The endoplasmic reticulum–plasma membrane tethering protein TMEM24 is a regulator of cellular Ca²⁺ homeostasis

Lipid exchange between the endoplasmic reticulum (ER) and the plasma membrane (PM) is facilitated by lipid transport proteins that are concentrated to, and participate in the formation of, ER–PM contact sites. These junctions are also important sites for cellular Ca²⁺ homeostasis, and the lipid transport is often coupled to changes in the cytosolic Ca²⁺ concentration (Balla, 2018; Chung et al., 2017; Saheki and De Camilli, 2017). The changes in membrane lipid concentrations occurring as a consequence of lipid transport may also influence Ca²⁺ influx or extrusion through modulation of Ca²⁺ channel activity or clustering (Johnson et al., 2018; Suh et al., 2010; Sun et al., 2019; Xie et al., 2016). TMEM24 is a recently characterized lipid transport protein that localizes to ER–PM contacts where it, through an N-terminal synaptotagmin-like mitochondrial-lipid-binding (SMP) domain, provides the PM with phosphatidylinositol, the precursor of the signaling lipids phosphatidylinositol-4-phosphate [PI(4)P] and phosphatidylinositol 4,5-bisphosphate [PI(4,5)P₂] (Lees et al., 2017; Sun et al., 2019). It binds negatively charged lipids in the PM via a C-terminal polybasic region, and neutralization of positively charged amino acids in this region by protein kinase C (PKC)-dependent phosphorylation results in TMEM24 dissociation from the PM. TMEM24 is also equipped with a C2 domain, which, however, seems dispensable for both PM binding and lipid transport. The spatial regulation of TMEM24 activity differs from other SMP- and C2-domain proteins, such as the extended synaptotagmins, whose interaction with the PM and lipid transport are instead triggered by Ca²⁺ elevation (Bian et al., 2018; Giordano et al., 2013; Idevall-Hagren et al., 2015).

Pancreatic β-cells are secretory cells that produce insulin and release this hormone to the circulation in response to elevated blood glucose levels. The mechanism controlling insulin secretion is well-characterized and involves glucose uptake and metabolism, resulting in an elevated ATP/ADP ratio which, in turn, closes ATP-sensitive K⁺ (K_ATP)-channels, causing membrane depolarization, opening of voltage-dependent Ca²⁺ channels, Ca²⁺ influx and the fusion of insulin-containing granules with the PM (Rorsman and Ashcroft, 2018). Ca²⁺ is the most important trigger of insulin secretion, but the response can also be modulated by many factors, including lipids and lipid-derived signaling molecules. Phosphoinositides in particular have been shown to regulate insulin secretion at several stages, including membrane depolarization, Ca²⁺ influx and granule docking and release (Wuttke, 2015). The phosphoinositide PI(4,5)P₂ also serves as a precursor of inositol 1,4,5-trisphosphate (IP3), which triggers Ca²⁺ release from the ER and may contribute to insulin granule exocytosis, and diacylglycerol (DAG), which together with Ca²⁺ amplifies secretion by stimulating the PKC activity (Wuttke et al., 2013, 2016). Lipid transport at ER–PM contact sites is important for the normal function of insulin-secreting β-cells. DAG transport by E-Syt1 was recently found to provide negative feedback on insulin secretion after being recruited to sites of Ca²⁺ influx, where it locally clears the PM of the pro-secretory lipid (Xie et al., 2019). β-cells with reduced E-Syt1 expression thus exhibit increased accumulation of PM DAG and excess insulin secretion in response to glucose. E-Syt1 occupies the same contact sites as TMEM24. However, their presence at these contacts does not overlap in time due to their opposite dependence of PM binding on the prevailing Ca²⁺ concentration (Xie et al., 2019). Interestingly, TMEM24 was recently found to be indispensable for glucose-stimulated insulin secretion (Lees et al., 2017; Pottekat et al., 2013). The complete loss of insulin secretion in clonal TMEM24-knockout (KO) β-cells is likely caused by an effect on β-cell Ca²⁺ homeostasis. β-cells lacking TMEM24 thus exhibit markedly suppressed glucose-induced Ca²⁺ oscillations, but the mechanism behind this is not clear (Lees et al., 2017). One possibility is that TMEM24 is necessary for maintaining the PM PI(4,5)P₂ level, which is required for ion channel gating during stimulation, although the global PI(4,5)P₂ levels are unaltered in TMEM24 KO cells (Lees et al., 2017). The dynamics of TMEM24 in glucose-stimulated β-cells is also difficult to reconcile with a role as a positive regulator of insulin secretion, since it is spatially separated from this process during glucose stimulation. Clarification of the role of TMEM24 in the regulation of insulin secretion is therefore required.

We now show that the subcellular distribution of TMEM24 is highly dynamic and influenced by both Ca²⁺ and PM DAG. In contrast to previous studies, we do not find an absolute requirement of TMEM24 for normal Ca²⁺-triggered insulin secretion. Instead, we show that TMEM24 supports sustained insulin secretion by regulating ER and mitochondrial Ca²⁺ handling and mitochondrial energy production.

