Introduction

Nucleic acids are critical biomarkers in molecular diagnostics, offering exceptional specificity and sensitivity for detecting and monitoring diseases. In infectious disease diagnostics, techniques such as polymerase chain reaction (PCR) are widely used to detect viral RNA, enabling accurate diagnoses of conditions like acquired immunodeficiency syndrome (AIDS), hepatitis C, and coronavirus disease 2019 (COVID-19)1,2,3. Similarly, tuberculosis (TB) is identified through the detection of Mycobacterium tuberculosis DNA in clinical samples4, facilitating early and precise intervention.

In cancer diagnostics, liquid biopsy technologies leverage nucleic acid biomarkers such as circulating tumor DNA (ctDNA) and cell-free RNA (cfRNA) for companion diagnostics, early detection, prognosis, and therapeutic monitoring5,6. The complexity of cancer, characterized by genetic and epigenetic heterogeneity, necessitates advanced tools. Next-generation sequencing (NGS), with its multiplexing capabilities, is increasingly favored over PCR for its ability to simultaneously detect multiple mutations, gene expressions, and epigenetic modifications, providing a comprehensive understanding of tumor biology and advancing personalized cancer care.

Highly specific biomarker detection is essential in clinical settings for effective disease management. Early detection of drug-resistant strains in infectious diseases, such as rifampin-resistant Mycobacterium tuberculosis or multidrug-resistant bacteria, enables timely therapeutic adjustments7,8,9,10, reducing mortality and curbing the spread of resistance. In cancer, detecting rare genetic alterations within the background of normal DNA is particularly challenging, especially when ctDNA fractions are below 1%11. Advanced methods like digital PCR and unique molecular identifier (UMI) strategies for NGS allow the detection of rare mutations at ctDNA fractions as low as 0.1%12,13, offering insights into tumor heterogeneity, treatment resistance, and disease progression. However, detecting even lower ctDNA fractions, as seen in early-stage cancers or minimal residual disease (MRD), remains a hurdle, necessitating further advancements in sensitivity and accuracy.

Emerging single-molecule fluorescence techniques such as SiMREPS (Single-Molecule Recognition through Equilibrium Poisson Sampling), Ago-FISH (Argonaut-based Fluorescence In-Situ Hybridization), and dynamic FRET-FISH (Fluorescence Resonance Energy Transfer-based Fluorescence In-Situ Hybridization) have shown promise in enhancing the specificity of nucleic acid biomarker detection14,15,16,17,18. By leveraging kinetic information from transient probe binding, these methods can distinguish highly homologous biomarkers with exceptional precision15,18,19. Additionally, multi-color imaging and probe exchange enable high multiplexing capabilities18,20. Despite these advances, detection speed remains a challenge; current methods require approximately 10 min of observation to achieve optimal analytic specificity15,16,18, which is insufficient for high-throughput diagnostics. Faster detection times are crucial for sequential multiple target analysis in clinical settings.

MicroRNAs (miRNAs), small non-coding RNAs about 22 nucleotides long, are key regulators of gene expression and serve as valuable biomarkers due to their association with diseases like cancer, cardiovascular disorders, sepsis, and neurological conditions21. Detectable in various biological fluids, miRNAs enable early diagnosis of cancers such as pancreatic, breast, and lung cancer with high accuracy22,23,24,25,26. However, their small size and the presence of homologous family or cluster members pose challenges for traditional detection methods like PCR and NGS, which often struggle with specificity, quantification accuracy, and multi-target detection, particularly at low miRNA concentrations.

We present Q-FISH (Quenching-based Fluorescence In-Situ Hybridization), an innovative technology designed for the rapid and highly specific detection of multiple nucleic acid biomarkers. Q-FISH operates with sub-second detection speeds, surpassing previous single-molecule miRNA detection methods like SiMREPS and Ago-FISH by over 600-fold. Demonstrating its capabilities, Q-FISH enabled the rapid discrimination of highly homologous miRNAs and the precise quantification of endogenous miRNAs, setting the stage for more efficient, accurate, and transformative advancements in molecular diagnostics.

