CEMs-SELEX: DNAzyme-powered de novo biomarker discovery and noninvasive cancer diagnosis

Abstract

The ability to interrogate complex biological environments without prior target knowledge is a key challenge in chemical sensing. Conventional methods are limited in discovering previously unidentified biomarkers. To overcome this, we developed CEMs-SELEX (cell extracellular mixtures–systematic evolution of ligands by exponential enrichment), a de novo selection platform that screens for functional DNAzymes responsive to disease-specific CEMs. Applied to tongue squamous cell carcinoma (TSCC), we identified C502, a DNAzyme activated by tumor secretions with high affinity (K_d = 24.49 nanomolar) and fast catalytic turnover (K_cat = 47.56 per second). Integrated multiomics and affinity pull-down identified annexin A2 (ANXA2) as its target, and molecular docking with mutagenesis elucidated the recognition mechanism. Engineering C502 into a DNA Walker enabled ultrasensitive detection of ANXA2 in clinical serum and saliva. In 80 individuals, the assay clearly differentiated patients with TSCC from healthy donors, highlighting its potential as a noninvasive diagnostic framework. Collectively, this work closes the loop from functional probe discovery to clinical assay, offering a generalizable platform for noninvasive cancer diagnostics.

CEMs-SELEX finds DNAzymes from tumor secretome, yielding ANXA2-activated C502 for noninvasive TSCC detection.

INTRODUCTION

Biomarker discovery represents a pivotal yet challenging frontier in cancer diagnostics, particularly for malignancies such as tongue squamous cell carcinoma (TSCC) that lack well-characterized molecular targets for noninvasive detection (1–5). Conventional approaches, including mass spectrometry (MS) and antibody-based assays, often require prior knowledge of candidate identities, which confines their utility to hypothesis-driven validation and limits the discovery of previously unidentified biomarkers within complex biological environments (6–9).

Functional nucleic acids, particularly DNAzymes, offer a unique chemical solution to this problem. As catalytically active DNA molecules, DNAzymes can cleave RNA substrates in a sequence-dependent manner, and through in vitro selection, they can be evolved to acquire allosteric responsiveness to specific molecular targets. This intrinsic coupling of target recognition and signal amplification makes them powerful tools for de novo biomarker discovery without prior target information (10–12). Seminal studies have been instrumental in advancing this paradigm. For instance, AptaBiD demonstrated aptamer-based biomarker discovery directly from whole-cell lysates (13), and the functional screening strategies developed by Li’s team [e.g., RNA-cleaving fluorogenic DNAzymes (RFDs)] used complex targets like secretomes (14–16), whereas the Gu team’s molecular recognition based on affinity and catalysis (MORAC) platform established a crucial link between target identification and molecular recognition through activity-based DNAzymes selection coupled with MS-based target deconvolution (17). These foundational studies firmly established the principle of moving beyond predefined targets.

However, translating this principle into a fully integrated pipeline, one that seamlessly unifies de novo DNAzyme selection from native secretomes, definitive target deconvolution, and the engineering of a clinically validated, amplification-based diagnostic system deployable in noninvasive biofluids, remains to be realized. This gap is particularly critical for cancers such as TSCC, which lack well-defined molecular signatures for noninvasive detection.

To address this gap, we introduce CEMs-SELEX (cell extracellular mixtures–systematic evolution of ligands by exponential enrichment) as an integrated platform that extends the activity-based discovery paradigm to the native tumor secretome. Applied to TSCC, this platform enables a complete diagnostic workflow: (i) the direct selection of a high-performance DNAzyme, C502, from the TSCC secretome; (ii) the identification of its specific protein target, annexin A2 (ANXA2), through integrated metabolomic and proteomic profiling and affinity pull-down; and (iii) the engineering of a C502-driven DNA Walker system for ultrasensitive detection of ANXA2 in clinical serum and saliva samples. Collectively, this work translates the secretome-based functional selection concept into a tangible diagnostic outcome, offering a generalizable and integrated pathway for noninvasive cancer diagnostics.

RESULTS

CEMs-SELEX of TSCC-specific DNAzymes

Current diagnostic methods for TSCC primarily rely on subjective visual examination and invasive biopsy, lacking noninvasive molecular tools for early screening (18, 19). To enable noninvasive early screening, we used CEMs-SELEX to isolated DNAzymes responsive to tumor secretomes, which directly target soluble biomarkers, unlike conventional Cell-SELEX that targets membrane proteins (20–22).

We used CEMs as complex targets for two key reasons: (i) They comprise a diverse array of biomolecules, including proteins and metabolites, thereby increasing the probability of identifying functional DNAzymes; and (ii) they bypass the need for prior target identification, substantially reducing both time and cost. Using CEMs derived from TSCC cells improves selection efficiency while ensuring clinical relevance.

The in vitro selection process is summarized in Fig. 1A and fig. S1. A DNA library (~10¹⁴ sequences) ligated to a phosphorylated substrate (P-Sub; table S1) was first subjected to negative selection against CEMs from normal cell lines (HOK16B and HOEC). Uncleaved sequences were then incubated with pooled CEMs from TSCC cell lines (CAL27, SCC4, and SCC9) for positive selection. Cleaved products were purified, polymerase chain reaction (PCR) amplified, and religated for subsequent rounds (figs. S2 to S5 and tables S2 to S4). Selection stringency was progressively increased over 12 rounds by reducing incubation time from overnight at 4°C to 2 hours at room temperature.

Fig. 1. Screening of RNA-cleaving fluorogenic DNAzymes responsive to TSCC secretomes.

(A) Workflow of the CEMs-SELEX process for functional DNAzyme selection. Key steps include (i) ligation of P-Sub to the DNA library, (ii) negative selection against CEMs from normal cell lines (HOK16B and HOEC), (iii) positive selection against pooled CEMs from TSCC cell lines (CAL27, SCC4, and SCC9), (iv) PCR amplification, and (v) high-throughput sequencing. (B) Schematic of the selection strategy. CEMs from live cells are fractionated and profiled; fractions inducing catalytic cleavage (visualized as a band shift on d-PAGE) are enriched over multiple rounds. (C) Enrichment of cleavage activity across selection rounds. DNA pools from rounds R1 to R12 were incubated with negative or positive CEMs for 2 hours and analyzed by d-PAGE. M, fluorescent DNA marker. Cleavage percentage (Clv%) was calculated as: Clv% = (Int[clv]/6)/(Int[clv]/6 + Int[unclv]) × 100%, where Int[clv] and Int[unclv] represent fluorescence intensities of cleaved and uncleaved bands, respectively. The factor of 6 accounts for fluorescence enhancement upon cleavage. nt, nucleotides. (D) Schematic of the microplate-based cleavage assay for rapid monitoring of selection progress. (E) Representative fluorescence images (top) and corresponding fluorescence intensity quantification (bottom) across selection rounds. Data are presented as means ± SEM (n = 3 independent experiments).

