Skip to Content

Tiny (And Not So Tiny) Changes

A huge amount of medicinal chemistry – and a huge amount of medicine – depends on small molecules binding to protein targets. Despite decades of study, though, with all the technology we can bring to bear on the topic, we still don’t have as clear a picture of the process as we’d like. Protein structure is well-known as an insanely tricky subject, and the interactions a protein can make with a small molecule are many, various, and subtle.
This gets re-emphasized in this new paper from the Shoichet group at UCSF. They’re using a well-studied model protein pocket (the L99A mutant of T4 lyzozyme, itself an extremely well-studied protein). That cavity is lined with hydrophobic residues, and (being a mutant at a site without function) it’s not evolutionarily adapted for any small-molecule ligands. It’s just a plain, generic roundish space inside a protein, and a number of nonpolar molecules have had X-ray structures determined inside it.
What this paper does is determine crystal structures (to about 1.6A or better) for a series of closely related compounds: benzene, toluene, ethylbenzene, n-propylbenzene, sec-butylbenzene, n-butylbenzene, n-pentylbenzene, and n-hexylbenzene. That’s about as nondescript a collection of aryl hydrocarbons as you could ask for, differing from each other only by the number and placement of methylene groups. (Three of these had already been determined in earlier studies). How does the protein cavity handle such similar compounds?
By doing different things. They found that one nearby part of the protein, the F helix, adopts two different conformations in the same crystal, in different proportions varying with the ligand. (The earlier structures from the 1990s show this, too, although it wasn’t realized at the time). The empty cavity and the benzene-bound one have one “closed” conformation, but even just moving up to toluene gives you about 20% of the intermediate one, with a shifted F-helix. By the time you get to n-butylbenzene, that conformation is now about 60% occupied in the crystal structure, with 10% of the “closed”, and now 30% of a third, “open” state. The pentyl- and hexyl-benzene structures are mostly in the open state. Digging through the PDB for other lyzozyme cavity structures turned up examples of all three forms.
These adjustments come via rearrangement of hydrogen bonds between the protein residues, and it apparently has a number of tiny rachet-like slips it can make to accommodate the ligands. And there’s the tricky part: these changes are all balances of free energies – the energy it takes for the protein to shift, and the energy differences between the various forms once the shift(s) have taken place, which include the interactions with the various ligands. The tradeoffs and payoffs of these sorts of movements are the nuts-and-bolts, assembly-language level of ligand binding.
And it has to be emphasized that this is a very simple case indeed. No polar interactions at all, no hydrogen bonds, no water molecules bridging or being displaced from the protein surface, no halogen bonds, no pi-stacking or edge-to-pi stuff. There are also, it has to be noted, other ways for proteins to deal with such small changes. The authors here, in fact, looked through the literature and the PDB for just such series to compare to, and found (for example) that the enzyme enoyl-ACP reductase (FabI) doesn’t take on such discrete states – instead, a key residue just sort of smoothly slides into a range of positions. That said, they also found examples where the behavior is more like the mode-switching seen here.
If that’s common, then calculating ligand binding gets more complicated, which is not what it was needing. These are about the smallest and least substantial ligand changes you can come up with, and here’s a protein shifting around quite noticeably between an ensemble of low-energy states to deal with them. The problem is, there are a huge number of such states available to most binding sites, and distinguishing them from each other, or from the original binding mode, by first principles is (in many cases) going to be beyond our capabilities for now.
Here’s Wavefunction’s take on these results – he says that “The conclusions of the paper are a bit discomforting. . “, and if I were a molecular modeler, I’d say the same thing!

17 comments on “Tiny (And Not So Tiny) Changes”

  1. Thanks for the plug Derek. I think cases like these pose good challenges for some of the recent FEP techniques we have been discussing which are supposed to get energies accurate down to 1 kcal/mol. In this case many of the ligands lead to a net energy change of 1.5 kcal/mol with a corresponding change in backbone conformation; both these aspects of the problem should provide a great testing field for any free energy technique worth its salt IMO. The other question is whether a technique like MD or FEP can get the distribution of discrete protein conformations right, even semi-quantitatively (an even more challenging problem).
    In any case, the good news is that problems like this will keep modelers busy for a long time in the foreseeable future.

  2. anon says:

    wait so we cant just calculate tox out of a molecule with design! preposterous! time to call watson!

  3. luysii says:

    [ Proc. Natl. Acad. Sci. vol. 110 pp. 9344 – 9349 ’13 ] The binding pockets of naturally occuring proteins fall into about 400 classes. The same is true for single domain proteins which were computationally generated to be compact. So pockets are a spandrel — a byproduct of compact protein structure (which can’t be fully compact). Natural selection can then act on these pockets to tune them for specific ligands.
    So — Ligand binding by proteins is an intrinsic property of proteins and not due to evolution
    Interesting !

  4. MMG says:

    I’m not sure this is quite as ‘disconforting’ as one might think. As anyone who has done docking or modeling has probably realized, a crystal structure only represents a single snapshot of a protein’s structure. Proteins are highly dynamic and assume many different conformations over time. I see this as highlighting the significant limitations of rigid receptor docking/modeling algorithms. I hope studies like this compel the computation community to progress onward toward MD and FEP like Wavefunction mentioned. Admittedly these are nascent technologies (still) but now is the time to really begin refining them and improving their accuracy as they offer a much better idea of how dynamic a protein can be in vivo.

