Structural model of human PORCN illuminates disease-associated variants and drug-binding sites

Wnts are secreted fatty acid-modified glycoproteins that play essential roles in both development and adult homeostasis (Loh et al., 2016; Zhong et al., 2020). Wnt ligand binding to receptors and co-receptors on the surface of target cells initiates an array of downstream signaling events, most notably β-catenin-mediated transcription (Jung and Park, 2020; Nusse and Clevers, 2017; Yu and Virshup, 2014). Owing to its important role in diverse processes, the Wnt pathway is tightly regulated at multiple levels. One essential control point is the production and transport of Wnt ligands from Wnt-producing cells to target cells in the local environment.

There are 19 Wnt genes in the mammalian genome, and each requires a post-translational acylation on an essential serine positioned at the tip of the disulphide bond-stabilized Wnt hairpin 2 (Bazan et al., 2012). This unique modification is catalyzed by the enzyme porcupine (PORCN), which is present in essentially all metazoans. PORCN is a member of the family of membrane-bound O-acyltransferases (MBOATs) (Hofmann, 2000) and catalyzes the transfer of a monounsaturated palmitoleate [(Z)-hexadec-9-enoate] from coenzyme A (CoA) to the hydroxyl of the serine residue (see model, Fig. 1A) (van den Heuvel et al., 1993; Riggleman et al., 1990; Takada et al., 2006). Acylated Wnts are then able to transfer to the Wnt transport protein WLS and be transported to the plasma membrane (Coombs et al., 2010; Nygaard et al., 2021). The palmitoleate modification is also required for Wnt binding to Frizzled (FZD) receptors (Hirai et al., 2019; Janda et al., 2012; Nile et al., 2017).

Because Wnt signaling drives an important subset of cancers and PORCN function is essential for Wnt activity, a growing number of small molecules that potently inhibit the catalytic activity of PORCN have been developed (Shah et al., 2021; Zhong and Virshup, 2020). These drugs block the palmitoleation of Wnts and inhibit the growth of Wnt-addicted cancers. Several of these PORCN inhibitors are in human clinical trials. How and where these drugs interact with PORCN to block its activity is not known, although pharmacophore models have been developed to identify critical features of the drugs (Poulsen et al., 2015).

Wnt signaling is also critically important during embryonic development (Loh et al., 2016). Loss-of-function mutations of PORCN and WLS that block the acylation and delivery of Wnts cause early embryonic lethality in mouse knockout models (Barrott et al., 2011; Biechele et al., 2011; Fu et al., 2009). Multiple congenital malformation syndromes are due to partial loss-of-function mutations in Wnt production. Among those, diverse germ line mutations in PORCN are associated with Goltz syndrome, an X-linked genetic disorder also known as focal dermal hypoplasia (FDH; OMIM 305600) (Bostwick et al., 2016; Goltz, 1992; Grzeschik et al., 2007; Wang et al., 2007). This rare disorder affects the development of ectodermal and mesodermal tissues, and affected individuals have variable defects in limbs, skin and bone as well as other abnormalities in internal organs. FDH is usually found in heterozygous females in whom the severity of the disease depends on the pattern of X-chromosome inactivation or the presence of somatic mosaicism. Males may also have FDH, usually in the setting of a mosaic mutation, although germline mutations in PORCN have been identified in hemizygous males with colobomata, microphthalmia, anophthalmia or other findings of FDH (Brady et al., 2015; Madan et al., 2017; Wawrocka et al., 2021).

Overall, more than 170 unique genomic variants in the PORCN locus linked to FDH have been listed in the Global Variome shared LOVD database (https://databases.lovd.nl/shared/genes/PORCN). Many of these are classified as variants of uncertain significance (VUSs) because they are missense mutations located outside of the known functional features. Determining whether these variants are deleterious can be facilitated by analysis of comparative evolutionary and, where available, structural features, and computationally by online tools such as PolyPhen-2 (Adzhubei et al., 2010), Missense3D (Khanna et al., 2021) and PROVEAN (Choi and Chan, 2015). Laboratory-based assessment of PORCN function, an approach we and others have used previously, is significantly more laborious (Proffitt and Virshup, 2012; Rios-Esteves et al., 2014). Prioritizing which mutants to test in cell-based assays would be facilitated by knowledge of where the variants lie in the PORCN structure. Although the PORCN structure has not been solved, the structures of a bacterial and several mammalian MBOAT proteins were recently determined (Jiang et al., 2021; Long et al., 2020; Qian et al., 2020; Sui et al., 2020; Wang et al., 2020), providing potential templates for a PORCN homology model.

Our interest in modeling the structure of PORCN was catalyzed by the assessment of a patient manifesting FDH phenotype with an uncharacterized PORCN variant. Taking advantage of advances in structural prediction algorithms, we developed a high-quality structural homology model of human PORCN (Sánchez and Šali, 1997; Vyas et al., 2012; White, 2009). This PORCN model is consistent with experimentally determined topology and functional residues, explains the effects of multiple FDH mutations, and predicts the consequences of new human variants of uncertain significance. Molecular docking of WNT8A and palmitoleoyl-CoA (PAM-CoA) in the structure positions the PORCN catalytic histidine 341 adjacent to the thioester bond and the serine in Wnt hairpin 2, providing a model for Wnt acylation. Multiple PORCN inhibitors including ETC-159, LGK974 and IWP-L6 can be docked into the PORCN active site in a manner consistent with an established pharmacophore model (Poulsen et al., 2015). Overall, this structural model provides mechanistic insights into PORCN substrate recognition and catalysis, as well as the pharmacologic inhibition of its enzymatic activity, and can facilitate the development of improved inhibitors and the rationalization of disease-relevant human PORCN variants.

