Structural Basis for Small G Protein Effector Interaction of Ras-related Protein 1 (Rap1) and Adaptor Protein Krev Interaction Trapped 1 (KRIT1)

Background: Rap1 GTPase binds the KRIT1 FERM domain resulting in relocalization. Results: The KRIT1-Rap1 co-crystal structure is determined to 1.95 Å resolution. Conclusion: Rap1 binds KRIT1 using an extended surface that encompasses both Switch I and Switch II loops in Rap1 and F1 and F2 lobes of KRIT1 FERM domain. Significance: We provide a structural framework to understand Rap1-mediated relocalization of KRIT1. Cerebral cavernous malformations (CCMs) affect 0.1–0.5% of the population resulting in leaky vasculature and severe neurological defects. KRIT1 (Krev interaction trapped-1) mutations associate with ∼40% of familial CCMs. KRIT1 is an effector of Ras-related protein 1 (Rap1) GTPase. Rap1 relocalizes KRIT1 from microtubules to cell membranes to impact integrin activation, potentially important for CCM pathology. We report the 1.95 Å co-crystal structure of KRIT1 FERM domain in complex with Rap1. Rap1-KRIT1 interaction encompasses an extended surface, including Rap1 Switch I and II and KRIT1 FERM F1 and F2 lobes. Rap1 binds KRIT1-F1 lobe using a GTPase-ubiquitin-like fold interaction but binds KRIT1-F2 lobe by a novel interaction. Point mutagenesis confirms the interaction. High similarity between KRIT1-F2/F3 and talin is revealed. Additionally, the mechanism for FERM domains acting as GTPase effectors is suggested. Finally, structure-based alignment of each lobe suggests classification of FERM domains as ERM-like and TMFK-like (talin-myosin-FAK-KRIT-like) and that FERM lobes resemble domain “modules.”

tive GDP-loaded form. This cycling is facilitated by guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs) (18). GEF-facilitated cycling to the GTPloaded form results in conformational changes in small GTPases, specifically in the Switch I and Switch II loops, that allow high affinity binding to downstream effector molecules, including Rap1, usually important for activation of signaling pathways. GAP-facilitated GTPase activation results in GTP hydrolysis, and this leads to conformational changes that cause dissociation from effector molecules and turn off downstream signaling. Rap1 is a key player in the signal transduction networks critical for formation and regulation of cell-cell junctions (19) and in control of ␤1, ␤2, and ␤3 integrin affinity, avidity, and signaling (20). As one of the roles of Rap1 is to trigger integrin activation (21), the model where activated Rap1 triggers KRIT1 translocation from microtubules to the membrane, where KRIT1 can release ICAP1-mediated repression of integrin activation, is attractive.
In this paper we provide the first molecular level description of the direct interaction between KRIT1 and Rap1. We show that Rap1 binds KRIT1 using both its Switch I and Switch II loops and that the surface Rap1 recognizes on KRIT1 encompasses both the F1 and F2 lobes of the FERM domain. We additionally show that the interaction interface in KRIT1 is completely conserved over evolution. Thus, we provide the first structural description of KRIT1. We also provide a template for FERM domains acting as GTPase effectors and the first example of intermolecular recognition by a FERM domain using both F1 and F2 lobes. Finally, we provide structure-based cla-dograms for the FERM domain lobes, which suggest the FERM lobes resemble protein domain "modules."

