Ca²⁺ Oscillations, Waves, and Networks in Islets From Human Donors With and Without Type 2 Diabetes

A major part of pancreatic islets comprises collectives of nutrient-sensing and insulin-secreting β-cells (1,2). They adapt to changes in metabolic demands with corresponding changes in intracellular stimulus-secretion coupling, and despite their heterogeneity, under normal conditions they respond in a synchronized manner (3–5). This is facilitated by gap junctions that allow for the exchange of ions and small metabolites between β-cells and coordinate electrical activity across the islets in the form of propagating intercellular waves (6–8). Noteworthy, gap-junctional communication is instrumental for appropriate glucose-induced insulin release, whereas its impairment is associated with disruptions in plasma insulin oscillations, similarly to models of metabolic diseases (9–11). Therefore, intercellular coupling and its modulation are increasingly recognized as key to normal islet function (12–14) and potentially viable targets to improve insulin secretion (15,16).

Changes in intracellular calcium concentration ([Ca²⁺]_i) play a central role in intracellular stimulus-secretion coupling, and most of our knowledge on [Ca²⁺]_i signaling in islets derives from mouse models. Although mouse and human islets share key structural and functional features, the intracellular electrical and [Ca²⁺]_i activity in human β-cells is more variable and differs from that of mouse islets (2,14). Furthermore, the collective β-cell rhythmicity is much less explored in human islets. Although there is a similar degree of gap-junctional coupling in mouse and human β-cells (17), the unique cytoarchitecture of human islets and less coherent [Ca²⁺]_i activity may explain why distinguished [Ca²⁺]_i oscillations and waves have been more difficult to detect than in mouse islets (18–22). Advances in optogenetics and computational tools have allowed assessment of multicellular β-cell behavior in mice, revealing that the mediating [Ca²⁺]_i waves are initiated from specific subregions of the islet that are relatively stable in time. These subpopulations are also called pacemaker or leader regions and are defined by local excitability and metabolic profiles (23–27). In human islets, the synchronization patterns are similar but more clustered than in mice and the collective dynamics was found to change in disease (22,26,28,29).

Our increasing awareness of the importance of intercellular communication has stimulated the development of advanced computational approaches. Network science has provided powerful tools to assess collective cellular behavior in islets (22,24,26,29–32). Typically, functional β-cell networks are constructed from thresholded pairwise correlations of [Ca²⁺]_i signals, and the subsequent analyses show that the β-cell collectives organize into efficient, modular, and heterogeneous networks (24,30,33), in which a subset of metabolically highly active hub cells is believed to crucially enhance the communication capacities of intercellular networks (24,34). The complex hub-like connectivity patterns arise from islet-wide [Ca²⁺]_i dynamics and are influenced by cellular heterogeneity, heterogeneous intercellular interactions, and interactions of cells with the environment (35,36). However, the extent to which specialized subpopulations of cells contribute to synchronized network dynamics, persistency, and initiation of intercellular signals, and how these parameters change in type 2 diabetes, is not entirely clear. Particularly, the reports about pacemaker cells and their relation to hub cells are inconsistent and require further clarification (37,38).

Here we combined camera-based [Ca²⁺]_i imaging of human islets from organ donors with and without type 2 diabetes with classical physiological and advanced network analysis to investigate whether and how glucose controls [Ca²⁺]_i oscillations and waves, how the [Ca²⁺]_i dynamics relates to functional connectivity in islets, and how these parameters change in diabetes.

