Massively multiplexed nucleic acid detection with Cas13

Abstract

The great majority of globally circulating pathogens go undetected, undermining patient care and hindering outbreak preparedness and response. To enable routine surveillance and comprehensive diagnostic applications, there is a need for detection technologies that can scale to test many samples^1-3 while simultaneously testing for many pathogens^4-6. Here, we develop Combinatorial Arrayed Reactions for Multiplexed Evaluation of Nucleic acids (CARMEN), a platform for scalable, multiplexed pathogen detection. In the CARMEN platform, nanolitre droplets containing CRISPR-based nucleic acid detection reagents⁷ self-organize in a microwell array⁸ to pair with droplets of amplified samples, testing each sample against each CRISPR RNA (crRNA) in replicate. The combination of CARMEN and Cas13 detection (CARMEN-Cas13) enables robust testing of more than 4,500 crRNA-target pairs on a single array. Using CARMEN-Cas13, we developed a multiplexed assay that simultaneously differentiates all 169 human-associated viruses with at least 10 published genome sequences and rapidly incorporated an additional crRNA to detect the causative agent of the 2020 COVID-19 pandemic. CARMEN-Cas13 further enables comprehensive subtyping of influenza A strains and multiplexed identification of dozens of HIV drug-resistance mutations. The intrinsic multiplexing and throughput capabilities of CARMEN make it practical to scale, as miniaturization decreases reagent cost per test by more than 300-fold. Scalable, highly multiplexed CRISPR-based nucleic acid detection shifts diagnostic and surveillance efforts from targeted testing of high-priority samples to comprehensive testing of large sample sets, greatly benefiting patients and public health^9-11.

Fig. 1. CARMEN–Cas13 achieves attomolar sensitivity.

a, Identification of multiple circulating pathogens in human and animal populations represents a large-scale detection problem. b, Schematic of CARMEN–Cas13 workflow. c, Zika cDNA is detected by a single CARMEN–Cas13 assay with attomolar sensitivity and tens of replicate droplet pairs (black dots, numbers of replicates are in blue); red lines show medians and are used to construct the heat map below. Top, representative droplet images. AU, arbitrary units.

Fig. 2. Comprehensive identification of HVs with CARMEN–Cas13.

a, The development and testing of a panel for all 169 HVs with at least 10 available genome sequences. b, Experimental design using pooled PCR amplification. c, Testing a comprehensive HV panel with synthetic targets using CARMEN–Cas13. PCR primer pools 1–15 and viral families are below and to the left of the heat map, respectively. Grey lines, crRNAs not tested. d, Multiplexed coronavirus panel, comprising human coronaviruses 229E, NL63 and HKU1 Middle East respiratory syndrome (MERS) coronavirus; severe acute respiratory syndrome (SARS) coronavirus; novel coronavirus SARS-CoV-2 (nCoV); and negative control (−). e, Testing the HV panel with patient samples (additional data in Extended Data Fig. 8a). Heat maps indicate background-subtracted fluorescence after 1 h (c) or 30 min (d), or fold change over background (e). NI, not interpretable. f, Concordance of CARMEN and NGS in patient sample testing. Each box displays the number of tests and the percentage of the total. g, Identification by NGS or CARMEN of any viral sequence from the known infections in patient samples (for example, detection of HIV in samples from patients with HIV). h, Identification by NGS or CARMEN of the crRNA target for each known infection. i, Positive test results in patient samples for viruses other than the known infections. DENV, dengue virus; ZIKV, Zika virus; HCV, hepatitis C virus; TLMV, Torque teno-like mini virus; Pegi A, pegivirus A; HPV4, human papillomavirus 4; KIPyV, KI polyomavirus; MCV, Merkel cell polyomavirus; SINV, Sindbis virus; γHPV, gamma human papillomavirus; βHPV, beta human papillomavirus 2; AROAV, aroa virus. CARMEN does not test for γHPV or AROAV because fewer than 10 γHPV or AROAV genomes had been published before 24 October 2018.

