Transforming RNA Degradation into Accurate RNA Detection: RNase I–Assisted Rolling Circle Amplification for Clinical Samples

Abstract

Ribonucleases are powerful degradative enzymes that rapidly cleave RNA and have long been regarded as obstacles to RNA diagnostics. Detecting large, structured RNA molecules in biological samples is further complicated by extensive secondary structures and abundant background RNA. Here we transform this destructive activity into a diagnostic advantage through a RNase I–assisted rolling circle amplification (RI‐RCA) strategy that integrates enzymatic RNA digestion with circular DNA templates (CDTs) to achieve direct, reverse‐transcription‐free RNA detection. In this system, Escherichia coli RNase I—a widely available endoribonuclease—selectively digests structured RNA to expose complementary CDT‐binding sites, which are immediately protected through duplex formation and converted into primers for RCA. The resulting reaction proceeds isothermally at room temperature in linear, quasi‐exponential, and exponential modes, achieving quantitative signal generation across concentrations from 50 aM to 50 fM and maintaining performance in 50% pooled human saliva. Applied to clinical saliva samples from SARS‐CoV‐2–positive and –negative individuals, RI‐RCA achieved near‐perfect sensitivity and 98–100% diagnostic accuracy. By coupling enzymatic RNA digestion with sequence‐specific protection via CDT hybridization, this strategy converts RNA instability from a liability into an analytical advantage, providing a simple, robust, and clinically translatable platform for accurate detection of structured RNA targets.

Here, we transform this otherwise destructive enzymatic activity into a powerful diagnostic advantage through an RNase I–assisted rolling circle amplification (RI‐RCA) strategy. By integrating controlled RNase I–mediated RNA digestion with circular DNA templates, this approach enables direct and highly sensitive detection of target RNA sequences. Using this strategy, we successfully identified COVID‐positive saliva samples by accurately detecting the presence of SARS‐CoV‐2 genomic RNA in patient saliva.

Introduction

Ribonucleic acids (RNA) play essential roles in gene regulation, catalysis, and pathogen biology,^{[ 1 ]} making them attractive molecular targets for diagnosing a wide spectrum of diseases, from cancer^{[ 2} , ³ , ⁴ , ⁵ , ⁶ , ^{7 ]} to viral infections such as HIV,^{[ 8 ]} Zika,^{[ 9 ]} avian influenza^{[ 10 ]} and SARS‐CoV‐2.^{[ 11} , ¹² , ¹³ , ^{14 ]} However, detecting RNA in its native form remains intrinsically challenging. RNA molecules are chemically fragile and often fold into stable secondary and tertiary structures that restrict probe accessibility. In biological samples, these structured RNAs coexist with abundant background RNA and are continually exposed to potent ribonucleases that degrade them rapidly. As a result, achieving accurate detection of large, structured RNA (lsRNA) species in clinical samples is notoriously difficult.

The dominant strategy for RNA detection couples reverse transcription (RT) with nucleic acid amplification. Reverse transcription‐polymerase chain reaction (RT‐PCR)^{[ 15 ]} remains the clinical gold standard for RNA detection because it provides high sensitivity, quantitative accuracy, and scalability.^{[ 16} , ^{17 ]} Yet, the approach is constrained by the need for multiple enzymatic steps, precise thermal cycling, and stringent contamination control, making it time‐consuming and less suited to point‐of‐care testing. Furthermore, incomplete reverse transcription or sequence cross‐reactivity can lead to false positives or negatives,^{[ 18 ]} reducing diagnostic reliability.

Several isothermal amplification methods—loop‐mediated isothermal amplification (LAMP),^{[ 19} , ²⁰ , ^{21 ]} signal‐mediated amplification of RNA technology (SMART),^{[ 22 ]} strand displacement amplification (SDA),^{[ 23} , ²⁴ , ²⁵ , ^{26 ]} recombinase polymerase amplification (RPA)^{[ 27} , ²⁸ , ^{29 ]} and hybridization chain reaction (HCR)^{[ 30} , ³¹ , ^{32 ]}—circumvent thermal cycling but often demand elaborate primer designs and pre‐amplification steps, resulting in trade‐offs between speed, sensitivity, and simplicity.^{[ 33 ]} Their dependence on reverse transcription or complex sequence engineering also limits broad adaptability and increases assay costs.

