA quantification by avoiding errors associated with reverse transcription and amplification, though large-scale testing of diverse samples would be needed to properly compare the two approaches. In conclusion, droplet-based Cas13a direct detection assays with kinetic barcoding represents a method for using fundamental properties of CRISPR-Cas13 reactions at a single-molecule level to enable rapid and sensitive RNA quantification that could be extended to multiple RNA viruses and RNA biomarkers in the future.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 15 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint and cr4tg depict the RNA target for the crRNA2 and crRNA4 respectively. If the whole N gene is loaded to a droplet containing both crRNAs, signal will accumulate twice as fast than the droplet containing only one target sequence (complete loading); if N gene is fragmented into two halves and loaded into two separate droplets, the number of positive droplets will be . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. were determined based on a two-tailed Student's t-test and are 5.0e-4, 2.2e-3, 0.12, 0.96, respectively, for results obtained with a single guide and 1.1e-6, 7.8e-4, 2.8e-5, 7.6e-3, respectively, for results obtained with a combination of 26 guides. ns = not significant, *p < 0.05, ** p < 0.005, ***p < 0.001.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 20 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint Table 1 ) and no target RNA. 31 individual trajectories from two replicate runs were measured in droplets ranging from 30 to 36 µm in diameter.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 23 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint

Protein purification was performed as previously described 1 

Patient samples were obtained from a UC Berkeley campus testing program, as previously described 2 . Briefly, human mid-turbinate nasal/oropharyngeal swabs were collected, processed, and tested by running qRT-PCR on extracted RNA using primers for SARS-CoV-2. Ct values were obtained using primers for the N gene, S gene, and Orf1ab of the SARS-CoV-2. When the Ct value for two out of the three genes was 37 or below, the sample was considered positive. Ct values for the samples used in this study can be found in Supplementary Table 2 . Sequence data for each sample was used to identify SARS-CoV-2 variants. The samples used in this study were non-identifiable biospecimens, for which there was no link to identifiable subject information. Only secondary data analysis was carried out on the samples, and the research data will not be held for inspection by nor submitted to the FDA.

All crRNAs targeting SARS-CoV-2, SARS-CoV-2 variants (D614G, S13I), or HCoV-NL-63 were ordered as desalted oligonucleotides from Synthego at a scale of 5 nmole (Supplementary Table   1 ). Select crRNAs have also been previously described 1, 3 . is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 28 RNA. For reactions using more than one crRNA, multiple crRNAs were combined at an equal concentration and subsequently the total crRNA mix was assembled with Cas13a at 133nM equimolar concentration. 25nM of RNP complex is used unless specified otherwise. The reaction mix is measured either in bulk or in droplet following emulsification (see droplet formation); For the bulk Cas13a assay, the reaction mix was loaded into a 0. imaging or incubated in a heating block at 37˚C before being transferred and imaged. In cases where imaging could not be performed immediately after the reaction, the emulsion was quenched on ice until imaging. In both cases, the emulsion was quickly separated by spinning in a speedcontrolled mini-centrifuge (~50 rpm) for 10 seconds, the oil was completely removed from the bottom of the tube, and the emulsion was transferred into a custom flow cell after several cycles of gentle manual mixing.

The sample flow cell was prepared by sandwiching double-sided tape (~20µm thick, 3M Cat# 9457) between an acrylic slide (75mm x 25mm x 2mm, laser cut from a 2mm-thick acrylic plate)

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 29 and a siliconized coverslip (22mm x 22mm x 0.22mm, Hampton research Cat# 500829). Both surfaces are hydrophobic, promoting thin layers of oil between droplets and the surface.

Siliconized coverslips were rinsed with isopropanol to remove any auto-fluorescent debris (20 minutes sonication) and spin dried prior to assembly. 15 µL of sample emulsion is loaded into the flow cell by capillary action, after which the inlet and outlet are sealed with Valap (1:1:1 vasoline:lanoline:paraffin).

Droplet imaging is carried out on an inverted Nikon Eclipse Ti microscope (Nikon Instruments) equipped with a Yokogawa CSU-X spinning disk. 

We used MATLAB (Mathworks R2020b) to detect positive droplets and quantify fluorescence signals from microscopy images. First, the grayscale images were converted to binary images based on a locally adaptive threshold. The threshold is defined generously to select all the positive droplets and potentially some negative droplets or debris at this stage. Second, connected droplets were separated by watershed transform. Third, individual droplets are identified by . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint looking for circular continuous regions and droplet parameters such as radius, circularity, and fluorescence signal are quantified. Fluorescence signal is quantified in two different ways: the mean fluorescence signal of a droplet reflecting the density of cleaved reporter; the total fluorescence signal reflecting the total amount of cleaved reporter within a droplet. Lastly, positive droplets are chosen based on their circularity and total fluorescence signal by applying a threshold that was consistently used throughout the experiments.

