A yeast two-hybrid system for the screening and characterization of small-molecule inhibitors of protein–protein interactions identifies a novel putative Mdm2-binding site in p53

Protein–protein interactions (PPIs) are fundamental to the growth and survival of cells and serve as excellent targets to develop inhibitors of biological processes such as host-pathogen interactions and cancer cell proliferation. However, isolation of PPI inhibitors is extremely challenging. While several in vitro assays to screen for PPI inhibitors are available, they are often expensive, cumbersome, and require large amounts of purified protein. In contrast, limited in vivo assays are available to screen for small-molecule inhibitors of PPI. We have engineered a yeast strain that is suitable for screening of small-molecule inhibitors of protein-protein interaction using the Yeast 2-hybrid Assay. We have optimised and validated the assay using inhibitors of the p53-Mdm2 interaction and identified a hitherto unreported putative Mdm2-binding domain in p53. We report a significantly improved and thoroughly validated yeast two-hybrid (Y2H) assay that can be used in a high throughput manner to screen for small-molecule PPI inhibitors. Using the p53-Mdm2 interaction to optimize the assay, we show that the p53-Mdm2 inhibitor nutlin-3 is a substrate for the yeast ATP-binding cassette (ABC) transporter Pdr5. By deleting nine ABC transporter-related genes, we generated a ABC9Δ yeast strain that is highly permeable to small molecules. In the ABC9Δ strain, p53-Mdm2 interaction inhibitors, like AMG232 and MI-773, completely inhibited the p53-Mdm2 interaction at nanomolar concentrations in the Y2H assay. In addition, we identified a conserved segment in the core DNA-binding domain of p53 that facilitates stable interaction with Mdm2 in yeast cells and in vitro. The Y2H assay can be utilized for high-throughput screening of small-molecule inhibitors of PPIs and to identify domains that stabilize PPIs.


Background
A multitude of biological processes are dependent on protein-protein interactions (PPIs). It has been estimated that approximately 350,000 types of PPIs occur in a human cell [1]. This offers innumerable opportunities to develop PPI inhibitors and thus to specifically control cellular processes. However, obtaining a small-molecule inhibitor of PPIs is not trivial. Generally, the interfaces involved in PPIs are flat and large, thereby reducing the probabilities of effective competition by a smallmolecule inhibitor [2]. However, there has been some success over the last couple of decades in obtaining small-molecule PPI inhibitors [3]. It has been observed that a few key residues in the PPI interface contribute to the bulk of the binding energy in a PPI [4] and, thus, could be targeted for inhibition by a small molecule.
In order to sample the entire conformational space for PPIs, a simple inexpensive screening assay usable in a highthroughput format is extremely desirable. Although there are in vitro assays involving methods such as isothermal titration calorimetry, surface plasmon resonance, microscale thermophoresis, enzyme-linked immunosorbent assay, fluorescence polarization, fluorescence resonance energy transfer, and the bead-based AlphaScreen assay (Perkin Elmer), they are often labour intensive, expensive, and require large amounts of purified protein [5]. Moreover, in vitro assays may identify compounds that are good PPI inhibitors but may have toxic effects or low cell permeability. Therefore, it is advantageous to have an economic, high throughput in vivo assay to screen for PPI inhibitors.
The yeast two-hybrid (Y2H) assay is a powerful tool to identify binary PPIs [6] by exploiting the modular nature of the yeast Gal4 transcription factor. In this assay, the DNAbinding domain and activation domain of Gal4 are fused to two proteins of interest. If the two proteins of interest physically interact, an active Gal4 transcription factor is generated, thereby driving expression of reporter genes under the control of the GAL promoter (Fig. 1a, left panel). Apart from confirming an interaction between two proteins, this assay has been pivotal in discovering novel binding proteins. The Y2H assay has been used in developing binary protein interactome maps in model organisms such as yeast [7] and humans [8].
