Functional 3-D modelling of G protein coupled receptors Uğur Sezerman Central Dogma DNA Transcription mRNA Translation PROTEINS Motivation • Knowing the structure of molecules enables us to understand its mechanism of function • Current experimental techniques – X-ray cystallography – NMR X-Ray Crystallography • crystallize and immobilize single, perfect protein • bombard with X-rays, record scattering diffraction patterns • determine electron density map from scattering and phase via Fourier transform: • use electron density and biochemical knowledge of the protein to refine and determine a model "All crystallographic models are not equal. ... The brightly colored stereo views of a protein model, which are in fact more akin to cartoons than to molecules, endow the model with a concreteness that exceeds the intentions of the thoughtful crystallographer. It is impossible for the crystallographer, with vivid recall of the massive labor that produced the model, to forget its shortcomings. It is all too easy for users of the model to be unaware of them. It is also all too easy for the user to be unaware that, through temperature factors, occupancies, undetected parts of the protein, and unexplained density, crystallography reveals more than a single molecular model shows.“ - Rhodes, “Crystallography Made Crystal Clear” p. 183. NMR Spectroscopy determining constraints using constraints to determine secondary structure • protein in aqueous solution, motile and tumbles/vibrates with thermal motion • • NMR detects chemical shifts of atomic nuclei with non-zero spin, shifts due to electronic environment nearby • determine distances between specific pairs of atoms based on shifts, “constraints” • use constraints and biochemical knowledge of the protein to determine an ensemble of models Primary Assembly Secondary Folding Tertiary Packing Quaternary Interaction PROCESS STRUCTURE Biology/Chemistry of Protein Structure Protein Assembly • occurs at the ribosome • involves dehydration synthesis and polymerization of amino acids attached to tRNA: + 3 2 n NH - {A + B A-B + H O} -COO • yields primary structure Amino Acids Forces driving protein folding • It is believed that hydrophobic collapse is a key driving force for protein folding – Hydrophobic core – Polar surface interacting with solvent • Minimum volume (no cavities) Van der Walls • Disulfide bond formation stabilizes • Hydrogen bonds • Polar and electrostatic interactions PROTEIN FOLDING PROBLEM • STARTING FROM AMINO ACID SEQUENCE FINDING THE STRUCTURE OF PROTEINS IS CALLED THE PROTEIN FOLDING PROBLEM Secondary Structure • non-linear • 3 dimensional • localized to regions of an amino acid chain • formed and stabilized by hydrogen bonding, electrostatic and van der Waals interactions The a-helix Ramachandran Plot • Pauling built models based on the following principles, codified by Ramachandran: (1) bond lengths and angles – should be similar to those found in individual amino acids and small peptides (2) peptide bond – should be planer (3) overlaps – not permitted, pairs of atoms no closer than sum of their covalent radii (4) stabilization – have sterics that permit hydrogen bonding • Two degrees of freedom: (1) (phi) angle = rotation about N – Ca (2) (psi) angle = rotation about Ca – C • A linear amino acid polymer with some folds is better but still not functional nor completely energetically favorable packing! Chou-Fasman Parameters Name Alanine Arginine Aspartic Acid Asparagine Cysteine Glutamic Acid Glutamine Glycine Histidine Isoleucine Leucine Lysine Methionine Phenylalanine Proline Serine Threonine Tryptophan Tyrosine Valine Abbrv A R D N C E Q G H I L K M F P S T W Y V P(a) 142 98 101 67 70 151 111 57 100 108 121 114 145 113 57 77 83 108 69 106 P(b) P(turn) 66 83 95 93 146 54 156 89 119 119 74 37 98 110 156 75 95 87 47 160 59 130 101 74 60 105 60 138 152 55 143 75 96 119 96 137 114 147 50 170 f(i) 0.06 0.07 0.147 0.161 0.149 0.056 0.074 0.102 0.14 0.043 0.061 0.055 0.068 0.059 0.102 0.12 0.086 0.077 0.082 0.062 f(i+1) 0.076 0.106 0.11 0.083 0.05 0.06 0.098 0.085 0.047 0.034 0.025 0.115 0.082 0.041 0.301 0.139 0.108 0.013 0.065 0.048 f(i+2) 0.035 0.099 0.179 0.191 0.117 0.077 0.037 0.19 0.093 0.013 0.036 0.072 0.014 0.065 0.034 0.125 0.065 0.064 0.114 0.028 f(i+3) 0.058 0.085 0.081 0.091 0.128 0.064 0.098 0.152 0.054 0.056 0.07 0.095 0.055 0.065 0.068 0.106 0.079 0.167 0.125 0.053 HOMOLOGY MODELLING • Using database search algorithms find the sequence with known structure that best matches the query sequence • Assign the structure of the core regions obtained from the structure database to the query sequence • Find the structure of the intervening loops using loop closure