To better understand the role of TMEM24 in the regulation of β-cell function, we investigated conditions that promote TMEM24 dissociation from the PM. Elevation of the glucose concentration from 3 to 20 mM in clonal MIN6 β-cells loaded with the Ca²⁺ indicator Cal590 resulted in regular cytosolic Ca²⁺ oscillations with elevations mirrored by synchronized TMEM24–GFP dissociations, seen as reductions in PM-proximal fluorescence by total internal reflection fluorescence (TIRF) microscopy (Fig. 1A–C), similar to what has previously been reported (Lees et al., 2017). Direct depolarization with 30 mM KCl also caused dissociation of TMEM24 from the PM, consistent with voltage-dependent Ca²⁺ influx being the trigger for dissociation (Fig. 1B–D). Similar TMEM24 dissociation was also seen in response to carbachol, which activates phospholipase C and triggers IP3-mediated release of Ca²⁺ from the ER (Fig. 1B,C). Passive depletion of ER Ca²⁺ using the sarcoendoplasmic reticulum (SR) Ca²⁺ transport ATPase (SERCA) inhibitor cyclopiazonic acid (CPA) also triggered TMEM24 dissociation, although less prominently than in response to carbachol (Fig. 1B,C). Glucose-induced TMEM24 dissociation has previously been shown to depend on PKC-mediated phosphorylation of C-terminal residues in the molecule (Lees et al., 2017). Although Ca²⁺ is a potent activator of PKC, some isoforms also require DAG. To determine to what extent DAG is involved in the spatial control of TMEM24, we stimulated MIN6 cells with 1 μM of the DAG analog phorbol 12-myristate 13-acetate (PMA). This resulted in an immediate dissociation of TMEM24 from the PM without apparent change in the cytosolic Ca²⁺ concentration (Fig. 1B,C; Fig. S1A,B). To further uncouple DAG formation from Ca²⁺ changes, we first exposed cells to CPA to deplete ER Ca²⁺ stores, followed by addition of carbachol to activate PLC and increase PM DAG. CPA caused an increase in cytosolic Ca²⁺ and a slight dissociation of GFP–TMEM24 from the PM. The subsequent addition of carbachol resulted in a pronounced dissociation of TMEM24 that occurred in the absence of noticeable changes in cytosolic Ca²⁺ concentration (Fig. S1C–E). These results indicate that TMEM24 is spatially controlled by both Ca²⁺ and DAG. We also noticed that whereas carbachol caused a homogenous dissociation of TMEM24 from the PM (Fig. 1E, right panels), the dissociation in response to direct depolarization with KCl was instead incomplete, more heterogenous and differed between sub-regions of the PM within the same cell (Fig. 1E, left panels). This pattern resembles that of the DAG biosensor GFP–C1aC1b_PKC, which reports DAG formation after autocrine activation of purinergic P2Y1 receptors by ATP co-secreted with insulin from the cells (Wuttke et al., 2013; Fig. S1F,G). Because of the close interplay between DAG and Ca²⁺, we determined the effect of known Ca²⁺ concentrations on TMEM24–mCherry PM binding in α-toxin-permeabilized MIN6 cells in which changes in Ca²⁺ could easily be uncoupled from those in DAG. We found that dissociation of TMEM24 occurred with an EC₅₀ of 358±70 nM (mean±s.e.m.; n=13) and was maximal at 30 μM (Fig. 1F,G). As mentioned above, addition of 1 μM PMA caused pronounced dissociation of TMEM24 from the PM and a subsequent stepwise increase in the Ca²⁺ concentration had little effect on TMEM24 PM binding (Fig. S1H,I). For comparison, the PM binding of E-Syt1–mCherry, another Ca²⁺-dependent ER–PM contact protein, was triggered at 1 μM Ca²⁺ and was maximal at 10 μM Ca²⁺ (EC₅₀ 1.99±0.51 μM, n=6) (Fig. 1G). When comparing the PM fluorescence intensity of TMEM24–mCherry in intact cells and following permeabilization in a Ca²⁺-free buffer, we found that removal of Ca²⁺ promoted further association of TMEM24 with the PM (13±5% increase in PM TMEM24–mCherry fluorescence, n=22, P<0.01) (Fig. 1H) while the opposite was observed for GFP–E-Syt1 (29±4% drop in PM fluorescence, n=31). These results indicate that TMEM24 PM binding might be partially reduced already at resting Ca²⁺ concentrations. Fluorescence recovery after photobleaching (FRAP) analysis revealed that TMEM24–GFP, although enriched at ER–PM contacts, exhibited pronounced dynamics with a large proportion belonging to a mobile fraction (Fig. 1I). Taken together, these results show that the association of TMEM24 with the PM is regulated by both DAG and Ca²⁺, and that dissociation can be triggered by modest nanomolar Ca²⁺ concentrations, resulting in relatively weak interactions between TMEM24 and the PM.

TMEM24 has previously been reported to be required for glucose-stimulated Ca²⁺ influx and insulin release, although the mechanisms are unclear (Lees et al., 2017). To examine its role in more detail, we reduced TMEM24 expression in MIN6 cells by siRNA-mediated transient knockdown (KD), which resulted in a 61.2±6.3% (mean±s.e.m.; n=4, P<0.01) reduction in TMEM24 expression as assessed by western blotting (86.7±5.2% reduction as assessed by qRT-PCR; data not shown) (Fig. 2A). We next determined the impact of reduced TMEM24 expression on Ca²⁺ handling using the fluorescent Ca²⁺ indicator Cal520. In control cells, elevation of the glucose concentration from 3 to 20 mM resulted in rapid lowering of the cytosolic Ca²⁺ concentration (−17.1±0.9%, n=148) due to ATP-driven sequestration of the ion into the ER (Roe et al., 1994), followed after 2.8±0.09 min by a pronounced Ca²⁺ increase and sustained elevation (Fig. 2B–E). Most cells showed slow regular Ca²⁺ oscillations in the continued presence of 20 mM glucose, whereas a smaller number showed more irregular responses (Fig. 2B). In TMEM24 KD cells, the initial Ca²⁺ lowering in response to 20 mM glucose was less pronounced (−12.4±0.8%, n=137, P=0.0016 for comparison to control) and was followed after 3.31±0.11 min (n=141, P=0.003 for comparison to control) by a sustained rise of Ca²⁺ with regular oscillations (Fig. 2B,C,E). The time-average increase in Ca²⁺ was slightly higher in TMEM24 KD cells compared to controls (Fig. 2D). Overexpression of TMEM24 resulted in an exaggerated initial lowering of the cytosolic Ca²⁺ concentration in response to glucose and a slightly increased sustained Ca²⁺ response when compared to non-transfected control cells (Fig. 2C–E). To ascertain that the relatively mild effect on Ca²⁺ handling in TMEM24 KD cells was not due to remaining low TMEM24 expression, we generated TMEM24 KO MIN6 cell lines using CRISPR/Cas9 (Fig. 2F). Both control and TMEM24 KO cells exhibited glucose-induced oscillations in the cytosolic Ca²⁺ concentration, although the time-averaged Ca²⁺ increase was reduced by 21±2% (P=1.47×10⁻⁹) in the TMEM24 KO cells (Fig. 2G–I). Similar to what was seen in TMEM24 KD, the KO cell line also exhibited strongly suppressed initial Ca²⁺ lowering in response to glucose (Fig. 2J,K). Both TMEM24 KD and TMEM24 KO cells exhibited Ca²⁺ responses to glucose that contained a component of rapid Ca²⁺ transients superimposed on top of the regular, slow Ca²⁺ oscillations (Fig. 2B,G).

