Binding of the Concave Surface of the Sds22 Superhelix to the α4/α5/α6-Triangle of Protein Phosphatase-1

Functional studies of the protein phosphatase-1 (PP1) regulator Sds22 suggest that it is indirectly and/or directly involved in one of the most ancient functions of PP1,i.e. reversing phosphorylation by the Aurora-related protein kinases. We predict that the conserved portion of Sds22 folds into a curved superhelix and demonstrate that mutation to alanine of any of eight residues (Asp148, Phe170, Glu192, Phe214, Asp280, Glu300, Trp302, or Tyr327) at the concave surface of this superhelix thwarts the interaction with PP1. Furthermore, we show that all mammalian isoforms of PP1 have the potential to bind Sds22. Interaction studies with truncated versions of PP1 and with chimeric proteins comprising fragments of PP1 and the yeast PP1-like protein phosphatase Ppz1 suggest that the site(s) required for the binding of Sds22 reside between residues 43 and 173 of PP1γ1. Within this region, a major interaction site was mapped to a triangular region delineated by the α4-, α5-, and α6-helices. Our data also show that well known regulatory binding sites of PP1, such as the RVXF-binding channel, the β12/β13-loop, and the acidic groove, are not essential for the interaction with Sds22.

Among the protein phosphatases that occur in all studied eukaryotic lineages, the Ser/Thr-specific protein phosphatases of type-1 are the best conserved, with more than 70% of their residues nearly invariant (1). This conservation extends well beyond structurally and catalytically important residues to include exposed residues involved in the binding of regulatory proteins. As a catalytic subunit, PP1 1 depends on the interaction with one or two regulatory subunits for subcellular localization, substrate specificity, and activity regulation (1,2). Eukaryotic cells contain a large variety of regulatory subunits of PP1, which account for the diversified action of this phosphatase. We have recently proposed that PP1 acquired an essential function during early eukaryotic evolu-tion by the development of sites for interaction with a primordial regulatory subunit(s) (1). This essential primordial function and the sequential acquirement of additional interaction sites and functions would then have impeded further mutation of the corresponding portion(s) of the surface. The phylogenetic distribution of PP1 indicates that this primordial function must have been acquired before the divergence of the extant eukaryotic lineages. One of the most ancient functions of PP1 is to dephosphorylate substrates of the Aurora-related protein kinases, and this is essential for the completion of mitosis (3). The regulatory subunit(s) associated with this function of PP1 remain unknown, but the protein Sds22 (38 kDa) has emerged as a prime candidate. First, both yeast and mammalian Sds22 have been shown to interact with PP1 and to be part of a complex with PP1 that is enriched in the nucleus (4 -7). Second, the Sds22 encoding gene was identified independently in fission and in budding yeast as an extra-copy suppressor of the temperaturesensitive mitotic arrest phenotypes that are associated with certain mutations of PP1 (4,5,8). Deletion of the Sds22encoding gene caused a similar mitotic arrest, and this phenotype could be complemented by the overexpression of PP1 (4,5,8). Third, the conditionally lethal phenotype in budding yeast that was conferred by a loss-of-function mutation of the Aurora-related kinase Ipl1 (Ipl1-2), was largely relieved by the expression of certain temperature-sensitive mutant versions of Sds22 or PP1 (9,10). The mutant Sds22 version that rescued the Ipl1-2 phenotype showed a decreased ability to interact with PP1. The expression of this mutant Sds22 did not affect the cellular levels of PP1 or Sds22, but drastically reduced the nuclear level of PP1 and caused a redistribution of the nuclear pool of PP1 (9).
Hitherto, little is known about the mechanism of interaction between Sds22 and PP1. Most known PP1 regulators contain a so-called "RVXF" motif with the consensus sequence [RK]X (0 -1) [VI]X [FW], which binds to a hydrophobic channel of the catalytic subunit (1,11). Sds22 lacks an RVXF motif but consists of a tandem array of leucine-rich repeats (LRRs), which are established protein interaction modules (12). In this paper, we demonstrate that the LRR-repeats of Sds22 are indeed essential for binding to PP1 and we propose that the LRRs assume the conformation of a curved superhelix ending in a C-terminal so-called LRR cap (13). Guided by this three-dimensional model and by the crystal structure of PP1, we have been able to map determinants of the Sds22-PP1 interaction at the concave surface of the Sds22 superhelix and in a triangle composed of ␣-helices 4, 5, and 6 of PP1. * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 1 The abbreviations used are: PP1, protein phosphatase-1; BLAST, basic local alignment search tool; CC, cysteine containing; EGFP, enhanced green fluorescent protein; LRR, leucine-rich repeat; RI, ribonuclease inhibitor; X-gal, 5-bromo-4-chloro-3-indolyl-␤-D-galactoside.

