Close this search box.

Interindividual- and blood-correlated sweat phenylalanine multimodal analytical biochips for tracking exercise metabolism – Nature Communications

Integrated system sensing strategy and applications

The biochip integrating multiple functional modules can be attached to the skin and samples the secreted sweat on the skin surface for multimodal sensing, data processing, and wireless transmission (Fig. 1a and Supplementary Fig. 1). All modules of the biochip can be prepared at a large scale through simple processing techniques, including thermal evaporation and laser engraving (Supplementary Figs. 2, 3).

To achieve high sensitivity and reliable detection of sweat Phe, we developed a Phe-imprinted MIP that mimics the functions of biological enzymes. Unlike regular MIPs that act as ‘molecular filters’ or ‘artificial antibodies’23, our MIP-based ‘artificial enzyme’ allows for direct and selective electrocatalytic oxidation of Phe on the electrode surface. This approach enables the direct electrochemical determination of sweat Phe concentration using differential pulse voltammetry (DPV). The enzyme-mimicking MIP can not only selectively bind with the target Phe in sweat through specific binding sites, but also directly electro-oxidize Phe via electrocatalytically active surface functional groups (Fig. 1c). The Phe-imprinted MIP electrode was synthesized by electropolymerizing polyaniline (PANI), which has been used for the chiral recognition of aromatic AAs (Tyr, Trp, and Phe) due to its own electrocatalytic activity44,45,46.

Successful template imprinting and extraction from the electropolymerized PANI matrix were verified by various characterization methods (AFM, SEM, and FTIR-ATR) and molecular interaction simulations (Supplementary Note 1 and Supplementary Figs. 4–8). Furthermore, an improved theoretical intermolecular simulation based on density functional theory (DFT) calculations for studying the electrochemical behavior of Phe on the MIP electrode surface supports that the electrocatalytic phenomenon of the PANI-MIP electrode for detecting Phe in this system is related to the carbonyl (C = O) groups on the quinone rings of the polymer chains generated by the electro-degradation of PANI (Fig. 1c and Supplementary Note 2). Specifically, under the same external electric field, a greater charge transfer number (Δq = 3.924e) is calculated between the electro-degraded MIP electrode (E-MIP) and a Phe molecule compared to other electrode cases (Supplementary Fig. 9).

To obtain stable and reliable Phe electrochemical responses in complex sweat samples with variable pH and electrolyte concentration conditions, our biochip system integrates a vertically assembled microfluidic module with a neutral pH buffer filter paper embedded in the sensing chamber (Fig. 1d). As sweat flows through the filter paper loaded with dried phosphate buffer (PB, pH 7.5, 20×), this embedded design maintains the solution environment at a constant pH and high ionic strength level to achieve stable Phe sensing responses. Notably, as shown in Fig. 1e, the introduction of filter paper in front of the electrodes had negligible effects on the DPV responses of the integrated Phe sensing system within the normal physiological concentration range compared to that of pristine electrodes in PBS. In addition, by incorporating the design of a serpentine outflow channel with a rough upper surface, the microfluidic module also allows visualization of sweat flow for sweat loss quantification, including sweat volume and rate (Supplementary Note 3).

We explored the feasibility of combining the measured bimodal signals (sweat Phe concentration and sweat rate) to analyze the Phe partitioning mechanism into sweat, assess exercise metabolic status, and investigate the relationship between sweat and serum Phe levels (Supplementary Fig. 10). In detail, by using the wearable system to collect data from multiple volunteers after 20 min of exercise, we observed a moderately negative correlation between sweat rate and Phe concentration (Fig. 1f), indicating that the mechanism of Phe partitioning into sweat relies on diffusion and its concentration may be affected by sweat dilution47. Remarkably, the correlation in the low sweat rate region was poorer than that in the high sweat rate region because the skin surface Phe content in sweat, which is affected by individual skin quality differences, was difficult to be exhausted by perspiration at low sweat rates20. Moreover, we also defined an ideal indicator for assessing exercise metabolic status with reduced interindividual variability, i.e., sweat Phe secretion rate (μmol min−1 m−2), and identified an individual with possible high metabolic risk during exercise whose secretion rate was greater than that of other volunteers (Fig. 1f). Here, we divided Fig. 1f into three metabolic risk areas, where lower than 1 μmol min−1 m−2 or higher than 2 μmol min−1 m−2 were defined as low risk or high risk, respectively, and between the two values was defined as medium risk. The division was based on a combination of actually measured values from subjects in this study and the similar sports science research20. Finally, through normalizing sweat indicator concentrations by sweat rates to reduce interindividual variability, similar correlations between sweat and serum Phe levels were observed in two different subjects (Fig. 1g), indicating the potential of sweat Phe quantification for non-invasive personalized healthcare management. While there have been several recent pioneering and significant advances in sweat AA sensing23,24,26, an in-depth investigation of AA partitioning mechanisms in conjunction with sweat rate detection and normalization has not yet been conducted due to the lack of multimodal analysis based on supplemental sweat rate measurements in parallel (Fig. 1h). Moreover, the use of other commonly measured sweat biomarkers and their analysis in combination with sweat rate measurements for health monitoring have also been lacking24,36,48,49,50,51,52,53,54.

