The mammalian CTLH complex is an E3 ubiquitin ligase that targets its subunit muskelin for degradation

The multi-subunit C-terminal to LisH (CTLH) complex is the mammalian homologue of the yeast Gid E3 ubiquitin ligase complex. In this study, we investigated the human CTLH complex and characterized its E3 ligase activity. We confirm that the complex immunoprecipitated from human cells comprises RanBPM, ARMC8 α/β, muskelin, WDR26, GID4 and the RING domain proteins RMND5A and MAEA. We find that loss of expression of individual subunits compromises the stability of other complex members and that MAEA and RMND5A protein levels are interdependent. Using in vitro ubiquitination assays, we demonstrate that the CTLH complex has E3 ligase activity which is dependent on RMND5A and MAEA. We report that the complex can pair with UBE2D1, UBE2D2 and UBE2D3 E2 enzymes and that recombinant RMND5A mediates K48 and K63 poly-ubiquitin chains. Finally, we show a proteasome-dependent increase in the protein levels of CTLH complex member muskelin in RMND5A KO cells. Furthermore, muskelin ubiquitination is dependent on RMND5A, suggesting that it may be a target of the complex. Overall, we further the characterization of the CTLH complex as an E3 ubiquitin ligase complex in human cells and reveal a potential autoregulation mechanism.


Results
WDR26 and GID4 are CTLH complex members. The initial characterization of the CTLH complex determined that it was composed of 6 subunits, RanBPM, TWA1, muskelin, ARMC8 and the RING domain proteins RMND5A and MAEA ( Fig. 1a) 13,14 . We confirmed the composition of the complex by immunoprecipitation of RanBPM in HEK293 cells and found that CTLH complex members remain associated with RanBPM even under stringent conditions (Fig. 1b, Supplementary Fig. 1). WD Repeat Domain 26 (WDR26) and human GID4 (also known as c17orf39), the homologues of the yeast Gid complex members Gid7 and Gid4, respectively, were not detected in the initial identification of the complex 13 . In contrast, we found that endogenous WDR26 associates with RanBPM ( Fig. 1b) and that CTLH complex members co-immunoprecipitate with exogenously expressed WDR26 and GID4 (Fig. 1c,d), consistent with recent interactome studies that revealed the human proteins do associate with the complex 24,25 . Taken together, this works shows that WDR26 and GID4 are CTLH complex subunits.
To compare the subcellular localizations of all complex members, we transfected HA or FLAG tagged constructs of RanBPM, TWA1, ARMC8, RMND5A, MAEA, muskelin, WDR26 and GID4 in HeLa cells (Fig. 2a). Consistent with a previous report 13 , RanBPM, TWA1, ARMC8 and RMND5A showed nucleocytoplasmic distribution, with a nuclear predominance, while muskelin appeared mostly cytoplasmic and MAEA nearly exclusively nuclear (Fig. 2b). Interestingly, GID4 displayed a near exclusive nuclear staining and WDR26 was primarily cytoplasmic (Fig. 2a,b). The differing subcellular localization of CTLH complex members suggests the possibility that several complexes of varying composition may co-exist in the nucleus and cytoplasm.
