Broad pH-resistant microdroplet SERS platform based on Ag@SiO₂@PVP NPs for real-time analysis of microbial metabolites

Abstract

Surface-enhanced Raman scattering (SERS) holds great promise for the real-time detection of microbial metabolites. However, pH fluctuations in fermentation environments can induce probe corrosion, aggregation, or precipitation, severely compromising the stability and reproducibility of SERS signals and thus diminishing system reliability for real-time analysis. To address this challenge, this study proposes a “material-system co-design” strategy to develop a wide pH-tolerant microdroplet SERS detection platform based on Ag@SiO₂@PVP nanoprobes. First, SERS nanoprobes with inherent pH tolerance were constructed through the synergistic effects of physical isolation provided by the SiO₂ shell and steric hindrance from the outer polyvinylpyrrolidone (PVP) coating. Subsequently, to achieve integrated operations including rapid mixing of nanoprobes and metabolites, homogeneous droplet generation, in-situ trapping, and SERS detection, a microfluidic chip incorporating a herringbone-shaped mixer, a droplet generation zone, an in-situ trapping zone, and a reflective substrate was designed. Using L-DOPA as a model molecule for performance validation, the results demonstrated a limit of detection as low as 9.5 × 10⁻⁹ g/mL (based on the 3σ method). The SERS signals remained highly stable and reproducible across the pH 3–11 range, with a linear correlation coefficient R² > 0.99 and a relative standard deviation (RSD) of characteristic peak intensities below 5% (n = 20). In tests using real samples from E. coli fermentation broth, the method maintained clear spectral recognition capability across the concentration range of 10⁻⁸ ~ 10⁻⁴ g/mL, exhibiting significantly superior detection performance and stability compared to bare silver nanoparticles. Preliminary tests with L-tyrosine and arbutin further validated the platform’s universality. Through its synergistic “dual-protection probe–multifunctional chip” strategy, this study provides a powerful tool for reliable, high-throughput analysis of metabolites in complex biological environments.

Introduction

In the transition of biomanufacturing from traditional fermentation to precision intelligence-driven manufacturing, high-throughput real-time analysis of microbial metabolites is a critical step for transforming from “black-box operation” to transparent perception and from experience-driven to data-driven paradigms^1–4. Surface-enhanced Raman spectroscopy (SERS), with its high sensitivity, fingerprint identification capability, rapid response, and in-situ analysis potential, enables sample interrogation in a near-pristine state, faithfully reflecting the dynamics of biological processes. Consequently, it is regarded as a highly promising analytical tool in this field^5–8. Concurrently, droplet microfluidics, leveraging its precise manipulation of fluids at microliter to nanoliter scales, offers a viable pathway for constructing highly integrated automated analysis platforms^5,9,10. By partitioning samples into thousands of independent micro-reactors, it provides an ideal platform for achieving high-throughput, parallel SERS analysis^7,11–14.

The enhancement mechanism of SERS primarily originates from electromagnetic field enhancement¹⁵. When molecules adsorb onto the surfaces of noble metal nanostructures such as gold or silver, their Raman signals can be amplified by factors of up to millions or even hundreds of millions, enabling trace-level detection and providing unique molecular structural information¹⁶. Among commonly used metal nano-enhancers, silver exhibits particularly outstanding enhancement performance due to its lower Ohmic loss in the visible region, which generates stronger and more highly localized surface electromagnetic fields, thereby creating more numerous and intense SERS “hot spots”^17,18. However, the practical application of silver nanomaterials in complex metabolic environments still faces significant challenges: (1) pH fluctuations in the system can compress the electrical double layer and neutralize the surface charge essential for stabilization^19–22, leading to uncontrolled particle aggregation and even permanent precipitation; (2) acidic or oxidative environments can induce corrosion and dissolution of silver, resulting in loss of SERS activity²³. Furthermore, pH changes can alter the adsorption behavior of analytes on the particle surface, causing significant variations in SERS signal intensity and spectral features^24,25, severely compromising detection reliability and reproducibility.

To address the interference caused by pH fluctuations in microbial metabolic systems on SERS detection, existing research has primarily focused on strategies such as external pH control or replacing silver nanoparticles. For example, Crooks et al.²⁶ achieved local electrochemical pH regulation within individual droplets by inducing water electrolysis using microelectrodes; Rosenbaum’s team²⁷ achieved global pH adjustment across numerous droplets through the diffusion of proton carriers (e.g., organic acids and amines) from the external oil phase; Jia et al.²⁸ developed a flexible pH regulator on a digital microfluidics (DMF) platform, combining droplet manipulation with image analysis for real-time pH monitoring and control in microliter-scale samples. However, pH control within microdroplets alters the original chemical environment of the sample, potentially causing signal distortion. More critically, in practical application scenarios, such intervention may even induce apoptosis of strains or cells within the droplets, fundamentally conflicting with the principle of “non-destructive in-situ” SERS analysis. Additionally, this strategy involves complex operation, difficulty in monitoring pH within microdroplets, and inability to achieve large-scale precise control, thus failing to meet the demands of real-time, high-throughput SERS analysis. On the other hand, while replacing silver nanoparticles can improve chemical stability, it often comes at the cost of sacrificing enhancement performance, leading to significantly reduced SERS signal intensity and difficulty in meeting the detection requirements for trace metabolites^29,30. Therefore, developing a SERS detection strategy that combines high enhancement performance with broad pH tolerance, while supporting real-time, in-situ analysis, has become an urgent need in this field.

