CHROMATQGRAPHIC SCEENCE SERIES
`
`VOLUME 88
`
`3673 a r: a fipti m izati 0 n In
`Preparative
`
`Princépieg and Biophag‘aaeutical Applications
`
`MTX1036
`
`1
`
`MTX1036
`
`

`

`Scale-Up and Optimization in
`Preparative Chromatography
`
`2
`
`

`

`Scale-Up and Optimization in
`Preparative Chromatography
`
`Principles and Biopharmaceutical Applications
`
`edited by
`Anurag S. Rathore
`Pharmacia Corporation
`Chesterfield, Missouri, U.S.A.
`
`Ajoy Vela yudhan
`Oregon State University
`Corvallis, Oregon, U.S.A.
`
`MAKCEL
`
`DEKKER
`
`MARCEL DEKKER, INc. (cid:9)
`
`NEW YORK. BASEL
`
`3
`
`

`

`ISBN: 0-8247-0826-1
`
`This book is printed on acid-free paper.
`
`Headquarters
`Marcel Dekker, Inc.
`270 Madison Avenue, New York, NY 10016
`tel: 212-696-9000; fax: 212-685-4540
`
`Eastern Hemisphere Distribution
`Marcel Dekker AG
`Hutgasse 4, Postfach 812, CH-4001 Basel, Switzerland
`tØl: 41-61-260-6300; fax: 41-61-260-6333
`
`World Wide Web
`http://www.dekker.com
`
`The publisher offers discounts on this ’book when ordered in bulk quantities. For more
`information, write to Special Sales/Professional Marketing at the headquarters address
`above.
`
`Copyright ' 2003 by Marcel Dekker, Inc. All Rights Reserved.
`
`Neither this book nor any part may be reproduced or transmitted, in any form or by
`any means, electronic or mechanical, including photocopying, microfilming, and re-
`cording, or by any information storage and retrieval system, without permission in
`writing from the publisher.
`
`Current printing (last digit):
`10 9 8 7 6 5 4 3 2 1
`
`PRINTED IN THE UNITED STATES OF AMERICA
`
`4
`
`

`

`ii
`An Overview of Scale-Up in
`Preparative Chromatography
`
`’Anurag S. Rathore
`Pharmacia Corporation, Chesterfield, Missouri,, U.S.A.
`
`Ajoy Velayudhan
`Oregon State University, Corvallis, Oregon, U.S.A.
`
`I INTRODUCTION
`
`(cid:149) Preparative chromatography continues’ to be the dominant purification tech-
`nique in the production of biological compounds, especially in the pharmaceu-
`tical and biotechnological industries. However, the conceptual complexity of
`a I purely theoretical approach to preparative chromatography is formidable,
`because we are dealing with systems of highly coupled, nonlinear partial dif-
`ferential equations [1,2]. Although theoretical work is progressing, it can cur-
`rently capture predictively only a few aspects of realistic biotechnological sep-
`arations, especially given the extremely complex biochemical feedstocks often
`used in these applications. It is not entirely coincidental that the current ap-
`proach to scale-up and optimization in industry is highly empirical. Although
`this is natural, especially given the constraints of process validation, the first
`few chapters of this book attempt to show that current theoretical understand-
`ing does, give insight into the, practical issues involved in scale-up and .optim-
`ization. These chapters show that a careful combination of basic theory with
`experiments can reduce the time needed to achieve an effective scale-up of a
`realistic chromatographic separation.
`
`1
`
`5
`
`

