Investigating Protein–Ligand Interactions by Solution Nuclear Magnetic Resonance Spectroscopy

Abstract Protein–ligand interactions are of fundamental importance in almost all processes in living organisms. The ligands comprise small molecules, drugs or biological macromolecules and their interaction strength varies over several orders of magnitude. Solution NMR spectroscopy offers a large repertoire of techniques to study such complexes. Here, we give an overview of the different NMR approaches available. The information they provide ranges from the simple information about the presence of binding or epitope mapping to the complete 3 D structure of the complex. NMR spectroscopy is particularly useful for the study of weak interactions and for the screening of binding ligands with atomic resolution.


Introduction
Interactions of proteins with other molecules essentially define their functions. [1] Such interactions are not only involved in almoste very process in biological systems, but are also key events, when the externalm odulation of protein functionb y drugs is desired. [2][3][4] The interaction partners pan aw ide range from ions, small molecules, lipids, peptides to other proteins or membranes. Nuclear magnetic resonance (NMR) spectroscopy is av ery efficient technique in order to get information about protein-ligand interactions at atomicr esolution. Besides providing structural information, it also allows for af ast screening, especially of weakly binding ligands. There are several approaches available, which can be used to study protein ligand interactions by solution NMR spectroscopy.T he methods can be primarily categorized into protein observed or ligand observed techniques. In ap rotein observed method, as pectrum of protein is acquired and the ligand is titrated. This provides information aboutt he residues in the protein, which are involved in the direct interaction with the ligand. In ligand observed methods as pectrum of the ligand is acquired and the protein is added. The ligand can be anything ranging from a small moleculel ike ac hemical compound or ap eptidet oa macromolecule like DNA or another interacting protein. The equilibriumi nvolved in as ingle two side protein (A)-ligand (B) complexf ormation A + B,AB is described by the dissociation constant K d [5] where [A],[ B],a nd [AB] are equilibration concentrations of the reactants Aa nd Ba nd the complex AB, respectively.F or ab imolecular interaction the units of thermodynamic equilibrium constant K d are in molar concentrations. Strongestb inding in biomolecular complexes has been found on the order of 10 À15 m for biotin binding to avidin, [6] and on the extreme, a physiologically active protein protein interaction in phosphorylation signaling with a K d of 25 mm has been reported. [7] The size of K d is determined by the on (k on )a nd off (k off )r ates of the ligand on its target according to The lifetime of the protein-ligand complex, which is given by 1/k off is responsible for the overall appearance of the NMR spectra. Twol imiting cases can be described:s low and fast chemicale xchange between the free and protein-bound form of the ligand. In thes low exchange regime the lifetimeo ft he protein-ligand complexi sm uch longert han the differencei n chemicals hifts between two signals observed for the free (w f ) and bound (w b )f orm that is, j w f Àw b j @ 1/k off .T his situation, which is found typically fors trong complexes resultsi nt wo observable NMR signals. On the other side aw eak binding event is found when j w f Àw b j ! 1/k off .I nt his case the ligand exchanges fast between the free and bound form, which leads to ac ollapse of the two signals into as ingle peak, as the lifetime is too short for the observation of individual signals on the NMR timescale. Some of the NMR techniques used for investigating protein-ligandi nteractions only work in the fast exchange regime, whileo thers are only possible for strongly interacting molecules in slow exchange. Their respective windows of interaction strength are discussed for the individual methods. Besides the exchange regime, the size of the protein and the ligand and of course the desired information have to be considered in selecting the mosta ppropriate NMR experiment for aparticular interaction.

Chemical Shift Mapping
Upon formation of protein-ligandi nteractions several physical parameters of both the protein and the ligand change. First of all there will be changes in the local electron density due to for example, differences in the hydrophobicity at the interaction surface. Differences in the electron density have an influence on the most easily observable NMR parameter-the chemical shift. Large changes in chemical shifts are also induced by the spatial proximity of groupsw ith magnetic susceptibility anisotropies, like aromatic rings.I mportantly,t he chemical shift is not only influenced by the change in the covalent molecular structure of ap rotein but also by the non-covalenti nteractions with ligands and solvent molecules. One of the most important protein observed methodsf or the investigationo fp rotein-Protein-ligand interactions are of fundamental importance in almosta ll processes in living organisms. The ligandsc omprise small molecules, drugs or biological macromolecules and their interaction strength varies over several orders of magnitude. Solution NMR spectroscopy offersalarge repertoire of techniques to study such complexes.H ere, we give an overview of the different NMR approaches available. The information they provider anges from the simple information about the presence of binding or epitope mapping to the complete 3D structure of the complex. NMR spectroscopy is particularly useful for the study of weak interactions and for the screening of bindingl igands with atomicr esolution.
