A wearable sensor for the detection of sodium and potassium in human sweat during exercise

© 2020 The Author(s) The SwEatch platform, a wearable sensor for sampling and measuring the concentration of electrolytes in human sweat in real time, has been improved in order to allow the sensing of two analytes. The solid contact ion-sensitive electrodes (ISEs) for the detection of Na+ and K+ have been developed in two alternative formulations, containing either poly(3,4-ethylenedioxythiophene) (PEDOT) or poly(3-octylthiophene-2,5-diyl) (POT) as a conductive polymer transducing component. The solution-processable POT formulation simplifies the fabrication process, and sensor to sensor reproducibility has been improved via partial automation using an Opentron® automated pipetting robot. The resulting electrodes showed good sensitivity (52.4 ± 6.3 mV/decade (PEDOT) and 56.4 ± 2.2 mV/ decade (POT) for Na+ ISEs, and 45.7 ± 7.4 mV/decade (PEDOT) and 54.3 ± 1.5 mV/decade (POT) for K+) and excellent selectivity towards potential interferents present in human sweat (H+, Na+, K+, Mg2+, Ca2+). The 3D printed SwEatch platform has been redesigned to incorporate a double, mirrored fluidic unit which is capable of drawing sweat from the skin through passive capillary action and bring it in contact with two independent electrodes. The potentiometric signal generated by the electrodes is measured by an integrated electronics board, digitised and transmitted via Bluetooth to a laptop. The results obtained from on-body trials on athletes during cycling show a relatively small increase in sodium (1.89 mM–2.97 mM) and potassium (3.31 mM–7.25 mM) concentrations during the exercise period of up to 90 min.


Introduction
Research into wearable sensors has experienced significant growth in recent years, enabled by advances in electronics, materials and analytical science [1][2][3]. However, the vast majority of devices currently on the market are based on physical sensors, such as temperature, pressure, movement and heart-rate monitors, as incorporation of biochemical sensors into wearable platforms poses complex challenges [4]. Accessing biological fluids such as blood and interstitial fluid also represents a significant obstacle, as the skin barrier must be breached in some manner to access the sample, leading to issues of wearer comfort and potential infection. In contrast, sweat represents a readily accessible biological fluid, whose composition includes electrolytes and metabolites which can provide information on the health status and physical condition of the individual [5]. Electrolytes such as sodium and potassium ions are related to hydration status, and can be used to diagnose and track the efficacy of treatments for conditions such as cystic fibrosis [6], and electrolyte imbalances [7]. Real-time tracking of electrolyte concentrations in sweat can therefore provide important information for biomedical diagnostics, assessing therapeutic interventions and sports performance optimisation. Electrochemical (potentiometric or galvanometric) sensors constitute one of the main approaches to wearable biochemical sensors. Obtaining the lifetime and stability for tracking key biochemical targets during normal daily activity is more challenging for biochemical sensors compared to the much better-behaved physical transducers. Current research on potentiometric, and other electrochemical sensors has been recently reviewed but new systems are reported regularly. Rogers and collaborators demonstrated a system capable of monitoring pH, lactate, glucose, and chloride; using a battery-free device. Javey and Crespo have reported patch sensors for the measurement of H+, Na + , K + and Cl − [8][9][10], while Bertsch focused on NH 4 + [11], and Andrade on total ionic content [12]. Gerasopoulos developed a lactate-sensing patch based on an organic electrochemical transistor and https://doi.org/10.1016/j.talanta.2020.121145 Received 1 April 2020; Received in revised form 2 May 2020; Accepted 8 May 2020 chronoamperometric sensors have also been reported for ethanol [13], urea [14] and caffeine [15]. We have previously reported a watch-type platform for sweat electrolyte monitoring called 'SwEatch' [16,17] and an analogous patch system [18], both of which are capable of drawing sweat from the skin of the wearer through thread fluidics, and tracking the concentration of Na + ions in exercise generated sweat through embedded potentiometric sensors. In this article, we report further advances which improve the design of the SwEatch platform and extend its use to the analysis of multiple electrolytes. The modular design of the platform easily allows the selection of the analyte by simply replacing the electrode component. The device was characterised via lab testing prior to use in trials with volunteer athletes for a period up to 90 min in the School of Health and Human Performance at Dublin City University.