This study focuses on how to enable established microfluidic-SERS analysis systems to operate stably and reliably within extremely dynamic chemical environments, such as a wide pH range. The innovation of this work lies not in inventing entirely new chip operating principles, but in achieving a “material-system co-design.” This involves deeply coupling intrinsically pH-tolerant Ag@SiO₂@PVP nanoprobes with a microfluidic chip integrating a herringbone-shaped mixer, droplet traps, and a reflective substrate for optical path enhancement, successfully constructing a microsystem with “passive robustness.” Specifically, core-shell Ag@SiO₂@PVP nanocomposites were first constructed: the inner SiO₂ shell effectively isolates the Ag core from corrosive agents like H⁺ and OH⁻, while the outer polyvinylpyrrolidone (PVP) layer prevents particle aggregation through steric hindrance. This ensures excellent probe dispersibility across a wide pH range, fundamentally creating a SERS signal source unaffected by environmental pH. Secondly, a T-junction microdroplet detection chip integrated with a herringbone-shaped mixer was designed, and a complete microdroplet SERS platform was established. This platform enables an integrated workflow encompassing rapid mixing of metabolites and nanoprobes, generation of homogeneous droplets, automatic trapping, and in-situ SERS detection. Systematic evaluation using the E. coli metabolite L-DOPA as a model molecule demonstrated that the platform achieves both high reproducibility (RSD < 5%) and high-throughput detection capability across the pH 3–11 range, without requiring any external active regulation. Validation using real fermentation broth further confirmed its practical potential. A comprehensive comparison with existing technologies (Supporting Information Table S1) reveals significant advantages of this work in terms of wide pH stability, reproducibility, detection throughput, and the absence of external intervention. In summary, this work aims to provide a robust and versatile solution for real-time, high-throughput analysis of metabolites in complex biological environments through the synergistic strategy of innovative probe material design and microsystem integration.

Experimental section

Materials and Instruments

Silver nitrate (AgNO₃, 99.8%) and trisodium citrate (C₆H₅Na₃O₇·2H₂O, 98%) were purchased from Chengdu Kelong Chemical Co., Ltd. (Chengdu, China). Levodopa (L-DOPA, C₉H₁₁NO₄, 99%), polyvinylpyrrolidone (PVP, average molecular weight 10,000, K13-18), and tetraethoxysilane (TEOS, (C₂H₅O)₄Si, ≥ 99.99%) were purchased from Shanghai Maclin Biochemical Technology Co., Ltd. (Shanghai, China). SU-8 negative photoresist was purchased from Xi’an Boyan Micro-Nano Information Technology Co., Ltd. (Xi’an, China). SYLGARD 184 polydimethylsiloxane (PDMS) and its matching curing agent were purchased from Shanghai Tengyan Chemical Trading Co., Ltd. (Shanghai, China).

The Raman detection system utilizes a 633 nm laser (Changchun New Industries Optoelectronics Technology Co., Ltd., China); Optical components and lens mounts were purchased from Base Optics Technology Co., Ltd. (Guangzhou, China); optical elements were sourced from Solarex Optoelectronics Technology (Shanghai) Co., Ltd.; optical fibers were supplied by Yangtze Optical Fibre and Cable Joint Stock Co., Ltd. (Wuhan, China); micro-syringe pumps were acquired from Baoding DICON Electronics Technology Co., Ltd. (China); plasma cleaners were purchased from Zhengzhou Ketan Instrument Equipment Co., Ltd. (China); and experimental water was prepared using the Smart-S ultrapure water system from Shanghai Hetao Instruments Co., Ltd.

Microfluidic chip fabrication

PDMS microfluidic chips were fabricated using soft photolithography (Supporting Information Fig. S1). All chip channels were uniformly 70 μm deep. The herringbone-shaped mixing unit comprised equilateral triangles, with a premixing channel width of 500 μm, subsequent serpentine bending channels at 300 μm, and all other channels at 150 μm. The chip uses a mirror substrate instead of a glass substrate for bonding, further improving the efficiency of Raman signal acquisition. To verify the contribution of the mirror substrate, we designed a control experiment using a 10⁻⁶g/mL Rhodamine 6 G solution to detect Raman signals on both the reflective chip and a standard PDMS chip (with a glass substrate).

Numerical simulation

Silver Using COMSOL Multiphysics software, simulations were conducted on the mixing, droplet generation, and capture processes of a three-component aqueous sample in a T-shaped reflective microfluidic chip. Mathematical relationships were established between droplet diameter D, capture time (i.e., Raman exposure time T_drop), and the continuous phase flow velocity U_c, as well as the dispersed phase flow velocity U_d (the sum of the flow velocities of the three aqueous samples). To verify the accuracy of the simulation, we compared the experimentally observed droplet diameters with the simulated predictions under three representative flow velocity conditions.

Microdroplet SERS platform design

The 633 nm laser output is transmitted through multimode fiber, collimated and filtered by a collimator and bandpass filter (Thorlabs), then directed into a 633 nm cage-type Raman probe. A dichroic beamsplitter (Chroma) within the probe reflects the laser onto a 50× objective lens (Olympus LMPLFLN 50X). The objective lens is precisely adjusted using a displacement stage to focus the laser beam onto the region where droplets are captured on the chip, enabling in-situ Raman detection. The Raman signal is collected by the same objective lens. After returning to the probe, it passes through a longpass filter (Thorlabs) to remove Rayleigh scattered light. Finally, the signal is guided through an optical fiber into a spectrometer for spectral acquisition and data processing. The chip’s specular-reflective substrate efficiently reflects the generated Raman scattered light back into the upward collection path, thereby enhancing signal collection efficiency.

