Carbon Nanotube-Induced Enzyme Polymerization for Wearable Multifunctional Biosensors in Sweat Analysis

Abstract

Wearable sweat sensors have emerged as promising tools for noninvasive health monitoring, yet the low analyte concentrations in sweat compared to blood pose significant challenges for the limit of detection. In this study, we developed a high-sensitivity electrochemical biosensor using carbon nanotube (CNT)-induced enzyme polymerization to detect uric acid and glucose with ultralow detection limits. The CNTs were functionalized via EDC/NHS to achieve covalent enzyme immobilization, enhancing catalytic efficiency, electron transfer, and sensor stability. To enable multifunctional sensing, we integrated glucose and uric acid detection with a pH sensor into a single wearable platform. A radially symmetric microfluidic module was designed through finite element analysis to optimize sweat flow and minimize refresh time, ensuring real-time biomarker tracking. The system also incorporated pH-based signal correction to improve detection accuracy in complex sweat environments. Finally, the sensing performance was validated through on-body sweat collection and analysis from six human volunteers, demonstrating its robustness, reliability, and potential for advancing next-generation personalized healthcare applications. This work provides a framework for designing multifunctional wearable sweat sensors and highlights the role of material and device innovations in overcoming key challenges in this field.

1. Introduction

Sweat-based electrochemical sensing has emerged as a promising noninvasive approach for monitoring various biomarkers, offering significant potential for real-time health monitoring and personalized medicine. Among sweat biomarkers, uric acid, glucose, and pH hold particular importance due to their strong correlations with metabolic, physiological, and pathological conditions. Uric acid is a key indicator of purine metabolism, and its abnormal levels are associated with conditions such as gout and cardiovascular diseases. Similarly, glucose monitoring is critical for assessing energy metabolism and managing diabetes, while pH levels provide insights into electrolyte balance and metabolic acidosis. However, the accurate and selective detection of these biomarkers in sweat poses substantial challenges. The trace concentrations of analytes, the complex composition of sweat, and the dynamic nature of sweat secretion necessitate the development of highly sensitive, selective, and stable sensing platforms. Carbon-based materials, particularly carbon nanotubes (CNTs), have drawn considerable attention in this context due to their exceptional electrical conductivity, high surface area, chemical stability, and excellent electrocatalytic properties. , These unique characteristics make CNTs an ideal platform for enhancing the sensitivity and selectivity of electrochemical sensors targeting uric acid, glucose and pH.

The growing demand for real-time, noninvasive health monitoring has accelerated research into wearable sweat sensors capable of continuous detection of these biomarkers. , In particular, real-time monitoring of blood glucose and uric acid levels in cancer patients is of great importance, as it can provide clinicians with timely and accurate information for disease management, enable early detection of metabolic complications, and facilitate personalized treatment adjustments. Electrochemical wearable sweat sensors could offer an attractive solution by integrating advanced materials, microfabrication techniques, and signal processing systems into compact, user-friendly devices. In particular, recent fabric- and textile-based electrochemical biosensors and health-managing textile systems have demonstrated the great potential of integrating sensing, long-term stability and wearability into a single platform. , However, translating the laboratory performance of these sensors into practical wearable applications involves significant challenges, especially in the design and integration of functional modules. Key components, such as the sweat collection units, sensing interface, signal transduction elements, and data transmission units, must be seamlessly incorporated into a flexible and biocompatible platform while maintaining high sensitivity, reliability, and mechanical resilience. Furthermore, issues such as signal interference, sensor fouling, and mechanical deformation during wear need to be addressed to ensure device stability and usability. Wearable sweat sensors have emerged as promising tools for real-time biomarker monitoring; however, their sensitivity presents a notable limitation, as evidenced by recent research. Current sensor technologies struggle to detect low-concentration biomarkers in sweat, emphasizing the necessity for improved sensitivity to achieve accurate monitoring. Material innovations have been investigated to enhance signal amplification, yet these efforts fall short of fully addressing sensitivity constraints. A miniaturized OECT system has also been developed, which offered high-resolution detection. However, it faces sensitivity challenges caused by background noise and interference, particularly in complex biofluids such as sweat. This highlights the need for advanced materials and optimized designs to improve sensor performance.

This study builds on the demand for continuous sweat analysis by leveraging the exceptional properties of carbon nanotubes to induce polymerase activity and enhance the performance of wearable electrochemical sensors. Specifically, we developed a multifunctional CNT-based sensing platform by covalently functionalizing EDC/NHS-activated CNTs with uricase and glucose oxidase enzymes, coupled with a high-sensitivity pH sensor. The high surface area of CNTs enables dense enzyme loading, while their conductive network facilitates rapid electron transfer, resulting in an 83% increase in sensitivity compared to Nafion-enzyme sensor. Additionally, the integration of a radially symmetric microfluidic channel (180° span) minimized sweat refresh time to 163 s, ensuring real-time biomarker tracking. To enhance wearability, the sensor was fabricated on a flexible substrate with a total thickness of 600 μm, enabling conformal adhesion to skin during dynamic movements. By integrating CNT-based sensing modules, this research aims to advance the next generation of wearable devices capable of accurate, real-time monitoring of uric acid, glucose, and pH in sweat. The findings from this work hold significant potential for advancing personalized healthcare and establishing wearable diagnostics as a cornerstone of modern medicine.

