Diffusional Contributions and Electrostatic Interactions during Ultrafiltration Andrew L. Zydney Department Head and Walter L. Robb Family Chair Department of Chemical Engineering The Pennsylvania State University
Ultrafiltration Ultrafiltration is used extensively for protein concentration, buffer exchange, and desalting Formulation of therapeutic proteins Processing of food and beverage products Recent studies have demonstrated the potential of using ultrafiltration for protein separations based on different rates of protein transport across semi-permeable membranes
Whey Protein Purification Absorbance 1.6 1.4 1.2 1.0 0.8 0.6 Initial feed α-la β-lg Ultrafiltration at ph 5.5 50 mm phosphate Final retentate 1.2 1.0 β-lg 0.8 0.6 0.4 0.4 0.2 0.2 0.0 420 450 480 510 540 570 600 Electromigration time, sec Final filtrate 0.40 0.35 α-la 0.0 420 450 480 510 540 570 600 Electromigration time, sec Protein MW (kd) pi β-lactoglobulin 18 5.2 α-lactalbumin 14 4.6 0.30 0.25 0.20 0.15 0.10 420 450 480 510 540 570 600 Electromigration time, sec Cheang and Zydney, J Membrane Sci, 231, 159 (2004)
Monoclonal Antibody Purification from Harvested Cell Culture Fluid Absorbance! 0.04" 0.03" 0.02" 0.01" Harvested Cell Culture Fluid Retentate after 2-Step Filtration Large! Impurities!! Mab! BSA! Small! Impurities! Mab! 0" 0" 10" 20" 30" 40" Retention Time (min)! 50" 60 0" 10" 20" 30" 40" Retention Time (min)! 50" 60" Purification with two membrane separations: - Step 1: ph 7, 150 mm phosphate (Mab in permeate) - Step 2: ph 4, 10 mm acetate (Mab in retentate)
Membrane Transport Use of membranes for protein purification has generated renewed interest in understanding protein transport through semipermeable membranes Diffusion / convection in highly constricted pores, including effects of pore geometry Effects of electrostatic interactions (solution ph, ionic strength, membrane charge)
Stirred Cell Apparatus Air Pressure Amicon Stirred Cell The image cannot be displayed. Your computer may not have enough memory to open the image, or the image may have been corrupted. Restart your computer, and then open the file again. If the red x still appears, you may have to delete the image and then insert it again. Ultrafiltration Membrane Filtrate Filtrate Flux: J v = Q A Observed Sieving Coefficient: S o = C filtrate C bulk
Ultracel Membrane Skin Support structure Composite regenerated cellulose membrane SEM cross-section provided by Millipore
Sieving Coefficient vs Flux Observed Sieving Coefficient, S o 1.2 1.0 0.8 0.6 0.4 0.2 0 Bovine BSA - 100 serum kd albumin (69 kda) Omega membrane 100 kda membrane Filtrate Flux, J v (m/s) 10-7 10-6 S o = C filtrate C bulk 10-5 Sobs 10-4
Effect of Ionic Strength Observed Sieving Coefficient, S o 1 0 0 1 0-1 1 0-2 1 0-3 1 0-7 0.1 5 M 0.0 5 M 0.0 1 5 M 0.0 0 5 M e v i n g C o e 0.0 0 2 M Bovine Serum Albumin MW = 67 kd, pi =4.8 10-6 10-5 Filtrate Flux, J v (m/s) ph 7 1 0-4
Physical Phenomena Protein sieving coefficient varies by over two orders of magnitude Significant effects of both filtrate flux and solution ionic strength Protein sieving coefficient shows a distinct minimum with increasing filtrate flux, approaching a value of one at high flux Concentration polarization (bulk transport) Diffusion / convection through membrane
Concentration Polarization
Concentration Polarization Filtrate Flux, V Back-Diffusion Increasing Filtrate Flux C wall C filtrate Accumulation of retained protein at upstream surface of membrane Extent of polarization determined by balance between convection towards membrane and diffusion back into bulk solution VC " D # $C $z = VC filtrate C bulk Membrane
Actual Sieving Coefficient Actual Sieving Coefficient, S a 10 0 10-1 10-2 10-7!"K d = 0.0040 S = 0.037 BSA - 5 g/l 100 K membrane S a = C filtrate C wall 10-6 Filtrate Flux, J v 10-5 (m/s) 10-4
Membrane Transport Solute Flux: N s = K c VC pore " K d D # $C pore $z K c = hindrance factor for convection K d = hindrance factor for diffusion Partition Coefficient: " = C pore,z=0 C wall = C pore,z=# m C filtrate Assumes equilibrium at pore entrance (z = 0) and pore exit (z = δ m )
Actual Sieving Coefficient S a = C filtrate C wall = S " exp( Pe ) m S " + exp( Pe ) m #1 Asymptotic Sieving Coefficient: S " = #K c Membrane Peclet Number: Pe m = K c K d $ V" m & % D # ' ) (
Actual Sieving Coefficient Actual Sieving Coefficient, S a 10 0 10-1 10-2 "K c = 0.037!"K d = 0.0040 S = 0.037 "K d = 0.0040 BSA - 5 g/l 100 K membrane S a = S " exp Pe m 10-7 10-6 10-5 10-4 Filtrate Flux, J v (m/s) ( ) ( ) #1 S " + exp Pe m
Hindrance Coefficients Model curves based on hydrodynamic models for spherical solutes
Multi-layer Membranes Effects of diffusion / convection are significantly more complicated in multilayer and asymmetric membranes Internal concentration polarization develops when flow is through the more open (larger pore size) region first Net result is a directional dependence to solute transport / sieving coefficient
Directional Dependent Transport Tight side up Open side up
Actual Sieving Coefficient: Two-Layer Membrane Tight side up Open side up
Dextran Sieving Dual Skinned Hollow Fiber Membrane Observed Sieving Coefficient, S o Shell-to-Lumen Solid curves are model calculations with: r lumen = 2.6 nm r shell = 13 nm Lumen-to-Shell Dextran Molecular Weight (kda)
Starling Flow - Backfiltration Feed Hollow Fiber Membrane Pressure, P Axial Position, z
Electrostatic Interactions Observed Sieving Coefficient, S o 1 0 0 1 0-1 1 0-2 1 0-3 1 0-7 0.1 5 M 0.0 5 M 0.0 1 5 M 0.0 0 5 M e v i n g C o e 0.0 0 2 M Bovine Serum Albumin MW = 67 kd, pi =4.8 10-6 10-5 Filtrate Flux, J v (m/s) ph 7 1 0-4
Electrostatic Interactions! = C pore C solution = (1 " #) 2 $ exp - E % kt & ' - - - - - steric interactions electrostatic interactions - - - - - - -! = r solute r pore E kt = solute to pore size ratio = dimensionless energy of interaction
Energy of Interaction E kt = A 2 1! protein 2 + A 2! pore + A 3! protein! pore! protein = surface charge density of protein! pore = surface charge density of pore wall Three terms associated with: (1) distortion of protein double layer (2) distortion of double layer in pore (3) direct charge-charge interactions
Protein Sieving -- ph Effects Actual Sieving Coefficient, S a Bovine Serum Albumin 10 mm KCl 2 A 2 " pore 2 A 1 " protein A 3 " protein " pore Note: Model only includes partitioning effects Solution ph
Hindrance Factor - Diffusion τ = κb Hindrance Factor, K d b = pore radius κ -1 = Debye length Ratio of Solute to Pore Size, λ from Dechadilok and Deen, J Membrane Sci, 336, 7 (2009)
Charged Membranes Opportunity to improve performance of traditional ultrafiltration processes by increasing protein retention Opportunities for highly-selective protein separations
Membrane Charge Effects Sieving Coefficient, S o 1 0.1 0.01 Cytochrome c ph 7 Zeta Potential 0.3 mv 3.2 mv 4.2 mv 5.8 mv 6.6 mv 0.001 1 10 100 1000 Solution Ionic Strength (mm)
Membrane Charge Effects Sieving Coefficient, S o 1 0.1 0.01 0.001 Neutral Membrane Positive Membrane Antibody BSA (MW = 160 kd, pi = 9.0) (MW = 68 kd, pi = 4.8) ph 5.0, 10 mm acetate buffer
Charged UF Membranes Number of Charge Groups Surface density of ligands Ligands with multiple charge groups Location of Charge Groups Effect of spacer arm length / branching Nature of Charge Groups Weak versus strong acid / base Detailed ligand and linkage chemistry
Epichlorohydrin Activation OH + CH 2 ClCH CH 2 0.1 M NaOH 45 0 C OCH 2 CH CH 2 Hydroxyl groups on cellulose O Epichlorohydrin O Activated Membrane Degree of modification controlled by reaction time
Diamine Addition OH + CH 2 ClCH CH 2 O 0.1 M NaOH 45 0 C OCH 2 CH CH 2 + O H 2 N-R-NH 2 Diamine Multiple chemistries and ligand lengths
Charge - Modified Membrane OH + CH 2 ClCH CH 2 O 0.1 M NaOH 45 0 C OCH 2 CH CH 2 + O ph 11 OCH 2 CHCH 2 HN-R-NH 2 OH Charge-Modified Membrane 45 0 C H 2 N-R-NH 2
Ligand Chemistry 2 N 4 N (a) Diaminodecane (1.4 nm), 13 C (b) Diaminobutane (1.8 nm), 14 C Sequential reaction 2 N 6 N 4 N (c) Pentaethylenehexamine (1.8 nm), 13 C (d) Quaternary amine (1.7 nm), 14 C Sequential reaction (e) Short quaternary amine (0.4 nm), 5 C
Effects of Ligand Chemistry 6 N Data for cytochrome c at ph 7 2 N 4 N
Separation of Protein Variants Protein variants differ at only a single amino acid residue (e.g., deamidation, oxidation of methionine, carbamylation, etc) Resulting variants can have different activity and immunogenicity --> very challenging separation Example: myoglobin variants produced by chemical modification of a single lysine (conversion of amine to carboxylic acid)
Myoglobin Sequence GLSDGEWQQV!!LNVWGKVEAD! IAGHGQEVLI!!RLFTGHPETL! EKFDKFKHLK!!TEAEMKASED! LKKHGTVVLT!!ALGGILKKKG! HHEAELKPLA!!QSHATKHKIP! IKYLEFISDA!!IIHVLHSKHP! GDFGADAQGA!!MTKALELFRN! DIAAKYKELG!!FQG! Chemical modification of a single lysine amino acid out of 153 residues
Selectivity for Myoglobin Variants Variants differ at 1 out of 153 amino acids
Separation of Myoglobin Variants 0.8 Feed Solution Purified Product Negatively-charged membrane Absorbance (AU) 0.6 0.4 0.2 Variant Product Product 0 600 650 700 750600 650 700 750 Electromigration Time (s) Electromigration Time (s)
Summary Diffusive transport can significantly alter protein transmission during ultrafiltration Bulk mass transport (concentration polarization) Enhanced transmission at low filtrate flux Directional dependence for multilayer membranes Electrostatic interactions provide unique opportunities for enhanced performance Repulsive interactions cause strong rejection Role of ligand structure still being explored
Acknowledgements Penn State Amit Mehta (Genentech) Russell Boyd (Merck) Doug Burns (Talecris) Mareia Ebersold Mahsa Rohani Genentech Robert van Reis Millipore Baxter Healthcare Financial Support Millipore - Ultracel Membranes Walter L. Robb Family Chair