Distinctive phosphoinositide- and Ca2+-binding properties of normal and cognitive performance–linked variant forms of KIBRA C2 domain

Kidney- and brain-expressed protein (KIBRA), a multifunctional scaffold protein with around 20 known binding partners, is involved in memory and cognition, organ size control via the Hippo pathway, cell polarity, and membrane trafficking. KIBRA includes tandem N-terminal WW domains, a C2 domain, and motifs for binding atypical PKC and PDZ domains. A naturally occurring human KIBRA variant involving residue changes at positions 734 (Met-to-Ile) and 735 (Ser-to-Ala) within the C2 domain affects cognitive performance. We have elucidated 3D structures and calcium- and phosphoinositide-binding properties of human KIBRA C2 domain. Both WT and variant C2 adopt a canonical type I topology C2 domain fold. Neither Ca2+ nor any other metal ion was bound to WT or variant KIBRA C2 in crystal structures, and Ca2+ titration produced no significant reproducible changes in NMR spectra. NMR and X-ray diffraction data indicate that KIBRA C2 binds phosphoinositides via an atypical site involving β-strands 5, 2, 1, and 8. Molecular dynamics simulations indicate that KIBRA C2 interacts with membranes via primary and secondary sites on the same domain face as the experimentally identified phosphoinositide-binding site. Our results indicate that KIBRA C2 domain association with membranes is calcium-independent and involves distinctive C2 domain–membrane relative orientations.

tional scaffold protein with around 20 known binding partners, is involved in memory and cognition, organ size control via the Hippo pathway, cell polarity, and membrane trafficking. KIBRA includes tandem N-terminal WW domains, a C2 domain, and motifs for binding atypical PKC and PDZ domains. A naturally occurring human KIBRA variant involving residue changes at positions 734 (Met-to-Ile) and 735 (Ser-to-Ala) within the C2 domain affects cognitive performance. We have elucidated 3D structures and calcium-and phosphoinositide-binding properties of human KIBRA C2 domain. Both WT and variant C2 adopt a canonical type I topology C2 domain fold. Neither Ca 2؉ nor any other metal ion was bound to WT or variant KIBRA C2 in crystal structures, and Ca 2؉ titration produced no significant reproducible changes in NMR spectra. NMR and X-ray diffraction data indicate that KIBRA C2 binds phosphoinositides via an atypical site involving ␤-strands 5, 2, 1, and 8. Molecular dynamics simulations indicate that KIBRA C2 interacts with membranes via primary and secondary sites on the same domain face as the experimentally identified phosphoinositide-binding site. Our results indicate that KIBRA C2 domain association with membranes is calcium-independent and involves distinctive C2 domain-membrane relative orientations.
KIBRA is linked to memory, cognition, and neurological disorders in humans and in rodent models (for a review, see Ref. 2). A single nucleotide polymorphism (SNP), rs17070145, in the ninth intron of KIBRA, for example, has been implicated in human cognition (15), a finding corroborated by numerous subsequent studies, including a meta-analysis (16). rs17070145 is associated with Alzheimer's disease (AD) (3,17). KIBRA, moreover, has additive and epistatic interactions with APOE (18), the ⑀4 allele of which is the strongest genetic risk factor for sporadic AD. The effect of SNP rs17070145 on memory and AD may arise due to differential activation of the MAPK pathway (19), important for memory and learning processes. In human, rat, and mouse brains, KIBRA is mainly expressed in memoryrelated regions, including hippocampus and cortex, as well as cerebellum and hypothalamus (15,20,21).
KIBRA acts in the same pathway as protein kinase M (PKM), a brain-specific kinase thought to be involved in long-term memory storage through its control of neuroreceptor trafficking, particularly trafficking of ␣-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors, the major excitatory neurotransmitter receptors in the brain (22). KIBRA binds, colocalizes with, and is phosphorylated by PKM (11,23), and KIBRA is involved in AMPA receptor trafficking (24). KIBRA counteracts proteasomal degradation of PKM (25). In addition, KIBRA binds to dendrin (26) and synaptopodin (13,27), postsynaptic cytoskeleton organizers important for synaptic transmission and cognition. Aberrant acetylation of tau, linked to cognitive deterioration in dementia, disrupts postsynaptic function by reducing postsynaptic KIBRA (4,28). Consistent with these accumulated observations, KIBRA overexpression and knockdown exert positive and negative effects, respectively, on key molecular and cellular processes in neurons (29). It has been shown by immunoprecipitation that KIBRA can form homodimers (20) and heterodimers with the two other WWC protein family members, WWC2 and WWC3 (10), although the functional significance of this remains unknown.
Here, we characterize the calcium-and phosphoinositidebinding properties of KIBRA C2 domain (Fig. 1), including a variant that arises from two SNPs (rs3822660G/T and rs3822659T/G) in human KIBRA that result in substitution of two adjacent residues (M734I,S735A) (30). Although the molecular consequences of these exonic SNPs have not been fully elucidated, they affect cognitive performance, and there is almost complete linkage disequilibrium between the aforementioned intronic SNP rs17070145 (15) and rs3822660G/T and rs3822659T/G (30).
The C2 domain is found in more than 125 different human proteins (31,32). Although most C2 domains are Ca 2ϩ -dependent membrane association domains, some C2 domains do not bind Ca 2ϩ , and some mediate protein-protein rather than protein-lipid interactions (31,32). Our experimental and computational investigations of WT and variant KIBRA C2 domains indicate that KIBRA C2 is a noncanonical C2 domain that in vitro both binds phosphoinositides and associates with membranes in an atypical, calcium-independent manner.