EXPERIMENTAL PROCEDURES
Expression and Purification of Rap1 and KRIT1-Human KRIT1 FERM domain (420 -736) was subcloned into pGEX6p-1 expression vector and then overexpressed in BL21(DE3) Escherichia coli. Protein expression was induced by 0.2 mM isopropyl 1-thio-␤-D-galactopyranoside at 16°C overnight. Cells were then harvested at 4°C and lysed in lysis buffer (20 mM Tris, 500 mM NaCl, 20 mM imidazole, 10% glycerol, pH 8.0) by freeze/thaw using dry ice with ethanol and warm water followed by three cycles of sonication on ice. The supernatant was then loaded onto glutathione-Sepharose 4B beads (GE Healthcare) and cleaved with PreScission protease overnight at 4°C on beads. Eluted proteins were then loaded onto a Resource-S ion-exchange chromatography column (GE Healthcare) with 20 mM Hepes, pH 7.0, and eluted over a 5-500 mM NaCl gradient. The fractions were then further purified by size-exclusion chromatography on a Superdex 200 column (GE Healthcare) using 20 mM Hepes, pH 7.0, buffer. Human Rap1B cDNA was purchased online from Thermo Scientific Open Biosystems. Rap1B(1-167) was subcloned into a modified pET-32 vector with an N-terminal His 6 tag and a tobacco etch virus cleavage site. Mutation of Gly-12 to valine was conducted using the QuikChange Lightning site-directed mutagenesis kit (Stratagene). Recombinant Rap1-G12V protein was overexpressed in E. coli Rosetta (Novagen) host cells by induction with 0.2 mM isopropyl 1-thio-␤-D-galactopyranoside at 16°C, A 600 0.6. Cell pellets were resuspended in lysis buffer and then lysed by freeze/thaw followed by three cycles of sonication on ice. The supernatant was applied to HisTrap HP chelating affinity column (GE Healthcare) initially and then eluted with 200 mM imidazole in lysis buffer. The fusion protein was cleaved by tobacco etch virus protease at 4°C overnight dialyzing against dialysis buffer (20 mM Tris, 500 mM NaCl, 10% glycerol, pH 8.0). The sample was then reloaded onto over a HisTrap HP column to remove the uncleaved fusion proteins. The elution fractions were then concentrated and loaded onto a size-exclusion chromatography Superdex 75 column with 20 mM Hepes, pH 7.0, buffer. Proteins were then concentrated using Millipore concentrators. Both KRIT1 and Rap1B proteins were tested by SDS-polyacrylamide gels and found to be higher than 95% purity.
Co-crystallization of Rap1-KRIT1-The FERM domain of KRIT1 was concentrated to 15 mg/ml in 200 mM NaCl, 20 mM Hepes, pH 7.0, 2 mM DTT. G12V mutant Rap1 was concentrated to 10 mg/ml in 150 mM NaCl, 20 mM Hepes, pH 7.0, 1 mM DTT. The two proteins were then mixed by carefully evaluating the molar ratio as 1:1. The protein mixture was then incubated on ice for more than 2 h with 1 mM GTP␥S and 5 mM MgCl 2 . Initial co-crystallization screens were performed using the JCSGϩ sparse matrix crystallization kit (Qiagen). We observed initial hits in conditions containing 0.2 mM potassium nitrate, 20% PEG3350. Optimization was then performed by carefully screening the concentration of these two chemicals in the crystallization drop. The optimized KRIT1-Rap1 co-crystals were grown using hanging-drop methodology at room temperature  (drop size, 2 l). Crystals grew after 3 days in the crystallization conditions of 0.1-0.2 mM KNO 3 , 15-20% PEG3350. Crystals were flash frozen in liquid nitrogen using the same crystallization conditions as cryoprotectant. Crystallographic data were collected at beamline 24-ID-C (NE-CAT) at Argonne National Laboratory, Advanced Photon Source. All data were processed using HKL2000 (22).
Structure Determination-The KRIT1-Rap1 complex cocrystallized into space group P2 1 with unit cell parameters of a ϭ 57.5 Å, b ϭ 78.0 Å, c ϭ 59.8 Å, ␣ ϭ ␥ ϭ 90°, ␤ ϭ 90.9°. We obtained initial phases by conducting molecular replacement using a crystal structure of the GTP-loaded form of Rap1 (Protein Data Bank (PDB) ID 1C1Y) (23) and the FERM domain of radixin (PDB ID 1GC6) (24). The Rap1 model was stripped of ligands, the radixin model was reduced to a polyalanine/serine/ glycine trace for each of the subdomains, using the program Chainsaw (25), and GTPase and each subdomain (F1, F2, and F3) were used as molecular replacement search models. Phaser (26) yielded translation Z-scores of 17.1 for Rap1 and 6.5 and 5.9 for two of the subdomains (F2 and F3) of radixin. The solution was then input into ARP/wARP (27), and following 200 cycles 254 residues were built with 128 in sequence and R and R free factors of 31.4 and 54.4%, respectively. This model was then input into the Phenix phase_and_build module (28) for two cycles, and 250 residues were built with 218 placed in sequence and R and R free factors of 41.7 and 44.6%, respectively. This model was then input into ARP/wARP using parameters identical to those of the previous round (200 cycles, add atoms above 3.1 , remove atoms below 0.9 , density modification at the start of autobuilding). This time ARP/wARP built a total of 473 residues in sequence and yielded R and R free factors of 24.8 and 30.9%, respectively. Standard refinement was then conducted in Refmac (29) and Phenix with TLS (28). Model building was conducted in COOT (30). The final model was validated by MolProbity (31). The final structure has been deposited in the Protein Data Bank under accession code 4DXA.
KRIT1-Rap1 Pulldown Assays-GST-KRIT1 and His-Rap1 were expressed in E. coli as described above. Glutathione-Sepharose 4B (GE Healthcare) resin was equilibrated in PBS with 1% Triton X-100 (PBST). 1.5 ml GST-KRIT1 lysate was added to 120 l of equilibrated resin in a 2-ml microcentrifuge tube and then incubated at 4°C with shaking for 1 h. The resin bound with fusion protein was washed three times with 1 ml of PBST. 10 g of pulldown resin was then incubated with 20 g of purified protein in 400 l of buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Triton X-100, 5 mM MgCl 2 , 1 mM GTP␥S tetralithium salt) at 4°C with shaking for 1.5 h. After incubation, the resin was washed three times with 1 ml of buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Triton X-100). Following washing, 30 l of 1ϫ SDS sample buffer was added to the resin, and samples were heated to 95°C for 5 min. The samples were then briefly centrifuged and loaded on a 15% SDS-polyacrylamide gel. The expression, purification, and pulldown protocols were identical for all KRIT1 and Rap1 mutant proteins. Mutations were generated using the QuikChange site-directed mutagenesis kit (Agilent Technologies).

