Cell-Free Synthetic Biosensors Based on CRISPR/Cas Mediated Cascade Signal Amplication for Precise RNA Detection

Synthetic biology has been harnessed to create new diagnostic technologies. However, most synthetic biosensors involve error-prone amplication steps and limitations of accuracy in RNA detection. Here, we report a cell-free synthetic biosensing platform, termed as SHARK (Synthetic Enzyme Shift RNA Signal Amplier Related Cas13a Knockdown Reaction), to eciently and accurately amplify RNA signal by leveraging the collateral cleavage of activated Cas13a to regulate cell-free enzyme synthesis. Based on cascade amplication and customized enzyme output, SHARK behaves a broad compatibility in different scenarios. Using a personal glucose meter, we detected 50 copies/μl SARS-CoV-2 on a SHARK-loaded paper. In addition, when combined with machine learning, SHARK can perform bio-computations and thus provide miRNA patterns for cancer diagnosis and staging. SHARK shows characteristics of precise recognition, cascade amplication and customizable signal outputting in one pot comparisons with established assays based on 64 clinical samples, presenting great potential in developing next-generation RNA detection technology.


Introduction
Rapid, accurate, and cost-effective detection of RNA biomarkers plays important roles in aiding pandemic prevention, medical treatment, and prognostic management [1][2][3][4] . Generally, quantitative reverse transcription PCR (qRT-PCR) has been engaged as the "gold standard" method for measuring gene strands in clinic 5 , but some challenges remain when detecting RNA species, due to their low abundance, easy degradation, short length, and single nucleotide mutations, etc 6,7 . Inadequate access to the qRT-PCR equipment has also limited the application for RNA detection in resource limited settings. Recently, various strategies, such as isothermal ampli cation methods 8 and nanoprobing techniques 9 , have been developed to sense RNA targets and output uorescent or colorimetric signals for readout. However, these strategies involve multiple ampli cation reactions that may cause nonspeci c ampli cation and false positive errors, yielding signals with poor accuracy and reliability 10 .
Synthetic biology based on cell-free gene expression (CFE) system has been harnessed to create new diagnostic technologies to detect a variety of pathogens and disease biomarkers 11 . The RNA detection process of synthetic biosensors is programmed to recognize the gene targets through base pairing, amplify signals via biosynthetic reactions, and express reporter proteins (e.g., uorescent proteins or various enzymes that can further augment the signals by catalyzing substrates), outputting tailor-made signals for different detection scenarios 12,13 . Usually, these synthetic sensors generate limited increase in signal (<10 3 fold), and thus in most cases a separate gene ampli cation step is needed to augment and convert the target signal [13][14][15] , leading to issues of poor accuracy and tedious operation. CRISPR/Cas13a can accurately identify and cleave speci c RNA sequences and further augment the signals by means of collateral nonspeci c catalytic cleavage, presenting excellent capabilities in accurate nucleic acid recognition and signal ampli cation 16 . For example, Cas13a can recognize and cleave single-stranded RNA targets, and its collateral activity is subsequently activated to cleave almost all reporter RNA, generating a uorescent signal as readout 17 . Although the dynamic range is over seven orders of magnitude of activated Cas13a-catalyzed cleavage 18 , these strategies still need to be combined with additional ampli cation steps to meet the detection needs. Considering that activated Cas13a may carry out unspeci c cleavage of mRNA in CFE process 19,20 , it is convincing that coupling the cell-free synthetic biosensor with CRISPR/Cas13a will solve the issues of target recognition accuracy and signal ampli cation integration, and realize portable and easy-to-use platform for RNA detection with customized signals.
In this study, we developed an RNA detection platform by coupling CRISPR/Cas13a with CFE system for precise recognition and cascade ampli cation of different RNA species, termed as SHARK (Synthetic Enzyme Shift RNA Signal Ampli er Related Cas13a Knockdown Reaction). In brief, trace RNA targets in tested samples were rst recognized by the CRISPR RNA (crRNA), and then activated the Cas13a to digest the mRNA in CFE system, reducing the expression of reporter enzymes and outputting electrochemical, colorimetric, or bioluminescent signals with a "turn-off" behavior. Integrating the SHARK into different point-of-care devices, we rapidly detected the virus RNA of SARS-CoV-2 via a portable glucose meter, and sensitively operated bio-computations with multiple microRNAs as inputs for cancer diagnosis in paper analytical devices and digital-chips, indicating that our SHARK platform holds a great potential for developing rapid, accurate, and multi-module strategies for disease diagnosis.

