A wearable fluidic collection patch and ion chromatography method for sweat electrolyte monitoring during exercise

This paper presents a method to continuously collect and reliably measure sweat analyte concentrations during exercise. The method can be used to validate newly developed sweat sensors and to obtain insight into intraindividual variations of sweat analytes in athletes. First, a novel design of a sweat collection system is created. The sweat collection patch, that is made from hydrophilized foil and a double-sided acrylate adhesive, consists of a reservoir array that collects samples consecutively in time. During a physiological experiment, sweat can be collected from the back of a participant and the ﬁ lling speed of the collector is monitored by using a camera. After the experiment, Na + , Cl (cid:1) and K + levels are measured with ion chromatography. Sweat analyte variations are measured during exercise for an hour at three di ﬀ erent locations on the back. The Na + and Cl (cid:1) variations show a similar trend and the absolute concentrations vary with the patch location. Na + and Cl (cid:1) concentrations increase and K + concentrations seem to decrease during this exercise. With this new sweat collection system, sweat Na + , Cl (cid:1) and K + concentrations can be collected over time during exercise at medium to high intensity, to analyse the trend in electrolyte variations per individual.


Introduction
Recent developments in exible and wearable electronics for sweat sensing offer new opportunities for gathering physiological information from human beings real-time. Current healthcare systems are moving towards connected care solutions and measuring the physical status of a patient in their home situation. For healthy persons, sweat sensors can potentially play a role in prevention of diseases and to motivate people to exercise. This research focuses on sweat sensing applications in a sports context. Until now, little is known about how sweat analytes change over time and how they can be used to monitor athletes. The physiological literature about this topic is scattered and papers oen show contradicting results. The application of sweat analytes in monitoring the physiological status of athletes is currently limited, because little is known about how sweat analyte concentrations relate to blood values. 1,2 An important reason for this is the lack of suitable sweat collection methods for physiologists, who are studying the physiological mechanisms behind sweating. Local methods include collection with absorbent patches, pouches and the Macroduct sweat collection system. [3][4][5][6] All these methods require repeated placement and removal of these devices when you want to measure electrolyte variations during an exercise. Several challenges of sweat measurement can be identied. Below, the most important challenges are reported, based on the literature and our own experiments. 3,7 (1) Sweat rates vary over time and inuence the sampled ion concentrations.
(2) Due to the low sweat rates per gland (nl min À1 mm À2 ), sample volumes are small.
(3) Skin and sweat gland metabolism inuence the concentration levels in sweat.
(4) Contamination of new sweat with old sweat or by skin contaminants can occur.
(5) Sample collection is difficult due to evaporation from the highly distributed sweat glands, and the irregular skin surface. (6) Several sweat analytes, such as glucose and proteins, are present in very low concentrations.

(A) Electrolytes of interest
This research focuses on measuring electrolytes in sweat to keep track of the physical status of an athlete. Stefaniak and Harvey gave an overview of the main constituents in sweat and their median concentrations. 8,9 According to the literature several constituents are of potential interest for monitoring the athlete's physical status during exercise.
Eccrine sweat mainly consists of water and NaCl. Na + and Cl À are secreted in the secretory coil of the sweat gland and a part is reabsorbed in the duct. When the sweat rate becomes higher, less ions are reabsorbed in the duct and the concentrations of Na + and Cl À are higher. 2,10 Since the concentration varies with the sweat rate, measuring Na + and Cl À levels may be useful in determining the sweat rate of an athlete. Several researchers state that Na + and Cl À increase when an athlete dehydrates and lower levels of Na + and Cl À can indicate electrolyte loss. 1,11 However, papers that present Na + and Cl À measurements together with changes in the hydration status, show mixed results. 12 One of the main reasons for this is the lack of standardized collection and analysis methods.
For K + , part of the secretion mechanism is known. However, contradictory results about K + variations in sweat and the relation of K + levels to the sweat rate are found in the literature. 13,14 It is also expected that K + levels are oen overestimated due to skin contamination. 15 More research is needed to nd out if this parameter turns out to be useful in monitoring the physical status of an athlete.
Although the literature shows contradictory results for NH 4 + measurements as well, some researchers state that NH 4 + levels are related to lactate levels in the blood, which makes this analyte potentially interesting for tracking muscle fatigue. 16 Gao et al. designed a sensor array that enables potentiometric measurements of electrolytes, like Na + and K + and amperometric measurements of metabolites such as lactate and glucose. Mugo and Alberkant developed a non-enzymatic cortisol sensor that makes use of molecular imprinted polymers for selectivity. 21 In addition, Yuan et al. created a sensor that can measure the sweat rate, total ionic charge and sodium concentration. 22 A few examples are presented above. To give an overview of current sweat sensor developments, several literature reviews are published. [23][24][25][26][27] Secondly, researchers use microuidics systems for sweat sensing purposes. Ma et al. designed a patch that contains a wick and a microuidic channel that collects the sweat and brings the sweat to the electrode of the sensor. 28 Koh et al. created microuidic channels from PDMS that enabled colorimetric measurement of sweat and Choi et al. created a micro-uidic network with small bursting valves for chrono-sampling of sweat. 29,30 Sample volumes of the collected sweat in these systems are small (e.g. 6 ml). Aranyosi et al. created a uidic system for single but larger samples. 31

