Journal of Chromatography A, 885 (2000) 17–39

www.elsevier.com / locate / chroma

Review

Contributions of theory to method development in solid-phase

extraction
Colin F. Poole*, Ajith D. Gunatilleka, Revathy Sethuraman
Department of Chemistry, Wayne State University, Rm. 171 Chemistry, Detroit, MI 48202, USA

Abstract

The kinetic and retention properties of solid-phase extraction devices are reviewed from the perspective of method
development strategies. Models based on frontal analysis are used to correct retention properties of solid-phase extraction
devices to account for the fact that too few theoretical plates are provided for retention to be independent of kinetic factors.
The available pressure drop for the sampling device largely dictates the choice of useful particle sizes and maximum bed
length. The use of octanol–water partition coefficients and extrapolated values of the retention factor obtained by liquid
chromatography are poor empirical models for the estimation of breakthrough volumes with water as the sample solvent. The
solvation parameter model provides an adequate description of sorbent retention for the estimation of breakthrough volumes,
rinse solvent volume and composition, and elution solvent volume and composition. Combining the frontal analysis and
solvation parameter models offers a comprehensive approach to computer-aided method development in solid-phase
extraction. This is the first step in the development of a structure-driven approach to method development in solid-phase
extraction that should be more reliable and less tedious than traditional trial and error approaches.  2000 Elsevier Science
B.V. All rights reserved.

Keywords: Reviews; Breakthrough volume; Sorbent retention; Solid-phase extraction; Method development

Contents

1. Introduction ............................................................................................................................................................................ 18
2. Identification of critical parameters for modeling in solid-phase extraction .................................................................................. 19
3. Breakthrough volumes ............................................................................................................................................................. 21
3.1. Experimental determination of breakthrough volumes........................................................................................................ 22
3.2. Methods for estimating breakthrough volumes .................................................................................................................. 23
3.3. Models for predicting breakthrough volumes .................................................................................................................... 23
3.4. Optimization of physical properties for sampling devices ................................................................................................... 24
4. Sorbent retention ..................................................................................................................................................................... 28
4.1. Experimental determination of retention factors ................................................................................................................ 28
4.2. Methods for estimating retention factors ........................................................................................................................... 29
4.3. Solvation parameter model .............................................................................................................................................. 30
5. Other approaches .................................................................................................................................................................... 37
6. Conclusions ............................................................................................................................................................................ 38
References .................................................................................................................................................................................. 38

*Corresponding author. Tel.: 11-313-577-2881; fax: 11-313-577-1377.

E-mail address: [email protected] (C.F. Poole)

0021-9673 / 00 / $ – see front matter  2000 Elsevier Science B.V. All rights reserved.
PII: S0021-9673( 00 )00224-7
18 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

1. Introduction Sample processing in solid-phase extraction con-

sists of four distinct steps. Initially, the sorbent is
Solid-phase extraction for liquid samples became a conditioned with solvent to improve the reproducibil-
widely used laboratory technique following the ity of analyte retention and to reduce the carry
introduction in the 1970s of disposable sorbent through of sorbent impurities at the elution stage
cartridges containing porous particles sized to allow [1–4,13]. The conditioning solvent is then replaced
sample processing by gentle suction [1–4]. A typical with the same solvent as the sample solvent and the
solid-phase extraction cartridge consists of a short sample passed through the sampling device at a
column (generally an open syringe barrel) containing controlled flow-rate. Optionally, after the sample has
a sorbent with a nominal particle size of 50–60 mm, been processed, the sorbent is rinsed with a weak
packed between porous metal or plastic frits. A large solvent to displace undesired matrix components
number of sorbents are in use today corresponding to from the sorbent without displacing the analytes.
the desire for general purpose, class-specific and Finally, the analytes of interest are eluted from the
even compound-specific extractions [5–7]. In addi- sorbent in a small volume of strong solvent for
tion, several approaches to automation of solid-phase subsequent determination. The conditioning step is
extraction based on robotics, dedicated instruments critically important for processing aqueous samples
using flow-processing, and on-line analyzers with using particle-loaded membranes but is less impor-
direct coupling of the extraction column to a chro- tant for other sampling devices except as a general
matographic instrument have gained acceptance and approach to minimize contamination of extracts by
are used in many laboratories [1,8,9]. sorbent impurities. The high surface tension of water
The early 1990s saw the development of alter- combined with the microporosity of the particle-
native sampling formats based on disk technology loaded membranes results in slow and uneven flow
that have prospered to challenge the supremacy of through the membrane and low analyte recovery if
cartridge devices for off-line sample processing the membranes are not first conditioned with an
[10,11]. Disk technology was conceived to tackle organic solvent. For large sample volumes a small
several problems encountered in the use of car- amount of the same organic solvent is usually added
tridges. Namely, slow sample processing rates, low to the sample to maintain a constant sample flow-
tolerance to blockage by particles, the inadequate rate. The drying step between processing aqueous
and variable packing density of cartridges, to mini- samples and eluting the retained analyte with a
mize non-specific matrix adsorption, and to miniatur- water-miscible organic solvent is also important. The
ize sampling devices for processing small sample purpose of the drying step is to reduce the volume of
volumes. At least three different disk devices are water retained by the eluting solvent. Excess water
commercially available. The particle-loaded mem- may interfere in further concentration of the eluent
branes consist of 8–10-mm sorbent particles sus- by the gas-blow down method. Other practical
pended in a web of PTFE microfibrils. The mem- considerations associated with sample processing by
branes have a homogeneous structure containing solid-phase extraction are summarized in Table 1.
about 80% (w / w) or more of sorbent particles The general approach to method development in
formed into circular disks 0.5 mm thick with diame- solid-phase extraction has remained rather empirical
ters from 4 to 96 mm. Particle-embedded glass fiber based on experimental trial-and-error procedures.
disks contain 10–30-mm sorbent particles woven into General guides for sorbent selection, such as Fig. 1,
a glass-fiber matrix available in a wide range of sizes can provide a useful starting point given a minimum
[12]. SpeedisksE consist of a thin sandwich of 10- amount of sample information. Structured empirical
mm sorbent particles held between two glass-fiber rules then permit the method development process to
filters with a screen to hold the filters in place [1]. be completed in an ordered if inexact fashion [1–4].
Disk technology has gained acceptance for process- The outcome is a recipe, hopefully for success, at the
ing large sample volumes and small diameter disks expense of the human and physical laboratory re-
for processing small samples, but otherwise cartridge sources consumed. In any event, it is a tedious and
devices are still dominant. time-consuming process bringing little personal
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 19

Table 1
Experimental variables that affect analyte recovery by solid-phase extraction
Conditioning solvent (typically 3–5 holdup volumes)
(a) Ensures reproducible retention and flow. Critical step for particle-loaded membranes
(b) Helps to minimize contamination of extracts by sorbent impurities
(c) Replace by sample solvent before processing sample
Flow rates (typical range 0.2–1.5 mm / s)
(a) More critical for cartridges than disks due to their variable and heterogeneous packing density (channeling)
(b) More critical when the sample volume exceeds the breakthrough volume as typical sampling devices provide too few theoretical
plates for flow independent retention
Sample properties
(a) Dilute viscous samples with a weak low viscosity solvent to reduce sample processing time
(b) Remove excessive particle matter by filtration or centrifugation to maintain a constant sample-processing rate
(c) Add small volume of organic solvent (1–3%, v / v) to large volume water samples to ensure sorbent remains solvated and to maintain a
constant (fast) sample-processing rate. Important for particle-loaded membranes
(d) Adjust pH to reduce ionization of weak acids and bases for reversed-phase sampling
(e) Maintain approximately constant ionic strength for samples and standards when using reversed-phase sampling conditions. Ionic
strength is a critical parameter for ion-exchange extraction
(f) Deproteination of biofluids may be required for acceptable recovery of low-molecular weight analytes for reversed-phase sampling
(g) Precipitation of inorganic acids (sulphate, phosphate, etc.) by barium hydroxide is sometimes required for acceptable recovery of
organic acids from biofluids using ion-exchange extraction
Drying time (typically 1–5 min, but sometimes considerably longer)
(a) Sufficient to remove all sample solvent trapped in the sorbent pores
(b) Excessive drying may result in low recovery of analytes from evaporation or retention in poorly solvated regions of the sorbent
Rinse solvent (optional)
(a) Small volume of intermediate strength solvent to elute matrix components. Analytes remain immobilized on the sorbent
(b) Biological fluids, plant extracts and soil extracts often require a rinse step but surface waters may not
Eluting solvent (ideally 2–3 holdup volumes but often larger)
(a) Should be a strong solvent able to displace all analyte from the sorbent in a small volume
(b) Normally should be volatile and miscible with the sample solvent

gratification. Alternative approaches based on com- pounds from water, this has been the application area
puter-aided strategies and simulation require an most extensively studied within the framework of
appropriate level of theory so that at decision steps in our review, and provides the major portion of the
the method development process fast simulation or work reported here.
calculation procedures can be used in place of trial-
and-error experiments. Solid-phase extraction seems
to have had minimal appeal to the theoretician in 2. Identification of critical parameters for
spite of its obvious importance as a general sample modeling in solid-phase extraction
preparation procedure. Real progress towards a theo-
retical framework for optimization of the design of Most of the parameters that describe the sequence
solid-phase extraction devices and the development of processing steps in solid-phase extraction are
of structure-based, computer-aided approaches to amenable to measurement by liquid chromatography
method development has taken place over the last or estimation using theoretical principles derived
few years with promises that a more rational ap- from the theory of liquid chromatography [4,14–16].
proach to solid-phase extraction will soon be avail- Analyte concentrations are generally low and the
able. These studies are the feature of this review. type and amount of sorbent required to isolate
Other articles in this issue cover solid-phase mi- sufficient analyte for its convenient determination is
croextration, ion-exchange and ion-pair extraction, indicated by the breakthrough volume for the sam-
automation, and gas-phase sorbent trapping, which pling device. Ultimately, the sample volume that
are not discussed here. Although not entirely re- must be processed depends on the analyte con-
stricted to the sorbent extraction of neutral com- centration and the operating characteristics of the
20 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

Fig. 1. Method selection guide for the isolation of organic compounds from solution. SAX, strong anion exchanger; SCX, strong cation
exchanger; WCX, weak cation exchanger; RP, reversed-phase sampling conditions; NP, normal-phase sampling conditions; IE, ion-exchange
sampling conditions.