Preparation of SERS-Enhanced Nanoparticles

Core-shell structured silver-coated silica nanoparticles (Ag@SiO₂ NPs) were employed as SERS substrates. Their preparation was based on pre-synthesized silver nanoparticles Ag NPs (Based on the method of Lee and Meisel³¹), This study used 60 nm Ag cores as the standard size for subsequent research³². See Supporting Information Fig. S2 for a schematic diagram of the process of preparing Ag@SiO₂ NPs³³. To enhance the dispersion and stability of nanoparticles within the chip, dynamic modification with PVP was employed. Dissolve 100 mg of solid PVP in 10 mL of deionized water, stirring and sonicating for 30 minutes to ensure complete dissolution. Prepare a 10 g/L PVP aqueous solution for later use³⁴.

Characterization of nanoparticles

To validate the structural integrity and parameter controllability of the materials, systematic characterization was performed on the synthesized Ag NPs, Ag@SiO₂ NPs with varying silica shell thicknesses, and Ag NPs, Ag@SiO₂ NPs, and Ag@SiO₂@PVP NPs treated under different pH conditions (pH 3, 7, 11). Morphology observation and particle size measurement were performed using a scanning transmission electron microscope (SU8600, Hitachi) and a field emission transmission electron microscope (Talos F200S, Thermo Fisher Scientific). Surface elemental composition and chemical states were analyzed by X-ray photoelectron spectroscopy (ESCALAB 250Xi, Thermo Fisher Scientific). The Zeta potential and hydrated particle size were determined using a nanoparticle size and Zeta potential analyzer (NanoBrook Omni, Brookhaven Instruments) to evaluate colloidal stability. Optical properties were characterized by obtaining ultraviolet-visible absorption spectra using a UV-visible-near-infrared spectrophotometer (UV-3600, SHIMADZU).

Sample preparation for metabolites under investigation

This study used the neuropharmacological precursor levodopa as the model analyte. The standard powder was dissolved in ultrapure water to prepare aqueous solutions with a concentration gradient ranging from 10⁻⁸ to 10⁻⁴ g/mL, which were reserved as standard samples.

The actual metabolite samples were provided by the Tianjin Institute of Industrial Biotechnology, Chinese Academy of Sciences. These were fermentation broths containing levodopa produced by Escherichia coli metabolism, used directly as unlabeled detection samples³⁵.

A single colony was inoculated into 2 mL of LB medium and incubated overnight at 37 °C and 250 rpm. The inoculum was then transferred to a 1-L shake flask containing 200 mL of LB medium and incubated for 12 hours at 37 °C and 250 rpm. Finally, the entire inoculum was transferred to a 5-L bioreactor (BIOTECH-5BG, Bxbio) containing 1.8 L of fermentation medium (composition: 30 g/L glucose, 1.6 g/L(NH₄)₂SO₄, 7.5 g/L K₂HPO₄·3H₂O, 2 g/L citric acid, 2 g/L MgSO₄·7H₂O, 4 mg/L CoCl₂·6H₂O, 0.0075 g/L FeSO₄·7H₂O, 20 mg/L Na₂SO₄, 0.6 mg/L Cu₂SO₄·5H₂O, 6.4 mg/L zinc sulfate, and 4.525 mg/L manganese sulfate hydrate) for fermentation. After 60 hours of fermentation, the concentration of L-DOPA in the fermentation broth was determined using high-performance liquid chromatography (HPLC). Actual fermentation broths of known concentrations were diluted to create samples with a concentration gradient ranging from 10⁻⁸ to 10⁻⁴g/mL. The fermentation pathways for L-tyrosine and arbutin were the same as described above.

Surface-enhanced Raman spectroscopy analysis

To validate that the optimized PVP modification strategy provides stability without compromising Raman activity, testing was conducted under neutral conditions using Rhodamine 6 G (R6G, 10⁻⁶mol/L) as the probe molecule on a self-built droplet SERS platform. R6G, Ag@SiO₂ NPs (silica shell thickness 5 ± 3 nm), PVP solution, and fluorinated oil (HFE-7500) were injected onto the chip via a syringe pump at preset flow rates (Table S1) Signals were collected upon excitation by a 633 nm laser (50 mW power, 500 ms exposure time).

To evaluate the SERS performance of Ag@SiO₂@PVP nanoparticles and their signal stability under varying pH conditions, real-time Raman detection of microdroplets was performed on levodopa aqueous solutions with concentration gradients ranging from 10⁻⁸g/ml to 10⁻⁴ g/ml at pH 3, 7, and 11. Standard samples, Ag@SiO₂ NPs (silica shell thickness 5 ± 3 nm), PVP solution, and HFE-7500 were injected onto the chip via a syringe pump at predetermined flow rates (Table S2). During detection, the system integration time was set between 15 and 500 ms, the corresponding droplet throughput is 120–4,000 drops per minute. The limit of detection (LOD) is calculated using the 3σ method: Measure the signal intensity at 1400 cm⁻¹ for 30 blank samples, calculate the standard deviation (σ), and take the concentration value on the calibration curve corresponding to three times the standard deviation.