Results

Scheme and demonstration of Q-FISH

The key principles of Q-FISH for nucleic acid detection are illustrated in Fig. 1a. The method employs two DNA probes: a fluorescent probe (F-probe) labeled with a fluorophore and a quencher probe (Q-probe) labeled with a quencher (Fig. 1a). The probe sequences are designed to bind in close proximity on the target, enabling efficient fluorescence quenching. To facilitate multiplexing via probe exchange, the binding of both probes must be transient; however, the F-probe should bind more stably than the Q-probe so that multiple quenching events can occur during a single F-probe binding event. A major advantage of using a quencher rather than a fluorophore is that it allows short DNA probes to be used at much higher concentrations for rapid detection of nucleic acid biomarkers without increasing fluorescence background. The higher probe concentration enhances the binding rate, while the shorter probe length increases the dissociation rate, together resulting in an overall increase in detection speed.

Fig. 1: Rapid nucleic acid biomarker detection using Q-FISH.
Fig. 1: Rapid nucleic acid biomarker detection using Q-FISH.The alternative text for this image may have been generated using AI.
Full size image

a Q-FISH Scheme. The detection chamber contains surface-immobilized target nucleic acids and freely floating F- and Q-probes. The F-probe is labeled with a fluorophore, while the Q-probe is labeled with a quencher. Upon the binding of the F-probe to the target, the Q-probe rapidly and transiently binds in close proximity, resulting in frequent fluorescence quenching. b miRNA Preparation for Q-FISH. The preparation involves poly(A)-tailing of miRNA followed by hybridization with biotinylated poly(T) DNA. The hybridized miRNA is then immobilized on a surface via biotin-streptavidin binding. c Sequences used for let-7a detection: target miRNA (let-7a), Cy3-labeled F-probe (orange), and BHQ2-labeled Q-probe (gray). d Fluorescence time trace. A representative Cy3 fluorescence intensity time trace (green) is shown, with a black trace representing a two-state fit generated using hidden Markov modeling. e Representative single-molecule fluorescence images at varying observation times. Molecules identified as let-7a based on kinetic analysis are marked with green circles. The criteria used for let-7a identification are provided in Supplementary Fig. 9. The observation time is indicated at the top of each image, and the number of identified let-7a molecules is shown at the bottom. f Comparison of detection speeds. Observation time dependency of the number of let-7a molecules detected is shown for Q-FISH (green square), dynamic FRET-FISH (blue diamond)17, and Ago-FISH (orange circle). For Q-FISH, 20 nM F-probe and 5 µM Q-probe were used. For dynamic FRET-FISH, 1.2 µM donor and 40 nM acceptor probes were used, while for Ago-FISH, 2 nM Argonaute-loaded DNA probes were applied. Error bars represent the standard deviation. To calculate the error bar for Q-FISH (too small to be visible in the figure), the experiments were repeated seven times, with an average of 252 molecules analyzed per experiment. Figure (f) adapted from “Rapid quantification of miRNAs using dynamic FRET-FISH” by Kim et al., used under CC BY 4.0. Data extracted from the original publication.

As a proof-of-concept experiment for Q-FISH, we detected let-7a miRNA using a custom-built total internal reflection fluorescence (TIRF) microscope (Fig. S1). For this detection, let-7a miRNA was poly(A)-tailed at its 3’-end and immobilized on a surface by hybridization to a complementary poly(T) DNA strand (Fig. 1b). F-probe, labeled with Cy3, was designed to bind to the seed region of let-7a miRNA (Fig. 1c, orange), while Q-probe, labeled with BHQ2, was designed to bind to the tail region of the miRNA (Fig. 1c, gray). The dissociation time of the F-probe was measured to be 22.5 s. To enable multiple binding and dissociation events during F-probe binding, the dissociation time of the Q-probe must be significantly shorter than that of the F-probe. Q-probes of 8, 9, and 10 nucleotides were evaluated, and the 8-nt probe, with a dissociation time of 0.11 s (Fig. S2), was selected because it enabled the fastest target detection (Fig. S3). This Q-probe is one nucleotide shorter than the tail probe used in our previous dynamic FRET-FISH study, resulting in a reduced dissociation time (Fig. S4). In addition, the Q-probe concentration was optimized to promote multiple binding and dissociation events. Concentrations of 0.5, 1.0, 2.0, and 5.0 μM were tested, and 5.0 μM was selected because it provided the fastest detection (Fig. S5). Under our experimental conditions, the photobleaching times of the fluorophores were measured to be longer than the binding lifetimes of the F-probes (Fig. S6). Moreover, blinking events prior to photobleaching were observed only rarely for Cy3 and were not observed at all for Cy5 (Fig. S6).

