Liquid-state NMR spectroscopy is a powerful technique to elucidate binding properties of ligands on proteins. Ligands binding in hydrophobic pockets are often in close proximity to methyl groups and binding can lead to subtle displacements of methyl containing side chains to accommodate the ligand. To establish whether pseudocontact shifts can be used to characterize ligand binding and the effects on methyl groups, the N-terminal domain of HSP90 was tagged with caged lan- thanoid NMR probe 5 at three positions and titrated with a ligand. Binding was monitored using the resonances of leucine and valine methyl groups. The pseudocontact shifts (PCS) caused by ytterbium result in enhanced dispersion of the methyl spectrum, allowing more resonances to be observed. The effects of tag attachment on the spectrum and ligand binding are small. Significant changes in PCS were observed upon ligand binding, indicating displacements of several methyl groups. By determining the cross-section of PCS iso-surfaces generated by two or three paramagnetic centers, the new position of a methyl group can be estimated, showing displacements in the range of 1–3 Å for methyl groups in the binding site. The information about such subtle but significant changes may be used to improve docking studies and can find application in fragment-based drug discovery. Keywords Isotope labeling · Pseudocontact shift · Methyl groups · NMR spectroscopy · Paramagnetic tag · Heat shock protein Abbreviations PCS Pseudocontact shift HSQC Heteronuclear single quantum coherence Electronic supplementary material The online version of this article (https ://doi.org/10.1007/s1085 8-018-0190-5) contains TROSY Transverse relaxation optimized supplementary material, which is available to authorized users. spectroscopy CLaNP-5 Caged lanthanoid NMR probe 5 * Marcellus Ubbink ntd-HSP90 N-terminal domain of heat shock protein 90 firstname.lastname@example.org Mathilde Lescanne email@example.com Introduction Puneet Ahuja firstname.lastname@example.org Fragment-based drug discovery (FBDD) has proven to be Anneloes Blok an effective method to develop medicinal drugs (Erlanson email@example.com 2012). FBDD is based on finding very small molecules that Monika Timmer bind to the target with a large average binding energy per firstname.lastname@example.org heavy atom (~ 0.3 kcal/mol/heavy atom). Such fragments Tomas Akerud still have a low affinity and then need to be elaborated to Tomas.Akerud@astrazeneca.com molecules with more negative binding free energies, either Leiden Institute of Chemistry, Leiden University, by linking fragment hits or by growing them (Bohacek Einsteinweg 55, 2333 CC Leiden, The Netherlands et al. 1996; Hajduk and Greer 2007). For the elaboration Structure, Biophysics & Fragment-Based Lead Generation, of hits into lead compounds FBDD depends heavily on Discovery Sciences, IMED Biotech Unit, AstraZeneca, structural analysis of fragment-target complexes by X-ray Gothenburg, Sweden Vol.:(0123456789) 1 3 276 Journal of Biomolecular NMR (2018) 71:275–285 diffraction of crystals or NMR spectroscopy. The former Here, we describe the use of PCS as structural restraints technique is most commonly used but NMR is an alterna- to probe at the same time binding kinetics and structural tive for structure determination and offers complementary changes of the protein ntd-HSP90 upon fragment binding. information. Structure determination of the complex by HSP90 is a target protein against cancer (Nagaraju et al. NMR requires a complete NOE analysis of protein and 2017) and its ATP binding site located in the N-terminal ligand, which is tedious but can be used in cases where domain (ntd) is targeted for inhibition (Li et al. 2012). X-ray crystallography fails (Pellecchia et al. 2008). Other, HSP90 is a molecular chaperone essential to prevent client less demanding methods are based on transferred NOEs, proteins from ubiquitin–proteasome system degradation. paramagnetic relaxation enhancements (PRE) or pseudoc- More than 200 client proteins of HSP90 have been identi- ontact shifts (PCS) to obtain information about the ligand fied, including oncoproteins (Murray et al. 2010). Therefore, bound state while benefitting from the narrow linewidths HSP90 is a cancer-target protein and inhibitors have been of the ligand in the free state (Viegas et al. 2011; Guan found to bind the N-terminal domain and/or the C-terminal et al. 2013; Jahnke et al. 2000, 2001). domain (Den and Lu 2012). Several potent molecules are We aim to investigate the possibilities of PCS to study clinical candidates for cancer treatment through inhibition of ligand–protein interactions. PCS have been used before the ATPase activity and FBDD has been successfully applied to obtain model of ligands bound to proteins (Guan et al. to HSP90, which led to a clinical trial (Murray et al. 2010). 2013; Tu and Gochin 1999; Saio et al. 2011; John et al. We find that ligand binding is only marginally affected by 2006). Before, we demonstrated that ligands that are in attaching the two-armed lanthanoid tag CLaNP-5 (Keizers fast exchange between bound and free state can exhibit et al. 2007, 2008) to ntd-HSP90 at three locations. Methyl transferred PCS caused by lanthanoid tags on the protein, group resonances show extensive chemical shift perturba- which can be used to determine a low-resolution model tions in the binding site, as well as further in the hydropho- of the ligand in the binding site, provided that a structure bic core of the protein. Several significant PCS changes are of the protein is available and under the assumption that observed upon ligand binding, which can be interpreted as ligand binding does not result in backbone conformational movements of the methyl groups of a