Near-infrared long lifetime upconversion nanoparticles for ultrasensitive microRNA detection via time-gated luminescence resonance energy transfer

Abstract

Upconversion nanoparticle (UCNP)-based luminescence resonance energy transfer (LRET) biosensing offers advantages such as wash-free detection and precise biomolecule quantification. However, its sensitivity remains limited due to continuous energy transfer in co-doped UCNPs during LRET. Here we present a time-gated LRET strategy using near-infrared (NIR) long-lived luminescent UCNP donors (L-TG-LRET), achieving an 8-fold increase in luminescence lifetime without compromising emission intensity. This prolonged energy migration and transfer pathway significantly enhances sensitivity by preventing rapid Tm³⁺ reactivation during LRET to IRDye800 acceptors. Applying this approach to microRNA (miRNA) detection, we achieve a 17.9-fold higher sensitivity than conventional steady-state methods. Furthermore, the L-TG-LRET successfully quantifies miRNA expression in cancer cells, plasma, and exosomes, enabling the differentiation of cancer patients from healthy donors. Notably, this approach outperforms polymerase chain reaction in detecting low-abundance exosomal miRNAs. These results highlight the potential of L-TG-LRET system as a valuable tool for sensitive biomolecular detection in clinical diagnostics.

Introduction

Lanthanide-doped upconversion nanoparticles (UCNPs), which can emit visible or shorter-wavelength near-infrared (NIR) luminescence under NIR excitation, offer considerable advantages, such as high photostability, non-photobleaching, sharp emission bands, and large anti-Stokes shifts^1,2,3,4,5. In particular, their long luminescence lifetimes in the micro- to millisecond range render them suitable for time-gated luminescence detection^6,7. The NIR excitation and emission properties of UCNPs, combined with time-gated luminescence detection, allow spectral and temporal elimination of background noise and provide autofluorescence-free detection.

Among the various applications of UCNPs in bioimaging^8,9, theragnostics^10,11, drug delivery^12,13, and super-resolution microscopy^14,15, recent studies have highlighted the potential of UCNP-based luminescence resonance energy transfer (LRET) assays in sensitive biosensing and disease diagnostics^16,17,18,19. LRET is a process of short-range, non-radiative resonance energy transfer between an energy donor and acceptor, which facilitates target identification without the need for complex procedural steps^20,21,22. The narrow, tunable excitation and emission bands of the UCNPs facilitate the selection of spectrally compatible acceptors for LRET. Moreover, their NIR properties and large anti-Stokes shifts prevent donor-acceptor co-excitation and photodamage to biomolecules. Consequently, the use of UCNPs as energy donors in LRET assays (UCNP-LRET) allows for a synergistic combination of the advantages of UCNPs with those of LRET, offering an opportunity to achieve high sensitivity, simplicity, and wash-free homogeneous detection for biosensing applications^16,17,18,19.

However, in practical applications, enhancing the sensitivity of UCNP-LRET is more challenging than the energy transfer of single-molecule fluorophores because of the complex processes of long-range excitation energy migration and transfer within the sensitizer-activator network, followed by subsequent energy transfer to the acceptor^23,24,25. This complexity arises because a single UCNP comprises numerous sensitizer and activator ions, with the activator ions serving as the actual donors that transfer energy to the acceptor. Given this complexity, recent studies have reported that the intensity of the optimal upconversion luminescence (UCL) may not necessarily coincide with the ideal conditions for optimal LRET. Consequently, it is imperative to consider multiple factors when attempting to improve LRET sensitivity. These factors include not only the intensity of the UCL but also the size of the UCNPs, their architectural configuration, the distance between the donor and acceptor molecules, and the surface coating^17,23,25. Furthermore, understanding the effects of photoexcitation laser modes, spatial distribution of dopants, and UCL lifetime is essential for developing sensitive and reliable UCNP-LRET biosensing systems.

Herein, we propose a strategy to significantly enhance the sensitivity of UCNP-LRET by incorporating an NIR long-lived luminescent UCNP donor (L-donor) into a time-gated LRET (TG-LRET) assay under 980 nm pulsed excitation. This approach, denoted as L-TG-LRET, was achieved through the synthesis of a sensitizing-core/activating-shell/inert-shell nanostructure comprising NaYbF₄@NaYF₄:10%Yb, 1%Tm@NaYF₄ UCNPs. By precisely adjusting the architectural configuration of sensitizer Yb³⁺ ions and activator Tm³⁺ ions, we were able to extend the NIR luminescence lifetime from 250 to 2080 μs, while maintaining a constant intensity under 980 nm excitation. In contrast to conventional UCNP nanostructures in which the sensitizer and activator ions are co-doped, the L-donor features spatially separated Yb³⁺ and Tm³⁺ ions. This configuration results in a significantly slower Yb³⁺-mediated energy migration and prevents the immediate reactivation of Tm³⁺ during LRET under pulsed excitation at 980 nm (Fig. 1). The unique property of NaYbF₄@NaYF₄:10%Yb, 1%Tm@NaYF₄ UCNPs enables distinct observation of changes in UCL intensity and decay upon LRET from UCNPs (donor, emitting at 800 nm) to IRDye800 (acceptor, absorbing between 650 and 850 nm), thereby enhancing LRET sensitivity. To enable its application in disease diagnostics, the L-TG-LRET system has been used for microRNA (miRNA) detection. MiRNAs are small, single-stranded, noncoding RNAs (18−25 nucleotides) that orchestrate gene expression at the post-translational level. Owing to their small size, miRNAs are inherently stable in various body fluids, making them promising biomarkers for liquid biopsies. The L-TG-LRET assay demonstrated the ability to detect three types of miRNAs at the attomolar level and successfully identified them in cancer cells, patient plasma, and exosomes. In particular, the L-TG-LRET-based detection strategy exhibited superior sensing performance compared to the polymerase chain reaction (PCR) for detecting low levels of exosomal miRNAs. These results highlighted the potential of the L-TG-LRET assay for precise diagnostic applications. It is anticipated that the L-TG-LRET assay can be extended to detect a range of biomarkers, thereby contributing to the diagnosis and monitoring of various diseases.

Fig. 1: Schematic illustration of miRNA detection based on LRET.

The pDNAs indicate probe DNAs that are complementary to the respective target miRNAs. IRDye800 is an acceptor. a TG-LRET using NIR long-lived UCNP as a donor exhibits higher sensitivity than (b) SS-LRET using NIR short-lived UCNP as a donor. UCNP, upconversion nanoparticle; LRET, luminescence resonance energy transfer; EM, energy migration; ET, energy transfer. In a, b dark blue and light blue spheres represent Yb³⁺ ions in the excited and ground states, respectively. Part of the figure was created in BioRender. https://BioRender.com/7185k6m.