As illustrated in Fig. 1B, functional DNAzymes activated by disease-relevant components were enriched over multiple rounds, with cleavage activity monitored by denaturing polyacrylamide gel electrophoresis (d-PAGE). After 12 rounds, the library showed enriched cleavage activity, peaking at R11 (23.9%) with a slight drop at R12, indicating convergence (Fig. 1C and figs. S6 and S7). Fluorescence intensity increased over rounds in microplate assays (Fig. 1, D and E). The R11 pool exhibited high specificity for TSCC-derived CEMs (CAL27 and SCC4) compared to other cancer cell lines and buffer controls (fig. S8). On the basis of its stable cleavage activity and excellent selectivity, the R11 pool was selected for cloning and sequencing to identify individual DNAzyme candidates.

Multitiered selection of an optimal TSCC-specific DNAzyme

High-throughput sequencing of the final selection pools identified nine dominant sequence clusters (C1 to C9; table S5). To evaluate functional performance beyond sequence abundance, we implemented a multitiered screening strategy. Initial screening showed that all DNAzymes except C1 exhibited significant cleavage activity in positive CEMs (fig. S9).

To streamline validation, we developed a magnetic bead–based flow cytometry assay that offers higher throughput and sensitivity than traditional gel-based methods (Fig. 2A). This assay revealed that C3 and C7 showed nonspecific cleavage, whereas C2, C5, and C9 were specifically activated by positive CEMs (Fig. 2B). Near-infrared spectroscopy confirmed their specificity, excluding C3 and C7 (fig. S10). The three lead DNAzymes (C2, C5, and C9) exhibited distinct sequence biases toward G- or C-richness (Fig. 2C). Flow cytometric quantification verified that all three bound significantly better to positive CEMs than to negative CEMs or blank controls (fig. S11). However, comparative analysis showed that C2 yielded a substantially weaker fluorescence response than C5 and C9, with the latter two demonstrating superior dose-dependent activity (Fig. 2D). Binding characterization via single-point binding curves yielded a dissociation constant (K_d) of 24.5 nM for C5, significantly lower than those of C2 and C9 (Fig. 2E), confirming its enhanced target-binding capability.

Fig. 2. Selection, characterization, and optimization of TSCC-specific DNAzymes.

(A) Schematic of the magnetic bead–based flow cytometry cleavage assay. (i) Probe preparation: Amino-modified DNAzymes were ligated to fluorescent P-Sub and purified. (ii) Immobilization: The ligated products were covalently conjugated to carboxylated magnetic beads. (iii) Cleavage assay: Functionalized beads were incubated with CEMs, washed, and analyzed by flow cytometry to quantify cleavage activity. (B) Screening of DNAzyme candidates C1 to C9. Cleavage activity of bead-immobilized candidates in the presence of negative or positive CEMs. The green peak indicates high-affinity binders. (C) Sequences of the three lead DNAzymes (C2, C5, and C9). F, FAM-labeled deoxythymidine; R, adenine ribonucleotide; Q, BHQ1-quenched deoxythymidine. (D) Comparison of lead DNAzyme activity. Fluorescence intensity of C2, C5, and C9 after 1 hour of incubation with positive CEMs, measured by flow cytometry. Data are presented as means ± SEM (n = 3), ***P < 0.005 (two-tailed unpaired Student’s t test). (E) Binding affinity. Single-point binding curves and dissociation constants (K_d) for C2, C5, and C9. (F) Structural optimization of C5. The secondary structure was refined using M-fold. Optimization strategies included removal of unpaired regions, incorporating of conserved motifs from C2 and C9, retention of essential hairpins, and selective base substitutions, yielding variants C501 and C502. (G) Catalytic constant of the optimized DNAzyme. Sequence and catalytic activity of C502 (K_cat = 47.56 s⁻¹). FP, forward primer; RP, reverse primer.

Structural optimization and kinetic characterization of DNAzyme C5

With the high-affinity DNAzyme C5 identified, we sought to engineer it into a more efficient catalytic probe for diagnostic applications. Using M-fold simulations (23, 24), we engineered C5 into C502 by trimming termini, incorporating conserved motifs from C2/C9, and reinforcing hairpins, yielding a more stable structure (ΔG = −8.20 kcal/mol; Fig. 2F). Under optimized conditions (37°C, 20 mM Mn²⁺, and 100 nM C502), C502 showed enhanced cleavage activity with high specificity (figs. S12 to S16). It was specifically activated by TSCC CEMs but not by normal or other cancer cells (fig. S17).

Last, we characterized the catalytic efficiency of C502 using Michaelis-Menten kinetics. With C502 fixed at 100 nM and Sub concentration ranging from 1 to 10 μM, initial rates increased with substrate load (fig. S18). Fitting initial velocities from the first 60 s of reactions (fig. S19) yielded a catalytic constant (K_cat) of 47.56 s⁻¹, indicating that each C502 molecule cleaves ~47 substrate molecules per second (Fig. 2G). These results confirm the high catalytic proficiency of C502 and its suitability for sensitive molecular detection.

Functional specificity of C502 and metabolic profiling of the TSCC secretome

Functional characterization of C502 across TSCC and normal cell lines confirmed its pronounced specificity for tumor-derived CEMs. Using bead-based flow cytometry, real-time kinetics, and steady-state fluorescence spectroscopy, we found that C502 was strongly activated by CEMs from TSCC lines (CAL27, SCC4, and SCC9) but not by those from normal controls (HOK16B and HOEC). Notably, SCC4-derived CEMs induced the most robust cleavage response, suggesting a higher abundance or accessibility of the target within the SCC4 secretome (Fig. 3, A to C).

Fig. 3. Functional specificity of C502 and metabolic profiling of the TSCC secretome.

(A) Flow cytometry analysis of the C502 cleavage activity. C502 immobilized on magnetic beads was incubated with CEMs from TSCC cell lines (CAL27, SCC4, and SCC9) and normal cell lines (HOK16B and HOEC). (B) Real-time fluorescence kinetics of C502 (100 nM) incubated with the indicated CEMs over 2400 s. (C) Endpoint fluorescence spectra after 1 hour of incubation, confirming the highest C502 activation by SCC4-derived CEMs. Statistical significance compared to other lines is indicated (****P < 0.0001). (D) Workflow of the untargeted metabolomics analysis of TSCC and normal CEMs. LC-MS, liquid chromatography–mass spectrometry. (E) PCA score plot showing clear separation between TSCC and normal CEMs. Percentages represent variance explained by each principal component. (F) VIP score plot (left) and volcano plot (right) highlighting the top 20 differentially abundant metabolites (red: up-regulated; green: down-regulated). (G) Hierarchical clustering heatmap of the top 20 metabolites after z-score normalization. Metabolite classes are numbered: 1, organic acid and its derivatives; 2, amino acids and their metabolites; 3, alcohols and amines; 4, nucleotides and their metabolites; 5, heterocyclic compounds; 6, alkaloids; and 7, benzene and its substituted derivatives. (H) Relative abundance of selected metabolites across TSCC cell lines. (I) Schematic summary of metabolic reprogramming in the TSCC secretome, highlighting dysregulation in glycolysis, amino acid metabolism, antioxidant response, TCA cycle, and intercellular signaling. ATP, adenosine triphosphate; CoA, coenzyme A; ROS, reactive oxygen species.

