A Brief Guide to...
Implementing Robustness Testing for HPLC Methods
Analytical methods need to be robust so that they can be used routinely without problems and can be easily transferred for use in another laboratory if necessary. The intrinsic robustness of an HPLC method depends on a range of variables related to the method parameters, such as preparation of test solutions, mobile phase composition, column, etc. A definition of robustness is provided by ICH: ‘The robustness of an analytical procedure is a measure of its capacity to remain unaffected by small, but deliberate variations in method parameters and provides an indication of its reliability during normal usage.’ Therefore a study of robustness will involve an investigation into the effect of these small, but deliberate variations.
Ideally robustness should be investigated as part of method development since a method is not complete without an evaluation of its reliability in routine use, but unfortunately it is often deferred, or completely overlooked, due to the time consuming nature of the study. When the method is being validated, a robustness study may then be carried out. In the ICH validation guidelines, robustness is not included in the tabular summary of required characteristics which should be tested during validation, which could lead to the mistaken belief that a study of robustness is not required. However, in the section of the guidelines relating to robustness, it is expected that ‘the evaluation of robustness should be considered during the development phase and depends on the type of procedure under study. It should show the reliability of an analysis with respect to deliberate variations in method parameters.’ So not only is robustness good practice in terms of developing fit for purpose analytical methods, it is also a regulatory requirement.
The emergence of the use of Quality by Design (QbD) principles in pharmaceutical manufacturing has led to the application of QbD to analytical methods. This in turn has highlighted the importance of robustness during method development since the design space concept of QbD translates into knowledge about the effect of each method parameter on the final analytical result. For many analytical chemists this means that the robustness of a method may now be considered during method development.
Selection of the robustness factors
Whether robustness is investigated as part of the development of the method, at the end of method development during optimisation, or after method development during validation, the method parameters which will be investigated remain the same, e.g. % organic in the mobile phase, pH of the mobile phase, etc. These are referred to as the robustness factors and the first step of robustness testing is to decide which factors are going to be studied. The number of factors selected determines how much information is gathered about the way in which changes in method parameters affect the results, but with increasing numbers of factors the time required to complete the investigation also increases.
Selection of the factor levels
Once a decision has been made regarding the method parameters, or factors, which will be investigated, the levels of the factors need to be defined. For example, if the percentage of acetonitrile in a mobile phase is the factor which will be deliberately varied, by how much should it be varied? The amount of variation is referred to as the factor level, typically limits around a nominal value are investigated and the magnitude of these limits has to be defined. When using a QbD approach to method development the aim is to understand the effect of changing a method parameter, whereas when determining the robustness of a method during validation the aim is to examine the variation which might be expected in routine use of the method. Thus the limits for the former may be expected to be wider than in the latter scenario.
Selection of the method responses
The factors and factor levels define the way in which the method (and thus HPLC system) will be set up for robustness experiments, but a way of measuring the effect of the method variations is also required. This needs to take into account quantitative aspects of the method such as an assay result, e.g. %w/w for main components or impurities, but also chromatographic criteria such as resolution of a critical pair or tailing of a major peak. These are referred to as the method responses.
Selection of experiments
There are two main ways in which the robustness experiments may be performed. Using a one factor at a time (OFAT) approach, each factor is varied without changing any of the other method parameters and thus the effect of the factor can be observed. An alternative approach is to use experimental design where factors are investigated simultaneously, the advantage being significant saving in the time required for the study. However, it is more difficult to interpret data obtained from experimental design compared to OFAT. In practice, a combination of both approaches may be beneficial.
Execution of experiments
Performing the robustness experiments is the time consuming step. The method parameters need to be set up correctly for each experiment using suitable test solutions and sequences of injections. A random order is advised when using experimental design but this may involve changeover between mobile phase components and columns. The order of experiments requires careful planning.
Interpretation of results
Whether the data is obtained by an OFAT approach or by experimental design, conclusions need to be derived about what it all means. Statistical and graphical methods may be used to aid interpretation. Factors which affect the method responses may need to be controlled. Robustness data enables the development of a method specific system suitability test.
The procedure described above for robustness testing is summarised below in Figure 1.
Figure 1 Summary of the procedure for robustness testing
Composition of the mobile phase
During the development of HPLC methods the composition of the mobile phase is modified to separate and control the retention times of the peaks of interest. In the case of reversed phase HPLC the volume fraction of the organic solvent in the mobile phase is related to the retention factor; increasing the proportion of organic solvent has the effect of decreasing the retention time. The ‘rule of 3’ (which applies to low molecular weight molecules analysed by reversed phase HPLC) allows for an approximate prediction of the effect of changes in the proportion of organic solvent in the mobile phase. A change of 10% in the organic solvent content results in the retention time changing by a factor of approximately 3. The implication of this for robustness testing is that composition of the mobile phase is a very important factor which should therefore be included in the study.
