Computational Modeling Explains the Multi Sterol Ligand Specificity of the N-Terminal Domain of Niemann–Pick C1-Like 1 Protein

Abstract

Niemann–Pick C1 like 1 (NPC1L1) is a sterol transporter expressed in the apical membrane of enterocytes and hepatocytes. NPC1L1 resembles the lysosomal NPC1 protein including an N-terminal domain (NTD), which binds a variety of sterols. The molecular mechanisms underlying this multiligand specificity of the NTD of NPC1L1 (NPC1L1–NTD) are not known. On the basis of the crystal structure of NPC1L1–NTD, we have investigated the structural details of protein–sterol interactions using molecular mechanics Poisson Boltzmann surface area calculations here. We found a good agreement between experimental and calculated binding affinities with similar ranking of various sterol ligands. We defined hydrogen bonding of sterol ligands via the 3′-β-hydroxy group inside the binding pose as instrumental in stabilizing the interaction. A leucine residue (LEU213) at the mouth of the binding pocket transiently opens to allow for the access of sterol into the binding pose. Our calculations also predict that NPC1L1–NTD binds polyene sterols, such as dehydroergosterol or cholestatrienol with high affinity, which validates their use in future experiments as close intrinsically fluorescent cholesterol analogs. A free energy decomposition and computational mutation analysis revealed that the binding of various sterols to NPC1L1–NTD depends critically on specific amino acid residues within the binding pocket. Some of these residues were previously detected as being relevant for intestinal cholesterol absorption. We show that clinically known mutations in the NPC1L1–NTD associated with lowered risk of coronary heart disease result in strongly reduced binding energies, providing a molecular explanation for the clinical phenotype.

Introduction

Intestinal cholesterol absorption and reuptake of cholesterol from the bile in hepatocytes depends critically on a particular sterol transporter named Niemann–Pick C1 like 1 (NPC1L1).¹ NPC1L1 was discovered as the target of the cholesterol absorption inhibitor ezetimibe in 2005, even though other targets have been determined in the brush border membrane in later studies.^2,3 NPC1L1 has 42% sequence identity and 51% similarity to the ubiquitous endo-lysosomal transporter NPC1, with which it also shares the overall membrane topology.⁴ In hepatocytes and polarized hepatoma cells, NPC1L1 locates almost exclusively to the apical canalicular membrane, where the protein is thought to mediate reabsorption of cholesterol from the lumen of biliary canaliculi.⁵⁻⁷ Intracellularly, NPC1L1 localizes to the endocytic recycling compartment (ERC) in nonpolarized hepatoma cells and to the subapical compartment—an organelle with similar composition and function as the ERC—in polarized hepatoma cells.^5,6 Both compartments are particularly rich in free cholesterol, which is in continuous exchange with the plasma membrane.^8,9

It has been shown that NPC1L1 contains at least one binding site for cholesterol and related sterols; this binding site is located in its N-terminal domain (NTD). In addition to the NTD, NPC1L1 contains, like many proteins involved in cholesterol homeostasis including NPC1, a sterol-sensing domain in its transmembrane region.^4,10,11 The NTD of NPC1L1 has been purified and crystallized showing close structural and functional similarity to the NTD of NPC1.¹²⁻¹⁴ Both of the proteins’ NTD bind not only the cholesterol but also biosynthetic cholesterol precursors and oxysterols; in addition, the NTD of NPC1L1 has also been implicated in binding and transport of vitamins.^15,16 The molecular basis for the broad ligand binding specificity of the NTD of NPC1L1 is not known. As mutation of NPC1L1 results in lowered plasma LDL levels and risk for developing coronary heart disease;^17,18 a molecular understanding of NPC1L1’s function in cholesterol transport is crucial.

Here, we have carried out a detailed and thorough computational analysis of the multiligand binding specificity of the NTD of NPC1L1. Using the available crystal structures of the NTDs of NPC1L1 and NPC1, we have built structural models of sterol–protein complexes and determined binding affinities using molecular mechanics-Poisson–Boltzmann-surface area (MM-PBSA)-based free energy calculations in direct comparison with the experimental binding data. We show that the promiscuity in sterol binding by NPC1L1 relies at least partly on the same key residues in the proteins NTD. We also predict that intrinsically fluorescent sterols, such as dehydroergosterol (DHE) or the related cholestatrienol (CTL), have comparable affinity to NPC1L1’s NTD as cholesterol. This makes them promising probes in future experiments and validates their use in cellular experiments of NPC1L1-mediated sterol transport.⁵ A free energy decomposition analysis combined with computational mutagenesis sheds light on the role played by different residues in stabilizing the binding of cholesterol to the NTD of NPC1L1, thereby directly linking the molecular interactions to clinically observed phenotypes of impaired intestinal cholesterol absorption.

