A Fragment-Based Ligand Screen Against Part of a Large Protein Machine

Abstract

The ubiquitous AAA+ ATPase p97 functions as a dynamic molecular machine driving several cellular processes. It is essential in regulating protein homeostasis, and it represents a potential drug target for cancer, particularly when there is a greater reliance on the endoplasmic reticulum–associated protein degradation pathway and ubiquitin–proteasome pathway to degrade an overabundance of secreted proteins. Here, we report a case study for using fragment-based ligand design approaches against this large and dynamic hexamer, which has multiple potential binding sites for small molecules. A screen of a fragment library was conducted by surface plasmon resonance (SPR) and followed up by nuclear magnetic resonance (NMR), two complementary biophysical techniques. Virtual screening was also carried out to examine possible binding sites for the experimental hits and evaluate the potential utility of fragment docking for this target. Out of this effort, 13 fragments were discovered that showed reversible binding with affinities between 140 µM and 1 mM, binding stoichiometries of 1:1 or 2:1, and good ligand efficiencies. Structural data for fragment–protein interactions were obtained with residue-specific [U-²H] ¹³CH₃-methyl-labeling NMR strategies, and these data were compared to poses from docking. The combination of virtual screening, SPR, and NMR enabled us to find and validate a number of interesting fragment hits and allowed us to gain an understanding of the structural nature of fragment binding.

Keywords

fragment-based drug design AAA+ ATPase p97 valosin-containing protein (VCP)

Introduction

p97 has an essential role in regulating protein homeostasis;^1,2 however, because it acts on only a subset of protein degradation pathways, it is a potential drug target for new cancer therapies.³ Functional p97 is hexameric, with each protomer consisting of three major domains that could be targeted to inhibit overall function. These domains include two highly conserved Walker A and B motif-containing ATPase domains (D1 and D2) and an N-terminal domain that interfaces with the D1 domain. The D2 domain exhibits strong ATPase activity in vitro^4,5 and has been a primary target for high-throughput screening (HTS) for p97 inhibition. Both reversible (DBeQ, ML240, and ML241) and irreversible (3,4-methylenedioxy-β-nitrostyrene Syk inhibitor III and NMS-859) inhibitors have been identified as potent, adenosine triphosphate (ATP)-competitive inhibitors of p97 ATPase activity in vitro and inhibitors of p97-dependent degradation in cells.^7–10 The activity of the D1 domain has been controversial, with recent work indicating that it is an active ATPase that can be inhibited differentially from D2.⁵ The N-domain functions as a protein–protein interaction site for binding adaptor proteins that direct p97 to particular cellular pathways.¹ There are therefore multiple potential binding sites for small molecules.

The N, D1, and D2 domains are highly dynamic and intimately connected,^11–14 making allosteric inhibition possible. The large size of the enzyme, combined with these intrinsic motions, adds additional challenges for drug discovery. One novel class of allosteric inhibitors (NMS-873) was recently reported to bind an interprotomer site proximal to the D1–D2 linker region.^7,15 Binding of NMS-873 prevents necessary conformational changes within p97 by freezing the D2 domain in an adenosine diphosphate (ADP)-bound conformation. Identifying and targeting additional allosteric sites (e.g., the N–D1 interface, within the D1 domain, or in the D1 interprotomer region) might also prevent p97 activity by locking the enzyme in a particular conformation.

Here, we describe a fragment-based ligand discovery (FBLD) strategy to identify compounds that bind anywhere within the N and D1 ATPase domains, against which (to our knowledge) no inhibitors have yet been identified. Compared to HTS, FBLD uses smaller compounds (10–20 heavy atoms) that can more efficiently sample larger regions of chemical space.¹⁶ Although, in general, fragments bind relatively weakly, they can achieve high ligand efficiency (defined as ΔG/heavy atom). In addition, fragments are more likely to explore different subsites within a given binding pocket, allowing for subsequent fragment linking and merging strategies to yield new chemotypes.¹⁷ FBLD approaches rely on sensitive biophysical techniques, including ligand-detected nuclear magnetic resonance (NMR) experiments, such as saturation transfer difference (STD)¹⁸ and 2D heteronuclear correlation–based chemical shift perturbation NMR methods that can provide specific structural data about ligand binding.¹⁹ Surface plasmon resonance (SPR) is another sensitive, high-throughput technique for identifying and quantifying binding parameters for fragment hits.²⁰ Finally, X-ray crystallography is commonly used to describe in detail the fragment pose and binding pocket; however, obtaining a crystal of ligand-bound p97 is capricious, and co-structures have yet to be reported.

In this work, we report a case study for using fragment-based ligand design approaches against the large and dynamic ND1 hexamer ( Fig. 1 ). A screen of a fragment library was conducted by SPR and followed up by NMR, two complementary biophysical techniques. We also ran a prospective virtual screen and compared the docking scores of fragments for the different sites we probed, and we evaluated the scores of the hits identified by SPR and NMR targeting these sites. This screening effort yielded 13 validated fragments that bind to either one or two sites in ND1–p97. We then used competition assays and isotopic methyl-labeling NMR strategies to obtain structural data for three fragments that bind the nucleotide site. Together, our results suggest that ND1 has several potential binding sites. Targeting different sites could elucidate the allosteric mechanisms by which p97 transmits information about adaptor binding and ATP turnover, and could lead to the development of anticancer therapies. The ND1 ligands identified here provide starting points to advance toward lead compounds through chemistry based on either their crystal structures or, in their absence, the testing of analogs (“SAR by catalog”) by SPR and NMR.

Figure 1.

Roadmap of the fragment-based drug discovery approach applied to the p97 ND1 domains.

Materials and Methods

Fragment Libraries

The UCSF Small Molecule Discovery Center (SMDC) fragment library includes 2485 compounds (Life Chemicals, Munich, Germany; Maybridge, Loughborough, UK; and Asinex, Winston-Salem, NC) with median values for molecular weight of 207 Daltons, heavy atoms n = 15, and a partition coefficient (ALogP) of 1.5. On average, the number of rotatable bonds is two, and the number of hydrogen bond donors and acceptors averages one and two, respectively. All fragments were included in the virtual screen.

Sixteen of the 20 confirmed SPR hits were available for repurchase. The remaining four compounds (11, 14, 15, and 19) were reconstituted as dry powder from the NMR screening stocks. To confirm compound identities and estimate purity, the molecular weights of the 20 compounds were determined by liquid chromatography–mass spectrometry (a Waters 2790 separation module and a Waters micromass ZQ single quadrupole mass spectrometer equipped with a Waters 2996 photodiode array detector, a Waters 2424 ELS detector, and an XBridge C18 column). All 20 compounds exhibited the expected molecular weight; estimated purity is listed in Supplementary Table S5.

