Diabetes relief in mice by glucose-sensing insulin-secreting human α-cells

Abstract

Cell-identity switches, in which terminally differentiated cells are converted into different cell types when stressed, represent a widespread regenerative strategy in animals, yet they are poorly documented in mammals. In mice, some glucagon-producing pancreatic α-cells and somatostatin-producing δ-cells become insulin-expressing cells after the ablation of insulin-secreting β-cells, thus promoting diabetes recovery. Whether human islets also display this plasticity, especially in diabetic conditions, remains unknown. Here we show that islet non-β-cells, namely α-cells and pancreatic polypeptide (PPY)-producing γ-cells, obtained from deceased non-diabetic or diabetic human donors, can be lineage-traced and reprogrammed by the transcription factors PDX1 and MAFA to produce and secrete insulin in response to glucose. When transplanted into diabetic mice, converted human α-cells reverse diabetes and continue to produce insulin even after six months. Notably, insulin-producing α-cells maintain expression of α-cell markers, as seen by deep transcriptomic and proteomic characterization. These observations provide conceptual evidence and a molecular framework for a mechanistic understanding of in situ cell plasticity as a treatment for diabetes and other degenerative diseases.

Extended Data Fig. 1.. Sorted islet cell-types are highly pure and efficiently labeled.

(a) Human islets were dissociated into single cells and antibody-labeled for FACS-sorting. Representative FACS plots illustrating cells labeled with pan-endocrine marker HIC1–2B4 and non-β endocrine marker HIC3–2D12 ^,. Sorted islet cell-fractions were immunostained for insulin (INS), glucagon (GCG), somatostatin (SST), PP (PPY) and ghrelin (GHRL), and counted. All sorted cells were mono-hormonal. Other staining data with higher magnification or tile scanning are shown in Supplementary Data 1–3. Scale bars: 250 μm. All FACS results (n = 42 different donors) are also summarized in Supplementary Table 2. (b) Sorting results showing cell purity of islet cell types. Purity of α-cells and β-cells in most islet preparations displayed a 99% purity (98.9±0.8% and 98.9±0.5%, respectively), but PPY⁺ γ-cells showed great batch-to-batch variability (up to 98% purity), with α- or ghrelin⁺ ε-cell contamination, but without β-cells (less than 0.5%). Only sorted cells with high purity (> 99% for α/β cells and > 90% for γ-cells) were used in experiments. (c) qPCR of hormonal expression (INS, GCG, SST, PPY) in all sorted fractions (α-, β-, γ-cells) that were used for experiments. qPCR of INS in sorted α-cells shows very rare contamination of β-cells, which is consistent with estimated purity calculated by previously published method . (d) Sorted α-cells were transduced with Ad-GFP, reaggregated into pseudoislets and cultured for 7 or 14 days. To evaluate transduction efficiency, pseudoislets were dissociated again into single cells and FACS-analyzed. More than 99% α-cells expressed GFP, while non-transduced α-cells did not. FACS plots are representative from 3 independent donor samples. All values of % purity or contamination (a-c) are mean ± SD. n = 41 donors for α-cells, n = 42 donors for β-cells and n = 5 donors for γ-cells. For details, see Supplementary Table 2.

Extended Data Fig. 2.. Reaggregation of dispersed purified human β-cells.

(a) Pure β-cells were labeled with GFP and traced in 3 different culture conditions: in monolayer (“single β”), β-cell-only aggregation (“β”), or β-cell aggregation with stromal cells including HUVECs and MSCs (“β+HM”). Live imaging at indicated days (middle panels) and immunofluorescence at day 7 (right panels) show β-cell-only pseudoislets were self-organized by Day 5, whereas β+HM aggregates were constituted in only 1 day. β-cells in β+HM pseudoislets located at the periphery, while HM cells formed the core of the aggregates (red and blue, respectively). (b) To determine the optimal number of β-cells per pseudoislet, GFP⁺ transduced β-cells were seeded on aggregation-plate-wells at the indicated densities. 7 days after culture, aggregates were harvested and analyzed. Aggregates were uniform in size. Pseudoislet size correlated with the number of cells seeded per well. It was reported that human islet cell aggregates with a diameter 100—150 μm, consisting of 1000 cells, show a comparable function to native islets ; we thus decided to perform reaggregation experiments at 1000 β-cells/pseudoislet (1000 β-cells, 129.6±3.1 μm). β-cell aggregates with HM were also analyzed. n = 8 pseudoislets for 500 β-cells, n = 8 pseudoislets for 1000 β-cells, n = 9 pseudoislets for 2000 β-cells, n = 8 pseudoislets for 3000 β-cells, and n = 52 pseudoislets for 1000 β-cells + 400 HUVECs + 100 MSCs. (c) Immunofluorescence at indicated time-points in β-cell pseudoislets and β-cell+HM pseudoislets. (d) TUNEL staining (green) showed rare apoptotic cells (0.8%) in β-cell aggregates after 7 day-culture. (e) qPCR analyses of INSULIN and PDX1 expression in monolayer and aggregated β-cells showing higher expression of β-cell markers in reaggregated β-cells. Data are expressed as fold-change relative to the value in single β-cells. *p = 0.022 in qPCR for INS, * p = 0.026 in qPCR for PDX1, Mann-Whitney test, two-tailed. n = 6 donor samples. (f) ELISA measurements of static glucose-stimulated human insulin release at 3 mM (Low) and 20 mM (Hi) glucose showing glucose-responsive C-peptide secretion in both β and β+HM aggregates, but not in single β-cells. ****p < 0.0001, **p = 0.0037, * p = 0.012, two-way RM ANOVA with Holm-Sidak’s multiple comparisons test, n = 5 for single β-cells, n =8 for β and β+HM, n = 6 for native islets (all are biological replications from different donors). (g) Stimulation index in glucose-stimulated insulin secretion in (f) exhibiting comparable values among pseudoislets of β and β+HM and native islets. **p = 0.0089, *p = 0.045, one-way ANOVA with Benjamini, Krieger and Yekutieli’s multiple comparisons test. All images are representative from 5 (a,c,d) or 3 (b) independent experiments. ns: no statistical significance. All data shown are means ± s.e.m. Scale bars: 25 μm (a), 50 μm (c,d), 100 μm (b).

