Recognition of physiological phosphorylation sites by p21-activated kinase 4

Abstract

Many serine/threonine protein kinases discriminate between serine and threonine substrates as a filter to control signaling output. Among these, the p21-activated kinase (PAK) group strongly favors phosphorylation of Ser over Thr residues. PAK4, a group II PAK, almost exclusively phosphorylates its substrates on serine residues. The only well documented exception is LIM domain kinase 1 (LIMK1), which is phosphorylated on an activation loop threonine (Thr508) to promote its catalytic activity. To understand the molecular and kinetic basis for PAK4 substrate selectivity we compared its mode of recognition of LIMK1 (Thr508) with that of a known serine substrate, β-catenin (Ser675). We determined X-ray crystal structures of PAK4 in complex with synthetic peptides corresponding to its phosphorylation sites in LIMK1 and β-catenin to 1.9 Å and 2.2 Å resolution, respectively. We found that the PAK4 DFG+1 residue, a key determinant of phosphoacceptor preference, adopts a sub-optimal orientation when bound to LIMK1 compared to β-catenin. In peptide kinase activity assays, we find that phosphoacceptor identity impacts catalytic efficiency but does not affect the K_m value for both phosphorylation sites. Although catalytic efficiency of wild-type LIMK1 and β-catenin are equivalent, T508S mutation of LIMK1 creates a highly efficient substrate. These results suggest suboptimal phosphorylation of LIMK1 as a mechanism for controlling the dynamics of substrate phosphorylation by PAK4.

Graphical Abstract

Introduction

For protein kinases to correctly propagate signals, they must selectively target their downstream substrates. This can occur through an array of mechanisms including co-localization, docking interactions (Ubersax and Ferrell, 2007), and direct binding of the kinase catalytic domain to the substrate (Miller and Turk, 2018). Among these interactions, short linear motifs comprising amino acids that flank the substrate residue can significantly contribute to selectivity for both serine/threonine and tyrosine kinases (Miller and Turk, 2018; Turk, 2008). The phosphoacceptor residue itself is also a key component of substrate specificity; for example, tyrosine kinases are structurally distinct from serine/threonine kinases (Taylor et al., 1995; Ubersax and Ferrell, 2007). Similarly, serine/threonine kinases often discriminate between serine and threonine residues, which is driven by the residue immediately following the conserved activation segment DFG motif, termed the DFG+1 residue (Chen et al., 2014). While most protein kinases display strong phosphorylation site preferences in vitro, many bona fide protein substrates are phosphorylated at sites would appear to be disfavored by the kinase (Chen et al., 2014; Pinna and Ruzzene, 1996). The mechanisms of this apparent disfavored phosphorylation are not generally well understood, so in this study we sought to understand an important disfavored kinase-substrate pair.

The p21-activated kinases (PAKs), are members of the sterile-20 family of serine-threonine kinases and are major downstream effectors of the Rho family of small GTPases (Arias-Romero and Chernoff, 2008; Ha et al., 2015; Hofmann et al., 2004; King et al., 2014; Kumar et al., 2017; Wells and Jones, 2010). The six members of the PAK group are classified by their domain organization into two sub-groups, the type I (PAK1-3) and type II (PAK4-6) PAKs, all of which play roles in cytoskeletal remodeling, cell motility, inhibition of apoptosis and transcription regulation (Ha et al., 2015; Kumar et al., 2017; Li et al., 2012). Of the type II PAKs, PAK4 is the best studied and has demonstrated roles in embryonic and neural development (Dart and Wells, 2013; Wells et al., 2010); its overexpression is also linked to tumorigenesis and metastasis, particularly in prostate and non-small cell lung cancers (Ahmed et al., 2008; Cai et al., 2015; Ha et al., 2015; Lu et al., 2017; Minden, 2012; Thillai et al., 2018; Wells et al., 2010).

