Exon Skipping in the RET Gene Encodes Novel Isoforms That Differentially Regulate RET Protein Signal Transduction

Abstract

Rearranged during transfection (RET), a receptor tyrosine kinase that is activated by the glial cell line-derived neurotrophic factor family ligands (GFLs), plays a crucial role in the development and function of the nervous system and additionally is required for kidney development and spermatogenesis. RET encodes a transmembrane receptor that is 20 exons long and produces two known protein isoforms differing in C-terminal amino acid composition, referred to as RET9 and RET51. Studies of human pheochromocytomas identified two additional novel transcripts involving the skipping of exon 3 or exons 3, 4, and 5 and are referred to as RET^ΔE3 and RET^ΔE345, respectively. Here we report the presence of Ret^ΔE3 and Ret^ΔE345 in zebrafish, mice, and rats and show that these transcripts are dynamically expressed throughout development of the CNS, peripheral nervous system, and kidneys. We further explore the biochemical properties of these isoforms, demonstrating that, like full-length RET, RET^ΔE3 and RET^ΔE345 are trafficked to the cell surface, interact with all four GFRα co-receptors, and have the ability to heterodimerize with full-length RET. Signaling experiments indicate that RET^ΔE3 is phosphorylated in a similar manner to full-length RET. RET^ΔE345, in contrast, displays higher baseline autophosphorylation, specifically on the catalytic tyrosine, Tyr⁹⁰⁵, and also on one of the most important signaling residues, Tyr¹⁰⁶². These data provide the first evidence for a physiologic role of these isoforms in RET pathway function.

Introduction

RET is a receptor tyrosine kinase that is critical for kidney morphogenesis, spermatogenesis, and development of the nervous system (1–4). RET is activated by a family of four growth factors known as the glial cell line-derived neurotrophic factor (GDNF)² family ligands (GFLs), which includes GDNF, neurturin, artemin, and persephin (1). Each GFL binds to one of four cognate glycosylphosphatidylinositol-anchored co-receptors known as the GDNF family receptor αs (GFRαs) (2, 5). The GFL-GFRα complex binds to RET, inducing RET dimerization and subsequent autophosphorylation on multiple tyrosine residues within the intracellular tyrosine kinase domain. This enhances tyrosine kinase activity and initiates the association of adaptor proteins and enzymes that trigger multiple second messenger cascades (6).

The presence of two major Ret isoforms, RET9 and RET51, has been extensively described in the literature, and a third isoform, RET43, has also been observed in humans (7–9). Ret is 20 exons long, and Ret9 and Ret51 transcripts differ in alternative splicing of intron 19. Intron 19 in the Ret51 transcript is excised properly, whereas, in the Ret9 transcript, the intron is retained, changing the reading frame and inserting a premature stop codon into the amino acid sequence. This creates a unique nine-amino acid C-terminal sequence for RET9 and a unique 51-amino acid C-terminal sequence for RET51. Interestingly, these two different isoforms display marked differences in their degradation and function (7, 10, 11).

Three additional Ret transcripts have been reported in various tumor sources as well as adult human tissues (12). These novel transcripts are a product of exon skipping in the 5′ region of RET, which encodes for the extracellular domain of the protein. Skipping of exons 3 (RET^ΔE3) or exons 3, 4 and 5 (RET^ΔE345) gives rise to transcripts that encode for full-length Ret proteins but with deletions in the extracellular domain, specifically cadherin-like domain 1 (CLD1) or CLD1–3, respectively. These deletions are hypothesized to change the extracellular domain structure as well as binding to GFL-GFRα complexes and may impact the overall stability of the proteins. The skipping of exons 3 and 4 results in a frameshift that culminates in a premature stop. This protein encodes only a small portion of the RET extracellular domain with no transmembrane or intracellular domain. Because this transcript is likely to be subjected to nonsense-mediated decay, it is unlikely to be involved in RET signal transduction.

Here we show that Ret^ΔE3 and Ret^ΔE345 transcripts are conserved in vertebrates and that the mRNA and proteins of these splice variants are expressed throughout the nervous system in mice. We also show that these isoforms are trafficked to the cell surface and that both isoforms interact with all four GFRαs. Additionally, we find that RET^ΔE3 is phosphorylated to a similar level as full-length RET and is activated in a GDNF-dependent manner. However, RET^ΔE345 displays higher baseline autophosphorylation, specifically on the catalytic tyrosine Tyr⁹⁰⁵, and also on an additional signaling tyrosine, Tyr¹⁰⁶². Interestingly, RET^ΔE345 is not activated in a GDNF-dependent manner. Taken together, these isoforms may have unique and important unidentified roles in the development and maintenance of the nervous system and kidneys as well as in the pathophysiology of neuroendocrine gland diseases.

Results

Exons 3, 4, and 5 of Ret Are Not Highly Conserved, but Ret^ΔE3 and Ret^ΔE345 Transcripts Are Observed in Several Organisms

It has been described previously that the intracellular domain of RET, which contains a tyrosine kinase domain, is more highly conserved than the extracellular domain (13). To better understand the level of conservation of amino acids encoded by each exon, we compared sequences from zebrafish (Danio rerio), mouse (Mus musculus), rat (Rattus norvegicus), and human (Homo sapiens). We found that the entire intracellular domain was more than 75% conserved at the amino acid level. This was not surprising because mutations in many of these highly conserved regions give rise to loss of function (e.g. Hirschsprung disease) or gain of function (e.g. multiple endocrine neoplasia type 2A (MEN2A) and 2B (MEN2B)) phenotypes (Fig. 1A). Interestingly, exons 3, 4, and 5 are less than 50% conserved between these species. Although this lack of conservation is suggestive that these regions of RET are not as functionally important, it could also be interpreted as allowing for amino acid flexibility between the different RET isoforms.

FIGURE 1.

Removal of exons 3 or 3, 4, and 5 of RET creates proteins with large deletions in the extracellular domain, and transcripts encoding these receptors are expressed in multiple organisms. A, comparison of the amino acid sequences of RET between zebrafish, mice, rats, and humans was performed, and the conservation within each exon is indicated by color. Interestingly, the amino acids encoded in exons 3, 4, and 5 are the least conserved. Additionally, the impact of exon skipping on the extracellular domain within the protein is shown. B, Ret^ΔE3 and Ret^ΔE345 transcripts are detectable in zebrafish (D. rerio), mice (M. musculus), and rats (R. norvegicus) by qPCR. Like human RET^ΔE3 and RET^ΔE345, alternative splicing of exon 3 or exons 3, 4, and 5 in these vertebrates are predicted to encode full-length, transmembrane RET proteins with deletions in the extracellular domain. C, deleted regions are shown in red, projected onto a structural model for RET cadherin domains 1–4 derived from PDB code 4UX8.

