Dynamin 2 regulates biphasic insulin secretion and plasma glucose homeostasis

Proper regulation of membrane trafficking balances material and signal exchange and is fundamental to cellular functions. Similar to nerve terminals, in which endocytosis supports high rates of synaptic vesicle recycling (1, 2), neuroendocrine cells undergo vigorous membrane trafficking to regulate large dense-core vesicle (LDCV) release and cellular functions. However, the mechanisms by which endocytosis influences secretory function are unclear, and the molecular nature of endocytosis in these cells remains poorly understood. Pancreatic β cells are the only cell type that releases insulin in humans. Failure of insulin secretion due to β cell loss or functional decline causes type 1 diabetes and type 2 diabetes (T2D), respectively (3). Declines in insulin secretion may arise from the defects of single or multiple steps in the insulin granule trafficking cycle, including granule biogenesis from the trans-Golgi network (TGN), subsequent maturation, recruitment to the plasma membrane (PM), exocytosis, endocytosis, and endosome-to-TGN traffic. It is imperative to understand how regulated membrane trafficking controls insulin secretion.

In response to continuous exposure to high concentrations of glucose, insulin release proceeds in two distinct temporal phases in both humans (4) and rodents (5, 6), including a fast, transient first phase and a slow, sustained second phase. In patients with T2D, the second phase is strongly reduced and the first phase is nearly abolished (7). Multiple factors contribute to the biphasic nature of glucose-stimulated insulin secretion (GSIS), including distinct pools of insulin granules, metabolic signaling, and actin cytoskeleton remodeling (8–12). The granules docked to the PM, also known as the readily releasable pool (RRP) granules, mainly contribute to the first phase. Direct total internal reflection fluorescence (TIRF) imaging suggests that the granules near the PM participate in the first phase and the “newcomer” granules mobilized from a reserve pool located a distance away from the PM produce the second phase of GSIS (12). Moreover, increasing evidence supports an emerging model in which the actin cytoskeleton plays a critical role in biphasic GSIS (10, 11) through granule mobilization and recruitment from the reserve pool to the PM. Many actin remodeling molecules, such as the Rho GTPase family proteins Cdc42 (13) and RAC1 (14), PAK1 (15), N-WASP (16), and PPARβ/δ (17), selectively regulate the second phase of GSIS, and perturbations of microtubules (18) and kinesin-1 (19) disrupt the second phase.

Upon glucose stimulation, β cells rapidly uptake glucose and generate ATP, which closes K_(ATP) channels and depolarizes the PM, and the subsequent intracellular Ca²⁺ increase triggers insulin granule exocytosis (3, 20). After exocytosis, the membrane components of insulin granules (such as lipids, v-SNAREs, synaptotagmin, transporters, and ATPases, etc.) are internalized and transported to the TGN in order to complete the cycle of granule membrane trafficking. The exocytosis-endocytosis coupling of insulin granules was first demonstrated by a pioneering electron microscopy (EM) study in 1973 (21). Since then, the study of β cell endocytosis has been largely stalled, in sharp contrast to the remarkable progress made in characterizing insulin exocytosis at both molecular and cellular levels. Individual insulin granules have two fates after fusion with the PM: pinching off intact following a transient fusion pore opening (termed “kiss and run,” refs. 22, 23, and/or “cavicapture,” ref. 24) or being retrieved through undefined mechanisms after a full collapse onto the PM. Membrane capacitance studies in β cells demonstrate heterogeneous endocytosis kinetics and different endocytosis modulation by Ca²⁺ (25, 26), IP₆ (27), G proteins (28), and actin (26). However, the molecular nature of endocytosis machinery in β cells is poorly understood, and how endocytosis regulates β cell function and glucose homeostasis in vivo is unknown.

Dynamin was originally identified as a microtubule-binding GTPase (29) and later identified as a pinchase to free endocytic vesicles from the PM (refs. 30, 31, and for review, see refs. 32, 33). In shibire mutants of Drosophila, nearly empty nerve terminals at a nonpermissive temperature after stimulation (34) suggest an essential role of dynamin in synaptic vesicle reformation. In mammals, dynamin is encoded by 3 genes (32): Dnm1, Dnm2, and Dnm3. Dynamin 1 is highly expressed in the nervous system, dynamin 2 is ubiquitous in all tissues, and dynamin 3 is found primarily in neuronal tissues and at low levels in testis (35, 36). At nerve terminals, dynamin 1 and 3 contribute to synaptic vesicle reformation and neurotransmission (36–40); synapses lacking dynamin 1 (and dynamin 3) have fewer vesicles and abundant clathrin-coated profiles (CCPs) accumulated on the PM (36, 38, 40). In nonneuronal cells, dynamin isoforms appear cell-type specific and may play distinct roles in multiple endocytosis pathways (32, 35, 41). Recent studies further highlighted the diverse functions of dynamin in other biological processes (42–44).

Here, we studied dynamin 2 function in insulin secretion and glucose homeostasis in vivo. We generated tamoxifen-inducible β cell–specific Dnm2 KO mice. The KO mice developed significant glucose intolerance and a pronounced reduction of the second phase of GSIS. β Cells lacking dynamin 2 contained abundant insulin granules, as did controls, but the clathrin-mediated endocytosis (CME) was strongly impaired at the later stage of membrane retrievals. Moreover, KO cells exhibited striking reorganization and enhancement in cortical F-actin, which can be fully rescued by reexpressing dynamin in these cells; TIRF imaging at single-cell levels revealed selective reduction of insulin secretion and granule recruitment at the later stage of glucose stimulation. These data provide direct genetic evidence that dynamin 2 regulates biphasic insulin secretion and glucose homeostasis in mammals, implying a potential link between dynamin 2–mediated membrane trafficking and the pathogenesis of diabetes mellitus.

Pancreatic β cells express multiple dynamin isoforms. Western blots of purified pancreatic islets revealed that dynamin 2 was abundant and dynamin 1 was not detectable (Figure 1, A–C, and Supplemental Figure 1, A and B; supplemental material available online with this article; doi:10.1172/JCI80652DS1). Interestingly, we found fairly rich levels of dynamin 3 (Figure 1C), which is reported primarily in neuronal tissues and at low levels in testis (35). The antibody specificity was verified using cell lysates from tissues with appropriate dynamin genotypes as negative and positive controls. Immunohistochemistry of pancreatic sections showed strong dynamin 2 (but not dynamin 1) fluorescence in islets (Figure 1, D and E) and low levels of dynamin 3 distributed equally in islets and exocrine areas of pancreas (Figure 1F). Immunocytochemistry of isolated individual β cells in cultures confirmed the coexpression of dynamin 2 and 3 in β cells (Supplemental Figure 1C). The dynamin 2 fluorescence was punctate at the PM (Figure 1G) but was more diffuse throughout cytosol, with a slightly enhanced signal in some perinuclear regions (Supplemental Figure 2, B and C). Dynamin 2 signal was largely separated from insulin granules, with only a limited number of dynamin 2 puncta colocalized with granules (Figure 1G).