WT and variant KIBRA C2 adopt a typical C2 fold
Our 2.6-Å resolution crystal structure (Protein Data Bank code 6FD0) shows that variant C2 (varC2) adopts a typical eight-stranded ␤-sandwich C2 domain fold. The crystal form comprises a parallel dimer (i.e. the two monomers are parallel to each other) formed via an intermonomer disulfide bond involving Cys-771, the fifth and only unpaired cysteine in KIBRA C2 ( Fig. 2A). The previously available crystal structure of WT KIBRA C2 (Protein Data Bank code 2Z0U) involves a similar parallel, Cys-771 disulfide-linked dimer. According to analytical ultracentrifugation (AUC) measurements (Fig. S1  and Table S2), WTC2 and varC2 samples both comprise a mixture of monomer and dimer in solution.
The backbone 1 H, 13 C, and 15 N NMR chemical shifts of C2(C771A) and varC2(C771A) match closely the corresponding chemical shifts back-calculated using SPARTAϩ (33) from the C2(C771A) crystal structure (Protein Data Bank code 6FB4). Correspondingly, the TALOSϩ (34) and SPARTAϩ (33) predictions are that the secondary structure compositions of C2(C771A) and varC2(C771A) in solution are very similar to the C2(C771A) crystal structure (Fig. S4). The ␤-sheet NOE patterns of C2(C771A) and varC2(C771A), moreover, are as expected based on the C2 topology I fold, allowing for peak overlap and absence. NMR data therefore indicate that  Because KIBRA dimerization has been indicated by cellbased data (10,20), MD simulations were conducted to compare conformational stabilities of the C2 dimers observed in crystal structures. Both parallel dimers (WTC2/2Z0U and varC2/6FD0) exhibited significant changes in relative monomer position (Fig. 3) with the changes varying between repeat simulations and force fields. In WTC2 dimer simulations, the mean r.m.s.d. value after three repeat simulations was 0.52 Ϯ 0.04 and 0.49 Ϯ 0.05 nm for GROMOS and OPLS force fields, respectively. The corresponding values in varC2 simulations were 0.51 Ϯ 0.08 nm and 1.4 Ϯ 0.23 nm. In equivalent simulations, in contrast, the C2(C771A) antiparallel dimer (Protein Data Bank code 6FB4) did not undergo significant changes in relative monomer position. There were no significant conformational changes within C2 monomers in any simulation (Fig. 3).