`

`2 (cid:9)
`
`Rathore and Velayudhan
`
`It will be convenient to introduce some terminology [3] to clarify the
`ensuing discussion. The various kinds of physiochemical interactions that are
`used in chromatography to produce selectivity are called modes of interaction.
`Examples include electrostatic interactions in ion-exchange or ion chromatog-
`raphy, hydrophobic interactions in reversed-phase and hydrophobic interaction
`chromatography, and specific interactions in affinity chromatography. Once a
`mode of interaction has been chosen, the various ways in which a separation
`can be achieved (isocratic or gradient elution, stepwise elution, displacement,
`frontal analysis) are called modes of operation. For many separations, the best
`mode of interaction is easily specified, and scale-up or optimization focuses
`on the choice of mode of operation.
`Finally, when the concentrations of all adsorbable components are low
`enough to lie within the linear or Henry’s law region of their respective adsorp-
`tion. isotherms, the separation is called linear. Even if one component’s con-
`centration reaches the nonlinear region of its (multicomponent) adsorption
`isotherm for some fraction of the separation, the process is called nonlinear.
`The basic ideas for scale-up and optimization given in the beginning
`chapters are applied to real separations in the subsequent chapters in which
`industrial case studies are presented. An issue of practical importance in a
`separation sequence is that of how to achieve the global optimum in the param-
`eter of interest (typically maximum productivity or maximum recovery or min-
`imum cost; mixed or combined optimization criteria are also possible). This
`issue is not discussed in detail in this chapter, because Chapter 3 deals with
`it comprehensively. Further, the case studies in subsequent chapters often al-
`lude to constraints from one separation step limiting or otherwise affecting
`the choice of conditions in other steps.
`The structure of this chapter is as follows. An introductory section on
`method development places in perspective the various steps involved in arriv-
`ing at an effective separation protocol at the bench scale. This is, of course,
`a necessary preliminary to scale-up, which by definition seeks to maintain
`upon scale-up the quality of a separation that has already been developed.
`Section ifi begins with heuristic rules for scale-up and then develops a simple
`quantitative model that clarifies when such heuristic rules can be used with
`reasonable accuracy. The issue of bed heterogeneity and its implications for
`scale-up are also discussed. In Section IV practical considerations characteris-
`tic of the various modes of interaction and operation are described briefly.
`Although considerations of space preclude the full discussion of all these is-
`sues, key points are brought out and important references in the literature are
`highlighted.
`
`6
`
`

`

`An Overview (cid:9)
`
`II. METHOD DEVELOPMENT
`
`3
`
`Method development Is a multistep process that precedes scale-up and opera-
`lion at large scale. The general practice is to perform optimization at small
`scale due to relatively smaller requirements, of material and resources as well
`as the ease of performing several runs in a parallel fashion.
`
`A. Decoupling of Thermodynamics from Kinetics
`
`A variety of parameters(cid:151)choice of stationary and mobile phases, the particle
`size of the stationary phase, the column dimensions, the flow rate, the feed
`loading(cid:151)affect the production rate and recovery obtained in a preparative
`separation. Trying to understand the interplay of all of these parameters simul-
`taneously is a daunting task. In addition, testing all the various possibilities
`experimentally is likely to be extremely tedious and is impractical under typi-
`cal industrial constraints. However, the following simplification is available
`to us at little cost. It is. likely that equilibrium parameters (the choice of station-
`ary and mobile phases, leading to selectivity) can be selected independent of
`"kinetic" parameters such as flow rate, feed loading, and particle size. Such
`a decoupling of thermodynamic and kinetic parameters is probably rigorously
`justifiable only in linear chromatography, but even in nonlinear chromatogra-
`phy it is likely that choosing the mobile and stationary phases first does not
`significantly decrease the attainable production rates and recoveries.
`The first step is therefore to choose the most effective mode of interac-
`tion. This is often clear from the fundamental properties of the feedstock or
`the product. Other important factors include the objective and nature of the
`separation problem, literature precedents, and prior experience with the prod-
`uct. Inputs from the vendors of chromatographic media and instrumentation
`may also be useful at this stage. The strategy is depicted schematically in Fig.
`1 [4].
`
`B. Optimization of Thermodynamics at Bench Scale
`
`We present here a simple and rapid approach to the thermodynamic component
`of method development. We take the view that for many separations the choice
`of stationary phase is far more important than the choice of mobile phase (this
`is particularly true of ion-exchange runs, where standard salts are used as
`mobile phase modulators). Of course, there are many cases where specific
`binding of various kinds can require the use of special additives for the mobile
`
`7
`
`

