Fundamental and

Regulatory Aspects of
UHPLC in Pharmaceutical
Analysis
Dennis Åsberg

Faculty of Health, Science and Technology

Chemistry

DOCTORAL THESIS | Karlstad University Studies | 2017:9

Fundamental and
Regulatory Aspects
of UHPLC in
Pharmaceutical Analysis
Dennis Åsberg

DOCTORAL THESIS | Karlstad University Studies | 2017:9

Fundamental and Regulatory Aspects of UHPLC in Pharmaceutical Analysis

Dennis Åsberg

DOCTORAL THESIS

Karlstad University Studies | 2017:9

urn:nbn:se:kau:diva-47852

ISSN 1403-8099

ISBN 978-91-7063-756-8 (print)

ISBN 978-91-7063-757-5 (pdf)

©
The author

Distribution:
Karlstad University
Faculty of Health, Science and Technology
Department of Engineering and Chemical Sciences
SE-651 88 Karlstad, Sweden
+46 54 700 10 00

Print: Universitetstryckeriet, Karlstad 2017

WWW.KAU.SE
“Data! Data! Data!” he cried impatiently. “I can’t make bricks without clay.”
— Arthur Conan Doyle in The Adventures of Sherlock Holmes

iii
Abstract

Ultra-high performance liquid chromatography (UHPLC) provides a consi-

derable increase in throughput compared to HPLC and a reduced solvent
consumption. The implementation of UHPLC in pharmaceutical analysis,
e.g. quality control, has accelerated in recent years and there is currently a
mix of HPLC and UHPLC instrumentation within pharmaceutical compa-
nies. There are, however, technical and regulatory challenges converting a
HPLC method to UHPLC making it difficult to take full advantage of UHPLC
in regulatory-focused applications like quality control in pharmaceutical
production.
Using chromatographic modelling and fundamental theory, this thesis
investigated method conversion between HPLC and UHPLC. It reports on
the influence of temperature gradients due to viscous heating, pressure
effects and stationary phase properties on the separation performance. It
also presents a regulatory concept for less regulatory interaction for minor
changes to approved methods to support efficient life cycle management.
The higher pressure in UHPLC gave a retention increase of up to 40%
as compared to conventional HPLC while viscous heating, instead, reduced
retention and the net result was very solute dependent. Selectivity shifts
were observed even between solutes with similar structure when switching
between HPLC and UHPLC and an experimental method to predict such
selectivity shifts was therefore developed. The peak shape was negatively
affected by the increase in pressure for some solutes since secondary
interactions between the solute and the stationary phase increased with
pressure.
With the upcoming ICH Q12 guideline, it will be possible for the
industry to convert existing methods to UHPLC in a more flexible way
using the deeper understanding and the regulatory concept presented here
as a case example.

iv
Swedish Summary

De flesta läkemedel är molekyler som framställs på kemisk väg genom

en rad delprocesser. Under framställningen bildas olika biprodukter med
snarlik struktur vilka ofta har en okänd fysiologisk effekt. Därför görs alltid
en kvalitetskontroll av läkemedlets renhet där mängden biprodukter mäts.
För att analysera vad ett läkemedel innehåller så är vätskekromatografi
den vanligaste tekniken och läkemedelstillverkare har hundratals sådana
instrument för att kunna analysera alla läkemedel de tillverkar. I vätskekro-
matografi löser man först upp provet, t.ex. en tablett, i ett lösningsmedel
och sedan injiceras det i ett vätskeflöde som pumpas genom ett poröst
material packat i ett rör som kallas kolonn. Provets komponenter fastnar
olika hårt på materialet i kolonnen vilket medför att det tar olika långt tid att
skölja ur dem och de kommer därmed ut ett och ett, dvs. de har separerats.
Allteftersom komponenterna kommer ut ur kolonnen detekteras de och
varje ämne bildar en topp i ett kromatogram.

v
 Jag har arbetat tillsammans med AstraZeneca för att bättre förstå hur
provkomponenterna separeras och med att utveckla effektivare kromato-
grafiska analysmetoder. Nyckeln till att designa en effektiv analysmetod
är att förstå hur de olika komponenterna fastnar, adsorberar, på de porösa
partiklarna som kolonnen är packad med. Jag har utvärderat och utvecklat
fysikaliska modeller som med hjälp av datorsimuleringar kan förutsäga
separationen. Resultaten jämfördes sedan med riktiga experiment för att
utvärdera hur bra modellen stämde överens med verkligheten.
Under de senaste åren har det varit en trend mot att packa kolonnen
med mindre och mindre partiklar eftersom topparna då blir skarpare och
analysen kan göras snabbare. De nyaste instrumenten kan utföra analyser
upp till 10 gånger snabbare och förbrukar mycket mindre lösningsmedel
vilket gör dem väldigt attraktiva för läkemedelsindustrin. Det finns dock
både tekniska och regulatoriska utmaningar med att byta ut de gamla
instrumenten. Att pumpa vätska genom en kolonn packad med så små
partiklar kräver ett högt tryck (ca 1 000 bar) och det höga trycket påverkar
hur ämnena adsorberar. Dessutom bildas det värme i kolonnen på grund
av friktionen mellan vätska och partiklar vilket ger en ojämn temperatur
i kolonnen som, tillsammans med trycket, påverkar separationen. En stor
del av avhandlingen handlar om att förstå och förutsäga hur separationen
ändras när analysmetoden överförs till ett instrument med små partiklar. Att
på ett förutsägbart och robust sätt överföra metoder mellan olika instrument
är viktigt då industrin har en blandning av nya och gamla instrument där
metodutvecklingen ofta sker på nya instrument och rutinanalys på äldre.
Läkemedelstillverkning är dessutom hårt reglerad av myndigheter och
en analysmetod som används för ett marknadsfört läkemedel får inte ändras
utan myndigheternas godkännande. Då ett läkemedel oftast är marknadsfört
i en mängd länder är det många myndigheter som ska granska och godkänna
ändringar och därför vill man från industrins sida undvika ändringar om
man kan. Det här har medfört att läkemedel i vissa fall analyseras med
10-20 år gammal utrustning. Under 2017 ska nya riktlinjer släppas för EU,
USA och Japan som tillåter viss flexibilitet, t.ex. möjliggör att ändra en
kromatografisk metod så att de nyaste instrumenten kan användas. Vi
har tagit fram ett förslag på hur metoder kan utvecklas och ändras i linje
med de nya riktlinjerna och demonstrerat det med en kvalitetskontroll
för läkemedlet Nexium. Min förhoppning är att den ökade flexibiliteten
tillsammans med den djupare förståelse för separationen som presenteras i
avhandlingen, ska underlätta för industrin att implementera ny teknik som
kan effektivisera kvalitetskontrollen av läkemedel.

vi
List of Papers

This thesis is based on the following papers, hereby referred to by their

roman numerals, and reprints are appended at the end of the thesis with
permission from Elsevier.

I Combining Chemometric Models with Adsorption Isotherm Mea-

surements to Study Omeprazole in RP-LC. D. Åsberg, M. Leśko, J.
Samuelsson, A. Karlsson, K. Kaczmarski and T. Fornstedt. Chromato-
graphia, 79, 1283-1291 (2016)

II Method Transfer from High-Pressure Liquid Chromatography to

Ultra-High-Pressure Liquid Chromatography. I. A Thermodyna-
mic Perspective. D. Åsberg, M. Leśko, J. Samuelsson, K. Kaczmarski
and T. Fornstedt. Journal of Chromatography A, 1362, 206–217 (2014)

III Method Transfer from High-Pressure Liquid Chromatography to

Ultra-High-Pressure Liquid Chromatography. II. Temperature and
Pressure Effects. D. Åsberg, J. Samuelsson, M. Leśko, A. Cavazzini,
K. Kaczmarski and T. Fornstedt. Journal of Chromatography A, 1401,
52–59 (2015)

IV A Quality Control Method Enhancement Concept—Continual Im-

provement of Regulatory Approved QC Methods. D. Åsberg, M.
Nilsson, S. Olsson, J. Samuelsson, O. Svensson, S. Klick, J. Ennis, P.
Butterworth, D. Watt, S. Iliadau, A. Karlsson, J. T. Walker, K. Arnot, N.
Ealer, K. Hernqvist, K. Svensson, A. Grinell, P.-O. Quist, A. Karlsson
and T. Fornstedt. Journal of Pharmaceutical and Biomedical Analysis, 129,
273–281 (2016)

vii
 V A Practical Approach for Predicting Retention Time Shifts due
to Pressure and Temperature Gradients in Ultra-High-Pressure Li-
quid Chromatography. D. Åsberg, M. Chutkowski, M. Leśko, J.
Samuelsson, K. Kaczmarski and T. Fornstedt. Journal of Chromato-
graphy A, 1479, 107–120 (2017)

VI A Fundamental Study of the Impact of Pressure on the Adsorption

Mechanism in Reversed-Phase Liquid Chromatography. D. Åsberg,
J. Samuelsson and T. Fornstedt. Journal of Chromatography A, 1457,
97–106 (2016)

My contributions to the papers included in the thesis were:

I: I did the planning, the experiments, the DoE calculations and wrote the
manuscript. II: I did the planning, the experiments, the PP/FA/AED calcu-
lations and wrote the manuscript. III: I did the planning, the experiments
and wrote the manuscript together with my coauthors. IV: I participated in
the planning, did a majority of the experiments, all calculations and wrote
the manuscript together with my coauthors. V: I did the planning, the
experiments, developed the empirical approach and wrote the manuscript.
VI: I did the planning, the experiments, the calculations and wrote the
manuscript.

Papers not Included in the Thesis

VII Fast Estimation of Adsorption Isotherm Parameters in Gradient

Elution Preparative Liquid Chromatography. I: The Single Com-
ponent Case. D. Åsberg, M. Leśko, M. Enmark, J. Samuelsson, K.
Kaczmarski and T. Fornstedt. Journal of Chromatography A, 1299, 64–70
(2013)

VIII Fast Estimation of Adsorption Isotherm Parameters in Gradient

Elution Preparative Liquid Chromatography. II: The Competitive
Case. D. Åsberg, M. Leśko, M. Enmark, J. Samuelsson, K. Kaczmarski
and T. Fornstedt. Journal of Chromatography A, 1314, 70–76 (2013)

IX Choice of Model for Estimation of Adsorption Isotherm Parame-

ters in Gradient Elution Preparative Liquid Chromatography. M.
Leśko, D. Åsberg, M. Enmark, J. Samuelsson, K. Kaczmarski and T.
Fornstedt. Chromatographia, 78, 1293–1297 (2015)

viii
 X Estimation of Nonlinear Adsorption Isotherms in Gradient Elution
RP-LC of Peptides in the Presence of an Adsorbing Additive. D.
Åsberg, M. Leśko, T. Leek, J. Samuelsson, K. Kaczmarski and T.
Fornstedt. Chromatographia, dx.doi.org/10.1007/s10337-017-3298-y
(2017)

XI The Importance of Ion-Pairing in Peptide Purification by Reversed-

Phase Liquid Chromatography. D. Åsberg, A. Langborg Weinmann,
T. Leek, R. J. Lewis, M. Klarqvist, M. Leśko, K. Kaczmarski, J. Samu-
elsson, and T. Fornstedt. Journal of Chromatography A, dx.doi.org/10.
1016/j.chroma.2017.03.041 (2017)

XII Evaluation of Co-Solvent Fraction, Pressure and Temperature Ef-

fects in Analytical and Preparative Supercritical Fluid Chromato-
graphy. D. Åsberg, M. Enmark, J. Samuelsson and T. Fornstedt.
Journal of Chromatography A, 1374, 254–260 (2014)

XIII The Effect of Temperature, Pressure and Co-Solvent on a Chiral

Supercritical Fluid Chromatography Separation. M. Enmark, D.
Åsberg, J. Samuelsson and T. Fornstedt. Chromatography Today, 7,
14–17 (2014)

XIV A Closer Study of Peak Distortions in Supercritical Fluid Chroma-

tography as Generated by the Injection. M. Enmark, D. Åsberg, A.
Shalliker, J. Samuelsson and T. Fornstedt. Journal of Chromatography A,
1400, 131–139 (2015)

XV Evaluation of Scale-Up from Analytical to Preparative Supercritical

Fluid Chromatography. M. Enmark, D. Åsberg, H. Leek, K. Öhlén,
M. Klarqvist, J. Samuelsson and T. Fornstedt Journal of Chromatography
A, 1425, 280–286 (2015)

XVI Analytical Method Development in the Quality be Design Frame-

work. D. Åsberg, A. Karlsson, J. Samuelsson, K. Kaczmarski and T.
Fornstedt. American Laboratory, 46, 12–15 (2014)

ix
Contents

Abstract iv

Swedish Summary v

List of Papers vii

1 Introduction 1
1.1 Trends in Liquid Chromatography . . . . . . . . . . . . . . . 2
1.2 Trends in Pharmaceutical Analysis . . . . . . . . . . . . . . . 3
1.3 Post-Approval Changes . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Models for Liquid Chromatography 8

2.1 The Mass Balance . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 The Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 The Adsorption Process . . . . . . . . . . . . . . . . . . . . . 11
2.4 The Stochastic Model . . . . . . . . . . . . . . . . . . . . . . . 17

3 UHPLC Fundamentals 20
3.1 Geometric Scaling . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Effects of Elevated Pressure . . . . . . . . . . . . . . . . . . . 22
3.3 Effects of Viscous Heating . . . . . . . . . . . . . . . . . . . . 24

4 Analytical Method Validation 26

4.1 Method Performance Parameters . . . . . . . . . . . . . . . . 26
4.2 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.3 Analytical Quality by Design . . . . . . . . . . . . . . . . . . 31

5 Results and Discussion 34

5.1 Paper I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Paper II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

x
 5.3 Paper III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.4 Paper IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.5 Paper V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.6 Paper VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

6 Concluding Remarks 57

Acknowledgment 59

References 60

Appendix 72
Paper I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Paper II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Paper III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Paper VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Paper V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Paper VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

xi
Chapter 1
Introduction

The pharmaceutical industry is regulated by governmental agencies to

ensure the safety, quality and efficacy of the produced drug products and
one aspect of the regulatory oversight is the drug application process [1]. In
the EU, drug applications are sent to the European Medicines Agency (EMA)
and, once approved, are valid in all member states [2]. Analytical methods
are a necessary part of the drug application and are used to generate data
for acceptance, release and stability of the drug, making pharmaceutical
analysis one of the most important branches of applied analytical chemistry
[3]. All analytical methods in the drug application have to be validated to
ensure that they produce consistent results of sufficient quality for their
intended purpose [4].
A majority of all analytical methods for pharmaceutical analysis are
based on high-performance liquid chromatography (HPLC) and the most
common mode is by far reversed-phase liquid chromatography (RPLC) [3,
5]. RPLC is a versatile separation technique well suited to separate and
quantify a wide range of organic molecules and ions commonly encountered
in pharmaceuticals [6]. HPLC was developed in the 60s but there has been
major advances throughout the years and it is still an active research
area. Among the more recent advances are ultra-high-performance liquid
chromatography (UHPLC), which takes advantage of the higher throughput
and/or better column efficiency offered by sub-2 µm stationary phase
particles, and core-shell particles having a solid core surrounded by a
porous shell [7]. However, it has been challenging to convert existing HPLC
methods to UHPLC in a predictable and robust way [8, 9]. In this thesis, I
investigate UHPLC in depth using advanced separation theory to map the
fundamental differences between UHPLC and HPLC in order to facilitate
the implementation of UHPLC in pharmaceutical analysis so that validation
meets modern regulatory guidelines.

1
1.1 Trends in Liquid Chromatography

There is a trend towards increasing the column efficiency and reducing

the analysis time in order to meet the ever increasing demands from the
chemical and pharmaceutical industry [7, 10]. Increasing the column
efficiency allows for higher peak capacities or separation of closely related
compounds while reducing the analysis time increases throughput by
making it possible to analyze more samples in the same time.
UHPLC takes advantage of narrow-bore columns packed with sub-
2 µm particles which offers shorter analysis time with preserved separation
efficiency [7, 10]. The smaller particles increases the pressure resistance of
the column since it is inversely proportional to the square of the particle
diameter. In addition, the optimal mobile phase velocity is also inversely
proportional to the particle diameter making the operating pressure even
higher at optimal kinetic performance [7]. In 2004, Waters Corp. introduced
the first commercial UHPLC instrument which was marketed under the
trade name Ultra-Performance Liquid Chromatography (UPLC) [11]. A UHPLC
instrument can operate at higher pressure (usually up to 1200 bar) than
HPLC instruments and has decreased extra-column volume to minimize
system dispersion [7, 10]. Generally, a five-fold fold increase in throughput
compared to conventional HPLC can be achieved [8, 12]. UHPLC columns
have an inner diameter of 2.1 mm instead of the conventional 4.6 mm which
results in a lower volumetric flow rate and a lower solvent consumption
[8, 13]. Reducing the solvent consumption is cost-effective and makes the
analysis more environmentally friendly [13].
The pharmaceutical industry recognized the benefits UHPLC soon
after its commercialization in 2004 [14]. However, it is unlikely that the
pharmaceutical industry will replace the majority of their HPLC equipment
with UHPLC in the near future due to the cost of investing in new equipment
and a still functioning HPLC park [6]. UHPLC is mainly used at sites
focusing on research and development while production facilities still
largely employ HPLC [8, 15]. A large pharmaceutical company may have
hundreds or even thousands of HPLC systems across its organization and it
is not economically feasible to replace all instruments at once [6]. Therefore,
there is a mix of HPLC and UHPLC equipment in most pharmaceutical
organization creating a demand for efficient and robust method conversion
between instruments.
Furthermore, it can be technically challenging to predictably convert
a method from HPLC to UHPLC. A very large pressure gradient over the

2
column is generated in UHPLC causing significant temperature gradients
due to viscous heating [7, 16]. Although well established scaling rules
exists when changing the column dimensions and the particle diameter [9],
the temperature gradients and the pressure itself can affect the retention,
selectivity and efficiency of the separation [16]. There has also been some
indications of stationary phase properties differing between 5 µm and
sub-2 µm particles which can further complicate matters [17].
In 2006, stationary phase particles with a solid core and a porous shell,
called core-shell particles, were reintroduced successfully to the market after
having gained little interest since the 60s [18, 19]. Modern core-shell particles
can show a reduction in minimum plate height of up to 25% compared to
fully porous particles of the same size and have a smaller C-term in the
van Deemter equation (mass transfer coefficient) [7, 20]. Thereby, core-shell
particles can either improve efficiency compared to fully porous particles
or lower the flow resistance keeping the efficiency constant by switching
to a larger particle size [19]. The efficiency gain from core-shell particles
at optimum flow rate is not completely understood, but believed to stem
from a reduction in the longitudinal diffusion term and a reduction of the
eddy dispersion term [19, 20]. The reduction in the longitudinal diffusion
term can be understood since a significant fraction of the column volume is
occupied by a non-porous material through which the solute cannot diffuse
while the reduction of the eddy dispersion term has been attributed to a
higher transcolumn homogeneity of the core-shell columns. Using UHPLC
instrumentation with columns packed with sub-2 µm core-shell particles
is the setup that today yields the lowest height equivalent to a theoretical
plate. However, it has been suggested that such high efficiencies are not
always necessary and core-shell particles have not yet seen widespread
use in pharmaceutical analysis [19]. Although there have been reports of
pharmaceutical applications demonstrating the use of core-shell particles
[21–25].

1.2 Trends in Pharmaceutical Analysis

Historically, the pharmaceutical industry has been slow to adopt new

technologies in their manufacturing and quality control of release batches
resulting in suboptimal performance. One reason for this was a perceived
hindrance from strict governmental regulations which made it difficult to
change an already approved process or analytical method [26]. If a method
would be changed after its approval, an assessment of the modified method

3
is needed and the change should be approved by the regulatory agencies in
the regions where the drug is marketed [27]. Additionally, the number of
batches that had to be discarded because they did not meet specifications was
high and the reason for the batch failures was often unknown. To address
these issues, the US Food and Drug Administration (FDA) introduced the
initiative Pharmaceutical cGMP for the 21st Century - A Risk-Based Approach
in 2004 [28]. The objective of the initiative was to enhance and modernize
regulation of pharmaceutical manufacturing and product quality. CMC1
and pharmaceutical regulatory programs were evaluated to encourage early
adoption of new technology and implementation of risk-based approaches,
i.e. focusing on factors critical to the safety and quality of the drug product.
The International Council for Harmonisation (ICH) releases guidelines
harmonizing the pharmaceutical development process in the EU, the US
and Japan [4], while it is the governmental agencies in each region that
evaluate and approve drug applications. Following the FDA initiative,
ICH developed a set of guidelines for pharmaceutical manufacturing and
outlined a number of concepts, among them Quality by Design (QbD), to
facilitate a science and risk-based approach to pharmaceutical manufactu-
ring [29–32]. The ICH Q8 guideline states that “pharmaceutical QbD is a
systematic approach to development that begins with predefined objectives
and emphasizes product and process understanding and control based
on sound science and quality risk management” [29]. Quality should be
built into the product by understanding the process and product giving
a more robust manufacturing process less prone to failure [33]. QbD is
also intended to give more regulatory flexibility to the manufacturing
process and, together with a modern quality management system, promote
continual improvement of the process during its lifetime.
The industry has begun implementing parts of the ICH Q8–Q11 gui-
delines and this is often denoted as an enhanced approach to product
development compared to a traditional approach. However, regulatory
agencies are not classifying drug applications as enhanced or traditional
applications as they often contains elements from both approaches. In can
be said to exist a continuum of approaches with the traditional approach,
based on compliance with set specifications, at one end and the science and
risk based, enhanced approach on the opposite end.
After the introduction of QbD in manufacturing, researchers began
1Chemistry, manufacturing and control are drug development activities before the clinical
trials which involves assessing safety, pharmacokinetics, dose, etc. as well as developing the
manufacturing process and suitable packaging.

4
applying QbD principles also to the development of analytical methods [34].
This extension of QbD is known as analytical QbD and the objective is to
increase the robustness of analytical methods and provide more regulatory
flexibility to make the life-cycle management of the methods more efficient
[35–39]. However, there is currently no international consensus from
regulatory agencies on analytical QbD and no official guidelines have been
released. Therefore, applications that includes analytical QbD elements are
evaluated on a case-by-case basis [40].
A timeline of important guidelines concerning drug development is
shown in Fig. 1.1. An annex to ICH Q8 was added in 2008 which discussed
QbD explicitly and an official FDA guideline referring to the ICH Q8–Q10
documents were released in 2011 [41]. Analytical QbD was first suggested
in 2007 [42] and has recently been elaborated in several book chapters [26,
43, 44]. The next ICH guideline, ICH Q12, is due 2017 and will be about
life-cycle management and post-approval changes, since the current ICH
guidelines have not had the intended impact on these subjects [45].

Figure 1.1: A timeline over important guidelines for pharmaceutical development.

QbD relies on the construction of a design space where the relationship

between critical factors and process quality is known. The design space is
part of the drug application and it is possible to change the process inside
the design space without the change counting as a post-approval change
[33]. Mathematical modeling is then needed to obtain the relationship
between quality attributes and input variables (settings, materiel attributes,
etc.) and the complexity of the model depends on the process. The model
forms the basis of the knowledge about the process on which to base the
design decisions in the drug application [29]. Therefore, there is also a
trend towards increased mathematical modeling and computer simulations
both in pharmaceutical manufacturing and analysis [46].

5
1.3 Post-Approval Changes

To convert an approved HPLC method to UHPLC, a post approval change

need to be filed to the regulatory agencies. FDA denotes a change to an
approved drug application as a post-approval change [47] while the EMA
denotes it as a post-authorization variation [48]. To avoid confusion, I will
use the FDA term, but the terms are interchangeable. Depending on the
potential of a change to have an effect on the safety or effectiveness of a drug
product, FDA divides changes into three types; major, moderate and minor
changes. A major change requires the submission of a supplement and
approval by FDA prior to distribution of the drug product made using the
change. A moderate change is divided into two subtypes, one requires the
submission of a supplement to FDA at least 30 days before the distribution of
the drug product made using the change and the other allows distribution
as soon as FDA receives the supplement. A minor change has only minimal
potential to have an adverse effect and need only be described in the annual
report. EMA recognizes similar types of changes but denotes them as type
IA, IB and II where type II is similar to a major change, type IB is similar to
a moderate change and type IA is similar to a minor change.
A change to an analytical method outside the specifications given in
the drug application is categorized as moderate or major depending on its
potential to have an adverse impact on the performance of the analytical
method [47]. Formally, converting a HPLC method to UHPLC rank as
type 1B in the EU, meaning that regulatory agencies have 30 days to send
questions or require more information and if they do not, the change can
be implemented when the time period expires [48]. However, in reality
there is always a dialog between the company and regulators before a post
approval change of this kind is implemented.
After a modification, the method should perform equally or better
compared to the original one. The modified method need to be revalidated
and an equivalence protocol between the modified and original method
should be established to get approval from the regulatory agencies [8].
However, it takes time to get the necessary approvals, since a method
can easily be marketed in over 100 countries [6]. Therefore, it may not be
feasible, under current regulations, to change a HPLC method to UHPLC
for a marketed product. Therefore, new drug applications can have two
analytical methods, one for HPLC and one for UHPLC, filed for the same
analysis to avoid changing from HPLC to UHPLC post approval [12, 49].

6
1.4 Aim

Implementation of new technology for continual improvement of pharma-

ceutical manufacturing and analysis is today essential to stay competitive
and regulatory agencies has began promoting adoption of new technology
in current and upcoming regulatory guidelines. UHPLC has great potential
to improve the throughput in pharmaceutical analysis but both technical
and regulatory challenges for a wide spread implementation exists.
This thesis addresses the technical challenge with predictable method
conversion between HPLC and UHPLC from a fundamental perspective
and the regulatory challenge of efficient post approval change management.
Paper [I] studies the separation of the drug substance omeprazole from two
impurities, and investigates how mechanistic modeling can complement
conventional chemometric models to elucidate the separation process. Ome-
prazole, Fig. 1.2, is a proton-pump inhibitor and the API in the blockbuster
drugs Losec and Nexium [50]. In paper [II] the potential differences in
stationary phase properties between sub-2 µm particles and conventional
HPLC particles were investigated. The effect of elevated pressure and
viscous heating on the retention and peak shape was studied in papers
[III,V,VI] while the conversion of a quality control method from HPLC
to UHPLC was studied in paper [IV]. Furthermore, in paper [IV] a new
concept for handling post-approval changes with minimal regulatory in-
teraction was presented and exemplified with a realistic case example, a
quality control method for the drug Nexium. In the end my hope is that the
knowledge gained from a fundamental understanding of UHPLC would
support analysts to reliably switch between HPLC and UHPLC based on
scientific decision making which, in turn, would increase the regulators’
trust. Our regulatory concept may serve as support for implementing the
upcoming ICH Q12 guideline handling life cycle management of analytical
methods.

Figure 1.2: The structure of the proton pump inhibitor S-omeprazole.

7
Chapter 2
Models for Liquid Chromatography

A mechanistic model is based on assumed physical or chemical processes

which are thought to have given rise to the phenomena which are being
studied. The parameters in a mechanistic model have a physical definition
and can be measured, in theory at least, independently of each other. In
contrast, a statistical model is based on an empirical relationship which
only seeks to best describe the data. The model parameters lack physical
meaning and does not attempt to explain underlying principles of the
studied process.
In liquid chromatography, models are mainly used for two purposes;
to increase the understanding of the chromatographic process and to find
optimal operating conditions for a separation. Mechanistic models have
been used extensively to increase the understanding of the adsorption [51,
52] and mass transfer [53, 54] processes in chromatography while statistical
models are common in method development and validation [43, 44].
For some models, it is important to distinguish between linear and
nonlinear chromatography because they may not be applicable to both [55].
Linear chromatography refers to conditions where the concentration of
adsorbed solute is directly proportional to the solute concentration in the
mobile phase, i.e. the adsorption isotherm is a linear function. This is the
situation often encountered in analytical applications of chromatography.
Nonlinear chromatography refers to a nonlinear relationship between the
solute concentration in the mobile phase and in the stationary phase,
i.e. a nonlinear adsorption isotherm, and is encountered in preparative
chromatography, but also in analytical separations where the peaks are
overloaded. Models of chromatography based on the mass balance of each
component can be applied to both linear and nonlinear conditions while
molecular dynamic models, which are based on a microscopic description
of the solute migration, are not suitable for nonlinear chromatography [56].

8
2.1 The Mass Balance

The mass balance of a solute is a partial differential equation describing

how the solute is transported through the chromatographic column and
has parameters such as flow rate, bed porosity and adsorption equilibrium
parameters. The equilibrium-dispersive (ED) model was used to describe the
mass balance in paper [I,II,VI]. The ED model assumes that the contribution
of all non-equilibrium effects can be lumped together into an apparent
axial dispersion coefficient, that the mobile phase is non-compressible, that
the pure solvent is not adsorbed on the stationary phase and isothermal
conditions [55]. For a radially homogeneous column the ED model is
written as
∂C ∂q ∂C ∂2 C
+F +w  Da 2 (2.1)
∂t ∂t ∂z ∂z
where z is the length coordinate, t the time coordinate, F is the ratio between
stationary and mobile phase volume and w is the interstitial velocity of the
mobile phase. If w is unknown, it can be replaced by the linear velocity. The
equilibrium concentration of the solute in the stationary phase, q, is related
to its corresponding mobile phase concentration, C, through the adsorption
isotherm. The apparent axial dispersion coefficient, Da , is calculated from

Lu
Da  (2.2)
2Nap

where L is the length of the column, u is the linear velocity and Nap apparent
plate number. The initial condition used in this work was that the column
was equilibrated with pure mobile phase before each experiment. The
Danckwerts boundary conditions were applied [57]:

∂C 

wC(0, t) − Da  wC 0 (t) (2.3a)
∂z z0
∂C 

0 (2.3b)
∂z  zL

where C0 (t) is the injection profile which is determined experimentally.

The ED model was solved numerically; orthogonal collocation on finite
elements were used to discretize the spatial derivatives [58] and the Adams-
Moulton method implemented in the VODE procedure was used to solve
the system of ordinary differential equations [59]. This numerical method is
robust and has been used extensively in the past to solve the ED model [55].

9
2.2 The Energy Balance

In UHPLC, heat is generated in the column due to viscous friction when

the mobile phase percolates the column. To describe this process, the
energy balance for the bed and the column wall in cylindrical coordinates
are needed [60, 61]. The energy balance for the bed describing the heat
generated by viscous friction and the energy evacuated from the column
under steady state conditions is
∂T ∂T 1 ∂ ∂T ∂ ∂T ∂P
   
Cp uz + Cp ur  rλ eff + λ eff − u z (1 − αT) (2.4)
∂z ∂r r ∂r ∂r ∂z ∂z ∂z
where C p is the mobile phase heat capacity, u z and u r are local super-
ficial mobile phase velocities in z and r directions respectively, T is the
temperature, P is the pressure, α is the mobile phase thermal expansion
coefficient and λ eff is the effective bed thermal conductivity which depends
on temperature and hence position. The left-hand side of Eq. (2.4) is the
flux of energy carried with the mobile phase by convection. The first and
second term on the right-hand side describes the energy change due to
conduction and the third therm is the heat generated by viscous heating
and lost by thermal expansion of the mobile phase. The energy balance for
the column wall in contact with the surrounding is also needed [60, 61]:
1 ∂T ∂2 T ∂2 T
 
0  λw + 2 + λw 2 (2.5)
r ∂r ∂r ∂z
where λw is the column wall conductivity. The column is no longer
isothermal and using the ED model for a radially heterogeneous bed in
cylindrical coordinates and accounting for spatial variations in u and Da
yields [62, 63]:
∂C ∂q 1 ∂(uC) ∂ ∂C 1 ∂ ∂C
   
+F +  Da,z + rDa,r (2.6)
∂t ∂t t ∂z ∂z ∂z r ∂r ∂r
t is the total porosity of the bed. When the column is packed with core-shell
particles instead of fully porous ones, the phase ratio is calculated as
(1 − e ) (1 − s ) 1 − ρ3


F (2.7)
t
where e is the external porosity, s is the shell porosity and ρ is the ratio of
the solid core diameter to the external diameter of the core-shell particle.
The ED model performs well modeling elution profiles in non-isothermal
conditions [64] when solving the steady state energy balance, Eq. (2.4) and
Eq. (2.5), for the column prior to the mass balance, Eq. (2.6).

10
2.3 The Adsorption Process

The adsorption process is the core of modeling RPLC, since it governs

the separation of the sample components and it is characterized by the
adsorption isotherm for each component. There exists a number of different
adsorption isotherm models, both mechanistic and empirical, but this
section only discusses the ones used in the thesis. The adsorption isotherm
needs to be determined experimentally and when doing this, calculating
the adsorption energy distribution is often helpful for selecting the most
appropriate isotherm model.

2.3.1 Adsorption Isotherm Models

The adsorption isotherm relates the concentration of a solute adsorbed on

the stationary phase to the concentration in the mobile phase. The most
common liquid-solid adsorption isotherm is the Langmuir isotherm [55]:

qs bC
q(C)  (2.8)
1 + bC
where q s is the monolayer saturation capacity and b is the association
equilibrium constant. At the limit C → ∞ we have q  q s , because the
Langmuir model assumes no adsorbate-adsorbate interaction and therefore
only a monolayer of solute molecules covers the stationary phase. The
retention factor, k, is related to the initial slope of the adsorption isotherm,
i.e. for Eq. (2.8) we have

dq 
k  F lim
0
 Fq s b  Fa (2.9)
C →0 dC C0

where a is called the Henry constant. If the stationary phase is heterogene-

ous, and contains two discrete types of adsorption sites, the bi-Langmuir
model can be used instead [55]:

q s,1 C qs,2 C
q(C)  + (2.10)
1 + b1 C 1 + b2 C

where subscripts 1 and 2 denotes the two adsorption sites and qs  q s,1 + q s,2 .
The bi-Langmuir model have been used in paper [II,VI]. If the solute
exhibits multi-layer adsorption, the liquid-solid BET model is suitable [65]:

bS C
q(C)  qs (2.11)
(1 − bL C) (1 − bL C + bS C)

11
where bS is the liquid-solid association equilibrium constant and b L is the
association equilibrium constant between layers of adsorbed molecules.
This adsorption isotherm was used to model the adsorption of omeprazole
in paper [II]. Eq. (2.11) can be limited to solute molecules adsorbing in two
layers and this version of the BET isotherm is written as [66]

bS C + 2bS bL C 2
q(C)  q s (2.12)
1 + bS C + bS bL C 2

which is mathematically equivalent to the quadratic model and the Moreau

model, but derived under different assumptions [55]. Eq. (2.12) was
employed in paper [I] to model the adsorption isotherm of the neutral form
of omeprazole.
Adsorption isotherms can be classified into different types based on their
shape using the van der Waals classification which was originally developed
for gas-solid equilibrium isotherms in the 40s [67]. Type I represents
isotherms which are convex upwards, like the Langmuir isotherm, and
are the most common type of liquid-solid equilibrium isotherms. Both
type II and type III isotherms are possible when there are significant
adsorbate-adsorbate interactions, but are uncommon in RPLC. In type II,
the interaction between the solute and the stationary phase dominates and
the solute will form a complete monolayer before starting to form the next
layer. If the adsorbate-adsorbate interactions dominates, incomplete layers
will form and we have a type III isotherm. The BET isotherm, Eq. (2.11) can
describe both type II and type III depending on the relative strength of the
equilibrium constants. Fig. 2.1 shows characteristic adsorption isotherms of
type I-III, their corresponding overloaded elution profiles and a schematic
illustration of the formation of solute layers as the concentration increases.
In gradient elution, the organic modifier fraction changes continuously
during the migration of the solute through the column. The parameters in
the isotherms are dependent on the fraction of organic modifier, φ, and will
therefore also vary during the run [VII-VIII]. Based on the linear solvent
strength theory, the Langmuir model can be extended to incorporate φ
giving
a0 e−Sa φ C
q(C, φ)  (2.13)
1 + b0 e−Sb φ C
where S a and S b are parameters describing the sensitivity to the organic
modifier and a0 and b0 are the isotherm parameters in pure water. The
modifier fraction can be included in the bi-Langmuir and BET models in an
analogues manner which was done in paper [II].

12
Figure 2.1: a) Adsorption isotherms corresponding to type I, II and III. The type
I isotherm was obtained with Eq. (2.8) and type II and III was obtained with
Eq. (2.11). The parameters were q s  100 g/L, b  b S  0.1 L/g and b L  0.04
L/g for type II and b L  0.06 L/g for type III. b) Corresponding elution profiles
for an 500 µL injection of a 10 g/L sample calculated with Eq. (2.1) using linear
velocity 15 cm/min and Da  0.058 cm2 /min. c) Illustration of the formation of
layers corresponding to the three types.

2.3.2 Adsorption Isotherm Acquisition

The adsorption isotherm parameters for a certain system cannot be deter-

mined from theory and so must be determined experimentally. There are
many different experimental methods for determining adsorption isotherms.
Frontal analysis (FA) and the perturbation peak (PP) method are accurate
and well established in the literature. However, they require relatively large
amounts of sample and many experiments [68]. In FA, the dependence
of retention time of breakthrough curves on the solute concentration is
used to determine single component adsorption isotherms. In the PP
method, the column is equilibrated with a certain solute concentration
and a perturbation is created by injecting a small excess or deficiency of
the solute. The retention time of the perturbation peak is related to the
slope of the adsorption isotherm and by introducing perturbations with
the column being equilibrated with different solute concentrations, the
adsorption isotherm can be determined.
Two methods which requires fewer experiments and less sample are the

13
elution by characteristic point (ECP) method [69] and the inverse method
(IM) [70, 71]. In the ECP method, the diffuse rear of an overloaded elution
profile is related to the adsorption isotherm. In the inverse method, an
adsorption isotherm model has to be chosen a priori, and the parameters
are estimated by optimizing the agreement between calculated and experi-
mental elution profiles. Both the ECP and IM method are limited by the
highest eluted concentration of the elution profile. The inverse method has
some advantages compared to other methods, namely that it can be used to
estimate not only adsorption isotherm parameters but also parameters in
the mass balance such as Da and that it can be used when the conditions in
the column are not constant. For example, the inverse method was used
to estimate adsorption isotherm parameters directly from elution profiles
obtained in gradient elution, Fig. 2.2, in paper [VII,VIII,IX].

Figure 2.2: An illustration of the inverse method in gradient elution; first an initial
guess of the adsorption isotherm parameters in Eq. (2.13) was done followed by a
numerical optimization to the experimental elution profile.

A proven strategy for determining how the adsorption isotherm for a

solute varies with experimental conditions such as the fraction of organic
modifier in the mobile phase, is to determine the the adsorption isotherm
with FA or PP at one reference condition and then use ECP or IM to
determine the isotherm parameters at new conditions [51]. This approach
was used to study the effect of pH [I], organic modifier [II] and pressure
[VI] on the adsorption isotherm.

14
2.3.3 The Adsorption Energy Distribution

The adsorption energy distribution (AED) assumes a continuous distribution

of adsorption energies and is used to characterize the heterogeneity of a
surface, thereby determining the number of different types of adsorption
sites present on the stationary phase surface. This information can then be
used to select an appropriate adsorption isotherm model [51].
The equilibrium constant is related to the adsorption energy, εa , through
the Arrhenius equation
b  b0 eεa /RT (2.14)
where b0 is a preexponential factor and R is the universal gas constant. A
fundamental equation of the global adsorption isotherm on heterogeneous
surfaces is [72] ∫
q(C)  f (ln b)θ(b, C) d(ln b) (2.15)

where q(C) is the amount of solute adsorbed at solute concentration C,

f (ln b) is the AED and θ(b, C) the local adsorption model. θ(b, C) is often
the Langmuir model, Eq. (2.8), but it is possible to use others like the BET
model, Eq. (2.11), which was done in paper [II]. To obtain the AED from
Eq. (2.15) for a set of experimental adsorption data (C and q), the expectation
maximization (EM) algorithm provides a robust method that converges
with high stability toward the maximum-likelihood estimate [72, 73]. It
does not require any prior knowledge of the distribution function and no
smoothing of the isotherm data. A unimodal and a bimodal AED is shown
in Fig. 2.3 with corresponding adsorption isotherms. It would be difficult
to distinguish between a homogeneous or heterogeneous model from only
the adsorption isotherm data.

2.3.4 Insights Into the Adsorption Mechanism

The stationary phase surface of a C18-bonded porous silica adsorbent is

always, to some extent, heterogeneous [74]. The heterogeneous surface
arise from different sources; e.g. heterogenities of the bare silica surface
due to elemental impurities or bond strains of the silicate tetrahedra [75],
residual silanol groups [76] or heterogenities in the structure of the C18
bonded phase itself [77].
Based on adsorption isotherm measurements and AED calculations, it
has been suggested that the heterogeneity of the stationary phase causes the
solute to adsorb through different interactions on the same adsorbent [74].
The solute interacts with different types of adsorption sites characterized by

15
 100
300 a) b)
80

60
q s [g/L]

200

q [g/L]
40
100
Langmuir
20
bi-Langmuir
0 0
-7 -6 -5 -4 -3 -2 -1 0 0 50 100 150
ln(b) C [g/L]

Figure 2.3: a) AED for surfaces with one and two types of adsorption sites. b)
Corresponding adsorption isotherms where Langmuir is a one site model and
bi-Langmuir is a two site model. (Langmuir: q s  110 g/L, b  0.027 L/g and
bi-Langmuir: q s,1  100 g/L, b 1  0.01 L/g, q s,2  10 g/L, b 2  0.2 L/g.)

their adsorption energies. From AED calculations, it is evident that there

are several discrete adsorption sites and low molecular weight compounds
usually show two distinct adsorption sites on an end-capped stationary
phase in RPLC [51, 74]. The adsorption site having the lowest energy, i.e.
the weakest equilibrium constant, has the highest saturation capacity. It
has been proposed that the low energy sites are dispersive interactions
at the interface between the mobile phase and the C18-bounded layer.
The adsorption sites having a higher adsorption energy and thereby a
larger equilibrium constant, usually have a lower saturation capacity and
is located deeper into the C18 layer, closer to the silica surface. The exact
nature of these adsorption sites are not completely understood and depends
on the properties of the solute molecule. A schematic description of a
solute adsorbing on a C18-bonded stationary phase having two different
adsorption sites is given in Fig. 2.4.
Even though the high energy sites are fewer in number, they can
contribute significantly to the retention factor [78]. If the bi-Langmuir
adsorption isotherm is assumed, the retention factor is related to the
adsorption isotherm through Eq. (2.9) and since the low energy sites
have high saturation capacity and low equilibrium constant and vice
versa for the high energy site, they contribute to a similar extent to the
retention factor. Furthermore, a well known reason for peak tailing in liquid
chromatography is column overloading and a solute having heterogeneous
adsorption can have a different linear isotherm range compared to a solute

16
Figure 2.4: Schematic description of the adsorption equilibria for a solute molecule
adsorbing at two types of adsorption sites; one located near the C18-bonded layer
interface and one located deeper in the C18 layer.

having homogeneous adsorption.

Molecular simulations supports the interpretation of multiple adsorp-
tion sites [52]. Calculations of the spatial distribution coefficient for the
solute in the direction perpendicular to the silica surface show how the
solute is distributed in different regions of the stationary phase. For a
polar compound, 1-propanol, the distribution coefficient exhibited two
preferential regions [79]. 1-Propanol adsorb on the interface of the C18
layer and the second region was close to the silica surface, deep into the
C18 bonded layer with the interface region being much larger.

2.4 The Stochastic Model

Another way of describing the chromatographic process is through molecu-

lar dynamic models which looks at the molecular level [56]. A molecular
dynamic model is based on a stochastic description of the molecule’s random
migration through the column. A molecule is assumed to perform a random
number of adsorption-desorption events, denoted n, and for each instance
it adsorbs, it spends a random amount of time adsorbed on the stationary
phase. The average time a solute molecule spends in the moving mobile
phase is called fly-time, τm , and the average time it spends adsorbed on

17
the stationary phase or in a stagnant part of the mobile phase is called
the sojourn time, τs . τs and τm are average quantities made up of the
individual values of the adsorption time for each adsorption event and the
time spent in the mobile phase between two adsorption events, which are
random quantities. The retention factor, k, is the ratio between τs and τm ,
i.e. k  τs /τm , and the retention time is calculated as tR  nτm + nτs , i.e.
the sum of the time spent moving in the mobile phase nτm and the time
spent in the stationary phase nτs after n adsorption-desorption events.
Stochastic models can be used to study the peak shape in linear chro-
matography and to better understand why some peaks are tailing even at
very lower concentrations. In line with the discussion about heterogeneous
adsorption in Section 2.3, molecular dynamic models also assumes multiple
adsorption sites [80]. Adsorption at a site with high adsorption energy are
characterized by a long τs while adsorption sites with a low adsorption
energy are characterized by a short τs . Adsorption on a heterogeneous
surface can result in peak tailing because the sites with short τs produce the
sharp front of the peak while the sites with long τs give rise to an elongated
tail.
In paper [III], the molecular dynamic model developed by Giddings and
Eyring [81] was used to study the peak shape as a function of pressure and
temperature for a tailing, cationic solute. In this model, the chromatographic
process is assumed to be a Poisson process and the chromatographic peak
is the probability density function of time spent in the column by the
molecules. If the stationary phase surface is assumed to be covered with
two types of adsorption sites and the relative amount of these sites are
denoted p i , the retention factor is

tR  nτm + n p1 τs,1 + p 2 τs,2

(2.16)

The column efficiency for an surface with two adsorption sites is given by

p 1 + p 2 τs,2 /τs,1

2  2
1 1 k
 + 2n

2 k + 1 (2.17)
N ND p 1 + p 2 τs,2 /τs,1

where ND is the dispersion effect from the mobile phase estimated from
an unretained marker. The skew is traditionally used to describe the peak
symmetry and is
3 p1 τs,1
3
+ p 2 τs,2
3
S √  (2.18)
2 n
 3/2
p1 τs,1
2
+ p 2 τs,2
2

18
A symmetrical chromatographic peak has S  1 and a tailing peak has
S > 1.
Direct evidence for the stochastic model assuming a multi-site adsorp-
tion process have been reported by single molecule spectroscopy [76, 82, 83].
Single molecule spectroscopy techniques provides experimental data of the
adsorption-desorption steps for individual molecules which is not possible
to obtain through conventional chromatographic experiments providing
only the ensemble-averaged information. When studying the adsorption of
a cationic molecule on a commercially available, end-capped C18-bonded
stationary phase with single molecule spectroscopy, strong, randomly dis-
tributed adsorption sites were revealed [84]. Additionally, the trajectory
of single molecules moving through C18-bonded, porous particles were
measured and from the adsorption-desorption events, there was evidence of
heterogeneity in the interaction with the surface [85]. Using single-molecule
data, it is also possible to approximate the overall adsorption site residence
time distribution for a process including both weak and strong binding
sites [86, 87]. With this approach, the experimental elution profiles could
be reproduced and the asymmetry of the peak at different mobile phase
compositions was correctly predicted.

19
Chapter 3
UHPLC Fundamentals

Sub-2 µm particles results in a high pressure resistance, a reduced C term

giving a flatter profile of the van Deemter curve and a lower optimal plate
height [88]. Fig. 3.1 compares the pressure drop and van Deemter curves
for columns packed with 1.7 µm, 3.5 µm and 5.0 µm particles. A higher
throughput and/or efficiency can be achieved in UHPLC at the cost of a
higher operating pressure [8]. This chapter discusses how to geometrically
scale a chromatographic method when the column dimensions and particle
size are changed. However, for a predictable conversion from HPLC to
UHPLC, the potential effects from the high pressure need to be taken into
account. The pressure can affect the chromatography by both influencing
the retention mechanism, and the the mobile and stationary phase by
compressing it [89]. Viscous heating is discussed separately from the other
consequences of high pressure since it is essentially a temperature effect.

a) b)

1.7 µm 5.0 µm
3.5 µm
Pressure

HETP

3.5 µm

5.0 µm 1.7 µm

Linear velocity Linear velocity

Figure 3.1: a) Comparing pressure drops over the column for three particle
diameters. b) Comparing van Deemter curves for these columns.

20
3.1 Geometric Scaling

Geometric scaling from one column dimension or particle size to another

with preserved selectivity is well established in liquid chromatography
[9]. UHPLC instruments have low extra-column volumes compared to
conventional HPLC to decrease extra-column band broadening and gradient
dwell time [7]. Therefore, these volumes need to be accounted for as well
when scaling a method from HPLC to UHPLC. In isocratic elution, the
injection volume, Vinj , and flow rate, Fv , need to be scaled when changing
the particle diameter, dp , and column diameter, dc [90]. The injection volume
for the changed method, denoted 2, is related to the injection volume of the
original method, denoted 1, through

V0,2
Vinj,2  Vinj,1 (3.1)
V0,1

where V0 is the column hold-up volume. The flow rate is scaled so that the
reduced linear velocity of the mobile phase is preserved, i.e.

dc,2
2
dp,1
Fv,2  2 d
Fv,1 (3.2)
dc,1 p,2

In gradient elution, the gradient dwell time, td , (the time before the gradient
reaches the column inlet) and gradient slope, β, must also be scaled and
the most straightforward approach is based on the linear solvent strength
theory [91, 92]. The scaled dwell time is then

Fv,1 V0,2
td,2  td,1 (3.3)
Fv,2 V0,1

where the Fv,2 is calculated with Eq. (3.2). For a linear solvent gradient, the
gradient volume should be preserved, i.e.

Fv,2 V0,1
β2  β1 (3.4)
Fv,1 V0,2

Experimental examples of geometric scaling can be found in [90, 91] and

was also performed in paper [IV].
However, even when following the geometric scaling rules rigorously,
the separation performance may shift between HPLC and UHPLC due to
the higher pressure in UHPLC [93–98]. Pressure and temperature effects are
discussed in the subsequent sections while the consequences for method
conversion are discussed in paper [II-V].

21
3.2 Effects of Elevated Pressure

At pressures around 1000 bar, the column steel casing expands, the stationary
phase is compressed and the mobile phase can no longer be regarded as
incompressible [89]. Although these effects have a relatively minor influence
on routine applications, they may need to be accounted for when performing
fundamental research since they affect important parameters such as the
mobile phase velocity and the bed porosity [99–101]. For conventional
columns, the expansion of the casing is increasing the volume by around
0.1% per 1000 bar [89]. The compression of a C18-bonded stationary phase
can decrease its volume by ca 1% when increasing the pressure 1000 bar
[89]. It is mainly the C18-bonded layer that is compressed and since most
of the surface area is located inside the pores, it is the internal porosity of
the bed which is most affected. The specific volume of the mobile phase
as a function pressure largely depends in its composition, with e.g. water
having a lower compressibility than acetonitrile, and the compressibility
of the mobile phase affects the local flow rate. However, the definition of
the flow rate set by the operator varies between instrument manufacturers
[7]. If the set flow rate is defined at the low pressure side of the pump, the
actual flow rate will be lower than the set one at the column inlet. On the
other hand, if it is defined at the high pressure side of the pump, the actual
flow rate will be higher than the set one at the column outlet.
In HPLC conditions, the pressure has been known to affect retention for a
long time [102–104], but has not gained much attention due to the moderate
pressure found in conventional HPLC. However, with the commercialization
of UHPLC, the effect of pressure on the retention mechanism became more
important since much larger pressure effects were encountered [93–96].
Since the pressure drop over the column is close to linear which means that
a pressure of 1000 bar during normal operation gives an average pressure
of 500 bar in the column [89], the solute will experience varying pressure
when it travels through the column. The retention factor can be related to
the pressure through [105]
 
∆V F
ln (k)  − P + ln + ln (k 0 ) (3.5)
RT F0
where ∆V is the difference in solute molar volume in the stationary phase
and in the mobile phase and subscript 0 indicates a reference condition. In
RPLC ∆V < 0, i.e. the solute molar volume is smaller when the solute is
adsorbed on the stationary phase, and an increase in retention with pressure
is observed. This can easily be understood through le Chatelier’s principle;

22
when the pressure is increasing the system strives to counteract the change
by increasing its volume, and hence lowering the pressure, which is achieved
when the solute spends a longer time in the stationary phase. Non-polar
solutes have a small pressure dependence while polar and charged solutes
have a larger pressure dependence [93–95]. This implies that ∆V is more
negative for polar and charged solutes, i.e. they experience a larger decrease
in volume when adsorbing. McCalley et al. [93–95] have suggested that
polar and charged solutes lose a larger part of their solvation layer when
adsorbing compared to nonpolar solutes (Fig. 3.2). Nonpolar solutes have
a solvation layer consisting of hydrophobic molecules like acetonitrile or
methanol which can more easily penetrate the C18-bonded layer when the
solute is adsorbed. Apart from the polarity, the molecular weight of the

Figure 3.2: Schematic description of the loss of solvation layer upon adsorption for
a polar compound on a C18-bonded stationary phase.

solute has a large influence on its pressure dependence. For example, for a
1000 bar pressure increase the retention factor increases 25-100% for small
molecules, 150% for peptides (∼ 1.3 kDa), 800% for insulin (∼ 6 kDa) and
3000% for myoglobin (∼ 17 kDa) [96].
The effect of pressure on column efficiency, assuming no viscous heating,
is not well investigated since it is difficult to study without causing viscous
heating. However, theoretical calculations shows that pressure has a
negative effect on column efficiency when working above the optimal flow
rate [106, 107]. The decrease in column efficiency was attributed to the
pressure dependence of the solute diffusion coefficient.

23
3.3 Effects of Viscous Heating

When the mobile phase percolates a column packed with very fine particles
at a sufficiently high flow rate, heat is generated due to viscous friction [108].
Viscous heating is negligible at HPLC pressures (< 400 bar), but becomes
significant in UHPLC and has therefore been given much attention in the
last decade [16]. The power generated by viscous friction is

Û  Fv ∆P
W (3.6)

making it directly proportional to both the flow rate and the pressure
drop [109]. The temperature profile in the column is dependent on how
the column is thermostated [60, 61]; in a still air environment the axial
temperature gradient prevails since the conditions are close to adiabatic
while in a water bath the heat is effectively removed from the column wall
and radial temperature gradients dominates (Fig. 3.3a). In still air conditions
the axial gradients are in the range of 10 ◦C to 20 ◦C and in a water bath the
radial gradients are in the range of 0.5 ◦C to 5 ◦C at around 1000 bar.

Figure 3.3: a) Schematic illustration of the temperature gradient obtained with two
different thermal environments for the column at high flow rate. b) The reduced
height equivalent to a theoretical plate (HETP) and retention factor for omeprazole
as a function of mobile phase velocity on a BEH C18, 1.7 µm column in a still air
column oven and a water bath.

24
 Column efficiency is mostly affected by radial temperature gradients
because they cause the velocity profile of the solute band to vary in the
radial direction which makes the solute band broader [61, 110]. Therefore,
the efficiency loss is much larger in a column thermostated with a water
bath since the majority of the generated heat is then removed through the
column walls. In still air conditions, the negative effect of viscous heating
on efficiency is generally quite small which is evident from the van Deemter
curves in Fig. 3.3b.
The retention factor [111, 112] and adsorption isotherm [113, 114] are
functions of temperature and are therefore affected by viscous heating. The
retention factor as a function of temperature is described by the van’t Hoff
equation  
∆H ∆S F
ln(k)  − + + ln + ln(k0 ) (3.7)
RT R F0
where H and S are the enthalpy and entropy of transfer of the solute
from the mobile phase to the stationary phase, respectively. Normally, the
retention decreases with increasing temperature in RPLC, i.e. the opposite
to pressure where the retention increases. Viscous heating affects retention
when the column is in adiabatic conditions since the temperature increase
is then the most severe (Fig. 3.3b). Since an increase in pressure increases
retention and an increase in temperature decreases retention, the net effect
for a certain solute is hard to predict and both a net increase and a net
decrease in retention has been observed [98, 115]

25
Chapter 4
Analytical Method Validation

In the manufacturing of a drug product, a control strategy is mandatory to

confirm the quality of the product. In the product control strategy, different
product quality attributes, such as purity and stability, are monitored with
an analytical procedure [27]. All analytical procedures used in the control
strategy are part of the drug application and are reviewed and approved
together with the manufacturing process by regulatory agencies. In the
drug application, all analytical procedures need to be shown to be suitable
for their intended purpose and this is done by validating that they meet their
performance criteria [4, 116]. In the EU, the drug application for a new drug
product is made to the EMA which appoints two member states to review
the submission and after their approval, the new drug product may be
marketed in all member states. For a drug product marketed in the US, the
approval from the FDA is needed. After the patent has expired for a drug
product, official analytical procedures are published in pharmacopoeias
like the European Pharmacopoeia (EP) and the US Pharmacopoeia (USP).
Validation guidelines varies somewhat but they are all similar to the ones
given in ICH Q2 [4, 27, 117].

4.1 Method Performance Parameters

The quality of the data generated by an analytical method is specified in a

number of performance parameters [118]. When validating an analytical
method, the performance parameters are evaluated to confirm that they
meet specifications. Some performance parameters like accuracy and
reproducibility are only evaluated during the validation while others, like
chromatographic resolution, may also be included in a system suitability test
(SST) [5]. The purpose of a SST is to continuously monitor the performance
of the method to assure that it delivers data of sufficient quality.

26
4.1.1 Specificity

Specificity is the ability of an analytical method to assess unequivocally the

analyte in the presence of components that are expected to be present in the
sample and must therefore be assured before other performance parameters
can be validated [4, 119]. In liquid chromatography it means that the analyte
peak should be separated from all other peaks and it is usually assessed by
measuring the resolution factor between the analyte peak and its closest
neighbors. If co-elution is expected, it will also be necessary to test the
purity of the analyte peak with, for example, mass spectrometry. When
assessing specificity, relevant impurities and excipients can be spiked in
realistic concentrations to a sample containing the pure drug substance
and this approach was used in paper [IV] to assess specificity of a quality
control method for the drug product Nexium.

4.1.2 Precision

The precision of a method is the variability or variance between independent

measurements of the same sample under prescribed conditions [4, 119]. The
variability of a method is made up of a number of different contributions
and is divided in four levels: system precision, repeatability, intermediate
precision and reproducibility. Since variances are additive, the higher levels
includes the contributions from the lower levels, e.g. reapeatibility includes
the variance from the system precision (Fig. 4.1).

Figure 4.1: Example of sources to variability in the different precision levels.

In liquid chromatography, system precision is assessed by repeated

injections of the same sample and describes the variability of the instrument
and is an important parameter in the SST. Repeatability is assessed by

27
comparing independent test results for identical samples obtained under
identical condition (same equipment, operator, etc.) within a relatively
short time period. It is the sum of instrument variability and variability
from the sample preparation. The intermediate precision assesses the
variability in the procedure when done in the same lab, i.e. also accounting
for different analysts and instruments under a longer time period. Reprodu-
cibility is the variability between laboratories using different analysts and
different equipment when analyzing the same type of sample during an
extended period of time. It is important for methods intended for multiple
laboratories and for official methods reported in e.g. the USP or EP. In
liquid chromatography, precision is often measured as the relative standard
deviation in peak area. In paper [IV], the system precision was assessed in
the SST and the reproducibility was assessed in the validation by analyzing
the same sample in two different laboratories.

4.1.3 Accuracy

The accuracy of a method is how close a measured value is to a true or

accepted reference value and validation of the accuracy should focus on any
systematic bias [4, 119]. The accuracy of an assay of a drug substance should
preferably be done with an absolute measuring technique like titration or
quantitative NMR. For a drug product assay on the other hand, accuracy is
often assessed by measuring the recovery of a spiked sample with liquid
chromatography. Impurities are present in a much lower concentrations
compared to the API so the accuracy of impurity assays are often lower
than for the API. The accuracy was validated by assessing the recovery
of the API omeprazole and two impurities by analyzing spiked samples
containing a placebo matrix in paper [IV].

