Chloroauric acid (HAuCl₄), sodium borohydride (NaBH₄), silver nitrate (AgNO₃), hydrochloric acid (HCl), ascorbic acid (AA), sodium oleate, anhydrous ethanol, palladium chloride (PdCl₂), bovine serum albumin (BSA), chloroplatinic acid (H₂PtCl₆), 5,5′-dithiobis (2-nitrobenzoic acid) (DTNB), and phosphate buffer solution (PBS) were purchased from Sinopharm Chemical Reagent Co. Cetyltrimethylammonium bromide (CTAB), cetyltrimethylammonium chloride (CTAC), tris(2-carboxyethyl)phosphine (TCEP), and TMB were purchased from Aladdin Biochemical Science and Technology Co. The nucleotide sequences listed in Table 1, including the VEGF aptamer and single-stranded DNAs (ssDNA1 and ssDNA2), were synthesized by Sangon Bioengineering (Shanghai) Co. The qRT-PCR kit was acquired from Getein Biotech (China). Deionized water (resistivity of 18.2 MΩ) was prepared using a Millipore purification system. Human serum samples were supplied by Yixing Hospital of Traditional Chinese Medicine.

S-4800II Field Emission Scanning Electron Microscope (Hitachi, Japan), Tecnai12 Transmission Electron Microscope (Philips, Holland), Tecnai G2 F30S-TWIN Field Emission Transmission Electron Microscope (FEI, United States), Cary60 UV-Vis Spectrophotometer (Agilent, United States), and DXRxi Micro-Raman Spectrometer (ThermoFisher, United States).

We optimized the experimental steps based on Tang et al. (2022). AuNRs with uniform morphology were synthesized using a seed-mediated growth method. Initially, 0.196 mL of HAuCl₄ solution (25.4 mM) was mixed with 9.78 mL of deionized water, Then, 9.9 mL of CTAB solution (0.2 M) was added and mixed thoroughly. Next, 2 mL of freshly prepared NaBH₄ solution (5.5 mM) was rapidly injected into the mixture under stirring at 1,100 rpm. Stirring was maintained for 120 s, during which the color changed from yellow to light brown. The solution was then left undisturbed for 1 h to complete seed formation. Subsequently, 500 mL of deionized water, preheated to 50°C, was used to dissolve CTAB (140 mM) and sodium oleate (26 mM). After cooling the solution to 30°C, 3.95 mM AgNO₃ solution was added and allowed to react for 15 min. With continuous stirring at 800 rpm, 500 mL of HAuCl₄ solution (1 mM) was introduced and stirred for 1.5 h until the solution became colorless. Then, 4.2 mL of 37% HCl solution (12.1 M) was added, followed by 2.5 mL of AA solution under stirring at 1,200 rpm. After 30 s, 0.8 mL of the previously prepared seed solution was added to the growth mixture, stirred at 1,300 rpm for 30 s, and then left undisturbed for 12 h. Finally, the product was collected by centrifugation at 9,000 rpm for 15 min, and the precipitate was redispersed in a 7 mM CTAB solution.

Pd shells were formed on the surface of AuNRs through reduction in an aqueous solution containing CTAB and AA. A 10 mM H₂PdCl₄ solution was first prepared by dissolving 35.4 mg of PdCl₂ powder in 2.4 mL of 0.2 M HCl using a 60°C water bath for 60 min, then diluted to a final volume of 20 mL with ultrapure water. To prepare the coating mixture, 4 mL of the 10 mM H₂PdCl₄ solution and 100 mM AA were added to 70 mL of 7 mM CTAB, followed by the addition of 80 mL of ultrapure water. The mixture was stirred vigorously for 5 min. Next, 3.4 mL of AuNRs solution was added to the prepared reaction mixture and stirred continuously at 25°C for 3 h. To remove excess CTAB, the resulting Au@Pd NRs were washed by centrifugation at 11,000 rpm for 15 min. The supernatant was discarded, and the precipitate was redispersed in 5 mL of ultrapure water. Subsequently, ssDNA1 (0.1 mM) activated by TCEP (1 M) was added to it and incubated for 2 h at 37°C. To block nonspecific binding sites on the particle surface, 15 μL of a 1 wt% BSA solution was added, and the mixture was incubated for an additional 1.5 h. Finally, the product was centrifuged at 12,000 rpm for 7 min to remove unbound reagents and nucleic acid strands, yielding the nanoenzyme probe.