Our initial sequence analysis suggested that Ret^ΔE3 and Ret^ΔE345 transcripts could be expressed in additional vertebrates other than H. sapiens. These transcripts were originally identified in human kidney and substantia nigra fetal tissues (12). Sequence analysis demonstrated that exon skipping also encodes potential full-length transcripts in zebrafish, mice, and rats (Fig. 1A). To determine whether these novel transcripts were expressed in these organisms, species-specific primers were created to identify each of the Ret transcripts: full-length Ret, Ret^ΔE3, and Ret^ΔE345. RT-PCR analysis identified the presence of Ret^ΔE3 and Ret^ΔE345 from 48 hours post fertilization zebrafish embryos, E19.5 rat dorsal root ganglia, and E18.5 mouse spinal cord (Fig. 1B). Sequence analysis confirmed these amplicons to be Ret^ΔE3 and Ret^ΔE345 (data not shown). This is the first time these transcripts have been identified in vertebrates other than H. sapiens. Taken together, these data indicate that 5′ exon skipping is conserved and results in the expression of Ret^ΔE3 and Ret^ΔE345 transcripts.

Exon Skipping Creates Large Deletions in the Extracellular Domain of RET

To understand the impact deletions in the extracellular domain would have on RET, we analyzed the deletions in terms of a CLD1–4 model derived from electron microscopy and small angle x-ray scattering analyses (14). The main consequence of deletion of exon 3 is the removal of half of CLD1 (residues 113–150) and half of CLD2 (residues 154–208). Each domain adopts a β sandwich cadherin fold comprised of two β sheets separated by numerous buried hydrophobic residues. Elimination of exon 3 removes one sheet from CLD1 and several β strands from both sheets of CLD2 (Fig. 1C). Therefore, it is likely that both CLD domains are substantially perturbed leaving β strands without the cadherin fold structure and exposing some hydrophobic residues. Exon 3 removal also eliminates a highly constrained Cis-Pro loop (Cys¹³⁷-Cys¹⁴²) unique to higher vertebrate RET sequences. The portion of CLD1–2 removed coincides with regions stabilizing the clam shell arrangement of CLD1–2 (15). This would result in a more open and extended arrangement for the remaining parts of CLD1 and CLD2 with much greater flexibility between the domains. An odd number of cysteine residues (three in total) are deleted by removing exon 3, including the Cys¹⁵⁷-Cys¹⁹⁷ disulfide. Cys¹⁶⁶ is also removed, leaving its partner Cys²⁴³ unpaired but buried. The two unpaired cysteines, Cys⁸⁷ and Cys²¹⁶, which constitute a known folding bottleneck in wild-type RET, are left untouched. We expect RET^ΔE3 to be a less stable transmembrane protein with a short half-life that retains at least some ligand-binding properties.

For RET^ΔE345, removal of exons 3, 4, and 5 effectively eliminates much of CLD1–3 but leaves CLD4-cysteine-rich domain (CRD) intact (Fig. 1C). This region by itself was shown to be insufficient to bind ligand but retains at least one important binding epitope (i.e. necessary for ligand binding but not sufficient) (14). Many receptor tyrosine kinases have extracellular domains that have autoinhibitory functions in the absence of ligand. Deletion of such regions can lead to autoactivation in a ligand-independent manner. By analogy with these other systems, removal of CLD1–3 could leave a truncated form of RET that would be constitutively activated in the absence of ligand and likely would be unresponsive to the GFLs.

Ret^ΔE3 and Ret^ΔE345 Are Co-expressed in Several Tissues with Ret throughout Development

To determine where and when Ret^ΔE3 and Ret^ΔE345 transcripts are expressed and whether their expression levels change over time, we analyzed tissues in which full-length Ret is expressed at several developmental time points. Within the central nervous system, we analyzed the brain, as signaling via Ret is important for the survival of central noradrenergic neurons and dopaminergic neurons in the ventral midbrain, as well as the spinal cord, where GDNF/GFRα1/Ret signaling is critical for axon guidance into the hind limbs and survival of γ motor neurons (16–21). In the peripheral nervous system, we analyzed sensory neurons of the dorsal root ganglia (DRG), where neurturin/GFRα2/Ret signaling is important for the survival and maintenance of mechanoreceptors during development (22, 23). Additionally, shortly after birth, a subclass of nociceptors transition from being TrkA-positive to expressing GFRα1/Ret to support the development and survival of nonpeptidergic nociceptors (24–26). Sympathetic neurons of the superior cervical ganglia (SCG) were also analyzed, as it has been well established that artemin/GFRα3/Ret signaling is necessary for proper sympathetic chain migration and axon guidance during development (27–29). Last, the kidney was analyzed because GDNF/GFRα1/Ret signaling is crucial for ureteric bud branching and morphogenesis of the kidney (30–34). Using qPCR, we examined the relative expression of Ret, Ret^ΔE3, and Ret^ΔE345 in each of these tissues in mice and found that they were detected in all five tissues. The relative expression of Ret, Ret^ΔE3, and Ret^ΔE345 was lower in the brain, spinal cord, and kidney (Fig. 2A) compared with the DRG and SCG (Fig. 2B). Interestingly, in the spinal cord, DRG, and SCG, there was a significant increase of Ret^ΔE3 expression at E19.5 that preceded a significant increase of Ret at P3. The expression of all three transcripts was highest in the DRG, particularly at E19.5 and P3, when the IB4+ subpopulation of nociceptors is emerging.

FIGURE 2.

