Discovery of Inhibitors for Proliferating Cell Nuclear Antigen Using a Computational-Based Linked-Multiple-Fragment Screen.

Proliferating cell nuclear antigen (PCNA) is a central factor in DNA replication and repair pathways that plays an essential role in genome stability. The functional roles of PCNA are mediated through an extensive list of protein-protein interactions, each of which transmits specific information in protein assemblies. The flexibility at the PCNA-protein interaction interfaces offers opportunities for the discovery of functionally selective inhibitors of DNA repair pathways. Current fragment-based drug design methodologies can be limited by the flexibility of protein interfaces. These factors motivated an approach to defining compounds that could leverage previously identified subpockets on PCNA that are suitable for fragment-binding sites. Methodologies for screening multiple connected fragment-binding events in distinct subpockets are deployed to improve the selection of fragment combinations. A flexible backbone based on N-alkyl-glycine amides offers a scaffold to combinatorically link multiple fragments for in silico screening libraries that explore the diversity of subpockets at protein interfaces. This approach was applied to discover new potential inhibitors of DNA replication and repair that target PCNA in a multiprotein recognition site. The screens of the libraries were designed to computationally filter ligands based upon the fragments and positions to <1%, which were synthesized and tested for direct binding to PCNA. Molecular dynamics simulations also revealed distinct features of these novel molecules that block key PCNA-protein interactions. Furthermore, a Bayesian classifier predicted 15 of the 16 new inhibitors to be modulators of protein-protein interactions, demonstrating the method's utility as an effective screening filter. The cellular activities of example ligands with similar affinity for PCNA demonstrate unique properties for novel selective synergy with therapeutic DNA-damaging agents in drug-resistant contexts.

Introduction

Proliferating cell nuclear antigen (PCNA) is a nuclear homotrimeric protein that encircles DNA with classic attributes of a processivity factor in DNA replication.[1] However, PCNA also mediates protein complex formation in base excision, mismatch, nucleotide excision, homologous recombination (HR), Fanconi anemia DNA repair, and translesion synthesis pathways.[2−5] PCNA can also participate as a regulator of cell cycle progression, chromatin remodeling, and transcription.[4] More than 200 proteins are currently proposed to interact with PCNA with a major subset involved in DNA damage repair.[6] Multiple key proteins share a binding motif for PCNA called the PCNA-interacting protein-like motif (PIPM).[7−12] PCNA is the central scaffold protein for DNA polymerase association and elongation of repaired DNA strand, as well as necessary for the final steps of ligation. Inhibition of PCNA association with PIPM-containing proteins could ultimately impair the cell’s ability to repair or replicate DNA. In support of this approach, the deletion of the PIPM within c-Abl disrupts the nuclear c-Abl apoptotic function in DNA-damaged cells.[13] Although various strategies currently exist for targeting DNA repair pathways, antagonists of specific PCNA functions could serve uniquely to inhibit specific DNA damage postreplication repair,[14] specifically disrupting RAD6-dependent translesion synthesis, as well as the “template switch” pathway.[15] This function is relevant in the repair of DNA double-strand breaks through HR, a pathway known to be overactivated in various tumor types.[16] The discovery of such an inhibitor would provide a unique probe for functional analysis with potential as a sensitizing or synergistic agent in the development of new combination therapies.

Although a rationale to better understand the functional roles of PCNA in normal and disease states has emerged, PCNA–protein interactions do not provide a traditional small-molecule target and motivate a new approach to modulate protein–protein interactions (PPIs). Structure-based design of PCNA inhibitors could benefit from structures of PIPM-containing peptides/proteins in complex with PCNA. The information to date has revealed that the core PIPM recognition sequence is involved in a 310-helix that binds in a hydrophobic surface pocket on PCNA. Small-molecule mimics of this topology are not always easily defined. Our previous studies of PCNA features responsible for specificity of protein recruitment implicate varied conformations of PCNA upon ligand engagement, which impart functional differences.[17] Finally, a recent high-throughput biomolecular screening effort led to the discovery of the thyroxine class (T3 or L-3,5,3′-triiodothyronine) of inhibitors that target the surface pocket in which the PIPM’s 310-helix binds.[18,19] The analog compound, T2AA ((S)-4-(4-(2-amino-3-hydroxypropyl)-2,6-diiodophenoxy)phenol) showed pharmacologic effects in combination with the DNA damaging agent, cisplatin in the drug-sensitive cell line, U2OS.[19,20]

Fragment-based drug design (FBDD) is a generalized approach commonly used in the discovery of small-molecule hit ligands for biological targets.[21] By screening low molecular-weight chemical fragments, which individually bind to their intended target sites, higher affinity drug-like molecules can be generated using the combined information. FBDD can have inherent advantages when factors such as commercial availability, ease of synthesis, and coverage of chemical space are considered.[22] Furthermore, libraries consisting of larger molecular weight entities can be disadvantaged by their “bulkiness”, which may preclude favorable binding ligand interactions in protein subpockets due to steric exclusion.[23] Screening single small fragments is useful in situations where 3D structural information of a target site is readily available. However, the methods and approaches in fragment-based screens have demonstrated variable outcomes with similar targets.[24] New and refined in silico approaches that utilize the concepts of fragment-based screens offer increased applications to a broader range of protein targets.

A challenge to the application of FBDD methods for PPIs is the conformational flexibility of protein docking sites, even in cases where localized “hot spots” are defined.[25,26] In cases where the fragment binding at one subpocket can influence the conformation of a second subpocket, a molecular screening system that maps multiple subpocket binding sites, in tandem, would provide advantages for defining new probe ligands.[23] Such an approach would offer opportunities to define prefiltered hit molecules with multiple fragment features to be considered for subsequent optimization of drug properties. Peptoids offer a chemical scaffold with conformational flexibility and potential for high degrees of fragment diversity. Traditional biomolecular screens using peptoid libraries have proven useful in identifying ligands for many types of protein targets.[27−33] Several of these efforts make use of trimeric peptoid libraries, which are generally considered small molecules.[34,35] A key attribute of these molecular types is that they effectively “tie” three chemical fragments together into a single backbone, enabling tandem multifragment screening. However, depending on target systems, it may not be efficient to construct large and diverse libraries of trimeric molecules for screening. An in silico method to reduce the number of compounds while maximizing the diversity of fragment features prior to a focused set of scalable syntheses is desired. The objective here is to develop such an approach to target the PIPM binding site in PCNA.

In this study, a chemical platform based on moderate molecular weight peptoids (450–1000 Da) is evaluated for application in the discovery of novel small-molecule PPI inhibitors of PCNA. This system employs computational-based peptoid library design and screening that accounts for the conformational flexibility of the receptor and ligands. In this analysis, molecular dynamics (MDs) simulations were used to identify potential key PCNA–ligand interactions. Together with the flexibility of the PIPM binding site of PCNA,[17] the results indicate the predicted adaptability of the protein toward multiple inhibitors. These inhibitors are shown for the first time to have an unusual selective synergy with doxorubicin in drug-resistant tumor cell lines. Ultimately, the approaches tested here could have applications in systems where distinct receptor–ligand conformations confer selective functional states that modulate the biological responses.