Fig. 3. Influenza subtype discrimination with CARMEN–Cas13.

a, Schematic of influenza A subtype discrimination using CARMEN–Cas13. b, Discrimination of H1–H16 using CARMEN–Cas13. c, Discrimination of N1–N9 using CARMEN–Cas13. d, Identification of H and N subtypes from patient samples and mixed synthetic targets (Methods, ‘Cas13 detection reactions’ under ‘General procedures’). Heat maps indicate background-subtracted fluorescence after 1 h (for H1–H16 discrimination) or 3 h (for N1–N9 discrimination) of Cas13 detection. In b–d, synthetic targets were used at 10⁴ copies per μl. Asterisks in d indicate a signal above threshold for a synthetic target or patient sample.

Fig. 4. Multiplexed DRM identification with CARMEN–Cas13.

a, Schematic for identification of DRMs in HIV reverse transcriptase using CARMEN–Cas13. b, Identification of six reverse transcriptase (RT) mutations using CARMEN–Cas13. c, d, Concordance between CARMEN–Cas13 and Sanger sequencing (c) and CARMEN–Cas13 and NGS (d), for identification of DRMs in patient plasma or serum samples for sequences with three or fewer mismatches relative to our design. NC, negative control. e, Identification of 21 integrase DRMs using CARMEN–Cas13. Heat maps indicate SNP indexes after 0.5–3 h of Cas13 detection, normalized by row. In b, e, synthetic targets were used at 10⁴ copies per μl. Asterisks indicate the synthetic target with the mutation.

Extended Data Fig. 1. The CARMEN workflow at the molecular and macroscopic scale.

a, Detailed molecular schematic of nucleic acid detection in CARMEN–Cas13. After amplification (with optional reverse transcription), detection is performed with Cas13, using in vitro transcription to convert amplified DNA into RNA. The resulting RNA is detected with exquisite sequence specificity by Cas13–crRNA complexes, and collateral cleavage activity of Cas13 produces a signal using a cleavage reporter RNA. b, Overview of the CARMEN workflow. Amplified samples and detection mixes are colour coded, emulsified and pooled into one tube. In a single pipetting step, the pool of droplets is loaded onto a chip, where the droplets self-organize into pairs. Fluorescence microscopy is used to read the colour code of each droplet, mapping the position of each sample and detection mix in the chip and droplets in each well are merged, initiating all reactions across the chip nearly instantaneously. After incubation, the reaction result for each well is read using fluorescence microscopy and mapped back to the colour codes of the sample and/or detection mix in each well. c, Microwell design optimized for droplets made from PCR products or detection mixes. d, Dimensions and layout of a standard chip. The area covered by the microwell array is shown in light blue. e, Photograph of a standard chip. f, Photograph of a standard chip sealed inside an acrylic loader, ready for imaging.

Extended Data Fig. 2. Detailed schematic of loader and chip function in CARMEN.

Step 1, samples are amplified, colour coded and emulsified. In parallel, detection mixes are assembled, colour coded, and emulsified. Step 2, droplets from each emulsion are pooled into a single tube and mixed by pipetting. The pooling step is rapid to minimize small molecule exchange between droplets (see Supplementary Discussion 4). Step 3, the droplets are loaded into the chip in a single pipetting step. Side view, the droplets are deposited through the loading slot into the flow space between the chip and glass. Tilting the loader moves the pool of droplets around the flow space, allowing the droplets to float up into the microwells. Step 4, the chip is clamped against glass, isolating the contents of each microwell, and imaged by fluorescence microscopy to identify the colour code and position of each droplet. Step 5, droplets are merged, initiating the detection reaction. Step 6, the detection reactions in each microwell are monitored over time (a few minutes to 3 h) by fluorescence microscopy.