Figure 1

Dynamin isoforms in mouse pancreatic islets. (A–C) Western blots for dynamin 1, 2, and 3 in purified mouse islet lysate (lane 1). The specificity of antibodies was verified using cell lysates with the indicated dynamin genotype as positive and negative controls (lanes 2 and 3). 110–150 islets were used for each lane. Each panel shows representative results from 3 to 5 independent experiments. MEF, mouse embryo fibroblasts. (D–F) Immunofluorescence of dynamin 1, 2, and 3 in mouse pancreatic islets. (G) Subcellular distribution of dynamin 2 in β cells (bottom optical section using spinning disk confocal microscopy views). Arrows indicate dynamin 2 puncta colocalized with insulin granules. Scale bars: 50 μm (D–F); 2 μm (G).

Temporally controlled dynamin 2 deletion in pancreatic β cells impairs glucose homeostasis in vivo. To study dynamin 2 function in β cells at both cellular and whole organism levels, we generated the Dnm2 KO mouse model, in which dynamin deletion was temporally controlled and limited in β cells. To this end, we crossed Dnm2^fl/fl mice (45) with Tg(Ins1-Cre/ERT)1Lphi (also termed MIP1-CreERT) transgenic mice (46) that selectively express Cre estrogen receptor fusion proteins driven by a mouse insulin-1 promoter (Ins1) in β cells. The latter has improved β cell selectivity (no ectopic expression in hypothalamus in the brain) compared with other β cell Cre or CreERT lines (47). After tamoxifen treatment, Cre recombinase translocated from the cytosol into nuclei and deleted the Dnm2 gene in β cells (Figure 2A). Dnm2 KO mice were viable and not grossly distinguishable from their littermate controls. Western blotting and immunohistochemistry experiments (Figure 2, B and C) demonstrated markedly lower levels of dynamin 2 in KO islets than in control islets, indicating efficient dynamin 2 depletion. The remaining fluorescence was presumably from the non–β cells in pancreatic islets, such as α, δ, and pancreatic polypeptide cells, or weak background fluorescence from β cells. We often saw some α cells in the center of KO islets, in contrast to the periphery distribution of α cells in control islets (Supplemental Figure 3). The dynamin 2 deletion, specifically in β cells but not in α cells, was confirmed (Supplemental Figure 4), consistent with the β cell–specific expression of Cre recombinase in MIP1-CreERT mice (46, 47).

Figure 2

Dynamin 2 is required to maintain normal plasma glucose homeostasis in vivo. (A) Scheme of temporally controlled, β cell–specific Dnm2 KO. Tamoxifen triggers the translocation of Cre recombinase driven by the MIP promoter into the nucleus and deletes Dnm2 in β cells. (B and C) Efficient dynamin 2 deletion in Dnm2 KO β cells demonstrated by (B) Western blots and (C) immunohistochemistry. Scale bar: 20 μm. (D) Similar body weight between control and KO mice (n = 13 for each group). (E) Significant glucose intolerance in Dnm2 KO mice compared with different controls (n = 11–13 mice for each group). (F) AUC of GTT curves (P = 0.039, 1-way ANOVA with Tukey honest significant difference). (G) ITTs (n = 8–9 mice for each group). (H–K) Body weight, GTT, and ITT after high-fat diet feeding (n = 6–8 mice for each group). (L and M) Similar body weight but significant glucose intolerance in Dnm2 KO mice following 3 glucose injections (every 30 minutes as indicated by arrows); n = 6 mice for each group except Dnm2^fl/fl plus tamoxifen (n = 5). (N) AUC comparison of GTT curves in M (P = 0.003, 1-way ANOVA). Unpaired 2-tailed Student’s t tests were used unless specified. T, tamoxifen; V, vehicle. *P < 0.05, **P < 0.01, ***P < 0.005.

We first examined how dynamin 2 deletion influences glucose homeostasis at the system level in vivo. The Dnm2 KO mice exhibited significant glucose intolerance following glucose administration (i.p.) compared with their littermate controls without MIP1-CreERT (Figure 2E), in spite of their similar body weights (Figure 2D). Further experiments in vehicle-treated Dnm2^fl/fl MIP1-CreERT mice and Dnm2^fl/fl mice showed no significant difference between different control groups. The AUC of glucose levels in KO mice was significantly higher than different controls (Figure 2F; P = 0.039, 1-way ANOVA). These data demonstrated the glucose intolerance observed in Dnm2 KO mice was caused by a direct dynamin 2 deletion in β cells rather than other nonspecific factors, such as gene background or tamoxifen treatment itself. Moreover, the glucose intolerance was further exacerbated by metabolic stress induced by high-fat feeding for 3 months (Figure 2, I and J). However, KO mice had similar body weight compared with that of their littermate controls (Figure 2, D and H), and the insulin sensitivity in KO mice remained normal in both physiologically and metabolically stressed conditions (Figure 2, G and K).

To further verify glucose homeostasis deficiency in dynamin 2 KO mice, we monitored plasma glucose responses following 3 glucose injections (i.p., every 30 minutes) (Figure 2, L–N). Different control groups had a comparable body weights and responded similarly to glucose stimulation. They maintained a plateau level of glucose, without further significant increase at 30 minutes after the first injection (Figure 2M), indicating a strong capacity of glucose handling in vivo to stabilize healthy levels of plasma glucose, even after strong glucose challenge. KO mice had significantly higher levels of plasma glucose than all control groups after injections (Figure 2, M and N), despite the fact that they responded with a similar plateau-increase pattern. This suggests a defect in continuous insulin secretion and glucose handling capacity during the prolonged glucose stimulation. The AUC in KO mice was significantly higher than that in controls (Figure 2N, P = 0.003, 1-way ANOVA). In summary, these in vivo experiments revealed a specific role of dynamin 2 in glucose homeostasis at the system level; the intact insulin sensitivity excludes a potential defect in insulin signaling in peripheral tissues of KO mice, pointing to insufficient insulin release from pancreatic β cells.

