Drosophila CORL is required for Smad2-mediated activation of Ecdysone Receptor expression in the mushroom body

During embryonic development in animals, secreted Transforming Growth Factor β (TGFβ) proteins perform a multitude of tasks. Later in life, mutations that disrupt TGFβ signaling often lead to tumor growth. One group of TGFβ pathway regulators is oncogenic Sno/Ski proteins. These bind Smad signal transducers downstream of TGFβ/Activin subfamily members. The initial model for Sno/Ski function was based on data from gain-of-function studies in mammalian cells and states that they are obligate antagonists of TGFβ signaling (Jahchan and Luo, 2010). This model has evolved to accommodate loss-of-function data from flies and nematodes, and RNAi studies in mammalian cells. In Drosophila, data from Sno (Snoo – FlyBase) mutants suggests that Sno acts as a pathway switch: Sno facilitates TGFβ/Activin signaling via a molecular mechanism that simultaneously antagonizes Dpp/BMP signaling (Takaesu et al., 2006). The fly data are supported by evidence that DAF-5 facilitates Smad signaling in nematodes (da Graca et al., 2003) and that Sno facilitates activin signaling in two mammalian cell types (Sarker et al., 2005; Sarker et al., 2008). Consistent with this dual role in TGFβ signaling (negative when overexpressed but positive at physiological levels), human SNO (SKIL – Human Gene Nomenclature Database) can function as both an oncogene and a tumor suppressor gene (Jahchan and Luo, 2010).

Previous analysis of known Sno/Ski-related proteins clustered them into three subfamilies (Sno/Ski, Dachsund and CORL) (Takaesu et al., 2006) the defining feature of which is a Sno homology domain that interacts with Smads. The CORL subfamily contained two members: the Drosophila predicted protein CG11093 and mouse Corl1. At that time, only mouse Corl1 had been cloned and it displayed neural-specific expression during development (in a region of the hindbrain that will become Purkinje cells of the cerebellum). In cell culture, mouse Corl1 functioned as a co-factor for the LBX1 homeodomain protein (Mizuhara et al., 2005). Mouse Corl2 was subsequently cloned and found to have neural-specific expression similar to mouse Corl1 – starting at embryonic day 10.5 in the cell bodies of Purkinje cell progenitors and in the Purkinje cells of adults (Minaki et al., 2008; Miyata et al., 2010). Interestingly, the cerebellum is the site of motor coordination and cerebellar defects are associated with movement disorders known as ataxias (Orr, 2010).

Human CORL proteins FUSSEL15 (SKOR1 in human, Corl1 in mouse) and FUSSEL18 (SKOR2 in human, Corl2 in mouse) display conserved expression in adult Purkinje cells. Both bind Smads non-specifically in a melanoma cell assay and luciferase assays suggest human SKOR2 inhibits while human SKOR1 has no effect on TGFβ signaling (Arndt et al., 2005; Arndt et al., 2007). A genome-wide association study suggested that mutations in the chromosomal region containing SKOR1 were linked to an ataxia known as restless leg syndrome (Kemlink et al., 2009).

The possibility that CORL plays a role in Drosophila TGFβ signaling intrigued us. We cloned CORL and noted that it had neural-specific expression. We generated a null mutant and found that EcR-B1 expression was absent from the MB of larval brains, a phenotype we mimicked with CORL-RNAi and Smad2-RNAi in wild type. In an epistasis study, constitutively active Baboon could not stimulate EcR-B1 expression in CORL mutant brains placing CORL in the TGFβ pathway. Together with evidence that mouse Corl1 specifically binds Smad3, we propose that CORL proteins are a new family of co-factors for Smad signaling.