All six PAKs strongly favor phosphorylation of serine over threonine, with type II PAKs including PAK4 appearing more selective than type I PAKs (Rennefahrt et al., 2007). Although PAK4 has multiple validated direct substrates (Table 1) (Bompard et al., 2010; Callow et al., 2002; Dan et al., 2001; Guo et al., 2014; Li et al., 2012; Li et al., 2010; Wells et al., 2010; Xu et al., 2016; Zhao et al., 2017; Zhuang et al., 2015), only one of these substrates is phosphorylated on a threonine residue, Thr508 of LIM domain kinase 1 (LIMK1) (Dan et al., 2001). This residue is phosphorylated by PAK4 (and other Rho-effector kinases) (Bernard, 2007; Scott and Olson, 2007), and its phosphorylation is required for full activation of LIMK1, which in turn phosphorylates and inactivates cofilin proteins - actin binding proteins that facilitate actin filament severing (Dan et al., 2001; Hamill et al., 2016). PAK4 therefore has an unusual exception to its strong preference for phosphorylation of serine substrates. The molecular mechanisms for this exception are unclear, especially because previous studies suggested the large and hydrophobic phenylalanine residue at PAK4’s DFG+1 position should strongly disfavor phosphorylation of LIMK1-Thr508 (Chen et al., 2014).

Table 1. Documented human PAK4 substrate sequences.

PAK4 typically phosphorylates Ser residues, but LIMK1 is the only known substrate that PAK4 phosphorylates at a Thr residue. BAD S75 is commonly referred to as S112 based on mouse nomenclature. Primary citations are listed (Bompard et al., 2010; Callow et al., 2005, Callow et al., 2002; Cammarano et al., 2005; Dan et al., 2001; Gnesutta et al., 2001; Guo et al., 2014; Li et al., 2012; Li et al., 2010; Wong et al., 2010; Xu et al., 2016; Zhao et al., 2017; Zhuang et al., 2015).

PAK4 Substrate	Phosphosite Sequence	Phosphosite	Citation
PAK4	KEVPRRKSLVGTPYW	S474	Callow et al. (2002)
β-catenin	QDYKKRLSVELTSSL	S675	Li et al. (2012)
GEF-H1 (ARHGEF2)	PVDPRRRSLPAGDAL	S810	Callow et al. (2005)
Paxillin	ELDELMASLSDFKFM	S272	Wells et al. (2010)
Estrogen Receptor- α	IKRSKKNSLALSLTA	S305	Zhuang et al. (2015)
N-WASP	KRSKAIHSSDEDEDE	S484,	Zhao et al. (2017)
	RSKAIHSSDEDEDED	S485
P53	DRNTFRHSVVVPYEP	S215	Xu et al. (2016)
Ran	DRKVKAKSIVFHRKK	S135	Bompard et al. (2010)
STMN2	KQINKRASGQAFELI	S50	Guo et al. (2014)
BAD	EIRSRHSSYPAGTED	S75	Gnesutta et al (2001)
Raf-1	RPRGQRDSSYYWEIE	S338	Cammarano et al. (2005)
p120-catenin	GLEDDQRSMGYDDLD	S288	Wong et al. (2010)
Integrin β5	REFAKFQSERSRARY	S759	Li et al. (2010)
	AKFQSERSRARYEMA	S762
LIMK1	PDRKKRYTVVGNPYW	T508	Dan et al. (2001)

To understand the mechanisms of PAK4’s atypical phosphorylation of LIMK1, we conducted a structural and functional study. We determined a 1.9 Å crystal structure of PAK4 catalytic domain in complex with a synthesized peptide corresponding to LIMK1’s substrate sequence (Thr508), and we compared this to a 2.2 Å crystal structure of PAK4 catalytic domain in complex with a synthesized peptide corresponding to a well characterized serine substrate of PAK4, β-catenin (Ser675). We find that similar to optimized ‘PAKtide’ substrates (Chen et al., 2014), the DFG+1 residue is re-oriented to accommodate the unusual Thr phosphoacceptor. We then assessed the kinetics of kinase activity for PAK4 against LIMK and β-catenin substrate peptides with either Ser or Thr as the phosphoacceptor residue. We find that phosphoacceptor identity increases catalytic efficiency but does not affect K_m, and although catalytic efficiency of wild-type LIMK1 and β-catenin are equivalent, T508S mutation of the LIMK1 peptide creates a highly efficient substrate.