Droplet images were acquired using an OLYMPUS BX51 microscope. Raman signal data processing and plotting were performed using OriginPro software.

Results and discussion

Establishment of a microdroplet SERS platform

To achieve stable and efficient detection of microbial metabolites in complex fermentation broths, this study designed and fabricated a T-junction reflective microfluidic chip integrated with a herringbone-shaped mixer (Supporting Information Fig. S3). As illustrated in Fig. 1a, the chip integrates three functional zones: a herringbone-shaped mixer mixing unit, a homogeneous droplet generation zone, and an active droplet trapping and in-situ SERS detection zone. The operational workflow is as follows: three aqueous phases-the metabolite sample, PVP dispersant, and Ag@SiO₂ nanoparticles-are injected through inlet (1) as shown in Fig. 1a. They undergo rapid and thorough mixing facilitated by the herringbone-shaped within the serpentine mixing channel (2), forming the dispersed phase. The continuous phase (HFE-7500 fluorinated oil) is introduced through inlet (3). By precisely controlling the flow rates of the two phases, uniformly sized droplets are generated at the T-junction (4). These droplets then flow into the optical detection zone (5), where they are automatically trapped and stably retained by the unique variable cross-section channel design. Here, they are excited by a 633 nm laser for Raman signal acquisition. Finally, the droplets from which measurements have been taken exit the chip via outlet (6) (see Supporting Information Fig. S4 for the dynamic process).

Fig. 1. Optimization of microfluidic chip parameters.

a Schematic diagram of droplet microfluidic chip structure and corresponding microscopic images: 1) Three aqueous phase (dispersed phase) inlets; 2) Sample mixing zone; 3) Oil phase (continuous phase) inlet; 4) Droplet generation zone; 5) Droplet capture and Raman detection zone; 6) Droplet outlet; COMSOL Multiphysics Simulation. b COMSOL Multiphysics simulation results showing the mixing of three-component aqueous samples, droplet generation, and capture process (droplet diameter: 2.2 × 10⁻⁴m). c Relationship between droplet diameter D and continuous phase flow velocity U_c, dispersed phase flow velocity U_d. d Relationship between droplet diameter D and droplet capture time (Raman exposure time) T_drop in the detection zone

The flow rates of the three aqueous phases are critical parameters that directly determine the uniformity and stability of droplet generation, as well as the reliability of droplet trapping within the detection zone, serving as the cornerstone for achieving efficient detection and complete Raman signal acquisition. Numerical simulations provide a theoretical basis for optimizing these parameters. The simulation results in Fig. 1b reveal the residence time of droplets in the trapping zone (which corresponds to the required Raman exposure time), offering a key reference for precisely matching signal acquisition with droplet dynamics. Furthermore, as shown in Fig. 1c, the droplet diameter D exhibits a regular dependence on the flow rates of the continuous phase U_c and the dispersed phase U_d, enabling precise control over droplet size by modulating these flow rates. Droplet size directly determines the extent of deformation and spreading within the variable cross-section trapping channel, thereby influencing the droplet residence time T_drop as shown in Fig. 1d. We systematically calculated the combinations of U_c and U_d required for different T_drop values (Supporting Information Eqs. S1–S9), providing clear guidance for flow rate control during experiments (Supporting Information Table S2). To validate the simulation accuracy, we compared droplet diameters observed via microscopy with those predicted by COMSOL simulations under three representative flow rate conditions. The results showed good agreement (Supporting Information Fig. S5), confirming the value of simulations in guiding experimental parameter selection.

Guided by the optimized design parameters, stable chip operation was successfully achieved. Experiments demonstrated that the three aqueous phases could be rapidly mixed within the herringbone-shaped mixing structure (see Supporting Information Fig. S4 and Video S1). By controlling the flow rates of the aqueous and oil phases, the droplet size and generation frequency could be effectively tuned, and stable droplet trapping was achieved in the chip’s capture zone (see Supporting Information Fig. S4 and Video S2). These results laid a solid foundation for subsequent reliable SERS detection.

Based on the microfluidic chip described above, this paper successfully constructed a highly integrated droplet microfluidic SERS detection platform. Structurally, the platform can be divided into two core functional modules: the Raman signal excitation module and the Raman signal collection module. The overall layout and optical path design are shown in Fig. 2 and Supporting Information Fig. S6. The core components of the Raman signal excitation module include the excitation fiber, collection fiber, 633 nm cage-style Raman probe, and the microfluidic chip mounted on a three-axis translation stage. The Raman signal collection module is integrated into a standalone chassis as shown in Fig. 2a, housing a 633 nm laser, a Raman spectrometer, a microcomputer, and a display screen, achieving a compact and integrated system.

Fig. 2. Schematic diagram of the microdroplet SERS platform principle.

a SERS signal collection module. b For droplet SERS signal detection module

The cage-style Raman probe (indicated by the orange dashed box in Fig. 2) integrates all optical components, including fiber collimators, into a standard 30 mm cage system (Supporting Information Fig. S7), significantly enhancing the mechanical stability and anti-interference capability of the external optical path. The optical system exclusively utilizes optical fiber transmission for both the excitation laser (depicted in red in Fig. 2) and the Raman scattering signal (depicted in blue in Fig. 2). This design allows for more flexible placement of the laser and signal collection system (green dashed box), further reducing the overall footprint. The SERS-based T-junction reflective chip in Fig. 2b is fixed on a three-axis translation stage, allowing fine adjustments to precisely align the focal point of the probe’s objective lens with the chip’s detection area.

