Preparation of a multifunctional and ultrasensitive Au@AgNPs-Van/PDMS �lm SERS substrate for rapid detection of foodborne pathogens in beef

We present a novel and super-sensitive vancomycin (Van) modi�ed Au@AgNP self-assembly with polydimethylsiloxane (PDMS) (Au@AgNPs/Van-PDMS) �lm surface-enhanced Raman scattering (SERS) substrate for identi�cation of foodborne pathogens in beef, without the need for in vitro bacterial culture. The results illustrated that the Au@AgNPs/Van-PDMS �lm exhibited high reproducibility and Raman enhancement effect (1.09×10 5 ). Clostridium perfringens, Bacillus subtilis and Staphylococcus aureus isolated from beef captured by Au@AgNPs/Van-PDMS �lm exhibited high reproducibility and signi�cant differences. Principal component analysis score plots distinguished classes of foodborne pathogens, and the qualitative identi�cation of linear discriminant analysis was correct at 100%. The detection limit of S. aureus in beef was as low as 3 CFU/mL. This method could provide an effective means and technical support for realizing the application of SERS technology for quick and highly sensitive detection of foodborne pathogens in complex environments.


Introduction
Infections caused by foodborne pathogens, particularly those caused by pathogenic bacteria, pose a substantial threat to global public health. If treatment is delayed until after the beginning of symptoms, patients' health can rapidly deteriorate. [1][2] . Rapid detection of foodborne pathogens is critical to food safety and human health. Therefore, developing rapid, sensitive and reliable detection methods for assaying of foodborne pathogens is an urgent requirement.
The most common techniques for detection of foodborne pathogenic bacteria include plate colony counts, polymerase chain reaction (PCR) and enzyme-linked immunosorbent assays (ELISAs). The traditional identi cation method of bacterial culture with speci c media is the standard method for detecting foodborne pathogens; however, it is laborious and time-consuming before reliable identi cation is achieved, which usually takes 24 − 72 hours [3][4] . ELISA-based technology is highly sensitive, but its analytical steps are complicated, time-consuming, expensive and labor-intensive 5 . PCR is a molecular technique that uses enzymatic replication to amplify DNA fragments in vitro, allowing for high sensitivity and low detection limits. However, it is also time-and labor-intensive, and requires operationally experienced personnel to perform and analyze the tests 6 . These identi cation techniques are not suitable for direct rapid detection and eld detection of foodborne pathogens in complex food media.
Many spectroscopic techniques have been developed for food quality and safety testing. For example, uorescence spectroscopy is fast for identi cation, but poor in speci city and susceptible to mutual elemental interference and overlapping peaks [7][8] . Near-infrared spectrum has the disadvantages of large spectral bandwidth, overlapping characteristic peaks and susceptibility to moisture interference 9 . Although hyperspectral techniques collect more comprehensive sample information, they also lead to a large amount of data and redundant information, making it di cult to accurately characterize microbiological information such as foodborne pathogens 10 . Raman spectroscopy is a light-scattering technique that has the advantages of being fast, real-time, non-damaging, easy to operate, and with low interference by moisture, but it has problems such as large spectral bandwidth and low intensity 11 . Compared with other spectroscopic