Instrumentation
A DEK 248 semi-automatic screen printer was used for the deposition of conductive and dielectric layers in the electrodes. PMMA parts were cut with an Epilog Zing Laser Cutter. Electrodeposition of PEDOT was carried out using a CH Instruments CHI630B electrochemical analyser. Potentiometric measurements were recorded using a multichannel Lawson Labs MCV potentiometer, with a double junction Ag + / AgCl electrode (Sigma-Aldrich, Ireland); cell potentials for the integrated electrodes were measured using the Shimmer PCB and acquired with Consensys v1.5.10 software (Shimmer Ltd.).

PEDOT based Na + and K + ion selective electrodes
The electrode fabrication procedures have been improved from those previously reported for the fabrication of PEDOT based Na + ion Selective Electrodes (ISEs) [16,17,19]. Furthermore, in this work, in addition to the Na + ISEs, analogous K + ISEs were also employed. Briefly, the electronically conductive layer (carbon ink) and dielectric layer were screen printed on Polyethylene terephthalate (PET) sheets. For the Poly(3,4-ethylenedioxythiophene) PEDOT formulations, PEDOT (~0.8 mg per electrode) was deposited on the exposed surfaces of the electrodes by constant-potential electro-polymerisation from 0.05 M solution of EDOT (97%) in EMIM NTf 2 (99.5%), using a three-electrode setup consisting of the carbon ink working electrode, a platinum wire counter electrode and a bare Ag wire as pseudo-reference electrode. Polymeric capping membranes were formed in the wells by dropcasting the THF-based cocktails. The cocktails for the Na + ion-selective electrode (Na + -ISE) and reference electrode (RE) used with the PEDOT interfacial layer were as previously reported [21]. The K + selective ISE cocktail was valinomycin (10% by weight), KTCPB (50% mol amount vs ionophore), and 1/2 ratio of PVC and plasticiser (DOS).

POT based Na + and K + ion selective electrodes
For the POT based Na + and K + ion selective electrodes both ISE and RE cocktails remained the same formulation as outlined with the addition of 10% by weight of POT directly to the membrane. These cocktails were used to prepare membranes with the same drop-casting protocol but removed the need for a separate mixed ion-electron conductivity layer with the PVC membrane deposited directly on the carbon ink sub-layer. Prior to drop-casting, a 20% solution of plasticiser (DOS for ISEs, 2 μL, DEHP for Res, 4 μL) in ethanol was added into the PMMA wells. The total amount of drop-cast membrane was 3.2 mg for the RE, and 1.4 mg for each of the ISEs (see Fig. 1).

Dropcasting automation
The deposition of the polymeric electrode membranes by drop casting is a labour-intensive process that can introduce sources of error which limit inter-electrode reproducibility. To try and address this issue, the drop casting procedure was automated via the use of an Opentrons® liquid handling robot (see Fig. 2). The robot was interfaced through an open source Python API, and a programme was developed in-house for automated drop casting of the electrode membranes. Different drop casting protocols were interactively defined through a graphical interface. After fabrication, each electrode was conditioned prior to use to facilitate hydration of the membranes and ensure equilibrium population of the membrane with the respective primary ions in the conditioning solution. This was achieved by immersing each electrode for 2 h in a 0.01 M solution of the primary ion (NaCl or KCl). Conditioning for shorter times or with lower concentrations resulted in a diminished response from the both the K + and Na + ISEs.

