Real-time nanomolar vitamin monitoring in sweat using an electrochemical skin-attached device

Abstract

Invasive and inconvenient blood tests hinder the diagnosis and treatment of micronutrient deficiencies, which affect billions worldwide. Wearable sweat sensors offer a noninvasive and real-time alternative. However, on-body monitoring of multiple low-abundance vitamins in sweat remains challenging. Here, we show a wearable electrochemical platform that utilizes bioreceptor-functionalized gold nanoflower/sulfur and nitrogen codoped carbon to detect six low-concentration vitamins in sweat (vitamins B₁, B₂, B₇, B₉, B₁₂, and D) with nanomolar-level sensitivity. Integrated system incorporates iontophoretic sweat induction, microfluidic sampling and real-time vitamin detection with calibration. Human studies revealed temporal profiles of VB₉ levels in sweat in response to oral VB₉ supplementation and VB₉-rich diets, and a strong sweat-serum correlation (r = 0.849) has been found. The significant differences in VB₉ levels observed between smokers and nonsmokers demonstrated the potential of this technology for vitamin monitoring in diverse populations. Our platform facilitates early detection of nutritional imbalances and advances personalized nutrition.

Nutrient deficiencies are widespread, but blood tests are often inconvenient. Here, the authors develop a wearable electrochemical platform for noninvasive, real-time detection of multiple low-concentration vitamins in sweat and validate its performance in human studies.

Introduction

Current global estimates indicate that more than 2 billion people are affected by hidden hunger, a condition characterized by adequate caloric intake alongside insufficient dietary micronutrients, leading to health impairments^1,2. Vitamins are central to this crisis; their deficiencies are intrinsically linked to a range of serious pathologies (Fig. 1a)^3–6. For instance, vitamin B₉ (VB₉, folic acid) or vitamin B₁₂ (VB₁₂) deficiency impairs cell proliferation and neural development, causing megaloblastic anemia and neural tube defects^7,8; vitamin D (VD) deficiency disrupts calcium‒phosphorus metabolism and skeletal health, leading to skeletal disorders and growth retardation in children^9,10. Compounding this problem, individual vitamin requirements vary substantially because of genetic polymorphisms, gut physiology, and lifestyle^11,12, creating an urgent need for precision nutrition strategies that can tailor interventions to meet individual needs.

Fig. 1. Design and mechanism of the wearable vitamin biosensor.

a Schematic of the pathologies linked to vitamin deficiency and the factors associated with vitamin status. b Schematic of the affinity-based electrochemical sensor for vitamin detection in human sweat, illustrating its construction and sensing strategy. BREs, biorecognition elements. c Multilayer design of the wearable vitamin biosensor for sweat induction via iontophoresis, sample collection, and biosensing. d Photograph of a multiplexed sensor array. Scale bar, 1 cm. e Block diagram of the electronic system of the wearable vitamin biosensor. DAC digital-to-analog converter, ADC analog-to-digital converter, UART universal asynchronous receiver-transmitter, CE counter electrode, RE reference electrode. f Photograph of the FPCB and the flexible wearable sensor patch. Scale bar, 1 cm.

However, the current paradigm for managing vitamin status is ill-equipped for this task. Guidelines often overlook interpersonal physiological variability, leading to widespread under- or oversupplementation^13,14. Persistent undersupplementation exacerbates deficiency-related pathologies, whereas chronic excessive intake may induce adverse effects and significantly increase specific health risks^15,16. Consequently, regular monitoring of vitamin status is critical for personalized nutritional intervention and minimizing potential risks associated with improper dosing. The gold standard for evaluating vitamin status relies on laboratory quantification of vitamins or their relevant biomarkers in blood (serum or plasma)^4,14. These assays, however, involve invasive blood collection, complex sample preprocessing, specialized instrumentation, and operation by trained personnel¹⁷. This time-consuming and intricate process inevitably delays intervention feedback, making real-time, personalized adjustments impossible.

Wearable sweat sensors have emerged as promising noninvasive alternatives for real-time monitoring of physical vital signs^18,19. Recent advances in wearable sweat sensors have demonstrated that sweat analysis allows real-time health monitoring^20–23, with some biomarkers showing strong correlations with blood levels^24–27. For instance, sweat uric acid levels increase significantly following a purine-rich diet, and the serum-sweat correlation supports its potential use as a biomarker for gout management²⁸. The levels of glucose^29,30 and amino acids^31–33 in sweat increase rapidly after food or supplement intake; their strong serum-sweat correlations indicate considerable potential for monitoring the risk of metabolic syndrome. However, detecting trace vitamins (including VB₉, VB₁₂, and VD) in sweat remains challenging due to their low physiological concentrations and the complexity of the sweat matrix^21,31. Current strategies such as enzymatic electrodes or direct oxidation of electroactive molecules lack the sensitivity required for low-concentration sweat vitamin monitoring^28,34,35. Therefore, a highly sensitive sensing platform is needed to detect trace amounts of vitamins in sweat. It should assess dynamic profiles and their correlations with serum levels, thereby facilitating noninvasive, real-time assessment of vitamin status in daily life.

In this work, we present a high-sensitivity wearable electrochemical sensor comprised of a unique combination of gold nanoflower/sulfur and nitrogen codoped carbon (AuNF/SNC) and affinity-based electrochemical sensing technology (Fig. 1b). AuNF/SNC exhibits exceptionally fast electron mobility and has a high specific surface area, providing an optimal platform for the immobilization of vitamin-specific biorecognition elements (BREs, antibodies/binding proteins) to achieve high-sensitivity detection. In this system, detection is performed using affinity-based electrochemical vitamin sensors, in which sweat vitamins and horseradish peroxidase-conjugated vitamins (HRP-vitamin) compete for binding to the BREs immobilized on AuNF/SNC electrodes. The resulting enzymatic reduction of hydrogen peroxide (H₂O₂) mediated by hydroquinone (HQ) generates a cathodic current that is inversely proportional to the vitamin concentration in sweat. The modular biosensing interface enables the facile replacement of BREs to quantify several trace vitamins, including VB₁, VB₂, VB₇, VB₉, VB₁₂, and VD (Supplementary Table 1). Such multianalyte quantification addresses a critical gap in wearable monitoring by overcoming the inherent limitations of single-analyte sensors. The patch contains a polydimethylsiloxane (PDMS) reagent chamber layer with two chambers, one preloaded with HRP-vitamin and the other with HQ/H₂O₂, and reagents are delivered on demand. The integrated system also incorporates transdermal iontophoresis for sweat induction, microfluidic channels for sampling, and wireless data transmission, while real-time monitoring of pH and ionic strength enables calibration of vitamin data to mitigate interindividual sample matrix variations. Validation of the sensor against the enzyme-linked immunosorbent assay (ELISA) reference reveals a strong correlation (Pearson’s correlation coefficient, r = 0.989). Given the clinical significance of VB₉, we present the temporal profiles of VB₉ levels in sweat. Our studies track the response of VB₉ levels in sweat to oral supplementation and VB₉-rich diets, revealing a strong correlation between sweat and serum (r = 0.849). This robust relationship supports the high potential of VB₉ in sweat as a noninvasive biomarker for personalized nutritional monitoring. Furthermore, the significant differences in VB₉ observed between smokers and nonsmokers highlight the potential of this technology for at-home vitamin management in diverse populations.

