Results

Yeast two-hybrid screens identify Alpha 4 as an interacting partner of both MID1 and MID2
In order to begin to determine the identity of potential interacting factors and the processes with which MID1 is involved, we performed a yeast two-hybrid assay utilising the full length MID1 protein (which shares 94.9% identity, 99.9% similarity with mouse Mid1) as bait in a screen of a mouse 10.5 dpc (days post-coitum) whole embryo cDNA library. At 10.5 dpc, expression of Mid1 is seen essentially throughout the embryo although strongest levels of expression are evident in highly proliferating tissues such as the developing craniofacial region [27]. The use of the mouse 10.5 dpc whole embryo library was therefore chosen to maximise the likelihood of detecting functionally relevant interactors.
Approximately 1 × 106 cDNAs were initially screened using MID1 as bait. Potential interacting clones were selected based on the activation of the three endogenous reporter genes (His, lacZ, and Leu). Recovery and sequencing of the positives from the library screen demonstrated nine cDNA clones with insert sizes ranging from 1.0 kb to 1.3 kb but representing the same gene. Five additional, singly represented putative interacting clones were also identified but do not constitute part of this report. Database searches identified the nine similar clones as encoding Alpha 4, a rapamycin-sensitive regulatory subunit of protein phosphatases 2A (PP2A) and other PP2-type phosphatases. By comparison with the published murine Alpha 4 sequence, all clones were judged as being essentially full-length. Notably, the nine in-frame cDNAs represented a minimum of seven independent clones as 5' fusion to the GAL4 activation domain occurred at either nucleotide +4, +7, +10 or +13 and most clones containing different length polyadenylated tails. To confirm that this interaction was not an artefact of the independent GAL4 activation domain and DNA binding domain fusion events, the full-length cDNAs were interchanged such that Alpha 4 was fused to the GAL4-DBD and MID1 to GAL4-AD. Co-transformation of these two constructs into MaV203 gave similar levels of growth on 3AT (Fig 1A), ie. a level of activation comparable to the strongest interacting proteins, Fos and Jun (Fig 1A, Control E – colony 10).
Figure 1  Alpha 4 interacts with MID1 and MID2. (A) Yeast two-hybrid analysis of the interaction between MID1 and Alpha 4 as well as MID2 and Alpha 4. Yeast agar plate (leu- trp- his-, 75 mM 3-AT) showing growth for MID1/Alpha 4 and MID2/Alpha 4 interactions as well as positive control two-hybrid combinations and no growth for negative controls. (B) Detection of full-length myc tagged-Alpha 4 when co-expressed with GFP-MID1 and GFP-MID2 fusion proteins in transiently transfected Cos1 cells. (a) GFP-MID1 fluorescence (green), (b) anti-myc antibody detecting myc-Alpha 4 localisation (red), (c) overlay of (a), (b) showing co-localisation of the myc-Alpha 4 fusion protein and GFP-MID1, with DAPI stain for DNA (blue). (d) GFP-MID2 fluorescence (green), (e) myc-Alpha 4 localisation (using same detection as b) (red), (f) overlay of (d), (e) with DAPI (blue) showing co-localisation of the myc-Alpha 4 fusion protein and GFP-MID2. (g) Detection of transiently expressed myc-Alpha 4 fusion protein in Cos1 cells, (h) overlay of (g) and DAPI stain showing cytoplasmic distribution of myc-Alpha 4 fusion protein.   Given that MID1 and MID2 share 77% overall amino acid identity (92% similarity), we wished to investigate whether MID2 was capable of binding to the same proteins, such as Alpha 4, or whether this interaction was specific for MID1. This was investigated using two approaches: a direct yeast two-hybrid test of the ability of MID2 to bind Alpha 4, and a full two-hybrid screen of the 10.5 dpc whole mouse embryo cDNA library this time using MID2 as bait. The direct test demonstrated that MID2 did indeed interact with Alpha 4. Comparison of the growth of the MID2/Alpha 4 transformed yeast on plates (SC -Leu, -Trp, -His) containing 75 mM 3AT indicated the interaction between MID2 and Alpha 4 was as strong, or stronger, than that observed for MID1 (Fig 1A). A similar result was obtained when assessing the conversion of Xgal (data not shown). In the full yeast two-hybrid screen, some potentially novel interactors were identified for MID2. However, the majority of the putative interacting clones, and again the only sequences represented more than once in the isolates, represented Alpha 4.