To bypass the RT step altogether, we previously developed a DNAzyme‐assisted approach for direct RNA detection. A 10–23 DNAzyme cleaved selected regions of lsRNA to generate a 3′ terminus capable of hybridizing with a circular DNA template (CDT), which in turn initiated rolling circle amplification (RCA) at room temperature.^{[ 34} , ³⁵ , ³⁶ , ^{37 ]} RCA is an isothermal amplification process that produces long DNA concatemers detectable through colorimetric or fluorometric readouts.^{[ 38} , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ^{47 ]} While this DNAzyme‐RCA platform demonstrated the feasibility of RT‐free RNA detection, its practical implementation was hindered by RNA inaccessibility: among 230 DNAzymes screened against the SARS‐CoV‐2 genome, only eight produced primers that initiated RCA, and only one achieved clinically usable performance.

To improve RNA accessibility, we explored antisense oligonucleotides (ASOs)^{[ 48 ]} designed to locally destabilize RNA structures and facilitate hybridization by DNAzymes^{[ 49 ]} or CDTs.^{[ 50 ]} Although effective in some cases, this approach relied on computational RNA‐folding predictions and iterative screening, both of which are time‐consuming and only moderately reliable.

In this work, we introduce a fundamentally different concept: repurposing the degradative activity of ribonucleases to enable RNA detection. We reasoned that controlled enzymatic digestion could fragment structured RNA molecules, transiently exposing previously inaccessible regions for recognition by a complementary CDT. Once hybridized, the CDT protects the RNA/DNA duplex from further cleavage, allowing the protected RNA region to serve as a primer for RCA. Among known ribonucleases, Escherichia coli RNase I is particularly suitable—it is inexpensive, cofactor‐independent, and highly efficient at cleaving accessible RNA strands.^{[ 51} , ^{52 ]} Importantly, we hypothesized that RNA engaged in a stable RNA/DNA duplex should be significantly less susceptible to RNase I cleavage than its single‐stranded counterpart, as RNase I and related enzymes evolved to recognize RNA‐only substrates rather than hybrids.

Our strategy, illustrated schematically in Figure 1, employs RNase I to degrade the bulk RNA population while selectively preserving CDT‐bound regions. The resulting protected RNA fragments serve directly as RCA primers, enabling amplification at room temperature without the need for reverse transcription or PCR. This one‐pot process—termed RNase I–assisted rolling circle amplification (RI‐RCA)—thus transforms RNA degradation from a destructive process into a constructive mechanism for selective signal generation.

Figure 1.

Schematic illustration of the RNAse I‐assisted RCA (RI‐RCA) mechanism. a) RNase I partially digests structured RNA, transiently exposing the CDT‐binding region. b) Hybridization with the circular DNA template (CDT) protects the target sequence from further degradation and allows the protected RNA fragment to serve as a primer for rolling circle amplification (RCA). The resulting RI‐RCA process generates amplified DNA products that can be visualized colorimetrically or fluorometrically.

We demonstrate RI‐RCA using four CDTs targeting distinct regions of the SARS‐CoV‐2 genome. The method performs robustly in linear, quasi‐exponential (QE), and exponential (E) amplification modes, detecting both short and long RNA transcripts over concentrations from 50 aM to 50 fM. Quasi‐exponential RCA (QE‐RCA) is an amplification mode in which linear RCA is first initiated by the RNA fragment, followed by hybridization of a secondary primer (P2) to the RCA product to trigger a second amplification step to enhance signal output, but without a tertiary primer or restriction enzyme to initiate a fully exponential feedback mechanism.^{[ 34 ]} Remarkably, RI‐RCA maintains performance in 50% pooled human saliva and achieves near‐perfect sensitivity and 98–100% diagnostic accuracy in clinical saliva samples.

By coupling enzymatic degradation with molecular self‐protection, RI‐RCA converts RNA instability from a liability into an analytical advantage. This conceptually distinct yet practically accessible strategy establishes a foundation for next‐generation RNA diagnostics that operate under ambient conditions and function effectively in complex biological samples.