Figure 1d shows a representative single-molecule fluorescence time trace of Cy3 when both F-probe and Q-probe were added to the detection chamber. Rapid and brief quenching events were clearly visible. In the absence of the Q-probe, Cy3 blinking was observed only infrequently (Fig. S7). In the absence of target miRNA, Cy3 spots were rarely observed (Fig. S8), indicating non-specific binding of F-probes on the surface was negligible. The fluorescence time traces were analyzed using two-state hidden Markov modeling, which provided the dwell times of the fluorescent and quenched states, as well as the number of quenching events observed. A trace was identified as a target when the dwell time of the fluorescent state was less than 0.1 s, the dwell time of the quenched state exceeded 0.04 s, and both fluorescent and quenched states occurred more than four times during observation.

The number of target molecules satisfying the criteria increased monotonically with observation time (Fig. 1e, Fig. S9), reaching a relative detection efficiency of 73.0% with just a 1-s observation period (Fig. 1f, green squares). In comparison, achieving the same detection efficiency required 20 s for FRET-FISH (Fig. 1f, blue diamond) and 750 s for Ago-FISH (Fig. 1f, orange circles), highlighting the ultrafast detection speed of Q-FISH.

Multiplexing capability of Q-FISH

For clinical applications, particularly in cancer diagnosis, it is crucial to simultaneously identify highly homologous multiple biomarkers present in limited specimen amounts. To demonstrate the multiplexing capability of Q-FISH by probe exchange (Fig. 2a), we selected let-7 family miRNAs as target molecules: let-7a, let-7b, let-7c, and let-7d. These miRNAs are highly homologous, sharing an identical seed region with only one or two nucleotide variations in the tail region (Fig. 2b). Additionally, they hold symbolic importance as the first miRNAs identified27,28, and play a critical role in regulating RAS expression, functioning as bona fide tumor suppressor genes29,30. These miRNAs are promising biomarkers for cancer diagnosis, as their expression levels vary depending on the type of cancer31,32.

Fig. 2: Multiplexing scheme utilizing probe exchange.
Fig. 2: Multiplexing scheme utilizing probe exchange.The alternative text for this image may have been generated using AI.
Full size image

a Schematic illustrating multiplexing by probe exchange. When probe sets are sequentially introduced into a detection chamber containing multiple target types (RNA 1-4) immobilized on a surface, the Q-probe selectively binds to its corresponding target, generating a blinking fluorescence signal. In contrast, non-target molecules display stable fluorescence intensity. b Sequences of let-7 family miRNAs and probe sets. The sequences of let-7 family miRNAs and their corresponding probe sets are shown. Red letters indicate sequence variations relative to let-7a, while orange and gray letters represent the sequences of the F-probes and Q-probes, respectively. c Fluorescence time traces. Representative fluorescence time traces (green) were observed when different probe sets were sequentially injected into a detection chamber containing surface-immobilized let-7 family miRNAs (let-7a, let-7b, let-7c, or let-7d). Black traces represent two-state fits generated using hidden Markov modeling. d Positive rate analysis for multiplexing by probe exchange. The positive rate table shows true positives (diagonal) and false positives (off-diagonal) for the detection of let-7 family miRNAs using Q-FISH with a 1-second observation time. The criteria used for target identification are provided in Supplementary Fig. 16. To construct the table, data from 247 to 3,115 molecules, obtained from experiments repeated five to seven times, were used. e Single-molecule fluorescence images obtained by sequentially injecting different probe sets into a detection chamber containing immobilized let-7a (50 pM), let-7b (50 pM), let-7c (50 pM), and let-7d (100 pM). Molecules identified as targets for each probe set are marked with green circles. The observation time was 1 second. The F-probe concentration was 20 nM, while the Q-probe concentrations were 5 µM for let-7a, let-7c, and let-7d, and 1 µM for let-7b.

