Cortical mitochondria regulate insulin secretion by local Ca²⁺ buffering in rodent beta cells

Pancreatic beta cells sense increases in the surrounding glucose concentration and secrete glucose-lowering insulin to counteract the elevation. The mechanism by which glucose controls insulin secretion involves uptake across the plasma membrane, followed by glycolysis and mitochondrial oxidative phosphorylation to produce ATP. This nucleotide closes ATP-sensitive K⁺ channels in the plasma membrane, leading to depolarization, opening of voltage-dependent Ca²⁺ channels and a rise in cytosolic Ca²⁺ that triggers insulin secretion (Rorsman and Ashcroft, 2018). Mitochondrial metabolism is thus required for glucose-stimulated insulin secretion, but this organelle also regulates insulin secretion downstream of ATP production by controlling the cytosolic Ca²⁺ concentration (Wiederkehr and Wollheim, 2008). The outer mitochondrial membrane is Ca²⁺ permeable, and Ca²⁺ is further transported into the mitochondrial matrix by the mitochondrial Ca²⁺ uniporter (mCU), in a process that is driven by the negative membrane potential (Patron et al., 2013). The mCU has a low affinity for Ca²⁺, and its activity under resting conditions is very low, but it can be activated by elevations in the cytosolic Ca²⁺ concentration and mitochondria contribute to Ca²⁺ homeostasis by acting as buffers. For the mCU to work efficiently, it needs Ca²⁺ in concentrations in excess of 10 µM. Such high concentrations are rarely seen in the bulk cytosol, but are frequently reached at sites of Ca²⁺ release or influx (Parekh, 2008). In recent years, numerous studies have shown that mitochondria localized close to Ca²⁺-release channels of the endoplasmic reticulum (ER) or voltage-dependent Ca²⁺ channels of the plasma membrane are exposed to much higher Ca²⁺ concentrations than those localized further away and can buffer changes in cytosolic Ca²⁺ (Giacomello et al., 2010; Malli et al., 2003; Quintana et al., 2011). Mitochondria are highly dynamic in cells, and their motility is largely dependent on GTPase-mediated transport along microtubules. The distribution of mitochondria in the cell is nonhomogenous, and it may also vary over time. This is particularly evident in cells with an asymmetric morphology, such as neurons. In these cells, mitochondria transport to axonal segments is required for local ATP production, which, in turn, is used for synaptic vesicle release (Verstreken et al., 2005). However, the presence of mitochondria in the presynaptic compartment also negatively regulates vesicle release by local buffering of synaptic Ca²⁺ elevations triggered by depolarization, in a process that is regulated by neuronal activity (Vaccaro et al., 2017). The basic mechanism of synaptic vesicle exocytosis and insulin granule exocytosis is the same, but whether mitochondria play a direct, local role in the regulation of insulin release is not known. Most studies on the interplay between mitochondria and Ca²⁺ have focused on the regulation of mitochondrial metabolism by the ion. The increase in Ca²⁺ that occurs in the mitochondrial matrix following voltage-dependent Ca²⁺ influx has been shown to have both stimulatory and inhibitory effects on ATP production (Alam et al., 2012; Kennedy et al., 1999; Krippeit-Drews et al., 2000; Li et al., 2013; Wiederkehr et al., 2011). Since insulin granules require microdomains of high Ca²⁺ for efficient exocytosis, it is possible that mitochondria play a more direct role in controlling insulin secretion by exerting a local buffering effect. In this study, we show that beta-cell mitochondria are abundant in the subplasmalemmal space under resting conditions, but that the density decreases following Ca²⁺ influx, in a process that requires intact cortical actin. Using light-driven molecular motors, we increased the number of subplasmalemmal mitochondria and found that this results in both reduced cytosolic Ca²⁺ elevations and exocytosis following depolarization.

