STARD10 regulates human pancreatic β cell differentiation and triglyceride metabolism

Summary

The STARD10 gene encodes for a lipid transfer protein and is associated with type 2 diabetes (T2D) risk and β cell function. However, the role of STARD10 in human β cell development and function are still unclear. Here, we deleted STARD10 in human embryonic stem cells and differentiated them into β-like cells. The deletion of STARD10 reduced the formation of INS⁺ β-like cells and proliferation. Lipidomics analyses revealed that STARD10-null β-like cells had higher triglyceride levels. Consistently, the expression levels of ETFB, which is involved in fatty acid β-oxidation, were severely reduced in STARD10-null β-like cells. STARD10-null β-like cells also had reduced glycolytic function, mitochondrial oxidative phosphorylation, and palmitate oxidation, which can likely contribute to slower proliferation and β cell dysfunction. Overall, our findings provided further insights into the role of STARD10 in human β cell development and function, and how its loss-of-function can contribute to increased T2D risk.

Graphical abstract

Highlights

• •STARD10 is increasingly expressed during human beta cell differentiation
• •STARD10 is involved in cell cycle and human beta cell formation
• •Loss of STARD10 results in triglyceride accumulation in hESC-derived beta cells
• •Loss of STARD10 impairs glycolysis, oxidative phosphorylation, and fatty acid oxidation

Teo and colleagues uncovered the role of T2D-associated STARD10 lipid transfer protein in human beta cells. They found STARD10 to be important for proliferation and INS+ beta cell formation. Loss of STARD10 increased intracellular triglyceride levels, decreased fatty acid beta-oxidation protein ETFB, reduced glycolysis, oxidative phosphorylation, and palmitate oxidation. These insights suggest how STARD10 loss can contribute to increased T2D risk.

Introduction

Meta-analyses of genome-wide association studies (GWASs) in the European population have identified the STARD10 locus as a type 2 diabetes (T2D) risk locus (Mahajan et al., 2018, 2022; Morris et al., 2012; Strawbridge et al., 2011; Voight et al., 2010). Risk variants in the STARD10 locus are also associated with higher fasting and 30 min post-glucose load plasma glucose levels (Nielsen et al., 2011; Scott et al., 2012; Strawbridge et al., 2011), lower fasting and 30 min post-glucose load plasma insulin levels (Nielsen et al., 2011; Strawbridge et al., 2011; Voight et al., 2010), lower insulinogenic index and HOMA-β but not HOMA-IR (Dimas et al., 2014; Nielsen et al., 2011), implying that they may affect the insulin secretory function of pancreatic β cells. Through expression quantitative trait locus (eQTL) analysis, the index variant at the STARD10 locus was found to be associated with lower STARD10 transcript expression in human islets (Carrat et al., 2017; van de Bunt et al., 2015). Notably, the genomic region containing the top most probable causal variants in the STARD10 risk locus was shown to interact with the STARD10 promoter via chromatin looping studies, and deletion of this genomic region lowered STARD10 expression and glucose-stimulated insulin secretion (GSIS) in EndoC-βH1 cells (Hu et al., 2021). These data suggest that STARD10 is likely to be the chief effector gene at this T2D risk locus and that lower STARD10 expression may impair β function.

STARD10 is a member of the steroidogenic acute regulatory protein (StAR)-related lipid transfer (START) domain family and can bind to multiple lipid species, such as phosphatidylcholine (PC), phosphatidylethanolamine (PE) (Olayioye et al., 2005), and phosphoinositide (Carrat et al., 2020). It has been shown that STARD10 binds to these phospholipids to mediate their transfer between membranes (Olayioye et al., 2005). To study the role of STARD10 in pancreatic β cells and its association with diabetes, Carrat and colleagues deleted Stard10 specifically in the β cells of mice and found that these mice had higher fed glycemia at 14 weeks of age. β cell-specific Stard10 KO mouse islets also showed a diminished cytosolic Ca²⁺ influx and insulin secretion in response to high glucose stimulation (Carrat et al., 2017). Similarly, deletion of STARD10 in the human β cell line EndoC-βH1 reduced GSIS (Hu et al., 2021). These findings indicate that STARD10 plays a role in regulating β cell function.

While the above findings implicate altered STARD10 expression in the β cell as a likely mediator of T2D risk, a more detailed understanding of its role in lipid biology and β cell development remain elusive, especially in the context of human β cells. Here, we used human embryonic stem cell (hESC)-derived β-like cells to investigate the role of STARD10. We found that STARD10 is increasingly expressed throughout β cell differentiation and regulates the formation of INS⁺ β-like cells from hESCs. To understand the global effects of losing STARD10 in human β cells, we performed transcriptomics and lipidomics analyses on STARD10-null β-like cells. Our findings indicate that the loss of STARD10 expression impairs β cell proliferation and triglyceride (TG) metabolism. Through Seahorse analyses, we showed that STARD10-null β-like cells had impaired glycolytic and mitochondrial respiratory functions, as well as reduced maximal palmitate oxidation. Overall, these perturbations may lead to reduced β cell mass and function, and ultimately to a higher risk of T2D.