Results

Tripeptoid Library Creation Using Computational Tools

The PIPM binding site on PCNA was examined to gain basic information on favorable chemical features of potential small molecule inhibitors. PCNA does not have any known natural small-molecule modulators, nor does it have an obvious ligand-targeting site. Even when compared to other protein–protein interfaces, such as the contact surface between MDM2 and p53, there is not a clear binding groove—instead there is a shallow, relatively small surface pocket where PIPM-containing proteins interact. Our previous work identified three key subpocket regions at the PIPM binding site that are important for the recognition and binding of PIPM-containing peptides/proteins.[17] It was hypothesized that tripeptoids would be sufficient in dimensionality to engage each of these subpockets. The general mass range of such materials are close to those for other known inhibitors of PPIs [e.g., Zhao, et al. (2015).[36] The size of these ligands also enhances synthetic accessibility, time to completion, and cost efficiency. More importantly, peptoids have more favorable properties to penetrate cell membranes and are resistant to solvents, temperature, chemical denaturants, and proteolysis.[37−40]

An initial set of 20 primary amines was selected (Figure , fragments labeled in blue) that varied in hydrophobicity, aromaticity, ability to form hydrogen bonds, and substructures present in clinically available drugs. Many of these amines are commercially available, although some require different degrees of in-house synthesis. The CombiGlide application within Maestro (Schrödinger)[41] was used to generate a virtual combinatorial library of trimeric peptoids that contained each of the 20 primary amines, plus hydrogen as a potential substituent, to give 9261 total compounds. The ligands were ionized (along with desalting and tautomeric generation) using Epik to generate all possible states within a pH range of 7.0 ± 2.0, and these were minimized using the OPLS-2005 force field.

Figure 1

Fragments used for the creation of virtual tripeptoid libraries. A list of 37 primary amines, along with T2AA, was used to create a combinatorial virtual library of peptoid-based compounds. N* indicates the location of the −NH2 group, which is the position of substitution into the tripeptoid backbone. Blue labels indicate fragments that were used in the initial screen of tripeptoids containing a combinatorial set of 20 primary amines.

The use of a single rigid protein structure for in silico docking can often produce skewed results due to geometric constraints and functional group directionality at the binding site. This is exacerbated when taking into account a protein such as PCNA, which has a significant degree of plasticity at the PIPM site,[17] increasing the likelihood of many false negatives using rigid docking. Although a flexible docking method such as Schrödinger Maestro’s induced-fit[42] can address this issue, the computational time required to screen large numbers of highly flexible ligands such as peptides and peptoids can make it less practical. As an alternative, four crystal structures of PCNA bound to four different ligands were selected for screening, each representing structural divergence at the PIPM binding site. These structures included PCNA complexed to the C-terminal tail of p21 (PDB ID: 1AXC), PCNA complexed to residues 331–350 of FEN1 (PDB ID: 1U7B), PCNA complexed with the T3 ligand (PDB ID: 3VKX), and PCNA complexed with a fragment of DNA polymerase η (PDB ID: 2ZVK). Using this approach, the enrichment of fragments in specific positions of the tripeptoid are expected to reflect the features of all four PCNA structures.

Each protein structure was minimized in complex with its respective ligand/peptide using the OPLS-2005 force field in Maestro. Cubic grid boxes were created using either T3 (3VKX) or a PIPM peptide amino acid at the hydrophobic site on PCNA, as the centroid, with a length of 30 Å. The prepared ligands were flexibly docked into each form of PCNA using the standard precision (SP) model in Glide,[43] and the top 10% of the ligand hits from each docking run were flexibly redocked into the respective form of PCNA using the extra precision (XP) Glide algorithm, which places harsher penalties on desolvation energy and ligand strain and has greater requirements for ligand–receptor shape complementarity. Upon completion of each docking exercise, the top 50 hits in each of the docking runs were manually analyzed. The frequency with which a specific fragment was present at a particular location along the tripeptoid backbone was tallied for each crystal structure of PCNA, and the top 50 hits from each run were compiled into a total list of 200 top hits.

The compiled screening results indicated a different population of candidate hit ligands from each crystal structure of PCNA. Most informative was the variation in fragment enrichments observed at each of the three positions along the peptoid backbone with an example summary in Figure . The 1st (N-terminal) position showed a preference for a set of side chains, including NLys, NArg, NTyr, NGln, NEba, and NBal. Two fragments that stood out most significantly were NLys and NArg, each of which were present in over 20% of the top hit list. The 2nd position showed less enrichment with nine fragments appearing, but the top 5% hits contained NArg (15%) and NPip (14%). In comparison to the first two positions, the 3rd position in the peptoid backbone showed the highest enrichment of particular side chains, notably aromatic groups that contained functionalities that allow for hydrogen bonding with protein amino acids. NBal especially stood out as a highly enriched fragment (43%) in the 3rd position along with two additional aromatic fragments (NTyr 13% and NPip 20%).

Figure 2

Glide docking-based frequency with which fragments appeared in the three substitution positions on a tripeptoid backbone. A set of 20 fragments, in addition to hydrogen, were virtually combinatorially incorporated into a tripeptoid backbone and were screened against four different crystal structures of PCNA (PDB IDs: 1AXC, 3VKX, 1U7B, and 2ZVK) in silico using the Glide SP and XP docking algorithms. The frequency that respective fragments appeared in positions R1, R2 or R3 (lower right) were tallied for the top 50 hits from each run involving a different crystal structure. The percentage of the cumulative total of substitution frequency (out of a possible 200) at a given position is shown above the stacked columns. Hydrogen as a substituent is labeled as “Gly”.

A second screen evaluated how the use of pre-existing ligands in combination with the associated crystal structure impart fragment selection. In previous studies, the small molecules, T3 and T2AA, as well as several synthetic variants, have demonstrated binding in a hydrophobic subpocket of the PIPM binding site.[18,19] Both compounds were considered as monomers compatible with submonomer peptoid synthesis conditions.[44] Either fragment could potentially serve as an anchor fragment, directing small peptoid fragments to additional subpockets in the PIPM binding site on PCNA. T2AA was selected as a better candidate for investigation based upon physicochemical properties and the lack of thyroid hormone properties.[18] A second extended virtual combinatorial library was created using the same methodology as previously discussed, with a set of 37 peptoid side chains along with NT2AA and CT2AA, which differ in attachment points (Figure ). As the hydrophobic subpocket on PCNA would be expected to sterically accommodate potential binding of the T2AA fragment, only the structure of PCNA bound to T3 was used (PDB ID: 3VKX). This constraint introduced an intentional bias to increase the likelihood of T2AA acting as an anchor. All ligands were prepared and screened as before, this time using Maestro’s HTVS scoring function. The top 50% of the hit compounds were screened again using Glide’s SP model, and the top 30% of the hits from the SP screen were docked again using Glide’s XP model. As before, the frequency that a particular side chain was present at a given substitution location on the peptoid backbone was tallied for the top 50 hits from the XP screen.

From the results (Figure S1), a significant enrichment for a small set of possible side chains was observed with NT2AA in the first position consistent with the ligand and structural bias introduced into the screen. Interestingly, the second position showed high enrichment of the ethanol amine side chain (NEal). In contrast to the previous screen, the third position did not show significant enrichment for any particular side chain. A review of the docking poses further revealed that fragments in the third position adopted multiple conformations taking on a variety of interactions in the region proximal to the PIPM glutamine binding site. As a result, the third position fragments did not offer any obvious stabilizing interactions with PCNA in contrast to the results of the first screen. The high frequency of NT2AA as a fragment in the first position reflects the expected favorable docking at the unique hydrophobic pocket. There is a reasonable likelihood that these interactions at the first position significantly influenced the interactions allowed at the second position which enriched primarily for NEal. A strategy for incorporation at the N-terminus of resin-bound peptoid without the use of protecting groups (Figure ) offered a pathway to synthesis of peptoids with T2AA at the first position.

