Using QbD and DoE for the Systematic Optimisation of a Telescoped Process

In this paper (Gonzalez et al., Org. Process Res. Dev. 2019, 23, 6, 1143-1151) from the Chemical and Synthetic Development team at Bristol-Myers Squibb, DoE and QbD methodologies have been used to develop a telescoped process for the API step in the route to a reversible inhibitor of BTK. (Scheme 1)

Scheme 1: Two step process to BTK inhibitor (4) and atropisomer impurity (5) through intermediate (3).

There are several benefits to adopting a telescoped process including fewer manipulations and therefore reduced process or plant time, reduced waste and potential cost savings. However, in this instance, the development of a telescoped process was driven by storage and stability concerns of a solid-state intermediate which could result in an increased level of impurity and affect the quality of the drug.

Due to the complexity and scale of the problem the team elected to optimise three separate unit operations in parallel using DoE and high-throughput experimentation, then bring together the optimised steps into a single process.  The experimental plan started at the end, by identifying the critical quality attributes (CQAs) which were important for delivering the required quality target product profile (QTPP) (Figure 1).  Investigation of the API reaction step would highlight the Critical Process Parameters (CPPs) for the penultimate reaction and the associated workup.

Figure 1: Experimental design plan guided by CQAs, CPPs and DoE responses.

Firstly, the API step was evaluated in order to determine the CPPs which needed to be controlled in the penultimate step and workup. The aim of the investigation was to understand which parameters affect the formation of the atropisomer and learn how to minimise this. A design was carried out looking at 7 factors at 2 levels and 2 additional measured factors (Table 1)

Table 1: The factors and levels investigated for the API reaction

Statistical analysis of the results suggested that the factors which affect the level of atropisomer are DMF content, water content of both the input and base solutions, and temperature. There is also an interaction between DMF and temperature which shows that at high DMF levels, a high temperature can reduce the levels of the isomer but if the level of DMF is low, temperature does not have a large effect. The best conditions to reduce the isomer are high temperature, low level of DMF and low water in the input stream & base solution. This understanding is significant because in the telescoped process there is less control over the DMF and water levels carried into the reaction. This information was used to increase the reaction temperature from 20 °C, the original condition, to 30 °C. As the level of DMF and water in the input stream may vary slightly from batch to batch, the increased operating temperature will reduce the risk of a higher level of atropisomer.

The design determined the CPPs to be the levels of residual DMF and water and the findings were taken into the optimisation of the coupling reaction (2 to 3, Scheme 1). The potential impact of residual DMF on quality and yield prompted the search for a different solvent but a straightforward solvent swap to MeTHF, the reaction solvent in the API step, was not possible due to solubility issues.

A design was carried out looking at the DMF volume fraction, the total amount of solvent, the equivalents of compound 2, the equivalents of 1-methylimidazole, the equivalents of MSA and the reaction temperature. To increase the efficiency of the investigation further, lab automation technology was used and all experiments were carried out in 1 day. The set of experiments showed that solvent mixture was not critical to formation of 3 with all experiments (where eq 2>1) proceeding to high conversion with low impurity formation. The DMF volume fraction did not affect quality so the reduction of DMF in the reaction by 40% was possible. This change did have an impact on a yield-impacting impurity, see the paper for more details.

The original workup of the coupling reaction involved 4 washes, each of which was to be optimised. Optimisation of a workup is generally easier once the reaction has been optimised and it is good practise to wait until the reaction is understood and optimised before improving the work up.

To ensure quality, strict control of residual DMF and water is required and low lying impurities, residual starting materials and reagent byproducts should be removed. Several designs were carried out using high-throughput extractions to maximise the amount of information gained in optimising the washes. The use of DoE was especially useful for wash 1 as the model suggested a high acid molarity and low volume of MeTHF but stability studies showed that strongly acidic conditions can cause degradation of 3. The molarity of the HCl wash was kept at 1M but the amount was increased to 10 mL/g. The increased wash combined with minimising the amount of extraction solvent improved removal of DMF from 60 to 90% in the first wash. Washes 2 and 3 polished the removal of DMF even more although there were only marginal improvements once the optimised conditions were used. The final wash was optimised to simplify the azeotropic distillation and lower the water content of the intermediate stream as this was important for the API step.

Using DoE to understand the impacts to the API reaction and high-throughput extractions to optimise the workup, a process was rapidly developed to achieve the same quality API with an overall 13% improvement in the yield and a 41% reduction in the PMI.

The paper states that “Since the introduction of the Food and Drug administration’s (FDA’s) guidance on quality by design (QbD) the use of statistical design of experiment, multivariate analysis, and the study of chemical kinetics for process understanding have become commonplace.” While that may be true for larger pharmaceutical companies who have statistics experts in-house, and DoE may be used routinely as part of QbD, it is generally used towards the end of the development timeline to check the robustness and stability of a process or product. While this is very useful, it doesn’t give additional knowledge about the process and knowledge is power. In PMCC’s experience, DoE is being used but not always correctly and we see many clients making simple mistakes which mean that a tool which should increase their efficiency may in fact be making things more complicated or take longer than it should. A real strength of design is getting more information out of fewer experiments. DoE should be used to approach a problem through increased process understanding as this supports QbD, gives more knowledge to the scientist and allows informed decision making.

About the Author:

Dr Laura Forfar,  is a consultant at PMCC where she provides support in the delivery of training courses in experimental design (DoE) and uses DoE to help clients from a variety of chemical industries.

Laura has industrial experience as an organic chemist developing metal scavengers and earned a PhD at the University of Bristol. She holds the position of Visiting Industrial Fellow at the University of Bath where she works with the CSCT providing DoE training and project support.

Join our training courses in QbD or DoE to find out more about how to apply these essential tools to enable you to work more efficiently and achieved enhanced process understanding.