The type 2 diabetes gene product STARD10 is a phosphoinositide-binding protein that controls insulin secretory granule biogenesis

Objective Risk alleles for type 2 diabetes at the STARD10 locus are associated with lowered STARD10 expression in the β-cell, impaired glucose-induced insulin secretion, and decreased circulating proinsulin:insulin ratios. Although likely to serve as a mediator of intracellular lipid transfer, the identity of the transported lipids and thus the pathways through which STARD10 regulates β-cell function are not understood. The aim of this study was to identify the lipids transported and affected by STARD10 in the β-cell and the role of the protein in controlling proinsulin processing and insulin granule biogenesis and maturation. Methods We used isolated islets from mice deleted selectively in the β-cell for Stard10 (βStard10KO) and performed electron microscopy, pulse-chase, RNA sequencing, and lipidomic analyses. Proteomic analysis of STARD10 binding partners was executed in the INS1 (832/13) cell line. X-ray crystallography followed by molecular docking and lipid overlay assay was performed on purified STARD10 protein. Results βStard10KO islets had a sharply altered dense core granule appearance, with a dramatic increase in the number of “rod-like” dense cores. Correspondingly, basal secretion of proinsulin was increased versus wild-type islets. The solution of the crystal structure of STARD10 to 2.3 Å resolution revealed a binding pocket capable of accommodating polyphosphoinositides, and STARD10 was shown to bind to inositides phosphorylated at the 3’ position. Lipidomic analysis of βStard10KO islets demonstrated changes in phosphatidylinositol levels, and the inositol lipid kinase PIP4K2C was identified as a STARD10 binding partner. Also consistent with roles for STARD10 in phosphoinositide signalling, the phosphoinositide-binding proteins Pirt and Synaptotagmin 1 were amongst the differentially expressed genes in βStard10KO islets. Conclusion Our data indicate that STARD10 binds to, and may transport, phosphatidylinositides, influencing membrane lipid composition, insulin granule biosynthesis, and insulin processing.


INTRODUCTION
Diabetes mellitus is characterised by high blood glucose and currently affects around 8.5% of the population worldwide. Normal glucose homeostasis requires the processing of proinsulin and the storage of the mature hormone within dense core granules in the pancreatic bcell [1]. Glucose-induced insulin secretion involves glucose uptake and metabolism through the glycolytic pathway, increased ATP production by the mitochondria, and the closure of ATP-sensitive K þ channels (K ATP ). Subsequent depolarisation of the plasma membrane leads to the opening of voltage-gated Ca 2þ channels, Ca 2þ -dependent assembly of the SNARE (SNAP (Soluble NSF Attachment Protein) Receptor) complex, and exocytosis [2]. In the intact islet, b-celleb-cell connections allow coordinated insulin secretion through the propagation of Ca 2þ and other signals [3e5], a process impaired by glucolipotoxicity [6] and affected by genes implicated in diabetes risk through genome-wide association studies (GWAS) such as ADCY5 [7] and TCF7L2 [8].
We have recently examined a T2D-associated locus adjacent to STARD10 on chromosome 11q13 [9,10]. Risk variants at this locus were associated with a decrease in STARD10 mRNA in human islets, with no concomitant change in the liver. Changes in the expression of the nearby ARAP1 gene were not associated with the possession of risk alleles in either tissue, pointing to STARD10 as the mediator of the effects of risk variants. Providing further compelling evidence for STARD10 as an "effector" gene, mice deleted for Stard10 specifically in the b-cell recapitulated the features observed in the human carriers of the risk allele, with an increase in fed glycemia and a decrease in the plasma proinsulin:insulin ratio. Islets isolated from the knockout mice also showed impaired glucose-induced Ca 2þ signalling and insulin secretion. Thus, b-cell STARD10 may be a useful therapeutic target in some forms of type 2 diabetes, particularly in risk allele carriers who may benefit from a tailored, pharmacogenetic approach. STARD10 (previously named phosphatidylcholine transfer protein-like, Pctp-l) is a phospholipid transfer protein possessing a steroidogenic acute regulatory protein-(StAR-) related lipid transfer (START) domain that facilitates the transport of phosphatidylcholine and phosphatidylethanolamine between intracellular membranes [11]. Nevertheless, the molecular mechanisms by which STARD10 regulates insulin secretion in the b-cell, as well as its subcellular localisation and target membranes, remain unknown. We, therefore, examined in detail here the role of STARD10 in controlling the lipid composition, granule maturation, proinsulin processing, and metal ion homeostasis in the mouse b-cell. We reveal an unexpected role for STARD10 in binding inositol phospholipids which may contribute to both secretory granule biogenesis and intracellular signalling.

MATERIAL AND METHODS
2.1. Generation and use of Stard10 null mice All animal procedures were approved by the UK Home Office according to the Animals (Scientific Procedures) Act 1986 of the United Kingdom (PPL PA03F7F0F to I. L.). Stard10 whole body and conditional KO mice (C57BL/6NTac background) were generated by the trans-NIH Knockout Mouse Project (KOMP) and obtained from the KOMP Repository via the International Mouse Phenotyping Consortium (IMPC). Mice homozygous for floxed Stard10 (Stard10 tm1c(KOMP)Wtsi , i.e., Stard10 ;fl/fl ) alleles were crossed to mice expressing Cre recombinase from the endogenous Ins1 locus (Ins1-Cre mice). This generated Stard10 fl/fl :Ins1Cre þ (bStard10KO) mice as in [9].

