Cytoplasmic-Nuclear Trafficking of G1/S Cell Cycle Molecules and Adult Human β-Cell Replication: A Revised Model of Human β-Cell G1/S Control

One hundred sixty-four different cadaveric islet preparations were used for these studies. The demographics and sources of the islets are described in another report (1). Dispersal of the human islets was performed as described in detail previously (1–5). Rat islets were isolated from 2- to 3-month-old Sprague-Dawley rats, dispersed, and cultured as detailed previously (5,6). Rat studies were approved in advance by the University of Pittsburgh Institutional Animal Care and Use Committee.

Adenovirus preparation has been described previously (1). The efficiency of adenoviral transduction, assessed using β-galactosidase and insulin costaining of human islets transduced with Ad.lacZ, was (mean ± SEM) 65.1 ± 3.0, 67.9 ± 2.5, and 75.7 ± 2.8% at 24, 48, and 72 h after transduction, respectively. In addition, to prepare a green fluorescent protein (GFP)-tagged cdk6 adenovirus, human cdk6 cDNA was subcloned into pcDNA3.1/CT-GFP plasmid (Invitrogen, Carlsbad, CA) using a GFP fusion TOPOTA expression kit (Invitrogen), which places the GFP at the C-terminus of cdk6. This was subcloned into the adenovirus shuttle vector, pACCMV, and adenovirus was prepared as described (1–7).

Islets were dispersed to single cells, fixed, and labeled as described (1–7). For studies with proliferating conditions, dispersed islets were transduced with either Ad.LacZ or Ad.cdk6 plus Ad.cyclin D3 (100 multiplicity of infection) for 2 h, cultured for 24, 48, and 72 h (as described in the figures), and immunolabeled using antisera as described in the Supplementary Table 1 of our accompanying original article (1). Labeled cells were visualized using laser confocal microscopy. Each experiment shown is representative of 3–6 human islet preparations.

Immunoblotting was performed as described (1–7). Antibodies used to detect the G1/S molecules are described in detail in Supplementary Table 1 of accompanying article (1). Each experiment shown is representative of 3–6 human islet preparations.

Rat insulinoma cells (Ins1 832/13) were washed in PBS twice and trypsinized for 5 min. Complete medium (RPMI medium; Gibco, Grand Isle, NY) containing 5.5 mmol/L glucose, 1% penicillin and streptomycin, 10% FBS, 10 mmol/L HEPES, 2 mmol/L glutamine, 1 mmol/L sodium pyruvate, and 50 μmol/L β-mercaptoethanol was added and a suspension of 200,000 cells was plated on a glass-bottom microwell dish (MatTek, Ashland, MA). The cell suspension was transduced with 100 multiplicity of infection of Ad.cdk6-GFP for 2 h. Human islets (200 islet equivalents) were dispersed as described and plated on a glass-bottom microwell dish (MatTek) and were transduced for 2 h with 100 multiplicity of infection Ad.cdk6-GFP. The transduction was stopped by adding 1 mL complete medium to the Ins1 cells or the dispersed human islets. Transduced Ins1 cells or dispersed human islets were imaged 24 h after infection using a Nikon A1 Confocal Live Cell System and the NLS-Element software (Nikon, Melville, NY).

Statistical analysis for Figs. 4 and 5 was performed using one-way ANOVA with Bonferroni post hoc correction. Student paired two-tailed t test was used in Fig. 2A and C. All values are expressed as means ± SEM. P ≤ 0.05 was considered significant.

In the absence of mitogenic growth factors or nutrients that are able to induce robust human β-cell proliferation, we have used adenoviral expression of D-cyclins, cyclin E, or cdk2, cdk4, and cdk6 to drive human β-cell proliferation (2–4,7). For the current studies, we selected the combination of Ad.cdk6 plus Ad.cyclin D3 because this combination yielded the highest rates of BrdU and Ki-67 incorporation (3). We asked whether proliferation associated with adenoviral expression of cyclins and cdks might influence the subcellular localization of G1/S components.

As previously reported (3,4), the expression of Ad.cdk6 and Ad.cyclin D3 led to β-cell proliferation and also the nuclear appearance of these two molecules (Fig. 1A). Specifically, after transduction with Ad.cdk6/cyclin D3, Ki67 labeling increased from 0.1% to ∼9.4% of β-cells; the percent of β-cells that contained nuclear cdk6 increased from 0.8 to 46.1% and the percent of β-cells that contained nuclear cyclin D3 increased from 0.3 to 29.2% (Fig. 2A).

