A commenter on my last post about the computer-aided discovery of a promising new morphine-like analog brought up a very interesting point that deserves some discussion. As the commenter who was the synthetic chemist on the study says,
"I particularly liked the part where the authors admit that no matter how smart their approach was, many unique properties of the lead compound were pure luck and wouldn't be possible to predict with current tools. And if Brian Shoichet writes this, I'm pretty sure that nobody in the world can claim otherwise. As a synthetic chemist behind PZM21, I confess that its synthesis in diastereomerically pure form and - especially - figuring out its absolute configuration were quite a luck, too."
This sentiment is confirmed from reading the paper where the authors state:
"Several caveats deserve to be mentioned. Although structure-based discovery succeeded in finding novel scaffolds and supported facile optimization, some of the properties of PZM21 were likely fortuitous. Biased signalling through G protein and arrestin pathways reflects the stabilization of conformations over 30 Å from the orthosteric site where PZM21 binds. We did not select molecules that preferentially stabilize these conformations, but instead relied on chemical novelty to confer new biological properties.
Receptor subtype selectivity was attained by simply selecting molecules that extended into variable regions of the receptor, a strategy that may not always work. Several aspects of the pharmacology presented here remain preliminary, including the metabolic stability studies and the pharmacokinetics, and it is not clear at this time whether the unprecedented in vivo activity of PZM21 reflects its biased and specific agonism, or some other feature conferred by its novel chemotype. Finally, identification of agonists from docking to an inactive state receptor structure cannot always be relied upon, though there is precedence for doing so against opioid receptors."
Where does all this interesting biology come from? Several factors may contribute, but recall that the top ligands form an extra hydrogen bond with an aspartate which hasn't been seen in opioid receptor ligands before. Knowing the cooperative effects which ligand binding can transmit through protein structures, it's not farfetched to believe that this tiny difference may have contributed to the biological effects. The general point here is that several biological features of the small molecule, especially biased agonism (differential interactions between GPCRs and their protein partners) and stabilization of conformations in remote parts of the protein were not predicted. However, the chemical features of the ligand were discovered more rationally by docking. Thus as the authors say, they used chemistry to discover novel biology.
This is an important point that should not be lost on critics as well as advocates of molecular modeling and medicinal chemistry. Sometimes novel biology can be discovered precisely because it's unpredictable. Now this is not some facetious cop out that puts a positive spin on the unpredictability of modeling downstream changes in protein conformation and inter-protein interactions. But what's happening here is that because chemistry is necessarily correlated to biology in highly non-linear and emergent ways, it's hard to tell beforehand what will happen when you discover a new ligand with novel interactions with a protein. On the other hand, this same unpredictability also provides a chance to make forays into unexplored biological territory.
It turns out this is actually very good news for chemists and drug designers. Operationally it's far easier to simply (relatively speaking) discover new ligands and then find out what they do: virtually docking a large screen of molecules against a complex protein like a GPCR is far easier than trying to predict what happens downstream after those molecules bind. Similarly, making analogs of these molecules is also quite easy. On the other hand, modeling any kind of remote conformational changes or biased agonism in this case would not just have been futile but would have likely led the researchers in the wrong direction, and this fact is more generally true of other targets too. It was far better to use the chemistry as the starting point, and then let the biology sort itself out.
This study therefore demonstrates a very fortuitous aspect of medicinal chemistry and modeling that is true of drug discovery in general: you can actually use the unpredictability of these techniques to discover new biology, because the more uncertain your starting point is, the more probable it is that it will lead to novelty. In uncertainty lies opportunity. All you need to do really is to make sure you are endowing the system with novel chemistry; the complex but deterministic relationship between well-understood chemistry and ill-understood biology will then take care of the rest. They key is to generate as many starting points for unexpected biology as possible. You may not always discover useful biology this way, but you certainly increase the chances of doing so.
Sometimes you need to give serendipity a deliberate platform to launch itself.