ZnT8 Loss-of-Function Accelerates Functional Maturation of hESC-Derived β Cells and Resists Metabolic Stress Induced Cell Death in Diabetes


 Human embryonic stem cells (hESCs) derived β cells (SC-β cells) hold great promise for diabetes treatment, yet how to achieve functional maturation of these SC-β cells and protect them against metabolic stresses such as glucotoxicity and lipotoxicity remain elusive. By single cell RNA-seq pseudotime analysis, we revealed that ZnT8 is involved in SC-β cells functional maturation process and its loss of function (LOF) accelerates functional maturation of SC-β cells. As a result, ZnT8 LOF improves glucose-stimulated insulin secretion (GSIS) and enhances proinsulin to insulin conversion efficiency in SC-β cells, both in vitro and in vivo, by releasing the negative feedback of zinc inhibition on insulin secretion. Furthermore, SC-β cells with ZnT8 LOF are resistant to metabolic stresses induced cell death, as lipotoxicity or glucotoxicity, displaying higher survival. Most importantly, transplantation of SLC30A8-/- SC-β cells into diabetic mice significantly improves glycemia restoration and SC-β cell survival with long-term stability. Therefore, our study offers an advanced cell replacement therapy for diabetes with both improved SC-β cell survival and function against metabolic stress.


Introduction
Pancreatic β cell loss and dysfunction underlie diabetes mellitus 1,2 . Although whole pancreas or islet transplantation offers an alternative to insulin treatment in diabetic patients, they are severely limited by the scarcity of donor organs. Stem cell-based cell therapy from human pluripotent stem cells (hPSCs) to functional pancreatic β cells (SC-β cells) turns to be a promising approach for diabetes treatment and drug development [3][4][5][6][7] . However, limited functional maturation of these SC-β cells hampers this strategy as a cell replacement therapy for diabetes, especially the lack of sensitivity to glucose stimulation 8,9 10 .
Moreover, the fragility of the SC-β cells remains as another major challenge. Even though the encapsulation of SC-β cells could prevent risks of immune destruction and oncogenic transformation, this physical barrier is still susceptible to glucotoxicity and lipotoxiciy and oxidative stress that directly annihilate β cells' function and viability. Therefore, it is also essential to protect transplanted SC-β cells' function and survival against the harmful environment recurrent in the diabetic patients to achieve long-term functionality and stability.
Human genome-wide association studied (GWAS) have identified various genetic predispositions related to both β cell malfunction and diabetes susceptibility, suggesting human β cells could be born with "fragility" and later be devastated under metabolic stresses 11,12 . Interestingly, recent studies have reported that loss-of-function (LOF) of SLC30A8 encoding ZnT8 (zinc transporter, enriched in islet β cell ) protects against human diabetes by enhancing insulin secretion capacity 1,2 , implying that some human pancreatic β cells could alternatively be born with "robustness" 13 . Based on these previous studies, we hypothesized that ZnT8 LOF could potentially promote functionality of the SC-β cells.
Furthermore, either lipotoxicity or glucotoxicity as metabolic stress is detrimental to human pancreatic β cell survival [14][15][16] , which is also recurrent in diabetic patient as a key challenge to cell replacement therapy jeopardizing the viability of transplanted SC-β cells. Therefore, it would be highly valuable to examine if human embryonic stem cells (hESCs)-derived β cells harboring ZnT8 LOF could potentially offer an advanced stem cell-based cell replacement therapy for diabetes with improved functional maturation and survival against metabolic stresses.
To test this, we applied CRISPR/Cas9 technology and introduced ZnT8 LOF mutation into genome of the hESCs which were step-wisely differentiated into functional pancreatic β cells. The generated pancreatic β cells with ZnT8 LOF obtain accelerated functional maturation, showing significantly improved insulin secretion capacity and proinsulin to insulin conversion efficiency both in vitro and in vivo. Moreover, ZnT8 LOF alleviated the glucose toxicity and palmitate-induced lipotoxicity with improved cell viability. Most importantly, the generated human pancreatic cells with ZnT8 LOF dramatically improved restoration of normoglycemia in diabetic mice with long-term stability, offering an advanced stem cell-based cell replacement therapy.