Materials and Methods

Computational Modeling

Homology Modeling and Structure Refinement

The NTD of NPC1L1 was crystallized in closed conformation without any sterol bound to it (PDB id: 3QNT, resolution 2.83 Å).¹² For binding affinity calculations, an open conformation NTD was built using the Swiss-model.¹⁹ Briefly, the sequence (22–284) of the extracellular NTD of NPC1L1 has been retrieved from the UniProt (ID: Q9UHC9) database²⁰ and used as a query for template-based sequence alignment. The atomic coordinates of the NTD of NPC1 (PDB id: 3GKJ, resolution 1.60 Å) was used as a template because it shares 32% sequence identity with the NTD of NPC1L1, possess 86% sequence coverage, and was crystallized in a complex with bound 25-hydroxycholesterol (25HCL). Models were built based on the target-template alignment using ProMod3 as a part of the Swiss-model (sequence conservation between NPC1L1–NTD and NPC1–NTD is provided in Supporting Information, Figure S1). The conserved structural coordinates between the query and the template were copied from the template to the final model, which includes 25HCL in the binding pocket. The final NPC1L1–NTD structural model was subsequently imported into the Maestro module, available in the Schrödinger package,²¹ and the structure was optimized using the protein preparation wizard.²² This optimization includes adding hydrogen atoms, assigning correct bond orders, and building disulfide bonds. The protonation states of all of the ionizable residues (at pH = 5.0) were predicted by PROPKA²³ provided in the protein preparation wizard. The structure model was finally optimized by energy minimization (i.e., only position of the hydrogen atoms) using the OPLS2005 force field.²⁴

Generation of Ligand Binding Conformation

Three-dimensional (3D) structures of a set of six sterols (cholesterol, lanosterol, 25HCL, β-sitosterol, stigmosterol, and epicholesterol) were prepared, as described before.²⁵ NPC1L1–NTD–sterol binding poses were generated using the molecular docking procedure Glide.²⁶ The ligand-binding site for docking was defined using the receptor grid generation module available in Glide. The centroid of the grid box (of size of 20 Å) was centered at the ligand, and water molecules at the active site beyond 3 Å from the bound ligand were deleted. Specific settings in docking and scoring function parameters used in this study are described in detail elsewhere.^25,27 A proper starting binding pose is crucial for molecular dynamics (MD) simulations and binding affinity prediction.²⁵ Therefore, each ligand was docked, and the best 10 ligand poses were saved for further binding pose analysis (docking scores are summarized in the Supporting Information, Table S1). As the NPC1L1–NTD open-form coordinates were modeled based on the open-form structure of the NTD of NPC1 bound to the ligand 25HCL,¹⁴ we adopted similar sterol binding modes also for NPC1L1’s NTD. For instance, the 3′-hydroxy group (designated here as the sterol head group) attached to the steroid A-ring deeply inserts into the NPC1L1–NTD binding pocket, exactly at the same site because it was found for 25HCL in the X-ray structure of the NPC1–NTD–25HCL complex. The binding poses for all sterols are quite similar to that for 25HCL with one exception; the plant sterol stigmasterol oriented preferentially its D-ring attached “tail” toward the NPC1L1–NTD binding pocket. Therefore, for stigmasterol, constrained docking runs were carried out in which the head group was constrained to be oriented toward the deep NPC1L1–NTD pocket, as observed for other sterols. To verify the correct binding mode of stigmasterol, we used both binding poses from constrained docking and unconstrained docking for the binding affinity calculations. Binding poses for all protein–sterol complexes as used for MD simulation are provided in the Supporting Information (Figure S2).