Human islets were isolated in the Alberta Diabetes Institute IsletCore (39) or the Clinical Islet Laboratory (40) at the University of Alberta, with appropriate ethics approval from the University of Alberta Human Research Ethics Board (Pro00013094 and Pro 00001754). Immediately after isolation, islets were cultured at 22°C in CMRL media containing 5 mmol/L glucose, 100 units/mL penicillin/streptomycin, 100 μg/mL BSA, 0.5% insulin-transferrin-selenium (10 mg/L insulin, 5.5 mg/L transferrin, 0.0067 mg/L selenium), and 2 mmol/L l-alanyl-l-glutamine dipeptide (GlutaMAX). Donor characteristics and culture time prior to experiments are indicated in Supplementary Table 1. Insulin secretion from each donor is shown in Supplementary Table 2. Prior to most experiments, islets were handpicked and cultured for an additional overnight in DMEM (Gibco) containing 5.5 mmol/L glucose, 10% FBS, and 5% penicillin/streptomycin. For one donor only (Supplementary Table 1) islets were cultured overnight in RPMI-1640 medium (Life Technologies) containing 7.5 mmol/L glucose, 10% FBS, and 5% penicillin/streptomycin.

Intact human islets were incubated in culture medium containing 5 μmol/L Fluo-4 (Invitrogen) for 60 min. Islets were mounted into a custom recording chamber and perifused with glucose-free RPMI-1640 growth medium supplemented with 10% FBS and 5% penicillin/streptomycin and glucose as indicated. The islets were continuously perifused with extracellular solution at a bath temperature of ∼32°C. Imaging of the upper islet surface was performed with a ZEISS SteREO Discovery V20 upright microscope with a PlanApo S 3.5x mono FWD 16 mm objective and ZEN acquisition software. Fluorescence was excited at 488 nm at 10% light intensity with use of an LED light source. Images were captured with an AxioCam MRm CCD camera at 0.33 Hz at 200 ms exposure and recorded for 1 h. Note that at higher acquisition rates the bleaching would be too intense to ensure such long recordings and that 0.33 Hz is sufficient to track the path of intercellular waves. Twelve-bit 1,388 × 1,040 pixels images were captured, where each pixel was 1.02 μm × 1.02 μm.

The [Ca²⁺]_i dynamics in islets was analyzed off-line with ImageJ and custom-made scripts in Python. After denoising and application of a stabilization algorithm, the F-to-F₀ ratio was computed. The image was then binarized and subdivided into a square mesh, i.e., islet subregions (ISRs), with an edge size of 15 µm, since the discrimination of individual cells was not possible. The mean grayscale values from all frames was exported from each ISR, and the extracted [Ca²⁺]_i traces were processed with a high-pass filter to retrieve baseline trends and possible artifacts. This enabled a binarization of [Ca²⁺]_i signals, which were then used to calculate the signaling parameters, and for the extraction of [Ca²⁺]_i waves. The latter was achieved by means of a space-time cluster analysis, as previously described (41). We calculated the average wave sizes and identified the initiating ISRs. All ISR waves were ranked in accord with their order of activation in a [Ca²⁺]_i wave. The ranks were represented as relative activation ranks; i.e., for each wave they were rescaled to the unit interval, with the lowest initiation rank signifying the first activated ISR in the given wave. The average initiation rank of a given ISR was calculated as the average over all relative initiation ranks from all Ca²⁺ waves, so that low average initiation ranks indicate that the given ISRs often served as wave initiators. How [Ca²⁺]_i signals were processed and analyzed is shown and explained in more detail in Supplementary Fig. 1 and how the [Ca²⁺]_i waves were extracted is demonstrated in Supplementary Video 1.

Filtered [Ca²⁺]_i signals were used for the correlation analysis and the subsequent network construction. Two ISRs were functionally connected if the pairwise correlation coefficient between the i-th and j-th ISR, R_ij, exceeded a given threshold value, as previously described (22,30,32). We used R_ij > 0.7 for the establishment of networks with a suitable number of connections for further diagnosis with conventional network metrics. We calculated the degree of individual ISRs, the average degree, and the relative degree distribution. To quantify network integration, we calculated the relative largest component, which signifies the fraction of ISRs that are either directly or indirectly connected and how many of them are isolated. To quantify functional segregation of islets, we calculated modularity, a measure of how effectively the network can be partitioned into submodules. Higher values of modularity indicate more segregated network structures (32).