During the detection process, the metabolite sample, PVP dispersant, and Ag@SiO₂ nanoparticles are separately injected into the chip. They mix rapidly within the microchannel, forming SERS “hot spots,” and are subsequently encapsulated into individual droplets. Upon entering the circular detection zone, the droplets are automatically captured and stably retained by the variable cross-section channel structure. At this point, the 633 nm excitation laser penetrates the PDMS layer and precisely illuminates the trapped droplet. The unique chip design, where the PDMS microchannel layer is bonded to a reflective substrate, efficiently reflects the generated Raman scattered light back towards the collection optics above, thereby enhancing signal collection efficiency^36,37. A comparison of R6G Raman signal intensities between the reflective chip and a standard chip (Supporting Information Fig. S8) showed that the signal intensity from the reflective chip was approximately 2.3 times higher (Supporting Information Eqs. S10-12), confirming the significant enhancement of signal collection efficiency afforded by the mirrored substrate. Finally, the Raman optical signal is transmitted in real-time to the spectrometer for analysis, with the corresponding spectrum simultaneously displayed on the screen. As demonstrated in Video S3, once a droplet exits the detection area, its corresponding spectrum disappears automatically, intuitively confirming the in-situ nature and reliability of the signal acquisition.

Characterization of Ag@SiO₂ NPs-enhanced particles

To enhance the stability of silver nanoparticles (Ag NPs) in complex pH environments, we successfully synthesized core-shell silica-coated silver nanoparticles (Ag@SiO₂). This structure provides physical isolation of the Ag core by the SiO₂ shell^38–40, effectively preventing direct contact between the silver core and the metabolic environment. This strategy fundamentally circumvents the issues of Ag NP oxidation and SERS performance degradation caused by pH fluctuations^41–45.

The as-prepared nanoparticles were systematically characterized for their morphology and structure. The SEM images in Fig. 3a show that the pristine Ag NPs are relatively spherical with an average diameter of 60 ± 5 nm. After coating with SiO₂ (using a 5 ± 3 nm silica shell as an example, Fig. 3b), the particles retain their spherical morphology, and the particle size increases uniformly with increasing shell thickness (Supporting Information Fig. S9). Further TEM and EDS analysis (Supporting Information Fig. S10) confirmed the successful construction of the core-shell structure. The TEM images of Ag@SiO₂ NPs (with a silica shell thickness of 5 ± 3 nm) (Fig. 3c–e) clearly reveal an approximately 60 nm Ag core (depicted as dark/pink regions) completely encased by a uniform and distinctly delineated SiO₂ shell (depicted as light/cyan regions), demonstrating the reliability of this synthesis strategy.

Fig. 3. Characterization of enhanced particles.

a SEM morphology of Ag NPs. b SEM image of Ag@SiO₂ NPs (silicon layer thickness 5 ± 3 nm). c TEM image of Ag@SiO₂ NPs (silicon layer thickness 5 ± 3 nm). d Elemental distribution map of Ag@SiO₂ NPs (silicon layer thickness 5 ± 3 nm). e High-resolution TEM image of the silica layer in Ag@SiO₂ NPs (silica layer thickness: 5 ± 3 nm). f UV-Vis absorption spectra of Ag NPs and Ag@SiO₂ NPs with different silica shell thicknesses (5 ± 3 nm, 10 ± 3 nm, 20 ± 3 nm, and 40 ± 3 nm) (insets show physical photographs of the samples). g DLS Testing and Curve Fitting. h Zeta potential and fitting curve. i XPS spectrum of Ag 3 d. j XPS spectrum of Si 2p (Note: No Si 2p signal is present in Ag NP samples). k Raman spectrum of R6G enhanced by Ag@SiO₂ NPs with different silica shell thicknesses. l Raman intensity comparison of R6G at 1511 cm⁻¹. Each data set was tested three times; error bars represent the mean and standard deviation of the three measurements

We subsequently investigated the influence of the SiO₂ shell on the physicochemical properties of the nanoparticles. UV-Vis absorption spectra Fig. 3f indicate that the shell thickness significantly modulates the LSPR properties: compared to Ag colloid (LSPR peak at 420 nm), coating with SiO₂ results in a red shift (to 452 nm), while further increasing the shell thickness leads to a blue shift (to 443 nm). This phenomenon aligns with the Mie theory for core-shell particles^46–48. Dynamic light scattering (DLS) results as shown in Fig. 3g and Fig. S11 demonstrate that both Ag NPs and Ag@SiO₂ NPs exhibit uniform size distribution and good dispersity, meeting the requirements for subsequent applications⁴⁹. Zeta potential analysis as shown in Fig. 3h and Fig. S12 reveals that the absolute surface charge of Ag@SiO₂ NPs is higher than that of bare Ag NPs. This originates from the negative charge introduced by the SiO₂ shell, which enables effective inhibition of particle aggregation over a wider pH range through enhanced electrostatic repulsion. X-ray photoelectron spectroscopy (XPS) depth profiling in Fig. 3i, j further confirms that the surface chemical state of Ag@SiO₂ NPs is predominantly silicon, indicating complete encapsulation of the silver core by the SiO₂ layer.