4.1.4 Linearity and Range

Linearity refers to the relationship between the measured signal and the
analyte concentration or amount [4, 119]. It is not mandatory that the
concentration is directly proportional to the signal, but it is the preferred
case since it is most practical. So, strictly speaking, the intended calibration
model should be assessed and not necessarily the linearity. The range
refers to the analyte concentrations intended for the method and depends
on the method’s purpose. For example, in the assay of a drug substance
the range is normally between 80 and 120% of the test concentration.
Linearity is evaluated in the concentration range by a calibration curve and

28
statistical measures like the residual values from linear regression or the
confidence intervals from the y-intercept. For LC-UV, it is usually enough
to graphically evaluate the deviation of experimental data from the fitted
calibration curve. Although not a measurement of linearity itself [120], the
correlation coefficient, R2 , is often given as support to an assessment of
linearity and was done so in paper [IV].

4.1.5 Detection and Quantification Limit

The detection limit of a method gives the lowest amount of analyte that can
be detected and the quantification limit gives the lowest amount of analyte
that can be quantified [4, 119]. Estimation of detection and quantification
limits can be done with different approaches but they are all based on the
variability of the method at very small analyte concentrations. In paper [IV],
the quantification limit was determined for the impurities in a drug product
based on the residual standard deviation (sr ) of the calibration curve as
QL  10sr /S where S is the slope of the calibration curve. The detection
limit can be determined with the same approach, i.e. DL  3.3sr /S.

4.2 Robustness

The robustness of a method is its capacity to remain unchanged by small

variations in procedure-related factors such as instrument settings, sample
preparation etc. and it indicates how sensitive the method is to changes in
these factors [4, 121]. Robustness is separated from ruggedness in the USP
with ruggedness being the variation sensitivity in non-procedure-related
factors such as operator, lab etc. and is assessed with the reproducibility
[121].
To assess a method’s robustness, factors are varied in a deliberate way,
often using DoE, and the quantitative influence on performance parameters
is studied [44]. In the case of liquid chromatography, factors that are often
investigated in a robustness test is eluent pH and composition, temperature,
flow rate and stationary phase batch. The robustness test is often performed
directly after the method development, i.e. before the method validation
activities given in Section 4.1, in order to avoid repeating the method
validation if the method should fail the robustness test.
A 2-level fractional factorial design us suitable for robustness studies
since the number of factors are usually large and the range of the factors
limited (in small intervals, curvature and interaction terms can often be
neglected). Examples of common responses in liquid chromatography are

29
resolution factor, tailing factor and retention factor of the first and last
eluting peaks. It can be advantageous to also include the responses in the
SST. Acceptance criteria are set for each response based on the required
performance and used to evaluate if the robustness is acceptable.
An example of a HPLC robustness assessment using DoE is given in
Fig. 4.2 which demonstrates four cases:

i The response is inside the acceptance criterion and the regression

model is insignificant. This is the ideal outcome which indicates that
the factors do not affect the response inside the design region. This
case is shown for the response impurity area in Fig. 4.2a where the
ANOVA p-value for the null hypothesis that there is no correlation
between the factors and the response is high and the variations in
impurity area are due to random noise.

ii The response is inside the acceptance criterion and the regression

model is significant. This outcome shows that the factors are affecting
the response and that a regression model can be established to
determine if the response is meeting the acceptance criterion in the
whole design region. This is the case for resolution factor and the
corresponding response surface, Fig. 4.2b, shows that it is above its
acceptance criterion in the design region.

iii The response is outside the acceptance criterion and the regression
model is significant. This outcome allows the regression model to
be used for determining a robust part of the original design region.
The retention factor in Fig. 4.2a has a significant regression model and
one experiment giving a value above the acceptance criterion. The
robust region can be determined from the confidence interval based
on the regression model’s variance and is shown in Fig. 4.2c as a
white, dashed line. Above this line the retention factor will be below
10 at 99% confidence level.

iv The response is outside the acceptance criterion and the regression

model is insignificant is the most problematic outcome. The response
is not robust in the design region and no regression model can be
obtained to determine a robust region. The tailing factor in Fig. 4.2a is
an example of this case and more experiments would be needed to
complete the evaluation of the robustness for this response.

30
Figure 4.2: A robustness test with three factors in a fractional factorial level III
design. a) Experimental results for the responses, the ANOVA p-value for the
model with insignificant model displayed in orange and acceptance criteria. b)
Response surface for the resolution factor where temperature was an insignificant
factor and c) response surface for the retention factor at 25 ◦C.

4.3 Analytical Quality by Design

Quality by Design (QbD) is a systematic approach to product development

wherein understanding the process is paramount and quality is built into
the process resulting in better robustness [29]. Researchers have suggested
that QbD principles can be applied also to the analytical procedures that are
part of the control strategy [34]. Development of an analytical procedure
based on QbD principles includes similar activities as for traditional method
development but are given in a more structured manner. The main difference
being that in analytical QbD the method performance is largely defined and
understood during the method development step instead of the validation
step [43, 44]. At the beginning of the method development, the method
goals and the performance required to achieve those goals are given in
an analytical target profile (ATP). The ATP specifies what to measure and

31
acceptance criteria for accuracy, precision, etc.
When a suitable method has been established, a risk assessment is
performed to identify critical method parameters that have a large impact
on the method’s performance. Traditionally, a risk assessment is often done
informally to determine which factors to include in the robustness study.
QbD, on the other hand, promotes a more formal risk assessment where the
result is supplied in the drug application and tools like Ishikawa diagrams,
failure mode effect analysis or mechanistic modeling are employed. An
example of an Ishikawa diagram is given in Fig. 4.3 describing a number
of factors for an HPLC method that can affect its performance. In a failure
mode effect analysis, potential factors are ranked according to severity,
likelihood and detectability to determine their importance [122]. Both these
tools requires knowledge about the analytical procedure to be effective
which can be problematic if there is a lack of experience of the analytical
technique. Mechanistic modeling can then be valuable since the analysis can
be simulated under varying conditions to investigate how certain variables
affects the performance [123].

Figure 4.3: Example of an Ishikawa diagram used to list potentially critical factors
for risk assessment of a HPLC method.

The design space is denoted method operable design region (MODR)

when applied to an analytical method and is essentially a range of settings
for critical parameters, for example pH, temperature and gradient slope
in liquid chromatography, where the method is shown to meet all its
performance criteria. A MODR is often determined using DoE and a
wider factor range is used compared to a conventional robustness study
resulting in significant regression models [43, 44]. The understanding of

32
how factors affects the responses are based on a mathematical model which
could be either mechanistic or statistical [124]. In the MODR, all acceptance
criteria are met and it is therefore allowed to change variables as long as the
method is operated inside the MODR. Finally a control strategy should be
established which ensures method performance each time the analytical
procedure is employed. The control strategy contains a SST, but the purpose
of the SST is not only to monitor the day-to-day performance of the method
but also to assess the method performance after intended changes. The
method development using the QbD concept is summarized in Fig. 4.4.
QbD promotes life cycle management in where the analytical method is
continuously improved. However, very few examples exists where this has
been realized and is one of the topics of paper [IV].

Figure 4.4: The steps in analytical method development withing the QbD frame-
work.

To appreciate the advantage of QbD, consider a quality control method

for batch release. If the method would fail, the drug cannot be released to
the market. So, if the original column brand becomes unavailable and the
new brand gives slightly worse performance, the success of the method
depends on its robustness. A traditional method may not be robust enough
to handle the reduced performance and the settings may be too tightly
specified to improve the performance. The method would need to be
redeveloped and a post approval change would have to be submitted the
regulatory agencies which could take months and in the meantime batches
will not be released. However, a QbD method would be more robust in the
first place and, if necessary, there is the possibility to do changes within the
MODR without regulatory interaction [43].

33
Chapter 5
Results and Discussion

Throughout paper [I-V], the API omeprazole has been investigated toget-
her with some of its impurities using a BEH C18 stationary phase from
Waters (Milford, MA, US) and a mobile phase consisting of acetonitrile
and phosphate buffer. In paper [I], the adsorption of omeprazole was
studied varying the pH and type of organic modifier and we show how
adsorption isotherm measurements can explain trends observed in a DoE
investigation. Paper [II-III] are concerned with the fundamental differences
between HPLC and UHPLC conditions, i.e. stationary phase chemistry,
pressure and viscous heating. Paper [IV] presents a case example of a
quality control method method being converted from HPLC to UHPLC as a
continual improvement and an approach for handling minor post-approval
changes with minimal regulatory interactions was developed. In paper [V]
an experimental method was developed to predict retention and selectivity
shifts due to the temperature and pressure gradients in UHPLC. Finally,
in paper [VI] the effect of pressure on the adsorption isotherm from a
fundamental point of view was investigated to explain some observations
made in the previous papers.

5.1 Paper I

In the development of an analytical procedure, DoE is often performed to

find the optimum operating conditions and in the validation step, DoE is an
essential part of the robustness study. Traditionally, the data generated in
DoE is modeled with a statistical regression model to find the relationship
between factors and responses. However, recent regulatory initiatives, like
QbD, encourage the understanding of the process or analytical procedure
and a statistical regression model may be inadequate to understand the
underlying reasons for the observed trends.

34
 Paper [I] demonstrated how adsorption isotherm modeling can compli-
ment an empirical DoE investigation to explain how eluent pH, temperature
and type of organic modifier affects the separation of omeprazole from
two of its impurities. Adsorption isotherm models provide a valuable tool
for studying the adsorption mechanism in liquid chromatography from a
mechanistic perspective. The separation was investigated in isocratic elution
with either acetonitrile or methanol as organic modifier. For each modifier,
an experimental design with pH and temperature as factors were performed
with a sample containing omeprazole and the impurities H168/66 and
H193/61 in relevant concentrations for a quality control method.
Response surfaces from the statistical models displayed some interesting
trends (Fig. 5.1). Methanol and acetonitrile gave quite different trends in
both resolution factor (Rs ), where methanol revealed a maximum resolution
at intermediate pH while acetonitrile gave highest resolution at high pH,
and tailing factor (Tf ) where omeprazole went from tailing to fronting
with acetonitrile as the pH increased and was relatively unaffected with
methanol.

Figure 5.1: Response surface for a) the resolution factor between omeprazole and
H168/66 and b) the tailing factor for omeprazole with acetonitrile as the organic
modifier. c) Resolution factor and d) tailing factor with methanol as organic
modifier.

The resolution factor between omeprazole and H168/66 was affected by

pH because both compounds behave as weak acids in the this pH interval.

35
By measuring the retention factor in a wide pH-interval it was evident
that when protonated, H168/66 has longer retention time than omeprazole
with acetonitrile and vice versa for methanol (Fig. 5.2). The crossing of
the lines in Fig. 5.2 represents a shift in elution order and is correlated to
the region where R s  0 in the response surface for acetonitrile (Fig. 5.1a).
Therefore, the difference in resolution with acetonitrile and methanol is
due to a selectivity difference for the protonated forms of H168/66 and
omeprazole.

10
a) MeCN b) MeOH
8
H168/66
6
Omeprazole
k

0
6 8 10 12 6 8 10 12
s s
w
pH w
pH

Figure 5.2: The retention factor as a function of sw pH for omeprazole and H168/66
with a) acetontrile and b) methanol as organic modifier at 30 ◦C.

The fronting of omeprazole was, at least partially, the result of overloa-

ding since the peak became more symmetrical as the sample concentration
decreased. The shape of an overloaded peak is governed by its nonlinear
adsorption isotherm and first the adsorption isotherm of the protonated
form of omeprazole was obtained using the inverse method. The 2-layer
BET model, Eq. (2.12), was selected as the isotherm model since it gave good
agreement with the experimental data both for acetonitrile and methanol.
Acetonitrile gave elution profiles consistent with a type III isotherm model
(diffuse front) while methanol gave elution profiles consistent with type
I (diffuse rear) as shown in Fig. 5.3a-b. This was reflected in the isotherm
parameters where the equilibrium constant for adsorption on the stationary
phase, b S , was about two times larger for methanol while the equilibrium
constant for adsorption on the solute layer, b L , was similar. The reason
for the weaker interaction with the stationary phase using acetonitrile was
believed to be the thick layer acetonitrile forms, compared to methanol,
at the stationary phase interface [125] which makes it more difficult for
omeprazole to get access to the stationary phase.
Since pH had a large influence on peak shape with acetonitrile, the

36
adsorption isotherm for omeprazole was also modeled as a function of
pH. The design region includes the pKa of omeprazole, so the adsorption
isotherm needed to account for this. Additionally, the buffer capacity was
not high enough to keep the pH constant when overloaded samples for
the inverse method were injected and the local variations in pH when
the sample zone moved through the column was also accounted for. An
semi-empirical model based on the general Langmuir isotherm, denoted
the pH model, was employed and showed that the fronting increases with
the fraction of deprotonated omeprazole for acetonitrile (Fig. 5.3c). As the
pH increases, the deprotonated form increases which has a lower saturation
capacity than the protonated form. For methanol, this gives deformed
peaks at high pH (Fig. 5.3d).

2.5
a) MeCN, pH = 7.6 4 b) MeOH, pH = 8.1
Concentration [g/L]

2
Experiments 3
1.5 BET model
2
1

0.5 1

0 0

2.5
c) MeCN, pH = 9.4 4 d) MeOH, pH = 9.5
Concentration [g/L]

2
Experiments
3
1.5 pH model

2
1

0.5 1

0 0
2 4 6 8 6 8 10 12 14
Time [min] Time [min]

Figure 5.3: Overloaded elution profiles at the a-b) lowest pH and c-d) highest pH
used in the experimental design obtained with either acetonitrile or methanol as
organic modifier at 30 ◦C.

To conclude, this paper investigated the adsorption mechanism for

omeprazole and how pH, temperature and type of organic modifier affected
it and how these effects translated to resolution and tailing. Adsorption
isotherm models were found to be valuable to understand the trends from
response surfaces obtained from DoE data.

37
5.2 Paper II

An assumption made when changing the particle size and column dimensi-
ons in chromatography, is that the stationary phase is the same. However,
by comparing columns packed with supposedly the same stationary phase
but different particle size (1.7 µm and 5 µm) using the Tanaka characteri-
zation protocol, significant selectivity shifts were observed with certain
column brands [17]. The hydrogen bonding capacity and the silanol activity
were two parameters that varied with the particle size and the result was
large shifts in retention, especially for basic compounds. Furthermore, the
temperature is often higher in UHPLC than in HPLC either due to viscous
heating or a combination of viscous heating and a higher set temperature
of the column oven to reduce the pressure drop over the column. Since
the temperature affects the adsorption isotherm parameters and hence the
retention, saturation capacity and peak shape, the higher temperature in
UHPLC may affect the separation performance.
Therefore the aim of paper [II] was to compare the adsorption isotherms
obtained at 20 ◦C with a column packed with 3.5 µm particles (denoted
condition I) with those obtained at 40 ◦C with a column packed with 1.7 µm
particles (condition II) from the same brand of stationary phase to evaluate
the effect of these changes on the adsorption mechanism. Adsorption
isotherms were acquired for one neutral (cycloheptanone, C7), one anionic
(sodium 2-naphtalenesulfonate, SNS) and one cationic (benzyltriethylam-
monium chloride, BTEAC) compound for the two conditions. The flow rate
was kept low with the 1.7 µm particles to avoid temperature and pressure
gradients because they complicates adsorption isotherm determination.
Additionally, we wanted to avoid studying too many factors at the same
time and pressure and temperature gradients were the subject of paper
[III].
The adsorption isotherms for BTEAC, C7 and SNS were determined
with the PP method and described by the bi-Langmuir model, Eq. (2.10)
(Table 5.1). The retention factor was lower in condition II for BTEAC
and SNS while it was, rather surprisingly, identical for C7. In RPLC, it is
expected that the retention factor decreases with temperature and the higher
temperature in condition II was believed to be the main reason for the lower
retention. The three compounds are affected differently by temperature,
with C7 being least affected according to additional experiments in paper
[III]. The total porosity was 0.60 in condition I and 0.55 for condition II,
i.e. the ratio between stationary phase and mobile phase was slightly

38
higher with the 1.7 µm particles, which increases the retention factor in
condition II. Therefore, the identical retention factors for C7 was most likely
a coincidence of the decrease in retention due to higher temperature and
the increase in retention due to lower porosity canceling each other out.
The saturation capacity, which can be seen as a measure of the number of
possible adsorption sites, was very similar for the two conditions while the
equilibrium constant, related to the strength of the interactions, decreased in
condition II. The equilibrium constant is related to the temperature through
Eq. (2.14) and is expected to decrease with increasing temperature as was
the case here. Based in these observations, the nature of the adsorption
mechanism seems very similar in the two conditions and no additional
interactions with, for example, residual silanols were evident.

Table 5.1: Adsorption isotherm parameters obtained at two conditions; I 20 ◦C and

3.5 µm particles and II 40 ◦C and 1.7 µm particles.

Solute Condition k a1 q s,1 b1 a2 qs,2 b2

I 10.5 3.72 18.8 0.198 11.8 3.07 3.82
BTEAC
II 9.33 3.22 20.6 0.157 8.34 2.67 3.12
I 5.94 1.38 50.3 0.0274 7.39 3.57 2.07
SNS
II 4.12 1.04 48.7 0.0213 4.09 2.64 1.55
I 4.21 3.93 40.2 0.0978 2.29 n.a. ≈0
C7
II 4.21 3.56 51.7 0.0688 1.68 n.a. ≈0

The adsorption isotherm for omeprazole (∼ 80% in protonated form1)

was described by the BET model, Eq. (2.11). Due to the limited solubility
of omeprazole in the eluent, there is not much curvature of the isotherm
but it is evident from the overloaded peaks that a type III model like
BET is suitable (Fig. 5.4a-b). The AED displayed one adsorption site for
both conditions which indicates homogeneous adsorption in the studied
concentration interval (Fig. 5.4c). The variation of the isotherm parameters
with acetonitrile was investigated (Fig. 5.4d-f) and was found to follow the
same trends in both conditions. The a parameter shows that the retention
time is slightly lower in condition II and that the difference increased at low
acetonitrile fractions. b S decreased with acetonitrile which was expected
since omeprazole becomes more soluble in the eluent as the acetonitrile
fraction increases. bL shows a maximum around 22% acetonitrile and it
was speculated that at lower fractions omeprazole binds very strongly to
1The paper erroneously states that omeprazole was negatively charged.

39
the stationary phase due to its low solubility in the eluent making it less
attractive to form multi-layers.

a) 3 b)
1.7 µm, 40°C
20
2 3.5 µm, 20°C
C [g/L]

q [g/L]
10
1

0 0
3 4 5 6 7 8 0 0.5 1 1.5 2 2.5
k C [g/L]
c) d) 4
4
3
q s [g/L]

2 ln(a) 2

0 1
-4 -3 -2 -1 0 0.18 0.22 0.26 0.3
ln(b) φ [v/v]
e) -6 f) -2

-7
ln(b S)

ln(b L )

-3

-8
-4
-9
0.18 0.22 0.26 0.3 0.18 0.22 0.26 0.3
φ [v/v] φ [v/v]

Figure 5.4: a) Overloaded omeprazole peaks normalized to retention factor. b)

Adsorption isotherms obtained with the ECP method at 25% acetonitrile. c) AED
calculations with the local BET model. d-f) Isotherm parameters in Eq. (2.11) as a
function of acetonitrile. Reprinted from D. Åsberg, et al., J. Chromatogr. A, 1362,
206–207, Copyright 2014, with permission from Elsevier.

We concluded that the adsorption isotherm model did not change

between condition I and II, but the values of the isotherm parameters did
change slightly. This was true for a wide range of acetonitrile fractions
since the isotherm parameters followed the same trend in both conditions
implying that geometric scaling in gradient elution should result in preser-
ved method performance. Since only minor differences in retention and
saturation capacity were observed for this experimental system, we should
not expect any complications related to stationary phase properties when
switching from 3.5 µm BEH C18 particles to the 1.7 µm variant of the same
particles.

40
5.3 Paper III

The elevated pressure and viscous heating in UHPLC may affect the perfor-
mance of a method when converting it from HPLC to UHPLC. The pressure
and temperature effects are convoluted under normal operation and only the
net effect is observed. In this paper we studied the individual contributions
from pressure and temperature to the overall change in retention and peak
shape. The peak symmetry as a function of pressure and temperature
was investigated with the stochastic model for BTEAC which was the only
compound displaying tailing at low concentrations. The pressure was
studied by using a low flow rate and a restrictor capillary installed after
the columns to avoid causing viscous heating. Different pressures were
obtained by varying the length of the capillary while keeping the flow rate
constant and the main part of the pressure drop was then in the capillary.
The retention factor of omeprazole and the same model compounds as in
paper [II] were determined at five pressures with the result for omeprazole
presented in Fig. 5.5a. The increase in retention for a 500 bar pressure
difference was about 40% for omeprazole which is quite significant2. A
20 ◦C increase in temperature resulted in a decrease in retention of around
20% for omeprazole (Fig. 5.5b). The temperature profile in the column was
calculated with the energy balance, Eq. (2.4), and validated through external
temperature sensors attached to the column surface. With the column in a
conventional still air oven, a 782 bar pressure drop over the column and
a flow rate of 1.2 mL/min, the temperature gradient in the axial direction
was 16 ◦C and the radial temperature gradient was 2 ◦C at most (Fig. 5.5c).
The relative effect of pressure and temperature on retention for ome-
prazole and the three model compounds was estimated by comparing the
retention factors at flow rates 0.13 mL/min and 1.2 mL/min. In Fig. 5.5d
the retention factor at 0.13 mL/min is taken as the reference value where
both pressure and viscous heating are negligible. The estimated change in
retention due to pressure (+300 bar) and due to temperature (8.5 ◦C) are
shown together with the experimentally obtained net effect. Even though
the individual effects are relatively large, they cancel each other out for
C7 and SNS and there is only a minor shift in retention when comparing
the experimental data. However, for OM and BTEAC the pressure effect
dominates with the retention factor increasing 10-15% at 1.2 mL/min. Since
2Note that there is an error in the abstract of this paper; the percent of increase in
retention is for a 300 bar pressure increase (corresponding to Fig. 5 in paper [III]), not 500
bar as is specified in the abstract.

41
there are large differences between the compounds it stands to reason that
selectivity shifts could occur which could cause problems when switching
between HPLC to UHPLC conditions. A case where this happened was
later demonstrated in paper [V].

Figure 5.5: Change in the retention factor of omeprazole as a function of a) pressure

and b) temperature. c) The temperature profile inside the column at 1.2 mL/min
(pressure drop 782 bar) with the oven at 40 ◦C. d) The calculated change in k due
to pressure (gray) and viscous heating (orange) when increasing the flow rate from
0.13 mL/min to 1.2 mL/min. The blue bars represent the experimentally obtained
net effect. Reprinted from D. Åsberg, et al., J. Chromatogr. A, 1401, 52–59, Copyright
2014, with permission from Elsevier.

BTEAC was tailing significantly even at low concentrations and the

elution profiles could be described by a 2-site stochastic model. The tailing
of a peak is related to the skew of the statistical distribution describing the
peak and the skew was studied as a function of temperature and pressure
for BTEAC (Fig. 5.6). The stochastic model revealed one adsorption site
with fast adsorption-desorption in the millisecond scale and n1 ≈ 10 000
which is common for symmetrical peaks in RPLC. The tailing was caused
by a second adsorption site which had slow adsorption-desorption with
the solute staying adsorbed up to several seconds before desorption. The
sojourn time, τs,2 , decreased with temperature for this type of site (Fig. 5.6a)
probably due to a faster diffusion at high temperature and a decrease in the

42
association equilibrium constant with temperature, Eq. (2.14). Increasing
the pressure resulted in a longer sojourn time (Fig. 5.6b) which can be
explained by results obtained in paper [VI] and will be discussed in relation
to that paper. The practical consequence is that the skew decreases with
temperature and increases with pressure which can be seen from the
experimental elution profiles in Fig. 5.6. That pressure could affect peak
symmetry had not been reported before and demonstrated that the pressure
effect is not completely understood in liquid chromatography.

7
a) Experiment b)
6
Linear fit
τs,2 [s]

3
30 35 40 45 50 0 200 400 600
Temperature [°C] Pressure [bar]
1.4
c) d)
1.35
Skew

1.3

1.25
30 35 40 45 50 0 200 400 600
Temperature [°C] Pressure [bar]

e) f) 524 bar
50°C

40°C 301 bar

30°C 51 bar

4 5 6 7 8 5 6 7 8 9
Time [min] Time [min]

Figure 5.6: Results from the stochastic model for BTEAC. The sojourn time for the
adsorption site with slow kinetic, τs,2 , as a function of a) temperature (pressure ca
50 bar) and b) pressure (temperature 40 ◦C). The skew is shown in c) as a function
of temperature and in d) as a function of pressure. e) Chromatograms with varying
temperature and f) with varying pressure. Reprinted from D. Åsberg, et al., J.
Chromatogr. A, 1401, 52–59, Copyright 2014, with permission from Elsevier.

43
5.4 Paper IV

Quality control (QC) methods plays an important role in the overall control
strategy of drug manufacturing. However, efficient life-cycle management
with continual improvements of QC methods are hindered by the lack of an
internationally harmonized legislation for handling post-approval changes.
Currently, a change to a QC method need to be approved by all regulatory
agencies where the drug product is marketed before it can be implemented
which is very time consuming and may result in the QC methods falling
behind the technical development. Fortunately, the ICH Q12 guidelines
are under development and set to be released in 2017 with the purpose
to harmonize the post-approval legislation and facilitate efficient life-cycle
management of both the manufacturing process and analytical procedures.
The aim of this paper was to develop a concept exemplifying how a post-
approval change to a QC method could be handled in practice with minimal
interaction with regulatory agencies before implementation.
We based the concept on QbD principles, but the possibility to do
changes inside the original MODR was too limited since it is seldom
possible to predict what flexibility is needed for implementing future
technology. The idea was that for a certain analytical technique, the
specified performance criteria, failure modes and controls would be valid
also after a change and a revalidation would ensure the preserved quality
of the method. The change and revalidation data would be recorded within
the company and presented upon inspection in a “do and tell” manner.
We distinguish between an ATP as defined in the QbD framework as a set
of performance criteria that are independent of the analytical technique
(measuring principle) and a method ATP (mATP) which we define as a set
of performance criteria for a specific technique, e.g. liquid chromatography.
The concept is demonstrated by a case study in which a QC method for
the drug product Nexium based on HPLC is converted to UHPLC as a
continual improvement strategy.

5.4.1 Method Development

The objective of the analytical procedure is specified along with the pro-
perties that need to be measured. The objective in the case study was to
determine the impurity and the API content in the drug product Nexium.
To this end, the concentration of omeprazole and four impurities needed to
be measured in a Nexium sample with HPLC. A set of method performance
criteria were determined which should be met for the method to give data

44
of sufficient quality. The mATP for the HPLC method is summarized
in Table 5.2 and includes criteria for specificity, accuracy, reproducibility,
quantification limit, linearity and relative response. After the mATP was
established and a suitable working point was found, it was validated and
the result is shown for omeprazole and one impurity in Table 5.2.

Table 5.2: mATP with method performance criteria and some results from the
HPLC method. The specificity was determined from the resolution between
the closest eluting peaks while the reproducibility was assessed by the relative
standard deviation of total impurity area for the same sample analyzed in two
laboratories.

Attribute Criteria Omeprazole H431/41

Specificity No interference R s  9.5 for closest peaks
Reproducibility ≥ 5.0% 2.56% in impurity area
Accuracy 90 − 110% 99.6% 103%
Quantification limit ≤ 0.05% 0.0085% 0.0064%
Linearity ≥ 0.9990 1.0000 1.0000
Relative response ≥ 0.7 1.00 1.29

5.4.2 Risk Assessment

The objective of the risk assessment is to identify critical factors that have a
serious impact on the method performance. Failure modes were identified
as the resolution between the critical peak pair being below 2, the tailing
factor of the main peak being outside the range 0.75-1.50, the first peak
eluting in the void and the retention factors for the last eluting peak being
above 30, corresponding to the method run time. Based on previous
experience of this type of sample, e.g. from paper [I], and experiments done
during the working point screening, the eluent pH, the eluent composition
and the column temperature was identified as critical factors. The failure
modes were included both in the robustness study of the MODR and in the
SST.

5.4.3 Method Validation

At this point, the method is known to meet the mATP criteria in the working
point and critical factors influencing the performance has been identified
along with appropriate failure modes. The main objective in the method
validation step is therefore to increase the knowledge of how the critical

45
factors affect the failure modes and determine the MODR. An expanded
robustness study was undertaken using DoE and statistical modeling. The
response surfaces for the method’s failure modes are shown in Fig. 5.7.
When more than two factors were significant, the settings giving the worst
case are presented. Only the resolution factor was outside specifications
in the design region (shown as the white dotted line in Fig. 5.7). The
MODR was therefore pH = 7.65-8.35 and T = 15.0-25.0◦C and the method
is specified in Table 5.3 with the tolerance levels for the the critical factors.
As for the understanding of the relationship between the critical factors
and the investigated responses, the trends are evident from Fig. 5.7 while
the physical explanation is discussed in depth in paper [I]. The resolution
factor was largely governed by the difference in pKa between omeprazole
and H168/66 while the fronting of omeprazole was due to a larger fraction
of omeprazole becoming deprotonated at high pH.

Table 5.3: The HPLC method given with the tolerance levels of the critical factors.

Column BEH C18, 3.5 µm, 4.6 × 100 mm

Buffer Phosphate, pH 8.0±0.35, 10 mM (A) and 3 mM (B)
Eluent A 10/90±5, v/v, acetonitrile/phophate buffer
Eluent B 80/20±5, v/v, acetonitrile/phophate buffer
Detection UV 302 nm
Gradient Linear from 0 to 100% B in 30 min
Gradient dwell time 1.04 min
Flow rate 1.00 mL/min
Temperature 20±5◦C
Injection volume 20 µL

5.4.4 Control Strategy

The control strategy ensures adequate performance of the method during its
lifetime, typically in the form of a system suitability test (SST). By carefully
constructing the SST so that it is also applicable outside the original MODR,
it can be used to assess the failure modes after a change to the method.
When converting the method to UHPLC we move outside the original
MODR and the SST should ensure method performance and generation of
reliable results.
The SST for the case study was taken to include the responses in the
robustness study as they were related to the failure modes and the system

46
Figure 5.7: Response surfaces for the HPLC robustness study. a) The resolution
factor between omeprazole and H168/66 with the robust region (the lower 95%-
confidence limit of the predicted resolution factor > 2) above the white dotted line.
b) The tailing factor of omeprazole. c) The apparent retention factor for H431/41
which is the first eluting peak. d) The apparent retention factor for H168/22 which
is the last eluting peak. Reprinted from D. Åsberg, et al., J. Pharm. Biomed. Anal.,
129, 273–281, Copyright 2016, with permission from Elsevier.

repeatability. The omeprazole peak was slightly fronting, with a tailing

factor of 0.93, and the resolution factor between omeprazole and H168/66
was 9.5 which is well above the acceptance criteria.

5.4.5 Continual Improvement

As a continual improvement of the method, the column is exchanged to an

UHPLC column (1.7 µm, 2.1 × 50 mm) and the temperature is increased
from 20 ◦C to 40 ◦C. The UHPLC column will increase the throughput by
shortening the analysis time while the higher temperature will reduce the
pressure and increase the resolution between omeprazole and H168/66
(Fig. 5.7a). This change is outside the original MODR and would normally
require the submission of a post-approval change to the regulators before
implementing it. However, what is proposed in the new concept is that the
change would be implemented directly and managed according to internal
change control procedures and presented upon inspection. The original

47
submission would specify how re-validation and method equivalence would
be handled after a minor change.
The original method was geometrically scaled to give the same separa-
tion with the new column and the resulting chromatograms are compared
in Fig. 5.8. From paper [II] we know that the adsorption mechanism are
almost unaffected when changing the particle size of the stationary phase.
When performing the geometrical scaling and keeping the temperature the
same, Fig. 5.8a, the retention increased slightly in the modified method
most likely due to a pressure increase [III]. However, when also increasing
the temperature, Fig. 5.8b, the retention factors decreased somewhat and
the selectivity between omeprazole and H168/66 improved. The analysis
time was reduced by a factor 4 and the solvent consumption was reduced by
a factor 10 when switching to UHPLC. The mATP needs to be re-validated
Normalized Response

a) HPLC
UHPLC
Normalized Response

b)

0 2 4 6 8 10 12 14 16
Apparent retention factor

Figure 5.8: Chromatograms from the original and modified QC method for a)
the same oven temperature, 30 ◦C, and b) the temperatures specified in the QC
methods, i.e. 20 ◦C and 40 ◦C. Reprinted from D. Åsberg, et al., J. Pharm. Biomed.
Anal., 129, 273–281, Copyright 2016, with permission from Elsevier.

for the modified method and this was achieved with similar results as for
the original method. The failure modes are equally valid for the modified
method and was not changed. Next, the MODR was adjusted based on the
settings of the modified method as center point by repeating the robustness
study. Since the resolution had improved, all of the design region was

48
now robust. The SST is still relevant for testing the failure modes and
the modified method passed the SST. As a test for method equivalence,
the original and modified methods were demonstrated to give the same
impurity area in the center point and the extreme points of the experimental
design.

5.4.6 Conclusions

The proposed concept enables implementation of minor post approval chan-

ges within the same measuring principle without prior regulatory approval.
The mATP, failure modes and controls given in the original drug application
are valid also for the modified method and a complete revalidation of
these elements are performed within the company’s pharmaceutical quality
system and the result archived for future regulatory inspections. If the
presented concept will be reality it means that the pharmaceutical industry
can save time and resources and still maintain high quality QC-methods
while it reduces the burden for regulatory agencies.

5.5 Paper V

We investigated the effect of elevated pressure and viscous heating in paper

[III] and showed that it could complicate method conversion between
HPLC to UHPLC. Here, a case where the selectivity, and hence resolution,
between omeprazole and the impurity H168/66 is seriously affected by
pressure and temperature gradients is presented. At the flow rate 0.08
mL/min omeprazole and H168/66 co-elutes while increasing the flow rate
to 0.50 mL/min results in baseline separation (Fig. 5.9a). This selectivity
shift is mainly due to a change in temperature caused by viscous heating
which is evident from an experimental design in pressure and temperature
space (Fig. 5.9b). In this case, the flow rate could be identified as a critical
factor when, in reality, it is temperature that is the underlying cause of the
selectivity shift.
The aim of this paper was to develop an easy-to-use, empirical method
to predict retention time shifts due to pressure and temperature gradients
inside the column. Such a method would be a valuable tool in routine met-
hod development to assess method robustness and aid method conversion
between HPLC to UHPLC. The strategy was based on the assumption that
by putting the column in a water jacket the generated heat was effectively
removed and the temperature could be kept nearly constant regardless of
flow rate. This assumption was investigated by calculating the temperature

49
Figure 5.9: The effect of flow rate on the selectivity between omeprazole (largest
peak) and H168/66. a) Chromatograms obtained at three flow rates giving average
column pressures of 93, 300 and 484 bar. The column oven was set to 308 K. b) The
response surface for selectivity in pressure and temperature space using a restrictor
capillary to adjust the pressure. Reprinted from D. Åsberg, et al., J. Chromatogr. A,
1479, 107–120, Copyright 2017, with permission from Elsevier.

profiles, in the same way as in paper [III], with the column kept in a still
air oven and in a water jacket. In still air, the axial temperature gradient
dominated and the temperature increased about 10 ◦C from the inlet to the
outlet of the column (Fig. 5.10a). When keeping the column in the water
jacket, the radial temperature gradient dominated and was about 0.7 ◦C
resulting in a case where the column could be controlled within ±1 ◦C
(Fig. 5.10b). In agreement with Fig. 5.9, no change in selectivity with flow
rate was observed when keeping the column in the water jacket since the
temperature could be kept nearly constant. There is also a clear difference
between water and air thermostating when plotting the retention factor as a
function of average column pressure controlled by the flow rate. In a water
bath, the increase in retention factor is solely due to the increase in pressure
while in the still air oven the retention factor increases more moderately
since an axial temperature gradient is also present (Fig. 5.10c-d). This is
another approach than the one in paper [III] to de-couple the pressure and
temperature effects in UHPLC.
The empirical approach describes the retention factor as a function of
average pressure, Pavg , and the average temperature, Tavg , in the column as

k(T, P)  a1 (Tavg − Tavg,0 ) + a2 (Pavg − Pavg,0 ) + k 0 (5.1)

where a i denotes empirical fitting parameters and subscript 0 denotes a

reference condition where the pressure drop over the column is around 100

50
Figure 5.10: Calculated temperature profiles in a) still air (522 bar) and b) water
jacket (494 bar) for the BEH column. Normalized retention factor for c) omeprazole
and d) H168/66. Dashed line is a linear fit to the experimental data (symbols).
Reprinted from D. Åsberg, et al., J. Chromatogr. A, 1479, 107–120, Copyright 2017,
with permission from Elsevier.

bar. The DoE investigation in pressure-temperature space confirmed that

Eq. (5.1) is a valid approximation of how the retention factor varies. Pavg
and Tavg are functions of flow rate:

Pavg (Fv )  a3 (Fv − Fv,0 ) + Pavg,0 (5.2)

Tavg (Fv )  a4 (Fv − Fv,0 ) + Tavg,0 (5.3)

The a i parameters have to be determined experimentally in four experiments:

i The first experiment is used to describe a reference condition where
the pressure and temperature gradients are negligible. It is an injection
at a low flow rate where the back pressure is around 100 bar.

ii The second experiment is for determining how the retention factor

depends on temperature. It is an injection at low flow rate, as in (i),
but with the column oven set to a temperature 20 ◦C higher than in
the first experiment.

51
 iii The third experiment is for determining how the retention factor
depends on pressure. It is an injection at high flow rate (back pressure
around 1000 bar) with the column placed in a water jacket. The
water jacket minimizes the axial temperature gradient and keeps the
temperature ±1 ◦C so the pressure effect can be measured.

iv The final experiment estimates the pressure and the average column
temperature as a function of flow rate with the column in the standard
still air oven. It is done at high flow rate using the column oven with
the same set temperature as for the reference condition in step (i).

The empirical approach was validated by predicting the retention factor for
omeprazole and three impurities at four flow rates and with two columns
(one with fully porous particles and one with core-shell particles). Rather
surprisingly, considering the approximations done deriving it, the relative
error between the predicted and experimental retention factor was always
less than 1% and in many cases below 0.5%. This level of accuracy is
enough to predict the selectivity shifts observed between omeprazole and
H168/66 (Fig. 5.11). To predict the elution profiles, the column efficiency
was estimated with linear interpolation of the elution profiles in step (i) and
(iv) which was found to be good enough.

a) 0.2 mL/min b) 0.5 mL/min

Prediction
Experiment

7.6 7.8 8 8.2 8.4 8.6 3.4 3.6 3.8

Time [min] Time [min]

Figure 5.11: Predicted and experimental chromatograms for a) 0.20 mL/min and
b) 0.50 mL/min for omeprazole and H168/66. Reprinted from D. Åsberg, et al., J.
Chromatogr. A, 1479, 107–120, Copyright 2017, with permission from Elsevier.

The empirical approach has the potential to be a practical tool in routine

method development, but there are still some parts left to investigate. While
gradient elution is very common in pharmaceutical analysis, the empirical
approach in its present form assumes isocratic elution where the eluent
viscosity is constant during the run. Extending it to gradient elution would

52
not be straightforward since there will exist different cases depending on
the viscosity of eluent A and B and was therefore outside the scope of this
paper. Solutes, like proteins and large peptides, which are more sensitive to
pressure and temperature would also be interesting to try in the empirical
approach to see if the linear relationship in Eq. (5.1) still holds.

5.6 Paper VI

A retention mechanism in RPLC taking the pressure into account has mainly
been studied by measuring the retention factor at different pressure levels.
However, in such measurements it is difficult to obtain information about
secondary interactions between the solute and the stationary phase which
are often responsible for tailing peaks. With nonlinear adsorption isotherm
measurements on the other hand, it is possible to elucidate secondary
interactions.
The aim of this paper was to determine adsorption isotherms at different
pressures ranging from 100 to 1000 bar and measure how the different
adsorption sites contribute to the overall pressure dependence of the solute.
More specifically, we were interested in why ionic solutes are more affected
by pressure than neutral solutes and why BTEAC in paper [III] became
more tailing as the pressure increased. To this end, a similar approach as
the one in paper [II] was employed where the adsorption isotherms were
determined for three compounds; one cation (BTEAC), one anion (SNS) and
one neutral (antipyrine, AP), but here the pressure was varied instead. AP
replaced cycloheptanone since it had similar molecular weight to BTEAC
and SNS. The pressure was controlled by the same approach as in paper
[III], i.e. by varying the length of a restrictor capillary attached after the
column while keeping the flow rate constant.
The total porosity of the stationary phase is an important property when
deriving adsorption isotherms from experimental data and it was therefore
measured at all pressure levels. A change from  t  0.552 to t  0.560 was
noted when the pressure increased from 100 to 900 bar mostly because of a
compression of the C18-bonded layer of the stationary phase. A variation
in total porosity in this range is negligible for most routine applications
but should be accounted for when determining adsorption isotherms. All
three compounds were found to be best described by the bi-Langmuir
adsorption isotherm when evaluating different models following a rigorous
approach with Scatchard plots, AED, model selection and verification by
predicting elution profiles and comparing them to experimental ones. For

53
AP, the saturation capacity for the high energy site was around 1% of
the total saturation capacity and the equilibrium constant was about ten
times higher compared to the low energy site. For BTEAC and SNS, the
saturation capacity for the high energy site was instead around 10% of the
total saturation capacity but the equilibrium constant was still about ten
times higher for the high energy site.
Two adsorption sites prevailed for all pressure levels in the AED
(Fig. 5.12) which indicates that the adsorption isotherm model per se does
not change with pressure, only the numerical parameters do. The ad-
sorption isotherms tended towards a slightly increased saturation capacity
and initial slope with increasing pressure. Based on previous studies (see
Section 2.3) we interpreted the low energy site as adsorption located at
the eluent-C18 layer interface while the high energy represents adsorption
deeper into the C18 layer. Some trends were evident from the isotherm pa-
rameters, although their exact values should be viewed critically since they
are obtained through nonlinear regression which can yield multiple sets of
parameters giving the same functional values. The association equilibrium
constant increases with increasing pressure for both sites most likely due
to the increased pressure driving the equilibrium between the mobile and
stationary phase towards the stationary phase to reduce the solute molar
volume in accordance with Eq. (3.5). The saturation capacity of the low
energy site is almost unchanged, which is to be expected since the surface
area of the stationary phase is not affected by pressure. The saturation
capacity of the high energy site increased with pressure and molecular
simulations have demonstrated an increased de-wetting of a C18 stationary
phase with increasing pressure [126]. We speculated that the increase in
saturation capacity was caused by a de-wetting of the stationary phase
which made adsorption sites near the silica surface, previously occupied by
water, available [127].
The change in partial molar volume of the solute associated with the
adsorption of each of the two sites were derived from the adsorption
isotherms using a combination of Eq. (3.5) and Eq. (2.9) for each site. For
the low energy site the solute molar volume was practically the same for the
three compounds and equal to ∆Vlow ≈ −6 cm3 /mol. For SNS and BTEAC,
the high energy site had almost twice as large volmue change while AP
displayed a 3.6 times larger volume change. Based on these results, a more
nuanced retention mechanism accounting for pressure was proposed. Polar
and ionic solutes will have a surrounding polar solvation layer in the mobile
phase and lose parts of this layer when adsorbing on the stationary phase.

54
 12 60

10
a) b)
8 0.2 40
q s [mM]

q [mM]
6
109 bar
0.1
4 316 bar
20
462 bar
2 0 612 bar
6 7 8 919 bar
0 0
3 4 5 6 7 8 9 10 11 12 0 1 2 3
12

1
c) 10
b)
0.8 8
q s [mM]

q [mM]
0.6 6 88 bar
0.4 263 bar
4
378 bar
0.2 2 508 bar
808 bar
0 0
4 5 6 7 8 9 10 11 12 13 0 0.5 1
2 12
e) 10
f)
0.2
1.5
8
q s [mM]

0.1
q [mM]

1 6
107 bar
0 311 bar
7 8 9 4
0.5 445 bar
2 590 bar
900 bar
0 0
4 5 6 7 8 9 10 11 12 13 0 0.5 1
ln(b) C [mM]

Figure 5.12: AED and adsorption isotherm for a-b) AP, c-d) BTEAC and e-f) SNS.
Reprinted from D. Åsberg, et al., J. Chromatogr. A, 1457, 97–106, Copyright 2016,
with permission from Elsevier.

It is the loss of this solvation layer that causes the change in the solute’s
molar volume. When the solute adsorbs on the low energy site located near
the C18 surface, only a small fraction of the solvation layer is lost resulting
in a small reduction of the molar volume. Adsorption on the high energy
sites, on the other hand, forces the solutes to penetrate into the C18 layer

55
and a larger portion of the solvation layer is lost making the molar volume
of the solute decrease more (Fig. 5.13). So, a solute with more adsorption
events at the high energy sites will have a retention factor that are more
sensitive to pressure.
Returning to paper [III], we are now better equipped to understand why
BTEAC exhibited increased tailing at higher pressures. We can rule out a
decreasing saturation capacity with pressure based Fig. 5.12 and calculations
presented in paper [VI] showed no reduction in column efficiency with
pressure due to overloading. If we assume that the nature of the adsorption
sites observed in the stochastic modeling and in the adsorption isotherm
model are of similar nature, then the increase in sojourn time for the high
energy site (Fig. 5.6b) can be correlated to the increase in the association
equilibrium constant with pressure for this site. To conclude, an increase in
pressure causes the equilibrium constant for the high energy site to increase
which in turn results in a longer sojourn time in this type of site. The longer
sojourn type increased the skew of the peak and hence a larger tailing factor
was observed.

Figure 5.13: Schematic illustration of adsorption from the mobile phase on to

the stationary phase for a low-energy site at the eluent-C18 layer interface (left)
and a high-energy site near the silica surface (right). The decrease of ∆V for the
solute upon adsorption on the high-energy site is explained by the larger loss of
its solvation layer when it penetrates deeper into the C18 layer. Reprinted from
D. Åsberg, et al., J. Chromatogr. A, 1457, 97–106, Copyright 2016, with permission
from Elsevier.

56
Chapter 6
Concluding Remarks

UHPLC, with its higher throughput and lower solvent consumption, has
been a great leap forward in the evolution of liquid chromatography.
However, the implementation of UHPLC in pharmaceutical analysis has not
been straightforward and there are both technical and regulatory challenges
to overcome. The goal of the thesis has been to study UHPLC in depth and
thereby solve these challenges. My approach was to implement adsorption
isotherms and mechanistic modeling to complement conventional DoE and
statistical regression models to satisfy the trend towards a more science-
based approach to pharmaceutical method development. I demonstrated
that the adsorption mechanism and the surface heterogeneity were not
seriously affected when switching from a HPLC particle size to a UHPLC
particle size using adsorption isotherm modeling. This is an important
finding since the assumption that the stationary phase remains unaltered is
a requirement for reliable geometrical scaling between particle sizes.
Elevated pressure and temperature gradients due to viscous heating are
the two most important factors to account for to achieve predictable method
conversion between HPLC and UHPLC. Their effect on retention and peak
shape was quantified and studied from a fundamental point of view to
investigate how they affected the adsorption mechanism and separation
performance. I found that a secondary, high energy, type of adsorption site
was the main reason for the pure pressure effect. Ionic solutes adsorbed to
a larger degree on the secondary sites and were hence more influenced by
pressure than neutral solutes. I also demonstrated that the combination
of pressure and viscous heating could cause selectivity shifts since the
retention of solutes, even with similar structure, were affected differently.
Therefore, an easy-to-use, experimental approach intended for method
development was developed to predict these selectivity shifts. Together,
these studies provides a better understanding of UHPLC and demonstrates

57
for the regulatory agencies an approach for studying and controlling critical
factors in UHPLC.
To change from HPLC to UHPLC in the control strategy of a marketed
drug products, a post approval change would need to be submitted and
approved by the regulatory agencies, which is time consuming. Together
with AstraZeneca, I developed a concept to tackle this issue and proposed
an approach for handling minor post approval changes such as this one
without prior regulatory approval. The idea was that as long as the same
analytical technique is used, the method controls (SST and mATP) and
failure modes would be relevant also for settings not included in the original
submission and by re-validating the controls, the method performance
would not be compromised.
The current ICH Q12 draft document provides a framework for post
approval changes in a more predictable and efficient manner across the pro-
ducts life cycle by making it more feasible to do changes with a downgraded
reporting category under the company’s pharmaceutical quality system
[128]. By demonstrating a good product and process knowledge, only
critical input and output parameters affecting the process performance and
product quality need regulatory approval prior to a change, i.e. noncritical
parameters may be changed without prior regulatory interactions. The
elements in the submission that may not be changed without approval
should include a method description with the critical aspects for the in-
tended performance (similar to the mATP) rather than detailed operating
conditions, along with parameters verifying the method continue being fit
for purpose (a SST). Changes to analytical methods for currently marketed
products can also be handled under ICH Q12 in a manner strikingly similar
to that presented paper [IV]. The method is shown to be fit for purpose
by revalidating it according to ICH Q2 (i.e. revalidating the mATP), the
SST is verified and the results are shown to be equivalent or better. If this
is done the change can be made with downgraded regulatory reporting.
AstraZeneca has already begun implementing such an approach for ana-
lytical methods in some of their drug applications, for example with the
drug product naloxegol (Movantik) which was recently approved by the
FDA and the EMA.
Drug manufacturing will be relaying more on science and risk-based
approaches in the coming years where the final quality of the product is put
in focus. I hope that this thesis can provide support for this trend and the
implementation of the ICH Q12 guideline while enabling implementation
of new technology, such as UHPLC, in pharmaceutical analysis.

58
Acknowledgment

Many people have supported me throughout my work with this thesis and
I am very grateful to

my main supervisor Torgny Fornstedt and my assistant supervisor Jörgen

Samuelsson for your support, inspiration and all our fruitful discussions,

Anders Karlsson for initializing this project and patiently teaching me about
the pharmaceutical industry,

Martin Enmark for great cooperation and nice company in Uppsala,

Krzysztof Kaczmarski and Marek Leśko for a long and fruitful collaboration,

all of my coauthors for your effort, expertise and insightful comments,

Mikael Andersen for your invaluable help in the lab and

my past and present colleagues at the department for your pleasant company.