To detect these miRNAs, we used the same Cy3-labeled F-probe targeting the seed region and distinct BHQ-labeled Q-probes targeting the tail regions of the let-7 family miRNAs (Fig. 2b). To fine-tune dissociation kinetics, some Q-probes were internally labeled, which further accelerated probe dissociation (Fig. S10). To assess the specificity of the probes, the probe sets were sequentially introduced into a detection chamber containing one of the four immobilized let-7 family miRNAs. Representative fluorescence time traces showed that quenching events occurred predominantly with the matching target-probe set, except in the case of probe set A (Fig. 2c, Fig. S11–14). Frequent quenching events were observed for both let-7a and let-7c miRNAs with probe set A. However, these quenching events could be effectively distinguished by applying a kinetic threshold, allowing differentiation between let-7a and let-7c (Fig. S15–16). By leveraging the kinetic information, we successfully identified the let-7 family miRNAs with high analytical specificity (Fig. S17). The false positive rates of Q-FISH for let-7 miRNA family are comparable to previously reported single-molecule miRNA detection techniques (Fig. 2d), such as Ago-FISH and dynamic FRET-FISH, and significantly outperforms commercial PCR-based products16,17,33.

To demonstrate the multiplexing capability of Q-FISH, we immobilized let-7a, let-7b, let-7c, and let-7d in the same detection chamber and detected them by sequentially introducing the corresponding probe sets (Fig. 2e). A similar number of spots were observed for the different targets, except for let-7d. Although the concentration of let-7d was twice that of the others, the number of detected let-7d molecules was the lowest, indicating that its probe was not fully optimized.

In single-molecule fluorescence detection, multiplexing can also be achieved through multi-color imaging (Fig. 3a). To demonstrate this approach for distinguishing let-7 family miRNAs, we used the same BHQ-labeled Q-probe targeting the seed region of let-7a and let-7b and distinct F-probes targeting the tail region: a Cy3-labeled probe for let-7a and a Cy5-labeled probe for let-7b (Fig. 3b).

Fig. 3: Multiplexing scheme utilizing multi-color imaging.
Fig. 3: Multiplexing scheme utilizing multi-color imaging.The alternative text for this image may have been generated using AI.
Full size image

a Schematic illustrating multiplexing by multi-color imaging. Different probe sets, each with differently labeled F-probes but sharing the same Q-probe, are simultaneously introduced into a detection chamber containing multiple target types (RNA 1–2) immobilized on a surface. The distinct F-probes selectively bind to their corresponding targets, generating unique fluorescence signals. Blinking induced by Q-probe binding is used to distinguish true targets from nonspecific binding. b Sequences of let-7a and let-7b miRNAs and probe sets. The sequences of let-7a and let-7b miRNAs, along with their corresponding probe sets, are shown. Red letters indicate sequence variations relative to let-7a. Orange and gray letters represent the sequences of the F-probes and Q-probes, respectively. Cy3-labeled F-probes were used for let-7a detection, while Cy5-labeled F-probes were employed for let-7b detection. c Fluorescence time traces. Representative fluorescence time traces of Cy3 (green) and Cy5 (red) were observed when the three DNA probes were injected into a detection chamber containing simultaneously immobilized let-7a and let-7b. Black traces show two-state fits generated using hidden Markov modeling. d Observation time dependency of the number of detected targets. The number of detected let-7a (green circle) and let-7b (red square) molecules was plotted as a function of observation time. The criteria used for target identification are detailed in Supplementary Fig. 18. Error bars represent the standard deviation. To calculate the error bars, the experiments were repeated 14 times, with each experiment analyzing an average of 68 to 106 molecules. e Single-molecule Fluorescence Images. Green and red circles represent molecules identified as let-7a and let-7b, respectively, within a 1-second observation time. Images were generated from 100 frames, each captured with a 10 ms exposure time. For the experiment, the same concentration (50 pM) of let-7a and let-7b were used for surface immobilization, and Cy3 and Cy5 were simultaneously exited using green and red lasers. f Positive rate analysis for multiplexing by multi-color imaging. The positive rate table shows true positives (diagonal) and false positives (off-diagonal) for let-7a and let-7b detection with a 1-s observation time. To construct the table, a total of 1307 to 1607 spots were analyzed from five independent experiments. For target identification, the same criteria as in (d) were used, as detailed in Supplementary Fig. 18. The F-probe concentration: 20 nM and the Q-probe concentration: 500 nM.