(C) A new sweat collection system
Although a large number of compact and continuous sweat sensing systems are presented in the recent literature, applications for monitoring an athlete's or patient's status are still limited.
Physiologists lack suitable collection and analysis systems for continuous monitoring to nd useful sweat biomarkers and how they change over time. In the technological literature, validation of novel sweat sensor systems in human trials appears difficult, because there are no standardized methods to perform reference measurements by chrono sampling sweat and analysing it in the lab aerwards. To solve these problems, we propose a new sweat collection device that can automatically collect a sequence of sweat samples (>100 ml). The system is easier to fabricate and the collection surface is larger than in previously presented solutions, which facilitates the use of the system by both engineers and physiologists and increases reliability of the measurements. Physiological experiments and chromatography measurements are executed to prove that this method can be used to analyse electrolyte variations over time.
The sweat collector device can in principle be connected to an electrochemical sensing system for continuous measurements. The testing of our new sweat collectors with an in situ analysis system will be discussed in a subsequent paper.

Method
In order to allow for continuous uptake of sweat volumes and to simplify the collector placement step, it was decided to develop a simple disposable foil type sensor patch that can automatically collect a sequence of 5 samples. The design, simulations of the inow of sweat and lab experiments are presented rst. Aer demonstrating the working principle of the collector, the new system is tested in a physiological experiment. Sweat is collected with the new patch during exercise and analysed in the lab with ion chromatography. This is presented in Section (B-D).

(A) Design & simulations
The uidic system contains a sequence of reservoirs that are created from two layers of hydrophilic lm (Visgard 275, 32 a PET lm with a PU coating) with a double-sided adhesive (3M 1522, 33 a PE tape with an acrylate adhesive) in between.
(1) Design. In Fig. 1, an exploded view and a front view of the nal design of the sweat collection system are presented. The sweat collection surface is 40 cm 2 . The funnel-shaped 2D structure of the skin adhesive ( Fig. 1, no. 1) guides the sweat to the inlet. In combination with the grating structure, it serves as a spacer to the skin too, to make sure that the sweat drops down.
Due to capillary forces, the sweat ows from the inlet (Fig. 1, no. 7.) to the reservoirs (Fig. 1, no. 10.). The walls of the capillaries are highly hydrophilic to ensure that the ow rate of sweat is not negatively inuenced by the channels. The sweat passes a T-junction ( sharp edges at the junction and by directing the le branch of the junction upwards, it is ensured that the sweat goes straight into the reservoir. When the reservoir is lled, the sweat will take the le turn and will ow to the next reservoir by capillary forces. Entrapment of air is prevented by placing air outlets ( Fig. 1, no. 9) at the right corner of each reservoir. The materials were cut into the desired shapes with a CO 2 laser system (Merlin Lasers, Lion Laser Systems, The Netherlands). The lm is rinsed with demineralized water (to remove the extra ionic surfactant that was applied to the lm by the manufacturer, because it inuenced the sodium measurements in preliminary experiments). Samples of the foil and adhesive were placed in vials with demineralized water (3 ml, 24 hours), and ion concentrations (Na + , Cl À and K + ) in these vials were analysed to ensure the absence of background contaminants from the materials. No signicant Na + , Cl À and K + peaks were detected in these samples. Contact angles of the materials were measured with optical tensiometry (KSV Instruments Ltd). The spacer tape has an average contact angle of 50.5 and the hydrophilic foil has an average contact angle of 92.2 (Fig. S1, ESI †).
(2) Simulations. Gravitation and capillary effects are used to make sure that the sweat ows into the reservoirs at a similar or faster volumetric ow rate than the sweat rate. The capillary effects depend on the surface tension and the geometry of the tube. A relation between the capillary pressure and the contact angle and dimensions of the microchannel with a certain height (h) and width (w) is given by the Young-Laplace equation: where P is the capillary pressure, g the surface tension of the liquid and q t , q b , q l , and q r are the contact angles of the top, bottom, le and right microchannel walls. For the design of capillary channels, there are a few practical guidelines. First, contact angles smaller than 60 are preferred. 34 Secondly, some of the walls of a microchannel can be made of a hydrophobic material if one takes into account that the ratio between, for example, the hydrophobic height and the hydrophilic width of the channel is very low. Furthermore, uid ow near corners needs to be considered since it can affect the lling of the channels negatively by entrapment of air bubbles. 35 It can be prevented by rounding the edges and corners of the channels, for example. Several designs were made and CFD simulations with COMSOL Multiphysics soware were performed to test the functioning of the devices. For the simulations it is assumed that water can ow freely into the channel (ideal case). The transport of the uid interface is given by a level set function. 36 This function is coupled with the Navier-Stokes equations to describe mass and momentum transport of the uid. The velocity of the uid is dependent on convection, pressure, diffusion, the surface tension and gravity in this model.
Initially, a small reservoir at the beginning of the microchannels is lled with water and the rest of the channels and the big reservoir are lled with air. The initial velocity is 0. A hydrostatic pressure is applied at the inlet of the channels (P ¼ r Â g Â y) and the pressure at the outlet is 0. Atmospheric pressure can be omitted since it is acting on both the inlet and the outlet. The gravity is added to the model as a volume force. The wetted wall feature is used to identify hydrophobic and hydrophilic walls. First, the inuences of the spacer tape and corner ow effects were simulated with a straight channel. Second, parts of the collection patch were simulated to test if the reservoirs are lled in the right way and to prevent entrapment of air.