Figure 3

Synthesis of T2AA- or non-T2AA-containing tripeptoids. (a) C2H3BrO2 (1 M in DMF, 20 equiv), DIC (19 equiv), 1 h, 35 °C; (b) H2N-R3 (1 M in DMF), 2 h, RT; (c) H2N-R2 (1 M in DMF), 2 h, RT; (d) H2N-R1 (1 M in DMF), 2 h, RT; (e) T2AA (19.5 equiv), DIEA (39 equiv), DMF, 16 h, RT; and (f) TFA/TIS/H2O (95:2.5:2.5), 1 or 3 h, RT.

Synthesis and in Vitro Screening of Tripeptoids

Fragments that appeared in ≥5% of each position on the tripeptoid backbone were assessed for incorporation and synthesis of select tripeptoids. Among the top hit fragments were several not commercially available. To streamline the process in these early steps, NEba and NGln were not considered for synthesis to avoid a need for incorporation of acid-labile protecting groups. The preparations of the remaining noncommercially available fragments with appropriate protecting groups are outlined in the Supporting Information. In all, 85 ligands were synthesized using the methodology shown in Figure . Most of the compounds contained the fragments that enriched from the virtual combinatorial screen; however, some ligands were also synthesized that were not predicted to be among the hits for purposes of comparison (e.g., tripeptoids containing NAem in the 1st position). The total list of synthesized peptoids are shown in Table S1. In some cases, chemical fragments that represented structural analogs were selected for incorporation, although they did not appear with high frequency in the virtual screen. This approach was deployed for comparisons with fragments that were top virtual hits. Examples of these fragments included NBza (substitute for NBal, NPip, or NTyr), NTrp (substitute for NPip), and NMma [(4-methylphenyl)methylamine, substituted for NBal, not shown in Figure ].

After synthesis and purification, the candidate peptoids were screened in a fluorescence polarization (FP) assay to identify ligands that disrupted the interaction of His-tagged PCNA and fluorescein-labeled 16-mer Pogo Ligase peptide (FAM-PL). This synthetic peptide is a high-affinity ligand previously used for a reporter in PCNA binding assays.[45] In the design of the assay, conditions were adopted from Pedley et al.[17] to improve the dynamic range between bound and unbound FAM-PL. Although an acceptable Z-score was determined using the original 10 nM peptide and 100 nM PCNA (Figure S2),[46] an improved dynamic range and reagent efficiency was established by adjusting the ligand and receptor concentrations (Figures S3 and S4). Using 5 nM FAM-PL peptide with 1 μM PCNA, conditions were established for >80% of the peptide binding sufficient for subsequent displacement assays and calculations of Ki values.

The compounds were initially screened in the FP assay at concentrations of 1 mM and 250 μM to define hits (data not shown) using T2AA as a positive control. From the screen at 250 μM, sixteen ligands were selected and each further analyzed in two-fold dilutions series to generate dose response curves (see Figure S14). Anisotropy was converted to fractional occupancy (FO), and IC50 values were determined for each peptoid by performing nonlinear regression fits of each dose response curve using eq . Inhibition constants (Ki values) were calculated from the resultant IC50 values using eq (Table ). Additional control experiments in the FP screen used individual fragments that composed the top hits including NGln, NLys, NPip, NTyr, NBal, NBza, and NEal, none of which were found to bind to PCNA (Figure S5). Examples of dose response curves for four of the top hits are shown in Figure .

Table 1

Hit Compound IC50 and Ki Values Measured by FPc

compound name	IC50 (μM)a	Ki (μM)a,b
T2AA	1.34 ± 0.33	0.128 ± 0.0318
Gly-NPip-NBal	7.74 ± 3.41	0.745 ± 0.328
NBal-NLys-NTyr	>600	>50
NLys-NPip-NBal	1.94 ± 0.51	0.186 ± 0.0491
NLys-NPip-NMma	12.93 ± 1.97	1.24 ± 0.190
NLys-NTyr-NBal	∼165	∼16
NMma-NPip-NBal	11.69 ± 2.55	1.12 ± 0.245
T2AA-Asn	7.20 ± 2.74	0.693 ± 0.264
T2AA-Gln	3.52 ± 1.65	0.339 ± 0.159
T2AA-Gly	2.91 ± 0.91	0.280 ± 0.0876
T2AA-Gly-NBal	5.66 ± 1.67	0.545 ± 0.161
T2AA-Gly-NPip	16.17 ± 3.71	1.56 ± 0.357
T2AA-NEal-Gly	1.17 ± 0.37	0.113 ± 0.0356
T2AA-NEal-NMma	1.18 ± 0.24	0.114 ± 0.0231
T2AA-NEal-NPip	1.82 ± 0.37	0.175 ± 0.0356
T2AA-NEal-NTyr	0.482 ± 0.328	0.0464 ± 0.0316
T2AA-NPip-NLys	6.13 ± 2.84	0.590 ± 0.273

Values are represented as the 95% confidence interval around the mean.

Calculated using eq .

Structures of T2AA-containing ligands can be found in Figure S6.

Figure 4

Dose response curves of four of the top hit peptoid-based compounds from the fluorescent polarization screen. Peptoid-based ligands were subjected to two-fold dose response analysis. Solutions were plated in duplicates of four, and error bars represent the standard error around the mean. T2AA (1 mM) in dimethyl sulfoxide (DMSO) serves as the positive control, and DMSO alone serves as the negative control. Data for all remaining compounds in Table can be found in the Supporting Information, Figure S7.

Identification of Molecular Recognition Features of PCNA for Peptoid Ligands

To understand the chemical features of the peptoids–PCNA interactions, MDs simulations were performed with selected hits and compared with T2AA in complex with PCNA. Our previous report established the utility of this approach, especially with regard to the T3 binding site.[17] As outlined in the experimental methods, each ligand was first docked into the cocrystal structure of PCNA and PL–peptide (PDB ID: 1VYJ), with the peptide removed, using the Glide induced-fit model in Maestro. These structures provided starting points for the MDs simulations and ensured that there were no conflicts on an atomic scale due to steric clashes and/or unfavorable ionic contacts. Each simulation was run for 5.0 ns, or until it converged, as judged by the change in protein Cα and side-chain root-mean-square deviation (rmsd) over time (Figures S8–S10). Upon completion, simulation trajectories were aligned to the first frame of their own simulation, and then the simulations were aligned to one another based on the position of their Cα atoms using VMD. The final fifty frames for each MD were averaged to give a final structure for each PCNA–ligand complex with example results shown in Figure .

Figure 5

Average structures of the final 50 frames of MDs simulations. The final 50 simulation trajectory frames for each analyzed peptoid ligand were averaged using VMD, and the resulting structures visualized using Pymol. The hydrophobic pocket (orange) and PIPM glutamine binding site (blue) are highlighted in each. The whole structure of the PCNA monomer (based on PDB ID: 1VYJ) is shown in the bottom right for comparative purposes.

