To acquire high-speed 3D time course imaging of β-cell Ca²⁺ oscillations within intact islets, we utilized a lateral-interference tilted excitation light-sheet system (Fadero et al., 2018) mounted on an inverted fluorescence microscope (Figure 1A and Methods). To image Ca²⁺ activity and spatially resolve individual β-cells in intact islets, islets were isolated from Ins1-Cre:ROSA26^{GCaMP6s/H2B-mCherry} mice that express cytosolic GCaMP6s Ca²⁺ biosensors and nuclear H2B-mCherry reporters selectively within β-cells. We first compared the images collected by the light-sheet system with a commercial spinning disk confocal using the same ×40 water immersion objective. Similar to a widefield microscope, the axial resolution of the light-sheet microscope is dictated by the numerical aperture (NA) of the objective lens (~1.1 μm for a 1.15 NA objective and GCaMP6s emission) (Fadero et al., 2018), whereas the spinning disk uses a pinhole array to enhance axial resolution. At a shallow depth of 24 µm from the coverslip, H2B-mCherry-labeled nuclei and GCaMP6s-labeled β-cells were resolved both by the light-sheet and the spinning disk confocal. However, the nuclei were only resolved by the light-sheet system at depths ≥60 μm due to the reduced light scatter from side illumination (Figure 1B). Thus, the main advantage of the light-sheet system is the ability to image the entire islet in 3D (Figure 1C).

Prior studies of β-cell Ca²⁺ oscillations utilized 1 Hz imaging to resolve phase shifts for single β-cell traces within a single 2D plane (Benninger et al., 2008; Hraha et al., 2014; Skyggebjerg, 1999; Stožer et al., 2013a). To image the entire islet at similar acquisition speeds, the hardware was operated under triggering mode to minimize communication delays (Figure 1—figure supplement 1). In this mode, the lasers and the piezo z-stage were triggered directly by the camera, which received a single set of instructions from the computer via the NiDAQ card. To image 132 µm into the islet at 2 Hz, 15 ms was allowed for photon collection and stage movement for each of the 4 µm z-steps. Initially, an hour-long delay was required to save 120,000 imaging files after running a continuous 30-min experiment. Because this delay is only observed after the first 3 min of imaging, it was possible to eliminate the delay by separating the acquisition into a series of 3-min loops (Figure 1—figure supplement 2).

Individual β-cell nuclei were located using the Spots function of Bitplane Imaris software (Figure 1D). For each nucleus, a cellular region of interest (ROI) was defined by a sphere of radius 4.65 µm around the ROI center based on results from a computational automated radius detection (Briggs et al., 2024). The mean GCaMP6s intensity of each cell ROI for each time point was calculated and exported as single-cell traces. We did not observe any significant photobleaching using continuous GCaMP6s and H2B-mCherry excitation over the course of the experiment. The combination of this light-sheet system, islet cell labeling, and analysis pipeline allows for imaging of Ca²⁺ from nearly all β-cells in the islet at speeds fast enough for spatio-temporal analyses to identify functionally heterogenous β-cell subpopulations.

To investigate the synchronization between β-cells across the islet in 3D space, we imaged and extracted Ca²⁺ time courses for Ins1-Cre:ROSA26^{GCaMP6s/H2B-mCherry} islets that exhibit slow oscillations. Following the network analysis methods set forth in Šterk et al., 2024, we calculated the correlation coefficient between every cell pair and defined an ‘edge’ between any cell pairs whose correlation coefficient was above threshold (Figure 2A). This threshold was set such that the average number of edges per cell, also called the ‘cell degree’, was equal to 7. A fixed average degree rather than fixed threshold was used to mitigate inter-islet heterogeneity (Šterk et al., 2024). An example 3D network for a single β-cell within an islet is shown (Figure 2B) along with the frequency distribution of all β-cells within the islet (Figure 2C). The high degree cells (top 10% of the total population, blue) and low degree cells (bottom 10% of the total population, red) were then mapped onto a 3D projection of the islet and onto the Ca²⁺ time course (Figure 2D). Compared to average degree cells, the high degree cells were consistently located at the center of the islet while the low degree cells were located on the periphery, indicating that the islet network is distributed in a radial pattern (Figure 2D, E).