[a] W. Becker, + K. C. Bhattiprolu, + N. Gubensäk, + Prof.K .Z angger Institute of Chemistry,U niversity of Graz, Heinrichstrasse 28, A-8010G raz (Austria) E-mail:klaus.zangger@uni-graz.at ligand interactions is the chemical shift mapping (CSM) also known as the chemical shift perturbation (CSP) or complexation induced changes in chemicals hifts (CIS). [8] Thereby,a series of NMR spectra of the protein are recorded in the absence and presence of varying amountso ft he binding ligand.
Due to its superior signal dispersion, the most common experiment which is used for chemical shift mapping is the 15 N-heteronuclear single quantumcorrelation (HSQC) experiment.T ypically,p roteins have to be uniformly labeled with 15 Nb yp roducing them in genetically engineered E. coli bacteria. Binding is most easily seen by overlaying all HSQCs recorded during the titration. If there is an interaction, the chemical shifts of the residues involved in the complex formation with ligand, seen as peaks in a 15 N-HSQC,a re displaced from their original position. As described above,t wo limiting cases are found. In the fast exchange limit the two signals collapse to one, whose chemicals hift represents the population averaged value of the free and ligand-saturated protein. Depending on the relative amount of protein, ligand andt he K d value the resulting signal is somewhere between the free and bound state. In the slow exchange regime both signals of bound and free state are observedw ith signal integralsr epresenting their relative amounts (see Figure1).
As an example of chemical shift mapping, at itration of lysozyme with histamine [9] is shown in Figure 2. For this titration,a series of 15 N-HSQC spectraw as acquired on natural abundance hen egg white lysozyme at ac oncentration of 5mm with 32 scans per increment,a mounting to at otal experimental time of just over 1hour for each two-dimensional spectrum. Despite the low natural abundance of 15 N( 0.4%), reasonable 15 N-HSQC spectra can be recorded at such high concentrations.
One essential requirement for chemical shift mapping is that both the protein and the ligand are dissolved in the exact same buffer and measured under the same conditions, since chemicals hifts, especially those of amide protons are very sen-  For binding in fast exchange, the signal of the free-proteinpeak at w F is movingt owards the fully ligand-saturated protein peak w B .Ins low exchange, only the relative signal intensities of free and bound protein peaks change. Adapted from Ref. [2] with permission. sitive to differences in pH value, temperature and buffer composition. The shifting of ap articulars ignal in CSM experiments does not always indicate that the corresponding residue is close to the binding interface. Conformational changes also lead to differences in resonance frequencies.T hese peak shifts provide information about allostericc hanges in the protein upon the binding of al igand. There is no direct way to distinguish these shifts from the shifts which resultf rom ad irect interaction. However,t he peak shifts which are due to ac onformationalc hange can be usually observed in ar egion of the protein target which is buried inside the structure or is located away from the interaction surface. Ac hemical shift titration can also be used to determine the dissociation constant for weakly bound ligands. The chemical shifts of any affected protein signal measured at different ligand concentrations can be used in an onlinear least-squaref itting to obtain the K d value using the equation below: where Dd obs is the change in the observed shift relative to the free state, Dd max is the maximum shift change in saturation, [A] t is total protein concentrationa nd [B] t is total ligand concentration.W hen CSM is carriedo ut on ap rotein with known resonance assignments,t he residues which are involved in the interactions with the ligand are revealed. Chemical shift mapping is also particularly useful for the screeningo fl igands, and is therefore often used in drug design. It not only provides information about bindinga nd the binding strength,b ut also about the location on the protein where the interaction takes place. Small ligandst hat bind weakly to nearby regionsc an then be synthetically connected int he search of am ore tightly bindingl ead compound in ap rocess called SAR (structure activity relationship)b yNMR. [10] In cases where ligand binds to multiple binding sites of a protein with different affinities, ac hange in the linearity of the HN peak shifting in 15 N HSQC spectra is observed. The signal shifts in as traight line until the primarys tronger binding site is saturated by the ligand andt hen changes the direction of the shift while the ligand is occupying the second weaker binding site. An example of an onlinearp eak shifting can be seen in Figure 3. The Figure shows ar egion in the overlaid 15 NHSQC spectra of the 15 Nl abelledT AZ2 domain of at ranscriptional coactivator titrated with the unlabeled tumors uppressor p53 domain AD1 domain. [11] The color of the peaks in the spectrum changes from black to magenta,b lack indicating the initial spectrum of the free protein and magenta is the final spectrum where the protein to ligand ratio is 1:5. The two binding events were fitted to K d values of 24 and 164 mm.
