Differential recruitment of E3 ubiquitin ligase complexes regulates RET isoform internalization

The RET receptor tyrosine kinase has important roles in development, maturation and maintenance of genitourinary, neural and neuroendocrine cell types (Mulligan, 2014). RET is activated by binding a glial-cell-line-derived neurotrophic factor (GDNF) family ligand and a cell-surface-bound co-receptor of the GDNF family receptors alpha (GFRα) to stimulate multiple downstream signalling pathways (Ibanez, 2013). However, RET is also a potent oncogene, contributing to tumour initiation and progression of multiple human cancers. Activating oncogenic RET mutations occur in thyroid and lung carcinomas, while expression and activation of wild-type RET are also linked to tumour metastasis and invasion in pancreatic and breast cancers, suggesting a broad oncogenic role for RET activity in these diseases (Mulligan, 2014).

RET is expressed as two major conserved isoforms, RET9 and RET51, generated by alternative splicing of 3′ exons, which share the first 1063 residues but differ in their unique 9 or 51 C-terminal amino acids, respectively (Myers et al., 1995; Tahira et al., 1990) (Fig. 1A). RET isoforms are generally coexpressed, but differ in their abilities to promote cell growth, motility and invasion (Lian et al., 2017). RET interactions with protein adaptors and signalling molecules are predominantly mediated through the key isoform-specific phosphorylated tyrosine residues pY1062 (in RET9) and pY1096 (in RET51) and a C-terminal PDZ-binding motif (FTRF₁₀₇₂ in RET9 only) that lie within these isoform-specific tails (Mulligan, 2014). As a result, RET9 and RET51 recruit distinct signalling and regulatory protein complexes (Tsui-Pierchala et al., 2002). We have previously shown that RET9 and RET51 differ in maturation, subcellular localization and protein trafficking after ligand stimulation (Crupi et al., 2015a; Richardson et al., 2012), suggesting that isoform-specific regulatory mechanisms also contribute to functional differences. In response to ligand, RET receptors at the cell surface are internalized into the endosomal network via clathrin-coated pits (CCPs) (Crupi et al., 2015a; Richardson et al., 2006). RET51 is internalized more rapidly and abundantly than RET9 but a subset of receptors is targeted for recycling back to the cell membrane (Crupi et al., 2015a; Richardson et al., 2012). In contrast, RET9 is internalized more slowly and is not recycled at appreciable levels (Crupi et al., 2015a; Richardson et al., 2012). Our previous studies show that clathrin-associated adaptor protein complex 2 (AP2; also known as AP2M1) is an important determinant of RET endocytosis, but suggest that additional isoform-specific protein interactions and/or modifications may facilitate internalization (Crupi et al., 2015a). The processes and interactions regulating these differences have not been explored.

Ligand-induced ubiquitylation of receptor tyrosine kinases (RTKs) by E3 ligases is a well-characterized mechanism promoting receptor downregulation (Cooper et al., 2015; Haglund and Dikic, 2012). Ubiquitylation by multiple E3 ligase families can promote attenuation of signalling and targeting of receptors for lysosomal or proteasomal degradation, but can also be an important determinant for recruitment to CCPs, internalization and/or endosomal sorting of some RTKs (Haglund and Dikic, 2012). The casitas B-lineage lymphoma (CBL) family are prominent contributors to RTK ubiquitylation (Mohapatra et al., 2013), but other E3 ligases, such as the NEDD4 (neural precursor cell expressed, developmentally downregulated) family have cell-type-specific regulatory roles in ubiquitylation of specific RTKs (Boase and Kumar, 2015). Previous studies have shown that RET recruits CBL family members via interactions with adaptor proteins such as GRB2, but that RET isoforms may differ in their ability to recruit and be ubiquitylated by CBL family members (Calco et al., 2014; Carniti et al., 2003; Kales et al., 2014; Scott et al., 2005). The roles of other ubiquitin ligases in RET ubiquitylation have not been well characterized.

In this study, we assessed the mechanisms regulating ubiquitylation of RET9 and RET51 isoforms, and their effects on RET endocytosis. We demonstrate that RET51 associates with, and is ubiquitylated by, the c-Cbl E3 ubiquitin ligase (hereafter denoted CBL), recruited via interactions with the GRB2 adaptor protein, and that GRB2 is essential for ubiquitylation and internalization of RET51. RET9 is not appreciably ubiquitylated by CBL, but recruits the NEDD4 ubiquitin ligase via a multiprotein complex including GRB10 (growth-factor-receptor-bound protein 10) and SHANK2 (SH3 and multiple ankyrin repeat domain 2) adaptors through its unique C-terminal sequences. Furthermore, we show that RET9 ubiquitylation by NEDD4 promotes localization to CCPs and internalization from the membrane. Our data indicate that ubiquitylation of RET9 and RET51 is regulated by distinct protein complexes that promote internalization, and modulate trafficking of each RET isoform.