The Y2H assay can also be used to identify domains and amino acid residues required for PPIs. Deletion or replacement of amino acid residues critical for PPI or treatment with small-molecule PPI inhibitors will result in loss of reporter gene activity (Fig. 1a, right panel). It is possible to have a positive selection for screening of mutations or compounds that affect PPIs. For example, by placing the URA3 gene under the GAL promoter, one could screen for mutations or PPI inhibitors that rescue the lethality of yeast cells grown on medium containing 5-fluoroorotic acid; this approach is referred to as the reverse Y2H assay and was proposed 20 years ago [9,10]. However, there are very few reports of its use in screening of PPI inhibitors [11,12]. It has been acknowledged that low permeability of yeast cells to small molecules could limit the use of Y2H methods to screen for PPI inhibitors [13].
To explore the use of the Y2H assay to screen for inhibitors of PPIs, we chose the p53-Mdm2 interaction, for which there are several small-molecule inhibitors available. p53 is a master transcription factor that plays a key role in the regulation of cell cycle arrest, DNA damage response, senescence, and apoptosis [14]; it is mutated in more than 50% of cancers [15]. p53 is inhibited by Mdm2, a ubiquitin ligase that is often overexpressed in tumors [16]. By binding to the N-terminal transactivation domain of p53, Mdm2 inhibits its transcriptional activity, ubiquitinates and targets it for proteosomal degradation, and excludes it from the nucleus. Inhibition of the p53-Mdm2 interaction leads to activation of p53 and an increase in its tumor suppressive ability. The p53-Mdm2 interaction can be attributed to three key hotspot residues (Phe19, Trp23, and Leu26) in p53 that bind to a hydrophobic pocket on the surface of the Mdm2's Nterminal domain [17] (Additional file 1: Figure S1A). Small-molecule inhibitors, such as nutlin, AMG232, and MI-773, bind to the hydrophobic pocket of Mdm2 and inhibit the p53-Mdm2 interaction by mimicking the interaction of the three hydrophobic residues [18][19][20][21] (Additional file 1: Figure S1B-D). Binding of Mdm2 to full-length p53 was observed to be approximately 10fold stronger than the N-terminal domain of p53 (amino acid residues 1-93) [22], indicating the presence of additional domains in p53 that interact with Mdm2. Two such domains have thus far been reported; the DNAbinding domain of p53 (residues 234-286 within the conserved Boxes IV and V) has been shown to interact with the central acidic domain of Mdm2 [23] and the carboxyl-terminal regulatory domain of p53 (residues 367-393) interacts with the N-terminal domain of Mdm2 (residues 10-139) [24].
In this paper, we report on an improved version of the Y2H assay that can be used to screen for inhibitors of PPI in a high throughput manner. By deleting nine genes involved in ABC transporter gene function and expression, we constructed a yeast strain that is highly permeable to small molecules. We show that nutlin, AMG232, and MI-773 can inhibit the p53-Mdm2 interaction in the Y2H assay. We demonstrate that nutlin is a substrate for the yeast ABC transporter Pdr5. In addition, we suggest that the Box II region in the core DNA-binding domain of p53 interacts with the N-terminal domain of Mdm2 and makes the p53-Mdm2 interaction recalcitrant to small-molecule inhibition in the Y2H assay.