These probes were injected into a detection chamber where equal amounts (50 pM) of let-7a and let-7b miRNAs were immobilized. As shown in the representative single-molecule fluorescence time traces (Fig. 3c), molecules exhibiting Cy3 quenching and Cy5 quenching were clearly distinguishable. To differentiate real targets from non-specific bindings, we utilized Q-probe binding kinetics information (Fig. S18). The number of target molecules meeting the criteria increased monotonically with observation time, achieving relative detection efficiencies of 63.5% for let-7a and 72.8% for let-7b within just a 1-second observation period (Fig. 3d). As expected from the same target concentration, similar numbers of let-7a and let-7b molecules were successfully identified (Fig. 3e, Fig. S19, Supplementary Video 1).

To evaluate the specificity of multiplexing using multi-color imaging, false positive rates were measured with two distinct probe sets in detection chambers containing either let-7a or let-7b (Fig. S20). Slightly higher false positive rates were observed compared to the probe exchange scheme (Fig. 3f). This is expected, as the binding kinetics information of F-probes cannot be utilized to differentiate targets in this multiplexing approach. However, this limitation does not apply when distinguishing less-homologous biomarkers, as distinct F-probes target unique sequences, and the binding kinetics of Q-probes can aid in biomarker identification.

Endogenous miRNA quantification

Finally, we evaluated the ability of Q-FISH to quantify endogenous miRNAs in human tissues. Synthetic let-7a or let-7c was spiked into total RNA extracted from human liver or lung tissue, followed by poly(A) tailing at the 3′-end. After RNA immobilization onto a surface via hybridization to complementary poly(T) DNA strands, Q-FISH was used to quantify miRNAs across a range of spiked-in concentrations. All datasets showed good linear fits, with distinct slopes and y-intercepts (Fig. 4a and S21). The non-zero y-intercepts indicate the presence of endogenous miRNAs in the absence of spiked-in miRNAs. Using the well-conserved linear relationship between spiked-in miRNA concentration and detected spot counts, we estimated endogenous miRNA concentrations in both liver and lung tissue. For liver samples, let-7a and let-7c were present at concentrations of 17.8 pM and 7.3 pM, respectively, when the total extracted RNA concentration was 7 ng/µl (Fig. 4b). The finding that let-7a was approximately 2.4-fold more abundant than let-7c is consistent with previous reports and further underscores the predominance of let-7a in human liver tissue34. In lung tissue, concentrations of both let-7a and let-7c were higher than in liver tissue—35.1 pM for let-7a and 11.9 pM for let-7c, revealing tissue-specific differences in miRNA expression (Fig. 4b).

Fig. 4: Quantification of endogenous miRNAs using Q-FISH.
Fig. 4: Quantification of endogenous miRNAs using Q-FISH.The alternative text for this image may have been generated using AI.
Full size image

a Synthetic let-7a and let-7c were spiked into total RNA extracted from human liver tissue. The detected signals for let-7a (red circle) and let-7c (blue square) showed linear responses, with each data point representing the mean ± SEM of three independent experiments. Based on the fitted calibration curves, the endogenous concentrations were estimated as 17.8 pM for let-7a and 7.3 pM for let-7c. b Comparison of measured molar concentration of let-7a (red circle) and let-7c (blue square) from human liver (left), lung tissue (middle) and DI water as a control (right), with each data point representing the mean ± SEM from three independent experiments. As expected, neither let-7a nor let-7c was detected in the control experiment. For the estimation of the tissue samples, the total endogenous RNA concentration was maintained at 7 ng/µl. The experiments were conducted using the probe sets A and C (Fig. 2).

Discussion

To address the complex challenges in modern molecular diagnostics, there is a growing need for more advanced techniques to detect nucleic acid biomarkers beyond PCR and NGS. For example, liquid biopsy—used to diagnose cancer by detecting trace amounts of cancer biomarkers in biofluids alongside wild-type molecules—requires analytical sensitivity and specificity exceeding the capabilities of PCR and NGS. In addition, achieving high diagnostic accuracy for cancer, a complex disease, necessitates the simultaneous identification of multiple biomarkers within a reasonable timeframe and cost.

