Structure–function insights into direct lipid transfer between membranes by Mmm1–Mdm12 of ERMES

Eukaryotic cells are compartmentalized into membrane-bound organelles, each of which performs distinct biochemical tasks. However, the recent identification of direct physical contacts between the membranes of two organelles (Kornmann et al., 2009; Elbaz-Alon et al., 2014; Hönscher et al., 2014; Lahiri et al., 2014; Murley et al., 2015) highlights the need for coordinated functions of distinct organelles through direct exchange of metabolites and information. The ER-mitochondrial encounter structure (ERMES) is one such interorganelle tethering entity, and it physically links the membranes of the ER and mitochondria in yeast cells (Kornmann et al., 2009).

ERMES consists of four core subunits: Mmm1, an N-anchor ER membrane protein (Burgess et al., 1994); Mdm10, a β-barrel mitochondrial outer membrane (OM) protein (Sogo and Yaffe, 1994); Mdm34/Mmm2, an OM protein without an obvious transmembrane segment (Youngman et al., 2004), and Mdm12, a peripheral OM protein (Berger et al., 1997). ERMES also consists of the peripheral components Gem1 (Frederick et al., 2004; Kornmann et al., 2011; Stroud et al., 2011) and Tom7 (Yamano et al., 2010). ERMES was proposed to facilitate lipid exchange between the ER and mitochondria (Kornmann et al., 2009). Because most phospholipid biosynthetic pathways are present in the ER and mitochondria, production of cellular phospholipids relies on efficient transport of their precursor lipids between these organelles (Scharwey et al., 2013; Tamura et al., 2014; Tatsuta and Langer, 2017). For instance, cardiolipin, a mitochondrial signature phospholipid, is generated in the mitochondrial inner membrane from phosphatidic acid (PA), a precursor phospholipid that is supplied by the ER. Similarly, phosphatidylserine (PS) generated in the ER is transported to mitochondria, where it is converted to phosphatidylethanolamine (PE). PE is then transported back from mitochondria to the ER and methylated to form phosphatidylcholine (PC). Three ERMES subunits (Mmm1, Mdm12, and Mdm34) contain a synaptotagmin-like mitochondrial lipid-binding protein (SMP) domain, which may belong to the TULIP superfamily (Kopec et al., 2010, 2011), suggesting that these proteins can bind to phospholipids. Indeed, the purified SMP domain of yeast Mdm12 and that of Mmm1 fused to maltose-binding protein (MBP) were shown to bind to phospholipids (AhYoung et al., 2015; Jeong et al., 2016). However, there is no compelling evidence for ERMES mediating direct lipid transfer between membranes.

Recently developed in vitro phospholipid transport assays using isolated membranes mainly derived from mitochondria and the ER revealed that lack of an ERMES component (e.g., Mmm1, Mdm12, or Mdm34) impairs PS transport from the ER to mitochondria, but not PE transport from mitochondria to the ER (Kojima et al., 2016). Mutations or deletions of ERMES components lead to altered morphology and/or cellular lipid composition (Dimmer et al., 2002; Kornmann et al., 2009; Osman et al., 2009; Nguyen et al., 2012; Tamura et al., 2012; Voss et al., 2012; Tan et al., 2013). Nevertheless, it is still difficult to judge whether ERMES directly mediates phospholipid transport between organelles or defects in other ERMES functions, like the shape of the ER and/or mitochondria, indirectly affect lipid exchange. Further complicating the interpretation is that an increase in other ER–mitochondria contacts and/or expansion of the area of the ER–vacuole contact region may compensate for a loss of ERMES (Elbaz-Alon et al., 2014, 2015; Lahiri et al., 2014; Lang et al., 2015; Murley et al., 2015). Therefore, a possible lipid transfer function of ERMES at ER–mitochondrial contact sites has not been clearly established yet.

Structural information on ERMES subunits became available only recently. On the basis of a 17-Å-resolution electron microscopy image, AhYoung et al. (2015) constructed a model of the complex between Mmm1 and Mdm12 as a central (likely head [Mmm1]-to-head [Mmm1]) dimer flanked by two molecules of Mdm12 in a tail (Mmm1)-to-head (Mdm12) arrangement. Jeong et al. (2016) determined a 3.1-Å-resolution x-ray structure of Saccharomyces cerevisiae Mdm12, which formed a head-to-head interlocking dimer through domain swapping of the N-terminal β-strand, yet they proposed a provocative model in which the Mdm12 dimer dissociates into a monomer to associate with a head (Mmm1)-to-head (Mmm1) dimer of Mmm1 (likely in a tail [Mmm1]-to-tail [Mdm12] arrangement) and with Mdm34 in a head (Mdm12)-to-head (Mdm34) interlocking manner. In this study, we determined the x-ray structures of Mdm12 from Kluyveromyces lactis and found that interlocking homodimer formation does not play an important role in Mdm12 function. We further demonstrated that Mdm12 or Mmm1 alone is only marginally active in lipid transfer, but the Mmm1–Mdm12 complex exerts efficient phospholipid transfer between membranes in vitro. These results support a model in which ERMES is responsible for not only membrane tethering but also facilitating efficient lipid exchange at ER–mitochondria contact sites, and the Mmm1–Mdm12 pair is central in the lipid transfer function of ERMES.

Mdm12 from yeast K. lactis is homologous in its amino acid sequence to Mdm12 from S. cerevisiae (Fig. S1) and can functionally replace S. cerevisiae Mdm12 in S. cerevisiae cells (Fig. 1 A). To gain insight into the functions of Mdm12, we crystallized and determined the structures of the K. lactis Mdm12 core protein lacking nonconserved C-terminal residues (residues 1–239, KlMdm12) at 3.1 Å resolution, its dimethyl-lysine modified form (KlMdm12dmLys) at 2.25 Å resolution, and K. lactis full-length Mdm12 (residues 1–305, KlMdm12FL) at 3.5 Å resolution (Figs. S2 and S3 and Table 1), and we compared them with that of Mdm12 from S. cerevisiae at 3.1 Å resolution (ScMdm12; Fig. S3; Jeong et al., 2016). The crystal structures of the K. lactis Mdm12 derivatives are essentially identical, except for the N-terminal and/or C-terminal segment (Fig. S3, A and B), and we hereafter refer to the Mdm12dmLys structure with the highest resolution as the KlMdm12 structure. The overall folding of KlMdm12 exhibits an inverted cone-like structure, which is similar to that of ScMdm12 (Fig. S3 C), reflecting their amino acid sequence similarities (Fig. S1). The nonconserved regions (residues 67–85 of KlMdm12 and residues 70–114 of ScMdm12; Fig. S3 A) are disordered in the determined structures.

