Preparative Chromatography for Separation of Proteins E-Book

• Herausgeber: John Wiley & Sons
• Kategorie: Wissenschaft und neue Technologien
• Serie: Wiley Series on Biotechnology
• Sprache: Englisch

Seitenzahl: 914

Veröffentlichungsjahr: 2017

Beschreibung

Preparative Chromatography for Separation of Proteins addresses a wide range of modeling, techniques, strategies, and case studies of industrial separation of proteins and peptides.

•    Covers broad aspects of preparative chromatography with a unique combination of academic and industrial perspectives
•    Presents Combines modeling with compliantce useing of Quality-by-Design (QbD) approaches including modeling
•    Features a variety of chromatographic case studies not readily accessible to the general public
•    Represents an essential reference resource for academic, industrial, and pharmaceutical researchers

Leseprobe

Table of Contents

Cover

Title Page

List of Contributors

Series Preface

Preface

1 Model‐Based Preparative Chromatography Process Development in the QbD Paradigm

1.1 Motivation

1.2 Regulatory Context of Preparative Chromatography and Process Understanding

1.3 Application of Mathematical Modeling to Preparative Chromatography

Acknowledgements

References

2 Adsorption Isotherms

2.1 Introduction

2.2 Definitions

2.3 The Solute Velocity Model

2.4 Introduction to the Theory of Equilibrium

2.5 Association Equilibria

2.6 The Classical Adsorption Isotherm

2.7 The Classical Ion Exchange Adsorption Isotherm

2.8 Hydrophobic Adsorbents, HIC and RPC

2.9 Protein–Protein Association and Adsorption Isotherms

2.10 The Adsorption Isotherm of a GLP‐1 Analogue

2.11 Concluding Remarks

Appendix 2.A Classical Thermodynamics

References

3 Simulation of Process Chromatography

3.1 Introduction

3.2 Simulation‐Based Prediction of Chromatographic Processes

3.3 Numerical Methods for Chromatography Simulation

3.4 Simulation‐Based Model Calibration and Parameter Estimation

3.5 Simulation‐Based Parametric Analysis of Chromatography

3.6 Simulation‐Based Optimization of Process Chromatography

3.7 Summary

Acknowledgement

References

4 Simplified Methods Based on Mechanistic Models for Understanding and Designing Chromatography Processes for Proteins and Other Biological Products‐Yamamoto Models and Yamamoto Approach

4.1 Introduction

4.2 HETP and Related Variables in Isocratic Elution

4.3 Linear Gradient Elution (LGE)

4.4 Applications of the Model

4.5 Summary

Appendix 4.A Mechanistic Models for Chromatography

Appendix 4.B Distribution Coefficient and Binding Sites [20]

References

5 Development of Continuous Capture Steps in Bioprocess Applications

5.1 Introduction

5.2 Economic Rationale for Continuous Processing

5.3 Developing a Continuous Capture Step

5.4 The Operation of MCC Systems

5.5 Modeling MCC Operation

5.6 Processing Bioreactor Feeds on a Capture MCC

5.7 The Future of MCC

References

6 Computational Modeling in Bioprocess Development

6.1 Linkage of Chromatographic Thermodynamics (Affinity, Kinetics, and Capacity)

6.2 Binding Maps and Coarse‐Grained Modeling

6.3 QSPR for Either Classification or Quantification Prediction

6.4 All Atoms MD Simulations for Free Solution Studies and Surfaces

6.5 Ensemble Average and Comparison of Binding of Different Proteins in Chromatographic Systems

6.6 Antibody Homology Modeling and Bioprocess Development

6.7 Summary of Gaps and Future State

Acknowledgment

References

7 Chromatographic Scale‐Up on a Volume Basis

7.1 Introduction

7.2 Theoretical Background

7.3 Proof of Concept Examples

7.4 Design Applications: How to Scale up from Development Data

7.5 Discussion

7.6 Recommendations

References

8 Scaling Up Industrial Protein Chromatography

8.1 Introduction

8.2 Packing Quality: Why and How to Ensure Column Packing Quality Across Scales

8.3 Process Equipment: Using CFD to Describe Effects of Equipment Design on Column Performance

8.4 Long‐Term Column Operation at Scale: Impact of Resin Lot‐to‐Lot Variability

8.5 Closing Remarks

References

9 High‐Throughput Process Development

9.1 Introduction to High‐Throughput Process Development in Chromatography

9.2 Process Development Approaches

9.3 Case Descriptions

9.4 Future Directions

References

10 High‐Throughput Column Chromatography Performed on Liquid Handling Stations

10.1 Introduction

10.2 Chromatographic Methods

10.3 Results and Discussion

10.4 Summary and Conclusion

Acknowledgements

References

11 Lab‐Scale Development of Chromatography Processes

11.1 Introduction

11.2 Methodology and Proposed Workflow

11.3 Conclusions

Acknowledgments

References

12 Problem Solving by Using Modeling

12.1 Introduction

12.2 Theory

12.3 Materials and Methods

12.4 Determination of Model Parameters

12.5 Optimization

In Silico

12.6 Extra‐Column Effects

Abbreviations

References

13 Modeling Preparative Cation Exchange Chromatography of Monoclonal Antibodies

13.1 Introduction

13.2 Theory

13.3 Model Development

13.4 Model Application

13.5 Conclusions

13.6 Acknowledgments

References

14 Model‐Based Process Development in the Biopharmaceutical Industry

14.1 Introduction

14.2 Molecule—FVIII

14.3 Overall Process Design

14.4 Use of Mathematical Models to Ensure Process Robustness

14.5 Experimental Design of Verification Experiments

14.6 Discussion

14.7 Conclusion

Acknowledgements

Appendix 14.A Practical MATLAB Guideline to SEC

Appendix 14.B Derivation of Models Used for Column Simulations

References

15 Dynamic Simulations as a Predictive Model for a Multicolumn Chromatography Separation

15.1 Introduction

15.2 BioSMB Technology

15.3 Protein A Model Description

15.4 Fitting the Model Parameters

15.5 Case Studies

15.6 Results for Continuous Chromatography

15.7 Conclusions

References

16 Chemometrics Applications in Process Chromatography

16.1 Introduction

16.2 Data Types

16.3 Data Preprocessing

16.4 Modeling Approaches

16.5 Case Studies of Use of Chemometrics in Process Chromatography

16.6 Guidance on Performing MVDA

References

17 Mid‐UV Protein Absorption Spectra and Partial Least Squares Regression as Screening and PAT Tool

17.1 Introduction

17.2 Mid‐UV Protein Absorption Spectra and Partial Least Squares Regression

17.3 Spectral Similarity and Prediction Precision

17.4 Application as a Screening Tool: Analytics for High‐Throughput Experiments

17.5 Application as a PAT Tool: Selective In‐line Quantification and Real‐Time Pooling

17.6 Case Studies

17.7 Conclusion and Outlook

References

18 Recent Progress Toward More Sustainable Biomanufacturing

18.1 Introduction

18.2 The Impact of Individualized Unit Operations versus Integrated Platform Technologies on Sustainable Manufacturing

18.3 Implications of Recycling and Reuse in Downstream Processing of Protein Products Generated by Biotechnological Processes: General Considerations

18.4 Metrics and Valorization Methods to Assess Process Sustainability

18.5 Conclusions and Perspectives

Acknowledgment

References

Index

End User License Agreement

List of Tables

Chapter 02

Table 2.1 Parameters for a GLP‐1 derivative and six impurities estimated from the isocratic retention measurements on a Source 30 Q adsorbent.

Table 2.2 Estimated parameters in

F

(

c

), Equation 2.59.

Table 2.3 Isocratic retention parameters for three insulin variants on Toyopearl Butyl‐650S and Toyopearl Phenyl‐650S adsorbents.

Table 2.4 The calculated additional loss of the wetted surface area parameter,

, of the insulin variants during adsorption on Toyopearl Butyl‐650S.

Table 2.5 The second derivative of the isotherm calculated from the slope of the leading edges of simulated elution profiles compared with the value calculated from the adsorption isotherm.

Chapter 03

Table 3.1 Process parameter set point, relative standard deviation, and sensitivity with ranking.

Chapter 04

Table 4.1 Basic equations and plate number equations for mechanistic chromatography models.

Table 4.2 Effect of initial salt concentration (

I

0

) and sample loading on peak position and width in linear gradient elution (LGE).

