The multi-C2 domain protein dysferlin localizes to the plasma membrane and the T-tubule system in skeletal muscle; however, its physiological mode of action is unknown. Mutations in the DYSF gene lead to autosomal recessive limb-girdle muscular dystrophy type 2B and Miyoshi myopathy. Here, we show that dysferlin has membrane tubulating capacity and that it shapes the T-tubule system. Dysferlin tubulates liposomes, generates a T-tubule-like membrane system in non-muscle cells, and links the recruitment of phosphatidylinositol 4,5-bisphosphate to the biogenesis of the T-tubule system. Pathogenic mutant forms interfere with all of these functions, indicating that muscular wasting and dystrophy are caused by the dysferlin mutants' inability to form a functional T-tubule membrane system.
The T-tubule system is an extensive network formed by invaginations of the plasma membrane and accounts for about 80% of the plasma membrane surface of striated muscle (Al-Qusairi and Laporte, 2011). It functions in excitation–contraction (EC) coupling by propagating the action potential from the plasma membrane into the muscle fiber, ultimately leading to Ca2+ release from the sarcoplasmic reticulum (SR) and to muscle contraction. The T-tubule system is involved in the regulation of muscle fatigue, muscle differentiation, intracellular trafficking and plasma membrane repair (Al-Qusairi and Laporte, 2011; Engel and Franzini-Armstrong, 2004; Kerr et al., 2014; Klinge et al., 2007; Krolenko and Lucy, 2002; Lee et al., 2002). However, the molecular mechanisms required for biogenesis, maintenance and function of this cellular organelle are only partially understood (Al-Qusairi and Laporte, 2011; Engel and Franzini-Armstrong, 2004). So far, two proteins directly involved in the biogenesis of the T-tubule system have been identified. Caveolin-3 is critical for the generation of beaded invaginations of the plasma membrane (Franzini-Armstrong, 1991), and the muscle isoform of Bin1 (also known as amphiphysin 2) directly acts in formation of the smooth tubular profiles of the T-tubule system (Lee et al., 2002; Posey et al., 2014). Mutations in genes coding for Bin1 and caveolin-3 each induce different muscle diseases (Bohm et al., 2014), with alterations in Ca2+ homeostasis arising from Bin1 deficiency (Tjondrokoesoemo et al., 2011). Indirect observations have pointed to a function of dysferlin in T-tubule biogenesis (Al-Qusairi and Laporte, 2011; Demonbreun et al., 2013; Kerr et al., 2014, 2013; Klinge et al., 2010b, 2007). Mutations in the DYSF gene lead to limb-girdle muscular dystrophy type 2B (LGMD2B) or Miyoshi myopathy (MM), characterized by progressive muscular weakness and wasting, which typically start in the second decade of life. Interestingly, about half of dysferlin-deficient patients show an unusual and ambiguous disease phenotype by displaying increased levels of fitness before the onset of symptoms. This is in stark contrast to patients with any other form of muscular dystrophy (Biondi et al., 2013; Klinge et al., 2010a). So far, no treatments for the cause of these diseases are available, but recent gene transfer approaches are able to restore proper localization of the dysferlin full-length protein (Escobar et al., 2016; Pryadkina et al., 2015; Sondergaard et al., 2015).
Dysferlin, together with otoferlin, myoferlin and Fer1L4 to Fer1L6, belongs to the ferlin family, a group of vesicle fusion proteins. Dysferlin is a 230 kDa type II transmembrane protein containing seven C2 domains. C2-domain proteins often promote Ca2+-dependent membrane interaction and fusion (Lemmon, 2008), and the C2A domain of dysferlin binds to phosphatidylinositol 4-phosphate and phosphatidylinositol 4,5-bisphosphate [PI(4,5)P2] (Davis et al., 2002; Fuson et al., 2014; Therrien et al., 2009), which is an important regulator of cellular trafficking, ion channels and Ca2+ homeostasis (Balla, 2013; Hilgemann et al., 2001). The first C2 domain, C2A, is essential for phosphatidylserine accumulation at the membrane repair site (Middel et al., 2016). Dysferlin localizes to the plasma membrane and to the T-tubule membrane in striated muscle (Ampong et al., 2005; Anderson et al., 1999; Klinge et al., 2008), and transits to late endosomes, similar to myoferlin (Redpath et al., 2016). Dysferlin acts at intracellular membranes (Bansal et al., 2003; Marty et al., 2013), and the alteration of this process is held responsible for a defect in membrane repair (Bansal et al., 2003; Cooper and Head, 2014) during muscle regeneration (Chiu et al., 2009), and for changes in T-tubule morphology and the function of dysferlin-deficient muscle (Klinge et al., 2010b; Roostalu and Strähle, 2012). However, it is unknown how the dysferlin protein is mechanistically engaged in these functions (Al-Qusairi and Laporte, 2011; Kerr et al., 2014).
Here, we dissect the cellular and molecular function of dysferlin with respect to its membrane binding and membrane-deforming properties. Further, we analyze the influence of disease-related mutations on these functions and develop a model of the morphogenetic role of dysferlin in T-tubule biogenesis.
Dysferlin mediates membrane tubulation in vivo and in vitro
To study the function and intracellular localization of dysferlin, we expressed an EGFP-tagged full-length dysferlin construct in non-muscle cells that do not endogenously express this protein. Using HeLa and COS-7 cells, we found that dysferlin localizes to highly dynamic tubular and vesicular structures (Fig. 1A; Movie 1). Cholesterol depletion disrupted these structures and confirmed that the sites of dysferlin localization are membranes (Fig. 1B). Nocodazole treatment led to disintegration of these membrane tubules, indicative of their association with microtubules (Fig. 1B). This is consistent with an interaction of the dysferlin protein with α-tubulin (Azakir et al., 2010). To test whether the dysferlin tubules are accessible to externally added dye, we expressed EGFP–dysferlin in HeLa cells and stained the cells with the membrane-impermeable fluorescent dye FM4-64 for 40 s. Colocalization of this dye with dysferlin-induced membrane tubules was observed by live imaging immediately after removing the dye (Fig. 1E). Quantification of intracellular FM4-64 fluorescence showed a significantly higher intensity in dysferlin-transfected cells compared to in EGFP-transfected cells (Fig. 1D). Thus, in non-muscle cells, the dysferlin tubules are accessible from the exterior, either through membrane continuity with the plasma membrane or through endocytosis. These results indicate that the dysferlin-induced structures are morphologically similar to the T-tubules in skeletal and cardiac muscle. Further, we wanted to know if the intracellular membrane structures to which dysferlin was found to localize were pre-existing cellular organelles or if dysferlin was able to induce these structures de novo by recruiting intracellular membranes. We coexpressed dysferlin–Myc with GFP–Rab7, furthermore, we then expressed EGFP–dysferlin and labeled the ER protein PDI, the Golgi protein 58K (also known as FTCD), peroxisomal membrane protein Pex14 and lysosomal membrane protein Lamp1. Additionally, we visualized mitochondria using MitoTracker and the cytoskeleton by staining tubulin and F-actin. We found that the tubules induced by dysferlin did not colocalize with organelle markers or the cytoskeleton (Fig. S1A,B).