Results

STARD10 expression increases during in vitro differentiation of human pancreatic β cells

To examine in detail the expression profile of STARD10 during the course of human pancreatic β cell development, we differentiated the hESC line MEL1 INS^+/GFP into pancreatic β-like cells (Figure 1A) (Pagliuca et al., 2014). We first evaluated STARD10 expression and found that it increases throughout β cell differentiation, in a manner similar to that observed for PDX1, NKX6.1, and INS (Figure 1B). In contrast, the expression of other STARD genes such as STARD2, STARD7, and STARD11, which are in the same subfamily as STARD10 (Alpy and Tomasetto, 2005), exhibited little to no increase (Figure 1C). Flow cytometry analyses confirmed that STARD10 is co-expressed with the pancreatic progenitor marker PDX1 at D13 and D20, as well as with insulin at D35 (Figure S1A). The population of STARD10⁺ cells at D13 was relatively low (∼5.8%) but this population peaked at ∼47% at D20 before decreasing to ∼26% at D35 (Figures S1B–S1E). Overall, these data indicated the relevance and possible involvement of STARD10 during the differentiation of INS⁺ human pancreatic β-like cells.

Figure 1.

Loss of STARD10 in hESCs impairs the formation of INS⁺ β-like cells

(A) Schematic illustration of the 35-day β cell differentiation protocol used in this study. DE, definitive endoderm; PGT, primitive gut tube; PP1, early pancreatic progenitor; PP2, late pancreatic progenitor; EN, endocrine progenitor.

(B and C) Relative mRNA expression at the four differentiation stages (D0, D13, D20, and D35) measured by qPCR (n = 3–5). (B) PDX1, ^∗; NKX6.1, #; INS, †; STARD10, ‡. Statistical significance for each gene was calculated by comparing against D0. (C) STARD10, ^∗; STARD11, #. Statistical significance for each gene was calculated by comparing against D0. No significant difference for STARD2 and STARD7 at all timepoints.

(D–H) Western blot analysis at D35 of β cell differentiation (one representative experiment out of three). Relative mRNA expression at (E–F) D20 and (G–H) D35 of β cell differentiation measured by qPCR (n = 3–4).

(I and J) Flow cytometry analyses at D35 of β cell differentiation (n = 3). STARD10 KO represents pooled data from #2 and #4. Error bars denote SEM. ^∗p < 0.05; ^∗∗p < 0.01; ^∗∗∗p < 0.001; ^∗∗∗∗p < 0.0001. Statistical analyses were performed using unpaired t test for two groups and one-way ANOVA, followed by Tukey’s multiple comparison test for three groups.

See also Figures S1 and S2.

Loss of STARD10 in hESCs impairs the formation of INS⁺ β-like cells

Next, to study the role of STARD10 during human β cell differentiation, we used CRISPR targeting to knock out STARD10 in MEL1 INS^+/GFP hESCs and successfully generated two STARD10 KO lines (labeled #2 and #4). We generated a Parental control line (P) by transducing the hESCs with the same lentiviral vector containing the Cas9 gene but without any gRNAs. The Parental line went through the same clonal selection process as the two STARD10 KO lines. Sanger sequencing of the genomic DNA verified that line #2 is a compound heterozygous KO and #4 is a homozygous KO for STARD10 (Figure S2A). We differentiated the three STARD10 isogenic hESC lines into pancreatic β-like cells and confirmed that STARD10 protein was eliminated in #2 and #4 D35 cells using western blot analyses (Figure 1D).

To investigate whether the loss of STARD10 may impair human β cell development, we differentiated the above three lines into pancreatic β-like cells. Morphologically, there were no obvious differences among Parental, #2 and #4 cells (Figure S2B). We then evaluated the transcript levels of several key markers throughout the differentiation. We found that KO lines #2 and #4 expressed lower levels of PDX1 and NKX6.1 at the D20 endocrine progenitor stage (Figures 1E and 1F), and lower levels of INS and NKX6.1 at the D35 β-like cell stage (Figures 1G and 1H). Consistently, #2 and #4 also expressed lower GFP (driven by one INS promoter in the MEL1 hESCs) levels in D35 cells (Figure S2C), suggesting decreased transcriptional activation of the INS gene promoter.

To validate these findings, we performed flow cytometry analyses of the key markers above at various developmental stages. At the D13 pancreatic progenitor and D20 endocrine progenitor stage, there was no significant difference in the proportion of PDX1⁺ cells among the three lines, although there was a slight reduction in the STARD10 KO lines in some instances (Figures S2D–S2G). Coupled with the transcript data, these could mean that the loss of STARD10 does not impair the proportion of endocrine progenitor cells but reduces the overall expression of key β cell developmental genes, which may have negative developmental consequences. Importantly, at the D35 β-like cell stage, we found the proportion of INS⁺ cells to be significantly lower in #2 and #4 as compared to Parental (Figures 1I and 1J). Overall, these data suggest that the loss of STARD10 in hESCs impairs human β cell differentiation and significantly impairs the formation of INS⁺ β-like cells.

Deficiency of STARD10 leads to downregulation of cell cycle and cell proliferation genes in hESC-derived β-like cells

To further understand the effects of losing STARD10 in human β-like cells and to investigate the mechanisms of how STARD10 affects β cell development and function, we performed transcriptomics analyses on D20 and D35 cells, when STARD10 is highly expressed (Figures 1B; Tables S1 and S2). UMAP analyses showed that the transcriptomes of STARD10 KO cells (#2 and #4) were distinctly different from those of Parental cells at both D20 and D35 (Figure S3A). Gene ontology analyses demonstrated that developmental processes were most downregulated in D20 STARD10 KO cells (Figure S3B) whereas ion transport and nervous system processes were reciprocally upregulated in the STARD10 KO pancreatic endocrine cells (Figure S3C) and β-like cells (Figure S3D), which may further support our hypothesis that the KO of STARD10 alters the cell fate toward β-like cells.