Chapter 06

Table 6.1 Table outlining the cost and benefits of using the computational methods discussed previously in process development.

Chapter 10

Table 10.1 Error correlations used in Monte Carlo simulations.

Table 10.2 Relation of flow rate and residence times during elution with 12 and 48 interruptions in HTCC experiments.

Table 10.3 SMA isotherm parameters for lysozyme and avidin using SP Sepharose FF at pH 4.5.

Table 10.4 Process model parameters for HTCC simulations.

Table 10.5 Deviations of retention time and peak width from the reference of a simulated “ideal” experiment without salt steps and flow interruptions (“pauses”) for different experimental setups.

Chapter 11

Table 11.1 Scales of chromatography used during process development.

Table 11.2 Experimental design for miniature column DOE.

Table 11.3 Experimental design for miniature column DOE.

Table 11.4 Peak characteristics and resolution values for Poros HS50 elutions.

Table 11.5 Experimental conditions for slurry‐based screen.

Table 11.6 Comparison of DBC values as evaluated with miniature and lab‐scale columns.

Table 11.7 Summary of runs performed using miniature and lab‐scale columns to evaluate impact of column loading and elution pH.

Table 11.8 Scale‐up of protein A chromatography step.

Table 11.9 Summary of runs performed using miniature and lab‐scale columns to evaluate impact of pH, conductivity, and column loading.

Table 11.10 Scale‐up of AEX chromatography step.

Table 11.11 Summary of CEX experimental conditions tested.

Table 11.12 Experimental results for pH comparison.

Table 11.13 Summary of conditions tested via DOE using lab‐scale column.

Table 11.14 Summary of LRV as tested for worst‐case scenarios.

Chapter 12

Table 12.1 Used column parameters for Source 30Q column.

Table 12.2 Protein parameters.

Table 12.3 Normal composition for the loading solution in the reference process and the modified process.

Chapter 13

Table 13.1 Typical physical and mass transport parameters for cation exchange chromatography model.

Table 13.2 Typical monoclonal antibody SMA parameters.

Chapter 14

Table 14.1 Lab scale calibration experiments used for model development.

Chapter 15

Table 15.1 Experimental conditions for the batch breakthrough experiments.

Table 15.2 Model parameters for the best fit of capturing MAb1 on MabSelect SuRe chromatography media.

Table 15.3 Model parameters for the best fit of capturing MAb1 on CaptivA PriMAB chromatography media.

Table 15.4 Model parameters for the best fit of capturing MAb2 on MabSelect SuRe chromatography media.

Table 15.5 Key experimental conditions and performance indicators for the BioSMB experiments used in this study: MAb1 on MabSelect SuRe.

Table 15.6 Key experimental conditions and performance indicators for the BioSMB experiments used in this study: MAb1 on CaptivA PriMAB.

Table 15.7 Key experimental conditions and performance indicators for the BioSMB experiments used in this study: MAb2 on MabSelect SuRe.

Table 15.8 Summary of simulation parameters fitted for Batch and BioSMB experiments: MAb1 on CaptivA PriMAB.

Table 15.9 Summary of simulation parameters fitted for Batch and BioSMB experiments: MAb2 on MabSelect SuRe.

Chapter 16

Table 16.1 Classification of data types based on the number of variables. For obtaining good quality models, the minimum number of variables included in the study should be five.

Table 16.2 Summary of the literature on the various preprocessing methods applied on datasets prior to chemometrics analysis.

Table 16.3 Summary and meaning of key PCA model outputs (Umetri [32] and Kirdar et al. [33]).

Table 16.4 Summary and meaning of key PLS model outputs (Umetri [32] and Kirdar et al. [33]).

Chapter 17

Table 17.1 Number of tryptophan, tyrosine, and phenylalanine residues in proteins without a heme group.

Chapter 18

Table 18.1 Representative order‐of‐magnitude estimates of the process water and chemicals and consumable materials used in manufacture of recombinant proteins.

Table 18.2 Three representative stages of process control and their impact on process sustainability.

Table 18.3 Several considerations related to the nexus between environmental and economic thinking.

List of Illustrations

Chapter 01

Figure 1.1 (Top) The framework of QbD. (Bottom) Example of QbD elements contained in the QbD framework for a preparative chromatography step.

Figure 1.2 General extent of knowledge and process understanding obtained employing various methodologies and approaches.

Figure 1.3 HMWP content after purification on SEC for a biopharmaceutical as a function of feed concentration. , experimental results; , model prediction by mechanistic model; and , model prediction by statistical model based on DoE.

Figure 1.4 Chromatogram and purity of a three‐column MCSGP unit as a function of time for a 23 h semi‐continuous chromatographic purification of insulin (L. Aumann et al., Chromacon AG, internal report to Novo Nordisk).

Figure 1.5 Application options of mechanistic modeling.

Chapter 02

Figure 2.1 A schematic representation of a unit volume of a chromatographic column. Gray shading represents the solid‐phase support which is unaffected by the interaction with mobile phase and solutes.

Figure 2.2 A log–log plot of the measured isocratic retention volumes of a GLP‐1 derivative () and six contaminants on a Source 30 Q adsorbent at various sodium chloride concentrations in the eluant.

Figure 2.3 The adsorption isotherm (full line) of a GLP‐1 derivative on a Source 30 Q adsorbent with 0.03 M sodium chloride in the eluant. The dashed–dotted line is the slope of the isotherm, and the dashed line is

q

/

c

.

Figure 2.4 The calculated relative velocities of the diffuse wave and the shock wave of a GLP‐1 derivative on a Source 30 Q adsorbent with 0.03 M sodium chloride in the eluant.

Figure 2.5 Simulated isocratic elution profiles of a GLP‐1 derivative on a Source 30 Q adsorbent with 0.03 M sodium chloride in the eluant. The numbers in the legend indicate the load in mM∙CV. The concentration of GLP‐1 in the feed was 0.2 mM. The solid line that extends above the top of the simulated elution profiles shows the shape of the trailing edge of the ideal elution profile created by the diffuse wave.

Figure 2.6 Simulated isocratic elution profiles of a GLP‐1 derivative on a Source 30 Q adsorbent with 0.03 M sodium chloride in the eluant and a load of 8.8 mM∙CV using two different feed concentrations. The first number in the legend is the number of column volumes loaded, and the second number is the mM concentration of GLP‐1 in the feed. G means the GLP‐1 elution profile, and S means the salt elution profile.

Figure 2.7 Simulated isocratic elution profiles of a GLP‐1 derivative on a Source 30 Q adsorbent using 60, 100, and 200 cells in the Craig model. The eluant was a 0.03 M sodium chloride solution, and the load was 0.2 mM∙CV.

Figure 2.8 The ideal elution profile of a GLP‐1 derivative on a Source 30 Q adsorbent with 0.03 M sodium chloride in the feed and in the eluant. The dotted square from 0 to 1 CV represents a load of 5 mM∙CV. The front of the feed pulse enters at

V

i

 = 0. The curved line is the trailing edge of the elution profile, and the vertical dashed–dotted line at

V

i

 = 1.75 CV is the leading edge of the elution profile. The arrows connected with a dashed–dotted line indicate the distance between the front of the feed pulse and the resulting shock wave. The arrows connected with a solid line indicate the rear of the feed pulse and the resulting trailing edge of the elution profile.

Figure 2.9 The ideal elution profile of a GLP‐1 derivative on a Source 30 Q adsorbent with 0.03 M sodium chloride in the feed and in the eluant. The dotted square from 0 to 0.5 CV represents a load of 2.5 mM∙CV. The front of the feed pulse enters at

V

i

 = 0. The curved line is the trailing edge of the elution profile, and the vertical dashed–dotted line at

V

i

 = 1.86 CV is the leading edge of the elution profile. The arrows connected with a dashed–dotted line indicate the distance between the front of the feed pulse and the resulting shock wave. The arrows connected with a solid line indicate the rear of the feed pulse and the resulting trailing edge of the elution profile.

Figure 2.10 The solid line shows the deviation between the second derivative of the isotherm and the second derivative of the

F

(

c

) function. Details are given in Section 2.7.1.4.

Figure 2.11 Experimental and modeled isocratic retention volumes of lysozyme on a Tosoh Butyl‐650S adsorbent at pH 7 with ammonium sulfate in the eluant. The triangles show the experimental data, and the solid line shows the correlation. The dashed line is the predicted retention volume at pH 4.

