Determination of the membrane topology of PORCN, an O-acyl transferase that modifies Wnt signalling proteins

Wnt gradients elicit distinct cellular responses, such as proliferation, specification, differentiation and survival in a dose-dependent manner. Porcupine (PORCN), a membrane-bound O-acyl transferase (MBOAT) that resides in the endoplasmic reticulum, catalyses the addition of monounsaturated palmitate to Wnt proteins and is required for Wnt gradient formation and signalling. In humans, PORCN mutations are causal for focal dermal hypoplasia (FDH), an X-linked dominant syndrome characterized by defects in mesodermal and endodermal tissues. PORCN is also an emerging target for cancer therapeutics. Despite the importance of this enzyme, its structure remains poorly understood. Recently, the crystal structure of DltB, an MBOAT family member from bacteria, was solved. In this report, we use experimental data along with homology modelling to DltB to determine the membrane topology of PORCN. Our studies reveal that PORCN has 11 membrane domains, comprising nine transmembrane spanning domains and two reentrant domains. The N-terminus is oriented towards the lumen while the C-terminus is oriented towards the cytosol. Like DltB, PORCN has a funnel-like structure that is encapsulated by multiple membrane-spanning helices. This new model for PORCN topology allows us to map residues that are important for biological activity (and implicated in FDH) onto its three-dimensional structure.

acyl group from an acyl-coenzyme to a substrate. PORCN belongs to a subgroup of enzymes that specifically promote the acylation of protein or peptide substrates. Other relevant members of this subgroup include hedgehog acyl transferase (HHAT) and ghrelin O-acyl transferase (GOAT), which catalyse the lipid modification of Hedgehog and Ghrelin, respectively. Each of these enzymes catalyses reactions across the ER membrane. While the Acyl-CoA substrate is localized to the cytosol [13], the protein substrate (Wnt, Hh or Ghrelin) is located in the lumen of the ER [14,15]. Most MBOAT enzymes, including PORCN, HHAT and GOAT share a highly conserved histidine residue [1]. Though many reports indicate that the histidine residue is required for catalysis [1,[16][17][18][19][20], subsequent data showing that one HHAT variant, with a mutated histidine, retains significant activity challenges this idea [21].
The only MBOAT for which the crystal structure has been reported is DltB, a bacterial enzyme that promotes the D-alanylation of cell wall teichoic acid [22]. The elucidated structure of DltB reveals that it is composed of a ring of 11 transmembrane domains, which forms a funnel embedded in the lipid bilayer [22]. The authors speculate that the funnel may be important for substrate binding or catalysis [22].
The existence of multiple PORCN mutations that cause FDH provide us with a rare opportunity to investigate the relationship between the structure and function of an MBOAT family member [29][30][31][32][33][34]. Although multiple groups have begun to address the structural features of PORCN, significant uncertainty about the membrane topology remains [25,29,36,[44][45][46].
In this study, we used four distinct methods to characterize the membrane topology of PORCN. First, we leveraged topology modelling software to predict the location of transmembrane spanning domains. Second, we used a differential solubilization technique to distinguish whether different epitopes are oriented towards the ER lumen or cytosol. For these methods, we used an antibody against PORCN generated in our laboratory as well as commercially available antibodies that recognize multiple epitope tags. Third, we created PORCN variants with engineered consensus sites for the addition of N-linked glycosylation, which only occurs in the lumen of the ER. We then assessed the glycosylation status of different variants. Fourth, we used homology modelling and molecular dynamics simulation to predict the threedimensional structure of PORCN. Taken together, these studies reveal that PORCN has nine transmembrane domains with two reentrant membrane domains.
Human PORCN variant D (hPORCND) was used to verify the specificity of the PORCN antibody. hPORCND was cloned into pcDNA3.1/MYC-His(-)A with the stop codon intact to prevent the addition of the MYC-His epitope [48]. HA-tagged hPORCND was created by replacing the 5 0 end of hPORCND with the 5 0 end of HA-tagged mouse PORCN variant D (mPORCND). The resulting protein has a 3x HA epitope tag, amino acid 2-114 mouse PORCN and amino acid 115-461 human PORCN. This was then cloned into pcDNA3.1/MYC-His(-)A with the stop codon intact to prevent the addition of the MYC-His epitope.
MPORCND was subcloned into p3xFLAG-CMV-14 to add the 3xFLAG epitope (DYKDHDGDYKDHDIDYKDDDDK) to the C-terminus (PORCN-3XFLAG) [48]. The MYC epitope (EQKLISEEDL) was added to the N-terminus of PORCN-3XFLAG and wild-type PORCN and the C-terminus of wild-type PORCN by PCR. All the internal MYC epitope tags with or without linkers were added to PORCN-3XFLAG in the p3xFLAG-CMV- 14  For the glycosylation studies, we used HA-mPORCN cloned into pcDNA3.1/MYC-His(-)A (note: the stop codon was not removed, therefore HA-mPORCN did not get the MYC-His epitope tags from the vector) and added the glycosylation sites by mutating the amino acids by overlapping extension PCR (H53N, R90N, A134N, P184N, R235N,  D283N, D384N, D423N, K438N). The C-terminus glycosylation site with the linker was added by PCR (gylNVTyv). In addition, we used eGFP cloned into pcDNA3.1/MYC-His(-)A (note: the stop codon was not removed, therefore eGFP did not receive the MYC-His tag).