(B) Syringe pump experiments
A syringe pump experiment is executed to test if the reservoirs will be lled one aer the other and to research if the sweat ow in the collector is the same as the actual sweat rate.
The syringe pump (KDScientic 200, USA) is set at a rate of 48 ml min À1 . Via the funnel-shaped structure, the sweat will drop down towards the inlet of the reservoirs.
The collector is designed to be placed on the back of an athlete. The back is chosen, because of the high sweat rate and the presence of eccrine sweat glands at this location. Smith and Havenith measured a sweat rate at the back of 1.2 mg cm À2 min À1 during a running exercise at around 155 bpm of 30 min (25.6 AE 0.4 C, 43.4 AE 7.6% relative humidity). 37 The patch will have a collection surface of 40 cm 2 , which means that the sweat rate will be around 48 ml min À1 .
Collector lling is recorded with a camera. Aer the experiment, stills are taken from the movie. Due to the blue colorant that was added to the uid, all blue pixels in the image can be counted to derive the volume that was lled with uid. This process was automated with a MATLAB program.

(C) Physiological experiments
The physiological experiments were approved by the Human Research Ethics Committee of Del University of Technology. The participants gave informed consent before the experiment. Healthy recreational athletes (n ¼ 5, 20-30 years) that play sports 2 to 5 times a week were asked to cycle for one hour on a cycle ergometer. The ergometer is equipped with a cadence meter, power meter (Garmin Vector 3s) and cycling computer, a Garmin Edge 820 (Garmin, USA), that is placed at the handlebar (Fig. 2a). The subject wears a heart rate monitor and sweat collectors are placed at the back. A camera is placed behind the cyclist and focused on the sweat collectors to measure the lling speed. Aer placement of the heart rate belt and the rst sweat patches, the subject is asked to cycle for one hour at a constant cadence and power output (in the nal experiment a cadence of: M ¼ 91, STD ¼ 5 rpm, heart rate: M ¼ 158, STD ¼ 11 bpm and power: M ¼ 284, STD ¼ 28 W were measured (Fig. S2, ESI †), the temperature of the room was 20 C). Two preliminary tests were performed before the nal experiment to improve the test setup. Preliminary results (Fig. S3, ESI †) show the importance of a strict and optimized test protocol. In the nal experiment, the skin was cleaned with a sterile gauze pad that was wetted with demineralized water. This was executed two times before each patch was placed to avoid accumulation of old sweat constituents.
In the rst two series of experiments the collectors were replaced at the same location, when a collector was lled completely. Unfortunately, a replaced collector does not start lling immediately (it can take approximately 10 minutes). To improve the continuity of the sweat measurements, the patches were placed as shown in Fig. 2b in the nal experiment. Patch B can now be placed 10 minutes before patch A is lled and more sweat can be collected. It is assumed that the sweat rates at locations le and right of the spine are identical.