59
References

[1] S. Cox Gad. Pharmaceutical Manufacturing Handbook: Regulations and

Quality. Hoboken: Wiley, 2008. doi: 10.1002/9780470259832
[2] The European Union. Commission Directive 91/507/EEC. Official
Journal of the European Communities, 34: 32–52, 1991.
[3] S. Görög. The changing face of pharmaceutical analysis. TrAC, 26:
12–17, 2007. doi: 10.1016/j.trac.2006.07.011
[4] International Council for Harmonisation. Q2(R1): Validation of Ana-
lytical Procedures: Text and Methodology. Nov. 2005. url: www.ich.org
[5] C. F. Richardson and F. Erni. “Regulatory Considerations in HPLC
Analysis”. In: Handbook of Pharmaceutical Analysis by HPLC. Ed. by
M. W. Dong and S. Ahuja. Vol. 6. Separation Science and Technology.
Academic Press, 2005. doi: 10.1016/S0149-6395(05)80054-6
[6] R. Holm and D. P. Elder. Analytical advances in pharmaceutical
impurity profiling. Eur. J. Pharm. Sci., 87: 118–135, 2016. doi: 10.1016/
j.ejps.2015.12.007
[7] J. De Vos, K. Broeckhoven, and S. Eeltink. Advances in Ultrahigh-
Pressure Liquid Chromatography Technology and System Design.
Anal. Chem., 88: 262–278, 2016. doi: 10.1021/acs.analchem.5b04381
[8] B. Debrus, E. Rozet, P. Hubert, J.-L. Veuthey, S. Rudaz, and D.
Guillarme. “Method Transfer Between Conventional HPLC and
UHPLC”. In: UHPLC in Life Sciences. Ed. by D. Guillarme and J.-
L. Veuthey. The Royal Society of Chemistry, 2012. doi: 10.1039/
9781849735490
[9] G. K. Webster, T. F. Cullen, and L. Kott. “Method Transfer Between
HPLC and UHPLC Platforms”. In: Ultra-High Performance Liquid
Chromatography and its Applications. Ed. by Q. A. Xu. Wiley, 2013. doi:
10.1002/9781118533956

60
[10] S. Fekete, J.-L. Veuthey, and D. Guillarme. Comparison of the most
recent chromatographic approaches applied for fast and high reso-
lution separations: Theory and practice. J. Chromatogr. A, 1408: 1–14,
2015. doi: 10.1016/j.chroma.2015.07.014
[11] J. R. Mazzeo, U. D. Neue, M. Kele, and R. S. Plumb. Advancing LC
performance with smaller particles and higher pressure. Anal. Chem.,
77: 460 A–467 A, 2005. doi: 10.1021/ac053516f
[12] M. W. Dong and K. Zhang. Ultra-high-pressure liquid chromato-
graphy (UHPLC) in method development. TrAC, 63: 21–30, 2014.
doi: 10.1016/j.trac.2014.06.019
[13] J. Cielecka-Piontek, P. Zalewski, A. Jelińska, and P. Garbacki. UHPLC:
The Greening Face of Liquid Chromatography. Chromatographia, 76:
1429–1437, 2013. doi: 10.1007/s10337-013-2434-6
[14] S. A. C. Wren and P. Tchelitcheff. Use of ultra-performance liquid
chromatography in pharmaceutical development. J. Chromatogr. A,
1119: 140–146, 2006. doi: 10.1016/j.chroma.2006.02.052
[15] S. Fekete, J. Schappler, J.-L. Veuthey, and D. Guillarme. Current and
future trends in UHPLC. TrAC, 63: 2–13, 2014. doi: 10.1016/j.trac.
2014.08.007
[16] D. V. McCalley. The impact of pressure and frictional heating on
retention, selectivity and efficiency in ultra-high-pressure liquid
chromatography. TrAC, 63: 31–43, 2014. doi: 10.1016/j.trac.2014.06.
024
[17] P. Petersson and M. R. Euerby. Characterisation of RPLC columns
packed with porous sub-2 µm particles. J. Sep. Science, 30: 2012–2024,
2007. doi: 10.1002/jssc.200700086
[18] C. G. Horvath, B. A. Preiss, and S. R. Lipsky. Fast liquid chromato-
graphy. Investigation of operating parameters and the separation of
nucleotides on pellicular ion exchangers. Anal. Chem., 39: 1422–1428,
1967. doi: 10.1021/ac60256a003
[19] N. Tanaka and D. V. McCalley. Core–Shell, Ultrasmall Particles,
Monoliths, and Other Support Materials in High-Performance Liquid
Chromatography. Anal. Chem., 88: 279–298, 2016. doi: 10.1021/acs.
analchem.5b04093
[20] G. Guiochon and F. Gritti. Shell particles, trials, tribulations and
triumphs. J. Chromatogr. A, 1218: 1915–1938, 2011. doi: 10.1016/j.
chroma.2011.01.080

61
[21] J. Ruta, D. Zurlino, C. Grivel, S. Heinisch, J.-L. Veuthey, and D.
Guillarme. Evaluation of columns packed with shell particles with
compounds of pharmaceutical interest. J. Chromatogr. A, 1228: 221–
231, 2012. doi: 10.1016/j.chroma.2011.09.013
[22] M. Biba, C. J. Welch, J. P. Foley, B. Mao, E. Vazquez, and R. A. Arvary.
Evaluation of core–shell particle columns for ion-pair reversed-phase
liquid chromatography analysis of oligonucleotides. J. Pharm. Biomed.
Anal., 72: 25–32, 2013. doi: 10.1016/j.jpba.2012.09.007
[23] J. Fibigr, D. Šatínský, L. Havlíková, and P. Solich. A new method
for rapid determination of indole-3-carbinol and its condensation
products in nutraceuticals using core–shell column chromatography
method. J. Pharm. Biomed. Anal., 120: 383–390, 2016. doi: 10.1016/j.
jpba.2015.12.039
[24] J. Fibigr, D. Šatínský, and P. Solich. A study of retention characte-
ristics and quality control of nutraceuticals containing resveratrol
and polydatin using fused-core column chromatography. J. Pharm.
Biomed. Anal., 120: 112–119, 2016. doi: 10.1016/j.jpba.2015.12.014
[25] J. W. Ludvigsson, A. Karlsson, and V. Kjellberg. Core–shell co-
lumn Tanaka characterization and additional tests using active
pharmaceutical ingredients. J. Sep. Sci., 39: 4520–4532, 2016. doi:
10.1002/jssc.201600769
[26] R. W. Bondi Jr. and J. K. Drennen III. “Quality by Design and the
Importance of PAT in QbD”. In: Handbook of Modern Pharmaceutical
Analysis. Ed. by S. Scypinski and S. Ahuja. Vol. 10. Separation Science
and Technology. Academic Press, 2011. doi: 10.1016/B978- 0- 12-
375680-0.00005-X
[27] M. M. Dantus and M. L. Wells. Regulatory Issues in Chromatographic
Analysis in the Pharmaceutical Industry. J. Liq. Chromatogr. Related
Technol., 27: 1413–1442, 2004. doi: 10.1081/JLC-120030610
[28] U.S Food and Drug Administration. Pharmaceutical cGMPs for the
21st Century – A Risk-Based Approach. Sept. 2004. url: www.fda.gov
[29] International Council for Harmonisation. Q8(R2): Pharmaceutical
Development. Aug. 2009. url: www.ich.org
[30] International Council for Harmonisation. Q9: Quality Risk Manage-
ment. Nov. 2005. url: www.ich.org
[31] International Council for Harmonisation. Q10: Pharmaceutical Quality
System. June 2008. url: www.ich.org

62
[32] International Council for Harmonisation. Q11: Development and
Manufacture of Drug Substances. May 2012. url: www.ich.org
[33] L. X. Yu, G. Amidon, M. A. Khan, S. W. Hoag, J. Polli, G. K. Raju,
and J. Woodcock. Understanding Pharmaceutical Quality by Design.
AAPS J., 16: 771–783, 2014. doi: 10.1208/s12248-014-9598-3
[34] F. G. Vogt and A. S. Kord. Development of quality-by-design analyti-
cal methods. J. Pharm. Sci., 100: 797–812, 2011. doi: 10.1002/jps.22325
[35] J. Kochling, W. Wu, Y. Hua, Q. Guan, and J. Castaneda-Merced. A
platform analytical quality by design (AQbD) approach for multiple
UHPLC-UV and UHPLC–MS methods development for protein
analysis. J. Pharm. Biomed. Anal., 125: 130–139, 2016. doi: 10.1016/j.
jpba.2016.03.031
[36] L. Taevernier, E. Wynendaele, M. D’Hondt, and B. De Spiegeleer.
Analytical quality-by-design approach for sample treatment of BSA-
containing solutions. J. Pharm. Anal., 5: 27–32, 2015. doi: 10.1016/j.
jpha.2014.06.001
[37] H. I. Mokhtar, R. A. Abdel-Salam, and G. M. Hadad. Design Space
Calculation by In Silico Robustness Simulation with Modeling Error
Propagation in QbD Framework of RP-HPLC Method Development.
Chromatographia, 78: 457–466, 2015. doi: 10.1007/s10337-015-2858-2
[38] R. Mallik, S. Raman, X. Liang, A. W. Grobin, and D. Choudhury.
Development and validation of a rapid ultra-high performance liquid
chromatography method for the assay of benzalkonium chloride
using a quality-by-design approach. J. Chromatogr. A, 1413: 22–32,
2015. doi: 10.1016/j.chroma.2015.08.010
[39] C. Boussès, L. Ferey, E. Vedrines, and K. Gaudin. Using an innovative
combination of quality-by-design and green analytical chemistry
approaches for the development of a stability indicating UHPLC
method in pharmaceutical products. J. Pharm. Biomed. Anal., 115:
114–122, 2015. doi: 10.1016/j.jpba.2015.07.003
[40] Document EMA/430501/2013. EMA-FDA pilot program for parallel
assessment of Quality-by-Design applications: lessons learnt and Q&A
resulting from the first parallel assessment. Aug. 20, 2013. url: www.
ema.europa.eu/ema/
[41] U.S Food and Drug Administration. Guidance for Industry – Process
Validation: General Principles and Practices. Jan. 2011. url: www.fda.
gov

63
[42] P. Borman, K. Truman, D. Thompson, P. Nethercote, and M. Chatfield.
The Application of Quality by Design to Analytical Methods. Pharm.
Technol., 31: 142–152, 2007.
[43] D. K. Lloyd and J. Bergum. “Application of Quality by Design (QbD)
to the Development and Validation of Analytical Methods”. In:
Specification of Drug Substances and Products. Ed. by C. M. Riley, T. W.
Rosanske, and S. R. Rabel Riley. Elsevier, 2014. doi: 10.1016/B978-0-
08-098350-9.00003-5
[44] R. LoBrutto. “Method Design and Understanding”. In: Method Vali-
dation in Pharmaceutical Analysis. Ed. by J. Ermer and P. W. Nethercote.
2:nd. Wiley, 2015. doi: 10.1002/3527604685
[45] International Council for Harmonisation. Final Concept Paper – Q12:
Technical and Regulatory Considerations for Pharmaceutical Product
Lifecycle Management. July 2014. url: www.ich.org
[46] J. Rantanen and J. Khinast. The Future of Pharmaceutical Manufac-
turing Sciences. J. Pharm. Sci., 104: 3612–3638, 2015. doi: 10.1002/jps.
24594
[47] U.S Food and Drug Administration. Guidence for Industry – Changes
to an Approved NDA or ANDA. Apr. 2004. url: www.fda.gov
[48] The European Union. Information and Notices 2013/C 223/01.
Official Journal of the European Communities, 56: 1–79, 2003.
[49] N. Wu, C. J. Welch, T. K. Natishan, H. Gao, T. Chnadrasekaran,
and L. Zhang. “Practical Aspects of Ultrahigh Performance Liquid
Chromatography”. In: Ultra-High Performance Liquid Chromatography
and its Applications. Ed. by Q. A. Xu. Wiley, 2013. doi: 10 . 1002 /
9781118533956
[50] L. Olbe, E. Carlsson, and P. Lindberg. A proton-pump inhibitor
expedition: the case histories of omeprazole and esomeprazole. Nat.
Rev. Drug Discov., 2: 132–139, 2003. doi: 10.1038/nrd1010
[51] F. Gritti and G. Guiochon. Critical contribution of nonlinear chroma-
tography to the understanding of retention mechanism in reversed-
phase liquid chromatography. J. Chromatogr. A, 1099: 1–42, 2005. doi:
10.1016/j.chroma.2005.09.082
[52] R. K. Lindsey, J. L. Rafferty, B. L. Eggimann, J. I. Siepmann, and
M. R. Schure. Molecular simulation studies of reversed-phase liquid
chromatography. J. Chromatogr. A, 1287: 60–82, 2013. doi: 10.1016/j.
chroma.2013.02.040

64
[53] F. Gritti and G. Guiochon. Mass transfer kinetics, band broadening
and column efficiency. J. Chromatogr. A, 1221: 2–40, 2012. doi: 10.
1016/j.chroma.2011.04.058
[54] D. Hlushkou, F. Gritti, A. Daneyko, G. Guiochon, and U. Tallarek.
How Microscopic Characteristics of the Adsorption Kinetics Impact
Macroscale Transport in Chromatographic Beds. J. Phys. Chem. C,
117: 22974–22985, 2013. doi: 10.1021/jp408362u
[55] G. Guiochon, A. Felinger, D. G. Shirazi, and A. M. Katti. Fundamentals
of Preparative and Nonlinear Chromatography. 2:nd. Academic Press,
2006.
[56] A. Felinger. Molecular dynamic theories in chromatography. J.
Chromatogr. A, 1184: 20–41, 2008. doi: 10.1016/j.chroma.2007.12.066
[57] P. V. Danckwerts. Continuous flow systems. Chem. Eng. Sci., 2: 1–13,
1953. doi: 10.1016/0009-2509(53)80001-1
[58] K. Kaczmarski, M. Mazzotti, G. Storti, and M. Morbidelli. Modeling
fixed-bed adsorption columns through orthogonal collocations on
moving finite elements. Comput. Chem Eng., 21: 641–660, 1997. doi:
10.1016/S0098-1354(96)00300-6
[59] P. Brown, G. Byrne, and A. Hindmarsh. VODE: A Variable-Coefficient
ODE Solver. SIAM J. Sci. and Stat. Comput., 10: 1038–1051, 1989. doi:
10.1137/0910062
[60] K. Kaczmarski, J. Kostka, W. Zapała, and G. Guiochon. Modeling
of thermal processes in high pressure liquid chromatography: I.
Low pressure onset of thermal heterogeneity. J. Chromatogr. A, 1216:
6560–6574, 2009. doi: 10.1016/j.chroma.2009.07.020
[61] K. Kaczmarski, F. Gritti, J. Kostka, and G. Guiochon. Modeling
of thermal processes in high pressure liquid chromatography: II.
Thermal heterogeneity at very high pressures. J. Chromatogr. A, 1216:
6575–6586, 2009. doi: 10.1016/j.chroma.2009.07.049
[62] J. Kostka, F. Gritti, K. Kaczmarski, and G. Guiochon. Modified
Equilibrium-Dispersive Model for the interpretation of the efficiency
of columns packed with core–shell particle. J. Chromatogr. A, 1218:
5449–5455, 2011. doi: 10.1016/j.chroma.2011.06.019
[63] K. Kaczmarski and G. Guiochon. Modeling of the Mass-Transfer Ki-
netics in Chromatographic Columns Packed with Shell and Pellicular
Particles. Anal. Chem., 79: 4648–4656, 2007. doi: 10.1021/ac070209w

65
[64] J. Kostka, F. Gritti, G. Guiochon, and K. Kaczmarski. Modeling of
thermal processes in very high pressure liquid chromatography
for column immersed in a water bath: Application of the selected
models. J. Chromatogr. A, 1217: 4704–4712, 2010. doi: 10 . 1016 / j .
chroma.2010.05.018
[65] F. Gritti, W. Piatkowski, and G. Guiochon. Comparison of the ad-
sorption equilibrium of a few low-molecular mass compounds on
a monolithic and a packed column in reversed-phase liquid chro-
matography. J. Chromatogr. A, 978: 81–107, 2002. doi: 10.1016/S0021-
9673(02)01279-7
[66] F. Gritti and G. Guiochon. Influence of a buffered solution on the
adsorption isotherm and overloaded band profiles of an ionizable
compound. J. Chromatogr. A, 1028: 197–210, 2004. doi: 10.1016/j.
chroma.2003.11.106
[67] S. Brunauer, L. S. Deming, W. E. Deming, and E. Teller. On a Theory
of the van der Waals Adsorption of Gases. J. Am. Chem. Soc., 62:
1723–1732, 1940. doi: 10.1021/ja01864a025
[68] A. Seidel-Morgenstern. Experimental determination of single solute
and competitive adsorption isotherms. J. Chromatogr. A, 1037: 255–
272, 2004. doi: 10.1016/j.chroma.2003.11.108
[69] J. Samuelsson, T. Undin, A. Törncrona, and T. Fornstedt. Impro-
vement in the generation of adsorption isotherm data in the elution
by characteristic points method—The ECP-slope approach. J. Chro-
matogr. A, 1217: 7215–7221, 2010. doi: 10.1016/j.chroma.2010.09.004
[70] E. V. Dose, S. Jacobson, and G. Guiochon. Determination of isotherms
from chromatographic peak shapes. Anal. Chem., 63: 833–839, 1991.
doi: 10.1021/ac00008a020
[71] K. Kaczmarski. Estimation of adsorption isotherm parameters with
inverse method—Possible problems. J. Chromatogr. A, 1176: 57–68,
2007. doi: 10.1016/j.chroma.2007.08.005
[72] B. J. Stanley and G. Guiochon. Numerical estimation of adsorp-
tion energy distributions from adsorption isotherm data with the
expectation-maximization method. J. Phys. Chem., 97: 8098–8104,
1993. doi: 10.1021/j100132a046

66
[73] B. J. Stanley, S. E. Bialkowski, and D. B. Marshall. Analysis of
first-order rate constant spectra with regularized least-squares and
expectation maximization. 1. Theory and numerical characterization.
Anal. Chem., 65: 259–267, 1993. doi: 10.1021/ac00051a013
[74] F. Gritti and G. Guiochon. Heterogeneity of the Adsorption Mecha-
nism of Low Molecular Weight Compounds in Reversed-Phase
Liquid Chromatography. Anal. Chem., 78: 5823–5834, 2006. doi:
10.1021/ac060392d
[75] U. D. Neue. “Silica Gel and its Derivatization for Liquid Chromato-
graphy”. In: Encyclopedia of Analytical Chemistry. Ed. by R. A. Meyer.
Wiley, 2009. doi: 10.1002/9780470027318.a5915.pub2
[76] L. Kisley and C. F. Landes. Molecular Approaches to Chromato-
graphy Using Single Molecule Spectroscopy. Anal. Chem., 87: 83–98,
2015. doi: 10.1021/ac5039225
[77] L. C. Sander, K. A. Lippa, and S. A. Wise. Order and disorder in
alkyl stationary phases. Anal. Bioanal. Chem., 382: 646–668, 2005. doi:
10.1007/s00216-005-3127-2
[78] F. Gritti and G. Guiochon. Effect of the Surface Heterogeneity of
the Stationary Phase on the Range of Concentrations for Linear
Chromatography. Anal. Chem., 77: 1020–1030, 2005. doi: 10.1021/
ac040163w
[79] J. L. Rafferty, L. Zhang, J. I. Siepmann, and M. R. Schure. Retention Me-
chanism in Reversed-Phase Liquid Chromatography: A Molecular
Perspective. Anal. Chem., 79: 6551–6558, 2007. doi: 10.1021/ac0705115
[80] A. Cavazzini, M. Remelli, F. Dondi, and A. Felinger. Stochastic
Theory of Multiple-Site Linear Adsorption Chromatography. Anal.
Chem., 71: 3453–3462, 1999. doi: 10.1021/ac990282p
[81] J. C. Giddings and H. Eyring. A Molecular Dynamic Theory of
Chromatography. J. Phys. Chem., 59: 416–421, 1955. doi: 10.1021/
j150527a009
[82] M. J. Wirth and D. J. Swinton. Single-Molecule Probing of Mixed-
Mode Adsorption at a Chromatographic Interface. Anal. Chem., 70:
5264–5271, 1998. doi: 10.1021/ac980632s
[83] M. J. Wirth, M. D. Ludes, and D. J. Swinton. Spectroscopic Obser-
vation of Adsorption to Active Silanols. Anal. Chem., 71: 3911–3917,
1999. doi: 10.1021/ac990382v

67
[84] Z. Zhong, M. Lowry, G. Wang, and L. Geng. Probing Strong Ad-
sorption of Solute onto C18-Silica Gel by Fluorescence Correlation
Imaging and Single-Molecule Spectroscopy under RPLC Conditions.
Anal. Chem., 77: 2303–2310, 2005. doi: 10.1021/ac048290f
[85] J. T. Cooper, E. M. Peterson, and J. M. Harris. Fluorescence Ima-
ging of Single-Molecule Retention Trajectories in Reversed-Phase
Chromatographic Particles. Anal. Chem., 85: 9363–9370, 2013. doi:
10.1021/ac402251r
[86] L. Pasti, N. Marchetti, R. Guzzinati, M. Catani, V. Bosi, F. Dondi,
A. Sepsey, A. Felinger, and A. Cavazzini. Microscopic models of
liquid chromatography: From ensemble-averaged information to
resolution of fundamental viewpoint at single-molecule level. TrAC,
81: 63–68, 2016. doi: 10.1016/j.trac.2015.08.007
[87] J. N. Mabry, M. J. Skaug, and D. K. Schwartz. Single-Molecule Insig-
hts into Retention at a Reversed-Phase Chromatographic Interface.
Anal. Chem., 86: 9451–9458, 2014. doi: 10.1021/ac5026418
[88] D. T. T. Nguyen, D. Guillarme, S. Rudaz, and J.-L. Veuthey. Chroma-
tographic behaviour and comparison of column packed with sub-2
µm stationary phases in liquid chromatography. J. Chromatogr. A,
1128: 105–113, 2006. doi: 10.1016/j.chroma.2006.06.069
[89] M. Martin and G. Guiochon. Effects of high pressure in liquid
chromatography. J. Chromatogr. A, 1090: 16–38, 2005. doi: 10.1016/j.
chroma.2005.06.005
[90] D. Guillarme, D. T. T. Nguyen, S. Rudaz, and J.-L. Veuthey. Method
transfer for fast liquid chromatography in pharmaceutical analysis:
Application to short columns packed with small particle. Part I:
Isocratic separation. Eur. J. Pharm. Biopharm., 66: 475–482, 2007. doi:
10.1016/j.ejpb.2006.11.027
[91] D. Guillarme, D. T. T. Nguyen, S. Rudaz, and J.-L. Veuthey. Method
transfer for fast liquid chromatography in pharmaceutical analysis:
Application to short columns packed with small particle. Part II:
Gradient experiments. Eur. J. Pharm. Biopharm., 68: 430–440, 2008.
doi: 10.1016/j.ejpb.2007.06.018
[92] L. R. Snyder and J. W. Dolan. High-Performance Gradient Elution.
Hoboken: Wiley, 2007. doi: 10.1002/0470055529

68
 [93] M. M. Fallas, U. D. Neue, M. R. Hadley, and D. V. McCalley. Inves-
tigation of the effect of pressure on retention of small molecules
using reversed-phase ultra-high-pressure liquid chromatography. J.
Chromatogr. A, 1209: 195–205, 2008. doi: 10.1016/j.chroma.2008.09.021
[94] M. M. Fallas, U. D. Neue, M. R. Hadley, and D. V. McCalley. Further
investigations of the effect of pressure on retention in ultra-high-
pressure liquid chromatography. J. Chromatogr. A, 1217: 276–284,
2010. doi: 10.1016/j.chroma.2009.11.041
[95] M. M. Fallas, N. Tanaka, S. M. C. Buckenmaier, and D. V. Mc-
Calley. Influence of phase type and solute structure on changes
in retention with pressure in reversed-phase high performance
liquid chromatography. J. Chromatogr. A, 1297: 37–45, 2013. doi:
10.1016/j.chroma.2013.04.006
[96] S. Fekete, J.-L. Veuthey, D. V. McCalley, and D. Guillarme. The effect
of pressure and mobile phase velocity on the retention properties
of small analytes and large biomolecules in ultra-high pressure
liquid chromatography. J. Chromatogr. A, 1270: 127–138, 2012. doi:
10.1016/j.chroma.2012.10.056
[97] M. R. Euerby, M. A. James, and P. Petersson. The influence of
stationary phase on pressure-induced retention, selectivity and
resolution changes in RP-LC. Anal. Bioanal. Chem., 405: 5557–5569,
2013. doi: 10.1007/s00216-013-6973-3
[98] L. Nováková, J. L. Veuthey, and D. Guillarme. Practical method
transfer from high performance liquid chromatography to ultra-high
performance liquid chromatography: The importance of frictional
heating. J. Chromatogr. A, 1218: 7971–7981, 2011. doi: 10 . 1016 / j .
chroma.2011.08.096
[99] F. Gritti and G. Guiochon. Influence of the pressure on the properties
of chromatographic columns: I. Measurement of the compressibility
of methanol–water mixtures on a mesoporous silica adsorbent. J.
Chromatogr. A, 1070: 1–12, 2005. doi: 10.1016/j.chroma.2005.02.007
[100] F. Gritti, M. Martin, and G. Guiochon. Influence of pressure on the
properties of chromatographic columns: II. The column hold-up
volume. J. Chromatogr. A, 1070: 13–22, 2005. doi: 10.1016/j.chroma.
2005.02.008

69
[101] F. Gritti and G. Guiochon. Influence of the pressure on the properties
of chromatographic columns: III. Retention volume of thiourea,
hold-up volume, and compressibility of the C18-bonded layer. J.
Chromatogr. A, 1075: 117–126, 2005. doi: 10.1016/j.chroma.2005.03.095
[102] N. Tanaka, T. Yoshimura, and M. Araki. Effect of pressure on solute
retention in the ionization control mode in reversed-phase liquid
chromatography. J. Chromatogr. A, 406: 247–256, 1987. doi: 10.1016/
S0021-9673(00)94033-0
[103] V. L. McGuffin and C. E. Evans. Influence of pressure on solute
retention in liquid chromatography. J. Microcolumn Sep., 3: 513–520,
1991. doi: 10.1002/mcs.1220030606
[104] V. L. McGuffin, C. E. Evans, and S.-H. Chen. Direct examination of
separation processes in liquid chromatography: Effect of temperature
and pressure on solute retention. J. Microcolumn Sep., 5: 3–10, 1993.
doi: 10.1002/mcs.1220050102
[105] V. L. McGuffin and S.-H. Chen. Molar enthalpy and molar volume
of methylene and benzene homologues in reversed-phase liquid
chromatography. J. Chromatogr. A, 762: 35–46, 1997. doi: 10.1016/
S0021-9673(96)00958-2
[106] M. Martin and G. Guiochon. Pressure dependence of diffusion
coefficient and effect on plate height in liquid chromatography. Anal.
Chem., 55: 2302–2309, 1983. doi: 10.1021/ac00264a023
[107] U. D. Neue and M. Kele. Performance of idealized column structures
under high pressure. J. Chromatogr. A, 1149: 236–244, 2007. doi:
10.1016/j.chroma.2007.03.042
[108] F. Gritti and G. Guiochon. Complete Temperature Profiles in Ultra-
High-Pressure Liquid Chromatography Columns. Anal. Chem., 80:
5009–5020, 2008. doi: 10.1021/ac800280c
[109] A. d. Villiers, H. Lauer, R. Szucs, S. Goodall, and P. Sandra. Influence
of frictional heating on temperature gradients in ultra-high-pressure
liquid chromatography on 2.1 mm I.D. columns. J. Chromatogr. A,
1113: 84–91, 2006. doi: 10.1016/j.chroma.2006.01.120
[110] F. Gritti, M. Martin, and G. Guiochon. Influence of Viscous Friction
Heating on the Efficiency of Columns Operated under Very High
Pressures. Anal. Chem., 81: 3365–3384, 2009. doi: 10.1021/ac802632x

70
[111] L. A. Cole and J. G. Dorsey. Temperature dependence of retention in
reversed-phase liquid chromatography. 1. Stationary-phase conside-
rations. Anal. Chem., 64: 1317–1323, 1992. doi: 10.1021/ac00037a004
[112] L. A. Cole, J. G. Dorsey, and K. A. Dill. Temperature dependence
of retention in reversed-phase liquid chromatography. 2. Mobile-
phase considerations. Anal. Chem., 64: 1324–1327, 1992. doi: 10.1021/
ac00037a005
[113] F. Gritti and G. Guiochon. Adsorption Mechanisms and Effect of
Temperature in Reversed-Phase Liquid Chromatography. Meaning
of the Classical Van’t Hoff Plot in Chromatography. Anal. Chem., 78:
4642–4653, 2006. doi: 10.1021/ac0602017
[114] I. Quiñones, J. C. Ford, and G. Guiochon. Multisolute adsorption
equilibria in a reversed-phase liquid chromatography system. Chem.
Eng. Sci., 55: 909–929, 2000. doi: 10.1016/S0009-2509(99)00393-0
[115] S. Fekete, J. Fekete, and D. Guillarme. Estimation of the effects of
longitudinal temperature gradients caused by frictional heating on
the solute retention using fully porous and superficially porous
sub-2 µm materials. J. Chromatogr. A, 1359: 124–130, 2014. doi: 10.
1016/j.chroma.2014.07.030
[116] U.S Food and Drug Administration. Guidance for Industry – Analytical
Procedures and Methods Validation for Drugs and Biologics. July 2015.
url: www.fda.gov
[117] G. A. Shabir. Validation of high-performance liquid chromatography
methods for pharmaceutical analysis: Understanding the differences
and similarities between validation requirements of the US Food and
Drug Administration, the US Pharmacopeia and the International
Conference on Harmonization. J. Chromatogr. A, 987: 57–66, 2003.
doi: 10.1016/S0021-9673(02)01536-4
[118] I. Taverniers, M. De Loose, and E. Van Bockstaele. Trends in quality in
the analytical laboratory. II. Analytical method validation and quality
assurance. TrAC, 23: 535–552, 2004. doi: 10.1016/j.trac.2004.04.001
[119] J. Ermer. “Method Performance Characteristics”. In: Method Valida-
tion in Pharmaceutical Analysis. Ed. by J. Ermer and P. W. Nethercote.
2:nd. Wiley, 2015. doi: 10.1002/3527604685
[120] P. Araujo. Key aspects of analytical method validation and linearity
evaluation. J. Chromatogr. B, 877: 2224–2234, 2009. doi: 10.1016/j.
jchromb.2008.09.030

71
[121] B. Dejaegher and Y. V. Heyden. Ruggedness and robustness testing. J.
Chromatogr. A, 1158: 138–157, 2007. doi: 10.1016/j.chroma.2007.02.086
[122] H.-C. Liu, L. Liu, and N. Liu. Risk evaluation approaches in failure
mode and effects analysis: A literature review. Expert Syst. Appl., 40:
828–838, 2013. doi: 10.1016/j.eswa.2012.08.010
[123] E. Stocker, G. Toschkoff, S. Sacher, and J. G. Khinast. Use of me-
chanistic simulations as a quantitative risk-ranking tool within the
quality by design framework. Int. J. Pharm., 475: 245–255, 2014. doi:
10.1016/j.ijpharm.2014.08.055
[124] D. M. Hallow, B. M. Mudryk, A. D. Braem, J. E. Tabora, O. K.
Lyngberg, J. S. Bergum, L. T. Rossano, and S. Tummala. An Example
of Utilizing Mechanistic and Empirical Modeling in Quality by
Design. J. Pharm. Innov., 5: 193–203, 2010. doi: 10.1007/s12247-010-
9094-y
[125] F. Gritti and G. Guiochon. Adsorption Mechanism in RPLC. Effect
of the Nature of the Organic Modifier. Anal. Chem., 77: 4257–4272,
2005. doi: 10.1021/ac0580058
[126] J. L. Rafferty, J. I. Siepmann, and M. R. Schure. The effects of
chain length, embedded polar groups, pressure, and pore shape on
structure and retention in reversed-phase liquid chromatography:
Molecular-level insights from Monte Carlo simulations. J. Chromatogr.
A, 1216: 2320–2331, 2009. doi: 10.1016/j.chroma.2008.12.088
[127] F. Chan, L. S. Yeung, R. LoBrutto, and Y. V. Kazakevich. Interpretation
of the excess adsorption isotherms of organic eluent components
on the surface of reversed-phase phenyl modified adsorbents. J.
Chromatogr. A, 1082: 158–165, 2005. doi: 10.1016/j.chroma.2005.05.078
[128] International Council for Harmonisation. Working Draft Version 6.0 –
Q12: Technical and Regulatory Considerations for Pharmaceutical Product
Lifecycle Management. Oct. 2016

72
Chromatographia (2016) 79:1283–1291
DOI 10.1007/s10337-016-3151-8
ORIGINAL
Combining Chemometric Models with Adsorption Isotherm
Measurements to Study Omeprazole in RP‑LC
Dennis Åsberg1 · Marek Leśko2 · Jörgen Samuelsson1   · Anders Karlsson3 ·
Krzysztof Kaczmarski2 · Torgny Fornstedt1   
Received: 20 March 2016 / Revised: 21 June 2016 / Accepted: 26 June 2016 / Published online: 12 August 2016
© The Author(s) 2016. This article is published with open access at Springerlink.com
Abstract The adsorption of the proton-pump inhibitor
omeprazole was investigated using RP-LC with chemo-
metric models combined with adsorption isotherm model-
ling to study the effect of pH and type of organic modifier
(i.e., acetonitrile or methanol). The chemometric approach
revealed that omeprazole was tailing with methanol and
fronting with acetonitrile along with increased fronting at
higher pH. The increased fronting with higher pH for ace-
tonitrile was explored using a pH-dependent adsorption iso-
therm model that was determined using the inverse method
and it agreed well with the experimental data. The model
indicated that the peaks exhibit more fronting at high pH
due to a larger fraction of charged omeprazole molecules.
This model could accurately predict the shape of elu-
tion profiles at arbitrary pH levels in the studied interval.
Using a two-layer adsorption isotherm model, the differ-
ence between acetonitrile and methanol was studied at the
lowest pH at which almost all omeprazole molecules are
neutral. Omeprazole had adsorbate–adsorbate interactions
that were similar in strength for the acetonitrile and metha-
nol mobile phases, while the solute–adsorbent interactions
Electronic supplementary material  The online version of this
article (doi:10.1007/s10337-016-3151-8) contains supplementary
material, which is available to authorized users.
* Jörgen Samuelsson
Jorgen.Samuelsson@kau.se
* Krzysztof Kaczmarski
kkaczmarski@prz.edu.pl
1
Department of Engineering and Chemical Sciences, Karlstad
University, 651 88 Karlstad, Sweden
2
Department of Chemical and Process Engineering, Rzeszów
University of Technology, 35 959 Rzeszow, Poland
3
AstraZeneca R&D Gothenburg, 431 83 Mölndal, Sweden
were almost twice as strong with methanol. The difference
in the relative strengths of these two interactions likely
explains the different peak asymmetries (i.e., tailing/front-
ing) in methanol and acetonitrile. In conclusion, thermody-
namic modelling can complement chemometric modeling
in HPLC method development and increase the understand-
ing of the separation.
Keywords  Liquid chromatography · pH · Adsorption
isotherm · Design of experiments · Omeprazole
Introduction
The pharmaceutical industry quality control (QC) meth-
ods, used continuously to release product batches for mar-
ket, must be validated and then approved by regulatory
agencies, such as the US Food and Drug Administration
(FDA) [1]. To enable minor post-approval variations to
be made to approved QC methods without having to alert
regulators, the quality by design (QbD) concept was intro-
duced several years ago [2]. A variant of this approach is
a QC-method enhancement concept, where a clear bound-
ary regarding critical attributes is defined together with a
principle of the testing technique and an exemplified QC
method. In addition to this, an extended robustness test-
ing using the design of experiment (DoE) is required. By
doing this, a regulatory flexibility can be given based on the
presentation of a deeper scientific knowledge regarding the
actual QC method [2, 3]. In QbD, the use of chemomet-
ric modeling is encouraged and has become a key strategy
[4, 5]. When DoE is used in analytical QbD, the goal is to
establish a method operable design region (MODR), where
the method performance criteria are met and where varia-
tions in responses are understood. Inside the MODR, it is
13
1284
then possible to change experimental conditions without
further validation or regulatory interaction [6, 7].
Sometimes other tools, based on firm physicochemical
theory are needed to obtain the necessary scientific under-
standing of a process, especially when considering chro-
matographic methods. In liquid chromatography, adsorp-
tion equilibrium information about a pure component is
the most important piece of information for understanding
an adsorption process regardless of how many components
are present in the system [8]. Recent research has illus-
trated the importance of investigating adsorption isotherms
for understanding the adsorption processes in analytical as
well as preparative chromatographic systems [9–13].
Previously, we have studied the proton-pump inhibi-
tor omeprazole [14] in a case study of the method trans-
fer from HPLC to UPLC [13, 15, 16]. The QC method for
omeprazole was thoroughly validated in accordance with
the ICH guidelines [16]. It was observed, but not reported,
that pH was important for the peak shape and retention of
omeprazole and that overloaded elution profiles of ome-
prazole were “anti-Langmuirian”-shaped (type-III [17])
when using acetonitrile as organic modifier. It was also
noted that changing the organic modifier to methanol
resulted in “Langmuirian”-shaped (type-I [17]) elution pro-
files. These observations agree with earlier observations
that “Langmuirian”-shaped isotherms are found with the
use of methanol and S-shaped isotherms that are common
with acetonitrile [18]. This was proposed to be because of
a multilayer of acetonitrile adsorbing on the surface of the
stationary phase compared to the single layer of methanol.
To describe the pH dependence, Gritti and Guiochon have
developed a model for acidic and basic compounds close to
their pKa values taking into account their ionization equilib-
ria [19–21]. Changes in the adsorption mechanism due to
the nature of the organic modifier and pH are important to
consider, but are difficult to study using only the linear part
of the adsorption isotherm (using diluted samples). There-
fore, we propose also studying the nonlinear part of the
adsorption isotherm to get a more complete understanding.
Our aim is to investigating the adsorption of omeprazole
as a function of pH and type of organic modifier using a
combined DoE and adsorption isotherm approach. We will
illustrate how adsorption isotherm modeling can comple-
ment chemometric modeling using DoE to obtain a firmer
understanding of the separation system.
Theory
The solute concentrations in the mobile (C) and stationary
(q) phases are related through the adsorption isotherm, and
in this work, two adsorption models were used: a two-layer
adsorption isotherm model [22] and a model taking the
13
D. Åsberg et al.
degree of omeprazole ionization into account [19–21], here
denoted the pH-dependent model, since the degree of ioni-
zation depends on pH. The two-layer adsorption isotherm
model assumes that adsorbate–adsorbate interactions occur
and can be expressed as [22–24]
bS C + 2bS bL C 2
q = qs , (1)
1 + bS C + bS b L C 2
where qs is the saturation capacity, bS is the association
equilibrium constant at the adsorbent surface, and bL is the
association equilibrium constant on the first adsorbed layer
to the column surface. The two-layer model is an expansion
of the Langmuir model incorporating adsorbate–adsorbate
interactions between the first established adsorbed solute
layer and non-adsorbed solute molecules. Equation (1) is
mathematically equivalent to the quadratic [25] and the
Moreau [26] adsorption isotherms, but is derived under
different assumptions. The adsorption isotherm model that
takes omeprazole ionization into account is expressed as
[19–21]
(ac (1 − α) + an α)C
q= , (2a)
1 + ((1 − α)bc + αbn )C
(ac (1 − α) + an α)C
q= , (2b)
1 − ((1 − α)bc + αbn )C
where a and b are adsorption isotherm parameters. Derived
from the general Langmuir model, Eq. (2a) yields Lang-
muirian peak shapes and Eq. (2b) yields anti-Langmuirian
peak shapes [27]. Indices c and n denote the charged and
uncharged fractions of omeprazole, respectively, while α is
the fraction of uncharged omeprazole that is, implicitly, a
function of pH. In this study, the mobile phase was weakly
buffered and the local pH in the solute band depended on
the solute concentration at high loads [19–21].
The set of parameters in the adsorption isotherm model
was determined using the inverse method [27] for each
experimental condition. In the inverse method, the param-
eters are estimated by minimizing the sum of squared dif-
ferences between the experimental and calculated elution
profiles [28]. Elution profiles were calculated using the
equilibrium-dispersive (ED) model [27], which can be
expressed as
∂C ∂q u ∂C ∂ 2C
+F + = Da 2 , (3)
∂t ∂t εt ∂z ∂z
where F = (1 − εt )/εt is the phase ratio, εt is the total
porosity, u is the superficial velocity, Da is the apparent
dispersion coefficient, t and z are the time and axial posi-
tions in the column, respectively, and C and q are the local
mobile and stationary phase solute concentrations, respec-
tively. The orthogonal collocation on the finite-element
method [29] was used to discretize the spatial derivatives of
Combining Chemometric Models with Adsorption Isotherm Measurements to Study Omeprazole in RP-LC
the ED model, while the Adams–Moulton method imple-
mented in the VODE procedure [30] was used to solve
the system of ordinary differential equations. At t = 0, the
stationary phase was in equilibrium with the pure mobile
phase. Danckwerts boundary conditions were used at
the column inlet and outlet [27] with an experimentally
obtained injection profile [13]. The minimization was done
using a modified least square Marquardt algorithm [31].
Calibration from response to concentration for the experi-
mental elution profiles was done by fitting the different
column loads for each experimental condition, so that the
injected mass equaled the eluted mass [13].
Materials and Methods
Chemicals
Gradient grade acetonitrile (VWR International, Radnor,
PA, USA) and HPLC grade methanol (Fischer Scientific,
Loughborough, UK) were used as organic modifiers. Water
with a conductivity of 18.2 MΩ cm from a Milli-Q Plus
185 water purification system (Merck Millipore, Billerica,
MA, USA), analytical grade sodium phosphate dibasic
dihydrate, sodium phosphate monobasic dihydrate (Sigma-
Aldrich, St. Louis, MO, USA), and ammonium bicarbo-
nate (99 %; J.T. Baker, Deventer, Netherlands) were used
to prepare the aqueous buffers. The solutes were omepra-
zole (>99 %), methyl-omeprazole (analytical reference
standard), and omeprazole sulfone (analytical reference
standard) and were gifts from AstraZeneca R&D (Mölndal,
Sweden). The column hold-up volume was determined by
means of pycnometry [32] using acetonitrile and dichlo-
romethane. The aqueous buffers and sample solutions were
filtered through a 0.2-μm nylon filter membrane (What-
man, Maidstone, UK) before use.
Instrumentation
The experiments were performed on an Agilent 1200 chro-
matograph (Agilent Technologies, Palo Alto, CA, USA)
equipped with a binary pump, an auto sampler with a 900-
μL sample loop, a diode-array UV-detector, and a ther-
mostated column oven. The extra column volume from
the auto sampler to the detector was 0.039 mL and was
subtracted from the experimental data. The column was
a 100 mm × 4.6 mm XBridge BEH C18 column (Waters,
Milford, MA, USA) with an average particle diameter of
3.5 µm. Two columns were used, the first in the DoE part
and the second in all other experiments, with column hold-
up volumes of 1.055 and 0.925 mL, respectively. The flow
rates were 1.0 and 0.7 mL min−1 for the acetonitrile and
methanol mobile phases, respectively. Analytical peaks
1285
were detected at 302 nm, while overloaded peaks were
detected at 342 nm.
Procedure
The mobile phases used were either 25/75, v/v, acetonitrile/
aqueous buffer or 45/55, v/v, methanol/aqueous buffer. The
aqueous buffers were 15 mM phosphate buffer of pH 7.0–
9.0 in the DoE part and adsorption isotherm experiments;
when estimating the pKa values, 15 mM ammonium bicar-
bonate buffers of pH between 9.0 and 11.0 were used along
with an additional phosphate buffer of pH 6.0.
s pH is pH measured in the eluent containing the organic
w
modifier, i.e., measured directly in the eluent, although the
pH-electrode is calibrated with water solutions [33]. To
experimentally estimate the sw pKa values of omeprazole
and omeprazole sulfone, which both behave as monoprotic
acids in the investigated pH range, at 30 °C, the retention
factor at different sw pH values were obtained and the results
were fitted to the approximate model [34]:
k0 + k1 × 10pH−pKa
k= , (4)
1 + 10pH−pKa
where k0 and k1 are the retention factors for the acid and
basic forms, respectively. Seven different pH levels were
used for the acetonitrile case, and eight different pH levels
were used for the methanol case. The retention factor was
determined with three replicate measurements at each pH
level. Equation (4) neglects changes in the surface proper-
ties of the adsorbents, but is sufficiently accurate to esti-
mate the pKa values of specific solvent mixtures and tem-
peratures in the sw pH range applicable here.
Column temperature (20–40 °C) and pH (w w pH 7.0–9.0)
were chosen as factors in the experimental design, which was
a full factorial design in three levels with three center points.
For each run, 10-μL samples containing 0.15 mg mL−1
omeprazole, 0.011 mg mL−1 omeprazole sulfone, and
0.007 mg mL−1 methyl-omeprazole were injected in dupli-
cate. The diluent was the corresponding mobile phase for
each run. As responses, the retention and resolution factors
for all components and the tailing factor for omeprazole were
used. Regression models were constructed in the software
MODDE 7 (Umetrics, Sweden) after first removing outliers
and insignificant coefficients at a 95 % confidence level.
To determine the adsorption isotherm of omeprazole
at different pH levels, overloaded, duplicate injections of
300, 400, and 500 μL were made at five pH levels in the
same range as used in the DoE. The column temperature
was 30 °C, and the experiments were performed with either
acetonitrile or methanol as organic modifier. The omepra-
zole concentration was 2.5 mg mL−1 with the acetonitrile
mobile phase and 4.0 mg mL−1 with the methanol mobile
phase, while the diluent was the mobile phase.
13
1286 D. Åsberg et al.

Results and Discussion

Chemometric Modeling

The design region was chosen to span common HPLC tem-

Fig. 1  Structure of the omeprazole and the investigated impuri-
ties along with the chromatogram obtained at the center point
(buffer ww pH  = 8.0, 30 °C) of the experimental design with ace-
tonitrile as organic modifier. The flow rate was 1.0 mL min−1, the
detection was conducted at 302 nm, and the injection was 10 μL of
0.15 mg mL−1 omeprazole, 0.011 mg mL−1 omeprazole sulfone, and
0.007 mg mL−1 methyl-omeprazole. The hydrogen of the benzimida-
zole group lost at high pH for omeprazole and omeprazole sulfone is
indicated in red
peratures and a relevant pH range for the separation sys-
tem (omeprazole quickly degrades below 7 [35]). From the
full factorial design, excellent regression models could be
determined for all responses. The regression coefficients
and statistics are presented in Electronic Supplementary
Material Tables S1 and S2. The structures of the solutes
and the chromatogram of the center point with acetonitrile
are shown in Fig. 1. From the regression models, response
surfaces were constructed for each response. Figure 2a, b
shows the retention factors for omeprazole with acetoni-
trile and methanol as organic modifiers, respectively, with
pH being the most important factor. The same trends are
present for acetonitrile and methanol, i.e., increasing pH
and temperature reduces the retention factor, but tempera-
ture has a comparatively larger effect with methanol. Ome-
prazole sulfone behaves similarly to omeprazole (Elec-
tronic Supplementary Material Fig. S1), i.e., the retentions
are decreasing with increasing temperature and pH. For
omeprazole and omeprazole sulfone, the retention factor
decreases with increasing pH, because the hydrogen on the
benzimidazole group is lost at high pH (Fig. 1) and the mol-
ecules go from being neutral to being negatively charged.
Since the stationary phase is apolar and the mobile-phase
polar in RP-LC, charged compounds are less retained than

10
(a) MeCN, k 5 (c) MeCN, Rs (e) MeCN, Tf 1.1
9
4.5 5
1
pH

8.5 4
w
s

0 0.9
3.5
8
3 0.8
−5
7.5
9.5
1.26
(b) MeOH, k 7 (d) MeOH Rs 9 (f) MeOH Tf
6 8 1.24
9
7
pH

5 1.22
w
s

8.5 6
4 1.2
5
3 1.18
8 4
20 25 30 35 40 20 25 30 35 40 20 25 30 35 40
T [°C] T [°C] T [°C]

Fig. 2  Response surfaces from the experimental designs with acetonitrile (a, c, e) and methanol (b, d, f) as organic modifier. a, b Retention fac-
tors, k, of omeprazole, c, d resolution factors, Rs, between omeprazole and H168/66, and e, f tailing factors, Tf, of omeprazole

13
Combining Chemometric Models with Adsorption Isotherm Measurements to Study Omeprazole in RP-LC 1287

neutral ones. The pKa values for omeprazole and omepra- 10

zole sulfone in the mobile phase at 30 °C were determined MeCN
experimentally by measuring the retention factors with 8 Omeprazole sulfone
three replicates at different pH values and fitting the data
to Eq. (4), and the results are shown in Fig. 3. The pKa val-
ues, given with 95 % confidence intervals, for omeprazole 6
were 9.21 ± 0.16 and 9.18 ± 0.14 with acetonitrile and Omeprazole

k
methanol, respectively. For omeprazole sulfone, the pKa 4
values were 8.15 ± 0.03 and 8.38 ± 0.29 with acetonitrile
and methanol, respectively. The temperature dependence of
2
the pKa value for the phosphate buffer is −0.0028 units/K
[36] and can be neglected in the studied temperature inter-
val. For secondary amines, it is around −0.01 units/K [37], 0
giving a change of ±0.1 units in the design region due to
10
temperature. For methyl-omeprazole, the retention factor is
almost unaffected by pH, since the acidic hydrogen on the MeOH
benzimidazole group is replaced with a methyl group. 8
Figure 2c, d shows response surfaces for the resolution
factor between omeprazole and omeprazole sulfone. For
6
acetonitrile, the elution order is changed with the peaks
k

co-eluting around sw pH = 7.8, while for methanol, omepra-

zole always elutes after omeprazole sulfone with a maxi- 4
mum resolution factor at around sw pH 8.5. In both the cases,
pH is the most important factor with temperature play- 2
ing a minor role. The difference in pKa can quantitatively
account for the reversal of peak order observed in Fig. 2c
(cf. Fig. 3). The resolution factors between omeprazole 0
6 7 8 9 10 11 12
and methyl-omeprazole versus pH and temperature are s
pH
w
shown in Electronic Supplementary Material Fig. S2. For
acetonitrile, the temperature is not significant, leading to a
one-factor model in which the resolution factor increases Fig. 3  Estimation of pKa values for omeprazole and omeprazole sul-
fone in 25/75, v/v, acetonitrile/water and 45/55, v/v methanol/water at
with pH, but is never below nine. With methanol, on the 30 °C. Symbols are experimental retention factors at different mobile
other hand, omeprazole and methyl-omeprazole nearly co- phase sw pH levels and solid lines are the best fit to Eq. (4)
elutes at the lowest pH and temperature indicating a change
in selectivity when switching modifier. One reason for this
is that acetonitrile forms a double layer on the stationary
0.150 mg/mL
phase surface, while methanol forms a monolayer [18]. The 150
0.080 mg/mL
differences between acetonitrile and methanol will be dis- 0.015 mg/mL
cussed further in “Adsorption isotherm modeling”.
Response

Response surfaces for the peak tailing, calculated 100

according to the USP definition, are shown in Fig. 2e and
f. With acetonitrile as modifier, the peak is tailing at low
pH and fronting (i.e., a tailing factor below one) at high 50

pH with the temperature only playing a minor role. Metha-

nol, in contrast, gives only tailing peaks (i.e., tailing factor
0
above one) with temperature being the most important fac-
3 3.1 3.2 3.3 3.4 3.5
tor. This difference cannot be directly explained from the
observations in the DoE investigation. By injecting samples Time [min.]
of different concentrations and observings if the peak shape
changes, it is possible to get an indication if the origin of Fig. 4  Elution profiles of omeprazole using different sample concen-
trations to illustrate that the fronting is due to thermodynamic over-
the peak asymmetry is thermodynamic or kinetic. The loading. The mobile phase with 25/75, v/v, acetonitrile/phosphate
results are presented in Fig. 4, where three concentrations buffer (w
w pH = 9.0) at a temperature of 30 °C and flow rate of 1.0 mL/
of omeprazole are injected at sw pH 9.38 and 40 °C. The min

13
1288 D. Åsberg et al.

peaks clearly become more symmetrical when the concen- (a)

tration is decreasing; therefore, the underlying reason for
the peak asymmetry is likely of thermodynamic in nature
[27]. Note that in QC methods, it is often necessary to have
concentrations in the 0.1 mg/mL range of the active phar-
maceutical ingredient to obtain sufficiently high signals for
the impurities [38, 39].

Adsorption Isotherm Modeling

This section seeks to explain the reason for the peak asym-
metry observed in Fig. 2e and f due to pH. To do this, the
adsorption isotherms for omeprazole, with methanol and
acetonitrile as organic modifiers, were determined directly (b)
from overloaded elution profiles using the inverse method.
Experimental overloaded elution profiles are shown in
Fig. 6 (blue lines). The profile with acetonitrile as modifier
is “anti-Langmuirian” in shape having a diffuse front and a
sharp rear, and at sw pH values above 9, the diffuse front of
the elution profile is increasingly curved. With methanol as
modifier, at sw pH values up to approximately 9, the elution
profiles are “Langmuirian” in shape having a sharp front
and a diffuse rear.
When determining adsorption isotherms using the
inverse method, one must properly select an adsorption
isotherm model a priori. One important characteristic of
the adsorption isotherm is the number of adsorption sites,
usually determined from the adsorption energy distribution Fig. 5  Fraction of the neutral form of omeprazole, α, and sw pH versus
(AED) obtained from experimental data using, for example, the concentration of omeprazole in the mobile phase with a 25/75,
the frontal analysis (FA) [27]. The AED for omeprazole, w pH = 9.0) and b 45/55, v/v, meth-
v/v, acetonitrile/phosphate buffer (w
using the same stationary and mobile phases (sw pH ≈ 8.5) anol/phosphate buffer (ww pH = 9.0)

as in this work, was previously determined from FA [13]

and found to be unimodal, i.e., containing only one type of
adsorption site. Since the DoE investigation revealed that
the retention factor was strongly pH dependent and that the
pKa value of omeprazole was inside the studied pH range,
the adsorption model should take pH into account. Further-
more, when injecting large amounts of omeprazole, the
buffer had insufficient capacity to keep the pH constant, so
the pH will depend on the local omeprazole concentration.
One adsorption isotherm model accounting for this
situation is the pH-dependent model, Eq. (2), derived by
Gritti and Guiochon [19–21]. Rather, lengthy calculations
are needed to determine the function α(C), i.e., the frac-
tion of uncharged omeprazole molecules as a function of
the total omeprazole concentration. Using activity coef-
ficients and the sw pKa derived in the previous section, the
results for the highest pH cases are shown in Fig. 5 and the
calculations are described in detail in the Electronic Sup-
plementary Material. The pH-dependent isotherm model,
Eq. (2), has the advantage that only one set of parameters
is needed for modeling at arbitrary pH (i.e., in the inverse
method) and only elution profiles at the highest and lowest
pH levels being needed. The estimated sets of parameters
for acetonitrile, Eq. (2b), and methanol, Eq. (2a), are pre-
sented in Electronic Supplementary Material Table S3
along with some details of the estimation procedure. The
agreement between calculated and experimental elution
profiles was good in both methanol and acetonitrile (see
red lines in Fig. 6). Note that the intermediate pH in Fig. 6
was not used in the inverse method; the profiles at this level
are predictions and, therefore, agree somewhat less with
the experimental elution profiles. That the pH-dependent
isotherm was able to describe the experimental elution
profiles well lends strength to the proposed mechanism
that the relationship between the charged and uncharged
forms of omeprazole causes the increased fronting at high
pH. When the charged form increases at high pH, the elu-
tion profiles become more deformed, with parts of the front
starting to move faster than the rest of the profile. We con-
cluded that the increased fronting of the omeprazole peaks
with increased pH seen in the DoE investigation with ace-
tonitrile as modifier can be explained by thermodynamic

13
Combining Chemometric Models with Adsorption Isotherm Measurements to Study Omeprazole in RP-LC 1289

2.5
s s s
MeCN pH = 7.59 MeCN pH = 8.51 MeCN pH = 9.38
Concentration [g/L]

w w w
2

1.5
Experiments
1 pH−dependent
2−layer
0.5

0
3 5 7 9 3 5 7 9 3 5 7 9
4
s s s
Concentration [g/L]

MeOH pH = 8.06 MeOH pH = 8.93 MeOH pH = 9.51

w w w

0
6 8 10 12 14 6 8 10 12 14 6 8 10 12 14
Time [min.] Time [min.] Time [min.]

Fig. 6  Comparison between experimental elution profiles (blue
lines) and elution profiles calculated using the pH-dependent iso-
therm model (red lines) and the quadratic isotherm model (green
lines) for omeprazole at different pH levels. In a–c 25/75, v/v, ace-
tonitrile/phosphate buffer is used as mobile phase, and in d–f 45/55,
v/v, methanol/phosphate buffer is used as mobile phase. For acetoni-
trile, the flow rate was 1.0 mL min−1 and injections were 0.5 mL of a
2.5 g L−1 solution. For methanol, the flow rate was 0.7 mL min−1 and
injections were 0.5 mL of a 4.0 g L−1 solution. The column temper-
ature was 30 °C in all experiments. Note that for the pH-dependent
model, one set of numerical parameters is used at all pH levels, while
for the quadratic model, a different parameter set is used at each pH

overloading combined with variation in the local eluent pH
due to the weakly buffered mobile phase.
To study the difference between methanol and acetoni-
trile, we consider the case in which most (>95 %) omepra-
zole molecules are uncharged, i.e., at the lowest pH (buffer
pH 7.0). The two-layer isotherm model, Eq. (1), is fitted to
each pH separately to obtain individual sets of parameters.
At higher pH where the fraction of uncharged omeprazole
decreases, the two-layer isotherm should only be seen as
an empirical model, since it does not take into account the
charge of omeprazole or the variation in the local eluent pH
due to the weakly buffered mobile phase. However, at the
lowest pH, it can yield certain physiochemical insights. The
adsorption isotherm parameters for the two-layer isotherm
is presented in Electronic Supplementary Material Table
S4, and compared with the parameters at the lowest pH for
methanol and acetonitrile, the following is observed: (1) the
saturation capacity is higher for acetonitrile, (2) the equi-
librium constant for the adsorbent, bs, is almost twice as
large for methanol, and (3) the equilibrium constant for the
adsorbed solute layer, bL, is almost equal to that for metha-
nol. Previously, it has been shown that acetonitrile adsorbs
in multilayers to the C18 chains, while methanol adsorbs
in a monolayer [18, 40]. The thicker acetonitrile layer can
dissolve more solute molecules from the bulk than can the
bonded C18 layer alone, which could be the reason for the
higher saturation capacity. The acetonitrile multilayer could
also explain the lower bs with acetonitrile, since omepra-
zole molecules have more difficulty interacting with the
C18 chains due to the thicker acetonitrile layer. From the
above speculation, one could expect that omeprazole has a
similar bL with acetonitrile and with methanol, since this
reflects the interaction between omeprazole molecules sur-
rounded by the mobile phase, which consists mainly of
water in both the cases. The magnitude of bs relative to bL
with the two modifiers indicates that adsorbate–adsorbate
interactions are more favored in acetonitrile than in metha-
nol, which is believed to be the main reason for the change
in peak shape for uncharged omeprazole when switching
organic modifier.
Conclusions
The adsorption of omeprazole as a function of pH for two
organic modifiers, acetonitrile and methanol, has been
investigated through adsorption isotherm characterization.
The aim was to determine an adsorption isotherm model
for the adsorption of omeprazole to demonstrate how such
knowledge could provide complementary information to
support the chemometric modeling commonly used in the
QbD framework. The system considered here contained

13
1290
omeprazole along with two of its impurities, and the buffer
pH and temperature were varied in a DoE investigation.
From the DoE results, it appeared that the selectivity
differed between acetonitrile and methanol as modifiers,
omeprazole having the critical resolution factor with ome-
prazole sulfone in acetonitrile and with methyl-omeprazole
in methanol. This was explained partly by the differences
in the mobile phase pH and pKa values of the solutes in
the acetonitrile and methanol mobile phases. Furthermore,
the DoE results also revealed that omeprazole was tailing
with methanol and fronting with acetonitrile, along with
increased fronting at high pH. These observations could
not be explained by the DoE results, so after confirming
that the underlying origin of the asymmetry was thermody-
namic, the adsorption isotherms were determined to deepen
the understanding.
The increase in fronting with pH in the acetonitrile case
was understood by fitting a pH-dependent adsorption iso-
therm simultaneously to all mobile-phase pH values. This
model contains the fractions of neutral omeprazole mole-
cules, which are, implicitly, a function of the local mobile-
phase pH in the solute band. The pH-dependent model
agreed well with the experimental data and indicated that
the peaks exhibit more fronting at high pH due to a larger
fraction of charged omeprazole molecules. This model
could also accurately predict overloaded elution profiles at
arbitrary pH in the studied interval.
The difference between acetonitrile and methanol was
studied at the lowest pH at which almost all omeprazole
molecules are in the neutral state, using a two-layer adsorp-
tion isotherm model. From the determined adsorption iso-
therms, we found that (1) the saturation capacity was larger
with acetonitrile, (2) the association equilibrium constant
for adsorbate-adsorbent interactions is about a factor two
higher with methanol, and (3) the association equilibrium
constant for adsorbate–adsorbate interactions is similar for
the two organic modifiers. Points (1) and (2) were believed
to be due to the adsorbed multilayers of acetonitrile making
it possible to dissolve more solute molecules from the bulk
than could the bonded layer alone making it more difficult
for omeprazole molecules to interact with the C18 chains
due to the thickness of the acetonitrile layer. The acetoni-
trile multilayer lowered the solute-adsorbent equilibrium
constant, since omeprazole molecules have more difficulty
interacting with the C18 chains due to the thick acetonitrile
layer. The difference in relative strength between the two
equilibrium constants for the two modifiers is believed to
cause the “Langmuir”/”anti-Langmuir” difference.
We strongly believe that thermodynamic modeling can
be a useful tool to complement chemometric models for
the HPLC method validation in the QbD framework. Addi-
tional scientific-based information beside the DoE investi-
gation is of high importance to present and find acceptance
13
D. Åsberg et al.
for an enhanced QC-method concept. Depths in scientific
knowledge make it possible for the regulatory agencies to
give the pharmaceutical industry an increased flexibility
that allow continuous improvement of regulatory approved
QC methods. Thereby, a high-quality release process of
product batches can be maintained during the whole life
cycle of the product. An improved understanding of the
separation process and the ability to predict the shape of
overloaded elution profiles can be achieved at the cost of
only a few more experiments.
Acknowledgments  This work was supported by the Swedish Knowl-
edge Foundation via the project “SOMI: Studies of Molecular Inter-
actions for Quality Assurance, Bio-Specific Measurement & Reliable
Supercritical Purification” (Grant Number 20140179), by the ÅForsk
Foundation via the project “Improved Purification Procedures to Sat-
isfy Modern Drug Quality Assurance and Environmental Criteria”
(Grant Number 15/497) and by Grant 2015/18/M/ST8/00349 from the
National Science Centre, Poland. The work was partly supported also
by the Swedish Research Council (VR) in the project “Fundamental
Studies on Molecular Interactions aimed at Preparative Separations
and Bio-Specific Measurements” (Grant Number 2015-04627). We
are also grateful to Mikael Nilsson at Cambrex R&D Karlskoga and
to Jakob Rajgård at Waters for most valuable discussions.
Compliance with ethical standards 
Conflict of interest  All authors declare that they have no conflict of
interest.
Ethical approval This article does not contain any studies with
human participants or animal performed by any of the authors.
Open Access  This article is distributed under the terms of the Crea-
tive Commons Attribution 4.0 International License (http://crea-
tivecommons.org/licenses/by/4.0/), which permits unrestricted use,
distribution, and reproduction in any medium, provided you give
appropriate credit to the original author(s) and the source, provide a
link to the Creative Commons license, and indicate if changes were
made.
References
1. Görög S (2007) The changing face of pharmaceutical analysis.
Trends Anal Chem 26:12–17. doi:10.1016/j.trac.2006.07.011
2. Yu LX, Amidon G, Khan MA et al (2014) Understanding phar-
maceutical quality by design. AAPS J 16:771–783. doi:10.1208/
s12248-014-9598-3
3. Yu LX (2008) Pharmaceutical quality by design: product and
process development, understanding, and control. Pharm Res
25:781–791. doi:10.1007/s11095-007-9511-1
4. Rozet E, Lebrun P, Hubert P et al (2013) Design spaces for ana-
lytical methods. Trends Anal Chem 42:157–167. doi:10.1016/j.
trac.2012.09.007
5. Schmidt AH, Molnár I (2013) Using an innovative quality-
by-design approach for development of a stability indicating
UHPLC method for ebastine in the API and pharmaceutical for-
mulations. J Pharm Biomed Anal 78–79:65–74. doi:10.1016/j.
jpba.2013.01.032
6. EMA/430501/2013 Document (2013) EMA-FDA pilot program
for parallel assessment of quality-by-design applications: lessons
Combining Chemometric Models with Adsorption Isotherm Measurements to Study Omeprazole in RP-LC
learnt and Q&A resulting from the first parallel assessment.
EMA, London. http://www.ema.europa.eu. Accessed 1 Jul 2016
7. Vogt FG, Kord AS (2011) Development of quality-by-design ana-
lytical methods. J Pharm Sci 100:797–812. doi:10.1002/jps.22325
8. Do DD (1998) Adsorption analysis: equilibria and kinetics.
Imperial College Press, London
9. Samuelsson J, Arnell R, Fornstedt T (2009) Potential of adsorp-
tion isotherm measurements for closer elucidating of binding in
chiral liquid chromatographic phase systems. J Sep Sci 32:1491–
1506. doi:10.1002/jssc.200900165
10. Samuelsson J, Franz A, Stanley BJ, Fornstedt T (2007) Thermo-
dynamic characterization of separations on alkaline-stable silica-
based C18 columns: why basic solutes may have better capacity
and peak performance at higher pH. J Chromatogr A 1163:177–
189. doi:10.1016/j.chroma.2007.06.026
11. Gritti F, Guiochon G (2004) Retention of ionizable com-
pounds in reversed-phase liquid chromatography. effect of the
ionic strength of the mobile phase and the nature of the salts
used on the overloading behavior. Anal Chem 76:4779–4789.
doi:10.1021/ac0304121
12. Enmark M, Samuelsson J, Undin T, Fornstedt T (2011) Charac-
terization of an unusual adsorption behavior of racemic methyl-
mandelate on a tris-(3,5-dimethylphenyl) carbamoyl cellulose
chiral stationary phase. J Chromatogr A 1218:6688–6696.
doi:10.1016/j.chroma.2011.07.064
13. Åsberg D, Leśko M, Samuelsson J et al (2014) Method transfer
from high-pressure liquid chromatography to ultra-high-pressure
liquid chromatography. I. A thermodynamic perspective. J Chro-
matogr A 1362:206–217. doi:10.1016/j.chroma.2014.08.051
14. Olbe L, Carlsson E, Lindberg P (2003) A proton-pump inhibitor
expedition: the case histories of omeprazole and esomeprazole.
Nat Rev Drug Discov 2:132–139. doi:10.1038/nrd1010
15. Åsberg D, Samuelsson J, Leśko M et al (2015) Method transfer
from high-pressure liquid chromatography to ultra-high-pressure
liquid chromatography. II. Temperature and pressure effects. J
Chromatogr A 1401:52–59. doi:10.1016/j.chroma.2015.05.002
16. Åsberg D, Nilsson M, Olsson S et al A quality control method
enhancement concept—continual improvement of regulatory
approved QC methods. J Pharm Biomed Anal. doi:10.1016/j.
jpba.2016.06.018
17. Sing KSW, Everett DH, Haul RAW et al (1985) Reporting phy-
sisorption data for gas/solid systems with special reference to
the determination of surface area and porosity. Pure Appl Chem
57:603–619. doi:10.1351/pac198557040603
18. Gritti F, Guiochon G (2005) Adsorption mechanism in RPLC.
Effect of the nature of the organic modifier. Anal Chem 77:4257–
4272. doi:10.1021/ac0580058
19. Gritti F, Guiochon G (2009) Peak shapes of acids and bases
under overloaded conditions in reversed-phase liquid chroma-
tography, with weakly buffered mobile phases of various pH:
a thermodynamic interpretation. J Chromatogr A 1216:63–78.
doi:10.1016/j.chroma.2008.11.020
20. Gritti F, Guiochon G (2009) Adsorption mechanism of acids and
bases in reversed-phase liquid chromatography in weak buff-
ered mobile phases designed for liquid chromatography/mass
spectrometry. J Chromatogr A 1216:1776–1788. doi:10.1016/j.
chroma.2008.10.064
21. Gritti F, Guiochon G (2009) Band profiles of reacting acido-
basic compounds with water-methanol eluents at different and
ionic strengths in reversed-phase liquid chromatography. J Chro-
matogr A 1216:3175–3184
22. Gritti F, Guiochon G (2004) Influence of a buffered solution on
the adsorption isotherm and overloaded band profiles of an ioniz-
able compound. J Chromatogr A 1028:197–210. doi:10.1016/j.
chroma.2003.11.106
1291
23. Mihlbachler K, Kaczmarski K, Seidel-Morgenstern A, Guiochon
G (2002) Measurement and modeling of the equilibrium behav-
ior of the Tröger’s base enantiomers on an amylose-based chi-
ral stationary phase. J Chromatogr A 955:35–52. doi:10.1016/
S0021-9673(02)00228-5
24. Mihlbachler K, De Jesús MA, Kaczmarski K et al (2006)
Adsorption behavior of the (±)-Tröger’s base enantiomers in
the phase system of a silica-based packing coated with amylose
tri(3,5-dimethyl carbamate) and 2-propanol and molecular mod-
eling interpretation. J Chromatogr A 1113:148–161
25. Ruthven DM (1984) Principles of adsorption and adsorption pro-
cesses. Wiley-Interscience, New York
26. Moreau M, Valentin P, Vidal-Madjar C et al (1991) Adsorp-
tion isotherm model for multicomponent adsorbate–adsorb-
ate interactions. J Colloid Interface Sci 141:127–136.
doi:10.1016/0021-9797(91)90308-U
27. Guiochon G, Shirazi DG, Felinger A, Katti AM (2006) Funda-
mentals of preparative and nonlinear chromatography, 2nd edn.
Academic Press, Boston
28. Kaczmarski K (2007) Estimation of adsorption isotherm param-
eters with inverse method—Possible problems. J Chromatogr A
1176:57–68. doi:10.1016/j.chroma.2007.08.005
29. Kaczmarski K, Mazzotti M, Storti G, Mobidelli M (1997) Mod-
eling fixed-bed adsorption columns through orthogonal colloca-
tions on moving finite elements. Comput Chem Eng 21:641–660.
doi:10.1016/S0098-1354(96)00300-6
30. Brown P, Byrne G, Hindmarsh A (1989) VODE: a variable-
coefficient ODE solver. SIAM J Sci Stat Comput 10:1038–1051.
doi:10.1137/0910062
31. Fletcher R (1971) A modified Marquardt sub-routine for non-lin-
ear least squares. Atomic Energy Research Establishment, Harwell
32. Gritti F, Kazakevich Y, Guiochon G (2007) Measurement of hold-
up volumes in reverse-phase liquid chromatography: definition
and comparison between static and dynamic methods. J Chroma-
togr A 1161:157–169. doi:10.1016/j.chroma.2007.05.102
33. Rosés M, Bosch E (2002) Influence of mobile phase acid-
base equilibria on the chromatographic behaviour of proto-
lytic compounds. J Chromatogr A 982:1–30. doi:10.1016/
S0021-9673(02)01444-9
34. Horvath C, Melander W, Molnar I (1977) Liquid chromatogra-
phy of ionogenic substances with nonpolar stationary phases.
Anal Chem 49:142–154. doi:10.1021/ac50009a044
35. Arvidsson T, Collijn E, Tivert A-M, Rosén L (1991) Peak distor-
tion in the column liquid chromatographic determination of ome-
prazole dissolved in borax buffer. J Chromatogr A 586:271–276.
doi:10.1016/0021-9673(91)85132-Y
36. Scopes RK (1994) Protein purification. Springer, New York
37. Reijenga JC, Gagliardi LG, Kenndler E (2007) Temperature
dependence of acidity constants, a tool to affect separation selec-
tivity in capillary electrophoresis. J Chromatogr A 1155:142–
145. doi:10.1016/j.chroma.2006.09.084
38. Gavin PF, Olsen BA (2008) A quality by design approach to
impurity method development for atomoxetine hydrochloride
(LY139603). J Pharm Biomed Anal 46:431–441. doi:10.1016/j.
jpba.2007.10.037
39. Schmidt AH, Wess C (2013) A QbD with design-of-experiments
approach for development of a state-of-the-art UPLC purity
method for carbamazepine. J Liq Chromatogr Relat Technol
37:2653–2666. doi:10.1080/10826076.2013.853312
40. Kazakevich Y, LoBrutto R, Chan F, Patel T (2001) Interpreta-
tion of the excess adsorption isotherms of organic eluent com-
ponents on the surface of reversed-phase adsorbents: effect on
the analyte retention. J Chromatogr A 913:75–87. doi:10.1016/
S0021-9673(00)01239-5
13
 D. Åsberg et al. /Chromatographia 79 (2016) 1283–1291 1

Supplementary data
Combining chemometric models with adsorption isotherm
measurements to study omeprazole in RP-LC
Dennis Åsberg, Marek Leśko, Jörgen Samuelsson, Anders Karlsson,
Krzysztof Kaczmarski and Torgny Fornstedt

Design of Experiments
Additional results obtained from the regression models derived from the design of experiments
investigation are presented below. In Table S1 and S2 the regression coefficients are listed along with the
quality of the fit in terms of variation explained by the model (R2adj) and variation predicted by the model
(Q2). The models were obtained after excluding outliers and insignificant coefficients using a 95%
confidence level. In Fig. S1, the response surfaces for the retention factor of the impurities are shown. Note
the larger effect of temperature seen when methanol is used as organic modifier, which is also reflected in
the absolute value of the regression coefficients for temperature seen in Table S1 and S2. Fig. S3 shows the
resolution factor of omeprazole and H193/61 which is much larger for acetonitrile than for methanol.
Table S1: Regression coefficients for the DoE in the T-pH space with acetonitrile as organic modifier. k is
the retention factor, Rs is the resolution factor and Tf is the tailing factor. The value 0 denotes that the
coefficient was statistically insignificant at 95% confidence level.
Coefficient kOM kH168/66 kH193/61 Rs, H168/66- Rs, H193/61- Tf, OM
OM OM
Constant -17.78 106.5 14.63 -419.9 173.1 4.732
T 0.02814 -0.4303 -0.03369 1.420 0 0.04026
pH 6.471 -20.11 -1.224 87.58 -42.64 -0.8536
T*T 0 0.0006125 -0.0002414 0 0 0
pH*pH -0.4355 0.9667 0.07127 -4.413 2.773 0.04857
T*pH -0.008330 0.03888 0 -0.1601 0 -0.004959
R2adj 1.000 1.000 0.999 0.999 0.992 0.966
Q2 0.999 0.998 0.997 0.997 0.987 0.911

Table S2: Regression coefficients for the DoE in the T-pH space with methanol as organic modifier. k is
the retention factor, Rs is the resolution factor and Tf is the tailing factor. The value 0 denotes that the
coefficient was statistically insignificant at 95% confidence level.
Coefficient kOM kH168/66 kH193/61 Rs, H168/66- Rs, H193/61- Tf, OM
OM OM
Constant -28.78 60.62 7.995 -505.0 310.3 2.453
T -0.4844 -0.5251 -0.2506 0.6262 -0.1804 0.01844
pH 11.77 -9.952 1.125 118.1 -77.42 -0.3815
T*T 0.001631 0.001971 0.001893 0 0 0
pH*pH -0.8273 0.4507 -0.07346 -6.763 4.796 0.02616
T*pH 0.02838 0.03478 0 -0.07388 0.03575 -0.001719
 D. Åsberg et al. /Chromatographia 79 (2016) 1283–1291 2

R2adj 0.999 1.000 1.000 0.998 0.999 0.894

Q2 0.992 0.999 1.000 0.994 0.998 0.837

Fig. S1: Response surfaces from the experimental design for the retention factor of the impurities. a) and
b) are H168/66 with acetonitrile and methanol as organic modifier, respectively. c) and d) is H193/61 with
acetonitrile and methanol respectively. Note that H193/61 only has a very slight dependence on pH.