Single cell RNA-seq pseudotime analysis reveals accelerated functional maturation of SC-β cells by ZnT8 LOF
To track the differentiation process of SC-β cells, we first generated a Mel1 NKX6.1:linker2a:mCherry human embryonic reporter stem cell line (NKX6.1 mCherry/mCherry -INS GFP/W hESCs) by using CRISPR/Cas9 17 knock-in strategy in Mel1 INS GFP/W hESCs 18 . Through a stepwise method 8,9 of pancreatic β cell differentiation, the expression of mCherry fluorescence driven by the endogenous NKX6.1 promoter was observed throughout the pancreatic progenitor cell stage (S4) to mature SC-β cell stage (S7); GFP expression under INS promoter was initially weak in endocrine progenitor cell stage (S5) and then highly expressed in mature SC-β cells (S7) which formed cell cluster structures (Fig. 1a). Differentiation efficiency was measured by specifically expressed genes of each stage (Fig.   1a).
In order to explore the effects of ZnT8 LOF on the differentiation and function of SC-β cells,  Table 1). This CRISPR/Cas9-based genome editing was designed to create a premature stop codon in the reporter line (Extended Data Fig. 1a). LOF of this nonsense SLC30A8 variant was validated by qRT-PCR, which showed that SLC30A8 mRNA expression in mature SC-β cells harboring this nonsense mutation (hereinafter referred to as SLC30A8 -/-SC-β cells) was almost abolished compared with its WT counterpart (hereinafter referred to as WT SC-β cells) (Extended Data  Fig. 1c, d). Compared to the WT counterpart, ZnT8 LOF did not affect hESCs' morphology or self-renew capacity. The gene expression of pluripotent stem cell markers and capacity of teratoma formation were comparable to those of WT hESCs (Extended Data Fig. 1e, f). Taken together, these data confirmed that the nonsense mutation of c.479_524 del, p. Y169fx is indeed a LOF variant and the SLC30A8 -/-hESCs were suitable for interrogating the effects of ZnT8 LOF on SC-β cell differentiation.
We examined how ZnT8 LOF affected cellular differentiation from definitive endoderm at each stage to mature SC-β cell. ZnT8 LOF also had no obvious effect on SC-β cell differentiation efficiency at the definitive endoderm stage (Extended Data Fig. 2a, b), pancreatic progenitor stage (Extended Data Fig. 2c, d) and immature SC-β cell stage (Extended Data Fig. 2e, f). This data, combined with the observation that SLC30A8 is specifically expressed during maturation process (from S6 to S7) (Fig.   1b), confirmed that ZnT8 LOF does not affect the early cell fate specification of SC-β cells.
Next, we applied Monocle v2.0 to reconstruct the maturation process to further illustrate ZnT8 LOF in increased maturation process in more detail and ordered all cells from 4 major clusters along the pseudotime trajectory. After annotating the cell identity and summarizing the distribution of S6WT, S6KO, S7WT and S7KO cells along the pseudotime trajectory, we found that both S6KO and S7KO cells were overrepresented towards the maturation trajectory comparing to corresponding S6WT and S7WT cells, suggesting the maturation process was accelerated in SLC30A8 -/-SC-β cells (Fig. 1g, h).
We then plotted several genes involved for SC-β cell maturation in both WT and SLC30A8 -/cells.
Interestingly, genes related to β cell function such as VAMP2, PCSK2, and PCSK1 had significant increased levels and proportions in either S6KO or S7KO, or both. Importantly, ITGA1 (CD49a), a recently identified surface marker for functional and mature β cells derived from hPSC 19 , was also significantly enriched in SLC30A8 -/-SC-β cells at both S6 and S7. By contrast, CPA1 and NEUROG3 enriched in pancreatic progenitors, had significantly decreased levels in SLC30A8 -/-SC-β cells, further supporting accelerated maturation in SLC30A8 -/-SC-β cells (Fig. 1i).