`

`/
`
`4
`
`Rathore and Velayudhan
`
`Pick mode of interaction based
`on separation problem,
`literature precedence , prior
`experience and vendor input
`
`Pick resin candidates for resin
`screening based on separation
`problem, literature precedence,
`prior experience and vendor
`input
`
`Perform Resin Screening
`
`Optimize operating conditions
`
`PTIMIZED CHROMATOGRAPHIC STEI
`
`Figure 1. Strategy for optimization of a chromatographic separation.
`
`phase, but we ignore these situations in order to make the general approach
`clear. We therefore intend to use a standard mobile .phase, and we wish to
`screen a variety of stationary phases rapidly and equitably, i.e.., we have re-
`duced the problem to one of resin screening
`Once a list of resin candidates has been prepared, screening is performed
`to select the best resin to perform a particular separation. Selection of the
`resin for a chromatography step is perhaps the most important step in method
`optimization [4-9]. A resin screening protocol is illustrated in Fig. 2. In most
`cases the primary criterion for resin screening, is selectivity.. However, other
`screening criteria may also be identified and used depending on the particular
`separation problem.
`The general approach is as follows (the specifics in what follows are
`for ion-exchange chromatography in the gradient mode of operation, but the
`arguments can easily be generalized to other contexts). The process takes place
`in two stages.
`Stage 1. All stationary phases are packed into columns of identical
`size. If possible, all columns should be run at the same flow rate. This is not
`always practical (e.g., if the particle sizes available for different stationary
`phases are markedly different, then pressure drop constraints may limit the
`range of flow rates). Run a test gradient that spans a wide range of modulator
`levels, so that feed retention is facilitated. Make the gradient as shallow as
`
`8
`
`(cid:9)
`

`

`5
`
`An Overview (cid:9)
`
`Pick pH and buffer
`
`ick a (cid:9)
`resin (cid:9)
`
`Step 1. Batch experiment
`for checkinq bind!
`
`I
`
`Yes
`
`Binding
`
`Yes (cid:9)
`
`Yes
`
`END
`/
`Remove resin
`from further
`
`IN
`
`END
`
`(Remove (cid:9) resin
`from further
`
`Step 3. Run a comparison gradient
`
`Calculate recovery
`and pool purity ) (cid:9)
`
`(Economic
`
`Resins
`
`Figure 2. Resin screening protocol. (Reprinted courtesy LCGC North America, Ad-
`vanstar Communications, Inc.)
`
`practicable, in order to simultaneously get as much resolution as possible un-
`der these conditions. Stationary phases that exhibit little or no retention of the
`product are excluded at this stage. In addition, if almost no resolution is found
`between the product and the primary impurities, these stationary phases may
`be excluded. This latter decision should be made carefully, because the test
`gradient may not be a fair indicator of a sorbent’ s resolution. In other words,
`a sorbent may provide poor resolution of the product under the test gradient
`but high resolution under another gradient. Thus, the latter decision is to be
`made only if there is good reason to believe that this sorvent is unlikely to
`be effective.
`Stage 2. For each of the stationary phases. remaining, determine a
`tailored comparison gradient that is intended to show each sorbent under its
`most effective conditions for the given feed mixture. Parameters such as the
`
`9
`
`(cid:9)
`

`

`6 (cid:9)
`
`Rathore and Velayudhan
`
`feed loading and equilibration buffer should be kept the same for all sta-
`tionary phases. If the flow rate was the same for all runs in stage 1, then it
`should be maintained in this stage. If different sorbents were run with differ-
`ent flow rates in stage 1, then use the same flow rate for each sorbent in this
`stage.
`The comparison gradient is centered around th(cid:231) modulator concentration
`at which the product eluted in the test gradient in stage 1. Then, making the
`assumption that the band spreading of the peaks is inversely proportional to
`the gradient slope, all other parameters being constant, we have
`cLw = 13m (cid:9)
`where a and w are respectively the gradient slope and product peak width in
`the test gradient, and 0 and m are the corresponding parameters in the compari-
`son gradient. If we require that all comparison gradients have the same time
`(for standardization), then the starting and ending modulator concentrations
`(C1 and ci,, respectively) can be determined from the equations
`
`(1)
`
`and
`
`antF
`- (cid:9)
`C. - c 0 - (cid:9)
`2mV
`
`Cy = Celution – antF (cid:9)
`2mV
`
`(2)
`
`(3)
`
`Where Celdon and tw are respectively the concentration and time at which the
`center of the product peak eluted in the test gradient, n is the number of column
`volumes over which the comparison gradient is run, F is the flow rate, and V
`is the column volume. Note that if the beginning concentration c,, is found to
`be negative from Eq. (2), it is set to zero.
`The comparison gradient provides an equitable way of comparing differ
`ent stationary phases for the given feed mixture, because each stationary phase
`is provided with a gradient that is optimized for its particular retention behav-
`ior. Now the usual quantitative parameters of production rates and recovery
`and purity can be used to determine which stationary phase is best.
`The simple approach described above provides a rapid way to choose
`the best resin. However, if the chromatography step is intended for operation
`at preparative scale, particulary for commercial manufacture, several other
`issues must be addressed before final resin selection. These include the cost
`of the resin, the physical and chemical stability of the resin at the bed height
`and the number of cycles to be used at the manufacturing plant, media avail-
`ability with respect to the demand at commercial scale, resin lifetime, leaching
`
`-
`
`10
`
`

