Cholesterol is an essential structural component of mammalian cell membranes and the precursor of steroid hormones, vitamin D, and bile salts. Whole-body homeostasis of this critical molecule is maintained through a balance of endogenous biosynthesis, intestinal absorption from foods, and net biliary excretion versus reabsorption (1, 2). Perturbations in these systems are widespread in humans and can have major impacts on the risk of disease. As an example, it is well known that elevated plasma levels of total cholesterol and low-density lipoprotein (LDL) cholesterol both show strong statistical correlation with the incidence of cardiovascular diseases (3, 4). The therapeutic interventions that reduce circulating cholesterol levels (5), such as statins inhibiting cholesterol biosynthesis (6) and PCSK9 inhibitors increasing LDL cholesterol clearance (7), provide significant clinical benefit to patients.
Ezetimibe is another drug prescribed for the treatment of hypercholesterolemia and coronary heart disease, which can reduce plasma cholesterol by 15 to 20% as a monotherapy (8, 9). The story of ezetimibe is unusual in that the drug was discovered and approved as a cholesterol-lowering therapy in 2002 based solely on its de facto activity inhibiting intestinal cholesterol absorption, although the molecular target and mechanism of action were unknown (10, 11). Mouse knockout studies have since established the cholesterol absorption protein Niemann-Pick C1-Like 1 (NPC1L1) as the primary facilitator of intestinal sterol uptake and demonstrated that ezetimibe acts through inhibition of this protein (12–14). NPC1L1 knockout mice showed a 70% reduction in net intestinal cholesterol absorption and were insensitive to ezetimibe treatment (13). In vitro studies further revealed that membranes from cells overexpressing NPC1L1 had high affinity for ezetimibe and its analogs (14, 15). More recently, it has been demonstrated that ezetimibe binds directly to purified NPC1L1 protein (16). Efficacy in humans has been confirmed by clinical studies, which showed that addition of ezetimibe treatment to statin therapy led to a statistically relevant reduction of major vascular event risk (17). Human genetic studies led further validation to the role of this target, revealing that heterozygous carriers of NPC1L1 inactivating mutations have a decrease of 12 mg/dl in LDL cholesterol and exhibit 53% reduced risk of cardiovascular diseases (18).
In mammals, NPC1L1 is abundantly expressed and functionally localized on the apical membrane of intestinal brush border enterocytes in the proximal jejunum of the small intestine (13, 19), where it plays a vital role in dietary sterol uptake. In addition, in humans and nonhuman primates, NPC1L1 is found on the canalicular membrane of hepatocytes (20) where it is thought to play a role in the reabsorption and enterohepatic recycling of sterols. The NPC1L1 sequence shares 42% identity and 51% similarity with that of NPC1 (21), a lysosomal and endosomal cholesterol transporter linked to lipid storage disorder Niemann-Pick disease type C1. It is predicted to have a transmembrane domain (TMD) with 13 transmembrane helices (TMs), and three luminal domains composed of an N-terminal domain (NTD), a middle luminal domain (MLD), and a C-terminal luminal domain (CTD). In addition, TM3 to TM7 constitute a sterol-sensing domain (SSD), which is conserved in cholesterol metabolism or signaling-related membrane proteins (22).
Previous studies suggested that cholesterol can bind to both the NTD (23, 24) and the SSD (25, 26) of NPC1L1, but neither of these domains is sufficient for ezetimibe binding. Instead, a loop in the MLD was shown to be important for ezetimibe binding to NPC1L1 (16). The existence of multiple purported sterol binding sites and the lack of overlap with the proposed ezetimibe binding site leave much uncertainty about how the process of cholesterol uptake and ezetimibe inhibition occurs mechanistically. There is considerable evidence to suggest a role for endocytosis in cholesterol absorption (27), but there are conflicting reports on the specific mechanism of how and when this process occurs (28). Recently, Gong et al. (29) reported a 4.4 Å cryo–electron microscopy (cryo-EM) structure of full-length human NPC1, which provides a starting point to explore the architecture of NPC1L1. Winkler et al. (30) determined the crystal structure of Niemann-Pick type C–related protein 1 (NCR1) from yeast and suggested an internal tunnel for sterol transport in NPC proteins. The mechanisms of cholesterol transport by NPC1L1 and ezetimibe inhibition, however, remain unresolved. Here, we present two cryo-EM structures of NPC1L1, one in apo form at 3.7 Å and the other one in complex with an ezetimibe analog, ezetimibe-PS, at 3.5 Å. These two structures provide unique molecular insights into the mechanism whereby NPC1L1 facilitates cholesterol transport at the cell membrane and reveals the highly sought mechanism of action of ezetimibe in blocking cholesterol passage and subsequently reducing the intestinal absorption of cholesterol.
