350 likes | 629 Views
Using Chemical Shift Perturbations to Study the Conformation of Protein-Ligand Complexes The J-SURF/SDILICON Approach Guillermo Moyna Department of Chemistry & Biochemistry, University of the Sciences in Philadelphia, Philadelphia, PA 19104-4495 Pfizer Global Research and Development
E N D
Using Chemical Shift Perturbations to Study the Conformation of Protein-Ligand Complexes The J-SURF/SDILICON Approach Guillermo Moyna Department of Chemistry & Biochemistry, University of the Sciences in Philadelphia, Philadelphia, PA 19104-4495 Pfizer Global Research and Development November 20th 2003
NMR in Drug Design • NMR-based methods such as SAR-by-NMR, STD-NMR, and Structure- • Based NMR Screening (SbN) are successful at finding mM-mM hits when • none are available from High-Throughput Screening (HTS). • Structures of these hits bound to their targets are needed to guide the • synthesis of higher affinity lead compounds. • Structures of complexes are difficult by NMR and/or X-ray, particularly for • poor binders. Chemists want to see the structure now… • New methods are needed to rapidly generate structures of weak hits • bound to their targets. 3D Structure Determination of mM-mM Protein-Ligand Complexes CADD Target Structures Leads
Rapid NMR-based Structure Determination • Chemical shift perturbations (Dds) can be used to determine residues • affected by ligand binding: Dd maps. • Advantages: • Very easy to generate and interpret. • Exquisitely sensitive to binding (mM). • Disadvantages: • Poor resolution. • Biased by large residues. Small or • buried groups are de-emphasized. • Can fast/accurate methods based solely on Dd be developed? Two new • tools will be discussed: • SDILICON: Dds replace/complement NOEs as intermolecular • constraints. • J-Surfaces: Dds are transformed into ligand spatial localization.
Ddin 3D Structure Refinement • By definition, chemical shifts are indicators of 3D structure. In proteins, • Dds (dobs - drcoil) are related to the protein’s 3D fold. • To employ Dd data in structure refinement • shielding equations are needed. The main • contributors are aromatic rings, peptide • groups, and charged moieties. • For example, effects from ring currents in • aromatic rings can be accounted for using • the Haigh-Mallion equation: _ + + _ H ri i rj pH j
Ddin Protein/Peptide 3D Structure Refinement • Relationships for other anisotropic groups in proteins have been • parametrized (Case/Williamson/Wishart). Final equations used in modeling: • We have used these to study small peptides (Fmoc-Pro-Pro-Xaa). Limited • NOE data, but large Dds (-0.7 to -1.2 ppm): • Moyna, G.; Williams, H. J.; Nachman, R. J.; Scott, A. I. J. Peptide Res. 1999, 53, 294. NOE + Dd NOE
Ddin Protein-Ligand 3D Structure Refinement • A similar approach can be used to study protein-ligand complexes if • certain assumptions are made: • Dd perturbations on protein nuclei are caused only by the ligands. • Limited conformational rearrangement of the protein upon binding. • Ligands have anisotropic groups (aromatic rings, carbonyls, etc.). • The first two are to some extent the case with weak (mM-mM) binders. • These are usually the hits missed by normal HTS approaches… • More than95% of all compounds in the MDL Drug Data Report (MDDR) • have aromatic rings. • We call the method Shift DIrected LIgand CONformation (SDILICON) Claritin Chlortrimeton
Running SDILICON • SDILICON uses protein Dds to optimize the orientation/conformation of the • ligand at the binding site. Sybyl mol2 or PDB files can be used. • A job control file (‘.sdl’ file) has • information on the ligand, perturbed • nuclei, ligand anisotropic groups • (rings, multiple bonds, charges), • ligand rotatable bonds, etc., etc. Ligand atom IDs (have to match the mol2/PDB file) Ligand rotatable bond atom pairs • (Too) Many command-line options • control the optimization. i.e., ‘-rc’ • controls the ring-current method, • ‘-ff’ what type of potential energy • function to use, etc., etc. • Making the control files by hand is • not bad, but its tedious and can lead • to many errors. Perturbed protein atoms (1H/13C/15N) Ligand anisotropic groups