Genomic information refers to the D. melanogaster chromosome 4 complete sequence (Release 5.1; GenBank Accession Number AE014135). cDNA LD43973 for the predicted CG11093 was from Drosophila Genomics Research Center and BAC BACR13D24 (GenBank Accession Number AC010838) was from Children’s Hospital Oakland Research Institute. 5′ RACE was carried out using the GeneRacer kit (Invitrogen). The complete cDNA sequence of CORL (clone p1.10LD43973) is available at GenBank (Accession Number JX126878). Northern and Southern blots were analyzed as described previously (Newfeld and Gelbart, 1995). Two-sided PCR to verify the FLP-FRT-generated deletion in Df(4)dCORL was conducted as described previously (Parks et al., 2004). We employed a 3′ flanking primer (ending at 956,124 bp) with the Piggy-bac WH3′minus primer plus a 5′ flanking primer (ending at 1,002,248 bp) with the Piggy-bac WH5′minus primer. UAS.dCORL was generated from LD43973 using the BglII and XhoI sites of pUAST. CORL-RNAi was constructed from a 799 bp PCR fragment from within exon 3 cloned into pWiz. CORL-scrambled-RNAi was constructed from an 800 bp PCR fragment spanning exon 3 and exon 4 that included a 54 bp intron cloned into pWiz. The intron creates a small mismatch between CORL-scrambled-RNAi and the CORL mRNA that prevents knock-down of CORL but demonstrates the absence of off-target effects for the CORL-RNAi construct with which it shares 93.25% sequence identity.

The FRT-bearing strains Pbac{WH}f07015 (956,754 bp) and Pbac{WH}f06253 (1,000,397 bp) were obtained from the Exelixis Collection. An FLP-FRT-based scheme that generates an exact interstitial deletion between two syntenic FRT elements using intrachromosomal recombination (Parks et al., 2004) was employed to create two independent strains of Df(4)dCORL. Precise deletion of four genes and the exclusion of toy from Df(4)dCORL was molecularly confirmed by two-sided PCR on single flies, as well as by genomic Southern and RNA in situ. Df(4)dCORL was characterized genetically in complementation tests with sphinx^720RW (Dai et al., 2008), toy^hdl (Pbac{XP}d0719) (Kronhamn et al., 2002) and Pbac{RB}e02096 (disrupts all predicted transcripts of CG32016). The stages of lethality tests were as described previously (Takaesu et al., 2006). Flip-out clones were as described previously (Struhl and Basler, 1993). Additional strains are: 238y.Gal4 (Aso et al., 2009), T80.Gal4 (Marquez et al., 2001), CA-Tkv (Haerry et al., 1998), CA-Babo (Brummel et al., 1999), UAS.dSno (Takaesu et al., 2006), UAS.Mad1 (Takaesu et al., 2005), GFP-Sphinx61 (M. Long, University of Chicago, IL, USA), dSmad2-RNAi and Medea-RNAi (M. O’Connor, University of Minnesota, MN, USA).

RNA in situ of embryos and larval brains with riboprobes of CORL (cDNA LD43973) and toy (cDNA GH14454) were as described previously (Takaesu et al., 2006). Embryo VC fillets were as described previously (Broadie and Bate, 1993). Antibody staining of larval brains was as described previously (Shimizu et al., 2011). Hybridoma Bank antibodies were: Mabdac1-1 (1:50), EcR-B1 (AD4.4 1:50), lacZ (40-1A 1:1000), Pros (MR1A 1:10) and Fas2 (1D4 1:5). Additional antibodies were: Drosophila Act (rabbit 1:50; Santa Cruz), Tll (rabbit 1:500; Kosman et al., 1998), Trio (rabbit 1:1000; Awasaki et al., 2000), GFP (mouse 1:10, rabbit 1:1000, rat 1:1000; Abcam) and pH3 (rabbit 1:500; Abcam). Secondary antibodies were goat anti-mouse, anti-rabbit, anti-rat and anti-guinea pig Alexa Fluor 488, 546 and 633 (Molecular Probes), and biotinylated goat anti-rabbit (Vector Labs).

Immunoprecipitation and immunoblotting were as described previously (Kawabata et al., 1998). 293T cells were transfected using FuGene6. Lysates were subject to immunoprecipitation using anti-FLAG M2 (Sigma). For mouse Corl1-Smad interactions, immunoprecipitation was followed by immunoblotting with anti-Myc 9E10 antibody (Santa Cruz) or lysates were directly blotted with anti-Myc, anti-FLAG or anti-HA.