We next tested whether these tubule-inducing properties were shared by other members of the ferlin family. Remarkably, the dysferlin paralogs otoferlin and myoferlin did not induce intracellular membranes in non-muscle cells when overexpressed under the same conditions (Fig. 1F). In contrast to dysferlin, myoferlin and otoferlin colocalized with the ER marker PDI (Fig. 1F).
To investigate the physiological relevance of these findings, we analyzed five known patient dysferlin mutations (V67D in C2A, G299W in C2B, R959W in C2D, R1331L located between C2D and C2E, L1341P in C2E) and truncated dysferlin constructs, including the so-called minidysferlins (Azakir et al., 2012; Krahn et al., 2010). All minidysferlins, truncated forms and point mutations in C2 domains failed to induce membrane tubulation in non-muscle cells. Notably, the dysferlin mutant R1331L, bearing an amino acid change in the domain connecting C2D and C2E, partly preserved membrane tubulation (Fig. 1G,H). These findings suggest that an alteration of dysferlin-dependent membrane tubulation is relevant to dysferlin-associated muscle pathology and that the C2 domains have distinct roles in the context of the protein.
In order to assess membrane-binding properties of dysferlin more directly and in vitro, we analyzed full-length dysferlin and mutants in a flotation experiment using liposomes prepared from a Folch lipid fraction. Protein and liposomes were loaded at the bottom of a sucrose step gradient, and liposomes were floated by centrifuging and analyzed for the presence of bound protein by western blotting. We found that wild-type, but neither the pathogenic dysferlinV67D nor minidysferlin-3 (MD3), bound to the membranes (Fig. 2A). This finding indicates that the integrity of the full-length protein is required for membrane attachment, because a mutation in the C2A domain prevented binding, but that the wild-type C2A, C2F and C2G domains (MD3) are insufficient to execute this function. Taken together, dysferlin tubulates cellular membranes and induces structures in non-muscle cells that resemble the T-tubule system. This function depends on full-length dysferlin and is highly sensitive to pathogenic mutations.
To visualize the effect of dysferlin on membranes more directly, we incubated liposomes with recombinant full-length dysferlin and analyzed these samples by negative staining and electron microscopy. Recombinant Bin1 was used as a positive control (Fig. 2B). Dysferlin extruded liposomes into membrane tubules, demonstrating the potent activity of dysferlin in membrane tubulation in vitro (Fig. 2B). Again, this activity was prevented in the dysferlinV67D point mutation (Fig. 2B) and the MD3 truncation (Fig. 2B). These findings support the hypothesis that defective tubulation is a crucial factor for the development of dysferlin-deficient muscle pathology.
To gain further insight into the functional relevance of dysferlin-dependent membrane tubulation, we studied a model of T-tubule biogenesis. Human dysferlin-deficient myoblasts were differentiated in culture into multinucleated fibers and their T-tubule system was stained with the lipophilic dye DiIC16(3). Highly compact tubular aggregates were observed in 36% of dysferlin-deficient fibers but not in any of the control fibers (Fig. 2C,D; Fig. S1C). This underlines the hypothesis that dysferlin is crucial for the generation of the T-tubule system in skeletal muscle.
The importance of dysferlin function in T-tubule biogenesis was additionally confirmed in an in vivo model of mouse muscle regeneration. The snake venom notexin induces breakdown of skeletal muscle followed by rapid regeneration, providing a model of accelerated muscle and T-tubule biogenesis in vivo. The T-tubule system was visualized using the previously described Ca2+-K+ ferrocyanide method and electron microscopy (Franzini-Armstrong, 1991). During notexin-induced muscle regeneration, large vacuoles appeared within T-tubules of DYSF gene-deleted mouse muscle (Fig. 2E), exacerbating the phenotype of abnormal T-tubule morphology in non-regenerating muscle (Klinge et al., 2010b). T-tubule alterations were not observed in dysferlin-competent controls and mdx mouse model muscle (containing a mutation in the dystrophin gene; a disease control sample) (Fig. 2E), indicating that the membrane tubulation effect of dysferlin that was observed in vitro is relevant in vivo.
Functional interaction of dysferlin with proteins involved in T-tubule biogenesis
Bin1, one of the proteins involved in T-tubule biogenesis, is also able to induce membrane tubulation in non-muscle cells (Lee et al., 2002). Both proteins, dysferlin and Bin1, induced tubulation in about 50% of transfected cells (Fig. 3A). We coexpressed Bin1 and dysferlin heterologously in COS-7 cells and found partial colocalization (Pearson colocalization coefficient=0.16±0.03, N=10 cells) at membrane tubules (Fig. 3B). Bin1 coexpression did not influence dysferlin's tubulation ability (Fig. 3C). To dissect Bin1 and dysferlin-induced tubules in more detail, we ‘skeletonized’ the tubule images and found that the number and length of tubule branches was similar for both proteins, but the number of end points was significantly increased in dysferlin-transfected cells, whereas the number of junctions was increased in Bin1-transfected cells (Fig. 3D,E). This finding demonstrates that dysferlin-induced tubules are shorter and less branched compared to Bin1 tubules. Three-dimensional (3D) reconstructions of Bin1 and dysferlin in COS-7 cells and C2C12 muscle cells showed partial colocalization at the same structures (Fig. 3F).
The Bin1-induced membrane tubules are known to attract dynamin2 (Lee et al., 2002). Interestingly, however, endogenous dynamin2 did not colocalize with those membrane tubules that had been induced by dysferlin alone (Fig. 3G,H). When dysferlin and Bin1 were coexpressed, colocalization between dysferlin and dynamin2 was significantly increased in comparison to that upon single transfection of dysferlin (Fig. 3G). However, the dynamin2 signal overlapped significantly more with that of GFP–Bin1 than with that of dysferlin–Myc. These findings indicate that Bin1 and dysferlin exert overlapping but distinguishable functions.
