In-ear integrated sensor array for the continuous monitoring of brain activity and of lactate in sweat

Owing to the proximity of the ear canal to the central nervous system, in-ear electrophysiological systems can be used to unobtrusively monitor brain states. Here, by taking advantage of the ear’s exocrine sweat glands, we describe an in-ear integrated array of electrochemical and electrophysiological sensors placed on a flexible substrate surrounding a user-generic earphone for the simultaneous monitoring of lactate concentration and brain states via electroencephalography, electrooculography and electrodermal activity. In volunteers performing an acute bout of exercise, the device detected elevated lactate levels in sweat concurrently with the modulation of brain activity across all electroencephalography frequency bands. Simultaneous and continuous unobtrusive in-ear monitoring of metabolic biomarkers and brain electrophysiology may allow for the discovery of dynamic and synergetic interactions between brain and body biomarkers in real-world settings for long-term health monitoring or for the detection or monitoring of neurodegenerative diseases.


Preparations of the materials Styrene-ethylene-butylene-styrene block copolymer (SEBS):
The SEBS resin was prepared by dispersing SEBS beads into toluene (weight ratio of 1: 2).The mixture was mixed with a linear shaker (Scilogex, SK-L180-E) overnight or until the mixture became transparent and homogeneous.The cast resin was first dried in an ambient environment for 1 hr, then cured in a conventional oven at 80 °C for 1 hr to remove the excess solvent.
Stretchable Silver: The stretchable Ag ink was synthesized by mixing SEBS (4 g to 10 mL toluene ratio) and Ag flakes in a weight ratio of 1:2, in a dual asymmetric centrifugal mixer (Flacktek Speedmixer, DAC 150.1 KV-K) with a speed of 1800 RPM for 10 minutes or until obtaining a homogeneous ink.Stretchable Prussian-blue (PB): The stretchable PB ink was synthesized by mixing super-P carbon black, graphite powder, PB, SEBS (4 g to 10 mL toluene ratio), and toluene, in a weight ratio of 0.5: 3: 1: 1.26: 4.74, in the mixer at 2150 RPM for 10 minutes or until the ink was homogeneous.Before printing the stretchable inks, the ink viscosity was analyzed visually, and if necessary, (~200 µL) toluene was added, and the ink was centrifuged before use.Phosphate-buffered saline (PBS): PBS used in the electrode modification was prepared in 0.1 M with a pH of 7.4.All bovine serum albumin BSA solutions were prepared with a concentration of 10 mg/mL in PBS.
Chitosan: The chitosan solution was prepared by dissolving chitosan in 0.1 M acetic acid with concentration of 0.5 wt%.

Printing procedures
The fabrication of the in-ear sensors was based on a cost-effective layer-by-layer screen printing method (Extended Data Fig. 2a-f).The screen-printing was carried out using a semi-automatic MPM-SPM printer (Speedline Technologies).Custom stainless-steel stencils were designed (AutoCAD, Autodesk) and fabricated (Metal Etch Services) with the dimensions 10 inches × 10 inches and a thickness of 150 μm.The electrodes were printed layer-by-layer, as illustrated in Extended Data Fig. 2m-p.Specifically, the following printer parameters were used for each layer: stretchable Ag layer: force 40, speed 4, SEBS layer: force 20, speed 2, PB layer: force 20, speed 3. Before printing the PB layer, a layer of universal release agent (Universal mold release, Smooth-on Inc.) was sprayed onto the back of the stencil to avoid adhesion between the stencil and the SEBS.After printing, a double-sided adhesive (Medical Tape 1509, 3M Medical) was affixed to the backside of the sensor for later assembly.

Three-dimensional electrophysiological electrode simulation and fabrication
A finite element model was built to illustrate the mechanism of the 3D electrophysiological electrode formation.In the FEM, one electrophysiological electrode was modeled using the actual size of the terephthalate polyurethane (TPU) substrate and the stretchable Ag.The materials in the FEM used literaturereported parameters.For TPU: Thermal conductivity: 0.2 W/(m•K), Density: 1050 kg/m 3 , Young's modulus: 0.5 GPa, Poisson's ratio: 0.42, heat capacity at constant pressure: 1800 J/(kg*K), Coefficient of thermal expansion: 1.7 -4 1/K 51,52 .For stretchable Ag: as an approximation, the stretchable Ag used the parameters characterized from a solution-derived nanoparticle Ag ink thin film 53,54 .Thermal conductivity: 429 W/(m•K), Density: 10500 kg/m 3 , Young's modulus: 110 GPa, Poisson's ratio: 0.37, heat capacity at constant pressure: 235 J/(kg*K), Coefficient of thermal expansion: 1.91 -4 1/K 51 .Both materials were modeled as linear elastic materials with fixed constraints along the TPU's edges.A boundary temperature condition of 373.15K (100°C) was applied to all the exposed surfaces of the model.The FEM sectional displacement results after applying the heat condition (Extended Data Fig. 2s) show a deflection of 53% of the TPU thickness (80 μm/150 μm TPU thickness).
The stretchable Ag 3D electrodes were molded using a 3D printed 750-μm-thick tough polylactic acid (PLA) mold (S5, Ultimaker) and stretchable Ag ink (Extended Data Fig. 2t-v).The procedures were as follows: sprayed the back of the mold with the universal release agent (Universal mold release, Smooth-on Inc.), aligned the mold over the in-ear sensors, filled each mold hole with stretchable Ag ink, scratched the excessive ink with a blazer to make the surface flat, lifted the mold as soon as possible to avoid adhesion to the mold, and cured in the oven at 100 ºC for 10 mins.The following steps created the curvature structure or "holes" under each electrophysiological electrode.These "holes" were then filled with SEBS and cured in the oven at 100 ºC for 10 minutes.