Total internal reflection fluorescence (TIRF) microscopy imaging of MIN6 beta cells expressing mitochondria-targeted mApple (mApple-Tomm20-N1) revealed that 18±2% of the subplasma membrane area was occupied by mitochondria. Most mitochondria exhibited dynamic movements, but there were sites that were occupied by mitochondria throughout the time course of the experiments. Similar observations have been made in neurons, in which only one-third of the mitochondrial population in axons is mobile (Kang et al., 2008; Sheng and Cai, 2012). To test whether the cortical mitochondria constitute a stable population in beta cells, we stimulated the cells with 20 mM glucose, the main physiological stimulus of insulin secretion. This stimulation was without effect on the overall cortical mitochondria density (Fig. 1A,B). However, simultaneous recordings of the intracellular Ca²⁺ concentration with GCaMP5G revealed that the cortical mitochondria exhibited frequent transient displacements that mostly (83%) coincided with temporary increases in the cytosolic Ca²⁺ concentration (Fig. 1C–F). To more directly test the effect of Ca²⁺ on cortical mitochondria, we depolarized the cells with 30 mM KCl. We found that the cortical mitochondria density was significantly reduced in depolarized cells (from 18±2% to 9±1%, n=22; P<0.01) (Fig. 1G,H). Indeed, depolarization induced a rapid 13±2% (n=32; P<0.01) reduction in subplasma membrane mitochondria fluorescence within 1 min (Fig. 1I), indicating that the localization of mitochondria within beta cells is regulated under conditions that promote insulin secretion.

To obtain control over mitochondria distribution, we developed an optogenetic tool to guide mitochondria to the plasma membrane based on the CRY2-CIBN dimerization system. Briefly, we fused the anterograde motor protein Kif1a to CIBN and co-expressed it with GFP-tagged Tomm20 fused to CRY2. Blue-light illumination promotes reversible CRY2-CIBN interaction, resulting in anterograde transport of mitochondria along microtubules (Fig. 1J). Blue-light illumination of MIN6 cells expressing these tools resulted in a time-dependent increase in subplasma membrane mitochondria density, with a maximal increase of up to 100% (Fig. 1J,K). This increase was reversed following interruption of the illumination for 2 h (Fig. 1L). Moreover, inhibition of microtubule polymerization with nocodazole prevented the recruitment of mitochondria to the plasma membrane upon illumination, demonstrating that the mitochondrial accumulation is due to active transport (Fig. 1M).

Beta cells, like most cells, have an extensive cortical actin network that can limit organelle accessibility to the plasma membrane (Hsieh et al., 2017; Orci et al., 1972; Papadopulos et al., 2015). To visualize cortical actin, we expressed an mCherry-tagged version of the calponin homology domain from Utrophin (mCherry-CH_Utr), which selectively labels F-actin, in clonal beta cells and imaged them by TIRF microscopy. Under basal conditions, cortical F-actin was seen as a meshwork with densely spaced focal contacts (Fig. 2A). Following KCl depolarization, there was a striking accumulation of cortical actin, often in distinct patches, evident as increased subplasma membrane F-actin reporter fluorescence (Fig. 2A–D). Addition of latrunculin A to depolymerize F-actin prevented this effect (Fig. 2A–D). A possible explanation for the cortical F-actin patches is local changes in Ca²⁺ concentration beneath the plasma membrane. To explore this possibility, we co-expressed the plasma membrane-immobilized Ca²⁺ indicator 5HT₆-G-GECO and mCherry-CH_Utr. Depolarization resulted in an immediate increase in subplasma membrane Ca²⁺ concentration, accompanied by F-actin accumulation (time difference 19±3 s, n=13), and, although we observed local changes in Ca²⁺ concentration, these did not coincide with the location of F-actin patch formation (Fig. 2E,F). Elevation of the glucose concentration from 3 mM to 11 mM resulted in periodic accumulations of cortical F-actin that, in most cases (82%), were preceded by increases in the cytosolic Ca²⁺ concentration. Addition of the K_ATP-channel opener diazoxide completely suppressed glucose-induced Ca²⁺ concentration changes and dramatically inhibited the F-actin rearrangements (Fig. 2G–I; Fig. S1). As for KCl, the glucose-induced cortical F-actin accumulations occurred nonhomogenously across the plasma membrane, but tended to reoccur in the same location during prolonged stimulations (Fig. 2G; Fig. S1). The magnitude of cortical actin accumulation positively correlated with the relative increase in cytosolic Ca²⁺ concentration, indicating a direct role of Ca²⁺ in actin redistribution (Fig. 2I). Addition of latrunculin A prevented the glucose-induced periodic accumulations of F-actin but did not affect the Ca²⁺ fluctuations (Fig. 2J). Interestingly, the latrunculin A-induced actin depolymerization, like microtubule disruption, prevented the light-induced recruitment of mitochondria to the plasma membrane, indicating a role of cortical actin in the transport process (Fig. 2K,L). The relationship between actin accumulation beneath the plasma membrane and the disappearance of mitochondria from the same volume indicates that actin biding to the plasma membrane might displace mitochondria. TIRF microscopy imaging of cortical mitochondria (GFP-Tomm20) and F-actin (mCherry-CH_Utr) showed that the depolarization-induced increase in cortical actin density coincided with a reduction in mitochondria density in the same compartment (Fig. S1), similar to what has been shown for actin and cortical ER (Hsieh et al., 2017). To test whether the observed accumulation of cortical actin is directly caused by Ca²⁺ influx, we depolarized cells in the absence or presence of the L-type voltage-dependent Ca²⁺ channel blocker verapamil. In the presence of the inhibitor, depolarization failed to induce either Ca²⁺ influx or redistribution of cortical actin (Fig. 3), demonstrating direct effects of Ca²⁺ on the cortical actin network.