EXPERIMENTAL PROCEDURES
Data Base Searches and Structural Modeling-BLAST searches (14) were launched via the web-interface of the National Center for Biotechnology Information (www.ncbi.nlm.nih.gov/BLAST). Protein-structure files were obtained from the European mirror site of the Protein Data-Bank (pdb.ccdc.cam.ac.uk/pdb/). For structural analysis and the modeling of the backbone of Sds22, Windows versions 3.53 and 3.7 of the DeepView/Swiss-PdbViewer program (15) were used. The side-chains were added and energy-minimized using the dead-end elimination method (16) run on a dual processor Octane 2 work station (Silicon Graphics). The scene represented in Fig. 5A was constructed in Deep-View and rendered with version 3.1 of the program POV-Ray. Fig. 2 was produced using version 1.0 of the ICM lite program (MolSoft).
Plasmid Construction-The full-length and truncated versions of the coding sequence of human Sds22 were PCR-amplified using Pwo polymerase (Roche Molecular Biochemicals) and introduced between the NcoI and BamHI sites of the pACT-II vector (17) at the 3Ј end of a cassette encoding the transcription-activation domain of Gal4. Similarly, the full-length and truncated coding sequences of rabbit PP1␣ and PP1␤/␦, rat PP1␥ 1 , and budding yeast Ppz1 were subcloned in the NdeI and BamHI sites of the pAS-2 vector (18), downstream of a cassette encoding the DNA-binding domain of Gal4. The PP1/Ppz1 chimeras were constructed by consecutive introduction of a fragment from one parent molecule between the SmaI and BamHI sites of pAS-2 and of the complementary fragment from the other parent molecule between the NdeI and SmaI sites.
The QuikChange protocol (Stratagene) was applied for site-directed mutagenesis. Human Sds22-T356A was used as the parent plasmid for the introduction of other point mutations in Sds22 because substitution of alanine for Thr 356 in full-length Sds22 occasionally yielded a moderately enhanced interaction with PP1. PP1␥ 1 -⌬286 -323 was chosen as the parent plasmid for site-directed mutagenesis of PP1 since disruption of the RVXF-binding channel dramatically improved the interaction with Sds22. The construction of pAS-2-based plasmids carrying wild-type and mutant alleles of the budding yeast PP1 gene (GLC7) has been described elsewhere (19). All mutations were verified by DNA sequencing, and the expression of the mutant versions of Sds22 and PP1 was confirmed by Western analysis in crude yeast lysates.
For expression in mammalian cells, wild-type, and mutated coding sequences of Sds22 were subcloned in the BamHI site of a pSG5 vector (Stratagene) with a triple FLAG tag cassette inserted in its EcoRI site. The full-length or truncated coding sequences of rabbit PP1␣ and PP1␤/␦ and of rat PP1␥ 1 were introduced between the XhoI and BamHI sites of pEGFP-C1 (Clontech), downstream of the enhanced green fluorescent protein (EGFP) cassette.
Yeast Two-hybrid Assays-The Y190 reporter strain was transformed with the bait and prey vectors using a lithium acetate transformation protocol adapted as described in Ref. 20. Briefly, a 350-ml culture of the reporter strain was harvested during log-phase growth, washed in 10 mM Tris-HCl and 1 mM EDTA at pH 7.5, and resuspended in 1.5 ml of 10 mM Tris-HCl, 1 mM EDTA, and 100 mM lithium acetate at pH 7.5 to make the cells competent. 100 l of competent cells were then transformed with 100 ng of both the bait and the prey vector and 100 g of denatured salmon-sperm carrier DNA. After the addition of 0.6 ml of 40% polyethylene glycol 4000 and 10 mM Tris-HCl, 1 mM EDTA, and 100 mM lithium acetate at pH 7.5, the cells were incubated at 30°C for 30 min while shaking at 200 rpm. Subsequently, 70 l of dimethyl sulfoxide was added, and the cells were heat-shocked at 42°C for 15 min. After chilling on ice for 2 min, the cells were resuspended in 0.5 ml of 10 mM Tris-HCl and 1 mM EDTA at pH 7.5, plated on selective agar plates containing synthetic dropout medium without leucine and tryptophan, and incubated at 30°C for 5 days.