Strikingly, cell cycle processes were clearly downregulated in the D35 STARD10 KO β-like cells versus Parental cells (Figure 2A). A closer inspection revealed that a total of 148 cell cycle genes were significantly downregulated (FC > 2; FDR < 0.05) in D35 STARD10 KO β-like cells (Figure 2B). We confirmed that MKI67, which encodes the proliferation marker Ki67, and several cyclin genes such as CCNA2, CCNB1, CCNB2, CCNE1, and CCNE2 were significantly downregulated in STARD10 KO β-like cells via qPCR analyses (Figures 2C–2H). Consistent with the transcript findings, we found that the percentage of Ki67⁺ cells was significantly lower in STARD10 KO than Parental β-like cells via flow cytometry analyses (Figures 2I and 2J), suggesting that there could be less proliferative β-like cells when STARD10 is deleted. The specific downregulation of cell cycle genes and processes in D35 but not D20 STARD10 KO cells suggest a unique involvement of STARD10 in regulating cell cycle processes during the late stage of β cell differentiation.

Figure 2.

Deficiency of STARD10 leads to downregulation of cell cycle and cell proliferation genes in β-like cells

(A) Top 10 downregulated biological processes in STARD10 KO (average of #2 and #4) β-like cells as compared to Parental β-like cells.

(B) Heatmap analysis of the 148 mitotic cell cycle genes that were significantly downregulated. Mitotic cell cycle genes that were selected for qPCR validation were highlighted (black box).

(C–H) Relative mRNA levels of cell cycle genes measured by qPCR in D35 β-like cells (n = 3).

(I and J) Flow cytometry analyses at D35 of β cell differentiation (n = 3). STARD10 KO represents pooled data from #2 and #4. Error bars denote SEM. ^∗p < 0.05; ^∗∗p < 0.01; ^∗∗∗p < 0.001. Statistical analyses were performed using unpaired t test for two groups and one-way ANOVA, followed by Tukey’s multiple comparison test for three groups.

See also Figure S3.

Deletion of STARD10 leads to increased intracellular triglyceride accumulation in hESC-derived β-like cells

Given that STARD10 is a lipid transfer protein, we next wanted to explore the role of STARD10 in lipid metabolism and investigate how the loss of STARD10 leads to the observed phenomenon shown above. To this end, we performed lipidomics analyses on STARD10 KO and Parental β-like cells to identify any changes in lipid composition. From these analyses, 22 classes of lipids containing a total of 281 different lipid species were evaluated: cholesteryl esters (CE), ceramides, diglycerides (DG), gangliosides (GM3), hexosylceramides (HexCer), dihexosylceramides (Hex2Cer), trihexosylceramides (Hex3Cer), lysophosphatidylcholines (LPC), ether- and plasmalogen-LPC (LPC O and LPC P) lysophosphatidylethanolamines (LPE), plasmalogen-LPE (LPE-P), lysophosphatidylserines (LPS), PC, PE, ether- and plasmalogen-PE (PE-O and PE-P), phosphatidylglycerols (PG), phosphatidylinositols (PI), phosphatidylserines (PS), sphingomyelins (SM), and triglycerides (TGs) (Table S3). Notably, there were no significant differences for all lipid classes (where one class is the sum of all species) between Parental and STARD10 KO endocrine progenitors (D20) and β-like cells (D35) (Figure S4A), except for total TGs (Figure 3A). Total TG levels, driven by several specific TG species (Figure S4C), were significantly higher in STARD10 KO β-like cells (Figure 3A) but not endocrine progenitors (Figure S4B), further suggesting the role that STARD10 plays in the later stage of human β cell differentiation. Out of the 29 TG species detected, we found four species (TG 16:1 18:1 18:1, TG 18:1 18:1 18:1, TG 52:3, and TG 52:4) to be significantly upregulated in D35 STARD10 KO β-like cells and one species (TG 58:8) to be upregulated in D20 STARD10 KO endocrine progenitor cells (Figure S4C). We were unable to clarify the specific structure of TG 52:3, TG 52:4, and TG 58:8, which could include different combinations of fatty acids. The levels of several individual species belonging to the DG, GM3, PE P, Hex2Cer, Hex3Cer, LPC, PI, PE P, and SM lipid classes were different between Parental and STARD10 KO β-like cells (Figure S4D). Oil Red O (ORO) staining in D35 cells revealed that STARD10 KO β-like cells had significantly higher levels of intracellular TGs than Parental β-like cells (Figures 3B and 3C). To further confirm that the lipid accumulation occurs in β-like cells, we performed immunofluorescence analysis of the INS-GFP signals with LipidSpot610 staining (which stains lipid droplets) on D35 cells. LipidSpot610 staining partially overlapped with INS-GFP signals and its intensity levels were modestly higher in STARD10 KO lines (Figures 3D and 3E), suggesting that the lipid levels were increased in STARD10 KO β-like cells. Overall, we found that STARD10 KO could affect the quantity of unsaturated TG species during human pancreatic β cell differentiation.

Figure 3.

Deletion of STARD10 leads to increased triglyceride accumulation and impaired cellular bioenergetics in β-like cells

(A) Lipidomics analyses of total TGs on D35 of β cell differentiation (n = 3).

(B) Oil red O staining of β-like cells (one representative experiment). Scale bars, 50 μm.

(C) Quantification of oil red O staining in Parental and STARD10 KO (#2 and #4 combined) β-like cells (n = 3).