We prepared Au TOHs by improving the method described by Zhao et al. (2024). To begin, 19.5 mL of CTAB (0.1 M), 0.3 mL of HAuCl₄ (10 mM), and 0.2 mL of H₂PtCl₆ (0.1 M) were mixed thoroughly. Then, 1.8 mL of freshly prepared ice-cold NaBH₄ (10 mM) solution was added under low-speed stirring at 300 rpm. The mixture was stirred until it turned light brown with slight foaming. From this, 40 μL of the reaction mixture was combined with 20 mL of water to prepare a seed dilution solution for the next step. Next, 1 mL of HAuCl₄ (10 mM), 10 μL of H₂PtCl₆ (10 mM), and 0.5 mL of AA (0.1 M) were added sequentially to 20 mL of CTAC (25 mM) and mixed well. Then, 1 mL of the prepared seed dilution was added, and the resulting solution was gently shaken and placed in an oven at 45°C for a static reaction lasting 7 min. A deep red solution indicated the successful formation of Au TOHs. To assemble the Au TOHs into ordered arrays, 2 mL of Au TOHs solution and 4 mL of n-hexane were added to a beaker, followed by 2 mL of ethanol. The mixture was left to stand for 120 s to allow the formation of an oil-water interface, where Au TOHs spontaneously self-assembled into a tightly packed, metallic film. This monolayer film was then transferred onto a silicon wafer previously treated with piranha solution and dried under constant temperature to form the Au TOHs array. For preparation of the capture substrates, ssDNA2 and aptamer were first hybridized in phosphate buffer (8 mM) containing MgCl₂ (0.8 mM) and NaCl (20 mM). After incubation for 1.5 h, 800 μL of ssDNA2 (100 μM) was mixed with 80 μL of TCEP (1 M) and left to react at room temperature for 60 min to activate the sulfhydryl groups. The activated duplexes were then dropwise added onto the Au TOHs arrays and incubated at 37°C for 2 h to allow full attachment to the surface. Finally, the modified arrays were washed thoroughly with PBS and deionized water, completing the fabrication of the capture substrates.

This study was approved by the Ethical Review Committee of Yixing Hospital of Traditional Chinese Medicine (Ethics Number: 2024-KY-054). Serum samples were collected from 20 patients diagnosed with DR and 20 healthy individuals admitted to the hospital. Participant selection was conducted in accordance with the criteria outlined in the Chinese Guidelines for the Prevention and Control of Type II Diabetes Mellitus (2020 edition), and all participants signed informed consent voluntarily. The exclusion criteria included individuals with concurrent eye diseases such as glaucoma or cataracts, those with type I diabetes, acute diabetic complications including ketoacidosis or hyperosmolar coma, severe organ dysfunction, autoimmune disorders, or recent retinal photocoagulation treatment. Collected samples were processed aseptically and sealed according to standard procedures before being stored at −80°C in an ultra-low temperature environment. All steps complied with laboratory regulations for cold-chain management and testing operations.

Patient serum samples and nanoenzyme probes were added dropwise to the SERS detection platform. VEGF present in the serum initiated the competitive recognition process, allowing more nanoenzyme probes to bind to the capture substrate. After incubation, unbound probes were washed off with PBS buffer and deionized water. A reaction mixture containing TMB (1.0 mM) and H₂O₂ (0.6 mM), adjusted to pH 4.5 using sodium acetate buffer, was then added dropwise to the platform. The reaction was conducted at 30°C for 14 min in a constant-temperature environment. When the solution turned visibly blue, SERS spectra were recorded using a laser with a wavelength of 785 nm, an exposure time of 10 s, and a power of 5 mW.