The proband initially presented to the Greenwood Genetic Center (GGC) clinic at 6 weeks of age for evaluation of congenital limb anomalies (Fig. 1B). She also had a history of mild right-branch peripheral pulmonary artery stenosis. Her physical examination was notable for abnormalities of both hands and the right foot. On the left hand, there was a hypoplastic left thumb containing a nail, cutaneous syndactyly of fingers 3–4 extending to the bases of the distal phalanges, and mild medial clinodactyly of the 5th finger. On the right hand, there was a hypoplastic thumb containing a nail, mild radial deviation of the 2nd finger with a slight crease between the bases of fingers 2–3, and a suggestion of syndactyly at the bases of fingers 3–4. The right foot had a medially incurved hallux, absence of toes 2 and 3 with a central cleft, and almost complete syndactyly of toes 4 and 5 with separate nails. The right leg also appeared slightly shorter than the left, and there was mild elevation of the external genitalia and lower buttock on the right side compared to the left. Other minor atypical features seen on physical examination included thinning of the left ear helix, mild asymmetry of the nasal alae, retrognathia and inverted nipples. The hair was thinner on the right parietal scalp than on the left, and no skin abnormalities were noted. A previous eye examination had shown mildly anomalous irises (without colobomas), a minimal congenital nuclear cataract in the right eye and bilateral astigmatism, while subsequent eye examinations showed the additional findings of bilateral nasolacrimal duct obstruction (ultimately requiring dilatation) and small optic nerves, but no evidence of a cataract.

Results from blood chromosome analysis and whole-genome chromosomal microarray were normal. Whole-exome sequencing was performed as previously described (Spellicy et al., 2019), with an initial focused analysis of the TP63 and SALL4 genes, implicated in split-hand/foot malformation, that were normal. Additional analysis of the remaining exome-sequencing data revealed a novel heterozygous p.Tyr334Cys (c.1001A>G) variant of uncertain significance in the PORCN gene located on the X chromosome. Tyr334 is highly conserved in PORCN across diverse species and is only seven residues away from the catalytic His341 (Fig. 1C) (Hofmann, 2000), suggesting that the mutation could be pathogenic. Targeted testing of the patient's mother showed her peripheral blood lymphocytes to be homozygous normal (c.1001A); the patient's father was unavailable for testing. X-chromosome inactivation analysis on the patient was deemed uninformative. Together with the proband's clinical features, her variant in PORCN was felt to be consistent with a diagnosis of FDH, an X-linked dominant condition in which split-hand/foot malformation and other anomalies can occur. Considering this diagnosis, additional evaluations were pursued to screen for other potential findings of FDH. An abdominal ultrasound showed no evidence of renal or diaphragmatic anomalies. An audiologic evaluation was normal, and an ENT evaluation showed no laryngeal papillomas.

This case adds to the many VUSs in PORCN that have been reported in various databases. Prediction of phenotype from genotype is an inexact science and testing every individual VUS in cell-based assays is not widely available. We speculated that a structural model of PORCN would aid in predicting the phenotypic consequences of the VUSs associated with FDH. PORCN is a multipass-transmembrane (TM) protein that localizes to the endoplasmic reticulum (ER), and although its topology has been experimentally established with protein tagging (Galli et al., 2021; Lee et al., 2019), there is currently no experimentally derived structure. The three-dimensional (3D) structures of a bacterial and, more recently, several human MBOAT family members have been determined, providing mechanistic insights into their enzymatic functions (Jiang et al., 2021; Long et al., 2020; Ma et al., 2018; Qian et al., 2020; Sui et al., 2020; Wang et al., 2020). In addition, recent advances in modeling are increasingly capable of producing accurate structures (Yang et al., 2020). We therefore undertook a comparative protein structure-modeling approach leveraging off these new human MBOAT structures.

To select the optimum template for PORCN modeling, we compared sequence alignments of PORCN with the MBOAT proteins with solved structures, including DltB, ACAT1 (also known as SOAT1) and DGAT1 and HHAT (Fig. S1). As a first approach to generate a PORCN structural model, human PORCN sequence was uploaded to several platforms including Phyre2 (Kelley et al., 2015), Rosetta (DiMaio et al., 2011), RaptorX (Källberg et al., 2014), Swiss-model (Waterhouse et al., 2018) and I-TASSER (Yang and Zhang, 2015a,b). These programs selected the bacterial MBOAT DltB (Ma et al., 2018) as the homology template for PORCN due to the highest score of identity (14%). However, none of the simulated structures from these platforms could be docked with PAM-CoA successfully.

We next used MODELLER (Webb and Sali, 2016) to generate 3D structure models for human PORCN by using the MBOAT structures diacylglycerol O-acyltransferase 1 (DGAT1) (Wang et al., 2020) and acyl-coenzyme A cholesterol acyltransferase (ACAT1) (Long et al., 2020; Qian et al., 2020) as templates. The C-terminal 40 amino acids of PORCN lack similarity to the other MBOATs and were therefore refined using I-TASSER, which is ideal for modeling sequences of low homology via thread methodology (see Materials and Methods) (Yang and Zhang, 2015b). The homology structure based on DGAT1 (Fig. 2) was deemed to be of highest quality, with 92.2% of residues in favored and another 6.1% in allowed Ramachandran regions. In addition, the structure could readily be docked with PAM-CoA and WNT8A (discussed below) and was therefore further evaluated.

Overall consistent with the experimental data of Lee et al. (2019) and Galli et al. (2021), the predicted PORCN model has ten TM helices with both the N- and C-termini in the cytoplasm. Superimposition of the PORCN structure with DGAT1 showed a root-mean-square deviation (RMSD) of only 2.71 Å (Cα backbone) (Fig. 2A,B). The PORCN structure also has similar TM structure and folds to DltB, HHAT, ACAT1 and GOAT (also known as MBOAT4) (Campaña et al., 2019; Jiang et al., 2021) (Table 1; Fig. S2A). The catalytic histidine (His341) of PORCN is located in the middle of a tunnel, in a similar position to the catalytic residue His415 in DGAT1 (Fig. 2A, inset). The model with the topology reported by the fluorescence protease protection studies of Lee et al. (2019) is shown in Fig. 2C.

As an independent approach to evaluate the quality of the model, we used it to map two features of PORCN evolution: amino acid conservation and amino acid co-evolution. Residues critical for the function of a protein tend to be conserved, and, indeed, mapping sequence conservation determined with ConSurf (Ashkenazy et al., 2016) onto the structure of PORCN showed clustering of highly conserved residues in both the catalytic core and the presumed Wnt and PAM-CoA binding pockets (Fig. S2B). Additional useful information can be deduced from how amino acid networks co-evolve. This is based on the principle first recognized by Darwin, that biological structures (or in the case of proteins, residues) that are functionally linked tend to evolve in tandem (reviewed in Göbel et al., 1994; de Juan et al., 2013; Ovchinnikov et al., 2015). Given the large number of sequences available, we used computational methods to visualize PORCN co-evolving blocks. Analysis of 468 PORCN amino acid sequences from species as diverse as sponge and human using the Leri server identified several co-evolving blocks (Cheung et al., 2021). The most highly co-evolving residues, indicated by the blue-outlined block in Fig. S2C (and see example in Fig. S2D), when mapped on to the PORCN structure (Fig. 2D) highlight the inner catalytic core of the enzyme, supporting the functional accuracy of the PORCN structure prediction.

PORCN is an O-acyltransferase that preferentially catalyzes the transfer of the cis-Δ9 unsaturated palmitoleate (C16:1) to a conserved serine on hairpin 2 of Wnts. Based on homology to ACAT1 and DGAT1, we examined the PORCN structure for the presence of tunnels that might accommodate these ligands. Indeed, CAVER software highlighted two tunnels, denoted 1 and 2, in the TM domain that open to the ER lumenal and cytoplasmic sides of the membrane, respectively (Fig. 3A). Supporting their relevance, the residues in PORCN that are most highly conserved (Ashkenazy et al., 2016) form the walls of these contiguous structural tunnels 1 and 2 (Fig. 3A).

The structure and cavity analysis suggests that the Wnt acylation reaction proceeds via PAM-CoA binding to tunnel 2 via the cytoplasmic face of PORCN, and the acceptor hairpin 2 of Wnt entering the catalytic PORCN core via tunnel 1 from the ER lumen. Prior biochemical analysis suggests that PORCN His341 serves as a base catalyst to facilitate the transfer of the palmitoleoyl group from CoA to the serine of Wnt hairpin 2 (Lee et al., 2019; Nile and Hannoush, 2016; Proffitt and Virshup, 2012; Rios-Esteves et al., 2014). To examine whether our model can provide a mechanistic rationale for the molecular events during Wnt signaling, we docked PAM-CoA into the homology model of PORCN (Fig. 3B–E; Fig. S3B). The CoA moiety anchors in a pocket on the cytoplasmic face (Fig. 3C), similar to how it interacts with ACAT1 (Qian et al., 2020), DGAT1 (Sui et al., 2020; Wang et al., 2020) and HHAT (Jiang et al., 2021). The PAM-CoA thioester bond lies at the base of tunnel 1 (Fig. 3B). The unsaturated fatty acid chain extends into tunnel 2, lined by conserved hydrophobic residues (Fig. 3D,E; Fig. S3C). This tunnel has a kink lined by PORCN Asn306 and Met309 that accommodates the cis double bond (Fig. S3D). We note that Asn306 is invariant in PORCN and has previously been proposed to assist in catalyzing the transesterification reaction. However, mutation of Asn306 has no effect on the catalytic activity of PORCN in cell-based assays (Proffitt and Virshup, 2012; Rios-Esteves et al., 2014) but markedly decreased activity in a purified system in which only PAM-CoA was available as a substrate (Lee et al., 2019). The structural model thus suggests that Asn306 could enforce the use of cis double bond-containing fatty acid substrates (Tuladhar et al., 2019). We also tried to dock the saturated palmitoyl-CoA into the PORCN tunnel 2. However, no biologically plausible results were obtained. Two representative failed docking results are shown in Fig. S3E,F. In both cases, the thioester bond in palmitoyl-CoA is far from the catalytic His341 residue in PORCN. This is consistent with our current understanding that PORCN utilizes unsaturated PAM-CoA instead of the saturated palmitoyl-CoA as its substrate (Rios-Esteves and Resh, 2013; Tuladhar et al., 2019).

We next examined how a Wnt might interact with PORCN. WNT8A [extracted from the WNT8A/WLS structure, Protein Data Bank (PDB) 7KC4] (Nygaard et al., 2021) was docked into the PORCN model using the ClusPro 2.0 Server (Fig. 3B; Fig. S3B). Supporting our hypothesis, the predicted complex shows WNT8A docking into tunnel 1 from the lumenal face of PORCN, with the hairpin 2 loop inserted deeply into the tunnel (Fig. 3B). Interestingly, WNT3 from the WNT3–mFZD8 CRD complex (PDB 6AHY) did not dock hairpin 2 as deeply into the tunnel due to interference from Wnt hairpin1. Unlike WNT8A in complex with WLS, in the WNT3–mFZD8 CRD complex hairpin 1 is tightly apposed to hairpin 2 (Hirai et al., 2019; Nygaard et al., 2021). When the modeled complex of WNT8A–PORCN was superimposed with the PORCN–PAM-CoA model, the hairpin 2 loop of WNT8A is in close contact with the thioester of the PAM-CoA (Fig. 3E). During Wnt acylation, PORCN catalyzes the transfer of the palmitoleoyl moiety from CoA to the conserved serine residue on Wnt hairpin 2 (Ser186 of WNT8A). In the modeled ternary complex, Ser186 is positioned 4.3 Å from the thioester bond of the PAM-CoA, and the conserved catalytic residue His341 from PORCN is well positioned to activate the hydroxyl group of Ser186 from WNT8A for nucleophilic attack on the thioester group of the substrate (Fig. 3E). Asn306, another highly conserved residue in PORCN, while close to the PAM-CoA, has a less clear role in catalysis (Lee et al., 2019; Proffitt and Virshup, 2012; Rios-Esteves et al., 2014). The structural prediction of PORCN docked with PAM-CoA and WNT8A provides a coherent and useful model to understand Wnt acylation.

Resh and co-workers reported an extensive mutational analysis of PORCN (Rios-Esteves et al., 2014). We examined where mutants that affected PORCN activity in their assays were in our predicted structure. Trp305 interacts with the thioester of PAM-CoA, Tyr316 stabilizes the interaction of IL2 with TM4, Tyr334 is altered in our patient and is discussed below, Ser337 interacts with the CoA moiety, and Leu340 is adjacent to the catalytic H341 and contacts Wnt hairpin 2 (Fig. S3G). Thus, these engineered loss-of-function mutants can also largely be explained by the predicted PORCN complexes.

The above analyses demonstrate that the PORCN homology model is predictive of function. Next, we assessed whether the model can be used to predict the consequences of PORCN sequence variants found in patients (Fig. 4A). First, we asked whether the Y334C variant had damaging consequences. Tyr334 is located on TM helix 7, the same helix as the catalytic residue His341, and appears to stabilize the PORCN catalytic core by interacting with multiple hydrophobic residues in TM8 (Val350, Leu354, Ile357) and on the membrane side of intracellular loop 2 (Val301 and Val302) (Fig. S4A). Consistent with this, the Missense3D server (http://missense3d.bc.ic.ac.uk/missense3d/) (Khanna et al., 2021) predicts that the Y334C substitution significantly alters the internal cavity of the protein. To experimentally test the stability and Wnt/β-catenin signaling activity of this PORCN variant, we used our previously described PORCN knockout HT1080 cell line. This is a human male fibrosarcoma cell line in which an introduced frameshift mutation inactivates the PORCN gene so that the cells neither palmitoleate nor secrete Wnt ligands (Proffitt and Virshup, 2012). Ectopic expression of PORCN rescues Wnt secretion and can therefore be used to test the activity of various PORCN mutants. Consistent with the structural and computational prediction, PORCN Y334C has decreased protein expression in cell lysates compared to wild-type (WT) PORCN (Fig. 4B). In the presence of PORCN Y334C, significantly less WNT3A was secreted into the culture medium compared to in the control (Fig. 4B), and the mutant also had decreased activity in a functional Wnt/β-catenin signaling assay (Fig. 4C,H). This instability is not due to the presence of the cysteine residue, as Rios-Esteves et al. (2014) reported that PORCN Y334A also had decreased stability and decreased activity. These findings demonstrate that VUS PORCN Y334C is a deleterious mutation responsible for the patient's disease and support the value of the PORCN model to predict functional consequences.

From the Global Variome shared LOVD database, we identified nine PORCN missense VUSs. These variants fall in diverse regions of the predicted protein structure similar to what has been recently proposed (Galli et al., 2021) (Fig. 4A). We examined the expression and activity of four of the VUSs, including V149E, G163S, M123R and T265M, that have not previously been biochemically characterized. G163S appeared to be unstable (Fig. 4D). Glycine in TM loops has been reported to facilitate helix packing (Javadpour et al., 1999), and indeed Gly163 lies on intracellular loop 1 where the opposite face packs tightly with TM helices 2 and 4 (Fig. S4B). M123R was also unstable (Fig. 4D). Met123 lies on the ER lumenal end of TM helix 4 interacting with two methionines (Met109 and Met112) on TM3, potentially forming a methionine triad, a structure previously described at the ends of TM loops in the copper transporter CTR1 (Ren et al., 2019) (Fig. S4C). T265M, which decreases PORCN enzymatic activity without a change in protein stability, has potential to disrupt the binding of PAM-CoA to PORCN as it lies at the end of tunnel 2, adjacent to the end of the fatty acid chain (Fig. S4D). Both M123R and T265M are predicted as possibly or probably damaging by both PolyPhen-2 (Kelley et al., 2015) and Missense 3.0. V149E, which also has decreased expression, was not predicted by PolyPhen-2 and Missense 3.0 to have deleterious consequences for this mutant. Val149E replaces a conserved hydrophobic residue with a charged residue in the cytoplasmic domain, distant from the catalytic core (Fig. 4A). We speculate that this impairs protein stability due to altered protein–protein interactions of the cytoplasmic loop. Consistent with the model prediction as well as the observed protein abundance, all the mutants have reduced Wnt/β-catenin signaling activity in both the autocrine and paracrine reporter assays (Fig. 4E,H).

Goltz syndrome is an X-linked dominant disorder, and it mainly occurs in females. However, male patients with PORCN mutations who do not display typical FDH symptoms have been reported (Brady et al., 2015; Madan et al., 2017; Wawrocka et al., 2021). In particular, one family of brothers (hemizygous Y245C) had anophthalmia or microphthalmia (Wawrocka et al., 2021), which is not typically seen in FDH patients. We analyzed two such mutations, Y245C (possibly damaging) and S250F (probably damaging), by Polyphen-2 (Madan et al., 2017). However, both mutants have very low expression levels, as assessed by western blotting (Fig. 4F), and thus reduced autocrine reporter activity (Fig. 4G). Tyr245 on TM helix 6 packs tightly with Leu214 and Phe218 on TM5 (Fig. S4E); its replacement by Cys may disrupt this network. S250F is predicted to be probably damaging. It lies in the catalytic core, and the serine is part of a network with His409 of TM9 and Wnt hairpin 2 (Fig. S4F). Replacement by the more bulky and hydrophobic phenylalanine may disrupt this network. To further differentiate between loss of catalytic activity and more substantial misfolding, we assessed the ability of the mutant PORCN proteins to co-immunoprecipitate WNT3A (Fig. 4I). Note that in this assay, we altered the amounts of plasmid transfected to equalize the protein expression. Mutants Y334C, M123R and V149 retained WNT3A binding, while G163S and others had markedly diminished binding, consistent with more profound effects on PORCN structure. In all cases, the mutant proteins retained ER localization, similar to the WT protein (Fig. S4G). Again, elucidating the structure of PORCN assists in understanding the structural consequences of patient-associated variants.

Pharmacologic inhibition of PORCN may be useful in Wnt-high disorders such as cancer, and a number of candidates have shown robust activity in various model systems (Zhong and Virshup, 2020). We examined whether our model could elucidate the binding modes of potent small-molecule inhibitors of PORCN. To this end, we used the standard precision mode of GLIDE (Friesner et al., 2006) to dock the PORCN inhibitors ETC-159 (Madan et al., 2016), Wnt-C59 (Proffitt et al., 2013), LGK-974 (also called WNT974) (Liu et al., 2013), IWP-2 and its derivative IWP-L6 (Wang et al., 2013) as well as ETC-131 in the homology model [see Table 2 for chemical structures and half-maximal inhibitory concentrations (IC₅₀) of the drugs]. Notably, all the active inhibitors adopt an inverted ‘L’ shape and occupy the same binding site within the TM core, partially overlapping with the PAM-CoA binding (Fig. 5A,B; Fig. S5A–F). The biaryl moiety (colored red in Table 2) of the inhibitors sits deeply in the binding cavity (Fig. 5B,D). In Wnt-C59, LGK-974 and ETC-159, the aromatic ring at the bottom is surrounded by several hydrophobic residues including Val302, Val350, Leu405 and, notably, Tyr334 (Fig. 5D; Fig. S5A,B). These predicted binding modes also put the cap xanthine ring (colored blue in Table 2) in conflict with Wnt hairpin 2 in the catalytic core of PORCN (Fig. 5A). Testing this, we find that ETC-159 indeed blocks the binding of WNT3A to PORCN (Fig. 5C). This decreased binding is not simply due to a lack of palmitoleation, because Wnts lacking the target serine, and catalytically inactive PORCN H341 binds WNT3A like WT (Fig. 5C) (Rios-Esteves et al., 2014).

We further tested the predictions of the model by mutating several of the residues involved in close contacts with inhibitors. Remarkably, and strongly supporting the accuracy of the model, mutating the Val302 to Ala, or Tyr334 to either Cys (i.e. the patient mutation) or Ala, made PORCN resistant to inhibition by ETC-159 (Fig. 5E,F; Fig. S5G). This is not a general feature of disease-associated variants, as Ser250 mutated to Phe, found in two males with FDH, did not confer drug resistance (Fig. 5E) (Madan et al., 2017). Previous structure–activity relationship (SAR) studies indicated that ETC-675, a close analog of ETC-159 that does not have the bottom aromatic ring, completely lost its inhibitory effect (IC₅₀>1 μM) in the TOPFlash assay (Table 2). These results, taken together, suggest that the interactions of the aromatic ring 1 with residues including Tyr334 and Val302 at the bottom of the binding site are indispensable for the inhibitors’ activity.

Towards the ER lumenal side of the catalytic pocket, the second aromatic ring (ring 2 in Fig. 5D; colored red in Table 2) of the active inhibitors is wedged in between Trp305 and His341 (Fig. 5D,G; Fig. S5A,B,F). The carbonyl oxygen of the amide linker was predicted to form a hydrogen bond with the imidazole NH of His341 (Poulsen et al., 2015). The cap xanthine ring of ETC-159 is sandwiched between Phe253 and His409 and forms π-stacking interactions with these two residues (Fig. 5D). Overall, the superimposed alignment of the predicted inhibitor binding modes agrees with a previously reported ligand-based pharmacophore model of the PORCN inhibitors (Poulsen et al., 2015). In contrast, our model predicted that ETC-131, IWP-2 and IWP-L6 bind at a position slightly higher than the docking site of the other inhibitors (Fig. 5G; Fig. S5D–F). If the binding mode is accurate, the bottom thiophene ring of ETC-131 will be further away from PORCN Tyr334 than the aromatic ring 1 of ETC-159. In other words, mutation of Tyr334 should have less effect on drug sensitivity for ETC-131 than ETC-159. We therefore tested this hypothesis experimentally. As shown in Fig. 5H, although Y334C is completely resistant to ETC-159, it is still sensitive to ETC-131, supporting the prediction based on the model. ETC-131 has a methyl group (S-isomer) attached to the carbon atom between the amide linker and the xanthine ring. Based on the binding mode of ETC-159, we observed that an extra methyl is difficult to accommodate due to the presence of two bulky residues, Phe257 and Trp305, next to the carbon atom between the amide linker and the xanthine ring. These findings may explain the slight difference of the binding mode as adopted by ETC-131 (Fig. 5G; Fig. S5D). More interestingly, molecular docking of ETC-130 (the R-isomer of ETC-131) positioned the ligand in the binding site with an upside-down flip orientation (Fig. S5C), which agrees with the experimental result showing that ETC-130 is inactive in the TOPFlash assay (Madan et al., 2016). Despite containing a biaryl moiety that can insert deeply into the binding cavity of PORCN, IWP-L6 (a derivative of IWP-2) (Chen et al., 2009; Wang et al., 2013) has a much bulkier group at the lumenal side of the protein binding tunnel. Not surprisingly, the docking results positioned IWP-L6 in a similar binding location relative to ETC-131 (Fig. S5F). Collectively, the modeling results strongly indicate that the model can provide useful insights and predictions on the molecular interactions between PORCN and its inhibitors.

PORCN is the sole acyltransferase responsible for the palmitoleation of all Wnts. It is a promising drug target for Wnt-addicted cancers, and several PORCN inhibitors are currently in clinical trials. Loss-of-function mutations in PORCN cause FDH predominantly in females due to loss of Wnt signaling during development. However, X-chromosome inactivation as well as the large number of alleles cause a wide range of manifestations of FDH, making both the clinical diagnosis and the classification of new VUSs challenging. Here, building on recent structures of other human MBOAT proteins and as part of an effort to characterize a VUS in PORCN, we developed a structural homology model. The usefulness of the model is supported by its ability to predict the docking of WNT8A and PAM-CoA, reveal the consequences of the patient's Y334C mutation, assist in understanding and classifying other VUSs, and, finally, assist in understanding the SAR of multiple small-molecule inhibitors of PORCN.

The PORCN model was templated on the recently published DGAT1 cryogenic electron microscopy (cryo-EM) structures. DGAT1 catalyzes the transfer of the acyl group from oleoyl-CoA (cis-C18:1) to cholesterol to form cholesterol esters. Like PORCN, DGAT1 prefers unsaturated fatty acids to saturated acyl groups. (Asciolla et al., 2017; Chang et al., 2010; Rios-Esteves and Resh, 2013; Tuladhar et al., 2019). The accuracy of the PORCN model is demonstrated by our ability to dock palmitoleoyl- but not saturated palmitoyl-CoA into the PORCN tunnel 2. We note that the kinked acyl chain tunnel was also observed in the structures of stearoyl CoA desaturase (SCD), the enzyme that desaturates stearoyl and palmitoyl-CoA, and Notum, an esterase that cleaves the palmitoleoyl moiety from Wnts (Bai et al., 2015; Kakugawa et al., 2015). In PORCN, tunnel 2 accommodating the palmitoleoyl chain is surrounded by both highly conserved and co-evolving residues (Fig. 3E), suggesting that this recognition of a mono-unsaturated substrate is important throughout the animal kingdom.

Identifying a correct and optimal template for building a 3D model backbone is the first and the most critical step in the process of homology modeling. Templates for modeling PORCN were initially selected by automated homology modeling programs like Phyre2 based on sequence alignment. The bacterial MBOAT protein DltB was first used due to slightly higher identity compared to other human MBOATs (14% for DltB and 13% for DGAT1 in Phyre2). However, the tunnels in DltB that accommodate small D-alanyl groups, are very different from those in the human MBOAT proteins (Fig. S3A). Furthermore, DltB is on the bacterial plasma membrane whereas PORCN, ACAT1 and DGAT1 are located on the mammalian ER membrane, and bacterial plasma and mammalian ER membranes have very different compositions (Holthuis and Menon, 2014). Thus, the homology model based on DltB appeared less predictive than that based on DGAT1.

The predicted structure also suggests a pathway for Wnts to be acylated by PORCN and then handed off to the carrier protein WLS in the ER membrane. In the model, Wnt hairpin 2 enters tunnel 1 from the lumenal face of the ER. Wnt hairpins 1 and 3 make additional contacts with the lumenal loops of PORCN to stabilize the interaction. Following PAM-CoA docking, the catalytic His341 activates WNT8A Ser186 for nucleophilic attack on the thioester bond. The next step, the transfer of acylated Wnt from PORCN to WLS, we speculate is direct (Yu et al., 2014). In both the WNT8A–PORCN complex (here) and the WNT8A–WLS complex (Nygaard et al., 2021), Wnt hairpin 2 and the attached palmitoleoyl group are inserted into the TM domain. This suggests that the most energetically favorable route for acylated Wnt is a lateral transfer of the hairpin 2 PAM from PORCN to WLS within the plane of the membrane. Inspecting the ternary PORCN–WNT8A–PAM-CoA complex, a possible route for Wnt transfer is to slide between PORCN TM helices 3 and 4 on one side, and TM10 on the other. Consistent with this mode of transfer, we note that the conformation of WNT8A used here was essentially unchanged from its conformation bound to WLS (Nygaard et al., 2021), and that WNT3 in its mFZD8 CRD-bound conformation (PDB 6AHY) could not dock to PORCN in a reasonable position (Hirai et al., 2019). This prediction of WNT lateral transfer from PORCN to WLS may be testable in future studies.

Although the current PORCN model has demonstrated its applicability in providing an overall framework of both Wnt palmitoleation and inhibitor binding, it has several caveats that warrant further cautions. First, the protein–protein docking model was not able to predict the bona fide binding geometry between the catalytic His341 and the Ser186 on the hairpin loop 2 of WNT8A, despite a distance of 6.3 Å between the NH of the imidazole from His341 and the hydroxyl group of Ser186 from WNT8A. The model did not fully explain the role of Asn306, which has been proposed to aid catalysis, but the mutation of which did not affect function in two cell-based assays. Finally, our in-house SAR analysis and a previous pharmacophore model of PORCN inhibitors indicated that the oxygen atom from the xanthine ring plays a role as a hydrogen bond acceptor during protein–ligand interaction (Poulsen et al., 2015). Our model failed to identify the exact protein residue that forms the hydrogen bond with the oxygen from the xanthine group. These uncertainties may arise from the availability of limited choice of template for the modeling, as well as the low sequence homology between the localized protein motifs. However, it is also possible that the pharmacophore model was too simplistic, assuming that the biphenyl substituents of all inhibitors are perfectly superimposable in the active site. Atomic resolution structures of MBOAT protein family members have only been available recently. Future work focusing on extensive model refinement with the availability of new structures deposited in the public PDB, together with the aid of site-directed mutagenesis, should facilitate the improvement of the model. Meanwhile, we believe that an experimentally determined 3D structure of the Wnt-PORCN complex or an inhibitor-bound PORCN structure will address the uncertainties presented herein.

The patient was seen at the GGC, Greenwood, SC, for genetic evaluation related to congenital limb anomalies. Informed consent was signed by the parent of the proband prior to participation in the research study. Consent included permission to publish clinical information and photographs. All procedures employed were in accordance with study parameters approved by the Institutional Review Board, and compliant with practices at the GGC. All clinical investigation was conducted according to the principles expressed in the Declaration of Helsinki.

HT1080 PORCN null cells were previously reported (Proffitt and Virshup, 2012). STF reporter cell line (HEK293 with stably expressing firefly luciferase gene under the control of eight tandem repeats of the TOPFlash TCF/LEF responsive promoter element) was a gift from Kang Zhang (University of California San Diego, La Jolla, CA) (Xu et al., 2004). Mycoplasma test was performed regularly using MycoAlert Mycoplasma detection kit (Lonza). Cells were cultured in Dulbecco's modified Eagle medium supplemented with 10% fetal bovine serum (FBS), L-glutamate, penicillin/streptomycin and sodium pyruvate. Super8xTopflash reporter was a gift from Randy Moon (University of Washington, Seattle, WA, USA). pGK-WNT3A plasmid was provided by Karl Willert (University of California, San Diego, CA, USA). pMKIT-3xHA-mPORCN-D was a gift from Tatsuhiko Kadowaki (Nagoya University, Japan) and was described previously (Proffitt and Virshup, 2012; Tanaka et al., 2000). This plasmid was used to introduce all patient-derived mutations using site-directed mutagenesis following the manufacturer's protocol (Stratagene), and the sequences were verified by Sanger sequencing. Anti-V5 antibody (clone SV5-Pk1) was from Bio-Rad (MCA1360) and used at 1:5000 dilution for western blotting. Antibody against WNT3A was a gift from Shinji Takada (National Institutes of Natural Sciences, Okazaki, Japan) as mouse hybridoma supernatant and used at 1:100 dilution. Anti-HA tag rabbit monoclonal antibody (C29F4) was from Cell Signaling Technology (#3724) and used at 1:2000 dilution. Lipofectamine 2000 was used for transfections. ETC-159 and ETC-131 were prepared as 1 mM stock in dimethyl sulfoxide (DMSO) and stored at −30°C.