Fig. S2: The regression models for the resolution factor between omeprazole and H193/61 for a)
acetonitrile as the organic modifier where temperature as a factor is statistically insignificant and b)
methanol. Note the large resolution found at all pH with acetonitrile and that the peaks nearly co-elute at
low pH and temperature with methanol.
 D. Åsberg et al. /Chromatographia 79 (2016) 1283–1291 3

pH Calculations
In this study, samples were prepared by dissolving the neutral form of omeprazole in the eluent. The
fraction of uncharged omeprazole in the sample depends both on the eluent and the sample concentration.
To calculate this, two equilibria are considered:
H2 PO4− ⇌ HPO2−
4 +H
+

HA ⇌ A− + H +
where HA is omeprazole in its neutral form and A- is the charged form. Other phosphate, omeprazole,
water and methanol equilibria are omitted based on Fig. S4.

Fig. S3: Approximate fraction of different protolytic species (α) in the mobile phase (MeOH) as a function
of pH. The dashed lines are the relevant area for the experiments.

This problem was solved by combining the two equilibria and assuming a change of x according to:

𝐇𝐇𝐇𝐇𝐇𝐇𝟐𝟐−
𝟒𝟒 𝐇𝐇𝐇𝐇 ⇌ 𝐇𝐇𝟐𝟐 𝐏𝐏𝐎𝐎−
𝟒𝟒 𝐀𝐀−
Start 𝐴𝐴 𝐷𝐷 ⇌ 𝐵𝐵 𝐸𝐸
Change −𝑥𝑥 −𝑥𝑥 ⇌ +𝑥𝑥 +𝑥𝑥
Final stage 𝐴𝐴 − 𝑥𝑥 𝐷𝐷 − 𝑥𝑥 ⇌ 𝐵𝐵 + 𝑥𝑥 𝐸𝐸 + 𝑥𝑥

which gave the equilibrium expression:

𝐾𝐾𝑎𝑎 𝛾𝛾𝐵𝐵 (𝐵𝐵 + 𝑥𝑥)𝛾𝛾𝐸𝐸 (𝐸𝐸 + 𝑥𝑥)
= (S1)
𝐾𝐾𝑠𝑠 𝛾𝛾𝐴𝐴 (𝐴𝐴 − 𝑥𝑥)𝛾𝛾𝐷𝐷 (𝐷𝐷 − 𝑥𝑥)
where γ is the activity coefficient. The dissolved omeprazole used in this study was uncharged so E = 0. D
represents the uncharged form of omeprazole so γD = 1 which gives the quadratic equation when solved for
x:
𝑘𝑘𝑘𝑘 + 𝐴𝐴 + 𝐷𝐷 𝐴𝐴𝐴𝐴 𝛾𝛾𝐵𝐵 𝛾𝛾𝐸𝐸 𝐾𝐾𝑎𝑎
𝑥𝑥 2 + 𝑥𝑥 − = 0, 𝑘𝑘 = (S2)
𝑘𝑘 − 1 𝑘𝑘 − 1 𝛾𝛾𝐴𝐴 𝐾𝐾𝑠𝑠
The pH, A and B were estimated using the Henderson-Hasselbalch equation:
 D. Åsberg et al. /Chromatographia 79 (2016) 1283–1291 4

𝛾𝛾𝐵𝐵 𝐵𝐵
pH = p𝐾𝐾𝑎𝑎 + log10 � � (S3)
𝛾𝛾𝐴𝐴 𝐴𝐴

The pKa-value for the solute and the buffer component is also affected by addition of organic modifier. In
this study the pKa-value for omeprazole were chromatographically estimated to 9.18 in the eluent with
45% (v/v) MeOH and 9.21 in the eluent with 25% (v/v) MeCN. Phosphate pKa-values were taken from
literature [1, 2].
The activity coefficients were estimated using the extended Debye-Hückel equation:

−𝐴𝐴𝐷𝐷𝐷𝐷 𝑧𝑧 2 √𝐼𝐼
log 𝛾𝛾𝑖𝑖 = (S4)
1 + 𝑎𝑎𝑖𝑖 𝐵𝐵𝐷𝐷𝐷𝐷 √𝐼𝐼
where a is the Debye radius and I the ionic strength. The ionic strength was solved iteratively and is
defined as:

𝐼𝐼 = 0.5 � 𝑧𝑧𝑖𝑖2 𝐶𝐶𝑖𝑖 = 0.5([Na+ ] + [H2 PO4− ] + 4[HPO2− − + −

4 ] + [A ] + [H3 O ] + [OH ]) (S5)

The Debye radius was taken from literature [3] except for charged omeprazole where the Bates-
Guggenheim convention, which considers a Debye radius of 4.56 Å, was used [4]. ADH and BDH are
constants and can be calculated as [2]:
1.8246 ∙ 106 50.29
𝐴𝐴𝐷𝐷𝐷𝐷 = , 𝐵𝐵𝐷𝐷𝐷𝐷 = (S6)
(𝜀𝜀𝜀𝜀)3/2 √𝜀𝜀𝜀𝜀
Where ε is the dielectric constant and T is the absolute temperature. The dielectric constants for methanol
water mixtures [5] and acetonitrile water mixtures [4] were taken from literature.

Adsorption Isotherm Parameters

The estimation was been done on the basis of two peaks: buffer pH 9.0, Vinj = 500 µL and pH 7.0, Vinj =
300 µL.
The effective (apparent) dispersion coefficient is a function of pH. More precisely the effective diffusion
coefficient (inside adsorbent) should be a function of pH, because the degree of dissociation depends on
pH. The ion and neutral components have different diffusivity inside adsorbent.
Therefore apart from the isotherm model parameters the apparent dispersion coefficients (only for the pH
at which peaks were taken for the estimation) has also been estimated. For other pH the apparent
dispersion coefficient was calculated using linear relationship. The result from the inverse method is
presented in Table S3.

Table S3: Estimated adsorption isotherm parameters for the pH-dependent model with acetonitrile, Eq.
(2b) and methanol, Eq. (2a), as organic modifiers.
Modifier an bn ac bc Da pH 7 Da pH 9
[-] [L/g] [-] [L/g] [cm2/min] [cm2/min]
Methanol 10.2 0.00574 3.44 0.158 0.0290 0.0141
Acetonitrile 9.16 0.0265 0.0380 0.0104 0.0195 0.0689
 D. Åsberg et al. /Chromatographia 79 (2016) 1283–1291 5

Table S4: Estimated adsorption isotherm parameters for the two-layer isotherm for omeprazole at
different pH with acetonitrile and methanol as organic modifiers.
Parameter Acetonitrile Methanol
swpH 7.59 8.08 8.51 9.02 9.38 8.06 8.56 8.93 9.32 9.51
qs [g/L] 119 135 173 151 189 76.1 92.5 93.8 93.6 57.7
bS [L/g] 0.0720 0.0615 0.0441 0.0422 0.0314 0.133 0.101 0.0906 0.0855 0.128
bL [L/g] 0.0704 0.0653 0.0666 0.0984 0.121 0.0572 0.0429 0.0514 0.0589 0.117

References
1. Rosés M, Canals I, Allemann H, et al. (1996) Retention of Ionizable Compounds on HPLC. 2. Effect of
pH, Ionic Strength, and Mobile Phase Composition on the Retention of Weak Acids. Anal Chem
68:4094–4100. doi: 10.1021/ac960105d
2. Bosch E, Espinosa S, Rosés M (1998) Retention of ionizable compounds on high-performance liquid
chromatography: III. Variation of pK values of acids and pH values of buffers in acetonitrile–water
mobile phases. Journal of Chromatography A 824:137–146. doi: 10.1016/S0021-9673(98)00647-5
3. Kielland J (1937) Individual Activity Coefficients of Ions in Aqueous Solutions. J Am Chem Soc
59:1675–1678. doi: 10.1021/ja01288a032
4. Gagliardi LG, Castells CB, Ràfols C, et al. (2007) Static Dielectric Constants of Acetonitrile/Water
Mixtures at Different Temperatures and Debye−Hückel A and a0B Parameters for Activity
Coefficients. J Chem Eng Data 52:1103–1107. doi: 10.1021/je700055p
5. Albright PS, Gosting LJ (1946) Dielectric Constants of the Methanol-Water System from 5 to 55°1. J
Am Chem Soc 68:1061–1063. doi: 10.1021/ja01210a043
6. Baciocchi R, Juza M, Classen J, et al. (2004) Determination of the Dimerization Equilibrium
Constants of Omeprazole and Pirkle’s Alcohol through Optical-Rotation Measurements. Helv Chim
Acta 87:1917–1926. doi: 10.1002/hlca.200490172
 Journal of Chromatography A, 1362 (2014) 206–217

Contents lists available at ScienceDirect

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

Method transfer from high-pressure liquid chromatography to

ultra-high-pressure liquid chromatography. I. A thermodynamic
perspective
Dennis Åsberg a , Marek Leśko a , Jörgen Samuelsson b , Krzysztof Kaczmarski b,∗,1 ,
Torgny Fornstedt a,∗,2
a
Department of Engineering and Chemical Sciences, Karlstad University, SE-651 88 Karlstad, Sweden
b
Department of Chemical Engineering, Rzeszów University of Technology, PL-35 959 Rzeszów, Poland

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

Article history: This is the first investigation in a series that aims to enhance the scientific knowledge needed for reliable
Received 31 May 2014 analytical method transfer between HPLC and UHPLC using the quality by design (QbD) framework. Here,
Received in revised form 12 August 2014 we investigated the differences and similarities from a thermodynamic point of view between RP-LC sep-
Accepted 14 August 2014
arations conducted with 3.5 ␮m (HPLC) and 1.7 ␮m (UHPLC) C18 particles. Three different model solutes
Available online 20 August 2014
and one pharmaceutical compound were used: the uncharged cycloheptanone, the cationic benzyltri-
ethylammonium chloride, the anionic sodium 2-naphatlene sulfonate and the pharmaceutical compound
Keywords:
omeprazole, which was anionic at the studied pH. Adsorption data were determined for the four solutes at
HPLC
UHPLC
varying fractions of organic modifier and in gradient elution in both the HPLC and UHPLC system, respec-
Quality by design tively. From the adsorption data, the adsorption energy distribution of each compound was calculated
Adsorption isotherm and the adsorption isotherm model was estimated. We found that the adsorption energy distribution was
Gradient elution similar, with only minor differences in degree of homogeneity, for HPLC and UHPLC stationary phases. The
Omeprazole. adsorption isotherm model did not change between HPLC and UHPLC, but the parameter values changed
considerably especially for the ionic compounds. The dependence of the organic modifier followed the
same trend in HPLC as in UHPLC. These results indicates that the adsorption mechanism of a solute is
the same on HPLC and UHPLC stationary phases which simplifies design of a single analytical method
applicable to both HPLC and UHPLC conditions within the QbD framework.
© 2014 Published by Elsevier B.V.

1. Introduction
Ultra-high-pressure liquid chromatography (UHPLC) instru-
mentation became commercially available in 2004 (Waters Acquity
UPLC) and now most instrument manufactures offers UHPLC
equipment [1]. UHPLC provides faster separations with lower sol-
vent consumptions as compared to HPLC, with preserved column
efficiency [1,2]; therefore, the interest from the pharmaceutical
industry has grown steadily the last years [3–8]. UHPLC is now well
established and the pharmaceutical industry wants to transfer their
HPLC methods to UHPLC without costly post-regulatory changes
∗ Corresponding authors.
E-mail addresses: kkaczmarski@prz.edu.pl (K. Kaczmarski),
Torgny.Fornstedt@kau.se (T. Fornstedt).
1
Tel.: +48 17 865 1295; fax: +48 17 854 3655.
2
Tel.: +46 54 700 1960; fax: +46 73 271 28 90.
[9]. Today, many pharmaceutical companies file both an HPLC and
an UHPLC method. However, the traditional process of getting an
analytical method approved by the regulators is tedious and time-
consuming. However, the design of analytical methods within the
quality by design (QbD) framework allow changes from HPLC to
UHPLC to be made without re-filing a new method if the changes are
within the design space [10]. The idea of QbD is to design methods
through scientific understanding rather than empirical know-how
[11]. It is therefore of utmost importance to have clear scientific
understanding of the design space the method has been developed
for, i.e. in this case the differences between HPLC and UHPLC, which
is the focus of this investigation.
From an instrumental perspective, the main difference between
HPLC and UHPLC is that smaller particles are used in the col-
umn packing material in UHPLC; as a consequence, pumps able
to manage pressures up to 1000 bar are required. This results in
at least three factors that can affect the chromatographic behav-
ior (retention, efficiency, resolution, etc.); (i) although the column

http://dx.doi.org/10.1016/j.chroma.2014.08.051
0021-9673/© 2014 Published by Elsevier B.V.
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217 207

manufactures aim to produce columns with the same column conditions were used at the column inlet and outlet [21]. For the
chemistry in HPLC and UHPLC, the reduced particle size can results modifier, the injection profile was described by
in different ligand density distribution, porosity, relative particle ⎧
⎪ CM,0 , 0 ≤ t < tp
size distribution, etc.; (ii) the retention time and resolution changes ⎨  
with pressure [12,13] and (iii) due to viscous friction, temperature CM (t, z = 0) = CM,0 + ˇ t − tp , tp ≤ t < tp + tg (3)
gradients arise in the column which can affect both retention and ⎪
⎩
efficiency [14,15]. The dwell volume of the system and the extra- CM,0 + CM , tp + tg ≤ t
column volume are also generally smaller in an UHPLC system.
where tp is the time when the gradient reaches the column inlet,
Omeprazole (“OM”), an important proton pump inhibitor
tg is the duration of the gradient, CM is the change in modifier
[16,17] used in the treatment of heartburn and stomach ulcers,
fraction between the beginning and the end of the gradient and
together with three model compounds, are investigated in realis-
ˇ = CM /tg is the slope of the gradient. The injection profiles for
tic conditions, i.e. gradient elution and buffered mobile phases. The
the solutes were determined experimentally and fitted to empirical
three model compounds are one uncharged (cycloheptanone, “C7”),
equations using nonlinear regression (see Appendix A).
one negatively charged (sodium 2-naphatlene sulfonate, “SNS”)
and one positively charged (benzyltriethylammonium chloride,
2.2. Adsorption isotherm models in gradient elution
“BTEAC”). This study is part of a longer project where the overreach-
ing aim is to discuss how the major differences between HPLC and
In this work, the bi-Langmuir model
UHPLC conditions affect the adsorption equilibrium and give guide-
lines that can be of use when transferring a method from HPLC to aI C aII C
q(c) = + (4)
UHPLC using QbD in a pharmaceutical environment. 1 + bI C 1 + bII C
This study aims at a detailed in-depth investigation on the sim-
best described the adsorption of BTEAC, C7 and SNS on BEH C18.
ilarities and differences between typical modern HPLC and UHPLC
aI and aII are the distribution constant for adsorption sites I and II
operational conditions, respectively, with a focus on the thermo-
and bI and bII are the association equilibrium constants for these
dynamics of the adsorption equilibrium; the flow rate in UHPLC is
sites. For OM, the extended solid-liquid BET model [22] best fit-
kept low for avoiding temperature gradients and ensure accurate
ted the data. The model takes into account adsorbate–adsorbate
adsorption isotherm determinations. Such a fundamental compar-
interactions and is written as
ison between HPLC and UHPLC has not been done before. The
adsorption isotherms of the four model compounds were deter- aC
q(c) = (5)
mined for many different fractions of modifier. This is important (1 − bL C)(1 − bL C + bC)
because most analytical separations are conducted using gradi- BET is just an expansion of the Langmuir adsorption model to
ent elution. Finally, data just from overloaded gradient elution multi-layer adsorption where bL is the association equilibrium con-
experiments are used by the inverse method in gradient elution stant for surface adsorption–desorption over a layer of adsorbate
mode [18,19] to determine the adsorption isotherm and to predict molecules.
overloaded gradient elution profiles for all model compounds, pre- The following relationship has been suggested [23] between the
viously we have only studied uncharged and compounds described retention factor, k, and the organic modifier, CM , in reversed-phase
with type I (convex) adsorption isotherms. LC:
This study is the first investigation in a series that aims to 2
enhance the scientific knowledge needed for reliable analytical k(CM ) = kw e−SCM +dCM (6)
method transfer between HPLC and UHPLC using the QbD frame-
where kw is the retention factor in pure water. S and d describe
work.
how the retention factor changes with the organic modifier. In some
cases, e.g. when a narrow range of modifier concentrations are con-
2. Theory
sidered, the d-term is omitted. Eq. (6) is also commonly used when
describing how the adsorption isotherm parameters depend on
2.1. Equilibrium-dispersive model in gradient elution
the organic modifier [18,24]. Omitting the d-term, the bi-Langmuir
model is written as
To calculate band profiles, the equilibrium-dispersive (ED)
model of chromatography [20] was used. It gives the following mass aI,0 e−SaI CM C aII,0 e−SaII CM C
q(C, CM ) = −Sb CM
+ −Sb CM
(7)
transport balance for component i: 1 + bI,0 e I C 1 + bII,0 e II C
∂Ci ∂q ∂C ∂2 C and the BET model becomes
+ F i + w i = Da 2i (1)
∂t ∂t ∂z ∂z
a e−Sa CM C
where F is the phase ratio, w the superficial velocity of the mobile q(C, CM ) =   0
−Sb CM −Sb CM
phase divided by the total porosity, Da the apparent dispersion 1 − bL,0 e L C 1 − bL,0 e L C + b0 e−Sb CM C
coefficient, t and z are the time and axial position and C and q are (8)
the mobile and stationary phase concentrations, respectively. Da is
calculated from the equation
wL
Da = (2) 2.3. Determining adsorption isotherms in gradient elution
2N
where L is the column length and N the number of theoretical plates. The inverse method is a method for acquiring adsorption data
When modelling single component band profiles in gradient elu- from few experiments with good agreement between the model
tion we have that q = q(C, CM ) where C is the solute concentration and experiments [25]. In the inverse method, calculated band pro-
and CM is the modifier concentration in the mobile phase, i.e. the files are fitted directly to experimental ones using an optimization
system contains two components: the solute and the modifier. The algorithm. Recently, the inverse method was extended to gradient
initial condition was that the column was filled with pure mobile elution, i.e. modifier-dependent adsorption isotherms were deter-
phase with the modifier fraction CM,0 . Danckwerts-type boundary mined directly from overloaded bands obtained in gradient elution
208 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217
[18,19]. This simplifies considerably the otherwise tedious exper-
imental procedure of modelling gradient elution in nonlinear LC
and also enables acquisition of adsorption data from solutes that
are very sensitive to the fraction of organic modifier in the eluent.
In the inverse method, the adsorption isotherm is estimated by
minimizing the sum of squared differences between experimental
and calculated band profiles. In this study, the minimization was
done by a modified least-squares Marquardt method [26] in three
steps:
i. The column was assumed to work in analytical conditions and
the parameters in the analytical version of the model, i.e. when
b = 0, was estimated. This was done by minimizing the differ-
ence in retention time between experimental and calculated
analytical peaks obtained at three different gradient slopes.
ii. Next the b-parameters were pre-estimated using the faster Rou-
chon algorithm to solve the mass balances, Eq. (1) on the base of
four gradient elution, overloaded band profiles obtained at two
different slopes.
iii. Finally, all parameters in the modifier-dependent adsorption
isotherm were estimated simultaneously on the basis of the
same four overloaded band profiles.
In the previous studies [18,19], only uncharged compounds
were investigated. In this study, anionic and cationic compounds
which displayed stronger and more complex nonlinearity were also
studied along with a compound where the adsorption is described
with a type III (“anti-Langmurian”) adsorption isotherm.
3. Experimental
3.1. Chemicals
The mobile phase consisted of gradient grade acetonitrile pur-
chased from VWR International (Radnor, PA, USA) and deionized
water with conductivity 18.2 M cm delivered from a Milli-Q
Plus 185 water purification system from Merck Millipore (Biller-
ica, MA, USA). For pycnometry dichloromethane, also purchased
from VWR International, was used in combination with acetoni-
trile. The 30 mM phosphate buffer at pH 8.00 (22 ◦ C, ≈1 atm) was
prepared from analytical-grade sodium phosphate dibasic dihy-
drate and sodium phosphate monobasic dihydrate purchased from
Sigma-Aldrich (St. Louis, MO, USA). The phosphate buffer was fil-
tered through a 0.2 ␮m nylon filter membrane purchased from
Whatman (Maidstone, UK) before use. Benzyltriethylammonium
chloride “BTEAC” (99%), cycloheptanone “C7” (99%), sodium 2-
naphtalenesulfonate “SNS” (≥95%), all from Sigma-Aldrich, and
omeprazole sodium monohydrate “OM” (>99%), kindly gifted by
AstraZeneca (Mölndal, Sweden), were used as solutes.
3.1.1. pH considerations
All experiments were performed with a phosphate buffer at
w pH = 8.00 and the s pH in the mobile phase varied between 8.07
w w
and 8.79 for 4–40% acetonitrile, respectively. The ww pH property is
the pH measured and calibrated in aqueous solutions and the sw pH
property is the pH measured in the organic aqueous mixtures and
calibrated in aqueous solutions [27]. At these pH conditions, BTEAC
is positively charged, C7 is uncharged, SNS is negatively charged
and OM (amphoteric compound, w w
w pKa,1 = 4.06 and w pKa,2 = 0.79
[17]) is negatively charged. Phosphate buffers have been shown to
keep pH relatively constant when temperature is changed moder-
ately [28], so the temperature effects of the pH was not considered
further.
3.1.2. Sample preparation
Solutions of BTEAC, C7 and SNS were prepared by dissolving the
pure compound directly in the eluent. In gradient elution, the sol-
vent was the same as the eluent at the start of the gradient. OM
had low solubility in the eluent and was hard to dissolve directly in
the eluent. Instead, solid OM was dissolved in pure acetonitrile by
heating the solution to 40 ◦ C. Then phosphate buffer, pre-heated to
40 ◦ C, was added and the solution was placed in an ultra-sonication
bath until all solid particles were dissolved. Using this method, the
maximum solubility was ca. 1.1 g/L in 18% acetonitrile and 3 g/L in
30% acetonitrile and no precipitation was observed at room temper-
ature at these concentrations. OM was stable in the mobile phase
for approximately 24 h; hence all OM solution used in this work
were prepared the same day as the chromatographic experiments
were performed. All solutions were filtered through a 0.2 ␮m PTFE
syringe filter purchased from Whatman (Maidstone, UK) before use.
3.2. HPLC instrumentation
The experiments denoted with HPLC were performed on an
Agilent 1200 chromatograph (Agilent Technologies, Palo Alto, CA,
USA) equipped with a binary pump, an auto sampler with a 900 ␮L
loop, a diode-array UV-detector and a column thermostat. The tem-
perature was 20 ◦ C and the flow rate was 1.00 mL/min. The dwell
volume from the pump to the column inlet was 2.50 mL (deter-
mined according to ref. [29]) and the extra column volume from the
auto sampler to the detector was 0.031 mL. The extra-column vol-
umes were adjusted for in all experimental data. The HPLC column
was a 100 × 4.6 mm XBridge C18 column with an average parti-
cle diameter of 3.5 ␮m. Two identical columns were used denoted
XBridge #1 and XBridge #2. The column hold-up volumes, deter-
mined with pycnometry [30], were 0.991 and 1.011 mL for XBridge
#1 and #2, respectively. XBridge #1 was used in all experiments
except those with OM as solute. The columns were kindly supplied
by Waters.
3.3. UHPLC instrumentation
The experiments denoted with UHPLC were performed on a
Waters Acquity UPLC H-class (Waters Corporation, Milford, MA,
USA) equipped with quaternary pump system, an auto sampler,
a diode-array UV-detector and a column thermostat. The temper-
ature was 40 ◦ C and the flow rate was 0.10 mL/min. At this flow
rate, the heat generation due to viscous friction is negligible. The
dwell volume from the pump to the column inlet was 0.38 mL and
the extra column volume from the auto sampler to the detector
was 0.029 mL; the extra-column volumes were adjusted for in all
experimental data. The UHPLC column was a 50 × 2.1 mm Acquity
UPLC BEH C18 column with an average particle diameter of 1.7 ␮m.
Two identical columns were used denoted BEH #1 and BEH #2. BEH
#1 was used for all perturbation peak experiments and BTEAC elu-
tion profiles while BEH #2 was used for all other experiments. The
columns were kindly gifted by Waters. The column hold-up vol-
umes, determined with pycnometry [30], were 0.096 and 0.112 mL
for the two BEH columns.
3.4. Adsorption isotherm measurements
3.4.1. Perturbation peaks
The perturbation peak (PP) method [20] was used for all compo-
nents, except for OM, to obtain the adsorption energy distribution
and adsorption isotherm for one fraction of modifier in the eluent.
For BTEAC, perturbation pulses were introduced on 19 concen-
tration plateaus ranging from 0 to 10 g/L at 5% acetonitrile in the
mobile phase. For C7, perturbation pulses were introduced on 10
concentration plateaus ranging from 0 to 25 g/L at 25% acetonitrile
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217 209

Table 1
Experimental settings for the gradient experiments.

Solute BTEAC BTEAC C7 C7 SNS SNS OM OM

System HPLC UHPLC HPLC UHPLC HPLC UHPLC HPLC UHPLC

Csample (g/L) 5.00 5.00 25.00 25.63 5.00 5.00 1.00 1.00
Vinj (␮L) 250, 500 25, 50 250, 500 25, 50 100, 200 6, 10, 12.5 300, 400, 500 30, 40, 50
CM,0 (%) 4 4 15 15 7 7 18 18
CM (%) 26 26 25 25 18 18 12 12
tp (min) 0.87 1.18 0.87 1.44 1.60 1.44 2.710 1.212
ˇ (%/min) 1, 2, 3 1, 2, 3 1, 3, 5 1, 3, 5 1, 3, 5 1, 3, 5 1, 2, 3 1, 2, 3
 (nm) 250 240 280 300 320 323 342 342

Csample is sample concentration, Vinj is injection volume and  is detection wavelength. For other notation, see Section 2.1.

in the mobile phase and for SNS perturbation pulses were intro-
duced on 18 concentration plateaus ranging from 0 to 15 g/L at 15%
acetonitrile in the mobile phase. The column was equilibrated for
30 min at each concentration plateau, replicate perturbation peaks
were recorded and the average retention time was used in the
calculations of the adsorption data.
3.4.2. Frontal analysis
There were some problems to detect the perturbation peaks
of OM at infinitesimally small disturbances using the PP method.
Since the error in the PP method increases when there are large
differences between the plateau concentration and that of the dis-
turbances [31,32], the FA method was instead used for OM; 20

BTEAC BTEAC
20
a) b)
15
0.4
qs [g/L]

q [g/L]

10
0.2
5 HPLC at 20oC
o
0 UHPLC at 40 C
0
−2 0 2 4 0 2 4 6 8 10

C7 C7
80
6
c) d)
60
qs [g/L]

4
q [g/L]

40
2
20
0
0
−3.5 −3 −2.5 −2 −1.5 0 5 10 15 20

SNS SNS
20
0.01
e) f)
15
0.4 0.005
qs [g/L]

q [g/L]

0 10
0.2 0 2 4
5
0
0
−2 0 2 4 0 5 10 15

OM OM
15 30
2
g) bL = 0.1 L/g
1 h)
10 20
q [g/L]

q [g/L]

0
0 0.1 0.2
s

5 10

0 0
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5
ln(b) C [g/L]

Fig. 1. Adsorption isotherms and calculated AED for BTEAC (circles), C7 (squares), SNS (diamonds) and OM (triangles) on HPLC (black symbols and lines) and UHPLC (gray
symbols and lines). Lines in (b), (d), (f) and (h) are the best adsorption model fit. All AED were calculated using 106 iterations and energy space were spanned with 400 grid
points.
210 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217

BTEAC C7
3 2.5
2
2
ln(k) 1.5
1
1
0.5
0 0
0 0.05 0.1 0.15 0.1 0.2 0.3 0.4

SNS OM
4 4
3 HPLC 20oC
3
UHPLC 40oC
2
ln(k)

2
1
0 1
−1
0 0.1 0.2 0.15 0.2 0.25 0.3
CM [v/v] CM [v/v]

Fig. 2. The retention factor as a function of fraction of acetonitrile (CM ) in the mobile phase is plotted. BTEAC (circles), C7 (squares), SNS (diamonds) and OM (triangles) on
HPLC (black symbols and lines) and UHPLC (gray symbols and lines). The line represents a model fit to Eq. (6); see Table 3 for parameter values.

fronts were recorded between 0 and 2.5 g/L (which is the maxi-
mum solubility) at 25% acetonitrile in the mobile phase. An increase
in pressure of 1 bar per front was observed in UHPLC which made
frontal analysis of OM infeasible in UHPLC, therefore only FA-data
for HPLC was obtained. The temperature was in this case changed
to 30 ◦ C to lie between HPLC and UHPLC temperatures.
The reason for this increased pressure is probably because some-
thing in the omeprazole solution clogged the column; this is not
surprising if we take into account the fact concentration of the
solute (here OM) in frontal analysis sometimes exceeded thou-
sand fold normal analytical levels. This increase in pressure was
not observed when doing overloaded injection; only when pump-
ing close to saturated solution through the system for longer times
as was necessary in the FA experiments. The column could be regen-
erated by pumping 80/20 acetonitrile/water which agrees with the
idea that the pressure increase are due to an impurity clogging the
column.
3.4.3. Elution by characteristic points
To study how the adsorption isotherms depend on the fraction
of organic modifier in the eluent, adsorption isotherm were deter-
mined at different fraction of organic modifier with the elution by
characteristic points (ECP) method. The ECP method only requires
one overload elution profile for each condition and system as well
as a calibration curve to be able to convert detector response to
eluted concentration [20,33]. Overloaded band profiles in isocratic
mode were obtained at five acetonitrile fractions. The acetonitrile
fractions were different for the three solutes and chosen so that
they covered the acetonitrile fractions used in the gradient mode.
Three overloaded band profiles were recorded for each substance
at each modifier fraction, giving a total of 45 overloaded band pro-
files for each system. The injection volume for UHPLC was scaled
according to
UHPLC
Vcol.
UHPLC HPLC
Vinj. = Vinj. × HPLC
(9)
Vcol.
where Vinj. is the injection volume and Vcol. is the geometric volume
of the column [2].
3.4.4. Extended inverse method
The gradient experiments were performed by recording ana-
lytical and overloaded band profiles in gradient mode for three
different linear gradients. Two overloaded profiles were recorded
for each gradient and the gradients for each solute are shown in
Table 1. Replicate measurements were done and the column was
equilibrated with 20 column-volumes between successive gradient
runs.
4. Computations
4.1. Adsorption energy distribution
The adsorption energy distribution (AED) is a tool to determine
the number of different adsorption sites, and their different energy
of interactions, before any rival model fit procedure. This informa-
tion is useful when selecting the adsorption model to be fitted to
the raw data; since we know the degree of column heterogene-
ity, we can reduce the number of models to use in the rival model
fitting procedure. This increases our understanding of the thermo-
dynamics behind the retention mechanism. The adsorption energy
distribution (AED) was calculated from raw slope data [34] acquired
using ECP or perturbation peak method or raw adsorption data
from frontal analysis experiments. In AED calculations, we need
to assume a local adoption isotherm model. In this study, we used
the Langmuir adsorption isotherm for all solutes showing type I
adsorption (e.g. Langmuir) behavior. However, omeprazole show-
ing type III adsorption behavior and we instead used the extended
liquid-solid BET adsorption isotherm as local model [35,36]. In AED
calculation, we can only span the classical association, K, so we
need to set the association equilibrium parameter between the
adsorbate layers, bL in Eq. (5) as a constant. In these calculations,
several different values of bL were used to observe how it affects
the AED, finally bL = 0.1 were selected to be presented. In all cal-
culations, the expectation–maximization method [37], which does
not require any a priori assumptions about the global adsorption
isotherm, was used to compute the AED. One million iterations
were used and the energy spaces were spanned with 400 grid points
in the calculations. The adsorption energy boundaries were taken
as 1/(10 × bmax ) and 10/bmin .
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217 211

0
H PLC
UHPLC a) b)
2

−1

−2

3 2 .0

c) d)

1 .2

0 0 .4
0 .0 4 0 .0 6 0 .0 8 0 .1 0 0 .1 2 0 .1 4 0 .0 5 0 .1 0 0 .1 5

Fig. 3. Dependence of the bi-Langmuir adsorption isotherm parameters, Eq. (5), on the modifier fraction for the adsorption of BTEAC in HPLC (black) and UHPLC (gray).
Symbols are adsorption isotherm parameters determined using the ECP method. The lines are best model fit to Eq. (6).

4.2. ECP
The ECP method was used to determine adsorption isotherms
for all different modifier fractions instead of FA and PP because ECP
requires fewer experiments. In this study, we used the slope ECP
method, previously presented in [38]. Recently, the slope ECP was
also expanded and demonstrated to handle adsorption isotherm of
type III (anti-Langmuirian) and adsorption isotherms with inflec-
tion points [39]. Using the ECP method it is necessary to have a
calibration curve to convert detector response (R) to concentration
(C). This was done by fitting the detector response for three differ-
ent column loads for each condition to Eq. (10) so that the injected
mass is equal to eluted mass.
 k
2 First the single-component adsorption isotherm for each solute
C = k1 log
k2 − R
where k1 and k2 are constants. As a reference, the adsorption
isotherm models determined using the FA and PP methods were
used. Except for the UHPLC OM case where only ECP data were
used due to experimental issues, discussed above.
4.3. Modeling of band profiles in gradient elution
The orthogonal collocation on finite elements method [40,41]
was used to discretize the spatial derivatives of the ED model, Eq.
(1), and the Adams–Moulton method implemented in the VODE
procedure [42] was used to solve the system of ordinary differential
equations. The number of subdomains was chosen as a tenth of the
average column efficiency while the numerical accuracy was fixed
at 10−6 . The column efficiency was taken as the average efficiency
at five isocratic levels spanning the gradient. This is a fair approxi-
mation because only the shape of heavily overloaded peaks, where
the Shirazi number is high, was considered [20]. The total porosity,
based on pycnometric measurements of the column hold-up vol-
ume, was assumed to be constant during the gradient run. This has
been shown to be a good approximation for moderate changes in
organic modifier [18].
5. Results and discussion
(10)
was determined at one acetonitrile fraction in the eluent in both
HPLC and UHPLC. This was done with the PP method for all solutes
except OM because perturbation peaks were difficult to detect for
OM; instead the FA method was used for OM-HPLC. In OM-UHPLC,
the fronts increased the pressure, so neither FA nor PP experi-
ments were possible; instead the ECP method was used. Then the
dependence of the organic modifier was studied and compared
for HPLC and UHPLC. This was done by estimating the adsorp-
tion isotherms for five acetonitrile fractions in the eluent using the
ECP method. Finally, the adsorption in gradient elution was stud-
ied using the inverse method and estimating modifier-dependent
adsorption isotherms directly from overloaded bands obtained in
gradient elution.
212 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217

Table 2
Estimated adsorption isotherm parameters for BTEAC, C7 and SNS with the PP 2. 5 H PLC
method. Parameters for OM were estimated with the FA method in HPLC and the a)
ECP method in UHPLC. UHPLC
Solute System aI (–) bI (L/g) aII (–) bII (L/g) bL (L/g)

BTEAC HPLC 3.72 0.198 11.8 3.82 –

BTEAC UHPLC 3.22 0.157 8.34 3.12 –
C7 HPLC 3.93 0.0978 2.29 – –
1. 5
C7 UHPLC 3.56 0.0688 1.68 – –
SNS HPLC 1.38 0.0274 7.39 2.07 –
SNS UHPLC 1.04 0.0213 4.09 1.55 –
OM HPLC 7.37 2.27 × 10−5 – – 2.80 × 10−2
OM UHPLC 8.04 4.45 × 10−4 – – 2.85 × 10−2

5.1. Adsorption isotherms 0. 5

−1
The adsorption data were analyzed with a rigorous method prior
to the model selection [43]. First Scatchard plots were constructed.
b)
Together with the shape of overloaded elution profiles, the type of
adsorption could be determined. Overloaded band profiles having
a sharp front and a diffuse rear is characteristic for type I (“Lang-
−2
muir”) models and a diffuse front and a sharp rear indicates type III
(“anti-Langmuir”) models. The adsorption was of type I for BTEAC,
C7 and SNS, which was studied with the PP method, and the adsorp-
tion energy distribution (AED) was calculated from the raw slope
data obtained using the Langmuir isotherm as the local adsorp-
tion model. OM showed type III adsorption and the adsorption data −3
were determined using FA in the HPLC case and ECP in the UHPLC
case. The AED for OM was calculated with the extended solid–liquid
BET isotherm as the local adsorption model [35] using slope of the
adsorption isotherm data for UHPLC and raw adsorption data in the 1. 4
HPLC case. The last step was the model selection which was done by c)
fitting the adsorption isotherm models that were consistent with
the observations from the Scatchard plot, overloaded band profiles
and AED to the raw data with nonlinear regression.
BTEAC, which is positively charged, has overloaded band profiles
of type I, a concave Scatchard plot and a bimodal AED with sym-
0. 7
metrical distributions in both HPLC and UHPLC. The AED is shown
in Fig. 1a. The adsorption energies and the saturation capacities
for the adsorption sites are similar for HPLC and UHPLC. The bi-
Langmuir model has a bimodal AED, is a type I adsorption isotherm
with a concave Scatchard plot [43]. Using nonlinear regression, the
bi-Langmuir model was fitted to the data (Fig. 1b). The model is
consistent with all observations and fitted the data very well. The 0. 0
adsorption isotherms in HPLC and UHPLC are very close with the 0. 15 0. 25 0. 35
UHPLC isotherm slightly lower. The numerical fitting parameters
in the bi-Langmuir models are presented in Table 2 and agree well
with the AED calculations. Fig. 4. Dependence of the Langmuir plus a linear term adsorption isotherm param-
The uncharged solute (C7) showed overloaded band profiles of eters, Eq. (11), on the modifier fraction for the adsorption of C7 in HPLC (black) and
type I, a concave Scatchard plot which only had a slight curvature in UHPLC (gray). Symbols are adsorption isotherm parameters determined using the
ECP method. The lines are best model fit to Eq. (6). See Appendix A for a table with
UHPLC and a unimodal AED (Fig. 1c). Adsorption isotherm models the numerical values of the adsorption isotherm parameters.
that have one adsorption site and type I band profiles are the Lang-
muir, the Jovanovich and the Tóth isotherms [20]. The Jovanovich
isotherm has a convex Scatchard plot and the Tóth model describes unsymmetrical unimodel energy distributions. Therefore, the Lang-
muir model was fitted to the C7 data, but the fit was only moderate
for HPLC, where the Scatchard plot had a more pronounced curva-
Table 3
ture, with an R2 value of 0.9227 and only slightly better for UHPLC
Parameters estimation of Eq. (6) for k(CM ).
with R2 equal to 0.9629. However, by adding a constant term to the
Solute System R2 kw S d Langmuir model
BTEAC HPLC 0.999 36.9 27.0 36.0
aI C
BTEAC UHPLC 0.999 36.4 25.0 38.2 q(c) = + aII C (11)
C7 HPLC 1.000 56.7 13.5 11.3 1 + bI C
C7 UHPLC 1.000 52.6 12.9 10.5
excellent agreement between experiments and calculations could
SNS HPLC 1.000 174 25.5 17.7
SNS UHPLC 1.000 91.2 22.8 18.8 be achieved (see Fig. 1d). It should be noted that this is essen-
OM HPLC 1.000 16.8 × 103 45.0 51.3 tially the bi-Langmuir model with the bII -term equal to zero, but
OM UHPLC 1.000 4.22 × 103 37.1 38.6 since neither the Jovanovich nor the Tóth isotherms were con-
sistent with all observations it was decided to keep Eq. (11). The
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217
3 .5
H PLC
UHPLC
1 .5
4 3 .5
1
−2
0 .0 5 0 .1 0 0 .1 5 0 .2 0
Fig. 5. Dependence of the bi-Langmuir adsorption isotherm parameters, Eq. (4), on the modifier fraction for the adsorption of SNS in HPLC (black) and UHPLC (gray). Symbols
are adsorption isotherm parameters determined using the ECP method. The lines are best model fit to Eq. (6). See Appendix A for a table with the numerical values of the
adsorption isotherm parameters.
difference between the adsorption isotherms in HPLC and UHPLC
is largest at high concentrations, but it is still quite small even at
20 g/L.
The negatively charged solute (SNS) showed type I band profiles
and concave Scatchard plots. The AED was bimodal in both HPLC in
UHPLC (Fig. 1e). However, the high-energy site had a very low sat-
uration capacity in UHPLC. The adoption isotherms are described
very well with the bi-Langmuir model for both HPLC and UHPLC
conditions (see Fig. 1f). The difference in AED is also evident for the
adsorption isotherms where the isotherm for UHPLC lies below the
one corresponding to HPLC. The rather large difference in adsorp-
tion behavior between HPLC and UHPLC may be attributed to the
temperature difference of 20 ◦ C between HPLC and UHPLC. The
temperature dependence will be further investigated in a compan-
ion paper.
Omeprazole, which is also negatively charged, had a lim-
ited solubility and only a slight curvature could be seen on
the adsorption isotherm. The overloaded band profiles was
anti-Langmurian and the Scatchard plot was close to linear
with a positive slope which is true for type III adsorp-
tion isotherms [43]. The AED calculation showed a single
low-energy site for both HPLC and UHPLC (Fig. 1g). The
extended solid–liquid BET adsorption isotherm is consistent
with these observation if bL ≥ b/2. The BET isotherm fit-
ted the experimental data excellently with this condition
fulfilled (Fig. 1h). Note that the UHPLC data is obtained
with the ECP method at 40 ◦ C (hence no grey symbols in
Fig. 1h) and that the FA data was acquired at 30 ◦ C in HPLC.
213
1
a) b)
−1
−2
c) d)
2 .0
0 .5
0 .2 5 0 .0 5 0 .1 5 0 .2 5
Note that for OM the UHPLC isotherm is above the HPLC
counterpart which is the opposite from the model solutes. Over-
all, the AED and adsorption isotherm are very similar for HPLC and
UHPLC.
In conclusion, no dramatic changes in the adsorption behavior
were seen when comparing typical HPLC and UHPLC conditions.
The heterogeneity was similar for the two stationary phases. One
should expect similar adsorption characteristics when switching
from HPLC to UHPLC and the same adsorption model could be used
in HPLC and UHPLC.
5.2. Modifier dependence
Gradient elution is the most common programming technique
in LC and is applied in a majority of the routine analysis. Therefore,
it is important to establish that the modifier dependence of the
adsorption does not seriously change when going from HPLC to
UHPLC. This was investigated on the basis of both analytical and
overloaded peaks.
The relationship shown in Eq. (6) was found to describe the
modifier dependence of the retention factor with excellent accu-
racy. In Table 3, the numerical parameters are listed for comparison
and the result is plotted in Fig. 2. OM is the only solute where the
kw , S and d parameters differ by more than 10% between HPLC
and UHPLC. The kw parameter, which is the retention factor in
pure water, should be reviewed critically in the case of OM and
SNS because it is an extrapolation of the data and since these
solutes are sensitive to the modifier fraction a small error in the
214 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217

4 200
H PLC a) 18%
UHPLC 20%
150 HPLC
22%
3

q [g/L]
25%
100 30%

2 50

0
0 1 2 3
1
−6
120
b)
100
UHPLC
80

q [g/L]
−7
60

40

20
−8

0
0 1 2 3

− 2 .6
C [g/L]
c) Fig. 7. BET adsorption isotherms at different fraction of acetonitrile in the eluent in
HPLC (black) and UHPLC (gray) for OM. See Appendix A for a table with the numerical
values of the adsorption isotherm parameters.