Results
SHARK enables accurate and sensitive RNA detection in a single reaction. In synthetic biosensors, the plasmid is transcribed into mRNA to initiate the protein expression, and the degradation of mRNA will obviously in uence the output signals. Here, in SHARK platform, CRISPR/Cas13a can precisely recognize the RNA targets and activate its collateral cleavage activity, which is then used to regulate the mRNA levels to generate different concentrations of enzymes, nally outputting different modules of signals (Fig. 1a). To testify our hypothesis, the activation of CRISPR/Cas13a was investigated by incubating the target RNA with unspeci c mRNA molecules produced by a HiScribe T7 High Yield RNA Synthesis Kit. With increasing target concentrations, most transcribed mRNA strands were digested into fragments, indicating that the target RNA recognized the crRNA and activated the catalytic cleavage (Fig. 1b). Then, we coupled the CFE system with CRISPR/Cas13a to build a series of SHARK platforms, which generated different kinds of enzymes that could be detected by many commercial kits, including sucrose invertase (sInv), glucose-6-phosphate dehydrogenase (G6PD) and luciferase (Luc) (Supplementary Fig. 1a). The gel results showed that compared with nonsense control RNA, the target RNA selectively activated the SHARK, and signi cantly repressed the enzyme expression (Fig. 1c). The kinetics of SHARK was also determined across a range of different concentrations of target RNA. For example, the activation of the SHARK sensor generating G6PD was easily visualized by naked eyes, and it exhibited much faster kinetics compared with that expressing enhanced green uorescent protein (eGFP) as reporters (Fig. 1d), suggesting that the cascade ampli cation of signals via mRNA translation and enzymatic reaction greatly bene ted the detection. The experimental (Supplementary Fig. 2a) and simulation results ( Supplementary Fig. 2b,c) con rmed that as low as 10 fM RNA targets could trigger signi cant signal distinction with high reliability (CV = 5.51%). Taken together, SHARK could sensitively detect trace RNA at femtomolar range, behaving a better performance compared with other strategies, such as Cas13a-based detection 17 and toehold switch synthetic gene networks 14 (Fig. 1e).
SHARK output different modules of signals. The great advantage of CFE system is the versatile capability in rational design and integration of different reporter proteins. However, so far, only limited enzymes have been explored as reporters in the synthetic bio-sensors, such as eGFP and β-galactosidase 14,21 .
Here, to facilitate the signal readout, we explored to introduce several enzymes (i.e., sInv, G6PD and Luc) that can be easily detected using commercial kits into SHARK as reporters (Supplementary Note 1). First, we used sInv as reporters to build a biosensor (sInv-SHARK). The addition of target RNA activated SHARK, degraded the mRNA, and reduced the sInv expression, which catalyzed the conversion of sucrose to glucose (Fig. 2a). After loading the samples, we conveniently and sensitively detected trace RNA through measuring the glucose levels using a personal glucose meter (PGM, Accu-Chek Performa) with a limit of detection (LOD) at 11.72 fM (Fig. 2b). In contrast, we also developed a G6PD-SHARK sensor that converted the RNA signal to G6PD, and nally to the production of formazan, outputting colorimetric signals that can be visually measured without need for any instruments ( Fig. 2d) with LOD at 9.50 fM ( Fig. 2e). Differently, in a sensor of Luc-SHARK, the target RNA reduced the expression of luciferase, which catalyzed the bioluminescence reaction ( Fig. 2g) with high activity, low background noise, and fast response time, bene ting the RNA detection with a large dynamic range (10 − 4 -10 2 nM) and LOD at 6.55 fM (Fig. 2h).