The ScMdm12 structure has a hydrophobic cavity with a crevice-like opening that spans the “base” and “edge” of the cone-shaped structure, in which bound diacyl glycerophospholipid was identified (Fig. S3 D; Jeong et al., 2016). The determined KlMdm12 structure also revealed the presence of a hydrophobic pocket mainly consisting of conserved hydrophobic residues in which bound phospholipid was unambiguously located (Figs. S1 and S4 A). Although the presence of a deep hydrophobic cavity for phospholipid binding is common to ScMdm12 and KlMdm12, there are significant differences in the positions of the openings of the hydrophobic cavities and bound lipid conformations between ScMdm12 and KlMdm12, which are shown in Fig. S3 (D–G). In particular, KlMdm12 has an opening at the base, but not the edge, of the cone-shaped structure (Fig. S3 E), so that the head group of the bound lipid is located at the pocket opening at the base.

In vitro lipid transfer assays using mitochondria and ER membranes showed that ERMES primarily transports PS, not PE, between the ER and mitochondria (Kojima et al., 2016). TLC analyses revealed that the lipids extracted from the recombinant KlMdm12 purified from E. coli cells were PE and PG, two of the major E. coli phospholipids (Fig. S4 B), suggesting that KlMdm12 can bind to phospholipids with no significant preference between PE and PG, but not to cardiolipin with four acyl chains, in E. coli cells (E. coli cells lack PC). Mass spectrometry analysis of lipids bound to recombinant ScMdm12 expressed in E. coli cells also showed that ScMdm12 does not have strong preference for acyl-chain length for binding (Fig. S4 C). Studies using in vitro lipid displacement assays and/or mass spectrometry analyses of bound lipids in S. cerevisiae cells have reported a moderate lipid binding preference of PC (and PG) to PA and to PS for ScMdm12 (AhYoung et al., 2015; Jeong et al., 2016). This weak substrate specificity for neutral phospholipid PC as compared with acidic phospholipids was ascribed to the acidic residues near the opening of the hydrophobic cavity at the edge of the corn, where the head group of the bound lipid is located (Fig. S3 E; Jeong et al., 2016). However, in contradiction to this proposal, the opening of the hydrophobic pocket in KlMdm12 at the base of the corn exhibits only limited presence of net negative charges (Fig. S3 H).

To test whether the structural properties of the lipid-binding pocket, including its opening, change upon removal of the lipid in KlMdm12, we produced dimethyl-lysine–modified KlMdm12 crystals in the presence of a detergent (FOS-MEA-10), which removed most bound lipid molecules from KlMdm12, and determined the structure of lipid-free Mdm12 at 3.3 Å resolution (Table 1 and Fig. S4 D). Although structural flexibility reflected in the crystallographic temperature factors and disordered residues (Fig. S4, A and D) was larger for lipid-free KlMdm12 than lipid-bound KlMdm12, the overall folding was not affected by removal of the bound lipid (Fig. S4 E). The size and opening of the lipid-binding pocket in lipid-free KlMdm12 did not differ significantly from those of the pocket in lipid-bound KlMdm12.

The N-terminal 10 residues of Mdm12 are well conserved among different organisms (Fig. S1 A), and the ScMdm12 mutant lacking the N-terminal 11 residues caused growth defects of S. cerevisiae cells (Fig. 1 A), suggesting that the N-terminal segment has an important role in Mdm12 function. Jeong et al. (2016) found that purified ScMdm12 forms a homodimer, yet ScMdm12 with the N-terminally attached His₆ tag or ScMdm12 lacking the N-terminal seven-residue segment dissociates into a monomer. Furthermore, the x-ray structure of ScMdm12 showed a homodimer with domain swapping of the N-terminal seven-residue β-strand (Figs. 1 C and S1 A).

We analyzed the oligomeric states of ScMdm12 and KlMdm12 derivatives by gel-filtration chromatography (Fig. 1 D). Unexpectedly, not only the presence of the His₆ or His₁₀ tag at the N or C terminus of ScMdm12 but also the following protease cleavage site sequences affected the monomer–dimer equilibrium, which is not consistent with the observations of Jeong et al. (2016). On the other hand, KlMdm12 behaved as a monomer irrespective of the position of the attached His₆ tag. Lipid-bound KlMdm12 derivatives did not form a domain-swapping dimer like lipid-bound ScMdm12, although two Mdm12 molecules are packed in a head-to-head arrangement in the crystal asymmetrical units (Fig. 1 C). FOS-MEA-10–treated lipid-free KlMdm12 formed an interlocking dimer with domain swapping of the N-terminal β-strand, which is similar to the case of lipid-bound ScMdm12 (Fig. 1 C), but this could be caused by crystal artifact. To assess the contribution of the Mdm12 interlocking dimer to its functions in vivo, we analyzed the effects of the I5P mutation in ScMdm12 on yeast cell growth (Fig. 1 A); the I5P mutation was previously shown to disrupt the N-terminal β-strand and convert the Mdm12 dimer to a monomer (Jeong et al., 2016). In contrast to the Mdm12 mutant lacking the conserved N-terminal 11 residues, Mdm12 with the I5P mutation allowed cells to grow as fast as wild-type Mdm12. Replacement of Pro5 with Ala, Asn, or Trp did not cause defects in yeast cell growth (Fig. 1 A). These results suggest that interlocking dimer formation is not essential for the functions of ScMdm12 in vivo and is therefore not an essential structural requirement for both ScMdm12 and KlMdm12.

Although ScMdm12 alone could be expressed efficiently in E. coli cells, KlMdm12 could be expressed to a sufficient protein level in soluble forms in E. coli cells only upon coexpression of residues 179–434 of K. lactis Mmm1 (simply referred to Mmm1s), which lacks the N-terminal residues constituting the ER luminal domain followed by the transmembrane (TM) segment. Upon coexpression with Mmm1s, the N-terminally His₆ tag–attached Mdm12 (HMdm12) was distributed as both a large hetero-oligomeric complex with Mmm1s (Mmm1s–HMdm12) and a monomeric (free) form without Mmm1s (Fig. 2 A, left). When the His₆ tag was N-terminally attached to Mmm1s (HMmm1s), Mdm12 was not dissociated from the smaller HMmm1s–Mdm12 complex (HMmm1s–Mdm12, likely [KlMdm12]₂–[Mmm1s]₂; Fig. 2 A, right) as compared with the Mmm1s–HMdm12 complex. These results suggest that KlMdm12 tends to form a hetero-oligomer with Mmm1s rather than a homodimer. We used a smaller HMmm1s–Mdm12 complex for subsequent lipid transfer assays.