Exploring whether other G1/S molecules altered subcellular location in response to induction of proliferation, three INK4 family members were found to be not altered: p15, p18, and p19 all remained cytoplasmic (Fig. 1B). However, p16, p21, and p27 all shifted into the nuclear compartment (Figs. 1C and 2A). In quantitative terms, p16, which was nuclear in only 8.4% of β-cells under basal conditions, became nuclear in 16.1%. Similarly, p21, which was nuclear in 7.9% of quiescent β-cells, became nuclear in 35.6%; p27, which had been nuclear in 1.9% of quiescent β-cells, became prominently nuclear in 11.2%. Surprisingly, p57, which had been so predominantly nuclear in quiescent cells (42.1%), became less prominent in the nucleus (23.8%) with induction of proliferation (Figs. 1C and 2A).

To determine whether the changes in nuclear abundance of p16, p21, and p27 and the decrease in p57 were a result of changes in the total abundance of these molecules, we examined their relative abundance with or without expression of cdk6 and cyclin D3. As shown in Fig. 2B and C, although these molecules shifted location with proliferation, there were no changes in their absolute abundance. Collectively, these results suggest that p16, p21, p27, and p57 markedly alter their subcellular localization, but not their overall abundance, in response to induction of proliferation, with p16, p21, and p27 displaying a pronounced increase in their nuclear frequency, whereas nuclear p57 declined in nuclear frequency.

We next studied G1/S molecule nuclear presence versus proliferation (Ki67) as well as the 72-h time courses of translocation of cdk6, cyclin D3, p16, p21, p27, and p57. Ki67 was selected rather than BrdU because it reflects cell cycle progression at the time of labeling, whereas BrdU incorporation reflects both active and past proliferation. Examples of these studies are shown in Fig. 3, and their quantifications are shown in Fig. 4. In Fig. 4, the blue bars represent nontransduced control β-cells at 24, 48, and 72 h after dispersal and plating. The subsequent three bars reveal events in β-cells transduced with Ad.cdk6 plus Ad.cyclin D3 at these same time points (white indicates the percent of β-cell nuclei that contain the G1/S molecule in question; gray indicates the percent of β-cells containing Ki67 in their nuclei; and black indicates the percent of β-cells that are doubly positive for Ki67 plus nuclear presence of the G1/S molecule in question).

The cyclin D3 (Figs. 3A and 4) in control (blue bar) β-cells appears in the nucleus of only 0.7% of β-cells at 24 h, increases to 2.8% at 48 h, and declines to 0.3% at 72 h. With Ad.cdk6 plus cyclin D3 transduction, 25–30% of β-cells display nuclear cyclin D3 at all three time points (white bars), and the percent of Ki67⁺ β-cells initially increases from 9.7 to 14.1% and then decreases to 7.5% (gray bars). Interestingly, cyclin D3 colocalizes in the nuclei with Ki67 in 5.0, 8.6, and 0.7% of β-cells at 24, 48, and 72 h (black bars), suggesting that it may enter the nucleus early, partner with cdk6, activate proliferation, and then may be exported, degraded, or both by 72 h; cdk6 reveals a similar biphasic pattern, suggesting that it enters the nucleus transiently and then declines through one of these mechanisms (Figs. 3B and 4).

For p16 (Figs. 3C and 4), in control nontransduced β-cells (blue bars), its nuclear abundance increases with time, increasing from 2.7 to 8.0% between 24 and 72 h. With activation of proliferation, nuclear p16 increases to levels above baseline and above controls by 72 h. Significantly, p16 is never in the nucleus of a β-cell that is also Ki67⁺ (black bars) at any of the three time points, suggesting that nuclear p16 is incompatible with proliferation.

p27 (Figs. 3D and 4) reveals a pattern that is generally similar to p16, with the principal observations being that it is only occasionally (1.5–4.0%) nuclear in control β-cells, it markedly increases with induction of proliferation at 72 h (10.5%), and it never is in the nucleus of β-cells that are proliferating (black bars), suggesting that its nuclear presence also may be incompatible with proliferation.

