Quantum dot–DNA microsphere aptamer biosensor with AI-assisted structural modeling for rapid detection of the lung cancer biomarker USE1

Abstract

Background

Lung cancer remains a leading cause of cancer-related mortality worldwide, highlighting the urgent need for rapid, accurate, and affordable diagnostic strategies. UBA6-specific E2 conjugating enzyme 1 (USE1) is overexpressed in lung cancer and contributes to tumorigenesis, yet no clinically applicable method exists for its detection.

Results

We developed an AI-assisted aptamer biosensing platform for antibody-free detection of USE1. High-affinity candidates (Aptamer 1e) were identified through systematic SELEX and rational truncation, and AlphaFold3-based modeling was subsequently applied post hoc to provide a structural hypothesis for the observed binding. For signal amplification and visualization, we engineered a nanostructured detection system composed of rolling-circle–amplified DNA microspheres (DNAMS) conjugated with streptavidin–quantum dots (STA-QDs). The DNAMS–STA-QD biosensor enabled strong fluorescence signals in USE1-positive cancer cells and produced a clear visual distinction between tumor and matched normal lung tissues. In 30 paired tissue samples, the biosensor achieved AUC = 0.961, with 86.7% sensitivity and 93.3% specificity for detecting lung cancer. The assay requires no antibodies, enzymatic amplification, or specialized instrumentation, and offers a rapid, low-cost workflow.

Conclusions

This study presents a clinically oriented nanobiosensing platform that integrates AI-assisted aptamer structural modeling with quantum dot–enhanced DNA nanostructures for sensitive detection of USE1. The approach offers a robust, antibody-free method for lung cancer diagnosis and demonstrates the potential of combining deep learning with nanobiotechnology to accelerate biomarker detection tool development.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12951-026-04384-4.

Highlights

• SELEX identifies a high-affinity USE1 aptamer, and post hoc AI-associated modeling provided structural hypotheses that supported interpretation of experimentally validated aptamer candidates.

• DNA microsphere–quantum dot system enables amplified fluorescence detection.

• Biosensor provides visual, antibody-free detection of lung cancer tissue.

• Clinical testing achieved AUC 0.923, with 100% sensitivity and 80% specificity.

• Platform demonstrates scalable nanobiotechnology for cancer diagnostics.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12951-026-04384-4.

Background

Lung cancer is the second most diagnosed cancer worldwide and the leading cause of cancer-related deaths [1]. In 2023, approximately 2.2 million new cases and 1.8 million deaths were recorded, underscoring its significant global health burden [1, 2]. Surgical resection is the primary treatment option for patients diagnosed at an early stage, yet only ~18% present with resectable disease at diagnosis [3]. Moreover, survival outcomes remain suboptimal even after surgery, with five-year survival rates ranging from 90% in stage I to as low as 12% in stage III disease. Consequently, biomarker-guided systemic therapies have become increasingly important in the management of both resectable and advanced-stage lung cancer [4, 5].

Biomarker-driven therapies have revolutionized lung cancer treatment, offering improved survival, reduced toxicity, and enhanced quality of life compared with conventional chemotherapy [6, 7]. Currently, several actionable genetic alterations including epidermal growth factor receptor (EGFR)-activating mutations, anaplastic lymphoma kinase (ALK) rearrangements, ROS1 gene fusions, BRAF^V600E mutations, and neurotrophic tropomyosin-related kinase (NTRK) gene fusions [8–12] are targeted by FDA-approved agents. These targeted therapies are now being incorporated not only in advanced settings but also in adjuvant strategies following curative-intent surgery. In addition, multiple ongoing clinical trials are evaluating their use in neoadjuvant settings.

Nevertheless, platinum-based chemotherapy remains widely used across disease stages and in combination regimens, and clinical outcomes can still be limited by systemic toxicities and treatment resistance [13, 14]. These challenges reinforce the need for practical biomarker detection platforms that can support timely, workflow-compatible decision-making and monitoring in real-world settings.

The UPS plays a crucial role in protein homeostasis and is involved in numerous cellular processes, including DNA repair, endocytic trafficking, and immune response. Dysregulation of this system is implicated in the pathogenesis of various human diseases, including multiple cancers [15]. Among the emerging molecular biomarkers in lung cancer treatment, UBA6-specific E2 conjugating enzyme 1 (USE1), a member of the E2 enzyme family within the ubiquitin–proteasome system (UPS), has gained attention as a novel therapeutic target. USE1 is unique in that it selectively interacts with the UBA6 E1 enzyme rather than the conventional ubiquitin-activating enzyme [16–18]. In our previous study, we demonstrated that USE1 is markedly overexpressed in lung cancer tissues, and that the knockdown of USE1 significantly reduced tumor proliferation, migration, and invasion. These findings establish USE1 as a biologically relevant biomarker and have led to the development of RNAi-based therapeutic agents targeting USE1 for treatment of lung cancer [19, 20].

Despite the growing number of validated biomarkers and related targeted therapies such as USE1, the clinical implementation of biomarker-guided treatment continues to be limited by the lack of rapid, accurate, and user-friendly detection platforms. Antibody-based assays, though widely used, suffer from batch-to-batch variability, limited stability, and high production costs. To overcome these limitations, synthetic aptamers—short, single-stranded oligonucleotides—have emerged as attractive alternatives due to their high binding affinity and specificity, sequence-defined reproducibility, and amenability to chemical modification [21, 22]. Aptamers can be readily adapted to diverse diagnostic platforms, including optical, fluorescence, electrochemical, and nanomaterial-based assays, and are increasingly applied in precision medicine [23–26]. Notably, aptamer-based biosensors for lung cancer biomarkers have been reported in a range of formats—from visually read, point-of-care–oriented assays to instrument-assisted quantitative sensors—highlighting the translational potential of aptamer recognition elements when paired with practical readout strategies (e.g., fluorescence/nanoparticle or electrochemical platforms) [27–30].

However, conventional in vitro SELEX often fails to fully recapitulate physiological conditions, complicating aptamer discovery for clinically relevant targets. This can result in reduced efficacy or off-target effects, limiting clinical translation [31]. Recent advances in deep learning–based structural prediction, such as AlphaFold3, now enable modeling of protein–nucleic acid complexes with unprecedented accuracy, offering an opportunity to rationalize and complement SELEX-based aptamer selection [32–34].

In this study, we used SELEX and rational truncation to isolate high-affinity DNA aptamers targeting USE1, and applied AI-assisted structural modeling post hoc to rationalize the experimentally observed binding mode. Among the selected candidates, Aptamer 1 and its truncated variant, Aptamer 1e demonstrated specific and reproducible binding to USE1 in both lung cancer cell lines and patient-derived tissue samples. Building on these findings, we developed a novel detection platform utilizing biotinylated, self-assembled DNA microstructures conjugated with streptavidin-coated quantum dots for visualization of Aptamer 1e binding to USE1 in tissue sample of lung cancer patients. Our results indicate that these aptamers exhibit strong and specific recognition of USE1, suggesting their potential integration into a clinically applicable detection kit for USE1—a novel and actionable biomarker in lung cancer.

Methods

Patients and lung cancer tissue samples

The experiments were conducted after obtaining informed consent from patients and approval from the Institutional Review Board of Asan Medical Center (2014 − 0960). Surgically resected human lung tissues were acquired from the Asan Bio Resource Center (Resource No. 2014-20 (89)).

