Journal of Chromatography A, 1362 (2014) 278293

Contents lists available at ScienceDirect

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

A scaling rule in supercritical uid chromatography. I. Theory for

isocratic systems
Abhijit Tarafder , Christopher Hudalla, Pamela Iraneta, Kenneth J. Fountain
Waters Corporation, 34 Maple Street, Milford, MA 01757, USA

a r t i c l e i n f o a b s t r a c t

Article history: Scaling is regularly done in chromatography either to transfer a successfully designed method of analysis
Received 25 February 2014 developed in one system to another system, or to scale-up a separation method developed in analytical
Received in revised form 11 July 2014 scale to preparative scale. For liquid chromatography there are well-tested guidelines for scaling, which
Accepted 3 August 2014
makes it a routine job. For supercritical uid chromatography (SFC), on the other hand, neither do we have
Available online 19 August 2014
any well-understood principles behind scaling nor do we know how far the strategies applied in LC could
be applicable to SFC. In this article, we have addressed these issues and proposed a rule applicable for
Keywords:
scaling isocratic methods between different SFC systems and column dimensions under commonly used
SFC
Method transfer operating temperatures and pressures. We have shown that the scale-up and method transfer techniques
Pressure drop used in LC can be applied to SFC, provided we ensure that both the original and the target systems in SFC
Density variation operate at the same average density. The current article will present the theory, discuss the extents of
Density modulation applicability of this rule, and outline its limitations. In an accompanying article implementation of this
Preparative rule in various practical situations will be presented.
Scaling 2014 Elsevier B.V. All rights reserved.
Scale-up

1. Introduction
Chromatography can provide a separation solution for almost
any mixture provided we detect the right combination of station-
ary and mobile phase chemistry. Determining the right conditions
(or method) for separating certain mixtures can be very challeng-
ing and may need considerable time, materials and expertise to
succeed. Naturally, we try to retain or reuse these methods even
when using a different system, for example, a system to run the
method faster, or need to change the scale of separation, for exam-
ple, analytical to preparative scale.
Such transfer of methods is routinely done in chromatogra-
phy with various motivations. For a transfer within the analytical
regime, more commonly called method-transfer, the common moti-
vations are to (1) transfer a slower method, previously performed
with larger columns and larger particle sizes, to faster analyses
employing smaller column dimensions with sub 2 m particle
sizes, (2) transfer methods between different laboratories or sites
having the same or different instrumentation to the original
method, etc. Method transfer from analytical to preparative sep-
arations is more commonly called scale-up. Scale-up is an integral
part of the design of preparative chromatographic processes. It is
Corresponding author. Tel.: +1 508 482 3328; fax: +1 508 482 3100.
E-mail address: abhijit tarafder@waters.com (A. Tarafder).
often too costly and time consuming to develop a suitable separa-
tion method directly in the preparative systems. In many situations,
for example, enantioseparation of the Active Pharmaceutical Ingre-
dients (APIs) during the drug discovery phase, the separation
specialists routinely do not have sufcient material to develop
chromatographic methods directly at the preparative scale. Very
commonly, a separation method is rst developed in the analyti-
cal scale and then scaled up for preparative separation. However,
in spite of the differences in purpose, method-transfer and scale-
up follow very similar basic principles. Which means any strategy
developed for method-transfer can be applied for scaling up as well,
and vice-versa. In this article we worked with these basic princi-
ples, applicable to both method-transfer and scale-up issues and
would mention any transfer of method as scaling.
The main application area of the scaling approach we dis-
cussed here is for Supercritical Fluid Chromatography or SFC. The
advantages of SFC as a separation technique has been discussed
in detail in various reviews [13]. In summary, because of the
application of CO2 as the principal mobile phase component, sep-
arations performed with SFC can be signicantly faster, cheaper
and environment friendly, hence more sustainable, compared to LC
separations. Undoubtedly, preparative separation, a.k.a prep sepa-
ration, with SFC is fast becoming the norm in many industries, both
for chiral and achiral separations [3]. Sample analysis through SFC
is also drawing strong interest because of signicantly faster sep-
aration and orthogonal elution behavior of the analytes, compared

http://dx.doi.org/10.1016/j.chroma.2014.08.009
0021-9673/ 2014 Elsevier B.V. All rights reserved.
 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293 279

to LC. Additionally, with the availability of better and more robust

instrumentation, there is an increasing interest of using smaller
particle size columns for even faster analysis [4]. A crucial compo-
nent for the success of all these endeavors is to have a technically
sound scaling strategy in SFC, which could be conveniently applied
in all practical situations.
In spite of this requirement we could not nd any suitable scal-
ing strategy reported in the literature specically for SFC method
development. In LC we have model-based optimization strategies
[58], mostly used in the academia. The model-based approach uses
mathematical models to simulate the physical behavior of the chro-
matographic system. This approach is described in more details in
Section 2.1. There are also technically sound rule-based approaches,
for example, the so-called L/dp and L/dp 2 rules (discussed in Sec-
tion 3), successfully used in scaling industrial separations. In SFC
the only reported literature on reliable scaling strategies are on
model-based approaches [2], which are efcient but not always
applicable in practical situations. There are no rule-based scaling
approaches or rigorous testing of the LC scaling rules showing their
general applicability to SFC systems. Do we need to understand and
develop separate scaling methods for SFC or we can apply LC rules
directly to SFC? The purpose of this article is to answer these ques-
tions and to propose a technically sound rule-based approach for
SFC scaling.
The report is organized in the following way. First we present a
literature survey of the available scaling strategies. This is followed
by a description of the physical mechanism behind the rule-based
approaches used in LC. In the succeeding section we discuss the
applicability of these rules in SFC in light of the differences between
the operations of these two techniques. We show that the main
difference between LC and SFC scaling originates from the inu-
ence of pressure drop along the column. Geometric scaling rules,
for example, the L/dp rule, lead to differences in column pressure
drops between the original and the scaled systems. Although this
difference in pressure drops does not affect chromatography in
LC, in SFC this leads to differences in solvent densities along the
columns, which affects chromatography. To mitigate this difference
in density between the original and the scaled column, we propose
a method which would enable the LC scaling rules applicable to
SFC as well. In the subsequent part of the report we discuss the
range of applicability of this method and its limitations. We demon-
strate that this approach can be applied for scaling SFC operations
between various systems, conditions and column congurations,
over a wide range of operating temperatures and pressures. Fig. 1. Flowchart showing the major steps in a simulation-based scale-up strategy.