The ␤-galactosidase activity was evaluated in colony-lift filter assays with 5-bromo-4-chloro-3-indolyl-␤-D-galactoside (X-gal) as a substrate. The colonies were lifted using a Whatman filter, which was subsequently submerged in liquid nitrogen for 10 s to permeabilize the cells. After thawing, the filter was placed on a second filter that had been presoaked in a phosphate-buffered X-gal solution and incubated at 30°C. Blue-coloring was visually assessed after 4, 8, and 24 h.
Antibodies-The 14 C-terminal residues of human Sds22, the 10 C-terminal residues of PP1␤/␦, or recombinant EGFP were coupled to keyhole limpet hemocyanin and injected in rabbits. Polyclonal antibodies were affinity-purified on albumin-coupled peptides that had been linked to CNBr-activated-Sepharose-4B (Amersham Biosciences). In addition, monoclonal mouse antibodies were purchased that specifically recognize the transcription-activation domain of Gal4 (Clontech), the DNA-binding domain of Gal4 (Santa Cruz), or a FLAG tag (Stratagene).

Lysate Preparation, Immunoprecipitation, and Western Analysis-
Yeast cells were harvested from agar plates with the appropriate synthetic dropout medium in 1 ml of 10 mM Tris-HCl and 1 mM EDTA at pH 7.5. After sedimentation of the cells by centrifugation (5 min at 16,000 ϫ g), the cells were lysed by a 10-min incubation on ice with an equal volume of 0.1 M KOH.
COS-1 cells were grown in Dulbecco's modified Eagle's medium with 10% fetal calf serum. FuGENE 6 (Roche Molecular Biochemicals) was used for transfection with mammalian expression vectors encoding FLAG-labeled Sds22 and an EGFP-PP1 fusion protein. 48 h after transfection, the cells were washed twice with ice-cold phosphate buffered saline and lysed in 50 mM Tris/HCl at pH 7.5, 0.5 mM dithiothreitol, 0.5% Triton X-100, 0.3 M NaCl, 0.5 mM phenylmethylsulfonyl fluoride, 0.5 mM benzamidine, and 5 M leupeptin. After centrifugation (5 min at 5000 ϫ g), the FLAG-tagged (mutated versions of) Sds22 in the supernatant were immunoprecipitated with the monoclonal anti-FLAG antibodies and protein-G-Sepharose (Amersham Biosciences). The immunoprecipitates were washed twice in Tris-buffered saline containing 0.1% Igepal CA-630 (Sigma) and twice in pure Tris-buffered saline.
After addition of Laemmli loading buffer to the lysates or immunoprecipitates and boiling, the samples were separated by SDS-PAGE and electroblotted onto polyvinylidene difluoride membranes. Enhanced chemiluminescence (PerkinElmer Life Sciences) was used for the visualization of epitopes after incubation with the appropriate primary and peroxidase-labeled secondary antibodies.

Optimization of the Detection of Interaction between Sds22
and PP1-Because the yeast two-hybrid system has been successful in demonstrating the interaction between Sds22 and PP1 in yeast (5), we decided to use this technique for the mapping of the interaction sites involved. Preliminary experiments showed that human Sds22 (Sds22-WT) interacted rather weakly with full-length yeast or mammalian PP1 isoforms ( Fig.  1). We wondered whether this weak interaction could be accounted for by the sequestration of the PP1 hybrid by endogenous yeast PP1 regulators. Because Sds22 lacks a canonical PP1-binding RVXF sequence, we reasoned that disruption of the RVXF-binding channel of PP1 might improve the interaction with Sds22 by eliminating competition with endogenous RVXF-containing regulators. To test this hypothesis, PP1␥ 1 was truncated just before (PP1␥ 1 -⌬286 -323) and just after (PP1␥ 1 -⌬297-323) the last ␤-strand, ␤14, which contributes half of the residues that line the RVXF-binding channel (11). The bar diagrams representing the tested proteins were drawn to scale and aligned. The hatching patterns refer to the wild-type proteins the fragments are derived from. ϩϩϩ, ϩϩ, and ϩ denote that lifted colonies turned blue within 4, 8, and 24 h, respectively, whereas Ϫ indicates that no bluecoloring was observed after 24 h. If no viable transformants were obtained, this is noted as NT.