Extended Data Fig. 3.. Assessment of the effect of transcription factor expression on insulin production in human α-cells.

(a) Representative immunofluorescence images at 7 days of reaggregation. To determine the best α-to-β-cell reprogramming factors, human α-cells were transduced with adenoviral vectors, including Pdx1, MafA and Nkx6.1, in all combinations, and then reaggregated. Images are representative from n = 38 donors for αGFP and αPM, n = 5 for αPdx1 and α3TFs, n = 3 for αMafA, αNkx6.1, αMN6 and αPN6. (b) qPCR analysis of human insulin expression in α-cells transduced with indicated reprogramming factors, 7 days after aggregation. PM: Pdx1+MafA, MN6: MafA+Nkx6.1, PN6: Pdx1+Nkx6.1, 3TFs: Pdx1+MafA+Nkx6.1. ****p < 0.0001, ***p = 0.0006 versus αGFP control, one-way ANOVA with Tukey’s multiple comparisons test. n = 7 for αGFP, αPdx1, αPM and α3TFs, n = 5 for αMafA, αNkx6.1, αMN6, αPN6: all are biological replications from different donors. (c) Percentages of bihormonal cells (expressing insulin and glucagon) in αPM single cells, αPM-only pseudoislets and αPM+HM pseudoislets. One-way ANOVA with Tukey’s multiple comparisons, n = 3 from different donors. (d) qPCR analyses in αGFP and αPM pseudoislets cultured for 7 days. αPM cells have less glucagon expression than αGFP pseudoislets, but ARX expression is still maintained. *p = 0.015, Mann-Whitney test, two-tailed, n = 6 from different donors per group. (e) qPCR analysis for insulin expression in αGFP and αPM cells cultured in monolayer or pseudoislets. αPM single cells have less insulin expression than αPM pseudoislets; αGFP controls display only background levels. ***p = 0.0002, †††p = 0.0009, one-way ANOVA with Tukey’s multiple comparisons test, n = 4 from different donors. (f) Live-imaging of in vitro pseudoislet formation using αPM and HM cells. Aggregation is faster with HM cells. (g) Single α-cells transduced with PM show very rare reprogramming events (i.e. insulin production), whilst re-aggregated α-cells display high reprogramming efficiency. α-cells (GFP⁺, green) locate at the periphery of pseudoislets containing also HM cells (HUVECs/MSCs: only DAPI⁺, white). (h) Immunofluorescence for PDX1, NKX6–1 and insulin on αGFP, αPM and αPM+HM pseudoislets after 7 days of culture. Reprogrammed α-cells express insulin (red) and PDX1 (green), but not NKX6–1 (blue) in αPM and αPM+HM aggregates. (i) TUNEL staining (green) reveal almost no apoptosis in α-cell pseudoislets in 7-day cultures (1.8% in αGFP, 1.4% in αPM, 1.6% in αPM+HM). (j) Proliferation marker pHH3 staining (green) reveal almost no proliferation (< 1%) both in αGFP and αPM pseudoislets after 7-day cultures. Images are representative from 3 different donors (f-i). ns: no significance. All data shown are means ± s.e.m. Scale bars: 25 μm.

Extended Data Fig. 4.. γ-cell reprogramming and in vitro kinetics of cell number and gene expression levels in pseudoislets.