SwEatch platform
The 3D printed SwEatch pod-like platform previously reported [17] was redesigned to incorporate a dual macro-duct that provides for twochannel direct electrolyte monitoring in sweat. The main 3D printed platform assembly was a three-part design that enabled easy replacement of electronics, electrodes and sorbent material during use. The 3D printed platform was separated into three main components (Fig. 3); the microfluidic unit ( Fig. 3A), platform body ( Fig. 3B) and the fully integrated wearable platform (Fig. 3C). The microfluidic unit incorporated the dual macro duct and the two-half cylindrical arrangement of the sweat reservoir (Figs. 3 and 5A) which enabled two independent sample channels for the sodium and potassium electrodes ( Fig. 3 and A). The electrodes were secured in location to prevent P. Pirovano, et al. Talanta 219 (2020) 121145 longitudinal and lateral displacement during trials. This was achieved using flexible polymer TANGOBLACK contact pads to produce a snug fit between the lid and the macro-duct base (Fig. 3 1A). Redesign of the microfluidic 3D printed sub-assembly provided improved contact between the electrodes and the capillary flow of the sample in the fluidic channel. The multi-material 3D prints also enabled better sealing and close fit between the microfluidic unit and main platform body (Figs. 3 and 4B). The two-part body was held in place with a nylon M3XM12mm nut and bolt arrangement (Figs. 3 and 2B) in which the nut was counter-sunk flush with the surface. The platform employed a vertical arrangement of these units to enable the components to be arranged in a compact form factor. The platform held two Shimmer single channel PCBs, and two 3.7 V 155mAh batteries ( Fig. 3 and B). The pebble shape added rigidity to the enclosed platform (Figs. 3 and 1C) by providing smooth design counters which avoid localised stress formation. The strap was fed through three loops located at the top and each side of the platform to ensure continuous contact with the skin and prevent movement during on-body trials.

On-body trials
The trials were conducted in association with the School of Health and Human Performance at DCU (full consent and ethical approval were obtained). The volunteer athlete was selected with a moderate to high fitness level. The fluidic device was primed by wetting the protruding thread with 30 μL of deionised water and fastened onto the upper arm of the subject with elastic straps, after cleaning the skin surface with a deionised/distilled water rinse and drying with a sterile gauze. On-body trails were carried out where the subject was exercising on a stationary bicycle for 90 min for the first and second trial, respectively. During each trial, contextual information such as the room temperature, humidity power exerted, arm movement and hydration regime were recorded. The amount of sweat collected was determined by weighing of the sorbent material before and after the trials. The electrodes were calibrated before and after the trials, and the potentiometric data was converted into analyte concentrations by applying the calibration data using a linear drift correction model.

Electrode fabrication
During manual drop-casting a number of defects were observed in both the polymeric membranes (see S1 Fig. 1). The defects such as air bubbles and membrane overlapping resulted in high failure rates of the reference and ISEs (see Table 1). To combat this issue, the process was automated using the Opentrons ® automated pipetting robot (see Fig. 3). This move toward automation resulted in a significantly reduced defect rate for both the reference and ISEs ( Table 1).
The electrodeposition of a transducing layer of poly(3,4-ethylenedioxythiophene) (PEDOT) was also found to be a bottleneck in the fabrication as it requires up to 1 h per electrode and is not easily scalable. For this reason, we investigated the use of solution-processable alternative, specifically poly(4-octylthiophene) (POT) [20]. As   mentioned above, for these electrodes, the mixed conduction polymer is added as a component of the membranes in a 10% content by mass, and the soft polymeric membrane formed directly on top of the carbon ink sub-layer, without creating a distinct interfacial transduction layer. This further simplified the fabrication process through integration of the mixed conduction and ion recognition layers into a single hybrid layer.

Characterisation of electrode response
Na + ISEs based on the calixarene ionophore, K + ISEs based on valinomycin, and REs based on the ionic liquid 1-Hexyl-3-methylimidazolium tris-(pentafluoroethyl)trifluorophosphate (HMIM FAP) [21], were first characterised using a commercial Ag/AgCl reference electrode in solutions with log ionic activity ranging from -4.0 to -1.0. The REs demonstrated a stable potential over all concentration ranges which can be expected to be encountered in sweat. The Na + ISEs incorporating PEDOT exhibited a linear response to Na + over the range log(a) = -4.0 to -1.0, with slopes of 52.4 ± 6.3 mV/decade (n = 8). The sensitivity of the corresponding K+ PEDOT electrodes was found to be somewhat lower at 45.7 ± 7.4 mV/decade (n = 8), with a linear range from log(a) = -4.0 to -0.5. The POT based formulations showed improved sensitivity in the log(a) = -4.0 to -1.0 range, specifically 56.4 ± 2.2 mV/decade for the Na + ISE, and 54.3 ± 1.5 mV/decade for the K + ISE (n = 8 in both cases). In the following sections, we focus particularly on the characteristics of the POT based electrodes, to explore whether the hybrid layer configuration affects the electrode performance.