Virtual Screening

Three-dimensional coordinates for each fragment were generated by the LigPrep module (Schrödinger Release 2013-1: LigPrep, version 2.6; Schrödinger, New York, NY) in the Schrödinger Maestro software suite (Schrödinger Release 2013-1: Maestro, version 9.4; Schrödinger) using the OPLS 2005 force field.²¹ Ionization states were generated at pH 7.0 ± 0.5 pH units, and tautomers were generated using Epik. Specified chiralities were retained, and one low-energy ring conformation was allowed. At most, four tautomers per ligand were generated. A total of 3115 structures was generated.

The coordinate file for p97 ND1 domains (PDB: 1E32) was prepared in the Maestro Protein Preparation “Wizard” to remove waters, assign bond orders, and add hydrogens. Orientations of side chain amide and hydroxyl groups in the protein, and the protonation state and tautomers of histidine, were also optimized. A receptor grid box of 30 × 30 × 30 Å with an inner box of 10 × 10 × 10 Å was generated for each of four binding sites as follows: the nucleotide-binding site was centered on ADP; a triangle site centered on the centroids of residues Asn199, His384, and Met158; the arginine fissure site centered on coordinates (48.52, 22.54, 21.78); and the N-domain site centered on (17.39, −8.28, 15.37).

Fragment docking was completed using the Glide eXtra Precision (XP) protocol (Suite 2012: Glide, version 5.8; Schrödinger)²² with flexible ligand sampling, nitrogen inversions, ring conformation sampling, penalizing nonplanar amide torsions, Epik state penalties, and no enhancement of planarity for conjugated π groups. Van der Waals radii of ligand atoms were scaled to 0.8 (default), and a partial charge cutoff of 0.15 was used. Postdocking minimization was performed using 10 poses per ligand, with a convergence threshold of 0.5 kcal/mol. All other settings were set to defaults. Strain correction terms were not applied. One pose per input ligand conformation was exported. Structure visualization was carried out with PyMol (version 1.5.0.4; Schrödinger).

Molecular Cloning, Protein Overexpression, and Purification

Human p97 constructs, including ND1 (residues 1–458) and N-terminal AviTagged ND1, were cloned as described.⁶ A C-terminal AviTagged construct of the N-domain (residues 1–213) was also cloned into the modified N-terminal 6xHisTag pET15b vector.⁶ All constructs were verified by Sanger sequencing.

To obtain site-specific biotinylated proteins for SPR, AviTag–p97 constructs were expressed and purified using a combination of affinity and size exclusion chromatography, as previously described.⁶

For methyl-TROSY (transverse relaxation optimized spectroscopy) HMQC (heteronuclear multiple quantum coherence) NMR experiments, [U-²H; Alaβ-¹³CH₃; Ileδ₁-¹³CH₃; Leuδ-¹³CH₃, Valγ-¹³CH₃]-labeled p97 ND1 (¹³CH₃-ILVMA p97 ND1) was expressed in Rosetta 2(DE3) cells (EMD Millipore, Billerica, MA). An initial culture grown in Luria broth (LB) was used to inoculate M9 media (100% H₂O) and grown at 37 °C to an OD₆₀₀ of 0.5. Cells were pelleted by centrifugation, resuspended in 100% D₂O M9 media, and cultured at 37 °C until the OD₆₀₀ reached 0.5. At this point, isotopically ¹³C-labeled Ile, Leu, Val, Ala, and Met precursors (50 mg 2-ketobutryic-4-¹³C,3,3-d₂ acid; 100 mg 2-keto-3-methyl-¹³C-butyric-4-¹³C,3-d acid; 250 mg L-methionine-(methyl-¹³C); and 100 mg L-alanine-3-¹³C,2-d) were added. After 30 min, expression was induced with 1 mM IPTG followed by overnight culture at 20 °C. ¹³CH₃-ILVMA p97 ND1 was purified as described for unlabeled AviTagged ND1, with the exception that phosphate buffered saline (PBS), 0.5 mM Tris (2-carboxyethyl) phosphine (TCEP) was used during the size exclusion chromatography step.

Surface Plasmon Resonance

SPR biosensor experiments were performed using a Biacore 4000 instrument (GE Healthcare, Little Chalfont, UK). CM5 sensor chips were prepared essentially as described,⁶ with the exception that proteins were immobilized to high levels (5000–14,000 resonance units) in buffer [10 mM HEPES (pH 7.5), 150 mM NaCl, 0.005% Tween 20] by injecting 100–400 µg/mL protein for 1–10 min at 25 °C.

For the primary screen, fragment stocks (5 mM in DMSO, in 384-well plates) were diluted 20-fold with buffer [25 mM Tris (pH 7.5), 150 mM NaCl, 20 mM MgCl₂, 1 mM DTT, and 0.005% Tween 20] resulting in a final concentration of 250 µM in 5% (v/v) DMSO buffer. Compounds were screened at 20 °C using a flow rate of 30 µL/min, and association and dissociation times of 60 s. ADP (20 µM) and buffer were injected every 10 cycles as positive and negative controls, respectively. A solvent correction curve was completed at the beginning and end of the run to account for buffer and DMSO mismatches. Data were double-referenced, scaled, and normalized as described.²⁰ Hits were defined as having a response unit (RU) greater than 2.5 standard deviations over the baseline, and desirable sensorgrams characterized as a square-shaped trace with fast on- and off-rates.

Fragment hits were confirmed in dose–response using a concentration series ranging from 19.5 µM to 1.25 mM and 2-fold dilutions. Binding of ADP (0.027 – 20 µM; 3-fold dilutions) and/or adenosine monophosphate (AMP; 12.5–400 µM; 2-fold dilutions) was measured at the beginning and end of each run to assess the quality of the protein surface. Raw sensorgrams were reduced, solvent-corrected, and double-referenced using the Biacore 4000 software, and fragments were classified as either well-behaved compounds or promiscuous binders.²³ The repurchased fragments and four compounds from the screening stocks were reassayed in duplicate (12.5–800 µM; 2-fold dilutions) for binding to the N and ND1 p97 constructs to confirm binding specificity and to obtain binding affinities. Sensorgrams were imported into Scrubber2 (Biologic Software, Campbell, Australia) for fitting. The stoichiometry of ADP binding to ND1 monomer was assumed to be 1:1; the experimental ratio (typically, 0.8:1) was used to define the fraction of active ND1 protein. This active fraction was then used in stoichiometry calculations. Responses were normalized for molecular weight (MW) using Eq. 1, and were fit to a 1:1 or 2:1 binding model (ND1), or a 1:1 binding model using a maximum response equivalent to the theoretical maximum response (TR_max; Eq. 2) for the N-domain.

