Structural Basis for the Disruption of the Cerebral Cavernous Malformations 2 (CCM2) Interaction with Krev Interaction Trapped 1 (KRIT1) by Disease-associated Mutations*

Background: Mutations in Krev interaction trapped 1 (KRIT1) and cerebral cavernous malformations 2 (CCM2) are associated with CCM disease. Results: The CCM2-KRIT1 interaction is characterized structurally and biochemically. Conclusion: CCM2 preferentially binds the third NPX(Y/F) motif of KRIT1, and disease-associated mutations destabilize this interaction. Significance: These data may inform future studies into the biology of CCM disease. Familial cerebral cavernous malformations (CCMs) are predominantly neurovascular lesions and are associated with mutations within the KRIT1, CCM2, and PDCD10 genes. The protein products of KRIT1 and CCM2 (Krev interaction trapped 1 (KRIT1) and cerebral cavernous malformations 2 (CCM2), respectively) directly interact with each other. Disease-associated mutations in KRIT1 and CCM2 mostly result in loss of their protein products, although rare missense point mutations can also occur. From gene sequencing of patients known or suspected to have one or more CCMs, we discover a series of missense point mutations in KRIT1 and CCM2 that result in missense mutations in the CCM2 and KRIT1 proteins. To place these mutations in the context of the molecular level interactions of CCM2 and KRIT1, we map the interaction of KRIT1 and CCM2 and find that the CCM2 phosphotyrosine binding (PTB) domain displays a preference toward the third of the three KRIT1 NPX(Y/F) motifs. We determine the 2.75 Å co-crystal structure of the CCM2 PTB domain with a peptide corresponding to KRIT1NPX(Y/F)3, revealing a Dab-like PTB fold for CCM2 and its interaction with KRIT1NPX(Y/F)3. We find that several disease-associated missense mutations in CCM2 have the potential to interrupt the KRIT1-CCM2 interaction by destabilizing the CCM2 PTB domain and that a KRIT1 mutation also disrupts this interaction. We therefore provide new insights into the architecture of CCM2 and how the CCM complex is disrupted in CCM disease.

Structurally, CCM2 is predicted to contain an N-terminal phosphotyrosine binding (PTB) domain (8,13) and was also recently discovered to contain a C-terminal harmonin homology domain (29). KRIT1 contains an N-terminal Nudix domain (25), three Asn-Pro-X-Tyr/Phe (NPX(Y/F)) motifs (24), an ankyrin repeat domain, and a C-terminal FERM (band 4.1, ezrin, radixin, moesin) domain (18) (Fig. 1A). Despite what their name implies, the specificity of PTB domains is not restricted to phosphotyrosine motifs, because many of them can also bind to NPXY or NPXF motifs either in place of or in addition to NPXpY (where pY represents phosphotyrosine). The modes of interaction between the PTB domain and Asn-Pro-X-Tyr/Phe/Tyr(P) (NPX(Y/F/pY)) motifs are similar (30). Previous studies have suggested that the putative CCM2 PTB domain is important for the CCM2 association with KRIT1 (31), and a point mutation in CCM2 PTB domain (F217A) based on canonical interactions between PTB domains and NPX(Y/F/pY) motifs reduces its binding to KRIT1 (32). The interaction site within KRIT1, however, is controversial. Previous studies have suggested either no involvement of the conserved KRIT1 NPX(Y/F) motifs (33) or that the CCM2 PTB domain binds equally to both the second and third of these motifs (34).
To investigate the impact of rare missense point mutations in the KRIT1 and CCM2 genes and to further elucidate the structural basis for recruitment of CCM2 to KRIT1, we took a twopronged approach. We conducted clinical sequencing of patients suspected of having one or more CCM lesions to identify novel missense point mutations and used biochemical and structural studies to probe the interaction between CCM2 and KRIT1. In our clinical sequencing of the three CCM genes in hundreds of affected patients, we identified small numbers of missense mutations amid the much larger group of chain termination mutations. Because many of the mutations we found were within the predicted PTB domain of CCM2, we then went on to investigate the structure of the CCM2 PTB domain and the basis of its interaction with KRIT1. We began by confirming that CCM2 and KRIT1 can interact with one another in cells and that mutation of both the second and third NPX(Y/F) motifs of KRIT1 (KRIT1 NPX(Y/F)2 and KRIT1 NPX(Y/F)3 ) reduces this interaction. We found that the CCM2 PTB domain exhibits preferential binding in vitro to KRIT1 NPX(Y/F)3 over either KRIT1 NPX(Y/F)1 or KRIT1 NPX(Y/F)2 , indicating that this is the dominant region of KRIT1 that mediates its interaction with CCM2. We then determined the 2.75 Å co-crystal structure of the CCM2 PTB domain in complex with KRIT1 NPX(Y/F)3 and validated the crystallographically observed interaction by pulldown. To explore the role of the potential disease-related missense mutations that we had identified within the CCM2 PTB domain and KRIT1 NPX(Y/F)3 , we tested the impact of these mutations on CCM2, KRIT1, and their interaction. We find that the missense mutations identified in the region of CCM2 that encodes its PTB domain significantly reduce the solubility of the CCM2 protein product and that a missense mutation in KRIT1 that results in a V244L mutation reduces the ability of KRIT1 to interact with CCM2. These results provide a significantly improved understanding of the CCM2-KRIT1 interaction and its disruption in CCM disease.