Ret^ΔE3 and Ret^ΔE345 are dynamically expressed throughout development, and RET proteins with similar molecular weights as RET^ΔE345 are detect in vivo. A, expression of Ret, Ret^ΔE3, and Ret^ΔE345 in mouse brain (E15.5, E19.5, P3), spinal cord (E15.5, E19.5, P3, adult), and kidney (E15.5, E19.5, P3, adult) by qPCR is shown as relative expression compared with actin. B, expression of Ret, Ret^ΔE3, and Ret^ΔE345 in mouse sympathetic (E19.5, P3, adult) and DRG sensory neurons (E15.5, E19.5, P3, adult) by qPCR is graphed as in A. C, cDNA from E19.5 DRGs was synthesized from poly(A) RNA, and RT-PCR was performed to determine the presence of Ret transcripts with both 5′ and 3′ alternative splicing. Ret9, Ret9^ΔE3, Ret9^ΔE345, Ret51, and Ret51^ΔE345 transcripts were identified. D, E18.5 mouse brains were lysed, and immunoprecipitations (IP) for RET9 or RET51 were performed. Immunoblotting for RET detected full-length RET9 and RET51, as expected. The presence of RET9 and RET51 proteins at the same molecular weight as RET^ΔE345 suggested the existence of RET9^ΔE345 and RET51^ΔE345 isoforms.

5′ and 3′ Alternative Splicing of Ret transcripts Are Not Mutually Exclusive

Alternative splicing of intron 19 in Ret allows for the translation of two isoforms of Ret, RET9 and RET51. We sought to determine whether alternative splicing could occur simultaneously, both 5′ and 3′ in Ret transcripts, allowing for increased diversity of encoded RET proteins. cDNA was isolated from E19.5 mouse DRGs, as we found this tissue to have the highest relative expression of Ret, Ret^ΔE3, and Ret^ΔE345 (Fig. 2B). Forward primers specific for each 5′ splicing event were paired with reverse primers that would detect either Ret9 or Ret51 alternative splicing. Full-length Ret9 and Ret51 transcripts were detected, as expected, at 3.33 kb and 4.38 kb, respectively. We observed Ret^ΔE3 and Ret^ΔE345 with intron 19 retention, thus encoding Ret^ΔE3 and Ret^ΔE345 transcripts with amplicons of 3.05 kb and 2.61 kb, respectively. However, we were only able to detect Ret^ΔE345 with a properly excised intron 19 at an amplicon size of 3.66 kb, encoding a Ret51^ΔE345 transcript. Although we were unable to detect a Ret51^ΔE3 transcript (4.10 kb), this is likely due to the generally low expression levels of Ret^ΔE3. Taken together, these data indicate that the Ret locus encodes for at least three previously unidentified transcripts and at least five total Ret isoforms: Ret9, Ret9^ΔE3, Ret9^ΔE345, Ret51, and Ret51^ΔE345.

RET Proteins of the Same Molecular Weight as RET^ΔE345 Are Detected in Vivo

Although we can detect Ret^ΔE3 and Ret^ΔE345 mRNA transcripts, which are likely being processed for translation, we sought to detect the presence of these isoforms at the protein level. Plasmids encoding human RET51, RET51^ΔE3, and RET51^ΔE345 were transfected into NIH/3T3 cells, and lysates were run as size standards to aid in detecting Ret isoforms of the appropriate molecular weights from E18.5 mouse brain. The molecular weight of native RET51 protein is 120 kDa, but because of posttranslational modifications, specifically glycosylation of the extracellular domain, RET51 has a molecular weight of ∼180 kDa. The native molecular weights of RET^ΔE3 and RET^ΔE345 were predicted to be 109 kDa and 93 kDa, respectively. In transfected cells, we consistently observe the mature, processed proteins to have molecular weights of ∼150 kDa (RET^ΔE3) and ∼125 kDa (RET^ΔE345) (Fig. 2D).

Whole brains from Ret^+/+ or Ret^−/− E18.5 littermate mice were lysed and divided in two, and immunoprecipitations were performed to isolate either RET9 or RET51 proteins. Although the qPCR data suggested that Ret transcripts have an overall lower relative expression in the brain compared with PNS tissues, the brain was chosen for analysis because of the abundant amount of protein that could be isolated. Ret^−/− mice were used as a control to demonstrate the specificity of RET immunoblotting and to show the potential presence of RET proteins at the same molecular weight as RET^ΔE3 and RET^ΔE345. Because we previously observed more highly expressed Ret9^ΔE345 and Ret51^ΔE345 transcripts compared with those of Ret^ΔE3, we hypothesized that we would be most able to detect both RET9^ΔE345 and RET51^ΔE345 proteins. By immunoblotting for RET, we were able to detect full-length RET9 and RET51 proteins, as expected (Fig. 2D). Additionally, we observed RET9 and RET51 bands with similar molecular weights to RET^ΔE345, suggesting the presence of RET9^ΔE345 and RET51^ΔE345 proteins in mouse brain (Fig. 2D). Because the antibodies used for immunoprecipitations of RET9 and RET51 are to the C terminus, and the RET antibody for immunoblotting is to the common N-terminal region, this RET protein is unlikely to be a degradation product. However, we cannot exclude the possibility that this is an immature, non-glycosylated, full-length RET protein.

RET^ΔE3 and RET^ΔE345 Are Trafficked to the Cell Surface

To examine the biochemical properties of RET^ΔE3 and RET^ΔE345, we transitioned to an in vitro system for these experiments. Although RET^ΔE3 and RET^ΔE345 encode proteins that may be capable of signaling, this does not guarantee that they will be correctly trafficked to the cell surface for activation. To determine whether RET^ΔE3 and RET^ΔE345 are trafficked to the plasma membrane, NIH/3T3 cells were transfected with C-terminally epitope-tagged Ret constructs that also contain an IRES-GFP to serve as an indicator for transfected cells. Taking advantage of a RET antibody specific for the extracellular domain and using an HA antibody to label the C terminus of RET, we were able to selectively visualize RET located on the plasma membrane. Full-length RET was detected on the cell surface of transfected NIH/3T3 cells, as expected (Fig. 3K). Permeabilization revealed the C-terminal HA tag, confirming the integrity of the plasma membrane (Fig. 3P). Similar to full-length RET, both RET^ΔE3 and RET^ΔE345 were trafficked to the cell surface, as determined with the extracellular RET antibody in non-permeabilized cells (Fig. 3, S and A'). Transfection of GFP alone showed no labeling with the RET and HA antibodies, regardless of permeabilization, confirming the specificity of our immunostaining (Fig. 3, A–H). A cell surface biotinylation experiment was performed as a secondary measure to confirm these results, and we again observed trafficking of full-length RET, RET^ΔE3, and RET^ΔE345 to the cell surface (data not shown). Thus, RET^ΔE3 and RET^ΔE345 are trafficked properly to the cell surface, where they can participate in GFL-mediated activation.

FIGURE 3.

