GLP-1 receptor partners with VAPB and SPHKAP at cell contact sites to control mitochondrial changes and function in insulin-producing cells

The interaction of active GLP-1R with VAPB was deemed functionally relevant, as it was no longer present when VAPB P56S, a self-aggregating loss-of-function VAPB mutant associated with amyotrophic lateral sclerosis (ALS)^16,17, was expressed instead of its wildtype (WT) counterpart (Supplementary Fig. 1g–i), and as both VAPB knockdown by RNAi (Fig. 1k and Supplementary Fig. 1j, k), and overexpression of VAPB P56S, but not of WT VAPB (Supplementary Fig. 1l), significantly reduced the capacity of GLP-1R to potentiate insulin secretion from INS-1 832/3 cells.

As stated above, besides VAPB, our interactome analysis highlighted several factors enriched for their association with active GLP-1R localised at the ER, mitochondria and/or ERMCSs, a location where VAPB plays a prominent organising role¹⁷, suggesting that ERMCSs could be a potential hub for endosomal GLP-1R signalling. To determine if GLP-1R induces cAMP generation specifically at VAPB locations, we modified a fluorescence resonance energy transfer (FRET)-based system of biosensors targeted to endogenous proteins (FluoSTEPs)²⁵, so that a FRET cAMP sensor would fully reconstitute only in the presence of the GFP₁₁ fragment fused to VAPB, resulting in functional biosensor assembly specifically at VAPB loci (Supplementary Fig. 3a). This modified FluoSTEP system was used to measure VAPB-localised cAMP generation in response to either exendin-4 or exendin-F1 stimulation (Fig. 2e). Results in WT INS-1 832/3 cells (expressing endogenous levels of GLP-1R) show VAPB-localised cAMP responses with both agonists, which are less pronounced for the slow internalising exendin-F1 compared to exendin-4, demonstrating that GLP-1R actively signals at VAPB loci. Next, we evaluated whether GLP-1R signalling would also trigger a pool of PKA activity at the surface of mitochondria, a process known to be important for the control of mitochondrial function²⁶. To this end, we employed either a global (whole cell) or a mitochondrially-restricted PKA biosensor expressed at the outer mitochondrial membrane (OMM)²⁷ (Supplementary Fig. 3b). Exendin-4 stimulation of WT INS-1 832/3 cells triggered, as expected, a robust global PKA response, which was also evident with the OMM-localised PKA biosensor, demonstrating that GLP-1R stimulation triggers PKA activity at the mitochondrial surface (Supplementary Fig. 3c). Furthermore, utilising the same biosensors as above, we detected reduced OMM-localised, but no change in global, PKA activity with slow internalising exendin-F1 versus exendin-4 (Fig. 2f), leading to increased mitochondrial over global PKA signalling propensity with exendin-4 at this acute (5-minute) stimulation time-point (Fig. 2g). This, together with the VAPB-localised cAMP FluoSTEP results from above, indicates that GLP-1R internalisation is important for GLP-1R-induced cAMP generation at VAPB loci and PKA signalling at OMMs, at least acutely.

Further high-resolution 3D imaging to confirm endosomal GLP-1R localisation to ERMCSs included in situ correlative light and electron microscopy (CLEM) at room temperature to correlate ultrastructural features with positive fluorescent signals for SNAP/FLAG-hGLP-1R, EGFP-VAPB and mitochondria. To this end, INS-1 832/3 SNAP/FLAG-hGLP-1R cells were labelled at the cell surface with a membrane impermeable fluorescent SNAP-tag probe prior to exendin-4 stimulation, with correlation of the confocal slice to the corresponding TEM tomogram (Fig. 3e) revealing two mitochondrial volumes in close contact with EGFP-VAPB-positive ER, itself docked to a vesicle positive for SNAP/FLAG-hGLP-1R, indicative of a GLP-1R-positive endosome (Fig. 3f and Supplementary Movie 3). Similar results were obtained using a cryo-CLEM approach, with a SNAP/FLAG-hGLP-1R endosome docked to the EGFP-VAPB-positive ER network, itself in contact with a nearby mitochondrion, following correlation of cryo-confocal and cryo-FIB-SEM data (Supplementary Fig. 4a–c).

These results were further replicated using an immuno-EM approach in primary mouse islets transduced with a SNAP/FLAG-hGLP-1R-expressing adenovirus by labelling cell surface SNAP/FLAG-hGLP-1R in living islets with membrane-impermeable SNAP-biotin followed by 5 nm gold-conjugated streptavidin prior to exendin-4 stimulation and TEM processing. TEM imaging of ultrathin (70 nm) sections from the resin-embedded islets from above revealed 5 nm gold-positive endosomes (indicative of the presence of SNAP/FLAG-hGLP-1R) forming contacts with the ER, itself contacting nearby mitochondria (Fig. 3g).

Amongst the interactors enriched for active GLP-1R binding in our MS interactome analysis, we identified both PKA-RIα/β and the A-kinase anchoring protein (AKAP) SPHKAP, a paralog of AKAP11 (also known as AKAP220) that binds to two RIα subunits of the PKA holoenzyme²⁹. Both SPHKAP and AKAP11 harbour a FFAT (two phenylalanines in an acidic tract) motif^30,31, a specific short protein sequence that binds to the Major Sperm Protein (MSP) domain of VAPs (Supplementary Fig. 5a), which, in the case of SPHKAP, can be regulated by phosphorylation (phosphoFFAT, or pFFAT)³¹. Agreeably, SPHKAP has previously been shown to interact with VAP proteins in neurons and assemble functionally relevant PKA-RI biomolecular condensates at VAPA/B-positive ER locations^32,33. SPHKAP is expressed in metabolically relevant tissues, including pancreatic islets and intestinal K and L cells, and has been previously shown to play important roles in both incretin secretion from the gut and incretin-dependent glucoregulation^34,35. Additionally, we found SPHKAP mRNA expression to be decreased in a published RNA-Seq database of islets from db/db versus WT mice (Supplementary Fig. 5b), and meta-analyses from four separate genome-wide association studies (GWAS) in humans uncovered genome-wide significant associations between SPHKAP gene variants, T2D risk, and high BMI^36–39 (Supplementary Fig. 5c, d). We found the effect of SPHKAP gene variants in T2D versus high BMI to be tightly correlated (Supplementary Fig. 5e), suggesting a shared mechanism of action for SPHKAP in both disorders where GLP-1RAs are also known to be effective. Our own genetic analysis also demonstrated a significant association between SPHKAP gene coding variants and random blood glucose levels in whole exome sequencing (WES) data from UK Biobank (UKBB) individuals without diabetes (Supplementary Fig. 5f), strongly supporting a role for SPHKAP in the regulation of β-cell function.