Besides Bin1, caveolin-3 is also involved in T-tubule biogenesis. This protein additionally is known to regulate intracellular dysferlin trafficking (Hernandez-Deviez et al., 2008). Therefore, we analyzed the influence of caveolin-3 on dysferlin-induced tubules. We expressed dysferlin together with caveolin-3 heterologously in HeLa and COS-7 cells and found reduced dysferlin-mediated membrane tubulation (Fig. 4A-C). This suggests a negative regulatory effect of caveolin-3 on dysferlin function.
Lipid specificity of dysferlin
T-tubules of striated muscle are enriched in PI(4,5)P2, and this phospholipid also marks the sites of Bin1 activity (Lee et al., 2002; Milting et al., 1994; Wu and Baumgart, 2014). In addition, the individual C2 domains of dysferlin are known to bind to phospholipids (Davis et al., 2002; Therrien et al., 2009).
We wanted to study the lipid membrane specificity of dysferlin in a cellular system and therefore coexpressed the GFP-labeled plextrin homology (PH) domain of phospholipase C δ1 (PLCδ1) as a sensor for PI(4,5)P2 with Myc-tagged dysferlin in C2C12 myoblasts. The PI(4,5)P2 sensor colocalized with dysferlin at the plasma membrane, and showed a remarkable colocalization at the T-tubule system (Fig. 5A). To test whether dysferlin-induced membrane tubules in non-muscle cells share this property, we coexpressed PH-PLCδ–GFP and dysferlin in COS-7 cells. Dysferlin expression led to a massive formation of PI(4,5)P2-containing tubules within the cytoplasm (Fig. 5B). The distribution of the PI(4,5)P2 sensor was shifted away from the plasma membrane towards an intracellular location (Fig. 5B,H), as compared to cells without dysferlin overexpression. This might indicate that dysferlin recruits PI(4,5)P2 from the plasma membrane towards intracellular membrane tubules. Additionally, dysferlin tubules colocalized with the YFP-tagged PH domain of Akt, a biosensor for PI(3,4,5)P3 (Fig. S2A). To further study the affinity of dysferlin for these phosphoinositides, we generated PI(4,5)P2-rich vacuoles through expression of a constitutively active Arf6 mutant, Arf6Q67L, or of phosphatidylinositol 4-phosphate 5-kinase (PI4P5 kinase) (Donaldson et al., 2003; Krauss et al., 2003) (Fig. S2B,C). The specific recruitment of dysferlin to these vacuoles corroborated the affinity of the protein for PI(4,5)P2 (Fig. 5C,D). The PI(4,5)P2–dysferlin association showed an intriguing sensitivity to pathogenic mutations, because dysferlinV67D and other deletions and truncations were not recruited to PI(4,5)P2 vacuoles (Fig. 5E,I; Fig. S2D,E). Importantly, expression of the PI(4,5)P2-degrading phosphatase derived from synaptojanin-1 blocked tubule formation in non-muscle cells and altered the T-tubule system in C2C12 myotubes (Fig. 5G). These results show that dysferlin-induced tubules are dependent on PI(4,5)P2 and, furthermore, that PI(4,5)P2 is important for T-tubule biogenesis.
Taken together, dysferlin binds to PI(4,5)P2 and specifically recruits this phospholipid into dysferlin-induced tubules. Moreover, the formation of these tubules is dependent on PI(4,5)P2. Therefore, dysferlin-induced membrane tubules in non-muscle cells share biochemical characteristics of the T-tubule system. The finding that known disease-related DYSF mutations interfere with these properties suggests that defects in lipid binding and tubule formation are relevant for the pathogenesis of dysferlin-deficient muscular dystrophy.
Dysferlin deficiency increases SR Ca2+ release and exercise capacity
As our data point to a surprisingly immediate role of dysferlin in T-tubulogenesis, and as the T-tubule system has a crucial function in EC coupling, we analyzed intracellular Ca2+ homeostasis in wild-type and DYSF knockout mice. In the mutant mice, the last three exons of the DYSF gene (12 kb) were deleted, as previously described (Bansal et al., 2003). We isolated single fibers from flexor digitorum brevis (FDB) muscle, Fura-2-AM-loaded the fibers and analyzed Ca2+ transients using epifluorescence microscopy. Ca2+ transient amplitude in DYSF gene-deleted muscle was significantly increased in comparison to that in wild-type fibers (Fig. 6A,B). The measurements also showed, however, that the function of the SR Ca2+-ATPase SERCA, which regulates Ca2+ transfer back into the SR was unaltered (Fig. S3A). In addition, coupling of L-type Ca2+ channel (also known as dihydropyridine receptor; DHPR) and ryanodine receptor (RyR) appeared to be unaltered as no differences in the time to the peak of Ca2+ transients was observed (Fig. S3A). Also, no differences in baseline fluorescence were detected, indicating similar resting cytoplasmic Ca2+ levels in wild-type and DYSF gene-deleted fibers (Fig. S3A). We considered that increased Ca2+ release was either due to increased SR Ca2+ content or due to increased store-operated Ca2+ entry (SOCE) and tested these hypotheses in turn. Caffeine was used to open the RyR and induce SR Ca2+ release. These Ca2+ transients did not differ in wild-type and DYSF gene-deleted muscle fibers, suggesting that SR Ca2+ content was unaltered (Fig. S3B,C). We measured SOCE after depletion of the SR Ca2+ with thapsigargin. Ca2+ entry after store depletion was found to be similar in DYSF gene-deleted and wild-type fibers (Fig. S3D). mdx mouse model fibers served as a control (Boittin et al., 2006) and showed increased SOCE (Fig. S3D). In order to test whether dysferlin deficiency leads to changes in the expression of proteins required for EC coupling, which then could be made responsible for altered Ca2+ release, we performed western blot and quantitative (q)PCR analyses in wild-type and DYSF gene-deleted muscle. DHPR, RyR1, RyR1-stabilizing protein calstabin1 (also known as FKBP1A), Bin1, SERCA (encoded by ATP2A2), two SOCE-components Orai1 and Stim1, the Ca2+-binding protein calsequestrin and the triad proteins junctophilin-1 and mitsugumin-29 (also known a s SYPL2) were tested. Protein and mRNA levels of proteins involved in EC coupling did not differ significantly between wild-type and DYSF gene-deleted muscle (Fig. S4A-C). This is in agreement with our hypothesis that the dysferlin protein exerts its function directly at the T-tubule system. Immunofluorescence staining of the two main proteins involved in the EC coupling process, RyR1 and DHPR, did not show altered localization in wild type compared to in DYSF gene-deleted muscle fibers (Fig. S4D).