Finally, using R6G (10⁻⁶mol/L) as a probe molecule, we systematically evaluated the SERS enhancement efficiency of Ag@SiO₂ NPs with different silica shell thicknesses (5 ± 3 nm, 10 ± 3 nm, 20 ± 3 nm, and 40 ± 3 nm) as shown in Fig. 3k, l. The results demonstrate that thinner silica shells yield superior Raman enhancement performance, consistent with literature reports⁵⁰. Based on these findings, Ag@SiO₂ NPs with a silica shell thickness of 5 ± 3 nm were ultimately selected as the SERS-enhancing substrate for all subsequent studies in this paper.

Mechanism exploration of the dual-synergistic stabilization mechanism

To ensure the long-term stability of SERS probes against pH fluctuations in complex metabolic environments, this study introduced PVP as a dispersant^51,52 into the thin-shell Ag@SiO₂ NP system, constructing Ag@SiO₂@PVP nanocomposites with a dual synergistic protection mechanism. As illustrated in Fig. 4a, unlike Ag NPs that rely solely on electrostatic stabilization (which are prone to aggregation due to electrical double layer compression caused by pH changes)⁵³, the non-ionic steric hindrance mechanism of PVP is highly insensitive to variations in pH and ionic strength. This endows the system with superior anti-interference performance compared to purely electrostatic stabilization. The stability of this structure arises from the following synergistic effects: the inner SiO₂ shell acts as a physical barrier, effectively isolating the silver core from chemical attack by H⁺ and OH⁻ while providing a stable negative surface charge; The outer PVP layer confers colloidal stability through steric hindrance⁵⁴. PVP, as a non-ionic polymer, achieves stabilization primarily via steric hindrance: when PVP-coated Ag@SiO₂ particles approach each other, their hydrophobic carbon chains extending into the solvent interact, causing overlap of the polymer layers and creating a physical spatial barrier between particles. This overlap increases the local segment concentration, generating an osmotic pressure difference that drives solvent molecules to diffuse into the overlap region, producing a repulsive force (i.e., osmotic repulsion)^51,55. Furthermore, PVP adsorbs onto the particle surface via its carbonyl groups, and its physicochemical properties remain inert across a wide pH range, thereby significantly enhancing the dispersion stability of the colloid over a broad pH spectrum⁵¹.

Fig. 4. Stabilization mechanism exploration.

a Schematic illustration of the dual synergistic stabilization mechanism. b Photographs of solutions containing Ag NPs, Ag@SiO₂ NPs, and Ag@SiO₂@PVP NPs at different pH values. UV-Vis absorption spectroscopy of (c) Ag NPs, (d) Ag@SiO₂ NPs, and (e) Ag@SiO₂@PVP NPs at different pH values. f SERS spectra of R6G enhanced by different particles. g SERS signal intensity at 1511 cm⁻¹ for randomly selected droplets within 30 min. Each data set was tested three times; error bars represent the mean and standard deviation of three measurements. h Reusability study of the Reusability study of microdroplet SERS platforms: trend of R6G (1511 cm⁻¹) SERS intensity variation under different cycle numbers

This study systematically validated the effectiveness of this design through experiments. As shown in Fig. 4b, across the pH range of 3 to 11, Ag NPs exhibited severe aggregation, while both Ag@SiO₂ NPs and Ag@SiO₂@PVP NPs remained dispersed. UV-Vis absorption spectroscopy, DLS, and Zeta potential analyses (Fig. 4c–e, Figs. S13, S14) further demonstrated that Ag NPs showed attenuated absorption signals, increased particle size, and Zeta potentials approaching zero under extreme pH conditions. Ag@SiO₂ NPs displayed improved stability but remained insufficient under harsh conditions. In contrast, Ag@SiO₂@PVP NPs exhibited exceptional stability under all tested conditions. Microscopy and SEM images (Fig. S15) visually confirmed that the introduction of PVP significantly improved the dispersity of Ag@SiO₂ NPs, reducing aggregation and exposing more Raman hot spots for interaction with target molecules, thereby ensuring detection sensitivity and reliability.

It is worth emphasizing that this optimization strategy confers stability while maintaining excellent Raman enhancement performance. As shown in Fig. 4f, on-chip experiments under neutral conditions revealed that Ag@SiO₂@PVP NPs exhibited superior enhancement effects for R6G compared to Ag@SiO₂ NPs, and far exceeded that of Ag NPs. This is primarily attributed to the significant improvement in colloidal dispersity afforded by PVP: in the absence of PVP, Ag@SiO₂ NPs may still undergo local aggregation within microfluidic channels, shielding portions of the silver cores and reducing the effective “hot spot” area. PVP, through steric hindrance, enables the particles to exist in a highly monodisperse state, thereby exposing more abundant and uniform SERS-active surfaces^56,57. The enhancement factor reached 1.2 × 10⁷ (Supporting Information Eq. S13), and PVP itself generated no interfering signals within the Raman detection window. Furthermore, the reflective microfluidic chip demonstrated excellent reusability: six consecutive sets of droplet SERS signals acquired over 30 minutes showed high consistency (Fig. 4g), and no residual signals were observed after six cycles of use (Fig. 4h), indicating good anti-fouling properties of the chip. In summary, through the synergistic design of “SiO₂ physical isolation + PVP steric hindrance,” Ag@SiO₂@PVP successfully establishes an intrinsically stable, pH-resistant SERS probe, providing a robust foundation for reliable metabolite analysis in complex biological environments.