To quantify signal accumulation in the same droplet over time, we associated droplets over time 

We analyzed the Cas13a reaction with a single crRNA (Fig. 1F) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint 31 /[ 0 ] is turnover frequency, or the reciprocal of the mean waiting time <1/t> in the singlemolecule Michaelis-Menten framework 5 , and it can be obtained from Supplementary Fig. 1C and D after converting the fluorescence signal to molar concentration of cleaved reporter based on a calibration.

We processed the raw signal in a series of steps prior to analysis: First, we corrected for the global signal fluctuation, which arise from a slight drift in z-focus even with the Perfect Focus System; we characterized the global signal from the background droplets as the mean of their pixel values and divided it from the positive droplet signal in each image frame. Second, we corrected for the photobleaching; we first characterized the signal decay rate from >200 Cas13a curves exhibiting negative slopes and positive initial signals. We modeled photobleaching as a linear function of initial signal based on the observed linear relationship between the decay rate versus initial signal (R 2 = 0.87). Using the model, we corrected for photobleaching in each trajectory point-by-point. Third, we filtered the trajectories with a weak Savitzky-Golay filter (order 5, frame length 9) to remove the high frequency measurement noise while preserving overall structure of the curve. Lastly, we calculated instantaneous slopes by dividing signal change between frames by the frame interval and removed single outliers exhibiting high positive or negative slopes. To characterize key parameters of Cas13a kinetics, we analyzed individual trajectories in two different domains: First, we determined the slope, Tinit, and RMSD from signal time trajectories by linear regression. Since Tinit indicates time since the droplet reaction was started, we added a constant time (12.5 minutes) that took from Cas13 droplet formation until the beginning of time course imaging. Second, we determined the slopefast, slopeslow, and a fraction spent in each period by fitting a gaussian pdf to the instantaneous slope distribution. We compared the model qualities between the single versus binary gaussian pdfs using Akaike's . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint Information Criterion (AIC) to determine whether a trajectory exhibits two different periods of slope of not.

We used the slope and RMSD of individual signal trajectories to compare Cas13a reactions between different target-crRNAs. We first performed binary classification of trajectories based on the Supported Vector Machine (SVM) in MATLAB. For this, we collected 200 to 400 signal trajectories in each condition, in two or more independent experiments per condition to prevent bias. We converted the trajectories into a 2D array consisting of the slope and RMSD, and divided the array into a training and a validation set. We then trained an algorithm using the training set with the known answers (i.e. target-crRNA condition) and classified the validation set. The accuracy of identifying individual trajectories were 75% for HCoV-NL-63 vs SARS-CoV-2 and 73% for WT vs D614G. To access significance between two groups of trajectories, we employed a twotailed Student's t-test to the predicted class and reported p-values.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint

Amplification-free detection of SARS-CoV-2 with CRISPR-Cas13a and mobile phone microscopy

Accelerated RNA detection using tandem CRISPR nucleases

Amplification-free RNA detection with CRISPR-Cas13

An Ultralocalized Cas13a Assay Enables Universal and Nucleic Acid Amplification-Free Single-Molecule RNA Diagnostics

Real time quantitative PCR

Analytical sensitivity and efficiency comparisons of SARS-CoV-2 RT-qPCR primer-probe sets

Two distinct RNase activities of CRISPR-C2c2 enable guide-RNA processing and RNA detection

METHOD REFERENCES

Amplification-free detection of SARS-CoV-2 with CRISPR-Cas13a and mobile phone microscopy

Blueprint for a pop-up SARS-CoV-2 testing lab

Accelerated RNA detection using tandem CRISPR nucleases

High-Resolution Structure of Cas13b and Biochemical Characterization of RNA Targeting and Cleavage

Fluctuating Enzymes: Lessons from Single-Molecule Studies

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

We thank all members of the Fletcher, Ott, and Doudna laboratories for helpful discussions and feedback on this project. We also thank Stacia Wyman and Bryan Bach for providing samples from the Innovative Genomics Institute, and Chaz Langelier for providing samples from the Chan-

The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint

The data that support the findings of this study are available from the corresponding author upon reasonable request.. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint Zuckerberg Biohub. Purified LbuCas13a was a kind gift from Shanghai ChemPartner, and we thank Synthego for support with synthetic crRNAs. We gratefully acknowledge support from . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted August 5, 2021. ; https://doi.org/10.1101/2021.08.02.21261509 doi: medRxiv preprint