NMR and X-ray diffraction data indicate that KIBRA C2-Ca 2؉ interaction is very low-affinity and nonspecific
Ca 2ϩ binding was investigated by NMR, X-ray crystallography (XRC), and simulation. In numerous 1 H-15 N HSQC-mon-itored titrations with C2(C771A) and one with WTC2, calcium chloride addition to 10 -20 mM and sometimes higher concentrations (up to 76 mM) produced no significant reproducible chemical shift change with the largest composite 1 H-15 N changes around 0.05 ppm. Such small changes may be due to a very weak Ca 2ϩ ion interaction (K d Ͼ 10 mM), possibly nonspecific Ca 2ϩ binding to oxygen-rich surface clusters. Such a nonspecific interaction is supported by simulation: when eight Ca 2ϩ ions were initially positioned randomly around KIBRA C2, the final Ca 2ϩ positions in eight simulations included the calcium-binding regions (CBRs) plus six other locations around the domain (Fig. S6).
It is also possible that the domain scavenged Ca 2ϩ during expression and purification and was therefore Ca 2ϩ -bound prior to the experiments, although this seems unlikely because particular effort was made during some C2 NMR sample preparations to exclude Ca 2ϩ . Initial 1 H-15 N HSQC spectra were very similar, furthermore, irrespective of whether or not measures were taken to exclude Ca 2ϩ from NMR samples.
In several crystal structures, including WTC2 (Protein Data Bank code 2Z0U) and our C2(C771A) and varC2 structures, no bound Ca 2ϩ has been observed, although the protein for 2Z0U was produced by cell-free synthesis and Protein Data Bank entry for 2Z0U does not mention any attempt to introduce metal ions. The crystal structures determined here used Escherichia coli-expressed protein with no attempt to exclude Ca 2ϩ , however, so the C2 domains presumably encountered Ca 2ϩ and other divalent metal ions during expression and purification. Several times C2 crystals were soaked in solutions containing divalent metal ions, including 20 mM CaCl 2 , 100 mM CaCl 2 , and 100 mM MnSO 4 . Soaking did not cause visible disintegration of the crystals. Diffraction after soaking was still good, but these experiments yielded only one metal-bound structure, which had one Ca 2ϩ located in the intermonomer interface of a C2 dimer rather than in any of the CBRs.
Overall, our NMR and XRC data are consistent with KIBRA C2 domain having very low (K d Ͼ 10 mM), nonspecific affinity for Ca 2ϩ . Together with Protein Data Bank code 2Z0U, moreover, our NMR and XRC data show that lack of bound Ca 2ϩ does not disrupt KIBRA C2 from adopting a typical C2 domain fold.

KIBRA C2: Distinctive phosphoinositide and Ca 2؉ binding
peaks plus two pairs of unassigned side-chain NH 2 peaks (Fig.  4). The K d for C2(C771A)-PI(3)P binding was determined as a representative case using 13 perturbed residues (Fig. S7); the average K d was 1 mM. After phosphoinositide titration, CaCl 2 was titrated into the same NMR samples to establish whether Ca 2ϩ ions influence C2-phosphoinositide interaction or vice versa; there was no further spectral change upon CaCl 2 addition.
Coarse-grained MD simulations were performed to further examine KIBRA C2 monomer and dimer association with phosphoinositides and membranes. In these simulations, which can predict the binding modes of peripheral proteins to model membrane (35,36), the three experimentally observed forms of KIBRA C2 (monomer, antiparallel dimer, and parallel dimer, all without Ca 2ϩ ) were displaced away from a preformed phosphoinositide-containing bilayer. In the primary (more frequent) binding mode observed with a C2 monomer, C2 interaction with phosphoinositides involves mainly Arg-661, Lys-671, Arg-772, Arg-776, and Arg-779 (Fig. 6), slightly shifted from the crystal structure/NMR titration binding site. In a small number of simulations, the C2-phosphoinositide interaction involves mainly Lys-667 and Arg-776, coincident with the crystal structure/NMR titration binding site (Figs. 4 and 5). The parallel dimer (WTC2/2Z0U) shows two binding modes with one phosphoinositide often located close to the crystal structure/NMR titration binding site (Fig. 7). The antiparallel dimer (C2(C771A)) exhibits a primary binding mode in which the main interactions with the membrane occur via one of the two C2 domains (Fig. 8). In this orientation, a phosphoinositide is observed near the crystal structure/NMR titration binding site. In all cases, KIBRA C2 domain binding to the bilayer causes a degree of phosphoinositide clustering around the domain. Phosphoinositides interact mainly with lysines and arginines that face the bilayer.