Results

Design and fabrication of the wearable vitamin biosensor

We developed a fully integrated wearable system for the dynamic on-body monitoring of vitamins in sweat. The system comprises a flexible, single-use sensor patch connected to a reusable flexible printed circuit board (FPCB). The disposable patch integrates several key components (Fig. 1c and Supplementary Fig. 1): a multiplexed electrochemical sensor array fabricated using low-cost, high-throughput screen printing for vitamin, pH, and ionic strength detection (Fig. 1d and Supplementary Figs. 2 and 3); iontophoretic electrodes consisting of screen-printed carbon anode and cathode electrodes arranged symmetrically and combined with iontophoresis gel for on-demand sweat extraction; a PDMS reagent chamber (Supplementary Fig. 4), and a laser-engraved microfluidic module with multiple inlets. The reusable FPCB controls the system operation by regulating iontophoresis with a constant current of 160 μA ( ≈ 2.6 μA mm^-2) for 5 min to induce sweat, after which multivariate signals are acquired in situ via a custom electrochemical analog front-end (AFE) that supports simultaneous measurement with multiple electrochemical techniques. A Bluetooth low energy (BLE) module with an integrated microcontroller unit (MCU) handles signal processing and wireless transmission to a smartphone application (Fig. 1e and Supplementary Figs. 5 and 6). Owing to its flexible design, the system maintains conformal skin contact and stable electrochemical performance during real-time on-body sweat analysis (Fig. 1f and Supplementary Movies 1–3).

Development and testing of biosensors for vitamin detection

Conventional electrochemical vitamin detection is limited by sluggish electron transfer kinetics and interference from coexisting electroactive species, limiting sensitivity and selectivity for trace vitamin detection in sweat^36,37. Affinity-based electrochemical sensing overcomes these constraints by leveraging BREs and incorporating nanomaterial-mediated signal amplification to enhance both sensitivity and selectivity^24,25,38. The preparation process of the vitamin sensor is illustrated in Fig. 2a. To enhance the electrochemical performance of the electrode, we synthesized SNC via a one-step electrochemical doping method under optimized conditions (Supplementary Fig. 7 and Supplementary Note 1). AuNFs were then electrodeposited onto SNC to enhance the electrochemical performance, resulting in the formation of a AuNF/SNC nanocomposite. The increased S 2p and N 1 s peaks in the XPS spectra confirmed the successful codoping of sulfur and nitrogen (Fig. 2b and Supplementary Fig. 8). Consistent with the successful codoping, the decrease in the I_D/I_G ratio in the Raman spectrum after sulfur and nitrogen codoping revealed an elevated defect concentration (Supplementary Fig. 9)²⁵. The abundant defect sites on the SNC surface effectively modulated the nucleation and growth of AuNFs during electrodeposition³⁹, yielding a more uniform distribution of AuNFs (Supplementary Fig. 10), thereby increasing the electrochemically active surface area and enhancing the electrode’s electrochemical performance. The peak at 289.2 eV in the XPS C 1 s spectra corresponds to surface carboxyl groups on the SNC within the AuNF/SNC, with a fraction of 4.68% (Supplementary Fig. 11). These carboxyl groups remain partially exposed and accessible after AuNFs electrodeposition, as AuNFs do not fully cover the SNC surface. Accordingly, the EDC/NHS activation step targets these carboxyl groups for covalent immobilization of BREs. The morphology and crystal structure of the AuNF/SNC nanocomposite were further characterized by scanning electron microscopy (SEM), transmission electron microscopy (TEM), atomic force microscopy (AFM), and X-ray diffraction (XRD) (Fig. 2c–f and Supplementary Fig. 12).

Fig. 2. Materials and electrochemical characterization of the multiplexed sensor.

a Schematic illustration of the fabrication of AuNF/SNC. b XPS spectra of carbon materials before and after codoping. c XRD patterns of C, SNC, and AuNF/SNC. d SEM images of AuNF/SNC. The images are representative of three independent experiments. Scale bar, 200 nm. e TEM and HRTEM images with EDS mapping of AuNF/SNC. Scale bars, 500 nm (TEM and EDS) and 5 nm (HRTEM). The images are representative of three independent experiments. f XPS patterns of Au 4 f, N 1 s, and S 2p for AuNF/SNC. g Water contact angles of C, SNC, and AuNF/SNC. h Nyquist plots of electrodes in 1× PBS solution containing 2.0 mM K₄Fe(CN)₆/K₃Fe(CN)₆ (1:1) after each material preparation step. Rs, solution resistance; Rct, charge-transfer resistance; Zw, Warburg impedance. CPE, constant phase element. Z, Z’ and Z” represent impedance, resistance and reactance, respectively. i Batch-to-batch variations in the electrochemical performance of AuNF/SNC electrodes (n = 3 biologically independent electrodes per condition). j represents the current density. Data are presented as mean values ± SD. j Nyquist plots of electrodes in 1× PBS solution containing 2.0 mM K₄Fe(CN)₆/K₃Fe(CN)₆ (1:1) after each surface-modification step. k–n DPV voltammograms of vitamin sensors for detection of VB₇ (k), VB₉ (l), VB₁₂ (m), and VD (n), recorded from −0.3 to 0.2 V vs. Ag/AgCl. I represents the current, and E represents the potential. o Calibration curve for multiple vitamins (n = 3 biologically independent electrodes per condition). Data are presented as mean values ± SD. p Comparison of the sensing performance of AuNF/SNC sensor and other wearable vitamin sensors. q Selectivity of the sensors against other vitamins (n = 3 biologically independent electrodes per condition). ΔI is the peak current change. Data are presented as mean values ± SD.