To gain more insight into the oligomer formation of KlMdm12, we probed the topological arrangement of Mdm12 in the Mmm1s–Mdm12 complex by site-specific cross-linking. KlMdm12 contains three Cys (C133, C135, and C150), which are all buried or only slightly exposed to the solvent in the x-ray structure. We thus introduced additional Cys at position 5 near the entrance of the hydrophobic pocket of KlMdm12 (Fig. 2 B). We isolated both a free and Mmm1s-complexed forms of KlMdm12-I5C and labeled the Cys sulfhydryl group with a photoreactive benzophenone group. Subsequent UV irradiation led to generation of a cross-linked product, which was detected by both antibodies against Mdm12 and Mmm1, whereas KlMdm12-I5C alone or wild-type KlMdm12 in the Mmm1s–KlMdm12 complex did not generate a cross-linked product (Fig. 2 B). Therefore, the entrance of the hydrophobic pocket of KlMdm12 faces Mmm1s in the Mmm1s–Mdm12 complex, which is consistent with the low-resolution EM structure of the heterotetramer of Mmm1 and Mdm12 (AhYoung et al., 2015).

Three ERMES core subunits (Mdm12, Mmm1, and Mdm34) have an SMP domain, which is likely capable of lipid binding (Kopec et al., 2010, 2011). Indeed, the overall backbone structures of KlMdm12 and ScMdm12 resemble that of the SMP domain of a human extended synaptotagmin 2 (E-SYT2), which functions at the ER and plasma membrane appositions, including the presence of hydrophobic cavities and bound PE (Schauder et al., 2014). An interesting structural feature of the E-SYT2 dimer is that it has a hydrophobic continuous tunnel running through the cylinder-like dimer and accommodates two lipid-like molecules per monomer, although the continuous tunnel is still too short to bridge two membranes for lipid transport (Schauder et al., 2014; Fig. 3, A and B). In contrast, ScMdm12 and KlMdm12 have a noncontinuous hydrophobic pocket instead of a continuous tunnel, which accepts only acyl chains of one phospholipid molecule (Fig. 3, A and B). To test the possibility that the lipid-binding pocket of KlMdm12 could transiently form a tunnel or conduit running through the molecule for possible lipid transfer, we performed molecular dynamics (MD) simulation of the KlMdm12 molecule. The simulation results show that the distances between the terminal methyl atoms of the bound PE acyl chains and the hydrophobic side-chain atoms at the bottom of the lipid-binding pocket do not significantly change during 120 ns (Fig. 3, C and D). This indicates that the bottom of the lipid-binding pocket of KlMdm12 is tightly closed and is unlikely to form a continuous conduit like E-SYT2.

The present study and other biochemical as well as structural analyses (AhYoung et al., 2015; Jeong et al., 2016) showed that Mdm12 is a phospholipid-binding protein with little or moderate substrate specificities. However, it has not been demonstrated that Mdm12 has the ability to transfer lipids between membranes efficiently. To address this essential point, we dissected the process of lipid transfer between membranes and tested the ability of KlMdm12 to mediate each step of lipid transfer in vitro. First, we analyzed the step of lipid extraction from membranes. We thus incubated a free (monomeric) form of KlMdm12 with donor liposomes containing fluorescent nitrobenzoxadiazole (NBD)-labeled PE and then separated proteins from liposomes by flotation with step-gradient centrifugation (Fig. 4 A). Free KlMdm12 caused only a minute shift of NBD fluorescence from floated liposome fractions to protein fractions (Fig. 4 B). Because free Mmm1s is not stable in solution, we used the fusion protein between MBP followed by HMmm1s (MBP-HMmm1s). MBP-HMmm1s caused only a marginal shift of NBD fluorescence from liposome fractions to protein fractions (Fig. 4 B). Because KlMdm12 formed a hetero-oligomeric complex with Mmm1s, we incubated the purified HMmm1s–Mdm12 complex with NBD-PE–containing liposomes as well. Surprisingly, the HMmm1s–Mdm12 complex caused a significantly enhanced shift of NBD fluorescence (Fig. 4 B). To gain information on the substrate lipid specificity of lipid extraction by HMmm1s–Mdm12, we analyzed competition by the liposomes with different phospholipid compositions in these NBD-PE extraction assays. Extraction of NBD-PE from liposomes was competed by addition of liposomes containing PC alone, PC + PE (PC/PE = 50/50), PC + PA (PC/PA = 50/50), or PC+PS (PC/PS = 50/50; Fig. 4 C). Competition was the least effective for PC-only liposomes.

As a second step of lipid transfer between membranes, we analyzed the lipid insertion from the proteins bearing a substrate lipid into membranes. We thus preloaded either free KlMdm12 or the HMmm1s–Mdm12 complex with NBD-PE and subsequently mixed them with liposomes without NBD-PE (Fig. 4 D). NBD-PE prebound to free KlMdm12 was transferred to liposomes, yet this transfer of NBD-PE was again significantly enhanced when the HMmm1s–Mdm12 complex was used instead of free Mdm12 (Fig. 4 E).

Because the lipid extraction and lipid insertion assays described in Fig. 4 do not allow evaluation of the kinetics of lipid transfer between membranes, we performed a fluorescent-based lipid transfer kinetics assay between membranes, which is similar to the PA or PS transport assay reported previously (Connerth et al., 2012; Watanabe et al., 2015; Miyata et al., 2016). Donor liposomes containing both NBD-PE and fluorescent rhodamine (Rhod)–labeled PE were incubated with free KlMdm12 or the HMmm1s–Mdm12 complex and acceptor liposomes without fluorescent lipids (Fig. 5 A). When NBD-PE is on the same liposome as Rhod-PE, NBD fluorescence is quenched by rhodamine. If free KlMdm12 or HMmm1s–Mdm12 transports NBD-PE from donor liposomes to acceptor liposomes, the fluorescence of NBD-PE will be dequenched and increase in intensity. As shown in Fig. 5 B, HMmm1s–Mdm12 increased the NBD-fluorescence of NBD-PE in a time-dependent manner, indicating that HMmm1s–Mdm12 has a phospholipid transport activity. Lipid-free, FOS-MEA-10–treated HMmm1s–Mdm12 showed a comparable lipid transfer activity. On the other hand, the free form of Mdm12 or the MBP-HMmm1s fusion protein showed only marginal lipid transfer activities. Next, we compared the transfer of NBD-PE with that of PS labeled with NBD (NBD-PS) between liposomes (Fig. 5, C and D). Efficiency of NBD-PS transfer between liposomes by HMmm1s–Mdm12 was threefold higher than that of NBD-PE transfer, although the structural basis of this difference in lipid transfer efficiency is not clear. Therefore, although free Mdm12 and Mmm1s like the MBP-HMmm1s fusion protein are capable of transferring phospholipid between liposomes inefficiently, the lipid-transfer activity of Mmm1s and/or Mdm12 becomes significantly higher when they form a complex.