As expected, the interaction signal of PP1␥ 1 -⌬297-323 was roughly equal to that of wild-type PP1␥ 1 , whereas PP1␥ 1 -⌬286 -323 interacted much more strongly with Sds22 ( Fig. 1). PP1␥ 1 -⌬286 -323 also failed to interact with the yeast protein Gac1, an RVXF-containing glycogen targeting subunit of PP1 (not shown). Collectively, these observations indicate that it is not the deletion of the flexible C terminus of PP1 but the disruption of the RVXF-binding channel that improves the detection of interaction between Sds22 and PP1. Therefore, PP1␥ 1 -⌬286 -323 was used for the further mapping of PP1binding residues of Sds22 and served as the starting point for further mutational studies of PP1.
Mapping of Sds22-binding Sites of PP1-A first issue to address was whether or not the interaction with Sds22 is isoform-or subtype-specific. The PP1␣-subtype encompasses three of the four mammalian PP1 isoforms, i.e. PP1␣, PP1␥ 1 , and PP1␥ 2 (21). The latter two are products from a single gene and differ only in their C terminus. The fourth isoform, PP1␤/␦, makes up the ␤-subtype. Interestingly, some regulatory subunits are selectively associated with a particular PP1 isoform or subtype, such as the PP1␤/␦-specific myosin-targeting subunits (22) or the ␣-subtype-specific neurabins (23). Thus far, the only mammalian PP1 isoform that has been shown to interact with Sds22 is PP1␥ 2 (24). We found that human Sds22 interacted with PP1␣ and PP1␥ 1 in a two-hybrid assay (Fig. 1). Unexpectedly, after double transformation of the yeast with pAS-2-PP1␤/␦ and either pACT-II-Sds22-WT or empty pACT-II no colonies were obtained. This observation suggests that the constitutive expression of PP1␤/␦ in yeast is lethal, possibly due to the uncontrolled dephosphorylation of critical substrates and/or the sequestration of essential endogenous PP1 regulators. Intriguingly, transformants expressing the same PP1␤/␦ hybrid in conjunction with a Gac1 hybrid were viable, indicating that the interaction with Gac1 neutralized the toxicity of PP1␤/␦. C-terminally truncated versions of all mammalian PP1 isoforms (PP1␣-⌬286 -330, PP1␤/␦-⌬285-327, and PP1␥ 1 -⌬286 -323) gave a similarly strong interaction signal, which was enhanced in comparison with that of the wild type (Fig. 1). Moreover, fusions of full-length PP1␣, PP1␤/␦, and PP1␥ 1 to EGFP all co-sedimented with FLAG-tagged Sds22 (see below), which leads to the conclusion that Sds22 can bind to all mammalian PP1 isoforms. In accordance with previous reports on the activity of the mammalian and yeast PP1 holoenzymes that contain Sds22 (7, 25), Sds22-associated PP1 showed little or no spontaneous phosphatase activity with glycogen phosphorylase as a substrate (not shown).
Using PP1␥ 1 -⌬286 -323 as a starting point, we then set out to evaluate the effect of additional deletions on the ability to bind Sds22. N-terminal deletions proved to be ill tolerated; loss of the first nine amino acids (PP1␥ 1 -⌬1-9/⌬286 -323) already resulted in a drastic reduction of the interaction signal, and amputation of the N-terminal 12 or 40 residues (PP1␥ 1 -⌬1-12/ ⌬286 -323 and PP1␥ 1 -⌬1-40/⌬286 -323) abolished interaction altogether (Fig. 1). This suggests that the N terminus of PP1 is required for the binding of Sds22, either as a site for interaction or for consolidation of the required structural conformation. By contrast, the C-terminal removal of an additional 16 or 37 residues, yielding PP1␥ 1 -⌬270 -323 and PP1␥ 1 -⌬249 -323, respectively, had no appreciable effect (Fig. 1), which rules out the ␤12/␤13-loop (Fig. 2) as an essential Sds22-binding site. The latter loop has been reported to intervene in the binding of other PP1 regulators, such as Inhibitor 2 and NIPP1 (26). PP1␥ 1 -⌬249 -323 moreover lacks an entire flank of the acidic groove (Fig. 2), another interaction site of PP1 that has been proposed to accommodate a basic stretch of the PP1 regulators Inhibitor 1 and DARPP-32 (27). Considering that residues 41-269 of PP1␣ have been shown to constitute the minimal fragment that retains phosphatase activity (28), these results also indicate that catalytic activity is no requirement for the binding of Sds22.