MD Simulation and Binding Affinity Calculations

All NPC1L1–NTD–sterol complexes were subjected to MD simulations, followed by binding affinity calculations, which were performed and analyzed using the Amber 14 software.²⁸ Prior to MD simulation of complexes, each ligand binding pose (only ligand) was subjected to geometry optimization at the level of HF/6-31G** using the Gaussian 09²⁹ and other computational methodology steps, such as charge calculation and MD simulations used in our previous study.²⁵ Briefly, atomic charges for all sterols were calculated from the electrostatic potential at the B3LYP/cc-pVTZ level of theory, and for both, the calculation of the electrostatic potential and the geometries, an implicit water environment was modeled with the IEF-PCM continuum solvation model.^30,31 These atomic charges were fitted using the RESP procedure being implemented in the Antechamber module of the Amber 14 software.²⁸ Subsequently, the tleap tool in the Amber suite was used to build coordinate and parameter files using the Amber ff14SB force field. TIP3P water molecules were added to solvate the structures with a 10 Å buffering distance between the edges of the truncated octahedron box.³² Energy minimization was carried out; first, using a steepest descent minimization with all heavy atoms restrained for up to 1000 cycles, followed by minimization of the entire system using no positional restraints for 200 cycles. MD simulations were initiated with velocities generated from a Maxwell–Boltzmann distribution at 100 K, and periodic boundary conditions were applied in all directions defined by the octahedron box.³² The system temperature was gradually increased to 300 K at a constant volume over a 200 ps MD simulation period. Thereafter, the systems were equilibrated for another 500 ps in the NPT ensemble with a temperature of 300 K and a pressure of 1 bar using the Berendsen coupling algorithm.³³ After equilibration, MD simulations were run for 20 ns using a time step of 2 fs. During that period, the SHAKE algorithm was used to constrain the lengths of all bonds involving hydrogen atoms.³⁴ For analysis, coordinates were saved every 10 ps from the 20 ns simulation, that is, a total of 2000 snapshots, and used for binding free energy calculations and investigations, such as backbone heavy atoms relative to their initial structure and radius of gyration of ligand (Figures S3 and S4). To understand the overall structural flexibility of NPC1L1–NTD and their binding poses, a ligand-free NPC1L1–NTD (called apo form of NPC1–NTD) was included in the MD simulations. Subsequently, MM-PBSA-based binding free energies were calculated to rank the experimentally observed binding affinity of various sterols to the NTD of NPC1L1. The theoretical basis of the MM-PBSA method used in the study is described in detail elsewhere.³⁵ Briefly, the binding free energy of a protein–ligand complex in an aqueous medium can be estimated based on the Gibbs free energy scheme; however, because of heavy computational cost to estimate solvent–solvent interactions, the binding free energy is calculated using a thermodynamic cycle (eq 1), which relates the Gibbs free energy change upon binding to the free energy difference between the bound and unbound states in aqueous and vacuum environment, respectively.

1

Each of the above contribution terms can be calculated as a sum of three terms, as in eqs 2 and 3.

2

3

where E_MM is the interaction energy of the molecules, a measure of the enthalpic contribution to the Gibbs free energy. E_MM is the sum of the internal energy (E_int) of the molecules (i.e., bonded terms including bond and torsion angles), E_El and E_vdW represent the intermolecular electrostatic and van der Waals interactions energies between the protein and the ligand, respectively. The term G_solv refers to the Gibbs free energy of solvation and consists of both polar and nonpolar solvation energies of the molecule. G_solv energies are estimated based on the Poisson–Boltzmann approximation combined with a solvent-accessible surface area calculation. The term S_MM is the conformational entropy of the complex and is typically estimated based on the harmonic approach calculated with normal-mode analysis at the MM level.³⁶ Because of high computational demands of this analysis and the fact that normal-mode calculations only account for some but not all relevant entropy contributions, we excluded the entropy term from the free energy difference calculations. This is further justified by the fact that we only consider relative binding affinities. The binding free energies, that is, change in Gibbs free energy upon binding (ΔG_bind), were calculated for all NPC1L1–NTD–sterol complexes using the MMPBSA.py script being part of the Amber 14 package.³⁷ The binding free energies of the ligand–protein complex were extracted based on the single trajectory approach, and no separate MD simulations were run for free ligands or the receptor. Gibbs free energy contributions to binding in the decomposition analysis can be negative or positive, meaning that a particular residue either stabilizes or destabilizes the interaction with the ligand.

Computational Mutation Analysis

To better understand the binding free energy difference of cholesterol in various clinically relevant NPC1L1 mutants, additional calculations were carried out. Relevant point mutations (i.e., T61M, I105A, L110M, T128A, N132S, F205A, P215A, and L216A) underwent binding affinity calculations. Each clinically relevant mutant structure was built from the wild-type protein structure model using the “mutate residue” option in the Maestro module in the Schrödinger suite.²² Subsequently, corresponding residues were energy minimized to reduce the atomic clash between neighboring residues using the OPLS-2005 force field,²⁴ and the resulting structures in complex with cholesterol were used as the starting poses for MD simulations followed by MM-PBSA calculation, as described above. Gibbs free energy contributions to binding in the mutation analysis can be negative or positive, meaning that a particular mutation either stabilizes or destabilizes the interaction with the ligand compared to the wild-type complex.