(D) Chemical analysis
Aer sweat collection, transfer to vials is done by perforating the bottom tip of the reservoir on one side. A syringe with air is placed at the air inlet of the reservoir and the sweat is forced out through the perforated tip to the vial. Before each perforation of a reservoir, the outside of the collector is cleaned to prevent one sample contaminating the other (Fig. S4, ESI †). Approximately 5% of the sweat in each reservoir is lost during transfer. The sweat is diluted with 3 ml of ultra-pure water. The samples are stored in a fridge at 7 C. The day aer, the sweat is analysed using ion chromatography/high pressure liquid chromatography (IC/HPLC) 39 The IC system consists of a Metrohm 881 Anion system and an 883 Basic IC plus system (Metrohm, Switzerland). The two systems work independently. Both devices contain an HPLC pump that pumps the eluent through a column that separates the ions. Aer separation, the conductivity of the uid is measured by the conductivity detector.
To analyse the cations, a Metrosep C6 -150/4.0 column is used for separation with a solution of 3 mM HNO 3 as the eluent. The ow rate is set at 0.9 ml min À1 . For the anions, a Metrosep A supp 5 -150/4.0 column is used, and chemical suppression is performed by using the Metrohm Suppressor Module, which is regenerated with 150 mM H 3 PO 4 . The eluent of the anion system contains 1 mM NaHCO 3 and 3.2 mM Na 2 CO 3 . The ow rate in this system is 0.7 ml min À1 . A sample loop of 20 ml is used in both systems.
Before each analysis, standards need to be made for all the ions of interest. Solutions with 0.1 till 100 ppm are made. The measurements of the standards are used for calibration. Aer assigning manually the right retention times to the peaks, the MagIC Net soware, automatically creates a calibration graph based on the peak areas which can be applied to the measurements.
To quantify the accuracy of our method, standard deviation experiments are executed. Standard solutions with 0.1 to 100 ppm Na + , NH 4 + and K + ions and separate solutions with Cl À ions are prepared. The solutions are divided into 5 samples and measured one by one to obtain an idea of the standard deviation of the method at different concentration levels.

Results & discussion
The results of the simulations and lab experiments are presented rst. Thereaer, the results from the physiological experiments are discussed.

(A) Design & simulations
A rst simulation was executed to nd out the effect of the hydrophobicity of a spacer tape (sidewalls of the channels) on the volumetric ow rate in the channels. The dimensions of the simulated channel are 2 Â 0.25 Â 4 mm. Contact angles of q ¼ 112.5 and q ¼ 67.5 are chosen for the hydrophobic walls and the hydrophilic sidewalls, respectively.
To study the effect of corner ow, channels with rounded corners and the same dimensions as the previous channel are simulated as well. One channel has hydrophobic rounded sidewalls and the other has hydrophilic rounded sidewalls. Fig. 3a shows the inuence of the rounded and sharp, hydrophilic, and hydrophobic sidewalls on the contact point position of the uid-air interface over time. The volumetric ow rate decreases signicantly when the hydrophobic sidewalls have rounded corners. Since the laser cut spacer tape is placed on top of another layer of foil, the corners will be relatively sharp.
Simulations are also performed to test whether the Tjunction works suitably. In Fig. 3b it can be seen that the uid rst ows down and once the water reaches the bottom (t ¼ 5 ms), it goes le to the next reservoir.
Lastly, a simulation was performed to check if entrapment of air would be a problem in the reservoirs. Fig. 3c shows that the uid does not block the air outlet during the lling process, so that air bubbles can be released. The lling time of a reservoir with a height of 10 mm, is now very small. This is due to the innite supply of water at the inlet. The sweat rate will limit the uid supply in reality. However, the simulation shows that even with an innite supply of water, the reservoir is lled in the right way. This accounts for the T-junction as well.
(B) Syringe pump experiments Fig. 1 shows stills from one of the movies that is made during the syringe pump experiments, which prove that the reservoirs are lled consecutively. The measurement results of the experiment are shown in Fig. 3d. A delay of 50 to 80 seconds is observed before the uid is in the reservoirs and recorded by using a camera. The average volumetric ow rate in the collector is 47 ml min À1 , which means that the ow rate in the collector is 2% lower. From this we can conclude that the uid inow is not signicantly inhibited by the resistance of the channels.