Further C-terminal deletions of PP1, such as in PP1␥ 1 -⌬208 -323 and PP1␥ 1 -⌬174 -323, were inconclusive as fusions of the Gal4 DNA-binding domain, and these PP1-fragments already induced reporter gene activity in the absence of the Sds22 hybrid. However, following their expression as tagged proteins in COS-1 cells, EGFP-fused PP1␥ 1 -⌬208 -323 and PP1␥ 1 -⌬174 -323, like wild-type PP1␥ 1 and PP1␥ 1 -⌬286 -323, co-precipitated from the cell lysates with FLAG-tagged Sds22 (data not shown). To corroborate these data, two chimeric proteins were constructed that combine one-half of PP1␥ 1 with the complementary half of the catalytic domain of Ppz1. The latter is a type-1-like protein phosphatase from budding yeast that comprises an N-terminal regulatory domain and a catalytic domain. The catalytic domain of Ppz1 is 67% identical to the conserved portion of PP1 (Fig. 3). Several regulators of PP1, such as Inhibitor 2 and Inhibitor 3, also interact with Ppz1 (29). However, we could not detect any interaction between the catalytic domain of Ppz1 (Ppz1-⌬1-360) and Sds22 (Fig. 1). The chimeric protein Ppz1-361-528/PP1␥ 1 -174 -323 also failed to bind Sds22, but in agreement with the co-precipitation results its counterpart PP1␥ 1 -1-173/Ppz1-529-692 yielded a strong interaction signal (Fig. 1). A third chimeric protein, consisting of the 39 N-terminal residues of the catalytic domain of Ppz1 fused to residues 43-323 of PP1␥ 1 (Ppz1-361-399/PP1␥ 1 -42-323), interacted equally well as did wild-type PP1␥ 1 (Fig. 1). Collectively, these results suggest that the essential Sds22binding site(s) reside between residues 43 and 173 and, consequently, that the presence of the N terminus of PP1 may be required mainly for conformational reasons. Surprisingly, 79% (104/131) of the residues in the essential fragment of PP1␥ 1 are identical in Ppz1, which is substantially more than the 67% identity overall (Fig. 3). Nevertheless, the 27 differing residues apparently prohibit binding of Sds22 to the catalytic domain of Ppz1. To identify charged residues of PP1 that are involved in the interaction with Sds22, we assayed a panel of 17 mutant alleles of GLC7, the gene encoding yeast PP1 (19), for their ability to interact with Sds22 in the two-hybrid system. The corresponding proteins collectively harbor 37 charged-to-alanine mutations (Table I). In keeping with a reported dominant lethal phenotype (19), no viable transformants were recovered that expressed the glc7-135 allele. As shown in Table I, the degree of interaction with Sds22 varied significantly between alleles. However, three mutants showed no interaction with Sds22. The residues altered in these mutants (Glc7-127, Glc7-129 and Glc7-131) lie between the yeast counterparts of Lys 111 and Asp 153 , in agreement with our tentative assignment of residues 43 through 173 as the putative Sds22 interaction domain. Glc7-128 was able to interact with Sds22. However, the yeast residues altered in this mutant correspond to the largely buried Arg 122 and Glu 126 of PP1␥ 1 . Among the products of glc7-127, glc7-129, and glc7-131, six charged residues have been neutralized, namely the yeast PP1 counterparts of residues Lys 111 , Lys 113 , Asp 138 , Glu 139 , Lys 150 , and Asp 154 of PP1␥ 1 . Two of these, Lys 111 and Glu 139 , are completely or largely buried in the available crystal structures of PP1 (27,31) and can therefore be ruled out as Sds22-binding residues. To study the individual contribution of each of the remaining four residues, Lys 113 , Asp 138 , Lys 150 , and Asp 154 of PP1␥ 1 -⌬286 -323 were separately mutated to alanine (Fig. 3). All detectable binding of Sds22 was lost after mutation of Lys 150 . The latter residue resides in the ␣6-helix, which together with the ␣4and ␣5-helices form a protruding triangle (residues 127-157 of PP1␥ 1 ), wedged between the N terminus and the contiguous catalytic site and hydrophobic groove (Fig. 2). To complement the exploration of the region, we mutated five additional charged and exposed residues within a radius of 15 Å from Lys 150 , namely Arg 36 , Lys 41 , Arg 132 , Lys 141 , and Lys 147 (Fig. 3). Mutation of the latter residue effected a major decrease in binding of Sds22. These results implicate at least part of the ␣4/␣5/␣6-triangle as a major site for interaction with Sds22.