(D) LipidSpot610 staining of D35 cryosections (one representative experiment). Scale bars, 100 μm.

(E) Mean fluorescence intensity quantification of LipidSpot610 staining between Parental and STARD10 KO (#2 and #4 combined) cells (n = 3).

(F) Extracellular acidification rate (ECAR) measurement of D35 β-like cells via Seahorse XF Glycolysis Stress Test Kit (n = 5–6).

(G) Oxygen consumption rate (OCR) measurement of D35 β-like cells using the Seahorse XF Cell Mito Stress Test Kit (n = 5–6).

(H) OCR measurement of D35 β-like cells treated with palmitate using the Seahorse XF Palmitate Oxidation Stress Test Kit (n = 4–5). Error bars denote SEM. ^∗p < 0.05; ^∗∗p < 0.01. Statistical analyses were performed using unpaired t test for two groups and one-way ANOVA, followed by Tukey’s multiple comparison test for three groups.

See also Figure S4.

Deletion of STARD10 impairs glycolysis, mitochondrial respiration, and fatty acid oxidation in hESC-derived β-like cells

As the increase in lipid accumulation in STARD10 KO β-like cells may suggest perturbations in cellular bioenergetics, we next wondered if STARD10 KO β-like cells have altered glycolysis and mitochondrial respiration activity, especially in the presence of fatty acids. Hence, we performed Seahorse assays, and found that STARD10 KO β-like cells have impaired glycolytic function and mitochondrial respiration (Figures 3F and 3G). When we performed the mitochondrial oxidative phosphorylation (oxphos) stress test with palmitate, to our surprise, we found that STARD10 KO β-like cells have similar oxygen consumption rate (OCR) as Parental control cells at baseline (time = 0 min) (Figure 3H), unlike the mitochondrial oxphos stress test without palmitate (Figure 3G). These data suggested that STARD10 KO β-like cells can still oxidize palmitate to achieve normal OCR at baseline conditions. However, the treatment of FCCP (which pushes cells to maximum respiratory capacity), did not increase the OCR in STARD10 KO β-like cells unlike the Parental control (Figure 3H), suggesting that STARD10 KO β-like cells have reduced maximal palmitate oxidation capacity. Overall, our data suggest that the deletion of STARD10 impaired the ability of hESC-derived β-like cells to perform glycolysis, mitochondrial respiration, and fatty acid oxidation.

STARD10 regulates INS expression, proliferation, and lipid accumulation via ETFB

Because TG levels were significantly higher in STARD10 KO β-like cells, we looked at genes that are involved in lipid metabolism in our RNA-Seq data and found that ETFB expression was almost abolished in STARD10 KO β-like cells (Tables S1 and S2). qPCR (Figure 4A) and western blot analyses (Figure 4B) confirmed that STARD10 KO β-like cells had very little to no expression of ETFB. ETFB encodes for the electron transport flavoprotein β subunit, which is required for β-oxidation of fatty acids (Watmough and Frerman, 2010). Interestingly, our timecourse transcript analyses revealed that ETFB expression levels increase as hESCs are differentiated from the embryonic stage to the endocrine progenitor and β-like cell stage (Figure 4C), suggesting that human β-like cells may acquire the capacity to utilize fatty acids as an energy source as they mature. The abolishment of ETFB expression in STARD10 KO β-like cells may explain the impairment in fatty acid β-oxidation, which is consistent with the increased accumulation of TGs.

Figure 4.

STARD10 regulates INS and Ki67 expression and lipid accumulation via ETFB

(A) Relative gene expression analysis in D35 β-like cells (n = 4).

(B) Western blot analysis of D35 β-like cells (one representative experiment out of three).

(C) mRNA time course analyses throughout β cell differentiation (n = 3).

(D) Immunofluorescence analysis of ETFB and INS in D35 cells transduced with empty vector (EV) or ETFB-overexpressing vector (one representative experiment). Scale bars, 100 μm. (E–F) Mean fluorescence intensity quantification of ETFB and INS staining (n = 3).

(G) Immunofluorescence analysis of Ki67 in D35 cells transduced with empty vector (EV) or ETFB-overexpressing vector (one representative experiment). Scale bars, 100 μm.

(H) Mean fluorescence intensity quantification of Ki67 staining (n = 3). (I) LipidSpot610 staining of D35 cells transduced with empty vector (EV) or ETFB-overexpressing vector (one representative experiment). Scale bars, 100 μm. (J) Mean fluorescence intensity quantification of LipidSpot610 staining (n = 3). Error bars denote SEM. ^∗p < 0.05; ^∗∗p < 0.01; ^∗∗∗p < 0.001; ^∗∗∗∗p < 0.0001. Statistical analyses were performed using unpaired t test for two groups and one-way ANOVA, followed by Tukey’s multiple comparison test for three groups.

Next, to show that ETFB is a downstream mediator of STARD10, we performed ETFB overexpression rescue experiments on STARD10 KO #4 D20 endocrine progenitors as STARD10 is most widely expressed at that stage of differentiation. We evaluated the cells at D35 of the differentiation and confirmed that ETFB was successfully overexpressed in #4 cells (Figures 4D and 4E). The overexpression of ETFB in #4 D20 cells rescued INS expression (Figures 4D and 4F) and proliferation as marked by the increase in percentage of Ki67⁺ cells (Figures 4G and 4H) in D35 cells. Of note, STARD10 KO #4 cells did not survive well after cell dissociation as compared to Parental cells, accounting for lesser DAPI⁺ cells (Figure 4G). Expectedly, the overexpression of ETFB in #4 D20 cells also decreased lipid accumulation in D35 cells (Figures 4I and 4J), consistent with its essential role in fatty acid β-oxidation. Overall, these suggest that STARD10 regulates INS expression, proliferation and lipid accumulation β-like cells via ETFB.