EXPERIMENTAL PROCEDURES
Sequencing-Nomenclature for sequence variants was taken from the Human Genome Variation Society recommendations. As required, DNA was extracted from the patient specimen using a 5 Prime ArchivePure DNA blood kit. PCR was used to amplify the indicated exons plus additional flanking intronic or other non-coding sequence. After cleaning of the PCR products, cycle sequencing was carried out using the ABI Big Dye Terminator version 3.0 kit. Products were resolved by electrophoresis on an ABI 3730xl capillary sequencer. Sequencing was performed separately in both the forward and reverse directions.
Protein Expression and Purification-Human CCM2 (Uni-Prot Q9BSQ5) cDNA corresponding to amino acid residues 51-228 was subcloned into a modified pET-32 vector with a tobacco etch virus protease-cleavable N-terminal hexahistidine (His 6 ) tag and transformed into Escherichia coli Rosetta (DE3) cells (Novagen). Overnight cultures of these cells were inoculated into Luria Broth medium until A 600 ϭ 0.6. Protein expression was induced by the addition of 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside at 18°C overnight. The cells were harvested the next day and were resuspended in lysis buffer (500 mM NaCl, 20 mM imidazole, 20 mM Tris, pH 8.0) supplemented with 1 mM DTT, protease inhibitors, and DNase I. Cell lysis was performed by three cycles of freeze/thaw in a dry ice/ ethanol bath followed by sonication. Following clarification, the supernatant was loaded onto a HisTrap HP column (GE Healthcare). His 6 -CCM2 PTB was eluted by lysis buffer supplemented with 500 mM imidazole and dialyzed overnight against 250 mM NaCl, 20 mM Tris, pH 8.0, 1 mM DTT in the presence of tobacco etch virus to remove the His 6 tag. Cleaved CCM2 PTB was loaded to a Superdex 75 column (GE Healthcare) and eluted as a monodisperse peak in buffer containing 250 mM NaCl, 20 mM Tris, 1 mM DTT, pH 8.0.
Point mutations were introduced using the QuikChange Lightning mutagenesis kit (Stratagene). The GST-KRIT1 NPX(Y/F)3 mutants were treated in the same way as the wild-type protein.
For the His 6 -CCM2 PTB mutants, an additional desalting step was added following affinity purification into a final buffer of 150 mM NaCl, 20 mM Tris, pH 8.0, and 0.5 mM tris-(2-carboxyethyl)phosphine.
Size Exclusion Chromatography with Multiangle Light Scattering-Purified CCM2 PTB at a concentration of 1 mg/ml was mixed with a synthesized and HPLC-purified 13-mer peptide corresponding to KRIT1 NPX(Y/F)3 (VDKVVINPYFGLG; KRIT1 NPX(Y/F)3 ) (Tufts University Core Facility) in a 1:1 molar ratio. After incubation on ice for 2 h, the sample was filtered, and 100 l was injected onto an SRT-300 column (Sepax Technologies) at 0.4 ml/min in a buffer composed of 150 mM NaCl, 20 mM Tris, pH 8.0, 1 mM tris(2-carboxyethyl)phosphine, and 0.02% sodium azide. The column was run on an HPLC system (Agilent Technologies 1260) in-line with a DAWN-HELEOS II multiangle light scattering detector (Wyatt Technology) and Optilab T-rEX differential refractometer (Wyatt Technology). Data were analyzed using the ASTRA 6 software from Wyatt Technology.
