A Wearable Multisensing Patch for Continuous Sweat Monitoring

In sport, exercise and healthcare settings, there is a need for continuous, non-invasive monitoring of biomarkers to assess human performance, health and wellbeing. Here we report the development of a flexible microfluidic platform with fully integrated sensing for on-body testing of human sweat. The system can simultaneously and selectively measure metabolite (e.g. lactate) and electrolytes (e.g. pH, sodium) together with temperature sensing for internal calibration. The construction of the platform is designed such that continuous flow of sweat can pass through an array of flexible microneedle type of sensors (50 μm diameter) incorporated in a microfluidic channel. Potentiometric sodium ion sensors were developed using a polyvinyl chloride (PVC) functional membrane deposited on an electrochemically deposited internal layer of Poly(3,4-ethylenedioxythiophene) (PEDOT) polymer. The pH sensing layer is based on a highly sensitive membrane of iridium oxide (IrOx). The amperometric-based lactate sensor consists of doped enzymes deposited on top of a semipermeable copolymer mebrane and outer polyurethane layers. Real-time data were collected from human subjects during cycle ergometry and treadmill running. A detailed comparison of sodium, lactate and cortisol from saliva is reported, demonstrating the potential of the multi-sensing platform for tracking these outcomes. In summary, a fully integrated


Introduction
One of the unique challenges in sport, exercise science and healthcare, is the need for continuous, non-invasive monitoring of biomarkers for assessing human performance, health and wellbeing (B. Lo 2011). For example, the monitoring of hydration status and other vital signs during sporting activities can provide a wealth of information regarding one's physiological capacity and efficiency under stress. Alternatively, it may assist in personalising programmes for optimal training gins and recovery. Biochemical markers, while being important for physiological and pathological characterisations, are typically measured with blood assays; however, these can be problematic due to the invasive nature of sample collection and associated risks, along with poor compliance rates among subjects where blood collection is undesirable and difficult.
Sweat has been recognised as an easily accessible bodily fluid that can provide important diagnostic information (Mena-Bravo and de Castro 2014; Raiszadeh et al. 2012). For example, certain genetic disorders (e.g. cystic fibrosis) can be diagnosed in infants from their sweat composition (Rock et al. 2014). Sweat also contains other proteins and metabolites linked to disease and infection (Mena-Bravo and de Castro 2014;Raiszadeh et al. 2012). Humans have two types of sweat glands; eccrine glands cover the entire body and produce sweat to regulate body temperature, whereas apocrine glands are mainly found in the armpits and have a limited role in body cooling. Sweat is a clear, hypotonic and odourless fluid often described as an ultrafiltrate of plasma. It contains mainly ions such as sodium, potassium, calcium, magnesium, chloride and lactate. Sweat is easily accessible, with a typical sweat rate of human males measuring 0.85 mg cm -2 min -1 at the lower back (Patterson et al. 2000).
Several key biomarkers relevant to human health and performance can be assessed in human sweat. Sodium, for instance, is a marker for electrolyte imbalance and important for monitoring athletic performance in hot and humid environments, where sweat (and electrolyte) losses can impair physiological function. In healthcare, a significant loss of sodium in patients with cystic fibrosis (Mena-Bravo and de Castro 2014) can also cause hyponatremia. Sweat pH is another indicator of health and wellness. Lactate concentration in sweat can also indicate the energy pathways contributing to physical activity or general metabolic efficiency. A recent study reported a correlation between lactate levels in blood and sweat during exercise (Sakharov et al. 2010). This biomarker has additional potential for examining oxygen supply to tissue in diseased states, as a fall in oxygen delivery will produce a concomitant rise in anaerobic metabolism and thus, sweat lactate concentration. One such example is the testing of oxidative metabolism under pressure-induced ischemia (Arena and Sietsema 2011).
Electrochemical sensors present a promising technique for continuous monitoring of localised biological changes in sweat, due to inherent advantages of miniaturisation and noninvasive sample collection (Gao et al. 2016;Kim et al. 2015;Lisak et al. 2015;Matzeu et al. 2015). One difficulty faced when developing wearable biosensors is the need for physical contact and continuous analyte sampling. Unlike spot measurements, where large quantities of sweat are collected in a time-lapse manner, continuous monitoring requires the management of continuous sweat flow. In addition, this must be achieved under those dynamic environments (e.g. exercise, training) that active a sweat response. Previous biosensors have some limitations, as they can only monitor a single analyte and do not have on-site signal processing circuitry and sensor calibration mechanisms Jia et al. 2013;Rose et al. 2015). Recently, Gao et al. (2016) reported a fully integrated system that could simultaneously and selectively measure different analytes, as well as skin temperature for realtime sensor calibration. We hoped to extend this work by achieving simultaneous and selective measurements using several integrated sensors, plus temperature control for internal calibration, along with additional capabilities to allow a constant sweat flow for analysis and wireless data transmission to make the platform robust, wearable and easy to use.
In this paper, we present the development of a wearable electronic sensor, complete with microfluidic sampling and wireless readout electronics, which simultaneously measures the concentrations of hydrogen and sodium ions, as well as lactate in human sweat. Additional temperature and humidity sensors were used for internal calibration. The proposed platform is tailored for sweat monitoring during exercise of extended periods by incorporating paper microfluidic channels and reservoirs (Thuo et al. 2014). Sweat enters the paper channel via a small window (10 x 10 mm) through capillary attraction. The small window size ensures that even a small amount of sweat is sufficient for detecting the parameters of interest. The natural liquid wicking properties of paper facilitates sweat sampling, circumventing the problem of sweat accumulation, which often leads to inaccurate acquisition during longer experimental periods (Gao et al. 2016). Sweat is continuously absorbed at the entry window of the smart microfluidic system while sweat collected at the exit ports (reservoirs) are evaporated. This provides a continuous flow through the sensor surface and avoids sweat accumulation.