Islet isolation and culture
Mice were euthanized by cervical dislocation and pancreatic islets isolated by collagenase digestion as previously described [12] and cultured in RPMI 1640 medium, 11 mM glucose, supplemented with 10% (v/v) fetal bovine serum plus penicillin (100 units/mL), and streptomycin (0.1 mg/mL) at 37 C in an atmosphere of humidified air (95%) and CO 2 (5%).

Transmission electron microscopy (EM) imaging
For conventional EM, islets were chemically fixed in 2% paraformaldehyde (EM grade), 2% glutaraldehyde, and 3 mM CaCl 2 in 0.1 M cacodylate buffer for 2 h at room temperature, then left overnight at 4 C in a fresh fixative solution, osmicated, enrobed in agarose plugs, dehydrated in ethanol, and embedded in Epon. Epon was polymerised overnight at 60 C. Ultrathin 70 nm sections were cut with a diamond knife (DiATOME) in a Leica Ultracut UCT ultramicrotome before the examination on an FEI Tecnai G2 Spirit TEM. Images were acquired in a charge-coupled device camera (Eagle) and processed using ImageJ. ] using the eCALWY4 sensor was carried out on mouse islets dispersed onto coverslips as previously described [13]. Cells were maintained at 37 C and Krebs-HEPES-bicarbonate buffer (11 mM) was perifused with additions as stated in the figure. Images were captured at 433 nm monochromatic excitation wavelength. The acquisition rate was 20 images/min. Image analysis was performed with ImageJ software [14] using a homemade macro, and the fluorescence emission ratios were derived after subtracting the background. We observed that during acquisition, photobleaching gradually decreased the steady-state ratio with a linear kinetic (not shown All spectra were acquired using an Orbitrap Fusion Tribrid Mass Spectrometer controlled by Xcalibur 3.0 software (Thermo Scientific) and operated in data-dependent acquisition mode using an SPS-MS3 workflow. FTMS1 spectra were collected at a resolution of 120 000, with an automatic gain control (AGC) target of 400 000 and max injection time of 100 ms. Precursors were filtered with an intensity range from 5000 to 1E20, according to charge state (to include charge states 2e6) and with monoisotopic peak determination set to the peptide. Previously interrogated precursors were excluded using a dynamic window (60 s AE 10 ppm). The MS2 precursors were isolated with a quadrupole isolation window of 1.2m/z. ITMS2 spectra were collected with an AGC target of 10 000, max injection time of 70 ms, and CID collision energy of 35%. For FTMS3 analysis, the Orbitrap was operated at a resolution of 30 000 with an AGC target of 50 000 and a max injection time of 105 ms. Precursors were fragmented by high-energy collision dissociation (HCD) at a normalised collision energy of 55% to ensure maximal TMT reporter ion yield. Synchronous Precursor Selection (SPS) was enabled to include up to 5 MS2 fragment ions in the FTMS3 scan.

Data analysis
The raw data files were processed and quantified using Proteome Discoverer software v2.1 (Thermo Scientific) and searched against the UniProt Rat database (downloaded in January 2019; 35759 entries) using the SEQUEST algorithm. Peptide precursor mass tolerance was set at 10 ppm, and MS/MS tolerance was set at 0.6 Da. Search criteria included oxidation of methionine (þ15.9949) as a variable modification and carbamidomethylation of cysteine (þ57.0214) and the addition of the TMT mass tag (þ229.163) to peptide N-termini and lysine as fixed modifications. Searches were performed with full tryptic digestion and a maximum of 2 missed cleavages were allowed. The reverse database search option was enabled and all data were filtered to satisfy false discovery rate (FDR) of 5%.

Lipid overlay assay
All incubation steps were performed at room temperature. PIP strips (Thermo Scientific) were blocked for 1 h in TBS containing 0.1% Tween-20 (TBS-T) supplemented by 3% fatty-acid-free BSA (Sigma Aldrich) before incubation with the purified STARD10 protein (1 mg/mL in TBS-T þ 3% BSA) for 1 h. The membrane was washed 5 times for 10 min in TBS-T and probed for 1 h with the polyclonal goat anti-STARD10 antibody (Santa Cruz sc54336; 1/1000 in TBS-T þ 3% BSA). After 5 washes for 10 min in TBS-T, the membrane was incubated for 1 h with the horseradish peroxidase-conjugated donkey antigoat IgG (Santa Cruz sc2020; 1/2000 in TBS-T þ 3%BSA). After 5 washes for 10 min in TBS-T, bound proteins were detected by ECL reagent (GE Healthcare).