Previous studies have shown that CBL E3 ubiquitin ligases promote RET ubiquitylation (Kales et al., 2014; Pierchala et al., 2006; Scott et al., 2005). Here, we examined the contribution of CBL to ubiquitylation of RET9 and RET51 isoforms. In HEK293 cells overexpressing a tagged CBL construct, wild-type RET51 was robustly ubiquitylated, whereas RET9 ubiquitylation was less pronounced (Fig. 1B). Kinase-dead mutants (K758M) of both RET isoforms were not significantly ubiquitylated. Using CRISPR/Cas9, we generated CBL knockouts (KO) in HEK293 cells. RET51 ubiquitylation was reduced in these cells, but was increased by re-expression of a CBL construct (Fig. 1C), indicating that CBL was required for RET51 ubiquitylation. In co-immunoprecipitation assays, RET51 showed a robust interaction with CBL whereas the kinase-dead RET51 mutant did not (Fig. 1D). We did not detect interactions between CBL and RET9 in these assays (Fig. 1D). CBL was phosphorylated on tyrosine residues in response to RET51 activation, but not in the presence of a kinase-dead RET mutant or vector control (Fig. 1E). Finally, RET51 ubiquitylation and interaction with CBL was ligand dependent and maximal after 15 min GDNF (Fig. 1F). Our data are consistent with previous reports showing CBL-mediated ubiquitylation of RET51 (Richardson et al., 2009; Scott et al., 2005) and suggest that RET9 ubiquitylation may be mediated by mechanisms distinct from those of RET51.

CBL has been shown to interact directly with some kinases through an N-terminal modified SH2 domain, the TKB (tyrosine kinase binding) domain, and/or indirectly via adaptor-mediated interactions with its C-terminal proline-rich region (Mohapatra et al., 2013). In co-immunoprecipitation assays, we showed that RET51 did not interact with a truncated CBL protein containing only the N-terminal TKB domain, but was precipitated with a C-terminal CBL protein containing the RING and ubiquitin ligase domains and proline-rich region (Fig. 2A). To further characterize the RET51–CBL interactions, we used a purified His-tagged CBL–SH2 protein domain and a purified GST-tagged intracellular (ic) kinase domain of RET51 in in vitro far western protein interaction assays (Fig. 2B). We have previously shown that GST–icRET proteins are constitutively phosphorylated and able to recruit adaptors in a phosphorylation-dependent fashion (Crupi et al., 2015a). Here, we show that the purified GST–icRET51 protein did not directly bind the CBL-SH2 domain or a negative control (PTB domain of DAB2), but did bind the purified SHC–PTB domain protein (Fig. 2B), a positive control known to interact with RET51 (Arighi et al., 1997; Crupi et al., 2015a). Our data show that RET51 does not interact directly with CBL through its TKB domain, suggesting that it binds indirectly through interactions with the CBL C-terminal domains, potentially by recruitment of additional adaptors.

RET9 and RET51 differ only in their 9 and 51 C-terminal amino acids, respectively. Thus, the unique RET51 phosphotyrosine pY1096 and a multifunctional docking site at pY1062 (Fig. 1A) that recruits distinct signalling complexes to RET9 and RET51 (Tsui-Pierchala et al., 2002) were evaluated as potential RET51–CBL interaction sites. We showed that RET51–CBL interaction was abrogated by a tyrosine to phenylalanine substitution at Y1096 (Y1096F), whereas the Y1062F substitution had little effect on RET51 co-immunoprecipitation with CBL (Fig. 2C). We confirmed these interactions using a panel of sequential RET51 C-terminal truncation mutants (Fig. 2D). Although we detected robust CBL interaction with the full-length RET51 (aa 1-1114), no interactions were detected with any of the RET51 truncation mutants, suggesting that pY1096 is the exclusive site for interaction of CBL and RET51 and that the interactions at that site are indispensable for CBL recruitment.

Our data suggest that CBL is recruited to RET51 through indirect interactions at pY1096, mediated through an adaptor protein, such as GRB2, which has previously been shown to bind at this site (Alberti et al., 1998). To determine whether GRB2 mediates CBL recruitment to RET51, we evaluated the effects of shRNA GRB2 knockdown (KD) in HEK293 cells. CBL interaction with RET51 was detected in the control (Mock KD), but not GRB2 KD cells (Fig. 3A). Furthermore, CBL-dependent RET51 ubiquitylation was reduced in GRB2 KD cells compared with Mock KD control cells (Fig. 3B). Interestingly, we saw no residual interaction of RET51 and CBL in the GRB2 KD cells (Fig. 3A), suggesting that GRB2 is the major adaptor required for this interaction.

As our data show that RET9 does not interact with, or become ubiquitylated by CBL (Fig. 1B,D), we assessed the possibility that another E3 ligase, NEDD4, which is known to mediate ubiquitylation of other neurotrophic factor receptors (Arévalo et al., 2006) could mediate RET9 ubiquitylation. In HEK293 cells transiently overexpressing NEDD4, RET9 was robustly ubiquitylated in a kinase-dependent manner, whereas RET51 ubiquitylation was not detected (Fig. 4A). RET9 interactions with NEDD4 were GDNF dependent, as was ubiquitylation, which was maximal after treatment with GDNF for 15 min (Fig. 4B). Interestingly, recruitment of NEDD4 to RET9 occurred in a similar time frame to the RET51–CBL interaction, but accumulation of ubiquitylated RET9 occurred more slowly. RET9 ubiquitylation was lost in the presence of a mutation of Y1062 (Y1062F), which lies in a unique SH2-binding phospho-tyrosine motif, suggesting that adaptor protein interactions with RET9 at that site were required for NEDD4-mediated ubiquitylation. Interestingly, a F1072A mutation, which disrupts a unique class I PDZ-binding motif (FTRF₁₀₇₂) at the C-terminus, also abrogated RET9 ubiquitylation (Fig. 4A). We further investigated the roles of the unique RET9 C-terminal amino acid motifs in RET9 interactions with NEDD4 in GST-pull down assays. We showed that GST–NEDD4 interacts with wild-type RET9, but not a kinase-dead (K758M) RET9 (Fig. 4C), and becomes phosphorylated in response to RET9 kinase activity (Fig. 4D). However, mutation of Y1062 (Y1062F) or of the C-terminal PDZ-binding motif (F1072A) abrogated this interaction (Fig. 4C), suggesting that multiple functional motifs may be required for NEDD4 recruitment. Mutation of other RET phospho-tyrosines (e.g. Y1015) had no effect on NEDD4 recruitment. We did not detect interactions between RET51 and NEDD4 (Fig. 4C), consistent with differential mechanisms for ubiquitylation of RET9 and RET51.