To explore the molecular basis of this specificity, we performed untargeted metabolomic profiling of TSCC secretome (Fig. 3D). Principal components analysis (PCA) revealed a clear separation between TSCC and normal CEMs, indicating distinct metabolic phenotypes (Fig. 3E). The top 20 differentially abundant metabolites were identified by variable importance in projection (VIP) scores and volcano plot analysis (Fig. 3F), and hierarchical clustering confirmed their consistent dysregulation in TSCC (Fig. 3G). Quantitative comparison across TSCC lines showed that SCC4 CEMs contained the highest levels of several tumor-associated metabolites (fig. S20 and Fig. 3H), consistent with the superior activation of C502. Pathway analysis revealed significant enrichment in glycolysis/gluconeogenesis (e.g., lactate), amino acid metabolism (e.g., l-allothreonine), and fatty acid oxidation intermediates (e.g., valeric acid), along with up-regulation of the phenolic antioxidant 4-hydroxy-3,5-dimethoxybenzaldehyde, indicative of enhanced glycolytic flux, metabolic reprogramming, and oxidative stress (25–31).

Collectively, these data delineate an SCC4-specific metabolic microenvironment characterized by accelerated glycolysis, perturbed amino acid and lipid metabolism, and an activated antioxidant response, features suggestive of tricarboxylic acid (TCA) cycle dysfunction and altered intercellular signaling (Fig. 3I). This distinct metabolic landscape not only provides a mechanistic basis for the high affinity and specificity of C502 but also suggests that this DNAzyme can serve as a chemical probe for the TSCC secretome, reporting on the underlying metabolic state of the tumor. Having established this functional and metabolic signature, we next sought to identify the specific molecular entity within the secretome that directly activates C502.

Identification of ANXA2 as the functional target of C502

Conventional omics methods often lack the sensitivity to enrich low-abundance targets in native environments. Allosteric DNAzymes overcome this limitation: Conformational switching upon target binding enables both specific capture and direct signal readout, with high discrimination against interferents (32, 33). To identify the molecular target of C502 in TSCC, we performed a pull-down assay using biotin-labeled C502 and streptavidin beads (Fig. 4A), followed by SDS-PAGE and MS. Enzyme perturbation assays revealed that C502 cleavage activity was markedly reduced upon treatment of TSCC CEMs with trypsin or proteinase K, indicating that the target is proteinaceous (Fig. 4B and fig. S21).

Fig. 4. Identification of ANXA2 as the functional target of C502.

(A) Schematic of the pull-down assay. (B) Protease sensitivity of the C502 target. Cleavage activity of C502 in response to TSCC CEMs treated with trypsin, proteinase K, or control buffer. Data are presented as means ± SEM (n = 3), ****P < 0.0001 (two-tailed unpaired Student’s t test); n.s., not significant. (C) SDS-PAGE analysis of pulled-down proteins from TSCC CEMs: Lane 1: total secreted proteins from positive CEMs; lanes 2 and 3, 4 and 5, and 6 and 7: proteins pulled down from CAL27, SCC4, and SCC9 CEMs, respectively; lane 8: pull-down with a random sequence; lane 9: blank control. Arrows indicate bands excised for MS analysis. (D) List of candidate proteins identified by MS (≥2 unique peptides, 25 to 40 kDa). (E) Heatmap of mRNA expression of 11 candidates in TSCC and normal cell lines. (F) Functional validation. Fluorescence kinetics of C502 (100 nM) incubated with recombinant ANXA2, LDHA, or HSP90AB1 (each at 100 nM) over 1800 s. (G) qPCR analysis of ANXA2 mRNA levels in the indicated cell lines. Data are presented as means ± SEM (n = 3), ****P < 0.0001 (two-tailed unpaired Student’s t test). (H) Western blot analysis of ANXA2 protein levels. β-Tubulin as a loading control. (I) Immunofluorescence showing the subcellular localization of ANXA2 (red) in TSCC and normal cells. Nuclei stained with 4′,6-diamidino-2-phenylindole (DAPI; blue). Scale bars, 10 μm.

As shown in Fig. 4C, SDS-PAGE analysis of pulled-down proteins from TSCC CEMs revealed specific bands in the range of 25 to 42 kDa, with the strongest enrichment in SCC4-derived samples (lanes 2 to 7). No comparable bands were detected in control pull-downs using a random DNA sequence or beads alone (lanes 8 and 9). Specific bands were excised and analyzed by MS (fig. S22A). Applying stringent filtering criteria (≥2 unique peptides, molecular weight: 25 to 42 kDa, and putative secretory function), we identified 11 high-confidence candidate proteins (Fig. 4D, fig. S22B, and table S6). Subsequent quantitative PCR (qPCR) analysis showed that ANXA2, LDHA, and HSP90AB1 were the most highly expressed among the candidates (Fig. 4E, fig. S23, and table S7), and they were prioritized for further validation.

To validate functional activation, we incubated C502 with each purified recombinant protein. Only ANXA2 induced a significant increase in fluorescence intensity (P = 0.0096; fig. S24) and markedly accelerated the cleavage rate of C502 in real-time kinetic assays, whereas control proteins showed negligible effects (Fig. 4F and fig. S25). These results establish ANXA2 as a functional activator of C502. To confirm that C502 also recognizes endogenous ANXA2 in its native secreted form, we performed small interfering RNA (siRNA)–mediated knockdown of ANXA2 in positive cells. Conditioned media from knockdown cells induced significantly lower C502 activation compared to controls (fig. S26), confirming specific recognition of endogenous ANXA2 in the TSCC secretome.

Having established functional dependence on ANXA2, we next verified its consistent expression across the three TSCC cell lines used in CEMs-SELEX. Reverse transcription qPCR (RT-qPCR) and Western blotting confirmed that ANXA2 mRNA and protein levels were significantly elevated in CAL27, SCC4, and SCC9 cells compared to normal cell lines (HOK16B and HOEC), with the highest expression in SCC4 cells (Fig. 4, G and H, and fig. S27). Immunofluorescence staining further revealed predominant cytoplasmic and membrane localization of ANXA2 in TSCC cells, with only weak background staining in normal controls (Fig. 4I and fig. S28). These data collectively demonstrate that ANXA2 is a consistently expressed, representative functional target for TSCC, rather than a cell line–specific marker.

