Human Insulin Vesicle Dynamics During Pulsatile Secretion

In the past 4 years, we have received 30 preparations of human islets, all of which were isolated from heart-beating donors who had no evidence of prior type 2 diabetes (see acknowledgments). On arrival, these islets were transferred to warm medium (RPMI with 5 mmol/l glucose and 10% fetal bovine serum [FBS] and 5% CO₂, 37°C); within 2 h, the islets were dispersed and maintained according to our standard protocols (15). Rat pancreatic β-cells were prepared and cultured as previously described (14), using procedures approved by the University of Southern California animal use committee. Cultured cells were used within 5 days. (As described below, we report results of studies using islets harvested from only four human donors [one female and three male subjects, aged 32 ± 6.1 years, BMI 28 ± 3.1 kg/m²] whose pancreatic β-cells displayed pulsatile insulin secretion in response to glucose stimulation.)

For inclusion in this study, islets and pancreatic β-cells had to satisfy all of the following morphologic and functional criteria.

Islets needed to have smooth edges, and the distribution of islet sizes had to span both small and large islets, including at least 10% of islets having diameters >300 μm (indicating that the tissue was not over- or underdigested). Moreover, pancreatic β-cells derived from islets needed to have smooth membranes, relatively small nuclei, and few large cytoplasmic organelles (<1 μm).

Pancreatic β-cells had to stick to glass cover slips and express fluorescent cargo proteins when infected with adenovirus. Moreover, >80% of cells had to respond to potassium, and more than two cells from the same preparation had to respond vigorously (secrete >30 vesicles) to glucose stimulation. As described below, only four preparations satisfied all of these criteria.

Insulin vesicles were labeled with one of four different fluorescent cargo proteins. Three were made by fusing insulin C-peptide to one of the following fluorescent proteins: emerald green fluorescent protein (C-emGFP), enhanced green fluorescent protein (C-EGFP), or the bright mutant of dsRed fluorescent protein (C-Bright). The fourth was made by fusing the rodent isoform of islet amyloid polypeptide to enhanced green fluorescent protein (rIAPP-EGFP). Construction of adenovirus containing the gene for either C-emGFP or rIAPP-EGFP has been described previously (12,14). To make adenovirus containing C-EGFP and C-Bright, we used standard molecular biological techniques to modify the C-emGFP construct by replacing emerald green fluorescent protein separately with either enhanced green fluorescent protein or the bright mutant of dsRed fluorescent protein. DNA sequences were confirmed using standard sequencing protocols (University of Pittsburgh/University of Southern California Microchemical Core Facility).

Once plated, cells were incubated overnight and then transduced with adenovirus (5 × 10⁷ plaque-forming units/ml in RPMI 1640 with 10% FBS and 5 mmol/l glucose) containing the gene of interest. After inoculation of the cells with adenovirus (4 h), the medium was removed and replaced with fresh medium (RPMI 1640 with 10% FBS and either 5 or 8 mmol/l glucose for human or rat β-cells, respectively). Cells were imaged 16–48 h later.

Extracellular solution used in imaging experiments was composed of the following (in mmol/l): 136 NaCl, 4.2 KCl, 2.4 CaCl₂, 1.2 MgSO₄, 1.2 KH₂PO₄, 1 l-glutamine, 4 glucose, and 10 HEPES (pH 7.4). In the high-potassium stimulating solution, an equal quantity of KCl replaced NaCl to yield a final potassium concentration of 50 mmol/l. In the high-glucose stimulating solution, we added enough glucose to the basal salt solution so that the final concentration of glucose was 8–20 mmol/l (see figure legends for specific concentrations in each experiment). All solutions were prepared with water from a commercial purification system (Nanopure Infinity; Barnstead Thermolyne, Dubuque, IA). The osmolalities of extracellular and stimulating solutions were matched and fell between 280 and 320 mOsm. All salts for solutions were obtained from Sigma Aldrich (St. Louis, MO) with the highest purity available.

For descriptions of our imaging setups for one- and two-color TIRF microscopy, see our previous publications (14,15). Throughout the imaging experiments, cells were maintained in a warmed solution (between 35 and 37°C) with continuous exchange (∼1 ml/min). Pulled glass pipettes (∼3 μm tip diameter) were positioned near cells for transient application of test solutions.