Structure solution of STARD10
2.11.1. Cloning The full-length gene encoding human STARD10 protein was amplified from cDNA by PCR and cloned into a modified pET28a expression vector, pET28-HMT, which contains a fused N-terminal 6 Â His-tag, an MBP-tag, and a TEV protease recognition site (His-MBP-TEV) by In-fusionÒ HD Cloning kit (Takara Bio, USA). The fidelity of the constructs was confirmed by gel electrophoresis and sequencing.

Protein preparation
In brief, transformed E. coli BL21 (DE3) clones were grown at 37 C in LB medium containing 50 mg/mL Kanamycin to an optical density at 600 nm of 0.8. Protein expression was induced at 30 C for 4 h by adding isopropyl-b-D-thiogalactopyranoside (IPTG) to a final concentration of 0.5 mM. After harvesting, cells were resuspended in lysis buffer (20 mM Tris (pH 8.0), 1 M NaCl, and 0.5 mM TCEP) with protease inhibitor, lysed by sonication, and centrifuged at 18 000Âg for 60 min at 4 C. The supernatant was loaded on a HisTrap HP column (GE Healthcare, Fairfield, CT), equilibrated with buffer A (20 mM Tris (pH 8.0), 1 M NaCl, 0.5 mM TCEP, and 5 mM imidazole), washed with 30 mM imidazole, and finally eluted with 500 mM imidazole. After His-MBP-TEV-tag removal using TEV protease, the protein was dialysed into buffer B (20 mM Tris (pH 8.0), 100 mM NaCl, and 0.5 mM TCEP) and reloaded onto the HisTrap HP column (GE Healthcare) to remove the tag, uncleaved protein, and TEV protease. The flow-through fractions were collected and loaded onto a MonoQ column (GE Healthcare) preequilibrated with buffer B. Some E. coli background protein and DNA rather than STARD10 can bind on the MonoQ. Then, the flowthrough fractions were collected and loaded onto a Heparin HP column (GE Healthcare) preequilibrated with buffer B. Fraction containing STARD10 protein was eluted with a linear gradient from 250 mM to 800 mM NaCl. The protein was finally purified by Superdex 75 10/ 300 GL column (GE Healthcare) with buffer B.

Crystallisation and structure determination
Crystallisation trials were carried out by sitting drop vapour diffusion method at 293 K. Freshly purified STARD10 was concentrated to w38 mg/mL and centrifuged to remove insoluble material before crystallisation. Single crystals appeared in condition containing 50% (v/ v) PEG 200, 100 mM sodium phosphate dibasic/potassium phosphate monobasic (pH 6.2), and 200 mM NaCl after one month. Cryo-freezing was achieved by stepwise soaking the crystals in reservoir solution Original Article containing 10, 20, and 30% (w/v) glycerol for 3 min and flash freezing in liquid nitrogen. X-ray diffraction data were collected on beamline I03 at the Diamond synchrotron X-ray source and were integrated and scaled with the Xia2 system [15]. The structure was determined by molecular replacement using the START domain structure from human STARD5 protein (PDB ID: 2R55) as a search model in CCP4, followed by rigid-body refinement by Refmac5 [16]. The structure was refined using PHENIX [17] and interspersed with manual model building using COOT [18]. The structure contains one STARD10 molecule in the asymmetric unit. 245 out of 291 residues (STARD10 full length) were successfully built into the density in the final structure. The statistics for data collection and model refinement are listed in Supplemental Table 2.

Molecular docking of ligands in STARD10 and STARD2
In silico molecular docking was performed using the Bristol University Docking Engine (BUDE 1.2.9). Firstly, a set of conformations for the head groups, glycero-inositol-1-phosphate and glycero-3phosphoinositol-1-phosphate, were generated using OpenBabel (2.4.1) giving 87 and 85 conformers, respectively. The docking grid was centred on the central cavity in STARD2 (1LNL) and STARD10 (6SER). Each conformer of the PI and PI(3)P head groups was docked into both structures. Each docking run found the best solutions by sampling a total of 1.1 million poses via BUDE's genetic algorithm. Final models were constructed by adding the 2 linoleoyl chains from the conformation in 1LN1 (STARD2 dilinoleoylphosphatidylcholine complex) and refined by energy minimisation with GROMACS.

Statistics
Data are expressed as mean AE SD. The normality of the data distribution was tested by D'Agostino and Pearson and ShapiroeWilk normality tests. For normally distributed data, significance was tested by Student's two-tailed t-test or Welch's t-test if the variances were found significantly different by F-test. ManneWhitney test was used for nonparametric data, and one-or two-way ANOVA with SIDAK multiple comparison test was used for comparison of more than two groups, using Graphpad Prism 7 software. P < 0.05 was considered significant.

