Molecular mechanism of mitochondrial phosphatidate transfer by Ups1/Mdm35

Cardiolipin plays many important roles for mitochondrial physiological function and is synthesized from phosphatidic acid (PA) at inner mitochondrial membrane (IMM). PA synthesized from endoplasmic reticulum needs to transfer to IMM via outer mitochondrial membrane (OMM). The transfer of PA between IMM and OMM is mediated by Ups1/Mdm35 protein family. Although there are many structures of this family available, the detailed molecular mechanism of how PA is transferred between membranes is yet unknown. Here, we report another crystal structures of Ups1/Mdm35 in the authentic monomeric apo state and the DHPA bound state. By combining subsequent all-atom molecular dynamics simulations, extensive structural comparisons and biophysical assays, we discovered the conformational changes of Ups1/Mdm35, identified key structural elements and residues during membrane binding and PA entry. We found the monomeric Ups1 on membrane is an important transit for the success of PA transfer, and the equilibrium between monomeric Ups1 and Ups1/Mdm35 complex on membrane affects the PA transfer rate and can be regulated by many factors including environmental pH.


INTRODUCTION 1
As the powerhouse of cell, mitochondria produce energy through oxidative 2 phosphorylation, a process in which ATP is created via electron transfer chain and 3 ATP synthase embedded in the inner membrane of mitochondria (IMM). IMM has 4 a very high concentration of cardiolipin (CL) that accounts for about 20% of the 5 total lipids of IMM 1 . CL not only interacts with a number of IMM proteins, but also 6 serves as an integral component of respiratory complexes and participates in their 7 folding 2 . In recent years, the role of CL in mitochondrial signaling pathways has 8 also been highlighted. Upon stress signals, CL is externalized to the outer 9 membrane of mitochondria (OMM) forming a binding platform for the specific 10 recruitment of signaling molecules that are required for mitophagy and apoptosis 3 . 11 In eukaryotes, CL is synthesized on the matrix side of IMM via an enzymatic 12 cascade starting from phosphatidic acid (PA) 4 . PA is a rare component of the 13 mitochondrial membrane (< 1% of the total lipids) that is predominantly 14 synthesized in endoplasmic reticulum (ER) and imported to OMM 1,5 . Although it 15 remains unclear how PA is transferred from ER to OMM, the hetero-dimeric 16 protein complex, termed Ups1/Mdm35 in yeast or PRELID1/TRIAP1 in 17 mammalian cells, has recently been identified as a lipid transfer complex that 18 mediates the transport of PA from OMM to IMM 6-10 . 19 Ups1 is located in mitochondrial intermembrane space (IMS) and was 20 identified as a member of the conserved UPS1/PRELI family proteins, which are 21 related to mitochondrial function 11 . Ups1 was initially found to affect the 22 biogenesis of Mgm1, which is the homologue of human OPA1 and required for 23 mitochondrial fusion 12 . Cells lacking Ups1 showed dramatically reduced amounts 24 of short-form Mgm1 (s-Mgm1) and had less tubular mitochondria 12 . Deletion of 25 Ups1 resulted in a considerable decrease in CL level 13 . However, this decrease 26 could be restored by simultaneous deletion of Ups2, which is the homolog of Ups1 1 in yeast. Mdm35 is another conserved yeast IMS protein and belongs to the twin 2 Cx9C protein family 14 . As the binding partner of the Ups proteins, Mdm35 not 3 only facilitates their import into mitochondria, but also protects them against 4 proteolysis 15,16 . The role of Ups1/Mdm35 complex in lipid transfer was discovered 5 in 2012 and for the first time Ups1 was recognized as a lipid transfer protein that 6 can shuttle PA between mitochondrial membranes 8 . In addition, the dynamic 7 assembly of Ups1 with Mdm35 was found to be essential for the PA transfer 8 process. The closest homologue of Ups1/Mdm35 in human is PRELID1/TRIAP1 9 that facilitates PA transfer in vitro 9 . In addition to PRELID1, human possesses 10 another three Ups1 homologues named PRELID2, SLMO1 and SLMO2, all of 11 which shares the conserved "PRELID" domain with PRELID1 and form a complex 12 with TRIAP1 (a homologue of Mdm35) 10 . 13 In 2015, two groups reported crystal structures of Ups1/Mdm35 with or 14 without PA, in which the interaction between Ups1 and Mdm35 and the PA 15 binding pocket of Ups1 were revealed 17,18 . In the same year, Miliara et al reported 16 the crystal structure of SLMO1/TRIAP1, which is structurally similar with 17 Ups1/Mdm35. Despite no significant sequence homology, Ups1/Mdm35 as a 18 whole shows a remote structural similarity to other lipid transfer proteins (LTPs) 19 including the phosphatidylinositol transfer proteins (PITPs) and the related 20 cholesterol-binding START domains 19,20 . Structural comparison highlights the 21 α2-loop in Ups1 that seems to be the equivalent region of the lipid exchange loop 22 of PITP-α or the Ω1-loop in the START domains 17,21 . These two loops were 23 thought to be the membrane-docking site for lipid loading and unloading and 24 function as a lid of the lipid-binding cavity. The outward and inward movements of 25 the lid cause the widening and closing of the cavity entrance, respectively, which 26 bound DHPA is close to the positively charged side chain of R25, and the side 1 chains of H33, K58 and N152 are within the distances that allow formation of 2 electrostatic interactions or hydrogen bonds with the phosphate oxygen atoms. 3 The sn-1 and sn-2 acyl chains of DHPA are stabilized via hydrophobic interaction 4 with T76, I78, N97, M104 and V106 (Fig.2b). By comparison with the 5 Ups1/Mdm35-DLPA structure, we found DLPA inserts into the positively charged 6 pocket more deeply than DHPA ( Supplementary Fig.6). This may be due to the 7 acyl chain of DLPA is longer than that of DHPA, so that DLPA needs to occupy 8 more space in the pocket. 9

