A Conformable Organic Electronic Device for Monitoring Epithelial Integrity at the Air Liquid Interface

Air liquid interfaced (ALI) epithelial barriers are essential for homeostatic functions such as nutrient transport and immunological protection. Dysfunction of such barriers are implicated in a variety of autoimmune and inflammatory disorders and, as such, sensors capable of monitoring barrier health are integral for disease modelling, diagnostics and drug screening applications. To date, gold‐standard electrical methods for detecting barrier resistance require rigid electrodes bathed in an electrolyte, which limits compatibility with biological architectures and is non‐physiological for ALI. This work presents a flexible all‐planar electronic device capable of monitoring barrier formation and perturbations in human respiratory and intestinal cells at ALI. By interrogating patient samples with electrochemical impedance spectroscopy and simple equivalent circuit models, disease‐specific and patient‐specific signatures are uncovered. Device readouts are validated against commercially available chopstick electrodes and show greater conformability, sensitivity and biocompatibility. The effect of electrode size on sensing efficiency is investigated and a cut‐off sensing area is established, which is one order of magnitude smaller than previously reported. This work provides the first steps in creating a physiologically relevant sensor capable of mapping local and real‐time changes of epithelial barrier function at ALI, which will have broad applications in toxicology and drug screening applications.


Introduction
Air liquid interface (ALI) epithelial barriers, such as those found in the lung, skin and gut provide selective, physical and immunological protection for the body against pathogenic or toxic stimuli.Homeostatic functions include regulation of ionic conductivity, transport of essential metabolites and nutrients, and secretion of immunoand neuro-regulatory compounds. [1]The permeability of barrier tissue is a tightly regulated and selective process that is essential for health and, if perturbed, may cause permeation of toxins, antigens and microbes into the blood stream.Paracellular permeability is principally governed by tight junctions, which regulate the transport of small molecules and ions, while adherens junctions and desmosomes regulate cell-cell adhesion and communication networks. [2]Changes in the expression or localization of tight junction and adherens junction proteins may lead to a "leaky" barrier phenotype which is implicated in a variety of inflammatory and autoimmune disorders. [1,3]For example, inhaling environmental pollutants or cigarette smoke is known to increase epithelial permeability and lung inflammation within 10 mins of exposure, in in vitro models. [4,5][7] In the gastrointestinal tract, patients with inflammatory bowel disease (IBD), such as Crohn's and ulcerative colitis, display disrupted colonic tight junction proteins, [8] fewer adherens proteins [9] and an increase in permittivity of tracer molecules through the intestinal wall. [10]The increase in intestinal permeability associated with IBD, [11] and other GI disorders such as caelic disease, [12] is often referred to as a "leaky" gut.which allows potentially harmful microbes to pass through the epithelium and initiate an inflammatory response.
In vitro models have become indispensable tools for modelling these disorders and have shed light on their underlying pathways and treatment methods.The most common in vitro model of epithelial barrier function consists of a cell layer cultured on a semi-permeable Transwell insert.[15] In the case of respiratory models, it also allows for physiological culture at the Air-liquid interface (ALI) in which cells are exposed to air apically.[17] The formation of mucus and coordinated cilia beats are essential for the physical and immunological protection of the underlying epithelium, with their absence causing damage and inflammation. [18,19]Thus, any method used to quantify barrier integrity should refrain from perturbing the air-interfaced apical surface, for studies to retain their physiological relevance.In the case of the gut mucosa, in vivo, it is highly vascularized due to its large oxygen demands. [20]The oxygen in the blood is accessed via the basal epithelial membrane. [21]In vitro, this basolateral supply can be accurately mimicked using the ALI on top, which reduces diffusion distances and increases oxygen supply. [22]ALI also ensures nutrient exchange through the basolateral membrane.In some studies enhanced differentiation and improved morphological characteristics have been observed by maintaining only a thin layer of media on the apical surface. [22]ither complete ALI or semi-wet conditions (with small amounts of media on top) may enhance mucin production, cell polarization and barrier properties. [23]ommonly used methods to detect barrier integrity often require invasive procedures such as fixing or labelling the sample.For example, immunostaining of tight and adherens junction proteins (e.g., Zonulin-1, E-Cadherin, respectively) and quantifying the permeation of fluorescent dyes (e.g., FITC-dextran, Lucifer yellow).Non-invasive electrical methods are also available, which measure the conductance of ions across the epithelium, such as Trans Epithelial Electrical Resistance (TEER).Commonly used systems include the epithelial voltmeter EVOM 3 (World precision instruments, US), the Ussing chamber system and the CellZscope (NanoAnalytics, Germany).These systems operate by placing an electrode on either side of the biological membrane, in an electrolyte, e.g., cell culture media, and applying a voltage.The resulting readout is a measurement of the resistance of the cell barrier to ion flow, indicative of cell membrane permeability.A less permeable (more resistive) barrier will impede ion flow to a greater extent than a more permeable (less resistive) barrier.Electrodes measure the ionic voltage at a fixed AC frequency (12.5 Hz; EVOM) or DC (Ussing chamber) and calculate resistance via Ohm's Law.In contrast, the CellZscope system applies electrical impedance spectroscopy (EIS) at an AC frequency between 10 −3 and 10 6 Hz.The impedance data is then fitted to an equivalent circuit where the cell barrier is modelled at a resistor and capacitor in parallel.It is worth mentioning, that since each measurement technique and sampling method is unique, the resistance values extracted and/or units of measurement cannot be directly compared between systems.Indeed, a range of reported TEER units can be found in the literature (Ω, Ω cm, Ω cm 2 , Ω 0.5 cm) [24][25][26][27][28] depending on the sampling method and/or the mathematical model used.If circuit modelling is used, the equivalent circuit chosen may be refined to suit each specific experimental set-up and question.Thus, if multiple units are presented in the text, or one should wish to compare measurements taken between different systems, the overall trend and/or sensitivity, rather than numerical values, should be compared.
Although these commercially available techniques can provide quick, quantifiable electrical readouts, traditional electronic materials are rigid and placed a distance away from the biological sample.This distance, combined with inhomogeneities in electric field distribution and cell barrier tightness, limits the spatial sensitivity of gold-standard sensing electrodes.The field of organic bioelectronics has yielded advanced solutions to monitoring epithelial cell behavior which, compared to the commercial systems mentioned above, offer more mechanical flexibility and sensitive readouts.Of note, are poly (3,4-ethlyenedioxythiophene) doped (p-type) with poly (styrene sulfonate) (PE-DOT: PSS) based electrodes and organic electrochemical transistors (OECTs).Utilizing organic mixed ionic-electronic polymers boasts several advantages such as chemical tuneability and signal amplification through the electronic coupling to biologyderived ionic current. [29][32][33][34] We've also shown the ability to combine electronic-optical-metabolite sensing and integrate multiple biomimetic configurations including microfluidics [35] and scaffolds. [36,37]The resolution of organic electrode arrays even extends to measuring the conductivity of specific epithelial ionchannels, [38] spatially mapping changes in response to epithelial injury [39] and measuring single unit activity from neuronal populations. [40]However, most experimental studies use planar arrays, patterned on glass substrates, which limit their amenability to interface with advanced in vitro and ex vivo models with 3D architectures.Additionally, in the case of ALI culture, design is limited, as it requires the non-physiological addition of apical media for measurements using an external counter/reference electrode.Electrical methods which characterize barrier integrity under submerged conditions, both gold standard and experimental, disregard ALI physiology and the impact it may have on toxicology studies.Indeed, any repetitive or lengthy addition of apical media to ALI culture drastically disrupts barrier function (Figure S2, Supporting Information), mucus production [41] and may lead to artefacts.Flexible thin-film electrodes have been used widely for in vivo ALI applications including wearable skin sensors for monitoring humidity, pH, metabolite secretion etc. [42] and flexible electrocorticogram (ECoG) arrays for monitoring epileptic spike activity in brain tissue. [43,44]However, this technology is yet to be fully exploited for measuring the impedance of barrier tissue in in vitro ALI culture models.A proof-of-concept study by our group demonstrated how organic electronic devices may be adapted for this application by applying a flexible-gate electrode to the apical side of an ALI culture while measuring from a PEDOT: PSS channel beneath. [45]The flexible gate-device utilized cell-secreted mucus as an electrolyte, negating the need for apical media addition with rigid electrodes.The present study significantly improves upon the design and sensitivity.Importantly, we report here a flexible electronic device capable of monitoring at ALI in an allplanar configuration, rather than as opposing electrodes as done previously.Through the incorporation of a multi-size electrode design, it is possible to allow for biology-specific selection of optimal sensing area.The working electrode area, capable of detecting barrier integrity in vitro, is one order of magnitude less than any previously reported, with reduced measurement error compared to gold standard TEER measurements.This demonstrates the improved sensing efficiency of our design with the aim of being minimally invasive to the biological system.The ability of the device to measure barrier integrity and perturbations, in realtime, was assessed using the human bronchial cell line (Calu-3), human colon cell line (Caco-2) and primary human bronchial cells from healthy or COPD patients.The effect of the device on cell biocompatibility and barrier function was assessed via goldstandard methods including immunostaining and TEER.Overall, we present a novel technology for the assessment of epithelial integrity, which mitigates the non-physiological addition of media or external counter/reference electrodes for measurements.The platform presents a powerful tool for studying the health and dysfunction of human air-interfaced epithelia and will have wide applications for ALI toxicology studies in advanced in vitro models and tissue biopsies.