Modeling of the Conserved Portion of Sds22-To guide the mapping of PP1 interaction sites on Sds22, we first conducted a structural analysis of this regulator. The full-length Sds22 sequences of 26 species from all major eukaryotic lineages could be assembled from the results of BLASTP and TBLASTN searches of public sequence databases with the protein sequence of human Sds22 as a query (not shown). The central three quarters of all these Sds22 species is made up of LRRs of typically 22 amino acids, as already reported for yeast and mammalian Sds22 (4 -6, 32). However, the previous delineation of the LRRs in Sds22 had to be reviewed in the light of new insights into the LRR-architecture that were obtained from crystal and NMR structures of LRR-containing proteins (33,34). Fitting of the redefined LRR-profile onto the Sds22 sequence shifts the earlier described 11 LRRs to the C terminus by 8 residues and reveals a twelfth, incomplete C-terminal repeat that ends in a largely conserved C terminus (Fig. 4). The latter represents a so-called LRR-cap motif of 16 -17 amino acids, which also terminates the array of LRRs in many other LRR-containing proteins (13).
Crystal structures of LRR proteins show that each LRR adopts a hairpin-like conformation. The first or consensus fragment of the repeat corresponds to a ␤-strand of the form LXX-LXL, followed by a sharp turn of the form XX[NC]XL, which is stabilized by the hydrogen-bonded Asn or Cys side-chain. The second fragment of the LRRs is much more variable. It generally consists of 9 to 18 residues that form a helicoidal fragment and two loops that connect the helicoidal fragment with the flanking consensus fragments. Conveniently, the sequence and length of the variable C-terminal part of the repeats has been used to discern at least six distinct classes of LRRs (33,34). All the repeats in an array stack as turns of a curved superhelix with a parallel ␤-sheet at the concave side and the helicoidal fragments at the convex side. The side-chains of the hallmark Leu (or Ile) residues make up the hydrophobic core of the protein. In several LRR-proteins, the C-terminal end of the hydrophobic core is shielded from the solvent by an LRR-cap (13). The Asn and Cys side-chains at the [NC] position form a hydrogen-bonded ladder that prevents contact of the buried hydrophobic side-chains with polar main chain moieties.
Two classes of LRRs, the ribonuclease inhibitor-like (RI-like) and the cysteine-containing (CC) repeats, are clearly distinct from the combined remaining classes. RI-like and CC repeats tend to be longer than repeats from other classes and often have a Cys residue at the [NC] position, whereas this position is invariantly occupied by an Asn residue in the other classes. Interestingly, these observations correlate with a more pro-  as detected by yeast two-hybrid assays ϩϩϩ, ϩϩ, and ϩ denote that lifted colonies turned blue within 4, 8, and 24 hours, respectively, whereas Ϫ indicates that no blue coloring was observed after 24 hours. If no viable transformants were obtained, this is noted as NT. Note that the residue numbering is that of yeast PP1, which has to be incremented by 1 to obtain that of mammalian PP1␥ 1 .
nounced curvature of the LRR superhelix of representatives of the RI and CC classes (34). Among the other classes of LRRs, including the Sds22-like class, the superhelix curvature is fairly constant. Therefore, we ruled out the available crystal structures of proteins with RI-like or CC LRRs as candidate templates for the modeling of Sds22. Three of the four crystallized LRR-proteins with repeats of the Sds22-like class, i.e. the spliceosomal protein U2AЈ, Rab geranylgeranyl transferase, and the mRNA export protein TAP, also contain an LRR-cap. The conformation, curvature, and twist of the ␤-sheet and of the turns around the asparagine-ladder, as well as the architecture of the attached LRR-cap is virtually identical in these eukaryotic proteins (35)(36)(37). The prokaryotic fourth protein with Sds22-like LRRs, Internalin B, lacks an LRR-cap and displays a slight right-handed twist of the otherwise similar concave surface of the superhelix (38).
The spliceosomal protein U2AЈ was selected as a template to model the conserved part of the twelve Sds22 repeats and the LRR-cap (Fig. 5A), because it offered the obvious advantage of containing an LRR-cap and because among the three crystallized LRR-cap proteins it is sequence-wise the most similar to Sds22. The backbone coordinates of the LRR-cap and of the consensus fragments of the four C-terminal repeats of U2AЈ were used as a starting point. Then, the model was extended in an iterative process using the consensus fragments of the three central repeats of U2AЈ. The last two of these three fragments were repeatedly fitted onto the two N-terminal fragments of the growing model. Finally, optimal rotamers for the Sds22 side-chains were chosen using the dead-end elimination theorem as implemented in the Brugel program (16), and the energy of the side-chains was minimized using a conjugate gradient algorithm.