Generation of anti-PORCN polyclonal antibody
Generation of polyclonal antibodies to human PORCN was done in a collaboration with Prosetta Biosciences. The peptide that was used as the immunogen correlates to amino acids 281-300. The sequence of the peptide was cEWDLTVSKPL NVELPRSMVE.

Topology prediction algorithms and visualization tools
We used the TOPCONS web server to predict the membrane topology of mPORCN variant D. We also used the ExPASY server to generate a Kyte and Doolittle hydropathy plot [49]. Lastly, we took advantage of the TOPO2 tool to assist with the visualization of topology models [50,51]

Click chemistry palmitoylation assay
HEK293T cells were seeded in a 6-well plate, incubated overnight and co-transfected with 1.65 µg of pcDNA3.1 encoding spGFP:WNT1 (209-239):F C fusion along with 1.65 µg of pcDNA 3.1 encoding eGFP (as filler) or mPORCND tagged constructs using the FUGENE HD transfection kit (Promega) according to standard protocol. The cells were incubated overnight and then metabolically labelled with 100 µM of Alkyne palmitic acid (Alk-C16) or treated with DMSO as a control. After metabolic labelling of cells with Alk-C16 for 20-24 h, the cells were washed once with PBS and lysed in 300 µl of 100 mM Sodium Phosphate, pH 7.5 with 150 mM NaCl, 1% NP-40 and 1× Halt protease and phosphatase inhibitor cocktail for 1 h at 4°C. The lysate was cleared by centrifuging for 10 min and 260 µl of supernatant was transferred to a new tube. The F C containing proteins were immunoprecipitated with Protein A/G beads. Proteins retained on the beads were subjected to click chemistry with biotin azide before separation by SDS-PAGE and analysis by western blot [11,12]. The blots were probed with IRDye800 conjugated streptavidin, and anti-GFP antibody followed by goat Alexa Fluor_680-conjugated antimouse secondary antibody. Blots were then scanned using an Odyssey CLx imaging system and analysed using Image Studio software.

Glycosylation analysis
HEK293T cells were split to 24-well plates 1 day prior to transfection to yield 80-90% confluence the next day. The cells were then transfected with the indicated construct with Fugene HD reagent according to the manufacturer's instructions. The cells were incubated for 4 h at 37°C 10% CO 2 . DMSO or tunicamycin was then added to yield a final tunicamycin concentration of 2.5 ng µl −1 (DMSO was diluted the same way as a control). The cells were incubated overnight, lysed in 150 µl of 100 mM Hepes pH7.4 + 100 mM NaCl + 1% royalsocietypublishing.org/journal/rsob Open Biol. 11: 200400 TX-100, transferred to a microfuge tube and heated to 60°C for 5 min with SDS-PAGE loading dye (final concentration: 0.32 M Tris pH6.8 with 20 mM DTT, 2% SDS, 10% glycerol and 0.004% Bromophenol Blue) added. The lysates were then triturated with a 25-gauge needle five times, ran on a 10% SDS-PAGE gel and electroblotted to PVDF Immobilon-FL membrane. The blot was probed with anti-HA antibody followed by goat Alexa Fluor_680-conjugated anti-mouse secondary antibody. Blots were scanned in an Odyssey CLx imaging system and analysed using Image Studio software.

Homology modelling of PORCN
A homology model of mPORCND was generated using YASARA (v. 19.12.14) (Vienna, Austria) in automated mode [57], using the sequence of mouse PORCN isoform D (accession code, NP_076127.1). The model primarily used coordinates from the X-ray crystal structure of DltB, an integral membrane protein from Streptococcus thermophiles (PDB = 6BUH, solved to 3.15 Å) as a homology template [22]. YASARA created an ensemble of five models (6BUH-C01 through 6BUH-C05) with Z-scores ranging from −1.313 to −1.591. YASARA also created a hybrid model of the five individual models. However, the Z-score of the hybrid was less favourable (−1.840) than for model 6BUH-C01 (Z-score = −1.313). As such, model 6BUH-C01 was designated as the final model for mPORCND.