MID1 and MID2 tether Alpha 4 to the microtubules
To investigate whether Alpha 4 also associated with MID1 and MID2 in a homologous (mammalian) system, we co-transfected both GFP-MID1 (or MID2) with myc epitope-tagged Alpha 4. Transfection of the myc-tagged Alpha 4 expression construct alone resulted in a diffuse distribution of the myc-Alpha 4 fusion protein throughout the cytoplasm in most cells (Fig 1B; g, h). However, in some cells, a faint filamentous distribution that resembled the appearance of microtubules could be seen along with the cytoplasmic protein (data not shown). In contrast, in all cells co-transfected with GFP-MID1 and myc-tagged Alpha 4, there was negligible diffuse cytoplasmic staining. Instead, essentially all the myc-tagged Alpha 4 protein displayed a filamentous meshwork of staining that completely co-localised with MID1 along the length of the microtubules (Fig 1B; a-c). This result also suggested that MID1 was a limiting factor in the tethering of Alpha 4 to the microtubule network, a conclusion supported by western analysis that shows low levels of MID1 in these cells (data not shown). Co-expression of GFP-MID2 and myc-tagged Alpha 4 similarly resulted in co-localisation along the microtubules (Fig 1B; d-f).

Alpha 4 does not co-localise with MID1 or MID2 proteins harboring in-frame B-box deletions
Both endogenous mutant MID1 protein in OS patient cells [18] and various transiently expressed mutant MID1-GFP fusion proteins form cytoplasmic clumps [5,15,17,18]. We chose to exploit this previous observation by co-transfecting GFP-tagged MID1ΔCTD (or MID2ΔCTD) with a construct expressing a myc-tagged Alpha 4 protein in order to investigate whether Alpha 4 still remained bound to MID1 within such aggregates. The results clearly showed a distribution of myc-Alpha 4 that was indistinguishable from the clumped MID1ΔCTD and MID2ΔCTD truncated proteins, indicating that Alpha 4 indeed aggregates with the truncated MID1 and MID2 proteins (Fig 2B; m-o).
Figure 2  MID1/Alpha 4 and MID2/Alpha 4 interactions are maintained in all MID domain-specific deletions except for those involving the B-boxes. (A) Yeast two-hybrid analysis shows that the MID B-boxes are required for interaction with Alpha 4. The interaction of Alpha 4 with MID1ΔBB (21), or MID2ΔBB (23), is compared to its interaction with MID1ΔCC (22), or MID2ΔCC (24). (B) Subcellular localisation of myc tagged-Alpha 4 when co-expressed in Cos-1 cells with MID1 domain-specific deletions as GFP fusion proteins. Fluorescence detection of GFP-MID1ΔRF (a), GFP-MID1ΔBB (d), GFP-MID1ΔCC (g), GFP-MID1ΔFNIII (j), and GFP-MID1ΔCTD (m). Anti-myc antibody detection of myc-Alpha 4 in the same cells as expressing the various MID1 domain-specific deletions (b,e,h,k,n). Overlay of the GFP and anti-myc images of the same cells merged with DAPI stain of nuclei (c,f,i,l,o). All merged images, with the exception of (f) show co-localisation of myc-Alpha 4 with the various MID1 domain deletions. In (f), myc-Alpha 4 fails to co-localise with GFP-MID1ΔBB in small cytoplasmic aggregates.   To further define the motif in MID1 (and MID2) responsible for the interaction with Alpha 4, we generated in-frame deletions of all other motifs (Table 1) and fused the resulting clones to GFP. We initially transfected each construct alone to examine the effect of each deletion on the intracellular localisation of the proteins. The results showed that each motif, or at least their conserved spacing or arrangement, was essential for the distribution of MID1 (and MID2) along the length of the microtubules. The distribution of the individual domain deleted MID proteins when transfected alone was indistinguishable from their distributions when transfected along with myc-Alpha 4 (see below). Consequently, the GFP fluorescence images in Fig 2B can also be considered a representation of each distribution pattern in the absence of co-transfected myc-Alpha 4. Like the ΔCTD constructs, deletion of either the B-boxes or the FNIII domain also resulted in cytoplasmic clumps or speckles although these usually appeared smaller and greater in number and the speckles still appeared to co-localise with microtubules (data not shown). Both ΔRING proteins, in contrast, showed variability in their distribution with most transfected cells still showing association along the length of the microtubules. However, the microtubule association of these ΔRING proteins often did not extend to the cell periphery. Notably, deletion of the coiled-coil motif in each protein resulted a diffuse cytoplasmic distribution, suggesting that these proteins had lost their ability to associate with microtubules (see Fig 2B; g-i).