2. Materials and Methods

2.1. Materials

Glucose oxidase from Aspergillus niger, uricase from Candida sp, bis­(2-ethylhexyl)­sebacate, chitosan, sodium tetrakis­[3,5-bis­(trifluoromethyl)­phenyl]­borate, hydrogen ionophore I (tridodecylamine), tetrahydrofuran and polyurethane were purchased also from Sigma-Aldrich. 1-methyl-3-octylimidazolium bis­(trifluoromethylsulfonyl)­imide was purchased from Aladdin. N-(3-(Dimethylamino)­propyl)-N′-ethylcarbodiimide, N-hydroxysuccinimide sodium were purchased from Macklin. Multiwalled carbon nanotube was purchased from XFNANO.

2.2. Fabrication of the Sensor Array

Figure A illustrates the glucose, uric acid, and pH sensors fabricated on a polyamide substrate. The glucose sensor shares a reference electrode with the uric acid sensor, while the pH sensor has its own reference electrode. All sensor arrays share a counter electrode, reducing the overall size of the sensor and improving wearer comfort. The working (WE) and counter (CE) electrodes are fabricated from carbon ink, while the reference electrodes (RE) are composed of Ag/AgCl. The glucose/uric acid working electrode is modified with three layers: The first layer is Prussian blue, which is one of the most commonly used electrochemical media in analytical applications, followed by a layer of EDC/NHS activated carbon nanotubes is then deposited for more efficient coupling enzymes, and finally, an enzyme layer consisting of chitosan and glucose/urate oxidase. A pH sensor consists of a working electrode and a reference electrode for potential readout. Apply pH selective membrane (pHSM) and reference membrane (RM) respectively.

4.

Design and optimization of our wearable sensor. (A) Schematic representation of the wearable sensor device, illustrating its layered structure. From top to bottom: a polyimide layer with an outlet, a sensor layer housing three working electrodes (WE1: glucose sensor, WE2: uric acid sensor, WE3: pH sensor), a reference electrode (RE), and a counter electrode (CE), a microfluidic channel layer designed to transport fluid from inlets to the sensing area, and an adhesive layer at the bottom for secure attachment. (B) Numerical simulation showing the effect of inlet number and geometric span on refresh time. (C) Solute concentration distribution over time for different inlet configurations.

2.2.1. Preparation of the Activated Carbon Nanotube Layer

In this study, we prepared an activated carbon nanotube (CNT) layer to enhance enzyme immobilization and improve sensor performance. The preparation involved dispersing 2 mg/mL CNTs in 20 mL deionized water and sonicating for 30 min to achieve uniform dispersion. Next, 120 mg of EDC was added to the CNT dispersion, and the solution was stirred magnetically for 1 h to activate the CNT surfaces. The activated CNT solution was then centrifuged at 10,000 rpm for 10 min, and the supernatant was discarded. The CNTs were resuspended in deionized water, and this washing step was repeated twice to remove residual EDC. Subsequently, 60 mg of NHS was added to the washed CNTs in 20 mL deionized water, and the mixture was stirred magnetically for 1 h to introduce NHS groups onto the CNT surfaces. The NHS-activated CNT solution was then centrifuged at 10,000 rpm for 10 min, and the supernatant was discarded. The CNTs were resuspended in deionized water, and this washing step was repeated twice to eliminate unreacted NHS. This activated CNT layer serves as a robust platform for enzyme immobilization, facilitating enhanced sensor functionality.

2.2.2. Preparation of Glucose/Uric Acid Biosensors

Glucose/uric acid biosensor was developed by preparing a chitosan solution, wherein chitosan was dissolved in a 2% acetic acid solution and stirred magnetically for 1 h to obtain a 1% solution. Multiwalled carbon nanotubes (MWCNTs) were then added at a concentration of 2 mg/mL, and the mixture was sonicated for 30 min to create a homogeneous chitosan-MWCNTs viscous solution. Glucose/uric acid oxidase (Gox/Uox) was incorporated into the chitosan-MWCNTs solution at a concentration of 10 mg/mL in phosphate-buffered saline (PBS) at pH 7.2, using a 2:1 (v/v) ratio of chitosan-MWCNTs solution to Gox/Uox solution, and thoroughly stirred for uniform distribution. A Prussian blue (PB) mediator layer was deposited onto a carbon electrode using cyclic voltammetry, where the electrode was immersed in a solution containing 2.5 mM FeCl₃, 100 mM KCl, 2.5 mM K₃Fe­(CN)₆, and 100 mM HCl, with cyclic voltammetry performed within a potential range of 0–0.5 V (vs Ag/AgCl) at a scan rate of 20 mV/s. The resulting PB film was used to drop-cast 3 μL of the chitosan-MWCNTs-Gox/Uox mixture onto the PB-modified carbon electrode, which was then dried overnight at 4 °C under dark conditions to ensure proper immobilization. The sensor was stored at 4 °C when not in use. The sensor structure was shown in Figure S1, and the CV characteristic curves of the glucose/uric acid sensor were shown in Figures S2 and S3. As shown in Figure S2, the cyclic voltammogram of the CNT/enzyme-modified glucose electrode confirms a stable Faradaic response of the Prussian blue mediator within the potential window used for the amperometric measurements. The well-defined redox behavior and low background current indicate that the mediator layer and CNT/enzyme composite form a reproducible electrochemical interface suitable for sensitive glucose detection. Similarly, Figure S3 shows the cyclic voltammogram of the uric acid sensor, which exhibits comparable mediator redox characteristics, indicating efficient coupling between the CNT network and the uricase layer. This stable electrochemical behavior prior to chronoamperometric testing ensures reliable current readout for uric acid over the concentration range investigated in Figure B. The Ag/AgCl reference electrode (RE) and carbon-based counter electrode (CE) were used directly, requiring no further modification. Prior to calibration, the sensor was exposed to artificial sweat to simulate on-body conditions.