Structure of full-length NPC1L1 in the apo form
The full-length rat NPC1L1 (residues 1 to 1331) with a C-terminal FLAG and His10 tag was expressed in human embryonic kidney (HEK) 293 glucosaminyltransferase I–negative (GnTI−) cells and was purified in detergent digitonin for cryo-EM analysis (fig. S1). We initially obtained a cryo-EM density map at 3.7 Å from 899,111 selected particles, but one of the luminal domains, NTD, was only visible at low contour levels, indicating a local structural mobility of the NTD (fig. S1C). To sort out particles with all three luminal domains visible, we subtracted the signal from the TMD and performed focus classification on the luminal domains following the procedure described by Bai et al. (31). A class that consisted of 144,908 particles and covered all luminal domains was isolated and processed to a final map with overall resolution of 3.7 Å according to the gold-standard Fourier shell correlation 0.143 criterion (fig. S1E). The resulting map showed clear side-chain densities for model building of full-length NPC1L1 (fig. S2), despite local resolution being about 4.5 to 5.5 Å in the NTD (fig. S1D).
The overall architecture of NPC1L1, measuring roughly 150 Å by 60 Å by 40 Å, comprises four distinct domains of NTD, MLD, CTD, and TMD (Fig. 1, A and B) and resembles the structural arrangement of NPC1 (29, 32). While superposition of the whole structure of NPC1L1 and NPC1 yielded a root mean square deviation (RMSD) value of 3.29 Å, the NTD, MLD, CTD, and TMD can be superimposed to the corresponding domains in NPC1 with RMSD values of 1.52, 1.18, 1.22, and 1.49 Å, respectively (fig. S3). The NTD composed of nine helices flanked by a mixed three-strand β sheet connects to TM1 through a long proline-rich segment of 264PVIPPPEALRP274. The central cavity formed by α3, α4, α7, and α8 in the NTD (fig. S3A) was suggested to be a binding pocket for cholesterol (24, 33), but no discernable density for cholesterol was found in the density map (fig. S2). To corroborate that this central cavity in the NTD is important for cholesterol binding in the context of the full-length protein, we performed mutagenesis studies using the full-length NPC1L1 in which two residues, Ile105 and Leu216, in the NTD central cavity were mutated to alanine and then assessed the cholesterol binding. The two variants, Ile105Ala and Leu216Ala, exhibited significantly decreased cholesterol binding by about 80% as compared with wild-type NPC1L1 (fig. S4), which validates that this cavity is a cholesterol binding site in NPC1L1.
Excluding the NTD and TM1, the MLD, CTD, and rest of 12 TMs exhibit an internal twofold pseudosymmetry similar to NPC1 (29, 32), NCR1 (30), PTCH1 (Patched1) (34–37), and prokaryotic resistance-nodulation-cell division (RND) transporters (38–41) (Fig. 1B and fig. S3D). The MLD and CTD, which share a conserved base domain consisting of two helices and three strands, connect to the first TM (TM2 and TM8, respectively) and the second TM (TM3 and TM9, respectively) in each pseudo-repeat (Fig 1B and fig. S3, B and C). TM3 to TM7 constitute the SSD, which forms a V-shaped hydrophobic cavity and is accessible from both luminal domains and lipid bilayer (fig. S3E). Although we observed extra densities around the SSD (fig. S2), these densities could be either phospholipids and/or cholesterols. Because of the uncertainty, we did not assign specific molecules to these densities. We observed a rod-like density at the intersection of the MLD, CTD, and TMD surrounded by the loops and helices connecting the MLD and CTD to the TMD (Fig. 1C). Residues within 4 Å of this rod-like density are mostly hydrophobic in nature, including Pro379, Val380, and Trp383 in the MLD and Pro898, Phe1102, and Tyr1103 in the CTD. It is worth mentioning that similar rod-like densities were also found at a comparable location in the cryo-EM structures of PTCH1 (35, 36) and the crystal structure of NCR1 and were modeled as a cholesterol and an ergosterol molecule, respectively (30). Given the density features and conserved nature of the cholesterol binding site within the NPC1L1 homologous structures, we modeled this rod-like density as a cholesterol molecule.
The MLD and CTD intertwine with each other, making extensive interaction interfaces (Fig. 1D). A more precise analysis showed that MLD-α1 and MLD-α2 interact with the strands of the base domain of the CTD and CTD-α9. On the other side, CTD-α1 and CTD-α2 interact with the strands of the base domain of the MLD and MLD-α12. In addition, the loop connecting MLD-α8 and MLD-α9 [LMLD(α8-α9)], and the loop connecting MLD-α5 and MLD-α6 [LMLD(α5-α6)] interact with the loop connecting CTD-α7 and CTD-β2 [LCTD(α7-β2)]. All these interactions amount to a total buried surface area of 4155 Å2. In contrast, the NTD makes scattered interactions with the MLD and CTD (Fig. 1E), which contribute only a total buried surface area of 719 and 771 Å2, respectively. Since the total buried surface area of most stable protein-protein interaction interfaces is around 1600 (±400) Å2 (42), this suggests that the NTD does not have stable interaction with either the MLD or CTD. The lack of close interactions of the NTD with the other two luminal domains is likely responsible for its high mobility and lower local resolution (fig. S1D).