Results and Discussion

Affinity Ranking of Sterol Ligands for NPC1L1–NTD in Experiments and Simulations

Accumulating evidence suggests that NPC1L1 mediates absorption of cholesterol into enterocytes from the intestinal lumen but also into hepatocytes from the biliary compartment in the liver.¹ NPC1L1 also mediates intestinal and hepatobiliary absorption of phytosterols, although to a lower extent than that of cholesterol.^1,38,39 These functions of the protein depend critically on sterol binding to the NTD of NPC1L1 (NPC1L1–NTD). In the crystal structure of the NPC1L1–NTD, the empty binding pocket appeared closed, that is, not open, in contrast to the binding pocket of the NTD of NPC1.¹² Here, we have modeled the sterol-bound conformation of the NTD of NPC1L1 based on that of NPC1 because that has been determined also in complex with two sterols; cholesterol and 25HCL (Figure S5).^12,14 After optimizing the sterol–NTD complex of NPC1L1 (see Materials and Methods section), we have carried out MD simulations and calculated the Gibbs free energy of binding using the MM-PBSA approach. This computational method allows for accurate ranking of various sterol ligands based on their binding free energy, as we recently showed for the Niemann Pick C2 (NPC2) protein.²⁵ Briefly, in MM-PBSA, one carries out MD simulations here for each sterol–NPC1L1–NTD complex in the presence of explicit water molecules using a classical force field. This generates an ensemble of conformations, from which valid statistical thermodynamic averages can be derived. To estimate various contributions to the Gibbs free energy of binding, water is removed and replaced by an implicit solvent description. Binding affinities are estimated by calculating Gibbs free energy terms for the complex, the protein, and the sterol ligand, separately. By performing such calculations for NPC1L1–NTD in complex with various sterol ligands, we were able to match the calculated binding affinity in good agreement to that determined in two experimental studies (Figure 1). In the study of Zhang et al. (2011), a titration assay was carried out at 4 °C for 24 h, in which ³H-cholesterol dissolved in an aqueous buffer containing ethanol and the detergent Nonidet P-40 was found to bind to NPC1L1–NTD in a sigmoidal fashion with an apparent K_D = 0.17 μM and a Hill coefficient of 1.8.⁴ In the experiments by Kwon et al. (2011), a similar binding assay was performed using ³H-cholesterol in a buffer containing small amounts of the detergent NP-40.¹² From the saturation binding curve, an apparent dissociation constant of K_D = 0.012 μM was inferred. At saturation, 0.5 pmol of NPC1L1–NTD bound to 0.48 ± 0.04 pmol of ³H-cholesterol suggests a 1:1 stoichiometry of cholesterol binding to NPC1L1–NTD.¹² Remarkably, Kwon et al. (2011) report a more than 10-fold higher affinity of NPC1L1–NTD to ³H-cholesterol compared to that by Song and co-workers. The reason for the discrepancy in these values is not clear. One possible explanation could lie in different oligomerization states of the NPC1L1–NTD in both studies; whereas Zhang et al. (2011) proposed that tetramerization of NPC1L1–NTD results in cooperative binding of sterols; Kwon et al. (2011) found a monomeric form of the NPC1L1–NTD, which bound sterols in a fashion typical for 1:1 protein–ligand complexes.^14,40 The sigmoid binding behavior observed in experiments by Zhang et al. (2011) could be caused by a conformational transition of an eventual NPC1L1–NTD tetramer from a low affinity to a high affinity state upon ligand binding. However, a sigmoid binding curve could also be an artifact caused by nonspecific binding of ³H-cholesterol to detergent micelles, formation of additional cholesterol aggregates to which NPC1L1–NTD also binds, or both.⁴¹ Finally, the protein solution contained an excess of bovine serum albumin in the experiments by Zhang et al. (2011), which could affect the shape of the binding curve as well as the estimated value of the dissociation constant.⁴⁰ Side-chain oxidized cholesterol derivatives similar to 25HCL efficiently competed for cholesterol binding to NPC1L1–NTD in both studies.^12,40 In our simulations, cholesterol and 25HCL are strong binders, and only one binding site could be identified, suggesting that the apparent cooperativity observed in the experimental studies is not because of multiple binding sites per NPC1L1–NTD molecule. In experiments and our calculations, epicholesterol, stigmasterol, and β-sitosterol bind with lower affinity compared to cholesterol (Figure 1A,B).^12,40 Epicholesterol has the 3′-hydroxy group in the α-configuration, whereas cholesterol and all other sterols used in this study have this group in the β-configuration (Figure 1C). The 3′-hydroxy group in β-configuration forms a hydrogen bond to Ser56 of the NPC1L1–NTD protein, which is absent for epicholesterol (Figure S5). We suggest that this hydrogen bond is an important factor in stabilizing sterol binding to NPC1L1–NTD. In fact, it has been shown that shielded hydrogen bonds inside binding pockets impact ligand residence times.⁴² This is because water molecules must replace the ligand in the hydrogen bond upon ligand dissociation, and water molecules are often hindered in their diffusion into a binding pocket.^42,43 Strong binding of 25HCL and cholesterol but not of epicholesterol is also found for the NTD of NPC1, where Asn41 plays the role of Ser56 in forming a hydrogen bound to the 3′-hydroxy group in β- but not in the α-configuration, emphasizing the similarity of both binding pockets (Figure S5).^13,14 Stigmasterol and β-sitosterol contain a branched propanyl side chain, which causes steric clashes with residues in the upper part of the NPC1L1–NTD binding pose (not shown). Consequently, their binding affinity is reduced compared to that of sterols with unbranched alkyl chain (Figure 1A,B). Lanosterol is highly efficient in competing for the binding of cholesterol to NPC1L1–NTD in experiments,¹⁴ whereas it has a somewhat lower affinity than cholesterol in our MM-PBSA calculations (Figure 1A,B). The reason for this discrepancy is not known at the moment.

