The Future of API Synthesis

Historically, given enough time and resources, almost any small molecule API will succumb to “total synthesis” – first in medicinal chemistry and then in process chemistry. The greatest challenge facing the process chemist, both then and now, is to produce the molecules under a state of control, with a reproducible impurity profile, in an economical fashion, and at increasing scale. These seemingly simple but high-level goals arguably cause the most problems and influence route design the most. In addition, the regulatory and toxicology landscape has changed and a much more rigorous approach to impurity control, especially of potential genotoxins, has emerged (and rightly so).

More recently, the industry has recognized that it needs to be able to produce its products in a sustainable manner. For example, in almost all cases, solvent use is the greatest contributor to waste during API syntheses, which adds not only cost to the process, but also an environmental burden that must be tackled – with solvent type and usage being the main areas of concern. As drug makers increasingly seek more potent, targeted and (often as a consequence) more complex molecules, the pressure on synthetic chemists to find cost-effective and sustainable routes to their drug candidates has grown.

In terms of challenging chemotypes, chirality is a heavily investigated area. In the early 1990s, the FDA specified that APIs should be produced in high isomeric purity. A command that resulted in many intermediates or APIs being classically resolved, resulting in at least a 50 percent increase in the cost of raw materials, increased waste, and potential impurity issues with residual off-isomers. As asymmetric catalytic methods (both chemo- and biocatalytic) have advanced, they have become indispensable tools in producing chiral APIs.

Advances Over Time

Most modern pharmaceutical molecules have one or more chiral centers at which the stereochemistry needs to be controlled. Compound chirality can produce some of the biggest challenges, and the synthetic community continues to deliver exquisite methodologies to tackle these. Consider the historical progression of producing chiral secondary alcohols: originally, these might have been produced as a racemic mixture and resolved by diasteroisomeric ester formation. HC Brown’s hydroboration methodology then enabled asymmetric ketone reduction, but required a stoichiometric chiral reagent. The Corey–Bakshi–Shibata system was another step forward as it used chiral oxazaborolidine catalysts and achiral, but potentially hazardous, borane complexes. After that, asymmetric catalytic hydrogenation of ketones, as pioneered by Noyori, was a powerful advance but is reliant on ruthenium, a low-abundance metal, hydrogen gas and high pressure. 

Following somewhat in the shadow of many of these methods has been the use of ketoreductases. These natural enzymes have been known for a long time, but saw sporadic use in large-scale API synthesis because they were perceived as unstable in typical reaction matrices, difficult to scale, or unproductive in terms of asset utilization. The advent of directed evolution methodologies for optimizing enzymes, however, has dramatically expanded the use of ketoreductases – and enzymes in general – as many of the perceived drawbacks vanish with engineered variants. Since then, highly productive processes for the manufacture of some of the world’s largest drugs have been delivered – atorvastatin, simvastatin and sitagliptin, for example. The use of ketoreductases is, therefore, a good example of a methodology, enabled by directed evolution, that tackles chirality, and at the same time provides advantages when it comes to impurity generation, safety, atom economy and sustainability. 

Secondly, cross-coupling methodologies, most notably the Suzuki cross coupling, have had a dramatic effect on the synthesis of APIs. The ability to control regiochemistry by selectively heterocoupling two chemically differentiated aromatic partners is immensely powerful. In recent years, concerns about the potential genotoxicity of intermediates, the use of low-abundance metals, and even the use of pre-functionalized aromatics has spurred the synthetic community to develop methods that can activate C-H bonds, or use earth-abundant metals, such as iron.

What Lies Ahead

For practical, large-scale syntheses, patient safety in the form of improved control or elimination of impurities will be a considerable driver for the future. Therefore, any reaction type with dramatically improved chemo-, regio- or enantioselectivity, that does not sacrifice other key attributes, will always be of interest. When these methods become chemo-selective enough, resulting in very low levels of impurities, it will become increasingly possible to run multistep chemical processes in single-vessel reaction cascades. This synthesis of complex molecules in just a few discrete operational steps will lead to a much better return on capital.

Methods utilizing more sustainable metals for cross-coupling chemistry will also increase in number and use in large-scale manufacturing in the future.

As mentioned earlier, sustainability is a key driver for major pharmaceutical companies. We, therefore, expect to see more and more technologies emerge that enable shorter synthetic routes and that do not depend on hazardous reagents or non-sustainable solvents. Biocatalysis, fueled by directed evolution techniques, will be increasingly deployed to heighten selectivity and increase sustainability.

Continuous or semi-continuous processing is now mainstream and will continue to expand in scope, and corresponding synthetic methods will be discovered and implemented. For example, the current resurgence in interest in methods to activate molecules in potentially non-conventional positions – for instance by photocatalysis or electrochemistry – are well-suited to continuous or semi-continuous processing. Recently, the use of photo-biocatalysis with nicotinamide dependent ketoreductases was demonstrated to provide non-natural reactivity (1).  The same group has also shown that photo stimulation of flavin dependent ene reductases can impart new non-natural activity (2). 

Finally, the difference between small molecule APIs and large molecule APIs has begun to blur, and improved catalytic and biocatalytic, biologic-compatible methods for processing complex macromolecules will continue to be developed. 

Data and knowledge lead the way

The future of synthesis may also be affected by high-throughput experimentation and data analysis via AI algorithms.  In general, efficient learning from experience provides tremendous advantages when designing novel synthetic routes, novel reagents that enable such routes, and novel processes that ultimately facilitate the implementation of such routes at scale.