We next determined that the exendin-4-induced GLP-1R–SPHKAP interaction requires VAPB, as it is abrogated by VAPB knockdown (Fig. 4a, b, VAPB RNAi conditions; Supplementary Fig. 6a). Confocal analysis of SPHKAP-EGFP localisation in WT INS-1 832/3 cells revealed that, in contrast with a previous report suggesting mitochondrial localisation⁴⁰, SPHKAP localises to an intracellular network occasionally in contact with, but not inside, mitochondria (Fig. 4c, Control RNAi panels; Supplementary Fig. 6d). Indeed, labelling of ER with the fluorescent dye ER-Tracker revealed the punctate localisation of SPHKAP-EGFP along ER membranes (Fig. 4d, e), in agreement with recent reports which also indicate SPHKAP localisation and PKA-RI targeting to the ER^32,33. Also in agreement with these reports, the localisation of SPHKAP-EGFP to ER loci required VAPB, as it was lost in VAPB RNAi-treated cells (Fig. 4c, VAPB RNAi panels). We next generated a mutant version of SPHKAP-EGFP harbouring a point mutation in its pFFAT motif (SPHKAP-EGFP A215E, DFLTASE to DFLTESE). This strategy was based on previous structural data showing that position 5 of the FFAT core motif is located within a hydrophobic pocket in the MSP domain of VAP⁴¹, with previous studies indicating that introducing a negative charge in this position causes a steric hindrance which effectively disrupts FFAT motif—VAP interaction⁴². Agreeably, our SPHKAP-EGFP pFFAT mutant lost both its binding to exendin-4-stimulated SNAP/FLAG-hGLP-1R (Supplementary Fig. 6e) and its localisation to the ER network (Supplementary Fig. 6f). We additionally demonstrated the co-existence of SNAP/FLAG-hGLP-1R, VAPB and SPHKAP in mitochondria-associated membranes (MAMs) purified from exendin-4-stimulated INS-1 832/3 SNAP/FLAG-hGLP-1R cells (Supplementary Fig. 6g), and a protein-protein interaction between SPHKAP and EGFP-VAPB in the same cells (Supplementary Fig. 6h). Taken as a whole, these results demonstrate the association of SPHKAP with active GLP-1R at ERMCSs in a VAPB-dependent manner via the pFFAT motif of SPHKAP.

We next tested the effect of SPHKAP downregulation on GLP-1R-dependent downstream signalling. SPHKAP knockdown efficiency in INS-1 832/3 cells was validated against both endogenous SPHKAP (Supplementary Fig. 6i) and SPHKAP-EGFP (Supplementary Fig. 6j). Loss of SPHKAP did not affect the interaction between the receptor and VAPB (Supplementary Fig. 7a-c), indicating that GLP-1R is recruited to VAPB-positive ERMCSs independently of SPHKAP. Loss of SPHKAP also did not affect the capacity for the receptor to generate cAMP following exendin-4 stimulation of WT INS-1 832/3 cells (Fig. 4f). In contrast, while global PKA responses to exendin-4 were not affected (Fig. 4g), exendin-4-induced OMM-localised PKA activity was blunted by SPHKAP downregulation (Fig. 4h). Finally, we employed a similar modified FluoSTEP system to the one shown in Fig. 2, but this time adapted to assess VAPB-localised PKA activity in response to exendin-4 stimulation in Control versus SPHKAP RNAi-treated WT INS-1 832/3 cells (see Supplementary Fig. 7d for a schematic of this method). Results showed that exendin-4 stimulation triggers VAPB-localised PKA activity in WT INS-1 832/3 cells, and that this activity requires SPHKAP (Fig. 4i), presumably via SPHKAP binding to PKA-RI and assembly of a PKA signalling hub at ERMCSs. Consequently, we could observe the appearance of PKA-RIα-EGFP-positive puncta, indicative of liquid-liquid phase separation (LLPS) and biomolecular condensate assembly²⁵, in exendin-4-stimulated WT INS-1 832/3 cells (Supplementary Fig. 7e).

As SPHKAP has previously been shown to bind to the mitochondrial cristae junction regulator and MICOS complex component MIC19⁴⁰, a prominent PKA substrate involved in both cristae assembly during mitochondrial remodelling⁴³ and in regulation of ERMCSs, we next tested whether exendin-4 stimulation would trigger MIC19 phosphorylation in WT INS-1 832/3 cells using a MIC19 immunoprecipitation strategy coupled with western blotting with an antibody against PKA phosphosites (RRxpS/T)⁴⁴. Exendin-4 exposure resulted in a significant increase in PKA-specific MIC19 phosphorylation, an effect that was potentiated by the adenylate cyclase activator forskolin (Fig. 5e). Dual knockdown of SPHKAP and VAPB in INS-1 832/3 cells led to widespread changes in PKA phosphorylation of several MICOS complex components downstream of GLP-1R activation, suggesting that ERMCS-localised GLP-1R signalling modulates the activity of the MICOS complex (Supplementary Fig. 7k).

As the main role of the MICOS complex is to orchestrate mitochondrial cristae assembly⁴⁵, and MIC19 plays an important role in the transfer of lipid precursors across ERMCSs during mitochondrial remodelling¹⁰, we next assessed whether GLP-1R stimulation would play a role in mitochondrial remodelling processes in INS-1 832/3 cells, and whether this would be affected by SPHKAP knockdown. We first investigated if acute exendin-4 stimulation triggers changes in mitochondrial turnover. Despite no detectable changes in mitochondrial DNA copy number after 30 minutes of exendin-4 exposure in Control versus SPHKAP RNAi-treated, or in GLP-1R KO INS-1 832/3 cells (Supplementary Fig. 7l), acute (10-minute) stimulation of WT INS-1 832/3 cells with exendin-4 triggered increased mitophagy, assessed using an mKeima-based assay, an effect not present following SPHKAP downregulation (Fig. 5f). Further evidence of exendin-4-induced mitophagy was obtained by CLEM analysis of INS-1 832/3 SNAP/FLAG-hGLP-1R cells, where we could find instances of SNAP/FLAG-hGLP-1R-positive endolysosomes surrounded by mitochondria (Supplementary Fig. 8a,b). We also detected increased recruitment of the mitochondrial fission protein mCherry-Drp1 to mito-BFP-expressing mitochondria in exendin-4-stimulated versus vehicle-exposed WT INS-1 832/3 cells (Fig. 5g, Supplementary Fig. 8c, Supplementary Movies 4 and 5). Drp1 phosphorylation at Serine 616, which, unlike Serine 637⁴⁶, is important for regulating mitochondrial fission and is not directly PKA-dependent^47,48, was, as expected, not modified by acute exendin-4 exposure in Control RNAi-treated cells, but was progressively lost 15 minutes post-exendin-4 stimulation in SPHKAP RNAi-treated WT INS-1 832/3 cells (Fig. 5h).

Next, we performed mitochondrial morphology analysis of WT INS-1 832/3 cells treated with Control versus SPHKAP RNAi prior to vehicle or exendin-4 exposure (Fig. 6c). While exendin-4 triggered a similar degree of mitochondrial network remodelling as above in Control RNAi-treated cells, SPHKAP knockdown resulted in disrupted mitochondrial morphology at vehicle conditions, which somewhat resembled that of exendin-4-stimulated Control RNAi-treated cells in terms of tubularity but had further visual abnormalities. Moreover, exendin-4 stimulation had no further effect on the mitochondrial morphology of SPHKAP RNAi-treated cells. These results were validated with the Mitochondria Analyzer plugin in Fiji, which showed reduced mitochondrial form factor and mean area following exendin-4 exposure of Control RNAi-treated cells, with SPHKAP RNAi-treated cells under vehicle conditions showing similar values to those of exendin-4-stimulated control cells (Supplementary Fig. 8e). However, with this analysis method, we also detected an apparent increase in form factor and mitochondrial mean area in exendin-4-stimulated versus vehicle-treated SPHKAP RNAi-treated cells, an effect opposite to that triggered by exendin-4 in control conditions.

To better visualise the changes in mitochondrial morphology under these different conditions, we performed z-series analysis of confocal sections across whole cells, represented as MIPs in Fig. 6d, as well as STED microscopy analysis of mitochondrial morphologies using the fixable STED-compatible mitochondrial dye PKmito ORANGE FX (Fig. 6e), with Control versus SPHKAP RNAi-treated cells showing similar changes in mitochondrial morphology as those described above.