techniques, surface-enhanced Raman scattering (SERS) based on plasma nanoparticles (NPs) is a promising method for detection of foodborne pathogens owing to its high sensitivity, reliability and ngerprinting capability. Therefore, SERS has become an ideal choice for the detection and analysis of foodborne pathogenic bacteria 12 . The most commonly used precious metal colloids for SERS detection are AuNPs, AgNPs or their composite nanomaterials. Due to the electronic ligand effect and the enhancement of the local electric eld in the core-shell structure, the SERS activity of composite NPs (Au@AgNPs) is better than that of monolithic NPs (Ag, Au, etc.) [13][14][15] .
The sensitivity, speci city and reproducibility of ngerprint signals of foodborne pathogens are affected by the environment. Therefore, unlabeled SERS substrates are di cult to use directly for SERS detection of foodborne pathogens in complex samples due to their poor anti-interference properties and lack of recognition capture 16-17 .
In recent years, researchers have begun to use surface-modi ed SERS substrate to isolate bacteria from complex environments. Most studies have modi ed the surface of metal nanostructures with recognition molecules such as antibodies and phages, and combined them with magnetic materials (Fe 3 O 4 ) to achieve bacterial isolation and enrichment. However, these identi cation molecules are costly and operationally demanding, and the magnetic separation technique tends to lead to inhomogeneous distribution and poor reproducibility of SERS detection [17][18][19][20] . Vancomycin (Van), as a glycopeptide antibacterial drug, has the advantages of good stability, low cost and controllable quality. This antibiotic can interact with Gram-positive bacteria through the ve-point hydrogen bond between the heptapeptide on its skeleton and the terminal D-alanyl-d-alanine (D-Ala-d-Ala) on the cell wall of Gram-positive bacteria, and Van-modi ed SERS substrate can be used for the capture and detection of foodborne pathogens [21][22] . Polydimethylsiloxane (PDMS) has received a lot of attention as a support material for SERS-active NPs, which are well stabilized, simple to prepare, and have good Raman inertness and analyte enrichment in aqueous solution 4 .
Herein, we report the preparation of a multifunctional SERS substrate (Au@AgNPs/Van-PDMS lm) that could capture and enrich foodborne pathogenic bacteria in beef. This method used PDMS membrane as a exible solid-state substrate, and organically combines Au@AgNPs, which had good SERS enhancement, and Van, which captured and identi es food-borne pathogenic bacteria, to synthesize a SERS substrate. The substrate captured and enriched bacteria in solution, and served as a SERS substrate to enhance the Raman signal of the captured bacteria. The substrate was combined with the ltration process to enhance the Raman spectrum, and the rapid identi cation of foodborne pathogens was realized through analysis of the Raman ngerprint. At the same time, by analyzing the correlation between the intensity of Raman ngerprints and the concentration of foodborne pathogenic bacteria, rapid bacterial quanti cation was achieved. This provided a scienti c basis for food safety risk prediction. The aim of the present research was to establish a rapid detection method for foodborne pathogens in beef without the need for in vitro bacterial culture, and to provide an effective means and technical support for realizing the application of SERS technology for quick and highly sensitive detection of foodborne pathogens in complex environments. The ow chart of the study is shown in Fig. 1.