Interdependence of CTLH complex subunit stability. The RanBPM and TWA1 yeast homologues Gid1 and Gid8, respectively, form the core of the Gid complex with MAEA and RMND5A homologues (Gid9 and Gid2), which form a heterodimer (Fig. 1a). The remaining 3 subunits, Gid7, Gid5 and Gid4, the yeast homologues of WDR26, ARMC8 and GID4/C17orf39, respectively, were predicted to be located on the periphery of the complex 15,26 . If this topology is similar in the CTLH complex, RanBPM and TWA1 would be expected to be critical for complex stability. To determine how CTLH complex subunit expression influences each other, we assessed the levels of CTLH individual subunits in stable shRNA knockdown or CRISPR knockout (KO) cell lines for six of the complex members. We found that depletion of RanBPM or TWA1 strongly affected each other's protein levels as well as that of MAEA and RMND5A (Fig. 3a,b). TWA1 knockout also had a surprising inhibitory effect on the stability of ARMC8β, while ARMC8α was not significantly changed (Fig. 3b). Interestingly, the RING dimer www.nature.com/scientificreports www.nature.com/scientificreports/ partners MAEA and RMND5A appeared to require each other for stability as knockout of each one individually significantly reduced the protein levels of the other (Fig. 3c,d). No other prominent change was seen in RMND5A and MAEA KO cell lines, except that both showed a significantly higher amount of muskelin (Fig. 3c,d). This was also observed in RanBPM shRNA cells, albeit to a lesser extent potentially owing to the partial downregulation of RanBPM in these cells (Fig. 3a). Knockout of ARMC8 resulted in downregulation of MAEA, RMND5A and TWA1 (Fig. 3e), suggesting that TWA1, MAEA/RMND5A and ARMC8 influence each other's stability. Finally, the knockout of muskelin only had subtle effects, if any, on protein levels of other complex members (Fig. 3f).
To determine whether these changes occurred at the mRNA level, we performed RT-qPCR analyses to evaluate whether the KO of individual subunits had an effect on the transcriptional regulation of other CTLH complex members. We did not detect any change in mRNA expression for most of the subunits tested, except for a small reduction of muskelin and MAEA mRNA in RMND5A KO cells, and a slight decrease for RMND5A mRNA in ARMC8 KO cells ( Supplementary Fig. 2). In all cases, these changes were much smaller than the effects observed at the protein level (or even opposite in the case of muskelin) and therefore unlikely to account for the full extent of the effect observed at the protein level. Altogether, this substantiates that the knockout of individual complex members affects the stability of other complex members mostly at the protein level.
Finally, we used a combination of transient siRNA knockdown and subunit re-expression in knockout cell lines to confirm that these changes were not due to off-targets effects. We confirmed that siRNA downregulation of TWA1, muskelin and RMND5A recapitulated the changes observed in TWA1, muskelin and RMND5A CRISPR KO cells ( Supplementary Fig. 3). Similarly, RanBPM KO cells showed similar changes in CTLH subunits as the RanBPM shRNA cells and transient re-introduction of Flag-MAEA in MAEA KO cells restored the expression of RanBPM, ARMC8 and muskelin close to the levels observed in WT cells ( Supplementary Fig. 3).
Characterization of the CTLH complex E3 ligase activity. To determine whether the mammalian CTLH complex has E3 ligase activity, we conducted in vitro ubiquitination assays with the CTLH complex  26 . Note that the position of muskelin in the complex has not been formally defined. (b) Subunits of the CTLH complex are present in RanBPM immunocomplexes. HEK293 whole cell extracts were incubated with a RanBPM antibody and immunoprecipitated. Immunoprecipitates were analyzed by Western blot with the indicated antibodies. IgG was used as a negative control. (c) WDR26 associates with the CTLH complex. Whole cell extracts were prepared from HeLa cells untransfected (−) or transfected with FLAG-tagged WDR26 (+). FLAG-WDR26 was immunoprecipitated with a FLAG antibody and immunoprecipitates were analyzed by Western blot with the indicated antibodies. (d) GID4 associates with CTLH complex. Whole cell extracts were prepared from HEK293 cells untransfected (−) or transfected with HA tagged GID4 (+). HA-GID4 was immunoprecipitated with an HA antibody and immunoprecipitates were analyzed by Western blot with the indicated antibodies. immunoprecipitated from HEK293 cells via a RanBPM antibody (as in Fig. 1b). For these assays, we supplemented the reactions with the E2 enzyme UBE2D2 (UbcH5b) because it paired with the yeast RMND5A counterpart (Gid2) in in vitro assays 7 and was also identified as an interacting partner for human RMND5B, a paralog of RMND5A, in a large protein interaction screen 27 . Ubiquitination products were observed when the CTLH complex was immunoprecipitated from wild-type HEK293 cells, but not in RMND5A knockout HEK293 cells (Fig. 4a). As the complex is intact in the RMND5A KO cells (save for RMND5A, Supplementary Fig. 4), this result demonstrates that RanBPM immunocomplexes have E3 ligase activity and that it is dependent on RMND5A, a RING domain CTLH complex subunit. To understand the contributions of MAEA to the E3 ligase activity of the CTLH complex, we conducted in vitro ubiquitination assays with the RanBPM immunocomplexes in control and MAEA knockout HEK293 cells. As anticipated, limited E3 ligase activity was observed in the MAEA KO cells (Fig. 4b); however, consistent with the yeast Gid complex topology 26 , co-IP of RMND5A was not detected in MAEA KO cells. Therefore, the loss of activity could be attributed to the absence of RMND5A.