instrument selected for the determination. The sam- from the sorbent while leaving the analytes of
pling device is required to accommodate this sample interest immobilized on the sorbent. This can often
volume to deliver sufficient analyte for a secure be restated as the identification of the strongest
determination. The analyte concentration is generally solvent that eliminates matrix components from the
unknown, and this is the likely reason for analyzing sorbent without loss of analyte. This is largely
the sample. Consequently, methods are generally dictated by analyte retention factors. Optimum con-
established to ensure that if the sample contained ditions can be established by selecting eluting con-
more than a certain minimum analyte concentration ditions that preserve a certain minimum value for the
then the analytical protocol would provide a secure retention factor of the least retained analyte of
method for its determination. Minimum analyte interest.
concentration limits may be set by regulatory au- To accomplish a significant concentration of the
thorities for monitoring purposes or are established analytes of interest with minimal further sample
from a knowledge of expected concentrations given manipulation it is desirable to recover the analytes in
the origin of the sample (e.g., a single dose of a drug a small solvent volume. For this purpose it is
given to an animal to study its clearance and necessary to identify a solvent composition in which
excretion). the analytes have minimal retention factors. General-
A rinse solvent may be selected for matrix simpli- ly the minimum elution volume that can be safely
fication. The type and amount of sorbent is fixed by employed, unless backflushing is used, is about 2–3
the isolation step so the need here is to optimize times the holdup volume for the sampling device.
solvent composition to displace matrix components This corresponds to a retention factor ,2. If a
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 21

solvent is chosen for analyte elution that leaves some

matrix components immobilized on the sorbent then
matrix simplification is also achieved.
To select and optimize the sample processing
conditions for solid-phase extraction requires knowl-
edge of the retention properties of the analytes on the
sorbent under different mobile phase conditions
corresponding to sample application, rinse (matrix
simplification) and elution. In addition, the sorption
mechanism for sample application occurs according
to the characteristics of frontal analysis and the rinse
and desorption step by elution. Typical solid-phase
extraction devices contain short sorbent beds and
cannot be expected to provide many theoretical
plates. Therefore, the possibility that retention may
be influenced by kinetic properties has to be consid-
ered. These features are discussed presently. For
completeness, we note early work on the optimi-
zation of solid-phase extraction processing condi-
tions using liquid chromatography by Weidolf and Fig. 2. Typical representation of a breakthrough curve. VB is the
Henion [17] and Casas et al. [18]. Weidolf and breakthrough volume, VR the chromatographic elution volume, VC
Henion used columns packed with a solid-phase the sample volume corresponding to the isolation of the maximum
extraction sorbent and HPLC to predict the com- amount of analyte, C0 the concentration of analyte in the sample,
and sV the standard deviation of the derivative curve for the plot.
position of rinse and elution solvents for the isolation
of conjugated steroids from equine urine. The vol-
ume and composition of the rinse solvent for matrix interest. The breakthrough volume is established
simplification was determined from the retention from a breakthrough curve, as indicated in Fig. 2. In
factor of the peak front of the analytes and the the initial sampling phase a sample of fixed con-
composition for elution from the retention factors for centration, and usually at a constant velocity, enters
the peak tails in appropriate mobile phases. Casas et the sampling device. The analytes are quantitatively
al. [18] proposed a method to establish the relation- retained during the initial sampling phase by the
ships between the composition of the strongest sorbent up to the point that the sample volume
binary solvent allowing complete sorption and the exceeds the retention capacity of the sorbent. Further
weakest binary solvent causing instantaneous elution sample entering the sorbent bed will not be quantita-
for a series of structurally related standards on a tively retained by the sorbent and eventually the
sorbent cartridge and the retention factor for the concentration of analytes entering and exiting the
same standards determined by HPLC. From these sampling device become identical. The point on the
relationships a model was obtained to predict the curve at which some arbitrary amount of sample is
most suitable rinse and elution solvents for other detected at the outlet of the sampling device is the
compounds with a family resemblance to the stan- breakthrough volume (VB ). Different specifications of
dards from their HPLC retention factors. the ratio of the inlet to outlet concentration in the
definition of the breakthrough volume result in
different values. Typically 1, 5, or 10% are selected.
3. Breakthrough volumes From an experimental point of view the chosen
definition reflects the measurement difficulty of
The breakthrough volume is the most important determining small changes in the concentration of
characteristic parameter to determine the suitability the analytes at the outlet of the sampling device.
of a sampling device for isolating the analytes of From the perspective of modeling the breakthrough
22 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

process a value of 1% is generally chosen in keeping column bed is recorded producing a signal similar to
with the desire to define a maximum sample volume that indicated in Fig. 2. The analyte concentration is
that can be processed with minimal (acceptable) loss selected such that it does not overload the sorbent
of analyte. while providing a reasonable detector response for
The shape of the breakthrough curve is sigmoidal. ease of recording the breakthrough curve. Since the
A second point on the breakthrough curve, VC , breakthrough volume may be flow-rate dependent the
corresponds to the sample volume at which the flow-rate used to record the breakthrough curve
retention capacity of the sorbent is saturated and the should be similar to the flow-rate used for sample
concentration of analyte exiting the sampling device application. In the absence of a suitable method for
is the same (actually 1002% used to define the on-line detection the effluent from the sorbent bed
breakthrough volume) as that entering the sampling can be collected by aliquots using a fraction collector
device. It corresponds to the minimum volume of and each aliquot analyzed separately to determine
sample that will result in the isolation of the maxi- when breakthrough occurred [22].
mum amount of analyte by the sorbent. The point of For standard cartridge and disk devices off-line
inflection for the breakthrough curve corresponds to sample processing is commonly used. Samples are
the chromatographic retention volume, VR , since the processed in aliquots and an off-line detection meth-
first derivative of the breakthrough curve is a Gaus- od used to determine the analyte concentration in the
sian distribution similar to the peak response ob- extracts recovered from each sample aliquot [13,23–
served during elution chromatography. This is true 25]. Aliquots are processed in the same way as
for typical columns used in HPLC but is not regular samples. Each aliquot contains the same
necessarily the case for sampling devices with low amount of analyte but in a different sample volume.
numbers of theoretical plates [19]. Initially, an approximate value of the breakthrough
In general, there are two common causes of volume is established by using decade changes in
premature breakthrough in frontal chromatography. aliquot volumes, followed by a more systematic
The retention capacity of the sorbent bed is over- experimental design. For compounds with an esti-
loaded due to a high concentration of either analyte mated breakthrough volume between 0 and 50 ml,
or sorbed matrix components. This should not be a measurements are made at 2.5-ml volume incre-
problem for trace analysis of environmental water ments, between 50 and 100 ml at 5-ml increments,
samples where analyte concentrations are generally 100 and 1000 ml at 10-ml increments, and .1000 ml
low and the matrix burden tolerable. Secondly, the at 100-ml increments. Plotting the observed recovery
sorbent bed may fail to adequately retain the analytes for the complete sampling process against the corre-
due to the provision of an insufficient number of sponding aliquot volumes generates the breakthrough
theoretical plates for retention volumes to be in- curve. The breakthrough volume is estimated from
dependent of the plate count for the sampling device the line representing the best fit through the ex-
(Section 3.4). perimental data.
A similar approach to the above has been used to
3.1. Experimental determination of breakthrough determine the breakthrough volume of precolumn
volumes traps in on-line SPE-HPLC [26–29]. In this case the
precolumn sorbent trap is connected to the analytical
Breakthrough volumes are typically determined in column through a selection valve that allows the
one of two ways [20,21]. The most straightforward is sample to by-pass the analytical column during
the direct method using either on-line or off-line loading and the eluent for recovery to be switched to
detection. This is particularly convenient for pre- the analytical column for separation and determi-
column sorbent traps used in on-line SPE-HPLC. A nation. A small sample volume containing the ana-
solution containing a constant concentration of ana- lytes is pumped through the precolumn and the
lyte is pumped at a constant flow-rate through the chromatogram obtained by on-line elution recorded.
precolumn, which is connected directly to the detec- The initial sample volume is selected to be less than
tor. The appearance of the analyte at the exit of the the breakthrough volume. Further sample volumes
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 23

are then prepared as multiples of the initial sample Many regulatory agencies require the determi-
volume with each one spiked with a constant amount nation of the octanol–water partition coefficient
of analyte. Each sample is then processed as for the (log P or log KOW ) as a component of the approval
initial sample volume. Initially there will be no process for the use of new chemical entities. The
change in peak areas up to the breakthrough volume, octanol–water partition coefficient has become a
then in each subsequent volume that exceeds the recognized parameter to estimate compound hydro-
breakthrough volume there will be a decrease in peak phobicity of general importance for modeling numer-
areas in the chromatogram representing the loss of ous environmental and biological properties associ-
analyte from the precolumn due to its breakthrough. ated with the use and disposal of organic compounds
Plotting peak area for each analyte against the [35,36]. Nakamura et al. [37] proposed a general
sample volume processed enables the breakthrough guide for the selection of sample processing con-
volume to be estimated from the breakthrough curve. ditions for the isolation and recovery of agricultural
A mathematical model for estimating the break- chemicals from water based on estimated log KOW
through curve for precolumns used in SPE-HPLC values. A group of empirical rules were presented in
was proposed by Ferrer et al. [30] based on the the form of a decision tree and said to be useful for
assumption that the continuous flow of a solution in selecting reversed-phase sorbents and organic sol-
frontal chromatography could be represented as the vents to aid experimental trial-and-error approaches
consecutive flow of many discrete volumes. Each to method development. Hennion et al. [15] investi-
volume injected gives a chromatographic peak sepa- gated the use of log KOW to estimate log k W (sor-
rated from the next peak by the injection repetition bent–water retention factor) for the purpose of
rate. The sum of the individual peak areas measured modeling breakthrough volumes but concluded that it
at different mobile phase compositions simulates a was of limited value. In Section 4.3 we will provide
breakthrough curve from which the breakthrough a fundamental argument for why log KOW provides a
volume can be estimated. poor surrogate model for the isolation of organic
compounds from water.
3.2. Methods for estimating breakthrough volumes
3.3. Models for predicting breakthrough volumes
The determination of breakthrough volumes, par-
ticularly by off-line methods, is time consuming and From the general theory of frontal chromatography
somewhat subjective. Consequently, methods that it is possible to derive a relationship between the
enable breakthrough volumes to be estimated from breakthrough volume and the sorption properties of
solute properties or calculated from models that solid-phase extraction devices [15,21,29,38–44]. At
require a minimal number of experimental measure- the 1% breakthrough level the breakthrough volume,
ments (Section 3.3) are useful. Early in the develop- VB , is related to the retention volume, VR , through Eq.
ment of solid-phase extraction various relationships (1):
were sought between the breakthrough volume and
the aqueous solubility of compounds [31–33]. These VB 5VR 2 2.3sV (1)
methods, although providing some rough agreement where sV is the standard deviation depending on the
in a few cases, completely ignored the contribution axial dispersion of the analyte along the sorbent bed
of the sorbent to the retention process and are further (see Fig. 2) and is evaluated through Eq. (2):
limited by the availability and quality of solubility
data, particularly for sparingly soluble compounds. sV 5VM (1 1 k) / œN (2)
Numerous methods exist for the estimation of aque-
where VM is the interparticle volume of the sorbent
ous solubility from structural parameters [34], which
bed (holdup volume), k the retention factor (capacity
could fill the availability gap, but the lack of any
factor), and N the plate number for the sorbent bed
quantitative fundamental relationship between solu-
calculated by Eq. (3):
bility and the breakthrough volume diminishes the
attraction of this approach. N 5VR (VR 2 sV ) /s 2V (3)
24 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