Molecular dynamics simulations
A molecular dynamics simulation of PORCN was carried out to determine the stability of the structure and to examine its equilibrated conformation using the Amber18 and Amber-Tools18 packages [58]. The protein was embedded into a simulated bilayer of 1-palmitoyl-2-oleoylphosphatidylcholine (POPC) molecules employing the CHARMM Membrane Builder tool [59]. Topology files for the protein were constructed in TLEAP [22] using the FF14SB [60], LIPID17 and GAFF force fields [61]. The membrane-embedded protein was solvated in explicit TIP3P water arranged in an 8 Å box from the surface of the protein-membrane complex [62]. Prior to minimization, equilibration and production run, the complexes were charge neutralized by the addition of chloride anions. The MD simulations were carried out using particle mesh Ewald (PME) graphical processor unitaccelerated MD with the PME MD Compute Unified Device Architecture module of Amber18 [63]. Initially, the protein was subjected to (i) minimization with the restraint of the protein for 1000 cycles; (ii) unrestrained minimization for 1000 cycles; (iii) equilibration while warming from 50 K to 300 K for 30 ps using a constant-volume periodic boundary condition; (iv) equilibration with the restraint of the protein at 310 K for 20 ps with a constant-pressure periodic boundary condition and using isotropic pressure scaling; and (v) unrestrained equilibration at 300 K for 500 ps. After minimization and equilibration, the protein was subjected to a production run at 300 K for 50 ns. Preparation steps 1 and 2 comprised 500 steps of steepest descent, followed by 500 steps of conjugate-gradient descent. Steps 3-5 and the production runs used Langevin temperature regulation with the collision frequency of 2.0 ps −1 [64], bonds involving hydrogen were constrained by the SHAKE algorithm [65]. A 12 Å cutoff was used for non-bonded interactions calculated by the PME method [66]. The MD simulations were monitored by examination of the internal energy and root mean standard deviation (RMSD) of the resulting trajectories (electronic supplementary material, figure S2). The production run trajectories were visualized using visual molecular dynamics (VMD; v.1.9.3) [67], while static snapshots of the protein were visualized using PYMOL (Schrödinger, San Diego, CA).

Bioinformatic analysis of PORCN topology
As a first step towards understanding the membrane topology of PORCN, we used six different algorithms to predict the location and number of transmembrane domains (figure 1). There was excellent consensus around the existence, length and location of six membrane domains (figure 1, shown in grey

Validation of differential solubilization methodology
We then sought to generate experimental evidence to support or refute the predictions from our bioinformatics analysis. To do this, we analysed wild-type and epitope-tagged variants of PORCN by differential solubilization [68][69][70]. For the differential solubilization method, fixed cells are solubilized with either TX-100 or digitonin before being subjected to immunostaining. Whereas TX-100 solubilizes both the plasma membrane and the ER membrane, digitonin only solubilizes the plasma membrane. In TX-100 cells, both cytosolic and luminal epitopes should be accessible to antibodies and thus, can be visualized by immunostaining. By contrast, in digitonin labelled cells, only epitopes that are oriented towards the cytosol should be accessible to antibodies. Because the cells were transiently transfected in this experiment, some cells received DNA while others did not. As such, the non-transfected cells (labelled with DAPI only) served as a control for nonspecific antibody staining. Examination of these cells reveals little to no nonspecific staining. To further validate this methodology, we tested the localization of the C-terminus of Frizzled10 (FZD10), which is a classic seven-transmembrane spanning protein resembling a G-protein coupled receptor, (figure 2, bottom panel, a-d) and Lysozyme-MYC-KDEL, an ER-resident protein (figure 2, bottom panel, e-h). As expected, our data show that the MYC epitope on the C-terminus of FZD10 was accessible to antibodies when cells were solubilized with either TX-100 or digitonin, thus localizing the C-terminus to the cytosol. Conversely, the MYC epitope on Lysozyme-MYC-KDEL was accessible when cells were solubilized in TX-100, but not digitonin, thus localizing it to the ER lumen. This important control was performed each time we carried out the differential solubilization experiment.

PORCN topology determination using an antibody against PORCN
As a first step towards understanding the membrane topology of PORCN, we first sought to raise polyclonal antibodies to royalsocietypublishing.org/journal/rsob Open Biol. 11: 200400 peptides found in different predicted loops of hPORCND.