Periodograms were constructed using standard methods (33). Briefly, to construct a periodogram, each point in an evenly spaced time series of n points was modeled as the sum of sine and cosine terms with amplitudes and frequencies determined according to the following equations:

where m = n/2, ω_k = π*k/m and

The magnitude, S(k), for each period in the periodogram is defined as

It can be shown that the variance of the original data points partitions among the various periods according to the following equation:

The term p(k) is defined by normalizing each of the magnitudes in the periodograms by the sum of the squares about the mean, expressed as

The p(k) values are then used as the basis for statistical testing of significance in randomization trials.

However, before determining whether the magnitude of any of the observed periods was statistically different from the expected distribution of p(k) values, we first determined the likelihood that the overall pattern was significantly different from that expected for a random series. For this step, we used the Kolmogorov-Smirnov test (33) as follows:

The significance levels of D for the observed data were determined by randomization trials. In each trial, the observed data were randomly permuted, and the value of D was calculated. At the end of the N − 1 trials, the value of D for the observed data was grouped with all of the values of D from the randomization trials. The significance level of D was then calculated as 100 × rank/N, where rank refers to the rank of D for the observed data within the set of N values for D. If the significance level of D for the observed data was less than 5%, then we tested individual periods for significance using a similar randomization method. However, when testing individual periods, the statistically significant level was adjusted by the Bonferonni correction (α = 5%/number of periods tested) (33).

In some cases, the data were first compressed using a simple nonoverlapping, box-car summation. In these cases, secretion data from X continuous frames were summed, where X represents the compression factor (see results for typical values of X). All calculations were performed using scripts written for R (34), a freely distributed variant of the statistical programming language S+. To verify our scripts, we reproduced the data presented in Table 11.5 of Manly (33).

To calculate the overall pulse interval reported in Table 1, we included data from all cells that showed pulsatile secretion (i.e., all cells that passed the first part of our two-part test), even those in which we could not identify a significant period of oscillation. Bursts were identified by sharp changes in the cumulative secretion plots. The distribution of pulse intervals was skewed to the right, so we report the median value for this parameter (see Table 1).

All other values are reported as the means ± SE, unless otherwise stated. Parameters in Table 1 were compared using Student's t test with the Welch modification to the degrees of freedom.

At ∼24–48 h after cells were infected with C-Bright adenovirus, they were incubated for 30 min at room temperature in external solution containing 1 μmol/l of the AM (acetoxymethyl) ester form of the calcium indicator Oregon Green 488 BAPTA-1 (Invitrogen, Carlsbad, CA). Thereafter, the cells were washed once with dye-free external solution and then kept in dye-free solution for 30–60 min at 37°C. To measure insulin vesicle secretion and calcium changes simultaneously, images were collected from the red and green channels of our two-color microscope (15). Green (543 nm) light from an He-Ne laser was configured for TIRF excitation of insulin vesicles containing C-Bright, whereas blue light (488 nm) from an Ar⁺ laser was configured for wide-field excitation of the cytoplasmic calcium indicator. Fluorescence images from blue and green excitation were captured separately. To minimize photobleaching of the calcium indicator, we used an asymmetric acquisition scheme: nine images with green excitation light were collected for every one image with blue excitation light. The overall acquisition rate was ∼2.5 Hz. Cross talk between fluorophores was minimal, and no effort was made to correct signals.

For capacitance measurements, pancreatic β-cells were bath perfused at 1 ml/min flow rate in external solution containing (in mmol/l): 138 NaCl, 5.6 KCl, 1.2 MgCl₂, 2.6 CaCl₂, 4 glucose, 10 HEPES, and 10 tetraethylammonium-Cl. Patch pipettes had initial resistances of 3–4 MΩ when filled with internal solution containing the following (in mmol/l): 10 NaCl, 145 Cs-glutamate, 10 HEPES, 1 MgCl₂, 2 Mg-ATP, and 0.3 Na-GTP. Whole-cell recordings were made using the EPC-9 patch-clamp amplifier and Pulse software (Heka, Lambrecht/Pfalz, Germany). For capacitance measurements, we used the “Sine+dc” mode of the “Lock-in” extension in Pulse software. In brief, a 1-kHz frequency sine-wave voltage, with a 40-mV peak-to-peak amplitude, was superimposed on the −70-mV holding potential. Stimulation trains consisted of eight voltage pulses, with each pulse depolarizing the cell membrane from −70 to 0 mV for 500 ms; pulses in the train were separated by 100 ms. Trains were repeated approximately every minute. Capacitance recordings and TIRF imaging were synchronized through transistor-transistor logic signals.