(a) Live imaging during reaggregation into pseudoislets of GFP-transduced γ-cells. Like for α-cell pseudoislets (Fig. 1), sorted γ-cells were transduced with adenoviral vectors, and then seeded on reaggregation plates. (b) Reprogramming efficiency into insulin expression (b, % of insulin⁺ cells out of GFP⁺ cells). Seven days after aggregation, γ-cells transduced with the indicated reprogramming factors show the highest reprogramming incidence (PM combination). ****p < 0.0001, *p = 0.046 vs γGFP control; †††p = 0.0008 versus γPM, one-way ANOVA with Tukey’s multiple comparisons test. n = 3 from different donors. (c) qPCR analysis of human insulin gene expression in α-cell pseudoislets. Seven days after aggregation, γ-cells transduced with the indicated reprogramming factors show the highest reprogramming incidence (PM combination). ***p = 0.0009 vs γGFP control, one-way ANOVA with Tukey’s multiple comparisons test. n = 3 from different donors. Data are mean ± s.e.m. (d) Immunofluorescence of γGFP and γPM pseudoislets after 7 days in culture. Most insulin-expressing reprogrammed γ-cells maintain PPY expression (blue in inset). (e) Live imaging of aggregated transduced γ-cells. γPM+HM pseudoislets form faster than γ-cell-only pseudoislets (a). Scale bar: 25 μm. (f, g) Control γ-cells in γGFP+HM pseudoislets do not reprogram (insulin production) (< 1%; left panel in f), yet PM-transduced γ-cells become insulin-producers in γPM+HM pseudoislets (right panel in f). The architecture of γPM+HM pseudoislets is similar that of αPM+HM pseudoislets; however, γ-cell reprogramming efficiency in γPM+HM clusters is lower than in γPM-only pseudoislets (g). *p = 0.029, Mann-Whitney test, two-tailed. n = 4 from different donors. Data are mean ± s.e.m. Scale bars: 25 μm. (h) Glucose-stimulated insulin secretion: γ-cells in γPM-only pseudoislets efficiently secrete insulin in response to glucose stimulation in vitro, but not γGFP pseudoislets. Interestingly, they secrete insulin better than α-cells in αPM+HM pseudoislets (1.24 fmol/10³ cells vs 0.27 fmol/10³ cells for converted α-cells; h and Fig. 1f). ns (not significant): p = 0.82, **p = 0.0043, two-way RM ANOVA with Holm-Sidak’s multiple comparisons test. n = 3 donor samples. All data shown are means ± s.e.m. Scale bars: 25 μm. All images are representative of 3 (a,e) or 4 (d,f) independent experiments. (i) DNA content was measured by Pico-green tests to assess cell number kinetics in pseudoislets in vitro. Cell numbers dropped mainly between 1 and 2 weeks in α- and β-cell pseudoislets. n = 3 from different donors. Data are mean ± SD. (j) Expression levels of INSULIN and adenoviral vector mouse Pdx1 and MafA were also evaluated by qPCR at indicated time-points. Surprisingly, insulin expression levels were increased with time. Exogenous adenoviral Pdx1 and MafA expression levels were also maintained for 8 weeks in vitro. n = 3 from different donors. Data are mean ± SD.

Extended Data Fig. 5.. αPM+HM pseudoislets secrete human insulin upon glucose stimulation in healthy NSG host mice.

(a) αPM+HM pseudoislets generated from non-diabetic donor samples were transplanted under the kidney capsule of non-diabetic NSG mice. (b) Four weeks after transplantation, in vivo glucose-stimulation tests were performed. Circulating human C-peptide levels are higher after glucose injection compared to before the injection. *p = 0.031, Mann-Whitney test, two-tailed. n = 10 from 6 different donors. (c) Immunostaining picture of control αGFP+HM grafts 4 weeks after transplantation under the kidney capsule of host NSG mice; there is almost no reprogramming into insulin production (< 1%). (d—g) αPM+HM grafts show vascular formation around the grafts and retain abundant GFP-labeled insulin-expressing cells two weeks (d) and 4 weeks (e) after transplantation. Interestingly, αPM+HM grafts display increased reprogramming efficiency (% of insulin⁺ cells out of GFP⁺ cells; e, f) and fractions of monohormonal INS⁺ cell (4 weeks after transplantation; e, g). *p = 0.029, Mann-Whitney test, two-tailed. n = 4 from different donors. All data shown are means ± s.e.m. Scale bars: 25 μm. All images are representative from 6 (a), 3 (c), 2 (d) or 4 (e) different donors.

Extended Data Fig. 6.. Small number of human α-cell pseudoislets is sufficient to ameliorate diabetes in mice.