His-Tag protein SELEX procedures

The human USE1-targeting aptamer was isolated using a His-tag protein SELEX method. In the initial round of SELEX, the single-strand DNA (ssDNA) library, consisting of 10¹⁵ molecules, was incubated with 2 µg of recombinant His-tagged human full-length USE1 (His-USE1) protein and nickel-nitrilotriacetic acid (Ni-NTA) magnetic agarose beads at 25 °C for 30 min in His-tag binding buffer (2.95 mM KCl, 1 mM MgCl₂, 2.4 mM CaCl₂, 0.1% Tween 20, and 10 mM imidazole in 1X PBS, pH 7.0). The Ni-NTA beads capturing His-USE1 and bound ssDNA were collected using a magnets, and unbound sequences were removed by repeated washing with binding buffer. His-USE1–associated ssDNA was eluted using His-tag elution buffer (500 mM imidazole and 0.02% Tween-20 in PBS). The recovered DNA pool was PCR-amplified, and ssDNA for the subsequent round was regenerated by asymmetric PCR. The resulting ssDNA pool was used as input for the next SELEX round, and the selection procedure (binding–washing–elution–amplification) was repeated for a total of 10 rounds. To impose selection stringency and reduce enrichment of sequences binding non-specifically to the matrix/His-tag, washing was performed under detergent-containing conditions (0.1% Tween-20) and low-imidazole binding conditions (10 mM imidazole), while competitive elution was carried out using high imidazole (500 mM). From the sequenced pool, candidate sequences were prioritized based on enrichment frequency and subsequently screened by ELONA, and the top-performing sequence was designated Aptamer 1.

Aptamer structure and aptamer/protein docking site prediction

The secondary structure of aptamers was predicted using the M-fold web software (available at M-fold). The potential docking sites for aptamer and USE1 protein interactions were identified using the HDOCK server (HDOCK).

ELONA (Enzyme-Linked OligoNucleotide assays)

An ELISA-like assay, ELONA (Enzyme-Linked OligoNucleotide Assays) [35] adapted for DNA aptamers, was performed. The USE1 protein was diluted to 723 nM using 0.1 M sodium phosphate buffer (pH 7.4). Thereafter, 100 µL/well of this solution was incubated overnight at 4 °C in a polystyrene 96-well microtiter plate (MaxiSorpTM; Th. Geyer). Following coating, the wells were washed three times with 200 µL of washing buffer (0.3 M NaCl/PBS containing 0.05% Tween 20) and then blocked with 200 µL of 2% BSA in blocking buffer (0.3 M NaCl/PBS containing 0.05% Tween 20) for 1 h at room temperature. After three additional washes with 200 µL of washing buffer, 100 µL of either 5′- or 3′-biotinylated aptamers (5′ Bio or 3′ Bio) diluted to 10,000 nM in binding buffer (100 mM NaCl, 20 mM Tris-HCl pH 7.6, 10 mM MgCl₂, 5 mM KCl, 1 mM CaCl₂, 0.005% Tween 20) was added to each well. The plate was incubated at room temperature for 1.5 h with mild shaking to facilitate the binding of the aptamers to the immobilized protein.

Following the binding reaction, the wells were washed three times with 200 µL of binding buffer to remove unbound oligonucleotides. A streptavidin-HRP conjugate solution (Thermo Scientific) was diluted 1:10,000 in binding buffer to 0.2–0.25 µg/mL. Thereafter, 100 µL of this solution was dispensed into each well; incubation was performed at room temperature for 1 h with mild shaking. After five washes with 200 µL of binding buffer, 100 µL of 3,3′,5,5′-tetramethylbenzidine (TMB) substrate solution (Thermo Fisher Scientific) was added to each well and incubated for 30 min at room temperature in the dark. The reaction was terminated by adding 50 µL/well of 0.5 M H₂SO₄. The optical density was measured at 370–450 nm using a SpectraMAX spectrophotometer.

Unless otherwise stated in the figure legends, ELONA data represent four independent experiments (independent plates prepared and assayed on different days). Within each plate, each condition was typically measured in duplicate wells (technical replicates). Data are presented as mean ± SD unless otherwise stated. Apparent dissociation constants (Kd) were estimated by nonlinear regression fitting of A450 values to a one-site specific binding model using GraphPad Prism. For ROC analyses, sensitivity, specificity, and 95% confidence intervals were calculated, and cutoffs were determined by the Youden index as specified in the relevant figure legends (Table 1).

Table 1.

Sequences of parent and truncated aptamers targeting USE1

No.	Size (mer)	Sequence (5’ to 3’)
1	95	CACCTAATACGACTCACTATAGCGGATCCGATGGGTGGGGGGGTGGGTAGGATCCGTTCGGGGTTTGGCATAGGGTCTGGCTCGAACAAGCTTGC
1a	18	CTGGCTCGAACAAGCTGC
1b	28	GCATAGGGTCTGGCTCGAACAAGCTTGC
1c	32	CCTAACACGATTCACTATAGCTCGGGGTTTGG
1d	36	CACCTAATACGACTCACTATAGTTCGGGGTTTGGCA
1e	36	CACCTAACACGATTCACTATAGCTCGGGGTTTGGCA
2	98	CACCTAATACGACTCACTATAGCGGACAGGGCTGGTGTGGCTGGCGTCCGGCTCGAACAAGGCTGGTGTGGCTGGCGTTCTGGCTCGAACAAGCTTGC
3	90	CACCTAATACGACTCACTATAGCGGATCCGACAGAATGCCATCACCATGTCTAGACCTATTGGCTTTGCGACTGGCTCGAACAAGCTTGC

AI-assisted, deep-learning based in silico modeling of USE1-Aptamer 1e complex formation

The direct interaction between the USE1 protein and the designed DNA Aptamer 1e was predicted by generating three-dimensional complex models using the AI-assisted, deep-learning-based structural prediction, Protenix (AlphaFold3-based) model [32, 33]. The modeling was conducted post hoc to generate an explanatory hypothesis for the experimentally observed binding, not to prospectively select the candidate. We used the canonical USE1 protein sequence (UniProt ID: Q9H832-1) and the DNA sequence of Aptamer 1e.

Key parameters for the prediction run were set to mmseqs2 for MSA Mode; 6 for Number Recycles; 5 for Diffusion Samples (resulting in five distinct prediction models); and 200 for Sampling Steps. No residue modifications were applied. The quality and confidence of the predicted complexes were evaluated using metrics generated by the platform, including the mean predicted Local Distance Difference Test (pLDDT) score, the predicted Template Modeling score (pTM), and the predicted interfacial Template Modeling score (ipTM). Proximity to the catalytic residue Cys188 was additionally assessed as a measure of interface plausibility. The detailed metrics for all five generated models are available in Supplementary Table 1.

Regarding cross-reactivity prediction analysis, canonical protein sequences of USE1 and comparator E2 enzymes were obtained from UniProt and truncated to the E2 core domain for appropriate comparison. Two DNA aptamers were used in this study: the full-length Aptamer 1 (95 nucleotides) and the truncated Aptamer 1e (36 nucleotides). For each protein–aptamer pair, unbiased structure predictions were generated under identical conditions using five diffusion samples, 200 sampling steps, and six recycling iterations to enable fair comparative analysis across E2 enzymes. For each protein–aptamer combination, multiple independent models were generated. Model quality and binding confidence were evaluated using the ranking score and predicted interfacial TM score (ipTM) provided by the Protenix (AlphaFold3-based). Structural convergence was assessed by comparing the reproducibility of predicted binding poses across independent predictions. In addition, binding surface area (BSA) was calculated as:

BSA = (SASAprotein + SASAaptamer) − SASAcomplex,

where solvent-accessible surface area (SASA) values were derived from the predicted complex structures. All structural visualizations and alignments were performed using PyMOL, with proteins aligned based on the canonical E2 core fold.

Development of the USE1 protein detection biosensing kit

Circularization of linear DNA

Phosphorylated linear DNA (92 bases) and primer DNA (22 bases) at a final concentration of 10 µM were mixed in nuclease-free water (Table 2). For denaturation and annealing, the mixture was heated to 95 °C for 2 min and cooled slowly to 25 °C for 1 h using a thermal cycler (Bio-Rad). To ligate the nick in the circularized DNA, the solution was incubated overnight with 0.06 U·µl^− 1 of T4 DNA ligase and ligase buffer (30 mM Tris-HCl (pH 7.8), 10 mM MgCl₂, 10 mM dithiothreitol (DTT), and 1 mM adenosine triphosphate (ATP)).