Results

Design and synthesis of NIR long-lived luminescent UCNPs

To achieve an efficient LRET, we adopted a sensitizing-core/activating-shell/inert-shell nanostructure. The sensitizing core was designed to be NaYF₄:x%Yb (x = 0, 50, and 100) and functioned as an energy migration (EM) core. The activating shell was designed to be NaYF₄:y%Yb,z%Tm ( y = 10, 50, 100 - z, and z = 1, 2), serving as an energy transfer upconversion (ETU) shell. The inert shell is composed of NaYF₄. The nanostructure is capable of spatially confining Tm³⁺ ions (activators), which serve as actual donors in LRET, within the ETU shell. This reduces the distance between the donor and acceptor, thereby facilitating an efficient LRET^26,27. As illustrated in Fig. 2a, the excitation energy at 980 nm was harvested by the Yb³⁺ ions (sensitizers) in the EM core, initiating significant EM processes within the Yb³⁺ sublattice. The harvested energy is then transferred via EM and ETU to sensitize the Tm³⁺ ion in the ETU shell, producing NIR luminescence at 800 nm (³H₄→³H₆) through a two-photon upconversion process^28,29,30. The designed UCNPs were synthesized using a thermal decomposition method. Representative transmission electron microscopy (TEM) images (Fig. 2b, c; left panels) showed that the two core/shell/shell structures of NaYbF₄@NaYF₄:10%Yb,1%Tm@NaYF₄ and NaYF₄@NaYbF₄:2%Tm@NaYF₄ exhibited uniform size, morphology, and layer thickness. This uniformity was consistent across all other particles in the series of nanostructures (Supplementary Fig. 1). The size of the EM core (NaYF₄:Yb) was ~13 nm, the thickness of the ETU shell (NaYF₄:Yb, Tm) was ~2 nm, and the inert shell (NaYF₄) exhibited a thickness of ~1.5 nm. High-resolution TEM (HR-TEM) images of the two types of UCNPs showed clear lattice fringes with a d-spacing of 0.52 nm, corresponding to the interplanar distance of the (100) crystalline plane^31,32 (Fig. 2b, c; upper right panel). The hexagonal phases of the samples were further confirmed by the X-ray diffraction (XRD) analysis^31,32 (Supplementary Fig. 2). The distinct contrast differences in the high-angle annular dark-field scanning TEM (HAADF-STEM) images (Fig. 2b, c; bottom right panel) reveal the distribution of lanthanide ions within the different core/shell/shell structures. Notably, encapsulation with an inert outermost shell minimized surface quenching^29,33 and thus enhanced the NIR luminescent intensity and prolonged lifetime (Supplementary Figs. 3−6 and Supplementary Tables 1 and 2).

Fig. 2: NIR long-lived luminescent UCNPs.

a Schematic illustration of energy transfer processes in a core/shell/shell UCNP. EM, energy migration; ETU, energy transfer upconversion; CR, cross-relaxation. Representative TEM images of (b) NaYbF₄@NaYF₄:10%Yb,1%Tm@NaYF₄ and (c) NaYF₄@NaYbF₄:2%Tm@NaYF₄. TEM image (top left), size distribution histogram (bottom left), HR-TEM image (top right), and STEM image (bottom right) of representative core/shell/shell NPs. The (100) plane with 0.52 nm interplanar spacing of (b) β-NaYbF₄ and (c) β-NaYF₄ is identified. Data in b, c are representative of n = 3 independent experiments. d Representative emission spectra and lifetime at 800 nm of NaYF₄:x%Yb@NaYF₄:98%Yb,2%Tm@NaYF₄ UCNPs with different Yb³⁺ doping concentration x in EM core under 980 nm excitation. e Representative emission spectra and lifetime at 800 nm of NaYF₄:x%Yb@NaYF₄:y%Yb,z%Tm@NaYF₄ UCNPs with different Yb³⁺ and Tm³⁺ doping concentration in EM core and ETU shell under 980 nm excitation. The lifetime values of a series of nanoparticles are presented in Supplementary Tables 1 and 2. Source data are provided as a Source data file.

The NIR luminescence properties of the UCNPs can be modulated by varying the doping concentrations of Yb³⁺ and Tm³⁺ ions, which affect the interionic distances between the dopants. When the Yb³⁺ doping concentration in the ETU shell, which was co-doped with Yb³⁺ and Tm³⁺, was reduced, the interionic distance between Yb³⁺ and Tm³⁺ increased, and the absorption of the NIR excitation light decreased. These changes led to a lower Yb³⁺-to-Tm³⁺ energy transfer rate and a reduction in the probability of back-energy transfer from Tm³⁺ to Yb^{3+ 29}, thereby prolonging the NIR luminescence rise and decay times. Furthermore, the NIR luminescence emission intensity decreases owing to the reduction in both the energy harvested by Yb³⁺ and the energy transferred to Tm³⁺. In the synthesized NaYF₄@NaYF₄:y%Yb,2%Tm@NaYF₄ nanostructures, the NIR emission intensity decreased 124-fold when the Yb³⁺ doping ratio (y) in the shell was reduced from 98 to 10 (Supplementary Fig. 5). Concurrently, the NIR rise and decay times of UCNPs exhibited a 45% increase (from 128 to 186 μs) and an 87% increase (from 250 to 467 μs), respectively, by reducing the Yb³⁺ doping ratio (Supplementary Fig. 6 and Supplementary Table 2). Moreover, the NIR luminescence properties were affected by the Tm³⁺ concentration in the ETU shell. A decrease in the Tm³⁺ concentration resulted in greater Tm−Tm ionic separation. This larger separation leads to weaker cross-relaxation (CR) processes, including transitions from (¹D₂, ³H₆) to (³F_2,3, ³F_2,3) and from (¹G₄, ³H₆) to (³H₄, ³H₅), thus decreasing the population of the intermediate state ³H₄^34,35. Consequently, the lifetime of the NIR luminescence (³H₄→³H₆) increases, while the NIR emission intensity decreases. In the experiment, when y in the shell was fixed (y = 10, 50, or 100-z) and the Tm³⁺ concentration decreased from 2% to 1%, the NIR emission intensity exhibited a 29-fold decrease (Supplementary Fig. 5). This was accompanied by an increase in the NIR rise and decay times of up to 100% and 180%, respectively (Supplementary Fig. 6). These results indicate that in nanostructures co-doped with a sensitizer and an activator, an increase in the lifetime is accompanied by a decrease in the UCL intensity.

Because the emission intensity of UCNPs is a critical factor influencing the UCNP-LRET efficiency, the design of UCNPs with both long lifetimes and strong emission intensities is crucial. In the NaYF₄@NaYF₄:y%Yb,2%Tm@NaYF₄ UCNPs, doping additional Yb³⁺ into the EM core, which is spatially separated from the activator, is a strategy that could both extend the luminescence lifetime and increase the UCL intensity. By increasing the concentration of Yb³⁺ in the core to 0%, 50%, and 100%, the energy input to the upconversion process increased. This also facilitates a reduction in interionic Yb separation according to Dexter’s theory, promoting a more efficient EM^32,36. The migrated energy underwent a relay ETU process to reach the activator Tm³⁺ in the shell, resulting in an impressive increase in the NIR intensity of up to 65-fold (Supplementary Fig. 5). In addition, the enhanced energy migration within the EM core increased both the number of migration steps and the migration distance, which in turn extended the rise and decay times by up to 165% and 84%, respectively (Supplementary Fig. 6). Based on these findings, we identified the optimal doping ratio that not only maximized NIR luminescence intensity and maintained a short decay lifetime of ~250 μs (Fig. 2d) but also extended the lifetime by more than 8-fold, up to 2080 μs, without compromising intensity (Fig. 2e). This was achieved by adjusting the nanoparticle composition and introducing an EM layer that was spatially separated from the activator ions.

TG-LRET using NIR long-lived luminescent UCNPs

To verify the influence of UCL lifetimes on the LRET phenomenon, three UCNPs exhibiting identical emission intensities but differing lifetimes were examined under both 5 nm short pulse and 5 ms long pulse excitation (Fig. 3a and Supplementary Fig. 7). These UCNPs also exhibited comparable quantum yields and extinction coefficients (Supplementary Fig. 8 and Supplementary Table 3). This approach allowed us to consider only the effect of the luminescence lifetimes of the LRET donor, while excluding the effect of the donor emission intensity on UCNP-LRET. The UCNP with the longest lifetime in Fig. 2e, NaYbF₄@NaYF₄:10%Yb,1%Tm@NaYF₄, was denoted as the L-donor. The one with the middle lifetime, NaYF₄:50%Yb@NaYF₄:50%Yb,1%Tm@NaYF₄, was denoted as the M-donor, and the one with the shortest lifetime, NaYF₄@NaYbF₄:2%Tm@NaYF₄, was denoted as the NIR short-lived luminescent UCNP donor (S-donor). To compare the efficiency of LRET among the three donors, IRDye800 was used as the LRET acceptor because of its high extinction coefficient and significant spectral overlap with the maximum emission of the LRET donors at 800 nm. The acceptors were attached directly to the donor surfaces, which were modified with hydrophilic amine groups via amide bonds (Supplementary Fig. 9).