Overall Structure
To determine the co-crystal structure of Rap1 GTPase and KRIT1 FERM domain, we expressed and purified Rap1B-G12V (residues 1-167) and KRIT1 (residues 420 -736) in the E. coli expression system as N-terminally tagged His 6 and GST fusion proteins, respectively. Following proteolytic removal of the affinity tags and separate purification, the two proteins were mixed and co-crystallized together. We determined the 1.95 Å resolution co-crystal structure of KRIT1 in complex with Rap1 ( Fig. 1C and Table 1).

Structure of Rap1
In the co-crystal structure of KRIT1 FERM domain in complex with Rap1B-G12V, the structure of Rap1 is highly similar

Co-crystal Structure of Rap1 in Complex with KRIT1
to previously determined active-state Rap1 structures, an expected result because the G12V mutation in Rap1 is a mutation that results in an activated GTPase. The structure superposes with an r.m.s.d. between 0.7 and 0.9 Å over 167 aligned residues (36) with the active-state Rap1 structures of Rap1 in complex with RAF-RBD (PDB IDs 1C1Y, 1GUA, and 3KUC) (23,37,38) and Rap1 in complex with RapGAP (PDB ID 3BRW) (39). Good electron density is observed throughout the structure, and the GTP␥S moiety binds in the manner expected for this nonhydrolyzable nucleotide. The regions of Rap1 with highest B-factors are the Switch I and Switch II loops. This illustrates the conformational flexibility inherent in these important functional parts of the protein.