Figure 2.12 Experimental and calculated solubility curves of lysozyme in an aqueous solution of ammonium sulfate. The squares and the triangles are experimental data, and the full line is the correlated solubility curve at pH 4. The dotted line is the predicted solubility curve at pH 8.

Figure 2.13 Calculated equilibrium excess function of lysozyme in four ammonium sulfate solutions at pH 4 (dashed–dotted lines) and at pH 8 (full lines). The ammonium sulfate concentrations are × 0.01 M, squares 0.3 M, diamonds 1 M, and triangles 1.2 M. The lines at 1 and 1.2 M salt terminate at the solubility limit of lysozyme.

Figure 2.14 Experimental and correlated thermodynamic retention factors of lysozyme at pH 7 on the Toyopearl Butyl‐650S (squares) and the Toyopearl Ether‐650S (diamonds) adsorbents. The symbols show the experimental values, and the lines are the correlations. The eluant is an aqueous solution of ammonium sulfate. Solvents: solid lines, pure water; dashed lines, water with an admixture of 6.1 weight % ethanol.

Figure 2.15 A comparison of the old and the new correlation of the thermodynamic retention factor of the insulin ester. The symbols represent the old correlation, and the lines represent the new correlation: circles 0% ethanol, squares 5% ethanol, and triangles 10% ethanol. Adsorbents, full lines, Toyopearl Butyl‐650S; dashed lines, Toyopearl Phenyl‐650S.

Figure 2.16 The parameters in the retention model of an insulin ester on the Toyopearl Butyl‐650S (squares) and the Toyopearl Phenyl‐650S (triangles) adsorbents with different concentrations of ethanol in the eluant. The open symbols are the values of the linear slopes of the retention model,

b

i

, and the solid symbols are the values of ln

A

0

,i

. The dashed lines are trend lines to indicate that

b

i

is not a linear function of the ethanol concentration.

Figure 2.17 The calculated adsorption isotherm (full line) of a GLP‐1 analogue on a Source 30 Q adsorbent with 0.065 M sodium chloride in the eluant. The dashed–dotted line is the slope of the isotherm, and the dashed line is

q

/

c

.

Figure 2.18 The calculated second derivative of the adsorption isotherm of a GLP‐1 analogue on a Source 30 Q adsorbent with 0.065 M sodium chloride in the eluant. The open square indicates the location of minimum of the second derivative.

Figure 2.19 The relative velocity of the diffuse wave and the shock wave of a GLP‐1 analogue on a Source 30 Q adsorbent with 0.065 M sodium chloride in the eluant.

Figure 2.20 Simulated isocratic elution profiles of a GLP‐1 analogue on a Source 30 Q adsorbent with 0.065 M sodium chloride in the eluant. The loads were 0.001 (full line), 0.002 (dotted line), 0.005 (full line), 0.02 (dashed line), and 0.05 (full line) mM∙CV. The concentration of GLP‐1 in the feed was 0.1 mM. The dashed–dotted line shows the calculated ideal elution profile created by the diffuse wave.

Figure 2.21 Simulated isocratic elution profiles of a GLP‐1 analogue on a Source 30 Q adsorbent with 0.065 M sodium chloride in the eluant. The loads were 0.2, 0.5, 1 (full lines), 1.5 (dotted line), 2 (dashed line), 5 (full line), 10 (dashed–dotted line), and 15 (full line) mM∙CV. The concentration of GLP‐1 in the feed was 1 mM. The crosses show the ideal elution profile created by the diffuse wave.

Figure 2.22 The ideal elution profile of a GLP‐1 analogue on a Source 30 Q adsorbent with 0.065 M sodium chloride in the feed and in the eluant. The vertical lines are shock waves at different loads (mM∙CV): short dashed, 1; dashed–dotted, 2; dashed–double‐dotted, 5; and long dashed, 10. The full vertical line at 3.25 CV is a shock wave common to all loads >1.1 mM∙CV.

Figure 2.23 The ideal elution profile (full line) and the simulated elution profile (dashed–dotted line) of a GLP‐1 analogue on a Source 30 Q adsorbent with 0.065 M sodium chloride in the eluant at a load of 5 mM∙CV. The curved full lines are the part of the elution profile calculated from the diffuse wave velocity, and the dashed line in between is the part that disappeared due to the shock wave formation at the upper part of the leading edge at 1.9 CV. The vertical line at 3.25 CV is a shock wave that terminates the rear of the ideal elution profile.

Chapter 03

Figure 3.1 Simulation of buffer change using a SEC model in Example 3.1. Three proteins enter in a buffer with low pH and high salt concentration and exit with a desired buffer.

Figure 3.2 Simulation of an IEX separation in Example 3.2.

Figure 3.3 Simulation of the IEX separations of three proteins at two different pH.

Figure 3.4 Simulation of insulin separation from product‐related impurities. (Top) Ethanol step elusion in RPC. (Bottom) Decreasing salt gradient in HIC.

Figure 3.5 Simulation of affinity‐based capture of an antibody.

Figure 3.6 Simulation of MMC separation in Example 3.6.

Figure 3.7 The front‐capturing of 16 different flux limiters are seen above, from the top: WENO, low order upwind, Lax‐Wendroff, Beam‐Warming, Fromm, van Leer, umist, Sweby15, Ospre, Osher15, mc, minmod, hquick, hcus, van Albaba2, van Albaba1.

Figure 3.8 One simulation of 113 model responses done during calibration.

Figure 3.9 Maximal pooling of the target protein with a purity requirement of 96%.

Figure 3.10 The chromatogram for the worst‐case scenario.

Figure 3.11 (Top) Scatterplot visualization of the distribution of the purity as a function of a set of process parameters, (bottom) the resulting purity distribution with average of 96%. There is 5% probability of purity below 95.5%.

Figure 3.12 Decision variable impact on the objective minima in Example 3.12.

Figure 3.13 Seven Pareto fronts for the multi‐objective optimization of separation problem in Example 3.3 based on increasing purity requirements from 96 to 99%. The triangles indicate the optimal point for the selected objective in Example 3.12.

Figure 3.14 The optimal general gradient for the HIC case from Example 3.4.

Chapter 04

Figure 4.1 Elution methods (operation modes) for chromatography. (a) Isocratic elution. (b) Linear gradient elution. (c) Breakthrough curve (frontal analysis). (d) Stepwise elution.

Figure 4.2 Schematic representation of separation of two components in chromatography (upper) and transport phenomena in the chromatography column (bottom).

Figure 4.3 Chromatography elution curves (chromatogram, elution profile). The peak width at

(half width) becomes 2.35

σ

.

Figure 4.4 Relationship between

h

(dimensionless HETP) and

ν

(dimensionless velocity).

Figure 4.5 Perfusion effect.

Figure 4.6 Resolution,

R

s

. (a, b)

R

s

 < 1. (c)

R

s

 ≈ 1. (d)

R

s

 ≈ 2.

Figure 4.7 Adsorption isotherms as a function of salt concentration.

Figure 4.8 Zone movement during isocratic and gradient elution.

Figure 4.9 GH–

I

R

plots and LGE experiments.

Figure 4.10 Linear gradient elution chromatography (zone‐spreading and zone‐sharpening effects).

Figure 4.11 Flow sheet for determining HETP from linear gradient elution (LGE) curves.

Figure 4.12 Separation time and relative elution volume relationships as a function of column length.

Figure 4.13 Illustration on how to adjust the flow velocity and the gradient slope for reducing separation time (upper figure) and buffer consumption (lower).

Figure 4.14 Typical types I and II elution behavior. Column: DEAE Sepharose CL6B. Sample: ovalbumin (0.2 kg/m

3

) sample volume = 6 cm

3

, column agarose‐based beads,

d

p

 = ca. 110 µm,

d

c

 = 1.5 cm,

Z

 = 10 cm,

u

0

 = 0.2 cm/min, pH 7.9. The inset shows the effect of zone‐sharpening effect. Compared with an isocratic elution peak (the plate number,

N

 = 100), type I elution peak (

N

 = 100) is very sharp.

Figure 4.15 Stepwise elution. In the column the protein zone moves in the front‐spreading boundary of the elution buffer in type I elution, and consequently at the column outlet the protein peak appears in the very steep salt gradient. For type II elution the peak retention volume

V

R

increases, and the peak becomes wider with decreasing buffer salt concentration

I

E

(increasing

K

E

).