AuNF/SNC electrodes were activated with 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC) and N-hydroxysulfosuccinimide (NHS) for covalent BREs immobilization, followed by blocking nonspecific binding sites with bovine serum albumin (BSA). Competitive binding occurred when the vitamin in the sweat and the HRP-vitamin bound to the immobilized BREs. Differential pulse voltammetry (DPV) measurements were performed in the presence of HQ and H₂O₂ as the detection substrate (Supplementary Fig. 13). The hydrophilicity of the AuNF/SNC surface enhances both the antifouling performance and the biocompatibility of the electrode (Fig. 2g)⁴⁰. Open-circuit potential electrochemical impedance spectroscopy (OCP-EIS) and cyclic voltammetry (CV) were employed to characterize the electrodes after each fabrication step. In the Nyquist plots, a larger semicircle diameter corresponds to a higher charge-transfer resistance (Rct). After codoping and electrodeposition, the semicircle diameter decreased, indicating a reduced Rct. This reduction reflects improved electron transfer kinetics between the redox probes in solution and the functionalized transducer surface (Fig. 2h and Supplementary Fig. 14)²⁷. The CV profile of the AuNF/SNC composite exhibited minor redox peaks, which are attributed to surface redox processes arising from heteroatom functional groups in the SNC^41,42. The AuNF/SNC electrodes also demonstrated excellent stability and exhibited high batch-to-batch reproducibility owing to scalable fabrication processes (Fig. 2i and Supplementary Fig. 15). As expected for an affinity-based assay, each modification step increased the Rct because of hindered electron transfer kinetics. These stepwise changes in OCP-EIS and CV analyses confirmed the superior electrochemical activity and effective immobilization of affinity bioreceptors of the nanocomposite (Fig. 2j and Supplementary Fig. 16). Collectively, these analyses confirmed the successful development of the sensing platform.

The electrochemical performance of the vitamin sensors was evaluated using DPV. The peak current of reduction is inversely proportional to the sweat vitamin concentration after a competitive binding reaction between the target vitamin and its corresponding HRP-conjugated form. The AuNF/SNC sensors modified with BREs exhibit log-linear responses to six vitamins (VB₁, VB₂, VB₇, VB₉, VB₁₂, and VD) across concentrations ranging from 1 to 100 nM, which corresponds to physiologically relevant target ranges in sweat (Fig. 2k–o and Supplementary Fig. 17). Each vitamin was detected using a specific BREs, including streptavidin for VB₇, folate binding protein (FBP) for VB₉, intrinsic factor (IF) for VB₁₂, and monoclonal antibodies for VB₁, VB₂, and VD (Supplementary Table 1). Compared with SNC or AuNF/C-modified electrodes, the AuNF/SNC coating significantly increased the effective electrode surface area and enhanced BREs immobilization, resulting in 3.3-fold and 3.5-fold greater current responses towards the same vitamin concentration, respectively (Supplementary Fig. 18). The sensors achieved low limits of detection (LOD) of 0.51 nM for VB₁, 0.62 nM for VB₂, 0.37 nM for VB₇, 0.33 nM for VB₉, 0.42 nM for VB₁₂, and 0.57 nM for VD, offering better performance than most existing wearable vitamin sensors (Fig. 2p and Supplementary Table 2)^{20,21,34,43,44}. The sensors demonstrated high selectivity to target vitamins over a variety of potential interferences present at much higher concentrations, as well as high selectivity with minimal cross-reactivity towards other vitamins (Fig. 2q, Supplementary Fig. 19, and Supplementary Note 2). Reproducibility tests revealed that the sensor performance was consistent across multiple batches (Supplementary Fig. 20). Prior literature suggests that wearable electrochemical bioaffinity sensors typically have storage lifetimes on the order of one to two weeks at 4 °C^24,25,27.

Sweat analysis from stimulation to calibrated sensing

The disposable sensor patch was engineered with key features—multiplexed electrodes, on-demand sweat induction, and microfluidic sampling—to autonomously and reliably manage the sweat analysis (Fig. 3a). The sensing process was initiated by collecting iontophoresis-extracted sweat and delivering it to the reagent chamber, where sweat vitamins were bound to the immobilized BREs. This was followed by the sequential introduction of HRP-vitamin and HQ/H₂O₂ solutions, enabling the competitive electrochemical detection of the sensor array (Supplementary Fig. 21). To ensure robust performance during this process, several design elements were incorporated. For instance, a symmetric Tesla valve integrated into the inlet layer prevents backflow towards the skin during solution delivery (Supplementary Fig. 22)⁴⁵, and the flexible PDMS fluidic reservoir layer not only enables pumping but also allows mechanical compression to drive fluid transport, significantly enhancing the system’s adaptability.

Fig. 3. Design and characterization of the fully integrated wearable vitamin sensor for sweat induction, sampling, and analysis.

a Design of the wearable microfluidic vitamin sensor patch. CE, counter electrode; RE, reference electrode; IMP, impedance electrodes. b Potentiometric response of the pH sensor. The inset is the corresponding calibration plot. c Impedimetric response of the ionic strength sensor in PBS. The inset is the corresponding calibration plot. d VB₉ sensor validation against ELISA with calibration (left) and without calibration (right). e Schematic of sweat induction via iontophoretic transdermal carbachol delivery. f Localized sweat rates in three subjects after iontophoresis. S, subject. g and h Dynamic responses of the pH sensor (g) and ionic strength sensor (h) upon solution switching at different flow rates. i Numerical simulation of the distribution of the VB₉ concentration within the microfluidic reservoir 300 s after the inlet fluid changed from 10 to 80 nM VB₉, with different inlet numbers, spacing distances, and inlet/outlet orientations. Flow rate: 1.5 μL min^-1. j On-body evaluation of the optimized flexible microfluidic patch. Sweat was induced via iontophoresis and sampled without exercise. Scale bar, 5 mm.

