3. NMR Methods for Drug Discovery and Drug Development
As stated, NMR spectroscopy can be fundamental in studying how drugs interact with their targets. This has been done mainly via the Fragment Based Drug Design (FBDD) approach, which has two sub-approaches: target- (i.e., protein) based, or ligand- (drug) based. Target based screening monitors how the target responds to binding molecules in a method called Structure Activity Relationship (“SAR”) by NMR. Ligand (drug)-based screening methods provide ways to observe the binding/non-binding behavior of the drug in approaches such as Saturation Transfer Difference (STD) and other Nuclear Overhauser Effect (NOE) type methods, diffusion-based methods, relaxation-based methods (i.e., T1 and T2). Target based screening, ligand (drug) based screening, and their respective methods, are discussed in detail below.

3.1. NMR in Fragment Based Drug Design (FBDD)
NMR-based drug discovery can be broadly classified into two groups: chemical and biological (in-cell) categories. One of the principal methods of drug discovery using NMR spectroscopy is called fragment-based drug design (FBDD) [194]. In-cell NMR (biological) based drug discovery techniques will be discussed later in this review.
FBDD was first reported in 1996 [195] and used throughout the late 1990s as evidenced by the use of keywords related to FBDD in papers published during this time [196]. The use of FBDD as a viable drug screening technique began to be widely adopted in the mid-2000s [197]. High Throughput Screening (HTS) is another technique widely used in drug discovery [198]. HTS analyzes molecules from a chemical library to see which ones are suitable leads [198,199,200,201] (see Figure 6). FBDD techniques will screen against a carefully designed fragment library composed of a few thousand molecules (for details on the choice of compounds and design of fragment libraries, see [202,203]) and identified hits are further developed via fragment growing, fragment merging, or fragment linking [194]. For examples of drugs derived from the FBDD approach that are currently in clinical trials, refer to Table 2.
HTS has been productive in drug design [204,205], but the method is time and resource intensive [206] and expensive [206] because of the numerous molecules to be examined (~100 million) [207]. Furthermore, the success rate is only estimated to be at ~50% [204,208]. Unlike traditional HTS, which can survey a large number of molecules ranging from a few hundred thousand to a few million [209], FBDD usually surveys a few thousand molecules (~1000–15000) from libraries with greater chemical diversity [209,210]. FBDD is a main-stream screening technique for drug discovery [207,209,211,212,213,214,215,216] and NMR is standard for many FBDD studies [209]. Additional methods and techniques such as SPR, X-ray crystallography [209,217,218,219,220] etc. have also been used in FBDD studies, accompanied or unaccompanied by NMR experiments. For examples of FBDD derived drugs using methods besides NMR, refer to Table 2.
At the time of writing, and to the best of our knowledge, there are three Food and Drug Administration (FDA)-approved drugs derived from the FBDD approach [221], and over 30 are in clinical trials [222]. The first marketed drug derived via the FBDD approach is vemurafenib [223]. Vemurafenib is also the first drug approved for treatment of BRAF-mutant cancer [224], and is reported to exhibit significant clinical benefit for patients with metastatic melanoma [224]. Venetoclax, a common drug used to treat patients with chronic lymphocytic leukemia [225], is considered the second drug to be discovered using the FBDD approach [221], and ribociclib, a CDK4 inhibitor, the third [221]. The names, structures, targets/applications, and clinical status of vemurafenib, venetoclax, ribociclib, and other drugs are listed in Table 2.
As mentioned, NMR spectroscopy can be used in FBDD in two different ways: (1) target (or receptor) based screening, and (2) ligand-based screening. With the stated advantages and disadvantages, researchers must select based on their available compounds.

3.1.1. Target Based Screening
Target based screening typically utilizes the “SAR by NMR” (structure-activity-relationship by nuclear magnetic resonance) approach [246]. SAR is primarily used to identify and develop extremely tight-binding ligands [247]. The ligand to target binding is traditionally monitored via chemical shift changes [247] using a correlation spectroscopy such as 1H-15N HSQC starting with the target and no ligand present [248]. Multiple spectra for the target are recorded in the presence and absence of ligands. The binding ligand will cause chemical shift perturbations in the target, and these perturbations are often easily visualized by overlaying the two spectra [247]. For example Hajduk et al. investigated the binding interactions of 2-phenylimidazole with the FKBP protein as shown in Figure 7 [249].
From the overlaid spectra, chemical shift changes are measured, and from the molecular location, extent, and rate of the chemical shift changes, the binding site and affinity of the ligand is calculated [250]. Then, by following a procedure completely analogous to that of FBDD (see Figure 6), a ligand developed from multiple fragments can be optimized for the binding site of interest, again by monitoring the changes in chemical shifts of the target. Several examples of the successful applications of SAR by NMR in drug design research are replete in the scientific literature [204,251,252].
SAR by NMR spectroscopy allows researchers to observe directly ligand binding [247] in both solution state and solid-state spectra [253], increasing the method’s versatility [254]. It works particularly well for targeting proteins with adjacent “subpocket” binding sites [248]. Furthermore, SAR by NMR is cost-effective when combined with HTS (High Throughput Screening) [255]. SAR by NMR can also be used even when atomic peak assignments in spectra are unknown, though it is much more powerful when the resonance frequency of each atom is known [254]. The main limitation of SAR by NMR, however, is its inability to distinguish between multiple binding modes (i.e., cleavage of covalent bonds or allosteric changes), and if multiple binding modes are present, it can be difficult to pinpoint the “true” binding site of the ligand solely using data obtained using SAR by NMR [254].

3.1.2. NMR Ligand-Based Screening
Ligand-based screening, the second approach of NMR in FBDD, has three main categories: 1) Saturation Transfer Difference (STD) and Nuclear Overhauser Effect (NOE) type methods, based on 2) diffusion methods, or 3) relaxation-based methods (i.e., T1 and T2).