Synthesis of Au@AgNPs
Deionized water (99 mL) and HAuCl 4 (1 mL, 1%, wt %) were added to a clean triangular ask and heated to boiling on an electromagnetic heating stirrer at 600 rpm. Sodium citrate solution (6 mL, 1%, wt %) was added to the boiling HAuCl 4 solution and heated continuously for 10 min to produce AuNPs solution as the silver-shell-coated seeds. Fifty microliters of the above prepared AuNPs solution was added to 150 mL deionized water and 2 mL sodium citrate (C 6 H 9 Na 3 O 9 , 1%, wt %), heated to boiling and stirred at a controlled speed of 600 rpm. AgNO 3 solution (2 mL, 20 mM) was added dropwise to the above boiling solution and allowed to boil for 30 min. The heater was turned-off, the reaction solution was stirred at room temperature, and deionized water was added to bring the volume back to 200 mL 23 .

Design and synthesis of the Au@AgNPs/Van-PDMS lm
Au@AgNPs microspheres (10 mg) were added to 10 mL MUA ethanol solution (40 µM), sonicated for 2 h, and washed three times with anhydrous ethanol and ultrapure water to remove unreacted MUA. MUA was used as a multifunctional modi cation chain with a sulfhydryl group at one end that was rmly bound to the Ag shell surface via an Ag-S bond and a carboxyl group at the other end that provided a site for coupling vancomycin. At the same time, the carbon chain of MUA extended the arm, effectively reducing the steric hindrance between vancomycin and bacteria (the chemical structural formulas of MUA and Van were given in Figure S1). Au@AgNPs-MUA pellets were dissolved in MES buffer (0.1 M, pH 5.5) and sonicated for 10 min to disperse them completely. EDC aqueous solution (1 mL, 10 mg/mL) and 1 mL Van solution (5 mg/mL) were added and the ultrasonic reaction was continued for 2 h. EDC activation facilitated and stably promoted the chemical coupling between vancomycin and MUA ( Figure S2, in the Supporting Information). Au@AgNPs/Van were prepared by washing with ultrapure water three times [24][25] .
The preparation process of Au@AgNPs/Van-PDMS lm SERS substrate was shown in Figure S3. PDMS membranes were surface-treated by a low pressure plasma system (Zepto, Diener Electronic) at 15 Pa vacuum pressure and 30 W radio power to improve their hydrophilicity. The hydrophilic-treated PDMS lm was placed in 5% APTMS, incubated overnight with shaking, and washed three times with anhydrous ethanol in pure water after removal. The APTMS solution could silanize the PDMS membrane surface. The methoxy group in APTMS was easy to hydrolyze, and the silicon hydroxyl group (Si-OH) produced was combined with the PDMS membrane hydroxyl group by hydrogen bonding, assembling APTMS on the PDMS lm, with an exposed amino group at the end 26-27 . When the treated membranes were placed in Au@AgNPs-Van solution, the strong chemical interaction between the Ag on the surface of Au@AgNPs and the exposed amino groups of APTMS triggered the self-assembly of Au@AgNPs-Van on the membrane surface for a duration of 4 h, forming a sandwich structure of bottom-up PDMS-Au@AgNPs-Van. The above process was repeated twice after washing with ultrapure water to obtain Au@Ag NPs/Van-PDMS lm SERS substrate and placed on a clean bench. with sterile water and dried for processing, and the number of bacteria remaining in the blood was studied via by plate count assay.

SERS detection
Raman spectra were collected using a LabRam HR Evolution Raman spectrometer (Horiba Scienti c Inc.).
Raman spectra were collected and the spectrometer was operated using LabSpec software (V6). Prior to sample detection, the 520.7 cm −1 peak was used as the reference for calibration of the Raman instrument using silicon wafers. SERS spectra were acquired using a 532.8 nm laser as the excitation light source, with the laser intensity set to the maximum intensity (15 mW), the objective set to 100×, and the integration time of 30 s for three times. The detected spectra were from 400 to 1800 cm −1 . A total of 15 replicate spectra were collected for each bacterium.

Characterization of materials
The different materials prepared in this experiment were characterized by transmission electron microscopy (JEM-2100; Jeol), scanning electron microscopy (JSM-IT100; Jeol), UV-vis spectrophotometer (UV-2600; Shimadzu) and laser particle size analyzer (Mastersizer 3000; Malvern panalytical), shown in Fig. 2. The particle size of Au@AgNPs was mainly distributed in the range of 35 − 85 nm, and the average size was about 57.85 nm ( Fig. 2A). TEM demonstrated that the Au@AgNPs particles exhibited an Au core-Ag shell sandwich sphere structure (Fig. 2B, C). UV-vis absorption spectra of Au@AgNPs showed SPR absorption peaks at 420 nm and 517 nm, and a representative UV-vis absorption peak of vancomycin at 280 nm after incubation with vancomycin appeared in Au@AgNPs/Van (Fig. 2D). After vancomycin was activated in EDC buffer solution, it bound to AgNPs on the surface of Au@AgNPs through Ag-S bonding, thus immobilizing Van on the surface of Au@AgNPs, suggesting that vancomycin molecules were tightly bound on the Au@AgNPs surface 21,23,29 . SEM characterization of Au@AgNPs/Van-PDMS lm showed that the PDMS surfaces were covered with a dense forest of Au@AgNPs/Van particles (Fig. 2E, F). The strong chemical interaction between the Ag on the surface of Au@AgNPs and the exposed amino groups on the surface of the silanized PDMS lm triggered selfassembly of Au@AgNPs/Van on the PDMS lm surface, forming uniformly ordered SERS substrates with a bottom-up sandwich structure of PDMS-Au@AgNPs/Van. The energy-dispersive X-ray elemental  To determine whether the Raman detection signal was enhanced by Au@AgNPs/Van-PDMS lm and to quantify its SERS enhancement, we used a laser confocal Raman microscope system to collect spectral data of different foodborne pathogenic bacteria samples as a blank group and foodborne pathogenic bacteria on Au@AgNPs/Van-PDMS lm as a control group. As shown in Figure S4, the Raman spectrum forAu@AgNPs/Van-PDMS lm did not exhibit any identical bands, which would not affect the subsequent SERS detection of microorganisms. It can be noted that the normal Raman intensities for food-borne pathogens are weak with few distinguishable characteristic peaks, so it is di cult to obtain effective identi cation results for different food-borne pathogens in Figure S4. However, after using Au@AgNPs/Van-PDMS lm as the enhanced substrate for SERS measurement, we observed a dramatic increase in peak intensities as well as in the number of peaks in the spectrum. These ndings suggested that Au@AgNPs/Van-PDMS lm exhibited good performance in the rapid detection of different foodborne pathogens. Subsequently, the intensity of the strongest SERS spectral peaks of three different foodborne pathogenic bacteria was quanti ed to determine the enhancement effect of Au@AgNPs/Van-PDMS lm (Table S1, in the Supporting Information). The AEF was calculated to be 1.09 × 10 5 for Au@AgNPs/Van-PDMS lm.

SERS spectroscopic assessment
The average Raman spectra of the three foodborne pathogens were calculated from 15 Raman spectra of each species. The tentative attribution of all observed major spectral bands for the three foodborne pathogens in Table 1   [amine III (proteins)], 1308-1318 cm −1 (amide III) and 1454-1488 cm −1 (nucleic acid mode) were seen in C. perfringens and B. subtilis, respectively. The Raman shift and peak intensity of the SERS spectra of the three foodborne pathogenic bacteria differed. The SERS technique based on Au@AgNPs/Van-PDMS lm can detect even subtle compositional differences in different microorganisms and distinguish them quickly, which has the ability to detect and distinguish different species of foodborne pathogenic bacteria from beef samples.
To examine the reproducibility of the Au@AgNPs/Van-PDMS lm SERS substrate for detection of the three foodborne pathogenic bacteria, 15 random acquisitions of each of the three pathogens were performed using a LabRam HR Evolution Raman spectrometer (Fig. 3B-D). The maximum RSD of the spectra of the three foodborne pathogens was calculated to be the Raman peak of B. subtilis at 1001 cm −1 of 5.28%. The results indicated that the SERS technique based on Au@AgNPs/Van-PDMS lm SERS substrate had good reproducibility for the detection of foodborne pathogens, which provided support for further rapid detection of foodborne pathogenic bacteria based on this SERS technique.

Multivariate statistical analysis
To minimize the effects of similar Raman vibrational spectra interfering with microbiological classi cation and to observe the effect of qualitative identi cation of distinct foodborne pathogens, principal component analysis (PCA) and linear discriminant analysis (LDA) were used to evaluate the identi cation and classi cation of three foodborne pathogens detected by SERS with Au@AgNPs/Van-PDMS lm. PCA is a well-known method for multivariate data analysis that can assist in the visualization of data information based on projection techniques. With this technique, dimensions of the original data are reduced to a smaller number of new variables (principal components, PCs). Each PC is a linear combination of the original variables, with the rst PC explaining the maximum amount of variation in the original data and each subsequent PC explaining the maximum amount of variation not explained by the previous PC 38 . The raw spectral data with 341 variables were calculated by PCA to obtain 341 principal components. A three-dimensional scatter plot was constructed using the rst three principal components to observe the classi cation trends of 45 samples (Fig. 4A). The cumulative variance contribution of the rst three principal components (PC1, PC2 and PC3) was 95.80%, which represented all the spectral information. Although there was a clear classi cation trend for the three foodborne pathogens, the classi cation effect was still not ideal. Therefore, to make the classi cation of the three foodborne pathogens more satisfactory, further qualitative discriminant analysis of the three foodborne pathogenic bacteria spectra was performed using LDA.
LDA is a widely used supervised pattern recognition method. To achieve maximum discrimination, it seeks a linear projection of the data to a low-dimensional subspace such that the ratio of interclass distance to between-class distance is maximized. By using the LDA algorithm, dimensionality reduction, feature extraction and discrimination can be achieved simultaneously 39 . All of the 45 samples of the three foodborne pathogenic bacteria samples were correctly identi ed in their class-space ( Fig. 4B and Table 2). All samples for each foodborne pathogen were concentrated in a small area near the center of the mass, without any overlapping areas. LDA performed well in classifying and identifying three foodborne pathogenic bacteria by SERS spectra, and was able to distinguish them in a robust and fast manner with 100% correct identi cation rate in both the Initial and predict group.

Quantitative detection
To assess the detection precision of the SERS substrate, S. aureus were spiked into beef samples and the accurate concentration of bacteria in samples was determined by traditional plate counting 22,28 . The capturing e ciency (CE) of the spiked recovery process of S. aureus in beef samples was calculated using Eq. (1), where S1 was the raw microbial count of S. aureus and S2 was the residual amount of S. aureus in the SERS substrate rinse solution. From Fig. 5A combined with Eq. (1), the recovery of S. aureus in beef samples was obtained as 94.97%. Figure 5B shows the SERS spectra of Au@AgNPs/Van-PDMS lm captured to different concentrations of S. aureus. The SERS spectra intensity gradually increased with increasing S. aureus concentrations, as expected. With different concentrations of 10-10 5 CFU/mL, the logarithmic value of S. aureus concentrations represented a good linear connection with Raman signal intensity at 1524 cm −1 (Fig. 5C), the calibration curve was y = 3859.58x − 1956.23 with R 2 of 0.96. The limit of detection (LOD) was calculated using the formula: LOD = 3M/W, where M represents the value of the standard deviation of blank samples and W is the slope of the standard curve within the linear range 40 . According to this formula, the LOD of the proposed method was 3 CFU/mL.

Conclusion
We used the Au@AgNPs/Van-PDMS lm as a ultra-sensitive and selective SERS substrate to provide a method for on-site detection of foodborne pathogens in beef. There were signi cant differences in the SERS spectra of three foodborne pathogenic bacteria detected by Au@AgNPs/Van-PDMS lm-based SERS technique. PCA and LDA were used to classify and identify the three pathogens with good effect, and the qualitative identi cation accuracy of LDA was 100%. This indicated that the Au@AgNPs/Van-PDMS lm SERS substrate had good identi cation of foodborne pathogenic bacteria in beef samples.

Declarations Supporting Information
The chemical structure formulae of MUA (a) and Van (b), respectively ( Figure S1); Schematic diagram of Au@AgNPs/Van synthesis ( Figure S2); Schematic diagram of preparation process of Au@AgNPs/Van-PDMS lm SERS substrate ( Figure S3); SERS detection performance for the three foodborne pathogens pathogens based on Au@AgNPs/Van-PDMS lm ( Figure S4), Calculation of AEF for Au@AgNPs/Van-PDMS lm (Table S1), and Combine Figure 4 and Table 1 to explain the calculation process of AEF.

Consent for publication
All the authors agree with the publication.
Availability of data and materials The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
None of the authors declares any kind of competing interests.   Identi cation results of SERS spectra for three foodborne pathogens by PCA (A) and LDA (B).