Besides the requirement to use 15 Ni sotopically labeled proteins (unless very high concentrations are used), chemical shift mapping relies on well-resolved protein peaks in the HSQC spectra.T he linewidthso ft he signals increasew hen going to larger protein due to faster transverse relaxation, resulting from slower molecular tumbling. Well-structured proteins beyond % 40-50 kDa typicallyy ield 15 N-HSQCs pectra whose quality is not good enough for chemical shiftm apping. Several methodological developments in the last decades like transverse relaxation-optimized spectroscopy (TROSY), [12] deuteration, [13] stereoarray isotope labeling (SAIL), [14] direct 13 C-detection [15] or methyl-TROSY [16] have expanded the size range of proteins that can be analyzed by solution-state NMR spectroscopy.H owever,t hey often requiree xpensive isotopic labeling strategies. It should also be noted that for intrinsically disordered proteins [17] much narrower NMR signals [18] are observed and protein-protein interactions [19] can also be investigated on  much larger systems. [20] However,t he chemical shift changes upon the interactions of intrinsically disordered proteins are smaller compared to structured ones.T herefore, the preservation of constant solutionconditions during the titrationise ven more important and the high resolution of pure shift spectra might be helpful. [18,21] The vast majority of chemical shift mapping experiments is done on 1 H-detecteds pin pairs-e ither with 15 No r 13 C. However, 19 Fi saparticularly useful nucleus for CSM experiments. [22,23] It is the only naturallyo ccurring fluorinei sotope and the sensitivity is second only to the proton. A 19 Fn ucleus is typically shielded by 9e lectrons unlike the proton ( 1 H), where the nucleus is shielded by one electron. Due to this difference, the range of fluorinec hemical shifts (over 400 ppm for organo-fluorine compounds) and the sensitivity to its environment is much higherw hen compared to hydrogen. Signal overlapi s rarely seen in 19 FNMR. Naturally occurring proteins do not contain any fluorine nuclei. However,s ynthetically fluorinated amino acid analogues can be incorporated into proteins. [24] Incorporation of fluorinated amino acid residue requiresabacterial strain for recombinant protein expression, which is auxotrophic for ag iven aminoa cid. For site-specific labelling, pairs of transferR NA (tRNA) and aminoacyl-tRNA synthetase have been developed. An engineered E.Coli strain introduced with this pair incorporates the desired synthetic amino acid in vivo during translation. [25] Another strategy to place fluorine into a protein is in vitro covalent attachmento fafluorinated reactant/fluorinated molecular probe to the functional groups of amino acids. Cysteines are favorable residues due to their ability to form covalentb onds through their side chain sulfhydryl group. Howevert he side chain NH 2 of lysine or hydroxyl group of serine/threonine can also be used as ap ossible site of reaction. [24] In order to investigate protein-ligand interactions by 19 FNMR al igand titration using as eries of simple 1D fluorine spectra has to be recorded. Fluorine NMR has been also successfully employed for screening potentiald rug candidates. One indirect way to find enzyme inhibitors is the FABS (fluorine atoms for biochemical screening) approach. [23] FABS focuses on the substrate conversion of an enzyme to screen for potential inhibitors. In the presence of an active enzyme, both the substrate and the product, which are both fluorine labelled, produce 19 Fs ignals at distinctive resonance frequencies. When ap ositive hitf or an enzyme inhibitor is present,t he 19 F signal for the product disappears( or becomes less intense). This is because the inhibited enzymec an no longerc atalyze the substrate to product or only al ittle amount of product is formed (see Figure 4).

Hydrogen Exchange
Exchange rates of backbone amide hydrogens with bulk water are often used to get residue-specific information aboutt he solventa ccessibilityi naprotein. It was proposed more than 40 years ago that changes in amide exchange rates upon the addition of ab inding partner couldp rovide an epitope mapping approach for protein-ligandi nteractions. [26] Amide protons protected from bulk water by the ligand show reduced exchange rates. Most easily the exchange rates are measured by monitoring signal intensity changes of amide protons upon dissolving the protein in deuterated water.D ue to necessary time consuming 2D NMR acquisitions, only slowly exchanging amide protons can be measured. This approachw as first applied to map the binding epitope of am onoclonal antibodyt o horse cytochrome c. [27] All residues, whose exchange rates are affected upon antibodyb inding, weref ound in ac ontiguous regiono nt he protein surfaceo fc ytochrome c( see Figure 5), indicating no major structuralc hanges upon interaction with the antibody.
