Review Article

Received: 10 April 2008, Revised: 16 April 2008, Accepted: 18 April 2008, Published online in Wiley InterScience: 10 June 2008

(www.interscience.wiley.com) DOI:10.1002/jmr.898

Protein purification using chromatography:

selection of type, modelling and optimization
of operating conditions
J. A. Asenjoa * and B. A. Andrewsa
To achieve a high level of purity in the purification of recombinant proteins for therapeutic or analytical application, it
is necessary to use several chromatographic steps. There is a range of techniques available including anion and cation
exchange, which can be carried out at different pHs, hydrophobic interaction chromatography, gel filtration and
affinity chromatography. In the case of a complex mixture of partially unknown proteins or a clarified cell extract,
there are many different routes one can take in order to choose the minimum and most efficient number of
purification steps to achieve a desired level of purity (e.g. 98%, 99.5% or 99.9%).
This review shows how an initial ’proteomic’ characterization of the complex mixture of target protein and protein
contaminants can be used to select the most efficient chromatographic separation steps in order to achieve a specific
level of purity with a minimum number of steps. The chosen methodology was implemented in a computer- based
Expert System. Two algorithms were developed, the first algorithm was used to select the most efficient purification
method to separate a protein from its contaminants based on the physicochemical properties of the protein product
and the protein contaminants and the second algorithm was used to predict the number and concentration of
contaminants after each separation as well as protein product purity.
The application of the Expert System approach was experimentally tested and validated with a mixture of four
proteins and the experimental validation was also carried out with a supernatant of Bacillus subtilis producing a
recombinant b-1,3-glucanase.
Once the type of chromatography is chosen, optimization of the operating conditions is essential. Chromatographic
elution curves for a three-protein mixture (a-lactoalbumin, ovalbumin and b-lactoglobulin), carried out under
different flow rates and ionic strength conditions, were simulated using two different mathematical models. These
models were the Plate Model and the more fundamentally based Rate Model. Simulated elution curves were
compared with experimental data not used for parameter identification. Deviation between experimental data
and the simulated curves using the Plate Model was less than 0.0189 (absorbance units); a slightly higher deviation
[0.0252 (absorbance units)] was obtained when the Rate Model was used. In order to optimize operating conditions, a
cost function was built that included the effect of the different production stages, namely fermentation, purification
and concentration. This cost function was also successfully used for the determination of the fraction of product to be
collected (peak cutting) in chromatography. It can be used for protein products with different characteristics and
qualities, such as purity and yield, by choosing the appropriate parameters. Copyright # 2008 John Wiley & Sons, Ltd.

Keywords: protein purification; chromatography; selection; modelling; optimization

INTRODUCTION Once the type of chromatography is chosen, optimization of

the operating conditions is essential. In chromatography, protein
Until now, it has been virtually impossible to select separation adsorption depends on composition and concentration of the
and purification operations for proteins either for therapeutic or mixture and also on operation conditions such as flow rate, ionic
analytical application in a rational manner due to a lack of strength gradient, sample load, physical properties of the
fundamental knowledge on the molecular properties of the
materials to be separated and also the lack of an efficient system
to organize such information. A range of techniques is available
* Correspondence to: J. A. Asenjo, Department of Chemical Engineering and
such as anion and cation exchange, which can be carried out at
Biotechnology, Centre for Biochemical Engineering and Biotechnology,
different pHs, hydrophobic interaction chromatography, gel Millennium Institute for Cell Dynamics and Biotechnology (ICDB), University
filtration and affinity chromatography in addition to high of Chile, Beauchef 861, Santiago, Chile.
performance liquid chromatography (HPLC) and aqueous E-mail: [email protected]
two-phase partitioning. Evidently, when we are confronted with
a J. Asenjo, B. Andrews
a complex mixture of partially unknown proteins or a clarified cell
Department of Chemical Engineering and Biotechnology, Centre for
extract, there are many different routes one can take in order to Biochemical Engineering and Biotechnology, Millennium Institute for Cell
choose the minimum and most efficient number of purification Dynamics and Biotechnology (ICDB), University of Chile, Beauchef 861,
steps to achieve a desired level of purity. Santiago, Chile
65

J. Mol. Recognit. 2009; 22: 65–76 Copyright # 2008 John Wiley & Sons, Ltd.
 J. A. ASENJO AND B. A. ANDREWS

Figure 2. Determination of the Resolution between two peaks.