3.

Electrochemical performance, pH effect, and selectivity of CNT-enzyme-based wearable sweat sensors. (A) Dynamic current response of the glucose sensor (applied potential: −0.20 V vs Ag/AgCl; inset: calibration curve averaged from 5 independent devices). (B) Dynamic current response of the uric acid sensor (under the same conditions as (A). (C) Open-circuit potential characteristics of the pH sensor (exhibiting a near-Nernstian slope in the pH range of 4.0–8.0 (51.1 mV pH^–1; R ² = 0.99) with a stable plateau). (D) Effect of pH on current at a fixed concentration (50 μM). (E–G) Selectivity in the presence of representative interferents, including K⁺, H⁺, and nontarget analytes.

2.2.3. Preparation of pH Sensor

For the pH sensor, a first layer of multiwalled carbon nanotubes (MWCNTs), serving as the ion-to-electron transducer, was deposited onto the screen-printed carbon electrode by drop-casting 3 × 3 μL of a MWCNTs dispersion in THF (1 mg/mL). Each drop was allowed to dry for 2 min, ensuring complete evaporation of THF, before the next deposition. Following this, the pH-selective membrane or reference membrane (RM) was deposited by drop-casting 3 × 2.5 μL on top of the MWCNTs layer. Each layer was dried for 3 min before the subsequent drop was added. Both membranes were then left to dry at room temperature overnight. Finally, the electrodes were conditioned overnight in a pH 4.5 solution. After the final deposition, the electrode was stored overnight at 4 °C in a dry environment.

2.3. Implementation of the Sensors into the Microfluidic Cell

The sensor array was affixed to a microfluidic cell using 3 M Medical Transfer Adhesive 4075, ensuring precise alignment of the electrodes with the microfluidic channel. For on-body measurements, the device was affixed to the subject’s arm using adhesive transfer tape, ensuring a secure attachment that prevents sweat leakage. This setup also maintains adequate pressure from the eccrine glands, facilitating continuous and passive sweat flow during perspiration.

2.4. Miniaturized Electrochemical Driving Circuit and Layout

Miniaturized electrochemical driving circuit and layout. The wearable readout adopts a two-board rigid-flex architecture. A 2 × 3 cm rigid analog board hosts the transimpedance and high-impedance front-ends (AD8606), a 24 bit ADC (ADS131A04) for simultaneous acquisition, and a 16 bit DAC (DAC80501) for potentiostatic biasing; a 5 × 3.5 cm flexible board integrates the BLE microcontroller (EFR32MG24; RF-BM-MG24B2), charger/power path management (BQ24045), multirail regulation and a 2.5 V precision reference. The electronics connect to the six-electrode screen-printed sensor via short shielded traces to minimize pickup. With the microfluidic layer (PDMS ≈600 μm) and medical adhesives, the assembled patch thickness is ≈1.2 mm, enabling epidermal conformity during motion. The amperometric channels operate at −0.2 V vs Ag/AgCl (PB-mediated H₂O₂ reduction), while the pH ISE is read in OCP with a high-Z buffer. Data are averaged on-device and streamed over BLE to the user interface for real-time visualization and storage.

3. Results and Discussion

The composition of sweat is highly dynamic and undergoes significant alterations in response to pathological states, making sweat analysis a valuable tool for assessing an individual’s metabolic profile and physiological status. , Previous studies have reported that sweat biomarkers (e.g., uric acid, glucose, pH) may reflect systemic oxidative stress , levels and localized inflammatory microenvironmental changes. Mechanical joints in the human body, such as those in the fingers, elbows, and knees, are particularly susceptible to osteoarthritis due to prolonged stress or repetitive use. Future research could explore the relationship between sweat composition and osteoarthritis progression through sweat analysis, thereby offering a noninvasive and dynamic monitoring approach. Here, we present a wearable patch that measures uric acid, glucose, and pH in situ (Figure ). The multifunctional design of this wearable sensing patch, combined with innovations in materials and device architecture, provides a framework to address key challenges in this field.

1.

Carbon nanotube-induced enzyme polymerization based wearable multifunctional sensor for noninvasive sweat collection and in situ analysis. (A) Wearable carbon nanotube-induced enzymatic polymer sensors mounted on the subject’s arm for analysis of glucose, uric acid, and pH in sweat, with wireless real-time transmission of detection results to the user interface. (B) Assembly of electrode sensing patch structure and schematic diagram of sensing principle for glucose, uric acid and pH value detection, from bottom to top: The adhesive layer with a sample inlet is designed, which is used to adhere the wearable sensing device to the skin surface; the microfluidic channel is used to collect sweat from the skin surface; the electrode chip is used for signal acquisition; the polyamide layer with a sample inlet is designed. (C) Photos of the miniature wearable electrochemical testing system, where the flexible circuit board electronic device combines an electrode sensing patch with an integrated microfluidic structure, is worn on the subject’s arm.