Ezetimibe binding revealed by the structure of ezetimibe-PS bound NPC1L1
To investigate the binding site of ezetimibe and its mechanism of action, we determined the structure of NPC1L1 in complex with an ezetimibe analog, ezetimibe-PS (Fig. 2A). The cryo-EM reconstruction of the complex was obtained from 252,262 selected particles with overall resolution of 3.5 Å according to the gold-standard Fourier shell correlation 0.143 criterion (fig. S5). The density is well resolved for the entire structure, including the NTD and ezetimibe-PS (fig. S6). We observed two additional rod-like densities in the transmembrane region, one in the V-shaped hydrophobic cavity of the SSD and the other one by TM10 and TM13 (Fig. 2A and fig. S6), which were also observed in the PTCH1 structures (34–37). We modeled a cholesteryl hemisuccinate (CHS) molecule to each density because CHS was used during protein preparation and fits well into these densities.
Ezetimibe-PS binds to a pocket constructed by the three luminal domains, the NTD, MLD, and CTD (Fig. 2A), making mostly van der Waals contacts with surrounding hydrophobic residues (Fig. 2, B and C, and table S2). In particular, the Phe532 (MLD) side chain makes edge-to-face interactions with the 1-phenyl and 4-phenyl groups and Met543 (MLD) engages with the 4-phenyl group of ezetimibe-PS, respectively, validating previous biochemical finding that Phe532 and Met543 are critical for ezetimibe binding (16). In addition to the van der Waals interactions, the ezetimibe-PS makes four direct hydrogen bonds with Thr106 (NTD), Ser187 (NTD), Pro1021 (CTD), and Asn1022 (CTD), respectively. It is well known that ezetimibe undergoes glucuronidation on the 4-hydroxyphenyl group soon after oral administration (fig. S7) (43). Using an in vitro cholesterol uptake assay, we demonstrated that the active metabolite, ezetimibe-glucuronide, has a higher potency in inhibiting cholesterol uptake by NPC1L1 with a median inhibitory concentration (IC50) of 682 nM compared with 3.86 μM for ezetimibe (Fig. 2D). Further modification of ezetimibe-glucuronide by replacing the fluoride on the 1-fluorophenyl group with propargyl sulfonamide results in the most potent form, ezetimibe-PS (fig. S7) (15), which showed an IC50 of 50.2 nM (Fig. 2D). We also used a thermostability assay to assess the impact of the binding of ezetimibe and its analogs to NPC1L1. Ezetimibe-PS significantly enhanced the thermostability of NPC1L1 proteins with the increase of the melting temperature (Tm) by 6.7° to 10.7°. Ezetimibe-glucuronide increased the Tm moderately up to 4.9°, while ezetimibe affected the Tm the least by −2.9° to 2.9°.
Revealed by the structure (Fig. 2, B and C, and fig. S7), the glucuronidation of ezetimibe donates a hydrogen bond to the backbone carbonyl of Pro1021 (CTD), which can contribute to the increase of the potency and the Tm of NPC1L1 by ezetimibe-glucuronide. The replacement of fluoride by propargyl sulfonamide yields not only additional van der Waals interactions with Ile105 (NTD), Met188 (NTD), Phe205 (NTD), and Leu213 (NTD) but also the hydrogen bonds with Thr106 (NTD) and Ser187 (NTD); both of which accounts for the further increase of potency and Tm of NPC1L1 by ezetimibe-PS. Our results provide the direct evidence on the molecular basis for the affinity modulation of the ezetimibe analogs.
Conformational changes of NPC1L1 upon the binding of ezetimibe-PS
Structural comparison between NPC1L1 in the apo form and in complex with ezetimibe-PS using the TMD as an alignment reference reveals remarkable interdomain conformational changes in the luminal domains, especially the displacement of the NTD (Fig. 3A and movie 1). On the contrary, the individual domains between the two structures align well with RMSD values of 0.95, 0.83, 0.66, and 1.28 Å for the NTD, MLD, CTD, and TMD, respectively, and show marginal intradomain rearrangements. The NTD rotates 57.8° around the center of mass and displaces 17.5 Å in the presence of ezetimibe-PS (Fig. 3B), which brings in the loop connecting NTD-α6 and NTD-β3 [LNTD(α6-β3)] to interact with the MLD, and NTD-α3, NTD-α9 and the proline-rich segment to interact with the CTD (fig. S8A). The binding of ezetimibe-PS to NPC1L1 locks the NTD in a closed conformation, which significantly increases the total buried surface area between the NTD and MLD from 719 to 1193 Å2 and the total buried surface area between the NTD and CTD from 771 to 2808 Å2.