3.1.
Stard10 deletion affects dense core granule ultrastructure As an initial approach to determining the target membranes for STARD10 action, we first explored the impact of deleting the Stard10 gene on b-cell ultrastructure. We have previously shown that crossing of Stard10 floxed mice to Ins1Cre knock-in mice [19] efficiently and selectively deletes STARD10 in the pancreatic b-cell [9]. Transmission EM images of b-cells in islets isolated from bStard10KO mice revealed a dramatic change in insulin granule morphology with a significant increase in "atypical" granules with a "rod-shaped" core ( Figure 1A The analyses of transmission EM images of human b-cells from partially pancreatectomised patient samples (Supplemental Figure 1A and Supplemental Table 1) showed no significant correlation between STARD10 expression measured by RNAseq and the percentage of mature granules (Supplemental Figure 1B) or the density of mature, immature, or total granules (Supplemental Figure 1C).
Thus, deleting Stard10 specifically in the b-cell in mice greatly affected the insulin granule ultrastructure and overall shape with an increase in granules with a "rod-shaped" core.

Islet zinc content is increased in bStard10KO mice
To determine whether the abnormalities in granule structure may result from altered Zn 2þ content of granules, itself a critical regulator of insulin crystallisation [1,20], we used the Förster resonance energy transfer-(FRET-) based sensor eCALWY4 [13]  3.3. Newly synthesised proinsulin is constitutively secreted by bStard10KO islets The characteristic decrease of plasma proinsulin:insulin ratio observed in human carriers of the risk alleles at this locus [21] and the observation of a similar phenotype in bStard10KO mice [9] suggest an action of STARD10 on proinsulin processing in the b-cell. We, therefore, next investigated this hypothesis by performing a metabolic labelling pulsechase experiment in isolated islets from WT and bStard10KO mice.
After 20 min of pulse-labelling with 35 S-amino acids, the islets of both genotypes were chased for 1.5 or 4 h in 5.5 mM glucose medium. Insulin was immunoprecipitated from both the cell lysate (C) and the secretion media (M) (Figure 2A). Labelled proinsulin secretion by bStard10KO islets was increased at low (5.5 mM) glucose compared to A similar experiment was carried out where the islets were chased after 30 min in 20 mM glucose ( Figure 2F). Although the proinsulin secretion and the secreted proinsulin:insulin ratio from the KO islets at 20 mM glucose tended to be increased versus wild-type islets, values were not significantly different between genotypes ( Figure 2G,H). It is worth noting that insulin granules ordinarily have a long lifespan in bcells, with a half-life of several days [22], indicating that the labelled peptides observed in this experiment originate from young granules.
In order to determine whether the increase in newly synthesised proinsulin secretion observed in the pulse-chase experiment reflected a change in total secreted proinsulin, we measured the total insulin and proinsulin in the secretion medium after 30 min in low (3 mM) or stimulating (17 mM) glucose concentrations. However, the secreted proinsulin:insulin ratios remained unchanged between genotypes at both glucose concentrations ( Figure 2I).

Preserved glucose-regulated membrane potential and b-celle b-cell connectivity in Stard10KO islets
Our previous study [9] showed that the deletion of Stard10 in the b-cell impaired glucose-induced cytoplasmic Ca 2þ increases and insulin secretion. To test for a potential upstream defect, e.g., in the closure of ATP-sensitive K þ (K ATP ) channels [23], we performed perforated patch-clamp electrophysiology [24] to measure plasma membrane potential in dispersed single b-cells. No change in glucose-induced membrane depolarisation was observed in b-cells from bStard10KO animals compared to WT littermates (Supplemental Figures 2A and  2B).
b-celleb-cell connections are essential for synchronised intraislet Ca 2þ influx and ultimately efficient insulin secretion [3]. We subjected Original Article

Solution of STARD10 3D structure and molecular docking identify PI and PI(3)P as potential ligands
To assess which membrane lipids may be bound by STARD10, we sought to obtain a crystal structure of the purified protein. We generated and purified the recombinant protein from a bacterial expression construct comprising the human STARD10 coding sequence fused with a 6His-MBP (Maltose Binding Protein) tag. Following the recombinant protein production in E. coli and protein purification, STARD10 generated a well-diffracting crystal. The crystal structure was then resolved at 2.3 Å resolution by molecular replacement using the structure of STARD5 (PDB ID: 2R55) ( Figure 3A; Supplemental Movie 1; Supplemental Table 2).  In the STARD10 crystal structure, the dimensions of the cavity are such that the protein is expected to readily accommodate phosphatidylinositols, whereas the equivalent STARD2 cavity, due to its smaller size, gives a poor predicted binding energy. On this basis, the STARD10 binding pocket is expected to bind PI and various phosphorylated PIs with higher affinity than STARD2.
To test this prediction, the head groups of PI and PI(3)P (glyceroinositol-1-phosphate and glycero-3-phosphoinositol-1-phosphate) were docked into the 3-dimensional structure of STARD10 (described above) and the closest family member STARD2 [26] by using the BUDE [27] ( Figure 3B, Supplemental Movie 2). The software was used to dock different representative conformations of the ligand to the target protein: 87 conformers for PI and 85 for PI(3)P. When these conformers were docked to STARD10, all final poses were located inside the binding cavity. On the other hand, for STARD2, only two conformers were placed by BUDE inside the protein, and the rest were docked onto the outer surface of the protein, suggesting that, contrary to STARD10, binding inside STARD2 by PI and PI(3)P is unfavourable. Likewise, the predicted binding energies of the PI and PI(3)P head groups to STARD10 are about 40 kJ mol À1 better than those to STARD2 ( Figure 3C presents a comparison between STARD10 (unliganded, in green) and STARD2 (bound to phosphatidylcholine, in grey)).
Supplementary video related to this article can be found at https://doi. org/10.1016/j.molmet.2020.101015 3.6. STARD10 binds to phosphatidylinositides In order to confirm the above inferences based on the STARD10 structure and molecular docking, we performed a lipid overlay assay, in which in vitro lipid binding to a range of phospholipids spotted onto a nitrocellulose membrane is assessed [28]. The efficiency and specificity of protein purification were monitored by SDS-PAGE staining with Coomassie Blue ( Figure 4A). Purified STARD10 protein was incubated with the PIP membrane strips and detected using an anti-STARD10 antibody and secondary anti-goat-HRP antibody ( Figure 4B).
Interestingly, STARD10 interacted with all PIP species interrogated, suggesting that STARD10 may bind to membranes containing these phospholipids.