Figure 1.

Comparison of experimental and computed binding affinities. The logarithms of experimental relative affinities of sterol ligands to NPC1L1–NTD, as measured by Zhang et al. (A) and by Kwon et al. (B) were plotted against Gibbs free energies of binding calculated using the MM-PBSA approach. Increasing the experimental affinity is well-correlated with more negative binding energies. (C) List of sterols used for NPC1L1 binding affinity predictions. Green, blue, and red indicate the head group, body, and tail of cholesterol. The 3′-hydroxy group is colored red in all sterol ligands.

Gating Mechanism and Flexibility of NPC1L1–NTD upon Ligand Binding

In the crystal structure, the NTD of NPC1L1 was shown to be composed of nine α-helices flanked by three β-strands and containing nine conserved disulfide bonds.¹² This secondary structure is very similar to that of the NTD of NPC1 indicating a similar sterol binding mode and molecular function.¹⁴ Kwon et al. observed a remarkable feature of the NTD of NPC1L1, namely that it is closed in its empty conformation by a lid formed by a few residues around Leu192 (i.e., Leu213 in the original publication¹²).

This is in contrast to the NTD of NPC1, which remains open and accessible to the solvent also after ligand release (see Figure S6 for the comparison of the NTDs of NPC1 and NPC1L1).¹⁴ We confirmed in MD simulations, that the site of sterol entrance in NPC1L1–NTD is formed by Leu213, which can reallocate to open and allow a sterol molecule to enter the binding pose (Figure 2A). An adjacent region formed by residues 189–194 (region 4; Figure 2B) and residues 168–175 (region 3) appeared to be highly flexible with large B-factors. Although Phe205, Gln206, and other residues reside on the exterior site of the binding pose, Ser56 and Thr61 locate to the interior of the binding pocket (Figure 2A). The flexibility of distinct regions of the NPC1L1–NTD was preserved between the open, the cholesterol-bound, and the closed ligand-free form, suggesting that sterol binding does not require significant breathing motion of the protein (Figure 2B).⁴⁴ In contrast, the temporal behavior of molecular fluctuations of the ligand-free NPC1L1–NTD is somehow different in the open and closed state: by fitting a biexponential function to the time evolution of the B-factor during MD simulations, one finds that the NPC1L1–NTD reaches thermodynamic equilibrium in less than 10 ns in the open form, whereas it took significantly longer for the closed conformation, likely exceeding the total simulation time of 20 ns (Figure S7). However, as we did not use the closed and ligand-free conformation any further, this difference was not followed up on for the purpose of this study. Together, we conclude that ligand binding causes major conformational alterations in NPC1L1–NTD similar to the reallocation of the gate centered at Leu213 resulting in an increased structural flexibility of region 4, accompanied by reduced mobility in regions 2 and 3 (Figure 2B).

Figure 2.

Gating mechanism for ligand entry into the NTD of NPC1L1. (A) Illustration of the conformational change when opening the binding pose for sterol ligands. (B) B-factor of protein backbone atoms as a function of residue number for all sterol–NTD complexes from the MD simulations in comparison with the empty closed form (red). Particularly flexible regions (root-mean-square fluctuation, RMSF, of residues) are numbered. The inset is a structural cartoon representing the R-factors according to the X-ray structure with blue: low flexibility, green-yellow: intermediate flexibility, and orange-red: high flexibility. Ligands are 25HCL, cholesterol (CHO), lanosterol (LNS), β-sitosterol (bSTL), stigmosterol (SLT), and epicholesterol (ECL).

NTD of NPC1L1 Binds Oxysterols and Intrinsically Fluorescent Sterols Strongly

Using fluorescent cholesterol analogs in binding assays and cellular transport studies requires minimizing chemical alteration, which is necessary to create a fluorescent cholesterol probe. For example, attaching a fluorophore to cholesterol can significantly impact the binding properties, as shown in a variety of studies reviewed in ref (45). DHE is an intrinsically fluorescent cholesterol analog, which contains only two additional conjugated double bonds in the steroid ring system and another double bond and an extra methyl group in the side chain compared to cholesterol (Figure S8).

