Improving the specificity of nucleic acid detection with endonuclease-actuated degradation

Nucleic acid detection is essential for numerous biomedical applications, but often requires complex protocols and/or suffers false-positive readouts. Here, we describe SENTINEL, an approach that combines isothermal amplification with a sequence-specific degradation method to detect nucleic acids with high sensitivity and sequence-specificity. Target single-stranded RNA or double-stranded DNA molecules are amplified by loop-mediated isothermal amplification (LAMP) and subsequently degraded by the combined action of lambda exonuclease and a sequence-specific DNA endonuclease (e.g., Cas9). By combining the sensitivity of LAMP with the precision of DNA endonucleases, the protocol achieves attomolar limits of detection while differentiating between sequences that differ by only one or two base pairs. The protocol requires less than an hour to complete using a 65 °C heat block and fluorometer, and detects SARS-CoV-2 virus particles in human saliva and nasopharyngeal swabs with high sensitivity.


Introduction
Line 34-36: The authors make the assertion that "Quantitative qPCR…may have difficulty distinguishing between closely related sequences". The opposite is true; probe-based qPCR is exquisitely sensitive when it comes to differentiating single nucleotide differences between related sequences. Figure 1 should be reorganized so that the panels flow in a more logical fashion. It is incredibly difficult to read. Figure 1b-d has multiple bands visible in the 19n wells, however, it is not stated which band is the correct amplicon size, and it is therefore difficult to determine the efficiency of the degradation as a strong band ~150bp is still present in the 19n reaction in Figure 1d. Figure S2a: What is the "expected target DNA sequence". Are you referring to the target sequence for the Cas9/gRNA or the target sequence for the LAMP primers? Also, the authors state that the difference in scores is "mainly driven by the difference in A (given B and C are similar)". They should state that it is the difference between A in the positive and negative reactions, and that B and C are similar in their respective positive and negative reactions. It reads as though the authors are claiming that B and C are similar in the positive reaction, and B and C are similar in the negative reaction. Based on the number of points plotted on the graph, it looks as though triplicates were performed. This is not a high enough sample size to determine the limit of detection when using 0, 10 and 100 copies / reaction. Figure S3 is incredibly difficult to interpret given the author jump between red and grey plot lines for 19n / bat. How does figure S3f differ from Figure 1l? Do they not both show SARS-CoV-2 RNA with it's corresponding on-target gRNA versus the non-target (bat) gRNA? Why is the room temperature reaction time data given in Fig. 1I as the SENTINEL score yet in Fig. S3f and S3g the data is given as a percent reduction in fluorescence?

Results
The data using the AfeI restriction enzyme is impressive yet the authors do not discuss that their results should a 5-10x increase in the SENTINEL score using the restriction enzyme. It would have been prudent to use the AfeI restriction enzyme with the saliva samples, which may have increased the LOD.
Reviewer #2 (Remarks to the Author): Generally, the authors fabricated an interesting strategy to improve the specificity of LAMP-based methods in detection of SARS-CoV-2 pathogenic nucleic acids. Specifically, various sequence specific endonucleases, including wild type Cas9, its engineered variants, Cas12a and AfeI, were used to introduce 5' phosphorylated double strand break or nicks, which subsequently digested by the highly processive Lamda Lambda exonuclease. In general, this is an interesting study, but there are some major points to be further modified before publishing in Communications Biology.
1. Besides false positive, the other drawback of LAMP method is the post-amplification contamination issue, in which amplified products with high concentration might easily splash or form aerosols during experimental operation, causing pollution to air, reagents, pipettes and the gloves and clothes of the operator. That's why in LAMP reactions, tube lids cannot be opened once the amplification initiated. However, in the protocol provided by the authors, LAMP amplified products has to be diluted and mixed with SENTINEL master mix, which may result in serious contamination for subsequent experiments in practical scenarios. The authors should discuss this issue in the manuscript.
2. I am curious that why not use RPA instead of LAMP. In RPA reactions, the authors may able to premix SENTINEL master mix with rt-RPA supplemented with 5'-OH labeled rt-primer and amplification primers to give a one-pot reaction at 37oC.
3. In the manuscript, readouts of detection have to be calculated by equation (1-A/B) * (B/C), in which 3 different SENTINEL reactions has to be set and performed and measured quantitively. This is unpractical in real diagnosis. I would expect value B and C to be quite constant since they are, experimentally speaking, negative controls. Can the authors give a simplified numeric score? For example, decreasing rates of reaction A (Slope A) after the addition of SENTINEL master mix.
4. The authors stated that "Furthermore, we found that this strategy could reliably detect down to approximately 100 ssRNA or dsDNA molecules per microliter of input sample, which is in the attomolar concentration range. The sensitivity of this assay is bounded by the sensitivity of the LAMP step, which has been shown to be comparable to gold standard quantitative PCR (qPCR)". When using pure nucleic acids fragments as template, LoD of LAMP reaction can easily reach 10 copies/reaction. However, when viral particles were used as templates, LoD of 10 particles per microliter was observed. The authors should explain possible reasons for this phenomenon, and provide representative photographs of LAMP results in SI.