− 3 .5
− 4 .4
0. 18 0. 24 0. 30
Fig. 6. Dependence of the BET adsorption isotherm parameters, Eq. (5), on the mod-
ifier fraction for the adsorption of OM in HPLC (black) and UHPLC (gray). Symbols
are adsorption isotherm parameters determined using the ECP method. The lines
are best model fit to Eq. (6). See Appendix A for a table with the numerical values of
the adsorption isotherm parameters.
experiments results in large errors in kw . The S and d parameters
explain how fast the retention factor changes with the modifier
fraction. For OM, the change is faster in HPLC. One reason for
this is that the sensitivity increases with decreasing temperature.
Another interesting feature is that retention factor for BTEAC
decreases faster in HPLC conditions which gives a lower retention
factor at 15% acetonitrile despite the higher temperature in UHPLC.
The adsorption model has been shown not to change with
the fraction of organic modifier—only the numerical values of the
parameters are affected [18,24]. Therefore, the adsorption isotherm
models found in Section 5.1 are assumed to be true at all modifier
fractions and the numerical parameters are estimated with the ECP
method for these models.
For BTEAC (Fig. 3), the b-parameters are generally larger in HPLC
which is to expect when the temperature is lower. For HPLC, the
modifier dependence is clearly not linear and the d-term must be
used in Eq. (6). For the high-energy site, site II, the a-parameters are
almost identical for all investigated modifier fractions while the a-
parameters for the low energy site, aI , decreases more rapidly for
HPLC which gives the faster decrease in retention time for HPLC
seen in Fig. 2. For C7 (Fig. 4), both constants a and b show very
similar trends for HPLC and UHPLC. This indicates that neither
the temperature nor the column chemistry results in any signifi-
cant difference in the adsorption at any modifier fractions from an
uncharged probes perspective. For SNS (Fig. 5), the aII -constant for
the high-energy site, site II, is much higher for HPLC which can be
correlated to the temperature difference between HPLC and UHPLC.
One important observation is that the b-constants are significantly
larger for UHPLC for acetonitrile fraction above 0.15 despite the
higher temperature.
Due to the limited solubility of omeprazole, the concentration
of the overloaded peak used in the ECP calculations are increasing
from 1 g/L at 18% acetonitrile to 3 g/L at 30% acetonitrile. The trend
for omeprazole (Fig. 6) is similar to those of BTEAC and C7 with
equilibrium parameters which are lower for UHPLC. The most inter-
esting feature in this case is the equilibrium constant, bL , between
solute layers. To our knowledge, the dependence of organic mod-
ifier for this type of interaction has never been studied before.
It seems like this equilibrium constant goes through a maximum
when k  10. This is true for both HPLC and UHPLC and the maxi-
mum is almost at the same modifier fraction. Without data for other
solutes, it is currently difficult to know if this is a unique feature for
this solute or if it applies to other type III adsorption isotherms. The
adsorption isotherms for OM at the different modifier fractions are
shown in Fig. 7. It is evident from the large difference between the
isotherms that OM is very sensitive to the fraction of acetonitrile
in the mobile phase. Changes in the acetonitrile fraction of only 2%
result in a large change in the adsorption, especially at the lower
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217
HPLC − BTEAC HPLC − SNS
2.5
2 10
1.5
1 5
Concentration [g/L]
0.5
0 0
4 6 8 4 6 8
UHPLC − BTEAC UHPLC − SNS
2
10
1.5
1
5 0.5
0.5
0 0
2 4 6 8 4 6 8 5
Time [min]
Fig. 8. Comparison between experimental (dashed lines) and predicted (solid lines) overloaded band profiles in gradient elution. Grey lines are UHPLC and black lines are
HPLC. The gradient slopes are 2, 3, 3 and 2%/min for BTEAC, C7, SNS and OM, respectively. Note that these gradient slopes were not used in the estimation of the model
parameters. See Table 1 for further details.
fractions. The trend of the adsorption isotherms is almost identical
in HPLC and UHPLC.
5.3. Modelling of gradient elution
The newly developed inverse method for gradient elution is, for
the first time, used on ionic solutes and type III (anti-Langmuirian)
adsorption isotherms. The aim is to model band profiles in gradi-
ent elution for both HPLC and UHPLC and predict overloaded band
profiles in gradient elution but also to verify that the modelling
methodology works for on ionic solutes and type III adsorption
isotherms.
The modifier dependence of the adsorption parameters are
investigated in Section 5.2 and fitted to Eq. (6) with the d-term
included. The parameters estimated with the extended inverse
method are empirical fitting parameters, so the model should con-
tain as few adjustable parameters as possible. Therefore, as a first
approach the d-term in Eq. (6) is omitted when describing the mod-
ifier dependence of the adsorption parameters. See Eqs. (7) and (8)
for the models used in the gradient elution modelling.
Very good agreement between the calculated and experimen-
tal band profiles, which was used as a basis in the inverse method,
could be achieved with the d-term in Eq. (6) omitted. The estimated
models were then used to predict band profiles obtained at new
gradient slopes and column loadings to verify their validity. The
result is shown in Fig. 8 and it is evident that the estimations agree
well with experiments for all cases. There were no significant dif-
ferences in overlap for neither ionic solutes versus the uncharged
one nor HPLC versus UHPLC. Note that the flow rate was kept low
in the UHPLC experiments to avoid temperature gradients in the
column. We conclude that both ionic solutes and type III models
could successfully be modelled in gradient elution with the inverse
method in gradient elution.
When the adsorption isotherms estimated in the inverse
method are used to calculate the isocratic band profiles used in the
ECP method (Section 5.2), excellent agreement was found in HPLC
for all solutes. However, in UHPLC the isocratic band profiles were
eluted slightly faster when the experimental ones. This is believed
to be due to an overestimation of the isocratic hold before the gra-
dient reaches the column inlet, tp in Eq. (3). Because the UHPLC
is run at a very low flow rate, 0.1 mL/min, tp is very sensitive to
215
HPLC − C7 HPLC − OM
1 1.5
1
0.5
0.5
0 0
5 6 7 9 9.5 10
UHPLC − C7 UHPLC − OM
0.6 1
0.4
0.2
0 0
6 7 9.5 10 10.5 11
the estimation of the system volume between the pumps and the
column inlet. This potential error is adjusted automatically in the
inverse method, so all band profiles obtained in gradient elution
are therefore well modelled. This hypothesis was investigated the-
oretically by calculating band profiles with known gradients and
adsorption isotherms. These band profiles were then used as a basis
for estimation by the inverse method with the column hold, tp ,
overestimated by 15% and the gradient time, tg , overestimated by
10%. Perfect agreement between estimated and true band profiles
was obtained in gradient elution. However, if the estimated model
was compared to the true band profiles in isocratic elution, the
overlap was only 90%, which is about the same as was seen for the
UHPLC case described above.
We conclude that the inverse method in gradient elution works
well for charge solutes and different types of adsorption isotherms.
Accurate predictions can be made both in HPLC and UHPLC.
6. Conclusions
For the first time, the adsorption differences and similarities
between HPLC and UHPLC were investigated and modelled in
detail using thermodynamics with the state-of-the-art physic-
ochemical theory. Four different solutes were investigated: one
uncharged, one positively charged, one negatively charged and the
negatively charged pharmaceutical compound omeprazole. The
temperature, fraction of organic modifier and the particle size of
the packing material were varied and a newly developed method
for fast estimation of adsorption isotherms [18,19] in gradient
elution mode was for the first time used with success for ionizable
compounds and type-III adsorption isotherms.
The major conclusions are:
1) The adsorption isotherm model does not change when going
from HPLC to UHPLC, only the values of the parameters. But
the parameter dependence on the fraction of organic modifier is
more complicated, and complex, for charged compounds than
for uncharged ones.
2) The adsorption energy distributions are similar for the HPLC and
UHPLC stationary phases, although some differences in homo-
geneity was observed.
216 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217
3) The dependence of the organic modifier followed the same trend
in UHPLC as in HPLC.
4) The dependence of the organic modifier on a type-III adsorp-
tion isotherm, the solid–liquid BET model, was investigated for
the first time. It was found that the equilibrium parameter for
solute–solute layers had a maximum in both HPLC and UHPLC,
which was independent of the temperature.
These conclusions indicate that there are small differences
between the stationary phases used in HPLC and UHPLC and not
any major changes in the adsorption mechanism of the solutes, i.e.
change in peak shape or selectivity should be expecting due to dif-
ferences in the stationary phases. This holds true for both isocratic
and gradient elution.
This study is the first in a series that aims to enhance the sci-
entific knowledge needed for reliable analytical method transfer
between HPLC and UHPLC using the quality by design framework;
the next studies will focus (i) on temperature and pressure effects
and (ii) on practical implementations of the knowledge gained in
this study in industrial analytical settings.
Acknowledgements
We are grateful to Waters that provided an ACQUITY UPLC H-
Class system with support and technical advice during the project.
We are also grateful to Anders Karlsson at AstraZeneca R&D Möl-
ndal for initiating this project and for most valuable discussions.
This work was supported by the Swedish Research Council (VR)
in the project “Fundamental studies on molecular interactions
aimed at preparative separations and biospecific measurements”
(grant number 621-2012-3978) and by (i) the Swedish Knowledge
Foundation for the KK HÖG 2011 project “Improved Purification
Processes to Satisfy Modern Drug Quality Assurance and Environ-
mental Criteria” (grant number 20110212).
Appendix A. Supplementary data
Supplementary data associated with this article can be
found, in the online version, at http://dx.doi.org/10.1016/j.chroma.
2014.08.051.
References
[1] S. Fekete, E. Oláh, J. Fekete, Fast liquid chromatography: the domination of
core–shell and very fine particles, J. Chromatogr. A 1228 (2012) 57–71.
[2] M. Szalka, J. Kostka, E. Rokaszewski, K. Kaczmarski, Analysis of related sub-
stances in bisoprolol fumarate on sub-2-␮m adsorbents, Acta Chromatogr. 24
(2012) 163–183.
[3] S.A.C. Wren, P. Tchelitcheff, Use of ultra-performance liquid chromatography
in pharmaceutical development, J. Chromatogr. A 1119 (2006) 140–146.
[4] S. Fekete, J. Fekete, K. Ganzler, Validated UPLC method for the fast and sen-
sitive determination of steroid residues in support of cleaning validation in
formulation area, J. Pharm. Biomed. Anal. 49 (2009) 833–838.
[5] A.M. Evans, C.D. DeHaven, T. Barrett, M. Mitchell, E. Milgram, Integrated nontar-
geted ultrahigh performance liquid chromatography/electrospray ionization
tandem mass spectrometry platform for the identification and relative quan-
tification of the small-molecule complement of biological systems, Anal. Chem.
81 (2009) 6656–6667.
[6] R.N.O. Tettey-Amlalo, I. Kanfer, Rapid UPLC–MS/MS method for the determina-
tion of ketoprofen in human dermal microdialysis samples, J. Pharm. Biomed.
Anal. 50 (2009) 580–586.
[7] A. Abrahim, M. Al-Sayah, P. Skrdla, Y. Bereznitski, Y. Chen, N. Wu, Practical
comparison of 2.7 ␮m fused-core silica particles and porous sub-2 ␮m particles
for fast separations in pharmaceutical process development, J. Pharm. Biomed.
Anal. 51 (2010) 131–137.
[8] S. Fekete, I. Kohler, S. Rudaz, D. Guillarme, Importance of instrumentation for
fast liquid chromatography in pharmaceutical analysis, J. Pharm. Biomed. Anal.
87 (2014) 105–119.
[9] J. Musters, L. van den Bos, E. Kellenbach, Applying QbD principles to develop a
generic UHPLC method which facilitates continual improvement and innova-
tion throughout the product lifecycle for a commercial API, Org. Process Res.
Dev. 17 (2013) 87–96.
[10] A.S. Rathore, H. Winkle, Quality by design for biopharmaceuticals, Nat. Biotech-
nol. 27 (2009) 26–34.
[11] A. Rathore, R. Mahtre, Quality by Design for Biopharmaceuticals: Principles and
Case Studies, Wiley, Hoboken, NJ, USA, 2009.
[12] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Further investigations of the
effect of pressure on retention in ultra-high-pressure liquid chromatography,
J. Chromatogr. A 1217 (2010) 276–284.
[13] S. Fekete, J.-L. Veuthey, D.V. McCalley, D. Guillarme, The effect of pressure and
mobile phase velocity on the retention properties of small analytes and large
biomolecules in ultra-high pressure liquid chromatography, J. Chromatogr. A
1270 (2012) 127–138.
[14] K. Kaczmarski, F. Gritti, J. Kostka, G. Guiochon, Modeling of thermal processes
in high pressure liquid chromatography: II. Thermal heterogeneity at very high
pressures, J. Chromatogr. A 1216 (2009) 6575–6586.
[15] F. Gritti, M. Martin, G. Guiochon, Influence of viscous friction heating on the effi-
ciency of columns operated under very high pressures, Anal. Chem. 81 (2009)
3365–3384.
[16] L. Olbe, E. Carlsson, P. Lindberg, A proton-pump inhibitor expedition: the case
histories of omeprazole and esomeprazole, Nat. Rev. Drug Discov. 2 (2003)
132–139.
[17] J.M. Shin, Y.M. Cho, G. Sachs, Chemistry of covalent inhibition of the gastric
(H+, K+)-ATPase by proton pump inhibitors, J. Am. Chem. Soc. 126 (2004)
7800–7811.
[18] D. Åsberg, M. Leśko, M. Enmark, J. Samuelsson, K. Kaczmarski, T. Fornstedt, Fast
estimation of adsorption isotherm parameters in gradient elution preparative
liquid chromatography. I: The single component case, J. Chromatogr. A 1299
(2013) 64–70.
[19] D. Åsberg, M. Leśko, M. Enmark, J. Samuelsson, K. Kaczmarski, T. Fornstedt, Fast
estimation of adsorption isotherm parameters in gradient elution preparative
liquid chromatography II: The competitive case, J. Chromatogr. A 1314 (2013)
70–76.
[20] G. Guiochon, D.G. Shirazi, A. Felinger, A.M. Katti, Fundamentals of Prepara-
tive and Nonlinear Chromatography, 2nd ed., Academic Press, Boston, MA,
2006.
[21] P.V. Danckwerts, Continuous flow system, Chem. Eng. Sci. 2 (1953) 1–18.
[22] F. Gritti, W. Piatkowski, G. Guiochon, Comparison of the adsorption equilib-
rium of a few low-molecular mass compounds on a monolithic and a packed
column in reversed-phase liquid chromatography, J. Chromatogr. A 978 (2002)
81–107.
[23] P.J. Schoenmakers, H.A.H. Billiet, L. De Galan, Influence of organic modi-
fiers on the retention behaviour in reversed-phase liquid chromatography
and its consequences for gradient elution, J. Chromatogr. 185 (1979)
179–195.
[24] P. Jandera, D. Komers, Fitting competitive adsorption isotherms to the exper-
imental distribution data in reversed-phase systems, J. Chromatogr. A 762
(1997) 3–13.
[25] E. Dose, S. Jacobson, G. Guiochon, Determination of isotherms from chromato-
graphic peak shapes, Anal. Chem. 63 (1991) 833–839.
[26] R. Fletcher, A modified Marquardt sub-routine for non-linear least squares,
Atomic Energy Research Establishment, Harwell, UK, 1971.
[27] M. Rosés, E. Bosch, Influence of mobile phase acid-base equilibria on the chro-
matographic behaviour of protolytic compounds, J. Chromatogr. A 982 (2002)
1–30.
[28] L.G. Gagliardi, C.B. Castells, C. Ràfols, M. Rosés, E. Bosch, Effect of
temperature on the chromatographic retention of ionizable compounds:
II. Acetonitrile–water mobile phases, J. Chromatogr. A 1077 (2005)
159–169.
[29] J.W. Dolan, Dwell Volume Revisited, LC GC N. A. 24 (2006) 458–466.
[30] F. Gritti, Y. Kazakevich, G. Guiochon, Measurement of hold-up vol-
umes in reverse-phase liquid chromatography: definition and comparison
between static and dynamic methods, J. Chromatogr. A 1161 (2007)
157–169.
[31] P. Forssén, J. Lindholm, T. Fornstedt, Theoretical and experimental study of
binary perturbation peaks with focus on peculiar retention behaviour and van-
ishing peaks in chiral liquid chromatography, J. Chromatogr. A 991 (2003)
31–45.
[32] J. Lindholm, P. Forssén, T. Fornstedt, Validation of the accuracy of the perturba-
tion peak method for determination of single and binary adsorption isotherm
parameters in LC, Anal. Chem. 76 (2004) 4856–4865.
[33] S. Ottiger, J. Kluge, A. Rajendran, M. Mazzotti, Enantioseparation of 1-phenyl-
1-propanol on cellulose-derived chiral stationary phase by supercritical
fluid chromatography. II. Non-linear isotherm, J. Chromatogr. A 1162 (2007)
74–82.
[34] J. Samuelsson, T. Fornstedt, Calculations of the energy distribution from pertur-
bation peak data—a new tool for characterization of chromatographic phases,
J. Chromatogr. A 1203 (2008) 177–184.
[35] M. Enmark, J. Samuelsson, T. Undin, T. Fornstedt, Characterization of an
unusual adsorption behavior of racemic methyl-mandelate on a tris-(3,5-
dimethylphenyl) carbamoyl cellulose chiral stationary phase, J. Chromatogr.
A 1218 (2011) 6688–6696.
[36] V. Agmo Hernández, J. Samuelsson, P. Forssén, T. Fornstedt, Enhanced interpre-
tation of adsorption data generated by liquid chromatography and by modern
biosensors, J. Chromatogr. A 1317 (2013) 22–31.
[37] B.J. Stanley, G. Guiochon, Numerical estimation of adsorption energy distri-
butions from adsorption isotherm data with the expectation-maximization
method, J. Phys. Chem. 97 (1993) 8098–8104.
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217 217

[38] J. Samuelsson, T. Undin, A. Törncrona, T. Fornstedt, Improvement in the gen-
eration of adsorption isotherm data in the elution by characteristic points
method—the ECP-slope approach, J. Chromatogr. A 1217 (2010) 7215–7221.
[39] J. Samuelsson, T. Undin, T. Fornstedt, Expanding the elution by characteristic
point method for determination of various types of adsorption isotherms, J.
Chromatogr. A 1218 (2011) 3737–3742.
[40] J. Villadsen, M.L. Michelsen, Solution of Differential Equation Models by Poly-
nomial Approximation, Prentice-Hall, Englewood Cliffs, NJ, 1978.
[41] K. Kaczmarski, M. Mazzotti, G. Storti, M. Mobidelli, Modeling fixed-bed adsorp-
tion columns through orthogonal collocations on moving finite elements,
Comput. Chem. Eng. 21 (1997) 641–660.
[42] P.N. Brown, A.C. Hindmarsh, G.D. Byrne, Variable-Coefficient Ordinary Differ-
ential Equation Solver, available at: http://netlib.org
[43] J. Samuelsson, R. Arnell, T. Fornstedt, Potential of adsorption isotherm mea-
surements for closer elucidating of binding in chiral liquid chromatographic
phase systems, J. Sep. Sci. 32 (2009) 1491–1506.
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217 1

Supplementary data
Method transfer from high-pressure liquid chromatography to ultra-
high-pressure liquid chromatography. I. A thermodynamic perspective
Dennis Åsberg, Marek Leśko, Jörgen Samuelsson, Krzysztof Kaczmarski and
Torgny Fornstedt

Injection profiles
The experimental injection profiles were recorded with the column replaced by a zero-volume union and
the result was fitted to a number of empirical equations. For low injection volumes (≈10 μL) the following
equation was used

p1  t − p3  
2

C (t , z = 0 )= × exp   −2 ×   (S1)
p2 π /2  p2  
 

where p1, p2 and p3 are adjustable parameters, C is the solute concentration and t and z are time and length
coordinates respectively. For moderate injection volumes (≈100 μL) the following equation was used

p1 − p2
C ( t ,= )
z 0= + p2 . (S2)
1 + ( t / p3 )
p4

For large injection profiles (≈500 μL) the equation

p4
 t − p2   t − p2 
C ( t , z = 0 ) = p1 × 1 − exp  −  × exp  −  (S3)
 p3   p5 
 

was used.
 D. Åsberg et al. / J. Chromatogr. A 1362 (2014) 206–217 2

Estimated adsorption isotherm parameters from the ECP method

The numerical values of the adsorption isotherm parameters found with the ECP method in HPLC and
UHPLC are presented in Table S1. For the components BTEAC and SNS the parameters refer to Eq. (4),
for C7 the parameters refer to Eq. (11) and for OM the parameters refer to Eq. (5).

Table S1: Estimated adsorption isotherm parameters from the ECP method.
HPLC UHPLC
Solute CM [%]
aI bI [L/g] aII bII, bL [L/g] aI bI [L/g] aII bII, bL [L/g]
4 9.22 0.631 12.48 6.95 6.78 0.345 13.35 4.57
7 4.33 0.291 5.93 3.21 5.06 0.276 6.09 3.63
BTEAC
10 2.65 0.190 3.34 2.23 3.44 0.159 3.53 2.31
15 1.69 0.161 1.55 1.87 2.42 0.102 1.77 1.80
15 11.6 0.235 3.65 10.1 0.205 3.50
22 5.52 0.122 2.68 4.97 0.110 2.62
C7 30 2.89 0.0693 1.87 2.94 0.0550 1.65
35 2.22 0.0472 1.41 2.28 0.0444 1.41
40 1.82 0.0447 1.12 1.95 0.0359 1.10
7 21.4 2.03 26.6 16.8 24.7 2.66 7.19 21.1
10 10.56 1.03 14.2 8.91 12.2 1.33 5.57 9.31
SNS 15 3.87 0.422 5.05 3.81 4.91 0.689 2.00 4.48
20 1.66 0.225 1.99 1.97 2.49 0.504 0.526 4.42
25 1.08 0.227 0.686 1.89 1.30 0.422 0.241 8.47
18 41.7 2.17E-03 4.34E-02 33.0 9.36E-04 2.00E-02
20 25.1 1.45E-03 5.28E-02 21.3 7.44E-04 2.55E-02
OM 22 15.7 1.17E-03 5.63E-02 13.9 7.32E-04 3.57E-02
25 8.58 6.37E-04 3.96E-02 8.04 4.45E-04 2.85E-02
30 3.84 3.08E-04 2.04E-02 3.68 2.62E-04 2.06E-02
 Journal of Chromatography A, 1401 (2015) 52–59

Contents lists available at ScienceDirect

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

Method transfer from high-pressure liquid chromatography to

ultra-high-pressure liquid chromatography. II. Temperature and
pressure effects
Dennis Åsberg a , Jörgen Samuelsson a,∗ , Marek Leśko b , Alberto Cavazzini c ,
Krzysztof Kaczmarski b,∗∗ , Torgny Fornstedt a
a
Department of Engineering and Chemical Sciences, INTERACT, Karlstad University, SE-651 88 Karlstad, Sweden
b
Department of Chemical and Process Engineering, Rzeszów University of Technology, PL-35 959 Rzeszów, Poland
c
Department of Chemical and Pharmaceutical Sciences, University of Ferrara, IT-44 121 Ferrara, Italy

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

Article history: The importance of the generated temperature and pressure gradients in ultra-high-pressure liquid chro-
Received 17 December 2014 matography (UHPLC) are investigated and compared to high-pressure liquid chromatography (HPLC).
Received in revised form 30 April 2015 The drug Omeprazole, together with three other model compounds (with different chemical character-
Accepted 1 May 2015
istics, namely uncharged, positively and negatively charged) were used. Calculations of the complete
Available online 11 May 2015
temperature profile in the column at UHPLC conditions showed, in our experiments, a temperature dif-
ference between the inlet and outlet of 16 ◦ C and a difference of 2 ◦ C between the column center and the
Keywords:
wall. Through van’t Hoff plots, this information was used to single out the decrease in retention factor
Liquid chromatography
Method transfer
(k) solely due to the temperature gradient. The uncharged solute was least affected by temperature with
UHPLC a decrease in k of about 5% while for charged solutes the effect was more pronounced, with k decreases
Pressure up to 14%. A pressure increase of 500 bar gave roughly 5% increase in k for the uncharged solute, while
Temperature omeprazole and the other two charged solutes gave about 25, 20 and 15% increases in k, respectively. The
Stochastic theory stochastic model of chromatography was applied to estimate the dependence of the average number of
adsorption/desorption events (n) and the average time spent by a molecule in the stationary phase ( s ) on
temperature and pressure on peak shape for the tailing, basic solute. Increasing the temperature yielded
an increase in n and decrease in  s which resulted in less skew at high temperatures. With increasing
pressure, the stochastic modeling gave interesting results for the basic solute showing that the skew
of the peak increased with pressure. The conclusion is that pressure effects are more pronounced for
both retention and peak shape than the temperature effects for the polar or charged compounds in our
study.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction
The interest from the industry to move analytical methods
from high-pressure liquid chromatography (HPLC) to ultra-high-
pressure liquid chromatography (UHPLC) has grown in the last five
years [1]. UHPLC provides faster separations and lower solvent
consumption compared to HPLC, with preserved column effi-
ciency [2,3]. This is achieved by decreasing the particle size of the
∗ Corresponding author. Tel.: +46 54 700 1620; fax: +46 73 271 2890.
∗∗ Corresponding author. Tel.: +48 17 865 1295; fax: +48 17 854 3655.
E-mail addresses: Jorgen.Samuelsson@kau.se (J. Samuelsson),
kkaczmarski@prz.edu.pl (K. Kaczmarski).
stationary phase and increasing the linear velocity of the mobile
phase. As a consequence, the pressure drop over the column is much
larger in UHPLC compared to HPLC, which leads to significant pres-
sure and temperature (due to viscous heating and solvent compres-
sion) gradients in the column. These gradients have been shown to
affect chromatographic performance and predictability [4].
Temperature gradient depends strongly on the method
employed to thermostat the column and has been calculated for
different conditions in UHPLC [5–9]. Longitudinal temperature gra-
dients prevail when the column compartment is close to adiabatic
(e.g. in still-air conditions); they essentially affect only retention
time, without compromising column efficiency. Radial temperature
gradients, on the other hand, arise in well-thermostated conditions
(e.g., with water thermostating), where the center of the column

http://dx.doi.org/10.1016/j.chroma.2015.05.002
0021-9673/© 2015 Elsevier B.V. All rights reserved.
 D. Åsberg et al. / J. Chromatogr. A 1401 (2015) 52–59
has a different temperature than the wall. Radial temperature gra-
dients result in decreased column efficiency and should therefore
be avoided.
The effect of typical pressures found in UHPLC (ca. 1000 bar) on
retention has been investigated for a number of small compounds
and large biomolecules [4,10,11]. For uncharged species, relatively
small changes (of up to 12%) in retention factor were observed for a
pressure increase of 500 bar. For polar or ionic solutes much larger
increases, up to 50%, were reported.
Stochastic models of chromatography describe the chromato-
graphic processes at a molecular level [12]. In these models, the
chromatographic migration is represented as a random process in
which each molecule, while migrating along the column, performs
a random number of adsorption/desorption steps of random dura-
tion. Using the Characteristic Function formalism in the Fourier
domain [13] the stochastic model have been used for studying
heterogeneous adsorption, to including the effect of mobile phase
dispersion [14,15], reversed phase [16], chiral [17], size-exclusion
[18], and ion-exchange separations [19].
In this study, the pressure effect will be investigated from a
stochastic point of view. It has previously been shown that both
efficiency and peak symmetry can decrease with increasing pres-
sure [20]. Applying the stochastic model at different pressures yield
valuable insight in how the peak shape and retention depend on
pressure. This is especially true for compounds exhibiting heteroge-
neous adsorption which, to our knowledge, have not been studied
at different pressures. For heterogeneous adsorption, the van’t Hoff
plot does not yield any relevant physic-chemical information [21],
so by also studying heterogeneous adsorption at different tem-
peratures using stochastic modeling the effect of pressure and
temperature gradients in UHPLC can be investigated for a hetero-
geneous adsorption processes.
Additionally, by combining calculations of the temperature pro-
file in the column with the pressure and temperature dependence
of retention factor individual contributions of pressure and tem-
perature effects can be determined in a more rigorous way than
previously described in the literature.
The aim of this study is to investigate the effect of pressure and
temperature gradients by calculating their gradients and deter-
mining the individual contributions to retention and peak shape.
Temperature and pressure effects are first investigated separately
and then the combined effect is studied using the same model com-
pounds as in part I [22] (an uncharged, a positively and a negatively
charged and the drug omeprazole). To enable the study of pres-
sure and temperature effects for compounds with heterogeneous
adsorption, stochastic modeling will be used. Finally we will com-
pare the retention time and efficiency between HPLC and UHPLC
for omeprazole and one degradation product.
2. Theory
The stochastic theory [15,23,24] describes the chromatographic
process in terms of random variables, namely the time spent
by a molecule on the stationary phase (sojourn time) and that
elapsed between two successive adsorption/desorption events
(flying time). The history of a molecule traveling through a col-
umn can be interpreted as the sum of a random number of
adsorption/desorption steps performed by that molecule inside
the column. Mathematically, for each molecule, this history will
be the convolution integral of the density functions of the time
spent on the site. By means of the properties of the Characteris-
tic Function (i.e., the inverse Fourier transform of the probability
density function) it is possible to substitute the convolution inte-
gral with the product of the elementary characteristic functions
and to obtain both the fundamental peak shape parameters (mean,
53
variance, skew, etc.) and the chromatogram itself. In the language
of stochastic models, the chromatographic process is a Poisson
process, the chromatographic peak being the probability density
function of time spent in the column by molecules. Let us define
the average number of adsorption/desorption events by n and the
average sojourn and flying times by  s and  m , respectively. Under
the previous hypotheses, it can be demonstrated that, for a homo-
geneous surface (that is one characterized by a single sorption site
type), the average retention time is given by:
tR = nm + ns (1)
where n m is the time spent, on average, by a molecule in the mobile
phase (i.e., the column hold-up time) and n s is the average time a
molecule spends in the stationary phase.
For a heterogeneous surface, on the other hand, the corrected
retention time will depend on both the characteristics of the dif-
ferent adsorption sites and their relative abundance on the phase.
In the simple case of a surface paved with only two types of adsorp-
tion sites (2-site heterogeneous model [15]), the average retention
time can be expressed as:
 
tR = nm + n p1 s,1 + p2 s,2 (2)
where pi is the relative amount of the ith site (i = 1, 2) and n = n1 + n2 .
Through the characteristic function method, the calculation of
statistical peak moments is straightforward [13,15]. Using this
statistical moments it is possible to calculate the parameters tra-
ditionally used to describe peak asymmetry, such as peak skew or
peak excess, and to obtain expressions for the height equivalent to
a theoretical plate (H) or the number of theoretical plates (N). The
skew (S) for the 1-site and the 2-site models are given by Eqs. (3a)
and (3b), respectively [15]:
3
S= √ (3a)
2 n
3 + p 3
p1 s,1
3 2 s,2
S= √   (3b)
2 n p  2 + p  2 3/2
1 s,1 2 s,2
Column efficiency is determined according to [15]:
 k 2
1 1 2
= + (4)
N ND NM k+1
where k is the retention factor and ND is the dispersion effect from
the mobile phase estimated from an unretained marker fitted to the
exponentially modified Gaussian distribution (EMG) and calculated
from the distribution’s mean and variance. NM is defined for the
1-site model by:
NM = n (5a)
and for the 2-site model as:
  2
p1 + p2 s,2 /s,1
NM = n  2 (5b)
p1 + p2 s,2 /s,1
To obtain pure data for calculations of adsorption/desorption
kinetics, the extra column contribution to the elution zone must
be removed. This was done by fitting the peak of the void volume
marker to an exponentially modified Gaussian distribution and by
deconvolution of the elution peak of the void volume contribution.
In this work the stochastic model parameters were estimated using
the super modified sequential simplex optimization by minimi-
zation of the least-squares errors [25]. In this context it is worth
mentioning that the main objective for the stochastic models are
to correlate the changes in peak asymmetry (tailing) with pres-
sure and temperature – not to study band broadening itself. For
the latter, if the band broadening is not tailed so much, general
models/approaches are sufficient [26].
54 D. Åsberg et al. / J. Chromatogr. A 1401 (2015) 52–59
Table 1
Physicochemical properties from the manufacturer of the columns.
Property XBridge BEH-C18 AQUITY BEH-C18
Average particle size [␮m] 3.5 1.7
Pore volume [cm3 /g] 0.71 0.70
Surface area [m2 /g] 184 179
Average pore diameter [Å] 138 141
Total carbon content [%] 17.88 17.40
Surface concentration [␮mol/m2 ] 3.36 3.07
3. Materials and methods
3.1. Chemicals
The mobile phase was acetonitrile/aqueous-buffer (15 mM
phosphate buffer, pH 8.00) mixtures. Gradient grade acetonitrile
was purchased from VWR International (Radnor, PA, USA). The
buffer was prepared from water with conductivity 5.5 ␮S/m deliv-
ered from a Milli-Q Plus 185 water purification system from Merck
Millipore (Billerica, MA, USA) and from analytical grade sodium
phosphate dibasic dihydrate and sodium phosphate monobasic
dihydrate purchased from Sigma–Aldrich (St. Louis, MO, USA).
The phosphate buffer was filtered through a 0.2 ␮m nylon filter
membrane purchased from Whatman (Maidstone, UK) before it
was mixed with acetonitrile. The amount of acetonitrile varied
from 7 to 25% (v/v), depending on the solute. Benzyltriethylam-
monium chloride, BTEAC (99%), cycloheptanone, C7 (99%), sodium
2-naphtalenesulfonate, SNS (≥95%), all from Sigma–Aldrich, and
omeprazole sodium monohydrate, OM (>99%), kindly gifted by
AstraZeneca (Mölndal, Sweden), were used as solutes. BTEAC is
positively charged, SNS is negatively charged and C7 and OM are
uncharged at pH 8. The column hold-up volume was determined
with sodium nitrate (≥99.0%) purchased from Sigma–Aldrich.
3.2. Chromatographic equipment
The HPLC system was an Agilent 1200 chromatograph (Agilent
Technologies, Palo Alto, CA, USA) equipped with a binary pump, an
auto sampler, a diode-array UV-detector and a thermostated still
air column oven. The extra column volume from the auto sampler to
the detector was 0.037 mL and has been subtracted from the exper-
imental data. The HPLC column was a 100 mm × 4.6 mm XBridge
BEH C18 column with an average particle diameter of 3.5 ␮m. The
physicochemical properties of the column are reported in Table 1.
As the column was thermostated in a still air compartment, it can
be assumed that it is under adiabatic conditions [9].
The UHPLC system was a Waters Acquity UPLC H-class (Waters
Corporation, Milford, MA, USA) equipped with quaternary pump
system, an auto sampler, a diode-array UV-detector and a ther-
mostated column oven. Also in this case, one may assume the
column to be under adiabatic conditions. The extra column vol-
ume from the auto sampler to the detector was 0.027 mL and
has been subtracted from the experimental data. The UHPLC
column was a 50 mm × 2.1 mm Acquity UPLC BEH C18 with an
average particle diameter of 1.7 ␮m. The physico-chemical prop-
erties of the column are given in Table 1. The flow rates of the
respective systems used depended on the goals of the experi-
ments and are mentioned in connection to the actual experiment,
below. As example, for fundamental studies in UHPLC we used a
very low flow rate to avoid pressure and temperature gradients
whereas for optimized UHPLC experiments we used a high flow
rate.
3.3. Temperature, mass flow and pressure measurements
The temperature of the column wall was measured by attach-
ing three PT-100 (4-wire) resistance temperature detectors from
Pentronic AB (Gunnebo, Sweden) directly on the column surface.
For the UHPLC column, they were attached on the column wall at
10.7, 21.9 and 35.0 mm from the column inlet and for HPLC they
were placed at 21 and 81 mm from the inlet. A thermal adhesive
from Arctic Silver Inc. (Visalia, CA, USA) was used to attach them.
The PT-100 elements had the accuracy ±0.2 ◦ C and were verified in
house against a reference thermometer.
The total mass flow was measured by connecting a mini CORI-
FLOW Coriolis mass flow meter after the detector which was
purchased from Bronkhorst High-Tech B.V. (Ruurlo, Netherlands)
and had accuracy equal to ±0.2% of the mass flow.
The pressure at the column inlet and outlet was determined by
repeating the experiments first with the capillary going to the col-
umn inlet reconnected directly to waste and then with the column
replaced by a zero-volume union. The temperature was measured
at the flow rates 0.25, 0.50, 1.00 and 1.20 mL/min for UHPLC and at
0.40 and 1.00 mL/min for HPLC. The mobile phase was 25% acetoni-
trile as this composition corresponds to the largest viscosity of the
mobile phase [27].
3.4. Calculating temperature profiles
The experimentally measured axial temperature difference
between column inlet and outlet was less than 0.5 ◦ C in HPLC at
flow rate ≤1 mL/min and in UHPLC at flow rate ≤0.25 mL/min. This
temperature difference is deemed negligible so these conditions
were not modeled.
The modeling of temperature profiles in chromatographic
columns was done with the same method as described in refs. [8,9].
This method combines models of heat and mass transfer and mobile
phase velocity distribution. In these calculations, the mass flow
measured externally was used in place of the set up volumetric
flow rate. The external heat transfer coefficient and the param-
eter in the Blake–Kozeny–Carman correlation were estimated by
minimizing the differences between calculated and experimental
values of column outlet pressure and temperature (measured at the
third temperature sensor). The external heat transfer coefficient
was equal to 30 W/(m2 K) and the Blake–Kozeny–Carman parame-
ter was 146. At the different flow rates, calculations were validated
by comparing the estimated temperature at the column wall (at the
positions of the temperature sensors) with the experimental tem-
peratures. The agreement between calculated and experimental
data was very satisfactory, with relative errors smaller than 0.5%.
3.5. Chromatographic experiments
Triplicate analytical injections of the four compounds were done
at different temperatures in HPLC and UHPLC and for different
pressures for UHPLC. BTEAC (0.01 g/L) was studied at 7% acetoni-
trile in the eluent, SNS (0.001 g/L) at 15% and C7 (1 g/L) and OM
(0.025 g/L) at 25%. The column hold-up volume was measured with
NaNO3 (0.005 g/L) before each injection. In HPLC it was equal to
1.00, 0.96 and 0.93 mL and in UHPLC it was 0.11, 0.10, 0.10 mL for
7%, 15% and 25%, respectively. Injection volumes were 5 ␮L in HPLC
and 2 ␮L in UHPLC while all compounds were detected at 220 nm
except C7, which was detected at 280 nm. The flow rates were 1.00
and 0.13 mL/min in HPLC and UHPLC, respectively, which resulted
in maximum pressure drops of 150 and 100 bar over the columns
(negligible pressure and temperature gradients).
When investigating the effect of temperature, the temperature
was changed in 5 ◦ C increments and analytical peaks were recorded
at each temperature. The interval for HPLC was 20–40 ◦ C and for
 D. Åsberg et al. / J. Chromatogr. A 1401 (2015) 52–59
BTEAC
C7
SNS
OM
Response
1.5 2 2.5 3 3.5 4 4.5 5 5.5
Column volumes
Fig. 1. Overlaid chromatograms for HPLC (black) and UHPLC (gray). All extra-
column volumes have been corrected for and the retention volumes have been
normalized with the HPLC and UHPLC column volumes, respectively. The column
temperature was set to 40 ◦ C and the pressure drop over the column in HPLC was
100 bar and in UHPLC 800 bar, respectively. This corresponded to a flow rate of
1.0 mL/min for HPLC and 1.2 mL/min for UHPLC.
UHPLC 30–50 ◦ C. The intervals were different due to different tech-
nical limitations: the HPLC column thermostat was unable to have
stable temperatures >40 ◦ C while the UHPLC thermostat could not
operate in a reliable way at temperature <25 ◦ C. The pressure was
studied by placing a restriction capillary between the column out-
let and the detector. The extra column contributions from the
restriction capillaries were measured and accounted for in the cal-
culations, by lifting out the column and injecting the sample at
each restriction capillary set up. As restriction capillaries LC PEEKsil
tubing with inner diameter 25 ± 1 ␮m from SGE Analytical Science
(Milton Keynes, U.K.) was used with lengths 5, 10, 15 and 20 cm.
This approach allows to minimizing pressure and temperature gra-
dients over the column. The pressure used in the calculations is
taken as the average pressure in the column when assuming a lin-
ear pressure drop over the column [28] and is denoted Pavg It is
calculated as:
Pcol. inlet − Pcol. outlet
Pavg = + Pafter col.
2
where Pcol.inlet and Pcol.outlet is the pressure at the column inlet
and outlet while Pafter col. is the total pressure drop from the
column outlet to atmosphere including the restriction capillary.
The pressures in Eq. (6) were determined separately for all experi-
mental systems. Five different pressures were investigated for each
solute; Pavg ,25% = 53, 174, 302, 420, 550 bar, Pavg ,15% = 53, 175, 314,
419, 552 bar and Pavg ,7% = 51, 167, 301, 398, 524 bar, where the “%”
denotes the fraction of acetonitrile in the mobile phase.
4. Results and discussion
To demonstrate that a direct method transfer from HPLC
to UHPLC is not always straightforward, the four compounds
employed in this work have been eluted on the two columns under
isocratic conditions and typical flow rates of HPLC and UHPLC (1.0
and 1.2 ml/min, respectively). It is worth to mention that these
flow rates were not obtained as the optimal flow rate in a van
Deemter curve, as the study of column efficiency was not the pur-
pose of this work. Temperature in both cases was set to 40 ◦ C.
Overlaid chromatograms are presented in Fig. 1 where, for the sake
of comparison, retention is expressed as column volumes instead of
retention time. It is evident from Fig. 1 that the retention in column
volumes is longer for UHPLC (gray lines) compared to HPLC (black
lines), especially for the late eluting peaks which also exhibit more
tailing in UHPLC. This difference in retention might be caused by
factors such as different pressures, temperature gradients and col-
umn chemistry. From calculating the efficiency with the moment
55
2.2
2
1.8
1.6
ln(k)
1.4
1.2
1
0.8
0 200 400 600
Pavg [bar]
Fig. 2. Experimental retention factors for BTEAC (circles), C7 (squares), SNS (dia-
monds), and OM (triangles) for different pressures. Mobile phase compositions were
7, 15 and 25% acetonitrile for BTEAC, SNS and C7/OM, respectively and flow rate was
0.13 mL/min.
method for HPLC and UHPLC in Fig. 1, it could be seen that, in this
case, HPLC condition yielded the highest efficiency. The average
efficiency of three injections were for HPLC 6230, 9430, 10730 and
9330 and for UHPLC 1610, 3560, 5050 and 4330 for BTEAC, SNS, C7
and OM, respectively. Reasons for the lower efficiency in UHPLC
are most likely that the HPLC column was twice as long as the
UHPLC one, the column loading was higher in the UHPLC exper-
iments and that the extra-column volume in the UHPLC system
was a larger fraction of the column volume. Temperature gradi-
ents causing a radial thermal heterogeneity and pressure gradients
could also affect the efficiency.
Through the aid of the stochastic modeling and by ln(k) vs. 1/T
and ln(k) vs. P plots we will try to investigate the individual contrib-
utions of temperature and pressure to retention and peak shape.
(6)
4.1. Temperature and pressure effects on retention
4.1.1. Pressure dependence
The slope of ln(k) vs. Pavg plots is equal to Vm = Vstat. − Vmob , i.e.
the change in solute molar volume associated with the transition
between the stationary and mobile phase [29]. Generally the solute
molar volume is larger in the mobile phase, i.e. Vm < 0. According
to Le Chatelier principle, when the pressure is increased the equilib-
rium between solute molecules in the mobile and stationary phase
will be pushed toward the stationary phase, resulting in increasing
retention time [29,30].
The logarithm of the retention factor for different pressures was
fitted with linear regression and the result is presented in Fig. 2.
The relationship between ln(k) and Pavg is linear with a R2 value
larger than 0.980. Calculated Vm values were −11.9 ± 1.0 cm3 /mol
for BTEAC, −15.5 ± 0.3 cm3 /mol for SNS, −3.4 ± 0.3 cm3 /mol for C7
and −18.4 ± 0.5 cm3 /mol for OM given with a 95% confidence inter-
val. These observations are in good agreement with those reported
by Fallas et al. [10,11] for a similar system. The negative Vm is
attributed to the partial loss of the solvation layer of the solutes
when they move from the mobile to the hydrophobic station-
ary phase. The difference in Vm between uncharged and ionized
solutes is believed to be due to the hydration of the ions which are
partially lost when entering the stationary phase. However, this
clearly shows that pressure is a factor that needs to be considered
because it could affect the selectivity as well as retention.
56 D. Åsberg et al. / J. Chromatogr. A 1401 (2015) 52–59
Fig. 3. Calculated temperature profile in UHPLC for the mobile phase 25/75, v/v
acetonitrile/phosphate buffer at flow rate 1.2 mL/min. The dotted line represents
the inner column wall. At 50 mm the center of the column is warmest (55 ◦ C) and
the column wall is at ca 52 ◦ C; radius 0 is the center of the column. The pressure
over the column is 782 bar.
4.1.2. Temperature dependence
To better understand the effect of viscous heating, the temper-
ature gradients in the column were quantified. The 2-dimensional
temperature profile corresponding to UHPLC conditions of 25%
acetonitrile and flow rate 1.2 mL/min (same as Fig. 1) has been cal-
culated and shown in Fig. 3. The dotted line at 1.05 mm represents
the inner column wall. Because the column temperature profile was
assumed to be radially symmetrical, only half of the temperature
contour plot is shown. As can be seen from Fig. 3, the tempera-
ture along the column, i.e. longitudinally, increased 16 ◦ C from
the inlet to the outlet and the temperature inside the column from
centrum to wall, i.e. radially, decreased 2 ◦ C at most. The values
calculated for this specific system are close to those reported in the
literature [6,9] for similar pressure drops over the column.
The effect of the temperature gradient on retention was esti-
mated by calculating the local propagation speeds along the column
by first calculating the geometric radial average temperature using
the temperature profile in Fig. 3. Then the temperature dependence
of the retention factor was determined by fitting the logarithm of
retention factors to the reciprocal temperature, Fig. 4. Linearity of
ln(k) vs. 1/T was observed in all cases with R2 -values above 0.990,
except for BTEAC for which a slight nonlinearity was observed with
R2 -value 0.960. The retention factor of all solutes decreases with
2 affected by pressure. These results also suggest that in the method
1.5
ln(k)
1
0.5
3.1 3.2 3.3
1000/T [1/K]
Fig. 4. Experimental retention factors for BTEAC (circles), C7 (squares), SNS (dia-
monds), and OM (triangles) for different temperatures. Mobile phase compositions
were 7, 15 and 25% acetonitrile for BTEAC, SNS and C7/OM, respectively and flow
rate was 0.13 mL/min.
1.3 Temp. Pressure Observed
1.2
Normalized k
1.1
0.9
BTEAC SNS C7 OM
Fig. 5. The retention factor is compared for four different cases in UHPLC. The base-
line is taken as the retention factor at low flow rate 0.13 mL/min where pressure and
temperature gradients are negligible; the bars denoted “only T” represent the effect
caused only by the temperature gradient; “only P” denotes the case with only the
pressure effect present and “observed” represents actual experimental result where
both pressure and temperature effects are present.
temperature, which is the common behavior in RP-LC for moderate
temperature variations. Combining this information the retention
time was estimated by integrating the solute local propagation
speed along the UHPLC column. The uncharged solute, C7, is least
affected by the temperature gradient with a decrease in k of roughly
5%. The positively charged solute, BTEAC, is somewhat more sensi-
tive and k decreases about 10% while the negatively charged solute,
SNS, is most affected by temperature and k decreases almost 14%.
The retention factor of OM decreases approximately 9%, which
places it in the same region as the other two charged solutes.
4.1.3. The relative importance of temperature and pressure
gradients
In an attempt to compare the relative importance of pressure
and temperature on retention in UHPLC, the results from Sections
4.1.1 and 4.1.2 have been combined in Fig. 5, where the retention
factors of the four investigated molecules have been normalized
against the respective retention factor at 0.13 mL/min (where pres-
sure and temperature gradients are negligible). In Fig. 5, white
bars show the estimated contribution for temperature, while gray
bars show the pressure contribution, which has been estimated by
assuming a linear pressure gradient over the column. Finally, black
bars represent the observed, experimental retention factors found
in UHPLC.
The pressure and temperature have opposite effects on reten-
tion and will therefore, to a certain degree, cancel each other. Under
typical UHPLC conditions, the pressure effect is always larger than
the temperature effect, except for C7. For C7, which is uncharged,
the pressure effect is very small. Both the pressure and tempera-
ture effects are solute dependent. In particular, our data show that
charged solutes and those with larger molecular weight are most
transfer from HPLC to UHPLC, especially if charged solutes are con-
sidered, the effect of the pressure gradient along the column more
than that of temperature gradient should be taken into account, as
it is the dominating one.
4.2. Stochastic modeling
The adsorption isotherms for the modeling compounds has
previously been investigated and it was found that C7 and OM
exhibited homogeneous adsorption while SNS and BTEAC had het-
erogeneous adsorption [22]. Since the peak shape was also very
symmetrical for C7, OM and SNS, they were described by 1-site
stochastic models. One must stress that heterogeneous adsorption
not necessarily results in measurable heterogeneous kinetics.
 D. Åsberg et al. / J. Chromatogr. A 1401 (2015) 52–59 57

14 3 5
a) c) e)
12
10
4

N × 103
n1 × 10

n2
8 2.5 2.5
6
4
2 2 0

8 6 1.4
b) d) f)
1.3
6 4
[ms]

τs,2 [s]
1.2

S
s,1

4 2
τ

1.1

2 0 1
20 30 40 50 20 30 40 50 20 30 40 50
T [°C] T [°C] T [°C]

Fig. 6. Stochastic modeling of BTEAC which is described by a 2-site model at different temperatures. N is the column efficiency determined with Eq. (4) and S is the skew
determined with Eq. (3b).

For very symmetrical peak shapes as those observed under our
conditions, the number of adsorption/desorption events (n) and
adsorption sojourn time ( s ) are strictly correlated, so that nonlin-
ear fitting cannot differentiate between them. Therefore, any trends
with temperature or pressure seen for this kind of peaks may be an
artifact of nonlinear fitting. As a consequence, only the range of
these parameters is given here. The conclusion is that for solutes
with very symmetrical peaks, the stochastic approach is unable to
single out pressure and temperature effects.
For BTEAC, the elution peaks are asymmetrical so the nonlinear
fitting could differentiate between contributions of n and  s .
4.2.1. Temperature dependence
For OM, C7 and SNS (1-site model) n is similar for all three solutes
(10 000–20 000) in HPLC and UHPLC with the values for UHPLC
being slightly lower.  s is between 10 and 20 ms and the values are
slightly higher for UHPLC than for HPLC. These observations reflect
the fact that the measured column efficiency is higher in HPLC.
BTEAC adsorption is described by a 2-site adsorption isotherm
[22] and the elution peaks are tailing so the 2-site stochastic model
has been employed to fit the chromatograms. Site-1 is where the
majority of all adsorption/desorption events take place and the
sojourn time for these sites are in the millisecond scale, Fig. 6. On the
other hand, at site-2 only a few adsorption/desorption events take
place. However, the sojourn time found on these sites is roughly one
thousand times longer than on site-1. We observe that the average
number of adsorption/desorption events increases with increasing
temperature and that the sojourn time decreases with increases
temperature for both sites and systems. From a fundamental per-
spective, increasing the temperature will result in an increased
diffusion coefficient both due to the temperature itself but also
due to a reduction in viscosity, as described by the Stokes–Einstein
equation. The faster kinetics will result in increased n and decreased
 s as observed. The scatter of the data points in Fig. 6, especially for
n1 and n2 , is most likely due to a combination of numerical errors
in the optimization and experimental errors.
The skew can be seen as a measure of the peak symmetry where
a large positive skew means that the peak is tailing and a nega-
tive skew means that the peak is fronting. In this case the skew
decreases with temperature for both HPLC and UHPLC, Fig. 6f, which
indicates that the peak shape becomes more symmetrical when the
temperature is increased. Relating the observation of decreasing
skew to Eq. (3b), it is due to the increase in n2 and the decrease in
 s,2 for the slow site with increasing temperature. This also clearly

7 2.3 900
a) c) e)
6 800
2.25
4
n1 × 10

n2

5 700
2.2
4 600

3 2.15 500

8 7 1.4
b) d) f)

7 6
τs,1 [ms]

τs,2 [s]

1.35
S

6 5

5 4 1.3
0 200 400 600 0 200 400 600 0 200 400 600
Pavg [bar] Pavg [bar] Pavg [bar]

Fig. 7. Stochastic modeling of BTEAC which is described by a 2-site model at different pressures. N is the column efficiency determined with Eq. (4) and S is the skew
determined with Eq. (3b).
58 D. Åsberg et al. / J. Chromatogr. A 1401 (2015) 52–59
10
HPLC
5 and Rs the resolution factor.
Response [mAU]
0
10
UHPLC
5 –
0
0 1 2 3
Retention time [min]
Fig. 8. Chromatogram from HPLC and UHPLC for omeprazole (large peak) and the
degradation product H168/66 (small peak) for identical mobile phase, stationary
phase and column temperature. Flow rates are 2 mL/min in HPLC and 0.86 mL/min
in UHPLC.
shows that the tailing observed for BTEAC (see Fig. 1) is due to
heterogeneous kinetics.
4.2.2. Pressure dependence
For C7, OM and SNS, that is the compounds described by 1-site
stochastic model, n has been found to be nearly constant while
 s increase slightly with increasing pressure. In particular, n was
between 8000 and 14 000 and  s between 10 and 22 ms. It has
previously been reported that increasing pressure could result in
decreasing efficiency [20]. One probable explanation to this could
be that the viscosity increases with the pressure [28] and as a
result through the Stokes–Einstein equation this will result in an
increased diffusion coefficient. From the stochastic model perspec-
tive this will result in reduction in n because slower diffusion results
in fewer adsorption/desorption events take place. The sojourn time
( s ) will increase because it takes longer time for the solute to dif-
fuse out from the stationary phase. In this study for C7, OM and SNS
we observed an increasing sojourn time but more or less no effect
on n.
The results for BTEAC are presented in Fig. 7 and show that on
the “fast” adsorption site there is an increase of n1 and a decrease
in  s,1 . While the “slow”, second site, presents a nearly 50% increase
in  s,2 with a 500 bar pressure increase. As a consequence, the peak
skew is increased, Fig. 7f and efficiency decrease when the pressure
is increased, Fig. 7e. This increase in peak skew is due to the fact
that the solute molecules spend on average a longer time adsorbed
on the slow, second site when the pressure increases which makes
the tailing more pronounced. To understand heterogeneous kinet-
ics pressure dependency better it should be necessary to include
more experiments with focus on several different kinds of basic
compounds, however this is outside the scope of the study. So the
presented result should be viewed as a first observation.
From Figs. 6 and 7, one may conclude that high temperature and
low pressure improve peak shape for BTEAC. We believe that, even
though more information must be gathered to draw any general
conclusions, this is an interesting finding.
4.3. Practical implications
To compare the performance of HPLC and UHPLC, retention
time, column efficiency and resolution for a degraded omepra-
zole sample containing the degradation product H168/66 in small
amounts were investigated. The same mobile phase, temper-
ature and stationary phase chemistry were used. The column
dimensions and particle diameters differed; the HPLC column
was 100 mm × 4.6 mm, 3.5 ␮m and the UHPLC column was
50 mm × 2.1 mm, 1.7 ␮m. The flow rate and injection volume for
Table 2
Parameters for the separation of omeprazole and the degradation product H168/66
using HPLC and UHPLC columns, respectively. tr is the retention time, N the efficiency
Solute Parameter HPLC UHPLC
OM tr [min] 2.61 0.72
H168/66 tr [min] 1.74 0.45
OM N 5970 4230
H168/66 N 6240 5460
Rs 9.87 8.99
HPLC was 2.0 mL/min and 10 ␮L and for UHPLC they were scaled
according to Ref. [31] which yielded 0.86 mL/min and 1 ␮L. The
scaling was done to keep the reduced linear velocity of the mobile
phase constant. At these flow rates the pressure drop over the HPLC
column was 165 bar and 570 bar over the UHPLC column.
The chromatograms are presented in Fig. 8, where the retention
time for omeprazole is 3.6 longer in HPLC. The retention times, effi-
ciencies (moment method) and resolution factors (half-height) are
presented in Table 2. The efficiency of omeprazole was 29% higher
in HPLC and efficiency for H168/66 was 13% higher, which resulted
in a resolution factor that was about 9% higher in HPLC. However,
the loss in efficiency and resolution when switching to UHPLC are
very minor compared to the reduction in run time and resulting
decrease in solvent consumption which makes UHPLC the primary
choice in this case. The individual effects of pressure and temper-
ature gradients in UHPLC are not directly observable since their
effects are convoluted when comparing the different modes in this
way.
5. Conclusions
The aim of this study has been to investigate how pressure
and temperature affect retention and peak shape of the solutes
in HPLC and UHPLC. To this end, the chromatographic behavior
of four model compounds with different physicochemical proper-
ties has been modeled from both a thermodynamic and a kinetic
(microscopic-stochastic) viewpoint. The thermodynamic models
showed that the difference in solute molar volume for adsorbed
and in bulk solution, which determines the pressure dependence
of the retention factor, was largest for the polar solute omeprazole
which also had the largest molecular weight. When combining the
calculated temperature gradient and the linear pressure gradient
the individual contributions on retention could be determined. The
effect of the pressure gradient was found to be the dominating one
and should therefore be taken into account when switching from
HPLC to UHPLC.
From the stochastic modeling of the tailing, basic solute it was
evident that an increase in temperature yielded an increase in
average number of adsorption/desorption events while the aver-
age time spent by a molecule in the stationary phase was slightly
decreasing. For increased pressure the effect was the opposite.
Therefore a high temperature and a low pressure yielded low tail-
ing. Even though from different perspectives, the conclusions of
these models converge in showing that the effect of pressure gra-
dient along the column is as important as that of viscous heating.
With charged and polar compounds, we found that the impact of
the pressure gradient is even more important than that of viscous
heating in UHPLC for the investigated experimental conditions.
Acknowledgements
We are grateful to Waters that provided an ACQUITY UPLC H-
Class system with support and technical advice during the project.
We are also grateful to Anders Karlsson at AstraZeneca R&D Möl-
ndal for initiating this project and for most valuable discussions.
 D. Åsberg et al. / J. Chromatogr. A 1401 (2015) 52–59
This work was supported by the Swedish Research Council (VR) in
the project “Fundamental studies on molecular interactions aimed
at preparative separations and biospecific measurements” (grant
number 621-2012-3978) and by the Swedish Knowledge Foun-
dation for the KK HÖG 2014 project “SOMI: Studies of Molecular
Interactions for Quality Assurance, Bio-Specific Measurement &
Reliable Supercritical Purification” (grants number 20140179). AC
thanks the Italian University and Scientific Research Ministry (PRIN
2012ATMNJ 003).
References
[1] S. Fekete, J. Schappler, J.-L. Veuthey, D. Guillarme, Current and future trends in
UHPLC, TrAC 63 (2014) 2–13.
[2] S. Fekete, E. Oláh, J. Fekete, Fast liquid chromatography: the domination of
core–shell and very fine particles, J. Chromatogr. A 1228 (2012) 57–71.
[3] M. Szalka, J. Kostka, E. Rokaszewski, K. Kaczmarski, Analysis of related sub-
stances in bisoprolol fumarate on sub-2-(m adsorbents, Acta Chromatogr. 24
(2012) 163–183.
[4] S. Fekete, J.-L. Veuthey, D.V. McCalley, D. Guillarme, The effect of pressure and
mobile phase velocity on the retention properties of small analytes and large
biomolecules in ultra-high pressure liquid chromatography, J. Chromatogr. A
1270 (2012) 127–138.
[5] H. Poppe, J.C. Kraak, Influence of thermal conditions on the efficiency of
high-performance liquid chromatographic columns, J. Chromatogr. 282 (1983)
399–412.
[6] F. Gritti, G. Guiochon, Complete temperature profiles in ultra-high-pressure
liquid chromatography columns, Anal. Chem. 80 (2008) 5009–5020.
[7] F. Gritti, M. Martin, G. Guiochon, Influence of viscous friction heating on the effi-
ciency of columns operated under very high pressures, Anal. Chem. 81 (2009)
3365–3384.
[8] K. Kaczmarski, J. Kostka, W. Zapała, G. Guiochon, Modeling of thermal pro-
cesses in high pressure liquid chromatography: I. low pressure onset of thermal
heterogeneity, J. Chromatogr. A 1216 (2009) 6560–6574.
[9] K. Kaczmarski, F. Gritti, J. Kostka, G. Guiochon, Modeling of thermal processes
in high pressure liquid chromatography: II. thermal heterogeneity at very high
pressures, J. Chromatogr. A 1216 (2009) 6575–6586.
[10] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Investigation of the
effect of pressure on retention of small molecules using reversed-phase
ultra-high-pressure liquid chromatography, J. Chromatogr. A 1209 (2008)
195–205.
[11] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Further investigations of the
effect of pressure on retention in ultra-high-pressure liquid chromatography,
J. Chromatogr. A 1217 (2010) 276–284.
[12] J.C. Giddings, H. Eyring, A molecular dynamic theory of chromatography, J. Phys.
Chem. 59 (1955) 416–421.
[13] F. Dondi, M. Remelli, The characteristic function method in the stochastic theory
of chromatography, J. Phys. Chem. 90 (1986) 1885–1891.
59
[14] A. Cavazzini, M. Remelli, F. Dondi, Stochastic theory of two-site adsorption
chromatography by the characteristic function method, J. Microcol. Sep. 9
(1997) 295–302.
[15] A. Cavazzini, M. Remelli, F. Dondi, A. Felinger, Stochastic theory of multiple-site
linear adsorption chromatography, Anal. Chem. 71 (1999) 3453–3462.
[16] A. Felinger, Molecular movement in an HPLC column: a stochastic analysis,
LCGC N. Am. 22 (2004) 642–647.
[17] M. Enmark, J. Samuelsson, T. Undin, T. Fornstedt, Characterization of an
unusual adsorption behavior of racemic methyl-mandelate on a tris-(3,5-
dimethylphenyl) carbamoyl cellulose chiral stationary phase, J. Chromatogr.
A 1218 (2011) 6688–6696.
[18] F. Dondi, A. Cavazzini, M. Remelli, A. Felinger, M. Martin, Stochastic theory
of size exclusion chromatography by the characteristic function approach, J.
Chromatogr. A 943 (2002) 185–207.
[19] K. Horváth, M. Olajos, A. Felinger, P. Hajós, Retention controlling and peak shape
simulation in anion chromatography using multiple equilibrium model and
stochastic theory, J. Chromatogr. A 1189 (2008) 42–51.
[20] K. Okusa, Y. Iwasaki, I. Kuroda, S. Miwa, M. Ohira, T. Nagai, et al., Effect of
pressure on the selectivity of polymeric C18 and C30 stationary phases in
reversed-phase liquid chromatography. Increased separation of isomeric fatty
acid methyl esters, triacylglycerols, and tocopherols at high pressure, J. Chro-
matogr. A 1339 (2014) 86–95.
[21] F. Gritti, G. Guiochon, Adsorption mechanisms and effect of temperature in
reversed-phase liquid chromatography. Meaning of the classical Van’t Hoff plot
in chromatography, Anal. Chem. 78 (2006) 4642–4653.
[22] D. Åsberg, M. Leśko, J. Samuelsson, K. Kaczmarski, T. Fornstedt, Method transfer
from high-pressure liquid chromatography to ultra-high-pressure liquid chro-
matography. I. A thermodynamic perspective, J. Chromatogr. A 1362 (2014)
206–217.
[23] A. Felinger, Molecular dynamic theories in chromatography, J. Chromatogr. A
1184 (2008) 20–41.
[24] J.C. Giddings, Dynamics of Chromatography: Principles and Theory, Marcel
Dekker, New York, 1965.
[25] E. Morgan, K.W. Burton, G. Nickless, Optimization using the super-modified
simplex method, Chemometr. Intell. Lab. Syst. 8 (1990) 97–107.
[26] F. Gritti, G. Guiochon, Mass transfer mechanism in chiral reversed phase liquid
chromatography, J. Chromatogr. A 1332 (2014) 35–45.
[27] J. Billen, K. Broeckhoven, A. Liekens, K. Choikhet, G. Rozing, G. Desmet, Influ-
ence of pressure and temperature on the physico-chemical properties of mobile
phase mixtures commonly used in high-performance liquid chromatography,
J. Chromatogr. A 1210 (2008) 30–44.
[28] M. Martin, G. Guiochon, Effects of high pressure in liquid chromatography, J.
Chromatogr. A 1090 (2005) 16–38.
[29] V.L. McGuffin, S.-H. Chen, Molar enthalpy and molar volume of methylene and
benzene homologues in reversed-phase liquid chromatography, J. Chromatogr.
A 762 (1997) 35–46.
[30] P. Atkins, J. De Paula, Physical Chemistry, 9th ed., Oxford University Press,
Oxford, 2010.
[31] 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) 475–482.
 Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281

Contents lists available at ScienceDirect

Journal of Pharmaceutical and Biomedical Analysis

journal homepage: www.elsevier.com/locate/jpba

A quality control method enhancement concept—Continual

improvement of regulatory approved QC methods
Dennis Åsberg a , Mikael Nilsson b , Susanne Olsson b , Jörgen Samuelsson a , Olof Svensson c ,
Silke Klick c , Julie Ennis d , Paul Butterworth d , Denise Watt d , Stavroula Iliadou d ,
Angelica Karlsson c , Joanne T. Walker e , Kate Arnot d , Norb Ealer c , Kerstin Hernqvist c ,
Karin Svensson f , Ali Grinell d , Per-Ola Quist f , Anders Karlsson c,∗ , Torgny Fornstedt a
a
Department of Engineering and Chemical Sciences, Karlstad University, SE-651 88 Karlstad, Sweden
b
Cambrex R & D, SE-691 33 Karlskoga, Sweden
c
AstraZeneca R & D, SE-431 83 Mölndal, Sweden
d
AstraZeneca, Charter Way, SK10 2NA Macclesfield, UK
e
Medimmune, Medimmune Way, MD 20878 Gaithersburg, USA
f
AstraZeneca, SE-152 57 Gärtuna, Sweden