algorithms Homology Modeling: How it works o Find template o Align target sequence with template o Generate model: - add loops - add sidechains o Refine model 1esr TURALIGN: Constrained Structural Alignment Tool For Structure Prediction Motif Alignment Using Dynamic Algorithm RESULTS • For all the experiments done, our algorithm perfectly matched functional sites and motifs given as input to the program. – 1csh vs 1iomA : • RMSD = 2.50 – 1csh vs 1k3pA • RMSD = 2.12 – 1k3pA vs 1iomA • RMSD = 3.03 – 1b6a vs 1xgsA • RMSD = 2.23 – 1fp2A vs 1fp1D • RMSD = 2.98 • At average we got the best results for 5 experiments: • RMSD = 2.57 with ac:0.4,sc:0.4,tc:0.2,cc:0 Thanks to • Tural Aksel Why Functional Classification? • Huge amount of data accumulated via genome sequencing projects. • Costly experimental structure prediction methods (X-ray & NMR), takes months/year. • Also computational structure prediction methods are not accurate enough. G-protein coupled receptors (GPCRs) • Vital protein bundles with versatile functions. • Play a key role in cellular signaling, regulation of basic physiological processes by interacting with more than 50% of prescription drugs. • Therefore excellent potential therapeutic target for drug design and the focus of current pharmaceutical research. GPCR Functional Classification Problem • Although thousands of GPCR sequences are known, the crystal structure solved only for one GPCR sequence at medium resolution to date. • For many of them, the activating ligand is unknown. • Functional classification methods for automated characterization of such GPCRs is imperative. Relationship between specific binding of GPCRs into their ligands and their functional classification • Subfamily classifications in GPCRDB are defined according to which ligands the receptor binds (based on chemical interactions rather than sequence homology). • According to the binding of GPCRs with different ligand types, GPCRs are classified into at least six families. • The correlation between sub-family classification and the specific binding of GPCRs to their ligands can be computationally explored for Level 2 subfamily classification of Amine Level 1 subfamily. Benchmark Dataset • Dataset – 352 amines, 595 peptides, 1898 olfactory, 355 rhodopsin, 56 prostanoid • Derive GPCR proteins from GPCRDB & SWISSPROT through internet – Group the proteins according to their ligand specificity (i.e amines, peptides, olfactory, rhodopsin, prostanoid) – Seperate proteins into train and test groups with 2:1 ratio respectively – Derive the ecto-domains by using TMHMM (i.e nterminal, loop1, loop2, loop3) – Rewrite the sequences using 11 letter alphabets Classification of Amino acids Class Amino Acids Class Amino Acids a I,V,L,M g G b R,K,H h W c D,E i C d Q,N j Y,F e S,T k P f A Snake plot of the human beta-2 adrenoceptor PROTEIN DATABASE Train proteins; Ligand group: amines ID NAME Sequence n-term Loop1 ... 1 5H1A_RAT MDVFSF... acajejgdgd... jdaadbhe... ... 2 5H1A_FUGRU MDLRATS... bekccbec... aakjiceeiba.. ... 3 5H1A_HUMAN MDVLSPG.. bdfbfcccaa... aibcfihjbaf... . ... 4 5H1B_PANTR MEEPGAQ.. acckgfdifk kaibcfihj ... 5 5H1B_RABIT MEEPGAQ.. acckgfdifkka ... ibcfihjbd ... FINDING MOST COMMON PATTERNS FOR EACH LIGAND GROUP • Form triplets for n-terminal, loop1, loop2 and loop3 seperately – For 11 letter alphabet 1331 different triplets • For each triplet find proteins in certain ligand group those containing the current triplet at a given location and keep the data in vectors • Find the ratio of occurence of each triplet in a given GPCR protein type(i.e amines) in a given location (i.e loop1) • Insert the triplets into SQL database with their ratios • Sort the triplets according to their ratios VECTORS ID WORD PROTEINS 1 aaa 5H1A_RAT, 5H1A_FUGRU, ... 2 aab 5HT1_APLCA, 5HTA_DROME, ... 3 aac 5HT1_APLCA, 5HTA_DROME, 5H1A_PONPY 4 aad none ... ... ... 1328 kkh 5H1B_FUGRU , 5HTA_DROME... 1329 kki none 1330 kkj 5H1F_RAT 1331 kkk none FINDING DISTINGUISHING MOTIFS I • Compare the ratios of triplets of a certain ligand group with the occurence of this triplet with the other ligand groups one by one(aaa in amines = 0.5; in peptides = 0.1 r = 0.5/0.1 • Keep the motifs with n(150) highest “r”s for each ligand group pairs. These are the motifs that distinguish given group from the other groups RESULTS • Success rates for Information theory CART RESULTS The classification table showing the only patterns determining amines from all others • • • • • • • • • Index Triplet Family 1 CAA Amine 2 AIB Amine 3 HIJ Prostanoid 4 AEA Hormone-protein 5 JAA Hormone-protein 6 AAD TRH 7 ADA TRH 8 JCK Melatonin i.e. Variable importance of the amine determining patterns Patterns Relative