The robustness factor to be investigated for isocratic separations is the volume fraction of the organic solvent in the mobile phase. This can be expressed simply as %B. For gradient methods, the method parameter which is analogous to %B in isocratic mode is that of gradient time, or tG, and thus may be selected as a robustness factor. However, this parameter will not be subject to variation when the method is in routine use. An alternative approach for gradient methods is to choose two factors for investigation which will be subject to variation, namely the volume fraction of the organic solvent in the mobile phase at the start and at the end of the gradient. These may be referred to as %BSTART and %BEND.
Resolution of the critical Pair
The magnitude of the effect of this factor is dependent on the resolution of the critical pair in the separation. If the resolution is low, e.g., only just baseline resolved (refer to the illustrative example in Figure 2a), then it is likely that small increases in the organic content of the mobile phase could result in a loss of resolution and possible co-elution of the critical pair (refer to the illustrative example in Figure 2b). If the critical pair is well separated then it is likely that the separation will be more robust to changes in the amount of organic solvent in the mobile phase.
Figure 2a The critical pair (peak 4 and peak 5) in this separation are only just baseline resolved
Figure 2b The critical pair (peak 4 and peak 5) are no longer resolved due to an increase in %B
Selecting factor levels
Having selected the composition of the mobile phase as a robustness factor, the next step is to select appropriate factor levels. This is usually expressed as an interval between the extremes which represents the potential variability when the method is in routine use. In this case the factor levels will consist of a range above and below the percentage of organic solvent (%B) which is written in the HPLC method. The potential variability depends on the errors which may occur during measurement of the volume whilst preparing the mobile phase.
Estimation of potential variation of %B
An approximation of the worst case scenario for the error is estimated as follows: The volumes required for mobile phase typically require the use of a 1000mL measuring cylinder which has a tolerance of ±2.5 mL at a specified temperature. The gradations in the cylinder are typically 2 mL and so a worst case error measurement may be as much as 4.5 mL. The smallest amount that this measuring cylinder is likely to be used to measure is 550 mL, this equates to a maximum error in volume of about 0.8%. If this error is carried through to the overall volume fraction of the organic solvent in mobile phase then a variation of 1% would seem to represent the maximum variability likely to occur, it appears unlikely that normal variation in preparation of mobile phase would exceed this amount. Thus the interval for the factor levels in the robustness study would be %B ±1%. This interval may also be applied to %BSTART and %BEND for gradient methods. Illustrative examples are shown in Table 1 and Table 2 for an isocratic method and a gradient method respectively.
Table 1 Example of factor levels for the volume fraction of the organic solvent in the mobile phase (%B). The mobile phase for the example method is A: 75% Water & B: 25% Methanol
|Method value||Low value||High value |
Table 2 Example of factor levels for the volume fraction of the organic solvent in the mobile phase at the start (%BSTART) and at the end (%BEND) of the gradient. The mobile phase for the example method is A: 90% phosphate buffer at pH 2.1 & BSTART: 10% Acetonitrile, changing to A: 60% phosphate buffer at pH 2.1 & BEND: 40% Acetonitrile over 20 minutes.
|Method value||Low value||High Value |
It is assumed that the laboratory has standard operating procedures in place that ensure that the mobile phase is always prepared in the same way. Typically, the procedure is summarised as follows: The volume of the aqueous portion of the mobile phase is measured and any buffers are added, then the pH is adjusted as required. The organic solvent is measured in a separate measuring cylinder and the two portions are mixed together thoroughly before being placed on the HPLC system.
Online mixing of the mobile phase
If the HPLC system is used to mix the mobile phase online, using separate vessels containing the components of the mobile phase, then the error is likely to be very much lower. However, it is beneficial to have robustness data relating to premixed mobile phase to allow maximum flexibility when operating the method routinely.
Capability of the method
The variation of %B ±1% should provide a reasonable indication of potential variability of the method in routine use. If the full method capability is of interest then a wider interval may be of interest. In this case the full range of the volume fraction of the organic solvent in the mobile phase over which the results of the method are acceptable is sought. The inherent problem in this type of investigation is the number of experiments, and thus the time taken to explore the acceptable range. Ideally an interval is predicted in advance and then confirmed experimentally. This may be achieved if computer modelling is used for the method development. A virtual model of a separation may be created using the data obtained from a small number of experiments, allowing optimisation of method conditions without having to inject any samples on the HPLC system. Software packages which enable this include ACD/LC Simulator (Advanced Chemistry Development, Inc.) and DryLab® (Molnár-Institute for applied chromatography).
pH Control in Reversed Phase Mobile Phase
The reason that pH control is important in reversed phase HPLC relates to the ionisation of the analytes. The retention mechanism for an ionised, and therefore relatively polar, molecule is different for that of the same molecule in an unionised, and relatively less polar, form. If the pH of the mobile phase is not controlled for an ionisable analyte then the chromatography may exhibit broad or misshapen peaks, and if reasonable peak shape is achieved then it is very likely that the chromatography will be prone to reproducibility problems, including retention time changes, and therefore compromise the overall robustness of the method.
It is common practice to control the pH of the mobile phase for ionisable analytes. However, ionisable interferences in the sample matrix may also give rise to chromatographic problems. If the peak due to a component in the sample matrix is misshapen, broad or has a changing retention time, then this may interfere with analyte peaks of interest, again compromising the robustness of the method. This explains why the presence of a buffer can make a method more robust, but choosing to control the pH leads to other choices, namely what value of pH will work best? And, what buffer should be used?
