Differentiation of Affinity-Purified Human Pancreatic Duct Cells to β-Cells

Human islet isolations were performed at Joslin Diabetes Center with the method of Ricordi et al. (19). After digestion and islet purification, the remaining tissue was used for the following study. This islet-depleted tissue usually had <1% of insulin-containing cells by dithizone (diphenylthiocarbozone) staining. Organ donors were 18–58 years old and did not have diabetes. Seventeen pancreata were used for experiments of CA19-9–purified cells; the first 9 were also used as crude (unpurified) preparations, and 6 were also used for purified CA19-9 plus added stromal cells. A schema of our experiments is shown in Fig. 1.

Immediately after islet isolation and purification, 5 ml islet-depleted tissue was divided into 20 50-ml conical tubes, with 40 ml CMRL1066 (5.6 mmol/l glucose; Invitrogen) medium supplemented with 10% fetal bovine serum and 100 units/ml penicillin and 100 μg/ml streptomycin (Mediatech) (CMRL medium). After suspension, tissue was allowed to settle for 5 min, and then the supernatant was aspirated to remove low-density components including dead cells and islets. For dispersion of cells, 2 ml 0.25% trypsin/EDTA solution (Invitrogen) and 10 ml PBS were added to each tube and the contents mixed by vortex for several seconds and incubated for 30 min at 37°C/200 rpm in an incubator shaker. Cold CMRL medium (10 ml/tube) was added to halt the enzymatic reaction. After pipetting the tissue up and down, tissue was filtered through 40 μm cell strainers (Falcon) to remove bigger clumps, and flow through from every set of 2 tubes of the 20 was collected into a new tube. Then, preparations were centrifuged at 1,000 rpm for 5 min, the supernatant aspirated, and pellets resuspended in 5 ml PBS supplemented with 750 mg/l EDTA and 5 g/l BSA (PBS solution). In a few cases, additional trypsin/EDTA digestion for 5 min was needed to obtain single cells. Most of these dispersed single cells excluded trypan-blue negative. These preparations were designated crude or unpurified duct cells (Fig. 1); a portion of these cells was then purified as follows.

CA19-9, an antibody to sialyl Lewis a antigen (sialylated lacto-N-fucopentose II), has been used to characterize human pancreatic ducts (20–22) and is expressed throughout the human pancreatic ductal tree. Dispersed cells were centrifuged at 1,000 rpm for 5 min, supernatant aspirated, and pellets resuspended with mouse antihuman CA19-9 antibody (1:200, clone 116-NS-19-9; Invitrogen) in 2 ml PBS solution. After 5 min incubation at 4°C, 10 ml PBS solution was added, and the cell suspension was mixed gently. Tubes were centrifuged at 2,000 rpm for 10 min, supernatant aspirated, 250 μl/tube goat antimouse IgG microbeads (1:5; Miltenyi Biotec) in PBS solution were added, and pellets were mixed well. After 15 min incubation at 4°C, additional 5 ml/tube PBS solution was added. Tubes were centrifuged at 1,000 rpm for 10 min, and pellets were resuspended in 20 ml/tube of cold PBS solution and passed through 40 μm cell strainers to remove newly formed clumps of cells.

MACS magnetic LS separation columns (Miltenyi Biotec) were prepared according to the manufacturer's instructions. The cell suspension of each tube was added to an LS column after priming with 5 ml PBS solution. After washing with 3 ml PBS solution three times, the magnetic field was removed and cells remaining in the LS columns (positive fraction) collected with 5 ml PBS solution. Keeping cells at low temperature (4°C or on ice) for a long time seemed to cause more dead cells and clumping of cells that reduced purity by nonspecific sticking; therefore, all procedures except incubation were done at room temperature. These cells were designated purified duct cells. Of these, 90–94% excluded trypan blue; 8% of the purified cells were CK19⁻, insulin⁻, vimentin⁻, or HSP27⁻ and seemed to have just nuclei without cytoplasm; these, we presume, are the trypan blue–positive dead/dying cells.

Dispersed cells, either crude or purified, were placed in nontissue culture–treated T-75 and T-25 flasks (Falcon) at a density of 10 million cells/25 cm² in CMRL medium and incubated in 5% CO₂ at 37°C. After 4 days, culture medium was transferred into 50 ml conical tubes. After settling for 5 min (crude) or 10 min (purified), according to the gravity method, supernatant was aspirated and pellets were returned to the original flasks with new CMRL medium. The expansion phase, ∼7 days, was ended when contaminating stromal cells started to proliferate rapidly in crude duct preparations or when most of epithelial cells stopped expanding in purified duct preparations. In parallel experiments, we found that at 2 days 17.0 ± 2.9% of the CK19⁺ cells were Ki67⁺ (29).

