The crystal structures of PSD95 GK and peptide binders were downloaded from the RCSB Protein Data Bank (PSD95 GK/p-LGL2a: 3WP0, PSD95 GK/p-LGL2b: 3WP1, PSD95 GK/p-SAPAP1: 5YPO, PSD95 GK/MAP1A: 5GNV, PSD95 GK/QSF: 5YPR). The water molecules in the complex were deleted, and then the missing amino acid residues and side chain bumps were fixed by the prepared protein structure module in Sybyl-X 2.0 (Tripos International, 2012). Then, the pK values of the residues were predicted and protonized by H++ server (http://biophysics.cs.vt.edu/) (Anandakrishnan et al., 2012). The structure of the mutated peptides was constructed by substituting the sequence of QSF and energy minimized by Sybyl-X 2.0. Then, the molecular dynamics simulations were performed for the complex of PSD95 GK and various peptide binders.

The detailed MD methods were described in our previous work (Shan et al., 2019; Li et al., 2020). In brief, the simulations were performed by the AMBER 2020 packages with ff14SB force field (Maier et al., 2015; Case et al., 2020). Water molecules were treated with the TIP3P water model (Price and Brooks, 2004). The AMBER parameters of phosphorylated serine were obtained from the AMBER parameter database (http://research.bmh.manchester.ac.uk/bryce/amber) (Homeyer et al., 2006). First, three minimization stages were carried out for the systems. Then, each system was gradually heated from 0 to 300 K using the Langevin thermostat during the heating stage and maintained at 300 K during the following equilibrium and production stages.

To calculate the binding free energies of PSD95 GK with the ligands, MD simulations were performed using the aforesaid MD protocol until the systems reached equilibrium. The binding free energies were calculated using the MMPBSA. py implemented in AMBER 2020 (Miller et al., 2012). Totally, 100 snapshots were extracted from the equilibrium trajectory for MM-GBSA free energy calculation. The binding free energy (ΔG _bind ) in MM-GBSA between the receptor and ligand is summarized as Equation 1 (Tsui and Case, 2000; Gohlke and Case, 2004). Per-residue energy decomposition was also performed to evaluate the energy contribution of each residue in the systems. All the other parameters were kept as the default value. For per-residue energy decomposition, the binding free energy (ΔG _bind ) is divided into three parts as shown in Equation 2, ΔE _mm denotes the complex gas-phase interaction energy, which comprises van der Waals (ΔE _vdw ) and electrostatic (ΔE _elec ) interactions and the internal energy variation ( Δ E i n t r a ) (Equation 3). ΔG _solv represents the difference of the solvation free energy, which is divided into the electrostatic solvation free energy (ΔG _GB ) and nonpolar contributions (ΔG _SA ). -TΔS represents the contribution of the entropy of the solute molecules. Since the calculation of the entropy is computationally expensive for large systems and tends to introduce low-accuracy approximations, -TΔS is not considered in the present work (Miller et al., 2012). Δ G b i n d = G c o m p l e x − ( G r e c e p t o r + G l i g a n d ) . (1) Δ G b i n d = Δ E m m + Δ G s o l v − T Δ S . (2) Δ E m m = Δ E v d w + Δ E e l e c + Δ E i n t r a . (3) Δ G s o l v = Δ G G B + Δ G S A . (4)

The PSD95 GK domain (a.a. 531-713) was cloned into a modified pET-15b vector with N-terminal His6-tag. The construct was expressed in BL21 (DE3) Escherichia coli cells and induced by 0.2 M isopropyl-β-D-thiogalactoside (IPTG) for 18 h at 16°C. His6-tagged protein was first purified using Ni-NTA agarose affinity chromatography (GE Healthcare), and PSD95 GK was then purified by size-exclusion chromatography and then further purified by size-exclusion chromatography (Superdex-200 26/60, GE Healthcare) in the buffer containing 50 mM Tris pH 8.0, 100 mM NaCl, 1 mM EDTA, and 1 mM DTT. The designed peptide inhibitors were commercially synthesized (China Peptides).