Interaction of Rap1 and KRIT1
The KRIT1 FERM domain can interact with the KRIT1 N terminus (10,43), the Heart-of-Glass adhesion receptor (44), and Rap1 (9,16,17). The KRIT1-Rap1 interaction described here is the first of these to be investigated crystallographically. Broadly speaking, this interaction utilizes the whole Rap1 Switch region (Switches I and II) to interact with sites on both the F1 and F2 lobes of KRIT1. The interaction is in the plane of the FERM domain trefoil, thus creating a diamond shape, or four-leafed clover. A total of 1750 Å 2 (849 Å 2 in KRIT1 and 901 A 2 in Rap1) of surface area is buried, with ϳ1100 Å 2 (ϳ63%) being contributed by the Rap1-F1 interaction and ϳ650 Å 2 (37%) being contributed by the Rap1-F2 interaction ( Fig. 2A) (as defined by the PISA server) (45). The shape complementarity of the KRIT1-Rap1 interaction is 0.66 (46). We suggest that the intricate and extended nature of the Rap1-KRIT1 interaction may be the reason for the reported unusual specificity of KRIT1 for Rap1 but not for Ras (9,47). Interestingly, analysis by the PISA server finds no examples of similar GTPase domain interactions with acyl-CoA-binding protein (F2)-like fold proteins. As described below, GTPase interactions with ubiquitin (F1)-like fold proteins are frequently observed and can be compared with the KRIT1-Rap1 complex.
In detail, the interaction of Rap1 with KRIT1 encompasses 11 hydrogen-bonding interactions and 90 nonbonded interactions (PDBsum) (48). These interactions cluster into two sites that we term the "Switch" site and the "Distal" site (Fig. 2B). The Switch site buries a greater surface area and encompasses both Rap1 Switch I and Switch II, and regions from both KRIT1 F1 and F2 lobes. The Distal site is smaller and is distal from the GTP binding site in Rap1, and it encompasses residues from the Rap1 ␤2-turn-␤3 hairpin and KRIT1 F2 (Fig. 2C). We mapped sequence conservation of KRIT1 over 29 species onto the surface of KRIT1 FERM domain using the CONSURF server (49). It is striking that the surface of KRIT1, which interacts with Rap1, is almost completely conserved through evolution (Figs.  2D and 3). Furthermore, this complete surface conservation is contiguous between both the Switch and Distal sites and extends past these regions.
The Switch site can be broadly subdivided into three interconnecting regions that encompass Switch I-F2 interactions, the ␤-sheet, and Switch II-F1 interactions (Fig. 4, A and B).
Switch I-F2-At the N terminus of the Rap1 Switch I loop a broadly hydrophobic interaction is found centered around Ile-27 RAP which stacks against KRIT1 F2 residues Pro-525 KRIT and Leu-526 KRIT . This surface is extended to the aliphatic region of Gln-25 RAP which interacts with Leu-529 KRIT . Gln-25 Rap also hydrogen bonds to Arg-452 KRIT , one of two KRIT1 arginine residues critical for the Rap1-KRIT1 complex.
␤-Sheet Region-The second critical arginine residue, Arg-432 KRIT , hydrogen bonds to the backbone carbonyl of Switch I residue Pro-34 RAP . It also takes part in an intricate hydrogenbonding network between Arg-432 KRIT , Arg-452 KRIT , Asp-38 RAP , Tyr-40 RAP , and multiple water molecules (Fig. 4A). C-terminal to Pro-34 RAP in the Rap1 Switch I loop, water-mediated hydrogen bonding is observed for Thr-35 RAP and Ile-36 RAP . The Rap1 ␤2 strand begins at Glu-37 RAP , and antiparallel ␤-sheet hydrogen bonding is observed between the Rap1 ␤2 strand and the KRIT1 ␤2 strand, mediated by Glu-37 RAP and Ser-39 KRIT in Rap1, and Ser-433 KRIT and Tyr-431 KRIT in KRIT1. This allows a continuous ␤-sheet to form between Rap1 and KRIT1 F1 domain, encompassing all ␤-strands in both domains. The side chain of Glu-37 RAP also hydrogen bonds to the side chains of Tyr-431 KRIT and Arg-423 KRIT .
Switch II-F1-For the Rap1 Switch II loop, hydrophobic interactions are observed between Met-67 RAP and Lys-421 KRIT , and Phe-64 RAP and Phe-419 KRIT . We note that although Phe-419 KRIT is vector-derived, it probably packs in a fashion similar to the wild-type KRIT1 residue, which is a tyrosine (Fig. 4A).
Distal Site-The Distal site is centered at the F2 lobe residue Tyr-563 KRIT . Tyr-563 KRIT hydrogen bonds to Gln-43 RAP and stacks against Gln-50 RAP . These residues are at the C terminus of the ␤2 strand and the N terminus of the ␤3 strand in Rap1. An intricate water-mediated hydrogen bonding network is formed between these residues, Glu-45 RAP and the C terminus of KRIT1 helix ␣8 (Fig. 4A).