Finally, we asked if dysferlin-deficient mice mirrored the remarkable phenotype of many dysferlin patients who show increased muscle prowess before onset of disease. We analyzed voluntary wheel running distance and velocity of dysferlin-null mice in a longitudinal experiment using young (4 weeks), adolescent (12, 20, 40 weeks) and aged mice (75, 90 weeks). Daily voluntary running behavior of individually housed mice was recorded using a conventional running wheel that was connected to a computer. Concordant with elevated Ca2+ transients, young and adolescent dysferlin-null mice had greater exercise capacity than wild-type mice (Fig. 6C,D). As there was no shift in animal weight and fiber-type spectra, and no fiber hypertrophy (Fig. S5E-G), increased myoplasmic Ca2+ release could be the origin of increased exercise capacity. After the age of 12 weeks, a decline in muscle function due to progressive muscle disease was observed (Fig. 6C,D), in agreement with reduced muscle strength after this age (Grose et al., 2012).
Dysferlin is involved in skeletal muscle membrane repair and T-tubule biogenesis, but its mechanism of action has remained unclear (Ampong et al., 2005; Bansal et al., 2003; Cooper and Head, 2014; Demonbreun et al., 2013; Kerr et al., 2013; Klinge et al., 2010b, 2007). We addressed the biochemical and cellular function of dysferlin in several experimental systems: first, dysferlin was expressed in non-muscle cells. This system reveals the phenotype of dysferlin expression in the absence of endogenous dysferlin and of other known T-tubule proteins, such as Bin1 or caveolin-3. The membrane-tubulating function of dysferlin in this system is a strong indicator of its portrayed role as a membrane morphogenetic factor. This model is supported by the finding that dysferlin tubules are accessible from the plasma membrane; dysferlin tubules are in contact with the plasma membrane either directly or through rapid endocytosis (Oulhen et al., 2013). The expression of dysferlin in non-muscle cells was further used to study the lipid-binding specificity of dysferlin, and we show that dysferlin recruits the phospholipid PI(4,5)P2. These findings are substantiated by the analysis and expression of dysferlin in C2C12 cells. C2C12 cells are mouse myoblast cells that show morphological and functional features of muscle cells upon differentiation. In these cells, dysferlin localized to the T-tubule system, which was highly dependent on PI(4,5)P2. Further, expression of dysferlin in vitro supports the findings of the other experimental systems with regard to membrane tubulation and lipid specificity. In those systems, human dysferlin-deficient myoblasts showed highly aggregated T-tubules, and regeneration after muscle injury in vivo revealed abnormal T-tubule morphology in DYSF gene-deleted mouse muscle, underlining the role of dysferlin in membrane tubulation. Lastly, analysis of the dysferlin-deficient mouse model pointed to abnormal Ca2+ homeostasis, explaining the unusual running behavior that we observed in our longitudinal experiment and the ambiguous disease phenotype of dysferlin-deficient patients. Importantly, all the functions were confirmed by the use of patient-derived mutations (in cellular and in in vitro systems).
Dysferlin tubulates membranes in vitro and in cellular systems
The membrane tubules generated by dysferlin in non-muscle cells exhibit morphological and biochemical characteristics similar to those of the T-tubule system of skeletal muscle, suggesting a specific effect of dysferlin in T-tubule formation. We propose that the capacity to shape the T-tubule system is the primary molecular and cellular function of the dysferlin protein.
A membrane curvature-generating function of dysferlin is in line with findings indicating a membrane sculpting function of the C2 domains of dysferlin (Marty et al., 2013). It has been proposed that the C2 domains in a multi-C2 domain protein act synergistically to create the membrane curvature that will lead to tubulation (McMahon et al., 2010). Three or more C2 domains are present in members of the ferlin family, ‘the multiple C2 domain and transmembrane region proteins’ (MCTPs) (Shin et al., 2005) and in the extended synaptotagmins (Min et al., 2007). The latter are able to mediate Ca2+-dependent phospholipid binding, whereas the MCTPs do not seem to interact with negatively charged or neutral phospholipids (Shin et al., 2005) but instead bind to Ca2+ with high affinity. We hypothesize that one or several of the C2 domains of dysferlin can insert into the proximate leaflet of the lipid bilayer like a wedge, and further, that several of these membrane-curvature generating domains are oriented longitudinally to create (T-)tubules. Our work, however, also shows that tubule formation is not a universal property of the ferlin-family members, as the expression of otoferlin or myoferlin does not induce membrane tubulation. Otoferlin and myoferlin are dysferlin paralogs with multiple C2 domains and a similar domain structure. This highlights a distinct function of dysferlin within the ferlin-protein family.
We show that dysferlin-induced membranes are able to recruit those phospholipids that are an important component of the T-tubule system. When we degraded PI(4,5)P2, dysferlin could no longer induce tubule formations in non-muscle cells, indicating that the presence of PI(4,5)P2 is an essential requirement for dysferlin-dependent T-tubule formation. This finding is in agreement with the enrichment of PI(4,5)P2 in T-tubules (Lee et al., 2002; Milting et al., 1994). We could show that PI(4,5)P2 depletion also altered the T-tubule system in C2C12 myotubes, reinforcing the link between dysferlin-dependent phospholipid binding and the biogenesis of the T-tubule system. Given the known morphological defect of the dysferlin-deficient T-tubule system (Al-Qusairi and Laporte, 2011; Demonbreun et al., 2013), the localization of the protein to the developing T-tubule system (Klinge et al., 2010b) and the membrane tubulation properties of dysferlin, a direct role of dysferlin in the formation/biogenesis of this cellular organelle is implicated. Disease-related DYSF mutations interfere with phospholipid binding and membrane tubulation properties, suggesting that defects in lipid binding and tubule formation are relevant for the pathogenesis of dysferlin-deficient muscular dystrophy. This is corroborated by the results of our cell model of T-tubule biogenesis and our in vivo muscle regeneration experiments, which demonstrate a T-tubule-shaping function of dysferlin.