As illustrated in Figure A, the wearable device is encapsulated in the form of a sensing patch for the analysis of glucose, uric acid, and pH in superficial sweat, with the final monitoring results output via a computer user interface. Carbon materials exhibit excellent stability under high scanning voltages, low cost, and favorable biocompatibility with human skin, making them a preferred choice for wearable sensing patches. Figure B illustrates a schematic diagram of a flexible screen-printed carbon electrode sensor for electrochemical sensing detection, along with the electrochemical reaction mechanism for sweat sensing. The sensor consists of three carbon working electrodes (WE1, WE2, and WE3), two Ag/AgCl reference electrodes (RE), and one carbon counter electrode (CE). WE1 and WE2 carry PB and CNT-immobilized enzymes (GO _x or UO _x ) for glucose and uric acid detection. GO _x oxidizes glucose and UO _x oxidizes uric acid, producing H₂O₂; we measure PB/H₂O₂ reduction current at −0.2 V vs Ag/AgCl. The WE3 as modified with highly conductive carbon nanotubes to enhance the conductivity of the carbon electrode, while a pH-selective membrane was further deposited on its surface to monitor pH variations in sweat. The flexible electrode chip was integrated with a microfluidic structure composed of (1) an adhesive layer featuring a sweat inlet, (2) microfluidic channels, and (3) a polyimide layer incorporating a sweat outlet. This design enables the passive and continuous capillary-driven collection of sweat from the human skin surface. The physical appearance and wearability of the sensor system are illustrated in Figure C. A miniaturized electrochemical driving circuit was developed, and a flexible printed circuit board (FPCB) was employed to enhance the system’s wearability. The fully integrated wearable sweat sensor incorporates multifunctional capabilities, including multianalyte data acquisition, signal processing, and wireless transmission. This enables real-time electrochemical signal transmission to a user interface, facilitating on-site monitoring of sweat composition.

The development of enzyme-based biosensors faces critical challenges in achieving high catalytic efficiency, long-term stability, and precise spatial control of enzyme immobilization. Conventional approaches, such as Nafion-mediated enzyme immobilization (Figure C), suffer from inherent limitations including low surface area, random dispersion of enzymes, and weak noncovalent interactions. These factors lead to suboptimal spatial utilization, reduced enzyme loading density, and compromised stability due to enzyme leaching under operational conditions. Furthermore, the insulating nature of Nafion can hinder electron transfer between the enzyme’s active site and the electrode surface, diminishing biosensor sensitivity. To address these issues, we employed CNTs to induce enzyme polymerization, thereby achieving high enzyme loading and efficient utilization of three-dimensional spatial configurations on the electrode surface. As illustrated in Figure A, the carboxyl groups on the CNT surface were first activated using a combination of EDC and NHS. In this process, EDC, a carbodiimide reagent, reacts with the carboxyl groups to form reactive O-acylisourea intermediates, which are subsequently stabilized by NHS to generate NHS esters. These NHS esters exhibit high reactivity toward primary amine groups on enzyme molecules, resulting in the formation of covalent amide bonds that anchor the enzymes onto the CNT surface. Compared to direct dehydration condensation reactions between carboxyl groups on CNTs and amine groups on enzymes, this EDC/NHS-mediated approach provides a more robust and efficient coupling strategy. By introducing intermediate NHS esters, the chemical compatibility and reactivity between functional groups are enhanced, ensuring better alignment of chemical bonding energies and reducing steric hindrance during the coupling reaction. This optimization not only improves the reaction efficiency but also enhances the structural organization and stability of the immobilized enzyme layer.

2.

Schematic diagram of CNTs induced enzyme polymerization and imaging analysis. (A) Reaction equation of EDC and NHS with CNTs. Schema of (B) enzyme immobilization on carbon nanotube and (C) enzyme directly immobilized on carbon electrode surface. SEM image of (D) enzyme, (E) carbon nanotubes and (F) enzyme polymerization induced by CNTs. Fluorescence imaging experiments of (G) CNTs, (H) FTIR labeled enzyme and (I) enzyme polymerization induced by CNTs immobilized on carbon electrode surface.

CNTs exhibit a high specific surface area (theoretically approximately 150 m²/g for multiwalled CNTs) and excellent electrical conductivity, making them highly suitable for electrochemical biosensing applications. As shown in Figure B, the integration of CNTs not only enhances enzyme loading capacity but also facilitates the efficient transfer of electrical signals generated during the catalytic interaction between the enzyme and the analyte to the electrode surface. This significantly improves the sensitivity of the biosensor by minimizing signal loss and optimizing electron transfer pathways. The morphological characteristics of the enzyme, CNTs, and the enzyme-polymerized CNT complexes are presented in Figure D–F via scanning electron microscopy (SEM). From Figure E, it can be observed that CNTs aggregate to form a three-dimensional (3D) interconnected network. This 3D nanostructure provides an extensive surface area and an open framework, allowing the polymerized enzyme layer to achieve better interaction with the analyte. In addition, Figure F demonstrates the microstructure of CNT-induced polymerized enzymes, which exhibit larger and more densely packed assemblies compared to individual enzymes or CNTs alone. This clearly illustrates the ability of CNTs to induce the formation of a polymerized enzyme network.