Crystallization, Data Collection, and Structure Determination-Purified CCM2 PTB was concentrated to 10 mg/ml and mixed with the synthetic KRIT1 NPX(Y/F)3 peptide in a 1:1 molar ratio. Following incubation on ice for 2 h, sparse matrix and grid crystallization screening was conducted using the sitting drop method and the Classics, pHClear, PEGs, and JCSG ϩ screening kits (Qiagen) in 1-l drops with a 1:1 ratio of protein and precipitant solutions. Crystals grew against 20% (w/v) PEG 6000 and 0.1 M HEPES, pH 7.0, in the pH Clear Suite (Qiagen). For cryoprotection, crystals were transferred to the precipitant condition supplemented with 10% ethylene glycol prior to flash cooling in liquid nitrogen.
Diffraction data were collected at Northeastern Collaborative Access Team beamline 24-ID-E at the Advanced Photon Source and were processed using HKL-2000 (36) to 2.75 Å resolution. An initial structure solution was obtained by molecular replacement using Phaser (37) with an ensemble search model of previously determined PTB domains (Protein Data Bank entries 4JIF, 3SO6, 3G9W, 3HQC, 3DXE, 2EJ8, 2V76, 1J0W, 1WVH, 2CY4, 2DYQ, 1QQG, 1X11, and 1AQC) edited to remove ligands, water molecules, and non-PTB peptide chains. Four copies of CCM2 PTB were found per asymmetric unit, and autobuilding was performed using PHENIX (37), which built 531 residues in 14 chains, including residues of KRIT1 NPX(Y/F)3 . Autobuilding yielded R and R free values of 26.3 and 28.9% respectively. Iterative rounds of refinement and model building were conducted using PHENIX (37), REFMAC5 (38) using jelly body refinement, and Coot (39). TLS and NCS were used. The KRIT1 peptide was clearly visible in the electron density, allowing unambiguous determination of its register. The structure displayed good geometry and was validated using MolProbity (40). Crystallographic programs were supported by the SBGrid Consortium (41). The atomic coordinates and structure factors are deposited in the Protein Data Bank with accession code 4WJ7.
Pull-down Experiments-Pull-down experiments were conducted by incubating 67.5 g of CCM2 PTB with 20 g of GST-KRIT1-loaded glutathione-Sepharose beads for 2 h at 4°C with agitation. Samples were washed three times with 1 ml of pulldown buffer (150 mM NaCl, 50 mM Tris, pH 7.5, 0.1% Triton) and visualized by SDS-PAGE. Quantification was performed using ImageJ.
Solubility Tests-CCM2 disease mutants (L113P, L115R, L155P, L198R, and L213P) were introduced into His 6 -CCM2 PTB . All of these mutants as well as the wild-type protein were transformed into BL21 cells. Overnight cultures of the cells were inoculated into 100 ml of Luria broth, and protein expression was induced with isopropyl 1-thio-␤-D-galactopyranoside when A 600 ϭ 0.6. Cells were grown overnight at 18°C, harvested, and resuspended in 1 ml of a buffer composed of 500 mM NaCl and 20 mM Tris, pH 8.0, supplemented with DTT, protease inhibitors, lysozyme, and DNase I. The resuspended cells were lysed by three cycles of freeze/thaw in a dry ice/ ethanol bath followed by sonication. After sonication, 10 l of each lysate was spun down at 13,000 rpm. The pellets and supernatants were each resuspended to a total volume of 50 l. One l of each sample was run on an SDS-PAGE gel and visualized by Coomassie staining.
Biolayer Interferometry-The biolayer interferometry technique using the BLItz system (ForteBio) was used to measure binding kinetics for the CCM2 PTB interaction with GST-KRIT1 NPX(Y/F)1 , GST-KRIT1 NPX(Y/F)2 , and GST-KRIT1 NPX(Y/F)3 and the following biotinylated peptides: Biotin-225 ADTCIYNPLFGSD 237 (KRIT1 biotin-NPX(Y/F)2 ), Biotin-241 TNRVDKVVINPYFG 254 (KRIT1 biotin-NPX(Y/F)3 ), and Biotin-225 ADTCIYNPLFGSDLQYTNRVDKVVINPYFG 254 (KRIT1 biotin-NPX(Y/F)2-3 ). Anti-GST biosensors (for the GST fusions) or streptavidin biosensors (for the biotinylated peptides) were hydrated in binding buffer (150 mM NaCl, 50 mM Tris, pH 7.5) for 10 min. For each CCM2 PTB concentration (ranging from 2.5 to 415 M), the following procedure was performed. An initial baseline was collected by immersing the biosensor in binding buffer for 1 min, and then 4 l of 50 g/ml KRIT1 was loaded to the biosensor for 5 min. The KRIT1loaded biosensor was returned to binding buffer for collection of a second baseline for 90 s and then placed in 4 l of His 6 -CCM2 PTB for a 5-min association step. Finally, the biosensor was returned to binding buffer for a final 2.5-min dissociation step. For each data point, the background binding was also measured. For each CCM2 PTB concentration, the difference in the signal (in nm) just prior to the association step and that at the end of the association step was subtracted from the difference in signal for background binding. These values were plotted, and the curves were fit using GraphPad Prism to obtain a dissociation constant.