Bin1 induces membrane curvature in concert with PI(4,5)P2 and plays an important role in T-tubule biogenesis (Lee et al., 2002; Wu and Baumgart, 2014). These data together with our results indicate that dysferlin and Bin1 are both required for membrane tubulation and T-tubule formation. In spite of the functional similarities of these proteins, the structural differences we find in Bin1- and dysferlin-dependent tubules suggest that these proteins act on different sub-elements of the T-tubule system. Based on these structural differences, we further speculate that dysferlin localized at T-tubules has an additional downstream role, possibly with regard to plasma membrane repair.
The findings of our study also support a role of caveolin-3, the third protein involved in T-tubule biogenesis, as a negative regulator of dysferlin by modulating its intracellular trafficking. Caveolins inhibit endocytosis of dysferlin (Hernandez-Deviez et al., 2008), and this explains the reduction of dysferlin-induced membrane tubulation upon caveolin-3 overexpression. The independence of dysferlin-induced membrane tubulation from the GTPase dynamin2 suggests that Bin1 and dysferlin act synergistically but with distinct mechanisms, as Bin1 has been shown to functionally interact with dynamin2 in the generation of membrane curvature (Nicot et al., 2007). Different modes of action of dysferlin and Bin1, although participating in the biogenesis of the same cell compartment, might explain the development of distinct clinical phenotypes upon protein deficiency.
The T-tubule system functions in EC coupling and myoplasmic Ca2+ homeostasis. Our results show that dysferlin deficiency leads to abnormal Ca2+ homeostasis with increased Ca2+ transient amplitudes after stimulation. This may also explain why muscle pathology in dysferlin deficiency can be ameliorated by blocking the L-type Ca2+ channel (Kerr et al., 2013) and by long-term suppression of SR Ca2+ release (Hattori et al., 2007).
A recent study found no difference in evoked Ca2+ transients in skeletal muscle of dysferlin-deficient mice and suggested that altered Ca2+ regulation following injury is the cause of T-tubule alterations (Kerr et al., 2013). As correction of membrane repair alone is not sufficient to improve dystrophic changes in dysferlin-deficient muscle (Lostal et al., 2012), we suggest a model in which dysferlin deficiency primarily leads to alterations of the T-tubule system. In our model, changes in Ca2+ transients are secondary to morphological and functional alterations of the T-tubule system due to dysferlin deficiency.
About half of the patients affected by dysferlin-deficient muscular dystrophy have an increased muscle performance and are good athletes before the onset of muscle weakness (Biondi et al., 2013; Klinge et al., 2010a). This phenotype is unique amongst the muscle wasting disorders, and the underlying pathophysiology is unknown. Increased Ca2+ transient amplitudes are not due to altered expression or localization of T-tubule or triad proteins, and they are not due to increased SR Ca2+ content or SOCE. They might rather be a result of altered SR Ca2+-release kinetics due to altered channel gating, clustering or trafficking processes as these mechanisms are under regulation of PI(4,5)P2 (Rodríguez-Menchaca et al., 2012), which in turn is controlled by dysferlin in skeletal muscle. We therefore suggest that mutant dysferlin induces the high Ca2+ transients responsible for good muscle prowess before the onset of weakness, whereas it is the primary alteration of T-tubule membranes in dysferlin-deficient muscle that underlies defective membrane repair and differentiation, ultimately leading to muscular dystrophy.
MATERIALS AND METHODS
DYSF gene-deleted mice, Dysftm1Kcam/Dysftm1Kcam, were kindly provided by Professor Kate Bushby, Newcastle University, UK, with approval of the Mutant Mouse Regional Resource Centers (MMRRC), USA. Wild-type littermates were used as controls. Mice were housed under standard conditions and were treated in accordance with the European convention for the protection of vertebrate animals used for experimental and other scientific purposes (ETS 123). The experiments were approved by permits issued by the Landesamt für Verbraucherschutz, Oldenburg, Germany (Aktenzeichen 33.14.42502-04/085/08 and 33.9-42502-04-10/0192).
pcDNA3.1-dysferlin-Myc was derived from pcDNA4-EGFP-DFL. The DYSF sequence was cloned into EcoRI and NotI sites of pcDNA3.1(+)Myc-His-vector, and the stop codon was deleted using primers OST630 and OST631. Pathogenic mutations were introduced into the dysferlin constructs by DpnI-mediated site-directed mutagenesis using primers OST907 and OST908 for pcDNA4-EGFP-DYSFV67D (C2A mutation), OST575 and OST576 for pcDNA4-EGFP-DYSFG299W (C2B mutation), OST577 and OST578 for pcDNA4-EGFP-DYSFR959W (C2D mutation), OST579 and OST580 for pcDNA4-EGFP-DYSFR1331L (mutation between C2D and C2E) and OST581 and OST582 for pcDNA4-EGFP-DYSFL1341P (C2E mutation). Two truncated constructs containing the last two C2 domains with (pcDNA3.1-EGFP-DYSF-C2FGTM) or without (pcDNA3.1-EGFP-DYSF-C2FG) the transmembrane (TM) domain were amplified using primers JH12 and JH3 (with TM) or JH10 (without TM). Another transmembrane-deletion mutant (pcDNA4-EGFP-DYSFΔ™) was generated by DpnI-mediated site-directed mutagenesis with primers OST616 and OST617. pcDNA4-EGFP-minidysferlin-3 (MD3) was generated by DpnI-mediated site-directed mutagenesis with primers JH53 and JH54 (Azakir et al., 2012). The TagRFP sequence was amplified using primers pKG9 and pKG10, and inserted into HindIII and EcoRI sites of pcDNA3.1-Myc-His-vector. The BIN1 sequence was amplified using primers pKG7 and pKG8, and inserted into EcoRI and NotI sites of pcDNA3.1-TagRFP-vector. For bacterial expression constructs of dysferlin, the pET41a vector was cleaved by NdeI and BglII and Klenow fill-in with following blunt ligation was performed to delete the glutathione S-transferase (GST) tag from the vector. N-terminal His6-tagged DYSF was then inserted into KpnI and NotI sites of the vector (pET41aΔN-B-His6-DYSF). Mutation DYSFV67D was inserted using DpnI mutagenesis using primers OST907 and OST908 (pET41aΔN-B-His6-DYSFV67D). MD3 was further cloned into EcoRI and NotI sites of pGEX6p1 vector (pGEX6p1-minidysferlin-3). The BIN1 sequence was cloned into BamHI and NotI sites of pGEX6p2-vector (pGEX6p2-Bin1). All constructs were verified by DNA sequencing. All oligonucleotides used in this study are listed in Table S1 and all constructs are listed in Table S3.
