|Home | About | Journals | Submit | Contact Us | Français|
Phospholipid-binding modules such as PH, C1, and C2 domains play crucial roles in location-dependent regulation of many protein kinases. Here, we identify the KA1 domain (kinase associated-1 domain), found at the C-terminus of yeast septin-associated kinases (Kcc4p, Gin4p and Hsl1p) and human MARK/PAR1 kinases, as a membrane association domain that binds acidic phospholipids. Membrane localization of isolated KA1 domains depends on phosphatidylserine. Using X-ray crystallography, we identified a structurally conserved binding site for anionic phospholipids in KA1 domains from Kcc4p and MARK1. Mutating this site impairs membrane association of both KA1 domains and intact proteins, and reveals the importance of phosphatidylserine for bud neck localization of yeast Kcc4p. Our data suggest that KA1 domains contribute to “coincidence detection”, allowing kinases to bind other regulators (such as septins) only at the membrane surface. These findings have important implications for understanding MARK/PAR1 kinases, which are implicated in Alzheimer’s disease, cancer, and autism.
Regulation of cellular processes requires precisely controlled intermolecular interactions that alter the location and/or activity of effector proteins (Scott and Pawson, 2009), typically driven by protein modules that recognize specific features of proteins, nucleic acids or membranes (Seet et al., 2006). Several protein modules recognize anionic membrane phospholipids, including PH, C2, PX and FYVE domains (Lemmon, 2008). Some recognize phosphoinositides (PtdInsPns), levels and locations of which are tightly regulated. Others bind phosphatidylserine (PtdSer), which is concentrated in the plasma membrane inner leaflet (Yeung et al., 2008) and constitutes approximately 20% of phospholipid (Stace and Ktistakis, 2006).
Many more cellular functions appear to depend on anionic phospholipids than can be explained by currently understood phospholipid-binding domains (Audhya et al., 2004; Halstead et al., 2005; McLaughlin and Murray, 2006; Yu et al., 2004). Indeed, over 100 proteins that contain no known lipid-binding domain were found to bind phosphoinositides using a microarray-based analysis of the expressed S. cerevisiae proteome (Zhu et al., 2001). Here, we describe an analysis of the membrane association properties of these yeast proteins, from which we have identified several additional potential phospholipid-binding domains. We focus in this report on a membrane-targeting domain found at the C-terminus of the S. cerevisiae septin-associated protein kinases Kcc4p, Gin4p and Hsl1p. These kinases are involved in septin organization or in the yeast morphogenesis checkpoint that coordinates cell cycle progression with bud formation (Lew, 2003; Longtine and Bi, 2003; Shulewitz et al., 1999). They become activated at the bud neck, and are involved in septin ring assembly and/or promote Swe1p degradation to allow entry into mitosis (Barral et al., 1999; Sakchaisri et al., 2004). The C-terminal phospholipid-binding domain of the septin-associated kinases is required for their bud neck localization and function, and appears to bind phosphatidylserine in vivo. Using X-ray crystallography, we found that this phospholipid-binding domain has the same fold as the KA1, or kinase associated-1 domain (Pfam accession PF02149), one of the only common domains in protein kinases to which no function has yet been ascribed (Manning et al., 2002; Tochio et al., 2006).
KA1 domains are also found at the C-termini of mammalian Ser/Thr kinases that phosphorylate microtubule associating proteins (MAPs) such as tau, promoting their detachment from microtubules and thus reducing microtubule stability (Drewes et al., 1997). These kinases comprise the MARK/PAR1 family, which includes MAP/microtubule affinity regulating kinase (MARK) and partitioning-defective 1 or PAR1 (Matenia and Mandelkow, 2009; Timm et al., 2008), as well as the S. cerevisiae Kin1/2 kinases (Tassan and Le, 2004). MARK/PAR1 kinases are related to the AMP-activated protein kinase (AMPK)/Snf1 family (Manning et al., 2002; Marx et al., 2010). They are frequently found associated with membrane structures, and participate in diverse processes from control of the cell cycle and polarity to intracellular signaling and microtubule stability (Marx et al., 2010; Tassan and Le, 2004). MARK/PAR1 kinases have been implicated in carcinomas, Alzheimer’s disease (through tau hyperphosphorylation), and autism (Gray et al., 2005; Hurov et al., 2007; Maussion et al., 2008; Timm et al., 2008). We establish here that KA1 domains from both yeast and human kinases bind anionic phospholipids, thus ascribing a function to this poorly understood domain and providing important clues as to how activation of these AMPK-related kinases may be directly coordinated with membrane localization.