Running SDILICON • SDILICON uses protein Dds to optimize the orientation/conformation of the • ligand at the binding site. Sybyl mol2 or PDB files can be used. • A job control file (‘.sdl’ file) has • information on the ligand, perturbed • nuclei, ligand anisotropic groups • (rings, multiple bonds, charges), • ligand rotatable bonds, etc., etc. • (Too) Many command-line options • control the optimization. i.e., ‘-rc’ • controls the ring-current method, • ‘-ff’ what type of potential energy • function to use, etc., etc. • Making the control files by hand is • not bad, but its tedious and can lead • to many errors. • Solved with a Sybyl ‘custom’ GUI…
SDILICON Pros/Cons • Pros… • SDILICON is a C/C++ command-line standalone. Runs on anything • with a decent C/C++ compiler (LINUX, IRIX, Mac OS X, etc.). • Simple code that is, for those willing, simple to modify and improve. • SDILICON is fast. Multiple ligands can be oriented in their binding site • in a matter of minutes. This include racemic mixtures… • A variety of optimization methods are available, including Line- • Minimization, RIPS, and Genetic Algorithms. • Cons… • Current version is ‘developmental’. A nicer interface would help… • No ligand flexibility. Good binding modes may be missed due to a • ligand fragment bumping against the protein. • Were do we put the ligands to begin with? • We have developed other tools that also use Dds to solve this last, perhaps • most important, problem.
Locating the Ligand • A Dd for a proton puts a geometric constraint on the location of the • perturbing group (i.e., the ligand). • Largest perturbations are due to aromatic rings. A magnetic point dipole • (Pople) can be used as a probe to locate the ligand. Depending on Dd: • If only one proton is perturbed, the ligand • can be anywhere in a sphere of radius rmax.
Locating the Ligand • A Dd for a proton puts a geometric constraint on the location of the • perturbing group (i.e., the ligand). • Largest perturbations are due to aromatic rings. A magnetic point dipole • (Pople) can be used as a probe to locate the ligand. Depending on Dd: • If only one proton is perturbed, the ligand • can be anywhere in a sphere of radius rmax. • If more than one proton is perturbed, the • probability of locating the ligand will be • higher in the intersection of the spheres. • McCoy, M. A.; Wyss, D. F. J. Am. Chem. Soc. 2002, 124, 11758.
Locating the Ligand - J-Surfaces • Since these surfaces describe the most likely location of the ligand’s • electron density we call them J-surfaces. • Intersection of spheres with equal densities would make small shifts • dominate the J-surface.Solved by using uniform density for all spheres • and considering the intersection point density. • The density r3 dependency balances the Dd 1/r3 dependency, providing • self-consistency (i.e., effects from protons with large Dds far from binding • site are de-emphasized…). • Apart from giving a clear spatial location for the ligand, J-surfaces give • excellent starting points for the SDILICON optimization algorithm. 1H Dd data from 1H-15N HSQC (HCV NS3 Protease) Dd Map J-Surface + vdW
Case Study I - Calmodulin/W7 Complex • Ca2+-bound calmodulin (CaM) binds to two molecules of inhibitor W7 with • similar affinity (~10 mM). The structure of the complex was determined by • Ikura and co-workers using intermolecular NOE constraints. • There are a total of 31 non-NH protons with |Dd| larger than 0.1 ppm, 19 on • the N-termini and 12 on the C-termini. Mapped skyblue (-) and red (+). • Osawa, M. et al. J. Mol. Biol. 1998, 276, 165. W7 C-Termini N-Termini
CaM/W7 Complex • Using the reported Dd perturbations a J-surface for the complex was • computed. • The ligands determined from NOE data intersect the highest density • J-surfaces. Computation time after entering shifts is less than 1 second… C-Termini N-Termini