3.1.3. Saturation Transfer Difference (STD)
Saturation Transfer Difference (STD) NMR depends on the Nuclear Overhauser Effect (NOE), which is often used to enhance the sensitivity of less sensitive nuclei such as 13C and 15N [256,257]. This increase in sensitivity is possible because of dipolar coupling (i.e., through space interactions of separate nuclei) [257]. The increase in sensitivity is actually brought about by applying a long, low power radiofrequency pulse that selectively saturates the magnetization [256] of a specific chemical group (i.e., the methyl groups on a protein), which is then given time to transfer to another chemical group via the NOE dipolar coupling within a few angstroms [258]. The transfer in magnetization is easily visualized on a NMR spectrum that takes the differences in the signal intensities from before and after the irradiation. This new spectrum is called a “difference spectrum”, and it reveals what chemical groups interact with the irradiated signal [259] (see Figure 8).
STD NMR is an application of NOE used to probe the binding of ligands to a specific site within the targeted proteins [256]. A generic example of detecting ligand binding via STD is presented in Figure 9a. The STD NMR method follows the same concepts as a normal NOE experiment: a spectrum of the ligand in the free, non-binding form is recorded, the ligand is allowed to bind to the protein, which has a functional group of interest (i.e., methyls) with a saturated signal from a previous selective radiofrequency pulse. The saturated signal travels to the ligand, increasing the intensity of a signal on the ligand spectrum and finally a difference spectrum is used to determine precisely which sections of the ligands bind. The difference in peak intensities proves the presence of ligand binding [260].
Water-Ligand Observed through Gradient Spectroscopy (WaterLOGSY) is a second type of STD (see Figure 9b). The main difference with normal STD NMR is that water is the saturated signal [261], and instead of observing lower peak intensities, peak inversions indicate the presence of ligand binding [209].
For STD NMR to work properly, the ligand concentration must be in large excess (often 100–1000 fold) over the receptor so that effective saturation transfer can take place [260]. This means that for STD NMR, and WaterLOGSY, only small amounts (µg) of protein are required to get results [261,262,263]. This is advantageous for researchers, as they can perform STD NMR on a protein of interest, and preserve the rest of the unused sample for future/other experiments. Also, the same sample can be used for multiple NMR measurements. STD NMR facilitates the differentiation of binding ligands from non-binding ligands because the change in signal (as determined by the difference spectrum) is easy to measure and observe, as shown in Figure 9. WaterLOGSY has been extended to study ligand interactions with DNA and RNA [261].
There are additional NOE-type experiments (trNOE, INPHARMA, SALMON, etc.) used for drug design, and specific details regarding individual methods are found in the scientific literature [264].
With the pressing search for new antiviral drugs, any techniques for identifying and characterizing novel leads has become increasingly important. Benie et al. [265] described the use of saturation transfer difference (STD) NMR spectroscopy [262,266,267,268,269,270,271] to identify and characterize the binding of an antiviral compound to native human rhinovirus serotype 2 (HRV2). The experiments demonstrated that it is possible to subject targets of the size and complexity of whole viruses (for a model of an HRV2 particle cut open, cf. the table of contents) to STD NMR experiments. The principles of STD NMR have been known for many years [267,268] but it was only recently that the potential of this technique for screening libraries for compounds with binding activity toward protein receptors has been realized [262,266]. The technique also permitted the analysis of epitopes of ligands bound to receptor proteins. Previous NMR studies of virus-ligand interactions used chemical shift titrations, which required very large quantities of the virus. This approach was unworkable when studying pathogenic viruses. Benie et al. [265] demonstrated that solution state STD methodology not only reduces the amount of virus required by approximately 2 orders of magnitude, but also allows for the identification and characterization of virus-ligand interactions with atomic resolution [272].
The very large size of viruses makes them particularly attractive for studies by STD NMR, as they inherently yield large line widths allowing for easy irradiation of the virus without affecting the ligand protons. Furthermore, because of the larger correlation time of a virus in comparison to an average-sized protein, spin diffusion, and thus saturation transfer, is very efficient. The large line width has additional benefits not just for STD-based NMR methods but also for transfer NOESY spectra, as protons from the virus capsid are invisible in the NMR spectra (for an example of a transfer NOESY spectrum, see [265]). Moreover, competitive STD titration experiments can be used to determine the Kd value of a ligand [271]. Analysis of the STD spectra using the group epitope mapping method [271] allows for the determination of the binding epitope. STD NMR methods can considerably speed up the determination of the binding epitope for potential antiviral lead compounds.
Simple STD NMR experiments provide substantial information on the binding of ligands to native viruses and require very small amounts of the virus with measurement times in the range of tens of minutes. This allows for a high throughput of ligand samples without significant consumption of viral material because it remains unaffected by the experiments and is easily separated from the low molecular weight ligands by ultra-filtration subsequently. In addition to the detection of binding, a complete mapping of the ligand-binding epitope can be achieved [265].
Noroviruses (NV) are non-enveloped, single-stranded, positive-sense RNA viruses that are the major cause of epidemic outbreaks of gastroenteritis worldwide [273,274,275]. The viral coat consists of a single protein, VP1, which assembles into a capsid with overall icosahedral symmetry [276,277,278]. Attachment of human noroviruses to histo-blood group antigens (HBGAs) is thought to be critical for the infection process [279]. The protruding domains of the VP1 proteins, called P-domains, harbor highly conserved binding sites for HBGAs. STD NMR-based epitope mapping was used [262,271] to identify structural features of different core types critical for the binding of synthetic A- and B-tetrasaccharides [280] to virus-like particles (VLPs) of a highly homologous GII.4 strain (Ast6139). STD NMR experiments provide a robust and straightforward technique for obtaining ligand binding epitopes at atomic resolution. Comparing binding epitopes of related ligands then delivers critical information about structural requirements for ligand recognition. Conversely, comparison of binding epitopes of a given ligand binding to wild type, and to mutant proteins reveals the importance of individual amino acids for binding. STD NMR experiments with L-Fuc and B-trisaccharide in the presence of wild type and mutant VLPs yield virtually identical binding epitopes and suggest that these two mutations do not significantly alter HBGA recognition. The STD NMR approach to characterize binding of HBGA ligands to noroviruses has employed VLPs as targets and thus taken advantage of the large size of VLPs yielding excellent signal-to-noise ratios of the corresponding STD NMR spectra, as demonstrated previously [281].

3.1.4. Transferred NOE (tr-NOE) in Ligand Based Screening
The application of the transferred NOE (Tr-NOE) effect was first demonstrated by Bothner-By [282]. The Tr-NOE is the nuclear Overhauser effect between ligand spins, which are in chemical exchange between the bound and unbound form with the protein or receptor. Ligands, which are a mixture of target molecules, are small in size (below 500–1000 Da). Since they are usually low molecular weight molecules, they exhibit much shorter correlation times when compared to the receptor and have slow NOE build-ups with no spin diffusion. This is the reason they show small positive NOEs in the free form. When binding to a protein receptor, the situation changes, where the ligand acquires large correlation times in the bound state with rapid NOE build-up. Then they show spin diffusion and a strong negative NOE, which is termed the transferred NOE. Signals arising from the protein are usually not observed for large proteins as they are generally kept low in concentration, with ligands in a high excess concentration. In addition, most of the time protein signals are suppressed by their very short T2 period. It is worthwhile to mention that ligands that are in fast exchange between the bound and the free form (dissociation constants ranging from μM to mM) get enough bound time to transfer the negative NOE from the protein complex to the population of the free molecules, yet usually retain the chemical shift of the free molecule along with the relaxation characteristics. In order to observe tr-NOEs, the following condition have to be fulfilled:(3) |Nb∂b|≫|Nf∂f| where N and ∂ represent the number of molecules and the cross-relaxation rate, respectively. The subscript b and f represent the bound and free form, respectively. Therefore, to observe the tr-NOEs, a high excess concentration of ligands over protein is maintained. On the other hand, if the ligand concentration is kept too high, the excess free ligand in solution will exhibit positive NOE, which can result in a significant reduction of the tr-NOESY enhancements due to negative NOE developed by the very small concentration of bound ligand. Hence, the preparation of the sample becomes tricky and an optimum ratio between 10–30 to 1 is maintained while considering the dissociation constant values. The binding of a ligand to a receptor protein can easily be identified by observing the sign and size of the NOEs. There are some distinct experimental features for the discrimination between tr-NOEs from the bound state and NOEs of the ligand in free states like the build-up rate. For tr-NOEs, this is in the range of 50 to 100 ms, whereas for small ligands it is much longer. There have been various instances of experimental implementations to quickly determine the binding activity of ligand libraries. One example was to find the ligand molecule among a library of 10 similar structure polysaccharides that is bioactive in binding with recombinant E-selectin [283]. This is a protein present in an IgG chimera with a molecular weight of about 220 kDa. In this case, two 2D NOESY spectra were recorded. The NOESY spectra for the ligand library was measured at several temperatures and it was found that most of the 10 compounds exhibited the weak positive NOEs at 310 K, which was then chosen to differentiate between trNOEs showing large negative values. The trNOESY spectra of the ligand library in the presence of protein was recorded at different ratios, such as 5:1, 8:1, 12:1, 15:1, and 20:1, at 310 K. In all the ratios, trNOEs were observed; however, the ratio of 15:1 represented the best-case scenario.