Importance Loop 1 ‘caa’ 100 Loop 1 ‘gbh’ 97.46 Loop 3 ‘iak’ 83.767 Loop 1 ‘gjh’ 64.62 Loop 1 ‘gda’ 51.101 Loop 2 ‘aed’ 44.942 Loop 1 ‘agj’ 43.636 Loop 1 ‘aag’ 31.099 Loop 1 ‘dca’ 22.736 Loop 3 ‘akc’ 17.737 Loop 1 ‘hjj’ 16.511 N-term ‘afa’ 12.811 N-term ‘eea’ 0 Occurence of EIG in Loop2 in Rhodopsin Family Triplet JJI at exo-loop 2 in olfactory sub-family. Conclusion • Exploiting the fact that there is a non-promiscuous relationship between the specific binding of GPCRs into their ligands and their functional classification, our method classifies Level 1 subfamilies of GPCRs with a high predictive accuracy of 98%. • The presented machine learning approach, bridges the gulf between the excess amount of GPCR sequence data and their poor functional characterization. • The method also finds binding motifs of GPCRs to their specific ligands which can be exploited for drug design to block these site • With such an accurate and automated GPCR classification method, we are hoping to accelerate the pace of identifying proper GPCRs and their ligand binding scheme to facilitate drug discovery especially for neurological diseases. • Ligand binding motifs and their site information can be used as contraints to build better models. • Highly conserved sites from alignment of GPCR families can also be used as constraints Thanks to • Murat Can Çobanoğlu Class A Rhodopsin like • The largest and most diverse family of GPCRs • Conserved sequence motifs • Unique signal-transduction activities • Important members: – – – – – – Adrenergic Receptors Adenosine Receptors Chemokine Receptors Dopamine Receptors Histamine Receptors Opsins Highlighted 4 GPCRs for Structure Comparison Species human GPCR β2AR Ligand inverse agonists carazolol (Adrenergic) avian β1AR antagonist cyanopindolol (Adrenergic) human A2A (Adenosine) antagonist ZM241385 bovine Rhodopsin inverse agonist 11-cis retinal Extracellular surfaces • The most significant structural divergences lie in the extracellular loops and ligand-binding region β2AR/β1A R A2A rhodopsin - contain a short α-helix that is stabilized by intra- and inter-loop disulphide bonds - N-terminal regions are disordered - lacks a predominant secondary structure and expose the ligand-binding cavity to extracellular bulk solvent -forms a short β-sheet that caps the ligand and shielding the chromophore from bulk solvent and preventing Schiff base hydrolysis - amino terminus glycosylated Ligand-Binding Pockets • For both adrenergic receptors and rhodopsin, ligand binding is mediated by polar and hydrophobic contact residues from TM3, TM5, TM6 and TM7. • Ligand superpositions are partly overlapping for β2AR, β1AR and rhodopsin, however, for β2AR and β1AR are slightly more extracellular than rhodopsin. • This difference results in a significant in key rotamer conformational Ligand-Binding Pockets • In contrast to the β2AR, β1AR and rhodopsin, the ligand of A2A ( Adenosin) receptor binds in a mode that is roughly perpendicular to the bilayer plane, and the packing interactions with the protein, mostly with TM6 and TM7. Ligand-Binding Pockets • Despite the highly conserved seven transmembrane architecture, GPCRs can support a wide variety of ligand-binding modes • Also high conservation in the ligand-binding pocket is observed as well as in other subfamilies of GPCRs probably explains some of the difficulty in obtaining potent subtype-selective compounds in pharmaceutical discovery programs Cytoplasmic surfaces of the GPCR structures • Major structural difference between the ligand-activated GPCRs and rhodopsin lies in the ‘ionic lock’ between the highly conserved E/DRY motif on TM3 and a glutamate residue on TM6. • Conserved among all family A GPCRs, these amino acids form a network of polar interactions that bridges the two transmembrane helices, stabilizing the inactive-state conformation. Cytoplasmic surfaces of the GPCR structures • One common feature is the chemical environment surrounding residues of the highly conserved NPXXY motif. The cytoplasmic end of TM7, in which this motif is located, participates in key conformational changes associated with GPCR activation. • The proline in this motif causes a distortion in the α-helical structure, and the tyrosine faces into a pocket lined by TM2, TM3, TM6 and TM7. Mechanism for Activation • Structures of opsin provide clues to the transmembrane helix rearrangements that can be expected as a result of agonist binding • Most importantly, the side chain of Trp 265 (the toggle switch) moves into space previously occupied by the ionone ring of retinal • The cytoplasmic end of TM6 is shifted more than 6 Å outwards from the centre of the bundle Snake plot of the human beta-2 adrenoceptor