FXYD2 marks and regulates maturity of β cells via ion channel-mediated signal transduction

Abstract

Human pancreatic islets regulate organ development and metabolic homeostasis, with dysfunction leading to diabetes. Human pluripotent stem cells (hPSCs) provide a potential alternative source to cadaveric human pancreatic islets for replacement therapy in diabetes. However, human islet-like organoids (HILOs) generated from hPSCs in vitro often exhibit heterogeneous immature phenotypes such as aberrant gene expression and inadequate insulin secretion in response to glucose. Here we show that FXYD Domain Containing Ion Transport Regulator 2 (FXYD2) marks and regulates functional maturation and heterogeneity of generated HILOs, by controlling the β cell transcriptome necessary for glucose-stimulated insulin secretion (GSIS). Despite its presence in mature β cells, FXYD2 is diminished in hPSC-derived β-like cells. Mechanistically, we find that FXYD2 physically interacts with SRC proto-oncogene, non-receptor tyrosine kinase (SRC) protein to regulate FXYD2-SRC-TEAD1 signaling to modulate β cell transcriptome. We demonstrate that FXYD2^High HILOs significantly outperform FXYD2^Low counterparts to improve hyperglycemia in STZ-induced diabetic immune deficient mice. These results suggest that FXYD2 marks and regulates human β cell maturation via channel-sensing signal transduction and that it can be used as a selection marker for functional heterogeneity of stem cell derived human islet organoids.

Introduction

Diabetes is a complex disease that affects more than 30 million people in the United States^1,2,3,4,5. In particular, Type 1 diabetes (T1D) is a significant burden that typically emerges during adolescence and requires life-long administration of insulin and blood glucose monitoring⁶. Pancreatic islet transplantation holds promise as a potential cure for T1D, but donor islets are in short supply^7,8.

Human pluripotent stem cells (hPSCs) such as human embryonic stem cells (hESCs) and human induced pluripotent stem cells (hiPSCs) provide a potential alternative source of pancreatic β cells and hold the promise to cure diabetes^{9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26}. While current approaches generate insulin producing β-like cells from hPSCs in vitro that resemble primary β cells from cadaveric islets, these cells still lack several hallmarks of full maturity, including gene expression and mounting of a proper insulin secretion response to glucose^{26,27,28,29,30,31}. Immature juvenile and neonate β cells undergo postnatal functional maturation, a process that enhances glucose sensitivity and GSIS function^19,32. This developmental pathway for postnatal functional islet maturation typically takes a few years after birth in humans³³. Postnatal maturation is influenced by both lineage-specification^{34,35,36,37,38} and metabolic shifts^{19,23,30,39,40,41,42,43,44}. β cell fate-determinant transcription factor MAFA regulates the expression of genes such as INS and G6PC2, enhancing insulin production and reducing the insulin secretion at low glucose concentrations, which ultimately increases the amplitude of GSIS^34,45. The metabolic shifts from glycolytic to more oxidative states increase the insulin secretion at high glucose concentrations, optimizing GSIS function postnatally and potentially improving the function of hiPSC-derived islets^{19,24,39,41,44,46,47}. Transplantation of hiPSC-derived islets further promote functional maturation, partly through the aforementioned postnatal maturation pathways, leading to improvements in GSIS in vivo^48,49,50.

Recent advances, including single-cell RNA-sequencing (scRNA-seq) and islet physiological analyses have highlighted notable differences in function and transcriptional profiles between primary β cells and hPSC-derived β-like cells (insulin positive cells in hPSC-derived islets), such as irregular glucose sensitivity, mitochondrial function and Na⁺ current^{26,27,28,50,51,52}. Despite this, heterogeneity in the hPSC-derived islets is often overlooked in studies focusing on pooled, rather than individual, islet samples. This heterogeneity exists at both the β cell⁵³ and islet multicell⁵⁴ levels and is predicted to be even more pronounced due to the complex and lengthy in vitro differentiation process. Thus, understanding this heterogeneity at generated islet organoid’s level is essential for addressing the maturity issues currently faced. However, the regulation of this heterogeneity and the potential for selection of high quality hiPSC-derived islets using defined markers remain unclear.

It is essential to understand the cellular diversity within the developing islet, as well as the complex cellular interactions that may influence functional differentiation of islets from hPSCs. With the pivotal role of 3-dimensional (3D) structure and cell communication in organ functionality and cell type specialization, we previously developed a method to cultivate 3D human islet-like organoids (HILOs) from hPSCs²⁶. We have introduced a cutting-edge organoid cultivation system specifically for pancreatic islets, utilizing a gellan gum-based functional polymer within a 3D culture system that is both scalable and cost-effective. Employing our protocol, we can consistently produce HILOs that closely approximate the size of natural human islets, ranging from 100 to 500 μm, within an ~30-day time frame²⁶.

Here, using this model, we have identified FXYD2 as a marker for functional maturation and heterogeneity, serving as an indicator to distinguish “good” and “bad” quality of organoids. Our findings demonstrate that FXYD2 is not only a marker but also a pivotal regulator of the maturity-related gene expression and GSIS function and that leads to a more mature status for human β cells in vitro. Our screening showed that glucocorticoid receptor (GR) signalling, thyroid hormone receptor (TR) signalling and NaCl stimulation synergistically regulates FXYD2 expression. Furthermore, we have established FXYD2-based screening of functionally mature HILOs (FXYD2^High HILOs), creating a screening method to identify optimally functioning hPSC-derived islets.

Results

FXYD2 links mineral absorption pathways to functional maturation

In our prior work, we highlighted differential gene expression between human islet-like organoids (HILOs) and primary human islets using single-cell RNA sequencing (scRNA-seq)²⁶. Further analyses revealed a significant low expression of FXYD2, a gene linked to mineral absorption pathways and a known sodium- and potassium-transporting ATPase gamma chain subunit, particularly in β-like and δ-like cells of HILOs compared to primary human islets (Supplementary Fig. 1a–c). FXYD genes encodes members of a family of small membrane proteins that share a 35-amino acid signature sequence domain, beginning with the sequence PFXYD which contains 7 invariant and 6 highly conserved amino acids. We found that FXYD2 expression is exclusive to β and δ cells within human islets, a pattern not found with other FXYD family members (Supplementary Fig. 1d, e). Notably, we found that rodent β cells predominantly express FXYD3, 5, and 6 as well as FXYD2, while human β cells predominantly express only FXYD2, contrasting the human and mouse islets (Supplementary Fig. 1f). We extended our investigation to public scRNA-seq datasets of human islets and hPSC-derived islets, combining data from several studies (Fig. 1a). This meta-analysis across a total of 200,106 individual cells showed a similar trend of low FXYD2 expression in hPSC-derived β-like cells^25,27,28,52. In the aggregated dataset, insulin-positive cells were distributed across various clusters and at distinct time points, indicating that the maturation of hPSC-derived β-like cells is heterogeneous (Fig. 1b, c). Differentially expressed gene (DEG) expression analyses revealed that although FXYD2 expression is gradually increased during human pancreas development, a similar down-regulation of FXYD2 and related sodium-potassium (NAK) channel genes such as ATP1A1, ATP1A4 and ATP1B1 was observed in the hPSC-derived β-like cells compared to primary human β cells (Fig. 1d). In addition, we observed insufficient induction of FXYD2 genes in both insulin-positive (INS+) and somatostatin-positive (SST+) cells. Poor FXYD2 expression was one of the leading features of hPSC-derived INS+ cells (Fig. 1e, f), suggesting that poor FXYD2 expression is a common feature of current hPSC-derived β cells/islets technology. Despite an increase in FXYD2 expression during in vitro islet organogenesis, its levels in HILOs remained significantly lower than in primary human islets (Fig. 1g, and Supplementary Fig. 1g, h). Localization studies confirmed FXYD2’s presence at the plasma membranes of human islet β cells and the EndoC-βH1, a clonal human β cell line (Fig. 1h, i). These results suggest that FXYD2 is a potential marker for functional maturity in hPSC-derived β-like cells.

Fig. 1: FXYD2 marks a functional maturation in hPSC-derived β cells.

a Schematic of randomized integrated scRNA-seq analyses of 200,106 cells (200 K). The indicated gene sets at each stage of differentiation (a–p) were obtained from publicly available data. b UMAP analyses identified cell type transitions at each stage of the 200,106 cells of scRNA-seq data. c Expression of INS, GCG and SST expression is visualized. d Heatmap analyses of selected differentially expressed genes in each stage of cells. Heatmap is shown by Z-score. e Violin plot showing the expression of FXYD2 (in INS+ cells or SST+ cells), FXYD5 (in GCG+ cells) and FXYD6 (in GCG+ cells). In the combined violin and box plot, the violin shape represents the kernel density estimation, illustrating the distribution of the data, wider sections indicate higher data density. The overlaid box plot shows the median (centre line), first (Q1) and third (Q3) quartiles (box limits), and whiskers extending to the smallest and largest values within 1.5 × the interquartile range (IQR). The datasets from PS = (a), EN = (b)(c), FG = (d), PP = (e), EP = (f)(g), imI = (h)(i), mI = (j)(k)(l), vivo1 = (m), vivo3 = (n), hislets = (p) were analyzed. f Volcano plot showing significantly downregulated genes in hPSC-derived INS+ cells compared to primary human β cells. g qPCR analyses of FXYD2 gene expression in hiPSCs (n = 3), HILOs (n = 10) and primary human islets (hislets) (n = 10). Unpaired two-tailed t-tests. (h, i). Representative immunofluorescence images of insulin (green) and FXYD2 (red) in primary human islets (h) and human β cell line (EndoC-βH1 cells) (i) (scale bars = 100 μm). Error bars indicate SEM.

FXYD2 positively regulates insulin secretion in human β cells

Previous studies with FXYD2 knockout (FXYD2KO) mice indicated improved glucose tolerance due to increased insulin secretion from β cell hyperplasia that proliferates as a compensatory response⁵⁵—a condition often observed in pre-diabetic states associated with immature β cell functionality⁵⁶. However, human β cells, unlike those of rodents, have a limited capacity for self-replication^57,58,59,60. This led us to hypothesize that FXYD2 is integral to the proper functioning of acute GSIS in human β cells. To identify the physiological role of FXYD2 in human β cells, we engineered doxycycline (DOX)-induced FXYD2 overexpression (dFXYD2OE) and siRNA-based knockdown (FXYD2KD) in EndoC-βH1 cells. Human c-peptide is the stable, surrogate marker of human insulin. We found that FXYD2KD decreased c-peptide secretion in response to 20 mM glucose (described as GSIS), 20 mM glucose + 100 nM GLP-1, 20 mM glucose + 100 nM GLP-1 + 10 mM glutamate and 20 mM KCl in EndoC-βH1 cells (Fig. 2a, Supplementary Fig. 2a, b). Furthermore, we established a stable EndoC-βH1 cell line tagged with proinsulin-NanoLuc⁶¹, which co-secretes insulin and Gaussia luciferase at a constant molar ratio to track insulin secretion, finding that pharmacological FXYD2 inhibition (FXYD2i) by digitoxin⁶² suppressed GSIS and KCl-induced Gaussia luciferase secretion (Fig. 2b). Conversely, dFXYD2OE (DOX +) showed enhanced GSIS and KCl depolarization-stimulated c-peptide secretion in EndoC-βH1 cells (Fig. 2c). This effect was recapitulated in primary human islets, in which insulin secretion dependent upon glucose and KCl concentrations was altered by digitoxin (Fig. 2d). These results support the hypothesis that FXYD2 is essential for functional insulin secretion in mature human β cells. Next, we tested the effect of dFXYD2OE (DOX +) during differentiation of HILOs. We found that DOX-treatment from days 20 to 30 in dFXYD2OE (DOX +) HILOs significantly enhanced insulin secretion under both 3 mM and 20 mM glucose stimulation in HILOs (Fig. 2e). A previous study has indicated that sodium currents are twice as pronounced in hPSC-derived β-like cells than in human β cells²⁸. We observed that intracellular sodium concentration remained high in both normal (110 mM) and high (180 mM) NaCl concentrations in dFXYD2OE (DOX −) HILOs. Conversely, it changed dynamically in response to external NaCl concentrations in dFXYD2OE (DOX +) HILOs and EndoC-βH1 cells. (Supplementary Fig. 3a). Furthermore, ATPase activity was dynamically altered by high NaCl concentrations (180 mM) in dFXYD2OE (DOX −) HILOs, but remained stable in dFXYD2OE (DOX +) HILOs and EndoC-βH1 cells (Supplementary Fig. 3b). These observations suggested that FXYD2 modulates differentiation, maturation and ion gradients and positively regulates insulin secretion in human β cells.