Discussion

Previous studies have demonstrated the causal role of Stard10 in T2D through the modulation of insulin secretion in rodent β cells (Carrat et al., 2017, 2020; Hu et al., 2021) but the mechanisms relating to human β cell development remain elusive. Here, we found that the KO of STARD10 in hESCs affects β cell development, as well as proliferation at the late stage of β cell differentiation. STARD10 also regulates the expression of the fatty acid oxidation enzyme ETFB and triglyceride metabolism. Consequently, STARD10 KO β-like cells have impaired glycolysis and mitochondrial respiration, which may contribute to insulin secretion dysfunction.

In this study, we have discovered the role of STARD10 in human β cell development: (1) STARD10, but not the other members of the same subfamily, is increasingly expressed throughout β cell differentiation; (2) deletion of STARD10 in hESCs reduced the transcript expression of several key pancreatic (PDX1) and β cell markers (NKX6.1 and INS) at the endocrine progenitor and β-like cell stage; and (3) impaired the formation of INS⁺ β-like cells. Interestingly, although Stard10 KO mice have poorer insulin secretory function, no reduction in β cell mass was reported (Carrat et al., 2017), indicating a potential difference between human and mouse STARD10. This may be explained by the differences in lipid metabolism between human and mouse islets. Human islets accumulate lipid droplets (LDs) much more readily than mouse islets when transplanted into NOD SCID mice (Dai et al., 2016). Similarly, human islets from aged donors often contain LDs but mouse islets from mice as old as 1.5 year old barely contain any LDs (Tong et al., 2020). Furthermore, human islets are shown to be less sensitive to high-fat diet (HFD)-induced stress than mouse islets (Dai et al., 2016). Further investigations are required to determine if STARD10 plays a different role in human versus mouse β cells.

To further study the effects of depleting STARD10 in β-like cells, RNA-Seq analysis was performed, revealing that STARD10 KO β-like cells were less proliferative as supported by lower expression of cell cycle genes and fewer Ki67⁺ cells. In support of our findings, the role of STARD10 in regulating cell proliferation has been reported. STARD10 was found to be upregulated in breast cancer tissues as compared to normal tissues (Floris et al., 2019; Olayioye et al., 2004). Mechanistically, the overexpression of STARD10 increased cell growth and migration by increasing calcium influx and activating the ERBB pathway (Floris et al., 2019), both of which have been shown to activate the ERK signaling pathway (Schmitt et al., 2004; Sirkisoon et al., 2016) which is a key regulator of the cell cycle. While no lipid-protein binding assays were performed in this study, STARD10 has been shown to bind to PC (Olayioye et al., 2005) and PI (Carrat et al., 2020); both have well-documented roles in regulating the ERK signaling pathway (Anderson, 2006; Payrastre et al., 2001). Thus, it is possible that STARD10 regulates cell cycle and cell proliferation by modulating the ERK signaling pathway via PC and/or PI.

To investigate the effects of STARD10 depletion in the lipid profile of hESC-derived β-like cells, lipidomic analyses were performed. STARD10 KO β-like cells were found to have higher levels of TGs, which was also confirmed by ORO and LipidSpot610 staining. In support of our data, the KO of Stard2, which is a PC-binding protein and in the same STARD domain subfamily as STARD10, resulted in TG accumulation in mouse hepatocytes (Perez et al., 2020). Crystal structure of STARD2 (also known as PC-TP) has revealed that STARD2 binds to a single well-ordered molecule of PC via a central tunnel formed by a curved β-sheet, α2, α3, α4, and loop Ω1(Roderick et al., 2002), which makes it unlikely to bind to TGs directly. Given that STARD2 and STARD10 are structurally similar and STARD10 also binds to PC (Olayioye et al., 2005), it is conceivable that the loss of STARD10 in hESC-derived β-like cells induces a similar effect, and that it is unlikely that STARD10 binds to TGs directly to regulate their metabolism.

Interestingly, we found that the reason for increased TG accumulation in STARD10 KO β-like cells is likely due to the loss of ETFB expression. ETFB, together with ETFA, forms the ETF complex in the mitochondrial matrix which serves as an electron acceptor for acyl-CoA dehydrogenase; a process that is essential for fatty acid β-oxidation (Watmough and Frerman, 2010). Importantly, deleterious mutations in the ETFB gene results in a disease called multiple acyl CoA dehydrogenase deficiency (MADD), where patients have impaired fatty acid β-oxidation (van Rijt et al., 2019), leading to pancreatitis in a subset of them (Elkhateeb et al., 2021).

It is well-documented that chronic exposure of β cells to fatty acids can impair GSIS and induce apoptosis (Bensellam et al., 2012; Doliba et al., 2017; Poitout and Robertson, 2008). Increased TG accumulation in rat islets by overexpressing Dgat1 has also been shown to impair GSIS (Kelpe et al., 2002). Moreover, chronic exposure to fatty acids or inhibiting fatty acid β-oxidation has been shown to reduce Ins mRNA levels in rat islets (Briaud et al., 2001). The increase in TG accumulation may help to further explain how the loss of STARD10 leads to reduced INS expression, fewer INS⁺ cells, and impaired GSIS (Hu et al., 2021) in human β cells. However, lipidomics analyses performed in mouse β cells from β cell-specific Stard10 KO mice did not show differences in TG levels (Carrat et al., 2020). Again, these data suggest that there could be differences between human and mouse β cells. However, it is also possible that the deletion of STARD10 in early development (prior to insulin gene expression, as achieved here) exerts different effects than when deleted in the adult β cell (Carrat et al., 2020).