To analyze the propagation and spatial orientation of Ca²⁺ wave in 3D space, we calculated the lagged correlation coefficient between every β-cell and the islet average, and identified the phase lag with maximum correlation (Figure 2F). The spatial distribution of phases of an example islet is shown (Figure 2G), along with the frequency distribution of all β-cells within the islet (Figure 2H). The early phase cells (top 10% of the total population that depolarize first and repolarize first, blue) and late phase cells (bottom 10% of the total population that depolarize last and repolarize last, red) were then mapped onto a 3D projection of the islet and onto the Ca²⁺ time course (Figure 2I). Unlike the islet network, for which the high degree cells emanate from the islet center (Figure 2D, E), the early and late phase cells were each located at the islet periphery, and show a clear temporal separation between depolarization and repolarization (Figure 2I, J).

We next assessed the stability of high degree cells and early phase cells over time, by assessing their presence across consecutive oscillations (Figure 3A, B; Figure 3—figure supplement 1). The high degree cells and early phase cells have a similar ~60% retention rate between oscillations (Figure 3C). Strikingly, when we examined the center of gravity for each β-cell subpopulation, we found that the center of gravity of the early phase cells moved significantly more than that of the high degree cells (Figure 3D). This indicates that the early phase cells tend to change their identity more with each oscillation. To further investigate the change in location of early phase cells, we used principal component analysis to identify the principal axis between early and late phase cells (wave axis) and calculated the rotation of the axis between each oscillation. Of the 25 islets examined, 57% show substantial changes in the wave axis over time (Figure 3E). Thus, β-cell depolarization is initiated at different locations within the islet over time, while the β-cell network location is relatively stable.

The analyses in Figure 3 are focused on the top and bottom 10% of the population. To understand the stability of all β-cells within the 3D network or the 3D wave propagation over time, we ranked every β-cell in the islet by their phase/degree (cellular consistency), as well as the spatial proximity of every β-cell to the center of gravity of the top 10% of the subpopulation (regional consistency) (Figure 4A). We quantified the change in these distributions using a normalized non-parametric, information-theoretic metric termed Kullback–Leibler (KL) divergence (see Methods). If the ranking of high to low degree cell (e.g., A > B > C) for the first oscillation remains the same in the second oscillation, the KL divergence will be 0, indicating the cell ranking is completely predictable between oscillations. Alternatively, if the cell ranking changes between oscillations (A > C > B), the KL divergence will be 1, indicating the cell ranking is completely random (Figure 4B). When examining the consistency of the network, the regional stability was much higher than the cellular stability over time (Figure 4C). In contrast, when examining the wave, the cellular stability was similar to the regional stability (Figure 4D). This analysis of KL divergence supports the previous conclusions that the β-cell network is regionally stable, but the wave can start at different locations. Additionally, because the wave was consistent cellularly, this analysis may imply that the wave is established by cellular properties, whereas the network is emergent.

To investigate whether 2D analysis, as performed in all prior studies (Johnston et al., 2016; Westacott et al., 2017) provides a similar level of robustness as the current 3D analysis, we performed network and wave analyses on a single plane at either ¼- or ½-depth of the z-stack (Figure 5A). Both the 2D and 3D analyses showed that the wave axis changes over time (Figure 5B). The analyses also agreed that the high degree cells are located at the center of the islet (Figure 5C) and that the early phase cells are located at the edge of the islet (Figure 5D). However, when we looked at the regional and cellular consistency of the β-cell network, the 2D analysis at both ¼- and ½-depth of the z-stack showed no difference for regional and cellular consistency (Figure 5E). This result contradicts with the 3D analysis which showed the regional consistency of the β-cell network is significantly more stable than cellular consistency. When analyzing the wave, both 3D and 2D analyses at ¼-depth showed that the cellular consistency is more stable than regional consistency, while the results from a plane at ½-depth showed no difference (Figure 5F). These findings indicate that 2D imaging at different planes of the islet can sometimes skew the results of the heterogeneity analysis.