Mitochondria are important Ca²⁺ regulators in many cells, acting as low-affinity buffers under conditions of elevated cytosolic Ca²⁺ concentrations. Next, we therefore determined whether the subplasma membrane mitochondria in beta cells might shape the Ca²⁺ response following voltage-dependent Ca²⁺ influx. To this end, we depolarized cells with normal or light-induced increase in subplasma membrane mitochondrial density and compared the resulting Ca²⁺ elevation. We found that the Ca²⁺ response gradually decreased as mitochondria redistributed to the cell cortex, stabilizing at 50% reduction after 30 min of illumination (Fig. 4A,B). The typical Ca²⁺ response to depolarization was a mixture between stable elevation, reflecting voltage-dependent influx and superimposed transient peaks due to Ca²⁺ release from the ER (Xie et al., 2018). The latter phenomenon is secondary to stimulated insulin secretion and co-release of ATP that acts on purinergic P2Y₁ receptors to generate Ca²⁺-mobilizing inositol 1,4,5-trisphosphate (IP₃) (Wuttke et al., 2013). Increased cortical mitochondrial density suppressed both components (Fig. 4B–F), indicating that the mitochondria buffer Ca²⁺ from both sources. Similar suppression following light-induced recruitment of mitochondria to the cell cortex was also seen in beta cells stimulated with 20 mM glucose (Fig. 4G,H). To directly show that mitochondria sequester Ca²⁺ in response to depolarization, we performed simultaneous confocal microscopy recordings of cytosolic (GCaMP5G) and mitochondrial (mito-LAR-GECO) Ca²⁺ concentrations, and observed parallel changes in Ca²⁺ concentration in the two compartments (Fig. S2).

Exocytosis of insulin-containing granules is triggered by Ca²⁺ influx. Since we found that Ca²⁺ has an influence on mitochondrial position, and that the mitochondrial position, in turn, has an impact on the cytosolic Ca²⁺ concentration, we next investigated the role of subplasma membrane mitochondria in the regulation of exocytosis. We used vesicle-associated membrane protein 2 (VAMP2) fused to the pH-sensitive fluorescent protein pHluorin as a detector of exocytosis. In cells co-expressing the mitochondrial marker mApple-Tomm20, we observed mitochondria in the proximity of, but not overlapping with, insulin granule fusion sites at the plasma membrane (Fig. 5A,B). To directly test the role of cortical mitochondria in the regulation of insulin secretion, we used the light-driven molecular motors to increase the cortical mitochondria density. To measure insulin secretion in real time at the single-cell level, we utilized a well-characterized feedback mechanism. The above-mentioned autocrine effect of ATP co-secreted with insulin in addition to IP₃ also generates diacylglycerol (DAG) in the plasma membrane (Fig. 5C) (Wuttke et al., 2013). DAG recordings can therefore be used as a proxy for insulin secretion. Depolarization of MIN6 cells expressing the DAG biosensor mCherry-C1aC1b_PKC resulted in pronounced DAG spikes, reflecting individual granule fusion events, and application of the P2Y₁ receptor antagonist MRS2179 completely inhibited this response (Fig. 5D,E). In cells with light-induced increase in cortical mitochondria density, there was a 50% reduction in DAG spike frequency, indicating that cortical mitochondria can restrict insulin secretion (Fig. 5F). Consistent with this notion, we found that glucose-induced reductions in cortical mitochondria density were associated with increased DAG spike frequency (Fig. S2).