Fluorescence polarization assay was performed in 384-well plates (United States Corning) using a SpectraMax Paradigm (Molecular Devices). The FITC-labeled fluorescent peptide (FITC-AARRE (pS)YLKATQ, FITC-SAPAP) and free-labeled peptide (AARRE (pS)YLKATQ) were ordered from China Peptides. The assay buffer contained 50 mM Tris and 100 mM NaCl at pH 7.5. The peptide inhibitors were serially diluted in assay buffer at a top concentration sufficient to yield an 8-point dose–response curve. For each assay, FITC-SAPAP (10 nM) and PSD95 GK protein (2 μM) were added to a final volume of 20 μl in the assay buffer. The polarization values were measured after a 1-h incubation using a SpectraMax Paradigm (Molecular Devices) with excitation and emission wavelengths of 485 and 535 nm, respectively. The FP value was quantified as the difference between the parallel (I _ǁ) and perpendicular (I _┴) emission intensities normalized by the total fluorescence intensity of the emission beam (Equation 5) (Hall et al., 2016). To accomplish the different sensitivity for measuring emission intensities from the I _ǁ and I _┴ channels, G-factor is often used to correct for any bias toward or against the I _┴ channel (Equation 6). The dissociation constant (K _d) for PSD95 GK/SAPAP was determined using a constant concentration of 10 nM of FITC-SAPAP, and a series of increasing concentrations of PSD95 GK varied from 0 to 20 μM. The K _d and IC₅₀ values were determined from the plot using nonlinear least-squares analysis, and K _i values were obtained by the formula that is depicted in the article (Nikolovska-Coleska et al., 2004). FP = I ‖ − I ┴ I ‖ + I ┴ . (5) FP ( mP ) =   ( I ‖ − I ┴ × G ) ( I ‖ + I ┴ × G ) × 1000. (6)

Microscale thermophoresis (MST) assay was conducted to determine the binding affinity of PSD95 GK and the top peptide F10W using the Monolith NT.115. The His-PSD95 GK protein was kept in the PBS buffer at a concentration of 500 nM and then labeled according to the protocol of the protein labeling kit RED-Tris-NTA. The peptide was diluted in PBS buffer containing 0.05% Tween 20 for the final MST assay. Then, the labeled protein was mixed with the same volume of the diluted peptide (10 μl:10 μl) from different concentrations and incubated for 10 min at 37°C in the dark. Binding was measured by monitoring the thermophoresis with 20% LED power on the Monolith NT.115. The data were analyzed using Mo. Affinity Analysis v2.2.4 software.

ITC measurements were carried out on a MicroCal-iTC200 system (Malvern) in a buffer containing 50 mM Tris pH 8.0, 100 mM NaCl, 1 mM EDTA, and 1 mM DTT at 25°C. The concentrations of proteins loaded into the syringe (F10W) and cell (PSD95 GK) were 0.5 and 0.05 mM, respectively. The titration data were analyzed using Origin7.0 from MicalCal and fitted by a one-site binding model.

To investigate the structural features of the peptide binders of PSD95 GK, we accumulated the reported binders and binder affinity data to our knowledge (Supplementary Table S1). These binding partners could be divided into two categories: phosphor-peptides (p-LGL2a, p-LGL2b, p-SAPAP1, and p-LGN) and unphosphorylated binders (MAP1A, MBS domain of KIF13B, QSF, MKL, and DLS). At first, we compared binding affinities between PSD95 GK and its target peptides. As is shown in Supplementary Table S1, phosphor-peptides generally demonstrate more potent binding affinity than unphosphorylated peptides. This is largely due to the substitution of the phosphorylated serine (p-Ser) by Asp or Glu in unphosphorylated peptides which weakened the hydrogen bonding at the phosphor-site formed by R568, R571, and Y609. However, the MBS domain of KIF13B which has 122 residues demonstrates stronger binding affinity than most of the phosphor-peptides resulting from its more extensive binding interface (Zhu et al., 2016). Second, the amino acid sequences of these peptides were further aligned, and the results revealed that the amino acids near the Ser(0) at position 1, 2, 3, and 4 have high amino acid sequence consensus, indicating that they may have the similar binding mode, which is in accordance with the result of our previous study that the GK domain shares a unique recognition mode with the binding partners (Figure 1B) (Li et al., 2020). In addition, the crystal structures of the reported peptides are also displayed in Figure 1A, and it could be found that these peptides all share a short α-helix and two loops at C- or N -terminal. The key residue p-Ser or Asp/Glu is located at the α-helix region.