Fig. 3: Comparative analysis of luminescence decays of L-donor and S-donor undergoing TG-LRET to IRDye800 acceptor.

a Schematic diagram illustrating the energy transfer of L-donor compared to S-donor, both with acceptor (IRDye800) bound to the surface. UCL decay curves of the (b) L-donor, (c) M-donor, and (d) S-donor as a function of acceptor concentration under a 980 nm short pulse (5 ns) excitation. UCL decay curves of the (e) L-donor, (f) M-donor, and (g) S-donor as a function of acceptor concentration under a 980 nm long pulse (5 ms). The changes in luminescence decay time values (-Δτ_D) were compared for (h) the L-donor samples in (b) and (e), for (i) the M-donor samples in (c) and (f), and for (j) the S-donor samples in (d) and (g). L-donor, NIR long-lived luminescent UCNP donor; M-donor, NIR intermediate-lived luminescent UCNP donor; S-donor, NIR short-lived luminescent UCNP donor; TG-LRET, time-gated LRET; SS-LRET, LRET under steady-state excitation. Source data are provided as a Source data file.

The UCL decays of the UCNP-LRET system were investigated by assessing the changes in Tm³⁺ luminescence lifetime (τ_D) at 800 nm for different acceptor concentrations. A 5 ns short pulse laser was used (Fig. 3a). For TG-LRET, the L-donor exhibited a significant decrease in τ_D of 32.0%, whereas the M-donor and S-donor exhibited decreases of 26.9% and 23.6%, respectively (Fig. 3b–d, and Supplementary Table 4). The L-donor exhibited a lower Yb³⁺ doping concentration in its ETU shell, with the majority of the Yb³⁺ ions located in the core. The spatial separation of Yb³⁺ from Tm³⁺ in the shell resulted in a calculated distance of 0.870 nm between the dopants within the shell of the L-donor, which is significantly larger than that of the M-donor (0.522 nm) and S-donor (0.417 nm). As the transition probability decreases with increasing distance between neighboring dopants^37,38, S-donors with shorter interdopant distances enable faster and more frequent EM/ETU pathways. The rapid EM and ETU processes mediated by Yb³⁺, enable the swift reactivation (i.e., recharging) of Tm³⁺, leading to not only in quenching by LRET to the acceptor but also in emission at 800 nm (marked as ① and ② in Fig. 3a). By contrast, M-donors and L-donors with relatively greater dopant spacing exhibit slower recharging, making them more sensitive to the presence of acceptor, thereby increasing the LRET-induced changes in τ_D.

We further compared the time-gated detection (short pulse excitation) with conventional steady-state detection (continuous wave (CW) excitation) by examining τ_D changes in the two donors against acceptor concentration. When the pulse width was extended to 5 ms, the luminescence kinetics reached a plateau (Supplementary Fig. 10), indicating equilibrium in the excitation and relaxation of Tm^3+ 39,40. Accordingly, a 5 ms pulse of excitation was used to simulate the CW-induced steady-state luminescence. In contrast to TG-LRET, LRET under steady-state excitation (denoted as SS-LRET) showed only slight decreases in τ_D for all three donors: 15.8% (from 723 to 609 μs) for L-donor, 13.0% (from 331 to 288 μs) for M-donor, and 15.5% (from 168 to 142 μs) for S-donor (Fig. 3e–g and Supplementary Table 4). This reduced sensitivity is attributed to the immediate recharging of Tm³⁺ by a constant energy input, regardless of the donor nanostructure (Supplementary Fig. 7). We also examined a donor (NaYF₄@NaYbF₄:1%Tm@NaYF₄) in which the Tm³⁺ doping concentration was reduced compared to the S-donor. Although this donor exhibited a 1.8-fold longer lifetime than the S-donor, the emission intensity was 2.2-fold lower than that of the other donors (Supplementary Fig. 11), resulting in a decrease in τ_D of only 19.3% for TG-LRET, and a slight decrease of 10.4% (from 200 to 179 μs) for SS-LRET (Supplementary Fig. 11 and Supplementary Table 4). Nevertheless, TG-LRET exhibited a 120% higher −Δτ_D value compared to SS-LRET with this donor (from 21 to 46 μs). These results suggest that, while extending the donor lifetime improves TG-LRET sensitivity relative to SS-LRET, maintaining high emission intensity is crucial for achieving pronounced luminescence changes. Notably, the L-TG-LRET exhibited a 200% higher -Δτ_D value compared to the SS-LRET using the L-donor (L-SS-LRET) (from 115 to 341 μs, Fig. 3h). Similarly, TG-LRET with M-donor (M-TG-LRET) showed a 190% higher -Δτ_D value compared to SS-LRET with M-donor (M-SS-LRET) (from 43 to 125 μs, Fig. 3i). By contrast, TG-LRET with S-donor (S-TG-LRET) showed only 20% higher −Δτ_D value compared to SS-LRET with S-donor (S-SS-LRET) (from 26 to 31 μs, Fig. 3j), owing to the high density of Yb³⁺ ions neighboring Tm³⁺, which enables efficient energy storage and rapid recharging. Consequently, the −Δτ_D value of the L-TG-LRET was 1000% higher than that of the S-TG-LRET, resulting in a significantly more sensitive LRET system. These results highlight the potential of UCNP-LRET and pave the way for the refined development of UCNP donors and photoexcitation modulation to achieve superior biosensing performance.

UCNP-LRET assay for miRNA detection

To apply UCNP-LRET to biomolecular sensing, we attempted to detect miRNAs that play a pivotal role as regulators of gene expression and govern biological processes such as cell differentiation, proliferation, and apoptosis. Aberrant miRNA expression is closely associated with several diseases, including cancer, cardiovascular diseases, and neurological diseases, underscoring the importance of accurate miRNA detection for a multitude of clinical and research applications. In this experiment, three representative miRNAs (miR-21, miR-155, and miR-375) that are dysregulated in breast and lung cancers were selected for application in the LRET assay. To detect miRNAs using the LRET system, probe DNAs (pDNAs) complementary to the 5′ and 3′ ends of the target miRNAs were conjugated to the surfaces of the donor and acceptor, respectively (Supplementary Note 2, Supplementary Fig. 12, and Supplementary Table 5). The pDNA-donor remained stable over 15 days, as indicated by the consistent zeta potential and absorbance at 260 nm (Supplementary Fig. 13). In the presence of a target miRNA, the LRET donor and acceptor probes specifically bind to the respective ends of the target miRNA and the miRNA serves as a bridge that brings the two probes into close proximity. This proximity enables efficient LRET from the donor to the acceptor, leading to a decrease in the luminescence intensity of the donor at 800 nm under excitation at 980 nm (Fig. 4a). The relative intensity, which reflects the extent of the decrease in luminescence, correlates with the concentration of the target miRNAs, thus enabling quantitative measurement of miRNA levels.

Fig. 4: UCNP-LRET assay for miRNA detection.

a Schematic representation of TG-LRET for miRNA detection. The LRET donor (L-donor and S-donor) and LRET acceptor were conjugated with complementary ssDNA probes (pDNAs) to target miR-21, miR-155, and miR-375, respectively. Calibration curves (dashed line) for miR-21 detection using (b) L-donor and (c) S-donor, and (d) comparison of their LODs. Calibration curves (dashed line) for miR-155 detection using (e) L-donor and (f) S-donor, and (g) comparison of their LODs. Calibration curves (dashed line) for miR-375 detection using (h) L-donor and (i) S-donor, and (j) comparison of their LODs. The cut-off values were determined based on the maximum value of the Youden index analysis; for miR-21 detection, 0.0502 (L-SS, light red circle), 0.0503 (L-TG, red square), 0.0794 (S-SS, light gray circle), and 0.0648 (S-TG, gray square); for miR-155 detection, 0.0621 (L-SS), 0.0685 (L-TG), 0.0782 (S-SS), and 0.0800 (S-TG); for miR-375 detection, 0.0919 (L-SS), 0.109 (L-TG), 0.0829 (S-SS), and 0.0919 (S–TG). All reported values represent the mean ± SD, n = 3 independent experiments. S-SS, SS-LRET with S-donor; L-SS, SS-LRET with L-donor, S-TG, TG-LRET with S-donor; L-TG, TG-LRET with L-donor. Source data are provided as a Source data file. Part of the figure was created in BioRender. https://BioRender.com/7185k6m.

