Journal of Chromatography A, 1375 (2015) 3341

Contents lists available at ScienceDirect

Journal of Chromatography A
journal homepage: www.elsevier.com/locate/chroma

Linear isotherm determination from linear gradient elution

experiments
David Pster, Fabian Steinebach, Massimo Morbidelli
Institute for Chemical and Bioengineering, Department of Chemistry and Applied Bioscience, ETH Zurich, 8093 Zurich, Switzerland

a r t i c l e

i n f o

Article history:
Received 27 August 2014
Received in revised form
25 November 2014
Accepted 25 November 2014
Available online 2 December 2014
Keywords:
Chromatography model
Linear gradient elution
Exclusion effect
Parameter estimation

a b s t r a c t
A procedure to estimate equilibrium adsorption parameters as a function of the modier concentration in
linear gradient elution chromatography is proposed and its reliability is investigated by comparison with
experimental data. Over the past decades, analytical solutions of the so-called equilibrium model under
linear gradient elution conditions were derived assuming that proteins and modier molecules access
the same fraction of the pore size distribution of the porous particles. The present approach developed in
this work accounts for the size exclusion effect resulting in different exclusions for proteins and modier.
A new analytical solution was derived by applying perturbation theory for differential equations, and the
1st-order approximated solution is presented in this work. Eventually, a turnkey and reliable procedure
to efciently estimate isotherm parameters as a function of modier concentration from linear gradient
elution experiments is proposed.
2014 Elsevier B.V. All rights reserved.

1. Introduction
Ion exchange chromatography (IEC) is a well-known and wellestablished technique in the downstream processing of therapeutic
proteins [13]. However, despite the great popularity of chromatographic processes, their development and up scaling are still often
based on heuristic knowledge and trial-and-error approaches [4].
Such design procedures are often company-specic, if not personspecic, and are therefore not answering anymore to the call of
the regulatory authorities for standardizing and generalizing the
process development procedures. Moreover, the trial-and-error
approach might become more and more costly and challenging with process scale-up. Therefore, model-based design could
provide competitive advantages in terms of time and efciency by
reducing the number of experiments to perform. Indeed, mechanistic models only require a limited number of experiments
to calibrate the model parameters. However, the model-based
approaches cannot be the silver bullet and despite their great
potential the bottleneck of this approach is the determination of
the model parameters whose reliability will directly be reected
into the reliability of the model.

Corresponding author at: Institute for Chemical and Bioengineering, ETH Zurich,
Vladimir-Prelog-Weg 1, CH-8093 Zurich, Switzerland. Tel.: +41 44 632 30 34;
fax: +41 44 632 10 82.
E-mail address: massimo.morbidelli@chem.ethz.ch (M. Morbidelli).
http://dx.doi.org/10.1016/j.chroma.2014.11.067
0021-9673/ 2014 Elsevier B.V. All rights reserved.

In this work, we focus on the determination of reliable equilibrium parameters of linear isotherms as a function of modier
concentration. Two different approaches are reported in the literature to obtain the Henry coefcient as a function of the modier
concentration: pulse injection followed by isocratic elution or linear gradient elution. Among the two, gradient elution is the most
convenient one, especially for proteins which suffer from severe
mass transfer hindrances. Indeed, the compression induced by the
salt gradient leads to sharper peaks thus improving the detectability and the precision of the measurements [5]. Moreover, and not
the least, the use of a gradient makes the methods less sensible
to experimental error in the buffer preparation compared to isocratic elution. Yamamoto et al. [68] suggested a simple technique
to estimate equilibrium parameters from a set of pulse injections
eluted at different gradient slopes. This technique has become very
popular over the past decades.
The novelty of this work is to rigorously account for the size
exclusion effect [9,10] thus leading to a new and more reliable
analytical solution of the equilibrium model. In particular, in the
Yamamoto et al. procedure it is assumed that both modier and
protein access the entire pore size distribution [68]. For small
peptides, or particles with large pores this assumption is reasonable and leads to very good results. However, it is often observed
that proteins are excluded from the smallest pores while the modier is not, and therefore a new estimation procedure is needed
which accounts for differences in size between modier and proteins. Obviously, it is much more advantageous to have an analytical

34

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

solution than numerically integrating the system of partial differential equations describing the solute transport in a packed bed.
The computation time is indeed dramatically reduced.
In particular, in this work we propose a more accurate analytical
solution for evaluating the retention time in linear gradient elution
chromatography. The isotherm parameters are obtained by tting
the present model on a set of experimental data. These estimated
parameters where then compared to the true ones obtained by
the Inverse Method (IM) and to the ones obtained by the classical
Yamamotos approach.
2. Theory
2.1. Equilibrium model
At equilibrium, the concentration of solute retained in the stationary phase (q) is related to the concentration of solute in the
mobile phase (c) by a soliduid equilibrium law named isotherm.
At low solute concentration this isotherm is often assumed to be
linear and the initial slope of the isotherm, known as the Henry
coefcient, is dened as follows:
q = Hc