Biological activity of tagged PORCN constructs
Because we had only a single PORCN antibody, it was necessary to create a panel of epitope-tagged PORCN constructs to localize the termini and additional loops (figure 1b). As tagging proteins often affects the structure and function of the protein, we first tested the biological activity of our tagged variants using a click chemistry-based palmitoylation assay (figure 3,a) [11,12,71,72]. Activity in this assay is assessed by comparing

Determination of the orientation of the amino and carboxy-termini of PORCN
We then sought to define the orientation of the N-and C-termini of PORCN. We first introduced a MYC tag onto the N-and C-termini of PORCN. When overexpressed in COS7 cells, the N-terminal MYC epitope was readily accessible to antibodies when cells were solubilized with TX-100 (figure 4a). However, upon solubilization with digitonin, the epitope was largely masked (figure 4b). Though some immunostaining was evident, it was not in the reticulated pattern of the ER. Thus, it seems likely that the N-terminus is oriented towards the lumen. By contrast, the C-terminally tagged PORCN was visualized when cells were solubilized with either TX-100 or digitonin (figure 4c,d), thus suggesting that the C-terminus is oriented towards the cytosol. Because different tags can sometimes yield different results, we also tagged PORCN on the C-terminus with 3XFLAG (in lieu of MYC). As before, immunostaining was evident in both TX-100 and digitonin labelled cells (figure 4e,f ). Lastly, we assessed the localization of the N-terminus in PORCN with dual tags, a MYC tag on the N-terminus and the 3XFLAG tag on the C-terminus. Again, we were able to visualize staining for the MYC tag in TX-100, but not digitonin, solubilized cells (figure 4h, k). The 3XFLAG epitope was successfully stained in both TX-100 and digitonin solubilized cells ( figure 4i,l ). Thus, we conclude that the N-terminus is oriented towards the lumen and the C-terminus is oriented towards the cytosol. We further conclude that the use of a dual tag system does not profoundly alter the topology of PORCN.  PORCN variants with C-terminal 3XFLAG tags and internal MYC tags were then used to investigate the orientation of predicted internal loops using the differential solubilization assay. As a first step, we sought to validate that both epitopes were available when cells were solubilized with TX-100. The MYC epitopes for the fusions at positions 118, 145, 181, 223, 289, 376 and 427 (figure 5e-e 0 ,g-g 0 ,i-i 0 ,k-k 0 ,m-m 0 ,q-q 0 ,s-s 0 ) were successfully visualized when cells were solubilized with TX-100. The results with digitonin solubilized cells suggest that the loops containing residues 118, 223 and 427 (figure 5f 0 ,l 0 ,t 0 ) are oriented towards the lumen while loops containing residues 145, 181, 289 and 376 (figure 5h 0 ,j 0 ,n 0 ,r 0 ) are oriented towards the cytoplasm. As the fusions in which MYC tags were inserted at residues 47, 92 and 323 (figure 5a-a 0 ,c-c 0 ,o-o 0 ) were not consistently immunostained when cells were solubilized with TX-100, the results from digitonin solubilized cells for these three constructs did not provide meaningful information.
To circumvent the possibility that the epitopes for 47 and 92 were near the membrane bilayer or constrained in an unfavourable conformation, we then transfected and immunostained cells expressing MYC tags flanked by flexible linkers in the same positions. This time, both the MYC and 3XFLAG tags were detected when TX-100 solubilized cells were immunostained (figure 6a 0 ,e 0 ). Data from digitonin solubilized cells suggest that both of these epitopes are oriented towards the cytoplasm (figure 6b 0 ,f 0 ).
To test for the existence of additional short loops predicted between residues 68-70 (OCTOPUS, SCAMPI and SPOCTOPUS) and 342-344 (OCTOPUS, Philius, Polyphobius, SPOCTOPUS and TOPCONS), we created two more MYC epitope constructs flanked with flexible linkers at position 69 or 343. Both the MYC and 3XFLAG tags were detected upon immunostaining of TX-100 solubilized cells (figure 6c 0 ,g 0 ). Digitonin solubilized cells showed no detectable immunostaining, thus indicating that these epitopes are oriented towards the lumen (figure 6d',h').
To our surprise, our data for one of the loops (114-151) yielded different results depending on which construct was used. When the MYC tag was inserted into PORCN-3XFLAG at residue 118, the epitope localized to the lumen. However, when the MYC tag was inserted at position 145, the epitope was localized to the cytoplasm. These data suggested the possibility that there could be an additional membrane-spanning region that was not identified by the topology algorithms.

