NISDA assay mechanism
As shown in Fig. 1, the key components of the assay include an Initiator and two probes (M1 and M2). The Initiator is a dsDNA composed of an INA strand26 with a toehold sequence which is partially complementary to the Template DNA, and completely complementary to the viral RNA. The probe M1 is labeled with 6-carboxyfluorescein (FAM) and bhq_1 quencher at the end of the stem. In the absence of the viral RNA, the spontaneous interactions between the Initiator and probes are kinetically blocked, and there is no fluorescence signal as the fluorophore and quencher are in close vicinity.
The workflow of the NISDA assay is as following. First, the total RNA is extracted from oropharyngeal swab samples collected from patients (Clinical isolation, Hvidovre Hospital, Denmark), then added to the reaction mixture following by incubation at 42 °C for 30 min and fluorescence measurement. When the target viral RNA is present (Fig. 1), it is recognized by the Initiator through the toehold sequence of the INA strand. The INA then starts to hybridize with the viral RNA, leading to the Template displacement. This step, termed “displacement”, is the process of exchanging a long viral RNA to a short DNA template. This step is critical, as the subsequent signal amplification is much more efficient with the short template.
Driven by the entropy, the released Template then hybridizes with the molecular beacon structure of probe M1 (Fig. 1). The hairpin structure of probe M1 is opened, and the bʹ sequence is accessible to function as a toehold-binding site for the subsequent hybridization with the b sequence of probe M2. Due to the strand-displacement process, probe M2 displaces the initial Template by hybridization with probe M1 to form a more stable duplex structure. Afterward, the displaced Template acts as a fuel to open another probe M1 hairpin and initiate a new hybridization cycle. The process is termed “amplification”. Unfolding the probe M1 results in a distinct change in the fluorescence intensity arising from FAM, as the fluorophore and quencher are brought further apart. After 30 min of reaction at 42 °C, more M1:M2 pairs are formed, and the fluorescence intensity of the solution increases significantly. Based on the intensity change, the positive and negative test results can be easily distinguished by an optical reader.
Design of the molecular structures
We selected two specific sequences of RNA-dependent RNA polymerase (RdRP) and nucleoprotein (N) genes, for the selective detection of SARS-CoV-2 with minimal cross-target affinity to other types of human coronaviruses (Supplementary Fig. 1). The sequences of the Initiator, and probe M1 and M2 are listed in (Supplementary Table 1). As mentioned above, the NISDA mechanism comprises two parts: “displacement” and “amplification”. At first, we had to make sure that the amplification part on a short Template sequence would work properly. Optimum 22 nt DNA sequence identical to the RdRP and N genes were chosen for the Template (Fig. 2 and Supplementary Table 1) and M1 and M2 were subsequently designed. The rational design of the M1 and M2 was a prerequisite to minimize their interaction in the absence of the Template. In our first design, the length of the Template was chosen to be 30 nt. As shown in the polyacrylamide gel electrophoresis (PAGE) analysis for N gene probes (Fig. 2a), given the lane V, it can be clearly seen that M1 and M2 did not interact/hybridize with each other before the Template addition. Lane IV shows that the M1 was easily opened and hybridized with Template, while the Template could not open the M2 in the absence of M1 (lane VI). However, after the addition of the Template in the mixture of M1 and M2, a new band was observed and no band was seen for the recycled Template (lane VII). We supposed that, the released Template from Template:M1 duplex might be hybridized with M2 to form Template:M1:M2 triplex, hence hampering the cascade reaction. Therefore, we speculated that reducing the length of the Template from its 3′-end (segment c′) would decrease its binding affinity to the leftover sequence of M2 at its 5′-end (segment c). Interestingly, decreasing the Template length to 28 nt and 26 nt resulted in the appearance of a new band under the Template:M1:M2 triplex (Supplementary Fig. 2a, b). However, still no band was observed for the recycled Template. A shorter Template of 24 nt (Supplementary Fig. 2c) resulted in the disappearance of the Template:M1:M2 triplex and having a more clear band at the lower position. Although, the efficiency was still low due to the existence of the Template:M1 band, illustrating that considerable amounts of the Templates are still engaged with M1 strands and cannot contribute in the cascade reaction. Decreasing the Template length to 22 nt led to a very proper result (Fig. 2b). Almost all of the M1 strands were consumed to successfully form M1:M2 duplex and the Template was effectively recycled. In order to ensure that the observed band at lane VII was exactly attributed to the M1:M2 duplex, we annealed M1 and M2 in the absence of Template and compared the result with the mixture of M1, M2, and Template. According to the gel data in Fig. 2b (right panel), the M1:M2 duplex was perfectly formed upon the Template addition. From the obtained data, the robustness of our design was evidential due to the fact that almost all of the M1 primers were consumed after the addition of Template (lane VIII), while there were still considerable amount of un-hybridized M1 after annealing with M2 (lane IX). Similar performance was observed for the RdRP gene probes (Supplementary Fig. 3).