To elucidate the key residues of the binder peptides at the complex interfaces, we first investigated the interactions between the phosphorylated peptides and PSD95 GK by using MD simulations of the reported crystal structures. The structures were generated by CPPTRAJ program in AMBER 2020 for further analysis after the system reached equilibrium state (Roe and Cheatham, 2013). For the PSD95 GK/p-LGL2a complex, at the beginning of the MD simulation, the N-terminal loops were exposed toward the solvent and moved freely, which may lead to a rather large fluctuation on the RMSF of 8.87Å (Supplementary Figure 1A). However, when the system reached equilibrium after the 35-ns simulation, the loops moved toward the region between the β3 and β4 sheets of GK forming salt bridges between R (-3) at p-LGL2a and D579_GK with the energy contribution of R (-3) to -5.17 kcal/mol, which stabilized the N-terminal loops of the p-LGL2a peptide. The salt bridges were also found between the carboxyl group of F10_p-LGL2a and the guanidine group of R637_GK at α4 helix, which confined the movement of C-terminal loops. The hydrophobic residues such as L1 and L5 also possessed crucial hydrophobic interactions which facilitate the assembly of the complex with a binding energy of –3.74 kcal/mol and –3.88 kcal/mol, respectively.

By comparison, GK/p-SAPAP1 and GK/p-LGL2b are more stable than PSD95 GK/p-LGL2b during MD simulation. After analyzing the binding free energy decomposition, we found that the salt bridges were formed between R2 at p-LGL2b and D545 and D549 of GK and also existed between R (-2) at p-LGL2b and S561 and D579 of GK, which both reinforced the binding affinity in the complex of PSD95 GK/p-LGL2b. In addition, the binding was further strengthened by hydrophobic interactions between F1_p-LGL2b and M4 _p-LGL2b and the residues A601_GK, G602_GK, and Y611_GK. Similarly, for the complex of PSD95 GK/p-SAPAP1, Y1 was found to present the favorite energy contribution of –8.20 kcal/mol, which was inferior to that of p-Ser of −8.60 kcal/mol by forming hydrogen bonds and hydrophobic interactions with T611 from the β6 sheet of the GK domain. The residue R (-3) _p-SAPAP1 at the loops of N-terminal displayed the hydrogen bonds with C562_GK and T611 _GK and formed salt bridges with D629 _GK, which confined the movement of the N-terminal loops of p-SAPAP1. A4_p-SAPAP1 possessed the hydrogen bonds and hydrophobic interactions with Q603_GK, while L9_p-SAPAP1 possessed the hydrogen bonds and hydrophobic interactions with R676_GK.

In brief, the phosphor-Ser (pSer0) from each of the three peptides had critical interactions forming a network of hydrogen bonds with R568, Y580, and Y609 of the phosphor-site, and the hydrophobic interactions from the other amino acids were also indispensable for their binding interaction with PSD95 GK.

MAP1A and QSF are two unphosphorylated peptides that bind to PSD95 GK. We first characterized these interactions. In the PSD95 GK/MAP1A system, D0_MAP1A formed hydrogen bonds with R568_GK, Y580_GK, and Y609_GK. In addition, hydrogen bonds were also excited between G (-8)_MAP1A and D629_GK, Y (-6)_MAP1A and D549_GK, L (-3)_MAP1A and Y580_GK, and R (2) _MAP1A and D549_GK. Especially, the hydrophobic residues Y (-6), P (-4), and L3 displayed the top three energy contributions with the value of -7.89, -5.27, and -4.45 kcal/mol, which is consistent with the reported experiments that the mutation of hydrophobic residues Y (-6), P (-4), or L3 disrupted the interactions between the peptide and PSD95 GK (Xia et al., 2017).