KIBRA C2 structure
We have elucidated structural and functional characteristics of four forms of KIBRA C2 domain: WT, naturally occurring variant, and C771A mutants of WT and variant. Variant KIBRA, involving two amino acid changes in the C2 domain (M734I,S735A), affects human cognitive performance and is in almost complete linkage disequilibrium (30) with a previously identified intronic SNP that affects cognition (15,16). WTC2 It has been shown previously that full-length WT KIBRA forms dimers in mammalian cells (20). Overexpression of FLAG-tagged KIBRA in HEK293 cells, moreover, resulted in large KIBRA-containing clusters (20). KIBRA also forms heterodimers with the two other WWC protein family members, WWC2 and WWC3 (10). Yeast two-hybrid mapping indicated that KIBRA dimerizes in an antiparallel orientation with the C2 domain and one or more regions N-terminal of C2 required for this interaction (20). Whether or not KIBRA can dimerize in mammalian cells via intermolecular disulfide bonding remains open to question. Because redox potentials, and therefore likelihood of intermolecular disulfide bonding, vary according to cell type, cell status, and cell compartment, a mixture of noncovalent and covalent WWC protein dimerization modes is possible in vivo. The degree of KIBRA dimerization and/or oligomerization in the cell could, for example, fine-tune KIBRA function as a hub for multiprotein complex assembly. Different dimerization tendencies of WTC2 and varC2 could then provide an indication of the molecular mechanism(s) underpinning the link between variant KIBRA C2 and improved cognition (30). Other components of the Hippo pathway are modulated by dimerization; indeed, YAP2L and TAZ form disulfide-mediated dimers that are more stable and more onco-

KIBRA C2: Distinctive phosphoinositide and Ca 2؉ binding
genic than the corresponding monomers (37). Finally, because oxidative stress can promote disulfide bond formation in cytoplasmic proteins (38), the extent of disulfide-mediated KIBRA dimerization, and hence KIBRA function, could be regulated by oxidative stress. It is potentially relevant in this context that KIBRA-associated pathways can be redox-modulated (39) and that oxidative stress is a hallmark of AD (40).

Calcium binding
Any KIBRA C2-Ca 2ϩ interaction is low-affinity (K d Ͼ 10 mM): we did not observe bound metal ions in any of our six C2 crystal structures with or without crystal soaking with Ca 2ϩ or other divalent metal ions or bound phosphoinositide, and there is no bound Ca 2ϩ in the previous WTC2 structure (Protein Data Bank code 2Z0U). In NMR-monitored Ca 2ϩ titrations, moreover, the C2(C771A) 1 H-15 N HSQC spectrum did not change significantly. Unlike KIBRA C2, most C2 domains bind Ca 2ϩ with sub-mM affinity. For example, synaptotagmin I C2B domain has a Ca 2ϩ affinity of about 500 M (41), perforin C2 has a Ca 2ϩ affinity of ϳ200 M (42), and rabphilin-3A C2B domain has a particularly high Ca 2ϩ affinity with K d values of 7 and 11 M (43) explained by contributions from C2-flanking residues (44). Like KIBRA C2, however, some C2 domains show very weak or no residual calcium binding; e.g. rat synaptotagmin 4 C2A domain binds Ca 2ϩ with a K d around 10 mM, and rat synaptotagmin 4 C2B domain essentially cannot bind Ca 2ϩ despite the lack of an obvious sequence reason (45). The very low Ca 2ϩ affinity of KIBRA C2 could at least partly arise from suboptimal sequences for Ca 2ϩ binding in CBR1 and CBR3 (Fig. S8).