3.1.5. The INPHARMA Method for Pharmacophore Mapping
The INPHARMA method (see Figure 10) was designed to determine the relative orientation between two competitive ligands in the receptor-binding pocket through the observation of inter-ligand NOE between the two ligands. It is a tr-NOE in nature as it is mediated by the bound conformation of the competing ligands and in exchange with the receptor protein. The first example was competitive binding and observation of inter-ligand NOE between baccatin III and epothilone A in the presence of tubulin, which acts as a receptor [284]. Since the observation is on the ligand site, it provides unique advantages. The detailed conformation of a ligand-protein complex can be addressed by conventional NMR. However, it is time-consuming and demands full solving of the structure and there is also a size limitation. From that aspect, ligand-based methods are more useful. The only limiting fact is that it should fulfill all the conditions of tr-NOE explained previously in terms of dissociation constant (Kd), fast exchange regime, and proper ligand to protein ratio. Then, information on the ligand structure can be derived from tr-NOE build up as a function of mixing time. This can be readily explained using the originally proposed schematics [284]. The NOESY spectrum of a mixture of the two ligands A and B in the presence of the common receptor (T) is recorded. Under the situation that each of A and B exhibit competitive binding in a fast exchange regime with the receptor T, intermolecular tr-NOE peaks between the two ligands A and B can then be observed in the NOESY spectrum due to extensive spin diffusion. During the NOESY mixing time, the first proton of ligand A (HA) binds to receptor T, which results in transfers of magnetization from HA to HT. Subsequently, the complex AT dissociates as they fulfill the dissociation constant range, which creates the opportunity for ligand B to bind to the receptor T at the same binding site. This results in the transfer of the magnetization of HT, which had been originally coming from HA, to HB. As a result, an inter-molecular correlation HA–HB can be seen, and this inter-molecular NOE will be a function of mixing time as described above. The detailed analysis of such intermolecular NOE peaks helps in assessing the relative orientation of each ligand in the binding pocket.

3.1.6. Diffusion Based Spectroscopy in Drug Design
Diffusion is the random, translational motion of molecules in solution as a consequence of their thermal energy [285]. This type of motion is often referred to as “Brownian motion”, a motion that describes molecular movement induced by random collisions between the molecules [286]. In the presence of a concentration gradient, molecules will naturally move from places of higher concentration to places of lower concentration [287] after a period of time, t, as shown in Figure 11. Fick’s Law can be used to model this type of movement [288]. The distribution of the diffusing molecules is accurately represented by a Gaussian curve, a normal distribution centered at a single point, which gradually “flattens” as t approaches infinity [213]. The extent to which a molecule diffuses is directly related to its shape, size, and mass [285]. In homogeneous isotropic solutions, the root mean square distance (zrms) traveled by a molecule is given by following equation [289,290]:zrms=(2Dt)12  where D is the diffusion coefficient of the molecule, and t is the diffusion time. Making the assumption that the molecules are solid rigid spheres, the value of D can be calculated according to the famous Einstein-Stokes equation (Equation (2)):(4) D=kbT6πηrs  where kb is the Boltzmann’s constant (1.3807 × 10−23 J/K), T is the absolute temperature, η is the solution viscosity, and rs is the hydrodynamic radius of the molecule [290]. Equation (1) and Equation (2), however, are not universally applicable; they only apply to molecules that are freely diffusing in isotropic, homogeneous solutions, and importantly that can be accurately described as hard, rigid spheres [285]. Different molecular geometries and additional modes of diffusion (i.e., restricted and anisotropic) require more advanced mathematics and theory [291,292], but the essential concepts of diffusion remain the same.
The earliest pulse sequence used to measure diffusion in NMR spectroscopy is the gradient spin echo sequence (SE), developed by Stejskal et al. [293]. The SE pulse sequence is shown in Figure 12. The SE pulse sequence uses a gradient (G) of the externally applied magnetic field, (pulsed field gradient), the first after the 90° pulse, and the other after the 180° refocusing pulse. The first gradient pulse (G1) labels or gradient-encodes the NMR-active nuclei based on their physical position in the sample tube. If the molecules diffuse during the time period they are not in the correct position to experience the second gradient which re-focuses the spins. This is detected via NMR as a signal intensity decrease. After a diffusion time (∆), the second gradient pulse is applied to decode the spatial labeling of NMR-active nuclei, obtaining a well-defined spectra of diffusing molecules in solution [294]. Additional NMR sequences are available for diffusion experiments [295], and are detailed in more comprehensive reviews dealing with the subject [296,297].
The signal intensity of the diffusing molecules depends on three factors, as described by Equation (3) [294]:(5) I=I0e−Dγ2g2δ2  where I is the observed intensity, I0 the reference intensity (unattenuated signal intensity), D is, of course, the diffusion coefficient referred to earlier, γ is the gyromagnetic ratio of the observed nucleus, g is the strength of the gradient, δ the length of the gradient, and ∆ the diffusion time [294]. From Equation (3), it is easy to see that the signal intensity decreases exponentially with time, so it is vital to optimize the values of g, δ, and ∆ for diffusion NMR measurements [294].
The drug design approach based on diffusion NMR is basically a screening technique used to differentiate the binding ligands (drug) from non-binding components [264]. Ligands able to bind should have significantly different diffusion coefficients (D) compared to non-binding ligands [297], i.e., the diffusion coefficients of binding ligands will be smaller than those of non-binding ligands [264]. Thus, diffusion-based NMR is a way of effectively “filtering” and identifying which ligands are binding [264].
Diffusion-based NMR spectroscopy has advantages in ligand based screening applied to drug discovery. For example, Diffusion Ordered Spectroscopy (DOSY) does not require prior separation/purification of the ligand/target solution [298]. Diffusion based NMR allows simultaneous determination of diffusion coefficients in multicomponent systems containing large molecules (i.e., proteins) and possible binding partners (i.e., small drug compounds) [285], and no special labeling or contrasting agents are required, though their use is not exclusively inhibited (for an example of the use of labeled compounds in diffusion NMR spectroscopy, see [299]). A problem occurs when there is significant chemical shift overlap between the binding molecule signals and the target. This situation makes it hard to distinguish the NMR signals [300], and the calculations typically assign an intermediate value to the diffusion rate (i.e., one gets a smear). Multidimensional diffusion NMR pulse sequences are available [301], which may help resolve spectral overlap in 1D experiments [300]. Another issue is that molecules in chemical databases may have generally low solubility [302,303]. Low solubility decreases the overall signal intensity and therefore makes accurately measuring diffusion experiments far more difficult [304].
There are many examples demonstrating the successful application of diffusion NMR in examining drugs of pharmaceutical interest [305], and ligand-target interactions [167]. Hajduk et al. [167] exploited the changes in diffusion rates to detect ligands that bind to the FK506 binding protein and the catalytic domain of stromelysin. Nishimura et al. [306] utilized DOSY, in combination with NOESY to determine the orientation of two guest molecules, p-ethoxyiodobenzene and 2-iodo-6-methoxynaphthalene, within a host composed of a tetrakis(4-hydroxyphenyl)-cavitand and a tetra(4-pyridyl)-cavitand.
Furthermore, Matthias et al. [307] used 1H molecular diffusion and 19F spin diffusion to probe the drug loading properties of the Rf-PEG hydrogel for 5-fluorouracil (FU) and 1,3-dimethyl-5-fluorouracil (DMFU), two anticancer drugs.
DOSY can be combined with Saturation Transfer Difference (STD, discussed earlier in this review) to yield new insights about ligand-target interactions. Kramer et al. [308] combined STD with DOSY to analyze a mixture composed of wheat germ agglutinin and two derivatives of N-acetyl glucosamine (ligands). Using this new technique they were able to obtain high quality spectra of the components in the mixture. Tanoli et al. [309] also combined STD and DOSY to explore the interactions of smaller molecules with bovine serum albumin.
These are just a few examples to show that diffusion NMR spectroscopy has played, and will continue to play, a prominent role in drug design.