Table 2.

DNA sequences for circular DNA synthesis and fabrication of DNAMS

Name	Sequence (5’ to 3’)	Modification
Linear DNA (92 bp)	ACG TAC GGG TGA CGA AAC GAC GTT CCA TCG CTG TTA GAC TCA GAT TGG TTG CAC TTT CAG CAC GGG TTA TTC CGA GTG AAC CGT CCA CCA TC	5’-Phosphate
Primer (20 bp)	CCC GTA CGT GAT GGT GGA CG

Rolling Circle Amplification (RCA) of biotin-modified DNAMS via RCA and Streptavidin Quantum Dot (STA-QD) conjugation

For rolling circle DNA replication (Supplementary Fig. 5 A), the circularized template (at a final concentration of 1 µM) was incubated with Phi29 DNA polymerase (10 U·µl^− 1), dNTP mix (0.8 mM), biotin-14-dCTP (0.04 mM), and reaction buffer (50 mM Tris-HCl, 10 mM MgCl2, 10 mM (NH4)2SO4, and 4 mM DTT) for 20 h at 30 °C. Following incubation, the solution was centrifuged at 3000 RCF for 5 min to remove the supernatant. The particles were then washed with nuclease-free water; washing was repeated twice to thoroughly remove RCA reagents. Hoechst 33,342 dye was added to the washed biotin-modified DNAMS, and the mixture was incubated overnight in a 1000D solution. The dyed DNAMS was then washed using the a forementioned method.

To conjugate quantum dots to the biotin-modified DNAMS, 500 ng of biotin-modified DNAMS was mixed with 0.4 µM of Qdot 565 Streptavidin Conjugate (Invitrogen) solution and incubated for at least 48 h at room temperature. The STA-QD DNAMS was then washed as previously described to remove any unbound reagents. Imaging of the STA-QD DNAMS was performed using an Eclipse Ti (Nikon) inverted fluorescence microscope.

Dot blot analysis with Aptamer 1e and STA-QD DNAMS

For dot blot analysis (Supplementary Fig. 5B), Polyvinylidene fluoride (PVDF) membrane was activated via soaking in methanol (100%) for 1 min. After activation, the membrane was washed with NFW. The protein (BSA or USE1) was loaded onto the PVDF membrane and incubated at 37 ℃ for 1 h. The membrane was washed with PBST (0.1% v/v) for 5 min. For denaturation and annealing of Aptamer 1e, 5 µM of diluted aptamer solution was heated to 95 °C for 2 min and cooled slowly to 25 °C for 1 h using a thermal cycler (Bio-Rad). Thereafter, the USE1 Aptamer 1e was loaded on the protein with different concentrations and incubated at 37 ℃ for 1 h. The membrane was washed with PBST as mentioned earlier. STA-QD (40 fmole) or STA-QD DNAMS (2 pmol) was loaded on the dot and incubated at 37 ℃ for 1 h. Finally, the membrane was washed with PBST overnight. A Dot Blot image was captured using the Gel Doc imaging system. Heatmaps were generated by analyzing the relative intensity of each spot relative to the control blot.

Fluorescence biosensing kit for USE1 protein detection

The kit scaffold was designed using TinkerCad and fabricated with a multijet 3D printer (3D Systems, ProJet 3510 HD) using a biocompatible UV-curable resin (3D Systems, Visijet M3 Crystal) (Supplementary Figs. 6 A). The dimensions of the kit body were 1.0 cm x 1.2 cm x 2.5 cm, with two wells for loading the solutions. An activated PVDF membrane was attached to the backside of the scaffold. For the experimental setup, 500 nM of streptavidin was loaded onto the control line, while 500 nM of BSA or USE1 protein was loaded onto the test line. Thereafter, incubation was performed at 37 °C for 1 h. After incubation, the lines were washed with PBST. Denaturation and annealing of Aptamer 1e were performed as described earlier. Subsequently, 20 pmol of USE1 Aptamer 1e was loaded onto each line and incubated at 37 °C for 1 h. After washing with PBST, 2 pmol of STA-QD DNAMS were loaded into the wells, which were then incubated at 37 °C for 1 h. Finally, the wells were washed with PBST, and images were captured using a Canon EOS M50 Mark II (Canon, Inc., Tokyo, Japan).

The RGB digital images were processed into HSB images using ImageJ software. The hue image was depicted in pseudocolor. For colorimetric analysis, the saturation intensity of the control (C) and test (T) lines based on the color image were analyzed using ImageJ software.

Cell culture and transfection

HEK293T and Lung cancer cell lines, including A549, H1270, H292, and H1299, were obtained from the American Type Culture Collection (ATCC) and cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. The cells were maintained at 37 °C in a humidified atmosphere containing 5% CO₂. Transfection was performed using the iNfect transfection reagent (Intron, Korea), according to the manufacturer’s instructions. Cells were harvested 48 h post-transfection for subsequent analyses. For siRNA-mediated knockdown, Lipofectamine RNAiMax (Invitrogen) was used according to the manufacturer’s protocol. The siRNA sequence of USE1 is “TCTGCTTGAGTATTCTAGGTA”.

Immunoblotting

Tissue and cell lysates were prepared using a 1% SDS lysis buffer (40 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% SDS, 1 mM EDTA) supplemented with protease inhibitors (Complete Mini, EDTA-free; Roche). The lysates were centrifuged at 13,500 ×g for 30 min, and the supernatants were collected. Western blotting was conducted using primary antibodies against His-tag (sc-803, Santa Cruz Biotechnology, 1:1000 dilution), biotin (A150-109 A, Thermo Fisher Scientific, 1:1000 dilution), α-Myc (sc-40, Santa Cruz Biotechnology, 1:1000 dilution), USE1 (1:1000 dilution), and α-alpha-tubulin (ab7291, Abcam, 1:5000 dilution), followed by horseradish peroxidase-conjugated secondary antibodies (Thermo Fisher Scientific). Detection was performed via enhanced chemiluminescence (Bio-Rad).

Fluorescence/quencher aptamer probe for USE1 protein detection

To construct the aptamer probe, 80 nM of Aptamer 1 (ADNA), 40 nM of fluorescence-tagged DNA (FDNA, sequence: 5’−3’ GCA AGC TTG TTC GAG CCA G/F/), and 120 nM of quencher DNA (QDNA, sequence: 5’−3’/Q/CCC TAT GCC AAA C) were combined. The mixture was initially heated to 90 °C and then allowed to reach thermal equilibrium. The temperature was gradually reduced by 1 °C per minute until a final temperature of 20 °C was achieved. Finally, the solution was incubated for an additional 30 min.

For the aptamer probe binding assay with the USE1 protein, 8 nM of the aptamer probe and 1 mM of MgCl₂ were pre-mixed in a total volume of 50 µL of PBS and incubated at 15 °C for 10 min. Prior to reaching the final incubation temperature of 37 °C, different concentrations of the USE1 protein (0, 0.1, 0.3, 1, and 3 µM) were added, and the mixture was incubated for 60 min. After incubation at 22 °C for 30 min, fluorescence measurements were performed every minute.

Results

Selection of aptamers targeting USE1 using SELEX

Supplementary Fig. 1 illustrates the schematic of the His-tag/Ni–NTA magnetic bead–based SELEX process used to isolate His-USE1–targeting DNA aptamers. Briefly, an ssDNA library (~10¹⁵molecules) was incubated with recombinant His-USE1 (2 µg per round) captured on Ni–NTA beads. Across 10 iterative rounds, bound sequences were enriched by repeated stringent washing in binding buffer containing 0.1% Tween-20 and 10 mM imidazole to reduce non-specific bead/His-tag interactions, followed by competitive elution with 500 mM imidazole. Enriched pools were PCR-amplified and ssDNA was regenerated by asymmetric PCR for the next round. After round 10, the enriched pool was subjected to DNA sequencing.

