Resolving Spatiotemporal Electrical Signaling Within the Islet via CMOS Microelectrode Arrays

Depolarization of the β-cell plasma membrane in the islet of Langerhans is a key property that underlies β-cell function. β-Cells are excitable and respond to nutrient stimuli; for example, glucose. These nutrients are metabolized, which drives an electrical response, including membrane depolarization and elevated cytosolic calcium, which subsequently triggers exocytosis and insulin release (1).

β-Cells in the islet form highly connected networks by connexin36 (Cx36) gap junction coupling (2,3). As a result, oscillatory dynamics of membrane depolarization and subsequent changes in cytosolic calcium are synchronized, ensuring robust glucose-regulated and pulsatile insulin release (3–5). Pancreatic β-cells are functionally heterogeneous, and analysis of calcium dynamics has been used to classify leader, hub, or first-responder cell subsets (6).

We asked whether it is possible to expand current knowledge and to determine such functional subsets based on the underlying electrical activity of the cells. Previous work demonstrated that a combination of patch-clamp technique and single-cell RNA sequencing identified genes associated with functional heterogeneity (7). However, these data were collected from dispersed single cells. To overcome this caveat, we investigated the use of complementary metal-oxide–semiconductor (CMOS)-multielectrode arrays (MEAs) that have a high density of electrodes that enables simultaneous recordings of the electrical activity with cell-level spatial resolution.

MEAs have been used previously to measure the field potentials of cells within intact islets (8–11). An advantage of MEAs is that they provide high temporal resolution (kHz) and are noninvasive. Previous MEA measurements (referred to as “conventional” MEAs) provided some limited spatial information with one to three electrodes per islet (11). CMOS-MEA arrays have an electrode diameter of 8 µm and a distance between the electrodes of 16 µm, which corresponds to the size of individual islet cells (1) and enables the measurement of signal propagation from cell to cell.

In this study we examined whether subsets of β-cells with differing electrical activity or dynamics can be identified and how glucolipotoxic conditions influence the spatiotemporal electrical dynamics.

Apart from Cx36 gap junctions, the N-methyl-d-aspartate (NMDA) receptor has also been proposed to play a role in the synchronization of calcium dynamics in the islet (12). The NMDA receptor is a voltage- and ligand-gated ion channel that, when inhibited, influences the membrane potential of the islet (and individual cells), depending on the selected inhibitor (13,14). How strong the influence of the inhibitor is depends on whether it is a nonspecific or NMDA receptor subtype-specific inhibitor. How each type of NMDA receptor inhibitor influences the spatiotemporal electrical dynamics and islet synchronization is unknown.

Laboratory animal care was followed according to German laws (Az. 53.5.32.7.1/MS-12668, Health and Veterinary Office, Münster, Germany). Islets were isolated from adult male and female C57BL/6N mice (Charles River, Sulzfeld, Germany, or in-house breeding, Münster, Germany). Mice were euthanized using CO₂, and islets were isolated by collagenase digestion. After isolation, islets were kept in RPMI 1640 medium (11.1 mmol/L glucose) supplemented with 10% FCS, 100 units/mL penicillin, and 100 μg/mL streptomycin at 37°C in 5% CO₂ humidified atmosphere.

At 1–2 days after isolation, islets were placed on the CMOS-MEA chips (Multi Channel Systems [MCS], Reutlingen, Germany) covered with 2 mL of culture medium. Islets attached spontaneously to the electrodes after 1–3 days of culture and covered on average 100 electrodes. For experiments with glucolipotoxicity: 1 day after islets were placed on the CMOS-MEA, the culture medium was replaced by control medium (RPMI with 10 mmol/L glucose and 0.28% fatty acid-free BSA) or by glucolipotoxic medium (RPMI with 25 mmol/L glucose, 100 μmol/L sodium palmitate, and 0.28% fatty acid-free BSA) for 72 h.

For measurements, the culture medium was replaced by 2 mL bath solution. The CMOS-MEA was placed into the head stage, with heating applied from below, and the array was covered with a dark chamber because sensors are light sensitive (Fig. 1A and B). Electrical activity was initially examined before specific experimental protocols were conducted (Supplementary Material). Data acquisition was conducted with the CMOS-MEA-Control 2.8.0 software (MCS). Data were filtered with a Butterworth second-order filter between 0.1 and 500 Hz, processed at a frequency of 1 kHz, and visualized using a custom Python script.