Given the relatively small effect of TMEM24 KD or KO on glucose-induced Ca²⁺ influx observed here, we decided to reinvestigate the previously proposed fundamental role of TMEM24 in insulin secretion (Lees et al., 2017; Pottekat et al., 2013). Using insulin ELISA, we found that basal secretion at 3 mM glucose and that stimulated by 20 mM glucose were similar in control and TMEM24 KD and KO cells (Fig. 2N). There were also no differences in insulin content between control cells and TMEM24 KD or KO cells (Fig. 2L,M). To investigate the secretion kinetics under conditions more similar to those of primary β-cells within islets of Langerhans, we allowed wild-type and TMEM24 KO cells to aggregate into islet-like cell clusters (pseudo-islets). This preparation enhances the secretory capacity of the MIN6 cells by enabling electrical coupling and upregulating glucose-sensing and metabolic pathways (Chowdhury et al., 2013). Perifusion of wild-type MIN6 pseudo-islets during a step-increase in glucose concentration from 3 mM to 20 mM resulted in an initial peak of insulin secretion (first phase), followed by sustained secretion at a level much above basal secretion (second phase). TMEM24 KO pseudo-islets displayed a similar biphasic secretion pattern but the second phase of secretion was less pronounced than in wild-type cells (Fig. 2O,P). Taken together, these results question the previously reported absolute requirement of TMEM24 for glucose-stimulated insulin secretion, and instead indicate that TMEM24 plays a regulatory role during sustained insulin secretion.

Since we observed a slightly impaired Ca²⁺ response to glucose in TMEM24 KO cells, we tested to what extent Ca²⁺ influx in response to direct depolarization was affected by the loss of TMEM24. The resting cytosolic Ca²⁺ concentration was similar in wild-type and TMEM24 KO cells. Depolarization with 30 mM KCl resulted in an immediate rise of cytosolic Ca²⁺ that averaged 361±5 nM (n=298) in wild-type cells. This response was increased to 441±8 nM (mean±s.e.m.; n=317, P=3.12×10⁻¹⁶) in TMEM24 KO cells (Fig. 3A,B). Consistent with more Ca²⁺ entering the cells in response to depolarization, we also found that insulin secretion in response to acute depolarization was increased by 60% in TMEM24 KO cells (n=12, P=0.019) (Fig. 3G). Similar results were obtained from cells where TMEM24 expression had instead been reduced by siRNA (Fig. 3D–F,H). Several observations made in TMEM24 KO cells, such as lack of initial Ca²⁺-lowering effect of glucose and the appearance of irregular Ca²⁺ spikes, pointed to potential changes in the ability of the ER to sequester and mobilize Ca²⁺. To test this more directly, we mobilized Ca²⁺ from the ER by the SERCA inhibitor CPA in the absence of extracellular Ca²⁺ while measuring changes in the cytosolic Ca²⁺ concentration. The addition of CPA resulted in a 32±1 nM (n=298) increase in the cytosolic Ca²⁺ concentration, reflecting the release of Ca²⁺ from the ER. This release was increased to 48±2 nM (n=306, P=9.8×10⁻¹²) in TMEM24 KO cells (Fig. 3I,J). The addition of 10 mM Ca²⁺ to the extracellular buffer triggered store-operated Ca²⁺ entry, which was not different in the two cell lines (Fig. 3K). Similar results were obtained in cells where TMEM24 expression was reduced by siRNA (Fig. S2) and when ER Ca²⁺-store depletion was instead triggered by the addition of thapsigargin in the presence of extracellular Ca²⁺. Importantly, the response in TMEM24 KO cells returned to normal upon the re-expression of wild-type TMEM24 (Fig. 3L), whereas overexpression of TMEM24 in wild-type cells resulted in reduced release of Ca²⁺ from the ER (Fig. S2). Direct measurements of ER Ca²⁺ using the FRET-based Ca²⁺ sensor D4ER confirmed that the ER of TMEM24 KO cells contained more Ca²⁺ than that of wild-type cells under resting conditions, and also that more Ca²⁺ was released from the ER following SERCA inhibition with CPA (Fig. 3M,N). We speculated that the larger rise of cytosolic Ca²⁺ in response to depolarization in TMEM24 KO cells might be due to simultaneous Ca²⁺-induced Ca²⁺ release from the ER. To test this alternative, we performed experiments where cytosolic Ca²⁺ was measured following two brief applications of 30 mM KCl, where the second application was preceded by SERCA inhibition with thapsigargin. Consistent with previous observations (Chen et al., 2003), prevention of Ca²⁺ sequestration into the ER by SERCA inhibition resulted in a more pronounced depolarization-induced Ca²⁺ increase in wild-type cells (Fig. 3N,O). This augmentation was even more apparent in TMEM24 KO cells, and could be restored to the level of wild-type cells by the re-expression of TMEM24 (Fig. 3P,Q). These results indicate that Ca²⁺ sequestration into the ER may be a way to compensate for excess Ca²⁺ increase in TMEM24 KO cells rather than being the cause of it.

Because Ca²⁺ influx, extrusion and organellar sequestration all depends on ATP generated primarily via mitochondrial metabolism, we decided to complement the Ca²⁺ imaging with direct measurement of glucose metabolism using the Seahorse XF technique. As expected from the Ca²⁺ imaging data, TMEM24 KO had little effect on the resting oxygen consumption rate (OCR) (Fig. 4A). In contrast, the accelerated OCR induced by a rise of the extracellular glucose concentration from 3 to 20 mM was markedly impaired in TMEM24 KO cells (Fig. 4B). However, neither the proton leak nor the maximal OCR was different between wild-type and TMEM24 KO cells, indicating that there was no gross impairment in overall mitochondrial function (Fig. 4C–F). This conclusion was supported by lack of apparent changes in mitochondrial morphology in TMEM24 KO cells as assessed by confocal microscopy of cells expressing mitochondria-targeted mApple (Tom20–mApple) (Fig. 4G,H). Next, we measured the mitochondrial membrane potential in wild-type and TMEM24 KO cells using the fluorescent membrane potential indicator TMRM. Application of the uncoupler FCCP resulted in the immediate loss of TMRM fluorescence, which reflects depolarization of the inner mitochondrial membrane. This response was significantly smaller in TMEM24 KO cells, indicating that these cells already have partially depolarized mitochondria (Fig. 4I,J). The mitochondrial membrane potential not only controls ATP production by directly affecting the electron transport chain but also by regulating the amounts of Ca²⁺ that are taken up and extruded by the organelle. To determine whether TMEM24 might be involved in the regulation of mitochondrial Ca²⁺, we measured Ca²⁺ concentration changes in response to both CPA-mediated release of Ca²⁺ from the ER and depolarization-induced Ca²⁺ influx in wild-type and TMEM24 KO cell using mitochondrially targeted LAR-GECO1.2. This low-affinity sensor has a K_d for Ca²⁺-binding of 12 µM, and its fluorescence is expected to increase little under normal conditions, since depolarization-induced Ca²⁺ influx typically results in Ca²⁺ concentrations in the low µM range (see e.g. Fig. 3B). Consistently, KCl depolarization of wild-type cells triggered a robust increase in the cytosolic Ca²⁺ concentration, measured with the organic dye Cal520, but had little impact on mito-LAR-GECO1.2 fluorescence in the same cells (Fig. 4K,L). In contrast, depolarization caused a pronounced increase in mito-LAR-GECO1.2 fluorescence in TMEM24 KO cells, which was reduced to the level of wild-type cells upon the re-expression of TMEM24 (Fig. 4K,L). Similar results were obtained when the Ca²⁺ increase was instead triggered by passive depletion from the ER through CPA-mediated SERCA inhibition (Fig. 4M,N). A possible explanation for these observations is that the resting mitochondrial Ca²⁺ concentration is higher in TMEM24 KO cells, thus bringing the concentration into a range better suited for detection with the low-affinity sensor. Consistent with this hypothesis, we observed higher resting mito-LAR-GECO1.2 fluorescence in TMEM24 KO cells compared to wild-type cells (Fig. 4O).