RET^ΔE3 and RET^ΔE345 are trafficked to the cell surface. NIH/3T3 cells were transfected with GFP (A–H), Ret51::HA (I–P), Ret51^ΔE3::HA (Q–X), or Ret51^ΔE345::HA (Y–F'). Constructs for the RET51 isoforms were bicistronic, allowing for expression of GFP as an indicator for positive transfection, which was encoded by an IRES-GFP following RET cDNA sequences. Cells were washed in 1× PBS and fixed briefly in 4% paraformaldehyde 24 h post-transfection. Cells were then blocked in immunofluorescence blocking solution with or without Triton X-100 to allow for cell membrane permeabilization. In the absence of Triton X-100, antibodies specific for the RET extracellular domain were able to bind to RET51::HA (K), RET51^ΔE3::HA (S), and RET51^ΔE345::HA (A'). However, using an antibody against the HA epitope that would detect the C terminus located intracellularly for RET51::HA (L), RET51^ΔE3::HA (T), and RET51^ΔE345::HA (B') showed no binding/fluorescence in the absence of Triton X-100. When immunofluorescence blocking solution with Triton X-100 was used, fluorescence was detected using the anti-HA antibody for RET51::HA (P), RET51^ΔE3::HA (X), and RET51^ΔE345::HA (F'). These data show that RET51^ΔE3 and RET51^ΔE345 are trafficked to the cell surface similarly as full-length RET51. Scale bars represent 20 μm.

RET^ΔE3 and RET^ΔE345 Bind to GFRα1, GFRα2, GFRα3, and GFRα4

Previous structural analysis (Fig. 1C) suggested that interactions between RET^ΔE3 and the GFRαs may be unstable and that RET^ΔE345 would be unlikely to bind to the co-receptors because of deletions of important binding residues in CLD1 and CLD3. To confirm these results, we co-transfected HEK293T cells with RET^ΔE3 or RET^ΔE345 and HA epitope-tagged GFRα constructs (HA::GFRα1 (mouse), HA::GFRα2 (rat), HA::GFRα3 (mouse), and HA::GFRα4 (chicken)). Immunoprecipitations were performed using a stringent immunoprecipitation buffer (modified radioimmune precipitation assay buffer) to eliminate any weak, nonspecific interactions. Surprisingly, we found that both RET^ΔE3 and RET^ΔE345 bind to each of the four GFRαs (Fig. 4). Co-transfection of the GFRαs with TrkA, another receptor tyrosine kinase, was performed as a negative control to show selectivity of binding between the GFRαs and RET (Fig. 4). As expected, TrkA did not associate with any of the GFRαs. There were no apparent differences between the four different GFRαs in regard to the extent of their association with RET^ΔE3 and RET^ΔE345, suggesting that these splice variants could potentially be activated by all four GFLs (Fig. 4).

FIGURE 4.

RET^ΔE3 and RET^ΔE345 bind to GFRα1, GFRα2, GFRα3, and GFRα4. RET51, RET51^ΔE3, and RET51^ΔE345 were co-transfected with HA epitope-tagged GFRα1 (A), GFRα2 (B), GFRα3 (C), or GFRα4 (D). Immunoprecipitations (IP) were performed using an HA antibody to select for GFRαs. Immunoblotting (IB) for RET showed interaction of RET51^ΔE3 and RET51^ΔE345 with all four GFRα proteins. Co-transfection of HA::GFRα1, HA::GFRα2, HA::GFRα3, or HA::GFRα4 with TrkA was performed as a negative control, and, as expected, TrkA did not associate with any of the GFRαs.

RET^ΔE3 and RET^ΔE345 Heterodimerize with Full-length RET

In addition to being potential signaling molecules on their own, we observed that these transcripts are normally expressed in tissues where full-length RET is also expressed (Fig. 2). Although these isoforms may be present in different cells within these tissues, these observations raise the question of whether RET^ΔE3 and RET^ΔE345 could heterodimerize with full-length RET and affect its function. To test this possibility, HEK293T cells were co-transfected with HA epitope-tagged RET51 and untagged RET^ΔE3 or RET^ΔE345. Immunoprecipitations were performed using an HA antibody to select for full-length RET51 and determine whether the different isoforms could interact with one another (Fig. 5A). In an additional experiment, FLAG-tagged RET51^ΔE345 (FLAG:: RET51^ΔE345) was co-transfected with RET51 and RET51^ΔE3, and immunoprecipitations were performed using a FLAG antibody to select for RET51^ΔE345 (Fig. 5B). We found that both RET51^ΔE3 and RET51^ΔE345 associated with full-length RET51 and that RET51^ΔE3 and RET51^ΔE345 could also bind to one another (Fig. 5). This is consistent with former experiments suggesting that overexpression of RET proteins allows for ligand-independent dimerization via the transmembrane domain (35).

FIGURE 5.

RET^ΔE3 and RET^ΔE345 can heterodimerize with full-length RET. A, HA epitope tagged RET51 (RET51::HA) was co-transfected with untagged RET51^ΔE3 or RET51^ΔE345 in HEK293T cells. Immunoprecipitations (IP) were performed with an HA antibody to select for RET51::HA. Immunoblotting (IB) for RET51 showed co-immunoprecipitation of RET51^ΔE3 and RET51^ΔE345 with full-length RET51. B, a similar experiment was performed using FLAG epitope tagged RET51^ΔE345 (FLAG:: RET51^ΔE345). Immunoprecipitations with an anti-FLAG antibody selected for FLAG:: RET51^ΔE345. Immunoblotting for RET51 demonstrated the association of RET51 and RET51^ΔE3 with FLAG:: RET51^ΔE345. Co-transfections with TrkA were performed for both experiments as a negative control, and immunoblotting for TrkA showed no binding with either RET51::HA or FLAG::RET51^ΔE345, as expected.