In principle it should be possible to calculate VB from phase extraction devices can now be discussed. The
Eqs. (1)–(3) by determining VM and N for the main requirements are the optimization of size (VM
sampling device and measuring VR (or k) for the and N), kinetic properties (N, particle size, flow-rate)
analytes of interest. N and VM are easily determined and retention (N, k and VM ).
for precolumn sorbent traps used in coupled SPE-
HPLC by direct measurement since the on-line
3.4. Optimization of physical properties for
arrangement allows the recording of the break-
sampling devices
through curves. There is no convenient way to make
measurements of these parameters for off-line sam-
To maximize the breakthrough volume in the
pling conditions, and therefore estimates must be
absence of any constraints Eqs. (1)–(4) indicate that
made from results obtained for the solid-phase
the sampling device should have a large holdup
extraction sorbent by HPLC [14,25,45]. Eqs. (1)–(3)
volume (equivalent to stating that the sorbent bed
are derived assuming that the conditions of linear
should be large). The selected sorbent should provide
chromatography apply and the plate count for the
a high retention of the analyte under the sampling
sorbent trap is reasonably large. Large sample
conditions (large k) combined with a sufficiently
amounts or strongly retained matrix components
large value for N that retention is largely indepen-
resulting in curved sorption isotherms [46] or sorbent
dent of N. One could hope to maximize retention by
beds with very low numbers of theoretical plates [19]
sorbent selection but the other characteristic parame-
may result in poor estimates of the breakthrough
ters of the sampling device will have to be arrived at
volume based on the above equations.
through compromise.
Lovkvist and Jonsson [19] proposed a model
For off-line or on-line sampling the sorbent bed
described by Eq. (4) to characterize the sampling
cannot be too large because it is desirable to recover
properties of sorbent beds with a small number of
the analytes in a small solvent volume. In addition,
theoretical plates that has been adopted by other
in off-line sampling the pressure drop available to
groups [14,23,45,47–50] to calculate breakthrough
provide transport of the sample through the sampling
volumes under solid-phase extraction conditions. The
device at an optimum or practical velocity is limited.
coefficients a 0 , a 1 , a 2 are characteristic of the
The pressure drop per unit length of sorbent bed is
breakthrough level and:
given by Eq. (5) [51]:
VB 5 (a 0 1 a 1 /N 1 a 2 /N 2 )21 / 2 (1 1 k)VM (4) DP/L 5 uhf /d 2P (5)

are summarized in Table 2 [19]. Just as for Eqs. where DP is the pressure drop across a sorbent bed
(1)–(3) the calculation of the breakthrough volume of length L, u the linear velocity of the sample
requires the determination of N and VM for the solution through the sorbent bed, h the viscosity of
sampling device and either a measurement or estima- the sample solution (¯1310 23 N?s m 22 for an
tion of the retention factor. The influence of these aqueous solution), f the flow resistance parameter
parameters with respect to the performance of solid- for the sorbent bed (typically 10 3 for cartridges and
particle-loaded membranes [12,14,45,49,52]) and d P
Table 2 the average particle size for the sorbent. A plot of
Coefficients for the Lovkvist and Jonsson model, Eq. (4), for
DP/L against d P is shown in Fig. 3 for different
sorbent traps with small N values
sample velocities corresponding to 0.11, 0.22, 0.43
Breakthrough Coefficients and 1.08 mm / s. These values are equivalent to a
level (%)
a0 a1 a2 sample flow-rate of 0.5, 1.0, 2.0 and 5.0, ml / min
0.1 0.998 29.12 57.54 through a 1-cm diameter cartridge. For off-line
0.5 0.990 17.92 26.74 sampling using cartridge devices the available pres-
1.0 0.980 13.59 17.60 sure drop is limited to about 0.9 atm (as indicated in
5.0 0.903 5.36 4.60 Fig. 3 for a 1-cm bed length). Although shorter and
10.0 0.810 2.88 1.94
wider cartridges could be used to increase the sample
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 25

This is conveniently achieved using particle-loaded

membranes and particle-embedded glass fiber disks
[12,49]. To operate with a low-pressure drop these
materials are short (0.5 mm) and packed with small
particles to provide reasonable efficiency. Mem-
branes and disks with a 1-cm height and packed with
cartridge-size sorbents would have an inconveniently
large holdup volume requiring large solvent volumes
for desorption. This represents a poor compromise
and if large diameter devices are to be used they
should be thin and packed with small diameter
particles.
Sorbent cartridges with short beds packed with
course particles, or for that matter, very short beds
packed with fine particles, cannot be expected to
provide large plate numbers. Fig. 4 provides a plot of
plate height against interparticle mobile phase ve-
locity, the velocity of mobile phase passing through
the interparticle space, for several cartridge sorbents

Fig. 3. Plot of (DP/L) against the average particle size (d P ) for

different linear velocities (u). A50.11 mm / s, B50.22 mm / s,
C50.43 mm / s and D51.08 mm / s. The horizontal line represents
the practical limit of 0.9 atm for a 1-cm bed length.

flow-rate the results shown in Fig. 3 are fairly typical

of actual sampling conditions. In practice the avail-
able pressure drop limits cartridge sampling devices
to particle sizes greater than about 40 mm with larger
particles providing a wider and higher range of
sample flow-rates. For a flow-rate of 1 ml / min and a
sorbent bed of 1 cm diameter and 1 cm high the
optimum particle size is ¯49 mm with a maximum
pressure drop of 0.9 atm. Faster sample flow-rates
require larger particles if the pressure drop cannot be
increased. For on-line sampling using high-pressure
pumps for sample application pressure is no longer
the factor that dictates the particle size. In this case,
particle sizes much smaller than those indicated for
off-line sampling can be accommodated. These
columns are also generally short to minimize sample Fig. 4. Plot of the plate height (mm) against mobile phase
desorption volumes. This results in conditions favor- interparticle velocity (mm / s) for silica-based cartridge sorbents.
able for high sample processing rates. For off-line The test compound was anthracene and the mobile phase metha-
nol–water (80:20, v / v). Identification: (1) octadecylsiloxane-
sampling the easiest solution to increase sample-
bonded sorbent (light loading); (2) spacer-bonded propanediol
processing rates is to increase the diameter of the sorbent; (3) cyanopropylsiloxane-bonded sorbent; and (4)
sampling device at a constant bed height since the butylsiloxane-bonded sorbent. (Reproduced with permission from
flow-rate is proportional to the diameter squared. Ref. [14]. Copyright Elsevier Science Publishers).
26 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

[14]. There is no observed minimum in the plots

over the interparticle velocity range of 0.5–5.0 mm / s
corresponding to a flow-rate of about 3–30 ml / min
through a 1-cm diameter cartridge. The main contri-
bution to the plate height arises from flow anisotropy
in the packed bed and resistance to mass transfer.
Over the velocity range indicated the various sor-
bents provide from five to 15 plates per centimeter of
bed length for the butylsiloxane-bonded sorbent, six
to 10 for the cyanopropylsiloxane-bonded sorbent,
10–20 for the spacer-bonded propanediol sorbent
and 15–40 for the lightly loaded octadecylsiloxane-
bonded sorbent [14,25,43,45,52–54]. If cartridges
with a 1-cm diameter are operated at a low sample
flow-rate of 1 ml / min they cannot be expected to
provide more than about 20 plates per centimeter of
bed length. At higher sample flow-rates only lower
plate numbers are expected. Significant variation in
the plate count as a function of the sample flow-rate
is a likely cause of flow-rate dependence for break-
through volumes (see later in this section). For
sorbent beds with a low packing density even smaller
plate numbers are to be anticipated. For example, it Fig. 5. Plot of the plate height (mm) against the interparticle
mobile phase velocity (mm / s) for a particle-loaded membrane
was shown that typical cartridges contain on average (velocity axis corresponds to a sample flow-rate of 0–100 ml / min
about 10–25% additional empty space compared to a for a disk with a 38-mm diameter active sampling area).
stable consolidated sorbent bed [14]. The efficiency
of these sorbent beds will be lower than indicated by
Fig. 4 due, at least in part, to increased flow The influence of low plate numbers on break-
anisotropy and channeling. through curves for solid-phase extraction devices can
The kinetic properties of particle-loaded mem- be calculated using Eq. (4). In Fig. 6, breakthrough
branes are different to those of cartridge sampling curves are given for a sampling device with a holdup
devices (Fig. 5) [49]. A minimum value for the plate volume of 0.42 ml, a solute retention factor of 100,
height, corresponding to about seven particle diame- providing 5, 20 and 100 plates per cm bed length.
ters, is observed at an optimum interparticle velocity Increasing plate numbers results in a sharper front
of about 0.19 mm / s. Fitting the data to the Knox boundary and a larger breakthrough volume (23 ml
equation indicates that the contribution of flow for N 5 5, 34 ml for N 5 20 and 40 ml for N 5 100).
anisotropy is about three times greater and the Larger plate numbers are desirable for maximum
contribution from resistance to mass transfer about sample retention but small values of N are capable of
an order of magnitude greater than predicted for a providing useful breakthrough volumes. In practice,
homogeneous consolidated bed. The heterogeneous typical sorbent cartridges and membrane devices
structure of the membranes probably contributes used for off-line sampling provide values of N in the
unfavorably to their kinetic properties. Over the range of 5–30 where the breakthrough volume
typical flow-rate range for a 47-mm diameter disk depends on the kinetic properties of the sampling
with a 38-mm active sampling area, 10–100 ml / min, device.
the particle-loaded membrane will provide about four The flow-rate dependence of the breakthrough
to nine theoretical plates, with the largest values in volume for cartridges arises because the plate height
the region of the optimum flow-rate (about 13 ml / is (generally) linearly dependent on the mobile phase
min). velocity for typical sample flow-rates. The break-
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 27