Dynamin 2 is essential for preserving normal biphasic insulin release. Next, we examined dynamin 2 function in biphasic GSIS by direct secretion measurement from pancreatic islets. The continuous glucose (20 mM) stimulation induced a typical biphasic pattern of insulin release in control islets (Figure 3A). The first phase peaked at approximately 5 minutes after stimulation and deceased within the following 5~10 minutes; the second phase was relatively slow and long lasting. Such biphasic GSIS in control islets closely mimicked insulin release in vivo, as demonstrated in awake mice under hyperglycemic clamp (6). However, GSIS in KO mice was markedly decreased in the second phase (Figure 3A). Interestingly, the first phase of GSIS in KO islets reached a comparable peak amplitude at a time similar to that in controls and then decreased faster. Insulin secretion induced by 30 mM KCl showed similar properties, with comparable kinetics and peak amplitude for both groups (Figure 3B). The AUC analysis showed that insulin secretion in the second phase (20–40 minutes) in KO mice was significantly lower than that of controls (Figure 3D, P = 0.0051); insulin secretion in the first phase (10–20 minutes) and during 30 mM KCl stimulation decreased only slightly (Figure 3, C and E; P = 0.07, n = 5 independent experiments), which may indicate some subtle defects in insulin secretion from the RRP. Thus, dynamin 2 regulates biphasic insulin secretion, with much greater impact on the second phase of GSIS than the first phase.

Figure 3

Dynamin 2 regulates biphasic insulin secretion induced by 20 mM glucose. (A) 20 mM glucose induced biphasic insulin secretion from both control and Dnm2 KO islets (n = 5 independent experiments for each group, n = 30–50 islets per experiment per genotype). (B) 30 mM KCl induced insulin secretion after 10 minutes rest from the same islet samples in A. (C and D) AUC analysis for the first phase (10–20 minutes, n = 5 each; P = 0.17, 2-tailed t test) and the second phase (20–40 minutes, n = 5 each, P = 0.016, 2-tailed t test) of GSIS shown in A. (E) AUC analysis of insulin secretion (P = 0.07, 2-tailed t test) induced by 30 mM KCl from B. (F) Time-lapse live-cell TIRFM of a representative fusion event in a control β cell expressing NPY-pHluorin (20 Hz). (G) Total granule fusion events under TIRFM during the first phase (during the first 0–6 minutes) and the second phase (during minutes 7–20) of GSIS under 20 mM glucose stimulation. Each square in the images marked a single fusion event. (H) The average number of fusion events in the first phase (n = 8 and 14 for control and KO cells, respectively; P = 0.30, 2-tailed t test) and second phase of GSIS (n = 7 and 6 for control and KO cells, respectively; P = 0.01, 2-tailed t test). Scale bars: 1 μm (F); 2 μm (G). **P < 0.01.

We next examined insulin granule fusion in individual β cells using TIRF microscopy (TIRFM). We generated NPY-pHluorin, a pH-sensitive fluorescent probe that selectively targets in insulin granules and shows strong fluorescence only in nonacidic environment, to sensitively detect insulin granule fusion. Each fusion event can be readily identified by transient brightening of fluorescent spots due to the pH increase in the granules upon fusion with the PM, and the signal-to-noise ratio of fluorescence is very high in our experiments. Figure 3F shows a typical fusion event visualized under TIRFM. During continuous perfusion of 20 mM glucose, individual β cells often responded with a burst of exocytosis events at the beginning, followed by sparse fusion events at later stages. Figure 3G shows the individual fusion events during the first 6 minutes (the first phase) and from minute 7 to minute 20 (the second phase) of 20 mM glucose stimulation in a β cell from control and KO mice. Quantitative analysis (Figure 3H) revealed a significant decrease in the second phase (P = 0.01; n = 8 and 14 cells for control and KO) and only a slight decrease (P = 0.30) in the first phase in KO mice. Thus, dynamin 2 is required to support biphasic insulin release at the single-cell level, in close parallel with the data from intact islet perifusion (Figure 3A).

Dynamin 2 regulates insulin secretion with normal intracellular Ca²⁺ levels and glucose metabolism. Insulin secretion can be regulated at different steps of insulin granule trafficking. We visualized the dynamics of intracellular Ca²⁺ concentration ([Ca²⁺]_i), a key factor frequently implicated in insulin secretion deficiency, during GSIS in intact islets (Figure 4, A–C). 12.8 mM glucose elicited a fast [Ca²⁺]_i rise for approximately 5 minutes and subsequent [Ca²⁺]_i oscillations in control islets, and the rises and oscillations all started with a small and transient [Ca²⁺]_i decrease (Figure 4B, arrowheads). Further, direct KCl (30 mM) stimulation generated a [Ca²⁺]_i increase with a shorter delay and higher amplitude than glucose stimulation. The [Ca²⁺]_i properties in KO islets were very similar to those in controls, albeit with slightly higher basal levels. Moreover, in good agreement with the Ca²⁺ imaging data, patch-clamp experiments demonstrated the comparable amplitude and density of Ca²⁺ currents between control and KO β cells (Figure 4, D and F).

Figure 4

Dnm2 KO β cells maintain normal Ca²⁺ signaling and glucose metabolism. (A) Transmission and fluorescence images of islets after fura-2 AM loading. Scale bar: 50 μm. (B) Representative cytosolic Ca²⁺ ([Ca²⁺]_i) dynamics in response to 12.8 mM glucose and 30 mM KCl. (C) Average [Ca²⁺]_i amplitude changes (n = 15 islets for each group; 2-tailed t test). (D) Representative Ca²⁺ current, (E) C_m, and (F) average amplitude and density of Ca²⁺ current (n = 14 and 15 cells for control and Dnm2 KO cells, respectively). (G) Excitation spectrum of NADPH fluorescence in islets and (H) the fluorescence increase induced by 20 mM glucose in individual islets. (I) Average NADPH signal at rest and during 20 mM glucose stimulation (n = 15 islets in control; n = 13 in Dnm2 KO). *P < 0.05.

We further examined the metabolic status of β cells using NADPH imaging (Figure 4G). NADPH fluorescence in islets showed large heterogeneity from islet to islet (Figure 4H); continuous NADPH fluorescence imaging at basal and glucose stimulation conditions demonstrated normal mitochondrial glucose metabolism in KO islets (Figure 4, H and I). Thus, [Ca²⁺]_i dynamics, glucose sensing, and mitochondrial metabolism were intact in Dnm2 KO β cells.

Intact RRP of insulin granules in the absence of dynamin 2 in β cells. The sizes of RRP and the reserve pool of insulin granules are key factors that affect GSIS dynamics in β cells. Next, we investigated insulin granule pool properties using EM. Insulin granules in KO cells were abundant and had similar size (~360 nm in diameter) (Figure 5, A–C) and subcellular spatial distribution as that in controls (Supplemental Figure 5, A–C). The docked granule number, as a morphological estimate of RRP, was comparable between control and KO groups (Figure 5, A, B, and D). Given the proposed dynamin function in endocytosis and vesicle biogenesis from the TGN (48–51), it is interesting that we found no significant decrease in granule density in KO β cells (Figure 5, B and C, and Supplemental Figure 5, B and C), even under 20 mM glucose stimulation for 20 minutes (Supplemental Figure 5D). Moreover, we observed no obvious cell surface expansion in KO cells, which is different from the severe vesicle depletion at nerve terminals after dynamin perturbations (31, 38).