Based on enrichment frequency and subsequent screening results (aptamer array), three candidates (Aptamer 1, 2, and 3) were prioritized for downstream characterization (Table 1). Predicted secondary structures and docking simulations (HDOCK) are presented in Fig. 1, providing supportive structural hypotheses for aptamer–USE1 recognition. The docking simulations suggested putative interaction of each candidate with the USE1–ubiquitin binding region (including C188) (Fig. 1D–G). In addition, equilibrium binding analyses showed that Aptamer 1 exhibited the highest apparent affinity among the three candidates, with Kd values of 10 ± 2.76 nM, 32 ± 2.54 nM, and 90 ± 2.42 nM for Aptamers 1, 2, and 3, respectively (mean ± SD, n = 3; Fig. 1H–J). Accordingly, Aptamer 1 was selected as the parent sequence for subsequent truncation and assay optimization.

Fig. 1.

Predicted secondary structures, docking simulation and estimated equilibrium dissociation curve of aptamers. (A-C) Predicted secondary structure of aptamers 1, 2, and 3 were predicted using M-fold. (D-G) Predicted 3D binding model of aptamers 1, 2, and 3 with USE1 protein generated using the HDOCK server. (H-J) Equilibrium binding curves used to estimate apparent dissociation constants (Kd) of aptamers 1–3 toward purified USE1. Data were fitted by nonlinear regression to a one-site binding model, and Kd values are reported as mean ± SD from independent experiments (n = 3)

Evaluation of the in vitro binding affinity of aptamers using ELONA

An aptamer-based ELONA was developed to evaluate the binding affinity of the aptamers for purified USE1 protein, as previously described [35]. Among the three aptamers, Aptamer 1 (10 µM) exhibited the highest intensity with purified USE1 protein in a time-dependent manner based on ELONA (Fig. 2A).

Fig. 2.

Evaluation of the binding affinity of Aptamer 1 and truncated Aptamer 1e with the USE1 protein using ELONA. (A) Identification of optimal biotinylation orientation for Aptamer 1–3 by ELONA. Microtiter plates were coated with purified USE1 protein (723 nM), and 10 µM of 5′- or 3′-biotinylated Aptamer 1–3 was incubated. Absorbance at 370 nm (A370) was recorded hourly for 15 h. Data points represent the mean of four independent experiments (each performed in duplicate wells). (B) ELONA time-course comparison of truncated Aptamer 1 variants. Plates were coated with USE1 (723 nM) and incubated with 10 µM of Aptamer 1a–1e (and blank control). A370 was measured hourly for 15 h. Data points represent the mean of four independent experiments. (C) Predicted secondary structures of full-length Aptamer 1 and truncated Aptamer 1e. Secondary structures were predicted using Mfold; the highlighted region indicates the truncated segment used to generate Aptamer 1e. (A) and (E) Determination of apparent dissociation constants (Kd) by ELONA titration. Plates were coated with USE1 (723 nM) and incubated with serial dilutions of Aptamer 1 (0–20 nM) or Aptamer 1e (1.25–40 µM). Bound aptamer was detected by streptavidin–HRP/TMB and read at A450. Representative binding curves for Aptamer 1 (D) and Aptamer 1e (E) are shown. Apparent Kd values were obtained by nonlinear regression fitting to a one-site binding model. (F) Summary of apparent Kd values for Aptamer 1 and Aptamer 1e (mean ± SD, n = 3 independent titration experiments). (G) ELONA time-course comparison of Aptamer 1 and Aptamer 1e at a fixed concentration (10 µM). Plates were coated with USE1 (723 nM), incubated with aptamers as indicated, and A370 was recorded over 20 h. Data points represent the mean of four independent experiments (each performed in duplicate wells). (H) Concentration-dependent ELONA time-course profiles for Aptamer 1 (0.625–20 nM) and Aptamer 1e (1.25–40 µM). A370 was measured over 20 h to visualize binding kinetics within each aptamer’s dynamic response range. Blank indicates no aptamer control

To improve manufacturability and reduce oligonucleotide synthesis burden (and to potentially improve tissue penetration and structural stability), full-length Aptamer 1 was further truncated into five sequences (Table 1; Fig. 2C). Subsequent ELONAs with each truncated aptamer (10 µM) and purified USE1 protein were conducted to compare binding affinities. Aptamer 1e exhibited the highest binding capacity compared to the other truncated forms, including 1a, 1b, 1c, and 1 d (Fig. 2B).

To quantitatively compare binding affinity, we performed equilibrium ELONA titrations of Aptamer 1 and Aptamer 1e against purified USE1 protein across a concentration series. Nonlinear regression fitting to a one-site binding model yielded apparent dissociation constants (Kd) of 7.20 ± 2.91 nM for Aptamer 1 and 3.26 ± 0.20 µM for Aptamer 1e (mean ± SD, n = 3), demonstrating that Aptamer 1 binds USE1 with higher affinity (Fig. 2D–F).

Based on these affinity differences, subsequent ELONA experiments were conducted using concentration ranges selected to fall within the dynamic response window of each aptamer (nanomolar range for Aptamer 1 versus micromolar range for Aptamer 1e), thereby enabling measurable signals across the concentration series. The binding signals of Aptamer 1 and 1e for the purified USE1 protein, demonstrated in time- and dose-dependent manners, are shown in Fig. 2G–H. Overall, Aptamer 1 demonstrated high-affinity binding to USE1, whereas the truncated Aptamer 1e showed reduced binding affinity but retained measurable and reproducible USE1 binding in ELONA, supporting its use as a shorter, more manufacturable candidate for downstream assay optimization.

Regard to stability in biological-fluid matrix, biotinylated Aptamer 1 and Aptamer 1e were stored in 10% FBS for 7 days at 4 °C, − 20 °C, or 25 °C, and residual USE1-binding activity was measured by ELONA. When normalized to the 4 °C 0 h condition, both aptamers retained substantial binding at 4 °C (Aptamer 1: 92.6 ± 8.5%; Aptamer 1e: 80.2 ± 4.7%) and − 20 °C (Aptamer 1: 92.2 ± 10.6%; Aptamer 1e: 93.3 ± 11.1%), whereas storage at 25 °C reduced binding signals (Aptamer 1: 53.1 ± 9.7%; Aptamer 1e: 57.9 ± 4.2%) (Supplementary Fig. 7 A; mean ± SD).

Direct interactions between USE1 and Aptamer 1e using AI assisted, deep learning based in silico modeling and immunoblotting

To provide most probable direct interactions between USE1 and Aptamer 1e, we employed in silico complex structure prediction using the AI assisted, deep learning based Protenix (AlphaFold3-based) model (see Methods). Five distinct prediction models were generated, revealing a range of binding orientations of Aptamer 1e with USE1 (Supplementary Fig. 2 A).

Overall, the in silico modeling successfully generated a set of structurally plausible prediction models, as indicated by the quality metrics summarized in Supplementary Table 1. These models exhibit generally high confidence scores, with aggregate scores ranging from 0.754 to 0.8991 and mean pLDDT values above 80, suggesting reliable overall fold predictions. The ipTM values (ranging from 0.72 to 0.89 in Supplementary Table 1) indicate potential interactions between USE1 and Aptamer 1e across all models, suggesting a reasonable degree of confidence in the predicted protein-aptamer interfaces. Importantly, the top three models (Models 1–3) commonly predict a complex model of Aptamer 1e near the USE1’s binding site (indicated by the black dotted circle in Supplementary Fig. 2B). These models also exhibit relatively high ipTM values (0.89, 0.83, and 0.82, respectively), further supporting the likelihood of this binding site interface.