CTL is an even closer analogue of cholesterol bearing only two additional double bonds in the ring system compared to cholesterol. DHE and CTL are often used in sterol binding and transfer assays, and with appropriate microscope adaptions also employed as sterol probe in live-cell imaging studies.⁴⁶ We carried out MM-PBSA calculations of NPC1L1–NTD in complex with DHE and CTL and found that both fluorescent sterols bind to NPC1L1–NTD strongly. For DHE, the binding affinity is predicted to be even higher than that for the strongly binding ligands, cholesterol, and 25HCL (Table 1). In addition, we observed that the complexes of NPC1L1–NTD with cholesterol, DHE, and 25HCL share the same overall structure stabilized by the same residues at the interface between binding pose and ligand (Figure 3). In all three cases, the free 3′-hydroxy group is buried inside the binding pocket, forming a hydrogen bond to Ser56. There is enough space in the upper part of the binding pose to account for the extra methyl group of DHE, such that binding of this fluorescent sterol to NPC1L1–NTD is predicted to be strong. To get further insights into the structural determinants of the strong binding of these sterols to NPC1L1–NTD, we decomposed the calculated binding free energy into contributions of individual amino acids (Figure 4). Overall, we find a similar pattern of contributions for cholesterol, 25HCL, and DHE (see correlation plots in Figure S9). Residues in the binding pocket such as T128 have a negative free energy contribution for all sterol ligands, despite varying magnitudes (Figures 4A and S10). Importantly, this residue is conserved in the NTD of NPC1, where the corresponding T112 was found to be important for sterol binding in experiments and simulations.^14,47 However, we also note that certain residues contribute differently to the binding free energy for the three sterol ligands. For example, the contribution of Glu38 is positive for cholesterol (+2.49 ± 0.06 kcal/mol) and 25HCL (+0.93 ± 0.07 kcal/mol), whereas it is negative for DHE (−2.06 ± 0.04 kcal/mol). There are also some differences for cholesterol (−2.11 ± 0.002 kcal/mol) compared to 25HCL (+0.48 ± 0.03 kcal/mol) and DHE (+0.32 ± 0.01 kcal/mol) at Gln95 (Figure 4A). This residue is preserved between NPC1L1 and NPC1 (Figure S5), where mutation of the corresponding Q79 into alanine completely abolishes binding of ³H-labeled 25HCL, whereas it strongly reduces the binding of ³H-cholesterol.¹³

Table 1. Summary of (Binding) Energies Obtained from Final MM-PBSA Model Using Various MD Simulations^a,b.

	Sterol
	cholesterol	lanosterol	25HCL	β-sitosterol	stigmosterol	epicholesterol	DHE	CTL
activity	Jin-Hui Z.	2.00	1.97	1.95	0.70	1.26	1.30
	Kwon H. J.	2.00	1.97	1.96	1.48	1.26	1.30
Energy (kcal/mol)
EvdW	–60.09	–59.36	–59.95	–61.61	–55.71	–60.59	–62.63	–59.49
Eele	–13.40	–9.13	–25.26	–14.60	–17.95	–9.96	–22.09	–14.55
ΔGSol	62.40	60.06	73.70	71.77	68.81	64.54	70.79	64.25
ΔGgas	–73.49	–68.50	–85.29	–76.21	–73.67	–70.55	–84.72	–74.04
ΔGMM-PBSA	–11.09 ± 5.5	–8.44 ± 4.1	–11.82 ± 4.8	–4.44 ± 5.4	–4.86 ± 5.3	–6.01 ± 4.8	–13.93	–9.78

^aAll energy components are extracted from the differences (average) of ΔG_complex – ΔG_receptor – ΔG_ligand. The results refer to averages over 2000 frames, and all units are reported in kcal/mol. Abbreviation: E_vdW = van der Waals energy, E_ele = electrostatic energy, ΔG_gas = sum of van der Waals energy + electrostatic energy + internal energy, and ΔG_solv = solvation energy (polar and nonpolar).

^bConstrained docking pose.

Figure 3.

Comparison of binding modes of cholesterol, DHE, and 25HCL to the NTD of NPC1L1. (A) Bound cholesterol is shown in yellow. (B) Bound DHE is shown in green. (C) Bound 25HCL is shown in violet. All three sterols bind with high similarity being stabilized by a hydrogen bond between Ser56 and the 3′-OH group of each sterol at the bottom of the binding pose. Other residues critical for binding are indicated.

Figure 4.

Decompostion of binding energies for cholesterol, DHE, and 25HCL to the NTD of NPC1L1. Residue-based Gibbs free energy contribution to binding (A) and representative structure of the NPC1L1–NTD with bound cholesterol (B) is shown. Cholesterol is colored pink in B, whereas energy contribution of important residues to binding is indicated in red.