K=

(2)

where K is the retention factor at innite modier concentration

dened by K = p , and and are the two parameters describing
the retention factor dependency on modier concentration. cM (z, t)
is the modier concentration at time t and position z in the column.
Particular attention should be paid on the different porosities
involved when dealing with porous particles. Let us rst dene the
bed porosity, , as the ratio between the volume of liquid outside the
porous particles and the volume of the column. Therefore, (1 )
represents the volume of solid (particles skeletons) plus the volume of liquid that stagnates in the particles pores over the column
volume. In addition, the intra-particle porosity, p , represents the
volume fraction of the porous particle accessible to the solute. Usually, when dealing with macromolecules whose radii are close from
the average radius of the pores, p is dened in terms of accessible porosity, p,i . Therefore, p,i (specic to the macromolecule) is
ranging from p obtained for small non-excluded tracers to 0 for
completely excluded molecules, which leads to dene K,i = p,i .
The total porosity is dened by the sum of the volumes of liquid
(mobile and stagnant) over the volume of the column:
t = + (1 )p

(3)

Or for partially excluded macromolecules:

t,i = + (1 )p,i

chromatographic column is described by the following equation

[13]:

(1)

H depends on the protein, the stationary phase, the ionic strength,

the temperature and the pH of the mobile phase. In classical ion
exchange chromatography, a salt (typically NaCl) is used as modier to modulate the retention of the solute. The dependency of the
Henry coefcient (or the retention factor K = H, where  = (1 )/
is the phase ratio) towards the modier concentration is usually
described by the stoichiometric displacement model (SD) [11,12]:

K + cM

Fig. 1. Total accessible porosity of the different stationary phases.

(4)

The fraction of the pore size distribution accessible to the protein

obviously depends on the size of the macromolecule and on the
pore size distribution of the stationary phase [9]. Fig. 1 shows the
porosity proles of the four stationary phases investigated in this
work.
In the frame of the equilibrium model, the chromatographic column is modelled as an ideal plug ow (negligible mass transfer
limitations and no axial dispersion) and the solute concentration,
both in the mobile uid and in the particles is assumed to be at
equilibrium, as dened by Eq. (2). Under these simplifying assumptions, the mass balance equation for a solute (or the modier) in the

c
c
q
+
+ (1 )
=0
z
t
t

(5)

which using Eqs. (1) and (2) reduces to the following equation:
c
K
c
+ (1 + K)
= c
u t
u t
z

(6)

where u is the linear velocity dened by u = Q/A. Eq. (6) can be solved
using the method of characteristics. This method has already shown
its great potential for solving rst order PDEs. The basic principle is
to solve the PDE equation on characteristic curves along which the
PDE can be written as a system of ordinary differential equations.
The transformation is obtained by constructing the characteristic
curves parameterized by s so that the curve {z(s), t(s), c(s)} satises
the set of characteristic equations dened by the following ODEs:

dz

ds = 1

dt

= (1 + K)

(7)

u
ds

dc

dK
= c
ds

u dt

The ratio between the second and the rst lines of Eq. (7) leads to
the equation describing the evolution of the solute retention time
along the column axis z. The resolution of the other equations was
not considered in this work. It comes:
dt

= (1 + K)
u
dz

(8)

Eq. (8) describes the evolution of the retention time from the
inlet of the column (z = 0) to the position z for a solute whose equilibrium with the stationary phase is given by Eq. (1).
Under the assumption of constant retention factor for the modier and in the absence of interactions with the ion exchange groups
(meaning that K = K = p ) the integration of Eq. (8) between 0 and
z is obvious. It comes:

t0,M (z) = (1 + K )z
(9)
u
where t0,M (z) is the time needed by the modier to cover the distance from 0 to z. Therefore, for linear gradients the concentration
of modier at the position z along the column as a function of time
and gradient slope is described by the following equation:
0
cM (z, t) = cM
+ g(t t0,M (z)),

t t0,M (z), 0 z L

(10)

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

0 in mM is the initial modier concentration at the
where cM
beginning of the gradient, and g the gradient slope in mM/min.
Combining Eqs. (2), (8) and (10) leads to the ODE describing the
evolution of the time needed by a retained solute (a protein in this
case) to cover a certain distance along the column:

dt

0
+ g(t t0,M (z))] },
= {1 + K,i + [cM
u
dz
t > t0,M (z), 0 z L

(11)

where K,i = p,i . The dimensionless form of Eq. (11) is derived

0 + g v (t/t (1 +
by dening the following variables: (, t) = (cM
0
K )))/g v ,  = z/L, g v = gt0 , where t0 = L/u, L is the length of the
column,  = K K,i = (p p,i ), and  = (g v ) . The previous
equation is simplied as follows:
d
=  +  
d