Figure 1. The combinatorial characteristic of choosing the sequence of
operations for protein purification. is defined as the product of the separation coefficient and this
relative concentration (SSC ¼ SC u ¼ DF
h
u).
adsorbent matrix and column dimensions. Composition is usually
determined in the production stage (fermentation, sometimes
Resolution and efficiency
cell disruption and recovery) and thus for a given adsorbent
matrix, operational conditions such as flow rate and ionic In chromatography, resolution is a variable used to measure
strength gradient have to be chosen in order to find a function column performance; it represents a means of interpreting
able to represent the performance of the process. Maximization column data and provides a basis for comparing results from
of this performance function can be carried out mathematically if different operating conditions (Leser et al., 1996). Considering
a model able to simulate the chromatography is available. two peaks in a chromatogram, chromatographic resolution is
Mathematical models for describing a chromatographic separ- defined as the distance between the peak maxima divided by the
ation can be classified, depending on the assumptions mean peak width, as shown in Figure 2 (Rs ¼ 2 (V2 –V1)/
considered in its derivation. Models such as the Plate Model (W1 þ W2)). As the separation coefficient is also a measure of the
can be used for predicting the retention time and the elution process performance, it can be assumed that it is proportional to
curve. More complex models are those based on thermodynamic the resolution (SC a Rs) (Asenjo and Andrews, 2004). When an
and transport phenomena; these models are called Rate Models. equal concentration of all proteins in a mixture is present, it can
For selecting the sequence of operations for high resolution be shown that the efficiency of the process can be defined
purification of proteins, there are a large number of options that (h ¼ SC/DF a Rs/DF). This concept is not based on rate or
can be chosen almost in any random order as shown in Figure 1 equilibrium analysis but corresponds to a semiempirical analysis
(Leser and Asenjo, 1994). This review paper describes how to use of separation of two components. Experimental data have shown
the physicochemical data of the product protein and other a relatively constant behaviour of the efficiency for each
proteins present (contaminants) to select an ‘optimal’ or particular separation process (Watanabe et al., 1994; Leser
suboptimal sequence of operations using the fewest number et al., 1996).
of operations and once the type of chromatography is chosen on
how optimization of operating conditions is carried out using
First algorithm: selection separation coefficients
appropriate mathematical models.
A DF for each individual property such as charge, molecular
weight and hydrophobicity of pairs of proteins was calculated.
SELECTION OF OPERATION The efficiency (h) reflects the unequal ability of different
separation processes (and/or different materials used) to exploit
A clear rationale for selection of high resolution purification differences in the deviation factor to separate the proteins. This
operations has been developed (Asenjo and Andrews, 2004). It value is relatively constant for each type of separation and
characterizes the ability of the separation operation to separate chromatographic material used and was found experimentally
one protein from another by using the theoretical concept of (Watanabe et al., 1994; Leser et al., 1996). The property chosen for
separation coefficients (Asenjo, 1990; Leser and Asenjo, 1992). It gel filtration was molecular weight (mw), for hydrophobicity it
uses a relationship between the separation coefficient was either the concentration of (NH4)2SO4 (M) at which the
(SC ¼ DF
h) and the variables that show the performance in a protein eluted or the chromatographic KD value using a specific
separation process: the deviation factor (DF) for differences hydrophobic matrix and a decreasing gradient of (NH4)2SO4. For
among physicochemical properties and the efficiency (h) of the ion-exchange chromatography (IEC), the property used to
process. DF has been defined as the difference in a particular evaluate the deviation factor was charge as a function of pH
physicochemical property (such as molecular weight, charge or and charge density. Clearly, charge density (charge/MW) gave a
hydrophobicity) between two proteins, which correspond to the better correlation as a function of retention time in ion-exchange
target protein and the particular contaminant protein being chromatography (Asenjo and Andrews, 2004).
considered (Asenjo and Andrews, 2004). To include the rule of To calculate the SSC, the system will read a database
thumb that reflects the logic of first separating impurities present containing the information on the properties of the main
in higher concentrations, a relative contaminant protein contaminant proteins present in the specific expression system
concentration (u) and the separation selection coefficient (SSC) used. The data on the 13 predominant proteins present in a
66

www.interscience.wiley.com/journal/jmr Copyright # 2008 John Wiley & Sons, Ltd. J. Mol. Recognit. 2009; 22: 65–76
PROTEIN PURIFICATION USING CHROMATOGRAPHY

Table 1. Values of S for the chromatographic processes used

in the system

Chromatographic process S
Size exclusion 0.46
Hydrophobic interaction 0.22
Ion exchange 0.15

the value of the SSC for all main contaminant proteins (e.g. 5, 10,
15) and all properties (e.g. 7–12) and choose the highest value of
the SSC to select the best separation process.

Second algorithm: elimination of protein contaminants; a

dynamic database
After each high resolution step, the concentration of contaminant
proteins decreases and the number of steps has to be sufficient
to eliminate contaminants until the product reaches the desired
level of purity. In order to find the new concentration of all
proteins after each separation step, a simple algorithm was
developed based on the behaviour of a chromatographic sepa-
ration that will give a very approximate value of the concen-
tration of each of the contaminants after each separation step.
The amount of a protein contaminant eliminated after a
chromatographic step is graphically shown in Figure 3 for three
Figure 3. Representation of the peaks of a chromatogram as triangles. different situations. Figure 3a shows how the representation of
(a) Adjusting a triangle to a peak. (b) Variation in DF leads to different the chromatographic peaks was simplified to a triangle. In
amounts of contaminant (triangle on the right) in the protein product Figure 3b the protein product corresponds to the triangle on the
(triangle on the left) (S ¼ 0.15 for ion exchange, 0.22 for HIC and 0.46 for left and the contaminant to the one on the right. The shaded area
GF).
(S ¼ ABC or ABCD) corresponds to the amount of contaminant left
commercial strain of Escherichia coli used for producing with the product in each case (Leser et al., 1996). The variable S,
recombinant proteins that was kindly donated by an industrial shown in Table 1, corresponds to the peak width and has been
source (Chiron) was used in the Expert System (Woolston, 1994; experimentally determined (Lienqueo et al., 1996). It was also
Asenjo and Andrews, 2004). This database was then used to found that under the conditions normally used for protein
select the first high resolution purification step. Thus, the first purification, which are well below saturation, the value of S is
algorithm that allows to choose a separation step would calculate virtually independent of protein concentration. Table 2 shows

Table 2. Concentration and relative concentration (%) of the main contaminants present in Escherichia coli and a model protein,
showing how these values evolve during a consultation

Loading After the first step After the second step After the third step

Weight 0 (g/L) Conc 0 (%) Weight 1 (g/L) Conc 1 (%) Weight 2 (g/L) Conc 2 (%) Weight 3 (g/L) Conc 3 (%)
Cont 1 11.24 14.97 0.22 1.62 0.00 0.00 0.00 0.00
Cont 2 7.06 9.40 0.06 0.44 0.00 0.00 0.00 0.00
Cont 3 4.63 6.17 0.24 1.76 0.01 0.19 0.01 0.20
Cont 4 5.58 7.43 0.11 0.81 0.00 0.00 0.00 0.00
Cont 5 4.83 6.43 0.09 0.66 0.00 0.00 0.00 0.00
Cont 6 2.48 3.30 0.04 0.29 0.00 0.00 0.00 0.00
Cont 7 7.70 10.25 0.02 0.11 0.02 0.39 0.00 0.00
Cont 8 6.80 9.05 0.13 0.95 0.00 0.00 0.00 0.00
Cont 9 7.53 10.03 7.56 55.51 0.15 2.89 0.00 0.00
Cont 10 6.05 8.06 0.12 0.88 0.01 0.19 0.00 0.00
Cont 11 3.89 5.18 0.00 0.00 0.00 0.00 0.00 0.00
Cont 12 1.48 1.97 0.02 0.15 0.00 0.00 0.00 0.00
Cont 13 0.83 1.11 0.01 0.07 0.00 0.00 0.00 0.00
Product 5.00 6.66 5.00 36.71 5.00 96.34 5.00 99.80
67