Then we investigated the speci city of SHARK. Most isothermal exponential ampli cation methods show poor performance in distinguishing gene strands with high sequence homology 10,22,23 . As designed, CRISPR/Cas13a system has the capability to discriminate single nucleotide mutations since that it requires strict crRNA-target paring to activate the collateral cleavage 18,24 (Supplementary Fig. 3a). To testify our hypothesis, we synthesized three RNAs analogs with single-base mutation. As expected, in all three kinds of sensors of SHARK, the RNA targets strictly recognized the cognate crRNA and yielded reduced signals, whereas those with one-base mismatch barely activated CRISPR/Cas13a system, and generated strong signals similar to those from scrambled RNA (Fig. 2c,f,i), indicating that SHARK possessed high speci city in discriminating single nucleotide alterations. Further, the interference of human total RNA in samples was also investigated, and the biosensors outputted reliable signals even with a high molar ratio of target RNA to total RNA at 1:1000 ( Supplementary Fig. 3b-d). Thus, our SHARK has provided customizable signals with high sensitivity, speci city, and reliability, inspiring us to apply it in point-of-care testing of various RNA species at different scenarios.
Paper-loaded sInv-SHARK sensor facilitates rapid and reliable detection of SARS-CoV-2 using PGM. Infectious diseases caused by different viruses exert great threats to human health, and to date, the COVID-19 coronavirus pandemic has caused more than two million deaths 25 . Similar to other deadly respiratory syndromes (e.g., Severe acute respiratory syndrome coronavirus (SARS), Middle East Respiratory Syndrome Coronavirus (MERS)), rapid and accurate detection of SARS-CoV-2 will greatly bene t the contact tracing, treatment options and isolation requirements, meaning the urgent need to develop point-of-care technologies available at a variety of scenarios 26,27 . To favor the testing of SARS-CoV-2 at home, we designed a paper-loaded freeze-drying sInv-SHARK sensor with rapid readout by a PGM, which is easily accessible on the market (Fig. 3a). First, we constructed contrived SARS-Cov-2 and synthesized three crRNAs targeting the genes of replicase polyprotein 1ab (Orf1ab), nucleocapsid (N) and envelope protein (E) in SARS-CoV-2, respectively (Supplementary Fig. 4) 27 . Here, the combination of multiple crRNAs could activate multiple nuclease-inactive ribonucleoprotein complex 23 , and thus effectively increase the active Cas13a concentrations so as to improve sensitivity. When mixing the three crRNAs in the detecting solution containing contrived SARS-Cov-2 samples, the sInv-SHARK sensor accurately identi ed virus concentrations as low as 1 fM, outputting a signal value of 21.5 ± 0.9 with high reliability (CV = 4.18%, Fig. 3b). Considering the actual viral loads in clinic samples 26 , we serially diluted the pseudovirus on throat swab (from 10 to 7×10 8 copies per swab) with scrambled RNA as controls. The sInv-SHARK sensor outputted a decreased signal between 29.7 and 1.4, with a dose-dependent response from 20.2 to 5.3 at a concentration range of 10 − 5 -10 2 nM (i.e., ~10 3 to 10 7 copies/µL) with high accuracy and reliability, setting a cut-off value at 22.5 (~ 50 copies/µL) (Fig. 3c).
To facilitate the detection at home, we integrated sInv-SHARK with a PGM into a disposable device previously developed in our lab 28 , in which the users only need to drop the samples on the loading paper and press the start button, outputting a digital result within 1.5 h (Fig. 3d, Supplementary Fig. 5). We used qRT-PCR method to calibrate the sInv-SHARK-based detection, and several dilutions of pseudovirus at a range of 1.5×10 2 to 10 8 copies/µL were measured with three negative swabs as controls. The sInv-SHARK device generated reliable and easy-to-readout results between 5.3 and 22.5 that was highly consistent with those from the Ct values of qRT-PCR between 14 and 37 (Pearson r coe cient = 0.96) ( Fig. 3e), meaning that the cut-off value of 22.5 was equivalent to Ct = 37 (the standard qRT-qPCR value from Disease Control and Prevention (CDC) of China). Further randomized double blind tests in laboratory con rmed that both the sInv-SHARK and qRT-PCR methods showed high coincidence in detecting positive samples ( Supplementary Fig. 6), as well as in discriminating negative control samples (Supplementary Table 1).