Computational Mutation Analysis Identifies Key Residues in Sterol Binding to NPC1L1–NTD

Song and co-workers (2011) based on a study of Cohen et al. (2006) characterized some point mutations in NPC1L1–NTD, which causes phenotypic alterations including low cholesterol absorption (T61M, N132S) and reduced cholesterol uptake in a cellular sterol uptake assay (T61M, L110F, I105A, T128A, N132S, F205A, P215A, and L216A).^17,48 With the exceptions of Thr61 and Asn132, all of these residues in wild-type NPC1L1–NTD give negative contributions to the binding affinity to cholesterol in our calculations (see Figures 4A and S10). In particular, residues belonging to the ligand facing alpha helix, Ile105 and Thr128, and those on the opposite site of the binding pose in close proximity to cholesterol, Phe205, and Pro215, give strongly negative energy contributions, thus, stabilizing the binding interaction.

We have estimated the difference in Gibbs free energy of binding (ΔΔG) of cholesterol to NPC1L1–NTDs in which these residues have been mutated compared to the wild-type NPC1L1–NTD (Figure 5). Clinically relevant mutations in T61, I105, L110, T128, P215, and L216 reduced the binding energy in accordance with the experimental findings.^40,48 Interestingly, such mutations are often associated with little change in the structure of the binding pose, as exemplified for T61M, I105A, and T128A in Figure S11. This has been similarly observed for mutations in the NTD of NPC1 using alanine mutagenesis and CD spectroscopy.^13,14 Even though the effect of the T61M mutation on the binding affinity is smaller compared to most of the other mutations, it remains to be determined why residues distal from the binding pocket such as T61 can affect the binding affinity. Because the free energy contribution of T61 to the complex of wild-type NPC1L1–NTD with cholesterol is small (Figure S10), a rather subtle effect would be expected, and it is possible that methionine with its more amphipathic character affects the structure locally. In the case of mutation N132S, a slight increase in the binding affinity of cholesterol to NPC1L1–NTD was found, whereas the mutation F205A increased the affinity significantly (Figure 5B). The reason for this discrepancy between experiment and simulations for these two residues is currently unknown. To better understand the binding energetics of the F205A mutant, we carried out an energy decomposition analysis in the presence of the ligand cholesterol and compared that to the decomposition for the wild-type NPC1L1–NTD (Figure 6). There are several residues, such as Val55, Ser56, Glu38, Asn54, Leu99, Tyr156, and Gln206, which have much more negative free energy contributions to the total binding energy in the F205A mutant compared to the wild type, suggesting that they compensate for the loss of F205. In addition, many residues give slightly more negative values in the mutant, suggesting that mutating the bulky Phe205 at the entrance of the binding pose into Ala205 allows for tighter overall fitting of cholesterol in the binding pocket of NPC1L1–NTD. Why does radioactive cholesterol bind less tightly in a biochemical binding assay to the F205A mutant compared to the wild-type NPC1L1–NTD? One explanation could be experimental discrepancies, for example, impaired folding stability of the mutant. Alternatively, one can speculate that Phe205 is important in guiding cholesterol into the binding pocket, that is, to lower the activation energy for binding and thereby increasing the binding rate constant, k_on. This process is not explicitly considered in MM-PBSA calculations, which—after docking and MD simulation—instead only considers the end states of the binding process. Combined experimental and computational analysis of binding kinetics could shed light on the role played by Phe205 in future studies.

Figure 5.

Computational mutation analysis of key residues of the NPC1L1–NTD and their effect on the affinity for cholesterol. (A) Residues whose mutation has been studied are indicated and the position of bound cholesterol is shown in pink. (B) Change in Gibbs free energy of binding upon mutation for the indicated residues. (C) Experimentally determined reduction in cholesterol absorption from the work of Song and colleagues^40,48 is compared to the calculated change in Gibbs free energy of binding upon mutation.

Figure 6.

Key residues with altered Gibbs free energy contribution upon mutation of Phe205 into Ala. (A) NPC1L1–NTD with the F205A mutation in which the indicated residues contribute significantly to the binding energy of cholesterol (negative partial ΔG), whereas the same residues contribute much less in the wild-type protein. (B) Correlation plot of per-residue Gibbs free energy in the wild-type (abscissa) vs the F205A mutant NPC1L1–NTD (ordinate). Linear regression was carried out for all residues (r² = 0.81), except those in red, which have significantly more negative contributions in the mutant. Two residues with strongly altered ΔG are shown in yellow.