We next asked how an ER-localized protein that primarily engages in lipid transport at ER–PM contact sites could impact mitochondrial function. One possibility is that TMEM24 can function at other membrane contact sites. As we show here, a large part of the PM-bound pool of TMEM24 is dynamic even under resting conditions and could therefore participate in reactions at other cellular membranes, such as those of mitochondria. To test this alternative, we expressed GFP-tagged TMEM24 together with mitochondrially localized Tom20–mApple in wild-type MIN6 cells. Using confocal microscopy, we observe enrichment of TMEM24–GFP fluorescence at mitochondria following KCl-induced Ca²⁺ increases that was reversed when the depolarizing agent was removed (Fig. 5A,B). Observing the same cells by TIRF microscopy, which has an axial resolution of ∼100 nm, we could detect overlap between ER-localized TMEM24–GFP and mitochondria in the submembrane space, indicating close proximity between the two structures (Fig. 5C). To test the hypothesis that TMEM24 may interact with mitochondria at ER–mitochondria contact sites, we utilized a membrane contact site reporter based on dimerization-dependent fluorescence (Alford et al., 2012). Briefly, the two monomers of a heteromeric fluorescent protein were anchored to the surface of the ER and mitochondria, respectively, and reconstitution of the red-light-emitting fluorescent protein only occurs when the two proteins are in proximity of each other, such as at contact sites between the ER and mitochondria (Fig. 5D; Fig. S3). When expressed in MIN6 cells, we observed red fluorescent structures that overlapped with MitoTracker, likely reflecting ER–mitochondria contacts (Fig. S3). Co-expression of TMEM24–GFP revealed little overlap under resting conditions, whereas accumulation at sites of ER–mitochondria proximity became apparent after depolarization-induced Ca²⁺ increases (Fig. 5E,F). These results show that TMEM24, in addition to acting at ER–PM contact sites, also might engage in reactions at ER–mitochondria contacts.

Previous work has established the ER-localized lipid-transport protein TMEM24 as an important regulator of insulin secretion from pancreatic β-cells (Lees et al., 2017; Pottekat et al., 2013). Transient reduction of TMEM24 expression or knockout of the TMEM24 gene have been found to impair insulin secretion from clonal rat and mouse β-cells. The underlying mechanism has been proposed to involve reduced Ca²⁺-independent recruitment of insulin granules to the PM (Pottekat et al., 2013) or disturbed regulation of voltage-dependent Ca²⁺ influx through alterations in the PM lipid composition (Lees et al., 2017). In this work, we further explored the mechanisms underlying TMEM24-dependent regulation of insulin secretion. In contrast to previous studies, we found that reduced expression of TMEM24 has little impact on voltage-dependent Ca²⁺ influx and only modestly impaired insulin secretion from β-cells. Instead, we identified TMEM24 as an important regulator of Ca²⁺ homeostasis at both the ER and the mitochondria, and show that this protein regulates mitochondrial ATP production.

TMEM24 is anchored to the ER through an N-terminal transmembrane domain and contains, in sequence, a lipid-binding SMP-domain, a C2-domain and a polybasic C-terminus. It is highly expressed in neuronal and endocrine tissue, including the pancreatic islets of Langerhans, where it is involved in tethering the ER to the PM (Lees et al., 2017; Sun et al., 2019). Its localization to the PM depends on electrostatic interactions between the C-terminus and negatively charged lipids in the PM, and neutralization of charged amino acids in TMEM24 by PKC-dependent phosphorylation results in TMEM24 dissociation. Using both intact and permeabilized cells, we found that the Ca²⁺-dependent dissociation of TMEM24 from the PM can be triggered even by nanomolar elevations of cytosolic Ca²⁺ through intracellular release or extracellular influx. Interestingly, we found that TMEM24 is already partially displaced from the PM under resting Ca²⁺ concentrations, perhaps due to constitutive PKC activity. This is consistent with findings from neuron-like cells, where endogenous TMEM24 has been found both at the PM and in the bulk ER under resting conditions (Sun et al., 2019). Such observations would support the hypothesis that TMEM24 acts at additional cellular locations, perhaps driven by interactions with lipids in the membranes of organelles known to form contacts with the ER (Cohen et al., 2018; Eden et al., 2010). By locally photobleaching a small region of the PM in cells expressing TMEM24–GFP, we could also estimate the mobility of TMEM24 at ER–PM junctions. We found that TMEM24 weakly interacts with the PM, and that a large fraction was highly dynamic under resting conditions, which is in sharp contrast to E-Syt2, another SMP-domain-containing protein constitutively localized to ER-PM contact sites (Xie et al., 2019). This observation is also consistent with TMEM24 executing functions at other cellular locations than the PM.