Data on donor characteristics and islet secretion data, together with a summary of main analyses on correlation between islet function and donor or islet characteristics, are included in Supplementary Table 1 and Supplementary Table 2. For analyses presented in all figures, except in Fig. 5, we analyzed altogether 40 islets from six normal donors and 36 islets from six donors with diabetes. For the preliminary experiments with exendin-4 and adrenaline presented in Fig. 5, we analyzed two islets from one additional donor with diabetes (Supplementary Table 1). Individual islets typically encompassed between 60 and 330 ISRs, with an average size of ∼150 ISRs, which did not significantly differ between islets from normal donors and donors with diabetes (Supplementary Fig. 2). The Shapiro-Wilk normality test and Brown-Forsythe equal variance test were used before further parametric or nonparametric tests. Differences in parameters between 8 mmol/L and 12 mmol/L glucose and between islet sizes were assessed with the t test or the Mann-Whitney U test. Two-way ANOVA was used to compare the parameter values between islets from control and normal donors in 8 mmol/L and 12 mmol/L glucose. One-way ANOVA or the Kruskal-Wallis test was used to compare activation time, active time, initiation rank, and node degree for ISRs with the upper, intermediate, and lower third parameter values. Post hoc pairwise comparisons were conducted with the Bonferroni t test, Tukey test (for equal group sizes), or Dunn test (for unequal group sizes), with correction for multiple comparisons taken into account. One-way repeated-measures ANOVA was used to compare the effects of stimulation by glucose and exendin-4 with stimulation by glucose only. Statistical significance is indicated with asterisks in figures, as follows: *P < 0.05; **P < 0.01; ***P < 0.001. The exact values are indicated for P values between 0.01 and 0.1.

The data sets generated during or analyzed during the current study are available from the corresponding author on reasonable request. Information on human islet donors used in this study is available in the Alberta Diabetes Institute IsletCore repository (www.isletcore.ca). The raw data used to produce the figures are available from Figshare (DOI: 10.6084/m9.figshare.20031947).

We analyzed [Ca²⁺]_i signals in isolated human islets from donors with and without type 2 diabetes after stimulation with glucose. Our protocol consisted of an initial ∼15 min of substimulatory 3 mmol/L glucose, a subsequent stepwise increase to stimulatory 8 or 12 mmol/L glucose for ∼45 min, and a decrease back to 3 mmol/L glucose. For the analysis of various signaling parameters, we selected ∼30-min intervals of stable [Ca²⁺]_i activity from each islet. Since the spatial resolution of the recording setup did not enable identification of individual cells, the islets were subdivided into ISRs as shown in Fig. 1A and E and Supplementary Fig. 1.

The processed [Ca²⁺]_i activities of selected regions when glucose increased from 3 mmol/L to 8 mmol/L or 12 mmol/L for ∼30 min are shown in Fig. 1B and F, respectively. To provide more quantitative and detailed insight into the [Ca²⁺]_i activity, we show raster plots of binarized activity for all ISRs within an islet (Fig. 1C and G), along with the temporal evolution of the average oscillation frequency and duration (Fig. 1D and H). Both stimulatory glucose levels led to a relatively stable plateau phase with sustained [Ca²⁺]_i oscillations. Pooled data are shown in Figs. 1I–K. An ∼20% higher active time was observed in 12 mmol/L compared with 8 mmol/L glucose, solely due to a higher average frequency. The average duration of oscillations under both stimulation levels was ∼12 s, whereas the average frequencies were 0.22 and 0.27 min⁻¹ for 8 mmol/L and 12 mmol/L, respectively (Fig. 1I and J). Moreover, we characterized ISR activations after switching from 3 mmol/L glucose to stimulatory levels. We considered an ISR to be activated when the first [Ca²⁺]_i oscillation occurred after the rise in glucose. The ISRs that activate among the first are by convention termed first responders (31,33). Higher glucose increased the rate of recruitment by ∼30% (Supplementary Fig. 3).