2. Scaling strategies in chromatography

All the commonly used scaling strategies were developed for
and used in the scaling of liquid chromatographic operations. Later,
some of them were used in scaling SFC operations. The princi-
ples behind the LC scaling strategies are relatively well-understood
and documented. Based on a literature survey we could divide the
scaling approaches broadly into two categories: (a) model-based
approaches, which are mainly practiced within the academic circle,
and (b) experimental or rule-based approaches, which are practiced
both in the academic labs as well as in industry.
2.1. Model-based approach
The model-based approach takes a holistic view of the task and
tries to predict the global optimum performance of the separation
at hand. Based on mathematical models of the physico-chemical
behavior of a chromatographic system and state-of-the-art com-
putational techniques, simulation-based approach works with a
virtual experimental setup. The approach follows the steps shown
in Fig. 1. The rst step is to have a reliable mathematical model hav-
ing either an analytical or numerical solution. For smaller molecules
the Equilibrium Dispersive (ED) model was found to be sufcient
[5,6]. More detailed models, for example, Lump-Kinetic (LK) [9],
General Rate (GR) [7,8] and distributed-pore models [10] were
found to be more suitable for larger molecules, typically greater
than 10kDa. To simulate the results of a particular system, the mod-
els are written as computer simulation programs. The program then
needs to be characterized; by which the values of some char-
acteristic parameters of the model are estimated from the data
generated by a set of experiments [5,9] performed on a real system.
This step makes the simulation program ready as a virtual experi-
mental setup, with which one can rapidly perform experiments to
detect optimum operating conditions. Most often an optimization
routine is employed, which, through modulating the main oper-
ating variables, tries to detect the global optimum conditions of
the system. It can be noted that although the simulation-based
approach does rely on experimental data performed on analytical
systems, it does not limit itself in detecting conditions to repro-
duce the analytical performance in the prep system. With this
280 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293
Selecvity Staonary phase chemistry
Mobile phase properes
Resoluon
Retenon factor Staonary phase chemistry and
geometry
Mobile phase properes
Eciency Staonary phase geometry
Mobile phase properes, condion
Fig. 2. Deconvoluting different effects which control chromatographic behavior.
approach one can develop a new method, different from the ana-
lytical methods, focused on meeting the preparative separation
objectives.
In spite of its capability in detecting global optimum conditions
for separation and detecting methods more suitable for prep con-
ditions, the simulation-based approach found advocacy mainly in
the academic laboratories [59,2,11]. Application of this approach
in industrial cases has been minimal although having the potential
of generating the highest benets. For a more widespread appli-
cation of this technique it appears that we need to have more
user-friendly, easy-to-use programs to perform the simulation and
optimization studies.
2.2. Experimental or rule-based approaches
A diverse set of experimental or rule-based approaches have
been reported in the literature - from strategies purely based on
experiments to developing and applying empirical models for opti-
mization and scale-up. These approaches are undoubtedly the ones
used by most practitioners, either in the industry or in the academia
[1219].
For analytical method transfer, the set of rules compiled by Guil-
larme et al. [12] is the most comprehensive. The rules are (a) the
stationary phase chemistry in the target system should be identical
to the original system, (b) the ratio between the column length (L)
and the particle diameter (dp ) of the target system should be the
same as the original system (the L/dp rule), (c) the ratio between the
injection volume and the column void volume in both the systems
should be the same, and (d) the reduced linear velocity in both the
systems should be the same. We will discuss some of these rules in
detail in Section 3 because most of them will be useful to develop
a scaling rule in SFC.
Experimental scaling strategies for process scale-up, on the
other hand, are mostly experience-driven. The method optimiza-
tion (or method screening) is rst performed in the analytical
system and the resulting method is used as a guide to develop a
suitable separation condition for the prep system. Different strate-
gies of obtaining a scaled-up separation have been reported. In
the following two paragraphs we present an account of the strate-
gies used in LC and SFC, respectively, based on selected published
reports.
For LC scale-up, Perera et al. [15] conducted a method screening
in an analytical system and used the same column and the same
method for prep separation, only increasing the feed concentration
and the injection volumes. Gaggeri et al. [14] employed both exper-
imental and empirical approaches during scale-up for determining
the ow rates, run time, loading factor etc. Ishihara and Yamamoto
Diol, C18, Phenol etc.
Solubility
Relave permivity
Paral molar volume
Etc
Same as in Selecvity.
Pore geometry.
Same as in Selecvity.
Parcle size, column dimensions
Viscosity
Diusivity
Flow rate
[20] developed and reported empirical models, tuned with exper-
imental results, for optimization and scale-up of bioseparation
processes using gradient chromatography. In practice, three main
approaches for scale-up in LC could be detected: (1) using a geo-
metric scaling rule-based approach (e.g. L/dp ) where the purpose is
to replicate the optimized conditions achieved in the analytical sys-
tem, in the prep system, (2) using results from analytical screening
to choose between a set number of pre-programmed segmented
gradients applicable in the prep system, and (3) using the screening
conditions to develop a focused gradient to specically isolate the
compound of interest. Note that the last two approaches do not
follow the classical or standard scaling methodologies which are
based on geometric factors, but depend on additional empirical
steps to achieve more effective results.
For SFC scale-up, Pettinello et al. [16] conducted a separation
study with the goal of determining operating conditions suitable
to enhance purity and yield of a large-scale SFC process. The pro-
cess was rst developed on a bench-scale and then scaled up to a
pilot plant sized to semi-industrial production. The main approach
in this scaling was to start with the operating conditions deter-
mined at the bench-scale and then employ additional modications
to achieve conditions more suitable for pilot scale. White [21] inves-
tigated fast screening and purication of chiral mixtures through
SFC. For designing chiral purication processes from analytical
screening studies the author used a trial-and-error approach to
detect useful separation conditions. The same author reported [17]
an experimental strategy, based on mapping the retention times of
a ve component calibration mixture, for achiral separation. Gahm
et al. [18] emphasized the role of determining the mixture solu-
bility in the mobile phase during the design of a prep separation.
Hamman et.al. [19] reported an empirical approach to scale-up
from analytical screening. From experience the authors found that
a specic analyte retention time with an isocratic run is the key to
maximize the separation within a minimum separation time.
In summary, we noted from the reports we reviewed on SFC scal-
ing [16,21,18,19,2,11] that the practitioners either tried to follow
the LC scaling rules in SFC, which probably led to limited success
and requiring further method modication; or, there are elabo-
rate theoretical approaches being published from academia [2,11]
which are not implemented in practical situations. To ll in this
gap we developed a geometric scaling rule in SFC, an equivalent
or an adaptation of the rules used in LC, that is technically sound
and could be implemented in practical situations without the need
for any detailed simulation studies. In the next sections rst we
will present a discussion on the L/dp and L/dp 2 rules in LC and then
demonstrate why this approach may not be directly applicable to
SFC scaling.
 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293 281

Staonary phase Select the same surface chemistry of the column

Chose either L or dp of the target system

Column specicaons Calculate the other from (L/dp)T = (L/dp)O
Note: the values of the right hand side of the equaon are known.

Linear velocity (u) Calculate u from (u)T = (u x dp)O / (dp)T

Flow rate (Q) Calculate Q from (Q)T = (u)T x (A)T

Fig. 3. Design steps for scaling in liquid chromatography. L is the column length, dp is the particle diameter, u is the linear velocity and Q is the ow rate of the mobile phase,
and A is the column cross-section area. Subscript T indicates the target system and subscript O indicates the original system.

3. Geometric scaling rules in LC As the column efciency in terms of N can be written as N = L/H,
Eq. (2) can be further modied as:
The main focus during scaling in LC is to maintain the same
or similar column efciencies, with the implicit assumption that      
1 H dp Dm udp
the stationary phase chemistry and the mobile phase compositions = = A+B +C (3)
N L L udp Dm
will be kept unchanged in the target system. A rule which is widely
applied in this direction is the so-called L/dp rule, which requires
maintaining the same ratio of the column length (L) and the particle In Eq. (3), the terms A, B, C and Dm are basically functions of the
size (dp ) in the original system as well as in the target system. This intrinsic properties of the mobile and the stationary phases. During
rule works quite well and is well accepted in the industry and also scaling in liquid chromatography, if the intrinsic factors do not vary,
in many academic laboratories [12]. There is also a L/dp2 rule [22] the terms A, B, C and Dm remain unaltered in the target system.
which has been used during scale-up in bioseparation industries. The rest of the parameters in Eq. (3), that is, L (column length), dp
(particle diameter) and u (linear velocity), are the variables we can
manipulate to reproduce the same efciency.
3.1. The L/dp rule

The scaling rule of the L/dp states that to ensure the same ef-
ciency (in terms of number of plates or N) in the target system, the
ratio between the column length (L) and the particle diameter (dp )
should be the same as in the original system. For example, if the
column length and particle diameter in the original system is 50
mm and 1.7 m respectively, and the column length in the target
system is 150 mm, then to maintain the same N in the target sys-
tem the particle diameter should be 1.7 3  5 m. An additional
criterion to fulll the L/dp rule accurately is to maintain the same
(u dp ) product, where u is the linear velocity of the mobile phase.
To understand the theory behind this approach, we have to refer
to the van Deemter equation in terms of the reduced plate height
(RPH) and the reduced linear velocity, as rst proposed by Giddings
[23]:
B in Fig. 3. For LC this scheme works quite well. An example of the
h=A+ + C
 application of this scheme in LC scaling is shown in Fig. 4. In this
where h and  are the RPH and the reduced velocity, respectively
and A, B, C are coefcients of Eq. (1).
The versatility of Eq. (1) lies in its universality. It expresses RPH
only as a function of the reduced velocity of the system, irrespective
of the mobile phase properties and column characteristics. In other
words, to reproduce the same column efciency we have to ensure
that the numerical values of h and  are the same in both the original
and the target system.
To ensure that in practice, we re-write Eq. (1) with more familiar
terms by replacing h with H/dp and  with udp /Dm :
   
H Dm udp
=A+B +C
dp udp Dm
where H is the height equivalent to the theoretical plate or HETP
of the column, Dm is the molecular diffusivity and u is the linear
velocity of the mobile phase.
3.1.1. Maintaining L/dp and u dp
We can observe from Eq. (3), that irrespective of the way we
select the L, dp and u values, to achieve the same efciency we have
to maintain the same L/dp ratio and the same (u dp ) product in
the target system, as it was in the original system. This is the main
theoretical basis of the L/dp approach.
The above discussion also indicates that while designing the tar-
get system we cannot vary the L, dp and u independently. Rather, to
maintain the same efciency, we have to obey both the conditions
set by the system. For example, to maintain the rst condition, L/dp ,
we have to either choose the L or the dp . Selecting any one of them
will automatically dene the other too. Similarly, once we select
the dp based on the rst rule, we have to select the u based on that.
A decision owchart of the implementation of this rule is shown
(1)
example [24], an ACQUITY BEH C18 2.1 mm50 mm, 1.7 m was
scaled up to a XBridge C18 19 mm150 mm, 5 m prep column. The
resolution between peaks could be maintained by selecting analyt-
ical and prep columns with the same L/dp ratio and the same u dp
product.
3.2. The L/dp 2 rule
The L/dp 2 rule is used during the scale-up procedure in the
biopharmaceutical industries. The approach has been discussed in
detail by Rathore and Velayudhan [22]. In this approach, the A and
(2)
the B terms of the van Deemter equation are neglected on the
assumption that in the industrial separation, where the target is
to maximize the throughput, we would operate the system at such
high linear velocity (u) that the contribution of the C term will far
outweigh the contributions of the A and the B terms.
282 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293
Fig. 4. An example where the L/dp rule was successfully applied to scale-up an LC system [24].
Accordingly, the van Deemter equation is approximated in the
following way:
     