All full-length Ski/Sno, Dac and CORL proteins from C. elegans, D. melanogaster, S. purpuratus, M. musculus and H. sapiens were included in an alignment of 18 sequences encompassing 2011 amino acids and a Maximum Likelihood tree was generated as described previously (Konikoff et al., 2010). Identification of conserved domains, coiled-coil motifs and alignment highlighting were conducted as described previously (Takaesu et al., 2006).

We identified a cDNA corresponding to CG11093, the only CORL family member in flies (Takaesu et al., 2006). Analysis of dCORL transcription by 5′RACE identified two initiation sites: a proximal start site that corresponds to the 5′ end of the cDNA; and a distal start site 10 kb upstream on the far side of the adjacent gene sphinx (Fig. 1A). These two transcripts contain distinct exons 1 and 2 with initiator methionines in exon 2. Both splice in-frame to the common exon 3 and are identical thereafter (Fig. 1B). Analysis of CORL reporter genes covering the region (N.T.T., unpublished observations) indicates that only the distal start site is meaningful.

Employing the extended ORF we conducted a new phylogenetic analysis (supplementary material Fig. S1A). Overall conservation between mammalian and fly CORL is notable (37.7% similarity) but conservation in the Sno homology domain is astonishingly high (Fig. 1C,D). There is an average of 85% similarity between Drosophila CORL and mouse Corl1/mouse Corl2 in the 197-residue Sno homology domain. Conservation between Drosophila CORL and mouse Corl1/2 in the Sno homology domain easily exceeds the similarity of the 188-residue ligand domain of fly Dpp and human BMP4 (77%) (Newfeld and Gelbart, 1995), two proteins that can function across species. Thus, members of the CORL subfamily probably have conserved functions and perhaps conserved expression patterns.

The similarity of CORL to its fly paralogs in the Sno and Dac family is less robust (supplementary material Fig. S1B,C). Comparing CORL to Sno, both the Dac and SAND domains are roughly 45% similar with the CORL domain not as well conserved. Alignments highlight similarities and differences between these proteins. Similarity in the N-terminal region of the Sno homology domain includes the APC site and Cys2-His2 motif. Differences in the C-terminal region of the Sno homology domain include the absence of the Smad4-interacting I-loop and its TCHW motif, suggesting that CORL is unlikely to bind Smad4. The last 20 residues are similar to a helix-turn motif in the dimerization domain of Siah-interacting protein (Sina in flies) (Santelli et al., 2005), suggesting a role in protein-protein interactions.

Sequence alignments identified additional similarities. Upstream of the Sno homology domain, Sno/Ski proteins contain a hydrophobic region that binds Smad2/3 (mSno 83-92; Lx₃Lx₄L). A similar region is found in CORL at a similar location (CORL 32-42; Vx₃LLx₄I; supplementary material Fig. S1D). Downstream of the Sno homology domain, Sno/Ski and CORL proteins contain a coiled-coil region that, for Sno/Ski, is involved in recruiting the co-repressor Sin3a. These conserved features suggest that CORL proteins can bind Smad2/3 and recruit co-factors.

At stage 12, CORL transcripts become detectable in the ventral cord (VC) of the central nervous system (CNS) in a single cell per segment. In stage 13 embryos, bilaterally symmetrical expression in one or two cells per VC segment is visible (Fig. 2A,B). Given that all VC neuroblasts (NBs) are present by late stage 11 (Doe, 1992), CORL-positive cells are unlikely to be NBs but instead are ganglion mother cells (GMCs) or mature cells derived from early NB divisions. During stages 14-15 CORL in the VC expands to groups of three cells per segment with larger groups in the posterior. These are located peripherally (towards the surface) roughly four to five cell diameters from the midline (Fig. 2C-E). At stage 16, one instance of new CORL VC expression is seen in a single pair of cells at the midline within abdominal segment 3 (Fig. 2G).