QSF is a synthetic unphosphorylated peptide that integrates the optimal p-SAPAP1 peptide with the C-terminal tails of p-LGN or p-LGL2a (Zhu et al., 2017). The interface between PSD95 GK and QSF was also investigated using MD simulation. The results showed that R2_QSF has the most favorite energy contribution of -11.06 kcal/mol by forming salt bridges with D545_GK and D629_GK. The following key residues are Y1_QSF, I (-4)_QSF, and F (10)_QSF which accommodate the hydrophobic site of GK manifesting the hydrophobic interactions with -6.44, -5.59, and -4.88 kcal/mol, respectively. In addition, Y1_QSF formed the hydrogen bonds with T611_GK which could also be found in PSD95 GK/p-SAPAP1 (Supplementary Figure 3). The residue E (0)_QSF formed hydrogen bonds with R568, R571, and Y609 of the GK phosphor-site with −3.66 kcal/mol. The hydrogen bond was found between Q8_QSF and Q603_GK with −1.46 kcal/mol (Figure 2). While most of the residues possess favorable binding affinity, R (-5) and N6 obviously displayed an unfavorable binding energy contribution of 4.2 kcal/mol and 0.59 kcal/mol, respectively (Figure 1C).

Among unphosphorylated peptides, Asp (MAP1A) or Glu (QSF) mimicking the p-Ser of phosphorylated peptides formed crucial hydrogen bond interactions with R568_GK, Y580_GK, and Y609_GK of the phosphor-site. Similarly, the hydrophobic interactions were also important for the binding interaction of unphosphorylated peptides.

Based on the abovementioned interaction analysis of PSD95 GK and phosphor-/unphosphorylated-peptides, it could be concluded that the binding affinity is mainly dominated by the important hydrogen bonds between p-Ser or Asp/Glu and residues at the phosphor-site of GK, the crucial salt bridges at R (-2)_p-LGL2b, R (-3)_{p-LGL2a, p-SAPAP1}, and R (2) _{p-LGL2b,MAP1A,QSF,} and the indispensable hydrophobic interactions. As for the unphosphorylated QSF peptide possessing relatively low molecular weight and moderate binding affinity, we regarded QSF as the parent peptides for further sequence optimization based on per-residue energy decomposition results.

MD simulation and the MM-GBSA method were utilized to calculate the binding free energy of the PSD95 GK/QSF system with a value of −96.74 ± 10.03 kcal/mol. According to the per-residue energy decomposition results, we found that the residue Y1 displayed a crucial binding energy contribution of −6.45 kcal/mol. In order to validate whether the MM-GBSA method was applicable for this study, it was intentionally mutated to arginine to interrupt the hydrophobic and hydrogen bonding interactions. As anticipated, this mutation as Y1R peptide showed a higher binding free energy with a value of −85.88 ± 6.46 kcal/mol than that of QSF (Figure 3B). These results met the theoretical expectation, and it was suggested this approach might be suitable to predict the binding affinity.

Analyzing the distribution of per-residue energy decomposition of PSD95 GK/QSF, R (-5) which is located at the hydrophobic pocket of PSD95 GK, displayed unfavorable energy contribution mainly due to guanidine of R (-5) exposed to the solvent, leading to the relatively high polar solvation energy (Supplementary Table S2). Thus, we modified R (-5) to hydrophobic amino acids as F, G, A, T, S, and V. It can be found that the substitution of hydrophobic amino acids (R-5F, R-5G, R-5A, and R-5V) for arginine at position (-5) increased the binding free energy, while the substitution of hydrophilic amino acids (R-5T, R-5S) decreased the binding interaction, suggesting that (-5) position substitutions were preferred for hydrophobic amino acids.

Similarly, N (6) at α helix displayed slightly unfavorable energy contribution, which was modified as H, E, and C to enhance the hydrophilic interaction. However, the binding free energies of N6H, N6C, and N6E were higher than those of QSF. In addition, the hydrophobic cradle of GK formed by I593, A601, and L608 accommodated the benzene ring of F10, while the hydrophobic space was not fully utilized. The peptide F10W, in which the phenylalanine was replaced by tryptophan, also demonstrated better binding affinity with −105.94 ± 7.68 kcal/mol than QSF (Figure 3B). Therefore, five of the top-ranked peptides with lower binding free energy than QSF were synthetized and evaluated by competitive fluorescence polarization assay.

