Skip to main content

Chemical News

Simple Rings, Simply Wrong

Medicinal chemists spend a lot of time thinking about the relative greasiness of their molecules. Being professional scientists, of course, we have come up with some slightly more quantitative phrases than “relative greasiness”, but that’s definitely the idea. How hydrophilic/hydrophobic a compound is determines not to what extent it will dissolve in water – it can also be a measure of much of its behavior after it’s dosed in a human patient. Note the weasel word “can” in that last sentence though.

“Greasiness” is usually expressed as logP, the ratio of how much compound dissolves in (greasy) n-octanol versus water, and people have been trying to relate experimental or calculated logP values to drug behavior for decades now. There are broad trends, for sure. Over a certain logP value (I’m not going to give a number!) a given compound is more likely to have trouble getting absorbed and distributed throughout the body. There are plenty of potential problems: it’ll stick to proteins and membranes that you don’t want it to, it might start to pile up in fatty tissue, if it’s greasy with an acidic group it might end up going back around in the bile duct, if it’s greasy with a bunch of carbon-hydrogen bonds it’ll get ripped up by metabolizing enzymes that are on the lookout for that sort of thing, and if you batten it down with a bunch of carbon-fluorine bonds instead it might have a long, trailing half-life of month because it doesn’t get metabolized much at all.

There are worries for the “under” side of the scale, too – wildly water-soluble compounds can just sluice straight out the kidneys sometimes, although that’s not a problem that we often have in drug discovery work. No, the usual struggle is to keep your compound from being a ball o’lard, because most of the time we medicinal chemists tend to make things bulkier and less polar the longer we have them in our hands. We fight against these tendencies a lot more than we used to, to the point that logP concerns can actually become overdone. Attempts have been made to directly correlate logP to likelihood of toxicity, etc., and the arguing has been intense (it’s probably too simple a measure to bear that much weight).


Let’s sidestep that stuff, though, and stipulate for the moment that that making your compound greasier for no real gain in some other area is probably a bad move. All this is prelude to a simple but annoying question: can we, as chemists, compare two different but broadly similar structures and say which one is more polar? OK, back up: sure, we’re a self-confident bunch, but can we say it with any hope of being right? I ask this because of this new paper in J. Med. Chem. It compares simple variations of some absolutely classic cyclic amines: the six-membered ones with the one-carbon-bridged bicyclo variants. You will find both of these all over the literature – if I had a dollar for every morpholine, piperidine, or piperazine claimed in a drug patent I’d be writing this from my private island, although that I assumes that I wouldn’t still be asleep. Anyway, the right-hand forms, drawn in both their “flat” and conformational styles, have one more methylene stuck into them. So does that make them less polar?

Well, if you just hang a methyl off of them, it sure does. And if you just add a methylene into a chain, it sure does. But in this case. . .it doesn’t. The measured polarity (as logD, which is logP done at pH 7.4) is greater every time. And this while calculated logP values either predict that the compounds get less polar or only very slightly more. Experimental data for the win (again and again)! The paper is from AstraZeneca, and it does an AstraZeneca-style tour (sorry guys) through all the methyl substitutions, all the one-carbon bridges, all the two-carbon bridges, etc. This is the big one, though, and it’s both a very meaningful change, one that medicinal chemists often want to put into their molecules, and one that both intuition and quick calculation would tell you isn’t there. But it is. The best explanations for this behavior are increased basicity of the nitrogen (which especially shows up in logD values, of course) and an increased in exposed polar functionality due to the conformational change.

The paper ends up recommending that any team putting one of the six-membered heterocycles into their compound series should investigate the one-carbon-bridged versions as a matter of course. I would definitely agree with that, based on the data shown. I would add a few more general recommendations on top of that one, though. (1) Don’t blindly trust calculated property values, because they’re not as accurate as you think they are. The corollary is even more general: (2) never talk yourself out of an easy experimental test of an idea. And (3) be willing to consider that your chemist’s intuition is wrong – make a few things that you think might not work, once in a while. Our brains are not as accurate as we think they are, either.

19 comments on “Simple Rings, Simply Wrong”

  1. Anon says:

    Strained carbon single bonds have more P character and are therefore more polar. Basic Orgo 101. Think cyclopropane.

  2. Peter Kenny says:

    Introduction of the the carbon bridge would be expected to affect the thru-bond interaction between the heteroatoms (I would guess that it will weaken what is an inductive effect making the lone pairs of the heteroatoms more available). Piperazines in which one nitrogen is unsubstituted and the other is alkylated tend to protonate on the former under normal physiological conditions (thus is discussed in article that I’ve linked as the URL for this comment). I would guess that this may explain the difference between NH and NMe.

    On a more general note, one needs to be careful when using logD as a measure of compound quality since this implies that you can make compounds better by simply increasing the extent to which they are ionized.

    1. Derek Lowe says:

      Good points. That is indeed the sting hidden in logD measurements, which just goes to show that even real experimental values can lead you down the wrong path if you don’t stay alert.