Finally, we investigated the cross-reactivity between SARS-CoV-2 and other respiratory diseases related pathogens (e.g., MERS, human coronavirus HKU1, In uenza A virus (IVA), etc.). The heat map (Fig. 3f) re ected that the SHARK device could easily discriminate the pathogen RNA from others with the same concentration (1 nM). In turn, by designing cognate crRNA, our device could also detect the RNA targets from different pathogens (Fig. 3g, h). Taken together, sInv-SHARK-based device could sensitively and accurately sense the SARS-CoV-2 RNA and output digital signals with high reliability and robustness against matrix effect and interference ( Supplementary Fig. 7a, b), presenting as a convenient and reliable strategy for rapid detection of SARS-CoV-2 at resource limited settings such as home or community.
Paper-based G6PD-SHARK sensor enables multiple miRNA detection for cancer staging. miRNA is a class of endogenous small noncoding RNA molecules, which serve as promising clinical biomarkers in various human diseases, including cancer 29 . However, the characteristics of miRNAs, such as short length, high sequence homology, and low abundance, make it a great challenge to develop sensitive and speci c sensors 30 . CRISPR/Cas systems possess special capability in recognizing short gene targets (20-40 nucleotides) with high speci city and reliability, making it possible to sensitively detect miRNAs.
Considering the exibility of our SHARK platform, we designed a pattern pad to detect multiple miRNA targets in serum from the patients of non-small cell lung cancer (NSCLC), offering abundant information to evaluate the cancer staging.
We rst built a data mining model to select biomarkers via bioinformatics tool, and seven miRNAs highly related to NSCLC (Fig. 4a) were selected with four upregulated (i.e., miR141, miR-30a, miR-20a, Let-7b) and three downregulated (i.e., miR-126, miR-328, miR-19b) compared with healthy people by analyzing the miRNA expression database of 915 NSCLC and 105 healthy individuals from The Cancer Genome Atlas (TCGA). Then we designed a G6PD-SHARK based paper origami device with two detection zone respectively for loading testing sample (T) and control standard sample (C). The G6PD-SHARK sensor was pre-packed in the top layer through a freeze-drying procedure, which was folded with the base layer together to assemble the device for sample loading and signal ampli cation, generating an easy-to-read visual output (Fig. 4b, Supplementary Fig. 8a, b). To easily extract the results, we quanti ed the colorimetric signals via Image J software and used the ratio of T to C (T/C) to re ect the miRNA concentrations (Fig. 4c). Using a synthetic miR-20a sample, we veri ed that the T/C signal was proportional to the miRNA concentration ranging from 10 − 5 and 10 − 1 nM with LOD values as low as 7.56 fM, and thus the sensor could be used to detect most miRNAs in serum (10 − 5~1 0 nM) 31,32 . Meantime, the sensor also showed excellent capability in discriminating different miRNA targets (Fig. 4d) even in the presence of 10-fold interferences with the signal deviation less than 5.21%. Especially, in the sensor targeting let-7b, which is highly overexpressed in most cancers 33 , we found other members of let-7 family (i.e., let-7a, let-7c, let-7d) with few mismatched bases had little interference (less than 4.97%) at the concentrations of 10, 50, and 100 pM ( Supplementary Fig. 9), suggesting that the G6PD-SHARK platform guaranteed the detection of miRNAs with high sensitivity and speci city to distinguish single nucleotide mutations.