Materials and Methods

Protein Expression and Purification

PAK4 expression and purification was conducted as previously described (Chen et al., 2014; Ha et al., 2012; Zhang et al., 2018). The catalytic domain of PAK4 (residues 109-426) (UniProt ID: O96013-2) was expressed using a modified pET28 vector with hexa-histidine (6xHis) tag, removable by TEV protease. Following nickel-affinity chromatography using a HisTrap chelating column (GE Healthcare) and gel filtration using a Superdex 200 10/300GL (GE Healthcare), purified PAK4 catalytic domain was concentrated to 4 mg/mL for co-crystallization, or to 0.98 μg/mL for kinase activity assays.

Peptide Synthesis.

Peptides for crystallization and kinase assays were synthesized on the 0.1 mM scale with N-terminal acetyl and C-terminal amide and HPLC purified at Tufts University Core Facility. Peptides were resuspended in distilled water and concentration monitored using a Nanodrop spectrophotometer (GE). For kinase assays, peptides of equal length were used for LIMK1 and β-catenin: LIMK1 peptide (RKKRYTVVGN), LIMK1^T508S (RKKRYSVVGN), β-catenin (YKKRLSVELT), β-catenin^S6757 (YKKRLTVELT). Peptides correspond to residues 503 to 512 of LIMK1 (UniProt ID: P53667) and residues 670-679 of β-catenin (UniProt UD: P35222). The PAKtide peptide was one amino acid residue longer at the C-terminus (RKRRNSLAYKK). For structure determination, a shorter β-catenin peptide (KKRLSVE) was used.

Structure determination

PAK4 catalytic domain crystals were grown against 0.1 M Tris-HCL (pH 7.5) and 2 M sodium acetate at room temperature using the hanging drop vapor diffusion method (well: 500 μL; drop: 1 μL protein mixed with 1 μL precipitant). Prior to mixture with precipitant, the protein solution contained 1 mM AMP-PNP and 5 mM MgCl₂. Following crystal growth, 0.2 μL of peptide solution was added to a final concentration of 2 mM LIMK1 peptide, or a final concentration of 5 mM β-catenin peptide. Soaking was conducted for 1 day at room temperature. Crystals were looped and cryoprotected in 3 M sodium acetate. Crystallographic data were collected on beamline 24-ID-E of the Advanced Photon Source (APS) and were processed using the HKL-2000 package (Otwinowski and Minor, 1997) and molecular replacement using Phaser (McCoy et al., 2007) to obtain an initial solution with a previous structure of PAK4 alone as the search model (PDB ID: 4FIJ (Ha et al., 2012)). Iterative refinement and model building were conducted using the Phenix package (Adams et al., 2010) and Coot (Emsley et al., 2010). For both structures, after refinement of the kinase domain alone, clear unbiased difference density was observed in the substrate binding site allowing bound peptide to be built. In both structures, there was insufficient electron density to build bound AMP-PNP. Final placement of peptide position was validated using simulated annealing omit maps. Model validation was conducted using MolProbity (Chen et al., 2010).