Fig. 6. Plot of the breakthrough curves for a sampling device with Fig. 7. Plot of the breakthrough volume against the mobile phase
a holdup volume of 0.42 ml, retention factor 100, and five (A), 20 interparticle velocity for a cyanopropylsiloxane-bonded sorbent
(B) and 100 (C) plates per cm of bed length. (A) and lightly loaded octadecylsiloxane-bonded sorbent (B). The
data was modeled for a 1-cm diameter cartridge with a 1-cm bed
length, holdup volume of 0.42 ml and a retention factor of 100.
Plate height values were taken from Fig. 4.
through volume as a function of the mobile phase
interparticle velocity (representing a flow-rate range
of about 1–40 ml / min for a 1-cm diameter cartridge) diameter active sampling area. At both higher and
for two sorbents is shown in Fig. 7. Curve A is for lower flow-rates a decrease in the breakthrough
the cyanopropylsiloxane-bonded sorbent and curve B volume is observed.
the lightly loaded octadecylsiloxane-bonded sorbent. When it comes to the recovery of analytes from
In terms of kinetic performance they represent the the sorbent by elution in a small solvent volume it is
extreme cases with the cyanopropylsiloxane-bonded necessary to consider the shape of the frontal elution
sorbent providing from three to 10 plates over the curve as well as the retention capacity of the sorbent
velocity range illustrated and the octadecylsiloxane- [42]. The required elution volume for 99% recovery
bonded sorbent nine to 27 plates. There is a strong of an analyte, VE , on a sorbent trap with a low plate
reduction in the breakthrough volume with increas- number is given by Eq. (6):
ing sample processing rates. There is no optimum
value but the largest breakthrough volume is ob- VE 5VM [1 1 k][1 1 (2.3 / œN)] (6)
tained at the lowest sample flow-rates. Similar
effects are observed for particle-loaded membranes The only practical way to minimize the volume of
except in this case there is a maximum value for the eluting solvent is to use a small sorbent bed (mini-
breakthrough volume corresponding to the optimum mize VM ) and a strong solvent (k , 3 and ideally 0).
sample flow-rate. Since the change in efficiency Sorbent traps with relatively large values of N
around the optimum value is shallow the break- provide sharper desorption front profiles and require
through volume is not strongly affected by flow-rate a smaller elution volume to quantitatively recover the
in the range 10–30 ml / min assuming a 38-mm analyte from the sorbent trap.
28 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

4. Sorbent retention extraction. The equilibrium method uses a fixed

volume of sample solution with a known concen-
It is convenient to rearrange Eq. (4) into the tration of analyte that is continuously circulated by a
general expression: mechanical pump through the sampling device and
returned to the sample solution reservoir until a
log V 5 log QVM 1 log(1 1 k S ) (7) steady state is reached [43,44,55–58]. On-line moni-
toring of the analyte solution exiting the sampling
where V is either the breakthrough volume, volume
device with a UV detector is one possible method to
of rinse solvent or elution volume, Q the contribu-
establish the time required to reach a steady state.
tion of kinetic factors to retention resulting from the
Since mass transfer in liquids is slow this usually
small number of theoretical plates, VM the holdup
requires many hours. The sorbed amount of analyte
volume for the sorbent bed, and k S the retention
for steady-state conditions is then determined by
factor for the analyte with the sample solvent, rinse
elution with a small volume of strong solvent using
solvent or elution solvent as mobile phase. For a
any suitable method for quantitation. Provided that
limited range of flow-rates and a specified sampling
the quantity of sorbed analyte does not exceed the
device, the product QVM is approximately constant.
linear portion of the sorption isotherm the retention
Given the typical numerical values for QVM it is easy
factor can be calculated without resort to additional
for log(11k S ) to become the dominant term in Eq.
data other than the volume and concentration of the
(7) and the most important factor determining the
analyte in the sample solution, the holdup volume for
breakthrough volume. Changes in the holdup volume
the sorbent bed, and the amount of analyte taken up
and plate number for the sorbent bed provide only an
by the sorbent [43,58].
incremental change in log V. On the other hand
The alternative method of determining the re-
differences in the retention factor result in a selective
tention factor is by direct measurement of retention
change in log V. Since the retention factor for a
in a typical chromatographic experiment
particular compound depends on its range of inter-
[25,41,45,50,52,53,57,59,60] or forced-flow planar
molecular interactions with the sorbent and sample
chromatography [12,49,61] in the case of particle-
solvent then sorbent selection should have a large
loaded and particle-embedded membranes. Columns
influence on log V and the practical values for
are prepared by dry packing for sorbents typically
breakthrough, rinse and elution solvent volumes [25].
used for off-line solid-phase extraction and by slurry
To maximize the breakthrough volume requires the
packing for on-line precolumns. The choice between
selection of sample processing conditions that maxi-
slurry packing and dry packing is based on the
mize the retention factor. The selection of a rinse
average particle size for the sorbent. Particles larger
solvent requires identification of experimental con-
than about 20 mm being easily dry packed and
ditions that preserve a sufficiently large retention
smaller diameter particles requiring slurry packing
factor to avoid loss of analyte during the rinse step.
[51]. Columns are usually 10–35 mm in length and
The selection of the elution conditions requires
of any convenient internal diameter. Larger diameter
identification of experimental conditions that mini-
columns can be operated at higher flow-rates while
mize the retention factor. Methods that allow the
minimizing extracolumn contributions to retention
measurement or estimation of retention factors for
measurements. Compounds with retention factors up
relevant sampling conditions are then of prime
to about 10 4 can be determined in this way. It is
importance for method development in solid-phase
possible to determine retention factors for more
extraction.
highly retained compounds but column residence
times are inconveniently long. In practice, com-
4.1. Experimental determination of retention pounds with k . 10 4 are more than adequately
factors retained for most likely sampling conditions and an
accurate determination of their retention factor is
There are two general approaches for determining rarely required. Of more general interest is the
the retention factor of sorbents used for solid-phase influence of sampling conditions and the choice of
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 29

sorbents in the k , 10 3 range, since this has a direct conform to contemporary practice, but is far from
bearing on the suitability of a particular method. adequate, and is difficult to justify.
Some typical plots of log k against mobile phase
4.2. Methods for estimating retention factors composition for an octadecylsiloxane-bonded silica
sorbent with acetonitrile–water as mobile phase and
There is nothing particularly difficult in determin- a cyanopropylsiloxane-bonded silica sorbent with
ing retention factors using either the equilibrium methanol–water as mobile phase are shown in Figs.
method or the chromatographic method. Both meth- 8 and 9, respectively. Most, but by no means all,
ods are time consuming, however, and few ex- log k against volume fraction of organic solvent plots
perimental retention factors for conditions germane are curved when volume fractions of organic solvent
to solid-phase extraction are available in the litera- close to zero are included. A large contribution to
ture compared to the number of sorbents in general this curvature is the change in phase ratio associated
use, even for compounds of broad interest. It has with significant changes in the composition and
been more common to determine retention factors for structure of the stationary phase in contact with
the purpose of establishing the validity of general predominantly aqueous mobile phases. For individ-
models for solid-phase extraction than to advocate ual compounds linear, convex and concave plots are
their use as a general tool in method development. observed for the same sorbent and binary mobile
Methods that allow for the quick estimation of phase. Different sorbents or mobile phases often
retention factors are then of interest to establish produce different shaped plots for the same com-
convenient approaches for method development in pound. Therefore, generalizations for individual sol-
solid-phase extraction using models such as Eq. (7). utes, sorbents, etc., cannot be made. For intermediate
The general approach for estimating retention mobile phase compositions, an approximate linear
factors with sample solutions containing predomi-
nantly water is by extrapolation from retention
factors determined at more convenient mobile phase
compositions providing shorter separation times.
Extrapolations are based on either linear, Eq. (8), or
quadratic, Eq. (9), models:

log k 5 log k W 1 SfS (8)

log k 5 log k W 1 S1 fS 1 S2 f 2S (9)

where k W can be considered as an equation intercept

or identified as the retention factor for the analyte
with water as the mobile phase, fS the volume
fraction of organic solvent in a binary mixture of
water and organic solvent, and the S coefficients are
regression constants obtained by fitting the ex-
perimental data to the models. A comprehensive
discussion of the validity of Eqs. (8) and (9) would
be out of place here and only duplicate accounts
provided elsewhere [4,15,36,57,62]. It is frequently
assumed that log k W , the value of the retention factor
required for the estimation of sampling conditions
Fig. 8. Plot of the retention factor against percent volume fraction
with water as a sample solvent, is independent of the of acetonitrile in the mobile phase on a IST heavily loaded
mobile phase composition range and the relationship octadecylsiloxane-bonded silica sorbent for solid-phase extraction.
used for the extrapolation. This simple picture may Identification: (1) 2-phenylethanol and (2) benzonitrile.
30 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