Our analysis used two types of signals: one signal referred to as a spike (Fig. 1C and D) and another signal as a wave with a negative and positive deflection (Fig. 1E). Details on the signal extraction were described previously (15,16) and in the Supplementary Material.

Pearson product–based network analysis was performed as previously described (17–19). For each electrode pair within the area covered by the islet, the Pearson correlation coefficient was calculated using time series data over the indicated time window. Alternatively, the coactivity correlation coefficient was calculated by normalizing the time series between 0 and 1 and subsequently binarizing the time series as “1” if >0.5 and “0” if <0.5 (12). The coefficients were computed to construct a correlation matrix (Fig. 4A). An adjacency matrix was calculated by applying a threshold (0.5, 0.75, or 0.9) to the correlation matrix.

Electrophysiological measurements were performed with a bath solution of (in mmol/L) 140 NaCl, 5 KCl, 1.2 MgCl₂, 2.5 CaCl₂, 10 HEPES, and glucose as indicated, pH 7.4. Collagenase P was ordered from Roche Diagnostics (Mannheim, Germany) and RPMI 1640, FCS, and penicillin/streptomycin from Life Technologies (Darmstadt, Germany). Memantine was from Fisher Scientific (Schwerte, Germany). BSA, diazoxide, nifedipine, tolbutamide, sodium palmitate, and Ro 25-6981 (Ro-25) were from Sigma-Aldrich (Taufkirchen, Germany) or Diagonal (Münster, Germany).

Data were collected from islets of at least three independent mouse preparations for each series of experiments. Data are presented as scatter plots and/or means ± SD. The statistical tests are indicated in the Figure legends and in the Supplementary Material. GraphPad Prism 9.5.1 software was used for analysis. Values of P ≤ 0.05 were considered statistically significant.

The separation between male and female mice did not yield a sufficient amount of data for all experiments to conduct reasonable statistical tests. Therefore, statistics comparing male and female mouse islets were performed for selected data sets (Supplementary Table 1).

All data sets generated during and/or analyzed during the current study are available from the corresponding author upon request.

The electrical activity of cells within intact islets was measured using a CMOS-MEA chip (Fig. 1B, left). Several islets can be placed on each chip (Fig. 1B, middle). For analysis, each islet was considered separately. Only the electrodes covered by the islet were included (Fig. 1B, right), identified by exhibiting a stronger signal than the background noise (Supplementary Figs. 1 and 2). Two different motifs were used for the analysis, one being very fast (∼millisecond) positively oriented peaks, referred to as spikes (Fig. 1C and D and Supplementary Fig. 1), and the other being pronounced waves consisting of a negative and a positive amplitude deflection (Fig. 1E and Supplementary Fig. 2). The first steep and short deflection probably reflects the depolarization, whereas the second reflects the repolarization of the β-cells. On a long time scale (15 min) (Fig. 1C), spikes tended to occur in clusters (or bursts). However, for shorter scale measurements (<4–5 min), we did not distinguish between pattern of spikes in this study, but only whether spikes were present or not (Supplementary Material).

First, we examined how the voltage spike signals depended on glucose stimulation. Islets were sequentially stimulated with increasing glucose concentrations, and electrical activity was recorded 10 min after the start of each glucose stimulation. Figure 2A shows an example of CMOS-MEA electrodes (small squares) covered by an islet (gray squares) stimulated with increasing glucose concentrations, with electrodes showing spike activity highlighted in blue. Spike activity is present if at least one spike occurs within the 2-min evaluation interval (see Research Design and Methods). The proportion of electrodes with spike activity, expressed as a percentage of all electrodes covered by an islet, showed significant glucose dependence (Fig. 2B). The mean spike frequency of those electrodes showing spikes rose with increased glucose concentration (Fig. 2C), albeit with some variability. When considering all electrodes (including electrodes with no spikes) covered by the islet, the frequency also increased with increasing glucose with less variability (Supplementary Fig. 3A).