p21 (Figs. 3E and 4) reveals a different pattern. It does not change substantially in control β-cells (blue bars) over time, but its nuclear abundance (white bars) increases dramatically in transduced β-cells at every time point, ranging from 9.4 to 13.9% in controls and to 34.8 to 62.5%. Additionally, p21 is the only cell cycle inhibitor that coincides with Ki67 in the nucleus of β-cells, and it may do this in a time-dependent manner, increasing from 8.6 to 13.3% of β-cells and then declining to 5.7% at 24, 48, and 72 h, respectively. These observations are compatible with the assembly/chaperon and nuclear translocation functions of p21 in other cell types.

p57 (Figs. 3F and 4) reveals a different pattern. It is the only INK4 or KIP/CIP molecule that is present in the nuclei of large numbers of control β-cells and increases with time in culture (blue bars). Also, although its nuclear abundance increases with induction of proliferation at every time point (white bars), it is always lower in transduced β-cells than in control β-cells. Like p16 and p27, it never is present in the nuclei of proliferating β-cells (black bars).

The preceding results suggest a model in which G1/S molecules may be able to traffic into (cdk6, cyclin D3, p16, p21, p27) or out of (cyclin D3, p57) the nucleus of β-cells in a manner that is associated with proliferation. To directly confirm whether G1/S molecule trafficking can occur, an adenovirus in which human cdk6 was coupled to green fluorescent protein (GFP) (Fig. 5A) was prepared so that trafficking could be directly observed in real time, if it occurred. Ad.cdk6-GFP was selected because cdk6 can colocalize with Ki67 in human β-cells (Figs. 3B and 4), and because Ad.cdk6 is capable of driving human β-cell proliferation in vitro and in vivo and enhances human β-cell engraftment in vivo (1–3). To confirm that the Ad.cdk6-GFP construct was fully biologically active, it was compared with an adenovirus expressing wild-type untagged human cdk6 (Ad.cdk6), examining their relative levels of expression and their abilities, with or without Ad.cyclin D3, to induce pRb phosphorylation and human β-cell proliferation. These studies confirmed the efficiency of expression, bioactivity, and potency, as well as induction of pRb phosphorylation (Fig. 5B) and β-cell proliferation (Fig. 7C), of the Ad.cdk6-GFP as compared with wild-type cdk6.

The rat insulinoma line, Ins1, and dispersed human islet cells were transduced with the Ad.cdk6-GFP construct and live cell imaging was performed beginning 24 h after transduction. As can be seen in the live cell time-lapse images in Fig. 5D and E (also seen in video format, see Supplementary Movies 1 and 2), cdk6-GFP can be observed clearly in the cytoplasmic compartment initially, and then it shifts into the nuclear compartment with time. That this is not simply the natural progression of cytoplasmic translation of what will ultimately be a protein that is then targeted to the nuclear compartment is evident from the observations in Figs. 1–4 (1,3,4) that cdk6 is present in the cytoplasm but does not appear in the nuclear compartment in quiescent cells. Thus, these studies document in dynamic terms that cdk6 is a principally cytoplasmic protein in quiescent nonproliferating human β-cells but it is able to shuttle to the nucleus in association with proliferation.

To determine whether the cytoplasmic G1/S molecule paradigm is restricted to human β-cells or if it might be a feature of rodent β-cells as well, we examined cyclin D1, cyclin D2, and cdk2 in rat islets (Fig. 6) and compared their localization to that observed in the rat insulinoma cell line, Ins1 832/13. We selected cyclin D1 and cdk2 because they are present in the cytoplasm of human islets, and we chose cyclin D2 because it is the most important D-cyclin in rat islets and therefore comparable with cyclin D3 in human islets. As can be seen in the figure, these three molecules are most intensely apparent in the cytoplasmic compartment in rat β-cells. However, using the same antisera, the same fixation methods, and the same species, these three molecules are readily apparent in the nuclei of Ins1 cells.

To determine whether the molecules that appeared cytoplasmic in nonproliferating human β-cells could be observed in the nucleus of transformed proliferating human cell lines using the same antisera and same fixation methods in a human homologous system, we re-examined cyclin D3, cyclin A, cdk2, p16, and p21 in human β-cells, and we compared them with the human colonic cancer cell line, HCT116, and with HEK293 human embryonal kidney cells. As can be seen in Fig. 7, in human β-cells, each of the molecules showed the same principally cytoplasmic pattern described. In contrast, in HCT116 and HEK293 cells, cyclin D3, cyclin A, cdk2, and p16 could be readily observed in the nuclei of many cells; p21, which is nuclear or cytoplasmic in human β-cells, also was observed in both nuclei and cytoplasm in HEK293 cells.