Cell culture, immunofluorescence, live staining and microscopy
HeLa and COS-7 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) containing 1% glutamine, 1% penicillin–streptavidin and 10% fetal calf serum (FCS). C2C12 cells were cultured in DMEM containing 15% FCS. Differentiation was induced by replacing FCS with 3% horse serum. DNA constructs were transfected using Effectene Transfection Reagent (Qiagen) and were observed after 48 h. For immunofluorescence staining, cells were fixed for 20 min with 4% paraformaldehyde (PFA) and blocked with 3% horse serum and 0.5% saponin in PBS. Primary and secondary antibodies were diluted in blocking solution, and cells were incubated with antibodies for 1 h at room temperature. Myc-tagged and hemagglutinin (HA)-tagged constructs were stained with an anti-Myc antibody (1:200, Cell Signaling, 2272S) or anti-HA antibody (1:100, ab9110, Abcam). Anti-PDI (1:500, ab2792), anti-Lamp1 (1:600, ab24170), anti-Golgi (1:100, Golgi protein 58K, ab27043), anti-dysferlin (1:500, ab124684) and anti-dynamin2 (1:500, ab151555) antibodies were purchased from Abcam; anti-Pex14 antibody was purchased from ProteinTech (1:500, 10594-1-AP). Anti-Cav1.2 (1:50, sc-103588) and anti-amphiphysin-II (anti-Bin1) (1:500, sc-13575) antibodies were purchased from Santa Cruz. Secondary antibodies (1:500) conjugated to Cy3 (Jackson ImmunoResearch) or Alexa-Fluor-488 (MoBiTech) were used. Colocalization was analyzed by calculating Pearson and Spearman coefficients (PSCs) using ImageJ PSC colocalization plugin (imagej.nih.gov). EGFP fluorescence intensity was measured using ImageJ. Fluorescence was multiplied by the analyzed area, and the percentage of the area covered with fluorescence was calculated. For detubulation experiments, cells were treated with 10 mM β-methyl cyclodextrin (β-MCD) or 30 μM nocodazole for 30 min and fixed afterwards. Three lots of 100 cells were analyzed per mutant. For wild type and the mutants G229W, R959W and R1331L, nine lots of 100 cells were analyzed. Cells were observed with Axioimager M1 (Zeiss) equipped with a Plan Neofluar 100×/1.3 Oil lens and an AxioCam HRm camera (AxioVision Rel. 4.8 software, Zeiss), and z-stacks were deconvoluted. Analysis of Bin1 and dysferlin tubules in COS-7 cells was performed using confocal images (Inverted IX81 Olympus Microscope, Software FV10-ASW, UPLANSApo 60×/1.35 oil-immersion objective). A skeleton analysis was modified from Morrison and Filosa (2013) using ImageJ. Images were converted into binary images, skeletonized and analyzed using the AnalyzeSkeleton plugin (imagejdocu.tudor.lu/) to derive the number of branch points and end points, as well as the branch length. 3D visualization of confocal z-stacks was performed using Imaris software. Plasma membrane staining with FM4-64 (Life Technologies) was performed according to the manufacturer's instructions. Cells on cover slips were incubated for 40 s with 10 μM FM4-64 in HBSS (137 mM NaCl, 5.4 mM KCl, 0.25 mM Na2HPO4, 0.44 mM KH2PO4, 1.3 mM CaCl2, 1.0 mM MgSO4, 4.2 mM NaHCO3), washed with HBSS and imaged immediately during the following 3 min [LSM 710, ZEN 2009 software (Zeiss) equipped with a Plan-Apochromat ×63/1.40 oil-immersion objective]. ImageJ was used to analyze intracellular FM4-64 intensity. For time-lapse experiments, 4-well Ph+ µ-slides (ibidi) were used, and experiments were performed with an Olympus IX 81 microscope (Camera OBS Mega/View; software, xcellence pro 1.2; UPLANSApo 60×/1.35 or 100×/1.4 oil-immersion objective). For immunofluorescence analysis of skeletal muscle, single fibers were isolated from FDB muscles of wild-type and DYSF gene-deleted mice, seeded on laminin-coated cover slips and incubated overnight in DMEM containing 5% FCS. Then, fibers were washed with cold PBS, fixed with ice-cold 100% ethanol in −20°C for 20 min, blocked with 5% bovine serum albumin (BSA) and 0.5% Triton X-100. Fibers were incubated with primary antibodies (anti-RyR1, 1:100, Cell Signaling, 8153; anti-DHPR, 1:50, Thermo Scientific, MA3-920) overnight at 4°C and with secondary antibodies (1:500) conjugated to Cy3 (Jackson ImmunoResearch) or Alexa-Fluor-488 (MoBiTech) for 1 h at room temperature.
Human myoblasts [Muscle Tissue Culture Collection (MTCC)] were cultured in skeletal muscle growth medium (Promocell) containing 10% FCS, 1.5% glutamax (100×, Gibco) and 50 µg/ml gentamycin. Differentiation was induced by replacing the growth medium with DMEM containing 5% horse serum for 7 days. The membrane of human myoblasts was stained with DiIC16(3) (Thermo Fisher Scientific). The differentiated cells were washed with sucrose cacodylate buffer (SCB; 0.1 M sucrose, 0.1 M sodium cacodylate, pH 7.4), incubated with DiIC16(3) diluted to 12.5 µg/ml in SCB for 10 min at room temperature, and visualized by confocal microscopy (Inverted IX81 Olympus Microscope, software FV10-ASW, UPSLANSApo 60×/1.35 oil-immersion objective).
Constructs expressing His6–DYSF, His6–DYSFV67D, His6–DYSFΔ™, GST–minidysferlin and GST–Bin1 were transformed into competent BL21-RIL or BL21star/rosetta Escherichia coli cells. A 50 ml overnight culture was grown in Luria Bertani (LB) containing antibiotics. This culture was used to inoculate the main culture of 500 ml LB that contained antibiotics. At an OD600 of 0.6, protein expression was induced with 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) for 3 h at 30°C. The cells were harvested at 4066 g in a Sorvall GSA centrifugation rotor for 20 min, washed with cold PBS, centrifuged at 1464 g for 20 min. The cells were frozen at −80°C, thawed on ice and, per 1 g of pellet, the cells were resuspended in 10 ml of 300 mM NaCl in PBS, pH 7.3 containing Complete Protease Inhibitors without EDTA (Roche), 50 U DNase I and 50 μg/ml lysozyme. The resuspended cells were incubated for 30 min at 4°C and the cell lysate was frozen in 1 ml aliquots at −80°C. The lysate was thawed on ice and sonicated (Ultrasonic processor, Hielscher) twice for 10 s with an amplitude setting of 30%.