Although this automated workflow enables in situ analysis, the reliability of trace vitamin quantification in sweat was challenged by the large interpersonal and intrapersonal variations in sweat composition⁴⁶. Sweating is a complex process involving fluctuating pH and ionic strength, with varying patterns and levels in different subjects^31,47,48. Notably, affinity-based sensors are inherently sensitive to reaction conditions like pH/ionic strength²⁴, causing inaccuracies in vitamin quantification. To address this, we integrated pH and ionic strength sensors within a multiplexed electrochemical array. This integration both supported a more comprehensive assessment of physiological status and enabled real-time calibration of the vitamin sensor to enhance its reliability. The pH sensor exhibited step responses in terms of potential across pH 4 to 8 in buffer solutions, with a linear correlation (R² = 0.994) between the measured potential and pH values (Fig. 3b and Supplementary Fig. 23). The accuracy of the sweat pH analysis was validated against that of a commercial pH meter (Supplementary Fig. 24). Moreover, the admittance measured by the ionic strength sensor was linearly related to the electrolyte concentration (Fig. 3c). Real-time calibration of the vitamin sensor was implemented using concurrent pH and ionic strength readings. For example, with pH and ionic strength coassisted calibration, the measured sweat VB₉ level demonstrated a stronger correlation with the ELISA reference (r = 0.989), significantly enhancing the accuracy of in situ VB₉ analysis compared with the uncalibrated protocol (Fig. 3d, Supplementary Figs. 25 and 26 and Supplementary Note 3). Additionally, the integrated electrode exhibited mechanical flexibility and maintained stable electrochemical responses under bending at a 2 cm radius, as further supported by bending tests under various bending times and radii (Supplementary Figs. 27–29).

To enhance the translational potential of this wearable platform—particularly for sedentary populations with low sweat rates—we engineered an iontophoresis module to ensure a sufficient supply of sweat without the need for vigorous exercise. It featured carbon electrodes integrated with a carbachol-loaded hydrogel matrix. When electrical current was applied, transdermal carbachol delivery was enabled to elicit sustained sweat gland activation, thereby facilitating controlled high-yield sweat extraction for continuous monitoring (Fig. 3e, f). The continuous sensing performance was validated by dynamic perfusion of analyte solutions at physiologically relevant sweat rates (1–4 µL min^-1). The sensor achieved rapid signal stabilization to the new fluid when the perfusate conditions changed, demonstrating the rapid temporal resolution of the system under simulated physiological perturbations (Fig. 3g, h). To maximize sweat sampling efficiency, we developed compact flexible microfluidic modules with varied geometric parameters. Numerical simulations were performed to optimize the microfluidic design, including the inlet number, spacing distance, orientation, and location of the outlet. These simulations generated surface plots visualizing the dynamic flow of the fluid to show fluid transport along the designed flow path through inlets and microchannels into the reservoir layer (Fig. 3i and Supplementary Figs. 30 and 31). For the optimized system simulating at a sweat rate of 1.5 µL min^-1, the refreshing time required to reach 95% of the new solute concentration was approximately 4.5 min. This optimized architecture enables efficient, localized sweat induction and collection. In on-body validation studies, we affixed the patch to the skin of the subject to autonomously induce sweat through iontophoresis at rest, and high-temporal-resolution sampling was achieved. This was evidenced by complete displacement of preloaded blue dye by fresh sweat (Fig. 3j).

On-body validation of the wearable vitamin-sensing platform

VB₉ was chosen for on-body evaluation of the wearable technology because of its critical physiological role in individuals with varying physiological states and its broad clinical relevance in areas spanning from nutritional deficiency disorders to chronic disease management. Furthermore, as an essential vitamin that must be obtained through diet or supplementation, it serves as a representative example for validating real-time nutrient monitoring. On-body evaluation was conducted under real-life scenarios with participants wearing wearable sensing patches to track dynamic sweat changes before and up to 3 h after taking 5 mg of the VB₉ supplement or consuming a VB₉-rich diet⁴⁹. The system wirelessly transmitted data to a smartphone application, where dedicated algorithms converted raw signals into real-time vitamin concentrations for visualization within a custom mobile app. The integrated wearable systems could be comfortably worn on different body parts, enabling dynamic tracking of responses from VB₉, pH, and ionic strength sensors at rest (Fig. 4a, b and Supplementary Fig. 32).

Fig. 4. On-body evaluation of the fully integrated wearable sensor.

a Photograph of the wearable sensor worn on the forearm with BLE signal transmission to a mobile phone. b Photographs of a healthy subject wearing the sensor patch on different parts of the body. c Schematic and corresponding data showing on-body multiplexed sweat sensing before and 3 h after VB₉ supplementation, with data from Subject 1 and Subject 2. DPV measurements were recorded from −0.3 to 0.2 V vs. Ag/AgCl. d Schematic and corresponding data showing on-body multiplexed sweat sensing before and 3 h after consumption of a VB₉-rich meal, with data from Subject 1 and Subject 2. DPV measurements were recorded from −0.3 to 0.2 V vs. Ag/AgCl.

Real-time monitoring revealed consistently low baseline VB₉ concentrations of less than 10 nM in the two subjects. Following intake of 5 mg of VB₉, the VB₉ concentration in the participant’s sweat increased by over 3.6-fold relative to the baseline, demonstrating the system’s ability to detect fluctuations in the concentration of VB₉ (Fig. 4c). Moreover, following the consumption of the VB₉-rich diet, subjects exhibited a greater than 1.6-fold increase in their VB₉ levels 3 h after intake, which corresponded to lower sensor currents (Fig. 4d). Control experiments confirmed both biological relevance and sensor selectivity in real sweat samples (Supplementary Fig. 33). Concurrently, the sweat pH and ionic strength values fluctuated within established physiological ranges for healthy human subjects. These results demonstrated the ability of sweat to reflect VB₉ consumption and suggested a fully integrated wearable sensor as a promising alternative to invasive and time-consuming blood analysis for remote, at-home vitamin tracking.