(C) Physiological experiments & chemical analysis
Aer sweat collection during the physiological experiments, the sweat in the collectors is analysed using ion chromatography/ high pressure liquid chromatography (IC/HPLC). In the chromatographs of our sweat samples, the peaks are nicely separated and there are no signicant unknown peaks that interfere with the peaks of the selected ions (Fig. S5, ESI †).
To get an idea of the standard deviation that is introduced by the IC measurement method, standard deviations of the method are included. These are determined by analysing ve samples of each standard (0.1-100 ppm) and plotting the average relative standard deviations (RSDs) in a graph (Fig. 4a). The tting curve is used to determine the SDs in the actual measurements. For Cl À and Na + the SDs are relatively small, for K + , the concentrations in sweat are around ten times lower so the RSDs of the measurement method become important to consider. The measured concentration levels of Na + , Cl À and K + of patch location 2 are shown in Fig. 4b. The Na + and Cl À concentrations show an initial steady increase which levels off. However, Cl À levels are systematically lower by an average of 16.4 mM (SD ¼ 2.5 mM) than the Na + levels. This deviation can possibly be explained by the presence of other negative ions like lactate and bicarbonate (C 3 H 5 O 3 À and HCO 3 À ) in sweat.
In Fig. 4c, the Na + , Cl À and K + levels of patch location 1 (between the shoulder blades) and 3 (lower back) are added. The Na + and Cl À levels are very similar for collector 2 and 3 (Cl À : M (mean) ¼ 1.6, SD¼ 2.9, Na + : M ¼ 0.4, SD ¼ 3.4 mM) while the Na + and Cl À levels of collector location 1 are signicantly higher . One of the reasons for this difference between the absolute concentrations of collector 1 and collector 2 and 3 is the variation in the sweat rate across the different locations. Smith and Havenith measured at the sides of the upper back a median sweat rate of 0.84 mg cm À2 min À1 , while at the sides of the lower back this sweat rate was 0.75 mg cm À2 min À1 . 37 Higher sweat rates decrease ion reabsorption per unit volume and therefore higher concentrations can be measured at the upper back. Fig. 4d shows the K + levels at the three locations over time. A decrease is measured over time and no difference between the three locations can be detected within the error margin. To test whether the trend is still a decreasing line, when taking the large SDs into account, random samples and their linear ts (n ¼ 1000) were created, assuming a normal distribution.
Within the 95% condence interval, the ts for location 1 and 2 were always a decreasing line. However, from the literature it was assumed that only minimal changes in K + levels occur in sweat. 9 Therefore, it is expected that elevated values aer the start of the exercise are due to the presence of old sweat or residue in the channels or other physiological effects that need further research.
Because an increased sweat rate leads to less ion reabsorption in the reabsorptive duct, sweat rate measurements are important as well. The sweat collection rate is measured at location 2 during the experiments with the help of a camera. The average sweat rate is 1.19 mg cm À2 min À1 . The sweat collection rate varied between 0.74 and 1.54 mg cm À2 min À1 (Fig. 5a).
The sweat rate was plotted against the concentration of Na + and Cl À , to check whether this relation was visible. The data allows us in principle to verify a possible relation between the sweat rate and Na + and Cl À concentrations. Although the plot of Fig. 5b suggests a mild increase in ion concentrations with increasing sweat rate, the scatter in the current data is too large to draw a solid conclusion. In a next phase of this project, other sweat rate measurement techniques will be explored by, for example, conductive or capacitive measurements. It would also be interesting to compare this new method of sweat rate measurement with conventional methods like the ventilated capsule measurement and the absorbent patch method in the same physiological experiment. 40

Conclusion
The results of the nal physiological experiment show that Na + , Cl À and K + can be measured accurately with this new collection system in combination with ion chromatography. Although the Na + and Cl À concentrations differ when the collector is placed at the upper back, all three collector locations show a similar trend in Na + and Cl À levels. K + measurements are also reproducible, and a decreasing trend is measured.
In the current study, data from only one physiological experiment of a single individual are compared, because it is well known that sweat analyte concentrations can differ considerably between individuals and that the day-to-day variability in sweat analyte concentrations can be large. Therefore, comparing, and interpreting data between individuals in small scale experiments would be less relevant.
However, the presently developed device will facilitate more systematic larger studies on the inter-individual variability as well as the daily variations of sweat electrolytes in a single individual. In this way, the new device can contribute to a better understanding of sweat mechanisms and their relation to blood values, to nd useful biomarkers in sweat to monitor the status of an athlete.