{RU}_{normal} = {RU/MW}_{fragment} \times 1000

{TR}_{\max} = ({MW}_{fragment} / {MW}_{protein}) {RU}_{protein immobilized}

To determine if any of the repurchased fragments bound in the nucleotide-binding site, the Biacore 4000 was used to simultaneously measure the binding of the repurchased fragments (250 µM) to ND1 in the standard SPR screening buffer and in an SPR screening buffer containing 100 µM ADP. AMP (250 µM) was run as a positive control. Fragments that bound competitively in the nucleotide-binding site were expected to show a reduction in binding in the ADP buffer.

NMR Spectroscopy

STD NMR spectroscopy

All 1D STD NMR spectra¹⁸ were acquired on a Bruker AVANCE DRX500 MHz spectrometer at 296.8 K (referenced to 4% CH₃OH–CH₃OD using a coefficient of 1.0183) with a 5 mm Bruker QCI Cryoprobe (¹H, ¹³C, ¹⁵N, and ³¹P) with actively shielded Z-gradients and a BACS-60 automated sample changer (Bruker Biospin, Billerica, MA). Samples (500 µL) containing 500 µM fragment (or less if the compound had limited solubility) and 50 µM unlabeled ND1 protein were prepared in Norell XR-55 glass NMR tubes (Norell, Marion, NC) in 100% D₂O PBS (pH 7.5), 1 mM NaN₃, and 1 mM DTT, 0.5% d₆-DMSO (Norell, Marion, NC). 10 µM disuccinimidyl suberate (DSS) was added as an internal ¹H chemical shift reference, and compound reference spectra were acquired prior to addition of protein. The “stddiffgp19.3” pulse program from the Bruker pulseprogram library (TopSpin 1.3 pl10) was used for data acquisition with an excitation-sculpting double gradient echo-based 3-9-19 composite pulse WATERGATE-5 element for water suppression. IconNMR and a BACS-60 sample changer were used to automate data collection, and Topspin 3.1 (Bruker Biospin) was used for data analysis. Further details of experimental parameters can be found in the Supplementary Material.

STD NMR ADP competition experiments

The NMR STD experiment (described above) was used to measure the binding of ligand (500 µM) to ND1 (50 µM) in the presence and absence of ADP (250 µM) in the standard NMR buffer (see above). AMP was used as a positive control.

Methyl-TROSY HMQC NMR spectroscopy

NMR experiments were performed on an 800 MHz Bruker AVANCE-I spectrometer with a TXI Cryoprobe equipped with an actively shielded Z-gradient at 298.2 K. Samples were buffer-exchanged into 100% D₂O PBS buffer by repeated ultrafiltration, concentrated to 200 µM (in 200 µL volume), and placed in a Shigemi advanced NMR microtube (part no. BMS-005TB; Shigemi, Allison Park, PA). Two-dimensional [¹³C, ¹H]-HMQC methyl correlation experiments²⁴ on ¹³CH₃-ILVMA ND1 were acquired with 86* and 768* complex points in the ¹³C and ¹H dimensions, respectively, with maximum acquisition times of 20 (t₁) and 53 ms (t₁), respectively. Presaturation was introduced into the experiment for water suppression. A relaxation delay of 1.5 s was used, along with 128 scans, giving rise to an acquisition time of 10 h. All 2D HMQC-TROSY spectra were processed with NMRpipe, converted to Azara format (pipe2azara), and imported into CCPN Analysis2 for data analysis. A 1% (v/v) d₆-DMSO control (no fragment) did not recapitulate the observed shifts or broadening patterns seen for nucleotide or fragment binding (data not shown).

Results

Virtual Screening

We began the exploration of p97 with a de novo search of the topology of the ND1 monomer by using the Sitemap²⁵ software to identify potential small molecule–binding sites ( Fig. 2A and 2B ; PDB: 1E32). This predicted five potential small molecule–binding sites (Sitemap scores, volumes, boundaries, and locations for these sites are listed in Suppl. Table S1). The nucleotide-binding site in the D1 domain received a sitescore of 0.97 (a sitescore of >0.8 is considered to be predictive of druggable sites) and has a volume of 283 Å³ ( Fig. 2C ). This site consists of a hydrophobic binding pocket with a small pore extending through the protein, and a region of positive polarity, extending away from the hydrophobic cavity of the site, to accommodate the phosphate groups of the ATP/ADP nucleotides. Sitemap identified a second site with a sitescore of 0.98 located in the highly charged interface between the N- and D1 domains. This very large interdomain site, with a noncontiguous volume of 907 Å³, was termed the arginine fissure site and represents a potential binding site for allosteric inhibitors ( Fig. 2D ). The three remaining sites have smaller volumes (<160 Å³; Suppl. Table S1), and they were discarded after visual inspection because they were either inaccessible to solvent or extremely small and highly polar. Consequently, only two sites identified by Sitemap, the nucleotide-binding and arginine fissure sites, were used for subsequent docking.

Figure 2.

Locations and docking results for the four binding sites targeted in the ND1 domains of p97. (A) Surface representation of the overall p97 hexamer with a single monomer shown in cartoon representation. The N, D1 and D2 domains are colored red, green, and blue, respectively. (B) Cartoon representation of a monomer of p97 showing the locations of the nucleotide and triangle sites (labeled C), the arginine fissure (labeled D), and the N-domain binding site (labeled E). (C) Close-up view of the ND1 nucleotide-binding site with adenosine diphosphate (ADP) bound, and the adjacent triangle docking site (right and above the bound ADP). (D) The arginine fissure interface-docking site between the N and D1 domains. (E) The N-domain ligand-binding site. All binding sites are represented as electrostatic surfaces with positive and negative potentials colored blue and red, respectively. (F) Distribution of Glide eXtra Precision scores by docking site.

Two more sites that were not identified by the computational search were also used for docking. A third site that might be targeted by fragments, located adjacent to the nucleotide-binding site, was identified by visual inspection ( Fig. 2C ). We named this site the triangle site because of its shape; it is bordered by three residues: Asn199, His384, and Met158. A cleft in the N-domain, also identified by visual inspection, was named the N-domain site ( Fig. 2E ) and selected as a fourth site to target in docking experiments.