J. Mol. Recognit. 2009; 22: 65–76 Copyright # 2008 John Wiley & Sons, Ltd. www.interscience.wiley.com/journal/jmr
 J. A. ASENJO AND B. A. ANDREWS

Table 3. Sequence suggested by the Expert System to obtain a purity superior to 94% in the purification

SSC Criterion chromatography steps Purity Purity criterion chromatography steps Purity
Cation exchange at pH 6.0 33.1% Anion exchange at pH 7.0 63.7%
Hydrophobic interaction 49.5% Hydrophobic interaction 94.5%
Anion exchange at pH 7.0 97.0%

how the concentration in grams per litre and the relative shows that the system has the necessary robustness to variations
concentration (%) of the protein product (purity) and of the main and possible errors in the experimental determination of the data
contaminants present in E. coli change during a consultation with (<10%) but is sensitive enough to larger variations in protein
the Expert System (Leser, 1996). properties (>20%) (Asenjo and Andrews, 2004).
In order to consider affinity chromatography as a viable
separation, as in many cases suitable affinity ligands for the Experimental tests: purity criterion
protein product are well known, it was assumed that if this
technique is chosen by the user, all contaminants will be reduced Considering that a key value in protein purification is the purity
by a fixed percentage (e.g. 90%) in the affinity separation step level after each step and that now an algorithm has been
(Leser, 1996). However, since affinity chromatography will have to developed to calculate the purity after a purification step, this
be analysed on a case-by-case basis, given the nature of the criterion was also implemented as a possible selection criterion in
different ligands that can be used (e.g. metal ions in IMAC, dye or addition to the SSC. The purity criterion compares the purity level
other), it was not included in the Expert System described here. of the protein product obtained after a particular chromato-
graphic technique has been applied.
Purity is defined as
Robustness and sensitivity
Concentration of the target protein
A consultation was carried out using the Expert System to find all Purity ¼ (1)
S Concentration of all the proteins present
the steps necessary to achieve the desired level of purity (e.g.
98%) for the purification of the protein somatotropin produced in After determining which chromatographic technique gives the
E. coli. Once a process was found, the original databases were highest purity level, the system chooses this as the technique to
randomly varied at the levels of 10% and 20% to see the effect on use at this step. It then compares the purity with that required. A
the process proposed, in terms of its robustness and sensitivity, sequence of steps is chosen until the required level of final purity
by the system (Asenjo and Andrews, 2004). is reached. Finally, the system creates a list with the defined
The sensitivity of the proposed process to random changes in sequence of operations.
the values determined experimentally was investigated in order Two examples have been tested experimentally: a model
to assess the robustness of the system to either variations in the protein mixture and a recombinant b-1,3-glucanase from Bacillus
properties of the contaminant proteins present in the E. coli cells subtilis culture (Lienqueo et al., 1999). The model purification
used or in the experimental measurements. When only the E. coli mixture consisted of the purification of BSA (product) from three
data or both sets of data (E. coli and protein product) were contaminants (Soybean Trypsin Inhibitor (SBTI), Ovalbumin and
randomly varied at the level of 10% the sequence of operations Thaumatin). The results obtained for a target of 94% purity are
was exactly the same. On the other hand, when the data was shown in Table 3. The SSC criterion selects a purification
varied at the level of 20%, the sequence changed. This clearly sequence based on the elimination of the contaminant and

Table 4. Physicochemical properties and concentration for the main proteins in Bacillus subtilis ToC46(pFF1)

Charge (Coulomb molecule1) 1025

Initial concentration Molecular Hydrophobicity
Proteins (mg cm3) weight (Da) [(NH4)2SO4] pH 4.0 pH 5.0 pH 6.0 pH 7.0 pH 8.0
b-1,3-glucanase 0.60 31 000 0.00 1.46 0.62 1.02 2.33 2.52
Contaminants:
Low hydrophobic:
Contaminant 1 2.74 41 000 1.50 0.26 0.87 1.65 2.04
Contaminant 2 2.74 32 000 1.50 0.00 2.70 3.51 3.51
Medium hydrophobic:
Contaminant 3 0.25 35 500 0.20 0.55 0.22 0.73 1.82
High hydrophobic:
Contaminant 4 0.42 62 500 0.00 1.06 1.17 2.79 3.32
Contaminant 5 0.25 40 600 0.00 0.55 0.22 0.73 1.82
Contaminant 6 0.25 69 600 0.00 0.55 0.22 0.73 1.82
Contaminant 7 0.09 40 600 0.00 1.46 0.47 1.06 1.04
Contaminant 8 0.09 69 600 0.00 1.46 0.47 1.06 1.04
68

www.interscience.wiley.com/journal/jmr Copyright # 2008 John Wiley & Sons, Ltd. J. Mol. Recognit. 2009; 22: 65–76
PROTEIN PURIFICATION USING CHROMATOGRAPHY