V

R

is quite sensitive to a small change in

I

E

(see Figure 4.14). (a) Elution curves, (b) zone movement in the column, and (c) peak trajectory.

Figure 4.16 GH–

I

R

curves and

K–I

curves for monoclonal antibodies.

Figure 4.17 Type I stepwise elution‐based on

K–I

curves from GH–

I

R

curves.

Figure 4.18 Extra‐column broadening in type I stepwise elution.

Figure 4.19 Flow‐through chromatography (FTC) of monomer separation in comparison with linear gradient elution.

Figure 4.20 Operation windows for flow‐through chromatography (FTC) on a distribution coefficient

K

versus salt concentration

I

graph. A similar conceptual graph was presented by Kelley et al. [39].

Figure 4.21 Relative elution volume versus ion exchange capacity. The reference value = 

V

R

/

V

t

 = 7.5, +5% value = 7.9, −5% value = 7.1.

Figure 4.22 Relative elution volume versus ionic strength. Curve

r

 = reference value

, curve 1:

, curve 2:

.

Source

: Yamamoto et. al. [36]. Reproduced by permission of Elsevier.

Figure 4.23 Relative elution volume versus pH. Reference value = pH 5.

Figure 4.24 Distribution coefficient

K

as a function of salt concentration

I

determined by different methods for Mab and “triple light chain” (3LC) variant. The p

I

values for Mab monomer and 3LC are almost the same. HTS batch binding: robotic system with a 96‐well filter plate batch adsorption method. Yamamoto model:

K–I

curves calculated from

GH

–

I

R

curves

.

Figure 4.25 (a)

B

as a function of pH and related separation behavior. (b) A schematic drawing of

B

or

Z

p

versus pH.

Figure 4.26 Stoichiometric displacement model or steric mass action (SMD) model.

Figure 4.27 GH–

I

R

curves for PEGylated bovine serum albumin (BSA), the

B

values and the calculated hydrodynamic radius. The hydrodynamic radius

R

values were calculated by the Fee and Van Alstine equation [54].

Figure 4.28 Possible retention mechanism of PEGylated proteins in linear gradient elution IEC.

Figure 4.29 LGE elution curves and GH–

I

R

curve of DNAs (poly‐T oligo DNA).

Figure 4.30

B

and

I

R

as a function of DNA

M

w

or charges.

Figure 4.31

B

as a function of DNA

M

w

for different IEC columns.

Figure 4.32 Simulated linear gradient elution curves of DNA.

Figure 4.A.1 Numerically calculated linear gradient elution curve.

Chapter 05

Figure 5.1 Determining dynamic binding capacity.

Figure 5.2 Loading a column in periodic counter‐current chromatography (PCC) versus SCC.

Figure 5.3 Management of process chromatography solutions in a batch chromatography system (left) and a continuous system (right).

Figure 5.4 Operation of a three‐column PCC system. (a) Switching sections of a PCC cycle operation. (b) UV monitoring of the PCC cycle.

Chapter 06

Figure 6.1 (a) Schematic diagram showing the three parameters, which define the orientation of a protein with respect to the chromatographic surface (

φ

,

θ

, and

H

P

). (b) Protein‐surface interaction map of lysozyme on a model ion exchange surface. (c) Preferred binding orientations of lysozyme on the surface as determined from the hotspots on the interaction map in (b).

Figure 6.2 Comparison of retention data from chromatography and number of highly favorable orientations obtained from protein–surface interaction maps for various mutants of cold shock protein B on an ion exchange surface.

Figure 6.3 Multimodal ligand surface binding residues determined by NMR with color‐coded dissociation constants for (a) “Capto ligand” surface and (b) “Nuvia ligand” surface. (c) SAP map and (d) EP map for ubiquitin.

Figure 6.4 Currently available molecular dynamics methodologies and time scale of protein motions that can be detected using such an approach. Most of these methods impose assumptions and simplification of protein dynamics. Hence appropriate data processing is needed to distinguish the different modes of motion sampled and how the results replicate realistic protein dynamic properties.

Figure 6.5 (a) MD snapshot of ubiquitin solvated in an aqueous solution of free Capto MMC ligands. (b) Various moieties present on Capto MMC.

Figure 6.6 Maps of various Capto MMCs and its various moieties binding to ubiquitin as obtained from free ligand simulations. (a) Entire ligand, (b) carboxylic acid moiety, (c) aliphatic tail, (d) phenyl group. Color code: blue → green → yellow → red indicates the order of lowest to highest binding.

Figure 6.7 Spatial binding maps of Capto MMC at various distances away from the surface of ubiquitin. Maps are based on density of ligand (

ρ

′) as evaluated by spherical harmonics analysis and normalized by bulk density of ligands.

Figure 6.8 Differences in binding of benzene subgroup on Capto MMC (a) and benzene molecule (b) dispersed freely in aqueous solution with ubiquitin.

Figure 6.9 Comparison of binding hotspots as determined from MD simulations (color coding same as in Figure 6.6) and NMR experiments (region within the dark black line showed maximum binding in NMR). Residues marked in boxes showed changes in retention time in chromatographic experiments upon mutation, whereas those in ovals did not.

Figure 6.10 Structure of IgG

1

antibody illustrating the structural features. The Fab region is composed of both light and heavy chain sequence. The CDR loops are in purple, red, and brown. The Fc region is in blue and linked to the Fab with the purple linker. Disulfide bonds are shown in yellow. The ribbon diagram for IgG

1

was generated from PDB file 1IGY [179].

Figure 6.11 Flow diagram for the development of antibody homology structure. The major steps in the process of developing a homology model of antibodies based on initial protein sequence.

Figure 6.12 Structure and biophysical analysis of two antibody molecules (IgG

1

and IgG2

a

), highlighting differences that can be used to improve process development. (a) Patch analysis of protein surface properties as a function of pH, which illustrate that as a function of increasing pH, the percent distribution of charge properties that can influence ligand binding varied between the two molecules. (b) A plot of protein net charge as a function of pH, which shows a similar trend for both molecules but subtle differences in the net charge from pH 4 to 8 that can be exploited to improve separation and process development.

Chapter 07

Figure 7.1 Scale‐up on a volume basis. The line to the left shows how the performance (plate number) increases with the bed height and reaches an asymptotic level. The line to the right shows the quadratic increase of the pressure drop with the bed height. The horizontal dashed lines are the lower limit for the performance and the upper limit for the pressure drop. The intercept of the solid and dashed lines gives the limits of the working range for the bed height.

Figure 7.2 Separation of β‐lactoglobulin A and B by AIEX on Source 30Q [3]. Dimensions are column diameter × bed height (cm). The dashed line is the gradient. The flow rate is constant at

Q

 = 30 CV/h, while the velocity (

v

in cm/h) and the bed height (cm) changes by a factor of 10. The separation for

L

 = 1.7 cm is slightly less than for the remaining bed heights in agreement with theory.

Figure 7.3 Purification of rFVII by AIEX [4]. The flow rate is constant at

Q

 = 40 CV/h, while the velocity and the bed height change by a factor of 10. The data for the two main impurities show identical results. The calculated pressure drop is changed by a factor of 100.

Figure 7.4 Separation of human insulin (HI) and deamidated HI on RPC [5]. Two cases (different gradients) are each at two bed heights (10 and 25 cm). The flow rate is constant at

Q

 = 12 CV/h, while the velocity and the bed height changed by a factor of 2.5. The separation is the same with a slightly increased performance for from

L

 = 10 to

L

 = 25 in case A.

Figure 7.5 Purification of a recombinant peptide using RPC and bed heights of 25, 10, and 5.7 cm [6]. The flow rate was constant at

Q

 = 12 CV/h. The impurity level was too high when the bed height was 10 cm and below. A plot of the corresponding plate numbers versus the bed height (bottom right) shows how the assumption of

A

/

L

 ⩽ 

CQ

does not hold in this case. A minimum bed height of 15 cm was later established.

Figure 7.6 Scale‐up of an RPC step for a recombinant peptide including a scaling factor of ~400 and the bed height × 2 (S. Kidal and L. Sejergaard, Private Communication). All the analytical data were well within specification. A slight improvement in purity was observed with increasing bed height. The difference in retention volume is due to the different dead volumes in the systems.