NEDD4 associates with several RTKs through interactions with the SH2-domain-containing adaptor protein, GRB10 (Boase and Kumar, 2015). RET has been previously shown to bind GRB10 upon activation, although the specific RET residues involved were not determined (Pandey et al., 1995). Using GST pull-downs, we showed that the GRB10 SH2 domain binds RET9 in a kinase-dependent manner, and that a RET9 Y1062F mutation abrogates this interaction (Fig. 5A). In contrast, mutation of the RET PDZ-binding motif (F1072A) had minimal effect on binding of the GRB10 SH2 domain to RET9, although it was required for NEDD4 binding (Fig. 4C), suggesting that this RET motif could be involved in additional mechanisms of NEDD4 recruitment.

The RET9 C-terminal PDZ binding motif has been shown to be a docking site for the PDZ domain protein SHANK3 (Schuetz et al., 2004); however, interactions with other PDZ proteins have not been investigated. Of note, another SHANK family member, SHANK2, localizes to clathrin-rich regions at the cell membrane and to early endosomes (Dobrinskikh et al., 2010), where ubiquitylated proteins are enriched, and has a similar tissue distribution to RET in vivo (Lim et al., 1999), suggesting it might also be a candidate RET binding protein. In co-immunoprecipitation assays, we showed that SHANK2 binds RET9 but not RET51 (Fig. 5B). This interaction was lost when the RET9 PDZ binding motif was mutated (F1072A) or deleted, but was not affected by mutations of RET9 phospho-tyrosines at Y1062 (GRB10 binding site) or Y1015 (PLC-γ binding site) (Fig. 5C). Together these data suggest that GRB10 and SHANK2 were recruited to distinct sites in the RET9 C-terminus. However, as these sites lie in close proximity (Fig. 1A), we investigated whether SHANK2 and GRB10 binding to the RET9 C-terminus were independent. In HEK293 cells transiently expressing RET9 and both full-length GRB10 and SHANK2, we found that both adaptors associated with wild-type RET9 (Fig. 5D). Consistent with our previous findings, SHANK2 bound the Y1062F but not the F1072A RET9 mutants. Interestingly, recruitment of full-length GRB10 to RET9 was abrogated by both these mutations in the presence of SHANK2 (Fig. 5D), suggesting that GRB10 binding at Y1062 was not independent of SHANK2, and that SHANK2 binding at the RET9 C-terminus may facilitate or stabilize the RET9–GRB10 interaction. Furthermore, GRB10 interacted with SHANK2, even in the absence of RET, and the RET F1072A mutation did not affect this interaction, suggesting that these adaptors may associate constitutively (Fig. 5D).

As our data suggest that adaptor binding at both RET9 Y1062 and the PDZ binding motif, FTRF₁₀₇₂, are required for NEDD4 binding, and that SHANK2 recruitment may affect GRB10 binding, we further characterized the roles of GRB10 and SHANK2 in NEDD4 recruitment. We showed that expression of increasing amounts of SHANK2 significantly increased interactions of RET9 and NEDD4 (Fig. 5E). In addition, we showed that expression of either excess GRB10 or SHANK2 enhanced the association of NEDD4 and RET9 (Fig. 5E,F) and expression of both these adaptors further enhanced the interaction (Fig. 5F). Together, our data suggest that a multi-protein complex assembles at the tail of RET9, which may enhance or stabilize recruitment of NEDD4 to RET.

The types of polyubiquitin chains decorating a substrate can promote different interactions or functional outcomes (Yau and Rape, 2016). We evaluated the specific types of polyubiquitin chains ligated onto RET9 and RET51 using ubiquitin chain restriction analysis (UbiCREST) with a panel of deubiquitylating (DUB) enzymes that recognize distinct patterns of ubiquitin linkage (Hospenthal et al., 2015). RET9 and RET51 transiently expressed in HEK293 cells were immunoprecipitated after treatment with GDNF for 15 min, subjected to DUB treatment and cleavage products resolved by SDS-PAGE. Released ubiquitin chains were identified by silver staining (Fig. 6A). We detected prominent mono- and di-ubiquitin cleavage products for both RET9 and RET51 treated with the DUB OTUB1, which recognizes K48 polyubiquitin chains, generally linked to protein degradation. We also identified ubiquitin cleavage products for the AMSH DUB, which cleaves K63 ubiquitin linkages, associated with recruitment of signalling complexes implicated in many cellular processes, including endosomal protein sorting (Yau and Rape, 2016). K63 linkages appeared to be relatively more prevalent in RET51 cleavage products compared with RET9. Furthermore, reductions in the intensity of the remaining ubiquitylation of RET9 after OTUB1 treatment, and for RET51 after both OTUB1 and AMSH treatments (Fig. 6B), confirmed release of ubiquitin chains by these DUBs and were consistent with relatively greater reductions in RET51 ubiquitylation than RET9 after cleavage of K63 linkages. We saw complete loss of ubiquitylation of RET9 and RET51 using a USP2 DUB control, which cleaves all ubiquitin chains (Fig. 6B) but did not detect appreciable ubiquitin chain cleavage by other DUBs for either RET9 or RET51 (Fig. 6A), suggesting that other linkages were not prevalent in these samples. Our results are consistent with previous data linking RET ubiquitylation to protein degradation (Scott et al., 2005), but suggest that it may also have roles in other processes, such as protein sorting and recruitment of endosomal trafficking adaptors.