KRIT1-Rap1 Pulldown Assays
To validate the crystallographically observed interaction, we generated point mutations in KRIT1 and Rap1 and conducted pulldown assays. We generated KRIT1 mutations S430E, R432E, and R452E, which we expected to compromise the interaction with Rap1, and Rap1 mutations Q25A and D38A, which we expected to compromise the interaction with KRIT1 (Fig. 4B). We also generated mutations outside of the binding interface, which we expected not to alter KRIT1-Rap1 binding but which are crystal-packing interactions in the structure (R501E in KRIT1 and Q137R in Rap1). We found that the KRIT1 mutations S430E, R432E, and R452E did indeed compromise the interaction with Rap1 (Fig. 4C) and that the Rap1 mutations Q25A and D38A compromised the interaction with KRIT1 (Fig. 4D). The control mutations did not affect the interaction by pulldown. These pulldown assays therefore confirm the interaction surface between KRIT1 and Rap1 and confirm the previous finding that R452E mutation in KRIT1 compromises KRIT1-Rap1 interaction by reducing the binding K d from 1.8 M 67 M (9).

Comparison of Rap1-KRIT1 Complex with Other GTPase Complex Structures
The Rap1 GTPase domain interacts with the F1 and F2 lobes of KRIT1. These lobes fold in the manner of ubiquitin-like (F1) and acyl-CoA-binding protein (F2) domains. Because ubiquitin-like fold proteins are frequently shown to interact with GTPases, we investigated whether the mode of binding between the KRIT1 F1 lobe and Rap1 was similar to that observed previously for ubiquitin-like fold GTPase-binding proteins. Indeed, as was first observed for Raf in complex with Rap1 (23), the ubiquitin fold ␤ sheet of KRIT1 F1 extends through the GTPase Switch I loop in an antiparallel fashion (Fig. 5). The F1-Rap1 interaction is the larger surface, but shows lower shape complementarity (0.64) than the F2-Rap1 interaction (0.69) (46). We do not find example crystal structures in the

Co-crystal Structure of Rap1 in Complex with KRIT1
protein data bank of F2 (acyl-CoA-binding protein)-like fold proteins interacting with GTPases, therefore we believe that the Rap1-F2 interaction defines a new GTPase interaction surface. Because KRIT1 displays unusual specificity for Rap1 over Ras (9,47) we propose that the mode of interaction between Rap1 and KRIT1 represents a multidentate mechanism for GTPases to obtain high effector specificity.