Bonding procedures
After manufacturing the in-ear sensors, the sensors and the 3M medical tape at its bottom were carved along its designed profile (Extended Data Fig. 2g).A standard polyimide substrate two-layer flexible printed circuit board (PCB) (designed with Altium Designer, Altium Inc., Australia, and fabricated by PCBWay, China) was aligned, attached, and soldered to the silver traces using silver liquid solder aided by a Kapton tape transfer layer (Extended Data Fig. 2h-j).The flexible PCB has a dedicated adhesive area at the bottom side for attaching to the bonding area of the sensors (Extended Data Fig. 2y).After bonding the sensors and the flexible PCB, the adhesive liners of the 3M double-sided adhesive at the bottom of the sensor were released, and the sensors were attached onto the earphone's target area (Extended Data Fig. 2k) and secured using the 3M medical double-sided adhesive as well as a silicone earphone hook.

Electrochemical lactate electrode modification and working mechanism
For lactate electrode modification, the chitosan solution was mixed with a lactate oxidase (LOx) (40 mg/mL) in BSA solution, in a volume ratio of 1: 1, followed by drop casting a 2 µL aliquot of the mixture onto the working electrode surface modification (Extended Data Fig. 2l).After drying at room temperature for 1 hr, 2 µL of the chitosan solution was drop cast to the previously modified surface.Upon completing the corresponding modification steps, the resulting sensors were stored at 4°C overnight before use.The lactate sensing mechanism is based on the reaction of LOx on the working electrode surface with lactate, resulting in the creation of hydrogen peroxide and pyruvate.Further, the PB-based electrode transducer transforms the hydrogen peroxide product to hydroxyl ions (OH -) for selective lactate detection (Extended Data Fig. 3a,b).

Polyvinyl alcohol-hydrogel preparation procedures
The fabrication of the porous Polyvinyl alcohol (PVA) hydrogel was based on the protocol from a previous report 55 .Solutions of PVA and potassium hydroxide (KOH) were prepared in deionized (DI) water in a weight ratio of 1:10 and 1:30, respectively.Next, 10 g of the PVA solution, 14 g of the KOH solution, and 2 mL of DI water containing 2.6 g of table sugar were mixed in a beaker under mild stirring.Afterward, 15 g of the solution was placed into a petri dish (diameter ∼9 cm) and left inside a vacuum desiccator overnight to remove the excess water and allow the mixture to crosslink.The cross-linked hydrogel material (thickness: 0.5 mm) was then immersed inside a 0.1 M PBS solution to remove the sugar and excess KOH (until the gel reached a pH of 7).The gel was then cut into 1 × 1 cm 2 pieces and stored in a bottle containing 0.1 M PBS for its storage.

Sensors' chemical component analysis
Component analysis was conducted to verify the safety of the sensors to be used in the ear.For structural materials that will be in contact with the skin, the electrophysiological electrodes were made with Ag composite materials which has been widely used and considered safe.For the electrochemical sensors, PVA gel was designed to be in contact with the skin.Apart from these, SEBS was used as the insulation layer which can also touch the skin.
The PVA hydrogel materials have been assessed for the KOH contents.The KOH content of the alkaline solution from the hydrogel membranes was removed by performing repetitive washing steps with phosphate buffer (pH 7.0).During every washing cycle, the pH of the solution containing the hydrogel membranes was recorded using commercial pH strips.These washing steps were optimized by examining the KOH content using FTIR (Fourier-transform infrared) spectroscopic methods.Supplementary Fig. 5a shows the disappearance of the peak at ~670 cm -1 and confirms the absence of the K-O bond, which corresponds to the absence of the KOH from the hydrogel.In addition, the PVA hydrogel has also been assessed by using the pH measurement at its surface every time before the experimentation, where the neutral pH ensures also corroborates the removal of the KOH irritant from the PVA-hydrogel.
For the SEBS, during the patch fabrication process, the material layers are subsequently heat cured at 80°C and 60°C upon the deposition of each layer.For developing the sensor patch, a total of 30 minutes (10+10+10 minutes) of heat treatment has been given, where for the SEBS layer gets 20 minutes of heat exposure (at 80°C) in the oven, which considerably evaporates the toluene from the patch.In order to verify, we have performed the material characterization using thermogravimetric analysis (TGA) (Supplementary Fig. 5b), where the SEBS film has been characterized after the curing.TGA curve demonstrates the mass loss at different temperatures.The consistent 100% mass retention till the temperature of 250°C indicates the SEBS film has no toluene traces which would have otherwise been seen with mass loss at its boiling point (~110°C).The mass loss starting from ~250 °C is attributed to the heat degradation of SEBS.