We also performed mitochondrial morphology assessments in dispersed islet cells from mice transfected with Control versus SPHKAP siRNA, which responded similarly to intact islets in terms of increased mitochondrial fragmentation following exendin-4 exposure in control conditions, but showed disturbed mitochondrial morphology under vehicle conditions and no further effect of exendin-4 exposure in SPHKAP siRNA-treated cells (Fig. 9d). These phenotypic changes were still present when assessed as changes in mitochondrial form factor and mean area using the Mitochondria Analyser Fiji plugin (Supplementary Fig. 9e).

INS-1 832/3 rat insulinoma cells were maintained in RPMI-1640 at 11 mM D-glucose, supplemented with 10% foetal bovine serum (FBS), 10 mM HEPES, 2 mM L-glutamine, 1 mM pyruvate, 50 μM β-mercaptoethanol and 1% penicillin/streptomycin. INS-1 832/3 cells stably expressing SNAP/FLAG-tagged human GLP-1R (hGLP-1R) or human GIPR (hGIPR) were generated by transfection of pSNAP-GLP-1R or pSNAP-GIPR (Cisbio) into INS-1 832/3 cells lacking endogenous GLP-1R or GIPR expression (INS-1 832/3 GLP-1R or GIPR KO, a gift from Dr Jacqueline Naylor, AstraZeneca⁹²), followed by G418 (1 mg/mL) selection. Mycoplasma testing was performed biannually. Cell culture reagents were obtained from Sigma or Thermo Fisher Scientific.

Exendin-4, GLP-1 and GIP were purchased from Bachem, semaglutide was from Imperial College London Healthcare NHS Trust pharmacy, and tirzepatide was a gift from Sun Pharmaceuticals. Custom analogues (exendin-F1 and exendin-D3) were from WuXi AppTec, China, and danuglipron and orforglipron were purchased from MedChemExpress. SNAP-Surface fluorescent probes were from New England Biolabs and MitoTracker™ Red CMXRos and ER-Tracker Red were from Thermo Fisher Scientific.

INS-1 832/3 SNAP/FLAG-hGLP-1R cells were treated in 6-cm dishes with vehicle or 100 nM GLP-1RA (GLP-1, exendin-4, exendin-F1 or exendin-D3) for 5 minutes. Following stimulation, cells were washed in ice cold PBS and osmotically lysed in lysis buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Triton X-100, 1 mM EDTA) with 1% protease inhibitor (Roche) and 1% phosphatase inhibitor cocktail (Sigma–Aldrich) at 4 °C. Cell lysates were centrifuged at 17,000 × g and 4 °C for 10 minutes, and supernatants added to 40 μL anti-FLAG M2 agarose beads (A2220, Sigma-Aldrich) for overnight rotating incubation at 4 °C to increase binding efficiency. Samples were centrifuged at 5000 × g for 30 seconds, and bead pellets washed 3× in 500 μL wash buffer (50 mM Tris-HCl, 150 mM NaCl, pH 7.4) before storage at −80 °C until further analysis. The number of biological replicates per condition was 4 for all conditions apart from GLP-1-stimulated samples, which included 3 replicates. Negative controls included INS-1 832/3 SNAP/FLAG-hGLP-1R cell lysates incubated with empty Sepharose beads (Sigma–Aldrich) and INS-1 832/3 cell lysates (not expressing the tagged receptor) incubated with anti-FLAG M2 beads.

Beads were resuspended in 150 μL 1 M urea in 20 mM HEPES pH 8.0 with 1 mM DTT, followed by addition of 1.5 μg Trypsin gold (Promega) with overnight digestion at 37 °C. Supernatants were subsequently acidified with 0.5% trifluoroacetic acid (TFA), followed by desalting using reversed-phase spin tips (Glygen Corp) and drying using vacuum centrifugation. Dried peptides were resuspended in 0.1% TFA and Liquid Chromatography Tandem Mass Spectrometry (LC-MS/MS) performed in single replicate injections as previously described⁹³ using an Ultimate 3000 nano-HPLC coupled to a Q-Exactive mass spectrometer (Thermo Scientific) via an EASY-Spray source. Peptides were loaded onto a trap column (Acclaim PepMap 100 C18, 100 μm × 2 cm) at 8 μL/minute in 2% acetonitrile and 0.1% TFA. Peptides were eluted on-line to an analytical column (Acclaim Pepmap RSLC C18, 75 μm × 75 cm). A 90-minute stepped gradient separation was used with 4–25% acetonitrile 80%, formic acid 0.1% for 60 minutes, followed by 25–45% for 30 minutes. Eluted peptides were analysed by Q-Exactive operating in positive polarity and data-dependent acquisition mode. Ions for fragmentation were determined from an initial MS1 survey scan at 70,000 resolution (at m/z 200), followed by higher-energy collisional dissociation of the top 12 most abundant ions at a resolution of 17,500. MS1 and MS2 scan AGC targets were set to 3e6 and 5e4 for a maximum injection time of 50 and 75 mseconds, respectively. A survey scan range of 350–1800 m/z was used, with a normalised collision energy set to 27%, minimum AGC of 1e3, charge state exclusion enabled for unassigned and +1 ions and a dynamic exclusion of 45 seconds.

Data was processed using the MaxQuant software platform (v1.6.10.43), with database searches carried out by the in-built Andromeda search engine against the Swiss-Prot Rattus Norvegicus database (Downloaded—21st May 2022; entries: 8132) concatenated with the human GLP-1R protein sequence. A reverse decoy database approach was used at a 1% false discovery rate (FDR) for peptide spectrum matches and protein identifications. Search parameters included: maximum missed cleavages set to 2, variable modifications of methionine oxidation, protein N-terminal acetylation, asparagine deamidation and cyclisation of glutamine to pyroglutamate.

Label-free quantification (LFQ) was enabled with a minimum ratio count of 1. Proteins were filtered for those containing at least one unique peptide and identified in all biological replicates included per condition. For normalisation, proteins were ranked by fold change increase in abundance compared with control (vehicle-treated) immunoprecipitates. The resulting data was pre-processed by elimination of proteins that are non-specifically bound and residual background proteins, such as common contaminants and reverse database proteins. To further facilitate the selection of GLP-1R interactors, the following criteria were established; (i) identified at least twice from four runs; (ii) not nuclear proteins; and (iii) previously identified in pancreatic β-cells. Untargeted LFQ intensity values, normalised to hGLP-1R levels in each immunoprecipitated sample, were analysed by LFQ Analyst⁹⁴ to identify differences in levels of individual protein-GLP-1R interactions between vehicle and GLP-1RA-treated conditions. Paired t-tests adjusted for multiple comparisons using Benjamin–Hochberg FDR correction and Perseus type imputations were used for statistical analysis. To generate novel information about the functional pathways represented by these potential GLP-1R interactors, we performed pathway enrichment analysis with the GLP-1R binding partners exhibiting changes between vehicle and exendin-4-stimulated conditions using gProfiler.