Table 5. Sequence suggested by the Expert System for both criteria

Both criteria chromatography steps Purity Experimental validation chromatography steps Purity
Hydrophobic interaction 32.7% Hydrophobic interaction 33%–38%
Anion exchange at pH 6.5 70.3% Anion exchange at pH 6.5 65%–70%

property that gives the highest SSC value. On the other hand, the hydrophobicity proteins’ (contaminants 1 and 2) and part of
purity criterion chooses the optimum chromatographic step the ‘medium hydrophobicity proteins’ (contaminant 3) from the
considering all contaminants present. For this reason, in this b-1,3-glucanase. In this first step, the main contaminants were
particular case, the purity criterion gave a better result (only two eliminated. Figure 4b shows the separation of contaminants 3–4,
steps). In this case, when the SSC criterion is used, the first 5–6 and 7–8 from the b-1,3-glucanase. Figure 4 shows that the
separation step chosen is determined by the protein thaumatin scheme for purification suggested by the Expert System is valid
(Asenjo and Andrews, 2004). Since it has a very high pI (>8.0), in for purification of this recombinant b-1,3-glucanase.
this step no other protein is even partially purified. The second
step takes care of ovalbumin and the third one of SBTI. On the
other hand, when the purity criterion is followed, the first step MODELLING AND SIMULATION
(anion exchange at pH 7.0) takes care of ovalbumin as well as
thaumatin. This discrepancy between both criteria does not Once the type of chromatography is chosen, optimization of the
happen often. In this particular case, the disagreement was operating conditions is essential. In chromatography, protein
produced because of the extremely high pI of one of the adsorption depends on composition and concentration of the
contaminant proteins and partly also because all proteins were mixture and also on operation conditions such as flow rate, ionic
present in the same concentration, which does not often strength gradient, sample load, physical properties of the
correspond to a ’real’ situation of contaminants. adsorbent matrix and column dimensions. Mathematical models
for describing a chromatographic separation have been
discussed in the Introduction. Two such models are the Plate
Purification of b-1,3-glucanase and experimental Model and the more fundamentally based Rate Model.
investigation
Assessments were done using both the SSC criterion and the
Plate and rate models
purity criterion implemented in the Expert System for purification
of a b-1,3-glucanase from B. subtilis ToC46 (pFFI) culture. In this The Plate Model is based on the plate theory. Briefly, the model
case, both criteria gave exactly the same sequence. The data on assumes that the chromatographic column is formed by a
this system are given in Table 4. The results obtained for 70% number of plates (Np), each of them having the same ratio
purity are shown in Table 5 (Lienqueo et al., 1999). between the stationary phase volume and the volume of the
The chromatograms from the purification sequence are shown mobile phase (H). For a defined column geometry and if the
in Figure 4a,b and Table 5. Figure 4a shows the separation of ‘low adsorption kinetics is known, the problem is reduced to solving

Figure 4. Steps suggested by both criteria for the purification of glucanase (o): glucanase activity. (a) First step suggested: hydrophobic interaction
chromatography. (b) Second step suggested: anion-exchange chromatography at pH 6.5.
69

J. Mol. Recognit. 2009; 22: 65–76 Copyright # 2008 John Wiley & Sons, Ltd. www.interscience.wiley.com/journal/jmr
 J. A. ASENJO AND B. A. ANDREWS

Table 6. Equations of the Plate Model

Protein at each plate (i ¼ 1,..Np)

dKðCi ;IÞ dI Vm
dCi Np ðCi1  Ci Þ  Ci H dI du u¼ C0 ¼ C1 ¼ . . . ¼ CNp ¼ 0
¼ V0
du d Vm
1 þ H½KðCi ;IÞ þ Ci KðCi ;IÞ  < u  0 C0 ¼ 1
dCi V0
u>0 C0 ¼ 0
Ionic strength at each plate (i ¼ 1,..Np)
i
I0 for u  ð1 þ HKsalt Þ

 N p
Ii ¼
i i
I0 þ G u  ð1 þ HKsalt Þ for u > ð1 þ HKsalt Þ
Np Np

the system of Np ordinary differential equations (ODE) shown in subject to the initial and boundary conditions given by
Table 6. In order to solve this ODE system, the ionic strength at
each plate (i ¼ 1. . .Np) has to be computed as a function time. t ¼ 0 cp ¼ cp ð0; r; zÞ
Table 6 shows the formula for computing this variable in the case @cp
that a constant ionic strength gradient is applied for protein r¼0 ¼0 (5)
@r
elution (Yamamoto et al., 1983).
At low protein concentration, the adsorption kinetics is @cp
r¼1 ¼ Bi½cb ðz; tÞ  cp ðr ¼ 1; z; tÞ
computed from the value of the distribution coefficient (K) that @r
depends on the ionic strength of the mobile phase (I). Protein
In relationship (3), Cf(t), the time-dependent feeding concen-
displacement in ion-exchange chromatography is due to changes
tration (for a protein, Cf(t), will be different from zero while the
in the ionic strength of the mobile phase and thus the
sample is loaded into the column; for the displacer, the feeding
distribution coefficient and the number of plates cannot be
concentration is often a function of time). Dimensionless
computed from the first and second normalized central moment
variables and parameters in relationships (2)–(5) are shown in
of the elution curve. However, because during the travelling of
Table 7. Since all mass transfer phenomena are taken into account
the protein through most part of the column, the protein zone is
in partial differential Equations 2 and 4, Rate Models can be used
subject to an ionic strength near to the one at which this emerges
for testing different chromatographic conditions (Gu, 1995; Lazo,
from the column (Imax), a relationship has been presented for
1999).
computing the number of plates (Yamamoto et al., 1983, Shene
Operational conditions in chromatography such as flow rate
et al., 2006):
and ionic strength gradient are taken into account in both
In the more fundamentally based Rate Model the dimension-
mathematical models and thus predicted elution curves depend
less elution curves are obtained from the solution of the following
on them. However, the quality of the product obtained in
partial differential equation:
chromatography is also dependent on external operational
conditions such as the flow rate as well as the size of the fraction
@cb @cb 1 @ 2 cb of the protein product collected, as shown in Figure 5 (also called
¼ þ  ji ðcb  cp;r¼1 Þ (2)
@t @z PeL @z 2 peak cutting). From the scheme presented in Figure 5, if the
target protein is protein A, the outlet flow can be collected from
subject to the initial and boundary conditions given by
t  ti until t  te, the period during which the concentration of A in
the outlet flow becomes important. However, during this time, a
t¼0 cb ¼ cb ð0;zÞ part of the contaminants is also eluted. A way to minimize the