3.2. NMR and In Silico Screening-Two Complementary Approaches
In silico (virtual) screening is now a standard technique in drug design and discovery [310] that has been in use since at least 1991 [311], though the exact origin of the phrase “in silico” is not clear [312]. The nearly ubiquitous use of virtual screening is due to its efficiency in searching massive chemical databases in order to generate lead molecules [313] that inhibit protein-protein interactions [314], and its ability to help identity ligand (drug) binding sites on the target of interest [310] to lend insight to the mechanisms of action for lead compounds [315,316]. Virtual screening is often accompanied by in vitro or in vivo techniques for pharmacology drug research [312], to increase drug throughput, helping to reduce the time and cost of developing novel drug candidates [317]. Virtual screening has also been used to identify candidates for anti-viral drugs [318] and anticancer drugs [319]. Several chemical databases are available both for public and academic use [320]. Virtual screening is properly identified as a high-throughput screening (HTS) technique [321], though using its full capacity as an HTS technique is not required for most purposes.
Virtual screening requires a minimum of two inputs, (1) a three-dimensional model of the ligand (drug), and (2) a three-dimensional model of the receptor (protein) [322], the latter generated from the atomic studies of proteins via X-ray crystallography or NMR spectroscopy [323]. Virtual screening is not a truly “stand-alone” technique and has often been combined with additional biophysical techniques besides NMR spectroscopy and/or X-ray crystallography [324], such as differential scanning fluorimetry [325], fluorescence polarization, and surface plasmon resonance [324]. In this section, we briefly introduce how virtual screening has been combined with NMR spectroscopy, and how they are complementary approaches to each other in drug design. The complete details of how virtual screening works, and how it applies to drug design outside of its combination with NMR is well documented in additional reviews [310,322,326,327,328,329,330].
A prime example of the complementarity between NMR screening and virtual docking is found in the work of Chen et al. [331], in which the authors sought to target the A2A adenosine receptor (A2AAR) protein, a drug target for the treatment of Parkinson’s disease [332]. They used virtual screening and an NMR-based screening method against the same 500 molecules in a fragment library so they could compare the results of both methods. The virtual screen successfully predicted (based on calculated binding affinities) four out of the five orthosteric ligands discovered by NMR that were within the top 5% of the fragment library, showing that the two separate methods can give similar and reliable results. Later on, Chen et al. discovered that virtual screening picked up three additional fragments that remained undetected by the NMR-based method, and were, in fact, A2AAR ligands; this shows that though neither method is flawless, they are still perfectly complementary approaches for drug design [322,331].
In another scientific work that integrated NMR with virtual screening, Di Lello et al. [333] found small molecular inhibitors of the enzyme ubiquitin specific protease 7 (USP7), a key regulator of the tumor suppressor protein, p53 [334]. A fragment screen by NMR revealed a series of small molecules that bind in the active site of USP7 near the catalytic cysteine (amino acid 223). A ligand-based virtual screen utilizing the fastROCS program identified ~30 hit molecules, several of which were further characterized by 1H-15N TROSY chemical shift perturbation and line broadening to probe the binding site of the active hits. Di Lello. also tested the active compounds against EOL-1 cells to verify the hits as identified by virtual screening and further characterized by NMR, showing that the active compounds do indeed inhibit USP7 activity. Through additional study of the active molecules and further optimization of their structures, they eventually discovered a series of ligands that bind in the “palm” region of the catalytic domain of USP7, inhibiting its catalytic activity [333]. This study clearly demonstrates that NMR screening-based techniques can be combined with virtual screening to find viable drugs for targets of interest.
Additional examples of the successful integration of NMR and virtual screening as applied to protein targets are also found in the literature, further demonstrating the practicality and complementarity of virtual screening and NMR [329,335,336,337]. For example, Li et al. [338] used virtual screening filtered by NMR to identify and characterize non-metal chelating metallo-β-lactamase (MBL) inhibitors, and in particular, Verona integron-encoded MBL (VIM)-2, when previously there were no clinically significant inhibitors of MBL, since MBL enzymes hydrolyse many, if not all, β-lactam antibacterials compounds specifically designed to inhibit their activity [339]. Furthermore, Shan et al. [340] and Bertini et al. [337] both used virtual screening and NMR, in their respective studies. Through the combined use of NMR and virtual screening, Shan et al. was able to identify, design, and synthesize novel PDZ domain inhibitors, which are proteins implicated in tumorigenesis [340]. Bertini et al. was able to combine NMR to study the interaction of ligands with metalloproteinases, using known inhibitors of metalloproteinases as a starting point [337]. While HSQC NOESY NMR data provided structural and spatial constraints for the proposed 3D models, virtual screening was used to refine the models, and to probe the ligand-protein interaction. In each case (i.e., ligand-protein interaction), Bertini et al. was able to obtain a well-defined ligand conformation in the protein binding site, thus offering a viable alternative to other approaches described in the literature [337]. Clearly, combining virtual screening with NMR-based methods is advantageous in studying how ligands (drugs) bind and interact with targets (proteins) of interest.