Extended Data Fig. 3. CARMEN multiplexed detection nomenclature and detection of Zika sequences.

a, Assay, test and droplet-pair replicate nomenclature. Each multiplexed assay consists of a matrix of tests, where the dimensions of the matrix are M samples × N detection mixes. Each test is the result of one sample being evaluated by one detection mix, where the result of the test is the median value of a set of replicate droplet pairs in the microwell array. b, Plate reader data for SHERLOCK detection of synthetic Zika sequences at 3 h (n = 3 replicates). c, Comparison of plate reader and droplet (Fig. 1c) data. Replicates: n = 3 for plate reader data. Numbers of replicates for droplets data are indicated in teal. Error bars represent s.e.m. d, Bootstrap analysis of Zika detection in droplets. e, ROC curve for Zika detection in droplets.

Extended Data Fig. 4. Design and characterization of 1,050 colour codes.

a, Design of 1,050 colour codes. b, Schematic for characterization of 210 colour codes and the 3-colour dimension of 1,050 colour codes. c, Raw data from characterization of 210 colour codes. d, Performance of 210 colour codes in 3-colour space. e, Performance of 1,050 colour codes in 3-colour space. f, Illustration of the sliding distance filter (circle) in 3-colour space. g, Characterization schematic and performance of 1,050 colour codes in the 4th colour dimension.

Extended Data Fig. 5. mChip and HV panel design schematic and statistics.

a, Dimensions and layout of mChip, compared to a standard chip. The area covered by the microwell array is shown in purple. b, AutoCAD rendering of acrylic moulds used for mChip fabrication. c, Photograph of an mChip. d, Left, AutoCAD rendering of each part of the mChip loader; middle, AutoCAD rendering of the set-up of an mChip loader; right, AutoCAD rendering of an mChip in a loader, ready to be loaded. e, Photograph of an mChip being loaded. f, Loading and sealing mChip, corresponding to steps in Extended Data Fig. 2 (step 3, mChip loading). Droplets are deposited at the edge of the chip into the flow space between the chip and the acrylic loader. Tilting the loader moves the pool of droplets around the flow space, allowing the droplets to float up into the microwells. Step 4, the chip and loader lid are removed from the base and sealed against PCR film. No glass is used to seal the mChip. The sealed mChip, suspended from the acrylic loader lid, can be placed directly onto the microscope for imaging. g, Photograph of an mChip sealed and ready to be imaged. h, HV panel design. At the time we designed the panel (October 2018), there were 576 HV species with at least 1 genome neighbour in NCBI, and 169 with ≥10 genome neighbours. We aligned genomes by segment and analysed the sequence diversity using ADAPT to determine optimal primer and crRNA binding sites (see Methods, ‘HV panel design’ for details). i, Number of species in each family in the HV panel design. j, Number of primer pairs required to capture at least 90% of the sequence diversity within each species. Two species required the use of primer pairs containing degenerate bases. k, Number of crRNAs required to capture at least 90% of the sequence diversity within each species. l, The fraction of sequences within each species covered by each designed crRNA set; we were able to design small crRNA sets with 90% or greater coverage for 164 of the 169 species. m, n, To compare expected and observed performance for the HV panel, primers (m) and crRNAs (n) were classified into on-target, low activity or cross-reactive by sequence analysis (blue or black) or on the basis of experimental data (orange).

Extended Data Fig. 6. crRNA performance during HV panel testing.

a, Individual guide performance in rounds 1 and 2. Redesign and re-dilution between rounds of testing are indicated between the data from rounds 1 and 2. On-target: reactivity above threshold for intended target only. Cross-reactive: off-target reactivity above threshold. Low activity: no reactivity above threshold. b, Summary bar graph of crRNA performance in rounds 1 and 2. c, Summary table of redesign, re-dilution and concordance between rounds 1 and 2 for unchanged tests. d, e, Round 1 (d) and round 2 (e) ranked AUCs for ROCs for on-target versus off-target reactivity in round 1 of testing. Representative on-target and off-target distributions are shown for the indicated ranks.

