BORMAN ET AL.

The use of open-access high-performance liquid chromatography (HPLC) systems in drug discovery and pharmaceutical development has steadily increased during the past decade. This growth has enabled chemists and engineers to perform a large proportion of their own analysis without dedicated support from analytical scientists. Within chemical development at GlaxoSmithKline (GSK), for example, about half of chromatographic analyses (250,000 samples per year) are performed by process chemists or chemical engineers using open-access systems. The uptake of these systems has been high and the application of traditional techniques for monitoring reactions such as thin-layer chromatography (TLC) has been superseded in chemical development at GSK. Implementation of a novel, global open-access approach using generic fast-LC and custom analytical methods on more than 100 HPLC-ultra-violet (UV) instruments has been described by Roberts et al. (1).

Most HPLC open-access systems within chemical development laboratories are used to generate impurity-profile information using area–percentage responses on non-quality critical samples of active pharmaceutical ingredients (APIs), intermediates, and starting materials. In addition to obtaining this data, synthetic chemists must generate purity data on samples to support route- and process-development. A new tool developed by GSK, the Yieldaliser, enables the rapid and reliable estimation of yields in solution and assays of isolated solids (in addition to generating impurity profiles) without any requirement for the user to prepare standards. The authors discuss the calibration approach, which is performed using an unrelated standard material analyzed at an optimal wavelength which may be different to the compound of interest.

External standardization across different wavelengths

Several standardization methods were investigated during the development of the tool with a focus on automation. External and internal standardization were explored using single and multiple standard sets, which were analyzed at a single wavelength or multiple wavelengths. To analyze compounds that contain chromophores in the typical UV range used in HPLC (210–300 nm), the standard and sample compound had to be analyzed at their respective λMax or λOpt (i.e., the most optimal/robust region within the UV spectrum). Additional functionality was integrated into the open-access HPLC platform (this term is used by Roberts et al. to describe a collection of actively managed and supported open-access instruments [1]), which enables chemists and engineers to perform their own assay analysis.

To generate assay data on a particular compound using the platform, an analytical chemist (i.e., the lead user) must first determine an accurate response factor for the compound of interest (analyzed using the compound's λOpt) against the benzophenone external standard (analyzed using its λmax). The Yieldaliser software uses the response factor in its calculations.

Benzophenone was chosen as the external standard because it is stable, can be weighed out accurately, and is commercially available at high purity. Benzophenone is also very soluble in compatible solvents, has good chromatographic properties on open-access gradient HPLC methods, and has a high extinction coefficient at its λmax of 254 nm.

Figure 1: Sample log-on interface of the Yieldaliser software (GlaxoSmithKline, GSK). (ALL FIGURES ARE COURTESY OF BORMAN ET AL.)

Developing sample preparation protocols

Two protocols were developed to enable the analysis of liquid samples (for generating yield in solution data) and solid samples (assay data). These protocols were created to meet users' expectations. For example, sample preparation had to be quick, easy to follow, and use vessels that could be disposed of at minimal cost. Chemists needed to be able to to generate data that were sufficiently precise (within ± 3% w/w of the actual purity) for the majority of their applications. Trained analytical chemists can perform assays with greater precision (e.g., typically ± 0.5% w/w), however this is far more resource intensive.

Figure 2: Screenshot showing the Yieldaliser (GSK) software's assay interface.

Solid protocol. The requirements for establishing a solid protocol were similar to those used to create the liquid protocol. The developers initially used four-decimal-place balances in the laboratories, Pressmatic dispensers, and EDP pipettes. The final protocol calls for ~15 mg of solid that is weighed accurately on a four- or five-decimal balance into a disposable polypropylene container. Fifteen miligrams was found to be the minimum weight required for achieving an accurate measurement. One mililiter of dimethyl sulfoxide (DMSO) was added to the solid with an EDP pipette, and then 40 mL of solvent was added from the Pressmatic dispenser. DMSO was chosen as the predissolution solvent based on its aproticity, which maximizes solubility while remaining compatible with the system.

Adding and selecting compounds. A similar protocol to the solid protocol can be used to assess the suitability of a compound for the Yieldaliser and to ensure accuracy in determining the response factor. A full assessment of safety, stability, repeatability, and applicability of the compound (e.g., presence of appropriate λOpt ) should be performed. Using five-decimal-place balances and volumetric glassware to determine response factors provides improved accuracy over the liquid and solid protocols described above with a percentage relative standard deviation of < 1.5% (n = 3). Accurate determination of the response factors is crucial for meeting customer expectations of the system's performance.

Figure 3: A proof-of-concept study demonstrated chemists can generate data with acceptable precision and accuracy.

Proof of concept. A proof-of-concept study was performed to demonstrate that chemists can produce data within ± 3% of the actual purity (chemists' expectations) for the previously defined protocols. Four trained chemists were able to generate yield in solution and assay results for samples of known concentration within these limits for several compounds. Figure 3 provides some of the data generated from the proof-of-concept study. More than 90% of the data were within the desired ± 3% limits. The clusters of results that fell outside the limits represent results from chemists who were not familiar with the sample preparation devices. After receiving additional training, the data from these chemists improved in precision and accuracy.