INS-1 832/3 cells were transfected with relevant amounts of plasmid cDNA outlined in Supplementary Table , or siRNA outlined in Supplementary Table , using Lipofectamine 2000 (Life Technologies) in media without antibiotics according to the manufacturer's instructions. Experiments were carried out 24–48 hours after plasmid cDNA, or 72 hours after siRNA transfection. 1 2

WT or GLP-1R KO mice (generated in house⁹⁵) were humanely euthanised before inflating the pancreas through the common bile duct with RPMI-1640 medium containing 1 mg/mL collagenase (Nordmark Biochemicals). The pancreas was then dissected, incubated at 37 °C for 10 minutes, and digestion halted with cold RPMI-1640 supplemented with 10% FBS (F7524, Sigma–Aldrich) and 1% (v/v) penicillin/streptomycin solution (15070-063, Invitrogen). Mouse islets were subsequently washed, purified using a Histopaque gradient (Histopaque-1119 and Histopaque-1083, Sigma–Aldrich) and rested overnight at 37 °C in a 95% O₂/5% CO₂ incubator before performing experiments in mouse islet culture medium (RPMI-1640 with 11 mM D-glucose, supplemented with 10% FBS and 1% penicillin/streptomycin).

Human islets were received from the isolation unit at the Centre Européen d'Etude du Diabète (CEED), Strasbourg, or from isolation centres ascribed to the Integrated Islet Distribution Program (IIDP), National Institute of Diabetes and Kidney Diseases (NIDDK), National Institutes of Health (NIH), USA, with all ethical approvals in place. Human islets were cultured on arrival in RPMI-1640 medium supplemented with L-glutamine, 10% FBS, 1% penicillin/streptomycin solution, 0.25 μg/mL amphotericin B, and 5.5 mM glucose at 37 °C in a 95% O₂/5% CO₂ incubator. Donor characteristics are indicated in Supplementary Table 3.

Mouse islets were transduced with Control or SPHKAP shRNA-expressing lentiviral particles (MISSION® shRNA, Target Clone ID TRCN0000088765 mouse SPHKAP, Sigma–Aldrich) at an MOI of 1 and cultured for 72 hours before experiments.

INS-1 832/3 cells previously transfected with the relevant plasmids, INS-1 832/3 SNAP/FLAG-hGLP-1R or SNAP/FLAG-hGIPR cells were seeded in 6-well plates at 1.2 × 10⁶ cells per well and incubated overnight. Cells were treated in normal media at 37 °C with vehicle, 100 nM exendin-4 for 5 minutes or 100 nM GIP for the indicated times, washed once with ice cold PBS and lysed with 1 mL lysis buffer supplemented with 1% phosphatase inhibitor cocktail (Sigma-Aldrich) and 1% complete EDTA-free protease inhibitor cocktail (Sigma–Aldrich) on ice. A cross-linking step [2 mM 3,3′-Dithiodipropionic acid di(N-hydroxysuccinimide ester) (DSP, Sigma-Aldrich) for 30 minutes at room temperature followed by quenching with 20 mM Tris HCl for 15 minutes and 3× PBS washes] was performed prior to cell lysis for the SPHKAP-EGFP co-immunoprecipitation experiments to preserve protein-protein interactions despite proteolytic cleavage of the full-length fusion protein during the IP process. Cell lysates were centrifuged at 17,000 × g for 10 minutes, and supernatants added to 40 μL anti-FLAG M2 Magnetic beads (Sigma–Aldrich), anti-HA Magnetic beads (Thermo Fisher Scientific), or ChromoTek GFP-Trap® Magnetic Particles (Proteintech), and incubated at 4 °C for 1 hour. Beads were then washed 3x in 500 μL wash buffer and proteins eluted with urea buffer (200 mM Tris-HCl, pH 6.8; 5% w/v SDS, 8 M urea, 100 mM dithiothreitol, 0.02% w/v bromophenol blue) at 37 °C for 10 minutes before SDS-PAGE and western blot analysis.

Samples were loaded onto 10% SDS-polyacrylamide gels then transferred to 0.45 μm PVDF membranes and blocked with 5% skimmed milk [or 5% bovine serum albumin (BSA) for anti-phospho antibodies] in TBS buffer supplemented with 0.1% Tween 20 (TBS-T). Membranes were incubated with primary antibody overnight diluted in TBS-T/milk (or TBS-T/BSA for anti-phospho antibodies) at 4 °C, washed in TBS-T and incubated for 1 hour at RT with the corresponding HRP-conjugated secondary antibody in TBS-T/milk. Following a further 3× washes in TBS-T, membranes were developed with Clarity Western ECL substrate (BioRad) and imaged on a Biorad ChemiDoc or in a Xograph Compact X5 processor with Fuji medical X-ray films. Specific band densities were quantified in Fiji. The following antibodies were used (source, catalogue number, dilution): rabbit anti-VAPB (Antibodies.com, A14703, 1:5000); rabbit anti-SNAP tag (New England Biolabs, P93105, 1:5000); mouse anti-SPHKAP³² (kind gift from Dr Nicholas Vierra, University of Utah, L131/17; 1:1000); rabbit anti-SPHKAP³² (kind gift from Dr Nicholas Vierra, University of Utah, 36690, 1:1000); mouse anti-Tubulin (Sigma-Aldrich, T-5168, 1:2000); rabbit anti-HA tag (Abcam, Ab9110, 1:1000); rabbit anti-GFP (Proteintech, 50430-2-AP, 1:2000); rabbit anti-phospho-PKA Substrate (RRXS*/T*) (Cell Signaling, 9624, 1:1000); rabbit anti-Drp1 (D6C7) (Cell Signaling, 8570S, 1:1000); rabbit anti-phospho-Drp1 Ser616 (D9A1) (Cell Signaling, 4494S, 1:1000); mouse anti-human Tom20 (F-10) (Santa Cruz Biotechnology, sc-17764, 1:1000); goat anti-mouse IgG-HRP (Abcam, ab205719, 1:5000); goat anti-rabbit IgG HRP (Abcam, ab205718, 1:2000).

INS-1 832/3 SNAP/FLAG-hGLP-1R cells seeded onto 35-mm glass-bottom dishes (MatTek) were transfected with the indicated constructs. SNAP-FLAG-hGLP-1R was labelled for 15 minutes at 37 °C with SNAP-Surface Alexa Fluor 546 or 647 in full media prior to stimulation with 100 nM exendin-4 and imaging in a Leica Stellaris 8 inverted confocal microscope with a 63×/1.40 NA oil immersion objective with Lightning confocal super resolution modality from the facility for imaging by light microscopy (FILM) at Imperial College London. Mouse islets were transduced 24 hours prior to imaging with an adenoviral SNAP/FLAG-hGLP-1R vector (custom made by VectorBuilder) at a MOI of 1 and subsequently labelled for 10 minutes at 37 °C with SNAP-Surface Alexa Fluor 647 before 5-minute stimulation with indicated agonists and imaging of whole islets in the same microscope as above. Images were analysed in Fiji.

INS-1 832/3 SNAP/FLAG-hGLP-1R cells were seeded onto 18-mm coverslips, labelled with SNAP-Surface Alexa Fluor 647 and stimulated with 100 nM exendin-4 as above. Cells were then fixed in 4% paraformaldehyde and processed for immunofluorescence with primary rabbit anti-VAPB antibody (Sigma-Aldrich, HPA013144, 1:150) followed by secondary anti-rabbit FLUX 680 antibody (custom-made at Abberior, 1:200). Directly before MINFLUX imaging, samples were incubated with 150 nm gold fiducials (BBI solutions) for 5 minutes. After three brief washes with PBS, the samples were incubated with 10 mM MgCl₂ in PBS for a further 5 minutes and then mounted on a well slide in freshly prepared GLOX buffer (100 mM Tris-HCl pH 8, 10% w/v glucose, 0.4 mg/mL glucose oxidase, 64 μg/mL catalase) supplemented with 10 mM MEA and sealed with two part silicon (eco-sil speed, Picodent).