In clinic, more than 70% of NSCLC patients are diagnosed at advanced stages with a poor ve-year survival rate (~ 19.40%) reported by the Surveillance, Epidemiology, and End Results program, and early diagnosis will greatly bene t the treatments 34 . We next used our G6PD-SHARK-based sensor to detect the serum samples from lung cancer patients (Supplementary Table 2). We observed the up-regulation of miR-30a (2.91 ± 0.12), miR-20a (2.82 ± 0.15), Let-7f (3.13 ± 0.15), and down-regulation of miR-126 (0.61 ± 0.05), miR-328 (0.24 ± 0.01) in the patients with moderate sensitivity (40.25%-70.71%) and speci city (29.28%-71.25%), showing good correlation to the bioinformatics pro les. ROC curve analysis showed that ve miRNAs among them could be promising biomarkers for NSCLC, with AUC values from 0.64 to 0.76. Further, logistic regression analysis was used to construct combined models to evaluate these 5 candidates to verify the NSCLC signatures. Combining the ve miRNAs (i.e., miR-30a (+), miR-20a (+), Let-7f (+), miR-126 (-), miR-328 (-)) enhanced the AUC to 0.88 (sensitivity = 84.24%, speci city = 86.51%), indicating that multiple miRNA patterns are more reliable in cancer diagnosis than the single biomarker ( Fig. 4e). Considering that the miRNA signatures may vary with cancer progress, we sought to nd the relationship between the miRNA pro les and TNM (tumor, nodes, and metastasis) stage of NSCLC. The miRNA signatures of different patients at different stages (I, II, III, IV) diagnosed by pathological section were pro led in Fig. 4f using G6PD-SHARK-based sensor and qRT-PCR. The miRNA pro les showed a high coincidence between our SHARK sensor and qRT-PCR method (Pearson r coe cient = 0.84) ( Supplementary Fig. 10a). The patients at the same stage shares similar miRNA signatures, which also dynamically changed along with the cancer progression. That is, miR-126, miR-328, and miR-30a were positive at advanced stage, but negative at stage I ( Supplementary Fig. 10b-d), while miR-20a exhibited higher expression at all stages compared with healthy people (Supplementary Fig. 10e). Overall, our G6PD-SHARK-based sensor showed high sensitivity and speci city in detecting fetamolar levels of short miRNAs in the serum of lung cancer patients, and provided unique miRNA patterns for cancer diagnosis and staging.
G6PD-SHARK-CPU sensor enables miRNA computation for cancer diagnosis. To realize early cancer diagnosis at home, we sought to optimize G6PD-SHARK sensor by outputting one simple diagnostic result from multiple miRNA detection through bio-computation (Fig. 5a). First, we built a computational classi er with a support-vector-machine (SVM) model according to the published work 35 , and trained the system using the miRNA pro les of NSCLC patient and healthy individuals from TCGA. To avoid overtting of machine learning, different models were evaluated to improve the classi cation performance ( Supplementary Fig. 11, Supplementary Note 2) with a training database (105 NSCLC and 60 healthy samples), and generated an optimized model including seven miRNA inputs (miR-761 (-), Let-7f (-), miR-126 (-), miR-20a (+), miR-515 (+), Let-7b (+), miR-5688 (+)) associated with weights ranging from inter values of -5 to + 9, achieving identi cation accuracy of 85.72%, sensitivity of 87.11%, and speci city of 78.12% (Fig. 5b, Supplementary Table 3).
To facilitate the computation using the up-regulated and down-regulated miRNAs, we integrated a Cas9 system with the G6PD-SHARK to develop central processing unit (G6PD-SHARK-CPU) to transform miRNAs information. We combined the transcriptional and translational regulation systems together to tune the signal by using the up-regulated miRNAs related NSCLC to turn off the signal via G6PD-SHARK (OFF) and using the down-regulated miRNAs to turn on the signal via the Cas9-RsgRNA (ON) (Supplementary Fig. 12a, Supplementary Note 3). To test the ON/OFF system, we introduced G6PD reporter plasmids and measured the colorimetric signal as an output (Absorbance < 0.5 as TRUE). We observed 30-fold activation with high sensitivity (0.1 pM) and speci city when the target miRNA present in the ON system (Fig. 5c left, Supplementary Fig. 13), and also signi cant suppression in the OFF system (Fig. 5c right). Next, we designed an A NIMPLY B gate with miR-126 (-) and miR-5688 (+) as inputs, and only the input of (0, 1) could output a TRUE signal (1) (Fig. 5d). Finally, we tested G6PD-SHARK-CPU system with clinical serum samples from NSCLC and healthy individuals. The optimal diagnostic point was calculated at the probability score of 0.55 as cut-off value (Fig. 5e) from the receiver operating characteristic (ROC) curves (Fig. 5f). Through bio-computation, 28 of 38 NSCLC patients were identi ed with a sensitivity of 77.78% (95% CI: 64.92%, 88.28%), while 1 of 26 healthy samples was misclassi ed with speci city of 96.15% (95% CI: 81.11%, 99.81%), and accuracy of 86.42% (95% CI: 65.82%, 96.14%) (Fig. 5g). Despite that we simpli ed the miRNA transformation, G6PD-SHARK-CPU sensor still presented as a promising user-friendly platform to favor the routine early cancer screening.