Zhu et al. (2001) reported phosphoinositide binding for 128 of 5,800 S. cerevisiae ORF protein products arrayed on proteome chips – excluding dubious ORFs and integral membrane proteins. We selected 62 of these for further analysis (15 of which were protein kinases), including all “strong binders” defined by Zhu et al. (2001) plus potentially interesting “weak binders”. We first tested in vivo membrane association of these 62 proteins using a S. cerevisiae Ras rescue assay (Isakoff et al., 1998; Yu et al., 2004). Each protein was fused to constitutively active (Q61L), non-farnesylated, Ha-Ras and expressed in cdc25ts yeast cells – which harbor a temperature-sensitive mutation in the Ras GEF Cdc25p. If the test protein drives plasma membrane recruitment of this Ha-Ras fusion, it promotes growth above the restrictive temperature (complementing the cdc25ts allele) by overcoming the block in endogenous Ras activation (Isakoff et al., 1998). Of the 62 proteins analyzed, 33 promoted membrane recruitment of constitutively active Ha-Ras (Figure S1A & Table S1A), consistent with them harboring a phospholipid-binding domain. In qualitative lipid overlays (Kavran et al., 1998), 21 of these 33 membrane-targeted proteins also interacted in vitro with filter-bound anionic phospholipids (Table S1A), displaying a broad range of specificities. Several of the candidate Ras rescue-positive proteins also showed punctate or plasma membrane fluorescence when expressed as GFP fusion proteins in yeast or HeLa cells (Table S1A). For five of the candidate proteins (Cam1p, Dps1p, Kcc4p, Rgd1p and Stp22p), Ras rescue analysis of deletion mutants identified regions or domains responsible for membrane targeting (Table S1B). We focus here on Kcc4p.
In studies of the septin-associated kinase Kcc4p, Ras rescue analysis identified a C-terminal 160aa fragment (aa 877–1037) that is sufficient to drive Ha-Ras membrane recruitment in yeast cells (Figure 1A). This fragment also displays strong plasma membrane association when overexpressed as a GFP fusion protein in either S. cerevisiae or human HeLa cells (Figure 1B), suggesting recognition of a lipid that is common to yeast and human cells, rather than association with a less abundant protein target at the membrane.
As shown in Figure 1C, purified protein corresponding to residues 901–1037 from the Kcc4p C-terminus (Kcc4p901-1037) binds ‘promiscuously’ to PtdIns(4,5)P2 and other acidic phospholipids in surface plasmon resonance (SPR) studies. Overlay studies of intact Kcc4p (Table S1A) showed a similar lack of specificity, consistent with the binding to several phosphoinositides reported previously by Zhu et al. (2001). Kcc4p901-1037 bound with similar affinities to membranes containing 10% (mole/mole) PtdIns(4,5)P2, 20% (mole/mole) phosphatidic acid (PA), or 20% (mole/mole) phosphatidylserine (PtdSer) – all in a dioleoylphosphatidylcholine (DOPC) background. The binding data fit well to simple hyperbolic curves with apparent KD values from 3–10 μM (Table S2), in the same range reported for several other phospholipid-interaction domains (Lemmon, 2008). The amount of Kcc4p901-1037 bound at saturation (Bmax) scaled with anionic phospholipid content for PtdIns(4,5)P2 or PtdSer (Figure 1D). Interestingly, in all studies, Bmax was proportional to the anticipated negative charge density on the SPR sensorchip surfaces (rather than number of lipid molecules), assuming charge valences of −4, −2 and −1 for PtdIns(4,5)P2, PA, and PtdSer respectively at pH 7.4 (McLaughlin and Murray, 2006). As shown in Figure 1C, Bmax was approximately 2000 RUs for membranes containing either 10% PtdIns(4,5)P2 (charge −4) or 20% PA (charge −2), and approximately 1000 RUs for membranes containing 20% PtdSer (charge −1). These observations suggest that, rather than forming simple 1:1 complexes, binding stoichiometry depends on lipid charge – each Kcc4p901-1037 chain binding four times more PtdSer molecules (charge −1) than PtdIns(4,5)P2 molecules (charge −4). We also analyzed Kcc4p901-1037 binding to small unilamellar vesicles using a centrifugation-based sedimentation assay (Kavran et al., 1998). Only background levels of Kcc4p901-1037 sedimented with vesicles with no net charge; i.e. those containing 100% phosphatidylcholine (PC) or 20% (mole/mole) phosphatidylethanolamine (PE) in a PC background (Figure 1E). By contrast, vesicles containing 20% (mole/mole) of the anionic phospholipids PtdSer or PtdIns sedimented the majority of the Kcc4p901-1037 when anionic lipid was present at ≥50μM. Divalent cations did not significantly alter the affinity or specificity of phospholipid binding by Kcc4p901-1037. Neither elevating divalent cation levels (by adding 10 μM CaCl2 and 1 mM MgCl2) nor depleting them (by adding 1 mM EDTA) changed apparent KD values by more than 2-fold (Table S2).