RET^ΔE345 Has Increased Ligand-independent Activation Compared with RET and RET^ΔE3

To determine levels of activation of RET^ΔE3 and RET^ΔE345, the basal phosphorylation of tyrosine residues of the RET isoforms were compared with full-length RET phosphorylation. Overexpression of the RET isoforms in vitro, either in the presence or absence of a GFRα, allows for ligand-independent dimerization and activation of RET. The level of activation can be assessed by determining the summated level of all of the phosphorylated tyrosines using a pan-phosphotyrosine antibody, or individual tyrosine residues can be analyzed using residue-specific RET phosphotyrosine antibodies. We evaluated the level of total phosphotyrosine (Tyr(P)) between the RET isoforms and individually evaluated phosphorylation at two tyrosine residues of RET, Tyr⁹⁰⁵ and Tyr¹⁰⁶². Tyrosine 905 in the RET kinase domain is an autocatalytic tyrosine that is conserved in many receptor tyrosine kinases and is a binding site for GRB7 and GRB10 (36–38). Additionally, mutation of tyrosine 905 to phenylalanine (Y905F) impairs the kinase activity of RET (39). Tyrosine 1062 is a binding site for SHC, Dok4/5, IRS-1, and FRS-2 when phosphorylated and is a binding site for Enigma in a phosphorylation-independent manner (36, 40–47). Interaction between Tyr¹⁰⁶² and these adaptor proteins leads to activation of the Ras/ERK and PI3K/AKT pathways (41, 42, 45, 46, 48).

To evaluate RET phosphorylation, NIH/3T3 cells were transfected with RET51, RET51^ΔE3, or RET51^ΔE345 constructs in the presence or absence of GFRα1, and the total levels of phosphorylated RET or the level of phosphorylation at Tyr⁹⁰⁵ and Tyr¹⁰⁶² were assessed. To analyze total Tyr(P) levels of RET, immunoprecipitations of RET51 were performed, followed by immunoblotting for Tyr(P) (Fig. 6A). An increase in the level of RET^ΔE345 phosphorylation compared with full-length RET was observed, but this was not statistically significant (p = 0.0837). We did detect, however, a significant increase in the Tyr(P) level of RET^ΔE345 co-expressed with GFRα1 compared with full-length RET co-expressed with GFRα1 (p = 0.0366). The phosphorylation of RET^ΔE3 co-expressed with GFRα1 or expressed alone was not significantly different from full-length Ret (p = 0.6231).

FIGURE 6.

RET^ΔE345 displays increased ligand-independent activation compared with full-length RET. A, RET51 was co-transfected with HA epitope-tagged GFRα1 in NIH/3T3 cells. Immunoprecipitations (IP) were performed with a RET51 antibody to select for the RET isoforms. Immunoblotting (IB) for phospho-tyrosine and RET was performed, and integrated density values were calculated for each. B, preimmunoprecipitation lysates from the previous experiment were subjected to SDS-PAGE, and Western blotting was performed to determine the levels of residue-specific tyrosine phosphorylation of RET. Immunoblotting was performed for phospho-RET (Tyr⁹⁰⁵), phospho-RET (Tyr¹⁰⁶²), and RET, and integrated density values were calculated for each. Ratios of phospho-RET/RET (C), phospho-RET (Tyr⁹⁰⁵)/RET (D), and phospho-RET (Tyr¹⁰⁶²)/RET (E) are reported, with values normalized to RET51. F, expression plasmids encoding oncogenic RET proteins, RET51^MEN2A and RET51^MEN2B, were transfected into NIH/3T3 cells, and integrated density values were calculated and analyzed for phospho-RET/RET for all RET isoforms. *, p < 0.05; **, p < 0.01; n.s., not significant.

To determine levels of Tyr(P)⁹⁰⁵ and Tyr(P)¹⁰⁶², pre-immunoprecipitation lysates were collected from samples in the previous experiments, and immunoblotting was performed using antibodies specific for these two phosphorylated tyrosine residues (Fig. 6B). Levels of Tyr(P)⁹⁰⁵ and Tyr(P)¹⁰⁶² were significantly elevated for RET^ΔE345 compared with RET (p = 0.0439 and p = 0.0411, respectively) and also for RET^ΔE345 co-expressed with GFRα1 compared with full-length RET co-expressed with GFRα1 (p = 0.0106 and p = 0.0096, respectively). However, levels of Tyr(P)⁹⁰⁵ and Tyr(P)¹⁰⁶² were unchanged for RET^ΔE3 compared with RET and also for RET^ΔE3 co-expressed with GFRα1 compared with full-length RET co-expressed with GFRα1. Taken together, these data suggest that RET^ΔE3 has a similar activation level as full-length RET, whereas RET^ΔE345 has an elevated activation level compared with RET in a ligand-independent manner. Additionally, the lower total phosphotyrosine level for RET^ΔE345 compared with Tyr(P)⁹⁰⁵ and Tyr(P)¹⁰⁶² suggests that one or more of the other tyrosine residues in RET is not phosphorylated as highly as Tyr⁹⁰⁵ and Tyr¹⁰⁶² compared with full-length RET.

Because we consistently observed an increased level of tyrosine phosphorylation of the RET^ΔE345 isoform compared with full-length RET, we wondered whether this level of receptor activation was similar to oncogenic RET mutations, particularly those found in MEN2. Human RET point mutations known to be causal for either MEN2A or MEN2B were introduced into the RET51 expression plasmid, specifically C634R (RET51^MEN2A) and M918T (RET51^MEN2B). NIH/3T3 cells were transfected with RET51, RET51^ΔE3, RET51^ΔE345, RET51^MEN2A, or RET51^MEN2B constructs. Immunoprecipitations for RET51 were performed to isolate RET proteins, followed by immunoblotting for Tyr(P) as described previously. We observed a significant increase in the level of phosphorylation of both RET^MEN2A and RET^MEN2B oncogenic proteins compared with full-length RET (p = 0.0282 and p = 0.0097, respectively) (Fig. 6F). The levels of phosphorylation of RET^MEN2A and RET^MEN2B compared with RET^ΔE345 showed no significant difference (p = 0.4898 and p = 0.1712, respectively). These data indicate that the RET^ΔE345 isoform is phosphorylated at similar levels as oncogenic RET mutations.

RET^ΔE3 Is Activated in a GDNF-dependent Manner but RET^ΔE345 Is Not

Although differences are observed in the level of ligand-independent activation for RET^ΔE3 and RET^ΔE345, we wondered whether these two receptors could be activated in a GDNF-dependent manner. To evaluate this, Neuro2A cells were co-transfected with HA epitope-tagged RET51, RET51^ΔE3, or RET51^ΔE345 and with untagged GFRα1. Twenty-four hours post-transfection, the culture medium was removed, and cells were stimulated with 200 ng/ml recombinant mouse GDNF for 30 min. Cells were detergent-extracted, and immunoprecipitations for HA were performed to isolate the exogenous RET isoforms. Immunoblotting for GFRα1 was performed to determine the level of GDNF-dependent association of GFRα1 to RET (Fig. 7A). For our positive control, where GFRα1 and RET51 were co-expressed, we observed a significant GDNF-dependent association of these proteins after 30 min of stimulation (p = 0.0373) (Fig. 7B). Under conditions where GFRα1 and RET51^ΔE3 or RET51^ΔE345 were co-expressed, we observed a trend of GDNF-dependent association of GFRα1 to both RET isoforms, but this ligand-dependent interaction was not statistically significant (p = 0.1750 and p = 0.3269, respectively).