(a) Experimental time-line of Exp. #2. NSG mice were made diabetic with STZ and a week later were transplanted with 200–1000 αPM+HM pseudoislets (“STZ αPM+HM”) obtained from 3 donors (n = 3). As controls, STZ-diabetic mice were implanted with either no graft (“STZ no graft”) or the same number of native human islets (“STZ islets”). Non-diabetic NSG control mice were also monitored. Nephrectomy (“Nx”) was performed 4 weeks after transplantation for graft removal. For details, see Supplementary Table 5. (b) Random-fed blood glucose throughout Exp. #2. There is no significant improvement of hyperglycemia in STZ αPM mice (see Extended Data Fig. 6a for the areas under the curves of the engraftment period). n = 3 mice grafted with 3 different donors in all groups. (c) Body weight changes after STZ injection. There is body weight gain after transplantation with intact islets or αPM pseudoislets, and continuous weight loss in untransplanted diabetic controls (see Extended Data Fig. 6b for the areas under the curves of the engraftment period). Graft removal breaks this trend. n = 3 mice grafted with 3 different donors in all groups. (d) The “area under the curve” of random-fed blood glucose measurements during the engraftment period (indicated in yellow in b) show significant hyperglycemia improvement (lowering) only in mice engrafted with native islets. *p = 0.022, one-way RM ANOVA with Holm-Sidak’s multiple comparisons test. n = 3 different donors. (e) The area under the curve of body weight changes during engraftment (indicated in yellow in c) show significant body weight gain after transplantation with intact islets (“STZ islets” group) and αPM+HM pseudoislets (“STZ αPM” group). ****p < 0.0001, **p = 0.003, *p = 0.042, one-way RM ANOVA with Holm-Sidak’s multiple comparisons test. n = 3 different donors. (f, g) Glucose tolerance test at 4 weeks after transplantation (f) and after graft removal (g). Engrafted mice display recovery after 3 hours (f), yet this capacity is lost upon graft removal (g). Analysis of the area under the curve in f and g are shown in Extended Data Fig. 6h. ****p < 0.0001, ***p = 0.0007, ###p = 0.0009, versus STZ no graft, two-way RM ANOVA with Dunnett’s multiple comparisons test. n = 3 mice grafted from different donors (STZ αPM, STZ islets), n = 7 mice (STZ no graft), n = 6 mice (no STZ no graft). (h) Area under the curve of ipGTT at 4 weeks after transplantation (see also f) and 2 weeks after graft removal (see also g). There is partial STZ-diabetes recovery in “STZ αPM” mice, but not in “STZ no graft” group (left in h). After graft removal, both “STZ islets” and “STZ αPM“ groups become hyperglycemic again (right in h), proving that improvement in glucose tolerance and weight gain is graft-dependent. ****p = 0.00007, **p = 0.0025, ####p = 0.00002, ##p = 0.0014, one-way ANOVA with Benjamini, Krieger and Yekutieli’s multiple comparisons test. n = 3 mice grafted from different donors (STZ αPM, STZ islets), n = 7 mice (STZ no graft), n = 6 mice (no STZ no graft). (i) Blood human C-peptide levels in mice measured before (“0 min”) and after (“15 min”) glucose injection. Glucose-responsive C-peptide release is observed in mice bearing human islets or αPM pseudoislets. n.d., undetectable. **p = 0.0015, ****p = 5 × 10⁻⁸, two-way RM ANOVA with Holm-Sidak’s multiple comparisons test. n = 3 mice grafted from different donors (STZ αPM, STZ islets), n = 7 mice (STZ no graft), n = 6 mice (no STZ no graft). (j) Immunofluorescence of pancreas in the NSG RIP-DTR mouse that was transplanted with αPM pseudoislets (Exp.#3: DT+αPM, Figure 3b) shows that endogenous mouse β-cells were well-ablated and did not regenerated, suggesting improvement of diabetic symptoms was dependent on human αPM graft. Images are representative from 9 different diabetic mice after DT injection. (k) Immunofluorescence of pseudoislet kidney grafts in “STZ αPM” (upper panels) and “STZ islets” mice (lower panels), 4 weeks after transplantation. Monohormonal insulin-producing cells with GFP-tracer are abundant in the engrafted αPM+HM pseudoislets. Images are representative from n =4 mice with different donors’ grafts. (l) Body weight changes in experimental animals of Figure 3b (Exp. #3). After DT injection, untransplanted diabetic controls (DT + no Graft) exhibited continuous weight loss, while there is body weight gain after transplantation with intact islets or αPM pseudoislets. Nx: nephrectomy for graft removal. Error bars: SD. (m,n) Intraperitoneal glucose tolerance test (ipGTT) at 7 weeks after transplantation (related to Fig. 3b-d), shows significantly improved glucose tolerance both in DT+αPM and DT+islets groups. ****p < 0.0001, **p = 0.002, *p = 0.0338, versus DT + no graft, two-way RM ANOVA with Dunnett’s multiple comparisons test. Groups are indicated by same colored lines as Fig. 3b,c. n = 3 for DT + islets, n = 1 for DT + αPM, n = 5 for DT + no graft, and n = 5 for no DT + no graft (l,m). n = 2 for DT + islets, n = 1 for DT + αPM, n = 3 for DT + no graft, and n = 4 for no DT + no graft (n) (o) αPM pseudoislets grafted into mouse kidney show innervation (TH⁺) and vascularization (CD31⁺) 1 month after transplantation. (p) Proliferation marker pHH3 staining on grafts of αPM+HM 4 weeks after transplantation, showing almost no proliferative cells in grafts (< 1%). (q,r) Immunofluorescence of grafted αPM pseudoislets shows reprogrammed α-cells express insulin as well as GFP tracer at 3 moths after transplantation (q) and 6 months after transplantation (r), confirming a stable phenotype of αPM cells under in vivo environment. Left panels in r are confocal tile-scan images which were merged as a maximum projection. We did not detect any SST, PPY or GHRL positive cells. Black-line: non-grafted diabetic mice; red-line: diabetic mice with αPM+HM graft; blue-line: diabetic mice with native islet graft; black-dotted-line: healthy mice in b-I, l-n. All data shown are means ± s.e.m. (except in l; bars are SD). Scale bars: 50 μm. All images are representative from 9 different diabetic mice after DT injection (j), from n =4 (k), n=3 (o,p), n=1 (q), or n=1 (r) donors.

Extended Data Fig. 7.. Reprogrammed α-cells from diabetic donors lead to diabetes remission in mice.