Next, we investigated the ion channels that might be responsible for the spike signals and how spike activity was influenced by modulating membrane depolarization. In the presence of 10 mmol/L glucose, the K_ATP channel opener diazoxide (250 µmol/L), which hyperpolarizes the membrane, reduced spike activity to zero (Fig. 2D). The L-type calcium channel blocker nifedipine (10 µmol/L), also reduced spike activity (Fig. 2E), but not as markedly as diazoxide. In the presence of 3 mmol/L glucose, addition of the K_ATP channel blocker tolbutamide (100 µmol/L) resulted in a significant increase in the recruitment of electrodes with spike activity (Fig. 2F). Similarly, in the presence of 3 mmol/L glucose, an elevated KCl concentration, as well as 20 mmol/L tetraethylammonium in the presence of 8 mmol/L glucose, also increased the spike activity (Supplementary Fig. 3B and C).

These data show that increasing glucose recruits more cells displaying spikes and that these signals depend on K_ATP and voltage-gated calcium channel activity.

In all analyzed data of this study, a single wave propagated through the islet during the recording (Fig. 1E, left, and Supplementary Fig. 2), but in some recordings, waves occurred in rapid succession despite the same stimulus of 6 mmol/L glucose (Fig. 1E, right). At 6–10 mmol/L glucose, the wave duration was in a range of ∼10 s (Fig. 3A and B), similar to that of fast calcium oscillations (3,20), but decreased in response to 15 mmol/L glucose. To account for potential variations in the velocity measurement by the distance or islet region used for the calculation, we chose three different distances from the wave start: 48 µm (3 electrodes), 80 µm (5 electrodes), and 160 µm (10 electrodes). Similar wave velocities were measured for each distance used, suggesting uniform electrical wave propagation across the islet (Fig. 3C and D). Thus, we used a distance of 80 µm for subsequent wave velocity calculations. The velocity was variable, therefore only a trend in the increase of the velocity was observed upon increasing glucose concentration (Fig. 3E). As suspected with mathematical approaches for calcium waves, the wave velocities could be overestimated (21), yet the velocity was in a similar range to that described previously for propagating calcium oscillations (70–150 µm/s (3,20). In contrast to the characteristics observed for calcium oscillations (3,20), we did not observe any correlation between islet size and electrical wave velocity (Fig. 3F).

Next, we examined whether the electrode time courses with respect to the wave and/or spike signals are synchronized, as is observed for calcium oscillations. For each pair of electrodes (covered by an islet), the Pearson correlation coefficient was determined (Fig. 4A). The average correlation per islet increases with elevating glucose concentration, but is highly variable (Fig. 4B). The lower correlation with 10 mmol/L glucose might originate from islet areas that react differently with respect to burst and interburst activity.

We next examined the extent to which the synchronization depended on wave and spike features using the data from Fig. 2B (8 mmol/L glucose stimulus + 1 additional islet). If the time series data contained a propagating wave (5 of 10 islets), the mean correlation was significantly and substantially higher (Fig. 4C). The mean correlation was also highly dependent on the velocity of the propagating wave, with islets showing fast propagating waves exhibiting a high mean correlation (Supplementary Fig. 4A). In contrast, the correlation was not affected by the spiking behavior: islets with <10% of electrodes (5 of 10 islets) showing spikes had a similar average correlation as islets with >10% of the electrodes showing spikes (Fig. 4D). The average correlation for islets was similar if we considered only electrodes with spikes or without spikes (Supplementary Fig. 4B).

Furthermore, there was no relation between the average correlation and spike frequency (Supplementary Fig. 4C). These data therefore suggest that the wave signal represents a marker or trigger for islet cell synchronization.

To further examine the link between electrodes (cells) that are highly synchronized and slow waves, functional networks were constructed for each glucose concentration. Each electrode was considered as a node within a network (Fig. 4E), with edges between nodes determined by a high correlation, above a threshold, which represents a functional connection. Consistent with calcium imaging data (18), the functional electrical connectivity within the islet was glucose-dependent, irrespective of the correlation threshold (Fig. 4F). Network properties, including the clustering coefficient and global efficiency and local efficiency all followed the same pattern independent of the correlation threshold (Supplementary Fig. 5A–C). Our data revealed a significant correlation between cell connectivity and the wave duration (Fig. 4G). Cells with the highest connectivity (normalized degree >0.6) tended to have longer wave duration (Fig. 4H) and showed significantly greater voltage amplitude in both positive and negative directions (Fig. 4I and J).