FIGURE 7.

RET^ΔE3 is activated in a GDNF-dependent manner but RET^ΔE345 is not. A, HA epitope-tagged RET constructs were co-transfected with GFRα1 into Neuro2A cells. Twenty-four hours post-transfection, the medium was removed, and fresh medium was added to cells containing 200 ng/ml GDNF. Cells were stimulated for 30 min, and immunoprecipitations (IP) were performed from lysates with an HA antibody to selectively pull down the RET isoforms. Immunoblotting (IB) for GFRα1 and RET was performed, and integrated density values were calculated for each. B, quantification the of GFRα1/RET ratios. C, immunoblotting for poly-ubiquitin was performed, and integrated density values were calculated for poly-ubiquitinated RET and total RET. D, quantification of poly-ubiquitinated RET/RET ratios. n.s., not significant. AU is arbitrary units. No Tx indicates cells that were not stimulated with GDNF. Asterisks indicate p < 0.05.

To determine whether the RET isoforms were activated in a GDNF-dependent manner, immunoblotting for poly-ubiquitin was performed. We have shown previously that, upon ligand activation and autophosphorylation, RET is rapidly poly-ubiquitinated and targeted for degradation (49). Because ectopic expression of RET results in high levels of autophosphorylation in the absence of ligand, we were not able to use phosphotyrosine immunoblotting as a measure of RET activation upon GDNF stimulation. We observed a significant increase in poly-ubiquitinated RET51 (p = 0.0370) following GDNF stimulation, as expected, and also observed a significant increase in poly-ubiquitinated RET51^ΔE3 (p = 0.0493), suggesting that this isoform is activated in a GDNF-dependent manner. Interestingly, we did not observe a difference in the level of poly-ubiquitinated RET51^ΔE345 following GDNF stimulation (p = 0.0696). These data suggest that, like full-length RET, RET^ΔE3 is activated in a GDNF-dependent manner but RET^ΔE345 is not.

Discussion

Alternative splicing of precursor mRNA is one of many processes that mediates gene regulation in metazoans, and the frequency of alternative splicing increases with species complexity (50, 51). For example, of the ∼25,000 genes encoded by the human genome, ∼95% produce transcripts that are alternatively spliced (50, 51). By expanding the proteome through the synthesis of various protein isoforms, alternative splicing allows for increased protein diversity, with isoforms performing different biological functions.

Here we provide evidence for two alternative splicing events in Ret that, in combination with Ret9 or Ret51 alternative splicing, give rise to at least five Ret transcripts (Fig. 2C). The splicing events for RET^ΔE3 and RET^ΔE345 have only been described previously in human tissues (12). Because an additional Ret transcript, RET43, is only expressed in humans, we examined the expression of RET^ΔE3 and RET^ΔE345 in additional vertebrate tissues to determine whether these might also be human-specific transcripts (8, 52). To this end, we discovered that RET^ΔE3 and RET^ΔE345 are not human-specific RET transcripts but are also expressed in normal developing tissues of zebrafish, mice, and rats (Fig. 1C). Functionally important transcripts are transcriptionally conserved between species, arguing that RET^ΔE3 and RET^ΔE345 have physiologically important functions.

Analysis of an experimentally validated structural model for the RET cadherin-like domains 1–4 was made to evaluate the consequence of exon deletions of RET alternative splicing on the ligand binding and activation characteristics of RET^ΔE3 and RET^ΔE345 (Fig. 1B). Previous experiments indicate that the GDNF-GFRα1 complex binds directly to CLD4 and the cysteine-rich domain (53). This suggested that both RET^ΔE3 and RET^ΔE345 should be able to bind to the GDNF-GFRα1 complex, and indeed this interaction was observed not only for GFRα1 but also for GFRα2, GFRα3, and GFRα4 (Fig. 4). It should be noted that, in our experiments, the chicken GFRα4 construct contained all three cysteine-rich domains, unlike mammalian GFRα4, which lacks domain 1. Although it was originally thought that CLD1–3 of RET was involved in direct binding from results obtained through mutagenesis studies, a direct interaction between CLD1–3 and the GDNF-GFRα1 complex has not been observed (53, 54). Instead, it has been proposed that CLD1–3 contributes to the stability of the tertiary structure of the GDNF-GFRα1-RET complex by forming a secondary dimerization site to trap the GDNF-GFRα1 binary complex (14). These studies predict that RET^ΔE3, which forms a fused CLD1-CLD2 region, likely gives rise to a less stable tertiary GDNF-GFRα1-RET complex compared with full-length RET. This may result in a receptor variant with differential ligand binding specificities. FGF receptors are a subfamily of four receptor tyrosine kinases that are receptors for FGFs, of which there are 18 ligands (55). Alternative splicing of the FGF receptors in the region coding for the extracellular ligand-binding domain of the receptors allows for differential ligand binding. It is possible that the alternative splicing responsible for RET^ΔE3 similarly modifies GFL-GFRα affinity. Alternatively, the truncated extracellular domain of RET^ΔE3 may cause alterations in the conformational changes that occur upon GFL-GFRα binding that induces receptor autophosphorylation. This would cause different autophosphorylation kinetics in RET^ΔE3 compared with full-length RET and, therefore, differential activation of downstream second messenger cascades. Biochemical experiments using an in vitro system that allows for the monitoring of autophosphorylation upon direct ligand activation of RET^ΔE3 or full-length RET in isolation will be necessary to distinguish between these possibilities.