Adherent cells were lifted off with nonenzymatic cell dissociation solution (Sigma). Additional incubation for a few minutes with 0.25% trypsin/EDTA was used as needed to dissociate fibroblastic stromal cells. For aggregation, dissociated cells were placed on six-well hydrophobic plates (Corning) in serum-free Dulbecco's modified Eagle's medium/F12 (3.15 g/l glucose; Invitrogen) supplemented with 2.0 g/l BSA, 5 mg/l human insulin (Sigma), 5 mg/l human transferrin (Sigma), 5 μg/l selenite (Sigma), 1.2 g/l (10 mmol/l) nicotinamide (Sigma), 100 units/ml penicillin, and 100 μg/ml streptomycin.

After 3 days, aggregates were transplanted under the kidney capsules of normoglycemic NOD/SCID mice (8–12 weeks old) (Taconic). After 4 weeks, the kidneys with grafts were excised under anesthesia, and the animals were killed. All procedures were approved by the Joslin Diabetes Center Institutional Animal Care and Use Committee. The mice were housed and fed under specific pathogen-free conditions.

Cells/aggregates were collected into 1.5-ml tubes and pelleted by centrifugation. Pellets were fixed in freshly made 4% paraformaldehyde (Fisher Scientific) in PBS and enrobed in Bacto-agar (Becton Dickinson). Kidneys with a graft were fixed in 4% paraformaldehyde. Fixed cells and grafts were embedded in paraffin and sectioned (5 μm) by the Joslin Diabetes and Endocrinology Research Center histology core. Sections were incubated with proteinase K (DAKO) for 5 min for antigen retrieval (except that microwaving in citrate buffer was used for vimentin), blocked for biotin, and then incubated overnight at 4°C with primary antibody (mouse anti-human CK19 [1:20; DAKO], mouse anti-human CA19-9 [1:20], mouse anti-human HSP27 [1:20; LaboVision/Neomaker], or mouse anti-human vimentin [1:100, clone V9; DAKO]). Then, sections were incubated with donkey biotinylated antimouse IgG antibody (1:200; Jackson ImmunoResearch) at room temperature for 1 h, followed by streptavidin-conjugated Alexafluor 488 (1:200; Molecular Probes) at room temperature for 1 h. The vimentin antibody was specific for humans and does not react with mouse vimentin.

For insulin, sections were incubated at 4°C for 1 h with guinea pig anti-bovine insulin antibody (1:100; Linco Research, St. Charles, MO), washed, and then incubated with Texas Red–conjugated donkey anti–guinea pig IgG (1:200; Jackson ImmunoResearch) at room temperature for 1 h. DAPI was used for nuclear staining. Sections were examined in confocal mode on a Zeiss LSM 410 microscope using appropriate filters.

Total RNA was extracted with RNeasy (Qiagen) and RNase-free DNase (Qiagen) according to the manufacturer's protocols. Grafted tissue was trimmed as much as possible of surrounding mouse tissue. RNA (500 ng) was reverse transcribed to complementary DNA (cDNA) in 25 μl solution containing 5 μl 5× first-strand buffer (Invitrogen), 10 mmol/l dithiothreitol (Invitrogen), 1 mmol/l dNTP (New England Biolab), 50 ng random hexamers (Invitrogen), and 200 units Superscript II RNase H⁻ reverse transcriptase (Invitrogen). Reverse transcription reactions were incubated for 10 min at 25°C, 60 min at 42°C, and 10 min at 95°C. cDNA products were then diluted to a concentration corresponding with 20 ng starting RNA per 3 μl (23).

