Small molecule inhibitors of DNA repair are emerging as potent and selective anticancer therapies, but the sheer magnitude of the protein networks involved in DNA repair processes poses obstacles to discovery of effective candidate drugs. To address this challenge, we used a subtractive combinatorial selection approach to identify a panel of peptide ligands that bind DNA repair complexes. Supporting the concept that these ligands have therapeutic potential, we show that one selected peptide specifically binds and noncompetitively inactivates DNA-PKcs, a protein kinase critical in double-strand DNA break repair. In doing so, this ligand sensitizes BRCA-deficient tumor cells to genotoxic therapy. Our findings establish a platform for large-scale parallel screening for ligand-directed DNA repair inhibitors, with immediate applicability to cancer therapy. Cancer Res; 71(5); 1816–24. ©2011 AACR.
Mammalian cells rely on vast interactive DNA repair networks to preserve genomic integrity in the face of constant genotoxic stress (1). Loss of DNA repair fidelity predisposes cells to mutation, genomic instability, and carcinogenesis (2). Consequently, tumors often bear intrinsic DNA repair deficiencies absent in noncancerous normal tissues and these defects may represent tumor-specific therapeutic targets. The recent success of PARP inhibitors against BRCA-deficient tumors has generated great interest in the pursuit of new tumor-specific drugs targeting DNA repair networks (3–5). However, both the magnitude and complexity of the DNA repair supramolecular networks pose a serious obstacle to this goal and, currently, there is an underdeveloped repertoire of readily translatable molecular targets for this purpose.
Functional ligand-directed screening may provide an opportunity to accelerate the discovery of molecular targets within these pathways. This approach offers several advantages for mapping the DNA repair protein network: (i) its use of degenerate peptide sequences yields theoretically complete coverage of all possible binding sites; (ii) as it is nonbiased, it has the potential to uncover targets within factors not known to be involved in DNA repair; and (iii) it selects for ligands binding accessible and active motifs, thus enriching for the recovery of functional small molecules potentially useful as DNA repair inhibitors. As such, we use here subtractive combinatorial ligand-directed mapping of DNA repair complexes to guide the selection of peptides potentially useful as DNA repair inhibitors.
Materials and Methods
A phage random peptide library displaying the insert XXXXYXXXX (X, any residue; Y, tyrosine) was generated. pSP72 plasmid DNA (Promega) was linearized, biotinylated, and bound to magnetic streptavidin-coated beads (Dynabeads M-280; Invitrogen) at a DNA:bead molar ratio of 60,000:1. Nuclear extracts were prepared as described elsewhere (6). For negative counterselection, 1 mg of DNA-coated magnetic beads was incubated with 108 phage particles in PBS for 30 minutes at room temperature (RT). Unbound phage were collected, and positive selection was subsequently carried out on 1 mg of DNA-coated magnetic beads prebound to 100 μg of nuclear protein (37°C for 30 minutes) and incubated in PBS for 30 minutes at RT. This method of using DNA-coated beads to capture functional DNA repair proteins from nuclear abstracts was adapted from extensively validated published techniques (7–9). Bound phage particles were collected by the BRASIL (Biopanning and Rapid Analysis of Selective Interactive Ligands) method (10). After 3 rounds of selection, the recovered phage library was introduced into host bacteria, individual clones were selected on LB agar plates, and the phage DNA corresponding to the peptide insert was sequenced with established primers by methods described previously (11).
For the first 500 peptides recovered from our screen, similarity analysis was done using the NCBI human protein BLAST database search engine. To test for clustering of BLAST hits within functional annotated pathways, the top 10 hits for each peptide (5,000 hits total) were subjected to Ingenuity Pathway Analysis (12). To increase the stringency of the similarity analysis, BLAST results were filtered by the Entrez query “DNA damage repair” and the analysis was limited to results with scores of more than 18.0 bits. These results were considered positive only when peptides resembled domains in DNA repair proteins that are known to be functionally conserved (according to the Conserved Domain Database) or with protein sequences highly conserved among vertebrates. These results were organized using an online protein clustering algorithm (13), and a heatmap was generated to show quantitative results.