1 H dp Dm udp
= = A+B +C C
N L L udp Dm
From the Eq. (4), we can observe that if we can approximate the
van Deemter equation only with the C-term contribution, the basic
requirement of scale-up will be to maintaining the same L/dp 2 ratio
to ensure similar plate counts. The additional criteria is to keep the
linear velocity u the same. This adequately suits the needs of the
biopharmaceutical industries where the residence time of the ana-
lyte molecules in the column (controlled by u) can be an important
factor in molecular stability.
4. Application of LC geometric scaling rules in SFC
Before we discuss on the applicability of the LC geometric scaling
rules in SFC, we present a discussion on the fundamental mecha-
nism which controls scaling in any chromatographic system. The
points from this discussion will be used to analyze the applicability
of the LC scaling rules in SFC.
4.1. Role of density in scaling
4.1.1. Deconvoluting the factors affecting scaling
Any scaling mechanism is about changing the geometrical
structure and boundaries of the container vessel (the extrinsic envi-
ronment) keeping all the intrinsic physico-chemical interactions
the same. For chromatography, the extrinsic factors are controlled
by the column dimensions and the size of the stationary phase
particles. The intrinsic factors, on the other hand, are controlled
by the particle surface chemistry and the mobile phase proper-
ties. The mobile phase properties directly affect chromatography by
inuencing the analyte characteristics inside the chromatographic
system, for example, solubility, diffusivity, partial molar volume,
etc. The surface chemistry controls the analyte retention and also
the analyte movement in and out of the particle pores.
The intrinsic properties and the extrinsic factors together con-
trol the basic mechanism of chromatography. In the table shown in
Fig. 2 we have listed these factors, showing the parameters which
ultimately control them. For example the resolution (Rs ), which is
the prime indicator of the separation performance in any chromato-
graphic system, is controlled by both the intrinsic and extrinsic
factors. Rs is a function of the retention factor (k ), the selectivity
() and the column efciency (N). The selectivity () of the column
 2
 towards the analytes is controlled only by the physical proper-
udp ties of the mobile phase and the particle surface chemistry, which
(4)
LDm are intrinsic properties. The mechanism of solute retention (k ),
although mainly dened by the physico-chemical interactions, is
also controlled by the total accessible pore volume of the column,
which is an extrinsic factor. The mechanism of band broadening
(controlling N) strongly depends on the column geometry and also
on the properties of the mobile and the stationary phases. For prep
chromatography an additional factor comes into effect, the satu-
ration capacity (qs ) of the stationary phase, which is controlled by
the intrinsic properties. In summary, the analyte selectivity and
the saturation capacity are controlled by the intrinsic properties,
whereas the retention factor and the column efciency are con-
trolled by both the intrinsic properties and extrinsic factors. This
indicates that for a perfect scaling we need to reproduce the same
intrinsic and extrinsic conditions in the column so that the same Rs
and qs are reproduced. This is the prime condition for any perfect
scaling.
4.1.2. Roles of pressuretemperature versus densitytemperature
All the intrinsic properties related to the mobile and the sta-
tionary phases are state functions. That means if we maintain the
same pressure and temperature distribution in the original and the
target columns, assuming we are maintaining the same mobile
phase composition, we should be able to reproduce the same
physico-chemical interactions or the same intrinsic characteris-
tics in the target column. During scaling, to reproduce the intrinsic
conditions we select the same stationary phase chemistry, the
same mobile phase composition and the same temperature and
pressure. Although we use the pressuretemperature combina-
tion to set the state of the column, it is the densitytemperature
combination which fundamentally controls the physical behav-
ior. Pressuretemperature combination is used because of the
relative ease in measuring these two parameters. In several of
the recent publications [2527] we have seen that although
the pressuretemperature combination is easier to measure, it
is the densitytemperature combination which directly controls
the state functions of a system. In other words, to generate
the same intrinsic properties in the target system we have to
maintain the same densitytemperature conditions in both the
columns.
 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293
The geometric scaling rules, described in Section 3, ensures that
the extrinsic conditions in the original and the scaled systems
remain the same. However, recreation of the extrinsic conditions
should not interfere with the steps of recreating the intrinsic conditions,
because we need both the factors to be reproduced. While recreat-
ing the extrinsic conditions with the geometric scaling rules, in LC
we do not signicantly alter the mobile phase properties (barring
the UPLC conditions); so it does not affect the intrinsic conditions.
But in SFC, application of the L/dp rule inadvertently alters the intrin-
sic conditions. This is the main reason why the scaling rules of
LC are not directly applicable to SFC. In the following subsections
we demonstrate how the implementation of L/dp rule affects the
intrinsic conditions in SFC and what should be done to readjust it.
4.2. Effect of geometric scaling rule on column pressure drop
4.2.1. The L/dp rule
Although the approach of maintaining the same L/dp ratio and (u
dp ) product (discussed in Section 3.1.1) should lead to the same
plate count, and thus, the same resolution in the target system, it
will result in very different pressure drops across the columns. To
estimate the variations of the pressure drops we refer to the Darcys
law, which is applicable over a wide range of pressuretemperature
conditions.
According to Darcys law [28],
uL  u L
P = 2
= (5)
k0 dp k0 dp dp
In Eq. (5),  is the viscosity and is an intrinsic property of the
mobile phase which stays constant in LC but will vary along the
column in SFC. k0 is the specic permeability which is a column
characteristic and can be expressed as a function of the external
void fraction, e . k0 can be assumed to stay constant for either LC
or SFC operations. If we maintain the same L/dp ratio and the same
(u dp ) product to maintain the same plate number (ref Eq. (3)),
from Eq. (5) we can see that we cannot maintain the same pres-
sure drop. For example, if we change the particle diameter (dp ) by
n times, according to L/dp rule we have to change L by n times.
According to the same rule, we have to change u by 1/n times. Now
if we put this criteria in Eq. (3), we get the same N in both the
original and the target system. But if we put this in Eq. (5), we get
the following relation:
1 Fig. 5 shows the isopycnic plots of water and methanol.
P target = P original (6)
n2
Fig. 5. Isopycnic plots of water and methanol. The contour curves in both the plots are the constant density curves and the numbers shown on the curve are the density (in
g/mL) of the corresponding curves. In both the gures isopycnic plots are spaced with a 0.05 g/mL difference. In both the gures cartoons of two chromatographic columns,
representing the analytical and the prep runs, are drawn to represent the total pressure drop incurred for those operations. The purpose is to demonstrate that over the entire
range of pressure drop, normally incurred in an HPLC or UHPLC operation, the resulting density variations, between the analytical and the prep runs, are almost negligible.
283
For example, if we increase the particle diameter by three times,
we also have to increase the column length by three times and
decrease the linear velocity u by three times. These changes should
result into a nine fold decrease in the pressure drop across the col-
umn and a three fold increase in run time on the target system. Note
that these conclusions are valid with the assumption that in both
the original and the target systems the mobile phase properties and
the column permeability remained the same.
4.2.2. The L/dp 2 rule
The L/dp 2 rule should not lead to any differences in pressure
drops between the original and the target system. This is because
from the Darcys law (Eq. (5)) we can observe that if we maintain
the same L/dp 2 ratio and the same linear velocity u, which is the
main scaling criteria of the L/dp 2 rule, the pressure drop should
remain the same in both the systems. This means the L/dp 2 rule is
directly applicable in scaling an LC or an SFC system and does not
need any further adjustment for SFC operations. This rule, however,
is unsuitable for scaling analytical systems and more applicable
during scale-up.
5. Consequences of different pressure drops
This section discusses on the consequences of having different
pressure drops, between the target and the original system, during
LC and SFC scaling. Note that although here we focus on the dif-
ferent pressure drops originating from application of the L/dp rule
(see Section 4.2.1), the discussion is also relevant to any other con-
ditions which may lead to different pressure drops between two
systems, for example, systems having (1) different ow rates, (2)
different column dimensions, different particle sizes and different
column bed density.
The main difference between LC and SFC is the much higher
compressibility of the mobile phase in the latter technique. For
compressible uids, changes in mobile phase pressure along the
column can lead to considerable changes in its density, compared
to incompressible uids. To understand these differences more
clearly, in the following subsections we will compare the isopy-
cnic (constant density) plots of some common solvents used in LC
and SFC.
5.1. Different pressure drops in LC
The motivation of presenting these plots is to demonstrate the
284 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293