We have previously demonstrated that RET9 and RET51 are internalized at different rates and that interactions with the AP2–clathrin adaptor complex, which promotes receptor recruitment to CCPs, is an important determinant of this process (Crupi et al., 2015a; Richardson et al., 2012). To further assess the individual contributions of NEDD4 and CBL to internalization of RET9 and RET51, respectively, we used total internal reflection fluorescence (TIRF) microscopy to visualize the localization of RET isoforms at the cell surface. We measured the total intensity of RET within CCPs in response to GDNF stimulation, using TIRF microscopy coupled to automatic CCP detection and analysis (cmeAnalysis; MatLab), as we have used previously to monitor RET isoform recruitment to CCPs (Crupi et al., 2015a). CCPs were detected by AP2 staining, which exclusively detects cell-surface clathrin structures (Aguet et al., 2013). Mock KD and NEDD4 or GRB2 KD or CBL KO were generated in the SH-SY5Y neuroblastoma cell line, which endogenously expresses both RET isoforms. Cells were stimulated with GDNF, then fixed and subjected to immunofluorescence staining for RET isoforms and AP2 (Fig. 6C-E). Based on the timing of RET isoform recruitment to CCPs in our previous studies (Crupi et al., 2015a), we assessed GDNF-mediated localization of RET to AP2 puncta after 5 min for RET51 and after 15 min for RET9 compared with no treatment (0 min GDNF). Consistent with our previous studies (Crupi et al., 2015a), GDNF stimulation resulted in time-dependent localization of endogenous RET receptors to AP2 puncta at the cell surface in Mock KD cells (Fig. 6C-E). We detected RET51 at AP2 puncta following 5 min of GDNF treatment, and RET9 after 15 min (Fig. 6C-E). Minimal overlap of AP2 puncta with RET staining was observed in the absence of GDNF. In NEDD4 KD cells, we detected few CCPs harbouring RET9, even upon GDNF stimulation (Fig. 6C). Automated detection of AP2 structures (Crupi et al., 2015a), followed by systematic quantification of RET localized therein, demonstrated that the GDNF-dependent RET9 recruitment to AP2 puncta was significantly impaired in NEDD4 KD cells (Fig. 6C). Interestingly, the ligand-dependent increase in localization of RET51 to AP2 puncta in response to GDNF was unaffected in both CBL KO (Fig. 6D) and GRB2 KD cells (Fig. 6E). Localization of RET51 to puncta was slightly greater in CBL KD cells under all conditions, probably as a result of the larger and flatter morphology of these cells. Conversely, RET51 localization to puncta was lower overall in GRB2 KD cells, likely reflecting the pleiotropic effects of GRB2 on diverse cellular processes. Together, our data suggest that ubiquitylation of RET9 by NEDD4 may promote its recruitment to CCPs and subsequent internalization, but that CBL-mediated ubiquitylation is not essential for these early steps in recruitment of RET51 to CCPs for internalization.

Ubiquitylation of RTKs is an important determinant of receptor internalization from CCPs at the cell membrane into endosomal compartments (Haglund and Dikic, 2012). The roles of CBL- and NEDD4-mediated ubiquitylation in RET isoform internalization from the cell surface have not been explored. We used surface biotinylation assays to investigate the effects of ubiquitylation on RET9 and RET51 internalization in CBL KO and GRB2 and NEDD4 shRNA KD SH-SY5Y cells compared with Mock KD control cells (Fig. 7). Consistent with our previous studies (Crupi et al., 2015a; Richardson et al., 2012), RET51 internalization was detected by 5 min of GDNF treatment in Mock KD cells, whereas RET9 internalized more slowly, with appreciable internalization detected only after 15 min (Fig. 7A-C). NEDD4 KD reduced RET9 internalization, such that appreciable internalization was only detectable after 30 min of GDNF treatment in these cells (Fig. 7A). RET51 internalization was not affected by knockdown of NEDD4. CBL KO had little effect on the rates of RET51 or RET9 internalization (Fig. 7B), suggesting that ubiquitylation is not essential for RET51 internalization, or that other CBL family members could compensate for loss of CBL in these cells. As our data indicate that depletion of the GRB2 adaptor protein abrogates CBL interaction with, and ubiquitylation of, RET51 (Fig. 3A,B) we assessed the effect of GRB2 KD on RET internalization (Fig. 7C). In GRB2 KD SH-SY5Y cells, RET51 internalized more slowly than in Mock KD cells, and appreciable internalization was detected only after 15 min of GDNF treatment (Fig. 7C). Internalization of RET9 was not affected by GRB2 KD. Together, our data identify isoform-specific differences in RET ubiquitylation and suggest that these differences may modulate internalization of RET9 and RET51 into endosomal compartments.