It has been established in previous studies that the loss of STARD10 impairs GSIS in rodent (Carrat et al., 2017, 2020) and human (Hu et al., 2021) β cells but the mechanisms remain unclear. Here, we show via Seahorse assays that hESC-derived β-like cells deficient of STARD10 have impaired glycolysis and mitochondrial respiration—both processes are important for ATP generation and GSIS in β cells. Reduced glycolysis and mitochondrial respiration in STARD10 KO β-like cells could also explain for the reduced proliferation, a process that is highly energy demanding. Interestingly, when treated with saturating levels of palmitate, STARD10 KO β-like cells could still oxidize palmitate to achieve normal OCR at baseline conditions despite having severely reduced ETFB expression, suggesting that the reduced levels of ETFB could still cope with the palmitate load. However, when energy demand is increased maximally using FCCP, STARD10 KO β-like cells could not ramp up palmitate oxidation as seen in Parental control β-like cells. This may suggest that the effects of losing STARD10 are more pronounced in increased energy demand conditions, which is reminiscent of T2D conditions where obesity and insulin resistance increase insulin secretory demand of β cells. Such phenotype is expected of a T2D risk-associated gene, where disease phenotype is only prominent under metabolic stress conditions.

It is noteworthy that the β-like cells generated in this study were proliferative (with high proportion of Ki67⁺ cells) and did not exhibit glucose-responsiveness at 2.8/16.7 mM glucose (data not shown), demonstrating immaturity. However, these β-like cells exhibit glucose-responsiveness at 0.5/16.7 mM glucose (data not shown). While β-like cells generated using older differentiation protocols, such as the one used in this study (Pagliuca et al., 2014), are known to be immature and typically do not have appropriate insulin secretion in response to glucose without in vivo maturation (Nair et al., 2019), they still represent as a good disease model as shown in various studies (Lau et al., 2023; Low et al., 2021). Nevertheless, the findings in this study should be interpreted with caution given the immaturity of the β-like cells, and further studies using more advanced differentiation protocols that can yield more mature β-like cells should be carried out. Another limitation of this study includes the possibility of non-β-like cells present in the differentiation cultures contributing to the observed phenotypes. This can include other endocrine cell types, exocrine, enterochromaffin, and undifferentiated progenitor cells (Cuesta-Gomez et al., 2022). Unfortunately, the INS-GFP signal in our MEL1 INS^+/GFP lines, although detectable by confocal microscopy, could not be detected by our in-house FACS machine for sorting. Hence, β-like cells could not be purified for analysis. Nonetheless, INS⁺ β-like cells represent around half of the population of our differentiation cultures and phenotypes such as triglyceride accumulation were indeed observed in INS⁺ β-like cells in our study.

In summary, we have provided further insights into the role of STARD10 during human β cell differentiation and provided additional evidence especially in a human cellular context to support STARD10 as the effector transcript underlying the T2D association at this locus. We found STARD10 deletion to affect human β cell formation, cellular bioenergetics, and fatty acid oxidation. Together, these changes seem likely to contribute to poorer islet insulin secretory function and an increased propensity for the development of T2D in people carrying STARD10 risk alleles.

Methods

Cell culture

The hESC line MEL1 INS^+/GFP with a GFP allele knocked into one of the INS locus was a gift from Dr. Edouard Stanley (Micallef et al., 2012). The line was maintained in TeSR-E8 on plates pre-coated with 0.1% gelatin followed by DMEM high glucose supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 1% MEM non-essential amino acids (NEAA). Media were changed daily and cells were passaged weekly either by handpicking or using ReLeSR. All cells were incubated in a humidified incubator at 37°C with 5% CO₂ and were validated to be mycoplasma-free using the MycoAlert PLUS Mycoplasma Detection Kit.

