Some of you may have noticed our latest publication that’s just out in the "just accepted" section of J. Med. Chem. which describes the use of our difluoromethylation reagent, DFMS, as a predictor of whether an azaheteroaromatic fragment will be susceptible to metabolism by aldehyde oxidase (AO). The actual chemistry is covered in depth in the paper, so I thought I’d write about some of the challenges and doubts we faced in developing the project from the initial concept to the finished paper.
When Phil asked me to join “AO project” – a collaboration with Pfizer – and explained that the idea was to show that it was possible to use DFMS to predict whether a potential drug molecule would be susceptible to metabolism by aldehyde oxidase, I was somewhat skeptical. Yes, there were early results where DFMS reacted in the same position as aldehyde oxidase was known to oxidise, and the products were resistant to further metabolism by AO, but there were a few other issues to consider:
(i) Some of the known AO substrates we needed to develop difluoromethylation conditions for were really tough substrates. While it eventually proved possible to difluoromethylate them, finding conditions that prevented the more amenable substrates vanishing into a complex poly-difluoromethylated mess was another matter. These conditions were harsh, and there was a real risk that under the same conditions almost any substrate might work to some degree.
|Some substrates difluoromethylated too well…|
(ii) Aldehyde oxidase is an enzyme which can show incredibly sensitive discrimination between substrates, even when changes are far from the reactive center. It can also do other things like oxidizing aldehydes. DFMS isn’t an enzyme, and can’t do these things.
|It seemed likely that DFMS would have difficulty distinguishing these two.|
(iii) Aldehyde oxidase invariably oxidizes heteroarenes adjacent to nitrogen. But we already knew that substrates such as 2,6-disubstituted pyridines could be difluoromethylated.
A further concern was whether it was possible to use one reaction where we didn’t always understand the selectivity to predict another reaction where the selectivity was also poorly understood. Since there is a limited data set of published AO substrates, it would have been reassuring to match the factors that control selectivity with DFMS with those that control selectivity with AO. However, although we’d made some progress in understanding reactivity, there was still some way to go before this could be consistently extended to complex substrates with multiple or fused heteroarenes. Our attempts to predict selectivity based on (admittedly very simple) computational methods had also failed.
I’ve already mentioned that this work involved extensive collaboration with Pfizer. Nearly everyone in the Baran lab ends up collaborating with industry at some point, and while this can have some bad points (incredibly early meetings/telecons), for me the major advantage was access to chemicals. While I was investigating regioselectivity, Mike (our main liaison from Pfizer) became my “Chemical Santa” providing the tiny samples of numerous expensive, differentially-substituted heteroarenes that I needed to test hypotheses. I could also test out the chemistry on drugs and agrochemicals that would have otherwise been difficult or too expensive to get hold of.
|Not available for $1 per gram.|
For the AO project, the collaboration with Pfizer was absolutely crucial because it meant that we could obtain and test series of related compounds that had already been tested for AO activity. Much of the previous work on the effect of systematic structural changes on susceptibility to AO has been carried out by Pfizer, and we were particularly interested in this paper which showed dramatic changes in AO metabolism by changing the heteroaromatic fragment. There was no obvious structural reason to explain why these patterns were seen, or suggest that DFMS might show a similar reactivity pattern. It was the perfect test. (To be honest, I thought that this would conclusively show that the project couldn’t work.)
Fortunately accessing these compounds was simple - and the test results were surprisingly good – reactivity with DFMS closely matched the AO activity. This was the point that we figured out that the project might really result in a useful test. DFMS may not be able to mimic every aspect of AO, and we may not fully understand the reasons behind the selectivities, but DFMS did seem to give a remarkably good indicator of whether complex azaheteroaromatics would be susceptible to oxidation by AO.
Even though we were now confident that the concept was sound, there were still loose ends to tie up. Reaction conditions were tuned and re-tuned to provide a clear yes/no response when looking for an M+50 peak. We knew that there would be false positives, even when just considering reactivity on the azaheteroaromatic part of the molecule, and we had to find some good examples.
|A ‘good’ false positive example. I so wanted this one to match up, but at least it showed that the final reaction conditions wouldn’t chew up anything with delicate functionality.|
We also subjected sorafenib to AO testing because although it had an unsubstituted position adjacent to a pyridine nitrogen and would be expected to be prone to AO oxidation, we predicted (and testing confirmed) and it would not be reactive with DFMS. This appeared likely to give a false negative, so waiting for the AO result on this last one was particularly nerve-wracking. Finally we had to check that the chemistry was still robust to a typical accuracy of weighing when setting up dozens of these small scale reactions (i.e. we aimed towards measuring in spatula tips…)
All in all, with extensive screening and testing, we managed to overcome the problems associated with using a simple chemical indicator for an enzymatic reaction, and we’re hoping that the test will be useful for medicinal chemists. If anyone has questions or comments about the project, we’ll be happy to answer them!