A method for the generation of human stem cell-derived alpha cells

The generation of pancreatic cell types from renewable cell sources holds promise for cell replacement therapies for diabetes. Although most effort has focused on generating pancreatic beta cells, considerable evidence indicates that glucagon secreting alpha cells are critically involved in disease progression and proper glucose control. Here we report on the generation of stem cell-derived human pancreatic alpha (SC-alpha) cells from pluripotent stem cells via a transient pre-alpha cell intermediate. These pre-alpha cells exhibit a transcriptional profile similar to mature alpha cells and although they produce proinsulin protein, they do not secrete significant amounts of processed insulin. Compound screening identified a protein kinase c activator that promotes maturation of pre-alpha cells into SC-alpha cells. The resulting SC-alpha cells do not express insulin, share an ultrastructure similar to cadaveric alpha cells, express and secrete glucagon in response to glucose and some glucagon secretagogues, and elevate blood glucose upon transplantation in mice.

A lthough diabetes primarily involves beta cell dysfunction, there is mounting evidence that alpha cell defects play a role in disease etiology 1,2 . Patients with type 1 diabetes must cope with significant fluctuation in blood glucose levels including acute hypoglycemia 3 .
In the healthy pancreas, hormone-expressing endocrine cells function within the islets of Langerhans to precisely regulate blood glucose and energy metabolism. During hypoglycemia (low blood glucose), islet alpha cells secrete glucagon 4 which raises blood glucose levels by increasing glycogenolysis and gluconeogenesis in the liver 5,6 . Although alpha cells persist in diabetic islets, these alpha cells are often incapable of mounting an appropriate glucagon response, perhaps due to the absence of alpha cell−beta cell interactions 7 . Recent studies implicate dysfunction of alpha cells as a contributing factor in the elevated blood glucose levels observed in diabetic patients 4,7 .
Published methods to make pancreatic beta cells [8][9][10][11] all report a minor portion of alpha and delta (somatostatin secreting) cells. Additionally, there are several reports on the conversion of various cell types into alpha cells via transdifferentiation [12][13][14] . In 2011, Rezania et al. 15 reported a protocol for generating glucagon-positive cells that exhibited some glucagon secretion in vitro; however, upon transplantation of 1.9 million cells into mice, these cells had limited physiological effects. Despite these early efforts to generate glucagon-positive cells, the production of alpha cells has not been reproduced nor widely adopted by the field.
We previously reported the generation of functional pancreatic beta cells from human pluripotent stem cells using a six-step directed differentiation protocol 8 . That protocol generates beta cells as well as side populations including polyhormonal cells and nonendocrine cells 11 . The presence of polyhormonal cells in pancreatic differentiations have also been observed in other reported protocols 9,16,17 . Several reports described polyhormonal cells as having features of immature beta cells 9,[17][18][19] ; however, beyond the expression of insulin protein, there has been little evidence to support the similarity of polyhormonal cells to beta cells, let alone their capacity to differentiate into beta cells. In contrast, others have reported that polyhormonal cells are present during development, contribute to alpha cells later in development, express several markers of alpha cells, and give rise to glucagon-expressing cells when transplanted 11,15,20 .
Reports from the literature have referred to INS+/GCG+ cells in turn as "bi-hormonal" 21 , "polyhormonal" 9,22-24 , "alpha-like cells" 17 , or "alpha cells" 25 . The nomenclature of the INS+/GCG+ cells produced in directed differentiation protocols has become equally confusing. In Veres et al. 11 , we performed extensive scRNA-seq studies on beta cell differentiations and observed a population of cells expressing many markers of alpha cells which we labeled as "alpha-like" cells irrespective of insulin expression. In this manuscript we refine this definition of "alpha-like" cells to distinguish between cells that have detectable insulin protein which we define as "pre-alpha" cells and those that do not express insulin, which we define as "SC-alpha cells". We favor the term "pre-alpha" rather than "bi-hormonal" or "polyhormonal" because it more accurately depicts the similarity of these cells to alpha cells yet distinguishes them from more mature alpha cells that do not express insulin (SC-alpha).
Here we build upon our previous reports to develop a protocol for the generation of SC-alpha cells. The protocol robustly produces a transient pre-alpha cell intermediate, which has a transcriptional signature similar to alpha cells except that they express insulin in addition to glucagon. We identify a small molecule capable of driving pre-alpha cells to an alpha cell identity and demonstrate the ability to produce approximately 30% SC-alpha cells in vitro. Our results show that these SC-alpha cells have a transcriptional signature similar to human cadaveric alpha cells, are responsive to some glucagon secretagogues, and elicit a robust physiological response within 4 weeks of transplantation in mice. The SC-alpha cells generated can recapitulate central aspects of alpha cell biology and represent an efficient and scalable avenue to produce alpha cells for use in islet organoids for cell replacement therapy, drug screening, disease modeling, and may accelerate exploration of alpha cell biology.

Results
Optimization of differentiation to generate pre-alpha cells. Our previously published protocol for generating stem cell-derived beta cells (SC-beta cells) produces a small but significant population of pre-alpha cells 8,11 . Figure 1a shows a typical result wherein 27% of the cells are glucose-responsive SC-beta cells and 9% are pre-alpha cells expressing both insulin and glucagon. Kelly et al. 16 suggested that these pre-alpha cells (referred to as polyhormonal in ref. 16 ) come from progenitors that fail to express NKX6.1. Thus, we sought to modify our beta cell protocol to prevent or reduce induction of NKX6.1 at stage 4. Using the HUES8 embryonic stem cell line, we observed that removal of KGF, SANT-1 and treatment with LDN only on day 2 of stage 3 results in a decrease of NKX6.1-positive cells 11 . In addition, we observed that treatment with LDN on day 1 of stage 4 and incubation with no factors for the remainder of stage 4 resulted in a significant population of Chromogranin A+/NKX6.1− cells. Together these protocol modifications (Fig. 1b) resulted in a large fraction of pre-alpha cells as marked by the coexpression of insulin and glucagon proteins (Fig. 1c). In the HUES8 cell line, this pre-alpha cell optimized protocol produces an average of 62.6 ± 2.3% insulin+ and glucagon+ coexpressing pre-alpha cells and a small percentage (<10%) of monohormonal glucagonexpressing SC-alpha cells ( Supplementary Fig. 1). Immunofluorescent staining confirms a high proportion of cells that express both insulin and glucagon protein (Fig. 1d) distributed throughout the cell clusters.