Table 1  The MID1 and MID2 deletion constructs used in pEGFP-C2 for cellular co-localisation analysis, co-immunoprecipitation and in pPC86/pDBLeu for interaction analysis with Alpha 4. Terms: RF denotes a C3HC4 RING finger; BB denotes C2H2 B-Boxes; CC denotes a Coiled-coil motif; FNIII denotes a Fibronectin Type III domain and CTD denotes a C-terminal domain (encompassing a SPRY domain).   We then individually co-transfected each MID1 (and MID2) deletion construct with myc-Alpha 4 in an attempt to define the domain responsible for the interaction with Alpha 4. Co-transfection of either the ΔRING or ΔFNIII proteins with myc-Alpha 4 (Fig 2B; a-c and j-l, respectively) resulted in co-localisation of Alpha 4 with the abnormally distributed MID1 and MID2 proteins, as seen with the ΔCTD proteins. Co-expression of either ΔCC construct with Alpha 4 resulted in both proteins exhibiting a diffuse cytoplasmic distribution although the pattern of each was still suggestive of the two proteins being able to interact (Fig 2B; g-i). Strikingly, when either of the ΔBB constructs was co-expressed with Alpha 4, the mutant MID1 and MID2 proteins still formed cytoplasmic clumps but, in both cases, Alpha 4 remained diffuse in the cytoplasm (Fig 2B; d-f). These results imply that the B-boxes (and/or the linker residues between the B-boxes and RING motifs) are primarily responsible for the interaction with Alpha 4.
To confirm these results, we cloned the MID1 and MID2 domain-deletion constructs in-frame into pPC86 and co-transformed into MaV203 with pDBLeu-Alpha 4. As expected from the immunofluorescence data, activation of all two-hybrid reporter genes, at a level similar to that seen with the full-length MID proteins, was observed with the ΔRING, ΔFNIII and ΔCTD proteins (results not shown), confirming that neither of these domains mediated binding to Alpha 4. Notably, both ΔCC proteins also interacted with Alpha 4 (Fig 2A) suggesting that the two proteins were indeed still interacting in the cytoplasm in the immunofluorescence experiments despite not being associated with microtubules. However, the strength of the interaction between the ΔCC proteins and Alpha 4 appeared to be reduced compared to the full-length, ΔRING, ΔFNIII and ΔCTD (data not shown). In contrast, but again confirming the immunofluorescence result, the ΔBB constructs did not activate reporter gene expression indicating the interaction between Alpha 4 and either MID protein was abolished by removal of this motif (Fig 2A).

The coiled-coil domains of MID1 & MID2 are required for homodimerisation and microtubule binding but not Alpha 4 interaction
Immunoprecipitation experiments have previously shown that MID1 can form homodimers/homomultimers [17]. To investigate whether homodimerisation was a prerequisite for binding Alpha 4 and/or association with the microtubules, we first tested whether MID1 can interact with itself in the context of the yeast two-hybrid system. The results clearly indicated that the MID1-MID1 interaction is strong (Fig 3A). Domain-deletion constructs (in pDBLeu and pPC86 vectors) were then introduced into the yeast two-hybrid system. Unlike the proteins harboring a deletion of either the RING, B-boxes or FNIII motifs, the MID1 protein lacking the coiled-coil motif had significantly reduced capacity to bind the wild-type MID1 protein, inferring that the coiled-coil domain is critical for efficient homodimerisation (result not shown). Assessment of MID1 using the MultiCoil algorithm [28] supports the conclusion of dimer formation (not trimer) and that the first of the two coiled-coils in this domain is largely responsible for this property. Interestingly, the MID1 protein in which the CTD motif was removed also demonstrated a reduced ability to bind the wild-type MID1 in this system, although this effect was not as marked as that seen for the ΔCC proteins. Notably, the yeast two-hybrid and immunofluorescence experiments demonstrate that Alpha 4 can still interact with the MID1 and MID2 ΔCC proteins. This observation indicates that Alpha 4 must be able to interact with MID monomers although its tethering to the microtubule network is dependent on MID dimerisation, that is; the MID proteins only associate with microtubules as dimers.
Figure 3  MID1 and MID2 can homo- and heterodimerise with one another. (A) Yeast two-hybrid assay for MID1 and MID2 multimerisation. Yeast agar plate (leu- trp- his-, 75 mM 3-AT) showing growth for MID1/MID1 (25), MID1/MID2 (26 and 27) and MID2/MID2 (28). (B) MID1 and MID2 co-localise to the microtubules. Co-expression of GFP-MID1 (a) and myc-MID2 (b) in transiently transfected Cos1 cells showing co-localisation to the microtubular cytoskeleton in an overlay (c) with a DAPI stained nucleus (blue). (C) Co-immunoprecipitation of MID1 and MID2 homo- and heterodimers. Shown are extracts from Cos1 cells, transfected with GFP-MID1 (lane 1), GFP-MID2 (lane 2), myc-MID1 (lane 3), myc-MID2 (lane 4), GFP-MID1 and myc-MID1 (lane 5), GFP-MID2 and myc-MID1 (lane 6), GFP-MID1 and myc-MID2 (lane 7) and GFP-MID2 and myc-MID2 (lane 8). Samples were immunoprecipitated with anti-GFP antibody/protein-A sepharose beads, separated on a 8% SDS polyacrylamide gel, transferred to a nitrocellulose membrane and blotted with anti-c-myc antibody to detect co-precipitate protein.

