Initially, the GABARAP/AnkB-LIR computational model was generated, starting from the three-dimensional data of the complex, available in the PDB. The simulation, refined by molecular modeling techniques (see
Section 3 for details), was then used to design new peptides endowed with high affinity for GABARAP. The following procedure was adopted:
2.1. Setup of the GABARAP Computational Model and the Identification of the AnkB-LIR Core Sequence
The GABARAP/AnkB-LIR complex model (
Figure 1A) was retrieved from the Protein Data Bank (PDB) (accession code 5YIR) [
11] and then refined by energy minimization and molecular dynamics (MDs) simulations, following the procedure reported in
Section 3. The AnkB-LIR peptide rapidly reached the geometrical stability over the 500 ns long MD simulation, as demonstrated by the protein Cα RMSD plot (
Figure 1B).
As expected, and verified by inspecting the MD trajectory frames, the LIR domain (sequence WVIV of AnkB-LIR) created numerous contacts with GABARAP (
Supplementary Materials Figure S1). Interestingly, the SDEE residues were also involved in productive contacts, including the electrostatic interactions between the side chains of the peptide glutamates and the positively charged area of GABARAP close to K
46 and R
47. Conversely, the remaining C-terminal residues were mainly involved in internal contacts, stabilizing the α helix shaped by the DEEIEEARQKA sequence.
Then, to exactly define the minimal portion of AnkB-LIR with the highest affinity for GABARAP, the peptides AnkB-LIR and AnkB-core (sequence WVIVSDEE) were subjected to MD simulations and MM-GBSA calculations to estimate their binding free energy. Desmond and Prime tools of Maestro were employed for these computations, which predicted ΔG* values of −135.1 and −107.9 kcal/mol for AnkB-LIR and AnkB-core, respectively (
Table 1). This result indicates that the 8 amino acids belonging to the AnkB-core contribute 75% of the overall interaction energy of the full AnkB-LIR peptide (composed of 20 residues). Then, to further define the contact area, MD simulations and MM-GBSA calculations were performed on the GABARAP/WVIV complex model, in which only the LIR motif (AnkB-wviv peptide) was simulated. As reported in
Table 1, residues WVIV contribute 65% of the overall binding free energy. This outcome confirms that the LIR motif, shared by all proteins involved in the autophagy machinery, displays the highest complementarity with the GABARAP-binding site and is responsible for the most significant protein–protein contacts. Subsequently, the same computational protocol (MD simulations and MM-GBSA calculations) was applied to study the interaction of peptide K1. Considering that its experimentally determined K
d lies in the nanomolar range (390 nM), this peptide could be considered as a reference inhibitor of GABARAP, together with AnkB-LIR. By our computations, the predicted ΔG* for peptide K1 on GABARAP was −118.9 kcal/mol (
Table 1), a value slightly higher than that of AnkB-LIR (−135.1 kcal/mol). This result is in line with the K
d values reported for the two peptides (0.27 and 390 nM, respectively).
2.2. Computational Design of AnkB-Core Analogs
Considering that the WVIVSDEE (AnkB-core) sequence accounts for 75% of the GABARAP/AnkB-LIR contacts, we proceeded to design small peptides endowed with high affinity for GABARAP using AnkB-core as a template. In this attempt, only the residues of the LIR domain (WVIV, positions 2–5 of AnkB-LIR) of AnkB-core (WVIVSDEE) were mutated, because of their direct involvement in the interaction with GABARAP. In this challenging effort, we tried to optimize the peptide sequence, also shared by other GABARAP binders, to improve the ligand/protein complementarity and selectivity. To this end, the “affinity maturation protocol”, implemented into the Prime module of the Maestro software, was utilized to mutate the VIVS residues into all possible natural amino acid combinations. To avoid the combinatorial explosion, the Monte Carlo optimization option was selected. By this option, 2000 peptides were randomly generated by Monte Carlo algorithm and the peptides with a maximum of three simultaneous mutations were accepted to create the output file containing 100 solutions. Then, the Prime module was also employed to establish whether the mutations led to a more favorable interaction with the biological counterpart, by calculating the mutant peptides binding free-energy (ΔAffinity) values [
13].