In contrast to previous studies (Lees et al., 2017; Pottekat et al., 2013), we did not find support for an absolute requirement of TMEM24 for normal insulin secretion. Both transient knockdown of TMEM24 using siRNA and CRISPR/Cas9-mediated knockout of TMEM24 had little effect on either glucose- or depolarization-induced Ca²⁺ influx and insulin secretion from MIN6 cells cultured as monolayers. If anything, cells with reduced expression performed slightly better than control cells. Interestingly, when we instead allowed MIN6 cells to aggregate into islet-like cell clusters (pseudo-islets) we found that the sustained glucose-stimulated insulin secretion was reduced following TMEM24 KO. One previous study, in which TMEM24 expression was stably reduced by shRNA, showed impaired glucose-stimulated insulin secretion from both clonal rat INS1 and mouse MIN6 β-cell pseudo-islets. The Ca²⁺ responses in these cells were normal, and secretion in response to direct depolarization was also unaffected by TMEM24 knockdown (Pottekat et al., 2013), which is similar to what we report here. CRISPR/Cas9-mediated knockout of TMEM24 in INS1 cells resulted in complete inhibition of both glucose-induced Ca²⁺ increases and insulin secretion, which was restored by re-expression of full-length TMEM24 (Lees et al., 2017). Although INS1 cells secrete insulin in response to glucose, the mechanism is likely different from that of primary β-cells in that it not only depends on K_ATP-channel closure (Herbst et al., 2002) and voltage-dependent Ca²⁺ influx (Dorff et al., 2002). Insulin secretion may instead be triggered by Ca²⁺ released from intracellular stores, since addition of the ER Ca²⁺ ATPase (SERCA) inhibitor thapsigargin causes Ca²⁺ oscillations in these cells (Herbst et al., 2002). Another possibility for the observed differences in insulin secretion may be that TMEM24 only regulate secretion under certain conditions. The observation here that TMEM24 KO impaired sustained insulin secretion from pseudo-islets despite having little effect on glucose-induced cytosolic Ca²⁺ changes indicate that it may be involved in the amplifying pathways of insulin, which operate in parallel with the Ca²⁺-dependent triggering pathway. The amplifying pathway requires mitochondrial metabolism, and is more prominent in pseudo-islets than cell monolayers (Chowdhury et al., 2013). Consistent with this, we observe impaired mitochondrial oxidative phosphorylation in TMEM24 KO cells, which is accompanied by depolarization of the inner mitochondrial membrane and in hyper-accumulation of Ca²⁺. The mitochondria were still able take up and release Ca²⁺ in response to changes in cytosolic Ca²⁺, arguing against defects in the major uptake and extrusion pathways. Regulation of mitochondrial Ca²⁺ is closely linked to ER Ca²⁺, and it is possible that the changes in mitochondrial function observed after TMEM24 KO are secondary to changes in ER Ca²⁺ handling. We found that TMEM24 KO cells have increased Ca²⁺ accumulation in the ER, observed by both measurements of cytosolic Ca²⁺ following ER-store depletion and by direct measurements of ER Ca²⁺. It is not clear how TMEM24 contributes to ER Ca²⁺ homeostasis. The major route of Ca²⁺ uptake into the ER of β-cells is the SERCA (Roe et al., 1994) but we did not find any evidence for increased SERCA activity in TMEM24 KO cells. If anything, the activity was slightly reduced, as indicated by the lack of an initial Ca²⁺ lowering effect of glucose in TMEM24 KO cells, although this might also reflect impaired ATP production or the steeper concentration gradient in these cells. There is a continuous leakage of Ca²⁺ from the ER, which results in rapid loss of Ca²⁺ upon SERCA inhibition, but the mechanism behind this leak is not clear. Numerous mediators of ER Ca²⁺ leak have been identified, including presenilin-1/2 and TMCO1 (Tu et al., 2006; Wang et al., 2016). Reduced ER Ca²⁺ leak could explain the increased accumulation of Ca²⁺ in the ER of TMEM24 KO cells. Interestingly, loss of both presenilin-1 and TMCO1, like TMEM24, results in ER Ca²⁺ overload and in impaired mitochondria function (Tu et al., 2006; Wang et al., 2016, 2019). Although TMEM24 unlikely functions as a Ca²⁺ channel, it may modulate other release mechanisms either through direct interactions or modulation of the lipid environment. Because of its dynamic nature, TMEM24 may provide means to acutely regulate ER Ca²⁺ permeability in response to increases in cytosolic Ca²⁺ or PM DAG concentrations. An alternative explanation for how TMEM24 regulate mitochondria function is by acting in trans at ER-mitochondria contact sites. Increases in the cytosolic Ca²⁺ concentration causes the dissociation of TMEM24 from the PM and is followed by accumulation of TMEM24 at mitochondria, as shown by the increased overlap between TMEM24-GFP and an ER-mitochondria proximity reporter. This is similar to other ER-localized lipid transport proteins, such as ORP5/8 and Vps13A/C, which have been shown to bind more than one organelle membrane (Galmes et al., 2016; Kumar et al., 2018). Although our observations are consistent with a role of TMEM24 at ER-mitochondria contacts, this would need to be confirmed using super-resolution microscopy or correlative light and electron microscopy. ER-mitochondria contacts are sites of Ca²⁺ and lipid exchange that control mitochondria Ca²⁺ levels, ATP production and morphology in β-cells (Rieusset, 2018). It is possible that TMEM24 controls mitochondria Ca²⁺ by modulating the Ca²⁺ transfer reaction at ER-mitochondria contact sites. These contact sites concentrate many of the key components of organellar Ca²⁺ homeostasis, including the MCU, the voltage-dependent anion channel (VDAC), SERCA, Presenilin-1 and IP3 receptors (Area-Gomez et al., 2009; Vecellio Reane et al., 2020). The MCU has low affinity for Ca²⁺ and is kept inactive under resting cytosolic Ca²⁺ concentrations (Marchi and Pinton, 2014). However, MCU is still important for maintaining basal energy production, likely via sensing Ca²⁺ microdomains formed by Ca²⁺ release from the ER at ER-mitochondria contacts (Rossi et al., 2019). It is possible that TMEM24 controls the transfer of Ca²⁺ between the two compartments either by directly modulating the function of one or several components of the contact sites, or indirectly through its effect on ER Ca²⁺ concentration and Ca²⁺ mobilization. Another intriguing possibility is that TMEM24 modulates the lipid composition of the mitochondria membranes, and that lack of this transport alters mitochondria function. TMEM24 has a strong preference for phosphatidylinositol, and the importance of this lipid for mitochondria function has been known since the 1960s (Vignais et al., 1963). More recently, it has been shown that that the outer mitochondrial membrane is indeed rich in phosphatidylinositol (Zewe et al., 2020), and that phosphorylated derivatives of this lipid are required for mitochondrial function (Nagashima et al., 2020; Rosivatz and Woscholski, 2011). However, it remains to be discovered how phosphatidylinositol is delivered to the mitochondria. One possibility is that TMEM24 contributes to this transport and couples it to changes in Ca²⁺ concentration and energy demand. Interestingly, one study found that depletion or masking of PI(4,5)P₂ on the mitochondria surface caused mitochondrial fragmentation, which could be prevented by PKC activation and would also trigger TMEM24 dissociation from the PM to enable interactions with the mitochondria (Rosivatz and Woscholski, 2011). A Ca²⁺-dependent feedback system to control mitochondria function may be particularly important in the β-cells, where mitochondrial metabolism is tightly coupled to Ca²⁺ influx in order to adjust insulin secretion and maintain blood glucose homeostasis.