For semiquantitative radioactive multiplex PCR, 4.5 μl cDNA (equivalent to 30 ng RNA) was amplified in 50-μl volumes containing 5 μl 10× GeneAmp PCR buffer II, 1.5 mmol/l MgCl₂ (Applied Biosystems), and 160 μmol/l dNTPs. Additions of oligonucleotide primers (Sigma Genosys) were 10 pmol for human insulin (forward TCACACCTGGTGGAAGCTC, reverse ACAATGCCACGCTTCTGC, product size 179 bp) and 10 pmol for human cyclophilin (forward CCCACCGTGTTCTTCGAC, reverse GATCTGGTGGTTAAGATAAAACAC, 522 bp). These primers were designed not to recognize mouse insulin or cyclophilin RNA/DNA. Finally, 3 units AmpliTaq Gold DNA polymerase (Applied Biosystems) and 2.5 μCi of [α-³²P] dCTP (3,000 Ci/mmol; Perkin Elmer) were added. PCR amplification was performed in a Perkin-Elmer 9700 thermocycler with a 5-min initial denaturing step, followed by cycling of 1 min at 94°C, 1 min at the annealing temperature of 58°C, and 1 min at 72°C. The final extension step was 10 min at 72°C. PCR cycles were 22 for cells and 28–30 for grafts (because some mouse kidney tissue was included). PCR products were separated on 6% polyacrylamide gel in 1× Tris borate EDTA buffer. Band intensity was measured with a Storm phosphoimaging system (Molecular Dynamics, Sunnyvale, CA) and quantified with ImageQuant software (Molecular Dynamics). Preliminary reactions were performed to optimize PCR conditions such that all PCR products were linearly amplified. Levels of insulin mRNA expression were normalized to those of the internal control cyclophilin.

ABI 7300 real-time PCR system (Applied Biosystems) was used according to the manufacturer's instruction. The probe and primer sets of human insulin (assay identification no. Hs00355773) and cyclophilin A (part no. 4326316E) were purchased from Applied Biosystems; they are specific for humans and do not bind to cDNA of mouse islet, kidney, or liver. Ten nanograms cDNA was applied to each well, and levels of mRNA were determined as the average of triplet aliquots. Levels of insulin mRNA expression were normalized to those of the internal control cyclophilin. Data were then compared with those of human islets run in parallel.

In crude duct preparations, some fibroblastic cells are seen during expansion in CMRL medium. In one crude duct preparation in which the adherent epithelial cells grew poorly, the stroma cells reached 80–90% confluency. These cells were harvested with 0.25% trypsin/EDTA and passaged (1:2 dilution). After several passages, no epithelial cells or aggregates of cells were seen. After >20 passages, cells were used as stromal cells.

ANOVA and Fisher's protected least significant difference were used for statistical analysis. Data are shown as means ± SE. P values <0.05 were considered significant.

Starting samples of crude duct preparation or cells obtained following purification were designated as stage A. In crude duct preparations, cells usually clumped before attaching (within 2 days) to the flask surface. Epithelial-like cells expanded in monolayer from the attached clumps (Fig. 2A), reaching 70–80% confluence around day 7 (stage B). Then cells were transferred to hydrophobic plates for aggregation; most cells readily aggregated within the 3 days (stage C) (Fig. 2E). These aggregates were transplanted under the kidney capsule of NOD/SCID mice, and, 4 weeks after transplantation, grafts were easily found between the kidney capsule and cortex (stage D).

However, in CA19-9–purified duct preparations, cells formed smaller clumps over 3–4 days before attaching and expanding. Cells were homogenous; no fibroblastic cells were seen. Their expansion was slower than in the crude preparations and usually stopped at about day 7; the expanding cells often had abnormal morphology, looking more like fried eggs than like the usual cobblestone epithelium (Fig. 2B). Aggregates were much smaller than those from crude duct cells (Fig. 2F). Purified duct preparations from 10 pancreata were transplanted, but at 4 weeks only four grafts (one big and three small grafts) were found. Overall, the purified duct preparations had slow expansion, poor aggregation, and poor engraftment compared with the crude duct preparations, even from the same pancreata.

Stromal/mesenchymal cells often provide growth factors and function as feeder layers in vitro. The poor growth of the purified duct preparations may have resulted from the lack of even the small number of stromal cells found in crude preparations. Therefore, we evaluated the addition of pancreatic stromal cells to the purified duct preparations. Cultured stromal cells after >20 passages (0.05–2.0%) (Fig. 2D) were added either after magnetic sorting or after expansion. The addition of stromal cells caused faster expansion and larger aggregates than purified duct alone. The addition of 0.1–0.3% stromal cells gave the morphological appearance closest to that of the monolayers of crude duct preparation. With only 0.1% stromal cells, cellular expansion was enhanced (Fig. 2C) and larger aggregates were formed (Fig. 2G), and transplanted grafts were easily found. Therefore, 0.1% stromal cells were added after sorting for subsequent experiments. Stromal cells cultured alone formed large, dark aggregates (Fig. 2H) with all cells immunostaining for human vimentin (Fig. 5C). Stromal cells had no insulin⁺ nor CK19⁺ cells by immunostaining and no insulin mRNA by quantitative RT-PCR.