Wireless biosensor for personalized VB₉ monitoring

Globally, both VB₉ deficiency and excess supplementation remain widespread public health issues, highlighting the need for regular monitoring to guide dietary intervention and supplementation strategies. VB₉ has the potential to diffuse freely in aqueous physiological environments and be excreted into sweat, making it a promising candidate for noninvasive nutritional assessment (Fig. 5a). However, the temporal profiles of VB₉ in sweat following supplementation and its correlation with blood concentrations remain unclear. Given that the VB₉ status is dynamically modulated by behavioral factors, including nutritional intake and lifestyle (e.g., smoking)^50–52, we evaluated the dynamic response of VB₉ in sweat to oral supplementation, a well-established method for elevating VB₉ levels in serum. Continuous 2-day monitoring revealed baseline VB₉ concentrations in sweat of below 10 nM. Following oral intake, concentrations increased more than 2.0-fold relative to baseline within 3 h, closely paralleling the VB₉ pharmacokinetics in serum (Fig. 5b–d)⁵². VB₉ levels remained significantly elevated above baseline even after 8 h. Both subjects exhibited similar VB₉ temporal profiles on Day 2 to those on Day 1. The ELISA results were in strong agreement with the sensor measurements, confirming that the wearable sensors reliably tracked these dynamic changes in sweat VB₉ concentrations. We also conducted a 3-day monitoring study, which revealed persistently low VB₉ levels in sweat during an initial period with no supplementation (Days 1–2) and a sharp increase only after taking 5 mg of VB₉ on Day 3 (Fig. 5e). This response strongly indicates that the sensor’s signal is primarily specific to VB₉ supplementation, with minimal contributions from other sweat constituents. An extended cohort study (n = 8) further confirmed that baseline VB₉ concentrations in sweat ranged from 1–9 nM and increased to 12–32 nM after the ingestion of 5 mg of VB₉ pills (Fig. 5f and Supplementary Fig. 34). Furthermore, analysis of sweat from distinct body sites revealed negligible variations; all measurements remained within statistically comparable ranges (Supplementary Fig. 35). To further explore the relationship between oral dose and VB₉ concentration in sweat, we investigated VB₉ levels in sweat 3 h after ingestion of 0, 5, 10, and 20 mg of VB₉. These trials rigorously incorporated adequate washout periods between doses. The results indicated that VB₉ levels in sweat increased with increasing supplemental dose, with no evidence of saturation across the 0–10 mg dose range. However, a smaller increase was observed after the 20 mg VB₉ dose, indicating that the absorption limits were approaching (Fig. 5g). Moreover, comparative analysis of concurrently collected sweat and serum samples demonstrated that VB₉ concentrations in sweat increased concomitantly with increasing VB₉ levels in serum, demonstrating a strong correlation between sweat and serum VB₉ levels (r = 0.849; Fig. 5h). The Bland-Altman analysis reveals a negative mean bias and proportional bias between sweat and serum VB₉ measurements, although most points fall within the 95% limits of agreement (Supplementary Fig. 36). These deviations limit direct clinical equivalence; however, sweat measurements remain suitable for non-invasive trend monitoring. Future calibration models, correction algorithms, or individualized mapping could help mitigate these biases by adjusting for mean bias, correcting proportional bias, and accounting for individual variability, thereby further improving quantitative alignment between sweat and serum VB₉ measurements. These findings substantiate the potential use of VB₉ in sweat as a noninvasive biomarker for assessing VB₉ status.

Fig. 5. Noninvasive VB₉ monitoring using a sweat sensor.

a Metabolic pathway of VB₉ and its fluctuations in serum and sweat following supplementation. b Timeline of VB₉ supplementation and sample collection over 48 h. c and d Temporal changes in VB₉ levels in the sweat of Subject 1 (c) and Subject 2 (d). e VB₉ levels in the sweat of a single subject over three days. f Comparison of VB₉ levels in the sweat of eight subjects before and after supplementation with 5 mg of VB₉ (n = 8 independent subjects). The bottom and top whiskers indicate the minimum and maximum values, the box boundaries represent the interquartile range (25th–75th percentiles), and the square in the box indicates the mean. g Relationship between sweat VB₉ levels and VB₉ supplementation dose. h Correlations between the VB₉ levels in sweat and serum (r = 0.849). i Mechanism of the smoking-induced reduction in VB₉ levels. MS methionine synthase, CAs catecholamines, ROS reactive oxygen species. j Box-and-whisker plot of measured VB₉ levels in sweat and serum from two groups of participants: current smokers (group I, n = 8 independent subjects) and never smokers (group II, n = 7 independent subjects). The bottom and top whiskers indicate the minimum and maximum values, the box boundaries represent the interquartile range (25th–75th percentiles), and the square in the box indicates the mean.

Building on the observed response of VB₉ levels in sweat to supplementation, we next investigated its response to a factor known to deplete VB₉: tobacco use. Tobacco use has been consistently linked to reduced VB₉ levels through multiple mechanisms^50,51. Specifically, components of cigarette smoke inhibit the activation of methionine synthase, thereby preventing the remethylation pathway and trapping VB₉ in the form of 5-methyltetrahydrofolate⁵⁰. Furthermore, tobacco smoke increases catecholamine levels and oxidative stress, which alters the basal metabolic rate and leads to the inactivation of VB₉, resulting in lower VB₉ levels (Fig. 5i)⁵¹. Consistent with these mechanisms, smokers had lower VB₉ levels in both serum and sweat than those who had never smoked (Fig. 5j and Supplementary Fig. 37). This agreement between the confirmed physiological effect and our experimental data validates the physiological relevance and accuracy of our sweat-based measurements. These findings collectively support the potential use of our biosensor for noninvasive assessment of nutritional status across diverse populations.

Discussion

Vitamin deficiencies, emblematic of the pervasive global challenge of hidden hunger, highlight the urgent need for dynamic nutritional monitoring. Translating complex laboratory vitamin analysis into real-time, wearable sweat-sensing technology offers a promising path toward personalized nutrition. However, this vision has been constrained by major challenges: (1) low vitamin concentrations in sweat challenge detection sensitivity; (2) complex sweat matrix effects impede accurate in situ quantification; and (3) a paucity of robust correlation data linking vitamin levels in sweat to the gold standard or established health outcomes.