A new interaction mode between Ups1 and Mdm35 10
In the previously reported Ups1/Mdm35 structures, all three α helices of 11 Mdm35 interact with Ups1. However, in the structure of Ups1/Mdm35-DHPA, only 12 αA and αB helices of Mdm35 interact with Ups1, giving a buried surface area of 13 about 790 Å 2 , which is about 8.1% of the whole surface area of Ups1. This is 14 about 65% of the buried surface area between Ups1 and Mdm35 in other 15 Ups1/Mdm35 structures. Thus, by changing the orientation of the αC helix, 16 Mdm35 can adjust its interaction strength with Ups1. 17 To examine whether the interaction strength between Ups1 and Mdm35 18 would affect the PA transfer activity, we made three Ups1/Mdm35 truncation 19 mutants (Fig.2d) for functional assays. The △C5 mutant lacks the C-terminal 5 20 residues (K82-K86) of Mdm35. The △C14 mutant lacks the C-terminal 14 21 residues (A73-K86), which are located just after the αC helix and untraced in the 22 crystal structures. Secondary structure prediction shows that this segment is 23 disordered. In the △(αC+C14) mutant, the αC helix of Mdm35 was further deleted, 24 as a result this mutant can only bind to Ups1 in a weak interaction. First, we 25 measured the PA transfer activities of these mutants by fluorescent-based PA 1 transfer assay (Fig.2e). The △C5 mutant was used as a negative control, since 2 Watanabe et al had proved this mutant has no effect on the PA transfer activity of 3 Ups1/Mdm35 17 . As shown in Fig.2f, the △C5 and △C14 truncations have little 4 effect on the NBD fluorescence increase, whereas the △(αC+C14) truncation 5 dramatically impairs the PA transfer activity. This indicates the αC helix of Mdm35 6 plays an important role in PA transfer process and suggests that as the weakened 7 interaction between Ups1 and Mdm35 decreases the PA transfer activity of 8 Ups1/Mdm35. 9 We wanted to know whether the interaction strength between Ups1 and 10 Mdm35 would affect the membrane binding ability of Ups1, thus the liposome 11 co-sedimentation assay was performed (Fig.2g). When a system composed of 12 proteins and the liposome-bound proteins can be separated by ultracentrifugation. 21 The resulting supernatant and precipitate fractions were analyzed by SDS-PAGE. 22 By quantifying the Ups1 and Mdm35 bands in the two fractions, the distribution of 23 the two different states of Ups1 on liposomes can be calculated. We found that 24 the overall membrane binding ability of Ups1 increases most significantly in 25 △(αC+C14) mutant, compared to that of WT (Fig.2h). This is mainly due to the 26 significant increase in the amount of membrane bound monomeric Ups1. This 1 suggests that once the interaction between Ups1 and Mdm35 is weakened, more 2 Ups1 monomer will tend to stay on the lipid membrane. Interestingly, we found the 3 amount of membrane-bound monomeric Ups1 in △C14 mutant was also 4 significantly increased compared to that of the WT. This suggests that the 5 C-terminal 14 residues of Mdm35 are likely to be involved in the interaction with 6 Ups1, although the high flexibility of this segment makes it invisible in the crystal 7 structures. 8 The interaction between Ups1/Mdm35 and membrane 9 By liposome co-sedimentation assay, we have detected that both 10 Ups1/Mdm35 and monomeric Ups1 could bind to the membrane. To investigate 11 how Ups1 interacts with the membranes in these two different states, we 12 performed all atom molecular dynamics (MD) simulations in which either an 13 Ups1/Mdm35 complex or a freestanding Ups1 (Ups1 free ) interacting with 14 DOPA-containing lipid bilayers (Supplementary Movies 1 and 2). The initial 15 orientation of Ups1 relative to the lipid membrane was defined according to the 16 structural comparison between Ups1/Mdm35 and PITP-α. Because the α2-loop of 17 Ups1 was thought to be the equivalent region of the lipid exchange loop of PITP-α, 18 it may interact with the membrane in the same orientation as the lipid exchange 19 loop. Thus we chose the initial orientation of Ups1 to make its α2-loop facing onto 20 the membrane (Fig. 3a). The time evolution RMSD profiles show that, the 21 conformational change of Ups1 free is relatively small and converges sufficiently 22 after 300 ns, whereas those of Ups1/Mdm35 complex are larger for both Ups1 23 (Ups1 complex ) and Mdm35 parts and seem not to converge within the simulation 24 time (Fig. 3b). Because Mdm35 was thought to stabilize residues at the interface 25 between Ups1 and Mdm35, the increased RMSD of Ups1 complex may have hinted for allosteric sites at their interface. This speculation was confirmed by the 1 root-mean-square-fluctuation (RMSF) profiles of Ups1 in these two simulations 2 (Fig.3c). As expected, Ups1 complex has a significant lower residual fluctuation at the 3 loop region (residues N43-N49) surrounded by Mdm35 complex (~1.5 Å) than that in 4 Ups1 free (~2.5 Å), and had relatively higher RMSF at some distant segments 5 (residues N28-H33, V66-L70, L100-G104 and K140-V163). The higher RMFS 6 would affect the subsequent membrane binding of Ups1. 