High‐content imaging and structure‐based predictions reveal functional differences between Niemann‐Pick C1 variants

The human Niemann‐Pick C1 (NPC1) gene encoding a 1278 amino acid protein is very heterogeneous. While some variants represent benign polymorphisms, NPC disease carriers and patients may possess rare variants, whose functional importance remains unknown. An NPC1 cDNA construct known as NPC1 wild‐type variant (WT‐V), distributed between laboratories and used as a WT control in several studies, also contains changes regarding specific amino acids compared to the NPC1 Genbank reference sequence. To improve the dissection of subtle functional differences, we generated human cells stably expressing NPC1 variants from the AAVS1 safe‐harbor locus on an NPC1‐null background engineered by CRISPR/Cas9 editing. We then employed high‐content imaging with automated image analysis to quantitatively assess LDL‐induced, time‐dependent changes in lysosomal cholesterol content and lipid droplet formation. Our results indicate that the L472P change present in NPC1 WT‐V compromises NPC1 functionality in lysosomal cholesterol export. All‐atom molecular dynamics simulations suggest that the L472P change alters the relative position of the NPC1 middle and the C‐terminal luminal domains, disrupting the recently characterized cholesterol efflux tunnel. These results reveal functional defects in NPC1 WT‐V and highlight the strength of simulations and quantitative imaging upon stable protein expression in elucidating subtle differences in protein function.

responds to alterations in cholesterol levels by adjusting cellular cholesterol synthesis and cholesterol uptake accordingly. 3 Excess cholesterol in the ER is esterified by the ER membrane protein sterol Oacyltransferase (SOAT, a.k.a. ACAT) to generate cholesterol esters that are stored in lipid droplets for later use in lipid poor conditions. 4 Niemann-Pick C1-protein (NPC1) plays a central role in the export of cholesterol from the endo-lysosomal system. 5,6 NPC1 is a glycoprotein with 13 transmembrane segments 7 and three large luminal domains, the N-terminal domain (NTD), the middle luminal domain (MLD), and the Cterminal luminal domain (CTD). 8,9 It locates in endo-lysosomal membranes, orchestrating cholesterol egress together with NPC2, a soluble lysosomal cholesterol-binding protein. According to the prevailing view, the MLD of NPC1 recruits NPC2 to deliver luminal cholesterol to the sterol binding NTD of NPC1. 6 Recent findings suggest the existence of an NPC1 intramolecular tunnel through which cholesterol passes the lysosomal glycocalyx prior to insertion to the lysosomal limiting membrane. 10 Loss of function mutations in the NPC1 gene lead to the build-up of cholesterol in the endo-lysosomal compartment and a progressive neurodegenerative disorder, NPC disease, which leads to premature death. 11 Genetic testing together with staining of patient cells with filipin, a fluorescent cholesterol-binding compound, are used to confirm the disease diagnosis. There are over 300 described NPC disease-causing mutations spanning nearly the whole NPC1 protein. 12,13 A specific mutation may present with different phenotypes indicating that a particular mutation only partially contributes to the clinical disease manifestations. In addition, some genetic variants represent benign polymorphisms but NPC disease carriers and patients may also possess rare variants of unknown significance.

| Stable expression of NPC1 variants
To investigate the function of NPC1 variants, we first disrupted the endogenous NPC1 locus using CRISPR/Cas9-mediated editing in human A431 cells that are well suited for genetic engineering and imaging ( Figure 1A). The generated NPC1-KO cells ( Figure S1A    We focused on the functional differences between three NPC1 variant cell types, hereafter referred to as NPC1-V1, NPC1-V2, and NPC1-WT that were compared to A431 wild-type (WT, non-engineered) and NPC1-KO cells. NPC1-V1 has the aforementioned L472P, M642T and S863P changes in comparison to the Genbank reference sequence (here called NPC1-WT sequence). NPC1-V2 has the 642T and 863P changes but retains 472L ( Figure 1A). All the NPC1 variants were expressed at similar mRNA levels ( Figure 1B). NPC1-V1 and NPC1-WT cells moderately overexpressed NPC1 compared to the endogenous A431 NPC1 protein level, while NPC1-V2 showed higher expression ( Figure 1C). The obvious cholesterol accumulation in NPC1-KO cells compared to A431-WT cells was reduced in the NPC1-overexpressing cells. However, NPC1-V2 and NPC1-WT appeared to revert the NPC1-KO phenotype more efficiently than NPC1-V1 (Figure 2A Instead, NPC1-V1 often displayed a reticular, ER-like pattern ( Figure 2D) and less colocalization with Dextran, suggesting that NPC1-V1 is partly mislocalized ( Figure 2C-E). NPC1-V1 mislocalization seemed to correlate with cholesterol accumulation, as cells with diffuse/reticular NPC1-V1-GFP pattern typically showed a stronger filipin staining than cells with more punctate NPC1-V1-GFP (Figure 2A,C).