Polyacrylamide gel electrophoresis (PAGE) was performed to confirm sandwich hybridization of miRNA with pDNA-D (donor pDNA) and pDNA-A (acceptor pDNA). As expected, the band corresponding to the pDNA-D/miRNA/pDNA-A complex formed only in the presence of the target miRNA (Supplementary Fig. 14). The analytical performances of the L-TG-LRET and S-TG-LRET assays for miRNA detection were evaluated using serially diluted concentrations of each target miRNA. Both assays demonstrated a linear relationship between the relative intensities and logarithmic concentrations of miR-21, ranging from 10 aM to 100 fM (Fig. 4b, c and Supplementary Fig. 15). Notably, the L-TG-LRET showed higher responses than the S-TG-LRET at equivalent concentrations owing to its greater LRET sensitivity, achieving a 3-fold lower limit of detection (LOD) of 1.6 aM compared to 4.8 aM (Fig. 4d). A parallel control study using L-SS-LRET and S-SS-LRET showed that the TG-LRET assays had significantly lower LODs: L-TG-LRET was 10.6-fold lower than L-SS-LRET, and S-TG-LRET was 5.7-fold lower than S-SS-LRET. Similar results were observed for the detection of miR-155 and miR-375, with L-TG-LRET yielding the highest sensitivities, with LODs of 2 aM (4.3-fold lower than that of S-TG-LRET) and 2.1 aM (6.6-fold lower than that of S-TG-LRET), respectively (Fig. 4e−j and Supplementary Figs. 15 and 16). Overall, L-TG-LRET demonstrated an average of 17.9-fold lower LOD compared to the S-SS-LRET for the three types of miRNAs. These sensitivities are superior to those of previously reported assays for miRNA detection (Supplementary Table 6). Subsequently, to investigate the specificity of the LRET systems, we used several non-target miRNAs in addition to the target miRNAs. The results demonstrated that luminescence changes were significant only in the presence of target miRNAs, whereas negligible luminescence changes were observed for all non-target miRNAs (Supplementary Fig. 18). In agreement with the findings from the sensitivity experiments, the intensity change in the heat maps for the target miRNAs was most pronounced when L-TG-LRET was used (Supplementary Fig. 18). These results verified the capability of the LRET assay to distinguish target miRNAs from non-target miRNAs. To further evaluate the quantitative accuracy of the TG-LRET assay, a spike-and-recovery test was conducted in buffer. The TG-LRET assay yielded recovery rates between 93.9% and 107.8%, with low coefficients of variation (CVs) under 15.1%, demonstrating reliable quantification performance (Supplementary Table 7). The TG-LRET assay also showed high repeatability over a 10-day period, and consistent performance in the presence of colored biological interferents, such as hemoglobin and bilirubin, with CVs below 7.7% (Supplementary Figs. 19 and 20). Notably, the developed TG-LRET sensor achieved excellent analytical performance within a short time frame (10 min) in one pot, without the requirement of complex multi-step enzymatic reactions. These advantages lay the groundwork for the future development of various precise and facile diagnostic assays based on TG-LRET sensors that can achieve high analytical sensitivity.

Application of L-TG-LRET assay to cancer diagnosis

Our findings indicated that the L-TG-LRET assay exhibited the most optimal performance for miRNA detection. Accordingly, this method has been used to detect miRNAs in biological and clinical samples. Initially, we attempted to detect miRNAs in various cells using the L-TG-LRET sensor. We measured the levels of miR-21, miR-155, and miR-375 in the lysates of HCC1954 (breast cancer), MCF7 (breast cancer), A549 (lung cancer), HL60 (leukemia), and HaCaT (human epidermal) cell lines. Total RNA extracted from each cell was analyzed using the L-TG-LRET assay, and the respective miRNA concentrations were determined using calibration curves (Fig. 4b, e, h and Supplementary Fig. 22). Quantitative reverse transcription PCR (qRT-PCR) was conducted in parallel. Given the discrepancy in the linear range of miRNA detection between the L-TG-LRET assay and qRT-PCR, samples used for qRT-PCR were diluted 1000-fold and subjected to the L-TG-LRET assay. As shown in Fig. 5a and S23, the expression level of miR-21 was notably upregulated in all cancer cell lines (HCC1954, MCF7, A549, and HL60) compared to HaCaT, a normal cell^41,42,43. The expression level of miR-155 was higher in breast cancer cell lines (HCC1954 and MCF7) and a lung cancer cell line (A549) than in HL60 and HaCaT^44,45,46 (Fig. 5a and Supplementary Fig. 23). The expression level of miR-375 was significantly lower in MCF-7, HL60, and HaCaT cells than in HCC1954 and A549 cell lines (Fig. 5a and Supplementary Fig. 23)^47,48,49,50. These findings are in accordance with previous reports indicating that miR-375 is oncogenic in HER2-positive breast cancer (HCC1954) and tumor-suppressive in non-HER2 breast cancer (MCF7)⁵¹. Furthermore, the L-TG-LRET assay results were consistent with the qRT-PCR results with Pearson’s correlation coefficients of 0.9756 (miR-21), 0.8352 (miR-155), and 0.9921 (miR-375), thereby substantiating the reliability of this methodology (Fig. 5b−d). Based on these results, we expect that the L-TG-LRET sensor can be used to robustly quantify miRNAs in complex biological matrices.

Fig. 5: Application of L-TG-LRET assay to cancer diagnosis.

a Heatmap of miR-21, miR-155, and miR-375 levels in cell lines (HCC1954, MCF7, A549, HL60, HaCaT) measured by L-TG-LRET. Linear correlations between L-TG-LRET and RT-PCR for (b) miR-21, (c) miR-155, and (d) miR-375 across cell lines. Dashed lines represent 95% confidence intervals (CI). e Heatmap of miRNA levels in plasma samples from healthy donors (n = 28), lung cancer (n = 18), and breast cancer (n = 18) patients by L-TG-LRET. Analysis of (f) miR-21 (p = 1.47 × 10⁻⁵), (g) miR-155 (p = 1.93 × 10⁻²⁰), and (h) miR-375 (p = 5.35 × 10⁻⁸) from the plasma samples of lung cancer patients and healthy donors by L-TG-LRET. i ROC curve for differentiation of lung cancer patients from healthy donors by detection of miR-21 (p = 2.23 × 10⁻⁸), miR-155 (p = 1.41 × 10⁻⁸), miR-375 (p = 1.40 × 10⁻⁵), and their combination (p = 3.20 × 10⁻²¹) in plasma samples by L-TG-LRET. Analysis of (j) miR-21 (p = 0.000396), (k) miR-155 (p = 1.53 × 10⁻⁶), and (l) miR-375 (p = 8.63 × 10⁻¹⁰) from the plasma samples of breast cancer patients and healthy donors by L-TG-LRET. m ROC curve for differentiation of breast cancer patients from healthy donors by detection of miR-21 (p = 0.002676), miR-155 (p = 3.96 × 10⁻⁵), miR-375 (p = 5.30 × 10⁻⁷), and their combination (p = 6.07 × 10⁻¹⁵) in plasma samples by L-TG-LRET. miRNA concentrations were calculated using calibration curves from Fig. 4 and Supplementary Fig. 22. Data distributions were assessed with the Shapiro–Wilk test. Based on normality, either two-tailed unpaired t-tests or Mann–Whitney U tests were applied. Benjamini–Hochberg correction (FDR = 0.05) was used for multiple comparisons. ROC analysis and AUCs were performed using GraphPad Prism 10 with 95% CI. Boxes represent the mean, 25th, and 75th percentiles, while whiskers show the largest and smallest values. All data points represent the mean of three independent measurements. Source data are provided as a Source data file.