Extended Data Fig. 7. Synthetic target testing with HV panel.

a, Sample handling and data analysis for unknown samples. Following multiplexed PCR with 15 pools, PCR products are combined into sets of 3 (PCR metapools). A subset of the crRNAs correspond to the primers in each PCR metapool, shown by the colours in the expanded heat map. Composite heat maps are generated by combining data from the metapools in the expanded heat map. b, Five synthetic targets (10⁴ copies per μl) were amplified with all primer pools and detected using 169 crRNAs from the HV panel plus HCV crRNA 2. The heat map indicates background-subtracted fluorescence after 1 h.

Extended Data Fig. 8. Testing of clinical samples with HV panel and performance of influenza A subtyping.

a, CARMEN testing of patient samples and healthy pooled controls using the HV panel. Colour bar indicates fold change above background at 1 h for most crRNAs (3 h time point is shown for HIV and HCV crRNAs). Tests that could not be interpreted owing to the presence of signal above background in the negative controls are coloured in dark grey (not interpretable). Sample types: N, throat and nasal swabs; O, pooled healthy controls; P, plasma; S, serum; and W, water. Orange asterisks indicate signal above threshold (sixfold higher than background). b, Comparison of results from CARMEN, RNA sequencing-based identification of the sequence targeted by the indicated crRNA (Seq_CAR.), RNA sequencing-based identification of any sequences from the indicated virus (Seq_All), RT–PCR for the indicated virus, and a priori expectation based on information from the patient sample provider (a priori) for 4 dengue, 4 Zika, 20 influenza A, 26 HIV and 4 HCV patient samples. CARMEN testing was done over three rounds (as indicated by vertical separation between sections). Threshold cut-offs for making calls were: CARMEN, sixfold higher than background; Seq_CAR., 2 reads; Seq_All, 1 read per million (RPM); RT–PCR, according to the manufacturer’s instructions. Tests were considered uninterpretable when signal above background was observed in healthy pooled control samples assayed in parallel with patient samples. Heat maps indicate background-subtracted fluorescence after 1 h for most crRNAs (3 h time point is shown for HIV and HCV crRNAs). c, Heat map showing the full set of crRNAs designed to capture influenza N- sequence diversity. We tested 35 synthetic targets (10⁴ copies per μl) using 35 crRNAs. Grey, below detection threshold; green, fluorescence counts above threshold; orange outlines, subtypes; lowest row displays which targets are detected. Time, 3 h.

Extended Data Fig. 9. HIV reverse transcriptase mutation detection and future directions for CARMEN–Cas13.

a, Distributions of droplet fluorescence for each HIV reverse transcriptase crRNA–target pair after 30 min in most cases; 3 h time point for V106M and M184V. SNP indices in Fig. 4b are calculated from the medians of these distributions. b, Comparison of prior expectation based on Sanger sequencing from the patient sample provider (Sanger), CARMEN testing (CARMEN), and NGS of RNA from each sample (NGS) for 22 patient samples infected with wild-type HIV (No DRMs) or HIV bearing known drug resistance mutations (known DRMs). In some cases, NGS revealed a high number of mismatches (MM) between the HIV sequence in the sample and the crRNA sequence used in the CARMEN HIV reverse transcriptase DRM panel. Summary tables at the right quantify concordance between CARMEN and Sanger sequencing or CARMEN and NGS. c, Quantitative CARMEN–Cas13 schematic showing amplification primers containing T7 or T3 promoters, leading to increased signal for the majority (T7) product after Cas13 detection. d, Increased dynamic range of detection using quantitative CARMEN–Cas13. Dynamic range is indicated using coloured bars above the graph. Error bars indicate s.e.m. Replicates (n) for T7 and T3 data are noted in colour-coded text beneath the plot.