Our integrated wearable platform addresses these limitations by leveraging an electrochemical sensing mechanism that incorporates nanomaterial-mediated signal amplification and BREs for enhanced sensitivity and selectivity. The platform combines three key advances: (1) an affinity-based biosensor featuring AuNF/SNC for signal amplification, which enables highly sensitive electrochemical detection of vitamins at nanomolar levels; (2) modular BREs that permit the quantification of six essential vitamins beyond single-analyte constraints; and (3) autonomous microfluidic operation with on-demand sweat induction and real-time detection with calibration for reliable on-body sweat monitoring. Critically, we demonstrated that vitamin levels in sweat were clinically relevant biomarkers, with VB₉ serving as a key example. Longitudinal analysis revealed a strong sweat-serum VB₉ correlation (r = 0.849), supporting the utility of sweat for noninvasive nutritional assessment. Furthermore, dynamic VB₉ profiling revealed individual absorption kinetics following supplementation, highlighting the potential for real-time monitoring to achieve personalized nutritional intervention. Notably, the marked difference in VB₉ levels between smokers and nonsmokers underscores the ability of this technology to be used across different human populations for nutritional management. This platform significantly reduces analytical turnaround time, enabling rapid, data-driven adjustments to diet or supplementation. We envision that this biosensing platform could be a promising tool for mitigating hidden hunger and optimizing personalized nutritional interventions. Beyond the detection of vitamins, the platform’s modular design and high sensitivity can lead to scalable precision health tools that can potentially extend to diverse low-abundance biomarkers for holistic physiological profiling in the emerging era of AI-driven digital health.

Methods

Materials and reagents

Sodium tetrakis[3,5-bis(trifluoromethyl)phenyl]borate (Na-TFPB), hydrogen ionophore I, and bis(2-ethylhexyl) sebacate (DOS) were purchased from Sigma‒Aldrich. FBP, HRP-conjugated VB₉ (HRP-VB₉), HRP-conjugated vitamin B₁₂ (HRP-VB₁₂), IF, and monoclonal antibodies to vitamin D (VD) were obtained from Baiming Biotechnology (catalog no. BM841001). HRP-conjugated vitamin B₁ (HRP-VB₁), HRP-conjugated vitamin B₂ (HRP-VB₂), monoclonal antibodies to VB₁ (catalog no. VB1Ab-KJ01R), and monoclonal antibodies to VB₂ (catalog no.VB2Ab-KJ01R) were obtained from Shenzhen Kejie Industrial Development Co., Ltd. Streptavidin, chitosan, EDC, NHS, and BSA were sourced from Shanghai Yuanye Bio-Technology Co., Ltd. Vitamin B₁ (VB₁) and 2-(N-morpholino)ethanesulfonic acid monohydrate (MES) were purchased from Shanghai Aladdin Biochemical Technology Co., Ltd. Vitamin B₂ (VB₂), vitamin B₇ (VB₇), vitamin B₁₂ (VB₁₂), tetrahydrofuran (THF), polyvinylpyrrolidone (PVP), (3-triethoxysilyl)propyl succinic anhydride (TESPSA), (3-aminopropyl)triethoxysilane (APTES), thiourea, vitamin B₉ (VB₉), uric acid, D-(+)-glucose, HQ, K₄[Fe(CN)₆], dopamine (DA), and 25-hydroxyvitamin D (VD) were obtained from Shanghai Macklin Biochemical Co., Ltd. Phosphate-buffered saline (PBS) was purchased from Beijing Solarbio Science and Technology Co., Ltd. HAuCl₄·3H₂O and K₃[Fe(CN)₆] were obtained from Heowns. The TC-205 carbon ink (with a phenolic resin binder) and the AG-7 silver/silver chloride ink were obtained from Poten Technology Co., Ltd. Poly(vinyl chloride) (PVC) was purchased from Adamas. HRP-conjugated vitamin B₇ (HRP-VB₇), carbachol, and agarose were obtained from MedChemExpress. The PDMS base and crosslinking agents were purchased from Dow Corning Co., Ltd. Additionally, 3 M 468MP adhesive tape was purchased from 3 M Company. ARflow 93049 adhesive tape was purchased from Adhesives Research. A human VB₉ ELISA kit was obtained from Jonln Biology.

Fabrication of the multiplexed electrode array

An electrode array was fabricated by screen printing, which included carbon ink working electrodes (WEs) for vitamin, pH, and ionic strength detection; a carbon counter electrode (CE); and an Ag/AgCl reference electrode (RE). In addition, carbon ink was used to print the iontophoresis electrodes for sweat induction. During fabrication, the reference electrode was printed using silver/silver chloride ink and cured at 60 °C for 30 min, and the carbon electrodes were printed and cured at 60 °C for 30 min. To activate the screen-printed electrode (SPE) array, remove surface oxides, and enhance electrochemical performance, 200 µL of 0.1 M H₂SO₄ was deposited onto the surface, followed by CV scanning for 10 cycles (potential range: -0.2 V to 1.2 V vs. Ag/AgCl; scan rate: 100 mV s^-1). The activated array was then rinsed thoroughly with ultrapure water and dried. Next, 200 µL of 0.6 M thiourea solution was applied to the SPE surface, and electrochemical codoping of sulfur and nitrogen was achieved in situ through CV for 30 cycles (potential range: -1.2 V to 0.2 V vs. Ag/AgCl; scan rate: 60 mV s^-1) to obtain SNC. Following doping, the electrode was rinsed with ultrapure water and dried. Subsequently, 200 µL of 2.5 mM HAuCl₄ (prepared in 1× PBS) was dispensed onto the SNC surface, and the AuNFs were electrodeposited in situ by CV for 10 cycles (potential range: -0.5 V to 0.4 V vs. Ag/AgCl; scan rate: 50 mV s^-1). Finally, the electrode surface was rinsed with ultrapure water and dried under a stream of nitrogen gas.

Functionalization of the electrode array

The vitamin-sensing electrode fabrication protocol began with the preparation of a fresh 0.1 M MES buffer, which was used to prepare a solution containing 0.4 M EDC and 0.1 M NHS. A 20 µL aliquot of this EDC/NHS mixture was dispensed onto the electrode surface and incubated in the dark at 25 °C for 30 min to activate the carboxyl groups. Following activation, the electrode was thoroughly rinsed three times with MES buffer (prechilled to 4 °C) and dried under nitrogen. For the VB₉ biosensor, 10 µL of FBP solution (75 µg mL^-1 in 1× PBS, pH 7.4) was applied to the activated surface, and covalent immobilization through amide bond formation was achieved by incubation at 37 °C for 40 min. The electrode was then treated with 1% BSA at 37 °C for 30 min to passivate nonspecific binding sites. Sensors for VB₁, VB₂, VB₇, VB₁₂, and VD were fabricated using this protocol, differing only in immobilized BREs.