solvent compositions provides better agreement but

some significant differences still exist. Note that the
curve fit results obtained using arbitrary functions
providing the best statistical fit to the experimental
data are superior to Eq. (9), which fails to provide a
good fit to the data in a significant number of cases.
It is unfortunate that no reliable method for de-
termining log k W by extrapolation exists, and the
available approaches discussed above, result in both
acceptable and poor estimates of log k W without a
method to distinguish into which category individual
results fall. Because extrapolation methods provide
unreliable estimates of log k W with large errors likely
in some cases, extrapolation methods cannot be used
to estimate log k W with any reasonable level of
confidence.
A substantial body of work exists on the estima-
tion of log k W values from octanol–water partition
coefficients [15,36,63–67]. Since these correlation
models have been formulated most commonly on
log k W values obtained by the linear extrapolation
Fig. 9. Plot of the retention factor against the volume fraction of method they possess all the deficiencies discussed
methanol in the mobile phase on a Bakerbond cyano- above for extrapolation methods combined with the
propylsiloxane-bonded silica sorbent for solid-phase extraction. added uncertainty that arises from the quality of the
Identification: (1) naphthalene, (2) bromobenzene; (3) fit of the correlation models. Individual models are
acetophenone; and (4) 2-phenylethanol.
needed for compounds lacking a family resemblance
suggesting that these models are not rigorous chemi-
relationship between log k and the volume fraction of cal models. Literature values for the octanol–water
organic solvent can almost always be found. The partition coefficient vary widely with the method of
intercept obtained by linear extrapolation, however, measurement or calculation adding to the uncertainty
is generally different to the value obtained by curve when they are used to estimate values of log k W
fitting the experimental data when measurements at a [35,68]. Hennion et al. [15] concluded that log KOW
low volume fraction of organic solvent are included. is of limited utility for the prediction of sample
Also, quite different slopes, and therefore intercepts, processing conditions for herbicides. Baltussen and
are possible when different segments of the com- co-workers [47,48] used log KOW to represent the
position range are used for the linear extrapolation unknown partition coefficient for pesticides and
(see Figs. 8 and 9). When values for log k W obtained polycyclic aromatic hydrocarbons between solid
by linear extrapolation, curve fitting, and experiment poly(dimethylsiloxane) particles and water. Eq. (4)
are compared there is often poor agreement between provided the model for the breakthrough volume.
the results (Table 3) [15,36]. The agreement usually Acceptable agreement between experimental and
improves when results for low volume fractions of model predicted values for the breakthrough volumes
water are included in the extrapolation but this rather was taken as confirmation that log KOW provided a
defeats the object of the extrapolation method since reasonable estimate of the poly(dimethylsiloxane)–
at low volume fractions of organic solvent sepa- water partition coefficient.
rations times are long. In general, the linear extrapo-
lation method provides a poor estimate of log k W for 4.3. Solvation parameter model
most compounds. Curve fitting the experimental
results for composition ranges including low organic The solvation parameter model provides a bridge
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 31

Table 3
Comparison of measured and extrapolated values for log k W for an octadecylsiloxane-bonded silica sorbent
Compound Log k W
Linear Eq. (8)a Quadratic Eq. (9)b Best fit c Experimental
Methanol–water:
2-Phenylethanol 2.00 2.36 2.36 2.45
4-Chlorophenol 2.52 2.50 2.45 2.73
4-Nitrobenzyl alcohol 1.58 2.15 2.43 2.40
Acetanilide 1.60 2.19 2.27 2.52
Acetophenone 2.26 2.81 2.89 3.01
Benzaldehyde 1.87 2.43 2.51 2.56
Hexan-2-one 1.91 2.49 2.57 2.62

Acetonitrile–water:
2-Phenylethanol 1.27 2.05 2.23 2.45
4-Chlorophenol 1.81 2.35 2.35 2.73
4-Nitrobenzyl alcohol 1.21 1.89 1.98 2.40
Acetanilide 1.03 1.92 2.12 2.52
Acetophenone 1.71 2.33 2.63 3.01
Benzaldehyde 1.66 1.94 2.41 2.56
Hexan-2-one 1.74 2.31 2.50 2.62

Tetrahdrofuran–water:
2-Phenylethanol 1.32 1.65 1.89 2.45
4-Chlorophenol 2.69 2.46 2.40 2.73
4-Nitrobenzyl alcohol 1.63 1.56 1.78 2.40
Acetanilide 1.04 1.39 1.60 2.52
Acetophenone 1.59 1.99 2.13 3.01
Benzaldehyde 1.56 1.74 1.77 2.56
a
From 40 to 70% (v / v) methanol and acetonitrile and 30 to 60% (v / v) tetrahydrofuran.
b
For the full data range from 1 to 100% (v / v) methanol and acetonitrile and 1 to 70% (v / v) tetrahydrofuran.
c
Arbitrary mathematical functions were used that gave the best statistical fit to the full range of data from 1 to 100% (v / v) methanol and
acetonitrile and 1 to 70% (v / v) tetrahydrofuran.

between compound structure and retention and can solute property (SP), in this case either the retention
be used to estimate retention factors or distribution factor (log k) or breakthrough, rinse and elution
constants in chromatography and other two-phase volumes (log V ) when log QVM in Eq. (7) is approxi-
distribution systems [36,69–78]. In solid-phase ex- mately constant.
traction it has been used to estimate breakthrough, The solute descriptors in Eq. (10) are McGowan’s
rinse and elution volumes [12,14,23– characteristic volume VX (in cm 3 mol 21 / 100), excess
25,45,50,52,53,59–61,79,80]. The appropriate form molar refraction R 2 (in cm 3 / 10), the solute’s dipo-
of the model for solid-phase extraction from a liquid larity / polarizability p H2 , and the solute’s effective
phase is given by Eq. (10): hydrogen-bond acidity and hydrogen-bond basicity,
H H H
Sa H2 and Sb H2 , respectively. McGowan’s charac-
log SP 5 c 1 mVX 1 rR 2 1 sp 2 1 aSa 2 1 bSb 2 teristic volume is an additive property and is easily
(10) calculated for any solute by summing the atomic
volumes for all atoms and subtracting a constant
The model equation is made up of product terms amount for each bond [69,81]. The solute’s excess
representing solute properties (descriptors) and sys- molar refraction is calculated from its refractive
tem properties characteristic of the sampling system. index and characteristic volume as the difference in
Each product term represents the contribution of a molar refraction of the solute and an n-alkane of
defined intermolecular interaction to the correlated identical volume [82]. The refractive index is almost
32 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

an additive property, and is easily estimated for any statistical and chemical validity of the model; cross-
compound of interest from fragmental constants. The correlation between individual descriptor values
solute dipolarity / polarizability parameter can be should be absent (R 2 , 0.8); and clustering of in-
obtained experimentally from gas–liquid chromato- dividual descriptor values should be avoided [78].
graphic data or water-solvent distribution constants For statistical soundness three varied values for each
[71,83,84]. The solute’s effective hydrogen-bond solute descriptor and the intercept are a reasonable
acidity and effective hydrogen-bond basicity were minimum number of data points to fit Eq. (10).
originally obtained directly from hydrogen-bond However, since each solute expresses several interac-
complexation constants in an inert solvent [85,86], tions simultaneously, the minimum number of sol-
but now that these scales are established, further utes required could be reduced from 18 to nine. Such
values can be obtained by gas–liquid chromatog- small data sets rarely provide valid models, however,
raphy or through the use of water–solvent distribu- because of the lack of compensation for the uneven
tion constants [71,84]. For distribution systems con- distribution of experimental error. It is, therefore,
taining water, some solutes (e.g., anilines, pyridines, both sensible and prudent to seek an exhaustive fit.
and sulfoxides) exhibit variable hydrogen-bond That is, a fit that shows little variation in the system
basicity. For these solutes two different descriptors, constants as small groups of randomly selected
Sb H2 and Sb 02 , are used, with selection of the most solutes are sequentially removed from the model.
appropriate descriptor based on the intended applica- This can usually be achieved using 20–30 varied
tion. For aqueous samples Sb 02 is the solute de- solutes.
scriptor usually chosen while Sb H2 is more appro- A plot of the system constants as a function of
priate for non-aqueous samples. Solute descriptors solvent composition provides a system map. System
are available for about 4000 compounds [72] with maps are the most useful approach for method
others available through parameter estimates and by development. A typical system map is shown in Fig.
computational approaches [69–71,87,88]. 10. The individual system constants usually change
The system constants in Eq. (10) are defined by smoothly with composition and can be fit to simple
their complementary interactions with the solute linear or polynomial functions for computer-aided
descriptors. The r constant determines the difference calculation of sampling properties. Once generated
in capacity of the sorbent and sample solution to the system maps are permanent and can be used to
interact with solute n- or p -electrons; the s constant estimate sample-processing conditions using Eqs. (7)
to the difference in capacity of the sorbent and and (10) for any analyte whose solute descriptors are
sample solution to take part in dipole–dipole and known or can be reasonably estimated.
dipole-induced dipole interactions; the a constant is a In general terms there are three regions of the
measure of the difference in hydrogen-bond basicity system map of interest for method development in
of the sorbent and the sample solution; the b constant solid-phase extraction. The left-hand side of the map,
is a measure of the difference in hydrogen-bond corresponding to low organic solvent, is the region
acidity of the sorbent and sample solution; and the m of interest for establishing a safe sampling volume.
constant is a measure of the relative ease of cavity From system maps for different sorbents the pre-
formation for the solute in the solvated sorbent and ferred sorbent for the isolation step can be identified.
sample solution together with contributions from The intermediate region of the system map is of
dispersion interactions that fail to cancel when the interest for selection of the rinse solvent. The right-
solute is transferred between phases. For any sam- hand side of the system map is the region of interest
pling system, the system constants can be obtained for selecting solvent compositions for elution of the
by multiple linear regression analysis of experimen- analytes from the sorbent by identifying conditions
tal retention properties acquired for a group of varied that minimize retention so that the sample can be
solutes with known descriptors. recovered in a small volume of solvent. System maps
The main criteria for solute selection to character- for a heavy (A.D. Gunatilleka, C.F. Poole, unpub-
ize a sampling system are that the solutes should be lished results) and light loaded [25] octa-
of sufficient number and variety to establish the decylsiloxane-bonded silica sorbent, for a butylsilox-
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 33

tures. Limited data are available for other sorbent

types, generally restricted to predominantly aqueous
organic solvents or water as the sample solvent.
Sorbent selection is based on identifying an appro-
priate sorbent that provides an adequate break-
through volume for the analytes of interest in the
solution to be processed. The required sample vol-
ume is established by the characteristics of the
determination step. The breakthrough volume should
be greater than the sample volume required to isolate
sufficient analyte for convenient detection, identifica-
tion and quantitation. The breakthrough volume can
be calculated from Eq. (7) when the parameters for
log QVM are known or confidently estimated, as is
generally the case, and only the retention factor need
be calculated using Eq. (10) [25]. The system
constants for a wide variety of sorbents used for
solid-phase extraction with water or water containing
1% (v / v) methanol as the sample solvent are summa-
rized in Table 4. Given that only system constants
with a positive sign contribute favorably to sorbent
retention we can obtain some general insight into the
Fig. 10. System map for a heavily loaded octadecylsiloxane-
bonded silica sorbent with methanol–water as the sample solvent.
factors responsible for retention under reversed-
phase sampling conditions. For all sorbents the
ane-bonded [60], cyanopropylsiloxane-bonded [59], cavity / dispersion term (m coefficient) is most im-
and spacer-bonded propanediol [53] silica sorbents, portant for promoting retention. All sorbents are at
and for a macroreticular porous polymer [50] are least as competitive as water for lone-pair electron
available for several aqueous–organic solvent mix- interactions (r50) and in some cases these interac-