Quadriceps femoris muscle was isolated from 9- to 11-week-old wild-type and DYSF gene-deleted mice, frozen in liquid nitrogen and stored at −80°C. Muscle samples were homogenized in homogenization buffer (12.5 mM sucrose, 0.3 mM NaN3, 10 mM NaHCO3, pH7, 0.1 mM phenylmethylsulfonyl fluoride, Complete protease inhibitors) with the TissueRuptor (Qiagen), and protein concentration was determined by BCA assay (Uptima). Samples were mixed with 4× sample buffer and incubated for 5 min at 95°C. 20 μg of protein was loaded onto 8-12% SDS gels (depending on protein size) and transferred onto nitrocellulose membranes. Proteins were detected by antibodies against dysferlin (1:500, NCL Hamlet, Novocastra), Bin1 (1:500, Santa Cruz), mitsugumin-29 (1:1000, ab106438, Abcam), SERCA (1:2000, ab2818, Abcam), FKBP12 (1:1000, ab24373), L-type Ca2+ channel (1:200, MA3-920, Thermo), RyR (1:500, 8153, Cell Signaling), TRPC3 (1:1000, ab51560, Abcam), Orai1 (1:200, ab59330, Abcam), Stim1 (1:200, 610954, BD Biosciences), Calsequestrin (1:1000, ab3516, Abcam) and GAPDH (1:50000, G8795, Sigma). Secondary antibodies (1:10000) were conjugated with horseradish peroxidase (HRP) anti-rabbit or anti-mouse IgG. Densitometry was performed using ImageJ analysis software, and intensities were normalized to those of GAPDH.
Total RNA was isolated from quadriceps femoris muscle of wild-type and DYSF gene-deleted mice using peqGold TriFast™ (PEQLAB) according to the manufacturer's protocol. For cDNA synthesis, 1 µg of RNA was used with oligo-dT primers and Superscript III reverse transcriptase (Invitrogen) according to the manufacturer's instructions. The cDNA was stored at −20°C or used directly for qPCR using the primers listed in Table S2. All reactions were performed using SYBR Green Supermix PCR reagent (BioRad) and MyiQ Single-Color Real-Time PCR Detection System (BioRad). mRNA levels were normalized to those of GAPDH.
Lipid binding experiments
For preparation of liposomes, 10 μl Folch-fraction lipids (100 mg/ml, Sigma) with or without 5% PI(4,5)P2 (Avanti Polar Lipids) were evaporated under a nitrogen stream. Lipids were then hydrated with 750 μl hydration solution (20 mM Tris-HCl, pH 7.4, 100 mM NaCl, 150 mM sucrose, 1 mM EDTA). The solution was added dropwise to the lipids with vigorous vortexing. For liposome tubulation experiments, liposomes were extruded 21 times through 100-nm membranes (Echelon) afterwards. The liposome flotation assay was modified from Klopfenstein and Vale (2004). 5 μl of Bin1 protein (2 µg/µl) and 40 µl of dysferlin lysates (5 µg/µl) were added to 100 μl of liposomes in the presence of 2.5 mM CaCl2, and the liposome protein mixture was incubated for 20 min at 37°C. Sucrose concentration was adjusted to 1.6 M using a 2 M stock, and the protein–liposome mixture was overlaid with 300 µl of 1.4 M sucrose, 150 µl of 0.4 M sucrose and 300 µl of 0.25 M sucrose. Gradient centrifugation was performed in a TLS55 rotor for 60 min at 201,000 g and 4°C. Fractions of 200 µl were collected and analyzed by SDS-PAGE and immunoblotting.
Liposome tubulation assay
For liposome tubulation experiments, liposomes were prepared in hydration solution containing 30 mM Tris-HCl pH 7.4, 150 mM NaCl, 300 mM sucrose and 1 mM EDTA. 2.5 μl of protein lysate (10 µg/µl) and Bin1 protein (2 µg/µl) was added to 7.5 μl of liposomes in the presence of 2.5 mM CaCl2, and the liposome protein mixture was incubated at 37°C for 20 min. The liposome–protein mixture was diluted 1:10 with hydration solution, and 6 μl aliquots were absorbed onto formvar- and carbon-coated copper grids (Polysciences) for 45 s. The solution was blotted off onto Whatman paper and stained for 20 s with 2% phosphotungstic acid (PTA). The PTA solution was blotted off, and the grid was washed with hydration solution, blotted again, air dried and analyzed by electron microscopy (LEO EM912 Omega, Zeiss) with an on-axis 2048×2048-CCD camera (Proscan).
For T-tubule staining, soleus muscle from Dysftm1Kcam/Dysftm1Kcam, wild-type littermates and mdx mice was dissected out, stretched to its in vivo length, pinned and fixed in 2% glutaraldehyde in Sorensen's buffer for 15 min. Then muscles were trimmed, and midbelly segments were cut into blocks of 1.0 mm×0.5 mm×0.5 mm. These samples were incubated at room temperature overnight in 2% glutaraldehyde in Sorensen's buffer. They were postfixed in 2% osmium tetroxide with 0.8% potassium ferrocyanide for 1 h at room temperature. After dehydration, specimens were embedded in epoxy resin, sectioned at 80-90 nm thickness on a Leica ultramicrotome, collected onto grids and examined at 100 kV without further staining on a transmission electron microscope (LEO EM912 Omega, Zeiss).
In vivo regeneration
A cycle of degeneration and regeneration was induced in 12-week-old male mice through subcutaneous injection of notexin using a 0.5 ml U-100 insulin syringe (50 µl at 4 mg/ml in 0.9% NaCl; Latoxan) into the dorsolateral aspect of the left hindlimb so as to bathe the underlying soleus muscle in toxin, as described previously (Klinge et al., 2010b). Analysis was undertaken 10 days after notexin injection.
Succinate dehydrogenase staining
For succinate dehydrogenase (SDH) staining, gastrocnemius muscles were isolated from wild-type and DYSF gene-deleted mice at the age of 8 weeks, and muscles were embedded in mounting medium (Tissue-Tek), frozen in isopentane that had been cooled in liquid nitrogen. Sections were cut on a microtome and were incubated for 30 min at 37°C in SDH staining solution (0.011 mg/ml C4H4Na2O4, 0.024 M Na2HPO4×2H20, 0.024 M KH2PO4, pH 7.4, 0.7 mg/ml nitroblue tetrazolium, 10 mM MgCl2, 0.23 mg/ml menadione) and washed in water for 2 min. The slices were then fixed in 10% formalin for 10 min, rinsed in water and mounted with Aquatex (Merck). Images were analyzed using Axioimager M1 and AxioVision Rel. 4.8 software (Zeiss).