Structural basis of the C502-ANXA2 interaction

To decipher the mechanism by which ANXA2 binding allosterically activates C502, we probed the structure-function relationship of its aptamer domain. Systematic segmental mutagenesis across the central N₄₀ loop identified two discrete regions, b (nucleotides 6 to 10) and e (nucleotides 21 to 25), as critical for catalytic activation; mutation of these regions reduced cleavage activity by ~50% (Fig. 5, A and B, and fig. S29). Single-nucleotide mutagenesis within these regions (Fig. 5C) followed by real-time kinetic monitoring (fig. S30) revealed that all mutants in region b exhibited substantially reduced reaction rates over 2400 s, with signals dropping to nearly one-eighth of wild-type (WT) C502; mutant b-5 matched the kinetic profile of the blank control. In region e, mutant e-1 showed the most pronounced loss of activity. Consistent with these results, the characteristic fluorescence shift upon C502-ANXA2 binding was absent in these key mutants (Fig. 5D), confirming that the “ATCCG” motif is essential for target recognition.

Fig. 5. The “ATCCG” motif in C502 is essential for target recognition and cleavage.

(A) Mutational strategy within the N₄₀ region of C502. Each consecutive 5-nt segment (a to h) was replaced with “TTTTT.” (B) Cleavage activity of segment mutants. Fluorescence intensity of C502 and mutants a to h after 1 hour of incubation with or without 100 nM ANXA2. Data are presented as means ± SEM (target: n = 5; no target: n = 3); statistical significance was assessed by two-tailed unpaired Student’s t test. (C) Single-nucleotide mutations within regions b and e (mutants b-1 to b-5 and e-1 to e-5). (D) Cleavage activity of single-nucleotide mutants. Fluorescence intensity after 1 hour of incubation with 100 nM ANXA2 (means ± SD, n = 4). (E) Molecular docking model of the C502-ANXA2 interaction. Key nucleotides (purple) and amino acid residues (yellow) are shown; hydrogen bonds are indicated by dashed lines. (F) Binding free energies (ΔG_bind) of WT and mutant ANXA2 complexes with C502, calculated from 500-ns molecular dynamics (MD) simulations. Data are presented as means ± SEM from trajectory block averaging. (G) Hydrogen bond occupancy between C502 and ANXA2 over 500-ns MD simulations. The average number of hydrogen bonds maintained during the trajectory is shown for WT and mutant complexes. Data are means ± SD. (H) Representative structural snapshots at the end of 500-ns MD simulations. The WT complex maintains stable interfacial contacts, whereas Y188A and R231A mutants show the pronounced detachment of C502 from the binding pocket; S305A retains partial contact.

Molecular docking suggested that the “ATCCG” motif docks within a shallow hydrophobic pocket on ANXA2, stabilized by a network of specific contacts involving residues Tyr¹⁸⁸, Arg²³¹, and Ser³⁰⁵ (Fig. 5E). To further validate the predicted binding interface from the protein side, we performed in silico alanine scanning followed by 500-ns explicit-solvent molecular dynamics simulations on the WT ANXA2-C502 complex and three key interfacial mutants: Try¹⁸⁸→Ala(Y188A), Arg²³¹→Ala(R231A), and Ser³⁰⁵→Ala(S305A). As shown in Fig. 5F, binding free energy calculations revealed that the WT complex exhibited the most favorable binding (ΔG = −53.63 ± 0.35 kcal/mol), whereas Y188A (−1.29 ± 1.53 kcal/mol) and R231A (3.50 ± 0.28 kcal/mol) nearly abolished binding, and S305A (−24.32 ± 0.29 kcal/mol) retained only partial affinity. Consistently, hydrogen bond occupancy analysis (Fig. 5G) showed that the WT complex maintained a significantly higher number of protein-DNA hydrogen bonds throughout the simulation, whereas all three mutants exhibited reduced hydrogen bonding, indicating disruption of the interfacial contact network. Representative snapshots at the end of the simulations (Fig. 5H) illustrated that the WT complex preserved the initial binding mode, whereas C502 substantially deviated from the binding pocket in the T188A and R231A mutants and, to a lesser extent, in the S305A mutant. Detailed analyses of structural stability [RMSD (root mean square deviation), RMSF (root mean square fluctuation), R_g (radius of gyration), and SASA (solvent-accessible surface area)] and full trajectory snapshots are provided in figs. S31 and S32.

Notably, these protein-side mutagenesis results are in excellent agreement with our experimental DNA-side mutagenesis data: Mutations that disrupt the “ATCCG” motif in C502 lead to loss of catalytic activity, and mutations at the corresponding interfacial residues in ANXA2 similarly abolish complex stability. This cross-validation strongly supports the proposed recognition interface as the functional allosteric switch for C502 activation.

Construction and optimization of the DNA Walker

To engineer a robust diagnostic device, we assembled a catalytic DNA Walker on gold nanoparticles (AuNPs). This design integrates two principles: (i) fluorescence quenching via fluorescence resonance energy transfer (FRET) between the AuNP and fluorescent dye 6-carboxyfluorescein (FAM)-labeled substrate to minimize background and (ii) processive, autonomous catalysis on a two-dimensional (2D) scaffold for signal amplification (34–36). As illustrated in Fig. 6A, the DNA Walker operates via a “step-and-cleave” cycle (table S8): Upon ANXA2 activation, C502 cleaves its adjacent substrate, releasing a fluorescent fragment and diffusing to cleave subsequent substrates, enabling a single molecule to turnover hundreds of targets for ultrasensitive detection (37–40). The AuNP scaffold confers nuclease resistance and confines reaction components to favor high local concentration and efficient catalysis. Successful conjugation was confirmed by ultraviolet-visible (UV-vis) spectroscopy, showing a red shift in the AuNP absorption peak from 520 to 526 nm (fig. S33), consistent with DNA attachment.

Fig. 6. Application of the DNA Walker for ANXA2 detection in clinical samples.