SERS analysis of Levodopa across a wide pH range

To systematically evaluate the analytical performance of the synergistic stabilization-based microdroplet SERS platform utilizing Ag@SiO₂@PVP NPs, we detected L-DOPA standard aqueous solutions across a concentration range of 10⁻⁸g/mL to 10⁻⁴g/mL under pH conditions of 3, 7, and 11 (Fig. 5a, e, i), with a uniform exposure time of 500 ms. Under all pH environments, the SERS spectra acquired at gradient concentrations clearly resolved the characteristic fingerprint peaks of L-DOPA^58,59: The band at 485 cm⁻¹ is attributed to the asymmetric stretching vibration of the benzene ring. The peak at 708 cm⁻¹ corresponds to C-C stretching vibrations of the L-DOPA benzene ring, with contributions from C = O in-plane bending and broader C-C stretching modes. The band at 949 cm⁻¹ arises from C-C-H in-plane bending and C-N-H in-plane bending vibrations. The peaks at 1400 cm⁻¹ and 1597 cm⁻¹ are assigned to ring deformation and C = C stretching modes of the benzene ring, respectively. These results demonstrate the platform’s capability to effectively acquire fingerprint SERS signals of L-DOPA across a broad pH range.

Fig. 5. SERS analysis results for L-DOPA.

a SERS spectrum of L-DOPA concentration gradient at pH = 3 (exposure time 500 ms). b Raman intensity spectra of L-DOPA at 1400 cm⁻¹ at different concentrations when pH = 3. c Corresponding calibration curve of SERS intensity versus -log(C_L-DOPA), where SERS intensity was recorded at 1400 cm⁻¹. d Intensity distribution of the Raman peak at 1400 cm⁻¹ for 20 randomly selected droplets at an L-DOPA concentration of 10⁻⁸g/ml (exposure time 500 ms). e SERS spectrum of L-DOPA concentration gradient at pH = 7 (exposure time 500 ms). f Raman intensity spectra of L-DOPA at 1400 cm⁻¹ at different concentrations when pH = 7. g Corresponding calibration curve of SERS intensity versus -log(C_L-DOPA), where SERS intensity was recorded at 1400 cm⁻¹. h Intensity distribution of the Raman peak at 1400 cm⁻¹ for 20 randomly selected droplets at an L-DOPA concentration of 10⁻⁸g/ml (exposure time 500 ms). i SERS spectrum of L-DOPA concentration gradient at pH = 11 (exposure time 500 ms). j Raman intensity spectra of L-DOPA at 1400 cm⁻¹ at different concentrations when pH = 11. k Corresponding calibration curve of SERS intensity versus -log(C_L-DOPA), where SERS intensity was recorded at 1400 cm⁻¹. l Intensity distribution of the Raman peak at 1400 cm⁻¹ for 20 randomly selected droplets at an L-DOPA concentration of 10⁻⁸g/ml (exposure time 500 ms). Each data set was tested three times. error bars represent the mean and standard deviation of the three measurements

Further analysis revealed that, under different pH conditions, the intensity of the characteristic L-DOPA peak at 1400 cm⁻¹ decreased systematically with decreasing concentration (Fig. 5b, f, j). More importantly, the peak intensity exhibited an excellent linear relationship with the negative logarithm of concentration (-log C) at pH 3, 7, and 11 (R² > 0.99), with the three calibration curves showing substantial overlap and similar slopes as shown in Fig. 5c, g, k. This result directly confirms that the developed SERS probe and detection platform successfully overcome the core challenge of signal instability in SERS technology under pH fluctuations. The limit of detection (LOD) was calculated based on the 3σ method (Supporting Information Fig. S16, Eqs. S14-15). By measuring the Raman intensity at 1400 cm⁻¹ for 30 blank samples at pH 7, the LOD was determined to be 9.5 × 10⁻⁹g/mL.

The reproducibility and stability of the platform were also systematically validated. SERS signals were randomly collected from 20 droplets under each pH condition (Fig. 5d, h, l). The relative standard deviations (RSD) of the characteristic peak intensities were 4.9% (pH = 3), 4.5% (pH = 7), and 4.4% (pH = 11), all below 5%, indicating excellent signal consistency. The absolute values of the error bars were similar across different concentration levels, reflecting the stability of system operation; systematic factors such as minor fluctuations in droplet trapping position or slight shifts in the laser focus exerted relatively uniform effects on all droplets. Furthermore, to assess platform performance under high-speed acquisition modes, we conducted tests with the exposure time reduced to 15 ms. The results demonstrated that even at such a short exposure time, the characteristic Raman peaks of L-DOPA at 10⁻⁸ g/mL remained clearly identifiable (Supporting Information Fig. S17), corresponding to a droplet throughput of up to 4000 droplets per minute. This fully substantiates the platform’s excellent application potential in high-throughput, real-time detection scenarios.

SERS analysis of authentic samples

To validate the practical application potential of the synergistic stabilization-based microdroplet SERS platform utilizing Ag@SiO₂@PVP NPs, it was employed for the detection of L-DOPA in real E. coli fermentation broth. As shown in Fig. 6a, in the authentic fermentation environment (pH ≈ 3), Ag NPs were susceptible to matrix interference, exhibiting selective adsorption and oxidative denaturation, which resulted in a significant decline in their ability to resolve characteristic Raman peaks. In contrast, Ag@SiO₂@PVP NPs maintained clear spectral recognition capability across the concentration range of 10⁻⁸ ~ 10⁻⁴ g/mL. The SERS characteristic peaks from the real samples were highly consistent with those of L-DOPA standards, demonstrating the excellent detection specificity and applicability of this method in complex biological matrices. This specificity originates from two key factors: 1) the fingerprint characteristics of SERS enable the unique characteristic peaks of L-DOPA to be identified against the background of impurities; 2) the inherent chemical stability of the probe prevents the introduction of complex background interference caused by probe degradation, starkly contrasting with the failed case of bare Ag NPs.

