Myotubularin and PtdIns3P remodel the sarcoplasmic reticulum in muscle in vivo

The sarcoplasmic reticulum (SR) is a specialized form of the endoplasmic reticulum (ER) present in skeletal muscle and a well-organized intracellular membrane system essential for calcium homeostasis (Rossi et al., 2008). The SR is a convoluted network that surrounds the myofibrils and contacts several organelles in the muscle fibers (Flucher, 1992; Peachey, 1965; Porter and Palade, 1957; Veratti, 1961). It is composed of a network of tubules and cisternae forming the longitudinal SR (lSR) and the junctional SR (jSR), respectively. The jSR contacts sarcolemma invaginations called T-tubules to form the triad, composed of two circumferentially oriented and dilated jSR cisternae located on both sides of a T-tubule. Triads are the membrane structure underlying the excitation–contraction coupling (EC coupling). The SR is the main calcium store. Calcium intake is preferentially accomplished by SR/ER Ca²⁺ ATPase (SERCA) calcium pumps on the lSR while calcium release into the cytoplasm is mainly performed by the ryanodine receptor (RyR) at the jSR upon activation through interaction with the dihydropyridine receptor (DHPR), the later located on the T-tubule.

While numerous studies investigated the EC coupling and calcium homeostasis in skeletal muscle or described the SR organization, the mechanisms that govern SR remodeling and maintenance of specific membrane subdomains are still elusive.

Over recent years, several proteins regulating the shape of the ER in non-muscle cells have been identified (Shibata et al., 2010). ER tubules are shaped by two classes of proteins called reticulons and DP1/Yop1p. ER sheets may be formed and maintained by an intraluminal bridge between transmembrane proteins as CLIMP-63 or P180, while the high curvature of the sheet edge could be created by reticulons and DP1. Atlastin, a dynamin-like protein, participates to the fusion of ER membrane and thus to its high dynamics.

In skeletal muscle, the organization of specific SR domains results from a developmental process of membrane reorganization; the SR undergoes a series of modifications passing from an apparently disordered accumulation of membranes in the embryonic muscles to the very precise organization and architecture observed in the adult (Cusimano et al., 2009; Franzini-Armstrong, 2009; Luff and Atwood, 1971; Rossi et al., 2008). Proteins specifically regulating these processes have not been thoroughly described. However, manipulation of several proteins of the SR component of the triads was shown to impact the triad shape (Jones et al., 1998; Tijskens et al., 2003).

Several SR proteins are implicated in human muscle disorders (Al-Qusairi and Laporte, 2011). For example, RyR1 mutations were found in central core disease, rods myopathy, malignant hyperthermia and other congenital myopathies (Jungbluth et al., 2011; Treves et al., 2008). SERCA1 mutations associate with recessive Brody myopathy (Odermatt et al., 1996). Histological and ultrastructural observations of muscle biopsies from these patients showed frequent defects in triad morphology and EC-coupling. Similarly, disorganization of T-tubules was recently associated with centronuclear myopathies (CNM) in animal models and patients (Al-Qusairi et al., 2009; Dowling et al., 2009; Toussaint et al., 2011). The X-linked form of CNM (XLCNM) is associated with mutations in the MTM1 gene encoding myotubularin, a ubiquitously expressed lipid phosphatase that acts on bioactive phosphoinositide (PI) second messengers (Laporte et al., 1996). MTM1 dephosphorylates phosphatidylinositol 3-monophosphate (PtdIns3P) to phosphatidylinositol and phosphatidylinositol 3,5-bisphosphate [PtdIns(3,5)P₂] to phosphatidylinositol 5-monophosphate (PtdIns5P) (Blondeau et al., 2000; Chaussade et al., 2003; Taylor et al., 2000; Tronchère et al., 2004). PIs are lipid messengers that contribute to the unique identity of an organelle by targeting the localization of protein effectors (Di Paolo and De Camilli, 2006; Falasca and Maffucci, 2009). In cultured cells and yeast, PtdIns3P is enriched on early endosomes and intraluminal vesicles of multi-vesicular bodies (Gillooly et al., 2000; Michell et al., 2006). The role and localization of PtdIns3P in muscle tissue and its involvement in the XLCNM development has not been investigated in mammals.

We hypothesized that MTM1 and its substrate PtdIns3P could be implicated in membrane shape and remodeling in skeletal muscle. To address this potential link, we applied a combination of biochemical and imaging approaches to assess the localization of MTM1 in skeletal muscle and found that MTM1 is enriched on the SR component of the triad. Characterization of XLCNM patients cells and Mtm1-null mouse muscles, and modulation of the MTM1 activity in vivo upon ectopic expression of MTM1 wild-type (WT) and phosphatase-dead mutant (MTM1-C375S) using Adeno-associated virus (AAV) identified an important role for MTM1 activity on SR remodeling. We found that MTM1-WT promotes stacks of flat SR membranes while MTM1-C375S leads to highly curved cubic SR membranes. Electron microscopy experiments supported by biochemical assessment of PdtIns3P level showed that MTM1 and PdtIns3P regulate SR membrane shape in muscle. Our results provide evidence of an unexpected PI-related function of MTM1 on the SR in skeletal muscle.

Endogenous and overexpressed MTM1 was previously suggested to colocalize with triad markers (Buj-Bello et al., 2008; Dowling et al., 2009). To address further the subcellular localization of the endogenous MTM1 protein in muscle fibers and decipher on which membrane subdomain it is localized, we performed immunolocalization and subcellular fractionation experiments. In mouse isolated muscle fibers, co-labeling of endogenous MTM1 and the Z-line marker α-actinin by immunofluorescence showed the presence of MTM1 on both sides of the Z-line (Fig. 1A). The specificity of the MTM1 antibody was confirmed on Mtm1-KO fibers where we did not observe any staining. MTM1 colocalized with RyR1, an SR protein enriched at the triads (Fig. 1A, right column) and both proteins displayed a consistent overlapping profile as indicated by the signal intensity analysis (Fig. 1B). Therefore, MTM1 seems to be enriched at the triads on both sides of the Z-line.