(A) Schematic of the DNA Walker design. C502 and FAM-labeled Sub were covalently immobilized onto AuNPs via a 5′ thiol modification (HS-C6 linker). A 20-nt poly(T) spacer was appended to the 3′ end of C502, and a 10-nt poly(T) spacer was appended to the 5′ end of the Sub. Sequences are provided in table S5. (B) Optimization of AuNP:C502:Sub molar ratio by fluorescence kinetics. Optimal ratio: 1:100:1200. (C) Analytical performance. Calibration curve of the fluorescence intensity versus ANXA2 concentration (0 to 200 pM) using 500 pM DNA Walker (means ± SEM, n = 3). (D) Detection of ANXA2 spiked into human serum (means ± SEM, n = 6). (E) Spike-and-recovery rates of ANXA2 in serum. (F) Workflow of the clinical sample collection and detection. (G) DNA Walker signals in serum from patients with TSCC (red, n = 43) and healthy donors (yellow, n = 37). Data are presented as mean, ****P < 0.0001 (two-tailed unpaired Student’s t test). (H) Confusion matrix of the diagnostic performance (43 TSCC and 37 healthy). (I) Specificity of the DNA Walker against benign oral inflammatory conditions in serum. Fluorescence signals of the DNA Walker in serum samples from patients with TSCC (n = 10), patients with inflammatory conditions [periodontitis and oral leukoplakia] (n = 10), and healthy donors (n = 10). Data are presented as mean (n = 3); statistical significance was assessed by two-tailed unpaired Student’s t test. (J) Correlation of signals between paired serum and saliva samples (red: TSCC; yellow: healthy). Correlation between serum and saliva was high in both groups.

Systematic optimization followed. The optimal molar ratio of AuNP:C502:Sub was 1:100:1200, balancing catalytic turnover and background (Fig. 6B and figs. S34 and S35). Mn²⁺ concentration was optimized to 10 mM (fig. S36). Fluorescence intensity correlated linearly with DNA Walker concentration from 0 to 1 nM (R² = 0.9966), and 500 pM was selected as optimal (fig. S37). The assay exhibited a linear response to ANXA2 from 0 to 200 pM, with a limit of detection of 1.98 pM (3σ/S), demonstrating high sensitivity suitable for TSCC biomarker detection (Fig. 6C and figs. S34 and S35).

To evaluate clinical potential, we first examined DNA Walker stability. The DNA Walker exhibited superior resistance to DNase I and fetal bovine serum (FBS) compared to unprotected single-stranded DNA (ssDNA), attributed to steric protection by the high-density DNA assembly on the AuNP surface (fig. S39). Given that human serum and saliva contain more complex nuclease/protease mixtures, we further assessed stability in these matrices. Samples were incubated at 37°C for up to 24 hours, with fluorescence measured at 0, 1, 2, 4, 8, 12, and 24 hours. As shown in fig. S40, ssDNA in human serum exhibited a sharp increase at 1 hour followed by rapid decline, retaining only ~26% of initial fluorescence by 24 hours. In contrast, the DNA Walker maintained stable fluorescence throughout, with ~93% signal retention in serum and ~95% in saliva, confirming effective protection by the AuNP scaffold.

Having confirmed stability, we evaluated analytical performance in complex matrices. Using pooled human serum diluted to various concentrations (10 to 100%; fig. S41), we spiked in recombinant ANXA2 (0 to 200 pM). The DNA Walker exhibited a dose-dependent fluorescence increase across all serum concentrations with minimal background (Fig. 6D). Spike-and-recovery experiments showed high recovery rates (95.0 to 126.0%) and low relative standard deviations (RSDs) (2.2 to 4.2%), confirming accuracy and reliability (Fig. 6E). Reproducibility was assessed using three serum samples covering low, medium, and high ANXA2 levels. Intra-assay coefficients of variation (CVs) (six replicates per sample) were 3.43, 6.56, and 4.12%; inter-assay CVs (3 consecutive days, duplicate measurements) were 2.36, 4.00, and 2.09% (table S9), meeting clinical diagnostic requirements.

To benchmark the DNA Walker against a gold-standard method, we performed parallel spike-and-recovery experiments using a commercial ANXA2 ELISA kit. As shown in fig. S42, the enzyme-linked immunosorbent assay (ELISA) standard curve exhibited good linearity from 0 to 5000 pg/ml (R² = 0.9901) but deviated substantially at higher concentrations (0 to 10,000 pg/ml, R² = 0.7603), with a hook effect requiring sample dilution above 5000 pg/ml. Within its linear range, ELISA showed excellent recovery (93.3 to 101.2%) and low RSDs (0.15 to 2.39%). In contrast, the DNA Walker demonstrated a broader linear range (0 to 200 pM, extending to 10 nM in buffer) without hook effect, achieving comparable recovery (95.0 to 126.0%) and precision (RSD: 2.2 to 4.2%) with a simpler workflow (≤1 hour versus 4 to 5 hours). These results underscore the advantages of the DNA Walker for rapid and reliable ANXA2 detection.

Diagnostic performance in clinical samples

We next applied the DNA Walker to clinical samples, including serum from 43 patients with TSCC, 10 patients with benign inflammatory conditions, and 37 healthy donors (Fig. 6F). Patient-derived samples produced significantly higher signals, with some exceeding control levels by more than 10-fold (P < 0.0001; Fig. 6G). Receiver operating characteristic (ROC) analysis based on patients with TSCC and healthy donors yielded an area under the curve (AUC) of 1.000 [95% confidence interval (CI): 1.000 to 1.000], indicating perfect discrimination in this pilot cohort (fig. S43). Using the Youden index, the optimal threshold was set at 694.6 arbitrary units (a.u.), achieving 100% sensitivity (43/43) and 100% specificity (37/37) (Fig. 6H and table S10).

To benchmark our platform against a gold-standard method, we analyzed the same serum samples using a commercial ANXA2 ELISA kit (fig. S44). ROC analysis of the ELISA data yielded an AUC of 0.942 (95% CI: 0.892 to 0.992), with an optimal threshold of 819.29 pg/ml (Youden index). At this threshold, the ELISA achieved a sensitivity of 95.3% (41/43), a specificity of 89.2% (33/37), and an accuracy of 92.5% (74/80). The precision (positive predictive value) was 91.1% (41/45). Notably, two TSCC samples with low ANXA2 concentrations (581.57 and 461.71 pg/ml) were missed by ELISA but correctly detected by the DNA Walker, whereas four healthy individuals with elevated ANXA2 levels (up to 1307 pg/ml) were falsely classified as positive by ELISA but correctly identified as negative by the DNA Walker (figs. S45 and S46 and table S11). These discrepancies likely reflect the fundamental difference between measuring protein abundance (ELISA) and functional activity (DNA Walker), with the latter potentially better capturing tumor-relevant biological states.

Stratified analysis by tumor stage using retrospectively retrieved clinical records showed that both early-stage (T1/T2, n = 24) and late-stage (T3/T4, n = 15) patients were clearly distinguishable from healthy donors, with all early-stage samples exceeding the healthy control threshold (fig. S47). No significant difference in fluorescence signals was observed between the two stage groups, confirming that C502 detects TSCC even at early disease stages. To assess clinical specificity, we evaluated the ability of the DNA Walker to distinguish TSCC from benign oral inflammatory conditions in saliva. A cohort of patients with periodontitis (n = 8) and leukoplakia (n = 2) was collected and analyzed. Fluorescence signals from all benign inflammatory samples were significantly lower than those from patients with TSCC (P < 0.0001) and comparable to healthy controls (Fig. 6I), indicating minimal cross-reactivity with inflammation-associated factors (table S12). These results suggest that C502 responds specifically to TSCC-associated ANXA2 elevation rather than nonspecific inflammatory signals.

