A Membrane-Assisted Mechanism for the Release of Ceramide from the CERT START Domain

Ceramide transfer protein CERT is the mediator of nonvesicular transfer of ceramide from the ER to Golgi. In CERT, START is the domain responsible for the binding and transport of ceramide. A wealth of structural data has revealed a helix-grip fold surrounding a large hydrophobic cavity holding the ceramide. Yet, little is known about the mechanisms by which START releases the ceramide through the polar region and into the packed environment of cellular membranes. As such events do not lend themselves easily to experimental investigations, we used multiple unbiased microsecond-long molecular simulations. We propose a membrane-assisted mechanism in which the membrane acts as an allosteric effector initiating the release of ceramide and where the passage of the ceramide acyl chains is facilitated by the intercalation of a single phosphatidylcholine lipid in the cavity, practically greasing the ceramide way out. We verify using free energy calculation and experimental lipidomics data that CERT forms stable complexes with phosphatidylcholine lipids, in addition to ceramide, thus providing validation for the proposed mechanism.

System preparation.START apo and START-Cer.The structure of the START-Cer complex was retrieved from the protein data bank (PDB ID: 2e3q) and processed with MMTSB 1 tools and CHARMM-GUI.The CHARMM simulation package 2 (v47a2) was then used with the CHARMM36 force field for proteins and lipids to generate the topology (psf) and coordinates (crd) files for the complex.The complex was energy-minimized (steepest descents: 100 steps, adopted basis Newton-Raphson: 1000 steps), solvated in a cubic box (a=b=c= 86.1 Å) with TIP3P water molecules (edge cutoff=10 Å) and neutralized by adding 3 counter K + ions, replacing 3 randomly chosen water molecules from the bulk.This was followed by energy-minimization (steepest descent: 200 steps, adopted basis Newton-Raphson: 1000 steps).Particle mesh Ewald (PME) was used for longrange electrostatic interactions and periodic boundary conditions with a non-bonded cutoff of 16 Å along with truncation (vswitch) of van der Waals interactions at 10 Å.The system was then gradually heated from 190 K to 310 K with increments of 1 K every 100 steps using a gaussian distribution for velocity assignments.SHAKE 3 was used to constrain bonds between heavy atoms and hydrogen atoms.A short 5 ns equilibration and then 400 ns production simulations were run in the NPT ensemble with a 2-fs integration time step and using BLaDE 4 in CHARMM 47a2.The temperature was set to 310 K and controlled using a Langevin thermostat (drag coefficient = 0.1 ps -1 ).A Monte Carlo barostat was used (target pressure=1 atm) with pressure coupling moves every 25 steps and the default (100 Å 3 ) standard deviation of the gaussian distribution was used for drawing volume changes.The trajectory obtained was recentered around the START domain using CHARMM v.47a2 and clustered, using ttclust 5 and the RMSD of the domain backbone atoms.A structure from the largest cluster was selected for free energy calculations.Cer was removed from that structure to generate the structure of apo START (except for the N504 mutations for which the cluster closer to the X-ray structure was used).

START-Cer-POPC and START-POPC.
The structure of the START-Cer-POPC complex was taken from the holo-Neu_r1 simulation.The conformation was chosen such that the Cer and POPC tails were present and before Cer leaves the binding site (t=886.9ns).The START-POPC structure was obtained from the above by removing Cer.

2.
Multisite lambda dynamics (MSD) The apo START structure and the START-Cer complex structure extracted from the cluster above were solvated (cutoff 15 Å) in a cubic box with TIP3 water molecules and neutralized by adding three K+ ions using the MMTSB toolset.These systems are denoted as START-Water and START-Cer-Water respectively in Table S5.The structures extracted from the holo-Neu r1 simulation were also similarly solvated and neutralized and are referred as START-Cer-POPC-Water and START-POPC-water in Table S6.The mutant atoms on the perturbation site were added using the PATCH command in the CHARMM program 6 .The perturbation site had all the atoms of the wild-type residue and all the atoms for the mutant (including backbone). 7All valence angles, dihedrals and impropers were removed between the alchemical groups.The alchemical system containing two substituents on a single site was set up using the BLOCK module 7 of CHARMM and atoms of both substituents were constrained using the SCAT (force constant = 118.4)facility.A FNEX value of 5.5 was used for the functional form of .Soft core potentials were used along with the ALF 8 biases for each of the substituents.The POPC tail outside the START cavity was harmonically restrained with a force constant of 2 kcal*mol -1 *Å -2 on heavy atoms.The START-Water and START-Cer-Water systems were simulated using BLaDE with  values saved every 20 fs and a non-bonded cutoff of 12 Å along with truncation (vswitch) of van der Waals interactions between 9-10 Å while keeping the remaining conditions and parameters same as for the equilibrium simulations described earlier in this section (Cf System preparation).The START-Cer-POPC-Water and START-POPC-Water systems were simulated with DOMDEC 9 since BLaDE does not support harmonic restraints.In these simulations the Nosé-Hoover thermostat (temp = 310 K; tmass = 1000 kcal*ps 2 ) was used for temperature control and a Langevin piston (friction coefficient: 20 ps -1 ) for pressure control.All the simulations either with BLaDE or DOMDEC were performed with CHARMM 47a2.MSD simulations were performed in an iterative fashion. 10,11The first phase consisted of 50-60 small 100 ps long simulations to get an idea regarding the biases and, the second phase consisted of 10-20 longer 1 ns simulation to flatten the biases.The ALF biases were calculated using weighted histogram analysis method (WHAM) 12 and optimized after every run.The production runs were started with the lambda landscape flattened in 5 independent replicas of 5 ns each.The final relative free energy differences are obtained via the following equation: Where, i and j are the wild-type and mutant residues respectively.P -probability  -bias free energy