`

`An Overview
`
`FA
`
`of ligands, regulatory support files offered by the vendor, batch-to-batch vari-
`ations in resin quality, etc.
`It should be noted that the assumption that peak width is inversely propor-
`tional to gradient slope is an approximation and is not always expected to be
`valid (e.g., significant competition among the product and impurities for bind-
`ing sites on the adsorbent could cause the assumption to fail). However, it is
`likely to be a reasonable approximation for many realistic separations. More
`detailed methods of this kind can be established (see, e.g., Quarry et al. [10])
`but would usually require more data for each sorbent. Similarly, Jandera et a1.
`[11] and Jandera [12] determined optimal gradients in normal and reversed-
`this
`phase systems through numerical optimization of the governing equations;
`is a significant advance in the field but is not yet at the level of accessibility
`where industrial practitioners would use it routinely. The method outlined here
`was chosen for its simplicity and ease of use in an industrial context.
`This approach to resin screening is now demonstrated in detail for a
`practical separation problem. Rathore [4] showed that the stationary phase of
`choice for an anion-exchange separation was found rapidly using this ap-
`proach. .
`
`1. Resin Screening for an Anion-Exchange Chromatography Column
`This case study presents data obtained during the optimization of an anion-
`exchange chromatography column used in the process, of purifying a protein
`molecule derived from microbial fermentation.
`-
`Nine anion-exchange resins(cid:151)BioRad High Q, BioRad DEAE, Phar
`macia DEAE FF, Pharmacia Q FF, Pharmacia Q HP, Whatman Q, Whatman
`QA52, Whatman DE53, and TosoHaas Q650M(cid:151)were chosen for screening.
`All chromatography experiments were performed using an Akta Explorer
`(Amersham Pharmacia Biotech). The buffer and other operating conditions
`were chosen on the basis of prior experience with the molecule. (Pre-equilibra-
`tion buffer: 1 M Tris, pH 8.5. Equilibration buffer: 50 mM Tris, pH 8.5. Protein
`loading: 10 mg/mL resin.) Because the objective of this chapter is to lay out
`an efficient resin screening protocol and not to recommend a particular resin,
`the resins that were used will be referred to as resins 1-9 (not in the order in
`which they are named above). The optimum resin is expected to vary with
`the separation problem.
`Columns were packed with 1 mL of resin and equilibrated for 30 mm
`with the equilibration buffer. Equilibration was followed by loading of protein
`solution containing 1-2 times the intended protein loading for the respective
`column (mg proteinlmL resin). After 30 min, a wash was performed with 5
`mL equilibration buffer and the "flow-through" stream was collected. Protein
`
`/
`
`11
`
`(cid:9)
`

