Simultaneous multi-location wireless monitoring of sweat lactate trends

Wearable device technologies for sweat analytics present a versatile application for monitoring physiological state, which can circumvent the requirement for inconvenient and invasive blood sampling. This paper reports a miniature electrochemical sensor platform for non-invasive and wireless real-time monitoring of lactate in exercise-induced human sweat. The conformal and low profile sensor platform is composed of (a) a flexible electronic readout tag with wireless charging and data acquisition, and (b) a disposable enzymatic amperometric biosensor patch with electrodes fabricated using high throughput roll-to-roll processing. Data were generated in real time from sensor response to lactate in exercise-induced sweat from multiple body regions simultaneously. The biosensor demonstrates current response proportional to lactate at physiological concentration range between 5 and 30 mM. This developed platform can be adapted for sensing of other sweat constituents including ions or metabolites, and therefore advances wearable technology for personalized physiological monitoring


Introduction
Lactate represents a key cellular biomarker, closely related to tissue oxygenation [1]. Monitoring lactate concentration of biological fluids is useful for investigating the balance between lactate production and lactate secretion under critical conditions [2] (i.e. during surgery, during physical activity, under shock/ trauma). To this end, lactate biosensors have evoked great scientific interest towards clinical [3] and sports applications [4][5][6].
Blood lactate concentration provides information to athletes on important physiological conditions such as anaerobic work capacity [7,8]. Commercialized lactate sensors typically rely on finger-stick blood collection. However, the intrusiveness and inconvenience of such blood sampling methods, as well as unsuitability for continuous real-time screening, warrants alternative non-invasive bio-sensing technologies using, for example, wearable devices for analysis of lactate in sweat [9][10][11]. During sports, sweat is the bodily fluid in which online wearable electrochemical measurements are easiest to perform. Lactate concentration in human eccrine sweat has been shown to increase in response to increased exercise intensity [12,13]. Although lactate in sweat is a product of sweat gland metabolism, and the majority of reports state that lactate concentrations in sweat and blood are not directly correlated [2,14] there are some publications reporting a correlation, especially when the sweat is sampled over the working muscle [15,16]. This could be supported by the reports that the lactate concentration in sweat varies between different body parts [14]. Consequently, it is possible that changes in sweat lactate concentration at multiple body sites during exercise can provide information that is more valuable for athletes than from a single body location. Therefore, it is of interest to develop technology capable of monitoring sweat lactate simultaneously in multiple body locations.
Over recent years, extensive effort has been exerted on advancing wearable sensors for non-invasive monitoring of lactate in sweat [17]. One effective approach relies on amperometric measurement, following an enzyme-catalyzed reaction. Sweat lactate sensors reliant on this mechanism have been demonstrated in various forms. In addition to conformal skin patches [5,6,18,19] including tattoos [4] or adhesive tapes [20], other form factors have been explored, such as eyeglasses [21], textile [22], and touch pad [23]. However, all of these studies investigate a single body region per test. In addition, none of these examples report electrode fabrication by high-throughput roll-to-roll (R2R) process (table S1 (available online at stacks.iop.org/FPE/6/ 034003/mmedia)), which is essential for large volume fabrication of single use sensor components.
In this work, we present a wearable, wireless, non-invasive sensor platform for monitoring general trends of lactate in exercise-induced sweat. The sensor consists of electrodes printed by R2R process, which are enzymatically functionalized based on earlier reports to form active sensing components and integrated into a disposable patch. We demonstrate multiple sensors simultaneously operating in different body parts during cycling-ergometer exercises. The sensors provide current responses that are proportional to lactate concentrations between 5 mM and 30 mM, the range typically found in exerciseinduced eccrine sweat. During the exercise, we also collect blood and sweat for offline analysis. To the best of our knowledge, this is the first report of wireless sweat lactate monitoring from multiple body locations simultaneously, and demonstrating this using electrodes formed by R2R process.

High-throughput R2R manufactured electrodes
Electrode structures were fabricated on polyethylene terephthalate (PET) (125 µm, Melinex ST506, Foiltech) substrate that was thermally treated (140 • C oven, 2 min) to reduce dimensional changes of the substrate during printing processes. The electrode material stack was formed by sequential ink layer deposition and curing using high-throughput R2R rotary screen-printing process at 4 m min −1 for (a) conductive silver contact tracks and reference electrode (11 µm, Asahi LS-411AW, 140 • C oven, 1 min), (b) graphite working and counter electrode (DuPont BQ242, 120 • C oven, 1 min), (c) Prussian blue (PB) impregnated carbon working electrode overlayer (Gwent C2070424P2, 120 • C oven, 1 min), and (d) transparent insulator (DuPont 5036, 55% UV) to prevent short circuit of electrodes if the region outside of the electrochemical sensing cell is exposed to moisture. Patterns for fabrication of the printed electrode stack and an array of printed electrodes are depicted in figures 1(a) and (b), respectively.