Biotinylated and nonbiotinylated synthetic custom peptides [GNFRYLAPP, RYPAGLPFL (scr), RQIKIWFQNRRMKWKK (pen), GNFRYLAPP-pen, and RYPAGLPFL-pen] were produced by PolyPeptide Labs and were reconstituted with sterile water (1 mmol/L).
DNA-PKcs precipitation assay
Biotinylated peptides (10 μmol/L) were incubated with 50 μg of pooled nuclear extract, with sonicated DNA (10 μg), ATP (1 μmol/L), and/or excess free nonbiotinylated GNFRYLAPP (100 μmol/L) at 37°C for 15 minutes. Fifty micrograms of streptavidin-bound beads was added to precipitate peptide-bound proteins. After 3 washes, recovered proteins were resolved on a polyacrylamide gel and analyzed via Coomassie staining and Western blotting.
Streptavidin-bound 96-well plates (Roche) were coated with 100 ng of biotinylated, linearized pSP72 or 10 ng of biotin-GNFRYLAPP (vs. biotin-scrambled control). MaxiSorp 96-well plates (Fisher) were coated with 250 units of pure DNA-PK (Promega). After blocking, prey was added to each baited well and incubated at 37°C for 1 hour with 1 μmol/L ATP, sonicated DNA (50 ng/μL), and a protease/phosphatase inhibitor cocktail (Sigma). To compete with GNFRYLAPP/DNA-PKcs interactions, we added nonbiotinylated GNFRYLAPP or scrambled control peptide. To compete with DNA-PKcs/DNA interactions, we added either GNFRYLAPP or scrambled control peptide. We quantified DNA-PKcs immunoreactivity with fluorophore-conjugated antibodies detected on a fluorescence microplate reader. We quantified GNFRYLAPP reactivity with FITC-conjugated streptavidin (1:500, #016-010-084; Jackson Immunoresearch).
DNA-PKcs kinase assay
DNA-PKcs kinase activity assays were carried out with the SignaTect kit (Promega) according to the manufacturer's instructions, using pure DNA-PK (Promega) as the catalyst. Nonbiotinylated GNFRYLAPP and wortmannin (WM; 10 μmol/L) served as controls.
DNA-PKcs autophosphorylation assay
Pure DNA-PK (250 units; Promega) was incubated for 7 minutes at 37°C in 1 mmol/L HEPES with 1 μmol/L γ-32P-ATP and protease/phosphatase inhibitor cocktail (Sigma), with or without sonicated DNA (50 ng/μL), GNFRYLAPP (10 μmol/L), DMNB (10 μmol/L), or WM (10 μmol/L). Reactions were resolved on a polyacrylamide gel and phosphorylation was detected by autoradiography. An identical experiment was run in parallel with cold ATP, and DNA-PKcs autophosphorylation was detected with a mouse anti-phospho-T2609 DNA-PKcs antibody (#ab18356; Abcam).
Electrophoretic mobility shift assay
Custom oligonucleotides (Sigma) were hybridized and end-labeled with α-32P-dCTP by Klenow (New England Biolabs). Free radioactive nucleotide was removed with a PCR cleanup column (Qiagen). Labeled probe (4,000 cpm) was incubated at 37°C for 10 minutes with 250 units of pure DNA-PK (Promega) in 1 mmol/L HEPES buffer, 1 μmol/L ATP, and protease/phosphatase inhibitor cocktail (Sigma). As indicated, GNFRYLAPP (10 μmol/L) or a 10-fold molar excess of cold DNA probe was added. Reactions were resolved on a nondenaturing 4% to 20% PAGE gel (Bio-Rad), and mobility shifts were detected by autoradiography.
Protease protection assay
Standard techniques were used, as previously described (14). Briefly, pure recombinant DNA-PK (250 units; Promega) was incubated for 7 minutes at 37°C in 1 mmol/L HEPES with 1 μmol/L ATP and a phosphatase inhibitor cocktail, with or without sonicated DNA (50 ng/μL), GNFRYLAPP (10 μmol/L), or scrambled control peptide (10 μmol/L). Trypsin (Gibco) was then added (0.01%, w:v), and the mixture was incubated for 8 minutes at 37°C. Reactions were terminated by boiling and resolved on a 5% SDS-PAGE gel (Bio-Rad). DNA-PKcs fragments were detected with a polyclonal antibody (1:1,000, #ab18356; Abcam).