As the distance between the SR and the T-tubule is about 12 nm, immunofluorescence does not allow to differentiate proteins located at the T-tubules or at the SR. To address on which membrane MTM1 is preferentially located, we performed a subcellular fractionation of triad membranes from rabbit skeletal muscles based on discontinuous sucrose gradient (Saito et al., 1984) (supplementary material Fig. S1). Collected fractions were identified with antibodies against T-tubule and sarcolemma (DHPRα, dystrophin, caveolin 3, dysferlin) markers, SR (SERCA1) and triad (RyR1, calsequestrin) markers (Fig. 1C). MTM1 paralleled the localization of DHPRα and RyR1, confirming its enrichment at the triads. Then we proceeded to the treatment of triad-enriched fractions (F3 and F4 mixture) with wheat germ agglutinin (WGA) to separate peripheral membranes, such as sarcolemma, and other associated membranes, such as T-tubules (WGA+ fraction), from SR-enriched fraction (WGA−) containing SR proteins (Fig. 1; supplementary material Fig. S1). MTM1 was co-enriched with RyR1 in the WGA− fraction, while the WGA+ fraction showed a strong enrichment for DHPRα and dystrophin (Fig. 1D). To further confirm our results we assessed the localization of MTM1 using the vesicle immunoisolation technique (Horgan and Kuypers, 1988; Rosemblatt et al., 1981). This protocol is based on the use of DHPRα antibody-coupled IgG–agarose beads to trap T-tubule membranes. We found that MTM1 is poorly enriched in this immunoabsorbed fraction (supplementary material Fig. S1). Altogether, our results showed that MTM1 is mainly associated with SR membranes at the triad in skeletal muscle.

To address how MTM1 deficiency affects the SR/ER network we examined SR/ER organization of skeletal muscle from wild-type and Mtm1-deficient mice using NADH-TR oxidative staining on tibialis anterior sections from wild-type and 6-week old Mtm1-null mice. Mtm1-null muscles exhibited heterogeneous NADH-TR labeling with cytosolic accumulation and dense subsarcolemmal labeling compared to a homogenous network in wild-type muscles (Fig. 2A). As NADH-TR also labels mitochondria, we used immunodetection of RyR1 as a jSR marker and electron microscopy. RyR1 localization revealed by anti-RyR1 antibody was well organized as doublet bands in wild-type muscle fibers, while it was severely affected in Mtm1-null muscle fibers (Fig. 2A). In line with these observations, EM analysis of the muscle ultrastructure showed a clear disorganization of the triads and the presence of dilated jSR in Mtm1-null muscles (Fig. 2B). In addition, the intermyofibrillar space was enriched with membrane structures in the Mtm1-null muscles. To further characterize the organization of the lSR, we analyzed cross-sections of muscles at the A-band level. SR profiles were less prominent in the Mtm1-null (arrows) compared to the wild-type muscles, and SR membranes were found abnormally located within myofibrils (Fig. 2B). These data sustained that lack of MTM1 is correlated to defects of the SR.

Previous studies suggested an alteration of the T-tubule organization in the absence of MTM1 in mouse and in zebrafish, while the SR was not thoroughly considered (Al-Qusairi et al., 2009; Dowling et al., 2009). To address if SR defects are secondary to T-tubules abnormalities observed in Mtm1-null muscle, we investigate the ER in mouse and human muscle cells (myoblast) as myoblast cells lacks T-tubule and triad structures. The ER network, labeled with an antibody recognizing the ER retention KDEL sequence, was uniformly distributed in control myoblasts while it showed a perinuclear collapse in Mtm1 knock-down (KD) cells (Fig. 2C). To confirm this observation, we next performed a study of the distribution of the ER network in cells. Five random plot profile per cell were created to scan the fluorescence using ImageJ software, measured from the nucleus towards the cell periphery, as illustrated in Fig. 2C. Using this method, we compared changes in the subcellular distribution of the ER network between different cell lines. In order to represent the distribution of fluorescence in different cell regions the measurements were normalized to the cell size and to the fluorescence intensity. The ER network collapse in the Mtm1 KD cell line was confirmed by quantifying the fluorescence distribution from the nucleus to the cell periphery (Fig. 2D). To confirm that the ER collapse is a phenotype reproduced in patient cells, we performed similar experiments on myoblasts from XLCNM patients carrying either a MTM1 frameshift mutation at position 238 or a R241C missense mutation and found consistent results (Fig. 2E,F). Thus, in addition to the previously reported T-tubule disorganization, we highlight here a defect of the SR, and experiments in cultured cells suggest that T-tubule disorganization is secondary to jSR defects. These data support an important role of MTM1 on SR/ER organization and shape maintenance.

To gain insight into the potential implication of MTM1 PI 3-phosphatase activity and to mimic and enhance the phosphoinositide-dependent process on SR shape in vivo, we injected tibialis anterior muscles of wild type mice with AAV-Mtm1-WT or AAV-Mtm1-C375S dead-phosphatase mutant. Importantly, the dead-phosphatase mutant MTM1-C375S lacks the catalytic cysteine and was shown to have no residual enzymatic activity (Blondeau et al., 2000; Taylor et al., 2000). Moreover, the corresponding Cys to Ser mutant in the close homolog MTMR2 was crystalized with PdtIns3P and was shown to be folded and to accommodate its lipid substrate in the catalytic pocket (Begley et al., 2006; Taylor et al., 2000; Tronchère et al., 2004). Injection conditions and AAV particles concentration used were set up to reach an overexpression level of the exogenous proteins about 4–5 fold higher than that of endogenous MTM1, to focus our study on the impact of the exogenous protein constructs (supplementary material Fig. S2). The AAV-empty vector was injected as a control in the other leg of the same animal.

To address whether the active-phosphatase MTM1-WT or dead-phosphatase mutant MTM1-C375S ectopic expression affects muscle physiology we performed histological and physiological analyses of tibialis anterior muscles from injected mice 4 weeks after injection (supplementary material Fig. S3). At the same level of expression (supplementary material Fig. S2), both injected constructs (AAV-Mtm1-WT and AAV-Mtm1-C375S) did not promote any detectable signs of muscle pathology according to muscle weight (supplementary material Fig. S3B), specific isometric force (supplementary material Fig. S3C) and fiber size measurements (supplementary material Fig. S3D), suggesting that ectopic expression of dead-phosphatase mutant MTM1-C375S in normal muscle does not have a dominant negative effect and does not create a centronuclear myopathy phenotype.