MID1 & MID2 can form heterodimers on microtubules
Given their high level of identity, we investigated whether MID1 and MID2 can also form heterodimers using the yeast two-hybrid system, immunofluorescence of transiently transfected Cos1 cells and co-immunoprecipitation. Co-transformation of the yeast MaV203 strain with both pDBLeu-MID1 and pPC86-MID2 (or in the reverse vector combination) resulted in high level activation of all reporter genes indicating a strong interaction that was comparable to the strength of the MID1-MID1 homo-interaction (Fig 3A). These findings are contradictory to initial reports from a study by Cainarca et al [17] but have been confirmed in experiments involving transient transfection of GFP-MID1 and myc-MID2 fusion constructs (Fig 3B) as well as by co-immunoprecipitation (Fig 3C). Introduction of individual domain-deletions of MID1 together with full-length or domain-deletion MID2, and vice versa, into Cos1 cells and the yeast strain MaV203 using the relevant constructs showed, as expected, that the coiled-coil motif was largely responsible for mediating this heterodimerisation (data not shown).

The B-boxes of MID1 & MID2 are sufficient to bind Alpha 4
To verify that the Alpha 4 interaction and the association of the complex with microtubules were indeed dependent on the B-boxes and coiled-coil, respectively, and not just an artefact of altering the relative spacing of remaining domains, we fused the MID1 B-boxes and coiled-coil motif, or the coiled-coil motif alone, in-frame to GFP and co-transfected with the myc-Alpha 4 construct. Of interest was the observation that the MID1 coiled-coil domain alone fused to GFP (GFP-M1CC) resulted in cytoplasmic clumping (Fig 4B; a), supporting the notion that the coiled-coil domain is required for MID1 dimerisation. Consistent with this is that the phenomenon of cytoplasmic clumping seen in most OS patients has only been observed in those cases where the expressed mutant MID1 protein harbours a mutation outside the coiled coil motif [5]. Notably, however, in cells co-transfected with the GFP-M1CC and myc-Alpha 4 constructs, Alpha 4 did not co-localise with these M1CC clumps (Fig 4B; a-c). Like the M1CC protein, the construct expressing the fusion between GFP and the MID1 B-boxes plus coiled-coil domains (GFP-M1BBCC) resulted in clumps within the cytoplasm. Importantly, and in contrast to the co-transfection of M1CC, Alpha 4 was found to co-localise with the GFP-M1BBCC fusion protein (see Fig 4B; d-f). We did not undertake the generation of a GFP-M1BB (B-boxes alone) construct as it was predicted that the resultant GFP fusion protein would show a diffuse cytoplasmic distribution largely indistinguishable from Alpha 4. Instead, the MID1 B-boxes were cloned into pDBLeu and tested directly for their ability to interact with Alpha 4 in the two-hybrid system. The result (Fig 4A) clearly supports the conclusion that the B-boxes are sufficient to bind Alpha 4.
Figure 4  The B-boxes of MID1 are sufficient to bind Alpha 4. (A) Yeast two-hybrid analysis shows that the MID1 B-boxes (pPC86-M1BB) alone can interact with Alpha 4 (pDBLeu-Alpha 4) (31). The wild-type MID1-Alpha 4 interaction (30) and a control with the pPC86-M1BB and no interaction partner (29) were included for comparison. (B) Immunofluorescence assay highlights the importance of the B-boxes for Alpha 4 binding. Subcellular distribution of the GFP-fused MID1 coiled coil domain, GFP-M1CC (a) and myc-Alpha 4 (b) in the same cell shows that the two proteins do not co-localise, as seen in the merged image (c). However, the MID1 fragment, GFP-M1BBCC (d) and the myc-Alpha 4 (e) do co-localise in cytoplasmic speckles as seen in the merged image (f). Both merged images, (c) and (f), also show DAPI stain (blue), which indicates the position of the nucleus.