We then tested the effects of mutations in Mmm1 on the lipid transfer activities of HMmm1s–Mdm12 and MBP-HMmm1s (Fig. 5 E). On the basis of the predicted homology model of Mmm1s, we introduced mutations (V190S, L274S, or I369S) in the putative hydrophobic pocket in Mmm1s (Fig. 5 G). I369S and L274S, but not V190S, mutations partially suppressed the lipid transfer activity of free MBP-HMmm1s and the HMmm1s–Mdm12 complex. We also tested the effects of similar mutations in Mdm12 on the lipid transfer activities of the Mmm1–Mdm12 complex (Fig. 5 F). However, because KlMdm12 mutants tend to destabilize the oligomeric structures of HMmm1s–Mdm12, we analyzed lipid transfer by the complex of ScHMmm1s and ScMdm12 instead of the K. lactis HMmm1s–Mdm12 complex. We prepared ScMdm12, L10S, V125S, V214S, and E255R in the hydrophobic pocket (Fig. 5 G); L10, V125, and V214 in Mdm12 correspond to the positions of V190, L274, and I369, respectively, in Mmm1, and E255R in ScMdm12 was previously shown to exhibit weaker PC binding activity than wild-type ScMdm12 (Jeong et al., 2016). Although ScHMmm1s–Mdm12 variants showed much lower lipid transfer activity than KlHMmm1s–Mdm12 (Fig. 5, E and F), all the mutations in ScMdm12, including L10S corresponding to V190S in Mmm1, lowered the lipid transfer activity of the Mmm1–Mdm12 complex. It should be noted that the N-terminal 11 residues, including L10 in Mdm12, are essential for Mdm12 functions in vivo (Fig. 1 A). These results indicate that both Mmm1s and Mdm12 are responsible for the high lipid transfer activity of the Mmm1–Mdm12 complex.

Finally, we asked whether the Mmm1 mutations (V190S, L274S, or I369S in K. lactis Mmm1s) and ScMdm12 mutations (L10S, V125S, V214S, or E255R) that mostly affect lipid transfer between liposomes to a different extent in vitro could also affect lipid transport between the ER and mitochondria via ERMES. For this purpose, we used the recently developed in vitro assay system with the isolated yeast heavy-membrane fraction (HMF) to monitor the phospholipid exchange between the ER and mitochondria directly (Kojima et al., 2016). In brief, incubation of radioactive [¹⁴C]-serine with the HMF containing mitochondria and the ER membrane allows formation of radiolabeled PS in the ER. Radiolabeled PS is transported from the ER to mitochondria for conversion to PE, which is then transported from mitochondria to the ER for conversion to PC via the intermediate phosphatidyldimethylethanolamine (Fig. 6 A).

We isolated HMF from mmm1Δ cells expressing Vps13-D716H and wild-type or mutant S. cerevisiae Mmm1 (V190S, L274S, L274S/L369S, or L369S) and from mdm12Δ cells expressing wild-type or mutant S. cerevisiae Mdm12 (L10S, V125S, V214S, or E255R). Expression of Vps13-D716H was reported to suppress the growth defects in ERMES-lacking cells (Lang et al., 2015). The protein levels of Mmm1 and Mdm12 derivatives, phospholipid synthetic enzymes (e.g., Cho1, Psd1, Cho2, and Opi3), and a control protein (Tim23) did not differ significantly among HMFs isolated from those strains (Fig. S5 A). We then incubated the HMFs with [¹⁴C]-serine for different periods of time and analyzed synthesized phospholipids by TLC followed by radioimaging. Radioactive phospholipids were quantified, and the amounts of each phospholipid to total radioactive phospholipids or PE were plotted against the incubation time (Fig. S5, B and C). The relative amounts of PS to total phospholipids (PS/total) efficiently decreased in a time-dependent manner in wild-type membrane fractions, whereas PS synthesis, which was reflected in the total phospholipids (total), lasted for 30 min of incubation (Fig. S5, B and C, WT, PS/total, and total). This decrease in PS/total reflects the PS transport from the ER to mitochondria. Notably, the loss or mutations (V190S, L274S, L369S, and L274S/L369S) of Mmm1 decreased PS transport rates compared with that of the wild-type membrane fractions (Fig. 6 B). The loss of mutations (L10S, V125S, and V214S), but not mutation E255R, of Mdm12 also decreased the PS transport rates compared with that of the wild-type membrane fractions (Fig. 6 C). These results indicate that phospholipid transfer between liposomes by the isolated Mmm1s–Mdm12 complex observed in vitro (Fig. 4) may reflect the lipid transport by ERMES between mitochondria and the ER.

A fundamental question concerning ERMES in yeast is whether it is directly involved in lipid trafficking between the ER and mitochondria, and if so, how hydrophobic lipid molecules can be transported between the organelles by crossing the aqueous cytosol. Here, we focused on K. lactis Mdm12 and Mmm1, the peripheral membrane protein subunit and the ER-anchored subunit of ERMES, respectively, and analyzed their roles in lipid transfer between membranes. We first determined the x-ray structures of KlMdm12 with and without bound phospholipid and compared them with that of ScMdm12 complexed with phospholipid. KlMdm12 and ScMdm12 have a similar hydrophobic pocket in the cone-shaped molecules into which phospholipid can bind. Mutational analyses, including that of the Mdm12 I5P mutant, showed that the head-to-head homodimer formation of Mdm12 observed for the ScMdm12 x-ray structure (Jeong et al., 2016) is not necessary for the function of Mdm12 in vivo.

The structures of ScMdm12 and KlMdm12 and identification of phospholipids in their hydrophobic cavities clearly established that Mdm12 is a lipid-binding protein. However, when tested in vitro, purified free KlMdm12 alone or Mmm1s as a purified fusion protein (MBP-HMmm1s) is capable of extracting phospholipid from donor liposomes and transferring it to acceptor liposomes, but this lipid transfer activity is very low. On the other hand, the ability to transfer lipids is significantly enhanced when KlMdm12 and Mmm1s form a hetero-oligomeric complex with each other. The lipid transfer activity of the Mmm1s–Mdm12 complex is impaired by mutations in both Mmm1s and Mdm12. Furthermore, these Mmm1 and Mdm12 mutations also impaired ERMES-dependent PS transport from the ER to mitochondria in the in vitro lipid transport assays with isolated mitochondria and the ER membranes. This indicates that the lipid transfer activity of the Mmm1s–Mdm12 complex between membranes arises from the cooperation of Mmm1s and Mdm12.