In this study, we set out to determine the importance of mitochondrial positioning for the regulation of insulin secretion from beta cells. We identified a pool of subplasma membrane-localized mitochondria that undergoes dynamic changes in localization in a Ca²⁺- and F-actin-dependent manner and contributes to the regulation of insulin secretion by buffering cytosolic Ca²⁺.

The importance of mitochondrial metabolism for normal glucose-stimulated insulin secretion is well established; ATP derived from oxidative phosphorylation is required for inducing the membrane depolarization that triggers exocytosis of insulin granules. ATP is also crucial for priming and replenishing the release-competent pool of insulin granules (Rorsman and Ashcroft, 2018). In addition, the mitochondria act as low-affinity Ca²⁺ buffers that sequester the ion under conditions of elevated cytosolic Ca²⁺, such as those triggering insulin secretion (Tarasov et al., 2012; Wiederkehr and Wollheim, 2012). To what extent the ensuing increase in mitochondrial Ca²⁺ impacts mitochondrial metabolism and subsequent insulin secretion is still a matter of debate, with studies supporting both stimulatory and inhibitory roles of the ion on metabolism (Kennedy et al., 1999; Krippeit-Drews et al., 2000; Li et al., 2013; Wiederkehr et al., 2011). Another layer of complexity in mitochondrial regulation of beta-cell function is the control of their position within the cells. The mitochondria are highly dynamic organelles that move along microtubules, and it has been shown, in neurons, that the presence of mitochondria in the synapses is required for normal neurotransmission (Vaccaro et al., 2017). The beta cells lack the complex morphology of neurons but still have regions in the plasma membrane that resemble the synaptic active zones (Low et al., 2014). Cortical mitochondria have been described in many cell types, and often consist of a mobile and more stable population. We find that beta cells under resting conditions have a high density of mitochondria beneath the plasma membrane, similar to what has been described in the related chromaffin cell (Villanueva et al., 2014). Glucose stimulation did not result in a stable change in mitochondrial density at the plasma membrane, but the glucose-induced Ca²⁺ oscillations were frequently associated with transient displacements of cortical mitochondria. Consistent with this, depolarization with ensuing Ca²⁺ influx resulted in a reduction in the number of cortical mitochondria. We find that this Ca²⁺-induced redistribution of mitochondria is likely due to an increase in the cortical F-actin density. Increased association between cortical F-actin and the plasma membrane under conditions promoting exocytosis has previously been shown in chromaffin cells, and found to be important for the replenishment of plasma membrane-docked granules (Papadopulos et al., 2015; Wen et al., 2011). The redistribution involves activation of the GTPase Cdc42 and the Arp2/3 complex (Gasman et al., 2004). Interestingly, Cdc42 activity was recently shown to be stimulated by Ca²⁺/calcineurin, providing a possible mechanism for the depolarization-induced actin remodeling (Ly and Cyert, 2017). F-actin has also been found to regulate insulin secretion through multiple pathways, including a direct interaction with syntaxin 4, a protein that is important for granule docking (Jewell et al., 2008; Tomas et al., 2006). In the case of secretory granules, the redistribution of F-actin results in increased granule density at the plasma membrane (Papadopulos et al., 2015), whereas we observe reduced mitochondrial density under the same conditions. The ER also shows, in many cell types, a prominent cortical distribution that is similarly restricted by the presence of F-actin (Hsieh et al., 2017). Acute increase in cortical actin density, as observed here, might therefore simply displace mitochondria that are not stably attached to the plasma membrane. Some mitochondria remain at the plasma membrane, and these might reflect a recently described population that is stably tethered (Lackner et al., 2013). Ca²⁺ increases have previously been shown to reduce the motility of mitochondria along microtubules (Yi et al., 2004), and it has been suggested that the stalling at sites of high Ca²⁺ would enable local ATP production. This mechanism could be of importance in asymmetrical cells in which free diffusion of ATP is restricted, but it is less clear as to what extent ATP microdomains exist in small symmetrical cells, such as beta cells (Kennedy et al., 1999).