Fig. 6. Isopycnic plots of pure CO2 and (mol/mol,%) mixture of 90/10 CO2 /MeOH. The contour curves in both the plots are the constant density curves and the numbers shown
on the curve are the density (in g/mL) of the corresponding curves. In both the gures isopycnic plots are spaced with a 0.05 g/mL difference. In both the gures cartoons
of two chromatographic columns, representing the analytical and the prep runs, are drawn to represent the total pressure drop incurred for those operations. The main
motivation is to demonstrate that over the entire range of pressure drop, normally incurred in an SFC operation, the resulting density variation can be signicant, between
the analytical and the prep runs, even when liquid co-solvents are added.

possibility of density variation in an LC operation. Water, which is
widely used in LC, is nearly incompressible; whereas, methanol is
relatively compressible among the liquid organic co-solvents used
in LC. In both the plots, cartoons of two columns with two different
heights and widths are drawn. Note that the heights of these car-
toon columns are not representative of their actual length; rather,
they represent the net pressure drops across these columns. Sim-
ilarly, the cartoon columns were drawn with two different widths
just to provide a visual aid to differentiate between the original
(which is often an analytical column with smaller diameter) and the
target (which during scale-up is a prep column having wider diam-
eter) system. Note that these cartoon columns can also be viewed as
representing any conditions leading to different pressure drops in
two different systems, for example, (1) smaller particle size (hence
having higher pressure drop) versus larger particle size (lesser pres-
sure drop), (2) higher ow rate (higher pressure drop) versus lesser
ow rate (lesser pressure drop), (3) longer column (higher pres-
sure drop) versus shorter column (lesser pressure drop), etc. The
operating temperature of both the columns is represented by the
dashed line passing through their respective axes. On both the
columns, in each gure, the average density of operations corre-
sponding to the original and to the target systems are drawn with
closed circles. The contour plots of isopycnic conditions were cre-
ated based on data available from NIST uid database REFPROP 9.1
[29].
From Fig. 5 we can observe that for liquid solvents for exam-
ple, water and methanol, the differences in the pressure drop
between the original and the target systems resulted into very
similar average operating density. Note (Fig. 5) that this situation
doesnt change even when the temperature of operation is varied
over a wide range. Although the compressibility data of other liq-
uid organic solvents are not currently available over such a wide
range, we can safely assume that they will behave in a similar
way.
This observation is important in the current context. It demon-
strates that in LC, even if applying the L/dp rule leads to signicantly
different pressure drops, there will not be any changes in the
chromatography. Because the mobile phases in LC are fairly
incompressible, any differences in net pressure drops, between
the original and the target system, should not lead to any dif-
ferences in the average operating density. Which means, if we
keep the temperature of operation and co-solvent percent the
same, we should be able to replicate the same mobile phase
properties and hence the same chromatography in both the
systems.
5.2. Different pressure drops in SFC
In SFC the situation is different because of the differences in
the operating density caused by different pressure drops. Fig. 6
shows the isopycnic plots of pure CO2 and 90/10 (mol/mol,%) of
CO2 /MeOH mixture, along with cartoon columns representing ana-
lytical and preparative systems. The assumptions and the features
of these columns are the same as that discussed in the last section
(Section 5.1). We can note from the example shown here that the
average operating density of the preparative system is very differ-
ent from the average operating density of the analytical system.
Note that the extent of this variation increases with lower percent
of co-solvents in CO2 and also with higher operating temperatures.
Although Fig. 6 shows an example of the effect of pressure dif-
ferences during scale-up, it also represents any situation in SFC
where the original and the target systems have different pressure
drops along the column. A detailed discussion on these effects is
presented in Section 7.
6. Mitigating density variation during SFC scaling
Based on the discussion presented in Section 4.1, we can con-
clude that for very accurate reproduction of performance during
scaling, we have to reproduce the same density variation prole
in the target system, as it was in the original system. This can be a
very difcult, if not impossible task especially because the nature of
density variation inside a system can be very different, depending
on the operating temperature and the inlet and outlet pressures.
As an easier alternative, we are proposing a simple approach of
density modulation to reproduce the average mobile phase densities
between the two systems, which can be conveniently ensured in
a commercially available system and should work reasonably well
within acceptable approximation.
6.1. Steps to match average densities
We dene the operational steps which should lead to matching
the average densities of the original and the target systems. The
approach taken here manipulates the ABPR (Automated Back Pres-
sure Regulator) pressure setting of the target system to match the
 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293 285

Fig. 7. Flowchart showing the density modulation scheme described in Section 6.1. Note, the gray boxes at the left hand side represent the steps to determine the average
density inside the original system. The subscript O has been used with all the parameters related to the original system. The white boxes represent the iterative approach
taken to achieve the same or very similar average density in the target system, where the subscript T has been used.

average density of the original system. A owchart showing the the target system, the next ABPR pressure setting can be estimated
steps described here are also presented in Fig. 7. following this formula:
(PABPR,T )i+1 = (PABPR,T )i
1. Note the inlet pressure (a.k.a. System pressure) and the outlet    
Psys,O + PABPR,O (Psys,T )i + (PABPR,T )i
pressure (a.k.a. ABPR pressure) of the original system from the + (7)
operating panel of the chromatography data software. Also note 2 2
the column oven temperature from the same source.
where (PABPR,T )i+1 and (PABPR,T )i are the ABPR pressures for the
2. Assuming that the measured System pressure approximates the
(i + 1)th and ith iteration; (PSys,T )i is the System pressure for the ith
pressure at the column inlet and the measured ABPR pressure
iteration, all representing the target system; and, Psys,O and PABPR,O
approximates the pressure at the column outlet; also assuming
are the respective system and ABPR pressures of the original sys-
that the pressure inside the system varies linearly, determine
tem.
the pressures at intermediate points inside the original column
through linear interpolation.
6.2. The main assumptions
3. Assume that temperature inside the column along the axial
direction is the same as the oven temperature.
The approach of manipulating the ABPR pressure to match the
4. From the interpolated pressure and the noted temperature, esti-
average density of both the columns should work under the follow-
mate the density prole inside the column using a suitable
ing assumptions:
equation of state (EOS).
5. Applying a weighted average method over the column length,
1. The variation of pressure prole across the column is linear.
determine the effective average density of the original column.
2. The retention factor of the analytes are linear functions of the
6. Run the target column with the same ABPR setting as the previ-
mobile phase density.
ous column.
3. All the other factors which control scaling are also linear func-
7. Follow steps 1 to 5 to determine the effective average density of
tions of the mobile phase density at a certain temperature.
the target column.
4. The temperature variations along the column is not signicant
8. If the average density in the target column is lower than the
enough to affect chromatography.
original then the ABPR pressure of the target system should be
increased. If the average density in the target system is more
These factors are crucial in the applicability of this rule. In the fol-
than the original, the ABPR pressure should be decreased.
lowing subsections the viability of these assumptions are discussed
in detail.
Using density modulation to match the average density in both
systems may need multiple iterations in the absence of any spe- 6.2.1. Linearity of pressure drop
cic data about the variations of pressure and density in the target The assumption that inside the column and in the entire system
column. However, an approach based on matching the average the pressure will vary linearly is a key assumption which enables
pressures of both systems may be useful in most of the situations. us to determine the density prole inside the SFC system with rel-
The target system can be started with the same ABPR pressure as in ative ease. In Section 6.1 we have described how, from the inlet
the original system. Then, based on the resultant System pressure of and the outlet pressure data of a particular run, we can estimate
286 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293