Table 4
System constants for the estimation of solute retention in solid-phase extraction for aqueous samples (water or 1%, v / v, methanol in water)
Sorbent a System constants Refs.
m r s a b c
IST C 18 (HL) 4.39 0 0 20.79 21.90 20.27 –b
JTB C 18 (LL) 3.92 0 20.11 20.54 21.53 20.90 [25]
IST C 8 4.91 0 20.84 21.43 22.27 20.54 –b
IST CHX 3.94 0 20.32 20.83 21.98 20.31 –b
IST CH 3.22 0.35 0 20.92 21.61 20.37 –b
JTB C 4 3.36 0 0 20.46 21.53 21.38 [60]
JTB CN 2.06 0.53 0 20.51 21.45 20.88 [59]
JTB DIOL 1.57 0.61 0 20.45 20.80 21.05 [53]
PLRP-S 5.22 0.84 20.49 21.39 24.01 20.18 [50]
PGC 5.62 0 1.35 0 23.54 22.78 –c
AC 3.81 0 0.59 0 23.06 21.83 [80]
a
IST, International Sorbent Technology; JTB, J.T. Baker; PLRP-S, Polymer Laboratories (styrene–divinylbenzene porous polymer); PGC,
hypersil porous graphitic carbon; AC, activated charcoal; HL, high loading; LL, light loading; CHX, cyclohexanesiloxane-bonded; CH,
phenylsiloxane-bonded; CN, cyanopropylsiloxane-bonded; and DIOL, spacer-bonded propane diol.
b
A.D. Gunatilleka, C.F. Poole, unpublished results.
c
R. Sethuraman, C.F. Poole, unpublished results.
34 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

tions contribute favorably to retention (r is positive), combination of factors, one of which is the phase
if only in a minor way compared to the cavity / ratio for the sampling system when the retention
dispersion term. In general, polar interactions favor factor or breakthrough volume is used as the depen-
transport in the sample solution and result in smaller dent variable in Eq. (10).
breakthrough volumes. The exception is the carbon- To illustrate the role of sorbent selection on
based materials which have a significantly larger retention for aqueous samples a few representative
capacity for dipole-type interactions than water (s is calculations are summarized in Table 5. The impor-
positive). This is the distinguishing feature of car- tance of the cavity / dispersion contribution (mVX ),
bon-based sorbents compared to porous polymer and solvent hydrogen-bond acidity and solute hydrogen-
chemically bonded silica sorbents and combined with bond basicity (bSb ) H2 and the model constant term
a favorable cavity / dispersion term is the root of their (c) to retention when water is the sample solvent is
high retention and different selectivity to the other clearly indicated. In terms of maximizing retention
sorbents. For the chemically bonded sorbents dipole- the octadecylsiloxane-bonded silica sorbent (IST C 18
type interactions are of limited importance (s is HL), the porous polymer (PLRP-S) and the porous
either zero or small and negative). This is perhaps graphitic carbon (PGC) are superior to the other
surprising for the cyanopropylsiloxane-bonded sor- sorbents. Short-chain alkanesiloxane-bonded sor-
bents but is probably due to either selective solvation bents and chemically bonded sorbents with polar
of the cyano group by water molecules or strong functional groups are not as retentive as the three
intermolecular interactions between neighboring sorbents identified above for sampling aqueous solu-
cyano groups diminishing their availability for inter- tions. In terms of selectivity the porous polymer and
actions with analyte molecules. The cyano- porous graphitic carbon sorbent are preferred for
propylsiloxane-bonded and spacer-bonded pro- solutes with low hydrogen-bond basicity. Porous
panediol sorbents have small values for the cavity / graphitic carbon is particularly useful for the isola-
dispersion term compared to other chemically tion of large solutes and solutes with significant
bonded sorbents indicating greater cohesion and a dipolar character and low hydrogen-bond basicity
lower retention capacity. The greater cohesion of (e.g., polycyclic aromatic compounds, steroids, etc.)
these sorbents easily eclipses their capacity for Porous graphitic carbon is limited for the isolation of
selective polar interactions with the result that they small molecules, unless they are considerably dipo-
are less effective for the isolation of polar analytes lar, such as acetanilide, because it has an unfavorable
from water than the alkylsiloxane-bonded sorbents. model constant (c term) which opposes the favorable
None of the sorbents are competitive with water for cavity / dispersion contribution to retention. Thus
hydrogen-bond interactions except for carbon-based porous graphitic carbon would be a good choice for
sorbents which are similar in their hydrogen-bond the isolation of acetanilide but not the best choice for
basicity (a50). The general difficulty in isolating the isolation of phenol. In fact, none of the sorbents
hydrogen-bond acid solutes arises because the b in Table 5 are particularly useful for the isolation of
constant for all sorbents is numerically quite large phenol due to its small size and capacity for hydro-
and negative. The system properties most important gen-bond interactions.
for controlling retention are the high cohesive energy The above discussion has been limited to the
and hydrogen-bond acidity of water for which the isolation of solutes from water by solid-phase ex-
solvated sorbents compete to varying extents but fail traction. There are several additional factors to note.
to dominate. Ironically, these properties are in oppo- The solvation parameter model is strictly applicable
sition with respect to sorbent retention. The high to neutral compounds and ionizable compounds in
cohesive energy of water promotes retention while their neutral form. Ionization tends to reduce re-
the hydrogen-bond acidity of water is the principle tention compared to the neutral form of the solute in
reason for low retention of hydrogen-bond bases. solid-phase extraction from aqueous solution. Re-
The equation constant (c term) is not related to cently, the solvation parameter model has been
fundamental properties of the analytes but is clearly extended to the prediction of the retention of phenols
important in establishing retention. It is a complex in various states of ionization in liquid chromatog-
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 35

Table 5
Contribution of intermolecular interactions to retention of some varied solutes in solid-phase extraction with water (or water containing 1%,
v / v, methanol) as the sample solvent
Compound Sorbent Contribution to log k W Predicted
mVX rR 2 sp H2 aSa H2 bSb H2 c log k W

n-Propylbenzene IST C 18 (HL) 5.00 20.29 20.27 4.40

JTB C 18 (LL) 4.46 20.06 20.23 20.90 3.27
IST C 8 5.59 20.42 20.34 20.54 4.29
IST CHX 4.49 20.16 20.30 20.31 3.72
IST CH 3.67 0.21 20.24 20.37 3.27
JTB C 4 3.82 20.23 21.38 2.22
JTB CN 2.20 0.33 20.22 20.88 1.44
JTB DIOL 1.79 0.37 20.12 21.05 0.99
PLRP-S 5.95 0.51 20.25 20.60 20.18 5.43
PGC 6.40 0.68 20.53 22.78 3.77
AC 4.34 0.30 20.46 21.83 2.35

Benzonitrile IST C 18 (HL) 3.82 20.63 20.27 2.92

JTB C 18 (LL) 3.41 20.12 20.50 20.90 1.89
IST C 8 4.27 20.93 20.75 20.54 2.05
IST CHX 3.43 20.36 20.65 20.31 2.11
IST CH 2.80 0.26 20.53 20.37 2.16
JTB C 4 2.93 20.51 21.38 1.04
JTB CN 1.68 0.40 20.49 20.88 0.71
JTB DIOL 1.37 0.45 20.26 21.05 0.51
PLRP-S 4.55 0.62 20.54 21.32 20.18 3.13
PGC 4.90 1.50 21.17 22.78 2.45
AC 3.32 0.65 21.10 21.83 1.04

Acetophenone IST C 18 (HL) 4.45 20.91 20.27 3.27

JTB C 18 (LL) 3.97 20.11 20.73 20.90 2.23
IST C 8 4.98 20.85 21.09 20.54 2.50
IST CHX 4.00 20.32 20.95 20.31 2.42
IST CH 3.27 0.29 20.77 20.37 2.42
JTB C 4 3.41 20.73 21.38 1.30
JTB CN 2.09 0.43 20.70 20.88 0.94
JTB DIOL 1.59 0.50 20.38 21.05 0.66
PLRP-S 5.29 0.69 20.49 21.92 20.18 3.39
PGC 5.70 1.36 21.70 22.78 2.58
AC 3.86 0.60 20.88 21.83 1.75

Acetanilide IST C 18 (HL) 4.89 20.40 21.27 20.27 2.95

JTB C 18 (LL) 4.37 20.15 20.27 21.03 20.90 2.02
IST C 8 5.46 21.17 20.71 21.52 20.54 1.52
IST CHX 4.39 20.45 20.42 21.33 20.31 1.88
IST CH 3.58 0.30 20.46 21.08 20.37 1.97
JTB C 4 3.74 20.23 21.03 21.38 1.11
JTB CN 2.15 0.47 20.26 20.98 20.88 0.50
JTB DIOL 1.75 0.53 20.23 20.54 21.05 0.47
PLRP-S 5.81 0.73 20.69 20.70 22.69 20.18 2.28
PGC 6.26 1.89 22.37 22.78 3.00
AC 4.24 0.83 22.05 21.83 1.19