| Cell treatments and high-content imaging
To assess the intracellular delivery of LDL-derived cholesterol, we performed lipid manipulations to challenge the cholesterol transport machinery in these cells. Cells were delipidated by a 24-hour treatment with 5% lipoprotein deficient serum (LPDS)-containing media to upregulate LDLreceptor expression ( Figure S2). Thereafter, cells were loaded with 100 μg/mL of LDL for increasing times, fixed, and stained ( Figure 3A).  Figure 4C). However, NPC1-WT and NPC1-V2 expressing cells did not appreciably accumulate cholesterol in the late endosomal organelles upon LDL loading ( Figure 4C). This speaks for the idea that cholesterol egress from late endosomes is enhanced upon overexpression of a functional NPC1 protein.
Together, these findings strongly suggest a defect in the cholesterol transporting capacity of NPC1-V1. This defect was not observed in NPC1-V2 cells, which acted similarly to NPC1-WT cells upon LDL-loading, implying that the L472P change affects NPC1 function.

| The effects of the mutations on NPC1 structure
To investigate the effects of individual mutations on the NPC1 structure, we performed all-atom MD simulations of the membraneembedded full-length NPC1 variants, that is, wild-type (NPC1-WT), L472P, M642T, and S863P (see Figure 6A for a representation of NPC1 structure and the definition of the domains). All-atom MD simulations complement experiments by providing exceptional insight into the dynamics of protein-lipid complexes. 28,29 In the same spirit, we analyzed the effects of the mutations on the local and global conformation of NPC1 in comparison to the NPC1-WT.
In all NPC1 variants, the transmembrane domain of NPC1 was found to be stable and remain close to the initial conformation: the F I G U R E 3 Cell treatments and high-content imaging. (A), Outline of cell treatments prior to imaging. Cells are seeded in 384-well high content imaging plates in 10% FBS-containing complete medium for 2 days, then treated with 5% LPDS for 24 hours, loaded with 100 μg/mL LDL for 0 to 8 hours, fixed with 4% PFA and stained with filipin, LAMP1 antibody, CellMask Green, and TO-PRO-3 to analyze free cholesterol in the late endosomal compartments, or with LD540 and DAPI to analyze the number of lipid droplets.  The M642T and S863P variants are accompanied with local conformational changes ( Figure 6C,D, respectively) and can result in a rotation of the MLDs. On the other hand, the continuity of the cholesterol transport tunnel remains mostly intact in these variants. The M642T mutation also results in the extension of the cytoplasmic loop 2 toward the aqueous phase ( Figure 6C). This is likely because the substitution of Met with Thr increases the hydrophilicity of this loop, which connects transmembrane helices 3 and 4. Figure 6C highlights the extensive contacts between the membrane lipids and the 642M residue. In contrast, 642T was positioned in the water phase and did not interact directly with the lipids. This increases the exposure of the neighboring 643K to the cytoplasmic milieu, possibly making it more prone to ubiquitylation, as previously suggested. 30 As ubiquitylation affects protein trafficking and stability, we speculate that this might be related to the higher protein expression of NPC1-V2 compared to NPC1-WT. On the other hand, the reason for the lower expression of NPC1-V1 that also possesses 642T might be its partial retention in the ER and faster turnover, as previously suggested for the NPC1 mutant I1061T. 31 Our results provide evidence that the dissection of subtle changes in NPC1 function is facilitated by stable, moderate protein expression in combination with high-content imaging and analysis. This is evident for the 472P-variant, as NPC1-V1 is distributed differently and has impaired function when compared to NPC1-V2 and NPC1-WT, both harboring 472L. Indeed, in earlier studies typically based on transient, high protein overexpression and analysis of smaller cell numbers, NPC1-V1 has been considered to behave as a wild-type protein. Of note, a relatively large fraction of NPC1-V1 has been reported to be endoglycosidase H sensitive, 25 in line with a potential defect in its ER exit. In general, based on the above information, using NPC1-V1 as "WT" background dampens the conclusions when dissecting the effects of individual NPC1 sequence variants.
The all-atom simulations suggest that 472P changes NPC1 conformation. The conformational changes in the protein may contribute to the apparent defect in the ER export of NPC1-V1 and affect its postulated cholesterol transport tunnel. Remarkably, the clinical relevance of the L472P-change is supported by a recent report of two Iranian NPC1-patients carrying this mutation. 32 The high-content LDL-cholesterol transport assays based on quantitative monitoring of lysosomal and lipid droplet lipid accumulation described here, may prove useful in several settings. Besides discovering novel functionmodulating variants of NPC1, they may be employed, for instance, to assess potential therapeutics for specific NPC-disease genotypes. [Invitrogen], D22914). Cell culture reagents and general reagents were purchased from GibCo/Thermo Fisher, Lonza, and Sigma-Aldrich. Lipoproteindeficient serum (LPDS) was made from fetal bovine serum (FBS) as previously described. 33 LDL was prepared from pooled plasma of healthy donors by sequential ultracentrifugation as previously described. 33   A431 NPC1 knockout cells were used to generate stable cell pools expressing NPC1-EGFP proteins. Briefly, pSH-NPC1-V1-EGFP, pSH-NPC1-V2-EGFP or pSH-NPC1-WT-EGFP were transfected with plasmid pCas9-sgAAVS1-2 35 (Addgene, #129727) for integration into the AAVS1 safe harbor locus 36 using the X-tremeGENE reagent HP DNA transfection reagent (Sigma), and the cells were grown in culture medium containing puromycin 1 μg/mL until a resistant cell pool was obtained.