Differentiation of hESCs into β-like cells

hESCs were differentiated to β-like cells following a previously published protocol (Pagliuca et al., 2014) with slight modifications. hESCs were dissociated into single cells using TrypLE Express and 3–4 million cells were seeded at a density of 1 million cells/ml in mTeSR1 with 10 μM of Y-27632 into each well of a non-treated 6-well plate. Cells were incubated on a shaker at 80 rpm in a humidified incubator at 37°C with 5% CO₂. The next day, differentiation was initiated. Cells were differentiated into the definitive endoderm stage in S1 media (MCDB131, with 8 mM glucose, 29.3 mM sodium bicarbonate, 2% fatty acid-free BSA, 0.25 mM L-Ascorbic acid, 0.002% Insulin-Transferrin-Selenium-Ethanolamine [ITS-X], 1% Glutamax, and 1% pen/strep), with 100 ng/mL Activin A and 3 μM of the glycogen synthase kinase 3 inhibitor CHIR 99021 for the first day, and with 100 ng/mL Activin A alone for the next 2 days. This was followed by differentiation into the primitive gut tube stage, which was done in S2 media (MCDB131 with 8 mM glucose, 14.6 mM sodium bicarbonate, 2% fatty acid-free BSA, 0.25 mM L-Ascorbic acid, 0.002% ITS-X, 1% glutamax, and 1% pen/strep), with 50 ng/mL of fibroblast growth factor 7 (FGF7) for 3 days. Then, cells were differentiated into the first pancreatic progenitor stage in S3 media (S2 media but with 0.5% ITS-X instead), with the addition of 50 ng/mL FGF7, 2 μM retinoic acid (RA), 0.25 μM of SANT1, 500 nM phorbol 12,13-dibutyrate (PDBu), and 200 nM of LDN193189 for one day, followed by 50 ng/mL FGF7, 2 μM RA, 0.25 μM of SANT1, and 500 nM PDBu the following day. Then, the cells were differentiated into the second pancreatic progenitor stage in S3 media with 50 ng/mL FGF7, 100 nM RA, and 0.25 μM of SANT1 for 5 days. Next, the cells were differentiated into endocrine progenitors in S5 media (MCDB131, with 20 mM glucose, 20.9 mM sodium bicarbonate, 2% fatty acid-free BSA, 0.25 mM L-Ascorbic acid, 0.5% ITS-X, 1% glutamax, 1% pen/strep and 10 μg/mL of heparin), with 100 nM of RA, 0.25 μM of SANT1, 1 μM of the γ-secretase inhibitor XXI, 10 μM of Alk5 inhibitor II (Alk5iII), 1 μM of L-3,3′,5-triiodothyronine (T3), and 20 ng/mL of human betacellulin (BTC) for 4 days, followed by S5 media with 25 nM of RA, 1 μM of XXI, 10 μM of Alk5iII, 1 μM of T3, and 20 ng/mL of BTC for another 3 days. Subsequently, the cells were differentiated into β-like cells in S6 media (CMRL 1066 with 10% FBS and 1% pen/strep) with 10 μM of Alk5iII and 1 μM of T3 for at least 2 weeks. Media were changed every other day if the same type of media was required.

ETFB overexpression rescue

The human ETFB full-length cDNA was cloned into the pCDH overexpressing vector and packaged into lentiviruses by transfecting into 293FT cells with pRC-CMV-Rev, pHDM-HIVgpm and pHDM-VSV-G using the CalPhos transfection mix. Virus-containing media were collected at 48 h and 72 h after transfection. After centrifugation and filtration through the 0.4 μm filter, lentiviruses were pelleted using ultracentrifugation at 23,000 rpm for 90 min. The concentration of lentiviruses was quantified using the Lenti-X p24 Rapid Titer Kit. At D19 of the differentiation, spheroids were dissociated using TrypLE Express for 5 min, and 1 million cells were seeded onto each well of a 12-well plate as monolayers. At D20 of the differentiation, cells were transduced with lentiviruses at an MOI of 50 with 8 μg/mL polybrene. Thereafter, the cells were subjected to the same differentiation protocol as monolayers and harvested at D35 for immunofluorescence analyses.

RNA-sequencing analysis

RNA-Seq was performed by the Molecular Engineering Lab. Poly-A mRNA was enriched from 100 ng of total RNA using oligo-dT beads and used to construct multiplexed strand-specific RNA-Seq libraries as per manufacturer’s instructions. The library quality was assessed using an Agilent 2100 Bioanalyzer, quantified using a QuBit 2.0 fluorometer, pooled for sequencing on a HiSeq 2000 (1 x 101 bp read), and quantified using the KAPA quantification kit. Trimmomatic (v.0.33) was used to trim adapter sequences and low quality bases in Fastq read sequences (parameters: LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36), and the filtered Fastq sequence reads were aligned to the human genome (hg19) using Tophat (v.2.0.14) (parameters:-no-coverage-search –library-type = fr-firststrand) and annotated with Ensembl gene IDs to obtain bam files. Feature read counts were then generated using these files and the Python package-based htseq-count of HTSeq (v.0.6.1p1) (parameters: default union-counting mode, -stranded = reverse). The read count matrix output from HTSeq was then used to perform differential expression analysis using the edgeR package (available in R [v.3.1.3]) in both “classic” and generalized linear model (glm) modes to contrast WT vs. KO. Procedures described in edgeR documentation were followed to calculate p values, FDR adjusted p values (q values) and fold-changes. A threshold of q value <0.05 and fold change ≥2 was used to define differentially expressed genes (DEGs). DEGs were analyzed for functional enrichment using DAVID (Sherman et al., 2022).

Lipidomics analysis

Pancreatic β-like cells were washed five times with PBS, before flash-freezing in liquid nitrogen. Samples were analyzed by the Singapore Lipidomics Incubator (SLING). Samples were treated with butanol/methanol (1:1, v/v), containing internal standards (Avanti Polar Lipids) to extract the lipids. The mixture was vortexed for 30 s, sonicated for 30 min at 20°C, and then centrifuged at 14,000 g for 10 min. The supernatant was used to measure lipid content using an Agilent 1290 Infinity II LC system combined with an Agilent 6495A triple quadrupole mass spectrometer. An Agilent Zorbax RRHD Eclipse Plus C19, 95 Å (50 × 2.1 mm, 1.8 μm) column maintained at 40°C was used to perform reverse phase chromatographic separation of 1 μL samples. The mobile phases consisted of (A) 10 mmol/L ammonium formate in acetonitrile/water (40:60, v/v) and (B) 10 mmol/L ammonium formate in acetonitrile/2-propanol (10:90, v/v). Using a flow rate of 0.4 mL/min, the gradient elution program started at 20% B, before being increased to 60% B at 2 min, and 100% B at 7 min. After which, it was maintained till 9 min, then re-equilibrated at 20% B for 1.8 min prior to the next injection. All samples were kept at 10°C in the autosampler. Lipid amounts were normalized to protein content. The Agilent MassHunter Quantitative Analysis software (version B.08.00) was used to process the MRM chromatograms obtained in positive ion mode. Peaks were annotated based on retention time and specific MRM transitions.

