The introduction of Quality by Design, a systematic approach to product and process design and development, has shifted the paradigm in HPLC methods development from a retrospective approach to a prospective, systematic, risk-based approach in order to develop enhanced method understanding.1
The basis for Quality by Design of chromatographic methods can be broadly categorized into two areas: knowledge space and design space. The knowledge space encompasses all considerations made, all experiments conducted, and all knowledge gained in the development of a method (i.e., column screening or pH screening experiments). The knowledge space forms the basis for delineating a design space within which one can modify the chromatographic factors (i.e., gradient slope, temperature, and buffer concentration, all at a defined pH) without significantly impacting the final quality of the method and making sure all of the chromatographic figures of merit can still be met (resolution of critical pairs, tailing factor, selectivity, etc.).
There is an inherent need to understand the main effects of critical method factors and their mutual interactions on the critical method attributes (the response variables that are the quality characteristics of the method). Since there are many factors a scientist must consider when developing or validating a robust stability-indicating chromatographic method, statistical experimental design and analysis allow for an efficient and effective means of execution. This will lay the basis for defining the design space boundaries and a control strategy to ensure that quality is built into the method and further ensure the integrity of the results.
In most cases, reversed-phase chromatographic separation can be achieved with an appropriate column selection and evaluation of different variants of mobile phase parameters. The first stage of chromatographic method development is to identify the most suitable column, mobile phase pH, aqueous phase buffer, and organic solvent to separate the active pharmaceutical ingredient (API) and its impurities in a particular sample. After initial starting conditions have been identified, the next stage is to determine optimal and robust separation conditions. This task becomes more complicated as the number of operating variables increases, which leads to a larger number of experimental runs required. To simplify and accelerate the optimization process, computer simulation software packages have been employed.2
Liquid chromatography simulation software such as DryLab® (Molnár Institute, Berlin, Germany), LC-Simulator® (Advanced Chemistry Development, Toronto, Ontario, Canada), and ChromSword® (Merck, Darmstadt, Germany) have been shown to be effective tools in modulating gradient and column temperature during method development. These programs can use a small set of well-defined experimental data on a particular stationary phase at a defined variable such as pH to predict optimal separation based on changes in mobile phase composition and temperature.3 DryLab software uses retention data from scouting runs for subsequent retention and resolution prediction via simulation.3 ChromSword, another optimization software, takes a somewhat different approach, using structure fragments and dipole–dipole interactions to predict retention behavior.4,5 Both of these methods work without any direct connection to the chromatographic apparatus. More sophisticated software utilizes artificial intelligence. An early example is the EluEx (CompuDrug, Budapest, Hungary), which can suggest initial experimental conditions based on chemical structures.5
Many authors reported successful use of chromatography simulation software for method development purposes. A polar neutral compound (ethinylestradiol, EE) and its related substances (6-alpha-hydroxy-EE; 6-betahydroxy-EE; 6-keto-EE; 9,11-didehydro-EE; and estradiol) were separated on a Pinnacle® C18 column (Restek, Bellefonte, PA) (50 × 2.1 mm, dp = 1.9 µm). Two gradients with different slopes (7- and 21-min gradient time) were run at two different column temperatures (35 and 65 °C). The predicted retention times (RT) were in close agreement with the experimental times; the average retention time error was 1.6% (minimum 0.44, maximum 3.09).6 Similar results were observed in the simultaneous optimization of gradient program and mobile phase pH for basic molecules separated on a Zorbax SB C18 column (Agilent Technologies, Santa Clara, CA) (50 × 2.1 mm, dp = 1.8 µm) with methanol and buffer as mobile phase. Two initial gradients with different slopes (7- and 21-min gradient time) were run within a narrow pH range at three different mobile phase pH values (pH 6.2, 6.6, and 7.0) at a constant temperature. The predicted retention times were also in excellent agreement with the experimental times. When mobile phase pH was optimized, the overall average retention time error was under 2% (minimum error was 0.44% and maximum was 8.66%). It was also concluded that the accuracy of computer prediction depends on the applied pressure (flow rate).6
In addition to gradient optimization, many others also reported using the software for mobile phase pH optimization. In order to obtain a 2-D resolution map, three combinations of parameters were considered: gradient time (tG) versus eluent composition of acetonitrile (ACN) and H2O, tG versus column temperature, and eluent composition versus pH.7 Analyte separation was further optimized using DryLab software with experimental data input from two linear gradient runs: tG = 30 min and tG = 60 min; %B run from 5 to 50% at column temperature of 30 °C and aqueous mobile phase pH of 5.5 (analyte pKa of 6.6–9.0). The overall difference between predicted and experimental retention time was 0.2%.7
There have been many success stories utilizing the software to achieve optimal separation. However, within the literature, there are limited discussions on the limitation and capability of the approach.
In one study,8 multiple classes of compounds such as steroids, pesticides, algal pigments, fatty acid methyl esters, and acrylate monomers were chosen to examine the effectiveness of varying column temperature and gradient steepness via DryLab. They achieved optimal separation and good prediction accuracy for most analytes. However, in the case of testosterones, isomers of monohydroxytestosterone were not adequately resolved (maximum resolution <0.5) by any choice of gradient-time or temperature modulation. During the same study, for a group of toxicological compounds in body fluids, accurate retention time prediction of seven basic compounds was not feasible due to early elution.8 It was also mentioned that the extent of extrapolation based on input data should be carefully chosen.