Reviewer #1 (Remarks to the Author):
The authors describe their SENTINEL assay, which combines LAMP of SARS-CoV-2 RNA with endonuclease digestion of the amplicon to improve LAMP specificity. The principle justification for this approach is that other nucleic acid detection methods suffer from false-positive results. Yet we know that one of the principle concerns surrounding SARS-CoV-2 diagnostics is the increased proportion of false negative results in nucleic acid-based diagnostics as a result of the emergency of new variants. I don't believe that this work reflects a substantial improvement on SARS-CoV-2 LAMP diagnostics as the authors have not made a compelling case that RT-LAMP specificity is current a cause for concern for SARS-CoV-2 detection. Response: We thank the reviewer for a detailed reading of our manuscript and very constructive comments that greatly strengthen our study. The main goal of this study is to present a conceptually novel approach for nucleic acid detection; we used SARS-CoV-2 as proof of concept to demonstrate features and advantages of our approach. One proposed advantage of our system is to avoid the false-positive outcomes, and we agree that in our original submission, we did not present direct evidence of such in our example use cases. In our revision, we collaborated with a clinical laboratory who provided 50 blinded nasopharyngeal (NP) swabs from patient suspected to have COVID-19, and performed SENTINEL versus RT-LAMP alone on these samples.

Introduction
Line 34-36: The authors make the assertion that "Quantitative qPCR…may have difficulty distinguishing between closely related sequences". The opposite is true; probe-based qPCR is exquisitely sensitive when it comes to differentiating single nucleotide differences between related sequences. Response: We thank the reviewer for this correct comment. We have removed the previous assertion. Figure 1 should be reorganized so that the panels flow in a more logical fashion. It is incredibly difficult to read. Figure 1b-d has multiple bands visible in the 19n wells, however, it is not stated which band is the correct amplicon size, and it is therefore difficult to determine the efficiency of the degradation as a strong band ~150bp is still present in the 19n reaction in Figure 1d. Response: We agree with the reviewer that Figure 1 was indeed very disorganized. Figure 1 is now split into three separate figures (Figures 1 to 3) to improve clarity and logical flow. We realized that we failed to indicate what RT-LAMP product normally looks like for the general readership. Figure 1b now shows the product from RT-LAMP alone, which reveals multiple bands of increasing molecular weight, as expected. The sequence of these products is highly repetitive and contain multiple sites of Cas9 cleavage. Therefore, the strong ~150bp band only appears after Cas9 digestion, suggesting that Cas9 efficiently breaks apart large molecular weight DNA into 150bp units. The degradation efficiency using lambda exo is not 100%, as seen by the residual 150bp band (19n) in the new Figure 1e   action. Blue circles at the 5' ends of the RT-LAMP product represent 5' phosphorothiolate modifications. Red circles at the 5' ends represent 5' phosphates exposed from endonuclease action. b-e) Agarose gel electrophoresis of (b) SARS-CoV-2 N-gene RT-LAMP product alone, or with additional (c) Cas9 cleavage using on-target (19n) and non-target (neg) gRNA, (d) digestion with λ-exo, and (e) unwinding with RepX superhelicase. The RT-LAMP product (panel b) is not altered with exposure to λ-exo ('neg' of panel d), suggesting that 5' phosphorothiolate effectively prevents λ-exo end-degradation. c) On-target gRNA led to downward shift in gel bands, consistent with cleavage of RT-LAMP product. d) Addition of λ-exo did not lead to appreciable change in cleaved RT-LAMP product. e) Addition of λ-exo with Rep-X led to significant degradation of RT-LAMP product, only for the sample with on-target gRNA. In contrast, the sample with off-target gRNA was consistent across panels c-e. f) Quantification of panels b-d across two biological replicates, by measuring fractional intensities of each lane on the agarose gel relative to the maximum intensity across all lanes. From left to right, the first two lanes quantify panel c, second two lanes quantify panel d, and last two lanes quantify panel e. Figure S2a: What is the "expected target DNA sequence". Are you referring to the target sequence for the Cas9/gRNA or the target sequence for the LAMP primers? Also, the authors state that the difference in scores is "mainly driven by the difference in A (given B and C are similar)". They should state that it is the difference between A in the positive and negative reactions, and that B and C are similar in their respective positive and negative reactions. It reads as though the authors are claiming that B and C are similar in the positive reaction, and B and C are similar in the negative reaction. Response: We thank the reviewer for the careful reading of this manuscript and constructive comment. We see the ambiguity in our original text; what we meant to say is that both + and -have positive LAMP amplification, but + has a site cleavable by Cas9/gRNA whereas -does not. We have modified the text in Figure S2a to appropriately convey this information. We have also modified the second part of Figure  S2a as this reviewer has helpfully suggested to avoid further ambiguities.

