HTE at AstraZeneca: A History from A-Z

 Citation:

Douglas, J.J.; Campbell, A.D., et. al. The Implementation and Impact of Chemical High-Throughput Experimentation at AstraZeneca. ACS Catal. 2015, 15, 5229-5256. 

https://pubs.acs.org/doi/10.1021/acscatal.4c07969

Summary Figure:

Background:

High-throughput experimentation is a tool/technique/method to figure out the optimized conditions for a reaction as fast as possible. This paper details the history of HTE at AstraZeneca, which I think offers a helpful description of the strengths and weaknesses of HTE and where the field currently stands. 

AstraZeneca first began using HTE around 2004-2005, with the establishment of a dedicated HTE lab in the UK. Setting up HTE is (or at least was in 2004) more complicated than buying a single instrument, as figuring out the right set up for them required the use of multiple different pieces of equipment. For example, a solid handling robot, a separate liquid handling robot, and a heating block/24-well plate for test tubes were housed inside dedicated gloveboxes. 

Over time this set up was optimized and new instruments were added to evaluate reactions (ex. a scavenger screening system to detect levels of trace metals from cross-couplings). The authors actually show a schematic of the HTE lab at their Gothenburg site in Sweden, one of their three (the others are in Cambridge, UK and Boston, USA). I'm somewhat surprised at how small this is- while the gloveboxes are quite large (8, 12, and 16 arms in total), there are still only 3.

How it works:

 The authors spend some time expounding their philosophy for HTE: "HTE as a means of reaction understanding rather than simple hit-finding." To break this down further, HTE as a means of hit finding would mean screening a wide variety of conditions trying to get from 0% to 10% yield (an initial hit). For example, the authors described the use of HTE to improve an asymmetric hydrogenation needed for the process route of a drug candidate. If the plan was to use HTE to "find a hit", then they might have screened a 96 well plate with literature conditions from 96 different papers. Even if these weren't the optimal conditions, it is likely that at least one of them would be close enough to provide a noticeable yield/enantioselectivity and offer a starting point for future optimization. These reactions would use different solvents, precatalysts, ligands, additives, and pressure/temperature/times, but would screen a lot of chemical space. Running those 96 reactions would be cumbersome for the average chemist, so HTE can be very helpful in automating this process. 

In contrast, AstraZeneca focused their screening on only Ru and Rh precatalysts and conducted an initial focused ligand screen to solely evaluate the effect of the ligand. Once they identified J505 as the optimal ligand, then they screened solvents and additives to arrive at their optimal conditions. The difference between these two processes can be subtle, and often likely would work well together. However, AstraZeneca prioritizes these more focused HTE experiments because they value the ability of HTE to increase yields from 10% to 90% over the ability to go from 0% to 10%. 

 

The authors describe two major reaction types for which HTE is used: cross-coupling reactions and biocatalysis. In general, 86% of the reactions optimized using HTE involved catalysts, with the two outlier types being amide coupling and SNAR. This makes sense, there is a much larger chemical space for catalyzed reactions because you must screen the catalyst in addition to all the other conditions. 

Additionally, it is very easy to keep a library of precatalysts + ligands, or a library of enzymes, in one of these gloveboxes and pull them out at will. Once the system is established, you can run your library for all your asymmetric hydrogenations, so the cost is mostly upfront. 

I believe that repeated screens of 24 well plates would get more information than a single 96 well plate. However, I think the article makes it clear that most reactions get 3 screens and they're done, which makes sense because time at the HTE lab is valuable. 

Initial Questions and Key Findings:

1. Can you build a robot that runs reactions faster than a PhD chemist?

A: Yes, but it is challenging. A lot of work goes into designing robots capable of weighing out small quantities of different types of materials and then running reactions and collecting data on the results. However, this isn't to say that the robots are slow- they are WAY faster than humans. It just means that there is a lot of room for improvement. This is particularly true for microscale reactions, potentially on 384 or even 1536 well plates, which would allow extremely fast screening of chemical space, but has physical limitations and issues with replicating process-like conditions. 

2. What processes benefit the most from HTE?

A: AstraZeneca believes that taking an initial hit and subjecting it to optimization through HTE is the best way to use it. You can use HTE to identify possible problems or side reactions and then figure out the way to resolve these issues. 

3. Are the machines taking our jobs?

A: No. HTE is a tool to make chemists more efficient, but the robots aren't able to plan the experiments and think through the key findings to turn data into something useful. Even the most cutting edge robots require significant parameters to be set beforehand, enough that I don't think it saves much time to tell the robot exactly what to do (https://pubs.acs.org/doi/10.1021/jacs.4c17738). HTE automates the really tedious activity of weighing out small quantities of powders. 

Takeaways:

HTE is really powerful because it takes what a good chemist is already doing (running reactions) and cranks it up to 96.