Stability and selectivity
Following initial characterisation described above, the selectivity and stability of the electrodes was examined in the presence of the most relevant interferents for the application, that is, Na + and K + , followed by H + , Mg 2+ and Ca 2+ [22,23]. A calibration experiment in which the response of the ISEs and of the RE to alternating additions of NaCl and KCl were recorded (Fig. 4, A). Each step corresponds to an increase in log(a M+ ) of either Na + or K + of 0.5 M, calculated using the Debye- Fig. 3  Hückel model. Each ISE displayed a rapid response even in the mixed analyte solution, with no evidence of cross-response, and reasonable sensitivity (Fig. 4, B). The potentiometric selectivity coefficients (Fig. 5) were determined via the fixed interference method, with a background concentration of interferent of 0.1 mol/L [24], giving values of log In the simulated Sweat C solution, a phosphate buffer was used to generate the slightly acidic pH which sometimes occurs in sweat.
The drift of a test batch of POT electrodes (n = 71) was measured consecutively for a period of 2 h in each solution. The results are reported in Table 2, for the Na + ISEs, K + ISEs, and REs. The Ag/AgCl reference was shown to be stable (< 1.0 mV/h) under the same conditions, by measuring its potential against a standard calomel reference electrode. Most electrodes presented a modest drift, with a random distribution varying randomly from one electrode to the next. Given the low magnitude of the drift over the typical period of an exercise event, the electrodes were deemed to have satisfactory stability for on-body measurements. Prior to unbody measurements the sensors were stored dry in the dark and re-calibrated before use as previously reported [18].

Fluidic design
The uptake and transport of the sweat, within the dual macro-duct, occurred in the same way as previously reported [17]. The mirrored fluidic unit was capable of drawing sweat from the skin through passive capillary action and bringing it into contact with both the Na + and K + combination electrodes (see supplementary information Fig. 3a). The flow rate is roughly constant with 1 or 2 threads (linear flow rate a ca., 1.0 and 7.0 μg/min respectively), whereas with a triple thread, saturation behaviour occurs, with gradually decreasing flow rate from an initial high value of 26 μg/min due to exhaustion of the available sample storage reservoir, which can accommodate up to 1.8g of sweat (See SI Fig. 3b). The use of a double thread was therefore judged to be   6. Response of Na + ISE (top row), K + ISE (middle row) and RE (bottom row) to different analytes/interferents. The electrodes were exposed to increasing concentrations of analyte/Interferent (10 −4 , 10 −3 , 10 −2 , 10 −1 mol/L). Addition times are indicated by the grey vertical lines. Initial potentials were normalised to E = 0. the most appropriate as it represents a medium to high sweat uptake which is greater than the expected rate of sweat generation reported for the sampling area and location [25,26]. The maximum rate of sample uptake into the SwEatch platform should be greater than the expected rate of sweat generation, in order to minimise sweat pooling at the skin contact area and ensure that there is minimal mixing from sweat emerging at varying times during on-body trials.