Multiple ubiquitin ligases are involved in the regulation of signal transduction pathways following ligand-mediated activation of receptor tyrosine kinases. Ubiquitylation of RTKs can promote endocytosis, as well as target receptors for degradation or recycling (Haglund and Dikic, 2012). Consistent with previous studies (Calco et al., 2014; Kales et al., 2014; Scott et al., 2005), we have shown that the RET9 and RET51 isoforms are ubiquitylated upon GDNF stimulation. Our data, which focus on ubiquitylation of wild-type RET isoforms rather than oncogenic activated mutants, allow us to evaluate early events in response to GDNF-mediated RET ubiquitylation and suggest there may be differences in these processes between RET9 and RET51.

We demonstrate that the CBL E3 ligase promotes ubiquitylation of RET51 but not RET9. Our data indicate that CBL is recruited to RET51 via interactions with the GRB2 adaptor protein at phospho-tyrosine pY1096, a site that is unique to RET51 (Fig. 8A). CBL plays well-characterized roles in ubiquitylation of RTKs such as EGFR and MET, and is implicated in RTK internalization and trafficking (Ma et al., 2013; Mohapatra et al., 2013). We showed that both CBL KO and GRB2 KD reduced RET51 ubiquitylation. We saw no residual interaction of RET51 and CBL in the GRB2 KD cells, suggesting that other adaptors or RET binding sites had minimal roles in linking CBL to RET51. Interestingly, neither CBL KO nor GRB2 KD inhibited ligand-dependent recruitment of RET51 to AP2 puncta, which are markers of CCPs. We have previously shown that RET51 directly binds the AP2 clathrin adaptor and that this is required for receptor internalization (Crupi et al., 2015a). Our data suggest that indirect interactions between RET51 and AP2, through recruiting ubiquitin-binding proteins, do not significantly enhance ligand-dependent RET51 localization to AP2 puncta. Interestingly, however, GRB2 KD, but not CBL KO, did lead to delayed internalization of RET51, suggesting that there may be roles for ubiquitylation in recruiting other endocytic proteins at these later steps. Interactions between RET and the Cbl-3/c and Cbl-b CBL family members have also been identified in some cell types (Calco et al., 2014; Kales et al., 2014; Pierchala et al., 2006), perhaps suggesting that other members of this family could compensate for CBL depletion but not GRB2 loss in RET51 ubiquitylation and internalization in SH-SY5Y. Together, our data implicate a GRB2–CBL complex in RET51 internalization in SH-SY5Y cells, consistent with recent studies demonstrating that depletion of multiple CBL family members was not sufficient to abrogate EGFR internalization but that GRB2 KD blocked receptor endocytosis (Fortian et al., 2015).

We showed that wild-type RET9 is not associated with, or appreciably ubiquitylated by CBL, suggesting that other ubiquitin ligases could be involved. NEDD4 is the prototypical member of an E3 ligase family with roles in fetal growth and nervous system development (Boase and Kumar, 2015). NEDD4 has previously been shown to ubiquitylate several RTKs, including neurotrophic tyrosine kinase receptor type 1 (NTRK1), targeting it for endocytosis and degradation (Arévalo et al., 2006; Cooper et al., 2015; Takahashi et al., 2011; Vecchione et al., 2003). Our data demonstrate that RET9 is primarily ubiquitylated by the NEDD4 ubiquitin ligase in response to GDNF. Furthermore, depletion of NEDD4 impaired ligand-dependent RET9 localization to CCPs and decreased the rate of RET internalization, suggesting that NEDD4-mediated ubiquitylation is important both for early recruitment of AP2 to RET9 and for subsequent endocytosis. We showed that the RET9–NEDD4 association required both the RET9-specific phospho-tyrosine binding site pY1062 and C-terminal PDZ-binding motif (FTRF₁₀₇₂), suggesting that multiple adaptors may be needed to enhance and stabilize this interaction.

The association of NEDD4 with RTKs is frequently mediated through interactions with the GRB10 adaptor, which binds receptors in a phosphorylation-dependent manner and recruits NEDD4 to facilitate receptor ubiquitylation (Huang and Szebenyi, 2010; Morrione et al., 1999; Vecchione et al., 2003). RET has been previously shown to bind GRB10 upon activation (Pandey et al., 1995), although the specific RET residues involved in the interaction were not determined. Here, we demonstrate that the GRB10 SH2 domain binds RET9 at pY1062, a site that is also required for NEDD4-mediated ubiquitylation of RET9. A similar GRB10-mediated interaction between NEDD4 and the insulin-like growth factor receptor (IGF-IR) promotes ubiquitylation and internalization of the IGF-IR (Monami et al., 2008; Vecchione et al., 2003). Interestingly, we found that binding of a full-length GRB10 protein was also impaired by mutation of the RET9 C-terminal PDZ-binding motif, suggesting that additional complex members were involved in NEDD4 recruitment.

