Inverse PI by NMR: Analysis of Ligand 1H-Chemical Shifts in the Protein-Bound State

The study of protein-ligand interactions via protein-based NMR generally relies on the detection of chemical-shift changes induced by ligand binding. However, the chemical shift of the ligand when bound to the protein is rarely discussed, since it is not readily detectable. In this work we use protein deuteration in combination with [1H-1H]-NOESY NMR to extract 1H chemical shift values of the ligand in the bound state. The chemical shift perturbations (CSPs) experienced by the proton ligand resonances (free vs bound) are an extremely rich source of information on protein-ligand complexes. Besides allowing for the detection of intermolecular CH-π interactions and elucidation of the protein-bound ligand conformation, the CSP information can be used to analyse (de)solvation effects in a site-specific manner. In conjunction with crystal structure information, it is possible to distinguish protons whose desolvation penalty is compensated for upon protein-binding, from those that are not. Combined with the previously reported PI by NMR technique for the protein-based detection of intermolecular CH-π interactions, this method represents another important step towards the ultimate goal of Interaction-Based Drug Discovery.


Introduction
The design of small molecules that bind target proteins with high specificity and affinity is based on the optimization of complementary interactions in the bound state versus the respective unbound solvated states. One crucial interaction in drug design is that between the π -electrons of aromatic ring-systems and aromatic or aliphatic CH containing groups. 1 These so-called CH-π interactions are the most common type of non-covalent interaction in protein-ligand interactions. 2 While they have relatively small individual interaction energies between -1.5 to -3 kcal, 3-7 due to cooperative behavior, 8 and minimal desolvation penalty, 9 the cumulative effect of several CH-π interactions in protein-ligand binding events can result in a major contribution to binding energy. 10 Often CH-π interactions are grouped under the rather broad term hydrophobic interactions.
The hydrogen donor of CH-π interactions in protein-ligand complexes can come from either the protein or ligand resulting in CH protein -π ligand and CH ligand -π protein interactions respectively. The hydrogen donors in proteins are either sp 2 -hybridized CHgroups of aromatic sidechains of Tyr, His, Phe and Trp, 11 aliphatic sp 3 -hybridized (CH 3 , CH 2 ) groups which exist in all natural amino acid sidechains or the protein backbone (α-CH). 12 For aromatic sp 2 -hybridized CH-groups, the interaction is most favorable when the donor proton is positioned directly above the centroid of the aromatic acceptor ring-system (central CH-π H-bonds) at a distance (hydrogen to ring centroid) of ~2.5 Å and has a calculated energy minimum of -2.46 kcal/mol as shown in theoretical works for the benzene dimer. [3][4][5] For aliphatic, sp 3 -hybridized hydrocarbons like the benzenemethane pair the interaction energy is smaller (-1,45 kcal/mol) but the donor group ideal positioning remains directly above the aromatic ring-system center. 6,7 Unlike classical H-bonds, CH-π interactions are neither highly directional, nor strongly distance dependent, 13 and hence still favorably contribute to binding when oriented above the periphery of a ring system rather than its center (peripheral CH-π H-bonds). 10 We previously presented the PI by NMR methodology which can quantify the strength of CH-π interactions in protein-ligand complexes. 11 The approach relies on the pronounced shielding effect exerted by aromatic ring-systems on interacting protons.
Using a combination of amino acid labeling strategies and sensitive 1 H- 13