The Effect of pKa on Retention
Ionisable functional groups within an analyte have an associated pKa value. If the pH of a solution containing the ionisable analyte is held at the pH value which is equal to the pKa then the ionised and unionised forms will be present in equal amounts. In figure 3 the molecule amphetamine is shown. It is in equilibrium between the free base and the ionised molecule. The pKa of amphetamine is 9.8, therefore if it is in solution at a pH of 9.8 there will be equals amounts of each present.
Figure 3 The basic drug, amphetamine is in equilibrium between the free base and the ionised form
The relationship between retention of amphetamine on a RP HPLC column and the pH of the mobile phase is shown in figure 4. At low pH the base is ionised and thus is relatively more polar, compared to the unionised base at high pH. The retention mechanism is predominantly based on hydrophobic interactions and therefore the ionised base at low pH is relatively less retained on the column than the free base at high pH. The selection of buffer pH therefore affects the retention time of the analyte.
Figure 4 The relationship between retention on RP-HPLC column and the pH of the mobile phase for the base, amphetamine
The middle of the steep section of the plot in figure 4 corresponds to when the pH of the buffer equals the pKa of the molecule. Throughout the steep region of pKa ±2 pH units, a small change in pH may result in a large change in retention time. The implication is that if the pH of the mobile phase is not controlled very carefully every time the method is performed, then the retention time may change from run to run, potentially causing problems with resolution if the peaks elutes close to another peak. This effect is demonstrated in figure 5. In figure 5a the pH of the buffer is at 9.5. When the pH is changed to 9.3, the peak due to component 2 has moved to a shorter retention time but the retention times of the other peaks have not changed. Therefore peak 2 is not robust to changes in pH.
Figure 5a The mobile phase pH was adjusted to 9.5.
Figure 5b The mobile phase pH was adjusted to 9.3. The peak due to component 2 has moved to a shorter retention time.
During method development a value of pH in the flat region of the plot is preferable, where the pH is greater than 2 pH units away from the pKa, since small changes in pH do not result in changes in the retention time. This may not always be possible, either because the pKa of the analyte is not known, or because the benefit of using HPLC is that it separates mixtures of components, and when multiple components, the pH of the buffer may not be suitable for all the analytes in the mixture.
For weakly ionisable compounds, a buffer may not be required, it may be sufficient to make the mobile phase acidic, or basic. Acids such as formic, acetic and trifluoroacetic are commonly used to prepare mobile phases, particularly for use with protein and peptide analysis by LC-MS.
Factors and Factor Levels for Mobile Phase pH
The parameters in an HPLC method which relate to the buffer in the mobile phase are the pH, and the buffer strength (the latter also applies to acid and base modifiers). To assess robustness these are the factors which need to be investigated, and to enable this suitable factor levels need to be selected. The factor levels should represent the interval between the extremes of potential variability for each factor when the method is in routine use.
If the actual pH of the mobile phase varies each time the buffer is prepared, then the retention time of any analyte which has a pKa close to the pH of the mobile phase may also vary. This can lead to problems with identification and also quantification, if the resolution of the analyte peak is adversely affected. The appropriate factor levels for investigation will depend on the potential variability of the pH adjustment when preparing the buffer solution, and will therefore be most likely due to the accuracy of the pH meter, and the mixing of the buffer solution during pH adjustment. The specifications of pH meters vary and therefore a suggested approach is to use a worst case approach of 0.2 pH units away from the nominal pH, unless information is available to suggest that more variation may be possible. Therefore, for a method with a nominal buffer pH of 3.2, extremes of 3.0 and 3.4 would be investigated.
If the amount of the buffer salt (or acid/base modifier) is insufficient to achieve the required pH in the presence of the analyte, then reproducibility problems are likely. The potential variation in routine use for this factor is related to the measurement, whether by volume or weight. The effect of these variations in buffer strength is unlikely to make the difference between sufficient and insufficient buffering capacity and therefore this is a factor where it may be useful to investigate an interval which is greater than the probable variation in routine use. For example, for a buffer which is prepared at a concentration of 25mM, it may be interesting to investigate an interval from 20mM to 30mM. The choice will depend on the actual buffer or acid/base modifier in use and whether the buffer capacity is known to have associated robustness issues.
In conclusion, buffer pH is an important robustness factor when an analyte is ionisable, and potentially when any of the components in a sample matrix are ionisable. Variation in retention time is a common HPLC method robustness problem and is often related to variability in pH adjustment combined with analytes which have a pKa close to the buffer pH. If a robustness investigation highlights this problem then the preparation of the buffer needs to be carefully controlled, including the use of suitably accurate pH meters and probes. This is usually done by providing careful directions regarding the buffer preparation in the method documentation.
- International Conference on Harmonisation (ICH) of Technical Requirements for Registration of Pharmaceuticals for Human Use, Topic Q2 (R1): Validation of Analytical Procedures: Text and Methodology, 2005, www.ich.org