Electrochemical characterization of E-MIP sensor

The E-MIP electrode showed superior electrocatalytic oxidation for Phe over the electro-degraded non-imprinted PANI (E-NIP) electrode due to fewer binding sites on the E-NIP electrode for Phe to access (Fig. 2a; see also Supplementary Note 1 and Supplementary Fig. 11). In addition, the gold electrode and polypyrrole (PPY) based MIP electrode had negligible responses to Phe due to the lack of effective functional groups. These experimental results indicate that the electro-oxidation of Phe occurs more readily on the E-MIP electrode than other tested electrodes. These results agree with theoretical intermolecular simulations (DFT) used to calculate the electron transfer number between electrodes and Phe molecules under an external electric field (Fig. 2b, c and Supplementary Fig. 9). The incorporation of specific binding sites and active functional groups within the E-MIP enabled the direct, selective, and sensitive electrocatalytic oxidation of Phe (Supplementary Table 2).

Fig. 2: Characterizations of E-MIP sensor for Phe detection.
figure 2

a DPV scans of a PANI-based E-MIP electrode (PANI-E-MIP), a PANI-based E-NIP electrode (PANI-E-NIP), a PPY-based E-MIP electrode (PPY-E-MIP) electrode, and a gold electrode (Au) in a 10× PBS containing 200 μM Phe. b, c Comparison of the electron transfer number between different electrodes and Phe molecule within an external electric field. d, e DPV scans of the E-MIP-based Phe sensor for direct Phe detection after baseline correction (d) and corresponding peak current readouts (e). Inset, the molecular electrostatic potential surfaces of the E-MIP electrode and Phe. The error bars (n = 3 measurements) correspond to the standard deviation (SD). f Response comparison between E-MIP and E-NIP electrodes to equivalent Phe concentrations in PBS. g CV scans of different MIP electrodes based on PANI and an Au electrode in a solution containing 5 mM [Fe(CN)6]3− and 0.2 M KCl. h EIS responses of an E-MIP electrode, a MIP electrode, and an Au electrode in a PBS containing 200 μM Phe. i Selectivity of the E-MIP sensor against other AAs. The following substances were added in succession: 1 mM Glycine (Gly), 1 mM Serine (Ser), 500 μM Alanine (Ala), 500 μM Histidine (His), 200 μM Tyr, 200 μM tryptophan (Trp), and 200 μM Phe. j Selectivity of the E-MIP sensor against common sweat interferents in presence of 200 μM Phe. The following interferents were added in succession: 100 μM glucose (Glu), 5 mM Urea, 5 mM lactate (LA), and 100 μM ascorbic acid (AS). k Batch-to-batch variation of the E-MIP sensor performance in the presence of 200 μM Phe. The error bars correspond to the standard deviation (n = 3 independent sensors). The center for the error bars represents the mean value. l Influence of pH changes in peak currents and respective peak potentials. The dashed box indicates the pH range with stable DPV responses.

A well-defined increase in peak current readouts could be detected by DPV scans in the presence of increasing Phe concentrations from 10 to 1500 μM (Fig. 2d). Two linear relationships were determined: (i) from 10 to 300 μM, which had a sensitivity of 1.4 nA μM−1 and a limit of detection (LOD) of 4.7 μM; and (ii) from 300 to 1000 μM, which had a sensitivity of 0.27 nA μM−1 (Fig. 2e). As a control, a comparative experiment using the E-NIP electrode demonstrated appreciably smaller responses to similar Phe concentrations (Fig. 2f). This difference between MIP and NIP was also characterized in [Fe(CN)6]3−/4− solution by cyclic voltammograms (CV) and electrochemical impedance spectroscopy (EIS) (Fig. 2g and Supplementary Fig. 12). Notably, EIS measured in 200 μM Phe further confirmed that the E-MIP electrode is a more suitable material for electro-oxidizing Phe because the fitting slope (−0.88) referring to Warburg impendence is close to −1 (Fig. 2h and Supplementary Note 1).