Furthermore, real-time kinetic analysis of saliva samples confirmed markedly higher signals in patients with TSCC compared to healthy controls (fig. S48). A strong correlation was observed between patient-matched serum and saliva signals (R = 0.9771; Fig. 6J), underscoring the consistency of biomarker detection across sample types and highlighting the DNA Walker as a clinically feasible platform for noninvasive cancer diagnosis.

DISCUSSION

This study establishes CEMs-SELEX, an integrated platform for de novo biomarker discovery in complex secretomes (Fig. 7). Applied to TSCC, it achieves three key advances: (i) direct selection of functional DNAzymes from native tumor secretomes, (ii) identification of secreted ANXA2 as a functionally defined biomarker via its specific activation of DNAzyme C502, and (iii) engineering of a C502-driven DNA Walker for sensitive, noninvasive TSCC detection in serum and saliva.

Fig. 7. Schematic overview of the CEMs-SELEX platform and its application to TSCC.

The workflow comprises three main stages: (i) probe screening by CEMs-SELEX, (ii) target identification via affinity pull-down and LC-MS/MS (liquid chromatography–tandem mass spectrometry); and (iii) cancer diagnosis of TSCC. This integrated pipeline closes the loop from functional probe discovery to clinical diagnosis.

Building on prior activity-based screening (AptaBiD, RFDs, and MORAC), CEMs-SELEX extends the paradigm in three distinctive directions: targeting the soluble secretome for noninvasive relevance, integrating definitive target deconvolution, and coupling discovery with clinical diagnostic engineering, representing a complete loop from discovery to application not previously demonstrated. Unlike abundance-based methods (e.g., MS or antibody arrays), CEMs-SELEX performs unbiased, activity-based selection within the native secretome, yielding probes that report on disease-specific biological activity (41–44). This workflow identified C502, whose activation led to ANXA2. Although ANXA2 is known in cancer (45–48), its role as a secreted functional activator of a DNAzyme and its utility for noninvasive saliva detection are new. Our work positions ANXA2 as a mechanistically grounded, functionality-derived biomarker.

The DNA Walker showed robust performance: linear response to 200 pM ANXA2 in serum, dynamic range extending to 10 nM in buffer, and correct classification of even high-concentration samples. Direct comparison with a commercial ELISA highlighted the advantage of functional detection: Despite good overall performance (AUC = 0.942), ELISA misclassified several samples due to reliance on absolute concentration, whereas the DNA Walker correctly identified all cases by detecting functionally active ANXA2.

Stratified analysis confirmed that C502 detects early-stage TSCC (T1/T2 = 24), with all samples exceeding the healthy threshold. Using serum from benign inflammatory conditions, we demonstrated minimal cross-reactivity, confirming specificity for TSCC-associated ANXA2 elevation. The strong serum-saliva correlation further underscores detection consistency across sample types. We acknowledge that information on smoking status was not systematically collected for this pilot cohort; this limitation highlights the need for standardized data collection protocols in future multicenter studies. Although these findings support clinical potential, validation in large multicenter cohorts with broader benign and early-stage samples remains essential.

The CEMs-SELEX workflow is inherently generalizable and is currently being applied to bladder cancer, with preliminary results supporting its broader utility. By coupling unbiased secretome screening with target identification and clinical-grade diagnostic engineering, this platform provides a blueprint for developing activity-based assays for cancers lacking defined molecular targets. In conclusion, CEMs-SELEX, synergized with DNA nanotechnology, closes the loop from functional probe discovery to clinical assay, offering a generalizable pathway for activity-based, noninvasive cancer diagnostics.

MATERIALS AND METHODS

Materials and reagents

All DNA oligonucleotides (sequences listed in tables S1, S5, and S8) and RT-qPCR primers (table S7) were custom synthesized and high-performance liquid chromatography (HPLC) purified by Sangon Biological Engineering Technology & Co. Ltd. (Shanghai, P. R. China). Phosphate-buffered saline (PBS; pH 7.4), Dulbecco’s PBS (D-PBS), and Opti-MEM were from Gibco (USA). Bovine serum albumin (BSA) and 2-mercaptoethanol (2-ME) were purchased from Sigma-Aldrich (USA). 6× Loading buffer, T4 ligase, and DNA marker were from TaKaRa Bio Inc. (Dalian, China). Yeast tRNA, glucose, and MgCl₂·6H₂O were from Sinopharm Chemical Reagent Co. Ltd. Streptavidin agarose beads (20353), Protein Transport Inhibitor cocktail (00-5980-93), MaxiSorp black 96-well microplates (437111), carboxyl magnetic beads (C37269), and Lipofectamine 3000 (L3000001) were from Thermo Fisher Scientific (USA). Protein Marker (PM2510) was from SMObio. 5× SDS loading buffer (P0015L), proteinase K (ST535), and bicinchoninic acid (BCA) protein assay kit (P0010S) were from Beyotime (China). For gene expression analysis, total RNA was extracted using the SteadyPure Mag Tissue & Cell RNA Extraction Kit (Accurate Biology, AG21023, China) and reverse transcribed with the Evo M-MLV Mix Kit with gDNA Clean for qPCR (Accurate Biology, AG11728, China) and SYBR Green Premix Pro Taq HS qPCR Kit (Accurate Biology, AG11701, China). Recombinant ANXA2 (Ag1777), LDHA (Ag16703), and HSP90AB1 (Ag28379) were from Proteintech (Wuhan, China). Rabbit monoclonal antibody to ANXA2 (56634) and mouse monoclonal antibody to β-tubulin (20374) was from ProMab Biotechnologies Inc. (China). AuNPs (20 nm) were from Pioneer Nanotechnology Co. Ltd. (China). DNase I (RNase-free) was from New England Biolab (USA). ANXA2 ELISA kits (FXs04815, 96T) were from Shanghai Fenxi Biotechnology Co. Ltd. (China). All other chemicals were of analytical grade and used without further purification. All solutions were prepared with ultrapure water unless otherwise stated.

Cell culture and sample collection

Human cell lines used in this study included the TSCC lines CAL27, SCC4, and SCC9 (target cells) and nontumorigenic oral epithelial cell lines HOEC and oral keratinocyte cell lines HOK16B (control cells). All cell lines were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Short tandem repeat (STR) profiling was performed by Abiowell (Changsha, China) and BFB Bio Co. Ltd. (Shanghai, China); supporting documentation is available upon request. Cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM; Gibco, USA) supplemented with 10% FBS and 1% penicillin/streptomycin at 37°C in a humidified atmosphere of 5% CO₂.