Digital-Luc-SHARK device achieves absolute quanti cation of miRNA. Usually, the fold change of miRNAs is subtle at early stage of cancer. Inspired by the high sensitivity and speci city of SHARK in distinguishing miRNAs, we next integrated QuantStudio™ chip (20K data points) to design a digital-Luc-SHARK device with luciferase generating bioluminescence as reporters (Fig. 6a). To improve the signal, we constructed a fusion protein consisting of luciferase and yellow uorescent protein (YFP), exhibiting a 7.45-fold enhancement over native luciferase via bioluminescence resonance energy transfer (BRET) 36 (Fig. 6b). Then we designed a portable luminescence detection device consisting of an intensi ed CCD camera, a mirror and a lter with magni er (Fig. 6c, Supplementary Fig. 14). After mixing the miRNA samples with the luc-SHARK reagents and dispersing it in the picoliter-scale reactors, we observed the increase of the positive spots on the chip with the increase of target miRNAs, which was consistent with Poisson distribution equation (Supplementary Fig. 15), presenting a dynamic range from 100 to 5 ×10 4 copies/µL (2 to 1000 aM) (Fig. 6d), and showing over 10 4 -fold enhancement in sensitivity compared with the tube-based SHARK process. Further, by employing an IC Capture software, the full view of the chip was collected and analyzed, generating a detection linear range of 200 − 20,000 copies/µL ( Supplementary Fig. 16a, b). By virtue of the strict recognition, the digital-Luc-SHARK device also showed high speci city in discriminating the members of Let-7 family with interference less than 0.51% (Fig. 6e).
We next evaluated the accuracy of digital-Luc-SHARK, digital PCR, and qRT-PCR in quantifying miR-20a in clinic serum samples from NSCLC patients. All tests generated positive results, and the relative expression results from digital SHARK showed similar variations to those measured with digital PCR (Fig. 6f, Supplementary Fig. 17). Finally, we used this handheld digital-Luc-SHARK device to precisely detect the miRNA markers for early diagnosis of NSCLC. Four miRNAs (i.e., let-7b, miR-126p, miR-20a and miR-30a), previously tested as NSCLC biomarkers, were picked for validation. Figure 6g re ected that the up-regulation of let-7b, miR-30a, and miR-20a was highly related to the early-stage patients (Stage I, II) with Let-7b and miR-30a four times higher than those in the healthy people. The accurate and reliable quanti cation of miRNAs using our handheld digital-Luc-SHARK device makes it promising to achieve convenient early NSCLC diagnosis at home.

Discussion
In the present study, we developed a cell-free synthetic biosensor platform, i.e., SHARK, for detection of a variety of RNA species. By leveraging the collateral cleavage of activated Cas13a to regulate the mRNA translation, we achieved strict recognition, cascade ampli cation, and customizable signal output in one pot. This SHARK platform shows excellent compatibility in integrating with various point-of-care devices, and outputs different modules of signals, enabling the diagnosis of different infectious and chronic diseases with high sensitivity and reliability at resource limited settings.