The large deletion in the extracellular domain of RET^ΔE345, including the Ca²⁺ binding domain, which is important for RET folding and ligand-dependent signaling, is predicted to have dramatic effects on RET^ΔE345 activation. Indeed, it is likely that RET^ΔE345 functions as a constitutively active form of RET, consistent with the mechanistic data shown in Fig. 6. Autophosphorylation of RET^ΔE345 is increased compared with full-length RET in the presence and absence of GFRα1 in vitro (Fig. 6B). Constitutively active receptor tyrosine kinases have not been found in normal tissues and usually arise from germline and somatic mutations, where they are most commonly found to cause neoplasia. Activating mutations in RET have been shown to be responsible for multiple endocrine neoplasia, type 2, including neoplasias such as medullary thyroid carcinoma and pheochromocytomas (56). Unlike other constitutively active forms of RET, RET^ΔE345 may be posttranscriptionally or translationally regulated in vivo in a manner that is not recapitulated in the in vitro system used here. In this way, the signaling of RET^ΔE345 may be under tight regulation for this isoform to perform specific functions in vivo.

To interrogate the functions of RET^ΔE3 and RET^ΔE345 in primary neurons and tissues, we attempted to generate isoform-specific antibodies. Likewise, we also tried to create siRNAs to selectively knock down Ret^ΔE3 and Ret^ΔE345 transcripts. Neither of these experimental approaches were successful at targeting RET^ΔE3 and RET^ΔE345 without also affecting full-length RET. Therefore, to determine the physiologic functions of RET^ΔE3 and RET^ΔE345, a transgenic approach in a vertebrate will be required to selectively remove these individual transcripts. This could be accomplished by deleting specific introns and fusing the flanking exons in-frame to inhibit exon skipping (e.g. remove intron 3 and fuse exons 2 and 3 together to inhibit alternative splicing for Ret^ΔE3). How this approach would affect RET expression and other splicing events, however, is unknown. In addition, because of the invariably coincident expression of Ret^ΔE3 and Ret^ΔE345 in the same tissues as Ret, it is possible that full-length RET may compensate for the functions of RET^ΔE3 and RET^ΔE345. As an alternative approach, production of knockin animals in which only Ret^ΔE3 or Ret^ΔE3 is expressed would determine whether these isoforms are capable of supporting the normal development of the nervous system and kidneys.

Last, to gain a mechanistic understanding of the factors responsible for the alternative splicing of Ret, it would be interesting to explore RNA motifs encoding exon splicing enhancers within the locus that impact the formation of Ret^ΔE3 and Ret^ΔE345. Understanding the spliceosome machinery involved in this process may help us understand why the expression of these transcripts are elevated in tumors, such as in pheochromocytomas (12). Overall, it may be necessary to return to a disease model, such as neuroendocrine gland tumors, where these RET transcripts were originally identified and use the pathobiology to understand the signaling capacities of RET^ΔE3 and RET^ΔE345 to identify their functions in normal development.

Experimental Procedures

Molecular Modeling

This analysis was performed as described previously (14). Models of cadherin-like domains 1–4 were taken from PDB code 4UX8, defined by a combination of EM and small angle x-ray scattering analyses. The figure was made using PyMOL software (Schrödinger).

RNA Isolation, RT-PCR, and qPCR

Total RNA was isolated from zebrafish, mouse, or rat tissues using TRIzol reagent (Thermo Fisher Scientific) according to the instructions of the manufacturer. cDNA synthesis was performed from 1 μg of total RNA to generate poly(A) first-strand cDNA using the SuperScript III First-Strand Synthesis System for RT-PCR kit (Life Technologies). For RT-PCR reactions with amplicons of less than 200 bp, GoTaq Green Master Mix (Promega) and species-specific oligonucleotide primers for Ret, Ret^Δ3, and Ret^Δ345 were used. RT-PCR reactions with amplicons greater than 200 bp were performed using the Phusion high-fidelity DNA polymerase kit (New England Biolabs) following the instructions of the manufacturer. qPCR was performed using FastStart Universal SYBR Green Master Mix (Roche) and run on a QuantStudio 6 Flex real-time PCR system (Thermo Fisher Scientific). Quantitative analyses were performed by calculating the ΔCt (Ct of transcript divided by Ct of the housekeeping gene actin) for each gene analyzed and transforming that value to log₂. These values are reported as the relative expression.

Primer Sequences

The primers used for this study were as follows: ZF.RetFL.F, 5′-TCG TAG TTT ACG CAG CGG CTC A-3′; ZF.RetFL.R, 5′-TCG CGA TTT TCA GTG ATG TG-3′; ZF.Ret3.F, 5′-CAG TAG TTT ACG ATC TTC TGT ACC G-3′; ZF.Ret3.R, 5′-TTC AGT GTC AGT CCC GTT GA-3′; ZF.Ret345.F, 5′-CAG TAG TTT ACA GCT GAA ACT CAG TC-3′; ZF.Ret345.R, 5′-TGA CAT TGG AGA AGC GAA TG-3′; Mouse.RetFL.F, 5′-AGC ATC CGC AAT GGT GGT TT-3′; Mouse.RetFL.R, 5′-TGT TCT CCC TGA CTC GGA AG-3′; Mouse.Ret3.F, 5′-AGC ATC CGC AGG GAT AGT CT-3′; Mouse.Ret3.R, 5′-ACA CTG TCA CTG GGA AGG AC-3′; Mouse.Ret345.F, 5′-AGC ATC CGC AAG CTG ATT CT-3′; Mouse.Ret345.R, 5′-TTC ACT GGG AAG GAG TAG GC-3′; Rat.RetFL.F, 5′-AGC ATC CGC AAT GGC GGC TT-3′; Rat.RetFL.R, 5′-TAG CAT GCG GAA CTG GTA GA-3′; Rat.Ret3.F, 5′-AGC ATC CGC AGG GAC GGT CT-3′; Rat.Ret3.R, 5′-CCG CTT AAA CTC CAC CAC AG-3′; Rat.Ret345.F, 5′-AGC ATC CGC AAG CTG GTT CT-3′; Rat.Ret345.R, 5′-TAG GCC ATG GGT AGG TTC AG-3′; Mouse.Ret9.R, 5′-ATT TAC TGT CCA TTG CAA GGC-3′; and Mouse.Ret51.R, 5′-CCT ATC AGT GCT TTA AGT CTG-3′.

Mice

Wild-type C57BL/6J mice were purchased from The Jackson Laboratories (Bar Harbor, ME). Ret^−/− mice have been described previously and were maintained in a mixed genetic background (57). For timed matings, noon of the day on which a vaginal plug was detected was considered E0.5. All housing and procedures performed on mice were approved by the University of Michigan Animal Care and Use Committee.