3.3. Paramagnetic Resonance in Drug Discovery
Paramagnetic NMR (PNMR) can also play a prominent role in drug discovery [341], as PNMR can provide key structural information in situations where crystal structures cannot due to the weak binding of ligands [341]. PNMR can be used to quantify the binding between ligands and large biomolecules such as proteins, DNA, and RNA [342].
PNMR depends on the presence of a group (called the paramagnetic center) with an unpaired electron [343], and since many naturally occurring biomolecules and organic compounds lack a paramagnetic center, one such as caged lanthanide (CLaNP) [344], must be introduced artificially [341]. Once the paramagnetic center (often a metal ion) is present, paramagnetic effects can be used to measure the distance and the relative orientation (i.e., angle) between molecules [345]. This information is crucial when it comes to determining how ligands and substrates bind. Thus, PNMR is quite a useful technique for drug discovery when a paramagnetic center is present. The most relevant consequence of PNMR for drug discovery is paramagnetic relaxation enhancement (PRE), although there are a number of studies demonstrating the use of pseudocontact shift (PCS) effect in drug discovery research [341].
Paramagnetic relaxation enhancement (PRE) is proportional to the inverse sixth power of the distance between the paramagnetic center and the nucleus of interest (i.e., 1H), although it does not reveal anything about relative orientation [341]. PRE can give quantitative information in the range of 10–25 Angstroms [346]. Several researchers have taken advantage of this outstanding property to study the structural and dynamic properties of complex biomolecular machineries in their native environment [347].
For example, Iwahara et al. (2003) demonstrated that a protein’s binding polarity to DNA can be determined by PRE, using EDTA-derivatized deoxythymidine (dT-EDTA) with a chelated metal ion (such as Cu2+ or Mn2+) as a probe. dT-EDTA with a chelated metal ion is a convenient choice, as it can be inserted into any position of a synthesized oligonucleotide. With data derived from the PRE effect, one can easily determine the polarity of the protein (or drug) binding to DNA [348]. Several researchers have investigated DNA as a drug target [349], and the study of Iwahara et al. clearly demonstrates, and even indicates, that PRE can potentially be used to study the interactions between a drug and DNA [348], provided that a paramagnetic center such as dT-EDTA or a metal ion is present.
Brasuń et al. [350] also used PRE derived distances between a paramagnetic center and a nucleus of interest. They replaced the Cys-S-S-Cys bridge found in oxytocin and vasopressin with the His-Cu2+-His motif to investigate if doing so would alter the stability of oxytocin and vasopressin. They determined the distances between the Cu2+ ion and 1H nuclei (possible because of PRE), and used these values to generate three-dimensional models of the His-Cu2+-His motifs in both oxytocin and vasopressin. In doing so, they indicated that such an approach using PRE can help in designing new biologically active compounds [350], and hence in drug discovery research, as many drug discovery studies require a reliable models for the successful generations of hit-lead molecules, especially in the case of in silico docking [351]. This study again proves the usefulness of PRE, and therefore, PNMR, in drug discovery research.
In two additional studies, Huang et al. [352,353] used PRE in their individual studies of protein binding and protein dynamics, respectively. In the Huang et al. case [352], these authors used PRE to establish a model of the binding between the G-actin protein, and thymosin β4, an actin- binding protein. Using PRE determined constraints (distances) and 1H-15N HSQC, they were able to establish a well-converging docking structure of the G-actin/thymonsin β4 complex [352]. On the other hand Huang et al. [353] did not measure protein binding, but studied the conformational changes and dynamics of select large membrane proteins utilizing 19F-NMR spectroscopy, and Ni2+ as the paramagnetic center. Through a series of extensive experiments, they showed that conformational exchange rates of membrane proteins can be determined from measurements of the metal-enhanced longitudinal relaxation (i.e., PRE) of the 19F nuclei [353], thus yielding additional information (i.e., protein conformation dynamics) that could be utilized in drug discovery projects targeting proteins (i.e., understanding how the protein changes shape based on its environment can be used to find potential binding sites for drug candidates).
All these examples prove that PNMR is powerful approach in drug discovery research, given that PRE can aid in generating trustworthy models of interacting molecules, and that it can help researchers understand better how the molecules interact in the first place.