 contaminant content in the collected volume is by decreasing the
@cb Cf ðtÞ
z¼0 ¼ PeL cb ð0;tÞ  (3) collecting interval considering the time elapsed between t1 and t2
@z Co
(Figure 5).
@cb
z¼L ¼0 In order to define a performance function for a chromatog-
@z raphy that can be used for choosing operational conditions for
the separation of a given protein mixture, several parameters
In order to solve the partial differential in Equation 2, the
should be taken into account:
dimensionless concentration profile for each component in
the liquid phase contained inside the particles, cp, has to be (a) Concentration of the target protein, xA which is given by:
computed. These concentration profiles are obtained from the R t2 R t2
solution of the following partial differential equation: t1 CA
Fdt CA dt
xA ¼ R t2 ¼ tR1 t1 (6Þ
  t1 Fdt t2 dt
@ 1 @ 2 @cp
ð"p cp þ ð1  "p Þcp Þ ¼ h 2 r (4) As shown in Figure 5 concentration xA depends on the
@t r @r @r
collecting time and on the resolution of the purification stage
70

www.interscience.wiley.com/journal/jmr Copyright # 2008 John Wiley & Sons, Ltd. J. Mol. Recognit. 2009; 22: 65–76
PROTEIN PURIFICATION USING CHROMATOGRAPHY

Table 7. Dimensionless variables and parameters of the Rate Model

Concentration of the mobile phase Cb

cb ¼
Cb
Concentration of the liquid inside the adsorbent particles Cp
cp ¼
Cp
Concentration of the adsorbed protein Cp
cp ¼
Co
Dimensionless time vt
t¼
L
Dimensionless position in the column Z
z¼
L
Dimensionless position in the particle R
r¼
Rp
Peclet number v
L
PeL ¼
Dz
Biot number k
Rp
Bi ¼
"p
Dp
" p
Dp
L
h¼
Rp
2
v
3
Bi
h
ð1  "b Þ
j¼
"b

fixed by the flow rate and the ionic strength gradient. Costs protein loaded into the column:
involved in the afterward concentration processes (ultrafil- R t2
tration, lyophilisation) are related to the value of xA. CA
Fdt
yieldA ¼ t1 (8Þ
(b) Purity of protein (A) is defined as the ratio between the mass CA0 V0
of protein A and that of all the proteins in the collected
The yield of a chromatography depends on the collecting
volume:
time and it can be used to estimate costs of the production
xi stages (fermentation).
Purity of the i component ¼ Pm (7Þ
j¼1 xj (d) Process time, is defined as the time at which all the proteins in
the mixture loaded into the column are eluted and after
Purity not only allows to establish the relevance of the which the column can be prepared for the treatment of a new
chromatography as a purification stage but it can also be load. Process time can be used as an estimation of the costs
used for estimating costs involved in the rest of the purifi- involved in the chromatography stage.
cation steps.
(c) Yield is defined as the ratio between the mass of the target Plate and Rate Models have been used for simulating elution
protein in the collected volume and the mass of the same curves of a three-protein mixture in IEC carried out under
different operational conditions (flow rates and ionic strength
gradients). Q Sepharose FF was used as the adsorbent matrix. A
cost function for the protein production process was proposed
and flow rate, ionic strength gradient and collection time are
selected in order to minimize the cost function for different type
of protein products.

Simulation
Elution curves of the three-protein mixture in IEC were
experimentally recorded for two values of the ionic strength
gradient (g) and different flow rates (F). Values for the purity and
retention times obtained from the IEC elution curves simulated
using the Plate and Rate Models are shown in Tables 8 and 9,
respectively.
Comparisons between experimental and simulated elution
curves and the ionic strength profile computed using the Plate
Model for the different flow rates and an ionic strength gradient
Figure 5. Elution curves of a mixture of proteins. A is the target protein of 0.055 M/mL is shown in Figure 6. The deviation between the
product and f the fraction of peak A collected. experimental and simulated elution curves is presented in
71

J. Mol. Recognit. 2009; 22: 65–76 Copyright # 2008 John Wiley & Sons, Ltd. www.interscience.wiley.com/journal/jmr
 J. A. ASENJO AND B. A. ANDREWS

Table 8. Results of the simulations for the separation of a three-protein mixture in IEC using the Plate Model for different flow rates,
F, and ionic strength gradients, g

Purity (%) Retention time (min)

Run F (mL/min) g (M/mL) P1 P2 P3 P1 P2 P3 Deviation

1-a 0.3 0.100 54.4 49.4 80.6 12.2 13.3 16.5 0.0149
1-b 0.7 0.100 50.9 42.6 54.8 5.3 5.8 7.1 0.0093
1-c 1.0 0.100 50.3 39.8 47.1 3.7 4.1 5.0 0.0109
2-a 0.3 0.055 63.6 54.0 84.6 17.4 19.8 24.9 0.0189
2-b 0.5 0.055 59.9 48.1 71.3 10.5 11.9 15.0 0.0053
2-c 1.0 0.055 55.2 42.0 51.9 5.2 6.0 7.5 0.0064
P1, a-lactoalbumin; P2, ovalbumin; P3, b-lactoglobulin.

Table 9. Results of the simulations for the separation of a three-protein mixture in IEC using the Rate Model for different flow rates, F,
and ionic strength gradients, g

Purity (%) Retention time (min)

Run F (mL/min) g (M/mL) P1 P2 P3 P1 P2 P3 Deviation

1-a 0.3 0.100 64.3 63.1 78.7 12.5 13.8 16.6 0.0192
1-b 0.7 0.100 55.5 55.5 59.6 5.3 5.9 7.1 0.0111
1-c 1.0 0.100 53.0 52.2 74.3 3.7 4.1 5.0 0.0152
2-a 0.3 0.055 78.3 47.5 67.9 17.0 19.6 24.4 0.0252
2-b 0.5 0.055 65.3 55.2 58.0 10.2 11.8 14.6 0.0110
2-c 1.0 0.055 56.8 48.5 60.4 5.0 5.9 7.3 0.0059
P1, a-lactoalbumin; P2, ovalbumin; P3, b-lactoglobulin.