Pre-alpha cell transcriptional profile. We investigated the transcriptional signature of the pre-alpha populations produced at the end of stage 5 by single-cell RNAseq. Using single-cell sequencing (inDrops) 26 , we profiled 2043 cells from a pre-alpha cell differentiation revealing four distinct cell populations (Fig. 1e). Confirming the immunostaining and flow cytometry analysis, we observed a population of cells that express both insulin and glucagon transcripts, although expression of insulin transcripts was significantly lower than glucagon transcripts (mean tpm of 649 vs. 214,320; Fig. 1f and Supplementary Fig. 2a), indicating that these cells have downregulated insulin expression. This pre-alpha cell population (pink in Fig. 1e) expresses a transcriptional signature more similar to alpha cells than to beta cells ( Supplementary Figs. 2b and 3). In addition to expressing insulin and glucagon transcripts, the pre-alpha cells also express transcripts for several markers of alpha cells and lack several key markers for beta cells. For example, pre-alpha cells express transcripts for ARX, IRX1, and IRX2, transcription factors that are expressed in endogenous alpha cells 27 ( Supplementary  Fig. 2b), but do not express transcripts for beta cell markers such as NKX6-1, PDX1, and PAX4 ( Supplementary Fig. 3). Figure 1f shows the relative transcript expression levels of pancreatic hormones in the pre-alpha cell population compared to the major endocrine cell types from human islets.
In addition to the pre-alpha cell population, two minor cell populations are present including a TPH1-positive enterochromaffin population 11 and a nonendocrine population. The presence of these minor populations compared to the large number of pre-alpha cells confirms at the transcriptional level our finding that the majority of cells produced with this protocol are pre-alpha cells. Since the transcriptional profile of pre-alpha cells closely resembles alpha cells, we hypothesized that the pre-alpha cell population may be useful in generating stem-cell-derived alpha (SC-alpha) cells.
Pre-alpha cells secrete glucagon but not insulin. Due to the transcriptional similarity of pre-alpha cells to primary (cadaveric) alpha cells, we evaluated the functional response of pre-alpha cells to glucose. Pre-alpha cell clusters secrete glucagon under lowglucose (2.8 mM) conditions and suppress glucagon secretion at high glucose (20 mM) concentrations (Fig. 1g). Membrane depolarization induced by high (30 mM) extracellular K + stimulated glucagon secretion in both pre-alpha cells and mature human islet alpha cells. When assessing insulin secretion from these cells with a human specific processed-insulin ELISA, prealpha cells do not secrete significant quantities of processed insulin compared to human islets. Although these cells express insulin protein (Fig. 1c), the c-peptide antibody used in flow cytometry and immunofluorescence experiments does not distinguish fully processed insulin from its precursor, indicating that pre-alpha cells do not process proinsulin into insulin.
To analyze the insulin processing in pre-alpha cells, we evaluated proinsulin processing by Western blot ( Supplementary  Fig. 2c). In human islets, the majority of proinsulin protein is converted to insulin, while in pre-alpha cells, the majority remains as proinsulin. We assessed transcriptional expression of the prohormone converting enzymes PC1 and PC2 (PCSK1 and PCSK2 genes) and found that pre-alpha cells expressed PCSK2 to a much higher degree than they express PCSK1 ( Supplementary  Fig. 2b). Thus, pre-alpha cells transcribe the insulin gene and produce proinsulin protein, but do not cleave proinsulin nor secrete mature insulin in significant quantities.

The pre-alpha cell is a transient state in vitro and in vivo.
Previous reports demonstrated the presence of a small population of alpha cells in grafts from transplanted SC-beta cell differentiations 8 . We postulated that these alpha cells were derived from the pre-alpha cell side populations present in these SC-beta cell differentiations. As such, we tested the ability of pre-alpha cells generated in our protocol to convert into SCalpha cells post transplant. We transplanted 5 million pre-alpha cells under the kidney capsule of (n = 10) immunocompromised (SCID-beige) mice and observed a reduced expression of insulin protein over the course of 4 weeks. Grafts were retrieved 14, 28, or 56 days after transplantation and assessed for hormone expression using antibodies that react with insulin and glucagon proteins. At 14 days after transplantation, pre-alpha cells continued to express insulin and glucagon proteins (Fig. 2a left, Pearson's R value = 0.57). When grafts were evaluated at 28 days, few insulin protein-expressing cells were observed, whereas glucagon protein-expressing cells persisted (Fig. 2a   Step 1 Step 2 Step 3 Step 4 Step 5 Activin  Repl. Pop. ratios These results suggest that insulin protein expression is reduced in pre-alpha cells and glucagon protein expression is maintained with extended time in vivo. This result is consistent with previous studies which concluded that cells expressing both insulin and glucagon can resolve into alpha cells 20,25,28,29 . To exclude the possibility that the increase in SC-alpha cells observed after transplantation was due to selective replication of an SC-alpha subpopulation and/or concomitant death of prealpha cells, we evaluated cell replication and apoptosis during this in vivo maturation ( Supplementary Fig. 4). Rarely were TUNEL+/glucagon+ cells observed. Although low levels of Ki67-positive replicating cells were observed, they occurred equally in cells expressing both insulin and glucagon and glucagon-only ( Supplementary Fig. 4). Given the ability of pre-alpha cells to reduce insulin protein expression after transplantation, we tested whether this conversion can occur in vitro. After generation of pre-alpha cells, we continued culturing the cells for an additional 4 weeks in media without growth factors. During this extended culture period, cells were sampled at 7-day intervals and assessed for insulin and glucagon protein expression by flow cytometry. As shown in Fig. 2b, the vast majority of these cells (>80%) expressed both insulin and glucagon proteins at the beginning of the culture period. After 14 days, little change had occurred as >80% of the cells still expressed insulin and glucagon protein. After 21 days, a population of insulin-, glucagon+ cells appears with~20% of the cells expressing only glucagon protein. This population of monohormonal SC-alpha cells that appeared at day 21, persists at day 28 of extended culture. These results indicate that upon extended culture in vitro, a fraction of pre-alpha cells reduce insulin protein expression and become SC-alpha cells.