PORCN hydropathy plot suggests one additional membrane-spanning domain
Because we had conflicting data for one of the predicted loops of PORCN, we created a schematic in which we overlaid the membrane domains predicted by bioinformatics analysis along with the orientation of different loops onto a Kyte and Doolittle hydropathy plot (figure 7). As expected, the membrane domains predicted by bioinformatics analysis overlapped with hydrophobic regions on the plot. We noted, however, the existence of one additional moderately hydrophobic region spanning residues 124-140 (figure 7, dark grey box), which was not identified by any of the six algorithms used for our bioinformatics analyses. Because this domain is flanked by sequences which we have experimentally shown to be on opposite sides of the membrane, the most parsimonious interpretation is that residues 124-170 do indeed represent a previously unidentified membrane-spanning  Figure 4. N-terminus of PORCN is in the lumen and the C-terminus is in the cytosol. We first tested the single tag MYC PORCN constructs tagged on either the N-terminus (a,b) or the C-terminus (c,d). While the C-terminus had immunostaining in both the TX-100 and digitonin permeabilization, the N-terminus did not, indicating that the C-terminus is localized to the cytosol and the N-terminus in the lumen. We also wanted to create an internal control for the internal MYC tags. Therefore, we replaced the C-Terminus MYC with 3x-FLAG (e,f ) and it also immunostained when digitonin permeabilization was used. Finally, a MYC-PORCN-3xFLAG construct was tested (g-l ) and repeated the result that the N-terminus is in the lumen and the C-terminus is in the cytosol. Each construct was tested at least twice. Scale bar is 20 µm.  Figure 5. Determination of internal MYC tagged loops. All internal MYC constructs with C-terminus 3xFLAG were transfected into COS7 cells. The C-terminus 3xFLAG served as an internal control to ensure the topology was not altered by the internal MYC tag. As such, we expected the 3xFLAG epitope to immunostain with both the TX-100 and digitonin permeabilization for all the constructs, which was visualized (b 00 -t 00 ). Immunostaining of the internal MYC with the TX-100 permeabilization was expected to yield the same immunostaining as the 3xFLAG epitope. However, this was not the case; 47MYC (a 0 ), 92MYC (c 0 ) and 323MYC (o 0 ) did not have the same immunostaining pattern as the 3xFLAG, therefore we concluded that we could not use these three tags for a diagnostic of where the loop was located. As for the other constructs permeabilized with digitonin and immunostained with anti-MYC, 145MYC (h 0 ), 181MYC ( j 0 ), 289MYC (n 0 ) and 376MYC (r 0 ) all showed the same immunostaining as 3xFLAG indicating that those loops are in the cytosol, while 118MYC ( f 0 ), 223MYC (l 0 ), 427MYC (t 0 ) did not have the same immunostaining pattern and therefore indicates that these loops are in the lumen. Each construct was tested at least twice. Scale bar is 20 µm. royalsocietypublishing.org/journal/rsob Open Biol. 11: 200400 domain. Thus, these data initially suggested to us that PORCN has a total of 12 membrane domains (figure 1).

Analysis of membrane topology using Nglycosylation consensus sequence tags
We then sought to use an independent mechanism to further assess the orientation of different domains. To do this, we created 10 variants in which consensus sites for N-glycosylation were added (NxT/S) to HA-PORCN (figure 1b). For the internal sites, we mutated codons for residues that preceded X and T/S residues such that they now coded for an arginine (N). For the C-terminal site, we added an NVT sequence with linkers flanking the glycosylation site at the C-terminus of PORCN. We then compared the migration of these variants to HA-PORCN using SDS-PAGE. Seven of the variants migrated as a single band of 42 kDa, suggesting that they were not glycosylated. This result, however, is not diagnostic as the epitope may not have been available for glycosylation. Three of the variants, A134N, R235N and D283N, migrated as a doublet, consistent with the addition of N-linked sugars (figure 8).
To verify that the slower migrating band was indeed a consequence of N-linked glycosylation, we treated the cells with tunicamycin and showed that HA-PORCN variant was reduced to a single band (figure 8). While the data showing that R235 is glycosylated is consistent with the data from our differential solubilization studies, the finding that A134 and D283 are glycosylated is not in agreement with the prediction that A134 is in a transmembrane spanning domain and D283 is in the cytosol. As such, we then turned to  Figure 6. Determination of four additional internal MYC tagged loops with flexible linker. Four additional internal MYC constructs were created with a flexible linker to further assess the topology of PORCN. The four MYC constructs with C-terminus 3xFLAG were transfected into COS7 cells. The C-terminus 3xFLAG served as an internal control to ensure the topology was not altered by the internal MYC tag. As such, we expected the 3xFLAG epitope to immunostain with both the TX-100 and digitonin permeabilization for all the constructs, which was visualized (a 00 -h 00 ). Immunostaining of the internal MYC with the TX-100 permeabilization was expected to yield the same immunostaining as the 3xFLAG epitope, which it did (a 0 ,c 0 ,e 0 ,g 0 ). Permeabilized with digitonin and immunostained with anti-MYC, 47MYC (b 0 ) and 92MYC ( f 0 ) showed the same immunostaining as 3xFLAG indicating that those loops are in the cytosol, while 69MYC (d 0 ) and 343MYC (h 0 ) did not have the same immunostaining pattern and therefore indicates that these loops are in the lumen. Each construct was tested at least twice. Scale bar is 20 µm.  Figure 7. Hydropathy plot reveals one additional transmembrane domain. Kyte and Doolittle hydropathy plot was created for mouse PORCN. Overlaid is a schematic of PORCN with the boxes illustrating the predicted transmembrane domains. Light grey boxes were used for the domains that all six of the transmembrane prediction algorithms predicted, white boxes for the domains that were predicted by some of the algorithms and finally one dark grey box that none of the algorithms predicted by but do have a hydrophobic region indicated in the hydropathy plot. royalsocietypublishing.org/journal/rsob Open Biol. 11: 200400 homology modelling and molecular dynamics simulations to resolve these inconsistencies.