`

`8 (cid:9)
`
`Rathore and Velayudhan.
`
`Table 1 (cid:9)
`
`Final Comparison of Resins Considered
`
`Resin (cid:9)
`
`Binding" (cid:9)
`
`Selectivity" (cid:9)
`
`Recovery .c (cid:9)
`(mAU/mL) (cid:9)
`
`Pool (cid:9)
`purityc (%) (cid:9)
`
`Final
`decision"
`
`V
`
`X (cid:9)
`J (cid:9)
`J (cid:9)
`J (cid:9)
`
`Resin 1 (cid:9)
`Resin
`Resin
`Resin 4. (cid:9)
`Resin 5 (cid:9)
`Resin
`Resin (cid:9)
`Resin
`Resin 9 (cid:9)
`
`X
`X
`X
`X
`J (cid:9)
`1 (cid:9)
`
`X
`754 (cid:9)
`253 (cid:9)
`81.5 (cid:9)
`
`956 (cid:9)
`100 (cid:9)
`97.7 (cid:9)
`
`J (cid:9)
`I (cid:9)
`X
`J (cid:9)
`X
`1 (cid:9)
`J (cid:9)
`J
`a Resins included Bio1ad DEAE BioRad High Q TosoHaas Q650M Whatman Q, Whatman
`QA52 Whatman DE53 Pharmacia DEAE FF Phannacia Q FF and Pharmacia Q HP not num-
`bered in this order.
`= Satisfactory column performance; X = unacceptable column performance.
`Based on measurements by anion exchange HPLC (AE-HPLC)
`Source Reprinted courtesy of LCGC North America Advanstar Communications Inc
`
`(cid:149)
`
`(cid:149)
`
`(cid:149)
`
`was eluted with 10 mL of elution buffer (1 M NaCl in the equilibration buffer),
`and the eluant was collected separately. The flow-throughs and the eluants
`were analyzed for protein by UV absorbance at 280 nm and for its purity
`by anion-exchange high performance liquid chromatography (AE-.HPLC). As
`shown in Table 1, it was found that most resins showed satisfactory binding
`characteristics with the product Only resin 1 showed anomalous behavior in
`that the product was not retained under these conditions, so resin 1 was not
`considered further.
`Next, columns were packed with 10 mL of the remanung eight resins,
`and separations were performed using an identical test gradient of 0-500 mM
`NaCl in 20 column volumes (CV) of equilibration buffer. Peak fractions were
`analyzed by AE-HPLC Figure 3 illustrates the performance of resins 3, 4,
`and 5 under a test gradient of 0-500 mM NaCl in 20 CV The Y axis in Figs
`3 and 4 denotes the peak area obtained upon analysis by AE-HPLC (mAU)
`per unit injection volume (pL). The flow velocity and the fraction sizes are
`given in the figure legends. It is evident that running identical gradients with
`different resins leads to very different elution profiles in terms of the peak
`
`(cid:149)
`
`. (cid:9) (cid:149) Figure 3
`Column performance under test gradients. (Reprinted courtesy LCGC
`North America, Advanstar Communications, Inc.)
`
`12
`
`(cid:9)
`(cid:9)
`(cid:9)
`(cid:9)
`

`

`Resin 3
`100 cm/hr, 0 -500 mM NaCl in 20 CVs, 3 ml fractions
`350
`
`- Protein AE-HPLC
`--- lmpurityAE-HPLC
`
`300 -
`
`250 -
`
`200 -
`
`=L 150
`
`10 (cid:9)
`
`12
`
`
`
`Fraction#
`
`Resin 4
`100 cm/hr, 0 - 500 mM NaCl in 20 CVs, 5 ml fractions
`350
`
`
`Impurity AE-HPLC
`-6-- Pr(cid:243)tein AE-HPLC
`
`300 (cid:9)
`
`g 250
`
`200
`
`. (cid:9)
`
`so
`
`. (cid:9)
`
`4 (cid:9)
`
`6
`
`a (cid:9)
`Fraction#
`
`10 (cid:9)
`
`12 (cid:9)
`
`14
`
`Resin 5
`50 cm/hr, 0 - 500mM NaCl in 20 CVs, 2 ml fractions
`
`.(cid:149),I...
`
`I-e-protein AE-HPLC]
`lrnpurItyAE-HPLCI
`
`350 (cid:9)
`
`300 (cid:9)
`
`.g 250
`
`200
`
`15O
`
`100
`
`2 (cid:9)
`
`4 (cid:9)
`
`6 (cid:9)
`
`8 (cid:9)
`Fraction#
`
`.10 (cid:9)
`
`12 (cid:9)
`
`14
`
`13
`
`(cid:9)
`(cid:9)
`

`

`10 (cid:9)
`
`Rathore and Velayudhan
`
`Table 2 Calculation of the Comparison Gradients
`
`Migration (cid:9)
`time (cid:9)
`(mm) (cid:9)
`
`5.21
`6.61
`7.35
`
`Resina (cid:9)
`
`Resin 7
`Resin 8
`Resin 9
`
`Comparison
`Test (cid:9)
`gradient,
`gradient, Gradient Column Flow (cid:9)
`Elution start(cid:151)end volume volume velocity start(cid:151)end
`(mL) (mL/min) (cid:9)
`(CV) (cid:9)
`(MM) (MM) (cid:9)
`HIM
`110
`100-200
`100-200
`109
`146
`100-300
`
`1.7
`0.9
`1.7
`
`90-130
`80-130
`100-200
`
`20
`20
`20
`
`10
`10
`10
`
`1:
`
`’ Same mTable 1.
`Source: Reprinted courtesy of LCGC North America, Advanstar Communications Inc.
`
`width and peak position in the overall gradient. The poor selectivity obtained
`with resins 3-5 led to their elimination from further consideration.
`As listed in Table 1, it was found that only resins 6-9 showed satisfac-
`tory selectivity between the product and the impurity. Moreover, because resin
`9 exhibited better resolution than resin 6 and they had identical matrix, and
`ligand chemistry, the former was chosen over the latter for further consider-
`ation.
`
`Comparison gradients were calculated according to the procedure de-
`scribed above for resins 7-9. Product recovery was defined as the sum of
`product peak areas (in mAU) in the pooled fractions (having >90% purity by
`AE-HPLC) per milliliter of injected sample. Pool purity was defined as the
`purity of the total pool formed by mixing the fractions that meet the pooling
`criteria. Table 2 shows the calculation of the comparison gradient for these
`three resins, and Fig. 4 illustrated the protein and impurity profiles that were
`obtained after fraction analysis by AE-HPLC
`Figure 4 reinforces the understanding that performing separations with
`the designed "comparison gradients" yields very similar elution profiles with
`different resins and leds to a fair comparison of resin performance. It also
`follows from Fig. 4 that resin 8 showed good purity but poor recovery. Resins
`7 and 9 showed comparable recovery and pool purity. However, because of
`its better selectivity, resin 9 was chosen as the resin for this purification process
`and selected for further optimization of buffer pH, protein loading, feed flow
`rate, elution flow rate, gradient slope, and column length.
`
`Figure 4. Column performance under comparison gradients. (Reprinted courtesy
`LCGC North America, Advanstar Communications, Inc.)
`
`14
`
`

`

`80
`
`70
`
`60
`
`50
`
`40
`
`10
`
`0
`
`80
`
`70
`
`60
`
`0
`
`is0
`(cid:149)j 40
`
`E
`
`30
`
`20
`
`10
`
`n
`
`80
`
`70
`
`60 I:
`
`10
`
`a
`
`Resin 7
`50 cmlhr,90 -130mM NaCl in 11 CVs
`Recovery: 75mAU!ml
`Pool purity: 96
`
`o (cid:9)
`
`-O- Protein AE-HPLC]
`Impurity AE-HPLCI
`
`(cid:149) 5 (cid:9)
`
`10 (cid:9)
`Fraction#
`
`15 (cid:9)
`
`20
`
`Resin 8
`50cm/hr,80-130 mM NaCl in 11 CVs
`(cid:149) (cid:9)
`
`Recovery: 25 rnAU/mI
`Pool purity: 100
`
`GProtein AE-HPLC
`Impurity AE-HPLC
`
`5 (cid:9)
`
`10 (cid:9)
`Fraction#
`
`15 (cid:9)
`
`20
`
`Resin 9
`50 cm/hr, 100 -200mM NaCl in 11 CVs
`
`Recovery: 82 mAU!ml
`Pool purity: 98 %
`-e- Protein AE-HPL]
`Impurity AE-HPLCIj
`
`5 (cid:9)
`
`10 (cid:9)
`Fraction#
`
`15 (cid:9)
`
`20
`
`15
`
`