p53 interacts with Mdm2 in the Y2H assay
The p53-Mdm2 interaction is of immense biomedical importance and is a focus of several screens as a target for small-molecule inhibitors [18]. As there are many smallmolecule inhibitors available, it is an ideal system to optimize the Y2H assay [18]. We first established a Y2H assay to detect the p53-Mdm2 interaction. We fused the full-length human p53 (Hp53) to the Gal4 activation domain (Gal4 AD) and the full-length human Mdm2 (hMdm2) to the Gal4 DNA-binding domain (Gal4 BD), and introduced plasmids encoding them into the Y2H reporter strain AH109. AH109 has HIS3 and ADE2 genes under the GAL4 promoter and an interaction between b c a Fig. 1 p53 interacts with Mdm2 in the yeast two-hybrid (Y2H) assay. a Schematic showing the use of the Y2H assay in identifying interacting proteins (left panel) and inhibitors of protein-protein interactions (right panel). b Log-phase cultures of AH109 yeast cells containing plasmids encoding either Gal4 AD-p53/Gal4 BD-Mdm2, Gal4 AD-p53/Gal4 BD, Gal4 AD/Gal4 BD-Mdm2, or Gal4 AD/Gal4 BD were washed in water and plated at different dilutions on non-selective (-Leu-Trp) and selective (-Leu-Trp-Ade-His) plates and incubated at 30°C for 3 days. c Overnight cultures of AH109 yeast cells containing plasmids encoding either Gal4 AD-p53/Gal4 BD-Mdm2, Gal4 AD-p53/Gal4 BD, Gal4 AD/Gal4 BD-Mdm2, Gal4 AD/Gal4 BD, Gal4 AD-p53-F19A/Gal4 BD-Mdm2, or Gal4 AD-p53(Δ42)/Gal4 BD-Mdm2 in non-selective medium were washed in water and inoculated into selective and non-selective medium at OD 600 = 0.2 in duplicates. For each strain, growth as measured by average OD 600 of duplicate cultures is plotted against time. Ends of the vertical bar indicate the OD 600 values of the duplicate cultures Gal4 AD-and Gal4 BD-fusion proteins results in growth of yeast cells in Synthetic Dropout (SD) medium lacking histidine and adenine (Fig. 1a, left panel). As expected, only cells containing both plasmids encoding Gal4 AD-p53 and Gal4 DBD-Mdm2, but not the control cells, were able to grow in SD medium lacking histidine and adenine (Fig. 1b). Either deletion of the transactivation domain  or replacement of the hydrophobic triad residue phenylalanine F19 (which is involved in binding to the hydrophobic pocket of Mdm2) with alanine completely abolished growth of Gal4 AD-p53/Gal4 BD-Mdm2expressing cells in selective medium, indicating the specificity of the interaction (Fig. 1c).
Nutlin does not inhibit the p53-Mdm2 interaction in wildtype and pdr1Δ pdr3Δ pdr5Δ yeast cells in the Y2H assay Nutlin is a cis-imidazoline analogue that binds to the p53-binding site in Mdm2 (Additional file 1: Figure S1) and inhibits the p53-Mdm2 interaction [19]. Thus, we tested whether nutlin inhibits the p53-Mdm2 interaction in the Y2H assay. Surprisingly, nutlin had no effect on the growth of p53-Mdm2-expressing cells in selective medium (Additional file 2: Figure S2A). We considered the possibility that nutlin was being expelled from yeast cells by ABC transporter-mediated efflux. We deleted three genes, PDR1, PDR3, and PDR5, which have been reported to restrict permeability of yeast cells to small molecules [25], to construct the pdr1Δ pdr3Δ pdr5Δ strain (referred to as ABC3Δ for the remainder of this paper). Pdr1 and Pdr3 are transcription factors for the pleiotropic drug response and Pdr5 is an ABC transporter. In the ABC3Δ strain, nutlin inhibited p53-Mdm2 cell growth at 200 μM. However, nutlin also inhibited the growth of control cells expressing Csm1/Dsn1, a pair of interacting yeast kinetochore proteins [23,26], indicating that the growth inhibition of yeast cells by nutlin was not specific (Additional file 2: Figure S2B).
Nutlin inhibits the interaction of p53 (1-52) with Mdm2 in pdr1Δ pdr3Δ pdr5Δ cells We then considered the nature of the p53-Mdm2 interaction. While the transactivation domain of p53 interacts with the N-terminal domain of Mdm2, the core domain and C-terminal regulatory domain of p53 also respectively interact with the acidic-and Nterminal domains of Mdm2 [23,24]. To eliminate the possible effect of the core domain and C-terminal tail, we expressed p53 (1-52) fused to Gal4 AD and examined the effect of nutlin on its interaction with Gal4 DBD-Mdm2 in ABC3Δ cells. The interaction of p53 (1-52) with Mdm2 was completely inhibited at 125 μM nutlin (Fig. 2a). In contrast, the interaction of fulllength p53 with Mdm2 was only partially inhibited (29%), even at 150 μM nutlin. Cells expressing Csm1/ Dsn1 in selective medium were unaffected by the addition of 150 μM nutlin, indicating that the inhibition of PPIs by nutlin was specific (Fig. 2a). Moreover, nutlin did not affect the growth of cells expressing p53 (1-52) and Mdm2 in non-selective medium (Additional file 2: Figure S2C). Nutlin did not inhibit the growth of wild-type cells expressing p53 (1-52) and Mdm2, confirming that ABC transporters drive nutlin out of yeast cells (Fig. 2a).