The 2485-member SMDC fragment library (see Materials and Methods for a description) was docked using Glide XP against the four sites ( Fig. 2C–E ). Figure 2F shows the distribution of Glide XP docking scores by binding site. An ADP control docked to the nucleotide site with a Glide XP score of −10.3. The distribution of fragment scores at the nucleotide-binding site was skewed toward more favorable scores (better than −6) compared to the other three sites, with 211 fragments (6.7% of the library) having scores for the nucleotide-binding site in the −6 to −9 range. Only 31 (1%) and 51 (1.6%) fragments docked to the arginine fissure and triangle sites with scores better than −6, respectively. The N-domain site had the fewest docked poses with scores of −6 or better (10 fragments; 0.3%). We concluded based on these results that the nucleotide-binding site was the most likely, out of the sites we identified, to bind fragments. At this site, 4 of the top 10 ranked poses had a benzimidazole or indole moiety, consistent with the adenine-binding function of this site.

SPR Primary Screen

The SMDC fragment library was screened by SPR for binding to the ND1 domains of p97. The screen was completed in two experimental runs in which N-terminally AviTagged ND1 was immobilized on a neutravin-coated CM5 sensor chip to ~14,000 and ~7000 RUs, respectively. Fragments were injected at a single concentration of 250 µM, with positive (ADP; 20 µM) and negative (buffer) controls included between each group of 10 fragments. The data were processed as described in the Materials and Methods section, and the results are shown as a scatterplot in Figure 3A . The average and standard deviation of the ADP positive controls and baseline were 98.9 ± 2.6 (n = 286) and 0.17 ± 8.2 (n = 2257), respectively, giving a Z′ value of 0.67 for the screen. Because ND1 was immobilized to a lower level in the second half of the screen, the binding response of the fragments was lower, resulting in more noise in the ADP positive control and baseline. Therefore, hits were defined as being greater than 2.5 σ above the mean assay response for each half of the screen, which was equal to a fractional occupancy of 16.7% and 23.4%, respectively ( Fig. 3A ). The sensorgram shapes for fragments with binding levels higher than this threshold were inspected individually (see Materials and Methods) to eliminate obvious promiscuous binders, resulting in 228 (9.2%) active fragments being selected for follow-up by both STD NMR and SPR dose–response analysis.

Figure 3.

Surface plasmon resonance (SPR) and saturation transfer difference (STD) nuclear magnetic resonance (NMR) primary fragment screens against ND1. (A) Scatterplot of the SPR primary screen with fractional occupancy (y-axis) for each fragment (x-axis) represented as a gray square. The adenosine diphosphate (ADP) positive (black-filled circles) and buffer negative controls (black triangles) correspond to 100% and 0% occupancy, respectively. The dashed line indicates the threshold (>2.5 σ) for selecting SPR-active fragments in the screen. Fragments that subsequently reconfirmed as hits in dose–response studies are circled. (B) Scatterplot of STD NMR values and SPR fractional occupancy for the 215 SPR-active fragments that were tested by both methods. Fragments that were confirmed by SPR dose–response analysis are circled.

Confirmation of Active Fragments

STD NMR was used to follow up the 228 SPR-active fragments and identify possible false positives. STD NMR¹⁸ experiments monitor ligand resonances and provide a direct measure of fragment binding to the ND1 domains (an example is shown in Suppl. Fig. S1), free in solution and without the requirement to tether the target protein, which could potentially interfere with conformational dynamics underlying functional allosteric processes. Compounds that form aggregates at the concentrations used in STD NMR can be detected readily because their NMR signals show either line broadening or a significant decrease in intensity (samples contained 10 µM DSS as a reference). Of the 228 primary SPR-active fragments, 215 could be followed up by STD NMR (13 were either no longer available or insoluble in the NMR buffer, which does not contain detergent due to signal overlap). The STD NMR values are shown on a scatterplot together with SPR fractional occupancy (black squares; Fig. 3B ). STD NMR values for three-quarters of the SPR-active fragments (155; 72%) were distributed fairly evenly over a range from >2 to 90%, corresponding to weak (~2% STD; millimolar) to moderate (>50% STD; micromolar) affinity (Suppl. Table S2). STD NMR values larger than 50% were observed for 84 (39%) of the SPR-active fragments. An STD value <2% was measured for 59 fragments, indicating no binding by NMR. This result parallels other work showing that the two methods often identify non-overlapped hits, possibly owing to differences in buffer conditions, the presence of detergents, pH, or selection criteria.²⁶

SPR dose–response analysis

To identify false positives among the 228 SPR-active fragments, these fragments were tested in a dose-dependent manner using concentrations ranging from 19.5 µM to 1.25 mM in 2-fold dilution steps. In some cases, the fragments were visually insoluble in the assay buffer at the highest concentrations, and thus could not be investigated in the dose–response analysis. Twenty of the 228 active fragments (9%) bound with a stoichiometry <3 (stoichiometry is defined here as: ligands/protomer) and were defined as hits (circled in Fig. 3A ). The majority of SPR-active fragments that confirmed as “hits” had fractional occupancies in the range of 20–50% in the primary SPR screen (circled in Fig. 3A ). Three-quarters of the fragments (160; 70%) were categorized according to Giannetti²³ as promiscuous binders because they exhibited concentration-dependent aggregation, nonstoichiometric binding (stoichiometry between 3 and 5), or superstoichiometric binding (stoichiometry >5). Of the remaining fragments that did not reconfirm as hits, 30 showed weak or no binding, 13 were irreversible binders, and 5 bound nonspecifically to the reference surface. This step identified 20 compounds as ND1 hits to repurchase for confirmation.

Post hoc analysis of SPR and STD NMR

With both sets of biophysical data in hand for the 228 SPR-active fragments (SPR dose–response and STD NMR), a post hoc analysis of the data showed that SPR fractional occupancy did not correlate well with STD NMR values ( Fig. 3B ). In contrast, out of the 17 fragments that were sufficiently soluble for NMR investigation from the 20 compounds selected by SPR for repurchase, 14 showed a strong signal by STD NMR (>50%).

SPR Dose–Response Analysis and STD NMR of Repurchased Fragments

We repurchased and retested the 20 compounds identified by SPR as specific ND1 binders (16 repurchased and 4 from the NMR screening stocks). We also measured their binding to the N-domain alone. To monitor the ND1 protein’s binding capacity and stability, dose–response analyses of ADP and AMP controls were performed at the end of the SPR run. Both ADP and AMP controls bound with the expected affinities and kinetics ( Fig. 4A and 4B ). Three-quarters of the 20 retested fragments (15; 75%) reconfirmed as binding to the ND1 domains, with dissociation constants ranging from 150 to 1700 µM ( Table 1 ). Examples of SPR dose–response data for fragments fitted either to a one-site model ( Fig. 4C–E ) or for a stoichiometry of N = 2 ( Fig. 4F–H ) are shown in Figure 4 . Sensorgrams for the remaining fragments are shown in Supplementary Figures S2 and S3.