(a) Human islets from T2D donors were dissociated into single cells and antibody-labeled for FACS-sorting. Representative FACS plots showing cells labeled with pan-endocrine marker HIC1–2B4 and non-β-cell endocrine marker HIC3–2D12 . The purity of sorted islet cells was evaluated. FACS plots are representative from 3 T2D donors. (b, c) Reprogramming efficiency into insulin production (% of Insulin⁺ GFP⁺/ GFP⁺ cells in b) and qPCR analysis of human insulin gene expression (c), 7 days after aggregation of α-cells transduced with the indicated reprogramming factors. Pdx1 and MafA combined (“αPM”) trigger the highest reprogramming efficiency. ****p < 0.0001, *p = 0.031 versus αGFP control; ## p = 0.0055 versus αPM, one-way ANOVA with Tukey’s multiple comparisons test. n = 3 different T2D donors. (d) Representative immunostaining at culture day 7 in αGFP and αPM pseudoislets from 3 T2D donors. (e) Cartoon depicting transplantation experiment using 2 consecutive islet preparations from 2 different T2D deceased patients. First, 2,300 reconstituted αPM+HM pseudoislets were transplanted under the capsule of the left kidney (ventral side) of an STZ-treated diabetic NSG host mouse. Fortuitously, 2 weeks later, T2D islets were again available, and 1,450 new αPM+HM pseudoislets were generated and engrafted into the dorsal side of the same kidney. (f) Experimental time-line of Exp. #5. Sequential transplantation was performed using human α-cells of T2D donors to rescue STZ-diabetes, followed by anti-Glucagon receptor antibody (GCGR-Ab) treatment for 2 weeks. Graft was removed 1 week after GCGR-Ab therapy. The following week, GCGR-Ab treatment was stopped. (g) Random-fed blood glucose throughout Exp. #5. Before glucagon inhibition, there is a mild amelioration of hyperglycemia in the mouse bearing 2 grafts of T2D αPM pseudoislets, yet is less marked than in the mouse that received T2D islets. Under glucagon receptor antibody treatment (“GCGR-Ab”), glycemia drastically and quickly drops in both engrafted mice. Graft removal quickly leads to hyperglycemia, even under glucagon signaling inhibition. (h) Glucose tolerance tests before the 2^nd transplantation (3 weeks after 1^st transplantation), 4 weeks after the 2^nd transplantation, and after graft removal (“post Nx”). There is improved glucose tolerance in diabetic mice transplanted with “αPM+HM” pseudoislets (red line in left and middle panels), relative to untransplanted diabetic controls (black line). (I,j) Circulating human C-peptide after 1^st transplantation (i). The data after 2^nd transplantation were also shown in Fig. 3e. In vivo stimulation index (of insulin secretion) following a glucose challenge is similar in native T2D islets and T2D αPM pseudoislets (j). (k) Immunofluorescence of engrafted T2D αPM pseudoislets. Insulin-expressing (red) reprogrammed α-cells (GFP⁺, green) are abundant and do not contain glucagon (blue). (l, m) Reprogramming efficiency (l) and percentage of monohormonal insulin-producing cells (m) in αPM pseudoislets from T2D donors before transplantation (“pre Tx”) and after transplantation (“post Tx”). ***p = 0.0005, ** p = 0.0082, paired t test, two-tailed. n = 3 donors with T2D from 1^st, 2^nd grafts and independent cohort. (n) Immunofluorescence for PDX1, MAFA and INS of the graft of T2D intact islets (left) or T2D αPM+HM cells (right) 9 weeks after transplantation. Reprogrammed α-cells express insulin (red), PDX1 (green) and MAFA (blue). (o) qPCR analyses in αGFP, αPM aggregates in vitro (before transplantation) and αPM+HM pseudoislets in vivo (after transplantation). Transplanted αPM cells express higher insulin compared to that before transplantation, but still maintained ARX expression. Although endogenous expression levels of human β-cell TFs (PDX1, MafA, NKX6.1) were not changed significantly 7 days after transduction in vitro, their expression in αPM grafts was significantly increased after transplantation. Gene expression levels were normalized to GFP expression. ***p < 0.001, **p < 0.01, *p < 0.05, one-way ANOVA with Holm-Sidak’s multiple comparisons test. n = 3 different T2D donors for αGFP and αPM in vitro. n = 2 different T2D donors for graft of αPM+HM. (p) Transmission electron micrographs of a β-cell in an engrafted T2D islet (left) and of 2 reprogrammed α-cells in engrafted αPM+HM pseudoislets (right). T2D β-cells do not contain abundant insulin granules, as previously reported . Reprogrammed α-cells contain abundant β-like granules, with the typical crystalized dense core surrounded by a clear halo. (q) TUNEL staining (green) showed very rare apoptosis events (less than 1%) in the graft of αPM+HM pseudoislets 9 weeks after transplantation. Black line: non-grafted STZ mice (n =4); red line: STZ mice (n = 1) with αPM+HM graft from 2 donors; blue line: STZ mice (n = 1) with native islet graft; black-dotted-line: healthy mice (n = 2) in g and h. n.d.: not detected. All data shown are means ± s.e.m. Scale bars: 25 μm (d, k, n, q); 500 nm (p). Images are representative from 3 different T2D donors’ grafts (k,n,q), and from 2 different T2D donors (p).

Extended Data Fig. 8.. Transcriptomic analyses.