Figure 7.7 Increasing the production capacity for human insulin on RPC by 33% by increasing the bed height from 30 to 40 cm at constant flow rate (validation at intermediate scale,

D

 = 5 cm (A. Laursen and E. Hansen, Private Communication)). A slight increase in purity is observed with increased bed height. As a bonus this increases the robustness of the step.

Figure 7.8 Scale‐up of an IEX step for a recombinant peptide including a scaling factor of ~550 and the bed height from 25 cm (lab, black) to 35 cm (pilot, gray) (S. Kidal and O.E. Jensen, Private Communication). The chromatograms show big differences (left figure) due to a difference in the system volume and saturation of the UV detector in the lab run (black). For better comparison the two profiles are aligned in the valley of the separation (right), and this shows that there is only a small difference in separation, with the pilot profile (gray) being slightly sharper. No significant difference was found in the analytical results.

Figure 7.9 Scalability study of a recombinant protein purification using HIC (T.B. Hansen, Private Communication). OD280 versus volume (CV) for four batches run in production (48 L, 16.5 cm bed height) and simultaneously in the laboratory (58 mL, 8.2 cm bed height). Elution starts at

V

 = 0. The scaling factor is 1208 with 2× for the bed height. Still, both the chromatograms and the analytical results are identical.

Figure 7.10 Process design for three columns in four scenarios. Numbers are the bed height and the columns IB, IIIA, and IVC are the bottleneck (E. Hansen and I. Christiansen, Private Communication) in the four scenarios.

Figure 7.11 Plate number for typical analysis (left) and preparative mode (right). Dashed lines are

L

/

A

(left) and 1/

CQ

(horizontal, right), respectively. For the preparative case the particle diameter is 10× and the flow rate 2× compared to the analytical case.

Chapter 08

Figure 8.1 Model simulations of the aggregate levels (a) and yield (b) in the elution pool of HIC columns packed with increasing number of theoretical plates (

N

).

Figure 8.2 Model predictions for the pressure–flow behavior of Phenyl Sepharose 6FF at different scales for gravity‐settled bed heights of (a) 15 cm and (b) 30 cm.

Figure 8.3 Column packing model predictions for Phenyl Sepharose 6FF showing the effect of column diameter (

D

) and gravity‐settled bed height (

L

0

) on the compression factor at critical velocity (CF

cr

) at the column center.

Figure 8.4 Column packing predictions for Phenyl Sepharose 6FF showing the effect of column radius (

r

) on the particle axial displacement profile for two different gravity‐settled bed heights (

L

0

). (a)

L

0

 = 15 cm. (b)

L

0

 = 30 cm.

Figure 8.5 The “smile” at the column center and the wall is clearly evident in the pictures at the bottom. (Top right) this figure illustrates the trailing dye profiles due to the retaining bolts. Both these features are captured in the simulation results (top left).

Figure 8.6 Comparison of actual and simulated tracer profiles in a 1.4 m column with a flat header design. The qualification fluid velocity was 100 cm/h.

Figure 8.7 Velocity contours in the packed bed for the flat (top) and ribbed (bottom) designs.

Figure 8.8 Pressure contours in the packed bed for the flat (left) and ribbed (right) designs.

Figure 8.9 Tracer profiles obtained with the “ribbed” (narrow peak) and flat (broader peak) designs. The dispersion coefficient was set to zero in both cases.

Figure 8.10 Tracer transition profiles at the column outlet(s) obtained with the “ribbed” (steeper curve) and flat designs. The ribbed design achieves 99% tracer saturation at the outlet after 1.32 CV compared with 1.50 CV for the flat design.

Figure 8.11 Tracer mass fraction contours on the

x

 = 0 plane of the packed bed at times

t

 = 0 CV (top),

t

 = 1 CV (middle), and

t

 = 1.25 CV (bottom) obtained with the “ribbed” and flat designs. Ribbed design contours utilize mirror imaging across the symmetry surfaces of the model to enable more equal comparisons of the two designs.

Chapter 09

Figure 9.1 Here, CQA that define the needed product quality are identified; an example could, for instance, be viral clearance or product aggregation. Subsequently, process parameters, CPP, whose variation could have an effect on these CQA leading to a high risk of falling outside the design space, are determined. Additionally, other process parameters, KPP, that are not influencing the CQA but do influence the process attributes, PA, which is, for instance, yield, have to be determined as well for a full description of QbD design space.

Figure 9.2 Using the trial and error approach, just a small part of the process design space can be explored due to the inefficient nature and the high consumption of resources during this approach. HTE allows the investigation of a bigger part of the design space while utilizing the same amount of resources due to its miniaturized experiments. Having a trustworthy mechanistic model available enables investigation of the design space by mere in silico simulation and, thus, eliminates the limitation due to resources. Combining mechanistic models with HTE techniques, the so‐called hybrid approach, makes it possible to tailor your model to your experimental system and, hence, to cover an even bigger part of the design space.

Figure 9.3 In the OFAT approach, one variable (

x

1

or

x

2

) is changed at a time which neglects all interactions between these variables and can thus lead to a suboptimal point. In the DoE approach, on the other hand,

x

1

and

x

2

are varied simultaneously, which allows to capture dependencies between the two and hence enables finding the direction of the true optimum (based on Mandenius and Brundin [11]).

Figure 9.4 Schematic drawing of the different porosities, velocities, and effects inside a column as captured by the equilibrium transport‐dispersive model including the mass transfer approximated via the liquid‐film linear driving force.

Figure 9.5 Workflow for applying the hybrid approach to a chromatographic separation.

Figure 9.6 Scheme on the establishment of an operating window allowing pH selection. Upper part: pH gradient fractionation to determine elution pH of protein of interest (PoI)—fractions for gel electrophoresis are marked with a black box. Lower part: gel electrophoresis of the fractions retrieved from the pH gradient fractionation—PoI marked with a black box.

Figure 9.7 Possible plate configurations for HTE resin screening in batch uptake experiments; tested resins, salt concentrations, pH values in case of AEX, and salt type in case of HIC are varied.

Figure 9.8 (a) and (b) Column model validation showing overlaid experimental and simulated elution profiles of serum proteins in AEX and HIC. Feed composition—1.5, 3.5, and 0.6 mg/mL of ALA, BLG, and BSA, respectively. ALA: molecular weight (Mw = 14.2 kDa, p

I

 = 4.2–5.1; BLG: Mw = 18.3 kDa; p

I

 = 5.2–5.4; BSA: Mw = 66–69 kDa, p

I

 = 4.9–5.1). (c) and (d) Optimized column profiles. (c) Lab scale AEX: column volume = 1.0 mL; pH = 7.5; resin = Capto Q. Optimized conditions: gradient length = 15 CV; linear velocity = 400.7 cm/h, sample load = 5.0 CV (8% binding capacity); lower cut point = 11.7 CV, upper cut point = 15 7 CV. (d) Prep scale AEX (column volume = 35.3 L; column length = 50 cm; internal diameter = 30 cm) in linear gradient elution. The pH sample load and all other optimized conditions are the same as in (c).

Figure 9.9 Workflow for applying the hybrid approach to cascaded process optimization.

Figure 9.10 Complex fractionation and characterization scheme to determine important model parameters for crude protein mixtures.

Figure 9.11 Tree diagram showing all possible combinations of different chromatographic operations (AEX, anion exchange; CEX, cation exchange; HIC, hydrophobic interaction; SEC, size exclusion) up to a level of two sequential units.

Chapter 10

Figure 10.1 Sequence of HTCC process steps in case of gradient elution experiments.

Figure 10.2 Fishbone diagram on potential parameters affecting the quality of HTCC experiments on a LHS.

Figure 10.3 Relative standard deviation as a function of the pipetted volume for a LHS. ‐•‐, 50 mM phosphate buffer, pH 7, with fixed tip, stainless steel non‐coated, 1 mL syringe; ‐

‐, 50 mM phosphate buffer, pH 7, with low volume fixed tip, stainless steel FEP‐coated, 0.25 mL syringe; ‐▶‐, 1 g/L avidin with fixed tip, stainless steel noncoated, 1 mL syringe; ‐

‐, 1 g/L avidin with low volume fixed tip, stainless steel FEP‐coated, 0.25 mL syringe.