Discovery of Missense Point Mutations in KRIT1 and
CCM2-From 2004 to 2012 at PreventionGenetics, a clinical DNA testing laboratory, 507 full KRIT1 gene sequencing tests and 288 full CCM2 gene sequencing tests were performed on patients suspected by their physicians of having one or more CCMs. Likely causative mutations were identified in 30% of the KRIT1 tests and 12% of the CCM2 tests. The great majority of causative mutations in both genes were chain termination mutations (nonsense, frameshift, or splicing). However, a small number of missense mutations were also identified in these patients (Table 1 and Fig. 1A). Note the preponderance of Leu substitutions in CCM2. None of the missense mutations listed in Table 1 were reported in normal controls except for 1 of 13,000 alleles for KRIT1 R16C. In addition, none of these missense mutations were observed at PreventionGenetics in more than one patient, except for CCM2 L155P, which was observed in several affected members of a large Australian kindred and which we therefore classify as likely pathogenic. The CCM2 L113P mutation was reported in a patient by Riant et al. (12), but the clinical significance of this mutation was listed as unknown. Similarly, the KRIT1 L238F mutation was previously reported in one family (42). Finally, the pathogenic mechanism of the KRIT1 D281G mutation was reported to be through the introduction of a new, alternative splice donor site (14). The clinical significance of the remaining missense mutations listed in Table 1 remains uncertain.
Two of the KRIT1 mutations are in regions outside of the known functional domains of the KRIT1 protein and therefore may not have a structural or functional impact; KRIT1 residue Leu 238 lies between KRIT1 NPX(Y/F)2 and KRIT1 NPX(Y/F)3 C-terminal to the KRIT1 NPX(Y/F)2 binding site for SNX17 (35). Residue Asp 281 lies in a predicted flexible region between NPX(Y/ F)3 and the N terminus of the KRIT1 ankyrin repeat domain (Fig. 1A). Mutations of R16C and G101R are within surfaceexposed regions of the Nudix domain that are not known to be important for KRIT1 function (Fig. 1B). The Q473H mutation is within the F1 lobe of KRIT1 and is one of the residues used in its interaction with HEG1, so this mutation could cause disruption of the HEG1 interaction. The L667R mutation may result in disruption of the hydrophobic core of the F3 lobe of the KRIT1 FERM domain, potentially decreasing the affinity of protein-protein interactions or reducing KRIT1 solubility (Fig.  1C). The V244L mutation in KRIT1 is close to its third NPX(Y/F) motif (Fig. 1A), and it is therefore unclear whether this mutation would affect KRIT1 associations with its binding partners.
Interestingly, the majority of the CCM2 missense mutations fall within its predicted PTB domain (Fig. 1A). Although the predicted CCM2 PTB domain is believed to be critical for the interaction of CCM2 with KRIT1, it has not been structurally characterized; nor has its interaction with KRIT1 been fully mapped. We therefore decided to undertake further studies to investigate the molecular level basis of the CCM2 interaction with KRIT1 to better understand these disease-associated mutations.
CCM2 Interacts with the NPX(Y/F) Region of KRIT1-CCM2 and KRIT1 have been shown to interact with one another by both co-immunoprecipitation (co-IP) (33) and yeast two-hy-

CCM2-KRIT1 Co-crystal Structure
brid screen (34). Although studies have pointed toward the possible importance of the KRIT1 NPX(Y/F) motifs in interacting with the CCM2 PTB domain (33), a consensus has not been reached on which if any of these motifs are needed to bind to CCM2. To probe the basis for the CCM2-KRIT1 interaction, we began by conducting co-immunoprecipitation experiments for full-length CCM2 and KRIT1. In agreement with previous reports, we observe a robust interaction between the two wildtype proteins (Fig. 2B). We generated mutations in each of the three NPX(Y/F) motifs within full-length KRIT1 Fig. 2A), but we did not observe an appreciable effect for any of these mutations individually. It was not surprising that mutation of KRIT1 NPX(Y/F)1 did not reduce the interaction with CCM2, because this motif has previously been shown to be used for the KRIT1 association with ICAP1 (25), and competition between ICAP1 and CCM2 for binding to KRIT1 has not been observed previously (33). When we introduced a double mutation in both the second and third motifs (KRIT1 F234A/Y252A/F253A ), however, we did observe a reduction in co-IP (Fig. 2B), supporting previous data (34) and suggesting that the NPX(Y/F) motifs of KRIT1 are involved in its interaction with the CCM2 PTB domain. It was, however, unclear from this overexpressed co-IP system whether CCM2 shows a preference for KRIT1 NPX(Y/F)2 , for KRIT1 NPX(Y/F)3 , or if it binds similarly to both.