Nutlin is a substrate for the ABC transporter Pdr5
Inhibition of the p53 (1-52)/Mdm2 interaction by nutlin in ABC3Δ cells but not in wild-type cells (Fig. 2a) suggested that Pdr1, Pdr3, and Pdr5 proteins, either singly or in combination, mediate nutlin efflux from yeast cells. To test which of Pdr1, Pdr3, and Pdr5 proteins contributes to nutlin efflux, we tested the effect of nutlin on the p53 (1-52)/ Mdm2 interaction in single mutants. Interestingly, nutlin inhibited the p53 (1-52)/Mdm2 interaction only in pdr5Δ cells but not in pdr1Δ and pdr3Δ cells (Additional file 4: Figure S4), indicating that Pdr5 causes efflux of nutlin from yeast cells.
The Saccharomyces cerevisiae proteome is predicted to have 22 ABC transporters [28]. Although only a few of these transporters have been extensively studied, they are known to exhibit broad specificity [28]. To evaluate the specificity of Pdr5 in nutlin efflux, we deleted six genes encoding ABC transporters, namely PDR15, PDR11, PDR10, YCF1, SNQ2, and YOR1, and tested the effect of nutlin on the p53-Mdm2 interaction in the mutants. Strikingly, the p53-Mdm2 interaction was sensitive to nutlin only in pdr5Δ cells but not in the other eight mutants (Fig. 3a, Additional file 5: Figure S5A and Additional file 6: Figure  S6). For an unrelated drug, rapamycin, pdr11Δ and yor1Δ were most sensitive (Additional file 5: Figure S5B), confirming that ABC transporters have distinct substrate specificities. Nutlin sensitivity of pdr5Δ cells expressing p53 (1-52)/Mdm2 was rescued by introduction of a 2-μ plasmid encoding Pdr5 under the ADH1 promoter (Fig. 3b). These results indicate that nutlin is a substrate of Pdr5.
To use the Y2H assay for screening of PPI inhibitors, it is desirable to have a yeast strain that is permeable to a diverse collection of compounds. As ABC transporters have differing substrate specificities, we constructed a strain that lacks several ABC transporter genes. A previous study showed that a yeast strain lacking seven ABC transporter genes (PDR5, PDR15, PDR11, PDR10, YCF1, SNQ2, and YOR1) and one gene encoding a transcription factor for ABC transporters (PDR3) was viable [29]. We generated the ABC9Δ strain, a derivative of the Y2H strain AH109, which contains the abovementioned eight gene deletions together with deletion of PDR1. The ABC9Δ strain a b Fig. 2 Nutlin inhibits the interaction of p53 (1-52) with Mdm2 in ABC3Δ cells. a Overnight cultures of wild-type or ABC3Δ cells containing plasmids encoding either Gal4 AD-Csm1/Gal4 BD-Dsn1 (used as a negative control), Gal4 AD-p53/Gal4 BD-Mdm2, or Gal4 AD-p53 (1-52)/Gal4 BD-Mdm2 in non-selective medium were washed in water and inoculated at OD 600 = 0.2 into selective medium containing DMSO or nutlin at the indicated concentrations in duplicate. For each strain, growth as measured by average OD 600 of duplicate cultures is plotted against time. Ends of the vertical bar indicate the OD 600 values of the duplicate cultures. b Overnight cultures of ABC3Δ cells containing plasmids encoding either Gal4 AD-Csm1/Gal4 BD-Dsn1, Gal4 AD-p53 (1-52)/Gal4 BD-Mdm2, Gal4 AD-p53 (1-52)/Gal4 BD-Mdm2 (1-125), or Gal4 AD-p53 (1-52)/Gal4 BD-Mdm2 (1-125)-M62A in non-selective medium were washed in water and inoculated at OD 600 = 0.2 into selective medium containing DMSO or nutlin at the indicated concentrations. Growth of the cultures was monitored by recording the absorbance at 600 nm at the indicated time points. Data from a repeat experiment are depicted in Additional file 3: Figure S3 was hypersensitive to doxorubicin, myriocin, cycloheximide, and nutlin in comparison to the wild-type strain, indicating that it is permeable to several compounds (Fig. 4a). This is consistent with earlier studies showing that ABC transporter mutants in yeast are sensitive to doxorubicin, myriocin, and cycloheximide [30,31]. In the ABC9Δ strain, the p53 (1-52)/Mdm2 interaction was sensitive to nutlin at 10 μM in comparison to 100 μM in the ABC3Δ strain (Figs. 4b and Additional file 7: Figure S7). This value (10 μM) fits closely with the potency of