All cell lines were acquired from the American Type Culture Collection and cultured according to their recommendations, with the exception of the human KS1767 Kaposi sarcoma cell line (a gift from Dr. S. Leventon-Kriss). All peptides were added to standard growth media 72 hours prior to irradiation or mock irradiation. Peptides were used at a concentration of 10 μmol/L, as this was 2-fold lower than the concentration needed to see any toxicity by MTT assay (not shown).
Cells were irradiated using a 137Cs source (Shepherd Instruments), with a dose rate of 333 cGy/min.
Cells were plated on coverslips. After irradiation, at the indicated times, cells were fixed for 3 minutes in a 1:1 mixture of ice-cold acetone:methanol. After washing, the cells were then blocked for 1 hour at RT with 5% goat serum. Primary antibodies, in 1% bovine serum albumin, were then incubated overnight at 4°C. After washing, secondary antibodies were incubated for 1 hour at RT. Coverslips were mounted onto slides and imaged on an inverted fluorescence microscope.
One hundred fifty minutes following 6 Gy radiation, 1,000 cells were suspended in 0.5% low-melting temperature agarose and plated on microscope slides precoated with 1% agarose. Mounted cells were incubated at 4°C for 30 minutes in neutral lysis buffer (2.5 mol/L NaCl, 100 mmol/L EDTA, 10 mmol/L Tris base, 1% sodium lauryl sarcosinate, and 1% Triton X-100, pH 7.6). The slides were next equilibrated in TAE (pH 7.6), followed by electrophoresis at 40 V for 30 minutes. The slides were fixed in 70% ethanol for 10 minutes, DNA was stained with SybrGreen (Invitrogen), comets were imaged on an inverted fluorescence microscope, and unrepaired DNA damage was scored by CometScore (TriTek). One hundred fifty comets were analyzed per treatment and quantified as previously described (15).
Cells were irradiated with the doses described after 72 hours preincubation in the indicated conditions. Twenty-four hours later, the cells were replated at low density, under normoxic conditions, without peptide and/or reagent. Between 7 and 14 days later, colonies containing at least 50 cells were scored and results were normalized to the plating efficiency observed for nonirradiated cells (16, 17).
Antibodies used were as follows: anti-DNA-PKcs (1:1,000, #ab230; Abcam), anti-53BP1 (1:1,000, #612523; BD Biosciences), anti-53BP1 (1:1,000, #ab21083, Abcam), anti-BRCA1 (1:1,000, #ab47429; Abcam), anti-phospho-T2609 DNA-PKcs (1:1,000, #ab18356; Abcam), anti-γH2AX (1:1,000, #ab2893, Abcam), anti-mouse IgG (1:10,000, #ab6789; Abcam), anti-mouse IgG (1:5,000, #ab6787; Abcam), anti-rabbit IgG (1:5,000, #ab6793; Abcam), anti-rabbit IgG (1:1,000, #ab6939; Abcam), and anti-rabbit IgG (1:10,000, #111-035-006; Jackson ImmunoResearch).
Tests for statistical significance between groups were carried out using 2-tailed Student's t tests, Spearman rank, or log-ranked analyses, as indicated, with the level of significance defined as a value of P ≥ 0.05.
Construction and characterization of a DNA repair ligand library
A linear phage display library of 108 unique insert sequences was produced with the general peptide arrangement XXXXYXXXX (X, any residue; Y, tyrosine) displayed on pIII. Peptide binding to DNA repair factors was enriched by 3 successive rounds of selection on immobilized DNA–nuclear protein complexes after an initial negative counterselection on immobilized DNA alone (Fig. 1). Blunt-ended, linear, double-stranded DNA served as bait for nuclear proteins to focus the screen on double-strand DNA break (DSB) repair factors.
To assess for enrichment of DNA repair–interacting ligands, similarity analyses were done for the first 500 peptides in the selected library versus an online database of human proteins (18). Functional pathway analyses (12) were done on the top 10 BLAST hits for each (5,000 hits total). This revealed significant enrichment of proteins clustering into known functional pathways, including DNA damage repair (Fig. 2A). To further increase the stringency of the analysis, BLAST hits were also separately reduced to only those wherein the queried peptide mimicked a conserved functional domain in the homologous protein. Among the peptides retained after this analysis, many resembled critical DNA repair factors whose absence causes recognized syndromes of genetic instability in humans, mapped to residues in these factors known to be mutated in human cancers, and overlapped with domains within these factors having defined roles in protein–protein interactions, protein–DNA interactions, or ATP binding (not shown). An interactivity map was created to display proteins with conserved domains mimicked by the recovered peptides, which shows extensive coverage of DSB repair pathways (Fig. 2B).
GNFRYLAPP is a noncompetitive DNA-PKcs inhibitor
To explore whether mapped ligands might serve as DNA repair inhibitors, we produced one using solid-phase Merrified synthesis for functional analysis (GNFRYLAPP). GNFRYLAPP was chosen because of its similarity to Ku Autoantigen Related Peptide 1 (KARP-1), a primate-specific protein expressed as an in-frame N-terminal extension on Ku80 (Fig. 3A). Through undefined mechanisms, KARP-1 is known to modulate the activity of DNA-PKcs (19), a phosphatidylinositol 3-kinase-related kinase (PIKK) family member that plays a critical role in DSB repair (20, 21). GNFRYLAPP resembles an uncharacterized conserved motif in the N-terminal half of KARP-1 between its putative leucine zipper and nuclear localization domains (Fig. 3B), potentially responsible for its regulation of DNA-PKcs.
We first examined whether GNFRYLAPP and DNA-PKcs physically interact. The synthetic peptide GNFRYLAPP precipitated DNA-PKcs from nuclear extracts and did so in a DNA- and ATP-dependent manner (Fig. 4A). This interaction was specific, as it was inhibited by excess free GNFRYLAPP but not by scrambled peptide; also, purified recombinant DNA-PK was sufficient for attracting GNFRYLAPP (Fig. 4B).
To determine whether this interaction has functional consequences, we next examined the effects of GNFRYLAPP on DNA-PKcs activation and inactivation. When functioning normally, DNA-PKcs is activated by DNA binding and phosphorylates its downstream targets, participates in the recombination of DNA ends, and then deactivates through autophosphorylation and disengaging from DNA (20–22). Increasing molar ratios of GNFRYLAPP progressively inhibit the ability of DNA-PKcs to phosphorylate its p53-derived consensus substrate (Fig. 4C), suggesting the peptide induces an activation defect in its target. GNFRYLAPP inhibits DNA-induced DNA-PKcs autophosphorylation as well (Fig. 4D and E), suggesting that it also affects target inactivation.
In considering how GNFRYLAPP causes multiple defects in DNA-PKcs function, we first hypothesized that the peptide might interfere with the recruitment of DNA-PK to DNA breaks. However, DNA precipitated DNA-PKcs from nuclear extracts equally well in the presence and absence of GNFRYLAPP (Fig. 4F). Moreover, GNFRYLAPP did not affect the electrophoretic mobility shift in DNA induced by incubation with purified recombinant DNA-PK (Fig. 4G).
We subsequently reasoned that GNFRYLAPP might function as an inhibitor of the conformational change induced in DNA-PKcs on binding to DNA, known to be required for its proper functioning (23, 24). To study its conformation, we examined patterns of protease protection in DNA-PKcs induced by its binding to DNA. In its open conformation, DNA-PKcs was digested by trypsin to an extent that no fragments remained detectable by immunoblot; in contrast, when bound to DNA, DNA-PKcs was partially protected from trypsin such that bands of approximately 80 and 130 kDa were detectable (Fig. 4H). GNFRYLAPP prevented the protection from trypsin digestion conferred on DNA-PKcs by DNA binding (Fig. 4H), suggesting that GNFRYLAPP blocks the aforementioned conformational change in DNA-PKcs. Taken together, these data show that GNFRYLAPP functions as a noncompetitive inhibitor of DNA-PKcs by blocking its DNA-induced conformational change and thereby suppressing its kinase activity.
GNFRYLAPP impairs DNA repair in homologous recombination-defective cells
To elucidate a biological role for GNFRYLAPP in mammalian cells, we synthesized a cell-internalizing version of the peptide by chemically fusing it in tandem to penetratin (pen), a 16-mer derived from the third helix of the homeodomain of the Drosophila melanogaster transcription factor Antennapedia. Added to the media of cells in tissue culture, penetratin-fused peptides undergo rapid cellular uptake and eventually distribute in various subcellular compartments, including the nucleus (25).
We first sought evidence for a physical and functional interaction between GNFRYLAPP and DNA-PKcs in cellulo. GNFRYLAPP-pen formed foci in the nuclei of irradiated tumor cells, and no such focus formation was seen in cells lacking DNA-PKcs (Fig. 5A, left). We observed delayed resolution of GNFRYLAPP-pen foci over time, with the majority of cells retaining nuclear foci as late as 12 hours following irradiation, potentially reflecting an acquired defect in DSB processing in the presence of GNFRYLAPP-pen (Fig. 5A, right). In support of this hypothesis, GNFRYLAPP-pen also suppressed DNA-PKcs autophosphorylation following irradiation of tumor cells (Fig. 5B), suggesting that the peptide blocks activation of DNA-PKcs in cellulo. We next hypothesized that GNFRYLAPP-pen would impede DSB repair in cells and thereby might sensitize cells to DSB-inducing cytotoxic therapy, such as ionizing radiation. In light of the availability of alternative repair pathways in DNA-PKcs–deficient cells (26), we hypothesized that GNFRYLAPP-pen would show maximal inhibition of DNA repair in homologous recombination-deficient cells. In agreement with this hypothesis, GNFRYLAPP-pen both impaired DSB repair in cells lacking functional BRCA1 or BRCA2 and sensitized these cells to ionizing radiation (Fig. 5C–E).
Considering the potential therapeutic application of this peptide ligand, we next sought to further define the structure necessary for DNA-PKcs inhibition (Fig. 6A). Alanine scanning revealed GNFRY as the minimal required structure for DNA-PKcs inhibition. Tyrosine phosphorylation abolished the peptide's DNA-PKcs–inhibiting activity, as did mutating it back to GSFRYL, the consensus KARP-1 sequence. Serine phosphorylation of the KARP-1 sequence (GpSFRYL) reestablished DNA-PKcs inhibition, suggesting that GNFRYL serves as a mimic for the native active structure GpSFRYL. A 3-dimensional model of KARP-1 was generated (Fig. 6B) to view the mimicked domain from the perspective of its DNA-PKcs–binding site (19), which shows a high degree of accessibility of the aforementioned residues for protein binding.
Although phase I clinical trials have shown that DNA repair inhibitors may represent powerful and specific anticancer treatments, reactivating mutations in targeted pathways will likely lead to therapeutic resistance (3, 27–29), making the discovery of additional synthetically lethal interactions in DNA repair pathways critically important for the ultimate success of this emerging class of drugs. The platform described here provides an opportunity to accelerate the development of potential drug targets in the DNA repair network by allowing the simultaneous (i) discovery of targetable and functional motifs, (ii) prioritization of targets by functional interaction, and (iii) unbiased exploration for proteins with unrecognized roles in DNA repair.
To validate the utility of this approach, we selected GNFRYLAPP as a lead candidate over other promising peptides returned in the screen. The basis for this choice was first that the peptide mimics a small protein, increasing the odds that an oligopeptide might be sufficient to recapitulate a minimal structural and functional domain. Second, the region within KARP-1 mimicked by GNFRYLAPP is well conserved among primates and is nestled between 2 putative localization domains—a nuclear localization signal and a leucine zipper domain (19). The short, interspersed domain is an ideal candidate, then, for carrying out the DNA-PKcs–regulating function of KARP-1.
As shown by our cellular data, GNFRYLAPP shows therapeutic potential as a DNA-PKcs inhibitor. Many DNA-PKcs inhibitors have shown promising anticancer activity in preclinical models, only to fail clinical development due to unacceptable toxicity levels. To date, there are no DNA-PKcs inhibitors available for clinical use. Potentially addressing an unmet clinical need, then, GNFRYLAPP may have immediate therapeutic relevance in oncology. Although toxicology aspects of its development are yet to mature, peptide-based drugs typically have favorable toxicity profiles and it is likely that GNFRYLAPP will represent a deliverable alternative to existing DNA-PKcs inhibitors. Further work is needed to determine whether this will be the case.
GNFRYLAPP may also provide added value to the repertoire of DNA-PKcs inhibitors by virtue of being the only available compound that functions through a noncompetitive mechanism. As a result of its unique mechanism of action, GNFRYLAPP has several distinctive biochemical functions. First, though GNFRYLAPP inactivates the kinase activity of DNA-PKcs (Fig. 4B and C), DNA-PKcs must first be stimulated by a DSB before it can physically interact with GNFRYLAPP (Fig. 4A). The fact that DNA-PKcs must first engage a DSB before binding to GNFRYLAPP is reinforced by the finding that the affinity of DNA-PKcs for DNA is unaffected by GNFRYLAPP (Fig. 4F and G). This behavior is reflected by the cellular data, which show that the formation of GNFRYLAPP-pen foci occurs only after DNA-PKcs is stimulated to bind to double-strand breaks following irradiation. All of these data point to the possibility of a “quasi”-activated state in DNA-PKcs, where binding to a DSB creates a biochemical environment suitable for further downstream protein–protein interaction (with KARP-1, for example) but during which time the conformational change required for kinase activation has not yet occurred. By preventing the conformational change induced in DNA-PKcs on binding DNA, GNFRYLAPP blocks activation of the enzyme at the DSB. This distinguishes its mechanism of action from the active site inhibitors and thus may establish a novel class of DNA-PKcs inhibitors with immediate clinical applicability.
Even in the event that GNFRYLAPP does not show an improved therapeutic ratio compared with existing DNA-PKcs inhibitors, the true value of having pursued combinatorial drug selection for DSB inhibitors may lie in the sheer volume of candidate molecules uncovered. Many of the as yet uncharacterized peptides recovered in our initial screen may yield additional novel DNA repair inhibitors. Priority for further validation may be given to those mimicking domains within proteins for which structural data are available, as these data provide a significant advantage for structure-based rational drug design. Our ability to annotate these domains as likely being functionally involved in DSB repair ought to substantially increase the efficiency of such efforts. However, it should be noted that the screening method described here is not likely to have covered all DSB reactome protein domains comprehensively. The screen was designed to select for high-affinity reactions at the DSB end, and these represent a subset of all reactions relevant to DSB repair. We reasoned, though, that these high-affinity interactions might be the most therapeutically relevant and therefore a reasonable focus for initial efforts; our results with GNFRYLAPP suggest that this reasoning is sound.
In summary, we have used here subtractive combinatorial selection to guide the discovery of GNFRYLAPP as a noncompetitive inhibitor of DNA-PKcs and a candidate anticancer therapy. This approach may establish a foundation for rapidly expanding the discovery and exploitation of ligand-based DNA repair inhibitors for cancer treatment.
Disclosure of Potential Conflicts of Interest
The University of Texas and some of its researchers (W. Arap and R. Pasqualini) have equity in Mercator Therapeutics, which is subjected to certain restrictions under university policy; the university manages and monitors the terms of these arrangements in accordance to its conflict-of-interest policies.
B.J. Moeller, R.L. Sidman, R. Pasqualini, and W. Arap designed research; B.J. Moeller conducted research; R. Pasqualini and W. Arap contributed reagents/analytic tools; B.J. Moeller, R.L. Sidman, R. Pasqualini, and W. Arap analyzed data; B.J. Moeller, R.L. Sidman, R. Pasqualini, and W. Arap wrote the manuscript.
The authors thank Drs. Lei Li, Roberto Rangel, and Lilianna Guzman-Rojas for discussions and Dr. E. Helene Sage for critical reading of the manuscript.
This work was supported by grants from NIH and DOD and by awards from the Gillson-Longenbaugh Foundation, AngelWorks, and the Marcus Foundation (W. Arap and R. Pasqualini).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.