Although this method works quite reasonably for stable, well-structured proteins [27,28] it should be interpreted with caution when more flexible, less stable proteins are investigated. Binding of the c-Src SH3 domain to as mallp eptide ligand led to reduced amide exchange rates throughout the protein. This overall reduction in exchange rates could be attributed to ar educed population of the protein in ah igh-energyu nfolded state once the ligand is bound. [29] Hydrogen exchange has been used to study lowly populated high energy conformations as it is av ery sensitivep robe for such structural Figure 4. Schematic representation of the FABS method. "S" represents a substrate and "P" ap roduct peak in the 19 FNMR spectrum. In as ample containingt he free enzyme, the fluorine-containings ubstrate and product peaks are visible, whereas the presenceo fa ninhibitor suppresses the product peak. changes. [30] To probe the actual influence of ligand binding on the exchange rates, especially for proteins with al arger proportion of high energy conformations, it is necessary to measure exchange rates at different ligand concentrations. [29]

Solvent Paramagnetic Relaxation Enhancements
Additiono fa ni nert, freely soluble paramagnetic agent to a protein solution leads to increased relaxation of protein nuclei. These solvent paramagnetic relaxation enhancements (sPREs) depend on the distance between the observed nucleusa nd the paramagnetic probesi nt he vicinity. The relaxation enhancement (both T 1 and T 2 )o fasingle paramagnetic center are proportional to 1/r 6 , [31,32] where r is the distance between the paramagnetic centera nd the observed nucleus. For a plane surface the effect of all paramagnetic molecules in solution has to be added up weighted with 1/r 6 ,w hich yields a1 / d 3 dependence of the sPRE, [32] where d is the distance to the surface. For ap rotein, whose surface is not really flat, the effect of all paramagnetic centers can be added up by a1 / r 6 weighted grid search. [33] Overall, nuclei closer to the surface, that is, more solvent exposed show highers PREs than nuclei furtheri nside ap rotein. Ab ound ligand "protects" the binding interface from enhanced relaxation and can therefore by detected by reduced sPREs. This approachw as used for example, for the interaction of matrixmetalloproteinase 3( MMP3)t o tissue inhibitor of metalloproteinase 1( TIMP-1). Thereby,s PREs were obtained by monitoring linewidth changes in the presence and absence of the inhibitor, using Gd(EDTA) as the sPRE agent. In this study,b roadening of signals could also be observed outsidet he binding pocket. They couldb ea ttributed to conformational changeso ft he protein upon binding. More quantitative sPREs, obtained from actual relaxation measurements in the absence and presence of the binding partner can be used as experimental input for protein-protein docking studies. [34] For exact sPREs the paramagnetic agent needs to be very inert towards the investigated system to prevent locally enhanced sPREs due to specific binding. [35] Furthermore, only sPREs of non-exchangeable protonss hould be used to avoid transfer of very high water sPREs onto the protein.

Saturation-TransferDifference
The saturation-transfer differenceN MR (STD-NMR) approach is al igand-based screening technique, builds upon the nuclear Overhauser effect (NOE) [36] and works for ligandsi nf ast exchange, with K d values in ar ange of 10 À8 -10 À3 mol L À1 .F or an STD experiment, the protein target is selectively irradiated by a radiofrequency field which only hits resonances of the protein and removes the magnetic polarization of these nuclei. This is also called the on-resonance (I SAT )s pectrum,w here no ligand resonances are irradiated and frequency values from around À1t oÀ1.5 ppm are typicallyc hosen. Alternatively,i ft he ligands do not show resonance signals in the aromatic region, the saturation frequency can be set up further downfield to around1 1-12 ppm. When al igand is in fast exchange between the free and protein-bound form, the saturation gets transferred through the protein to the bound ligand and by exchange, that saturation is carriedo nt ot he free ligand where it is detected with high resolution ( Figure 6).
As the method name already implies as econd spectrum, where saturation takes place off-resonance (I 0 ), has to be acquired. In this experiment the irradiation frequency is set "outside" of any resonancef rom ligand and protein, for example at 40 ppm and yields an ormals pectrum of am ixture. In the differences pectrum (I STD = I 0 ÀI SAT )o nly signals from saturated ligands that interactw ith the protein will remain. All other components,w hich do not bind to the protein and consequently are not saturated, will be absent. The saturation through the protein and to the ligand is very fast (on the order of % 100 ms). Thus, if the off rate of the ligand is fast, the saturation gets quickly into the solution.I falarge excesso fl igand is used, the saturation of free ligands in solution gets amplified because the relaxation of small molecules is slower than the saturation transfer.T he saturation transfer depends mostly on the off rate and therefore, larger off rates produce larger STD signals. However,i ft he dissociation constant reachesavalue of approximately 10 mm,S TD signals become very weak as the saturation transfer is not efficient enough. Whenever,c ompoundm ixtures become more complex, additional information is needed. In this case, any two-dimensional NMR experiments can be combined with STD. STD-NMR is often associated with group epitope mapping (GEM). [37] Largest signal intensity changes can be found for protons that are in close proximity to their interacting protein. The knowledge about the epitope of the ligand is the starting point for designing and optimizing new drugs. [38,39] In addition, STD NMR has also been appliedt o characterize the interactions betweenl igands in context of membrane protein, [40] living cells, [41] viruses [42] and microtubule assemblies. [43]

. Water LOGSY
Av ariation of STD NMRs pectroscopy is WaterLOGSY (waterligand observation with gradient spectroscopy). [44] Water plays ac rucial role in the protein-ligand, protein-protein and protein-DNA/RNA interaction mechanism.W ater is presenta ti nterfaces of interacting molecules. [45] Water-ligand NOEs are negative, meaningt hat the residence time of water molcules is longer than 1ns. [46] This water can be either squeezed in between ligand and protein or located in aw ater shell surrounding the ligand.B ased on these observations, WaterLOGSY was developed to use bulk water to reveal the binding mechanism of ligands to proteins.B ya nalogy to STD NMR, on-and off-resonances pectra are acquired. The on-resonance saturation is appliedatthe water chemical shift and the off-resonance is applied outside of any ligand/protein resonances. After subtracting the on-resonance spectra from the off-resonances pectra, a negative NOE in case of ab inding eventc an be observed. In order to maximize the magnetization transfer rate, WaterLOG-SY is using all magnetizationt ransfer pathways such as spin diffusion, etc.. [47] Whenever,t he residence time is longer than % 300 ps the NOEs change sign and increase in magnitude.I n general,t he bigger the protein the longer the rotational correlation time and consequently,t he magnetization transfer is more efficient. The correlation time of the protein can be increasedb yd ecreasing the temperature or increasing the solution viscosity.