CCM2 Exhibits Binding Preference for KRIT1 NPX(Y/F)3 Motif in Vitro-We therefore decided to explore whether purified CCM2 PTB domain exhibits a binding preference to any of the KRIT1 NPX(Y/F) motifs in an in vitro system with purified proteins by conducting pull-down experiments. We expressed and purified the PTB domain of CCM2 encoding residues 51-228 (CCM2 PTB ) and designed GST fusion constructs corresponding to each of the three NPX(Y/F) motifs in KRIT1: GST-KRIT1 NPX(Y/F)1 , GST-KRIT1 NPX(Y/F)2 , and GST-KRIT1 NPX(Y/F)3 ( Fig. 2A). We expressed and purified the GST-KRIT1 NPX(Y/F) motif proteins and bound them to glutathione-Sepharose beads. We then tested the ability of purified CCM2 PTB to interact with each of the KRIT1 NPX(Y/F) motifs. We found that CCM2 PTB directly interacts with KRIT1 NPX(Y/F)3 and that this interaction is stronger than with either KRIT1 NPX(Y/F)1 or KRIT1 NPX(Y/F)2 (Fig. 2, C and D). These are the first data to suggest that CCM2 exhibits a preference among the KRIT1 NPX(Y/F) motifs.
Structural Basis for the CCM2 Interaction with KRIT1-The structure of the predicted CCM2 PTB domain, as well as its mode of binding to KRIT1, was previously unknown. Based on our pull-down experiments, we therefore set out to determine the structure of CCM2 PTB in complex with a 13-residue peptide corresponding to KRIT1 NPX(Y/F)3 ( 244 VDKVVINPYFGLG 256 ). We analyzed this complex by size exclusion chromatography with multiangle light scattering and found that it elutes as a monodisperse peak with a molecular mass of 20.8 kDa Ϯ 1.1%, consistent with a 1:1 CCM2 PTB -KRIT1 NPX(Y/F)3 complex with an expected molecular mass of 20.8 kDa (Fig. 2E). We conducted co-crystallization trials and obtained a 2.75 Å resolution (I/I is 1.2 at 2.75 Å and 2.75 at 2.96 Å resolution) data set that allowed structure determination by the molecular replacement method. Four copies of the CCM2 PTB -KRIT1 NPX(Y/F)3 complex are observed per asymmetric unit (Tables 2 and 3). The peptide chains for CCM2 PTB and the KRIT1 NPX(Y/F)3 peptide were first built using automated procedures and subsequently refined by manual model building. Because the register of each of the chains was clear, this allowed for complete assignment of built residues (Fig. 3B). We find that CCM2 PTB adopts a Dab-like PH/PTB fold that interacts with KRIT1 NPX(Y/F)3 in 1:1 stoichiometry (Fig. 3A). As predicted by structure-based sequence alignment (43), CCM2 PTB has the highest structural homology to ICAP1, a protein that binds both integrin ␤1 cytoplasmic tail and KRIT1 NPX(Y/F)1 (25, 26) (root mean square deviation 2.5 Å over 131 residues and a z score of 17.7 calculated by the Dali server (44)). Based on our crystal structure, CCM2 PTB comprises residues Ser 60 -Tyr 221 with a 32-residue flexible loop (loop ␤6/␤7) inserted between strands ␤6 and ␤7 (residues Asp 160 -Glu 191 ), for which we do not observe density in any of the four copies of CCM2 PTB (Fig. 3A). Like other PH/PTB fold domains, CCM2 PTB comprises a C-terminal ␣-helix and seven ␤-strands arranged in two anti-parallel ␤-sheets with strand ␤1 a constituent of both ␤-sheets (␤1-␤7-␤6-␤5 and ␤1-␤2-␤3-␤4). Similar to other Dab-like PTB domains, CCM2 PTB contains an additional ␣-helix, ␣1, inserted between strands ␤1 and ␤2, and a 3 10 helix, which we term helix ␣1Ј, connecting strands ␤4 and ␤5. We do not observe a basic patch at the PH fold phosphoinositide headgroup binding site located between loops ␤1/␤2, ␤3/␤4, and ␤6/␤7 (30) or the Dab-like headgroup binding site (30). Our structure therefore places the PTB domain of CCM2 into the Dab-like family of NPX(Y/F/pY) motif-binding PTB domains and suggests that it does not interact with phosphoinositide headgroups. Both the surface of CCM2 that interacts with KRIT1 and the sequence of KRIT1 NPX(Y/F)3 itself are well conserved over evolution (Fig. 3,  C and D). Interestingly, there is reduced sequence conservation in the 32-residue flexible ␤6/␤7 loop of CCM2, and serine residues (Ser 164 , Ser 166 , and Ser 168 ) in this loop have previously been found to be phosphorylated (45).