Synthetic biology has provided a powerful tool to develop diagnostic technologies to sense various targets (e.g., gene sequences 15 , biomarkers 37 , chemical contaminants 38 ), regulate the ampli cation via transcription and translation, express a series of enzymes, and output diverse detectable signals. One limitation of our SHARK platform is that it needs 1.5-3 hours to complete the entire detection process. Through optimizing components in SHARK, it is promising to further optimize the ampli cation processes, and select enzymes with high catalytic kinetics to realize rapid detection 39 . Another challenge is how to linearize the ampli cation response of the SHARK to t the bio-computation with multiple inputs 40 . For broad clinical use, mathematical models and programmable automations should be developed for user-de ned diagnosis. Taken together, we envision that our SHARK will be a versatile plugand-play platform to develop sensitive, low-cost and easy-to-use diagnosis technologies. One pot SHARK assay for RNA detection. One pot SHARK assay was performed using LbuCas13a for the trans-cleavage assay and cell free enzyme expression for signal transduction of RNA targets. The nal concentration of each reagent in the system was optimized (Supplementary Note 1). In brief, 11 µL of cell free enzyme expression reaction system was mixed with 4 µL of a CRISPR reaction buffer, and then we added 5 µL of target RNA and reaction buffer. After incubation at 37 ℃ for 90 min, the corresponding enzyme substrates were added for signal readout.

Methods
Colorimetric assay readout. The colorimetric assay was performed in 10 µL Tris buffer (50 mM, pH 8.5) containing 0.5 mM NADP + , 10 mM WST-8, 10 mM mPMS. After signal ampli cation in G6PD-SHARK assay, these mixtures were added and incubated for 5 min at 37 ℃ to produce orange yellow formazan. The absorbance at 450 nm was measured with a multimode microplate reader (Tecan Spark 10M).
PGM assay readout. Sucrose was hydrolyzed to glucose by sInv at 37 ℃, which was then detected by PGM. After signal ampli cation in sInv-SHARK assay, 10 µL 0.5 M sucrose was added and then incubated for 1 min. Finally, 5 µL of the reaction solution was tested by a PGM.
Bioluminescence assay readout. After signal ampli cation in Luc-SHARK assay, 20 µL of bioluminescent substrate (Beyotime, RG051M) was added, and the bioluminescent signal was measured using a bioluminescence reader after 5 min at 37 ℃ in dark.
Mathematical modeling. Rate equation model was established based on the transcription, translation and collateral cleaving process for SHARK assay as follows: (1-4).
The Michaelis-Menten constants and parameter values were depicted in Supplementary Table 6.
Preparation of contrived SARS-CoV-2. The contrived SARS-CoV-2 contains partial ORF1ab gene sequence, E Gene and N Gene coding region sequences of SARS-CoV-2. These target fragments were synthesized and cloned into a retroviral vector to perform as pseudovirus. The pseudovirus was synthesized and inactivated by Beijing Tsingke Biotechnology Co., Ltd.. We spiked inactivated virus into human plasma/serum (IRMM, ERM-DA470K), Saliva (Solarbio A7990) or simulated lung uid (Tegent, 1700 − 0800) and diluted to simulate clinical samples. Then we followed the instructions of RNA extraction kit (Qiagen, 52904) to extract pseudovirus RNA.
Fabrication of integrated device for SARS-CoV-2 detection. The integrated device was composed of three modules, i.e., paper-based nucleic acid extraction module, SHARK module and PGM detection module.
The whole device was designed by SolidWorks 28 . The detection module contains one PGM and one test strip.
Multi-channel paper-based device fabrication. Paper Analytical Device (PAD) for multiple miRNAs detection consisted of Whatman paper and two layer doubled-stick tapes. The detection pattern of PAD was designed with CorelDRAW and printed with laser engraving machine. The two layer doubled-stick tapes were enabled to isolate different channel area and form a liquid reservoir for sample loading.
Bioinformatics analysis. NSCLC related miRNA expression data come from TCGA. The screening search formula was follows, (NSCLC OR Non-small cell lung cancer OR Lung squamous cell carcinoma OR Lung adenocarcinoma) AND (Serum OR Plasma OR Free nucleic acid) AND (miRNA OR microRNA OR miR_). The adjusted P value was less than 0.05, and differential expression fold change was > 2 or < -2. R language package (edgeR) was used to screen differentially expressed miRNAs.  Table 7). The mixture was pre-incubated for 10 min at 37 ℃, and then different concentrations of target RNA (10 − 5 -10 3 nM) were added. The Reaction was incubated for 30 min at 37°C. The cleavage ability was analyzed by gel electrophoresis.