Altered lipidomic profile in bStard10KO islets
The above observations suggested that the loss of lipid binding (including phosphoinositides) and transport by STARD10 in bStard10KO cells might result in changes in granule or cellular lipid composition which in turn may affect granule biogenesis ( Figure 1A). Since granule lipids could not readily be quantified given the relatively small numbers of b-cells that can usually be isolated from the mouse pancreas (w200 000, or w20 mg protein), precluding granule isolation at the scale needed, we limited our studies to whole islets.
The lipid composition of Ctl and Stard10-null b-cells was explored by targeted mass spectrometry. Of 24 classes and 280 species of lipids measured (Supplemental Table 3 Figure 5), suggesting that STARD10 is involved in the regulation of the turnover of these lipid species.
3.8. STARD10 is localised to the cytoplasm and nucleus of the bcell As an alternative approach to address which intracellular membranes may be affected by alterations in STARD10 expression, and hence lipid distribution, we next performed a confocal analysis of the subcellular localisation of STARD10. In the absence of a commercial antibody able to detect endogenous STARD10 by immunocytochemistry, we overexpressed the protein tagged with the GFP either in carboxy-or aminoterminus in the human b-cell line EndoC-bH1 [29]. STARD10 was present homogenously in the cytosol and nucleus, and we saw no particular localisation to a given organelle or membrane ( Figure 6). Immunodetection of bound protein was performed using a primary anti-STARD10 antibody (Santa Cruz) and a secondary anti-goat-HRP antibody (Santa Cruz). STARD10 was bound to all PIP species.