Functionalization of electrode surface
Screen-printed electrode structures were functionalized for amperometric determination of lactate. Two types of electrodes were used. (a) Commercial carbon-based screen-printed electrodes for generating data in figure S1(a) (Ref. 710, DropSens), and (b) in-house fabricated R2R screen-printed electrodes on PET. PB mediated carbon working electrode was modified by additional PB film electrodeposited by cyclic voltammetry (CV) from −0.5 V to 0.6 V (vs Ag/AgCl or Ag) for 10 cycles at 50 mV s −1 in a fresh solution containing 2.5 mM FeCl 3 , 100 mM KCl, 2.5 mM K 3 Fe(CN) 6 , and 100 mM HCl. Electrodes were then rinsed with purified water and dried under nitrogen. Lactate oxidase (LOx, LCO-301, >80 U mg −1 , Toyobo)/chitosan (Sigma-Aldrich, low molecular weight) matrix was prepared according to earlier reports [6,18]. Six microliters of this mixture was drop cast onto the working electrode and dried at 4 • C overnight.

Disposable structure for on-body sweat monitoring
Single use electrochemical sensing patches were fabricated from a stack of components, to be attached directly to skin, with a footprint of 4 × 2 cm. Figure 1(c) illustrates how the patch is composed of (a) absorption pad (2 × 1.5 cm, Whatman CF6 cotton linter, 56 s/4 cm capillary flow rate, 136 mg cm −2 water absorption capacity, GE Healthcare) for disposal of analyzed sweat. (b) Functionalized electrode layer. (c) Hydrophilic membrane (Whatman 1 chromatography paper with pore radius 5.4 µm (GE Healthcare) to prevent direct skin contact with the enzyme sensing matrix and to transport fluid away from the sensing region through an exit port. (d) Adhesive tape (3M 1577 polyester double sided medical tape) for attachment to skin and defined sweat collection area of 30 mm 2 . Sweat sensing patch constituent materials were cut using laser, with manual patch assembly assisted using a customized jig.

Wireless data acquisition
A wireless wearable potentiostat, FlexPot, was developed as shown in the block diagram depicted in figure S2. The device is assembled on a flexible polyimide printed circuit board (PCB) and comprises a Texas Instruments LMP91000 potentiostat analogue frontend, an anti-aliasing filter, nRF51 Bluetooth low energy (BLE) radio micro controller unit with an analogue-to-digital converter, a flash memory, a rechargeable battery, wireless charging circuitry and a zero-insertion-force connector for a disposable electrochemical sensor. Table S2 lists the most important performance characteristics of FlexPot. A photograph of the potentiostat with a disposable electrochemical sensor can be seen in figure 1(d). (a) Electrode patterns for sequential roll-to-roll rotary screen printing of (i) silver conductive tracks, reference electrode and sensor outline with alignments for custom assembly jig, (ii) carbon working and counter electrode, (iii) Prussian blue mediated carbon working electrode, (iv) transparent insulator. (b) An array of electrodes printed on flexible and transparent PET substrate. (c) Exploded illustration of the conformal and disposable electrochemical patch for monitoring of lactate in sweat, composed of (i) absorption pad, (ii) functionalized electrode layer, (iii) hydrophilic membrane, (iv) adhesive, (v) skin. Hydrophilic membrane fills the electrochemical cell, with the narrow strip inserted out from the electrode layer exit port, to facilitate direct contact with the outer absorption pad. (d) FlexPot wireless potentiostat with a disposable lactate sensor patch omitting the absorption pad to enable viewing of underlying components.
A mobile phone App was developed to control the potentiostat and to receive, store, transfer and visualize measurement results. The Android device communicates with the potentiostat via BLE. The App sets the measurement parameters, starts and stops measurements, and receives streamed measurement results. A graph of the potentiostat current can be viewed in real time. The measurement results can also be stored on the Android device and transferred to cloud or sent via email.