Viewing from the luminal side, the MLD rotates 9.0° around the center of mass and displaces 3.2 Å clockwise, while the CTD rotates 10.8° around the center of mass and displaces 3.4 Å clockwise in the presence of ezetimibe-PS (Fig 3C). The rotation and displacement of the MLD and CTD results in dilating the cavity between the MLD and CTD to accommodate the binding of ezetimibe-PS. In contrast to the significant increases of the buried surface area between the NTD and MLD and between the NTD and CTD, the total buried surface area between the MLD and CTD is 4105 Å2 and remains unchanged upon the binding of ezetimibe-PS (fig. S8B). In addition, several TMs move in concert to the structural rearrangements in the luminal domains in the presence of ezetimibe-PS (Fig. 3D). TM1 tilts about 8.1° in response to the marked movement of the NTD and proline-rich segment. All TMs in the second pseudo-repeat, i.e., TM8 to TM13, tilt more than 5° in response to the movements of the MLD and CTD. In summary, all these domain movements again demonstrate the high mobility of the NTD in the apo open state, and the binding of ezetimibe-PS locks the NPC1L1 in an NTD closed state conformation.
Ezetimibe blocks a tunnel connecting NTD to SSD
Inspired by the recent structures of PTCH1, NCR1, and RND transporters, which unveiled internal tunnels for sterol or compound transport (30, 34, 35, 38, 39, 44), we surveyed the two NPC1L1 structures for potential cholesterol binding cavities and cholesterol delivering tunnels. For NPC1L1 in the apo form, a cavity in the NTD and a tunnel extending from the intersection between the upper subdomains of the MLD and CTD to the luminal side of the SSD were found (Fig. 4A). The cavity in the NTD is about 21 Å in length and 3.5 Å in radius and was suggested to be the cholesterol binding pocket according to the apo structure of the NPC1L1 NTD fragment and the cholesterol bound structure of the NPC1 NTD fragment (24, 33). The tunnel is about 63 Å in length with a minimum radius of 3 Å and is mostly hydrophobic in nature, which could allow cholesterol molecules to move down within.
In the ezetimibe-PS bound structure, we found a continuous tunnel spanning from the cholesterol binding cavity in the NTD to the luminal side of the SSD when omitting the ezetimibe-PS in the pocket constructed by the NTD, MLD, and CTD (Fig. 4B). This tunnel is about 90 Å in length with a minimum radius of 3 Å, resulted from joining the cavity and the tunnel found in the NPC1L1 in the apo form. Since this tunnel connects the NTD to the SSD, both of which have a cholesterol binding cavity, it is conceivable that cholesterol is transported from lumen to plasma membrane via this tunnel in NPC1L1. The ezetimibe-PS is found in the middle of the tunnel neither occupying the cholesterol binding cavity in the NTD nor overlapping with the putative cholesterol found in the structure of the apo-form NPC1L1 (Fig. 4C), suggesting that the binding of ezetimibe does not compete with the binding of cholesterol. To further support this observation, we assessed the cholesterol binding with competition of ezetimibe and its analogs. NPC1L1 treated with ezetimibe, ezetimibe-glucuronide, or ezetimibe-PS maintained similar cholesterol binding capability as NPC1L1 in the absence of these compounds, while cholesterol, β-sitosterol, and stigmasterol competed with the isotope-labeled cholesterol to bind NPC1L1 (Fig. 4D). The fact that ezetimibe and its analogs effectively inhibit cholesterol uptake by NPC1L1 but does not compete with the binding of cholesterol supports the hypothesis that cholesterol is transported via this tunnel and that ezetimibe blocks the cholesterol transport by occluding this tunnel.
Mechanism of cholesterol transport by NPC1L1 and ezetimibe inhibition
The structures of NPC1L1 presented here capture two important states within the cholesterol uptake cycle and allow us to propose a mechanism of cholesterol transport by NPC1L1 and ezetimibe’s mode of action (Fig. 5). The structure of the apo-form NPC1L1 represents an open state of NPC1L1, which has the cholesterol binding cavity in the NTD open for cholesterol loading. The structure of the ezetimibe-PS bound NPC1L1 represents a closed state of NPC1L1, which has the complete tunnel from the NTD to the SSD for cholesterol delivering from lumen to plasma membrane. With NPC1L1 in the open state, the cholesterol binding cavity in the NTD is open to micellar cholesterol in the small intestine. The dynamic nature of the NTD provides the opportunity for this domain to scavenge micellar cholesterol in the open state and to connect the NTD cholesterol binding cavity to the cholesterol delivering tunnel reaching to the SSD in the closed state. With the cholesterol bound to the NTD and the structural movements facilitating connection to the tunnel complete, the cholesterol can then make the passage and be delivered to the SSD buried in the plasma membrane. Because the cholesterol binding cavity in the SSD is open to the outer leaflet of the lipid bilayer, cholesterol can diffuse freely and accumulate locally in this leaflet for following cholesterol uptake processes.
