Enzyme Engineering Research Technology Development Enzymatic Catalysis Group

Enzyme Engineering Research & Technology Development Enzymatic Catalysis Group, PMC Advanced Technology

Overview of objectives Quantitative understanding of enzyme evolution (academic publications) • explaining origin of natural active site sequence distributions (benchmarking on MSA and pdb data) Redesign enzyme active sites (designer enzyme products) • modify substrate selectivity, product inhibition, etc • for industrial biocatalysis, biotechnology and biotherapeutics (with experiment) To advance the state-of-the-art in enzyme design technology (design software) • through the application of high-resolution physics-based methods for active site modeling using: 1) High-res protein structure prediction (OPLS + SGB): loop prediction for reshaping active sites, side chain optimization 2) Semiempirical enzyme-substrate binding affinity scoring (Km), substrate pose sampling 3) Refinement based on details of electronic structure: scoring activation energies (kcat)

Schematic of computational enzyme design technology Software Patents Design Protocol Patents

Zymzyne Enzyme Design and Optimization Platform Design Computationally Input information Target chemical Desired raw material Refine Experimentally System Output Zymzyne™ Computational Design Process ~1000 potential candidates expected catalytic activity Zymzyne™ Experimental Optimization Optimized Biocatalyst Existing synthetic pathways 1030 candidates screened 500 candidates screened Existing biocatalysts Software Patents Design Protocol Patents

A model fitness measure for enzyme sequence optimization substrate binding catalysis product release • Maximize free energy of substrate binding over sequence space § Represent catalysis through constraints on interatomic distances of catalytic side chains • Minimize total energy of complex for any sequence • To start, omit selection pressure for product release

Active site sequence optimization requires accurate energy functions, solvation models, and search algorithms 10 o resolution rotamer library (297 proteins) Xiang, Z. and Honig, B. (2001) J. Mol. Biol. 311: 421 -430.

Active site sequence optimization requires accurate energy functions, solvation models, and search algorithms S-GB continuum solvation 10 o resolution rotamer library (297 proteins) Xiang, Z. and Honig, B. (2001) J. Mol. Biol. 311: 421 -430. Ghosh, A. , Rapp, C. S. & Friesner, R. A. (1998) J. Phys Chem. B 102, 10983 -10990.

Active site sequence optimization requires accurate energy functions, solvation models, and search algorithms S-GB continuum solvation 10 o resolution rotamer library (297 proteins) Xiang, Z. and Honig, B. (2001) J. Mol. Biol. 311: 421 -430. Ghosh, A. , Rapp, C. S. & Friesner, R. A. (1998) J. Phys Chem. B 102, 10983 -10990. OPLS-AA molecular mechanics force field + Glidescore semiempirical binding affinity scoring function Friesner, R. A, Banks, J. L. , Murphy, R. B. , Halgren, T. A. et al. (2004) J. Med. Chem. 47, 1739 -1749. Jacobson, M. P. , Kaminski, G. A. Rapp, C. S. & Friesner, R. A. (2002) J. Phys. Chem. B 106, 11673 -11680.

φ, ψ = the backbone torsion angles Backbone = the sequence of (COOH)-[N-(CH-Ri)-(C=O)]N-NH 2 , where Ri is the i'th side chain. 2 N torsion angles specify the backbone configuration. Side-chains have their own rotamers too! These angles are represented by χi. Some side chains have no χ angles. Some have quite a few, such as the lysine above with χ1 -χ4.

Computational sequence optimization correctly predicts most residues in ligand-binding sites… Streptavidin Native – 10. 04 kcal/mol Chakrabarti, R. , Klibanov, A. M. and Friesner, R. A. Computational prediction of native protein ligand-binding and enzyme active site sequences. PNAS, 2005.

Computational sequence optimization correctly predicts most residues in ligand-binding sites… Streptavidin Native – 10. 04 kcal/mol Chakrabarti, R. , Klibanov, A. M. and Friesner, R. A. Computational prediction of native protein ligand-binding and enzyme active site sequences. PNAS, 2005.

Computational sequence optimization correctly predicts most residues in ligand-binding sites… Streptavidin Native – 10. 04 kcal/mol CO 2 - is covalent attachment site for biomolecules 9 / 10 residues predicted correctly in top 0. 5 kcal/mol of sequences Easy to exptly screen libraries of this size Chakrabarti, R. , Klibanov, A. M. and Friesner, R. A. Computational prediction of native protein ligand-binding and enzyme active site sequences. PNAS, 2005.

…and enzyme active sites R 61 DD-peptidase Native – 10. 02 kcal/mol

…and enzyme active sites R 61 DD-peptidase Native – 10. 02 kcal/mol High MSA variability T 123 highly degenerate in multiple sequence alignment

Computational enzyme sequence optimization: sugar catalysis b-galactosidase kcal/mol Native – 9. 13 Computed • Native amino acid is generally one of top 3 most frequently predicted • Could be used to focus combinatorial libraries (3 N vs 20 N, N = # of residues)

Computed amino acid distributions contain detailed evolutionary information Glucose-binding protein kcal/mol Native – 8. 81

Computed amino acid distributions contain detailed evolutionary information Glucose-binding protein kcal/mol Native – 8. 81 Observed (sequence alignment)

Computed amino acid distributions contain detailed evolutionary information Glucose-binding protein Native – 8. 81 kcal/mol Epimeric promiscuity Observed (sequence alignment) Anomeric promiscuity Computed OH OH • Computed residue frequencies often mirror natural frequencies

Summary of recent results: classical sequence optimization (Side chain prediction/ Binding affinity calculation / Sequence opt) Acid/base Y 159 Electrostatic stabilizer Lys 65 Nucleophile Ser 62 R 61 DD-peptidase T 123 highly degenerate in multiple sequence alignment

Computed amino acid distributions contain detailed evolutionary information Glucose-binding protein Native – 8. 81 kcal/mol Epimeric promiscuity Observed (sequence alignment) Anomeric promiscuity Computed OH OH • Computed residue frequencies often mirror natural frequencies

High-resolution sequence optimization is robust across diverse functional families Peptide Nucleotide Sugar

Active Site Design of Enzymes with Nucleotide Substrates: Cytidine Kinase

Multisubstrate enzyme active site sequences represent superpositions of computational predictions d. TMP HSV-1 thymidine kinase

Multisubstrate enzyme active site sequences represent superpositions of computational predictions d. TMP Ganciclovir (d. G analog)

Multisubstrate enzyme active site sequences represent superpositions of computational predictions d. TMP Ganciclovir (d. G analog) Thymidine Apply multiobjective sequence search algorithms to accommodate several substrates

Multisubstrate enzyme active site sequences represent superpositions of computational predictions d. TMP Ganciclovir (d. G analog) Native sequence = superposition of optimal sequences for multiple substrates Thymidine

Catalytic hydrogen-bonding networks can be incorporated into sequence optimization GLU 272 LYS 315 Cephalothin W 402 b-Lactamase : cephalothin b GLU 272 d c a SER 62 e g f LYS 67 h TYR 150 ARG 148 ASN 152 GLN 120

Catalytic hydrogen-bonding networks can be incorporated into sequence optimization +2 kcal/mol +1 kcal/mol Constrained + Filtered LYS 315 Cephalothin W 402 b GLU 272 d c a SER 62 e g f LYS 67 h TYR 150 ARG 148 ASN 152 GLN 120 Chakrabarti, R. , Klibanov, A. M. and Friesner, R. A. Sequence optimization and designability of enzyme active sites. PNAS, 2005.

Refining the scoring function: quantum chemical transition state calculations Enzyme WT kcat (s-1) KM (μM) kcat/KM (% Wild-type) 150 14 100 N 152 S 3 7 4. 3 N 152 D 0. 12 24 0. 05 N 152 S/Q 120 F 3 4. 6 6. 7 N 152 S/Q 120 H 20 11. 4 16. 3 Predicted 14. 3 kcal/mol Measured 14. 3 kcal/mol

Active Site Designability: The Number of Sequences that Solve a Given Design Problem +2 kcal/mol + 1 kcal/mol Constrained Number of residues correctly predicted General acid/base Y 159 Electrostatic stabilizer Lys 65 Catalytic nucleophile Glu-299 Catalytic Nucleophile Ser 62 DD-peptidase General acid/base Glu-200 b-gal

Patents: computational sequence optimization / experimental mutagenesis Example of screening focused library of sequence variants 3 permissible mutations identified by modeling at a target position 3 positions subject to mutagenesis 43 mutation combinations = 64 sequence variations Synthetic gene assembly and variant library construction via DNA synthesis Biological selection of variant library New enzymes Improved catalytic turnover Altered substrate selectivity

Patents: algorithms in development Protein structure Substrate binding Loop Sidechain Glidescore Pose sampling New algorithms for side chain optimization QM sequence refinement Classical Sequence Optimization (fixed ligand) Active site reshaping • scores desired loop against other low-energy excitations Reactive chemistry • for QM/MM refinement Calculating mutant enzyme of enzyme design • speeding up mutant reaction rates TS searches Classical Sequence Optimization (free ligand) • Hierarchical pose screening • Locates global seq/struct optima for a given active site/ligand comb • Estimates “designability” of active site (fixed backbone)

Testing: Current experimental projects Exptl project Methods applied Notes 1) Accommodate NAD+ Sirtuin redesign for enhanced activity 1) Active site backbone reshaping, multiobjective genetic and monte carlo sequence search 2) Reduce binding affinity to NAM (reaction product) to reduce product inhibition 2) Selection via in vivo complementation 3) In vitro kinetics of engineered enzymes Mutant activation barrier predictions in PBP b-lactamase 1) Side chain structure prediction + QM/MM activation barrier calculation 2) In vitro kinetics of mutants: compare Kd and kcat to computed values 1) To establish foundation for computational refinement of activity 2) Basis for future work on rapid algorithms for QM refinement of enzyme design

Discussion Points • NEB combinatorial screening protocols • NEB DNA enzyme engineering challenge problems • Scope for interaction: – Technology Platform to be used by both parties? – Engineered DNA Enzyme Products? Cosolvent-resistant polymerases? – IP: Software, Designability-Based Screening Protocols (compare Maxygen, Diversa), and Engineered Enzymes

Sirtuin – mutant production, selection, protein expression and enzyme assay Main objective - Develop genetic and biochemical assay systems to screen sirtuin mutant library and quantify enzymatic activities. Main steps - q q q Model mutations in the active site residues of bacterial sir 2 Tm. Generate a set of mutations using wild-type sirtuin as template based on computation-guided structural modeling. Transform the mutants into host strains with sirtuin deletion. Assay growth of mutant transformants under carbon source limitation. Select mutant constructs which can complement the growth defects resulted from sirtuin deficiency, which are manifested under carbon limitation. Purify the wild-type and active mutant enzymes and quantify their kinetic properties.

Sirtuin – mutant generation Model mutations in the active sites of sirtuin genes by computational analysis. Construct mutations in the wild-type sir 2 Tm plasmid (2 potential methods) By synthetic gene method – Generate sequence map for proposed nucleotide changes in the wild-type template (sir 2 Tm). Work with gene synthesis groups to make synthetic constructs for the mutant collection, e. g. how to get efficient oligo assembly to cover all the mutations. Obtain suitable plasmid vectors and clone the mutant constructs into the vectors. The vectors would depend on the host cells in which the mutant constructs would be expressed and selected, e. g. yeast, salmonella have different vectors to allow highlevel expression. By multi-site directed mutagenesis method – Use reagents including cells, enzymes and mutagenic primers to generate mutation in the wild-type sir. Tm template. Verify mutations by DNA sequencing. Both procedures for mutagenesis depend on the actual mutations to be made and how many constructs are needed to allow for effective functional screening.

Sirtuin – mutant library screening assay When the mutant collection is generated, transform the constructs into host cell with sirtuin deletion. Make competent cells for the host strain so they can take up DNA. Transform the wild-type plasmid into host as positive control. Transform the mutant plasmids into host cells. Assay whether the transformants could grow on carbon-limited media, such as with acetate or propionate as sources. If there is complementation, characterize the growth features of these cells. Verify the specific mutations by DNA sequencing. Transform the mutant construct into protein-expression host, such as Ecoli BL 21. Grow cultures and purify sufficient quantities of proteins. Set up enzymatic assays to quantify kinetic properties of wild-type and selected mutants.

Beta-lactamase – mutant selection, protein expression and activity assay q Model mutations in the active site residues of P 99 beta-lactamase. q Construct mutations in the wild-type P 99 beta-lactamase gene. q Obtain bacterial host strains suitable for screening beta-lactam antibiotic resistance. q Transform bacteria host cells with wild-type and mutant constructs. q Select transformed cells in the presence of beta-lactam antibiotics. q Identify the mutant clones which can grow in beta-lactam and thus retain beta -lactamase activities. q Express and purify the wild-type and mutant beta-lactamases and quantify their kinetic properties.

Beta-lactamase – mutant generation Model mutations in the active site of P 99 beta-lactamase based on computation. Construct mutations in the wild-type P 99 beta-lactamase plasmid. The actual processes would depend on what the mutations are and how many mutants are to be made. By synthetic gene method – Work with gene synthesis group to construct synthetic constructs, esp. in how to set up efficient oligonucleotides coverage for all the mutations. Clone all mutant constructs into suitable bacterial expression vector. By multi-site directed mutagenesis method Need to obtain mutagenic reagents such as cells, enzymes and primers to generate a set of mutations. Verify mutant production by DNA sequencing of individual clones.

Beta-lactamase – mutant selection With the bacterial host strains used for selection, make competent cells so that they can take up plasmid DNA. Transform wild-type P 99 beta-lactamase plasmid into host cells as positive control. Transform the mutant plasmids into host cells to select for active constructs. Make agar plates containing different types of beta-lactam compounds and at different concentration. Grow bacteria transformed with beta-lactamase plasmids on these plates and monitor colony formation. Identify the clones with good growth characteristics so they would be the candidates to provide hydrolytic activities on a variety of beta-lactam substrates. Verify specific mutations by DNA sequencing. Proceed to protein expression, purification and activity quantitation.

A General Framework for Computationally Directed Biocatalyst Design slack variable Enzyme-substrate binding affinity Catalytic constraint: interatomic distances rij < hbond dist • Minimize J over sequence space • Represent dynamical constraint with requirement that total energy of complex minimized for any sequence • Omits selection pressure for product release

Assessment of active site designability § Need to assess number of sequences that are structurally similar to native § Requires sampling over ligand conformations

Computationally directed active site sequence library generation Two approaches: § Marginal distributions (as shown) using top m (m constant) as shown or setting m_i according to exp(shannon entropy). Choose T based on exptl tractability. Assumes independence, but easier for exptlst to implement out-of-box. Note S in this case cannot be interpreted as number of microstates since LLN does not hold b) Joint distribution: sample m sequences from joint distribution for specified T’s. S computed based on moments of objectives. Compare D=exp(S) for several T’s, look for transition to region where denser sampling possible (heat capacity analogy). LLN holds, allowing interpretation of designability as relative number of microstates

Computed sequence entropies suggest equilibrium in sequence space Comparable Shannon site entropies suggest equilibrium for same fitness measure and provide concise comparison of distributions at all positions (rather than showing pdf at each position) Shannon sequence entropy: Si = - S(a=1. . . 20) [f(ia) ln f(ia)] Computed Catalytic constraint Observed Penicillium sp. b-galactosidase

Marginal active site sequence distributions Shannon site entropies: Computed based on marginal distributions; unlike joint cannot be expressed in closed form in terms of exp fn. Two approaches to estimating distribution – a) in terms of marginal moments of functions of f_i’s; b) in terms of explicit f_i’s (used here). Both based on drawing m samples from joint Extensions/modifications to PNAS paper figures: Better to display K-L relative entropies rather than site entropies for marginal distributions at each position Instead compare K-L relative entropies (joint distribution) wrt MSAs for models w different objectives, on same plot; alternatively use approach based on marginal distributions on Shannon entropy slide For such figures, compare K-L rel entropies (here marginal)

Plan for development of designability theory and experimental application (to be described in conclusion of our early papers) § Apply designability theory to all major enzyme families from PNAS papers; extend to designability of modified sirtuins experimentally § Could id the catalytic constraints and focus on objective for reducing inhibition (NAM binding affinity); estimate latter temperature. § Compare designability of NAD site to that of other enzyme classes studied, for same T’s. Check designability at lower T for NAM inhibition § Designability approach will help determine viability of drug development efforts more effectively than comb chem

Components of energy function Surface-area term Covalent bond potential Torsional potential H-bonding (sometimes) H 2 O H 2 H OO HO 22 HO O 2 H HO 22 OH OH O H H 22 O 2 H HOHOO 2 2 2 H H 22 O 2 H OHOHOO H 2 OH 2 HO 2 O 2 H 2 O 2 2 Non-bonding terms (Van Der Waals) The effect of water (a rude fellow!) Electrostatic potential http: //tinyurl. com/63 gt 3 lm

Computational active site optimization is structurally accurate to near-crystallographic resolution

Future plans • Understanding differences between PLOP/Glide/Qsite energies for summing energy calcs to calculate Km, kcat • Modeling the denatured state of proteins to estimate folding free energy for core sequence optimization Integration with other current developments • Induced fit + Backbone reshaping to start with globally-relaxed backbone shapes for unnatural ligands • MD treatment of loops + Backbone reshaping + Classical affinity opt for antibody engineering