In 2006, we found we needed better directed-evolution methodologies to enable a manufacturing process for hydroxynitrile, the chiral atorvastatin starting material. Traditional methodologies did not yield sufficient improvements, so we introduced ProSAR, a machine-learning algorithm, to the field of directed evolution (see sidebar, Protein Engineering). At that time, the urgent need for a better enzyme on our end and the step changes taking place in the cost of DNA sequencing prompted and enabled this development. The resultant process reduced the cost of manufacture by an estimated 50 percent (3).

These days, data from high-throughput experimentation is the learning material that feeds AI algorithms to provide potentially improved versions of the route, reagents, and process. In our hydroxynitrile example, such datasets included structural data (enzyme sequence) and activity data from a range of different reaction conditions. For the first time, mutations that appeared beneficial for activity were also found to be deleterious for stability, and such knowledge led to the broad adoption of ProSAR, ensuring that only truly beneficial mutations were retained.

With the introduction of ever-faster analytical instrumentation and the adoption of automated screening workflows, larger and more information-rich datasets are now being generated. Databases that store the experimental information in properly structured and easily searchable form now inform and predict the experimental path for creating the desired enzyme faster and faster.

With highly targeted, high-throughput screening and machine-learning-based directed evolution, we have successfully engineered ketoreductases for commercial manufacture of a broad range of chiral alcohols, which are often key intermediates in the synthesis of pharmaceutical ingredients. Such experimental datasets for a wide diversity of compounds are now used to train AI algorithms so that the physical interactions between a substrate and an enzyme can be modelled to yield increasingly accurate predictions for new reactions.        

As we continue to amass data on the substrate scope of different enzyme classes, computational tools will increasingly guide the selection of enzymes capable of catalyzing a target reaction under the required reaction conditions. Highly desirable characteristics, such as solvent stability, thermostability and pH can be deduced from previous characterizations of related enzymes using machine learning algorithms. With growing, high quality datasets and improved AI tools, directed evolution of enzymes will become increasingly rapid through more targeted computational predictions and faster decision making by the scientist. 

Some companies are content with their current synthesis methods – even though they may not be the best solutions! But the pharmaceutical industry needs better, often more chemically complex, molecules that provide increased efficacy and are safe to use. They also need to produce these molecules in an environment where tremendous societal pressure demands lower cost products. To fulfil the promise of providing more complex molecules that are produced at lower cost, innovation is required at multiple steps in the drug development process, including the medicinal and process chemistry steps.

Biocatalysis, enabled by directed evolution, is an innovation that enables the synthesis of complex molecules at any stage of the drug development process. The use of enzymes allows the development of short synthetic routes that create difficult-to-form bonds, as well as introducing chirality. These techniques are useful for the medicinal chemist, and as a candidate advances, better enzymes can be developed for ease of use, for cost, or for enabling (semi-) continuous processes. Suddenly, a lot of risk can be circumvented by making the right molecules early on, using a process that is close to scalable right from the start.

More and more companies are stressing themselves to find new ways to solve chemical manufacturing challenges and to use the power of big data to help overcome them more quickly. At the same time, these companies are engaging in partnerships to drive innovations that neither party could achieve alone. By applying the full breadth of synthetic tools available, by thinking holistically about route design from the early stages of drug discovery to the mature environment of branded drugs, and by leveraging multidisciplinary teams inside and outside their organizations, commercial chemists can implement truly disruptive innovation more readily, and with less risk and cost, than many might think.

Protein Engineering

Enzymes are immensely powerful catalysts which, in theory, allow chemists to address many of the key challenges associated with API synthesis. Enzymes have excellent chemo, regio- and stereoselectivity, and drive high-yielding, environmentally friendly processes. The enzymes themselves are sustainably produced by fermentation, are biodegradable and are generally used in aqueous media. However, natural enzymes do have limitations; for instance, their selectivity for pharmaceutically relevant substrates can be limited as these enzymes did not evolve to accept such molecules naturally.

Similarly, if the natural enzyme produces the wrong enantiomer, then the enzyme is of no use, synthetically. Natural enzymes are often inhibited by process-relevant concentrations of substrate or product, can be unstable in the conditions under which chemists would ideally like to use them and — when used in large amounts — can cause problems during work-up.

All these drawbacks can effectively be addressed by enzyme engineering. Directed evolution facilitates the rapid engineering of enzymes for the desired substrate under the desired conditions, as defined by the process chemist. This notion is hugely important; whenever an enzyme shows a trace of a desired activity, that activity can be amplified using directed evolution. From an enzyme that provided one turnover every five days, for example, directed evolution aided the development of a transaminase that is now used for the commercial scale production of sitagliptin, the API in Merck Sharp & Dohme’s Januvia.

Highly selective catalysts can often enable the chemist to redesign routes, removing redundant steps and increasing efficiency. This ability to conceive of a route on paper and then make the biocatalysts perform the desired functions often has a far larger impact on productivity than simply replacing a chemocatalyst with a biological equivalent.

Furthermore, chemists aren’t necessarily limited by the chemistry that nature performs. Chemists with an understanding of the mechanism of an enzyme’s native activity can “hijack” it and modify it to perform a new transformation, such as repurposing cytochrome P450 monooxygenases to perform cyclopropanation, as Nobel Laureate Frances Arnold did. The key realization is that the novel activity seen initially can be extremely low, yet measurable, and then through directed evolution, be escalated to the desired activity for synthetic use.