Figure 4.

Validation of fragment hits by surface plasmon resonance (SPR) and nuclear magnetic resonance (NMR). (A, B) Dose–response analysis for the adenosine diphosphate (ADP; 0.0032–50 µM; in 5-fold dilution steps) and adenosine monophosphate (AMP; 2.74–666 µM; in 3-fold dilution steps) binding controls. (C–H) Sensorgrams for fragment binding (12.5–800 µM; 2-fold dilution steps in duplicate) to the ND1 domains of p97. To obtain binding constants, response data were fit to an equilibrium model using a maximum response equal to two-site (N=2; C–E) or one-site (N=1; F–H) relative to ADP binding. (I) Scatterplot of affinity measured by SPR and saturation transfer difference (STD) NMR values of repurchased fragments. The AMP, ADP, and ATPγS controls (open symbols) are shown along with the 13 repurchased fragments that were confirmed by both assays (filled circles).

Table 1.

20 repurchased fragment hits; ND1 and N domain binding.

					K _D ^c	K _D	LE^d		Nucleotide site		Triangle Site		Arg. Fissure		N-Domain site
ID #	Structure	MW	LogD^a	Type^b	ND1	N	ND1	STD^e	Rank	Score^f	Rank	Score	Rank	Score	Rank	Score
1		239	2.99	Nuc	146 ± 3	890 ± 20	0.32	87 ± 3	96	−6.51	1783	−3.39	3000	−2.17	262	−4.85
2		232	1.79	Nuc	220 ± 10	1300 ± 300	0.29	77 ± 5	55	−6.83	1264	−3.70	844	−3.94	1836	−3.36
3		274	2.12	Nuc	710 ± 60	1960 ± 70	0.26	83 ± 4	326	−5.70	65	−5.91	1923	−3.24	2302	−3.03
4		259	2.80	N	186 ± 7	250 ± 3	0.28	77 ± 2	2492	−2.97	1748	−3.40	549	−4.33	765	−4.13
5		233	1.71	N	216 ± 8	241 ± 4	0.31	87 ± 2	30	−7.11	612	−4.25	997	−3.81	171	−5.04
6		246	3.13	D1	180 ± 20	NB	0.30	85 ± 1	2986	−1.81	2134	−3.19	2309	−2.98	1450	−3.61
7		254	2.36	D1	230 ± 10	NB	0.30	ND	410	−5.53	2002	−3.26	589	−4.25	474	−4.44
8		222	3.61	D1	250 ± 40	840 ± 20	0.28	76 ± 4	1604	−4.21	2672	−3.08	2839	−3.10	1433	−4.22
9		245	1.27	D1	350 ± 40	NB	0.27	75 ± 6	12	−7.63	1479	−3.56	1970	−3.21	994	−3.94
10		199	1.70	D1	380 ± 80	NB	0.31	74 ± 6	1008	−4.68	1804	−3.38	1377	−3.54	1359	−3.66
11		244	3.01	D1	500 ± 20	NB	0.26	78 ± 1	649	−5.15	692	−4.14	613	−4.23	75	−5.35
12		256	2.46	D1	520 ± 90	NB	0.23	50 ± 14	1012	−4.64	3007	−2.33	2292	−2.99	476	−4.43
13		226	2.05	D1	700 ± 200	NB	0.28	45 ± 1	61	−6.77	1441	−3.57	980	−3.82	280	−4.77
14		287	1.69	D1	1600 ± 100	NB	0.19	ND	2003	−3.55	634	−4.21	296	−4.87	1991	−3.26
15		259	1.56	D1	1700 ± 100	NB	0.21	82 ± 1	1960	−3.59	883	−3.97	692	−4.13	672	−4.20
16		258	2.21		SS	SS	—	34.5 ± 0.5	2859	−2.24	1618	−3.48	2794	−2.56	2242	−3.08
17		258	1.31		SS	SS	—	ND	20	−7.25	226	−5.09	2128	−3.10	281	−4.77
18		153	−0.72		NB	NB	—	0	739	−5.03	1864	−3.34	26	−6.10	1808	−3.38
19		256	2.13		NB	NB	—	74 ± 1	1216	−4.18	388	−5.05	1249	−4.04	1864	−3.76
20		254	1.40		NB	NB	—	74 ± 4	657	−5.15	567	−4.31	1241	−3.63	221	−4.93

NB, no binding; ND, not determined owing to lack of solubility; SS, superstoichiometric binding (N > 3).

Log D octanol/water distribution coefficient calculated using Pipeline Pilot.

Classification of fragments as D1 nucleotide-site binding (Nuc), D1 binding (D1), or N-domain binding (N).

K_D in micromolar. Reported errors are the standard error of the mean of the fit (n = 2).

Ligand efficiency, calculated as ΔG/(#heavy atoms) (kcal/mol*atom).

The largest observed percentage reduction in signal at the ligand resonances.

GLIDE XP score.

Although most fragments were found to bind to the ND1 construct with a 1:1 stoichiometry, we identified five fragments that bound to ND1 with a 2:1 stoichiometry ( Table 1 and Suppl. Table S3), suggesting that these fragments may have two distinct binding sites. When tested against the N-domain alone, all five of these fragments, and one additional fragment (8; 1:1 stoichiometry with ND1), bound the N-domain. When fit to a 1:1 stoichiometry, their affinities for the N-domain ranged from 200 µM to 2 mM. Two of these fragments (4 and 5) have similar affinities (in the 200 µM range) for both the ND1 protein and the N-domain alone (Suppl. Table S3), indicating they may bind preferentially to the N-domain ( Table 1 ; type III binders). Indeed, in a separate study, we identified similar fragments that bound in a cleft immediately adjacent to the adaptor-binding site in a 2D ¹⁵N-HSQC-based NMR chemical shift perturbation (CSP) screen targeting solely the N-domain (manuscript in preparation). The remainder (1, 2, 3, and 8) bound more strongly to the ND1 protein than to the N-domain alone (Suppl. Table S3), and thus the higher affinity of the two sites is likely located in the D1 domain ( Table 1 ; D1 binders). ADP competition assays (see below) allowed us to further classify some D1 domain binders as nucleotide-site binders (Nuc).