Accelerated functional maturation by ZnT8 LOF in SC-β cells is manifested by improved sensitivity to glucose stimulation and proinsulin to insulin conversion
Pancreatic β cells gain the capacity of glucose-stimulated insulin secretion (GSIS) as they mature 20 , and GSIS is one of the basic features to evaluate β cell function upon maturation. Single cell RNA-seq data showed that several insulin secretion pathway-related genes were upregulated in SLC30A8 -/-SCβ cells, such as NKX6.1, PDX1, CCND2, VAMP2, SYT13, RAB3B, CADPS, INS (Extended Data Fig.   5a). The gene set enrichment analysis (GSEA) also confirmed that the insulin secretion pathway was dramatically more enriched in SLC30A8 -/-SC-β cells at both the S6 and S7 than that in WT SC-β cells (Fig. 2a). Moreover, pancreas development, glucose homeostasis, carbohydrate homeostasis and peptide hormone secretion are among the top 10 enriched biological processes from gene ontology (GO) enrichment analysis at S7KO (relative to S7WT) (Fig. 2b). The qRT-PCR assays also revealed increased relative insulin expression level in SLC30A8 -/-SC-β cells (Fig. 2c). Our single cell RNA-seq revealed that ZnT8 LOF contributes to upregulation of insulin secretion-related genes and pathways.
In order to verify whether SC-β cells with ZnT8 LOF gain more sensitivity of insulin secretion, we analyzed the GSIS of SC-β cells at S7 by scoring the ratio of insulin release in high glucose (20 mM) to that in low glucose (2 mM) in 2 consecutive cycles. We observed that GSIS was significantly enhanced in SLC30A8 -/-SC-β cells upon high glucose stimulation in both cycle1 (WT, 1.30 ± 0.10;  Fig. 5b). Moreover, the PCSK1 encoding prohormone convertase PC1/3 that activates the proinsulin to insulin conversion by proteolysis 21-23 was increasingly expressed from our single cell transcriptomic profiling at the S6, indicating that the conversion efficiency might be also enhanced in SLC30A8 -/mature SC-β cell. To validate this, we checked the proinsulin conversion efficiency in the SLC30A8 -/-SC-β cells by comparing intracellular total proinsulin to Cpeptide ratio, and found that the ratio was significantly decreased in the SLC30A -/-SC-β cells (Fig. 2h, p = 0.0016), suggesting an increased conversion efficiency compared to WT β cells. To further confirm this in vivo, we transplanted both WT and SLC30A8 -/-SC-β cells into the kidneys of immune compromised mice (SCID-Beige), and the serum levels of human proinsulin and C-peptide were measured (Fig. 2i). Notably, human proinsulin/C-peptide ration was significantly decreased in mice with SLC30A8 -/-SC-β cell transplants (8.82% ± 1.59%) compared to that with WT SC-β cells (13.94% ± 1.14%) further confirming that proinsulin conversion efficiency was also enhanced in SLC30A8 -/-SC-β cells (Fig. 2j, p = 0.0397) in vivo. Moreover, the GSIS (human insulin) was also significantly increased in mice transplanted with SLC30A8 -/-SC-β cells compared to that with WT SC-β cells (Fig.   2k, p = 0.0477). Taken together, both in vitro and in vivo results further corroborate that the SC-β cells with ZnT8 LOF show advanced functional maturation.