few Ångström. These changes. Fragments, as well as larger compounds often changes can be translated into structural restraints that may bind in hydrophobic pockets on the protein where they be used in ligand docking studies. do not alter the positions of backbone atoms significantly. Methyl groups are often found in such pockets and, thus, are in direct contact with the ligand. They are prone to Materials and methods experience chemical shift changes due to changes in the chemical environment upon ligand binding, and may also Sample preparation more readily than backbone atoms show conformational changes due to the rotational freedom of side-chains Three double cysteine mutants of the ntd-HSP90 were (Wiesnerl and Sprangers 2015). In this way, they can help designed on the surface of the protein, S50C/D54C, A101C/ to accommodate ligand binding, enabling it to form opti- N105C and T149C/I187C (Lescanne et al. 2017). Ntd- mal interactions. Therefore, we wondered whether such HSP90 does not have any native cysteines. The protein was changes could be detected by using PCS. These shifts are produced labelled with Leu-δ1-δ2/Val-γ1-γ2-[ CH ] and caused by the interaction of the nuclear spin and the spin purified and tagged with CLaNP5 according to a published of unpaired electron(s) in a paramagnetic center. They protocol (Lescanne et al. 2017). CLaNP-5 was synthesized depend on the distance between the spin and the center to as described before (Keizers et al. 2007, 2008). the third power as well as on the orientation of the spin in the frame of the anisotropic component of the magnetic NMR titration susceptibility of the unpaired electron(s) (Otting et al. 2010; Liu et al. 2014). With a probe that is rigid relative Ntd-HSP90 in 50 mM Tris-HCl and 50 mM NaCl buffer, pH to the protein and a proper diamagnetic control, the PCS 7.7, was titrated with 4-(2-Fluorophenyl)-2-pyrimidinamine, can be predicted and measured with very high accuracy 1, (Fig. 1), a 189 Da known ligand of ntd-HSP90 that was and small changes in the location of the spin relative to kindly provided by AstraZeneca (Göteborg, Sweden). the paramagnetic center can result in measurable PCS Titrations were performed with three ntd-HSP90 mutants, changes. Methyl PCS can be observed in sensitive 2D S50C/D54C, A101C/N105C and T149C/I187C tagged with 3+ 3+ NMR spectra, potentially also on large systems by apply- Lu -CLaNP-5 or Yb -CLaNP-5. The concentrations of ing selective labelling in a deuterated background and by S50C/D54C, A101C/N105C T149C/I187C were 20, 103 and using TROSY-based experiments (Tugarinov et al. 2003; 65 µM, respectively, for both diamagnetic and paramagnetic Tugarinov and Kay 2005; Sprangers et al. 2008). forms of the protein. Concentrations (µM) of 1 for titrations 1 3 Journal of Biomolecular NMR (2018) 71:275–285 277 Q and Qa factors Q and Qa factors were used to quantify deviation between experimental and predicted data. Q and Qa were calculated according to Eqs. 1 and 2, respectively. Q is the usual meas- ure for goodness of fit, Qa is, however, less sensitive to bias toward cases in which the predicted value is much larger than the observed one, as compared to the opposite case, in which the predicted value is much smaller than the observed Fig. 1 Structure of 1 used to titrate ntd-HSP90 one (Bashir et al. 2010). In cases of a good fit, Qa ≈ 0.5Q. � � with S50C/D54C, A101C/N105C and T149C/I187C were ped exp [0, 39, 59, 88, 132, 198, 296, 444, 667, 1000], [0, 20, 40, � PCS,i PCS,i Q = (1) � � 121, 364, 1111] and [0, 40, 89, 200, 442, 665, 1008, 1897], � exp respectively. The NMR sample volume was 595 µL for all PCS,i samples, and dilution was neglected, because the biggest 13 1 volume of ligand solution added was < 5 µL. C- H HSQC (Palmer et al. 1991; Kay et al. 1992; Schleucher et al. 1994) � � � pred exp spectra were acquired at each titration point, on a Bruker PCS,i PCS,i Avance III 800 MHz spectrometer, equipped with a cryogen- Qa = (2) � � � 2 � exp � � pred � ically cooled TXI-probe head, operating at 298 K. Spectra � � � � PCS,i PCS,i � � � � were processed with nmrpipe (Delaglio et al. 1995) using the exponential EM apodization function for analysis with TITAN (Waudby et al. 2016). A similar titration was per- Results formed with WT ntd-HSP90 observing the amide groups. 2 2 0.5 Average CSP were calculated as (ΔH + (ΔC/10) ) for Tagging effects 2 2 0.5 methyl groups and (ΔH + (ΔN/6) ) for amides. PCS were measured using H and for calculations the geometrical aver- Ntd-HSP90 was tagged at three sites using the Caged Lan- 3+ age of the proton coordinates in methyl groups were used. thanoid NMR probe #5 (CLaNP-5), containing either Lu 3+ as a diamagnetic control or Yb as a paramagnetic center. The tagging sites, double mutants 50C/54C, 101C/105C and Assignments 149C/187C, have been described before (Lescanne et al. 13 1 2017). Assignments are shown in Fig. S1. Methyl C- H 3+ Methyl groups assignments have been performed before HSQC spectra of WT and CLaNP-5(Lu ) tagged mutants with traditional through-backbone NMR techniques and con- are very similar except for the resonance of a few methyl firmed by PARAssign (Skinner et al. 2013; Lescanne et al. groups very close to the tags (Lescanne et al. 2017), indicat- 2017). PARAssign provided the stereo-specific assignment ing that the tags do not have large effects on the structure of with high reliability for 14 Leu/Val methyl groups (Lescanne the protein. A first comparison of the methyl group spectra et al. 2017), based on PCS generated by CLaNP-5 attached of the paramagnetic and diamagnetic samples illustrates the at two distinct positions, S50C/D54C and A101C/N105C. increased dispersion of the resonances for the paramagnetic samples (Fig. 2), which has