The 20 repurchased SPR hits were also retested for binding to ND1 by STD NMR ( Table 1 and Suppl. Table S3). Thirteen of these exhibited strong STD NMR values (45–87%; Table 1 , Suppl. Table S3, and Fig. 4I ) and were confirmed as hits by both methods (SPR K_D <2 mM, N ≤2 and a STD value >45%). Two of the 15 repurchased fragment hits that were reconfirmed by SPR (7 and 14) could not be tested by STD NMR due to their low solubility in the NMR buffer. Three more of the repurchased fragments exhibited binding by STD NMR, but either were nonbinders by SPR (19: 73% STD; and 20: 74% STD) or were classified as superstoichiometric by SPR (16: 35% STD). One fragment (18) showed no binding by either method. One characteristic of STD NMR is that it cannot identify compounds that exhibit superstoichiometric binding.

Figure 4I shows the relationship between the affinity measured by SPR and the STD NMR value (at a ligand-to-protein ratio of 10:1). Except for the nucleotides (ADP and ATPγS), which have higher affinities and hence longer resonance times in the binding site, and one fragment with lower affinity (15; 1.7 mM), there is a rough trend correlating greater STD values with lower SPR dissociation constants ( Fig. 4I ). Although the STD NMR value does not predict the K_D, strong STD NMR values were seen for fragments with affinities in the micromolar to millimolar range (other NMR techniques are more suited to investigate compounds with affinities in the sub-micromolar range).¹⁹ We would not have expected to observe even this loose correlation because many factors contribute to the measured STD value,¹⁸ and this is probably due to the relaxation of the ligand protons being dominated by strong dipolar interactions with the protein protons in the bound state (due to the slow tumbling rate of the large ND1 324 kDa hexamer in solution; more details on physical factors contributing to these findings are given in the Supplementary Material).

ADP Competition Assays

To investigate if the 13 validated fragments bound to the D1 nucleotide-binding site, we ran SPR competition assays in the absence and presence of a saturating amount of ADP (100 µM; K_D = 200 nM). Fragments that bind in the nucleotide site should be displaced by the tighter binding ADP molecule, as is seen for the control AMP ( Fig. 5A ). By this measure, only three fragments 1, 2 ( Fig. 5A ), and 3 (data not shown) were found to bind the nucleotide-binding site. For the other 11 compounds (including 6 and 7), SPR traces were similar with and without ADP, indicating that they do not compete with nucleotide binding.

Figure 5.

Mechanisms of fragment hit binding. (A) Surface plasmon resonance (SPR) sensorgrams for binding of adenosine monophosphate (AMP) or fragments in the presence (red trace) and absence (black trace) of 100 µM adenosine diphosphate (ADP). Fragments that bind in the D1 nucleotide-binding site show a reduction in response to the presence of ADP. (B–D) Regions from 2D methyl-TROSY (transverse relaxation optimized spectroscopy) spectra of U-[²H], [¹³CH₃]-ILVMA p97 ND1, in which chemical shift changes and/or line broadening are observed on nucleotide and fragment binding. Panels B–D show parts of the spectra where affected signals from isoleucine (B; δ¹³C = 11.4–12.5 and δ¹H = 0.5–0.85 ppm), alanine (C; δ¹³C = 15.6–18.1 and δ¹H = 0.3–1.0 ppm), or leucine (D; δ¹³C = 25.0–27.0 and δ¹H = −0.3 to 0.3 ppm) appear. U-[²H], [¹³CH₃]-ILVMA p97 ND1 plus either nucleotide or fragment (red) is shown overlaid on the reference spectrum (black). (E) Glide eXtra Precision (XP) docking poses for AMP and four fragment hits to the D1 nucleotide-binding site. Glide XP docking scores are shown in the top left corner.

Six of the 13 fragments were also tested using STD NMR with ADP competition (250 µM ADP). AMP was used as a positive control, and the STD NMR value decreased from 65% to 12% STD on ADP addition. Four compounds tested (4, 8, 9, and 13) showed little change in STD NMR values or the line widths of their NMR signals on ADP addition, indicating that they are not competitive with ADP. Fragments 1 and 2 showed a reduction in line broadening on addition of ADP, which would be consistent with moderate-affinity (~100 µM) binding in the nucleotide site. They did not, however, show significant changes in STD NMR values on ADP addition. These data are compatible with fragments 1 and 2 being displaced by ADP from a micromolar affinity nucleotide-binding site; however, an unchanged STD value may be due to binding to additional low-affinity (mM) sites, as indicated by the SPR stoichiometries (Suppl. Table S3; additional details of physical factors contributing to these results are discussed in the Supplementary Material).

Based on SPR binding, stoichiometry ratios, and ADP competition data obtained with the ND1 and N-domain proteins, we put forward a hypothesis classifying fragments as D1 nucleotide-site binders (Nuc), D1 binders (D1), or N-domain binders (N; Table 1 and Suppl. Table S4). Further structural studies are needed to investigate the specific binding sites of these fragments within the D1- and N-domains.

Fragment Hit Binding Site Localization by Methyl-TROSY NMR

We used 2D [¹³C, ¹H]-methyl-TROSY NMR methods (M-TROSY), which allow NMR studies of large protein complexes, to probe the sites of fragment binding. These experiments, in which precursors of Ile, Leu, Val, Met, and Ala amino acids with a terminal ¹³C-labeled protonated methyl group are incorporated into a highly deuterated background to produce [U-²H; Alaβ-¹³CH₃; Ileδ1-¹³CH₃; Leuδ-¹³CH₃, Valγ-¹³CH₃]-labeled p97 ND1 (¹³CH₃-ILVMA ND1), can achieve excellent resolution and sensitivity for terminal side-chain methyl group signals, even in large protein complexes.²⁷ Previous work had shown well-resolved signals in M-TROSY spectra of the ND1 hexamer labeled with ¹³CH₃-Ile and ¹³CH₃-Ala, and identified chemical shift changes in alanine methyl groups in the N-domain of ND1 due to adaptor binding.²⁸ As the p97 ND1 domains form a symmetrical hexamer in solution, the nuclei in each ILVMA methyl group in individual protomers have identical chemical environments, giving rise to a single NMR signal for each residue in the protomer. This leads to a high-quality M-TROSY spectrum for the ¹³CH₃-ILVMA-labeled p97 ND1 hexamer despite the large size of the complex (324 kDa). Although many of the NMR signals for ¹³CH₃-ILVMA ND1 near the vertical centerline of the spectrum (~0.85 ppm in ¹H) are overlapped, other signals are dispersed (owing to the proximity of methyl groups to aromatic residues), and these can be used as “reporter signals” to follow changes induced by ligand binding.