Isolated fibers from transduced muscles were labeled with MTM1 and RyR1 antibodies. As for the endogenous MTM1 (See Fig. 1), ectopically expressed active-phosphatase MTM1-WT and dead-phosphatase mutant MTM1-C375S co-localized with RyR1 at triads (Fig. 3A). Highly expressing transduced isolated fibers showed that, in addition to the triad localization, exogenous active-phosphatase MTM1-WT and dead-phosphatase mutant MTM1-C375S displayed strong accumulations at the peripheral region or inside the fiber, respectively (Fig. 3B–C).

To address whether the accumulation of MTM1 is affecting the SR organization, we checked the localization of the SR marker RyR1 and SERCA1 in these fibers. RyR1 and SERCA1 were also accumulating at specific locations, similarly as MTM1, in fibers that showed a high expression of MTM1-WT and dead-phosphatase mutant MTM1-C375S (Fig. 3D). Calnexin also colocalized with the active-phosphatase MTM1-WT and dead-phosphatase mutant MTM1-C375S induced structures (supplementary material Fig. S4). These results suggest that ectopic expression of active-phosphatase MTM1-WT and dead-phosphatase mutant MTM1-C375S promotes an abnormal accumulation of SR at specific sites in muscle fibers.

In order to characterize these MTM1-positive SR accumulations in the transduced muscles, we performed ultrastructural analysis using electron microscopy. The wild-type muscles injected with empty vector, AAV-Mtm1-WT and AAV-Mtm1-C375S showed proper organization of the sarcomere arrangement and the triad structure (Fig. 4). In line with our previous immunolabeling experiments on isolated fibers and muscle biospies, we observed that active-phosphatase MTM1-WT and dead-phosphatase mutant MTM1-C375S promoted SR membrane remodeling toward different structures depending on the phosphatase activity, while muscles transduced with the AAV-empty constructs did not exhibit any form of membrane remodeling (Fig. 4A). Active-phosphatase MTM1-WT promoted the formation of SR membrane stacks (Fig. 4A, open arrowhead). However, dead-phosphatase mutant MTM1-C375S induced the formation of highly organized membrane structures within the muscle fibers (Fig. 4A, closed arrowhead). Such membrane structures have been reported previously in the literature in other systems as ‘undulating membranes’, ‘crystalloid membranes’, ‘paracrystalline ER’, ‘tubuloreticular structures’, ‘honeycomb-like structures’ or ‘cubic membranes’ (Almsherqi et al., 2006). We further named the highly organized SR structures formed by the dead-phosphatase mutant MTM1-C375S as ‘cubic membrane’. We then performed immunogold stainings and EM with anti-MTM1 antibodies and found that MTM1 is highly and specifically localized at stacks and cubic membrane structures, respectively (Fig. 4B). These observations suggest that expression of MTM1 directly promotes the formation of these SR structures.

Immunolabeling and ultrastructural analyses strongly supported that cubic membranes have an SR origin. However, as MTM1 defect is accompanied by T-tubule disorganization, we addressed whether they also have a sarcolemma origin or connection. We localized a sarcolemmal marker (dystrophin) in the AAV-Mtm1-WT-GFP- and AAV-Mtm1-C375S-GFP-injected muscles. Dystrophin did not localize to either types of MTM1 accumulations (supplementary material Fig. S5).The sarcolemma and T-tubule marker caveolin-3 (Cav-3) did not label the cubic membranes and showed a normal localization in the muscle fiber (supplementary material Fig. S6). Furthermore, we performed a potassium ferrocyanide staining to reveal sites connected to the extracellular space, like the intraluminal space of T-tubules. T-tubules were labeled with the ferrocyanide precipitate as black dots at the A/I bands transition on longitudinal sections of muscles injected with AAV-empty or both AAV-Mtm1 constructs (Fig. 4C). Cubic membranes in MTM1-C375S expressing muscles were not labeled with potassium ferrocyanide. Altogether, these data support that both the sub-sarcolemma membrane stacks triggered by MTM1-WT and the cubic membranes triggered by the MTM1-C375S dead-phosphatase have a SR origin.

To better understand the complexity of cubic membranes induced in vivo with the dead-phosphatase MTM1 mutant, we characterized them further by electron microscopy. Cubic membranes were composed of highly organized sinusoidal double membranes (Fig. 5A). Moreover, we noted that the cubic membranes exhibited a similar organization on longitudinal and transverse sections (Fig. 5A). Altogether, it suggests that these structures are not membrane stacks nor tubes but follow a tridimensional curvature.

Two main strategies could be applied to determine the spatial characteristics of cubic membranes: tomographical imaging and mathematical modeling. Cubic membranes adopt mathematically well-defined 3D morphologies; they can be modeled as triply period minimal surfaces. In nature, one typically finds the primitive (P), double diamond (D), and gyroid (G) surfaces (Almsherqi et al., 2006). We based our theoretical modeling on an existing cubic membrane simulation software [‘cubic membrane simulation projection software’ (Deng et al., 1999; Landh, 1995)] to create a similar program. The underlying method is based on pattern and symmetry recognition. The surface P, D and G are approximated by their corresponding nodal surfaces from which the program generates two-dimensional (2D) projection catalogs with various surface parameters, projection directions, viewing angles and section thicknesses of an EM specimen (supplementary material Fig. S7). In a second step, we performed EM tomography analysis of 200 nm-thick muscle sections by intermediate voltage transmission-EM at multiple tilted angles, yielding a large number of projections (Fig. 5B). Theses images were matched to theoretical computer-generated projections (Almsherqi et al., 2009; Deng et al., 1999; Landh, 1995) (Fig. 5C). The comparison suggested a gyroid cubic membrane configuration. Indeed, superimposition of computer-generated projections with subdomains of the EM images perfectly matched (Fig. 5B,C). This theoretical modeling strongly supports the notion that these membrane structures adopt a cubic membrane morphology.

Reconstructing the EM tomography projections by computational image analysis into a 3D representation confirmed the G-type configuration and revealed the curved profile of the membrane that was highly ordered and repeated in the Z projections (Fig. 5E,F).