Results and Discussion

Selecting RI‐RCA systems

To experimentally validate the RI‐RCA concept, we first designed a series of model systems based on genomic RNA fragments from the SARS‐CoV‐2 genome. Both short and long RNA transcripts were chosen to evaluate the generality of the method across targets of varying size and structural complexity. As an initial model, we used the N2 gene transcript (ls101), a CDC‐approved RT‐PCR marker,^{[ 53 ]} which contains stable secondary structures that limit hybridization to a complementary CDT, and therefore requires RNase‐mediated processing to generate a primer suitable for RCA. This system is designated as System 1.

To examine scalability, larger transcripts corresponding to the NSP8, NSP13, and NSP14 genes were also selected—designated ls584, ls1805, and ls1583, respectively (Table S1)—and are referred to as Systems 2–4. The transcript nomenclature follows a consistent convention in which “ls” denotes large, structured RNA derived from the SARS‐CoV‐2 genome, and the numeric suffix indicates transcript length in nucleotides.

For System 1, three CDTs—CDT1a, CDT1b, and CDT1c—were designed to protect the anti‐CDT region of ls101 through RNA/DNA duplex formation of 20, 30, or 40 base pairs (bp), respectively. The minimum‐free‐energy secondary structures and base‐pairing probabilities of all lsRNA targets were predicted using the Mathews Lab RNAstructure web server. Detailed experimental results for System 1 are presented in the main text, whereas data for Systems 2–4 are provided in the Supporting Information.

Figure 2 illustrates the predicted hybridization between the folded ls101 RNA and each CDT, with the anti‐CDT1 region highlighted in red. In the presence of RNase I, the enzyme progressively digests accessible regions of the structured RNA, leading to partial disruption of nearby base‐pairing interactions and transient exposure of the anti‐CDT1 site. This newly accessible segment allows the CDT (binding region in green) to compete effectively with intramolecular pairing and hybridize to its complementary RNA sequence. Among the designs, CDT1c, which forms the longest duplex (40 bp), shows the highest predicted binding affinity and the greatest ability to invade secondary structure and protect the target region. Binding affinity is defined here by RNA secondary structure–derived hybridization free energy (ΔG) calculations and not by experimentally measured binding constants (K_d). Continued RNase I digestion eliminates surrounding unbound RNA, leaving a protected RNA fragment 20–40 nucleotides in length that subsequently serves as the primer for RCA. This modular design principle—combining partial enzymatic digestion with sequence‐specific protection—was extended to all four systems, enabling systematic evaluation of RI‐RCA performance across different RNA lengths and structural contexts (Figures S1–S3).

Figure 2.

Predicted hybridization between ls101 RNA and CDT1 variants. Predicted secondary structures of RNA1 (ls101) and its hybridization with CDTs 1a–1c. The anti‐CDT1 region is highlighted in red, and the corresponding CDT‐binding regions are shown in green. RNase I digestion destabilizes local RNA structures surrounding this region, allowing CDT binding and protection of the complementary sequence. Among the designs, CDT1c (40 bp duplex) exhibits the highest predicted binding affinity and accessibility. Secondary structures of RNA were predicted using the Mathews Lab RNA structure web server.

RNase I Impact on Binding Between CDT and RNA

As illustrated in Figure 1, the RI‐RCA mechanism relies on two interdependent processes: RNase I partially digests structured RNA molecules, transiently exposing regions that were previously inaccessible, and CDTs hybridize to their complementary sequences within these exposed regions, forming RNA/DNA duplexes that resist further enzymatic degradation. To validate this protection mechanism, CDT–RNA interactions were examined using native gel electrophoresis (nPAGE).

Truncated RNA constructs containing the CDT‐binding regions were synthesized for each system to enable clear visualization of mobility shifts. For System 1, a 75‐nt RNA fragment (RNA1) encompassing the CDT1c‐binding site was used as a representative model, while analogous fragments were prepared for Systems 2–4. Shorter constructs were necessary because mobility differences between full‐length RNA (e.g., 584 nt for System 2) and its CDT complex (676 nt) are too small to resolve on nPAGE, whereas shorter fragments (e.g., 75 nt versus 167 nt) yield distinct shifts while preserving the relevant local structures governing binding.

