Competition NMR for Detection of Hit/Lead Inhibitors of Protein–Protein Interactions

Screening for small-molecule fragments that can lead to potent inhibitors of protein–protein interactions (PPIs) is often a laborious step as the fragments cannot dissociate the targeted PPI due to their low μM–mM affinities. Here, we describe an NMR competition assay called w-AIDA-NMR (weak-antagonist induced dissociation assay-NMR), which is sensitive to weak μM–mM ligand–protein interactions and which can be used in initial fragment screening campaigns. By introducing point mutations in the complex’s protein that is not targeted by the inhibitor, we lower the effective affinity of the complex, allowing for short fragments to dissociate the complex. We illustrate the method with the compounds that block the Mdm2/X-p53 and PD-1/PD-L1 oncogenic interactions. Targeting the PD-/PD-L1 PPI has profoundly advanced the treatment of different types of cancers.

The tumor suppressor p53 protein is a key player in protecting the organism from cancer and was therefore coined a term: "the guardian of the genome. To escape the control system mediated by p53, majority of human cancers has either mutation within p53 (50% all cancers), whereas the rest compromises the effectiveness of the p53 pathway 19,20 . In tumors with unmodified wildtype p53, the p53 pathway is inactivated by the Mdm2 and MdmX proteins 21,22 .
Therefore, the disruption of the Mdm2-p53/MdmX-p53 interactions that leads to the restoration of the impaired function of p53, poses a new approach to anticancer therapies across a broad spectrum of cancers.
Another important protein target in cancer is the PD-1/PD-L1 system. PD-1 (Programmed cell death protein 1) is expressed on activated T cells and plays a critical role in modulation of the host's immune response 23,24 . The principal PD-1 ligand, PD-L1, is expressed on macrophages, monocytes and cancer cells. Cancer cells exploit this ligand protein to avoid immune attack by T cells 25 . This seminal finding of how cancer cells use binding between PD-L1 and PD-1 to inhibit the killing of tumor cells by T cell have now been translated into effective medical treatment 17,26-4 30 . Blocking the immune checkpoint PD-L1 or PD-1 allows for the T cell killing of tumor cells, and immune checkpoint inhibitors targeting the PD-1/PD-L1 interaction have revolutionized modern cancer therapy for advanced cancer 18,[31][32][33] .
The interaction of PD-L1/PD-1 and p53-Mdm2/X present challenging cases in detecting weak binders, because the KD of this interaction is 8.2 and 0.2-0.6 µM, respectively 34,35 . We show herein that by introducing designed mutations in the component protein(s) of these proteinprotein complexes, we could weaken the affinity of the PPI interaction. The mutations are introduced within the binding partner that is not targeted with inhibitors, therefore not compromising the binding interface of targeted protein and tested ligand. For example, in the case of p53/Mdm2 complex, we have mutated p53 to determine fragments binding to Mdm2 that dissociates the complex. Since there are over 30 000 known naturally occurring missense mutations to p53 36,37 such a system is physiologically still relevant. Mutations within the nontargeted binding partner allow for the proteins that build up the PPI complex to be sensitive to weak binding compounds i.e. the modified assay can be used in the fragment-based screening.
We named this variant of the AIDA experiment "w-AIDA-NMR", where "w" stands for "weak".