Kinase activity assays

PAK4 activity was determined by a filter binding assay performed in triplicate using radiolabeled [γ-³²P]ATP, modified from (Chen et al., 2014; Hastie et al., 2006). PAK4 kinase domain (25 nM) was incubated with peptides at varying concentrations in kinase buffer (50 mM HEPES, pH 7.4, 10 mM MgCl₂, 12.5 mM NaCl, 1 mM MnCl₂, 1 mM DTT, 0.1 mM EGTA). Peptide concentrations were titrated in two-fold increments across the following concentration ranges: LIMK1, 25 μM - 1600 μM; LIMK1^T508S 2.5 μM - 320 μM; β-catenin, 56.3 μM - 3600 μM; β-catenin^S675T, 56.3 μM - 1800 μM; PAKtide, 6.25 μM - 400 μM (Miller et al., 2019). Reactions were initiated by addition of ATP to reach a final concentration of 20 μM with 0.025 μCi/μL [γ-³²P]ATP incubated at 30°C for 10 min and 20 min. Aliquots of each reaction mixture (5 μL) were pipetted onto P81 phosphocellulose filter paper that was then quenched in 75 mM phosphoric acid. Filters were washed twice for 5 min in 75 mM phosphoric acid, washed once briefly in acetone, air-dried, suspended in 10 mL Optifluor scintillation fluid, and analyzed with a liquid scintillation counter. The concentration of product formed was calculated using standards made from unwashed filters. Initial reaction velocities were found by fitting the [product] vs time data to a line with Excel (Microsoft). Catalytic constants were calculated by fitting the initial reaction velocities to the Michaelis-Menten equation using GraphPad Prism (version 8).

Results

To date, no structures have been determined of PAK family members in complex with regions of physiologically relevant substrates. Almost all type II PAK substrates are phosphorylated on a serine residue, with the sole exception of LIM domain kinase 1, which is phosphorylated on its activation loop threonine (Thr508) (Table 1, Figure 1A). This exception to the strong preference of type II PAKs towards serine as a substrate caused us to wonder whether there is a structural basis for this difference. We therefore determined crystal structures of the PAK4 catalytic domain in complex with synthetic peptide substrates derived from its physiological substrates LIMK1 and β-catenin.

Figure 1. Crystal structures of PAK4 with LIMK1 and β-catenin peptides.

A) Schematic of PAK4 and its known substrates grouped according to substrate residue. B) Crystal structure of PAK4 catalytic domain (light yellow cartoon) in complex with LIMK1 peptide (red sticks). N- and C-lobes of the kinase are indicated. C) Close up showing 2F_obs-F_calc electron density contoured at 1σ for LIMK1 peptide. D) Crystal structure of PAK4 catalytic domain (light yellow cartoon) in complex with β-catenin peptide (blue sticks). N- and C-lobes of the kinase are indicated. E) Close up showing 2F_obs-F_calc electron density contoured at 1σ for β-catenin peptide. Images generated with CCP4mg (McNicholas et al., 2011).

The crystal structure of PAK4 catalytic domain (residues 109-426) in complex with the LIMK1 peptide substrate (RKKRYT⁵⁰⁸VVGN) was determined to 1.9 Å resolution (Table 2), with good electron density observed for the bound substrate (Figure 1B,C). Similarly, the crystal structure of PAK4 catalytic domain in complex with a synthetic peptide corresponding to its major phosphorylation site in β-catenin (KKRLS⁶⁹⁵VE) was determined to 2.2 Å resolution (Table 2) with good electron density observed for the bound substrate (Figure 1D,E). Both structures exhibit the typical two lobe kinase fold, consisting of an N-lobe with five β-sheets and one α-helix and a C-lobe with mostly α-helices (Figure 1B,D). PAK4 in both structures is in the active state ‘DFG-in’ conformation, and the activation loop is phosphorylated on Ser474, similarly to other E.coli expressed PAK4 crystal structures (Baskaran et al., 2015; Chen et al., 2014; Ha and Boggon, 2018; Ha et al., 2012; Zhang et al., 2018). Root-mean-deviation between PAK4 in the structures is 0.2 Å over 292 residues (McNicholas et al., 2011).

Table 2. Data collection and refinement statistics.