MINFLUX imaging was performed on an Abberior instruments 3D MINFLUX microscope¹⁴. Briefly, the microscope is built on a motorised inverted microscope body (IX83, Olympus) with a CoolLED pE-4000 for epifluorescence illumination. The system is equipped with 405 nm, 485 nm, 561 nm, and 642 nm laser lines, as well as a 980 nm IR laser and xyz piezo (Piezoconcept) for active sample stabilisation. Images were acquired with a 100×/1.45 NA UPLXAPO oil objective (Olympus). During acquisition, the pinhole was set to 0.83 AU. Detection was performed with two APDs in spectral windows Cy5 near (650–685 nm) and Cy5 far (685–740 nm). MINFLUX imaging was performed with the default 2D imaging sequence supplied with the system. In brief, the sequence defines parameters used in the iterative localising process of MINFLUX, culminating in a final iteration with a targeted coordinate pattern (TCP) diameter of 40 nm and minimum photon threshold of 150 photons.

Two-colour spectral separation was performed based on detection channel ratio (DCR) values accounting for background inhomogeneity. DCR values from the HeadStart iterations were weighted averaged by the corresponding effective photon counts at the MINFLUX pattern offset. A two-component Gaussian Mixture Model was then fitted to the resulting values and the assigned probability of belonging to distribution 1 or 2 was retrieved. A localisation was then assigned to one colour if its probability was greater than 95% or otherwise set to a 'not defined' category. All localisations associated with the same molecule were subsequently assigned to the colour category that occurred most frequently within that molecule's trace. Molecules for which most localisations were not defined were not accounted for in the two-colour data. After two-colour separation, localisations were filtered by efo (<80 kHz) and number of localisations per trace (>3). The centroid of each remaining trace was calculated and DBSCAN cluster analysis used to identify centroid clusters (minimum points = 10, epsilon = 150 nm). Nearest neighbour analysis was performed between FLUX 680 and Alexa Fluor 647 clusters using the K-D Tree neighbour search algorithm.

Cells and islets were prepared for TEM and imaged as in ref. ¹⁸. Briefly, for SNAP/FLAG-hGLP-1R gold labelling, INS-1 832/3 SNAP/FLAG-hGLP-1R cells grown on Thermanox coverslips (Agar Scientific), or mouse islets transduced with SNAP/FLAG-hGLP-1R adenoviruses, were labelled with 2 μM cleavable SNAP-Surface biotin probe (kind gift of Dr Ivan Corrêa Jr, New England Biolabs) in full media, followed by incubation with 5 μg/mL NaN₃-free Alexa Fluor 488 Streptavidin, 5 or 10 nm colloidal gold conjugate (Thermo Fisher Scientific) to label SNAP/FLAG-hGLP-1R prior to stimulation with 100 nM exendin-4, fixation and processing for conventional TEM. Cells or islets were fixed at room temperature for 20 minutes in 2% PFA + 2% glutaraldehyde in 0.1 M sodium cacodylate, washed 2× with 0.1 M sodium cacodylate and 1× with ddH₂O, and then post-stained on ice with 2% osmium tetroxide (OsO₄) for 1 hour followed by 1% tannic acid at room temperature for 45 minutes, and washed 2× in 0.05 M sodium cacodylate before stepwise dehydration in 70%, 90%, and 100% EtOH and Epon resin embedding. Cells mounted on Epon stubs and resin-embedded islets in moulds were baked at 65 °C overnight. Samples were trimmed and cut en face with a diamond knife (DiATOME) in a Leica Ultracut UCT 6 ultramicrotome before examination on an FEI Tecnai G2-Spirit TEM. Images were acquired in a charge-coupled device camera (Eagle) and analysed in Fiji. Mitochondrial area, number of cristae per mitochondria and ERMCS length was determined as in ref. ⁹⁶ by manually drawing mitochondria and ERMCSs visible from the TEM micrographs.

INS-1 832/3 cells were co-transfected with 1 µg Venus-VAPB (generated in house from EGFP-VAPB) and 0.5 µg GLP-1R-NanoLuc⁹⁷. NanoBRET assays were carried out as previously described⁹⁷. Briefly, 24 hours after transfection, cells were washed with PBS, detached, and resuspended in NanoGlo Live Cell Reagent (Promega) with furimazine (1:40 dilution in HBSS) and seeded into white 96-well half-area plates. A baseline reading was taken for 5 minutes at 37 °C at 460 nm for NanoLuc emission and at 535 nm for Venus emission in a Flexstation 3 microplate reader. Cells were then stimulated with or without 1 µM of the indicated agonist and luminescent signal read for 30 minutes. Ratio (535/460 nm) BRET readings were normalised to baseline and to average vehicle readings for agonist-specific responses. Area under the curve (AUC) was calculated per agonist response using GraphPad Prism 10.2.1.

INS-1 832/3 cells were seeded onto 35-mm glass-bottom dishes (MatTek) and transfected with enhanced excitation-ratiometric PKA biosensor pcDNA3.1(+)-ExRai-AKAR2 (for global PKA) or pcDNA3-mitoExRai-AKAR2 (for mitochondrial PKA)²⁷. Cells were washed with PBS 24 hours after transfection and imaged in Krebs-Ringer bicarbonate-Hepes (KRBH) buffer (140 mM NaCl, 3.6 mM KCl, 1.5 mM CaCl₂, 0.5 mM MgSO₄, 0.5 mM NaH₂PO₄, 2 mM NaHCO₃, 10 mM Hepes, 0.1% BSA, saturated with 95% O₂/5% CO₂, pH 7.4) supplemented with 11 mM glucose and imaged ratiometrically at excitation wavelengths of 405 and 488 nm and emission of 530 nm using a Zeiss LSM-780 inverted confocal laser-scanning microscope with a 63× oil immersion objective, with images taken every 6 seconds, in a 37 °C heated chamber. A baseline reading was taken for 1 minute followed by stimulation with 1 µM exendin-4 and imaging for 5 minutes before addition of 10 μM forskolin (FSK) + 100 μM isobutyl methylxanthine (IBMX) for the last 2 minutes of acquisition to record maximal responses. AUCs were calculated using GraphPad Prism 10.2.1.

INS-1 832/3 cells were seeded onto 35-mm glass-bottom dishes (MatTek) and incubated overnight with the Green Up cADDis biosensor in a BacMam vector (Montana Molecular) according to the manufacturer's instructions. Cells were washed with PBS and imaged in 11 mM glucose KRBH buffer using a Nikon Eclipse Ti microscope with an ORCA-Flash 4.0 camera (Hamamatsu) and Metamorph software (Molecular Devices) with a 20× air objective and 488 nm laser, with images taken every 6 seconds, on a 37 °C stage. A baseline reading was taken for 1 minute followed by stimulation with 1 µM exendin-4 for 5 minutes, and 10 μM FSK + 100 μM IBMX addition for the last 2 minutes of the acquisition for maximal responses. Raw fluorescent intensities were extracted in Fiji to plot an average intensity trace and AUCs calculated in GraphPad Prism 10.2.1.