Figure 10.4 (a) Absorption difference

A

h

 = 

A

990nm

 − 

A

900nm

as a function of pipetted volumes of 100 mM phosphate buffer (marker squares) and 0.6 g/L lysozyme (marker diamonds) in half‐area MTPs (open marker) and full‐area MTPs (filled marker). Fitted by linear regression,

(eight replicates for each volume level). (b) Absorption difference

A

h

 = 

A

990nm

 − 

A

900nm

obtained in half‐area (‐

‐) and full‐area MTPs (‐♦‐) at nominal volumes of 75 and 150 μL, respectively, as a function of the lysozyme concentration in 100 mM phosphate buffer, pH 7.0. The

y

‐axis to the right is obtained relying on a calibration curve established by using a lysozyme solution of 0.6 g/L as displayed in Figure 10.4a.

Figure 10.5 Illustration of the Monte Carlo simulation procedure. (1) Fit of experimentally derived elution (in this case the asymmetric peak). Characteristics of this peak serve as reference values. (2)

In silico

calculation of theoretical concentrations in fractions of varying fraction volume. (3) For each fractionation, randomly distributed errors of different levels are added on volume (

x

‐coordinate) and concentration (

y

‐coordinate) in 10 000 combinations. (4) Each combination of coordinates is fitted for the determination of peak retention time and peak area. (5) Retention time and area of each simulated peak is related to the reference peak. Finally, the 95% confidence intervals of the mean retention time and mean peak area of all simulated peaks per fraction volume are calculated.

Figure 10.6 Effect of the number of fractions per peak width on the peak area (b, d, f) and peak retention time (c, e, g) determined after peak fitting for a symmetrical elution peak displayed by 95% confidence intervals. The confidence intervals were calculated for peak area and retention time of the simulated peaks related to the values of the reference peak (a). Lower and upper boundaries of the intervals are plotted in the same color. Introduced error categories were (b, c), determination of volumes; (d, e), determination of protein concentration; and (f, g), dilution procedures. The following errors introduced were set constant: (b, c),

c

2 %

; (d, e),

V

2 %

, 

V

meniscus

; (f, g),

V

0.5 %

, 

V

meniscus

, 

C

0.5 %

.

Figure 10.7 Effect of the number of fractions per peak width on the peak area (b, d, f) and peak retention time (c, e, g) determined after peak fitting for an asymmetrical elution peak displayed by 95% confidence intervals. The confidence intervals were calculated for peak area and retention time of the simulated peaks related to the values of the reference peak (a). Lower and upper boundaries of the intervals are plotted in the same color. Introduced error categories were (b, c), determination of volumes; (d, e), determination of protein concentration; and (f, g), dilution procedures. The following errors introduced were set constant: (b, c),

c

2 %

; (d, e),

V

2 %

, 

V

meniscus

; (f, g),

V

0.5 %

, 

V

meniscus

, 

C

0.5 %

.

Figure 10.8 Influence of the used fluid flow pattern on peak retention time (a) and (d), peak width at 50% peak height (b) and (e), and peak asymmetry (c) and (f) for experiments using lysozyme and avidin, respectively. For each fluid flow pattern, the same four columns were used in parallel for each protein and are indicated with numbers. The determined values (|, numbers), the mean (|, long), and the 95 and 99% confidence intervals widths (horizontal bars, thick and light gray, respectively) are displayed. A trend was determined if the 95% confidence intervals did not overlay vertically. The effect was considered significant, if also the 99% confidence intervals for each flow pattern did not overlay vertically.

Figure 10.9 Separation of lysozyme (gray) from avidin (black) simulated using different NaCl concentration steps for elution. Flow rate: 54 cm/h (3 μL/s).

Figure 10.10 Effect of salt step height, fractionation, and flow rate on the elution profile for the separation of lysozyme (gray) from avidin (black). (a) 54 cm/h. (b) 144 cm/h. The gradient slope was +525 mM NaCl over 18 CV. Solid lines: simulated elution profiles for corresponding salt step height. Markers (circles: lysozyme, boxes: avidin): concentration values obtained in theoretical fractions of corresponding size (300 or 75 μL). Dashed lines: fit of the fraction concentrations using the EGH function. The step size of 11 and 43.8 mM corresponded to a step length of 75 and 300 μL, respectively. The “ideal” chromatogram displays the simulation result assuming a linear gradient without flow interruptions.

Figure 10.11 Relative deviations of the determined output parameters from the simulated “ideal” gradient experiment for varying fraction volumes and flow rates grouped by a certain salt step height. The effect on the determined retention time and peak asymmetry is given for lysozyme (a, b) and avidin (c, d). The effect on the peak width is only displayed for avidin (e). The effect on the resolution is displayed in subplot (f).

Figure 10.12 Chromatograms of the separation of avidin (black) from lysozyme (gray) at flow rates 144 cm/h (a) and 54 cm/h (b) on SP Sepharose FF (CV = 200 μL) using a salt step height of 21.9 mM and fractionation in 75 μL. Separations were performed on a LHS in four parallel experiments (1a–1d), simulated including flow interruptions and salt steps (2) and performed on a laboratory LC system (ideal gradient and without flow interruptions, 3).

Chapter 11

Figure 11.1 Resin screening methodology via use of high‐throughput process development (HTPD) and lab‐scale column runs. The partition coefficient (

K

p

) is defined as the amount of product bound to the resin in equilibrium with the amount remaining in the liquid phase.

Figure 11.2 Workflow for lab‐scale chromatography process development.

Figure 11.3 Flow‐through AEX screenings for step yield (AEX load = 100 mg/mL resin).

Figure 11.4 Poros HS50 chromatograms at pH 5.0–6.0 with an approximate NaCl linear gradient.

Figure 11.5 Caliper Labchip CE‐SDS (nonreduced) results for Poros HS50 fractions at pH 5.0. The simulated gel shows the product at ~150 kDa with fragments at ~120, 68, and 48 kDa.

Figure 11.6 Butyl Sepharose 4 FF chromatograms with an approximated ammonium sulfate gradient. (a) pH 6.5, (b) pH 7.0, (c) pH 7.5, and (d) 8.0 with the dashed line representing the gradient.

Figure 11.7 Caliper Labchip CE‐SDS (nonreduced) results for Butyl Sepharose 4 FF fractions at (a) pH 7.0 and (b) pH 8.0. The simulated gel shows the product at ~150 kDa with fragments at ~120, 68, and 48 kDa.

Figure 11.8 HP‐SEC retention times for the Butyl Sepharose 4 FF pH 7.0 pools. Peak 1 is the first fragment, peak 2 is the second fragment, and peak 3 is the product.

Figure 11.9 Response surfaces for (a) yield and (b) aggregate content as a function of pH and conductivity at all loadings. A total of 16 points, as indicated by the black stars, were used to determine the response surface.

Figure 11.10 Desirability plot showing the target operating region (dark area) to achieve a yield of >80% and <2% aggregates, minimizing residual HCP (ppm) and DNA (ppb). Dots represent experimental conditions.

Figure 11.11 Miniature column chromatograms showing (a) the full chromatogram and (b) a focus on the initial breakthrough.

Figure 11.12 Factor interaction plots for (a) aggregate levels as a function of feed conductivity and (b) fragment levels as a function of feed pH.

Figure 11.13 Contour plot of capacity with respect to pH and salt concentration.

Figure 11.14 Poros HS50 chromatograms for all six pHs.

Figure 11.15 Resolution chromatograms of monomer and aggregates at pH (a) 4.5, (b), 4.7, (c) 4.9, (d) 5.1, (e) 5.3, and (f) 5.5.

Figure 11.16 Yield as a function of pH and conductivity. The black stars indicate the experimental points used to construct the response surface.

Figure 11.17 Partitioning coefficient values calculated over the range of screening conditions.

Figure 11.18 (a) Aggregates as a function of pH and conductivity. (b) HCP as a function of pH and conductivity. The black stars indicate the experimental conditions used to construct the response surface.

Figure 11.19 Comparison of breakthrough curves as observed at miniature and lab‐scale columns (operated at 300 cm/h) using monoclonal antibody loads at two different concentrations, set point, and 10× higher concentration.

Figure 11.20 Interaction plots showing the impact of column load and elution pH on aggregate levels.

Figure 11.21 Comparison of operating window as obtained by lab‐ and miniature‐scale experiments (dark section depicts the operating window).