Most previous models of human β-cell replication assumed that all G1/S molecules are present in the nuclear compartment, as depicted in Fig. 8A, awaiting activation (1–21). Instead, in quiescent adult human β-cells, most are found in the cytoplasm, as shown in the model in Fig. 8B (1). The only three abundant nuclear molecules in quiescent human β-cells are the following cell cycle inhibitors: pRb; p57, and, in some cases, p21 (1). All of which would be anticipated to discourage cell cycle progression. Further, the quiescent human β-cell model depicted in Fig. 8B is not static, but one that is active, motile, fluid, and characterized by shuttling of G1/S molecules to and from the cytoplasm into the nuclear compartment in association with β-cell proliferation, as shown in the models in Fig. 8C and D. These models demonstrate that several G1/S molecules can be induced to traffic into the nucleus in association with induction of proliferation. Further, the observations suggest that this trafficking may be selective, with some G1/S molecules engaging in nuclear trafficking (e.g., cdks, cyclins, p16, p21, p27, p57), whereas others may not (p15, p18, p19), and also suggest a directionality to this trafficking, with some molecules entering the nucleus (cyclins, cdks, p16, p21, p27) and some others exiting, degrading, or otherwise disappearing (cdk6, cyclin D3, p57).

The evidence for G1/S molecule cytoplasmic-nuclear trafficking seems clear. Specifically, cdk6 and cyclin D3 clearly become more abundant in the nucleus, as we have shown previously for cyclins D1 and D2, when overexpressed (1–4). In parallel, the cell cycle inhibitors p16, p21, and p27 also become more abundant in the nuclear compartment, without an overall increase in the amounts of these molecules. Conversely, the nuclear abundance of p57 declines, without an overall change in the amount of p57. In contrast yet again, p15, p18, and p19 reveal the following third pattern: no shift in intracellular location in response to proliferation. The most unequivocal evidence for trafficking was observed for cdk6-GFP, which is absent from the nuclear compartment under basal conditions but can be observed trafficking into the nucleus using live-cell GFP imaging approaches in Ins1 and human islet cells.

Superficially, the mechanism of proliferation seems obvious; cdk6 and cyclin D3 enter the nucleus, phosphorylate resident pRb, and cell cycle progresses. On reflection, however, the models raise fundamental questions about the control of trafficking of cell cycle molecules. In a widely held model (11,12,22–27), cdk6, cdk4, and the D-cyclins lack a nuclear localization signal (NLS) and require assembly in the cytosol by p21 and p27 (and perhaps p57) after their translation. Paradoxically, p21 and p27 serve not only as chaperones but also as members of cyclin–cdk inhibitor complexes. These complexes are guided into the nuclear compartment by the NLS that is present in both p21 and p27. This NLS is in proximity to phosphorylation sites, suggesting that kinases may phosphorylate p21 or p27 and activate the nuclear translocation of these molecules and their cdk–cyclin cargo. In this model, the abundance of p21 and p27 “titrates” proliferation, such that complete lack of p21 and p27 causes cell cycle arrest because cyclins/cdks cannot enter the nucleus and, conversely, marked upregulation of p21 and p27 causes cell cycle arrest because of their ability to block cyclin A/cyclin E/cdk1/cdk2 activity. Proliferation only occurs when p21 and p27 are present in the correctly titrated concentration. Of course, most of these data are derived from rapidly proliferating genetically modified mouse embryonic fibroblasts and cancer cell lines, in which all of these cyclins and cdks as well as CIPs/KIPs are nuclear (22–27). These events may be specific to cell type.