Cross Saturation/Transferred-Cross Saturation
Cross-saturation( CS) is at echnique, related to STD, for mapping the binding area between two proteins. [48] The transferred cross-saturation (TCS) methodi sa ne xtension of CS and enables the location of the interface between protein ligands and huge complexes (> 150 kDa). [49] Cross-saturation requires specific isotopic labeling of one protein. The chosen labeling strategy determines what experiment should be used and which residues are detected. One possible approachi st ou niformly label one of the proteins, protein I, with 2 Ha nd 15 N,w hereas the other one is unlabeled and as as olvent1 0% H 2 Oa nd 90 % D 2 Oi su sed. Under these conditions protein Ih as low proton density therefore spin diffusion in protein Ii ss uppressed. The complexi si rradiated at af requency,w hich only affects protein II, typically aliphatic protonr esonances are selected (see Figure 7). Since protein II has ah ighp roton density,s pin diffusion takes place and the saturation is immediatelyt ransferred to the other protonso fprotein II and further on to the binding area of protein Iv ia cross-saturation. There is no spin-diffusion in protein I, because of the low protond ensity,s ot he saturation cannotb et ransferred further than the interface. Typically, 1 H-15 NH SQC spectra are recorded beforea nd after the irradiation of protein II. Because of the low H 2 Oc ontent in the solvent, the amide protonso fp rotein Ia re partially protonated, so it is possible to detect them but of course with low sensitivity.T he residues of protein I, which are at the binding interface can be identified through ar eduction in intensity.
The transferredc ross saturation (TCS) method is used for huge complexes with low affinity binding. An excess of smaller protein Ii su sed to achieveafast exchange rate between bound and free state. Ar eductioni ns ignal intensity can be observed in the free form of the protein I. This technique can be used, for example, for interaction mapping betweens mall soluble and membrane-attached proteins. As an example of the TCS approach, Shimada and co-workersu sed it to characterize the weak binding site of insulino nt he insulinr eceptor (IR). [50] Transferred CS had to be applieds ince the IR has aM W of 460 kDa. The used labeling schemes differed slightly from the one described above. Twoi nsulin samples were prepared: one labeled with 2 H, 15 Na nd methyl-1 H, 13 Cw hich was used to detect only the methyl containing residues and the other sample was partially deuterated and 15 N 13 Cl abeled only at aromatic residues.T he experimentsw ere carried out in 99 % D 2 O. For studying the IR-insulini nteraction, the ligand insulin was used at an 15:1 excesst oa chievef ast exchange between the free andb ound state. In Figure 8az oom of the aromatic HSQCsb efore and after the irradiationi ss hown, revealing the intensity reductionso ft he aromatic residues located in the binding epitope of insulin.A ll residues of insulin showing signalr eductions of at least 30 %t hrough TCS from the insulin receptora re indicated on the structure. They are all exposed on one sideofi nsulin, which contains the binding epitope.

Transferred NOE
The NOE is used for measuring through-space distances. [51] It results from ad irect dipolar cross-relaxationb etween neighboring nuclear spins, whichd rops off sharplyw ith increasing distance. Usually NOEs betweenp rotons can be observed up to % 5 .T he sign and size of the NOE depends on the rotational correlation time and therefore on the size of the molecule. Proton-proton NOEs are small and positive for small molecules and large and negative for large molecules. Being ar elaxation phenomenon NOEs also take somet ime to build up and this takes much longer for small compared to large molecules. These properties of the NOE are exploited in transferred NOE (trNOE) experiments. [52] As mall ligand in fast exchange between free and bound to al arge protein developsr elatively fast al arge negative NOE while it is bound to the protein.