3.4. Solid State NMR in Drug Discovery
Since the late 1970s solid state NMR (ssNMR) has demonstrated its usefulness in complex biomolecular systems such as collagen or lipid bilayers [354]. However, over the past years ssNMR has gained attention in the field of drug design and is slowly becoming a commonly used technique as its proving to be a powerful tool for structural analysis of membrane proteins and amyloid fibrils [354,355,356].
ssNMR is becoming a more attractive alternative for several different reasons. One of them is the fact that it enables the characterization of a chemical compound in a solid-state form such as in a tablet/pill [356,357,358]. Moreover, ssNMR is not only restricted to analyzing the chemical structure but it can also provide insight into the physical properties of a compound such as polymorphism (different crystalline structures of the same compound), disorder (crystal defects and amorphous solids in the compound) or the presence of cocrystals (multicomponent crystal made of a compound and one or more small organic molecules) [356,357]. ssNMR can also be used to quantify the amount of crystalline against the amount of amorphous material in the sample to establish phase purity (the amount of desired phase separated from other, undesirable phase) [356,357,358].
ssNMR differs from liquid state NMR by the presence of anisotropic interactions. In liquids NMR these effects are averaged to zero as a consequence of rapid molecular tumbling. In solid state however, the molecules are not tumbling rapidly and the residual effects of anisotropic (orientation depended) interactions such as anisotropic chemical shift, magnetic dipolar coupling, and quadrupolar coupling could be observed in the form of broad peaks, with could be much wider than the chemical shift range of the nucleus [355,358,359]. As a results, there has been a constant effort to improve the sensitivity and resolution of solid state NMR spectra, which increased the potential of ssNMR in future applications [360]. One of the methods that works for nuclei with spin value of I = 1/2 is called magic-angle spinning (MAS). It increases the resolution by rapidly rotating the sample around a fixed (or so-called magic) angle of 54.736° [360]. This method can be combined with decoupling, to remove the dipolar couplings between spins. This is done by applying radiofrequency pulses or cross-polarization (CP) transfer of magnetization from abundant and sensitive nuclei such as 1H to less sensitive such as 13C [328,333]. A broader comparison between ssNMR and liquid state NMR is provided in [361].
As mentioned before, ssNMR can provide information about membranes and membrane proteins. For this reason, ssNMR can be used to detect interactions of ligands with receptors embedded to the membrane which enables the mapping of binding site of a receptor by utilizing CP-MAS (cross-polarization magic-angle spinning) NMR and site specific mutagenesis [355]. ssNMR can provide the conformation of ligands bound to the receptor which can then be used to optimize future drug in terms of better affinity and efficiency [355]. Since ssNMR is also applicable to amyloid research, it can be used for probing polypeptide structures of amyloid and intermolecular contacts between fibrils. The potential is for the design a drug that will inhibit the process of aggregation of proteins and peptides [355]. Lastly, since ssNMR gains insight into physical properties of a chemical compound it can be used for control of the process of formulation and processing of a drug to help assess the purity of a compound [358].
An example of ssNMR application related to drug design is the work of Callari et al., who monitored the effect of drug loading on the properties of micelles [362]. Polymer micelles are widely used as nano-carries for drug delivery, but so far the effects of drug loading on the morphology of a drug carrier had not been thoroughly investigated [362]. They created a model consisting of a fructose hydrophilic block and a PMAA block (micelle), to which a different amount of platinum complex was anchored. The results from this experiment showed that micelles loaded with a higher amount of platinum complex had reduced cellular uptake, release, and cytotoxicity. The micelles with a lower load (LL) of platinum complex were more effective at targeting cancer cells (of cell lines MDA-MB-231 (breast cancer) and A549 (lung cancer) than the micelles with a higher load (HL) of the platinum complex. This is evidenced by the lower IC50 (half maximal inhibitory concentration) values of the LL micelles as compared to the HL micelles. Both of those results could be related to the micellar structure and their potential for interaction between the sugar moieties and the cell wall [362].
Another example of practical application of ssNMR is the work of Lee and colleagues [363] in which they investigated the structure of a designed zinc-binding amyloid fibril that catalyzed ester hydrolysis. Metals ions such as zinc where found to affect the process of protein aggregation which resulted in arise of amyloid like structures. Therefore, understanding the processes of aggregation and the factors related to them is crucial for creation of new drugs for amyloid related diseases [364]. In the experiment Lee et al. used Ac-IHVHLQI-CONH2 peptide (referred as HHQ) to form fibrils with varying Zn2+:HHQ molar ratios. The results showed that Zn2+-bound HHQ fibrils form parallel-in-register form of packing β-strand in each sheet and His residues are coordinated to Zn2+ via Nδ1, while half of the His residues are also coordinated to Zn2+ via Nε2. Additionally, Zn2+ binds in a 1:1 metal ion/peptide ratio. After further analysis using structural bioinformatics, it was concluded that each zinc ion was coordinated by three histidine nitrogens from two adjacent strands. Half of all histidines bridged to Zn2+ ions forming a metal–imidazolate chain [363].

3.5. NMR Validation in Drug Design
A “hit” is a molecule identified from a screening technique (HTS, FBDD, etc) as having a desirable effect (i.e., decreased cellular growth, high affinity score) on a target [365,366]. However, the question of whether the activity is related to actual binding to the target, or to interference with one of the components of the assay readout mechanism, is uncertain. Thus, a validation step is required. Hit-validation is therefore the process of confirming, or validating, that the molecule(s) identified previously have on target activity and selectivity [367,368]. One of the highest-impacts of NMR on drug discovery is the use as a hit-validation tool. Though the hit-validation or confirmation of drugs is mostly limited to the solution state [369], this aspect of NMR truly is a “gold standard” technique in drug discovery.
NMR by itself is a powerful tool for drug validation as in the case of Sharma et al. (2012) [370] who sought to identify potential drug-like inhibitors against L-Aspartate α-Decarboxylase (ADC) an enzyme responsible for the decarboxylation of L-aspartate in order to generate β-alanine and carbon dioxide [371], in Mycobacterium tuberculosis. They began with known inhibitors of ADC, and developed a protocol to measure the enzymatic activity of ADC. Upon addition of ADC to a solution of L-aspartate, L-aspartate gradually disappeared because ADC was converted to L-aspartate to β-alanine; therefore the peak intensity of L-aspartate decreased, and the peak intensity of β-alanine increased in the presence of ADC (no inhibitor drug present). Using this newly developed NMR-based protocol allowed direct measurement of ADC enzymatic activity, and Sharma et al. were able to confirm the enzymatic inhibiting activity of seven previously discovered inhibitors of ADC [370]. This study demonstrated that NMR can be an effective validation tool for known drugs and for new drugs generated by a screening approach.
NMR is also able to remove false positives that emerge from biochemical screens [372]. For example, an aptly named technique called A La Assay to detect Reactive Molecules by Nuclear Magnetic Resonance (ALARM NMR) is able to eliminate false positives from HTS methods [373], and in the presence of a test compound or mixture, measures dithiothreitol (DTT)-dependent 13C chemical shift changes of the human La antigen [373]. Dahlin et al. provided an updated protocol of ALARM NMR to aid researchers in the production of the 13C-labeled La antigen reporter protein, in testing compounds with the La protein, and in the analysis of obtained NMR spectra. Using ALARM NMR prioritized hits identified from HTS screening [374]. An example of ALARM NMR is found in the work of Dahlin et al., where they used this technique to test molecules that were assumed to be inhibitors of histone acetyltransferase (HAT) inhibitors, and from their studies, actually discovered that 65% (15 out of 23) of the most commonly reported HAT inhibitors were actually faulty. They were actually nonselective interference compounds, not necessarily specific to the inhibition of HAT [375]. Thus, ALARM NMR (and NMR in general) served as a useful validation method, especially for unvalidated hits identified from biochemical screens [372] or other screening techniques.
The last example highlights the need for cross validation, or the combination of two or more techniques to verify identified chemical hits. Of course, NMR is not the sole technique used for drug validation. Most often, NMR drug validation is coupled with additional methods [367] such as surface plasmon resonance (SPR) [376,377] X-ray crystallography [377,378,379], isothermal calorimetry (ITC) [379], UV-Vis and/or fluorescence spectroscopy [380].
The work of Goudreau et al. is an excellent example of combining NMR with another biophysical technique, in this case X-ray crystallography, for drug validation [378]. A series of benzodiazepine inhibitors of Human immunodeficiency virus 1 (HIV-1) was identified using an in vitro capsid assembly assay, and further characterized by 19F-NMR. Analysis of the chemical shift perturbation and line broadening effect on the 19F-NMR spectra of the benzodiazepine inhibitors revealed the specificity and reversibility of the binding inhibitors. The same set of 19F-NMR spectra were used to identify the N-terminal domain of the capsid as the binding site of the benzodiazepine inhibitors. The specific amino acids involved in the binding of the benzodiazepine inhibitors were identified from the chemical shift perturbation of 1H,15N-TROSY NMR spectra. Later, use of X-ray co-crystallography confirmed binding locations of the benzodiazepine inhibitors and their binding modes, which was useful for further development and optimization of the benzodiazepine inhibitors [378]. The work of Goudreau et al. therefore showed how NMR could be used as a co-validation technique with another biophysical method [378].
NMR can be also coupled with multiple biophysical techniques to validate a molecule’s ability to inhibit protein-protein interactions (PPIs) [367]. An example of the combination of NMR with SPR and X-ray crystallography can be found in the work of Fry et al., where the authors sought to understand how the nutlin molecule inhibits MDM2-p53, a protein-protein interaction that has been an important cancer therapy target for several years [381,382,383]. Fry et al. [377] gradually deconstructed RG7112, the first nutlin molecule to enter clinical trials [384], into 11 fragments so they could study the inhibitory effect of RG7112 on the MDM2-p53 interaction by SPR, NMR, and X-ray crystallography. SPR was used to determine the Kd values of the RG7112 fragments and confirmed that RG7112 and some of its fragments do bind to MDM2, inhibiting the MDM2-p53 interaction. 1H,15N-HSQC NMR chemical shift perturbation was also used to assess and verify binding identified by SPR. Of the six fragments of RG7112 confirmed by 1H,15N-HSQC NMR as binding to MDM2, SPR showed binding for five of them; thus, the two separate techniques were in good agreement with each other. The fragments of RG7112 that were confirmed to bind by both SPR and 1H,15N-HSQC NMR were further studied with X-ray crystallography, which can tell precisely where and how the molecules bind to the protein. Using co-crystallization, Fry et al. were able to obtain structures for several of the verified binding fragments in complex with MDM2 and were able to visualize the binding of the fragments to the MDM2 protein [377]. NMR is obviously a powerful drug binding validation tool, but it becomes much more powerful when coupled with additional biophysical techniques, as seen in the work of Fry et al. [377].
Dias et al. [379] took a similar approach as Fry et al. [377] in that they took known inhibitors of a protein-protein interaction, and dissected them into individual fragments to assess a protein’s drug-ability. The interaction studied was that between the proteins von Hippel–Lindau (VHL), and the alpha subunit of hypoxia-inducible factor 1 (HIF-1α). Twelve compounds (known inhibitors and derived fragments) were developed using a crystal structure of HIF-1α peptide bound to the stable multiprotein complex pVHL-elongin C:elongin B (VCB). Each of these compounds was screened using three separate NMR techniques, Saturation Transfer Difference (STD), Carr–Purcell–Meiboom–Gill (CPMG) relaxation experiments, and WaterLOGSY (to assess drug binding and to predict drug binding mode. Each compound that was unambiguously detected (i.e., the molecule was identified as successfully binding by at least two of the three NMR methods of STD, CPMG, and WaterLOGSY) was subjected to further analysis by ITC and X-ray crystallography. ITC was used to determine the dissociation constants of binding molecules, and X-ray crystallography was used to confirm the binding mode predicted by the NMR studies. Generally speaking, the designed fragments had similar ligands efficacies compared to the parent molecules but had much higher dissociation constants (Kd values), meaning that the fragments bound less tightly than the original parent molecule [379]. With this example, it is possible to see the strength of using NMR as its own hit-validation tool (i.e., three different NMR techniques were used for screening compounds [379]), and yet, the follow-up of NMR studies with ITC and X-ray crystallography was useful in providing a basis for assessing the drug-ability of a protein-protein interaction [385,386,387,388]. Thus, it is clear to see that NMR is a prominent method of hit-validation in drug discovery research, especially in combination with other biophysical techniques.