Plotting of [Ca²⁺]_i oscillations as individual events in space-time demonstrated that [Ca²⁺]_i waves are the mechanism of interregional synchronization (Fig. 2A and D). We then calculated the average sizes of [Ca²⁺]_i waves and the relative SD of initiation ranks of ISRs for all islets and for whole intervals of sustained activity. The latter was used as a measure for the persistency of wave courses and initiator regions. Very small [Ca²⁺]_i events (size <30 s, where the size is defined as the number of activated ISRs multiplied by the time each ISR spent in the active phase, which corresponds to the volume of space-time clusters shown in Fig. 2A and D) were excluded from further analysis, since some short and isolated ISR activations were due to unavoidable noise in the [Ca²⁺]_i signals. Within the remaining [Ca²⁺]_i waves, we ranked all ISRs by the sequence of activations, as illustrated with raster plots in Fig. 2B and E. Under 8 mmol/L glucose, [Ca²⁺]_i waves of very different sizes were observed, whereas under 12 mmol/L glucose, the waves were less heterogeneous and more often encompassed substantial parts of the islet (Fig. 2C). Moreover, [Ca²⁺]_i wave initiation occurred in different ISRs and their courses did not follow the same sequence of activation. The initiator role often changed from one event to another. Comparing the relative SD of initiation ranks indicated that the sequence of activations was more persistent and thus the initiator regions were more stable in 12 mmol/L glucose (Fig. 2F).

In Fig. 2G and H, ISRs are presented with color-coded values of average initiation ranks. It can be observed that for both glucose concentrations the wave-initiating regions are scattered across the islet without an obvious pattern. Moreover, in Fig. 2I and J the average active times are plotted as a function of the relative initiation ranks. Only the values of ISRs that participated in at least 10% of all [Ca²⁺]_i waves were considered, and the active times of all ISRs were normalized with the corresponding average values within the given islet to enable pooling of data from different islets. First, it can be noticed that the relative initiation ranks are scattered between 0.25 and 0.75, whereas ISRs with very low or high initiation ranks were detected only exceptionally. This suggests that the roles of ISRs in the wave initiation process were rather dynamic, with a given ISR sometimes acting as an initiator and sometimes being activated among the last within a wave. Analyzing the relationship between an ISR’s active time and its role in wave initiation revealed that the waves were rarely triggered by ISRs with the lowest active times.

To further characterize the collective activity, we constructed functional connectivity maps for both glucose concentrations. [Ca²⁺]_i signals from all ISRs were statistically compared in a pairwise manner to build correlation matrices (Fig. 3B and E), which were then thresholded to obtain functional networks. For both stimulation levels, the network architectures appeared rather regular and lattice like (Fig. 3C and F), in contrast to β-cell networks inferred from mouse islets, which exhibit more long-range connections (24,26,30–32). Comparing networks in 8 mmol/L and 12 mmol/L glucose showed that under higher stimulation, the networks were denser (higher average degree [Fig. 3H]), more cohesive (higher relative largest components [Fig. 3I]), and less fragmented (lower modularity [Fig. 3J]). These properties reflect the fact that the [Ca²⁺]_i waves provoked by 12 mmol/L glucose typically encompassed larger parts of an islet, whereas in 8 mmol/L glucose, they were typically limited to smaller parts of an islet. The node degree distributions in Fig. 3G show that under both glucose concentrations, the islet networks were rather heterogeneous. A relatively small fraction of ISRs existed that were very well connected, with so-called hub areas being functionally correlated to up to 40% of all other ISRs.

To systematically compare ISRs with different node degree values with respect to their active times and wave initiation ranks, we pooled data from altogether 2,697 regions for 8 mmol/L glucose and 3,205 regions for 12 mmol/L glucose and divided them into three groups based on node degrees. In Fig. 3K–N the relationships between the above parameters are shown for all ISRs with semitransparent dots, whereas the three box plots indicate the least connected, the intermediately connected, and the most connected ISRs. Note that the node degrees and active times were normalized with the average values of the given islet to enable the pooling of ISRs from islets of different sizes and average activities. Evidently, for both glucose concentrations, the ISRs with the lowest of active times had the least connections, while the ISRs with the most active times harbored the most connections (Fig. 3K and L). The differences between active times between the least and most connected ISRs in 8 mmol/L and 12 mmol/L glucose were ∼40% and 30%, respectively. In contrast, no clear relationship between the average initiation rank and node degree could be found (Fig. 3M and N). The values of initiation ranks differed by <2.5% from the average, irrespective of node degree and glucose level. This corroborates the findings in Fig. 2 that the initiation of [Ca²⁺]_i waves was dynamic and occurred in different parts of an islet, in this case by either weakly or highly functionally connected ISRs with roughly the same probability.