the pressures at any point along the column through linear inter-
polation, from which we can calculate the density prole directly.
The basic assumption behind this relatively simple approach was
that during an SFC run, pressure varies linearly across the system.
This assumption saved us from the task of calculating the pressure
prole through solving a differential equation with no analytical
solution. In this section, we will discuss the viability of this assump-
tion and also its limitations.
To develop a general understanding on the pressure drop vari-
ations in SFC we can explore Darcys law. Reports from previous
publications [30,31] demonstrate that pressure drop estimations
based on the Darcys equation are sufciently accurate. The gen-
eral Darcy equation was written for pressure drop estimations of
incompressible uids. For SFC, however, the equation should be
expressed in the differential form [32,30]. The standard Darcys
equation considers the linear velocity (u) of the column to be
directly proportional to the pressure drop (Eq. (5)). Although this
relation still holds in SFC, the linear velocity should not be consid- Fig. 8. Variation of the kinematic viscosity and density of 100% CO2 as a function of
ered as constant inside the column. Because of the change in mobile pressure and temperature. Bold curves represent contour plots of kinematic viscos-
ity of unit centistokes. Dashed curves represent iso-pycnic plots of 100% CO2 , unit
phase density across the column and the constant mass ow rate
g/mL.
[30], the linear velocity should also change. Keeping this in mind,
for SFC we have to write the Darcys equation as the following:
under high pressure drop. With increasing percent of co-solvent
dP  G
= (8) (methanol) in the system (Figs. 9 and 10), the constant  curves
k0 dp Ae 
dL 2
become increasingly aligned with the constant-temperature lines.
where G is the mass ow rate of the mobile phase, A is the column This shows that the  value should vary to a lesser degree for the
cross-sectional area, e is the external void fraction and  is the same pressure drop at constant temperature operations. From the
density of the mobile phase. The above equation can be rearranged isopycnic plots of Figs. 9 and 10 we see that the regions where 
and written as: varies the most with pressure drop, that is, at lower temperatures,
the density variation is low. Which means even in the zones where
dP G 
= 2 
(9) the pressure prole along the column is not linear, we can assume
dL Ae k0 dp linearity because that should not signicantly affect the estimation
where, except the / part, the rest of the right hand side can be of the average density value. The most signicant variation of both
assumed to be constant during any time of a particular operation.  and density occurs at the low pressure high temperature zones,
The above equation can be simplied as: where the assumption of linear pressure may not work.
In summary, Figs. 810 demonstrate that the assumption of lin-
dP ear pressure variation inside the column to calculate the average
= C (10)
dL density should work over a wide range of pressuretemperature
where C is a constant representing C = G/(Ae k0 dp 2 ) and  is the condition, except at temperatures higher than 100 C with
kinematic viscosity of the mobile phase. pressures lower than 2000 psi.
The assumption of linear variation of pressure across the column
should be viewed from two angles: (1) the pressure variation is 6.2.2. Linearity of retention factor variations
truly linear or approximately linear, and (2) the pressure variation Another implicit assumption, on which the applicability of this
is not linear but an assumption of linearity is acceptable because approach depends, is that by matching the average densities of
it would not signicantly affect the estimation of average mobile
phase density in the column, which is the main objective of this
exercise.
Based on Eq. (10) we can conclude that the assumption that pres-
sure varies linearly across the column is valid only when  remains
constant. If  remains constant, the entire right hand side becomes
constant, which should lead to a linear variation of pressure across
the column. In other words, to understand how far our assumption
of linear pressure variation across the column is true, we have to
explore how far an assumption that  remains constant across the
column is true for an operation.
To understand how  varies as a function of pressure, tempera-
ture and mobile phase composition, we are presenting constant 
plots of 100% CO2 and (90/10,%) and (80/20,%) of CO2 /MeOH mix-
ture in Figs. 810 respectively. The gures also present isopycnic
plots of the respective mobile phase compositions. The data pre-
sented in these gures were obtained from the REFPROP software
of NIST [29]. From Fig. 8 we can note that the value of  remains
invariable or barely changing over a wide pressuretemperature
Fig. 9. Variation of the kinematic viscosity (bold curves) of 90% CO2 + 10% methanol
zone, especially above 60 C and pressures above 2000 psi. Below
as a function of pressure and temperature. The numbers on the contour plots repre-
60 C  varies roughly 69% over a pressure drop of 1500 psi, which sent the value of the kinematic viscosity of that zone in centistokes. Dashed curves
means the pressure prole along the column can be non-linear represent iso-pycnic plots, unit g/mL.
 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293 287

When we match the average density of the original and the

target system, we ensure that the following relation is true:
i=n1
i=1
((i+1 + i )/2)T (xi+1 xi ) i=n1
i=1
((i+1 + i )/2)O (xi+1 xi )
= (13)
L L
where the sufxes T and O represent the target and the original
systems, respectively.
At a certain temperature, the retention factor k can be written
as a function of , or k = f(). So, if we were able to measure k at
each section of the column, we could have calculated the average
k from the following relationship, just like the way we calculated
the average :

i=n1 (k i+1 + k i )/(2)(xi+1 xi )

k ave = i=1
(14)
L
Similar to the previous situation of matching the average densi-
ties of two systems, to match the average k of the original and the
target systems, we should have the following relation valid:
Fig. 10. Variation of the kinematic viscosity (bold curves) of 80% CO2 + 20% methanol
as a function of pressure and temperature. The numbers on the contour plots repre- i=n1
i=1
((k i+1 + k i )/2)T (xi+1 xi ) i=n1
i=1
((k i+1 + k i )/2)O (xi+1 xi )
sent the value of the kinematic viscosity of that zone in centistokes. Dashed curves = (15)
L L
represent iso-pycnic plots, unit g/mL.
If we assume that k is a linear function of , e.g. k = A + B, we
can write the average k of the target system as a function of density
the original and the target columns we are matching their average by replacing k i+1 with Ai+1 + B and k i with Ai + B:
retention factors. As a rst step in this direction, let us see how we    
i=n1 ((i+1 + i )/2)T (xi+1 xi ) i=n1 (xi+1 xi )
are determining the average density of the two systems and how (k ave )T = A i=1
+B i=1
(16)
L L
we are matching it. The average density of a system can be deter-
mined from the following integral (Eq. (11)) which is basically a From Eq. (13) we can replace the average density of the target
weighted average of the local densities over the column length: system with that of the original system to write as:
L  i=n1  i=n1
dx ((i+1 + i )/2)T (xi+1 xi ) (xi+1 xi )
0 i=1 i=1
ave = (11) A +B
L L L
 i=n1  i=n1 (17)
where ave is the average density of the mobile phase, x is the vari- ((i+1 + i )/2)O (xi+1 xi ) (xi+1 xi )
i=1 i=1
able representing column dimension,  is the density as a function A +B
L L
of x and L is the column length.
This integral can be numerically solved following a simple rect- Which means,
angle rule. Using the column position (x) vs density () data we can
write: (k ave )T = (k ave )O

(18)

i=n1 (i+1 + i )/(2)(xi+1 xi )
i=1
ave =
L
where sufx i represents the ith row of the x vs  data and n is
the total number of rows. Eq. (12) can be easily implemented on
an excel spreadsheet to calculate the average density. A sample of
calculating average density is presented in Appendix A.
From the above steps we can conclude that if is a linear func- k
(12) tion of density in both the systems, matching the average density
will always lead to matching the average k .
Presenting a general view on the range of validity of the assump-
tion that k is a linear function of  is difcult because we do not
have any general equation which relates the k with density. Data
published in the literature [32] shows that the Henrys constant

Fig. 11. Left: Henrys constant versus density data published by Rajendran et al. [32], Right: Shadowed areas showing examples of the extent of piece-wise linearity which
can be approximated from the power function.
288 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293
(H), which is a function of k at constant total void volume in a
column (t ), can be expressed as a power function of density. The
power function, over limited density ranges can actually be approx-
imated by linear functions following Taylor series expansion. A
visual demonstration of such approximations is shown in Fig. 11.
Note that although the H varied non-linearly over the entire density
range, we can linearize this relation over limited ranges of density
variation. In other words, if the density variation across the column
is not signicant, we can assume that k is a linear function of den-
sity. Note that this condition does not mean that the variation of
density across the column should be linear. Even if the density vari-
ation across the column is non-linear, as long as k variation with
density can be approximated with a linear variation, the average
density matching should lead to average k matching. The more the
k vs  relation deviates from linearity, the more this approach will fail
to maintain the original retention factors in the target system.
In conclusion, the density matching method proposed here
should be more applicable under conditions where the pressure
drop along the column does not lead to signicant density vari-
ation, for example, at high-temperature-low-pressure conditions.
This method should be more applicable at lower temperatures,
higher pressures and with higher co-solvent concentration.
6.2.3. Insignicant variation of temperature
An implicit assumption which is made in the above two assump-
tions is that the temperature during the chromatographic operation
will remain the same. Although some recent publications [33]
have shown that in SFC we can expect considerable temperature
drop due to pressure drop across the column, recently [34] it has
been shown that high temperature drop is limited only to highly
compressible ares on the pressuretemperature plane. Over the
pressuretemperature ranges where most of the SFC operations
are carried out, the temperature drop is not signicant, especially
when liquid co-solvents are added to the mobile phase.
7. Verication of density modulation approach
In this section, the theory described above on regaining the
intrinsic properties by matching the average densities of the orig-
inal and the target systems will be experimentally veried with
some example problems.
(1) Caffeine, (2) Carbamazepine, (3) Uracil, (4) Hydrocortisone, (5) Prednisolone, and (6) Sulfanilamide
0.30 1
3 6
0.25
2
0.20
4
0.15
AU
0.10
0.05
0.00
0.00 0.50 1.00 1.50
0.30
0.25
2,3
0.20
AU
1 Lower Density
0.15
0.10
0.05
0.00
0.00 0.50 1.00 1.50
Fig. 12. Increasing the particle size, keeping all the other experimental conditions the same, signicantly affected the retention factors, the selectivity and the column
efciency in SFC. Note, a change in particle size would have affected only the column efciency in LC.
7.1. Materials and methods
In all the examples either an 1.7 m ACQUITY UPC2 or 5.0 m
Viridis SFC column was used. All the data corresponding to
the analytical experiments were collected on an ACQUITY UPC2
System, and data corresponding to the prep experiments were col-
lected on a Waters Prep100q SFC System. For all the experiments
a standard sample mix was prepared with (1) caffeine, (2) carba-
mazepine, (3) uracil, (4) hydrocortisone, (5) prednisolone and (6)
sulfanilamide. The concentration of each of these analytes in the
mixture was 0.2 mg/mL where methanol was used as the sample
solvent. The numbers corresponding to the analytes are used to
label all the chromatograms presented here.
7.2. Effect of different particle sizes
Pressure drop along the column depends inversely to the square
of the particle diameter (Eq. (5)). So any change in particle diame-
ter can signicantly alter the average mobile phase density in SFC.
The differences in the operating densities resulting from the dif-
ferent pressure drops across the original and the target systems
can lead to very different chromatography. This is because differ-
ent average densities will result in different intrinsic properties of
the mobile phase, which may lead to different retention factors and
selectivities (Section 4.1).
7.2.1. Without matching average densities
This set of experiments was conducted with two columns hav-
ing the same dimensions (2.1 mm 150 mm) and chemistry (BEH
2-EP) but different particle sizes (1.7 m and 5.0 m). The other
experimental conditions, which were kept the same for both the
columns were: ow rate = 1.4 mL/min; Temperature = 40 C; ABPR
pressure = 1500 psi. Because of the differences in the particle sizes,
the columns experienced different pressure drops under the same
operational conditions, which resulted into the System pressure
with the 5.0 m column being 2004 psi and with the 1.7 m column
5888 psi.
Fig. 12 shows the separation of the standard mix in these two
columns under isocratic conditions (10% MeOH). It can be noted
from Fig. 12 that although the column conguration was changed
only by changing the stationary phase particle size, the result-
ing chromatography is signicantly different, showing changes in
Higher Density 2.1 x 150 mm Column
5 Isocratic: 10% MeOH
Mobile Phase
1.4 mL/min
ABPR 1500 psi
ABPR=
1.7 m
2.00 2.50 3.00 3.50 4.00 4.50 5.00
2.1 x 150 mm Column
Isocratic: 10% MeOH
1.4 mL/min
/
Mobile Phase
6 ABPR = 1500 psi
4
5
5 m
2.00 2.50 3.00 3.50 4.00 4.50 5.00
Minutes
 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293 289