With this background, the data summarized in Figs. 3 and 4 may indicate that among the cell cycle inhibitors studied, p21 is the one most commonly present in the nucleus of Ki67⁺ β-cells and therefore may serve the chaperone/nuclear translocator function in the human β-cell. Conversely, since p27 and p57, like p16, essentially never appear in the nucleus of Ki67⁺ β-cells, they may serve a conventional cell cycle inhibitor function. A similar trafficking-mediated proliferation-inhibitory function also has been suggested for p21 (28). Further, the observations that p16, p21, p27, and p57 all increase in nuclear frequency or abundance with the induction of proliferation (although for p57, always lower than in control cells) may reflect chaperone/nuclear translocator functions, e.g., for p21, or an overall reactive increase in inhibitory cell cycle “tone” in cells driven to proliferate. This model raises the question of what is restraining the cdks and cyclins, the CIP/KIP family, and the INK4 family in the cytoplasmic compartment of quiescent adult human β-cells and preventing them from entering the nucleus? Perhaps they are assembled in the cytoplasm by the CIP/KIP molecules, but for some reason (e.g., possibly lack of peri-NLS phosphorylation) these complexes are constrained to the cytoplasm and cannot transport the cdk/cyclins into the nucleus. Perhaps the INK4s or other as-yet-unidentified cytoplasmic molecules anchor the cdk/cyclins in the cytoplasm. Unraveling these issues will require silencing the CIPs/KIPs and INK4s to see which, if any, influences trafficking of cdks/cyclins or proliferation. It also will require immunoprecipitation and proteomic studies to identify additional molecules that are bound to the cdks/cyclins. These studies will be particularly challenging given the complexity of human islet cellular composition (2,29,30), the relative paucity of β-cells within intact islets, and the lack of availability of large numbers of pure populations of human β-cells for study.

As good working models should, these models raise many additional questions. For example, we have not studied the trafficking of all potential G1/S molecules. Do E2Fs or late G1/S molecules (cyclins A/E, cdks1/2) or p107/p130 also traffic to the nucleus when proliferation is activated? Does pRb exit the nucleus? If so, then what orchestrates all of this? Or, if the E2Fs remain in the cytoplasm, then how can the cell cycle become active? Or, as suggested by some (31,32), are E2Fs dispensable for proliferation? As another example, Figs. 3 and 4 suggest that cdk–cyclin D complexes activate proliferation in certain cells that lack cell cycle inhibitors in the nuclear compartment, but what decides whether a given β-cell does or does not contain nuclear cell cycle inhibitors? And, if cdk6 and cyclin D3 are present in the nucleus of 30–40% of transduced cells (Figs. 2–4), then why are all of these cells not proliferating? Of course, we could not study the subcellular localization and partnering of every G1/S molecule in a single β-cell, so that knowledge of the precise molecular complexes and the stoichiometry of their components within the cytoplasm or nucleus of a given β-cell remains incomplete. Finally, these studies were performed using adult human β-cells, which are recalcitrant to proliferation. Because replication is believed to occur at greater rates in fetal and neonatal human β-cells (17,20,33), it will be informative to develop a model such as that shown in Fig. 8 using these types of islets.

The observation that G1/S molecules that are cytoplasmic in primary rat β-cells can be nuclear in proliferating rat Ins1 insulinoma cells (Fig. 6) is consonant with the concept that the cytoplasmic-to-nuclear translocation occurs in association with activation of proliferation. Coupled with data regarding intact human pancreas in another article (1), it is further suggest that the subcellular localization to the cytoplasm in human β-cells is not an artifact of fixation, antiserum choice, brain death, or islet isolation, because all of these variables can be controlled for in the rodent models. The cytoplasmic G1/S quiescence model may apply more broadly beyond the human or rodent β-cell, extending to quiescent differentiated mammalian cells in general. Thus, in addition to the data of our other article documenting cytoplasmic localization of G1/S molecules in differentiated quiescent non–β-cells, the concept is further supported by Fig. 7, contrasting quiescent human β-cells to proliferating human HEK293 and HCT116 cancer cells immunolabeled under identical conditions.