Subsequently, the expression levels of miR-21, miR-155, and miR-375 were evaluated in 64 clinical plasma samples (28 from healthy donors, 18 from breast cancer patients, and 18 from lung cancer patients) using the L-TG-LRET assay. miR-21 and miR-155 were relatively upregulated in both lung (p < 0.0001 for both) and breast cancer patients (p < 0.001 and p < 0.0001, respectively) compared to healthy donors, consistent with previous studies (Fig. 5f, g, j, k)^{46,52,53,54,55,56}. In contrast, miR-375 was downregulated in lung and breast cancer patients (p < 0.0001 for both), corroborating earlier reports^57,58 (Fig. 5h, j). The miRNA concentrations measured by L-TG-LRET closely matched those obtained by qRT-PCR, confirming the reliability of the developed method for miRNA detection in complex biological samples (Supplementary Figs. 24 and 25). Receiver operating characteristic (ROC) curve analysis for the L-TG-LRET assay demonstrated that miR-21, miR-155, and miR-375 effectively distinguished lung cancer patients from healthy individuals, with area under the ROC curve (AUC) values of 0.9931, 1.0000, and 0.9778, respectively. The corresponding 95% confidence intervals (CI) were 0.8886–0.9989 for miR-21, 0.9180–1.0000 for miR-155, and 0.8407–0.9955 for miR-375. At the optimal thresholds determined by Youden’s index, the sensitivities were 94.4%, 100%, and 94.4%, and specificities were 87.5%, 95.0%, and 90.0% for miR-21, miR-155, and miR-375, respectively. For breast cancer patients, the AUC values for miR-21, miR-155, and miR-375 were 0.7611, 0.8574, and 0.9361, with 95% CIs of 0.5543–0.8734, 0.6984–1.0000, and 0.8934–1.0000, respectively (Fig. 5i, m). At the optimal thresholds, the sensitivities were 77.78%, 83.33%, and 100.0%, and specificities were 72.5%, 95.0%, and 92.5% for miR-21, miR-155, and miR-375, respectively. These AUC values were comparable to those obtained using qRT-PCR (Supplementary Fig. 26). Given the complex regulatory roles of multiple miRNAs in tumorigenesis, we further evaluated the combined diagnostic performance of all three miRNAs. The combined panel achieved AUCs of 0.9778 for lung cancer and 0.8943 for breast cancer, with sensitivities of 96.3% and 87.0%, and specificities of 94.1% and 88.1%, respectively (p < 0.0001 for both). The corresponding 95% CIs were 0.8970–0.9744 for lung cancer and 0.8207–0.9430 for breast cancer (Fig. 5i, m).

To further investigate the applicability of the L-TG-LRET assay for the detection of trace exosomal miRNAs, exosomes were isolated from clinical plasma samples (30 from healthy donors, 15 from breast cancer patients, and 15 from lung cancer patients) and analyzed (Fig. 6a). Exosomes are small extracellular vesicles with diameters of ~30–150 nm secreted by cells. They contain proteins, RNA (especially miRNAs), and DNA fragments that reflect the status and characteristics of the cells of origin. Additionally, they play pivotal roles in signal transduction, material exchange, and a multitude of physiological processes between cells. Therefore, exosomes are important biomarkers for the early diagnosis of disease and therapeutic monitoring. Initially, we characterized the isolated human plasma-derived exosomes. Figure 6b−d illustrates the size distribution of the exosomes as determined by nanoparticle tracking analysis (NTA). The diameter of the exosomes ranged from 61 to 158 nm, with an average concentration of 1.65 × 10¹² exosomes/mL in the isolated plasma samples. As illustrated in the insets of Fig. 6b−d, the TEM images demonstrate the characteristic round morphology of exosomes with a diameter of ~100 nm. Exosomes isolated from human plasma were further confirmed by flow cytometry analysis using CD63, a typical exosome marker (Fig. 6e). The results of the flow cytometry analysis demonstrated the presence of a fluorescence signal derived from exosomes tagged with a PE-conjugated anti-CD63 antibody (green in Fig. 6e). Furthermore, flow cytometry histograms indicated that exosome samples isolated from the plasma of healthy donors, lung cancer patients, and breast cancer patients exhibited a significant increase in mean fluorescence intensity (MFI) compared to exosome-free samples (Fig. 6f). These findings support the reliability of exosomes isolated from clinical human plasma. Next, total RNA extracted from the exosomes contained in 100 μL of clinical plasma was applied to the L-TG-LRET sensor to detect miR-21, miR-155, and miR-375 (Fig. 6g−o and Supplementary Fig. 27). The results demonstrated that the expression levels of miR-21 in exosomes derived from patients with lung cancer (1.83-fold) and breast cancer (2.27-fold) were markedly elevated compared with those in exosomes from healthy donors, which is in accordance with previous reports (Fig. 6g, h, k)^59,60,61,62. The expression levels of miR-155 were observed to be 3.73-fold higher in lung cancer patients and 3.98-fold higher in breast cancer patients^59,60,63. Conversely, miR-375 was downregulated to 0.61-fold and 0.55-fold in lung and breast cancer patients, respectively^47,64,65. Notably, the qRT-PCR results for exosomal miR-155 did not show statistically significant differences between cancer patients and healthy controls (Fig. 6i, m), likely due to the miRNA concentration in exosomes derived from 100 μL of plasma falling below the detection limit (~100 fM) of conventional qRT-PCR (Supplementary Fig. 22). In contrast, the L-TG-LRET assay, with a linear detection range of 10 aM to 100 fM, enabled sensitive quantification of exosomal miRNAs from minimal input. ROC curve analysis further confirmed the diagnostic potential of the L-TG-LRET assay for plasma-derived exosomal miRNAs. For lung cancer, miR-21, miR-155, and miR-375 yielded AUC values of 0.8311, 1.0000, and 0.9200, respectively, while in breast cancer, the corresponding AUCs were 0.9000, 1.0000, and 0.8756 (Fig. 6k, o). When the three miRNAs were combined, the L-TG-LRET assay achieved AUCs of 0.9244 for lung cancer and 0.9491 for breast cancer, with sensitivities of 95.6% for both, and specificities of 82.2% and 81.1%, respectively (p < 0.0001 for both). The corresponding 95% CI were 0.8824–0.9665 for lung cancer and 0.9173–0.9810 for breast cancer. Notably, the L-TG-LRET assay outperformed qRT-PCR in distinguishing cancer patients from healthy individuals, as evidenced by higher AUC values and statistical significance (Supplementary Fig. 28). These findings highlight the superior analytical sensitivity of the L-TG-LRET platform and support its potential for liquid biopsy-based cancer diagnostics through the detection of trace exosomal miRNAs.

Fig. 6: L-TG-LRET assay for exosomal miRNA detection.

a Schematic of lung and breast cancer diagnosis using L-TG-LRET detection of exosomal miRNAs. b–d Size distributions of plasma-derived exosomes (20,000× dilution) characterized by NTA. Insets show TEM images. e Flow cytometry histogram of CD63 in exosomes from plasma. f Quantification of CD63-PE MFI in plasma-derived exosomes. g Heatmap of miR-21, miR-155, and miR-375 levels in exosomes from healthy donors (n = 30), lung cancer (n = 15), and breast cancer (n = 15) patients obtained by L-TG-LRET. L-TG-LRET and PCR quantification of (h) miR-21 (p = 3.35 × 10⁻⁵ and p = 0.0331), (i) miR-155 (p = 1.11 × 10⁻¹⁷ and p = 0.853), and (j) miR-375 (p = 7.37 × 10⁻⁵ and p = 0.001218) in lung cancer vs. healthy donors. k ROC curves for lung cancer detection using miR-21 (p = 0.000334), miR-155 (p = 6.05 × 10⁻⁸), miR-375 (p = 5.35 × 10⁻⁶), and their combination (p = 1.03 × 10⁻¹⁵) in exosome by L-TG-LRET. L-TG-LRET and PCR quantification of (l) miR-21 (p = 0.000233 and p = 0.000621), (m) miR-155 (p = 7.65 × 10⁻⁷ and p = 0.141), and (n) miR-375 (p = 1.02 × 10⁻⁵ and p = 4.55 × 10⁻⁷) in breast cancer vs. healthy donors. o ROC curves for breast cancer detection using miR-21 (p = 1.47 × 10⁻⁵), miR-155 (p = 6.00 × 10⁻⁸), miR-375 (p = 4.72 × 10^-5), and their combination (p = 2.06 × 10⁻¹⁷) in exosome by L-TG-LRET. Left y-axis: miRNA concentration by L-TG-LRET; right y-axis: PCR C_t value. miRNA levels were calculated from Fig. 4 calibration curves. Data distributions were assessed using the Shapiro–Wilk test. Based on normality, two-tailed unpaired t-tests or Mann–Whitney U tests were used. Benjamini–Hochberg correction (FDR = 0.05) was used for multiple comparisons. ROC analyses and AUCs were generated in GraphPad Prism 10 with 95% CI. All tests were two-sided. Boxes represent the mean, 25th, and 75th percentiles, while whiskers show the largest and smallest values. All data points represent the mean of three independent measurements. Source data are provided as a Source data file. Part of the figure was created in BioRender. https://BioRender.com/7185k6m.