To calculate the diffusion coefficients of single insulin vesicles, we first tracked the positions of single insulin vesicles using commercially available tracking software (Metamorph; Molecular Devices, Sunnyvale, CA). For a vesicle tracked for n frames, we then calculated the vesicle's net displacement over i frames, where i = 1 to n − 1, using frame j as the starting frame, where j = 1 to n − i. For each i, we calculated the average of the square of the vesicle's net displacement. The mean square displacement was then plotted as a function of time, and the data were fitted using a weighted linear regression; points were weighted proportionally with respect to the number of points that contributed to the average. The slope of the fit is directly proportional to the two-dimensional diffusion coefficient (D₂ = slope/4, where D₂ is the two-dimensional diffusion coefficient). Calculations and curve-fittings were performed using scripts written in R.

We prepared primary cultures of pancreatic β-cells from rat and human islets and transduced them with a low titer of a viral vector encoding a fluorescently tagged vesicle cargo protein (see below for specific constructs). Within 48 h, we imaged individual fluorescent vesicles with a TIRF microscope, a microscope that selectively illuminates membrane-proximal vesicles at the base of the cell (13). Each of the cells we selected for study resided within a small cluster of cells. Typically, a single transduced cell contained hundreds of fluorescent vesicles near the plasma membrane, which were scattered about at a modest density (∼0.7 fluorescent vesicles per μm²) (see Table 1). In resting cells, most of the membrane-proximal fluorescent vesicles moved only slightly, jostling about a central point (two-dimensional diffusion coefficient of 4.9 ± 0.5 × 10⁻¹² cm²/s, n = 78 vesicles, eight human pancreatic β-cells) (35,36). We then characterized secretion by stimulating cells with solutions containing elevated concentrations of potassium or glucose.

Figure 1A shows a rat pancreatic β-cell expressing C-emGFP, a probe that consistently leaves insulin vesicles rapidly and completely during a single round of exocytosis (12,14,15). When this cell was transiently stimulated with potassium, it responded quickly (<5 s) (Table 1, “lag”) by secreting ∼50 vesicles, all fusing with the plasma membrane within a span of 10 s (Fig. 1B–E). We repeated this experiment for a number of rat and human pancreatic β-cells (Table 1) and found that potassium consistently stimulated a single short-duration, large-amplitude burst of insulin vesicle fusion. Rat and human pancreatic β-cells secreted insulin vesicles at similar rates (see Table 1, which summarizes characteristics of the burst).

Note that with a TIRF microscope, we are imaging only the footprint of a cell adhering to a glass coverslip—a portion of the plasma membrane encompassing <50% of the cell's total membrane surface area (for a precise estimate, see online Appendix 1 [available at http://dx.doi.org/10.2337/db06-0367]). Previous work has shown that vesicles fuse with the plasma membrane not only where cells contact the glass surface, but also where cells contact each other (15). If we assume that the rate of secretion is uniform over the entire cell, then extrapolation predicts that the number of vesicles secreted by a potassium-stimulated human pancreatic β-cell is ∼200, which represents the size of the readily releasable pool. In rats the estimate of the readily releasable pool size is ∼90.

Next, we investigated insulin vesicle exocytosis stimulated by glucose. Fewer rat and human pancreatic β-cells responded to glucose than to potassium (for human pancreatic β-cells, 45 of 49 cells responded to potassium, and 12 of 42 cells responded to glucose), and several features of the glucose-stimulated response differed from the potassium-stimulated response. First, when cells responded to glucose, there was a significant delay (Table 1, “lag”), in contrast to the immediate response to potassium stimulation. The delay is expected because glucose stimulates secretion through a pathway that is more complicated than the one activated by potassium. Once stimulated, cells secreted more slowly in response to glucose than to potassium (Table 1). Finally, whereas cells responded to potassium with a single, short-duration burst of secretion, they responded to glucose with widely varying temporal patterns of vesicle fusion.

In general, the temporal patterns of secretion were not random and included some cases in which vesicles fused in well-separated bursts, consisting of 10–15 vesicles (in the cell footprint) that all fused within a span of 15 s (Figs. 2 and 3 and Table 1). Extrapolated to the entire cell (as was performed above for potassium stimulation), we estimate that each glucose-stimulated burst consists of ∼70 vesicles in human β-cells and 30 vesicles in rat β-cells. As expected, the bursts of insulin vesicle fusion coincided with transient increases in intracellular calcium (Fig. 4).