(a,b) Gene-set enrichment analysis (GSEA) using the transcriptomes of sorted α-cells and αGFP pseudoislets. There is a significant enriched expression of genes associated with β-cell function such as mitochondrial “oxidative phosphorylation” and “respiratory chain” in αGFP pseudoislets compared to sorted α-cells (a). Heatmaps of transcriptomic expression levels of listed genes tested in GSEA (b). Gene sets were taken from GO gene sets of Molecular Signatures Database v6.0. (c) Top 15 Canonical pathways that differ between sorted single α-cells and αGFP pseudoislets. Inflammatory/stress-related pathways were downregulated in αGFP pseudoislets compared to sorted α-cell singlets. Pathway analyses were done by IPA. (d) Heatmaps of transcriptomic expression levels of gene sets in αPM pseudoislets and sorted α-cells (related to GSEA in Figure 4g). (e) Volcano plot representing differentially-expressed genes (DEGs) in sorted α-cells and β-cells (FDR < 0.05, FC > 2). 887 α-cell-enriched genes were identified. The blue box delimitates α-cell-enriched genes, i.e. the “α-cell signature” (Supplementary Table 14). (f-h) Changes in expression of ‘α-cell signature’ genes caused by reaggregation and PM effects. Volcano plots showing DEGs in αGFP pseudoislets relative to sorted α-cells (f) that characterize the cell reaggregation effect, in αPM pseudoislets relative to αGFP pseudoislets (g), reflecting the effect of Pdx1 and MafA expression, and in αPM pseudoislets relative to sorted α-cells (h), which reflects the combined effect of cell reaggregation and transcription factor expression. Downregulated DEGs in each condition (colored squares on volcano plots of f-h) were overlapped with the α-cell-enriched gene list of e, as a measure of repressed α-cell signature. The Venn diagrams show that 218 α-enriched genes are downregulated in α-cells upon aggregation (f, Supplementary Table 15), 120 genes upon Pdx1 and MafA activation (g, Supplementary Table 16), and in total 272 “α-like genes” are downregulated in α-cells as a result of the combined effect of cell aggregation and transcription factor expression (h, Supplementary Table 17). DEGs: FDR < 0.05. (i) In vivo effect on αPM cells at 1 month after transplantation. DEGs (FDR < 0.05) between αPM pseudoislets before transplantation (n = 7) and in grafted αPM pseudoislets (n = 5) were analysed with IPA to identify downstream effects. Several pathways were activated, including “synthesis of hormone”, “secretion of secretory granules” and “innervation”. See details in Supplementary Table 20. n = 5 grafts from 9 non-diabetic donors were retrieved from the mouse renal capsules, FACS-sorted with GFP, and analyzed by bulk RNA-Seq. (j) Effect of cell aggregation and reprogramming factor expression on human α-cell identity. Reaggregation of dispersed α-cells and expression of the transcription factors Pdx1 and MafA (PM) promotes the upregulation of a subset of β-cell-enriched genes (“β-cell signature”), which is sufficient to confer a glucose-stimulated insulin secretion activity (GSIS) to α-cells in monotypic αPM pseudoislets. Concomitantly, some but not all the α-cell-enriched genes (“α-cell signature”) are downregulated in αPM pseudoislets, leading to a hybrid “α/β” signature. Figures and histograms represent the number of genes affected. (k) Hierarchical clustering analysis in proteomic in vitro samples showing protein signatures of αPM pseudoislets are closer to β-cells.

Extended Data Fig. 9.. Single-cell RNA-Sequencing.

(a) Schematic of scRNA-Seq analyses and pseudotemporal ordering. Monotypic pseudoislets containing labeled human α- (“αGFP” or “αPM”) or β-cells (“βGFP”) were cultured for 1 week and sorted into single cells. Microfluidic device encapsulated each cell individually with a barcoded primer bead in a droplet. cDNA libraries were constructed and sequenced. In silico cell-lineage reconstruction during reprogramming was performed by pseudotime analysis, to dissect the reprogramming path/trajectory. αGFP and βGFP pseudoislets were analyzed as controls. (b) Gene expression of GCG and INS on t-SNE of single-cell transcriptomes from αGFP (n = 47), αPM (n = 434) and βGFP pseudoislets (n = 51) after 1-week-culture (related to Fig. 5a). (c) Cell clustering of αPM cells (n = 434) based on the state along pseudotime trajectory (related to Fig. 5b), showing 10 different states. Although 4 small branches were detected near the main path, most cells were distributed along main stem. (d) Gene expression of INS and GCG on pseudotime trajectory of αPM cells (related to Fig. 5b). (e) Cell distributions of pseudotime-based “early”, “mid” and “late” αPM cells on t-SNE map (related to Fig. 5f). (f) Kinetics of gene expression along pseudotime progression in αPM cells (n = 434) (related to Fig. 5e). (g) ARX expression in cell clusters from t-SNE and pseudotemporal ordering analysis.

Extended Data Fig. 10.. Evaluation of the specificity and cytotoxic properties of CTL clones.