Figure 4A–F shows the activity of two islets from the same donor with type 2 diabetes. While the [Ca²⁺]_i responses in one islet are similar to those in normal islets, in the other there is less activity, the [Ca²⁺]_i waves are smaller, and the functional networks are sparser and much more segregated. To summarize the general behavior of islets from donors with diabetes, we show data for all islets examined in Fig. 4G–N. For facilitation of comparison with control islets, box plots showing the values for normal islets are also displayed in a thinner and brighter form. Notably, for most parameters, the islets from donors with type 2 diabetes exhibited a glucose dependency similar to that of normal islets. However, in both glucose concentrations they were on average significantly less active (by 30–35%), predominantly due to a lower frequency of oscillations (Fig. 4G–I). Interestingly, the decrease in average active time could be attributed in part to a much higher proportion of ISRs with a very weak activity. More specifically, while the islets from control donors exhibited a normal-like distribution of ISR activity, in islets from donors with diabetes the distribution was right skewed with a higher ratio of weakly active ISRs (Fig. 4–Q), implying that the functioning of some regions was much more impaired than of others. The islets from donors with diabetes also responded to stimulatory glucose levels more slowly, especially to 8 mmol/L (Supplementary Fig. 3). Moreover, their [Ca²⁺]_i waves were smaller and more localized compared with normal responses in both glucose concentrations (Fig. 4J) and the wave initiation sites and their courses were more erratic, as the relative SD in initiation ranks was significantly higher (Fig. 4K).

The observed differences in the spatiotemporal [Ca²⁺]_i activity were accompanied by changes in functional connectivity. The degree distributions shown in Fig. 4R indicate that islets from donors with diabetes lack very well-connected ISRs, i.e., hubs. Due to fewer number of large [Ca²⁺]_i waves, there was no long-range coordination of intercellular signals, and consequently hub ISRs were less pronounced. In both glucose concentrations the average degree of highly connected ISRs, i.e., regions connected to at least 10% of all ISRs in the given islet, was significantly lower in islets from donors with type 2 diabetes (Fig. 4S and T), while the average node degree over all ISRs was not significantly lower (Fig. 4L). Since the functional networks in human islets exhibited a rather lattice-like structure, the average number of connections per ISR reflects essentially the local synchronization level, which was not altered in diabetes. Moreover, no significant differences were found between the relative largest components of islets from donors with and without diabetes, but there was a much higher fraction of islets with a very low level of cohesion in both glucose concentrations in the islets from donors with diabetes (Fig. 4M). Resulting from a lack of global [Ca²⁺]_i waves, the islets from diabetic donors were also functionally more segregated, although the difference seemed to be more pronounced under 8 mmol/L glucose (Fig. 4N). Finally, the relationships of active time, initiation ranks, and number of functional connections in islets from donors with diabetes were very similar to what was observed in islets from donors without diabetes (Supplementary Fig. 4).