To monitor Wnt secretion into the culture medium, cells were grown in 12-well plates and transfected with 500 ng of WNT3A expression plasmid, along with 50 ng and 100 ng of expression plasmids encoding 3xHA-tagged WT or mutant mouse PORCN-D as indicated using Lipofectamine 2000 (Thermo Fisher Scientific). Twenty-four hours after transfection, the medium was changed to low serum (2% FBS) at half volume (0.5 ml). For Wnt secretion experiments, medium was collected after another 24 h and subjected to western blot analysis. For SuperTOPFlash assays, experiments were performed in 24-well plates, transfecting 200 ng SuperTOPFlash plasmid together with 50 ng of WNT3A, 100 ng of mCherry and up to 20 ng of PORCN expression plasmids. Twenty-four hours after transfection, luciferase activity was quantitated as previously reported (Yu et al., 2020). For each experiment, a negative control of only transfecting the WNT3A without any exogenous PORCN was included. This served as the basal level of luciferase activity in the cells and was set to 1. The relative TOPFlash activity of the PORCN (WT or mutant) is the fold change from the basal level. If Wnt inhibitors were used, the cells were treated with the indicated inhibitors beginning 6 h after transfection and incubated for another 18 h before harvesting for the reporter assay.

Homology modeling was performed using the MODELLER (Webb and Sali, 2016) program from trRosetta (Yang et al., 2020) and (PS)² (Huang et al., 2015) by query-template alignment. Two recently published human MBOAT protein structures, DGAT1 (PDB 6VP0) and ACAT1 (PDB 6P2P), were chosen as templates to guide the modeling with the human PORCN amino acid sequence (UniProt Q9H237-1/NCBI Q9H237.2, also known as splice isoform D, 461 amino acids), using the T-Coffee homology extension (PSI-coffee) algorithm (di Tommaso et al., 2011). To refine the PORCN simulation, I-TASSER (Yang and Zhang, 2015b) was used by assigning the template generated from trRosetta and (PS)² with specified secondary structure for the C-terminal 41 amino acids (Val421 to Gly461). Lastly, the simulated structure was refined by GalaxyRefine2 from GalaxyWEB (Afgan et al., 2018; Boekel et al., 2015) to generate the final predicted structure.

Protein tunnels were identified and assessed using Caver 3.03 (Pavelka et al., 2016) as a Pymol Plugin using a probe radius of 2 Å, shell radius of 4 Å and shell depth of 3 Å. The clustering threshold was set at 3.5. The starting point was defined as the point between His341, Asn306 and Phe257.

System preparation and receptor–ligand docking calculations were performed using the Schrödinger Suite package (version 2020–4), using default parameters unless otherwise noted. The homology model of human PORCN as a receptor was first prepared using Protein Preparation Wizard (Sastry et al., 2013). The atom and bond types were corrected and the protonation states of ionizable species adjusted to pH 7.4 by Epik (Greenwood et al., 2010). During receptor grid generation, the ligand to be docked was confined to an enclosing box centered on the middle of the tunnel, using Ser262, Glu300, Asn315 and His341 to set the box center with an external box of 36 Å for PAM-CoA docking, and a ligand-sized box centered on Glu293, Val302 and His409 for inhibitor docking. The PAM-CoA structure and PORCN inhibitors were generated from PubChem, prepared using the LigPrep module to convert the 2D sdf format into 3D molecular structures, and docked to the orthosteric site of PORCN with the Glide program (Friesner et al., 2006) using the standard precision (SP) mode. The best binding pose predicted for human PORCN-PAM-CoA complex and PORCN-inhibitors docking was saved for further analysis for molecular interactions.

To predict protein–protein interactions, the structure of WNT8A was extracted from the WLS/WNT8A structure (PDB 7KC4) (Nygaard et al., 2021), and the ClusPro 2.0 Server (Kozakov et al., 2017) was used to predict its interaction with PORCN. We used the hydrophobic-favored potential in the server. A near-native state of protein conformations was chosen during protein–protein docking. A model of the PORCN–WNT8A–PAM-CoA structure is available upon request.

The authors thank members of the Virshup laboratory, including Babita Madan for discussions on PORCN structure and potential enzymatic mechanisms, and Yunka Wong for technical assistance. We had invaluable assistance with Leri from Ngaam J. Cheung (aka Yaan), University of Oxford.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.259383.