Phosphoinositide binding and membrane association
C2 domains have been shown to bind phospholipids via the Ca 2ϩ -binding regions (31,46) where the phospholipid contributes to the Ca 2ϩ coordination sphere. C2 domains also bind phospholipids and phosphoinositides via a ␤3-␤4 lysine-rich cluster (31,47,48). Our XRC data involving C2(C771A), however, indicate that KIBRA C2 domain binds phosphoinositide in a novel way involving residues from strands ␤1, ␤2, and ␤8 ( Fig. 5; Protein Data Bank codes 6FJC and 6FJD). Given this unusual binding mode, we checked whether C2 dimerization and/or crystal packing hinders phosphoinositide binding at the previously identified C2-phosphoinositide-binding sites: antiparallel C2 domain dimerization with strands ␤7 and ␤8 forming the intermonomer interface, as observed in our C2(C771A) crystal structures, obscures neither CBRs nor the ␤3-␤4 lysine-rich cluster. Similarly, phosphoinositide interaction at the lysine-rich cluster is not prevented by crystal packing: the cluster is not occluded by symmetry-related molecules. Hence, steric factors do not explain why the previously observed phospholipid-binding modes were not observed for KIBRA C2.
Possible reasons why phosphoinositide interaction with KIBRA C2 does not occur at previously identified C2 phosphoinositide-binding sites include the fact that KIBRA C2 does not conform to the consensus sequence of the ␤3-␤4 lysine-rich cluster (Fig. S9) and may not have sufficient positively charged residues in the vicinity to support phosphoinositide binding. The lack of a lysine-rich cluster may reflect the fact that KIBRA is not required to promote membrane fusion and hence does not need to induce membrane curvature (49). Also, clashes occur when C2(C771A) and PI(4,5)P 2 -bound rabphilin 3A structures are overlaid; the side chain of KIBRA residue Glu-757, for example, lies between phosphates 4 and 5 of PI(4,5)P 2 and apparently would repel these phosphate groups of the phosphoinositide; a glutamate occurs at this position in the C2 domains of all three WWC proteins, but an asparagine occurs here in seven of the eight other aligned C2 domains (Fig. S9). In addition, our observation that KIBRA C2 binds calcium very weakly, or perhaps not at all, with consequent lack of electrostatic bridging between KIBRA C2 negatively charged side chains and negatively charged phosphoinositides, helps to explain why phosphoinositide binding is not observed at the CBRs in KIBRA C2. Although the phosphoinositide headgroup is normally expected to be the key moiety, it is also worth considering whether the type of nonpolar tail influences binding: we used diC 4 forms of phosphoinositides for reasonable solubility while retaining an aliphatic tail. Previous C2 studies have used a range of phosphoinositides, including just the PI(4,5)P 2 headgroup (50, 51), diC 8 tails (49), and seemingly full-length tails (47), all resulting in typical C2 phosphoinositide-binding sites. It seems unlikely therefore that phosphoinositide tail type is a significant factor in our observation of an unusual phosphoinositide-binding site in KIBRA C2.
MD simulations indicate that, when associated with membranes, KIBRA C2 can bind to multiple phosphoinositides. As observed for other membrane-bound proteins (52), this results in clustering of phosphoinositide lipids around the protein with

KIBRA C2: Distinctive phosphoinositide and Ca 2؉ binding
phosphoinositides binding to lysines and arginines that face the membrane. Such phosphoinositide lipid clustering around KIBRA C2 is expected to change the local lipid environment. This, together with the fact that in some simulations we observed secondary orientations of KIBRA C2 relative to the membrane, could influence partner protein recruitment and consequently processes such as trafficking of receptors involved in learning and memory.
In summary, experimental and simulation data indicate that KIBRA C2 is a nonclassical C2 domain that can associate with membranes in a distinctive side-on, Ca 2ϩ -independent manner. Further investigation is required to tease out the detailed functional significance of this unusual mode of membrane association and consequently its implications for molecular mechanisms of learning and memory, organ size control, and major diseases.
Finally, we caution that we have observed this interaction mode using an isolated C2 domain in vitro. Although we have identified several factors that explain the observed binding mode, in the cell it is possible that other factors such as proteinbinding partners and membrane composition/structure alter the KIBRA C2-membrane association mode.