ZnT8 LOF promotes sensitivity to glucose stimulation of SC-β cells by releasing zinc inhibition
To decipher how ZnT8 LOF improves SC-β cell function, we focused on the its zinc-transportation regulation. We used a sensitive zinc fluorescent indicator "Zinquin" to monitor the zinc level inside the insulin secretory granules 24 and observed that the typical "dotted" pattern of granular zinc found in WT SC-β cells was largely missing in the SLC30A8 -/-SC-β cells (Fig. 3a). Next, we examined whether the lack of granular zinc contributes to enhancement of insulin secretion.
It has been demonstrated that ZnT8 LOF affects zinc-dependent insulin hexamer formation, which yields the inactive insulin form in the insulin secretory granules [25][26][27][28][29][30] . Since the dense-core structure is the characteristic of the insulin secretory granules that contain insulin hexamers facilitated with granular zinc 31-33 , we scored the dense-core granules in both WT and SLC30A8 -/-SC-β cells with electron microscopy (EM). We found the number of dense-core structures of insulin-zinc crystals were more in WT than those in SLC30A8 -/-SC-β cells at S7 (Fig. 3b, c, p = 0.0139). Similarly, the immature insulin granules in SLC30A8 -/-SC-β cell looked paler than those in WT at S6 (Extended Data Fig. 5c, d). These results suggested that the insulin hexamers and dense-core structure formation could be disrupted in the SLC30A8 -/-SC-β cells due to decreased zinc level. Furthermore, we examined whether zinc-dependent insulin hexamer formation interruption could enhance insulin secretion. To address this, we constructed another hESC line "B10" that carried a human insulin variant with a mutation of "His-B10-Asp" (Extended Data Fig. 6a). This mutation abolished insulin hexamer formation due to lack of the conserved zinc-coordinating B10 histidine [34][35][36][37] . We found that although SC-β cell differentiation efficiency and the ability of zinc transport were not affected by the B10 mutation (Extended Data Fig. 6b-d), insulin secretion was severely impaired rather than improved with the absence of hexamer formation (Extended Data Fig. 6e), ruling out the possibility that disruption of insulin hexamer formation contribute to improved insulin secretion in the SLC30A8 -/-SC-β cells.
It is known that extracellular zinc can be endocytosed by the Zip family members to the cytoplasm of pancreatic β cells [38][39][40] . Cytosolic zinc is further transported by ZnT8 to the insulin secretory granules and eventually co-secreted with insulin into the extracellular spaces. A high level of extracellular zinc has an inhibitory effect on GSIS in rodent islets [41][42][43] , raising another possibility that co-secreted zinc might be involved in the negative feedback of insulin secretion in an autocrine manner. To test this, we added zinc to the zinc-free medium containing SLC30A8 -/mature SC-β cells to monitor insulin release upon glucose stimulation (Fig. 3d). Ectopically added zinc inhibited insulin secretion in SLC30A8 -/-SC-β cells in a dose-dependent manner as measured by the high glucose/low glucose insulin release ratio with added zinc (control, 4.19; 125 μM, 2.28; 250 μM, 0.78; n ≥ 60 SC-β cell clusters for each dose, Fig. 3e). In addition, the dynamic response of C-peptide release from SLC30A8 -/-SC-β cells to 20 mM glucose was inhibited by added 250 μM zinc in the perifusion assay (Fig. 3f).
To determine how the granular zinc level correlated with insulin release, we analyzed the insulin  Fig. 6f, g). All these results supported our conclusion that ZnT8 LOF reduces granular zinc storage and consequently releases the negative feedback inhibition on insulin secretion and upregulates insulin expression in SC-β cells.

Less zinc transportation activity contributes to more functional maturation of SC-β cells
We demonstrated that ZnT8 LOF improves functional maturation of SC-β cells by abolishing zinc transportation, however, there was still one discrepancy of ZnT8 zinc transportation activity on diabetes occurrence. The WT SLC30A8 R325 variant 44,45 has long been considered to possess a zinc transportation activity inferior to that of the W325 variant (in which arginine is replaced by the tryptophan at position 325 of the ZnT8) 27,46 and is associated with higher risk of diabetes than the W325 variant, which contradicts the protective role of SLC30A8 LOF against diabetes 1 . Nevertheless, this conclusion was based on ectopic overexpression of the human R325 and W325 variants by plasmid transfection or adenovirus infection in either a mouse or rat tumor cell line 27,46 . In order to reconcile this unsolved discrepancy, we introduced the W325 mutation into the Mel1 NKX6.1 mCherry/mCherry -INS GFP/W hESCs by CRISPR/Cas9 based genome editing and differentiated the R325 (WT) and the W325 variant hESCs into β cells (Extended Data Fig. 7a). Although R325 did not significantly differ from W325 in SC-β cell differentiation efficiency as measured by NKX6.1-mCherry and INS-GFP (Extended Data Fig. 7b), we found that when Zinquin was used to assess the granular zinc level, the R325 SC-β cells displayed significantly stronger zinc intensity than the W325 SC-β cells at S7 stage (Extended Data Fig. 7c, d). According to the EM images, the insulin granules in W325 SC-β cells were paler than that in R325 (Extended Data Fig. 7e)This result clearly demonstrated that the R325 variant actually possesses zinc transportation capacity superior to that of the W325 variant, which is in line with recent biochemistry study 47 . Finally, the perifusion assays confirmed that the W325 SC-β cells are more sensitive to the glucose stimulation than the R325 SC-β cells. The AUC analysis confirmed that insulin release upon high glucose stimulation is significantly enhanced in the W325 SC-β cells (Extended Data Fig. 7f). These results reconciled the long-standing discrepancy of previous human genetic studies on the contradicting diabetes risks of these two variants of SLC30A8 1, 44 , and further confirmed that granular zinc less zinc transportation activity contributes to more functional maturation of SC-β cells.