While all three luminal domains are involved in ezetimibe-PS recognition, it is worthy to note that the inhibitor binding pocket is mostly contributed by the MLD and CLD (Fig. 2, A to C, and table S2), especially for ezetimibe and ezetimibe-glucuronide. It is likely that ezetimibe binds to this pocket in the open state of NPC1L1 (Fig. 5). The additional interactions from ezetimibe and its analogs to the NTD could further lock the NTD in the closed conformation, which is the state captured by ezetimibe-PS in this study. The binding of ezetimibe and its analogs occludes the movement of cholesterol in the cholesterol delivering tunnel and subsequently prevents cholesterol absorption.
Ezetimibe was initially identified from a high-throughput screening for inhibitors of acyl-CoA:cholesterol acyl transferase (ACAT) in the early 1990s (11). While the compound was only a weak inhibitor of ACAT, it demonstrated potent activity to disrupt intestinal cell cholesterol uptake and garnered the Food and Drug Administration approval in 2002 based on the merits of this functional impact, despite the fact that its true target and mechanism of action were unknown. Since the validation of NPC1L1 as the molecular target of ezetimibe in 2004 (13, 14), important questions on the mechanism of how NPC1L1 functions as a cholesterol transporter and how ezetimibe inhibits this process remain unanswered. Here, we present two high-resolution cryo-EM structures of NPC1L1 that address these long-standing questions at the molecular level. Our structures describe the existence of a continuous tunnel connecting the NTD to the membrane SSD in NPC1L1 as a key element for cholesterol transport and demonstrate that ezetimibe acts through occlusion of this tunnel, which prevents the delivery of cholesterol to the membrane for accumulation but does not block cholesterol binding.
At the point of first contact between NPC1L1 and cholesterol emulsified in bile acid micelles in the lumen of the small intestine, we propose that the NTD may play a scavenger role. Using the NTD fragment alone, Zhang et al. (23) showed that mutations on the NTD cholesterol binding cavity abolished the cholesterol binding. Here, we obtained similar results using the full-length NPC1L1 (fig. S3). Moreover, they showed that either mutations on the NTD cholesterol binding cavity or deletion of the NTD reduced the cholesterol uptake by NPC1L1 markedly and suggested that the cholesterol binding by the NTD is required for NPC1L1-mediated cholesterol uptake (23). Structural comparisons between NPC1, NCR1, and our two NPC1L1 structures show significant NTD conformational heterogeneity and the NTD being the most flexible component in each structure (fig. S9A). Such a flexibility could be advantageous in both increasing the likelihood of substrate engagement and the success of capture. Given that the NTD is indispensable in the process of cholesterol transport by NPC1L1 and is spatially dynamic, it is plausible that this domain functions in the initial recruitment of cholesterol to the NTD cholesterol binding cavity. On the basis of our structures, the NTD cholesterol binding cavity articulates cleanly with the tunnel, which extends into the membrane near the SSD, providing a subsequent path for captured cholesterol to enter the lipid bilayer.
Recently, similar tunnels that are believed to convey substrates have been identified in the related transporters, such as PTCH1, NCR1, and RND family transporters (30, 34, 35, 38, 39, 44). It is worth mentioning that the sterol transport mechanism of NCR1 proposed by Winkler et al. (30) is analogous to the cholesterol transport mechanism to that seen in NPC1L1. The crystal structure that they published of NCR1 outlines a tunnel running from the NTD to the SSD to convey sterols. However, the junction between the NTD cavity and the tunnel formed by the two other luminal domains in the NCR1 structure seems to be narrower and more solvent-exposed (fig. S9B). Compared to our structures, the conformation of the NCR1 structure is more similar to the apo-form NPC1L1. In contrast, the tunnel identified in the ezetimibe-PS bound NPC1L1 structure is closed and continuous at the junction between the NTD and other two luminal domains of the MLD and CLD, establishing a plausible structural basis for the proposed role in cholesterol conveyance. Two important pieces of evidence highlight the key role of the tunnel to the NPC1L1 transport function. First, our structure unambiguously demonstrates that the known cholesterol absorption inhibitor ezetimibe binds to and occludes the central area of the tunnel. Additional evidence comes from mapping human mutations that were identified from human low dietary cholesterol absorbers (45) to the NPC1L1 structure (fig. S10). A significant percentage, 8 of the 19 low cholesterol absorption mutations, are either near the tunnel or in the SSD.
Several studies have shown that the cholesterol uptake mediated by NPC1L1 involves clathrin/adaptor protein 2 (AP2)–dependent endocytosis (27, 46), and a working hypothesis was proposed to address the uptake and delivery of cholesterol to intracellular compartments (27). In this model, NPC1L1 deposits cholesterol to the plasma membrane resulting in cholesterol-enriched microdomains. Subsequently, the NPC1L1 and cholesterol-enriched microdomains are internalized to endocytic recycling compartments mediated by clathrin/AP2. Ezetimibe inhibits the internalization of NPC1L1 and blocks the cholesterol uptake by NPC1L1 (27). The cholesterol transport mechanism mediated by NPC1L1 that we present here is compatible with this working model and substantiates the first part of this process, i.e., the cholesterol harvest and delivery to plasma membrane by NPC1L1. According to the cholesterol transport mechanism that we propose, cholesterol on intestinal micelles is captured by the NTD and delivered to the SSD through the tunnel. After cholesterol is deposited to the SSD, it can freely diffuse to the surrounding outer leaflet of the lipid bilayer, leading to a cholesterol-enriched microdomain. Presumably, the accumulation of cholesterol in the outer leaflet of the lipid bilayer affects the structure and curvatures of the plasma membrane locally, which could serve as a signal to trigger clathrin/AP2-dependent endocytosis. Furthermore, our data provide a key molecular evidence of how ezetimibe inhibits cholesterol transport by blocking the tunnel in NPC1L1, which prevents the passage of cholesterol and consequently its accumulation in the outer leaflet of the lipid bilayer and therefore prevents the internalization of NPC1L1.