Figure 5

Intact insulin granule biogenesis, docking, and RRP in Dnm2 KO β cells. (A and B) EM of a control and Dnm2 KO β cell. The arrow indicates endocytic intermediates underneath the PM. Note the abundant insulin granules with similar diameter in Dnm2 KO β cells. Asterisks denote granules docked to the PM. M, mitochondria; G, insulin granule; IG, immature granule. (C) Morphological analysis of insulin granule density and size, (D) the docked granules, and (E) immature granule density (P = 0.017, 2-tailed t test) (n = 8–10 cell sections randomly selected). (F) Proinsulin granules visualized in the z-stack projections of confocal images (at a 300 nm z-step). (G) Average proinsulin granule density in control (n = 24) and KO (n = 17) β cells (P = 0.0297, 2-tailed t test). (H) Insulin granules visualized under TIRFM in a representative control cell and KO β cells. (I) Average insulin granule density under TIRFM from control (n = 23) and dynamin 2 KO (n = 14) cells. (J) Average C_m traces and (K) C_m changes induced by different short depolarizing pulses. The cell numbers in each experiment are shown above the bars (P > 0.05, 2-tailed t test). Scale bars: 1 μm (A and B); 100 nm (B, inset); 2 μm (F and H). *P < 0.05.

Mature and immature types of insulin granules are readily distinguishable in our preparations based on their characteristic ultrastructures. Typical mature granules contained a dense core with a halo; immature granules were fully filled with homogeneous and much lighter electron density and had tight-fitting membrane (Supplemental Figure 6). These observations were consistent with previous immuno-EM studies (52, 53). The immature granules were more frequently observed in Dnm2 KO β cells (Supplemental Figure 5B), and their density was slightly higher than that in controls (Figure 5E, P < 0.05, 2-tailed t test). Likewise, proinsulin fluorescence imaging revealed a small but significant increase of proinsulin granule density in KO β cells (Figure 5, F and G). These data indicate a potential role of dynamin 2 in insulin granule biogenesis or maturation in β cells, which is in line with the proposed dynamin function in intracellular membrane trafficking in nonsecretory cells (refs. 48, 49, 54, 55 but see also ref. 50). However, the presence of abundant mature granules in dynamin 2 KO β cells does not support an indispensable role of dynamin 2 in insulin granule biogenesis at normal conditions, at least it is not the rate-limiting step in the granule trafficking cycle.

We next estimated the docked insulin granule number by TIRFM that detects granules within approximately 100 nm above the PM. We found no difference in granule density between control and KO β cells (Figure 5, H and I), consistent with the intact granule docking under EM. Moreover, we evaluated the functional RRP size (56) by cell membrane capacitance (C_m) under whole-cell patch-clamp (57) (by single pulse with variable length, Figure 5, J and K). A 200-ms pulse elicited a similar C_m increase in both control (n = 10) and KO β cells (n = 17), respectively (P > 0.05); the longer pulses induced larger C_m elevations, but there was no significant difference between control and KO cells (n = 7 and 5 for 0.5 s; n = 12 and 14 for 1 s; P > 0.05 for both cases, t test). Therefore, the functional RRP size was not significantly changed in KO β cells (Figure 5, J and K), which is in line with the similar insulin secretion induced by 30 mM KCl (Figure 3, B and E).

Dynamin 2 is critical for efficient exocytosis-endocytosis coupling in pancreatic β cells. The molecular nature of endocytosis is poorly understood in β cells. We directly monitored the membrane retrieval right after insulin exocytosis using capacitance measurements. β Cells were first identified by their relatively larger size compared to other islet cells and the characteristic Na⁺ channel inactivation (ref. 57 and Supplemental Figure 7A). Endocytosis in control β cells was largely heterogeneous, both in kinetics and in amplitude from cell to cell (Figure 6, A–E), although it can be fairly consistent in a given cell with the same stimulation (Supplemental Figure 7C). C_m often decayed with single or double exponential kinetics, accompanied occasionally by large downward C_m steps (arrowheads in Figure 6E). Similar properties were observed in Dnm2 KO β cells, and the average C_m trace still decayed rapidly. However, the decay time constant of the average C_m trace was much slower in KO cells (τ = 4.3 s, n = 14) than in control cells (τ = 2.7 s, n = 15) (Figure 6F). Given the critical role of dynamin 2 in multiple forms of endocytosis in nonneuronal cells (32, 45, 54, 58), it is surprising that the membrane retrieval was not blocked by the depletion of dynamin 2, an abundant dynamin isoform in β cells, indicating the presence of dynamin 2 independent endocytosis in these cells.

Figure 6

Impaired exocytosis-endocytosis coupling in β cells in the absence of dynamin 2. (A–E) Real-time C_m recordings showed endocytosis heterogeneity in amplitude and kinetics in different wild-type β cells. Arrows indicate the depolarizing pulse (0 mV, 1-second pulse in A–D; 0.5-second pulse in E); arrowheads in E indicate the large, stepwise C_m decreases. (F) Average C_m traces triggered by a single depolarization (0 mV, 1 second) from control (n = 15 cells) and Dnm2 KO cells (n = 14). The time constants of C_m decays are 2.7 seconds and 4.3 seconds for control and KO cells, respectively. The right panel shows the normalized C_m traces. Note the slower C_m decay and incomplete C_m recovery in KO cells. (G) The amplitude and speed of C_m decay at 2 seconds (P < 0.01) and 10 seconds (P < 0.005) after the depolarization (n = 15 and 14 for control and KO, respectively, unpaired 2-tailed t test). (H) Exocytosis-dominant (type-1) and endocytosis-dominant (type-2) C_m changes elicited by a train of pulses (ten 500-ms pulses, 0 mV, 1 Hz) in different β cells from wild-type mice. (I) The average C_m traces from control (n = 16) and Dnm2 KO (n = 21) groups. (J–L) Cumulative exocytosis and endocytosis and endocytosis rate during the train stimulation. **P < 0.01, ***P < 0.005.

To mimic the burst electrical activity in islets, we applied a train of pulses (0 mV, ten 500-ms pulses at 1 Hz) to β cells (Figure 6H, top). Consistent with previous reports (25, 26), we observed largely variable C_m changes during whole-cell C_m recordings. Figure 6H shows two types of typical C_m responses: an exocytosis-dominant C_m response (type-1, which has a continuous C_m increase and subsequent C_m plateau) and an endocytosis-dominant C_m response (type-2, which shows an initial C_m increase followed by a pronounced C_m decrease). The majority of recordings exhibited a mixture of these two types of C_m changes. To avoid potential bias in data classification, we simply pooled the C_m traces together for each genotypes (n = 16 in control; n = 21 in KO) (Figure 6I). The averaged C_m in KO cells displayed a continuous C_m increase (type-1) during train stimulation, which contrasts with the C_m decline following an initial, transient C_m increase (type-2 like) in control cells. The impaired endocytosis in KO cells may explain this difference, since C_m reflects the net change of surface area. Further quantitative analysis by extracting the exocytosis and endocytosis components of C_m data (37) suggested a marked reduction of endocytosis, but not exocytosis, in KO cells (Figure 6, J–L). Thus, dynamin 2 deletion impairs the efficiency of exocytosis and endocytosis coupling in β cells, and dynamin 2 is crucial for the optimal rate of membrane retrieval but is not indispensable.