  5. anon says:

    @luysii:
    “So — Ligand binding by proteins is an intrinsic property of proteins and not due to evolution”
    nothing about that summary you provided says that ligand binding isnt influences by evolution… not sure where you think your going with that claim though, but its not supported by the evidence supplied.

  6. Anonymous says:

    So, how far are we from accepting that computational models might be able to explain some experimental facts, inspire new hypotheses, and make great slides, but claiming they have predictive power is not realistic?

  7. luysii says:

    I guess I should have reversed the last two sentences in what I wrote.
    Here they are reversed (with a bit more)
    So — Ligand binding by proteins is an intrinsic property of proteins and not due to evolution
    Natural selection can then act on these pockets to tune them for specific ligands. (If that isn’t said to be the mechanism evolution, what is?).
    That pockets exist is due to the inherently imperfect packing of side chains in the interior of a protein. They exist because proteins exist.
    Hope that clears things up

  8. Anonzy says:

    #6: I think prediction is overrated, especially in drug discovery. Most success in drug discovery comes from pursuing well or ill defined hypotheses from a wide variety of disciplines and thinking strategies, flashes of inspiration, brute force approaches and just dumb luck. I don’t think *anyone* in drug discovery can truly make reliable predictions that result in the discovery of important new therapeutics. Even if computational methods can help me narrow down possibilities and generate hypotheses to test and do nothing else, I will take it.

  9. Some idiot says:

    #4 I think you are on the right track with the “snapshot” thing here… One thing that would be fantastic here (and probably well beyond our technical abilities at the moment, for a variety of reasons) would be to get NMR solution structures of such complexes, and the n pin down the relative mobilities of the different parts of the pocket. In the presence of the different ligands… Is that “ratchet” real, or just an artefact of a crystallisation process trying to sample real, dynamical life?
    Although probably impossible, it is a useful thought experiment.
    Great bit of work these guys have done… Solid data on the table.

  10. David Mobley says:

    Just a quick comment re Curious Wavefunction’s remark – we’d actually looked at similar helix motion in one of of the lysozyme binding sites with free energy calculations (what some call FEP) some years ago and found that we weren’t sampling the motion of this helix on simulation timescales (which at the time were fairly short). So my answer to #1 at this point is that, “when we tried it, no.” I did already point the paper out to my group, though, and we would like to look at this as this is exactly the kind of thing which is (a) going to be important a lot, (b) going to happen a lot more than people like to think, and (c) pose a lot of interesting and important challenges.
    The naive solution to “we don’t see that motion in simulations” is, of course, to simulate longer (basically it’s a somewhat “slow” motion for us, we can tell), which may work (and we can try). But we’d often rather be more intelligent about things and find a better way to deal with it than just brute force.
    So anyway, I agree with #1 that this is an important thing for us to explore with simulations/free energy methods.

  11. Anonymous #6 says:

    @ Anonzy,
    I don’t disagree at a high level with your views. Nobody knows for sure how to discover drugs, starting with me. But I don’t think that means all inputs have equal value.
    IMHO, as an example, if an in vitro binding assay tells me compound A has 100 times higher affinity than compound B I see that as more reliable information than when a computer simulation tells me that.
    Whatever we do that helps us think creatively, build hypotheses and test them is welcome.
    Just thinking out loud…

  12. julien michel says:

    @10 yes agree fully, that must be part of the next level of validation. I guess time for the free energy crowd to revisit T4L !

  13. JFT says:

    Is anyone surprised though? I have always thought crystal structures of proteins/enzymes are more a guide than actually useful. You are crystallizing a protein first-proteins don’t pack like that in vivo. Then you are settling on the conformation that freezes out- is this the lowest energy form? Then you are ignoring hte fact that these are highly dynamic structures that are constantly moving, sliding, shaking, adjusting. Then if you put a ligand in there you get the lowest energy form-which is what it would look like after reaction and before ligand dissociation- this is quite different from what it can look like at other stages of the reaction. And God help you if it has a helper protein or some other allosteric co-factor required for conversion.
    So, is anyone really surprised? These things are very complicated, and readily adjust and change to their environment-a methyl for hydrogen substitution is not negligible. If it was, there would be no need for Alanine in place of Glycine.

  14. Anonymous says:

    @13.
    Ask medicinal chemists working with structure based design.

  15. lluysii says:

    The floppiness of proteins continues to amaze us. For an example of this currently occurring in every neuron you posses see https://luysii.wordpress.com/2015/06/21/the-uses-of-disorder/

  16. Rich Rostrom says:

    As a computer guy, I found it amusing and interesting that you used the metaphor “assembly-language level” quite casually. Not only do you understand it, you expect your readers to understand it.
    Not what I would have expected in writing for an audience of chemists.

  17. Derek Lowe says:

    Hey, we’re an erudite bunch around here, amirite?

Comments are closed.