Glycolysis exerts strong control over the timing of β-cell Ca²⁺ oscillations (Merrins et al., 2022; Tornheim, 1997). Glucokinase, as the ‘glucose sensor’ for the β-cell, controls the input of glucose carbons into glycolysis (Merrins et al., 2022; Matschinsky and Ellerman, 1968), and the downstream action of pyruvate kinase controls membrane depolarization by closing K_ATP channels (Merrins et al., 2022; Lewandowski et al., 2020; Foster et al., 2022; Ho et al., 2023; Quesada et al., 1999; Figure 6A). We applied glucokinase activator (GKa, 50 nM RO-28-1675) and pyruvate kinase activator (PKa, 10 µM TEPP-46) (Lewandowski et al., 2020; Foster et al., 2022) to determine the effects of these enzymes on β-cell subpopulations during glucose-stimulated oscillations.

Since biochemically distinct processes occur during the silent phase (i.e., the electrically silent period when K_ATP channels close and Ca²⁺ remains low) and the active phase (i.e., the electrically active period Ca²⁺ is elevated and secretion occurs) (Merrins et al., 2022), we quantified the duration of each phase along with the oscillation period and duty cycle (the ratio of active phase to full cycle) (Figure 6B). In the presence of vehicle control (0.1% DMSO), the Ca²⁺ duty cycle remained stable, although an outlier in the control group resulted in a small decrease in the silent phase duration (Figure 6C). In 3 of 15 islets, GKa induced a Ca²⁺ plateau (duty cycle = 1.0); out of necessity these islets were removed from the oscillation analysis. In the majority of islets, GKa increased the oscillation period and duty cycle (Figure 6D). The duty cycle increase was driven by an increase in the active phase duration, with no impact on the silent phase, the time when K_ATP channels close (Figure 6D). In contrast with activation of glucokinase, PKa increased the oscillation frequency by reducing the silent and the active phase duration in equal proportions (Figure 6E). The absence of any PKa effect on the duty cycle is expected since fuel input is controlled by glucokinase, whereas silent phase shortening is expected based on the ability of PKa to reduce the time required to close K_ATP channels and depolarize the plasma membrane (Lewandowski et al., 2020; Foster et al., 2022). Thus, a single-cell 3D analysis of β-cell Ca²⁺ oscillation upon GKa and PKa stimulation provides similar conclusions to prior 2D studies of intact islets.

Previous 2D studies have found metabolic differences along the Ca²⁺ wave, as measured by NAD(P)H fluorescence (Westacott et al., 2017). We measured the 3D position of early or late phase cells in response to glucokinase or pyruvate kinase activation. A positional analysis showed that GKa strongly reinforced the islet region corresponding to early and late phase cells, while vehicle and PKa had no discernable effect (Figure 7A). The KL divergence for wave propagation was correspondingly reduced by GKa (Figure 7B), indicating increased consistency, and the wave axis was significantly stabilized by GKa (Figure 7C). Again, PKa had no discernable effect on the KL divergence for wave propagation or wave axis stability, a likely indication that the Ca²⁺ wave origin is primarily, if not exclusively, controlled by glucokinase patterning.

As a second approach, we examined the percentage of early or late phase cells maintained following activation of glucokinase or pyruvate kinase. Early phase cells were maintained to a greater degree upon GKa application, indicating greater consistency, but again showed no change upon vehicle or PKa application (Figure 7D). Late phase cells showed no difference in their maintenance upon any of the treatments (Figure 7E), suggesting that the earliest phase cells drive the consistency of the wave propagation. The time lag between early and late phase cells was increased upon GKa application (Figure 7F), showing that GK activation can enlarge the differences between early and late phase cells.

While metabolic differences have been suggested to underlie functional heterogeneity in the β-cell network (Johnston et al., 2016; Briggs et al., 2023), we observed no changes in the consistency of the islet network upon either GKa or PKa application (Figure 7—figure supplement 1A). Similarly, the consistency of high or low degree cells also did not change upon either GKa or PKa application (Figure 7—figure supplement 1B, C). Collectively, these findings implicate glucokinase as the key determinant of the Ca²⁺ wave in 3D space, whereas metabolic perturbations have little influence on the islet network.