7 The time evolution profiles of the distances between the membrane surface 8 phosphate plane (MSPP) and the center-of-mass (COM) of Ups1 in these two 9 simulations show a sharp decrease of the distance for Ups1 free . After 400 ns 10 Ups1 free has a stable distance to the membrane, which is significantly different to 11 Ups1 complex (Fig. 3d). This is also reflected by the converged RMSD profile of 12 Ups1 free after 400 ns. And Ups1 free is found to establish stable protein-membrane 13 interaction by inserting F69 that is located at the α2-loop into the lipid bilayers (Fig.  14   3d). The insertion of F69 into the membrane causes the COM of Ups1 free to 15 approach the membrane bilayer by 9 Å, which should facilitate PA entering the 16 lipid-binding pocket of Ups1. As expected, the direct interaction between PA and 17 Ups1 free was observed after 800 ns (Supplementary Movie 3). A snapshot of 18 Ups1 free interacting with the membrane at 1,023 ns is studied here (Fig. 3e), in 19 which a PA molecule is trying to enter the lipid-binding pocket of Ups1 free . The 20 phosphate head of PA is surrounded by four positively charged residues H33, K61, 21 K148 and K155, which had been determined important for the PA transfer 17,18 . 22 Ups1 free interacts with the lipid membrane through the hydrophobic and positively 23 charged residues located in the L2-loop (N28-H33), α2-loop (K61-R71) and the 24 C-terminal long α3-helix (K140, R146, K154, M158, F162, K166 and R171). To be 25 noted that, these membrane-bound structural elements contain exactly those 26 residues with the high RMSF in Ups1 complex (Fig. 3c). The L2-loop is conserved, 1 indicating its importance for Ups1 function ( Supplementary Fig.7). The α2-loop is 2 believed the lid for the PA binding pocket and may participate in membrane 3 binding 21 . As for the C-terminal long α3-helix, despite no significant sequence 4 homology, its secondary structure is conserved among the members of PRELID 5 family. Interestingly, we found although Ups1/Mdm35 does not establish a stable 6 interaction with the membrane, it tries to approach the membrane in the same 7 orientation as Ups1 free after 200 ns (Supplementary movie 1). 8 To test whether Ups1 interacts with the membrane as the MD simulation 9 predicted, we performed the following experiments. First, we tested whether the 10 C-terminal α3-helix of Ups1 is involved in membrane binding. Previous studies 11 revealed that Ups1 exclusively binds to liposomes containing negatively charged 12 phospholipids 8 . And our MD simulation shows that a number of positively 13 charged residues (K140, R146, K154, K166 and R171) in the α3-helix contact the 14 negatively charged membrane surface. If these basic residues really participate in 15 membrane binding, mutating them into Glu should impair the PA transfer activity 16 of Ups1/Mdm35. We constructed a series of mutations on these residues and 17 tested their abilities of membrane binding and PA transfer. alone is enough to completely eliminate the PA transfer activity of Ups1 (Fig.3f). 22 Since structural studies of Ups/Mdm35 had excluded the possibility of interaction 23 of these basic residues with Mdm35 or PA, the loss of PA transfer activity in these 24 mutants would be due to disruption of the Ups1-membrane interaction. This 25 hypothesis was ascertained by the result of liposome co-sedimentation assay. As 26 shown in Fig. 3g, for these mutants, the amount of membrane-bound 1 Ups1/Mdm35 (defined as State A) decrease significantly, which leads to the 2 decrease of their overall membrane binding abilities, compared to that of WT. It is 3 worth noting that for these mutants the amount of membrane bound monomeric 4 Ups1 (defined as State B) significantly increase, compared to that of WT, which 5 will be discussed later. 6 Then we examined whether F69 is involved in membrane binding of 7 Ups1/Mdm35. F69 is located in the α2-loop, which contains positively charged 8 and hydrophobic residues, but no negatively charged residues. Its amino acid 9 composition and structural similarity with the lipid exchange loop of PITP-α make 10 it possible to participate in binding to the negatively charged membrane 11 ( Supplementary Fig.7). If F69 is involved in membrane binding, mutating F69 to 12 the negatively charged residue Glu should impair the membrane binding ability of 13 Ups1/Mdm35, whereas mutating F69 to a residue with a similar side chain, such 14 as Leu, should not have much effect on its membrane binding ability. The result of 15 liposome sedimentation assay confirmed our hypothesis (Fig.3h). The overall 16 membrane binding ability of F69E mutant is significantly impaired. Meanwhile, 17 F69E mutation also causes a large accumulation of monomeric Ups1 on the 18 membrane. As for F69L mutant, the amount of Ups1/Mdm35 and monomeric 19 Ups1 on the membrane is similar to that of WT. To test whether it is possible for 20 F69 to be buried into the membrane during the PA transfer process, we further 21 mutated F69 into small hydrophobic (Ala), bulky uncharged polar (Tyr) and bulky 22 hydrophobic (Trp) side residues and measured their PA transfer activities (Fig.3i). 