A complete assignment of the ILVMA ¹³CH₃ terminal groups in the ND1 hexamer is outside the scope of this work. However, we hypothesized that specific NMR signals were associated with terminal methyl groups of particular ILVMA residues near the ND1 nucleotide-binding site based on (1) published Ala assignments;²⁸ (2) statistical chemical shift ranges (¹H, ¹³C) for ILVMA methyl groups; (3) comparison of spectra from ILV-¹³CH₃- and ILVMA-¹³CH₃-labeled proteins (Suppl. Fig. S3); (4) titration with different nucleotides (AMP, ADP, and ATPγS); and (5) correlation of the observed spectral changes with contacts made by different bound nucleotides in crystal structures. We proposed that the isoleucine, alanine, and leucine methyl reporter signals (IleA, IleB, AlaC, AlaD, AlaE, UnkF, LeuG, and LeuF) affected by nucleotide binding ( Fig. 5 ) corresponded to Ile380, Ile383, and Leu253 and alanine residues close to the nucleotide-binding site.

Using the set of reporter signals, we sought to locate the binding site(s) for several fragments with <250 µM affinity (1, 2, 6, and 7). Binding of these fragments to the ND1 domains was accompanied by changes in M-TROSY spectra ( Fig. 5 ) that resembled AMP or ADP binding. For two fragments (1 and 2), the set of signals affected was very similar to those for the nucleotides. In particular, the signals for LeuG and LeuH, which in our hypothesis we associated to the CH₃δ_1,2 groups of Leu253, and both IleA and IleB (associated with the CH₃δ₁ groups of Ile380 and Ile383) showed significant broadening, consistent with them having a close proximity to the bound fragments. In contrast, the binding of two other fragments (6 and 7) affected only a subset of the reporter signals ( Fig. 5 ), indicating that they bound differently to 1 and 2. A more detailed comparison of the spectra from nucleotides and fragments, and explanation of the M-TROSY NMR results, can be found in the Supplementary Material.

Comparison of the Virtual Screen and the Experimentally Validated Hits

Virtual screening performed prior to the biophysical experiments suggested that the nucleotide-binding site was the most likely, out of four potential sites that were examined, to bind fragments. However, we could only confirm, by ADP competition experiments and TROSY NMR, binding at this site for 3 of the 13 validated hit fragments. Two of these 3 fragments were among the top 100 docking hits (1: Glide XP score −6.5, rank 96; and 2: Glide XP score −6.8, rank 55), and the third received a reasonable score (3: Glide XP score −5.7, rank 326). The predicted docking poses of 1 and 2 are shown in Figure 5E . They show that the fragments attempt to satisfy polar interactions at the back of the pocket with the Gly207 backbone carbonyl O-atom. The chlorine atom from 1 partially fills the hydrophobic pocket created by Ile383. The remaining top docking hits for the nucleotide-binding site represent false positives; although some of these did show binding in the initial SPR screen, they were later eliminated due to nonstoichiometric binding.

The locations of the binding sites for the 10 other confirmed hit fragments remain unknown. The docking results at the three other sites that we considered generated very few fragment poses with good docking scores, including those for confirmed hits. Sitemap did not identify any other plausible binding sites in either the D1 or N domain. One possible explanation is that conformational changes, possibly associated with functional motions, could open other binding pockets that are not apparent in the available crystal structures.

Discussion

p97 is a large and dynamic protein machine. Data from electron microscopy (EM)²⁹ suggest large D1–D2 interdomain movements during the catalytic cycle; X-ray crystallography has also indicated N–D1 interdomain conformations. This dynamic behavior has significant consequences for developing approaches to inhibiting the enzyme. In addition to the conformational complexity, the large size of the protein and the rugged features of the surfaces of the three domains could combine to present a large number of potential sites that might bind small fragments and potentially be targeted.

A combination of SPR and NMR identified 13 confirmed ND1-binding fragments from a library of ~2500 compounds with affinities <1 mM (seven with affinities <250 µM), stoichiometric binding (N ≤2), NMR STD values >45%, and good ligand efficiencies (~0.3). Eleven fragments likely bind the D1 domain, and two the N-domain. Sitemap calculations and visual inspection were used to identify four possible binding sites for small molecules in the p97 protomer, but our studies were able to confirm binding at only two of these, where three fragments were found to bind in the nucleotide-binding site and two likely target the N-domain site. It is possible that additional sites in the ND1 domains remain undiscovered. Indeed, the identification of cryptic allosteric sites by computational methods is a challenging task and is the subject of active development.³⁰

As a screening and validation tool, STD NMR proved to be an effective confirmatory method, and, with a judicious choice of a cutoff level, it has the potential to exclude fragments that bind nonspecifically or with high stoichiometries. SPR hits with a strong STD value (>45% for this study) were much more likely to show stoichiometric binding in SPR dose–response studies (the assessment of “strong” STD value generally scales with protein size and shape). In this study, tighter-binding fragments generally gave larger STD values for most of the specific, reversible binders as long as the affinity was higher than the single-digit micromolar range. We would generally not expect the STD value to correlate with ligand affinity because many variables affect the value. STD NMR can also identify weaker binders, and these may not be identified as positives by our SPR assay.

Recent advances in NMR have enabled investigations of large protein complexes.²⁴ By applying high levels of deuteration together with ¹³CH₃-ILVMA residue type-specific labeling and M-TROSY, we could probe fragment binding to the 320 kDa p97 ND1 hexamer. In general, the M-TROSY results correlated well with computational and ADP-competition experiments, allowing us to develop a hypothesis of how two fragments could bind in the D1 nucleotide site. In the absence of X-ray crystallography, high-field NMR can provide structural information describing weak binding fragments.

The combination of virtual screening, SPR, and STD NMR enabled us to discover and validate a number of interesting fragment hits. The docking, SPR, and M-TROSY NMR allowed us to gain an understanding of the nature of fragment binding. The structural information, for the fragments that we hypothesize bind in the nucleotide site, will help guide fragment hit-to-lead development and directions for chemical synthesis to improve potency.

Footnotes

Acknowledgements

The content of this publication does not necessarily reflect the views or policies of the U.S. Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. government. We thank John Irwin for help identifying potential docking sites to target.

Supplementary material for this article is available on the Journal of Biomolecular Screening Web site at .

Declaration of Conflicting Interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: MPJ is a consultant to Schrödinger LLC.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project has been funded in whole or in part with federal funds from the National Cancer Institute, National Institutes of Health, under Chemical Biology Consortium Contract No. HHSN261200800001E.

References

Meyer

Bug

Bremer

Emerging Functions of the VCP/p97 AAA-ATPase in the Ubiquitin System. Nat. Cell Biol. 2012, 14, 117–123.