ZnT8 LOF SC-β cells are resistant to lipotoxicity and glucotoxicity induced cell death
In diabetic patients, lipotoxicity and glucotoxicity are two crucial risk factors associated with β cell  (Fig. 5c, d). The body weight gain and blood glucose level were measured every 1 week and 10 days respectively after transplantation. Furthermore, mice transplanted with SLC30A8 -/-SC-β cells showed significantly increased body weight gain and decreased blood glucose level compared to mice transplanted with WT SC-β cells, suggesting improved glycemia restoring capacity of the SLC30A8 -/-SC-β cells. (Supplementary Fig. 5e, f). After transplantation for 90 days, blood glucose level in three of five mice transplanted SLC30A8 -/-SC-β cells were restored to normal (a level below 200 mg/dl) whereas only one of five mice transplanted WT SC-β cells showed restored normal glucose level (Fig. 5g). For the Intra-peritoneal glucose tolerance test (i.p.GTT), the mice were fasted for 16h and consequently challenged with 3g/kg glucose by intraperitoneal injection.
The blood glucose change during the 120 min after glucose intraperitoneal injection revealed that the mice carrying SLC30A8 -/-SC-β cells responded to the glucose challenge similarly as the nondiabetic untreated mice (Fig. 5h). Moreover, the kidney grafts of SLC30A8 -/-SC-β cell were analyzed with immunostaining at 6 months after transplantation, which showed an architecture of intermingled insulin and glucagon positive cells and few somatostatin positive cells (yellow arrows on the image) like human islets (Fig. 5i), demonstrating the long-term stability of the SLC30A8 -/-SC-β cells in diabetic mice. Altogether, our results provided the first evidence that SLC30A8 -/-SC-β cells could be applied as advanced stem cell-based cell replacement therapy for diabetes due to both improved function and survival.

Discussion
Recent concerted research advances of generating sustainable SC-β cells offer a promising approach of stem cell-based cell replacement therapy for diabetes treatment. Nonetheless, the transplanted β cells possess limited functional maturation and still face retained or recurrent metabolic stress in the diabetic patients, thus the remaining challenge is to achieve long-term functionality and survival of these SC-β cells. A recent elegant study demonstrated that functional maturation can be achieved by FACS sorting and reaggregating the immature SC-β cells 10 . Nevertheless, the relatively complicated procedure still needs to be optimized to makes it practical for large-scale production.
Given that ZnT8 LOF in pancreatic β cells results in increased insulin secretion, it is highly valuable to explore if and how the stem cell-derived islet β cells harboring ZnT8 LOF could be of advanced therapeutic importance against diabetes. Although it was recently reported that human induced pluripotent stem cell (iPSC) with ZnT8 haploinsufficiency displayed reduced SLC30A8 expression 13 , further investigation into the therapeutic application of ZnT8 LOF β cells derived from hPSCs was still missing, especially possible improved function or viability of these β cells harboring ZnT8 LOF.
Therefore, our study took this initiative and engineered hESCs harboring SLC30A8 LOF mutation followed by step-wisely differentiation into functional SC-β cells. Importantly, we demonstrated Furthermore, it has been a long-standing discrepancy of the diabetes risk variant ZnT8 Arg325 and Trp325. ZnT8 Arg325 variant was considered to possess reduced zinc transporting capacity compared to the protective Trp325 variant, which contradicted the protective role of ZnT8 LOF 1,2,44 . Instead, we demonstrated that the Arg325 has hyperactive zinc transporting activity concluded from the functional assays on β cells derived from hESCs harboring Arg325 versus W325 variants. We therefore reconciled the paradoxical diabetes risk of ZnT8 with different mutation variants and showed that ZnT8 mediated granular zinc transportation has a negative impact on β cell functional maturation.
Taken together, in light of the discovery that inhibition of ZnT8 zinc transportation activity improves insulin release, our study provided a conceptual advance suggesting that ZnT8 is a potential druggable target for treating human diabetes. Further studies will be necessary to perform compound screening to identify ZnT8 inhibitors, which could be utilized as innovative therapeutics for diabetes prevention and treatment.