(12)

Eq. (12) can be solved analytically by series expansion. In the

next section the rst two terms of this series, corresponding to the
0th and 1st-order approximated analytical solutions, are derived.
Accordingly, the solution of Eq. (12) is assumed to be approximated
by a series expansion of the perturbation parameter   =  / . In
essence,   represents the difference between the intra-particle
porosity accessible to a tracer (or to the modier in this case), p ,
and the intra-particle porosity effectively accessible to the protein,
p,i .
 =  (0) +    (1) + O(  )
2

(13)


The 0th-order approximation (i.e.

= 0) represents the case
where no pore exclusion is present for the protein which therefore
access all available pores just like the modier. This corresponds to
the case addressed by Yamamoto [6,7].
2.2. Zeroth-order approximation
Since the pioneering work of Freiling in 1955 [14] many authors
worked on the derivation of the 0th-order approximation as
Yamamoto [6,7], Parente and Wetlaufer [15], and later Carta and
Jungbauer [5].
In the frame of the 0th-order approximation proposed by
Yamamoto, the difference between the volume fraction of stationary phase accessible to the modier and to the protein is neglected.
Therefore, p,i = p and  = 0. Eq. (12) simplies as follows:
d (0)
d

 ( (0) )

with V0 = AL = t0 Q, V0 being the elution volume of the a tracer

excluded from all the pores. VR is simply tR Q. It is worth noting that from Eq. (10), the concentration of modier at the outlet
of the column when the protein is eluting is easily computed by
R = c (L, t ) = g v (t (0) t
cM
M
R
0,M (L))/t0 and Yamamotos equation is
R
nally obtained:
gv =

R)
(cM

1+

(17)

(1 + )

The superscript R refers to tR , time at which the protein is eluting.

Eq. (16) is traditionally rearranged and linearized by logarithmic
transformation and the parameters and are estimated from the
slope and the intercept of the following curve:
R
log (cM
)=

1
1
log (g v ) +
log ((1 + ))
1+
1+

(18)

2.3. First-order approximation

When increasing the size of the protein of interest, easily going
up to several nanometers (as for monoclonal antibodies), the exclusion effect is no more negligible and the above mentioned 0th-order
approximation could lead to large discrepancies between experimental and simulated data.
Therefore, the derivation of a 1st-order analytical solution is
proposed to account for this exclusion effect. Perturbation theory for ordinary differential equations has been applied with the
perturbation parameter   =  / [16]. The previous assumption
K,i K is less closely to be satised the larger the value of   .
By introducing Eq. (13) in Eq. (12) a rst order expansion in  
is obtained as follows:

d (0)
d (1)
+ 
=  +  ( (0) ) (1 +   ( (1) / (0) ))
d
d

(19)

The superscript (1) stands for 1st-order approximation. Only

rst order terms in   are considered. Therefore, the Taylor expansion of the right hand side of Eq. (19) reduces to:
(1 +   ( (1) / (0) ))

= 1   ( (1) / (0) ) + O(  )

2

(20)

Combining Eqs. (19) and (20) leads to the nal equation to be

solved.

(1+) (1)
d (0)
d (1)

+ 
=  +  ( (0) )    ( (0) )
d
d

(14)

(21)

Recalling that  (0) is the solution of the unperturbed problem

where the superscript (0) stands for 0th-order approximation. Eq.

(14) can easily be integrated between the inlet and the dimensionless position  along the column:


 (0) (, t) =

(1 + )


  +

(0) 1+
(0 )

1/(1+)
(15)

1+

(0)

where  =   0 is the covered dimensionless distance and 0 =

 (0) (0 , 0). Eq. (15) can also be used in the case where piecewise
linear gradient are considered. In the present case, only simple lin(0)
ear gradients are considered so that  0 = 0 and 0 = 0. Eq. (15) is
evaluated at the outlet of the column ( = 1 and t = tR , where tR is the
retention time of the protein) considering that the gradient starts
0 = 0); the solution given by Eq. (15) simplies to the
at 0 mM (cM
following equation:

(0)
VR

35

= V0

(g v (1 + ))
1 + K +
gv

1/(1+)

(d (0) /d =  ( (0) ) ), and dividing by   , Eq. (21) simplies to a

rst order ODE describing the evolution of  (1) as a function of :
(1+) (1)
d (1)

=   ( (0) )
d

(22)

The ODE can be integrated analytically between the inlet ( = 0)

and the dimensionless position  along the column for any time
t t0,M (z) with the boundary condition  (1) (0, t) = 0. It comes:

(1)

(, t) =

1+
1 + 2

  +

(0) 1+

(0 )

1+


(0) 1+2
(0 )

(0) 1+

 +

(/1+)

(0 )

1+

(23)

(16)

Eq. (23) is evaluated at the end of the column ( = 1) with a linear

0 = 0 mM, and t = t . The solution simplies
gradient starting from cM
R

36

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

as follows:

3. Materials and methods


(1)

VR

= V0 1 + K +

(g v (1 + ))
gv

1/(1+)

3.1. Materials

1+
(K K,i )
1 + 2

(24)

It is worth noting that the 1st-order approximated solution

remains quite simple but provides a much better approximation
for and as discussed shortly in detail.
2.4. Data tting
Least square regression was applied to estimate optimal and
parameters that minimize the difference between simulated and
experimental data. The distance between the experimental and
simulated data was evaluated by computing the residual sum of
squares (RSS). The optimization problem consisted then in nding optimal (, ) parameters by minimizing the RSS dened as
follows:
D(, ) =

n


exp

exp

(VR,i VR,i (, )) = [VR

exp

VR ] [VR

VR ]

(25)

i=1
exp

In this equation, VR represents the vector of experimentally

measured elution volumes while VR (, ) are the corresponding
values calculated with parameters estimated by one of the following approaches:
- 0th-order approximated solution (Yamamotos equation (16)).
- 1st-order approximated solution (Eq. (24)).
- Inverse method (Eq. (12)).
It is worth noting that the values of and estimated by
integrating the full model (12) which corresponds to the inverse
method (IM) [17], are considered as the true values (IM , IM ). In
the following, we compare the and values obtained using the
various model approximations to see which one of them is able to
best approximate these parameters.
2.5. Condence interval
The quantication of the quality of the approximated parameters was performed by estimating their condence intervals. To do
so, the null hypothesis (, ) = (i , i ) was tested. (i , i ) correspond
to the optimal equilibrium parameters obtained with the different
approximated solutions. These optimal values were obtained from
(i = 0) Eq. (16) (thus corresponding to (0 , 0 ), in which 0 stands for
the 0th-order approximation), and (i = 1) Eq. (24) (thus corresponding to (1 , 1 ), in which 1 stands for the 1st-order approximation).
The tting was obtained by minimizing the RSS using Lagarias et al.
simplex algorithm implemented in Matlab .
For a total of n experiments, a F-test was applied to nd points
that reject the null hypothesis within a q% condence interval [18].
D(, ) D(i , i )
2
Fq,2,n2

n2
D(i , i )

(26)

Searching for all the (, ) that respect this inequality leads to

the following condence region, where q is the (1 q) quantile of
the F-distribution with 2 and n 2 degrees of freedom [18,19].
{(, )|D(, ) D(i , i )[1 + (2/(n 2))Fq,2,n2 ]}

(27)