The sensitivity of the platform was assessed by determining the Limit of Detection (LOD), defined as the lowest reliably distinguishable concentration of the analyte. The LOD was calculated using the following formula: L O D = 3 S D / K where K is the slope obtained from linear regression of the signal-concentration curve, and SD is the standard deviation of the SERS intensity at 1,602 cm⁻¹ for the blank sample.

Au TOHs with strong SERS activity were prepared as substrates for the detection platform. The addition of H₂PtCl₆ to the seed and growth solutions effectively enhances the purity and monodispersity of Au TOHs with negligible effects on the gold nanostructures. Pt selectively deposits on the high-energy crystalline surfaces, thereby inhibiting their growth without doping into the gold nanostructures (Zhao et al., 2024). Figures 2A,B present the SEM and TEM images of the synthesized Au TOHs, confirming their regular morphology, uniform dispersion, and sharp edge features. Figure 2C displays the HRTEM and SAED images, showing clear lattice fringes with a spacing of approximately 0.231 nm, indicating a single-crystalline structure. The UV-Vis-NIR absorption spectrum in Figure 2D reveals a prominent absorption peak near 546 nm. Figure 2E shows that Au TOHs formed tightly packed, ordered arrays via self-assembly at the oil-water interface. In addition, Au TOHs labeled using DTNB (10 nM) were compared with pure DTNB (10 mM) Figure 2F according to Eq: E F = I S E R S / C S E R S I R a m a n / C R a m a n , the value of EF was 1.88 × 10⁹, indicating that the Au TOHs array had a significant SERS enhancement effect. Figure 2G shows the SERS spectra collected from 20 randomly selected positions on the DTNB labeled Au TOHs array, indicating high signal uniformity. The scatter plot in Figure 2H represents the intensity values at 1,340 cm⁻¹ across these positions, with a relative standard deviation (RSD) of 4.22%, further confirming the reproducibility of the array. In terms of stability, Figure 2I presents the SERS spectra of DTNB labeled Au TOHs arrays stored at room temperature over 0, 5, 10, and 15 days. A gradual decrease in signal intensity was observed, with an overall drop of about 10.11% compared to the initial state, suggesting that the arrays retain good stability over time.

The transmission electron microscopy image in Figure 3A shows that the synthesized Au@Pd NRs exhibit uniform rod-like morphology with consistent size and shape. As seen in the HRTEM image in Figure 3B, the nanorods display a Pd-coated Au core-shell structure, featuring a rough surface and lattice fringe spacing of 0.192 nm. The average length and width of the Au@Pd NRs are approximately 80 nm and 32 nm, respectively, giving an aspect ratio of around 2.5:1, as illustrated in Figures 3C,D. The UV-Vis-NIR absorption spectra in Figure 3E show that the Au@Pd NRs solution has a distinct peak at 747 nm. When TMB was used as the substrate and in the presence of H₂O₂, Au@Pd NRs could catalyze the oxidation of TMB to produce a blue product and showed a strong absorption peak at 646 nm, whereas the solution was unchanged when TMB and H₂O₂ were present alone, which demonstrated its good catalytic activity. Elemental composition analysis through EDX and mapping (Figure 3F) confirmed that Au and Pd are the primary components of the nanorods. The detected Cu signal originates from the supporting TEM grid. As shown in Figure 3G, SERS spectra of DTNB labeled Au@Pd NRs were compared with those of pure DTNB (10 mM). The Au@Pd NRs displayed a strong SERS enhancement, with an enhancement factor (EF) calculated as 9.9 × 10⁸. To evaluate the catalytic activity of the nanoenzymes, we analyzed the Michaelis-Menten kinetics of Au@Pd NRs by fixing the H₂O₂ concentration, adjusting the TMB concentration, and setting up three replicate wells for each group (Figures 3H,I). The maximum reaction rate was determined to be 0.663 mM s⁻¹, and the Michaelis constant (Km) was 0.351 mM for the Au@Pd NRs nanoenzyme.