RET has been previously shown to recruit adaptor proteins through interactions with the PDZ-binding motif at the RET9 C-terminus (Schuetz et al., 2004). The SHANK2 PDZ scaffolding protein is recruited to a number of cell surface receptors and has been shown to internalize with clathrin from the cell membrane and localize to endosomes and lysosomes (Dobrinskikh et al., 2010), making it a candidate for RET binding. Here, we show that SHANK2 binds the RET9 C-terminal PDZ-binding motif in a kinase-dependent manner and that NEDD4 recruitment to RET9 is increased in the presence of excess SHANK2. RET9–NEDD4 association was also increased in the presence of excess GRB10 and further enhanced by GRB10 and SHANK2 co-expression. Together, our data suggest that both SHANK2 and GRB10 are involved in the recruitment of NEDD4 to activated RET9 receptors and that the NEDD4–GRB10–SHANK2 complex plays important roles in RET9 ubiquitylation and endocytosis (Fig. 8B).

Although traditionally PDZ protein binding was thought to be phosphorylation independent, our data show that binding of SHANK2 to RET9 requires an active kinase but does not occur through the pY1062 site unique to RET9, suggesting that conformational changes in the RET9 C-terminus associated with autophosphorylation may be required for SHANK2 recruitment (Fig. 8B). Recent studies have shown that RET9 kinase activation leads to a cascade of autophosphorylation events commencing with Y1062 phosphorylation (Plaza-Menacho et al., 2014). These data suggest that RET9 Y1062 accessibility could contribute to its more rapid site phosphorylation and recruitment of signalling proteins (Plaza-Menacho et al., 2014). As we showed that GRB10 and SHANK2 associate constitutively, SHANK2–RET binding may facilitate association of GRB10 with the conformationally activated RET9, promoting subsequent recruitment of NEDD4 to the complex via GRB10. Our data suggest cis-regulatory roles for phosphorylation of RET9 C-terminal tyrosines by also enhancing accessibility for PDZ protein binding to the C-terminus of RET9.

Ubiquitylation of membrane receptors, originally identified as a signal for degradation, also plays important roles in regulating protein interactions required for diverse cellular processes (Yau and Rape, 2016). Covalently attached polyubiquitin chains can be formed through multiple ubiquitin linkages, most frequently through lysines K48 or K63 which have been associated with intracellular routing of ubiquitylated substrates (Swatek and Komander, 2016; Yau and Rape, 2016). K48 linkages are able to target proteins for proteasomal degradation, while K63 linkages appear to have more diverse cellular functions and may orchestrate receptor signal transduction, endocytosis and intracellular trafficking to multiple compartments as well as targeting them for degradation (Erpapazoglou et al., 2014; Hanyaloglu et al., 2005; Raiborg et al., 2002). Here, we showed that both RET isoforms are associated with K48 and K63 ubiquitin chains in response to GDNF. Interestingly, we showed that K63 linkages are relatively more prevalent for RET51 than RET9. K63 polyubiquitylation of substrates recruits ubiquitin-binding proteins that are important for sorting RTKs for proteasomal or lysosomal degradation, or recycling to the cell membrane (Erpapazoglou et al., 2014; Swatek and Komander, 2016). We have previously shown that internalized RET51, but not RET9, can be sorted to RAB11-positive recycling endosomes for return to the cell surface, or targeted for degradation (Crupi et al., 2015b; Richardson et al., 2012), suggesting that decoration with specific polyubiquitin chains may provide an important layer of regulation for RET51 sorting in the early and late endosomes.

Our data demonstrate that RET9 and RET51 have differential mechanisms of receptor ubiquitylation in response to activation by GDNF. We show that RET51 ubiquitylation is modulated by the CBL E3 ligase recruited through interactions with the GRB2 adaptor protein. Loss of GRB2-mediated interactions abrogates ubiquitylation and reduces RET51 internalization from the cell membrane. In contrast, RET9 recruits the NEDD4 ubiquitin ligase via a multicomponent adaptor protein complex including SHANK2 and GRB10 that interacts with multiple RET9 C-terminal binding motifs. Our data are consistent with previous studies showing distinct kinetics of RET9 and RET51 localization to CCPs in response to GDNF, which suggests the involvement of differential internalization signals for each RET isoform (Crupi et al., 2015a). Modification of RET isoforms by different ubiquitin ligases may promote differential interactions of RET isoforms with ubiquitin-binding endosomal trafficking molecules, that could, in part, contribute to the differential trafficking and signalling patterns downstream of RET activation that have previously been observed (Crupi et al., 2015a; Richardson et al., 2006, 2012). Our data, described here, support a model in which ubiquitylation may contribute to these processes, and these contributions may differ between RET9 and RET51, but other factors including membrane and compartment localization and signalling protein complex interactions are also important players. RET's roles in both development and cancer suggest that its precise regulation is critical for maintaining cellular homeostasis, making it an important target for anti-cancer therapies (Mulligan, 2014). Understanding of the mechanisms by which RET is down-regulated following ligand-mediated activation are key to the development of such potential new treatments.

Expression constructs encoding the full-length RET9 and RET51 isoforms, and GFRα1, have been previously described (Richardson et al., 2006, 2012). Truncations and point mutations were introduced by PCR-based site-directed mutagenesis and sequences verified. The GST-fusion construct encoding human icRET51 (residues 664-1114) has been previously described (Crupi et al., 2015a).