To prepare conditioned media, cells were cultured to 90% confluence, washed twice with PBS, and detached using 0.2% EDTA (without trypsin). After centrifugation at 2000 rpm for 4 min, cell pellets were resuspended in complete DMEM and transferred to new culture flasks. Supernatants were collected after 36 hours of culture, filtered through a 0.22-μm membrane, and stored at −80°C. Supernatants from positive lines (CAL27, SCC4, and SCC9) or negative lines (HOK16B and HOEC) were pooled, mixed, and aliquoted. A minimum volume of 50 ml per cell type was collected.

Selection of DNAzymes via CEMs-SELEX

To minimize selection bias and explore a broader sequence space, we used an undefined DNA library against complex CEMs targets. d-PAGE was used for DNA separation and purification throughout the selection process, reducing PCR amplification cycles and improving screening efficiency.

The key steps are summarized in fig. S1: (1) annealing of the DNA library, P-Sub, and linker at a 1:1:2 molar ratio, followed by T4 DNA ligase–mediated ligation; (2) purification of the ligated products by d-PAGE and ethanol precipitation; (3) negative selection using CEMs from normal cell lines (HOK16B and HOEC) in SELEX buffer, followed by ethanol precipitation; (4) purification by d-PAGE; (5) positive selection using pooled CEMs from TSCC cell lines (CAL27, SCC4, and SCC9) in SELEX buffer, followed by ethanol precipitation; (6) a second round of positive selection (repeat of step 5); (7) limited preamplification (10 cycles) followed by bulk PCR with cycle number optimized and monitored by d-PAGE; (8) ethanol precipitation, strand separation by d-PAGE, and recovery of the Cy5-labeled strand; and (9) quantification of ssDNA, reconstruction of the triple-stranded complex, and storage of residual ssDNA at −20°C. Steps 2 to 9 were repeated iteratively with increasing stringency (reducing incubation time from overnight at 4°C to 2 hours at room temperature) until cleavage activity plateaued. The final enriched library was amplified and subjected to high-throughput sequencing.

Selection and validation of TSCC-specific DNAzymes

1) d-PAGE analysis of cleavage efficiency. Catalytic activity was monitored by band shift on d-PAGE. Cleavage percentage (Clv%) was calculated as: Clv% = (Int[clv]/6)/(Int[clv]/6 + Int[unclv]) × 100%, where Int[clv] and Int[unclv] represent the grayscale intensities of cleaved and uncleaved bands, respectively (Fig. 1B). In the first selection round, only negative selection was performed to minimize nonspecific amplification.

2) Microplate fluorometric monitoring. To track enrichment of catalytic activity across rounds, 10 μl of CEMs from each round was incubated with fluorogenic substrate for 45 min. Fluorescence was measured using a microplate reader (Ex/Em = 488/520 nm). Although high background from uncleaved substrate limited qualitative analysis, a consistent increase in fluorescence over successive rounds indicated successful enrichment of active DNAzymes (Fig. 1D).

3) Bead-based flow cytometric DNAzyme assay. For rapid quantitative screening, fluorescently labeled DNAzymes were conjugated to carboxyl magnetic beads and analyzed by flow cytometry (Fig. 2A). The procedure included: (i) probe preparation, amino-modified (3′-NH₂-C6) DNAzymes were ligated to the fluorescent P-Sub using the T4 DNA ligase and purified by d-PAGE; (ii) immobilization, the purified DNAzyme-substrate complex was covalently coupled to carboxylated magnetic beads (6 μm) via EDC/NHS [1-ethyl-3-(3-dimethylaminopropyl)carbodiimide/N-hydroxysuccinimide] chemistry in MES buffer (incubated at 1200 rpm for 24 hours), followed by washing; and (iii) cleavage assay, functionalized beads were incubated with CEMs in SELEX buffer for 1 hour, washed, resuspended, and analyzed by flow cytometry [fluorescein isothiocyanate (FITC) channel). Mean fluorescence intensity from at least 10,000 beads was used to quantify cleavage activity. This method reduced total detection time to ~2 hours.

Target affinity verification and condition optimization

To establish optimal assay conditions for C502, the following parameters were systematically evaluated: temperature (26°C versus 37°C), divalent cations (Mn²⁺, Na⁺, K⁺, Mg²⁺, Ca²⁺, Pb²⁺, and Zn²⁺ at 10 mM), Mn²⁺ concentration (5 to 200 mM), and C502 concentration (50 to 1000 nM). Optimal conditions were defined as those yielding the highest signal-to-background ratio. Mn²⁺ (20 mM) and 37°C were selected as optimal, with 100 nM C502 used for subsequent experiments.

For steady-state measurements, reactions were incubated at 37°C for 1 hour, and fluorescence emission spectra were recorded from 505 to 650 nm (excitation: 499 nm). For kinetic measurements, temporal scanning was performed using a spectrofluorometer, and initial rates were derived from the first 60 s of reaction. Michaelis-Menten parameters were determined by varying substrate concentration from 1 to 10 μM while keeping C502 fixed at 100 nM under optimized buffer conditions (20 mM Mn²⁺ and 37°C). Initial velocities were calculated from the first 60 s of reaction traces.

Metabolic analysis and target identification

Target proteins bound to DNAzyme C502 were captured from TSCC CEMs using streptavidin beads. After extensive washing, bound proteins were eluted with 2× SDS loading buffer, denatured at 100°C for 10 min, and separated by 12% SDS-PAGE (120 V and 90 min). Gels were stained with Coomassie brilliant blue, and specific bands were excised for in-gel tryptic digestion followed by MS analysis. High-confidence candidate proteins were identified on the basis of ≥2 unique peptides, with keratins and common contaminants excluded. Untargeted secretome metabolomics was performed by Sangon Biological Engineering Technology & Co. Ltd. (Shanghai, China). MS analysis was conducted by Luming Biological Technology Co. Ltd. (Shanghai, China).

RT-qPCR and Western blot analysis

mRNA levels of candidate genes (ANXA2, RACK1, CLIC1, LDHA, VIM, HSP90AB1, C3, THB S1, PSNA, PKM, and EN01) were quantified by RT-qPCR using the SYBR Green Master Mix according to the manufacturer’s instructions (table S7). Total RNA was extracted from cells using the Cell RNA Extraction Kit and then reverse transcribed into cDNA using a reverse transcription kit. RT-qPCR reactions were performed on a CFX96 Touch system in triplicate. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as an internal control for normalization. Relative gene expression was calculated using the 2^−ΔΔCt method.