Results and Discussion
In this work we present a fast and reliable detection method of bound ligand proton CSPs for the identification and optimization of favorable CH-π interactions between ligand hydrocarbons (CH ligand ) and protein aromatic amino acids (π protein ). The strategy combines replacement of non-exchangeable protons by deuteration, and ligand based 1 H NMR spectroscopy to probe CH ligand -π protein interactions in protein-ligand interaction sites. Protein deuteration has revolutionized NMR structural biology by greatly extending the molecular weight limit of protein systems amenable to NMR spectroscopy. 16 Additionally, measurements in D 2 O solution eliminate exchangeable (amide and hydroxyl) protons thereby further simplifying the NMR spectra and reducing undesirable relaxation pathways. The combined effect of spectral simplification and improved relaxation properties make the NMR observation of ligands bound to protein targets feasible. [17][18][19] In order to obtain atom-specific CSP data for each individual ligand proton, the assignments of ligand signals both in the free and protein-   Originally the [ 1 H-1 H]-NOESY experiment was designed to probe spatial proximities of protons based on the r -6 dependence of the NOE effect. This information encoded in NOE cross peaks forms the basis for the extraction of distance constraints and is routinely used in NMR for structure determination. In drug design this is valuable information for medicinal chemists since it enables modeling of the ligand conformation when bound to the protein, especially in the absence of co-crystal structure information. 23 Importantly for our Inverse PI by NMR strategy presented here, chemical exchange cross peaks between the free and bound forms can be also observed in the  To widen the scope of this analysis to weaker affinity ligands often dealt with in the early stages of the drug discovery programs, we attempted this analysis with Ligand 2 which has an affinity of 9.7 µM (PDB ID: 6G2E) (Figure 1g). 15  type experiments. 37 The EXSY pattern for the weaker binding Ligand 2 is almost congruent with the more decorated nM Ligand 1 for the chemical moieties shared between the two. This suggests that meaningful information about the presence of distinct CH-π interactions can be obtained in early stages of a drug development program even for weak-to-moderate binding affinities.
In addition to valuable CSP information, the NOESY experiment also gives access to distance information of the ligand-bound state encoded in NOE cross peaks,  Figure 3a shows NOESY cross peaks (dotted blue) from ligand proton 3 to bound ligand signals 5, 10 and 12. Figure 3b shows the corresponding spatial proximity information directly mapped onto the crystal structure of NSD3 in complex with Ligand 1. The thereby extracted distance information provides insight into the relative positioning of ligand protons to each other, which can be especially relevant to determine the bound conformation of flexible ligands. Although this is not the prime motivation of our approach, the easy access to additional structural information about the bound conformation of the ligand can additionally support drug design efforts.

Results and Discussion
We present a straightforward workflow to determine CSP data for ligand protons when bound to their target protein under saturating conditions. A wealth of information can be gained by careful analysis of the ligand's free vs. bound CSPs, which have been largely untapped until now. While the analysis presented here is by no means exhaustive, our data suggests that we can contribute valuable information to important interactions such as CH-π, (de)solvation and its compensation in the bound state. We are not completely independent of a protein-ligand model or a complex structure at this point but are working on additional experiments to perform this type of analysis also in absence of a high-resolution crystal structure and on a timescale that is competitive in the fast-moving environment of medicinal chemistry. Thus, we have complemented the protein-derived PI by NMR approach described in our previous report, 11 with a ligandbased version (inverse PI by NMR), that allows to visualize CH donors on the side of the ligand. This will allow a more comprehensive mapping of relevant CH-π interactions in the binding pocket. Large upfield 1 H chemical shift changes in the ligand are induced by aromatic sidechains of the protein, provided that an energetically favorable CH-π interaction exists. In addition, our data suggests, that desolvation of CH groups that are not compensated by either forming a CH-π or a CH-O interaction in the bound state can also lead to significant loss of deshielding with CSP values of up to -1 ppm. In addition, we could show that even in the early stages of drug discovery programs, where ligands with moderate affinities need to be optimized, the (ligandbased) observation of chemical-shift changes induced by aromatic ligand moieties can be utilized to identify fragments which form the best CH-π interactions.
We anticipate that the implementation of both ligand-and protein-detected PI by NMR strategies will provide novel opportunities in future drug design, particularly for difficult-to-drug targets, by offering direct access to relevant molecular interactions in the protein ligand interface and strategic guidance for medicinal chemistry -both of which have the potential to change the way drugs will be designed in the future.