Device Characterization
Devices were fabricated using standard deposition and photolithography-based methods [46] (details in the Experimental Section).Briefly, 100 nm thick gold tracks were deposited between two, 2 μm thin, layers of Parylene C (PaC), which contained perfusion holes to improve contact and allow for the exchange of metabolites and oxygen, when measuring a biological sample.Gold working electrodes were patterned with radii ranging from 25 to 1000 μm (1.9 × 10 2 -3.1 × 10 6 μm 2 ) and an internal reference ring electrode of 8.47 × 10 6 μm 2 (Figure 1Ai

-iii).
A thin layer of PEDOT: PSS was spin-coated onto the surface of electrodes, using the sacrificial peel-off technique, [46] which has an established effect of introducing volumetric capacitance and reducing electrode impedance, in the frequency range important for biological signals. [40,47]The PEDOT: PSS layer exhibits a rainbow-like shine and possesses a diameter slightly smaller than that of the gold working electrodes (Figure 1Aii-iii).The perfusion holes traverse both layers of PaC and are discernible as the rectangular and triangular shapes shown in Figure 1Aiii.The deposition of thin film gold within a 4 μm PaC substrate means that the entire body of the device, shown in Figure 1Ai-iv, is mechanically flexible, ensuring conformability to a range of shapes and surfaces.The flexible body of the device is connected to a rigid FCC cable that allows for interfacing between the device and external electronics (Figure S1A, Supporting Information).To validate the device's flexibility and its capacity to endure deformation, the body of the device was subjected to substantial mechanical crumpling (Figure S1A-C, Supporting Information) and electrical characteristics were compared, following the protocol developed by Khodagholy et al. [48] In this instance, the device was placed in PBS, EIS was performed, and impedance measurements were modeled using a two-element circuit as described previously, [47] where R 1 = The combined resistance of the solution and electrode (Ω), C 1 = Capacitance of PEDOT thin film (F) and f C = cut-off frequency (Hz) which is defined as f C = 1/2∏R 1 C 1 .The electrical performance of the device was maintained upon deformation, with no significant differences in impedance curves (pre-crumple in black and post-crumple in red) or extracted values after crumpling (Figure S1D-F, Supporting Information).
To characterize device response relative to electrode size, scaling or normalization approaches are commonly used.Using the aforementioned two-element circuit model, when plotting against electrode area (A), R 1 , C 1 and f C scale as expected for PE-DOT: PSS coated electrodes of varying sizes [47] ; R 1 scales with A −0.4 , C 1 scales with A 1 and f C scales with A −0.4 (Figure 1B).To further validate our chosen model system, a normalized master curve of bode plots for all electrode sizes was plotted where Z′ = Z/R 1 and f′ = f/f C .When normalized, all curves overlay (Figure 1Bi,ii), confirming that the device and model behave as expected in the solution.

Modelling Device Operation with Phantom Barriers at ALI
Prior to testing on cellular systems, we validated the ability of our device to measure changes in barrier resistance using a "phantom" gelatin layer.A gelatin hydrogel was chosen, as this material is often used in the development of bioelectronic sensors to simulate the dielectric properties of tissue and biological barriers. [49,50]t is also possible to tune the conductivity of gelatin hydrogels to mimic the variation in resistivity observed in biological barriers of different origins.To this end, 10% (w/v) gelatin was prepared with either 0.01 m or 3 m KCl to mimic a barrier with high or low resistance, respectively.Gelatin hydrogels, of consistent volume and thickness, were injected into a Transwell submerged in PBS in a cell culture plate, to mimic the set-up of in vitro ALI cultures (Figure 1Aiv; Inset).To measure at the ALI, devices were first placed in PBS (200 μL) in the apical compartment, which was aspirated until the device was in contact with gelatin, where it remained for the duration of the measurements (Figure 1Aiv).A two-electrochemical cell set-up was used for EIS measurements with an internal, all-planar, CE/RE.Each electrode size, ranging from 25 to 1000 μm radius (1.9 × 10 2 -3.1 × 10 6 μm 2 ), was sequentially measured via a custom-built printed circuit board (PCB) connected to a potentiostat.Microscopy image, as a close-up of the white dotted rectangle in (Ai), showing PEDOT: PSS coated electrodes, gold tracks and perfusion holes (iii).Gelatin phantom placed in a 24 mm Transwell insert inside of a 6-well cell culture plate (iv).Scale bars = 2 mm for A, ii and iv; 50 μm for iii.Log-log plot of R 1, C 1 and f c against electrode area when measured in PBS (Bi), modelled using the equivalent circuit in inset.Data are fitted to a log-log robust non-linear regression where n = 6 for each condition.Log-log complex impedance plot of each electrode measured in PBS (Bii) and the corresponding plot when normalized to electrode area (Biii).Electrode radii (μm) are colour-coded and labelled on the left-hand side of the plot.Log-log plot of R 1 ′ and C 1 ′ against electrode area when measuring gelatin 0.01 m (Ci, dashed lines) or 3 m KCl (Ci, solid lines), as phantoms for barrier tissue of differing resistive properties.Data are fitted to a log-log robust non-linear regression.Log-log complex impedance plot of electrodes, normalized for area, when monitoring gelatin phantoms (Cii).Electrode radii are colour-coded and labelled on the left-hand side of the plot.The plot of extracted R 1 ′ normalized to electrode area (circles) against gold standard TEER (squares) for gelatin phantoms of 0.01 or 3 m KCl (Ciii).Data are presented as the mean ± SD where n = 6-8.
Although gelatin acts as a dielectric and may be modelled as a resistor and capacitor in parallel, at 10% (w/v), the capacitive element of gelatin is negligible compared to the overall capacitance of the system (Table S1, Supporting Information).Thus, the Bode plots are fitted to the simplified equivalent RC series circuit, where R 1 ′ = The combined resistance of the solution, electrode, gel and electrode-substrate interface, and C 1 ′ = The combined capacitance of the electrode PEDOT: PSS thin film and the gel.Indeed, the aim of circuit fitting should always be to use the smallest number of elements that describe all system features, to prevent over-fitting. [27]A comparison of extracted R and C values, showing how values scale with electrode area, in PBS

Scales with electrode area (A
Gel (0.01 m KCl) Gel (3 m KCl) a) The complete composition of PBS is listed in materials and methods.
versus gelatin phantoms, are shown in Table 1 and discussed below.The extracted capacitance of the device in contact with gelatin (C 1 ′) scales with electrode area (Figure 1 Ci; Blue line; A 0.9 ).Although this relationship is not strictly linear compared with C 1 device measurements in PBS (Figure 1Bi; Blue line; A 1 ), this can explained by the analogy of a parallel plate capacitor, where the capacitance is a function of plate area (in this case electrode area), permittivity of the substrate (in this case gelatin) and distance between the plates (in this case CE/RE and W/E).Thus, as the working electrode area increases capacitance values, of the electrode and gel combined (C 1 ′), will change as a function of sensing area, distance and the volume of gelatin the current passes through.
The extracted resistance values (R 1 ′) of the device in contact with gelatin scale inversely with electrode area (Figure 1 Ci; Red lines; A −0.2 ), but are less linear compared to R 1 device measurements in PBS (Figure 1Bi; Red line; A −0.4 ).This is due to the combined series resistance of R 1 ′, which is a function of both the electrode area, the volume of gel measured and the electrodesubstrate interface.By increasing the ionic conductivity (decreasing resistivity) of the gel from 0.01 m to 3 m KCl, we observed an expected reduction in the overall magnitude of R 1 ′ that is proportional across all electrode sizes (Figure 1 Ci; Red dotted vs opaque line).Changing ionic resistivity was not expected to change the capacitance of the gel and this can be seen in Figure 1 Ci (blue dotted vs opaque line).To validate the performance of the device, and to confirm that the circuit model chosen is appropriate for measuring resistive changes in gelatin phantoms, a master Bode plot, where Z′ = Z/R 1 and f′ = f × f C , was plotted (Figure 1Cii).When normalized, all curves overlap except for smaller sizes when measuring the highly resistive gel (Figure 1Cii; Dotted lines).This suggests that the assumptions made, and models used, are accurate for detecting resistive changes in tissue phantoms in electrode sizes above a ≥ 50 μm radius.Electrodes below this size would likely be less accurate in sensing resistive changes, due to the inverse scaling of electrode material resistance and electrode area, or would require an alternative model for fitting and extracting R 1 ′ values.
Furthermore, to validate our device against gold standard techniques, TEER values were taken in parallel using chopstick electrodes (EVOM 3; World precision instruments).Although it is important to validate any new device against a gold standard for proof-of-concept, the units of measurement between these techniques cannot, and should not, be directly compared.As mentioned in the introduction, units can vary depending on each individual experimental setup and, as such, the overall trends and sensitivity of measurements should be compared.In this instance, conventional TEER measurements are taken with chopstick electrodes placed on either side of, and some distance away from, the cell membrane and are calculated by multiplying ohmic resistance by cell growth area.However, because our device is placed in direct contact with the cell layer, ohmic resistance is normalized to the working electrode area, based on the relationship established in the non-linear regression in Figure 1 Ci (red lines) and Table 1 (Details of normalization can be found in the methods and in Table S2, Supporting Information).When plotting normalized resistance values (R 1 ″) against TEER, values obtained follow analogous numerical trends (Figure 1Cii).R 1 ″ values for the 0.01 m KCl gel were 753.9 ± 78.3 Ω cm 0.4 and for the 3 m KCl gel 135.0 ± 22.8 Ω cm 0.4 .TEER values obtained by measuring the same gels were 1374.2 ± 52.1 Ω cm 2 and 444.3 ± 14.1 Ω cm 2 respectively.This validates the ability of our device to detect gels of varying resistive properties to, at least, the same degree of accuracy as gold standards, but with the additional benefit of measuring at ALI in an all-planar configuration.Additionally, when comparing the percentage change in resistance, between the 0.01 m and 3 m KCl gels, our device showed a greater ability to detect differences in resistance compared to the gold standard (82.1% vs 66.7%, respectively).This suggests that our device may be more sensitive to measuring changes in resistive properties than TEER measured via the chopstick electrodes, a finding that is reproduced when performing cell experiments below.

Optimizing Electrode Size for Monitoring Human Epithelial Barriers at ALI
Epithelial barriers range in their resistive properties, depending on their location in the body.26] As such, it is important to determine an optimal electrode size suited to the models used here and where the non-sensing, cutoff regime lies.Although this has been characterized previously in set-ups with submerged planar multi-electrode arrays and external CE/RE, [30,38,39,51,52] here we perform such a characterization at ALI with an all planar WE/CE/RE.ALI cultures which subsequently have apical media added show a drastic reduction in barrier resistance, which we hypothesize concurrently reduces physiological relevance.This is a time-dependent reduction which can decrease initial TEER values by as much as 60% in 24 h (Figure S2, Supporting Information).
To model epithelial barriers of differing resistive values, we used two well-characterized cell lines; The bronchiole epithelial cell line Calu-3 [53,54] and the colon epithelial cell line Caco-2. [55,56]he representative Bode plots in Figure 2A and Figure 2B show EIS measurements of all electrode sizes in contact with Calu-3 and Caco-2 cells, respectively.A cut-off size was established, where, below 400 μm radius (≈5 × 10 5 μm 2 ), electrodes were unable to sense the impedance of the cell barrier.This is identifiable by the upward shift in impedance magnitude and phase angle, in the mid-frequency range, of electrode size >400 μm radius (Figure 2A,B).This frequency range is well established as relevant for detecting biological signals including epithelial/endothelial barrier formation, with lower or higher frequencies relating to device or electrolyte impedance, respectively. [40,47]maller sizes (˂400 μm radius) show impedance spectra equivalent to those taken at baseline in PBS, with no change in magnitude at f c when in contact with the cell layer.As impedance scales inversely with electrode area, this is likely due to the inherently high impedance of the smallest electrodes which mask any signal coming from the biological phenomena.This is further highlighted in Figure 2C where a representative electrode is chosen to demonstrate the EIS spectra of a sensing (500 μm radius; 5 × 10 5 μm 2 ) versus non-sensing (100 μm radii; 3.1 × 10 4 μm 2 ) electrode size when overlayed with the baseline (no cell) control.
Compared to previous studies using OECT or microelectrode devices to measure barrier integrity of in vitro lung and gut models, our sensing area (5 × 10 5 μm 2 ) is an order of magnitude smaller than previously reported. [45,52]This demonstrates the improved sensing efficiency of our design with the aim of being minimally invasive to the biological system.One reason for this improved sensitivity is that the device is in direct contact with the biological sample.By reducing the amount of electrolyte, and distance at the electrode/sample interface, ionic flow is restricted to the sensing electrode site.Additionally, noise from inhomogeneous electric field distributions at the electrode/electrolyte interface is reduced compared to traditional electrodes which are submerged in an electrolyte some distance away from the cell layer.This is demonstrated in Figure S3 (Supporting Information) where the addition of apical media, and consequent poor device contact, renders the device incapable of detecting barrier properties.In contrast, by reducing the amount of electrolyte atop the cell layer, and forcing closer contact, sensitive measurements are enabled (Figure S3, Supporting Information).Furthermore, the ability of the internal, all-planar setup (internal RE/CE in the same plane as WE), to measure the same degree of sensitivity as an external Pt RE/CE (Figure 2D) demonstrates the unique ability of our device to measure sensitively at ALI.Compared to previous experimental and gold standard set ups, that require submerged external RE/CE or source/drain electrodes, our device requires fewer components and mitigates the non-physiological addition of apical media.This is of great advantage for both in vivo and ex vivo set-ups, where manipulating the tissue to fit external electrodes on either side of the tissue is cumbersome and invasive.
In order to extract a resistance value, commonly used to quantify and compare barrier tissue, EIS measurements were modelled with a simple four-element equivalent circuit.This circuit model is well established for monitoring barrier function using electrical cell-substrate impedance sensing. [27]Here, R 1 = The combined resistance of the device and solution, C 1 = The capacitance of PEDOT: PSS film, R 2 = The combined resistance of the device-cell interface, cell barrier and cell-Transwell interface and C 2 = The capacitance of the cell barrier (Figure 2E).When plotting R 2 we obtain values of 3400.3 ± 82.6 Ω for Calu-3 cells and values of 6538.3 ± 856.7 (Ω) for Caco-2 cells.Gold standard TEER values, measured in the same wells, obtained values of 1993.9 ± 123.3 Ω cm 2 and 3820.4 ± 969 Ω cm 2 , respectively (Figure 2F).TEER values are comparable to those reported in literature, of Calu-3 grown in 6 well inserts of between 2000-2500 Ω cm 2 , [57,58] and Caco-2 cells of between 3000-4000 Ω cm 2 . [59,60]Although TEER values of Calu-3 and Caco-2 cells are reported anywhere in the range 100-2500 Ω cm 2 and 250-4000 Ω cm 2 , respectively, [28] these values largely depend on the cell growth area and electrode type used.In conventional calculations of TEER, ohmic resistance is multiplied by cell growth area so that, theoretically, experiments using arbitrary cell areas can be compared.However, when using chopstick electrodes, inhomogeneities in electric field/current distribution and barrier tightness, over the cell culture area, are known to influence specific resistivity. [61]These inhomogeneities have been shown to scale with cell culture area, but are not corrected for in conventional representations of TEER, and can lead to a one-order of magnitude difference in reported TEER values. [61]In contrast, our device enables local measurements with smaller electrode sizes in direct contact with the cell layer.Thus, there is the potential to mitigate cofounding variables such as inhomogeneous electric field distributions or barrier tightness.For example, using an electrode-array or by moving the device across locations, a spatial map of the barrier function may be created, which is not possible with indirect measurements such as TEER.The above demonstrates the ability of our device to reproducibly detect epithelial barrier properties, to the same degree of accuracy as gold standards, but with the additional benefit of reduced electrode size and measuring at ALI which is more physiologically relevant with respect to the biology.

Device Biocompatibility
The transparent nature of the device allowed for combined electrical-optical measurements and can be visualized, using phase contrast microscopy (Figure 3A-F).Calu-3 (Figure 3A-C) and Caco-2 cells (Figure 3D-F) can be seen in real-time before, during and after device contact, with the morphology of the cells remaining intact throughout.Confocal microscopy images further demonstrate this, where cell nuclei (in blue) can be seen when the device is placed in apical contact with the cell layer (Figure 3G).Device biocompatibility is confirmed with the live/dead viability assay, where live cells are shown in green and dead cells are shown in red (Figure 3H).The presence of the tight junction protein, ZO-1, after contact with the device for the duration of the experiment (sample fixed straight after removal), showed typical spatial localization to the perimeter of the cells (Figure 3I,J).The Z-stack images, shown in the insets of Figure 3I,J, demonstrate that the device contact did not affect the polarization of the cells, as can be seen by the apical polarization of ZO-1, above the nuclei, and the confinement of the nuclei to the basement of the cell.Additionally, the presence of the device for the duration of experiments (≈2 h) had no significant effect on TEER values of Calu-3 (Figure 3K; P = 0.38) or Caco-2 cells (Figure 3L; P = 0.42).Overall, we conclude that the device does not negatively affect the viability or barrier function of the cell models used.Further studies could look to establish if long-term measurements (days), as appose to acute measurements (minshours), are possible and at which time-point (if any) the device becomes disruptive to the biological sample.

Monitoring Epithelial Barrier Perturbation at ALI
After validating the device's ability to detect epithelial barriers and defining a cut-off sensing regime electrode size (5 × 10 5 μm 2 ), we tested the ability of the device to detect changes in epithelial integrity by the addition of ethylene Glycol Tetraacetic Acid (EGTA).EGTA is a calcium chelator commonly used to transiently disrupt calcium-dependent tight-and adherens-junctions by depleting the media of Ca 2+ . [32,56]Upon replenishing the system with calcium-containing media, the intercellular junctional network reforms along with barrier integrity.This calcium switch assay allows the in vitro assessment of barrier permeability, with lower TEER values corresponding to a disrupted epithelium and vice versa.EGTA was added to the basal compartment of wells, in order to keep the system at ALI.For acute measurements, 5 mm EGTA was added and measurements were taken at baseline (≈0 mins EGTA), disruption (≈30 mins EGTA) and recovery (≈2 h after media replacement).At 30 mins of treatment, there was a drop in impedance magnitude in the mid-frequency range, when a device is placed in apical contact for 2 h (H), stained using the live/dead viability assay kit (L3224; Thermofisher).Confocal microscopy image of Calu-3 cells which were fixed immediately after device removal (I,J; J is a higher magnification of the same sample as I; Below insets are orthogonal views at the same magnification).Cells were immunostained for the tight junction protein ZO-1 with primary rabbit polyclonal anti-ZO-1 antibody (617 300; Thermofisher) and secondary anti-rabbit antibody conjugated to Alexa Fluor 488 (green; ab150077, Abcam).Cells were counterstained for the nucleus using Hoechst 33 342 (blue; ab228511, Abcam).Scale bars = 50 μm.The presence of the device has no significant effect on TEER, measured with chopstick electrodes, for either Calu-3 (K) or Caco-2 (L) cell lines.Statistical analysis was performed using a one-way ANOVA with repeated measures.n = 3-9 for each group and presented as the mean ± SD; ns = P = 0.38 for Calu-3 and P = 0.42 for Caco-2.and a right downward shift of the phase, which resembled baseline (no cell) values (Figure 4A,B).EGTA had no effect on device or media controls (Figure S4, Supporting Information) and, after the addition of fresh Ca 2+-containing media, the impedance magnitude of cell monolayers recovered to pre-treatment values.
Extracted R 2 values were normalized to the pre-treatment condition and presented as % change from baseline (Figure 4C,D).EGTA significantly (***; P = 0.0001) reduced resistance values by −67.9 ± 10.4% for Calu-3 (***; P = 0.0001) and −79.7 ± 6.1% for Caco-2 cells (****; P < 0.0001).Upon addition of Ca 2+ -containing media, cells recovered to within 4.6 ± 14.4% and 1.1 ± 6.7% of pre-treatment baseline values, for Calu-3 and Caco-2 cells respectively.The normalization of resistance values allows for comparison to gold standard TEER measurements, which demonstrated comparable numerical trends (Figure S5, Supporting Information).This validates the ability of the device to successfully sense to, at least, the same degree of accuracy as gold standards, but with the additional benefit of measuring at ALI in an all-planar configuration.
Additionally, when comparing the percentage change between pre-treatment and EGTA-treated cells, our device showed greater sensitivity to barrier perturbations (Figure S5, Table S3, Supporting Information) and less measurement error (standard deviation) than the gold standard (Figure S5, Table S3, Supporting Information).Similarly to the data collected with gelatin phantoms (as described in Figure 1Ciii), this suggests that our device may be more sensitive to measuring changes in the resistive properties of barrier tissues.The increased accuracy in measuring recovery may also have a biological origin, in that the airway cells can remain in their native ALI environment during measurements, whereas the repetitive addition of apical media required for gold standard TEER measurements induces artefacts and/or barrier damage (Figure S2, Supporting Information).Changes in normalized C 2 values revealed a significant increase upon EGTA treatment with both Calu-3 (Figure 4E; P = 0.015, *) and Caco-2 cells (Figure 4F; P = 0.001, **) that returned to baseline after 2 h with fresh media.An EGTA-dependent increase in cell layer capacitance has been reported previously in Caco-2 monolayers and is hypothesized to take place due to transient rearrangements in actin microfilaments and cell morphology. [62]he large standard deviation in EGTA-treated membrane capacitance values, but not resistance values (Figure 4E,F), have also been reported. [62]This may be due to the reduced ability to reliably detect changes in membrane capacitance, at low membrane resistances. [63]Additionally, since EGTA predominantly disrupts paracellular pathways, that is, resistance across the cell layer, and does not change cell number or thickness, capacitance may be a less reliable measurement of this specific chemically induced barrier disruption.
After validating the device's ability to detect chemicallyinduced barrier disruption at acute time-points, the ability to measure changes in real-time was explored.The device was placed on the apical surface of the ALI culture and maintained in the same location throughout the duration of the experiment.All measurements were automated and carried out inside an incubator, to minimize any unnecessary disruption to cells.Here, 5 min intervals were chosen to represent "real-time" measurements, due to the time required for a frequency sweep (10 −1 -10 5 Hz), lasting ≈3 min.If required, more frequent measurements may be taken by reducing the frequency range used.
EGTA was added in excess (50 mm) to the basal compartment of cells and disruption was recorded every 5 mins, over 30 mins.The Bode plot in Figure 5A demonstrates a drop in impedance magnitude and a right downward shift of the phase, which resembled baseline (no cell) values after 30 mins.This is consistent with acute measurements (Figure 4) but with the added benefit of obtaining time-resolved signals.For clarity, Figure 5B shows a scaled version of the Bode plot in Figure 5A, to highlight the mid-frequency range.Similarly, extracted R 2 values showed an incremental decrease over time, reaching a value of −63.8 ± 7.7% of the baseline after 30 mins (Figure 5C).Extracted C 2 values showed the inverse trend with an incremental increase over time (Figure 5D).The large standard deviation in C 2 values seen at 30 mins is consistent with the acute EGTA-induced changes described above; as the resistance decreases and the cell layer becomes "leakier", the ability of the cell (or cell membrane) to store charge is compromised.Upon addition of Ca 2+ -containing media, recovery was measured every 5 mins over a total of 150 mins.The Bode plot in Figure 5E demonstrates the incremental recovery of the impedance magnitude and phase to pre-treatment values.Figure 5F shows a scaled version of this Bode plot, to highlight the signals in the mid-frequency range.Extracted R 2 (Figure 5G) and C 2 (Figure 5E) values showed an incremental increase and decrease, respectively, to baseline conditions.
In contrast to the acute measurements shown in Figure 4, it is not possible to compare the real-time data to gold standard EVOM measurements.The reason behind this is multifactorial.First, EVOM chopstick electrodes require submersion in media for operation.However, as discussed previously, prolonged or repetitive apical media in ALI culture drastically disrupts barrier function (Figure S2, Supporting Information).Thus, the continuous additional/removal of media for "real-time" EVOM measurements, is not compatible with ALI culture.Second, EVOM electrode operation is a manual and un-automated process, where chopstick electrodes have to be moved between wells for each measurement.Aside from the human error that may be caused, by inconsistent placement of electrodes, this process is not compatible with operation inside an incubator.Thus, long-term continuous measurements, where ALI cultures are to remain viable, are not currently possible with the EVOM system.Other electrode systems exist, namely the CellZscope, which offers automation and built-in circuit modelling.The issue remains that media submersion is required for electrode operation, making this system incompatible with continuous measurements at ALI.Furthermore, the temporal resolution of CellZscope measurements is limited to 1-12 data points per hour (depending on the model), [64] although as with our system, this could theoretically be reduced by restricting the range of frequencies measured.
In summary, when measuring acute timepoints, the device demonstrates the ability to successfully measure chemical disruption and recovery of epithelial cell barriers to the same degree of accuracy as gold standards.More significant, however, is the capacity of the device to measure ALI culture continuously, in real-time, to achieve a temporal resolution that is not yet possible with gold standards.This presents a non-trivial technological advance in sensor capabilities for monitoring epithelial physiology and may have a far-reaching impact on the design of future bioelectronic systems for both in vitro and in vivo applications.

Monitoring Disease-Specific Electrical Signals of Patient Cells Grown at the ALI
After validating the ability of the device to measure chemicallyinduced barrier perturbations in cell lines, we tested the ability of the device to differentiate between respiratory epithelial cells grown from healthy and diseased patients.To this end, normal human bronchial epithelial (NHBE) and COPD human bronchial epithelial (COPD HBE) cells were grown and differentiated at ALI (as described in methods).Experimental setup and cell culture membranes were kept consistent between cell line and primary cell experiments.The epithelium of patients with lung disease, such as COPD, is known to display altered morphology compared to healthy individuals.[7] Additionally, both COPD patient tissue (in situ) and cells grown at ALI (in vitro, as in this study) reproducibly show altered differentiation for ciliated cells, including changes in ciliated cell number, gene/protein expression and cilia beat frequency, compared to healthy controls. [65]As with all primary tissue samples, there is large donor-to-donor variability due to factors such as severity of disease, genetic makeup and lifestyle factors, with variability observed in healthy and diseased patients alike. [65]Thus, the ability to instantly detect epithelial barrier abnormalities in vitro or ex vivo (e.g., from patient biopsies) would present an invaluable technological advance in diagnostic and drug screening applications.Indeed, current gold-standards for detecting/confirming cancer or disease require lengthy and highly skilled techniques such as histology and immunohistochemistry of tissue biopsies.Here we present a plug-and-play device, that can be placed directly atop the patient-derived biological tissue, to detect changes in barrier properties associated with respiratory disease.
Bode plots revealed EIS curves, typical of barrier-forming cells, when measuring primary cells (Figure 6A); This included an upward shift in the impedance magnitude and phase angle, in the range responsible for biological or barrier signals (as described above; mid-range frequencies), that was significantly different from no cell controls.Extracted R 2 values highlighted differences between healthy and COPD cells showing values of 5898.3 ± 768.7 Ω and 4207 ± 1729.43 Ω, respectively (Figure 6B).The relatively large variation (SD) in values is a common phenomenon in primary cells, due to the large inhomogeneities in patient populations, as described above.Although R 2 values did not reach significance in this experiment, the numerical trend of COPD NHBEs, having a lower resistance than NHBEs, is more consistent with the literature [5][6][7] compared to the values obtained in the same wells with TEER, which indicated the opposite trend (Figure 6B).One explanation for lower TEER values with NHBEs (1280.9 ± 673.4 Ω cm 2 ) compared to COPD HBEs (2739.9 ± 447.8 Ω cm 2 ) is the observation that primary cells, especially NHBEs, grow less uniformly in a larger surface area: The probability of non-uniform cell coverage, including the formation of cell clusters, was more likely when grown in 6 well inserts (4.67 cm 2 ) compared to 24 well inserts (0.33 cm 2 ) (Figure S6, Supporting Information).Although 6 well inserts (4.67 cm 2 ) were originally chosen here, to keep experimental setups consistent throughout the manuscript, the accuracy and resolution of conventional TEER techniques may depend on cell growth area.As mentioned earlier, when using chopstick electrodes, inhomogeneities in an electric field/current distribution and barrier tightness, over the cell culture area, are known to influence specific resistivity and can lead to an order of magnitude difference in calculated TEER values, especially since the measurement is a bulk measurement averaged over the entire area. [61]It is expected these inhomogeneities would be considerably amplified when measuring primary cell culture due to less uniform growth over large areas, compared to cell lines, as well as donor variability.In contrast, our device enables local measurements in direct contact with the tissue layer.Thus, there is the potential to mitigate cofounding variables such as non-uniform barrier tightness or electric field distribution.To confirm this hypothesis, and the validity of our technique, a second set of experiments was carried out on additional NHBE and COPD donors, shown in Figure S7 (Supporting Information).Here, the cells were grown on a smaller growth area (24 well inserts, 0.33 cm 2 ) to minimize the effect that a large surface area has on uniform cell growth and the associated spatial/electric field inhomogeneities with chopstick electrodes.As expected, a significant decrease in R 2 from NHBE (4630.8 ± 1094.4 Ω) to COPD (1528.5 ± 355.9 Ω) was observed and this trend was indeed matched with the gold standard TEER measurement (693.3 ± 93.1 Ω cm 2 and 225.5 ± 69.3 Ω cm 2 respectively) which are also consistent with reported values in the same growth area (24 well insert). [5,7]he ability of our devices to measure electrical impedance means that we have access to a very rich data set compared to gold-standard devices.As mentioned previously, due to existing chopstick electrode limitations which measure resistance alone, at a single frequency, literature reports have a simplified picture of cell morphology changes between NHBE and COPD, and the resistance measurement provides little nuance in terms of donor variability and degree of disease.We noted, that in the case where the resistance change between NHBE and COPD was minimal (Figure 6B), there was an increase in impedance magnitude and phase angle compared to healthy samples (Figure 6A).Additionally, COPD HBEs showed a significant reduction in membrane capacitance versus NHBEs (***, P = 0.0009) with values of 1.7 × 10 −8 ± 5.5 × 10 −9 and 1.2 × 10 −7 ± 3.2 × 10 −8 , respectively (Figure 6C).In contrast, in the case where resistance was markedly decreased for COPD HBEs, in the second set of donors (Figure S7, Supporting Information), the capacitance was highly variable.Consistent with the capacitance trends described above for EGTA treatment, and those reported in literature, [63] the ability to reliably detect capacitance may be impaired when membrane resistance is reduced below a certain threshold.We posit, that the capacitance can be relied on as a useful parameter when the resistance is above a certain level, which must be determined on a case-by-case basis.The question arises, however, what the changes in resistance and capacitance collectively indicate for cell morphology.Given the different trends in resistance and capacitance between different donors of NHBEs VS COPD HBEs, immunofluorescence was used to investigate the morphology of the cell layers.
For the case of the first set of donors (Figure 6), the increased high-frequency Bode magnitudes and cell membrane capacitance, may be a result of COPD-related cilia dysfunction.Indeed, a range of cilia abnormalities are associated with COPD including cilia dyskinesia, altered cilia beat frequency and changes in cilia cell expression, number and/or structure. [66,67]This hypothesis is depicted in the schematic of Figure 6D and can be visualized when comparing the cytoskeleton structures of NHBE (Figure 6Ei) and COPD HBE (Figure 6Fi) samples.NHBEs display actin filaments, with punctate localization, on the apical cell surface (Figure 6Ei,ii).This is further highlighted in the Z-stack image, where a pseudostratified polarized cell layer is observed: Epithelial cells are elongated, nuclei are confined to the basal side and highly condensed actin filaments, indicative of uniform cilia distribution, are localized to the apical surface (Figure 6Ei; inset below).In contrast, COPD HBEs show very few, and spatially sparse, actin localizations on the apical cell surface (Figure 6F).These changes in filamentous actin localization were quantified and showed a significant reduction (*, P = 0.018), from 60.0 ± 9.4% to 9.1 ± 2.8%, in apical actin density in COPD samples (Figure 6G).
Changes in cell morphology or ciliated cell number may influence the membrane capacitance values observed.Indeed, previous reports demonstrate up to a ≈50% change in membrane capacitance due to changes in a number of cilia present [68] and, in an intestinal model, electrical capacitance was correlated to the number of epithelial projections. [69]However, ciliated cell expression can be highly variable, both within and between patient populations, [65] and, in our case, the second set of donors (Figure S7, Supporting Information) displayed different filamentous actin localizations and cell morphologies compared to the first (Figure 6).Additionally, EIS is a global measurement of the entire biological system.Thus, other cell-specific phenotypes may also contribute to the electrical signal.For example, Z-stack images revealed a COPD-dependent reduction in epithelial cell height and a change in cell shape and size across all donors (Figure 6G; Figure S7, Supporting Information).Indeed, epithelial cell height was significantly reduced, from 22.7 ± 0.9 μm to 15.3 ± 2.4 μm (*, P = 0.044), while cell nuclei size and aspect ratio was significantly increased, from 51.5 ± 19.0 μm 2 to 93.9 ± 29.4 μm 2 (****, P = <0.0001)and from 1.4 ± 0.3 to 1.6 ± 0.3 (****, P = <0.0001),respectively (Figure 6G).Nucleus elongation and changes in size and height of epithelial cells are indicative of epithelial-to-mesenchymal transition, a well-known pathophysiological process involved in airway remodeling and COPD, [70] and may also contribute to changes in electrical signal observed.
As described in the introduction, gold standard TEER measurements interrogate the sample by recording a square waveform at a single frequency.Thus, important biological information at other frequencies may be missed.By interrogating patient samples with our device using EIS, disease-specific and patient-specific signatures have been uncovered.Additionally, circuit modelling revealed significant changes in membrane capacitance of COPD samples, that may be linked to disease-related cell morphologies.We envision that this technology can contribute to increased drug screening of a range of patient-specific diseases and to the field of personalized medicine.In terms of work flow, an initial EIS measurement could be used to "diagnose" the severity of disease and be followed up with biomolecular assays to confirm tissue morphology based on predictions from the electrical data.