Discussion

In this study, we developed an L-TG-LRET system and demonstrated that an appropriate UCNP design in conjunction with time-gated analysis can significantly enhance the sensitivity of UCNP-LRET assays. The L-donor, which features spatially separated Yb³⁺ and Tm³⁺ ions, distinguishes itself from the S-donor by facilitating lower EM/ETU. This extended energy transfer pathway, along with the introduction of an EM layer, allowed for an 8-fold increase in the luminescence lifetime without compromising the emission intensity. The incorporation of time-gated analysis with short pulse excitation also improved sensitivity. In the SS-LRET analysis, continuous energy migration and rapid recharging of Tm³⁺ ions reduced the responsiveness to acceptors, obstructing subtle energy transfer measurements. By contrast, by slowing down the EM and ETU in the L-donor under short pulse excitation, the L-TG-LRET system prevented the rapid reactivation of Tm³⁺ ions during LRET. This is evidenced by the 1000% higher -Δτ_D value compared to S-SS-LRET, indicating a strong response even to small amounts of acceptor. Notably, the design of the L-donor, with a lower density of Yb³⁺ ions near Tm³⁺, resulted in the L-TG-LRET exhibiting a 200% higher -Δτ_D value compared to the L-SS-LRET, whereas the S-TG-LRET showed only a 20% increase compared to the S-SS-LRET. These results suggest that further exploration of the UCNP configurations and analytical methods will pave the way for more sensitive and reliable UCNP-LRET applications.

Building on this enhanced sensitivity, the L-TG-LRET system was successfully applied to the detection of three model cancer-associated miRNAs—miR-21, miR-155, and miR-375—achieving LODs of 1.6, 2, and 2.1 aM, respectively, in one-pot within 10 min. As shown in Supplementary Table 6, the L-TG-LRET assay for miRNA detection exhibited superior sensitivity, which was lower than that of the alternative assays, a considerable proportion of which exhibited LODs in the pM or nM range. In addition, the LOD of this assay was superior to that of the standard qRT-PCR method. This remarkable sensitivity demonstrates the capacity of the assay to detect ultra-low levels of miRNA, which is critical for early disease diagnosis and monitoring of minimal residual disease. Moreover, the assay time of only 10 min signifies a notable enhancement over alternative methods that frequently require longer incubation periods. These advantages make L-TG-LRET a promising platform for miRNA detection in clinical diagnostics, offering both exceptional sensitivity and practical speed.

This system successfully detected the miRNA levels in both biological and clinical samples, demonstrating its clinical potential. Notably, the developed system demonstrated superior sensitivity for low-abundance miRNAs, particularly in exosomal fractions, where conventional methods often encounter difficulties owing to the limited quantity of RNA available. The ability to differentiate between cancer patients and healthy donors based on miRNA expression profiles underscores the potential of L-TG-LRET for the detection and monitoring of cancer. In the future, the adaptability of the L-TG-LRET system offers the potential for its application to extend beyond the detection of miRNAs. The modular design of the assay allows for adaptability in the detection of other nucleic acid targets or small biomolecules through modification of recognition elements. Further research should focus on the expansion of UCNP synthesis and the streamlining of the LRET process with the objective of enabling routine clinical use. Furthermore, the exploration of alternative lanthanide ions or donor−acceptor pairs may facilitate additional enhancements in sensitivity and expand the scope of detectable targets.

In conclusion, the L-TG-LRET system represents a significant advancement in biosensing technology, particularly for applications that require high sensitivity and minimal background interference. Its successful application to both biological and clinical samples establishes a foundation for its use in precision diagnostics, potentially enabling more timely and accurate health interventions. Future developments in UCNP-based biosensors could lead to impactful contributions to the fields of disease diagnosis, treatment monitoring, and personalized medicine.

Methods

Ethical statement

The methodology adopted for this study was reviewed and approved by the Institutional Review Board of Samsung Medical Center (SMC 2021-06-083-015) and the Ethics Committee of the Korea Research Institute of Bioscience and Biotechnology (IRB #P01-202211-02-001), and written informed consent was obtained from all patients prior to blood sampling. All specimens were anonymized with a unique number, and access rights were restricted to only relevant researchers.

Materials

Oleic acid (OA, 90%) 1-octadecene (ODE, 90%), yttrium(III) acetate hydrate (99.9%), ytterbium(III) acetate hydrate (99.9%), thulium(III) acetate hydrate (99.95%), sodium hydroxide (NaOH, ≥98%), ammonium fluoride (NH₄F, ≥99.9%), cyclohexane (≥99%), ethanol (absolute), methanol (≥99.8%), tetrahydrofuran (THF, ≥99.9%), dopamine hydrochloride (≥99.9%), hydrochloric acid (HCl, 37%), dimethyl sulfoxide (DMSO, ≥99.9%), hydroxylamine hydrochloride (HH, 99%), and tris(2-carboxyethyl)phosphine hydrochloride (TCEP) were purchased from Sigma-Aldrich (St Louis, MO, USA). 1 M hydroxyethyl piperazine ethane sulfonicacid (HEPES) buffer, sulfosuccinimidyl 4-(N-maleimidomethyl) cyclohexane-1-carboxylate (sulfo-SMCC), and Total Exosome Isolation Kit Total were purchased from Thermo Fisher Scientific (Waltham, MA, USA). Oligonucleotides were synthesized and purified by Genotech (Daejeon, Korea), Bioneer® (Daejeon, Korea), and Integrated DNA Technologies (IDT) (Coralville, IA, USA). Luna® Universal One-Step RT-qPCR Kit was purchased from New England Biolabs (Ipswich, MA, USA). Dulbecco’s modified Eagle’s medium (DMEM) and Fetalgro Bovine Growth Serum (FBS) were purchased from Welgene Inc. (Gyeongsan, Korea) and RMBIO® (Missoula, MT, USA), respectively. A549, HCC1954, MCF7, and HL60 cell lines were obtained from Korean Cell Line Bank (Seoul, Korea). HaCaT cell line was kindly provided by Professor Hyun Gyu Park (Korea Advanced Institute of Science and Technology). miRNeasy Mini Kit was purchased from Qiagen (Hilden, Germany). Maxwell® RSC miRNA Plasma and Serum Kit was obtained from Promega (Madison, WI, USA). PS Capture™ Exosome Flow Cytometry Kit was purchased from FUJIFILM Wako Pure Chemical Corporation (Osaka, Japan). BD Pharmingen™ PE Mouse Anti-Human CD63 was purchased from BD Biosciences (San Jose, CA, USA). DEPC-treated distilled water and Phosphate-buffered Saline (PBS) were purchased from Bioneer® and used in all experiments. All other chemicals were of analytical grade and used without further purification.