Fig. 13. (a) Showing the estimated density proles along the 1.7 m and 5.0 m columns described in Section 7.2.1. A detailed description of the calculation steps is provided
in Section 7.2.2. Note that when using the same ABPR pressure (1500 psi) the effective average density of the 1.7 m column was 0.88 g/mL whereas the same for the 5.0 m
column was 0.8 g/mL. (b) Showing the estimated density proles along the 1.7 m and 5.0 m columns described in Section 7.2.2. Note that by increasing the ABPR pressure
of the 5.0 m column the average densities of both the systems were matched.

retention factors and selectivities of the analytes. Note how carba-
mazepine (2) and uracil (3) almost co-eluted in the 5.0 m column,
whereas they were well resolved and had good resolution in the
1.7 m column. In an analogous LC experiment, the retention and
selectivity should have remained the same, although the efciency
should have changed because of the differences in particle sizes.
proles after iteratively matching the average densities is shown
in Fig. 13b. Keeping all the other conditions the same, as described
in Section 7.2.1, only by changing the ABPR pressure of the 5.0 m
particle size column from 1500 to 3390 psi, we could approx-
imately regain the retention factors and the selectivity of the
analytes, which can be noted from Fig. 14.

7.2.2. Matching average densities 7.3. Effect of different ow rates

The main reason behind the signicant alteration of the reten-
tion factors and the selectivities can be attributed to the change in
the average operating densities of the two systems. Fig. 13a shows
the estimated density proles along the columns with 1.7 m and
5.0 m particle sizes respectively, both having 1500 psi as the ABPR
pressure. Note that the average operating density of the 1.7 m
column is 0.08 g/mL higher than the 5.0 m column, which is cer-
tainly not negligible. Calculation steps of these density proles are
provided in Appendix A.
To match the average density we applied the iterative approach,
described in Section 6.1. Note that the iterative approach was
directed at determining a new ABPR pressure for the 5.0 m
column so that the average densities of both the systems were
matched. The resulting ABPR pressure (obtained through iteration)
of the 5.0 m particle size column was 3390 psi which resulted in
an average density more closely matched to the average density
of the 1.7 m column (Tables 4 and 3). The resultant density
The rule of matching the average density can be applied more
broadly to deal with any system or method alteration which can
have a direct impact on the density prole of the system. For
example, changing the mobile phase ow rate results in different
pressure drops across the column and hence different density pro-
les. When a set of experiments with a 2.1 mm 150 mm, 5 m
BEH-2EP column was conducted with ow rates (a) 0.48 mL/min,
(b) 1.4 mL/min, and (c) 4.0 mL/min, very different retention fac-
tors and analyte selectivities were obtained (results not shown
here). After applying the density modulation methods we could
recover these factors back, which are demonstrated in the exper-
imental results shown in Fig. 15. For the faster ow rates (cases
(b) and (c)), the ABPR pressures were adjusted to yield the same
average density across the column as in the 0.48 mL/min experi-
ment. For 0.48 mL/min ow rate the ABPR pressure was 3600 psi
and System pressure 3776 psi. To achieve the same average density,

(1) Caffeine, (2) Carbamazepine, (3) Uracil, (4) Hydrocortisone, (5) Prednisolone, and (6) Sulfanilamide

0.30 1
3 6
0.25 2
2.1
2 1 x 150 mm Column
0.20
Isocratic: 10% MeOH
4 1.4 mL/min
AU

0.15 5 ABPR = 1500 psi

0.10
NUSP(Sulfan) = 19,809
0.05
1.7 m
0.00
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00
0.30
2.1 x 150 mm Column
0.25
1 3 6
Isocratic: 10% MeOH
0.20 2 1.4 mL/min
ABPR = 3390 psi
AU

0.15
4
0.10 5 NUSP(Sulfan) = 6,561

0.05
5 m
0.00
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00

Minutes

Fig. 14. Implementing the density modulation approach in the 5 m column resulted in retention factors and selectivity very similar to that obtained in the 1.7 m column.
Note that the decrease in efciency (from plate number N = 19,809 to 6561) is commensurate with the standard relationship between N and dp as observed in LC.
290 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293

Density Profiles 0.30

0.48 mL/min
3
with Matched Density Averages 1
2
6 NUSP(Sulfanilamide)
0.95 0.20
(for 2 1 x 150 mm,
2.1 mm 5 m column) 9 883
9,883

AU
4
5
4.0 mL/min 0.10

Average Density 0.00

0.89 g/mL Average
g Densityy 0.00 2.00 4.00 6.00 8.00 10.00
1.4 mL/min 0.89 g/mL Minutes
nsity (g/mL)

0.30
0.90

0.48 mL/min 0.20 1.4 mL/min

NUSP( Sulfanilamide )

AU
6,561
0.10
bile Phase Den

Average Density
0.89 g/mL
0.00
0.85 0.00 1.00 2.00 3.00 4.00
0.30 Minutes
Mob

0.20 4.0 mL/min

AU
0.10 NUSP(Sulfanilamide)
3,203
0.80
0 25 50 75 100 125 150 0 00
0.00
0.00 0.20 0.40 0.60 0.80 1.00 1.20
Column Position (mm)
Minutes

Fig. 15. The left hand side of the gure shows estimated density proles across the column at three different ow rates. The ABPR pressures of the 1.4 and 4.0 mL/min ow
rate experiments were adjusted in a way so that the average densities match with the average density of the experiment with 0.48 mL/min ow rate, which was 0.89 g/mL.
The right hand side of the gure shows the chromatograms with three different ow rates but matched average densities. Note, although the number of plates reduced with
increasing ow rates the selectivity of the analytes remained similar.