Besides cholesterol, NPC1L1 was also reported to be responsible for the transport of plant sterols (12). Although it was reported that β-sitosterol and stigmasterol were not able to compete with cholesterol binding to the NPC1L1 NTD fragment (23), our data shown here establish that when it comes to the full-length protein, β-sitosterol and stigmasterol do competitively block cholesterol binding (Fig. 4D). Additional unpublished observations in our laboratory using radiolabeled plant sterols support the direct binding of these plant sterols to the full-length protein, and analysis of our structures shows no major steric hindrance to the binding of common plant sterols in the NTD cholesterol binding cavity. These points lead us to suggest that NPC1L1 is capable of binding a variety of sterols in the NTD binding cavity, while there might be slight differences in how NPC1L1 recognizes cholesterol versus plant sterols. With this structural work, a much clearer picture for the complete mechanism of NPC1L1-mediated sterol absorption arises. This, along with the demonstration of where ezetimibe acts in the process, might find applications in future drug development effort of more effective therapies for cardiometabolic diseases.
MATERIALS AND METHODS
Expression and purification
The complementary DNA of full-length rat NPC1L1 (NM_001002025.1) was subcloned into a pTT5 expression vector (National Research Council of Canada) with a C-terminal tobacco etch virus (TEV) protease cleavage site followed by a FLAG tag and a 10× His tag. HEK 293S GnTI− cells were grown in suspension in FreeStyle 293 medium (Gibco) at 37°C with 5% CO2. When the cell density reached 2.0 × 106 cells/ml, 0.5 mg of plasmids and 2 mg of polyethylenimines (Polysciences) were used to transiently transfect per liter of HEK 293S GnTI− cells. Eight to 16 hours after transfection, Yeastolate (50× stock from Biological Industries, 166.6 g/liter) and 10 mM sodium butyrate were added to the cells, and the temperature was reduced to 30°C. The cells were harvested after another 48-hour incubation. The cell pellet from 5-liter cell culture was resuspended in 100 ml of lysis buffer [500 mM NaCl, 5% glycerol, 50 mM Hepes (pH 7.5), and one SIGMAFAST protease inhibitor tablet (Sigma-Aldrich)] and lysed with a microfluidizer. The lysate was clarified by centrifugation at 10,000g at 4°C for 15 min, and the membranes were collected by ultracentrifugation at 125,171g at 4°C for 2 hours. The membranes were then solubilized in 100 ml of solubilization buffer [500 mM NaCl, 5% glycerol, 50 mM Hepes (pH 7.5), 2% n-dodecyl-β-d-maltopyranoside (DDM), and 0.2% CHS at 4°C for 2 hours]. Insoluble material was removed by ultracentrifugation at 125,171g at 4°C for 30 min. ANTI-FLAG M2 affinity agarose resin (Sigma-Aldrich) was added to the supernatant and rotated at 4°C for 2 hours. The resin was washed with 10 column volumes of wash buffer [500 mM NaCl, 5% glycerol, 50 mM Hepes (pH 7.5), and 0.05% digitonin]. The bound protein was eluted in wash buffer to which FLAG peptide (100 μg/ml) was added. The eluted protein was concentrated to ~6 mg/ml and subjected to size exclusion chromatography on a Superose 6 10/300 GL increase column (GE Healthcare) equilibrated with the buffer containing 150 mM NaCl, 25 mM Hepes (pH 7.5), and 0.05% digitonin. For the NPC1L1 in complex with ezetimibe-PS, the eluted protein was incubated with 200 μM ezetimibe-PS on ice for 30 min before the size exclusion chromatography. The peak fractions corresponding to NPC1L1 monomer were collected and concentrated to ~6 mg/ml for cryo-EM grid preparation.
Cryo-EM sample preparation and data collection
To prepare cryo-EM grids, 3 μl of NPC1L1 sample (~5 mg/ml) was applied to glow-discharged Quantifoil R1.2/1.3 300 mesh grids. The grids were blotted for 5 s at 4°C and 100% relative humidity and vitrified with a FEI Vitrobot Mark IV. For data collection of NPC1L1 in the apo form, images were recorded with a Gatan K3 camera on a Titan Krios microscope operated at 300 kV. Superresolution movie frames (0.06 s per frame with 1.8-s total exposure) were recorded at a nominal magnification of 105,000× (corresponding to a physical pixel size of 0.862 Å), with a total electron dose of 46.76 electrons/Å2. For data collection of NPC1L1 in complex with ezetimibe-PS, images were recorded with a Gatan K2 camera on a Titan Krios microscope operated at 300 kV. Superresolution movie frames (0.2 s per frame with 7.2-s total exposure) were recorded at a nominal magnification of ×130,000 (corresponding to a physical pixel size of 1.059 Å), with a total electron dose of 48.61 electrons/Å2.