Fig. 2: FXYD2 enhances insulin secretion in human β cells.

a Human c-peptide secretion (pM) from EndoC-βH1 cells is shown. FXYD2 knockdown (siFXYD2/FXYD2KD) suppresses insulin secretion in EndoC-βH1 cells. siRNA was transfected 72 h prior to the GSIS assay. n = 3. Unpaired two-tailed t-tests. b Dose-dependent suppression of insulin secretion by digitoxin, measured by Gaussian Proinsulin Nano-Luc system in EndoC-βH1 cells. Digitoxin (0, 0.1, 0.5, 1, 2,5, 10 nM) was treated for 24 h prior to GSIS assay. n = 4. One-way ANOVA. c Human c-peptide secretion (pM) from EndoC-βH1 cells is shown. DOX-induced FXYD2 induction enhances insulin secretion in EndoC-βH1 cells. DOX (+) was treated 72 h prior to GSIS assay. n = 3. Unpaired two-tailed t-tests. d Human c-peptide secretion (pM) from primary human islets is shown. Pharmacological FXYD2 inhibition for 24 h by digitoxin (0, 0.1, 1, 10, 100 nM) suppresses insulin secretion in primary human islets. n = 3. One-way ANOVA. e Human c-peptide secretion (pM) from dFXYD2OE-HILOs is shown. DOX (+) was treated every other day from day 20 to day 30 differentiated HILOs. n = 6. Unpaired two-tailed t-tests. G3 = Glucose 3 mM, G20 = Glucose 20 mM, KCl20 = KCl 20 mM, GLP-1 = GLP-1 100 μM. Error bars indicate SEM.

Next, to explore the influence of FXYD2 on β cell identity and maturity, and its potential enhancement of insulin secretion function, we conducted bulk RNA sequencing (seq) with dFXYD2OE (DOX +) in EndoC-βH1 cells. We identified 3,461 upregulated genes and 3,592 downregulated genes in dFXYD2OE (DOX +) with respect to parental EndoC-βH1 cells (Fig. 3a). Kyoto Encyclopaedia of Genes and Genomes (KEGG) pathway analyses revealed that FXYD2 broadly regulates the gene expression related to insulin secretion in EndoC-βH1 cells (Fig. 3b, left). Gene Set Enrichment Analysis (GSEA) of the transcriptome of dFXYD2OE (DOX +) EndoC-βH1 cells revealed that positive enrichment of the mineral absorption pathway (Fig. 3b, right). We noted an enrichment among dFXYD2OE (DOX +) augmented transcripts of genes involved in insulin release, such as those in the mineral absorption pathway, vesicle transport, glycolysis, the TCA cycle and other metabolic processes (Fig. 3c). In contrast, dFXYD2OE (DOX +) downregulated genes were enriched for functions associated with cell cycle progression, reflecting transcriptional changes seen in postnatal β cell maturation¹⁹. Conversely, pharmacological FXYD2 inhibition led to diminished gene clusters tied to metabolism and insulin secretion (Fig. 3d, Supplementary Fig. 4a, b). These results suggest that FXYD2 expression positively influences genes related to insulin secretion as well as subsequent GSIS function. We previously observed the expression of transcriptional factors such as Mafa, Esrrg, Pdx-1 and Neurod1 and their related downstream genes is increased during maturation in mouse islets¹⁹. Promoter motif analyses indicated that the cluster of genes induced by dFXYD2OE (DOX +) is regulated by these transcriptional factors including MAFA, ESRRG, PDX-1 and NEUROD1 in EndoC-βH1 cells (Fig. 3e, left). In contrast, dFXYD2OE (DOX +) suppressed the cluster of genes that are regulated by E2F, a known β cell replication factor in EndoC-βH1 cells (Fig. 3e, right). While dFXYD2OE (DOX +) showed limited impact on β cell identity related genes such as INS, the constitutively active FXYD2 overexpressing (cFXYD2OE) EndoC-βH1 cells for over 30 days significantly increased INSgene expression in EndoC-βH1 cells (Supplementary Fig. 4c, d). Consistent with this result, we found that insulin secretion was increased by cFXYD2OE in EndoC-βH1 cells stimulated by 3 mM glucose, 20 mM glucose and 20 mM KCl-stimulated EndoC-βH1 cells (Supplementary Fig. 4e). Those enhancement of β cell lineage gene expression and reduction of cell cycle progression are indicative of β cell maturation. To further validate that FXYD2 regulates human β cell maturation, we used HILOs to test whether dFXYD2OE (DOX +) and FXYD2KD are sufficient to modulate maturity-related gene expression in human islet organogenesis. dFXYD2OE (DOX +) during HILO differentiation from days 27 to 31 led to significantly increased expression of β cell lineage-related genes expression such as INS, MAFA, MAFB, SIX2, UCN3, and metabolic genes related to maturation such as COX6A2, WNT4, ESRRG, GCK, whereas FXYD2KD had the opposite effect (Fig. 3f). To decipher the FXYD2-mediated transcriptional regulation, we conducted the bulk RNA-seq comparing the transcriptome in dFXYD2OE, FXYD2KD and cFXYD2OE EndoC-βH1 cells. Our optimizing experiments revealed that DOX (+)-induced FXYD2 expression monitored by mCherry expression reached to plateau within 120 h and became unstable (Supplementary Fig. 5a). In contrast cFXYD2OE system stably expressed FXYD2 monitored by GFP expression (Supplementary Fig. 5b). We found that DOX (+) slightly increases Casp3/7 activity when it’s treated longer than 192 h in EndoC-βH1 cells (Supplementary Fig. 5c). From these results, we reasoned the comparison of transcriptome at a shorter time point (3 days) with dFXYD2OE system and longer time point (30 days) with cFXYD2OE system, together with FXYD2KD system (6 days) (Supplementary Fig. 5d). We found that 2920 genes were upregulated and 3397 genes were downregulated by cFXYD2OE, and 2204 genes were upregulated and 2024 genes were downregulated by FXYD2KD in EndoC-βH1 cells (Supplementary Fig. 5e). The KEGG analyses showed that cFXYD2OE increased insulin secretion and thyroid hormone signalling pathway and decreased axon guidance pathway (Supplementary Fig. 5f). In contrast FXYD2KD increased axon guidance and decreased metabolic pathways (Supplementary Fig. 5g). We then analysed 125 genes commonly upregulated by dFXYD2OE (DOX +) and cFXYD2OE, while downregulated by FXYD2KD, which were mostly metabolic pathway enriched genes (Supplementary Fig. 5h). Similarly, 163 commonly downregulated by dFXYD2OE (DOX +) and cFXYD2OE, while upregulated by FXYD2KD, which highlighted the axon guidance pathway enriched genes (Supplementary Fig. 5i). Interestingly, both insulin secretion and mineral absorption pathways are commonly upregulated by dFXYD2OE (DOX +) and cFXYD2OE, however, only cFXYD2OE upregulates differentiation/maturation related genes such as INS, MAFA, IAPP and SIX2 (Supplementary Fig. 5j, k).

Fig. 3: FXYD2 overexpression and pharmacological inhibition inversely regulate the transcriptome in human β cells.

a Volcano plot showing differentially expressed genes by DOX-inducible FXYD2 overexpression (dFXYD2OE) in EndoC-βH1 cells. -log₁₀ (p-value) > 1.3. b Kyoto Encyclopaedia of Genes and Genomes (KEGG) pathway analysis of upregulated (Up) or downregulated (Down) regulated genes by dFXYD2OE. GSEA analyses for mineral absorption pathway (NES > 1.35, FDR q-value < 0.17) are shown in the right. c The heatmap show dFXYD2OE (DOX +) upregulated the gene necessary for functional β cells, while downregulated cell cycle related genes. d The heatmap show metabolic pathway related gene expression is upregulated by dFXYD2OE (DOX +), while downregulated by pharmacological FXYD2 inhibition (FXYD2i). e Motif analyses revealed that the promoter region of FXYD2-regulated gene clusters often contains the transcription factors (TFs) binding sites listed above. f dFXYD2OE (DOX +) or dFXYD2KD (DOX +) HILOs at day 27 were treated with DOX (+) for 4 days and then gene expression was analysed by qPCR. n = 3. Heatmaps are shown by Z-score.

Taken together, these results suggested that FXYD2 orchestrates the transcriptional landscape governing β cell differentiation, maturation, and functionality.

FXYD2 enhances the SRC signalling-dependent transcriptional pathway by physical interaction