Impaired CME and abnormal endocytic ultrastructure in β cells in the absence of dynamin 2. Distinct membrane retrieval kinetics in β cells suggests the presence of multiple endocytic pathways. We next used a transferrin uptake assay to evaluate CME in the absence of dynamin 2. Although weak transferrin fluorescence was still visible beneath the PM in Dnm2 KO β cells, transferrin–Alexa Fluor 568 internalization was significantly reduced (Figure 7, A and B), suggesting the impaired CME in KO cells.

Figure 7

Strong defect in CME and abnormal accumulation of endocytic intermediates in Dnm2 KO β cells. (A) Impaired transferrin uptake in Dnm2 KO β cells. Representative images after 10 minutes of transferrin uptake. (B) Relative changes of fluorescence intensity (n = 3 independent experiments, 6–12 β cells were analyzed from each group in each experiment; P < 0.01, 2-tailed t test). (C) Accumulated endocytic intermediates (arrows) adjacent to the PM in KO cells. Two clathrin-coated vesicles interlinked together through a narrow membrane tubule (arrowhead) are shown in the inset. (D and E) A gallery of endocytic intermediates from Dnm2 KO and control cells. Note the CCPs with elongated necks/tubules in KO. (F) Significant increase of CCPs in Dnm2 KO cells at rest (n = 22 and 36 EM sections for control and KO) and after 20 mM glucose stimulation (n = 29 and 20 for control and KO cells, respectively) (P < 0.005, 2-tailed t test). (G) 3D reconstruction of endocytic profiles in a Dnm2 KO β cell using EM tomography. Cells were stimulated with 20 mM glucose for 20 minutes at 37°C before fixation. Note the abundant CCPs (red) and the elongated, narrow tubules (light green) that connected the CCPs to the PM (dark green) directly or through a membrane expansion. (H and I) Simple CCPs with elongated tubules connected to the cell surface. (J–M) Complex structures of endocytic intermediates that connected to PM with large, branched membrane vacuoles. Scale bars: 2 μm (A); 200 nm (C and G); 100 nm (C, inset, D, and E); 50 nm (H–M). **P < 0.01, ***P < 0.005.

We next characterized β cell endocytosis at ultrastructural levels. Close inspections of EM images revealed that more abundant endocytic intermediates accumulated on the PM in KO β cells than in controls (Figure 7, C–F), and this difference was further enhanced after 20 mM glucose stimulation for 20 minutes (Figure 7F and Supplemental Figure 5D). CCPs in KO β cells tended to have narrower and longer necks than those in controls (Figure 7, C and D), and CCPs were occasionally interconnected with a narrow membrane tubule (Figure 7C, inset) which was unique to Dnm2 KO cells.

Further EM tomography analysis revealed a striking accumulation of CCPs and unique endocytic intermediate structures in 3 dimensions that were connected to the PM through narrow tubules in Dnm2 KO cells (Figure 7, G–M). Some CCPs were directly connected to the PM with a single narrow tubule (Figure 7, H and I), and many endocytic intermediates formed more complex structures, in which multiple CCPs interconnected to the same tubule or large vacuoles still connected to cell surface by a narrow tubule (Figure 7, J–M). The meandering tubules in KO β cells were approximately 20 to 80 nm in diameter and up to hundreds of nanometers in length. These structures are reminiscent of endocytic defects in central synapses lacking dynamin 1 (and dynamin 3) (36, 38, 40) but more complicated than those in fibroblasts lacking dynamin (45). A unique feature of these endocytic intermediates is that each tubule was capped by a CCP, suggesting a severe defect in the membrane fission reactions of CME.

Dynamin 2 regulates F-actin organization and insulin granule recruitment to the PM. The pronounced impairment of the second phase of GSIS in Dnm2 KO mice is strikingly similar to the effect of perturbations in actin regulatory proteins, such as Cdc42 (13), PAK1 (15), RAC1 (14), and N-WASP (16). We therefore examined the actin organization in Dnm2 KO β cells. Indeed, F-actin fluorescence (Alexa Fluor 488–conjugated phalloidin) was much stronger in KO β cells than in controls, with distinct spatial reorganization (Figure 8, A and B). F-actin filaments in control cells were relatively long and smooth, and they formed a loose and thin meshwork beneath the PM in 3 dimensions (Figure 8C). In contrast, KO β cells showed a strikingly disorganized and thicker F-actin network that was composed of crowded, shorter actin filaments and abundant, dense actin patches/puncta in all dimensions (Figure 8, A and C).

Figure 8

Significant reorganization of F-actin in Dnm2 KO β cells and impaired insulin granule recruitment to the PM during GSIS. (A) TIRFM showing a strong fluorescence increase and structural reorganization of the F-actin network (phalloidin) in a Dnm2 KO β cell. Insets show the boxed regions at high magnification. (B) Average fluorescence intensity of F-actin (n = 10 and 11 cells in control and KO; P < 0.01, 2-tailed t test). (C) 3D views of the F-actin network under confocal imaging (at 200 nm z-step). (D) F-actin changes in Dnm2 KO β cells were fully rescued by expressing dynamin 2–RFP. (E) Relative changes of phalloidin fluorescence intensity in control (n = 36), Dnm2 KO (n = 15), and Dnm2 KO cells transfected with dynamin 2–RFP (n = 17 cells; P = 0.002, 1-way ANOVA). (F) The number of insulin granules that were recruited to the PM under TIRFM. Cells were equally stimulated with 20 mM glucose and 10 nM GLP-1 for 1 hour at 37°C and fixed immediately. (G) Significant decrease in the number of insulin granules recruited to the PM after glucose stimulation (n = 25 and 35 cells for control and KO, respectively; P < 0.001, 2-tailed t test). (H) The model depicting dynamin regulation of biphasic GSIS through an actin-dependent mechanism in pancreatic β cells. The dashed line denotes other CME-mediated signaling pathways that may also involved. Scale bars: 5 μm (A and D); 2 μm (F); 1 μm (A, insets). **P < 0.01, ***P < 0.005.