For Western blotting, cells were lysed in radioimmunoprecipitation assay (RIPA) buffer with protease inhibitors, and protein concentrations were determined by BCA assay. Equal amounts of protein (20 μg) were separated by SDS-PAGE, transferred to polyvinylidene difluoride (PVDF) membranes, and blocked with 5% skim milk. Membranes were incubated overnight at 4°C with primary antibodies against ANXA2 (1:1000) and β-tubulin (1:10,000), followed by horseradish peroxidase (HRP)–conjugated secondary antibodies (1:5000). Protein bands were visualized by ECL and quantified using ImageJ, with ANXA2 levels normalized to β-tubulin.

siRNA-mediated knockdown of ANXA2 in TSCC cells

To silence ANXA2 expression in TSCC cells (CAL27, SCC4, and SCC9), cells were seeded in 6-well plates at 2 × 10⁵ cells per well and cultured overnight to reach 70 to 80% confluence. ANXA2-specific siRNA (sense: 5′-CGGGAUGCUUUGAACAUUGAATT-3′; antisense: 5′-UUCAAUGUUCAAAGCAUCCCGTT-3′) and negative control siRNA were purchased from Sangon Biological Engineering Technology & Co. Ltd. (Shanghai, P. R. China). Transfection was performed using Lipofectamine 3000 according to the manufacturer’s instructions. Briefly, for each well, 5 μl of Lipofectamine 3000 was diluted in 125 μl of Opti-MEM, and 5 μl of siRNA (20 μM stock) was diluted in 125 μl of Opti-MEM. The diluted siRNA and Lipofectamine 3000 were mixed (1:1 ratio), incubated for 15 min at room temperature, and added dropwise to the cells. After 48 hours of transfection, conditioned media were collected for C502 activation assays.

Construction and characterization of the DNA Walker

The DNA Walker was assembled on AuNPs using a salt-aging method. AuNPs (20 nm) were functionalized with thiolated-C502 and FAM-labeled Sub at various molar ratios (detailed sequences in table S8). Before conjugation, C502 and Sub were reduced with 10 mM tris(2-carboxyethyl)phosphine (TCEP) for 30 min at room temperature to cleave disulfide bonds. The reduced strands were then added to 1 nM AuNP solution containing 4% Tween 80 and incubated for 1 hour. FAM-labeled Sub was added at a molar ratio of 1:1200 (AuNPs:Sub) and incubated for 30 min. Subsequently, 700 μl of 5 M NaCl was introduced stepwise, followed by sonication for 1 min and incubation for 1 hour; this sonication step was repeated three times. Unbound strands were removed by centrifugation (13,000 rpm, 20 min) in washing buffer (10 mM tris-HCl, 100 mM NaCl, and 0.05% Tween 20), repeated three times. The functionalized AuNPs were resuspended in storage buffer (10 mM tris-HCl with 100 mM NaCl) and stored in the dark at 4°C.

To determine the average number of substrate molecules per AuNP, conjugated Sub was released by adding 10 μl of 50 mM 2-ME to the DNA Walker solution. After 30-min incubation at room temperature, the mixture was diluted to 100 μl with 100 mM tris-HCl buffer, and fluorescence intensity was measured using a microplate reader (Ex/Em: 488/520 nm). The substrate concentration was calculated from a standard curve, and the number of substrate molecules per AuNP was derived accordingly.

UV-vis spectra of bare AuNPs and DNA Walker were recorded using a UV-2600 spectrophotometer (Shimadzu) over 400 to 650 nm. The characteristic plasmon peak of AuNPs shifted from 520 to 526 nm after DNA conjugation, confirming successful surface modification.

Functional evaluation and stability of the DNA Walker

For ANXA2 detection, the DNA Walker (500 pM) was incubated with recombinant ANXA2 at indicated concentrations in reaction buffer [50 mM tris-HCl (pH 7.5), 50 mM NaCl, 20 mM MnCl₂, and BSA (0.5 mg/ml)] at 37°C. Fluorescence kinetics were monitored over time using a spectrofluorometer (Ex/Em: 488/520 nm). A control without ANXA2 was included in each experiment.

To assess stability, DNA Walker (1 nM) or free ssDNA (100 nM C502 and 1200 nM Sub) was incubated in 100% FBS, DNase I (1 U/ml), pooled human serum, or human saliva at 37°C for up to 24 hours. Fluorescence intensity was measured at 0, 1, 2, 4, 8, 12, and 24 hours using an F-7000 fluorescence spectrophotometer (Ex/Em: 488/520 nm). All experiments were performed in triplicate.

Quantification of ANXA2 in human serum by ELISA

ANXA2 concentrations in serum were measured using a commercial human ANXA2 ELISA kit (Shanghai Fenxi Biotechnology Co. Ltd., FXs04815) according to the manufacturer’s protocol. Briefly, 100 μl of serum samples or standards was added to precoated 96-well plates and incubated for 1 hour at room temperature. After washing, biotinylated detection antibody was added for 1 hour, followed by streptavidin-HRP for 30 min. After washing, tetramethylbenzidine (TMB) substrate was added for 15 min, and the reaction was stopped with sulfuric acid. Absorbance was read at 450 nm using a microplate reader. All samples were assayed in duplicate. A standard curve was generated using recombinant ANXA2, and sample concentrations were interpolated. The linear range was 0 to 5000 pg/ml; samples exceeding this range were diluted and reanalyzed.

Clinical sample collection and processing

Clinical serum and saliva samples were obtained from the Hunan Cancer Hospital with approval from its Institutional Review Board (approval no. 2024-KS-0019). All participants provided written informed consent. The cohort included serum from 43 patients with pathologically confirmed TSCC, 37 age-matched and gender-matched healthy donors, and 10 patients with benign oral conditions (periodontitis and leukoplakia). Saliva was collected from a subset of 10 patients with TSCC, 10 patients with inflammatory conditions, and 10 healthy donors. Measurements performed by operators blinded to sample identity. Complete tumor-node-metastasis (TNM) staging information was available for most patients with TSCC. For some, records were incomplete (e.g., TX or NX), and three outpatients had no recorded staging information (only age and sex documented). For cases with confirmed distant metastasis (M1), disease was classified as late stage (stage IV); patients without staging information were excluded from stage-stratified analyses but included in overall diagnostic performance evaluations. Detailed clinical characteristics are provided in table S10. Serum was aliquoted immediately after collection and stored at −80°C until analysis. Saliva was mixed with RIPA lysis buffer containing 1% protease inhibitor (1:5 ratio), sonicated on ice (70 Hz, 10 s, and three cycles), and centrifuged at 10,000 rpm for 20 min at 4°C. The supernatant was collected, normalized to total protein (1 mg/ml) using a BCA assay, and stored at −80°C until use.

Supplementary Materials

This PDF file includes:

Figs. S1 to S48

Tables S1 to S12

Erratum published 9 October 2026:

The Supplementary Materials PDF has been replaced with the corrected versions of tables S10 and S11. The authors’ conclusions are not affected by this correction. The original version is available here: