Protein Structure and Prediction

Michael Strong, Ph.D.Integrated Center for Genes, Environment, and HealthNational Jewish Health

Why do we care about protein structuresStructure leads to Function – Enzymes, Structural (Cell Wall), Protein Interactions (host-pathogen), Replication, Transcription, Translation

(Binds to receptors on a cell surface, attached carbohydrades help it evade immune system)

Shapes the virus, enables budding

Protects the RNA viral genome

Ebola Virus-RNA genome-Encodes 7 proteins

To make new copies of RNA genome

Why do we care about protein structuresCombining Structure and Genomic Information- Help us understand phylogeny and implications of mutations

Science 12 September 2014: Vol. 345 no. 6202 pp. 1369-1372

Why do we care about protein structuresCombining Structure and Genomic Information- Help us understand phylogeny and implications of mutations

Drug Resistance

Rifampin

Tuberculosis

rpoB drug target

Why do we care about protein structuresCombining Structure and Genomic Information- Help us understand phylogeny and implications of mutations

Homology model of CFTR structure, with common mutation F508 .PNAS 3256–3261, 105:9

Experimental Approach

PQITLWKRPLVTIRIGGQLKEALLDTGADDTVLEEMNLPGKWKPKMIGGIGGFIKVRQYDQIPIEICGHKAIGTVLVGPT PVNIIGRNLLTQIGCTLNF

From Sequence to Structure

HIV Protease

HIV ProteaseWith Inhibitor

X-ray CrystallographyNMRCryo-Electron Microscopy

Computational Approach

From Sequence to Structure

MNPNQKIITIGSVCMTIGMANLILQIGNIISIWISHSIQLGNQN QIETCNQSVITYENNTWVNQTYVNISNTNFAAGQSVVSVKLAGNSSLCPVSGWAIYSK DNSVRIGSKGDVFVIREPFISCSPLECRTFFLTQGALLNDKHSNGTIKDRSPYRTLMS CPIGEVPSPYNSRFESVAWSASACHDGINWLTIGISGPDNGAVAVLKYNGIITDTIKS WRNNILRTQESECACVNGSCFTVMTDGPSNGQASYKIFRIEKGKIVKSVEMNAPNYHY EECSCYPDSSEITCVCRDNWHGSNRPWVSFNQNLEYQIGYICSGIFGDNPRPNDKTGS CGPVSSNGANGVKGFSFKYGNGVWIGRTKSISSRNGFEMIWDPNGWTGTDNNFSIKQD IVGINEWSGYSGSFVQHPELTGLDCIRPCFWVELIRGRPKENTIWTSGSSISFCGVNS DTVGWSWPDGAELPFTIDK"

H1N1 NA

Homology ModelingProtein ThreadingAb initio

Protein Building Blocks

Typical Protein Sequence MNPNQKIITIGSVCMTIGMANLILQIGNIISIWISHSIQLGNQN

Protein Building Blocks

Amino Acid Side Chain (R groups)

Amino Acid Side Chain (R groups)

DisulfideBonds

Amino Acid Side Chain (R groups)

-acidic

+basic

DNA

Folded Protein

Most Proteins Spontaneously Fold

RNA

Some proteins need chaperones for correct folding

Transcribed by RNA polymerase

Translated by Ribosome

native state, Folded protein

spontaneous self-organisation (~1 second)

Most Proteins Spontaneously Fold

Folded protein

Unfolded protein

Denaturing conditions

Native conditions

ChristianAnfinsen’s Experiment1950s

Most Proteins Spontaneously Fold

Important to Computational Biologists, because this suggests that all information relating to the correct folding of a protein is contained in it’s primary amino acid sequence, but …

Most Proteins Spontaneously Fold

But Proteins lack easy rules for folding as compared to DNA

ProteinDNA

Many Factors Influence Protein Folding

Protein

Proteins Assume the Lowest Energy Structure

Factors that influence folding include:1.Hydrophobic Interactions / collapse (particularly within the core)2.Hydrogen bonds – lead to secondary structures3.Disulfide Bonds (Cysteine residues)4.Salt Bridges / Ionic Interactions (among charged residues)5.Multimeric interactions with same type or other proteins

Common Secondary StructuresAlpha helix

Common Secondary StructuresBeta Sheet

Common Secondary StructuresLoop Regions

Loop

Strong M et al, Proc National Academy of Sciences vol. 103 no. 21, 8060–8065, 2006

Example - Hemoglobin

Fluoroquinolone Target gyrACrystal Structure

Rifampin targetrpoBHomology Model

Diversity of Protein Structures

Streptomycin resistancegidB Homology model

Isoniazid TargetinhACrystal Structure

EthionamideTarget, inhACrystal StructureIsoniazid Activating

Enzyme, KatGCrystal Structure

Streptomycin ResistancerpsL Homology model

PyrazinamideActivating enzymepncACrystal Structure

A B C D

E F G H

Experimental Methods of Structure DeterminationX-ray crystallographyHigh resolution structure determination

Grow a protein Crystal

Experimental Methods of Structure DeterminationX-ray crystallographyHigh resolution structure determination

Experimental Methods of Structure DeterminationX-ray crystallographyHigh resolution structure determination

•Intensities and phases of all reflections are combined in a Fourier transform to provide maps of electron density

Phases determined by using heavy metals or selenomethionine (MAD)

• Smaller Proteins than X-ray • Distances between pairs of hydrogen

atoms• Lots of information about dynamics• Requires soluble, non-aggregating

material• Assignment sometimes

difficult

Experimental Methods of Structure DeterminationNMR – Nuclear Magnetic ResonanceHigh resolution structure determination

NOE cross-peak if they are within 5.0 Å

• Low to medium resolution ~10-15Å

• Limited information about dynamics

• Can be used for very large molecules and complexes

Experimental Methods of Structure DeterminationCryo Electron MicroscopyLow to medium resolution structure determination

Database of Protein StructuresPDB – Protein Data Bank

Database of Protein StructuresPDB – Protein Data Bank

104,125 structures as of 10/20/2014

Database of Protein StructuresPDB – Protein Data Bank

Even so, the number of solved structures greatly lags behind the rate of new genes being sequenced … Solution: Computational Structural Methods

GenBank Sequences

• Atoms in pdb files are defined by their Cartesian coordinates:

Database of Protein StructuresPDB – Protein Data Bank Files

Visualization of PDB filesPymol, Jmol, Chimera, etc

Visualization of PDB filesPymol, Jmol, Chimera, etc

DALI Structural AlignmentsAlign Protein Structures, Structure SuperpositionGenerates a comparison matrix (transform protein into a 2D array of distances between C-alpha atoms. Z score reflects reliability, lowest RMSD identified

Computational Approach

From Sequence to Structure

MNPNQKIITIGSVCMTIGMANLILQIGNIISIWISHSIQLGNQN QIETCNQSVITYENNTWVNQTYVNISNTNFAAGQSVVSVKLAGNSSLCPVSGWAIYSK DNSVRIGSKGDVFVIREPFISCSPLECRTFFLTQGALLNDKHSNGTIKDRSPYRTLMS CPIGEVPSPYNSRFESVAWSASACHDGINWLTIGISGPDNGAVAVLKYNGIITDTIKS WRNNILRTQESECACVNGSCFTVMTDGPSNGQASYKIFRIEKGKIVKSVEMNAPNYHY EECSCYPDSSEITCVCRDNWHGSNRPWVSFNQNLEYQIGYICSGIFGDNPRPNDKTGS CGPVSSNGANGVKGFSFKYGNGVWIGRTKSISSRNGFEMIWDPNGWTGTDNNFSIKQD IVGINEWSGYSGSFVQHPELTGLDCIRPCFWVELIRGRPKENTIWTSGSSISFCGVNS DTVGWSWPDGAELPFTIDK"

H1N1 NA

Secondary Structure PredictionAlpha Helix, Beta Strand, or Other

Tertiary Predictions:

1.Homology Modeling2.Fold Recognition3.De Novo Protein Structure Prediction

Secondary Structure Prediction-probability of amino acid to be in a alpha helix, Beta strand, or other (coil/loop) based on known structures.