PKC activation promotes maturation of SC-alpha cells. Given that pre-alpha cells can convert to SC-alpha cells in vivo and in vitro, we sought to identify signals that promote this conversion by performing a small-molecule screen. HUES8 embryonic stem cells were differentiated using the pre-alpha cell optimized protocol to the end of stage 5 (stage 6 day 1), arrayed into 384well plates and incubated with a custom 43-member smallmolecule library (Supplementary Table 3) where compounds were chosen for their activity in targeting signaling pathways. After 96 h of treatment (stage 6 day 5), cells were fixed and stained for insulin and glucagon (Fig. 3a). High-content imaging was used to quantify the dispersed (2D) cell populations for the percentage of cells expressing each hormone individually and the percentage of pre-alpha cells marked by expression of both hormones (Fig. 3b, c). The effect of each compound was evaluated in quintuplicate assays. The protein kinase c (PKC) activator phorbol 12,13-dibutyrate (PDBu) decreased the percentage of insulin protein-expressing cells and increased the percentage of glucagon-expressing cells compared to vehicle controls. To confirm that the effects of this compound were not unique to the planar assay format, PDBu was evaluated in the final stage of the 3D directed differentiation protocol. Over the course of a 28-day treatment, PDBu induced a significant population of pre-alpha cell clusters to reduce insulin protein expression compared to control (33 ± 6% vs. 16 ± 5%; n = 5; Fig. 3d and Supplementary  Fig. 5). The resulting cell population contained significantly more alpha cells that did not coexpress the insulin protein (Fig. 3e).
To further explore the specificity of PKC activation, a small library of known PKC activators was evaluated. This structurally diverse set of compounds have all been reported to have PKC activating abilities. We added pre-alpha cell clusters (stage 6 day 1) to six-well plates and treated with either vehicle or each PKC activator for 28 days and assessed the percentage of monohormonal, glucagon-expressing alpha cells ( Fig. 3f and Supplementary  Fig. 6). While the control condition resulted in only 13.7% SCalpha cells, treatment with each PKC activator resulted in a significant increase in the percentage of SC-alpha cells, though none more so than PDBu. To evaluate the stability of the shift from pre-alpha cells to SC-alpha cells, we withdrew PDBu from cells for       Fig. 7). Withdrawal of PDBu did not significantly affect the percentage of SC-alpha cells in our population. These results demonstrate that activation of PKC stably reduces the expression of insulin in the INS+/GCG+ population thereby accelerating the conversion process of pre-alpha cells to SC-alpha cells in vitro (Fig. 3g). As previously described, differentiation of hPSCs to beta cells results in a side population of pre-alpha cells. To this end, we evaluated the effect of PKC activation on the differentiation of SC-beta cells. Treatment of stage 6 SC-beta cells with the PKC activator PDBu did not have a significant effect on the number of beta cells in the population (Supplementary Fig. 8a). However, when treated with PDBu, the pre-alpha side population in the differentiation was reduced and the percentage of SC-alpha cells was increased ( Supplementary Fig. 8b). Because previous reports pointed to variability when using different cell lines 30 , the robustness of this SC-alpha protocol was evaluated in the 1016 iPS cell line. Differentiated 1016 cells generate a similar percentage of SC-alpha cells ( Supplementary Fig. 9) showing that the protocol is able to direct another stem cell line to SCalpha cells using this protocol. To establish the performance of our SC-alpha cell protocol in comparison to previously published reports, we performed a head-to-head comparison where SCalpha cells were generated with our protocol or the Rezania protocol 15 in order to assess protocol efficiency and the functional performance of the resulting cell populations (Supplementary Figs. 1,10,11,12,13). Using the HUES8 cell line, the percentage of SC-alpha cells produced was improved more than eightfold over previously reported protocols and demonstrates improved functional responses to glucose.
Molecular characterization of pre-alpha and SC-alpha cells. To investigate the transcriptional changes that occur as cells transition from pre-alpha cells to SC-alpha cells, we performed singlecell RNAseq on pre-alpha cells and SC-alpha cells (Fig. 4). On a global level, the transcriptional profile of pre-alpha cells and SCalpha cells are remarkably similar with insulin expression being a notable exception (Fig. 4a). Analysis of 82 islet specific genes in comparison to human islet expression patterns further confirms the similarity of both pre-alpha cells and SC-alpha cells to primary human alpha cells (Fig. 4b). These results suggest that the conversion of pre-alpha cells to SC-alpha cells represents a subtle maturation process rather than a cell fate change. Consistent with our global gene expression analysis, a KEGG and gene ontology pathway analysis demonstrates significant similarity between prealpha and SC-alpha cells (Fig. 4c, d and Supplementary Figs. 14 and 15) with the most significant pathways common between both cell types. These top pathways are consistent with the function of alpha cells and include protein export, protein processing in the ER, protein folding, and cellular signaling pathways. Despite these similarities, some differences were observed. Pre-alpha cells uniquely express genes enriched in the insulin secretion and metabolic stress pathways, while SC-alpha cells uniquely express genes enriched in the glucagon secretion and hormone action pathways.