We defined early phase cells as those that depolarized and repolarized first. We also assessed whether the results were consistent for cells that only depolarized first (while ignoring repolarization). Similar to early phase cells, GKa increased the retention of β-cells that depolarized first (Figure 7—figure supplement 2A) and their regional consistency (Figure 7—figure supplement 2B). However, GKa did not influence the wave axis change (Figure 7—figure supplement 2C), indicating that the cells that depolarize first are a more unstable population than those that depolarize and repolarize first.

Ins1-Cre mice (Thorens et al., 2015) (Jax 026801) were crossed with GCaMP6s mice (Jax 028866), a Cre-dependent Ca²⁺ indicator strain, and H2B-mCherry mice, a Cre-dependent nuclear indicator strain (Blum et al., 2014). The resulting Ins1-Cre:ROSA26^{GCaMP6s/H2B-mCherry} mice were genotyped by Transnetyx. Mice were sacrificed by CO₂ asphyxiation followed by cervical dislocation at 12–15 weeks of age, and islets were isolated and cultured as detailed in Ho et al., 2023. All procedures involving animals were approved by the Institutional Animal Care and Use Committees of the William S. Middleton Memorial Veterans Hospital and followed the NIH Guide for the Care and Use of Laboratory Animals.

The stage of a Nikon Ti inverted epifluorescence microscope was replaced with a Mizar TILT M-21N lateral-interference tilted excitation light-sheet generator (Fadero et al., 2018) equipped with an ASI MS-2000 piezo z-stage and Okolab stagetop incubator. The sample chamber consisted of an Ibidi 4-well No. 1.5 glass bottom chamber slide with optically clear sides. Excitation from a Vortran Stradus VeraLase 4-channel (445/488/561/637) single mode fiber-coupled laser and CDRH control box (AVR Optics) was passed through the TILT cylindrical lens to generate a light-sheet with a beam waist of 4.3 μm directly over the objective’s field of view. Similar to a widefield microscope, the axial (z) resolution of the light-sheet microscope is dictated by the NA of the objective (~1.1 μm for our Nikon CFI Apo LWD Lambda S 40XC water immersion objective with an NA of 1.15). Fluorescence emission was passed through an optical beamsplitter (OptoSplit III, 89 North) and collected by an ORCA-Flash4.0 v3 digital CMOS camera (Hamamatsu C13440-20CU) equipped with a PoCL camera link cable. To achieve high-speed triggered acquisition, the laser and piezo z-stage were triggered directly by the camera, which received a single packet of instructions from the NIS-Elements JOBS module via a PCI express NiDAQ card (PCIe-6323, National Instruments). Electronic components (DAQ card, camera, lasers, stage) were linked by a Nikon ‘standard cable’ via a Nikon BNC breakout box; cable assembly is diagrammed in Figure 1—figure supplement 1. Images were streamed to a Dell computer equipped with an Intel Xeon Silver 4214R CPU, 256 GB RAM, XG5 NVMe SSD, and NVIDIA Quadro Pro 8 GB graphics card and Bitplane Imaris Software (Andor).

Reagents were obtained from Sigma-Aldrich unless indicated otherwise. Islets isolated from Ins1-Cre:ROSA26^{GCaMP6s/H2B-mCherry} mice were incubated overnight and loaded into an Ibidi µ-slide 4-well No. 1.5 glass bottom chamber slide and maintained by an Okolab stagetop incubator at 37°C. The bath solution contained, in mM: 135 NaCl, 4.8 KCl, 2.5 CaCl₂, 1.2 MgCl₂, 20 HEPES, 10 glucose, 0.18 glutamine, 0.15 leucine, 0.06 arginine, 0.6 alanine, pH 7.35. Glucokinase activator (50 nM RO-28-1675, Axon), pyruvate kinase activator (10 μM TEPP-46, Calbiochem), and vehicle control (0.1% DMSO) were added as indicated. GCaMP6s (488 nm, 5% power, 50 mW Vortran Stradus Versalase) and H2B-mCherry (561 nm, 20% power, 50 mW) were simultaneously excited and emission was simultaneously collected on a single camera chip using an optical beamsplitter (Optosplit III, 89 North) containing a dichroic mirror (ZT568rdc, Chroma) and emission filters for GCaMP6s (ET525/40, Chroma) and mCherry (ET650/60, Chroma). The exposure time was set to 15 ms in NIS-Elements JOBS, which includes ~10 ms camera integration time and ~5 ms stage dwell time. This was sufficiently fast to image intact islets an axial (z) depth of 132 μm at 2.02 Hz (33 z-steps every 4 μm). Raw NIS-Elements ND2 files were imported into Bitplane Imaris analysis software. The location of each cell was marked using H2B-mCherry nuclear signal and a sphere mask was created based on average β-cell nuclear diameter. Nuclear ROIs were mathematically expanded to 9.3 μm to avoid overlapping cells, which was determined by point-scanning confocal imaging (Briggs et al., 2024). Masks were propagated to all the time points and mean Ca²⁺ levels were used to generate single-cell traces that were exported from Imaris to Microsoft Excel. Quantitative analyses of the β-cell network and Ca²⁺ wave were performed in MATLAB as described below.