been noted before (Sattler and Fesik 1997). For example, a crowded spectral region with Cross‑section refinement a width of 0.5 ppm in the H dimension in the spectrum of the diamagnetic sample, disperses over 1.0 ppm for mutant A home-written python script was used to define possible 50C/54C and up to 3.0 ppm for mutant 101C/105C in the locations of the bound methyl groups locations. PCS iso-sur- spectrum of the paramagnetic sample. Increased dispersion faces were calculated for a grid of 15 × 15 × 15 Å, with 100 is of interest for methyl group HSQC spectra, because the points per dimension and centered on the methyl group posi- resonances are often more crowded than in amide HSQC tion in the crystal structure of free ntd-HSP90 [PDB entry spectra, in particular for Leu and Val methyl hydrogens and 3t0h (Li et al. 2012)]. Cross-sections of iso-surfaces from to a lesser extent methyl carbons. In principle, lanthanoids 3+ 3+ 3+ different tags were defined by finding the positions within with larger paramagnetic effects (Tm, Dy, Tb ) provide the grid that matched the required PCS of all tags within an even more dispersion but also cause considerable paramag- error of 0.02 ppm (0.03 ppm in one case, see below). netic relaxation farther from the metal than in the case of 1 3 278 Journal of Biomolecular NMR (2018) 71:275–285 Fig. 2 Enhancement of spectral dispersion by PCS. An overlay is shown of Leu/Val methyl HSQC spectra of ntd-HSP90101C/105C tagged with 3+ 3+ CLaNP-5 loaded with Lu (black contours) or Yb (red contours). The inset shows a detail and the lines connect equivalent resonances 3+ Yb . Thus, such lanthanoids are more appropriate to gener- are the overlap areas of two or three iso-surfaces from dif- ate dispersion in spectra of bigger proteins. ferent tag locations. A cross-section was calculated such Magnetic susceptibility (Δχ) tensors were refined pre- that its thickness reflects 0.02 ppm uncertainty. Thus, large viously using amide proton PCS (Lescanne et al. 2017). cross-sections (Fig. 4a) indicate a weak PCS gradient, as Methyl group PCS were predicted based on these ten- is observed far from the paramagnetic centers. Thin cross- sors and the ligand-free structure [PDB entry 3t0h (Li sections (Fig. 4b) report on a steep PCS gradient, closer et al. 2012)] and compared to the experimental ones. For to the paramagnetic center or close to where the PCS mutants 50C/54C and 101C/105C most predicted PCS fit changes sign. Cross-sections for the iso-surfaces of pairs the experimental values well (Fig. 3). Mutant 149C/187C of mutants are shown in Fig. 4, panels A and B, for the shows a poorer fit but that is in line with the results for methyl groups from residues Leu70 and Val150. The pre- the amides (Fig. S2), which was attributed to the fact that dicted PCS of these methyl groups match the experimental the tag crosslinks two β-strands and appears to assume PCS within 0.02 ppm. The methyl groups observed in the two conformations (Lescanne et al. 2017). To illustrate crystal structure are located at the intersection of the two how differences between experimental and predicted PCS cross-sections, i.e., at the position that matches the experi- translate into differences in the expected locations of the mental PCS for all three mutants within 0.02 ppm. For nuclei in the protein, cross-sections of experimental PCS some methyl groups the discrepancy between experimental iso-surfaces for the different mutants were calculated PCS and PCS predicted on the basis of the crystal struc- (Fig. 4). The iso-surface identifies all locations around a ture is larger than 0.02 ppm. For instance, the experimen- paramagnetic center with a given PCS. The cross-sections tal PCS of Val136 γ1 and γ2 are − 1.03 and − 1.12 ppm, Fig. 3 Prediction of methyl group PCS. The PCS for the Leu/Val the experimental PCS. No fitting was performed. a For mutant methyl groups of ntd-HSP90 were predicted using the published 50C/54C the Q factor (Eq. 1) is 0.14 (Q = 0.07, Eq. 2) b For mutant amide based Δχ tensor parameters (Lescanne et al. 2017) and the 101C/105C Q = 0.05 (Q =0.025). c For mutant 149C/187C Q = 0.24 structure with PDB entry 3t0h (Li et al. 2012) and plotted against (Q =0.12). The red line represents a perfect correlation 1 3 Journal of Biomolecular NMR (2018) 71:275–285 279 Fig. 4 Cross-sections of iso-surfaces of experimental free PCS, with Leu70 δ1 and δ2, respectively. b Val150 methyl groups. The grey the free structure 3t0h. Large grey and black spheres represent the areas are the γ1 cross-sections from mutant 101C/105C with mutant crystal structure locations of methyl groups, centred on the carbon 149C/187C and from mutant 50C/54C with 149C/187C, the black methyl group with radius of 1 Å, for Valγ1/Leuδ1 and Valγ2/Leuδ2, area is the V150 γ2 cross-section for mutant 50C/54C with mutant respectively. Grey and black spheres represent the experimental PCS 101C/105C. c Val136 methyl groups. In black spheres the PCS iso- cross-sections for Valγ1/Leuδ1 and Valγ2/Leuδ2, respectively for the surfaces cross-section of mutant 50C/54C with mutant 101C/105C free ntd-HSP90. Each cross-section was calculated using 0.02 ppm is shown for methyl γ2. In grey spheres are the two PCS iso-surface error on the PCS, for a cubic grid with sides of 50 points over 5 Å, cross-sections of mutant 50C/54C with 101C/105C and mutant centred on the methyl group of interest. a Leu70 methyl groups. The 101C/105C with mutant 149C/187C. Both crystal structure methyl two grey and black areas are cross-sections from mutant 50C/54C groups are on the edge of the cross-section with mutant 101C/105C and mutant 50C/54C with 149C/187C for respectively, and the predicted values show deviations of For residue Leu107, the predicted PCS correlated