Detergent Extraction and Preclearing of Whole Tissues

Tissues were harvested, placed in a 2-ml tube with 250 μl of immunoprecipitation buffer lacking Nonidet P-40 (TBS (pH 7.4), 10% glycerol, 500 μm sodium vanadate, and protease inhibitors) along with a steel grinding ball (5 mm, 69989, Qiagen, Valencia, CA) and mechanically homogenized using a TissueLyzer II (Qiagen) for 2 min at a frequency of 20 Hz. Following this, 250 μl of 2% Nonidet P-40-containing immunoprecipitation buffer was added to homogenates and incubated for 1 h at 4 °C under gentle agitation. Homogenates were then centrifuged for 10 min at 16,000 × g and subjected to an initial preclearing step with protein A and protein G alone at 4 °C for 2 h under gentle agitation. A second preclearing step was performed with protein A, protein G, and a species-matched nonspecific control IgG for 2 h under gentle agitation.

Immunoprecipitations and Immunoblotting

Plates were placed on ice, gently washed twice with 1× PBS (pH 7.4), and lysed with modified radioimmune precipitation assay buffer (TBS (pH 7.4), 10% glycerol, 1% Triton X-100, 0.1% SDS, 500 μm sodium vanadate, and protease inhibitors). Proteins were immunoprecipitated using anti-RET9 (goat, Santa Cruz Biotechnology), anti-RET51 (goat, Santa Cruz Biotechnology), anti-HA (mouse, Millipore), or anti-FLAG (rabbit, Cell Signaling Technology) selective antibodies. Immunoprecipitates were subjected to SDS-PAGE, followed by electroblotting onto PVDF membranes (Immobilon P, Millipore). Immunoblotting was performed on blots using antibodies selective for RET (1:500, goat, R&D Systems), HA (1:5000, rabbit, Cell Signaling Technology), Trk (1:1000, rabbit, Santa Cruz Biotechnology), phospho-tyrosine (1:1000, mouse, Millipore), phospho-RET (Tyr⁹⁰⁵) (1:500, rabbit), phospho-RET (Tyr¹⁰⁶², 1:500, rabbit), GFRα1 (1:500, goat, R&D Systems), or poly-ubiquitin (1:500, mouse, Enzo Life Sciences). Phospho-RET (Tyr⁹⁰⁵) and phospho-RET (Tyr¹⁰⁶²) have been described previously (58). Lysates collected prior to immunoprecipitations served as loading controls to assess protein expression levels and were also subjected to immunoblotting for actin (1:1000, goat, Santa Cruz Biotechnology). Each biochemical experiment was performed three to four times with similar results.

Mammalian Cell Culture and Transfections

NIH/3T3 (ATCC), HEK293T (ATCC), and Neuro2A cells (ATCC) were maintained at 37 °C with 5% CO₂ in DMEM supplemented with FBS (10% by volume) and a penicillin/streptomycin/glutamine mixture. Cells were plated into 6-well tissue culture plates and allowed to proliferate until cells were 70% confluent. Transfections of cells were performed using Lipofectamine 2000 according to the instructions of the manufacturer (Invitrogen). For experiments testing the interaction of proteins, a total of 4 μg of plasmid DNA was added per well using a GFP plasmid as a control for transfecting equal amounts of DNA. For in vitro ligand-independent activation experiments, 2 μg of HA::GFRα1 and 1 μg of one of the RET constructs were transfected for a total of 3 μg of DNA per condition, with a GFP plasmid used as a control for transfecting equal amounts of DNA. RET51^ΔE3 and RET51^ΔE345 constructs were subcloned from RET51 constructs into pcDNA3.1 (Thermo Fisher Scientific). A Ret51 construct encoding mouse RET51 fused at the C terminus with an HA epitope tag (RET51::HA) was acquired from Dr. Ben Allen. This plasmid was subcloned to create similarly tagged Ret51^ΔE3 and Ret51^ΔE345 constructs, which were subcloned into pLentilox-IRES-GFP (University of Michigan Vector Core). For in vitro ligand-dependent activation experiments, 3 μg of HA epitope-tagged Ret51, Ret51^ΔE3, or Ret51^ΔE345 was co-transfected with 3 μg of untagged GFRα1. Twenty-four hours post-transfection, the medium was aspirated, and cells were treated with 200 ng/ml recombinant mouse GDNF (Peprotech) in fresh cell culture medium for 30 min. The HA::GFRα1 (mouse) plasmid was generously donated by Dr. Ben Allen. The HA::GFRα2 (rat), HA::GFRα3 (mouse), and HA::GFRα4 (chicken) plasmids were generously donated by Dr. Carlos Ibáñez. The TrkA (rat) plasmid was generously donated by Dr. Christin Carter-Su. Constructs encoding human RET51 harboring MEN2A (C634R) and MEN2B (M918T) point mutations in RET were created using the QuikChange II XL site-directed mutagenesis kit (Agilent).

Quantification of Immunoblots

Scanned images of x-ray films were imported into ImageJ (National Institutes of Health) and processed using the gel analysis tool. Integrated density values obtained from immunoblotting were reported as mean ± S.E. with arbitrary units on the vertical axis. Values were normalized to the level of RET51 phosphorylation in each experiment. Statistical analyses were performed using Student's t test for each experiment, and all biochemical experiments were performed at least three times with similar results.

Immunocytochemistry

Transfected cells that were plated on glass coverslips were fixed for 8 min in 4% paraformaldehyde. Cells were washed twice in 1× PBS and briefly blocked in immunofluorescence blocking solution (3% BSA and 1% normal donkey serum in 1× PBS (pH 7.4)) with or without 0.1% Triton X-100. Primary antibodies were diluted in blocking buffer (RET (1:50, goat, R&D Systems), HA (1:1000, mouse, Millipore)), and slides were incubated at 4 °C overnight. Cells were washed twice with 1× PBS and then incubated at room temperature for 1 h with secondary antibodies (donkey anti-mouse Alexa Fluor 633 (Thermo Fisher Scientific), donkey anti-goat 543 (Biotium)) diluted in blocking solution. Last, cells were washed three times in 1× PBS and mounted with DAPI Fluoromount-G (Southern Biotech). Cells were imaged using a Zeiss Axiovert 200 M epifluorescence microscope with a ×40 objective.

Author Contributions

N. A. G. conducted the experiments, analyzed the results, and wrote the paper. J. K. V. and S. S. N. performed experiments. N. Q. M. performed the structure-based predictions of the impact of exon deletions. B. A. P. conceived the idea for the project, contributed intellectual advice, and edited the manuscript.