To investigate how the membrane-localized protein FXYD2 can modulate gene expression, we examined the proteins interacting with FXYD2 in human β cells. To this end, we constructed a lentiviral vector encoding the FXYD2 protein tagged with V5 and TurboID⁶³ (pLV-V5-FXYD2-TurboID-EGFP) and expressed it in EndoC-βH1 cells (Fig. 4a). Prior to immunoprecipitation (IP) of FXYD2 using the V5-tag, we treated cells with 500 nM biotin to label proteins proximal to FXYD2. We observed numerous biotinylated proteins, visualized by streptavidin antibody (SA) in the IP samples (Fig. 4b). NanoLC-MS/MS analyses of potential FXYD2-binding proteins revealed that proteins such as FXYD2 (7-33 kDa), TRIM21 (52 kDa), Erlin1/2 (complex ~75 kDa), SLC5A9 ( ~ 74 kDa), ATP1A1 ( ~ 110 kDa), SRC proto-oncogene, non-receptor tyrosine kinase (SRC)(60 kDa) and SRC substrate cortactin (CTTN) ( ~ 80 kDa) were present in IP samples (Fig. 4c). We confirmed SRC, CTTN and ATP1A1 as interacting partners of FXYD2, which were pulled down in both SA:IP and V5:IP samples (Fig. 4d, Supplementary Fig. 6a). To further explore how FXYD2 enhances kinase activity, we conducted phosphoproteomic analyses using a kinase microarray (KAM-1325 microarrays, Kinexus). With a p-value < 0.05 and fold change (FC) > 1.5 as threshold values, we found 285 upregulated phosphosites in cFXYD2OE EndoC-βH1 cells compared to control EndoC-βH1 cells (Fig. 4e). KEGG pathway analyses revealed that pathways involving serine/threonine and tyrosine kinase, as well as Src homology 2 domains, were altered in cFXYD2OE EndoC-βH1 cells (Fig. 4f). Pathway analyses showed that MAPK and PI3K-AKT signalling, along with Thyroid hormone receptor (TR) signalling were upregulated by cFXYD2OE (Fig. 4f, g). Interestingly, STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) analyses positioned the SRC signalling pathways at the centre of these identified activated pathways (Fig. 4h). Further analyses confirmed that phosphorylation levels of SRC at Y419 and AKT1 at serine 473 were increased by cFXYD2OE (Fig. 4i). SRC, a non-receptor cytoplasmic tyrosine kinase, becomes activated following stimulation by plasma membrane receptors, including receptor tyrosine kinases and integrins, and plays a crucial role in multiple physiological homoeostatic pathways. SRC has a dual function: it maintains dynamic gene expression and metabolism in normal healthy cells and contributes to tumorigenesis and metastasis in cancer cells⁶⁴. It also regulates insulin secretion in β cells in a context-dependent manner^65,66,67. The activity of SRC is regulated by tyrosine phosphorylation at two sites, leading to opposite effects: phosphorylation at tyrosine (Y) 416 and Y419 (Y416/Y419) in the activation loop of the kinase domain increases enzyme activity while phosphorylation at Y527 in the SH2 domain renders the enzyme less active⁶⁸. We discovered that DOX-dependent induction of FXYD2 led to enhanced SRC phosphorylation at Y416 and the downstream AKT pathway while reducing SRC phosphorylation at Y527 (Fig. 4j). These results suggest that FXYD2 directly interacts with and regulates the SRC signalling pathway. FXYD2 is known to interact with Na⁺/K⁺ ATPase (NKA) as a suppressor of NKA activity⁶⁹. We further explored the relationship between FXYD2, ion channel activity and SRC signalling regulation. Therefore, we selected the Ion Channel Compound Library (APEX Ion Channel Compound Library) for screening. We pretreated the proinsulin NanoLuc system⁶¹ integrated EndoC-βH1 cells with 1 μM compounds for 24 h. Our small molecules screening using 199 ion-channel modulator compounds and 33 DMSO controls identified many sodium channel inhibitors (sodium ci) that enhanced insulin secretion in EndoC-βH1 cells (Supplementary Fig. 6b). These sodium cis, including triamterene, oxcarbazepine, propafenone, and ouabain, led to enhanced insulin secretion under low (3 mM) and high (20 mM) glucose stimulation in primary human islets (Supplementary Fig. 6c). Notably, we found that these sodium cis also led to enhanced phosphorylation at Y416 in SRC (Supplementary Fig. 6d). Interestingly, when we knocked down FXYD2 by shRNA or siRNA, propafenone-mediated SRC activation was abrogated (Supplementary Fig. 6e, f). To test whether sodium cis-induced SRC activation is related to the amplification of insulin secretion, we pretreated SRC modulators and/or propafenone in primary human islets. We found pretreatment of propafenone enhanced high-glucose and KCl-induced insulin secretion, but the SRC inhibitor (dasatinib) partially suppressed insulin secretion in human islets (Supplementary Fig. 6g). These results suggested that the FXYD2/NKA/SRC axis regulates physiological human β cell insulin secretion. It is noteworthy that although propafenone treatment partially mimicked cFXYD2OE function, a kinase microarray also showed that unique kinase activities were induced by propafenone and cFXYD2OE, respectively, and a low significance synergistic effect was identified (Supplementary Fig. 7a–e). These results suggested that FXYD2 activates SRC and associated signalling pathways in human β cells.

Fig. 4: FXYD2 regulates signal transduction in functional β cells.

a Schematic of TurboID proximal labelling for FXYD2. EGFP expression (green) is shown in pLV-V5-hFXYD2-TurboID-EGFP expressing EndoC-βH1 cells. Scale bars = 10 μm. b TurboID proximal labelling to identify biotinylated proximal proteins of FXYD2. Immunoprecipitation (IP) by V5-tag and Immunoblot (IB) by streptavidin (SA). Immunoblot result shown is representative of three independent biological replicates. c Proteomics analyses identified unique proteins that co-precipitated with V5-FXYD2. Identified proteins are shown. d IP for SA in control and V5-FXYD2-TurboID-EGFP expressed EndoC-βH1 cells. IB for V5, SRC, CTTN, ATP1A1 and SA. Immunoblot result shown is representative of three independent biological replicates. e Phospho-antibody microarray of control and constitutive FXYD2 overexpressing (cFXYD2OE) EndoC-βH1 cell lysate. f Pathway Ontology (PO) analyses of cFXYD2OE upregulated pathways. Selective kinases (left) and KEGG pathway (right) are shown. g Thyroid hormone signalling-related kinase protein expression. h STRING analyses showing the network of proteins differentially phosphorylated by cFXYD2OE. SRC pathways are highlighted by a pink circle. i Selected heatmaps of differentially regulated phosphosites belonging to SRC pathways in indicated conditions. j Representative IB analysis and quantification for SRC signalling in DOX-induced FXYD2 overexpression (DOX +) in EndoC-βH1 cells (n = 3). DOX (+) was treated for 72 h. Bar graph shows the quantification of IB (n = 3−6). Unpaired two-tailed t-tests. Error bars indicate SEM. Heatmaps are shown by Z-score.

TEAD1 regulates the key β cell gene expression

SRC signaling is known to regulate the downstream transcription factor TEAD1 through both direct and indirect mechanisms. We found that cFXYD2OE showed enhanced TEAD1 expression accompanied by SRC activation in EndoC-βH1 cells (Fig. 5a). To test whether TEAD1 acts as a transcription factor to regulate maturity related genes, we engineered doxycycline inducible TEAD1 overexpression (dTEAD1OE, DOX + ) and TEAD1 knockdown (dTEAD1KD, DOX +) EndoC-βH1 cells. dTEAD1OE (DOX +) increased the gene expression of UCN3 and SIX2. In contrast, TEAD1KD (DOX +) led to decreased MAFA, UCN3 and SIX2 gene expression in EndoC-βH1 cells (Fig. 5b). TEAD1KD reduced both low glucose and high glucose-induced insulin secretion in cFXYD2OE EndoC-βH1 cells (Fig. 5c), suggesting that TEAD1 plays a critical role in enhancing maturity related gene expression and insulin secretion function in human β cells. To further validate the importance of TEAD1 function under FXYD2 signalling in β cells, we have performed additional knockdown experiments using 3 differential siRNA targeting to TEAD1. We confirmed that TEAD1 knockdown decreases TEAD1, MAFA, UCN3 and IAPP expression in EndoC-βH1 cells (Supplementary Fig. 8a). We found that three differential FXYD2 siRNA or TEAD1 siRNA knockdown decreased 20 mM high Glucose + GLP-1 + Glutamate and KCl-stimulated insulin secretion in control and cFXYD2OE EndoC-βH1 cells (Supplementary Fig. 8b, c). Consistent with TEAD1 shRNA experiments (Fig. 5c), three differential FXYD2 siRNA or TEAD1 siRNA knockdown showed tendency to decrease 3 mM low Glucose-stimulated insulin secretion, this was observed significant in cFXYD2OE EndoC-βH1 cells (Supplementary Fig. 8c). Importantly, dTEAD1OE (DOX + ) showed partial rescue of these insulin secretion under the FXYD2 siRNA knockdown (Supplementary Fig. 8d). To determine whether TEAD1 directly regulates these genes, we performed ChIP-seq of TEAD1 occupancy in control and cFXYD2OE EndoC-βH1 cells. 4,753 peaks were identified commonly occupied in control and cFXYD2OE EndoC-βH1 cells (Fig. 5d). TF binding motifs related to β cell function, such as GLIS3, COUP-TFII, THRa, NKX2-2, MAFB, and RORg, were enriched in control EndoC-βH1 cells (Fig. 5e). In addition, cFXYD2OE led to enhanced the TF binding motifs related to β cell maturation such as MafA (Fig. 5e). The commonly regulated pathways with or without cFXYD2OE are mainly related to β cell function such as the cAMP signalling pathway and insulin secretion, which is further upregulated by cFXYD2OE in EndoC-βH1 cells (Fig. 5f). Concomitantly, recruitment of TEAD1 to the promoter region of MAFA, PDX-1, NKX6-1, UCN3, IAPP and NEUROD1 were increased by cFXYD2OE (Fig. 5g). To explore if the increase of TEAD1 expression in cFXYD2OE EndoC-βH1 cells is associated with the activation of SRC signalling, we inhibited SRC signalling in cFXYD2OE EndoC-βH1 cells using dasatinib, a known potent SRC inhibitor (SRCi). We found that SRCi suppressed phosphorylated SRC (pSRC) in control and cFXYD2OE EndoC-βH1 cells (Supplementary Fig. 9a). TEAD1 expression is decreased in both control and cFXYD2OE EndoC-βH1 cells. In contrast, DOX-treated dTEAD1OE as well as dTEAD1KD did not alter pSRC (Y416) nor Non-pSRC activation (Supplementary Fig. 9b), suggesting that SRC signalling regulates TEAD1 expression in EndoC-βH1 cells. Together, these data indicate that FXYD2 enhances the recruitment of TEAD1 to maturity related genes in human β cells.

Fig. 5: TEAD1 regulates the key human β cell gene expression.

a Representative IB analysis of indicated phosphorylated and total proteins in EndoC-βH1 cells under the indicated conditions. b qPCR analyses of indicated gene expression in EndoC-βH1 cells. DOX-inducible TEAD1 overexpression (dTEAD1OE, DOX +) and knockdown (dTEADKD, DOX +) regulate the expression of key β cell genes, including MAFA, UCN3, and IAPP. n = 3. Unpaired two-tailed t-tests. c, Human c-peptide secretion assay (fold change to DOX (–) G3 stimulation) from dTEAD1KD EndoC-βH1 cells. n = 12. Unpaired two-tailed t-tests. d–g TEAD1 ChIP-seq analyses. The distribution of relative gene positions at the peak summit (d), enriched motif analyses (e), KEGG pathway analyses (f) in control and cFXYD2OE EndoC-βH1 cells. g Browser tracks showing TEAD1 ChIP-seq peaks at MAFA, PDX-1, NKX6-1, UCN3, IAPP, NEUROD1 loci in control and cFXYD2OE EndoC-βH1 cells. Error bars indicate SEM.

Taken together, these results suggest that the FXYD2-SRC-TEAD1 pathway may regulate the β cell transcriptomes for maturation and associated insulin secretion.