poorly 0.19 (− 0.84 ppm) and 0.16 ppm (− 0.96 ppm) for mutant with the experimental ones for mutants 50C/54C and 50C/54C for γ1 and γ2, respectively. Because of the strong 149C/187C. The predicted PCS are clearly distinct for the PCS gradient, these large PCS deviations translate in δ1 and δ2 methyl groups (0.072 and 0.252 ppm for mutant only a small displacement of about 0.6 Å as compared to 50C/54C and 0.68 and 0.94 ppm for mutant 149C/187C), the crystal structure (Fig. 4c). In the case of the mutant whereas the experimental values are very similar (0.221 and 149C/187C, the experimental PCS of Val136 γ1 and γ2 0.194 ppm for mutant 50C/54C and 0.822 and 0.842 ppm for equal 0.27 and 0.27 and the deviations from the predicted mutant 149C/187C). The similarity of the two values sug- values are only 0.03 (0.23 ppm) and 0.05 ppm (0.22 ppm), gests a kind of averaging. The linewidth of the resonances respectively, yet these translate into sizeable displacements in the diamagnetic sample also suggests a form of exchange. of 1.5 Å, due to the weak PCS gradient at this position. To establish whether population of more than one rotamer This remarkable difference is visualized in Fig. S3. Thus, explains the deviating PCS, the populations of each of the these findings demonstrate that deviations larger than the three rotamers were determined. Each rotamer was modelled measurement error of PCS (usually 0.02 ppm or less) can in the structure with Pymol (DeLano 2009) (ignoring steric be caused by very subtle differences in structure, whereas clashes) and the PCS were predicted. The best fit was found small deviations may still reflect a more sizeable mismatch for an exchange between two rotamers populated at 53 and between observed PCS and the structure used for PCS pre- 47%, with the third rotamer not being populated, Fig. 5. The diction. The PCS gradient at the methyl group position can Qa fit quality factor (see Eq. 2) for Leu107 drops from 0.14 be used to evaluate the structural relevance for observed to 0.03 when predicted PCS are calculated as a combina- differences between experimental and predicted PCS. To tion of rotamer 2 (+ 0°) and rotamer 3 (+ 120°), for mutants establish to what degree the uncertainty in the Δχ tensor 50C/54C and 149C/187C. For mutant 101C/105C, rotamer parameters affects the iso-surfaces, the Δχ tensors were 2 fits the data well and admixture of rotamer 3 reduces the also calculated using the methyl PCS as input, rather than fit quality. However, Leu107 is located very close to the tag the amide PCS. These independent data report on the same in this mutant, and near a reported flexible region (Didenko tensor, so differences are a measure for the uncertainty. et al. 2012), so either the methyl group location or the The fitting involves eight parameters that are not com- exchange populations of rotamers could be influenced by pletely uncorrelated, so slightly different solutions can be the CLaNP tag. It is interesting that the PCS seem to provide found. It was found that the difference are indeed small evidence for rotamer exchange and can be used to estimate (Fig. S4). The effect on the calculated iso-surfaces only is populations. Such dynamics is not obvious from the chemi- significant for residues with a steep PCS gradient, Fig. S5. cal shift. Definitive evidence, however, would require further 1 3 280 Journal of Biomolecular NMR (2018) 71:275–285 Fig. 5 a Experimental (black) and predicted PCS for the two methyl cyan sticks. Rotamers 1 and 3 are in green and magenta, respectively. groups of L107 for rotamer 2, observed in the crystal structure [PDB The rotamers were generated with Pymol (DeLano 2009). Note that entry 3t0h (Li et al. 2012)] in cyan and for the combination of rota- the idealized, generated rotamers shows slightly different angles com- mer 2 and 3 in grey, for the two mutants 50C/54C and 149C/187C. b pared to the crystal structure rotamer Residue L107. The rotamer found in the crystal structure is shown in experiments, which were beyond the scope of this work. No Binding parameters other methyl groups were found to be experiencing the same phenomenon. The dissociation constant K as well as the dissociation D, rate constant (k ) were calculated with TITAN soft- OFF ware (Waudby et al. 2016) using a 1:1 binding model Ligand titration (A + B ⇆ A − B). TITAN fits both the equilibrium con- stant (K ) and the dissociation rate constant (k ) on the D OFF First, ntd-HSP90 mutants tagged with diamagnetic CLaNP-5 basis of the CSP and line broadening. A global fit of the were titrated with the weakly binding ligand 1. Three regions resonances of five methyl groups (Table S1) was performed of the protein exhibit chemical shift perturbations (CSP) and then one at the time was taken out and the fit repeated. upon titration (Fig. 6), revealing a localized binding site, in a The WT K was calculated using a titration performed on cleft of the protein, but with small CSP being observed rela- the N labelled protein, on the basis of four amide protons tively far away in the core of the protein, Fig. 7. The affected (Table S1). The ranges of values obtained in this way are regions are similar for the three mutants (Fig. 6), suggesting reported in Table 1 and provide a more realistic error range that the tag at different locations does not alter the binding than the fit error. Differences are observed between the dif- location. Moreover, most peaks show similar CSP directions ferent mutants, suggesting that the affinity is influenced by in the two-dimensional HSQC spectra (Fig. S6), indicating the tags to some degree, in particular for mutant 149C/187C. similar changes in the chemical environment of the methyl K values of 200 and 150 µM were used for the calcula- groups