      1. Peter Kenny says:

        Generally, I think that alkane/water logP provides a better measure of polarity and I’ve linked another article as the URL for this comment. This article shows how polarity of individual hydrogen bond acceptors and donors can be estimated from alkane/water logP measured for structurally prototypical model compounds.

  3. John Wayne says:

    I TAed physical organic for a few years, and I can tell you how we taught this. There are three effects that could contribute to the overall polarity of the molecules when you add this bridging methylene. In order of importance:

    1. Heteroatoms presented 1,4 across a six membered saturated ring interact with each other. This is apparent in the pKa trend from piperidine to morpholine. If you look at the bottom left structure above, the ‘axial’ lone pair on atom X mixes across the top of the chair with the antibonding orbital of the lone pair on the nitrogen making a new molecular orbital and strengthening/shortens all those bonds. When you add this bridge, minimizing all the steric interactions removes this stabilizing interaction – the two heteroatoms become more independent from each other and more polar. *

    2. Ring strain increases polarity a bit, but this is probably a minor contributor in this case. We could contract the resulting 5,5 systems and see. The strong effects on microsomal stability you can see from going to small alkyl rings is mostly from the increasing homolytic C-H energy of those strained methylenes (and the removal of those methylenes as well).

    3. The inductive interactions between these heteroatoms is fairly similar between the two compounds.

    * You can rationalize the anomeric effect the same way; this is over fewer atoms.

    * The alpha effect (two heteroatoms next to each other, Ex. hydrazine) is the opposite of this effect. The two adjacent heteroatoms cannot interact and their lone pairs repel each other making them more reactive.

    Anyway, this is the way I was taught (and did teach) this stuff; I hope it helps.

  4. Wavefunction says:

    One of the most important – and longstanding – lessons from this study is how medicinal chemists ignore or downplay conformational effects at their own peril.

  5. cynical1 says:

    Okay, the world cannot reliably predict/calculate a simple logP of a small heterocycle but somehow I’m supposed to believe that AI is going to save the day and start discovering drugs for us in the near future.

    1. Dionysius Rex says:

      Burn the witch!

    2. achemist says:

      Just use an AI on Big data to synthetize new molecules by photoredox catalyzed electrochemistry. Preferred targets are the microbiome and CRISPR/CAS9 to develop personalized medicine.

      Let me know if I forgot any of the current Buzzwords

      1. franny says:

        biocatalysis 💁‍♂️

      2. Scott says:

        You need to have at least 24 buzzwords on your bingo card (5×5 grid with the center spot free).

        More buzzwords are needed!

      3. WildCation says:


      4. loupgarous says:

        and the proteosome (an active buzzword when looking at the thalidomide analogs as novel treatments for odd cancers)

    3. Wavefunction says:

      Well, AI isn’t trying to calculate anything; it’s simply trying to identify trends. If this kind of phenomenon is common enough, it will be part of the data and therefore of an AI’s predictive model. The question here is really one about the data rather than the algorithm.

    4. Gallery says:

      “I’m supposed to believe that AI is going to save the day”

      The obvious response is to replace “AI” with “medicinal chemists”. The point of the article was that our intuition fails at this task. You might think that AI won’t work, but there’s no reason to believe that people are better! Once the experiment been run, the same data is available to both the AI and the person and at least there’s a better chance that the AI read the paper.

  6. Mike Turner says:

    Trivia time – in the first paragraph I think you mean “How hydrophilic/hydrophobic a compound is determines not JUST to what extent it will dissolve in water”?

  7. Mark says:

    I’m enjoying the blog and the commentary! As a non-chemist it’s interesting to watch how, after an unexpected effect has been found experimentally, it’s something that one would obviously expect from theory…
    As an LC-analyst, I’m also interested, because this presumably affects the accuracy of software products that aim to predict retention on reverse phase columns.

    1. John Wayne says:

      The majority of software programs that calculate values for you are of varied quality; they are only as good as the training sets used to make them. In an ideal world, the user would know how applicable a given program is to the problem they are looking at. In the real world, people jam structures into them they have no business querying. As you correctly point out, about zero programs are going to make this call.

  8. Li says:

    Never made much sense to me why n-octanol was the solvent of choice for one of the two partitioning solvents. Why not methane? (or at least an alkane – heck, even benzene). It’s obvious that n-octanol IS polar. I guess it’s because it is itself substantially insoluble in water. …although come to think of it, why water as the ‘prototype’ polar solvent? I mean it is exceptional in so many ways, it’s a pretty bad choice for a standard. (Not meaning that in biologically relevant systems water isn’t a good standard, but for first principles solvency/polarity isn’t it TOO unusual? H2S or NH3 maybe? SO2? Last I looked into solubility prediction, there were like half a dozen (often more) parameters you needed to get a reasonable handle on water solubility. Way too many to make it practical. Why wouldn’t density functional calculations make short work of this, I wonder?

Comments are closed.