Consequently, we focused on the structural features of Models 1–3 to analyze aptamer interactions with C188 (a key residue in ubiquitination) within the protein-aptamer complex (Figs. 3A-C). Close-up views of these binding modes reveal close distances between the Aptamer 1e and C188 in USE1 (4.3 Å, 3.0 Å, and 2.9 Å in Figs. 3D-F). In contrast, Models 4 and 5 (Supplementary Fig. 2C-F) show a greater distance between the Aptamer 1e and C188 (23.5 Å and 22.7 Å minimum distance to DNA, respectively), suggesting less direct interactions at the ubiquitination site compared to the top-ranked Models 1–3.

Fig. 3.

Direct interaction between USE1 protein and aptamer 1 or 1e by AI-assisted deep learning in silico modeling and in vitro immunoblotting. (A–C) Top 3 predicted binding modes of USE1-Aptamer 1e complex. Proteins are shown in green (A: Model 1), cyan (B: Model 2), and magenta (C: Model 3), while the aptamer 1e is colored in orange and blue. (D–F) Close-up views of the interaction between C188 and the aptamer 1e in each model. Predicted distances between the C188 and the closer aptamer atom are 4.3 Å (D: Model 1), 3.0 Å (E: Model 2), and 2.9 Å (F: Model 3), respectively. (A) Ni-NTA pull-down assay demonstrating the interaction between His-tagged USE1 protein and biotinylated aptamer 1. (B) Ni-NTA pull-down assay showing the binding between His-tagged USE1 protein and biotinylated aptamer 1e

In conclusion, the in silico USE1-Aptamer 1e complex structure prediction by Protenix (AlphaFold3-based) reveals the structurally plausible direct interactions between USE1 and Aptamer 1e. Specifically, Models 1–3 predict a binding mode where the Aptamer 1e interacts with the C188 region (the active site for ubiquitination) of USE1.

To further explore the direct interaction between USE1 and the aptamers, in vitro Ni-NTA and streptavidin pull-down assays were performed using purified His-USE1 protein and biotinylated aptamers. These in vitro assays confirmed that Aptamer 1 and Aptamer 1e exhibit direct interactions with the USE1 protein (Figs. 3G-J). Consistent with these findings, immunoprecipitation assays conducted in 293 T cells overexpressing USE1 further validated the direct binding of Aptamers 1 and 1e to the USE1 protein (Supplementary Figs. 3 A and 3B). Furthermore, aptamer fluorescence quenching assays demonstrated time- and concentration-dependent binding of Aptamer 1 (Supplementary Fig. 4 C and 4D) or Aptamer 1e (Supplementary Fig. 4 F and 4G) to the USE1 protein. Taken together, these data demonstrate direct interactions between Aptamers 1 or 1e and the USE1 protein through complementary in silico modeling, in vitro assays, and in vivo validation.

In silico assessment and biochemical streptavidin pull-down validation of potential cross-reactivity of USE1 aptamers with other E2 enzymes

Given that E2 ubiquitin-conjugating enzymes share a conserved core domain, we assessed whether the USE1 aptamers may exhibit potential off-target binding to other E2 family members. To address this concern, we performed (i) an in silico E2-enzyme panel comparison under identical modeling conditions and (ii) orthogonal biochemical streptavidin pull-down assays to support preferential binding to USE1.

We applied the AI-assisted Protenix (AlphaFold3-based) complex prediction workflow against multiple E2 ubiquitin-conjugating enzymes and structurally related proteins. Given the high structural conservation of the canonical E2 core domain (~145–160 amino acids) across the E2 enzyme family, this analysis aimed to assess whether the aptamers exhibit preferential and structurally stable binding to USE1 relative to other E2 members. A representative panel of E2 enzymes, including UBE2K, UBE2E1, UBE2N, UBE2D2, BIRC6, CDC34, UBE2L3, and UBE2G2, was selected for comparison. Although these proteins share a conserved E2 fold consisting of a central β-sheet flanked by α-helices, their overall sequence identity is relatively low (< 30%). Structural alignment based on the canonical E2 core domain confirmed the overall fold similarity among the selected E2 enzymes, establishing a stringent structural context for assessing potential cross-reactivity (Supplementary Fig. 8 A). Unbiased protein–aptamer complex predictions were generated using the Protenix (AlphaFold3-based) model for USE1 and each comparator E2 enzyme under identical conditions. Both the full-length Aptamer 1 (95 mer) and the truncated Aptamer 1e (36 mer) were independently evaluated to determine whether truncation affected binding specificity. Quantitative comparison of prediction confidence metrics revealed that USE1 consistently exhibited higher ranking scores and interfacial predicted TM scores (ipTM) across independent prediction replicates, whereas other E2 enzymes showed lower mean values accompanied by substantially greater variability (Supplementary Fig. 8B). These results indicate that aptamer binding to USE1 is predicted with higher confidence and structural consistency compared to other E2 family members. Consistent with these confidence metrics, the binding surface area (BSA) analysis demonstrated that USE1–aptamer complexes formed substantially larger and more reproducible interaction interfaces than thnose involving other E2 enzymes (Supplementary Fig. 8E; Supplementary Table 2). Notably, USE1 exhibited not only the largest mean BSA values for both Aptamer 1 and Aptamer 1e, but also comparatively lower standard deviations, indicating a more stable and well-defined interaction interface. In contrast, non-USE1 E2 enzymes displayed smaller BSAs with greater dispersion, suggesting less extensive and less reproducible interaction interfaces. These trends were preserved upon aptamer truncation, demonstrating that the reduced-length Aptamer 1e retained USE1-specific binding behavior. Collectively, these results demonstrate that USE1-targeting aptamers exhibit a strong binding preference and enhanced structural stability toward USE1, with minimal propensity for off-target interactions among structurally related E2 enzymes.

To biochemically assess cross-reactivity in a dense proteomic background, we incubated biotinylated Aptamer 1e with 293 T cell lysate and captured aptamer-associated complexes using streptavidin beads. Under these conditions, the aptamer pull-down robustly recovered USE1, whereas signals for other E2 enzymes including UBE2L3, UBE2N and UBE2G2 were absent or markedly reduced relative to USE1 (Supplementary Fig. 8 F), supporting Aptamer 1e’s preferential recognition to USE1 in a complex lysate environment. Taken together, the in silico E2-enzyme panel comparison and orthogonal streptavidin pull-down validation support preferential recognition of USE1 by Aptamer 1/1e and strengthen the specificity of the USE1 aptamer probes (Supplementary Fig. 8; Supplementary Table 2).

Detection of the use1 protein in lung cancer cells and tissue using Aptamer 1 and 1e

Previous studies have demonstrated that the protein level of USE1 is elevated in 92% of patients with lung cancer, underscoring its significance as a biomarker for the proliferation, invasion, and migration of lung cancer. Consequently, we evaluated the detection capabilities of Aptamer 1 and 1e for the USE1 protein in lung cancer cells and tissues, employing these aptamers as biosensors.

Using ELONA, Aptamer 1 (10 nM) and 1e (1 µM) successfully detected the USE1 protein in 100 ng of lysate from A549 lung cancer cells. This detection was consistent even after overexpression of USE1 with an HA-USE1 plasmid or knockdown with USE1 siRNA (Figs. 4A-C). Furthermore, these aptamers identified the USE1 protein in smaller amounts (50 ng) of lysates from various lung cancer cell lines, including A549, H1270, H292, and H1299 (Supplementary Fig. 3C-F).

Fig. 4.