CaM/W7 Complex • Is the J-surface more informative than the regular Dd map regarding the • spatial location of the ligands? Looking at the C-termini binding site: • Clearly yes…
CaM/W7 Complex • Once the spatial locations of the two W7 ligands in the protein were • determined, SDILICON was used to optimize their binding site orientation. • Both ligands optimized simultaneously. Only 3 minutes of computation… • There is good agreement between SDILICON and NOE structures. • Initially, differences in the N-terminal binding mode assumed to be due to • conformational rearrangement upon W7 binding. C-Termini N-Termini
CaM/W7 Complex • The optimization was repeated after ligand rotatable bonds were • implemented into SDILICON. GAs were used to obtain a ‘global minimum’, • and the resulting structure line-minimized. • Better agreement with NOE structure. The rigid ligand side-chain was • bumping against the N-termini binding site. Rotatable bonds are needed, • even if the ligand’s anisotropic groups are part of a rigid framework… C-Termini N-Termini
CaM/W7 Complex • How consistent with the observed Dd data are the structures obtained from • SDILICON calculations? • We can back-calculate shift perturbations for all/some protein protons from • the SDILICON structure, use them to compute a theoretical J-surface (Jcalc), • and compare it to the observed J-surface (Jobs). • There is > 30% overlap between Jobs and Jcalc, indicating that the • SDILICON 3D model is consistent with the observed shift perturbations. Observed (Jobs) Calculated (Jcalc) Intersection (JobsJcalc)
Case Study II - Neocarzinostatin (NCS) • CaM was an ideal case. Sulfur-aromatic interactions between methionines • and naphtalenes create large Dds that guide optimization. No NHs used. • apoNCS-CH9 complex studied by Caddick and co-workers. 38 NH and CH • protons with |Dd| larger than 0.1 ppm, only one |Dd| larger than 0.7 ppm. • Urbaniak, M. D. et al. Biochemistry2002, 41, 11731. CH9
apoNCS Complexes • Once again, the J-surface clearly points to the spatial location of the • ligand in the protein binding site. CH9 CH9 vdW-Accessible J-Surface
apoNCS Complexes • Once again, the J-surface clearly points to the spatial location of the • ligand in the protein binding site. • At lower densities, conformational rearrangement is detected. Phe78 - Act as flaps over binding site CH9 Thr85 Raw J-Surface - Lower density
apoNCS Complexes • The SDILICON calculations were done filtering out Dd perturbations not • contributing to the high density J-surface. Low-energy structures of CH9 • obtained from a GA search used as starting points for line-minimization. • Clearly, the structure that puts the ligand further away from the binding site • is ‘wrong’. However, we need a non-biased method to confirm this.
apoNCS Complexes • Again, this can be done by comparing the Jobs surface to Jcalc surfaces • derived from both models. • Jcalc surfaces that deviate substantially from the Jobs surface indicate • structures inconsistent with the observed Dd data. These models can be • eliminated from the structure pool. 20% intersection 0% intersection
apoNCS Complexes • The previous examples use full Dd assignments for the J-surface and • SDILICON calculations. These are great, but take a long time to gather. • The minDd method is an alternative for quick/tentative assignments. NH • minDds for CH9 and three additional apoNCS binders were available. • minDds cannot be used with SDILICON (signs are lost, miss-assignments, • etc.). However, they show perturbations ideal for J-surface calculations… • Williamson, R. A. et al. Biochemistry1997, 36, 13882. CH3 CH5 CH7
apoNCS Complexes • J-surfaces derived from minDds for all four ligands… • Accurate, even when iffy perturbations are used. Ideal for ‘automation’… CH9 (Dd) CH3 (minDd) CH9 (minDd) CH7 (minDd) CH5 (minDd)
Some Real Data • Previous examples used polished data from academia. What about some • real-life stuff? Data from compounds deemed non-leads in SPRI HCV • Protease program (80 mM - 1 mM binders). • Only NH Dds available. The number of Dds varies from 3 (SCH17865) to • 22 (SCH10386), and their ranges are as small as -0.15 to -0.07 ppm • (SCH17865), to -0.59 to 0.63 ppm (SCH92). SCH10363 SCH17865 SCH92 SCH9301 SCH415425