The H⁺-selective electrode was fabricated by drop-casting a membrane cocktail onto the pH electrode. This cocktail was formulated by dissolving 2 mg of hydrogen ionophore I, 2 mg of Na-TFPB, 100 mg of PVC, and 200 mg of DOS in 2 mL of THF, followed by overnight solvent evaporation. To ensure long-term continuous monitoring capability, the H⁺-selective sensor was conditioned overnight in 0.1 M HCl. For ionic strength detection, a AuNF/SNC nanocomposite electrode was employed without additional functionalization.

The material morphology was characterized using scanning electron microscopy (SEM; ZEISS Gemini SEM 300, Germany), transmission electron microscopy (TEM; JEOL JEM-F200, Japan), and atomic force microscopy (AFM; Bruker Dimension Icon, Germany). Compositional and valence state analyses were performed by X-ray diffraction (XRD; Rigaku Smartlab 3 kW, Japan) and X-ray photoelectron spectroscopy (XPS; Thermo Scientific K-Alpha, USA). Raman spectra of the electrode materials were acquired using a Thermo Scientific DXR 3Xi Raman microscope (USA). Surface wettability was assessed via contact angle measurements (JY-82C, Chengde Dingsheng, China).

Electrochemical characterization of sensors

All the electrochemical measurements—including CV, OCP-EIS, and DPV—were performed at ambient temperature using a Multi Autolab (based on Autolab PGSTAT204, Metrohm AG, Switzerland).

To characterize the electrochemical performance of different materials and validate the success of each surface modification step, CV and OCP-EIS were performed following every preparation or functionalization procedure. Specifically, CV measurements were conducted in 1× PBS containing 0.5 mM K₃[Fe(CN)₆] (potential range: −0.4 to 0.6 V vs. Ag/AgCl; scan rate: 50 mV s^-1), whereas OCP-EIS characterization employed 1× PBS with 2.0 mM equimolar K₄[Fe(CN)₆]/K₃[Fe(CN)₆] (frequency range: 10^-2–10⁶Hz). Vitamin detection was carried out via DPV in PBS containing 2.0 mM HQ and 1 mM H₂O₂ (potential range: −0.3 to 0.2 V vs. Ag/AgCl; step: 4 mV; modulation amplitude: 50 mV; modulation time: 0.05 s; interval time: 0.5 s). pH sensors were tested using PBS with varying pH to characterize sensor performance.

Fabrication of the microfluidic system

The microfluidic patch featured a multilaminate architecture with four functional layers for sweat collection and a PDMS reagent chamber. Each layer was precision-patterned using a CO₂ laser cutter (DF-60D, Xinglihua, Beijing), with sequential assembly from the skin interface upwards as follows: (1) an accumulation layer fabricated from high-viscosity double-sided adhesive tape (3 M 468MP; laser parameters: 60% power, 80% speed, 1000 PPI), (2) an inlet layer formed from 0.025 mm PET film (laser parameters: 20% power, 100% speed, 1000 PPI), (3) a channel layer fabricated from AR hydrophilic single-sided adhesive (AR 93049; laser parameters: 40% power, 90% speed, 1000 PPI), and (4) a reservoir layer constructed with 3 M 468MP adhesive (laser parameters: 60% power, 80% speed, 1000 PPI). All layers were patterned with iontophoretic gel contours, ensuring that an electrical current from the top electrode was applied to induce on-body iontophoretic sweat. The microfluidic sensing system was tested using a syringe pump (PHD ULTRA, Harvard Apparatus) at different flow rates. To test the sweat-collection capability and refresh performance of the microfluidic patch, preassembled patches containing blue dye preloaded in the reservoir layer were affixed to participants’ skin. Sweating was then induced through iontophoresis.

Fabrication of the PDMS reservoir layer commenced with spray-coating an acrylic mould with a mould-release agent (ER-200), followed by 30 min of ambient drying. Liquid PDMS mixed at a 10:1 (w/w) base-to-curing-agent ratio was then cast into the mould and thermally cured at 80 °C for 4 h to form the PDMS sheets. To achieve bonding between the PDMS and the PET substrate, both surfaces were rinsed with methanol and deionized water and then treated with oxygen plasma for 60 s. Subsequently, the PDMS and PET contact interfaces were treated with aqueous solutions of TESPSA (1 wt%) and APTES (1 wt%), respectively, for 30 min at room temperature⁵³. Following surface modification, the substrates were rinsed with water and isopropyl alcohol, air-dried, and brought into conformal contact at room temperature. Ports beneath the PDMS layer enabled efficient fluidic filling.

Formation of hydrogels for iontophoresis

Anodal and cathodal iontophoresis hydrogels were fabricated using 3 wt% agarose. Agarose was dissolved in ultrapure water by heating to 250 °C, then cooled to approximately 160 °C. At this stage, 1 wt% carbachol was incorporated into the anodal gel formulation, while 1 wt% KCl was added to the cathodal gel formulation. The mixtures were then cast into iontophoresis electrode-shaped moulds and solidified at room temperature. All the fabricated hydrogels were stored at 4 °C prior to use.

Numerical simulation of the microfluidic module

Numerical simulations were performed using COMSOL to analyse the times for different designs of the microfluidic sweat collection module to refresh. Three-dimensional models with dimensions identical to those of the actual devices were created in SOLIDWORKS for distinct microfluidic architectures. These models were subsequently imported into COMSOL Multiphysics. The fluid flow in the microfluidic module was laminar and incompressible, and its behavior was described by the Navier-Stokes equation.

$$\rho \left(\frac{\partial v}{\partial t}+\left(v\mathrm{\cdot }\nabla \right)v\right)=-\nabla p+\mu {\nabla }^{2}v$$ 1

$$\nabla \mathrm{\cdot }v=0$$ 2

Here, $\rho$, $v$, $t$, $p$, and $\mu$ denote the liquid density, flow rate, time, pressure, and viscosity, respectively. The total flow rate of the sweat was 1.5 µL min^-1.

The process of refreshing the channel with solutions of different concentrations involves mass transfer. This process used a dilute species transport model with the transfer mechanism set to convection, and the solute diffusion process followed Fick’s first law. Convection-diffusion was described by the following equation:

$$\frac{\partial C}{\partial t}+v\cdot \nabla C=D{\nabla }^{2}C$$ 3

where $v$ denotes the velocity vector of the fluid, $\nabla C$ denotes the concentration gradient, and $D$ denotes the diffusion coefficient.