The E-MIP-based Phe sensor showed selective recognition of the Phe target, and effective discrimination against other AAs in sweat at physiologically relevant high concentrations (Fig. 2i), along with a variety of common sweat interferents (Fig. 2j and Supplementary Fig. 13). Meanwhile, it also exhibited the chiral recognition selectivity for the L enantiomer (L-Phe) compared to the D enantiomer (D-Phe) due to the use of L-Phe as the imprinted template (Supplementary Fig. 13). The all electro-processed MIP layer could be formed scalably on evaporated gold electrodes, which resulted in high reproducibility in terms of batch-to-batch consistency and continuous successive measurement stability (Fig. 2k and Supplementary Figs. 14, 15). The effect of pH changes on the responses of the Phe sensor was evaluated from pH 6 to pH 11, which indicated a relatively stable response between pH 7.0 and pH 9.5 (Fig. 2l, Supplementary Fig. 16, and Supplementary Note 1). However, human sweat is weakly acidic (pH 5 to pH 7), which motivated us to incorporate a microfluidic method for neutral pH buffering to accurately measure sweat Phe in our biochip system.

Design and performance characterization of multipurpose microfluidics

The multipurpose microfluidic module for wearable sweat Phe sensors was fabricated by laser-engraving, which involved the patterning of channels and filter papers as well as surface roughening (see Methods for details). Each functional layer of microfluidics was integrated in a vertical direction, rendering a delicate 3D structure (Fig. 3a). The resulting microfluidics were capable of not only improving sweat sampling with higher temporal resolution for sweat sensing, but also serving the following purposes: 1) inlet and chamber optimization for fast sweat collection and refreshing; 2) visualized and serpentine outflow channels for in situ assessment of sweat loss status; and 3) maintenance of a pH-neutral and high ionic strength solution environment for accurate sweat Phe detection.

Fig. 3: Design and characterizations of multipurpose microfluidics for sweat sampling.
figure 3

a Layer-by-layer view of microfluidic device design. b Numerical stimulation of time required to fill chamber for different numbers and shapes of inlets with embedded filter paper. c Comparison between simulated and experimental results of the sweat sampling/filling process. d Time evolution of the average Phe concentration for refreshing process in the chamber without or with embedded filter paper. e FEA of microfluidic refreshing process without or with embedded filter paper. The area marked by the dashed line is represented as an anomalous hard-to-refreshing area. f Photographs of the microfluidics during exercise on the skin (right) and optical micrograph of sweat flowing in the visualized microchannel (left, magnified view of the sweat front). Scale bar, 1 cm. g Optical reflectivity of empty channels with or without μ-dots and filled channels with μ-dots. h Measured flow rates by naked eye at different fluid filling positions under pump injection rates of 0.5, 1, or 2 μL min−1. i Sweat rates measured by the algorithm form two body parts of eight healthy subjects during 10 to 20 min of exercise. j Neutral pH buffering capability of the embedded filter paper in the microfluidic chamber. Note: there were no error bars here, but the pH ranges measured by colorimetric pH test papers. k, l Sensor response changes (k) and corresponding DPV scans (l) caused by injection of different sweat sample volumes. Inset, photograph of the biochip. m DPV scans of the integrated wireless system at different flow rates (from 0.5 to 2 μL min−1).

First, the combination of an increased number of inlets and an elliptic inlet design instead of a circle one increased the sampling area to collect sweat. Simultaneously, the presence of filter paper within the sensing chamber also decreased the chamber volume that needed to be filled with collected sweat (Supplementary Note 3 and Supplementary Fig. 17). With an experimentally measured sweat rate (2 μL min−1) as the inlet flow rate, the synergy of these two structural optimizations achieved rapid filling of the sensing chamber in the case of twelve inlets (around 8 min from starting exercise) (Fig. 3b). Finite element analysis (FEA) of the two-phase water/air filling process (Fig. 3c, Supplementary Fig. 18, and Supplementary Video 1) shows accordance with experimental time-lapse images of sweat filling in a volunteer during exercise. Furthermore, numerical simulations also showed that the embedding of filter paper can accelerate and homogenize the refreshing process, which is referred to as the refreshing time taken for the old solute concentration in the chamber to adjust to a new concentration during sweat inflow (Fig. 3d, e, Supplementary Fig. 19, and Supplementary Video 2).