Finally, we set out to check whether the loss in [Ca²⁺]_i activity observed in islets from a donor with type 2 diabetes could at least partly be improved by incretin hormone receptor signaling and to quantify the extent to which the less active ISRs correspond to regions occupied by non-β-cells. To this purpose, in two islets from a donor with type 2 diabetes, we used a functional protocol where stimulation by glucose was followed by stimulation by glucose and the glucagon-like peptide 1 (GLP-1) receptor agonist exendin-4, low glucose, low glucose and the K_ATP channel opener diazoxide, and finally adrenaline, as indicated in Fig. 5. Evidently, GLP-1 receptor activation increased the average active time and enhanced interconnectedness of ISRs, particularly due to a higher proportion of hub ISRs; however, the sizes of [Ca²⁺]_i waves did not change significantly on average, but the sizes of the largest waves increased, and this probably accounts for the increase in functional connections (Fig. 5B–D). Exendin-4 elicited qualitatively the same effect in an islet from a control subject (see Supplementary Fig. 5). With regard to putative non-β ISRs, approximately one-sixth of ISRs responded to stimulation by adrenaline with an increase in [Ca²⁺]_i, as would be expected of α-cells. These regions were found throughout the islets clustered in groups and were much less active under stimulation with glucose than the ISRs that did not respond to adrenaline with an increase in [Ca²⁺]_i activity. They also formed much less functional connections. It is worth pointing out that the putative β-cell ISRs would be expected to respond to adrenaline by decreasing their [Ca²⁺]_i activity; however, in our protocol this was achieved already by low glucose or the combination of low glucose and diazoxide.

In the current study we aimed to advance our understanding of spatiotemporal [Ca²⁺]_i patterns in human islets by systematically quantifying the glucose-dependent collective activity through the use of intensive computational approaches. As recently reviewed (14), previous studies involving [Ca²⁺]_i imaging in human islets have included reports of globally (19,20) or locally (21) synchronized [Ca²⁺]_i oscillations, the presence of some [Ca²⁺]_i oscillations without quantifying them or their synchronicity in detail (17,22,24,42), or no clearly discernible oscillations (18). Two studies have included evaluation of [Ca²⁺]_i oscillations in human islets quantitatively, one with use of a single stimulatory glucose concentration (11 mmol/L) (21) and the second with two different concentrations (11 mmol/L and 16.7 mmol/L) (20). The frequencies of oscillations in our study are comparable with those of these previous studies and are in the range of 0.1–1.0 min⁻¹. Taking into account the increase in frequency of oscillations from 8 mmol/L to 12 mmol/L glucose in our study and the increase in duration from 11 mmol/L to 16.7 mmol/L glucose in the study by Martín and Soria (20), it seems possible that human islets respond to increasing glucose by first increasing the frequency and then the duration of oscillations, which would qualitatively resemble the behavior in some mouse islets (33). Moreover, the range of oscillation frequencies observed in our study is compatible with the period of electrical, metabolic, and secretory oscillations measured in isolated human islets (43–45), as well as with the period of insulin pulses in vivo (10), which are all in the range of 3–5 min, i.e., frequencies of 0.20–0.33 min⁻¹. It should also be pointed out that further studies, possibly with higher temporal resolution, are required to clarify whether human islets also display faster [Ca²⁺]_i oscillations, like mouse islets do, and to more precisely quantify the active time (10,33). Finally, we assumed that most ISRs responding to glucose would be β-cells and the majority of weakly active ISRs that did not participate in waves and were disconnected in functional networks were non-β-cells. In future studies, identification of cells with use of immunofluorescence could help specifically identify smaller ISRs corresponding to individual β and non-β-cells. However, the functional discrimination protocol preliminarily used in an islet was able to identify putative ISRs containing α-cells, the distribution and proportion of which correspond with findings of previous morphological studies in human islets (1,18,46).

To our knowledge, no previous study quantified the characteristics of [Ca²⁺]_i waves and functional networks in human islets to the extent that we could make a comparison with our findings. We wish to point out that the velocity of waves observed in this study (∼10 µm/s), is approximately an order of magnitude slower than the velocity of fast calcium waves in mouse islets (∼ 100 µm/s) (6–8), but comparable with slow waves in mouse islets, where lags of several seconds or tens of seconds could be observed between beginnings of slow oscillations (31). The observed behavior of [Ca²⁺]_i waves and functional networks, as well as their response to increasing glucose, can be attributed to a relatively less pronounced cellular heterogeneity at elevated glucose, which in turn diminished the spatiotemporal variations in excitability, making the courses of [Ca²⁺]_i waves more stable. Moreover, the distribution of connections between ISRs was found to obey an exponential function, indicating a lower heterogeneity of human islet functional networks compared with mouse islets, in which the degree distribution typically follows a truncated power law (24,29,30,33). This discrepancy can at least partly be attributed to structural and functional differences between mouse and human islets and a somewhat different nature of [Ca²⁺]_i wave organization (1,14,18,46). More specifically, [Ca²⁺]_i waves in mouse islets commonly spread over the whole islet (6,27), whereas in human islets we found them to be much more heterogeneous in size and often encompass only smaller regions (Fig. 2). This also implies a lower network efficiency and a less pronounced small-world character of the human islets. Furthermore, together with donor and method heterogeneity (47,48), this may also help explain why in previous studies investigators were often unable to detect [Ca²⁺]_i oscillations and waves in human islets and why the existing estimates of their characteristics differ (14).