Evaluation of the binding ability of Aptamer 1 and truncated Aptamer 1e with the USE1 protein with lung cancer cell line and paired tissue sample using ELONA. (A-C) Microtiter plates were coated with 100 ng of A549 cells subjected to different conditions (A; untreated, USE1 overexpressed, siRNA control, USE1 knockdown) and incubated with 10 nM of Aptamer 1 (B) or 1 µM of truncated Aptamer 1e (C). Optical density was measured 1 h post-reaction. Data points represent the average of four independent experiments, each with two replicates. (D) Immunoblot analysis of USE1 expression in paired lung tumor (T) and adjacent non-tumor (N) tissues (pooled cohort, n = 60 pairs; #1–60). (E) ELONA binding signals measured in paired tissue lysates using Aptamer 1e across the pooled cohort (n = 60 pairs; #1–60). (F) ROC curve (95% confidence intervals) for distinguishing tumor from adjacent non-tumor tissues based on ELONA signals in the pooled cohort. The optimal cutoff (2.8) was determined by the Youden index (Sensitivity 100.0%, Specificity 88.3%)

For clinical tissues, ELONA signals were measured in a pooled cohort of 60 paired tumor and adjacent non-tumor tissues (#1–60), and tumor lysates consistently showed higher binding signals than matched adjacent tissues (Fig. 4D–E). ROC analysis (95% confidence intervals) identified an optimal cut-off value of 2.8 (Youden index), yielding sensitivity 100.0% (95% CI 93.98–100.0%) and specificity 88.3% (95% CI 77.82–94.23%) for distinguishing tumor from adjacent non-tumor tissues (Fig. 4F; AUC = 0.96, p < 0.001). Overall, these results support robust detection of elevated USE1 in lung cancer cells and paired tumor tissues using ELONA.

Streptavidin-Quantum Dot (STA-QD) DNA Microsphere (DNAMS) dot blot analysis for the targeting of USE1 with Aptamer 1e

In the realm of molecular diagnostics for various diseases, the in vitro amplification of nucleic acids forms the foundation of all modern methods. Among the diverse amplification techniques, rolling circle amplification (RCA) enables the synthesis of nucleic acid products with predetermined nucleotide sequences. Therefore, we developed a dot blot analysis to detect USE1-targeting Aptamer 1e using self-assembled biotin-modified DNAMS and STA-QD conjugation (STA-QD DNAMS) (Supplementary Fig. 5). Dynamic light scattering analysis confirmed a mean diameter of approximately 1,000 nm for the biotin-modified DNAMS. Following the conjugation of STA-QD to biotin-modified DNAMS, the size increased compared to that of the biotin-modified DNAMS alone. Fluorescence microscopy validated the successful conjugation of STA-QD to biotin-modified DNAMS. The biotin-modified DNAMS were labeled using Hoechst 33,342, exhibiting blue fluorescence, while the conjugated STA-QD displayed red fluorescence (Fig. 5A).

Fig. 5.

STA-QD DNAMS dot blot analysis for USE1 targeting aptamer 1e detection. (A) Dynamic light scattering analysis revealing the size distribution of biotin DNAMS and streptavidin quantum dot conjugated DNAMS (STA-QD DNAMS). Hoechst 33,342 stain (blue) highlights the core of STA-QD DNANS and the red signal indicates the conjugation of STA-QD. (B) Dot blot analysis with BSA and USE1 protein for optimization of aptamer 1e concentration with STA-QD. (C) Dot blot analysis with BSA and USE1 protein for optimization of aptamer 1e concentration with STA-QD DNAMS. (D) Dot blot analysis for the limit of detection of BSA and USE1 protein with STA-QD DNAMS. (E) Immunoblotting of 15 paired tumor and adjacent non-tumor tissues analyzed to detect USE1 protein expression. (F) Dot blot analysis with STA-QD DNAMS performed using lysates from 15 paired tumor and non-tumor adjacent tissues to assess detection ability. (G) Intensity plot of dot blot analysis with STA-QD DNAMS performed using lysates from 15 paired tumor and non-tumor adjacent tissues to assess detection ability

To evaluate the effectiveness and optimization of the polyvinylidene fluoride (PVDF) membrane with a biotinylated aptamer model, an experiment was conducted using biotinylated Aptamer 1e and STA-QD (Fig. 5B). As the amount of biotinylated Aptamer 1e increased from 10 to 20 pmol, the intensity of the USE1 protein signal generally increased, whereas the BSA control dot showed minimal change. However, at 50 pmol of aptamer, the signal intensity in the control group (BSA) was elevated to a level similar to that in the USE1 group, resulting in no significant difference between the BSA and USE1 groups. Consequently, we concluded that the optimal amount of Aptamer 1e for detecting USE1 is 20 pmol.

Although truncation reduced intrinsic affinity (Fig. 2D–F), dot blot/kit performance is also format-dependent. In the QD–DNAMS format, Aptamer 1e showed higher signal intensity than Aptamer 1 at the same aptamer input amount (Supplementary Fig. 5 C and D), likely due to reporter accessibility/steric effects; therefore, Aptamer 1e was selected and optimized to maximize signal-to-noise.

Following validation of the dot blot analysis setup for detecting the USE1 protein with STA-QD, we optimized the use of biotinylated Aptamer 1e for effective detection of the USE1 protein with STA-QD DNAMS using the a forementioned method. Consistently, 20 pmol of biotinylated Aptamer 1e was identified as the optimal condition for detecting USE1 (Fig. 5C). Notably, even at 50 pmol of biotinylated Aptamer 1e, the intensity difference with BSA was significant, unlike with STA-QD. This discrepancy might be attributed to the larger size of STA-QD DNAMS and the numerical advantage of multiple STA-QD conjugations on DNAMS, which reduce off-target effects and enhance signal amplification.

Using the optimized conditions from previous experiments, we conducted dot blot assays to determine the limit of detection (LOD) for the USE1 protein with STA-QD DNAMS. Aptamer 1e and STA-QD DNAMS were used to detect the USE1 protein at concentrations of 0.05, 0.5, 5, 50, and 500 nM in a dose-dependent manner (Fig. 5D). The LOD for the USE1 protein was determined to be 10 nM, as the intensity difference was insignificant below this concentration. We also examined storage stability of the QD–DNAMS biosensing reagents/kit performance. After 7 days of storage with 10% FBS, dot blot performance was largely maintained at 4 °C, partially preserved at −20 °C (59.4 ± 22.2% of the 4 °C 7-day condition), and markedly decreased at 25 °C (28.1 ± 2.4%) (Supplementary Fig. 7B). Representative dot blot images further showed clear discrimination between USE1 and BSA under 4 °C and −20 °C storage, whereas 25 °C storage reduced contrast and increased background (Supplementary Fig. 7 C), supporting 4 °C/−20 °C as recommended storage conditions.

The detection capability of the dot blot assays was also assessed in 15 lung cancer tissues with elevated USE1 protein levels and their paired adjacent non-tumoral lung tissues (Fig. 5E) and all samples showed higher fluorescent intensity compared to their adjacent non-tumor tissues (Figs. 5F-G).

In conclusion, the combination of biotinylated aptamer models and STA-QD DNAMS provides a sensitive and reliable platform for detecting the USE1 protein with lung cancer samples.

Fluorescence biosensing kit for the targeting of USE1 via Aptamer 1e detection

Fluorescence-based biosensing is an effective method for detecting cancer biomarkers [36]. To facilitate the clinical application of this technology for biomarker detection in lung cancer, we developed a fluorescence biosensing kit using STA-QD DNAMS targeting USE1 with Aptamer 1e. A PVDF membrane was integrated into a 3D-printed scaffold with two wells: control and test lines and the fluorescence signal were measured using a UV lamp (Supplementary Fig. 6 A).

The control setup of the kit contained streptavidin as a positive control, whereas the test setup included bovine serum albumin (BSA) as a negative control and USE1 as the target protein. A fluorescence signal was observed in the test line containing the USE1 protein, indicating successful detection by Aptamer 1e through STA-QD DNAMS. In contrast, the BSA loaded test line showed only minimal fluorescence, confirming the specificity of the detection system. The signal differences were readily visible to the naked eye (Fig. 6A). To obtain more quantitative results, the mean fluorescence intensity (MFI) was measured using ImageJ software. As expected, the USE1 test line had an intensity comparable to the control line, while the BSA test line exhibited negligible signal (Fig. 6A). To assess the detection capability of USE1 protein in clinical specimens we applied the biosensing kit to human lung cancer tissue. The test line loaded with tumor lysates exhibited a markedly stronger fluorescence signal compared to that of the paired adjacent non-tumoral lung tissues (Fig. 6B, Supplementary Fig. 6B and C), consistent with elevated USE1 expression in tumor samples.