When it comes off the protein, as mall positive NOE forms, but that takes much longer and is therefore rather insignificant. The negative NOE, which is transferred from the bound conformation can be observed in nicely resolved spectra with sharp signals at the positiono ft he free ligand, but at the same time getting important NOE distance information of the ligand in the bound state (see Figure 9).
The NOE could be transferred further on to another competitive ligand, whichi so bserved in the INPHARMA approach. [53] Thereby,l igands that bind consecutively in the same binding side can be identified by transfer of the NOE from one ligand via the protein to the other ligand.
An example application of the transferred NOE is the binding of d-gluco-dihydroacarbose (GAC1), which acts as an inhibitor,t og lucoamylase. [54] The availablec rystal structure of GAC1 in complex with the catalytic domain of glucoamlyase reveals an unexpected bond conformation of the N-glycosidic linkage. GAC1 has two conformations, which are pH dependent, con-formationAis favored under basic conditions (pH 9.0) whereas conformation Bi sp resent at low pH (pH 3.0). Althought he crystal structure mentioned before was determinedu nder low pH conditions, the bound conformation of the solid-state complex resembles conformation A. To exclude possible artifacts from crystal packing forces at ransferred NOE experiment was used to confirm the unexpected conformation of GAC1, which is selected by the enzyme upon binding. Figure 10 showsb oth conformations of GAC1 at the N-glycosidic linkage: (a) presents the bound conformationA(b) reveals the inverted conformation B, found in solution. On the right side the trNOE spectrum of GAC1 and glucoamylasea tp H4.5 is shown, revealing inter-glycosidictrNOEswhich confirm conformationA .

NOE Editing/Filtering
For tight protein-ligandi nteractions NOEs can be used for regular structure determinations as long as the size of the complex is within the size limitations of NMR experiments. In order to simplify NOE assignments within the protein, within the ligand and between them,s everal different isotope labeling ( 2 H, 15 N, 13 C) schemes are typically used. Combined with isotope editing and filtering NMR experiments one can obtain intramolecular cross-peaks from either the labelled protein or the unlabeled ligand in the complex, or intermolecularN OEs from their interface region only. [55] For example, by mixing a 13 C, 15 Nl abeled protein with an unlabeled ligand, one can record intramolecularN OEs of the protein with 15 N, 13 C-edited NOESY spectra. On the other hand, using 13 C, 15 Ni sotope filtering methods the signals of protons bound these nuclei are filtered out, leaving only NOEs of the unlabeled ligand. For detecting intermolecular contacts a 13 Ce dited, 15 N/ 13 Cf iltered NOESY experiments shows only NOEs between 13 Cb ound protons of the labeled protein and 12 Ca nd 14 N-bound protons of the ligand. [56] As an example the NOE-based structure of the complex betweent he homodimeric bacterial antitoxin CcdA with its cognate DNA [57] is shown in Figure 11.H ere, intermo-lecularN OEs were not only essential to determine the binding interface between the protein and DNA, but also to define the even larger interaction region between the two monomers of CcdA. A 13 C-edited, 13 C, 15 Nf iltered 3D NOESY-HSQC was ac- Figure 9. Schematic representation of the transferred NOE experiment. Large negative NOEs build up for al igand bound to al arge protein. When the ligand is in fast exchange between free and bound form, the NOEs are transferred to the free ligand, where they can be observed with high resolution. Figure 10. Conformations of GAC1 at the N-glycosidic linkage(a) conformation Af ound by transferred NOEs( b) "inverted" conformation B, which is foundi naqueous solution at the samebuffer conditions. In c) the trNOESY spectrum is shown revealing the interglycosidic trNOEs which confirm conformation Ai nt he complex. Reproduced from Ref. [54] with permission. Reviews quired on as ample that contained an equimolar mixture of unlabeled and 13 C, 15 N-labeled CcdA. Upon mixing, three different kinds of homodimers form:u nlabeled-unlabeled, labeledunlabeled and labeled-labeled. Using the edited, filtered NOESY only intermolecular NOEs of the 50 %l abeled-unlabeled homodimers are recorded.