Using pseudotime analysis, we established a sequence of transcriptional changes that occur as cells progress from prealpha to SC-alpha cells (Supplementary Fig. 16). The transition from pre-alpha to SC-alpha results in the sequential decrease of stress-related genes VTN and AFP followed by beta and delta cell genes INS, TSPAN1, GAL. Finally, alpha-cell-associated genes PLIN2, TXNIP, and GC are upregulated. To further examine the role of PKC activation on these transcriptional changes, we performed an additional scRNA-seq experiment to compare the transcriptional profile of SC-alpha cells generated with PDBu to untreated SC-alpha cells which resulted from a spontaneous conversion from pre-alpha to SC-alpha cells. ( Supplementary  Fig. 17a). The tSNE plots from stage 6 day 28 cells generated with and without PDBu resulted in three major overlapping cell populations indicating significant similarity in their transcriptional profile. The predominant population expresses transcripts that are known markers of cadaveric alpha cells (ARX, IRX1, IRX2) 27 (Supplementary Fig. 17b). Very few genes are differentially expressed between the SC-alpha generated with and without PDBu ( Supplementary Fig. 17c and Supplementary Table 2). Of the genes that are differentially expressed, insulin transcript is downregulated 2.3-fold in cells treated with PDBu. Glucagon transcripts were not significantly affected by PDBu treatment.
To evaluate the protein expression levels of insulin and glucagon, we performed Western blots on cell lysates from cells treated with PDBu ( Supplementary Fig. 18). Cells treated with PDBu had a higher expression of glucagon protein compared to untreated cells. In untreated samples, glucagon expression is decreased. As we observed in our flow cytometry analysis (Fig. 3d), the Western blot confirms that PDBu treatment decreases the level of insulin protein expression ( Supplementary  Fig. 18).
Structural and functional characterization of SC-alpha cells. SC-alpha cells express known markers of alpha cells including GCG, PDX1, IRX, PC2, but do not express beta cell-specific markers PC1 or NKX6-1 (Fig. 5a−j). Electron microscopy reveals that the ultrastructure of SC-alpha cells resembles that of human primary alpha cells. In human islets, glucagon granules in alpha cells are dark and diffuse with round cores with an average size of 242 ± 8 nm (Fig. 5k), which is smaller than the condensed insulin granules of the beta cell 31 . In SC-alpha cells, the secretory granules have a similar ultrastructure to alpha cells with an average size of 214 ± 12 nm (Fig. 5l).
To evaluate the ability of SC-alpha cells to respond to physiological conditions, we screened a number of glucagon secretion modulators (Fig. 5m). Like mature human alpha cells [32][33][34][35][36] , glucagon secretion in SC-alpha cells is inhibited by glucose, an effect that is maximal at 7 mM glucose with higher glucose concentrations being less inhibitory (Fig. 5m). This pattern of glucagon secretion in response to increasing glucose concentrations has been reported by others 37 . Glucagon secretion from SC-alpha cells is suppressed in the presence of exogenous somatostatin, normally released within the islets by the delta cells, unaffected by exogenous insulin and arginine and stimulated by veratridine, an activator of voltage-gated Na + channels. Arginine evokes a transient stimulation of glucagon secretion ex vivo in mature alpha cells and time-resolved measurements may therefore be necessary to detect its stimulatory effect 38 .
We also sought to evaluate the stimulatory capacity of SCalpha cells using electrophysiology. Using a fluorescent reporter cell line that expresses mCherry under the endogenous glucagon promoter, we identified alpha cells for patch clamp experiments and compared the electrical activity of these cells to primary human alpha cells in response to changing glucose concentrations. Like primary alpha cells, SC-alpha cells generate spontaneous overshooting action potentials in the presence of 1 mM glucose (Fig. 5n, o). In the SC-alpha cells, these action potentials reflect activation of voltage-gated Na + , Ca 2+ , and K + channels, similar to what has been demonstrated in adult human alpha cells 37 . The effects of high glucose (10−11 mM) were also similar: slight depolarization of the interspike membrane potential, an increase in action potential frequency and a reduction in action 11 26 7 Pre-Alpha SC-Alpha  GCG  GJA3  LOXL4  CRYBA2  TM4SF4  TMEM236  IGFBP2  COTL1  PAX6  POPDC3  ARX  STK32B  LDHA  MUC13  F10  BVES  NPNT  EPHA4  FXYD5  TTR  ADCY2  TMEM176B  ALDH1A1  CHGB  ISL1  PLCE1  PEMT  ARRDC4  RGS4  SMIM24  DPP4  GC  FAP  FAM83B  SERPINA1  SPINK4  SERTM1  APOH  LPAR1  CRH  MYO10  SPOCK3  SYTL5  TMEM45B FXYD3 potential height (Fig. 5n, o inlays). These effects of high glucose are similar to those reported in mouse alpha cells and indicate that glucagon secretion may be influenced by action potential firing in a amplitude-rather than frequency-dependent fashion 39 . In all, the results demonstrate that SC-alpha cells resemble human primary alpha cells in their transcriptional profile, glucagon granule morphology, electrophysiology and physiological response to several (but not all) glucagon secretion modulators in static incubations.

SC-alpha cells reduce hypoglycemia in transplantation models.