`

`12
`
`Rathore and Velayudhan
`
`C. Optimization of Kinetics (Operating Conditions)
`at Bench. Scale
`
`Once the stationary and mobile phases have been chosen, we turn to the deter-
`mination of optimal operating conditions, i.e., determination of the kinetic, as
`opposed to thermodynamic, contributions. Thus, the particle size and column
`dimensions are determined in these studies, along with the optimal gradient
`slope and feed loadings. The following general approach is suggested.
`First, experiments are performed to evaluate the effect of various op-
`erating parameters that affect resin performance in terms of the selectivity and
`protein loading. These parameters may include the mobile phase conditions
`(pH, organic content, buffer composition, etc.) and the gradient slope and de-
`sign. Optimum mobile phase conditions and the gradient design are chosen
`from the experimental data obtained.
`Next, the effect of flow velocity and protein loading on the quality of
`separation is evaluated and, on the basis of resin performance, the bed height,
`protein loading, and flow velocity are chosen to obtain satisfactory resolution
`and cycle time. It is desirable that laboratory experiments be done at the bed
`height that will be used at pilot scale in order to obtain comparable column’
`performance at large scale.
`A detailed analysis of the interaction among these kinetic parameters is
`complicated and is not described here. Many of the underlying issues are
`brought out clearly by Felinger in his chapter on optimization (Chap. 3). In
`industrial practice, a heuristic approach similar to the one just described is
`often used. It is likely to produce effective, if not necessarily optimal, op-
`erating conditions in the hands of an experienced practitioner.. More details
`of these practical approaches are given in several of the industrial case studies
`in this book.
`This separation of very large molecules and particles such as viruses is
`an important industrial topic and is beginning to be addressed in the literature
`[13,14]. However, the field is still in its infancy and is likely to change rapidly.
`We therefore do not feel that it would be appropriate to attempt a summary
`here, and we refer the reader to the growing literature on this subject.
`
`Ill. THEORETICAL CONSIDERATIONS IN SCALE-UP
`
`A. Physical Overview
`
`The performance of a chromatography column depends on a variety of design
`and operating factors. In order to have a successful scale-up it is desirable to
`
`16
`
`(cid:9)
`

`

`An Overview
`
`13
`
`C. Commercial scale
`50 cm x 20 cm
`40L
`
`Figure 5. Three different scales of columns used frequently during process develop-
`ment.
`
`maintain kinetic (particle size, pore size, ligand chemistry, temperature, mo-
`bile phase) and dynamic (bed height, flow velocity, packing density) equiva-
`lence between the chromatography columns used in the laboratory and the
`pilot plant. This objective can be accomplished by using identical stationary
`and mobile phases in the two columns and operating them at identical bed
`height, linear flow velocity, protein loading (mg protein per mL of resin), feed
`conditions, gradient length, and gradient slope [6]. To handle the increased
`volume of load at pilot scale, the most common procedure used to increase
`column volume is to increase the column diameter so that the column volume
`increases proportionately [8,15]. This keeps the residence time of the product
`constant and avoids causing any product stability issues.
`Figure. 5 is a schematic illustration of the three sizes of columns that
`are often used at laboratory, pilot plant, and commerical scales. Scouting ex-
`periments in the laboratory are mostly done in small columns to conserve
`the materials and also because several experiments can be done in parallel
`simultaneously at lab scale. However, as discussed above, it is extremely im-
`portant to maintain bed height constant while scaling up, so the best approach
`is to perform the final optimization steps at the bed height that will later be
`used at the pilot plant and commercial scale. This approach is illustrated in
`Fig. 6 and 7. .
`These general considerations are frequently used in industry as the basis
`for scale-up. In the next. section, a quantitative analysis is given that shows..
`when such simple "volumetric" scale-up can be used and describes alternatives
`that are appropriate when the column length must be changed on scale-up.
`The van Deemter equation is widely used to characterize band broaden-
`ing in. a chromatography column and is expressed as
`
`17
`
`

`

`14
`
`Rathore and Velayudhan
`
`INCREASE
`Column diameter ) (cid:9)
`Bed height*
`
`IMITATIONS
`olumn pressure
`(cid:9) distribution
`(Flow
`Uniform packing
`
`’CONSTANT
`Linear flow velocity
`Stationary/ mobile phase
`Protein loading
`Feed conditions
`Gradient
`’.Bed height..-’
`
`COLUMN
`Scale (cid:9)
`Dimensions (cid:9)
`Volume (cid:9)
`
`A. Lab scale
`8 cm x 2.6 cm
`40 mL
`
`B. Pilot plant scale
`25 cm x 20 cm
`1OL
`
`Figure 6. Scaling from laboratory (or bench) to pilot-plant scale.
`
`INCREASE
`
`Column diameter W
`
`B. Pilot plant scale
`25 cm x 20 cm
`10L
`
`C. Commercial scale
`50 cm x 20 cm
`40L
`
`7 CONSTANT
`Linear flow velocity
`Stationary/ mobile phase
`Protein loading
`Feed conditions
`Gradient
`Bed height
`
`\ (cid:9)
`’ (cid:9)
`
`COLUMN
`Scale
`Dimensions
`Volume
`
`Figure 7. Scaling up from pilot-plant scale.
`
`18
`
`(cid:9)
`(cid:9)
`(cid:9)
`(cid:9)
`(cid:9)
`