The only clearly recognizable protein module in Kcc4p according to the SMART, Pfam and UniProt databases is the N-terminal kinase domain (Figure 1A). However, BLAST searches (Altschul et al., 1990) identify a ~130 amino acid region related to Kcc4p901-1037 at the C-termini of the functionally related S. cerevisiae kinases Gin4p and Hsl1p (Figures 2A and S2). The Gin4p C-terminus (residues 1007–1142) shares 41% sequence identity with Kcc4p901-1037, and the Hsl1p C-terminus (residues 1379–1518) is more distantly related (sharing just 16% identity with Kcc4p901-1037). As shown in Figure 2B, fusing these C-terminal regions from Gin4p or Hsl1p to Q61L Ha-Ras allowed complementation of the cdc25ts allele in Ras rescue assays. The Gin4p and Hsl1p C-termini also showed robust plasma membrane localization when expressed in yeast cells as GFP fusion proteins (Figure 2C). Moreover, the Gin4p C-terminal domain (expressed in E. coli as a GST fusion protein) bound PA, PtdIns(4,5)P2 and PtdSer in SPR studies (Figure 2D), resembling the in vitro interactions seen for Kcc4p901-1037 (although with different charge dependence, interpretation of which is complicated by dimerization of the fused GST). The Gin4p and Hsl1p C-termini therefore have broadly similar membrane-binding properties to those seen for Kcc4p901-1037. It is important to point out that Gin4p and Hsl1p were not found in the proteome-wide screen of yeast phospholipid-binding proteins described by Zhu et al. (2001), arguing that additional, as-yet-unidentified, S. cerevisiae phospholipid binding proteins may exist.
To determine which cellular phospholipids are important for in vivo membrane association of the C-termini from Kcc4p, Gin4p and Hsl1p, we assessed their localization (as GFP fusion proteins) in S. cerevisiae mutants harboring specific phospholipid synthesis defects. Plasma membrane localization was not detectably altered when levels of PtdIns(4,5)P2 or PtdIns4P were reduced by manipulation of temperature sensitive yeast strains (Stefan et al., 2002), arguing that neither of these phosphoinositides plays a dominant role (Figure S3). By contrast, in cho1Δ cells that lack PtdSer (Hikiji et al., 1988) the degree of plasma membrane association of each domain was reduced significantly (Figure 3). Ratios of plasma membrane to cytosolic fluorescence (FPM/FCyt: see Experimental Procedures) in wild-type cells were 1.4 ± 0.35, 1.5 ± 0.08 and 2.9 ± 1.0 respectively for GFP-Kcc4p877-1037, GFP-Gin4p1003-1142, and GFP-Hsl1p1358-1518; similar to the FPM/FCyt ratio of 1.5 ± 0.16 measured for the lactadherin discoidin-type C2 domain previously characterized as a specific PtdSer probe (Yeung et al., 2008). Loss of PtdSer in cho1Δ cells reduced FPM/FCyt ratios to 0.53 ± 0.15 (Kcc4p877-1037), 0.93 ± 0.20 (Gin4p1003-1142), and 0.95 ± 0.13 (Hsl1p1358-1518) – mirroring the effect on the PtdSer-specific lactadherin C2 domain (FPM/FCyt = 0.61 ± 0.20).
Previous studies employing fluorescent surface-potential probes and the lactadherin C2 domain have shown that the plasma membrane inner leaflet is the most negatively charged of cytoplasmic-facing membranes, and that PtdSer is the primary determinant of this surface charge (Yeung et al., 2008; Yeung et al., 2006). C-terminal domains from the septin-associated kinases appear to resemble these non-specific surface-potential probes. They show preferential targeting to the plasma membrane that is dependent on PtdSer, although they do not specifically recognize this lipid. The residual plasma membrane association seen in cho1Δ cells for these domains (Figure 3) may reflect their ability to bind either PtdIns (see Figure 1E), levels of which are known to be elevated in cho1Δ cells (Hikiji et al., 1988), or other less abundant anionic plasma membrane phospholipids.