FLAG- and HA-tagged CBL expression constructs were gifts from Drs Ye-Guang Chen (Zuo et al., 2013) and Wallace Langdon (Peschard et al., 2001), respectively. Full-length NEDD4 obtained from Dr Madhavi Kadakia (Leonard et al., 2013) was modified to include an N-terminal 2× FLAG tag. GST–NEDD4 was provided by Dr Xiaolong Yang (Queen's University, Kingston, ON, Canada). The HA–ubiquitin construct (Treier et al., 1994) was provided by Dr Dirk Bohmann. GST–GRB10 was provided by Dr Justus Duyster (Bai et al., 1998). Full-length GRB10 (isoform A) was generated by PCR and sequence verified. The HA–SHANK2 construct was obtained from Dr Min Goo Lee (Kim et al., 2004). His-tagged DAB2, SHC PTB domain and CBL–SH2 constructs were obtained from Dr Gavin MacBeath (Kaushansky et al., 2010; Koytiger et al., 2013). shRNAs against human NEDD4 and GRB2 were obtained from Dharmacon (GE Healthcare, Mississauga ON, Canada). CRISPR constructs targeting CBL were obtained from Horizon Discovery Group (Cambridge, UK).

HEK293 (CRL-1573) and SH-SY5Y (CRL-2266) cells were obtained from ATCC (Manassus, VA, Canada). HEK293 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) (Millipore Sigma, Oakville, ON, Canada) supplemented with 10% fetal bovine serum (FBS) (Millipore Sigma). Cells were transfected using Lipofectamine 2000 according to the manufacturer's instructions (Life Technologies, Burlington, ON, Canada). Endogenous RET expression was stimulated in SH-SY5Y cells with 10 µM retinoic acid (Millipore Sigma) overnight prior to each experiment. RET activation was induced with 50 ng/ml GDNF (Peprotech, Rocky Hill, NJ, USA) for the indicated times.

Polyclonal shRNA knockdown (KD) or empty vector control (Mock KD) cell lines were generated by lentiviral transduction as previously described (Crupi et al., 2015a) and selected with puromycin (1.5 µg/ml) to generate stable polyclonal KD HEK293 and SH-SY5Y cell lines. CRISPR/Cas9 gene editing was used to generate CBL knockout (CBL KO) in HEK293 and SH-SY5Y cells.

Cells were washed with phosphate-buffered saline (PBS) and lysed in lysis buffer [20 mM Tris-HCl (pH 7.8), 150 mM NaCl, 1 mM sodium orthovanadate, 1% Igepal, 2 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 10 µg/ml aprotinin, and 10 µg/ml leupeptin], as previously described (Richardson et al., 2006, 2012). For UbiCREST analyses, 10 mM N-ethylmaleimide and 100 mM NaF were added to the lysis buffer. Protein concentrations were determined using the BCA protein assay kit, (ThermoFisher Scientific, Waltham, MA, USA) according to the manufacturer's instructions. For immunoprecipitations, 1 mg whole cell lysate was incubated with 1 μg antibody for 2-4 h at 4°C with agitation, followed by addition of 40 μl Protein A/G agarose (Santa Cruz Biotechnologies, Santa Cruz, CA, USA) and overnight incubation at 4°C with agitation. Protein complexes were collected by centrifugation at 2000 g, washed four times with lysis buffer and resuspended in Laemmli buffer (180 mM Tris-HCl, pH 6.8, 2% SDS, 10% glycerol, 0.01% Bromophenol Blue) with 3% 2-mercaptoethanol. Proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes (Bio-Rad Laboratories, Mississauga, ON, Canada), as described (Richardson et al., 2006, 2012).

RET expression was detected using C-19G (RET9), C-20G (RET51) or H300 (pan-RET) (Santa Cruz Biotechnology), D3D8R or E1N8X (pan-RET) (Cell Signaling Technology, Beverly, MA) antibodies. RET tyrosine phosphorylation was detected using the anti-phospho-RET (pY905, Cell Signaling, #3221) antibody. NEDD4 (H-135), CBL (C-15), GRB2 (C-23) and SHANK2 (H-150) antibodies were from Santa Cruz Biotechnology. The GRB10 antibody was from Cell Signaling (#3702). Anti-γ-tubulin and anti-phospho-tyrosine (4G10) antibodies were obtained from Millipore Sigma. Antibodies for detection of epitope tags were: GST (S-tag-05); HIS (H-15); HA (F7 or Y11) (Santa Cruz Biotechnology); FLAG-M2 (Millipore Sigma).

HEK293 cells were transiently co-transfected with the indicated constructs and HA-tagged ubiquitin. Cells were serum starved overnight followed by treatment for 30 min with MG132 (5 μM; Millipore Sigma) prior to activation of RET with GDNF for 15 min. Cell lysates were collected and immunoprecipitated with the indicated antibodies and immunoblotted, as described.

Ubiquitin chain restriction analysis was performed using the UbiCREST kit (BostonBiochem, Cambridge, MA, USA) according to the manufacturer's instructions. Briefly, RET9 or RET51 was immunoprecipitated from whole cell lysates of transiently transfected HEK293 cells treated with GDNF (15 min), as described above. Precipitated protein complexes were washed three times in PBS with 0.05% Tween-20 and with deubiquitylase (DUB) dilution buffer. Beads were distributed into reaction tubes, suspended in DUB dilution buffer containing individual DUB enzymes and incubated at 37°C for 30 min. Supernatants were collected, resolved by SDS-PAGE and visualized using a Bio-Rad Silver Staining Plus kit. Residual beads were suspended in Laemmli buffer with 1% 2-mercaptoethanol and analysed by western blotting. DUB enzymes used, and their cleavage specificities were: OTUD3 (cleaves K⁶- and K¹¹-linked polyubiquitin); CEZANNE (K¹¹); His6-TRABID (K²⁹, K³³ and K⁶³); OTUB1 (K⁴⁸); GST-AMSH (K⁶³); YOD1 (K⁶, K¹¹, K²⁷, K²⁹, and K³³); OTULIN (linear polyubiquitin only); and USP2, which cleaves most ubiquitin linkages (used as a positive control).