To determine the role of cortical mitochondria in the regulation of beta-cell function, we developed a light-regulated transport system, similar to what has recently been described (van Bergeijk et al., 2015), and, using this, we were able to increase the cortical mitochondria density twofold. Since the light-induced mitochondria translocation is based on kinesin-dependent transport along microtubules, it was completely blocked by the microtubule destabilizing drug nocodazole. The reaction was reversed within 2 h, when illumination was interrupted. The light-induced changes in cortical mitochondria operate on a significantly longer timescale than the rapid dynamics induced by Ca²⁺, likely reflecting both the slow reversibility of the optogenetic dimers (Kennedy et al., 2010) and the requirement of active, retrograde transport. Surprisingly, the light-induced transport also required intact F-actin, as it was completely inhibited by the actin depolymerizing drug latrunculin A. It has been shown that, although transport along microtubules is the main mechanism, short-distance mitochondrial movements and immobilization of the organelle at the cell cortex often require the actin cytoskeleton (Boldogh and Pon, 2006; Chada and Hollenbeck, 2004; Saxton and Hollenbeck, 2012). Furthermore, mitochondrial movement is enhanced in the absence of F-actin (Morris and Hollenbeck, 1995). The light-induced accumulation of cortical mitochondria resulted in pronounced suppression of both glucose- and depolarization-induced elevation of the cytosolic Ca²⁺ concentration and of insulin granule exocytosis. Mitochondria have previously been ascribed local buffering capacity at ER–mitochondria contact sites, where they limit the spread of Ca²⁺ released by IP₃ receptors (Rizzuto et al., 1998). The findings here suggest a similar role of mitochondria in controlling a pool of Ca²⁺ of importance for regulated secretion, and that this buffering capacity is under the control of Ca²⁺ itself. We propose a model in which cortical mitochondria accumulate close to sites of exocytosis, perhaps to fuel energy-demanding processes related to granule priming and release. Upon influx of Ca²⁺, the cortical actin density is increased, causing rapid displacement of mitochondria with resulting loss of Ca²⁺ buffering capacity near the influx sites, thereby supporting continued exocytosis (Fig. 6).

Nocodazole, latrunculin A, verapamil and diazoxide were from Sigma-Aldrich (Germany). The following plasmids were used in this study: GFP-CRY2-OCRL (Idevall-Hagren et al., 2012), CIBN-GFP-CAAX (Idevall-Hagren et al., 2012), mApple-Tomm20-N1 (Addgene plasmid #54955), R-GECO [Addgene plasmid #32444 (Zhao et al., 2011)], GCaMP5G [Addgene plasmid #31788 (Akerboom et al., 2012)], mCherry-CH_Utr [Addgene plasmid #26740 (Burkel et al., 2007)], mCherry-C1aC1b_PKC (Wuttke et al., 2013), iRFP-PH-PLC_δ1 (Idevall-Hagren et al., 2012), mito-LAR-GECO [Addgene plasmid #61245 (Wu et al., 2014)] and 5HT6-G-GECO [Addgene plasmid #47499 (Su et al., 2013)].

Kif1a-GFP-CIBN was generated by cloning mouse Kif1a into GFP-CIBN using the following primers: Kif1a-NheI-for (5′-TATAGCTAGCGCCACCATGGCTGGGGCCTCTGT-3′), Kif1a-AgeI-rev (5′-TATAACCGGTCCCAGCAGATCTCGCAGCC-3′). Kif1a-iRFP-CIBN was generated by replacing GFP in Kif1a-GFP-CIBN with iRFP using the following primers: AgeI-iRFP-for (5′-CTCACCGGTAGCTGAAGGATCCGTCGCCAG-3′), HindIII-iRFP-rev (5′-CTCAAGCTTGCTCTTCCATCACGCCGATCTG-3′). Tomm20-GFP-C1-CRY2 was generated by amplification of the N-terminal region of mouse Tomm20 using the following primers: Tomm20-Nhe1-for (5′-TATAGCTAGCACCATGGTGGGTCGGA-3′), Tomm20-Nhe1-rev (5′-TATAGCTAGCAGACCACCAGATTGAAAAT-3′). Subsequently, CRY2 was inserted into the multiple cloning site after the GFP using the restriction sites Kpn1 and Mfe1 (CRY2-Kpnl-for 5′-TATAGGTACCATGAAGATGGACAAAA-3′, CRY2-MfeI-rev 5′-TATACAATTGATCTGCTGCTCCGATCATGA-3′). Tomm20-mApple-CRY2 was generated by inserting CRY2 into the Tomm20-mApple plasmid (Addgene plasmid #54955) via the restriction sites Not1 and Xba1 (NotI-Cry2-for 5′-AACGCGGCCGCCATGAAGATGGACAAAAAGACT-3′, XbaI-Cry2-rev 5′-AACTCTAGATGCTGCTCCGATCATGATCT-3′). A stop codon C-terminal to mApple was deleted using a Q5 mutagenesis kit (New England Biolabs), Q5SDM_mCh-Tomm20-for 5′-GCGGCCGCCATGAAGATG-3′ and Q5SDM_mCh-Tomm20-rev 5′-CTTGTACAGCTCGTCCATGC-3′. All plasmids were verified by sequencing.