Comparison of KRIT1 FERM Subdomains
As discussed above, the KRIT1 FERM domain is most similar to myosin-X; however, to investigate whether the subdomains of KRIT1 are similar to other FERM domain lobes we submitted F1, F2, and F3 individually to the Dali server. Similar to the whole KRIT1 FERM domain, the F1 lobe has highest structural Cavia porcellus (guinea pig), XP_003496996 Cricetulus griseus (hamster), XP_002196126 Taeniopygia guttata (zebra finch), NP_001026144 Gallus gallus (chicken), XP_003452511 Oreochromis niloticus (Nile tilapia). Alignment was conducted using ClustalW (34). Figure was generated using Aline (60). Interacting residues were defined using the PISA server (45). Center panel shows KRIT1 on the left side (purple) and Rap1 on the right. Rap1 is colored green, with the Switch I loop colored orange and the Switch II region colored blue. Expanded views show interactions of residues discussed in the text for the Distal site and the three interconnecting regions of the Switch site (labeled Switch II-F1Ј, Switch I-F2Ј, and ␤-sheet). B, schematic illustrating interactions of KRIT1 and Rap1. H-bonds are shown with blue lines and contacts with orange dotted lines. Rap1 residues are flanked by KRIT1 F2 residues on the left and KRIT1 F1 residues on the right. Residues in Rap1 Switch I and the following ␤2 strand are shaded green, and those in Rap1 Switch II and the preceding ␤3 strand shaded blue. Residues mutated in this study are shown in red. C, Pulldown assay for KRIT1 mutations. Pulldown assays performed with GST-KRIT1-coated beads and purified Rap1. Pulldowns were conducted using wild-type GST-KRIT1 (WT) and GST-KRIT1 mutated to include the following mutations: S430E, Ser-430 points toward a hydrophobic region so mutation to glutamic acid is predicted to disrupt the interaction; R432E, a charge reversal predicted to introduce charge repulsion; R452E, predicted to impact hydrogen bonding with Gln-25 RAP and introduce charge repulsion; and R501E, a residue involved in crystal contacts and not predicted to impact Rap1 binding. KRIT1 mutations S430E, R432E, and R452E all impact Rap1 pulldown, but as expected, KRIT1 mutation R501E does not. GST-alone pulldown control and input lanes are shown on the right. D, pulldown assay for Rap1 mutations. Pulldown assays performed with GST-KRIT1 (K) or GST-alone (G) coated beads and purified Rap1. Pulldowns were conducted with Rap1 wild-type (WT) and mutations Q137R, a residue involved in crystal contacts and not predicted to impact KRIT1 binding; D38A, predicted to impact a hydrogen bonding network involving Arg-432 KRIT ; and Q25A, predicted to impact hydrogen bonding involving Arg-452 KRIT . Both D38A and Q25A Rap1 mutations compromise binding to KRIT1, but as expected, Q137R does not. Input lanes are shown on the right. . Surprisingly, however, both the F2 and F3 lobes show highest similarity to talin F2 and F3 lobes. For the KRIT1 F2 lobe, the most similar structure in the PDB is talin-1 (PDB ID 1MIX) with a Z-score of 15.3, an r.m.s.d. of 2.0 Å over 109 aligned residues, and a sequence identity of 25%. For the KRIT1 F3 lobe, the 7 most similar structures in the PDB are of talin-1 and talin-2, with Z-scores ranging between 13.4 and 14.1, r.m.s.d. values ranging between 1.3 Å and 1.7 Å over 83 or 84 aligned residues, and sequence identities between 14 and 15%. We also noted that KRIT1 is well conserved in the phosphotyrosine binding-like binding site of F3 (Fig. 2D) and that on superposition of the KRIT1 F3 lobe to the F3 lobes of talin-1 and talin-2, there is reasonable structural similarity (Fig. 6). As the talin F3 lobe binds the short cytoplasmic tails of the ␤ subunits of integrin adhesion receptors in diverse conformations (50,51), and KRIT1 is implicated in modulation of integrin function (11), we tested whether KRIT1 could also bind ␤ integrin cytoplasmic tails. We conducted pulldown assays using purified recombinant KRIT1 FERM, well characterized recombinant His-tagged integrin cytoplasmic tail constructs and previously described protocols (32,33). However, despite the structural similarities between KRIT1 and talin, we could not detect a specific interaction between KRIT1 and integrin ␤1 or ␤3 (data not shown). Similar results were obtained using GST-tagged ␤1 tails. When considered in light of the amino acid conservation of the KRIT1 F3 subdomain (data not shown), this suggests that binding of another, nonintegrin, ligand at the canonical phosphotyrosine binding domain ligand binding site may play important roles in KRIT1 activity. This could potentially be the N-terminal NPXY motif region, previously shown to bind KRIT1 FERM (10,43). We therefore tested whether KRIT1 FERM could bind the KRIT1 N terminus, and as expected we observed a direct interaction (data not shown).