nutlin a b Fig. 3 Nutlin is a substrate for the ABC transporter Pdr5. a Overnight cultures of either wild-type strain or the indicated deletion strains (pdr5Δ, pdr15Δ, pdr1Δ, etc.) containing plasmids encoding Gal4 AD-p53 (1-52)/Gal4 BD-Mdm2 were washed in water and inoculated at OD 600 = 0.2 into selective medium containing DMSO or nutlin at the indicated concentrations. Growth of the cultures was monitored by recording the absorbance at 600 nm after 0, 24, 44, and 68 h following inoculation. Growth at 68 h is indicated in the plot. Additional file 5: Figure S5A contains the complete dataset and Additional file 6: Figure S6 contains data from a repeat experiment. b Overnight cultures of pdr5Δ cells containing plasmids encoding Gal4 AD-p53 (1-52)/Gal4 BD-Mdm2 and Pdr5 or Gal4 AD-p53 (1-52)/Gal4 BD-Mdm2 alone were washed in water and inoculated at OD 600 = 0.2 into selective medium containing DMSO or nutlin at the indicated concentrations in duplicate. For both strains, growth as measured by average OD 600 of duplicate cultures is plotted against time. Ends of the vertical bar indicate the OD 600 values of the duplicate cultures in mammalian cells, indicating that the ABC9Δ strain is an excellent system for small-molecule PPI inhibitor screens [32,33].
AMG232 and MI-773 inhibit the p53 (1-52)/Mdm2 interaction in the Y2H assay We then tested whether the Y2H assay for the p53 (1-52)/Mdm2 interaction optimized for nutlin was extendable to other p53-Mdm2 interaction inhibitors, namely AMG232 and MI-773. Like nutlin, AMG232 and MI-773 bind to the p53-binding site in Mdm2 and block the p53-Mdm2 interaction (Additional file 1: Figure S1). Both AMG232 and MI-773 abolished the p53 (1-52)/ Mdm2 interaction in wild-type, ABC3Δ, and ABC9Δ strains (Fig. 5a, b). However, the p53 (1-52)/Mdm2 interaction was approximately 25-fold more sensitive to MI-773 and AMG232 in the ABC9Δ strain in comparison to the wild-type strain (Fig. 5a, b).  The full-length p53/Mdm2 interaction is more resistant to small-molecule inhibitors in comparison to the p53 (1-52)/ Mdm2 interaction As the p53 (1-52)/Mdm2 interaction in the ABC9Δ strain was more sensitive to small-molecule inhibitors in comparison to wild-type and ABC3Δ strains (Fig. 5), we tested whether the full-length p53/Mdm2 interaction was sensitive to small-molecule inhibitors in the ABC9Δ strain. Nutlin, MI-773, and AMG232 inhibited the interaction of full-length p53 with Mdm2 by approximately 70-75% (Fig. 6). Inhibition of the p53-FL/Mdm2 interaction by nutlin in the ABC9Δ strain was approximately 2-to 3-fold better than that achieved in the ABC3Δ strain (Fig. 2a). In contrast, the p53 (1-52)/Mdm2 interaction was inhibited by approximately 90-95% by the three inhibitors (Fig. 6). Taken  Highly conserved Box II in the DNA-binding domain of p53 strengthens the p53-Mdm2 interaction To rationalize our observation that nutlin, AMG232, and MI-773 inhibited the p53 (1-52)/Mdm2 interaction much more efficiently than the full-length p53/Mdm2 interaction in the Y2H assay we hypothesized that there are additional domains in p53 besides the transactivation domain that interact with Mdm2. To map the region of p53 that contributes to nutlin resistance of the p53-Mdm2 interaction, we made a series of deletions of fulllength p53 and assayed the nutlin sensitivity of their interaction with Mdm2. The results are summarized in Fig. 7a. Deletion of the C-terminal tail or the DNAbinding domain Box V of p53, which have been implicated in the interaction with Mdm2, had no effect on nutlin sensitivity. Interestingly, p53 (1-160), which has the first 60 residues of the DNA-binding domain, was resistant to nutlin and p53 (1-143) was partially resistant to nutlin. The interaction of p53 (1-160) with Mdm2 was abolished by the F19A mutation and by deletion of the transactivation domain (Additional file 8: Figure S8). These results indicate that, although the primary interaction of p53 (1-160) with Mdm2 is via the transactivation domain, the DNA-binding domain interacts with Mdm2 and stabilizes the interaction.