Second, an upper three-layer was introduced into the vertical microfluidic structure to obtain adequate outflow channel volume for readable sweat loss while controlling the overall device footprint (Supplementary Note 3 and Supplementary Fig. 20). In brief, the differential reflectivity/transmittance to visible light at μ-dots leads to an apparent visual color change of the serpentine outflow channel (one meander corresponding to 1 μL, volume resolution of 0.5 μL) in empty and water-filled states55,56 (Fig. 3f, g and Supplementary Figs. 21, 22), which enables visualized flow readouts for estimating sweat rate/volume on the skin. To validate this capacity, a physiological range of constant flow rates (from 0.5 to 2 μL min−1) was injected into the microfluidics by syringe pump. The readable fluid filling positions over time were converted to flow rates measured by the naked eye, showing correspondence between injected and measured flow rates (Fig. 3h and Supplementary Fig. 23). Additionally, a computer vision algorithm was used for automatic sweat loss reading (Supplementary Note 3). Based on this, the exercise sweat rates of two identical body parts (forehead and forearm) of eight volunteers were measured to demonstrate the practicality of this visualization design (Fig. 3i, Supplementary Fig. 24, and Supplementary Table 3), which can be used to comprehensively investigate the relationship between AA loss and water loss during exercise.

Third, a disposable and replaceable neutral pH-buffering filter paper was laser-cut and embedded within the sensing chamber of the microfluidics. The injected 80 μL lactate acid (LA) solution (pH 5.0) was continuously buffered to around pH 7.0 (Fig. 3j), showing the ability to buffer sweat to neutral pH conditions for more than 40 min during intense exercise with a persistently high sweat rate of 2 μL min−1. After incorporating the multipurpose microfluidics for sweat sampling and buffering as well as a matching flexible circuit for wearable DPV measurement (Fig. 3k, inset), wireless and continuous Phe sensing of the multimodal sensors was validated via injection of successive target solutions (raw sweat samples) at physiological sweat rates (Fig. 3k–m and Supplementary Fig. 25). When continuously injecting sweat samples, the sensor performance maintained a stable response (less than 10% variation) with the first 80 μL of sweat injection (Fig. 3k, l), corresponding to the aforementioned neutral pH buffering capability of the microfluidics. The results shown that the embedded pH-buffering filter paper in microfluidics can effectively alleviate the influence of pH as well as ionic strength changes on the sensor response within the design expectation. Furthermore, A stable response of the sensor was been obtained in the skin physiological temperature range (Supplementary Fig. 26). More importantly, the integrated sensor also exhibited stable DPV readouts for Phe sensing at different flow rates from 0.5 to 2 μL min−1 (Fig. 3m), revealing that changes in sweat rate had a negligible effect on sensor responses. We also integrated an ion-selective Cl sensor into this multimodal system for more comprehensive exercise monitoring and management (Supplementary Fig. 27).

Sensor evaluation for assessing exercise metabolic risk

The integrated multimodal sweat sensor could be suitably worn for in situ measurement and display of sweat Phe, chloride, and sweat loss levels during exercise (Fig. 4a–c, Supplementary Video 3, and Video 4). The matching flexible circuit can implement excitation DPV potential and differential OCP measurement as well as corresponding multimodal bio-signal retrieval, processing, and transmission (Fig. 4d). As a demonstration of the sensor validation (Supplementary Video 3), the multimodal sensor was placed on the forehead of a healthy volunteer (subject # 1) during a cycling exercise trial at moderate intensity [around 80% maximal heart rate (HRmax)] for 40 min in a climate-controlled room (22 to 25 °C, 20 to 40% relative humidity).

Fig. 4: On-body evaluation of wearable multimodal biochip for dynamic exercise sweat analysis and assessment.
figure 4