Standard model proteins were considered: hen egg white

lysozyme (HEWL) and -lactalbumin form bovine milk (-Lac)
were obtained from SigmaAldrich and cytochrome C (Cyt-C) from
horse heart muscle was purchased from Acros Organics.
For the PEGylation of -Lac, methoxypolyethylene glycol succinimidyl propionic acid (mPEG-SPA) from JenKem was used, with
a nominal average molecular weight of 10 kDa.
Fractogel EMD SO3 (M) resin from Merck Millipore
(crosslinked polymetacrylate, sulfonic group, particle diameter 4090 m), SP SepharoseTM BB resin from GE Healthcare
(6% cross-linked agarose, sulphopropyl group, particle diameter 100300 m), POROS 50 HS from Applied Biosystems
(polystyrenedivinylbenzene, sulphonic group, particle diameter
80 m) and UNOsphereTM S resin from Bio-Rad Laboratories (sulfonic group, particle diameter 80 m) were used. The prepacked
columns were purchased from ATOLL GmbH. All the columns have
the same dimensions 50 mm 5 mm, for a total bed volume of 1 mL.
In addition, Q Sepharose High Performance resin from GE Healthcare (crosslinked agarose beads, quaternary strong anion exchange
groups, particle diameter 34 m) was used for the LGE experiments with the PEGylated -Lac. The stationary phase was packed
according to the manufacturers protocol in a Tricorn column (GE
Healthcare). The resulting bed length was 142 mm for an internal
diameter of 5 mm.
The mobile phase was either a 25 mM phosphate buffer (or a
25 mM Tris buffer when using the Q Sepharose High Performance
resin). The buffers were prepared with sodium dihydrogen phosphate and disodium hydrogen phosphate, (or TrisHCl and HCl) for
buffer A while buffer B was also containing 1 M sodium chloride
as modier. The buffers were adjusted to pH 7.0 and ltered on
Durapore membrane lters, type HVLP, 0.45 m (Millipore). The
water was deionized and further puried with a Millipore Simpack
2 purication pack.
3.2. PEGylation of -lactalbumin
1 mg of -Lac and 3-fold molar excess mPEG-SPA were dissolved in 2.5 mL of 25 mM phosphate buffer adjusted at pH 7.0.
The reaction took place at 25 C in a stirred beaker during 3 h.
The crude solution from the reaction was fractionated by SEC on
an analytical TricornTM Superdex 200 10/300 GL (GE Healthcare).
As mobile phase a 25 mM sodium phosphate buffer, containing
100 mM Na2 SO4 at pH 7.0 was used. The elution ow rate was set to
0.5 mL/min. Five fractions were considered, going from the non conjugated protein to the 4-times PEGylated -Lac (P0 to P4 ). The purity
of the fractions was checked by size exclusion chromatography.
3.3. Estimation of model parameters
Measurement of the proteins retention times were performed
on an Agilent 1100 Series equipped with a quaternary pump, a
degasser, an autosampler, a column oven and a diode-array detector. The peaks were detected at 280 nm for HEWL and -Lac and
415 nm for Cyt-C. 10 L injections from 1 mg/mL protein solution
were xed. The retention times were estimated by computing the
rst order moment of the chromatograms.
Pulse injections experiments were performed with six different
linear gradient slopes from 5 to 100 column volumes (CV) going
from 0% to 100% B. The gradient slopes g were chosen as 10, 20, 33,
50, 100 and 200 mM/min. At the end of the gradient, the columns
were re-equilibrated with buffer A for 5 CV. The volumetric ow
rate was 1 mL/min and the experiments were performed at 25 C.

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

37

Inverse size exclusion chromatography (iSEC) was applied and

tracers with a wide range of molecular sizes (salt, PEG of 0.2 kDa
and dextrans of 1, 5, 12, 50, 150, 270, 410, 670 and 2000 kDa) were
considered in order to measure the total accessible porosity prole
of the packing materials. This selection of tracers, with radii from
0.1 nm to 37.2 nm, enables to measure the complete porosity prole. The radii of the dextrans were computed with the following
0.498 , M being the molecular weight of the
equation: R = 0.845 Mw
w
dextran polymer in kDa [10]. The results of the iSEC are presented
in Fig. 1.
4. Results and discussion
4.1. Numerical comparison
(i)

The numerical relative error ((VR VR )/VR ) between the

approximated solution (i is either 0 or 1 depending if the 0th-order
or the 1st-order approximation is used) and numerical integration
of Eq. (11), simply noted VR , is shown in Fig. 2. As it has been
previously observed in Fig. 1, when increasing the size (hydrodynamic radius) of the solute, it is more and more excluded from the
porous particles. Indeed, when the size of the solute exceeded the
diameter of the largest pores, the solute can no longer enter the
porous particles. Therefore, it can be seen that when increasing the
size of the protein, meaning that the accessible volume fraction
goes from the one of the modier to the bed porosity, the relative
difference between the approximated solution and the numerical
integration of Eq. (11) increases. In this graph, only the accessible
particle porosity, p,i , was varied in order to cover the range of K,i
from K = 1 to 0. When the total accessible porosity of the solute is
equal to the total porosity (measured with a non excluded tracer as
salt in the present case), both approximated models are equivalent
to the full model dened by Eq. (11) and the numerical relative
error tends to 0. Decreasing the partition coefcient K,i leads to an
increase of the numerical error but it is always at least two orders
of magnitude lower for the 1st-order approximation than for the

Fig. 2. Relative difference between approximated solutions (0th and 1st-order

approximation) and numerical integration of Eq. (11) for different accessible porosities. Parameters for the calculation are: = 108 , = 3.50, = 0.40, t = 0.80, K = 1,
g = 50 mL/min and L = 5 cm.

0th-order solution and never exceeds 0.5% while the prediction

from the 0th-order approximated solution are approaching 10%
in the worst case. It should be noticed that this error is purely
numerical and does not include experimental variability.
4.2. Accuracy of the model predictions
All (, ) values estimated by the 3 different approaches, 2 different proteins (HEWL and Cyt-C) and the 4 cation exchange resins
are reported in Table 1.
They were all compared to the optimal (IM , IM ) obtained by the
IM. It is worth mentioning that the 1st-order approximation which
accounts for the exclusion effect in the case of protein with size
comparable to the average pore size, provides parameters that are
always in very good agreement with the true values. In the worst

Fig. 3. Relative error on parameter: (i IM )/IM (top) and parameter (i IM )/IM (down) for lysozyme (left) and cytochrome C (right).