FXYD2^High insulin producing cells of HILOs marks more mature transcriptome and function

Identifying mechanisms controlling the expression of FXYD2 in human islets may contribute to further improvements in the generation of functional hPSC-derived islets. To investigate the regulatory elements controlling FXYD2 gene expression, we engineered a FXYD2 promoter-driven dual-reporter system (FXYD2-fLuc-mCherry) in EndoC-βH1 cells. This system enabled us to screen a library of over 500 FDA-approved drugs and chemicals. Our screening identified prominent modulators within the nuclear receptor ligand library, notably triiodothyronine (T3, a thyroid hormone receptor ligand) and dexamethasone (Dex, a glucocorticoid receptor ligand) (Fig. 6a). Notably, a synergistic increase in FXYD2 expression was detected when Dex and T3 were combined. Additionally, this upregulation, mediated by nuclear receptors, was further amplified in the presence of retinoic acids (RA, retinoic receptor ligands) (Fig. 6b). Furthermore, we found that FXYD2 expression was stimulated by NaCl in EndoC-βH1 cells and hiPSC derived β-like cells (2D cultured HILOs), respectively (Fig. 6c). Dex and NaCl led to further increase in FXYD2 expression when stimulated with T3 in EndoC-βH1 cells (Fig. 6b, d). We further found that NaCl-stimulation led to increased FXYD2 protein levels and SRC phosphorylation at Y416, along with global epigenetic changes regulation such as an increase in H3K4me3 and H3K27ac in EndoC-βH1 cells (Supplementary Fig. 10a). In addition, NaCl-stimulation influenced the gene expression associated with TR signalling and FXYD2 expression (Supplementary Fig. 10b–e). These results revealed that NaCl stimulation influences gene expression at the chromatin level and synergizes with nuclear receptor signalling in human β cells. We investigated the induction of FXYD2 under the high NaCl-mediated osmotic stress condition act as a defence mechanism of the cells. To test this idea, we compared the osmotic resistance in dFXYD2OE (DOX +) and control EndoC-βH1 cells at different NaCl concentrations. The physiological range of hyperosmotic condition (~250 mM) is often used for osmotic stress-resistance tests^70,71,72. We found that this hyperosmotic condition led to reduced cellular ATP levels in a dose-dependent manner and caused osmotic stress-induced cell death in EndoC-βH1 cells (Fig. 6e). Furthermore, we found that dFXYD2OE (DOX +) conferred protection, whereas FXYD2KD by siRNA led to accelerated NaCl osmotic stress-induced cell death at NaCl concentrations in the range of 165 mM to 250 mM (Fig. 6e). Our HILOs were maintained in differentiation media containing 110 mM NaCl. Using these findings, we combined transient stimulation by high osmolarity (180 mM NaCl) with Dex, and T3 to increase FXYD2 expression for further screening of FXYD2 highly expressing HILOs (Fig. 6f). The batch-to-batch variation in hPSC-derived human islet organogenesis may be improved by enhancing FXYD2 expression and depleting FXYD2-negative cells from HILOs. To visualize FXYD2 expression, we constructed the FXYD2-fLuc-mCherry in hiPSCs. Our new protocol facilitated the generation of FXYD2 high-expressing HILO clusters, and heterogeneous mCherry expression could be observed (Fig. 6g). This enabled us to isolate FXYD2 high-expressing (FXYD2^High) and low-expressing (FXYD2^Low) HILOs respectively. To conceptualize the functional heterogeneity distinguished by FXYD2, we sorted FXYD2^High (mCherry +) and FXYD2^Low (mCherry-) β-like cells (hINS-GFP +) and created size-matched pseudo HILOs (pHILOs) (Fig. 6g). Importantly, FXYD2^High pHILOs exhibited greater GSIS and KCl-induced insulin secretion compared to FXYD2^Low pHILOs, indicating that FXYD2 may serve as a functional marker for HILOs (Fig. 6h). These findings suggested the potential of FXYD2 as a marker for functional heterogeneity in human β cells and offers a strategy for in vitro screening of FXYD2^High functional HILOs.

Fig. 6: Generation of FXYD2^High HILOs.

a Drug screening (1 μM, 24 h stimulation) in FXYD2 promoter driven luciferase and mCherry expressing EndoC-βH1 cells. Dex and T3 upregulated FXYD2 promoter activity. The heatmap is shown by three independent screening data with fold change to control (DMSO, n = 18). b qPCR analyses of FXYD2 gene expression in EndoC-βH1 cells. Dex (10 μM for 24 h), T3 (1 μM for 24 h) and retinoic acids (RA, 2 μM for 24 h) synergistically enhanced FXYD2 gene expression in EndoC-βH1 cells. Brain-derived neurotrophic factor (BDNF, 10 μg/ml for 24 h) and ciliary neurotrophic factor (CNTF, 10 μg/ml for 24 h) are used as negative control. n = 3. Unpaired two-tailed t-tests. c qPCR analyses of indicated gene expression in EndoC-βH1 cells and hiPSC-derived β-like cells. NaCl stimulation enhanced FXYD2 and ATP1B1 gene expression in EndoC-βH1 cells and hiPSC-derived β-like cells (dispersed HILOs in 2D). n = 3. Unpaired two-tailed t-tests. d qPCR analyses of FXYD2 gene expression in EndoC-βH1 cells. Dex and NaCl synergistically stimulate FXYD2 expression in EndoC-βH1 cells. n = 3. Unpaired two-tailed t-tests. e Cell survival was measured by cellular ATP level (ratio). dFXYD2OE (DOX +) showed resistance to NaCl (osmotic pressure)-induced cell death in EndoC-βH1 cells. In contrast, FXYD2KD shows increased sensitivity to NaCl induced cell death in EndoC-βH1 cells. n = 3. Two-way ANOVA. f Schematics of generation of FXYD2^High and FXYD2^Low HILOs. Transient Dex 10 μM and NaCl 180 mM stimulation was used to improve visualization of FXYD2. g Representative fluorescence images of FXYD2-driven mCherry (red) and corresponding optical image. FXYD2^High and FXYD2^Low HILOs can be identified by FXYD2-fLuc-mCherry expression. FXYD2^High and FXYD2^Low enriched insulin positive cells were isolated from HILOs. Size-matched pseudo-HILOs (pHILOs) were generated from FXYD2^High and FXYD2^Low expressing HILOs. h Human c-peptide secretion response (pM/10,000 cells) in FXYD2^High and FXYD2^Low pHILOs. n = 3. Unpaired two-tailed t-tests. Scale bars = 100 μm. G3 = Glucose 3 mM, G20 = Glucose 20 mM, KCl20 = KCl 20 mM. Error bars indicate SEM.

We noticed that FXYD2 expression is limited in insulin producing cells of HILOs compared to a pure human β cell line (Supplementary Fig. 10f). Therefore, we investigated the transcriptome features of the FXYD2^High and FXYD2^Low insulin producing cells of HILOs. We purified INS^High/FXYD2^Low and INS^High/FXYD2^High cells using a dual reporter for insulin promoter driven GFP and FXYD2 promoter driven mCherry at day 30 of HILOs (Fig. 7a). Bulk RNA-seq analyses revealed that INS^High/FXYD2^Low cells were characterized by higher glycolytic transcriptomes, while INS^High/FXYD2^High cells were characterized by hyper insulin secretion, β cell identity (Type II diabetes mellitus) and mineral absorption transcriptomes (Fig. 7b). Specificially, INS^High/FXYD2^High cells were enriched for the β cell identity and maturity related genes such as MAFA, UCN3, SIX3, WNT4, G6PC2, IAPP, FXYD2 and INS (Fig. 7c, d). We expanded these findings in integrated scRNA-seq data sets (Fig. 1a). FXYD2 expression is induced in in vitro maturation and in vivo maturation data sets (Supplementary Fig. 11a). The population analyses revealed that INS^High/FXYD2^High population in hPSC-derived islets range from 7 – 54% during in vitro differentiation and in vivo maturation, and from 50.4 – 82.5 % in primary human β cells (Supplementary Figs. 11b, 12b). Importantly, we observed that INS^High/FXYD2^High population is enriched with maturity-related genes, including G6PC2, IAPP, MAFA or MAFB, based on in vitro and in vivo maturation, as well as primary islet data sets. In contrast, the INS^High/FXYD2^Low population did not exhibit enrichment for other endocrine hormonal markers, such as GCG or SST, nor for progenitor markers like NGN3 or SOX9 (Supplementary Fig. 11c). Supporting this notion that UMAP clustering analyses revealed that the INS^High/FXYD2^High and INS^High/FXYD2^Low population is widely dispersed in immature islet status, but it is more enriched in similar UMAP area based on in vitro and in vivo maturation and primary human islets (Supplementary Figs. 11d, 12a, b). In particular, we observed INS^High/FXYD2^High and INS^High/FXYD2^Low population located in the almost identical UMAP area in primary human islets, while it is showed significant transcriptome differences including FXYD2, INS, IAPP and NKX2-2 (Supplementary Fig. 12c).

Fig. 7: INS^High/FXYD2^High HILOs exhibit enriched transcriptomic profiles indicative of functional maturation.

a Sorting strategy of INS^High/FXYD2^Low and INS^High/FXYD2^High HILOs. b–d Bulk RNA-seq analyses of INS^High/FXYD2^Low and INS^High/FXYD2^High HILOs. KEGG pathway analysis (b), Differential expression gene (DEG) analyses with -log₁₀ (padj = p-value adjusted) > 1.3. (c) and Heatmap analyses (d) of INS^High/FXYD2^Low and INS^High/FXYD2^High HILOs are shown. Samples were obtained from n = 3 independent batches of differentiations.

These results suggest that functional differences between INS^High/FXYD2^High and INS^High/FXYD2^Low population is not likely due to the enrichment of multihormonal cells, rather than the enrichment of functionality and maturity of the cells.

FXYD2^High but not FXYD2^Low HILOs restores euglycemia in diabetic mice in vivo

Functional immaturity is the hallmark of hPSC-derived islets. However, it has not yet been demonstrated whether the GSIS response in each individual hPSC-derived islet is homogeneous or heterogeneous. Therefore, our investigation sought to identify functionality at the single organoid level and assess FXYD2’s potential to serve as an indicator of functional diversity within islets derived from stem cells. To achieve this, we established a triple reporter assay system, with insulin expression indicated by GFP, insulin secretion functionality tested by Gaussian proinsulin NanoLuc, and FXYD2 expression highlighted through a firefly luciferase and mCherry dual-reporter system. This innovative setup revealed a correlation between the variance in GSIS and the levels of intracellular FXYD2 activity, as evidenced by firefly luciferase, along with mCherry expression. The Gaussia NanoLuc assay showed varying degrees of GSIS response in each individual HILO, suggesting functional heterogeneity in generated hPSC-derived islets (Fig. 8a). We found that secreted Gaussia luciferase activity correlated with intracellular FXYD2 firefly luciferase activity, suggesting that FXYD2 marks the functional heterogeneity of HILOs (Fig. 8b). To streamline the process, we adopted a manual approach to separate the HILOs based on the intensity of FXYD2 expression as visualized by mCherry fluorescence under insulin-GFP expression (Fig. 8c). This stratification revealed that FXYD2^High HILOs demonstrated a superior insulin secretion capability in comparison to FXYD2^Low HILOs (Fig. 8d). The culmination of this approach was the successful isolation of a more uniform population of HILOs distinguished by their pronounced GSIS functionality. These findings suggested that FXYD2 may be a reliable marker of functional diversity within HILO clusters.

Fig. 8: FXYD2^High HILOs restores euglycemia in diabetic mice in vivo.

a Secreted Gaussia luciferase and intracellular firefly luciferase activity in each HILOs is shown. Individual HILO insulin secretion was measured by the Gaussia Nano-Luc system and intracellular FXYD2 gene activity was measured by firefly luciferase. n = 48. b Correlation of secreted Gaussia Nano-Luc and intracellular FXYD2 promoter driven firefly luciferase. XY analyses is shown by simple linear regression. R² = 0.509, p < 0.0001. c INSULIN promoter driven GFP (green), FXYD2 promoter driven mCherry and optical image of isolated FXYD2 enriched (FXYD2^High) and less enriched (FXYD2^Low) HILOs (scale bars = 300 μm). d Human insulin secretion (mU/L/10,000 cells) from FXYD2^High and FXYD2^Low HILOs. n = 9. Unpaired two-tailed t-tests. e Schematic of transplantation. f Fed ad lib, blood glucose levels in 500 clusters (~1500 IEQ) of FXYD2^High (n = 6), FXYD2^Low (n = 5) or Sham (n = 2) transplanted STZ-NOD-SCID mice for ~12 weeks. Graft removal (nephrectomy, Nx) was performed on day 84 post-transplantation. Two-way ANOVA. g % of body weight change during the transplantation study is shown. FXYD2^High (n = 6), FXYD2^Low (n = 5) or sham (n = 2). Unpaired two-tailed t-tests. h Serum human c-peptide levels (pM) were measured before and 15 min after i.p. glucose injection at 8 weeks after transplantation. The right panel shows the stimulation index to GSIS (fold). FXYD2^High (n = 6), FXYD2^Low (n = 5). Unpaired two-tailed t-tests. i Representative immunofluorescence images of insulin (green) and FXYD2 (red) in transplanted kidney sections 84 days after transplantation (scale bars = 100 μm). G3 = Glucose 3 mM, G20 = Glucose 20 mM. Error bars indicate SEM.