Our previous studies highlighted how the PCNA PIPM domain can adopt a variety of conformations to optimize ligand binding.[17] The results in Figure indicate significant differences in PCNA conformation with the peptoid ligands when compared to the cocrystal structure of PCNA–PL. These differences are amplified in the PIPM binding regions for each protein complex. Regions on PCNA that appeared to drive the conformational difference between each structure most substantially were between residues 80–86, 93–97, 104–111, 117–136, 162–166, 172–177, 181–194, and 251–257 (Figure S11). Perhaps unsurprisingly, each of these regions was found in either a β-turn or unordered loop structure, given these motifs’ propensity to be flexible. The PIPM binding site itself is surrounded by four distinct flexible regions comprising β-turn residues 40–46, the disordered interdomain connecting loop residues 117–136, β-turn residues 229–235, and disordered loop residues 251–257. Of the protein residues that looked to be the most important for direct interaction with the peptoid ligands, His44, Pro129, Pro234, Ala252, Pro253, and Ile255, each interacted with ligands to varied degrees in each of the MDs (Figure S12). Many of the significant contacts were shared with T2AA and the PL peptide, indicating that the inhibitors occupy many of the “anchoring” contacts between PCNA and PL.

To further evaluate the differences between each MD output structure, a principal component analysis (PCA) of the trajectory snapshots for the Cα atoms of each PCNA–ligand complex was performed (Figure ). The principle components are based upon orthogonal eigenvectors to describe the axes of maximal variance in the distribution of structures. The percentage of variance of fluctuation in protein atom positions in each dimension is characterized by a corresponding eigenvalue. In addition to the outputs from the MDs simulations performed here, trajectories of PCNA bound to various known peptide ligands, as used in the study by Pedley, et al.,[17] were included in the PCA for comparison. In this additional set were trajectories of PCNA in complexes with the DNA polymerase δ, PL, p85-α, p21, Apo, Akt, or Abl peptide. By clustering structures in principal component space, the focus is on the relationships between different structures in terms of their major structural displacements. In the context of this work, clustering along principal components 1, 2, and 3 allows for the comparison of the significant structural differences between each conformation of PCNA which covers more than 50% of their conformational variance (Figure , bottom right panel).

Figure 6

PCA of PCNA topology variance. The final 100 frames of each PCNA–ligand MD trajectory were aligned based on the position of the PCNA Cα backbone atoms. A PCA of the aligned trajectories shows differential clustering of PCNA conformations (residues 1–257) when in complex with either a peptoid-based ligand or a PIPM-containing peptide. Principal components 1–3 (PC1, PC2, and PC3) for each structure were clustered and plotted along with the proportion of variance for each principal component.

The PCA indicated distinct differences in the topology of the PCNA–ligand interaction sites. Of all complexes, PCNA–T2AA and PCNA–NLys-NPip-NBal displayed the most distinct principal component values. Of note, the PCNA–T2AA complex is also well separated from the tripeptoid PCNA–T2AA–NEal-NTyr, implicating the influence of the second and third fragments outside of the hydrophobic T2AA binding pocket. Interestingly, this same tripeptoid containing T2AA appears to take on features like the PCNA–p21 complex. In contrast, this cluster is more dissimilar from the PCNA–NLys-NPip-NBal complex. Although the structures were separated from one another, some similarities in the eigenvectors are also noteworthy. For example, although PCNA–PL and PCNA–NLys-NPip-NBal were well separated in the clustering space, they both had nearly equivalent second-principal components. Likewise, the population distribution of PCNA–PL and PCNA–T2AA had nearly equivalent first-principal components. These results implicate similar conformations only in certain dimensions. The potential utility for predicting inhibitory efficiency from this PCA are not yet clear, but this information may be useful for classifying ligand–receptor complexes with respect to the level of inhibition.

Further analysis of the MDs simulations indicates that the peptoid inhibitors are able to disrupt key interactions between PCNA and the PL peptide (Figures and S13). Although these compounds provide a geometric hindrance to PIPM binding, it is also significant that they prevent PCNA from forming important anchoring contacts with PIPM residues. Computational results from Pedley, et al.[17] demonstrate that the conserved amino acids of the PIPM—glutamine in position 1, a hydrophobic residue in position 4, and aromatic residues in positions 7 and 8—act as anchoring residues that drive conformational stability of the complex between PCNA and PIPM-containing peptides/proteins. Disrupting these points of contact should substantially weaken their interactions and prevent binding.

Figure 7

Peptoid inhibitors disrupt key PCNA–PIPM interactions. (Top) Competitive peptoid inhibitors, such as NLys-NPip-NBal (yellow sticks), are projected to bind at the PIPM-binding site on PCNA (gray surface; PDB ID: 1VYJ), which overlaps with the PL peptide (green cartoon and sticks). (Bottom left) NLys-NPip-NBal overlays key contact points between the PL peptide’s PIPM and PCNA. Spheres represent PIPM amino acid residues. Colors indicate direct disruption of key (red) or nonkey (orange) residues and nondisruption of key (yellow) or nonkey (white) residues. (Bottom right) Results from Pedley, et al. (2014) demonstrating that residues 1, 4, 7, and 8 of the PL peptide’s PIPM act as anchoring residues. ** Changes in surface accessible surface area were calculated with ANCHOR,[50] measuring the differences between bound and unbound forms of the PL peptide.

Classifying Peptoid-Based Molecules as Inhibitors of PPIs

Inhibitors of protein–protein interactions (iPPIs) have characteristics that distinguish them from other traditional inhibitors in that they display features such as higher molecular weight, higher hydrophobicity, and a larger number of aromatic rings.[47,48] iPPIs also demonstrate higher degrees of globularity, lower distribution of hydrophilic regions, smaller proportions of exposed hydrophilic regions, and stronger capacities to bind hydrophobic patches at the core of protein–protein interfaces as compared to inhibitors of classical targets such as enzymes.[49] Although peptoid-like molecules have been demonstrated to disrupt PPIs,[51,52] it was not clear how the fragment compositions in the combinatorial libraries influence the features that would be predicted to be classified as iPPIs prima facie.

All fragments shown in Figure , including both variants of T2AA, were used to create a virtual combinatorial set of tripeptoids as before. The peptoids were characterized by implementing a Bayesian classifier method analogous to the one used in Morelli et al. (2011).[48] To perform the Bayesian classification, the 2P2I Hunter data set,[48,53,54] which is a library of molecules that contained (at the time) 40 known iPPIs and 1018 small molecules that are not inhibitors of PPIs, was first obtained. Next, four descriptors were calculated for each compound—globularity, CW2, EDmin3, and IW4—using the same methodology as Kuenemann et al. (2014).[55−57] These descriptors measure the following factors, respectively: (1) three-dimensional shape globularity; (2) ratio between the surface of the hydrophilic regions calculated at −0.5 kcal/mol and the total molecular surface [it is proportional to the concentration of hydrophilic regions (involved in weak potential polar interactions]) compared to the total surface area]; (3) unbalance between the center of mass of a molecule and the barycenter of its hydrophilic (IW) interacting regions (a high integy moment is a clear concentration of hydrophilic interacting regions at one extremity of the compound). (4) third lowest local minimum of the interaction energy (in kcal/mol) of a dry probe (it measures the potential interaction energy of the ligand with a hydrophobic object). This final criterion was selected because it often shows the greatest distinctions of iPPIs from other known classical inhibitors.