We evaluated the potential utility of these cells to modulate physiology in vivo by transplanting SC-alpha cells. Using a continuous glucose monitor (CGM), we evaluated the interstitial glucose concentrations in animals at 5-min intervals during a period of fasting-induced hypoglycemia. Control animals (sham surgery) exhibited hypoglycemia after 6 h of fasting while mice transplanted with SC-alpha cells were protected from hypoglycemia (Fig. 6a). In SC-alpha-transplanted animals, blood glucose levels begin to return to baseline before food was restored while control animals continued to decrease blood glucose until food was restored. Furthermore, when administered an i.p. bolus of exogenous insulin, SC-alpha-transplanted animals are protected from hypoglycemia (Fig. 6b). To evaluate the effect of transplanted SC-alpha cells on long-term blood glucose values, we measured interstitial blood glucose continuously starting 4 weeks after transplantation and continuing until week 8. During this time, mice were maintained under normal housing conditions with 24-h light/dark cycles and feeding ad libitum. At the end of the 4-week observation period, CGM glucose readings for all mice in a cohort were averaged and represented as an average blood glucose value per 5-min interval throughout a standard 24-h period with the associated standard error (Fig. 6c). As has previously been reported [40][41][42][43] , control mice exhibited a characteristic fluctuation pattern in blood glucose values over a standard 24-h period with elevated blood glucose levels during active/feeding periods (dark) and lower blood glucose levels during resting periods (light). In SC-alpha-transplanted mice, we observed a similar pattern of glucose fluctuations, although the average blood glucose concentrations were elevated compared to control mice (Fig. 6d). Transplantation of SC-alpha cells does not perturb the normal circadian regulation of blood glucose concentrations in these animals but the presence of additional glucagon secreting cells raises the basal blood glucose concentrations in these animals. Further evaluation of circadian and ultradian patterns in blood glucose levels confirms that there is no significant shift in periodicity of these previously reported glucose rhythms (Supplementary Fig. 19). The sum result of SC-alpha cell transplantation is an elevation of blood glucose levels and a reduction in the proportion of time that mice spend with blood glucose levels in the hypoglycemic range (<70 mg/dl). As shown in Fig. 6e, mice that have received SC-alpha cell transplants spend on average 5 ± 2% of the time in hypoglycemia as compared to 15 ± 4% for control animals. These results suggest that transplantation of SC-alpha cells may have the ability to offer some protection from hypoglycemia.
To further demonstrate that the protective effect of SC-alpha cells was due to glucagon signaling, we evaluated the ability of SCalpha cells to secrete glucagon in vivo in response to arginine stimulation and in response to a mixed meal. Control mice and mice transplanted with SC-alpha cells were injected i.p. with an arginine solution to stimulate glucagon secretion from SC-alpha cells. After arginine injection, serum glucagon levels raised more than twofold (Fig. 6f and Supplementary Fig. 20). In comparison to control animals, both the basal and stimulated glucagon levels in these animals are several fold higher in the SC-alphatransplanted animals suggesting that the majority of glucagon observed in these studies was secreted from the transplanted SCalpha cells. As has been previously reported for alpha cells 44,45 , SC-alpha cells secrete glucagon in response to administration of a mixed meal. After refeeding, SC-alpha transplanted mice demonstrated a greater than twofold increase in serum glucagon (Fig. 6g). Given the significantly higher circulating glucagon observed in SC-alpha-transplanted animals under both fasting and fed conditions, we sought to evaluate the influence of transplanted SC-alpha cells on glucose tolerance and clearance. In a glucose tolerance test (GTT), SC-alpha-cell-transplanted mice demonstrated elevated fasting blood glucose and peak glucose levels, with impaired glucose clearance (Fig. 6h). Given that circulating glucagon levels in these animals (basal and stimulated) are several fold higher than control animals (Fig. 6f, g), it is not surprising that blood glucose levels are elevated during glucose challenge. Taken together these results demonstrate the function of SC-alpha cells upon transplantation and the ability of these cells to modulate blood glucose metabolism and physiology.

Discussion
The promise of regenerative medicine lies in the ability to produce tissues to replace damaged or diseased tissues. Diabetes has been a leading candidate for this approach because a single-cell type, the beta cell, plays such a critical role in the disease. This led to a focus on the generation of pancreatic beta cells and although much progress has been made, there are potential advantages in adding other cell types to treat diabetes.
Numerous studies highlight the importance of multiple islet cell types in regulating insulin secretion 46,47 . Paracrine signaling from alpha cells establishes a set point for insulin secretion 48 , and glucagon is important for beta cell regeneration in a zebrafish model 49 . Recent studies suggest that alpha cell dysfunction plays a role in chronic hyperglycemia in patients with T1D due to hyperglucagonemia 4 . Additionally, a defective alpha cell response to hypoglycemia is part of the impaired counter-regulatory response to hypoglycemia observed in T1D 7 . After cessation of insulin secretion in nondiabetics, glucagon is the first line of defense against hypoglycemia prior to activation of sympathetic responses. However, this response is altered in T1D due to alpha cell dysfunction, resulting in increased episodes of hypoglycemia. In T1D, insulin concentrations are dependent on insulin dosing while the sympathetic counter-regulatory response is impaired by prior hypoglycemia 50 . The ability of SC-alpha cells to protect against hypoglycemia and increase circulating glucagon levels in response to physiological cues may represent an avenue for therapeutic intervention in hypoglycemia. Although we have not evaluated the long-term impacts of SC-alpha cell transplantation in these studies, the high levels of circulating glucagon observed in response to SC-alpha cell transplantation are likely to result in compensatory glucagon resistance 15 .