Next, to test whether FXYD2^High HILOs show a more stable glucose lowering effect than FXYD2^Low HILOs in vivo, we transplanted 500 clusters of FXYD2^High or FXYD2^Low HILOs into streptozotocin (STZ)-induced diabetic NOD-SCID mice (STZ-NOD-SCID) (Fig. 8e). We found that the STZ-NOD-SCID mice transplanted with FXYD2^High HILOs exhibited improved hyperglycaemia and reversed body weight loss compared to those receiving FXYD2^Low HILOs (Fig. 8f, g). Graft removal (Nephrectomy, Nx) at 84 days post-transplantation led to reinstated hyperglycaemia in STZ-NOD-SCID mice. (Fig. 8f). These results suggested that FXYD2^High HILOs are more functionally homogeneous for the treatment of diabetes. Although FXYD2^Low HILOs were less effective in glycemic control compared to FXYD2^High HILOs, they still led to slightly improved hyperglycaemia in comparison with a sham control, suggesting that FXYD2^Low HILOs could undergo functional maturation in vivo. We performed an intraperitoneal glucose tolerance test (i.p. GTT) to evaluate the GSIS function in vivo in the mice receiving transplanted FXYD2^High or FXYD2^Low HILOs at 8 weeks post-transplantation. The i.p. GTT indicated higher human c-peptide secretion in mice transplanted with FXYD2^High HILOs mice compared to those with FXYD2^Low HILOs (Fig. 8h). Insulin positive cells from the explanted FXYD2^High HILOs ~3 months later retained augmented FXYD2 expression compared with FXYD2^Low HILOs (Fig. 8i). Supporting this notion, we found that both FXYD2^High and FXYD2^Low HILOs grafts still expressed FXYD2 at 3 months post-transplantation (Fig. 8i, Supplementary Fig. 13a). Collectively, the transplantation study revealed that FXYD2^High HILO-enriched clusters show profound glycemic control compared to FXYD2^Low HILO-enriched clusters. These results suggest that FXYD2 marks functional heterogeneity in the hPSC-derived islet organoid system and that FXYD2 may be a reliable marker for functional diversity in HILOs.

Discussion

The generation of functional human islet organoids has been a focal point of research for decades, yet challenges remain, particularly the issue of batch-to-batch variability, which hampers reproducibility. Our study provides novel insights into these challenges. We discovered that the functional immaturity of these islets is not uniformly distributed, resulting in heterogeneous insulin secretion capabilities (Fig. 8a). Controlling this heterogeneity may be possible by targeting markers of maturation. We observed that the mineral absorption pathway under the control of FXYD2 is one of common downregulated features in hPSC-derived β-like cells compared to that of human primary β cells (Supplementary Fig. 1d). Analysis of public scRNA-seq data confirmed that poor FXYD2 expression is a widespread issue among hPSC-derived β-like cells, irrespective of their origin.Through our investigations, FXYD2 emerged as a significant marker indicative of functional maturation and heterogeneity within HILOs. This is also consistent with the recent observation that FXYD2 expression is increased during in vivo maturation of sc-islets⁵¹ and is recognized as β cell specific biomarker⁷³, highlighting its significance in improving islet functionality.

Interestingly, our research also revealed that T3 enhances FXYD2 expression in human β cells (Fig. 6a) and even in the absence of T3 stimulation, overexpression of FXYD2 leads to the amplification of TR signalling (Fig. 4g). Considering this positive feedback loop by FXYD2, it is not surprising that FXYD2^Low HILOs failed to show efficient in vivo maturation (Fig. 8f), which typically occur within 2-3 months after transplantation, since the TR signalling pathway has been recognized for its role in facilitating the maturation of β cells in vivo^48,49, binds to the promoter region of MAFA, a master regulator of β cell function⁴⁵. As mechanistic insights, we found that FXYD2 physically interacts with SRC to regulate the TEAD1 signalling pathway to increase key β cell maturation genes. YAP1, a known interacting partner of TEAD1, is implicated in promoting pancreatic progenitor proliferation^74,75 therefore YAP1 inhibition in pancreatic progenitor stage has been known to promote β cell differentiation^74,76. It is notable that YAP1 is not present in mature β cells and exhibits low expression in EndoC-βH1 cells and HILOs (Supplementary Data 2), despite its typical association with TEAD1 expression. In contrast, it is acknowledged that marked expression of TEAD1 in mature β cells⁷⁷. This may suggest that TEAD1 plays an important role in mature β cells, independent with YAP pathway. Our findings resonate with this perspective, as we observed that TEAD1 directly binding to the promoter region of MAFA, PDX-1, NKX6-1, UCN3, IAPP and NEUROD1 to regulate the expression of maturity-associated β cell genes in human β cells. This aligns with recent studies that have pinpointed that β cell-specific TEAD1 knockout mice exhibited β cell growth defects and developed diabetes⁷⁷. Additionally, SRC has been implicated to enhance insulin secretion in β cells⁶⁶. Our results suggested that possible TEAD1 regulation under the control of SRC in mature β cells (Supplementary Fig. 9a). Although SRCi almost completely abolished pSRC, TEAD1 expression was only partially reduced, suggesting the potential involvement of alternative regulatory mechanisms. In β cells, VGLL4 and MENIN proteins are suggested directly bind to TEAD1 and regulate gene expression involved in β cell differentiation and function⁷⁸. VGLL4 and MENIN has been implicated in pancreatic islet development and might enhance β cell maturation through TEAD1 activation. BRD4 is an epigenetic regulator known to promote transcription of key β cell genes. Studies suggest that TEAD1 activates WNT4, which is suppressed by BRD4 by direct protein-protein interaction in cardiac fibroblasts⁷⁹. Given its role in modulating β cell maturation by WNT4, BRD4 may compensate for reduced TEAD1 expression by upregulating TEAD1 target genes or recruiting other transcriptional activators. Together, these alternative pathways suggest that VGLL4 and MENIN or BRD4 pathway might interfere for the FXYD2-SRC-TEAD1 pathway by maintaining TEAD1 transcriptional activity and regulating β cell maturation and insulin secretion. Further studies will be necessary to confirm their contributions to TEAD1-dependent β cell function. FXYD2 is also known to modulate Na⁺/K⁺ ATPase activity by blocking Na⁺-current⁸⁰, and previous reports suggest that Na⁺ currents are approximately twice as high in hPSC-derived β-like cells, indicating aberrant Na⁺/K⁺ ATPase regulation²⁸. This aberrant regulation may contribute to the functional immaturity observed in these cells. These observations suggest that FXYD2 serves a dual role as both a channel modulator and a pivotal hub of signal transduction in β cells.

Balancing cell identity and cell cycle progression are part of a key aspect in the maturity of β cells⁵⁶. It is noteworthy that FXYD2 knockout (KO) mice exhibited β cell hypertrophy and improved glucose tolerance, implicating β cell hypertrophy as a compensatory mechanism for insulin secretion in response to glucose unresponsiveness for FXYD2KO’s phenotype⁵⁵. Human β cells, unlike those of rodents, have a limited capacity for self-replication^57,58,59,60. This led us to hypothesize that FXYD2 is integral to the proper functioning of acute GSIS in human β cells. Indeed, our dFXYD2OE in HILOs and EndoC-βH1 cells showed enhanced gene expression associated with β cell function and maturation, while reduced gene expression associated with β cell cycle progression (Fig. 3c). Supporting this notion that from recent advanced human islet data base⁸¹, we confirmed that the positive correlation between FXYD2 expression and human β cell function (Supplementary Data 1). We also found that FXYD2 is the solo FXYDs family member enriched in human β cells, while FXYD2, 3, 4, 6 are enriched in mouse β cells (Supplementary Fig. 1f). This suggests that FXYD2 function may be a more indispensable function to positively regulate insulin secretion in human β cells rather than in mouse β cells. However, the degree of this positive regulation of human β cell function by FXYD2 may differ in the stage of developments and/or the level and length of expression, as per what we observed that in the different experimental models (Supplementary Data 2). Recent studies also identified HNF1A as a key driver of intra-donor heterogeneity among β cells, with reduced expression in β cells from Type 2 diabetes (T2D) donors. Given that FXYD2 is a known target of HNF1A⁸², its activity could influence β cell functionality by modulating sodium currents, potentially affecting insulin secretion during T2D development.

Our luciferase assay revealed that FXYD2 expression and GSIS function positively correlated in individual HILOs. This suggests that the ratio of FXYD2 highly expressing insulin-producing cells in each cluster of HILOs might be a part of the determining factor of functional heterogeneity in individual HILOs. One limitation of hPSC-derived islets is the presence of multihormonal immature cells, which are normally only observed in the foetal stage in development, which raises the question of whether the observed better function of the INS^High/FXYD2^High population compared to the INS^High/FXYD2^Low population is due to the increase of multihormonal cells in INS^High/FXYD2^Low population. scRNA-seq analyses of the INS^High/FXYD2^High and INS^High/FXYD2^Low populations revealed no increase in the expression of multihormonal markers GCG and SST, or the endocrine progenitor marker NGN3, in the INS^High/FXYD2^Low population. This suggests that cell plasticity is not the primary driver of the observed functional differences. Previous studies identified that CD63, CD81, and UCN3 (urocortin 3) mark functional heterogeneity within the β cells^83,84,85. CD63 has been implicated in the trafficking of secretory granules within β cells and marked functional heterogeneity within the β cells⁸⁵. In contrast, CD81 has been shown to mark immature and dedifferentiated β cells⁸⁴. Evidence suggests that another maturation marker UCN3 also participates in the local regulation of insulin release, potentially through autocrine or paracrine signalling within the islets⁸⁶. Somatostatin produced from δ cells is known as an important counter regulatory signal for β cell insulin secretion. Interestingly, FXYD2 is expressed not only in β cells but also δ cells (Supplementary Fig. 1e), however the potential role in paracrine regulation of β cells by FXYD2 in δ cells is currently unknown. Future studies are needed to identify the role of FXYD2 in δ cells. FXYD2, unlike these other markers, acts as a not only heterogeneity marker, but also as a pivotal regulator for β cell maturation, providing novel insights on the regulation of maturity by membrane protein.

The concentration of Na⁺ inside and outside cells is tightly regulated. Typically, the concentration of Na⁺ is higher outside cells. This gradient is maintained by the sodium-potassium pump (Na⁺/K⁺ ATPase), which actively transports Na⁺ out of the cell and K⁺ into the cell. Changes in intracellular Na⁺ concentration can influence cellular metabolism and ionic balance, which can indirectly affect epigenetic modifications. For example, changes in cellular metabolism can alter the availability of substrates needed for epigenetic modifications like DNA methylation or histone acetylation. Our study sheds light on the fact that elevated NaCl concentration enhances the expression of open chromatin markers, FXYD2 and SRC signal pathway members, and alters gene expression in EndoC-βH1 cells. This is also in line with recent studies that have underscored disparities in chromatin accessibility and gene expression at the single-cell level when comparing stem cell-derived islets to primary islets²⁹. These results suggest that the membrane protein FXYD2 can contribute to gene expression regulation through ion-channel mediated signal transduction. Further investigation to optimize the degree of gene expression regulation by natural or synthetic ligands and Na⁺ concentration is required to further characterize the precise mechanism of gene regulation in human β cells.