As shown in Figure 3, lanes 1 and 2 contain CDT1c and RNA1 alone, respectively. RNase I treatment of CDT1c alone (lane 3) confirmed that the CDT is stable. In contrast, RNase I digestion of unprotected RNA1 (lane 4) resulted in complete degradation. When CDT1c and RNA1 were incubated without RNase I (lane 5), no band shift was observed, indicating that the secondary structure of RNA1 prevented CDT access. Upon addition of RNase I (lane 6), a distinct shifted band appeared, accompanied by a decrease in the free CDT signal, signifying formation of a CDT1c–RNA1 duplex.

Figure 3.

Validation of CDT protection against RNase I digestion by native PAGE. Native gel electrophoresis of a truncated 75 nt RNA fragment containing the CDT1c binding site, with or without CDT1c and RNase I treatment. Lanes 1–6 correspond to CDT1c only, RNA only, CDT1c + RNase I, RNA + RNase I, RNA + CDT1c (no RNase I), and RNA + CDT1c + RNase I. Complete RNA degradation is observed in the absence of CDT protection, whereas inclusion of CDT1c produces a distinct band shift corresponding to the RNA/CDT duplex, confirming protection from RNase I digestion.

To further examine the locality and extent of CDT‐mediated protection during RNase I digestion, denaturing PAGE analysis was performed using radiolabeled RNA1 in the presence of CDTs with different lengths of complementary regions (Figure S4). Two additional reactions were included to enable fragment size determination: a full RNA ladder generated by partial alkaline hydrolysis, representing all ribonucleotides (lane 1), and an RNase T1–generated ladder producing fragments terminating at guanosine residues (lane 2). These ladders were used to assign fragment sizes in the RNase I digestion patterns. In the absence of CDT, RNase I digestion resulted in extensive RNA degradation (lane 3). In contrast, the presence of CDTs produced discrete protected RNA fragments whose sizes depended on the length of the RNA–DNA duplex (lanes 4–6). Specifically, CDT1a (20‐nt complementary region) afforded only weak protection (lane 4), whereas CDT1b (30‐nt complementary region) and CDT1c (40‐nt complementary region) generated prominent protected fragments of approximately 30 nt and 40 nt, respectively (lanes 5 and 6). These observations indicate that RNase I activity is locally attenuated within sufficiently long RNA–DNA duplex regions while remaining active elsewhere along the transcript.

These results confirm the proposed mechanism: RNase I cleaves structured RNA stochastically, transiently exposing the CDT binding site, which is then captured and protected through RNA/DNA duplex formation. The protected RNA:DNA complex remains intact, while surrounding regions are fully digested. Similar protection patterns were observed for Systems 2–4 (Figures S5–S7), demonstrating that RNase I digestion and CDT hybridization act cooperatively to convert degradative processing into selective preservation of target RNA sequences that subsequently serve as primers for RCA.

RNase I Impact on RCA

We first optimized the concentration of RNase I required to degrade unstructured RNA and expose the anti‐CDT region for CDT binding and subsequent RCA. Following CDT addition, varying concentrations of RNase I (0.5–10 mU µL⁻¹) were introduced. Linear RCA was then initiated by adding dNTPs, polynucleotide kinase (PNK), Φ29 DNA polymerase, and the intercalating dye SYTO 9. RNase I was added 5 min after CDT to ensure that the CDT was available to bind the anti‐CDT region as soon as it became accessible through RNase‐mediated digestion.

Fluorescence arising from SYTO 9 binding to RCA products (RPs) was monitored over 60 min. Based on the emission intensity at 30 min (F₃₀), 0.5 mU µL⁻¹ RNase I produced the highest RCA output (Figure S8). Doubling this concentration sharply reduced the signal, and minimal or no products were observed at 5 and 10 mU µL⁻¹. These results suggest that 0.5 mU µL⁻¹ RNase I provides sufficient digestion to remove unstructured regions and expose the target site while preserving the CDT‐protected fragment. At higher concentrations, excessive RNase I likely competes with the CDT for binding or completely degrades the accessible segment before protection occurs, thereby inhibiting primer formation.