At the end of these calculations, we visually inspected the results for the first 100 peptides with the highest predicted affinity. Among the predicted peptides, we noted that 7 of them displayed ΔAffinity values lower than 2 kcal/mol with respect to the initial template (AnkB-core). In these peptides, position 2 was substituted by Arg, Glu, or Ile; positions 3 and 4 contained only Ile; while position 5 included only alkaline residues, such as His and Arg. Among them, only one candidate, WEIIHDEE, named Pep-sol4, was further investigated by MD simulations and MM-GBSA calculations, because it presented an interesting Glu in position 2. Through this acidic amino acid, the peptide could potentially interact with the positive area shaped by GABARAP-K
46 and GABARAP-R
67 (two conserved residues among Atg8 proteins). Moreover, GABARAP-K
46 is considered to be a universal gate-keeper, regulating the entrance of ligands interacting through the LIR motif [
14]. The structural alignment of the GABARAP/Pep-sol4 complex to the GABARAPL2/UBA5 NMR structure (PDB accession code 6H8C) [
15] confirmed that the glutamate in position 2 of Pep-sol4 could reproduce the interaction displayed by E
15 (GAMEIIHEDNEWGI
ELVSE) of the “ubiquitin-like modifier activating enzyme 5” (UBA5) with GABARAP-K
46.
By applying the computational protocol previously adopted for the reference inhibitors, the binding free-energy value of Pep-sol4 was calculated to be slightly lower than that of AnkB-core (−103.3 vs. −101.4 kcal/mol), suggesting that the new peptide could bind GABARAP with a similar binding affinity (
Table 2).
Moreover, the visual inspection of the GABARAP/Pep-sol4 MD trajectory and the root mean square fluctuation (RMSF) plot of the ligand atoms over the simulation time suggested that the C-terminal portion of the peptide was not firmly bound to the GABARAP surface, thus preventing a stable and productive interaction with the protein (
Figure 2). Consequently, considering that the side chains of I
3 and D
8 were spatially close in the binding mode adopted by Pep-sol4, we designed a cyclic peptide in which I
3 and D
8 were mutated into two Cys residues bound by a disulfide bond. This modification aimed at reducing the conformational flexibility of the ligand, generating a more stable binding mode on the GABARAP surface. The resulting peptide (named Pep-sol4cc, WE
CIHDE
C) was again analyzed in the complex with GABARAP by MD simulations and MM-GBSA calculations. At the end of these computational procedures, the estimated ΔG* of Pep-sol4cc was −103.7 kcal/mol, a value comparable to that of Pep-sol4 (−103.3 kcal/mol). This information led us to conclude that the structural rigidification did not affect the affinity of the peptide; however, as demonstrated by the ligand RMSF plot (
Figure 2), an improvement of the conformational stability was successfully achieved.
2.3. Computational Design of the WC8 and WC10 Peptides
Then, with the aim of improving the theoretical binding affinity of Pep-sol4cc, H
5 was mutated into a Phe, supposing that it could better fill the hydrophobic area sized by P
52, L
55, and Q
59. The resulting peptide (WC8, sequence WE
CIFDE
C) was analyzed by MD simulations and MM-GBSA calculations, which led to a ΔG* value of 12 kcal/mol, lower than that of the originator (
Table 2). The cluster analysis performed on the MD trajectory frames displayed that, in the structure representative of the most populated cluster of GABARAP/WC8 conformations (accounting for 79% of conformational ensembles explored), the ligand was stably bound on the GABARAP surface (see
Figure 2 for the RMSF plot), forming numerous interactions (
Table 3 and
Figure 3A).
In detail, several H-bonds were established, and salt bridges formed between WC8-E2 and the side chains of GABARAP-K46 and -R67, and between the Cter of WC8-C8 and the side chain of GABARAP-R28.
Regarding the hydrophobic contacts, the indole ring of WC8-W
1 was positioned in a pocket formed by residues I
23, I
32, P
30, K
48, and F
104, while the side chain of WC8-I
4 pointed toward another pocket delimited by Y
49, V
51, F
60, L
63, and I
64, establishing van der Waals (vdW) interactions. Finally, similar hydrophobic contacts were also observed between WC8-F
5 and the GABARAP area shaped by P
52, L
55, and L
63. The complete 2D representation of the interaction network between WC8 and GABARAP is shown in the
Supplementary Materials Figure S2A.