Figure 6. Experimental and simulated elution curves of the three protein mixture using the Plate Model (P1: a-lactoalbumin; P2: ovalbumin; P3:
b-lactoglobulin) for an ionic strength gradient of 0.055 M/mL and different flow rates. (a) 0.3 mL/min NP1 ¼ 18; NP2 ¼ 6; NP3 ¼ 6 (b) 0.7 mL/min NP1 ¼ 11;
NP2 ¼ 4; NP3 ¼ 4 (c) 1 mL/min NP1 ¼ 6; NP2 ¼ 2; NP3 ¼ 2.
72

www.interscience.wiley.com/journal/jmr Copyright # 2008 John Wiley & Sons, Ltd. J. Mol. Recognit. 2009; 22: 65–76
PROTEIN PURIFICATION USING CHROMATOGRAPHY

Figure 7. Experimental and simulated elution curves of the three-protein mixture using the Rate Model (P1: a-lactoalbumin; P2: ovalbumin; P3:
b-lactoglobulin) obtained for an ionic strength gradient of 0.1 M/mL and different flow rates. (a) 0.3 mL/min hP1 ¼ 10; hP2 ¼ 9.5; hP3 ¼ 4.5 (b) 0.7 mL/min
hP1 ¼ 6; hP2 ¼ 6; hP3 ¼ 3 (c) 1 mL/min hP1 ¼ 4; hP2 ¼ 4; hP3 ¼ 4.

Table 8. Maximum value for the deviation was 0.0189 (absorbance chosen, the peak width increases, the maximum protein
units). concentration decreases and the retention time increases. Since
Results in Figure 6 and Table 8 indicate that the Plate Model can IEC is in many cases one step in a protein purification process, its
be used for simulating the elution curve of a protein mixture in output will affect other steps. The best way of relating how the
IEC. It is important to note that all parameters in the model such results obtained in one chromatography step (process time,
as those needed for computing the number of plates and the concentration of the target protein in the collected volume and
distribution coefficients for each protein were obtained form purity and yield of the target protein) affect other stages in the
independent experiments. purification process is through a cost function since in many
The comparison of the experimental and simulated elution cases the value of the product is fixed by the market and thus an
curves and ionic strength profile computed using the Rate Model important way to increase the profit is through the reduction of
is shown in Figure 7. In these simulations the adsorption kinetics the processing costs. A cost function for a protein production
and parameters were used (Shene et al., 2006). Table 9 shows the process and how the chromatography performance affects it,
deviation between simulated and computed values; the similar to that proposed previously (Huenupi et al., 1999) for a
maximum deviation was 0.0252 (absorbance units). From protein extraction process, can be defined as follows:
the comparison of the results presented in Tables 8 and 9, the  
average prediction deviation obtained with the Rate Model was B1 Purity B3 tend
Cost ¼ a1 þ a2 1  þ a3 þ a4 (9)
slightly higher than that obtained using the Plate Model. Yield B2 CA B4
The Rate Model has several advantages over the Plate Model,
the most important being that it can be extended for simulating The first term in Equation 9 takes into account costs involved in
elution curves of more concentrated protein mixtures, where the fermentation in such a way that the chromatography’s step
protein interaction effects could be significant and more complex yield decreases and more protein mixture will be needed to fulfill
relationships for the adsorption kinetics must be used. Never- the required production level. The second term represents costs
theless, for the case under study the Plate Model is easier to involved in further purification steps, for instance, hydrophobic
implement computationally and also has a lower CPU demand interaction chromatography (Asenjo and Andrews, 2004). As the
due to the small size of the ODE system that has to be solved. purity of the product eluted from the chromatography increases,
these costs decrease and they become equal to zero in the case
where the separation is accomplished in this stage only. Costs
OPTIMAL SELECTION OF OPERATING related to concentration processes, such as dialysis, ultrafiltration
CONDITIONS or freeze–drying, are inversely related to the concentration of the
product obtained in the chromatography, which corresponds to
Results in Tables 8 and 9 show that flow rate and the ionic the third term in Equation 9. The last term in Equation 9 takes into
strength gradient affect the purity and retention time of the account the costs of the chromatography determined by the
different proteins in the mixture. A higher purity is obtained by processing time. A longer processing time may result in higher
using a small ionic strength gradient for a given flow rate. resolution but this will determine the use of a larger unit or more
However, when a small value of the ionic strength gradient is than one unit in parallel, thus increasing the cost for the specified
73

J. Mol. Recognit. 2009; 22: 65–76 Copyright # 2008 John Wiley & Sons, Ltd. www.interscience.wiley.com/journal/jmr
 J. A. ASENJO AND B. A. ANDREWS