To further characterize the E3 ligase activity of the RING domain subunits RMND5A and MAEA, we conducted in vitro ubiquitination assays with bacterially expressed proteins. Initial experiments using purified GST-RMND5A and SUMO-MAEA failed to show any detectable activity (data not shown), possibly due to poor folding or insolubility of these enzymes. Therefore, we omitted the purification step and conducted E3 ligase assays using crude bacterial extracts as previously done to test the activity of Gid2 and Gid9 7,11 . In these conditions, human GST-RMND5A exhibited weak, but observable substrate and self-ubiquitination activity (Fig. 5a).
Previous studies reported that the bacterially expressed yeast Gid9, the homologue of MAEA, had no detectable E3 ligase activity in vitro 11 . However, in vivo, a cysteine mutation in the MAEA RING domain abolished Gid complex ubiquitination of FBPase, suggesting that the RING domain of Gid9 is required for the Gid complex activity 11 . Surprisingly, a SUMO tagged recombinant version of human MAEA expressed in Escherichia coli exhibited some E3 ubiquitin ligase activity (Fig. 5b), albeit weaker than that of RMND5A. Thus, both recombinant RMND5A and MAEA display E3 ligase activity in vitro.
Characterization of E2 pairings and lysine linkage. As E3 ligase ubiquitination activity is dependent on a specific E2 enzyme, we sought to determine which E2 enzymes function optimally with the CTLH complex. In a panel of 11 E2 enzymes, the GST-RMND5A fusion protein exhibited E3 ligase activity only when UBE2D1 (UbcH5a) or UBE2D2 (UbcH5b) are present in the reaction (Fig. 6a, Supplementary Fig. 5), while RanBPM immunocomplexes were able to function with UBE2D1, UBE2D2 and UBE2D3 (Fig. 6b). Interestingly, the complex or GST-RMND5A did not exhibit activity when paired with UBE2H (UbcH2) (Fig. 6a,b), the human homologue of yeast Gid3 28 (also known as ubc8), which is the E2 required for the glucose-induced ubiquitination of FBPase 29,30 . Similarly, no activity was detected with CDC34 ( Fig. 6b), which has a C-terminal extension of acidic residues similar to that of Gid3 that is critical for its activity 31 .
To assess which type of chain linkage is being mediated by RMND5A when paired with UBE2D2, we trypsin/ LysC digested the GST-RMND5A ubiquitination assay reactions and analyzed the modified ubiquitin peptides using mass spectrometry. In the GST-RMND5A assays with UBE2D2, approximately 80% of modified ubiquitin www.nature.com/scientificreports www.nature.com/scientificreports/ was ubiquitinated on K48, while the remaining 20% was K63 (Fig. 6c, Supplementary Fig. 6). No other lysine linkages were detected. This suggests that RMND5A can mediate both K48 and K63 ubiquitin chains, with a preference for K48.

Muskelin is a target of the CTLH complex.