Fig. 6.

Fluorescence biosensing kit for USE1 targeting aptamer 1e detection. (A) Representative fluorescence images of the biosensing kit incorporating aptamer 1e conjugated with STA–QD DNAMS. The control line was loaded with streptavidin as a positive control, while the test lines were loaded with either bovine serum albumin (BSA) as a negative control or recombinant USE1 protein as the target. (B) Representative fluorescence detection in human lung cancer tissue samples. The test line was loaded with tumor lysates, and the control line was loaded with lysates from paired adjacent non-tumorous lung tissue. (C) MFI quantification across the independent prospective cohort (n = 30 pairs; #31–60) comparing tumor versus adjacent non-tumor tissues. (D) ROC curve (95% confidence intervals) for classifying tumor versus adjacent non-tumor tissues based on MFI values from the kit in the independent prospective cohort. The optimal cutoff (50.9) was determined by the Youden index (sensitivity 86.7%, specificity 93.3%)

To assess clinical applicability, we applied the biosensing kit to an independent prospective cohort of 30 paired lung tumor and adjacent non-tumor tissues (#31–60). Tumor lysates consistently produced higher fluorescence signals than paired adjacent tissues (Fig. 6C). ROC analysis based on MFI values identified an optimal cutoff of 50.9 (Youden index), yielding sensitivity 86.7% (95% CI 70.32–94.69%) and specificity 93.3% (95% CI 78.68–98.82%) (Fig. 6D; AUC = 0.961, p < 0.001). These results support the potential utility of the DNAMS–STA–QD platform as a rapid, visual, and clinically applicable method for USE1 detection as a biomarker in lung cancer tissue samples.

Discussion

To expend access of more efficacious and less toxic therapies to patients with lung cancer, the advancement of biomarker-guided precision medicine has emerged as a major priority in lung cancer treatment [4]. A central challenge in this area is the identification of clinically actionable biomarkers grounded in tumor pathogenesis, along with the development of robust tools for their accurate detection. In this study, we sought to address these challenges by developing a detection system targeting UBA6-specific E2 conjugating enzyme 1 (USE1)—a previously validated biomarker implicated in lung cancer progression and treatment target [19, 20].

We successfully identified and validated a DNA aptamer with high affinity and specificity for USE1. The binding properties of the aptamers were confirmed using multiple methods, including ELONA, immunoblotting, fluorescence quenching assays, and AI-assisted, deep learning–based in-silico structural modeling. Assays were carried out with purified USE1 protein, lung cancer cell lines, and patient-derived lung cancer tissue samples. To enhance detection sensitivity and allow for easy visualization, we engineered a fluorescence-based sensing platform incorporating self-assembled, biotinylated DNAMS conjugated with STA-QD, ultimately leading to the creation of a prototype detection kit targeting USE1 in lung cancer tissues.

Unlike conventional antibodies, aptamers are non-immunogenic, chemically synthesized, and easily modified or conjugated with various therapeutic agents or carriers [37]. These properties have facilitated aptamer adoption in diagnostics and therapy [22, 38], and support their application in the present study.

Since the introduction of Cell-SELEX for targeting lung cancer cells [39], several aptamers have been identified for the detection of lung cancer biomarkers using lung cancer cell lines, tissue lysates, or patient blood samples. Zhou et al. [40] identified the aptamer C12 targeting high-density lipoprotein binding protein (HDLBP), a novel biomarker for small-cell lung cancer (SCLC); knockdown of HDLBP suppressed tumor growth and metastasis both in vitro and in vivo. More recently, the AP-9R aptamer was developed to target cancer stem cells of lung cancer, with annexin A2 identified as the target protein. The expression of annexin A2 was associated with stemness, metastasis, and poor clinical outcomes in lung cancer, suggesting its role as a cancer stem cell (CSC) marker and regulator [41].

While these studies demonstrated the feasibility of aptamer-based biomarker detection, most cell-based SELEX-derived aptamers lack initial target specificity and may cross-react with unrelated proteins or cancer types. In contrast, our study began with a well-characterized, mechanistically validated biomarker, USE1 [20], and employed SELEX to generate aptamers with high specificity. By direct targeting a purified, disease-relevant USE1 protein, our approach enhances diagnostic precision and supports integration with precision medicine for lung cancer.

Among the three aptamers developed in this study, Aptamer 1 showed the strongest USE1 binding affinity (Fig. 1H; Kd = 10.0 ± 2.76 nM). This result suggested that its folded structure presents the USE1-binding motif in a more favorable orientation that others. Based on this screening result, Aptamer 1 was selected as the parent sequence for truncation and downstream optimization.

To reduce oligonucleotide synthesis burden and enable a more practical probe format, Aptamer 1 was truncated to Aptamer 1e. While Aptamer 1e retained specific USE1 recognition, truncation reduced its apparent affinity in ELONA and therefore required a higher working concentration to achieve robust signal-to-background ratios (Fig. 2D-F).

To support structural hypotheses for the experimentally observed binding behavior after truncation, we applied AI-assisted, deep learning–based modeling (Protenix) post hoc. Recent computational and AI-assisted approaches in aptamer research span multiple stages of the discovery pipeline, ranging from sequence/NGS-guided prioritization and deep learning–based aptamer discovery/refinement to sequence–binding prediction frameworks [42–44]. More recently, AlphaFold3 has enabled joint modeling of biomolecular complexes including proteins and nucleic acids, and initial studies have evaluated its utility for direct modeling of DNA/RNA aptamers and aptamer–protein complexes to generate structural hypotheses [33, 45]. Our predicted complex models provided supportive hypotheses that may explain (i) Aptamer 1e–USE1 recognition and the putative binding mode near the USE1 ubiquitin-binding region (Fig. 3A–F) and (ii) a suggested preference for USE1 over other E2 enzymes within the tested in silico panel (Supplementary Fig. 8A–D; Supplementary Table 2), rather than serving as the sole driver of candidate selection.

Despite recent advances, predicting nucleic acid–protein complexes remains challenging. Multiple binding modes may be plausible for a given ssDNA aptamer, and predictions can depend on input settings while not capturing the full conformational ensemble. In addition, modeling does not explicitly account for experimental context such as ionic strength, divalent cations (e.g., Mg²⁺), buffer composition, surface immobilization effects in plate-based assays, or potential multivalent interactions that can influence apparent affinity and working concentration. Finally, in the absence of experimental restraints (e.g., crosslinking-MS, mutational scanning, or high-resolution structural data), predicted interfaces should be interpreted cautiously as hypotheses. For these reasons, we emphasize that Protenix modeling complements, rather than replaces, empirical measurements and orthogonal biochemical validation.

Accordingly, specific binding and selectivity were further supported by orthogonal biochemical validation, including reciprocal pull-down assays using His-tag capture (Ni-NTA) and biotin capture (streptavidin), which consistently supported complex formation between His-USE1 and the biotinylated aptamers. Although band intensities in these assays can be influenced by capture geometry and biotin accessibility (in addition to intrinsic affinity), the generally stronger signals observed with Aptamer 1 were consistent with its higher ELONA affinity, whereas differences in streptavidin-based assays likely reflect steric/accessibility effects associated with aptamer length and folding. In addition, streptavidin-based aptamer pull-down from 293 T lysate robustly recovered USE1, whereas representative E2 enzymes tested showed absent or minimal co-precipitation (Supplementary Fig. 8E), reducing concerns regarding broad off-target binding within the E2 family, although only a representative subset of E2 enzymes was evaluated (Supplementary Fig. 8; Supplementary Table 2).