PAGE (12% TBM) analysis of a, amplification and b, displacement parts of the NISDA assay for N gene. a, b Investigating the role of Template length on the occurrence of the cascade reaction (amplification); decreasing the Template length from 30 to 22 nt resulted in disappearing of the unwanted bands and emergence of a new bands corresponding to the target M1:M2 duplex and recycling of the Template. c Typical gel image of the Initiator (T/INAt_15) before and after the PAGE purification together with the PAGE analysis on the displacement of Template by mimic SARS-CoV-2 (85 nt) strand. T denotes for Template. All the molecular weight markers are 100 bp. Each experiment was repeated at least two times independently.
For the displacement part, it was necessary to have a balanced design on the Initiator, composed of the Template and a DNA strand with enhanced affinity (INA). The critical issue was to design the INA to have (i) high binding affinity to efficiently and specifically bind to the whole RNA genome, and (ii) sufficient stability to facilitate the strand displacement. Therefore, the best condition was to increase the affinity only at the toehold sequence of the complementary DNA. Although, one can use any other types of high-affinity nucleic acids such as peptide nucleic acids (PNA) and locked nucleic acid (LNA) as they have been widely used for performance enhancement of DNA biosensors27,28,29. The complementary DNA harboring different toehold lengths (8 and 15 nt) with and without the INA were synthesized by PentaBase A/S. PentaBase’s proprietary INA® technology is on the basis of insertion of a base unit (intercalator) that is intercalated into nucleobases without disrupting or substituting any nucleotide in the nucleic acid sequence. The integrated INA® increases the stacking of the duplex helix, which enhances the specificity and affinity of the oligonucleotide. Although, the numbers, types, and positions of the intercalators need to be rationally designed such that a minimum degree of flexibility and a maximum duplex stability is achieved. According to the reported article26, the best conformation would be having three or four nucleotides distance between each two intercalators. Therefore, we decided to have the 8 nt and the 15 nt toehold sequences modified with two and four intercalators, respectively. However, the type and the positions of intercalators were calculated and optimized by PentaBase A/S to minimize secondary structures and achieve higher duplex stabilities. As clearly seen in Supplementary Fig. 4, compared to the natural DNA strands, the corresponding INAt_8 (harboring high-affinity 8 nt toehold sequence; t_8 indicates the toehold length) and INAt-15 showed higher duplex stabilizations as determined by the raise in the melting temperatures (ΔTm (on average) = + 3.2 °C for INAt_8, and + 8.7 °C for INAt-15).
To illustrate the displacement part using PAGE analysis, we used a mimic SARS-CoV-2 DNA sequence (85 nt) and mixed it with different purified Initiators. Figure 2c (right panel), demonstrates a typical PAGE analysis on the Template (T) displacement by mimic SARS-CoV-2 using Initiator (T/INAt_15). From Fig. 2c (left and middle panels), it is clear that the T/INAt_15 was well purified, as there was almost no visible band corresponding to the Template or INAt_15 in the middle panel as well as the lane III of the right panel. However, incubation of T/INAt_15 with the mimic SARS-CoV-2 (85 nt) resulted in the emergence of a higher band in the gel (lane V), demonstrating the formation of mimic SARS-CoV-2 (85 nt):INAt_15 following by the successful displacement of the Template. Although since the mimic SARS-CoV-2 (85 nt) was a DNA strand and not that long to undergo complicated self-folding/loop structures, the other Initiators could also perform the Template displacement (Supplementary Fig. 5). Further analysis was also carried out later in this study to ensure the Initiators’ efficiency on the Template displacement by the whole SARS-CoV-2 RNA genome.