Synthesis of the core UCNPs

The core UCNPs were synthesized using a high temperature injection method. For preparing the core precursor, 0.8 mmol of Ln(CH₃CO₂)₃ (Ln = Y and Yb) was mixed with OA (4 mL) and ODE (6 mL) in a 50 mL flask. The mixture was heated at 153 °C with constant stirring for 1 h 30 min for formation of lanthanide-oleate complex. The solution was then cooled down to 50 °C, and then methanol solution of NaOH (1 M, 2 mL) and methanol solution of NH₄F (0.4 M, 8 mL) were added consecutively with stirring for 1 h. The solution was then heated to 110 °C and degassed under vacuum conditions at 110 °C for 15 min to remove residual methanol. The precursor solution was then cooled to room temperature.

To synthesize the core nanoparticles, OA (6 mL) and ODE (4 mL) were loaded into a 50 mL flask and heated at 110 °C for 15 min. The mixture was degassed under vacuum conditions, and then heated to 310 °C under argon environment. When the temperature of the solution reached 310 °C, half of the core precursor solution was rapidly injected in one shot, and the solution was stirred at 310 °C for 1 h 30 min. The resultant core nanoparticles were collected by centrifugation, washed with ethanol, and dispersed in cyclohexane (2 mL) after cooling to room temperature.

Synthesis of the core/shell/shell UCNPs

The core/shell UCNPs were prepared by the following seed-mediated epitaxial growth procedure. 0.2 mmol of Ln(CH₃CO₂)₃ (Ln = Y, Yb, and Tm) was mixed with 3 mL of OA and 7 mL of ODE in a 50 mL flask. The mixture was stirred constantly and heated to 153 °C for 1 h 30 min, to allow the formation of lanthanide-oleate complex. The solution was then cooled down to 50 °C, and then 1 mL of the as-prepared core nanoparticles was added, along with 0.5 mL methanol solution of NaOH (1 M) and 2 mL methanol solution of NH₄F (0.4 M). The reaction mixture was stirred for 1 h, and then heated to 110 °C under vacuum conditions to remove residual methanol. Then, the solution was heated to 300 °C under argon environment and held for 1 h 30 min. The resultant nanoparticles were collected by centrifugation, washed with ethanol, and dispersed in cyclohexane (1 mL) after cooling to room temperature. The similar procedure was applied to synthesize core/shell/shell UCNPs, with the exception that 0.1 mmol of Y(CH₃CO₂)₃ (0.1 mmol) was used.

Characterization of the UCNPs

The TEM measurements were carried out on a NEO ARM (JEOL Ltd., Tokyo, Japan) installed at Hanyang LINC3.0 Analytical Equipment Center (Hanyang University, Korea) at an acceleration voltage of 200 kV. The scanning electron microscopy (SEM) measurements were carried out on a Sigma 300 (Zeiss, Oberkochen, Germany) installed at the Center for Polymers and Composite Materials (Hanyang University, Korea). The XRD patterns were characterized by a SmartLab diffractometer (Rigaku, Tokyo, Japan). The zeta potential was measured by a Zetasizer Nano ZSP (Malvern Co., Malvern, UK). The Fourier transform infrared (FT-IR) spectra were obtained by a Nicolet iS50 spectrophotometer (Thermo Fisher Scientific). The luminescence spectra were obtained by Kymera 193i spectrometer (Andor Technology, Belfast, UK) and an intensified sCMOS detector (ISTAR-SCMOS-18F-73; Andor Technology) with a shortpass filter (ff-01-950/sp-25; Semrock, NY, USA). The luminescence lifetimes were obtained using a photomultiplier tube (H10721-01; Hamamatsu Photonics, Hamamatsu, Japan) with a band pass filter (ff-01-800/12-25; Semrock). We used two types of lasers: MDL-III-980-2W for CW or long pulse excitation at 980 nm controlled with TTL, and EKSPLA NT342 (OPO laser, repetition rate 10 Hz, pulse duration 5 ns) for short pulse excitation at 980 nm. The average Yb-Tm distance (d) can be approximately calculated by the equation below^35,66:

$$d={\left(\frac{\frac{{a}^{2}c\sqrt{3}}{2}}{1.5\left(x+y\right)}\right)}^{\frac{1}{3}}$$

(1)

x and y denote the Tm³⁺ and Yb³⁺ concentrations, respectively. For the hexagonal NaYF₄ unit cell, a = 0.596 nm and c = 0.353 nm.

Molar extinction coefficient measurement

In this work, molar extinction coefficient (ε_UCNP) of UCNPs at 980 nm was determined using the following equation:

$${\varepsilon }_{{UCNP}}(980)=\frac{A(980)}{l{\rm{\cdot }}{C}_{{UCNP}}}$$

(2)

where A(980) is the absorbance at 980 nm measured using a UV-Vis-NIR spectrophotometer (Cary 5000, Agilent Technologies), l is the optical path length of the cuvette (l = 1 cm), and C_UCNP is the molar concentration of UCNPs. C_UCNP was estimated based on the molecular weight of each nanoparticle and its mass concentration the solution (Supplementary Note 1).

Quantum yield measurement

Quantum yields (Φ) for nanoparticles were measured using an integrating sphere coupled with a fiber optic spectrometer (Supplementary Fig. 8). Initially, the system was calibrated with a Xenon Arc light source (Thorlabs) to obtain a correction factor, which was then applied to adjust the sample spectra. Next, a micro quartz cuvette containing a cyclohexane dispersion of the nanoparticles was placed inside the integrating sphere, connected to a fiber optic spectrometer (Flame, Ocean Optics) to record both emission and excitation signals under 980 nm excitation. A cyclohexane solution without nanoparticles was used as a reference. The quantum yield was calculated using the following equation:

$$\varPhi={N}_{{em}}/{N}_{{abs}}=\int {P}_{{em}}(\lambda ) \lambda d \lambda \,/{\int} [{P}_{{ref}}(\lambda )-{P}_{{sample}}(\lambda )] \lambda d \lambda$$

(3)

where P_em(λ) is the emission intensity spectrum of the sample, and P_ref(λ) and P_sample(λ) are the intensity spectra of the excitation light recorded after passing through the reference solution and the nanoparticle dispersion, respectively.

Preparation of amine-modified UCNPs

Two solutions of UCNPs (15 mg, dissolved in THF) and dopamine hydrochloride (50 mg, dissolved in water) were mixed in 50 mL flask under argon atmosphere, and heated to 50 °C with vigorous stirring. After 5 h of incubation, HCl (1 M) was added, and resulting amine-modified UCNPs were obtained through several washing steps with water.

Comparative analysis of luminescence decays of L-donor and S-donor

Amine-modified donor (50 µg) was mixed with IRDye800 NHS ester (LI-COR Biosciences, Lincoln, NE, USA) at various concentrations (0, 10^−0.5, 10⁰, 10^0.5, 10¹ µM) in HEPES buffer (10 mM, pH 7.4). The mixture was incubated at room temperature for 30 min while shaking at 600 rpm. After the incubation, NIR luminescence lifetimes were obtained.

Preparation of pDNA-modified UCNPs

To prepare maleimide-modified UCNPs, the amine-modified UCNPs (1 mg) were incubated with sulfo-SMCC in HEPES buffer (10 mM, pH 7.4) and incubated for 5 h. The maleimide-modified UCNPs were then collected through several washing steps with HEPES buffer. pDNA-modified UCNPs were obtained using a thiol-maleimide reaction. The mixture of disulfide DNA (Genotech) and TCEP was reacted for 30 min. The maleimide-modified UCNPs were then added to this mixture and incubated overnight. Finally, the pDNA-modified UCNPs were obtained by several washing steps with HEPES buffer. Sequences used for the pDNA-modified UCNPs are listed in Supplementary Table 5.

PAGE analysis

The reaction solution (10 μL) was resolved on 15% polyacrylamide gel for PAGE analysis, run at 130 V for 100 min with 1× TBE as the running buffer. After RedSafe (Intron Biotechnology, Sungnam, Korea) staining, the gel was scanned using ChemiDoc™ (Bio-Rad Laboratories, Hercules, CA, USA).