Fluorescence imaging experiments were conducted to further characterize the mechanism of CNTs-induced enzyme polymerization. In this study, uricase was labeled with FITC, and three types of samples were prepared and analyzed: pristine CNTs, enzyme modified directly with Nafion, and the CNTs-induced enzyme polymerization system, as shown in Figure G,H,I, respectively. The imaging results demonstrate that pristine CNTs exhibit no detectable fluorescence signal. In contrast, the sample modified by direct immobilization with Nafion displays a moderate fluorescence intensity, which indicates a limited enzyme loading on the sensor surface. Notably, the CNTs-induced enzyme polymerization system shows a significantly enhanced fluorescence signal, which suggests a much higher enzyme loading compared to the other two approaches. This outcome highlights the effectiveness of the polymerization strategy in maximizing the amount of enzyme immobilized on the CNTs-modified electrode, which is expected to contribute to improved analytical performance of the biosensor. To further verify the formation of the CNT–enzyme composite, Fourier transform infrared (FTIR) spectra were recorded for pristine CNTs, free enzyme and the CNT–enzyme film (Figure S8). The spectrum of the CNT–enzyme composite shows the characteristic amide I and amide II bands of the enzyme together with the typical vibrational bands of CNTs, confirming the coexistence of an enzyme-rich layer and the CNT network in the hybrid film. Compared with pristine CNTs, the emergence of pronounced amide bands after EDC/NHS activation and enzyme coupling is consistent with successful immobilization of the enzyme onto the CNT surface. These FTIR results, in combination with the SEM and fluorescence imaging in Figure , provide additional evidence supporting the CNT-induced enzyme polymerization and the formation of a three-dimensional enzyme network on the CNT-modified electrode.

The structure of our sensors was shown in Figure S1. Figure S1 depicts the multifunctional electrochemical sensor, where three working electrodes dedicated to glucose, uric acid and pH sensing share integrated Ag/AgCl reference and carbon counter electrodes on the same flexible substrate. This compact layout enables simultaneous multianalyte detection in sweat while maintaining a small footprint compatible with wearable applications. We initially investigated individual glucose, uric acid, and pH sensors to evaluate their standalone performance, as presented in Figure . For the glucose sensor, a continuous analyte addition test was conducted (Figure A), where glucose concentrations were incrementally increased from 0 to 100 μM over 150 s. The sensor exhibited a stepwise increase in current, demonstrating its capability for continuous operation. To ensure the reproducibility of these results, five identical sensors were fabricated, and each concentration was tested repeatedly. The averaged data, shown in the inset, indicate a sensitivity of 2.07 nA/μM with a high linear correlation between current and concentration (R ² = 0.99). These findings highlight the sensor’s precision and its potential suitability for real-time, continuous monitoring in wearable applications. To further evaluate the role of CNTs in enhancing sensor performance, we compared the sensitivity of the CNT-enzyme-modified glucose sensor with that of a sensor modified only with glucose oxidase. Remarkably, the integration of CNTs significantly improved the sensor’s performance, resulting in an 83% increase in sensitivity compared to the enzyme-only configuration. This enhancement is evident across the entire tested concentration range, with the CNT-enzyme sensor achieving a current response of approximately 0.21 μA at 100 μM, compared to 0.11 μA for the enzyme-only sensor. Additionally, the CNT-modified sensor demonstrated a detection limit as low as 0.1 μM (Figure S4), enabling reliable detection of trace glucose levels. The chronoamperometric I–t curves in Figure S4 show rapid and stepwise increases in current upon successive additions of glucose, with well-defined plateaus and negligible baseline drift. These features demonstrate that CNT-induced enzyme polymerization not only lowers the detection limit but also affords fast, stable responses across the concentration range relevant to sweat glucose analysis. In contrast, the enzyme-only sensor exhibited inconsistent and unreliable responses at concentrations below 10 μM.

Similarly, the performance of the uric acid sensor was evaluated, as shown in Figure B. The sensor exhibited a stepwise increase in current with incremental additions of uric acid, ranging from 0 to 100 μM over 150 s. This behavior demonstrates the sensor’s ability to respond reliably to varying analyte concentrations. The calculated sensitivity of the uric acid sensor is 22.17 nA/μM, with a strong linear correlation between current and concentration (R ² = 0.99), confirming its precision and responsiveness. These results indicate the sensor’s capability for accurate uric acid detection, making it suitable for applications requiring precise quantification in biofluids. To further explore the role of CNTs in enhancing the sensor’s performance, a direct comparison was made between the CNT-enzyme-modified uric acid sensor and the enzyme-only sensor. The CNT-modified sensor demonstrated significantly improved performance, with a current response of approximately 2.0 μA at 100 μM, compared to 1.5 μA for the unmodified sensor. This enhancement is consistent with the trends observed for the glucose sensor, further highlighting the beneficial effects of CNT integration. Notably, the detection limit of the CNT-enhanced uric acid sensor was reduced to as low as 0.1 μM (Figure S5), enabling reliable detection of trace levels of uric acid. Consistently, the I–t traces in Figure S5 display clear and reproducible current steps upon successive uric acid additions down to 0.1 μM, indicating efficient electron transfer through the CNT–uricase network. The stable current plateaus at low concentrations further confirm that the CNT-based architecture maintains high sensitivity and signal fidelity in the dilute uric acid regime characteristic of human sweat. In contrast, the enzyme-only sensor exhibited poor accuracy and inconsistent responses at concentrations below 10 μM, indicating its limitations in detecting low analyte levels.

To facilitate the detection of sweat pH and enable necessary corrections to the data from the glucose and uric acid sensors based on pH variations, we also developed a pH sensor utilizing a hydrogen ion-selective permeable membrane. As shown in Figure C, the baseline performance of the pH sensor provides a reliable reference for evaluating the behavior of the analyte sensors across a pH range of 4.0–8.0. The sensor exhibits a linear response to decreasing pH levels, with a sensitivity of 51.1 mV/pH and a correlation coefficient of 0.99. Furthermore, the sensor demonstrates stable potential plateaus over time, indicating consistent performance and reliability in physiological conditions. Glucose current increased with pH (Figure D; R ² = 0.99), underscoring the need for pH correction. Such pH dependence highlights the necessity of precise pH monitoring and calibration to ensure accurate glucose quantification in variable pH environments, such as human sweat.