To determine the optimal CDT protection length, CDTs ranging from 20–60 nt were tested with 0.5 mU µL⁻¹ RNase I. A 40 nt protection length provided maximal RCA output (Figure S9), whereas increasing the duplex length to 50 or 60 nt yielded no additional benefit. Thus, all subsequent studies used 0.5 mU µL⁻¹ RNase I and a 40 nt CDT (CDTc).

Time‐dependent fluorescence was recorded for each system with and without RNase I. For System 1, four conditions were tested: (i) a short 40 nt transcript (positive control, PC), (ii) CDT alone (negative control, NC), (iii) ls101 RNA + CDT1c without RNase I, and (iv) ls101 RNA + CDT1c with RNase I in buffer (B). The sample without RNase I verified RNA integrity and confirmed RCA dependence on RNase I. Figure 4a shows that minimal fluorescence was obtained without RNase I, while inclusion of RNase I led to a marked increase in signal. Figure 4b quantifies this enhancement: in the absence of RNase I, RCA efficiency was only 1% of the PC but increased to 61% upon RNase I treatment. Comparable trends were observed for Systems 2–4 (Figure S10), with RCA efficiencies ranging from 59–99%. Lower efficiencies in Systems 1 and 2 may reflect persistent local structures that resist digestion or weaker CDT–RNA interactions. Overall, RNase I consistently enhanced RCA efficiency across all systems, validating its role in removing structural barriers and generating effective primers for RCA.

Figure 4.

Enhancement of RCA by RNase I. a) Real‐time fluorescence monitoring of RCA in buffer for System 1 with or without RNase I. b) Normalized fluorescence intensities at 30 min (F₃₀) relative to the positive control (PC). c) Real‐time fluorescence in 50% pooled saliva. d) Corresponding normalized F₃₀ values. Addition of RNase I significantly enhanced RCA output in both buffer and saliva, confirming its role in generating accessible primers from structured RNA.

To assess potential interferences in clinical samples, the same experiments were repeated in 50% diluted, SARS‐CoV‐2–negative pooled saliva (S) under identical conditions. As shown in Figure 4c,d, RNase I enhanced RCA output in saliva, although the overall signal was lower than in buffer. A similar trend was seen for System 3, while Systems 2 and 4 retained comparable performance in both media. These results confirm that RNase I remains active in complex biological matrices and can enhance RCA efficiency even in saliva‐based assays. Fluorescence curves and F₃₀ values for all systems in saliva are presented in Figure S11.

The trend of decreasing RCA efficiency with shorter CDT protection lengths was consistently observed in both buffer and saliva (Figures S12 and S13). Gel analysis (Figure S14) further confirmed that longer CDTs (CDTb and CDTc) produced stronger and more distinct RCA bands, whereas the shortest design (CDTa) yielded only faint signals. These results indicate that extended CDT protection enhances primer stability and amplification efficiency under optimal RNase I conditions.

The selectivity of RCA is well established, as CDTs hybridize exclusively with their complementary primer sequences.^{[ 54 ]} Even minor sequence mismatches disrupt hybridization and prevent RCA initiation. In the RI‐RCA assay, this specificity is further reinforced by the RNase I digestion step: without correct CDT protection, unbound RNA is degraded, eliminating false priming. Cross‐reactivity experiments, where each lsRNA target was paired with all four CDTs, confirmed this behavior. Significant fluorescence was observed only for matching CDT–RNA pairs, with no amplification in noncomplementary combinations (Figure S15), verifying that RI‐RCA maintains high specificity.

To benchmark performance, RI‐RCA was compared with our previously reported antisense oligonucleotide–assisted RCA (ASO‐RCA) assay.^{[ 50 ]} ASO‐RCA uses 80 nt ASOs to destabilize secondary structures near the CDT binding site following a DNAzyme cleavage step. Because System 1 is too short for effective ASO targeting, the comparison focused on Systems 2–4. For System 2, ASO‐RCA gave slightly higher output, whereas for Systems 3 and 4, RI‐RCA yielded approximately twice the fluorescence of ASO‐RCA (Figure S16). This difference likely reflects greater efficiency of RNase I in removing structural barriers compared to ASOs, especially for longer or more complex RNA targets. Notably, ASO‐RCA requires additional oligonucleotides and a 30 min DNAzyme cleavage step prior to RCA, both of which are unnecessary in RI‐RCA, highlighting the simplicity of the latter method.