UCNP-LRET assay for miRNA detection

To evaluate the analytical sensitivity of the LRET assay, synthetic miRNA (3 μL) was mixed with the LRET donor (5 µg, pDNA-modified UCNP) and the LRET acceptor (8 pmol, pDNA-modified IRDye800) (IDT) in HEPES buffer (10 mM HEPES, 50 mM NaCl, pH 7.4) (total volume of 150 μL). The mixture was incubated at 22 °C for 10 min while shaking at 600 rpm. The NIR emission intensity of the LRET donor was measured by the intensified sCMOS detector under CW excitation and pulse excitation, respectively. For the LRET assay under CW excitation (SS-LRET), collection width was 200 ms. For the TG-LRET, time-gated luminescence signals were recorded with a delay time of 100 ms, a gate time of 5 ms, and 40 accumulations, under short pulse excitation at 980 nm with a repetition rate of 10 Hz. In the presence of target miRNA, hybridization of the target miRNA with pDNAs-D/A conjugated to the donor and acceptor brings them into proximity, leading to quenching of the LRET donor luminescence by the acceptor. The relative intensity was calculated from the emission spectra by the equation below:

$${Relative\; intensity}=1-\frac{{I}_{{DA}}}{{I}_{D}}$$

(4)

where I_D and I_DA represent the NIR intensity of the mixture of LRET donor/acceptor pair, before and after reaction with miRNAs.

qRT-PCR

Conventional stem-loop qRT-PCR using Luna® Universal One-Step RT-qPCR Kit was used for PCR analysis. In detail, 12 μL solution containing of DEPC-treated distilled water (2.84 μL), Stem-loop primer (SL primer) (10 μM, 0.6 μL), Forward primer (F primer) (10 μM, 0.48 μL), Reverse primer (R primer) (10 μM, 0.48 μL), Luna® Universal One-Step Reaction Mix (New England Biolabs) (6 μL), Luna® WarmStart® RT Enzyme Mix (New England Biolabs) (0.6 μL), and the analyte solution (1 μL) was subjected to the qRT-PCR protocol programmed for 10 min at 55 °C and 1 min at 95 °C, followed by 45 cycles of 10 s at 95 °C and 30 s at 60 °C. Fluorescence was measured at the end of every extension step with CFX Connect™ RealTime PCR Detection System (Bio-Rad Laboratories). A 1 μL volume of total RNA was used as the analyte solution for each sample type: RNA from cancer cells (100 ng/μL), plasma RNA extracted from 100 μL of patient and healthy donor plasma, and exosomal RNA isolated from exosomes derived from 100 μL of patient and healthy donor plasma. The overhang length of SL primers was optimized to ensure accurate analysis of miRNAs, as demonstrated in Supplementary Fig. 21. The sequences used for the qRT-PCR are listed in Supplementary Table 5 and calibration curves obtained by subjecting synthetic miRNAs at varying concentrations are shown in Supplementary Fig. 22.

Application of L-TG-LRET assay to cancer diagnosis

All cell lines were cultured in DMEM at 37 °C in 5% CO₂. All culture media included 10% FBS. During the exponential growth phase, cell number was determined with LUNA-II^TM automated cell counter from Logos Biosystems (Anyang, Korea). Total RNA was extracted from 1 × 10⁶ cells using miRNeasy Mini Kit (Qiagen) following the instructions of the manufacturer and stored at −20 °C. Extracted RNA concentration was measured by using a NanoDrop® spectrophotometer (Thermo Fisher Scientific) and diluted to 100 ng/μL, which was further used as analyte. Plasma samples from lung cancer patients (n = 18) were obtained from Samsung Medical Center and plasma samples from breast cancer patients (n = 18) were provided by the Biobank of Ajou University Hospital (approval No. AJHB-2024-09). Single donor human plasma samples were obtained from Innovative Research, Inc. (Novi, MI, USA). Total RNA was extracted from 100 μL of patient or healthy donor plasma samples using the Maxwell® 96 Instrument (Promega) with the Maxwell® RSC miRNA Plasma and Serum Kit (Promega), following the manufacturer’s instructions. Exosomes were isolated from 100 μL of patient or healthy donor plasma samples by the Total Exosome Isolation Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions and stored at −20 °C. From the isolated exosomes, total RNA was extracted using the Maxwell® 96 Instrument and Maxwell® RSC miRNA Plasma and Serum Kit following the instruction of the manufacturer. All RNA samples were stored at −20 °C and used as analytes for analyses.

To assess the clinical applicability of the L-TG-LRET, the total RNA samples from cancer cells, patient plasma, and exosomes were used. The total RNA samples (3 μL) were mixed with the LRET donor (5 µg) and the LRET acceptor (8 pmol) in HEPES buffer (10 mM HEPES, 50 mM NaCl, pH 7.4), achieving total volume of 150 μL. The mixture was incubated at 22 °C for 10 min with shaking at 600 rpm. Then, miRNA detection was performed as described above. The cut-off values were determined based on the maximum value of the Youden index analysis.

Characterization of exosomes

Particle number and size distribution of exosomes were determined by NTA using Zeta View PMX 430 (Particle Metrix, Inning am Ammersee, Germany). Isolated exosome samples were appropriately diluted using 1× PBS buffer to track ~100 particles in the field of view. NTA measurement was recorded and analyzed at 11 positions with 488 nm laser.

Negative staining method was used for exosome imaging by TEM. Isolated exosomes were applied on a freshly glow-discharged carbon-coated 200-mesh grid (EMS, Hatfield, PA, USA). Next, 20 µL of 2% uranyl acetate staining solution (EMS) was applied to the sample, and excess staining solution was blotted using the wedge of a filter paper. The grids were allowed to dry and then examined at 120 kV using a FEI Tecnai G2 (Thermo Fisher Scientific). Images were recorded using a Gatan Orius camera (Pleasanton, CA, USA) and processed digitally.

Exosomes used for flow cytometry analysis were isolated from human plasma samples with PS Capture™ Exosome Flow Cytometry Kit (FUJIFILM Wako Pure Chemical Corporation) following manufacturer’s instructions. Exosomes bound on the exosome capture beads were labeled with PE-conjugated anti-human CD63 antibody (BD Pharmingen™, BD Biosciences) for 1 hour at 4 °C in the dark. A total of 20 µL of the pre-diluted antibody was added to 80 µL of bead suspension, following the manufacturer’s recommended usage. After washing twice with washing buffer, the exosome capture beads were resuspended in 600 µL of washing buffer, and a total of 5000 events were recorded on BD LSRFortessa^TM Cell Analyzer (BD Biosciences).

Statistical analysis

Statistical analyses were performed using Origin Pro 2016, IBM SPSS (version 27), and GraphPad Prism 10. Mean values and ± standard deviations were calculated from at least triplicate independent measurements. LODs and linear dynamic ranges were determined using linear regression methods^20,67, involving analysis of the slope of the regression line and the standard deviation of the intercept. Data distributions were assessed for normality using the Shapiro–Wilk test. Based on the distribution, either two-tailed unpaired t-tests or Mann–Whitney U tests were applied to compare miRNA expression levels between cancer patients and healthy donors. Box-and-whisker plots were used for data visualization, with boxes representing the 25th and 75th percentiles, the central line indicating the mean, and whiskers extending to the minimum and maximum values. To assess the concordance between the L-TG-LRET assay and qRT-PCR, both Pearson’s correlation coefficient (r) and Spearman’s rank correlation coefficient (ρ) were calculated. Pearson’s r was used to evaluate the strength of linear relationships, with effect size thresholds of 0.1 (small), 0.3 (medium), and 0.5 (large). Spearman’s ρ was additionally computed to assess monotonic associations without assuming normality. ROC curve analyses, including AUC values, 95% CIs, and p-values, were performed using GraphPad Prism. For multivariable ROC analyses, predicted probabilities were obtained via logistic regression modeling. Outlier assessment was performed using the ROUT method (Q = 1%) in GraphPad Prism. No data points were excluded, as no outliers were identified. Where applicable, p-values were corrected for multiple comparisons using the Benjamini–Hochberg procedure (FDR = 0.05). All statistical tests were conducted at a 95% confidence level. Statistical significance is indicated as follows: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***), and p < 0.0001 (****).

Reporting summary

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