Finally, in paired clinical tissue lysates, Aptamer 1e achieved 100% sensitivity and 88.3% specificity for lung cancer detection in the pooled cohort (#1–60). These results confirm that Aptamer 1e is effective detectors of the USE1 protein not only in vitro and in lung cancer cell lines but also in actual patient cancer tissues and supporting the translational potential of USE1 aptamer probes.

To enable visualization and amplification of detection signals, we employed rolling circle amplification (RCA), a method that synthesizes long single-stranded DNA using a circular template [46]. This technique not only amplifies signal output but also generates functionally active DNA microstructures that can be further modified [47]. Self-assembled DNAMS was produced via RCA with several functionalized dNTPs, which can exhibit DNAMS fluorescence and allow the attachment of other functional moieties through strong interactions, such as chemical conjugation or biotin–streptavidin interaction [46, 47].

The analytical limit of detection (LOD) of our DNAMS–STA-QD dot-blot assay was 10 nM for USE1 protein (corresponding to ~305 ng/mL assuming ~30.5 kDa). This sensitivity is well aligned with our current application to tissue/lysate-based measurements [48, 49], where USE1 is overexpressed in tumors and where rapid, instrument-light readout is advantageous.

In comparison with conventional antibody-based assays such as ELISA, which can achieve high analytical sensitivity but require validated antibody pairs, plate readers, and longer workflows, the DNAMS–STA–QD format emphasizes a low-instrumentation workflow with optional quantitative MFI analysis and simplified reagent handling. Notably, the nM-range analytical LOD in the DNAMS–STA–QD format should be distinguished from intrinsic binding affinity (Kd) measured in ELONA. Although the truncated Aptamer 1e shows reduced apparent affinity in plate-based ELONA (low-µM Kd) relative to full-length Aptamer 1, the DNAMS–STA–QD platform enables lower practical LOD through multivalent signal amplification (multiple STA–QD per DNAMS) and improved signal-to-background characteristics. Building on this analytical performance, we constructed a detection kit combining USE1-targeting Aptamer 1e with self-assembled biotin-modified DNAMS and STA-QD conjugates. The resulting platform provides a visualizable readout with quantitative MFI analysis, supporting practical implementation of the DNAMS–STA-QD format for biomarker detection. To strengthen clinical relevance, we evaluated the kit in an independent prospective cohort (#31–60), where it consistently distinguished tumor from paired adjacent tissues.

For interpretability, key assay conditions are summarized briefly here. ELONA assays used USE1-coated wells and biotinylated aptamers with HRP/TMB readout (A450) and/or time-course monitoring (A370), and most datasets represent four independent experiments with duplicate wells per condition unless otherwise stated; the DNAMS–STA–QD kit used PVDF lines with Aptamer 1e and STA–QD DNAMS incubations followed by ImageJ-based MFI quantification (see Methods for details).

In the broader landscape of aptamer-based biosensors for lung cancer biomarkers, platforms vary widely in transduction and readout requirements [27]. Reported approaches span instrument-free or minimal-instrument visual assays (e.g., strip-type formats) as well as instrument-assisted quantitative aptasensors, each with trade-offs in workflow complexity, instrumentation needs, and matrix tolerance [28–30]. In this context, our DNAMS–STA–QD format leverages DNAMS-based signal amplification and an instrument-light workflow with optional MFI quantification to support practical measurements in tissue/lysate samples. Several aptamer-based biosensing approaches have also been reported for minimally invasive testing in blood/plasma matrices in lung cancer [50, 51], illustrating the translational interest in liquid-biopsy–type workflows. In contrast, USE1/UBE2Z is primarily an intracellular UPS-related marker, and thus a single consensus “physiological concentration” in circulation is not well established as for classical secreted plasma biomarkers. Accordingly, our current platform was optimized for tumor-derived tissue/lysate samples, where USE1 is overexpressed and assay readout can be implemented within routine workflows. Extending our platform to blood/plasma would likely require higher sensitivity and improved robustness against matrix effects, and would involve prospective specimen collection under ethical approval and standardized pre-analytical handling; thus, future work will focus on enrichment and matrix-optimized assay conditions to support blood-based detection.

From a clinical workflow perspective, the DNAMS–STA–QD kit is designed as an instrument-light assay with a visually interpretable readout and optional quantitative analysis (MFI). For tissue/lysate applications, sample preparation follows standard protein-lysis workflows (e.g., homogenization/lysis and clarification), which can typically be completed within ~1 h depending on specimen handling and batch size, followed by a short incubation/readout step. In terms of cost, the platform leverages pmol-scale aptamer input and common biochemical reagents (PVDF membrane and streptavidin–QD conjugates), suggesting feasibility for low- to moderate-cost implementation compared with instrument-intensive analytical platforms; however, a formal cost analysis will be required in future translational studies. Finally, regarding operational stability, both aptamers and the kit reagents retained substantial performance after 7-day storage at 4 °C–− 20 °C, whereas 25 °C storage reduced performance (Supplementary Fig. 7), supporting practical handling under standard cold-chain conditions.

Despite certain limitations, including a relatively small number of patient tissue samples for kit validation, and the fact that AI-assisted insights for 1e binding were explanatory rather than confirmatory, the current clinical evaluation should be interpreted as a pilot/proof-of-concept study. Larger independent cohorts (ideally multi-center) will be required to confirm generalizability and refine operating thresholds. To our knowledge, this study is the first to report the development of a USE1-specific aptamer and its integration into a fluorescence-based biosensing kit using STA-QD DNAMS. The consistency between experimental results and AI-assisted, deep learning–based structural predictions strengthens confidence in the validity of Aptamer 1e. These findings support the continued development of biomarker-driven diagnostic platforms and may contribute to the precision management of lung cancer.

Conclusion

In this study, we successfully developed and validated DNA aptamers targeting USE1, a key biomarker implicated in lung cancer pathogenesis and therapeutic targeting. Both the full-length Aptamer 1 and its truncated derivative Aptamer 1e exhibited specific binding affinity to USE1. These aptamers demonstrated strong performance in vitro assays and were able to distinguish lung cancer tissues from non-cancerous samples with high sensitivity and good specificity. Building on these findings, we engineered a novel diagnostic platform combining rolling circle amplification (RCA)-derived, biotinylated DNA microstructures with streptavidin-conjugated quantum dots (STA-QD) to detect Aptamer 1e binding via fluorescence-based visualization. The resulting detection kit offers a rapid, instrument-light method for identifying USE1 in tissue/lysate clinical samples, warranting further validation in larger cohorts.

This work also highlights the utility of AI-assisted, deep learning–based structural modeling as a complementary and explanatory tool that supports experimental findings. Future studies may extend such modeling to additional truncated variants and integrate prospective biophysical simulations, thereby advancing aptamer design pipelines that are both experimental and computational. Collectively, our approach underscores the potential of USE1-targeted aptamers as clinically actionable diagnostic probes and contributes toward biomarker-guided precision management of lung cancer.

Author contributions

K.M.J., K.H.Y., and D.J.K. performed the experiments with assistance from I.S.J.; G.D.L. and M.J.K. analyzed the data; A.R.P and W.P.I. performed the simulations with assistance from M.J.J.; PCW. L., J.B.L., W.P.I. and J.O.J. supervised the project. All authors wrote the manuscript.

Funding

This study was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MIST) (RS-2023-00208173, RS-2023-00207868) and by the Asan Institute for Life Sciences, Seoul, Republic of Korea (2023IP0114). This work was also supported by the 2024 Research Fund of the University of Seoul for Jong Bum Lee.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable requests.

Declarations

Ethics approval and consent to participate

The experiments were conducted after obtaining informed consent from patients and approval from the Institutional Review Board of Asan Medical Center (2014 − 0960). Surgically resected human lung tissues were acquired from the Asan Bio Resource Center (Resource No. 2014-20 (89)).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.