Figure 11.22 Worst‐case run with HMW and HCP breakthrough.

Figure 11.23 Yield, aggregates, and HCP as a function of elution pH and conductivity as obtained by (a) miniature and (b) lab‐scale columns (dashed box shows the ranges for operating parameters set for large‐scale experiments).

Figure 11.24 Desirability plots for maximizing yield while decreasing aggregate and HCP levels as obtained by (a) miniature and (b) lab‐scale columns (dashed box shows the ranges for operating parameters set for large‐scale experiments).

Figure 11.25 Resolution of clipped species and aggregates at pH 5.6 with step elution at 15 mS/cm.

Figure 11.26 Yield, reduction of clipped species, and aggregates as a function of elution pH and conductivity.

Figure 11.27 HP‐SEC results of the protein A product low pH hold stability.

Chapter 12

Figure 12.1 Retention volume as a function of salt concentration of the components B, C, and D.

Figure 12.2 Measured capacity from breakthrough experiments and fitted isotherm for compound C.

Figure 12.3 Charge curves for components A, B, C, and D. The process step is run at pH = 7.6. The charge curve is calculated based on the p

K

a

values of Ref. [5] and the corresponding dissociation reactions.

Figure 12.4 Reference chromatogram for process before the

in silico

optimization.

Figure 12.5 Chromatogram for process after the

in silico

optimization. The full line is the absorbance response (referring to the ordinate axis on the left), and the dashed line is the salt concentration (referring to the ordinate axis on the right).

Figure 12.6 Resulting salt concentrations and chromatograms for systems with differences in the mixing volume (0.1, 0.4, and 1 CV).

Figure 12.7 Overlay of experimental reference chromatogram, dashed line, and deviation chromatogram where the main peak is eluting early, full line.

Figure 12.8 Overlay of experimental early elution run and

in silico

simulation of process with higher salt concentration in buffer 1. Only the four components named in Table 12.2 are simulated as described earlier.

Figure 12.9

Top

: experimental chromatogram from new pilot facility with pulse of regeneration solvent.

Bottom

: modeled chromatogram of process with high salt concentration in wash phase.

Figure 12.10

Top

: bound protein concentrations in column after a normal wash phase.

Bottom

: bound protein concentrations after a wash phase containing a small pulse of regeneration solvent.

Figure 12.11

Top

: resulting chromatogram with reference gradient and high‐impurity batch.

Bottom

: resulting chromatogram with modified gradient.

Chapter 13

Figure 13.1 Process chromatogram for a typical monoclonal antibody–cation exchange step.

Figure 13.2 Fundamental mechanisms of chromatography.

Figure 13.3 Prediction of nonbinding monoclonal antibody transport behavior as a function of flow rate.

Figure 13.4 Fractionation of typical cation exchange chromatogram showing product‐related species.

Figure 13.5 Model fit to typical cation exchange fractionation study.

Figure 13.6 Model fit to typical CEX dimer profile.

Figure 13.7 Equilibrium binding constants for monoclonal antibody species as a function of pH (dimer: upper curve; main monomer: lower curve).

Figure 13.8 Model predicted versus experimental pool volumes.

Figure 13.9 Model fit to cation exchange fractionation data at pH 5.0 for Fractogel SO

3

(top) and Fractogel COO resins (bottom).

Figure 13.10 Impact of pH and gradient slope on cation exchange pool dimer levels and pool concentration.

Figure 13.11 Selectivity curves for Fractogel COO (top) and Fractogel SO

3

(bottom).

Figure 13.12 Impact of operating parameters on cation exchange pool concentration (top), step yield (middle), and pool dimer levels (bottom).

Figure 13.13 Impact of operating parameters on dimer levels.

Figure 13.14 Expected distribution of step yield (top) and dimer levels (bottom) by Monte Carlo simulation.

Figure 13.15 Impact of load dimer level and stop collect on pool dimer levels.

Figure 13.16 Expected distribution of step yield (top) and dimer levels (bottom) with alternative stop collect strategy.

Figure 13.17 Impact of raw material‐related parameters on dimer clearance.

Chapter 14

Figure 14.1 Domain structure and

in vivo

activation of endogenous FVIII and the rearrangement of the different domains. M denotes a metal.

Figure 14.2 Overview of downstream process steps.

Figure 14.3 Simulated chromatograms compared to calibration experiments. Blue lines are experimental chromatograms; other colors are simulations. Vertical lines represent suggested peak collection criteria to obtain required purity and concentration.

Figure 14.4 Results for the 216 computer simulations in terms of total HMWP (a), pool concentration (b), purity (c), and yield (d) as function of the load concentration.

Figure 14.5 Verification results compared to simulations for total HMWP (top) and pool concentration (bottom).

Figure 14.6 Correlation between HCP and HMWP reduction.

Figure 14.A.1 Example of a discretized PDE.

Figure 14.A.2 Breakthrough curve for

N

 = 6.

Figure 14.A.3 Breakthrough curve for

N

 = 1000.

Figure 14.A.4 Breakthrough curve for

N

 = 1000.

Figure 14.A.5 Size exclusion example with four parameters (fictional parameters).

Figure 14.B.1 Column model for convection in an ideal tube.

Figure 14.B.2 Column model for convection in an ideal packed bed.

Figure 14.B.3 Column model for diffusion in a packed bed.

Figure 14.B.4 Column model for mass transfer in the packed bed mobile phase and pore phase.

Figure 14.B.5 Column model for adsorption in a packed bed.

Figure 14.B.6 Column model for all contributions.

Figure 14.B.7 Model for residence time.

Figure 14.B.8 Model for dimensionless time.

Figure 14.B.9 Model for space velocity.

Figure 14.B.10 Equations for dimensionless model.

Figure 14.B.11 Model for scale‐up (axial dispersion)—see also Chapter 7.

Figure 14.B.12 Model for scale‐up (mass transfer and adsorption)—see also Chapter 7.

Chapter 15

Figure 15.1 Global layout of the BioSMB system with multiple chromatography columns. The valves (shown in the center) are held in a single disposable cassette.

Figure 15.2 Combining the deviations in horizontal and vertical direction.

Figure 15.3 Breakthrough curves for MAb1 on MabSelect SuRe chromatography media. The markers represent the experimental data at various linear velocities, and the curves represent the model curve according to the simulation model, using one single parameter set that was optimized for the entire dataset (10–40 CV).

Figure 15.4 Breakthrough curves for MAb1 on CaptivA PriMAB chromatography media. The markers represent the experimental data at various linear velocities, and the curves represent the model curve according to the simulation model, using one single parameter set that was optimized for the entire dataset (10–40 CV).

Figure 15.5 Breakthrough curves for MAb2 on MabSelect SuRe chromatography media. The markers represent the experimental data at various linear velocities, and the curves represent the model curve according to the simulation model, using one single parameter set that was optimized for the entire dataset (10–40 CV).

Figure 15.6 Capture efficiencies for MAb1 on MabSelect SuRe according to in silico BioSMB (curves) and derived from the experiments (data points). The capture efficiency is plotted against the operating capacity (or load capacity).

Figure 15.7 Capture efficiencies for MAb1 on CaptivA PriMAB according to in silico BioSMB (curves) and derived from the experiments (data points). The capture efficiency is plotted against the operating capacity (or load capacity).

Figure 15.8 Capture efficiencies for MAb2 on MabSelect SuRe according to in silico BioSMB (curves) and derived from the experiments (data points). The capture efficiency is plotted against the operating capacity (or load capacity).

Figure 15.9 Parity plot for the capture efficiency of MAb1 on MabSelect SuRe as predicted by the numerical simulation model (in silico BioSMB) and from the experiments.

Chapter 16

Figure 16.1 Illustration of the basic structure of a typical chromatogram.

Figure 16.2 An illustration of typical chromatographic data and the possible multivariate data modeling approach. The higher dimensional data matrix is unfolded by preserving the direction of variables. This gives lower dimensional matrix containing data points that are representative of the actual process.

Figure 16.3 A visual representation of PCA showing two orthogonal components accounting for maximum possible variability of the data.

Figure 16.4 Score plots for second and third PCs calculated from UV elution curves: (a) training set cycles 1–36, (b) training set cycles 1–36 plus test set cycles 37–48. Figure adapted from Hou et al. [9] with copyright permission. It is evident that column performance starts deteriorating right from 42nd cycle even though experimentally it was observed only at 48th cycle.