(a) Design of cytotoxic T lymphocyte (CTL) killing assays. As target cells, monotypic pseudoislets (αGFP, αPM, or βGFP) after 1–2 week culture were dissociated and labelled with chromium (⁵¹Cr). In some control conditions, islet cells were loaded with either DRiP or PPI peptide epitopes. Then, target cells were co-cultured with effector cells (CTLs), which were either CMV-directed (CMV: negative control clones), DRiP-directed (DRiP: targeting stressed β-cells), PPI-directed (PPI: recognizing preproinsulin), or alloreactive (HLA-A2: positive control) CTL clones at 3 different effector/target (E/T) ratios. DRiP and PPI CTLs are autoreactive T-cell clones derived from T1D patients. After 4h-coculture, the release of ⁵¹Cr from islet-cells was measured with γ-counter to calculate the specific cell-lysis. (See Extended Data Fig. 10 to validate the specific killing capability of CTL clones). (b) Schema of validation for CTLs. To evaluate the specificity and function of CTL clones, JY cells, Epstein-Barr virus (EBV)-immortalised B lymphoblastoid cell line (HLA class-I A2⁺), were used as target cells. As positive control groups, JY cells were loaded with either INS-DRiP_1–9 (DRiP) or preproinsulin (PPI) peptide epitope and labelled with chromium (⁵¹Cr). Then they were co-cultured with effector cells (CTLs), which is either CMV-directed (CMV), DRiP-directed (DRiP), PPI-directed (PPI), or alloreactive (HLA-A2) CTL clone. (b) CTL killing assay against JY cells. JY cells were killed by the alloreactive HLA-A2 CTLs, but not by CMV-directed CTL, β-cell-specific CTLs anti-PPI or anti-DRiP CTLs. When target cells were loaded with the PPI or DRiP peptide epitope, those JY cells were killed by the respective CTLs, confirming that the specific CTLs function and kill when they recognize their epitope. Each dot represents independent measurement from 3 independent experiments. ****p < 0.0001, one-way ANOVA with Holm-Sidak’s multiple comparisons test.

Figure 1.. Glucagon-expressing α-cells efficiently engage insulin production.

(a) Generation and analysis of pseudoislets composed of labeled human islet endocrine cells. Highly pure cell preparations were labeled with GFP alone or in combination with reprogramming factors (“TFs”) via adenoviral transduction (see Extended Data Fig. 1 and Supplementary Table 2). Labeled islet cells were reaggregated into pseudoislets and analyzed in vitro and in vivo after transplantation into immunodeficient mice to examine their functionality, molecular profiling, and immunogenicity. (b) Live-imaging during reaggregation of GFP-transduced α-cells. (c) Insulin protein expression in α-cells 7 days after transduction and aggregation. PM: Pdx1+MafA, MN6: MafA+Nkx6.1, PN6: Pdx1+Nkx6.1, 3TFs: Pdx1+MafA+Nkx6.1. ****p<0.0001 vs. αGFP control; ††††p<0.0001, †p=0.047 vs. αPM, one-way ANOVA with Tukey’s multiple comparisons test. n=10 (αGFP, αPM), n=5 (αPdx1, α3TFs), n=3 (αMafA, αNkx6.1, αMN6 and αPN6). All samples are replications from different donors. (d) Immunofluorescence of 2 pseudoislets made of α-cells transduced with GFP (αGFP) or GFP and Pdx1+MafA (αPM) (see Extended Data Fig. 3a). Most reprogrammed insulin-producing α-cells express glucagon 1 week after transduction. (e) Reaggregation significantly increases insulin expression in αPM-cells. Only 4% of αPM cells contain insulin if kept as single α-cells. Upon reaggregation, either alone or in combination with HUVECs & MSCs (“HM”), 35% become insulin+. ****p<0.0001, ns (not significant): p=0.38 (αGFP singlets vs αPM singlets), p=0.40 (αPM vs αPM+HM), one-way ANOVA with Tukey’s multiple comparisons test. n=3 donor samples in each condition. (f) GSIS. C-peptide release from αPM cells is enhanced in the presence of HM cells; the dotted line indicates C-peptide background level in medium-only αGFP controls. ****p<0.0001, **p=0.0068, *p=0.028, two-way RM ANOVA with Holm-Sidak’s multiple comparisons test. n=5 donor samples in each condition. Scale bars: 25 μm. Images are representative from 5 (b,f) or 38 (d) independent experiments. All data are mean ± s.e.m.

Figure 2.. Insulin-producing human α-cells reverse murine diabetes.

(a) Experimental design. NSG or NSG-RIP-DTR mice were made diabetic with streptozotocin (STZ) or diphtheria toxin (DT); αPM pseudoislets were transplanted under the renal capsule, either from single (Exp.#2; Extended Data Fig. 6) or multiple donors (Exps.#3 & #4; b,c). Grafts were removed after 4 weeks or up to 24 weeks in the longest experiment (“Nx” in b,c). (b) Random-fed glycemia in Exp.#3. Glycemia was decreased to normal (dotted lines) after engraftment of 6,000 pseudoislets generated from 6 donors (red line), in 3 transplantations (“Tx”: 2,150+3,100+750), like controls receiving human islets (4,000 IEQ, blue lines). Untransplanted diabetics remained hyperglycemic (black lines). n=5 (DT+noGraft), n=3 from 3 donors (DT+islets), n=1 from 6 donors (DT+αPM), n=5 mice (noDT+noGraft). (c) Random-fed glycemia in Exp.#4. 4,000 pseudoislets made from 3 donors were transplanted, leading to complete rescue. n=3 (DT+noGraft), n=2 from 2 donors (DT+islets), n=1 from 3 donors (DT+αPM), n=4 mice (noDT+noGraft). (d,e) Human C-peptide blood levels before and after glucose injection. Data from non-diabetic (d; Exps.#3,#4) and T2D donors (e; Exp.#5) are shown. n.d., undetectable. ****p<0.0001, two-way RM ANOVA with Holm-Sidak’s multiple-comparison tests. In d, n=8 (DT+noGraft), n=5 from 5 donors (DT+islets), n=1 from 6 donors (DT+αPM, Exp.#3), n=1 from 3 donors (DT+αPM, Exp.#4), n=9 mice (noDT+noGraft). In e, n=4 (STZ+noGraft), n=1 from 1 donor (STZ+islets), n=1 from 2 donors (STZ+αPM, Exp.#5), n=2 mice (noDT+noGraft). (f) Immunofluorescence on transplanted pseudoislets. Monohormonal insulin-expressing (red) α-cells (GFP⁺, green) are abundant. (g,h) Reprogramming efficiency (g) and percentage of monohormonal insulin-producing cells (h) in αPM pseudoislets before and after transplantation. *p=0.029, two-tailed Mann-Whitney test. n=4 donors. (i) Immunofluorescence on pseudoislets engrafted for 6-months. n=1 (6 months; i, Extended Data Fig. 6r) and n=1 (3 months; Extended Data Fig. 6q). Scale bars: 25 (f), 50 μm (i). Data are mean ± s.e.m. (d,e,g,h).