`

`where u is the linear flow velocity, H is the plate height of the column, and
`A, B, and C are constants. The plate height H is equal to the length of the
`column divided by the total number, of plates N, so H is smaller for a more
`efficient column. A reflebts the quality of the packing of the column and is
`independent of the linear flow velocity. A is small when the column is packed
`well and is homogeneous throughout its length. B is a measure of the band
`broadening due to longitudinal diffusion of the sample components along the
`edge of their respective bands as they travel across the column. It decreases
`with increasing linear flow velocity because the sample components spend
`less time undergoing diffusion inside the column. C includes contributions
`from the binding kinetics (adsorption/desorption) as well as the mass transfer
`of the sample components to and from the packing particles.
`Preparative chromatography is usually carried out at high flow velocity
`in order to increase throughput. Then the C term usually dominates Eq. (4),
`leading to the simplified form
`
`H Cu, (cid:9)
`
`or, equivalently, (cid:9)
`
`N = L (1/Cu) (cid:9)
`
`(5)
`
`In an ideal case, when the column packing and operating conditions are
`kept the same while scaling up (C is a constant), the scald-up involves just a
`volumetric increase in column dimensions. For such a case, Eq. (5) can be
`rewritten as
`
`L/u=CN (cid:9)
`
`(6)
`
`To preserve the efficacy of separation, the total number of plates is to be kept.
`constant, so it follows from Eq. (6) that if the bed height needs to be increased
`or decreased for some reason (e.g., pressure drops too high), the linear flow
`velocity might also be altered appropriately so as to keep the ratio of L/u
`(N), and
`constant. This ensures a constant number of plates in the column
`the column performance is maintained. This simple analysis is expanded and
`generalized in the following section. In particular, if the particle size needs
`to be changed upon scale-up (for economic or other reasons), the more general
`treatment must be used.
`This very simple physical introduction to scale-up sets the scene for a
`straightforward quantitative analysis of the problem in the next section.
`
`B. Simple Scale-Up Calculation
`
`The basic idea behind scale-up is to preserve the quality of the separation
`achieved at small scale [16,17]. Implicit in this approach is the admission that
`
`7.
`
`19
`
`

`

`16 (cid:9)
`
`Rathore and Velayudhan
`
`we are not yet able to determine optimal operating conditions a priori for
`different scales of operation. Thus, we. settle for determining effective, near-
`optimal operating conditions at bench sale. Effective scale-up rules should
`then produce comparable results at larger scale. (The alternative approach of
`finding optimal operating conditions is currently practicable for some impor-
`tant classes of separation problems; this approach is discussed in Chapter 3).
`A typical scale-up from laboratory to pilot plant is on the order of 50-
`100-fold. This is frequently followed by a 10-50-fold scale-up from pilot plant
`to final commercial manufacturing scale.
`The usual approach is to hold the plate count constant upon scale-up
`and increase the feed volume and column volume proportionately. This ap-
`proach was originally based on the assumption of linear adsorption. Later in
`this section we discuss how this assumption can be relaxed.
`If the. subscripts b and 1 are used to describe parameters at bench and
`large scale, respectively, we have
`
`NjNb (cid:9)
`
`Vf,1 = Vfee.j,b
`VCOJU
`,b
`
`Vcoiunm,i (cid:9)
`
`(7)
`
`If band’ spreading is dominated by pore diffusion, as is often the case
`in realistic separations [18-20], then the plate count can be described by
`
`
`N -
`
`
`
`
`
`
`--- - -- (cid:9)
`
`L
`Ud2
`P
`
`
`
`
`. (cid:9)
`
`(9
`
`where L is the column length, u the mobile phase linear v