38

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

Table 1
Comparison between equilibrium parameters obtained by the 3 different approaches.
FG EMD SO3

K,i
, p

Sepharose BB

POROS 50 HS

UNOsphere S

HEWL

Cyt-C

HEWL

Cyt-C

HEWL

Cyt-C

HEWL

0.50

0.43

0.65

0.46

1.26

1.00

1.20

0.49, 0.81

0.33, 0.82

0.39, 0.83

Cyt-C
0.95
0.39, 0.89

0th-order
equation (16)

( 109 )

13.65
3.67

4.68
3.68

90.64
4.16

4.34
3.73

89.90
4.02

7.92
3.80

308.7
4.14

5.96
3.75

1st-order
equation (24)

( 109 )

6.72
3.53

1.68
3.48

14.10
3.80

0.56
3.32

83.10
4.01

4.32
3.68

227.5
4.09

2.58
3.58

IM
equation (12)

( 109 )

5.33
3.58

1.14
3.41

13.19
3.79

0.62
3.33

71.61
3.98

4.34
3.67

224.2
4.08

2.72
3.58

case (Cyt-C on SP Sepharose BB), 10% difference is observed for the

parameter, and only 0.3% for , while for the same conditions the
estimates from the 0th-order approximated solution are more than
600% off for and 12% for . These results are summarized in Fig. 3.
When using parameters estimated by the tting of the 0thorder approximated solution, the worst predictions are obtained
with SP Sepharose BB resin. As expected, for this resin the difference between the total porosity accessible to the salt (0.88) and to
the lysozyme and cytochrome C (0.55 and 0.48, respectively) is the
largest among the investigated resins. In other words, the exclusion effect is more pronounced for this resin. For comparison, Fig. 4
presents the elution volume calculated from Eq. (11) using (0 , 0 )
and using (1 , 1 ). Even if the results obtained with (0 , 0 ) are
already in good agreement with the experimental data, it is worth
noting that the prediction are further improved when using the 1storder correction. These results are also due to the strong correlation
between and , which can compensate each other differences to
produce very similar Henry constant values.
To get a better feeling of the improvement obtained when using
(1 , 1 ) instead of (0 , 0 ) the relative errors with respect to the
exp
exp
experimental data (computed as (VR VR )/VR ) are shown in
Fig. 5. It can be seen that the error is systematically larger for the
0th-order approximated solution, while the parameters obtained
with the 1st-order approximation give errors that are similar to
the one obtained by the IM, which is in the range of the experimental error. Not surprisingly, the error obtained with the parameters
(0 , 0 ) is maximal for the steepest gradients. Indeed, for steep
R is larger than for shallow gradients, thus increasing
gradients cM

error is divided by about 2 when using the parameters (1 , 1 )

instead of (0 , 0 ).
4.3. Parameter reliability and condence intervals
The RSS were computed and shown in Fig. 6. Not surprisingly,
the precision of the calculated elution volumes is directly correlated
to the complexity of the model: 0th-order approximation < 1storder approximation < IM. RSS calculated using the parameters (0 ,
0 ) are increasing with increasing   values, while those calculated
from the 1st-order approximation remain more or less constants

R)
which amplify the error
the inuence of K compared to (cM
introduced by the assumption K,i K . On average, the relative

exp

Fig. 4. Elution volume calculated with (0 , 0 ) (dashed lines) and (1 , 1 ) (plain

lines). The experimental data corresponds to lysozyme ( ) and cytochrome C ( )
gradient elution experiments on UNOsphere S.

(i)

exp

Fig. 5. Relative error (VR VR )/VR between experimental and simulated data
calculated by numerical integration of Eq. (11) with (i , i ) obtained from Eq. (16)
( ), and Eq. (24) ( ). Data calculated with (IM , IM ) are also presented ( ). Experimental data correspond to lysozyme (top) and cytochrome C (down) elution on SP
Sepharose BB.

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

Fig. 6. RSS as a function of   for lysozyme (up) cytochrome C (down).   values

were evaluated with g = 200 mM/min.