Figure 3.. Transcriptomic and proteomic analyses of insulin-producing human α-cells.

Principal component (a) and Pearson correlation (b) analyses of RNA-Seq samples, showing a gene signature shift from α- to β-cells. Each dot in (a) represents one donor. (c) Volcano plot representing DEGs between sorted α-cells and β-cells (FDR < 0.05, FC > 2). 887 α-cell-enriched genes and 587 β-cell-enriched genes were identified (Supplementary Table 10). (d-f) Volcano plots showing DEGs in: (d) αGFP pseudoislets relative to sorted α-cells, characterizing the cell aggregation effect, (e) αPM relative to αGFP pseudoislets, reflecting the effect of PM overexpression, and (f) αPM pseudoislets relative to sorted α-cells, reflecting the combined effect of reaggregation and PM overexpression. Upregulated DEGs in each condition (colored squares in d-f) were overlapped with β-cell-enriched genes from c, as a measure of β-cell trait acquisition. The Venn diagrams show that 128 “β-like genes” were upregulated in α-cells upon aggregation (d), 115 genes upon PM activation (e), and 268 “β-like genes” by the combined effect of aggregation and PM activation (f). Subsequent proteomic analyses validated many of the identified genes, which were also expressed at the protein level: 16 out of the 128 proteins in d, 18 of 115 in e, and 31 of 268 in f. (g) Gene-set enrichment analysis (GSEA) of αPM pseudoislets compared to sorted α-cells revealed an enhanced expression of β-like gene-sets and genes involved in regulating insulin secretion. (h) Quantitative proteomic analysis of “acquired β-cell signature proteins” in d-f. These proteins were more abundant in αPM than in αGFP pseudoislets. Proteins known to be important for β-cell function are highlighted in red. n=5 (sorted α), n=5 (sorted β), n=7 (αGFP), n=7 (αPM), n=6 (βGFP), n=5 (grafted αPM) in a-g. n=3 from 4 donors (αPM), n=2 from 4 donors (αGFP) in h. Data are mean ± SD.

Figure 4.. scRNA-Seq analysis of insulin-producing human α-cells.

(a) t-SNE visualization of single-cell transcriptomes of pseudoislets; 47 αGFP, 434 αPM and 51 βGFP cells form 3 distinct clusters. (b) In silico pseudotime ordering of αPM cells (n=434) shows 3 different states along a main pseudotemporal trajectory: “early” (135 cells), “mid” (213 cells) and “late” (86 cells). Each dot represents one cell. Most αPM cells allocate along the main path from “early” to “late” based on reprogramming progression. (c) Clustering of differentially modulated genes by pseudotime progression of αPM cells shows distinct kinetics of gene responses to cell conversion: increase in expression of β-cell genes (‘pro-conversion’ genes) and increase in expression of α-cell genes (‘resistant’ genes). Only α-/β-related genes are shown. (d) Dot plot showing gene signature shifts among different pseudotime stages. (e) Gene expression kinetics along pseudotime progression of representative genes. Green: α-cell-related genes, red: β-cell-related genes (c-e). (f) Superimposition of pseudotime categories on t-SNE map reveals an early-to-late transition of αPM cells.

Figure 5:. Immunogenicity tests on insulin-producing human α-cells.

(a) Cytotoxic T lymphocyte (CTL) killing assay against β-cells from βGFP pseudoislets. Anti-PPI (preproinsulin) CTLs lyse β-cells, but not anti-DRiP CTLs. (c) Assay against α-cells (αGFP pseudoislets). αGFP cells are lysed by the alloreactive HLA-A2 CTLs, but not by β-cell-specific anti-PPI or anti-DRiP CTLs. When loaded with PPI or DRiP peptide epitopes, αGFP cells are lysed by the corresponding CTLs. (d) Assay against α-cells from αPM pseudoislets. A fraction of αPM cells are lysed by PPI-directed CTLs (due to their insulin production), but not by anti-DRiP CTLs recognizing stressed β-cells. αPM cells are lysed by anti-DRiP CTLs if pulsed with exogenous DRiP peptide. CMV: CMV-directed CTLs as negative controls, HLA-A2: alloreactive CTLs as positive controls. E/T: effector/target. Each dot represents an independent measurement from 3 independent experiments using in total 7 different donor samples. **p=0.0022 (a); **p=0.0078 (b); ***p=0.0007, **p=0.0076, †† p=0.0069(c); ****p<0.0001 (b,c); one-way ANOVA with Holm-Sidak’s multiple comparisons test. All data are mean ± s.e.m. Sample information in Supplementary Table 22.