How does the Mmm1–Mdm12 complex mediate phospholipid transfer between membranes? An attractive model is that Mmm1 has the ability to extract lipids from and insert lipids into the ER membrane and thereby functions as an extractor or inserter of lipids in the ER membrane. For lipid transport from the ER to the mitochondrial OMs, Mmm1 should pass the extracted lipid over to Mdm34 via Mdm12. The revealed geometry of the entrance of the lipid-binding pocket of Mdm12 in contact with Mmm1 supports this idea. For the transfer of the lipid from Mmm1 to Mdm12, two possible models can be considered (Fig. 7). In the lipid carrier model (Fig. 7 A), Mmm1 may change the relative geometry to Mdm12 from the tail (Mmm1)-to-head (Mdm12) contact to the head (Mmm1)-to-head (Mdm12) contact, so that the phospholipid molecule bound to the hydrophobic pocket of Mmm1 can be transferred to the one of Mdm12 through the outlets of the both pockets at the head-to-head interface between Mmm1 and Mdm12. Upon receiving the phospholipid molecule from Mmm1, Mdm12 may switch the partner from Mmm1 to Mdm34, with changing the relative geometry to Mdm34 from the tail (Mdm12)-to-head (Mdm34) to the head (Mdm12)-to-head (Mdm34) contact for further lipid transfer. Mdm34 also contains an SMP domain and may thus be another possible lipid-binding component on the mitochondrial surface like Mmm1 in the ER membrane. Therefore, a switch between the heterodimeric structural units consisting of the SMP domains of Mdm12 and Mmm1 and of Mdm12 and Mdm34, with changing relative geometry to each other, may facilitate lipid transfer between the ER and mitochondrial OMs. Another possible model for lipid transport from ER to mitochondria, the continuous conduit model (Fig. 7 B), assumes the presence of a continuous hydrophobic tunnel running all the way from Mmm1 to Mdm34 via Mdm12. The phospholipid molecule extracted by Mmm1 from the ER membrane may diffuse through the hydrophobic conduit via Mdm12 to reach Mdm34 and is then inserted into the mitochondrial OM by Mdm34. Our observation that the bottom of the hydrophobic pocket of Mdm12 is tightly closed favors the lipid carrier model, yet stable complex formation between Mmm1 and Mdm12, without any dynamic exchange of Mmm1 or Mdm12 with their free forms, favors the continuous conduit model. The question of which of these speculative models is correct should be addressed experimentally in future studies.

The plasmids used here were constructed by standard recombinant DNA techniques. For expression of Mdm12(1–239) from K. lactis (KlMdm12) in E. coli cells, the DNA fragment for KlMdm12 was amplified from the genome of K. lactis and cloned into pET-21d (for nontagged Mdm12; Merck), pET-15b (for KlMdm12 with the N-terminal His₆ tag [HMdm12]), or pET-16b (for KlMdm12 with the N-terminal His₁₀ tag). For expression of His-thrombin-ScMdm12, ScMdm12-TEV-His, and His-TEV-ScMdm12 in E. coli cells, the DNA fragments for corresponding ScMdm12 derivatives were amplified and ligated into pET-15b, pET-22b, and pRSF-1b, respectively. The plasmids for coexpression of the soluble domain of Mmm1 from S. cerevisiae and ScMdm12 were prepared as follows. A DNA fragment for Mmm1 from S. cerevisiae (184–426, ScMmm1s) was cloned into pET-15b at the NdeI and BamHI sites and amplified by PCR with a DNA fragments for the His₆ tag and thrombin cleavage sequence. The resulting DNA fragment was cloned into pRSF-1b. The expression plasmid for ScMdm12 without the His₆ tag was constructed after cloning into pET-22b using the NdeI and BamHI sites. For expression of Mmm1(179–434) from K. lactis (Mmm1s), the DNA fragment for Mmm1s was amplified and ligated into pET-15b (for Mmm1 with the N-terminal His₆ tag [HMmm1s]) or pACYC-Duet (for nontagged Mmm1). For expression of S. cerevisiae Mdm12 (ScMdm12) for lipidome analyses, DNA for ScMdm12 was PCR amplified and cloned into pET-30a(+) (Novagen). Plasmids for expression of KlMdm12 or Mmm1 in yeast cells were prepared as follows. The MDM12 and MMM1 genes containing ∼500- and 300-bp regions upstream and downstream of the initiation and stop codons in the open reading frame, respectively, were PCR amplified using the S. cerevisiae genome DNA as template and cloned into the TRP1-containing pRS314 or pRS316 vector.

The mdm12Δ and mmm1Δ strains (with expression of Vps13-D716H) were described in Tamura et al. (2012) and Kojima et al. (2016), respectively. The mdm12Δ strain, which lacks the chromosomal MDM12 gene and is complemented with the URA3 plasmid bearing the MDM12 gene, was transformed with a TRP1-containing single-copy plasmid harboring the genes for Mdm12 from S. cerevisiae and Mdm12 mutants, including Mdm12 (1–239), from K. lactis. The resulting strains were cultured on SCD-Trp plates containing 5′-fluoroorotic acid to obtain strains expressing S. cerevisiae Mdm12 and K. lactis Mdm12 mutants. Cells were grown in SCD (0.67% yeast nitrogen base without amino acids, 0.5% casamino acid, and 2% glucose) or SCLac (0.67% yeast nitrogen base without amino acids, 0.5% casamino acid, and 2% lactate) media with appropriate supplements.

For crystallization, KlMdm12 (residues 1–239) and KlMdm12FL (residues 1–305) were coexpressed with K. lactis Mmm1s (residues 179–434) in E. coli strain Rosetta (DE3) or BL21(DE3) by incubation at 16°C for 16 h after addition of 0.5 mM IPTG (isopropyl-d-thiogalactoside). Free KlMdm12 or KlMdm12FL dissociated from the Mmm1s-Mdm12(FL) complex was purified by affinity chromatography for the His₆ tag attached to the N-terminus of KlMdm12(FL).