In calculating the Bayesian cutoff, each sample was left out one at a time, and a model built using the results of the samples, with that model used to predict the left-out sample. Once all samples had predictions, a ROC plot was generated, and the area under the curve (ROC AUC) calculated (Figure S14). The best split was determined by picking the split that minimized the sum of the percent misclassified for category members and for category nonmembers, using the cross-validated score for each sample. A contingency table was constructed, containing the number of true positives, false negatives, false positives, and true negatives. Based on the resulting calculated cutoff of −0.188, 38 out of the 40 iPPIs as well as 244 out of the 1018 non-iPPIs were predicted to be true iPPIs. Although the selected cutoff produced a false positive hit rate of approximately 24%, this was considered acceptable given that fewer false negatives would theoretically result, and the ROC score was rated as good (Figure S14). It is expected that increasing the number of compounds to more than those found in the 2P2I Hunter set in the future could produce an even more robust model.

Next, this same model was applied to the library of tripeptoid ligands, and the same four descriptors were calculated for each molecule—globularity, CW2, EDmin3, and IW4. The peptoids were prepared in the same way as the compounds in the training set, with all possible ionization states at a pH of 7. When considering the entire set of tripeptoids, both those that are predicted to be iPPIs and those that are not, the statistics for the distribution of descriptor scores (Tables S2–S4) indicate that those molecules are similar in globularity to other known iPPIs than non-iPPIs. For CW2, the tripeptoids had a higher score than either iPPIs or non-iPPIs, likely due to the fact that these peptoids, on average, have more exposed hydrophilic regions than what would be expected for classical drugs. Overall, the full set of tripeptoids was not strongly associated with either iPPIs or non-iPPIs, as there are ligands that strongly share features with other iPPIs, and those that do not. When only the tripeptoids that showed up as experimental hits were considered, 15 out of 16 peptoids (94%), along with T2AA, were predicted to be iPPIs, with the lone exception being Gly-NPip-NBal. Separately, a random set of 31 tripeptoids that were synthesized, but did not show up as experimental hits, were analyzed using the classification method as before. From these, 22 ligands (71%) were predicted to be non-iPPIs, whereas 9 (29%) were predicted to be iPPIs. From the results, it appeared more likely that tripeptoids that share traits with other known iPPIs would bind to PCNA in a manner sufficient to disrupt the binding of the PIPM-containing PL peptide.

In Vitro Assessment of Peptoid-Based PCNA Ligands for Synergy with DNA-Damaging Agents in Cancer Cell Lines

PCNA is involved with numerous DNA damage repair pathways, including base excision repair, mismatch repair, HR, and the Fanconi anemia pathway.[2−5] Each of these pathways involve different protein partners in complex with PCNA. T2AA has been shown to synergize with cisplatin, which causes DNA single-strand and double-strand breaks in a drug-sensitive cell line.[19,20] Inhibiting PCNA appeared to increase the amount of DNA damage as well as increase the length of time for repair in these studies. Further in vitro cell assessments of the new multifragment ligands were pursued to compare their effects with the PCNA inhibitor, T2AA.

Tumor cells that are resistant to the well-established therapeutic DNA damaging agents, cisplatin and doxorubicin show differences in their dependences on DNA repair pathways.[58] The high-affinity tripeptoids NLys-NPip-NBal and one containing the T2AA fragment (T2AA-NEal-NTyr) were resynthesized and tested for effects on tumor cell growth and viability alone and in combinations with these DNA-damage agents. When exposed to tumor cell lines as single agents, these compounds had no meaningful effects on cell growth up to 50 μM in MDA-MB-231 (Figure S15). In combination with DNA-damaging agents, cisplatin and doxorubicin, these two tripeptoids were able to enhance the effects of both DNA-damaging agents at concentrations as low as 3 μM in the drug-resistant A549 and MDA-MB-231 cell lines (Figures S16 and S17). Although the effects with cisplatin combinations are measurable using the tripeptoid PCNA inhibitors, T2AA effects were not observed in these two cell lines. However, all of the agents were able to elicit strikingly high synergistic effects of doxorubicin on total growth inhibition (TGI) over three doubling periods from 20 to 100-fold as summarized in Table . To test these effects on cytotoxicity more robustly, clonogenic assays were executed using 24 h time points of drug incubation. These data confirmed the results of the cell growth assays (Figures and S18), which clearly show the potencies of the doxorubicin–PCNA inhibitor combinations on single cell viability over a 24 h period.

Table 2

PCNA Inhibitor GI50 Values in Combination with DNA Damaging Agents

	A549				MDA-MB 231
treatment	GI50 (μM)	GI50 M/D	TGI (μM)	TGI M/D	GI50 (μM)	GI50 M/D	TGI (μM)	TGI M/D
cisplatin	14.3	N/A	21.4	N/A	4.18	N/A	9.30	N/A
cisplatin 3 μM T2AA	14.5	0.99	20.9	1.02	3.18	1.32	8.13	1.14
cisplatin 10 μM T2AA	13.8	1.04	20.1	1.07	2.99	1.40	5.95	1.56
cisplatin 30 μM T2AA	12.8	1.11	20.5	1.04	3.38	1.24	7.53	1.24
cisplatin 3 μM NLys-NPip-NBal	11.0	1.29	12.6	1.70	1.94	2.15	3.16	2.47
cisplatin 10 μM NLys-NPip-NBal	6.08	2.35	8.36	2.56	1.52	2.74	3.65	2.55
cisplatin 30 μM NLys-NPip-NBal	4.63	3.08	6.23	3.44	1.18	3.54	2.51	3.71
cisplatin 3 μM T2AA-NEal-NTyr	10.8	1.32	12.6	1.71	2.01	2.08	3.81	2.44
cisplatin 10 μM T2AA-NEal-NTyr	5.71	2.50	6.88	3.11	1.65	2.53	3.08	3.02
cisplatin 30 μM T2AA-NEal-NTyr	4.97	2.87	6.49	3.30	1.15	3.64	2.41	3.86
doxorubicin	5.66	N/A	17.2	N/A	53.3 nM	N/A	123 nM	N/A
doxorubicin 3 μM T2AA	2.75	2.06	3.78	4.55	0.87 nM	61.1	2.68 nM	45.9
doxorubicin 10 μM T2AA	2.54	2.23	3.75	4.59	0.91 nM	58.7	2.27 nM	54.2
doxorubicin 30 μM T2AA	0.50	11.1	0.59	29.1	0.62 nM	86.9	1.17 nM	105
doxorubicin 3 μM NLys-NPip-NBal	3.58	1.58	8.84	1.95	5.26 nM	10.1	6.15 nM	20.0
doxorubicin 10 μM NLys-NPip-NBal	3.19	1.78	5.64	3.05	6.74 nM	7.91	8.94 nM	13.8
doxorubicin 30 μM NLys-NPip-NBal	0.48	11.9	0.85	20.3	1.76 nM	30.4	6.02 nM	20.4
doxorubicin 3 μM T2AA-NEal-NTyr	4.24	1.34	8.83	1.95	4.31 nM	12.3	3.69 nM	33.3
doxorubicin 10 μM T2AA-NEal-NTyr	2.43	2.33	2.76	6.24	1.34 nM	39.7	3.07 nM	40.1
doxorubicin 30 μM T2AA-XEal-NTyr	0.43	13.3	0.67	25.5	0 56 nM	95.6	1.20 nM	103

Figure 8

PCNA inhibitors synergism with doxorubicin. T2AA, T2AA-NEal-NTyr, and NLys-NPip-NBal synergized with doxorubicin both in A549 and MDA-MB-231. These data used a clonogenic assay format with cell counts normalized to the plating efficiency value as 100%.