The tendency of overexpressed ER transmembrane proteins to reorganize ER tubules into stacked ER cisternae, also called organized smooth ER (OSER) structures, has been described by several groups (Snapp et al., 2003). Several lines of evidence suggested that oligomerization of the cytoplasmic domains of ER transmembrane proteins on apposed ER membranes was necessary for membrane stacking (Gong et al., 1996; Snapp et al., 2003; Yamamoto et al., 1996). As recombinant MTM1 could assemble into heptamers we hypothesized that MTM1 oligomerization could promote membrane stacking observed both with expression of active-phosphatase MTM1-WT and dead-phosphatase mutant MTM1-C375S (Schaletzky et al., 2003). We therefore assessed the oligomerization potential of active-phosphatase MTM1-WT and dead-phosphatase mutant MTM1-C375S in vitro and in muscle. Equimolar GST-MTM1-WT was incubated with in vitro translated B10-flagged MTM1-WT and MTM1-C375S. Pull-down experiments were revealed with anti-B10 antibody and showed that MTM1 can oligomerize independently of its phosphatase activity (Fig. 6A). Moreover, we performed an immunoprecipitation with anti-GFP antibody using protein extracts from wild-type muscles injected with AAV-Mtm1-WT-GFP, AAV-Mtm1-C375S-GFP and AAV-GFP according to the same procedure as the untagged constructs. Endogenous MTM1 immunoprecipitated with both MTM1-WT-GFP and MTM1-C375S-GFP, confirming that both constructs can oligomerize (Fig. 6B). Protein extracts from muscles injected with AAV-Mtm1-WT-GFP and AAV-Mtm1-C375S-GFP displayed a tendency of higher protein levels for the endogenous MTM1 compared to the muscle expressing GFP, suggesting a stabilization of the endogenous MTM1 protein through the oligomerization with MTM1-WT-GFP and MTM1-C375S-GFP proteins, as suggested for the MTMR8-MTMR9 myotubularins dimer (Zou et al., 2012).

As active-phosphatase MTM1-WT induced stacking of flat membranes while the dead-phosphatase mutant MTM1-C375S generated highly curved cubic membranes, we hypothesized that the MTM1 substrate, PtdIns3P, could be determinant for the cubic symmetry and more generally for SR curvature. As muscle tissues from living mice cannot be metabolically labeled, we used a novel sensitive mass assay (Chicanne et al., 2012) to measure the level of PtdIns3P in WT skeletal muscles injected with AAV-empty, AAV-Mtm1-WT and AAV-Mtm1-C375S and compared with Mtm1-null muscles. As expected, the PtdIns3P level was increased in the Mtm1-null muscle (2.2-fold increase) compared to the wild-type muscle (Fig. 6C). The level of PtdIns3P decreased in AAV-Mtm1-WT-injected muscle, while it was significantly increased in AAV-Mtm1-CS-injected muscle to the level of Mtm1-null muscle (1.65-fold decrease for MTM1-WT and 1.7-fold increase for MTM1-C375S). Comparison of wild-type and Mtm1-null muscles supports that PtdIns3P is a major physiological substrate of MTM1. These PIs measurements strongly suggest that the dead-phosphatase mutant MTM1-C375S protects PtdIns3P from dephosphorylation by endogenous MTM1, and that the formation of cubic membranes is paralleled by an increase in PtdIns3P.

To our knowledge, PtdIns3P localization was never assessed in mammalian skeletal muscle. We performed a direct labeling of PtdIns3P in skeletal muscle injected with AAV-Empty, AAV-Mtm1-WT and AAV- Mtm1-C375S using the PtdIns3P-specific tandem FYVE domain biosensor (2XFYVE) (Gillooly and Stenmark, 2001; Gillooly et al., 2000; Gillooly et al., 2003; Raiborg et al., 2001). Biotinylated 2XFYVE recombinant probe was applied on ultrathin muscle cryosections followed by immunogold detection to localize PtdIns3P at the ultrastructural level. PtdIns3P was detected at the triad in all injected muscles and preferentially located to the SR cisternae rather than to the T-tubule (Fig. 6E). Localization of PtdIns3P to the triad in Mtm1-null muscle could not be assessed due to their strong disorganization. Quantification of immunogold particles per triad showed a significant increase in PtdIns3P upon dead-phosphatase mutant MTM1-C375S expression (1.43±0.38 gold particle/SR cisternae) compared to AAV-empty and AAV-Mtm1-WT transduced muscles (0.73±0.36 and 0.63±0.18 gold particle/SR cisternae, respectively) (Fig. 6D). These data are in agreement with the enrichment of endogenous MTM1 at the jSR (see Fig. 1) and suggest that MTM1 regulates the level of PtdIns3P at the jSR.

We also investigated the presence of PtdIns3P on the remodeled SR structures. A massive accumulation of PtdIns3P was detected on cubic membranes in muscle injected with AAV-Mtm1-C375S, while flat membrane stacks observed in muscle injected with AAV-Mtm1-WT were not labeled with the 2XFYVE probe (Fig. 6F). Thus, flat SR membrane stacks are induced by expression of catalytically active MTM1 and are devoid of PtdIns3P, while highly curved cubic membranes are induced by the dead-phosphatase mutant MTM1-C375S and enriched in PtdIns3P, supporting that PtdIns3P directly contributes to membrane curvature at the SR.

To gain insight into the implication of PtdIns3P on the shape of junctional SR in vivo, we injected tibialis anterior muscles of wild-type mice with AAV-2XFYVE or AAV-2XFYVE-C215S (2XFYVECS) mutant. Importantly, the 2XFYVECS mutant was shown to be unable to bind the PtdIns3P (Gaullier et al., 1998). In order to characterize junctional SR structure in the transduced muscles, we performed ultrastructural analysis using electron microscopy. We observed that 2XFYVE-GFP expression promoted enlargement of junctional SR membrane, while muscles transduced with the AAV-2XFYVECS-GFP constructs did not exhibit enlargements of membrane (Fig. 7A,B). T-tubule shape or the connection between t-tubule and jSR were not altered. These results confirmed the presence of the PtdIns3P at junctional SR and its potential role in the shape modulation of this membrane compartment.

The SR forms a complex membrane network associating a range of different curvatures and is key for intracellular organization and calcium homeostasis in skeletal muscle. However, the regulation of SR curvature and shape is barely understood. In this study we identified the phosphatase myotubularin and its lipid substrate PtdIns3P as key regulator of SR remodeling. We showed that MTM1 is enriched at the jSR in skeletal muscle using immunofluorescence and biochemical approaches. MTM1 deletion, knockdown or mutation in mouse and in patients with centronuclear myopathy lead to disorganization of SR or ER in muscle and myoblast cells, respectively. Moreover, in vivo modulation of MTM1 lipid phosphatase activity alters PtdIns3P levels and promotes SR remodeling. Altogether, these data support a role for MTM1 enzymatic activity in the remodeling and maintenance of jSR membranes in skeletal muscle.