To determine the cell composition of in vitro crude, CA19-9 purified, and CA19-9 purified plus 0.1% stromal cell preparations, insulin⁺ and CK19⁺ cells were quantified on immunostained sections of pelleted cells and aggregates (Table 1). At least several thousand cells were counted in each in vitro stage (stages A–C) of each preparation, and >1,000 CK19⁺ cells were counted in each graft (stage D). In crude duct preparations, the population of CK19⁺ cells increased through expansion (stage A to B), and insulin⁺ cells decreased (as a percentage of total cells) from the very low initial 0.38 ± 0.09%. Another way of calculating the proportion of insulin⁺ cells was as a percentage of total CK19⁺ cells (Fig. 3A); with this denominator, the percentage of insulin⁺ cells decreased through expansion (initial 0.69 ± 0.15% in stage A to 0.07 ± 0.02% in stage B), increased through aggregation (to 0.43 ± 0.07% in stage C), and further increased in grafts (to 6.12 ± 0.35% in stage D). Representative CK19 and insulin staining in grafts (stage D) of crude duct preparations are shown in Fig. 4A and B.

In purified duct preparations without added stroma, no insulin⁺ cells were found in vitro (stages A–C) (Table 1). Seventeen pancreata, including the four shown in Table 1, were used for immunomagnetic purification. Although at least 4,000 (usually >10,000) cells were evaluated for each in vitro stage (stages A–C) of each preparation, no insulin⁺ cells were found (Fig. 5A), except in two preparations in which several insulin⁺ cells were found. Both of these preparations had many more dead cells than usual, with resultant clumping during sorting and poor purification, and were therefore not used as purified preparations. Of the four grafts obtained from the purified duct preparations, one had ∼1% insulin⁺ cells of total CK19⁺ cells (>1,000 CK19⁺ cells were counted); the other three were very small grafts (data not shown).

In purified duct plus stroma preparations, no insulin⁺ cells were found in starting samples (stage A) or expanded cells (stage B), but 0.1 ± 0.0% insulin⁺ cells were found in aggregates (stage C) (Table 1). Insulin⁺ cells as a percentage of total CK19⁺ cells increased in grafts (1.10 ± 0.07% in stage D) (Fig. 3). Again, several thousand cells were counted in each stage of each in vitro preparation (stages A–C), and >1,000 CK19⁺ cells were also counted in each graft (stage D). Representative CK19 staining in purified duct plus stroma preparations are shown in Fig. 5D and F. All cells of aggregates of cultured stromal cells immunostained for human vimentin (Fig. 5C). In their aggregates and grafts, no insulin⁺ cells were found (Fig. 5C and E) and no insulin mRNA detected with RT-PCR (Fig. 6C). Additionally, no CK19⁺ cells were found in vitro or within these grafts (data not shown).

In grafts (stage D) of crude duct preparations, 85 ± 4.5% of insulin⁺ cells were in duct-like structures, and in grafts (stage D) of purified duct plus stroma preparations, all insulin⁺ cells were in duct-like structures. To test if this localization was a random event of contaminating β-cells, 1,000 isolated human islets, dispersed with trypsin, were added to one crude duct preparation just after dispersion. After expansion, this preparation had ∼2% insulin⁺ cells of total cells; RT-PCR data are shown in Fig. 6A. In the three grafts examined from this preparation, only one-half of the insulin⁺ cells were in duct-like structures, with many far from duct-like structures (Fig. 5G). The proportion of insulin⁺ cells outside duct-like structures in grafts from crude duct or crude duct plus added islets may reflect expansion of contaminating insulin⁺ cells; however, the localization of all insulin⁺ cells in duct-like structures in the grafts from purified duct plus stroma suggest that these cells were derived from ductal cells.