The single-molecule detection of nucleic acid biomarkers using fluorescently labeled short DNA probes shows great potential for overcoming challenges in modern molecular diagnostics. By leveraging transient binding kinetics, ultra-specific biomarker detection becomes feasible. Furthermore, multi-color imaging and probe exchange facilitate high multiplexing. However, to achieve widespread adoption in high-throughput clinical applications, detection speed must be further optimized. While the detection time for a single biomarker is significantly shorter than that of PCR (10 min vs. several hours), the sequential detection of multiple biomarkers increases the total diagnostic time proportionally with the number of biomarkers. To address this limitation, Q-FISH has been developed as a technique that enables ultrafast detection of multiple biomarkers.

In Q-FISH, a short DNA probe labeled with a quencher is paired with a longer DNA probe labeled with a fluorophore, and the rapid quenching kinetics of the directly excited fluorescence are used for target identification. While fluorescence quenching can be considered a form of FRET—since the excited energy of the fluorophore is transferred to a non-fluorescent quencher through dipole-dipole interaction as in FRET—Q-FISH differs from our previously developed nucleic acid detection method, dynamic FRET-FISH, in several important ways. In dynamic FRET-FISH, the acceptor signal generated by FRET from the directly excited donor fluorophore is used for target identification. Although the acceptor fluorophore is not nominally excited by the laser, minor but non-negligible direct excitation of the acceptor always occurs. Furthermore, donor signal bleed-through into the acceptor channel becomes significant at high donor probe concentrations. As a result, the concentration of the acceptor and donor probes cannot be increased indefinitely, since background fluorescence in the acceptor channel becomes substantial at high acceptor or donor probe concentrations, ultimately limiting the detection speed of dynamic FRET-FISH. In contrast, the quencher in Q-FISH produces no background fluorescence, allowing for the use of much higher concentrations of the Q-probe and significantly improving target detection speed (Fig. S4c, d).

For multiplexing, Q-FISH can take advantage of two key features of single-molecule imaging: multi-color imaging and sequential probe exchange. By leveraging these capabilities, we demonstrated the rapid and precise detection of multiple highly homologous miRNAs. In this study, we employed only two fluorophore–quencher pairs, Cy3–BHQ2 and Cy5–BHQ2; however, we anticipate that the range of pairs can be expanded to include blue and infrared fluorophores such as Cy2 and Cy7, which would significantly enhance diagnostic throughput.

Although miRNAs were used solely to demonstrate the capabilities of Q-FISH in this study, we expect the technique to be readily adaptable for detecting a broad range of biomarkers, including ctDNA and cfRNA, in liquid biopsy applications. Like other single-molecule techniques such as SiMREPS, Ago-FISH, and dynamic FRET-FISH, Q-FISH exploits probe binding and dissociation kinetics to identify authentic targets. For deployment in molecular diagnostics with large test panels, these approaches require probe-specific threshold optimization and calibration for quantitative analysis. Q-FISH reduced detection time by approximately 600-fold compared with SiMREPS and Ago-FISH, rendering sample preparation steps, such as poly(A)-tailing, the primary limitation on overall turnaround time. Further optimization of sample preparation procedures is therefore expected to substantially reduce the total assay time for single-molecule–based biomarker detection.

Methods

Oligonucleotides preparation

DNA and RNA strands were purchased from Integrated DNA Technology (IDT, Coralville, IA), Bio Basic Inc. (Markham, ON), and Bioneer Inc. (Daejeon, Korea), and their detailed sequences are listed in Supplementary Table 1. Amine-modified DNA probes were labeled with Cy3 or Cy5 mono NHS-esters by incubating 0.059 mM DNA with 1.05 mg/ml fluorophores in a reaction buffer (100 mM sodium tetraborate, pH 8.5) for one hour. To achieve higher labeling efficiency, the fluorophore concentration was increased to 2.1 mg/ml by adding additional fluorophores, and the reaction continued for another hour. Excess dyes were removed via ethanol precipitation, and the labeled oligonucleotides were stored in deionized water. BHQ2-labeled DNA oligonucleotides were obtained from Bio Basic Inc. and Bioneer Inc. Based on absorption measurements, the labeling efficiencies of the probes were estimated to exceed 95% (Fig. S22).