INS-1 832/3 cells were seeded at 50,000 cells per well in 96-well glass-bottom imaging plates (Sigma–Aldrich) coated with poly-L-lysine and laminin (1:50). Primary mouse islets were dispersed by trituration with trypsin-EDTA 0.25% for 3 minutes and seeded as above, or imaged whole without dispersion. Both cells and islets were stimulated with 100 nM exendin-4 or vehicle for 10 minutes and subsequently labelled with MitoTracker Red CMXRos (Thermo Fisher Scientific), washed 1× with PBS and imaged in phenol red free RPMI-1640 in a Leica Stellaris 8 inverted confocal microscope with 63×/1.40 NA (islets) or 100×/1.40 NA (cells) oil immersion objectives. Five images were collected per condition and analysed using the Mitochondria Analyzer plugin for Fiji to determine mitochondrial mean area and form factor. Cells were also visually classified by predominant mitochondrial phenotype as in refs. ^49,98 into (a) hyperfused (highly interconnected and elongated), (b) tubular (mixed) or (c) fragmented mitochondrial types. For INS-1 832/3 and dispersed islet cell experiments, 30–130 cells were analysed from n = 4, 5 independent repeats per experiment; for intact islet experiments, all visible cells in a total of 15–30 islets from n = 5, 6 independent repeats were analysed per experiment.

INS-1 832-3 cells were seeded in glass-bottom dishes (MatTek), labelled with 500 nM PKmito Orange FX (Spirochrome) for 30 minutes, washed 3× in PBS and incubated in complete media for 30 minutes to recover. Cells were then treated with vehicle or 100 nM exendin-4 in complete media for 10 minutes prior to fixation in 4% PFA and imaging in PBS using a Leica Stellaris 8 microscope in TauSTED mode with a 100×/1.40 NA oil immersion objective. Cells were excited at 591 nm and pulsed depletion performed at 775 nm with 750 ps delay at 50% intensity.

INS-1 832/3 cells were seeded in 35-mm glass-bottom dishes (MatTek), incubated in 3 mM glucose KRBH for 1 hour, followed by incubation with 50 µM tetramethylrhodamine (TMRE) in 3 mM glucose KRBH for 15 minutes to label active mitochondria⁹⁹. Cells were then washed with KRBH and imaged in 3 mM glucose KRBH with images taken every 12 seconds in a 60×/1.4 NA oil immersion objective with a 561 nm laser in a Nikon Eclipse Ti microscope with an ORCA-Flash 4.0 camera (Hamamatsu) and Metamorph software (Molecular Devices). A baseline reading was recorded for 1 minute followed by stimulation with 1 µM exendin-4 in 6 mM glucose KRBH for 2 minutes. Glucose levels were then increased to 20 mM and cells imaged for a further 5 minutes, followed by mitochondrial hyperpolarisation using 20 µg/mL oligomycin for 5 minutes and depolarisation using 20 µM carbonyl cyanide 4-trifluoromethoxyphenylhydrazone (FCCP) for 5 minutes. Raw fluorescent intensities were extracted in Fiji to obtain average intensity traces and AUCs calculated using GraphPad Prism 10.2.1.

INS-1 832/3 cells were seeded in 6-well plates at 1.2 × 10⁶ cells per well. For VAPB WT versus P56S experiments, cells were co-transfected with 40 ng proinsulin-NLuc and 200 ng of either pEGFP-VAPB, pEGFP-VAPB-P56S [generated in house by site directed mutagenesis from pEGFP-VAPB with Fwd 5'-CGATGATTCCGCTGTTGCTCCTCACACAGTACCTAC-3' and Rev 5'- GTAGGTACTGTGTGAGGAGCAACAGCGGAATCATCG-3' mutagenesis primers and the Quick-Change Lightning kit (Agilent)], or pcDNA3.1+ as a control. For SPHKAP-EGFP experiments, cells were co-transfected as above with proinsulin-NLuc and either empty EGFP or SPHKAP-EGFP. For RNAi experiments, cells were transfected with Control versus Target (VAPB or SPHKAP) siRNA. Cells were incubated overnight in INS-1 832/3 media at 3 mM glucose 24 hours after cDNA and 48 hours after siRNA transfections and then pre-incubated for 30 minutes in KRBH buffer supplemented with 3 mM glucose before stimulation in KRBH buffer supplemented with 11 mM glucose ± 100 nM exendin-4 for 60 minutes. Cell supernatants were collected to assess secreted insulin and cells lysed in KRBH + 0.1% Triton X-100 with sonication for total insulin extraction. For cDNA-transfected cells, NanoGlo Live Cell Reagent with furimazine (1:40 dilution in HBSS) was added and relative luciferase units (RLU) recorded at 460 nm using a Flexstation 3 microplate reader.

Mouse islets were incubated in KRBH buffer supplemented with 3 mM glucose for 1 hour followed by stimulation in KRBH buffer supplemented with 11 mM glucose ± 100 nM exendin-4 for 30 minutes. Supernatants were collected and centrifuged at 150 × g for 3 minutes. Islets were lysed in acidic ethanol (75% ethanol, 15 mM HCl), sonicated 3 × 10 seconds in a water bath and centrifuged at 17,000 × g for 10 minutes to determine intracellular insulin concentrations.

For RNAi-transfected cells and shRNA-transduced islets, sample insulin concentration was determined using the Insulin Ultra-Sensitive HTRF Assay kit (Cisbio) according to the manufacturer's instructions.

INS-1 832/3 SNAP/FLAG-hGLP-1R cells seeded onto coverslips were incubated with or without 5 µM thapsigargin in vehicle or 100 nM exendin-4-supplemented media for 8 hours before fixation and terminal deoxynucleotidyl transferase dUTP nick-end labelling (TUNEL) assay performed using the DeadEnd^TM Fluorometric TUNEL system (Promega) following the manufacturer's instructions. Coverslips were mounted with ProLong™ Diamond Antifade Mountant with DAPI (Thermo Fisher Scientific) and imaged for fluorescein and DAPI (to assess total cell number) in a Zeiss LSM-780 inverted confocal laser-scanning microscope with a 20×/0/8 NA air objective. The percentage of TUNEL-positive cells was calculated in Fiji from 5 regions per coverslip.

Mouse and human islets were stimulated with vehicle or 100 nM exendin-4 in full media for 10 minutes at 37 °C. Islets were washed with PBS and fixed for a minimum of 1 hour. Islets were washed again with PBS and processed by PLA with the NaveniFlex Tissue MR Red kit (Navinci) following the manufacturer's instructions, with an extra permeabilization step. Antibody pairs for mouse islets were mouse anti-mouse GLP-1R Ab Mab 7F38 (DSHB, 5 µg/mL) and either rabbit anti-VAPB (A14703, Antibodies.com, 1:100), or rabbit anti-mouse SPHKAP (36690, kind gift from Dr Nicholas Vierra, University of Utah³², 1:100); antibody pairs for human islets were mouse anti-human GLP-1R Mab 3F52 (DSHB, 5 µg/mL) and either rabbit anti VAPB (A14703, Antibodies.com, 1:100), or rabbit anti-human SPHKAP (HPA042499, Atlas Antibodies, 1:500). For insulin co-staining, islets were further incubated with anti-insulin antibody (IR002, Dako Agilent Technologies, undiluted) overnight followed by secondary anti-guinea pig Alexa Fluor 647 (A21450, Life Technologies, 1:200) for 1 hour. Islets were imaged using the confocal settings on a Leica Stellaris STED FALCON microscope and a 40×/0.8 air objective.