Motion correction was performed using the MotionCor2 program (47), and the contrast transfer function (CTF) parameters of the micrographs were estimated using the Gctf program (48). All subsequent image processing was performed using RELION-3.0 (49). Initially, about 1000 particles were manually picked from a few micrographs. Class averages representing projections of NPC1L1 in different orientations were selected from the two-dimensional (2D) classification of the manually picked particles and used as templates for automated particle picking. For the dataset of NPC1L1 in the apo form, 2,941,713 particles were automatically picked from 11,531 micrographs and subjected to two rounds of 2D classification. A total of 1,447,590 particles were selected for two rounds of 3D classification using the initial model generated by RELION as the reference. The best class, containing 899,111 particles, provided a 3.7-Å map after 3D refinement, but the NTD was only visible at low contour levels, indicating a local structural heterogeneity in the NTD. Therefore, we performed a focused 3D classification with density subtraction (31) to improve the density of the NTD. In this approach, the densities corresponding to the TMD and the detergent micelle were subtracted from the original particles. The subsequent 3D classification on the subtracted particles was carried out by applying a mask around the luminal domains and having all the orientations fixed at the value determined in the initial 3D refinement. After this round of classification, one class, containing 144,908 particles, with better density in the NTD was selected for per-particle CTF refinement, followed by 3D refinement and postprocessing with a soft mask around the full density. The resulting 3D reconstruction was estimated to be 3.7 Å using gold-standard Fourier shell correlation 0.143 criterion. For the dataset of NPC1L1 in complex with ezetimibe-PS, 1,997,040 particles were automatically picked from 9698 micrographs and subjected to one round of 2D classification. A total of 1,200,019 particles were selected for two rounds of 3D classification using the initial model generated by RELION as the reference. A total of 252,262 particles from the best class of the second round of 3D classification were selected for per-particle CTF refinement and Bayesian polishing, followed by 3D refinement and postprocessing with a soft mask around the full density. The resulting 3D reconstruction was estimated to be 3.5 Å using gold-standard Fourier shell correlation 0.143 criterion. ResMap (50) was used to estimate the local resolution.
Model building, refinement, and validation
Because NPC1L1 has 42% identity to NPC1, the crystal structure of NPC1 [Protein Data Bank (PDB) code 5U74], which contains MLD, CTD, and TMD, was used to make a model of these three domains of NPC1L1 using phenix.sculptor program in PHENIX (51). The crystal structure of the NTD of NPC1L1 (PDB code 3QNT) and the model containing MLD, CTD, and TMD were docked into the density map as rigid bodies using phenix.dock_in_map program in PHENIX, followed by Rosetta FastRelax protocol (52) to morph the domains to the density map. The fitted model was improved automatically by Rosetta iterative local rebuilding protocol (53) and then manually adjusted in Coot (54). Sequence assignment was guided and validated by bulky side residues such as Phe, Tyr, Trp, and Arg and predicted glycosylation sites. One or two N-acetylglucosamine moieties were built to each site on the basis of the densities. A cholesterol molecule was assigned to the density lobe surrounded by the MLD, CTD, and TMD. For the NPC1L1 in complex with ezetimibe-PS, two CHS molecules were assigned in the two density lobes surrounding the TMD. The ezetimibe-PS binding site was identified, and the ezetimibe-PS was modeled using AFITT (OpenEye Scientific Software). These final atomic models were refined against the corresponding density maps using phenix.real_space_refine program in PHENIX with secondary structure and geometry restraints. The final atomic models were evaluated using MolProbity (55) and EMringer (56). The internal cavities and tunnels were found using MOLE 2.0 (57). Structural figures were made in PyMOL (Schrodinger, LLC) or UCSF Chimera (58). PISA was used to calculate buried surface areas (59).