Plasmids for expression of TMEM24–EGFP (Lees et al., 2017), E-Syt1–GFP (Giordano et al., 2013) and mRFP–PH-PLCδ1 were gifts from Pietro De Camilli (Yale University, New Haven, Connecticut, USA), as were those for ER–oxGFP, GFP–CAAX and E-Syt2–GFP (Giordano et al., 2013). VAMP2–pHluorin and NPY–mCherry were gifts from Sebastian Barg (Uppsala University, Uppsala, Sweden). GFP–P4M-SidM was Addgene plasmid #51472 (deposited by Gerald Hammond, University of Pittsburgh; Hammond et al., 2014). GFP–C1aC1b_PKC was a gift from Anders Tengholm (Uppsala University; Wuttke et al., 2013). R-GECO and mito-LAR-GECO were Addgene plasmids #32444 and #61245 (deposited by Robert Campbell, University of Alberta, Edmonton, Canada; Wu et al., 2014; Zhao et al., 2011). mApple–Tomm20 was Addgene plasmid #54955 (deposited by Michel Davidson). TMEM24–mCherry was generated by PCR amplification of human TMEM24 with flanking Nhe1 and EcoR1 followed by ligation into the mCherry-N1 vector (gift from Pietro De Camilli) using the following primers: TMEM24-Nhe1-fwd, 5′-CTAGCTAGCATGGATCCGGGCTGGGGGCA-3′; TMEM24-EcoR1-rev, 5′-CCGGAATTCTGAGCTGGGGGCTGGGGTT-3′. RA-Sec61b and GB-Dcp1b Addgene plasmids #153978 and #153979 (deposited by Gia Voeltz; Lee et al., 2020). GB was PCR-amplified using a 5′ primer containing a AgeI site, 5′-CGCTAGCACCGGTGGCCACCATCA-3′ as well as a 3′ primer containing a NotI site and a stop codon, 5′-GCGGCCGCTTATCCGGACTTGTACC-3′. The PCR-generated fragment was subcloned into the AgeI/NotI sites of the TOM20–mApple vector, replacing the mApple sequence and forming a tandem TOM20-GB construct. All salts, HEPES, poly-L-lysin, EGTA, α-toxin, thapsigargin, nitrilotriacetic acid (NTA) and diazoxide were from Sigma-Aldrich. DMEM, penicillin, streptomycin, glutamine and fetal bovine serum (FBS) were from Life Technologies. Carbachol, phorbol 12-myristate 13-acetate (PMA), cyanide-p-trifluoromethoxyphenylhydrazone (FCCP), rotenone, antimycin A and cyclopiazonic acid (CPA) were from Sigma-Aldrich. Bafilomycin was from TOCRIS Bioscience. Fluo-4-AM, Fura-2-AM and TMRM were from Life Technologies. Cal-520-AM and Cal590-AM were from AAT Bioquest.

The mouse β-cell line MIN6 (passages 18–30; gift from Jun-ichi Miyazaki, Kumomoto University, Japan) (Miyazaki et al., 1990) was cultured in DMEM (Life Technologies) supplemented with 25 mmol/l glucose, 15% FBS, 2 mmol/l L-glutamine, 50 μmol/l 2-mercaptoethanol, 100 U/ml penicillin and 100 μg/ml streptomycin (complete culture medium). The cells were kept at 37°C and 5% CO₂ in a humidified incubator. Prior to imaging, 0.2×10⁶ cells were resuspended in 100 µl Opti-MEM-I medium (Life Technologies) with 0.2 µg plasmid (total) and 0.5 µl Lipofectamine 2000 (Life Technologies) and seeded in the center of a 25-mm poly-L-lysine-coated coverslip. The transfection reaction was terminated after 4–6 h by the addition of 2 ml complete culture medium and cells were imaged 18–24 h later. MIN6 pseudo-islets were generated by resuspending 1×10⁶ MIN6 cells in 1 ml complete culture medium and seeding them into a non-stick 12-well cell culture plate (Falcon plastics, VWR, Sweden) followed by culturing for 48–72 h at 37°C and 5% CO₂ in a humidified incubator. This resulted in the spontaneous formation of islet-like cell clusters. Cell cultures were routinely tested for mycoplasma infections (Eurofins genomics).

Transfected MIN6 cells, grown on 25-mm poly-L-lysine-coated glass coverslips, were incubated with 0.5 µM of the AM-ester form of Cal520 (AAT Bioquest, Sunnyvale, CA) for 30 min at 37°C. The coverslips were subsequently used as exchangeable bottoms in a modified Sykes–Moore open superfusion chamber that was mounted on the stage of a TIRF microscope (described below) and connected to a peristaltic pump that allowed rapid change of medium. Following a change from normal, extracellular-like, medium (125 mM NaCl, 4.9 mM KCl, 1.3 mM MgCl₂, 1.2 mM CaCl₂, 25 mM HEPES, 1 mg/ml BSA with pH set to 7.4) to an intracellular-like medium (see below), the superfusion was interrupted and α-toxin was added directly to the chamber (final concentration ≈50 μg/ml). Permeabilization was considered complete when the Cal520 fluorescence had decreased by >90%, which typically took 2–5 min. Superfusion was then started again and the cells were exposed to intracellular-like buffers containing calibrated Ca²⁺ concentrations while fluorescence from both remaining Cal520 and mCherry-tagged fusion proteins was recorded. These experiments were performed at ambient temperature (21–23°C).

Intracellular-like medium with buffered pH, [Ca²⁺] and [Mg²⁺] used in α-toxin permeabilization experiments contained: 6 mM Na⁺, 140 mM K⁺, 1 mM (free) Mg²⁺, 0–100 μM (free) Ca²⁺, 1 mM Mg-ATP, 10 mM HEPES, 2 mM (total) EGTA and 2 mM (total) nitrilotriacetic acid (NTA) with pH adjusted to 7.00 at 22°C with 2 M KOH. The total concentration of Ca²⁺ and Mg²⁺ was calculated using the online version of MaxChelator (http://www.stanford.edu/~cpatton/webmaxcS.htm). Media were made fresh on the day of experiment and kept on ice. To validate the media composition, cells loaded with the Ca²⁺ indicator Cal520 were mounted on a TIRF microscope, permeabilized and exposed to media with increasing Ca²⁺ concentrations. From these data a dose–response curve was generated and the EC₅₀ for Ca²⁺-binding to Cal520 was estimated to be 980 nM, an in situ estimation 3-fold higher than the reported in vitro K_D of 320 nM.