NMR
NMR data for resonance assignment (Fig. S3) were acquired at 35°C on a 600-MHz Varian Unity INOVA spectrometer with an ambient temperature probe at University of Bath or on cryoprobe-equipped 700-MHz Bruker or 800-MHz Varian/Agilent spectrometers at the Medical Research Council Biomedical NMR Centre, Mill Hill. Protein concentration used for 3D NMR experiments was in the range 0.5-0.8 mM. NMR data were processed using NMRPipe/NMRDraw (54) and analyzed using CCPN Analysis (55). Backbone chemical shifts for C2(C771A) and varC2(C771A) were deposited in Biological Magnetic Resonance Bank (BMRB) under accession numbers 27429 and 27430, respectively. TALOSϩ (34) was used to predict C2 domain secondary structure from the assigned NMR chemical shifts. For comparison, secondary structure was pre-dicted from crystal structures using SPARTAϩ (33) to convert the coordinates to chemical shifts and then using those to predict secondary structure.
Numerous 1 H-15 N HSQC-monitored titrations of C2(C771A) and one of WTC2 with CaCl 2 were conducted, including attempts to produce NMR samples that were initially calciumfree, which involved buffer solutions made with commercial calcium-free water and treatment with EGTA. Typical initial sample conditions for these experiments were 50 mM MES, pH 6.5, 50 mM NaCl, and 10 mM MES, pH 6.5, 150 mM NaCl, sometimes with EGTA (1 or 2 mM) as an additional measure to try to ensure that C2 was not Ca 2ϩ -bound to begin with. Final CaCl 2 concentrations in some of these titrations exceeded 15 mM, in one case reaching 76 mM.

Analytical ultracentrifugation
Sedimentation velocity scans were recorded for four (varC2) or five (WTC2, C2(C771A), and varC2(C771A)) concentrations of each construct; concentrations are listed in each panel of Fig. S1. All experiments were performed at 50,000 rpm using a Beckman XL-I analytical ultracentrifuge with an An-50Ti rotor. Data were recorded using the absorbance (at 280 nm) and interference optical detection systems. The density and viscosity of the buffers were measured using a DMA 5000M densitometer equipped with a Lovis 200ME viscometer module. The partial specific volume for each protein was calculated using Sednterp from the amino acid sequence. Data were processed using SEDFIT (57), fitting to the c(s) model (Table S2).  (58). Molecular replacement was carried out with BALBES (59), and the model was refined using Coot (60) and PHENIX (61). The resultant structures were evaluated using MolProbity (62,63). Data collection and processing statistics are given in Table S4.

Coarse-grained molecular dynamics (CG-MD) simulations
CG-MD simulations were performed using the Martini 2.1 force field (64,65) and GROMACS (66). In the CG-MD simulations, the protein (in monomeric and dimeric forms) was displaced away from a preformed bilayer. Protein orientation relative to the bilayer was different at the beginning of each repeat simulation. 25 repeat simulations of 2 s each were conducted. The bilayer consisted of 1-palmitoyl-2-oleoyl-sn-glycero-3phosphocholine (ϳ73%), 1-palmitoyl-2-oleoyl-sn-glycero-3phosphoserine (ϳ20%), PI(4,5)P 2 (ϳ5%), and PI(3,4,5)P 3 (ϳ2%). All systems were energy-minimized and subsequently equilibrated (for 500 ns) with the protein backbone particle restrained. For the production simulation, the time step was 20 fs, the pressure was 1 bar, and the temperature was 323 K. Berendsen's algorithm (67) was used to control the pressure and temperature. An elastic network model was applied to all backbone particles with a cutoff distance of 0.7 nm (68). The LINCS algorithm was used to constrain bond lengths (69), and the Lennard-Jones interactions were shifted to zero between 0.9 and 1.2 nm. Coulombic interactions were shifted to zero between 0 and 1.2 nm.

Atomistic MD simulations
Atomistic MD simulations were run at 310 K using GRO-MACS and two different force fields: the GROMOS96 43a1 force field (70) was used with SPC water molecules, and the OPLS-AA force field was used with TIP4P water molecules. The velocity rescaling method (71) was used to control the temperature, and the Parrinello-Rahman barostat (72) was used for pressure control. Isotropic pressure coupling was used. Bond lengths were constrained to equilibrium lengths using the LINCS method, and the particle mesh Ewald method was used to model the electrostatic interactions. The time step was 2 fs. WTC2, varC2, and varC2(C771A) structures were used for these simulations. Two simulations of 100 ns and one simulation of 300 ns were performed for each force field. We also performed simulations conducted with monomeric KIBRA C2 domain in solution in which eight calcium ions were randomly added. Eight repeat 100-ns simulations were performed, each starting from different initial configurations, using the OPLS-AA force field with TIP4P water molecules.