The synergy observed for the PCNA inhibitors compelled testing of the selectivity for the cell context. The drug-resistant tumor cells would be expected to exhibit a dependence on DNA repair processes due to the stress induced by mutational status that promotes a reduced cell cycle regulation. To validate the selectivity of these effects to the tumor cells, the same combinations of PCNA inhibitors and DNA-damage agents were tested for synergy in growth inhibitory effects using the noncancer human derived cell line, HEK293. This cell line is not highly sensitive to either doxorubicin or cisplatin, similar to the cancer cell lines evaluated in Figure . The data in Figures and S19 demonstrate that the PCNA inhibitors were not able to sensitize these cells to either of these DNA damaging agents.

Figure 9

PCNA inhibitors’ effect on noncancer cell growth in combination with DNA-damage agents. None of the PCNA antagonists tested showed the ability to enhance either doxorubicin (top) or cisplatin (bottom) effects on cell growth in HEK293 cells. These studies used the (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) (MTT) proliferation assay with the standard dosing described for each DNA-damaging agent.

T2AA, T2AA-NEal-NTyr, and NLys-NPip-NBal were used in combinations with cisplatin and doxorubicin. Cisplatin and doxorubicin were used at concentrations following a 3-fold dilution series over 10-steps starting at 1 mM and 200 μM, respectively. The dosing of these second agents were 3, 10, and 30 μM in the MDA-MB-231 breast cancer cell line and A549 lung cancer cell line. The GI50 and TGI were assessed after 3 d growth. The M/D ratio is the ratio of the effect of the DNA-damage agent alone versus in combination with the PCNA inhibitors. Italicized values were not significantly different from the effects of the DNA-damaging agent alone.

Discussion

FBDD is a contemporary method for small-molecule hit discovery and/or chemical probe development due to its distinct advantages over traditional higher molecular-weight chemical libraries.[22] There are challenges to optimize this approach to achieve the theoretical outcome of an exponential increase in binding affinities for two independent fragments when linked into one molecule. Protein–protein interfaces often present noncontinuous subpockets, each distinct in the chemical environments for fragment recognition. The linked multiple-fragment screens approach for simultaneous binding in distinct subpockets in a protein target offers an advancement to current methodologies. The efforts to use this approach were motivated by our prior analysis of peptide–PCNA interactions.[17,23] This process for ligand discovery benefited from significant structural information regarding the PCNA-ligand interaction sites. Using distinct receptor conformations as starting points offered a minimally biased avenue for finding the novel inhibitors. By linking multiple fragments in a single peptoid backbone for in silico screening, efficiency in selection of synthetic targets for in vitro testing to discover inhibitors of PCNA–protein interactions has been demonstrated. In addition to increasing the throughput of iterative screens, multifragment-binding approaches can enhance second and/or third fragment effects in environments where induced-fit or conformational selection are operative.[23] This can be true when searching for novel fragment pairs demonstrated by the de novo discovery of NLys-NPip-NBal. Previous studies were able to identify the small-molecule T2AA as an inhibitor in cellular systems.[18,19] This study has also been able to leverage the insights from the T2AA–PCNA binding pocket to identify second-fragment binding sites and prepare variants with suitable affinity to disrupt the binding between PCNA and a PIPM-containing peptide. In this case, the use of a more biased starting point with the T2AA–PCNA complex provided a pathway to modify the known inhibitor and reveal distinct high-affinity ligands including T2AA-NEal-NTyr.

The screening approach used here has the potential to be applied to numerous protein targets where an ideal ligand would need a high globularity and cover a relatively large surface area. Improvements in the approach could be achieved using additional in silico methods that can account for the level of receptor structure data available to define spatial orientations of subpockets. Certain limitations will persist with any scaffold for protein interaction modulators including the peptoid backbone; restrictions on molecular geometry, degrees of freedom, and C-terminal functional groups are all critical factors to consider. Interestingly, the classification analysis of peptoids that were experimental hits predicted almost all of them to be iPPIs, whereas the majority of nonhits were predicted to be non-iPPIs. It is possible that this could be used in the future as an additional computational filter based on whether a compound is predicted to be an iPPI, and this could further narrow the list of compounds to be synthesized once favorable side chains are identified.

The flexibility of PCNA is likely involved in distinct biological roles of protein partner recognition. The MD simulations performed in this work demonstrate that PCNA can adopt very unique conformations depending on the partner ligand. Our previous efforts demonstrated a range of affinities among a series of PIPM-containing peptides, and MD simulations implicated the selection of distinct PCNA conformations in the bound states.[17] These observations are substantiated by the recent report of results from more extensive MD simulations to evaluate the role of structurally disordered regions of PCNA–protein complexes. The studies reveal that a PCNA conformation-selection process by the protein or peptide ligands is consistent with the results to explain the diversity of PCNA–ligand interactions.[59] Another potential source of selective binding effects could be produced by PCNA binding at one site of the trimer that negatively modulates protein binding at other sites. This property is reflected in the variable occupancy in the trimer, as we observed previously with peptides,[17] and again here with several of the tripeptoids. Also, in the extended MD simulations, there was additional evidence for cooperativity that resulted in asymmetry of the PCNA trimer.[59] These properties are examples of features not revealed by any of the available protein crystal structures. Equally important are the observations from the recent structure–function evaluations of the naturally occurring PCNA point mutants associated with DNA repair dysfunction.[60,61] These mutant proteins demonstrated remarkable selectivity in their functional losses. For the hypomorphic PCNA mutation, Ser288Ile, which occurs at the PIPM-binding site, selective partner protein binding was also supported by the protein crystal structure analysis of the mutant protein.[62,63] A prediction of how structural features of ligand–PCNA interactions influence the efficacy of a small molecule targeting the PIPM binding site represents a continued challenge for structure-based design tools. The chemical information gained from the tripeptoid ligands, and subsequent study of the molecular contacts will provide useful insights for next generation PCNA inhibitors.

The hypothesis that a protein or ligand engagement with the PIPM binding site can impart a functionally selective impact on PCNA is one of importance and novelty. Although continued studies of the cellular mechanisms are warranted, the results of the cell-based assays support the feasibility that enhanced sensitivity is dependent upon the type of DNA damage. The results demonstrate, for the first time, a strong synergy with the novel PCNA inhibitors in highly drug resistant cellular contexts, whereas on their own, the PCNA inhibitors display no meaningful effects. The DNA-damage agents, cisplatin and doxorubicin produce damage that exhibits dependence upon separate repair pathways.[64−71] PCNA’s role in these processes are distinct and implicate the synergy to be associated with double-strand break repair.[5] This type of synergy is distinct from that discussed in previous reports for PIPM-targeted agents, T2AA[20] and AOH1160[72] where cell lines already sensitive to DNA-damage agents also showed sensitivity to PCNA inhibitors alone. In addition, the differential synergy effects with cisplatin observed here with the tripeptoids suggests a role for PCNA–ligand complexes that impart distinct effects from T2AA alone. These synergies in resistant tumor cell contexts implicate extended usage of combinations in new disease contexts for existing therapies. As a result, the potential for the approach to date argues for the value in the continued discovery and development of a family of PCNA ligands that bind in the PIPM recognition region.