Since the spike signals do not contribute to islet synchronization, we investigated whether the same islet cells (electrodes) respond consistently by showing spike signals upon a repeated stimulus (Fig. 5A, and Supplementary Material Protocol 3). Consistent increases and decreases in spike activity were observed upon alternating between 3 and 8 mmol/L glucose (Fig. 5B). For the three stimuli, the number of times spikes were detected on an electrode was counted and compared between all electrodes covered by an islet (Fig. 5D, white bars). On average, 28 ± 18% of the electrodes covered by islet cells showed spikes within the first minutes of stimulation. The percentage of electrodes that always showed spikes (three of three stimulations) was 11 ± 9% under control conditions, suggestive of a population of highly responsive cells.

Islet cell dysfunction occurs in type 2 diabetes and glucolipotoxic conditions (22); therefore, we tested whether glucolipotoxicity affects the number of highly responsive cells by culturing islets with medium supplemented with 25 mmol/L glucose and 100 µmol/L palmitate for 72 h (GLT medium). For GLT-cultured islets, the number of electrodes that showed no spikes significantly decreased compared with control conditions (Fig. 5D), and the number of electrodes that showed spike activity (one of three times) tended to increase. In contrast, the number of electrodes that showed highly responsive cells was similar between GLT and control conditions (three of three stimulations) (Fig. 5D, gray bar).

Compared with control conditions, the spike frequency in response to each stimulation with 8 mmol/L glucose was significantly lower in GLT-treated islets (Fig. 5E), but not in the presence of 3 mmol/L glucose (Supplementary Fig. 5D). In addition, islets exposed to GLT medium showed a higher percentage of electrodes with spike activity when glucose was initially lowered to 3 mmol/L (first 3 mmol/L glucose treatment) (Fig. 5B and C), which normalized in the following intervals (Fig. 5C).

The velocity of the wave signals was significantly reduced in GLT-treated islets (Fig. 5F). The correlation coefficient was increased at 3 mmol/L and unchanged at 8 mmol/L (Fig. 5G).

A possible candidate to be involved into the network of β-cells within the islet is the NMDA receptor ion channel (12). To evaluate the NMDA receptor activity within the islet network, we examined the impact of acute activation and inhibition. Following addition of NMDA and the coactivator glycine to the measurement buffer, there was a small decrease in the number of electrodes with spike activity (Fig. 6A), which is consistent with cell hyperpolarization (14). There was no effect on the wave velocity or correlation across the islet (Fig. 6A). Inhibition of all NMDA receptor subtypes by memantine (Fig. 6B) or those containing the GluN2B subunit by Ro-25 (Fig. 6C) led to a significant decrease in the number of electrodes with spike activity but no change in the wave velocity or correlation. This suggests that some of the spikes are caused by NMDA receptor–driven calcium influx. The lack of changes in wave propagation or correlation suggests that NMDA receptors are not involved in slow waves and thus in synchronization, consistent with a lack of influence of spikes on synchronization. Of note, inhibition of NMDA receptors by Ro-25 clearly elevated wave velocity in five of seven experiments, pointing to a special role of GluN2B-containing NMDA receptors. An increased correlation with Ro-25 due to a higher velocity becomes clear when only the values of the active time (signal >50% of the signal range) are correlated (Supplementary Fig. 6).

The CMOS-MEA enables recordings of the spatial-temporal electrical activity via measurement of the field potential. We have demonstrated that this approach can be applied to the islet to characterize the spatiotemporal electrical response and heterogeneous responses throughout the islet. We analyzed two different voltage signals: spikes, which are signals generated by voltage-dependent calcium channels, and voltage waves resembling the characteristics of calcium oscillations. The calcium-permeable NMDA receptor appears to be a modulator of spike activity but has minor effects on wave velocity. Spikes and waves are both glucose-dependent and show altered behavior in response to glucolipotoxicity. Through network analysis, we showed that the overall synchronization of islet cells depends on propagating waves rather than spike activity.