Mechanical robustness
Since the variation of ear geometries between human participants is notable, different mechanical deformations such as bending and twisting may be applied to the sensors before, during, and after insertion.Therefore, the mechanical robustness and stability of the in-ear integrated sensors was evaluated to ensure functionality over an extended period.To improve the flexibility of the sensor, all the materials used during the screen-printing process were chosen to have stretchability.To characterize the sensors' resilience to possible insertion deformation, the sensors were tested with a repetitive dual-direction uni-axial strain of 20%.The mechanical stretching was performed in a controlled setting with a programmable motorized linear stage (X-LRQ, Zaber Technologies Inc.).The stage clamped both edges of the in-ear integrated sensors (Fig. 1g) and stretched with controlled strain and speed for a pre-programmed distance.One stretching cycle consisted of stretching in both the longitudinal direction and the latitudinal direction sequentially.The stretching distance was 120% the length of the direction being stretched.The stretching speed was set at 3 mm/s for both the stretching and releasing stages.Additionally, the resiliency of the sensors against insertion was visually inspected by inserting it into and retracting it out of a silicone ear phantom repetitively (Fig. 1j).
To characterize the pre-stretching and post-stretching material structure and morphology, scanning electron microscope (SEM) images (FEI Quanta 250) were taken before and after the 100 cycles of mechanical stretching of the same in-ear integrated sensors.To characterize the electrodes' connectivity before and after stretching, pre-stretching and post-stretching impedance measurements were performed with the potentiostat in a three-electrode setup (Supplementary Fig. 6i).Here the working electrode (WE) was the electrophysiological or electrochemical electrode to be measured using the impedance spectroscopy method, with an Eac = 10 mV and a frequency range of 1-100 kHz.The reference electrode (RE) and the counter electrode (CE) used an Ag/AgCl reference electrode (Gamry Instruments) and a platinum wire, respectively.To characterize the lactate sensing electrodes' functionality after stretching, pre-stretching and post-stretching in-vitro lactate sensing measurements were taken on each cycling by running chronoamperometry (CA) steps of 60 s under a fixed concentration of 2 mM lactate.
Supplementary Fig. 6a-h shows the pre-and post-stretching SEM microscopic structure of the electrodes on the sensors.By design, there were two types of electrodes classified by the outermost layer materials.The Ag-type electrodes included the electrophysiological electrodes and the electrochemical reference electrode, which were made with stretchable Ag.The PB-type electrodes included the electrochemical working electrode and counter electrode, which had a surface coating of PB.The SEM showed no apparent structural damage or cracks for both the Ag-type and PB-type electrodes after 100 stretching cycles.
Supplementary Fig. 6j,k shows pre-and post-stretching Ag-type and Pb-type electrode impedance measured in the 1 mol/L PBS.After stretching, the Ag-type electrode maintained a low (<2 kΩ @ 10 Hz) impedance and demonstrated a less capacitive behavior, indicated by a reduced phase angle at low frequency.The PB-type electrode retained its behavior at low frequency (<100 Hz) and demonstrated a less capacitive behavior at high frequency.In both cases, stretching the electrode structure resulted in a more porous structure, which promoted more direct charge transfers across the surface and a smaller double capacitance.
Extended Data Fig. 8b shows the pre-and post-stretching in-vitro lactate sensing performance.By comparing the chronoamperometric intensity of each stretching cycle, minimal current changes (<8%) were observed as the stretching cycles progressed.Such negligible differences demonstrate the efficient combination of stretchable components integrated into the costume inks during fabrication.

Robustness under different temperature and humidity conditions
To demonstrate the robustness and stability of the in-ear integrated sensors under varying temperature and humidity conditions inside the ear, we characterized the sensors under an expanded testing condition in comparison to the reported normal temperature (36.4 ± 0.6 °C40 ) and humidity (relative humidity: 40% -70% 41 ) profiles inside the ear.For characterization of stability, the conductivity of the electrophysiological electrodes was measured using a three-electrode electrochemical impedance spectroscopy (EIS) setup in which a droplet of 1 M PBS was placed on top of one electrophysiological electrode; the EIS setup was the same to that described in the section on mechanical robustness.The sensitivity of the electrochemical lactate electrodes was evaluated using the chronoamperometry method described in the section on the electrochemical lactate sensing method, with different lactate concentrations added to the electrochemical electrode surface (0 mM as a baseline, 5 mM, 10 mM, 15 mM, 20 mM, and 25 mM).Supplementary Fig. 7a shows the setup of the temperature testing.The sensors were positioned atop a hotplate.Using an infrared thermometer, the temperature of the electrode-electrolyte interface was measured for the electrophysiological electrode conductivity.The following temperatures were measured: 25 °C, 30 °C, 32.5 °C, 35 °C, 37.5 °C, and 40 °C.Supplementary Fig. 7b shows the setup of the humidity testing.The sensors were placed inside a closed acrylic chamber (12 × 12 × 12 inch), along with a mini humidifier inside a beaker of water, a digital temperature and humidity monitor, and fixation helper hands to hold the counter and reference electrodes.Four humidity conditions were measured, including 40%, 50%, 60% and 70% (relative humidity).
Supplementary Fig. 7c,d show the conductivity of the electrophysiological electrodes at various temperatures.Both the impedance and phase of the electrode-PBS interface reveal no significant changes between 25 °C and 40 °C, which corresponds to the normal range of ear temperature.Due to the fact that we were assessing the electrophysiological electrode interface with PBS covering the electrode, the humidity test configuration did not apply to the electrophysiological electrodes.Nonetheless, the stability of the electrophysiological electrodes under varying humidity levels inside the ear can be confirmed by the valid electroencephalogram (EEG) recording following exercise and perspiration production.
Supplementary Fig. 7e-n show the sensors' electrochemical stability characterization results under variable temperature and humidity.Simultaneously with the electrophysiological sensor, the robustness of electrochemical sensors were evaluated for the stability at variable temperature and humidity by performing the sensitivity assessment.Significantly similar sensitivities of the sensors were observed at different temperatures which is attributed by the intact activity of the LOx enzyme and the stable sensor development.Also, the sensitivities at different humidity levels are fairly consistent, which is most likely due to the wet PVAgel on the top of it which resists the impact of fluctuation of the humidity by marinating the sensing surface wet (100% humidity).Supplementary Fig. 7e-j show the dose dependent response of the electrochemical sensor under different temperatures.Supplementary Fig. 7k-n shows the dose dependent response of the electrochemical sensor under different humidity levels.
The adhesive at the bottom of the sensors stayed firm, showing no sign of detachment.