A two-step statistical analysis revealed that these bursts of secretion were separated by significant periods ranging from 15 to 45 s (Figs. 2 and 5). See research design and methods for details of the analysis, and see the legend for Fig. 5 for exact values of significant periods. Our analysis also revealed statistically significant periods in cells in which the pulses of vesicle fusion were not obvious on initial inspection of the raw data (Fig. 6).

Before, during, and after stimulation with either potassium or glucose, new fluorescent insulin vesicles arrived at the plasma membrane at ∼1 vesicle/μm² per h—much more slowly than the rate at which fluorescent vesicles were secreted (Table 1). Thus, during vigorous stimulation, there was a net decrease in the number of fluorescent vesicles near the plasma membrane (Fig. 7).

We next investigated the kinetics of refilling the readily releasable pool of vesicles. For these experiments, we monitored secretion from individual cells by simultaneous application of two different techniques: TIRF imaging and patch-clamp measurements of membrane capacitance. TIRF imaging selectively reports exocytosis of vesicles containing fluorescent cargo protein (age <48 h, the time since viral transduction). In contrast, membrane capacitance measurements monitor changes in the membrane surface area, which increases on addition of membrane from any vesicle, whether it contains fluorescent cargo (age <48 h) or not (age >48 h).

To stimulate secretion, we applied a train of membrane-depolarizing voltage steps to cells whose plasma membrane was voltage-clamped in the whole-cell configuration. Depolarization is known to activate voltage-gated calcium channels and to trigger increases in membrane capacitance (37), which should be accompanied by exocytosis of fluorescent vesicles. Indeed, as seen in Fig. 8, individual depolarizations elicit a capacitance step, and during a train of depolarizations, the size of the capacitance steps progressively diminishes until there is no change for the last few stimuli. This indicates that the readily releasable pool of vesicles is exhausted at the end of the train.

To calculate the contribution of fluorescent vesicles to overall secretion during a train of depolarizations, we counted the number of vesicles that disappeared and assumed that each vesicle contributed 2 fF of membrane capacitance (estimates range from 1 to 3.8 fF per vesicle) (38–40). We also assumed that the rate of secretion of fluorescent vesicles in the footprint is representative of the rate across the entire surface of the cell. With these assumptions, we estimate that fluorescent vesicles contributed at least half of the vesicles secreted in response to the first train of depolarizing voltage steps. (If we were to use the estimate of 3.5 fF per vesicle, a large estimate of unitary capacitance [41], then all secretion for the first train would have been of new fluorescent vesicles.)

After the first train of depolarizations, we allowed the cells to rest for ∼1 min and then applied a second train. As shown in Fig. 8, cells had largely restored their ability to secrete, as judged by the reappearance of capacitance changes stimulated by a second train of depolarizing voltage steps. However, the fraction of secretion attributed to fluorescent vesicles decreased significantly for the second train compared with the first train (Fig. 8 and supplemental Appendix 2). In one exceptionally stable cell, we were able to apply four separate trains of stimuli (consecutive trains separated by 1 min), and we observed that the fraction of secretion attributed to fluorescent vesicles decreased progressively for succeeding trains (Fig. 8B, lower panel).

These results demonstrate directly that human pancreatic β-cells contain both young (fluorescent) and old (nonfluorescent) vesicles in their readily releasable pools. Remarkably, the β-cells first secrete their young (fluorescent) vesicles, and only later use their older (nonfluorescent) vesicles to refill their pool of secretion-competent vesicles.

In the current study, we have characterized the dynamics of the functional vesicle pools in individual human pancreatic β-cells. Our work was made possible because we had access to a number of human islet preparations, including islets that were of transplant quality—prepared for use in treating patients with type 1 diabetes (42). In addition, we developed a strategy to study single cells located in small cell clusters because previous work suggests that isolated β-cells function differently from β-cells found within islets (11). We transduced cells using a low titer of virus, and we chose for study only those clusters in which one cell, at most, was infected. In addition, we calculated rates of vesicle arrival and secretion, normalized to a square micron of surface area, so that we could obtain whole-cell estimates for vesicle behavior by multiplying the normalized rates by the total surface area of cells, derived from membrane capacitance measurements.