Experimental Procedures

Reagents and solvents were purchased from Sigma Aldrich unless otherwise noted. Materials were repurified via recrystallization or distillation as necessary before use. NMR experiments were performed on Bruker (Bruker Corp., Billerica, MA) ARX300 (300 MHz), ARX400 (400 MHz), or DRX500 (500 MHz) instruments. Low resolution electrospray ionization (ESI) and atmospheric pressure chemical ionization (APCI) studies were carried out on an Agilent 6320 Ion Trap (Agilent Labs, Santa Clara, CA) mass spectrometer. High-resolution mass measurements were obtained on a LTQ Orbitrap XL mass spectrometer (Thermo Scientific Corp.), utilizing ESI. Molecular masses and sequences of peptides or peptoids were validated on an Applied Biosystems (Framingham, MA) matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF)/TOF 4800 mass analyzer or Applied Biosystems Voyager DE PRO mass spectrometer using either 2,5-dihydroxy benzoic acid or α-cyano-4-hydroxy cinnamic acid matrices. Thin-layer chromatography analyses were performed on Merck aluminum-backed F254 silica gel plates. Protein and peptide concentrations were determined by UV absorbance at 280 nm. Fluorescent peptide concentrations were determined by absorbance at 494 nm. Stock solutions of each polypeptoid were made by measuring the dry mass of each in predried, preweighed screw-cap vials and adding the volume of DMSO necessary to give 10 mM solutions. Stock solutions of compounds containing N-terminal 5-carboxyfluorescein were made by measuring the absorbance at 494 nm, using an extinction coefficient of 79 000 L mol–1 cm–1 and Beer’s law (A = εbc) to calculate concentration. Data analyses and graphical representations were performed in Microsoft Excel, GraphPad Prism 6, or OriginPro 2015.

General Method for Synthesis of Peptoid Trimers

Trimeric peptoids were synthesized using an adapted procedure for submonomer peptoid synthesis.[44] In summary, 0.05 mmol of Rink amide AM or MBHA resin (loading: 0.65 or 0.71 mmol/g, respectively) was transferred to a 25 mL glass fritted peptide reaction vessel and was swelled with dimethylformamide (DMF) for 30 minutes. The resin was then deprotected using two 2.5 mL portions of 20% piperidine in DMF with incubation times of 15 min for each addition at room temperature. A solution of 1.5 mL of 1 M bromoacetic acid (30 equiv) in DMF and 230 μL (29.4 equiv) of N,N′-diisopropylcarbodiimide was added, and the resin was placed on an orbital shaker for 1 h at 37 °C. At that time, the resin was washed with DMF (6×) and DCM (3×), and a solution of 1 M respective primary amine (2 M for commercially available primary amines) in DMF was added, with incubation on an orbital shaker for 2 h at 37 °C. These steps were repeated with washing steps in between to produce the desired peptoid sequence. For the coupling of T2AA, peptoids were first synthesized up to the final bromoacetic acid addition. A solution of 500 mg (19.5 equiv) of T2AA and 34 μL (39 equiv) of DIEA in 2.5 mL of DMF was added, and the resin was incubated overnight at room temperature on an orbital shaker. Peptoids were cleaved from resin using a solution of TFA/TIS/water (95:2.5:2.5), incubating the resin at room temperature for either 1 or 3 h in the case of peptoids containing NArg, Neal, or NBal side chains. TFA was removed with a steady stream of blowing air, and the remaining residue was dissolved in ACN/H2O (50:50) with 0.1% TFA, frozen and lyophilized. The peptoids were purified via high-performance liquid chromatography (HPLC) (Beckman Coulter System Gold 166 or 168) using an increasing gradient of ACN/H2O with 0.1% TFA (5:95) to (100:0) over 30 min on an Agilent ZORBAX SB-C18 reverse-phase semi-preparative column. Molecular masses were validated via low resolution ESI or APCI experiments, and exact masses were obtained by high resolution ESI or MALDI-TOF.

FP Z′-Factor Analysis

FAM-PL (10 μL, 20 nM) in FP binding buffer [25 mM N-(2-hydroxyethyl)piperazine-N′-ethanesulfonic acid (HEPES) at pH 7.4, 10% glycerol, 0.01% Triton X-100] was combined with either 10 μL of 200 nM recombinant (His)6-PCNA protein in binding buffer or 10 μL of binding buffer in each of 48 wells on a ProxiPlate-384 F Plus low volume, black, opaque plate (24 replicates per set). The plate was allowed to incubate at room temperature in the dark for 30 min prior to fluorescent measurement. FP and resultant anisotropy were measured on a BioTek Synergy 4 multi-detection microplate reader (BioTek Instruments Inc., Winooski, VT) using an excitation filter of 485 nm and an emission filter of 530 nm, each with a 20 nm band-pass. The average of each control set was calculated along with the standard deviation. The Z′-factor was calculated using eq where σ+ is the standard deviation of the positive control (FAM–PL peptide in the presence of PCNA protein), σ– is the standard deviation of the negative control (FAM–PL peptide in the absence of PCNA protein), and μ+ and μ– are the mean anisotropy values of the positive and negative controls, respectively.

FP Binding Assay

Increasing amounts of recombinant (His)6-PCNA protein were prepared in the FP binding buffer (25 mM HEPES at pH 7.4, 10% glycerol, 0.01% Triton X-100), with an 11-step 2-fold dilution series, and a top concentration of 30 μM. Each solution (10 μL) was combined with 10 μL of 20 nM FAM-PL peptide formulated in the FP binding buffer in a single well of a ProxiPlate-384 F Plus low volume, black, opaque plate (PerkinElmer). Each concentration of protein was plated in a replicate of four, and the plate was allowed to incubate at room temperature in the dark for 30 min prior to fluorescent measurement. FP and resultant anisotropy were measured on a BioTek Synergy 4 multi-detection microplate reader using an excitation filter of 485 nm and an emission filter of 530 nm, each with a 20 nm band-pass. The parallel and perpendicular intensity values for each sample (n = 4) were used to calculate FO of the FAM–PL peptide bound to monomeric PCNA using eq .where ,, , and fb is the fraction of FAM–PL bound to PCNA, r is the observed anisotropy value, rf is the anisotropy of free unbound FAM–PL peptide, rb is the anisotropy of FAM–PL peptide saturated with PCNA protein, Q is the ratio of quantum yield of bound (qb) to free (qf) FAM–PL peptide, ∥f and ∥b are the parallel intensities of free unbound and saturated FAM–PL peptides, respectively, and ⊥f and ⊥b are the perpendicular intensities of free unbound and saturated FAM–PL peptides, respectively.

FO values were analyzed using nonlinear regression statistics in OriginPro 2015, representing them as the mean ± standard error of the mean (Y) and plotting them as a function of the monomeric PCNA protein concentration (X). From this, eq was used to obtain a dissociation constant (Kd) for FAM–PL.where n is the Hill slope.

FP Competition Assay

Solutions of the competitive ligand were formulated from DMSO stocks (10 mM for tripeptoids, 20 mM for T2AA) into FP binding buffer (25 mM HEPES at pH 7.4, 10% glycerol, 0.01% Triton X-100) at appropriate 2× (relative to the desired effective screening concentration) concentrations. Each competitive ligand (10 μL) was combined with 5 μL of 4 μM recombinant (His)6-PCNA protein in binding buffer and 5 μL of 40 nM FAM–PL in binding buffer into each well of a ProxiPlate-384 F Plus low volume, black, opaque plate, in replicates of four. DMSO at an equivalent concentration in binding buffer to the concentration of DMSO in the competitive ligand sample was used as a negative control; T2AA at 1 mM was used as a positive control. The plate was allowed to incubate at room temperature in the dark for 30 min prior to fluorescent measurement. FP and resultant anisotropy were measured on a BioTek Synergy 4 multi-detection microplate reader using an excitation filter of 485 nm and an emission filter of 530 nm, each with a 20 nm band-pass. Anisotropy values were converted to FO using eq , and IC50 values were calculated by fitting the data to eq .where n is the Hill slope.