In the developing brain, at stage 13 CORL first appears in one or two cells within the ventral-most part of the trito- and deutocerebrum, and in small groups of cells in lateral regions of the ocular protocerebrum (Fig. 2A, lateral; 2H, dorsal). During stages 14-16, the number of CORL cells significantly expands in the ocular protocerebrum (Fig. 2C,F, lateral; 2H-J, dorsal). The CORL cells are present in roughly the same region of the ocular protocerebrum as those expressing the adjacent gene twin of eyeless (toy; Fig. 2J,K). Interestingly, at this stage toy is present in MB neurons and toy mutants die with visible MB defects (Furukubo-Tokunaga et al., 2009). The MB is the site of many cognitive functions and receives input from a variety of sources (Truman et al., 1994; Davis, 1996).

During stages 13-15 CORL transcripts become detectable anterior to the brain in pairs of cells that flank the pharynx (Fig. 2C, lateral; 2H,I, dorsal). By stage 16, CORL cells in the ocular protocerebrum form a five-pointed star pattern with a large cluster in the center and smaller groups located at the tips (Fig. 2F, lateral; 2J, dorsal). A thorough analysis of CORL expression in the embryo will be reported elsewhere. Northerns revealed a single CORL transcript of 2.6 kb present at all stages except second and third larval instars (Fig. 2L). Overall, these studies suggest that CORL expression is restricted to specific subsets of CNS cells.

CORL RNA is ubiquitously visible in the CNS of third instar larvae at a low level with several areas of strong expression. In the brain, CORL is prominent in roughly circular groups of cells in the dorsal/anterior region of the brain where the MB is located (Fig. 3A). CORL is also notable in subesophageal cells in the dorsal/posterior region of the brain. In the VC, three to five CORL cells per segment are evident but their identity is unknown. No expression is seen in the optic lobe of the brain, in imaginal disks or in any other larval tissue.

CORL is located in chromosome region 102D on the difficult to study chromosome 4. Nevertheless, we generated a small deletion removing CORL [Df(4)dCORL; Fig. 3B,C] using the FLP-FRT method of Parks et al. (Parks et al., 2004). FLP-induced recombination between two syntenic FRT sites precisely deletes the intervening region and leaves behind a functional FRT. We confirmed the endpoints of Df(4)dCORL via two-sided PCR and verified that the CORL locus is absent by genomic Southern blotting (supplementary material Fig. S2A,B). In addition to CORL, Glu-RA (metabotropic glutamate receptor), sphinx (non-coding RNA) and CG32016 are removed by Df(4)dCORL. toy is adjacent to CORL on the distal side (Furukubo-Tokunaga et al., 2009) but is not affected by Df(4)dCORL (Fig. 3C).

The stage of lethality studies revealed that 57% of Df(4)dCORL homozygotes die as larvae and an additional 22% die as pupae with 21% surviving to adulthood. This is significantly more lethality at both stages than in the parental strains. We conclude that the lethality of homozygous individuals is due to the loss of CORL because, as described in detail below, tissue-specific expression of UAS.dCORL rescues Df(4)dCORL mutant phenotypes in that tissue (Fig. 4G). Consistent with this conclusion, we noted that the parental FRT strains are fully viable, indicating an absence of secondary lethals and that adults homozygous for deletions of Glu-RA (Bogdanik et al., 2004), sphinx (Dai et al., 2008) or Pbac{RB}e02096 in CG32016 are fully viable. Thus, we consider all Df(4)dCORL pre-adult phenotypes to result from the loss of dCORL.

Df(4)dCORL mutants display a complex phenotype. Df(4)dCORL larvae exhibit poor growth, developmental delay, wander away from food precociously and prematurely evert their anterior spiracles (Fig. 3D). Df(4)dCORL adults have greatly reduced lifespan and poor female fertility. Given the presence of dCORL transcripts in the anterior/dorsal region of the brain where the MB is located and the role of the MB in a variety of behaviors, we examined the MB in Df(4)dCORL adults. These MBs display numerous structural defects (Fig. 3E-J). Among the defects are: shortened, misshapen and missing dorsal lobes, β neurons that cross the midline and γ-lobe pruning defects that can reflect the loss of EcR-B1 expression (Zheng et al., 2003).