Typically, when stimulated with >6 mmol/L glucose, the membrane potential of a mouse β-cell depolarizes from −70 to −50 mV (1). Islet cells in our experiments exhibited spikes over an extended glucose concentration range (6 to 30 mmol/L). Our data suggest that the spikes result from voltage-dependent L-type calcium channels (Fig. 2D–F), which correlates to the fact that L-type calcium channels drive mouse β-cell action potentials (APs) (1). However, because nifedipine does not completely eliminate spike activity, other calcium channels, such as R, N, or P/Q, may also play a role (1). Furthermore, the calcium-gating NMDA receptor influences the spike activity (Fig. 6A–C). Nevertheless, our determined spike frequencies (Fig. 2C) are clearly different from any AP signals and should not be equated with APs recorded during electrical bursts of whole islets with conventional techniques. This discrepancy is because extracellular electrodes detect the integral depolarization vector, which may consist of contributions from a few cells in the recording plane as well as above it. In addition, APs of a burst also spread from cell to cell, but presumably not all in the same direction.

Overall, the CMOS-MEA measurements have the advantage to provide additional spatial information at the level of individual cells.

Calcium imaging data has often restricted temporal resolution up to 4–8 Hz (3,23), which the CMOS-MEA chip far exceeds with a frame rate of 1 kHz (can be extended to 25 kHz) while measuring a plane of cells. Field potential measurements have the additional advantage over calcium imaging in that they give more information about cell-to-cell connectivity controlled by the plasma membrane as cytosolic calcium is influenced by transmembrane currents as well as by release from intracellular stores and export from the cytosol. We note that calcium can also indirectly influence the membrane potential (e.g., via calcium-dependent potassium channels).

A limitation of this method is that a simultaneous measurement of intracellular calcium (from which MEA measurements are recorded) or insulin secretion is not possible with our setup to establish a better link to stimulus-secretion coupling.

We observed a high level of synchrony or correlation between voltage time-courses of electrodes covered by islet cells, which are qualitatively similar to the synchrony in calcium measurements between β-cells that has been previously reported and depends on gap junction coupling (19). Our network analysis demonstrated that islets have synchronized electrical activity (Fig. 4B). Importantly, the electrical waves were associated with substantially higher correlated electrical activity (Fig. 4C). In contrast, the faster spikes do not contribute to the correlation of islet electrical activity (Fig. 4D).

Functional analysis of calcium dynamics has classified subsets of β-cells (6), such as hub cells, which are defined as having more functional connections to other cells (24). Regarding our electrophysiological characterization, we identified electrodes with more functional connections that were associated with a much stronger wave signal (Fig. 4H–J). This further supports a role for hub cells to provide greater electrical coordination across the islet. We also identified a subset of most responsive cells that always showed spike activity upon repeated glucose stimulation (Fig. 5D). These cells made ∼11% of islet cells (electrodes). Whether these two subgroups (hubs and highly responsive) overlap remains to be clarified. However, the lack of a link between spiking and synchronization suggests that these consistently spiking cells represent a different population or subset of cells. A similar finding was made based on calcium imaging data (25).

When network analysis was performed with a series of elevations of glucose concentration, at 10 mmol/L glucose, there was a small drop in the average correlation and for each network parameter (Fig. 4B and Supplementary Fig. 5A–C). The greatest variation in wave velocity was also observed at 10 mmol/L glucose (Fig. 3E), where wave velocity and average correlation are related (Supplementary Fig. 4A). At 10 mmol/L glucose, islets reached a maximum spike frequency (Fig. 2C), and these fast spikes lack any influence on the overall signal correlation (Supplementary Fig. 4B). Thus, an increase in spike signal frequency generates an apparent decrease in synchronization, whereas at further increases in glucose concentration are associated with increased velocity of slower waves that would provide an increase in the synchronization of electrical activity.