MIN6 cells were resuspended with 25 nM antiTMEM24 siRNA smartpool (Dharmacon, siGENOME cat. no. L-061885-01-0020) pre-mixed with 1.6 µl/ml Lipofectamine RNAiMAX (Life Technologies) in Opti-MEM I medium. After 3 h, the Opti-MEM I medium was replaced by DMEM growth medium containing 25 nM siRNA and 1.6 µl/ml Lipofectamine 3000 (Life Technologies) and incubated overnight. Cells transfected with a scrambled sequence siRNA (Dharmacon, siGENOME) following the same procedure were used as controls. After 48 h, cells were transfected for imaging as described above. The knockdown efficiency was evaluated by western blotting and real-time PCR (see below).

TMEM24 KO MIN6 were generated using a mix of three TMEM24 CRISPR/Cas9 plasmids (Santa Cruz Biotechnology, sc-428047), containing GFP and the following gRNAs: 5′-TTACCATGGTGCGCTCTGAT-3′, 5′-CAGCACTCAGCCCGCCATGA-3′ and 5′-GTCCCCCGCTGCCGTCTCCA-3′. TMEM24 CRISPR/Cas9 plasmids were transfected into MIN6 by Lipofectamine 3000 (Thermo Fisher, L3000015). After 24-h, single cells positive (knockout) and negative (control) for GFP were sorted using a BD FACSariaIII CellSorter and subsequently plated into 96-well plates. Cells were cultured in standard MIN6 growth medium. The clones were tested for the presence of TMEM24 by western blotting utilizing anti-TMEM24 antibody (Thermo Fisher, A304-764A; 1:1000).

A medium containing 125 mM NaCl, 4.9 mM KCl, 1.3 mM MgCl₂, 1.2 mM CaCl₂, 25 mM HEPES, 1 mg/ml BSA with pH set to 7.4 at 37°C was used in all microscopy experiments. Cells were pre-incubated in this medium supplemented with 3 mM glucose for 30 min followed by perifusion with the same medium during recordings. A previously described custom-build prism-type TIRF microscope setup equipped with a 16×/0.8 NA water immersion objective (Nikon) was used to observe large populations of cells (Idevall-Hagren et al., 2010). It was built around an E600FN upright microscope (Nikon) enclosed in a Perspex box thermostated at 37°C. 491-nm and 561-nm DPSS lasers (Cobolt, Sweden) were used to excite GFP and mCherry, respectively. The laser beams were merged by dichroic mirrors (Chroma Technology) and homogenized by a rotating light shaping diffuser (Physic Optics Corps) before being refocused through a dove prism (Axicon) with a 70° angle to achieve total internal reflection. Laser lines were selected with interference filters (Semrock) in a motorized filter wheel equipped with a shutter (Sutter Instruments) blocking the beam between image captures. Emission light was detected at 530/50 nm (Semrock interference filters) for GFP or 597LP (Melles Griot glass filter) for mCherry using a CCD camera (Hamamatsu ORCA-AG) controlled by MetaFluor software (Molecular Devices). For high-resolution TIRF imaging, we used an Eclipse TiE microscope (Nikon) equipped with a TIRF illuminator and a 60×/1.45 NA or 100×/1.49 NA objectives as previously described (Xie et al., 2019). Confocal microscopy was performed on an Eclipse TE2000 microscope (Nikon) equipped with a Yokogawa CSU-10 spinning disc confocal unit and a 100×/1.49 NA plan Apochromat objective (Nikon) (Idevall-Hagren et al., 2013). Briefly, GFP and mCherry were excited by 491-nm and 561-nm DPSS lasers (Cobolt, Sweden) and were detected through 530/50 nm interference filter and 597LP filter, respectively, through a black-illuminated EM-CCD camera (DU-888; Andor Technology).

After three washes with phosphate-buffered saline (PBS), MIN6 cells were homogenized and lysed in RIPA buffer (50 mM Tris-HCl, pH=7.4, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS and 150 mM EDTA), placed on ice for 30 min or agitated on a rotator at 4°C for 30 min. Cell lysates were mixed with 3× sample buffer (6% SDS, 15% 2-mercaptoethanol, 30% glycerol, 0.006% Bromophenol Blue and 0.15 M Tris-HCl) and boiled at 95°C for 10 min. Protein content was determined by a detergent-compatible protein assay (Bio-Rad, Hercules, CA, USA) and proteins were separated by SDS-PAGE (5–20%) and blotted onto PVDF membrane using semi-dry transfer. Membranes were blocked in 4% milk dissolved in Tris-buffered saline with 0.1% Tween 20. The following antibodies and dilutions were used: anti-TMEM24 (Bethyl Laboratories, A304-764, 1:1000, specificity confirmed by KD and KO in this study), anti-GAPDH (Cell Signaling Technology, 14C10, 1:1000), anti-rabbit-IgG secondary antibody (GE Healthcare, 1:10,000). Membranes were developed with the Odyssey Fc Imaging system (LI-COR Bioscience).

mRNA from control and TMEM24 knockdown cells were extracted using the NucleoSpin RNAPlus kit (Macherey-Nagel). RT-PCR was performed by QuantiTect SYBR Green RT-PCR kit (Qiagen) using the following primer: GAPDH-fwd, 5′-ACTCCACTCACGGCAAATTC-3′; GAPDH-rev, 5′-TCTCCATGGTGGTGAAGACA-3′; TMEM24-fwd, 5′-CGCCCAGAACTCAGCCTAAA, TEME24-rev, 5′-GGGTAGGTCTGGGGATGGAT-3′. The PCR was performed in a Light Cycler 2.0 (Roche). Results are presented as ΔΔC_t, normalized to GAPDH expression in control and TMEM24 KD cells.