Because the dynamin 2 deletion in β cells is temporally controlled by acute administration of tamoxifen at adulthood, it seems unlikely that the actin reorganization arises from chronic compensation in response to dynamin 2 loss. To further exclude nonspecific effects of Dnm2 KO, we tested whether reexpression of dynamin 2 rescues the abnormal F-actin phenotype in KO β cells. Three days after dynamin 2–red fluorescent protein (dynamin 2–RFP) transfection, KO β cells expressing dynamin 2–RFP fully restored the disorganized F-actin to normal, which contrasted to those cells containing no dynamin 2–RFP in the same cultures (Figure 8, D and E). These data suggest a specific role of dynamin 2 in F-actin reorganization in β cells, consistent with the complex feedback interactions between dynamin and actin during CME (45, 59, 60).

The molecular mechanisms underlying actin regulation of biphasic insulin secretion (10, 61) are poorly understood. One possibility is that the reorganized F-actin inhibits insulin release as a physical barrier by impeding granule access to the PM or decrease RRP size, as proposed in chromaffin cells (62, 63). However, this was not supported by intact docked granules and RRP size in KO cells. The second possibility is that dynamin-induced actin reorganization retards granule recruitment at a later stage of secretion. This is supported by the positive role of actin in vesicle transport (64–67) and the pronounced defect in the second-phase secretion (Figure 3), which largely relies on new granule recruitment to the PM (8, 11). To directly test this idea, we incubated β cells for 1 hour with 20 mM glucose and 10 nM glucagon-like peptide-1 (GLP-1), an incretin peptide that increases insulin secretion through cAMP- and VAMP8-dependent mechanisms (68). Indeed, the treatment revealed a significant decrease of insulin granule density under TIRFM in KO β cells (Figure 8, F and G), suggesting a defect of insulin granule recruitment during GSIS, which is consistent with the proposed actin function in granule replenishment (11, 61).

Given the tight link between dynamin and actin during endocytosis (69) and the critical role of actin in the second phase of GSIS (11, 61), it is conceivable that dynamin 2–dependent endocytosis regulates biphasic insulin release, at least in part, through actin-dependent recruitment of new granules to the PM at the late stage of GSIS (Figure 8H). It will be of special interest to investigate the molecular interactions among dynamin, actin, and insulin granules during GSIS in the future.

In this study, we provide the first genetic evidence in vivo to our knowledge that dynamin 2 is required to preserve normal glucose homeostasis and biphasic insulin secretion through endocytosis regulation of granule membrane trafficking. Interestingly, dynamin 2 played a prominent role in the second phase of GSIS, with little effect on insulin granule docking and RRP size. Moreover, dynamin 2 was critical for the efficient coupling of insulin granule exocytosis/endocytosis and the membrane fission reactions at the late stage of CME, but it was not essential for insulin granule biogenesis from the TGN in β cells under basal conditions. Finally, dynamin 2 deletion led to striking F-actin reorganization in KO β cells that can be fully restored by reexpression of dynamin 2. F-actin remodeling contributed to the recruitment of insulin granules to the PM for release during GSIS. Thus, this study revealed a regulatory function of dynamin in biphasic insulin secretion in pancreatic β cells and indicated a potential link between endocytosis and diabetes through the homeostatic control of membrane trafficking.

The molecular nature of endocytic machinery in pancreatic β cells. We found that both dynamin 2 and 3 are coexpressed in mouse pancreatic β cells; this is different from other types of LDCV-secreting cells, such as chromaffin cells (41), PC12 cells (27), HIT-T15 cells, and INS-1 cells (70), in which both dynamin 1 and 2 are expressed. The 3 dynamin isoforms in mammals contain the same domain organization, with 80% identity in encoding DNA sequences, and they independently contribute to membrane endocytosis with largely overlapping function (32). It will be interesting to examine whether dynamin 2 and 3 have differential functions in β cells.

In spite of the vigorous insulin granule recycling in β cells and high levels of dynamin 2 expression, dynamin 2 deletion caused neither granule reduction nor cell surface expansion (Figure 5, A–C, and Figure 4E). Exocytosis-endocytosis coupling was impaired but not blocked in Dnm2 KO cells, since a substantial amount of endocytosis remained (Figure 6). These results were unexpected. It is unlikely that developmental compensation plays a significant role here, because Dnm2 was deleted selectively in β cells under temporal control at adulthood. The conditionally targeted gene deletion can also rule out the potential complications of alternative loss-of-function approaches (71), such as off-target effects (of siRNA knockdown, pharmacological inhibitors, ref. 71; and acute photoinactivation, ref. 72) or dominant-negative effects (73) of dynamin mutations. Moreover, the full restoration of normal F-actin organization by dynamin 2 reexpression in KO β cells suggests a specific role of dynamin 2. Thus, dynamin 2 is not essential for β cell endocytosis per se but is critical for an optimal rate of membrane retrieval.

In Drosophila shibire mutants (31, 74), endocytosis deficiency is very severe, as demonstrated by a block of endocytosis, vesicle depletion, and extensive cell surface expansion of nerve terminals after stimulation. In comparison, we observed different results in dynamin 2 KO β cells. The simple explanation for this discrepancy is that β cells have access to dynamin 2–independent forms of endocytosis. For example, low levels of dynamin 3 (and possibly extremely low levels of dynamin 1) may, at least partially, support endocytosis in Dnm2 KO β cells. A similar scenario has been reported at rodent nerve terminals lacking dynamin 1 (36, 75), in which dynamin 3 partially supports vesicle recycling (38, 39). It is unclear whether the dynamin-independent endocytosis at some nonsecretory cells (45, 71) and nerve terminals (i.e., bulk endocytosis) (40, 72, 76) also occurs in dynamin 2 KO β cells. Future study using triple dynamin KO cells will shed new light on this question.

Strongly impaired CME in the absence of dynamin 2 in β cells. Both impaired transferrin uptake and aberrant ultrastructure revealed by EM in Dnm2 KO β cells demonstrated a crucial function of dynamin 2 in CME. Further, 3D EM tomography revealed that most PM-detached CCPs captured in individual EM sections were actually connected to the PM in KO β cells and represented endocytic intermediates after insulin release, suggesting the impairment of membrane fission at a late stage of CME in the absence of dynamin 2.

The endocytic intermediate structures captured by EM in dynamin 2 KO β cells (Figure 7, G–M) also shed new light on the fate of insulin granules after their full fusion. After insulin granule exocytosis and collapse onto the PM, the granule membrane components may be recaptured selectively as a whole unit or internalized by CME after being broken into smaller pieces. Both scenarios of LDCV membrane retrieval have been proposed in chromaffin cells (77, 78). Abundant accumulation of CCPs indicates the significant role of CME among other forms of endocytosis in β cells. The size of CCPs was approximately 100 nm in diameter, which was only about one-third the size of insulin granules (360 ± 12 nm, n = 6 control cells). Thus, membrane components of insulin granules after full fusion most likely split and resort into smaller pieces for CME.