Table 1 Fig. 15 shows the column efciency values (as dened by the
Matching average density to mitigate effect of different ow rates.
United States Pharmacopeia) for the sulfanilamide peaks on the
Flow rate System pressure ABPR pressure Average pressure chromatograms of all the ow rates. The values are different
(mL/min) (psi) (psi) (psi) because at ow rates faster than the optimum linear velocity, chro-
0.48 3776 3600 3688 matographic efciency decreases, similar to what is observed for LC
1.40 3996 3390 3693 applications.
4.00 5066 2322 3694

7.4. Effect of different column dimensions scale-up to prep

the ABPR pressure of the 1.4 mL/min experiment had to be set at
3390 psi, and for the 4.0 mL/min experiment 2322 psi. The System
pressures of these two experiments were 3996 psi and 5066 psi,
respectively. Data corresponding to these adjustments are sum-
marized in Table 1.
systems
This example demonstrates the density matching approach dur-
ing the scale-up of applications to preparative chromatography.
Fig. 16a demonstrates the initial separation method, developed on

Table 2 Table 3
Variation of pressure and density across the 5 m column with 1500 psi ABPR Variation of pressure and density across the 1.7 m column with 1500 psi ABPR
pressure. pressure.

x Temperature Estimated Estimated Sectional x Temperature Estimated Estimated Sectional

pressure density average density pressure density average density
(mm) ( C) (psi) (g/mL) (g/mL) (mm) ( C) (psi) (g/mL) (g/mL)

0 40 2004.0 0.818 0.817 0 40 5888.0 0.946 0.943

10 40 1970.4 0.816 0.815 10 40 5595.5 0.940 0.936
20 40 1936.8 0.814 0.813 20 40 5302.9 0.933 0.930
30 40 1903.2 0.811 0.810 30 40 5010.4 0.926 0.923
40 40 1869.6 0.809 0.808 40 40 4717.9 0.919 0.916
50 40 1836.0 0.807 0.806 50 40 4425.3 0.912 0.908
60 40 1802.4 0.805 0.803 60 40 4132.8 0.904 0.899
70 40 1768.8 0.802 0.801 70 40 3840.3 0.895 0.890
80 40 1735.2 0.800 0.798 80 40 3547.7 0.886 0.881
90 40 1701.6 0.797 0.796 90 40 3255.2 0.876 0.870
100 40 1668.0 0.794 0.793 100 40 2962.7 0.865 0.859
110 40 1634.4 0.792 0.790 110 40 2670.1 0.852 0.846
120 40 1600.8 0.789 0.787 120 40 2377.6 0.839 0.831
130 40 1567.2 0.786 0.784 130 40 2085.1 0.823 0.813
140 40 1533.6 0.783 0.781 140 40 1792.5 0.804 0.792
150 40 1500.0 0.780 150 40 1500.0 0.780

Average density (ave ) 0.800 Average density (ave ) 0.882

 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293 291

Densi ty Profil es 0.30 a) 1 3 6

1.00 0.25 2
0.20

AU
4
0.15 5
0.10
Mobile Phase Density (g/mL)
0.95 a) 0.05
0.00
Average Density 0.0 1.0 2.0 3.0 4.0
0.89 g/mL Average Density Minutes
c) 0.89 g/mL 0.30
b)
0.90 0.25
0.20

AU
b) 0.15
Average Density 0.10
0.88 g/mL 0.05
0.85
0.00
0.0 2.0 4.0 6.0 8.0 10.0
Minutes
0.60 c)
0.80 2.1 x 150 mm, 1.7 m (1.4 mL/min) 0.50
0.40

AU
2.1 x 150 mm, 5 m (0.48 mL/min)
0.30
19 x 150 mm, 5 m (~83 mL/min) 0.20
0.75 0.10
0 25 50 75 100 125 150 0.00
0.0 1.0 2.0 3.0 4.0 5.0 6.0
Column Position (mm)
Minutes

Fig. 16. The left hand side of the gure shows estimated density proles across three different columns with three different ow rates. The ABPR pressures of the experiments
were adjusted in a way such that the average densities of all these operations are very similar. The right hand side of the gure shows the chromatograms corresponding to
these experiments. Note, although the number of plates are varying the selectivity of the analytes remained similar.