We clearly observe binding of a KRIT1 NPX(Y/F)3 peptide to each of the four CCM2 PTB molecules in the asymmetric unit, with KRIT1 residues Val 244 -Leu 254 built in each copy (Fig. 4A). The mode of binding we observe is similar to other structures of canonical PTB domains in complex with NPX(Y/F) motifs (30). A ␤-strand is formed by KRIT1 residues Lys 246 -Ile 249 that hydrogen-bonds to ␤5 of CCM2 PTB , extending the ␤1-␤7-␤6-␤5 anti-parallel ␤-sheet. This KRIT1 ␤-strand also stacks against helix ␣2 of CCM2 PTB (Fig. 4B). At the C terminus of KRIT1 NPX(Y/F)3 , the NPYF motif forms a type III turn and packs against helix ␣1Ј. Multiple hydrophobic and hydrogen-bonding contacts are made between KRIT1 and CCM2. Of note, KRIT1   (Fig. 4, B and C). The interface between CCM2 PTB and KRIT1 NPX(Y/F)3 comprises eight or nine hydrogen bonds for each of the four copies and buries an average of 1359 Å 2 (between 576 and 637 Å 2 for the four copies of CCM2 PTB and between 700 and 802 Å 2 for the four copies of KRIT1 NPX(Y/F)3 ) as calculated by PDBsum (46). Structure-guided Mutagenesis to Disrupt CCM2 PTB -KRIT1 NPX(Y/F)3 Interaction-We conducted biolayer interferometry to measure the affinity of the interaction between CCM2 PTB and GST-KRIT1 NPX(Y/F)3 and observed a K D of 13.0 Ϯ 2.5 M (Fig.  5A). This affinity value is similar to measurements previously observed for PTB domain interactions with NPX(Y/F) motifs (47)(48)(49). We were not able to observe binding for CCM2 PTB to GST-KRIT1 NPX(Y/F)1 or GST-KRIT1 NPX(Y/F)2 ( Fig. 5A and Table 4). Longer GST fusions encompassing both KRIT1 NPX(Y/F)2 and NPX(Y/F)3 were severely proteolyzed upon expression in E. coli, so to compare the affinity of CCM2 PTB for the entire NPX(Y/F)2-3 region of KRIT1 with the individual NPX(Y/F)2 and NPX(Y/F)3 motifs, we synthesized biotinylated KRIT1 peptides encompassing NPX(Y/F)2, NPX(Y/F)3, or NPX(Y/ F)2 and NPX(Y/F)3 (KRIT1 Biotin-NPX(Y/F)2 , KRIT1 Biotin-NPX(Y/F)3 , and KRIT1 Biotin-NPX(Y/F)2-3 , respectively). We again used the biolayer interferometry method to measure CCM2 PTB affinities for these biotinylated synthesized KRIT1 peptides. We found that although we could not detect a measurable interaction between CCM2 PTB and KRIT1 Biotin-NPX(Y/F)2 , the affinities of CCM2 PTB for binding to either KRIT1 Biotin-NPX(Y/F)3 or KRIT1 Biotin-NPX(Y/F)2-3 were very similar (15.5 Ϯ 4.0 and 14.5 Ϯ 2.9 M, respectively) and also similar to that measured for the GST-KRIT1 NPX(Y/F)3 protein expressed in E. coli (Table 4). To validate the structurally defined interaction between CCM2 PTB and KRIT1 NPX(Y/F)3 , we used a pull-down assay where GSTfused KRIT1 peptide was bound to glutathione-Sepharose beads and purified CCM2 PTB was added. Based on the crystal structure, we then introduced point mutations that we pre-dicted would disrupt the CCM2 PTB -KRIT1 NPX(Y/F)3 interaction (indicated with blue labels in Fig. 4B). In KRIT1, we introduced a mutation of V248D to interrupt the hydrophobic interaction with CCM2 helix ␣2. We also mutated the NPYF motif to NPAA (Y252A/F253A) as we had done for the full-length protein because mutation of the NPX(Y/F) motifs is often used to interrupt their interactions with PTB domains. These mutations significantly impaired the ability of CCM2 PTB to pull down with GST-KRIT1 NPX(Y/F)3 (Fig. 5, B and C). In CCM2, we introduced three sets of mutations: C210D, I136A/I139A, and F217A. Cys 210 is located within CCM2 helix ␣2 and oriented toward the KRIT1 binding site, so we hypothesized that its mutation would interrupt the CCM2-KRIT1 interaction. Ile 136 and Ile 139 are located at the N and C termini of the 3 10 helix ␣1Ј that interacts with KRIT1 NPX(Y/F)3 , so we expected that this double alanine mutation would destabilize the