To further improve detection sensitivity, we employed two amplification strategies: QE‐RCA and exponential‐RCA (E‐RCA). As illustrated in Figure 5a, QE‐RCA uses linear RCA products as templates for a P2 primer that reinitiates amplification, boosting signal output.^{[ 34 ]} In contrast, E‐RCA utilizes a nicking‐enzyme–based feedback system.^{[ 55} , ⁵⁶ , ^{57 ]} Specifically, a short DNA primer (Tmp2) complementary to the RCA product contains a recognition site for Nb.BtsI and thiophosphate modifications at its 3′ end to prevent polymerase extension. Once bound, Nb.BtsI cleaves the RCA product at the Tmp2 site, generating monomeric fragments that act as new primers while releasing Tmp2 to participate in further amplification (Figure 5d). Using QE‐RCA under the same buffer conditions, fluorescence output increased dramatically (Figure 5b), allowing detection of ls101 RNA at 500 aM (∼200 copies µL⁻¹) after 6 h (Figure 5c)—an improvement of over six orders of magnitude relative to linear RCA (Figure S17). E‐RCA achieved even greater sensitivity, detecting 50 aM of ls101 RNA after 6 h (Figure 5e), consistent with exponential kinetics as verified by point‐of‐inflection (POI) analysis of fluorescence versus log [RNA] (Figure 5f). POI was determined by fitting fluorescence–time traces to a sigmoidal model and identifying the time point at which the second derivative equals zero, corresponding to the maximum amplification rate. This kinetic parameter showed a consistent correlation with input RNA concentration across the tested dynamic range and was less influenced by baseline fluctuations or endpoint saturation than fixed‐time fluorescence readouts. Together, these results demonstrate that coupling RI‐RCA with QE‐RCA or E‐RCA substantially enhances detection sensitivity, enabling reliable quantification of structured RNA at clinically relevant concentrations.

Figure 5.

Sensitivity enhancement through quasi‐exponential RCA (QE‐RCA) and exponential RCA (E‐RCA). a) Schematics of QE‐RCA and E‐RCA mechanisms. QE‐RCA employs a secondary primer (P2) that reinitiates amplification from linear RCA products, while E‐RCA uses a nicking‐enzyme–based feedback loop involving Nb.BtsI and primer Tmp2. b) Real‐time fluorescence curves for QE‐RCA in buffer. c) Calibration curve showing detection of ls101 RNA down to 500 aM. d) Schematic of the E‐RCA feedback cycle. e) Real‐time fluorescence curves for E‐RCA showing exponential amplification and detection at 50 aM. f) Plot of point of inflection (POI) versus log [RNA] confirming exponential amplification behavior. Error bars represent one standard deviation (n = 3).

Clinical Validation

Given that both QE‐RCA and E‐RCA modes achieved attomolar detection limits, QE‐RCA was selected for downstream validation because it offers operational simplicity, requiring only an additional primer (P2) rather than both a nicking enzyme and modified primer. To confirm compatibility between RI‐RCA and QE‐RCA, all four systems were tested in 50% pooled negative saliva spiked with their respective lsRNA targets. As shown in Figure S18, inclusion of RNase I resulted in robust amplification, with endpoint fluorescence (F₃₀) values between 10 000 and 25 000 RFU—comparable to, and in several cases exceeding, those observed for linear RCA (Figure S11). These results verify that RNase I does not interfere with downstream amplification or secondary primer binding, confirming the seamless integration of RI‐RCA with QE‐RCA.