Fig. 6. Microdroplet SERS detection and analysis of authentic strain metabolites.

a SERS spectrum enhanced by Ag NPs. b SERS spectrum of L-DOPA concentration gradient enhanced by Ag@SiO₂ NPs. c Raman intensity at 1400 cm⁻¹ for different concentrations of L-DOPA. Each data set was tested three times. Error bars represent the mean and standard deviation of three measurements. d Calibration curve of SERS intensity at 1400 cm⁻¹ versus -log(C_L-DOPA). e SERS spectra of 10⁻⁴ g/ml L-DOPA collected every 5 min over 60 min. f Distribution of the characteristic peak intensity at 1400 cm⁻¹ for the Raman signal of 10⁻⁴g/mL L-DOPA, collected every 5 min over a 60 min period

Although this method primarily targets rapid detection, its semi-quantitative potential was also evaluated. Gradient concentration samples of actual fermentation broth (detailed in the Experimental section, representing real fermentation broth with L-DOPA concentrations verified by HPLC) were analyzed using the microdroplet SERS platform. The results revealed a clear positive correlation trend between the SERS signal intensity and the L-DOPA concentration determined by HPLC as shown in Fig. 6b, demonstrating the method’s utility for rapidly distinguishing samples with different concentration levels. Furthermore, the calibration curve constructed between the characteristic peak intensity and -log C exhibited good linear correlation (R² > 0.97, Fig. 6c, d), further confirming its semi-quantitative potential.

To assess the operational stability of the platform, a fermentation broth sample with a concentration of 10⁻⁴ g/mL was continuously monitored for 60 minutes, with data acquired every 5 min (Fig. 6e). The results showed that the intensity of the characteristic L-DOPA peak at 1400 cm⁻¹ remained stable, with an RSD of 9.7% (Fig. 6f), confirming the platform’s satisfactory stability and repeatability for continuous monitoring of real samples. To further validate the platform’s universality, E. coli metabolites L-tyrosine (Tyr) and arbutin were employed as second and third model analytes, detected under their respective authentic fermentation conditions. The results demonstrated that the platform could also sensitively detect samples at 10⁻⁸g/mL, with calibration curves between characteristic peak intensity and -log C showing good linear correlation (R² > 0.94, Supporting Information Fig. S18).

These results indicate that the dynamic dispersion strategy based on Ag@SiO₂@PVP NPs, coupled with the microdroplet SERS platform, can effectively achieve accurate identification of specific metabolites in complex samples. Through material-system co-design, this study has successfully constructed a microsystem exhibiting excellent performance in terms of pH interference resistance, detection sensitivity, and analytical throughput. Unlike common approaches that create novel analytical capabilities by introducing new operating principles or functional modules into the microsystem architecture^60–62, this work, while achieving superior analytical performance, simultaneously enhances the overall robustness and reliability of the system in complex environments from another dimension. This is of significant value for optimizing biomanufacturing processes.

Conclusion

In this study proposed and developed a microdroplet SERS detection platform based on Ag@SiO₂@PVP nanoprobes for high-throughput, automated real-time analysis of microbial metabolites. Through the dual protection mechanism of “SiO₂ physical isolation + PVP steric hindrance,” the core challenge of signal instability inherent to conventional silver nanoparticles under pH-fluctuating environments was effectively addressed. The designed multifunctional reflective microfluidic chip integrated rapid mixing, droplet generation, in-situ trapping, and SERS detection, enabling integrated operation. Through the synergistic design of probe material innovation and chip integration, the platform exhibited excellent analytical performance-achieving a Raman enhancement factor of 1.2 × 10⁷, a limit of detection as low as 9.5 × 10⁻⁹ g/mL (based on 3σ calculation), and a minimum exposure time of only 15 ms (corresponding to a droplet throughput of 4000 droplets per minute). Furthermore, the platform demonstrated exceptional stability across the pH 3–11 range (R² > 0.99, RSD < 5%) and was successfully applied for the detection of the metabolite L-DOPA in real E. coli fermentation broth. Preliminary tests with L-tyrosine and arbutin also validated its general applicability to a certain extent. Moreover, compared to existing pH control strategies, this work achieves “regulation-free” wide pH tolerance through the inherent stability of the materials, avoiding sample disturbance caused by external intervention and offering significant advantages in detection repeatability and efficiency. Therefore, this microdroplet SERS platform holds significant application potential as a real-time analysis technology for microbial metabolites.

Author contributions

Haixia Zhao performed the experiments and completed the manuscript. Jizhuo Liu designed and fabricated the platform structural housing and revised the manuscript. Huiling Yuan provided samples of metabolites from actual microbial strains. Yi Duan and Ping Wen participated in the characterization experiments. Yi Xu, Qinhong Wang, and Li Chen provided project guidance and assisted in the preparation of the manuscript. All authors reviewed and approved the final manuscript. Special thanks to Prof. Qinhong Wang and Dr. Huiling Yuan of Tianjin Institute of Industrial Biotechnology, Chinese Academy of Sciences for their guidance and assistance.

Conflict of interest

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41378-026-01311-3.