helix and consequent binding to KRIT1. Phe 217 is at the C terminus of helix ␣2 stacking against KRIT1 residue Asn 250 , and its mutation has previously been shown to reduce the ability of CCM2 to bind KRIT1 (32). Each of these mutations significantly impaired the ability of CCM2 PTB to pull down with GST-KRIT1 NPX(Y/F)3 (Fig. 5, D and E). Introduction of the crystallographically defined mutations C210D and I136A/I139A into full-length CCM2 also interrupted the interaction by co-IP (Fig. 5F). Therefore, because a single point mutation in the PTB domain of CCM2 disrupts binding to KRIT1, we conclude that CCM2 contains a single binding site for KRIT1.
Missense Point Mutations Destabilize the CCM2 PTB Domain-Having shown through biochemical and biophysical assays that KRIT1 NPX(Y/F)3 is the predominant site of interaction and having determined the structure of the CCM2 PTB domain in complex with the KRIT1 NPX(Y/F)3 peptide, we then interactions with CCM2. Stick format is used for KRIT1 and CCM2 residues discussed in this work. Residues labeled in blue are mutated in this study, and the underlined label indicates CCM disease-associated mutation. C, schematic map of interactions between KRIT1 and CCM2. Dashed red lines, hydrogen bonds; blue lines, non-bonded contacts. Secondary structure elements are labeled. Blue labels, residues mutated in this study; underlined label, CCM disease-associated mutation. All interactions shown are observed in at least three of the four copies in the asymmetric unit, with the exception of KRIT1 residue Pro 251 , which makes contacts with CCM2 in only one copy.
asked where the CCM2 disease-associated mutations identified here and elsewhere map to the structure of the PTB domain and whether they affect KRIT1 binding. In addition to the L113P, L115R, L155P, and L213P point mutations identified through our sequencing efforts, other studies have independently found a L198R (13,33) as well as additional L113P (12) mutations in CCM2 in patient populations. The L198R mutation has been shown to reduce the interaction with KRIT1 (33), but the molecular basis for this is unknown. The effect of the L113P mutation has not yet been analyzed. To better understand the impact of missense mutations in CCM2 and KRIT1, we mapped CCM2 mutations L113P, L115R, L155P, L198R, and L213P and KRIT1 mutation V244L onto our co-crystal structure. Each of these CCM2 leucine residues is located within the hydrophobic core of CCM2 PTB , where they interact with one another to form a tight hydrophobic cluster (Fig. 6A). Based on the structural mapping, we postulated that introduction of the CCM2 mutations could negatively affect protein stability. We therefore generated L113P, L115R, L155P, L198R, and L213P point mutations in our His 6 -CCM2 PTB construct and tested their solubility. We found that introduction of each of these mutations resulted in significantly reduced solubility upon overexpression in E. coli (Fig. 6B), suggesting a negative impact on the structural integrity of CCM2 PTB . In contrast, KRIT1 residue Val 244 packs tightly into a pocket formed by the CCM2 PTB domain N terminus of helix ␣1, Ser 83 , and strand ␤5 (Fig. 6A). To test the impact of the KRIT1 mutant V244L, we introduced the mutation into our GST-KRIT1 NPX(Y/F)3 construct and conducted pull-down assays with CCM2 PTB . We found that the KRIT1 V244L mutation causes a significant reduction in the ability of KRIT1 NPX(Y/F)3 to pull down CCM2 (Fig. 6, C and D), and we were not able to detect binding for this interaction by biolayer interferometry (data not shown). These results suggest that these disease-associated missense mutations in CCM2 or KRIT1 either disrupt protein solubility and stability or destabilize the binding between CCM2 and KRIT1.