3.6. Other Methods Used to Determine the Drug-Target Complexes
Substantial progress has been made in the NMR field over the past 5–10 years, and various methods were established to determine the drug-target complexes. Most of them utilize either NOE or chemical shift perturbations (CSP) although in silico models/programs, using NMR-derivate data also exist.

3.6.1. DIRECTION
One of the methods called difference of inversion recovery rate with and without target irradiation (DIRECTION) is used to map pharmacophores and can be an alternative to STD experiments. This method uses the difference between longitudinal relaxation rates of ligand protons with- and with-out irradiation of the protons of the target protein. The DIRECTION approach, however cannot be used for slowly exchanging (strong binding) ligands. The practical approach of this method was demonstrated on the experiment when analyzed the interactions between p38 MAPK (p38 a mitogen-activated protein kinase) and its inhibitor-SB203580 [389,390]. The results from this experiment showed that protons H1, H4, H5, and H6 of SB203580, are in close neighborhood with the protons of p38 MAPK when compared with H2, H3, and methyl protons. It indicates that two aromatic rings (a pyridine ring and fluorophenyl ring) of SB203580 interact tightly with p38 MAPK. The results were later confirmed with proton density map of each ligand’s proton, based on the crystal structure of SB203580–p38 MAPK complex [391]. Moreover, the same authors already created a new and improved protein–ligand docking method by combining the DIRECTION obtained NMR data with docking software. [392].

3.6.2. ILOE
A second method that can be used to map pharmacophores is called inter-ligand nuclear Overhauser effect (ILOE). This 2D NMR experiment detects when two ligands bind simultaneously to adjacent sites on a protein surface although both of the ligands do not have to bind to the same binding pocket (opposite to INPHARMA, see above) [5,393]. A negative ligand−ligand NOE signal will be created when ligands bind in close proximity to each other whereas ligands that do not bind will show no NOEs, or at most very weak positive ones [372,394]. ILOE also enables determination of the ligand orientations with respect to one another [393]. As in the case of INPHARMA, ILOE can be utilized even in the absence of a 3D protein structure and used with large proteins. Additionally, ILOE differs from INPHARMA in mixing times—for ILOE the mixing times are typically in the range of 600–800 ms [345]. Application of ILOE was first shown on glycolate+NAD+ in the presence of porcine heart lactatedehydrogenase, and by glucose-6-phosphate+NADPH in the presence of L. mesenteroides glucose-6-phosphatedehydrogenase and from that time it has been widely used [393,395,396].

3.6.3. SOS-NMR
A third method called structural information using Overhauser effects and selective labeling (SOS-NMR), relies of STD experiments performed on ligand complexes with different protein samples that have been fully deuterated excluding a specific type of amino acid. In other words, the data obtained by SOS-NMR gives insight into the ligand-binding amino acid composition and when taken into consideration the 3D structure of targeted protein can be used to establish the structure of protein-ligand complex. This approach has been demonstrated using two complexes—FKBP complexed to 2-(3′-pyridyl)-benzimidazole and MurA complexed to uridine diphosphateN-acetylglucosamine (UDP-GlcNAc). The results showed that for FKBP and MurA, only four and three amino acids (FKBP: Ile, Val, Leu, Met; MurA: Trp, Phe, His) were needed to be selectively protonated in perdeuterated samples to establish the ligand-binding site. Additionally, on average only 6 amino acids were required for accurate identification of ligand-binding surface. According to authors SOS-NMR can greatly improve the early stages of the drug discovery process [397]. Moreover, combining SOS-NMR with other methods can even further increase chances for a positive outcome of an experiment [398].