In summary, through physiological assays, transcriptome analysis, and in vitro human islet organogenesis, we identified FXYD2 as a pivotal regulator of functional maturation and a marker for both maturation and heterogeneity in HILOs (Supplementary Fig. 13b). These findings may offer valuable insights that could lead to novel strategies for generating and screening fully functional human islet organoids.

Methods

Resources

All critical resources are summarized in Supplementary Data 3.

Maintenance of mouse lines

Animals were maintained in a pathogen-free animal facility on a 12-h light-dark cycle at an ambient temperature of 23 °C. Water and food were provided ad lib. Age- and background-matched male NOD-SCID mice (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ, stock number 005557) were used for experiments, and all procedures were performed in accordance with protocols approved by the IACUC and the Animal Resources Department of the Lundquist Institute at Harbour-UCLA Medical Center (#32082).

Cell culture

Human pluripotent stem cells (hPSCs) such as human induced pluripotent stem cells (ChiPSC12, TAKARA) and human embryonic stem cells (H1ES, WiCell) were maintained in Cultrex RGF Basement Membrane Extract (Bio-techne, 3433-010-01)-coated plates with mTeSR Plus (STEMCell Technologies, 100-027). hPSCs were passaged every 4–7 days with ReLeSR (STEMCell Technologies, 05872). Single cell suspensions were prepared using Accutase (STEMCell Technologies, 07920), washed in PBS, and collected by centrifugation (153 g/1300 rpm for 5 min). EndoC-βH1 cells (UniverCell) were maintained in Cultrex RGF Basement Membrane Extract-coated (4 μg/ml at 37 °C for 1 h–overnight pretreatment) plates with EndoC-βH1 culture media with the following ingredients: MCDB131 media (Sigma, M8537-10X1L); 0.1745% NaHCO3 (Spectrum, S1147); 2% Free fatty acid free BSA fraction V (GoldBio, A-421-1); 2 mM Glutamax-I (Gibco, 35050-061); 1% Penicillin/Streptomycin (Gibco, 15140-122); 0.25 mM Ascorbic acids (Sigma, A4544-25G); 10 μM Zinc sulphate (Sigma, Z0251-100G); 55 μM 2-Mercaptoethanol (Gibco, 21985-023); 1 × MEM-NEAA (Gibco, 11140-050); 10 mM HEPES (Gibco, 15630-080); 11 mM glucose (Figher, BP350-1). EndoC-βH1 cells were passaged 1–2 times every week. HEK293LTV cells (Cell Biolabs, LTV-100) were cultured in DMEM (Sigma, D6546-500ML) containing 10% FBS (Genesee Scientific, 25-525H), 1% Glutamax-I and 1% Penicillin/Streptomycin. HEK293LTV cells were passaged every 2–4 days. hPSCs, EndoC-βH1 cells and HEK293LTV cells were cryopreserved with Bambanker (FujiFilm Wako, 302-14681) in a –80 °C freezer or liquid nitrogen. All human pluripotent stem cell studies were conducted under the oversight of the Cedars Sinai Stem Cell Research Oversight (SCRO) as part of collaborative effort with Lundquist Institute (STUDY00000916).

Generation of human islet-like organoids (HILOs)

Human islet-like organoids (HILOs) were generated according to a previous study¹ with minor modifications. Briefly, hPSCs were cultured on Cultrex gel-coated plates until it reached approximately 90% confluence. hPSC suspensions were prepared using ReLeSR or Accutase and cultured in ultra-low attached 6-well plates (2 × 10⁶ cells/mL) with orbital shaking (~60 rpm for 2 days), EZSPHERE (IWAKI, 4810-900SP), Aggrewell 400 or 800 (StemCell Technologies, 34860 or 34825) at 5000–10,000 desired cell numbers per microwell in the presence of the ROCK inhibitor (10 μM Y-27632, StemCell Technologies, 72308). Cells were then resuspended with 3D Kelco Gel Stem TeSR™ Base Medium in the presence of the ROCK inhibitor (10 μM) for 0-2 days until spheroids reached ~200 μm in diameter. The medium was then replaced with 0.015% Kelco gel containing 0.3% methylcellulose and supplemented with 100 ng/ml human Activin A and 3 μM CHIR99021 in differentiation medium (S1) for 1 day and then 100 ng/ml human Activin A in differentiation medium (S1) for another 2 days (stage 1, definitive endoderm). Subsequently, the medium was changed to differentiation medium (S2) with 50 ng/ml FGF7 for 2 days (stage 2, foregut/FG), then to differentiation medium (S3) with 50 ng/ml FGF7, 0.25 μM SANT-1, 1 μM retinoic acid, 100 nM LDN193189, 10 μM RepSox, and 200 nM α-amyloid precursor protein modulator TPB for 2 days (stage 3, pancreatic progenitor/PP), then 50 ng/ml FGF7, 0.25 μM SANT-1, 100 nM retinoic acid, 100 nM LDN193189, 10 μM RepSox, and 100 nM of the α-amyloid precursor protein modulator TPB for 3 days. Subsequently, medium was replaced with differentiation medium (S4) with 0.25 μM SANT-1, 50 nM retinoic acid, 100 nM LDN193189, 10 μM RepSox, 1 μM T3 for 3 days (stage 4, endocrine progenitor/EP). Subsequently, the medium was replaced with differentiation medium (S5) containing 100 nM LDN193189, 100 nM γ-secretase inhibitor XX, 10 μM RepSox, and 1 μM T3 for 7 days (stage 5, immature β-like cells/imβ). Subsequently, the medium was replaced with differentiation medium (S5) containing 10 μM Trolox, 2 μM R428, 1 mM N-acetyl cysteine, 10 μM RepSox and 1 μM T3 for an additional 5 days (day 25) (stage 6, immature β-like cells/imβ). From day 26, the medium was replaced with differentiation media (S5) with 1 μM T3 and WNT4 supernatant (Final volume 10%, ~100 ng WNT4) (stage 7, mature β-like cells/β). WNT4 supernatant was generated using DOX-inducible WNT4 (pLV[Exp]-Bsd-TRE>hWNT4[ORF010485](ns):T2A:mCherry and human Ubiquitin C promoter-rtTA-IRES-puromycinR) expressing HEK293. After 24 h of DOX (+) treatment, medium was replaced with S5 Media and subsequentially stimulated with DOX (+) for 3 – 4 days. Supernatant was then filtered using 0.22 μm filter and the batch containing >100 mg/ml of WNT4 was frozen as stock until further use. After 30 days, the HILOs were maintained in differentiation media (S6) containing 1 μM T3. For 2D culture, HILOs were single-cell dissociated by TrypLE with 10 ng/ml DNase I for 12 min at 37 °C in a shaking water bath and then cultured in Cultrex gel-coated plates or cryopreserved using Bambanker at 10–20 million cells/1 ml until used for experiments. Media information and differentiation strategy is given Supplementary Data 4.

Single cell RNA-seq analysis

To comprehensively examine the gene expression in HILOs and human islets, publicly available datasets were downloaded from the previous studies^2,3,4,5; GSM4274565 (a), GSM4274567 (b), GSM4274569 (c), GSM4274574 (d), GSM4274575 (e), GSM4274578 (f), GSM5114460 (g), GSM4274580 (h), GSM3141962 (i), GSM3141967 (j), GSM4274583 (k), GSM5114461 (l), GSM5114467 (m), GSM5114469 (n), GSM5114472 (o), GSM3142001 (p). Using the Seurat package (version 5.0.0) in R, individual Seurat objects were established for each file via the CreateSeuratObject command, allowing efficient organization and preprocessing of each dataset. These individual objects were then merged into a single, integrated dataset with the merge function using default settings. This merging step aligned the datasets for downstream comparative analyses to explore potential differences and similarities in gene expression profiles between HILOs and human islets. A standardized preprocessing pipeline was implemented to ensure consistency across samples. First, data normalization was performed with Seurat’s NormalizeData function, which standardizes gene expression values across cells, facilitating inter-sample comparisons. The FindVariableFeatures function was then applied to identify variable features, focusing on genes with high variation as they often contain the most biologically relevant signals for clustering and downstream analyses. Following the identification of variable features, data scaling was conducted using the ScaleData function in preparation for further analyses. Dimensionality reduction was achieved through principal component analysis (PCA) using the RunPCA function, which retained the principal components capturing the majority of variance within the data. To address potential batch effects and effectively integrate multiple samples, the harmony package was applied. Harmony was specifically chosen for its robustness in correcting batch effects in high-dimensional datasets, aligning data across diverse samples. The reduction dimensions were set to 1:50, optimizing data integration while preserving meaningful biological differences. After integration, cells from distinct datasets were evaluated for connectivity as a quality control step, which was critical to reducing technical noise and enhancing the biological interpretability of the integrated data, as well as minimizing batch effects that could obscure true biological variation. Visualization of the integrated dataset was performed using Uniform Manifold Approximation and Projection (UMAP) to obtain a clear and interpretable representation of cell clusters. UMAP plots, generated through Seurat’s DimPlot function, provided insights into the distribution of cell populations across samples. For more granular analysis of gene expression, the FeaturePlot function visualized spatial expression patterns of specific genes, aiding in the identification of cell types and subpopulations within the dataset. Average expression levels of key genes across clusters were calculated and compared using the AverageExpression function, offering a deeper understanding of gene regulation dynamics in HILOs and human islets. To further investigate β-cell-like populations, cells expressing the INS gene, a marker for insulin-producing cells, were isolated using the WhichCells function, selecting those above the mean expression level. This INS+ (INS^High) population was then stratified into FXYD2^High and FXYD2^Low subgroups based on the mean expression level of FXYD2. Functional differences between these high and low FXYD2-expressing cells were analysed, potentially revealing distinct β-cell-like states or functional subtypes. For comparative analysis between FXYD2^High and FXYD2^Low groups, the FindMarkers function in Seurat was utilized, identifying differentially expressed genes based on fold change and statistical significance for robustness. Significant genes were further examined for functional relevance, potentially providing insights into molecular mechanisms associated with FXYD2 expression levels. Gene expression distributions within each group were visualized with the VlnPlot function, offering a clear view of expression changes over time or under different conditions. To investigate potential lineage trajectories among cell types, we used the Slingshot package, which is well-suited for trajectory inference in scRNA-seq data. Slingshot was applied to model the developmental pathways and differentiation states present within the dataset. Using the getLineages function, we specified human embryonic stem cells (hESCs) and human induced pluripotent stem cells (hiPSCs) as the initial cluster, serving as the starting point for trajectory analysis. This enabled us to track the differentiation pathways and lineage relationships, providing insights into the developmental processes of HILOs and human islet cells. We further refined our trajectory analysis by utilizing the getCurves function to smooth lineage pathways and the plotGeneCount function for visualizing gene expression patterns along these inferred trajectories. This approach facilitated the identification of key genes associated with different developmental stages, as well as the transitions between cell states, contributing to a more comprehensive understanding of cell lineage dynamics in this system. For scRNA-seq data preprocessing and differential expression analysis, data normalization was performed using the LogNormalize method with a scaling factor of 10,000. Highly variable genes (HVGs) were identified using the FindVariableFeatures function with default parameters.　Cells in the top 50% for INS or FXYD2 gene expression were independently defined as positive for each gene. Subsets of INS+ and FXYD2+ cells were selected using the WhichCells command. DEGs between INS + /FXYD2+ cells and INS + /FXYD2- cells were identified using the FindMarkers function. A Wilcoxon rank-sum test was used with an adjusted p-value threshold of 0.05.