For lipidomics analysis of bound lipids for ScMdm12, the expression plasmid was transformed into BL21(DE3) +pRARE bacteria (Promega). The cells were grown in 2 l kanamycin- and chloramphenicol-containing LB medium to an OD of 0.6. The cells were then cooled down on ice, induced with 1 mM IPTG, and allowed to express the protein for 16–20 h at 15°C. The cells were then harvested by centrifugation, washed with lysis buffer (1 M NaCl, 5 mM MgCl₂, 50 mM Tris-HCl, pH 7.4, 30 mM imidazole, 1 mM PMSF, and 1 mM DTT), and processed for Mdm12 purification. Cell pellet was resuspended in 40 ml lysis buffer and sonicated on ice. One tenth (4 ml) of the sonicated cell suspension was kept for total cellular lipid quantification. After centrifugation, the supernatant was applied to a 5-ml HisTrap column (GE Healthcare), washed with 60 ml of lysis buffer and eluted in the same buffer containing 300 mM imidazole. The fractions containing the protein were pooled and dialyzed against the same buffer without imidazole and PMSF but containing 0.5 M NaCl. The protein was then applied to a HiPrep Superdex 200 26/60 column (GE Healthcare). The fractions containing the monomeric peak were pooled.

For preparation of KlMdm12 variants, the E. coli strain Rosetta (DE3) or BL21(DE3) (Merck) bearing plasmids for expression was cultured at 37°C in LB media containing ampicillin or chloramphenicol. Expression of proteins was induced by adding 0.5 mM IPTG when OD₆₀₀ reached to 0.5, and then the cells were further incubated at 16°C for 16 h. The selenomethionine (SeMet)-labeled proteins were prepared from E. coli cells cultured in minimal media containing 50 mg/l SeMet. The cells were harvested, washed with 20 mM Tris-HCl, pH 7.4, and 300 mM NaCl (TBS300), and disrupted by sonication in TBS300. After centrifugation, the supernatant was loaded onto a Ni-NTA (QIAGEN) column and washed with TBS300 containing 20 mM imidazole, pH 8.0. The target proteins were eluted by TBS300 containing 500 mM imidazole, pH 8.0, and further purified by a Hiload 26/600 Superdex 200 pg column (GE Healthcare) with 20 mM Tris-HCl, pH 7.4, and 150 mM NaCl (TBS150). For high-quality crystal production, the eluted His₁₀-KlMdm12 (HMdm12) fractions were combined and treated with factor Xa protease (QIAGEN) to remove the N-terminal affinity tags. The cleaved proteins were subjected to reductive methylation of Lys and applied to a Superdex 200 10/300 GL column to obtain purified target proteins.

Reductive methylation of Lys in KlMdm12 was performed according to methods reported previously (Bokoch et al., 2010), with minor modifications. In brief, 10 ml of 1 mg/ml Mdm12 in 20 mM Hepes-NaOH, pH 7.4, and 100 mM NaCl was mixed with 500 µl of 2% (wt/vol) dimethylaminoborane and 30 µl formaldehyde and incubated for 4°C for 16 h, and then 200 µl of 2% (wt/vol) dimethylaminoborane and 20 µl formaldehyde were added for further incubation at 4°C for 4 h. The reaction was quenched by 20 mM Tris-HCl, pH 7.4, and the modified proteins were purified by gel filtration chromatography.

The plasmid for expression of the MBP-KlaMmm1s, a fusion protein of MBP followed by K. lactis Mmm1s (179–434) with a linker segment consisting of the TEV protease recognition sequence, His₆ tag, and thrombin recognition sequence between them was constructed as follows. The pET-15b plasmid bearing DNA for K. lactis Mmm1s (179–434) was used to amplify the fragment with the coding sequence for His₆ tag followed by the thrombin recognition sequence, and the resultant DNA fragment was cloned into the pMal-c2x vector (NEB). The TEV protease recognition sequence was inserted between MBP and the His₆ tag at the DNA level. MBP-KlaMmm1s was expressed and purified according to essentially the same procedure used for the Mmm1–Mdm12 complex. MBP was expressed in the same manner as other proteins and purified by amylose resin (NEB) in TBS150 containing 1 mM EDTA by using elution buffer (20 mM maltose and 1 mM EDTA in TBS150). The eluate was collected and loaded onto a Hiload 26/600 Superdex 200pg column in TBS150, and eluted MBP was pooled and concentrated. Mutagenesis of K. lactis Mmm1 was performed with oligonucleotide-directed PCR mutagenesis using adequate primers. Mmm1s–HMdm12 and MBP-HMmm1s containing mutant Mmm1s were expressed in E. coli cells and prepared by the same procedures as those containing wild-type Mmm1s.

The N-terminally His₆-tagged ScHMmm1s and ScMdm12 were coexpressed in the BL21(DE3) strain and purified as KlHMmm1s and KlMdm12 by gel filtration and ion exchange chromatography using a COSMOGEL IEX type Q column (Nacalai Tesque).

Purified KlMdm12 derivatives were concentrated to ∼10 mg/ml and screened with manufactured crystallization screening kits (Hampton research) using an automated robot system mosquito (TTP Labtech) by the sitting-drop vapor-diffusion method at 25°C. Successful conditions that generate crystals were refined to obtain crystals suitable for x-ray diffraction experiments using the hanging-drop vapor-diffusion method. X-ray diffraction data collection were performed for the HMdm12 crystals grown in 0.2 M malate-2Na, 0.1 M Tris-HCl, pH 8.0, and 20% polyethylene glycol 3350. The SeMet-labeled HMdm12 derivative was crystallized under the similar condition. The chemically modified KlMdm12 was crystallized in 1.4–1.6 M sodium potassium phosphate, pH 8.2, and 20% glycerol with or without 5.3 mM detergent FOS-MEA-10 (Anatrace). To remove FOS-MEA-10 from the grown crystals, the crystals were washed with Biobeads (Bio-Rad) in fresh mother liquors for twice and left in the same buffer for at least 3 wk. X-ray diffraction data collection was performed at beamline BL38B1 and BL44XU in SPring-8. The crystals were mounted on the cryo-loop (Hampton Research) and flush cooled in the liquid nitrogen stream. The data were collected up to 3.1 and 3.5 Å resolution for native and SeMet-labeled HMdm12, respectively. The chemically modified protein crystals gave diffraction up to 2.25 Å resolution, and the crystals grown in the presence of FOS-MEA-10 up to 3.3 Å resolution. The diffraction data were processed with HKL2000 (Otwinowski and Minor, 1997). The statistics of the data collection are summarized in Table 1.