On-body trials
Four-point calibrations of the solid-state combination electrodes for Na + and K + (POT) electrodes were performed prior to commencing onbody trials. Calibrations for both Na + and K + were carried out by using 10 −4 ,10 −3 ,10 −2 ,10 −1 M NaCl and KCl. The known solutions were pipetted into the inlet of the platform where the thread brought the liquid in contact with the sorbent material at the reservoir which wicked the liquid across the Na + and K + combination electrodes. Results were transmitted to the laptop via Bluetooth and processed using the Consenys software by taking 100 data points from at each single decade interval (Fig. 7). The pre-trial slopes and intercepts for Na + were 61.13 mV/decade, 293.74 mV and for K + were 57.61 mV/decade, 183.78 mV, respectively.
During the trials, the SwEatch device was placed on the athlete's upper arm (Fig. 8) as this site produces significant sweat while also restricting movement of the device during the trail. The volunteer cycled for 90 min with an average speed of 33.8 km h −1 at an effort of 149-159 W leading to profuse sweating from the individual during the session. The volunteer's heart rate was also monitored throughout the trial (mean = 156 bpm), and the temperature in the room was kept relatively constant at 25.5°C and 19.1%, respectively (additional information can be found in supplementary information Table 1).
The outputs of the fully integrated SwEatch platform measuring Na + and K + for the duration of the session are shown in Fig. 9. The data clearly shows the sharp increase in the signal for both Na + and K + corresponding to sweat reaching both electrodes at ca. 8 min as previously reported [16,17,27]. For Na + after the initial rise the signal remains relatively stable from ca. 11 min with a slight increase at ca. 34 min until ca. 48 min at which time the signal starts to decline. This response pattern for Na + is consistent with what we and others have previously published [8,17,28] during on-body trials. However, the equivalent concentration range for Na + was slightly below the typical physiological range (10 mM-90 mM Na + ) [29]. The Na + concentration showed the expected increase from a baseline of 0.03mM (1) to 1.89 mM (2) corresponding to sweat entering the device at ca. 8 min.
The concentration rises to 2.97 mM at ca. 34 min (3) before decreasing to 2.21 mM (4) at ca. 58 min and further to 0.61 mM (5) at ca. 78 min. The low concentration observed in Na + levels throughout the trail can be attributed to a number of factors which include; the subjects sweating rate, skin temperature, electrolyte reabsorption amongst others. Further validation of the SwEatch platform against the 'gold standard analytical techniques' combined with increased trials are planned to fully assess the patterns observed. This will therefore allow us to further elucidate the potential of these type of wearable platforms to provide reliable data on the mechanisms of sweating and the physiology status.
For K + the results also showed increase in concentration from a baseline of 0.13 mM (1) to 3.31 mM (2) at ca. 11 min again corresponding to the introduction of sweat into the SwEatch platform, the slight differences in time reflecting the slightly different sweat transfer rates through the device pathways to each sensor. The concentration starts to decrease after ca. 14 min to a low of 1.65 mM (3) after which the concentration increases to 2.63 mM (4) at ca. 52 min and further to 7.25 mM (5) at ca. 88 min. This concentration range falls within the typical physiological range reported for K + (between 2 and 10 mM) [29]. No significant movement artefact was observed when the subject raised the arm to which the SwEatch platform was attached at ca. 18 min and at ca. 46 min, suggesting that the improved design is more robust to movement previously reported [16,17]. To confirm the stability of the electrode during the trials, post-trial calibrations were compared to pre-trial calibrations (Fig. 10). Virtually no change was  exhibited, indicating that the electrode response remained stable throughout the trial. These results add to the body of knowledge of sweat electrolyte monitoring using on-body devices. Given that the electrodes appeared to be inherently stable over the course of the trials, and movement artefact was significantly reduced, there can be a certain degree of confidence in reliability of the signals obtained. An integrated calibration facility that can provide in-trial validation measurements at regular intervals would be ideal, but this is difficult to achieve practically, as it would involve storing at least one on-device standard, and switching this into contact with the electrodes, increasing the fluidic system/electronics complexity. Such added functionality may become available as fluidics based on biomimetic principles become more available and mature. Alternatively, it may be possible to periodically perturb the electrodes using an impressed signal e.g. apply a voltage and check current obtained in the presence of the sample to estimate the system impedance/resistance. Over time, as more groups publish trial-based data using on-body chemical sensors, the patterns of electrolyte concentration variations in sweat during exercise will become much more established and contribute to a clearer interpreation of the physiological basis of these patterns. This will be part of a much bigger effort to expand real-time measurements using a wide range of on-body biochemical sensing, as is already happening with on-body patch-type sensing of glucose for personalised diabetes management [30].

Conclusion
In summary we have demonstrated the use of a wearable platform for the simultaneous detection of Na + and K + in sweat. The platform presented has differentiating factors compared to previously developed SwEatch platforms [17,31] such as simpler and partially automated electrode fabrication, improved electrode stability, platform redesign to accommodate two-channel direct electrolyte monitoring in sweat and resistance to movement artefact. This combination of changes to the platform have generated a new set of real-time data for measuring electrolytes in sweat in high performance athletes. However, it is acknowledged that these results should be interpreted with caution as the use of such technologies can be the source of unwanted variability in sweat monitoring. Increased validation of the response patterns obtained for electrolyte variations is required to establish if these variations are real or accurate. Validation of the response patterns will only emerge when a substantial body of trial-based time series becomes available that exhibit similar trends. Establishing the true concentration ranges will require further creative thinking to address how to perform parallel measurements ideally on-device (same sample) during on-body trials.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.