Numerous studies have described the organization of SR in muscles from different organisms and have shown that SR plays an important role not only on calcium homeostasis, as the main calcium store and calcium release site, and muscle contraction but also on the positioning and maintenance of other organelles and muscle compartments (Boncompagni et al., 2009; Franzini-Amstrong, 1991; Franzini-Armstrong, 2007; Luff and Atwood, 1971; Rossi et al., 2008; Takekura et al., 2001; Takekura et al., 1993). Boncompagni et al. suggested that during postnatal development the SR contacts and maintains mitochondria near the triad (Boncompagni et al., 2009). The alteration of SR organization and dynamics in skeletal muscle might thus generate pleiotropic defects. Muscles from patients with centronuclear myopathy and MTM1 mutations are characterized by a broad intracellular disorganization including centralized nuclei surrounded by an accumulation of sarcoreticulum membranes and glycogen granules, mis-localized mitochondria, longitudinal orientation of T-tubules and abnormal triad structure (Hnia et al., 2011; Romero, 2010). As we found an enrichment of MTM1 at the junctional SR and as the absence of MTM1 in Mtm1-deficient mouse and myoblasts from XLCNM patients affects the SR/ER network, we conclude that SR alteration most probably represents a primary cause of most of the organelle positioning defects reported in muscles biopsies from XLCNM patients. This does not exclude that MTM1 regulates organelle positioning through additional pathways like through interaction with desmin as it was recently suggested (Hnia et al., 2011). Moreover, in vivo modulation of MTM1 and comparison with Mtm1-deficient muscles support the involvement of MTM1 and PtdIns3P in the SR membrane shape, that could consequently affect triads organization. Altogether, these SR organization and shape defects could generate abnormal orientation of T-tubules, calcium homeostasis and muscle contraction defects, culminating in muscle weakness and hypotonia reported in patients and animal models (Al-Qusairi et al., 2009; Beggs et al., 2010; Dowling et al., 2009; Toussaint et al. 2011).

To address the role of myotubularin in the maintenance of the SR system in skeletal muscle in vivo, we compared Mtm1-null muscles with muscles overexpressing MTM1 constructs. In particular, we modulated MTM1 PI-phosphatase activity by ectopic expression of MTM1-WT or the dead-phosphatase MTM1-C375S mutant. Both constructs affected SR shape and promoted membrane stacking, albeit MTM1-WT was associated to flat membranes while the MTM1-C375S was associated to highly curved membrane structures called cubic membranes. Membrane stacks formed by muscle transduction with AAV-Mtm1-WT was previously reported in mouse in a study aiming to use this construct for a gene therapy approach (Buj-Bello et al., 2008). However, the origin of these structures and their dependence on the MTM1 enzymatic activity were not assessed. We showed that these membrane structures originated at least in part from the SR, the endogenous localization of MTM1, by electron microscopy and immunolabeling with SR markers.

The tendency of overexpressed ER proteins to reorganize ER tubules into organized smooth ER (OSER) with flat or cubic membrane structures has been described by several groups (Almsherqi et al., 2006; Snapp et al., 2003; Yamamoto et al., 1996). The authors showed that OSER were formed following dimerization of resident ER proteins; addition of GFP but not monomeric GFP to the cytoplasmic domains of ER transmembrane proteins was sufficient to induce OSER structures. In such studies, ER transmembrane proteins were used. Indeed we showed that MTM1-WT and MTM1-C375S could oligomerize in vitro and ex vivo, suggesting that MTM1 oligomerization is sufficient to promote membrane stacking, independently of its enzymatic activity. In the studies dealing with OSER structures, ER transmembrane proteins were used and formation of OSER was dependent on ER anchorage. MTM1 has no transmembrane domain; we hypothesize that its enrichment at jSR and its capacity to induce SR membranes stacking depends on its interaction with an unidentified resident SR protein. Alternatively or concomitantly, MTM1 may bind to phospholipids via its PH-GRAM domain, although the lipid-binding property of the PH-GRAM domain of myotubularins is still controversial (Choudhury et al., 2006; Schaletzky et al., 2003; Tsujita et al., 2004).

We further addressed the role of PtdIns3P, a physiological PI substrate of MTM1, on the regulation of SR curvature in muscle. The localization and function of PtdIns3P in mammalian skeletal muscle had not been characterized. We focused our observations on the triads and jSR where we located MTM1. Immunogold labeling and biochemical analysis showed that PtdIns3P localized at the jSR, as MTM1, and was enriched at this location in muscles injected with the dead-phosphatase MTM1-C375S, supporting that MTM1 regulates PtdIns3P at the jSR. The role of PtdIns3P in ER remodeling had not been thoroughly investigated. Knockout of the endogenous MTM1 and in vivo modulation of the MTM1 phosphatase activity had a strong impact on the observed SR structures: MTM1 deficiency led to altered SR structures and MTM1-WT promoted flat membrane stacks while the dead-phosphatase MTM1-C375S induced highly curved SR cubic membranes. Using similar experimental approaches, we found that PtdIns3P was absent from flat membrane stacks while it was enriched on cubic membranes. Both PtdIns3P and MTM1 dimerization are needed for increased curvature of SR membranes. Moreover, the 2XFYVE-GFP expression in the muscle confirmed the importance of PtdIns3P for the shape of junctional SR. The PtdIns3P level at SR membranes appears directly correlated with the degree of curvature. Further works need to investigate whether membrane curvature results from a direct effect of PtdIns3P accumulation on the membrane or an indirect effect via the recruitment of PtdIns3P binding proteins. Previous studies suggested an association between PtdIns3P and curved membranes in cultured cells (Axe et al., 2008; Hamasaki and Yoshimori, 2010). Axe and colleagues speculated that PtdIns3P accumulation at the ER alters membrane curvature to create the so-called ‘omegasome’, a highly curved sub-domain involved in the first step of autophagosome biogenesis. In addition, Fan et al. suggested a critical role of PI3KC3 (hVPS34) and Barkor/Atg14(L) for the creation and the stabilization of PtdIns3P-enriched omegasomes at ER membranes, respectively (Fan et al., 2011). In this study, we implicate for the first time MTM1 and PtdIns3P in membrane curvature of SR in the muscle tissue. It is thus possible that regulated accumulation of PtdIns3P on ER or SR leads to increased curvature in both cases, and is necessary for omegasome formation at the first step of autophagy (for ER) and for SR shape to build a correct triad in muscle.