In contrast, the uric acid sensor (Figure D) demonstrates remarkable stability across the same pH range, with its current response fluctuating narrowly between 0.6 and 1.0 μA without any clear linear trend. This pH insensitivity underscores the sensor’s robustness and adaptability, making it a reliable option for uric acid detection under diverse physiological conditions. The contrasting pH behaviors of these two analyte sensors reinforce the critical role of the pH sensor in enabling real-time pH monitoring and providing corrective adjustments for accurate analyte quantification. By integrating the pH sensor with the glucose and uric acid sensors, the system achieves enhanced reliability and precision, ensuring its suitability for noninvasive, sweat-based health monitoring applications.

The selectivity of the three developed sensors was systematically evaluated using representative interfering species commonly found in human sweat. Glucose, uric acid, potassium ions (K⁺), and hydrogen ions (H⁺) were chosen as interferents to reflect relevant physiological conditions. As shown in the experimental results, both the glucose and uric acid sensors display excellent selectivity, which can be attributed to the high catalytic specificity of the corresponding enzymes. To further verify their robustness against other typical sweat constituents, we also examined the responses of the glucose and uric acid channels to lactate, ascorbic acid, and urea (Figure S11). In all cases, the current changes induced by these species were negligible compared with those produced by the target analyte. The pH sensor also exhibits strong anti-interference capability, which is achieved by employing an ion-selective membrane that allows selective permeation of H⁺ ions. It is important to note that the glucose sensor (Figure E) is significantly affected by variations in hydrogen ion concentration, which necessitates the use of the pH sensor for signal correction. This observation is consistent with the data presented in Figure D, which further underscores the importance of simultaneous pH monitoring for accurate glucose detection under physiological conditions. Beyond the instantaneous selectivity, we also assessed the long-term storage stability of the CNT-based glucose and uric acid sensors. Five replicate sensors were prepared and periodically tested in 50 μM glucose or 50 μM uric acid (in PBS) over a period of 15 days, with each data point measured in quadruplicate (n = 4). As shown in Figure S9 (Supporting Information), the initial current responses of the glucose and uric acid sensors (0.298 μA and 0.863 μA, respectively) decreased by only ∼8.7% and ∼4.7% after 15 days of storage, indicating that the covalent immobilization strategy based on CNTs effectively suppresses enzyme activity loss during long-term storage.

To advance the wearable applicability of our sensors, we integrated individual electrochemical sensors into a unified system and developed a multilayer architecture to facilitate efficient sweat collection and analysis. Conventional microfluidic designs impose persistent limitations on the advancement of wearable sweat sensors for real-time biomarker monitoring, including inefficient handling of small fluid volumes, prolonged refresh times, and contamination risks arising from residual sweat accumulation. Traditional systems often employ simplistic channel geometries with limited inlets and nonoptimized flow paths, resulting in stagnant zones and uneven solute distribution. These shortcomings lead to delayed temporal resolutionwhere sensor outputs lag behind actual sweat secretionand reduced accuracy, particularly for low-abundance biomarkers like glucose and uric acid. To address these challenges, the electrochemical sensing layer was designed to synergize with microfluidics, ensuring both spatial efficiency and signal fidelity. As Figure A showed, the sensor layer integrates three working electrodes (WEs) for glucose (WE1), uric acid (WE2), and pH (WE3) detection, while minimizing cross-talk through shared and isolated reference electrode (RE) configurations. Specifically, WE1 and WE2 share a common RE to reduce parasitic resistance and component count, whereas WE3 employs a dedicated RE and counter electrode (CE) to accommodate the high-impedance requirements of pH sensing.

Building on this optimized electrochemical foundation, a compact, flexible microfluidic module was designed to isolate the sweat sampling area from the sensing electrodes, ensuring contamination-free delivery of freshly secreted sweat for real-time detection. The module’s design prioritized practical constraints such as sensor miniaturization (64 mm² footprint), user comfort (200 μm channel thickness for epidermal conformability), mechanical stability, and sweat collection efficiency. A six-inlet configuration was selected to balance influx uniformity and device compactness, as preliminary tests revealed that fewer inlets (3–4) caused flow bias in low-volume sampling (<2 μL/min), while excessive inlets (>8) compromised skin-device interfacial stability. Circular inlet arrangements were adopted to leverage radial flow symmetry, which reduced shear-induced protein adsorption on channel walls compared to linear arrays.