Although our study includes only a small number of cells, the cells we selected for analysis displayed a number of features suggesting that they were of exceptionally good health. Most importantly, they displayed oscillations of insulin vesicle fusion that coincided with oscillations in intracellular calcium levels. Additional features that corroborated the good health of the cells included their ability to express fluorescent cargo protein when transduced with adenovirus, their tendency to flatten when maintained in culture, and their ability to refill the readily releasable pool of vesicles rapidly when stimulated to secrete (see below). Screening for such cellular characteristics may prove useful as a means of verifying the health of islets for transplantation.

Using potassium stimulation, we obtained estimates of the size of the readily releasable pool of vesicles in pancreatic β-cells. In human β-cells, potassium stimulation rapidly triggered a single, short-duration (∼10 s) burst of exocytosis consisting of ∼55 vesicles per cell footprint, which extrapolates to ∼200 vesicles for the entire cell. This estimate of the readily releasable pool in human β-cells is similar to previous estimates of the readily releasable pool in rodent β-cells (22). Indeed, in parallel experiments, we measured the readily releasable pool of vesicles in rat β-cells to be ∼90 vesicles for the entire cell. These estimates may be slight underestimates because they are based on imaging fluorescent insulin vesicles, which are a subset of all insulin vesicles. However, as we note below, most insulin vesicles undergoing exocytosis for a first stimulus are young fluorescent insulin vesicles, so our estimates are likely to be fairly accurate.

Despite the preponderance of evidence suggesting that individual pancreatic β-cells should secrete insulin in discrete pulses or bursts, there has been little direct evidence of such pulsatile insulin secretion at the level of the single cell and single vesicle. In fact, most previous studies of insulin vesicle fusion at the single-cell level reported that pancreatic β-cells respond to glucose by secreting insulin vesicles at a relatively steady rate (27,30,32).

Here, we have shown directly that individual pancreatic β-cells can and do secrete insulin vesicles in short-duration, small-amplitude bursts that recur regularly and that consist of ∼70 vesicles per burst for human β-cells and ∼30 vesicles per burst for rat β-cells (extrapolated in both cases to the entire cell). As noted above, these estimates may be slight underestimates. Furthermore, individual glucose-stimulated bursts do not fully deplete the readily releasable pool. Instead, they release only about a third of the vesicles in the readily releasable pool. Our observations reveal the single-cell contribution to the oscillations of insulin secretion observed at higher levels of tissue organization.

Intracellular calcium concentration is a key regulator of insulin vesicle secretion (17,43), and in previous publications, oscillations in intracellular calcium concentration have been recorded in both intact islets (44) and in single pancreatic β-cells (45). In isolated β-cells from mice, three classes of intracellular calcium oscillation have been described, with periods ranging from tens of seconds to a few minutes (46). As we have shown (Fig. 4), bursts of insulin vesicle fusion coincide with transient increases in intracellular calcium concentration, and the periods separating the bursts fit within the previously reported range for periods of oscillations in intracellular calcium concentration (46).

Not all studies of clonal β-cells (36) and isolated pancreatic islets (30) have shown that intracellular calcium levels oscillate in response to glucose stimulation. Rather, some studies have reported that intracellular calcium concentration increased suddenly and remained elevated. Although it is not clear why only certain cells responded to glucose with easily identified oscillations, the loss of pulsatile insulin secretion observed during the development of both type 1 (9) and type 2 (10) diabetes suggests that the cells and tissue responding with oscillations are the more healthy cells.

We characterized refilling of the readily releasable pool in primary cultures of human and rat pancreatic β-cells. Contrary to what was reported previously in pancreatic beta tumor cell lines (27), we did not observe a dramatic increase in the rate of fluorescent vesicle arrival at the plasma membrane during a human β-cell's response to either potassium or glucose. In fact, we even observed a net decrease in the overall number of fluorescent vesicles at the plasma membrane because the rates of secretion exceeded by at least an order of magnitude the rates of arrival of new fluorescent insulin vesicles (Table 1 and Fig. 7).