The CLEM protocol was adapted from ref. ¹⁰¹. Briefly, INS-1 832/3 SNAP/FLAG-hGLP-1R cells were seeded on gridded coverslips (MatTek) and transfected with EGFP-VAPB before labelling with MitoTracker Red CMXRos and SNAP-Surface Alexa Fluor 647 and stimulation with 100 nM exendin-4 for 5 minutes. Cells were initially fixed for confocal microscopy with 4% PFA + 0.1% glutaraldehyde, and confocal stacks acquired with a C2+ confocal microscope (Nikon) and a 60x oil immersion objective. After fluorescence imaging, cells were additionally fixed for TEM with 4% PFA + 2.5% glutaraldehyde in sodium phosphate buffer. Samples were further processed according to a protocol modified from ref. ¹⁰² using OsO₄ and thiocarbohydrazide (TCH). In brief, samples were incubated in a 2% aqueous OsO₄ solution containing 1.5% potassium ferrocyanide and 2 mM CaCl₂ (30 minutes on ice), washed in 1% TCH, rinsed in ddH₂O and incubated again in 2% OsO₄. After washing, samples were stained en bloc with 1% uranyl acetate (UA), rinsed in ddH₂O and dehydrated in a graded series of ethanol. Samples were embedded into epoxy resin Embed812 (Science Services). After resin curing at 60 °C, the coverslip was removed by dipping into liquid nitrogen. The block was trimmed to a region identified during confocal imaging and serial sections with a thickness of 300 nm cut. Section post-staining was performed with 2% UA in methanol and 1% lead citrate in ddH₂O. After application of 15 nm colloidal gold beads, sections were imaged in a 300 kV Thermo Fisher F30 electron microscope. Tilt series were acquired at steps of 1° ranging from −63° to +63°. Tomograms were reconstructed with the IMOD software package¹⁰³. Correlation of confocal images and tomogram slices was done with the EC-CLEM plugin¹⁰⁴ in Icy¹⁰⁵. Segmentation of the tomograms was performed manually with Microscopy Image Browser¹⁰⁶. 3D rendering of the segmentation was performed with ORS Dragonfly (www.theobjects.com/dragonfly/index.html↗).

INS-1 832/3 SNAP/FLAG-hGLP-1R cells previously transfected with EGFP-VAPB were plated onto glow discharged (30 seconds, 20 mA) poly-L-lysine-coated Quantifoil R2/2 200 mesh gold EM grids (Electron Microscopy Sciences). Cells were labelled with 500 nM PKmito Orange FX (Spirochrome) and SNAP-Surface Alexa Fluor 647 before stimulation with 100 nM exendin-4 in complete media for 10 minutes. Grids were subsequently plunge-frozen in condensed ethane using a Vitrobot and stored in liquid N₂ for further analysis.

The clipped plunge-frozen grids were loaded into the Zeiss correlative cryo-holder which can be transferred into the cryo-LSM and the cryo-FIB-SEM without further removal from the holder. For cryo-confocal data acquisition, the cryo-holder was transferred onto the cryo-Linkam stage (Linkam Scientific) and mounted onto a ZEISS Axio Imager.Z2 fitted with a LSM900 confocal scan head equipped with 3 confocal channels, including the Airyscan 2 detector for super resolution. Low magnification objectives (C Epiplan-Apochromat 5×/0.2 NA DIC, C Epiplan-Apochromat 10×/0.4 NA DIC) were used for overviews, and an LD EC Epiplan-Neofluar 100×/0.75 NA DIC used for high resolution. Images were acquired using the following settings: 488 nm, 561 nm and 640 nm excitation with emission bands of 493–551 nm, 553–612 nm and 613–700 nm. Reflection and brightfield channels were additionally acquired for correlation of light microscopy with SEM images. Confocal images were processed with LSM Plus (linear Wiener deconvolution). For highest resolution, Airyscan detection was used with the following settings: 488 nm, 561 nm and 640 nm excitation with emission bands of 492–549 nm, 564–611 nm and 644–700 nm and images processed using the Airyscan joint deconvolution algorithm. For cryo FIB-SEM data acquisition, the correlative cryo-holder was transferred into the Zeiss Crossbeam 550 equipped with a Quorum cryo stage. The transfer is done through the Quorum Prep Chamber attached to the FIB-SEM chamber. Inside the chamber, the sample surface was sputter-coated with platinum with a current of 5 mA for 45 seconds to make the surface conductive for SEM imaging and FIB milling. After insertion, the FIB-SEM coordinate system was aligned with respect to the LSM dataset using the ZEN software (Zeiss) and ROIs acquired in the LSM relocated and imaged in the FIB-SEM. The sample was then tilted to 54° (perpendicular to the FIB) and a cold platinum deposition applied to minimize the curtaining effect (vertical stripes in the SEM images that result from uneven milling of the sample due to roughness of the sample surface). For this purpose, the gas injection system (GIS) nozzle was inserted at ∼2.5 mm from the sample surface and the unheated GIS opened for about 60 seconds resulting in a protection layer with a thickness of ∼0.5 µm. Trenches were milled at the ROIs with a FIB current of 3 nA (Gallium ion energy 30 keV) to generate cross-sections. FIB-SEM volume imaging was performed with a milling current of 300 pA (Gallium ion energy 30 keV) using a slice thickness of 30 nm to obtain SEM image stacks representing a volume of 16 × 12 × 9 µm (XYZ). The SEM images were acquired with an electron energy of 2.3 keV and an electron current of 30 pA with a pixel size of 8 nm using the InLens SE detector. The resulting FIB-SEM datasets were imported into ZEN project. After alignment of the image stack, images were processed using "destriping" (ZEN) based on the VSNR algorithm¹⁰⁷. Processed image stacks were cropped to an appropriate size and a 3D alignment of the SEM and LSM stacks performed. A selected area in the aligned region of the FIB-SEM data was manually segmented, slice by slice, in Fiji, using the freehand selection tool. Outlines were saved as masks, and 3D reconstruction performed in Icy¹⁰⁵.

ROS was determined using the ROS-Glo™ H₂O₂ Assay (G8820, Promega) according to the manufacturer's instructions. Briefly, islets were incubated overnight in full media containing 1 µM thapsigargin ± 100 nM exendin-4 at 37 °C and subsequently incubated in 25 µM H₂O₂ Substrate diluted in H₂O₂ Substrate Dilution Buffer plus 1 µM thapsigargin ± 100 nM exendin-4 in a final volume of 100 µL for 3 hours at 37 °C in a 5% CO₂ incubator. Media (50 µL) was combined with 50 µL ROS-Glo™ Detection Solution in a 96-well opaque plate and incubated for 20 minutes at room temperature. Relative luminescence units (RLU) were recorded in a Flexstation 3 plate reader and results normalised to islet cell counts.

INS-1 832/3 cells were transfected with the dual-excitation ratiometric biosensor mKeima-Red-Mito-7 and seeded onto 35-mm glass-bottom dishes (MatTek). Cells were imaged 24 hours after seeding in RPMI-1640 buffer without phenol red ± 100 nM exendin-4 by time-lapse microscopy every 12 seconds for 11 minutes in a 37 °C heated chamber using a Leica Stellaris 8 inverted confocal microscope with a 40×/1.2 NA oil immersion objective with excitation wavelengths of 440 and 580 nm and emission wavelength of 620 nm. Ratio fluorescence intensities for the two excitation wavelengths (580/440 nm) was calculated to obtain a response trace as in ref. ¹⁰⁸.

INS-1 832/3 cells were transfected with pCAG G-CEPIA1er (for ER calcium) or pCAG mito-RCaMP1h (for mitochondrial calcium)¹⁰⁹. Twenty four hours after transfection, cells were washed 1× with PBS and imaged by time-lapse microscopy in RPMI-1640 without phenol red in a 40×/1.2 NA oil immersion objective Nikon Eclipse Ti microscope with an ORCA-Flash 4.0 camera (Hamamatsu) and Metamorph software (Molecular Devices). Mitochondrial calcium was imaged with 561 nm, and ER calcium with 488 nm excitation wavelengths, with images taken every 3 seconds. A baseline reading was recorded for 1 minute followed by 1 µM exendin-4 stimulation for 3 minutes. Raw fluorescent intensities were extracted using Fiji to obtain average intensity traces and AUCs calculated using GraphPad Prism 10.2.1.