Electrode-ear impedance equivalent circuit models
The electrode-ear impedance showed a gradual decrease after insertion, which reached a plateau (impedance: <1 MΩ, phase: approximately −21°) after approximately 2 minutes (Fig. 2a,b).Extended Data Fig. 4a-f reveals granular structure in the surface morphology and the elemental composition of the electrophysiological electrodes after 1 hour of in-ear use.The sodium content of these granules indicated sweat residue, confirming that the decreasing trend in electrode-ear impedance was mostly due to sweat deposition at the electrode-ear contact, converting it from a dry to a wet electrode interface.EIS results in Fig. 2c,d show a capacitive to resistive shift of the electrode-ear impedance with decreasing frequency.Such behavior was modeled as an equivalent circuit model comprised of layered conductors and constant phase element (CPE) with an angle <90°.
Several equivalent circuit models have been developed for electrode-skin contact under different conditions 17,25 .The RRC model models the dry contact interface, consisting of a spread resistance Rs, in series to a parallel combination of charge-transfer resistance Rct and double layer capacitance Cdl.The RRCRC model models the wet contact interface, adding another RC element to the RRC model to represent the sweat contribution to the interface.The CPE model replaces the Cdl with a constant phase element, an imperfect capacitor representing the non-faradaic impedance with a phase shift < 90°.The impedance of a constant phase element is 1/(jω) n Q, where 0 < n ≤ 1, Q measures the magnitude of ZCPE while n fits the bilayer phase and n = 1 represents a capacitor.The total impedance of the three models is listed below: The equivalent circuit fitting took the averaged electrode-ear EIS data (Fig. 2c,d) as the input.Levenberg-Marquard algorithm was used for an amplitude weighing minimization function for the electrode-ear EIS data fitting: Here ra is the function to be minimized.N = 50 is the number of EIS data points across the 1-10 kHz frequency range.M is the number of parameters, M = 3 for the RRC and CPE model, M = 5 for the RRCRC model.f is the frequency.P1 … PM are the parameters to be fitted.

Electrophysiological sensing validation
The electrophysiological sensing performance of the in-ear integrated sensors was validated by conducting a parallel measurement with a 32-channel dry contact EEG headset (CGX Quick-32, Cognionics) on 2 participants.The setup of the in-ear integrated sensors remained the same as described in the method section.Separately, the EEG headset was mounted over the scalp of the participants (Extended Data Fig. 7a).Data transmission was through Bluetooth to the host computer in parallel to the in-ear sensors' data acquisition (DAQ) system.A 500 Hz sampling rate was configured.The impedance of each headset electrode to the skin was monitored prior to the measurement.The acceptable impedance threshold was lower than 2 MΩ, which indicated a solid electrode-to-skin contact.After removing unused ExG and accelerometer channels, the available number of EEG channels was 29.The parallel electrophysiological experiments included the alpha modulation, ASSR, and the eye movement experiments.All the experiments followed the same protocol as those with the in-ear integrated sensors alone.Event markers for the alpha modulation and ASSR experiments were added to the data stream using the EEG headset bundled data acquisition software (CGX Acquisition software, Cognionics).
Alpha modulation: Extended Data Fig. 7c,d show the 2-participant (4 ears in total) average alpha modulation power spectral densities from the in-ear integrated sensors and the EEG headset.Here the grand average results included datasets from all participants' electrophysiological channels.Not all the EEG headset channels recorded significant alpha modulation signals because alpha modulation originates from the specific brain region occipital lobe.27 channels with the alpha modulation ratio > 1.2 were selected for the grand average calculation.After channel selection, the alpha modulation ratio measured from the EEG headset was RAM = 4.40 (±2.36) (V 2 /V 2 ).By importing the channel location map to the EEGLAB toolbox in Matlab, scalp spectra at 8 Hz, 10 Hz, and 12 Hz were plotted with the locations of headset electrodes.The 10 Hz and 12 Hz spectra clearly showed the alpha modulation activities originated from the occipital lobe, while the 8 Hz spectral drifted to the central frontal region where theta band wave was originated 56 .The same alpha modulation signal was measured with the in-ear integrated sensors as a controlled and parallel experiment.The measured alpha modulation ratios with contralateral and ipsilateral referencing were RAM = 3.38 (±2.01) (V 2 /V 2 ) and RAM = 2.41 (±0.31) (V 2 /V 2 ), respectively.These results show that the EEG headset obtained a higher amplitude modulation ratio, due to the more than 9 times more channels utilized than the in-ear integrated sensors.Some channels from the EEG headset were close to the source of alpha modulation and produced a much higher amplitude modulation ratio (e.g., channel O1 located around the occipital lobe got a RAM = 10.63).With the contralateral referencing setup, the in-ear integrated sensors practically measured the potential difference across the scalp, which was comparable to that of the EEG headset.This can be demonstrated by the similar noise floor (described as power spectral density (PSD) at 15-20 Hz in Extended Data Fig. 7c,d, both 2-4 dB (rel.to μV/√Hz)) from the EEG headset and the contralateral referencing in-ear measurements.In-ear measurements using ipsilateral referencing had the lowest amplitude modulation ratio, which was consistent with the fact that ipsilateral referencing assessed mostly the local potential field surrounding the ear.
Auditory steady-state response: Extended Data Fig. 7e,f show the 2-participant (4 ears in total) average 40-Hz ASSR power spectral densities from the in-ear integrated sensors and the EEG headset.Here the grand average results included datasets from all participants' electrophysiological channels.Not all the EEG headset channels recorded significant ASSR signals because ASSR originates from the specific brain region, the temporal auditory cortices.19 channels with the ASSR signal to noise ratio (SNR) > 1.5 were selected for the grand average calculation.After channel selection, the ASSR SNR measured from the EEG headset was 3.70 (±2.36) dB.By importing the channel location map to the EEGLAB toolbox in Matlab, the 40 Hz scalp spectra were plotted with headset electrodes' locations, which showed the auditory responses originated from the temporal auditory cortices.The same ASSR signal was measured with the in-ear integrated sensors in a controlled and parallel experiment.The ASSR SNR obtained with ipsilateral referencing was 7.38 (±1.72 dB).While the ASSR signal was not observed with contralateral referencing.These results show that the ipsilateral referencing in-ear measurements obtained a higher ASSR SNR due to proximity to the auditory cortices.Nevertheless, channels of the EEG headset close to the auditory cortex also produced a significant ASSR (e.g., channel CP6 located around the right auditory cortex got an ASSR SNR of 10.63 dB).Contralateral referencing in-ear measurements were comparable to the EEG headset with a comparable noise floor (described as PSD at 40-80 Hz in Extended Data Fig. 7e,f, both −2 to −4 dB (rel.to μV/√Hz)).
In-ear Ipsilateral referencing outperformed contralateral referencing regarding ASSR measurements.
Eye movement: Extended Data Fig. 7b shows the synchronous eye movement signals measured by the in-ear integrated sensors and the EEG headset.For eyeball movements including glancing up, glancing down, glancing left, and glancing right, the EEG headset recorded an averaged amplitude of 175 μV, 129 μV, 124 μV, and 151 μV, respectively, which accounted for a grand average amplitude of 144.75 μV.And for heavy eye blinks and normal eye blinks, the EEG headset recorded an average amplitude of 994 μV and 276 μV, which accounted for a grand average amplitude of 635 μV.The EEG headset recorded a significantly larger amplitude when comparing the eye movement signals measured from the in-ear integrated sensors and the dry contact EEG headset.For eyeball movements, the headset measured an amplitude 29 times higher than the ipsilateral referencing measurements and 130% higher than the contralateral referencing measurements.For eye blinking, the headset measured an amplitude 20 times higher than the ipsilateral referencing measurements and 8 times higher than in-ear contralateral referencing measurements.Apart from the instructed 6 eye movements, the EEG headset also recorded more EMG artifacts as shown at t = 35.78s, 62.87 s, 62.07 s, 75.46 s, and 84.14 s, which were not observed from the in-ear measurements.In summary, in-ear ipsilateral referencing measurements recorded the smallest eye movement signals and practically no eyeball movement signals.In-ear contralateral referencing measurement recorded both the eyeball movements and eye blinking.However, its amplitude was much less than the dry contact EEG headset, which had the highest sensitivity and recorded more EMG artifacts.