Muskelin is the only member of the CTLH complex that does not have a yeast homologue in the Gid complex and therefore stands out as a unique difference between the two complexes 10 . Interestingly, muskelin protein levels were significantly increased in both RMND5A and MAEA KO cells compared to control HeLa cells and this was reversed by the reintroduction of RMND5A and MAEA into their respective knockout cells by transient transfection (Figs 3c,d and 7a and Supplementary Fig. 3). Treatment with the proteasome inhibitor MG132 resulted in increased muskelin levels in control cells relative to . RanBPM and TWA1 are essential for complex stability. Whole cell extracts prepared from control shRNA and RanBPM shRNA HEK293 cells (a), or from control (labelled as C), TWA1, RMND5A, MAEA, ARMC8 and muskelin HEK293 CRISPR knockout cells (b-f) were analyzed by Western blot with antibodies to CTLH complex members, as indicated. Vinculin was used as a loading control. Quantifications are shown below each blot and protein levels are shown relative to control cells set to 1 and normalized to Vinculin levels. Data represent averages from three separate experiments, with error bars indicating SD. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. www.nature.com/scientificreports www.nature.com/scientificreports/ DMSO control treatment, but not in RMND5A KO HeLa cells (Fig. 7b), suggesting that the increase in muskelin in RMND5A KO cells is proteasome dependent. To confirm that muskelin protein stability was dependent on the CTLH complex, we performed cycloheximide (CHX) treatments to compare muskelin degradation of WT and RMND5A KO cells. Muskelin had a half-life of about 24 hours in WT cells, whereas no significant degradation was detected in RMND5A KO cells, even 36 hours following addition of CHX, suggesting that muskelin degradation is dependent on the CTLH complex activity (Fig. 7c). This led us to suspect that muskelin could be a ubiquitination target of the complex leading to proteasomal degradation.
To evaluate if ubiquitination of muskelin is regulated by the CTLH complex, we co-transfected HA-ubiquitin and a construct encoding FLAG-muskelin in control or RMND5A KO HEK293 cells, followed by treatment with MG132. FLAG pulldown of transfected muskelin and hybridization with an HA antibody showed a poly-ubiquitination pattern in control cells which was reduced to background levels in RMND5A KO cells (Fig. 7d). Furthermore, endogenous ubiquitinated muskelin co-immunoprecipitated with transfected HA-ubiquitin in control cells but not in RMND5A KO cells (Fig. 7e). Together, this suggests that the CTLH complex is required for muskelin ubiquitination in vivo. Overall, the data suggests that the CTLH complex regulates ubiquitination and protein levels of one of its own subunits, muskelin.

Discussion
In this study, we have characterized the composition and stability of the CTLH complex and demonstrated, using in vitro assays, that the complex has E3 ligase activity. We confirmed that human GID4 and WDR26 associate with the complex and that RanBPM, TWA1, MAEA, RMND5A and ARMC8 each have roles in maintaining complex stability. Importantly, RMND5A and MAEA are completely co-dependent on each other. We found that ubiquitination activity of the complex is dependent on the RING subunits RMND5A and MAEA. Furthermore, both recombinant RMND5A and MAEA also exhibit activity. Additionally, we determined that the complex can pair with ubcH5 family of E2 enzymes and GST-RMND5A can catalyze K48 and K63 ubiquitin chains. Finally, we revealed that CTLH complex regulates ubiquitination and proteasomal degradation of its peripheral subunit muskelin.
Our data suggests that protein expression of several CTLH complex subunits is interdependent, which has previously been observed to some extent for the yeast Gid complex. In the Gid complex, Gid1 (RanBPM) was deemed essential for the stability of the complex as its deletion resulted in decreased levels of Gid8 and Gid2 26 . www.nature.com/scientificreports www.nature.com/scientificreports/ Also, Gid2 and Gid9 were found to stabilize each other 26 . Our analyses of CTLH complex subunits knockdown/ knockout cells suggest that this is also true for the CTLH complex. RanBPM knockdown resulted in the decrease of TWA1, RMND5A and MAEA protein levels and RMND5A and MAEA required each other, suggesting that, like in the Gid complex, the two RING domain subunits are stabilized through heterodimerization. TWA1, like RanBPM, is central to CTLH complex formation, as its knockout affected RanBPM, MAEA, RMND5A and, curiously, ARMC8β. Finally, muskelin knockout did not significantly affect other CTLH complex members' protein levels. Previous microarray analyses revealed no changes in CTLH complex members' RNA expression in RanBPM shRNA cells 32 , and this was confirmed through qPCR analyses. We did not detect any significant changes in mRNA levels in most other CTLH complex knockout cells, except in 2 cases (RMND5A and ARMC8) where slight changes in mRNA expression were detected but were much more subtle than the effect observed at the protein level. Therefore the changes in protein expression induced by the knockout of individual CTLH subunits reflect mostly changes in protein stability rather than at the transcriptional level.
We found that both ARMC8α and β were present in the CTLH complex. These 2 isoforms originate from alternative splicing of the same gene product and have previously been identified as being part of the CTLH complex 13 . However, it raises the question as to whether both isoforms are present together in the complex or that sub-complexes contain one or the other isoform. Interestingly, the loss of TWA1 had different effects on the two ARMC8 isoforms, with ARMC8β being strongly decreased and ARMC8α slightly increased, albeit not significantly, inferring that the alternate splicing event may be regulated by the CTLH complex. Reversely, ARMC8 knockout significantly reduced TWA1, MAEA and RMND5A levels, suggesting that ARMC8 may stabilize the association of the core complex members.
Immunofluorescence analyses showed that the CTLH complex subunits are localized to different extents in the cytoplasm and nucleus. The subcellular localization observed with ectopically expressed proteins appear to match that of several endogenously expressed CTLH complex members reported in a previous study 13 . Adding to that, we found that GID4 is nearly exclusively nuclear, whereas WDR26 is mostly cytoplasmic. This suggests the possibility that several distinct CTLH complex variants exist in the cell. However, the fact that the knockout of MAEA, which is mostly nuclear, affects the levels of muskelin, which is mostly cytoplasmic, implies that these two seemingly differentially localized members are connected. Therefore, the dynamics of the nucleocytoplasmic localization of CTLH complex(es) will need further investigation. Also, our analysis involved ectopically expressed proteins, and while our results show that transfected CTLH complex members can interact with the endogenous CTLH complex (see Fig. 1), it is possible that a significant fraction of the transfected proteins did not associate with the complex and that their observed localization was dictated by their own localization properties rather than by those of the complex as a whole.
Our analyses of the CTLH complex E3 activity through in vitro ubiquitination assays using RanBPM immunocomplexes showed that, as expected, RMND5A is essential for activity. MAEA is also likely essential since its KO reduces RMND5A protein levels to background, therefore preventing CTLH complex activity. Also, based www.nature.com/scientificreports www.nature.com/scientificreports/ on the yeast topology, MAEA links TWA1 to RMND5A and potentially ARMC8 and therefore is essential for complex formation 26 .
We found that both bacterially-expressed recombinant RMND5A and MAEA have E3 ligase activity. This was previously reported for the Xenopus laevis and yeast homologues of RMND5A 7,12 , however, this is the first time that MAEA is shown to have intrinsic activity. The activity of both RMND5A and MAEA was low, and the reason for this may be that both contain RING domains that diverge considerably from the RING consensus 10 . Analysis of the consensus sequence for RMND5A homologues suggests its RING domain resembles that of TRIM32, but contains serines instead of cysteines in its second active site 10 . TRIM32 exists as a tetramer formed by two separate dimerization events that occur through the RING domains and the coiled-coil region 33 . TRIM32 RING dimerization is essential for catalytic activity 33,34 . This suggests that this may also be true for the RMND5A-MAEA dimer. Previous analyses with the yeast complex found that the interaction of Gid2 and Gid9 still occurs when the Gid2 CTLH and LisH domains are deleted, suggesting that the interaction may be mediated by the RING domains 11 . In our in vitro assays, we expressed RMND5A as a fusion protein with GST. Since GST dimerizes, that could be fulfilling the pre-requisite for dimerization. Alternatively, the LisH and CTLH domains could mediate dimerization and/or the RING domain may have similar dimerization capabilities as TRIM32. The MAEA RING domain is longer than conventional RING domains, is not predicted to coordinate zinc and has no obvious similarities to other eukaryotic RING motifs 10 . As it was expressed as a SUMO-fusion protein, dimerization, if required for activity, could only have been mediated by the RING and/or the LisH/CTLH domains.
Another study recently described the activity of the CTLH complex through biochemical reconstitution with purified components 35 . While some of our conclusions are in agreement with their findings, notably with respect to the requirements for MAEA and the other core complex subunits RanBPM, TWA1 and RMND5A for E3 ligase activity, some of our results differ, in particular regarding the identification of the E2 requirement for complex activity. While we detected robust activity with UBE2D1, UBE2D2 and UBE2D3 variants, we did not observe any activity with UBE2H, or CDC34. In agreement with our data, Lampert et al. did not report any activity with CDC34, however, they found that UBE2H was the most efficient E2 and they observed little activity with Samples were run on a 10% SDS-PAGE gel and analyzed by Western blot with the indicated antibodies. Muskelin quantifications are shown below relative to vinculin levels and normalized to untreated samples which were set to 1. N = 3, error bars indicate SD. *P < 0.05. The error bar for the KO samples at 24 h is not visible because it is smaller than symbol size. (d) Ubiquitination of transfected muskelin is regulated by RMND5A. Control or RMND5A KO HEK293 were co-transfected with FLAG-muskelin or HA-ubiquitin (HA-Ub) for 24 hours, followed by 10 μM MG132 treatment for 8 hours. FLAG-muskelin was pulled down in denaturing conditions. Note that transfected muskelin migrates higher than endogenous muskelin due to the added tags that amount to about 7 kDa. (e) Ubiquitination of endogenous muskelin is regulated by RMND5A. Control or RMND5A KO HEK293 were transfected with HA-Ub for 24 hours, followed by 10 μM MG132 treatment for 8 hours. HA-Ub was pulled down in denaturing conditions. (2019) 9:9864 | https://doi.org/10.1038/s41598-019-46279-5 www.nature.com/scientificreports www.nature.com/scientificreports/ UBE2D3. While UBE2H was identified as the human homologue of Gid3 28 , it lacks the C-terminal extension of acidic residues found in Gid3. This acidic tail is reminiscent of CDC34 in which the tail both binds the E3 ligase and promotes ubiquitin transfer 31 . Most of their assays involved the in vitro reconstituted CTLH complex with subunits purified from insect cells, whereas our assays involved bacterially-expressed RMND5A and the endogenous CTLH complex immunoprecipitated from HEK293 cells. Another difference is that we assayed complex activity from endogenous RanBPM immunoprecipitates, whereas they assayed the E3 activity from the complex immunoprecipitated through transfected tagged-RMND5A. The use of different experimental models (recombinant versus native complex) or the difference in complex isolation procedure (potentially yielding different complex conformations) as well as the relative activities of E2 enzymes could potentially account for these differences. However, our in vitro assays using recombinant RMND5A are consistent with previous in vitro ubiquitination assays which showed that yeast Gid2 (homolog of RMND5A) was able to function with UBE2D2 7 .
Our finding that muskelin protein levels and ubiquitination are regulated by the CTLH complex provides insight into a possible function and/or regulatory module of the complex. It also raises the question as to whether muskelin has a role in CTLH complex function or if it is just one of its targets. If the former, its appearance later in evolution suggests that it may provide a function to the complex distinct from the yeast Gid complex. As shown in this study and others, muskelin localizes mostly to the cytosol, which is dependent on its LisH domain-mediated dimerization 13,20,[36][37][38] . Interestingly, its overexpression has previously been shown to promote relocalization of other CTLH nuclear complex members to the cytoplasm 13 . Therefore, we speculate that the function of muskelin in the CTLH complex may be to direct the complex to specific cytoplasmic targets. Ubiquitination and subsequent degradation of muskelin by the complex could therefore serve as a feedback mechanism under certain conditions. Self-ubiquitination is a common feature of most E3 ligases and in many cases is used as a mechanism of autoregulation 39,40 .