Materials and Methods
This study was conducted as a two-step process. First, a microfluidic platform was designed to create a constant flow of sweat within a robust and flexible embodiment for the chemical transducers. Low power wireless electronics were also developed and successfully integrated. The miniature electrochemical sensors were calibrated in the laboratory and checked against standard equipment. Second, the sensors were validated on human volunteers during running and cycling activities of increasing intensities. Saliva samples were also collected across exercise for tracking and comparing sodium, lactate and cortisol in this fluid, with an expected trend towards elevated biomarker levels with exercise intensity.

Microfluidic patch design
Several in vitro experiments were conducted with a novel thin flexible wearable patch that incorporates a paper microfluidic channel, with embedded flexible microneedle-based sensors connected with a wireless potentiostat. The advantage of using a microfluidic system is the possibility of using micro volumes of sweat for analyte testing. The fluid system is simple and the size can be modified and tuned depending on the attachment site on the body. An integrated sensing system inside the microfluidic channel also gives the advantage of minimum delay in sampling and subsequent analysis of the sample fluid.
The microfluidic patch consists of layers of polymer, paper microfluidics, and flexible sensors. A CO 2 laser fabrication system (Model VLS 6.6, Universal Laser System) was used to cut out structures on the various polymer layers, which includes a 80 µm thick layer of pressure sensitive adhesive (PSA) (layer 2 from the bottom, Figure 1) that was laminated onto a layer of 50 µm thick Poly (methyl methacrylate) (PMMA) (layer 1 and 4 from the bottom, Figure 1). The sensors were placed inside the microfluidic channel ( Figure 1) which draws a constant flow of fresh sweat passing through them. The continuous delivery of fresh sweat towards the biosensing area is essential for the correct operation of the biosensors during prolonged periods of exercise. However, the use of conventional or miniaturised pumping system (e.g. peristaltic pumps) is not practical for wearable devices. Therefore, a hybrid microfluidic device was produced by simply incorporating paper structures with specific absorption rates inside the microchannel of the platform. The use of paper channels ensures fresh sweat from the skin surface are driven through the microfluidic channel by capillary action towards the electrochemical sensors for measurements. Flow control was achieved by incorporating two different grades of paper; Whatman 4 is a fast absorbing paper while Whatman grade 113 is a high capacity absorbing paper. The Whatman 4 was employed in the straight section of the microfluidic platform (see Figure 1). Conversely, four round pads of the Whatman 113 were located at the terminal points of the channel to collect the analysed sweat and to drive more fresh sweat from the skin through the microchannel. The initial time for the sweat to cover all three sensors is estimated to be 15µl (Fu et al. 2011  To enhance subject comfort during the exercise experiments, the overall thickness of the microfluidic patch was limited to 180µm and a layer of stick-to-skin® double-sided adhesive were used to ensure the device is completely adherent to the skin when exercising. No delamination of the device was observed during the physical exercise of the subjects, while fresh sweat is sampled through the inlet port of the patch (Figure 1). The small transmission circuit is connected with all the sensors on the back of the platform.

Sensor design and fabrication
Platinum and silver wires (diameter 50 µm, Goodfellow Corp.) were used in the fabrication process. The electrode surface was polished sequentially using 1, 0.

Enzymatic membrane transducer
Lactate oxidase from Pediococcus, lyophilized powder, and all common chemicals were obtained from Sigma (UK). Polyurethane membrane was based on a proprietary prepolymer Trixene SC7602 (Baxenden Chemicals Ltd, Accrington, UK). The cleaned electrode was drop-casted with an enzyme crosslinked to bovine serum albumin using glutaraldehyde. To achieve high selectivity for lactate, an internal copolymer of sulphonated polyesther ether sulphonepolyether sulphone (SPEES/PES) membrane is used as internal layer of the electrode. External 10% polyurethane coatings prepared in tetrahydrofuran were applied.