The phase problem was solved by an automated software suite (PHENIX; Adams et al., 2010) using single-wavelength anomalous dispersion data from the crystals of SeMet-labeled HMdm12. The structure was refined by refmac5 (Murshudov et al., 2011) and CNS (Brünger et al., 1998), and manual model building was performed by Coot (Emsley et al., 2010). The optimization of secondary structures and energy minimization were performed using a 3D refine web service (Bhattacharya and Cheng, 2013). The structures of native or chemically modified proteins were solved by the molecular replacement method using the constructed model from SeMet-labeled HMdm12 as a search model by PHASER (McCoy et al., 2007) in the CCP4 suite (Collaborative Computational Project, Number 4, 1994). The determined structures were validated by Rampage (Lovell et al., 2003), and secondary structure analysis was performed by DSSP (Kabsch and Sander, 1983; Joosten et al., 2011). Structural images were drawn using PyMOL Molecular Graphics System (version 1.7.0; Schrödinger).

Atomic coordinates and structure factors files have been deposited in the Protein Data Bank under accession numbers 5H54 (KlMdm12(1–239), native), 5H55 (KlMdm12FL from K. lactis, full-length protein), 5H5C (KlMdm12(1–239) with dimethyl lysine modification and FOS-MEA-10 treatment), and 5H5A (KlMdm12(1–239)dmLys with dimethyl-lysine modification).

Lipids bound to free KlMdm12 were extracted as follows. The free form of KlMdm12 was purified from E. coli cells expressing both KlMdm12 and Mmm1s and was vigorously vortexed in 2:1 chloroform/methanol for 20 min. One-fourth volume of 0.1M HCl and 0.1 M KCl was then added, and the sample was further vortexed for 10 min. The organic phase separated by a low-speed spin was dried under a nitrogen stream and subjected to TLC analysis.

ScMdm12-bound lipids as well as total lipids were extracted from E. coli cells, without coexpression of Mmm1, using the Folch procedure (Folch et al., 1957). Lipidomic analysis was performed as described previously (Guan et al., 2010) on a TSQ Vantage triple quadrupole mass spectrometer (Thermo Fisher Scientific). Quantifications of relative amounts of lipids were performed by calculating the area under the curve in the full-scan mass spectrogram for each lipid species analyzed.

Chemical modification of the amino group of phospholipids with 4-fluoro-7-nitrobenzofurazan (NBD-F; Dojindo) was performed according to the manufacturer’s protocol. E. coli total phospholipids in chloroform (Avanti Polar Lipids) were dried up thoroughly by nitrogen gas to form thin films, which were then hydrated by 50 mM boric acid buffer, pH 8.0. The hydrated phospholipids were modified by NBD-F at 65°C for 1 min. The reaction was stopped by cooling on ice. The labeled phospholipids, including crude NBD-PE, were extracted according to the methods reported previously (Ichihara et al., 2011). In brief, methyl tert-butyl ether and methanol were added to the sample (2:1:2 vol/vol) and subjected to a low-speed centrifuge. The resulting upper layer was transferred to new test tubes, dried by a nitrogen stream, and then dissolved in TBS300 to be hydrated. The hydrated crude NBD-PE was mixed with the cell extracts containing free Mdm12 or Mmm1s-Mdm12. Proteins loaded with NBD-PE were purified by the protein purification procedures described in the Protein expression, purification, and modification section.

Phospholipids were obtained from Avanti Polar Lipids except for milk PE from Nagara Science. Lipids in stock solutions in chloroform were mixed at a desired molar ratio, and the solvent was evaporated. The lipid film was hydrated in appropriate buffer. The lipid suspension was incubated at room temperature for 30 min and extruded through polycarbonate 0.1-µm filter using a miniextruder (Avanti Polar Lipids) to prepare liposomes.

The liposome flotation assay was performed as described previously (Krick et al., 2012), with minor modifications. In brief, liposome/protein mixture was put into 5 ml of an ultracentrifuge tube and mixed with 750 µl TBS150 containing 80% (wt/vol) Nycodenz AG (Axis-Shield). The mixture was subsequently overlaid by 750 µl TBS150 containing 30% (wt/vol) Nycodenz and then 450 µl TBS150 without Nycodenz (Fig. 4, A and D). For competition experiments, the mixture was overlaid by 1,000 µl TBS150 containing 15% (wt/vol) sucrose, 1,000 µl TBS containing 7.5% (wt/vol) sucrose, and by 1,000 µl TBS150 without sucrose (Fig. 4 C). After centrifugation at 276,000 g for 1.5 h, fractions were collected from the top, and each fraction was analyzed for NBD fluorescence or protein amounts.

To monitor lipid extraction from liposomes by KlMdm12 or Mmm1s–Mdm12, 10 µM protein was mixed with 200 µM donor liposome, whose lipid composition is egg PC (840051C; Avanti)/NBD-PE (L-α-phosphatidylethanolamine-N-[7-nitro-2-1,3-benzoxadiazol-4-yl], 810118C; Avanti = 80/20), in 150 µl TBS150 and incubated at 37°C for 30 min. The NBD-PE used here is a chicken egg PE derivative containing a head group with an NBD group and acyl chains of 18:1 and 16:0 predominantly. After incubation, the reaction mixtures were diluted with 600 µl ice-cold TBS150 and subjected to liposome flotation (described in the Liposome flotation section). After ultracentrifugation, protein-containing fractions were analyzed for fluorescence excited at 470 nm and observed at 537.5 nm with a F-2100 spectrofluorometer (Hitachi). For liposome competition experiments, 200 µM donor liposomes in 200 µl TBS150 was incubated with 10 µM Mmm1s–HMdm12 in the presence or absence of 200 µM competitor liposomes (PC/PE = 100/0 or 50/50; PE is from milk; Nagara Science) at 37°C for 30 min. Minor fluorescence contribution by the liposome to protein fractions was subtracted. Total NBD fluorescence was estimated from the fluorescence intensity of the NBD-PE–loaded liposomes solubilized with 0.1% Triton X-100.

To monitor lipid transfer from proteins to liposomes, proteins preloaded with 20 µM NBD-PE (see Preparation of proteins preloaded with NBD-PE) were mixed with 400 µM acceptor liposomes containing egg PC alone in TBS150 and incubated at 37°C for 30 min. After incubation, 150 µl of reaction mixture was diluted with 600 µl ice-cold TBS150 and subjected to liposome flotation (described above). After ultracentrifugation, every 250-µl fraction from the top was collected and mixed with 250 µl TBS150 containing 0.2% Triton X-100 for NBD fluorescence measurements.