We then investigated the role of CORL in larval MBs. Here, circulating Myoglianin (Myo) signals via the type I receptor Baboon (Babo) and Smad2 to activate expression of EcR-B1 (Zheng et al., 2003; Zheng et al., 2006; Awasaki et al., 2011). We examined EcR-B1 MB expression in both Df(4)dCORL and CORL-RNAi flip-out clone genotypes. The presence of a Smad2-binding hydrophobic region suggested the hypothesis that CORL may cooperate with Smad2 upstream of EcR-B1. A prediction of this hypothesis was that loss of CORL would lead to loss of EcR-B1 in the MB.

An alternative hypothesis we tested first is that CORL is a general MB factor rather than one specific for Smad2-dependent events. We examined three Smad2-independent MB markers in Df(4)dCORL larvae: Dachshund (Dac; nuclear in neurons with functions in α and β lobe maturation) (Martini et al., 2000), Tailless (Tll; nuclear in NBs and GMCs with functions in proliferation) (Kurusu et al., 2009) and Prospero (Pros; nuclear in GMCs then cortical in subset of neurons with functions in differentiation) (Doe et al., 1991). Expression of these proteins in wild-type and Df(4)dCORL larval MB was indistinguishable (supplementary material Fig. S2C-L), as were the levels of mitotic marker phospho-Histone3 (not shown). There is no evidence that CORL is a general regulator of transcription in the MB. A second alternative hypothesis we examined was that CORL is required for expression of Activin (Act). This seemed reasonable because Act has significant expression in a pair of subesophageal neurons whose axons trace a path similar to the distribution of CORL cells in the larval brain. We found that Act expression is unaltered in Df(4)dCORL (supplementary material Fig. S4M-O).

By contrast, experiments with the MB markers Trio (all MB neurons in larvae; Fig. 4A,B) or 238y.Gal4 (all MB neurons; Fig. 4C,D) in Df(4)dCORL larvae revealed that EcR-B1 was absent in the MB of most individuals. EcR-B1 in non-brain neural tissues was unaffected in Df(4)dCORL larvae (e.g. ring gland; supplementary material Fig. S2K,L). EcR-B1 expression in the VC was greatly reduced (supplementary material Fig. S2M,O) in Df(4)dCORL larvae, suggesting that CORL plays roles in the CNS outside the MB. An analysis of those functions will be reported elsewhere.

The loss of EcR-B1 expression in the MB of Df(4)dCORL larvae is phenocopied by 238y.Gal4 driving dCORL-RNAi in wild-type larvae (Fig. 4E,F). Rescue experiments employing 238y.Gal4 driving UAS.dCORL in Df(4)dCORL larvae significantly restored EcR-B1 expression only in 238y.Gal4 expressing cells (compare Fig. 4G,D). The rescue experiment formally demonstrates that it is the deletion of CORL that leads to the loss of EcR-B1 in Df(4)dCORL larvae. The data are consistent with the hypothesis that CORL cooperates with Smad2 upstream of EcR-B1.

