SAVE THE DATE: May 16 th 17 th, 2018 NIIMBL National Meeting.

Size: px

Start display at page:

Download "SAVE THE DATE: May 16 th 17 th, 2018 NIIMBL National Meeting."

Austin Summers
5 years ago
Views:

Approaches and challenges for quantifying multi-body interactions as higher-order structure for predicting product attributes Chris Roberts Professor, Chemical & Biomolecular Engineering, University

1 Approaches and challenges for quantifying multi-body interactions as higher-order structure for predicting product attributes Chris Roberts Professor, Chemical & Biomolecular Engineering, University of Delaware Associate Institute Director, NIIMBL Director, Biomolecular Interaction Technologies Center Acknowledgements: Cesar Calero-Rubio (formerly UD) Mahlet Woldeyes (UD) Glenn Ferreira (UD) Chris O Brien (formerly UD) Ranendu Ghosh (formerly UD) Eric Furst (UD) Atul Saluja (formerly BMS) Vladimir Razinkov (Amgen) Wei Qi (Amgen) Rick Remmele (RemSciPharma) Anne Robinson (Tulane) Funding: Amgen, Bristol-Myers Squibb, MedImmune, NSF, NIH, BITC

2 SAVE THE DATE: May 16 th 17 th, 2018 NIIMBL National Meeting

3 SAVE THE DATE: January 6 th 11 th, 2019 Inaugural GRC-Biotherapeutics and Vaccines Development

$5 x 10 5 Da) Only small fraction of structure is active Protein Conc n ~ 10 2 g/l Net: large doses needed: ~ 1 mg protein : kg of patient body wt.$

4 Context: Protein formulation, delivery, & selection 0.15 Monoclonal Antibody ê Solubility [NaCl] / M Agg n? Chem. Deg n ph Very large molecules (~ 1.5 x 10 5 Da) Only small fraction of structure is active Protein Conc n ~ 10 2 g/l Net: large doses needed: ~ 1 mg protein : kg of patient body wt. Market share à Patient self administration /0,0,171,20469,640,433,83b95ef1.jpg

5 Protein-protein interactions and phase behavior W 22 = f(r,ω) K D ~ M B 22 / B 22,Steric 1 0 WEAKLY ATTRACTIVE fl_realurl_image/proteinkristallisation-bieten-01-pr.jpg -5 to -10 K D ~ M Dumetz et al., Biophys. Journal (2008)

6 Protein-Protein Interactions (PPI) influence a number of stages in aggregation pathways 200 nm 50 nm 200 nm 50 nm Weiss WF IV, Hodgdon T, et al. Biophys J (2007). Roberts CJ, Curr. Opin. Biotech. (2014).

7 Viscosity and weak interactions / cluster formation Viscosity has shown correlations with weak interactions (k d ) (?) Connolly, B.; et. al. Biophys. J. 2012, 103, Sharma, V. K.; et. al. Proc. Natl. Acad. Sci. 2014, 111, If one is in the dilute limit, k D = h 0 + 2B 22

8 B 22 and k D are often used interchangeably to capture colloidal interactions If one is in the dilute limit, theoretically k D = h 0 + 2B 22 Electrostatic interactions can be repulsive or attractive, but it is difficult to actually predict when this will occur Roberts D, Keeling R, Tracka M, van der Walle C, Uddin S, Warwicker J, Curtis R. Mol. Pharm. 2014

9 1-1 Interactions: Spanning the scales from weak to strong interactions K D ~ nanomolar spanning to Kd ~ mm or higher conc For strongly attractive conditions, Kd values scale ~ -1/B 22 K D ~ micromolar è B 22 or k D ~ 1000 ml/g At low conc, G 22 = -2B 22 (See next few slides)

10 Multi-body interactions: G 22 vs. B weak interactions & concentration effects Kirkwood and Buff, J. Chem. Phys (1951) B 22 f([protein]); dilute solution weak interac n G 22 = f([protein]); any conc. weak / strong interac n Kirkwood-Buff Solution Theory EXACT, but not originally developed for proteins

11 At low concentrations, these are equivalent measures Dynamic Light Scattering D c (q 0,c 2 ) = D 0 H (q 0,c 2 ) 1+ c 2 G 22 Equilibrium AUC µ 2 c 2 = T,V,µ j 2 Neutron / x-ray small-angle scattering I(q 0) ~ c 2 (1+ c 2 G 22 ) kt c 2 (1+ c 2 G 22 ) Do not need to be in the dilute limit G 22 is valid at any c 2 (for a 1-phase system) S 0 = 1+c 2 G 22

12 Experimental protein aggregation (SEC) from low to high concentration ph 6.5 ph 5 k obs = rate coeff. for monomer loss via SEC Ghosh, R., et. al. J. Pharm. Sci. 2016

13 Protein aggregation and weak interactions Ghosh, R., et. al. J. Pharm. Sci More attractive / less repulsive conditions (higher ionic strength) have higher aggregation rates

Protein-Protein weak interactions quantified with Rayleigh

osmotic 2 nd coefficient (independent of c 2 ) G 22 =

Rayleigh profiles ex R 90 K = M 2,appc 2 + M 2 G (m) 2 22 c 2

2 lim c 2 0G 22 If one uses k D k D = ω + 2B 22 Subindeces: 1

1] dv V Could be obtained from molecular simulations Ghosh, R.

14 Protein-Protein weak interactions quantified with Rayleigh scattering c 2 = protein concentration B 22 = protein-protein osmotic 2 nd coefficient (independent of c 2 ) G 22 = protein-protein KB integral (depends on c 2 ) Experimental Rayleigh profiles ex R 90 K = M 2,appc 2 + M 2 G (m) 2 22 c 2 (m) S q=0 = 1+ c 2 G 22 M 2,app M 2 If one uses B 22 = 1 B 22 2 lim c 2 0G 22 If one uses k D k D = ω + 2B 22 Subindeces: 1 - Solvent (water) 2 Protein 3 & on - Excipients G 22 = [g(r) 1] dv V Could be obtained from molecular simulations Ghosh, R., et. al. J. Pharm. Sci Blanco, M., et. al. J. Chem. Phys. 2011, 134(22),

15 Mayer Sampling with Overlap Sampling (MSOS): B 22 Ø Open 2-particle system (N=2, V, T) Ø Molecular simulations for protein interactions low protein concentrations Mayer functions integrated with respect to a known reference: γ 22 π B 22 = B π / γ 22 22,ref ref γ 22 π π,ref / γ 22 over over π π π π,ref Rubio, C.C. et al. J. Phys. Chem B 2016 Shaul, K.R.S., et. al. J. Chem. Phys Errington, J.R. J. Chem. Phys Ben-Naim, A. Statisitical Thermodynamics for Chemist and Biochemists Plenum Press. 1992

Reweighting μ S q=0 = f(protein conc.) Rubio, C.C. et al. J. Phys. Chem B 2016 Shaul, K.R.S., et. al. J. Chem. Phys. 2011 Errington, J.

16 Molecular simulations for protein interactions High concentrations Transition matrix Monte Carlo (TMMC): G 22 Ø Grand-canonical ensemble Ø Pr(N μvt) is reconstructed and reweighted µ 2 N 2 T,V,µ' = kt ( ) N 2 1+ N 2 G (m) 22 /V Reweighting μ S q=0 = f(protein conc.) Rubio, C.C. et al. J. Phys. Chem B 2016 Shaul, K.R.S., et. al. J. Chem. Phys Errington, J.R. J. Chem. Phys Ben-Naim, A. Statisitical Thermodynamics for Chemist and Biochemists Plenum Press. 1992

17 Different ranges of CG models Calero-Rubio et al. J Phys Chem B 2016

18 Coarse-graining: balancing computational cost & accuracy Same volume, different packing n 2 scaling DODECA HEXA Asymptotic behavior towards all-atom result. HEXA and DODECA balance accuracy + speed Calero-Rubio et al. J Phys Chem B 2016

Overview of interaction model (illustrated with HEXA example) Q Fv Q C1 Sterics: Hard-sphere potential Short-range: Modified Lennard-Jones potential Electrostatics: Modified screened-coulomb

19 Overview of interaction model (illustrated with HEXA example) Q Fv Q C1 Sterics: Hard-sphere potential Short-range: Modified Lennard-Jones potential Electrostatics: Modified screened-coulomb potential Q CH2 Hinge: Harmonic potential Q CH3 Bead size: σ ST (nm) Ionic strength: I (mm) Bead hydrophobicity: ε SR (k B T) Structural INPUT: 1 sequence 2 3d struc if available Calero-Rubio et al. J Phys Chem B 2016

determined Q CH2 Ø Bead-bead distances: crystal structure This leaves two parameters to tune: Q CH3 1.