Development of NISDA assay
Prior to the whole-genome detection, we tested the assay on serial dilutions of the Template. The reaction mixture was prepared composed of probe M1 labeled with FAM/bhq-1 and unmodified probe M2. The reaction temperature was fixed at the optimum 42 ± 1 °C and the fluorescence signal arising from FAM was recorded over time. It should be noted that the assay performance at lower reaction temperature was poor, possibly due to the hybridization of the recycled Template with M1:M2 duplex at the c segment to hamper the successive amplification cycles. As depicted in Fig. 3a, two sets of probes were designed, attributed to the RdRP gene (14,083–14,119 nt) and N gene (28,694–28,730 nt). As seen in Fig. 3b and c, almost similar assay responses were observed for both probe sets. After the addition of the Template into the reaction mixture (M1 and M2), in <20 min the fluorescence signals reached their maximum values demonstrating that all of the M1 probes are in the form of M1:M2 duplex. An ultrafast response of the assay can be clearly seen even at the very low concentration of the target Template (300 aM). The higher concentration of the Template resulted in the faster fluorescence signal saturation, as the assay could detect 300 nM in 10–13 min and 3 fM in 17–20 min for both genes. Therefore, for the quantitative measurement, the slope (Ɵ) of the fluorescence signal enhancement over the saturation time was assigned for the detection signal. A very wide linear response of 300 nM to 3 fM was observed from both assays designed for RdRP and N genes with the LoD of 133 aM and 181 aM, respectively (Fig. 3b, c).
a Genome map of the SARS-CoV-2 isolate Wuhan-Hu-1, depicting two studied regions of RdRP and N genes and their corresponding nucleotide positions (red bars). Typical curves illustrating the kinetics of the fluorescence signal of the assay, designed for b, RdRP gene and c, N gene, over time for serial dilutions of the Template from 300 nM down to 300 aM in TES buffer (pH 7.8). The linear regression plots from 300 nM down to 3 fM are shown for each corresponding curves. Three independent experiments were run (n = 3). The assay comprising M1 and M2 probes was launched at the constant temperature of 42 °C while the fluorescence intensity of 6-FAM was recording each 3.5 min. RFU and Scr denote for relative fluorescence unit and scrambled DNA sequence, respectively. The Slope values (Ɵ) were calculated based on the value of saturated fluorescence intensity recorded for each graph (indicated as filled circles) dividing by the corresponding time intervals. Error bars represent mean ± SD. The solid boxes represent mean values for buffer ± SD. Source data for (b, c) are provided as a source data file.
Function of Initiator in the Template displacement
Before the final development of the NISDA assay for the whole genome, exchanging the RdRP and N genes of synthetic SARS-CoV-2 RNA (TWIST Bioscience) into the corresponding Templates were evaluated using different types of Initiators (denoted as T/nucleic acidt-x duplex; T stands for Template, x stands for the length of toehold sequence). Reaction mixtures composed of M1, M2, and three types of Initiators (T/DNAt-15, T/INAt-8, and T/INAt-15) were prepared and incubated with SARS-CoV-2 RNA (106 copies per µL) while the FAM fluorescence was recorded over time at 42 °C. As illustrated in Fig. 4, for both RdRP and N genes, upon the SARS-CoV-2 addition only the reaction mixtures containing T/INAt-15 showed fluorescence signal enhancement, whereas the other Initiators (T/DNAt-15 and T/INAt-8) did not show any significant signals, meaning that the T/INAt-15 could capably function to exchange the whole genome to the corresponding Template and further signal amplification. This might be due to the fact that the longer toehold (15 nt) sequence provides more binding sites and higher stability hence greater specificity and binding efficiency to the target sequence at 42 °C. Although, INAs with longer toeholds (>15 nt) did not show good performances, possibly because of self-folding and dimer formations. Moreover, given the inefficiency of the unmodified T/DNAt-15, increasing the affinity of the toehold sequence using INA technology was necessary for the assay to successfully undergo the strand displacement. This might be due to the unpredicted 3D structure and inherit self-folding (stem-loop formation) of the whole RNA genome to hamper a proper hybridization of DNA to its corresponding binding site. More importantly, compared to the RdRP gene, the NISDA assay on the N gene showed faster (~30 min) and more sensitive performance (dramatic signal enhancement).