We gave particular emphasis to the identification of specialized functional subpopulations of islet regions, their potential overlap, and their characteristics in terms of [Ca²⁺]_i activity. The heterogeneous distributions of functional connections support the existence of so-called hub regions. Although the extent to which they coordinate the collective response and the temporal persistence of their role are not yet clear (37,49), our finding that hub ISRs have the highest active times is consistent with recent findings in mouse islets (33) and the view that hub cells are metabolically highly active and participate in most of the [Ca²⁺]_i waves (4,24,25,34). Furthermore, previous studies in mouse islets have shown that [Ca²⁺]_i waves originate from pacemaker regions with elevated excitability (6,27), whereby the pacemaker cells were identified as metabolically less active with higher intrinsic frequencies (25). In our analyses we did not identify a clear relation with the activity of ISRs, except that the regions with low active times seldom initiated the waves. The triggering of waves was rather changeable and switched between areas scattered throughout the islets, especially under lower glucose, where intercellular variability was more pronounced. These concepts are in parallel with the idea that presence of pacemakers in networks of excitable cells reflects emergent dynamical behavior (50). Most importantly, we noticed no clear relationship between wave initiation and hub ISRs; i.e., the waves could be triggered from either weakly or highly functionally connected ISRs. Apparently, hub and wave-initiating regions are both genuine features of human islets, but their roles do not necessarily overlap and should not be considered equal. Importantly, these two populations were also suggested to differ from the so-called first responders, i.e., cells that lead the first transient phase of [Ca²⁺]_i elevation (31,33). In our analyses presented in Supplementary Fig. 6 we identified that the first responding ISRs tended to exhibit greater-than-average values of active time during the plateau phase of oscillations, especially for the lower glucose concentration. The first-responding ISRs were not exceptional in terms of the number of functional connections, but there was a slight tendency for them to be more likely among the wave initiators.

To our knowledge, no studies directly addressed how [Ca²⁺]_i oscillations, waves, and coordinated network activity change in diabetes, yet a body of circumstantial evidence suggests that impaired coupling plays an important role (13,14) and that a decline in coordinated islet activity is observed in diabetes conditions (22,26,28). Noteworthy, one other study demonstrates regular calcium oscillations in diabetic islets, with frequencies similar to those in our study (42). Additionally, clear secretory oscillations with similar frequencies have been reported in islets isolated from T2D donors (44) and insulin levels seem to oscillate at a similar frequency, but reduced amplitude, in vivo in T2D (10). In our analyses on islets from donors with type 2 diabetes we found that the active time is decreased, mostly due to decreases in frequency of oscillations and heterogeneous dysfunction of some regions within the islets, that the [Ca²⁺]_i waves are smaller, and that the islet functional networks are more segregated, compared with normal islets. The decreased frequency of [Ca²⁺]_i oscillations in our hands is at odds with the view that the frequency of oscillations in insulin does not change in type 2 diabetes and is worth pursuing further in future studies. However, it has to be kept in mind that in vivo oscillations are a result of interislet synchronization mediated by intrapancreatic nerves or by systemic feedback mechanisms, and changes in frequency may be a consequence of defects in these mechanisms or of changes in islet entrainability rather than in intrinsic islet frequencies (10). In contrast, our data are consistent with the finding of decreased insulin oscillation amplitudes in T2D. The shorter oscillation duration namely means less active time per oscillation and smaller wave sizes mean fewer active cells per oscillation and consequently probably less insulin secreted per oscillation.