The second half of the LRRs is highly variable, both between and within the three eukaryotic LRR-cap proteins that have been crystallized, and none of the in total thirteen available repeats display the Sds22-specific consensus sequence. In the absence of a sufficiently similar template, we have chosen not to incorporate any of the (very similar) hypothetical models of a pair of Sds22 repeats that account for the well conserved hydrophobic and [NG] positions in the variable second half of these repeats (34,39), even though the suggested general interior/exterior orientation of the side-chains appears highly plausible.
The conservation of the consensus residues among Sds22 repeats in general points at a structural role for these residues, which was considered in the modeling of Sds22. On the other hand, residues that are conserved in a particular Sds22 repeat most likely serve a functional purpose. Strikingly, these resi-dues are almost exclusively exposed at the concave surface of the LRR superhelix (Fig. 5B). As described below, we could indeed experimentally confirm that the prime PP1 interaction site of Sds22 corresponds to the region that could be modeled confidently in terms of general architecture and curvature.
Mapping of PP1-binding Residues of Sds22-Using the Cterminally truncated PP1␥ 1 -⌬286 -323 in yeast two-hybrid assays, we first explored the importance of the ill-conserved N terminus of Sds22 in the interaction with PP1. A truncated version of fission yeast Sds22 that lacked the ill-conserved N terminus and the first half of the first LRR, has been reported to co-precipitate with PP1 (6). Using a similarly truncated version of human Sds22 (Sds22-⌬1-88), we confirmed that the conserved C-terminal three-fourths of Sds22, i.e. the LRR-array and the LRR-cap, suffice for the interaction with PP1 (not illustrated). However, the N-terminally truncated Sds22 interacted considerably weaker with PP1 than did wild-type Sds22.
To study the participation of the concave surface of the Sds22 superhelix in the binding to PP1, 36 exposed residues of repeats 2-12 were separately targeted for mutation to alanine, or in the case of Asp 148 to valine, and the effect of these mutations on the interaction with PP1 was evaluated by two-hybrid assays (Fig. 5B). The mutation of Asp 148 , Glu 192 , Phe 214 , Glu 300 , Trp 302 , or Tyr 327 completely abolished the interaction with PP1 while mutation of Phe 170 or Asp 280 severely impaired the binding of PP1 (Fig. 5B). These eight residues are conserved and cluster in two patches. The same eight mutations also impeded  (13), the side-chains of these residues were added. The LRR repeats are numbered, and the arrowhead marks the Nterminal end of the Asn ladder. B, schematic representation of the residues exposed at the concave surface of Sds22 in one-letter codes. The residues are organized in columns that correspond to LRRs and are topped by the repeat number. Residues that are conserved in at least 75% of all known Sds22 species are written in bold, and all non-alanine residues included in the box were individually mutated to alanine. Lifted colonies turned blue within 4, 8, and 24 h after mutation of residues with a white, light gray and dark gray background, respectively, while a black background indicates the absence of blue coloring after 24 h. the interaction of Sds22 with full-length PP1 in the yeast-two hybrid system (not shown).
To verify some key findings of the two-hybrid assays in a mammalian expression system, four mutant versions of Sds22 with a triple N-terminal FLAG tag were co-expressed with an EGFP-PP1␤/␦ fusion protein in COS-1 cells. In complete agreement with the results from the yeast two-hybrid assays, Western analysis revealed co-sedimentation of EGFP-PP1␤/␦ with immunoprecipitated FLAG-tagged wild-type Sds22, Sds22-V172A, and Sds22-S236A, but not with Sds22-E192A or Sds22-W302A (Fig. 6).

DISCUSSION
The present study established that the conserved C-terminal three-fourths of Sds22, which are proposed to form a curved LRR superhelix fused to a C-terminal LRR-cap, suffice for binding to PP1. Nevertheless, the affinity of Sds22 for PP1 was significantly reduced by removal of the N terminus of Sds22. Strikingly, a similar N-terminal amputation of fission yeast Sds22 confers a temperature-sensitive mitotic defect (6), which may also correlate with compromised binding to PP1. Possibly, the N terminus of Sds22 folds into an N-terminal LRR-cap, like the one observed in the structure of Internalin B, a prokaryotic LRR protein of the Sds22-like family (38). Deletion of such a shielding cap may induce a distortion of some of the LRRs to avoid exposure of the hydrophobic core of the LRR-superhelix.
Furthermore, a major bipartite interaction site for PP1, comprising residues Asp 148 , Phe 170 , Glu 192 , Phe 214 , Asp 280 , Glu 300 , Trp 302 , and Tyr 327 of Sds22, was mapped to the concave surface of the superhelix. Indeed, mutation of any of these residues severely or completely compromised the interaction with PP1. The reproducibility of a selection of these yeast two-hybrid results was confirmed in co-immunoprecipitation experiments from mammalian cell lysates. Interestingly, in the four available crystal structures of LRR proteins in complex with macromolecular ligands, the concave side of the LRR-superhelix also functions as a binding site (35,37,40,41).