a Location of the toehold-binding sites for both RdRP and N genes. Typical curves illustrating the fluorescence signal kinetics of the NISDA assay over time, designed for b, RdRP gene and c, N gene using different Initiators of T/DNAt-15 (solid blue circle), T/INAt-8 (hollow red circle), and T/INAt-15 (hollow green square). Water (hollow black triangle) means no Initiator, only M1 and M2. Source data for (b, c) are provided as a source data file.
To address why the NISDA assay functioned more efficiently for the N gene rather than the RdRP gene, we carried out an RNA structure analysis at 42 °C using RNAfold30. By comparing the folded-binding sites, we observed that the RdRP binding site folded into a single hairpin with about twice as strong folding energy as the N-binding site which folded into two smaller stems (Supplementary Discussion 1 and Supplementary Fig. 6a). For the folding of the binding sites with an additional 30 nt and 60 nt up- and downstream, the RdRP again folded stronger than N (Supplementary Discussion 1). Moreover for RdRP gene, the whole binding site was in a structured region, whereas for the N gene, the toehold part of the binding site was in a high probable unpaired region (Supplementary Fig. 6b, c). We furthermore observed that RdRP did not have alternative foldings in contrast to N (Supplementary Fig. 7). These observations prompted us to explore the joint binding and folding patterns of the reverse complement of the binding site alone and the binding sites with context. For this purpose, we employed IntaRNA31 at 42 °C on the RdRP and N gene pairs, respectively. We found that both the hybridization and the energy of the duplex indeed was lower for the N gene than the RdRP (Supplementary Fig. 8). These observations are consistent with the N probe functioning better than the RdRP probe. The obtained results demonstrate that it is very important to select a target sequence within the whole genome to be not only specific but also accessible at 42 °C. Therefore, for our further experiments, we used probes designed for N gene to perform the assay evaluation and clinical validation.
Sensitivity and selectivity of the NISDA assay
The NISDA assay was developed for the N gene and its performance was further studied. The sensitivity analysis was carried out using serial dilutions of synthetic SARS-CoV-2 RNA (TWIST Bioscience) in Nuclease-free water and spiked in known negative oropharyngeal RNA extracts (provided by clinical microbiology department, Hvidovre Hospital, Denmark). Figure 5a and b illustrates the NISDA assay response to different concentrations of SARS-CoV-2 RNA diluted in water and negative matrix, respectively. The ultrafast response of the assay is evidential as the fluorescence intensity could dramatically reach to its maximum value within the first minutes of the reaction. Considering the plots of fluorescence enhancement slope as the function of RNA concentration, the semi-quantitative behavior of the assay can be comprehended. Although, sudden enhancement of the fluorescence signals resulting from rapid amplification makes it difficult to accurately define the slopes thereby calculating the LoD. As a result, in order to calculate the sensitivity of our assay, we performed a quantification study based on the endpoint fluorescence readings (tables in Fig. 5a, b). The cutoff value was defined as 95% of confidence interval (i.e., the mean value of the signals obtained from the negative samples plus two times of their standard deviation (mean + 2 SD)12, and the lowest concentration of the SARS-CoV-2 RNA that showed a greater signal than the cutoff value was defined as the LoD. Therefore, the LoD of NISDA assay was five and ten copies µL−1 of SARS-CoV-2 RNA diluted in water and negative matrix, respectively (one copy µL−1 corresponds to five copies per reaction). In addition, the developed NISDA assay did not show any false-negative result. Moreover, no cross-reactivity was observed for the NISDA assay (prepared for SARS-CoV-2) tested on other various respiratory viral nucleic acids of human coronaviruses 229E (H CoV 229E), influenza A virus subtype H1N1 and H3N2, Boca virus 1, H enterovirus 68, H rhinovirus 89, and Mumps (Fig. 5c, d).