Although secretion stimulation did not dramatically enhance the arrival of fluorescent vesicles at the plasma membrane, we did observe the appearance of fluorescent vesicles at the plasma membrane, with an arrival rate of ∼1 vesicle/μm² per h. This apparently slow rate of arrival is actually more than sufficient to account for the daily rate of insulin secretion at a typical human pancreatic β-cell—over a 24-h period, 24,000 vesicles would be expected to reach the plasma membrane of a single human β-cell (1 vesicle/μm² per h × 1,000 μm² per cell × 24 h). (Note, however, that this rate may be an overestimate of the true rate of vesicle delivery at the membrane because some vesicles appear not to dock firmly and may actually undergo repeated appearances at the membrane.) In comparison, the average daily insulin secretion in healthy humans has been reported (47) to be ∼75 pmol/min, corresponding to secretion of ∼120 insulin vesicles per β-cell per day (assuming 10⁶ islets per human pancreas (42), 10³ β-cells per human islet (37) and 5 × 10⁵ insulin molecules per insulin vesicle (48).

The second observation that explains why fluorescent vesicles are not needed for rapid resupply of the readily releasable pool is that nonfluorescent vesicles are capable of rapid priming to join the readily releasable pool. We were able to study this process by combining TIRF microscopy and membrane capacitance measurements (Fig. 8). For a first train of depolarizations, at least half of the vesicles contributing to secretion were fluorescent—surprising, given that the fluorescent vesicles were vastly outnumbered by nonfluorescent vesicles. This observation corroborates the finding in another type of endocrine cell that new vesicles are preferentially secreted (49). After allowing a minute for replenishment of the readily releasable pool, a second train used to probe the makeup of the readily releasable pool showed that a significant fraction of the resupplied vesicles were nonfluorescent, or older, vesicles.

This result implies that, over short time spans (on the order of several minutes), pancreatic β-cells preferentially secrete their young (labeled) vesicles (∼50% of the vesicles secreted in response to the initial train of depolarizations) (see supplemental Appendix 2) and then recruit older (unlabeled) vesicles to refill the readily releasable pool of insulin vesicles. Over longer time spans, especially between meals when secretion stimulation is much diminished, we expect that new (fluorescent) vesicles will repopulate the membrane-proximal pool, arriving there shortly after they bud from the Golgi body (as has been suggested for other types of endocrine cells) (50).

The rapid and directed movement of new vesicles from the Golgi body to the plasma membrane is distinct from the relatively slow and random movement of vesicles once they are near the plasma membrane (50). Of note, the average diffusion coefficient of insulin vesicles in clonal β-cells (36) is approximately two orders of magnitude greater than the average diffusion coefficient we report here (4.9 ± 0.5 × 10⁻¹² cm²/s) in primary cultures of human pancreatic β-cells. On the other hand, the value we report agrees with previous studies of mobility for large dense-core vesicles in other types of endocrine cells (35,51,52). The greater mobility of insulin vesicles in clonal cells might account for the higher rate of vesicle arrival on glucose or potassium stimulation in comparison with what we observed in primary cultures of human pancreatic β-cells.

An ultrastructural study of fixed pancreatic β-cells suggested that the density of membrane-proximal insulin vesicles increases by ∼50% after 40 min of glucose stimulation (21). It is important to note that electron microscopes visualize all insulin vesicles and cannot distinguish young from old vesicles. The insulin vesicles studied by TIRF microscopy and transient expression of a fluorescent cargo protein are young (<48 h old) and represent a subset of the insulin vesicles studied by the electron microscopy approach. Thus, the results from electron microscopy, TIRF microscopy, and membrane capacitance measurements taken together imply that large numbers of older insulin vesicles move closer to the plasma membrane in a glucose-stimulated pancreatic β-cell after the cell first secretes its youngest vesicles.

In summary, we have demonstrated for the first time that glucose stimulates some pancreatic β-cells to secrete insulin vesicles in discrete bursts that recur regularly. Furthermore, we report that β-cells preferentially secrete first their youngest insulin vesicles and then recruit older insulin vesicles for rapid release.

Funding for this work was provided by generous gifts from the family of Betty Smith Rose and by grants from the U.S. National Institutes of Health provided to D.J.M. (DK10181) and R.H.C. (DK60623) and from the American Heart Association to R.H.C.

We thank Robert Ritzel for helpful comments on this manuscript. Human islets for basic research were made available by the Islet Cell Resource Centers (Washington University, University of Pennsylvania, University of Minnesota, and City of Hope Medical Center [Duarte, CA]) as well as Emory University (Atlanta, GA). We thank all members of the Islet Cell Resource Centers for their time and efforts.