upon ligand binding. An exception is the resonance tion of the 100% bound state CSP for mutant 50C/54C and for Leu103 δ1, for which the CSP is not the same for all 101C/105C respectively. For these two mutants, a 50 µM mutants. The discrepancy is largest for mutant 101C/105C, variation of the K results in a change of the extrapolated in which the Leu is very close to the tag, being located in CSP of 0.01 ppm at most, half the error used for the further between the two engineered Cys residues. Thus, the tag at calculations. For mutant 149C/187C a K of 50 µM was this position appears to have some effect on ligand binding. used to extrapolate the CSP. A titration of the WT ntd-HSP90 in which the amide groups were observed was also performed. The results are in line Methyl group re‑orientation with the methyl titrations, with the largest effects near the binding cleft, in particular in the long α-helix that lines the The titrations were repeated with ntd-HSP90 tagged with 3+ binding site. Also for the amides, smaller effects are seen CLaNP-5 (Yb ). The binding characteristics were the far from the binding site, suggesting that binding effects are same as for the diamagnetic sample. To obtain the PCS transmitted into the rest of the protein (Fig. 7). for the ligand bound state, the maximal CSP, representing 1 3 Journal of Biomolecular NMR (2018) 71:275–285 281 Fig. 6 Ligand binding. |CSP| are plotted for methyl group resonances bound state. The red and blue lines mark 0.02 and 0.2 ppm, respec- 1 13 upon binding of 1. The H and C CSP are shown in red and blue tively. The light blue ovals highlight the most perturbed methyl bars, respectively. The CSP have been extrapolated to the 100% groups and back-predicted PCS were compared with those of the ntd-HSP90 fully saturated with 1, were calculated for both diamagnetic and paramagnetic proteins by extrapolation free protein. The correlation is very good, with Qa values of 0.03, 0.04 and 0.16 for mutants 50C/54C, 101C/105C and from the CSP data in the last point of the titration, by using the K values. The extrapolated CSP were subtracted to 149C/187C, respectively (Fig. S7). The Qa values are very similar to those found between the experimental PCS and obtain the PCS of the bound state. Ligand binding causes changes in some PCS up to 0.1 ppm, Fig. 8. All PCS val- the back-calculated PCS after tensor refinement (Lescanne et al. 2017). Thus, any change of the Δχ tensors is within ues are provided in Table S2. Methyl groups exhibiting sig- nificant ΔPCS, larger than 0.04 ppm for at least two of the the precision of its parameters. The fact that the largest PCS changes map to methyl groups located in the binding site paramagnetic centers, were selected for further analysis, comprising L48, L103, L107 and V136 (Table 2). also provides evidence that not a tensor change but move- δ2 δ1 δ2 γ1 ment of the methyl groups is the cause of the PCS changes. These residues surround the binding site of ntd-HSP90, Fig. 7. Changes in PCS could be caused by changes in the Under that assumption, the following calculations were done. We wondered whether the PCS can provide informa- position of the methyl groups due to interactions with the ligand, by changes in the tag position or orientation or by a tion about the distance range of the rearrangement as well as the new location of the methyl groups. Using the PCS of combination of both effects. The Δχ tensors were calculated from the PCS of the complex of ntd-HSP90 and 1 (Table S3) the bound form, the possible new positions can be calculated 1 3 282 Journal of Biomolecular NMR (2018) 71:275–285 Fig. 7 CSP map for binding of 1 to ntd-HSP90. The smaller spheres represent the Leu and Val methyl groups, coloured in a gradient of white to blue for increasing CSP as observed in the titration of mutant 50C/54C 3+ tagged with CLaNP-5 (Lu ). The larger spheres represent the amide nitrogens that were observed in a titration of WT ntd-HSP90, coloured according the average amide CSP (from white to red). The residues L48, L103, L107 and V136, harboring a methyl group with |ΔPCS| > 0.04 ppm for at least two mutants, are shown in yellow sticks. The structure is taken from PDB entry 3t0z. The ATP ligand observed in that structure is indicated by semi-tranparent, cyan sticks to indicate the binding site Table 1 Parameters for binding of 1 to ntd-HSP90 derived from the result in only two possible positions for the Leu that can titrations with TITAN software, for the WT and three mutants match simultaneously the PCS of both methyl groups to the −1 experimental ones. The center of the triple mutant cross- K (µM) k (s ) D OFF section area for the δ1 methyl is shifted by 3 Å upon binding WT 160–220 1400–1800 of the ligand, whereas for the δ2 methyl it does not change, 50C/54C 195–238 1500–2300 which could indicate that the sidechain rotates. Similarly, the 101C/105C 140–171 1075–1300 γ1 methyl group of Val136 experiences a significant change 149C/187C 41–45 1300–1900 in PCS, whereas the γ2 methyl group does not. The cross- section center of the γ1 methyl group for mutant 50C/54C with mutant 101C/105C shifted 1.2 Å from the position in the free protein. In Fig. 9c, it can be that the cross-section as an iso-surface of the new PCS around the lanthanoid. By determining the cross-sections of such iso-surfaces from two area for the bound state (yellow) has moved relative to the one for the free state (light gray). For Leu107, the two iso- or even three mutants, the location space can be reduced, as was shown above for the free protein. surfaces, from mutants 50C/54C and 149C/187C are sim- ilar because the two tensors happen to be nearly parallel All cross-sections for the ligand bound state could be determined using an uncertainty of 0.02 