| Imaging and image analysis
Cells were cultured on 384-well high content imaging plates (Corning) and imaged with PerkinElmer Opera Phenix automatic spinning-disk confocal microscope using a 63 x water objective, NA 1.15 or with a Nicon Eclipse Ti-E inverted widefield fluorescence microscope using a 40 x air objective, NA 0.75, and 1.5 x zoom. Z-stacks were acquired to span whole cells on filipin, LAMP1, CellMask, TOPRO-3, LD540, and DAPI channels. Widefield image stacks were automatically deconvolved using the Huygens batch processing application (https://svi.nl/ HuygensSoftware). Image stacks were maximum intensity projected by custom MATLAB (http://www.mathworks.com/products/matlab/) scripts. LAMP1 organelles and lipid droplets were detected with CellProfiler. 37 Automatic cell segmentation and automatic image analysis were based on a previously described protocol. 38 Cells were segmented in a hierarchial manner starting with detection of cell nuclei in either TOPRO-3 or DAPI images based on the MoG adaptive and Otsu adaptive thresholding methods, respectively. Touching nuclei were separated by built-in intensity methods. Following nuclear detection, cytoplasm was detected in CellMask Green images or DAPI images using intensity propagation based on the Otsu global and Otsu adaptive thresholding methods, respectively, using the identified nuclei from the first step as a seed point. To detect lysosomes/late endosomes, LAMP1 images were thresholded using Otsu global method and lipid droplets were detected using a previously described CellProfiler module. 39 Integrated filipin intensity in each detected LAMP1-structure was measured and a mean value calculated per cell. All parameters were normalized to the mean of A431-WT 0 h LDL load condition. Lipid droplet counts were normalized to cell area, to exclude the effect of variation in cell size, and to the mean of A431-WT 8 h LDL load condition. Exemplary images' brightness and contrast are adjusted in ImageJ FIJI (http://fiji.sc) similarly in each figure except in Figure 2C (GFP) and 4B (LAMP1).
In Figure 2B, cellular cholesterol accumulation in complete medium was analyzed by quantifying the area of filipin signal after subtracting the cytoplasmic background and manually outlining the cells, leaving the plasma membrane outside the region of interest. In Figure 2D, cells were imaged live with Leica TCS SP8 X inverted confocal microscope using a 63 x water objective, NA 1.20, in an environmental chamber at 37 C, 5% CO 2 . Images were deconvolved using HyVolution automated deconvolution. Cells were manually outlined and background was subtracted on Dextran and GFP channels. Colocalization of NPC1-GFPpositive organelles with Dextran-positive organelles was quantified using Manders colocalization coefficient in Fiji.
After 20 minutes incubation on ice, the lysate was centrifuged to get rid of nuclei. Equal amounts of the postnuclear lysates were loaded onto SDS-PAGE and transferred to nitrocellulose membrane (Protran, Whamann, PerkinElmerTM). The blots were incubated with appropriate antibodies. Goat-anti-mouse-HRP or goat anti-rabbit-HRP was used as secondary antibody. ECL was done according to the manufacturer's protocol (Bio-Rad). Images were acquired using the ChemiDocTM MP Imaging System (Bio-Rad).
Quantitative reverse transcription-PCR was performed using Light