Complex	PAK4 with LIMK1 peptide	PAK4 with β-catenin peptide
PDB accession code	6WLY	6WLX
Data collection
Space group	P41212	P41212

PAK4 Substrate	Phosphosite Sequence	Phosphosite	Citation
PAK4	KEVPRRKSLVGTPYW	S474	Callow et al. (2002)
β-catenin	QDYKKRLSVELTSSL	S675	Li et al. (2012)
GEF-H1 (ARHGEF2)	PVDPRRRSLPAGDAL	S810	Callow et al. (2005)
Paxillin	ELDELMASLSDFKFM	S272	Wells et al. (2010)
Estrogen Receptor- α	IKRSKKNSLALSLTA	S305	Zhuang et al. (2015)
N-WASP	KRSKAIHSSDEDEDE RSKAIHSSDEDEDED	S484, S485	Zhao et al. (2017)
p53	DRNTFRHSVVVPYEP	S215	Xu et al. (2016)
Ran	DRKVKAKSIVFHRKK	S135	Bompard et al. (2010)
STMN2	KQINKRASGQAFELI	S50	Guo et al. (2014)
BAD	EIRSRHSSYPAGTED	S75	Gnesutta et al (2001)
Raf-1	RPRGQRDSSYYWEIE	S338	Cammarano et al. (2005)
p120-catenin	GLEDDQRSMGYDDLD	S288	Wong et al. (2010)
Integrin β5	REFAKFQSERSRARY AKFQSERSRARYEMA	S759 S762	Li et al. (2010)
LIMK1	PDRKKRYTVVGNPYW	T508	Dan et al. (2001)
X-ray source	APS NE-CAT-E		APS NE-CAT-E
Detector	Eiger 16M PAD		Eiger 16M PAD
Wavelength (Å)	0.97918		0.97918
Unit cell
a, b, c, (Å)	61.6, 61.6, 179.6		61.6, 61.6, 179.4
α, β, γ, (°)	90.0, 90.0, 90.0		90.0, 90.0, 90.0
Resolution range (Å)*	50.00-1.90 (1.97-1.90)		50.00-2.20 (2.28-2.20)
No. of unique reflections	28404 (2747)		18577 (1799)
Multiplicity*	20.3 (15.3)		11.2 (11.8)
Completeness (%)*	99.9 (99.9)		99.8 (100)
Mean I/σ(I)*	20.8 (1.8)		20.7 (2.0)
Wilson B factor (Å2)	38.5		47.5
Rpim (%)*	2.7 (38.6)		3.2 (34.7)
CC 1/2*	0.999 (0.615)		0.977 (0.701)
Refinement statistics
Reflections used in refinement	28228 (2621)	18366 (2419)
Resolution range* (Å)	43.58-1.90 (1.97-1.90)		44.85-2.20 (2.32-2.20)
Rwork (%)*	18.6 (31.5)		19.7 (29.5)
Rfree (%)*	20.5 (32.1)		22.9 (35.2)
Free R reflections (%)*	5.0 (5.0)		5.0 (5.0)
Free R reflections, no.*	1412 (138)		919 (88)
Residue range built	PAK4 / 299-590, LIMK1 / 504-512		PAK4 / 299-590, β-catenin / 672-676
Soaked peptide / Built peptide	RKKRYTVVGN / KKRYTVVGN		KKRLSVE / KRLSV
No. water molecules	135		54
Mean B factor, Å2
All atoms	45.9		57.42
PAK4 (chain A)	45.1		57.16
Peptide (chain B)	66.3		82.06
H2O	47.7		49.88
Model statistics
RMSD bond lengths (Å)	0.005		0.010
RMSD bond angles (°)	0.776		1.094
Ramachandran plot: favored/allowed/outliers (%)	98.6 / 1.4 / 0.00		97.59 / 2.41 / 0.00
MolProbity clashscore	3.21 (98th percentile)		5.24 (98th percentile)

^*Parentheses indicate highest resolution shell.

Crystallographic data and refinement statistics indicate good quality diffraction data and refined crystal structures.