New Mode of FERM Domain Intermolecular Recognition Defined by KRIT1 FERM Domain F1 and F2 Lobes
FERM domain intermolecular recognition and targeting have predominantly been shown to be mediated by the F3 lobe, as illustrated for radixin (Fig. 6A). In contrast, intramolecular mechanisms of regulation for FERM domains so far seem to be mediated by the F1 and F2 lobes, as illustrated for FAK (52) and moesin (53) (Fig. 6A). The KRIT1-Rap1 complex represents the first description of intermolecular recognition by a FERM domain using F1 and F2 lobes. Because Rap1 has previously been shown to release KRIT1 from microtubules (10), the finding that KRIT1 is extremely conserved on the contiguous surfaces of the F1 and F2 lobes (Fig. 2D) potentially makes sense from an evolutionary standpoint; it is harder for the sequence of a protein to wander through evolution if there is a requirement to bind multiple partners. We therefore propose that the F1 and F2 surface is utilized by multiple binding partners, including Rap1, and that Rap1 directly competes with a binding partner that localizes KRIT1 to the microtubule machinery.
FERM domain interactions with GTPases may be more common than previously thought. Recent work has shown that several sorting nexin endosomal trafficking proteins (SNX17, SNX27, and SNX31) contain FERM domains. These FERM domains can bind activated Ras in vitro and may be involved in endosomal Ras signaling processes (54). Furthermore, the atypical FERM domain of talin has also been shown to bind GTPases by its F0 lobe, a ubiquitin-like fold domain similar to the F1 lobe (55). We therefore propose that the KRIT1-Rap1 interaction represents a template to understand FERM domain containing protein-GTPase interactions. It is interesting to note that the KRIT1-Rap1 interaction surface significantly overlaps with the FAK kinase-FERM autoinhibitory interaction surface (Fig. 6A) (52). Demonstration of this mode of FERM-GTPase domain interaction raises the possibility that GTPases may play undiscovered functionally important roles in FDCP conformational rearrangements, which could impact regulation of the FERM domain-containing enzymes. It will therefore be interesting to discover whether FERM domain-containing kinases, (including the JAKs, FAK, and PYK2) and FERM domain-containing phosphatases (including PTPN3, PTPN4, and PTPN13) can bind and be regulated by GTPases.

Considerations for FERM Domain Classification
Primary sequence alignment analysis has classified FDCPs into three subgroups named for their prominent members: kindlin and talin; ERM proteins, GEFs, kinases, and phosphatases; and myosin and KRIT1 proteins (56). Based on our structure-based Dali analysis that shows highest similarity to myosin-X, the overall structure of KRIT1 would support the previous classification. In contrast, however, our structurebased analysis of the F2 and F3 sublobes of KRIT1 supports the notion that KRIT1 could potentially also be included in the talin and kindlin FDCP group or may represent a crossover FDCP with similarities to both talin/kindlin and to myosin. To investigate the classification of FERM domains further, we therefore constructed structure-based cladograms using each of the F1, F2, and F3 lobes for examples of all FERM domains that exist in the PDB (Fig. 6B). The cladograms suggest two groups of FERM domain in the current structural library, which we term ERMlike and TMFK-like (talin-myosin-FAK-KRIT1-like). The pairwise similarity of the TMFK-like FERMs varies per lobe. For example, talin clusters with FAK in the F1 cladogram, but with myosin-VII when analyzing F2 and F3. Although these results are somewhat different from the previous classification based only on primary sequence for whole FERM domains (56), together both studies raise the interesting prospect that FERM domain lobes resemble domain modules within the FDCP family of proteins. Structure-based classification of the FERM domain lobes may be particularly useful as the functional similarities and differences of FERM domains are dissected.

CONCLUSIONS
The loss of KRIT1 is associated with acquisition of CCMs and is incompatible with life (7). The functions of this protein are, however, only now becoming understood. One of these functions is to provide a binding partner and sink for ICAP1; KRIT1 sequesters this regulator of integrin activation away from integrin ␤1 tails, thus allowing talin-and kindlin-mediated integrin activation (57). Clearly, the localization of KRIT1 to the membrane must therefore be under tight regulation to maintain proper balances of integrin activation. The small GTPase Rap1 has been suggested to be the primary player in relocalizing KRIT1 from microtubules to the membrane (9). Interestingly, Rap1 also interacts with RIAM (Rap1-GTP-interacting adaptor molecule). As one of the roles of RIAM is to attract talin to integrin tails to facilitate integrin activation (58), an elegant system is now revealing itself. In this model, Rap1 facilitates KRIT1 relocalization to the membrane where it releases an integrin suppressor, ICAP1, from binding integrin cytoplasmic tails, concomitantly Rap1 binds RIAM which recruits talin to integrin. Thus, Rap1 plays a key role in regulating proper spa-tial-temporal regulation of ␤1 integrin activation, and to do this it requires its binding partner, KRIT1. The mechanisms revealed here for Rap1 and KRIT1 interaction are therefore an important step toward understanding molecular basis for this elegant system for integrin regulation.