| Model Preparation
The model for the full-length wild type NPC1 was constructed using RosettaCM 40 for cryo-EM density guided comparative modeling. 41 The crystal structure of the NTD (PDBID: 3GKH 6 ) and that of the transmembrane and luminal domains (PDBID: 5 U74 42 ) were used as templates. The initial model for the backbone of transmembrane helix 1 was adopted from the cryo-EM structure (PDBID: 3JD8 8 ). Rigid body docking of the templates into the density 8 was performed using Chimera. 43 RosettaCM 40 was then used for the density-guided combination of the templates and modeling of the missing residues. One thousand models were generated, and the best scoring model that did not contain structural defects was subjected to a density-guided fast relax protocol of Rosetta. The intrinsic pK a of the protein residues were determined using PROPKA 3.1 44 on the selected model. The protonation state of each residue was assigned based on the pK a estimations and visual inspection considering the residue's position (pH~5 for luminal and pH~7 for transmembrane and cytoplasmic residues). VMD 45 psfgen plugin was used for building the all-hydrogen model for the wild type NPC1, and to model post-translational modifications and the point mutations: L472P, M642T and S863P. The topology format conversion to GROMACS was performed using the VMD TopoTools plugin. 46

| Simulation systems
Each of the four variants was independently embedded in a hexagonal bilayer containing 260 POPC and 140 cholesterol molecules with random lateral distribution (35 mol% cholesterol) using in-house scripts.
Each system was then solvated and neutralized with 0.15 M NaCl solution (62 000 water molecules, 167 Na + , and 167 Cl − ions). Next, the systems were independently subjected to a staged minimization and equilibration protocol using GROMACS 2016.x. 47 The heavy atoms of the protein and the head groups of the lipids were initially restrained (total of~7 ns). The integration time step was ramped up from 1 to 2 fs and the restraints were released gradually during this protocol. The systems were heated to 310 K and pressurized at 1 atm using the Berendsen thermostat and barostat. 48 The final equilibrated configurations were used to start the production simulations.

| Simulation Protocols
All simulations were carried out using GROMACS 2016.x. 47 The protein and lipids were described using the Charmm36m 49 and Charmm36 50 force fields, respectively. TIP3P 51 water model and ion parameter set 52 distributed with the forcefield were used. The equations of motion were integrated using a leap-frog algorithm with a 2-fs time step. All bonds between hydrogen and heavy atoms were constrained using the LINCS algorithm. 53  The NPC1-WT system was simulated for 2.5 μs and the mutant NPC1 systems for about 500 to 700 ns. Another shorter repeat (~300-400 ns) was performed for each mutant starting from the final configuration of the NPC1-WT system.

| Analyses
All analyses (RMSD, distances, etc.) were performed using tools distributed with GROMACS 2016 47 and VMD. 45 The analysis of the cholesterol efflux tunnels was performed using the HOLE program. 58 All molecular images were generated using VMD. 45