DISCUSSION
The disease cerebral cavernous malformations is closely linked to mutations in the KRIT1, CCM2, and PDCD10/CCM3 genes and their resulting proteins KRIT1, CCM2, and CCM3. We investigated the impact of germ line mutations in patients who harbor multiple CCM lesions and discovered missense mutations in both KRIT1 and CCM2, with many of the CCM2 missense mutations clustering within its PTB domain. Because  the molecular basis for how CCM2 and KRIT1 interact was not known and was the subject of controversy, we mapped the interaction, determined its structural basis, and validated it using in vitro assays. We then conducted further analysis of the CCM2 and KRIT1 missense mutations in the context of these new data, finding that many of these disease-associated mutations have a negative impact on folding of the CCM2 PTB domain or affect its associated with KRIT1. Our study therefore provides a significant improvement in our understanding of the molecular level impact of CCM-associated mutations.
In the in vitro system we used to investigate the interactions of the purified CCM2 PTB domain with each of the individual KRIT1 NPX(Y/F) motifs, we see a clear preference of CCM2 for binding to KRIT1 NPX(Y/F)3 and cannot detect binding of the CCM2 PTB domain to either isolated KRIT1 NPX(Y/F)1 or KRIT1 NPX(Y/F)2 (Fig. 2, C and D). The affinity measured for CCM2 PTB interaction with KRIT1 NPX(Y/F)3 is within the range previously observed for PTB interactions with NPX(Y/F) motifs (47)(48)(49) and is also similar to the affinity we measured with KRIT1 NPX(Y/F)2-3 . Additionally, our structure-directed point mutations within CCM2 that interrupt the interaction with KRIT1 show that CCM2 utilizes a single binding site within its PTB domain to interact with KRIT1. We therefore conclude that the predominant interaction between KRIT1 and CCM2 is by the CCM2 PTB domain binding to KRIT1 NPX(Y/F)3 via a canonical PTB-NPX(Y/F) motif binding site. In the overexpression system we use in Fig. 2B, however, we observe that to interrupt KRIT1 interaction with CCM2 by co-immunoprecipitation, there seems to be a requirement to simultaneously mutate both the KRIT1 NPX(Y/F)2 and KRIT1 NPX(Y/F)3 motifs. This is in accordance with previous yeast two-hybrid studies suggesting that CCM2 has no clear preference among the KRIT1 NPX(Y/F) motifs (34) and could be an artifact of this overexpression system. Although our biochemical studies do not suggest a role for the KRIT1 NPX(Y/F)2 in this interaction, they do not rule out the possibility of weaker secondary KRIT1 interaction sites for CCM2, so further studies to explore the role of a potential secondary interaction of CCM2 with KRIT1 could be merited.
Our structural analysis provides the first atomic resolution picture of the N-terminal PTB domain of CCM2, confirming previous predictions that CCM2 contains a PH/PTB domain fold in the Dab-like subfamily. Among other PTB domains, the CCM2 PTB domain has highest structural similarity to that of ICAP1 (25). In contrast to previous predictions, we do not observe canonical or non-canonical phosphoinositide binding sites on the CCM2 PTB structure (43), a finding that may impact ongoing research into potential membrane recruitment of CCM2. This structure additionally shows that the CCM2 PTB domain interacts with the third NPX(Y/F) motif of KRIT1 via a canonical PTB-NPX(Y/F) interaction. It is also possible that the CCM2 PTB domain interacts with other CCM2 binding partners, such as SMURF1 (28). If other proteins can bind at this site using a similar mode of binding, it will also be interesting to determine whether there is competition among CCM2 binding partners.
Interestingly, a number of the patient-derived mutations identified in our large scale sequencing effort result in missense mutation of leucine amino acids within the hydrophobic ore of the CCM2 PTB domain to either proline or arginine. Our solubility tests show that these mutations destabilize the CCM2 PTB domain, potentially implying that these mutations act as pseudononsense mutations. In contrast, the large scale sequencing of KRIT1 revealed a missense mutation, V244L, within the CCM2-binding region N-terminal of KRIT1 NPX(Y/F)3 , a location that had not previously been discovered to be mutated in CCM disease. We find that Val 244 is important for the interaction of KRIT1 and CCM2, and even a conservative mutation to leucine results in weakening the interaction between CCM2 and KRIT1.
Taken together, our results provide a molecular level understanding of how the CCM complex is formed around the CCM2-KRIT1 interaction. The results also suggest that proper folding of CCM2 is lost for some of the missense mutations found in CCM disease patients. Therefore, we anticipate that our data will allow for an improved understanding of the cellular function of the CCM2-KRIT1 interaction and other CCM2 PTB domain interactions, providing the foundation for future studies into the mechanisms by which mutations in KRIT1 and CCM2 might lead to CCM disease.