Inhibition constants (Ki) for the competitive ligands were determined using eq , which is a modified form of the Cheng–Prusoff equation, previously reported for FP assays.[73]where [I]50 is the concentration of each competitive peptoid at 50% inhibition, [L]50 is the concentration of the FITC-PL peptide at 50% inhibition, [P]0 is the concentration of monomeric PCNA protein at 0% inhibition, and Kd is the dissociation constant obtained from eq .

Molecular Dynamics Simulations

In preparation for MDs simulations, selected hit peptoids were flexibly docked into the PIPM binding site of the cocrystal structure of PCNA–Pogo Ligase (PL) peptide (PDB ID: 1VYJ), with the peptide itself removed, using the Schrödinger Glide’s induced-fit docking model. The PCNA crystal structure was prepared using the Protein Preparation Wizard in Maestro, with PCNA minimized in complex with the PL peptide using the OPLS-2005 force field and implicit solvation. PCNA–peptoid complexes were then explicitly solvated in Schrödinger’s Desmond[74] using the TI3P water model in the presence of 0.15 M sodium chloride buffer to generate orthorhombic water boxes that contained a 10 Å buffer region. Each system was then minimized with the OPLS-2005 force field.

The MDs simulations were performed in the same way as described in Pedley, et al.,[17] using Desmond and the OPLS-2005 force field. In summary, long-range electrostatic interactions were determined using a smooth particle mesh Ewald method with a grid spacing of 0.8 Å. For nonbonded van der Waals interactions, a cut off of 9.0 Å was set. All simulations were performed for 5.0 ns, except in cases where simulations did not fully converge after 5.0 ns (simulations were extended by 2.5 ns in those situations), using the Desmond NPT ensemble with a six-step slow relaxation protocol prior to the MDs run: (i) 2000 step limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) minimization with a loose convergence restraint of 50 kcal/mol/Å; (ii) 2000 step L-BFGS minimization with a convergence constraint of 5 kcal/mol/Å; (iii) a 12 ps Berendsen NVT simulation at a temperature of 10 K with restraints on solute heavy atoms; (iv) a 12 ps Berendsen NPT ensemble at a temperature of 10 K and pressure at 1.01325 bar with restraints on solute heavy atoms; (v) a 24 ps Berendsen NPT ensemble at a temperature of 300 K and a pressure at 1.01325 bar with restraints on solute heavy atoms; (vi) a 24 ps Berendsen NPT ensemble at a temperature of 300 K and a pressure at 1.01325 bar with restraints on residues beyond 15 Å of the restrained ligand. The 5.0 ns MDs simulation run was performed using the NPT ensemble. Temperature of the simulation was kept at 300 K using a Nosé–Hoover thermostat. Pressure was maintained at 1.01325 bar using the Martyna–Tobias–Klein method. Energy and trajectory data were recorded at every 1.2 and 5.0 ps, respectively.

Upon completion of each simulation, PCNA trajectory data were processed in VMD[75] after removal of each peptoid ligand. For each trajectory, the protein backbone Cα atoms were aligned to the first frame of the simulation to generate rmsd and Cα fluctuations (RMSF) values. Simulations were determined to be converged once rmsd values had stabilized (slope of the rmsd curve leveled off horizontally over the period of the final 0.5–1.0 ns). In preparation for PCA, trajectories for each PCNA–peptoid system were overlaid using the alignment tools in VMD. Additionally, trajectories previously generated from Pedley, et al.,[17] including the systems where PCNA is in complex with the polymerase δ, PL, p85α, p21, Apo, Akt, or Abl peptides, were overlaid with the PCNA–peptoid trajectories for purposes of comparative analysis. PCA were performed using the Bio3D package[76] in R to analyze the conformational differences between the aligned trajectories over the period of the final 0.5 ns for each simulation (100 snapshots per PCNA–ligand system). The first two orthogonal eigenvectors (principal components—PC1 and PC2) were plotted. Average trajectory coordinates for the final 50 frames of each simulation were performed in VMD to generate average overall conformations of PCNA in complex with each peptoid.

Bayesian Classifier Method

The Bayesian classifier method was generated using the 2P2I Hunter database containing a total of 1058 compounds, 40 of which are iPPIs (actives) and 1018 are non-iPPIs (decoys).[48,53,54] All compounds in that set were converted from 2D to 3D and minimized using the Accelrys Discovery Studio 4.1 Visualizer (Accelrys, San Diego, CA).[77] The tripeptoid ligands consisting the test set were prepared using LigPrep in the Maestro Schrödinger software suite.[41] Four descriptors (Glob, EDmin3, IW4, and CW2) were calculated for each compound according to the method developed by Kuenemann, et al.[49] A Laplacian-corrected Bayesian classifier model was generated using the Discovery Studio visualizer.

MTT Proliferation Assay

MDA-MB-231 and A549 cell proliferation was measured using the MTT cell proliferation reagent. Cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) (phenol-red free) supplemented with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin at 37 °C with 5% CO2. Cells were then plated on flat bottom 96-well plates at a density of 5 × 103 cells per 80 μL per well. Cells were attached for 4 h and then treated with 20 μL of each compound for 72 h. For each experiment involving doxorubicin, we used 200 μM and then performed a 9-step dilution series using a factor of 3 to produce 10 concentrations. For each experiment involving cisplatin, a 1 mM stock was diluted in a 9-step series using a factor of 3 to produce 10 concentrations. Ten microliters of MTT (5 mg/mL in DMEM phenol-red free media) were added to each well at a final concentration of (0.5 mg/mL). After 4 h incubation at 37 °C, 50 μL of the solubilization solution [10% Triton-X 100, acidic isopropanol (0.1 N HCl)] was added and the plates were sealed and stored from light for 3 d. Absorbance was read at 570 nm, and the percent cell growth was normalized by comparison to a day zero control plate with no drug for each respective cell line. The day zero plate standardized 0 and 100% cell viability, with columns plated in triplicates with either 100 μL of DMEM alone or 5 × 103 cells in 100 μL of DMEM. No inhibitors were added; instead, 10 μL of MTT reagent was added to each well after which cells were allowed to attach. After 4 h, 50 μL of solubilization solution was added to the plate, before sealing and storage in the dark for 3 d. Data were then analyzed in Prism to calculate TGI and GI50. Synergy effects were validated utilizing the Choy–Talalay method.[78]

Clonogenic Assay

MDA-MB-231 and A549 long-term survival was studied using a clonogenic assay. Cells were grown in DMEM (phenol-red free) supplemented with 10% FBS at 37 °C with 5% CO2. They were then plated on flat bottom 6-well plates at a density of 1 × 103 cells with 1 mL per well. Cells were allowed to attach for 4 h before treatment with 5 concentrations of each DNA-damaging agent and incubation for 24 h. For drug combination studies, cells were also exposed to a single constant dosage of 3, 10, or 30 μM PCNA inhibitor. Cells were then washed with PBS and incubated in 1 mL DMEM with 10% FBS at 37 °C under 5% CO2. When visible colonies could be identified (typically 6–7 d), the cells were fixed in formaldehyde and stained with DAPI solution for counting using a Cytation 3. The results were normalized by comparison with the counts obtained using control cells incubated in drug-free medium.