23 The F69E mutant nearly abolished the PA transfer activity. Both F69A and F69Y 24 mutants markedly impair the PA transfer activities. As for F69W and F69L 25 mutants, they keep the same PA transfer activities with WT. Since F69 can only 26 be replaced by residues with large hydrophobic residues, it should be in a 1 completely hydrophobic environment at some stage at least during the PA 2 transfer process. Structural analysis shows that F69 is unlikely to interact with the 3 hydrophobic residues of Ups1 itself, thus a reasonable explanation is that this 4 residue would be buried in the membrane during the PA transfer process, as it is 5 predicted from MD simulation. 6 Potential conformational change upon PA binding 7 To explore how PA enters the PA-binding pocket of Ups1 and whether Ups1 8 has a conformational change before and after PA binding, we compared the 9 crystal structures of Ups1 in different states, including the MD simulated 10 membrane-bound state (Ups1 free -1023ns), the apo state (PDB codes 5JQM and 11 4YTW), the DLPA-bound (PDB code 4YTX, chain B) and DHPA-bound state 12 (PDB code 5JQO). In addition, Y. Watanabe et al observed another state of Ups1 13 (PDB code 4YTX, chain L), in which Ups1 has a more open α2-loop, yet has no 14 PA in its lipid binding pocket 17 , which is referred to X-state here. 15 After superimposing the five crystal structures to the MD simulated 16 membrane bound structure of Ups1, we found although the overall structures of 17 the six molecules are very similar, the conformations of the α2-loop and the 18 N-terminus of the α3-helix (N-α3-helix) are obviously different (Fig.4a). In the 19 Ups1 free -1023ns and apo state structures (PDB codes 5JQM and 4YTW), the 20 hydrophobic side chains of P63 and W65 in the α2-loop bind to the hydrophobic 21 region formed by I137, V141 and W144 of the N-α3-helix, giving a buried surface 22 area of 499 Å 2 , 344 Å 2 and 234 Å 2 (~31%, 24% and 15% of the total surface area 23 of the α2-loop), respectively ( Fig. 4b and Table 1). However, in the PA-bound and 24 the X-state structures, both the α2-loop and N-α3-helix swing outward, which 25 increases the distance between the α2-loop and N-α3-helix and destroys their 1 hydrophobic interaction (Fig. 4a). To better compare the changes of proximity 2 between the α2-loop and N-α3-helix in these structures, we selected four adjacent 3 residues (P63, W65, V141 and W144) located in the α2-loop and N-α3-helix (Fig.  4 4b) and measured the distance between them (P63-V141, P63-W144, W65-V141, 5 and W65-W144). As shown in Table 1, the four distances are the smallest in 6 Ups1 free -1023ns structure, followed by the apo-state, PA-bound state, and finally 7 the X-state. Because both α2-loop and N-α3-helix are involved in membrane 8 binding, the change of their proximity leads to the change of the opening size on 9 the membrane interaction interface of Ups1 (Fig. 4c). For the 10 membrane-anchored Ups1 free -1023ns and the apo state structure, there is only a 11 small hole on this interface, which is lined by residues H33, K61, K148 and K155 12 and might be involved in recognizing or binding the head of PA. For the two 13 PA-bound Ups1 structures, the hole is enlarged into a cleft. We speculated it 14 might be due to the PA occupation in the pocket. In the X-state, the cleft is 15 continued to expand into a wedge-shaped cavity. Although there is no density of 16 DLPA in its PA binding pocket, we speculated the structure might represent an 17 intermediate state in which PA is entering or exiting from the pocket, thus 18 requiring a wider channel on the membrane interaction interface to accommodate 19 the two long acyl chains of PA. To test our hypothesis, we re-analyzed the 20 electron density of the X-state and found an additional density near the membrane 21 contact surface, which could accommodate a DLPA molecule well (Fig.4d). 22 The structural comparison further reveals that the positions of F69 and W65 23 vary among different states (Fig.4b). In PA-bound and X-state structures, it is W65 24 that is inserted into the membrane to anchor Ups1. In Ups1 free -1023ns structure, 25 F69 is inserted into the membrane. In apo-state structures, both F69 and W65 are 26 not inserted into the membrane. It seems that the conformational changes of the 1 α2-loop and N-α3-helix result in the membrane-anchoring residue of Ups1 2 alternating between W65 and F69 during the PA transfer process (Supplementary 3 movie 4). 4 To test whether the hydrophobic interaction between the α2-loop and 5 N-α3-helix will affect the function of Ups1/Mdm35, we generated a W65A mutant 6 and an I137A/V141A/W144A triple mutant of Ups1. Both of them have an 7 impaired PA transfer activity (Fig.4e). We also constructed a W65A/F69A double 8 mutant and found the PA transfer activity of Ups1 is nearly abolished (Fig.4e). 