Despite mounting evidence that defects in beta cell function do not fully account for the dysregulation observed in diabetes 4,7 , current therapies for diabetes (both type 1 and type 2) have focused almost exclusively on the insulin-producing beta cell. For example, insulin replacement therapy used in patients with type 1 and type 2 diabetes seeks to replace insulin that would normally be produced by beta cells. Few therapies are currently on the market that target other cell types in the pancreas. GLP-1 receptor agonists such as exenatide have been shown to reduce glucagon secretion from the pancreas, and it has been suggested that exenatide acts directly through alpha cells 51 . Furthermore, the mechanism by which glucose regulates glucagon secretion remains unclear and has been attributed to both intrinsic and  Fig. 6 In vivo characterization of SC-alpha cell transplantation. a Fasting-induced hypoglycemia in control (n = 9 animals) and SC-alpha transplanted mice (n = 9 animals). Continuous glucose monitoring for 24-h period at 5 weeks post transplant. Food was removed after 15 h and restored at 22 h. SC-alpha-cell-transplanted mice are protected from hypoglycemia. b Insulin tolerance test on control (n = 9 animals) and SC-alpha (n = 9 animals) transplanted mice at 6 weeks post transplant. SC-alpha-cell-transplanted mice are protected from hypoglycema. c Continuous glucose monitoring of mice for 4 weeks reveals that mice transplanted with SC-alpha cells (n = 10 animals) have elevated blood glucose compared to control animals (sham surgery, n = 10 animals). Graph represents the average daily blood glucose value at 5-min intervals for all animals in each treatment group. Each data point represents the average of 280 CGM readings (10 mice × 28 days). d In transplanted mice, SC-alpha cells reduce instances of hypoglycemia in mice. Average blood glucose reading by CGM over 28-day period (n = 10 animals for each group). e Percentage of time over 4 weeks spent in range (≥70 mg/dL) and hypoglycemic (<70 mg/dL) (n = 10 animals for each group). f Total serum glucagon levels in response to an arginine bolus at 4 weeks post transplant. SC-alpha-transplanted mice (n = 5 animals) and control animals (n = 5 animals) were fed ad libitum and injected i.p. with arginine. Total serum glucagon levels were measured at time 0 (T0), 10 and 30 min post injection. g Fasting and fed total serum glucagon levels for control (n = 5 animals) and SC-alpha (n = 5 animals) transplanted animals. Fasted mice were deprived of food for 6 h before blood collection. Blood was collected from fed animals 2 h after food was restored. h Glucose tolerance test in control mice (n = 5 animals) and SC-alpha-transplanted mice (n = 5 animals). All data are presented as mean ± SEM. Significance was calculated using a two-tailed unpaired Student's t test.
paracrine effects. The ability to generate SC-alpha cells enables the exploration of these mechanisms in the absence of paracrine interactions from beta and delta cells.
Reduced glucagon levels have been observed following treatment of type 2 diabetic patients with DPP-4 inhibitors, and since DPP-4 inhibition results in increased circulating endogenous GLP-1, a similar mechanism for GLP-1 receptor agonists has been suggested 52 . Furthermore, there have been multiple attempts to produce glucagon receptor antagonists, since the high glucagon levels observed in type 2 diabetics may contribute to hyperglycemia. On the other hand, glucagon secretagogues, such as the TRG5 agonist ABT-777, could potentially be used in the treatment of hypoglycemia for patients with type 1 diabetes. Despite these advances, much remains unknown regarding the druggability of alpha cells and no good models for functional human alpha cells exist for in vitro studies. Future advances could be accelerated by the availability of renewable sources of the alpha cells reported here.
In this study, we focus on the generation of pancreatic alpha cells as a step toward making all islet cell types, development of cell-based therapies that prevent hypoglycemia and better regulate plasma glucose concentrations, and co-culture experiments of alpha and beta cells to better understand the interplay between these cell types in islet function and disease. The scalable 3D protocol described here has some advantages over previous reports 15 , including the absence of exogenous growth factor matrices such as Matrigel™. Moreover, this protocol enables the production of hundreds of millions of SC alpha cells (~1 SCalpha cell for every three cells seeded), which is much more efficient that 2D cultures reported.
In producing SC-alpha cells in vitro, we describe the generation of pre-alpha cells that express both insulin and glucagon proteins. Similar polyhormonal cells have been described in a number of different situations, including normal development 25 , in pancreatitis and under stress conditions 23 , and as an unexpected side population in directed differentiation protocols 8,11 . Uncertainty regarding the nature of these cells has existed with some reports describing these cells as immature beta cells, others describing these cells as intermediates for alpha cells and still others describing these cells as a hybrid cell type with features of both alpha and beta cells. Cells expressing both insulin and glucagon have been observed early in mouse, rat, and human development during the primary transition of endocrine differentiation in the pancreas 28,53,54 . These early instances of polyhormonal cells do not persist into adulthood and have been postulated to resolve into alpha cells 25 . Furthermore, in humans, the existence of polyhormonal cells in the healthy adult pancreas has not been reported, although there are reports of polyhormonal cells in human pancreatic tissues associated with stress and disease 23,24 . We propose a nomenclature for alpha-like cell populations allowing the term "alpha-like" to describe any cell that has a transcriptional signature resembling an alpha cell (GCG+/ARX +/IRX1+) and designating two subsets of alpha-like cells: "prealpha" cells and "SC-alpha cells" defined by their expression or lack of INS. While alternate mechanisms for generating human pancreatic alpha cells may not require the generation of this prealpha intermediate, the present study demonstrates the potential of pancreatic pre-alpha cells to produce functional alpha cells.
As shown here, activation of PKC resulted in the maturation of pre-alpha cells into glucagon-expressing SC-alpha cells. PKC activation has been reported to enhance glucagon secretion from mouse and human islets through calcium-dependent exocytosis 51,55 . Further, a role for PKC in the transcription and protein expression of glucagon has been reported 56 . Our studies further establish the importance of PKC in alpha cells and establishes a role for PKC activation in alpha cell identity and maturation. The transcriptional changes occurring as cells mature from pre-alpha cells to SC-alpha cells contain several genes implicated in PKC signaling. PKC activation plays a role in cell stress response 57 , tetraspanin regulation 58 , and a role in Vitamin D and calcium regulation 59 . Our studies identifying PKC activation as an important signaling pathway in alpha cell identity and maturation is limited by the lack of published studies linking PKC signaling to alpha cell function. Further studies are needed to identify the precise mechanistic action of PKC activation in alpha cell maturation. These studies, as well as studies combining SC-alpha and SC-beta cells, will now be possible.

Methods
Cell culture. Human pluripotent stem cells (HUES8 and 1016 cell lines) were maintained as previously described 8 . Briefly, cells were adapted to a highly scalable 3D culture system using spinner flasks and maintained using mTeSR media. Suspension cultures were established by seeding 150 million cells in mTeSR media with 10 μM Y27632 and maintained at 70 rpm in a humidified incubator at 37°C and 5% CO 2 . Media was changed at 48 h to mTeSR without Y27632. Cells were passaged every 72 h by dispersing to single cells using Accutase and seeded into fresh mTeSR with Y27632. The Harvard University Embryonic Stem Cell Research Oversight Committee and the Mayo Clinic Stem Cell Research Oversight Committee reviewed and approved all work involving human pluripotent stem cells carried out in this manuscript including review of compliance with informed consent.