Fluorescence polarization assay has been widely implemented in high-throughput screening (HTS) and drug discovery (Banco et al., 2018; Miyoshi, 2018; Perveen et al., 2021). The assay could detect the alterations in the apparent molecular weight of a fluorescent probe (or tracer) in solution by observing changes in the polarization of the sample’s emitted light. Therefore, it is usually utilized to interrogate a variety of biological processes, such as a binding event or the enzymatic cleavage of a substrate (Hall et al., 2016). In the competitive FP-based assay, we developed fluorescein isothiocyanate (FITC)–conjugated SAPAP peptide as the fluorescent probe and a His-tagged PSD95 GK domain (His-GK) as the protein. The complex of His-GK with the FITC-SAPAP peptide produced a high FP value that was decreased following the displacement of the labeled peptide by either the unlabeled peptide or inhibitor compounds (Figure 3A). We next chose a low concentration of 10 nM of the probe which could yield a reasonable fluorescent signal and a stable polarization value. To determine the dissociation constant (K _d) for His-GK/FITC-SAPAP, we used a constant concentration of 10 nM of FITC-SAPAP, and a series of increasing concentrations of His-GK varied from 0 to 20 μM. The K _d of binding between the FITC-SAPAP probe and His-GK was determined to be 4.28 μM, with a dynamic range (ΔmP = mP of bound peptide – mP of free peptide) of 160 ± 5.23 mP. The stability of the FP assay is critical for high-throughput screening, so we incubated the plate at room temperature over a 12-h period. The results showed that the assay is stable and suitable for further high-throughput screening of GK inhibitors, as evidenced by the binding curves (Supplementary Figure S2). Then, the optimal concentration of His-GK was determined to 2 μM with 78 ± 2.34 mP for the competitive binding assay. The competitive binding assay was conducted with 10 nM of FITC-SAPAP, 2 μM of the PSD95 GK domain, and a series of increasing concentrations of the designed peptides. The unlabeled peptide SAPAP was treated as the positive control. The results of competitive fluorescence polarization assay showed that the abovementioned five peptides displayed better inhibitory activity than QSF peptide. Especially, the peptide F10W displayed the most potent inhibitory activity of PSD95/SAPAP with a K _i value of 0.73 μM (7 times than K _i value of QSF). To further confirm the binding affinity, the direct interaction between PSD95 GK and F10W was tested by MST and ITC assay with a K _d value of 0.54 ± 0.21 μM and 0.92 ± 0.45 μM, respectively, which is lower than that of QSF (K _d = 1.14 ± 0.14 µM in ITC assay).

To elucidate the difference of inhibitory activity among QSF, F10W, and R5A, the per-residue energy decomposition using the MM-GBSA method was applied. As expected, after the displacement of R (-5) with phenylalanine, glycine, alanine, or valine, the peptide displayed a more favorable energy contribution due to the decreased polar solvation energy (Supplementary Table S5–S8). These hydrophobic residues coupled with I (-4) increased the hydrophobic interactions with the hydrophobic pocket of GK (Figure 4C). In addition, the residue A (-5) of R (-5)A formed the hydrogen bond with D549_GK, which reinforced the binding affinity (3.36 kcal/mol in R (-5)A vs. 4.2 kcal/mol in QSF) (Figure 4A). In addition, the energy decomposition of R5V, R5F, and R5G were shown in Supplementary Material (Supplementary Table S6–S8). For F10W, the results showed that tryptophan at (+10) position played a more favorable interaction with the hydrophobic cradle of GK (−5.70 kcal/mol in F10W vs. −4.88 kcal/mol in QSF) (Figures 4A,B). Unexpectedly, the arginine at -3 position of F10W showed obvious improvement on energy contribution compared with QSF (−7.22 kcal/mol in F10W vs. −1.41 kcal/mol in QSF), which was offset by the unfavorable energy contribution of R (-5) with 7.33 kcal/mol. Therefore, we anticipate developing a more potent inhibitory peptide by double-position mutations of F10 and R (-5) in the future research work.

As reported, QSF did not have obvious selectivity for PSD95 GK against DLG1 GK by MD simulation and ITC experiment (Li et al., 2020). These results also suggested that the key residues for the binding of QSF with DLGs are conserved and might not have selectivity for other DLG subfamily members (DLG1-5 GK). Analyzing the binding interactions from the Trp of F10W, we found that the key amino acids (I593, A601, and L608) are identical except for DLG5, suggesting that the modification of QSF to F10W might only affect the selectivity of F10W for DLG5 (Supplementary Figure S6).