9 To test whether the alternate insertion of F69 and W65 into the membrane is 10 essential for the function of Ups1/Mdm35, we wanted to lock the position of W65 11 or F69 so that the membrane anchoring of Ups1 can only be accomplished by one 12 of them (Fig.4f). First, we designed the L70C/I103C mutant in which F69 cannot 13 be inserted into the membrane. In the PA-bound structures, W65 interacts with 14 I137, V141 and W144, and F69 is located above the membrane surface. The 15 distance between the Cα atoms of L70 and I103 in the DLPA-bound structure 16 (PDB code 4YTX, chain B) is 8.8 Å. If a disulfide bond is formed between L70 and 17 I103 by mutating to cysteines, the distance between their Cα atoms should be 18 reduced to 5.5-5.7 Å, which should make the reside (L70C) away from the 19 membrane. As a result, the adjacent residue F69 should also be away from the 20 membrane. Meanwhile, it should be less affected for W65 that still has a chance 21 to insert into the membrane. Second, we designed another T64C/K140C mutant 22 based on the structure of Ups1free-1023ns, in which W65 cannot be inserted into 23 the membrane (Fig.4f). The position of W65 is locked to interact with I137, V141 24 and W144 in this mutant while F69 is almost unaffected and still has the ability to 25 insert into the membrane. 26 After purifying the two mutants, we found that each pair of the introduced 1 cysteines can form an intramolecular disulfide bond spontaneously as expected. 2 As shown in Fig. 4g, the bands of L70C/I103C and T64C/K140C mutants shift 3 upward upon DTT treatment. Furthermore, the disulfide bonds between L70C and 4 I103C, and between T64C and K140C were validated by mass spectrum 5 ( Supplementary Fig. 8). Although these two intra-cross-linked mutants could 6 hardly transfer PA, their PA transfer activities were restored after 10 mM DTT 7 treatment (Fig.4h). These data suggest that the conformational changes 8 (membrane insertion) of both F69 and W65 are important for PA transfer activity 9 of Ups1/Mdm35. Interestingly, for the T64C/K140C mutant, we also detected a 10 small amount of inter-cross-linked domain-swapped dimer, which was not 11 detected in the previous N28C /G159C mutant 17 . 12

Regulation of PA transfer activity by pH 13
Because the pH of IMS is significantly influenced by the respiratory chain 14 function 25 , we guessed that the pH value would regulate the PA transfer activity 15 of Ups1/Mdm35. After measuring the PA transfer activity of Ups1/Mdm35 under 16 different pH, we found that the optimum pH of Ups1/Mdm35 is 7.0 ( Fig. 5a). At pH 17 6.5 and pH 7.5, the activity is slightly lower. At pH 6.0, the activity is significantly 18 lower than that at pH 7.0. And at pH 5.5, the activity is even less and falls to less 19 than one-third. Considering there are many basic residues located in the 20 membrane binding surface of Ups1, we suspect that the pH may affect the affinity 21 between Ups1/Mdm35 and the membrane by changing the charge on these 22 residues. To test this, the electrostatic potentials and net charge of Ups1 23 molecules in different states under different pH values were calculated (Table 2) 24 using the PDB2PQR server 26 . We found all of the molecules carry the largest 25 number of positive charges at pH 5.5 and the number of charges carried by Ups1 varies according to different states and pH. The change in electrostatic potentials 1 under different pH mainly occurs at the entrance of the PA binding pocket (Fig.  2   5b). It seems that under lower pH, Ups1 would have a stronger membrane binding 3 ability due to its increased positively charge. This hypothesis was confirmed by 4 the result of liposome co-sedimentation assay (Fig.5c) that at lower pH more 5 monomeric Ups1 accumulates on the membrane, which would be the reason for 6 its decreased PA transfer activity (see discussions below). 7

DISCUSSIONS 8
Although there have been many crystal structures of Ups1/Mdm35 (or its 9 homologous) in different states reported 17,18,21,23 , the molecular details of how PA 10 is transferred between membranes is yet unknown. In the present study, we 11 reported another two crystal structures of Ups1/Mdm35 in a truly monomeric apo 12 state and a novel DHPA bound state, which provides us an opportunity to 13 compare all available structures systematically and gain further insights into the 14 conformational changes of Ups1/Mdm35. 15 Besides, we developed the liposome co-sedimentation method to 16 quantitatively detect membrane bound Ups1 in Mdm35 complexed and 17 none-complexed free states, which discovered the relationship between the 18 portion of membrane bound Ups1 in free state and the PA transfer activity of 19 Ups1/Mdm35. Our dedicative all-atom molecular dynamics simulations further 20 indicated that the none-complexed Ups1 can form a more stable interaction with 21 membrane in comparison with Ups1/Mdm35. MD simulations further disclosed 22 novel key elements, the α2-loop, α3-helix and L2-loop, involved in membrane 23 binding and potential PA entry, which were subsequently verified by mutagenesis 24 experiments. By comparing the PA transfer activities among different truncations 25 of Mdm35, we found the interaction strength between Mdm35 and Ups1 also 1 regulates the PA transfer activity of Ups1/Mdm35. 