Simultaneous electrophysiological sensing
The Integrated in-ear sensors are capable of simultaneous sensing of multiplexed electrophysiological modalities.Particularly, electrodermal activity (EDA), electrooculography (EOG), and EEG features including alpha, ASSR reported in Fig. 2 can be measured at the same time with the same electrophysiological sensors.This multiplexing of electrophysiological signals in the ear allows for a more complete knowledge of physiological processes that occur throughout different cognitive and emotional states.EEG can offer information on cognitive processes, such as attention and arousal, and assist with the identification of neurological illnesses by measuring electrical activity in the brain.EOG detects electrical activity in the muscles that regulate eye movements and can offer information regarding eye movements during a variety of cognitive activities, including reading and visual search.And EDA monitors the electrical conductance of the skin and can offer information about the activity of the sympathetic nervous system, which is related to emotional arousal and stress 57 .
Baseline: To demonstrate the feasibility of electrophysiological sensing multiplexing, a number of experiments were conducted on one participant using the ipsilateral referencing setup to simultaneously measure EEG and other electrophysiological modalities.First, a baseline experiment was conducted to evaluate solely the EEG-related 40-Hz ASSR response.Using the same experimental setup as outlined in "Methods: electrophysiological measurement system integration and the on-body setup", the electrophysiological signals were sampled at 500 Hz.The participant was presented with 40-Hz ASSR stimuli for one minute, during which time the participant's eyes remained closed.Extended Data Fig. 6a,b shows the baseline EEG streaming data in 1 minute and the obtained 40 Hz ASSR SNR to be 15.28 (±4.00) dB.

EEG + EOG:
The purpose of the combined EEG and EOG experiment was to show the measurement of eye movements, including eye blinks, while simultaneously collecting characteristic EEG features.As indicated in Extended Data Fig. 5, ipsilateral referencing captured eye blink signals with a much lower amplitude than the contralateral referencing configuration (and also EEG headset).The measurement was conducted in the same manner as the baseline experiment, with the participant instructed to blink every five seconds.Extended Data Fig. 6c shows eye blinking waveforms with a <100 μV, 5-s interval that are similar with the ipsilateral referencing EOG results in Extended Data Fig. 5. Extended Data Fig. 6d shows the PSD of the 1-minute EEG and EOG measurements with a 40-Hz ASSR SNR to be 17.36 (±2.86) dB.The result demonstrates simultaneous EEG and EOG data streaming, with EOG artifacts having negligible impact on EEG feature extractions.In this experiment, the EOG artifacts increased the <10 Hz band power by 7.40 dB.
Particularly, EOG artifacts are generally low-frequency signals (<10 Hz) that may overlap with EEG activity in this frequency range.This may lead to erroneous identification of low-frequency EEG activity, although the 40 Hz ASSR characteristic in this example remained unaffected, as corroborated by the ASSR SNR.