-Chou-Fasman (short runs of amino acids), GOR (Bayesian, takes neighbors into account)

- helices – no prolines, periodicity 3.6 residues/turn- strands – alternating hydropathy, or ends hydrophillic and

center hydrophobic-other – small, polar, flexible residues, and prolines

But, stalled at 55- 60% accuracy

Also used position specific profiles based on multiple sequence alignments (evolutionary information) (ie insertion/deletion more likely to be in coil/turn), PSI BLAST and HMM, NN and SVM (improved to about 75-80%)

Secondary Structure Prediction

But we really want to know how the protein folds in three dimensions

CASP - Critical Assessment of Techniques for Protein Structure Prediction

• Started in 1994, Helped push the field of structure prediction•“Contest-like” setup•Catagories include:

•Homology Modeling / Comparative Modeling•Fold Recognition / Threading•Ab Initio, De novo•Partially vs. Automated Methods (now quite similar results)

Goal: Predict structures of solved but unpublished/unreleased structures (used to evaluate predictions. Every year, predictions / algorithms get better

Comparative Modeling “Homology Modeling”• Proteins that have similar sequences (i.e., related by evolution) are likely to have similar three-dimensional structures

1. BLAST sequence of Interest against PDB to identify a template•Multiple templates can be used if desired•Templates with Ligands bound can be used to identify binding sites and interacting residues in the homology model

Sequence identity required depends on protein length. A good rule of thumb is to have at least 40% sequence identity. Higher sequence identity is best. Lower than 25% is not reliable (zone of uncertainty)

Above 75% sequence identity, usually quite reliable homology model

Accurate sequence alignments very important

Programs include Modeller and Swiss Model

Comparative Modeling “Homology Modeling”

Steps include:1.Template recognition and initial alignment2.Alignment Correction (Multiple Sequence Alignment can Help)3.Backbone Generation (transfer coordinates from template)4.Loop Modeling (loops hard to predict with insertions)5.Side Chain Modeling (usually similar tortion angles at high sequenc ID)6.Model Optimization (minor energy minimization steps or restrain some atom positions)7.Model Validation (Higher ID more accurate usually, Calculate energy, or normality index (bond length, tortion angles))8.Iteration (to refine)

Protein Threading§ Generalization of homology modeling method• Homology Modeling: Align sequence to sequence• Threading: Align sequence to structure (templates)For each alignment, the probability that that each amino acid residue would occur in such an environment is calculated based on observed preferences in determined structures.§ Rationale:• Limited number of basic folds found in nature• Amino acid preferences for different structural environments provides sufficient information to choose the best-fitting protein fold (structure)

Protein Threading, Fold RecognitionOften, seemingly unrelated proteins adopt similar folds.-Divergent evolution, convergent evolution. For sequences with low or no sequence homology

Fold recognition• The number of possible protein structures/folds is limited (large number of sequences but relatively few folds (some estimate ~1000)) (most apparent when 50% of structures with no seq homology were solved and had folds similar to known structures) 90% of new structures deposited in PDB have similar folds to those already known

• Proteins that do not have similar sequences sometimes have similar three-dimensional structures (such as B-barrel TIM fold)

• A sequence whose structure is not known is fitted directly (or “threaded”) onto a known structure and the “goodness of fit” is evaluated using a discriminatory function

• Need ways to move model closer to the native structure

3.6 Å5% ID

NK-lysin (1nkl) Bacteriocin T102/as48 (1e68)

Ab initio prediction of protein structure – concept

Difficult because search space is huge. Much larger conformational space

Goal: Predict Structure only given its amino acid sequenceIn theory: Lowest Energy Conformation

• Go from sequence to structure by sampling the conformational space in a reasonable manner and select a native-like conformation using a good discrimination function

Difficult for sequences larger that 150aa

Rosetta (David Baker lab) one of best (CASP evaluation)

Rosetta structure prediction2 phases1.Low-resolution phase – statistical scoring function and fragment assembly

A. local structure conformations using info from PDB (3 and 9mer stretches)

B. multiple fragment substitution simulated annealing – to find best arrangement of the fragments (Monte Carlo Search)

C. low resolution ensemble of decoy conformations

2. Atomic refinement phase using rotamers and small backbone angle moves (in populated regions of Ramachandran plot)

A. RefinementB. Then structures clustered based on RMSD C. Center of the Largest Clusters chosen as

representative folds (likely to be correct fold)

Quality AssessmentRamachandran Plot – Phi Psi anglesTo identify residues that may be in wrong conformationProcheck, What_check