2 Based on all available crystal structures, MD simulations and all biophysical 3 assays in this study, we could attribute different structures of Ups1/Mdm35 into 4 membrane-free apo state (PDB codes, 5JQM and 4YTW), membrane-bound apo 5 state (MD simulated structure), membrane-bound PA entry state (PDB code, 6 4YTX) and PA bound state (PDB codes, 5JQO and 4YTX). Both liposome 7 co-sedimentation experiments and MD simulations indicated the none-complexed 8 Ups1 is the important transit during Ups1/Mdm35 mediated PA transfer, which 9 was also suggested previously that the extraction of PA from membrane or 10 delivery of PA into membrane is performed by Ups1 independently 8 . With these 11 basic concepts as well as the key elements for PA transfer discovered in this 12 study, we could propose a detailed molecular model for Ups1/Mdm35 mediated 13 PA transport ( Fig.6 and Supplementary Movie 4). 14 First, when Ups1/Mdm35 approaches the lipid bilayers, Ups1/Mdm35 intends 15 to interact with membrane through the membrane-binding residues of Ups1. 16 These residues include N28-H33 of the L2-loop, and the hydrophobic and 17 positively charged residues located in the α2-loop and the C-terminal long 18 α3-helix. Although these membrane-binding residues are highly flexible in the 19 presence of Mdm35, they can still interact with the membrane and allow 20 Ups1/Mdm35 associated onto the membrane as an integral part 21 Second, Mdm35 will dissociate from Ups1, which begins with the αC-helix 23 detachment, followed by the detachment of αA and αB helices. This liberates the 24 residues of Ups1 located at the Ups1-Mdm35 interface and the resulting allosteric 25 effects alleviate the flexibility of the membrane-binding residues of Ups1. Under 1 this circumstance, the hydrophobic interaction (mediated by P63, W65, I137, 2 V141 and W144) between the α2-loop and the N-α3-helix is enhanced, resulting 3 in the insertion of F69 into the membrane. This event makes the COM of Ups1 4 further approaching to the lipid bilayer (membrane-anchored Ups1), which is 5 important for PA extraction. 6 Third, the negatively charged phosphate head of PA is attracted to the 7 positive charged residues H33, K61, K148 and K155, and interacts with them, 8 which triggers conformational changes of the N-α3-helix and α2-loop. They move 9 in opposite directions to enlarge the entrance of the PA-binding pocket, allowing 10 the two long acyl tails of PA to enter into the pocket. During this process W65 is 11 inserted into the membrane as a substitute of F69. When PA enters into the 12 binding pocket completely, the N-α3-helix swings inwards a little to avoid the exit 13 of PA from the pocket. 14 Fourth, Mdm35 interacts again with PA bound Ups1 to form Ups1/Mdm35-PA. 15 In the fifth and final step, Ups1/Mdm35-PA releases from the membrane to the 16 solution. The PA release process by Ups1/Mdm35 is just a reverse process of PA 17 extraction and will not be discussed further. has an impaired PA transfer activity, whereas is normal in binding to membrane 17 . 22 They thought the α2-loop is not involved in membrane binding, but only plays as a 23 gate for PA entry and exit. However, considering PA comes from the membrane, 24 it's difficult to imagine how α2-loop acts as a gate without membrane interaction. 25 forward/reverse reaction constants for each step are described as k 1 /k -1 , k 2 /k -2 , 14 k 3 /k -3 , and k 4 /k -4 , and the association/dissociation constants of PA to Ups1 are 15 denoted as k a /k d . The reaction (a) represents the process that Ups1/Mdm35 binds 16 onto donor membrane and performs PA extraction while the reaction (b) 17 represents the process that Ups1/Mdm35 binds to acceptor membrane and 18 performs PA release. The reaction (b) is just a reverse procedure of the reaction 19 , thus sharing the same reaction constants. In the following discussion, we 1 assume that the association/dissociation reaction rates between U/M and 2 membrane are similar with that between U-PA /M and membrane. Thus, k 4 /k -4 3 approximately equal to k -1 /k 1 . 4 If steady state conditions are reached, ten equations can be derived from represents the concentration of membrane (see Fig. 2g). 13 Then we could calculate the binding constant between Ups1/Mdm35 and 14 membrane by utilizing the experimental data from liposome co-sedimentation 15 ( Supplementary Fig.9). We found the K1 values of 2E, 3E and 4E mutants are 16 much lower than that of WT, indicating that mutating the positively charged 17 residues at Ups1-membrane binding interface to negatively charged residues 18 decreases the membrane binding ability of Ups1/Mdm35, which is in consistency 19 with our prediction. Furthermore, the K1 value of △(αC+C14) mutant is much 20 higher than that of WT, suggesting the mutant binds to the membrane more easily, 21 which is also in consistency with our prediction that Ups1 needs to disassociate 22 from Mdm35 and form a stable interaction with membrane, and the weaker 23 binding between Mdm35 and Ups1 decreases the energy barrier of disassociation 24 and increases the possibility for Ups1 binding to membrane. As a result, the 1 above model of the detailed PA transfer by Ups1/Mdm35 is fairly consistent with 2 our experimental data. 3 From the liposome co-sedimentation experiments and the PA transfer assays, 4 we could found that weakening the interaction between Mdm35 and Ups1 5 increases the amount of total membrane bound Ups1, which is majorly 6 contributed from the amount of monomeric membrane bound Ups1 (Fig. 2h). 