The mode of peptide binding in both structures is broadly similar. Each peptide is posed in the canonical kinase peptide substrate binding site (Figure 1B,D), similar to other PAK-substrate or PAK-pseudosubstrate structures (Baskaran et al., 2015; Chen et al., 2014; Ha and Boggon, 2018; Ha et al., 2012), and the positioning of the substrate residues is broadly conserved (Figure 2A). Concordantly, both peptides share similar peptide backbone conformations, indicating that the identity of the phosphoacceptor residue does not significantly impact peptide backbone conformation (Figure 2B,C). A common feature of the previously determined PAK-substrate and PAK-pseudosubstrate structures is PAK recognition of a basic residue N-terminal to the phosphorylation site (Ha et al., 2015; Rennefahrt et al., 2007). In PAK4 this is achieved through an acidic cradle defined by residues Asp444 and Glu507 (Ha et al., 2015; Rennefahrt et al., 2007). For both PAK4-LIMK1 and PAK4-β-catenin, the acidic cradle makes salt bridge interactions with the headgroup of an Arg residue two residues N-terminal to the phosphorylation site (the P-2 Arg) (Figure 2D,E).

Figure 2. Details of the PAK4-peptide interactions.

A) Sequences of LIMK and β-catenin peptides showing the linear motif position. B,C) Close ups showing the LIMK1 peptide bound to PAK4 (B) and β-catenin (C) with the side-chain of Phe461 shown as spheres. D,E) Extent of the bound peptide interactions showing hydrogen bonds and residues discussed in the text.

The phosphoacceptor residue hydroxyl group for both LIMK1 and β-catenin peptide substrates is poised for phosphotransfer, although nucleotide is not visible in either structure. As with synthetic optimized peptides (Chen et al., 2014), the consequences of correct orientation of the phosphoacceptor residue are structurally observed in the orientation of the residue immediately C-terminal to the conserved DFG motif, the DFG+1 residue (Phe461). In the β-catenin bound structure, the phenyl ring is oriented approximately perpendicular to the peptide backbone, but in the LIMK1 bound structure Phe461 rotates to accommodate the methyl group on Thr508 (Figure 2B,C). Similar orientation changes were observed for peptides optimized for PAK phosphorylation (PAKtides) but have not previously been observed for peptides corresponding to physiologically relevant substrates. These structures therefore illustrate that peptides derived from physiological substrates follow similar recognition rules to those observed for optimized substrates (Chen et al., 2014).

To assess the enzymatic impact of the unusual threonine phosphoacceptor in LIM kinases compared to the strong preference of PAK4 for serine, we conducted kinase activity assays. Using radiolabeled [γ-³²P]ATP we determined Michaelis-Menten kinetics and obtained K_m and k_cat values for PAK4 catalytic domain phosphorylation of 5 distinct peptides (Figure 3, Table 3). We first compared phosphorylation of synthetic peptides corresponding to β-catenin (YKKRLS⁶⁷⁵VELT) and LIMK1 (RKKRYT⁵⁰⁸VVGN). We found that unexpectedly the catalytic efficiency for PAK4 phosphorylation of the LIMK1 and β-catenin peptides was broadly equivalent (k_cat/K_m ≈ 300) (Figure 3A, inset i, Table 3).

Figure 3. PAK4 kinase assay results.

A) Michaelis-Menten best-fit equations for all five peptides plotted with mean reaction velocities. Insets: (i) Comparison of β-catenin and LIMK1 curves. (ii) Comparison of LIMK1^T508S and LIMK1 curves. (iii) Comparison of β-catenin and β-catenin^S675T curves. (iv) Comparison of LIMK1^T508S and optimized PAKtide curves. B) Michaelis constant, K_m, values for all five peptides. C) Maximum reaction velocities for all five peptides. All error bars are standard deviations.

Table 3. Catalytic constants for phosphorylation of peptide substrates by PAK4 and exchange mutants.

Underlined residue is the phosphoacceptor of each peptide. Data show the Michaelis-Menten non-linear regression best-fit values from three kinase assays with standard deviation indicated. Welch’s unpaired t-test (two-tailed) was used to calculate significance; all pairs are significant (p < 0.05) except those indicated with the same symbolic superscript (0.1 < p > 0.05), or with an alphabetical superscript (p > 0.1). 95% confidence interval (CI) shown for K_m.