Diffusion Editing
The diffusion behavior of small molecules is significantly different from the one of large biomolecules, sinced iffusion coefficients are inversely proportional tot he hydrodynamic radius. Translational diffusionc an be measured by NMR spectroscopy quite conveniently using pulsedf ield gradients (PFG), which produce al inear magnetic field variation across the NMR sample. The precession frequencydependsonthe overall magnetic field. AP FG leads to varying precession frequenciesf or a particular signal across the NMR sample and therefore defocusing of its magnetization. Refocusing is achieved by applying another PFG for exactely the same duration and with the same strength after inverting the magnetization with a1 808 radio frequency pulse. This refocusing into observable magnetization only works if there is no diffusion in the NMR sample. Translational diffusionr esults in ad ecay of the magnetization according to where I is the observed intensity after the application of the two gradients, I 0 the intensity with zero gradient strength, D the diffusion coefficient, g the gyromagnetic ratio, d the duration of the gradient with strength G and D the time between the two gradients. The diffusion coefficient of the ligand in equilibrium between free and protein-bound form is given by: where D e is the experimentally determinedd iffusion coefficient of the ligand, D f the one of the free ligand and D b the ligand bound to the protein (so for small ligands the one of the protein), A b and A f are the mole fraction of bound and free ligand, respectively,a nd A T is the total amount of ligand in solution (A T = A b + A f ). Fast diffusings mall molecules can be removed from the spectrum through their faster decay in ad iffusion filter.C ompounds that bind to al arge biomolecule show significantly slowerd iffusion and remain in the spectrum. This approach,w hichh as also been called "affinity NMR" allows the screening of small molecule libraries for compounds to bind to am acromolecule. As an example, the binding of 4-cyano-4' hydroxybiphenyl tos tromelysin [59] is shown in Figure 12. Diffusion editing only works if there is as ignificant size difference between the protein and ligand since the hydrodynamic radiusi sp roportionalt othe cubic root of the molecular weight(MW 1/3 ). Figure 11. Solution structureo ft he bacterial antitoxin CcdA bound to its cognate DNA. 13 C-edited, 13 C, 15 N-filteredN OESY spectra were used to distinguish intra-from intermolecular NOEs.

Relaxation Editing
Besides translational diffusion, rotational tumbling also depends on the molecular size. Al arge biomoleculeh as am uch longer rotational correlation time than as mall molecule. This leads to shorter T 2 relaxation times (broader lines) for larger molecules, whereas T 1 relaxationt imes can be comparable. Due to different relaxation mechanismst his effect is even strongerf or 19 Fa sc ompared to 1 H. Binding of small molecules can be observed through line-width changes,u pont he addition of ab inding protein. [2,60] Dissociation constants can be obtained by monitoringt he linewidth or signali ntensities as a functiono fb inding partner concentrations. Instead of the linewidth, the actual T 2 relaxation times can be followed, or to avoid problems due to scalar couplinge volution, T 1 1 relaxation times are frequently used. T 1 1 relaxation is active during a spin-lock field, which supresses the evolution of homonuclear scalar coupling. Relaxation-editing is achieved typicallyu sing a Carr-Purcell-Meiboom-Gill (CPMG) sequence, which allows for T 2 relaxation time measurements by following the signal decay as af unction of the relaxation delay.S etting this delay to severalh undreds of ms removes fast relaxing signals from the spectrum.T os creen for binding ligands relaxation-edited spectra in the absence of presence of the protein have to be recorded. Bound ligands are removed from the relaxationedited spectrum in the presence of protein and show up after subtraction of the two spectra. Aw ide range of binding affinities can be investigated by relaxation-editing. Typically short CPMG times will be used for high affinity ligands, since their relaxation is more strongly influenced upon binding, while longer relaxation delays have to be used for weakly binding ligands.

Paramagnetic Tags
Spin labels, specifically organic radicals or paramagnetic lanthanideb ased labels have al ong history in NMR and have been used as chemical shift mediators and line broadening agents to determine conformational and structuralc hanges in proteins since the 1970s. [61] Ap aramagnetic center in ap rotein can be used to probe for ligand binding in its vicinity.U npaired electrons of paramagnetic probesc ause an increase of relaxation rates of nuclei up to ad istance of about 20 .T his phenomenon is known as paramagnetic relaxation enhancement (PRE) and causes line broadening in the spectra. The magnitude of the PRE depends on the square of the gyromagnetic ratio, the inverse sixth power of the interspin distance and the correlation time and can be described by the transverse relaxation rate enhancement R 2 [Eq. (6)]. [62] Si st he electron spin, g I the protong yromagnetic ratio, gt he electron gf actor, b the Bohr magneton, r the distance between the electron spin and the nuclear spin, w I the resonance frequency of protons, and t c the correlation time of the vector connectingt he electron and nuclear spins. The correlation time t c can be described by Equation (7) and depends on the rotational correlation time of the protein-ligand complex t r , the electronic relaxation time t s and the lifetimeo ft he complex t m .