Directed differentiation. Differentiations were initiated 72 h after seeding into mTeSR at a density of 0.5 million cells per mL. Clusters were allowed to settle at the bottom of the spinner flask and the media was removed by aspiration. Protocolspecific media was introduced to the spinner flask with appropriate growth factors and the flask was returned to the incubator with stirring. Cells were directed sequentially to a definitive endoderm (DE), gut tube endoderm (GTE), pancreatic progenitor (PP), endocrine progenitor (EP) and pre-alpha (PA) population as described below. Liquid media (S1, S2, S3) composition is as previously reported 8 . Media changes are as follows-Day 1: S1 + 100 ng/mL Activin A + 3 μM CHIR99021. Day 2: S1 + 100 ng/mL Activin A.  Table 1.
Flow cytometry. Cells were collected for flow cytometry analysis at various time points throughout the differentiation process. Cells were dispersed with TrypLE Express at 37°C for 15 min after which TrypLE was quenched with S3 media. Cells were fixed in 4% paraformaldehyde (PFA) for 20 min and subsequently stored in PBS at 4°C until staining. Staining was performed by permeabilizing the cells in block solution (PBS + 0.1% Triton X-100 + 5% donkey serum) for 40 min. Cells were then incubated with primary antibodies in block solution for 1 h at room temperature (RT), washed twice with PBST and incubated with secondary antibody in block solution for 1 h at RT. Cells were then washed three times and resuspended in PBST at a concentration of 1 × 10 6 cells/mL. Stained cells were analyzed using the Accuri C6 or Attune flow cytometers. FlowJo v10 and Attune NxT ver 2.6 software were used to process the respective flow cytometry data. Insulin was assessed using an antibody that recognizes the c-peptide epitope from the Iowa Hybridoma Bank at a concentration of 1:300 (DSHB, GN-ID4). Glucagon was assessed using an antibody that recognizes the GLP-2 (Santa Cruz, SC-7781) or glucagon (Sigma, G2654) epitopes, and used at a concentration of 1:300 or 1:1000, respectively. The antibodies used recognize epitopes for both proinsulin and processed insulin as well as proglucagon and processed glucagon. For simplicity, the nomenclature insulin and glucagon is used throughout the text. Secondary antibodies used were donkey-anti-rat-Alexa Fluor 488 (Invitrogen A-21208, 1:500) and either donkey-anti-goat-Alexa Fluor 647 for GLP-2 (Invitrogen A-21447, 1:500) or donkey-anti-mouse-Alexa Fluor 647 for glucagon (Invitrogen A-31571, 1:500). The gating strategy for flow plots is shown in Supplementary Fig. 21.
Immunofluorescence. Cell clusters or kidney grafts were fixed in 4% PFA for 1 h, washed and embedded in Histogel (clusters) and subsequently embedded in paraffin and sectioned. Sections were stained for insulin and glucagon using the antibodies described above and mounted using Fluoromount-G with DAPI (4′,6diamidino-2-phenlindole). Clusters were imaged using an EVOS FL Auto 2 imaging station.
Single-cell RNA sequencing. Single-cell mRNA sequencing was carried out using the inDrops platform 26,60 on 5000 cells, as previously described using "v3" beads (1-cell bio), or the 10X Genomics Chromium system using the 3′ v3 library kit on >2000 cells per sample. Samples were collected at stage 6 day 0 and stage 6 day 28 from both PDBu+ and PDBu− conditions. Libraries were prepared from each sample and sequenced on a NextSeq 500 or HiSeq 4000. Sequencing was processed using the inDrops pipeline (https://github.com/indrops/indrops/) to generate UMIFM read counts or Cell Ranger v.3.0.2 (https://software.10xgenomics.com/ single-cell/overview/welcome) to perform barcode processing and single-cell gene UMI counting.
For inDrops: Read counts were converted to transcript per million (tpm) values, excluding genes that contribute more than 2% of reads in any given cell. Highvariance genes were identified as previously described 26 and used for principal component analysis (PCA). tSNE projection was carried out using the first 25 components (not rescaled to unit variance), using the Python wrapper of the Barnes-Hut C tSNE implementation (https://github.com/lvdmaaten/bhtsne). Clustering was performed using the principal components as input for diffusion map embedding and Louvain community detection, as implemented in Scanpy 61 . For 10X Chromium: Data were analyzed using Seurat v.3 62,63 . Briefly, cells were removed if they had <500 detected genes or contained >25% mitochondrial content. The remaining cells were regressed for variance in mitochondrial genes. Raw UMI counts were normalized per million total counts and log-transformed, with additional batch correction performed using the Seurat functions FindIntegrationAnchors and IntegrateData. A tSNE projection was carried out using the first 20 components, defining clusters which correlated with pre-alpha (GCG+, INS+) cells and alpha (GCG+, INS−) cells. Only these clusters were input into Monocle 2 for further analysis. Trajectory analysis was performed in Monocle 2 using semi-supervised ordering, identifying 50 genes that covaried with a change in insulin expression 64 . These genes were used to order the pre-alpha and alpha cells along a pseudotime trajectory. Hierarchical clustering was performed using the hclust function in R, which uses an agglomerative approach to group using the default complete linkage method. Venn diagrams and heatmaps were created by performing gene ontology analysis using David 6. Hormone secretion assays. Human islets (Prodo Laboratories) or differentiated pre-alpha or SC-alpha cells were washed twice in low-glucose (2.8 mM) Krebs Ringer (KRB) buffer (128 mM NaCl, 5 mM KCl, 2.7 mM CaCl 2 , 1.2 mM MgSO 4 , 1 mM Na 2 HPO 4 , 1.2 mM KH 2 PO 4 , 5 mM NaHCO 3 , 10 mM HEPES (4-(2-hydroxyethyl)-1-piperazineethanesufonic acid), 0.1% BSA (bovine serum albumin) in MilliQ). Cells were then loaded into 24-well transwell inserts and fasted in lowglucose KRB for 1 h at 37°C. Clusters were washed once in low-glucose KRB and then incubated in low-glucose KRB for 1 h at 37°C. After incubation, the supernatant was collected and stored at −20°C until analysis. Cells were then transferred to high glucose KRB (20 mM) for 1 h at 37°C and the supernatant was collected and stored. Cells were then transferred to low-glucose KRB with 30 mM KCl to observe depolarization conditions. Cells were incubated in this buffer for 1 h and the supernatant was collected. Cells were then dispersed and counted using a Vi-CELL automated cell counter. Collected supernatants were analyzed by ELISA for human insulin (ALPCO) and/or total glucagon (Mercodia) concentrations and normalized for cell number. For animal transplant studies, total serum glucagon is measured since mouse and human glucagon sequences are identical.