RESULTS
p53 mutants. The complexes of wt-p53 with Mdm2 and MdmX have KD values 0.60 µM and 0.24 µM, respectively 35 . These KD's determine the weakest inhibitor, which can be tested with the AIDA-NMR methods. For the compounds that weakly bind to Mdm2/X, which still could be good initial scaffolds for further optimization in drug development process, the method would not be sensitive enough for detecting inhibition of the p53-Mdm2 and p53-MdmX PPI.
Three key residues Phe19, Trp23 and Leu26 of p53 make the highest contribution to the binding energy of p53 with Mdm2/X [38][39][40] . Among them, Trp23 is the most important, and any mutations 5 of this residue completely abrogate the binding between p53 and Mdm2/X 39 . Mutations of Phe19 have similar effects, but not as strong as those of Trp23. Both residues are essential for the p53/Mdm2 complex formation. They are buried within pockets of Mdm2/X with strong π-π stabilizing interactions (Fig. 1a,b; Supplementary Fig. 1) 39 . Since the aim of the research was not to block the interaction, but only to slightly weaken it, Phe19 and Trp23 were not touched, and Leu22 and Leu26 were chosen as most plausible targets of the mutations (Fig. 1a The most significant lowering of the Mdm2/X-p53 interaction was observed for the L22A mutation, where the KD value increased six fold compared to that of the p53-wt/Mdm2 complex. Surprisingly, modifications of Leu26, which is one of the key amino acids of the p53 binding to Mdm2/X, were less efficient than mutations of Leu22. Moreover the L26I mutant generated lower KD value (0.31 µM) than in the Mdm2/p53-wt interaction (0.60 µM). Although the interaction with the L26V mutant was weakened, the affinities of the proteins were still too strong to use the preformed complexes for the investigation of the weak binding compounds. Therefore, we designed the double mutants (L22IL26V and L22VL26V) which combine above- dimensional proton NMR spectra of all mutants were almost identical to that of the wt-p53 spectra. This indicated that the mutated p53 constructs were correctly folded (data not shown).

Inhibitors of the wt-p53/Mdm2 and wt-p53/MdmX interactions.
A large number of lowmolecular-weight compounds that bind to Mdm2 and MdmX [42][43][44] have been tested in clinical trials for several years 45,46 . Among the most advanced are cis-imidazoline derivatives called Application of mutants L22A and L22IL26V increased the sensitivity of 1D AIDA-NMR assay.
The Mdm2 inhibitors 1 and 2 caused almost full recovery of the Trp23 signal with the L22A mutant ( Fig. 2). For MdmX, all tested MdmX inhibitors released the p53 from its complex with MdmX ( Fig. 3). Better results were obtained for the L22A mutant than for the L22II26V mutant.

PD-1 mutants.
The second PPI that we tested with the w-AIDA-NMR is the interaction of human PD-L1 with PD-1. Here we use the entire ectodomain of PD-1 and the PD-1 binding domain of the ectodomain of PD-L1 (residues 18-134) ( Supplementary Fig. 11). Although the PD-1/PD-L1 complex is characterized by a relatively high affinity constant (Supplementary Fig.   12), it is, nevertheless, too strong to be used in the NMR screening of "weak" fragments, which usually have two or three orders of magnitude lower affinities. Therefore, we designed a series of the PD-1 mutants based on the structure of the PD-1/PD-L1 complex 52 . We identified the amino acid candidates on PD-1 for the mutations using the same approach as that described for the p53/Mdm2/MdmX systems. The mutations should weaken the binding of PD-1 to PD-L1 in a minimal invasive manner (Fig. 1c, Supplementary Figs. 1 and 13). The mutated PD-1 N66A, 8 Y68A, E135A and the double mutant N66AY68A were then expressed in E. coli (Supplementary Table 1) and checked by NMR whether the mutations affected the folding of the protein. From these tested mutants only the N66A mutant was folded (the data for other mutants not shown). Protein melting analysis showed that the N66A PD-1 mutant has similar midpoint of thermal transition (melting temperature), indicating that the changes in its stability against denaturation were insignificant; in fact, the N66A mutant was slightly more temperature stable than wt-PD-1 ( Supplementary Fig. 14). The affinities of the PD-1/PD-L1 and (N66A)PD-1/PD-L1 complexes was determined using the Microscale Thermophoresis by titrating the labeled PD-1 with the unlabeled PD-L1. The resulting data was next fitted with KD models to yield affinities. For the (N66A)PD-1/PD-L1 complex, we were not able to reach the top plateau as we could not concentrate PD-L1 solution above 183 µM ( Supplementary Fig. 12). Therefore, the KD is estimated to be above 100 µM, considerably higher than for the native complex, allowing weaker small-molecular fragments to dissociate it. The N66A mutation did not affect the overall structure of the interface of the PD-1/PD-L1 complex as depicted in the Supplementary Fig. 13.
Asn66 is only involved in one hydrogen bond with Ala121 from PD-L1 and when mutated to alanine does not significantly change the space in the complex ( Supplementary Fig. 13 inset).
Moreover, when overlaid on top of each other the critical for interactions residues from hPD-L1 are virtually in the same position regardless to the mutations of PD-1 including murine PD-1human PD-L1 complex ( Supplementary Fig. 13).