Phenol IST C 18 (HL) 3.40 20.47 20.57 20.27 2.09

JTB C 18 (LL) 3.04 20.10 20.32 20.47 20.90 1.25
IST C 8 3.80 20.74 20.86 20.68 20.54 0.98
IST CHX 3.05 20.28 20.50 20.59 20.31 1.37
IST CH 2.50 0.28 20.55 20.48 20.37 1.38
JTB C 4 2.60 20.28 20.47 21.38 0.47
JTB CN 1.49 0.43 20.31 20.44 20.88 0.29
JTB DIOL 1.22 0.49 20.27 20.24 21.05 0.15
PLRP-S 4.05 0.68 20.44 20.83 21.24 20.18 2.04
PGC 4.36 1.20 21.06 22.78 1.72
AC 2.95 0.53 20.92 21.83 0.73
36 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

raphy by use of an additional solute descriptor in Eq. maps were used to determine breakthrough volumes
(10) [89,90]. This approach has not been applied to for sample processing, the composition and volume
solid-phase extraction but there is no reason to of rinse solvents for matrix simplification, and the
believe that it would not be equally applicable. composition and volume of elution solvents. The plot
Silica-based sorbents contain a low concentration of of sample solvent composition against log V for three
silanol groups with sufficient ion-exchange capacity estrogens on an octadecylsiloxane-bonded silica sor-
to adsorb basic solutes [91–94]. The properties of bent is shown in Fig. 11. A sample volume of 45 ml
these sites are not included in the solvation parame- of urine was required to provide a sufficient amount
ter model and are a potential source of disagreement of estrogens for analysis. Consequently, a minimum
between experiment and model predictions for easily breakthrough volume for estriol of 45 ml, the least
protonated solutes. A general cause of low recovery retained of the estrogens, is required of the sampling
of these compounds is not premature breakthrough device. As shown in Fig. 11 this can be achieved at
but strong retention that prevents high recovery of any sample solvent composition containing less than
the solutes by elution with an organic solvent. This 25% (v / v) methanol in water. A rinse solvent
problem can sometimes be circumvented by addition volume of 6 ml was selected for matrix simplifica-
of a competing base to the elution solvent. It is tion. From Fig. 11 a rinse solvent containing up to
common practice to use binary solvent mixtures for 40% (v / v) methanol can be used without break-
method development in solid-phase extraction. Re- through of estriol. The most retained of the estrogens
tention in ternary solvent mixtures can be predicted is estrone and the solvent composition required for
using the solvation parameter model and a mixture elution of all estrogens can be predicted from the
design approach to define the system surfaces ob- retention properties of this compound. Pure methanol
tained [77,95].
System maps with methanol–water mixtures are
available for particle-loaded membranes [14,61] and
particle-embedded glass fiber discs [12,14] contain-
ing octadecylsiloxane-bonded silica sorbents. Par-
ticle-loaded membranes containing porous polymer
sorbent particles are more retentive for most solutes
than those containing octadecylsiloxane-bonded sor-
bent particles [61]. Solvent effects are very signifi-
cant for the porous polymer containing membranes,
which are capable of absorbing organic solvent from
the sample solution changing their selectivity and
phase ratio [24]. One percent organic solvent is used
to increase the sample processing rate with particle-
loaded membranes and the selective uptake of the
organic solvent by the porous polymer particles
resulted in differences in breakthrough volumes
exceeding an order of magnitude depending on the
identity of the solute and organic solvent. The
solvation parameter model provided the necessary
theoretical framework to enable the most favored
sample processing solvent to be identified for a
particular application.
Fig. 11. A plot of methanol–water composition against log V (Eq.
As a demonstration experiment Seibert and Poole
7) for estriol (A), 17b-estradiol (B), and estrone (C) on an
[25] used Eq. (7) and the solvation parameter model, octadecylsiloxane-bonded silica sorbent. VB indicates the required
Eq. (10), to predict the sample processing conditions breakthrough volume and Vrinse the composition of the rinse
for the isolation of estrogens from urine. System solvent for matrix simplification.
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 37

was selected as the elution solvent to minimize water a significant capacity for dipole-type and hydrogen-
contamination of the extract, which interfered in the bond interactions are expected to exhibit different
derivatization step required for gas chromatography. solvation properties towards wet octanol and the
It was estimated that about 4 column void volumes sorbents indicated in Table 6. The general case is
were required for quantitative recovery of estrone. that log KOW is not expected to provide a reasonable
The good agreement between model predictions and model for the estimation of sampling conditions for
experiment was taken as confirmation of the success the common sorbents used in solid-phase extraction.
of the approach. The authors note that the simple System maps and single solvent composition
theoretical framework provides a significant step models for normal-phase sampling conditions are
forward towards computer-aided method develop- few in number [53,79,96–98]. The driving force for
ment in solid-phase extraction. retention under normal-phase conditions is the
The solvation parameter model provides a general capacity of the solvated sorbent for polar interac-
framework to explain why the octanol–water dis- tions; in contrast to extraction from water the cavity /
tribution constant provides a poor tool for the dispersion term (m constant) is generally unimpor-
prediction of sample processing conditions in solid- tant or negative. As would be expected, the main
phase extraction of aqueous samples. The system contribution to retention on aminopropylsiloxane-
constants for the octanol–water partition coefficient bonded phases is the high basicity of the solvated
(log KOW ) are well established [36,71]. To construct sorbent (a constant). For the cyanopropylsiloxane-
a correlation model so that log KOW can be used to bonded sorbent it is the capacity of the solvated
estimate log V or log k the minimum requirement is sorbent for dipole-type interactions (s constant). For
that the ratio of the system constants for the systems the spacer-bonded propanediol sorbent it is the
to be correlated are (nearly) identical. The system capacity for hydrogen-bond interactions, particularly
constant ratios for the octanol–water partition coeffi- as a hydrogen-bond acid (b constant), combined with
cient and sorbents for solid-phase extraction are a significant capacity for dipole-type interactions (s
summarized in Table 6. For all 11 sorbents there is constant) that are important. In effect, the reasons for
only a poor match between the solvation properties choosing a particular sorbent for extraction from an
of wet octanol and the sorption properties of the organic solvent are completely different to those for
solvated sorbents. The only likely correlation is for water with most of this difference attributed to the
solutes with minimum polarity since all solutes with unique properties of water. Although the solvation
parameter model can model retention on chemically
Table 6 bonded sorbents using organic solvents it is possible
System constant ratios for the octanol–water partition coefficient that the model does not correctly account for differ-
and solid-phase extraction sorbents with water or 1% (v / v) ences in solute size and sorbent site-specific interac-
methanol in water as the sample solvent tions when silica gel is the sorbent [96].
System System constant ratios (solvation parameter
model)
r /m s /m a /m b /m 5. Other approaches
Octanol–water 0.147 20.276 0.08 20.908
IST C 18 (HL) 0 0 20.18 20.43 Hughes and Gunton [99] proposed a graphical
JTB C 18 (LL) 0 20.03 20.14 20.39
method to estimate the volume and solvent strength
IST C 8 0 20.17 20.29 20.46
IST CHX 0 20.08 20.21 20.50 of rinse and elution solvents in solid-phase extraction
IST CH 0.11 0 20.29 20.50 based on an extension of the classical theory of
JTB C 4 0 0 20.14 20.46 liquid–liquid extraction. This leads to the equation:
JTB CN 0.26 0 20.25 20.70
JTB DIOL 0.39 0 20.29 20.51 ln(1 2 R T ) 5 n ln(1 2 R 0 ) 1 ln R RET (11)
PLRP-S 0.16 20.09 20.27 20.77
PGC 0 0.24 0 20.63 where R T is the total extraction recovery, n the
AC 0 0.15 0 20.80
number of extractions, R 0 the extraction efficiency,
38 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39

and R RET the limiting extraction recovery used to [3] J.S. Fritz, Analytical Solid-Phase Extraction, Wiley, New
York, 1999.
account for irreversible solute retention by the
[4] N.J.K. Simpson (Ed.), Solid-Phase Extraction: Principles,
sorbent. A plot of 2ln(1 2 R T ) against n (or the Strategies, and Applications, Marcel Dekker, New York,
equivalent volume of rinse or elution solvent) yields 2000.
a straight line from which the extraction efficiency [5] N. Masque, R.M. Marce, F. Borrull, Trends Anal. Chem. 17
and the R RET factor are calculated from the slope and (1998) 384.
[6] D. Stevenson, Trends Anal. Chem. 18 (1999) 154.
intercept, respectively. Within this model favorable [7] V. Pinchon, M. Bouzige, C. Miege, M.-C. Hennion, Trends
conditions for matrix simplification are recognized as Anal. Chem. 18 (1999) 219.
those in which the extraction efficiency (slope of the [8] A.J.H. Louter, J.J. Vreuls, U.A.Th. Brinkman, J. Chromatogr.
plot) is low for the analytes and high for the matrix A 842 (1999) 391.
[9] E.R. Brouwer, S. Kofman, U.A.Th. Brinkman, J. Chroma-
in the rinse step and high for the analytes in the togr. A 703 (1995) 167.
elution step. In practice the plots are often non-linear [10] D.F. Hagen, C.G. Markell, G.A. Schmitt, D.B. Blevins, Anal.
but this does not prevent a qualitative assessment of Chim. Acta 236 (1990) 157.
the recovery and matrix simplification procedure [11] H. Lingeman, S.J.F. Hoekstra-Oussoren, J. Chromatogr. A
689 (1997) 221.
being graphically evaluated from the experimentally
[12] M.L. Mayer, C.F. Poole, M.P. Henry, J. Chromatogr. A 695
derived curves. (1995) 267.
[13] M.L. Mayer, C.F. Poole, Anal. Chim. Acta 294 (1994) 113.
[14] C.F. Poole, S.K. Poole, D.S. Seibert, C.M. Chapman, J.
Chromatogr. B 689 (1997) 245.
6. Conclusions [15] M.-C. Hennion, C. Cau-Dit-Coumes, V. Pichon, J. Chroma-
togr. A 823 (1998) 147.
Steady progress has been made towards the de- [16] O.V. Rodinkov, L.N. Moskvin, J. Anal. Chem. 54 (1999)
velopment of computer-aided method development 130.
strategies for solid-phase extraction as a replacement [17] L.O.G. Weidolf, J.D. Henion, Anal. Chem. 59 (1987) 1980.
[18] M. Casas, L.A. Berrueta, B. Gallo, F. Vicente, Chromato-
for traditional trial and error approaches. Contempor- graphia 34 (1992) 79.
ary models based on frontal analysis to accommodate [19] P. Lovkvist, J.A. Jonsson, Anal. Chem. 59 (1987) 818.
the contribution of kinetic factors to retention re- [20] I. Liska, J. Chromatogr. A 655 (1993) 163.
sulting from the small number of theoretical plates [21] M.-C. Hennion, V. Pichon, Environ. Sci. Technol. 28 (1994)
provided by sorbent sampling devices and the solva- 576A.
[22] I. Liska, A. Kuthan, J. Krupcik, J. Chromatogr. 509 (1990)
tion parameter model for the prediction of sorbent 123.
retention are the most promising. Current evolution- [23] M.L. Larrivee, C.F. Poole, Anal. Chem. 66 (1994) 139.
ary development of the solvation parameter model [24] S.K. Poole, C.F. Poole, Analyst 120 (1995) 1733.
towards computational approaches to the estimation [25] D.S. Seibert, C.F. Poole, J. High Resolut. Chromatogr. 21
of solute descriptors from compound structure would (1998) 481.
[26] N. Masque, M. Galio, R.M. Marce, F. Borrull, J. Chroma-
further enhance its use in method development for togr. A 771 (1997) 55.
solid-phase extraction. At the same time estimation [27] V. Pichon, M.-C. Hennion, J. Chromatogr. 665 (1994) 269.
methods based on octanol–water partition coefficient [28] V. Pichon, L. Chen, S. Guenu, M.-C. Hennion, J. Chroma-
and extrapolated retention factors (log k W ) have to togr. A 711 (1995) 257.
be considered more circumspect as lacking a sound [29] P. Subra, M.-C. Hennion, R. Rosset, R.W. Frei, J. Chroma-
togr. 456 (1988) 121.
theoretical basis as well as associated experimental [30] R. Ferrer, J.L. Beltran, J. Guiteras, Anal. Chim. Acta 346
uncertainties. (1997) 253.
[31] E.M. Thurman, R.L. Malcom, G.R. Aiken, Anal. Chem. 50
(1978) 775.
[32] C.M. Josefson, J.B. Johnston, R. Tubey, Anal. Chem. 56
References (1984) 764.
[33] J. Przyazny, J. Chromatogr. 346 (1985) 61.
[1] E.M. Thurman, M.S. Mills, Solid-Phase Extraction. Princi- [34] M.H. Abraham, J. Le, J. Pharm. Sci. 88 (1999) 868.
ples and Practice, Wiley, New York, 1998. [35] C. Hansch, A. Leo, D. Hoekman, in: Exploring QSAR:
[2] J.R. Dean, Extraction Methods for Environmental Analysis, Hydrophobic, Electronic and Steric Constraints, Vols. 1 and
Wiley, Chichester, 1998. 2, American Chemical Society, Washington, DC, 1995.
 C.F. Poole et al. / J. Chromatogr. A 885 (2000) 17 – 39 39