and close to the RSS estimated with the true parameters. The
parameter   can be simply seen as the difference between the total
porosity and the accessible porosity for the protein. Moreover,   is
a function of the gradient slope, increasing with steeper gradients.
Therefore, the more the solute is excluded from the porous particles the bigger   and the steeper the gradient the bigger   . Large  
values implies large numerical error when using the approximated
solutions. Nevertheless, it seems that the 1st-order corrective term
is enough to obtain very good agreement between simulated and
experimental data without increasing the computational effort. In
all cases, the RSS estimated with (1 , 1 ) is about 0.1. We propose
this value as a threshold above which the 1st-order approximation
should be used. According to Fig. 6 this corresponds to a   values
greater than 104 .
To evaluate the reliability of the estimated values, the 95%
condence intervals were computed for the 0th-order and the 1storder approximated solution. The smaller the condence region,
the more reliable are the estimated parameters. Fig. 7 clearly shows
a narrower region in the (, ) plane for the 1st-order approximated
solution compared to the 0th-order which is an additional proof
that the 1st-order model is more reliable than the 0th-order one.
Remarkably, it can be noticed that (1 , 1 ) are really close from the
optimal (IM , IM ), much closer than (0 , 0 ).
As it can be seen in Fig. 7, the condence regions are very elongated, meaning that all couples (, ) that seems to fall the line
ln() = f() dened by the longer axis of the ellipsoidal condence
region are equivalently good (within the 5% condence interval)
in predicting the retention times. However, even though the (0 ,
0 ) values seem to compensate the exclusion effect so that the predicted retention times are still in very good agreement with the

39

Fig. 7. Optimal values and 95% condence interval obtained for the 0th-order
approximated solution ( and dash line) and 1st-order approximated solution (
and plain line). (IM , IM ) are also depicted ( ). The experimental values correspond
to lysozyme (top) and cytochrome C (down) elution experiments on SP Sepharose
BB.

experimental data, this does not mean that these parameters make
totally sense from a physical point of view. By using Eq. (24) instead
of Eq. (16), the isotherm parameters and are retrieving their
physical meaning.
4.4. Strongly excluded solutes
To emphasize even more this last observation, the same LGE
experiments were performed with PEGylated -Lac on a strong
anion exchange resin (Q Sepharose). Each PEGamer is known to
be constituted of several positional isomers, but under the investigated conditions single peaks were obtained for each PEGamer.
Therefore, they were considered as homogeneous pseudo-species.
The exclusion effect is expected to be more and more pronounced when increasing the degree of PEGylation of the protein
(number of PEG chains attached to the protein). This is clearly
shown in Fig. 8.
The values of (0 , 0 ), as well as (1 , 1 ), are shown in Fig. 9. It
can be seen that the parameters obtained from Eqs. (16) and (24),
respectively, present radically different trends. While the (0 , 0 )
are increasing with the number of conjugated PEG chains, the (1 ,
1 ) are decreasing. An increase of with the degree of PEGylation
would signicate an increase of the apparent charge of the protein interacting with the ligands. This would contradict previous
ndings. Indeed, Seely and Richey [20] observed that the retention time of PEGylated proteins was inversely proportional to the
degree of PEGylation in both anion and cation exchange chromatography. It is now commonly accepted in the literature that this is the

40

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

becomes signicant when the solute size approaches the pore size.
Indeed, while usually neglected, the exclusion effect can become
signicant when dealing with macromolecules. This has been
exemplied with different protein and PEGylated proteins. The new
approach is proposed as an extension of the classical Yamamotos
equation, for which exclusion effect are now accounted for. This has
been made possible by introducing in the classical equation (11) a
rst order correcting term, dened as the difference between total
and effectively accessible intra-particle porosity, and by solving the
resulting ODE with the perturbation method. The new approach
remains simple, as it does not require any additional parameter
compared to the classical Yamamotos equation. However, despite
its similarity with the classical approach, the proposed equation,
now accounting for the exclusion effect, has been proved for proteins and PEGylated proteins to deliver much better, reliable and
physically meaningful linear isotherm parameters.
Acknowledgments
Fig. 8. Total accessible porosity of the Q Sepharose High Performance column. The
dextran polymers are represented by ( ), while the -Lac and PEGylated -Lac are
depicted with ( ). The radius of the PEGylated proteins where estimated with the
equation proposed by Fee and van Alstine [22].

One of the authors (David Pster) is indebted to Prof. G. Carta

and Prof. A. Jungbauer for introducing him to the theory of gradient elution chromatography and for fruitful discussions about
Yamamotos method during the summer class at the BOKU University, Vienna. We are also grateful to Oliver Ingold for his meticulous
work with the -lactalbumin PEGylation. This work was nancially
supported by Sano, Vitry sur Seine, France.
References

Fig. 9. (open symbols) and (lled symbols) as a function of the degree of PEGylation from the unmodied -Lac to the 4-times PEGylated protein.

results of a charge-shielding effect of the neutral polymer chains.