Yamanaka

Sasagawa

Ogura

Recent Advances in p97/VCP/Cdc48 Cellular Functions. Biochimica et Biophysica. Acta. 2012, 1823, 130–137.

Deshaies

R. J.

Proteotoxic Crisis, the Ubiquitin-Proteasome System, and Cancer. BMC Biol. 2014, 12, 94.

Song

Wang

C. C. H.

. ATPase Activity of p97-Valosin-Containing Protein (VCP): D2 Mediates the Major Enzyme Activity, and D1 Contributes to the Heat-Induced Activity. J. Biol. Chem. 2003, 278, 3648–3655.

Briggs

L. C.

Baldwin

G. S.

Miyata

. Analysis of Nucleotide Binding to P97 Reveals the Properties of a Tandem AAA Hexameric ATPase. J. Biol. Chem. 2008, 283, 13745–13752.

Chou

T. F.

Bulfer

S. L.

Weihl

C. C.

. Specific Inhibition of p97/VCP ATPase and Kinetic Analysis Demonstrate Interaction between D1 and D2 ATPase Domains. J. Mol. Biol. 2014, 28, 2886–2899.

Magnaghi

D’Alessio

Valsasina

. Covalent and Allosteric Inhibitors of the ATPase VCP/p97 Induce Cancer Cell Death. Nat. Chem. Biol. 2013, 9, 548–556.

Chou

T. F.

Frankowski

K. J.

. Structure-Activity Relationship Study Reveals ML240 and ML241 as Potent and Selective Inhibitors of p97 ATPase. Chem. Med. Chem. 2013, 8, 297–312.

Chou

T. F.

Brown

S. J.

Minond

. Reversible Inhibitor of p97, DBeQ, Impairs Both Ubiquitin-Dependent and Autophagic Protein Clearance Pathways. PNAS. 2011, 108, 4834–4839.

10.

Chou

T. F.

Deshaies

R. J.

Quantitative Cell-Based Protein Degradation Assays to Identify and Classify Drugs That Target the Ubiquitin-Proteasome System. J. Biol. Chem. 2011, 286, 16546–16554.

11.

Niwa

Ewens

C. A.

Tsang

. The Role of the N-Domain in the ATPase Activity of the Mammalian AAA ATPase p97/VCP. J. Biol. Chem. 2012, 287, 8561–8570.

12.

Chia

W. S.

Chia

D. X.

Rao

. ATP Binding to p97/VCP D1 Domain Regulates Selective Recruitment of Adaptors to Its Proximal N-Domain. PLoS ONE. 2012, 7, e50490.

13.

Huang

Zhao

. Interprotomer Motion-Transmission Mechanism for the Hexameric AAA ATPase p97. PNAS. 2012, 109, 3737–3741.

14.

Huang

Lennarz

W. J.

Dynamic Flexibility of the ATPase p97 Is Important for Its Interprotomer Motion Transmission. PNAS. 2012, 109, 9792–9797.

15.

Polucci

Magnaghi

Angiolini

. Alkylsulfanyl-1,2,4-Triazoles, a New Class of Allosteric Valosine Containing Protein Inhibitors. Synthesis and Structure-Activity Relationships. J. Med. Chem. 2013, 56, 437–450.

16.

Huang

Jacobson

M. P.

Binding-Site Assessment by Virtual Fragment Screening. PLoS ONE. 2012, 5, e10109.

17.

Lanter

Zhang

Sui

Medicinal Chemistry Inspired Fragment-Based Drug Discovery. Methods Enzymol. 2011, 493, 421–445.

18.

Mayer

Characterization of Ligand Binding by Saturation Transfer Difference NMR Spectroscopy. Angewandte. Chemie-Int. Ed. 1999, 38, 1784–1788.

19.

Pellecchia

Bertini

Cowburn

. Perspectives on NMR in Drug Discovery: A Technique Comes of Age. Nat. Rev. Drug. Discov. 2008, 7, 738–745.

20.

Giannetti

A. M.

From Experimental Design to Validated Hits a Comprehensive Walk-Through of Fragment Lead Identification Using Surface Plasmon Resonance. Methods Enzymol. 2011, 493, 169–218.

21.

Jorgensen

W. L.

Maxwell

D. S.

Tirado-Rives

Development and Testing of the OPLS All-Atom Force Field on Conformational Energetics and Properties of Organic Liquids, J. Am. Chem. Soc. 1996, 118, 11225–11236.

22.

Friesner

R. A.

Murphy

R. B.

Repasky

M. P.

. Extra Precision Glide: Docking and Scoring Incorporating a Model of Hydrophobic Enclosure for Protein-Ligand Complexes. J. Med. Chem. 2006, 49, 6177–6196.

23.

Giannetti

A. M.

Koch

B. D.

Browner

M. F.

Surface Plasmon Resonance Based Assay for the Detection and Characterization of Promiscuous Inhibitors. J. Med. Chem. 2008, 51, 574–580.

24.

Tugarinov

Hwang

P. M.

Ollerenshaw

J. E.

. Cross-Correlated Relaxation Enhanced ¹H-¹³C NMR Spectroscopy of Methyl Groups in Very High Molecular Weight Proteins and Protein Complexes. J. Am. Chem. Soc. 2003, 125, 10420–10428.

25.

Halgren

T. A.

Identifying and Characterizing Binding Sites and Assessing Druggability. J. Chem. Inf. Model. 2009, 49, 377–389.

26.

Wielens

Headey

S. J.

Rhodes

D. I.

. Parallel Screening of Low Molecular Weight Fragment Libraries: Do Differences in Methodology Affect Hit Identification? J. Biomol. Screen. 2013, 18, 147–159.

27.

Goto

N. K.

Kay

L. E.

New Developments in Isotope Labeling Strategies for Protein Solution NMR Spectroscopy. Curr. Opin. Struct. Biol. 2000, 10, 585–592.

28.

Isaacson

R. L.

Simpson

P. J.

Liu

. A New Labeling Method for Methyl Transverse Relaxation-Optimized Spectroscopy NMR Spectra of Alanine Residues. J. Am. Chem. Soc. 2007, 129, 15428–15429.

29.

Yeung

H. O.

Foerster

Bebeacua

. Inter-ring Rotations of AAA+ ATPase p97 Revealed by Electron Cryomicroscopy. Open Biol. 2014, 4, 130142–130142.

30.

McClendon

C. L.

Friedland

Mobley

D. L.

. Quantifying Correlations between Allosteric Sites in Thermodynamic Ensembles. J. Chem. Theory Comput. 2009, 5, 2486–2502.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

2.11 MB

0.00 MB