Conveniently, the concave surface can be quite confidently modeled in terms of general architecture and curvature because the proposed rigid ladder of hydrogen-bonded asparagine residues also restricts the curvature of the attached parallel ␤-sheet. It must, however, be noted that some uncertainty remains in terms of the twist of the sheet. While we have opted for a virtually untwisted sheet as observed in other eukaryotic proteins with Sds22-like LRRs and an LRR-cap, a modest right-handed twist of the sheet cannot be excluded. Such a twist occurs in the proteins Internalin B, which consists largely of Sds22-like LRRs (38), and YopM, which has shorter LRRs of the so-called bacterial class (42). The right-handed twist of these proteins has been tentatively explained by the repulsion of exposed negatively charged side-chains that occupy a fixed position in the second half of neighboring repeats (34). The Glu position in the consensus sequence of the Sds22 repeats may therefore induce a similar twist.
We have also demonstrated that Sds22 can bind all mammalian isoforms of PP1. Given that most of the variance between these isoforms is concentrated in the N-terminal 40 and Cterminal 30 residues, this observation is in line with our conclusion that the N terminus of PP1 does not function as a binding site for Sds22. Nevertheless, removal of the N terminus proved detrimental to the interaction, probably because of a structural distortion of the truncated protein. Further deletion and alanine-scanning studies and experiments with chimeric PP1/Ppz1 proteins suggest that the essential Sds22-binding sites are located between residues 43-173 of PP1␥ 1 , which rules out several classical interaction sites such as the RVXFbinding channel, the ␤12/␤13-loop, and the acidic groove as essential determinants of the interaction. In accordance with our results, it has recently been shown that point mutations in the RVXF-binding channel that impeded binding of RVXFcontaining PP1 regulators did not compromise the interaction with Sds22 (43). In fact, for our mapping we exploited the observation that PP1␥ 1 -⌬286 -323 yielded a stronger interaction signal in yeast two-hybrid assays than did the wild-type. Because no stronger signal was obtained with PP1␥ 1 -⌬297-323, this effect cannot be ascribed to the deletion of the flexible C terminus of PP1 but is indeed caused by the deletion of residues 286 -296 that include the ␤14-strand. Given that RVXF-binding is disrupted by this deletion, it is likely that the binding of Sds22 to this truncated version of PP1 is facilitated by the elimination of interactions with RVXF-containing subunits that are incompatible with Sds22 binding. However, it cannot be ruled out that deletion of residues 286 -296 relieves some sterical hindrance of the Sds22-PP1 interaction. A third explanation, namely that this deletion introduces an artificial binding site for Sds22 is highly unlikely. Indeed, such a site would comprise previously unexposed elements, and the crystal structure of PP1 (11) shows that these elements must have been eliminated in PP1␥ 1 -⌬270 -323 and PP1␥ 1 -⌬249 -323, which yield a similarly enhanced interaction signal. Moreover, the point mutations in Sds22 that compromise the binding to PP1␥ 1 -⌬286 -323 also impede binding to wild-type PP1␥ 1 .
The observation that the C-terminal half of PP1, including all residues that contribute to the RVXF-binding channel, is not required for the interaction with Sds22 raises the interesting possibility that in Sds22-associated PP1 this channel is free for interaction with a specific additional RVXF-containing subunit. For instance, the unidentified phosphoprotein of 25 kDa that co-purifies with Sds22 and PP1 from fission yeast (6), may constitute a third component of the Sds22 complex. Other trimeric PP1 holoenzymes are known that contain an RVXFcontaining and an RVXF-less regulator, such as the CPI-17⅐Mypt1⅐PP1 complex (30). We also confirmed earlier observations that Sds22-associated PP1 largely lacks phosphorylase phosphatase activity (7,25). This, however, does not necessarily imply that Sds22 is a negative regulator of PP1 since the effects of various regulators is well known to be substrate-dependent (2). It can indeed be envisaged that the binding of Sds22 to the ␣4/␣5/␣6-triangle of PP1 restricts the accessibility of the catalytic site to its physiological substrate(s) and thus blocks out other PP1 substrates, such as phosphorylase a.
In summary, we have gained insight into the complex interaction mechanism that governs the binding of Sds22 to PP1 and have charted major binding sites on both interaction partners.