Muskelin promotes cellular prion protein (PrP C ) turnover via the lysosome in neurons and muskelin knockout accelerates Prion disease induced by prion infection in mice 19 . Interestingly, a fragment of RanBPM is found overexpressed in Alzheimer's Disease (AD) patients and its overexpression promotes Amyloid beta (Aβ) generation and hallmarks of AD [41][42][43][44][45] . As PrP C propagates the neurotoxic signaling effects of Aβ [46][47][48] , this could implicate a mechanism whereby the effects observed by RanBPM overexpression in AD are, at least partially, a result of increased ubiquitination and proteasomal degradation of muskelin, leading to a defect in PrP C lysosomal degradation. It will be of interest to assess adult mouse models of other CTLH complex members, especially in the context of neurodegenerative diseases, to determine if there is an interplay between the CTLH complex and muskelin ubiquitination that contributes to disease pathogenesis.
In summary, we have extended the characterization of the mammalian CTLH complex, a unique E3 ligase complex containing a RING heterodimer. We confirmed its function as an E3 ubiquitin ligase and determined the requirements for its stability and activity. We have also revealed it regulates ubiquitination of its subunit muskelin, adding to its list of possible targets and identifying a potential autoregulatory mechanism that awaits further investigation. A comprehensive study of the substrates of the CTLH complex will be informative to understand the biological function of this multi-subunit E3 ubiquitin ligase.

Methods
Cell culture, transfections and treatments. Wild-type HeLa and HEK293 cells, and control shRNA and RanBPM shRNA stable HEK293 cells have been described previously 49,50 . All cells were cultured in high glucose Dulbecco's modified Eagle's medium (Wisent Bioproducts, St. Bruno, Quebec, Canada) supplemented with 10% fetal bovine serum (Wisent Bioproducts) at 37 °C in 5% CO 2 . Cells were treated with 10 μM MG132 (EMD-CalBiochem, San Diego, CA) for the indicated time points. For cycloheximide (CHX) treatment, cells were treated with 100 μg/ml cycloheximide (BioShop, Burlington, ON, Canada) and collected at the indicated time points. Plasmid transfections were carried out with jetPRIME (Polypus Transfection, Illkirch, France) according to the manufacturer's protocol. siRNA transfections were carried out as described previously 51 for siMuskelin and siRMND5A and siTWA1 (Silencer, AM16708A, 25822, Ambion, Life Technologies, Burlington, ON, Canada) transfections were performed using the same conditions.
To assess ubiquitination of FLAG-muskelin, control or RMND5A KO HEK293 cells were co-transfected with pCDNA-FLAG-muskelin and pMT123 plasmid expressing HA-ubiquitin 52 . Twenty-four hours after transfection, cells were MG132 treated for 8 hours. Cells were lysed in denaturing buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 1% Triton, 1% SDS, 1 mM Na3VO4, 10 mM NaF, 1 mM phenylmethylsulfonyl fluoride (PMSF), 1 μg/ml of aprotinin, 10 μg/ml of peptatin, 1 μg/ml of leupeptin and 25 mM NEM (N-Ethylamalide, Bioshop Canada, Burlington, ON, Canada)), passed through a 23 G needle ten times and incubated on ice for 30 minutes. For immunoprecipitation, lysates were diluted 1:10 in buffer A (50 mM Tris, pH 7.5, 150 mM NaCl, 1% Triton, 1 mM Na3VO4, 10 mM NaF and 25 mM NEM) and incubated with anti-FLAG (M2; F1804; Sigma-Aldrich) for 2 hours at 4 °C, www.nature.com/scientificreports www.nature.com/scientificreports/ followed by incubation with Dynabeads Protein G for 1 hour. Beads were then washed five times in buffer A and resuspended and boiled in SDS loading dye. Eluates were split in half when loading on SDS-PAGE so that HA and FLAG immunoblotting could be analyzed separately (to avoid overlapping signal). The same procedure was employed for the HA-ubiquitin immunoprecipitation, except that cells were transfected with only HA-ubiquitin plasmid and anti-HA (HA-7, H9658, Sigma-Aldrich) was used for the pulldown.
Statistical analysis. Statistical analyses were performed using GraphPad PRISM (GraphPad Software Inc., La Jolla, CA, USA). Differences between two groups were compared using unpaired two-tailed t test. Results were considered significant when P < 0.05.