Numerical simulations systematically optimized the geometric angle spana critical parameter for minimizing solute refresh time (defined as the duration for reservoir concentration to reach 90% of a new target value, 0 μM → 80 μM) at a physiologically relevant flow rate of 2 μL/min. The 180° span achieved the fastest refresh time of 163 s (Figure B), outperforming narrower spans (30°: 415 s; 60°: 352 s; 90°: 300 s; 120°: 212 s) and the 270° configuration (211 s) by 2.5–5 times. This performance disparity arises from distinct hydrodynamic mechanisms: narrower spans (e.g., 30°) exhibit tortuous flow paths that restrict convective momentum transfer, forcing solute homogenization to rely on slow diffusion processes, as evidenced by persistent concentration gradients (blue-to-red transitions beyond 400 s) in spatially resolved profiles (Figure C). Conversely, the 270° design, despite near-complete circumferential symmetry, introduces redundant flow pathways and localized vortices (transitional red-to-blue zones in Figure C), which dissipate kinetic energy and prolong mixing. The increased length of the channel in the 270° configuration results in higher frictional resistance, thereby decreasing the bulk flow rate (Figure S6). In contrast, the 180° span optimally balances radial symmetry and compact geometry, minimizing stagnant zones while preserving convective momentum. This enables rapid solute refresh through coordinated advection-diffusion coupling, as confirmed by the steep concentration–time curve (Figure B) and uniform spatial distribution (90% homogeneity at 163 s). To experimentally validate the simulated flow behavior and sample delivery efficiency, we further carried out a dyed-sweat visualization test in an 180°-span PDMS microfluidic chip. Methylene-blue-labeled artificial sweat was periodically introduced through the six inlets, and time-lapse images confirmed that the dye front uniformly filled the central sensing chamber within 180 s with good radial symmetry, in good agreement with the COMSOL-predicted refresh time and concentration profiles (Figure S10). With the design finalized, the next step involved fabricating a high-precision mold using CNC (Computer Numerical Control) machining. Following mold fabrication, the flexible microfluidic chip was cast in PDMS (polydimethylsiloxane), which is a widely used biocompatible polymer known for its optical transparency, flexibility, and compatibility with biological samples. These characteristics make PDMS particularly well-suited for wearable applications, allowing the chip to conform seamlessly to the skin while enabling real-time monitoring of sweat biomarkers through optical or electrochemical methods (Figure S7).

To further enhance the sweat detection capabilities of the sensor, we developed a miniaturized electrochemical driving circuit and utilized a flexible printed circuit board (FPCB) to improve the wearability of the system. The compact design of the driving circuit ensures seamless integration with the sensor platform, enabling efficient signal acquisition, conditioning, and transmission within a lightweight and portable form factor. The system-level architecture in Figure A integrates multianalyte data acquisition, processing, and wireless transmission for continuous physiological monitoring. Data acquisition is initiated by electrochemical sensors: a glucose sensor generates a redox current proportional to glucose concentration, while a pH sensor produces a voltage signal via ion-selective potentiometry (pH 4–8 range). Both signals undergo transimpedance amplification and high-impedance buffering to preserve signal fidelity. A low-pass filter suppresses high-frequency noise, and differential sampling eliminates common-mode interference, ensuring baseline stability. The conditioned analog signals are digitized by a high-resolution ADC, achieving ±1% full-scale accuracy. A microcontroller executes real-time calibration algorithms and converts raw data into physiological metrics. Validated results are wirelessly transmitted via a Bluetooth module to a user interface for real-time visualization. The physical appearance and practical wearability of the sensor system are presented in Figure B.

5.

Miniaturized electrochemical testing module and in vivo sweat validation. (A) System block diagram. (B) Photograph of the flexible hardware: The module is integrated on a flexible circuit board, coupled with multianalyte sensing patches and a microfluidic module, enabling passive and continuous sweat collection. (C–E) Comparison with commercial benchtop analyzers: In artificial sweat, the selectivity and detection characteristics of glucose, uric acid (amperometric), and pH (potentiometric) are consistent with those of benchtop instruments. (F–G) In vivo human tests (n = 15): Sensing patches were attached to the arm with medical adhesive, and sweat was passively introduced into the detection zone through radially symmetric microchannels. Ethical approval: PJ2024–65 (SYXK 2024–0017); volunteers provided informed consent.

To compare the performance of our miniaturized electrochemical testing module with that of a commercial electrochemical analyzer, we conducted real-time selectivity tests on all three sensors using glucose, uric acid, K⁺, and H⁺ as target analytes and potential interferents. The results shown in Figure C–E indicate that the miniaturized testing module achieves detection capabilities that are comparable to those of the commercial instrument, which is evidenced by the similar selectivity profiles observed for each sensor. This finding demonstrates the reliability and accuracy of our portable device, which supports its application in subsequent on-body sweat analysis and highlights its potential for use in practical wearable biosensing scenarios.