Conclusions

NPC1L1 is highly expressed in the apical membrane of enterocytes and hepatocytes, where this protein is involved in cholesterol transport from mixed bile salt micelles across the extracellular glycocalyx and into the apical plasma membrane. A key to this function is the sterol binding capacity of the NPC1L1–NTD. In this study, we have carried out a detailed computational analysis of the binding mechanisms of various sterol ligands to the NPC1L1–NTD. We can confirm the experimentally determined ranking of binding affinities of various sterols and provide structural details of the binding mechanisms. Binding of sterols bearing their 3′-hydroxy group in β-configuration, such as cholesterol or 25HCL, is stabilized by a hydrogen bond at Ser56, which is replaced by Asn41 in the NTD of NPC1 (Figure S6 and ref (14)). We find that some of the key residues mediating sterol binding are conserved between NPC1L1–NTD and NPC1–NTD (e.g., Q95 vs Q79, E30 vs E38, and T112 vs T128 (Figure S6)). Those residues have been implicated in control of the sterol transfer between NPC1–NTD and NPC2.^14,47,49 However, it is important to note that there are also residues in NPC1–NTD whose mutation leads to an NPC disease phenotype and which have no counterparts in NPC1L1–NTD. For example, Q92, which corresponds to A108 in NPC1L1–NTD, is located next to the pocket entrance (Figure S12), where it stabilizes the opening and overall structure of the NPC1–NTD binding pose.⁵⁰ Mutations Q92R and Q92S in NPC1–NTD are known to give a severe clinical phenotype.^50,51 It is likely that the majority of these nonconserved residues leading to NPC disease is involved in docking to or sterol transfer from NPC2 or plays other functional roles apart from sterol binding. This is supported by the observations that distinct subdomains in NPC1–NTD and NPC2 are involved in sterol binding versus sterol transfer.^14,49

It is possible that NPC1L1 needs a sterol binding protein aka NPC2 to take up cholesterol from the lumen of intestine or bile compartment, but so far no such interaction partner has been reported. Cholesterol, and to a lower extent 25HCL and other oxysterols, are hydrophobic molecules, and NPC2 likely binds them in the lysosome to prevent formation of sterol micelles or aggregates.⁵² Because both the digestive and bile juice contain bile salts, cholesterol is likely emulsified by these detergents making sterol binding proteins as a counterpart to NPC2 obsolete. On the other hand, residues in NPC1L1–NTD whose mutation leads to impaired cholesterol uptake into enterocytes have been described. We demonstrate that particular residues whose mutation is associated with this phenotype, such as low intestinal cholesterol absorption and plasma LDL levels (e.g., T61, L110F, and N132), defective cholesterol binding (e.g., L216A), or impaired cholesterol transport function shown for the related NPC1–NTD (i.e., Q95), contribute significantly to the overall binding energy of cholesterol to NPC1L1–NTD. We also show that most clinically observed mutations of NPC1L1 result in reduced binding energies to cholesterol, thereby providing a molecular explanation for the observed phenotypes. There are also few selected mutants which increase the binding affinity in our calculations, and we show in a free energy decomposition analysis that this is because of distant residues whose interaction with the sterol ligand increases, thereby compensating for the mutation. Additional defects in clinical mutants, such as impaired endocytosis or recycling through endosomes compared to wild-type NPC1L1,⁴⁸ seem to be independent of the sterol binding capacity of the NPC1L1–NTD, as also suggested in recent experiments.⁵³ Together, our study provides a molecular picture of sterol binding to NPC1L1 and of alterations in this process associated with clinical phenotypes.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.9b01668.

• Sequence conservation between NPC1L1–NTD and NPC1–NTD; summary of docking simulations; final binding poses used for MD simulations and MM-PBSA calculations; RMSD of protein backbone heavy atoms in NPC1L1–NTD–ligand complexes relative to their initial structure (protein–ligand complex); radius of gyration of ligand atoms relative to their initial structure; illustration of binding mode of 25HCL (PDB ID: 3GKJ) in NPC1–NTD and epicholesterol in the NPC1L1–NTD binding pocket; open and closed conformation of the NTD of NPC1 and NPC1L1; comparison of the open and closed form of the NPC1L1–NTD; structure of fluorescent cholesterol analogs; comparison of binding affinities for various ligands; contribution of residues being mutated in low cholesterol absorbers to the binding free energy in NPC1L1–NTD; position of representative residues whose mutation leads to impaired binding of cholesterol; and location of Ala108 in NPC1L1–NTD and Q92 in NPC1–NTD relative to the binding pocket (PDF)

Author Present Address

^§ Biomedicinskt Centrum (BMC), Department of Chemistry, Uppsala University, Husargatan 3, 75237 Uppsala, Sweden.

We acknowledge the Danish Council for Independent Research for financial support (grant id: DFF-7014-00050B).

The authors declare no competing financial interest.

Notes

Starting structure of protein–ligand complex (NPC1L1–25-OH cholesterol) and trajectory analysis and MM-PBSA data are available at https://figshare.com (10.6084/m9.figshare.9901046). Further information is available from the authors upon request.