STARD10 interactome analysis identifies proteins involved in phosphatidylinositide signalling
To gain further insights into the possible mechanisms through which STARD10 may influence granule lipid composition, and other processes pertinent to the control of insulin secretion, protein interacting partners of STARD10 were identified by immunoprecipitation of the endogenous protein in the rat pancreatic b-cell line INS1 (832/13) and liquid chromatographyetandem mass spectrometry analysis after TMT labelling of the digested samples. 303 significantly enriched proteins were detected in the STARD10 pull-down condition compared with the nontargeting control IgG (P < 0.05 and FDR<0.05, paired t-test) ( Table 1). Amongst these significant interactions, we observed proteins well known to play a role in b-cell physiology: the pore-forming subunit Kir6.2 (Kcnj11, ATP-sensitive inward rectifier potassium channel 11) and its auxiliary subunit, the sulfonylurea receptor SUR1 (Abcc8, ATPbinding cassette subfamily C member 8) of the ATP-sensitive K þ (K ATP ) channel [30]; the phosphoinositide PI(4,5)P 2 -interacting protein granuphilin (Sytl4; synaptotagmin-like 4) associates with and is involved in the docking of the insulin granules [31]; the polypyrimidine tractbinding protein 1 (Ptbp1) is required for the translation of proteins localised in the insulin granule in response to glucose [32]. Gene ontology analysis of the identified STARD10 binding partners reflected an enrichment in proteins associated with RNA binding and processing, gene expression, and splicing (Supplemental Tables 4, 5, and 6). Interestingly, STARD10 also interacted with the inositol lipid kinase phosphatidylinositol 5-phosphate 4-kinase type-2 gamma (Pip4k2c) enzyme, which converts phosphatidylinositol 5-phosphate (PI(5)P) into phosphatidylinositol 4,5-bisphosphate (PI(4,5)P 2 ) [33].
The efficiency and specificity of STARD10 immunoprecipitation by the antibody and the co-immunoprecipitation of the Kir6.2 channel were confirmed by Western (immuno-)blot analysis (Supplemental Figure 4) 3.10. Altered expression of genes controlling inositol phospholipid signalling in bStard10KO islets STARD10 has previously been reported to influence the activity of peroxisome proliferator-activated factor (PPARa) in the liver [34], suggesting a possible influence on gene expression in the b-cell. In order to assess this possibility, we performed massive parallel RNA sequencing (RNAseq) on islets from WT and bStard10KO mice. We identified 88 differentially regulated genes (padj<0.05, n ¼ 6 animals), with 33 being upregulated and 55 being downregulated in KO versus WT islets. The list of the 20 most significantly regulated genes is presented in Supplementary Figure 5A and Table 2 (padj ranging from 1.71 Â 10 À157 to 2.25 Â 10 À3 ). As expected, a decrease in Ins1 expression in the bStard10KO mice carrying an Ins1-Cre knock-in allele was observed. Interestingly, Scg2, encoding secretogranin 2, a member of the granin protein family, localised in secretory vesicles [35], and Rasd2, also known as Ras homologue enriched in striatum (Rhes), were upregulated. Interestingly, consistent with possible changes in phosphoinositide signalling, Pirt (phosphoinositide-interacting regulator of transient receptor potential channels), a regulator of transient receptor potential Ca 2þ channels [36], and Syt1 (synaptotagmin 1) regulating fast exocytosis and endocytosis in INS1 cells [37] were downregulated in bStard10KO islets.
Enrichment analysis through the gene ontology consortium website (http://www.geneontology.org/) identified genes whose products are localised in neuronal projections (11 genes, fold enrichment ¼ 4.57, P < 0.05) and neurons (12 genes, fold enrichment ¼ 3.95, P < 0.05) amongst the downregulated genes (Supplemental Figure 5B, Supplemental Table 7). As secretory cells, neurons share their transport and secretory machinery with b-cells, suggesting that these genes play a role in the insulin secretion process. Amongst the genes identified, several are implicated in insulin secretion and/or b-cell survival: Chl1 (cell adhesion molecule L1-like), whose silencing has previously been shown to reduce glucose-induced insulin secretion in INS1 cells [38] and is downregulated in islets from type 2 diabetic subjects [39]; Nos1 (neuronal nitric oxide synthase), implicated in insulin secretion and b-cell survival [40]; Adcyap1r1 (adenylate cyclaseactivating polypeptide 1 receptor 1), encoding the PAC1 receptor of the pituitary adenylate cyclase-activating polypeptide (PACAP) Figure 5: Altered lipidomic profile in bStard10KO islets. The deletion of Stard10 in mouse pancreatic b-cells significantly increased the total cholesteryl esters, phosphatidylinositols, and the phosphatidylethanolamine species 34:0 in islets (n ¼ 5 animals; *P < 0.05, **P < 0.01, paired t-test). See Supplemental Table 3 for a complete list of all the lipids measured. Original Article neuroendocrine factor [41]; and Cplx1 (complexin 1) [42]. None of several key b-cell signature genes involved in the regulation of Ins expression and b-cell identity (Supplemental Figure 5C), or of the "disallowed" b-cell gene family (Supplemental Figure 5D) [3], were affected by Stard10 deletion.