MID1 is phosphorylated on serine and threonine residues
As Alpha 4 is a regulator of PP2-type serine/threonine phosphatases, we investigated whether MID1 and MID2 might themselves be phosphorylated and hence possible targets of Alpha 4/phosphatase action. Due to the low levels of endogenous MID1 and MID2 in all tested cultured cell lines, we performed western analysis of immunoprecipitated MID-GFP proteins using extracts of Cos1 cells that had been transiently transfected with the various expression constructs. Immunoprecipitation with anti-GFP antibodies followed by western analysis using anti-phosphoserine and anti-phosphothreonine antibodies showed that MID1- and MID2-GFP fusion proteins were phosphorylated on both serine and threonine residues (Fig 5A). Similar analysis using anti-phosphotyrosine antibodies failed to demonstrate phosphorylation of tyrosine residues on either protein (result not shown).
Figure 5  MID1 contains phosphorylated serine and threonine residues. (A) Extracts from untransfected Cos1 cells (lanes 1 & 3) or Cos1 cells transfected with GFP-MID1 (lanes 2 & 4) were immunoprecipitated with anti-GFP antibody/protein-A sepharose beads and analysed by western blot analysis using either an anti-phosphoserine antibody (lanes 1 & 2) or anti-phosphothreonine antibody (lanes 3 & 4). Protein bands in lanes 2 and 4 indicated that GFP-MID1 contains both phosphoserine and phosphothreonine residues. (B) Domain-specific deletions of MID1 were used in an attempt to crudely map locations of the phosphorylated serine and threonine residues. Shown are extracts from Cos1 cells transfected with full-length GFP-MID1 (lane 1), GFP-MID1ΔRF (lane 2), GFP-MID1ΔBB (lane 3), GFP-MID1ΔCC (lane 4), GFP-MID1ΔFNIII (lane 5), and GFP-MID1ΔCTD (lane 6). The samples were immunoprecipitated with anti-GFP antibody/-protein-A sepharose beads, separated on 8% SDS polyacrylamide gels, transferred to nitrocellulose membranes and blotted with anti-phosphoserine antibody (top panel) or anti-phosphothreonine antibody (bottom panel). (C) Computer assisted detection of potential serine/threonine phosphorylation sites in MID1. Potential serine/threonine kinase consensus phosphorylation sites in MID1 were identified using NetPhos 2.0 software [38]. Examination of a multiple alignment of available MID1 and MID2 sequences was carried out to identify fully conserved putative phosphorylation sites. A diagrammatic representation of this analysis shows 16 conserved sites depicted as dots (red for serine and blue for threonine) along the length of a representative MID1 protein (numbered residue positions are also depicted at the top of diagram). The actual kinases that recognise the residues are listed on the left of the figure; CaMII (R-X-X-S/T-X), CKI (Sp/Tp-X2–3-S/T-X), CKII (X-S/T-X-X-D/E), GSK3 (X-S/T-X-X-X-Sp), MAPK (P-X-S/T-P), PKA (R-X1–2-S/T-X), PKC (X-S/T-X-R/K), PKG ((R/K)2–3-X-S/T-X).   In an attempt to define the location of the sites in MID1 that were phosphorylated, we performed similar immunoprecipitation and western analysis but this time using extracts of Cos1 cells that had been transfected with the individual domain-specific deletion constructs (Fig 5B). That the overall phosphorylation of the MID1 fusion protein was not significantly affected by deletion of any individual domain may suggest that MID1 is phosphorylated at multiple threonine residues along the protein. However, using the anti-phosphoserine antibody, no serine phosphorylation of the ΔBB protein was detected suggesting that most serine phosphorylation in MID1 occurs at or near the B-boxes.

Identification of potential sites of phosphorylation in MID1 and MID2
Computer prediction of potential target residues for phosphorylation by serine/threonine kinases not surprisingly identified numerous consensus sites throughout both human MID1 and MID2. Given the conservation of the rapamycin-sensitive pathway from yeast to mammals and the fact that Alpha 4 binds to both MID1 and MID2 which are only 77% identical, we reasoned that any functionally relevant phosphorylation site should be conserved across species and in both MID1 and MID2 proteins. Examination of available orthologous MID1 and MID2 sequences (from seven and three species, respectively) showed that sixteen of these sites (6 threonines and 10 serines) were fully conserved between all MID proteins (Fig 5C). Interestingly, two of these sites (Ser92 and Ser96), which represent consensus phosphorylation sites for GSK3 and MAPK/CKI/CKII respectively, are the only two conserved serine residues in the amino-terminal half of the protein and, as both are located in the region deleted in the ΔBB constructs of MID1 and MID2, are likely to be the primary sites of serine phosphorylation in these proteins.