All lanthanides have similarc hemical behavior and are inert. Diamagnetic control samples can be easily prepared using different lanthanides.S trong paramagnetism is observedf or Dy 3 + ,T b 3 + ,T m 3 + ,a nd moderate paramagnetism in Er 3 + ,H o 3 + , Yb 3 + ,w hereas La 3 + ,a nd Lu 3 + are diamagnetic. The vast majority of naturalp rotein do not contain ap aramagnetic center. However,s everal approaches are availablet oi ntroduce ap aramagnetic probe into ap rotein. One way is to bind as ynthetic paramagnetic probe site-specifically onto ap rotein surface via reactive amino acids, such as cysteineo ra rtificial amino acids. Other cysteines, which might be presenti nt he protein would need to be mutated away.H owever,c ysteine mutations are often found to destabilize the protein fold and some cysteines are essential for the protein function. As an alternative it is possible to employ non-natural amino acids for site-specific labeling of proteins. [63] For instance ap aramagnetic nitroxyl centerc an be attached to p-azido-l-phenylalanine by click chemistry. [64] Metal-containing proteinso ffer another approach to paramagnetic labelling. Anaturally bound,non paramagnetic metal in ap rotein could be replaced with ap aramagnetic metal ion. This strategy has been successfully used to determine the structure of the 30 kDa exonuclease domain e of E.coli DNA polymerase III (Pol III) in complex with another Pol III subdomain, q. [65] It is also possible to use al anthanide binding peptide(LBP) which can be genetically engineered into the protein. This approachw as used for the screening of low-and high affinity ligands of the Src homology 2( SH2)d omain of growth factor receptor-bound protein 2( Grb2). [66] Binding of ligands to ap aramagnetically tagged protein can be identified using regularr elaxation-editing approaches. [67] With paramagnetic tags on ap rotein, especially when several of them are used at different locations on the protein surfacei ti sp ossible to determine the site and orientation of the ligand on the protein. Ligand signals closer to the paramagneticc entere xperience larger relaxatione nhancements compared to the ones on the opposite side of the ligand. [68] 3.9. Residual Dipolar Couplings When molecules are partially alignedi namagnetic field, an additional splitting can be observed which depends on the orientation of the bond between the coupled spin pair relative to the magnetic field. The reason for this splitting is the dipolar coupling, ad irect interaction between the nuclear magnetic moments, which is very strong in the solid state, leading to very broad signals but averages to zero in isotropic solution. Several meansa re available to produce small degrees of order- ing, like bicelles,b acteriophages, polymer gels or the inherent anisotropy of the magnetic susceptibility of some molecules. [69,70] The resultingr esidual dipolarc ouplings( RDCs) provide structurali nformation,w hich is complementary to the distance information provided by the NOE. [70] In favourable cases the mode of binding of al igand to ap rotein can be determined by the measurement of RDCs of the ligand in the presence of the protein in an alignment medium. Since the RDC also dependso nt he distance between the dipolar coupled nuclei, only one-bond 13 C-1 Ha nd 15 N-1 Hw ith knownb ondlength are used. Therefore, the ligand has to be labelled isotopically when using RDC information. The first applicationo f RDCs in protein-ligand interactions was carried out by Prestegard and co-workers. [71] They observed residual dipolar couplings of 1 H- 13 Cs pin pairs in a-methyl mannoside in the presence of mannose-bindingp rotein-A (MBP) in af ield-oriented aqueous liquid crystal.G enerally much smaller splittlings were observedf or a-methyl mannoside in the same alignment mediumw ithout MBP,b ut they are not scaled down and vary from site to site, indicating ad ifferent orientation of free and MBP bound ligand.S ince a-methyl mannoside is aw eak binding ligand,t he observed RDCs represent ap opulation weighted average of those in the free and bound form. RDCs originating from the bound state can then be calculated with the known dissociation constant. Using the known structure of amethylm annoside and MBP ab inding mode of this complex could be deduced from five experimental RDCs with singular value decomposition. [72] In favourable cases,r esidual dipolar couplings can also be used when the protein target is very large or even embedded in an oriented membrane bilayer.A n example is the transientb inding of the C-terminal transducing undecapeptide, which was selectively 15 N-labelled at Leu-5 and Gly-9. [73] Fast exchange between the free and bound form of the peptide enabled its partial alignment and measureable RDCs.

Summary and Outlook
Protein-ligand interactions play ap ivotal role in almosta ll process in biologya nd for drug development. Their elucidation is often more important for understanding the function(s) of a protein than its 3D structure. NMR spectroscopy is at the forefront of methods for investigating protein-ligand interactions as it provides ap lethora of technique to reveal with atomic resolution the interaction mechanismsf or both weakly and tightly bound ligands. The information provided could be anywhere from the simple confirmationo fb inding to the 3D structureo ft he complex. Although some methods, like the transferredn uclear Overhauser effect (NOE) or saturation-transfer difference (STD) can provide information even for extremely large complexes, protein-derived informationi ss till constrained by the NMR size limit, rendering structural information of proteins beyond % 50 kDa often impossible to obtain. Proteinligand investigations by NMR are inherently in vitro experiments.N MR solution studies on such interactions in vivo, in living cells are justb eginning to emerge [74] and could open new and excitingw ays for future studies.