We propose the following model for the regulation of SR membrane curvature by myotubularin and PtdIns3P (Fig. 8). Phosphatases from the dual-specificity class, including MTM1, proceed through the following steps: substrate recognition, formation of a phosphoenzyme intermediate where the cysteine forms a covalent thio-ester bound with the phosphorylated substrate, release of the unphosphorylated substrate, and regeneration of the enzyme with a water molecule (Denu et al., 1996). While dephosphorylation of PtdIns3P promotes membrane flattening, we used the MTM1-C375S mutant to mimic and freeze an enzyme-substrate intermediate complex and increase PtdIns3P local concentration. Although the C375S mutant does not have the cysteine residue to form a covalent bound with the substrate, it cannot dephosphorylate PtdIns3P and thus may still bind to it, an hypothesis sustained by the crystallization of PtdIns3P in complex with MTM1-C375S (Begley et al., 2006). Increased PtdIns3P level and increased binding of MTM1 (i.e. decreased enzymatic activity) favor the equilibrium toward a curved membrane. In vivo, it is expected that the fine-tuning of MTM1 enzymatic activity modulates the level of PtdIns3P and subsequently the degree of curvature. These results are consistent with a recent study showing that ectopic expression of active and dead-phosphatase MTM1 at endogenous level in the Mtm1-null mouse lead to a similar improvement of most of the XLCNM-like phenotypes except the triad shape (Amoasii et al., 2012). These results sustained an involvement of the phosphatase activity of MTM1 in the membrane shape of the triad.

In conclusion, our data provide a new perspective on the function of myotubularin in SR organization and its role jointly with PtdIns3P in modulating membrane curvature. The tight regulation of PtdIns3P on specific membrane subdomains by myotubularin-related proteins may be a general mechanism to control membrane curvature in different tissues. We propose alterations of this mechanism is a main primary cause for the development of the X-linked centronuclear myopathy.

Animals were housed in a temperature-controlled room (19–22°C) with a 12 hour:12 hour light/dark cycle. Mice were humanely killed by CO₂ inhalation followed by cervical dislocation, according to national and European legislations CEE86/609 on animal experimentation, and protocols approved by our institutional IGBMC ethical committee.

Full-length mouse isoform MTM1 cDNA was cloned into pENTR1A (Invitrogen, Carlsbad, CA) and then recombined into a pAAV-MCS and pAAV-GFP vector using the Gateway system (Invitrogen). The C375S mutation was introduced by primer-directed PCR mutagenesis. All constructs were verified by sequencing. pXR1 (AAV1) plasmid was a gift from Jude Samulski at the Gene Therapy Center, the University of North Carolina at Chapel Hill. 2XFYVE and 2XFYVE-C215S construct provided by H. Stenmark (Department of Biochemistry, Institute for Cancer Research, Oslo, Norway) was recloned into pENTR1A (Invitrogen, Carlsbad, CA) and then recombined into a pAAV-GFP vector using the Gateway system (Invitrogen).

Primary antibodies used were mouse anti-α-actinin (EA-53; Sigma-Aldrich, St. Louis, MO), DHPRa1 (Cav1.1) subunit (MA3-920; Affinity Bioreagents, Golden, CO), glyceraldehyde-3-phosphate dehydrogenase (GAPDH; MAB374; Chemicon, Temecula, CA), RyR1 (clone 34C; Sigma-Aldrich, St. Louis, MO), SERCA1 ATPase (MA3– 911; ABR), anti-calsequestrin (IgG2b clone VIIID12, Fisher scientific SAS), anti-dysferlin (clone Ham3/17B2, Novocastra, Wetzlar, Germany) anti-KDEL ER marker (10C3, Santa Cruz Biotechnology, Santa Cruz, CA); goat anti-caveolin 3 (N-18, Santa Cruz Biotechnology, Santa Cruz, CA). The rabbit anti-dystrophin was provided by Dr Mornet (INSERM U1046, Montpellier France) and characterized previously (Hnia et al., 2006). Rabbit anti-MTM1 antibodies were made onsite at the polyclonal antibody facility of the Institut de Génétique et Biologie Moléculaire et Cellulaire (IGBMC). Alexa-conjugated secondary antibodies (goat anti-mouse or goat anti-rabbit or donkey anti-mouse or donkey anti-rabbit coupled to Alexa Fluor 488 or Alexa Fluor 595) were purshased from Invitrogen. Secondary antibodies against mouse and rabbit IgG, conjugated with horseradish peroxidase (HRP), were obtained from Jackson ImmunoResearch Laboratories (West Grove, PA). Hoechst was purchased from Sigma-Aldrich (B2883). The AAV Helper-Free system was purchased from Stratagene (La Jolla, CA; catalog no. 240071).

Knockdown (KD) Mtm1 and control C2C12 myoblast cell lines used were generated as described (Hnia et al., 2011). Human and mouse primary myoblasts were generated from human biopsy explants (provided by O.M. Dorchies, University of Geneva, Geneva, Switzerland), as described previously (Dorchies et al., 2001; Hnia et al., 2011).

AAV2/1 vectors were generated by a triple transfection of AAV-293 cell line with pAAV2 insert containing the insert under the control of the CMV promoter and flanked by serotype two inverted terminal repeats, pXR1 containing rep and cap genes of AAV serotype 1, and pHelper encoding the adenovirus helper functions. The AAV2/1 vectors were produced and purified as previously described (Cowling et al., 2011).

Two- to 3-week-old wild-type male 129PAS mice were anesthetized by intraperitoneal injection of 5 ml/g of ketamine (20 mg/mL; Virbac, Carros, France) and xylazine (0.4%, Rompun; Bayer, Wuppertal, Germany). Tibialis anterior (TA) muscles were injected with 20 ml of AAV2/1 preparations or sterile AAV2/1 empty virus. The TA muscles were dissected under anesthesia, when required for transmission electron microscopy (TEM), 4 weeks post injection (PI) and frozen in nitrogen-cooled isopentane and liquid nitrogen for histological and immunoblot assays, respectively.