Initially, expression of insulin mRNA was evaluated with semiquantitative, radioactive RT-PCR using primers for human insulin and human cyclophilin. Insulin expression increased with engraftment both in crude preparations (0.38 ± 0.07 in stage C to 4.44 ± 0.83 in stage D) and in purified duct plus stroma preparations (undetectable in stage C to 0.44 ± 0.10 in stage D) (Fig. 6A). For more quantitative analysis, particularly of the starting tissues, we used quantitative real-time PCR (Fig. 6B and C). Insulin mRNA in initial stage A was only 1.0 ± 0.4 × 10⁻⁵ that of adult human islets in purified duct preparations and 1.0 ± 0.9 × 10⁻⁵ in purified duct plus stroma preparations; those in expanded stage B in purified duct plus stroma preparations were 1.2 ± 0.4 × 10⁻⁶ of islets. These data provide strong evidence of the lack of even degranulated β-cells in the initial purified duct preparations. Measured by quantitative RT-PCR, insulin mRNA increased in grafts compared with that of aggregates; grafts from crude preparations (0.46 ± 0.05% of islets in aggregate stage C to 5.76 ± 0.71% of islets in graft stage D) and purified duct plus stroma preparations (0.06 ± 0.01% in aggregate stage C to 1.22 ± 0.16% of islets in graft stage D) showed substantial increases in insulin mRNA. (No insulin mRNA was detected by quantitative PCR in stage B, C, or D of the stroma—only preparation.)

If the insulin⁺ cells differentiated in vivo from the transplanted duct cells as we hypothesized, then one might expect to find some transitional cells. In fact, some cells in aggregates (stage C) and grafts (stage D) (Fig. 7) show immunostaining of both insulin and ductal markers, either CK19, CA19-9, or HSP27. Anti-human HSP27 antibody has been used to distinguish transplanted human tissue from mouse kidney (24). We found that within the human pancreas, all pancreatic cells (epithelial and stromal) other than the islet endocrine cells expressed human HSP27 (Fig. 7A). Within aggregates (stage C) (Fig. 7C) and grafts (stage D) (Fig. 7D), some insulin⁺ cells also expressed HSP27. Our criteria for a transitional cell were that CK19 staining must be along most of the plasma membrane of an insulin⁺ cell (Fig. 7B) or overlap with the immunoreactivity of insulin. More frequently, CK19 immunostaining was along only part of the membrane of an insulin⁺ cell or the cell was not embedded within the CK19⁺ epithelium; such cells were not considered as colocalizing or transitional even though they may have been at a late stage of differentiation. Similarly, some cells in the grafts (stage D) coexpressed insulin and the ductal marker CA19-9 (data not shown).

Our main finding was that human pancreatic duct cells can differentiate into insulin-producing cells. We succeeded in obtaining very purified pancreatic duct cells from human islet–depleted pancreatic digests by magnetic sorting with CA19-9 antibody; quantitative RT-PCR detected only 1 of 100,000 of the insulin of adult human islets in these purified preparations. After sorting, these purified duct preparations contained no insulin⁺ or vimentin⁺ cells. When 0.1% stromal cells were added to the purified duct preparation to enhance expansion, the grafts were found to consist of 1% insulin⁺ cells. Moreover, all insulin⁺ cells were in duct-like structures, and some coexpressed CK19, CA19-9, or HSP27, markers seen in ductal cells but not in β-cells in sections of human pancreas. These findings provide strong support that these insulin⁺ cells were newly generated and derived from pancreatic duct cells. In our original study initially establishing this model, we showed that, by ultrastructure and by secretion studies, the islet cells were well developed but perhaps not yet fully mature (12). Gao et al. (13) showed that the secretion was glucose responsive and closely resembled mature β-cells. Here, however, the insulin⁺ cells may still be immature in phenotype, since many are still expressing ductal markers.

Our aim had been to identify the origin of insulin⁺ cells, and the exclusion of other cells leads us to conclude that pancreatic duct cells (CA19-9⁺ cells) could become insulin⁺ cells. A strength of our study is that duct cells were purified immediately after human islet isolation, with the purification taking only 3–4 h. Thus, transdifferentiation of acinar cells (25,26) or dedifferentiation of β-cells into epithelial phenotype (14,16,17) was very unlikely to have occurred before duct purification. Hao et al. (27) similarly reported that a CK19-rich fraction derived from an islet-depleted fraction could become β-cells when transplanted with fetal pancreatic cells into NOD/SCID mice. However, in that study, transdifferentiated acinar cells or dedifferentiated β-cells could have developed an epithelial (CK19⁺) phenotype during shipment from isolation centers or during the 4-day culture in G418, used to eliminate mesenchymal cells. Gao et al. (28) suggested that the insulin⁺ cells may have been derived from the residual β-cells in their preparations, since depletion of neural cell adhesion molecule (NCAM)⁺ cells, which were presumed to be only islet cells, resulted in few insulin⁺ cells. However, while NCAM is expressed selectively in rodent islets, we found that in human pancreas, NCAM was also expressed at both the mRNA and protein level in the duct epithelium (S.B.-W., C. Nienaber, unpublished data).