Nonetheless, this work has important limitations. First, we have used cdk6 and cyclin D3 to force proliferation, an artificial approach. Although it would have been more “physiological” to have used growth factors or nutrients to induce proliferation, no growth factors or nutrients that can induce robust proliferation in adult human β-cells have been identified, so the cdk/cyclin approach represents the only currently feasible approach. In addition, although it seems reasonable to assume that G1/S molecules traffic into and out of the nucleus, most of the data provided are descriptive and not functional or mechanistic, with the exception of the live cell imaging for cdk6. Along these lines, whereas several G1/S molecules appear in the nucleus and are less frequent at later time points, we have not documented that this nuclear diminution is a result of nuclear exit, intranuclear degradation, or another process, all of which are known to occur in other systems (26,27,34–38). Also, it also should be emphasized that the majority of studies were performed in isolated, dispersed, cultured, cadaveric β-cells, all conditions that might alter the normal distribution, function, and immunodetection of cell cycle molecules. This was unavoidable because the usual harsh fixation methods used for intact pancreas pathology specimens, e.g., long-term 10% formalin, prevented immunolabeling for most of the G1/S molecules studied. However, we were able to confirm the subcellular location in intact human pancreas of pRb p18, p57, p107, and p16, i.e., in five of five cases in which immunolabeling of intact pancreas was possible (1). Further, the data are complimented by similar data in nonproliferating rodent β-cells (Fig. 7) (36,37), so these observations seem firm. Also, whereas we use the broad term “G1/S” throughout, and although many of these same molecules participate in the G0/G1 transition, whether there are additional molecules that may “gate” the G0/G1 interface in β-cells is unknown. Importantly, we also have used terms such as “proliferation” and “replication” advisedly, using a surrogate for β-cell proliferation, Ki67, rather than authentic expansion of β-cell numbers. Unfortunately, this is unavoidable because no laboratory has developed techniques that provide unequivocal evidence of productive human β-cell expansion in vitro or in vivo (14). In a related manner, we variously use the terms “quiescent,” “senescent,” and “irreversible” interchangeably in the context of nonproliferating human β-cells. Although some would argue that these are different states, distinguishing among them in cadaveric human β-cells is difficult or impossible. Another important consideration is that although we have demonstrated an association between nuclear-cytoplasmic trafficking and cell cycle activation, we have not proven in formal terms that this is required for cell cycle entry in the human β-cell. It also is important to emphasize that whereas it is likely that G1/S molecule trafficking is one important regulatory process involved in β-cell cycle entry, it is only one of many important mechanisms, with other examples being phosphorylation and dephosphorylation of key substrates, transcriptional and posttranscriptional induction, or stabilization/destabilization of cyclins/cdks, or comparable reductions in cell cycle inhibitors, to name a few. Finally, although we believe we have controlled as well as reasonably possible for immunocytochemical/histochemical variation and error, antibody specificity can be troublesome. Thus, it must be emphasized that the models in Fig. 8 are just that, working models that can be used for confirmation, modification, correction, extension, and hypothesis-generation; they likely will evolve further with time.

In summary, these studies provide three new models of human β-cell G1/S cell cycle control, one in quiescence (Fig. 8B), one for proliferating β-cells induced by cdk6/cyclin D3 (Fig. 8C), and one for β-cells that fail to proliferate despite cdk6/cyclin D3 expression (Fig. 8D). The models should be useful for hypothesis-generation and provide new questions and potential therapeutic approaches to human β-cell expansion, such as high-throughput drug screens examining nuclear translocation or kinome profiling to identify kinases that phosphorylate key cell cycle inhibitors and induce their translocation.

This work was supported by the National Institute of Diabetes and Digestive and Kidney Diseases and the Beta Cell Biology Consortium through National Institutes of Health grants U-01 DK 089538, R-01 DK55023, R56 DK065149, and T32-07052, by Juvenile Diabetes Research Fund grants 1-2008-39 and 34-2008-630, by American Diabetes Association grant 7-12-BS-046, and by a University of Pittsburgh Junior Faculty Award (to N.M.F.-T.).

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

N.M.F.-T. researched data, contributed to the discussion, and wrote the manuscript. J.W.K., F.G.S., R.T., R.W., M.T., G.C., A.E.C., K.K.T., and H.S. researched data. D.K.S. contributed to the discussion. A.F.S. contributed to the discussion and wrote the manuscript. N.M.F.-T. and A.F.S. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

The authors thank Adolfo Garcia-Ocaña, PhD, Rupangi C. Vasavada, PhD, and Laura C. Alonso, MD, at the University of Pittsburgh School of Medicine for many helpful discussions during the preparation of these studies. The authors also thank the Kroh and Wagner Family Foundations for their support of this work. Human islets were generously supplied by the National Institute of Diabetes and Digestive and Kidney Diseases–supported and Juvenile Diabetes Research Fund–supported Integrated Islet Distribution Program, and by Tatsuya Kin, MD, at the University of Alberta.