S2. Replicate MD Simulations
We conducted two replicates of the simulations for each START-lipid bilayers systems (see Table S1); the results of the first replicates are presented in the main text.In this section, we report our observations from the second replicates.Our findings suggest that only POPC can insert a tail into the cavity, positioning its head group between the Ω1 and Ω4 loops.

Apo-ER_r2.
A POPE lipid occupied the site between Ω1 and Ω4, and the cavity remained closed throughout the simulation.No gate opening was observed during the simulation.
Apo-neutral_r1.a POPC lipid occupied the site between Ω1 and Ω4 snorkeling to insert a tail in the open cavity.Simultaneously another POPC lipid blocked the entrance into the cavity by placing its headgroup near the cavity.No tail insertion was observed during the simulation.
Holo_neutral_r2 / holo-Golgi_r2.we also observed the opening of the cavity and lipid snorkeling events.In both simulations, a POPC occupied the site between Ω1 and Ω4 loops and snorkels around the entrance to the cavity, but neither tail insertion nor ceramide release were observed.

Fig. S1 .
Fig. S1.Alternative starting orientations of START with respect to bilayers.(A) Numbers under the arrows report the initial minimum distance between START and bilayers.All simulations (including the replicates) were run for 250 ns and lead to the START domain anchored at the bilayer through the C-terminus α4 helix, the Ω1, Ω2 and Ω4 loops.The Orientation_0 corresponds to the starting orientation used in the main text (See also Fig 2 in main text).(B-D) Minimum distance between START and bilayers through the simulations.(E-G) Depth of insertion of each START amino acids (Cα of glycine and Cβ of others) into the bilayers for both replicas (r1 and r2) in each case, calculated using the last 50 ns of each simulation.

Fig. S2 .
Fig. S2.RMSD plots.RMSD for both replicas (r1 and r2) in each case, calculated for START domain excluding the N terminus helix, resid 362-391.(A) Apo in the ER-like bilayer (B) Holo in the Golgi-like bilayer (C) Apo in the neutral bilayer (D) Holo in the neutral bilayer.

Fig. S3 .
Fig. S3.Tilt angle of apo and holo forms on the different bilayers.The tilt is defined using the START domain α4 vector with respect to the bilayer normal.

Fig. S4 .
Fig. S4.START domain and bilayer distance.Minimum distance between the START domain and the bilayer along the 2 µs simulations: (A) apo and ER bilayer, (B) holo and Golgi bilayer, (C) apo and neutral bilayer and (D) holo and neutral.

Fig. S5 .
Fig. S5.Gate opening through displacements of Ω1, Ω4 and α4.(A, D, and G) Time series of distances between W473 (Ω1) and P564 (Ω4) (plain lines), and S476 (Ω1) and W564 (Ω4) (dotted lines) in simulations of START with the ER (A), Golgi (A), and neutral (D, G) bilayers.(B, C, E, F, H, and I) Estimated probability density (KDE function) of the closed and open states of START calculated from the simulation trajectories with the ER bilayer (B), the Golgi bilayer (C), the neutral bilayer (Neu) (E, F, H, and I).The KDE (Kernel density estimation) represents the data using a continuous probability density curve.The bar at the right side shows the intensity of data values along the KDE curve.
Fig. S6.Hydrophobic contacts and hydrogen bonds.Time series of hydrogen bonds and hydrophobic contacts between W562 and its nearest neighbors in the Ω1 and Ω4 loops.(A-H) Hydrophobic contacts (I) Hydrogen bonds.The plots display a bar if there is at least one hydrophobic contact (A-H) or hydrogen bond (I).

Fig. S7 .
Fig. S7.Backbone RMSD for apo START in water in three replicate simulations.The N-terminus helix (residues number 362-391) is not included.

Fig. S8 .Fig
Fig. S8.Gate opening through displacements of Ω1, Ω4 and α4.(A-C) Time series of the W473(Ω1)-P564(Ω4) in blue and S476(Ω1)-W562(Ω4) in red distances in each of the three replicate simulations of apo START in water.(D-F) Density distributions (KDE function) of the distances plotted in A-C and characterizing the closed and open states of START.The KDE (Kernel density estimation) represents the data using a continuous probability density curve.The bar at the right side shows the intensity of data values along the KDE curve.