Measurement of exercise-induced sweat
All on-body experiments were performed in compliance with the protocol approved by VTT research ethics committee. Stationary cycling was employed as the form of exercise, with heart rate (bpm), power (W), and cadence (rpm) ascertained using a wrist monitor (Fenix 3R, Garmin) coupled with power meter foot pedals (Vector 3S, Garmin). Exercise protocol was divided into regions according to workload, including (a) warm up (8 min, ∼70 W), (b) increasing load (3 min intervals from 150 to 350 W), and (c) cool down (1 min rest followed by 9 min at ∼70 W).
Skin was cleaned with 10% ethanol in water prior to applying electrochemical sensors for monitoring of lactate in sweat. Sensors were attached directly to the skin using adhesive tape in mirrored locations on both sides of the body at sites of inner forearm and outer thigh. As shown in figure S3, FlexPot readout tags were connected to each sensor using a linking cable and encapsulated with plastic film to prevent moisture damage from direct contact with perspiration. Each sensor and FlexPot were then further secured in position using additional adhesive to retain intimate skin connection and electrical contact between sensor and read out tag during exercise-induced motion. Generated data was observed and collected in real time, then transferred wirelessly via mobile phone application for later interpretation.

Reference methods
Complementary analysis methods were used in addition to the enzymatic electrochemical sensors developed during this study. Skin was cleaned with ethanol/water solution prior to and between each collection of blood or sweat.
Sweat was collected from skin close to the wearable sensor on the right forearm, using plastic pipette, then fluid transferred into autoclaved Eppendorf tubes. For offline analysis of collected exercise induced sweat, lactate in sweat was oxidized by lactate oxidase, with hydrogen peroxide evolved in the reaction determined by a colorimetric hydrogen peroxide assay (Quantitative Peroxide Assay kit, Thermo Scientific). Hydrogen peroxide concentration was determined in a spectrophotometric assay by measuring absorbance at 595 nm using a plate reader [19].
Blood was withdrawn from the fingertip by spring-loaded lancet, with the resulting droplet immediately analyzed using a portable commercial lactate monitor (Lactate Scout+, EKF Diagnostics).

Biosensor development and optimization
When LOx is in the presence of oxygen, it has the ability to catalyze lactic acid into hydrogen peroxide (H 2 O 2 ) and pyruvic acid. This generated H 2 O 2 is identified electrochemically by way of reduction to hydroxide ion (OH − ), where the reduction is driven by release of electrons from a working electrode and these electrons may be quantified as current signal. A limitation of this approach is that a potential of around +0.65 V (vs Ag/AgCl reference electrode) is required for amperometric H 2 O 2 detection using a carbon working electrode surface [24]. This operation voltage results in oxidation of other sweat constituents such as ascorbic acid [25] and uric acid [26], which could interfere with the signal and reduce the sensor selectivity for lactate. In addition, for miniaturized devices, power consumption requires consideration in order to minimize battery size, with an operation potential close to zero V being preferable. PB can be employed as an electron redox mediator for electrochemical reduction of H 2 O 2 , assisting electron transfer and enabling lower operation voltage of the sensor [5,6]. Figure S4 depicts the operation mechanism of a sensor for LOx catalyzed oxidation of lactic acid, assisted by PB. Note that figure S4 depicts L-lactic acid as the analyte. This is because this enantiomer dominates in abundance over D-lactic acid in the human body. Under physiological conditions, L-lactic acid dissociates by ionization of a carboxyl proton to form the lactate ion. Therefore, throughout this report we refer to lactate as the analyte of interest, rather than L-lactic acid.
CV was used to identify a suitable reduction potential to be applied during chronoamperometric determination of lactate. A sensor was prepared using a working electrode formed from ink containing PB mediator. CV at 50 mV s −1 in 2.5 mM lactate revealed a clear reduction peak signal at approximately −0.1 V as presented in figure S1(a). This demonstrates the ability of PB to shift the sensor operation potential close to zero volts, conserving required power input and alleviating interference from oxidation of interferents at high cell potential. The identified reduction potential of −0.1 V was used for subsequent electrochemical determination of lactate by chronoamperometry.