a Assay response to different dilutions of synthetic SARS-CoV-2 RNA from 100 copies µL−1 to 5 copies µL−1 (blue line: 100; yellow line: 50; red line: 25 and black line: 5) in nucleases free water. b Assay response to different dilutions of synthetic SARS-CoV-2 RNA from 100 copies µL−1 to 10 copies µL−1 (blue line: 100; yellow line: 50; red line: 25 and black line: 10) in oropharyngeal RNA extract, together with their corresponding quantification graphs and dataset based on the calculated signal enhancement slope (Ɵ) and fluorescence readout at 30 min of reaction, respectively. FL, CI, N, and SD denote for fluorescence, confidence interval, negative control, and standard deviation, respectively. c Typical fluorescence kinetics recorded over time for SARS-CoV-2 RNA (10 copies µL−1) and different respiratory viral nucleic acids (500 copies µL−1) spiked in validated CoV-19-negative oropharyngeal RNA extract matrix (Hvidovre Hospital). d Corresponding plot of panel c demonstrating the recorded fluorescence intensities after 30 min of reaction for the studied respiratory viral nucleic acids. Three independent experiments were run (n = 3) and the P values were calculated based on the unpaired two-tailed t test (p < 0.01 for SARS-CoV-2 vs. non-targets; P < 0.001 for SARS-CoV-2 vs. negative matrix). Error bars represent mean value (center line) ± SD. copies µL−1 is the numbers of RNA strands per µL (one copy µL−1 corresponds to five RNA copies per reaction). Source data are provided as a source data file.
Clinical validation of the NISDA assay
We performed the clinical validation of the NISDA assay using N gene of the targeted sequence for the detection of SARS-CoV-2 RNA from oropharyngeal RNA extract specimens. A total of 127 clinical samples (62 verified CoV-19 positive with diverse Ct values and 65 verified CoV-19 negative) were tested. A qualitative test was performed based on the endpoint fluorescence signal. The extracted RNA samples provided by Hvidovre Hospital were stored at 80 °C before the assay operation. The cutoff value obtained by performing the assay on all of the negative samples (Fig. 6) was 76.01 RFU which was far below the 95% of the confidence interval of the positive samples (161.1 RFU). As shown in Fig. 6a, the NISDA assay was able to successfully identify all the 65 negative samples with no false-positive rate and 60 positives out of 62 CoV-19 positive samples (3.2 % false-negative rate) in only 30 min (Supplementary Fig. 9). The false negatives might be due to the degradation of the extracted RNA samples in the laboratory, even though, a false-negative occurrence at Ct > 37 does not rise a major concern, as it is a common issue for the qRT-PCR32. An alternative strategy would be to design different probe sets for the E gene and run the experiments simultaneously to avoid probable false negatives in a back-to-back supporting approach.
a Plotting the NISDA assay data points obtained in the laboratory for 127 clinical oropharyngeal RNA extracts (65 validated CoV-19 negatives and 62 validated CoV-19 positives) together with the corresponding Ct values of the positive samples; blue solid circles indicate samples considered positive by NISDA and red hollow circles indicate samples considered negative by NISDA; mean value difference between the two populations was statistically significant (P < 0.001; calculated based on the unpaired two-tailed t test). b Reproducibility test of the NISDA assay showing the fluorescence kinetics of three identical measurements on a CoV-19-positive sample (Ct = 30). CI denotes for the confidence interval. Error bars represent mean ± SD. c On-site performance of the NISDA assay at Bispebjerg Hospital using Real-Time PCR System (CFX96) for monitoring the fluorescence kinetics of 37 clinical oropharyngeal RNA extracts (31 validated CoV-19 negatives (gray solid lines under the cutoff value) and 6 validated CoV-19 positives (blue solid lines above the cutoff value)). d Numbers of false-negative and positive rates for two clinical validations in the laboratory and the hospital (on-site). Source data are provided as a source data file.