DISCUSSION
Our earlier demonstration [10] that STARD10 is the likely causal gene at the T2D risk locus on chromosome 11q13 has emphasised the probable importance of lipid transfer for the normal physiology of the pancreatic b-cell. Indeed, the role of lipid transfer proteins has emerged recently as an important element in cell biology [43]. Several lines of evidence presented in our studies support a role for STARD10 in PI/PIPs binding and potentially transport between intracellular membranes. (a) Resolution of the crystal structure of STARD10 and molecular docking predict the binding of PI/PIPs in its binding pocket ( Figure 3). (b) Purified recombinant STARD10 binds to phosphatidylinositide species in a lipid overlay assay (Figure 4). (c) The phosphatidylinositol content was increased in bStard10KO compared to control islets ( Figure 5). (d) Our proteomic study identified the phosphatidylinositol 5-phosphate 4-kinase type-2 gamma (Pip4k2c) enzyme and the PI(4,5)P 2 -binding protein granuphilin [44] as STARD10 binding partners. (e) The expression of the phosphoinositide-binding proteins Syt1 and Pirt were decreased in bStard10KO islets (Supplemental Figure 5). Pirt encodes a phosphoinositide-interacting regulator of the transient receptor potential (TRP) Ca 2þ channels.
Several TRP channels are expressed in the b-cell and an increasing number of data points towards the role of these channels in the regulation of insulin secretion [45]. A previous study [46] has shown that female Pirt À/mice are heavier than controls and develop glucose intolerance.  Taken together, these data suggest that STARD10 may bind PI and/ or PIPs either in the lipid-binding pocket, as predicted by the crystal structure and molecular docking (Figure 3), or at the surface of the protein. We note that the findings from the lipid overlay assay are consistent with a binding of STARD10 to membranes containing PIP lipids (with an affinity which appears higher for (3)P containing species). This could be important for the targeting of STARD10 to specific membrane compartments. In contrast to its closest relative STARD2, which has been shown to selectively bind phosphatidylcholine [47], our crystal structure determination and molecular docking indicate that STARD10, due to its larger binding pocket, is predicted to bind phosphatidylinositol (PI) and its various phosphorylated forms (PIPs) (Figure 3) in addition to its previously recognised ligands phosphatidylcholine and phosphatidylethanolamine [11]. Consistent with likely mistargeting of lipids to membranes involved in insulin secretion, we demonstrate that STARD10 is required for normal secretory granule maturation in murine b-cells. Of note, recent findings [48] have indicated that cholesterol is also required for normal granule maturation. In addition, in INS-1 (832/13) cells, the insulin secretory granule was found to contain a high proportion of phosphatidylinositol (w20%, 5-fold that in the whole cell) [48]. It is thus conceivable that the changes in secretory granule biogenesis may occur as a result of changes in lipid delivery to the granule membrane and/or transcriptional events (i.e., altered gene expression). In contrast, we were not able to obtain evidence for an effect of STARD10 variation on human b-cell granules, perhaps reflecting a greater diversity in granule structure observed in man, as well as the limited number of subjects available for analysis. Proteomic analysis of the STARD10 binding partners in INS1(832/13) pancreatic b-cells (Table 1) identified several proteins involved in insulin secretion such as the subunits of the K ATP channel Kir6.2 and SUR1 and the secretory vesicle protein granuphilin, which possesses a C2 domain capable of binding the phosphoinositide PIP 2 [49]. Another binding partner, PTBP1 (polypyrimidine tract-binding protein 1), regulates the translation of insulin granule proteins in response to glucose [32]. The potential regulation by STARD10 of the activity of this protein could conceivably affect granule structure. Given the known function of STARD proteins in lipid transport, it is possible that, in the absence of STARD10, phospholipids are not addressed to the appropriate membranes in cellular subcompartments. It is, thus, conceivable that the altered lipidomic profile might impact the localisation of STARD10 interacting partners and, therefore, alter their function in the b-cell and in insulin secretion.
RNAseq analysis of the transcriptome of WT and bStard10KO islets identified differentially expressed genes which are likely to play a role in adult b-cell physiology (Supplemental Figure 5). Of note, in liver, Stard10 knockout affects the activity of PPARa [34], and it is conceivable that changes in lipid distribution impact lipid-regulated factors such as those in the b-cell. There is also evidence that another Stard family member, STARD2, binds directly to transcription factors [50], so this mechanism is conceivable for STARD10. The gene products of Syt1 (synaptotagmin 1) and Cplx1 (complexin 1) are part of the molecular machinery driving vesicle exocytosis and form a tripartite complex with Soluble N-ethylmaleimide sensitive factor Attachment protein Receptor (SNARE) proteins [51]. However, none of these genes are readily linked to a change in insulin crystallisation or granule biogenesis. Importantly, we also identified Pirt, a phosphoinositideinteracting regulator of transient receptor potential channels, as a binding partner of STARD10 and thus likely either to sense or to influence phosphoinositides levels in cells. Changes of the product of these genes at the protein level by Western blotting and/or immunocytochemistry would provide additional validation and information and will be the focus of further studies. We also considered the possibility that altered granule composition may reflect altered intracellular Zn 2þ levels given that similar changes are observed after the deletion of the gene encoding the Zn 2þ transporter ZnT8, Slc30a8 [52]. Thus, the total Zn 2þ content, likely reflecting intragranular Zn 2þ , assessed by ICP/MS, was increased in the islets of bStard10KO versus WT mice (Figure 1). The latter appears unlikely to be due to changes in the expression of ZnT8 (encoded by Slc30a8), which we did not observe, but might reflect altered intrinsic activity, or granule recruitment, of ZnT8 due to altered membrane lipid composition [53]. Consistent with this possibility, anionic phosphatidylinositols increase, whereas nonbilayer phospholipids (promoting membrane curvature), lysophosphatidylcholine (LPC) and phosphatidylethanolamine (PE), lower ZnT8 activity [53].