Conclusion
This work highlights the development of an all-planar flexible electronic device that can monitor epithelial barrier function and perturbations at the air-liquid interface.The device is userfriendly and is compatible with cell biology workflows, where tissues can be built up in vitro and may, for example, be utilized during the course of an intervention to study the response of the tissue, without compromising the ALI physiology.The mechanical flexibility of the device allowed for direct and conformable contact with biological surfaces, to provide a quantifiable electrical readout under physiological conditions.The multi-electrode design established a minimum electrode radius (400 μm) needed to capture epithelial properties, and disruption, of bronchiole (Calu-3) and intestinal (Caco-2) cell lines and primary cells from healthy and COPD patients.As this device was the first of its kind, readouts were validated against gold-standard TEER measurements.Although units cannot be directly compared, our device showed comparable trends and increased sensitivity compared to gold standards, but with the advantage of being in direct contact with the biological sample, without the need for apical media additions.As such, the sensing area achieved is one order of magnitude smaller than previously reported, highlighting the increased sensing efficiency of the design.Additionally, we confirm the device is biocompatible and does not negatively affect the barrier properties under study, for example, tight junction expression, cell morphology, polarization and TEER values.Further studies could look to establish the capability of this device to monitor long-term changes in biological signals (days-weeks) in order to enrich and widen the applications that it could be useful for.
Although the equivalent circuits chosen in this study were simple and fit the experimental question, future work could look to improve the accuracy of the extracted resistance values.For example, our extracted R 2 values represent the combined resistance of the cell barrier and the cell-substrate interface.If desired, for a more specific application, advanced mathematical modelling may be used to distinguish the resistance values attributed specifically to the cell barrier or the cell-substrate interface. [27,71,72]Future work includes a multi-electrode grid design, in which local changes in epithelial barrier properties can be spatial mapped across growth areas.The spatial resolution, and real-time readout, that this technology offers could be extremely beneficial when used alongside established work flows in drug screening and diagnostic applications.For example, in vitro drug screening platforms still rely on end-point assays which are disruptive to the sample.This device could be used, alongside these traditional assays, to offer a quick non-invasive assessment/early indicator of drug performance in real-time.This technology may also be used to aid a pathologist's assessment of biopsy tissue, allowing quick, real-time measurements on live tissue, prior to initiating lengthy histopathological processes.Patient-to-patient variability and degree of severity of disease inevitably mean there are changes in cell morphology between samples.The discriminating ability of our device to consider both paracellular permeability and ultrastructural features of cells, such as cilia, is an advantage in directing downstream studies and assays.One future avenue of research will also be spatial recording of the impedance of tissues, which could give information on intersample variation.Ultimately, we hope this technology will help speed up the drug-discovery and diagnostic process and could extend from respiratory epithelial/cilia disorder detection (as demonstrated here), all the way to cancer detection.Future work also includes the application of this versatile sensing platform to other models and tissue biopsies including the gut, esophagus, skin and eye.This technology has the potential to replace gold standard measurement apparatus, for example, chopstick electrodes, Ussing Chambers and could even be used for in vivo applications where assessing barrier health in situ could aid medical diagnosis and treatment.

Experimental Section
Fabrication of Devices: Devices were fabricated on 4′ silicon wafers (PDS 2010 lab coater 2; Speciality Coating Systems).A 2 μm insulating layer of Parylene C (PaC) was deposited.Negative photoresist (AZnLOF, 6 μm thickness) was spin-coated onto the wafer (5 s at 500 rpm then 30 s at 6000 rpm), and pre-baked at 110 °C for 2 min UV exposure (Contact Mask Aligner, MA/BA6; Suss) through a light mask (JD Photo Data) and subsequent development with AZ826 created valleys with no photoresist.To im-prove the adhesion of titanium to the PaC, the wafer was treated with O 2 plasma (O 2 gas 30% MFC2, 100 W; Diener electronic GmbH & Co) 1 min.An E-beam evaporator (Lesker) was used to deposit 10 nm titanium and 100 nm gold onto the sample.The sample was developed in acetone for 30 min, leaving gold tracks in the valleys.
Another layer of PaC (2 μm) was deposited.A dark mask and positive photoresist (AZ10XT, spin-coated at 5 s at 500 rpm then 30 s at 4000 rpm to give a 6 μm thickness) were used with development for 4.5 min in 400 K developer to form the device outline.The sample was etched (Plasma Pro 80 RIE; Oxford Instruments) for 23 min to remove PaC around the device outline.The excess photoresist was then removed, and three layers of soap (3% micro 90 solutions; Cole-Palmer) in deionized water were spin-coated on top of the device.A sacrificial layer of PaC (2 μm) was deposited.The photolithography and etching steps were repeated, with the aim of exposing the gold active sites, the electrodes and the contact pads.The wafer was activated with O 2 plasma.
Three layers of PEDOT: PSS were spun-coat on the device, then soft baked at 110 °C for 30 s between each layer.The PEDOT: PSS was prepared from 1% wt dispersion (PH1000; Heraeus, USA), 5% (w/v) of ethylene glycol, 0.05% (w/v) dodecylbenzene sulfonic acid, and 1% (w/v) glycidyloxypropyl trimethoxysilane and filtered with a 0.45 μm polytetrafluoroethylene filter.The sacrificial layer was peeled off, and the wafer hard-baked for 1 h at 130 °C to allow the PEDOT: PSS to cross-link.The devices were left in deionized water overnight to remove excess PSS.The devices were bonded (Finetech Lambda 1) to an FCC cable (Premo-Flex 15 015 Series; Molex) using 5um ACF tape (ACFFilm).
Device Sterilization: Devices were soaked in 70% ethanol for at least 24 h then allowed to air dry in sterile conditions.Devices were then washed thoroughly with sterile DI water and soaked in media for 30 mins prior to use.
Gelatin Phantom Preparation: Gelatin powder (Porcine skin; G1890; Sigma-Aldrich, USA) at 10% (w/v) was dissolved in aqueous KCl solution (0.1 or 3 m) at room temperature and mixed with a magnetic stir bar at 60 Hz in a water bath maintained at 35 °C for at least 1 h.Using a syringe, 2 ml of the gelatin solution was injected into the apical compartment of a Transwell and placed on ice until needed.