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

Article history: Quality Control methods (QC-methods) play an important role in the overall control strategy for drug
Received 12 April 2016 manufacturing. However, efficient life-cycle management and continual improvement are hindered due
Received in revised form 8 June 2016 to a variety of post-approval variation legislations across territories and a lack of harmonization of the
Accepted 10 June 2016
requirements. As a result, many QC-methods fall behind the technical development. Developing the
Available online 14 June 2016
QC-method in accordance with the Quality by Design guidelines gives the possibility to do continual
improvements inside the original Method Operable Design Region (MODR). However, often it is necessary
Keywords:
to do changes outside the MODR, e.g. to incorporate new technology that was not available at the time
Method enhancement concept
Quality by design
the original method was development. Here, we present a method enhancement concept which allows
Continual improvement minor adjustments, within the same measuring principle, outside the original MODR without interaction
HPLC with regulatory agencies. The feasibility of the concept is illustrated by a case study of a QC-method based
Method transfer on HPLC, assumed to be developed before the introduction of UHPLC, where the switch from HPLC to
UHPLC is necessary as a continual improvement strategy. The concept relies on the assumption that the
System Suitability Test (SST) and failure modes are relevant for other conditions outside the MODR as
well when the same measuring principle is used. It follows that it should be possible to move outside the
MODR as long as the SST has passed. All minor modifications of the original, approved QC-method must
be re-validated according to a template given in the original submission and a statistical equivalence
should be shown between the original and modified QC-methods. To summarize, revalidation is handled
within the pharmaceutical quality control system according to internal change control procedures, but
without interaction with regulating agencies.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction and life-cycle management of the QC-methods ensure proper qual-

Quality Control methods (QC-methods) are an important part of
the chemistry, manufacturing and controls documentation submit-
ted to regulatory agencies to support marketing applications, and
as such they are part of the manufacturer’s commitment towards
the regulators. Careful and well-planned development, validation
∗ Corresponding author.
E-mail address: Anders.AS.Karlsson@astrazeneca.com (A. Karlsson).
ity control testing of drug substance and drug product batches [1].
The purpose of the QC-method is to reliably measure drug sub-
stance and drug product quality attributes in order to secure safety
and efficacy aspects for the patient. QC-methods are therefore an
important part of the overall control strategy. However, for a prod-
uct registered in multiple markets efficient lifecycle management,
innovation and continual improvement are hindered due to a vari-
ety of post-approval variation legislations across territories and
a lack of harmonization of the requirements [2]. Even relatively
minor changes to a registered method require some kind of reg-

http://dx.doi.org/10.1016/j.jpba.2016.06.018
0731-7085/© 2016 Elsevier B.V. All rights reserved.
274 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281

ulatory action, e.g. reporting in an annual report, notification, or

submission of a variation that requires regulatory approval prior
to implementation. As a principle, implementation of a change to a
QC-method has to wait for the last territory to approve the varia-
tion, which can take months, sometimes even years. The amount of
regulatory notifications and prior-approval supplements, in com-
bination with the time until a change can actually be implemented,
is an efficient blocker to innovation and continual improvement,
and as a result many QC-methods have lagged behind the technical
development. A typical example where a modification of the origi-
nal QC-method may become desirable due to availability of better
equipment is the change of chromatographic columns, including
stationary phase chemistry, particle size and column dimensions.
Quality by Design (QbD) is a framework jointly developed by
regulators and the pharmaceutical industry, the principles being
outlined in ICH guidelines Q8-11 [3–6]. The overall purpose of QbD
is to facilitate a more science and risk based development, facilitate
continual improvement and ultimately ensure quality products and
a reliable supply to patients [2].
While QbD is a framework to be applied to the development of
a product and its manufacturing process, the principles laid out in
ICH Q8-11 may, in a wider sense, be applicable to QC-methods as
well. An attempt was made in 2010 to apply QbD to QC-methods
by PhRMA and EFPIA who developed a concept where QC-methods
are described by their Analytical Target Profile (ATP) [7]. The ATP is
a set of acceptance criteria for method performance and allows any
method where compliance with the criteria could be demonstrated
as a QC-method [7]. This concept was not endorsed by regulators.
However, in 2013 as part of a Q&A document to the EMA-FDA pilot
program for parallel assessment of Quality-by-Design applications
[8] the FDA and EMA stated their view of an ATP as being “accept-
able as a qualifier of the expected method performance by analogy
to the Quality Target Product Profile as defined in ICH Q8 (R2)”, but
not when switching between methods with different principles,
e.g. HPLC to NIR [8].
An approach for analytical QC-method development and life Fig. 1. Schematic description of QC-method development in accordance with the
QbD framework.
cycle management using principles outlined in ICH Q8-11 is sum-
marized in Fig. 1. Step 1 is to determine what the objective is and
what should be measured. Then an appropriate technique is chosen long as failure modes are the same and the SST has been constructed
and the desired performance of the method is detailed in a num- under careful consideration of the failure the SST can be constructed
ber of criteria, written as a method Analytical Target Profile (mATP, in a way that it is relevant not only for the MODR studied but also for
defined as a “within principle” ATP) that must be fulfilled. In step other conditions provided the same measurement principle is used.
2, risk assessments are performed as part of method development For measurement of the same Quality Attribute it can be assumed
and validation, where failure modes are assessed, evaluated, and that failure modes are the same when using the same measure-
appropriate measure taken to ensure the long term performance ment principle. Therefore, it should be possible to move outside
of the method in accordance with the mATP. This should be done the MODR as long as a passed SST ensures method performance
by involving analysts and experts with experience of the method and generation of reliable results. For chromatographic methods
being developed and prior knowledge of the technique and similar such a move could be the adjustment of parameters outside the
methods. In step 3 factors identified in the risk assessment should MODR, but also the use of a different type of stationary phase or
be evaluated, e.g. by employing Design of Experiments (DoE) with column dimensions.
multivariate statistical tools. Based on the risk assessments, the To illustrate the above concept, it can be assumed that a QC-
subsequent evaluation of failure modes and the responses stud- method uses High Performance Liquid Chromatography (HPLC) and
ied in DoE, a set of parameter ranges called the Method Operable was developed before the introduction of Ultra High Performance
Design Region (MODR) can be defined within which performance Liquid Chromatography (UHPLC) in 2004. UHPLC parameters would
of the method is fully understood. Step 4 involves construction of not fit into the original MODR. Since HPLC and UHPLC is essentially
controls to ensure adequate performance in each operation of the the same technique with the main difference being that smaller
method, typically in the form of a System Suitability Test (SST). The particles are used in the column packing material in UHPLC which
final step is continual improvement which can be done in the form results in higher back pressures [15], failure modes are the same
of adjustments of the working point inside the validated MODR. A and a change from HPLC to UHPLC could be seen as continual
number of case studies following this approach can be found in the improvement of an already approved analytical QC-method. The
literature [9–14]. challenge is to enable adjustments outside the original MODR,
The approach described above allows continual improvement impossible to incorporate when the method was first developed,
of the method inside the MODR, in a similar manner as a manu- as a continual improvement strategy.
facturing process may be moved within a registered Design Space. This paper describes a concept for enhanced QC-methods where
However, unlike a manufacturing process, QC-testing of a drug sub- a science based method control strategy is put in focus to enable
stance or drug product may commence only if the SST has passed. As “within principle” continual improvement. The method control
 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281
strategy includes an SST which is constructed based on the fail-
ure modes to ensure performance in each operation of the method,
and a set of method performance verification criteria that ensure
long term performance of the method after minor adjustments of
the method. The latter will be a re-validation managed within the
Pharmaceutical Quality System and according to internal change
control procedures, without interaction with regulatory agencies.
Statistical equivalence between results obtained before and after
the adjustment should also be part of the Control Strategy.
The aim of this study is to illustrate the method enhancement
concept with a case study where a QC-method for the proton-pump
inhibitor esomeprazole [16] is developed for HPLC and then trans-
ferred to UHPLC as a continual improvement strategy. We aim to
show the feasibility of making changes to an approved QC-method
outside the MODR without, but within the same measuring princi-
ple, without regulatory interaction. The screening and optimization
steps of a LC-method are outside the scope of this study and the
reader is referred to other work [10–13]. Also, all the impurities
listed in the pharmacopeias for esomeprazole is not considered
here, only those relevant to adequately demonstrate the method
enhancement concept.
2. Material and methods
2.1. Chemicals
The solvents used were gradient grade acetonitrile and
methanol. Phosphate buffer was prepared from MilliQ-water with
conductivity 18.2 M cm and from analytical grade sodium phos-
phate dibasic and sodium phosphate monobasic. The phosphate
buffer was filtered through 0.2 ␮m nylon filter membrane before
use. Omeprazole and the impurities H431/41, H168/66, H193/61
and H168/22 were reference standards from AstraZeneca (Mölndal,
Sweden) and the Nexium® pills were commercially available 20 mg
delayed-release capsules stored in commercial packages at 40 ◦ C,
75% RH for 6 months. The column hold-up time marker, sodium
nitrate (≥99.0%), was purchased from Sigma-Aldrich (St. Louis, MO,
USA).
2.2. Standard and sample preparation
Omeprazole and the impurities were first dissolved in methanol
which was subsequently mixed with phosphate buffer (25 mM,
pH 11) to 10/90, v/v methanol/buffer. Standard concentra-
tion of omeprazole was 0.150 mg/mL and for the impurities
0.00015 mg/mL which corresponds to 0.1% of the omeprazole con-
centration. For the Nexium pills, the capsules were emptied and
the granules from 20 capsules were pooled and ground to a fine
powder, which was stored in darkness. A standard sample corre-
sponded to about 0.6 mg Nexium powder/mL. The Nexium sample
was placed in an ultra-sonication bath prior to filtration to ensure
complete dissolution. All sample solutions were filtered through a
0.2 ␮m PTFE syringe filter and stored in brown laboratory glass to
protect from light.
2.3. Chromatographic instrumentation
The experiments denoted with HPLC were performed on an Agi-
lent 1200 chromatograph (Agilent Technologies, Palo Alto, CA, USA)
equipped with a binary pump, an auto sampler, a diode-array UV-
detector and a thermostated column oven. The dwell volume from
the pump to the column inlet was 1.04 mL (determined according
to ref. [17]) and the extra-column volume from the auto sampler to
the detector was 0.039 mL. The HPLC column was a 100 × 4.6 mm
XBridge BEH C18 column from Waters (Milford, MA, USA) with an
275
average particle diameter of 3.5 ␮m. The column hold-up volume,
estimated with sodium nitrate, was 1.040 mL.
The experiments denoted with UHPLC were performed on a
Waters Acquity UPLC H-class (Waters Corporation, Milford, MA,
USA) equipped with quaternary pump system, an auto sampler,
a diode-array UV-detector and a thermostated column oven. The
dwell volume from the pump to the column inlet was 0.38 mL
and the extra-column volume from the auto sampler to the detec-
tor was 0.029 mL. The UHPLC column was a 50 × 2.1 mm Acquity
UPLC BEH C18 column with an average particle diameter of 1.7 ␮m.
The column hold-up volume, estimated with sodium nitrate, was
0.106 mL.
For both HPLC and UHPLC, the UV detection was done at 302 nm.
Eluent A was 10/90, v/v, acetonitrile/phosphate buffer (pH 8.0,
10 mM) and eluent B was 80/20, v/v, acetonitrile/phosphate buffer
(pH 8.0, 3 mM). Column temperature were 20 ◦ C for HPLC and 40 ◦ C
for UHPLC while the flow rates were 1.00 mL/min and 0.43 mL/min,
respectively. The chromatographic methods are discussed in detail
in Section 3.
2.4. Procedure
Linearity was assessed with standard curves which were
obtained by two replicate injections at the levels 0.04, 0.08, 0.1, 0.3,
0.5 and 0.6% for the impurities and at the levels 0.04, 10, 30, 60, 100
and 120% for omeprazole. The limit of quantitation for each com-
pound was determined from the standard deviation of the standard
curves [18]. The system repeatability was evaluated by six consec-
utive standard Nexium injections and the area for the four studied
impurities was added together and the relative standard deviation
was calculated. The reproducibility was assessed by analyzing the
same Nexium sample both at the Karlstad University and Cambrex
laboratories and pooling the standard deviation for the area-% of
the four impurities. For accuracy measurements Nexium placebo
pills containing no omeprazole or impurities were used to get the
same sample matrix as the real Nexium samples. The samples were
then spiked with 0.150 mg/mL omeprazole and impurities H431/41
and H168/22 for which the accuracy was tested at four levels; 0.05,
0.1, 0.25 and 0.5%.
The experimental design and the regression models for the
robustness study were obtained with the software MODDE 7
(Umetrics, Sweden). The design was a four-factor, two-level full
factorial design with three center points extended with three addi-
tional experiments in order to estimate a previously known square
term in pH, giving a total of 22 runs. Outliers and insignificant coeffi-
cients were removed from the regression models at 95% confidence
level. The four factors were buffer pH, column temperature and
acetonitrile fraction in eluent A and B. The buffer pH was varied
between 7.5 and 8.5, the column temperature was varied between
15 ◦ C and 25 ◦ C in HPLC and between 35 ◦ C and 45 ◦ C in UHPLC
while the acetonitrile fraction in eluent A was between 5% and
15% and in eluent B it was varied between 75% and 85%. At each
point in the experimental design two blank injections followed by
two standard injections were performed. As responses the appar-
ent retention factor for H431/41 and H168/22, the resolution factor
between omeprazole and H168/66 (closest peak pair) and the tail-
ing factor of omeprazole. To further compare the HPLC and UHPLC
methods, a Nexium sample was analyzed at the low (5%, 75%, pH 7.5,
15/35 ◦ C), center (10%, 80%, pH 8.0, 20/40 ◦ C) and high (15%, 85%, pH
8.5, 25/45 ◦ C) points in the HPLC and UHPLC designs, respectively.
The areas of the peaks were used as the response.
276 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281

Table 1
Method descriptions of the HPLC and UHPLC methods and tolerance levels from the robustness testing for critical parameters.

Setting HPLC UHPLC

Column BEH C18, 3.5 ␮m, 4.6 × 100 mm BEH C18, 1.7 ␮m, 2.1 × 50 mm
Eluent A 10/90 ± 5, v/v, acetonitrile/phosphate buffer (pH 8.0 ± 0.35, 10 mM) 10/90 ± 5, v/v, acetonitrile/phosphate buffer (pH 8.00 ± 0.5, 10 mM)
Eluent B 80/20 ± 5, v/v, acetonitrile/phosphate buffer (pH 8.0 ± 0.35, 3 mM) 80/20 ± 5, v/v, acetonitrile/phosphate buffer (pH 8.00 ± 0.5, 3 mM)
Detection UV 302 nm UV 302 nm
Gradient Linear from 0 to 100% B in 30 min Linear from 0 to 100% B in 7.15 min
Gradient starting time 1.04 min 0.25 min
Flow rate 1.00 mL/min 0.43 mL/min
Column temp. 20 ± 5 ◦ C 40 ± 5 ◦ C
Injection volume 20 ␮L 2 ␮L

Table 2
Validation of the method analytical target profile for the HPLC method. The Quantitation Limit (QL) is given in percent of omeprazole concentration in a standard injection,
accuracy is given as average recovery from two samples at four levels and precision is given as relative standard deviation in area for six injections.

Attribute Criteria Omeprazole H431/41 H168/66 H193/61 H168/22

Specificity Resolution factor between closest eluting peaks = 9.5

Linearity ≥0.9990 1.0000 1.0000 1.0000 1.0000 1.0000
Rel. response ≥0.7 1.00 1.29 0.92 0.83 1.02
QL ≤0.05% 0.0085% 0.0064% 0.0077% 0.0083% 0.012%
Accuracy 90–110% 99.6% 103% 104%
Reproducibility ≤5.0% RSD for impurity area in two laboratories = 2.56%

Table 3
Omeprazole Accuracy measurements for validation of the original HPLC method.
Response

H168/22
H431/41 H168/66
H193/61
0 5 10 15
Time [min]
Fig. 2. Illustrative chromatogram obtained at 302 nm at the working point of the
HPLC method showing omeprazole and the four investigated impurities. The flow
rate was 1.00 mL/min, column temperature was 20.0 ◦ C, and sample concentrations
were between 0.005-0.025 mg/mL.
3. Results and discussion
3.1. Development of the original QC-method
3.1.1. Validation of mATP and SST
After the initial screening a working point was determined
which is summarized in Table 1 (the tolerance levels given in Table 1
are derived from the robustness study and will be discussed in Sec-
tion 3.1.2). A C18 -stationary phase with a linear gradient and UV
detection was used to separate and quantify omeprazole and four of
its potential impurities. The detection wavelength was selected so
that the relative response should meet the acceptance criterion in
the mATP, Table 2. The flow rate, column temperature and injection
volume were chosen to correspond to values typically employed in
HPLC. An illustrative chromatogram obtained at the working point
is shown in Fig. 2.
After having found a suitable working point, it was verified that
it met the mATP. The result from the mATP validation is presented
in Table 2. The specificity demand was interpreted as no inter-
ference between the investigated compounds, i.e. all five peaks
should be separated and it was quantified by measuring the resolu-
tion factor between the peaks. The critical resolution factor, which
is reported in Table 2, is between the closest eluting peak pair;
omeprazole and H168/66. The linearity was determined with the
squared correlation coefficient, R2 , which was larger than 0.9990 for
all components and the relative response is within the set limit. The
quantitation limit was found to be 0.0064–0.012% of the omepra-
zole concentration in a standard injection which is well below the
Preparation Level % of std. inj. Recovery [%] H431/41 Recovery [%] H168/22
1 0.05 102 103
1 0.1 104 101
1 0.25 102 105
1 0.5 102 104
2 0.05 103 105
2 0.1 104 106
2 0.25 104 107
2 0.5 103 103
set limit. The accuracy was determined for omeprazole and the first
and last eluting impurities H431/41 and H168/22, respectively and,
given in Table 2, is the average values for all levels for two different
samples. The recovery of the individual injections is presented in
Table 3 and all injections yielded recoveries in the interval specified
in the mATP. The reproducibility was 2.6% given as the pooled RSD
of the area-% of the four impurities in a Nexium sample analyzed in
two laboratories and the criterion that it should be below 5% were
met.
To ensure that the analytical method performs over time, a con-
trol strategy is needed. In this case the control strategy is a SST
which is written for this specific principle, i.e. LC. The SST contains
the USP tailing factor for omeprazole to monitor the peak shape,
the resolution factor between omeprazole and H168/66 since this
was the critical resolution factor, the system repeatability in total
impurity area of 6 consecutive injections of a Nexium sample and
the apparent retention factor for the first and last eluting peak to
ensure that they are eluted during the method run time. A HPLC-
chromatogram of the Nexium sample is shown in Fig. 3(a). The SST,
acceptance criteria and the results are presented in Table 4. As can
be seen, all criteria are within their acceptable limits.
3.1.2. Risk assessment and robustness study
In the risk assessment, critical factors that affect the perfor-
mance of the method should be identified. In this case the eluent
composition, buffer pH and column temperature were identified as
the most important factors governing the separation and therefore
selected as factors in the robustness study. The ranges of the investi-
gated factors are larger than normally applied in a robustness study
to ensure models with predictive power from the DoE and hence
increase the understanding of the chromatographic system.
 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281 277

a)

Response

b)
Response

2 4 6 8 10 12 14
Column volumes

Fig. 3. Chromatograms for (a) the original HPLC methods with flow rate 1.00 mL/min and temperature 20 ◦ C and (b) the modified UHPLC method with flow rate 0.43 mL/min
and temperature 40 ◦ C for a Nexium standard sample.

Fig. 4. Response surfaces for the HPLC robustness study. (a) The resolution factor (Rs ) between omeprazole and H168/66 with the robust region (the lower 95%-confidence
limit of the predicted resolution factor > 2) above the white dotted line. (b) The tailing factor (Tf ) of omeprazole. Note that the tailing factor is <1 at some conditions which
means the peak is fronting. (c) The apparent retention factor for H431/41 (k1 ) which is the first eluting peak. The Acetonitrile fraction in eluent B is equal to 75% and the pH
is 8.5 which give the lowest values of k1 . (d) The apparent retention factor for H168/22 (k5 ) which is the last eluting peak.

The resolution factor between omeprazole and H168/66 was
the critical one and a resolution factor larger than 2 was required
for a robust method. The only factors that were significant for the
resolution factor were the pH and the temperature, although the
model contained both the quadratic pH-term and the interaction
term between pH and temperature resulting in a nonlinear model.
The response surface is presented in Fig. 4(a). A high pH and a high
temperature yield a high resolution factor, but it is evident that
the SST criterion on the resolution factor is not met at the lowest
temperature and pH. Since the regression model is good with high
predictable power, Table 5, the robust region is determined as the
region in which the whole 95% confidence interval of the predicted
resolution factor is larger than 2 [19]. This boundary is shown as a

Table 4
System suitability test and the result from the HPLC and UHPLC methods, respectively. The results are derived from the three center points in the robustness testing and
given with a 95% confidence interval.

System suitability test Acceptance criteria HPLC UHPLC

Tailing factor of the omeprazole peak from a standard injection 0.75–1.50 0.93 ± 0.06 1.03 ± 0.01
Resolution factor between omeprazole and H168/66 >2.0 9.5 ± 1.4 11.2 ± 0.1
Retention factor of H431/41 >0.5 3.42 ± 0.02 2.36 ± 0.01
Retention factor of H168/22 <30 13.42 ± 0.04 12.51 ± 0.05
System repeatability, RSD for total area of impurities in a Nexium sample (n = 6) <3.0% 2.61% 1.42%
278 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281

Table 5
Statistics for the multivariate regression models for the original and modified method. R2 is the fraction of the variation of the response explained by the model and Q2 is the
fraction of the variation of the response predicted by the model.

Statistic Retention factor Rs Tf

H431/41 H168/66 Omeprazole H193/61 H168/22

HPLC − R2 0.9993 0.9977 0.9961 0.9748 0.9646 0.9904 0.9315

HPLC − Q2 0.9982 0.6819a 0.9864 0.9646 0.9498 0.9848 0.9075
UHPLC − R2 0.9981 0.9940 0.9963 0.9967 0.9975 0.9834 0.9938
UHPLC − Q2 0.9924 0.9786 0.9891 0.9891 0.9889 0.9678 0.9880
a
Due to co-elution with omeprazole some points in the DoE are missing, thereby the low Q2 .

Table 6
Validation of the method analytical target profile for the UHPLC method. The Quantitation Limit (QL) is given in percent of omeprazole concentration in a standard injection,
accuracy is given as average recovery from two samples at four levels and precision is given as relative standard deviation in area for six injections.

Attribute Criteria Omeprazole H431/41 H168/66 H193/61 H168/22

Specificity Resolution factor between closest eluting peaks = 11

Linearity ≥0.9990 0.9997 1.0000 0.9995 0.9995 0.9998
Rel. response ≥0.7 1.00 1.40 1.00 0.83 1.11
QL ≤0.05% 0.016% 0.046% 0.038% 0.048% 0.039%
Accuracy 90–110% 104% 99% 98%
Reproducibility ≤5.0% RSD for impurity area in two laboratories = 2.55%

dotted white line in Fig. 4(a) and a robust symmetric region around Table 7
Accuracy measurements for validation of the modified UHPLC method.
the center point became pH = 7.65–8.35 and T = 15.0–25.0 ◦ C.
The tailing factor of the omeprazole peak also has two significant Preparation Level % of std. inj. Recovery H431/41 Recovery H168/22
factors, namely the temperature and pH and the response surface 1 0.05 107 93
is shown in Fig. 4(b). The tailing factor is less than one at high pH 1 0.1 100 102
and temperature which means that the peak is fronting instead of 1 0.25 98 103
tailing at these conditions. The shift from tailing to fronting with 1 0.5 100 103
2 0.05 102 91
higher pH and temperature is probably due to a change in the proto-
2 0.1 97 94
nation of omeprazole (pKa = 9.3 [20]) which affects the adsorption 2 0.25 96 99
of omeprazole to the stationary phase. To have a satisfactorily sym- 2 0.5 96 102
metric peak shape the tailing factor should be in the range 0.75–1.5
which it is in the complete region with a predicted extreme value
equal to 0.79 ± 0.02 (95% confidence interval). the resolution between omeprazole and H168/66 increased with
The response surface for the first eluting component, H431/41, temperature, which could be seen in Fig. 4(a).
is presented in Fig. 4(c); here it is can be revealed that the acetoni- Utilizing the equations for transferring chromatographic meth-
trile fraction in eluent A is the most important factor and that the ods between different columns presented in [21], the gradient slope
temperature second most important factor. The acetonitrile frac- and isocratic hold, flow rate and injection volume from the original
tion in eluent B and the pH were also significant and set to their HPLC method is scaled for the UHPLC column. Eluent composition,
low and high values, respectively, in order to give the smallest pos- buffer pH and detection wavelength is unaltered and the working
sible values of the apparent retention factor. The SST criterion in point for the modified method is presented in Table 1.
this case was that the apparent retention factor should be larger As the first part of the method performance verification criteria,
than 0.5, i.e. the peak should elute after the column hold-up time. the mATP is re-validated for the modified method. The results from
The criterion is met with the minimum value equal to 1.40 ± 0.07 the mATP re-validation are presented in Table 6 with the recov-
(95% confidence interval). The response surface for the last eluting ery of the individual injections presented in Table 7. Note that the
component, H168/22, is shown in Fig. 4(d) and the SST criterion specificity, i.e. the resolution factor, between omeprazole and the
in this case is that the apparent retention factor should be smaller closest eluting impurity H168/66 is higher in the UHPLC method
than 30, i.e. the peak should elute during the run time. The criterion compared to the HPLC method. From Table 6 it is evident that all
is met and the maximum value is predicted to be 15.3 ± 0.2. criteria are met in the mATP which indicates that the scaling to the
The HPLC method is presented with the tolerance levels for the new column dimensions was successful.
studied factors in Table 1. The method meets the mATP, SST and is Because both methods were based on the same principle, i.e.
robust in the given region. LC, the same SST and failure modes were used. The result from the
UHPLC method is shown in Table 4 alongside the HPLC method
for easy comparison. The reproducibility was 2.6% given as the
3.2. Development of the modified method pooled RSD of the impurity area from two laboratories. An UHPLC-
chromatogram of the Nexium sample is shown in Fig. 3(b). All
3.2.1. Validation of mATP and SST criteria are fulfilled with satisfactorily safety margins which are
The changes to the original HPLC method include a new column somewhat better than the original HPLC method.
with smaller particle size and other dimensions, new instrumen-
tation with smaller system volumes and an increased column 3.2.2. Robustness study
temperature. The narrower and shorter column will allow the anal- As the second part of the method performance verification cri-
ysis time and solvent consumption to be significantly reduced with teria, the robustness study is repeated for the modified method. In
maintained performance. There are two advantages with increasing Fig. 5 the response surfaces for the resolution, tailing and appar-
the column temperature; first the viscosity of the eluent is reduced ent retention factors are shown for the two factors which had the
which result in a lower back pressure and, for this particular case, largest influence of each of the responses. The other significant fac-
 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281 279

Fig. 5. Response surfaces for UHPLC robustness study. (a) The resolution factor (Rs ) between omeprazole and H168/66 with acetonitrile fraction in eluent A equal to 5% and
in eluent B equal to 85% which give the lowest values of Rs . (b) The tailing factor (Tf ) of omeprazole with acetonitrile fraction in eluent B equal to 5% and temperature equal to
45 ◦ C which give the lowest values of Tf . (c) The apparent retention factor for H431/41 (k1 ) with acetonitrile fraction in eluent B equal to 75% and temperature equal to 45 ◦ C
which give the lowest values of k1 . (d) The apparent retention factor for H168/22 (k5 ) with temperature equal to 35 ◦ C and pH equal to 7.5 which give the highest values of k5 .

tors were set so to achieve the extreme values of the responses. The
Normalized Response

statistics of the regression models are presented in Table 5. a) HPLC

The resolution factor between omeprazole and H168/66, UHPLC
Fig. 5(a), was at its minimum when temperature, pH and acetoni-
trile in eluent A were at their lowest values and the acetonitrile
fraction in eluent B was at its highest. It was then equal to 5.1 ± 0.5
which is well above the criterion in the SST.
The tailing factor for omeprazole, Fig. 5(b), was at its minimum
when temperature, pH and acetonitrile in eluent A were at their
highest values and the acetonitrile fraction in eluent B was at its
lowest value. It was then equal to 0.86 ± 0.01 which is well above
the criterion in the SST. It is interesting to note that the tailing factor
Normalized Response

is closest to one, i.e. a symmetric peak, close to the center point of b)

the design region which is favorable.
The apparent retention factor for H431/41, Fig. 5(c), was
0.9 ± 0.1 at its minimum which corresponds to a high acetonitrile
fraction in eluent A and a pH around 7.6, i.e. the extreme value was
not in the corner of the experimental design. The maximum value of
the apparent retention factor for H168/22, Fig. 5(d), was 15.6 ± 0.2.
This happened at high acetonitrile fractions in the eluents and low
pH and temperature.
To summarize the robustness study; all responses were well 0 2 4 6 8 10 12 14 16
within the criteria specified in the mATP and the SST and the whole Apparent retention factor
experimental design was therefore robust which gave the ranges
for the factors shown in Table 1. Fig. 6. Comparing HPLC and UHPLC chromatograms with the time normalized to
retention factor for a) the same column temperature, i.e. 30 ◦ C and b) the tempera-
tures specified in the QC-methods, i.e. 20.0 ◦ C for HPLC and 40.0 ◦ C for UHPLC.
3.3. A comparison between the original and modified QC-method

The observed differences in apparent retention time between
the two methods could be explained using chromatographic theory.
To make the comparison as clear as possible, the column temper-
ature was changed to 30 ◦ C for both methods, all extra column
volumes were accounted for and the concentrations of omeprazole
and the impurities were changed to 0.005–0.025 mg/mL to avoid
column overloading. The overlaid chromatograms are presented in
Fig. 6(a) where the time axis has been normalized to retention fac-
tor (i.e. (t − t0 )/t0 ) and the response is normalized by dividing with
the highest peak. By normalizing the time in the chromatograms
with the column hold-up time according to the definition of the
retention factor, differences in column volume and total porosity
are accounted for. The HPLC peaks are eluted somewhat faster than
the UHPLC peaks, but the difference is shifting between different
peaks. This observation can be explained by the higher pressure
in UHPLC compared to HPLC, because the stationary phases in the
two columns per se have been shown to have very similar adsorp-
tion behavior [22,23]. The column inlet pressure at the beginning
of the gradient run is 100 and 400 bar for HPLC and UHPLC, respec-
tively. Omeprazole has been shown to have a significant pressure
dependence in this experimental system [23] which is not unusual
for certain types of compounds [24,25]. Since the impurities are
280 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281
1
a)
0.8
Response [AU] 0.6
0.4
0.2
0
10.5 11
Time [min]
Fig. 7. Comparing of the resulting omeprazole peaks in (a) HPLC and in (b) UHPC, injection of a Nexium sample. The concentration was 0.154 mg/mL and the injection volume
was 20 ␮L in (a) HPLC and 2 ␮L in (b) UHPLC.
similar in structure to omeprazole, it is likely that they also have
a significant pressure dependence. The axial temperature gradient
through the UHPLC column caused by viscous heating [26,27] was
estimated to 3–4 ◦ C at 400 bar inlet pressure [23]. In the robust-
ness study, it was observed that the retention for omeprazole and
the impurities decreased with increasing temperature. Therefore,
the increase in retention seen in UHPLC, Fig. 6(a), cannot be due to
viscous friction. Instead, the difference between HPLC and UHPLC
method would probably be larger if there was no viscous heating.
The combination of pressure and temperature differences between
HPLC and UHPLC explains the difference in retention seen for the
compounds, because different compounds respond differently to
pressure and temperature [24].
Fig. 6(b) is the same as Fig. 6(a) except that here the tempera-
tures are the ones specified in the methods, i.e. 20 ◦ C and 40 ◦ C for
HPLC and UHPLC, respectively. The order of the HPLC and UHPLC
peaks has been reversed for most of the compounds, which can be
expected from the larger difference in temperature. The effect of
pressure seen in Fig. 6(a) is not visible because here the temper-
ature effect dominates. That the pressure and temperature effects
are convoluted is important to remember if one should be able to
understand the shifts in retention times observed when switching
from HPLC to UHPLC.
If the method performance from the mATP and SST is compared
for the HPLC and UHPLC method, Tables 2–7, it is evident that both
methods fulfill all the set criteria in the working point. The quantita-
tion limit is lower in HPLC than it is in UHPLC for all the components,
but it still is within the acceptable region for UHPLC. In Fig. 7 the
omeprazole peak for the Nexium sample is presented in actual
time and response units to compare the peak width and height.
The UHPLC peak is much narrower and has a higher response. The
slope of the calibration curve of omeprazole was 5.0 mVs/ng and
6.0 mVs/ng for HPLC and UHPLC, respectively. Based on these obser-
vation, the lower quantitation limit in HPLC is probably due to a
higher linearity in the standard curves (Tables 2 and 6) which gives
a smaller standard deviation in the calculation of the LoQ. The res-
olution factor between omeprazole and H168/66 is larger in UHPLC
due to the increased temperature. When the robustness studies are
compared, Figs. 4–5, the acceptable region for the resolution factor
is larger in UHPLC, which means that the UHPLC method is more
robust towards changes in factors. This improvement in resolution
is most probably because UHPLC is operating in higher tempera-
tures and in this particular case, it is favorable. Considering all the
responses, there were different factors that were most important
in HPLC and UHPLC. This is likely due to confounding effects in the
b)
11.5 2 2.5 3
Time [min]
Table 8
Total amount of organic impurities in a Nexium sample stored in commercial pack-
age at 40 ◦ C/75%RH for 6 months. Results from three levels, low, center and high, in
the HPLC and UHPLC DoE.
Level in the DoE HPLC (area%) UHPLC (area%)
Low (5%, 75%, pH7.5, 15 ◦ C) 0.6 0.6
Centre (10%, 80%, pH8.0, 20 ◦ C) 0.6 0.6
High (15%, 85%, pH8.5, 25 ◦ C) 0.6 0.6
factors because of the choice of experimental design, i.e. there is
probably no physical reason for the differences seen. Overall, the
analysis time was reduced by a factor 4 and the solvent consump-
tion was reduced by a factor 10 when switching from the original
to the modified method.
Finally, the similarities between the HPLC and UHPLC methods
are shown by comparing the total area of organic impurities in a
Nexium sample stored at accelerated conditions, Table 8. This sam-
ple was injected at low, center and high settings, Table 8, in the
two DoE used in the robustness study. The total amount of organic
impurities in the original HPLC method at all the three tested set-
tings was found to be 0.6%. For the modified UHPLC method the
total amount of organic impurities was also found to be 0.6% at the
three tested settings. Together with the results from the mATP and
the SST this shows that the two methods are equivalent from all
relevant perspectives.
4. Conclusions
QC-methods are an important part in the control strategy for
the manufacturing of drug substances. However, efficient life-cycle
management and continual improvement are hindered due to a
variety of post-approval variation legislations across territories and
a lack of harmonization of the requirements. As a result, many QC-
methods have fallen behind the technical development. We have
presented a concept for QC-method development which addresses
the challenge of enabling minor adjustments, within the same mea-
suring principle, outside the original MODR which was impossible
to incorporate when the method was first developed. Such minor
adjustments would be done without any interaction with regulat-
ing agencies.
Building on the QbD framework for analytical methods which
enables changes inside the MODR, the new concept relies on the
assumption that the SST and failure modes are relevant for other
conditions outside the MODR when the same measuring principle is
used. It follows that it should be possible to move outside the MODR
 D. Åsberg et al. / Journal of Pharmaceutical and Biomedical Analysis 129 (2016) 273–281
as long as the SST is passed. The new method control strategy for
minor adjustments outside the MODR consists of: (i) a SST for each
operation, (ii) a statistical equivalence between the original and
modified methods and (iii) a set of method performance verification
criteria. Revalidation is handled within the pharmaceutical quality
control system according to internal change control procedures,
but without interaction with regulating agencies.
This concept has been exemplified by a case study where a QC-
method for the drug Nexium has been developed originally for
HPLC and then, as a continual improvement, a switch to UHPLC
was done. The SST and revalidation were passed for the modified
method and it was shown to be statistically equivalent to the orig-
inal method. The analysis time was reduced by a factor 4 and the
solvent consumption was reduced by a factor 10.
If the presented method enhancement concept will be reality
it means that the pharmaceutical industry can save money and
time and still maintain high quality QC-methods. Less burden and
decreased effort put on regulatory agencies will also be a reality. The
concept will help to maintain high quality of QC- methods used for
release testing of important medicines and in the end secure that
high quality medicines are delivered to the patients.
Acknowledgements
We are grateful to Waters that provided an ACQUITY UPLC H-
Class system with support and technical advice during the project
and who also donated the columns. This work was supported by
the Swedish Knowledge Foundation for the KK HÖG 2014 project
“SOMI: Studies of Molecular Interactions for Quality Assurance, Bio-
Specific Measurement & Reliable Supercritical Purification” (grant
number 20140179), by the Swedish Research Council (VR) in the
project “Fundamental Studies on Molecular Interactions aimed
at Preparative Separations and Biospecific Measurements” (grant
number 2015-04627) and by The ÅForsk Foundation for the project
“Improved Purification Procedures to Satisfy Modern Drug Quality
Assurance and Environmental Criteria” (grant number 15/497).
References
[1] S. Görög, The changing face of pharmaceutical analysis, TrAC 26 (2007) 12–17.
[2] ICH Harmonised Tripartite Guideline, Technical and Regulatory
Considerations for Pharmaceutical Product Lifecycle Management Q12,
(2014).
[3] ICH Harmonised Tripartite Guideline, Pharmaceutical Development Q8(R2),
(2009).
[4] ICH Harmonised Tripartite Guideline Quality Risk Management Q9, (2005).
[5] ICH Harmonised Tripartite Guideline Pharmaceutical Quality System Q10,
(2008).
[6] ICH Harmonised Tripartite Guideline, Development and Manufacture of Drug
Substances Q11, (2012).
[7] M. Schweitzer, M. Pohl, M. Hanna-Brown, P. Nethercote, P. Borman, G.
Hansen, et al., Implications and opportunities of applying QbD principles to
analytical measurements, Pharm. Tech. 34 (2010) 52–59.
281
[8] EMA/430501/2013 Document, EMA-FDA pilot program for parallel
assessment of Quality-by-Design applications: lessons learnt and Q&A
resulting from the first parallel assessment, (2013).
[9] J. Musters, L. van den Bos, E. Kellenbach, Applying QbD principles to develop a
generic UHPLC method which facilitates continual improvement and
innovation throughout the product lifecycle for a commercial API, Org.
Process Res. Dev. 17 (2013) 87–96.
[10] K. Monks, I. Molnár, H.-J. Rieger, B. Bogáti, E. Szabó, Quality by Design:
multidimensional exploration of the design space in high performance liquid
chromatography method development for better robustness before
validation, J. Chromatogr. A 1232 (2012) 218–230.
[11] P.F. Gavin, B.A. Olsen, A quality by design approach to impurity method
development for atomoxetine hydrochloride (LY139603), J. Pharm. Biomed.
Anal. 46 (2008) 431–441.
[12] A.H. Schmidt, I. Molnár, Using an innovative Quality-by-Design approach for
development of a stability indicating UHPLC method for ebastine in the API
and pharmaceutical formulations, J. Pharm. Biomed. Anal. 78–79 (2013)
65–74.
[13] A.H. Schmidt, M. Stanic, Rapid UHPLC method development for omeprazole
analysis in a quality-by-design framework and transfer to HPLC using
chromatographic modeling, LCGC N. Am. 32 (2014) 126–148.
[14] C. Boussès, L. Ferey, E. Vedrines, K. Gaudin, Using an innovative combination
of quality-by-design and green analytical chemistry approaches for the
development of a stability indicating UHPLC method in pharmaceutical
products, J. Pharm. Biomed. Anal. 115 (2015) 114–122.
[15] S. Fekete, I. Kohler, S. Rudaz, D. Guillarme, Importance of instrumentation for
fast liquid chromatography in pharmaceutical analysis, J. Pharm. Biomed.
Anal. 87 (2014) 105–119.
[16] L. Olbe, E. Carlsson, P. Lindberg, A proton-pump inhibitor expedition: the case
histories of omeprazole and esomeprazole, Nat. Rev. Drug Discov. 2 (2003)
132–139.
[17] J.W. Dolan, Dwell volume revisited, LC GC N. Am. 24 (2006) 458–466.
[18] ICH, Validation of Analytical Procedures: Text and Methodology Q2(R1),
(2005).
[19] E. Rozet, P. Lebrun, P. Hubert, B. Debrus, B. Boulanger, Design Spaces for
analytical methods, TrAC 42 (2013) 157–167.
[20] Chemicalize.org, Kft. ChemAxon, Budapest Hungary, (2015).
[21] D. Guillarme, D.T.T. Nguyen, S. Rudaz, J.-L. Veuthey, Method transfer for fast
liquid chromatography in pharmaceutical analysis: application to short
columns packed with small particle. Part II: gradient experiments, Eur. J.
Pharm. Biopharm. 68 (2008) 430–440.
[22] D. Åsberg, M. Leśko, J. Samuelsson, K. Kaczmarski, T. Fornstedt, Method
transfer from high-pressure liquid chromatography to ultra-high-pressure
liquid chromatography. I. A thermodynamic perspective, J. Chromatogr. A
1362 (2014) 206–217.
[23] D. Åsberg, J. Samuelsson, M. Leśko, A. Cavazzini, K. Kaczmarski, T. Fornstedt,
Method transfer from high-pressure liquid chromatography to
ultra-high-pressure liquid chromatography. II. Temperature and pressure
effects, J. Chromatogr. A 1401 (2015) 52–59.
[24] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Further investigations of
the effect of pressure on retention in ultra-high-pressure liquid
chromatography, J. Chromatogr. A 1217 (2010) 276–284.
[25] S. Fekete, J.-L. Veuthey, D.V. McCalley, D. Guillarme, The effect of pressure and
mobile phase velocity on the retention properties of small analytes and large
biomolecules in ultra-high pressure liquid chromatography, J. Chromatogr. A
1270 (2012) 127–138.
[26] F. Gritti, M. Martin, G. Guiochon, Influence of viscous friction heating on the
efficiency of columns operated under very high pressures, Anal. Chem. 81
(2009) 3365–3384.
[27] K. Kaczmarski, F. Gritti, J. Kostka, G. Guiochon, Modeling of thermal processes
in high pressure liquid chromatography: II. Thermal heterogeneity at very
high pressures, J. Chromatogr. A 1216 (2009) 6575–6586.
 Journal of Chromatography A, 1479 (2017) 107–120

Contents lists available at ScienceDirect

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

A practical approach for predicting retention time shifts due to

pressure and temperature gradients in ultra-high-pressure liquid
chromatography
Dennis Åsberg a , Marcin Chutkowski b , Marek Leśko b , Jörgen Samuelsson a ,
Krzysztof Kaczmarski b,∗ , Torgny Fornstedt a,∗
a
Department of Engineering and Chemical Science, Karlstad University, SE-651 88 Karlstad, Sweden
b
Department of Chemical and Process Engineering, Rzeszów University of Technology, PL-359 59 Rzeszów, Poland

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

Article history: Large pressure gradients are generated in ultra-high-pressure liquid chromatography (UHPLC) using
Received 14 September 2016 sub–2 ␮m particles causing significant temperature gradients over the column due to viscous heating.
Received in revised form These pressure and temperature gradients affect retention and ultimately result in important selectivity
22 November 2016
shifts. In this study, we developed an approach for predicting the retention time shifts due to these
Accepted 24 November 2016
gradients. The approach is presented as a step-by-step procedure and it is based on empirical linear
Available online 10 December 2016
relationships describing how retention varies as a function of temperature and pressure and how the
average column temperature increases with the flow rate. It requires only four experiments on standard
Keywords:
UHPLC
equipment, is based on straightforward calculations, and is therefore easy to use in method development.
Temperature The approach was rigorously validated against experimental data obtained with a quality control method
Pressure effects for the active pharmaceutical ingredient omeprazole. The accuracy of retention time predictions was
Gradient effects very good with relative errors always less than 1% and in many cases around 0.5% (n = 32). Selectivity
Method development shifts observed between omeprazole and the related impurities when changing the flow rate could also
Retention time be accurately predicted resulting in good estimates of the resolution between critical peak pairs. The
approximations which the presented approach are based on were all justified. The retention factor as
a function of pressure and temperature was studied in an experimental design while the temperature
distribution in the column was obtained by solving the fundamental heat and mass balance equations for
the different experimental conditions. We strongly believe that this approach is sufficiently accurate and
experimentally feasible for this separation to be a valuable tool when developing a UHPLC method. After
further validation with other separation systems, it could become a useful approach in UHPLC method
development, especially in the pharmaceutical industry where demands are high for robustness and
regulatory oversight.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction
Ultra-high-pressure liquid chromatography (UHPLC) offers a
3–10 fold increase in throughput and a significantly lower sol-
vent consumption than conventional UHPLC [1,2] and has recently
gained ground in pharmaceutical quality control [3,4]. Combin-
∗ Corresponding authors.
E-mail addresses: kkaczmarski@prz.edu.pl (K. Kaczmarski),
torgny.fornstedt@kau.se (T. Fornstedt).
ing UHPLC with columns packed with coreshell particles, having
reduced intra-particle diffusion path and low eddy dispersion as
compared to fully porous particles resulting in higher efficien-
cies at high flow rates [5–7] gives an additional advantage over
conventional HPLC. High throughput and high performance are par-
ticularly important for quality control methods of release batches
in the pharmaceutical industry [8]. However, due to strict regula-
tory oversight and high demands for robustness, it can be difficult
for the pharmaceutical industry to develop quality control methods
utilizing the advantages of UHPLC [1,3,4].

http://dx.doi.org/10.1016/j.chroma.2016.11.050
0021-9673/© 2016 Elsevier B.V. All rights reserved.
108 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120

We have, in a series of paper [3,9,10], investigated fundamen- Table 1

Physico-chemical properties of the columns given by the manufacturer and mea-
tal aspects of UHPLC including regulatory issues such as how the
sured in our lab.
pharmaceutical industry can switch from HPLC to UHPLC for qual-
ity control methods. Understanding the role of viscous heating and BEH Cortecs
elevated pressure was found to be a key for predictable and effi- Base particle
cient method development. Due to the very high pressures used in Mean particle size [␮m] 1.87 1.79
UHPLC (many times exceeding 1000 bar), viscous friction generates ω = rc /rp 0.000 0.687
Average pore diameter [Å] 129 87
heat in the column which increases its temperature. The generated
Surface area [m2 /g] 174 101
heat leads to axial and radial temperature gradients in the col- Pore volume [cm3 /g] 0.64 0.25
umn which can affect both retention and efficiency [11–15]. One
Bonded phase
advantage of columns packed with coreshell particles is that the Total carbon [%] 17.09 6.45
higher thermal conductivity of the solid core improves heat dissi- Surface coverage [␮mol/m2 ] 3.18 2.71
pation and in that way reduces the radial temperature gradients in End-capping yes yes
the column [16–18]. In addition to the temperature gradients, the Packed column
pressure per se impacts retention by altering the adsorption of the Dimension [mm × mm] 2.1 × 100 2.1 × 100
solute on the stationary phase [19–22]. Since individual solutes are External porositya 0.358 0.396
affected differently by pressure and temperature, selectivity shifts Total porosityb 0.589 0.529
Particle porosityc 0.360 0.220
can occur in UHPLC when the flow rate, and hence the pressure, is Shell porosity N.A. 0.326
changed [23]. Such selectivity shifts are not seen in HPLC and need Thermal conductivityd [W/m K] 0.985 1.150
to be understood and predicted for reliable UHPLC method devel- a
Measured by inverse size exclusion chromatography in our lab.
opment. However, to our knowledge, a feasible and easy-to-use b
Measured by pycnometry in our lab.
methodology for method development to predict retention time c
The particle porosity includes the volume of the solid silica core.
d
shifts due to pressure and temperature gradients does not exist. Calculated according to Eq. (40) in Ref. [13].
The approach available, developed by Kaczmarski et al., requires
advanced calculations and is therefore more suitable for research ature and mobile phase velocity profiles that was thereafter used
applications [13,14]. to solve the mass transfer in part (iii).
The pressure and temperature effects in UHPLC have been de-
coupled using restrictor capillaries [15,24] or a water bath [25] to 2.1. Heat transfer and velocity distribution
estimate each contribution to the retention when the flow rate is
changed. This methodology could be used as a basis for an empirical The heat transfer model for the column bed, Eq. (1), and that for
approach to predict the retention factor as a function of flow rate the column wall, Eq. (2), at steady-state are described as:
in UHPLC.    
∂T ∂T 1 ∂ ∂T ∂ ∂T ∂P
The aim of this study was to develop an easy-to-use, empirical Cp,m ux + Cp,m ur = reff + eff − ux (1 − ˛T ) (1)
∂x ∂r r ∂r ∂r ∂x ∂x ∂x
approach for predicating retention time shifts due to temperature  
2 2
and pressure gradients, suitable for routine method development. 1 ∂T ∂ T ∂ T
0 = w + + w (2)
The approach should require few experiments, standard chromato- r ∂r ∂r 2 ∂x2
graphic equipment and straightforward calculations. As in our
previous studies, we will base this investigation on data from a where Cp,m is the mobile phase heat capacity, ux and ur are super-
quality control method for the drug omeprazole [3,9,10] as it is a ficial mobile phase velocities in x and r directions respectively, T is
relevant case and two columns will be evaluated; one with fully the temperature, P is the pressure, ␣ is the mobile phase thermal
porous particles and one with coreshell particles. expansion coefficient, w is the wall thermal conductivity coeffi-
To justify the approximations needed for the empirical cient and eff is the effective bed thermal conductivity. eff was
approach, the retention as a function of pressure and tempera- estimated with the method described in [14] where the thermal
ture will be studied in an experimental design. Following this, the conductivity of the mobile phase is dependent on the local tem-
temperature distribution over the column placed in a still air col- perature and pressure and the thermal conductivity of the solid
umn oven or in a water jacket will be calculated by the approach particles are estimated from silica and octadecane. The estimated
developed by Kaczmarski et al. [13,14] based on a mechanistic thermal conductivity for the two types of stationary phase parti-
description of the mass balance and heat transfer process (we refer cles used in this work is presented in Table 1. Eqs. (1) and (2) were
to this approach as the mechanistic approach). From these find- solved with the same boundary conditions as in [13,14].
ings, we will justify the approximations and experimental methods The local velocity along the column was calculated from
which to base our empirical approach on. Finally, the empirical ∂P ε3e dp2 1
approach will be verified against experimental data and compared ux (r, x) = − (3)
∂x (1 − εe )2  (r, x)
to the results from the mechanistic model.
where εe the external porosity of the bed, dp the particle diameter, 
is the Blake-Kozeny-Carman parameter and  is the local viscosity.
2. Theory We calculated temperature and velocity distributions in the column
according to Eqs. (1–3). The heat convection coefficient from the
The mechanistic model used in this work for calculating elution external column surface to the surrounding, denoted hext , and 
profiles in UHPLC is very similar to that described in [13,14], there- were estimated based on experiments by minimizing the difference
fore the reader is referred to those papers for additional details. The between the calculated and measured column inlet pressure and
mechanistic model consists of three parts: (i) a model of the steady between the calculated and measured surface temperature at ca
state heat transfer which describes how the heat is generated by 8 cm from the inlet, for the highest flow rate experiments for each
viscous friction and how it is evacuated from the column under system (Supplementary Data Table S5). The mobile phase density
steady-state conditions (ii) a model for the mobile phase velocity and viscosity were taken from reliable data published in [26] for the
distribution and (iii) a model of solute mass transfer. First part (i) acetonitrile-water mixture. The mobile phase thermal expansion
and (ii) was solved simultaneously to obtain a steady state temper- coefficient, ˛, and heat capacity, Cp,m , as functions of pressure and
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 109

temperature were calculated according to the correlations given in where

1 is a geometrical constant describing the packing of the
[14]. bed and it was assumed that
1 = 0.7 [31]. The external mass trans-
fer coefficient (kext ) was calculated from the Wilson-Geankoplis
2.2. Mass balance correlation [32]:
 1/3
1.09 2
uDm
The model of mass transfer accounts for the propagation of kext = (10)
a solute band along a column. For very high pressure drops, the εe dp2
column is not isothermal because heat is generated by the elu-
The radial dispersion coefficient (Da,r ) was calculated on the base
ent going through the bed. The heat transfer out of the column
of the plate height equation derived by Knox [33]:
generates temperature gradients. Due to these axial and radial tem-
perature gradients, the other physico-chemical parameters, e.g. 0.03dp u
Da,r = + 0.7Dm (11)
the equilibrium constant and the migration rate, depend on the εt
actual position inside the column. To take this into account, the Finally, the tortuosity parameter ( ) was calculated from the
mass balance should be written in two space dimensions using correlation
cylindrical coordinates. In this work we have applied the modified
equilibrium-dispersive (ED) model for totally porous or coreshell = εs + 1.5 1 (1 − εs ) (12)
particles [27]. The mass balance equation is written as:
where 1 is estimated together with the adsorption isotherm
 
3 parameters from experimental elution profiles using the inverse
∂c (1 − εe ) (1 − εs ) 1 −  ∂q 1 ∂ (uc)
+ + method.
∂t εt ∂t εt ∂x
    The mass balance model must also be combined with an appro-
∂ ∂c 1 ∂ ∂c priate adsorption isotherm model. In this work, we consider a
= Da,x + rDa,r (4)
∂x ∂x r ∂r ∂r model where the Henry constant (H) is a function of temperature
and pressure. Hence, the adsorption isotherm is given by:
where c and q are the solute concentrations in the mobile and in the  
stationary phases, respectively, Da,x and Da,r are the apparent axial E − Vm P E − Vm P0,avg
q = H0 exp − c (13)
and radial dispersion coefficients, respectively, t is the time and  is RT RT0,avg
the ratio of the solid core diameter to the external diameter of the
where E the activation energy of adsorption, R the universal gas con-
coreshell particle. εt is the total porosity and εs is the shell porosity
stant and Vm the difference between the partial molar volumes
of the coreshell particle which is calculated from [27,28]:
of the solute in the adsorbed layer and in the liquid phase. The sub-
εt − εe script “0” denotes the reference experimental conditions here taken
εs =   (5)
(1 − εe ) 1 − 3 as the lowest flow rate experiments where the temperature gradi-
ents in the column are very small. H0 was estimated on the base of
For a totally porous particle we have  = 0 and the shell porosity the elution profile with the lowest flow rate (0.08 mL/min) for each
is replaced by the particle porosity in Eq. (5). All porosities are listed solute and experimental system and P0,avg and T0,avg are the aver-
in Table 1 for the columns employed in this work. The apparent axial age pressure and temperature in the column for these experiments.
dispersion coefficient (Da,x ) can be expressed as [27,29]:

DL εe u2 dp
 k 2 dp 1 + 2 + 32 − 3 − 54 1
z
Da,x = +  2 + (6)
εt 6εt (1 − εe ) 1 + kz 10εe Dm 1 +  + 2 kext

where Dm is the molecular diffusion coefficient, DL is the local dis-

persion coefficient, kext is the external mass transfer coefficient and In the next step, E and Vm along with 1 were estimated from the
is the tortuosity coefficient. The zone retention factor, kz , is equal elution profile with the highest flow rate for each experimental
to: system.
  The mass balance equation, Eq. (4), coupled with heat balance
1 − εe ∂q  
kz = εs + (1 − εs ) 1 − 3 (7) was solved by the finite element method previously described in
εe ∂c [13,14]. Namely, first the steady-state distributions of the tem-
perature and pressure profiles were derived. Afterwards the time
To solve the mass balance, Eq. (4), the following parameters
dependent mass balance equation was solved using the tempera-
have to be calculated: Dm , DL , kext , Da,r and . The molecular diffu-
ture and the pressure obtained in the first step.
sion coefficient, Dm , was estimated with the Wilke-Chang equation
locally [30]:
3. Material and methods

−8 T (x, r) f1 A1 M1 + f2 A2 M2
Dm (x, r) = 7.4 × 10 (8) 3.1. Chemicals
(x, r)Vb0.6

where f is the molar fraction of the solvent components in the
mobile phase, A is the solvent association factor (2.6 for water and
1.0 for acetonitrile [16]), M is the molecular mass of the solvent
components and Vb the solute molar volume at its boiling point. Vb
was estimated with the Le Bas method [30]. Dm was estimated for
the column inlet and outlet conditions and an average value was
used in the calculations. The local dispersion coefficient, DL , was
calculated from the relationship [31]:
DL =
1 Dm
Omeprazole (99.8%) and analytical reference standards of the
related compounds H153/73 (CAS: 176219-04-8), H168/66 (CAS:
88546-55-8) and H193/61 (CAS: 89352-76-1) were donated by
AstraZeneca R&D, Gothenburg, Sweden. H193/61 is a mixture of
5-methyl omeprazole and 6-methyl omeprazole (Supplementary
Data Fig. S1) and in this study we refer to 6-methyl omeprazole
when talking about the retention time of H193/61. Sodium nitrate
(≥99%), sodium phosphate dibasic dihydrate (99.5%), sodium phos-
phate monobasic dihydrate (99.0%) and polystyrene standards
(9) (400-2 500 000 g/mol) were purchased from SigmaAldrich (St.
110 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120
Louis, MO, USA). Acetonitrile (gradient grade), dichloromethane
(reag. ph. eur.) and sodium hydrogen carbonate (100.0%) were pur-
chased from VWR International (Radnor, PA, USA). Tetrahydrofuran
(HPLC grade), acetic acid (≥99.8%), sodium acetate (≥99.0%) and
sodium carbonate (≥99.5%) were purchased from Merck (Darm-
stadt, Germany). Deionized water with conductivity 18.2 M
cm the column below 100 bar. To adjust the pressure, keeping the flow
was delivered from a Milli-Q Plus 185 system (Millipore, Billerica,
MA, USA).
3.2. Instrumentation
An Acquity H-Class Bio system from Waters (Milford, MA, USA)
with a quaternary solvent manager, an auto sampler with a 10 ␮L
flow-through needle, a still air column oven with a mobile phase
pre-heater and a PDA detector with a 500 nL flow-cell was used
throughout the study. The extra-column volume from the auto
sampler to the detector was 19 ␮L and a delay between the start
of the recording in the Empower 3 software and the switching
of the injection valve of 0.75 s was measured, i.e. the recording
started 0.75 s before the actual injection. In experiments, men-
tioned as water jacket experiments, the column was placed in a
turbulent water jacket controlled by a MultiTemp 3 water ther-
mostat from Pharmacia Biotech (Uppsala, Sweden) instead of the
column oven, although the mobile-phase pre-heater was still in
operation. This setup increased the extra-column volume to 25 ␮L
because of longer capillaries (ca 80 cm 0.003 i.d.); all extra-column
volumes and the injection delay time were adjusted for in the data
processing. The column temperature was set to 35.0 ± 0.1 ◦ C in all
experiments, except for the experimental design. Two columns
were used in this work; one containing fully porous particles and
one containing coreshell particles. The column containing fully
porous particles was an Acquity BEH-C18 column (2.1 × 100 mm,
nominal particle size 1.7 ␮m) and the coreshell column was a
Cortecs UPLC C18 column (2.1 × 100 mm, nominal particle size
1.6 ␮m), both from Waters. Detailed information about the columns
is presented in Table 1.
3.3. Procedure
The mobile phase consisted of 20/80, w/w, acetoni-
trile/phosphate buffer. The phosphate buffer was 30 mM, pH
7.23 and filtered through 0.2 ␮m nylon filters form Whatman
(Maidstone, UK). The samples were dissolved in mobile phase with
the concentration of omeprazole equal to 0.04 g/L and 0.02 g/L for
the other compounds. Due to limited stability of omeprazole in
the mobile phase, the samples were exchanged with 12 h intervals
and the auto sampler temperature was kept at 8 ◦ C. All samples
were filtered through a 0.2 ␮m polypropylene syringe filter from
VWR International prior to use. The injection volume was 1 ␮L,
UV-detection was done at 220 nm and three replicates were done
for each injection.
Pycnometry [34] with acetonitrile and dichloromethane, was
used to determine the total porosity of the columns. The external
porosity was determined by inverse size exclusion chromatography
(ISEC) to determine the external porosity; 12 polystyrene standards
were dissolved in tetrahydrofuran and eluted with neat tetrahydro-
furan at 0.10 mL/min [16]. The experimental data obtained from
the ISEC experiments are presented in Supplementary Data Fig. S2.
The temperature of the column surface was measured by attaching
a PT-100 (4-wire) resistance temperature detector from Pentronic
AB (Gunnebo, Sweden) directly on the column surface ca 8 cm from
the inlet. The PT-100 elements have the accuracy ±0.2 ◦ C.
A three-level, two-factor, full factorial experimental design with
two center points was used to investigate how the retention factor
and selectivity changed with pressure and temperature for each
solute. Calculations were performed with MODDE 11 (Umetrics,
Sweden) and a 95% confidence level was used to remove insignifi-
cant model terms. The temperature was varied between 35 ◦ C and
59 ◦ C while the average column pressure was varied between ca
100 and 800 bar. The flow rate was low (ca 0.1 mL/min) to mini-
mize temperature gradients and keep the pressure gradient over
rate constant, LC PEEKsil tubing with inner diameter 25 ␮m from
SGE Analytical Science (Milton Keynes, U.K.) with lengths 5–50 cm,
was installed after the column.
Omeprazole, H168/66 and H153/73 acts as weak acids near
pH 7 and their pKa values in the mobile phase were estimated
with chromatography as described in [35]. Acetate buffer with pH
4.68, phosphate buffers with pH 6.22, 7.16 and 8.07 and carbonate
buffers with pH 9.52 and 10.83 were mixed in 20/80, w/w ace-
tonitrile/buffer. Buffer concentration was 30 mM in all cases. The
pH was measured in each mobile phase at 35 ◦ C with a glass elec-
trode calibrated with water solutions. Apparent pKa values were
then determined using chromatography at the following points in
the experimental design: 35 ◦ C, 100 bar, 35 ◦ C, 800 bar and 59 ◦ C,
100 bar. These apparent pKa values are approximately the same for
similar types of columns, e.g. C18, and was therefore determined
with the BEH column only [36].
As for the main objective, to predict how the retention factor
varied with flow rate due to increasing temperature and pressure
gradients, elution profiles were recorded at 0.08, 0.20, 0.30, 0.40,
0.50 and 0.54 mL/min for the four solutes. For the Cortecs column,
the highest flow rate was 0.58 mL/min instead. At each flow rate, the
surface temperature of the column and the extra-column pressure
drop were measured. The experiments were then repeated with the
columns in the turbulent water jacket instead of the column oven
which gives a much more efficient heat transfer. The maximum
flow rate for the water jacket experiments were 0.50–0.54 mL/min
due to a somewhat higher pressure drop over the columns.
The column efficiency was calculated with the moment method
[31] using an in-house algorithm implemented in MATLAB R2015b
(MathWorks Inc. Natick, MA, USA). The threshold value of the inte-
gration boundaries were set to 0.2% of the maximum peak height.
4. Results and discussion
The separation of omeprazole and four related compounds on
two C18 columns, one with fully porous particles (BEH) and one
with coreshell particles (Cortecs) was investigated at different flow
rates. The selectivity between omeprazole and H168/66 changed
significantly with flow rate (Fig. 1) which resulted in co-elution
at some flow rates and baseline separation at others. This makes
the method sensitive to variations in flow rate; for example, see
the shifts in selectivity between peaks 2 and 3 in Fig. 1. At high
flow rates, with pressure drops of 700–800 bar over the column,
the axial temperature gradient in the column is significant and the
pressure itself also affects the retention [10]. Therefore, the shifts in
retention and selectivity seen in Fig. 1 are most likely a result of the
change in pressure and temperature in the column rather than the
flow rate per se. By comparing the peak width for the two columns,
the gain in efficiency for the coreshell particles mentioned in the
introduction, is also evident.
First, the retention factor and selectivity as functions of pres-
sure and temperature are investigated in an experimental design
to confirm that pressure and temperature can indeed give rise to
selectivity shifts and to justify the use of a linear function between
retention factor and pressure and temperature (Section 4.1). To jus-
tify the use of a water jacket to reduce axial temperature gradients
and quantitate the temperature gradients obtained in a standard
still air oven, the mechanistic heat balance is solved to calculate
temperature distributions for the different experimental condi-
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 111

Fig. 1. Chromatograms showing the separation of omeprazole and four related compounds at flow rates (a) 0.08 mL/min (b) 0.30 mL/min and (c) 0.50 mL/min using two
columns: a BEH C18 column with fully porous particles and a Cortecs C18 column containing core shell particles. The inlet pressures for the BEH column were 152, 515
and 833 bar, and the inlet pressures for the Cortecs C18 column were 152, 506 and 796 bar. The structures of the solutes are shown to right: (1) H153/73 (2) H168/66 (3)
omeprazole (4) H193/61 (5-methyl omeprazole) and (5) H193/61 (6-methyl omeprazole). Acidic hydrogens with pKa close to 8 are shown in red. Column oven temperature
was set to 308 K and the time was converted to eluted volume for an easier comparison. The first eluting peak is NaNO3 indicating the column hold-up time. Note the
selectivity changes for peak 2 and 3.

tions (Section 4.2). A fast and accurate empirical approach is finally
developed based on the results in the previous sections and it is
verified by comparing predictions of elution profiles with experi-
mental data (Section 4.3).
4.1. Experimental design in pressure and temperature
To investigate the combined effect of temperature and pressure
on retention and selectivity, an experimental design in pressure-
temperature space was constructed. Experiments were performed
using a low flow rate and the pressure was adjusted by installing
a restrictor capillary after the column while the temperature was
adjusted using the column oven. This methodology has the benefit
of allowing nearly constant pressures and temperatures inside the
column.
The retention factor of each solute and the selectivity between
omeprazole and H168/66 were used responses. A model contain-
ing quadratic and interaction terms was fitted to each response
and excellent agreement was achieved between the regression
models and the experimental data with squared regression coef-
ficients above 0.998 for all models (Supplementary Data Table S1).
The models were very similar for the two columns which are evi-
dent from the similarity between the scaled and centered model
coefficients for the BEH column (Fig. 2a) and for the Cortecs col-
umn (Fig. 2b), respectively. Since the relative change in retention
with pressure and temperature is described in Fig. 2, absolute dif-
ferences in retention between the columns, e.g. due to different
surface coverage or total porosity, are not visible. Both columns
were C18 bonded resulting in the same primary retention mecha-
nism and hence similar temperature and pressure dependence in
the two columns [19]. The solutes under investigation here are not
positively charged and therefore less sensitive to silanol activity
making it difficult to notice differences in silanol activity between
the two columns. Considering the relative magnitude of the coef-
ficients (Fig. 2), we see that temperature has a negative effect on
retention while pressure has a positive effect which is the com-
mon behavior in reversed-phase LC [23,31]. Of more interest is the
fact that the solutes have widely different coefficients meaning that
they respond to changes in pressure and temperature differently.
This is rather unexpected since they have similar structure and size
and it indicates that chemically related solutes can demonstrate
selectivity shifts. For example, H193/61 differs from omeprazole
only by a methyl group and has a pressure coefficient almost twice
that of omeprazole. The interaction term has some importance for
the models and introduces a small curvature while the quadratic
terms are insignificant or of very low importance.
All compounds have a maximum in retention at high pressure
and low temperature while the pressure has the largest effect
except for H168/66 where pressure and temperature are equally
important (Fig. 3 and Supplementary Data Fig. S3). Omeprazole has
an increase of 36% in retention factor with the BEH column and
47% with the Cortecs columns for a 500 bar increase in pressure
measured at 35 ◦ C. Comparing the effect of pressure on retention
with previously reported values for compounds of similar molec-
ular weight [20,21], we see that a 50% increase in retention with
500 bar is quite common in RPLC. The selectivity between omepra-
zole and H168/66 increases with temperature, Fig. 3c and Fig. 3f,
which confirms that pressure and temperature variations in the
column can give rise to the selectivity shift observed when chang-
ing flow rate (Fig. 1 compound 2 and 3). At high flow rates the axial
temperature gradient makes the column warmer than the set tem-
perature in the column oven and thereby increasing the selectivity
112 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120

Fig. 2. Coefficients for the model terms in the regression models from the experimental design data for the retention factor responses using (a) The BEH column (b) the
Cortecs column. Insignificant terms at 95% confidence level have been removed from the models. T is column temperature and P is average column pressure. The differences
between the compounds reflect the differences in sensitivity for pressure and temperature variations.