In vitro differentiation of pancreatic β cells derived from hESCs.
hESCs were differentiated into pancreatic β-cells using the protocol adapted from published study 9,49 .
In brief, the hESCs were cultured on 1:3 Matrigel in hESCs medium until 90% confluency to start differentiation. The differentiation process includes planar culture through stages 1 to 4 to generate NKX6.1 + progenitors at high efficiency and suspension culture from stage 5 to 7 to generate maturing β-cells (Table S3 and S4).
A 10-fold dilution series of human genomic DNA standards ranging from 100 ng/μL to 0.1 ng/μL was used to evaluate the efficiency of the PCR and calculate the copy number of each gene relative to the house keeping gene TATA box binding protein (TBP) 50 . The table S5 shows q-PCR primers.

Immunofluorescence staining.
Adherent cultured or 3D cultured cells (S5-S7) were washed once with precooled PBS and fixed with

Insulin secretion assay.
SC-β cell clusters were starved in Krebs-Ringer buffer 9 supplemented with 2 mM glucose for 2 h in a 5% CO2/ 37°C incubator. For GSIS, the cell clusters would be stimulated alternately by Krebs-Ringer buffer with low-glucose or high glucose for 1 or 2 cycles. Supernatants were collected after 30min of each stimulation and the pallet lysed overnight in acidified alcohol (75% alcohol, 1.5% HCl) at -20 °C for insulin content measure. The same procedure was carried out for treatments with 30 mM KCl or 200 μM tolbutamide. The insulin (secreted or content) was measured by human insulin ELISA kit (ALPCO, #80-INSHU-E01.1).

Proinsulin to C-peptide content analysis.
To measure the proinsulin or C-peptide content, the SC-β cell clusters were cleared with PBS by brief centrifugation and lysed overnight in acidified alcohol (75% alcohol, 1.5% HCl) at -20 °C after 2h starving in the low glucose Krebs-Ringer buffer. The proinsulin and C-peptide were measured by human proinsulin ELISA kit (Mercodia, #1118-1-10) and human ultrasensitive C-peptide ELISA kit (Mercodia, #0111-1-10).

Dynamic perifusion assay for insulin secretion.
60-65 healthy S7-cell clusters (~200 μm-diameter) were handpicked, washed with Krebs-Ringer buffer twice and suspended in 2 mM glucose Krebs-Ringer buffer. Then clusters were loaded onto each chamber of an automated Biorep Perifusion System between two layers of Bio-Gel P-4 polyacrylamide beads. Under temperature-controlled conditions, clusters were perfused at a flow rate of 100 μl/min and samples were collected with 1 min collection points. Prior to sample collection, clusters were equilibrated under basal (2 mM glucose) conditions for 60min. For sample collection, clusters were exposed to low glucose (2 mM) Krebs-Ringer buffer for 15 min, followed by high glucose (20 mM) challenge for 30 min, then were exposed to 10 nM Extendin-4 high glucose condition for 15 min, and were finally exposed to low glucose Krebs-Ringer buffer for an additional 15 min. For the zinc inhibition assay, 250 μM ZnSO4 was added during the high glucose challenge for 30 min. Finally, the clusters were perfused by low glucose and 30 mM KCl for 10 min. Supernatant samples were frozen at -80 ℃ and hormone levels were measured using human C-peptide ultrasensitive ELISA kit (Mercodia, #10-1141-01). For normalization, the DNA content was determined by KAPA extraction and Quant-iT PicoGreen dsDNA Kit (Invitrogen, #P7589).

