Pathogenic FAM83G palmoplantar keratoderma mutations inhibit the PAWS1:CK1α association and attenuate Wnt signalling.

Background: Two recessive mutations in the FAM83G gene, causing A34E and R52P amino acid substitutions in the DUF1669 domain of the PAWS1 protein, are associated with palmoplantar keratoderma (PPK) in humans and dogs respectively. We have previously reported that PAWS1 associates with the Ser/Thr protein kinase CK1α through the DUF1669 domain to mediate canonical Wnt signalling. Methods: Co-immunoprecipitation was used to investigate possible changes to PAWS1 interactors caused by the mutations. We also compared the stability of wild-type and mutant PAWS1 in cycloheximide-treated cells. Effects on Wnt signalling were determined using the TOPflash luciferase reporter assay in U2OS cells expressing PAWS1 mutant proteins. The ability of PAWS1 to induce axis duplication in Xenopus embryos was also tested. Finally, we knocked-in the A34E mutation at the native gene locus and measured Wnt-induced AXIN2 gene expression by RT-qPCR. Results: We show that these PAWS1 A34E and PAWS1 R52P mutants fail to interact with CK1α but, like the wild-type protein, do interact with CD2AP and SMAD1. Like cells carrying a PAWS1 F296A mutation, which also abolishes CK1α binding, cells carrying the A34E and R52P mutants respond poorly to Wnt signalling to an extent resembling that observed in FAM83G gene knockout cells. Consistent with this observation, these mutants, in contrast to the wild-type protein, fail to induce axis duplication in Xenopus embryos. We also found that the A34E and R52P mutant proteins are less abundant than the native protein and appear to be less stable, both when overexpressed in FAM83G-knockout cells and when knocked-in at the native FAM83G locus. Ala 34 of PAWS1 is conserved in all FAM83 proteins and mutating the equivalent residue in FAM83H (A31E) also abolishes interaction with CK1 isoforms. Conclusions: We propose that mutations in PAWS1 cause PPK pathogenesis through disruption of the CK1α interaction and attenuation of Wnt signalling.

The first clue to possible physiological functions of PAWS1 came in 2013 from a 'woolly' mouse phenotype, in which a large deletion of the FAM83G gene (probably resulting in a severely truncated protein) was linked to a rough and matted appearance of the coat 4 . No further studies analysing biochemical or other possible phenotypic abnormalities in these mice have been reported. Another study reported a single homozygous missense mutation in the FAM83G gene (c.155C>G), which results in the substitution of a conserved arginine into proline (p.R52P) in the PAWS1 protein. This is the causative genetic defect for hereditary footpad hyperkeratosis (HFH), an autosomal recessive disease affecting several dog breeds, in which gradual thickening of the footpad epidermis leads to the development of painful cracks and fissures. Dogs with HFH also exhibit a softer, duller coat appearance 5,6 , reminiscent of the "woolly" mouse phenotypes. The HFH phenotypes also occur in human patients and are broadly described as palmoplantar keratodermas (PPK), which represent a group of skin conditions characterised by thickening of the skin on the palms of the hands and soles of the feet. PPKs often arise from mutations in genes encoding for the keratin cytoskeleton or cell junctions, although there are many cases for which the molecular basis has yet to be established 7,8 . Recently, a study reported a single homozygous missense mutation in the FAM83G gene (c.101C>A), which results in a substitution of a conserved alanine into glutamate (p.A34E) on the PAWS1 protein. The two human patients were siblings and both presented with palmoplantar keratoderma and thick, exuberant scalp hair 9 . Both A34E (human) and R52P (dog) mutations in PAWS1 lie within the conserved DUF1669 domain. The high degree of similarity between the phenotypes seen in mice, dogs, and humans provides compelling genetic evidence for the involvement of PAWS1 in skin and hair homeostasis.
In the last few years, we have made progress in understanding the biochemical functions and regulation of PAWS1 and other FAM83 proteins. We discovered that FAM83 proteins, through their DUF1669 domains, interact with distinct sets of CK1α, δ, and ε isoforms to direct them to distinct subcellular locations, thereby, perhaps, regulating the diverse roles of CK1 isoforms 10 . In particular, we found that PAWS1 interacts with CK1α and that this interaction is essential to promote canonical Wnt signalling in human cell lines and Xenopus embryos by accentuating the nuclear accumulation of β-catenin 1 . In the nucleus, β-catenin forms a complex with TCF/LEF transcription factors to activate Wnt-dependent target gene expression 11 . Because the Wnt signalling pathway plays crucial roles at several stages of epithelial and hair development (reviewed in 12,13), we asked whether the pathogenic palmoplantar keratoderma effects of the PAWS1 mutations might be due to dysregulation of Wnt signalling or, alternatively, to other activities of the PAWS1 protein. These other activities include the ability of PAWS1 to associate with the transcription factor SMAD1 to control a subset of non-canonical bone morphogenetic protein (BMP)-induced gene transcription, as well as its ability to interact with the SH3 adaptor CD2AP to regulate actin cytoskeleton remodelling 14,15 .