ppm, except for (Lescanne et al. 2017). Consequently, the cross-section is a curve area rather than a line like in the cases discussed Leu107, for which the uncertainty had to be raised to 0.03 ppm to find a cross-section. The cross-sections are above (Fig. S8). This cross-section area in the bound state is at a large distance from the conformation of Leu107 in shown in Fig. 9 in grey (γ1/δ1) and black (γ2/δ2) for the PCS of free protein and in yellow (γ1/δ1) and orange (γ2/ the crystal structure of the free protein, the minimal dis- tance between the δ2 methyl group and the cross-section δ2) for the PCS observed for the bound state. For Leu48, iso-surfaces could be calculated for mutants 50C/54C and being more than 5 Å, while it is only 0.6 Å for the free state. Thus, a large change in position is suggested by the 101C/105C and the cross-sections are shown for the δ1 and δ2 methyl groups, Fig. 9a. The conformation of the Leu in observed change in the PCS. In different crystal structures of ntd-HSP90, Leu 107 is found in very different orientations, the free protein is shown in sticks. The Leu methyl groups need to move at least 1.3 Å to move into the cross-section with 3.0 Å between Cα atoms in the structure of the free protein [PDB entry 3t0h (Li et al. 2012)] and the one with area of the bound state. Similarly, Fig. 9b shows the cross- sections for Leu103. In this case, data from all three mutants ntd-HSP90 bound to a close analogue of 1 [PDB entry: 2xdk 1 3 Journal of Biomolecular NMR (2018) 71:275–285 283 Fig. 8 PCS changes upon ligand binding. |ΔPCS| for Leu/Val methyl estimated error. The conservative maximum experimental error in the groups for the bound state are shown as blue, green and red bars for chemical shifts is estimated to be 0.02 ppm. The propagated error in the mutants 50C/54C, 101C/105C and 149C/187C, respectively. The ΔPCS is thus 0.04 ppm red dashed line represents the threshold of 0.04 ppm, based on the (Murray et al. 2010)]. This observation suggests that Leu107 Table 2 Significant ΔPCS (ppm). Methyl groups for which two |ΔPCS| > 0.04 ppm are listed is very sensitive to ligand binding. Therefore, the observed change in methyl group position is plausible, although it 50C/54C 101C/105C 149C/187C should be treated with caution because of the high degree of L48 − 0.046 − 0.058 No data δ2 correlation between the two Δχ tensors involved. The data L103 − 0.032 − 0.065 0.064 δ1 for mutant 101C/105C cannot be used in this case because L107 0.097 No data 0.067 δ2 the tag is too close to this Leu 107. V136 0.091 0.039 − 0.046 γ1 Fig. 9 Positions for methyl groups in the ntd-HSP90 bound to 1. mutant 50C/54C with 101C/105C and 149C/187C. The grey and Small spheres represent PCS iso-surfaces cross-sections in the free black spheres represent the triple-mutant cross-sections for the free state, at a distance of 1.6 Å (δ1) and 2.3 Å (δ2) from the crys- state, grey for Valγ1/Leuδ1 and black for Valγ2/Leuδ2, and in the tal structure positions. Note that the orange and black region overlap bound state, yellow for Valγ1/Leuδ1 and orange for Valγ2/Leuδ2. whereas the grey and yellow are bordering, suggesting that methyl The residues as found in the crystal structure of the free protein (PDB δ1 moves, whereas δ2 does not. c Val136, the grey and yellow small entry 3t0h) are shown in cyan sticks, with the methyl groups in the spheres represent the triple cross-section for methyl group γ1 in free analogous colors shown as 0.7 Å radius spheres centred on the carbon and bound state respectively. The small black spheres represent the methyl. a Leu48, with cross-sections for PCS from mutants 50C/54C free double cross-section between mutant 50C/54C and 101C/105C and 101C/105C. The minimal distance from the crystal structure con- of group γ2. The orange cross-section is not shown because the γ2 formation to the free cross-sections is 0.8 Å (δ1) and 0.6 Å (δ2), and methyl group showed insignificant PCS changes to the bound cross-sections is 1.3 Å (δ1) and 1.2 Å (δ2). b Leu103, the yellow and orange spheres represent the triple cross-sections of 1 3 284 Journal of Biomolecular NMR (2018) 71:275–285 In conclusion, the small but significant changes in the axes, cross-intersections of iso-surfaces are less resolved or PCS yield consistent information about the change in methyl non-existing. When using PCS for methyl localization, it group position that can be in the range of 1–5 Å. The iso- is important to realize that the PCS is highly anisotropic surface approach yields a limited set of possible conforma- and falls off with the third power of the distance between tions, in particular in the case when data from three tag the nucleus and the paramagnetic center. Localized, high locations are available. These data could be translated into PCS gradients make the PCS very sensitive to the methyl distance restraints, for example for ligand docking studies. group position. That is an advantage but also a danger. The higher the PCS gradient, the larger the effect of any PCS error will be. The use of several tag positions can counteract Conclusions this problem. However, this approach also demonstrated that positions predicted from PCS from two or three tags do not Methyl groups are widely used NMR probes but the appli- always match perfectly with those observed in the crystal cations of PCS to study them remain limited (Brewer et al. state of the protein, even in the absence of ligand. The rea- 2015). A straightforward application of paramagnetic tag- sons are not clear. It could be that the average position of ging is the additional dispersion induced by PCS. Paramag- these methyl groups is slightly different in the solution state netic tags have already been used to disperse resonances but it is