Fig. 3. Response surfaces for omeprazole and H168/66 obtained from the regression models of the experimental design data from the two columns. The subfigures shows
the dependency of the pressure and temperature on the retention factors for (a, d) Omeprazole and for (b, e) H168/66 for the BEH column and the Cortecs C18 column,
respectively; the dependency on these peaks selectivity factors are shown in (c) and (f) for the two columns.

between omeprazole and H168/66. The response surfaces show compounds is shown as red in Fig. 1. The pKa -values were estimated
that the selectivity, in our case, is almost unaffected by pressure. through fitting experimental data to [35]:
On the other hand, due to the difference in pressure dependence s
between omeprazole and H193/61 (Fig. 2), the selectivity increases k0 + k1 × 10w pH−pKa
k= s pH−pK (14)
about 5% with pressure between these two compounds. The selec- 1 + 10 w a

tivity between omeprazole and H153/73 also changes slightly with

where k0 and k1 are the retention factors for the acid and corre-
both temperature and pressure.
sponding basic forms, respectively. Eq. (14) neglects changes in
It has been reported that the pressure and temperature can
the surface properties of the adsorbent, but is sufficiently accurate
affect the degree of protonation of solutes when they are close to
to estimate the pKa -values of specific solvent mixtures and tem-
their pKa [21]. Omeprazole, H168/66 and H153/73 all have pKa -
peratures in the pH range applicable here. The pKa -values were
values near the mobile phase pH and the acidic hydrogen of these
estimated at three corner points in the experimental design; low
temperature, low pressure and high temperature, low pressure and
low temperature, high pressure. The sw pH (pH-scale where the pH
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120
Fig. 4. Retention factor on the BEH column as a function of sw pH in the mobile phase
for the compounds (a) Omeprazole (b) H153/73 (c) H168/66, having pKa -values close
to the working point, sw pH = 7.77 (dotted line). Filled symbols represent experimental
data and lines are fits to Eq. (14) while pKa -values are shown as empty symbols.
Pressures given in the figure are average column pressures and the flow rate was
0.10 mL/min in all cases.
is measured in the organic aqueous eluent and the electrode is
calibrated using aqueous solution) for each mobile phase was mea-
sured at 308 K and atmospheric pressure with the assumption that
it did not change in the chromatograph. This is not strictly true, but
due to difficulties to measure pH online or at elevated pressures; it
was deemed an acceptable approximation.
Fig. 4 shows the retention factor of omeprazole, H153/73 and
H168/66 as functions of sw pH in the mobile phase at different
pressures and temperatures. For a 24 K increase in temperature
from 308 K to 332 K, the pKa changed by −0.26, −0.23 and −0.27
units for omeprazole, H153/73 and H168/66, respectively (pKa open
symbols in Fig. 4), resulting in a decrease in protonation with
an increase in temperature. Increasing the pressure 629 bar (from
107 bar to 736 bar) gave changes in pKa of 0.18, 0.15 and 0.10 units
for omeprazole, H153/73 and H168/66, respectively, i.e. increas-
ing the pressure increases the protonation. While omeprazole and
H153/73 are almost completely protonated at the pH of the mobile
phase used in the experimental design (dotted line in Fig. 4), the
small shift in pKa -value does not significantly affects the reten-
tion and hence the retention factor in the experimental design
is affected only by pure pressure and temperature effects. How-
ever, the experimental design pH is very close to the pKa -value
of H168/66 and a small shift would cause measurable changes in
113
retention for this compound. Most likely, the comparatively large
temperature coefficient observed in the coefficient plot (Fig. 2) for
H168/66 is due to a change in protonation with temperature since
the change with temperature is about three times as large as the
change with pressure. This means that the temperature and pres-
sure gradients encountered in UHPLC can give rise to change in the
degree of protonation in addition to the pure temperature and pres-
sure effects. Previous results have shown that temperature affects
the degree of protonation of a compound if it is close to its pKa
value [37,38] and since viscous heating can increase the column
temperature up to 20 ◦ C at very high pressures [14,16], it is impor-
tant to take this into consideration for example when assessing
the robustness of an analytical method. The pressure has also been
shown to affect protonation in certain scenarios [39,40], but it is
less investigated compared to temperature. The pressure effect can
be of importance when converting a method from UHPLC to HPLC,
which is common in the pharmaceutical industry where method
development is done on UHPLC and routine analysis in the quality
control labs mostly uses HPLC [1,4].
4.2. The mechanistic modelling approach
First, the pressure and temperature profiles over the column
were calculated. To verify the calculations, the pressure at the col-
umn inlet and the temperature measured at the column surface
near the outlet of the column were compared to calculated val-
ues (Tables 2 and 3). Note that the highest flow rate experiment
for each column was used to obtain values of the Blake-Kozeny-
Carman parameter () and the external heat conductivity (hext ) so
the calculated column surface temperature and inlet pressure for
this flow rate are therefore model fits rather than model predic-
tions. The error in pressure was at most a few bar, giving a relative
error below 5% in all cases, which is acceptable. The surface tem-
perature of the columns was predicted with relative errors of less
than 0.5% in all cases, which is excellent.
4.2.1. Estimation of elution profiles
The temperature profile over the two columns for the high-
est flow rate shows an axial temperature gradient of 9 K and a
radial gradient of 0.3–0.4 K for both columns (Fig. 5). The axial tem-
perature gradient reported in previous studies [14,41] for a BEH
C18 columns with the same dimensions was ca 18 K. This is in
agreement with our findings since the eluent used in those studies
[14,41] was neat acetonitrile and hence the flow rate which was
needed to give a pressure drop of around 900 bar was 1.45 mL/min.
The power of generated heat is directly proportional to the flow
rate, see last term in Eq. (1), and since the flow rate in our case was
0.54 mL/min for the BEH column, a little under half the temper-
ature increase reported in [14,41] is expected. It should be noted
that the difference in heat capacity and heat conductivity of the
eluent used here and neat acetonitrile also affects the temperature
gradient making the comparison approximate.
There are very small visible differences between the temper-
ature profiles of the two columns (Fig. 5) which agree with the
measured temperature on the column surface (Table 2 and 3). The
temperature profiles for all flow rates for both columns are pre-
sented in Supplementary Data Figs. S4 and S5 . We conclude that
for this experimental system, the temperature gradients are very
similar in the BEH column containing fully porous particles and the
Cortecs column containing coreshell particles.
Elution profiles were predicted for all flow rates and compared
to the experimental ones. The retention time was predicted with a
relative error below 0.5% for the BEH column (Table 2), a relative
error below 0.8% for the Cortecs column (Table 3) and in many cases
the relative error was much lower. This agreement is very good
and can achieve excellent predictions of selectivity. The agreement
114 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120

Table 2
Comparing experimental (exp) and calculated (calc) column inlet pressures (Pin ), column surface temperature (T) and retention time (tr ) for omeprazole (OM) and H168/66
(H168) using the mechanistic models for the BEH column in still air conditions. Fv is set volumetric flow rate and RE denotes the relative error for each property. The
temperature at the column surface was measured at 86 mm from the column inlet. Note that the flow rate 0.54 mL/min was used to estimate the adsorption isotherm
parameters and other physico-chemical parameters.

Fv [mL/min] Pin,exp [bar] Pin,calc [bar] RE [%] Texp [K] Tcalc [K] RE [%] tr,OM,exp [min] tr,OM,calc [min] RE [%] tr,H168,exp [min] tr,H168,calc [min] RE [%]

0.08 152 149 1.91 308.0 308.3 −0.09 19.483 19.492 −0.04 19.342 19.345 −0.02
0.2 351 347 1.25 309.0 309.7 −0.22 8.187 8.177 0.12 8.033 8.003 0.37
0.3 515 512 0.50 311.1 311.4 −0.11 5.712 5.685 0.47 5.515 5.493 0.39
0.4 676 673 0.47 313.4 313.5 −0.03 4.455 4.437 0.41 4.247 4.228 0.43
0.5 833 833 −0.05 315.7 315.7 0.01 3.697 3.695 0.05 3.487 3.480 0.19
0.54 897 897 0.03 316.6 316.6 0.01 3.478 3.478 0.00 3.262 3.260 0.05

Table 3
Comparing experimental (exp) and calculated (calc) column inlet pressures (Pin ), column surface temperatures (T) and retention times (tr ) for omeprazole (OM) and H168/66
(H168) using the mechanistic models for the Cortecs column in still air conditions. Fv is set volumetric flow rate and RE denotes the relative error for each property. The
temperature at the column surface was measured at 78 mm from the column inlet. Note that the flow rate 0.58 mL/min was used to estimate the adsorption isotherm
parameters and other physico-chemical parameters.

Fv [mL/min] Pin,exp [bar] Pin,calc [bar] RE [%] Texp [K] Tcalc [K] RE [%] tr,OM,exp [min] tr,OM,calc [min] RE [%] tr,H168,exp [min] tr,H168,calc [min] RE [%]

0.08 152 145 4.80 308.0 308.3 −0.09 15.263 15.252 0.08 15.780 15.782 −0.01
0.20 348 335 3.85 308.5 309.5 −0.34 6.435 6.427 0.13 6.533 6.490 0.66
0.30 506 490 3.08 310.5 311.2 −0.21 4.468 4.445 0.52 4.450 4.415 0.79
0.40 656 642 2.10 312.7 313.0 −0.11 3.475 3.452 0.67 3.402 3.380 0.64
0.50 796 790 0.79 315.0 315.1 −0.02 2.873 2.863 0.35 2.768 2.762 0.24
0.58 905 905 0.01 316.7 316.7 0.00 2.538 2.538 0.00 2.418 2.418 0.00

between the shape of experimental and calculated elution profiles
were also compared and the result was good (Supplementary Data
Figs. S6 and S7). The calculated elution profiles were somewhat
sharper possibly due to extra-column band broadening.
4.2.2. Water jacket experiments
Exchanging the standard still air column oven of the Acquity
UPLC system with a water jacket with turbulent flow increases the
external heat conductivity (hext ) of the column about 175 times
(Supplementary Data Table S5). The heat generated by viscous fric-
tion is more effectively removed making the axial temperature
gradient decrease and the radial temperature gradient increase
[25,42,43]. This will cause significant losses in column efficiency
and it is therefore not recommended to use a water jacket or water
bath in routine analysis [25,42,43]. However, for some experiments
where the column efficiency is not important, the water jacket can
be a useful option, as we will discuss here and in the next section.
Temperature distributions for the two columns in water jackets at
pressure drops of around 900 bar were calculated and they showed
that the temperature inside the column never exceeded 0.6 K above
the set temperature (308 K) of the water thermostat and mobile
phase pre-heater (Fig. 6). The BEH and Cortecs columns behaved
very similarly with a somewhat smaller radial gradient in the
Cortecs column, most likely due to its higher bed thermal conduc-
tivity (Table 1). The temperature profiles with the columns in water
jackets for all flow rates are presented in Supplementary Data Fig.
S8 and Fig. S9. The radial temperature gradient calculated here are a
few degrees lower than previous results for similar pressure drops
[14]. We believe this to be due to the higher heat conductivity of
the eluent used here (20/80, w/w acetonitrile/buffer) compared to
that used in the previous study (neat acetonitrile) which makes the
generated heat more effectively evacuated. To validate the calcula-
tions, omeprazole elution profiles were calculated and compared to
experimental ones for different flow rates. The agreement in both
retention time and column efficiency was very good (Supplemen-
tary Data Fig. S10). The relative error in retention time was below
0.9% in all cases except for 0.50 mL/min with the BEH column (1.6%)
(Supplementary Data Tables S3 and S4).
Based on the observation that in our experimental setup, the
temperature in the column can be controlled to ±1 K, even at very
high pressures, using a water jacket makes it possible to isolate
the pressure effects on the retention without the need for restric-
tor capillaries. There are some advantages of using a water jacket
instead of a restrictor capillary; (i) for each pressure level, the length
of the restrictor capillary must be adjusted manually by an opera-
tor making automation difficult. (ii) To avoid leakage, steel screws
and ferrules must be used which cannot be removed once attached.
After adjusting the length of the capillary, new steel ferrules must
be added and the old ones discarded. (iii) Due to their small diam-
eter, solid particles can easily get stuck in the capillaries making
the pressure resistance change dramatically which makes it hard
to replicate the experiment with the same pressure resistance or
use the same restrictor for a long time. A water jacket or water
bath is usually available in larger laboratories and we believe that a
laboratory engineer used to routine work has an easier time chang-
ing to a water bath than installing restrictor capillaries. However,
restrictor capillaries can still be useful for research where there are
other demands.
Some UHPLC systems have the option of forced air mode for
the column oven (e.g. the Vanquish UHPLC system from Thermo
Fisher Scientific). We did not investigate how forced air condi-
tions compared with a water jacket in this study. However, forced
air conditions were compared to an insulated column (similar to
still air conditions) in SFC by Kaczmarski et al. [44]. They found
large efficiency losses when the column was operated under forced
air conditions compared to insulated conditions which indicate a
larger radial temperature gradient. This is also the result when still
air and water jacket setups are compared [42] indicating that forced
air conditions and water jacket give similar results in this regard.
The heat transfer coefficient from the column surface to the sur-
rounding was estimated to ca 20 W m−2 K−1 for still air conditions
and 3500 W m−2 K−1 for water jacket conditions in this study, while
it has been reported to be around 100 W m−2 K−1 for forced air
conditions [45]. However, the value for forced air is largely depen-
dent on the equipment and could vary much between systems. We
believe care should be taken if forced air is used instead of a water
jacket and additional experiments should be done to confirm that
the axial temperature gradient is negligible in forced air mode.
Comparing the retention factor at different flow rates for
omeprazole and H168/66 obtained with the still air column oven to
the water jacket setup clearly shows the difference between these
two thermal environments (Fig. 7). The retention factor has been
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120
Fig. 5. Calculated temperature distributions, in K, inside the columns placed in the
still-air column oven. a) The BEH C18 column containing fully porous particles at flow
rate 0.54 mL/min which results in a pressure drop over the column of 751 bar. b) The
Cortecs C18 column containing coreshell particles at flow rate 0.58 mL/min which
results in a pressure drop over the column of 781 bar. The column oven and mobile
phase pre-heater was set to 308 K and the mobile phase was 20/80, w/w acetoni-
trile/phosphate buffer. Note that the axial temperature gradient is dominating and
that the columns behave very similar.
normalized with the retention factor for the lowest flow rate in
each case to make for an easier comparison. For the water jacket
setup, the retention factor increases only due to an increase in
pressure. For the still air oven, there is also an increase in tem-
perature along the column which decreases the retention factor
compared to the water jacket setup. The difference between the
water and air setups is larger for H168/66 than for omeprazole
which is in agreement with the observation from the experimental
design investigation in Section 4.1, where the coefficient for tem-
perature was largest for H168/66 (Fig. 2). The regression models
for H168/66 from the experimental design data showed that the
temperature had a significant quadratic term. This is indeed also
noted when considering Fig. 7. Comparing the adjusted coefficient
of determination [46] for a linear and a quadratic polynomial fit to
the still air data in Fig. 7 for H168/66, shows that a quadratic fit has
the highest adjusted coefficient of determination (0.9994 vs. 0.9999
with the BEH column and 0.9964 vs 0.9995 with the Cortecs column
for a first and second degree polynomial, respectively). However, a
linear polynomial is still a good approximation for H168/66 and can
be used without much loss of accuracy. In the case of omeprazole, a
quadratic polynomial gives no noticeable improvement compared
115
Fig. 6. Calculated temperature distributions, in K, inside the columns placed in the
turbulent-water jacket. a) The BEH C18 column containing fully porous particles at
flow rate 0.50 mL/min which results in a pressure drop over the column of 741 bar.
b) The Cortecs C18 column containing coreshell particles at flow rate 0.50 mL/min
which results in a pressure drop over the column of 764 bar. The column thermostat
and mobile phase pre-heater was set to 308 K and the mobile phase was 20/80,
w/w acetonitrile/phosphate buffer. Note that the radial temperature gradient is
dominating and that the temperature inside the column never exceeds 309 K.
to a linear polynomial (0.9998 vs. 0.9997 with the BEH column and
0.9999 vs 1.0000 with the Cortecs column for a first and second
degree polynomial, respectively).
Experimental elution profiles for omeprazole and H168/66,
obtained with the column in a water jacket, are shown for a high
and a low flow rate with the BEH column in Fig. 8. No selectivity
shift is apparent between omeprazole and H168/66 with flow rate
changes when a water jacket is used (Fig. 8). Since the selectivity
shift observed with the still air column oven (Fig. 1) was mainly
due to the change of the axial temperature gradient, the disap-
pearance of the selectivity shift strengthens our result that there
is no axial temperature gradient in the column when using a water
jacket. So, controlling the column temperature with a water jacket,
or equivalently a water bath with stirring, provides a simple and
convenient experimental approach to determine the retention fac-
tor as a function of pressure at constant temperature. However,
care should be taken because some eluents having very low heat
conductivity, like neat acetonitrile, can give rise to significant radial
temperature gradients making this approach inaccurate.
116 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120

1.35
a) BEH 0.08 mL/min
a) OM
Normalized retention factor

1.3

1.25

1.2
Omeprazole
1.15 H168/66

1.1 BEH air

BEH water b) BEH 0.50 mL/min
Cortecs air
1.05
Cortecs water

1
1.35

b) H168/66
Normalized retention factor

1.3

1.25
1.4 1.6 1.8 2 2.2
1.2 Elution volume [mL]

1.15 Fig. 8. Experimental elution profiles obtained with the BEH column in a water jacket
holding the temperature constant at 308 K changing the flow rate: (a) 0.08 mL/min
1.1 (b) 0,50 mL/min. The increase in retention volume with flow rate is only due to
the increase in average column pressure (90 bar and 494 bar for 0.08 mL/min and
0.50 mL/min, respectively). Note that the selectivity shift, caused by the axial tem-
1.05 perature gradient, when the column was in the still air column oven (Fig. 1) is not
present in this case, indicating that the axial temperature gradient is very small with
1 the column in a water jacket.
100 200 300 400 500
Average pressure [bar] The reference condition is chosen as a flow rate where the temper-
ature in the column is equal to the set temperature and the average
Fig. 7. The retention factor as a function of the average column pressure for the
pressure in the column is around 100 bar. The next step is to deter-
water jacket and column oven thermostat options for the solutes (a) Omeprazole
and (b) H168/66. The retention factor for the lowest pressure was used for nor- mine how the average column pressure and the average column
malization of the retention factor making for easier comparisons. For the turbulent temperature changes with flow rate. Linear functions for pressure
water setup, the retention factor is a function of pressure only since the tempera- and temperature as functions of flow rate are assumed:
ture inside the column is 308 ± 1 K (Fig. 6). For the still air setup, the temperature
inside the column increases from 308 K to 317 K in the axial direction (Fig. 5) which Pavg (Fv ) = a3 (Fv − Fv,0 ) + Pavg,0 (16)
reduces the retention factor compared to the waters setup.
and
4.3. The empirical modelling approach Tavg (Fv ) = a4 (Fv − Fv,0 ) + Tavg,0 (17)

The goal with the empirical approach is to develop a method-
ology that, given a certain flow rate, estimates the average
temperature increase due to viscous heating and predicts the shifts
in retention for the solutes due to the pressure and temperature
gradients. This approach targets method development in an indus-
trial setting using a system with a conventional still air column
oven, and so we aspire to have few experiments, straightforward
calculations and conventional equipment. Technically, the exper-
iments described here to study pressure could be done with a
water jacket or restrictor capillaries. We suggest the use of a water
jacket before restrictor capillaries for the reasons discussed in Sec-
tion 4.2. Regarding the efficiency, neither the water jacket nor the
restrictor would yield efficiencies consistent with the conventional
still air oven. The radial temperature gradients present with the
water jacket would reduce efficiencies [43] and the low flow rate
and extra-column dispersion with the restrictor would also yield a
lower efficiency [7].
The idea is to first approximate how the retention factor varies
with pressure and temperature using a linear expression,
   
k (T, P) = a1 T − Tavg,0 + a2 P − Pavg,0 + k0
where subscript 0 refers to a reference condition and subscript avg
refers to an average pressure or temperature inside the column.
where Pavg,0 and Tavg,0 are the average column pressure and tem-
perature at the reference condition. By knowing the coefficients
a1 –a4 , the retention factor at arbitrary flow rates can be predicted.
These coefficients can be determined from only four injections fol-
lowing the procedure outlined below.
First, the injection at 0.08 mL/min, with the column oven set to
308 K, was taken as the reference condition and from this injection
all constants with subscript 0 in Eqs. (15–17) were determined.
To determine a1 , an injection was done at 0.08 mL/min but with
the column oven set to 332 K which gave a negligible change in
pressure. To determine a2 , an injection at 0.50 mL/min was done
with the column in a water jacket with the temperature 308 K. Since
the average column temperature was 308 ± 1 K even at high flow
rates with the water jacket (Fig. 6), the temperature can be taken as
constant. Next, an injection at the highest flow rate (0.54 mL/min
for BEH and 0.58 mL/min for Cortecs) was done with the standard
still air column oven set to 308 K. With the average pressure from
this injection, a3 could be determined. To determine a4 , the average
column temperature for this injection was found by solving for the
temperature in Eq. (15) since the retention factor is known, i.e.
(15)
   
k Tavg , Pavg − a2 Pavg − Pavg,0 − k0
Tavg = + Tavg,0 (18)
a1
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 117

The retention factor for omeprazole, the main component, was

used in Eq. (18). Now, with only four injections per column, all a) 0.08 mL/min Experiments
unknown constants have been determined. Mechanistic
The results for omeprazole and H168/66 are listed in Table 4 Empirical
for the BEH column and Table 5 for the Cortecs columns while
results for H153/73 and H193/61 are presented in Supplemen-
tary Data Table S6. Surprisingly, with this minimalistic, empirical
approach, it was possible to predict the retention factors of omepra-
zole, H153/73 and H193/61 with relative errors less than 1% which
is very good considering the approximations done in the procedure
by assuming a linear axial temperature gradient and no interaction 18.5 19 19.5 20 20.5
term in Eq. (15). In fact, in many cases the relative errors were below
0.5% which is excellent. However, the relative errors for H168/66
were around 2% for highest flow rates (not shown), while this is b) 0.30 mL/min
still acceptable, it could be improved. The large relative errors for
H168/66 are probably due to the additional effect of pKa varia-
tions (Fig. 4) discussed in Section 4.1 giving a nonlinear relationship
between the retention factor and temperature (Fig. 2). One way to
account for this is to calculate an apparent average column temper-
ature with Eq. (18) valid only for H168/66. By doing this, the relative
errors decreased to below 1% for all cases (Table 4 and Table 5).
So, for solutes close to their pKa , calculating an apparent average
temperature seems to be a way to improve the predictions. 5.2 5.4 5.6 5.8 6
Comparing the average column temperature predicted with the
empirical model with the average temperature along the center of
the column found with the mechanistic model (Table 4 and Table 5) c) 0.50 mL/min
revealed a relative error around −0.25% for the BEH column and
around −1% for the Cortecs column. The relative errors in the pre-
dicted average pressure were also larger for the Cortecs column
(Table 5). We do not have any logical explanation for the difference
between the BEH and Cortecs column in this regard. It could just
be that the average column temperature is predicted with relative
errors between 0.25–1% with this approach and variations in this
interval are random. Experiments on more systems are needed to
establish if the difference is due to random noise or has a physi- 3.3 3.4 3.5 3.6 3.7 3.8 3.9
cal explanation. However, we think that the predictions from the Time [min]
empirical approach of the average column temperature are good
enough for method development to get an approximation of the Fig. 9. Predicted elution profiles with the empirical and mechanistic approaches
actual temperature in the column. For example, if the quality con- compared with corresponding experimental elution profiles obtained with the BEH
trol method in this study for the BEH column, run at the highest column; (a) 0.08 mL/min (b) 0.30 mL/min (c) 0.50 mL/min. The large peak is omepra-
zole and the smaller one is H168/66. Standard still air column oven set to 308 K.
flow, would be described, we would say that the column oven was
set to 308 K and that, due to viscous heating, the average temper-
ature inside the column was 315 K. This was also investigated by Figs. 9 and 10 compares elution profiles for omeprazole and
Nováková et al. [47] who suggested lowering the set temperature H168/66 calculated from the empirical and mechanistic approaches
in UHPLC when converting a method from HPLC to UHPLC. with experimental ones. The reference condition, 0.08 mL/min, and
To predict elution profiles of Gaussian shape, the column effi- two predictions (0.30 mL/min and 0.50 mL/min) are shown and the
ciency is needed and to avoid additional experiments the efficiency agreement with experiments is very good and accurately predicts
was described by a linear relationship between the reference con- the selectivity shift observed in Fig. 1. In e.g. quality control meth-
dition and the highest flow rate. This is a fair approximation if the ods in the pharmaceutical industry, the resolution between critical
reference state is not taken at an extremely low flow rate where peak pairs is one of the most important factors in a robustness
the eddy diffusion term dominates in the van Deemter curve [28]. study [3]. The predicted peaks in Figs. 9 and 10 shows that the
From van Deemter curves given in dimensionless reduced inter- retention time and separation efficiency could be predicted accu-
stitial velocity available in the literature, it is possible to estimate rately enough to estimate the resolution between omeprazole and
approximately where the linear region is for a certain column. It H168/66; i.e. the critical peak pair in this separation. If predictions
could also be the case that the van Deemter curve is known for show that the resolution falls below the acceptance criterion for
the system under investigation, or a similar system, from the ini- certain flow rates, we would recommend that the actual experi-
tial method development phase. If so, the minimum HETP could be ment is performed at this flow rate since the predictions of the
used for the efficiency which would underestimate the peak width resolution have a larger uncertainty compared to the selectivity
at the lowest flow rates, but these are often the least interesting because both retention and efficiency are needed to calculate the
peaks since the pressure and temperature gradients are very small resolution.
at low flow rates. A third approach, if there is a need to increase To summarize, the mechanistic approach gave the most accurate
the accuracy of the efficiency estimate is to record one additional results and is based on fundamental theory making it excellent for
peak at intermediate flow rate and use a second-degree polynomial research into UHPLC, but is not practical to use in routine work since
function for interpolation instead of a linear function. This could no commercial software implementing the equations are available
be undertaken if the initial predictions indicate that the resolution today. On the other hand, the empirical approach gives predictions
criterion between a critical peak-pair is not met at some flow rates. with good accuracy, requires few experiments and simple calcu-
118 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120

Table 4
Calculated average column pressures (Pavg ), average column temperatures (Tavg ) and retention factors (k) for omeprazole (OM) and H168/66 (H168) using the empirical
approach for the BEH column compared with experimental values. Fv is set volumetric flow rate and RE denotes the relative error for each property. The highest and lowest
flow rate experiments were used in the calculations which explain the exact estimations for these two cases.

Fv [mL/min] Pavg,exp [bar] Pavg,calc [bar] RE [%] Tavg,exp a [K] Tavg,calc [K] RE [%] kOM,exp kOM,calc RE [%] kH168,exp kH168,calc RE [%]

0.08 93 93 0.00 308.3 308.0 0.10 6.643 6.643 0.00 6.591 6.591 0.00
0.20 206 205 0.41 309.4 309.9 −0.18 7.032 7.052 −0.28 6.883 6.862 0.30
0.30 300 298 0.48 310.8 311.6 −0.24 7.399 7.393 0.08 7.113 7.089 0.34
0.40 392 391 0.08 312.3 313.2 −0.28 7.733 7.733 −0.01 7.327 7.315 0.17
0.50 484 485 −0.13 313.9 314.8 −0.29 8.069 8.074 −0.06 7.545 7.541 0.05
0.54 522 522 0.00 314.5 315.4 −0.30 8.211 8.211 0.00 7.632 7.632 0.00
a
This value is the average temperature along the center of the column calculated with the mechanistic model.

Table 5
Calculated average column pressures (Pavg ), average column temperatures (Tavg ) and retention factors (k) for omeprazole (OM) and H168/66 (H168) using the empirical
approach for the Cortecs column compared with experimental values. Fv is set volumetric flow rate and RE denotes the relative error for each property. The highest and
lowest flow rate experiments were used in the calculations which explain the exact estimations for these two cases.

Fv [mL/min] Pavg,exp [bar] Pavg,calc [bar] RE [%] Tavg,exp a [K] Tavg,calc [K] RE [%] kOM,exp kOM,calc RE [%] kH168,exp kH168,calc RE [%]

0.08 93 93 0.00 308.2 308.0 0.06 5.659 5.659 0.00 5.898 5.898 0.00
0.20 202 194 4.00 309.3 310.6 −0.41 6.019 5.988 0.51 6.152 6.098 0.88
0.30 290 278 3.99 310.6 312.7 −0.68 6.311 6.263 0.76 6.322 6.264 0.92
0.40 374 363 2.97 312.1 314.9 −0.89 6.583 6.537 0.70 6.478 6.431 0.72
0.50 453 447 1.36 313.6 317.0 −1.09 6.838 6.812 0.39 6.622 6.597 0.37
0.58 514 514 0.00 314.8 318.7 −1.25 7.031 7.031 0.00 6.731 6.731 0.00
a
This value is the average temperature along the center of the column calculated with the mechanistic model.

lations making it more suitable for routine use in UHPLC method
development. However, this approach should be investigated fur-
ther before being applied in an industrial setting. A separation
system containing compounds like peptides and proteins which,
due to their larger molecular weight, are more sensitive to changes
in pressure could be studied to determine if the assumptions about
linearity in Eqs. (15)–(17) are still valid. In routine analysis, gradient
elution is very common and the extension of empirical approach to
gradient elution would be very valuable. If the mobile phase com-
position is changed moderately and mobile phase properties like
density and viscosity are almost constant, the empirical approach,
as presented here, could work satisfactorily. With large variation
in mobile phase composition, we would most likely need to derive
new relations who take this into account. However, this is outside
the scope of the current study and could be a subject for a future
investigation.
5. Conclusions
they were very similar for the two columns. However, when the
column was placed in a water jacket instead of the column oven,
the axial temperature gradient disappeared and the radial gradient
was ca 0.6 ◦ C since the external heat conductivity increases more
than 100 times in the water jacket. Therefore, by using a water
jacket, the retention factor could be studied as a function of pressure
instead of a restrictor capillary installed after the column. Restrictor
capillaries make experiments hard to automate since they require
manual handling to adjust length and hence pressure resistance. It
should be stressed that the water jacket option decreases column
efficiency due to a larger radial temperature gradient compared to a
still air oven and is therefore not recommended for routine analysis
[42,43]. However, it is useful for determining certain parameters or
studying certain phenomena such as the influence of pressure on
the retention factor.
The new approach developed here requires only four exper-
iments by assuming a linear relationship between the retention
factor and pressure and temperature:

i The first experiment is used to describe a reference condition

Due to the very high pressures in UHPLC, significant tem-
where the pressure and temperature gradients are negligible.
perature and pressure gradients develop in the column. We
It is an injection at a low flow rate where the back pressure is
demonstrated, with a quality control method for the drug omepra-
around 100 bar.
zole, that such gradients can give rise to retention and selectivity
ii The second experiment is for determining how the retention fac-
shifts when the flow rate, and hence back pressure, is changed. At
tor depends on temperature. It is an injection at low flow rate,
high flow rates, there was baseline separation between omepra-
as in (i), but with the column oven set to a temperature 20 ◦ C
zole and a related impurity (H168/66) and at low flow rates there
higher than in the first experiment.
was co-elution. By an experimental design in pressure-temperature
iii The third experiment is for determining how the retention fac-
space, we were able show how the selectivity varied with pres-
tor depends on pressure. It is an injection at high flow rate (back
sure and temperature and found that the temperature was most
pressure around 1000 bar) with the column placed in a water
important, but that the pressure also affected selectivity to some
jacket. The water jacket minimizes the axial temperature gradi-
degree.
ent so the pressure effect is dominating.
To predict selectivity shifts, we developed and validated an
iv The final experiment estimates the pressure and the average col-
accurate yet practical and easy-to-use, empirical approach for pred-
umn temperature as a function of flow rate with the column in
ication of retention time shifts in UHPLC due to variations in the
the standard still air oven. It is done at high flow rate using the
pressure and temperature gradients. To this end, the complete
standard column oven with the same set temperature as for the
temperature distribution for two types of columns, one packed
reference condition in step (i).
with fully porous particles and one packed with coreshell parti-
cles, were calculated with 20/80, w/w acetonitrile/buffer as eluent.
When using a conventional still air column oven the axial tempera- Retention predictions made by the empirical approach was val-
ture gradient was ca 9 ◦ C and the radial gradient was 0.3–0.4 ◦ C and idated against experiments and the agreement was found to be
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 119

perature on retention and making it possible to adjust factors such

a) 0.08 mL/min Experiments as the column oven temperature accordingly.
Mechanistic
Empirical
Acknowledgements

This work was supported by the Swedish Knowledge Founda-

tion via the project “Predictive Separation of Biopharmaceuticals”
(grant number 20150233), by the ÅForsk Foundation via the project
“Improved Purification Procedures to Satisfy Modern Drug Quality
Assurance and Environmental Criteria” (grant number 15/497), by
14.5 15 15.5 16 16.5 the Swedish Research Council (VR) via the project “Fundamental
Studies on Molecular Interactions aimed at Preparative Separations
and Biospecific Measurements” (grant number 2015-04627) and
b) 0.30 mL/min by the grant 2015/18/M/ST8/00349 from the National Science Cen-
tre, Poland. Kevin Jenkins and Fabrice Gritti at Waters, Milford, MA,
USA, are acknowledged for their excellent technical support regard-
ing the column properties. We are also grateful to Anders Karlsson
and Tomas Leek at AstraZeneca R&D, Gothenburg, Sweden for the
gift of the Coreshell column and valuable discussions.

Appendix A. Supplementary data

4.2 4.3 4.4 4.5 4.6
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.chroma.2016.11.
c) 0.50 mL/min 050.

References

[1] M.W. Dong, K. Zhang, Ultra-high-pressure liquid chromatography (UHPLC) in

2.65 2.7 2.75 2.8 2.85 2.9 2.95
Time [min]
Fig. 10. Predicted elution profiles with the empirical and mechanistic approaches
compared with corresponding experimental elution profiles obtained with the
Cortecs column: (a) 0.08 mL/min (b) 0.30 mL/min (c) 0.50 mL/min. The large peak
is omeprazole and the smaller one is H168/66. Standard still air column oven set to
308 K.
very good with relative errors below 1% for all solutes. In many
cases the relative errors were even below 0.5% which is excellent
considering how few experiments are needed for the predications
and the simplicity of the model. The estimated average column
temperature was validated against the one calculated from the
thermodynamic heat balance and was found to have relative errors
between 0.2–1.25%.
The presented approach was shown to work as intended for
the separation system of omeprazole and its impurities. However,
before recommending it as a general method, it should be evaluated
also for a separation system containing more pressure sensitive
compounds like proteins and peptides. Many separations are done
in gradient elution where the mobile phase composition changes
continuously during the analysis. To increase its usefulness, the
approach should be extended to gradient elution conditions and
this can be the subject of a follow-up study.
We believe that the empirical approach has the potential to be
used in routine UHPLC method development to assess robustness
for flow rate variations and to estimate the actual column temper-
ature, i.e. the temperature increase due to the viscous heating. This
approach has also potential to aid the method transfer from a HPLC
to an UHPLC method by estimating the effect of pressure and tem-
method development, TrAC 63 (2014) 21–30, http://dx.doi.org/10.1016/j.trac.
2014.06.019.
[2] M. Szalka, J. Kostka, E. Rokaszewski, K. Kaczmarski, Analysis of related
substances in bisoprolol fumarate on sub-2-␮m adsorbents, Acta Chromatogr.
24 (2012) 163–183, http://dx.doi.org/10.1556/AChrom.24.2012.2.2.
[3] D. Åsberg, M. Nilsson, S. Olsson, J. Samuelsson, O. Svensson, S. Klick, J. Ennis, P.
Butterworth, D. Watt, S. Iliadou, A. Karlsson, J.T. Walker, K. Arnot, N. Ealer, K.
Hernqvist, K. Svensson, A. Grinell, P.-O. Quist, A. Karlsson, T. Fornstedt, A
quality control method enhancement concept—continual improvement of
regulatory approved QC methods, J. Pharm. Biomed. Anal. 129 (2016)
273–281, http://dx.doi.org/10.1016/j.jpba.2016.06.018.
[4] R. Holm, D.P. Elder, Analytical advances in pharmaceutical impurity profiling,
Eur. J. Pharm. Sci. 87 (2016) 118–135, http://dx.doi.org/10.1016/j.ejps.2015.
12.007.
[5] V. González-Ruiz, A.I. Olives, M.A. Martín, Core-shell particles lead the way to
renewing high-performance liquid chromatography, TrAC 64 (2015) 17–28,
http://dx.doi.org/10.1016/j.trac.2014.08.008.
[6] J. De Vos, M. De Pra, G. Desmet, R. Swart, T. Edge, F. Steiner, S. Eeltink,
High-speed isocratic and gradient liquid-chromatography separations at
1500 bar, J. Chromatogr. A 1409 (2015) 138–145, http://dx.doi.org/10.1016/j.
chroma.2015.07.043.
[7] N. Tanaka, D.V. McCalley, Core–shell, ultrasmall particles, monoliths, and
other support materials in high-performance liquid chromatography, Anal.
Chem. 88 (2016) 279–298, http://dx.doi.org/10.1021/acs.analchem.5b04093.
[8] S. Görög, The changing face of pharmaceutical analysis, TrAC 26 (2007) 12–17,
http://dx.doi.org/10.1016/j.trac.2006.07.011.
[9] D. Åsberg, M. Leśko, J. Samuelsson, K. Kaczmarski, T. Fornstedt, Method
transfer from high-pressure liquid chromatography to ultra-high-pressure
liquid chromatography. I. A thermodynamic perspective, J. Chromatogr. A
1362 (2014) 206–217, http://dx.doi.org/10.1016/j.chroma.2014.08.051.
[10] D. Åsberg, J. Samuelsson, M. Leśko, A. Cavazzini, K. Kaczmarski, T. Fornstedt,
Method transfer from high-pressure liquid chromatography to
ultra-high-pressure liquid chromatography. II. Temperature and pressure
effects, J. Chromatogr. A 1401 (2015) 52–59, http://dx.doi.org/10.1016/j.
chroma.2015.05.002.
[11] F. Gritti, G. Guiochon, Complete temperature profiles in ultra-high-pressure
liquid chromatography columns, Anal. Chem. 80 (2008) 5009–5020, http://dx.
doi.org/10.1021/ac800280c.
[12] F. Gritti, M. Martin, G. Guiochon, Influence of viscous friction heating on the
efficiency of columns operated under very high pressures, Anal. Chem. 81
(2009) 3365–3384, http://dx.doi.org/10.1021/ac802632x.
[13] K. Kaczmarski, J. Kostka, W. Zapała, G. Guiochon, Modeling of thermal
processes in high pressure liquid chromatography: I. Low pressure onset of
thermal heterogeneity, J. Chromatogr. A 1216 (2009) 6560–6574, http://dx.
doi.org/10.1016/j.chroma.2009.07.020.
[14] K. Kaczmarski, F. Gritti, J. Kostka, G. Guiochon, Modeling of thermal processes
in high pressure liquid chromatography: II. Thermal heterogeneity at very
120 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120
high pressures, J. Chromatogr. A 1216 (2009) 6575–6586, http://dx.doi.org/
10.1016/j.chroma.2009.07.049.
[15] S. Fekete, J. Fekete, D. Guillarme, Estimation of the effects of longitudinal
temperature gradients caused by frictional heating on the solute retention
using fully porous and superficially porous sub-2 ␮m materials, J. Chromatogr.
A 1359 (2014) 124–130, http://dx.doi.org/10.1016/j.chroma.2014.07.030.
[16] F. Gritti, G. Guiochon, Mass transfer resistance in narrow-bore columns
packed with 1.7 ␮m particles in very high pressure liquid chromatography, J.
Chromatogr. A 1217 (2010) 5069–5083, http://dx.doi.org/10.1016/j.chroma.
2010.05.059.
[17] J. Kostka, F. Gritti, K. Kaczmarski, G. Guiochon, Modified
Equilibrium-Dispersive Model for the interpretation of the efficiency of
columns packed with core–shell particle, J. Chromatogr. A 1218 (2011)
5449–5455, http://dx.doi.org/10.1016/j.chroma.2011.06.019.
[18] F. Gritti, G. Guiochon, Comparison of heat friction effects in narrow-bore
columns packed with core–shell and totally porous particles, Chem. Eng. Sci.
65 (2010) 6310–6319, http://dx.doi.org/10.1016/j.ces.2010.09.019.
[19] D. Åsberg, J. Samuelsson, T. Fornstedt, A fundamental study of the impact of
pressure on the adsorption mechanism in reversed-phase liquid
chromatography, J. Chromatogr. A 1457 (2016) 97–106, http://dx.doi.org/10.
1016/j.chroma.2016.06.036.
[20] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Investigation of the effect
of pressure on retention of small molecules using reversed-phase
ultra-high-pressure liquid chromatography, J. Chromatogr. A 1209 (2008)
195–205, http://dx.doi.org/10.1016/j.chroma.2008.09.021.
[21] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Further investigations of
the effect of pressure on retention in ultra-high-pressure liquid
chromatography, J. Chromatogr. A 1217 (2010) 276–284, http://dx.doi.org/10.
1016/j.chroma.2009.11.041.
[22] S. Fekete, J.-L. Veuthey, D.V. McCalley, D. Guillarme, The effect of pressure and
mobile phase velocity on the retention properties of small analytes and large
biomolecules in ultra-high pressure liquid chromatography, J. Chromatogr. A
1270 (2012) 127–138, http://dx.doi.org/10.1016/j.chroma.2012.10.056.
[23] D.V. McCalley, The impact of pressure and frictional heating on retention,
selectivity and efficiency in ultra-high-pressure liquid chromatography, TrAC
63 (2014) 31–43, http://dx.doi.org/10.1016/j.trac.2014.06.024.
[24] A. Makarov, R. LoBrutto, P. Karpinski, Y. Kazakevich, C. Christodoulatos, A.K.
Ganguly, Investigation of the effect of pressure and liophilic mobile phase
additives on retention of small molecules and proteins using reversed-phase
ultrahigh pressure liquid chromatography, J. Liq. Chromatogr. Related
Technol. 35 (2012) 407–427, http://dx.doi.org/10.1080/10826076.2011.
601494.
[25] A. de Villiers, H. Lauer, R. Szucs, S. Goodall, P. Sandra, Influence of frictional
heating on temperature gradients in ultra-high-pressure liquid
chromatography on 2.1 mm I. D. columns, J. Chromatogr. A 1113 (2006)
84–91, http://dx.doi.org/10.1016/j.chroma.2006.01.120.
[26] J. Billen, K. Broeckhoven, A. Liekens, K. Choikhet, G. Rozing, G. Desmet,
Influence of pressure and temperature on the physico-chemical properties of
mobile phase mixtures commonly used in high-performance liquid
chromatography, J. Chromatogr. A 1210 (2008) 30–44, http://dx.doi.org/10.
1016/j.chroma.2008.09.056.
[27] K. Kaczmarski, G. Guiochon, Modeling of the mass-transfer kinetics in
chromatographic columns packed with shell and pellicular particles, Anal.
Chem. 79 (2007) 4648–4656, http://dx.doi.org/10.1021/ac070209w.
[28] F. Gritti, A. Cavazzini, N. Marchetti, G. Guiochon, Comparison between the
efficiencies of columns packed with fully and partially porous C18-bonded
silica materials, J. Chromatogr. A 1157 (2007) 289–303, http://dx.doi.org/10.
1016/j.chroma.2007.05.030.
[29] D. Antos, K. Kaczmarski, P. Wojciech, A. Seidel-Morgenstern, Concentration
dependence of lumped mass transfer coefficients: linear versus non-linear
chromatography and isocratic versus gradient operation, J. Chromatogr. A
1006 (2003) 61–76, http://dx.doi.org/10.1016/S0021-9673(03)00948-8.
[30] B.E. Poling, J.M. Prausnitz, J.P. O’Connell, The Properties of Gases and Liquids,
5th ed., McGraw-Hill Professional, 2001, 2016, http://dx.doi.org/10.1036/
0070116822.
[31] G. Guiochon, D.G. Shirazi, A. Felinger, A.M. Katti, Fundamentals of Preparative
and Nonlinear Chromatography, 2nd ed., Academic Press, Boston, MA, 2006.
[32] E.J. Wilson, C.J. Geankoplis, Liquid mass transfer at very low reynolds
numbers in packed beds, Ind. Eng. Chem. Fundam. 5 (1966) 9–14, http://dx.
doi.org/10.1021/i160017a002.
[33] J.H. Knox, G.R. Laird, P.A. Raven, Interaction of radial and axial dispersion in
liquid chromatography in relation to the infinite diameter effect, J.
Chromatogr. A 122 (1976) 129–145, http://dx.doi.org/10.1016/S0021-
9673(00)82240-2.
[34] F. Gritti, Y. Kazakevich, G. Guiochon, Measurement of hold-up volumes in
reverse-phase liquid chromatography: definition and comparison between
static and dynamic methods, J. Chromatogr. A 1161 (2007) 157–169, http://
dx.doi.org/10.1016/j.chroma.2007.05.102.
[35] D. Åsberg, M. Leśko, J. Samuelsson, A. Karlsson, K. Kaczmarski, T. Fornstedt,
Combining chemometric models with adsorption isotherm measurements to
study omeprazole in RP-LC, Chromatographia 79 (2016) 1283–1291, http://
dx.doi.org/10.1007/s10337-016-3151-8.
[36] T. Undin, J. Samuelsson, A. Törncrona, T. Fornstedt, Evaluation of a combined
linear–nonlinear approach for column characterization using modern
alkaline-stable columns as model, J. Sep. Sci. 36 (2013) 1753–1761, http://dx.
doi.org/10.1002/jssc.201201132.
[37] E. Bosch, P. Bou, H. Allemann, M. Rosés, Retention of ionizable compounds on
HPLC. pH scale in methanol-water and the pK and pH values of buffers, Anal.
Chem. 68 (1996) 3651–3657, http://dx.doi.org/10.1021/ac960104l.
[38] S.M.C. Buckenmaier, D.V. McCalley, M.R. Euerby, Determination of ionisation
constants of organic bases in aqueous methanol solutions using capillary
electrophoresis, J. Chromatogr. A 1026 (2004) 251–259, http://dx.doi.org/10.
1016/j.chroma.2003.11.007.
[39] C.P. Samaranayake, S.K. Sastry, In situ measurement of pH under high
pressure, J. Phys. Chem. B 114 (2010) 13326–13332, http://dx.doi.org/10.
1021/jp1037602.
[40] K. Sue, T. Morita, K. Totsuka, Y. Takebayashi, S. Yoda, T. Furuya, T. Hiaki,
Determination of dissociation constants of hexanoic, heptanoic, and benzoic
acids to 673 K and 30 MPa by potentiometric pH measurements, J. Chem. Eng.
Data 55 (2010) 4823–4826, http://dx.doi.org/10.1021/je1004164.
[41] F. Gritti, G. Guiochon, Heat exchanges in fast, high-performance liquid
chromatography. A complete thermodynamic study, Anal. Chem. 80 (2008)
6488–6499, http://dx.doi.org/10.1021/ac8003902.
[42] J. Kostka, F. Gritti, G. Guiochon, K. Kaczmarski, Modeling of thermal processes
in very high pressure liquid chromatography for column immersed in a water
bath: application of the selected models, J. Chromatogr. A 1217 (2010)
4704–4712, http://dx.doi.org/10.1016/j.chroma.2010.05.018.
[43] F. Gritti, G. Guiochon, Optimization of the thermal environment of columns
packed with very fine particles, J. Chromatogr. A 1216 (2009) 1353–1362,
http://dx.doi.org/10.1016/j.chroma.2008.12.072.
[44] K. Kaczmarski, D.P. Poe, A. Tarafder, G. Guiochon, Efficiency of supercritical
fluid chromatography columns in different thermal environments, J.
Chromatogr. A 1291 (2013) 155–173, http://dx.doi.org/10.1016/j.chroma.
2013.03.024.
[45] K. Kaczmarski, D.P. Poe, G. Guiochon, Numerical modeling of elution peak
profiles in supercritical fluid chromatography. Part I-Elution of an unretained
tracer, J. Chromatogr. A 1217 (2010) 6578–6587, http://dx.doi.org/10.1016/j.
chroma.2010.08.035.
[46] J.N. Miller, J.C. Miller, Statistics and Chemometrics for Analytical Chemistry,
6th ed., Pearson Education, Harlow, UK, 2010.
[47] L. Nováková, J.L. Veuthey, D. Guillarme, Practical method transfer from high
performance liquid chromatography to ultra-high performance liquid
chromatography: the importance of frictional heating, J. Chromatogr. A 1218
(2011) 7971–7981, http://dx.doi.org/10.1016/j.chroma.2011.08.096.
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 1

Supplementary data
A practical approach for predicting retention time shifts due to
pressure and temperature gradients in ultra-high-pressure liquid
chromatography
Dennis Åsberg, Marcin Chutkowski, Marek Leśko, Jörgen Samuelsson,
Krzysztof Kaczmarski and Torgny Fornstedt

2
1
1

Fig. S1: Separation of (1) 5-methyl omeprazole and (2) 6-methyl omeprazole on the Cortecs column
using 40/60, w/w, methanol/phosphate buffer (30 mM, pH 7.25) as mobile phase. Flow rate was 0.1
mL/min, detection was at 220 nm and temperature was 35°C while 1 µL of 0.02 g/L of methyl
omeprazole mixture was injected.

Fig. S2: ISEC to determine the external porosity of the columns using polymer standards in neat
tetrahydrofuran. Dashed lines are linear regression to the four polymer standards with highest molecular
weight which are totally excluded and extrapolation to zero molecular weight gives the external volume of
the bed.
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 2

Table S1: Quality of regression (R2) and quality of prediction (Q2) for the regression models obtained with
the software MODDE 11 (Umetrics, Sweden) in temperature-pressure space.

Column Statistic Retention factor αOM-H168/66

OM H168/66 H193/61 H153/73
R2 0.9999 0.9982 0.9999 0.9998 0.9991
BEH
Q2 0.9995 0.9876 0.9993 0.9989 0.9988
R2 0.9999 1.0000 1.0000 1.0000 1.0000
Cortecs
Q2 0.9997 0.9995 0.9998 0.9998 1.0000

Fig. S3: Response surfaces for the retention factor of H153/73 and H193/61 in pressure-temperature
space for the BEH column and the Cortecs column.
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 3

Fig. S4: Calculated temperature distributions using the mechanistic model for the BEH column in still air
column oven for flow rates 0.08, 0.20, 0.30, 0.40, 0.50 and 0.54 mL/min. Column pressure drops
indicated in figure.

Fig. S5: Calculated temperature distributions using the mechanistic model for the Cortecs column in still
air column oven for flow rates 0.08, 0.20, 0.30, 0.40, 0.50 and 0.58 mL/min. Column pressure drops
indicated in figure.
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 4

Fig. S6: Calculated peaks using the mechanistic model compared to experimental ones for the BEH
column in still air column oven. Flow rates, from left to right; 0.08, 0.20, 0.30, 0.40, 0.50 and 0.54
mL/min.