To compare β-cell subpopulations over multiple oscillations, we developed a semi-automated oscillation identifier to ensure that the results did not depend on manual identification of oscillation start and end times. First, the approximate time corresponding to the peak of each oscillation was manually identified based on the average islet signal. The time course around each oscillation peak was automatically extracted as seconds before the islet begins depolarization x/2 s after the islet completes repolarization, where x = ¼ oscillation duty cycle. Depolarization and repolarization were then automatically identified using the derivatives of the Ca²⁺ time course and MATLAB’s findpeaks function. All oscillation time courses were manually confirmed. For studies of glycolysis, we ensured that all pre- and post-glycolytic activator treatments had the same number of oscillations. All islets analyzed exhibited slow Ca²⁺ oscillations (period = 6.77 ± 0.36 min).

Network analysis was conducted as described in Briggs et al., 2023, with the caveat that the functional network was recalculated for each oscillation. The correlation threshold was calculated such that the average degree was 7 when averaged over all oscillations (Šterk et al., 2024).

Lagged cross correlation between the normalized Ca²⁺ dynamics of each cell and the islet mean was calculated for each oscillation. Each cell was assigned a cell phase, defined as the time lag that maximized the cross correlation.

Wave axis was defined as the primary axis between the early (top 10%) and late (bottom 10%) of cells in the Ca²⁺ wave. The primary axis was identified using principal component analysis. To ensure the axis was not confounded by spuriously located cells, cells were not included in the analysis if they were greater than 50 μm from the center of gravity (calculated using Euclidean distance) of their respective group (early or late phase). Variability of the wave axis over oscillations was defined as the squared Euclidean distance between each wave axis. To compare across islets of differences sizes, wave axis variability was normalized by the maximum variability possible for each islet. This maximum variability was identified by repeating the wave axis calculation 50,000 times for randomly selected early and late phase cells.

To calculate consistency over oscillations of the entire islet, network and wave analyses were conducted and cells were ranked for each oscillation (i) based on (1) their degree or phase (for cellular consistency) and (2) their Euclidean distance to the center of gravity (for regional consistency) of the high degree or early phase cells. The probability density functions (Pi) of these rankings were calculated using the MATLAB normpdf function. The KL divergence (Kullback and Leibler, 1951) (DKL) between cell ranks for each oscillation was calculated, where (n) is the index of each cell, and (i,j) are indices for each oscillation.(1)DKL(Pi∥Pj)=∑nPilogPiPj,

To compare across islets of differences sizes, we normalized the KL divergence by the maximum possible KL divergence for each islet identified by shuffling the cell ranks and calculating KL divergence 100 times.

Quarter- and half-depth 2D planes were selected from each 3D islet. Cells were included in the 2D plane if their location on the z-axis was within 3 μm of the plane.

Stastistical analysis was conducted using GraphPad PRISM 9.0 software. Significance was tested by first testing normality using Anderson–Darling and Kolmogorov–Smirnov normality tests and then using paired Wilcox tests, Student’s two-tailed t-tests or ANOVA as indicated. p < 0.05 was considered significant and errors signify ± SEM.