Lactate concentration of exercise-induced sweat is around 5-30 mM [13,27,28]. A sensor developed for this application requires a corresponding operation range, with large concentrations of lactate demanding sufficient quantities of both active LOx enzyme and PB mediator. We observed that an electrodeposition process of ten CV cycles shown in figure S1(b), displays increasing redox activity during early deposition cycles, and then stabilization of redox currents by the tenth cycle as the working electrode surface is covered by PB film, indicated by marginal difference between reductive peak minima in cycles nine and ten.
Electrochemical deposition of PB is time and resource expensive when considering processing steps towards mass manufacture such as R2R fabrication environments. It would therefore be a benefit to deposit the PB film by alternative means such as printing. To investigate this, we compare the operation of a sensor with rotary screen-printed commercial PB mediated carbon ink, to a sensor with additional PB grown by CV as described above. The results presented in figure 2(a) show that the sensor without electrodeposited PB does not operate at lactate concentrations above 15 mM, whereas the sensor possessing CV grown PB operates up to lactate concentration of 30 mM. It is possible that the electrodeposited PB film surface is less smooth than the PB mediated carbon ink. A rougher PB surface could assist attachment of the chitosan matrix to the working electrode surface. We suspect that in figure 2(a), the chitosan matrix becomes delaminated from the printed PB mediated carbon surface, resulting in the observed lack of sensor operation above 15 mM lactate. This motivates the incorporation of an electrodeposited layer of PB, and as such, this was included as a step in the sensor fabrication process.
Provided with a target operation range of up to 30 mM lactate [13,27,28], current response to standard solutions of lactate in PBS was tested using sensors produced during this work. Figure 2(b) shows that current response is proportional to lactate concentration from 5 to 30 mM, (data recorded from n = 7 sensors at 60 s for each concentration, sensitivity of 0.027 ± 0.002 µA mM −1 , R 2 = 0.99). Individual sensor operation and repeatability were found to be good, as shown in figure S5. The inter-sensor variability (error bars in figure 2(b)) is tolerable for monitoring general trends in changing lactate concentration and may be attributed to inconsistencies in the manual drop-casting of enzymatic sensing matrix [29].
A study was performed to ascertain the extent of interference on the biosensor chronoamperometric operation from a range of species other than lactate that are commonly present in sweat.  figure 2(c) shows that the measured current signal from tested interferents does not deviate substantially from that of PBS background. Addition of 5 mM lactate does cause measured current to deviate substantially from the background. This demonstrates the selectivity of the sensor towards the lactate target substrate.
Beyond the above-described proof of operation testing, extensive tests with model solutions were not performed, because it is known that enzymatic sensors behave differently in a real sweat matrix compared to that of model solutions such as buffers or synthetic sweat [29], for example due to pH, a complex mixture of components, and physiological salts affecting enzyme activity [30]. Matrix effects from sweat, including fluctuating or differing composition, complicates the generation of valid calibration data. For example, generation of calibration data by using lactate spiked sweat from a given individual human subject may not be reliable for analysis of sweat from another individual, or possibly even the same individual due to differing sweat composition between subjects and in the same subject over time. Therefore, having demonstrated proof of operation in model lactate solutions, our next step was to apply the sensor for monitoring general trends of changes in lactate from exercise-induced sweat, without attempting to convert acquired measurement signal into absolute lactate concentration.