SVM training and validation. To train the SVM classi er, we obtained the miRNA pro le data of GSE40738 from TCGA, which was used to identify the up or down-regulated miRNAs in NSCLC compared with healthy individuals (The process details in Supplementary Note 2). Then the selected miRNAs were validated with associated weights. To get simple result by G6PD-SHARK-CPU, the up-regulated miRNAs (+) associated with NSCLC were responded to the SHARK assay, and the down-regulated miRNAs (-) were responded to the Cas9 assay. The concentrations of crRNAs and RsgRNAs were tuned to the relevant weights.
qRT-PCR. We reversed transcribed contrived SARS-CoV-2 RNA using Goldenstar RT6 cDNA synthesis kit (TsingKe, TSK301). miRNAs were reversely transcribed using reagents from the TaqMan miRNA Reverse Transcription Kit (Applied BioSystems, A25576) together with the stem-loop primers (Supplementary Table 8). The reverse transcription thermal-cycling procedure was performed at 16°C for 30 min, 42°C for 30 min, and 85°C for 5 min. And then PCR experiments were performed on an Applied Biosystems 7500fast instrument with the ChamQ universal qPCR master system (Vazyme, Q711-02).
Digital RT-PCR. Digital RT-PCR reaction was prepared by combining target RNA and QuantStudio 3D Digital PCR Master Mix (PN A26358) according to the manufacturer's instructions. Then the reaction solution was loaded in one QuantStudio 3D Digital PCR Chip (PN 100027736). End-point uorescence data were collected on the QuantStudio 3D Digital PCR instrument and analyzed using the QuantStudio 3D Analysis Suite software (Thermo Fisher Scienti c).
Digital SHARK assay. The digital SHARK reaction mixture (10 µL) was prepared by mixing the following reagents: 5.5 µL of CFE system mixed with 2 µL of a Cas13a reaction mixture, the volume was then brought up with target miRNA and nuclease free water. The optimized plasmid concentration in microwell   Point-of-care testing of SARS-CoV-2 with paper-based slnc-SHARK using PGM. a Work ow for analysis of paper-based slnc-SHARK sensor. The spiked pseudoviruses were extracted from the samples by the fully integrated paper-based, and then captured by three different crRNAs, which activates the Cas13a to cleave the mRNA and thus reduce the sInv expression, outputting an electrochemical signal collected by  Paper-based G6PD-SHARK outputting colorimetric signal for multiplex miRNAs analysis. a Bioinformatics analysis of miRNA expression pro les to predict target miRNAs for NSCLC diagnosis. b Schematics of the paper-based G6PD-SHARK for multiplexed detection of miRNAs (Top). Images of the paper-based colorimetric output of patient and reference sample (Down). c Colorimetric detection of different concentrations of RNA inputs. T/C values indicate the ratios of the signals from testing sample (T) to that from control buffers (C). d Orthogonality of the device for highly speci c detection of target miRNAs.
e Expression levels of miRNAs in cancer (Orange) and normal serum (White). Two downregulated and three upregulated miRNAs were all signi cant between NSCLC with non-NSCLC. The AUC scores were identi ed and compared with logistic regression model. All values were quantitatively measured using ImageJ. All data are represented as mean ± s.d.. * P<0.05 (two-tailed Student's t-test). f Postnormalization heat map of miRNA expression from NSCLC patients' serum con rmed by RT-qPCR and G6PD-SHARK. RT-PCR were performed in triplicate and Ct values were averaged. The heat map is scaled versus the control group with the up-regulated miRNAs shown as red (>1) and the down-regulated as blue (<1).  Absolute quanti cation of miRNA using digital-Luc-SHARK. a Schematic of the digital-Luc-SHARK assay.
Target miRNA together with SHARK mixture are emulsi ed with oil into the QuantStudio™ chip. If one droplet doesn't contain miRNA, the enzyme can be expressed and generate bioluminescent signal (Termed "0"). In turn, the miRNAs activate the Cas13a to digest the mRNA, thus yielding a positive droplet (Termed "1"). By coupling a handheld device, the signals on the digital SHARK chip can be easily Let-7f, miR-30a, and miR-20a for healthy people (blue) and early-stage (red) NSCLC patients detected using digital SHARK.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.