Figure 16.5 Illustration 1: Resulting chromatograms of the inline measurement and off‐line analytical chromatography. Illustration 2: Resulting chromatograms of the product purity‐based real‐time pooling during a chromatographic separation of lys, cyt c, and rib A on SP Sepharose FF. It is observed that the developed model delivers precise results for the retention times and peak flanks for a ternary model protein system.

Figure 16.6 Flowchart illustrating the general workflow of multivariate data analysis (MVDA) of bioprocessing data.

Chapter 17

Figure 17.1 Absorbance in the far‐ and mid‐UV range of peptide bonds in different conformations.

Figure 17.2 Spectra of the components mainly contributing to protein mid‐UV absorption spectra. The measurements were performed with a Lambda 35 spectrophotometer (Perkin Elmer, Waltham, USA).

Figure 17.3 UV absorption spectra of 24 proteins. The spectra were recorded in protein solutions of 1.0 g/L. The spectra are normalized to equal total intensity for better visual comparison. The measurements were performed with a Lambda 35 spectrophotometer (Perkin Elmer, Waltham, USA).

Figure 17.4 UV spectra representative of the extremes within the 24 measured spectra (proteins containing a heme group were omitted). For more details on the protein spectra, see Figure 17.11. The measurements were performed with a Lambda 35 spectrophotometer (Perkin Elmer, Waltham, USA).

Figure 17.5 UV absorption spectra of

L

‐tryptophan and a tryptophan‐rich protein. The spectra have been normalized to equal total intensity. The measurements were performed with a Lambda 35 spectrophotometer (Perkin Elmer, Waltham, USA).

Figure 17.6 UV absorption spectra of insulin detemir, insulin aspart, and human insulin. The spectra have been normalized to equal total intensity. The measurements were performed with a Lambda 35 spectrophotometer (Perkin Elmer, Waltham, USA).

Figure 17.7 Partial least square regression between factor matrix

X

(mid‐UV absorption spectra) and response matrix

Y

(selective protein concentrations). The matrix

B

contains the regression coefficients and the matrix

E

the residuals (information not included in the regression).

Figure 17.8 PLS model calibration from pure protein components (left side) or process data (right side): Pure components can be applied to generate mixtures according to a DoE setup, while fraction analytics of chromatography runs has to be performed to obtain

Y

, if no pure components are available. Finally, the mid‐UV absorption spectra of the mixtures or fractions are measured to obtain

X

.

Figure 17.9 Spectra of 24 proteins measured in microtiter plates with a multimode microplate reader (Infinite 200, TECAN, Germany). All spectra are normalized to equal total intensity to facilitate comparison. Spectra are colored according to their similarity. (a) Spectra for all proteins. (b) Heme proteins. (c) Spectra of proteins without tryptophan. (d–f) Spectra of proteins with increasing ratio of tryptophan to tyrosine.

Figure 17.10 Protein spectra and calibration validation of three binary protein combinations. (a–c) Absorption spectra measured in microtiter plates with a multimode microplate reader (Infinite 200, TECAN, Germany). Spectra have been normalized to equal total intensity. (d–f) Visual display of validation results.

Figure 17.11 Validation result of a calibration of two monoclonal IgGs. (a–b) Absorption and absorption difference spectra of the two IgGs. (c–d) Validation results and resulting coefficient of determination.

Figure 17.12 Result of a HTE resin screen for separation of three proteins with 200 μL packed columns (Atoll, Germany). The separation was performed using a sodium chloride gradient (0–500 mM over 24 CV) and the eluate was collected in fractions of 200 μL. UV absorption spectra of each fraction were measured using an infinite plate reader (Tecan), and protein concentrations were calculated based on a previously calibrated PLS model. Peak fitting was performed to calculate resolution that was used for method validation purposes employing an alternative analytical method (analytical chromatography).

Figure 17.13 UV absorption spectra of human breast milk lysozyme (lysozyme HU) and hen egg white lysozyme (lysozyme AV). The spectra have been normalized to equal total intensity. (a) UV absorption spectra of lysozyme HU and lysozyme AV. (b) Result of selective concentration determination in validation samples containing lysozyme HU and lysozyme AV. (c–j) HTE chromatography with 200 μL columns prepacked with Macro‐Prep ceramic hydroxyapatite type 1, 80 µm (Atoll, Germany). Load: no protein (C/G), 1.0 g/L resin of each protein (D/H), 1.0 g/L resin of lysozyme HU (E/I), 1.0 g/L resin of of lysozyme AV (F/J). Load buffer: 1 mM phosphate buffer pH 6.8. Elution: linear step gradient from 1 to 300 mM phosphate at pH 6.8 over 18 CV. The eluate was collected in fractions of 1 CV.

Figure 17.14 PAT tool consisting of a Dionex Ultimate 3000 DAD, an ÄKTA purifier 10, a DAQ device, and a software tool chain. The software tool chain can communicate with the DAD and the ÄKTA. Communication with the ÄKTA is handled via the DAQ device using digital signals. The software tool chain consists of MATLAB and an Excel Visual Basic for Application Macro that is using classes of the Chromeleon Software Developer Kit. Copyright by 1: National Instruments, 2 MathWorks, 3 Lenovo, 4 Microsoft, and 5 Thermo Fisher Scientific.

Figure 17.15 Resulting chromatograms of the in‐line peak deconvolution and off‐line analytical chromatography. (a) Absorption at 280 and 527 nm plotted against the retention volume. Absorption was measured using a path length of 0.4 mm and then recalculated for a path length of 10 mm. (b) Predicted protein concentrations by the PLS model (solid lines) and determined protein concentrations in collected fractions using off‐line analytical chromatography (dashed lines) plotted against the retention volume.

Figure 17.16 Resulting chromatograms of real‐time pooling during a chromatographic separation of lysozyme, cytochrome c, and ribonuclease A. The black, vertical lines in the chromatograms visualize the pooling criteria that were automatically detected by MATLAB and forwarded to the control software of the ÄKTA. (a) Ribonuclease A was declared as target protein and separated with a gradient length of three CVs. (b) Cytochrome c was declared as target protein and separated with a gradient length of five CVs.

Figure 17.17 Validation run of the case study with mAb monomer, HMW, and LMW. (a) Change in the mid‐UV absorption spectrum due to increasing aggregate level. Absorption beyond 320 nm indicates scattering effects. Absorption was measured using a path length of 0.4 mm and then recalculated for a path length of 10 mm. (b) In‐line peak deconvolution for the 20 CV gradient validation run. (c) PLS model prediction for LMW and reference analytics. (d) PLS model prediction for HMW and reference analytics.

Figure 17.18 Case study with serum proteins. (a) PLS model prediction and corresponding results of the reference analytics for IgG and transferrin. (b) PLS model prediction and corresponding results of the reference analytics for IgA and IgM.

Figure 17.19 Mid‐UV absorption spectra of native insulin aspart and highly enriched deamidated insulin aspart (97% aspB3 and isoaspB3). The measurements were performed with a Lambda 35 spectrophotometer (Perkin Elmer, Waltham, USA).

Figure 17.20 Validation result of a calibration of highly enriched deamidated insulin aspart and native insulin aspart.

Chapter 18

Figure 18.1 An integrated set of mega trend drivers that currently impact on and significantly underpin developments in the sustainable manufacturing of bioproducts.

Figure 18.2 Schematic representation of the transition that is occurring from (a) the traditional manufacturing approaches solely based on cost and performance where local optimization of individual unit operations were practiced to (b) the paradigm shift associated with sustainable manufacturing where the overall environmental footprint of the product and global optimization of the overall process are also incorporated.

Figure 18.3 Operating regions related to the product yield and processing rate for the purification of human serum albumin with a DEAE Trisacryl M ion exchange chromatographic column. The open circles are the maximum production rates (0.87 and 1.46 mg/mL per min) for the 100 and 99% product yield curves derived from the SAM approach.

Figure 18.4 Schematic illustration of the water deployment within a typical monoclonal antibody production facility and the different exit ports where waste water recovery could be undertaken singly or in aggregation.

Figure 18.5 Correlation between the PMI expressed in terms of kilogram raw materials per kilogram API and (a) the Global Warming Potential/Carbon Footprint and (b) aqueous mass intensity for APIs in a development portfolio.