Results
Line 141: What are "appropriate concentrations of template"? For Fig 1g-h and Fig 2g, the n is not given in the figure legend or elsewhere in the manuscript. Based on the number of points plotted on the graph, it looks as though triplicates were performed. This is not a high enough sample size to determine the limit of detection when using 0, 10 and 100 copies / reaction. Response: We thank the reviewer for this comment. In this revision, we have clarified "appropriate concentrations of template" in the text as "input concentrations above 1 copy per microliter". Furthermore, we repeated the experiment using an optimized protocol with larger reaction volumes, which resulted in improved limits of detections. We also used 6 replicates for each sample, ensuring a larger sample size to better determine limit of detection.  Fig. 1I as the SENTINEL score yet in Fig. S3f and S3g the data is given as a percent reduction in fluorescence? Response: We thank the reviewer for this detailed comment. We agree that the previous version of the figures is very difficult to interpret -we believe we have greatly improved the clarity of our figures in this revision. The many red and grey line plots have been combined with improved organization; the new version of these plots can be found in the new Figure 3 (shown below).
Response: SENTINEL may also be compatible with RPA. Empirically, we found LAMP to produce the most robust and specific pre-amplification. Indeed, this may be a reason why related methods such as SHERLOCK and DETECTR also utilize LAMP instead of RPA for SARS-CoV-2 detection. LAMP was also more convenient to obtain, having multiple commercial sources locally in the US (Thermo Fisher, New England BioLabs) compared to RPA, which is only available from a company in the UK. We are currently working on follow-up directions that aim to combine SENTINEL with other isothermal amplification strategies such as RPA and helicase-dependent amplification in a one-pot reaction.
3. In the manuscript, readouts of detection have to be calculated by equation ( We thank the reviewer for this detailed comment and great suggestions. While three measurements are used in the formula, only 2 reactions are needed per sample, which is more practical.
As the reviewer notes, Reaction C is a negative control, and the same value can be used across the entire set of experiments. However, Reaction B is not a true negative control, with the value of B/C reporting on the extent of the isothermal amplification step. Only using Reaction A would lead to difficulty distinguishing between appropriate amplification followed by degradation, from lack of amplification. Looking at reaction A slope is very creative and interesting, but based on our understanding, would require repeated measurements of DNA concentration.
4. The authors stated that "Furthermore, we found that this strategy could reliably detect down to approximately 100 ssRNA or dsDNA molecules per microliter of input sample, which is in the attomolar concentration range. The sensitivity of this assay is bounded by the sensitivity of the LAMP step, which has been shown to be comparable to gold standard quantitative PCR (qPCR)". When using pure nucleic acids fragments as template, LoD of LAMP reaction can easily reach 10 copies/reaction. However, when viral particles were used as templates, LoD of 10 particles per microliter was observed. The authors should explain possible reasons for this phenomenon, and provide representative photographs of LAMP results in SI.