EEG + EDA:
The purpose of the combined EEG and EDA experiment was to show the measurement of earelectrode impedance while simultaneously recording EEG-specific characteristics.As introduced in the Methods section, the electrophysiological electrodes #1, #2, and #3 were connected to a potentiostat as WE, RE, and CE while still being connected to the electrophysiological DAQ system.The impedance was measured by applying 100 nA of sinusoidally alternating current between the WE and CE while monitoring the voltage drop between the WE and RE.Continuous impedance measurements were performed at three frequencies: 10 Hz, 100 Hz, and 500 Hz.Extended Data Fig. 6e,g,i show the simultaneous streaming of EEG and EDA data.EEG stream from electrophysiological sensor #2 was presented here since it only served as the RE in the impedance measurement and did not carry any current.The existence of periodic sinusoidal waveforms indicates current injection for impedance measurement.Measuring impedance at a higher frequency reduces artifacts in both the temporal and frequency domains of the EEG signal.This is shown by the decreasing amplitude of sinusoidal waveforms in the zoomed-in insets of Extended Data Fig. 6e,g,i, which occurred due to the stray capacitance in the potentiostat I/E converter (transimpedance amplifier) forming a low-pass RC-filter and also the capacitive component in the electrode-ear interface.In addition, since the majority of EEG-related frequency characteristics are less than 100 Hz, it is advantageous to perform impedance measurements for applications such as sweat analysis and lead off detection at a higher frequency band, as impedance measurements at lower frequencies do not carry more meaningful information.Extended Data Fig. 6f,h,j show 40-Hz ASSR SNR with 10 Hz, 100 Hz, and 500 Hz concurrent impedance measurement to be 11.37 (±1.62) dB, 11.07 (±1.02) dB, and 16.25 (±0.55) dB, respectively.Here Extended Data Fig. 6e shows a peak around 10 Hz resulted from the 10 Hz impedance measurement, while Extended Data Fig. 6j demonstrates that the out-of-band (500 Hz sampling frequency) 500 Hz impedance measurement has negligible influence on the EEG PSD, as evidenced by a similar 40-Hz ASSR to the baseline result.
EEG Motion artifact reduction EEG data collected from participants in motion is susceptible to degradation due to motion-induced artifacts.
Motion artifacts are primarily attributable to two sources, first is charge buildup at the ear-electrode interface from motion-induced displacement of electrodes relative to the ear canal, and second is cable movement, altering electrical characteristics at the contact site, and generating electromagnetic interference that could be picked up by the electrodes 58 .This study's experiments included both extremes of motion-artifact induction, where participants either engaged in intense physical activity (as with the combined EEG and lactate sensing experiment), or were instructed to remain motionless (for data segments used for alpha modulation and cognitive state analyses).Artifact rejection was performed by visual examination of EEG data and manual rejection of motion-related segments.
Removing motion artifacts is a challenging topic in electrophysiological research, usually requiring hardware and software solutions.Previous research 59 has demonstrated the efficacy of various signal processing methods for reducing motion artifacts, including independent component analysis (ICA), common average referencing, adaptive filtering, spectral (wavelet) decomposition, and automatic subspace reconstruction (ASR) 60 .These signal processing techniques could be applicable to recordings acquired from the in-ear integrated sensors used in this study, and here we investigate the feasibility of ASR for reducing 61 motion artifacts in in-ear EEG applications.
ASR is an adaptive spatial filtering technique that can be run online to remove high amplitude data components, relative to an artifact-free reference window that could be collected as a short (1 minute) calibration period 62 .This method performs a series of linear subspace projections on a sliding window of data to identify and remove principal components exceeding spatially varying thresholds determined during calibration.The removed components are interpolated and projected back to channel space.ASR provides tunable parameters including a cutoff parameter for adjusting channel variance thresholds that decide when ASR corrections would apply, and a window length of data for computing channel variances, which is expected to be on a similar timescale as the artifacts to be removed.ASR is available as part of the clean_rawdata plugin in EEGLAB.
The following experiment was performed to investigate the feasibility of using ASR with this study's in-ear system: a participant wearing the in-ear integrated sensors was instructed to stand motionless for about 1 minute, followed by a minute of uncued feet stamping.The participant was asked to keep their eyes closed for this entire duration, while listening to a 40 Hz ASSR stimulus generated and delivered as described previously.The recorded data was imported into MNE-Python, and bandpass filtered with cutoffs 0.1 and 50 Hz.The filtered time series for the entire experiment is shown in Supplementary Fig. 8a, with boxes identifying two manually selected 1 minute time windows corresponding to the baseline (minimal artifacts) and motion (stamping) segments.The baseline segment (zoomed in Supplementary Fig. 8b was aligned with the start of the experiment, and the motion segment (zoomed in Supplementary Fig. 8d) was selected to be a few seconds into the stamping activity, to prevent the relatively smaller motion artifacts from the commencement of physical activity from favoring ASR evaluation.
The filtered data was then imported into EEGLAB to run ASR calibration on the baseline segment using the asr_calibrate function with the cutoff parameter set to its default value of 5, and a reduced window length (window_len) of 0.1, to allow multiple windows to fit within one stamping cycle.Subsequently, the motion segment was filtered using the asr_process function (Supplementary Fig. 8e).PSDs were calculated for all three channels (EEGLAB's spectopo function, using a 10-s time window with no overlap), linearized, averaged, and converted back to log scale to get the channel mean shown in Supplementary Fig. 8c,f.The 40 Hz ASSR peak and the 31.2Hz (for lead-off detection) impedance measurement peaks are identifiable for the baseline period (Supplementary Fig. 8c, but become occluded by the elevated noise floor during stamping motion (Supplementary Fig. 8f).Running ASR is able to push the noise floor down such as the ASSR and impedance measurement peaks become visible again.Comprehensive hardware and software design considerations should be included for better motion artifact rejection, for example, an immediate future direction for this work is to further integrate the in-ear sensors with the backend DAQ system to eliminate dangling wires.

Supplementary Note 4 | Characterization of the combined electrophysiological and electrochemical sensing crosstalk EEG sensing to lactate sensing crosstalk analysis
Since EEG measurement was a passive potentiometric measurement, its effects on the lactate sensing were expected to be negligible.Under uncommon circumstances, the EEG sensors may be disconnected, resulting in transient charge transfers that may affect the lactate CA current measurements.Besides, for common-mode rejection, the driven right leg (DRL) electrode used in the EEG measurement sensed the common-mode voltage from the 3 electrophysiological electrodes and actively drove the concha cavum with an inverted common-mode signal to create a negative feedback loop.This driven signal could potentially propagate to the electrochemical electrodes and affect the lactate sensing results.A lactate sensing environment was created by drop casting 10 µL 0.1 mol/L PBS with 5 mmol/L Lactate to the PVA hydrogel to characterize these effects.A 180-s lactate sensing CA measurement was taken on two participants with a synchronous EEG measurement session.During this period, the following changes were made to the EEG measurement setup: physically disconnecting electrophysiological sensors from the DAQ, reconnecting electrophysiological sensors to the DAQ, disconnecting the DRL sensor from the DAQ, and powering off the DAQ.
Supplementary Fig. 10 shows the synchronous lactate CA current and EEG waveforms.First, the steadystate lactate CA current was approximately −1.9 μA, larger than those in Extended Data Fig. 9e,