Fig. S7: Calculated peaks using the mechanistic model compared to experimental ones for the coreshell
Cortecs column in still air column oven. Flow rates, from left to right; 0.08, 0.20, 0.30, 0.40, 0.50 and
0.58 mL/min.
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 5

Fig. S8: Calculated temperature distributions using the mechanistic model for the BEH column turbulent
water jacket for flow rates 0.08, 0.20, 0.30, 0.40 and 0.50 mL/min. Column pressure drops indicated in
figure.

Fig. S9: Calculated temperature distributions using the mechanistic model for the Cortecs column in
turbulent water jacket for flow rates 0.08, 0.20, 0.30, 0.40, 0.50 and 0.54 mL/min. Column pressure
drops indicated in figure.
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 6

Fig. S10: Calculated omeprazole peaks using the mechanistic model versus experimental ones for the two
columns with turbulent water jackets. a) The BEH C18 column with flow rates, from left to right; 0.08,
0.20, 0.30, 0.40 and 0.50 mL/min. b) The Cortecs column with flow rates, from left to right; 0.08, 0.20,
0.30, 0.40, 0.50 and 0.54 mL/min. Note that the highest pressure drop gives the largest elution volume.

Table S3: Validation of the mechanistic model for the BEH column in turbulent water jacket. Fv is
volumetric flow rate, Pin is the pressure at the column inlet and tr is the retention time of omeprazole while
RE denotes the relative error.

Fv Pin,exp Pin,calc RE tr,exp tr,calc RE

[mL/min] [bar] [bar] [%] [s] [s] [%]
0.08 147 147 0.20 19.233 19.250 -0.09
0.20 348 351 -0.72 8.195 8.188 0.08
0.30 523 527 -0.80 5.807 5.780 0.46
0.40 705 712 -1.02 4.615 4.612 0.07
0.50 864 900 -4.11 3.877 3.938 -1.59

Table S4: Validation of the mechanistic model for the Cortecs column in turbulent water jacket.

Fv Pin,exp Pin,calc RE tr,exp tr,calc RE

[mL/min] [bar] [bar] [%] [s] [s] [%]
0.08 147 144 2.24 15.132 15.132 0.00
0.20 346 339 1.97 6.465 6.422 0.67
0.30 516 509 1.38 4.567 4.527 0.88
0.40 689 684 0.78 3.632 3.600 0.87
0.50 865 862 0.38 3.077 3.063 0.43
0.54 935 935 0.00 2.915 2.915 0.00
 D. Åsberg et al. / J. Chromatogr. A 1479 (2017) 107–120 7

Table S5: Estimated parameters in the mechanistic modelling obtained from the inverse method.
Notation are as follows: ξ is the Blake-Kozeny-Carman parameter, hext is external heat conductivity, τ1 is a
tortuosity parameter, T0,avg and P0,avg are the average temperature and pressure at the lowest flow rate
(0.08 mL/min), H0 is the Henry constant at the lowest flow rate, E is the enthalpy of adsorption, ΔVm is
solute partial molar volume difference between stationary and mobile phase. OM denotes omeprazole and
H168 denotes H168/66 while Water means turbulent water jacket column thermostat and Air means still
air column oven thermostat.

Parameter BEH C18 Cortecs C18

System OM-Water OM-Air H168-Air OM-Water OM-Air H168-Air
ξ 146.2 146.2 146.2 193.6 191.0 191.0
hext [W m-2 K-1] 3500 18.39 18.39 3500 19.58 19.58
τ1 2.735 2.922 3.341 1.372 1.314 2.166
T0,avg [K] 308.01 308.27 308.27 308.11 308.25 308.25
P0,avg [Pa] 8918038 9188155 9188155 8717611 8920533 8920533
H0 9.27 9.42 9.34 10.66 10.76 11.21
E [J mol-1] 4862 4862 6064 2580 21380 6959
ΔVm [mol cm-3] -17.78 -16.94 -13.74 -18.64 -24.90 -13.40

Table S6: Relative error (RE) in estimated retention factors (k) from the empirical approach for H153/73
(EP impurity E) and H193/61 (6-methyl omeprazole).

BEH C18 Cortecs C18

Fv [mL/min] RE kH153/73 [%] RE kH193/61 [%] Fv [mL/min] RE kH153/73 [%] RE kH193/61 [%]
0.08 0.00 0.00 0.08 0.00 0.00
0.20 0.06 -0.17 0.20 0.63 0.32
0.30 0.40 0.06 0.30 0.99 0.62
0.40 0.55 0.13 0.40 0.95 0.57
0.50 0.73 0.15 0.50 0.72 0.27
0.54 0.61 0.15 0.58 0.33 -0.11
 Journal of Chromatography A, 1457 (2016) 97–106

Contents lists available at ScienceDirect

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

A fundamental study of the impact of pressure on the adsorption

mechanism in reversed-phase liquid chromatography
Dennis Åsberg, Jörgen Samuelsson ∗ , Torgny Fornstedt ∗
Department of Engineering and Chemical Sciences, Karlstad University, SE-65188 Karlstad, Sweden

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

Article history: A fundamental investigation of the pressure effect on individual adsorption sites was undertaken
Received 18 April 2016 based on adsorption energy distribution and adsorption isotherm measurements. For this purpose,
Received in revised form 10 June 2016 we measured adsorption equilibrium data at pressures ranging from 100 to 1000 bar at constant flow
Accepted 12 June 2016
and over a wide concentration range for three low-molecular-weight solutes, antipyrine, sodium 2-
Available online 15 June 2016
naphthalenesulfonate, and benzyltriethylammonium chloride, on an Eternity C18 stationary phase. The
adsorption energy distribution was bimodal for all solutes, remaining clearly so at all pressures. The
Keywords:
bi-Langmuir model best described the adsorption in these systems and two types of adsorption sites
UHPLC
Pressure were identified, one with a low and another with a high energy of interaction. Evidence exists that the
Adsorption isotherm low-energy interactions occur at the interface between the mobile and stationary phases and that the
Retention factor high-energy interactions occur nearer the silica surface, deeper in the C18 layer. The contribution of each
Loading capacity type of adsorption site to the retention factor was calculated and the change in solute molar volume
from the mobile to stationary phase during the adsorption process was estimated for each type of site.
The change in solute molar volume was 2–4 times larger at the high-energy site, likely because of the
greater loss of solute solvation layer when penetrating deeper into the C18 layer. The association equi-
librium constant increased with increasing pressure while the saturation capacity of the low-energy site
remained almost unchanged. The observed increase in saturation capacity for the high-energy site did
not affect the column loading capacity, which was almost identical at 50- and 950-bar pressure drops
over the column.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction
To increase the throughput while maintaining column effi-
ciency in liquid chromatography (LC), the current trend is toward
smaller stationary phase particles and higher pressure limits [1,2].
Ultra-high-pressure liquid chromatography (UHPLC) is now close
to becoming the mainstream technique in method development,
giving a 3–10-fold increase in throughput [3]. In UHPLC, 1000-bar
pressure drops over the column are common, which influences
retention, selectivity, and efficiency [4]. In reversed-phase liquid
chromatography (RPLC), retention normally increases with increas-
ing pressure and decreases with increasing temperature, while the
effects of frictional heating and pressure are similar in magnitude
for small solutes under typical UHPLC conditions [5,6]. While fric-
tional heating effects have been thoroughly studied using various
∗ Corresponding authors.
E-mail addresses: Jorgen.Samuelsson@kau.se (J. Samuelsson),
Torgny.Fornstedt@kau.se (T. Fornstedt).
approaches [7–10], the pressure effect in UHPLC has only been
investigated by measuring the pressure dependence of the reten-
tion factor [11–17]. The retention factor cannot give information of
different adsorption sites and is only related to the initial slope of
the adsorption isotherm.
For low-molecular-weight compounds in RPLC, increases of
up to 50% in retention factor have been observed with 500-bar
pressure increases [12–14]. According to classical thermodynamic
reasoning, the driving force behind this increased retention is that
the solute has a smaller molar volume when adsorbed into the
stationary phase than into the mobile phase [18]. While in the
mobile phase, the solute is surrounded by a solvation layer that
is partially lost when entering the stationary phase. Le Chatelier’s
principle states that the system, when pressure is increasing, will
strive to counteract the change by lowering the pressure, decreas-
ing its volume and driving the solute mobile–stationary phase
equilibrium toward the stationary phase. The pressure effect com-
monly increases with molecular size and for polar or ionic species
[14,16,17], so a larger change in partial molar volume is observed
for such solutes. McCalley et al. [12–14] proposed that the larger

http://dx.doi.org/10.1016/j.chroma.2016.06.036
0021-9673/© 2016 Elsevier B.V. All rights reserved.
98 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106
change in partial molar volume for polar and ionic solutes is due
to a more substantial loss of the solvation layer surrounding the
solute when entering the C18 layer. Neutral solutes are in principle
solvated by the organic solvent in the mobile phase, so the com-
plex can penetrate the C18 layer without much loss of its solvation
layer. Aside from molecular volume and polarity, the solute struc-
ture [14], mobile phase additives [16], and stationary phase type
[13,15] can also influence the pressure dependence. Experimental
parameters usually assumed to be constant in LC, such as column
dimensions and packing material volume, also depend slightly on
pressure and must be accounted for in UHPLC to correctly deter-
mine the total column porosity [19,20].
Better understanding the types of interactions in LC calls for
adsorption isotherm data over a wide concentration range [21].
From such data, the adsorption energy distribution (AED) and the
adsorption isotherm model can be determined, giving detailed
information about the various interactions taking place [22,23].
Adsorption isotherm data have previously been used to study the
effects of temperature [7], organic modifier [24], pH [25], and ionic
strength [26] in RPLC. However, to our knowledge, the only study
reporting adsorption isotherm data for different pressures was con-
ducted in a limited pressure interval (i.e., 50–250 bar) and with a
protein type of solute, insulin, and without the important AED-tool
[27].
The aim of this study is to measure adsorption data for three
small solutes, two ionic (i.e., one organic cation and one organic
anion) and one neutral, at five different pressure plateaus between
100 and 1000 bar using restriction capillaries to adjust the pressure.
The effect of pressure on the AED and on the adsorption isotherm
parameters will then be investigated. For heterogeneous adsorp-
tions, i.e., a solute that adsorbs on multiple types of adsorption
sites with different adsorption energies, the change in solute molar
volume upon adsorption will be determined for each type of site.
Finally, since an increase in saturation capacity can give an increase
in column loading capacity, the adsorption isotherms will be used
to simulate elution profiles with different sample loads at low and
high pressures. Plotting the sample load against the apparent plate
number will give a measure of the loading capacity.
2. Theory
From classical thermodynamics we can derive the following
relationship between the retention factor, k, and the local pressure,
P, [18]:
k V
ln =− P − ln F0 + ln K0 (1)
F RT
where V = Vstat − Vmob is the difference in solute molar volume
between the stationary and mobile phases, T is the absolute tem-
perature, F is the phase ratio (i.e., the ratio between the stationary
and mobile phase volumes), which is pressure dependent, R is the
universal gas constant, and the subscript 0 indicates a reference
pressure. In RPLC, an increase in retention with pressure is usu-
ally observed, so V < 0, which indicates a smaller solute molar
volume in the stationary phase than in the mobile phase [12–14].
The retention factor is directly proportional to the initial slope of
the adsorption isotherm. Most polar or ionic solutes adsorb on two
or more types of adsorption sites, which have different adsorption
energies [7,21,28]. A simple model describing two-site adsorption
of a solute is the bi-Langmuir adsorption isotherm [21],
b1 C b2 C
q(C) = qs,1 + qs,2 (2)
1 + b1 C 1 + b2 C
where qs,i denotes the monolayer saturation capacity of site i,
bi is the association equilibrium constant for site i, and q and C
are the stationary- and mobile-phase concentrations, respectively.
Without losing any generality, the first adsorption site of the bi-
Langmuir model is defined as the low-energy site and the second
as the high-energy site (b2 ≥ b1 ). The retention factor is related to
the Henry’s constants (i.e. distribution constant), Hi , through
dq
k = lim F = F (qs,1 b1 + qs,2 b2 ) = F (H1 + H2 ) = FHtot (3)
C→0 dC
It follows from Eq. (3) that the retention factor is the sum of the
contributions of all adsorption sites, making V in Eq. (1) a kind
of average for all adsorption sites. A similar situation arises when
considering the molar enthalpies and entropies derived from the
van’t Hoff equation for solutes adsorbing on two of more adsorption
sites [7].
The perturbation peak (PP) method [29] is an accurate method
for acquiring adsorption isotherm data. The column is equilibrated
with a solute concentration, C’, and then the retention volume, VR ,
of a small PP is related to the local slope of the adsorption isotherm
at that concentration through
dq VR (C  ) − V0
| = (4)
dC C=C  FV0
where V0 is the column hold-up time. The major drawback of the PP
method is that considerable time and solute are required to deter-
mine the whole adsorption isotherm. A fast and precise alternative
is the elution by characteristic points (ECP) method [21], which
determines adsorption isotherm data from the diffuse rear of an
overloaded elution profile using an equation similar to Eq. (4) [30].
The drawback of the ECP method compared with the PP method
is that the maximum achievable solute concentration is lower due
to the dilution of the sample band. A proven strategy is to use an
accurate method, such as the PP method, for a certain reference con-
dition and then to change a certain parameter, such as pressure, in
a step-wise fashion, determining the adsorption isotherm using a
rapid method, such as ECP, for each new condition. This approach
has been used to study variations in the adsorption isotherm with,
for example, changing modifier content [24], pH [25] and ionic
strength [26].
3. Material and methods
3.1. Chemicals
The mobile phase was HPLC-grade methanol (Fischer Scien-
tific, Loughborough, UK) and acetate buffer (30 mM, pH 4.70)
prepared from water with a conductivity of 18.2 M cm deliv-
ered by a Milli-Q Plus 185 water purification system (Merck
Millipore, Billerica, MA, USA) and from sodium acetate (≥99.0%)
and acetic acid (≥99.8%), both purchased from Merck (Darmstadt,
Germany). Sodium nitrate (≥99%) used as the column hold-up time
marker, benzyltriethylammonium chloride “BTEAC” (99%) used as
the cationic solute, antipyrine “AP” (≥99.0%) used as the neutral
solute, and sodium 2-naphthalenesulfonate “SNS” (≥99.0%) used
as the anionic solute, were all from Sigma-Aldrich (St. Louis, MO,
USA). The aqueous buffer and all sample solutions were filtered
through 0.2-␮m nylon filter membranes (Whatman, Maidstone,
UK) before use. The mobile phase was prepared by weighing so that
the methanol fraction was 10.43% w/w, 9.35% w/w, and 4.01% w/w
for AP, SNS, and BTEAC, respectively.
3.2. Instrumentation
An Acquity H-Class Bio chromatograph (Waters, Milford, MA,
USA) with a quaternary solvent manager, a 10-␮L flow-through
needle, a column oven, and a PDA detector with a 500-nL flow cell
was used in all experiments. The extra-column volume from the
needle to the detector was 17 ␮L and a delay of 0.75 s was measured
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106
between the start of data acquisition and opening of the injection
valve. Both the extra-column volume and the delay were adjusted
for in the experimental data. The flow rate was 200 ␮L/min and the
column oven was set to 40 ◦ C. Under these conditions, the pressure
drop across the column was approximately 100 bar, meaning that
there are no temperature gradients in the column due to frictional
heating [5].
The Acquity H-Class Bio chromatograph delivers a set volumet-
ric flow rate at the pressurized side of the pump [31], i.e., the
expansion of the eluent will occur after the restriction capillary,
leading to a constant eluent velocity equaling the set velocity from
the instrument up to the restriction capillary. To confirm that the
volumetric flow rate was correct regardless of the back pressure,
the mass flow was measured after the detector using a mini CORI-
FLOW Coriolis mass flow meter (Bronkhorst High-Tech B.V., Ruurlo,
Netherlands) with an accuracy of ±0.2% of the mass flow. The mass
flow was converted to volumetric flow using the density of the
mobile phase at current pressure and temperature, calculated using
REFPROP v 9.1 from the US National Institute of Standards and Tech-
nologies. The volumetric flow rate at the high pressure side of the
pump was confirmed to be independent of the back pressure. The
relative error in the volumetric flow was 0.3%, so the long-term
accuracy of the flow rate was estimated to be 0.6 ␮L/min at a flow
rate of approximately 200 ␮L/min.
The column was a 50 × 2.1 mm Kromasil Eternity C18 column
(AkzoNobel Pulp and Performance Chemicals, Bohus, Sweden) with
an average particle size of 2.5 ␮m. According to the manufac-
turer, the specific surface area was 233 m2 /g, the pore volume was
0.53 cm3 /g, the C18 surface coverage was 1.78 ␮mol/m2 , the pack-
ing density was 0.69 g/cm3 , and the total carbon content was 8.38%
(w/w). The column hold-up volume estimated using NaNO3 as a
dead volume marker was 95.6 ␮L at approximately 100 bar and the
mobile phase composition 10.43%, w/w methanol.
To adjust the pressure at constant flow rate, LC PEEKsil tubing
with inner diameter 25 ␮m (SGE Analytical Science, Milton Keynes,
U.K.) was used in lengths of 5, 10, 15, and 20 cm and installed after
the column. The additional extra-column volume due to the restric-
tor capillary and the union connecting the column outlet capillary
to the restrictor capillary (approximately 0.7 ␮L) was adjusted for
in the experimental data.
3.3. Procedure
The PP method was used for measuring adsorption isotherms in
the reference state with no restriction capillary attached. The con-
centration range was 0–79.5 mM for AP, 0–22.0 mM for BTEAC, and
0–30.4 mM for SNS with 20–25 concentration plateaus in between.
Perturbation peaks were conducted in duplicate 0.5-␮L injections
at a sample concentration of approximately 90% of the plateau con-
centration under investigation. At each pressure plateau (i.e., 0-,
5-, 10-, 15-, and 20-cm restriction capillary), analytical peaks were
obtained with 0.5-␮L injections and overloaded peaks with 10-␮L
injections. The concentrations of the overloaded injections were
79.5 and 8.0 mM for AP, 22.0 and 4.4 mM for BTEAC, and 30.4 and
3 mM for SNS. UV detection was done at 290, 304, and 262 nm for
AP, SNS, and BTEAC, respectively, with linear response.
3.4. Calculations
The slope of the adsorption isotherms were calculated using
the slope-ECP method [30] from the experimental elution pro-
file with the highest sample load for each solute and pressure.
The AED was calculated using the expectation-maximization (EM)
method [32] directly from the slope of the adsorption isotherm
[33]. The energy space was spaced in 500 grid points and 105 itera-
tions were used for the PP data and 108 iterations were used for
99
the ECP data. The energy space was determined by the inverse
maximum and minimum concentration data extrapolated a factor
10 [32]. Nonlinear parameter estimation was conducted using the
trust-region-reflective algorithm implemented in MATLAB 2015b
(MathWorks, Natick, MA, USA) with 100 random starting guesses.
The Equilibrium-Dispersive (ED) model [21] was used to describe
the mass balance of the solute in the column. Orthogonal collo-
cation on finite elements [34] was used to discretize the spatial
derivatives of the ED model and the Adams-Moulton method
implemented in the VODE procedure [35] was used to solve the
system of ordinary differential equations. The initial condition was
that the stationary phase was in equilibrium with pure mobile
phase, and Danckwerts-type boundary conditions were used at the
column inlet and outlet [21]. The pressure gradient, which was
assumed to be linear [19], was included as an additional component
specified by its initial condition and with infinitely low diffusion
and linear velocity equal to zero. In this way, the pressure depen-
dence of the adsorption isotherm could be modeled using the same
approach as in gradient elution [36,37].
4. Results and discussion
The investigation is organized as follows. First, how the column
porosity depends on pressure is investigated (section 4.1); then
the AEDs and adsorption isotherms are presented and the varia-
tion with pressure is discussed (section 4.2). From the adsorption
isotherm, the change in solute molar volume upon adsorption can
be estimated for each type of adsorption site and the results can be
related to the theory of heterogeneous adsorption. In section 4.3,
the effect of the pressure gradient on loading capacity is evaluated
using the acquired adsorption isotherm to simulate elution profiles
under varying loads.
4.1. Column porosity
When calculating the adsorption isotherm and the AED from
experimental data, the total porosity, εt , is needed in order to
derive the phase ratio, F = (1 − εt )/εt . However, it is not straight-
forward to determine the total porosity when operating in a wide
pressure range, because the volume of the steel column tube and
of the packing material will change with pressure [19]. Equations
for calculating the expansion of the steel column tube due to the
pressure difference, P, between the external and internal col-
umn walls have been derived by Martin and Guiochon [19]. In the
present work, the main pressure drop occurs after the column in the
restriction capillary, so the longitudinal stress on the column tube
is neglected. The column volume, (Vcol ), is therefore directly pro-
portional to the cross-sectional area, which in turn is proportional
to the radius squared. The ratio of the inner radius at pressure P to
a reference state at pressure P0 is given by [19]:
  2
rin 1
2
4 rext, 0 rin, 0 P
=1+ + (5)
rin, 0 3 3 r2 2 2
rext, 0 + rin, E
in, 0 0
A Young’s modulus, E, for steel equal to 2 × 106 bar [20] was used
and the external radius of the column was measured and found to
be 3.25 mm. In this study, the pressure ranges from 109 to 919 bar,
so we will use these as the pressure points. A pressure increase from
109 bar to 919 bar yields Vcol,919 /Vcol,109 = 1.00124, which equals a
0.21-␮L increase in the volume of the column used in this work.
The compression of the packing material, which in this work
mainly consist of silica and octadecane, can be calculated using the
compressibility, ˇ, of pure silica (1 × 10−6 bar−1 ), and pure octade-
100
cane (1 × 10−4 bar−1 ),
and then estimating the compressibility of
the mixture, ˇm , from Gritti et al. [20]:
ˇm = ␸Si ˇSi + ␸C18 ˇC18
where ϕ is the volume fraction. The volume fraction of octadecane
was approximated from the carbon weight fraction and from the
densities of silica (2.0 g/cm3 ) and liquid octadecane (0.78 g/cm3 )
[20]. The ratio of stationary phase volume at pressure P to a refer-
ence state at pressure P0 is then given by:
VS 1 − e−2ˇm P
= (7)
VS,0 2ˇm P
A pressure increase from 109 bar to 919 bar yields
Vs,919 /Vs,109 = 0.984. At 109 bar, the stationary phase volume
was measured to be 95.6 ␮L with NaNO3 , giving a 1.23-␮L
decrease in the stationary phase volume at 919 bar. The reduction
in stationary phase volume is caused mainly by the compression of
the octadecane layer over the silica backbone. The column hold-up
volume was measured at five pressures between 109 and 919 bar;
taking the column expansion into account, the total porosity was
determined from experimental data at these pressures. Compar-
ing the experimental results with the change in total porosity
predicted from Eq. (5) and Eq. (7) in Fig. 1 indicated very good
agreement between the experimentally obtained total porosity
and that estimated from the combination of steel tube expansion
and packing material compression. In Fig. 1, it is obvious that
the packing material compression is the most important factor,
the column expansion being more or less negligible. The small
difference at high pressure (see Fig. 1) could be due to uncertainty
in the compressibility of the silica and octadecane mixture. A linear
function was fitted to the experimental data with an R2 value of
0.9972:
εt (P) = 1.0229 × 10−5 × P + 0.5509
which is used to calculate the total porosity at arbitrary average
column pressures given in bar. In the following calculations of the
adsorption isotherms and AEDs, the phase ratio will be derived from
Eq. (8) for each pressure.
0.562
0.56
0.558
Total porosity
0.556
0.554
0.552
0.55
0 100
Fig. 1. Total porosity versus average column pressure for the Eternity C18 column measured experimentally with NaNO3 (black dots), estimated considering only expansion
of the steel column tube (dashed-dotted line), estimated considering only the compression of the packing material (dashed line), and estimated taking both expansion of the
steel tube and compression of the packing material into account (solid line).
D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106
4.2. AED and adsorption isotherm
To better understand the adsorption mechanism in RPLC, a set of
(6) adsorption data can be acquired to permit study of the evolution of
the adsorption isotherm parameters with pressure. The advantage
over studying only the retention factor is that the heterogeneity of
the adsorbent and the contribution of individual adsorption sites
to the overall retention factor of the solute can be assessed [7,21].
The adsorption equilibrium data for the solutes were measured
at low pressure over a wide concentration range using the PP
method. The data was analyzed using a rigorous approach based
on Scatchard plots, AED, model selection and verification of the
model by comparing experimental and calculated elution profiles
[22,23]. The results are shown in Fig. 2 for AP and in Supplemen-
tary Data Figs. S1 and S2 for BTEAC, and SNS, respectively. The
overloaded elution profiles and Scatchard plots, Fig. 2a, indicate
convex upwards adsorption isotherms for all solutes. The AED,
Fig. 2b, did not converge at low energy, but at least two adsorption
sites were clearly indicated. The lower limit of the AED is inversely
proportional to the maximum solute concentration [32], and it is
not experimentally feasible in this case to increase the concentra-
tion further due to either solubility or ionic strength issues [38].
The bi-Langmuir and the bi-Tóth models are examples of isotherm
models that have two different adsorption sites and are convex
upwards. The more common simpler one, the bi-Langmuir model,
has a symmetrical bi-modal AED while the bi-Tóth model has an
unsymmetrical bimodal distribution [22,39]. Because the AEDs in
this case are symmetrical and not unnaturally broad, there is no
indication of the bi-Tóth model. Therefore, the bi-Langmuir model
was fitted to the slope data obtained from the PP data, since this
model has a bimodal AED which agrees with our observations and
has been reported to accurately describes the adsorption of polar
solutes and ions in buffered mobile phases [24–26]. The estimated
parameters are presented in Supplementary Data Table S1 and the
(8)
fit were excellent for AP, Fig. 2c, and very good for SNS and BTEAC.
Using the estimated adsorption isotherm parameters, overloaded
elution profiles were calculated using the ED model and compared
with experimental ones, revealing good agreement. The overlap
areas [23] were 96.2% and 96.8% for 10-␮L injections of 8.0 and
Column tube
Packing material
Column tube + Packing material
Experimental
200 300 400 500 600 700 800 900 1000
Average column pressure [bar]
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106 101

a) b)

c) d)

Fig. 2. Characterization of antipyrine on Eternity-C18 at 0.20 mL/min and 30 ◦ C from perturbation pulse experiments. a) Scatchard plot from integrated raw data, b) AED
calculation using the EM-method with 105 iterations, 500 grid points in energy space and b in L/mol, c) Best fit to the adsorption isotherm derivative (symbols) using the
differentiated bi-Langmuir model (line) and d) Calculated elution profiles based on the fitting to the perturbation data are compared with experimental elution profiles for
10 ␮L injections of 8.0 and 79.5 mM antipyrine. The area overlap is 96.3 and 96.8% between experimental and calculated elution profiles for 8.0 and 79.5 mM antipyrine,
respectively.

Table 1
Adsorption isotherm parameters for antipyrine (AP), benzyltriethylammonium
chloride (BTEAC) and sodium 2-naphthalenesulfonate (SNS) on the Eternity C18 col-
umn at different pressures estimated using the ECP method. Site 1 is the low-energy
adsorption site and site 2 is the high-energy adsorption site.

Compound Pavg [bar] qs,1 [mM] b1 [L/mmol] qs,2 [mM] b2 [L/mmol]

AP 109 194 0.114 1.40 1.60

316 190 0.124 1.44 1.71
462 191 0.127 1.88 1.51
612 192 0.129 2.38 1.39
919 191 0.138 2.65 1.42

BTEAC 88 22.0 0.539 2.00 5.59

263 22.4 0.549 2.31 5.62
378 22.9 0.555 2.37 5.63
508 22.1 0.634 2.08 6.25
808 23.1 0.612 2.53 6.25

SNS 107 24.2 0.416 2.37 4.99

311 24.5 0.431 2.55 5.08
445 25.2 0.420 2.75 5.04
590 26.5 0.388 3.20 4.69
900 25.5 0.472 2.88 5.57

79.5 mM antipyrine (Fig. 2d), 93.4% and 93.0% for 10-␮L injections
of 4.4 and 22.0 mM BTEAC, and 89.5% and 96.8% for 10-␮L injections
of 3.04 and 30.4 mM SNS, respectively.
The bi-Langmuir model found at the reference state with the PP
method (as discussed above) was used as a starting point when
studying the parameter variation with pressure. The adsorption
isotherm model was assumed to be the same for all pressures and
only the numerical parameters may change. The fits were very
good, which is evident in Fig. 3 where experimental and calculated
elution profiles with the estimated bi-Langmuir model parameters
(Table 1) are compared. When comparing the elution profiles at
the lowest and highest pressures, the difference mainly represents Fig. 3. Experimentally obtained elution profiles (dotted lines) at two different col-
a shift toward longer retention times with no apparent change umn pressures obtained using restriction capillaries after the column and calculated
in peak shape. Fig. 4 shows the adsorption isotherms and their elution profiles using the bi-Langmuir model parameters estimated with the ECP
corresponding AEDs at different pressures for the three model com- method (solid lines). The injection volume was 10 ␮L and the sample concentrations
were 79.5 mM, 22.0 mM, and 30.4 mM for AP, BTEAC, and SNS, respectively.
pounds. Interestingly, the corresponding AED plots indicates the
two adsorption sites prevail at all pressures (Fig. 4a–c), in line with
the assumption (see above) that the adsorption isotherm model
102 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106

a) d)

b) e)

c) f)

Fig. 4. The AED (left column) and the estimated bi-Langmuir adsorption isotherm (right column) at the five pressure plateaus obtained using the slope-ECP method. The
AED calculation was done using the EM-method with 108 iterations, 500 grid points in energy space and b in L/mol.

per se does not change with pressure, only the numerical parame-
ters do. Both the high- and low-energy distributions are relatively
narrow and symmetrical which is consistent with the bi-Langmuir
model. For SNS, the AED did not converge for the low-energy dis-
tribution for the high pressure data. This is most likely because
of lack of sufficiently high concentration data, as discussed ear-
lier. This makes the low-energy distribution seem broader, but
this is for numerical reasons since the energy space is extrapo-
lated and the AED did not converge at the low-energy boundary.
A third adsorption site is present for SNS at the highest pressure,
which could indicate an increased heterogeneity of the adsorbent
or more probably merely be a numerical artifact. One should note
that the purpose of the AEDs shown in Fig. 4a–c is to compare the
difference between pressure plateaus and not revalidate the basic
adsorption isotherm model. The adsorption isotherms at different
pressures are shown in Fig. 4d–f and the trend is toward slightly
increasing overall saturation capacity and slightly increasing steep-
ness of the initial slope with increasing pressure. The adsorption
isotherms are all convex upwards, indicating that pressure does
not induce any adsorbate–adsorbate interactions. The trend in the
adsorption isotherms towards increasing overall adsorption is in
line with studies of the retention factor, representing the initial
slopes of these isotherms [12–14].
Two types of adsorption sites are seen in our experimental sys-
tem. Molecular simulation of adsorption in RPLC [40] has indicated
that 1-propanol on a C18 stationary phase with methanol–water
as the mobile phase has a bimodal density distribution. Most of
the molecules are located at the eluent–C18 layer interface, while
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106
Table 2
Change in solute molar volume upon adsorption on the stationary phase for the
two types of adsorption sites given with 95%-confidence intervals. VAnalytical is
the change in solute molar volume upon adsorption estimated directly from the
retention factor obtained from analytical peaks using Eq. (1).
Compound AP SNS BTEAC
VHigh [cm3 /mol] − 18 ± 2 − 9.9 ± 0.6 − 12 ± 2
VLow [cm3 /mol] − 5.3 ± 0.4 − 5.7 ± 0.7 − 6.4 ± 0.3
VAnalytical [cm3 /mol] − 7.1 ± 0.4 − 8.9 ± 0.2 − 9.9 ± 1.4
a small portion penetrates deep into the stationary phase and is
located near the silica surface. The adsorption of solutes on multiple
adsorption sites located at different depths in the stationary phase,
in close agreement with molecular simulations, has also been sug-
gested based on experimental results [28,41]. The adsorption site
at the eluent–C18 layer interface has a relatively low adsorption
energy and a high saturation capacity, while the adsorption site
deeper in the C18 layer has a higher adsorption energy but a much
lower saturation capacity [28,41]. There are many possible sources
of surface heterogeneity in C18-bonded silica, such as elemental
impurities, bond strains of the silica tetrahedron, residual silanol
groups, and disordered C18 chains, which could result in different
adsorption sites [28]. Table 1 presents the adsorption parameters
for the bi-Langmuir model. Note that these are numerical fitting
parameters and the bi-Langmuir model can have multiple sets of
parameters giving almost the same isotherm shape, therefore the
physical meaning of each parameter should be viewed critically,
especially for AP where the isotherm has low curvature making
the parameter estimation difficult. However, some general trends
are apparent. The association equilibrium constant increases with
increasing pressure for both sites. This can be related to Le Chate-
lier’s principle, i.e., the pressure driving the equilibrium between
the mobile and stationary phases toward the stationary phase is due
to the reduction in solute molar volume. The saturation capacity of
site 1, the low-energy site, is nearly independent of pressure, which
is expected because the eluent–C18 layer interface remains largely
unchanged with pressure [42]; this seems reasonable considering
that the interfacial area between the mobile and stationary phases
should not change substantially at the studied pressures. The satu-
ration capacity of site 2, the high-energy site, was found to increase
20–25% with pressure for SNS and BTEAC (the increase of 90% for AP
is doubtful and may be due to uncertainty in the nonlinear regres-
sion). Molecular simulation has indicated an increased dewetting
of the C18 stationary phase when pressure is increased from 1
to 1000 bar [42]. Dewetting can make sites previously occupied
by water molecules available [43], thereby increase the saturation
capacity of such sites. This could be the reason why we observe
the increase in saturation capacity for the high-energy site. These
trends are also in agreement with those previously reported for
inulin in RPLC [27].
The Henry’s constants, shown in Fig. 5, increase with pressure for
both types of adsorption site, the contribution from the high-energy
site being larger for the ionic solutes SNS and BTEAC. The change in
solute volume upon adsorption for each type of site can be calcu-
lated using Eq. (1) and the result is shown in Table 2. The R2 values
for Eq. (1) were 0.971, 0.994, and 0.988 for the low-energy adsorp-
tion sites and 0.982, 0.956, and 0.967 for the high-energy sites
for AP, BTEAC, and SNS, respectively. These R2 values are satisfac-
tory because the individual Henry’s constants were obtained from
the bi-Langmuirian fit at each pressure. A large negative change in
solute molar volume, V, indicates that pressure has a large effect
on the retention. From Table 2 it is evident that all three solutes have
Vlow ≈ − 6 cm3 /mol. SNS and BTEAC have almost twice as large
volume change when adsorbing on the high-energy sites, while
AP has a 3.6 times larger volume change. The V for each site is
103
compared to the value obtained directly from the retention factor,
which is a kind of average of all available sites.
To understand the difference in V between the low- and high-
energy sites, consider the loss of solute solvation layer during
adsorption on the two sites. A polar or ionic solute will have a sur-
rounding polar solvation layer when in the mobile phase and will
lose parts of this layer when entering the hydrophobic stationary
phase. However, during adsorption at the eluent–C18 layer inter-
face, much less of the solvation layer is lost than during adsorption
deeper in the C18 layer, where the solute and its solvation layer
are completely surrounded by the C18 chains. This is schematically
described in Fig. 6 for a particular solute.
4.3. Column loading capacity
So far we have seen that the monolayer saturation capacity
increases with increasing pressure. The question is how this affects
the peak width of an eluted compound. One way to study this
is to apply McCalley’s [38] approach in which column efficiency
as a function of sample load is plotted and the sample load at
50% efficiency is taken as a measure of column loading capacity.
Experimentally, it is difficult to conduct experiments studying only
pressure in the most realistic situation in which the pressure drops
over the column because of frictional heating and variations in
efficiency with flow rate. Here we employ an in silico approach,
using the experimentally obtained adsorption isotherms to sim-
ulate elution profiles with a pressure drop over the column, no
frictional heating, and a constant flow rate. An empirical equa-
tion for the pressure-dependent adsorption isotherms was used to
obtain the best possible fit to the experimental data. Each parame-
ter of the bi-Langmuir model, Eq. (2), was taken as a linear function
of pressure, which was fitted to the ECP data simultaneously at all
pressures. The set of numerical constants for each solute is shown
in Supplementary Data Table S2. Elution profiles for 10 sample
loads at two pressure drops over the column, i.e., 150–50 bar and
1000–50 bar, were calculated for the three solutes. The efficiency
was determined using the moment method, which is correct for
non-symmetrical peaks as well [21]. We also acknowledge that
the efficiency calculated from overloaded elution zones is not cor-
rect. One can regard the estimated column loading capacity as a
rough measurement rather than an exact value; however, it is good
enough for comparison between separations conducted at low and
high pressures.
The elution profiles for SNS are shown in Fig. 7 for loads between
0.01 nmol and 20 nmol, while those for BTEAC and AP can be found
in Supplementary Data Figs. S5 and S6 . Comparing the efficiency
with column load, it is evident that the effect of pressure is very
low (see Fig. 8). This was observed for all three solutes, although
the loading capacity was lower for the ionic solutes, as expected
from the saturation capacities in Table 1. It is the low-energy site
which has the highest saturation capacity and it was about 10 times
higher for AP compared to the ionic SNS and BTEAC. From Fig. 8,
this can be related to the lower loading capacity (about a factor 10)
observed for SNS and BTEAC. The reason why BTEAC have a lower
efficiency at low loads was investigated in a previous study [5] and
was believed to be due to kinetic effects. Only three solutes, all with
molecular weights of approximately 200 g/mol, were investigated,
so no general conclusions can be drawn from these results. The
results nevertheless indicate that the pressure drop itself should
be of minor importance for the loading capacity of low-molecular-
weight compounds in RPLC. However, because of the difference
between ionic and neutral solutes, when the mobile phase pH is
near the pKa of the solute, the loading capacity may be affected
by pressure because the pKa can change with pressure [4]. It also
104 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106

Fig. 5. The Henry’s constants (H) from the bi-Langmuir models for the low-energy adsorption site (gray) and the high-energy adsorption site (black) at five different pressures
obtained using restriction capillaries after the column.

Fig. 6. Schematic of adsorption from the mobile phase on the stationary phase for the low-energy site at the eluent-C18 layer interface (left) and high-energy site near the
surface of the silica (right) for a particular solute. The increase of V for the high-energy adsorption site is explained by the larger loss of solvation layer when the solute
penetrates deeper into the C18 layer.

remains to be investigated how the loading capacities of larger
molecules, such as proteins, are affected by pressure.
5. Conclusions
To deepen our understanding of how the adsorption mech-
anism in RPLC is affected by elevated pressures, the adsorption
isotherms of three low-molecular-weight compounds were deter-
mined at pressure plateaus between 100 and 1000 bar. The
adsorption isotherms for the polar uncharged antipyrine, the
positively charged benzyltriethylammonium chloride, and the neg-
atively charged sodium 2-naphthalenesulfonate had bimodal AEDs
and were best described by a bi-Langmuir model. The AED was
bimodal irrespective of pressure and only the numerical parame-
ters of the bi-Langmuir model changed, not the adsorption model
per se. The pressure dependence of the Henry’s constant for each
adsorption site was determined and the change in solute molar
volume between the mobile and stationary phases was calculated
for each type of adsorption site. Weak interactions occurring at the
low-energy sites are related to adsorption in the interface region
between the mobile phase and C18 layer, while strong interactions
occur at the high-energy sites located nearer the silica surface,
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106
a)
b)
c)
Fig. 7. Simulated chromatograms for 1-␮L injections of SNS with a) 0.01 nmol, b)
0.1, 0.5 and 1 nmol, and c) 2, 5, 10 and 20 nmol loads at two linear pressure gradients
over the column.
Fig. 8. Column efficiency versus column loading for AP, BTEAC and SNS from cal-
culated elution profiles with an injection volume of 1 ␮L at two linear pressure
gradients over the column.
deeper in the C18 layer [28,40,41]. For the two ionic solutes, the
change in solute molar volume upon adsorption was approxi-
mately twice as high when adsorbing on the high-energy sites and
3.6 times higher for antipyrine. This is believed to be due to the
greater loss of solvation layer during adsorption on high-energy
sites because of the need to penetrate deeper into the C18 layer.
105
The association equilibrium constant increased slightly with pres-
sure for both sites, which agrees with Le Chatelier’s principle and
previous results for insulin [27]. The saturation capacity for the low-
energy sites was nearly independent of pressure while it increased
for the high-energy sites. This increase could depend on the dewet-
ting of the stationary phase at high pressure [42], which made
available adsorption sites previously occupied by water molecules.
Through simulated elution profiles at two different pressure gradi-
ents over the column, the pure pressure dependence of the column
loading capacity was assessed assuming no frictional heating. For
the solutes studied here, the column loading capacity, defined as
the dimidiation of the column efficiency [38], did not change when
the pressure gradient increased from 50 to 950 bar. The difference
in saturation capacity found through the adsorption isotherm mod-
elling between the ionic and neutral species, agrees well with the
difference in observed loading capacity.
Acknowledgements
This work was supported by the Swedish Knowledge Founda-
tion for the project “SOMI: Studies of Molecular Interactions for
Quality Assurance, Bio-Specific Measurement & Reliable Super-
critical Purification” (grant number 20140179), by the Swedish
Research Council for the project “Fundamental Studies on Molecu-
lar Interactions aimed at Preparative Separations and Biospecific
Measurements” (grant number 2015-04627) and by The ÅForsk
Foundation for the project “Improved Purification Procedures to
Satisfy Modern Drug Quality Assurance and Environmental Crite-
ria” (grant number 15/497).
We are also grateful to Joakim Högblom and Per Jageland at
AkzoNobel PPC, Bohus, Sweden for the gift of the column and for
their excellent technical support.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.chroma.2016.06.
036.
References
[1] S. Fekete, J.-L. Veuthey, D. Guillarme, Comparison of the most recent
chromatographic approaches applied for fast and high resolution separations:
theory and practice, J. Chromatogr. A. 1408 (2015) 1–14, http://dx.doi.org/10.
1016/j.chroma.2015.07.014.
[2] J. De Vos, K. Broeckhoven, S. Eeltink, Advances in ultrahigh-pressure liquid
chromatography technology and system design, Anal. Chem. 88 (2016)
262–278, http://dx.doi.org/10.1021/acs.analchem.5b04381.
[3] M.W. Dong, K. Zhang, Ultra-high-pressure liquid chromatography (UHPLC) in
method development, TrAC 63 (2014) 21–30, http://dx.doi.org/10.1016/j.trac.
2014.06.019.
[4] D.V. McCalley, The impact of pressure and frictional heating on retention,
selectivity and efficiency in ultra-high-pressure liquid chromatography, TrAC
63 (2014) 31–43, http://dx.doi.org/10.1016/j.trac.2014.06.024.
[5] D. Åsberg, J. Samuelsson, M. Leśko, A. Cavazzini, K. Kaczmarski, T. Fornstedt,
Method transfer from high-pressure liquid chromatography to
ultra-high-pressure liquid chromatography. II. Temperature and pressure
effects, J. Chromatogr. A. 1401 (2015) 52–59, http://dx.doi.org/10.1016/j.
chroma.2015.05.002.
[6] S. Fekete, D. Guillarme, Estimation of pressure-, temperature- and frictional
heating-related effects on proteins’ retention under ultra-high-pressure
liquid chromatographic conditions, J. Chromatogr. A. 1393 (2015) 73–80,
http://dx.doi.org/10.1016/j.chroma.2015.03.023.
[7] F. Gritti, G. Guiochon, Adsorption mechanisms and effect of temperature in
reversed-Phase liquid chromatography. Meaning of the classical van’t hoff
plot in chromatography, Anal. Chem. 78 (2006) 4642–4653, http://dx.doi.org/
10.1021/ac0602017.
[8] F. Gritti, G. Guiochon, Heat exchanges in fast, high-performance liquid
chromatography. a complete thermodynamic study, Anal. Chem. 80 (2008)
6488–6499, http://dx.doi.org/10.1021/ac8003902.
[9] K. Kaczmarski, F. Gritti, J. Kostka, G. Guiochon, Modeling of thermal processes
in high pressure liquid chromatography: II. Thermal heterogeneity at very
106 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106
high pressures, J. Chromatogr. A. 1216 (2009) 6575–6586, http://dx.doi.org/
10.1016/j.chroma.2009.07.049.
[10] S. Fekete, J. Fekete, D. Guillarme, Estimation of the effects of longitudinal
temperature gradients caused by frictional heating on the solute retention
using fully porous and superficially porous sub-2 ␮m materials, J. Chromatogr.
A. 1359 (2014) 124–130, http://dx.doi.org/10.1016/j.chroma.2014.07.030.
[11] J.E. MacNair, K.D. Patel, J.W. Jorgenson, Ultrahigh-Pressure reversed-phase
capillary liquid chromatography: isocratic and gradient elution using
columns packed with 1.0-␮m particles, Anal. Chem. 71 (1999) 700–708,
http://dx.doi.org/10.1021/ac9807013.
[12] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Investigation of the effect
of pressure on retention of small molecules using reversed-phase
ultra-high-pressure liquid chromatography, J. Chromatogr. A. 1209 (2008)
195–205, http://dx.doi.org/10.1016/j.chroma.2008.09.021.
[13] M.M. Fallas, U.D. Neue, M.R. Hadley, D.V. McCalley, Further investigations of
the effect of pressure on retention in ultra-high-pressure liquid
chromatography, J. Chromatogr. A. 1217 (2010) 276–284, http://dx.doi.org/
10.1016/j.chroma.2009.11.041.
[14] M.M. Fallas, N. Tanaka, S.M.C. Buckenmaier, D.V. McCalley, Influence of phase
type and solute structure on changes in retention with pressure in
reversed-phase high performance liquid chromatography, J. Chromatogr. A.
1297 (2013) 37–45, http://dx.doi.org/10.1016/j.chroma.2013.04.006.
[15] M.R. Euerby, M.A. James, P. Petersson, The influence of stationary phase on
pressure-induced retention, selectivity and resolution changes in RP-LC, Anal
Bioanal. Chem. 405 (2013) 5557–5569, http://dx.doi.org/10.1007/s00216-
013-6973-3.
[16] A. Makarov, R. LoBrutto, P. Karpinski, Y. Kazakevich, C. Christodoulatos, A.K.
Ganguly, Investigation of the effect of pressure and liophilic mobile phase
additives on retention of small molecules and proteins using reversed-phase
ultrahigh pressure liquid chromatography, J. Liq. Chromatogr. Related
Technol. 35 (2012) 407–427, http://dx.doi.org/10.1080/10826076.2011.
601494.
[17] S. Fekete, J.-L. Veuthey, D.V. McCalley, D. Guillarme, The effect of pressure and
mobile phase velocity on the retention properties of small analytes and large
biomolecules in ultra-high pressure liquid chromatography, J. Chromatogr. A.
1270 (2012) 127–138, http://dx.doi.org/10.1016/j.chroma.2012.10.056.
[18] V.L. McGuffin, S.-H. Chen, Molar enthalpy and molar volume of methylene
and benzene homologues in reversed-phase liquid chromatography, J.
Chromatogr. A. 762 (1997) 35–46, http://dx.doi.org/10.1016/S0021-
9673(96)00958-2.
[19] M. Martin, G. Guiochon, Effects of high pressure in liquid chromatography, J.
Chromatogr. A. 1090 (2005) 16–38, http://dx.doi.org/10.1016/j.chroma.2005.
06.005.
[20] F. Gritti, M. Martin, G. Guiochon, Influence of pressure on the properties of
chromatographic columns: II. The column hold-up volume, J. Chromatogr. A.
1070 (2005) 13–22, http://dx.doi.org/10.1016/j.chroma.2005.02.008.
[21] G. Guiochon, D.G. Shirazi, A. Felinger, A.M. Katti, Fundamentals of Preparative
and Nonlinear Chromatography, 2nd ed., Academic Press, Boston, MA, 2006.
[22] F. Gritti, G. Guiochon, Critical contribution of nonlinear chromatography to
the understanding of retention mechanism in reversed-phase liquid
chromatography, J. Chromatogr. A. 1099 (2005) 1–42, http://dx.doi.org/10.
1016/j.chroma.2005.09.082.
[23] J. Samuelsson, R. Arnell, T. Fornstedt, Potential of adsorption isotherm
measurements for closer elucidating of binding in chiral liquid
chromatographic phase systems, J. Sep. Sci. 32 (2009) 1491–1506, http://dx.
doi.org/10.1002/jssc.200900165.
[24] D. Åsberg, M. Leśko, J. Samuelsson, K. Kaczmarski, T. Fornstedt, Method
transfer from high-pressure liquid chromatography to ultra-high-pressure
liquid chromatography. I. A thermodynamic perspective, J. Chromatogr. A.
1362 (2014) 206–217, http://dx.doi.org/10.1016/j.chroma.2014.08.051.
[25] F. Gritti, G. Guiochon, Effect of the pH, the concentration and the nature of the
buffer on the adsorption mechanism of an ionic compound in reversed-phase
liquid chromatography: II. Analytical and overloaded band profiles on
Symmetry-C18 and Xterra-C18, J.Chromatogr. A. 1041 (2004) 63–75, http://
dx.doi.org/10.1016/j.chroma.2004.05.004.
[26] F. Gritti, G. Guiochon, Retention of ionizable compounds in reversed-phase
liquid chromatography. effect of the ionic strength of the mobile phase and
the nature of the salts used on the overloading behavior, Anal. Chem. 76
(2004) 4779–4789, http://dx.doi.org/10.1021/ac0304121.
[27] X. Liu, P. Szabelski, K. Kaczmarski, D. Zhou, G. Guiochon, Influence of pressure
on the chromatographic behavior of insulin variants under nonlinear
conditions, J. Chromatogr. A. 988 (2003) 205–218, http://dx.doi.org/10.1016/
S0021-9673(03)00002-5.
[28] F. Gritti, G. Guiochon, Heterogeneity of the adsorption mechanism of low
molecular weight compounds in reversed-phase liquid chromatography,
Anal. Chem. 78 (2006) 5823–5834, http://dx.doi.org/10.1021/ac060392d.
[29] J. Lindholm, P. Forssén, T. Fornstedt, Validation of the accuracy of the
perturbation peak method for determination of single and binary adsorption
isotherm parameters in LC, Anal. Chem. 76 (2004) 4856–4865, http://dx.doi.
org/10.1021/ac0497407.
[30] J. Samuelsson, T. Undin, A. Törncrona, T. Fornstedt, Improvement in the
generation of adsorption isotherm data in the elution by characteristic points
method—the ECP-slope approach, J. Chromatogr. A. 1217 (2010) 7215–7221,
http://dx.doi.org/10.1016/j.chroma.2010.09.004.
[31] ACQUITY UPLC System, Operator’s Guide, (71500082502/Revision E).
[32] B.J. Stanley, G. Guiochon, Numerical estimation of adsorption energy
distributions from adsorption isotherm data with the
expectation-maximization method, J. Phys. Chem. 97 (1993) 8098–8104,
http://dx.doi.org/10.1021/j100132a046.
[33] J. Samuelsson, T. Fornstedt, Calculations of the energy distribution from
perturbation peak data–A new tool for characterization of chromatographic
phases, J. Chromatogr. A 1203 (2008) 177–184, http://dx.doi.org/10.1016/j.
chroma.2008.07.045.
[34] K. Kaczmarski, M. Mazzotti, G. Storti, M. Mobidelli, Modeling fixed-bed
adsorption columns through orthogonal collocations on moving finite
elements, Comput. Chem. Eng. 21 (1997) 641–660, http://dx.doi.org/10.1016/
S0098-1354(96)00300-6.
[35] P. Brown, G. Byrne, A. Hindmarsh, VODE: a variable-coefficient ODE solver,
SIAM J. Sci. Stat. Comput. 10 (1989) 1038–1051, http://dx.doi.org/10.1137/
0910062.
[36] D. Åsberg, M. Leśko, M. Enmark, J. Samuelsson, K. Kaczmarski, T. Fornstedt,
Fast estimation of adsorption isotherm parameters in gradient elution
preparative liquid chromatography. I: The single component case, J.
Chromatogr. A. 1299 (2013) 64–70, http://dx.doi.org/10.1016/j.chroma.2013.
05.041.
[37] D. Åsberg, M. Leśko, M. Enmark, J. Samuelsson, K. Kaczmarski, T. Fornstedt,
Fast estimation of adsorption isotherm parameters in gradient elution
preparative liquid chromatography II: The competitive case, J. Chromatogr. A.
1314 (2013) 70–76, http://dx.doi.org/10.1016/j.chroma.2013.09.003.
[38] D.V. McCalley, Overload for ionized solutes in reversed-phase
high-performance liquid chromatography, Anal. Chem. 78 (2006) 2532–2538,
http://dx.doi.org/10.1021/ac052098b.
[39] X. Zhang, J. Samuelsson, J.-C. Janson, C. Wang, Z. Su, M. Gu, T. Fornstedt,
Investigation of the adsorption behavior of glycine peptides on 12%
cross-linked agarose gel media, J. Chromatogr. A. 1217 (2010) 1916–1925,
http://dx.doi.org/10.1016/j.chroma.2010.01.058.
[40] J.L. Rafferty, L. Zhang, J.I. Siepmann, M.R. Schure, Retention mechanism in
reversed-phase liquid chromatography: a molecular perspective, Anal. Chem.
79 (2007) 6551–6558, http://dx.doi.org/10.1021/ac0705115.
[41] F. Gritti, G. Guiochon, Adsorption mechanism in RPLC. Effect of the nature of
the organic modifier, Anal. Chem. 77 (2005) 4257–4272, http://dx.doi.org/10.
1021/ac0580058.
[42] J.L. Rafferty, J.I. Siepmann, M.R. Schure, The effects of chain length, embedded
polar groups, pressure, and pore shape on structure and retention in
reversed-phase liquid chromatography: molecular-level insights from Monte
Carlo simulations, J.Chromatogr. A. 1216 (2009) 2320–2331, http://dx.doi.org/
10.1016/j.chroma.2008.12.088.
[43] F. Chan, L.S. Yeung, R. LoBrutto, Y.V. Kazakevich, Interpretation of the excess
adsorption isotherms of organic eluent components on the surface of
reversed-phase phenyl modified adsorbents, J. Chromatogr. A. 1082 (2005)
158–165, http://dx.doi.org/10.1016/j.chroma.2005.05.078.
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106 1

Supplementary data
A fundamental study of the impact of pressure on the adsorption
mechanism in reversed-phase liquid chromatography
Dennis Åsberg, Jörgen Samuelsson and Torgny Fornstedt

S1. AED and adsorption isotherm measurements

Fig. S2: Characterization of butyltriethylammonium chloride on Eternity-C18 at 0.20 mL/min and 30°C
from perturbation pulse experiments. a) Scatchard plot from integrated raw data, b) AED calculation
using the EM-method with 105 iterations, 500 grid points in energy space and b in L/mol, c) Best fit to the
adsorption isotherm derivative (symbols) using the differentiated bi-Langmuir model (line) and d)
Calculated elution profiles compared with experimental elution profiles for 10 μL injections of 4.4 and
22.0 mM butyltriethylammonium chloride. The area overlap is 93.4 and 93.0% between experimental and
calculated elution profiles for 4.4 and 22.0 mM butyltriethylammonium chloride, respectively.
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106 2

Fig. S3: Characterization of sodium-2-naphthalenesulfonic acid on Eternity-C18 at 0.20 mL/min and

30°C from perturbation pulse experiments. a) Scatchard plot from integrated raw data, b) AED calculation
using the EM-method with 105 iterations, 500 grid points in energy space and b in L/mol. c) Best fit to the
adsorption isotherm derivative (symbols) using the differentiated bi-Langmuir model (line) and d)
Calculated elution profiles compared with experimental elution profiles for 10 μL injections of 3.04 and
30.4 mM sodium-2-naphthalenesulfonicacid. The area overlap is 89.5% and 96.8% between experimental
and calculated elution profiles for 3.04 and 30.4 mM.

Table S1: Estimated isotherm model parameters from the perturbation peak data for antipyrine (AP),
benzyltriethylammonium chloride (BTEAC) and sodium 2-naphthalenesulfonate (SNS) on the Eternity-
C18 column. RSS is the residual square sum and Fischer is the Fischer number.
Compound RSS Fischer qs,1 [mM] b1 [L/mmol] qs,2 [mM] b2 [L/mmol]
AP 0.2583 4563 410 0.0172 80.1 0.202
BTEAC 1.3056 433 85.6 0.0494 7.16 2.43
SNS 0.7705 861 171 0.0229 7.39 2.23
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106 3

S2. Column overload

From the raw slope adsorption data determined with the ECP method at different pressures, a function
dq/dC = f(C,P) was determined with nonlinear least square regression. The function was on the form

∂q(C , P ) b1 ( P ) b2 ( P )
= qs,1 ( P ) + qs,2 ( P )
∂C (1 + b ( P )C )2
1 (1 + b ( P )C )2 2
P ) α1 P + α 2
qs,1 (=
P ) α3 P + α4
b1 (= (S1)
P ) α5 P + α6
qs,2 (=
P ) α7 P + α8
b2 (=

This type of function could model to data accurately and converged to a solution when starting guesses
from linear regression of the adsorption isotherms from the pressures plateaus, Table 1, was used.
Numerical parameters are presented in Table S2.

Table 2: Numerical parameters to describe the pressure dependence of the adsorption isotherm, Eq. (S1).
α1 α2 α3 α4 α5 α6 α7 α8
[mM/bar] [mM] [L/mM/bar] [L/mM] [mM/bar] [mM] [L/mM/bar] [L/mM]
AP -8.79E-03 1.96E+02 3.86E-05 1.07E-01 -5.65E-04 2.42E+00 9.24E-04 9.66E-01
SNS -7.26E-05 2.75E+01 1.06E-04 3.04E-01 2.99E-04 3.07E+00 1.05E-03 4.14E+00
BTEAC -4.65E-04 2.90E+00 2.39E-03 4.29E+00 -2.20E-03 2.49E+01 2.85E-04 3.81E-01

Fig. S4: Adsorption isotherm of sodium-2-naphthalenesulfonic acid on Eternity-C18 as a continuous

function of pressure calculated from Eq. (S1).
 D. Åsberg et al. / J. Chromatogr. A 1457 (2016) 97–106 4

Fig. S5: AP simulated at different loads. Fig. S6: BTEAC simulated at different loads.
Fundamental and Regulatory Aspects of
UHPLC in Pharmaceutical Analysis
Ultra-high performance liquid chromatography (UHPLC) provides a considerable
increase in throughput compared to conventional HPLC and a reduced solvent
consumption. The implementation of UHPLC in pharmaceutical analysis has
accelerated in recent years and currently both instruments are used. There are,
however, technical and regulatory challenges converting a HPLC method to
UHPLC making it difficult to take full advantage of UHPLC in regulatory-
focused applications like quality control in pharmaceutical production. In
UHPLC, the column is packed with smaller particles than in HPLC resulting
in higher pressure and viscous heating. Both the higher pressure and the higher
temperature may cause changes in retention and selectivity making method
conversion unpredictable.

Using chromatographic modelling and fundamental theory, this thesis

investigates method conversion between HPLC and UHPLC. It reports on the
influence of temperature gradients due to viscous heating, pressure effects and
stationary phase properties on the separation performance. It also presents a
regulatory concept for less regulatory interaction for minor changes to approved
quality control methods and how predicable method conversion is achieved by
improved understanding.

ISBN 978-91-7063-756-8 (print)

ISBN 978-91-7063-757-5 (pdf)

ISSN 1403-8099

DOCTORAL THESIS | Karlstad University Studies | 2017:9