Total, soluble, and insoluble proteins were extracted from the skeletal muscle of mice. Muscles were homogenized in 50 mM Tris, 10% glycerol, 1 mM EDTA, 50 mM KCl, 10 mM β-glycerophosphate, 10 mM NaF, 1 mM Na3VO4, 0.1% SDS, 2% Triton X-100 and protease inhibitors (Roche Diagnostics) using a Polytron® homogenizer (Kinematica Inc.) then extracted for 30 minutes at 4°C, and used for western blotting. Protein concentration was determined using a DC protein assay kit (Bio-Rad Laboratories, Hercules, CA) and lysates analyzed by SDS–PAGE and western blotting on nitrocellulose membrane. Primary antibodies (see antibodies section) were used at the appropriate dilution followed by secondary antibodies incubation and extensive washing. Membranes were revealed by ECL chemiluminescent reaction kit (Supersignal west pico kit, Thermo scientific, Pierce).

cDNA corresponding to full length wild-type or C375S mutant MTM1 sequences were cloned into pDEST15 (N-ter GST fusion) destination vector. All constructs were verified by Sanger sequencing. GST fusion proteins were expressed in the Escherichia coli strain BL21-Rosetta 2 (Novagen) and extracted from bacterial pellets by adding the extraction buffer (50 mM Tris-HCl pH 8.0, 100 mM NaCl, 5 mM EDTA, 1 mM EDTA) supplemented with 1 mg/ml lysosyme, a cocktail of protease inhibitors (Complete EDTA free, Roche) and 1 mM PMSF. After 30 minutes incubation on ice, detergents were added (0.01% N-laurylsarcosine and 0.5% Triton X-100) and lysates were incubated overnight at 4°C to obtain high solubilization. Then, lysates were centrifuged at 16.000 g for 30 minutes. GST fusion proteins were purified by incubation with glutathione-Sepharose 4B beads (GE Healthcare) overnight followed by extensive washing with extraction buffer plus 0.5% Triton X-1000. The purified GST-fusion proteins coupled to glutathione beads were then incubated overnight with pSG5-MTM1-B10 or pSG5-MTM1C375S-B10 translated in vitro (TNT Coupled Reticulocyte Lysate Systems, Promega) according to the manufacturer’s protocol. Briefly, plasmids were incubated with Methionine and TNT T7 quick master mix for 2 hours at 30°C. Homogenates were diluted with the extraction buffer and incubated with the recombinant proteins on beads. After washing three times with extraction buffer, bound proteins were analyzed by western blot.

This protocol is based on previously reported methods used in triad isolation from rabbit muscle membranes with slight modifications (Saito et al., 1984; Zorzano and Camps, 2006). Twelve grams of rabbit skeletal muscle were minced into 2–4 mm² pieces and fat and tendon were removed. Muscles were then homogenized in 10 ml ice cold homogenization buffer (20 mM Tris-HCl pH 7.4, 250 mM sucrose, 1 mM EDTA, 1 mM pepstatin and 1 mM leupeptin) using a Polyton homogenizer at low speed (5000 rpm). Homogenates were centrifuged for 20 minutes at 12,000 g in fixed angle rotor at 4°C. The supernatant (S1) was collected and pellet resuspended in 5 ml homogenization buffer and submitted to a second centrifugation 20 minutes at 12,000 g. The resulting supernatant (S2) was pooled with S1 and the pellet was solubilized in homogenization buffer with 8 M urea. An aliquot was mixed with 4× Laemmli buffer and frozen for analysis (pellet fraction, P). The S1 and S2 mixture was submitted to myofibril solubilization by adding KCl to a final concentration of 0.6 M and then incubated for 1 hour at 4°C on an orbital shaker. A microsomal fraction (M) from S1+S2 was obtained by centrifugation for 2 hours at 110,000 g at 4°C. Microsomes were resuspended in homogenization buffer using a Dounce homogenizer and layered onto a sucrose gradient. The gradient steps, 8 ml each, consisted of 45% (wt/wt) sucrose (1.6 M), 38% (1.3 M), 32% (1.1 M) and 27% (0.8 M) all buffered with 20 mM Tris-HCl pH 7.4. After a 16 hours centrifugation at 77,000 g at 4°C (in Beckman SW27 rotor), fractions at the interfaces between the gradients were collected (see supplementary material Fig. S1) and diluted twofold with homogenization buffer and further centrifuged for 2 hours at 120,000 g at 4°C. Fraction 1 (top of the 27%) contained mostly longitudinal SR, with some T-tubules and sarcolemma, fraction 2 (27/32% interface), was enriched with longitudinal SR and sarcolemma, fraction 3 (32/38% interface) contained a mixture of longitudinal and junctional SR and some sarcolemma, and fraction 4 (38/45% interface) consisted of highly enriched jSR cisternae. Additional isolation of terminal cisternae was performed on fraction 3 mixed with fraction 4 using wheat germ agglutinin (WGA), which binds with affinity for N-acetyl-D-glucosamine and sialic acid membranes present in the sarcolemma. In our case the WGA+ fractions referred to sarcolemma and T-tubules contaminant, as the DHPRα is mainly present in this fraction (see Fig. 1). The WGA− fractions were enriched with longitudinal and junctional SR containing a high amount of RyR1. Protein concentration was estimated using Bradford assay. Muscle fractions were separated by SDS/PAGE gel, and validated using several protein markers for SR, T-tubule and sarcolemma. The caveolin 1 and actin was used as loading control as they are associated to both fractions (Zorzano and Camps, 2006). Further, we assessed the presence of MTM1 in F3+F4 fractions using the vesicle immunoisolation technique (Horgan and Kuypers, 1988; Rosemblatt et al., 1981). This protocol is based on the used of DHPRα antibody coupled IgG-agarose beads to trap T-tubule membranes. This leads to efficient separation of T-tubule membranes from sarcolemmal membranes and, in addition, permits the separation of T-tubule from intracellular membranes (SR and ER). The IgG antibody was used as negative control.

Co-immunoprecipitation (co-IP) experiments were performed from fresh murine tibialis anterior muscles injected with pAAV-Mtm1-GFP, pAAV-Mtm1-C375S-GFP or pAAV-GFP (empty vector). Muscles were dissected and homogenized with a dounce homogenizer in ice-cold co-IP buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 5 mM EDTA, 5 mM EGTA, 1 mM DTT, 0.5% Triton X-100, 2 mM PMSF) supplemented with 0.05% (w/v) SDS. Lysates were centrifuged at 14.000 g at 4°C and precleared with 50 µl of G-Sepharose beads (GE Healthcare) and subsequently incubated with monoclonal anti-GFP (1∶1000, IGBMC) for 12–24 hours at 4°C. Protein-G–Sepharose beads were then added for 4 hours at 4°C to capture the immune complexes. Beads were washed four times with co-IP buffer and once with high stringency co-IP buffer (with 300 mM NaCl). For all experiments, two negative controls consisted of a sample lacking the primary antibody (Beads) and a sample incubated with IgG (IGBMC, Illkirch, France). Resulting immune-bound complexes were eluted in Laemmli buffer and submitted to SDS-PAGE and western blot analysis.