Lactate sensing to EEG sensing crosstalk analysis
Major crosstalk was expected from the lactate measurement to the EEG measurement since the lactate CA measurements required applying a −0.2 V potential between the WE and RE, which may drift the nearby electrophysiological sensor recordings (Extended Data Fig. 9a).Several design considerations were made to minimize such potential crosstalk.An insulation SEBS layer was first built to insulate all the electrodes.The electrophysiological 3D electrodes and electrochemical electrodes had a ~750 μm height difference, which was made up by the height of the PVA hydrogel covering the electrochemical electrodes, serving as a sweat collector.Due to the existence of the PVA hydrogel, which produced a low-impedance pathway under a sweaty condition, most of the lactate CA current would flow through the PVA hydrogel instead of the skin, which could affect the EEG measurements.
Such potential crosstalk effects from the lactate measurement to the EEG measurement were evaluated by recording the corresponding signals during on-body operations.The experiment considered two factors.The first factor was the sweat in the ear, which affected the conductivity of the ear canal, the electrode-ear interface, and the PVA hydrogel.After cleaning the participants' ears with alcohol pads before the experiment, both a less sweaty and a sweaty environment were simulated by drop casting deionized (DI) water (Extended Data Fig. 9b) or saline (Extended Data Fig. 9c) to the PVA hydrogel covering the electrochemical sensor.The second factor was the electrochemical CA setup.By design, EEG measurements permit continuous monitoring of cognitive brain state for BCI applications, throughout extended time periods of wearable use.In contrast, lactate measurement produces an intermittent signal for fitness applications, at times when sweat is being generated.Following these two factors, under both the less sweaty and the sweaty conditions, the co-sensing crosstalk experiments consisted of a 70-s 40-Hz ASSR measurement session, followed by a 60-s electrochemical CA measurement session starting at 5 s after the onset of the EEG session and ending at 5 s before the EEG session ended.Before the crosstalk experiment, a 70-s ASSR baseline experiment was first conducted without any CA measurement session in parallel.Extended Data Fig. 9d shows the 3-participant-averaged baseline 40 Hz ASSR PSD, with an 18.72 (±1.40) dB SNR.
Extended Data Fig. 9e,f and Extended Data Fig. 9h,i show the 3-trial-averaged synchronous CA and EEG waveforms in the less sweaty condition and sweaty condition, respectively.The electrochemical electrodes measured steady-state currents of −0.25 μA and −0.39 μA under the two conditions, with their difference due to the conductive saline utilized in the sweaty condition.The −0.2 V potential was applied at the onset of the lactate sensing session (t = 5 s), resulting in a transient artifact and a discharging phase to the EEG data stream.For electrophysiological sensors 1 to 3, the resulting transient artifact had an amplitude of −17.7 mV, −10.2 mV, and −1.9 mV under less sweaty conditions, and a more prominent amplitude of −75.4 mV, −33.2 mV, and −2.1 mV under sweaty conditions.This amplitude distribution revealed a location-dependent relationship: the closer the electrophysiological electrode was to the electrochemical electrodes, the greater the amplitude of the transient artifact.From the observation of the waveform, it took approximately 5 s (less sweaty condition) and 7 s (sweaty condition) for all the three electrophysiological sensors to return to their pre-artifact state (Supplementary Fig. 9a,f).From the spectral analyses (Supplementary Fig. 9b-e and Supplementary Fig. 9g-j), the transient artifacts, which had a broadband frequency spectrum, elevated the noise floor of the ASSR PSD and corrupted the corresponding 40-Hz ASSR peak.The artifact blanking method was used to remove the artifact.Here a period of the EEG time series data was rejected for EEG analyses (Extended Data Fig. 9e,h red shaded region represents the data rejection period; the blue shaded region represents the data acceptance period).After a 3-trial-averaged 9.63-s data rejection (less sweaty condition), and a 3-trial-averaged 10.78-s data rejection period (sweaty condition), Extended Data Fig. 9g and Extended Data Fig. 9j shows a corrected ASSR PSD (as derived in Supplementary Text Note 4) comparable to the EEG-only baseline result (Extended Data Fig. 9d) with an SNR of 18.63 (±0.84) dB and 19.69 (±0.81) dB, respectively.
Spearman correlation analysis was used to determine the correlation between the corrected co-sensing ASSR PSD results obtained using the artifact blanking method and the baseline ASSR PSD results.
Between the EEG-only baseline and the corrected ASSR, the Spearman's p-values were 0.957 and 0.950 in the less sweaty and sweaty conditions, respectively.These findings reveal that after removing the transient artifact from the EEG data stream, the spectral analyses of the EEG readings were hardly affected by the electrochemical CA measurement.