Fig. S10 .
Fig. S10.Lipid snorkeling within 3Å of the START domain in the ER and neutral simulations.Z value of the C218 of the POPX (X: C, E, I, and S) and the C18S atoms of the ceramide with respect to the average plane of the phosphorus atoms (Z=0) (A) apo+ER_r1 (B) holo+Neu_r1.Z value of the C316 of the POPX (X: C, E, I, and S) and the C18F atoms of the ceramide with respect to the average plane of the phosphorus atoms (Z=0) (C) apo+ER_r1 (D) holo+Neu_r1.

Fig. S11 .
Fig. S11.Average number of snorkeling events.Number of times the C218 and C316 atoms of the POPX (X: C, E, I, and S) and the C18S and C18F atoms of the ceramide are at least 5 Å higher than the average plane of the phosphorus atoms.The numbers are average per frame in windows of 100 ns each.Data shown for lipids from (A) the upper leaflet (to which the START domain is bound) and (B) the lower leaflet in the apo+ER_r1 simulation; and for lipids in (C) the upper leaflet (with the START domain) and (D) the lower leaflet in the holo+Neu_r1 simulation.The lower leaflets are shown for reference.

Fig. S12 .
Fig. S12.Hydrophobic contacts between bound ceramide and lipid bilayer.Number of hydrophobic contacts between the START-bound ceramide and the lipids of the bilayer in the holo+Golgi and Holo+Neu simulations.Color bar indicates the number of hydrophobic contacts in each frame; Blue: 1-5 hydrophobic contacts, red: 6 or more hydrophobic contacts.

Fig
Fig S13.POPC (1-palmitoyl-2-oleoyl-phosphatidylcholine) rearrangement induced by START domain in opened state detected by MD simulation.(A) Closeup view of the binding conformation of POPC within apo START domain, and the amino acids in the cavity involved in hydrophobic contacts with the POPC tail in the neutral bilayer.(B) Distribution of the POPC tail angle with respect to membrane normal in the neutral bilayer and their estimated probability density (KDE function) are presented.(C) Close-up view of the binding conformation of POPC within START-Cer complex, and the amino acids in the cavity involved in hydrophobic contacts with the POPC tail in the Golgi bilayer.(D) Distribution of the POPC tail angle with respect to membrane normal in the Golgi bilayer and their estimated probability density (KDE function) are presented.The KDE (Kernel density estimation) represents the data using a continuous probability density curve.The bar at the right side shows the intensity of data values along the KDE curve.The times at which the POPC tail inserts into and exits the cavity are given below the snapshots.Ceramide colored yellow, POPC colored purple, Ω1 colored orange and Ω4 colored cyan.START domain is shown as cartoon and colored grey.Membrane is shown as grey transparent surface.The gray dots represent all the sampled tilt angles.

Fig. S14 .
Fig.S14.The POPC lipid located between Ω1 and Ω4.The POPC head group is locked in place by a salt bridge between its phosphate group and R478, and by hydrogen bonds between the S476 side chain and the POPC phosphate group.

Fig. S16 .
Fig. S16.All hydrophobic contacts in the holo+Neu_r1 simulation.Time series of contacts between (A) POPC143 and the START domain, between (B) ceramide and the START domain.

Table S1 . Composition and size of the simulated systems
. "r" stands for replicate.See Material and Methods for detailed simulation protocols.

Table S2 . Candidate atoms for the hydrophobic interaction analysis.
The atom names correspond to the nomenclature of the CHARMM36 force field.

Table S3 : Inventory of interactions between START and the bilayer lipids.
The data presented in the table are averages over two replicas in each case, calculated from START binding to the bilayers to the end of the simulations.SSE: Secondary structure elements.AA: amino acid, α: helix, Ω: loop, β: beta-strand, occ: occupancy.

Table S4 . Occupancies (%) of lipid-tryptophane (W473, W562) hydrogen bonds.
Hydrogen bonds between the two exposed tryptophanes and the lipid bilayers, reported in percentage of simulation time.

Table S5 . Occupancies (%) of Cer-START hydrogen bonds.
Inventory of hydrogen bonds between ceramide and residues of the START domain cavity in simulations of the holo START domain on Golgi-like and neutral bilayers.Numbers are occupancies reported in percentage of simulation time.

Table S6 . MS𝜆D START-Cer relative binding free energy (ΔΔG) for the Y553F, N504A and V480A substitutions.
The production simulations were run in five replicates (40 ns each).

Table S7 . MS𝜆D START-Cer relative binding free energies for the V480A substitution in the presence of a POPC tail in the binding pocket.
The production simulations were run in 5 replicates (30 ns each).

Table S8 : Phosphatidylcholine (PC) species observed with CERT and other known PC-binding lipid transfer proteins by LC- MS/MS.
The different columns describe in order: the lipid transfer protein of interest; the PC species, the adduct as which the PC was observed, and the full identity of the PC with possible extra information on the fatty acyl composition; the MS intensity by which the PC was observed; its total carbon chain length and unsaturation; the MS ion mode in which the PC was observed and at what m/z and LC retention time range.