20 Selecting physically realistic parameters Q Fv Q C1 Ø σ ST : matching B 22,ST from all-atom simulations Ø Charges (Q i ): theoretical charge (sequence + ph) Ø Flexibility: rigid molecule (speed reasons) Ø Ionic strength: experimentally determined Q CH2 Ø Bead-bead distances: crystal structure This leaves two parameters to tune: Q CH3 1. Bead hydrophobicity / van der Waals attractions: well depth, ε SR 2. Effective charge/theoretical charge: ε cc Calero-Rubio et al. J Phys Chem B 2016

21 Training the model: B 22 vs total ionic strength (TIS) ph 5.0 Buffer only Buffer + 5% sucrose Calero-Rubio et al. J. Pharm. Sci. (2018)

22 Training the model: B 22 vs total ionic strength (TIS) ph 6.5 Buffer + 5% sucrose Buffer only Calero-Rubio et al. J. Pharm. Sci. (2018)

23 Predicting high c 2 with only low-c 2 data ph 5.0 Lines are predictions of high c 2 behavior based on low c 2 behavior Higher c 2 are achievable with fancier sampling algorithm Calero-Rubio et al. J. Pharm. Sci. (2018)

24 Predicting high c 2 with only low-c 2 data ph 6.5 Repulsive conditions: easier to model, more accurate (faster convergence) Attractive conditions: required more configurations (slower to converge) Calero-Rubio et al. J. Pharm. Sci. (2018)

25 What about when electrostatic interactions are strongly attractive? Manuscript in preparation; please contact C. Roberts for requests for preprints once they have been cleared internally

26 Surface charge distributions attractive dipole and higher multi-pole contributions can dominate Manuscript in preparation; please contact C. Roberts for requests for preprints once they have been cleared internally

27 Predictions of high concentration interactions from B 22 and MC simulations Manuscript in preparation; please contact C. Roberts (cjr@udel.edu) for requests for preprints once they have been cleared internally

28 What about higher resolution CG models? Calero-Rubio et al. J Phys Chem B 2016

29 B 22 response surfaces: potential developability index Manuscript in preparation; please contact C. Roberts for requests for preprints once they have been cleared internally

30 Single Chain Variable Fragments (scfv) can pose a different challenge Static light scattering ph 7, 10 mm NaPhos Equilibrium AUC ph 7, 10 mm NaPhos Dimer-monomer mix Dimer-monomer mix Mw ~ 28 kda Expected monomer LS and AUC data to show dimerization at typical formulation ph / ionic strength; but reverts to monomer at higher TIS O Brien, Calero-Rubio et al., Protein Science (2018)

31 Single Chain Variable Fragments (scfv) can pose a different challenge Equilbrium AUC ph 7, buffer mm NaCl Equilibrium AUC ph 7, 10 mm NaPhos Dimer-monomer mix Expected monomer Expected monomer LS and AUC data to show dimerization at typical formulation ph / ionic strength; but reverts to monomer at higher TIS O Brien, Calero-Rubio et al., Protein Science (2018)

32 scfv has a net neutral but charged linker, with multiple possible configurations O Brien, Calero-Rubio et al., Protein Science (2018)

33 The flexible linker is able to cause strong electrostatically driven attractions between the linker and the VH and VL domains O Brien, Calero-Rubio et al., Protein Science (2018)

34 Viscosity and weak interactions How predictive are these experimental or simulated interactions of potential problems with high viscosity? Current practice = assume B 22 (~ k D ) correlates with high viscosity Protein Conc n ~ 10 2 g/l Connolly, B.; et. al. Biophys. J. 2012, 103 (1), Sharma, V. K.; et. al. Proc. Natl. Acad. Sci. 2014, 111 (52), /0,0,171,20469,640,433,83b95ef1.jpg

35 IgG candidates can display a wide range of electrostatic attractive / repulsive behavior Manuscript under review; please contact C. Roberts (cjr@udel.edu) for requests for preprints once they have been approved by the journal Woldeyes, Razinkov, Qi, Battistoni, Furst, Roberts under review

36 Summary Funding: Amgen, BMS, NIH, NSF, BITC

Protein-Protein Interaction Measurement: Balancing Complete Characterization with High- Throughput Capability in Formulation Development

Protein-Protein Interaction Measurement: Balancing Complete Characterization with High- Throughput Capability in Formulation Development Introduction High throughput screening in the world of pharmaceutical