Inhibitors of the immunocheckpoint PD-1/PD-L1 interaction.
A sizeable number of smallmolecule binders to PD-L1 that inhibit this PPI has now been described (ca. 1000) 53,54 . We have recently published a series of studies on the affinities of the small-molecule inhibitors of the PD-1/PD-L1 interaction developed by Bristol-Myers Squibb (called herein the BMS compounds 55,56 ). The BMS compounds are based on the hydrophobic biphenyl core scaffold 53,57 .
Our NMR studies indicate that the binding of the BMS compounds to PD-L1 induces oligomerization of the protein [57][58][59] . The line width broadening in the NMR signals of PD-L1 after addition of BMS compounds, both in proton and 1 H-15 N 2D HMQCs is so extensive that it is impossible to estimate dissociation constants also for "weaker" binding precursors of the BMS compounds (KD's double digit μM). One of the NMR techniques that would allow validating the binding and estimating the value of KD is the w-AIDA-NMR assay.
For the start in the PD-1/PD-L1 system, we used two smallest fragments of the BMS-1166 inhibitor (fragments 5 and 6 in Supplementary Fig. 15) 59 . BMS-1166 is one of the most potent small-molecule ICB inhibitors for the PD-1/PD-L1 developed until now 53 . It binds to PD-L1 and efficiently dissociates the human PD-1/PD-L1 complex in vitro 59 . In the ICB cell models, it activates the effector T cells, which are attenuated by both soluble and membrane-bound PD-L1 presented by antigen-presenting cells 59 .

AIDA-NMR on the wt-PD-1/wt-PD-L1 complex.
For the start in the PD-1/PD-L1 system, we used two smallest fragments of the BMS-1166 inhibitor (fragments 5 and 6 in Supplementary   Fig. 15) that in our previous tests showed the interaction with the PD-L1 protein in the 1 H-15 N HMQC NMR 59 . To perform the 2D AIDA experiment for 5 and 6, as the so-called reporter protein, which should be 15 N isotopically labeled, we use the 15 N-labeled PD-1 (13.2 kDa) ( Supplementary Fig. 16a). After addition of PD-L1 (14.9 kDa) in the molar ratio 1:1, most of the cross-peaks in the 1 H-15 N HMQC spectrum of PD-1 became broader, their intensities decreased and most of the cross peaks disappeared (Supplementary Fig. 16b). This result confirms the forming of the complex with the molecular weight ca. 30 kDa. The AIDA-NMR assay was then applied to test the dissociating capabilities of 5 and 6. The compounds that despite being "active" in the binary interaction with PD-L1, were "inactive" in the AIDA test. No recovery signals from of the 15 N labeled PD-1 protein was observed in 2D and 1D spectra ( Supplementary   Figs. 16c-d and 17). This shows that the tested compounds are not able to dissociate the PD-1/PD-L1 complex; their dissociation constants with PD-L1 are higher than dissociation constant of the native human PD-1/PD-L1 complex (8 µM - Supplementary Fig. 12). We also performed a positive control and the full recovery of 2D HMQC spectrum of the 15 N PD-1 was observed after adding BMS-1166 with KD = 1.4 nM to sample with compound 5 (Supplementary Fig.   16e).
w-AIDA-NMR for the (N66A)PD-1/PD-L1 complex. In the same way as for the wt-PD-1 proteins, we performed an 2D AIDA-NMR experiment using the 15 N-labeled mutated protein (N66A)PD-1 (13.2 kDa) and wt-PD-L1 (Figs. 5a and 5b). In contrast to the pervious experiment, the resulting complex of the (N66A)PD-1/PD-L1 proteins (28.1 kDa) has smaller number of broadening/disappearing signals in the HMQC spectra due to the higher KD value. However, most noticeable changes of the chemical shifts can be observed in the range ca. 8.8-9.4 and 122-127 ppm for hydrogen and nitrogen, respectively (Fig. 5b). Addition of an equimolar amount of 5 or 6 results in dissociating of the PD-1/PD-L1 complex (Figs. 5c and 5d) with appreciable recovery of the 2D signals observed for 5. This suggests that the fragment 6 is less potent than 5 despite of the addition of the aromatic system. Next, for a positive control, to the sample of 6 an equimolar amount of BMS-1166 was added and a full recovery of 2D HMQC spectrum of the 15 N PD-1 was observed.
Similar results can also be obtained by analyzing the 1D 1 H NMR spectra ( Supplementary Fig.   18). As a result of the complex formation, significant signal broadening of the spectral lines can be observed. This is clearly visible for the NMR signals of PD-1 with the chemical shifts at -0.2 and -0.7 ppm. After addition of the tested compounds, those signals have partially recovered and significant sharpening of signals between 0.0 and -0.1 ppm were observed. The full recovery of the PD-1 signals was observed only after addition of BMS-1166.
We could estimate that the dissociation constant of PD-L1/fragment 5 interaction, which is in the range 50 ± 20 µM. In the case of fragment 6, for which the w-AIDA-NMR indicated less recovery of the NMR signals, we determine that the KD is around 120 ± 40 µM (the intensities 11 of the recovered resonances are used to obtain an approximate dissociation constant. Errors are quantified from signal-to-noise in NMR spectra). These results correlate with the Homogeneous Time-Resolved Fluorescence (HTRF) assay, where the IC50's for 5 and 6 were determined at 34 µM and above 100 µM, respectively (Supplementary Fig. 19).