Peptide	Sequence	kcat (S−1)	Km (μM) (95% CI)	kcat/Km (M−1S−1)
LIMK1	RKKRYTVVGN	0.038 ± 0.004	150 ± 60a* (70 to 320)	300 ± 100+
LIMK1T508S	RKKRYSVVGN	0.29 ± 0.04*	60 ± 20ab+ (25 to 135)	5000 ± 2000*
β-catenin	YKKRLSVELT	0.15 ± 0.02a	600 ± 300c* (260 to 1400)	300 ±100+
β-cateninS675T	YKKRLTVELT	0.010 ± 0.003	1100 ± 600ac+# (400 to 5300)	9 ± 6
PAKtide	RKRRNSLAYKK	0.20 ± 0.02a*	60 ± 20ab# (25 to 125)	4000 ± 1000*

We then assessed the impact of switching the phosphorylation site residue for both β-catenin and LIMK1 derived peptides. For β-catenin we generated a Thr-substrate peptide (YKKRLT⁶⁷⁵VELT) which we term β-catenin^S675T, and for LIMK1 we generated a Ser-substrate peptide (RKKRYS⁵⁰⁸VVGN) which we term LIMK1^T508S. Comparison of the kinetics for β-catenin and β-catenin^S675T indicates the K_m values are not significantly different, but the k_cat and catalytic efficiency are reduced by approximately an order of magnitude for the non-wild type Thr-substrate peptide (Figure 3A, inset ii). Likewise, comparison of the kinetics for LIMK1 and LIMK1^T508S indicates that a Thr-substrate is disfavored. Again, we find the K_m values are not significantly different but k_cat and catalytic efficiency increase by approximately an order of magnitude for the non-wild type Ser-substrate peptide (Figure 3A, inset iii). These results demonstrate that Thr as a substrate residue is highly disfavored, primarily by reduction of catalytic efficiency.

Lastly, we compared the catalytic efficiency of LIMK1^T508S with an optimized PAKtide peptide (RKRRNSLAYKK) that was previously designed based on linear peptide motif analysis (Miller et al., 2019). We find that catalytic efficiency of LIMK1^T508S is broadly similar to the optimized PAKtide peptide (Figure 3A, inset iv). This suggests that LIM kinase with a serine phosphorylation site on its activation loop would represent a highly efficient substrate for PAK4.

Discussion

PAK4 phosphorylates all of its known substrates on serine with the sole known exception of LIM domain kinase 1, which it phosphorylates on threonine. In this study we sought to understand the molecular and enzymatic reasons for these exceptions by studying PAK4 phosphorylation of a well-known serine substrate, β-catenin, and comparing this to its unusual substrate, LIMK1. We reveal the molecular basis for selectivity by showing that the DFG+1 residue is important for discriminating between serine and threonine phosphoacceptors, but LIMK1 phosphorylation by PAK4 proceeds at a similar catalytic efficiency as β-catenin. We also assessed the impact of a LIMK1 optimizing mutation, T508S, and find that this yields a good PAK4 substrate with catalytic efficiency equivalent to a designed PAKtide peptide. These studies therefore provide new insights into PAK phosphorylation of physiologically relevant substrates.

At the molecular level, we find that PAK4’s interaction with physiologically relevant Ser/Thr phosphoacceptors seems to be associated with its differences in enzymatic activity towards these substrates. The DFG+1 residue, immediately C-terminal to the conserved DFG motif, was previously found in conformations that seem to depend on the identity of the substrate residue. For designed peptide substrates that are optimized as PAKtide peptides, a mutation from Ser to Thr as the phosphoacceptor results in reorientation of Phe461 around its β-γ bond. This is thought to interfere with placement of ATP γ-phosphate and to hinder phosphate transfer (Chen et al., 2014). We show that similar conformational movements in the DFG+1 residue occur for the physiologically relevant substrates β-catenin and LIMK1, indicating that molecular level discrimination between Ser and Thr substrates occurs at the DFG+1 residue for these substrates.