The previous studies have established a positive correlation between the concentrations of biomarkers in sweat and blood, which provides a theoretical foundation for the application of wearable sweat sensors. For example, Umesha Mogera et al. reported the development of a wearable plasmonic paper-based microfluidic device for continuous sweat analysis, which enabled the simultaneous monitoring of multiple biomarkers in sweat and confirmed the positive correlation between uric acid levels in sweat and blood. Similarly, Karpova et al. proposed a noninvasive diabetes monitoring method through continuous analysis of undiluted sweat immediately, revealing a positive correlation between glucose concentrations in sweat and blood. These studies provide strong support for the use of sweat as an alternative medium for noninvasive monitoring of blood biomarkers, while also highlighting that factors such as sweat pH, individual variability, and sweat secretion rate may influence the observed correlations. Building on these previous findings, we utilized our self-developed wearable sweat sensor in conjunction with conventional blood sampling to investigate the correlation between sweat and blood concentrations of uric acid and glucose, as well as to evaluate the accuracy of our sensor. In this study, we measured the glucose concentrations in the sweat of 15 volunteers using wearable sensors, while analyzing their corresponding blood samples via standard laboratory methods. Figure F,G show the comparative results of sweat glucose and blood glucose concentrations among the 15 volunteers. It can be observed that the glucose concentration in sweat ranges from tens to hundreds of μM, whereas that in blood is in the mM range. Although there are significant differences in absolute concentrations between the two body fluids, the variation trends of sweat glucose and blood glucose across different volunteers are essentially consistent. This indicates that sweat glucose levels can dynamically reflect the relative changes in blood glucose, which is consistent with the conclusions reported in previous literature that glucose concentrations in sweat and blood are positively correlated. This further verifies the reliability and application potential of the wearable sweat sensor in this work for real human testing. Comparative analysis with state-of-the-art wearable sensors. To benchmark the performance of our CNT-induced enzyme polymerization strategy, we compared the key analytical figures of merit of the present wearable patch with representative state-of-the-art sweat sensors reported in the literature (summarized in Table S1). For glucose, our CNT–enzyme patch achieves a sensitivity of 2.07 nA μM^–1 and an LOD of 0.1 μM over a linear range of 0–100 μM (R ² = 0.99), which covers the physiologically relevant sweat glucose concentrations (50–300 μM) observed in our volunteers. These values are comparable to the sensitivity of a recent Ti₃C₂T _x /PANI/GO _x -based flexible MXene glucose sensor (25.16 μA·mM^–1·cm^–2 over 0.05–1.0 mM in artificial sweat), while providing more than a two-orders-of-magnitude lower LOD than that device (21 μM, S/N = 3). For uric acid, our CNT–enzyme sensor displays a sensitivity of 22.17 nA μM^–1 with an LOD of 0.1 μM in the 0–100 μM range (R ² = 0.99), which is highly relevant to the typical sweat UA levels (20–60 μM). Compared with a nonenzymatic PyTS@Ti₃C₂T _x -based wearable UA sensor, which exhibits a detection range of 5–100 μM and an LOD of 0.48 μM (S/N = 3) with a DPV sensitivity of ∼57.7 nA μM^–1, our CNT-induced enzyme polymerization strategy yields a substantially lower LOD while maintaining a comparable sensitivity regime. For pH monitoring, our potentiometric pH sensor offers a near-Nernstian response of 51.1 mV·pH^–1 over pH 4.0–8.0 with stable potential plateaus, which is slightly higher than the 39.52 mV·pH^–1 reported for the as-spun pH sensing yarn in a recent multibiosensing hairband platform, while both devices show excellent linearity (R ² > 0.99) over the relevant pH range. In terms of stability, our device delivers reproducible multistep chronoamperometric responses for both glucose and uric acid across five independently fabricated sensors and maintains stable signals during continuous on-body measurements in 15 volunteers without noticeable drift. Although the present study focuses mainly on short-term operational stability, the covalent CNT–enzyme immobilization approach used here is consistent with strategies that have enabled multiday and even multiweek stable operation in integrated wearable sweat-sensing platforms. By contrast, MXene-based and textile fiber-based systems typically demonstrate longer-term stability via material and mechanical design, such as 15 h continuous operation and 10 day storage for the MXene glucose sensor 2 or 24 h drift rates below 0.2 mV·h^–1 in the hairband biosensor array. Taken together, these comparisons indicate that the CNT-induced enzyme polymerization strategy yields analytical performance that is competitive with, and in terms of detection limit often superior to, current wearable sweat biosensors while maintaining the stability required for real-time, on-body physiological monitoring. All human experiments were performed with approval from the Shanghai Experimental Ethics Committee (PJ2024–65, certificate no. SYXK 2024–0017). Written consents were acquired from the volunteers of the research.

Our miniaturized system digitizes amperometric (glucose, uric acid) and potentiometric (pH) signals on-board and streams them wirelessly for real-time visualization. To enhance scalability without hardware changes, the display layer will adopt a channel-agnostic schema (timestamp, value, units, channel label, calibration tag) and support offline-first caching with automatic reconnection for intermittent links. For field-deployable adaptability, we prioritize configurable, pH-aware visualization (linking each glucose trace to concurrent pH), multisession logging with simple CSV/JSON export, and basic security (authenticated pairing and encrypted transfer). These software-centric steps are compatible with the current pipeline and provide a practical path to scaling across users, biomarkers, and concurrent wearable patches in future studies.

4. Conclusions

In summary, this study presents a multifunctional wearable electrochemical sweat sensor that utilizes carbon nanotube-induced enzyme polymerization, which significantly enhances sensitivity and lowers the detection limit for glucose and uric acid in sweat. The sensor platform employs EDC/NHS-activated carbon nanotubes for the covalent immobilization of enzymes, which improves catalytic efficiency and electron transfer while also ensuring long-term stability. The integration of a radially symmetric microfluidic channel, which optimizes sweat flow and reduces refresh time, enables real-time biomarker monitoring in dynamic conditions. Additionally, a flexible polyimide substrate is utilized, which provides conformal adhesion to the skin and mechanical robustness during active movement. On-body validation demonstrates that the system achieves reliable and accurate measurements of glucose, uric acid, and pH, which confirms its suitability for noninvasive and continuous health monitoring. This work establishes a scalable framework for wearable biosensing, which addresses key challenges in sensitivity, selectivity, and integration, and lays the foundation for future advancements in personalized healthcare and real-time physiological monitoring.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacsau.5c01340.

• Schematic diagrams of sensor and microfluidic structure, electrochemical characterization curves of biosensors, FTIR spectra and stability test results, microfluidic performance data, selectivity test figures, and sensor performance comparison table. These materials are available free of charge at (PDF)

CRediT: Chun Bi data curation, investigation, writing - original draft; Wen Fei data curation, investigation, writing - original draft; Jian Song conceptualization, methodology, resources, supervision, writing - review & editing; Shixing Chen formal analysis, methodology, writing - original draft; Yanzhi Dou data curation, formal analysis, methodology, writing - original draft; Shiping Song conceptualization, funding acquisition, project administration, supervision, writing - review & editing; Yifu Zhuang conceptualization, project administration, resources, supervision, writing - review & editing; Lei Cao conceptualization, project administration, resources, supervision, writing - review & editing.

The authors declare no competing financial interest.