7 Besides, alteration of Ups1-membrane binding interface not only decreases the 8 amount of membrane bound Mdm35/Ups1 complex but also increases the 9 amount of monomeric membrane bound Ups1 (Figs. 3g and 3h). We could further 10 found that the amount of monomeric membrane-bound Ups1 is inversely 11 correlated with the overall PA transfer activity of Mdm35/Ups1 (Figs. 2f, 3f and 3i). 12 According to our above molecular model of PA transfer, we could give a rational 13 explanation of this inverse correlation. The success of PA transfer not only needs 14 the binding of Ups1 to the membrane but also needs the successful release of 15 Ups1 from the membrane. The weakened interaction between Ups1 and Mdm35 16 increases the energy barrier of releasing PA-bound Ups1 from the membrane, 17 which is aided by Mdm35. Thus, we could observe more portion of monomeric 18 Ups1 on the membrane. As a result, the overall PA transfer activity is decreased. 19 The alteration of Ups1-membrane binding interface not only decreases the 20 membrane binding ability of Ups1/Mdm35 but also could affect negatively the 21 efficiency of PA entry and binding (to be noted that the PA entry site is coupled 22 with the Ups1 membrane binding site), thus most monomeric Ups1 are trapped on 23 the membrane without PA binding. As a result, the overall PA transfer activity is 24 nearly eliminated. The inverse correlation between the amount of monomeric 25 membrane bound Ups1 and the PA transfer activity could also be observed from 1 the pH dependent experiments (Fig. 5) . 2 The regulation of pH value to the PA transfer activity of Ups1/Mdm35 is very 3 likely one of regulating mechanisms in vivo. In mitochondrion, protons are 4 pumped into IMS, which would lower the pH value of IMS. Since CL is important 5 for the assembly and stability of the respiratory chain complexes, lower pH 6 suggests the respiratory chain functions well and there is enough CL to contribute. 7 Therefore, the PA transfer activity of Ups1/Mdm35 needs to be reduced to 8 decrease the synthesis of CL at IMM. In another case, an increased pH of IMS 9 suggests a potential decreased activity of the respiratory chain, potentially due to 10 the limited amount of CL. As a result, the higher PA transfer activity of 11 Ups1/Mdm35 would be important to balance the synthesis of CL at IMM. 12 Overall, our present structural and biophysical studies give a deep insight to 13 understand the PA transfer mechanism by Ups1/Mdm35, which provides 14 additional knowledge of mitochondrial physiological homeostasis. 15 16 17

Molecular cloning, protein expression and purification 2
To construct the co-expression vectors for N-terminal 6×His tagged full-length 3 Ups1 and Mdm35, the corresponding genes were amplified by PCR from 4 S.cerevisiae genome and cloned into the plasmid pETDute-1 (Novagen). To 5 construct N-terminal 6×His tagged Ups1-Mdm35 fusion protein, the recognition 6 sequence of PreScission protease (LEVLFQGP) was inserted between the 7 C-terminus of Ups1 and the N-terminus of Mdm35 to produce an Ups1-Mdm35 8 fusion gene, which was also inserted into the plasmid pETDuet-1. The constructs 9 containing point mutations were generated by PCR-based sitedirected 10 mutagenesis. All constructs were sequenced to confirm their identities. 11 The target proteins were expressed in E.coli Trans B cells (TransGen Biotech) 12 that were cultured in LB medium. After addition of 0.1 mM 13 isopropyl-D-thiogalactoside (IPTG), the cells were cultured at 30 °C for 5 hr. Cells 14 were collected and re-suspended in buffer A (50 mM Tris-HCl pH8.0, 300 mM 15 NaCl, 20 mM imidazole, and 1 mM PMSF) and then disrupted using the 16 high-pressure method at 120 MPa. Cell lysate was centrifuged with a JA25.5 rotor 17 (Beckman Coulter) at 39,000 g for 40min at 4 °C. The supernatant was loaded 18 onto a column of Ni-chelated Sepharose 6 Fast Flow (GE Healthcare) that was 19 pre-equilibrated with buffer A. The target protein was eluted using buffer B (50 20 mM Tris-HCl pH8.0, 300 mM NaCl, and 150 mM imidazole) and concentrated by 21 ultrafiltration (30 kDa cutoff; Amicon Ultra). The elution was further purified by size 22 exclusion chromatography with Superdex75 (10/300) column (GE Healthcare) in 23 buffer C (20 mM Tris-HCl, pH7.5 and 150 mM NaCl). The peak fractions were 24 collected, concentrated to ~20 mg/mL and stored in -80 °C. for further 25 experiments. For Ups1-Mdm35 fusion protein, the elution was purified with 1 Superdex75 in Buffer D (20 mM HEPES, pH7.0, and 150 mM NaCl). Then the 2 peak fractions were desalted and further purified by ion-exchange 3 chromatography using 1-ml Resource S column (GE Healthcare) in buffer E (20 4 mM mM HEPES, pH7.0) and buffer F (20 mM mM HEPES, pH7.0, and 1M NaCl). 5 Two peaks were obtained using a linear gradient of NaCl from 0 to 1 M in 20 6 column volumes. Fractions corresponding to the first peak were concentrated and 7 stored in -80 °C.for further experiments.