We then examined flip-out clones of CORL-RNAi, Smad2-RNAi and Medea-RNAi. If these studies phenocopy the loss of EcR-B1 MB expression in Df(4)dCORL larvae (which is rescued by UAS.dCORL) it would further support our hypothesis that CORL facilitates Smad2 signaling upstream of EcR-B1. We include Medea owing to its established role as a partner for Smad2, although a role in EcR-B1 activation has not yet been formally shown. In the ventral/posterior region near the calyx, EcR-B1 is prominent in the cell bodies of MB neurons. EcR-B1 at this location is unaffected by co-expression of GFP and CORL-scrambled-RNAi in flip-out control clones (Fig. 5A). With CORL-RNAi, EcR-B1 expression is lost in a large MB clone but not in single cell MB clones or in clones outside the MB (Fig. 5B). Large clones containing multiple cells are caused by flip-out in an MB neuroblast. In all neurons derived from the flipped-out NB, CORL-RNAi would be present prior to Smad2 signaling upstream of EcR-B1 activation. In these cells, CORL-RNAi prevents Smad2 activation of EcR-B1. In cells adjacent to large clones, EcR-B1 is not affected, indicting that knockdown of CORL cell-autonomously eliminates EcR-B1. Alternatively, single cell clones reflect a flip-out in a differentiated MB neuron. In these flipped-out neurons, EcR-B1 was already activated by Smad2 signaling or by either of two non-Smad pathways upstream of EcR-B1 [Rho-GTPases (Ng, 2008); Ftz-F1 (Boulanger et al., 2011)]. In differentiated MB neurons, knockdown of CORL does not affect EcR-B1, indicating that dCORL does not participate in EcR-B1 maintenance.

The phenotype of CORL-RNAi clones mimics exactly the phenotype seen in Smad2-RNAi (Fig. 5C) and Medea-RNAi clones (Fig. 5D). In these three RNAi genotypes, EcR-B1 expression is lost in large clones generated by flip-out in an MB neuroblast but not in single cell clones where the flip-out occurred in a differentiated MB neuron or in clones outside the MB. To further improve our confidence in the data, we conducted a second set of flip-out studies in the anterior/dorsal region of the MB adjacent to the NB/GMC cluster. Data from these anterior/dorsal studies perfectly mimicked our analysis of the ventral/posterior region (supplementary material Fig. S3A-D). Both sets of flip-out clone data further support the hypothesis that CORL contributes to Smad2 signaling upstream of EcR-B1 activation in the larval MB.

In a rigorous test of this hypothesis, we examined whether loss of CORL could prevent a constitutively active form of Babo (CA-Babo; Brummel et al., 1999) from stimulating EcR-B1 expression in the MB of Df(4)dCORL larvae. The hypothesis predicts that EcR-B1 will be absent in this experiment, which was observed when employing 238y.Gal4 to drive CA-Babo (compare Fig. 4H with 4G). This epistasis data strongly demonstrates a requirement for CORL in Babo-Smad2 signaling upstream of EcR-B1 activation in larval MB.

To determine whether the requirement for CORL in EcR-B1 activation was based on physical interactions between CORL and Smad2, we analyzed mouse Corl1 in biochemical studies. We examined the ability of mouse Corl1 to bind every Smad protein and the influence of receptor activation on mouse Corl1-Smad binding. First, we found that mouse Corl1 interacted strongly with mouse Smad3 (a homolog of Drosophila Smad2) and weakly with mouse Smad8 (homolog of Mad; Fig. 6A). Second, we demonstrated that mouse Corl1-Smad3 complex formation is strongly increased by TGFβ receptor activation. Comparison with the well-known TGFβ repressor Ski (which binds mouse Smad2/3 and mouse Smad4) revealed that Ski interacts with mouse Smad3 more strongly than mouse Corl1 (Fig. 6B). The extraordinary sequence conservation of CORL proteins and the biochemical data for mouse Corl1 strongly suggest that the requirement for CORL in Smad2 signaling is based upon CORL-Smad2 physical interactions.

Given that Sno/Ski proteins have both negative and positive effects on TGFβ signaling, we examined CORL flip-out clones in the MB. In both anterior/dorsal (Fig. 7A) and ventral/posterior clones (Fig. 7B), overexpression of CORL led to cell-autonomous loss of EcR-B1 expression. Consistent with these results, luciferase assays of TGFβ/Activin signaling showed that mouse Corl1 and Ski can repress TGFβ/Activin responses in a dose-dependent manner (supplementary material Fig. S4). This led us to wonder whether CORL overexpression functions as a general Smad inhibitor. We drove CORL in wing disks, where it is not normally expressed, and found that this led to vein truncations that phenocopy expression of Sno and dominant-negative Mad (protein encoded by the Mad¹ allele; Fig. 7C-F). Thus, ectopic CORL can inhibit Dpp-Tkv-Mad signaling that is required for vein formation. To rigorously test this hypothesis, we examined the ability of ectopic CORL to block CA-Tkv (Haerry et al., 1998) from stimulating vein overgrowth (Fig. 7G). Co-expression of CORL and CA-Tkv significantly, though not completely, suppressed the CA-Tkv phenotype (Fig. 7H), demonstrating that ectopic CORL can non-specifically antagonize Smad signaling.