3.6.4. Tert-butyl Labelling
A completely different approach to this topic was taken by Chen et al. [399,400]. Instead of using isotope labeling, Chen’s group decided to use a tert-butyl group contained within ligand-1 to obtain structural information about the protein-ligand complex [400]. The tert-butyl group formed an intense singlet in 1.0 to 1.5 ppm range thanks to rapid methyl rotation and methyl reorientation within that group. When compared with the protein’s 1H-NMR signal, the tert-butyl signal tended to be much narrower and resulted in easy detection without the need for isotopic enrichment even in protein complexes of high molecular mass such as Bacillus stearothermophilus DnaB hexamer (320 kDa) [399]. Additionally, the tert-butyl group produces intense NOESY cross peaks that can be observed even in the situations where normally cross-peaks of the proteins are barely detectable. This is partially because the signal corresponded to nine protons within tert-butyl group. Those aspects enable measurements of pseudo-contact shifts generated by paramagnetic tags attached to the protein. As a result, it allows positioning of the ligand on the protein. An example of this approach, is dengue virus NS2B-NS3 protease from serotype 2 (referred as DENpro) in complexed with ligand containing a tert-butyl group. The result of this experiment showed NOEs between the tert-butyl group of ligand-1 and residue Val155 from DENpro [400].

3.6.5. SALMON
Solvent accessibility, ligand binding, and mapping of ligand orientation by NMR spectroscopy (SALMON) is another method based on the data obtained via nuclear Overhauser effect. This method utilizes WaterLOGSY [401] to probe for solvent accessibility to the ligand and determine the orientation of the ligand by analyzing signal changes in WaterLOGSY spectra (positive signal from unbound ligand vs. negative for protein-bound ligands). This method was first used to determine the orientation of prodrug called tretazicar ((5-(aziridin-1-yl)-2,4-dinitrobenzamide) known as CB1954 in NQO2 (quinone oxidoreductase 2) binding site. Previous attempts had been made to obtain the orientation of tretazicar bounded to NQO2, however the results obtained from X-ray crystallography were inconclusive as two orientations of tretazicar could be possible. The information obtained via SALMON showed that the side chain of asparagine at position 161 formed a hydrogen bond with 2-nitrogroup of tretazicar, and that the aziridine moiety of tretazicar pointed toward the solvent [401].

3.6.6. LOGSY Titration
Another variant of WaterLOGSY method called LOGSY utilizes the titration slopes as a measure of solvent accessibility. The titration slopes are created by a constant increase of protein concentrations. This method also provides more insight into the process of ligand solvation by checking the influence of protein concentration onto the process. This approach was used on the bromodomain 1 of protein 4 (Brd4-BD1) by mapping epitopes of two ligands interacting with Brd4-BD1 and predicting ligands position. The results showed that the triazolopyridazine moiety of both ligands was implanted into the binding pocket of the Brd4. Additionally, the results from LOGSY titration showed that methyl-group 1 of ligand 1, aromatic proton 8 of ligand 2 and aromatic proton 8 of ligand 1 exhibit strong water NOE. This information enabled researchers to utilize a chemical replacement strategy (substitute bound water molecules by suitable functional groups) for aromatic proton 8 in a series of ligands containing the triazolopyridazine ring. Those protons were replaced with an amino or aminomethyl groups and as a result, the binding affinity of those ligands increased 100-fold. Finally, the results obtained from X-ray crystallography for ligands with such modifications allowed to find the binding mode of the triazolopyridazine ring of ligand 1 (with methyl group pointing internally) and the substituted amino group was found to create hydrogen bond to the side chain of Asn140 of Brd4-BD1 [402].

3.6.7. Nuclear Magnetic Resonance Molecular Replacement (NMR2)
The most recent approach called Nuclear Magnetic Resonance Molecular Replacement (NMR2) utilizes spatial data obtained through solution-state NMR in order to locate the binding pocket of a complex structure. For that, it uses a receptor model, e.g., a X-ray structure of a homolog, to conduct an analysis and at the same time excluding the need for protein resonance assignment. To conduct an experiment using such an approach requires a few steps. First, either the protein or ligand used in the complex must be uniformly 13C and 15N labeled. Then, an experiment to assign the ligand is needed such as 2D 13C,1H-HMQC or 13C,1H-HMBC. The next step is the evaluation of ligand intra- and ligand–protein intermolecular distances through NOE cross peaks obtained from F1-15N,13C-filtered 1H,1H-NOESY. Lastly, choosing a proper input structure is required which can be either X-ray or NMR structures in apo form, with another bound ligand, or a homolog to the protein of interest. Then the NMR2 program analyzes for all possible partial assignments (such as methyl groups of a protein) and calculates the complex structures for all options [403,404]. This method was already successfully used to resolve complex structures in case of slow and fast exchange ligands [403,404,405,406].

3.6.8. HECSP
In silico methods combined with NMR derived information can also be used to determine accurate drug-target complexes. 1H empirical chemical shift perturbation (HECSP) is an empirical model that is based on chemical shift perturbation (CSP) of a protein. CSP represents the change in chemical shifts in a protein due to alteration of its chemical environment (which can happen upon ligand binding). The CSP of a target protein is obtained by a series of 2D HSQC experiments with a set of ligand titrations involving samples that contain 15N-labelled protein. The calculation of 1H-CSPs inside the protein are based on four contributors: 1) ring current, 2) electric field, 3) hydrogen bonding, and last 4) magnetic anisotropy. To show the value of the HECSP model two CSP examples were used: apo-neocarzinostatin (apoNCS)-naphthoate ester complex, and human intestinal fatty acid binding protein (hIFABP)-ketorolac-ANS complex. The results from the experiment showed that HECSP model can distinguish native ligand from decoys and more clearly define protein-ligand complex structures with NMR derived information [407].

3.6.9. SAMPLEX
Another program that can utilize CSP called Smoothed Automatic Mapping of Protein from Listed Extremes (SAMPLEX) can help to determine the interaction surface of proteins complexes. SAMPLEX takes the chemical shifts of the protein of interests in both the free and bound state and corresponding 3D structure of a protein in the free state. The programs returns a confidence value for each residue to be in a perturbed or unperturbed state (0.05 as being in a perturbed state, −0.05 as remaining in their unperturbed state). This approach was tested on five examples, one of which was Subtilisin BPN’ (serine protease) complexed with its inhibitor–chymotrypsin inhibitor 2. The results showed that residue 2, and residues 56–62 of chymotrypsin inhibitor-2 were perturbed and residue 63 was in an ambiguous state. To compare, the X-ray crystallography data showed residues 50 and 54–61 to be involved in the interaction. For subtilisin BPN’ the program predicted residues 33, 97, 99–109, 126-128, 141, 154–156, 167–171 and 218–219 to perturbed and residues 65, 98 and 220 to be in ambiguous state. That information was also confronted with the X-ray crystallography data which shown residues 99–104, 125–128, 154–157, 167, 218–221 to be perturbed [408].