Data collection and structural determination 22
Diffraction data were collected at 100 K on beamline BL17U and BL18U of 23 SSRF (Shanghai Synchrotron Radiation Facility) and processed with HKL2000 24 package 28 . Single wavelength anomalous (SAD) diffraction data of the SeMet 25 substituted Ups1/Mdm35 crystal was collected with a wavelength of 0.9791 Å. 1 Selenium sites were determined using the program SHELXD 29 . Phases were 2 calculated and refined using SOLVE 30 and RESOLVE 31 . An initial model was 3 built using COOT 32 and further refined with PHENIX 33 . Structures of 4 Ups1/Mdm35-DHPA and Ups1-Mdm35 fusion protein were solved by the 5 molecular replacement method and further refined with PHENIX. The quality of 6 the structures was validated with MolProbity 34 . The statistics of data collection 7 and structure refinement are summarized in Supplementary Table 1. 8

PA transfer activity assay 22
PA-transfer activities of the Ups1/Mdm35 complex and its mutants were 23 measured by the fluorescent de-quenching assay as described previously 8,17 . Ups1/Mdm35 and 8 mM liposomes, respectively. The mixture was then 10 ultra-centrifuged at 250,000 g in a S140AT rotor (Doga Limited) on a micro 11 ultracentrifuge (Hitachi Koki himac CS-FNX series) at 25 °C for 30min. Then the 12 supernatants and pellets were subjected to 16.5% Tricine SDS-PAGE analysis. 13 The stained gels were scanned and analyzed using Gel Doc™ EZ Gel 14 Documentation System (Bio-rad). 15

Molecular dynamics simulations and Data analysis 16
Multi-component lipid bilayers were constructed using CHARMM-GUI 35 with a 17 composition of DOPC/DOPE/DOPI/TOCL2/DOPA/DOPG/DOPS (8:6:2:1:1:1:1). 18 Ups1 free and Ups1/Mdm35 complex systems were set up by placing the protein 19 ~8Å above the surface of the membrane. The systems were then solvated in 20 explicit TIP3P 36 water molecules, with potassium and chloride ions in order to 21 achieve a neutral and 0.18 M ionic solvent, using the solvate and auto-ionize tools 22 of VMD 37 . The MD simulations were performed using NAMD 2.11 package 38 and 23 the CHARMM 36 force field 39,40 with CMAP correction 41 . Electrostatic interactions 24 were calculated using the particle mesh Ewald sum method 42 with a cutoff of 12 Å. 25 All hydrogen-containing covalent bonds were constrained by the SHAKE 1 algorithm 43 except that SETTLE algorithm 44 was used for waters, therefore 2 allowing an integration time step of 2 fs. Before production runs, the system was 3 minimized in energy, heated to 310 K, and pre-equilibrated in the canonical 4 ensemble while the protein backbone, and water oxygen atoms harmonically 5 restrained with spring constant of 10 kcal mol -1 Å 2 . Simulations were then 6 continued in the constant NPT ensemble (310K and 1 atm). Langevin thermostats 7 with a damping coefficient of 0.5 ps -1 were used to control the system temperature. 8 A Langevin-piston 45 barostat with a piston period of 2 ps and a damping time of 2 9 ps was used to control the pressure. Constraints were next released step-wise 10 (with spring constant gradually decrease from 10 to 0 kcal mol -1 Å 2 ) before starting 11 the production runs. A total of 600 ns and 1040 ns of data were generated for 12 Ups1/Mdm35 complex and Ups1 free systems respectively. Only data after 13 reaching equilibrium would be taken for further analysis. cloning, protein expression, purification, crystallization, data collection and 8 structure determination. K.C. and J.F. performed molecular dynamics simulations. 9 L.Y. performed mutagenesis, biophysical assays and PA transfer activity assay. 10 Y.Z. analyzed the data. Y.Z., J.F. and F.S. wrote the manuscript. 11
Step 2, the dissociation of Mdm35 from Ups1 leads to the insertion of F69 6 into the membrane, anchoring Ups1 onto membrane.
Step 3, the PA entrance and 7 binding leads to the conformational change of Ups1, in which W65 is inserted into 8 the membrane as a substitute of F69.

Supplementary Text
Kinetic equations of Ups1/Mdm35-mediated PA transport and the derivation of K1 constant from the liposome co-sedimentation assay.