[36] C.F. Poole, S.K. Poole, A.D. Gunatilleka, Adv. Chromatogr. [69] M.H. Abraham, Chem. Soc. Rev. 22 (1993) 73.
40 (2000) 159. [70] M.H. Abraham, in: P. Politzer, J.S. Murray (Eds.), Quantita-
[37] M. Nakamura, M. Nakamura, S. Yamada, Analyst 121 tive Treatments of Solute / Solvent Interactions, Elsevier,
(1996) 469. Amsterdam, 1994, pp. 83–134.
[38] C.E. Werkhoven-Goewie, U.A.Th. Brinkman, R.W. Frei, [71] M.H. Abraham, H.S. Chadha, in: V. Liska, B. Testa, H. van
Anal. Chem. 53 (1981) 2072. de Waterbeemed (Eds.), Lipophilicity in Drug Action and
[39] H.G.J. Mol, J. Staniewski, H.-G. Jansson, C.A. Cramers, Technology, VCH, Weinheim, 1996, pp. 311–337.
R.T. Ghijsen, U.A.Th. Brinkman, J. Chromatogr. 630 (1993) [72] M.H. Abraham, C.F. Poole, S.K. Poole, J. Chromatogr. A
201. 842 (1999) 79.
[40] H.G.J. Mol, H.-G. Janssen, C.A. Cramers, J. High Resolut. [73] A.D. Gunatilleka, C.F. Poole, Anal. Commun. 36 (1999)
Chromatogr. 16 (1993) 413. 235.
[41] M.-C. Hennion, V. Coquart, J. Chromatogr. 642 (1993) 211. [74] S.K. Poole, C.F. Poole, J. Chromatogr. 845 (1999) 381.
[42] S. Guenu, M.-C. Hennion, J. Chromatogr. A 725 (1996) 57.
[75] W. Kiridena, C.F. Poole, J. Planar Chromatogr. 12 (1999) 13.
[43] A. Gelencser, G. Kiss, Z. Krivacsy, Z. Varga-Puchony, J.
[76] T.M. Pawlowski, C.F. Poole, Anal. Commun. 36 (1999) 71.
Hlavay, J. Chromatogr. A 693 (1995) 217.
[77] W. Kiridena, C.F. Poole, Chromatographia 48 (1998) 607.
[44] A. Gelencser, G. Kiss, Z. Krivacsy, Z. Varga-Puchony, J.
Hlavay, J. Chromatogr. A 693 (1995) 227. [78] C.F. Poole, S.K. Poole, M.H. Abraham, J. Chromatogr. A
[45] K.G. Miller, C.F. Poole, J. High Resolut. Chromatogr. 17 798 (1998) 207.
(1994) 125. [79] S.K. Poole, C.F. Poole, Anal. Commun. 33 (1996) 353.
[46] S. Bitteur, R. Rosset, Chromatographia 23 (1987) 163. [80] S.K. Poole, C.F. Poole, Anal. Commun. 34 (1997) 247.
[47] E. Baltussen, F. David, P. Sandra, H.-G. Janssen, C.A. [81] M.H. Abraham, J.C. McGowan, Chromatographia 23 (1987)
Cramers, J. Chromatogr. A 805 (1998) 237. 243.
[48] E. Baltussen, H. Snijders, H.-G. Janssen, P. Sandra, C.A. [82] M.H. Abraham, G.S. Whiting, R.M. Doherty, W.J. Shuely, J.
Cramers, J. Chromatogr. A 802 (1998) 285. Chem. Soc. Perkin Trans. 2 (1990) 1451.
[49] W.P.N. Fernando, M.L. Larrivee, C.F. Poole, Anal. Chem. 65 [83] M.H. Abraham, J. Chromatogr. 644 (1993) 95.
(1993) 588. [84] M.H. Abraham, J. Phys. Org. Chem. 6 (1993) 660.
[50] D. Bolliet, C.F. Poole, Chromatographia 46 (1997) 381. [85] M.H. Abraham, P.L. Grellier, D.V. Prior, J.J. Morris, P.J.
[51] C.F. Poole, S.K. Poole, in: Chromatography Today, Elsevier, Taylor, J. Chem. Soc. Perkin Trans 2 (1989) 699.
Amsterdam, 1991. [86] M.H. Abraham, P.L. Grellier, D.V. Prior, J.J. Morris, P.J.
[52] D.S. Seibert, C.F. Poole, J. High Resolut. Chromatogr. 18 Taylor, J. Chem. Soc. Perkin Trans. 2 (1990) 521.
(1995) 226. [87] C.M. Du, K. Valko, C. Bevan, D. Reynolds, M.H. Abraham,
[53] D.S. Seibert, C.F. Poole, M.H. Abraham, Analyst 121 (1996) Anal. Chem. 70 (1998) 4228.
511.
[88] M. Plass, K. Valko, M.H. Abraham, J. Chromatogr. A 803
[54] I. Johannes, L. Molder, L. Tiikma, Oil Shale 14 (1997) 41.
(1998) 51.
[55] R.E. Shoup, G.S. Mayer, Anal. Chem. 54 (1982) 1164.
[89] D. Bolliet, C.F. Poole, M. Roses, Anal. Chim. Acta 368
[56] J.W. Carr, J.M. Harris, Anal. Chem. 60 (1988) 698.
(1998) 129.
[57] P. Jandra, J. Kubat, J. Chromatogr. 500 (1990) 281.
[90] M. Roses, D. Bolliet, C.F. Poole, J. Chromatogr. A 829
[58] C.E. Green, M.H. Abraham, J. Chromatogr. A 885 (2000)
(1998) 29.
41.
[91] P. Martin, A. Taberner, A. Fairbrother, I.D. Wilson, J. Pharm.
[59] D.S. Seibert, C.F. Poole, Chromatographia 41 (1995) 51.
Biomed. Anal. 11 (1993) 671.
[60] D.S. Seibert, C.F. Poole, Anal. Commun. 35 (1998) 147.
[61] M.L. Mayer, S.K. Poole, C.F. Poole, J. Chromatogr. A 697 [92] P. Martin, E.D. Morgan, I.D. Wilson, Anal. Proc. 32 (1995)
(1995) 89. 179.
[62] L.C. Tran, P.W. Carr, J. Chromatogr. A 656 (1993) 521. [93] K. Albert, R. Brindle, P. Martin, I.D. Wilson, J. Chromatogr.
[63] W.J. Lambert, J. Chromatogr. A 656 (1993) 469. A 665 (1994) 253.
[64] H. van de Waterbeemed, M. Kansy, B. Wagner, H. Fischer, [94] B. Law, S. Weir, N.A. Ward, J. Pharm. Biomed. Anal. 10
Methods Prin. Med. Chem. 4 (1996) 73. (1992) 167.
[65] M.-M. Hsieh, J.G. Dorsey, J. Chromatogr. 631 (1993) 63. [95] D. Bolliet, C.F. Poole, Anal. Commun. 35 (1998) 253.
[66] P. Vallat, W. Fan, N. El Tayar, P.-A. Carrupt, B. Testa, J. Liq. [96] W. Kiridena, C.F. Poole, Analyst 123 (1998) 1265.
Chromatogr. 15 (1992) 2133. [97] J.H. Park, M.H. Yoon, Y.K. Ryu, B.E. Kim, J.W. Ryu, M.D.
[67] Y. Zhang, W. Shi, L. Zhang, H. Zou, J. Chromatogr. A 802 Jang, J. Chromatogr. A 796 (1998) 249.
(1998) 59. [98] F.Z. Oumada, M. Roses, E. Bosch, M.H. Abraham, Anal.
[68] L.G. Danielsson, Y.-H. Zhang, Trends Anal. Chem. 15 Chim. Acta 382 (1999) 301.
(1996) 188. [99] D.M. Hughes, K.E. Gunton, Anal. Chem. 67 (1995) 1191.