Electron microscopy.
SC-β cell clusters were spun down to remove excess medium and washed with PB (PBS without sodium chloride) for 2 times, followed by addition of ice-cold fixative (2.5% glutaraldehyde) over night. The clusters were further processed using electron microscopy sample making protocol by the Electron Microscopy Facility at the School of Life Sciences and Technology, Tongji University. The samples were taken picture (insulin granules) by JEOL JEM-1230 transmission electron microscope.
Single cell RNA sequencing Alignment. A 16 bp cell barcode and 9 bp unique molecular identifier (UMI) were contained in read 2 of our paired-end sequencing data, while all genomic information was contained in read 1. We used the Drop-seq pipeline to perform the alignment (version 1.13). Briefly, we added the barcode and UMI information into tag, trimmed adaptor, 5' primer and 3' polyA from reads. Then the trimmed reads were aligned to the human genome with transcriptome annotation (GRCh38, ensemble 93) using STAR (version 2.6.0a). After that, the tags were merged with the aligned SAM file. The barcode substitution and indel errors were repaired. Finally, a digital gene expression matrix was obtained from barcode-UMI-gene triplets. The remaining cell barcodes were consistent with the ones we used as input.
Scater was used to filter out the low-quality cells. Cells with library sizes greater than 2000 UMI counts, with a number of detected genes greater than 1000 and a percentage of mitochondrial genes less than 25% were used for subsequent analysis. In total, 1543 cells passed the criteria, while 89 cells failed.
Cell clustering. Seurat (version 2.3.4) was used to identify cell clusters. Firstly, First, the UMI counts matrix as input and kept genes expressed in at least 2 cells, leaving 25426 out of 31097 genes for downstream analysis. UMI counts were normalized and scaled by the default parameters. Genes with average expression between 0.0125 and 5 and dispersion more than 0.25 were considered variable genes. Then, used scaled data with variable genes were used as input to perform principal component analysis (PCA). Used jackstraw procedure with 1000 replicates to identify significant PCs with a strong enrichment of differences to separate the cells. PCs 1-12 were selected to run t-SNE and perform the clustering through the FindClusters function.

Differential expression analysis and GO enrichment.
To identify DEGs between two clusters, we used the Seurat function FindMarkers with the following parameters: logfc.threshold = 0.25, test.use = "wilcox", min.pct = 0.1. The cell clusters used for the comparisons were those four clusters corresponding to four given cell types. Genes with adjusted p value no more than 0.05 were selected.

Mouse studies.
SCID-Beige mice were obtained from Charles river and maintained in the animal facility of Tongji University, Shanghai, China. All experiments were performed in accordance with the University of Health Guide for the Care and Use of Laboratory Animals and approved by the Biological Research Ethics Committee of Tongji University. Male mice aged 6-8 weeks were randomly assigned to treatment groups, such that the body weight and blood glucose levels were matched before treatment.
For hyperglycemia restoration, the mice of diabetic disease modeling were made by multiple low-dose Streptozocin (STZ) injection (60 mg/kg/d, for 5 days) and transplanted with the same amount of WT or SLC30A8 -/-SC-β cells under the kidney capsule (3 million cells per animal). Some of nontransplantation diabetic mice were used as control. The body weight gain and random fed blood glucose were measured following transplantation. For SC-β cell with ZnT8 LOF function test, same amount of WT or SLC30A8 -/-SC-β cells were transplanted into healthy SCID-Beige mice under the kidney capsule (2 million cells per animal). Two weeks after transplantation, in vivo GSIS was performed every two weeks by fasting the animals for 16 h and collecting serum from the eye socket before and 30 min following intraperitoneal injection of glucose at 3 g/kg body weight. The proinsulin and C-peptide were also measured in the serum from the eye socket by using human proinsulin ELISA kit and human ultrasensitive C-peptide ELISA kit. Human ultrasensitive insulin ELISA (ALPCO, #80-INSHUU-E01.1) was used to measure the human insulin content in the mouse serum.

Statistical analysis.
Data are derived from at least three independent biological replicates, unless otherwise specifically indicated. Mean ± s.e.m. was used to present quantification data. p-values were calculated by twotailed unpaired student's t-test if not otherwise specifically indicated. For multiple comparisons, pvalues were calculated by One-way repeated measures ANOVA or two-way repeated measures ANOVA. *p < 0.05, **p < 0.01, ***p < 0.001, n.s. not significant.    In perifusion assay, p-values are calculated by two-way repeated measures ANOVA with Bonferroni