Although the core elements of the TGFβ signal transduction pathway (ligands, receptors and Smads) and their various subfamily specificities have been known for over a decade, efforts to uncover new mechanisms influencing this potent cell fate/cell cycle regulator continue unabated. Previously unknown mechanisms of truly general application are occasionally uncovered (e.g. mono- and deubiquitylation) (Dupont et al., 2012) but the majority of interest and effort in the fields of developmental biology and oncology focus on the identification of new factors that influence TGFβ-dependent cell fate or cell cycle decisions in a single cell type. Analysis of these new pathway components will provide information on as yet unknown developmental processes and suggest potential targets for therapeutics.

Here, we have moved this effort forwards significantly by demonstrating that CORL is a new player in TGFβ signal transduction: as a facilitator of Smad2 signaling in loss-of-function studies; and as an antagonist of Smad2 and Mad signaling when ectopically expressed. This finding has wide implications because CORL belongs to a highly conserved gene family that is closely related to the family of Sno/Ski proteins, which also appear to have a dual role in regulating TGFβ signaling. There are clear similarities between CORL and Sno data from fly loss- and gain-of-function phenotypes. Loss-of-function studies showed that, in both Df(4)dCORL and CORL-RNAi genotypes, the loss of CORL led to the loss of Babo-Smad2-dependent EcR-B1 expression in the larval MB. In Sno^Ex4B/Sno^sh1402 genotypes, the loss of Sno led to reduced expression of cell cycle markers such as phospho-histone 3 in the larval optic lobe (Takaesu et al., 2006). Gain-of-function studies showed that overexpression of CORL or Sno in wing disks antagonized Mad signaling, leading to vein defects.

This functional correspondence does not extend to CORL and Sno expression during development, suggesting that although they may have similar effects on TGFβ signaling, they do so at different times and in different tissues. CORL is probably expressed as a single transcript that encodes one protein and is restricted to the embryonic and larval CNS where it is present at high levels in a small number of cells. Sno is expressed as multiple transcripts that become multiple proteins very broadly in both embryos and larvae. Sno is expressed in many tissues that do not express CORL, such as the embryonic epidermis and larval imaginal disks (Takaesu et al., 2006). In the embryonic CNS, CORL expression precedes Sno by two developmental stages (12 versus 14) but given its ubiquity in this tissue, Sno is almost certainly present in cells that express CORL beginning at stage 14, although this has not been formally shown. Within the third instar larval CNS, the only cells with significant Sno expression are in the optic lobe (Quijano et al., 2010) where CORL is not expressed.

Biochemical differences between CORL and Sno proteins dictate that they accomplish their similar effects on TGFβ signaling via distinct mechanisms. At the amino acid level, the key distinction is likely to be the absence in CORL of the Smad4-binding TCHW motif in Sno (supplementary material Fig. S1C). Only the threonine residue is present in mammalian CORL proteins (Fig. 1C). By contrast, all Sno/Ski and CORL family members contain a hydrophobic region upstream of the Sno homology domain that strongly binds Smad2/3. Thus, the ability of CORL family members to bind preferentially to the TGFβ/Activin transducer Smad3 (mouse CORL1 explicitly and Drosophila CORL by analogy, based on sequence conservation exceeding that of the cross-functional proteins Dpp and BMP4) is distinct from the ability of Sno to bind Smad2, Mad and Medea.