In addition, the reproducibility of the NISDA assay on N gene was tested by conducting three identical experiments on one validated CoV-19 positive (Ct = 30) and one validated CoV-19-negative oropharyngeal RNA extract sample. A sound reproducibility can be interpreted from Fig. 6b. The developed NISDA assay also showed great robustness, as it could retain 99.2% and 98.9% of its initial fluorescence signals after being stored in a dark humid chamber for a month at RT and for 2 months at 4 °C, respectively. We also conducted a similar experiment on the RdRP gene, and the assay showed a very poor sensitivity of 61.2%, relatively long response time (60 min), and the recorded data points for the true positives were close to the cutoff value (Supplementary Fig. 10). This, as mentioned before, might be due to the difficulty for the Initiator to have proper access to its target sequence to successfully undergo the Template displacement (Supplementary Discussion 1).
In order to better evaluate the NISDA assay and compare its performance with qRT-PCR, we performed another clinical study at Bispebjerg Hospital, Denmark. For the temperature control (42 °C) and fluorescence readout, we used an identical BioRad Real-Time PCR System (CFX96) which was being used for daily CoV-19 tests at the Bispebjerg Hospital (CoV-19 Diagnostics Division, Department of Clinical Biochemistry). The NISDA assay was performed on 37 randomized oropharyngeal RNA extract specimens without knowing their qRT-PCR results. The NISDA assay represented 100% sensitivity and 100% specificity with no false-positive or false-negative rates (Fig. 6c, d), signifying that the RNA degradation might be the issue that we encountered during the laboratory testing. Although, a relatively high value for the confidence bands (Table 1) would be attributed to the low number of population.
Comparing with qRT-PCR, it was found that NISDA assay represented 100% sensitivity and 100% specificity for the detection of SARS-CoV-2 N gene when the target was spiked in negative matrix within the detection range (Table 1). Compared with other SARS-CoV-2-detection methods33, NISDA assay showed a considerably high value of diagnostic odds ratios (DORs) for the detection of SARS-CoV-2 (6.61–8.75).
In order to further investigate the advantages of our approach against qRT-PCR and similar isothermal techniques (RT-LAMP and RPA), we performed side-by-side experiments. As discussed in Supplementary Discussion 2 and 3 and illustrated in Supplementary Figs. 11–13, compared to qRT-PCR, NISDA was more user-friendly, faster and required simpler data analysis. Moreover, NISDA was more sensitive than RPA and compared to RT-LAMP, NISDA did not rely on a specific sample preparation method and was more robust at RT. A comparison analysis is also provided in Table 2 remarking the advantages of NISDA assay against similar molecular diagnostic methods, being rapid, nonlaborious (one-pot detection), robust, and cost-effective (nonenzymatic).
In addition, we carried out a set of experiments to evaluate whether the NISDA assay is capable of performing directly on the saliva specimens. We prepared some mimic samples by spiking synthetic SARS-CoV-2 RNA (100 copies µL−1) into healthy saliva. As seen in Supplementary Fig. 14 (Protocols 1 and 2), the assay showed high background signals and did not show any response when performing on spike-in samples, even on the lysed samples. That might be due to the high density of saliva matrix to either promote nonspecific bindings or hamper proper fluorescence light transmissions through the media. Therefore, we added one more step of saliva centrifugation to obtain a cell-free matrix (Supplementary Fig. 14, Protocol 3). This time the NISDA assay showed a response, however, only two of the five spike-in samples were detected positive. The observed performance demonstrates the promising potential of NISDA assay to become more convenient and faster, nonetheless, comprehensive optimization experiments are required to increase its performance.
