Self-optimisation of automated continuous reactors

• Main Author: Holmes, Nicholas
• Other Authors: Bourne, Richard ; Blacker, John
• Published: University of Leeds 2017
• Subjects: 615.1
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.718807

The optimisation of problematic reaction steps in the synthesis of a drug compound is crucial for pharmaceutical process development. In recent traditions, this has carried out using design of experiments (DoE), which shows the key reaction variables and provides optimum reaction conditions. The process can require a lot of experiments and be time and resource consuming. The speed of optimisation experiments can be increased by using automated platforms complete with online analysis, which carry out reactions and acquire analytical samples without any human intervention. If these experiments can be carried out in continuous reactors then they will benefit from faster kinetics, enhanced heat and mass transfer, improved safety and higher productivity over their batch counterparts. An automated self-optimising flow reactor combines a continuous reactor with online analysis and feedback loop. The feedback loop contains full computerised control and monitoring of all equipment as well as a minimising algorithm, which will use the results from the online analysis to predict new optimum conditions. The technique has been shown to optimise the synthesis of small organic compounds but has, so far, yet to be widely used in pharmaceutical process development. This thesis has improved self-optimising technologies in order to make it a useful technique in pharmaceutical process development. First, the final bond forming step in the synthesis of an active pharmaceutical ingredient was optimised for yield. Studies were primarily carried out on a model compound in order to establish the correct reactor setup before transferring to the active compound, which found an optimum yield of 89%. The work also provided mechanistic evidence for generation of impurities. Next, response surface models were successfully fitted to the data obtained from a branch and fit algorithm optimisation of a Claisen-Schmidt condensation. In depth statistical calculations show how DoE models can be generated from self-optimisation data with good fit and predictability (R2 > 0.95, Q2 > 0.90), and with the aid of commercial DoE software. Further work developed the use of direct mass spectrometry (MS) as the online analytical method. The short method times and real-time analysis of MS allowed a steady state detection function to be built, followed by a linear calibration model of all the species in the amidation of a methyl ester. The reaction was optimised for yield using branch and fit algorithm, and DoE, with excellent agreement between the two techniques in both optimum conditions and responses. Finally, changes were made to the optimisation program to reduce the amount of material required for automated optimisations. Reaction pulses of sub-reactor volumes were pumped through the reactor, dispersed in a continuous phase of miscible solvent. Residence time distribution experiments were carried out to characterise the dispersion of the reactor and calculate the minimum reactor pulse volume. Optimisations were primarily carried out using pattern search algorithm and a multi-objective evolutionary algorithm, the latter of which generated a three target function optimum, reducing the amount of waste by 81%. Overall this work has shown how self-optimisation can be a valuable tool for pharmaceutical process development. The existing technique has been improved by demonstrating its use in the synthesis of pharmaceutical compounds, combining it with existing DoE techniques, adding new forms of online analysis, and reducing the amount of material required to deliver a multi-target optimum.