The mass transport process was simulated by numerically solving the Navier-Stokes equation and convection‒diffusion equations for incompressible flow. The total inlet flow rate was 1.5 µL min^-1, with no-slip solid phase wall boundary conditions. The concentration of VB₉ ranged from 10 nM to 80 nM. The times required for the average volume concentration to reach 90% and 95% of the new solute concentration are shown in Supplementary Table 3. The refreshing time was defined as the time to achieve the new target concentration.

Electronic system design and integration

The wearable sensing platform featured an FPCB with a rectangular footprint (40 mm × 20 mm) fabricated on a polyimide substrate. The system-level block diagram and the detailed circuit schematic are presented in Fig. 1e and Supplementary Fig. 6, respectively. Electronic components were integrated onto the top layer via surface-mounted technology (SMT), including a microcontroller unit (MCU; STM32L4, STMicroelectronics) for system control, a low-energy Bluetooth module (BLE; CH9140, WCH) for wireless communication, a digital-to-analog converter (DAC; MCP4728, Microchip Technology), an analog-to-digital converter (ADC; ADS1115, Texas Instruments), and a custom electrochemical analog front-end (AFE) based on the precision operational amplifier AD8606 (Analog Devices).

A custom-developed Android application, which was created using the App Inventor platform, established a bidirectional BLE link with the device. User-defined operational parameters and commands from the application were received and parsed by the MCU via universal asynchronous receiver‒transmitter (UART) communication, enabling the dynamic execution of diverse electrochemical detection protocols. The acquired sensor data were subsequently streamed wirelessly via BLE back to the mobile application for real-time visualization and persistent data storage.

Recruitment of participants

Human subject evaluation of the wearable biosensing system for noninvasive sweat analysis was conducted under a protocol (ID 020250306) approved by the China Agricultural University Human Research Ethics Committee. All participants provided written informed consent prior to enrollment. Two cohorts were recruited through advertisements: (1) a healthy cohort (n = 9) comprising individuals aged 18–35 years with a BMI of 18.5–24.9 kg m^-2, fasting blood glucose concentration <100 mg dL^-1, and no smoking history to demonstrate sensor performance and investigate sweat-serum VB₉ correlations; and (2) a smoker cohort (n = 8) meeting identical criteria but with a smoking history ( ≥ 1 year, ≥ 10 cigarettes/day), specifically recruited to examine the effects of tobacco use on VB₉ levels.

Wearable sensor validation in human subjects

After an overnight fast, the subjects arrived at the laboratory for on-body validation. Participants were administered VB₉ tablets. Paired sweat and venous blood samples were then collected at designated timepoints. Prior to attaching the wearable sensor patch, the skin was cleaned with alcohol wipes followed by lint-free wipes. Sweat induction was performed via iontophoresis, and paired venous blood samples were collected during the same period. Serum was isolated from clotted blood through centrifugation (980 × g, 10 min, 4 °C) and immediately cryopreserved at -80 °C.

As exemplified in Fig. 5b, the first batch of sweat and blood samples was collected from subjects at 8:30 AM. The subjects then took 5 mg VB₉ tablets at 9:00 AM, followed by a second sample collection 3 h later (12:00 PM). The final batch was collected at 17:30. The same procedure was repeated the next day. For on-body evaluation of the sensor, sweat analyte dynamics were monitored before and 3 h after the subjects ingested 5 mg VB₉ tablets. Additionally, sweat composition was tracked before and approximately 3 h after consumption of a VB₉-rich diet (∼1000 kcal, ∼3 mg of VB₉). All sweat samples were collected via iontophoresis. To minimize confounding variables, the intake of VB₉-rich foods was restricted throughout the study.

During on-body sensing, the multiplexed sensor wirelessly transmitted impedance and open-circuit potential (OCP) data from integrated ionic strength and pH sensors to a mobile phone interface via Bluetooth. Upon signal stabilization, the system initiated competitive incubation followed by DPV analysis within the prestored HQ/H₂O₂. The VB₉ concentration was calculated using correction algorithms that incorporated concurrently measured ionic strength and pH data to correct for interference effects. Each sensor functioned as a single-use device to ensure measurement fidelity.

ELISA analysis of sweat samples for sensor validation

We quantified the VB₉ concentration in sweat using a competitive ELISA for biosensor validation. The analytical protocol entailed sequential addition of sweat samples and standards to microtiter wells precoated with the VB₉ antigen, followed by biotinylated antibodies and HRP-VB₉. After incubation and washing, the enzymatic activity was quantified through the use of a 3,3’,5,5’-tetramethylbenzidine (TMB) substrate. TMB was oxidized by HRP to produce a blue product, which was then converted to a yellow end product upon acidification. The absorbance was measured at 450 nm, and concentrations were calculated from the standard curve.

Analysis of serum samples

Serum VB₉ quantification was performed via a chemiluminescence competitive assay (Baiming Biotechnology). Samples were treated with extraction buffer to liberate protein-bound VB₉, followed by incubation in streptavidin-coated microplates with sequential additions of HRP-VB₉ and biotinylated FBP to form immobilized complexes via biotin‒streptavidin conjugation. During incubation, endogenous VB₉ and HRP-VB₉ competed for FBP binding sites. After being washed to remove unbound constituents, the addition of chemiluminescent substrate generated HRP-catalyzed signals quantified as relative luminescence units. VB₉ concentrations were determined against the calibration curve.

Ethics

Every experiment involving animals, human participants, or clinical samples has been carried out in accordance with a protocol approved by an ethics committee. Each participant gave informed written consent. No identifiable personal information is included in this study. The study included 9 healthy adults (3 females and 6 males; aged 18–35 years) and 8 smokers (all males; aged 18–35 years). Participants received monetary compensation for their participation.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

H.Z. and X.W. conceived and designed the study. H.Z. and F.R. supervised the project. X.W., Y.W., Y.L., and Y.S. performed sensor characterization, validation, and sample analysis. X.W., Y.W., XM.W., and Y.S. analysed the data. X.W., S.L., and J.O. designed the human clinical studies. P.M. and X.W. contributed to the numerical simulation. The manuscript was co-written by X.W. and H.Z. All authors discussed the results and contributed to the final manuscript.

Peer review

Peer review information

Nature Communications thanks Ulkuhan Guler, Ahmed Khorshed, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data supporting the findings of this study are available within the article and its supplementary files. Any additional requests for information can be directed to and will be fulfilled by the corresponding authors. Source data are provided with this paper.

Code availability

The codes used for collecting data are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72356-1.