Retroviral transduction of cells for the stable expression of target proteins
Retroviral pBabe-puromycin vectors encoding GFP or the desired target protein (6 μg) were co-transfected with pCMV5gag-pol (3.2 μg, Cell Biolabs, RV-111) and pCMV5-VSV-G (2.8 μg, Cell Biolabs, RV-110) into a 10 cm-diameter dish of ~70% confluent HEK293-FT cells. Briefly, plasmids were added to 1 ml Opti-MEM medium (Thermo Fisher Scientific, 31985062) to which 24 μl of 1 mg/ml polyethylenimine (PEI; diluted in 25 mM HEPES pH 7.5) was added. Following a brief vortex mix and incubation at room temperature for 20 min, the transfection mix was added dropwise to the HEK293-FT cells. 16 h post-transfection, fresh medium was added to the cells. 24 h later, the retroviral medium was collected and passed through 0.45 μm filters. Target PAWS1-KO HaCaT or U2OS cells (~60% confluent) were infected with the optimised titre of the retroviral medium diluted in fresh medium (typically 1:5 -1:10) containing 8 μg/ml polybrene (Sigma, H9268) for 24 h. The retroviral infection medium was then replaced with fresh medium, and 24 h later, the medium was again replaced with fresh medium containing 2 μg/ml puromycin (Sigma, P9620) for selection of cells which had integrated the rescue constructs. until the 4-cell stage was reached. Embryos were then injected into a single ventral blastomere with 500 pg of the indicated capped RNA, synthesised using SP6 mMessage mMachine kit (Invitrogen, AM1340), in a total volume of 5 nl. Embryos were then allowed to develop to approximately stage 34-35 at 21°C, before being fixed in 4% paraformaldehyde (PFA). Embryos were then counted and scored for the secondary axis phenotype: secondary axis complete with 2 × cement gland = complete secondary axis; secondary axis apparent but 1 × cement gland = partial secondary axis; enlarged cement gland/rostral structures = dorsalised; comparable to WT = WT. As phenotypes were distinct, no blinding/randomisation was undertaken. For each experiment, approximately 35-40 embryos were injected to ensure sufficient statistical power and a similar number of uninjected embryos were kept under the same conditions as controls. The experiment was repeated three times, twice on the same day (morning and afternoon) using eggs from two different females and testes from the same male, the third experiment was undertaken on a separate day using eggs from a third female and testes from a second male.

Mass spectrometry
For western blotting, embryos were obtained as described above and injected with 500 pg of the indicated capped RNA into the animal hemisphere at the one-cell stage. Embryos were allowed to develop to stage 10 at 18°C before being snap frozen on dry ice and stored at -20°C for later protein extraction.

Statistical analysis
Graphing and statistical tests were performed using Prism 7 software (GraphPad). Unless otherwise noted, data are presented as the mean ± standard deviation of at least three biological replicates. Specific tests used are described in the respective figure legends. Significance levels are as follows: *P<0.05, **P<0.005, ***P<0.001. See underlying data for data underlying all presented figures 18 .
PPK phenotypes are associated with abnormal epidermis and often result from epidermal hyperplasia 7,19,20 . To investigate the impact of the PPK mutants in a physiologically relevant cell line model, we used CRISPR/Cas9 genome editing to generate PAWS1-knockout (KO) HaCaT cells, which are a spontaneously transformed human keratinocyte cell line ( Figure 1C & Figure S1 (extended data 18 )). These cells were then stably restored with near-endogenous levels of wild-type PAWS1, or with equivalent levels of the two pathogenic mutants (A34E & R52P) and two CK1-interaction deficient mutants (D262A & F296A) 1 or the GFP control ( Figure 1C). As we had previously reported with U2OS cells 1 , PAWS1 WT displayed predominantly diffused cytoplasmic localisation in HaCaT cells, and no obvious differences in localisation patterns were observed with PAWS1 A34E or PAWS1 R52P ( Figure S2 (extended data 18 )). Endogenous CK1α was detected in PAWS1 IPs from cells rescued with PAWS1 WT but not from those rescued with the pathogenic mutants or with the CK1-interaction deficient mutants ( Figure 1D).
To ask whether the pathogenic PPK PAWS1 mutants affect PAWS1 function through additional changes to interacting partners, we undertook an unbiased proteomic approach to identify interactors of these mutants. To this end, PAWS1-KO HaCaT cells were rescued with GFP control, PAWS1-GFP, PAWS1 A34E -GFP, PAWS1 R52P -GFP or PAWS1 F296A -GFP, and anti-GFP IPs were subjected to proteomic analyses. A Coomassie-stained gel revealed that the disappearance of a band at ~41 kDa, the predicted size of endogenous CK1α, from IPs of PAWS1 A34E , PAWS1 R52P and PAWS1 F296A was the only striking difference from the wild-type control ( Figure 1E). Proteomic analysis of interacting proteins from each IP confirmed that the only difference between PAWS1 WT and the three mutants was in the abundance of CK1α ( Figure 1F), suggesting that noninteraction of the mutants with CK1α is likely to be a key factor  in pathogenesis of PPK. This was further verified by immunoblotting, which showed that PAWS1 WT interacts with endogenous CK1α while the mutants do not ( Figure 1G).
Intriguingly, the mutant proteins had a lower apparent molecular weight on SDS-PAGE compared with the WT protein ( Figure 1C, D, E & G). As PAWS1 is a known substrate of CK1α 1 , we investigated if this mobility shift reflected a change in the phosphorylation status. Indeed, PAWS1 WT was observed as a single, faster migrating band following treatment with lambda protein phosphatase in vitro ( Figure S3 (extended data 18 )). PAWS1 A34E and PAWS1 R52P also appear to be phosphorylated in cells, albeit to a lesser extent. Meanwhile, the mobility shift of PAWS1 S614A (which interacts with, but is not phosphorylated by CK1α 1 ) is indistinguishable from PAWS1 WT . Thus, PAWS1 phosphorylation by other kinases can occur in both CK1α-dependent and independent manner.
Furthermore, when PAWS1-KO cells were rescued with PAWS1 WT or with the PPK pathogenic mutants, we consistently observed a lower abundance of PAWS1 A34E and PAWS1 R52P proteins compared with PAWS1 WT (Figure 1C, D, E & G). This suggests that the A34E and R52P mutations might affect the stability of PAWS1 protein.

PPK PAWS1 mutants exhibit reduced protein stability
To ask whether the pathogenic mutations of PAWS1 affect its stability, we first transiently transfected PAWS1 WT , PAWS1 A34E , and PAWS1 R52P into PAWS1-KO U2OS osteosarcoma cells to achieve comparable starting levels of the respective proteins. The stability of the proteins was tested over 9 h following inhibition of protein synthesis with cycloheximide ( Figure 2A&B). We found that following cycloheximide treatment, PAW-S1 A34E and PAWS1 R52P protein levels declined more rapidly (t 1/2 = 3 h) than PAWS1 WT (t 1/2 > 9 h). After 9 h of cycloheximide treatment, PAWS1 A34E and PAWS1 R52P protein levels were reduced to 10-20% of the levels of their untreated controls (0 h), while about 60% of PAWS1 WT remained ( Figure 2A&B). As a control, c-myc protein levels were undetectable within 3 h of cycloheximide treatment ( Figure 2B). The reductions in PAWS1 A34E , and PAWS1 R52P protein levels, as well as that of c-myc, was rescued by co-treatment with the proteasome inhibitor bortezomib ( Figure 2C&D).
One possible explanation for the decreased stability of the PAW-S1 A34E and PAWS1 R52P proteins is that inability to interact with CK1α prevents their phosphorylation by CK1α. However, PAWS1 S614A is not phosphorylated by CK1α 1 , and its stability is unaffected in the cycloheximide assay ( Figure 2E, F), arguing that PAWS1 phosphorylation by CK1α does not regulate its stability.

Canonical Wnt signalling is impaired by PPK PAWS1 mutations
The PAWS1-CK1α complex is an important mediator of the Wnt signalling pathway, so we sought to determine if canonical Wnt signalling is affected by the PAWS1 A34E and PAWS1 R52P mutants. Because HaCaT cells did not respond to stimulation with Wnt3a ( Figure S4 (extended data 18 )), we turned to the U2OS cells in which we have previously studied canonical Wnt signalling 1 . We co-expressed PAWS1 WT , PAWS1 A34E , PAWS1 R52P , or PAWS1 F296A with a TOPflash Wnt/ β-catenin luciferase reporter in U2OS cells 1 , and measured luciferase reporter activity following stimulation with control-or Wnt3A-conditioned medium ( Figure 3A&B). Consistent with our previous report 1 , overexpression of PAWS1 WT increased both basal and Wnt3A-stimulated reporter activity compared with GFP and PAWS1 F296A controls ( Figure 3A&B). Under these conditions, overexpression of the PAWS1 A34E and PAWS1 R52P mutants, at similar levels to that of PAWS1 WT , did not enhance either basal luciferase reporter activity or that induced by Wnt3A ( Figure 3A, B), suggesting that these mutants are unable to mediate Wnt signalling in U2OS cells.
Consistent with its Wnt-activating role, we have previously demonstrated that ectopic delivery of PAWS1 WT mRNA into a single ventral blastomere at the 4-cell stage Xenopus embryo results in the formation of a complete secondary axis, resembling that formed in response to ectopic xWnt8. This axis-inducing ability of PAWS1 requires CK1α-binding because the PAWS1 D262A and PAWS1 F296A mutants fail to induce axis duplication 1 . In order to test the axis-inducing ability of PAWS1 A34E and PAWS1 R52P mutants, we microinjected PAWS WT or the mutant mRNAs into Xenopus embryos and assessed the formation of a secondary body axis at the tadpole stage. While PAWS WT induced partial or complete axis duplication in ~80% of embryos, PAWS1 A34E and PAWS1 R52P mutants did not ( Figure 3C-E), further confirming the failure of these mutants to activate Wnt signalling. We also observed lower levels of PAWS1 A34E and PAWS1 R52P protein relative to PAWS1 WT in these tadpoles despite the embryos being injected with the same amounts of mRNA ( Figure 3C). Thus, we normalised the amount of PAWS1 WT and mutant proteins expressed in the embryos by titrating the amount of mRNA injected. Axis duplication phenotypes were still observed in ~50% of embryos injected with half the amount of PAWS1 WT mRNA, indicating that the lack of phenotype observed with the mutants is unlikely to be explained by reduced protein levels ( Figure S5 (extended data 18 )).
To circumvent potential artefacts of the overexpression systems used above, we used CRISPR/Cas9 genome editing to replace the endogenous PAWS1 protein of U2OS cells with PAWS1 A34E . To achieve this, we used a novel donor strategy to knock in a polycistronic cassette consisting of GFP cDNA, an internal ribosome entry site (IRES) element, and PAWS1 A34E cDNA directly downstream of the native FAM83G promoter ( Figure 4A). GFP-positive clones were isolated and homozygous insertion of the PAWS1 A34E mutation was verified by PCR and genomic sequencing of one of the clones, which was then selected for further investigation ( Figure S6 (extended data 18 )). Consistent with the destabilising effect of the PAWS1 A34E mutation demonstrated earlier (Figure 2A-D), PAWS1 protein levels but not mRNA levels in U2OS A34E cells were substantially lower than in U2OS WT cells ( Figure 4B&C). We note, however, that this may also be due in part to reduced efficiency of translation initiated by the IRES relative to the wildtype mRNA sequence 21,22 . Interestingly, the PAWS1 mRNA levels in PAWS1-KO U2OS cells, also generated by CRISPR/Cas9 genome editing, were much lower than in U2OS WT and U2OS A34E cells ( Figure 4C), probably because of nonsensemediated decay of the PAWS1-KO transcript caused by a premature stop codon.
We measured Wnt-induced expression of the canonical Wnt target gene AXIN2 in U2OS WT , U2OS KO , and U2OS A34E cells. Wnt3a treatment induced a robust 5-fold upregulation of AXIN2 mRNA in U2OS WT cells relative to control ( Figure 4D). In contrast, in both U2OS KO and U2OS A34E cells, Wnt3A-induced upregulation of AXIN2 mRNA was significantly reduced compared with U2OS WT cells; p<0.001 ( Figure 4D).
Alanine 34 of PAWS1 is conserved in FAM83 proteins and appears functionally analogous in FAM83H Both Ala 34 and Arg 52 of PAWS1 lie in the DUF1669 domain, which is conserved and located at the N-terminus of FAM83 proteins and is required for binding to CK1 kinases 10 . Whilst the Arg 52 residue of PAWS1 is conserved only in FAM83D and FAM83E, Ala 34 is completely conserved across all FAM83 members, and may therefore serve a similar and important function for all FAM83 members ( Figure 5A). With this in mind, we made an analogous mutation on FAM83H (A31E) and introduced FLAG-FAM83H WT or FLAG-FAM83H A31E into FAM83H KO U2OS cells. Interestingly, we observed lower levels of FAM83H A31E protein than of FAM83H WT ( Figure 5B), reminiscent of the observation that PAWS1 A34E is less stable than PAWS1 WT . Consistent with our previous report 10 , IPs of FAM83H WT co-precipitated endogenous CK1α, δ, and ε isoforms ( Figure 5B). However, FAM83H A31E IPs did not co-precipitate CK1α, δ, or ε isoforms ( Figure 5B), suggesting that this residue in FAM83 proteins is necessary for binding to CK1 isoforms.

Discussion
The almost identical hyperproliferative epidermis and hair phenotypes reported in human PPK patients carrying the homozygous PAWS1 A34E mutation 9 and in HFH dogs carrying the PAWS1 R52P mutation 5,6 hint at a common mechanism for disease pathogenesis. Our results strongly suggest that this common mechanism involves the inability of PAWS1 A34E and PAWS1 R52P to associate with CK1α, which reduces their ability to activate Wnt signalling.
CK1α (and other CK1 isoforms) regulates Wnt signalling both positively and negatively by phosphorylating many components of the pathway. For example, CK1α phosphorylates cytoplasmic β-catenin at Ser45, which allows GSK3β to phosphorylate Thr41, Ser37 and Ser33 and mark it for proteasomal degradation, thus down-regulating Wnt/β-catenin signalling 23, 24 . In contrast, CK1 kinases can also positively regulate Wnt signalling 25 . For example, in response to the binding of Wnt ligand to the LRP5/6 receptor, CK1α phosphorylates LRP5/6 and p120catenin at the plasma membrane, both of which events are needed for full activation of signalling 26,27 .
We have previously shown that CK1α can exist in distinct complexes with all FAM83 members in addition to PAWS1 10 . Individual FAM83 proteins deliver CK1α or other CK1 isoforms to distinct subcellular compartments, and potentially to specific CK1 substrates, to influence specific cellular processes. For example, FAM83D delivers CK1α to the mitotic spindle to ensure proper spindle orientation and timely mitotic progression 28 . The PAWS1-CK1α complex appears to regulate Wnt signalling by controlling the nuclear accumulation of β-catenin downstream of the β-catenin destruction complex through as-yet-unknown mechanisms 1 . Establishing PAWS1-dependent CK1α substrates involved in mediating Wnt signalling will shed light on the mechanisms by which the pathogenic PPK PAWS1 mutants malfunction in Wnt signalling.
Despite the diverse functions of CK1α and associated FAM83 complexes, it is interesting to note that the ablation of CK1α from keratinocytes in mice resulted in palmoplantar and hair phenotypes similar to those associated with the two PAWS1 mutations 29 . Although these phenotypes were not characterised in detail, it would be interesting to compare them morphologically and at the molecular level with those from PAWS1-mutant PPK phenotypes from human patients and dogs. As chemical inhibitors of CK1 isoforms are known to affect Wnt 30-34 and p53 signalling 35 , it is not surprising that complete ablation of CK1α from keratinocytes leads to the activation of both Wnt and p53 signalling. We suggest that differences in the relative levels of Wnt signalling components that are positively or negatively regulated by CK1α between cell types/tissues may ultimately determine the phenotype caused by PAWS1-CK1α dysregulation. We also cannot rule out the contributions of other signalling pathways known to be active in skin, including, but not limited to, TGF-β/BMP, FGF, and YAP/TAZ 12,36-38 . Given that PAWS1 is highly expressed in the epidermal layer and inner root sheath of hair follicles 9,39 , it may be that PAWS1 is required for tissue-specific regulation of CK1α activity and modulation of signalling responses in these compartments. Characterisation of PAWS1 function in animal models or skin organotypic cultures will hopefully provide more definitive evidence of this in the future.
Finally, we also report here that PAWS1 mutant proteins have significantly shorter half-lives in cells. Consistent with these findings, immunostaining of skin sections revealed reduced levels of PAWS1 A34E protein in a patient suffering from PPK 9 . Computational structure predictions suggest that residue A34 is positioned at the centre of an alpha helix, and R52 at the amino-terminal boundary of the following alpha helix 40,41 . Taken together with the radical nature of the A-E and R-P amino acid substitutions-hydrophobic to negative/hydrophilic, and positive/hydrophilic to hydrophobic respectively-it is likely that PAWS1 A34E and PAWS1 R52P are misfolded and subsequently degraded in a proteasome-dependent manner. Determination of the structure of PAWS1 or DUF1669 in complex with CK1α will allow accurate mapping of the residues that directly form the interface, and, given the high degree of conservation of the DUF1669, will no doubt be invaluable in understanding broader aspects of FAM83 and CK1 kinase biology. This project contains the following underlying data • Immunoblots°  • Mass spectrometry° KWu 181203.sf3 (Scaffold file of mass spectrometry data shown in Figure 1E-G)

Data availability
• The work brings together the recent genetic data from dog and human FAM83G/PAWS1 mutations (Ref 9), which are associated with keratoderma and excess hair. The study is essentially complete, is high quality, has an excellent Materials and Methods section, and raises lots of interesting questions for the future that will be useful for various groups interested in CK1, other FAM83 gene members and broader Wnt and CK1-dependent signalling mechanisms.
The major output is to evaluate the interesting, potentially disruptive point mutations in the PAWS1/FAM83G 'DUF' (more likely CK1-binding) domain (R52P and A34E) by putting them through their paces in the context of CK1 and CD2AP/SMAD1 binding, broader proteomic interactomes, FAM83G cellular stability in several cellular models, axis duplication in frog embryos in vivo and also through generation of a very useful panel of knock-out/A34E knock-in lines.
I do not have any specific criticisms, the work is of a high quality, well explained and (most importantly) is very well controlled (often using previously characterized D262A/F296A FAM83G/PAWS1 mutants), and comes to sensible conclusions, clearly implicating the conserved Ala residue in the DUF domain of other FAM83 members as being involved in CK1 isoform targeting.
Discussion point: The reduced stability of A34E and R52P FAM83G might have been compared side-by-side/or titrated at the mRNA level with D262A/F296A (or similar) in the Xenopus model, since the lack of phenotype here could, in theory, be explained by the decreased amount of FAM83G protein present (certainly <1/2 of the WT). The F296A is also 'more' stable (based on data in Figure 1C/E/G, all other things being equal).
Obviously, it will be interesting to learn more about FA83G turnover, and how this might be controlled by proteome interactions (including CK1, but perhaps not through phosphorylation?). I would also purify recombinant FAM83G and the various mutants, and analyse whether they do indeed exhibit different thermal or chemical stability, or if the change in half-life is actually caused by the huge changes in partner interactions, or changes in PTMs. In this context, the clear difference between band-shifting of all the PAWS1 mutants tested in Figure 1C/D (HACAT KO cells) is distinct from the lack of change in shifting in Figure 1A/B (HEK-293), where the endogenous PAWS1 is presumably present.

Is the work clearly and accurately presented and does it cite the current literature? Yes
Is the study design appropriate and is the work technically sound? Yes

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
No competing interests were disclosed. Competing Interests: I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.