An unexpected finding from the current study is that bStard10KO islets secreted more radiolabeled proinsulin versus WT after 4 h of a chase in 5.5 mM glucose medium, whilst no difference was observed between genotypes at 20 mM glucose. These results contrast with a decrease in the proinsulin:insulin ratio observed in plasma from bStard10KO mice [9] and a similarly lowered ratio in human risk variant carriers versus controls [21]. One possible explanation is that increased proinsulin secretion in unstimulated cells may enrich, for the storage of mature, processed insulin, hence producing a lowered proinsulin:insulin release during stimulation with high glucose. According to studies using radioactive labelling [54] or live-cell imaging of a SNAP-tag fused with insulin [55], young, newly synthesised insulin granules are preferentially secreted upon glucose stimulation. Given their reduced transit time, the proportion of unprocessed proinsulin in the young granules could be higher than that in the older ones. Other evidence points towards the existence of vesicular trafficking pathways leading to unstimulated, "constitutive-like" secretion of proinsulin [56]. It has been speculated [57] that this involves routing to the endosomal compartment and either cargo degradation in the lysosome or secretion at the plasma membrane through a "constitutive-like" pathway [58]. By changing the lipid composition of granules, STARD10 may modulate their trafficking properties, affect the age of secreted granules, and/or decrease their targeting to the lysosomes. On the other hand, the total proinsulin:insulin ratio in the secretion medium of isolated islets remained unchanged between the two genotypes ( Figure 2I), and the cellular (total immunoreactive) proinsulin:insulin ratios were similar in WT and bStard10KO islets ( Figure 2D).
In addition to its role in insulin crystallisation in dense core granules, zinc ions, which are cosecreted with insulin, inhibit insulin secretion in an autocrine and paracrine manner [59], at least in part through the activation of the pancreatic K ATP channel [60,61]. Zn 2þ coreleased with insulin also attenuates insulin hepatic clearance, an effect that has been proposed to be mediated through the inhibition of the clathrinmediated insulin receptor endocytosis by the liver [59]. An increase in Zn 2þ secretion by the b-cell in the presence of a reduced STARD10 expression could, thus, participate in the decrease both in insulin secretion and in the plasma proinsulin:insulin ratio observed in the bStard10KO mice [9] and the carriers of the risk alleles [21,62].
A striking finding of our studies is the change in lipid composition observed in the bStard10KO islets ( Figure 5). The increased levels of cholesterol, phosphatidylinositol, and PE 34:0 observed may be explained by less efficient transport of lipids, their precursors, or breakdown products between sites of synthesis and degradation in the absence of Stard10. Although it was not possible to assess which intracellular membranes may be affected given the limited sensitivity Original Article of the bulk analysis performed and study of changes in lipid profiles in subcompartments such as insulin granule being limited in mouse islet due to the small amount of material compared with cultured cell lines, previous studies have made a link between cholesterol and insulin granule biogenesis and secretion. For example, according to Bogan et al. [63], insulin granules are the major site of cholesterol accumulation inside the b-cell. An excess of cholesterol altered the granule ultrastructure with an increase in mean granule diameter and the retention of immature granule proteins to the granule, changes associated with impaired insulin secretion. In addition, loss of the cholesterol transporter ABCG1from the b-cell led to a decrease in the cholesterol content of the granules and was associated with abnormal, enlarged insulin granules [64].
The impact of altering b-cell cholesterol content and homeostasis, leading to an excess or deficiency in cellular cholesterol content, has previously been achieved through various means, including exogenous addition of cholesterol [65], targeting either cholesterol transporters [48,66,67] or biosynthetic enzymes [68], or the use of an ApoEdeficient mice model [69]. Impaired glucose-induced insulin secretion was observed in all of these cases. In addition to cholesterol, the insulin secretory granule was found to contain a high proportion of phosphatidylinositol (w20%, 5-fold that in the whole cell) in INS-1 (832/13) cells [70]. Phosphatidylinositol transfer protein has been shown to increase secretory vesicle formation [71]. Interestingly, the phosphoinositide PI(4)P is present at the granule membrane and its dephosphorylation into PI by the phosphatase Sac 2 is required for efficient insulin secretion by the b-cell [72]. The synthesis of another phosphoinositide, PI(3)P, occurs on secretory vesicles in neurosecretory cells [73], and the inhibition of class II phosphoinositide 3 kinase responsible for the synthesis of PI(3)P from PI impaired insulin secretion [74]. Several additional studies have pointed to the role of phosphoinositides in insulin secretion [75e77]. Moreover, a connection between cholesterol and phosphoinositides in cellular processes has previously been suggested. For example, cholesterol stabilizes phosphoinositide domains, increasing the recruitment of target proteins to the PIP domains [78], and transport proteins with dual phosphoinositide/cholesterol ligands play an important role in the cellular localisation of these lipids [48,79]. In bcells, cholesterol regulates insulin secretion through PI(4,5)P 2 [65]. In addition, the PI transporter TMEM24 regulates pulsatile insulin secretion by replenishing the phosphoinositide pool at the endoplasmic reticulum-plasma membrane contact sites [80]. It is, thus, tempting to speculate that STARD10 might affect insulin secretion in a similar fashion.
In conclusion, we identify phosphatidylinositols as potential new ligands for STARD10. Changes in STARD10 expression in carriers of type 2 diabetes risk alleles may consequently affect the b-cell lipid composition and alter granule maturation and, ultimately, insulin synthesis and secretion. performed the interactome analysis. R.B.S. performed the STARD10 ligand docking studies. F.A. provided the STARD10-GFP and GFP-STARD10 constructs. A.C. performed the ICP/MS analysis. I.L. was the holder of the Home Office project licence and responsible for the work carried on animals in this study. G.A.R. conceived the study and cowrote the manuscript. G.A.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

STUDY APPROVAL
All in vivo procedures were approved by the UK Home Office according to Animals (Scientific Procedures) Act 1986 (HO Licence PPL 70/7349) and were performed at the Central Biomedical Service, Imperial College, London, UK. Pancreatectomised patient samples were from the IMIDIA consortium (www.imidia.org; Solimena, M, personal communication) with appropriate permissions from donors and/or families and approval by the local ethics committee. A written informed consent was received from participants prior to the inclusion in the study.

DATA AND RESOURCE AVAILABILITY
The RNAseq raw sequence data on bStard10KO and WT mice will be made available via deposition to ArrayExpress. Lipidomic mass spectrometry raw data can be provided upon request and will be deposited in MetaboLights. The resources generated or analysed during the current study are available from the corresponding author upon reasonable request.