Participants
Six healthy males (aged 25-40 years) were recruited for this study. Pre-screening indicated that they were physically fit and healthy, with no known medical disorders or conditions that would either confound the study results or prevent them from exercising safely. Written consent was sought prior to commencement of the study and ethical approval was provided by the National Research Ethics Service, UK (Reference 10/H0808/124).

Saliva collection and analysis
Saliva samples (~1 mL) were taken across exercise using a passive drool method without stimulation. Samples were provided before, immediately after each exercise bout, and 10 minutes into the recovery period. Saliva was collected into plastic vials and stored at -80C until assay. After thawing and centrifugation, the samples were analysed in duplicate for cortisol concentrations using a commercial immunoassay kit (Salimetrics, Europe). All samples were analysed within a single assay plate, so no inter-assay variation is reported.
Intra-assay CV's on duplicate samples are typically <5% for this assay.

Miniature wearable wireless instrumentation
For data acquisition, custom wireless sensing electronics based on the nRF51822 (Nordic semiconductors) IC with Blue-Tooth Smart wireless transmission were employed. Analogue sensor interfaces were designed with a LT1638 (Linear Technologies) operational amplifier for measurements from the lactate sensor and a LMP7721 (Texas instruments) amplifier for voltage measurements from the ionic sensors. The system developed also caters for impedance and temperature measurements when connected to a thermistor (Epcos B57891M0333K000).
An Android application was developed to capture the data from the wearable sensors. The electronics were housed in a 3D printed case (Objet Eden). The overall size of the wearable package measures 12mm x 30mm x 30mm.

Results
We carried out independent validation of the pH sensor for sweat monitoring during the exercise testing of human subjects ( Figure   3A), a result shown in ex situ using collected sweat samples (Patterson et al. 2000) and in other human studies (Matzeu et al. 2016). The measured sodium concentration varied during exercise and reached a level of 42 ± 1.2 mM (see Figure 3A). The pH measure showed a realtime response that varied from 5 to 6.5, but generally increased as the time of exercise increased.
During cycling exercise, followed by subsequent running on a treadmill, the flexible platform was placed against the skin in the lumbar region of the lower back. As shown in Figure 5C, the sensor readings are presented following different exercise phases from warming up to intense spinning, followed by intense running. The data from the remaining 4 volunteers are summarised in Table 1. These results showed that the concentration of lactate, sodium, pH varied between the volunteers. In general, we noted a rise in the mean values of pH, sodium and lactate across each exercise over time, followed by a slight reduction during the recovery period. The reliability of these measures were acceptable for this study, with CV's ranging from 6% (Na), 7% (pH) and 9% (Lactate).

04
An increase in muscle heat conductance to the skin causes an increase in skin temperature.
To account for this, the sensors were placed inside the microfluidic channel and on the body where the sensors were kept in conditioning solutions before use. Depending on the individual, perspiration started during the initial 10-15 min of the warm up. Initially, the sweat on the skin is slightly acidic, but as exercise continues the sweat rate increases due to thermoregulation and, in this period, we observed a reduction in the reabsorption process (Magalhaes et al. 2010); therefore, sweat pH increases along with sodium concentration ( Figure 4). Sodium levels increased with the increased intensity of the physical activity. The level went up to 35 ± 1.5 mmol·L-1 Na + ( Figure 7B) and dropped down to 15 mmol·L-1 during recovery. These results are consistent with previous studies (Matzeu et al. 2016). A correlation between lactate and sodium levels in sweat was also observed. Lactate concentration (Figure 4 A, B) increased with a higher level of exercise effort. Sweat lactate is a function from the eccrine gland energy metabolism. As a result, increasing exercise intensity leads to an increase production of sweat lactate .
The sensor output was stable during exercise testing and the results were reproducible with a maximum standard deviation of ± 0.215 for pH sensor and ± 0.312 for the sodium based on 6 measurements. The differences in sensor readings between volunteers are consistent with Page 13 of 19 expected variation due to factors such as sweat rates and fitness levels (Havenith and Vanmiddendorp 1990). Lactate concentrations also varied between subjects, which is probably related to differences in lactate excretion and sweat rates with increasing workloads (Buono et al. 2010). The stabilisation of our readings ( Figure 4A) is likely to correspond to the stabilisation of the physiological responses (Derbyshire et al. 2012). Interestingly, further lactate elevations were observed ( Figure 4A and 4B) during the next stage of experimental testing (completed consecutively) and again followed by a slow decline. In general, we noticed a rise in sweating as the cycling workload increased and this was followed by a rise in skin temperature, sweat lactate and sodium levels (Schabmueller et al. 2006).
Temperature readings were taken for re-calibration of the sensors as this affects enzyme activity in the amperometric (lactate) sensors, as well as (to a lesser extent) the potentiometric pH and sodium sensors. We recorded temperature changes between 20 to 40 o C across the volunteers during exercise. This information was used to compensate for the sensor readings. In addition to the sweat sensing devices, we tested salivary cortisol as a key indicator of physiological stress. Intuitively, this hormone should follow the sweat biomarker responses as they change with higher exercising intensities. It has been reported that the levels of sodium and lactate found in saliva are also correlated with blood measurements (Sonner et al. 2015).