WC8 exhibited an estimated ΔG* value close to that of K1; hence, with the aim of designing a more potent peptide, we included two additional N-terminal residues. This hypothesis was supported by the fact that the AnkB-LIR peptide (
EEWVIVSDEEIEEARQKA), used as a template, contains two glutamate residues before the AnkB-core (WVIVSDEE). For this reason, we speculated that the homologation of WC8 on the N
ter could lead to a more potent peptide, considering that the new atoms could create an additional bond network. Our objective was to reach the region sized by I
32, Y
5, and K
47, close to the W site, on the GABARAP surface (the yellow area in
Supplementary Materials Figure S3A). Therefore, to find the optimal
N-terminal sequence, two glycines were initially added to WC8 (GGWE
CIFDE
C); then, the application of the “affinity maturation protocol” on the first Gly residue led to the identification of Tyr (YC10,
Table 2) and Trp (WC10,
Table 2) as the most suitable substitutions. In this attempt, the glycine in position 2 was not mutated to allow a certain conformational mobility on the
N-terminal tail of the new peptide. Interestingly, the
N-terminal residues (WG) and the Glu in position 4 (E
4) of WC10 (
WGW
ECIFDEC) reproduced the interactions displayed by UBA5 (GAMEIIHEDNE
WGI
ELVSE) in the complex with GABARAP and GABARAPL2 [
15] (
Supplementary Materials Figure S3B).
MD simulations and MM-GBSA calculations on the GABARAP/YC10 and GABARAP/WC10 complexes revealed that the latter possessed the highest affinity, with a predicted ΔG* value almost 7 kcal/mol lower than that of WC8 (
Table 2). Cluster analysis was then performed on the GABARAP/WC10 MD trajectory frames; the representative structure of the most populated cluster of conformations (which accounts for the 88% of total conformational ensemble explored) is represented in
Figure 3B. Notably, the visual inspection of the GABARAP/WC10 most representative structure highlighted that the side chain of the newly added residue (W
1) formed a cation–π interaction with GABARAP-K
46, while the carbonyl group of the same residue established an H-bond with the side chain of GABARAP-K
48 (
Figure 3B). Surprisingly, the side chain of W
1 did not occupy the expected region of GABARAP, but the new additional cation–π interaction greatly contributed to the calculated binding affinity of the peptide. In addition, WC10 (1) shares all the interaction networks established by the GABARAP/WC8 complex, (2) is able to orientate GABARAP-K
48 in order to establish additional cation–π interactions with WC10-W
3, and (3) is able to form an additional H-bond interaction between the I
6(C=O) and GABARAP-R
67(=NH
2+) (
Figure 3). The 2D representation of the interaction network between WC10 and GABARAP is showed in the
Supplementary Materials Figure S2B.
To conclude, we designed two new cyclic peptides (WC8 and WC10) endowed with a reduced conformational mobility and calculated binding free-energy values in a lower range than those estimated for the reference peptides, AnkB-core and K1. In light of these data, WC8 and WC10 could exhibit higher experimental affinities compared to the reference peptides. Nevertheless, it must be considered that our computations did not account for the entropic contributions to the binding free energy; hence, they should be regarded as a starting point for further experimental studies.
2.4. Experimental Validation and Biophysical Experiments
Based on the results of the computational study, the K1, AnkB-core, WC8, and WC10 peptides were purchased by Proteogenix for the experimental investigations. In detail, MST and SPR assays were conducted on the peptides displaying a sufficient stability in water and PBS buffer. Then, the anticancer potential of the most promising candidates was investigated. Initially, we verified that the peptides were water soluble and stable in the buffer in which the recombinant GABARAP protein was solved. Unfortunately, AnkB-core was not soluble in water; thus, it was impossible to use it as a reference peptide. Conversely, K1, WC8, and WC10 displayed an excellent stability in water and PBS, so they were tested by biophysical methods.
In detail, MST and SPR experiments were carried out with the aim of measuring their K
d values on GABARAP. As a preliminary step, the K
d of the reference peptide K1 was determined in order to (1) check the experimental procedure and verify the result against data reported in the literature by Weiergräber et al. (K
d = 390 nM) [
9], and (2) obtain a reference value to compare the K
d measured for the new peptides. MST experiments were conducted on a Monolith NT.115 instrument (
Figure 4A), while SPR analyses were performed using a BIAcore 8K system, applying the protocol reported by Weiergräber et al. (see
Section 3 for details) [
9] (
Figure 4B).