Some Real Data • We start seeing problems. When we have a limited number of small Dds, • it’s hard to pin down a binding site using J-surfaces. For example, • SCH17865, with only three Dds in the -0.15 to -0.07 range: • We basically have only three very large spheres (1/r3 dependency), which • results in a very large intersection volume (> 202 Å3). • SDILICON will not give us a unique family of conformers, but several • located in a large region of space…
Some Real Data • However, compounds SCH9301 and SCH92 give well defined J-surfaces • (< 50 and 30 Å3) in the same region of space (50 % intersection): • Similar results for SCH415425. All share a common binding site. SCH9301 SCH92
Some Real Data • The orientation/conformation of SCH9301 was then computed with • SDILICON. A single ‘global-minimum’ was obtained with GAs and further • optimized by line-minimization. • In order to validate the resulting structure, we again looked at the Jobs • surface versus the Jcalc surface derived from the optimized structure… H57 S138 K136 L135 R155 Y134 A156 A157
Some Real Data • Since only NH data was used, only back-calculated NH Dds were used to • compute Jcalc. Observed (Jobs) Calculated (Jcalc)
Some Real Data • Since only NH data was used, only back-calculated NH Dds were used to • compute Jcalc. • Although it won’t, this 3D model could be used to design new lead • compounds. Since no X-Ray data are available for these complexes, this • example shows the potential of the J-SURF/SDILICON approach in SbN. Observed (Jobs) Intersection (JobsJcalc) - 75% overlap
Something for SMASH People… • The J-SURF/SDILICON approach is not limited to ligand- and • protein-protein complexes.We applied it successfully to the study of • perylene oligomerization. • In these compounds there is a • concentration-dependent upfield • shift of the Ha and Hb protons. • In the ‘dimer’, DdHa = -0.31 ppm • and DdHb = -0.51 ppm. • We are obviously dealing with ring- • currents and, to a lesser extent, • amide group anisotropy effects. • Ideal for J-SURF/SDILICON… • Wang, W.; Li, L.-S.; Helms, G.; Zhou, H.-H.; Li, A. D. Q.; J. Am. Chem. Soc. 2003,125, 1120. Ha Hb
Shift-Minimized Perylene dimer • These are the J-surface obtained for one of the monomers and the shift- • minimized dimer structure. The back-calculated Dd values for Ha and Hb • protons are -0.32 and -0.48 ppm respectively. • As expected, the highest J-density is right on top (bottom) of the rings. The • distance between rings obtained using Dd constraints is 3.51 Å, almost • identical to the distance obtained from ab initio calculations (3.55 Å). 3.51 Å
Conclusions and Future Work • J-surfaces are a simple and rapid way to spatially locate ligands from Dds. • The method also identifies protein regions which undergo rearrangement • upon ligand binding or mutation. A web-based ‘J-server’ coming soon… • SDILICON rapidly docks ligands based solely on Dd perturbations and • intermolecular non-bonded interactions. Structures obtained are similar in • quality to those determined from intermolecular NOEs. • Combined they provide a quick approach to locate and dock ligands in the • protein binding site. Ideal for high-throughput structure determination. • Current version (11/03) allows for rotation around ligand single bonds, and • for exhaustive conformational search with GAs. • http://tonga.usip.edu/gmoyna/sdilicon/ • Parameters for anisotropic groups other than aromatic rings and amides, • such as sulfones, carboxylates, multiple bonds, 15N, etc., are required.
Acknowledgments People Dr. Mark McCoy (SPRI) Prof. Stephen Caddick (U. of Sussex) Prof. Alexander DeQuan Li (WSU) Zhijian Li (USP - SDILICON GA) Edward P. O’Brien (USP - J-SURF - Currently UMD) Adam Wenocur (USP - J-SURF) Prof. Randy J. Zauhar (USP - My Own C/C++ Guru…) Funding Schering-Plough Research Institute Office of the VP of Academic Affairs, USP