DISCUSSION
Fragment screening is frequently the first step for the identification and development of molecules that modulate the activity of therapeutic targets. Numerous biophysical methods exist for the identification of fragment hits [60][61][62][63] . Among them, the NMR experiment is recognized as a highly robust technique for fragment screening against protein targets 6,8,[64][65][66] . Here, we have described a competition NMR experiment -w-AIDA-NMR -that is sensitive to weak μM-mM interactions and directly shows whether an antagonist releases proteins from their PPI interaction. We believe that w-AIDA-NMR is a valuable complement to the renowned binary ligand-protein SAR-by-NMR assay 6,64,65 , but also to the saturation transfer difference (STD) 12 NMR experiment. For the STD, AIDA-NMR avoids weak points of that experiment as the AIDA-NMR offers checking for compound aggregation and protein instability, two situations leading to false positives. Moreover, by introducing the mutations into non-binding partner, w-AIDA-NMR is performed under physiological conditions as target protein/ligand interface is not compromised. This was further validated using the native PD-1/PD-L1-Long complex that was Syntheses. Nutlin-3a, compounds 1, 2, 3, and 4 were purchased. Compounds 5 and 6 were synthesized according to the methods described by Guzik et al. 57 .

Mutagenesis.
Site directed mutagenesis of p53 and PD-1 were performed using PCR. The mutagenic primers for p53 (Supplementary Table 1) were designed for QuickChange Sitedirected mutagenesis kit (Stratagen). Vectors pET28a with human p53 were used as templates.
The mutagenic primers used for PD-1 (Supplementary Table 1) were designed using an inversed PCR approach. Vectors pET24d and pET21b with human PD-1, respectively, were used in the same manner.  16 from 416 nM to 183 µM. Samples were premixed and incubated for 2h at RT in dark before loading into capillaries. Data processing is described in the legend of Supplementary Fig. 11.
Homogenous Time-Resolved Fluorescence (HTRF). The HTRF assay was performed using the certified Cis-Bio assay kit at 20 µL final volume using their standard protocol as described by Musielak et al. 54 . Measurements were performed on individual dilution series to determine the half maximal inhibitory concentration (IC50) of tested compounds. After mixing all components according to the Cis-Bio protocol, the plate was incubated for 2 h at RT. TR-FRET measurement was performed on the Tecan Spark 20M. Collected data was background subtracted on the negative control, normalized on the positive control, averaged and fitted with normalized Hill's equation to determine the IC50 value using Mathematica 12.
Fluorescence Polarization Assay (FP). FP competition assay is based on the displacement of p53 mutant peptide called P2 from the complex with Mdm2/X as previously described 68 . All measurements were conducted on Tecan Infinite® 200 PRO plate reader. Buffer formulation was as follow: 50 mM NaCl, 10 mM Tris pH 8.0, 1 mM EDTA, 5% DMSO. FP was determined at λ = 485 nm excitation and λ = 535 nm emission 15 min after mixing all assay components.
All tests were performed using Corning black 96-well NBS assay plates at room temperature.