Surprisingly, the K
d measured for K1 was close to 3 µM, a value 7 times higher than the one reported in the literature. However, all the techniques employed in this study agreed on this value. The data obtained for WC8 revealed a K
d of 22 µM (
Figure 5A,B), consistent among the different biophysical approaches. Remarkably, WC10 displayed a K
d in the same range of the reference peptide K1, with a value close to 4 µM, obtained by both MST and SPR (
Figure 5C,D).
Because the K
d value of peptide K1 proved to be higher than the one reported in the literature, we decided to validate our data by repeating the MST-binding affinity experiments using another Monolith instrument (Monolith NT.115Pico), located in a different laboratory. The new results confirmed our previous findings, with all the peptides displaying K
d values consistent with those determined earlier (
Supplementary Materials Figure S4).
According to the theoretical predictions, WC10 should be more active than K1 (ΔG* = −122.0 vs. −118.9 kcal/mol, respectively), and WC8 less active than the others (ΔG* = −115.7 kcal/mol), as shown in
Table 2. Considering the confidence range of the experimental K
d and the omission of the entropic contribution to the estimated binding free-energy values, the computations predicted the affinity trend of the selected peptides well.
2.5. Biological Experiments
Finally, K1, WC8, and WC10 were tested in vitro on PC-3 prostate cancer cells, to evaluate their potential antitumor effects (
Figure 6). Prostate cancer is the second most commonly diagnosed malignancy in men worldwide. Considering that the probability of developing the disease during a man’s lifetime is 15% and that prostate tumor cells can also spread to the lungs and bones via angiogenesis [
16], we decided to evaluate the biological activity of the peptides in vitro on a prostate cancer model. PC-3 cells were chosen for the screening due to their highly metastatic nature, effectively mimicking an aggressive form of the disease. Notably, it has recently been demonstrated that prostate cancer models show a significant upregulation of autophagy [
6,
8,
17].
Therefore, the biological activity of different concentrations of K1, WC8, and WC10 (from 0.5 to 10 µM) was evaluated with an MTS cell viability assay on PC-3 cells and non-cancerous PNT2 prostate cells (
Figure 6). The results reported in
Figure 6A show that, 96 h post-treatment, none of the tested samples displayed a significant cytotoxicity (cell availability > 90%), confirming the excellent biocompatibility and potential pharmacological selectivity for tumor cells. Indeed, as shown in
Figure 6B, a reduction in cell viability (expressed as percentage % of viable cells) was observed in PC-3 cells treated with K1, WC8, and WC10 compared to the untreated control. Interestingly, the treatments of PC-3 cells with WC8 and WC10 (from 1 to 10 µM) display high efficacy, when compared to Paclitaxel (
Figure 6B). The in vitro data demonstrate that the compounds exhibited a considerable anticancer activity, especially at the highest tested concentration (cell viability 27.16% for K1, 24.06% for WC8, and 22.5% for WC10).
The biological data on PC-3 cells indicate that all peptides possess IC
50 values close to 5 µM, consistent with the K
d estimated by the biophysical experiments. Surprisingly, WC8 displayed a better activity profile than the reference peptide K1. Based on this finding, we may speculate that some other biochemical mechanism or additional activity on different mAtg8 subfamilies could improve the activity of the new peptides [
8]. Nevertheless, since the work presented here is a proof-of-concept study, the peptides have been preliminary tested in a non-cancerous and subsequently in a cancer cell line, to exclude possible off-target cytotoxicity and perform an initial pilot study to evaluate the in vitro anti-cancer efficacy. However, we are planning to extend the screening to other cancer cell lines in the upcoming further evaluation of the peptides and their antineoplastic mechanism of action. Furthermore, to shed light on the possible secondary targets, we have planned biological and biophysical experiments on LC3B to evaluate if our peptides could show any affinity to it. Moreover, we should remember that the data on PC-3 cells represent a preliminary assessment that merely suggests the potential application of GABARAP inhibitors as anticancer agents. Further biological assays are needed to unveil the mechanism by which these peptides trigger cell death.