We next evaluated the diagnostic performance of the RI‐RCA platform using saliva samples from 20 SARS‐CoV‐2–positive and 20 negative individuals previously classified by RT‐qPCR (Ct values and variants are summarized in Table S2). Following a 10 min saliva processing step, each sample underwent CDT incubation (5 min), followed by RNase I digestion and QE‐RCA amplification, with endpoint fluorescence at 60 min (F₆₀) serving as the diagnostic readout (total assay time of 75 min). As summarized in Figure 6, all four systems clearly discriminated positive from negative samples, regardless of the viral variant. Receiver‐operator characteristic (ROC)–derived thresholds (Panel A (i–iv)) yielded excellent diagnostic metrics: System 1 achieved 95% sensitivity and 100% specificity, while Systems 2–4 achieved 100% accuracy. Box plots (Panel B (i–iv)) showed well‐separated fluorescence distributions, and ROC analyses (Panel C (i–iv)) confirmed high diagnostic accuracy (Panel D): 98% for System 1 and 100% for Systems 2–4. These results demonstrate that RI‐RCA provides reliable SARS‐CoV‐2 detection in minimally processed clinical saliva.

Figure 6.

Clinical validation of RI‐RCA with saliva samples. a) Endpoint fluorescence (F₆₀) values for 20 SARS‐CoV‐2–positive and 20 negative saliva samples analyzed for Systems 1–4, with ROC‐based cutoffs indicated by dotted lines. b) Box plots showing separation between positive and negative cohorts. c) ROC curves for each system demonstrating excellent diagnostic discrimination with area‐under‐the‐curve (AUC) values of 1.00 for all systems. d) Summary of overall diagnostic performance. All four systems exhibited clear signal separation between positive and negative samples, achieving sensitivities of 95–100% and accuracies of 98–100%.

To further assess assay specificity against non‐target viral species, negative pooled saliva samples were individually spiked with three human coronaviruses—OC43, 229E, and NL63. For each virus, a final concentration of 12500 PFU mL^{− 1} was used, and samples were analyzed using all four RI‐RCA systems following CDT incubation, RNase I digestion, and QE‐RCA amplification. Robust fluorescence signals were observed exclusively for SARS‐CoV‐2–positive samples, while no detectable amplification was observed for any of the non‐target human coronaviruses (Figure S19A). In addition, concentration titration experiments spanning 0–25000 PFU mL^{− 1} for OC43, 229E, and NL63 produced no detectable amplification across the tested range (Figure S19B). These results demonstrate that RI‐RCA exhibits high specificity for SARS‐CoV‐2 without cross‐reactivity toward other human coronaviruses.

To confirm that RI‐RCA is also compatible with exponential RCA, we analyzed saliva samples from the first ten positive and ten negative patients using E‐RCA. Each sample underwent CDT incubation and RNase I digestion, followed by amplification with Tmp2 and the nicking enzyme Nb.BtsI. As shown in Figure S20, all systems again exhibited clear positive–negative discrimination. Endpoint fluorescence, box plots, and ROC curves demonstrated 100% diagnostic accuracy, indicating that E‐RCA can likewise be integrated with RI‐RCA in clinical matrices. Figure S21 compares the F₆₀ values obtained for these samples using QE‐RCA and E‐RCA. No significant differences in endpoint fluorescence were observed between the two amplification modes, except for a small number of samples in system 4. These discrepancies may be attributed to experimental variability, differences in enzyme activity, or variations in fluorescent dye brightness. Importantly, these results demonstrate that both amplification strategies are suitable for validating clinical samples.

Finally, RI‐RCA was benchmarked against our previously reported ASO‐RCA. Because System 1 lacks sufficient structure for ASO targeting, comparisons were restricted to Systems 2–4. Clinical ASO‐RCA data for Systems 2–3 was taken from prior studies, while System 4 was newly evaluated here using an 80‐nt ASO adjacent to the CDT priming site (Figure S22). Across Systems 2–4, both assays achieved 100% sensitivity except System 2 (95% for RI‐RCA), but RI‐RCA consistently delivered superior specificity (100% versus 80–100%) and overall accuracy (97.5–100% versus 90–100%). RI‐RCA also exhibited lower background fluorescence in both positive and negative samples, attributable to RNase I degradation of endogenous non‐specific RNA that would otherwise generate false‐positive signals. Furthermore, the RI‐RCA workflow was faster, eliminating the 30‐min DNAzyme cleavage step required in ASO‐RCA.