also possible that (one of) the tags have subtle struc- of an intrinsically disordered protein, presenting a crowded tural effects causing a small mismatch between PCS of the amide proton spectrum (Gobl et al. 2016). Our data dem- native and tagged protein variants. Consequently, to study onstrate again that additional dispersion is also relevant for the effects of ligand binding it is recommended to use the folded, larger proteins that are being studied with methyl difference in PCS between free and bound states to derive group NMR probes, because two-dimensional spectra usu- distance restraints for methyl movement, as was discussed ally show considerable crowding for methyl resonances in above. To obtain reliable PCS the use of a probe that is the H region of 0.8–1.0 ppm, as was reported and used rigid relative to the protein is important. The high rigidity of before (Sattler and Fesik 1997). Ytterbium was the lantha- two-armed CLaNP-5 enables the measurement of accurate noid of choice for a 25 kDa protein. For larger proteins, PCS and the prediction of the tensors parameters accurately. lanthanoid of different paramagnetic ‘strength’ could be The choice of the lanthanoid determines the optimal region. used to benefit from a high dispersion, while limiting loss Nuclei too close to the tag will experience PRE, nuclei too of information because of PRE. far experience small changes in PCS upon displacement. 3+ Furthermore, PCS were used to provide evidence for With Yb used here, distances between nucleus and tag in small-scale movements (1–3 Å) of methyl groups upon the range of 50 Å yielded PCS changes of 0.04–0.1 ppm that ligand binding, by using a triangulation approach. This appear to match with displacements of 1–3 Å. The ability makes the PCS a powerful tool to observe re-arrangement to observe and characterize small but significant changes in of the side chains solely on the basis of 2D NMR spectra. the methyl group positions can potentially also be applied Of course, assignments are required for the interpretation to large proteins, because it is well-established that with of such data. Recently, we have demonstrated that the same deuteration methyl groups can be detected in very large paramagnetic constructs used in the ligand titrations could systems (Saio et al. 2014; Kerfah et al. 2015). We also note be used to obtain partial assignments of the methyl groups that this approach is not limited to ligands that are in fast (Lescanne et al. 2017). Thus, this application of PCS could exchange. As long as assignments are available for free and complement the use of NOESY experiments, or indeed crys- bound states of the protein, PCS can also be obtained in tal structures, for structure determination of ligand–protein slow-exchange systems. complexes. An advantage is that for measurement of PCS Funding The research leading to these results has received funding only low sample concentrations are required. It is clear that from the People Programme (Marie Curie Actions) of the European a single paramagnetic center is not sufficient to define a rel- Union’s Seventh Framework Programme FP7/2007–2013 under evant area for the methyl group position in the bound state. Research Executive Agency Grant 317127 (pNMR). With two paramagnetic centers and the use of both methyl Open Access This article is distributed under the terms of the Crea- groups of valine and/or leucine the positions can be approxi- tive Commons Attribution 4.0 International License (http://creat iveco mated, because of the additional steric constraint that the mmons.or g/licenses/b y/4.0/), which permits unrestricted use, distribu- two methyl groups must be at a distance of 2.5 Å. With tion, and reproduction in any medium, provided you give appropriate three tags, the location becomes quite restrained, although credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. two or three quite different positions may be found due to the shape of PCS iso-surfaces. Attention should be paid to the relative orientations of the susceptibility tensors used to generate the PCS. In case of tensors with parallel main 1 3 Journal of Biomolecular NMR (2018) 71:275–285 285 Li J et al (2012) Structure insights into mechanisms of ATP hydrolysis References and the activation of human heat-shock protein 90. Acta Biochim Biophys Sin 44:300–306. https ://doi.org/10.1093/abbs/gms00 1 Bashir Q, Volkov AN, Ullmann GM, Ubbink M (2010) Visualization of Liu WM, Overhand M, Ubbink M (2014) The application of paramag- the encounter ensemble of the transient electron transfer complex netic lanthanoid ions in NMR spectroscopy on proteins. Coord of cytochrome c and cytochrome c peroxidase. J Am Chem Soc Chem Rev 273:2–12. https ://doi.org/10.1016/j.ccr.2013.10.018 132:241–247. https ://doi.org/10.1021/ja906 4574 Murray CW et al (2010) Fragment-based drug discovery applied to Bohacek, R.S., McMartin, C. & Guida, W.C. The art and practice of Hsp90. Discovery of two lead series with high ligand efficiency. structure-based drug design: a molecular modeling perspective. J Med Chem 53:5942–5955. https ://doi.org/10.1021/jm100 059d Med. Res. Rev. 16, 3–50 (1996) Nagaraju GP et al (2017) Epigenetic effects of inhibition of heat shock Brewer KD et al (2015) Dynamic binding mode of a Synaptotagmin- protein 90 (HSP90) in human pancreatic and colon cancer. Cancer 1-SNARE complex in solution. Nat Struct Mol Biol 22:555. https Lett 402:110–116. https ://doi.org/10.1016/j.canle t.2017.05.021 ://doi.org/10.1038/nsmb.3035 Otting G (2010) Protein NMR using paramagnetic ions. Annu Rev Delaglio F et al (1995) Nmrpipe—a multidimensional spectral process- Biophys Biomol Struct 39:387–405. https://doi.or g/10.1146/annur ing system based on unix pipes. J Biomol NMR 6:277–293. https