Endocytic control of biphasic GSIS in mouse pancreatic β cells. This study is the first demonstration to our knowledge that dynamin 2–mediated endocytosis regulates plasma glucose homeostasis through changing the insulin secretion capacity of β cells. It underscores the crucial role of dynamin 2 in biphasic insulin secretion and suggests endocytic control of hormone secretion in vivo. Along the same line, two recent studies provided extra evidence that aberrant endocytic trafficking in β cells may contribute to the pathophysiology of diabetes (79, 80).

The strong decline of insulin secretion in Dnm2 KO islets was in agreement with the previous work showing the secretion inhibition by overexpression of dynamin K44A or dynamin 2 siRNA knockdown in insulin-secreting cell lines (23). Moreover, interestingly, the decline occurred prominently in the second phase of GSIS. This secretion deficit is unlikely to have resulted from impaired insulin granule biogenesis, because insulin granules in the KO β cells were still abundant in our experiments (Figure 5, A–C). This phenotype apparently differs from that of nerve terminals lacking the dominant dynamin isoform (dynamin 1), in which synaptic vesicle number significantly decreases due to impaired vesicle reformation (36, 75). Thus, the mechanisms underlying dynamin-mediated secretion function between the LDCVs in endocrine cells and the small synaptic vesicles at nerve terminals may be different.

The selective reduction of the second phase of GSIS in Dnm2 KO mice shows striking parallel with the reduced insulin secretion after perturbations of actin remodeling (13–15, 81), a key factor that selectively contributes to the second phase (10, 11). Accordingly, enhanced actin disassembly by PPARβ/δ (17) or the actin-capping protein gelsolin (81) increases the second phase. In Dnm2 KO β cells, F-actin was largely reorganized at the cortical layer (Figure 8, A–C), and this change can be fully rescued by reintroducing dynamin 2 into these cells (Figure 8, D and E), suggesting a specific causal relationship between dynamin 2 and actin reorganization in β cells. These data support a specific role of dynamin 2–mediated actin remodeling in the regulation of GSIS in β cells and highlight its potential relevance to the pathogenesis of T2D. In support of this idea, human islets of patients with T2D exhibit a strong decrease (80% less than controls) in P21-activated kinase PAK1, a downstream effector of Cdc42 and RAC1 signaling that controls cytoskeletal remodeling. Furthermore, PAK1 inhibition in nondiabetic islets leads to a selective reduction of the second phase of GSIS (15). Finally, diabetic islets from humans with T2D (82) contained ~10 times higher levels of total actin than controls, and a similar higher level of actin has also been reported in Goto-Kakizaki rats (83, 84).

Multiple roles of actin have been increasingly discovered to regulate membrane trafficking (85), but its exact molecular function remains poorly understood. Both a negative role as a physical fusion barrier and a positive role of promoting vesicle transport and exocytosis have been proposed (86, 87). In insulin-secreting cells, a fairly selective role of actin remodeling in the second phase of GSIS has been reported (11, 13, 14, 16, 61). Similarly, we observed a selective decrease in the second phase of GSIS along with strong actin reorganization in Dnm2 KO β cells. Our morphological and functional data suggest that insulin granule docking and RRP size were normal at rest in Dnm2 KO cells (Figure 5); this is consistent with the minor difference between control and KO islets in initial release induced by both glucose (the first phase) and direct depolarization (30 mM KCl or short depolarization). Therefore, the F-actin reorganization in KO β cells does not differentially affect RRP granules compared with control cells (Figure 3). On the other hand, TIRFM demonstrated a significant decrease in fusion events during the second phase (Figure 3), with a significant decrease in the number of insulin granules beneath the PM at the later stage of GSIS in KO cells (Figure 8, F and G). These data demonstrated a defect in insulin granule recruitment from the reserve pool to the PM in Dnm2 KO β cells after the RRP granules run out, providing a direct cellular explanation for the selective decline of the second phase of GSIS in KO mice.

Dynamin may regulate actin function through both direct and indirect mechanisms in β cells. First, dynamin interacts with actin cytoskeleton through direct binding to F-actin (88) and microtubules (89) or through indirect interactions with actin-binding/regulatory molecules (90–92). For example, dynamin can directly bind actin filaments through a conserved actin-binding domain between amino acid 399 and 444, and the point mutation of this domain specifically decreases the dynamin-binding affinity to actin without affecting endocytosis (88). Dynamin self-oligomerization after actin binding promotes the elongation of short F-actin by removing the actin-capping protein gelsolin from the barbed end of actin filaments. The shorter disorganized F-actin observed in dynamin 2 KO β cells is in line with a defect of dynamin-dependent actin elongation. Second, dynamin regulates actin reorganization through a mechanism downstream of the CME, where the complex actin-dynamin interactions occur (45, 59, 60, 69). For instance, clathrin coats arrested on the PM are required for nucleating the actin regulators and polymerizing actin at CCPs (45). Third, some CME-mediated signaling pathways may also contribute to the actin remodeling. In fibroblasts lacking dynamin, the activated Cdc42-associated kinase (Ack) is activated and concentrated at clathrin-coated pits through direct binding to clathrin heavy chain (93).

Figure 8H depicts a model based on the above data. During insulin granule trafficking, dynamin 2 regulates endocytosis as the major membrane fission machinery, particularly in CME. Clathrin and dynamin 2 on CCPs orchestrate actin function through direct and/or indirect molecular interactions to optimize its role in typical biphasic GSIS. In Dnm2 KO β cells, accumulation of the arrested CCPs on the PM due to fission defects impairs normal F-actin network, and this actin disorganization further causes a defect in granule recruitment to the PM that is critical for the second phase of GSIS. This model explains the impaired biphasic GSIS and glucose intolerance in Dnm2 KO mice. However, it is important to note that dynamin 2–mediated endocytosis controls the abundance and turnover of cell surface molecules, some of which may also be involved in GSIS through different signaling pathways (Figure 8H, dashed line).

In summary, using temporally controlled and β cell–specific gene ablation, we have provided compelling evidence both in vivo and in vitro that dynamin 2 is required for normal glucose homeostasis through direct regulation of biphasic insulin secretion from pancreatic β cells. Depletion of dynamin 2 caused a severe CME defect and subsequent aberrant actin reorganization; impaired granule recruitment from the reserve pool to the PM contributed to the prominent decline of the second phase of GSIS in Dnm2 KO mice (Figure 8H). This study has uncovered a pivotal role of dynamin 2–mediated endocytosis in regulating pancreatic β cell secretory function by orchestrating membrane trafficking homeostasis and F-actin remodeling, indicating a potential link between endocytosis and the pathophysiology of diabetes.

