Chapter 1
Introduction: The First Steps of Method Development in Liquid Chromatography
Imre MolnĂĄrâ,⥠and Szabolcs Feketeâ
âMolnĂĄr-Institute, Institute for Applied Chromatography,
Schneeglöckchenstrasse 47, D-10407 Berlin, Germany
â School of Pharmaceutical Sciences, University of Geneva,
University of Lausanne, CMU â Rue Michel Servet 1,
1211 Geneva 4, Switzerland
1.1Introduction
Modeling in high-performance liquid chromatography (HPLC) was started in 1975 by Csaba HorvĂĄth at Yale University, Haven, CT, USA. He purchased a PDP-11 computer and studied the theory of band spreading in HPLC with this computer [1]. The fundamentals of reversed-phase liquid chromatography (RPLC) were established based on compounds of relevance to the field of life science, such as catecholamines and their derivatives and metabolites [2]. The analysis time for 100 organic acids could be reduced from 48 h to less than 30 min using RPLC [3]. A few months later, the separation of amino acids and peptides was achieved first time on an octadecyl (C18) stationary phase [4]. The basic relationships were investigated and systematic work was carried out, typically one-factor-at-a-time (OFAT), to be able to understand the reason for peak movements in a chromatogram during the optimization process. The observed relationships created a new theory of solvophobic interactions that was reported in a series of papers which explained the significant differences observed between the solute retentions observed in water-rich and organic-modifier-rich mobile phases [5â7]. The power of water to enable retention on a C18 stationary phase â due to its high surface tension â can be reduced by adding methanol (MeOH) or acetonitrile (ACN), as is usually done in RP gradient elution. These generic gradients have the advantage of being able to elute almost any organic compound by a continuous increase in the content of the organic modifier in the aqueous mobile phase, thus leading to a new culture of gradient elution in HPLC.
In the 1980s, Snyder and Kirkland studied column properties by measuring ca. 1000 columns at DuPont to understand how solute diffusion in the pores influences band spreading. They included the van Deemter and the Knox equations into their models. In 1985 Lloyd Snyder (LC Resources) and Imre MolnĂĄr (MolnĂĄr-Institute) began to develop a software to assist method development for HPLC separations. The software was first an isocratic variant which could calculate tables of retention factor (k) ranges and resolution (Rs) values based only on a few experiments and could predict the optimal mobile phase composition to achieve the highest resolution between the compounds in the shortest possible time. Snyder named the software âDryLab Iâ (I for isocratic) [8]. It could set retention limits for a separation (1 < k < 10) and also introduced the principle of âequal band spacingâ (EBS) based on the resolution of the least well-separated peak pair, called the âcritical peak pairâ (Rs,crit). If the critical peak pair is separated with baseline resolution (Rs > 1.5), it can be presumed that all other peak pairs will also be atleast baseline separated.
A short time later, the development of a software version working in gradient elution mode was created, based on the theory developed by Snyder, under the name DryLab G (G for gradient) [9, 10]. Here, the initial and final mobile phase composition could be varied â with up to 10 gradient steps â and the influence of column dimensions (L, ID, dp), flowrate and instrument factors (dwell volume, extra-column volume) could also be considered. It was the first attempt to calculate multifactorial influences on chromatographic separation in gradient elution.
In 1989, Snyder and Joseph Glajch provided an impressive collection of contributions from experienced method development professionals regarding their work. A set of 43 papers was the final result, with a premium selection of articles by well-recognized authors like Berridge, Deming, Billiet, Galan, Snyder, Dolan, Jandera, Lankmayer, Schoenmakers, Massart, ValkĂł, Jinno and others who contributed important research to this field [8].
Unlike the initial interest on band spreading, it was only somewhat later that other variables like mobile phase pH, temperature, buffer composition, ternary eluent composition, ion-pair concentration, etc., were added to be able to study the influence of multifactorial changes on the position of peaks in a chromatogram, and even later that they were commercially available under the name DryLab Imp (Isocratic multiparameter version). The impact of column length, inner diameter, particle size, flow rate, system dwell volume and extra-column volume on the separation could also be calculated. DryLabÂź was therefore a multifactorial tool even during 1988â1990 in separation modeling and helped to understand how retention is changing and how to control selectivity changes of moving peaks to be able to comply with regulatory expectations.
To study the influence of two measured variables at the same time, 2D models were developed. These 2D resolution maps showed the separation of the critical peaks in the chromatogram as a function of two simultaneously changed experimental variables. The gradient time (tG) and the mobile phase temperature (T) were chosen as the two most relevant variables (tG-T-model) as a simple method for peak tracking. The concept of a âmethod operable design regionâ (MODR) or âdesign spaceâ (DS) was laid down for HPLC here the first time. A more detailed summary of the contributions is compiled in Ref. [11].
After the great success of the tG-T-model, advocated by Snyder and Dolan, they introduced this concept in 2000 for column characterization [12].