Homology modelling and molecular dynamics
simulations of PORCN predict topology that aligns with differential solubilization data A homology model of mPORCND was generated based on the high-resolution X-ray crystal structure of DltB, an MBOAT family member from S. thermophiles, which was solved at 2.5 Å (PDB = 4EZD; [22]. It is important to note that mPORCND and DltB are only 15% identical and 36% similar using the BLOSUM62 alignment matrix [73]. However, this is not unexpected as it has been established that membrane proteins are dramatically less well conserved than water-soluble proteins even though the functions, and presumably the structures, are well conserved [74]. Homology modelling has also been used to validate the structure of GOAT, which shares only 12% identity and 26% similarity with DltB (see figure 10c) [75]. However, Campaña et al. [75] only used regions sharing substantial homology between GOAT and DltB for the pairwise structural alignments. In addition to the primary template, the homology modeller in YASARA employed an ensemble of 137 proteins to model various domains of the final structure. We then carried out a molecular dynamics (MD) simulation to refine the homology model of PORCN under equilibrating conditions within a bi-lipid membrane. The protein produced from homology modelling was embedded into a simulated membrane comprised POPC molecules using the CHARMM membrane builder utility [59], with the orientation and position of the protein within the lipid determined using the automated PPM server [76]. The lipid-protein complex was subjected to a 50 ns production run at 300 K. We observed that the majority of the equilibration was complete within 25 ns for the membrane-protein complex, although the simulation was continued for a total of 50 ns (electronic supplementary material, figure S2). The root-mean-square deviation at equilibrium suggested it was stabilized at approximately 2.80 Å. The MD-refined homology model is depicted in figure 9, with the approximate membrane position as was determined by the PPM server during the initial membrane embedding step. Consistent with the crystal structure of DltB, the homology modelling clearly shows the presence of a funnel composed of a ring of 10 membrane domains, including eight transmembrane domains and two reentrant membrane domains. This funnel is thought to serve as the conduit for acyl-CoA to be translocated from the cytoplasmic side of the endoplasmic reticulum to the luminal side, where neonascent Wnts are located [14,22,75].
We then compared the position of N-and C-termini, and key amino acids observed in the homology model, with the position of the analogous features in MYC tagged experimental model (table 1). The MD-refined homology model predicts Q69, V118, G223 and E427 in the lumen, and L47, S92, E145, Q181, P289, R323 and I376 extending into the cytosolic space. F343 is predicted to reside in the membrane-spanning domain. Furthermore, the MD-refined homology model suggests that both the N-and C-termini reside in the lumen.
The MD-refined homology model is remarkably consistent with our differential solubilization data, but not the glycosylation variant data. Specifically, the MD-refined homology model and the differential solubilization data are in full agreement with respect to the localization of the N-terminus and residues L47, Q69, S92, V118, E145, Q181, G223, P289, I376 and E427. One minor inconsistency is that the differential solubilization data indicate that F343 is oriented in the lumen while the MD-refined homology model indicates that it is positioned towards the luminal side of the membrane bilayer (table 1). However, it is important to note that the F343 MYC-tagged variant has flexible linkers on each side of the MYC tag (figure 6) and thus it is possible that these extra sequences may have caused the MYC tag to protrude out of the membrane bilayer into the lumen. The only significant discrepancy between the differential solubilization data and the MD-refined homology model is in the position of the C-terminus (figures 4 and 9). Whereas the homology model places the C-terminus in the lumen, the differential solubilization data indicate that it resides in the cytosol (figures 4 and 9). It is important to note that the C-terminus of PORCN extends beyond the sequences in DltB. As such, the homology model had no basis for assigning the C-terminus of PORCN to the cytosol or the lumen. As the differential solubilization data with both MYC and 3XFLAG tagged PORCN clearly show that the C-terminus is oriented towards the cytosol, we favour the idea that the C-terminus is indeed localized to the cytosol (figure 4).