An alternative interpretation might be that the insulin⁺ cells found in the aggregates and grafts were progeny of a few residual β-cells from the starting preparations. It has been suggested that human β-cells can undergo epithelial-mesenchymal transition (EMT) and greatly expand in culture with a loss of insulin expression and that, with further manipulations, insulin and other markers can be reexpressed, albeit at extremely low levels (14,16,17). However, the actual cell of origin was obscured in these studies by the lack of marking of the human β-cells and by the significant contamination of human islet preparations by many cell types. Recent studies have addressed this issue with genetically marked mouse islets and have shown that while murine β-cells may undergo EMT in culture, their expansion in vitro was minimal (33–35). Even so, β-cells can replicate, so theoretically if β-cells were initially present, they could expand in culture and, with a concomitant dying off of ductal cells, become enriched and thereby could account for the insulin⁺ cells found in the aggregates and grafts. While this possibility cannot be ruled out, we think it is highly unlikely, based on several of our findings. First, in the purified duct with or without stroma samples, with at least 4,000 cells counted at each cultured stage for each pancreas and at least 1,000 CK19⁺ cells per graft, no insulin⁺ cells were found until aggregation and grafts. Second, in these samples, insulin mRNA was essentially absent, as only 1/100,000 of that of human islets. Third, there is no evidence of the massive cell death in either the aggregates or the grafts, which would be necessary for a β-cell enrichment to the frequency of insulin⁺ cells in the grafts. Thus, while we cannot rule out the possiblility without in vitro lineage tracing, it is highly unlikely that the insulin⁺ cells found in the grafts with transitional characteristics were residual β-cells or their progeny.

We should discuss the possibility that the added stromal cells were the source of the insulin⁺ cells. It has been reported that islet-derived mesenchymal (fibroblast-like) cells can become β-cells in vitro (14,16,17). While the morphology of those cells and our stromal cells was similar, their origins are likely different. The starting tissue of our stromal cells was islet-depleted (crude duct) pancreatic digest, in which the epithelial cells did not expand, but initially included stromal (fibroblast-like) cells, while the mesenchymal cells of the above-cited studies were derived from islet-rich preparations and assumed to originate from islet cells. Even so, under the same conditions as our purified ducts, passaged stromal cells did not contain any insulin⁺ cells at any stage of our experimental protocol, including grafts; by quantitative RT-PCR, no insulin mRNA was detected in the stromal cells. Therefore, it is highly unlikely that the added stromal cells became the insulin⁺ cells.

The role of stromal cells in our study remains unclear. They improved expansion, aggregation, and engraftment of purified duct cells, perhaps by secreting extracellular matrix and/or growth factors (29). In a preliminary study, expansion was improved when matrix (collagen) coated the flasks. Stromal cells may also secrete morphogens or provide a microenvironment that promotes differentiation into insulin⁺ cells. Inductive signals from the mesenchyme are essential for the proliferation of embryonic pancreatic epithelial cells (30,31). Moreover, it was reported that fibroblast growth factor 10, produced by stromal cells, is important for the proliferation of epithelial progenitor cells (32). Further studies are underway to investigate these possibilities.

In conclusion, we show that human pancreatic duct cells can become insulin⁺ cells in vitro. Our method should be useful to study further manipulations that promote neogenesis of β-cells from duct cells.

This study was supported by grants from the National Institutes of Health (DK44523 and DK74879 to S.B.W.), the Juvenile Diabetes Research Foundation (IN 05-1077 to S.B.W.), the Diabetes Research and Wellness Foundation, and an important group of generous private donors. Support was provided by the media and advanced microscopy cores of the Joslin Diabetes and Endocrinology Research Center (NIH and P30 DK36836-16). S.Y. was supported by a Manpei Suzuki Diabetes Foundation fellowship and by a Juvenile Diabetes Research Foundation fellowship.

The islet-depleted human tissue was provided by the Joslin Islet Cell Resource Center (NCRR U42 RR16606).