Generation of temporally controlled, β cell–specific Dnm2 KO mice. Mice carrying floxed sites in the exon 2 of dynamin 2 gene (Dnm2^fl/fl mice) (45) were crossed with Tg(Ins1-cre/ERT)1Lpi (also termed MIP1-CreERT) transgenic mice (46). Dnm2^fl/fl mice were provided by Pietro De Camilli (Yale University, New Haven, Connecticut, USA). Mouse genotypes were determined by PCR. Dnm2^fl/flMIP1-CreER mice and age-matched littermate controls (Dnm2^fl/fl) were paired for experiments whenever possible; age-matched black-6 mice were occasionally used in some earlier experiments. Mice were administrated at an age of 1 to 3 months with tamoxifen (47) (0.15 g/kg body weight dissolved in corn oil as vehicle, injected i.p., once a day) for 10 days, and experiments started approximately 4 weeks after the first treatment. Approximately 76% to 94.3% of β cells showed efficient dynamin 2 depletion, depending on the experimental conditions.

Glucose tolerance tests and insulin tolerance tests. For glucose tolerance tests (GTTs), blood was sampled from mouse tail veins at 0, 15, 30, 60, and 120 minutes after glucose injection (i.p., 2 g/kg body weight) in the fasted mice (for 14 hours), and glucose levels were measured by glucose meters (OneTouch-Ultra2, LifeScan Inc., or TRUEresult, Nipro Diagnostics Inc.). For insulin tolerance tests (ITTs), mice fasted for approximately 6 hours were i.p. injected with Novolin R (0.75 units/kg body weight; Novo Nordisk Inc.) before blood sampling. Age-matched littermates (Dnm2^fl/flMIP1-CreERT and Dnm2^fl/fl mice after injections of vehicle with and without tamoxifen) were used for GTTs to minimize the variation in gene background, intrauterine environment, etc. For high-fat diet experiments, mice were fed with high-fat diet (D12331, 58 kcal% fat with sucrose, Research Diets Inc.) for 3 months before experiments (for details, see the Supplemental Methods).

Pancreatic islet isolation and β cell cultures. Pancreatic islets of Langerhans and individual islet cells were isolated and cultured as previously described (57) with minor modifications (see the Supplemental Methods).

Islet perifusion, insulin secretion assays, and Western blotting. Islet perifusion was performed at 37°C using a system designed in-house. Isolated islets of equal number (between 30 and 50) and size were paired between control and KO mice in each experiment. Islets were perifused with Krebs-Ringer bicarbonate buffer (KRBB); fractions collected in 96-well plates were used for insulin tests with insulin AlphaLISA (94). Purified islets were homogenized in lysis buffer for Western blotting (see the Supplemental Methods for details).

Patch-clamp and C_m measurements. β Cells grown on coverslips were patch clamped at 34°C to 36°C, and whole-cell capacitance measurements were performed as previously described (37, 57) (see the Supplemental Methods for details).

Immunofluorescence staining. Pancreata were fixed with 4% paraformaldehyde (PFA) plus 4% sucrose in 0.12 M sodium phosphate buffer, embedded in OCT, and cut in 8-μM frozen sections. Cells grown on coverslips were fixed with 2% PFA. Samples were first processed for immunostaining with the primary antibodies and then with fluorescence-conjugated secondary antibodies (see the Supplemental Methods).

Spinning disk confocal and TIRF imaging. Fluorescence imaging was performed with the confocal and TIRF imaging system based on a Nikon Ti-E Eclipse inverted microscope (see the Supplemental Methods). Imaging 3D reconstruction, fluorescence analysis, and quantification were done using NIS-Elements (Nikon). Images for comparisons were acquired and displayed in the same settings for both control and Dnm2 KO cells unless specified. The insulin granules under TIRF were define as docked granules and counted using NIS spot detection function. For Figure 8F, cells preincubated with 10 nM GLP-1 for 30 minutes were stimulated with 20 mM glucose and 10 nM GLP-1 for 1 hour at 37°C and fixed immediately. For F-actin rescue experiments, β cell cultures were transfected with dynamin 2–RFP (provided by Pietro De Camilli) using Lipofectamine 2000 and fixed after 3 days of expression.

Live-cell TIRF imaging of insulin granule fusion. NPY-pHluorin was generated by replacing EGFP in NPY-EGFP plasmid with the pH-sensitive pHluorin (both NPY-EGFP and pHluorin were provided by Pietro De Camilli). NPY-pHluorin has the advantage of a higher signal/noise ratio for fusion events due to its pH sensitivity, and it is more sensitive to granule fusion events than NPY-EGFP, although it provides no information on granule dynamics before fusion. β Cells were transfected with NPY-pHluorin with Lipofectamine 2000 (Life Technology) and used within approximately 40 hours. Cells were TIRF imaged at 5 Hz (200-ms exposure), with 2 × 2 pixel binning at 35°C to 37°C, and the continuous perfusion of KRBB or 20 mM glucose in KRBB (pH = 7.3) was applied. All fusion events detected between 0 and 6 minutes and 7 and 20 minutes during 20 mM glucose perfusion were defined as the first and the second phase, respectively, and were normalized to the individual cell footprint areas (for details, see the Supplemental Methods).

Transferrin uptake assay. Cultured β cells were incubated with Alexa Fluor 568–conjugated transferrin (25 μg/ml, Invitrogen, T23365) in normal culture medium for 10 minutes (37°C) after preincubation in serum-free DMEM for an hour and then washed with acidic buffer and fixed with 2% PFA and imaged later (See the Supplemental Methods).

Cytosolic Ca²⁺ and NADPH imaging. [Ca²⁺]_i was measured by fura-2 ratiometric fluorescence imaging as described previously (39, 95). NADPH fluorescence imaging was carried out with the same system as the Ca²⁺ imaging (see the Supplemental Methods for details).

EM and EM tomography. EM was carried out as described previously (37), with minor modifications. Purified pancreatic islets were fixed with 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer for 2 hours and post-fixed in 1% OsO₄, 1.5% K₄Fe(CN)₆, and 0.1 M sodium cacodylate for 1 hour. Islets were en bloc stained, dehydrated, embedded, and cut into ultrathin sections (70–90 nm), followed by imaging and analysis (see the Supplemental Methods for details).

For electron tomography, tissue sections (200–250 nm) were imaged in a TECNAI TF20 transmission electron microscope (FEI) operated at 200 kV (Yale Center for Cellular and Molecular Imaging, Yale University) as described previously (40). Contours of membranes were traced using IMOD software manually.

Statistics. All values are presented as mean ± SEM unless otherwise indicated. Statistical analysis was performed using 2-tailed Student’s t test for 2 groups. One-way ANOVA with post-hoc Tukey honest significant difference test was used to analyze the differences among 3 or more samples. P values of less than 0.05 were considered statistically significant.

Study approval. All procedures involving animals were approved by the Institutional Animal Care and Use Committee of the University of Wisconsin-Madison and followed the NIH Guide for the Care and Use of Laboratory Animals (8th ed. The National Academies Press. 2011.).