Discussion
This paper presents novel work on the development, validation and application of a flexible microfluidic platform with incorporated sensors for direct contact onto the human body, thereby allowing prolonged and continuous measurements of key analytes during exercise. To achieve platform delivery, several areas of specialisation were merged including human physiology, electrochemistry, material engineering, microfluidics, electronics and computer science.
Compared to the previously reported non-invasive sweat sensors (Jia et al. 2013;Rose et al. 2015;Schazmann et al. 2010), we developed a fully integrated system for continuous and simultaneous detection of several physiological parameters inside a passively-activated microfluidic system. A constant potential of +0.65 V (so that to detect hydrogen peroxide, a product of enzymatic reaction of lactate with lactate oxidase) was used for calibrating the lactate sensors, which was performed in vitro in phosphate-buffered saline solution. The linear range (up to 28 mM) covered the physiological values of interest. The internal SPEES/PES layer of the sensor is negatively charged, which makes it highly selective for targeted sensing.
The outer biocompatible membrane protects the sensing layer and prevents enzyme leaching.
Dry storage of IrOx electrodes led to noticeable sensitivity decrease of typically 10-20%.
Soaking these electrodes for 1 h in buffer solution restore their sensitivity to the original value (Ges et al. 2005). Na + and pH sensors had relative standard deviations (R.S.D.) ~ 0.6 % and the lactate sensor has RSD of ~4 % in sensitivity.
Sensitivity of the pH sensor is 71.90 ± 0.8 mV/unit and for [Na + ] is 56 ± 1 mV/unit. Longterm studies showed stability and repeatability of over 8 months for the pH sensor, over 5 weeks for the sodium sensor and over 4 weeks for the enzymatic sensor if kept properly (Modali et al. 2016). The signals from both types of sensors reach steady state within 10 s of starting any measurements. In this work, the sensors were tested both inside and outside the channels, but we observed no differences to suggest that cross-contamination or accumulation occurred. The lactate and sodium sensors also showed excellent selectivity in the presence of varying amounts of different analytes, such as 100 µl glucose, 50 µl uric acid, 50 µl ascorbic acid, 10 mM NaCl for lactate sensor and 10 mM NaCl, 5 mM KCl, 5 mM NH 4 CL, 0.5 mM MgCl 2 , 0.5 mM CaCl 2 for sodium sensors.
Overall, we have developed a flexible platform that can house several sensors simultaneously, with extended capabilities for wireless data collection. Real-time, multiparameter analysis of sweat during exercise provides valuable information about the health and fitness level of an individual, as well as subject-specific responses. In addition, we have shown that the concurrent measurement of lactate, sodium and cortisol can be achieved noninvasively (across a relatively intense bout of exercise) with little or no discomfort to the subjects being tested. This type of monitoring overcomes many of the problems associated with blood sample collection, with much higher compliance and adherence rates among different populations (e.g. athletes, patients).
We do acknowledge some of the shortcomings of this study. For example, the small number of men tested to date using the wearable sensors and the relatively short duration of exercise monitoring. To achieve a better understanding of the health and clinical meaning of the data collected, more studies are needed to examine individuals under conditions likely to promote both physiological and pathophysiological changes. Other directions for research include the deployment of more biomarkers inside the microfluidic platform and sensor validation in other non-physical environments associated with an increase in sweat production (e.g. stress induction and anxiety, hyperhidrosis, temperature and humidity changes).

Conclusions
A highly sensitive wearable system was developed for the continuous measurement of clinically important parameters in human sweat. This integrated system can also transmit information wirelessly for ease of collection and storage, with the potential for real-time data analytics. To advance this work, a large scale population-based exercise study using our wearable platform will be performed.

Hihglights
 Novel results on the development of multisensing flexible platform.
 A wearable patch with incorporated microfluidic channel and fast responding sensors.
 Robust structure allowing an undisturbed sweat flow and huge implications for continuous monitoring.