Figure 1 .
Figure 1.Characterization and mode of operation of flexible devices.Photograph (Ai) and brightfield microscopy image (ii) showing PEDOT: PSS coated working electrodes of radii 25-1000 μm and an internal ring reference/counter electrode of 8.5 × 10 6 μm 2 .Microscopy image, as a close-up of the white dotted rectangle in (Ai), showing PEDOT: PSS coated electrodes, gold tracks and perfusion holes (iii).Gelatin phantom placed in a 24 mm Transwell insert inside of a 6-well cell culture plate (iv).Scale bars = 2 mm for A, ii and iv; 50 μm for iii.Log-log plot of R 1, C 1 and f c against electrode area when measured in PBS (Bi), modelled using the equivalent circuit in inset.Data are fitted to a log-log robust non-linear regression where n = 6 for each condition.Log-log complex impedance plot of each electrode measured in PBS (Bii) and the corresponding plot when normalized to electrode area (Biii).Electrode radii (μm) are colour-coded and labelled on the left-hand side of the plot.Log-log plot of R 1 ′ and C 1 ′ against electrode area when measuring gelatin 0.01 m (Ci, dashed lines) or 3 m KCl (Ci, solid lines), as phantoms for barrier tissue of differing resistive properties.Data are fitted to a log-log robust non-linear regression.Log-log complex impedance plot of electrodes, normalized for area, when monitoring gelatin phantoms (Cii).Electrode radii are colour-coded and labelled on the left-hand side of the plot.The plot of extracted R 1 ′ normalized to electrode area (circles) against gold standard TEER (squares) for gelatin phantoms of 0.01 or 3 m KCl (Ciii).Data are presented as the mean ± SD where n = 6-8.

Figure 2 .
Figure 2. Optimizing electrode size for monitoring epithelial barriers at ALI. Representative Bode plots of Calu-3 (A) and Caco-2 (B) barrier measurements with each electrode radii (μm) size color-coded and labeled on the left-hand side of the plot.Bode plot of an electrode size capable of sensing (500 μm radii; purple line) and an electrode unable to sense (100 μm radii; red line) biological barrier impedance at f c (C). Plots in black correspond to no-cell controls for respective electrode sizes.Bode plot showing the sensitivity of the planar internal CE/RE compared to an external Pt CE/RE when measuring with an electrode size capable of sensing (500 μm; blue line) and not capable of sensing (100 μm; red line).(D) Dashed lines indicate measurements taken with the internal CE/WE ring on top of the cell layer and opaque lines indicate measurements taken with an external Pt mesh CE/WE below the cell layer.Schematic of the device set-up and four-element equivalent circuit used to model epithelial cell barriers (E); R 1 = combined resistance of the device, insert and basal media; C 1 = capacitance of the device and; R 2 /C 2 in parallel = the resistance and capacitance of the cell barrier respectively.Plot of extracted R 2 values against gold standard TEER values for cells grown in 6 well Transwell inserts (F).Data are presented as the mean ± SD where n = 4-9 biological replicates and 2-3 device replicates.

Figure 3 .
Figure 3. Optical characterization of device biocompatibility.Phase contrast microscopy (A-F) of Calu-3 (A-C) and Caco-2 cells (D-F) before, during and after device contact at ALI for 2 h.Confocal microscopy image of Calu-3 cell nuclei (stained and shown in blue; Hoechst 33 342, ab228511, Abcam) with a device (shown by the dotted lines) in apical contact (G).Confocal microscopy image of live (shown in green) versus dead (shown in red) Caco-2 cellswhen a device is placed in apical contact for 2 h (H), stained using the live/dead viability assay kit (L3224; Thermofisher).Confocal microscopy image of Calu-3 cells which were fixed immediately after device removal (I,J; J is a higher magnification of the same sample as I; Below insets are orthogonal views at the same magnification).Cells were immunostained for the tight junction protein ZO-1 with primary rabbit polyclonal anti-ZO-1 antibody (617 300; Thermofisher) and secondary anti-rabbit antibody conjugated to Alexa Fluor 488 (green; ab150077, Abcam).Cells were counterstained for the nucleus using Hoechst 33 342 (blue; ab228511, Abcam).Scale bars = 50 μm.The presence of the device has no significant effect on TEER, measured with chopstick electrodes, for either Calu-3 (K) or Caco-2 (L) cell lines.Statistical analysis was performed using a one-way ANOVA with repeated measures.n = 3-9 for each group and presented as the mean ± SD; ns = P = 0.38 for Calu-3 and P = 0.42 for Caco-2.

Figure 4 .
Figure 4. Monitoring acute epithelial barrier perturbations at the ALI.Representative Bode plot of Calu-3 (A) and Caco-2 (B) cells before (purple/green line), during (blue line) and after recovery (purple/green dashed line) addition of 5 mm EGTA for 30 mins.Percentage change in extracted R 2 values of barrier disruption and recovery, when normalized to pre-treatment values for Calu-3 (C; ***, P = 0.0001) and Caco-2 cells (D; ****, P = <0.0001).Percentage change in extracted C 2 values of barrier disruption and recovery, when normalized to pre-treatment values for Calu-3 (E; * P = 0.015) and Caco-2 (F; **, P = 0.001) cells.Statistical analysis was performed using a one-way ANOVA with Dunnett's multiple comparisons test.Data are presented as the mean ± SD where n = 4-5 biological replicates and 2-3 device replicates.

Figure 5 .
Figure 5. Monitoring real-time epithelial barrier perturbations at ALI. Representative Bode plot of Caco-2 cell disruption when treated with excess EGTA, measured every 5mins for 30mins (A).Signals due to biological disruption can be seen more clearly when highlighting the mid-range frequencies (B; This is the same Bode plot as A but with reduced axes).Percentage change in extracted R 2 (C) and C 2 (D) values of barrier disruption, when normalized to pre-treatment values.Representative Bode plot of Caco-2 cell recovery, measured every 5 mins for 150 mins (E).The mid-range frequencies are highlighted in (F; This is the same Bode plot as A but with a reduced axis).Percentage change in extracted R 2 (G) and C 2 (H) values of barrier recovery, when normalized to pre-treatment values.Data are presented as the mean ± SD where n = 4-5 biological replicates and 2-3 device replicates.

Figure 6 .
Figure 6.Monitoring disease-specific electrical signals of primary ALI cultures.Representative Bode plot (A), extracted R 2 (B) and extracted C 2 (C) values of primary human bronchial cells from healthy (NHBE; Green) and diseased (COPD HBE; Orange) patients.Statistical analysis was performed using an unpaired t-test; ns = P = 0.14; ***, P = 0.0009.n = 2 donors and 4 biological replicates for each condition.(D) Schematic showing a healthy bronchial epitithelium (left) and epithelial abnormalities associated with COPD (right) including cilia dysfunction and cytoskeleton re-arrangement (D).Confocal microscopy images of NHBE (E) and COPD HBE (F) cells fixed immediately after measurements (ii is a 2.5× magnification of the dashed white box in i; Below insets are orthogonal views of the same sample).Cells were immunostained for f-actin (Red; Alexa Fluor 594-conjugated Phalloidin, Thermofisher) and counterstained for the nucleus using Hoechst 33 342 (blue; ab228511, Abcam).Scale bars = 50 μm.Quantification of epithelial cell morphology including apical actin density, cell height, nucleus size and nucleus aspect ratio (G).Statistical analysis performed using an unpaired t-test; *, P = 0.018; *, P = 0.044; ****, P = <0.0001.n = 3-4 Z-stacks for apical actin density and epithelial height and n = 300-550 nuclei for size and aspect ratio.

Table 1 .
Non-linear relationship between extracted R 1 / C 1 values and gelatin phantoms, measured in PBS.