In vitro 3H-cholesterol uptake assay
Madin-Darby Canine Kidney II (MDCKII) cells stably expressing human NPC1L1 (hNPC1L1/MDCKII) were maintained in Dulbecco’s modified Eagle medium (DMEM) (Gibco) containing 1% penicillin/streptomycin (Gibco), 2 mM l-glutamine (Gibco), and blasticidin (5 μg/ml; Invitrogen) supplemented with 5% fetal bovine serum (Gibco). hNPC1L1/MDCKII cells were seeded in 384-well culture plates at a density of 5 × 103 cells per well for a cholesterol uptake assay. The cells were grown at 37°C overnight in a humidified 5% CO2 incubator. Next day, the cell culture medium was completely removed and replaced with DMEM containing test compounds [ezetimibe (Eze), ezetimibe-glucuronide (Eze-Glu), and ezetimibe-PS (Eze-PS); American Radiolabeled Chemicals]. Cells were incubated with compounds at 37°C for 1 hour in a 5% CO2 incubator, followed by addition of 5 nM [1,2-3H(N)]-cholesterol (PerkinElmer) complexed to 10 mM taurocholate (Sigma-Aldrich) in DMEM. After a 90-min incubation for cholesterol uptake, the cells were washed once with phosphate-buffered saline and lysed with MicroScint-20 (PerkinElmer) at room temperature for 1 hour. Radioactivity in the lysate was determined by liquid scintillation counting (TopCount NXT HTS, PerkinElmer).
Differential scanning fluorimetry
Differential scanning fluorimetry (DSF) was performed as described previously (60). 7-Diethylamino-3-(4’-Maleimidylphenyl)-4-Methylcoumarin (CPM) dye from Invitrogen was dissolved in dimethyl sulfoxide (DMSO) (Sigma-Aldrich) at 4 mg/ml as the dye stock. The dye stock was diluted 1:40 in the reaction buffer [0.5 M NaCl, 0.05 M Hepes (pH 7.5), 5% glycerol, 0.05% DDM, and 0.005% CHS] before use. To make reaction mixture, 20 μl of NPC1L1 protein (0.1 mg/ml), 2 μl of the diluted dye, and 8 μl of reaction buffer containing 10 μM ezetimibe or its analogs were mixed well in a polymerase chain reaction tube. DSF measurements were performed using a Rotor-Gene Q (Qiagen). Samples were equilibrated at 25°C for 90 s before ramping from 25° to 99°C at 4 s/°C. The apparent Tm, corresponding to the inflection point of the curve, was derived from analysis using the Rotor-Gene Q software.
Cholesterol binding assay
Radiolabeled [1,2-3H(N)]-cholesterol stock in 100% ethanol (PerkinElmer) for each binding reaction (0.5 μM) was aliquoted to glass test tubes. Unlabeled sterol samples in 100% ethanol, used to test competitive inhibition, were then added (20 μM). The solvent was evaporated under a stream of nitrogen gas. Sterols were resolubilized in 120 μl of 25 mM tris (pH 7.4) with 150 mM NaCl and placed in a sonicating bath for 10 min. Ezetimibe series compounds in DMSO were added (20 μM) followed by NPC1L1 protein (0.62 μM). Samples were incubated at 23°C overnight with gentle agitation to allow binding to occur. NPC1L1 protein was then isolated by fast protein liquid chromatography using 100 μl of binding sample fractionated on a Superdex 200 5/150 GL column (GE Healthcare). The column was pulsed with 50 mM taurocholic acid to remove free tracer and reequilibrated after each run. NPC1L1 peak fractions were determined by ultraviolet monitoring, and bound tracer was quantified by liquid scintillation counting of 10-μl fraction with 120 μl of MicroScint-20 (PerkinElmer) and reading the plates on a TopCount NXT HTS (PerkinElmer).
Data from in vitro cholesterol uptake assay, DSF, and cholesterol binding assay were analyzed using GraphPad Prism 7.04 (GraphPad Software Inc., San Diego, CA). The n of each specific experiment is listed in the relevant figure legend. For the in vitro cholesterol uptake assay, a nonlinear regression of a dose-response curve was fitted to the data according to the variable slope model
Here, Top and Bottom are plateaus in the units of the Y axis. IC50 is the concentration of agonist that gives a response half way between Bottom and Top. Hillslope describes the steepness of the family of curves. For the DSF and cholesterol binding assay, data are presented as means with error bars indicating SE. The number of replicates is reported in figure legends.
Acknowledgments: We thank C. Xu and K. Song at the UMass Cryo-EM Facility for help with electron microscopy data collection. We thank F. Garces and S. Jackson for critical discussions and helpful comments. Funding: This work is supported by the NIH (P41GM103832 to W.C.) and the NRF (2019R1C1C004598 and 2019R1A6A7076042 for S.-H.R.). Author contributions: M.Z., X.M., and Z.W. conceived the project. C.-S.H. prepared EM samples, analyzed cryo-EM data, and built the models. X.Y. prepared cryo-EM grids and collected and analyzed cryo-EM data. B.C.C. performed molecular cloning, initial protein expression, and thermostability experiments. K.C. performed cholesterol uptake assay. P.F. performed cholesterol binding assay. S.-H.R. collected and analyzed cryo-EM data. W.C. provided guidance to collect and analyze cryo-EM data. C.-S.H., P.F., X.M., and Z.W. wrote the manuscript with input from all authors. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The 3D cryo-EM density maps have been deposited in the Electron Microscopy Data Bank under accession numbers EMD-21035 and EMD-21037. Atomic coordinates for the atomic models have been deposited in the PDB under accession numbers 6V3F and 6V3H. All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.