It was in 2009 that it first became possible to study three measured variables at the same time and calculate the influence of an additional six/eight other parameters, such as flow-rate, column dimensions, instrument parameters and gradient conditions, in the so-called âcubeâ. Launching this 3D critical resolution map [13], new avenues were opened not only in method development but also in robustness testing. Several HPLC instruments (e.g. Waters, Shimadzu, Thermo) can be controlled today with the DryLabÂź software for the automated processing of the necessary experiments to enable quicker separations in an automated fashion. Nowadays, experiments that require build-up of four resolution cubes (on four columns providing different selectivity) can be performed in one single day as a complete method screening and optimization protocol [14].
The latest version of the software (DryLabÂź4, version 4.3) enables âdryâ modeled robustness testing. From the DS, as defined in a 3D resolution map, it is possible to obtain robustness information for the measured parameters, including tG, T, tC (ternary composition) and mobile phase pH. In addition, based on the models included in the software, the retention time of any compound can be calculated to account for the influence of additional variables, such as flow rate or initial and final mobile phase composition (expressed in %B) through the gradient. Consequently, the impact of changes in any of these six variables on the resolution can be assessed using simulated 26 or 36 type factorial designs. No additional experiments are necessary for performing the simulated robustness calculation [13]. The possible deviations from the nominal values just need to be defined and then the software makes the calculations for 26 = 64 or 36 = 729 conditions. With two additional gradient points (each gradient point corresponds to two additional variables), the variants sum up to 210 = ca. 59,000 chromatograms, which are calculated and evaluated in less than 1 min. At the end, the software provides a âfrequency distribution graphâ showing how often a certain critical resolution value occurs under any combination of the possible parameters. This graph also shows the failure rate, i.e. number of experiments that could fall outside the required critical resolution in routine work. On the other hand, âregression coefficientsâ can also be obtained to show the effect of each variables, related to the selected deviation from the nominal value, for the critical resolution. The robustness feature allows to reduce the âout of specificationâ (OoS) results.
In this book, the current situation of HPLC retentionâresolution modeling and its applications in the bio- and pharmaceutical industry are reviewed.
1.2Modeling Alternatives
Currently, there are many other commercially available software packages on the market such as DryLabÂź4 modules (PeakTracking, 3D-Cube, Robustness Module, Know-ledge Management Module, Column Comparison Module, MolnĂĄr-Institute, Germany), ChromSwordÂź packages (Developer, AutoRobust, Scout, ChromSword, Latvia), ACD packages (LC simulator, ChromGenius, AutoChrom, ACD/Labs, Canada), Fusion (S-Matrix, USA) and Osiris (Datalys, France). Some of them mainly focus on the quantitation process using statistical approaches to find out if a method is not robust or to support method screening (Fusion). Other packages, like DryLab, explain why a method is not robust and how to change conditions to get back to the validated region (DS).
Some tools start from molecular structure and try to derive an approximation of the retention time at which a molecule would elute from the column (ChromSword). Other tools (e.g. EluEx) use logD, logP and pKa values to approximate the mobile phase composition and pH range for a decent resolution. Other packages offer a mathematical statistics-oriented development approach (Fusion).
Presently, the trend is to look at robust conditions in a multivariable space in different ways. First of all, it is most important to understand peak movements in HPLC separations, which are based on a sufficiently wide range of eluent properties (pH, gradient time and program, flowrate, etc.), before time-consuming experiments based on trial & error can be investigated. Multifactorial modeling simplifies and speeds up the process of developing reliable chromatographic separations by allowing the user to model changes in the separation conditions using a computer. As an example, the analysis time of an old pharmacopeia method was cut down from 160 min to less than 3 min [15]. The most important advantage is, however, the better understanding of the scientific process of the separation and to prove the suitability of the HPLC method for communication of results to the regulatory agencies (FDA, EMA) in order to receive commercial authorization for drugs.
1.3What is the Purpose of Method Development?
Method development in HPLC means the search for the optimal chromatographic operating conditions (type of mobile and stationary phase, temperature, gradient steepness, pH, ionic strength, etc.) resulting in the proper separation of a mixture into its constituents within a given analysis time frame [16]. Because of the high probability for peak overlap and the high dependence of the retention time on the employed chromatographic parameters, the method development process is often tedious and timeconsuming (up to several weeks of work) [17]. It requires the knowledge and expertise of the analyst, but still involves a lot of trial-and-error processes. Computer-assisted method development however has the potential to speed up the process significantly, if adequate retention models exist. The gain in analysis time can be particularly significant for regulated laboratories (pharmaceutical, food and environmental analytical laboratories) to prove that the method is based on solid science. Before beginning method development, the chromatographers need to review what is known about the sample [18]. The goals of the separation should also be defined at this point. The chemical composition of the sample can provide valuable clues for the best choice of initial conditions for an HPLC separation. The analytical target profiles (ATPâs) of the HPLC sep...