Two different beta-cell lines were used in this study: the mouse cell line MIN6 (Ishihara et al., 1993) and the rat cell line INS-1E (Merglen et al., 2004). Cell lines were tested for mycoplasma at regular intervals. MIN6 cells were cultured in Dulbecco's modified Eagle medium (4.5 g/l D-glucose), supplemented with 15% fetal calf serum (FCS) and 50 µM β-mercaptoethanol. The INS-1E cells were kept in RPMI-1640 medium supplemented with 50 µM β-mercaptoethanol, 1 mM sodium pyruvate, 10 mM HEPES and 10% FCS (all Life Technologies, The Netherlands). All cells were cultured at 37°C and 5% CO₂.

Cells were plated on 25-mm glass coverslips and grown for 24–48 h until reaching ∼60% confluence. Expression vectors were subsequently transfected into cells using Lipofectamine 2000 (Life Technologies) according to the manufacturer's instructions. Cells were used for experimentation 24–48 h after interruption of the transfection reaction.

Blue light-induced recruitment of mitochondria to the subplasma membrane region of cells was accomplished by epifluorescence illumination at the stage of a TIRF microscope using a custom-built illuminator. Briefly, 470-nm light provided by a light-emitting diode (LED) (200 mW, ThorLabs, Newton, NJ) was collimated and focused to a 10-mm (∅) spot that was projected to the cells through the bottom of a quartz prism used to generate evanescent wave excitation (Idevall-Hagren et al., 2010). Illumination (10% output power) was provided as 100-ms pulses every 10 s. In some experiments, cells were instead exposed to blue-light illumination for 10–60 min inside a cell culture incubator using a collimated 470-nm blue-light LED as previously described (Xie et al., 2016). Following illumination, cells were placed at the stage of a TIRF microscope and continuously illuminated with 491-nm blue light derived through the evanescent field.

Selective visualization of the plasma membrane/subplasma membrane region was performed by TIRF microscopy, using either an inverted microscope and a 60×/1.45 NA objective or a prism-based system built around an upright microscope equipped with a 16×/0.8 NA objective, as previously described (Idevall-Hagren et al., 2010; Xie et al., 2016). Before all imaging experiments, cells were pre-incubated for 30 min at 37°C in a buffer containing 125 mM NaCl, 4.9 mM KCl, 1.3 mM CaCl₂, 1.2 mM MgCl₂ (all Merck, Germany), 10 mM HEPES, 0.1 mg/ml bovine serum albumin (Sigma-Aldrich) and 3 mM glucose (pH 7.40) (Sigma-Aldrich). The cells were subsequently mounted in modified Sykes–Moore chambers, placed at the stage of the microscope and superfused with buffer. Temperature was maintained at 37°C throughout all experiments.

Image analysis was performed using Fiji software (Schindelin et al., 2012). For quantifications of subplasma membrane changes in fluorescence, regions of interest were identified on the TIRF micrographs based on the cell footprint as visualized by a general plasma membrane stain (iRFP-PH-PLC_δ1). The changes in fluorescence of other markers within this region over the time course of an experiment were subsequently determined, and the data were normalized to the pre-stimulatory level using Excel (Microsoft Corp., Redmond, WA). All experiments were repeated a minimum of three times, unless stated otherwise, and statistical analysis – using one-way ANOVA for multiple comparisons or two-tailed paired or unpaired Student's t-tests – was performed using Prism (version 7.00 for Mac, GraphPad Software, La Jolla, CA). P<0.05 was considered statistically significant.

We thank Antje Thonig for excellent technical assistance and Erik Gylfe for critical reading of the manuscript.