Poly(A)-tailing of miRNAs

For the experiments using synthetic miRNAs, miRNAs were incubated at 37 °C for 1 hour in a reaction buffer containing 20 mM Tris-HCl (pH 7.0), 0.6 mM MnCl2, 20 μM EDTA, 0.2 mM DTT, 100 μg/mL acetylated BSA, 10% glycerol, 5 mM rATP, 90 U/µl Yeast Poly(A) Polymerase (74225Z25KU, Thermo Fisher Scientific), and 2 U/µl RNase inhibitor (N2615, Promega). The reaction was terminated by heating the mixture at 65 °C for 15 min. For quantification of endogenous miRNAs, total RNAs from human liver and human lung tissues were purchased from Thermo Fisher Scientific (AM7960, AM7968). Extracted total RNAs (200 ng/µl) were incubated at 37 °C for 1 hour with synthetic miRNA let-7a or let-7c at varying concentrations in a reaction buffer containing 20 mM Tris-HCl (pH 7.0), 0.6 mM MnCl2, 20 μM EDTA, 0.2 mM DTT, 100 μg/mL acetylated BSA, 10% glycerol, 5 mM rATP, 90U/µl yeast Poly(A) Polymerase (74225Z25KU, Thermo Fisher Scientific), and 2U/µl RNase inhibitor (N2615, Promega). The reaction was terminated by heating at 65 °C for 15 min.

Single-molecule experiment

Single-molecule experiments were conducted using a custom-built total internal reflection fluorescence microscope. Detection chambers were prepared by adhering double-sided tape to the surface of a quartz slide to create flow cells, followed by sealing with a coverslip35. To minimize nonspecific molecular binding, the surfaces of the detection chamber were coated with a mixture of PEG and biotin-PEG in a 9.2:1 ratio.

Poly(A)-tailed RNAs were diluted two-fold in a buffer containing 1.25 mM EDTA and 2U/µl RNase inhibitor, then annealed with 5 µM biotinylated poly(T30) at 55 °C for 5 min. The mixture was gradually cooled to 4 °C at a rate of −1 °C per 30 s. To prepare RNA solutions at the specified concentrations in a total volume of 20 μl, 1.4 μl of the annealed RNAs was diluted with T50 buffer (20 mM Tris-HCl, pH 8.0; 50 mM NaCl), and immobilized on the detection chamber surface via streptavidin–biotin interactions.

F-probes (Cy3 or Cy5) and Q-probes (BHQ2) in an imaging buffer (40 mM Tris-HCl, pH 8.0; 400 mM NaCl; 0.5% formamide; 80 mM urea; 0.4 U/µl RNase inhibitor; and PCD oxygen scavenger system, 46852004, Oriental Yeast Co. LTD.) were injected into the detection chamber. The optimal probe concentrations for miRNA quantification were determined empirically through a trial-and-error approach, as shown in Supplementary Fig. 5. Single-molecule imaging was performed at 30 °C, with Cy3 and Cy5 excited by 532-nm (Compass 215M-50, Coherent) and 632-nm (Cube640–100 C, Coherent) lasers, respectively. Fluorescence signals were collected using a water-immersion objective (UPlanSApo 60x, Olympus), separated by using dichroic mirrors (635dcxr, Chroma), and imaged on an EM-CCD camera (iXon DU-897U, Andor) with a 2 × 2 binned image sensor and a 10 ms exposure time (Fig. S1). All data are obtained from more than three experiments, and error bars are calculated by the standard error of the mean.

Digital image processing and analysis

Fluorescence images were captured using a custom program written in LabView (2009, National Instruments). Spatial positions of fluorophores were extracted using WindSTORM36, and image drift was corrected with the Mean-Shift-Drift-Correction algorithm37. Fluorescence time traces were analyzed with Divisive Segmentation and Clustering (DISC)38. Data analysis was performed using IDL (7.0, NV5 Geospatial Software), MATLAB (R2020a, MathWorks), and Origin (8.5, OriginLab).

Statistics and reproducibility

All data were obtained from at least three independent experiments. The number of analyzed molecules is indicated in the corresponding figure captions. Error bars represent either the standard error of the mean (SEM) or the standard error (SE), as specified in the figure captions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.