production level. Values of coefficients a1, a2, a3 and a4 in Table 10. Operational conditions, protein yield and purity in
relationship (9) give the relative weights to the different terms in a chromatography separation for minimum production cost
the cost function; the sum of these coefficients is constrained to based on the cost function given by Equation 9 and different
be 1. Parameters B1–B4 in relationship (9) are introduced in order relative weights (ai) for the different production stages
to scale the different variables. Values for these parameters will
depend on the system geometry and the range of operational
conditions that can be used in a given system. Relationship (9) Case 1 Case 2 Case 3
states that costs of the different stages, given by the different a1 0.30 0.55 0.30
terms, are linearly related to the variables. However, scaling a2 0.30 0.20 0.50
indexes similar to those used for equipment scale-up (exponents a3 0.10 0.05 0.05
in the different terms) can be introduced in order to build a more a4 0.30 0.20 0.15
rigorous model (Huenupi et al., 1999). Min Cost 0.593 0.651 0.510
The cost function in relationship (9) was evaluated considering Operational conditions*
the case in which the target protein is ovalbumin, a protein F (mL/min) 0.6 0.6 0.3
whose retention time was found to be between those of the g (M/mL) 0.105 0.105 0.065
other two proteins in the mixture, as a way to consider the worst f (—) 0.50 0.75 0.45
case in a given protein purification process. It was assumed that Results
flow rate and ionic strength gradients are constrained to take Yield 0.8195 0.9455 0.7605
values between 0.3 and 1.0 mL/min and 0.055 and 0.105 M/mL, Purity (%) 53.10 47.32 73.67
respectively. From the scheme shown in Figure 5, the fraction ( f ) CA (mg/mL) 0.0260 0.0200 0.0243
of the peak collected is given by tend (min) 9.82 9.82 25.34
t2  t1 *Operational conditions were constrained to take values of
f ¼ (10) 0.3–1.0 mL/min for F, 0.055 –0.105 for g and 0.0–1.0 for f (with
te  ti increments of 0.1, 0.005 and 0.05, respectively).
For the range of operational conditions, tested values for
parameters B1, B2, B3 and B4, in relationship (9) were chosen so
that each term in expression (9) could reach a maximum value of
1. Hence B1, B2, B3 and B4 were 0.72, 77%, 0.0345 g/L and
28.54 min, respectively. Case 2 (Table 10) shows the operational conditions in the
Values of coefficients a1, a2, a3 and a4 depend on the chromatography for the production of a target protein having
characteristics of the target protein such as the required final very high fermentation costs (ai ¼ 0.55). Subsequent purification
purity and its synthesis during the fermentation. Costs involved in stages (a2 ¼ 0.2) and those involved in the IEC (a4 ¼ 0.2) are of the
a fermentation process for protein synthesis can be assumed to same magnitude and lower than in the previous case. This could
represent between 30 and 70% of total production cost (Huenupi be the case of an intracellular target protein with low substrate to
et al., 1999). Purification costs (a2 þ a4) can represent between target protein yield. In this case costs for cell disruption and
10–50% of a protein production process. Costs for the separation from cell debris are assumed to be included in those
concentration stages (a3) can be considered lower than those for the fermentation. Results indicate that during the IEC, a higher
involved in the purification stages (between 10 and 30%). fraction of the volume should be collected ( f ¼ 0.75). This results
in an even lower purity and higher yield than in the previous case.
Chromatography operational conditions for the production of
Selection of peak size (peak cutting)
a target protein required with a high final purity, for instance, a
Simulations were carried out using different values of the flow pharmaceutical product, are shown in Case 3 (Table 10). Costs
rate and ionic strength gradients and the cost function was involved in the subsequent purification stages are set as 50% of
evaluated for different values of f [fraction of product peak the total production costs (a2 ¼ 0.5). Operational conditions in
collected (Eq. 10)]. Three different combinations for the ai the chromatography stage for this case are those of a product
(i ¼ 1,..,4) coefficients were considered in order to simulate with a high purity (73.67%). In order to achieve the high purity, IEC
conditions for the production of different types of target proteins. must be carried out at low flow rate (0.3 mL/min) and with small
Table 10 shows the flow rate, ionic strength gradient and the ionic strength gradient (0.065 M/mL). As fermentation costs are
value for f found, for which the minimum value of each of the cost relatively low, collection of the eluted protein corresponds to a
functions was obtained. smaller fraction of total elution time ( f ¼ 0.45) in order to obtain a
In Case 1, the operational conditions for the chromatography high purity protein fraction. This results in a higher purity and
in a process in which costs for the fermentation process, lower yield than in the previous two cases and a longer
subsequent purification steps, and the chromatography contrib- processing time that results in a higher resolution. The actual
ute in the same degree to the total production costs fractions collected in all cases are clearly shown in Figure 8.
(ai ¼ a2 ¼ a4 ¼ 0.30) are presented (0.30). This could be the
case of an enzyme required with a low purity and for which
Use of mathematical models for process validation
purification is carried out in order to eliminate contaminants that
decrease its activity, for instance an industrial enzyme. For this A recent paper (Kaltenbrunner et al., 2007) shows how the Plate
case flow rate, ionic strength gradient and fraction collected were Model has been used for an industrial practical application of
equal to 0.6 L/min, 0.105 M/mL and 0.5, respectively. As expected chromatographic theory for process characterization towards
the final purity is relatively low, 53.1%, and a high yield is validation of an ion-exchange operation. When a chromato-
obtained. graphic operation utilized to purify a human therapeutic protein
74

www.interscience.wiley.com/journal/jmr Copyright # 2008 John Wiley & Sons, Ltd. J. Mol. Recognit. 2009; 22: 65–76
PROTEIN PURIFICATION USING CHROMATOGRAPHY

Figure 9. Process flow chart from a decision to commercialize a bio-

logic to product launch. The sub-processes of process characterization
and screening experiments are outlined (Kaltenbrunner et al., 2007).

chromatographic theory to prioritize operating parameters. Both

methodologies, empirical and rational, were performed to
evaluate a specific ion-exchange chromatography operation
for the preparative separation of closely related protein species.
The paper shows that the application of the rational approach
has the potential to accelerate the evaluation and significantly
reduce the amount of analytical testing needed.

Figure 8. Elution curves of the mixture of proteins in a chromatography

separation for the minimum production cost given in Table 8. Between CONCLUSIONS
the arrows, the fraction of protein product collected for (a) Case 1, (b) Case
2 and (c) Case 3. The three cases correspond to protein products with In this paper we have reviewed and discussed the proteomic
different characteristics as described in the text. approach to select a purification process for proteins based on
physicochemical properties as well as the use of mathematical
modelling for the optimization of performance (actual operating
conditions) in a chromatographic step. The methodology
is prepared for validation before commercial production, described constitutes a rational proteomic procedure to separate
numerous tests have to be performed to establish the relative the main contaminant proteins with a minimum number of steps
importance of each operating parameter to define its future role and the optimization of such steps.
and importance in the framework on in-process controls. This An algorithm to calculate the SSC, a parameter used to select
prioritization process is usually performed using an entirely the actual purification at each step was developed, and the
empirical approach. The process flow chart from a decision to translation of physicochemical data of the proteins to chromato-
commercialize a biologic to product launch is shown schema- graphic behaviour was also carried out for ion-exchange
tically as a process flow-sheet in Figure 9. This paper chromatography, hydrophobic-interaction chromatography and
demonstrates the application of a rational approach based on gel filtration.
75