Chemical screen. Pre-alpha cell clusters were dispersed using TrypLE Express for 15 min and quenched using S3 media. The resulting single-cell suspension was arrayed into a 384-well plate, which had been coated with Matrigel with 50,000 cells in each well. Compounds were introduced each to five separate wells and incubated for 96 h. Media was changed after 48 h and at 96 h cells were fixed with 4% PFA and stained for insulin and glucagon as described above with the following modifications. Antibody incubations and washes were performed using a BioTek EL405 plate washer and dispenser. Cells were imaged with the Cellomics ArrayScan high-content imaging system. Nine fields of view were captured for each well and analyzed for cell number and fluorescent content. The percentage of cells expressing insulin and glucagon protein was calculated for each well.
Electron microscopy. To analyze the granule ultrastructure, human islets or differentiated SC-alpha cells were fixed for 2 h at RT in fixative (2.5% glutaraldehyde, 1.25% PFA and 0.03% picric acid in 0.1 M sodium cacodylate buffer, pH 7.4). Subsequently, islets or SC-alpha cells were washed in 0.1 M cacodylate buffer and post-fixed with 1% Osmium tetroxide (OsO 4 )/1.5% potassium ferrocyanide (KFeCN 6 ) for 1 h, washed 2× in water, 1× Maleate buffer (MB) and incubated in 1% uranyl acetate in MB for 1 h followed by two washes in water and subsequent dehydration in grades of alcohol (10 min each; 50%, 70%, 90%, 2 × 10 min 100%). The samples were then put in propylene oxide for 1 h and infiltrated overnight in a 1:1 mixture of propylene oxide and TAAB Epon (Marivac Canada Inc.). The following day, the samples were embedded in TAAB Epon and polymerized at 60°C for 48 h.
Cell transplantation. All animal studies were conducted with approval from appropriate institutional oversight committees: Mayo Clinic Institutional Animal Care and Use Committee; Harvard University/Faculty of Arts & Sciences (HU/ FAS) Standing Committee on the Use of Animals in Research and Teaching, Local Ethics Review Committee for Animal Experiments (Gothenburg, Sweden). Presurgery, animals were housed in groups within sterile cages with unrestricted access to food and water. Ambient temperature was maintained between 18 and 25°C, humidity 30−70% with 12 h light/dark cycles. Transplantation of cell clusters was performed as previously described 8 . Briefly, 5 × 10 6 cells were injected under the kidney capsule of male SCID-beige mice. Control mice underwent a mock surgery where saline was injected into the kidney capsule. Post-surgery, mice were single housed and monitored for up to 8 weeks after transplantation. Kidney grafts were harvested at various time points and stained for insulin and glucagon as described above.
Continuous glucose monitoring. Transplanted mice were monitored using a Dexcom G4 continuous glucose monitoring (CGM) system. Each mouse was anesthetized with isoflurane prior to application of the CGM sensor and transmitter. Sensors were inserted into the subcutaneous space on the back of the mice and attached using wound clips. The adhesive patch for each sensor was trimmed to approximately 5 mm around the sensor and wound clips were used to strengthen adhesion. Sensors were calibrated twice a day using a handheld glucometer (Accu-Chek) and tail bleeds. Sensors were reset (stopped and restarted) every 7 days throughout the study (4 weeks total). After reset, sensors went through a 2-h calibration period in which no interstitial glucose readings were collected. As has been previously described 65 interstitial glucose readings were recorded and used to approximate blood glucose concentrations.
Electrophysiology. Electrical activity was recorded from alpha cells in intact human pancreatic islets using methods similar to those previously established for mouse islets 39 . Measurements from SC-alpha cells were made essentially as described for dispersed human islet cells 37 . For both cell types, measurements were performed from metabolically intact cell using the perforated patch whole-cell technique.
Data analysis and calculations. Data processing of the CGM-measured glucose concentrations was conducted using MATLAB 2015b (MathWorks). The ultradian period of the CGM-measured glucose rhythms was quantified using autocorrelation analysis as previously described for the characterization of pulsatile insulin secretion 42 . MATLAB 2015b was used to calculate correlation coefficients of the CGM-measured glucose concentrations. The dominant circadian period in glucose concentration was assessed using Lomb-Scargle periodogram analysis using ClockLab (Actimetrics).
Statistics and data reproducibility. Statistical analysis was performed using either ANOVA or a paired or unpaired Student's t test, where appropriate (GraphPad Prism v.8.0). Representative images from Figs. 1d, 3e and Supplementary Fig. S10e, f were performed in three biologically independent samples. Images from Fig. 5a−i and Supplementary Fig. S4 were performed on one sample and are representative of three separate images obtained per sample. Western blots performed in Supplementary Figs. S2c and S18 were performed once.

Data availability
Full scans of Western blots are shown in Supplementary Fig. 22. Raw and processed single-cell RNA sequencing data that support the findings of this study are available in the Gene Expression Omnibus under the accession number GSE138857. Human islet data were obtained from publicly available sources, GEO: GSE84133 and GEO: GSE114297. All other data are available from the corresponding authors upon reasonable request.