To monitor lipid transfer between liposomes via the HMmm1s–Mdm12 complex and MBP-HMmm1s, we adopted fluorescent dequenching assays as described previously (Connerth et al., 2012; Watanabe et al., 2015; Miyata et al., 2016). The lipid compositions of the donor liposomes and acceptor liposomes are egg PC/milk PE/egg Liss-Rhod-PE (810146C; Avanti)/NBD-PE (810118C; Avanti) = 50/40/2/8 and egg PC/milk PE = 50/50, respectively. 12.5 µM donor liposomes was incubated with 50 µM acceptor liposomes in the presence or absence of the indicated proteins in 600 µl assay buffer (20 mM Tris-HCl, pH 7.5, and 150 mM NaCl) at room temperature. For NBD-PS transfer assays, NBD-PE in the aforementioned assays was replaced with NBD-PS (1,2-dioleoyl-sn-glycero-3-phospho-L-serine-N-(7-nitro-2-1,3-benzoxadiazol-4-yl; 810198; Avanti). The increase in dequenched NBD fluorescence was monitored using a F-2100 spectrofluorometer (Hitachi).

In vitro phospholipid transport assays were performed according to the method described previously (Kojima et al., 2016). For preparation of HMFs, the mmm1Δ and mdm12Δ strains used here (Tamura et al., 2012) were cultured on synthetic media containing lactate as a carbon source to eliminate ρ⁰ or ρ⁻ strain before membrane isolation. HMFs were isolated from mmm1Δ cells expressing Vps13-D716H from the pRS314 plasmid under the control of the GPD promoter with wild-type Mmm1 or mutant Mmm1s in the mmm1Δ strain or wild-type Mdm12 or mutant Mdm12s in the mdm12Δ strain from the pRS316 plasmid with its own promoter, which were cultivated in SCD medium at 30°C for 15 h (Kojima et al., 2016).

For in vitro cross-linking, the Mdm12–Mmm1s complex was prepared as follows. The DNA fragment for KlaMdm12(1–239) or its I5C mutant was cloned into the pET-22b vector to make a fusion protein with the TEV protease cleavage site followed by the His₁₀ tag at the C terminus and that for KlaMmm1(179–434) into the pRSF-1b vector. The E. coli BL21(DE3) strain was transformed with those plasmids. Wild-type and I5C mutant of Mdm12 complexed with Mmm1s was expressed and purified from the E. coli BL21(DE3) strain transformed with those plasmids with the same procedure used for the other Mdm12–Mmm1s complexes. Purified Mdm12–Mmm1s complexes were mixed with a 5 M excess of 4-N-maleimidobenzophenone (NMBP; Sigma) in the presence of 1 mM tris(2-carboxyethyl)phosphine and incubated for 2 h at room temperature under dark conditions. The reaction was stopped by the addition of 5 mM DTT, and the proteins were separated from the reagents by gel-filtration chromatography. The NMBP-modified Mdm12(I5C)–Mmm1s complex was transferred to a 96-well dish and subjected to UV irradiation for 15 min on ice. The cross-linked products were analyzed by SDS-PAGE followed by immunoblotting.

We performed MD simulations for two systems using the AMBER12 program package (Case et al., 2005). One is the simulation of Mdm12 including a 1-palmitoyl-2-oleyl-sn-glycero-3-phosphocholine (POPC) molecule and explicit water molecules, and the total number of atoms was 52,280. Another is that of only Mdm12 with explicit water molecules, and the total number of atoms was 52,221. Here, 16 Na⁺ ions were included for each system to neutralize the systems. The force fields were AMBER ff99SB (Hornak et al., 2006), Lipid11 (Skjevik et al., 2012), and TIP3P water model (Jorgensen et al., 1983) for Mdm12, POPC, and explicit water molecules, respectively. Initial structures were built using the LEaP module in AmberTools (Case et al., 2005) and then minimized and equilibrated in three stages: heating for 200 ps from 10 K to 300 K with heavy atom positional restraints of strength 10 kcal mol⁻¹ Å⁻², equilibration for 5 ns with the restraints, and equilibration for 40 ns without restraints. After that, the production runs for each system were performed for 120 ns. The MD simulations under the isobaric-isothermal ensemble with periodic boundary conditions using particle mesh Ewald method (Darden et al., 1993) to treat long-range electrostatics and a real space cutoff of 12 Å. Temperature was controlled with a Langevin thermostat with a collision frequency of 1 ps⁻¹. Pressure regulation was achieved with isotropic position scaling, a Berendsen barostat (Berendsen et al., 1984) of ∼1 atm, and a pressure relaxation time of 1 ps. Bonds involving hydrogen were constrained using the SHAKE algorithm (Ryckaert et al., 1977).

Fig. S1 shows the sequence alignment with secondary structures of Mdm12 proteins and E-SYT2. Fig. S2 documents diagrams of the Mdm12 and Mmm1 derivatives used in this study. Fig. S3 shows the comparison of the structures of KlMdm12 derivatives, indicating that they are essentially similar, and compares the structures of KlMdm12 with that of ScMdm12. Fig. S4 compares the structures of KlMdm12 with and without phospholipid. Although lipid-free KlMdm12 exhibits more structure flexibility, the size and shape of the lipid-binding pocket do not differ significantly between lipid-bound and lipid-free KlMdm12. Fig. S5 shows the results of in vitro phospholipid transport assays using HMFs containing the ER and mitochondrial membranes with various Mmm1 or Mdm12 mutants.

We thank the members of the T. Endo laboratory for discussions and critical comments on the manuscript. We also thank Drs. Kaori Yunoki and Chika Saito for technical assistance. The synchrotron radiation experiments were performed at beam lines BL38B1 and BL44XU at SPring-8, Japan, with the approval of the Japan Synchrotron Radiation Research Institute (proposals 2012B1092 and 2013A1032).

This work was supported by Japan Society of the Promotion of Science Grants-in-Aid for Scientific Research (25840020 and 17K18230 to S. Kawano, 15H05595 to Y. Tamura, and 15H05705 and 2222703 to T. Endo), Japan Science and Technology Agency/Core Research for Evolutional Science and Technology (grant JPMJCR12M1 for T. Endo), the Swiss National Science Foundation (H. Riezman), and the National Centre of Competence in Research Chemical Biology (H. Riezman). R. Kojima was a Research Fellow of the Japan Society of the Promotion of Science.

The authors declare no competing financial interests.

Author contributions: S. Kawano and T. Endo designed the research and wrote the paper. Y. Tamura designed a part of the research and wrote the paper. S. Kawano performed most experiments, and Y. Tamura and R. Kojima performed a portion of the experiments (except for lipid analyses and MD calculations). S. Bala and E. Asai prepared materials. B. Kornmann and H. Riezman designed and A.H. Michel and I. Riezman performed experiments for lipid analyses of S. cerevisiae Mdm12. For MD simulations, Y. Sakae and Y. Okamoto designed and Y. Sakae performed the computations.