Neither activation nor inhibition of the NMDA receptor affected electrical synchronization, largely because only spike activity is significantly affected (Fig. 6A–C), which does not contribute to a higher correlation (Fig. 4D). Prior work that investigated the influence of NMDA receptor blockers on calcium waves also reported a large interislet variability regarding wave velocity (12). This work described increased synchronized activity due to the inhibition of all NMDA receptors, which was suggested to not result from a higher calcium wave velocity since the wave velocity was not affected by the inhibitor MK-801. In our case, velocity and correlation of electrical waves were not significantly changed by the NMDA receptor inhibitor memantine (Fig. 6B). On average, the velocity decreased slightly with memantine, which fits with studies in neurons showing that activation of NMDA receptors opens Cx36 channels (26). However, the trend toward increased velocity by the GluN2B-specific inhibitor Ro-25 suggests an interaction between GluN2B-containing receptors and Cx36 channels, perhaps because both interact with Ca²⁺/calmodulin-dependent protein kinase II (27). However, this needs to be studied in more detail for β-cells.

Glucolipotoxic environments have detrimental effects on β-cell function and cellular mass (22). A previous study by our group using a conventional MEA demonstrated via a single electrode recording of islet electrical activity that glucolipotoxicity increased the overall electrical activity of the islet (28). The more detailed analysis with CMOS-MEAs shows that a glucolipotoxic environment increased the number of cells exhibiting spikes. Furthermore, the number of highly responsive cells in terms of spiking activity was not altered by the glucolipotoxic conditions (Fig. 5D), suggesting that some cells are pushed to a lower glucose threshold but that this process is reversible upon the repetitive stimulation. We also observed that islets exposed to glucolipotoxic conditions had a lower spike frequency (Fig. 5E), presumably due to more cells being recruited but with low activity (Fig. 5D), thus lowering the frequency average. These combined effects where single-cell activity increased but the spike frequency decreased may reflect a compensatory mechanism that enables the glucolipotoxic environment to not massively disrupt insulin release (29).

Lipotoxic or glucolipotoxic milieus were previously shown to affect connectivity with the islet by reducing the proportion of functional links between cells (24), reducing Cx36 gap junction coupling (30), and disrupting calcium oscillation synchronization (31). Supporting the hypothesis of reduced coupling, we observed that the wave velocity decreased when islets were challenged by a glucolipotoxic environment for 72 h (Fig. 5F). In contrast, our islets had a higher average correlation in electrical activity (Fig. 5G) when exposed to glucolipotoxicity. This was observed as a significant effect at 3 mmol/L and as a trend at 8 mmol/L glucose and may result from the lower glucose threshold increasing activity (Fig. 5G). In summary, our data indicate that GLT conditions affect fast spiking by decreasing spike frequency and slow wave electrical signals by slowing the propagation velocity. This might not have a major effect on intracellular calcium levels initially, given the additional influence of store-released Ca²⁺ release but could lead to a loss of coordination between cells over a longer term.

We demonstrate that CMOS-MEA enables high-resolution recording of spatiotemporal electrical activity across the pancreatic islet. We identified subsets of islet cells with a high electrical response to glucose and a high robustness to a glucolipotoxic environment. This environment decreased signal propagation across the islet and affected the fast spike behavior of other islet cells. NMDA receptors were identified as modulators of the fast electrical spikes as well, but have heterogeneous effects on signal propagation, and, therefore, do not influence the average correlation of islet cells under physiological conditions.

Acknowledgments. The authors thank Dr. Sven Schönecker (MCS) for his continuous support on the CMOS-MEA device.

Funding. This study was funded by Deutsche Forschungsgemeinschaft (M.D., Research Training Group GRK 2515, Chemical Biology of Ion Channels). The data analysis research was supported by the research training group “Dataninja” (C.B., Trustworthy AI for Seamless Problem Solving: Next Generation Intelligence Joins Robust Data Analysis) funded by the German federal state of North Rhine-Westphalia. Furthermore, funding was provided by the National Institute of Health National Institute of Diabetes and Digestive and Kidney Diseases grants R01 DK102950 (R.K.P.B.) and R01 DK106412 (R.K.P.B.) and the National Science Foundation (NSF) Graduate Research Fellowship DGE-1938058_Briggs (J.K.B.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.G. wrote the manuscript. A.G. analyzed the data. A.G. and J.O. performed study experiments. A.G., R.K.P.B., and M.D. edited the manuscript. A.G. and M.D. designed the study. J.D.H., J.K.B., T.B., R.K.P.B., T.D., and C.B. wrote the code that was used to analyze the data. A.G. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented as poster at the 83rd Scientific Session of the American Diabetes Association in San Diego, CA, 23–26 June 2023.