Seahorse assay

Extracellular acidification rate (ECAR) and oxygen consumption rate (OCR) were measured using the Seahorse XFe 96 Analyzer (Agilent Technologies). At D35 of β cell differentiation, cells were dissociated using TrypLE Express for 5 min and filtered through a 40 μm filter. 100,000 cells were seeded on each well of a Seahorse XFe 96 well plate, precoated with DMEM High Glucose +2 μg/mL of fibronectin and 1% of ECM for at least a day. The following day, the Seahorse XF Glycolysis Stress Test, Seahorse XF Mito Stress Test, or Seahorse XF Palmitate Oxidation Stress Test were carried out using kits from Agilent Technologies. After the Seahorse assay, cells were lysed for protein normalization.

Statistical analyses

Data were presented as mean ± standard error of the mean (SEM). Statistical analyses performed include unpaired t test for two groups analysis, one-way ANOVA followed by Tukey’s multiple comparisons test for three or more groups analysis. Each dot in a graph represents one independent experiment, and the number of independent experiments (n) are indicated in figure legends. All graphs and statistical analyses were done using Graphpad PRISM 7.

Resource availability

Lead contact

Further information and requests for resources and regents should be directed to and will be fulfilled by the lead contact, Adrian Kee Keong Teo (adrian_teo@a-star.edu.sg).

Materials availability

All unique/stable reagents generated in this study are available from the lead contact without restriction.

Data and code availability

RNA-Seq data are uploaded onto GEO under record GSE153466. The lipidomics raw data and acquisition methods are deposited to MetaboLights open data repository under study identifier MTBLS13689.

Acknowledgments

The authors thank members of the Teo laboratory for the critical reading of this manuscript. W.X.T., N.S.A., and E.X.H.L. were supported by the National University of Singapore (NUS) Research Scholarship (RS). W.X.T. is also supported by NMRC OFYIRG24jan-0053. S.M., F.T., and P.I.B. are supported via the Life Sciences Institute (LSI) at NUS, National Research Foundation (NRFI2015-05, NRFSBP-P4) and A^∗STAR (I1901E0040). A.A.P.S. was supported by the NUS Integrative Sciences and Engineering Program Scholarship. N.H.J.N. was supported by NMRC OFYIRG18may040 and the first A^∗STAR Career Development Fund 2019. H.H.L. was supported by the Institute of Molecular and Cell Biology (IMCB) Scientific Staff Development Award (SSDA) for her part-time Ph.D. G.A.R. and A.K.K.T. are supported by an MRC Program grant MR/R022259/1. G.A.R. was also supported by a Wellcome Trust Investigator Award (212625/Z/18/Z), Diabetes UK (BDA/15/0005275, BDA 16/0005485) grants, a start-up grant from the CRCHUM, Université de Montréal, and a John R. Evans Leader Award from Innovation Canada. This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking, under grant agreement no. 115881 (RHAPSODY). This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation program and EFPIA. A.K.K.T. is also supported by IMCB, A^∗STAR, FY2019 SingHealth Duke-NUS Surgery Academic Clinical Program Research Support Program Grant, Precision Medicine and Personalised Therapeutics Joint Research Grant 2019, the 2nd A^∗STAR-AMED Joint Grant Call 192B9002, HLTRP/2022/NUS-IMCB-02, Paris-NUS grant 2021-06-R/UP-NUS (ANR-18-IDEX-0001), OFIRG21jun-0097, CSASI21jun-0006, MTCIRG21-0071, SDDC/FY2021/EX/93-A147, HLCA23Feb-0031, SC36/19-000801-A044, H24G1a0015, M24N2K0087, M25P1a0054, and SFS_RND_2_FS_0005.

Author contributions

Conceptualization, W.X.T., N.S.A., N.H.J.N., G.A.R., and A.K.K.T.; methodology, W.X.T., N.S.A., M.H., N.H.J.N., G.A.R., and A.K.K.T.; software, S.M., P.I.B., F.T., and S.H.; formal analysis, W.X.T., N.S.A., S.M., P.I.B., F.T., S.H., and A.K.K.T.; investigation, W.X.T., N.S.A., E.X.H.L., A.A.P.S., H.H.L., and N.H.J.N.; resources, N.H.J.N., G.A.R., and A.K.K.T.; writing – original draft, W.X.T., N.S.A., and A.K.K.T.; writing – review and editing, W.X.T., N.S.A., E.X.H.L., M.H., S.M., P.I.B., F.T., S.H., A.A.P.S., H.H.L., N.H.J.N., G.A.R., and A.K.K.T.; supervision, N.H.J.N., G.A.R., and A.K.K.T.; project administration, N.H.J.N., G.A.R., and A.K.K.T.; funding acquisition, N.H.J.N., G.A.R., and A.K.K.T.

Declaration of interests

N.H.J.N. and A.K.K.T. are co-founders and shareholders of BetaLife Pte Ltd. N.H.J.N., but not A.K.K.T., is employed by BetaLife Pte Ltd. G.A.R. has received grant funding from, and is a consultant for, Sun Pharmaceuticals Inc. All other authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.

Published: February 12, 2026