In an effort to understand anionic phospholipid binding by C-terminal domains from the septin-associated kinases, we determined the X-ray crystal structure of Kcc4p917-1037 to 1.7 Å resolution (see Table S3). The domain contains two interacting α helices (α1 and α2) that lie on the concave surface of a five-stranded antiparallel β-sheet (Figures 4A and S4). A short β-strand (β1) precedes helix α1, which is then followed by a 4-stranded β-meander (β2-β5) and a C-terminal α-helix (α2). Remarkably, the structure of Kcc4p917-1037 is very similar to that of the extended Kinase-Associated 1 domain (KA1) from the MARK3 human MAP/microtubule affinity regulating kinase (Tochio et al., 2006), depicted in Figure 4B (PDB ID 1UL7). KA1 domains were initially defined as a Pfam domain family of ~50 amino acids (PF02149) at the C-termini of kinases from the MARK/PAR1/Kin family (Matenia and Mandelkow, 2009; Tassan and Le, 2004; Timm et al., 2008). NMR structural studies (Tochio et al., 2006) showed that the stable KA1 domain in MARK3 actually contains ~100 amino acids. The 118-residue phospholipid-binding domain at the Kcc4p C-terminus that we have identified here also appears to be a KA1 domain. It contains all secondary structure elements seen in MARK3-KA1, plus a short additional α-helix at its amino terminus (αN). As shown in Figure 4C, the core (~100 amino acid) Kcc4p and MARK3 KA1 domains overlay very well (with Cα position rms deviation of just 2.4 Å), despite sharing only 10% sequence identity – explaining the failure to identify this domain through sequence analysis. A structure-based sequence alignment of KA1 domains from the MARK/PAR1/Kin family and the Kcc4p/Gin4p/Hsl1p kinases is shown in Figure S2.
Although speculated to participate in autoregulatory intramolecular interactions in MARK/PAR1 kinases (Marx et al., 2010), no clear function has been ascribed to KA1 domains. Having identified the Kcc4p KA1 domain as a phospholipid-binding domain, we next asked whether previously-recognized KA1 domains from human MARK1, MARK3 and MELK (Maternal Embryonic Leucine zipper Kinase) also associate with cell membranes and bind phospholipids. As shown in Figure 5A, all of these KA1 domains recruit Q61L Ha-Ras fusions to yeast cell membranes, complementing the cdc25ts mutation in Ras rescue assays. GFP fusions of the MARK1 and MARK3 KA1 domains showed substantial plasma membrane localization in HeLa cells (Figure 5B). Moreover, the MARK1, MARK3 and MELK KA1 domains (as GFP fusions) showed robust plasma membrane localization in S. cerevisiae, with FPM/FCyt ratios ranging from 1.8 to 3.1 (Figure 5C). Again, these values were reduced by ~50% in PtdSer-deficient cho1Δ cells (Figure 5C), but were not significantly altered in mutant yeast strains with reduced phosphoinositide levels (Figure S5). The subcellular localization properties of KA1 domains from human MARK1, MARK3 and MELK therefore appear similar to those seen for the Kcc4p, Gin4p and Hsl1p KA1 domains identified here. In addition, purified monomeric MARK1-KA1 showed essentially the same in vitro phospholipid-binding characteristics as Kcc4p-KA1, binding to vesicles that contain PtdSer, PA or PtdIns(4,5)P2 (Figure 5D) with KD values in the 2.3 μM – 8.9 μM range (Table S2), and with Bmax values that scale with membrane charge density. The KA1 domains from MARK/PAR1 family kinases thus appear to be phospholipid-binding domains that are likely to promote membrane association of their host proteins in cells. Indeed, Alessi and colleagues (Göransson et al., 2006) previously implicated the KA1 domain as an important membrane localization determinant in MARK3 mutants that fail to bind 14-3-3 proteins. Our findings suggest that this observation reflects MARK3-KA1 binding to acidic phospholipids, and argue that the KA1 domain should be considered as a bona fide membrane-targeting/anionic phospholipid-binding module.
To understand how KA1 domains interact with negatively charged membranes, we analyzed features common to the structure of the yeast Kcc4p KA1 domain and a crystal structure of the human MARK1 KA1 domain that we determined to 1.7 Å resolution (see Table S3). Both have notable positively charged patches and/or crevices on their surfaces (Figure 6) that result from basic side-chain arrangements reminiscent of headgroup-binding sites in other phospholipid-interaction domains (Hurley, 2006; Lemmon, 2008).
For Kcc4p-KA1, clear electron density could be seen for two bound sulfate ions, 27 Å apart, which lie on either side of a positively-charged region that stretches across the width of the domain in the orientation shown in Figure 6A, and encircles the β3/β4 loop that projects prominently from its surface. One of these sulfates (SO4#1) interacts primarily with lysine side-chains in the αN/β1 loop (K932), and β5 (K1010), and lies close to K1016 in the amino-terminal part of helix α2 (Figures 6A and S4A). Adjacent electron density (~3 Å away) is fit well with a glycerol molecule that contacts K1010 in strand β5 plus serine and threonine side-chains (S1014 and T1015) at the beginning of helix α2 (Figure S4A). Intriguingly, in a second crystal form (Table S3) density for a tartrate ion replaces both SO4#1 and the bound glycerol (Figure S4B), implicating this region as an important anion binding-site in Kcc4p-KA1. The second sulfate in Figure 6A (SO4#2) lies in a basic pocket on the Kcc4p-KA1 surface formed largely by side-chains from the helix α1 C-terminus (K959) and the α1/β2 loop (K964).
The locations of bound anions in crystal structures of membrane-targeting domains frequently reveal the binding sites for phospholipid headgroups (Hurley, 2006; Lemmon, 2008; Wood et al., 2009). We therefore used mutagenesis to investigate the importance of the SO4#1 and SO4#2 binding sites for in vivo membrane association of Kcc4p-KA1. When pairs of basic residues were mutated (Figure 6A), plasma membrane localization of GFP/Kcc4p-KA1 was only impaired when one or both mutated residues contributed to binding of one of these sulfates (K932, K1007, K1010, K1016, K1020, K964, and K978 were implicated). Importantly, mutations at both sulfate-binding sites diminished membrane recruitment, suggesting that the KA1 domain makes multiple contacts with the bilayer surface. Engaging both the SO4#1 and SO4#2 sites in binding to a membrane surface is difficult to envision without the β3/β4 loop penetrating the bilayer. This loop contains several hydrophobic side-chains (with sequence VNDSILFL), and resembles ‘membrane insertion loops’ reported in C2, PX, and FYVE domains (Cho and Stahelin, 2005; Lemmon, 2008). As shown in Figure S6A, Kcc4p-KA1 can indeed penetrate acidic phospholipid-containing monolayers that have packing densities similar to those estimated for cell membranes (Demel, 1994; Marsh, 1996) – resembling C2, PX, FYVE, and some PH domains in this respect (Stahelin et al., 2007).
Only the SO4#1/glycerol binding site of Kcc4p-KA1 is conserved in the hMARK1 KA1 domain – in location, charge characteristics (Figure 6B) and sequence (Figure S2). It lies in the most sequence-conserved region of aligned KA1 domains that encompasses strand β5, helix α2, and the loop that connects them. In addition to conserved positive charge in this region (in β5), all KA1 domains have serine and/or threonine residues at the beginning of helix α2 that contact bound glycerol in Kcc4p-KA1 (Figure S4A) and may interact similarly with the glycerol backbone of bound phospholipids. As anticipated from these observations, hMARK1-KA1 mutations in the basic patch corresponding to the Kcc4p SO4#1 binding site impaired both plasma membrane association (Figure 6B) and in vitro binding to anionic phospholipids (Figure S6B). K773 and R774 in strand β5 of hMARK1-KA1 appear to be important for membrane association. Moreover, an R698S/R701S double mutation close to the hMARK1-KA1 N-terminus prevented plasma membrane association and vesicle binding, suggesting that the basic patch extending to the bottom left of hMARK1-KA1 in Figure 6B makes additional contributions – perhaps functionally replacing the SO4#2 binding site of Kcc4p-KA1. Thus, membrane association of both the Kcc4p and the MARK1 KA1 domains appears to involve cooperation of more than one positively-charged binding region – centered on the conserved SO4#1 binding site seen in Kcc4p-KA1. Similar utilization of multiple binding sites has previously been described for annexins, as well as PKC-type C2, PX, and PH domains (Lemmon, 2008).
Double mutations (K1007S/K1010S or K1016S/K1020S) that abolish membrane localization of isolated Kcc4p-KA1 in Figure 6A did not prevent intact Kcc4p from being targeted to the bud neck when overexpressed in wild-type yeast cells (Figure 7A). However, background cytoplasmic fluorescence was increased to some extent, and simultaneous introduction of all four KA1 domain mutations into intact Kcc4p abolished its targeting to bud necks.
Hypothesizing that the KA1 domain must cooperate with other domains in targeting intact Kcc4p to bud necks, we surmised that residual low-affinity PtdSer binding by K1007S/K1010S- or K1016S/K1020S-mutated KA1 domains might be sufficient to drive normal Kcc4p targeting in this overexpression study. We therefore re-examined localization of the intact GFP-Kcc4p variants in cells lacking PtdSer. As suspected, PtdSer loss (in cho1Δ cells) completely abrogated bud neck localization of K1007S/K1010S-mutated GFP-Kcc4p (Figure 7A: see also Figure S7A). In other words, K1007S/K1010S-mutated Kcc4p is dependent on normal plasma membrane PtdSer levels for its targeting to the bud neck, implicating PtdSer as an important determinant of Kcc4p localization. Bud neck localization was still seen for wild-type and K1016S/K1020S-mutated GFP-Kcc4p in cho1Δ cells (although cytosolic fluorescence was increased) – suggesting that the elevated PtdIns levels found in these cells (Hikiji et al., 1988) may be sufficient. Western blotting confirmed that all GFP-Kcc4p variants were expressed at or above wild-type levels (Figure S7B). Taken together, these data show that bud neck targeting of intact GFP-Kcc4p can be abolished either by mutating basic residues in the KA1 domain’s anionic phospholipid binding site or – importantly – by simultaneously reducing anionic phospholipid levels in the plasma membrane inner leaflet and mutating the KA1 domain.
The lack of a clear phenotype for KCC4 mutations (Longtine et al., 2000) prevented us from being able to assess functional consequences of the KA1 domain mutations described above. However, studies of Gin4p demonstrated a functional requirement for the KA1 domain (Figure 7B). Deleting the GIN4 (or HSL1, but not KCC4) gene in S. cerevisiae leads to an elongated bud phenotype characteristic of a G2/M delay due to morphogenesis checkpoint failure (Longtine et al., 1998). In gin4Δ cells, this elongated bud phenotype can be rescued by overexpressing a wild-type Gin4p GFP fusion (Figure 7B), and the protein is found at bud necks. However, when just the KA1 domain (but not septin-binding region) is deleted, the GFP-Gin4pΔKA1 fusion fails to rescue gin4Δ cells, and is diffusely localized (Figure 7B) in much the same way as GFP-Kcc4p harboring multiple KA1 domain mutations.
Our search for previously undescribed phosphoinositide/phospholipid binding domains identified a small C-terminal domain in S. cerevisiae septin-associated kinases that binds acidic phospholipids. Crystallographic studies revealed that this is a KA1 domain, a module previously identified at the C-termini of kinases from the mammalian MARK/PAR1 family. We show that KA1 domains from both yeast and human kinases bind acidic phospholipids including PtdSer. For yeast Kcc4p, we also present data using KA1 domain mutations that implicate PtdSer as an important determinant for targeting this kinase to its site of action at the bud neck.
Our findings with Kcc4p and Gin4p argue that – in addition to its documented dependence on septin binding (Barral et al., 1999; Longtine et al., 1998) – bud neck localization of septin-associated kinases requires KA1 domain/phospholipid interactions. On their own, neither the KA1 domain nor the septin-binding region of Kcc4p/Gin4p/Hsl1p is sufficient for specific bud neck targeting – but C-terminal fragments encompassing both are efficiently localized to bud necks (Crutchley et al., 2009; Longtine et al., 1998; Okuzaki and Nojima, 2001). Thus, simultaneous engagement of the septin- and phospholipid-binding domains appears to be required for Kcc4p, Gin4p and Hsl1p recruitment to septin assemblies at the bud neck for kinase activation. This combination of septin-binding and phospholipid-binding domains may function as an effective “coincidence detector”, allowing the kinases to bind septins only at membrane locations. The septins themselves also bind weakly to anionic phospholipids (Casamayor and Snyder, 2003; Zhang et al., 1999), suggesting further that kinase/phospholipid, kinase/septin and septin/phospholipid interactions all cooperate to organize a well-defined assembly at the bud neck. Coincidence-detection of this sort, in which multivalent interactions involving both protein-binding and lipid-binding domains drive complex formation, has been suggested for several systems (Carlton and Cullen, 2005; Lemmon, 2008). It is particularly interesting for Kcc4p that the KA1 domain can promote kinase targeting to a specific location (the bud neck) despite binding non-specifically to anionic phospholipids: it appears to restrict the ability of Kcc4p to bind septins only in the context of a negatively-charged membrane surface, as a logical ‘AND’ gate. Similar coincidence-detection mechanisms may also be relevant for specific membrane targeting of human MARK/PAR1 family proteins. Indeed, we show here that – like their structural counterparts in the yeast septin-associated kinases – KA1 domains of human MARK/PAR1 family proteins bind acidic phospholipids in cells and in vitro.