At the enzymatic level, our comparison of PAK4’s physiological substrates shows that despite possessing a Thr phosphoacceptor residue LIMK1 is a favorable substrate phosphorylated by PAK4 with similar catalytic efficiency to β-catenin, a Ser phosphoacceptor substrate. It is not immediately clear which residues of the linear motif drive the tighter K_m observed for LIMK1 compared to β-catenin (Table 3). Nevertheless, recent studies have shown that interactions between the kinase β3-αC loop and substrate residues C-terminal to the phosphoacceptor are important for controlling catalytic efficiency (Miller et al., 2019), implying that Val-Gly at positions +2 and +3 in LIMK1 are more preferable than Glu-Leu in β-catenin. This position preference is further supported by peptide array profiling (Rennefahrt et al., 2007).

The effect of switching the phosphoacceptor residue in both the β-catenin and LIMK1 peptide yields results of similar magnitude. For β-catenin, mutation of the phosphoacceptor Ser to Thr results in an approximately 10-fold reduction in k_cat and commensurate change in catalytic efficiency, and similarly, for LIMK1, mutation of the phosphoacceptor Thr to Ser results in an approximately 10-fold increase in k_cat and change in catalytic efficiency. For LIMK1, however, this increase in catalytic efficiency results in a peptide that compares favorably with a substrate peptide designed to be optimal for phosphorylation by PAKs based on peptide array profiling (Miller et al., 2019; Rennefahrt et al., 2007). Compared to PAK2, PAK4’s phosphorylation of a LIMK1 peptide has a low k_cat (Wu and Wang, 2003). This may be because PAK4, unlike PAK2, has a noncanonical KPEN motif containing a Ser instead of an Asn residue (Taylor et al., 1992). Mutation of this Ser to an Asn increases PAK4 activity (Miller et al., 2019; Qu et al., 2001) implying that the noncanonical KPEN motif hinders PAK4 activity. It is interesting to hypothesize that the weak kinase activity of PAK4 may increase the importance of substrate recognition for allowing efficient catalysis and consequent downstream signaling.

The presence of a Thr phosphoacceptor site on LIMKs is highly conserved evolutionarily, and many of their upstream kinases preferentially phosphorylate threonine over serine, including the ROCK and MRCK families (Amano et al., 2001; Maekawa et al., 1999; Ohashi et al., 2000; Sumi et al., 2001b). Although our study does not establish disfavor for the PAK-LIMK pathway in a cellular context, in other systems the presence of a suboptimal phosphoacceptor residue confers sensitivity to inhibitors and tempers response to activating stimuli (Kang et al., 2013). It is tempting to speculate on the impact of optimizing PAK-LIMK by altering LIM kinase phosphoacceptor identity and tuning phosphosite specificity. This may allow simultaneous enhancement of PAK-driven signals to the cytoskeleton from RAC/CDC42 and suppression of ROCK-driven signals from RHO. If, however, LIMK1 requires Thr in its activation loop for full activity, substituting of the phosphoacceptor residue for Ser could negatively impact its function (Kaldis et al., 2000). Considering LIMK1’s and LIMK2’s distinct subcellular localization (Acevedo et al., 2006; Sumi et al., 2006) and that ROCK phosphorylates LIMK2 (Sumi et al., 2001a), the possibility that LIMK1 and LIMK2 have distinct upstream pathways remains open, and we note that while PAK phosphorylation of LIMK1 is well-established (Dan et al., 2001; Edwards et al., 1999), to date, PAK phosphorylation of LIMK2 has not been directly shown.

In summary, we find that PAK4 phosphorylation of peptide substrates is significantly impacted by the identity of the phosphoacceptor residue, but both β-catenin and LIMK1 fall into a similar range of catalytic efficiency. Although catalytic efficiency of wild-type LIMK1 and β-catenin are equivalent, T508S mutation of LIMK1 creates a highly efficient substrate suggesting that inefficient phosphorylation of LIMK1 is a mechanism for controlling normal substrate dynamics.

Highlights.

• Molecular basis for PAK4 phosphorylation of physiological substrates

• Phosphoacceptor identity impacts catalytic efficiency but does not affect the K_m

• Suggests suboptimal phosphorylation of LIMK1 as a mechanism for controlling the dynamics of substrate phosphorylation by PAK4