Spectral analyses of the EEG with chronoamperometric induced artifacts
The co-sensing crosstalk results showed that a transient artifact and a discharging phase were introduced to the EEG measurement at the start of the CA measurement.Since the transient artifact, or in this case, an impulse, was a constant value across all frequencies between 0 to fs/2 Hz, the PSD of the corresponding EEG signal will have an elevated noise floor.The artifact blanking method was used to remove such artifacts.As shown in Extended Data Fig. 9e,h, a portion of the EEG time series data was rejected for analyses after the onset of the transient artifact.This section analyzes how different selection of this data rejection period affects the final EEG spectral analyses.
To investigate the noise performance of the artifact, unlike the methods used for alpha modulation or ASSR characterizations, no time-domain averaging (resulting in a smoothed noise floor) was used here for power spectral density calculation, and a longer window length of 50 s was used to achieve higher frequency resolution.The calculated PSD was shown in the log-log scale to demonstrate the 1/f noise common for biological systems.Four EEG data rejection periods were selected: 1.No data rejection: the whole 70-s EEG data stream was used to analyze the artifact's full effect on the EEG PSD. 2. 15 ms data rejection period: the EEG data stream 15 ms after the onset of the electrochemical CA measurement included the data stream after the peak of the transient artifact, analyzing the effect of the discharging phase.3. 1 s data rejection period: an intermediate stage.4. 5 s (less sweaty condition) or 7 s (sweaty condition) data rejection period: after investigating the waveforms (Extended Data Fig. 9e,h), the whole transient artifact along with the discharging phase was rejected for EEG analyses.
Supplementary Fig. 9b-e and Supplementary Fig. 9g-j show the PSDs calculated with the abovementioned four data rejection periods in the less sweaty and sweaty conditions.The results showed the artifact's location dependence, which was also shown by the amplitude of the transient artifact itself.With the increase of the EEG data rejection period, the elevated noise floor of the three sensors gradually decreased to the same level (Also indicated by the decrease of 1/f noise corner frequency).The artifact in the sweaty condition brought in more interference than the less sweaty condition, which was indicated by the elevated noise floor and the longer data rejection period needed to remove the artifact.After 5 s in the less sweaty condition, or 7 s in the sweaty condition, the discharging phase of the artifact appeared to end from the time domain, and the PSDs of the three electrophysiological sensors almost overlapped from the frequency domain.

Data rejection period calculation:
As a quantitative and more accurate approach, the length of the data rejection period was further determined by an iteration approach.For each of the 3 crosstalk EEG measurement datasets, the data rejection length increased by 0.1 s in each iteration.The resulting EEG PSD was correlated with its EEG-only baseline EEG PSD until it first reached a >0.95 Spearman's p-value.As a result, data rejection periods of 8.9 s, 11.2 s, and 8.8 s (9.63 s on average) were required for the three EEG datasets in the less sweating condition.The required data rejection times for the sweating condition were 10.5 s, 11 s, and 10.8 s (10.78 s on average).Using the averaged data rejection periods, the corrected 40 Hz ASSR PSDs are shown in Extended Data Fig. 9g,j.The resulting PSDs in the co-sensing experiments were comparable to the EEG-only baseline results (Spearman's p-value > 0.95) with close ASSR SNRs.In conclusion, with a conservative data rejection period of 11 s, the artifact resulting from the electrochemical CA measurement will not affect the EEG spectral analyses results.
j. Sensitivity testing of the electrochemical electrodes under different temperatures with different concentrations of lactate added on.k-n.Sensitivity testing of the electrochemical electrodes under different humidities with different concentrations of lactate added on.Here the temperature conditions were chosen as 25 °C, 30 °C, 32.5 °C, 35 °C, 37.5 °C, and 40 °C, respectively.The humidity conditions were chosen as 40%, 50%, 60% and 70% (relative humidity), respectively.

Matsumoto
Fig.4j-kshows the fitted equivalent circuit models' EIS.The CPE model produced the best modeling accuracy because it can model phase shifts <90°.The RRCRC model has higher accuracy than the RC model for bringing in the wet interface, which corroborates the presence of sweat, as shown in Extended Data Fig.4d-f.Eye movement signal measurement results and analysisTwo sensing modalities were explored with the in-ear integrated sensors in both ears.Ipsilateral referencing sensing used electrophysiological electrodes and REF electrodes from the same ear, measuring potential differences locally around the ear.Contralateral referencing sensing used electrophysiological electrodes and the REF electrode from the contralateral ear, measuring potential differences across the scalp in a similar manner to the scalp EEG.Due to the transmission attenuation of corneal-retinal signals, local field potentials around the ipsilateral ear are less affected by eye movements.In contrast, contralateral ear potentials are subject to greater eye-motion-induced signal components, as corroborated by a 60-s eye movement sensing session in Extended Data Fig.5a,b, which showed a higher noise floor along with much larger eye movement responses from the contralateral referencing setup.Four common eyeball movements and two eye blinks were characterized with the two referencing configurations, with the averaged time series results from multiple participants shown in Extended Data Fig.5c.

Design and fabrication of the sensors
Ipsilateral referencing measured 4.84 (±0.31) μV for eyeball movements, including glancing up, down, left, and right, which was 4.60% of the contralateral reference measures of 105.27 (±11.77)μV.Ipsilateral referencing contributed more electromyography (EMG) signals from eyelid movements and measured 30.84 (±2.58) μV for normal or heavy eye blinking, which was 38.89% of the contralateral referencing measures of 79.31 (±3.23) μV.Given that EEG amplitudes are typically less than 30 μV, the results suggest that ipsilateral referencing, or employing both the sensing and reference electrodes in one ear, has better eyeball movement artifact resistance than contralateral referencing, or scalp EEG.Both sensing modalities measured blinking signals, but ipsilateral referencing still recorded a smaller amplitude.According to the findings, ipsilateral referencing is more appropriate for EEG measurements involving eye movements.Contralateral reference is more appropriate for measurements in which the objects of inquiry are eye movement signals.
h because of the presence of lactate in the PVA hydrogel.Disconnecting electrophysiological sensors resulted in the saturation of corresponding electrophysiological sensors and a ~0.3 s, <50 nA fluctuation of lactate sensing current.Disconnecting DRL resulted in more noise in the EEG waveform but had a negligible effect on the lactate CA current.Powering off the DAQ also did not have a discernible effect on the lactate CA current.In conclusion, since the lactate sensing relies on the steady-state CA current, none of the EEG actions resulted in interference with the lactate sensing results.