ev.bioph ys.09300 8.13132 1 ://doi.org/10.1007/BF001 97809 Palmer AG, Cavanagh J, Wright PE, Rance M (1991) Sensitivity DeLano WL (2009) PyMOL molecular viewer: updates and refine- improvement in proton-detected 2-dimensional heteronuclear ments. Abstracts of Papers of the Am Chem Soc 238 correlation Nmr-spectroscopy. J Magn Reson 93:151–170. https Den RB, Lu B (2012) Heat shock protein 90 inhibition: rationale and ://doi.org/10.1016/0022-2364(91)90036 -S clinical potential. Ther Adv Med Oncol 4:211–218. https ://doi. Pellecchia M et al (2008) Perspectives on NMR in drug discovery: a org/10.1177/17588 34012 44557 4 technique comes of age. Nat Rev Drug Discov 7:738–745. https Didenko T, Suarte MSA, Karagöz EG, Rüdiger SGD (2012) Hsp90 ://doi.org/10.1038/nrd26 06 structure and function studied by NMR spectroscopy. Biochim Saio T et al (2011) An NMR strategy for fragment-based ligand screen- Biophys Acta 1823:636–647. https ://doi.or g/10.1016/j.bbamc ing utilizing a paramagnetic lanthanide probe. J Biomol NMR r.2011.11.009 51:395–408. https ://doi.org/10.1007/s1085 8-011-9566-5 Erlanson DA (2012) Introduction to fragment-based drug discovery. Saio T, Guan X, Rossi P, Economou A, Kalodimos CG (2014) Struc- Top Curr Chem 317:1–32. https://doi.or g/10.1007/128_2011_180 tural basis for protein antiaggregation activity of the trigger factor Gobl C et al (2016) Increasing the chemical-shift dispersion of unstruc- chaperone. Science 344:1250494. https ://doi.org/10.1126/scien tured proteins with a covalent lanthanide shift reagent. Angew ce.12504 94 Chem Int Ed 55:14847–14851. https://doi.or g/10.1002/anie.20160 Sattler M, Fesik SW (1997) Resolving resonance overlap in the NMR spectra of proteins from differential lanthanide-induced shifts. Guan JY et al (2013) Small-molecule binding sites on proteins estab- J Am Chem Soc 119:7885–7886. https ://doi.org/10.1021/ja971 lished by paramagnetic NMR spectroscopy. J Am Chem Soc 356m 135:5859–5868. https ://doi.org/10.1021/ja401 323m Schleucher J et al (1994) A general enhancement scheme in heteronu- Hajduk PJ, Greer, J (2007) A decade of fragment-based drug design: clear multidimensional NMR employing pulsed field gradients. strategic advances and lessons learned. Nat Rev Drug Discov J Biomol NMR 4:301–306. https ://doi.org/10.1007/BF001 75254 6:211–219. https ://doi.org/10.1038/nrd22 20 Skinner SP, Moshev M, Hass MA, Keizers PH, Ubbink M (2013) PAR- Jahnke W et al (2000) Second-site NMR screening with a spin- Assign–paramagnetic NMR assignments of protein nuclei on the labeled first ligand. J Am Chem Soc 122:7394–7395. https ://doi. basis of pseudocontact shifts. J Biomol NMR 55:379–389. https org/10.1021/ja001 241+ ://doi.org/10.1007/s1085 8-013-9722-1 Jahnke W, Rudisser S, Zurini M (2001) Spin label enhanced NMR Sprangers R et al (2008) TROSY-based NMR evidence for a novel class screening. J Am Chem Soc 123:3149–3150. h ttps : //do i. of 20S proteasome inhibitors. Biochemistry 47:6727–6734. https org/10.1021/ja005 836g ://doi.org/10.1021/bi800 5913 John M, Pintacuda G, Park AY, Dixon NE, Otting G (2006) Structure Tu K, Gochin M (1999) Structure determination by restrained molecu- determination of protein-ligand complexes by transferred para- lar dynamics using NMR pseudocontact shifts as experimentally magnetic shifts. J Am Chem Soc 128:12910–12916. https ://doi. determined constraints. J Am Chem Soc 121:9276–9285. https :// org/10.1021/ja063 584z doi.org/10.1021/ja990 4540 Kay LE, Keifer P, Saarinen T (1992) Pure absorption gradient Tugarinov V, Kay LE (2005) Methyl groups as probes of structure and enhanced heteronuclear single quantum correlation spectroscopy dynamics in NMR studies of high-molecular-weight proteins. Chem- with improved sensitivity. J Am Chem Soc 114:10663–10665. biochem 6:1567–1577. https ://doi.org/10.1002/cbic.20050 0110 https ://doi.org/10.1021/ja000 52a08 8 Tugarinov V, Hwang PM, Ollerenshaw JE, Kay LE (2003) Cross- Keizers PHJ, Desreux JF, Overhand M, Ubbink M (2007) Increased correlated relaxation enhanced 1H-13C NMR spectroscopy of paramagnetic effect of a lanthanide protein probe by two- methyl groups in very high molecular weight proteins and pro- point attachment. J Am Chem Soc 129:9292–9293. https ://doi. tein complexes. J Am Chem Soc 125:10420–10428. https ://doi. org/10.1021/ja072 5201 org/10.1021/ja030 153x Keizers PHJ, Saragliadis A, Hiruma Y, Overhand M, Ubbink M (2008) Viegas A, Manso J, Nobrega FL, Cabrita EJ (2011) Saturation-trans- Design, synthesis, and evaluation of a lanthanide chelating pro- fer difference (STD) NMR: a simple and fast method for ligand tein probe: CLaNP-5 yields predictable paramagnetic effects inde- screening and characterization of protein binding. J Chem Educ pendent of environment. J Am Chem Soc 130:14802–14812. https 88:990–994. https ://doi.org/10.1021/ed101 169t ://doi.org/10.1021/ja805 4832 Waudby CA, Ramos A, Cabrita LD, Christodoulou J (2016) Two- Kerfah R, Plevin MJ, Sounier R, Gans P, Boisbouvier J (2015) Methyl- dimensional NMR lineshape analysis. Sci Rep. https ://doi. specific isotopic labeling: a molecular tool box for solution NMR org/10.1038/srep2 4826 studies of large proteins. Curr Opin Struct Biol 32:113–122. https Wiesnerl S, Sprangers R (2015) Methyl groups as NMR probes for ://doi.org/10.1016/j.sbi.2015.03.009 biomolecular interactions. Curr Opin Struct Biol 35:60–67. https Lescanne M et al (2017) Methyl group assignment using pseudocontact ://doi.org/10.1016/j.sbi.2015.08.010 shifts with PARAssign. J Biomol NMR. 69(4):183-195. 1 3
Journal of Biomolecular NMR – Springer Journals
Published: Jun 2, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
All the latest content is available, no embargo periods.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud