The effect of the R957C substitution was analyzed using SNPs and GO and I-Mutant tools. SNPs and GO is a tool which predicts disease associated amino acid substitution at a single position in a specific protein including functional classifications with >82% prediction accuracy. I-Mutant predicts the effects of single point mutation with 80% accuracy. A probability score higher than 0.5 reveals a disease related effect of the mutation on protein function. The input given to the SNPs and GO and I-Mutant tools was the UniProt accession number (Q6ZNA4) of Arkadia isoform-1 protein, the sequence position of the wild type amino acid and the mutated amino acid.

A human RNF111 gene encoding amino acids 927–994 of the full length Arkadia was sub-cloned in a pGEX-4T-1 vector and transformed in Escherichia coli (E. cloni®) EXPRESS BL21 (DE3) cells (Lucigen). Cells were grown at 37°C in minimal medium (M9) supplemented with 1 g/L¹⁵N ammonium chloride, 4 g/L glucose or ¹³C-glucose and 1 ml/L¹⁵N or ¹⁵N/¹³C Bioxpress™ (CIL) for single or double labeled samples, respectively. Cell cultures were induced at an optical absorption of 0.6–0.9 with 1 mM IPTG (final concentration). After 4 h incubation at 37°C cells were harvested and the resulting pellet was lysed by sonication in PBS (Phosphate Buffered Saline) pH 7.4 ± 0.2 containing a protease inhibitor cocktail (Sigma) and DNAase I. The cell lysate was cleared by centrifugation at 20.000rpm (rounds per minute) for 30 min. For protein purification the supernatant was loaded onto a GST-trap 5 ml column (GE Healthcare). The GST-tag and Arkadia^927-994 were separated after overnight incubation with the protease thrombin (Merck Millipore) at 4°C. Arkadia^927-994 was eluted in PBS pH 7.4 ± 0.2 and concentrated using Amicon® Ultra centrifugal filter units (3 kDa cutoff) (Merck Millipore). The protein was further purified by size exclusion chromatography on a Superdex75 column (GE Healthcare).

The E2 enzyme UBCH5B (UBE2D2) was expressed and purified as described elsewhere (Birkou et al., 2017) and ubiquitin was expressed in E. coli (E. cloni®) EXPRESS BL21 (DE3) cells (Lucigen). Cells were grown at 37°C in minimal medium (M9) supplemented with 4 g/L glucose and 1 g/L NH₄Cl and induced with 1 mM (final concentration) IPTG. After 4 h of incubation at 37°C cells were harvested and lysed by sonication in PBS pH 7.4. The cell lysate was cleared by centrifugation at 20.000 rpm for 30 min. The supernatant was heated at 85°C for 15 min and centrifuged at 10.000 rpm for 20 min. For further purification ubiquitin was passed through a Superdex75 column (GE Healthcare).

NMR spectra were recorded on a Bruker Avance 600 MHz spectrometer equipped with a TXI cryoprobe and Bruker Avance III 700 MHz spectrometer equipped with a four-channel 5 mm cryogenically cooled TCI gradient probe. Protein samples were prepared in a mixed solvent of 90% H₂0 (50 mM K₂HPO₄, 50 mM KH₂PO₄ pH 7), 10% D₂0, 1 mM NaN₃ and 0.25 mM DSS (4,4-dimethyl-4-silapentane-1-sulfonic acid) as internal chemical shift standard. All NMR experiments are included in the Bruker pulse program library. Data were processed with Topspin 3.5 pl5 software and analyzed with CARA (Keller, 2004) and XEASY (Bartels et al., 1995). The sequence specific assignment of R957C Arkadia^927−994 domain was obtained using conventional backbone and side chain assignment methods (Wüthrich, 1986; Ferentz and Wagner, 2000) and were deposited in the BioMagResBank (BMRB; accession no: 50,985). The NMR solution structure of the R957C mutant was determined with DYANA (Güntert et al., 1997) and the ensemble of 30 models with the lowest RMSD and target function values was deposited in the ProteinDataBank after energy minimization with AMBER (Pearlman et al., 1995) (PDB, accession no: 7P2K; statistical analysis is reported in Supplementary Table S2). The assignment of the backbone ¹H-¹⁵N resonances of all non-proline residues of E2 enzyme UBCH5B was obtained from the BMRB database (accession no: 6,277).

The backbone mobility of the R957C Arkadia^927-994 mutant on the ps-ns time scale was investigated through ¹⁵N relaxation measurements (¹⁵N T ₁ and T ₂, {¹H^N}-¹⁵N NOE at 298 K) on Bruker Avance 600 and Avance III 700 spectrometers (experiments are reported in Supplementary Table S3). Standard pulse programs of the Bruker library were used. The overall delays between scans were 1.8 s for the T₁ and T₂ measurements and 4 s for the interleaved heteronuclear NOE experiment, respectively. The delays used for the T₁ experiments were 7, 18, 40, 85, 150, 230, 350, 500, 680, and 900 ms, whereas delays of 17, 34, 51, 68, 85, 102, 136, 187, and 221 ms were used for the T₂ experiments. ¹⁵N relaxation data were analyzed according to the model-free approach as implemented in the Tensor 2 program (Dosset et al., 2000) (R ₁, R ₂ and (¹H) ¹⁵N NOE diagrams are shown in Supplementary Figure S2).

To identify whether two zinc ions are bound to the R957C mutant, atomic absorption spectroscopy was performed (Perkin Elmer AAnalyst 700). For the zinc concentration determination proteins were purified as described above. Zinc ions standards of 0.1–1 ppm (Zinc Pure Standard, 1000 μg/ml, 2% HNO₃, Perkin Elmer) were used and an appropriate calibration curve was constructed (Supplementary Figure S1). The concentration of zinc in the protein samples was determined dividing the intercept by the slope of calibration curve.

To evaluate the stability of the R957C Arkadia^927–994 domain metal-chelation experiments were performed with the ethylenediaminetetraacetic acid (EDTA). ¹H-¹⁵N HSQC spectrums were recorded after each EDTA addition at a ratio 1:0.5, 1:1, 1:1.5 and 1:2 (Zn: EDTA).

In order to identify the interaction interface between E3 R957C Arkadia^927–994 and E2 UBCH5B, titration experiments were monitored by ¹H–¹⁵N HSQC spectra of labeled ¹⁵N R957C Arkadia^927–994 or ¹⁵N UBCH5B after each addition of the unlabeled protein partner. The unlabeled protein was added in eight steps in order to reach the following ratios and saturation of labeled/unlabeled protein: 1:0.25, 1:0.5, 1:0.75, 1:1, 1:1.25, 1:1.5, 1:1.75, 1:2. Combined chemical shift perturbations (CSPs) after binding were calculated using the equation: Δ δ p p m = ( Δ δ H N ) 2 + ( Δ δ Ν 5 ) 2 (Garrett et al., 1997; Baker et al., 2006; Marousis et al., 2018).

For the ubiquitination assays the Arkadia^876–994 polypeptide bearing the R957C mutation was used. Briefly, ubiquitination assay was performed by incubating 0.5 μM hUbe1 enzyme (Boston Biochem), 5 μM UBCH5B, 100–150 μM ubiquitin and 15 μM wt/R957C Arkadia^876-994 in 20 mM Tris-HCl pH 7.5, 50 mM NaCl, 5 mM ATP, 2 mM MgCl₂ and 2 mM DTT at 37°C for 0–60 min. Time points were collected after the addition of ATP and the reactions were stopped by addition of SDS loading buffer. Samples were resolved by 15% SDS-PAGE and visualized by western blotting for ubiquitin using the anti-Ub antibody P4D1 (Santa Cruz Biotechnologies, SCBT) and the goat anti-mouse m-IgGk BP-HRP sc-516102 (SCBT).

Luciferase assays were conducted in HEK293T cells (Cancer Research UK). using the Dual Luciferase reporter system (Promega), as previously described by Levy at al. 2007. Briefly, cells were transfected with the appropriate combination of promoter-reporter constructs and expression plasmids using FuGENE transfection reagent (Promega) and were cultured for 24 h after transfection. Luciferase activities in the cell lysates were measured following the manufacturer’s protocol. The experiments were repeated at least three times.

Atomic absorption spectroscopy was performed on R957C mutant to measure the Zn (II) content. A protein sample of 0.3 mM gave zinc concentration of 0.71 mM. These concentrations correspond to two zinc ions per protein molecule.

According to the structural role of the Zn (II) in RING domains, addition of the metal-chelating agent EDTA to a final ratio 1:2 (Zn: EDTA) led to the complete unfolding of the protein after 48 h as indicated by the loss of chemical shift dispersion in the ¹H-¹⁵N HSQC spectra (Figure 2A). Similarly, addition of EDTA to the wt RING domain caused complete unfolding of the protein after 48 h of incubation with the chelating agent (Figure 2B). These results suggest that the addition of one more Cys residue in the RING domain of Arkadia has not affected the zinc ions binding and the stability of the mutated protein. According to the C_α (56 ppm) and C_β (27 ppm) chemical shifts Cys957 in reduced state and does not participate in zinc ions binding or in disulfide bond formations (Sharma and Rajarathnam, 2000).

The NMR solution structure of the R957C mutant was determined based on a total of 906 NOEs distance constraints. The final family energy minimized NMR models contains two β-strands, namely β1 (Val955-Leu958) and β2 (His962-His965) forming an antiparallel β-sheet, two zinc binding loops and a 3-turn α-helix encompassing the residues Gln966-Thr975 (Figure 3). The two β-strands of the R957C mutant are one amino acid longer compared to wt RING Val955-Arg957 and Leu963-His965 β-strands. The a-helix observed in the R957C mutant is identical to the wt Arkadia RING helix in length and sequence position (Chasapis et al., 2012). The secondary structure elements of the R957C RING domain exhibit a ββα topology as observed in the wt Arkadia, RNF24 (PDB: 2EP4) and RNF168 RING domains (Zhang et al., 2013). The average distance between the two metal centers in the final 30 NMR models is 13.8 ± 0.4 Å (Figure 3). A detailed view of the coordination of the two Zn (II) ions is shown in Figure 4. The overall backbone and heavy atom RMSDs of the 30 energy-minimized models of R957C are 1.22 ± 0.48 Å and 2.08 ± 0.45 Å, respectively, for the core region between Glu939 and Ile986. Quality assessment of the final models reveals that 100% of the non-glycine/non-proline residues fall into favorable or allowed regions of the φ/ψ dihedral angle space in the Ramachandran plot (Supplementary Table S2).

¹⁵N relaxation studies of R957C Arkadia^927–994 revealed that mutant is a monomer in solution since the correlation time for isotropic tumbling measured based on the R ₂/R ₁ ratio is 4.69 ns corresponding to a MW of ∼7.81 kDa (theoretical MW of R957C RING is 7767 Da). In addition, model-free analysis of ¹⁵N relaxation data as implemented in the Tensor2 program showed that the core exhibited a rather rigid structure (Figure 5). The order parameters of the region Lys941-Ile986 (average S ² = 0.77) are higher than those of the N- and C-terminal residues (Lys927-Glu940 and Glu987-Ser994, respectively) closely resembling the ¹⁵N-relaxation properties of the wt Arkadia RING (Chasapis et al., 2012; Birkou et al., 2017) (Figure 5). The above data clearly support that there are no significant differences in the structure and dynamic properties between wt and R957C Arkadia RING domain.

To determine whether the R957C mutation affects the interaction of the RING with E2 we used NMR driven interaction studies of ¹⁵N R957C with the E2 enzyme UBCH5B, which functions as a partner of Arkadia in vitro (Mavrakis et al., 2007). The analysis showed that the resonances exhibit either fast or intermediate exchange behavior, suggesting a moderate affinity for the two proteins. The largest chemical shift changes were observed for the sequential stretches Cys942–Thr943, Cys945-Ile947, Val955-Leu958, Phe964-His965, Val967-Asp970 and Trp972-Asn976 and for residue Ile981 (Figure 6A). The first three sequential stretches comprise the first and the third Zn(II)-binding motif, while the fourth and fifth are located at the C-end of the α-helix Gln966-Thr975. The above regions are essentially identical with those exhibiting the largest chemical shift changes upon interaction with UBCH5B in the native RING (Chasapis et al., 2012; Birkou et al., 2017).

The addition of unlabeled R957C RING to ¹⁵N-labeled UBCH5B resulted in the loss of four UBCH5B amide resonances (Ser94, Ile99, Ser100 and Leu103) in the ¹H–¹⁵N HSQC spectra during the titration. The largest CSPs in UBCH5B were observed for residues Ala2-Leu3, Ile6-Glu9, Asn11-Asp12, Arg15-Asp16 (N-terminal helix α1) Ala19, Ser22 (β1 strand), Ile37 (β2 strand), Phe62-Lys63 (loop L1 between the β3 and β4 strands), Ser91-Glu92, Ala96-Thr98 (loop L2b), Lys101, and Leu104 (the first half of helix α2) (Figure 6B). All these residues were also found to participate in wt Arkadia RING–UBCH5B interactions (Chasapis et al., 2012; Birkou et al., 2017), strongly suggesting that UBCH5B utilizes the same interface for its interaction with the Arkadia RING mutant R957C.

The above data show that R957C Arkadia mutant and UBCH5B E2 enzyme interact via the surfaces that are expected to be ivolved in a canonical E2-E3 RING interaction (Chasapis et al., 2012; Gundogdu and Walden, 2019). However, it seems that the CSPs observed for the mutant are somewhat larger compared with those observed for the wt-E2 interaction [0.09/0.06 ppm for the R957C and wt RING, respectively and 0.042/0.038 ppm for the interacting UBCH5B with R957C and wt RING, respectively (Figure 5 and Birkou et al., 2017)].

To examine the actual activity of the mutant protein we developed an in vitro ubiquitination assay. For the auto-ubiquitination assays, wt or mutated polypeptides containing NRG and TIER segments along with the RING domain (Arkadia^876–994) were used, because the RING domain alone does not auto-ubiquitinate. We performed the assay in the presence of E2 UBCH5B and ubiquitin. Immunoblot analysis of these reactions, with anti-ubiquitin antibodies, showed that both wt and R957C Arkadia ^876–994 were functional in this in vitro auto-ubiquitination assay, as indicated from the ladder-like bands that appeared after addition of ATP to the reaction mixture (Figure 7) due to the formation of a polyubiquitin chain on the substrates wt or R957C Arkadia ^876–994. We found that the R957C Arkadia^876–994 exhibit significantly rapid and enhanced autoubiquitination (Figure 7), indicating a gain of function and not loss of function.

Moreover, R957C Arkadia’s functionality was tested in vivo by luciferase assays. Luciferase experiments were carried out in HEK293T cells using the TGF-β SMAD-dependent reporter CAGA₁₂-Luc, empty plasmid, the full length wt, C937A and R957C Arkadia proteins as described in Birkou et al., 2017. Overexpression of the R957C Arkadia protein enhanced reporter expression a slightly higher than that of the wt Arkadia, however, this was not statistical significant in this assay (Figure 8). These in vitro and in vivo results show that the R957C mutation on the RING domain does not reduce the activity of Arkadia as E3 ubiquitin ligase.