Analysis of lactate in exercise-induced sweat
On-body evaluation of the developed lactate sensor was performed by continuous, real-time measurement during stationary cycling. The protocol was performed twice by the same test subject. On both occasions, one lactate sensor was attached to each outer thigh and inner forearm of the test subject, enabling wireless recording of current response to lactate in sweat simultaneously from four bodily locations, yielding the profiles in figure 3. In all sensors, current remains close to zero until the onset of perspiration at around 15 min into the exercise. At this point, eccrine sweat is released from the sweat glands with sufficient volume to wet the working electrochemical cell where lactate is selectively detected, providing measureable current signal. Figure 3(a) presents data where for sensors situated on the right side of the body the measured signal increases with increasing workload over the course of around 6 min, before gradually returning towards the baseline. Sensors on the left side of the body exhibit the same initial trend of increase and decrease, however a second increased negative signal event occurs from around 25 min that is not seen in sensors on the right side of the body. This second signal event continues until the cool down period, where the signal approaches the baseline. Results from a repeat experiment ( figure 3(b)) do not as clearly present the same asymmetry in measurements from left versus right side of the body. Here, sensors on the left thigh and both arms present an initial signal increase, followed by signal fluctuation, before decreasing towards the baseline later in the experiment. The absolute current generated from the sensor on the right thigh is approximately twice that recorded from the sensor in  the same location during the first experiment. However, both of these sensors present the same general current trend. The differences in recorded signal across all sensors might be due to (a) differing lactate concentrations in sweat from different bodily locations and from different experimental occasions, or (b) sensitivity variations between sensors, for instance due to different sweat pH or temperature [5] in different bodily regions and experimental locations. Nevertheless, the general trend across all sensors is the largest current right after the onset of sweating, with a decline of signal during latter phases of the exercise. This is in agreement with reported measurements made using a similar biosensor [6], where decrease in signal during latter phases of exercise was attributed to the dilution effect of increased sweat rate burying the lactate excretion rate [13] that is proportional to the product of the sensor signal and the sweat rate.
In addition to the real-time monitoring of lactate in sweat, samples of sweat were collected from the right forearm during both experiments for offline analysis. To alleviate contamination from earlier time periods, the collection region and surrounding area were cleaned with ethanol/water mixture immediately following each collection. Figure 4 displays how for collected sweat from both experiments at the onset of sweating, the lactate concentration is elevated, gradually decreasing over the remainder of the exercise, with the majority of samples within the expected range of 5-30 mM [13,27,28]. However, the initial collected sweat sample in figure 4(a) is around 40 mM. It is plausible that this was caused by evaporation of sweat from the skin surface during the first 17 min of exercise, resulting in accumulation of lactate on the skin surface. The observed trend of initial high lactate concentration, followed by constant decrease, is in agreement with that recorded by the wearable sensors located on the right forearm close to the vicinity of the skin region where sweat was collected for analysis (figure 4 plots signal from wearable sensors as positive current to assist visual comparison with other data). Figure 4 provides evidence that the wearable sensors operate as intended to provide a current signal that is proportional to the concentration of lactate in sweat. However, we propose that varying sweat rate means that the obtained data does not reveal accurate information on lactate excretion over the duration of the exercise. Therefore, a method is required for real-time monitoring of localized changes in sweat rate to enable determination of lactate excretion rate. This will be explored in future studies.
Lactate concentration in capillary blood from the fingertip of the right hand was analyzed for comparison with online and offline sweat from the right forearm. Data plotted in figure 4(a) exhibits an inflection point around 2 min after the onset of sweating. It is possible that this event indicates shift from aerobic to anaerobic condition [8]. From this point, the lactate concentration in blood rises from 2.2 mM gradually up to 10.6 mM by the end of the high intensity exercise period, then decreases to 6.7 mM during the 10 min cool down period, where the power of exercise was 20% of the maximum load. The temporal profile of blood lactate reflected the increase of gradient loading during exercise. It is of note that determination of lactate threshold unequivocally from the blood lactate data presented here can be difficult, mainly because of a limited number of samples during the course of the exercise, and variation arising from fingerpick blood collection. Figure 4(a) compares blood lactate to online (sensor signal) and offline (collected) sweat lactate from the right forearm. Correlations of the sensor signals with the blood lactate are shown in figure S6. The left side sensor signals had a greater correlation with the blood lactate. For the left side, we also saw a larger correlation for the working muscle as compared to the latent muscle as reported in [16]. It is known that the correlation between blood and sweat lactate can be overpowered by lactate also being a byproduct of sweat gland metabolism [14]. However, at least one article [27] that was used to draw this conclusion relies on data recorded below the lactate threshold (below anaerobic condition, where blood lactate level experiences minimal exercise related increase). There have been reported examples suggesting a correlation between lactate concentrations in sweat and blood [15,16,31]. Identification of any correlation requires distinguishing absolute lactate concentration of sweat, from that of lactate excretion independent of sweat rate. This requires a reliable method for real-time quantification of localized sweat rate. In the absence of an accurate method for quantification of sweat rate, we did not analyze lactate in blood during the second experiment ( figure 4(b)).
We discovered that in order to extract useful information from data generated on lactate in sweat, it is important to differentiate absolute concentration of lactate in sweat from that of lactate excretion rate. It is known that decreased concentration of lactate in sweat at high exercise intensity is likely a result of increased sweat production rate, with increased sweat rate diluting the lactate and challenging the ability to gauge the extent of lactate excretion [13]. This motivates development of methods for quantification of sweat rate. A promising approach for realizing this is by using admittance-based monitoring of sweat advancing into a micro-capillary [32]. In addition, it would be highly beneficial to incorporate a system for quantifying skin moisture evaporation [33], for insight into whether this affects perceived sweat rate.

Conclusions
This work demonstrates a miniature wearable electrochemical sensor for non-invasive and wireless monitoring of trends in lactate levels encountered in exercise-induced human eccrine sweat. The developed sensing platform demonstrated the determination of absolute lactate concentration, however in the absence of sweat rate quantification, only relative changes in lactate level were possible to determine. Lactate in sweat was simultaneously monitored on multiple bodily locations, with data transmitted wirelessly for visualization and collection by Android phone app. This platform can be adapted to electrochemical monitoring of other biomarkers in sweat. Further work will focus on reduction of device size to enhance wearability, and the development of an effective system for real-time monitoring of sweat rate, to enable determination of sweat biomarker excretion rate, irrespective of fluctuations in sweat rate.

Data availability statement
The data generated and/or analyzed during the current study are not publicly available for legal/ethical reasons but are available from the corresponding author on reasonable request.