Final model for PORCN membrane topology
We then used a combination of our differential solubilization data and our MD-refined homology modelling data to arrive at our final model for PORCN topology. Our final model for PORCN has nine transmembrane domains and two reentrant membrane domains ( figure 1a and 10). We further propose that the N-terminus is oriented towards the lumen while the C-terminus is oriented towards the cytosol. The conserved N306 residue is localized to a cytoplasmic loop while the conserved H341 residue is positioned near the apex of the funnel. 4. Discussion

Summary of topology
Our studies using both experimental and computational approaches have provided us with a reliable membrane topology for PORCN that will undoubtedly advance the design of novel PORCN inhibitors and allow for deeper inquiry into the structure and function of this critical enzyme. Though we had set out to exclusively use anti-PORCN antibodies developed against different regions of PORCN for our experimental studies, only one of our PORCN peptide immunogens yielded usable antibodies. As such, we introduced multiple epitope tags throughout the protein in positions that were suggested to be located in the loops between the membrane domains. Because epitope tags are known to cause changes in protein structure, we were very circumspect about these studies. However, several pieces of data served to increase our confidence. First, the introduction of internal MYC tags never altered the position of our C-terminal 3xFLAG tag (figures 5 and 6). Second, the anti-PORCN antibody showed the same localization as the 289MYC tag ( figure 5). And lastly, our structural modelling experiments in which we modelled PORCN based on the crystal structure of DltB validated all of our experimental data using epitope tags, except at the C-terminus. Recall that PORCN is significantly longer than DltB; as such, the modelling software had no basis for predicting the position of C-terminus of PORCN. Our final PORCN model indicates the presence of nine transmembrane domains and two reentrant membrane domains. As reentrant membrane domains are common in transmembrane proteins that are pores or channels, the presence of two reentrant membrane domains in PORCN is consistent with the presence of a funnel that shuttles palmitoleoyl-CoA from the cytoplasm to the lumen [79]. Interestingly, the introduction of consensus sites for the addition of N-linked glycosylation at 10 different positions yielded somewhat inconsistent results. Specifically, our differential solubilization and homology modelling data suggest that A134 is positioned in the membrane bilayer while D283 is localized in the cytosol. Notably, the differential solubilization data placing D283 in the cytosol is our strongest data as we were able to use anti-PORCN antibodies to  royalsocietypublishing.org/journal/rsob Open Biol. 11: 200400 define the orientation of wild-type (untagged) PORCN for that particular loop. However, our glycosylation data indicate that A134 and D283 are localized to the lumen. Taken together, these results provide more confidence in the differential solubilization data coupled with the MD-refined Homology model than in the results from studies in which we inserted new glycosylation sites. As such, we recommend that previous studies having relied on the insertion of glycosylation sites to provide insights about membrane topology be interpreted with caution. Though it is not clear why this method yields inconsistent results, one could speculate that the glycosylation of this site occurred while PORCN was still being translated and inserted into the membrane.
Having said that, we cannot completely rule out the possibility that the low levels of homology between PORCN and DltB compromised the validity of the homology modelling and that PORCN has alternate conformers.

Comparison of our model with other PORCN models
Though our PORCN topology model is similar to other reported models in many respects, there are clear differences.
Whereas we demonstrate the presence of nine transmembrane domains with two reentrant membrane domains, other models propose anywhere between 8 and 11 transmembrane domains [25,29,36,[44][45][46]. And while our studies clearly indicate that the N-terminus is in the lumen, all four of the previous models show it in the cytoplasm. On the other hand, three of the four previous models agree with us that the C-terminus in the cytoplasm. Important experimental differences between our study and previous studies are that (i) we used a combination of experimental data and homology modelling and (ii) we created and used more epitope tags than any of the previous reports. The addition of additional epitope tags allowed us to identify a (a) current PORCN model (b) PORCN Lee et al. [44] (c) GOAT Campaña et al. [75] (e) GOAT Taylor et al. [77] (e) HHAT Figure 10. PORCN experimentally determined topology is similar to HHAT and GOAT topology. TOPO2 was used to create the topology of mouse PORCN that match the experimental data collected and homology modelling. The topologies for different reported models of PORCN (b), GOAT (c,d ) and HHAT(e) are also shown for comparison [44,75,77,78].
royalsocietypublishing.org/journal/rsob Open Biol. 11: 200400 transmembrane domain that had not been predicted by any of the topology prediction algorithms.