The ability of Sno to bind multiple Smads allows it to function as a ‘pathway switch’ by shunting Medea between pathways – reducing Dpp/Mad while stimulating Activin/Smad2. Positive and negative actions on TGFβ signaling are inherent in the mechanism of action of Sno. The effect of CORL on Smad2 signaling cannot be explained in that way. Our model proposes instead that the dual activity of CORL is based on its function as a dose-sensitive co-factor for Smad2, as suggested by the mouse Corl1 luciferase assays. In this model, loss or gain of CORL in an otherwise wild-type individual leads to aberrant Smad2 signaling and loss of EcR-B1, either through the loss of a required co-factor or by the sequestration of Smad2 into nonfunctional complexes when overwhelmed by excess CORL. CORL-mediated inhibition of Dpp-Tkv-Mad signaling when ectopically expressed may arise from its ability to weakly recognize Mad, as suggested by the weak binding of mouse Corl1 to Smad8 (Fig. 6A) and its sequestration into non-functional complexes.

This model explains the cell-type specificity of EcR-B1 activation in MB neurons by circulating Myo (Awasaki et al., 2011) via ubiquitous babo and Smad2 (Zheng et al., 2003; Brummel et al., 1999): MB neurons are the only cells exposed to Myo that have sufficient CORL function to facilitate Smad2 transcriptional activity. However, the model requires two caveats. First, CORL loss of function dose effects are only penetrant if the loss exceeds 50% as heterozygous Df(4)dCORL flies appear wild type. Second, CORL is not a universal Smad2 co-factor as Smad2 regulates Ecdysone biosynthesis in the ring gland without CORL (Gibbens et al., 2011).

Significant levels of conservation for both neural-specific expression and amino acid sequence for CORL proteins suggest that the function of CORL in TGFβ/Activin signaling (facilitation of Smad2 activity) and CORL developmental roles (tissue-specific activation of gene expression) will be conserved in vertebrates. This hypothesis is supported by a recent paper describing a loss-of-function phenotype for mouse Corl2 in mice (Wang et al., 2011). In addition to gross cerebellar defects, homozygous mouse Corl2 mutant mice lack sonic hedgehog expression in Purkinje cells. The authors view this result through the lens of overexpression studies showing that mouse Corl2 represses BMP signaling in cell culture to generate a two-step explanation for the phenotype: BMP repression is necessary for sonic hedgehog activation. Viewed through the lens of the CORL loss-of-function MB phenotype, a one-step explanation is obtained: mouse Corl2 facilitates TGFβ/Activin signaling upstream of sonic hedgehog activation.

In adult mammals, the presence of human and mouse CORL proteins in Purkinje cells implies a role in motor control, the primary physiological function of the cerebellum. A large number of ataxias are caused by disruption of normal Purkinje cell function (Orr, 2010). A study connecting mutations in the chromosomal region containing FUSSEL15 (human CORL1/SKOR1) to the familial movement disorder restless leg syndrome (Kemlink et al., 2009) suggests the possibility that loss of human CORL1 is a factor in the genesis and/or progression of a subset of ataxias. The possibility that Df(4)dCORL adults may be a new model for ataxia is under investigation.

In summary, loss-of-function studies demonstrated a formal requirement for CORL in Babo-Smad2 signaling, while gain-of-function studies showed that ectopic expression of CORL antagonizes Smad signaling nonspecifically. Our model for the dual activity of CORL proposes that it functions as a tissue-specific, dose-sensitive co-factor for Smad2. The conservation of neural expression and strong sequence homology for all CORL proteins suggests that they are a new family of co-factors for Smads.

We thank Yuto Kamiya, Charlotte Konikoff, Tomohiro Ogami, Barrett Pfeiffer, Alice Schmid and Robert Wisotzkey for technical assistance. We appreciate insightful discussions and/or support from Kevin Cook, Manyuan Long, Mike O’Connor and Tetsuya Tabata. Reagents were provided by Children’s Hospital Oakland, Developmental Studies Hybridoma Bank, Drosophila Genomics Resource Center, Exelixis Collection, Indiana Stock Center, Manyuan Long, Mike O’Connor, Yuichi Ono, John Reinitz and Tetsuya Tabata.