a Photographs of a subject wearing the biochip on forehead and a smartphone app interface. b Screenshot of a frame of the sensor validation video. c Ultrathin and flexible demonstration of the sensor performance. Scale bar, 1 cm. d Hardware block diagram of the flexible circuit for the sweat sensor. e Real-time continuous monitoring of sweat loss (left top), sweat chloride (left bottom) and Phe concentrations (right) along with corresponding DPV data from 0.4 to 0.6 V per scan obtained from the forehead of subject #1. f, g Dynamic sweat Phe measurement (f) and corresponding box-and-whisker plot (g) in two groups of male subjects: lean/normal group (n = 4 subjects) and overweight group (n = 4 subjects). Difference in sweat Phe levels collected for two groups is statistically significant (two-tailed Wilcoxon rank-sum test, W = 378; ***P < 0.001; P = 0.00000169). The box ends represent the 25th and 75th percentiles. The horizontal line in each box represents the median. The upper and lower whiskers represent the maxima and minima, respectively, which refer to the range of non-outlier data values. h Sweat Phe secretion rates of eight subjects calculated by sweat rate and Phe concentration during 10 to 20 min of exercise. i High-positive correlation of sweat AAs and Phe levels for the validation of sweat Phe secretion rate as an indicator to reflect sweat AA loss for exercise-related metabolic risk assessment. The data are based on 28 sweat samples collected from the two groups (lack of data on Subject #3 due to his low forehead sweat volume). The solid line represents the linear-fitted trendline.

The sweat rate was computed from the captured sweat volume and reached a maximum level during the initial stage of perspiration (10 min), and then decreased until stabilization (Fig. 4e). This trend stems from thermoregulation, adaption, and equilibration of the body’s response to the exercise load as expected57,58. Meanwhile, there was a rising trend in the sweat chloride concentrations during exercise. More importantly, the real-time Phe measurement also showed a decreasing trend (Fig. 4e) because the endogenous Phe loss from the plasma gradually dominates while the contribution from skin AAs diminishes20. Similar trends as mentioned above were also observed in a low intensity exercise of jogging (around 60% HRmax) with lower sweat rates (Supplementary Fig. 28). The simultaneous measurement of the above three types of sweat indictors demonstrates the capability of our integrated biochip for multimodal sweat sensing. Among them, Simultaneous measurement of Phe concentration and sweat rate aided a quantitative assessment of exercise-induced AA loss and thereby helps to identify individuals with exercise-related metabolic risk, which have relatively high Phe secretion rate (Fig. 4f)20.

To demonstrate practical application, we selected eight male volunteers with different physiological characteristics from the previously recruited 16 subjects for monitoring sweat Phe levels under the same trial conditions as described above. An analogous decrease trend was observed in all subjects, while there was lower sweat Phe concentrations in overweight subjects (BMI ≥ 25) than in lean subjects (BMI < 25) according to BMI classification (Fig. 4f and Supplementary Table 3). A non-parametric analysis of variance (Wilcoxon rank-sum test) was performed for statistical comparison, revealing a significant difference between the two groups (W = 378; ***P < 0.001) (Fig. 4g). This difference supports that overweight people tend to have lower levels of sweat Phe than lean people due to the concentration dilution effect caused by excessive sweat volume/rate59. Combined with the previous negative correlation between sweat rate and Phe concentration (Fig. 1b), these findings suggest that the partitioning mechanism of sweat Phe may mainly be attribute to passive transport from interstitial fluid (ISF) or blood15,36 in addition to from the skin itself.

Sweat Phe secretion rates of all subjects were calculating by measuring and multiplying sweat rates and Phe concentrations (Fig. 4h), which is an important and suitable indicator for assessing the level of AA loss during exercise without interindividual variability according to statistical analysis (Supplementary Fig. 29). In order to verify this approach, we determined the corresponding AA concentrations of all sweat samples in order to analyze the relation between sweat AA and Phe levels and obtained a high Pearson correlation coefficient of 0.862 (Fig. 4i). Therefore, as expected, the sweat Phe secretion rate can reliably indicate the sweat-facilitated loss of amino acids during exercise, which is useful for assessing exercise metabolic risk and for potentially guiding nutritional supplementation to address losses in sweat and to maintain nitrogen balance20. Owing to an appreciably larger sweat Phe secretion rate compared to other volunteers (Fig. 4h), subject # 2 was identified as an individual with high sweat-facilitated loss of AAs and a candidate for protein supplementation after exercise.

Evaluation of sweat Phe sensor for diet management and serum correlation

As confirmed above and reported before14,20, the levels of AAs in sweat is mainly attributed to ISF or blood partitioning as sweating progresses. Accordingly, beyond its practical use in exercise management, the sweat Phe sensor has additional potential applications to understand the correlation of Phe levels in sweat versus serum such as in the case of diet management (Fig. 5a). Although it has been reported that sweat AA levels are associated with blood AA levels30,60, their metabolic correlation during exercise has not been well studied, especially for non-NMF AAs.