For scRNA-seq of primary human islets, data alignment and quantification were performed following established procedures. The FASTQ files were obtained from SRR8199931 and processed using Cell Ranger (version 8.0.1). The analysis was performed to align reads to the reference genome GRCh38, quantify gene expression, and generate count matrices using the “cellranger count” command. The output gene expression matrix, barcode information, and summary statistics were used for downstream analysis in R. In addition to SRR8199931 Cell Ranger outputs, scRNA-seq gene expression datasets from GSM3142001 and GSM5603741 were obtained and processed using the Seurat package (version 5.1.0) in R (version 4.4.1). Cells were filtered to retain those expressing between 200 and 5000 genes, and cells with mitochondrial gene expression exceeding 10% were excluded to remove low-quality or dying cells. Data normalization was performed using the LogNormalize method with a scaling factor of 10,000. Highly variable genes (HVGs) were identified using the FindVariableFeatures function with default parameters. Cells in the top 25% for INS or FXYD2 gene expression were independently defined as positive for each gene. Subsets of INS+ and FXYD2+ cells were selected using the WhichCells command. DEGs between INS + /FXYD2+ cells and INS + /FXYD2- cells were identified using the FindMarkers function. A Wilcoxon rank-sum test was used with an adjusted p-value threshold of 0.05.

Virus production

Lentiviruses were produced using second- or third-generation lentiviral systems in a HEK293LTV cell line. Briefly, virus core plasmids pMD2.G (Addgene, 12259) and psPax2 (Addgene, 12260) or pMD2.G (Addgene, 12259), pRSV-Rev (Addgene, 12253) and pMDLg/pRRE (Addgene, 12551) and target lentivirus plasmids (Plasmid information is listed in Supplementary Data 5) were transfected into 70% confluent HEK293LTV cells in T75 flasks using Lipofectamine 3000 (Invitrogen, L3000015) and Opti-MEM (Gibco, 31985070) following manufacture’s protocol. One day after transfection, the media were changed to fresh HEK293LTV culture media (Day 0). Day 1 and Day 3 supernatants were collected and then cell debris were removed by centrifugation (153 g/1300 rpm, 5 min). Lentivirus containing supernatants were filtered using a 0.45 μm filter, then the lentivirus was used as a fresh virus or concentrated 10 × using Lenti X Concentrator (TAKARA, 631232) and DMEM/F12 (GIBCO,) for long-term storage at −80 °C in a deep freezer. Polybrene (Santa Cruz, sc-134220), was used as scaffolds for virus infection with spinfection (800 × g, 1 h). For dFXYD2OE or dTEAD1OE in EndoC-βH1 cells or ChiPSC12 cells, after 3 days co-infection of pLV[Tet]-Bsd-TRE>hFXYD2[NM_001127489.1](ns):T2A:mCherry, Tet-pLKO-puro-hFXYD2 shRNA, pLV[Exp]-Bsd-TRE>hTEAD1[NM_021961.6](ns)*:T2A:TurboRFP or pLV[3miR30]-Bsd-TRE>TurboRFP:{hTEAD1[miR30-shRNA#1]}:{hTEAD1[miR30-shRNA#2]}:{hTEAD1[miR30-shRNA#3]} and human Ubiquitin C promoter-rtTA-IRES-puromycinR, cells were treated with 0.8 μg/ml of puromycin (InvivoGen, ant-pr) and/or 10 μg/ml of blasticidin (InvivoGen, ant-bl) for 3 days in culture media to establish the stable lines. Similarly, for constitutive FXYD2 overexpression, pLV[Exp]-EF1A > hFXYD2[NM_001680.5](ns)/V5:(TurboID):T2A:EGFP; for Gaussia proinsulin nanoLuc, Proinsulin-NanoLuc in pLX304 were infected in EndoC-βH1 cells.

Transfection

EndoC-βH1 cells were transfected with negative control siRNA, FXYD2 siRNA or TEAD1 siRNA (Supplementary Data 5) for 72–144 h prior of RNA isolation, protein purification or insulin secretion assays. Lipofectamine 3000 (Invitrogen, L3000001) and Opti-MEM (Gibco, 51985-034) following manufacturer’s protocol. One day after transfection, the media were changed to fresh EndoC-βH1 Media.

Insulin/c-peptide secretion assays

Insulin secretion assay from primary human islets and HILOs was measured as previously described^1,6. Briefly, human islets (n = 10 each), HILOs (n = 20 each) were pre-cultured at 37 °C for 30 min in (Krebs-Ringer bicarbonate buffer (KRBH) containing 129.4 mM NaCl, 3.7 mM KCl, 2.7 mM CaCl₂, 1.3 mM KH₂PO₄, 1.3 mM MgSO₄, 24.8 mM NaHCO₃ (equilibrated with 5% CO₂, 95% O₂, pH7.4), 10 mM HEPES and 0.2% (v/v) BSA (fraction V, Sigma) with 3 mM glucose. Then, human islets (n = 10 each), HILOs (n = 20 each) were incubated in KRBH buffer (500 μl/1 ~ 20 clusters) supplemented with 3 mM glucose for 1 h (G3) and then with 20 mM glucose (G20) for another 1 h. Subsequently, the cell clusters were incubated with 20 mM KCl/3 mM glucose (KCl). The supernatant was collected after each stimulation. Finally, the clusters were pelleted. For Insulin secretion assay from EndoC-βH1 cells, we split EndoC-βH1cells 100,000–500,000 cells/well in Cultrex RGF Basement Membrane Extract-coated (4 μg/ml at 37 °C for 1 h pretreatment) 24 well plates. EndoC-βH1 cells were incubated in KRBH buffer (500 μl/1 ~ 20 clusters) supplemented with 3 mM glucose for 1 h (G3) and then with 20 mM glucose (G20), G20 + GLP-1 100 nM or G20 + GLP-1 100 nM + Glutamate 10 mM for another 1 h. Subsequently, the cell clusters were incubated with 20 mM KCl/3 mM glucose (KCl). The supernatant was collected after each stimulation. Intracellular and secreted insulin and c-peptide levels were determined by ultrasensitive Insulin ELISA KIT (Mercodia, 10-1132-01) or ultrasensitive c-peptide ELISA Kit (Mercodia, 10-1141-01).

Gaussia Luciferase and Firefly Luciferase assays

Gaussia luciferase assay was performed using Pierce Gaussia Luciferase Glow Assay Kit (Thermo Scientific, 16161). Firefly luciferase assay was performed using Firefly Luciferase Assay Kit 2.0 (Biotium, 30085-2). Experiments were performed following the manufacturer’s instructions.

Na⁺/K⁺ ATPase assay and Sodium assays

Na⁺/K⁺ ATPase assay was performed using ATPase assay Kit (Abcam, ab234055). To determine the intracellular sodium concentration, Sodium Assay Kit (Abcam, ab211096) was used. Briefly, 200,000 cells were treated with 110 mM or 180 mM NaCl containing media for 1–2 h and the experiments were performed following the manufacturer’s instructions.

Quantitative RT-PCR analysis

Total RNA was extracted using TRIzol (Invitrogen, 15596018) and the Aurum Total RNA mini kit (Bio-Rad, 7326820). Reverse transcription was performed using the iScript cDNA synthesis kit (Bio-Rad, 1708891 BUN). Real-time quantitative RT-PCR (qPCR) was performed using SYBR Green (Bio-Rad, 1725122) and analyses were performed using the ABI7900HT Fast Real-Time PCR System or the ABI StepOnePlus Real-Time PCR System. Primer information is provided in Supplementary Data 5.

Generation of bulk RNA-Seq libraries

Total RNA was isolated from cell pellets treated with RNAlater (Invitrogen) using the Aurum Total RNA Mini Kit (Bio-Rad, 7326820) and treated with DNaseI for 30 min at room temperature (RT). Sequencing libraries were prepared from 100-500 ng of total RNA using the TruSeq RNA Sample Preparation Kit v2 (Illumina) according to the manufacturer’s protocol. Briefly, mRNA was purified, fragmented, and used for first- and second-strand cDNA synthesis followed by adenylation of 3′ end adenylation. Samples were ligated to unique adaptors and amplified by PCR. Libraries were then validated on the 2100 Bioanalyzer (Agilent), and normalized and pooled for sequencing. Sequencing was performed by Novogene, Inc.

High-throughput sequencing and analysis

RNA-seq libraries prepared from three biological replicates for each experimental condition were sequenced on the Nova-seq 6000 PE150. Image analysis and base calling were performed automatically using the Illumina HiSeq real-time analysis software. Short read sequences were mapped to a UCSC hg38 reference sequence using the RNA-seq aligner STAR⁷. The hg38 associated known splice junctions were provided to aligner, and de novo junction discovery was also allowed. Differential gene expression analysis, statistical testing and annotation were performed using Cuffdiff 2⁸. Transcript expression was calculated as gene-level relative abundance in fragments per kilobase of exon model per million (fpkm) mapped fragments and correction for transcript abundance bias was applied⁹. Heatmaps were generated using the R-Script with heatmap.2 (gplot) software. The scale of heatmaps was determined by the Z-score.

Chromatin Immunoprecipitation (ChIP)-seq analyses

20 − 30 × 10⁶ EndoC-βH1 cells were fixed using 1% paraformaldehyde (Electron Microscopy Sciences, 15710) and disuccinimidyl glutarate (DSG, Thermo Scientific, PI20593). ChIP assay was performed using Pierce™ Magnetic ChIP Kit (Thermo Scientific, 26157) following manufacture’s protocol. Briefly, 40 million reads per sample were sequenced by Nova-seq 6000. The short DNA reads were aligned against the human hg38 reference genome using the Illumina CASAVA-1.8. Reads were aligned using the Bowtie aligner, allowing up to two mismatches in the read. Only tags that map uniquely to the genome were considered for further analysis. Subsequent peak calling and motif analysis were conducted using HOMER, a software suite for ChIP-seq analysis. The methods for HOMER, which are described below, have been implemented and are freely available at http://homer.ucsd.edu. One tag from each unique position was considered to eliminate peaks resulting from clonal amplification of fragments during the ChIP-seq protocol. Peaks were identified by MACS2 software. The threshold for the number of tags that determine a valid peak was selected for an FDR < 0.005, as empirically determined by repeating the peak finding procedure using randomized tag positions. Peaks are required to have at least 4-fold more tags (normalized to total count) than input or immunoglobulin G control samples and 4-fold more tags relative to the local background region (10 kb) to avoid identifying regions with genomic duplications or nonlocalized binding. Differential TEAD1 peaks are called by Homer using a threshold of 1.5 fold change and p value < 0.0001. Peaks are annotated to gene products by identifying the nearest RefSeq transcriptional start site. Visualization of ChIP-seq results was achieved using Integrative Genomics Viewer (IGV). Gene ontology analysis was performed using DAVID.