The sparser functional networks observed in islets from donors with diabetes could be intuitively associated with lower activity; however, we argue that this is not the main reason. Rather, due to the presence of inactive regions, there is a loss of globally synchronized oscillations and therefore fewer functional connections in hub regions can to a large extent be explained by the lack of global [Ca²⁺]_i waves. This could be a consequence of heterogeneously decreased gap-junctional coupling (11,17,22,28,36). These results also imply that hub ISRs are not only often part of intercellular waves but also seem to play a vital role by mediating these waves across the multicellular network, even though they principally do not serve as wave initiators. Moreover, theoretically it has been shown that the overall activity and functional connectivity are intimately linked with [Ca²⁺]_i wave propagation in heterogeneously coupled systems (36). Importantly, we also found that GLP-1 receptor stimulation is able to increase the active time and densify the functional networks not only in control islets but to a similar extent also in islets from a donor with type 2 diabetes (Fig. 5 and Supplementary Fig. 5). Although preliminary, this finding is consistent with other recent reports on islets and the therapeutic effects of GLP-1 receptor agonism in vivo (15,22).

Finally, it should be noted that we observed a rather large variability in responses of islets from both normal donors and donors with diabetes, with some of islets of the latter exhibiting striking pathophysiological changes and others behaving rather normally. This supports the view that heterogeneity is an important aspect of normal islet functioning (4) and influences the susceptibility of different islets toward diabetogenic insults (51). Therefore, it should be considered as an important aspect in conducting research on human islets in general, also beyond [Ca²⁺]_i signalization and intercellular connectivity. In this study, we used [Ca²⁺]_i as the only proxy for assessing stimulus-secretion and intercellular coupling, and we argue that further studies are needed to understand the mechanistic relationship between the observed changes in [Ca²⁺]_i signals and other functional parameters, such as changes in metabolism, electrical activity, insulin secretion, and paracrine interactions, to name only a few (29,52,53). Future studies shall also help us better understand the temporal evolution of the described changes in [Ca²⁺]_i signals during progression of diabetes and their relationship with previously described pathophysiological factors characterizing β-cell dysfunction (54).

Acknowledgments. The authors thank the Human Organ Procurement and Exchange Program (HOPE) and Trillium Gift of Life Network (TGLN) for their work in procuring human donor pancreas for research. The authors also thank Drs. James Shapiro and Tatsuya Kin (University of Alberta Clinical Islet Program) for contributing some islet preparations for this study and James Lyon (Alberta Diabetes IsletCore, University of Alberta) for excellent technical assistance in islet isolation. Finally, the authors especially thank the organ donors and their families for their kind gift in support of diabetes research and wish to dedicate this article to the late Mathias Braun, a pioneer of human islet electrophysiology.

Funding. H.L. was supported by a Sino-Canadian Studentship from Shantou University. Research was funded in part by a Human Islet Research Network (HIRN) grant from the National Institutes of Health (U01DK120447). P.E.M. holds the Canada Research Chair in Islet Biology. The authors also acknowledge support of the Slovenian Research Agency (research core funding nos. P3-0396 and I0-0029, as well as research projects nos. J1-2457, N3-0133, J3-9289, and J3-3077).

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

Author Contributions. M.G., R.Y.-D., H.L., P.E.M., and A.S. designed the study and contributed to study conceptualization. R.Y.-D. and H.L. performed the experiments. M.G. analyzed data. M.G., R.Y.-D., P.E.M., and A.S. contributed to the interpretation of the results. M.G. wrote the original draft of the manuscript and prepared the figures. P.E.M. and A.S. contributed to discussion and reviewed and edited the manuscript. P.E.M. and A.S. were responsible for study supervision, funding acquisition, and project administration. A.S. 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 in abstract form at the HIRN 2019 Annual Investigator Meeting, Washington, DC, 28 April 2019 to 1 May 2019, and 2020 Annual Investigator Meeting, 30 September 2020 to 1 October 2020.