Together, these findings establish RI‐RCA, combined with either QE‐RCA or E‐RCA, as a robust, accurate, and clinically viable diagnostic platform. By transforming enzymatic RNA degradation into a sequence‐selective amplification process, RI‐RCA achieves high sensitivity, absolute specificity, and operational simplicity, providing a powerful alternative to conventional RT‐ or ASO‐based RNA assays.

Conclusions

In summary, we introduce a simple and robust strategy for RNA detection that couples E. coli RNase I digestion with a CDT to transform enzymatic RNA degradation into a selective signal‐generation mechanism. In this system, RNase I partially digests structured RNA, transiently exposing complementary regions that are immediately protected by CDT hybridization, producing short RNA fragments that serve as primers for RCA. This one‐pot process operates efficiently at room temperature in linear, quasi‐exponential and exponential formats, functions robustly in complex biological matrices such as saliva, and achieves detection limits as low as 50 aM. By enzymatically removing non‐target RNA species, the assay minimizes background interference and effectively eliminates the likelihood of false‐positive signals.

Importantly, the RI‐RCA method maintained strong performance in 50% pooled saliva and delivered excellent clinical performance. Across four independent systems, fluorescence readouts from clinical saliva samples demonstrated clear separation between positive and negative cohorts, with sensitivities of 95–100% and overall accuracies of 98–100%. Compared with recently reported saliva‐based molecular assays for SARS‐CoV‐2 detection (Table S3), RI‐RCA offers superior simplicity, a shorter turnaround time, and comparable or higher diagnostic accuracy under ambient conditions.

Saliva is an attractive diagnostic specimen due to its non‐invasive collection and suitability for point‐of‐care testing; however, minimally processed saliva poses significant challenges for conventional RNA‐based assays. High false‐negative rates in saliva‐based RT‐PCR have been attributed to active RNases, unidentified inhibitory components, and substantial inter‐sample variability in viscosity and composition, all of which compromise RNA integrity and enzymatic amplification.^{[ 58 ]} In contrast, RI‐RCA transforms these limitations into analytical advantages by intentionally digesting long, structured genomic RNA into short, accessible regions that can hybridize to complementary CDTs. Target recognition is thus independent of full‐length RNA integrity, and signal generation is shifted to a DNA‐only RCA process that is more tolerant of salivary inhibitors and matrix heterogeneity, enabling reliable molecular detection directly from complex saliva samples.

Beyond SARS‐CoV‐2 detection, the RI‐RCA platform is broadly adaptable to diverse RNA biomarkers associated with viral, bacterial, and oncogenic processes. However, non‐structured RNA targets such as microRNAs do not require an RNase digestion step, as they are approximately 20–25 nucleotides in length and do not form secondary structures that would inhibit RCA. Accordingly, RI‐RCA is particularly well suited for the detection of long, structured RNA targets—such as viral genomic RNAs—where secondary structure and inaccessibility present major challenges for conventional amplification strategies. Conceptually, this work represents the first demonstration of harnessing the digestive power of ribonucleases to enable nucleic acid diagnostics. By converting nuclease activity—traditionally viewed as detrimental—into a constructive force for target recognition, RI‐RCA establishes a new paradigm in RNA analysis. Because many nucleases share similar catalytic properties and rapid turnover rates, the principle is readily extendable to other enzymes and targets. Together, the combination of mechanistic generality, analytical sensitivity, and operational simplicity positions RI‐RCA as a versatile foundation for next‐generation biosensing and clinical diagnostic technologies.

Supporting Information

All relevant data presented in this study are provided in the article and its Supplementary Information. Source data are provided with this paper.

Author Contributions

Y.L., J.D.B., and A.M. conceptualized the idea, designed the experiments and interpreted the data. C.L., Z.Z., and J.G. designed some experiments. A.M. performed all RI‐RCA experiments and analyzed the data. Y.L., J.D.B., and A.M. wrote the manuscript. All authors edited the manuscript.

Conflict of Interests

The authors declare no conflict of interest.

Supporting information

Supporting Information

Data Availability Statement

The data that support the findings of this study are available in the Supporting Information of this article.