Transverse cryosections (8 µm) of mouse TA skeletal muscles were stained with Hematoxylin and Eosin (HE), succinate dehydrogenase (SDH) and viewed with a fluorescence microscope (DM4000; Leica Microsystems, Sunnyvale, CA). Cross-sectional area (CSA) was analyzed on HE sections from TA mouse skeletal muscles, using the RoiManager plugin of ImageJ image analysis software (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http://rsb.info.nih.gov/ij/, 1997–2009). The percentage of TA muscle fibers with centralized or internalized nuclei was counted using the cell counter plugin of ImageJ image analysis software. Transverse cryosections (8 µm) sections of mouse TA skeletal muscles were prepared, fixed, and stained with the antibodies. We used monoclonal antibodies directed against RyR (clone 34C; Sigma) and SERCA1 ATPase (MA3– 911; ABR). Nuclei were detected by co-staining with Hoechst (Sigma-Aldrich) for 10 minutes. Sections were viewed using a fluorescence microscope (DM4000; Leica Microsystems, Sunnyvale, CA). For isolated fibers, TA muscle was dissected from the hind limb and fixed, permeabilized and stained as described previously (Cowling et al., 2011).

Muscle force measurements were evaluated by measuring in situ muscle contraction in response to nerve and muscle stimulations, as described previously (Cowling et al., 2011).

Muscle biopsies from TA muscles of anesthetized mice were fixed with 4% PFA and 2.5% glutaraldehyde in 0.1 M phosphate buffer (pH 7.2) and processed as described (Cowling et al., 2011). C2C12 Mtm1-KD and control cells were fixed with 2.5% (v/v) glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.2) for 24 hours at 4°C and processed as described (Hnia et al., 2011).

Ultrathin cryosections of 70 nm were immunogold labeled using an automat (EM IGL, Leica Microsystems), which consisted of 15 minutes fixative quenching in 150 mM glycine, rinses in PBS, 30 minutes blocking in 0.1% BSA and 0.1% cold water fish skin gelatine (FSG, Aurion), 1 hour incubation in primary antibody 1/100 in PBS containing 0.1% FSG, 1 hour incubation with protein-A coupled with 10 nm colloidal gold particles (University Medical Center, Utrecht, NL) and postfixation in 2.5% glutaraldehyde.

Muscle biopsy specimens from hind limbs were fixed with 2.5% glutaraldehyde in 0.1 mol/l cacodylate buffer (pH 7.2) and processed for T-tubule staining as described (Al-Qusairi et al., 2009).

Muscle biopsies from TA muscles of anesthetized mice were fixed with 4% PFA and 0.2% glutaraldehyde in 0.2 M PIPES buffer (pH 7.2). After two washes using PHEM buffer (PIPES, HEPES, EGTA, MgSO₄) the muscle fragments were infiltrated in sucrose 2.3 M overnight at 4°C. Small pieces of muscle were mounted on silver pins and frozen in liquid nitrogen before cutting ultrathin cryosections on a Leica UC ultramicrotome with an EM FC7 cryochamber. Sections were cut at −110°C and retrieved with a 1∶1 mixture of 2.3 M sucrose/2% methylcellulose. The sections were allowed to thaw on a glace dish on ice. The following immunolabeling procedure was performed entirely on ice with pre-cooled solutions. The grids were first incubated on PBS for 10 minutes, on PBS containing 0.5% BSA for 10 minutes and then on 7 µl droplets of 2XFYVE biotinylated domain/PBS for 20 minutes. Following four washes over 15 minutes in PBS, the probe was localized using anti-biotin antibodies, rabbit-anti-mouse secondary antibody (DAKO, Denmark) and Protein A–gold (UCM, Utrecht, Netherlands). Sections were then embedded/contrasted using methylcellulose/uranyl acetate (2.0% and 0.4% respectively) and viewed on a JEOL-JEM 1230 electron microscope at 60–80 kV. Determination of the densities of gold labeling over cell membranes was accomplished on images at 30,000× magnification. Gold particles that were assigned to SR compartments of the triads were counted on 25 random images of two different muscles groups from independent experiments. The structure was identified using morphological criteria on the longitudinal sections of the muscle. Gold density on SR/triads structures on the cryosections was calculated by dividing gold beads on SR/triads structure by total triads present on the section. On each section the total triads present were a minimum of 20.

For electron tomography, semithin sections (200–250 nm) were cut using an ultracut UC6 (Leica Microsystem), collected on formvarcarbon-coated copper hexagonal 50 mesh grids and post-stained with 2% uranyl acetate (W/V) and lead citrate. Ten nanometer colloidal gold particles were applied on one side of the grid to be used as fiducial markers. Automated data acquisition of the tilt series through an angular range of −65° to +65° with 18 increments was performed using an electron microscope equipped with a field emission gun and operating at 200 kV (Tecnai F20, FEI Company, Eindhoven, The Netherlands) and Xplore 3D. Tomograms were computed using the IMOD software.

TA muscles were homogenized with the Dounce homogenizer and total lipids were extracted using the modified method of Bligh and Dyer (Bligh and Dyer, 1959) and prepared for the PtdIns3P mass assay (Chicanne et al., 2012). Total phospholipids were quantified by phosphorus measurement according to Fiske and Subbarow method (Fiske and Subbarow, 1925).

All values are given as mean standard error. Differences between two groups were assessed using unpaired two-tailed Student’s t-tests. P<0.05 was regarded as significant. Statistical analysis was performed in Excel (Microsoft).

We acknowledge Olivier Dorchies for patient cells, Kinga Tomczak and Alan Beggs for Mtm1 knockdown myoblasts, Jude Samulski for vectors, Christine Kretz for mouse genotyping, Anthony Bonfrate for help with the theoretical modeling, and Coralie Spiegelhalter, Nadia Messaddeq and Marc Koch for excellent technical assistance.