Insulin secretion was measured from monolayers of MIN6 cells grown in 12-well plates. 4×10⁵ control or TMEM24 knockdown cells were prepared as described above in 12-well plates 72 h before secretion measurements. Cells were pre-incubated in 3 mM glucose imaging buffer at 37°C for 30 min. Cells were subsequently incubated in buffers containing 3 mM glucose, 20 mM glucose or 3 mM glucose supplemented with 30 mM KCl, immediately followed by buffer collection. In the end, cells were released by the addition of 80 µl typsin and mixed with 120 µl complete growth medium. 100 µl of the collected cell suspension was mixed with 100 µl acidic ethanol, sonicated on ice and neutralized by the addition of 900 μl Tris buffer (pH 8) to determine insulin content. The insulin concentration in the samples was measured by a mouse insulin AlphaLISA kit (Perkin-Elmer). For perifusion experiments, groups of 30 pseudo-islets were placed in a 10 µl teflon tube chamber and perifused at a rate of 60 µl/min using a pressurized air system (AutoMate Scientific, Berkeley, CA, USA) and equilibrated in experimental buffer containing 3 mM glucose during 45 min. The perifusate was subsequently collected in 5-min fractions into ice-chilled 96-well plates with a non-binding-surface (Corning Inc Kennebunk, ME) while changing the glucose concentration to 20 mM. The islets were retrieved and briefly sonicated in acid ethanol to determine insulin content. Insulin concentrations in the samples were determined with ELISA according to the manufacturer's instructions (Mercodia AB, Uppsala, Sweden).

The oxygen consumption rate (OCR) in MIN6 cells were monitored by an Extracellular Flux Analyzer (XFe96, Agilent Technologies, CA, USA). 3×10⁴ cells (wild-type and TMEM24 KO) were seeded in each well of XFe96 microplates and cultured for an additional 24 h. Cells were thereafter pre-incubated with Seahorse XF DMEM (Agilent Technologies) containing 3 mM glucose (pH 7.4) for 1 h at 37°C before the microplate was inserted into the XFe96 Analyzer. For each experiment, 4–6 replicates of each treatment were measured. OCR at 3 mM glucose was measured for 30 min, followed by another 30 min with either 3 mM or 20 mM glucose. Then, the proportions of respiration driving ATP synthesis and proton leak were determined by blocking ATP synthase with 2 μM oligomycin (Sigma-Aldrich). Subsequently, 2 μM of the mitochondrial uncoupler cyanide-p-trifluoromethoxyphenylhydrazone (FCCP) was added to determine the maximal respiratory capacity. Finally, 2 μM rotenone and 5 μM antimycin A were added together to block transfer of electrons from the mitochondrial respiratory chain complex I and III to determine the remaining non-mitochondria-dependent respiration. To calculate the mitochondrial respiration, non-mitochondrial OCR was subtracted from the total OCR. Data was normalized to protein content which was determined by the DC protein assay (Bio-Rad Laboratories, USA).

The mitochondrial membrane potential (ΔΨ_m) was monitored by the lipophilic cationic dye tetramethylrhodamine methyl ester (TMRM) via fluorescence time-lapse imaging using an inverted microscope (Eclipse TE2000U; Nikon, Kanagawa, Japan). The epifluorescence microscope is equipped with a high-power LED light source (Omicron LedHUB; Photonlines Ltd, Newcastle, UK) which, connected with a 5-mm diameter liquid light guide, provided excitation light at 540 nm. Emission was measured at 560 nm (5 nm half-bandwidth) and the fluorescence signal was detected by an Evolve EMCCD camera (Photometrics, Arizona, USA). MetaFluor software (Molecular Devices Corp.), which was allowed to control the microscope setup and to acquire images every 5 s. Prior to imaging, the cells were seeded onto 25 mm round coverslips and loaded with TMRM at 10 nM during a 30-min incubation at 37°C in an imaging buffer containing 125 mM NaCl, 5 mM KCl, 1.3 mM CaCl₂, 1.2 mM MgCl₂, and 25 mM HEPES with pH adjusted to 7.40 with NaOH. The coverslips were placed on the bottom of an open Sykes–More chamber. On the top of the coverslip, a thin 25-mm diameter stainless steel plate with a 4-mm wide and 7-mm long opening pressed the 1-mm thick silicon rubber gasket with identical dimensions and central opening to the coverslip. The temperature of the chamber holder and the CFI S Fluor 40×/1.3 NA oil immersion objective (Nikon) was stable at 37°C during the experiment using custom-built thermostats. Fixed on the stainless-steel plate, inlet and outlet cannulas maintained a laminar superfusion at a rate of 2.0 ml/min with the imaging buffer containing 10 nM TMRM.

Cytosolic [Ca²⁺] was measured on an epifluorescence microscope setup (described above) using the ratiometric dye Fura-2 or the green-fluorescent dye Cal520. The cells were preincubated for 30 min at 37°C in imaging buffer supplemented with 1 µM of the AM-ester-form of the indicator, followed by repeated washing in imaging buffer and imaging. Mitochondrial Ca²⁺ was measured using the genetically encoded red fluorescence, low-affinity indicator mito-LAR-GECO (Wu et al., 2014), whereas ER Ca²⁺ was measured using the FRET-based indicator D4ER (Ravier et al., 2011). Both indicators were delivered to cells by transient transfection as described above, and experiments were performed 18–36 h post transfection.

To determine the morphology of the mitochondria, wild-type MIN6 cells and TMEM24 KO cells expressing mApple–Tom20 were observed under the spinning disc confocal microscope. The images were analyzed with the open-source image analysis software CellProfiler (Carpenter et al., 2006) using several modules that were placed in a sequential order to create a flexible image analysis pipeline. First, processing filters were applied to enhance the fluorescence signal of the regions that display higher intensity relative to its immediate neighborhood. This allowed the better identification of the mitochondria as separate objects during the following module. The Otsu's automatic threshold method permitted the assignment of the threshold value by including the pixels of the image either in the ‘background’ class or the ‘foreground’ class. The objects/mitochondria detected using the pipeline were not de-clumped to enable the identification of potential network formation, and the size and shape of the mitochondria was determined and the data was exported to Microsoft Excel. Eccentricity, which distinguish mitochondria based on shape from tubular (1) to circular (0), was used as an overall determinant of the mitochondria shape.

TIRF microscopy and confocal microscopy images were analyzed offline by Fiji (Schindelin et al., 2012). To determine fluorescence changes, the ROIs and background regions were first manually identified. Fluorescence intensity changes within these regions were recorded and the data was exported to Excel. All data points were background corrected and normalized to the initial fluorescence intensity (F/F₀).

One-way ANOVA followed by Tukey's post hoc test (for normally distributed data), Kruskal–Wallis one-way ANOVA followed by a Mann–Whitney U-test (for non-parametric data) or a two-tailed (paired or unpaired) Student’s t-test were used. All data presented are from at least three biological replicates. For imaging data, n numbers represent individual cells and N numbers represents independent experiments. For all other data presented, n numbers indicate independent observations.

We are grateful to Mrs Antje Thonig and Mrs Parvin Ahooghalandari for excellent help with molecular biology work and insulin secretion measurements and to Prof. Erik Gylfe and all members of the O.I.-H. laboratory for input on the work.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.259073.