Fig. 5: In situ sweat Phe analysis for assessing serum levels and protein diet intake effects.
figure 5

a Metabolic pathway of Phe along with serum and sweat Phe fluctuations caused by protein intake during exercise. b Dynamic changes of sweat Phe levels from two subjects with different BMI in three periods of exercise, including before and after protein diet intake as well as after rest. DPV data from 0.4 to 0.6 V. c Correlation between sweat AAs and Phe levels in the intake-exercise experiment from two different subjects. d Sensor-measured Phe concentrations in sweat samples versus corresponding ELISA readouts. Data was measured from all sweat samples collected in the above evaluation experiment. The solid line represents the linear-fitted trendline. e Comparative study of sweat and serum Phe levels in the three periods of exercise from subject #1 (top) and subject #5 (bottom). f Correlations between sweat and serum Phe levels before (top) and after (bottom) the sweat rate normalization from two different subjects. Lines represent the fitted trendlines. g, h Fluorescence microscopy images (g) and relative cell viability (h) of HaCaT cells after 2 or 4 days incubation in biocompatibility test (n = 2 independent experiments). Scale bar, 100 μm.

To evaluate the use of our sweat sensors for non-invasively assessing serum Phe levels, a pilot study was conducted for continuous sweat Phe monitoring and corresponding serum Phe quantification on representatives (Subjects# 1 and 5) of two BMI groups before and after protein intake (Fig. 5b–f). In each later stage of exercise sweating (plateau after 30 min), protein diet intake resulted in elevated sweat Phe levels in both subjects, while decreased levels were measured after rest (Fig. 5b). Moreover, the successive decrease of Phe levels in the initial stage of exercise sweating (10 min) points to the consumption of skin Phe and its untimely replenishment (Supplementary Fig. 30). However, Phe levels measured after 20 min did not show this change trend. Combined with the above two different phenomena, it could be inferred that skin Phe in sweat is no longer dominant at this time, but is replaced by the contribution from blood. Importantly, there was a greater extent of Phe percentage fluctuation (especially for sweat changes) in the overweight subject than in the lean subject, which is likely due to different metabolic conditions (Supplementary Fig. 31).

In addition, there was also a strong positive correlation between sweat Phe and AA levels (Fig. 5c), demonstrating the potential of using our sweat sensor for diet management by assessing both changes in serum Phe levels and sweat AA loss. Here, the accuracy of the sweat Phe sensor for testing human sweat samples was validated by enzyme-linked immunosorbent assay (ELISA) using commercial Phe kits (Fig. 5d). Taking the plateau of sweat Phe level in exercise as a comparative index with corresponding serum Phe levels quantified by ultra-performance liquid chromatography with mass spectrometry (LC-MS) (Supplementary Fig. 32), good agreement in the level changes between sweat and serum Phe before and after protein intake was observed (Fig. 5e). High Pearson correlation coefficients of 0.878 (Subject #1) and 0.947 (Subject #5) were observed between sweat and serum Phe levels (Fig. 5f, top). However, the findings also suggested interindividual variability in the correlation between serum and sweat Phe levels due to the large difference in the fitted line. To reduce the interindividual variability, sweat Phe concentrations were normalized to the sweat Phe secretion rate by multiplying by the individual stabilized sweat rate during exercise. This normalized sweat Phe indicator showed a strong correlation with serum Phe levels while also having a similar slope of the fitted line between the two subjects (Fig. 5f, bottom). The difference in the line intercept is likely due to the difference in the physiological properties of subjects and Phe content on the skin surface. Furthermore, the positive interindividual correlation between sweat and serum Phe levels became stronger after sweat rate normalization (Supplementary Fig. 33). In short, by introducing sweat rates to reduce interindividual variability, sweat Phe secretion rates are a potentially suitable indicator to investigate the correlation between serum and sweat Phe. As such, the ability of our sensors to detect both sweat Phe levels and sweat rates displays a superior capacity to assess serum Phe levels over other available sensor options, and supports the feasibility of exploring phenylalanine as a sweat biomarker.


Considering that the biochip may need to be worn for a long period of time in diet management applications, we also conducted a biocompatibility test of the biochip. As a representative skin cell line, human immortalized keratinocytes (HaCaT) were cultured on the biochip as the experimental group, whereas HaCaT was cultured in a blank medium as the control group. As shown in Fig. 5g, the fluorescence microscopy images show the survival status of the HaCaT cells after 2 or 4 days incubation. There was no significant reduction in relative cell viability of HaCaT cells on the biochip compared to the control group (Fig. 5h), indicating that the biochip is safe for prolonged wear and suitable for sweat multimodal detection in diet management.