In a recent work, while investigating the retention of PEGylated
lysozyme on strong cation exchange resin, Yamamoto et al. conrmed that the effective charge, , effectively decreases with the
degree of PEGylation [21].
While the (, ) estimated from both models were apparently equivalently good in estimating the retention time, the (1 ,
1 ) retrieve some physical meaning. As expected, they decrease
with the degree of PEGylation. For large macromolecules, strongly
excluded from the intra-particle volume, the difference in pore
accessibility between the modier and the proteins has to be taken
into account in order to derive physically reasonable and meaningful parameters.
5. Conclusion
In this study we showed how to use approximate solution of the
equilibrium model in conjunction with suitable LGE experiments to
estimate linear isotherm parameters in the form:

K = K + cM

A new procedure to interpret these results, based on Eq. (24) has

been developed in order to account for the exclusion effect which

[1] M.A. Desai, Downstream Processing of Proteins, Humana Press, Totowa, New
Jersey, 2000.
[2] P. Cutler, Protein Purication Protocols, Humana Press, Totowa,
New Jersey, 2004.
[3] L. Hagel, G. Jagschies, G. Sofer, Handbook of Process Chromatography: Development, Manufacturing, Validation and Economics, Academic Press, Amsterdam,
The Netherlands, 2007.
[4] T. Ishihara, T. Kadoya, S. Yamamoto, Application of a chromatography model
with linear gradient elution experimental data to the rapid scale-up in ionexchange process chromatography of proteins, J. Chromatogr. A 1162 (2007)
3440.
[5] G. Carta, A. Jungbauer, Protein Chromatography, Wiley-VCH Verlag GmbH &
Co. KGaA, Weinheim, 2010, pp. 277308.
[6] S. Yamamoto, K. Nakanishi, R. Matsuno, T. Kamikubo, Ion exchange chromatography of proteins-prediction of elution curves and operating conditions. I.
Theoretical considerations, Biotechnol. Bioeng. 25 (1983) 14651483.
[7] S. Yamamoto, K. Nakanishi, R. Matsuno, T. Kamijubo, Ion exchange chromatography of proteins-predictions of elution curves and operating conditions. II.
Experimental verication, Biotechnol. Bioeng. 25 (1983) 13731391.
[8] T. Ishihara, S. Yamamoto, Optimization of monoclonal antibody purication by ion-exchange chromatography: application of simple methods with
linear gradient elution experimental data, J. Chromatogr. A 1069 (2005)
99106.
[9] B.C. de Neuville, A. Tarafder, M. Morbidelli, Distributed pore model for biomolecule chromatography, J. Chromatogr. A 1298 (2013) 2634.
[10] P. DePhillips, A.M. Lenhoff, Pore size distributions of cation-exchange adsorbents determined by inverse size-exclusion chromatography, J. Chromatogr. A
883 (2000) 3954.
[11] S. Yamamoto, K. Nakanishi, R. Matsuno, Ion-Exchange Chromatography of Proteins, CRC Press, New York, 1988.
[12] C.A. Brooks, S.M. Cramer, Steric mass-action ion exchange: displacement proles and induced salt gradients, AIChE J. 38 (1992) 19691978.
[13] G. Guiochon, A. Felinger, D.G. Shirazi, A.M. Katti, Fundamentals of Preparative
and Nonlinear Chromatography, Academic Press, Amsterdam, The Netherlands,
2006.
[14] E.C. Freiling, Ion exchange as a separation method. 9. Gradient elution theory,
J. Am. Chem. Soc. 77 (1955) 20672071.
[15] E.S. Parente, D.B. Wetlaufer, Relationship between isocratic and gradient retention times in the high-performance ion-exchange chromatography of proteins:
theory and experiment, J. Chromatogr. 355 (1986) 2940.
[16] A.H. Nayfeh, Introduction to Perturbation Techniques, Wiley-VCH Verlag GmbH
& Co. KGaA, Weinheim, 1993.
[17] K. Kaczmarski, Estimation of adsorption isotherm parameters with inverse
method-possible problems, J. Chromatogr. A 1176 (2007) 5768.
[18] D.M. Bates, D.G. Watts, Nonlinear Regression Analysis and Its Applications, John
Wiley & Sons, Inc., New York, 2008.
[19] W.A. Duckworth, W.R. Stephenson, Beyond traditional statistical methods, Am.
Stat. 56 (2002) 230233.

D. Pster et al. / J. Chromatogr. A 1375 (2015) 3341

[20] J.E. Seely, C.W. Richey, Use of ion-exchange chromatography and hydrophobic
interaction chromatography in the preparation and recovery of polyethylene
glycol-linked proteins, J. Chromatogr. A 908 (2001) 235241.
[21] S. Yamamoto, S. Fujii, N. Yoshimoto, P. Akbarzadehlaleh, Effects of protein conformational changes on separation performance in electrostatic interaction

41

chromatography: unfolded proteins and PEGylated proteins, J. Biotechnol. 132

(2007) 196201.
[22] C.J. Fee, J.M. Van Alstine, Prediction of the viscosity radius and the size exclusion chromatography behavior of PEGylated proteins, Bioconj. Chem. 15 (2004)
13041313.