Isolation of single muscle fibers
FDB muscles were excised from 8- to 15-week old mice after isoflurane anesthesia and following cervical dislocation. Isolation of single skeletal muscle fibers was modified from Capote et al. (2005). FDB muscles were incubated in modified Tyrode solution (145 mM NaCl, 2.5 mM KCl, 1 mM MgSO4, 10 mM HEPES, 10 mM glucose, 2 mM CaCl2, pH 7.4) containing 4 mg/ml collagenase type 2 (Worthington) for 50 min at 37°C for enzymatic dissociation of muscle fibers. Muscle fibers were then stripped off from tendons mechanically and were dissociated mechanically by pipetting.
Measurement of intracellular Ca2+ transients, SR Ca2+ and SOCE
Ca2+ transients were measured using epifluorescence microscopy at room temperature. Isolated single skeletal muscle fibers were plated on laminin-coated glass chambers and incubated for 30 min until fibers were settled. Fibers were then incubated for 20 min with the Ca2+ indicator dye Fura-2-AM (10 μmol/l, MoBiTec) and mounted on the stage of a Nicon Eclipse TE 200-U inverted microscope. Cells were superfused with modified Tyrode solution at a constant flow of 80 ml/h. Fibers were stimulated in an electric field at 1 Hz. Fura-2-AM was alternatingly excited at 340 and 380 nm using a 75 W xenon arc lamp, and fluorescence emission at 510 nm was recorded using a photomultiplier (IonOptix Corp, Milton). Background fluorescence at 340 and 380 nm was subtracted, and the ratio of fluorescence emission was used as a measure of intracellular Ca2+ concentration. Analysis of Ca2+ transients was performed with IONWizard Version 5.0 (IonOptix). SR Ca2+ content was determined by measuring caffeine-induced Ca2+ transients. FDB fibers were incubated for 4 min in Tyrode's solution without Ca2+-containing 20 µM N-benzyl-p-toluene-sulphonamide (BTS). Release of SR Ca2+ content was evoked by addition of 30 mM caffeine and 20 µM BTS in Tyrode's solution. For measurement of SOCE, the SR Ca2+ store of FDB fibers was depleted through incubation with 5 µM thapsigargin, and Ca2+ entry was measured after re-addition of Ca2+ in Tyrode's solution. For Ca2+ measurements, N=4 mice and 67 (wild type), 61 (knockout) or 36 (mdx mouse model) fibers were analyzed. For caffeine experiments, N=8 mice and 29 (wild type), 27 (knockout) or 15 (mdx mouse model) fibers were analyzed.
Running wheel experiment
Wild-type and dysferlin-null mice were housed individually, and cages were mounted with a conventional running wheel when mice were at the age of 4, 12, 20, 40, 75 and 90 weeks over a period of 3 weeks. The running wheel was connected to a computer, and daily voluntary running activity was monitored. One revolution corresponds to a running distance of 0.38 m. A rotation sensor with a resolution of 16 per turn was connected to the wheel axis. Wheel activity was recorded using LabVIEW™-based custom software (National Instruments Corp.). In the first running period, the mice reached the maximum running performance at day 12 and in the following running periods at day 8. Average daily distance (m) and average daily running velocity (m/s) were calculated from all values above a threshold of 40% of the mean running distance started from day 12 or day 8, respectively. Running behavior of 17-20 wild-type and 14-23 dysferlin-null mice was analyzed in this experiment at time points 4-75 weeks. Some mice had to be excluded from the analysis at the different time points because the cage was not available for recordings due to technical reasons during the long time recordings (defect rotation sensor, computer break down). In the last running period, 13 wild-type and 9 dysferlin-null mice were analyzed as several animals had died before this point.
Statistical analysis was performed with Excel or GraphPad Prism 4 using Student's t-test or two-way ANOVA for repeated measurements. Data are presented as mean±s.e.m. P-values<0.05 were considered statistically significant.
We thank Irmgard Cierny, Elisabeth Ehbrecht and Marc Ziegenbein for technical assistance; Jens Schmidt for mdx mice; Gertrude Bunt (Molecular and Optical Live Cell Imaging facility Göttingen) for help with microscopy; Stephan Lehnart for access to the LSM 710 microscope; Tolga Soykan for advice on liposome preparations; Michael Meinecke for discussions; Steve Laval, Hanns Lochmüller and Ira Milosevic for critical reading of the manuscript; and Mikael Simons, Schanila Nawaz, Ira Milosevic, Ellen Reisinger, Tobias Meyer, Julie G. Donaldson and Pietro de Camilli for provision of plasmids. We thank the Muscle Tissue Culture Collection (MTCC) for providing the samples. The Muscle Tissue Culture Collection is part of the German network on muscular dystrophies (MD-NET, service structure S1, 01GM0601) and the German network for mitochondrial disorders (mito-NET, project D2, 01GM0862) funded by the German ministry of education and research (BMBF, Bonn, Germany). The Muscle Tissue Culture Collection is a partner of EuroBioBank and TREAT-NMD.
J.H. designed and performed experiments, interpreted data and contributed to writing of the manuscript. K.B. performed experiments, interpreted data and contributed to writing of the manuscript. R.B. and M.D. performed experiments and interpreted data. D.L. contributed to running wheel experiments and data analysis. V.O.N., S.W., L.S.M. designed experiments and interpreted data. J.G. contributed to the development of the study. S.T. and L.K. conceived the study, designed experiments, interpreted data and wrote the manuscript.
This work was supported by the Deutsche Forschungsgemeinschaft (DFG) [KL1868/2-1 and KL1868/5-1 to K.L., TH 1538/1-1 to S.T. and SFB1002/2 TP A10 to S.T.]; a PhD fellowship of the Hunsmann-Stiftung [to K.B.], the State of Lower-Saxony, Hannover, Germany to [ZN2921 to S.T.], the DFG International Research Training Group [GRK 1816 to S.W. and L.S.M.], the Deutsches Zentrum für Herz-Kreislaufforschung (German Centre for Cardiovascular Research) (to L.S.M., and S.W.) and and the Eva Luise and Horst Köhler Foundation.
The authors declare no competing or financial interests.