Table 4 neither the density nor the temperature is substantially varied. Any
Variation of pressure and density across the 5 m column with 3390 psi ABPR
variation of pressure drops between the original and the target sys-
pressure.
tem should not result into any signicant difference in density. Not
x Temperature Estimated Estimated Sectional having any variation of the density and the temperature ensures
pressure density average density
recreation of the same intrinsic properties in the target system.
(mm) ( C) (psi) (g/mL) (g/mL)
This indicates that we do not need to take any explicit measure to
0 40 3996.0 0.900 0.899 match the intrinsic properties. We just need to address the extrinsic
10 40 3955.6 0.899 0.898
changes brought into the system by changing the column dimen-
20 40 3915.2 0.897 0.897
30 40 3874.8 0.896 0.895 sions and the particle sizes according to the scaling rules.
40 40 3834.4 0.895 0.894 The main difference between scaling in LC and SFC originates
50 40 3794.0 0.894 0.893 from the high compressibility of the mobile phase used in SFC. For
60 40 3753.6 0.892 0.892
separations using CO2 as the principal mobile phase component,
70 40 3713.2 0.891 0.890
80 40 3672.8 0.890 0.889 analyte retention factors are inuenced largely by the mobile phase
90 40 3632.4 0.889 0.888 density and temperature. Because of the high compressibility of
100 40 3592.0 0.887 0.887 CO2 under standard operating conditions, the density can change
110 40 3551.6 0.886 0.885 signicantly with changes in pressure (under isothermal condi-
120 40 3511.2 0.885 0.884
tions), with retention factors increasing with decreasing mobile
130 40 3470.8 0.883 0.883
140 40 3430.4 0.882 0.881 phase density (caused by decreasing pressure). In addition, the
150 40 3390.0 0.880 selectivity and resolution of the analytes may be impacted as they
Average density (ave ) 0.890
respond differently to the same changes in mobile phase density.
This can present a challenge when attempting to transfer a method
between different column congurations that involve changes in
a 1.7 m particle size column, Fig. 16b demonstrates a subsequent column dimensions, ow rates or stationary phase particle size,
transfer of the method to an analytical 5 m column while scaling which in turn alters the pressure (density) prole across the col-
down the ow rate for the lower optimum linear velocity of the umn.
larger particle. The method is further transferred to a preparative We have shown in this article that during scaling if we can gener-
scale, 19 mm 150 mm, 5 m column (Fig. 16c), again by matching ate the same average density inside the column of the target system,
the average density prole of the initial separation on the 1.7 m as it was in the original system, we can approximately create the
particle size column. same intrinsic conditions in both the systems, which should lead to
similar chromatography. Modulation of the average density of the
8. Conclusion target system can be achieved by manipulating the ABPR pressure.
The approach described here should be applicable for an isocratic
Understanding the scaling mechanism in SFC lies in proper system within most of the operating pressuretemperature condi-
understanding of the factors controlling chromatography. The basic tions used in SFC. This approach, however, may not be applicable to
chromatographic parameters, retention and efciency, are con- systems with solvent gradient operations where both the mobile
trolled by intrinsic properties of the mobile phase and some phase concentration and the density vary as a function of time. We
extrinsic factors related to the column dimensions and particle size. need to expand the current theory to address this issue which will
To reproduce the same conditions in the target system we have to be the next step.
ensure that the intrinsic and the extrinsic conditions are the same
in both the systems. Appendix A
These intrinsic properties controlling the chromatography, in
turn, are controlled by the density and the temperature of the sys- If it can be assumed that pressure varies linearly across the
tem. During any scaling operation in LC, except the UPLC conditions, column, the entire pressure prole inside the column can be
292 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293
estimated by interpolating the System (column inlet) and the ABPR
(column outlet) pressures. Let us take the example of the condi-
tions described in Section 7.2.1. In that example the pressures along
the column axis for the 5 m column with 1500 psi ABPR pres-
sure (Table 2) were estimated by linearly interpolating 2004 and
1500 psi, which are the measured System and ABPR pressures of
this system, respectively. The temperature was assumed to have
remained the same (40 o C) along the column (Table 2). We note
from Table 2 that based on these two assumptions we can esti-
mate the complete temperature and pressure proles along the
column. Based on the temperature and pressure proles we can
estimate the density prole by calculating the density as a function
of pressure and temperature at each point. For example the density
of a CO2 /MeOH mixture (90/10,mol/mol(%)) at 40 C and 2004 psi
is 0.8179 g/mL (Table 2), calculated by the REFPROP software of
NIST.
REFPROP calculates the neat CO2 density following the Span
and Wagner equation of state (EOS) [35] and calculates the
CO2 /methanol mixture density using the Kunz and Wagner model
[36]. The errors in the estimation of CO2 density using Span and
Wagner EOS range between 0.03 and 0.05% for CO2 pressures up
to 4350 psi and temperatures up to 250 C [29]. For methanol,
the errors on the values provided by REFPROP are 1% for the
density of the dilute gas and between 0.6 and 3% for that
of the liquid at pressures up to 14500 psi and temperatures
between 0 and 70 C [29]. No specic information regarding the
estimation errors made by Kunz and Wagner mixing rule is avail-
able.
After we calculate the density prole along the column from the
estimated pressure and temperature proles, we can calculate the
average density across the column following the mid-point rectan-
gle rule of numerical integration, which is shown in Eq. (12). Note
that to detect the mid-point of the rectangle, the average density
in each section (i+1 + i )/(2) in Eq. (12) is calculated. Lets call it
sectional average density. For example the sectional average den-
sity value corresponding to the rst row in Table 2 is 0.817 g/mL,
which was obtained as: 8.1685 = (0.8179 + 0.8158)/(2). To obtain
the average density ave across the entire column, individual sec-
tional average density should be multiplied by the sectional length
(10 cm for all the intervals here), then all the values added up and
divided by the column length, L (Eq. (12)). The average density of the
5 m column with 1500 psi ABPR pressure (Table 2) is 0.80 g/mL, as
shown in Fig. 13. Calculations of all the density proles and the aver-
age densities referred in Section 13 followed the same procedure as
described in this appendix.
Acknowledgements
Prof Arvind Rajendran of University of Alberta for providing the
electronic copy of Fig. 11. Steve Collier of Waters Corporation for
helpful discussion and Dr. Jeff Kiplinger of Averica Discoveries for
helpful discussion.
References
[1] G. Guiochon, A. Tarafder, Fundamental challenges and opportunities for
preparative supercritical uid chromatography, J. Chromatogr. A 1218 (2011)
10371114.
[2] A. Rajendran, Design of preparative-supercritical uid chromatography, J. Chro-
matogr. A 1250 (2012) 227249.
[3] L. Miller, Preparative enantioseparations using supercritical uid chromatog-
raphy, J. Chromatogr. A 1250 (2012) 250255.
[4] S. Khater, C. West, E. Lesellier, Characterization of ve chemistries and three
particle sizes of stationary phases used in supercritical uid chromatography,
J. Chromatogr. A 1319 (2013) 148159.
[5] G. Guiochon, A. Felinger, A.M. Katti, D.G. Shirazi, Fundamentals of Prepar-
ative and Nonlinear Chromatography, 2nd ed., Elsevier, Amsterdam, The
Netherlands, 2006.
[6] C. Heuer, P. Hugo, G. Mann, A. Seidel-Morgenstern, Scale up in preparative
chromatography, J. Chromatogr. A 752 (1996) 1929.
[7] T. Gu, K.-H. Hsu, M.-J. Syu, Scale-up of afnity chromatography for puri-
cation of enzymes and other proteins, Enzym. Microb. Technol. 33 (2003)
430437.
[8] T. Gu, Y. Zheng, A study of the scale-up of reversed-phase liquid chromatogra-
phy, Sep. Purif. Technol. 15 (1999) 4158.
[9] A. Tarafder, L. Aumann, T. Muller-Spath, M. Morbidelli, Improvement of an
overloaded, multi-component, solvent gradient bioseparation through multi-
objective optimization, J. Chromatogr. A 1167 (2007) 4253.
[10] B.C. de Neuville, A. Tarafder, M. Morbidelli, Distributed pore model for bio-
molecule chromatography, J. Chromatogr. A 1298 (2013) 2634.
[11] M. Enmark, P. Forssen, J. Samuelsson, T. Fornstedt, Determination of adsorption
isotherms in supercritical uid chromatography, J. Chromatogr. A 1312 (2013)
124133.
[12] D. Guillarme, D.T.-T. Nguyen, S. Rudaz, J.-L. Veuthey, Method transfer for fast liq-
uid chromatography in pharmaceutical analysis: application to short columns
packed with small particle. Part I: Isocratic separation, Eur. J. Pharm. Biopharm.
66 (2007) 475482.
[13] E. Rozet, W. Dewb, E. Ziemonsa, A. Bouklouzec, B. Boulangerd, P. Hubert,
Methodologies for the transfer of analytical methods: a review, J. Chromatogr.
B 877 (2009) 22142223.
[14] R. Gaggeri, D. Rossi, S. Collina, B. Mannucci, M. Baierl, M. Juza, Quick
development of an analytical enantioselective high performance liquid chro-
matography separation and preparative scale-up for the avonoid naringenin,
J. Chromatogr. A 1218 (32) (2011) 54145422.
[15] R.W.H. Perera, et al., Screening approach, optimisation and scale-up for
chiral liquid chromatography of cathinones, J. Chromatogr. A 1269 (2012)
189197.
[16] G. Pettinello, A. Bertucco, P. Pallado, A. Stassi, Production of EPA enriched mix-
tures by supercritical uid chromatography: from the laboratory scale to the
pilot plant, J. of Supercrit. Fluids 19 (2000) 5160.
[17] C. White, J. Burnett, Integration of supercritical uid chromatography into drug
discovery as a routine support tool: II. investigation and evaluation of super-
critical uid chromatography for achiral batch purication, J. Chromatogr. A
1074 (2005) 175185.
[18] K.H. Gahm, et al., Purication method development for chiral separation
in supercritical uid chromatography with the solubilities in supercritical
uid chromatographic mobile phases, J. Pharm. Biomed. Anal. 46 (2008)
831838.
[19] C. Hamman, M. Wong, M. Hayes, P. Gibbons, A high throughput approach
to purifying chiral molecules using 3 m analytical chiral stationary phases
via supercritical uid chromatography, J. Chromatogr. A 1218 (2011)
35293536.
[20] T. Ishihara, T. Kadoya, S. Yamamoto, Application of a chromatography model
with linear gradient elution experimental data to the rapid scale-up in ion-
exchange process chromatography of proteins, J. Chromatogr. A 1162 (2007)
3440.
[21] C. White, Integration of supercritical uid chromatography into drug discov-
ery as a routine support tool: Part I. fast chiral screening and purication, J.
Chromatogr. A 1074 (2005) 163173.
[22] A. Rathore, A. Velayudhan, Scale-Up and Optimization in Preparative Chro-
matography: Principles and Biopharmaceutical Applications, Marcel Dekker,
Inc., New York, USA, 2002.
[23] J.C. Giddings, Liquid chromatography with operating conditions analogous to
those of gas chromatography, Anal. Chem. 35 (1963) 2215.
[24] D.M. Diehl, F. Xia, J.Y. Cavanaugh, K.J. Fountain, J.-A.M. Jablonski, T.E. Wheat,
Method Migration from UPLC Technology to Preparative HPLC. Application
Note, Waters Corporation, 2007, pp. 16, Library No. 720002375EN.
[25] A. Tarafder, G. Guiochon, Use of isopycnic plots in designing operations of
supercritical uid chromatography: I. The critical role of density in determining
the characteristics of the mobile phase in supercritical uid chromatography,
J. Chromatogr. A 1218 (2011) 45694575.
[26] A. Tarafder, G. Guiochon, Use of isopycnic plots in designing operations of
supercritical uid chromatography: II. The isopycnic plots and the selection
of the operating pressuretemperature zone in supercritical uid chromatog-
raphy, J. Chromatogr. A 1218 (2011) 45764585.
[27] A. Tarafder, G. Guiochon, Use of isopycnic plots in designing operations of
supercritical uid chromatography. III: Reason for the low column efciency in
the critical region, J. Chromatogr. A 1218 (2011) 71897195.
[28] H. Darcy, Les Fontaines Publiques de la Ville de Dijon, Dalmont, Paris, France,
1856.
[29] E.W. Lemmon, M. Huber, M.O. McLinden, D.G. Friend, Standard Reference
Data Program, National Institute of Standards and Technology, Gaithersburg,
2014.
[30] A. Tarafder, K. Kaczmarski, M. Ranger, D.P. Poe, G. Guiochon, Use of the isopy-
cnic plots in designing operations of supercritical uid chromatography: IV.
Pressure and density drops along columns, J. Chromatogr. A 1238 (2012)
132145.
[31] D.P. Poe, D. Veit, M. Ranger, K. Kaczmarski, A. Tarafder, G. Guiochon, Pres-
sure, temperature and density drops along supercritical uid chromatography
columns. I. Experimental results for neat carbon dioxide and columns packed
with 3- and 5-micron particles, J. Chromatogr. A 1250 (2012) 105114.
[32] A. Rajendran, O. Kruchi, M. Mazzotti, M. Morbidelli, Effect of pressure drop on
solute retention and column efciency in supercritical uid chromatography,
J. Chromatogr. A 1092 (2005) 149160.
 A. Tarafder et al. / J. Chromatogr. A 1362 (2014) 278293 293

[33] D.P. Poe, J.J. Schroden, Effect of pressure drop on solute retention and column
efciency in supercritical uid chromatography, J. Chromatogr. A 1216 (2009)
79157926.
[34] A. Tarafder, P. Iraneta, G. Guiochon, K. Kaczmarski, D.P. Poe, Estimations of
temperature deviations in chromatographic columns using isenthalpic plots. I.
Theory, J. Chromatogr. A (2014).
[35] R. Span, W. Wagner, A new equation of state for carbon dioxide covering the
uid region from triple-point temperature to 1100 K at pressures up to 800
MPa, J. Phys. Chem. Ref. Data 25 (1996) 15091596.
[36] O. Kunz, W. Wagner, The GERG-2008 wide-range equation of state for natural
gases and other mixtures: an expansion of GERG-2004, J. Chem. Eng. Data 57
(2012) 30323091.