2 × 10⁷ INS-1 832/3 SNAP/FLAG-hGLP-1R cells were seeded in a T75 flask in complete media and cultured overnight. Cells were then stimulated in complete media containing 100 nM exendin-4 for 10 minutes. The crude mitochondrial fraction containing ER-associated membranes was isolated using the mitochondria isolation kit for cultured cells (89874, Thermo Fisher Scientific) according to the manufacturer's protocol. MAMs were subsequently separated from mitochondria by ultracentrifugation at 95,000 × g for 30 minutes in an OptiPrep™ Density Gradient Medium (Sigma–Aldrich). The MAM fraction was diluted 1:10 in mitochondrial resuspension buffer (250 mM mannitol, 5 mM HEPES, 0.5 M EGTA, pH 7.4), pelleted by ultracentrifugation at 100,000 × g for 1 hour and prepared for western blot analysis as described above.

pcDNA3-GFP11(x7)-VAPB was cloned in house from pEGFP-C1-hVAPB and pcDNA3-GFP11(x7)-Actin by PCR amplification of human VAPB with addition of 5'-BglII and 3'-XhoI restriction sites using Fwd 5'-GTACTCAGATCTCGAGCTCAAGC-3' and Rev 5'- CTCATCTCGAGGAATTGCTACAAG-3' primers. The second BglII site at position 7107 in pcDNA3-GFP11(x7)-Actin was removed by site directed mutagenesis using Fwd 5'- TGACGTCGACGGATCGGGAAATATCCCGATCCC-3' and Rev 5'- GGGATCGGGATATTTCCCGATCCGTCGACGTCA-3' primers with the Quick-Change Lightning kit (Agilent) prior to in frame cloning of human VAPB using T4 DNA ligase (Promega).

The FluoSTEP system was used to determine VAPB-localised cAMP and PKA signalling as in ref. ²⁵. INS-1 832/3 cells were transfected with pcDNA3-GFP11(x7)-VAPB and pcDNA3.1(+)-FluoSTEP-ICUE (for cAMP) or pcDNA3.1(+)-FluoSTEP-AKAR (for PKA) at a 1:1 ratio. Cells were seeded onto glass-bottom dishes (MatTek) 24 hours before imaging in RPMI-1640 without phenol red + 10% FBS. Cells were imaged by time-lapse microscopy using a Zeiss LSM-780 inverted confocal laser-scanning microscope with a 63×/1.4 NA oil immersion objective and a 488 nm excitation wavelength and emissions at 508 and 583 nm for 2 minutes to record a baseline, followed by agonist stimulation at 100 nM for 5 minutes and 2-minute exposure to 10 μM FSK + 100 μM IBMX to generate a maximal response. FRET emission ratios were calculated and normalised to mean baseline values using GraphPad Prism 10.2.1.

INS-1 832/3 cells were plated onto Matrigel (Gibco) coated glass-bottom imaging chambers (Ibidi) and transfected with 4xMTS-Halo, ER-RspA-NFAST and OMM-short-RspA-CFAST⁵⁰. After 18 hours, cells were labelled with 10 nM JFX646 Halo ligand (kind gift from the Lavis lab, Janelia) in optiMEM (Gibco) for 3 minutes, followed by 3× washes in mKRB buffer (140 mM NaCl, 2.8 mM KCl, 2 mM MgCl₂, 1 mM CaCl₂, 10 mM HEPES, 10 mM glucose, pH 7.4) at 37 °C. The Lime splitFAST ligand (Twinkle Factory) was prepared at 5 μM in mKRB buffer and passed through a 0.22 μm filter before being added to cells. Exendin-4 was used at 1 μM in Lime splitFAST ligand in mKRB and added to cells for 10 minutes prior to imaging with an LSM980/Airyscan2 point-scanning confocal microscope (Zeiss) in LSMPlus mode. Simultaneous excitation was achieved using solid-state diode lasers at 488 nm/10 mW (0.9% max power) and 639 nm/7.5 mW (1.0% max power), with imaging conditions experimentally selected to minimise crosstalk. The resulting fluorescence was collected using a 63×/1.4 NA oil immersion objective and detected on a 34-channel spectral GaAsP array detector with the gain set at 650 V and split into two ranges of 490–632 nm and 641–694 nm. Samples were imaged at room temperature in a humidified imaging chamber with a Definite Focus module (Zeiss) employed for thermal drift correction and ZEN Blue v3.2 (Zeiss) software used for acquisition and deconvolution (LSMPlus function). The number of contact sites per mitochondrial area was quantified in Fiji. Firstly, the acquired images were Gaussian blurred (sigma radius 1.5) to aid in the generation of reliable downstream masks and puncta detection. An automated Otsu threshold was used to generate mitochondrial masks and quantitate mitochondrial area. These same masks were then applied to the splitFAST channel to ensure that only true contact sites (i.e. puncta on mitochondria) were counted in the analysis. A blinded count of total number of ERMCSs in each cell was taken and normalised to the mitochondrial area.

Our research complies with all relevant ethical regulations. Human islets were obtained from authorised islet isolation facilities at CEED and IIDP, with all ethical approvals in place and informed relative consent obtained. These islets have been refused for transplantation, have been quality-controlled, and meet specific criteria for research purposes. Mouse islets were purified from mice bred at the Central Biological Services unit of Imperial College London. All animal procedures were approved by the British Home Office under the UK animals (Scientific Procedures) Act 1986 (Project License number PP7151519 to Dr Aida Martinez-Sanchez, Imperial College London, UK) and from the local ethical committee (Animal Welfare and Ethics Review Board) at the Central Biological Services unit of Imperial College London.

All statistical analyses and graph generation was performed using GraphPad Prism 10.2.1 (Dotmatics). Statistical tests used are indicated in the corresponding figure legends. The number of replicates per comparison represents biological replicates, with technical replicates within biological replicates averaged before statistical testing unless indicated. Data are presented as mean ± SEM. The p value threshold for statistical significance was set at 0.05.

To compare the effects of SPHKAP gene region variants on T2D versus BMI, we used GWAS meta-analyses summary statistics from DIAMANTE consortium (http://www.diagram-consortium.org↗) for T2D by Mahajan et al.³⁶, and from GIANT↗ for BMI by Locke et al.³⁹. We report SNPs from the region surrounding SPHKAP gene (Chr2:228,844,666-229,046,361, NCBI human genome build 37). Data pre-processing involved merging and aligning variant effects to T2D risk-increasing allele. We extracted β values for BMI and T2D signals and calculated respective odds ratios (OR) for T2D using the formula OR = exp(β). A scatter plot of SNPs with at least nominally significant (p < 0.05) effects on both BMI and T2D was generated using RStudio with the ggplot2 package, and Pearson correlation between BMI and T2D-associated signals calculated in R using the cor() function.

Statistical analysis of association between random blood glucose levels (RG) and coding DNA variants from whole-exome sequencing UKBB European ancestry individuals' data for SPHKAP gene variants was performed in PLINK2.0¹¹² assuming an additive genetic model. The model included covariates age, sex, time since last meal (t), accounted for as t, t2 and t3, and six principal components. Individuals were excluded if they had self-reported diabetes or diabetes medication, were pregnant, or had RG ≥20 mmol/L. Chromosome positions are given according to NCBI human genome build 38. The RG phenotype definition is provided in detail in ref. ¹¹³.

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