To map the domain of Mdm2 that is required for increased nutlin resistance of its interaction with fulllength p53, we expressed the minimal Mdm2 (25-109) required for interaction with p53. The interaction of Mdm2 (25-109) with p53 (1-160) and p53 (full length) was also resistant to nutlin. To test whether the decreased sensitivity of p53-Mdm2 interaction was due to increased p53 levels, we compared the protein amounts of strains expressing different p53 variants by western blotting (Additional file 9: Figure S9). Full-length p53 was expressed to a lower extent in comparison to the other p53 variants, ruling out this possibility.
To directly compare the strength of interaction between Mdm2 and different variants of p53, we used an in vitro binding assay. We prepared HA-tagged Mdm2, full-length p53, p53 (1-52), and p53 (1-160) by in vitro translation. We then mixed HA-Mdm2 with the p53 variants and assayed the interaction by coimmunoprecipitation. Full-length p53 bound specifically to Mdm2 (Fig. 7b). Interestingly, p53 (1-160) bound to Mdm2 approximately 10-fold more efficiently than p53 (1-52). These results are consistent with our Y2H results, which show that the interaction of p53 (1-52) with Mdm2 is more sensitive to small-molecule inhibitors compared to the p53 (1-160)/full-length p53 and Mdm2 interactions. We also tested the ability of p53 (1-116) and p53 (1-143) to bind to Mdm2 in the in vitro binding assay. Only p53 (1-143), but not p53 (1-116), bound to Mdm2 (Fig. 7b). Comparable amounts of Mdm2-HA were immunoprecipitated in all the binding reactions (Fig. 7c). These results are in excellent agreement with our Y2H assay results.
To identify residues involved in the additional p53-Mdm2 interaction, we examined the conserved residues around the region (116-143) of p53 that is minimally required for nutlin resistance in the Y2H assay and stable interaction with Mdm2 in the in vitro binding assay (Fig. 8a). We split the conserved regions into three segments (Seg1-3). Interestingly, these segments lie very close to DNA in the p53-DNA co-crystal structure [34] (Additional file 10: Figure S10). We replaced the conserved residues in each segment with alanine. Mutations in segments 1 and 2, but not 3, rendered the p53 (1-143)/Mdm2 interaction nutlin sensitive (Fig. 8b). We further replaced the subset of residues in Segments 1 and 2 with alanine and determined their effect on the nutlin sensitivity of the p53 (1-143)/Mdm2 interaction. F113A (Seg1AA), G117A T118A (Seg1BA), and K120A S121A V122A (Seg1CA) mutations in Segment 1 conferred nutlin sensitivity (Additional file 11: Figure S11A). However, mutations T125A Y126A S127A (Seg2BA) and L130A N131A K132A (Seg2CA) in Segment 2 had no effect on nutlin resistance (Additional file 11: Figure  S11A). The T123A C124A T125A Y126A S127A (Seg2AA) mutation in Segment 2 abolished growth in selective medium (Additional file 11: Figure S11A). We confirmed that the wild-type and mutant p53  proteins were expressed at comparable levels by western blotting (Additional file 11: Figure S11B). As multiple mutations in Segment 1 conferred nutlin sensitivity, we focused on the Seg1 mutant for further analysis.
We tested whether the Seg1A mutant was defective in interaction with Mdm2 in the in vitro binding assay. Crucially, only wild-type p53 (1-143), but not its Seg1A mutant variant, interacted with Mdm2 in vitro (Fig. 8c). However, the Seg1A mutation did not have an effect on the full-length p53-Mdm2 interaction (Fig. 8c). Comparable amounts of Mdm2-HA were immunoprecipitated in all the binding reactions (Additional file 12: Figure S12). In agreement with this result, the Seg1A mutation also did not have any effect on the p53-FL/Mdm2 interaction in the Y2H assay (Additional file 13: Figure S13).
We suggest that additional domains in p53 that interact with Mdm2 mask the effect of Seg1A mutation on the p53-Mdm2 interaction. This is consistent with our observation that full-length p53 binds more strongly to Mdm2 in comparison to p53 (1-143) (Figs. 7b and 8c). Although the physiological significance of the Seg1/ Mdm2 interaction remains to be clarified, there was an excellent correspondence between sensitivity of the p53-Mdm2 interaction to small-molecule inhibitors in the Y2H assay and their stable association in the in vitro binding assay.
A two-phased strategy for screening of PPI inhibitors using the Y2H assay The enhanced permeability of the ABC9Δ strain coupled with economic aspects of high-throughput, growth- based assays makes the Y2H assay suitable for screening of PPI inhibitors. In Fig. 9, we depict a methodological flowchart using a two-phased Y2H assay to screen for inhibitors of PPI. In the first phase, a Y2H assay for PPIs of interest is established and validated by mutagenesis. Based on our experience with the p53-Mdm2 interaction, it would be prudent to identify domains minimally required for PPIs by truncations as weak secondary PPIs affect the sensitivity to small-molecule inhibition. Experiments for the first phase can be performed in the standard wild-type yeast strain for two-hybrid analysis.
In the second phase, the plasmids encoding the minimal set of specifically interacting protein domains are introduced into the ABC9Δ strain, which is permeable to small molecules. Transformants pre-grown in nonselective medium are split into two and transferred into selective and non-selective media containing smallmolecule compounds in a microtiter plate. Compounds toxic to yeast cells will inhibit growth in both selective and non-selective media. However, compounds that Fig. 9 A two-phased strategy to screen for small-molecule inhibitors of PPI using the Y2H assay. Please see the text for details inhibit the growth of yeast cells only in selective media are putative PPI inhibitors and can be taken up for further investigation.

Discussion
By using the p53-Mdm2 interaction, which is amenable to small-molecule inhibition, as a model, we have developed and validated an improved version of the Y2H assay that can be used in a high-throughput manner.
To accomplish this, we made two crucial changes. Firstly, we used the minimal Mdm2-interacting domain in p53 instead of full-length p53. Secondly, we increased the permeability of yeast cells to small molecules by disabling ABC transporters. By analyzing the p53 domains required for conferring resistance to inhibition by small molecules, we identified a hitherto unreported Mdm2-interacting segment overlapping with the highly conserved Box II region in the core domain of p53. Mdm2 inhibits the DNA binding activity of p53, but precisely how this occurs remains unknown. Our observations offer an explanation for this. As the DNA domain of p53 is engaged in interactions with Mdm2, p53-Mdm2 complexes may have a restricted ability to bind DNA. It will be interesting to test the effect of mutating segment 1 in the core domain of p53 on the Mdm2 inhibition of the DNA-binding activity of p53 and on the function and stability of p53 in mammalian cells.
Why do small-molecule inhibitors inhibit the full-length p53-Mdm2 interaction in mammalian cells but not in the Y2H system? Since a good correlation between the strength of the p53-Mdm2 interaction in vitro and an ease of inhibition by small-molecule inhibitors in the Y2H assay was observed, we suggest that there are factors in mammalian cells that compete with Mdm2 in binding to p53. Displacement of the transcriptional activation domain of p53 from Mdm2 by small-molecule inhibitors might create a binding site for proteins like p300, which are not present in yeast cells. Alternatively, posttranslational modifications in either p53 or Mdm2 in mammalian cells may weaken the p53-Mdm2 interaction. Our observations might explain why the p53-Mdm2 interaction is only partially inhibited by nutlin in vitro [35].
Complete inhibition of the p53 (1-52) interaction with Mdm2 by AMG232 and MI-773 at nanomolar concentrations suggests that the assay is highly sensitive. AMG232 was by far the most potent of the three inhibitors tested. This is consistent with its dissociation constant (0.045 nM) being lower than that of MI-773 (62 nM) and nutlin (90 nM). In contrast to nutlin, AMG232 and MI773 were able to inhibit the p53 (1-52)/Mdm2 interaction in wildtype cells. Nutlin thus appears to be a better substrate for yeast ABC transporters than MI-773 and AMG232. As many naturally occurring compounds, such as alkaloids, have an imidazole ring like nutlin, it is tempting to speculate that yeast ABC transporters have evolved to recognize such compounds. As human cells contain several ABC transporters, it will be interesting to test whether cotreatment of cancer cells with ABC transporter inhibitors improves the anti-cancer activity of nutlin.
We observed an excellent correlation between resistance of the p53-Mdm2 interaction to small-molecule inhibitors and the strength of the PPI in the in vitro binding assay. Binding of p53 (1-52/1-116) to Mdm2 is sensitive to nutlin in the Y2H assay and is too weak to be detected in the binding assay. In contrast, the binding of p53 (1-143/1-160) is resistant to nutlin in the Y2H assay and can be robustly detected in the binding assay. Binding of full-length p53 to Mdm2 was the strongest in the in vitro binding assay and resistant to nutlin in the Y2H assay. It is worth highlighting that the expression of p53 was approximately 10-fold lower compared to other p53 variants in yeast cells and its resistance to small-molecule inhibitors in the Y2H assay could be underestimated. Mutations in the Box II region of p53 (1-143) that render sensitivity to nutlin in the Y2H assay also abolish its association with Mdm2 in the in vitro binding assay. We therefore propose that the Y2H assay is not only useful in screening for PPI inhibitors but could also help to delineate domains involved in stabilizing PPIs.

Conclusion
In summary, by disabling several ABC transporters in yeast, we developed a significantly improved Y2H assay to screen for small-molecule inhibitors of PPIs. Using inhibitors of the p53-Mdm2 interaction to validate the Y2H assay and demonstrate its suitability for highthroughput screening, we have also identified a novel putative Mdm2-binding site in the highly conserved Box II region of p53. The ease and economic nature of the assay, coupled with advantages of an in vivo-based system, makes it a valuable addition to the existing stockpile of in vitro tools to screen for PPI inhibitors.

Yeast strain construction
Derivatives of the yeast strain AH109 in the MATCH-MAKER GAL4 Two-Hybrid system (Clontech Laboratories, CA, USA) were used in this study. ABC transporter-related genes were deleted by standard methods using PCR product-mediated homologous recombination. The ABC9Δ strain was generated by crossing single deletion strains with each other. A list of yeast strains used in the study is available in Additional file 15: Table S2.
Drug sensitivity test using the Y2H assay (±)-Nutlin-3 (Cayman Chemicals, Ann Arbor, MI, USA) was dissolved in DMSO to make 50 mM stock solution. MI-773 (MedChem Express Co. Ltd., China) and AMG232 (MedChem Express) were dissolved in DMSO to make 10 mM stock solutions. Overnight yeast cultures were washed and then diluted to OD 600 = 0.2 in 1 mL SD-Leu-Trp-His-Ade medium containing the desired concentration of the drug. The cultures were then placed in a 30°C incubator with shaking for 3 days. Growth of the cultures was monitored by measuring the absorbance at 600 nm using the Implen Nanophotometer. This assay was also performed in a 96-well plate format with 200 μL SD-Leu-Trp-His-Ade medium containing the desired concentration of the drug in duplicate. OD 600 at different time points was measured using a Gen 5 TM (BIO-TEK Instrument, Vermont, USA) microplate reader. The average of the two OD 600 readings was calculated and used in the analysis.