Application

Following the implementation of the Yieldaliser across chemical development at GSK—including sites within the US, UK, and Italy—a wide range of tasks were performed by chemists and engineers using the approach. Three types of analysis—mass-balance determiniation, crystallization endpoints, and solubility curves—exemplify how data provided by the Yieldaliser can be useful in chemical development.

Mass-balance determinations. These determinations involve identifying where major material losses occur during solubilization. For example, during a recrystallization of an API, it is useful to gain an understanding of the efficiency of the initial crystallization along with the subsequent solvent wash steps.

Crystallization endpoints. The Yiedaliser can be used to determine whether visually determined API crystallization endpoints are sound by analyzing the supernatant. Results can be used to determine whether the input batch and crystallization parameters affect the rate of crystallization.

Solubility curves. To aid in the development of the final API-stage crystallization, accurate solubility curves (concentration versus temperature) are required. Several curves can be generated using Yieldaliser data. These curves represent the solubility at several crucial solvent-mixture ratios. Key points in the process can be identified (e.g., seeding composition, point of supersaturation, and completion of crystallization).

Figure 4: Active pharmaceutical ingredient solubility at various acetone:water ratios, including exponential trendline (–10 to +60 °C).

Key points in the crystallization process are shown in Figure 4. For example, the line labeled "post-distillation" represents the concentration and composition of the solution at the end of the atmospheric distillation. The line shows that there is an approximate 7 °C window between the solution cooling to the point of supersaturation (dark blue line) and the boiling point. Also marked in Figure 4 is the seeding point at 53 °C, which is approximately 5 °C inside the saturated solubility curve (pink line) for this composition. The red line represents the isolation composition and shows that very little material (1.2 mg/mL) is left in solution once the slurry has been cooled to –10 °C. This same line shows that most of the API has crystallized during the water addition and before the cool-down from 47 ° to –10 °C. Therefore, it is more important to examine changes when water is added. Users were able to define the final process parameters with the help of these solubility curves.

Continuous improvement

The integration of the Yieldaliser into the open-access HPLC platform has enabled chemists to generate impurity-profile and assay data simultaneously from a single analysis. The additional assay data obtained from the same sample preparation and chromatographic run is therefore generated with little additional cost. Users can perform the sample preparation in ~1 minute, thereby enabling results to be generated and emailed to their desks within 10 minutes of the sample preparation (i.e., when the sample queue is empty). A common open-access HPLC platform also allows chemists to generate Yieldaliser data that can support processes that have been transferred from the laboratory scale into larger production facilities regardless of where in the world the analysis is required.

Figure 5: Annual metrics collected using the Yieldaliser from GSK's Stevenage, Hertfordshire, United Kingdom, facility in 2009.

Figure 5 highlights the most frequent types of assay analysis performed by users at GSK's Stevenage site. More than 97% of the analyses focused on generating yield in solution using the typical protocol or dilute protocol (see Figure 2). The data also show that chemists that used the calibration tool throughout 2009 often did not need to determine the purity of isolated solids (more precise data is often required on solids and therefore generated by trained analytical chemists on dedicated HPLC systems).

Future plans

The examples documented by the authors demonstrate the use of the Yieldaliser within GSK's chemical development division. Implementation of an equivalent validated system, applicable to current good manufacturing/laboratory practice analysis, may help standardize operating methods for determining HPLC assays. Evolving quality by design approaches for analytical methods will potentially provide greater flexibility for changing methods that are already registered (2, 3). Improvement to the sample-preparation protocols is also under investigation (e.g., use of liquid handling devices that can better handle 20 µl over a wide range of solvents) to automate further sample analysis and improve accuracy and precision. Lastly, automated methodology is being explored for automating fault detection and troubleshooting across the open-access HPLC platform (4–6).

Phil Borman* and John Roberts are managers in analytical sciences/chemical development at GlaxoSmithKline (GSK), Gunnels Wood Road, Stevenage, Hertfordshire, SG1 2NY, United Kingdom, tel. +44 1438 763713, fax +44 1438 764414, phil.j.borman@gsk.com
. Barbara O'Reilly is a statistical programming analyst at Roche (Welwyn Garden City, UK). Robin Attrill is a manager in synthetic chemistry/chemical development at GSK in Stevenage. Ian Barylski is a manager and Keith Freebairn is a director, both in particle/process sciences and engineering/chemical development, at GSK in Stevenage.

References

1. J. Roberts et al., Amer. Pharm. Rev. 13 (2), 38–44 (2010).

2. P. Borman et al., Pharm. Technol. 31 (10), 142–152 (2007).

3. M. Schweitzer et al., Pharm. Technol. 34 (2), 52–59 (2010).

4. H. Eriksson and P. Larses, J. Chem. Inf. Comput. Sci. 32, 139–144 (1992).

5. T. Wessa et al., Organic Process R&D 4, 102–106 (2000).

6. L. Kaminski et al., J. of Pharm. and Biomedic. Anal. 51 (3), 557–564 (2010).

Editor's Note: This article is being simultaneously published in Pharmaceutical Technology Europe's October 2010 issue.