Topology comparisons reveal that PORCN has a similar topology to GOAT and HHAT
Because PORCN, GOAT and HHAT share common functions as O-acyl transferases, we predicted that the three enzymes would also share structural features. As such, we then compared our new topology model to those reported for GOAT and HHAT (figure 10) [75,77,78,80]. The N-termini of both PORCN and GOAT are oriented towards the lumen while that of HHAT is reported to be oriented towards the cytosol. Consistent with the notion that the catalytic domain includes a conserved histidine (H341 in PORCN), the membrane topologies of PORCN, GOAT and HHAT are most similar in the C-terminal halves of the proteins, with all three proteins having their C-termini localized to the cytosol. In addition, PORCN transmembrane domains 1-5, reentrant membrane domain 1 and transmembrane domains 6 and 7 roughly align with those of GOAT, with only the lengths of the loops varying [75,77]. Notably, the conserved histidine in PORCN, GOAT and HHAT is positioned in the membrane bilayer. However, our homology modelling data indicate that the residues immediately following the histidine remain embedded in the membrane while those for GOAT and HHAT are predicted to extend into the lumen. Our studies place the conserved asparagine residue (N306 in PORCN), which is hypothesized to be important for binding to acyl-CoA, to a cytoplasmic loop. This aligns well with the GOAT topology model from Taylor  4.4. Many residues implicated in focal dermal hypoplasia are localized to the funnel Because our MD-refined homology model showed close agreement with our differential solubilization data, we used it to assess the location of key residues in PORCN, for which missense mutations are known to cause FDH (figures 9g-i, 10a). The MDrefined homology model shows that many of the residues that are involved in FDH are localized to the funnel, including S136, G168, H252, V258, S297, L331, H341, E361 and R365. These data are consistent with those from Campaña et al. [75], who suggested that surface-exposed side chains in GOAT are less important than those within the interior of the enzyme. In addition, our model places the H341 residue, which has specifically been implicated in catalysis, near the apex of the funnel in the membrane bilayer [81]. The localization of many of the residues implicated in FDH to the funnel is also consistent with the previous studies suggesting that the funnel is important for shuttling acyl-CoA from the cytoplasmic face of the lipid bilayer to the lumen [13,22,75]. Cumulatively, these data are consistent with the importance of the funnel for the PORCN function.

Future directions
Our proposed topology also raises some interesting questions for future investigations. The presence of the funnel offers up a mechanism for transporting acyl-CoA to the active site, marked by H341. Given that H341 is embedded in the membrane, how do Wnts gain access to the active site? We and others have previously shown that the Wnt sequences immediately surrounding the palmitoylation site form a thumb-like projection that is held together by two disulfide bonds, with the palmit(e)oylated serine being localized to the distal-most aspect of the thumb [10,23,44]. Thus, the palmit(e)oylation site is part of a rigid structure that could be inserted into the membrane. We note that many of the amino acids flanking the serine have positively charged side groups and thus, speculate that this may promote interactions of the thumb with negatively charged phospholipid head groups. Are these positively charged amino acids critical for binding? In addition, although mouse PORCN is not palmitoylated, human PORCN has been shown to be palmitoylated on cysteine 187 [12]. What class of palmitoyl acyl transferases might be responsible for this modification? As our topology places this residue in a cytosolic loop, we speculate that a member of the DHHC family is likely to be responsible for this modification. The region of PORCN that has splice variants (A-D) is located between transmembrane domains 6 and 7 in a luminal loop [48]. What is the role of these splice variants? As it turns out, splice variant A, which is the shortest of the variants, appears to have approximately 50% less activity than B-D [46]. As such, it is possible that the composition or length of the variant is important for Wnt binding to PORCN. Lastly, we believe that aspects of the MD-refined homology model could be validated or improved by computational modelling studies of known PORCN inhibitors such as LKG-974 [82], GNF-6231 [83] and characterized analogues of each. While we believe the C-terminal region of our homology model is misassigned to the lumen, we are more confident on the structure of the transmembrane alpha-helices and reentrant loops, due to successful localization of residues in the cytosol and lumen consistent with differential solubilization data. As such, we Table 1. Summary of molecular dynamics-comparing refined homology model and experimentally derived model for residues that were MYC tagged. royalsocietypublishing.org/journal/rsob Open Biol. 11: 200400 predict the central cavities in the cytosolic and luminal domains contain plausible binding sites for small molecule inhibitors. Once binding modes are hypothesized, binding affinities can be estimated using MM-GBSA and MM-PBSA methods [84], and can be compared with experimental inhibitory potency, over a series of compounds. We expect that reasonable inhibitor-bound structures could be further refined by MD to improve the structure of PORCN, and could also be used as a template for virtual screening experiments.
In sum, the determination of the PORCN membrane topology along with the identification of residues that are critical for PORCN function (from FDH) will allow us to further our understanding of the relationship between structure and function for this important enzyme. We look forward to better understanding the roles of these residues in PORCN function. performed all of the homology modelling and molecular dynamics simulations. J.G.F., L.S. and R.B. contributed to experimental design and carried out differential solubilization and glycosylation experiments. D.N.H. validated the anti-PORCN antibody. E.U.M. helped make the PORCN-3XFLAG construct used in most of the studies. V.R.L. assisted with the generation of anti-PORCN antibody and provided input into experimental design. L.W.B. designed and carried out some experiments and took the lead role in writing the manuscript.