J. Mol. Recognit. 2009; 22: 65–76 Copyright # 2008 John Wiley & Sons, Ltd. www.interscience.wiley.com/journal/jmr
 J. A. ASENJO AND B. A. ANDREWS

Another algorithm used to estimate concentration of each mathematical models. These models were the Plate Model and
protein contaminant after a chromatographic process is the more fundamentally based Rate Model. Simulated elution
performed was also developed. The methodology described, curves were compared with experimental data not used for
which was handled by a computer-based Expert System, was parameter identification. Deviation between experimental data
tested with recombinant proteins produced in E. coli, with a good and the simulated curves using the Plate Model was less than
database for the main protein contaminants and purification of a 0.0189 (absorbance units); a slightly higher deviation [0.0252
recombinant protein product with good results. (absorbance units)] was obtained when the Rate Model was used.
The system was robust to errors <10%, which is the range that Simulation of IEC for protein purification can be used as a tool
can be found in the experimental determination of the properties for choosing operational conditions such as flow rate, ionic
in the database of product and contaminants. On the other hand, strength gradient and the externally fixed operational condition
the system was sensitive, both to larger variations (>20%) in the that in this work was termed the collecting time (fraction
properties of the contaminant database and the protein product. collected, f ). In order to do this, a performance function for
The purification strategy proposed was experimentally tested ion-exchange chromatography has to be defined. However, since
and validated with a mixture of four proteins and the a purification stage such as ion-exchange chromatography is
experimental validation was also carried out with a supernatant integrated into the protein production process, its performance is
of B. subtilis producing a recombinant b-1,3-glucanase. affected by previous and possibly subsequent processing and
In addition to SSC, final purity can also be used as a selection purification stages. In this work a cost function for the whole
criteria, given the fact that it is also calculated after each protein production process that can be used for the selection of
separation step is performed, to give the new protein the operational conditions as well as the fraction of the product
contaminant concentrations in the database. Although both to be collected (peak cutting) in chromatography was built and
criteria, SSC and purity, will in most cases give similar results, tested. This function can be used for protein products with
purity may give fewer steps (and thus a better process) when different characteristics and qualities such as purity and yield by
concentrations of contaminant proteins are similar in the crude choosing the appropriate parameters.
starting material.
Once the type of chromatography is chosen, optimization of
the operating conditions is essential. Chromatographic elution
curves for a three-protein mixture (a-lactoalbumin, ovalbumin
Acknowledgements
and b-lactoglobulin), carried out under different flow rates and The authors acknowledge support of the Millennium Scientific
ionic strength conditions, were simulated using two different Initiative (project P05-001-F) and Fondef (project D04I1374).

REFERENCES

Asenjo JA. 1990. Selection of operations in separation processes. In Leser EW, Lienqueo ME, Asenjo JA. 1996. Implementation in an
Separation Processes in Biotechnology. Asenjo JA (ed.) Marcel Dekker. expert system of selection rationale for purification processes
New York; 3–16. for recombinant proteins. Ann. N.Y. Acad. Sci. 782: 441–
Asenjo JA, Andrews BA. 2004. Is there a rational method to purify proteins? 455.
From expert systems to proteomics. J. Mol. Recognit. 17: 236–247. Lienqueo ME, Leser EW, Asenjo JA. 1996. An expert system for the
Gu T. 1995. Mathematical Modeling and Scale-Up of Liquid Chromatog- selection and synthesis of multistep protein separation processes.
raphy. Springer: Berlin. Comput. Chem. Eng. 20: S189–S194.
Huenupi E, Gómez A, Andrews BA, Asenjo JA. 1999. Optimization and Lienqueo ME, Salgado JC, Asenjo JA. 1999. An expert system for selection
design considerations of two-phase continuous protein separation. of protein purification processes: experimental validation. J. Chem.
J. Chem. Technol. Biotechnol. 74: 256–263. Technol. Biotechnol. 74: 293–299.
Kaltenbrunner O, Giaverinni O, Woehle D, Asenjo JA. 2007. Application of Shene C, Andrews BA, Lucero A, Asenjo JA. 2006. Mathematical modelling
chromatographic theory for process characterization towards vali- of elution curves for a protein mixture in ion exchange chromatog-
dation of an ion-exchange operation. Biotechnol. Bioeng. 98: 201–210. raphy and for the optimal selection of operational conditions. Bio-
Lazo C. 1999. Simulation of Liquid Chromatography and Simulated technol. Bioeng. 95: 704–713.
Moving Bed (SMB) Systems. Sudienarbeit Technische Universität: Watanabe E, Tsoka S, Asenjo JA. 1994. Selection of chromatographic
Hamburg-Harburg. protein purification operations based on physicochemical properties.
Leser EW. 1996. Prot_Ex: an expert system for selecting the sequence of Ann N.Y. Acad. Sci. 721: 348–364.
processes for the downstream purification of proteins. Ph.D. Thesis, Woolston PW. 1994. A physicochemical database for an expert system for
University of Reading. the selection of recombinant protein purification processes. Ph.D
Leser EW, Asenjo JA. 1992. The rational design of purification processes Thesis, University of Reading.
for recombinant proteins. J. Chromatogr. 584: 35–42. Yamamoto S, Nakanishi K, Matsuno R. 1983. Ion exchange chromatog-
Leser EW, Asenjo JA. 1994. The rational selection of purification processes raphy of proteins predictions of elution curves and operational
for proteins: an expert system for downstream processing design. conditions. I. theoretical considerations. Biotechnol. Bieng. 25:
Ann. N.Y. Acad. Sci. 721: 337–347. 1465–1483.
76

www.interscience.wiley.com/journal/jmr Copyright # 2008 John Wiley & Sons, Ltd. J. Mol. Recognit. 2009; 22: 65–76