Expression constructs for GST or GST-tagged proteins and His-tagged SH2 and PTB protein domains were grown in BL21 codon plus –RP Escherichia coli. Harvest and purification was performed as previously described (Crupi et al., 2015a).

Purified His-tagged proteins (2 μg) were resolved by SDS-PAGE gels and transferred to nitrocellulose membranes. Membranes were blocked in 5% milk in TBS with 0.1% Tween-20 (TBS-T) for 1-2 h at room temperature. Purified GST-icRET51 or GST-alone (1 μg/ml) proteins were incubated with blocked membranes overnight at 4°C. Membranes were washed in TBS-T and probed with an anti-GST or anti-His antibodies for 2 h at room temperature. Blots were washed with TBS-T and incubated with secondary antibody for 1 h.

5 µg GST-fusion protein was re-bound to 10 µl glutathione–Sepharose 4B beads (GE Healthcare) and incubated with 1 mg whole cell lysates for 2-4 h. Beads were washed four times with lysis buffer prior to the addition of Laemmli buffer and analysis by SDS-PAGE and immunoblotting.

Biotinylation of cell surface proteins was performed in SH-SY5Y cells with the indicated shRNA knockdown or with CRISPR/Cas9-generated knockout of CBL, as previously described (Crupi et al., 2015b; Richardson et al., 2012). Briefly, cells were serum starved for 3 h prior to labelling with biotin and activation of RET with 50 ng/ml GDNF for the indicated times. Remaining biotin label was stripped from the cell surface using 50 mM MeSNa. Cells were lysed and the biotinylated proteins recovered with streptavidin-conjugated agarose (Thermo Fisher Scientific).

Experiments were repeated independently a minimum of 3 times. Means and standard errors (±s.e.m.) were calculated for each experimental condition, and significance between conditions was determined using a two-tailed Student's t-test for α=0.05. Densitometry was performed using ImagePro (Media Cybernetics, Rockville, MD, USA).

SH-SY5Y cell lines were seeded onto poly-L-lysine-coated #1.5 glass coverslips (Electron Microscopy Sciences, Hatfield, PA, USA) at 350,000 cells per well in six-well plates. Cells were treated with retinoic acid (10 μM) overnight and serum starved for 3 h prior to addition of GDNF ligand (50 ng/ml) for the indicated times. Cells were fixed with 3% paraformaldehyde, quenched with 100 mM glycine in PBS, permeabilized in 0.1% Triton X-100 and blocked in 3% BSA. Cells were incubated with primary antibodies against AP2 (AP6, Abcam) (1:400) and RET9 (C19R, Santa Cruz) or RET51 (EPR2871, Abcam, Cambridge, UK) (1:50) for 2 h at room temperature followed by Hoechst stain (1:500), Alexa-Fluor-594 and -488 secondary antibodies (1:200) (Thermo Fisher Scientific) for 1 h at room temperature. Coverslips were wet mounted on glass slides in PBS for imaging. Some TIRF microscopy experiments were performed using a 150× (NA 1.45) objective on an Olympus IX81 instrument equipped with Cell TIRF modules (Olympus Canada, Richmond Hill, ON) using 491 nm (50 mW) and 561 nm (50 mW) laser illumination and 520/30, 624/50 emission filters. Images were acquired using a C9100-13 EM-CCD camera (Hamamatsu Corp., Bridgewater, NJ, USA). Additional TIRF microscopy experiments were performed on a Quorum Diskovery TIRF microscope (Guelph, ON, Canada), comprising a Leica DMi8 microscope equipped with a 63× (NA 1.49) TIRF objective with a 1.8× camera relay (total 108× magnification). Imaging was done using 488 nm and 561 nm laser illumination using 527/30 and 630/75 emission filters. Images were acquired using a Zyla 4.2Plus sCMOS camera (Hamamatsu).

Automated detection of AP2 puncta and robust analyses were implemented in MATLAB (MathWorks) cmeAnalysis software (lccb.hms.harvard.edu/software.html). Briefly, this method uses a Gaussian-model-based approach to detect and analyse diffraction-limited AP2 structures (Crupi et al., 2015a), in this case determined in single frames within the green AP2 channel (‘master’ channel). RET fluorescence intensities at AP2 puncta, corresponding to the amplitude of the Gaussian model in the red channel (‘slave’ channel) within detected AP2 puncta were also obtained for each AP2 object. Average values of RET fluorescence within AP2 puncta in fixed cells were normalized to total RET expression in the TIRF field and are shown as means for a minimum of 50,000 AP2 puncta per condition. The statistics (ANOVA and Bonferroni post test) show differences in the relevant comparisons in a minimum of four independent experiments.

The authors would like to thank Drs Xiaolong Yang, Steven Smith and Scott Andrew for helpful discussion, and M. Woodside and P. Paroutis at the Hospital for Sick Children Imaging Facility for technical assistance.