Co-immunoprecipitation and western blot analyses

V5-FXYD2-TurboID-EGFP overexpressed EndoC-βH1 cells were scraped and collected from the cell culture plate, then lysed with ice-cold Pierce RIPA buffer (Thermo Scientific, 89900) containing Halt protease and phosphatase inhibitor cocktail (Thermo Scientific, 78442). Subsequently, the samples were gently sonicated, then placed on ice for 15 min before being centrifuged at 15,000 × g for 15 min. Supernatants were collected and incubated with 5 µg of anti-V5 or anti-IgG antibody overnight at 4 °C, and then precipitated with Dynabeads Protein G (Thermo Scientific, 10003D) for 1 h at 4 °C with rotation (15 rpm). The Dynabeads were washed three times with ice-cold PBS, and the proteins were eluted by adding IgG elution buffer (Thermo Scientific, 21004), and LDS-sample buffer (Alfa Aesar, J61942.AD) for 5 min at 80 °C. The eluted samples were separated on a Bolt 4–12% Bis-Tris Plus Gel (Thermo Scientific, NW04125BOX) at 200 V for 22 min, and then transferred to an iBlot2 PVDF mini stack (Thermo Scientific, IB24002) using the iBlot2 transfer system. The membrane was blocked with 2% (w/v) nonfat dry milk (Apex, 20-241), washed five times with TBST (Bioland Scientific, TBST03-02) using SNAP i.d. 2.0 (Millipore) and then incubated in the primary antibody overnight at 4 °C. Anti-mouse or anti-rabbit-IgG-HRP was used as secondary antibody to detect undenatured antibodies. The membrane was stained with ProSignal Femto (Prometheus, 20-302B) and visualized using the Bio-Rad ChemiDoc MP imaging system (Bio-Rad).

Protein purification, NanoLC-MS/MS

20 − 30 × 10⁶ EndoC-βH1 cells expressing V5-FXYD2-TurboID-EGFP were washed with ice-cold Phosphate Buffered Saline (PBS) and collected by cell scraper. Cells were lysed in RIPA buffer along with halt protease/phosphatase inhibitor cocktail (ThermoFisher scientific, #78442). After mild sonication, cells were incubated for 15 min on ice before being centrifuged at 15,000 × g for 15 min to prepare cell lysate samples. Total protein was measured via Pierce BCA assay kit (ThermoFisher scientific, #23225). The total protein was normalized for each condition and treatment, then incubated with 5 μg of antibody overnight, before incubation with Dynabeads protein G (Invitrogen, #10004D) for an additional 60 min. The Dynabeads were collected magnetically and washed three times with ice cold PBS, transferred into a new tube, and stored at −80 °C. For nanoLC-MS/MS, equivalent amounts of pull-down samples were treated with 8 M urea, 10 mM TCEP, 100 mM TEAB pH 8.5 to solubilize bound protein. Samples were diluted to 6 M urea using 100 mM TEAB, and digested overnight at 37 °C with 0.04 µg of LysC (Promega - R-LysC), with constant rotation. Samples were subsequently diluted to 1 M urea with 100 mM TEAB, then incubated with 0.04 µg of Trypsin Gold (Promega) with continuous rotation. The resulting peptides were acidified (with 100% formic acid) to a final concentration of 2% formic acid (with 100% formic acid), isolated using a stacked C18/SCX STAGE tip¹⁰, and eluted with 5% ammonium hydroxide, 80% acetonitrile. Elutions were evaporated to dryness in a Speedvac vacuum concentrator and redissolved in 0.1% FA in water. NanoLC-MS/MS analyses were performed in UCI proteomic core. 75 micron inside diameter × 25 cm nanocapillary columns were packed in-house with ReproSil-Pur C18-AQ beads (1.9 μm diameter; Dr. Maisch GmbH). Using an EASY-nLC 1200, columns were equilibrated with 0.1% FA (solvent A) in water followed by sample injection and a discontinuous gradient from 0 to 5% solvent B (93% CH3CN in 0.1% FA/water) over 5 min then to 23% B over 60 min then to 35% B over 15 min, at a flow rate of 250 nL/min. Column eluate was electrosprayed directly from the column end into an LTQ Orbitrap Velos Pro mass spectrometer, acquiring FTMS precursor spectra at 60,000 resolution followed by MS2 spectra (FTMS, 7500 resolution) of the 15 most intense > +1 charged precursor ions with intensity > 2000 by HCD (30% NCE). After two fragmentations within 30 sec ions were dynamically excluded for 40 s via a 500-entry list with early expiration from the list after a detection within the exclusion period that fell below S/N = 2.0. MS and MS/MS data were searched for tryptic peptides using Mascot 2.7 against a database comprising UniProt_Human, UniProt_Mouse and common contaminants, allowing 20 ppm mass tolerance, 2+ , 3+ and 4+ charge states and Deamidated (NQ) and Oxidation (M) variable modifications.

Kinase microarray

2 × 10⁶ Control and cFXYD2OE EndoC-βH1 cells were treated with or without Propafenone (1 µM for 24 h; APExBIO B2281), then washed with ice-cold PBS. Cells were harvested with 100 μl ice-cold lysis buffer (Kinexus Lysis buffer with complete inhibitors). After mild sonication, samples were supplemented with 6 mM NTCB (2-nitro-5-thiocyanatobenzoic acid) followed by incubation in a 37 °C water bath for 15 min. The resulting lysate homogenate was centrifuged at 15,000 × g or greater for 30 min at 25 °C. Biological triplicated samples were pooled and supernatants were then snap-frozen and stored at –80 °C for subsequent study at Kinexus. Analyses were performed with technical duplicates.

Immunohistochemistry (IHC) and confocal microscopy analyses

IHC and confocal analyses were performed on frozen or paraffin-embedded sections of human islet-like organoids (HILOs) using the following antibodies: insulin (1/100, Abcam ab7842), FXYD2 (1/100, Sigma WH0000486M1-100UG or Invitrogen PA5-104165), CHGA (1/100, Abcam ab15160), glucagon (1/100, Abcam ab10988), somatostatin (1/100, Abcam ab108456). Secondary antibodies were coupled to Alexa 488, 568, 647 (Life Technologies) fluorophore and visualized by confocal microscopy (Leica SP8) or fluorescence microscopy (Zeiss, Axiovert 40 CFL). Hoechst 33342 (Thermo Scientific, 62249, 1 μg/ml final concentration) was used for nuclear staining.

FACS analysis and cell sorting

HILOs were dissociated by TrypLE (Gibco, 12604-013) and 100  μg/ml of DNase I (Bioland, EPD01-01) for 12 min with gentle shaking in 37 °C water bath. Cells were fixed and permeabilized by Transcription Factor Buffer Set (BD Pharmingen, 562574). Cells were labelled wither reporter system (Supplementary Data 2) or specific antibodies include Insulin (C27C9) Rabbit Antibody (Alexa Fluor® 488 Conjugate) (1/100, Cell Signalling, #9016), Insulin (C27C9) Rabbit Antibody (Alexa Fluor® 647 Conjugate) (1/100, Cell Signalling, #9008), FXYD2 Antibody [FITC] (1/100, Bio-Techne, NBP2-97415F) or FXYD2 Antibody [1/100, Alexa Fluor® 647] (Bio-Techne, NBP2-97415AF647). FACS analyses were performed using BD FACSymphony and cell sorting were performed using BD FACS Alia III.

Drug screening

To identify small molecules that could upregulate FXYD2 expression, we treated FXYD2-fLuc-mCherry EndoC-βH1 cells with 1 µM of compounds from the SCREEN-WELL Nuclear Receptor ligand library (Enzo BML-2802-0100), dispensing 100,000 cells into each well of a 96-well plate, and incubated them for 24 h. Subsequently, the cells were lysed to perform the Biotium Firefly Luciferase assay. Similarly, to discover small molecules that enhances insulin secretion, we treated Proinsulin NanoLuc EndoC-βH1 cells with 1 µM of compounds from the DiscoveryProbe Ion Channel Compound Library (APExBIO L1030), dispensing 100,000 cells into each well of a 96-well plate, and incubated them for 24 h. Afterward, the cells were incubated in 3 mM glucose KRBH buffer for 1 h, followed by incubation in 20 mM glucose KRBH buffer for another hour. The supernatant was then collected for the Pierce Gaussia Luciferase assay.

FXYD2^High and FXYD2^Low HILO transplantation

A three-step procedure including isolation of FXYD2 enriched HILOs was performed as follows: 1) Induction of FXYD2 using 10 μM Dex, 1 μM T3 and WNT4 containing maturation media for 5 days (Days 25–30); 2) Enrichment of FXYD2-positive β cells with HILOs by 180 mM NaCl + selection procedure (Days 30–Day 31); and 3) Isolation of FXYD2-enriched HILO clusters by fluorescent reporter (Day 31 -). Immunodeficient NOD-SCID mice were acquired from The Jackson Laboratory and housed in autoclaved cages in an SPF facility at the Lundquist Institute. Mice were rendered diabetic by a single high dose (180 mg/kg) injection of streptozotocin (STZ; i.p., Sigma S0130–500MG). Within one week after STZ injection, mice with blood glucose levels above 300 mg/dl were used as transplant recipients. FXYD2^High and FXYD2^Low HILOs were manually isolated by fluorescence microscopy from three different batches of HILOs (day 30 - day 40). 500 clusters of HILOs were resuspended in 200 µl of S5 Media, loaded into laboratory tubing (SiLastic, 508-004), and centrifuged (400 × g for 1−2 min) to generate cell clusters in the centre of the tubing. Cell clusters (approximately 30−50 μl) were transplanted under the kidney capsules in 8−16 week old STZ-injected diabetic mice. Ketamine (80 mg/kg) and xylazine (10 mg/kg) were used as surgical anaesthesia and the mice were placed on 37 °C heating pads for recovery. Blood glucose was monitored with a Nova Max Plus. Nephrectomies (Nx) and graft removal experiments were performed to confirm the efficacy of glucose regulation in the transplanted FXYD2^High and FXYD2^Low HILOs at the indicated days. The kidney with the graft was ligated to the renal hilum with 4-0 silk (LOOK, SP116), and then resected. Removed grafts were processed for IHC analyses.

Glucose tolerance test (GTT)

For the GTTs, mice were fasted for 6 h then given 2 g/kg body weight of glucose diluted with 0.9% NaCl saline solution by i.p. injection. Tail blood samples were collected both prior to and 15 min after the glucose injection to measure human c-peptide in the mouse serum. Blood glucose was monitored at 0, 15, 30, 60, 120 min after glucose injection. All mice were randomly assigned to experimental groups for further analysis and no samples were excluded from the analysis.

Software and programmes used for bioinformatics analysis

For genomic data analysis, the following software or programmes were used: R studio (https://posit.co/), Seurat (https://satijalab.org/seurat/)^11,12,13, DAVID¹⁴, UCSC genome browser (http://genome.ucsc.edu)¹⁵, Homer (http://homer.ucsd.edu/homer/)¹⁶, IGV¹⁷.

Statistics and reproducibility

All quantitative data are presented as mean ± standard error of the mean (SEM), unless otherwise indicated. Statistical analyses were performed using GraphPad Prism version 10. Unpaired, two-tailed t-tests, one-way ANOVA and two-way ANOVA tests were used as appropriate for hypothesis testing. Sample sizes, exact p values, and the number of biological replicates (n) are reported in the corresponding figure legends. For qPCR analyses, all data were obtained from biological triplicates, with each individual sample normalized using technical triplicates. All experiments were independently repeated at least three times with consistent results, unless otherwise stated. No data were excluded from the analyses, and no statistical methods were used to predetermine sample size. The investigators were not blinded to group allocation.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.