During the target recognition process using the nanoenzymatic SERS bifunctional detection platform, both the pH and substrate concentrations significantly influence nanoenzyme activity. To ensure reliable testing, this study optimized key parameters using five detection platforms from the same batch. As shown in Figures 4A,B, the SERS signal intensity increased initially with rising temperature and pH, peaked, and then declined. The optimal conditions were determined to be a temperature of 30°C and a pH of 4.5. Figure 4C shows that the SERS signal intensity increased steadily over time, with the rate of increase slowing and reaching a plateau after 14 min. Therefore, 14 min was selected as the optimal incubation time. As indicated in Figures 4D,E, the highest SERS signal intensity was observed at a TMB concentration of 1.0 mM and an H₂O₂ concentration of 0.6 mM.

The specificity and reproducibility of the detection platform were assessed under optimized conditions. As shown in Figures 5A,B, SERS spectra of five groups of VEGF were measured using independently prepared five-batch detection platforms. The RSD was 2.34%, indicating strong reproducibility and minimal variation across batches. To assess specificity, the platform was tested against structurally unrelated proteins, including three sets of the same concentrations of BSA, IgG, CEA and PCT. The results, shown in Figures 5C,D, revealed strong SERS signals only in the presence of VEGF. In contrast, signals for other analytes closely resembled the blank, confirming that these interferents did not affect detection. Furthermore, nonspecific binding was minimized by pretreating serum samples to remove non-target molecules and by thorough PBS rinsing of the platform. These measures supported the high specificity and reproducibility of the SERS detection platform developed in this study.

Under optimized experimental conditions, the nanoenzymatic SERS bifunctional detection platform was applied to detect and quantify VEGF in serum across a range of concentrations. In order to accurately determine the lower limit of detection of this assay platform, in this study, we adjusted the concentration of healthy human serum samples to 1 mg/mL by pretreatment and addition of VEGF, followed by a gradient dilution method, adjusting the concentration of samples to 10 pg/ml-1 μg/ml. As shown in Figure 6A, with increasing VEGF levels, more Au@Pd NRs were recruited to the detection platform, leading to a progressive rise in the SERS signal intensity of catalytically generated oxTMB. This change in signal intensity corresponded to the concentration of the target protein. A linear regression curve was generated between the logarithm of the VEGF concentration obtained and the SERS intensity at 1,602 cm⁻¹ by performing 3 sets of SERS assays for 7 different gradients of VEGF concentration (Figure 6B). The LOD was calculated to be as low as 0.11 pg/mL. When compared to previously reported detection strategies (Table 2), the method described here exhibits a substantially lower detection limit and strong specificity, enabling accurate measurement of VEGF in complex serum samples.

Early-stage DR can be screened through imaging and further confirmed through pathological diagnosis. Figures 7A,B show scanning laser ophthalmoscopy (SLO) images of a DR patient, revealing hemorrhagic spots and proliferative membranes in the fundus. Figures 7C,D present fundus fluorescence angiography images displaying widespread retinal nonperfusion and fluorescent leakage from neovascular formations in both eyes. Figure 7E is an OCT scan that shows macular edema and detachment of the retinal neuroepithelial layer. While imaging technologies such as OCT provide valuable diagnostic insights, their high cost, complex operation, and limited availability hinder large-scale screening. The nanoenzymatic SERS bifunctional detection platform developed in this study offers a more accessible approach for early DR screening.

In Figure 8A, we compared the differences between 20 DR patients and 20 healthy individuals by detecting VEGF. The results showed that the VEGF concentration in the DR group was significantly higher than that in the healthy group (P < 0.001). We analyzed serum samples from patients with early-stage DR and healthy individuals using the nanoenzymatic SERS bifunctional detection platform. As shown in Figures 8B,C, a clear difference in SERS signal intensity was observed between the two groups when detecting VEGF. The SERS signal at 1,602 cm⁻¹ was substituted into the linear regression equation to calculate the VEGF concentration (ng/mL). To assess accuracy, the results obtained from the SERS platform were compared with those from the ELISA method, which served as the reference standard in this study. As shown in Table 3, the calculated concentrations exhibited relative deviations that remained within an acceptable range, demonstrating the reliability of the detection platform.