Several reports have suggested that the C-terminal tail of MARK/PAR1 kinases (which includes the KA1 domain) plays a role in reversible autoinhibition of kinase activity (Elbert et al., 2005; Marx et al., 2010; Timm et al., 2008). For example, the C-terminal KA1 domain-containing region of the S. cerevisiae Kin1 and Kin2 kinases was reported to interact with the N-terminal catalytic domain (Elbert et al., 2005) – suggesting direct intramolecular autoinhibitory interactions. A similar model was also proposed for S. cerevisiae Hsl1p (Hanrahan and Snyder, 2003), and septins were suggested to activate Hsl1p by binding close to the C-terminal region and disrupting autoinhibitory intramolecular interactions. One concern raised about this model (Crutchley et al., 2009; Szkotnicki et al., 2008) is that it cannot explain why Hs1lp is activated only by assembled septins at the bud neck, and not by free septin complexes. Our findings provide an explanation: that the C-terminal region of Hsl1p (and other septin-associated kinases) must bind to both septins and anionic membrane phospholipids (via its KA1 domain) to drive the protein to the bud neck and relieve the proposed intramolecular autoinhibition.
Reversing intramolecular autoinhibitory interactions by engaging one or more phospholipid-binding domains is a recurring theme in kinase regulation, with protein kinase C (PKC) and other AGC kinases providing well characterized examples (Newton, 2009). Our studies suggest that the mechanistic role of the KA1 domain in septin-associated kinases may be broadly analogous to that of C1 and C2 domains in PKC or the PH domain in Akt (Newton, 2009). The KA1 domain lacks the lipid selectivity of these other modules, but appears to restrict specific recognition of other targets (such as septins) to a membrane context. Extending our observations to the MARK/PAR1 family kinases, the KA1 domain was previously implicated as a determinant of membrane localization for MARK3 (Göransson et al., 2006), and dissociation of hMARK2 from the plasma membrane coincides with reduced activity (Hurov et al., 2004). Thus, phospholipid-engagement of the KA1 domain may also play a role in the activation of these kinases at particular membrane locations. Intriguingly, the KA1 domain fold has recently been seen in additional kinase contexts that warrant further investigation. A C-terminal domain in the Arabidopsis AtSOS2 kinase has a KA1 domain fold (Sánchez-Barrena et al., 2007), and includes a protein phosphatase-interacting (PPI) motif (in strand β1 and helix α1). It is not known whether this domain binds phospholipids. A C-terminal domain in the α-subunit of heterotrimeric AMPK orthologs also has a KA1 domain fold, and is intimately associated with the C-terminal region of the β-subunit (Townley and Shapiro, 2007). Since KA1 domain-containing proteins are implicated in a wide range of diseases, from Alzheimer’s disease to cancer to diabetes, understanding the regulatory role of this domain is an important goal. Our studies show that at least a subgroup of KA1 domains bind non-specifically to acidic phospholipids and allow kinase activation to be coordinated with membrane association, in an unexpected variation of a theme used by other kinases that employ C1, C2, PH, and other domains.
Ras rescue assays were performed exactly as described (Yu et al., 2004). Briefly, DNA encoding candidate proteins or fragments was PCR amplified from S. cerevisiae (BY4741) genomic DNA or a HeLa cell cDNA library, and subcloned into modified p3S0BL2 (Isakoff et al., 1998) to generate plasmids encoding Ha-Ras Q61L fusions. Plasmids were transformed into cdc25ts yeast cells, and rescue of the growth defect at 37°C assessed as described (Yu et al., 2004).
For yeast studies, DNA fragments encoding candidate proteins or domains were subcloned into modified pGO-GFP (Cowles et al., 1997) and transformed into wild-type (BY4741) or cho1Δ BY4743 cells as described (Audhya and Emr, 2002). Images were collected at 100X magnification using a Leica-DMIRBE microscope and processed using Volocity deconvolution software (Improvision). All images of yeast cells are representative of >90% of cells expressing the relevant GFP fusion protein (from over 100 cells in at least three experiments). Analysis of full-length (or ΔKA1) Gin4p was performed in YEF1238 gin4Δ::TRP1 (YEF473A) cells (Longtine et al., 1998). To quantify plasma membrane localization, lines were drawn across individual cells using ImageJ and mean values for fluorescence in the plasma membrane (FPM) and cytosolic (FCyt) regions were determined along the length of these lines as described (Szentpetery et al., 2009). The ratio of these means (FPM/FCyt) was used as a measure of plasma membrane localization.
For analysis of subcellular localization in mammalian cells, domains of interest were subcloned into pEGFP-C1 (Clontech) and transiently transfected into HeLa cells using Lipofectamine 2000 (Invitrogen). Cells were imaged at 40X, and images processed as above. All microscopy images presented are representative of at least three independent experiments, assessing over 100 cells each.
Phospholipid-binding experiments were performed using surface plasmon resonance (SPR) exactly as described previously (Yu et al., 2004) or sedimentation assays (Kavran et al., 1998). For SPR studies, vesicles contained dioleoylphosphatidylcholine (DOPC) alone, or the noted percent (mole/mole) of test lipid in a DOPC background, and were immobilized on L1 sensor chip surfaces (BIAcore). Purified test proteins were flowed over these surfaces at a series of concentrations, determined by absorbance at 280nm using calculated extinction coefficients. SPR signals for each experiment were corrected for background (DOPC) binding and plotted against protein concentration to yield binding curves that were fit to simple hyperbolae. Experiments were performed in 25 mM HEPES, pH 7.4, containing 150 mM NaCl. For sedimentation assays, brominated PC was used as the background lipid and experiments were performed exactly as described (Kavran et al., 1998).
DNA encoding the KA1 domains from Kcc4p (residues 917–1037) and MARK1 (residues 683–795), plus an N-terminal hexahistidine tag were subcloned into pET21a (Novagen) for expression in E. coli BL21 (DE3) cells. For generating selenomethionine (SeMet)-containing Kcc4p-KA1 protein, a third methionine was introduced by substitution at L936, and protein was produced from B834(DE3) methionine auxotrophs in MOPS-based minimal medium supplemented with SeMet. Proteins were purified from cell lysates in three steps, using Ni-NTA resin (Qiagen), cation exchange chromatography, and a Superdex 75 size exclusion column (GE Healthcare). Crystals were grown at 21°C using the hanging drop vapor diffusion method by mixing equal parts of protein (at 300–400μM) and reservoir solutions. MARK1-KA1 crystals were obtained from 0.1 M Na acetate, pH 4.6, containing 0.04 M CaCl2, and 15–25% (w/v) PEG 3350. Kcc4p-KA1 crystals were obtained both from 0.1 M HEPES, pH 7.4, containing 0.2 M (NH4)2SO4 plus 20% (w/v) PEG3350 (for both native and SeMet protein), and from 1.0 M K/Na tartrate, 0.1 M Tris, pH 7.0, with 0.2 M LiSO4. Crystals were cryo-protected by direct transfer into reservoir solution containing 15% (w/v) glycerol, and were flash frozen in liquid nitrogen. Data were collected at the Advanced Photon Source (Argonne, IL) beamlines 23ID-D and 23ID-B, or the Cornell High Energy Synchrotron Source (CHESS) beamline F2, and were processed using HKL2000 (Otwinowski and Minor, 1997).
Experimental phase information was obtained for Kcc4p-KA1 using data collected from the SeMet-containing Kcc4p-KA1/L934M crystals, with single-wavelength anomalous diffraction (SAD) methods implemented in SHELX C/D/E (Schneider and Sheldrick, 2002). The resulting experimentally phased map was excellent, and allowed all but the first eight amino acids (including the His6 tag) to be traced with Coot (Emsley and Cowtan, 2004). The resulting model was used to identify molecular replacement (MR) solutions for datasets obtained with native protein using the program Phaser (CCP4, 1994). For MARK1-KA1, the structure was solved using MR with a search model based on the mouse MARK3-KA1 domain NMR structure (PDB ID 1UL7) (Tochio et al., 2006), using Phaser (CCP4, 1994). Model building employed Coot (Emsley and Cowtan, 2004), following each round of refinement using Refmac (CCP4, 1994). Data collection and refinement statistics are presented in Table S3. Structure figures were generated using PyMol (DeLano, 2002).
We thank members of the Lemmon, Ferguson, and Bi laboratories, Ben Black, Jim Shorter and Greg Van Duyne for constructive comments. Erfei Bi, Scott Emr and Daryll DeWald provided yeast strains used in this study. Crystallographic data were collected in part at the GM/CA Collaborative Access Team at the Advanced Photon Source (APS), funded by NCI (Y1-CO-1020) and NIGMS (Y1-GM-1104). Use of APS was supported by the U.S. Department of Energy, under contract No. W-31-109-ENG-38. Additional crystallographic data were collected at beamline F2 at the Cornell High Energy Synchrotron Source (CHESS), supported by NIGMS and the NSF (under award DMR-0225180), using the Macromolecular Diffraction at CHESS (MacCHESS) facility, supported by the NIH (award RR-01646). This work was funded in part by NIH grant R01-GM056846 (to M.A.L.), and a predoctoral fellowship from the American Heart Association Great Rivers Affiliate (K.M.). Coordinates and structure factors have been deposited in the Protein Data Bank (www.rcsb.org/pdb) with identification numbers 3OSE (MARK1-KA1), 3OSM (Kcc4p-KA1 with bound tartrate), and 3OST (Kcc4p-KA1 with bound sulfates).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.