Skip to main content

Analytical Chemistry

How Close is Cryo-EM To Riding Over the Horizon?

OK, all this cryo-electron microscopy stuff is great, new protein structures, things that are huge, that can’t be crystallized, fine, fine: but when, the medicinal chemists in the audience ask, will we be able to use it for structure-based drug discovery? This new review tries to answer just that question.

The first thing that you think about is resolution. The closer you can pin down the location of every single atom around the binding site, the better the shape you’re in for SBDD. As it stands right now, according to the authors, there are 672 structures in the EMDB archives at 4Å resolution or better. But only 75 of those are better than 3Å, and only 12 are at 2.5Å or better. And there are only five structures below 3Å with small molecule ligands in them, so there’s the state of the art. Here they are: paromomycin with the Leishmania ribosome, GSK2837808A with lactate dehydrogenase B, resiniferatoxin with TRPV1, UPCDC30245 with p97, and PETG with beta-galactosidase. And there’s only one paper (mefloquine with the Plasmodium 80S ribosome, from last year) that shows use of cryo-EM structural data to propose and test new compounds, although you’d have to think that this is happening industrially, outside the current range of open publication. But by any standards, it’s early. Very early.

The paper identifies some bottlenecks that are going to have to be widened for this technique to become a regular part of the process. Sample preparation and data collection are the first. As it is now, cryo-EM is largely a handcrafted art form, especially at the high-resolution end of things. In particular, my impression is that collecting data to get down to below 3Å resolution can really send you out on the long tail as you fill in gaps in the data set. Significant automation needs to be brought in for higher throughput (people are actively working on this), and there need to be improvements in both the processes and the hardware in order to increase the success rates at the same time.

Once you get past that, the next barrier is working up all those data points. As it stands, say the authors, the best software (while semi-automated) is far from being able to generate structures without significant human oversight and intervention. It’s another art form at this point, and it needs experienced people with good judgment at the controls. The progress of instrumentation, though, has always been to make such people less valuable and necessary for any given machine, which seems a bit perverse until you consider what short supply they’re in. Work is underway in this area as well, to try to make things as robust as X-ray data processing has become, but it’s not there yet.

The authors also emphasize a key point: making the whole process faster is not only useful, it’s essential if cryo-EM is to have the maximum impact in a drug discovery organization. There are many, many techniques out there that started out being able to provide interesting and useful information, but too slowly to have any effect on actual drug projects. Some of them picked up speed and are now able to catch the bus (as it takes off in a cloud of dust), and others haven’t, or not yet. But projects move right along (or should), and if you’re always running behind the crowd, yelling “Hold on, guys! I think I found something else!” you’re not going to be as effective as you could be.

Overall, though, rapid progress is expected. The experience of X-ray crystallography helps a great deal: we know what high-thoughput systems look like in that (related) field, so we know what to aim for and what the differences and weak points probably are. And the demand for this technique is huge: a really robust cryo-EM setup would be a huge help in drug discovery and would illuminate binding events for targets that are almost complete black boxes as things stand now. Structure-based methods are not the instant answer to all your problems, of course, but they’re still pretty damned useful, especially for the kinds of proteins that cryo-EM can target. So speed the day. Revisiting this topic in a couple of years should show some real advances.

13 comments on “How Close is Cryo-EM To Riding Over the Horizon?”

  1. Mobio says:

    My sense in examining many Cryo-EM membrane protein structures closely is that even with ~3.5 A resolution, key side-chains in binding pockets are frequently not explicitly defined (missing in maps) and may have been modeled instead in the final PDB file.

  2. hn says:

    At that resolution (3.5A) you can’t be that confident of your small molecule binding mode. Cryo-EM will be more useful for antibody drugs.

  3. Anonymous says:

    Derek began with “… pin down the location of every single atom around the binding SITE” but ended with “… illuminate binding EVENTS”, emphasis added. Many proteins and complexes can show detectable structural changes highly remote from the actual binding site. Those events are often the important ones, too. A flap that doesn’t flip, a pore that no longer pours, an unhinged hinge, etc.. Those events can provide insight into mechanism of action and other ways to target the function other than at the original binding site.

    In addition, many small molecules are not bound to the outer surfaces of receptors or targets where they might show up as useful bumps in the Cryo-EM. Many are embedded deeply in the interior folds at the native binding sites or other interior allosteric binding sites. I think those would remain invisible to Cryo-EM.

    I am mostly thinking back to the use of Cryo-EM to provide structural info about huge complexes (motor protein complexes, etc.), viruses, ribosomes, etc. that could not be obtained any other way. I am less “up” on the use of Cryo-EM for smaller X-ray competitive structures. But I know that they are doing some amazing things with Cryo-EM.

    1. Brian says:

      I agree with your point to use CryoEM for the study of macromolecular complexes that elude structure determination by crystallography.

      An issue that I see with CryoEM arises in data processing. You pick particles, assign them to particular classes, then use 2-D classes for a 3-D reconstruction. (Albeit, this is a crude overview.) The problem specifically arises in the process of picking particles; if you don’t have a completely homogeneous population of protein-inhibitor complex, you will be using both protein with and protein without inhibitor bound for your reconstruction. The exceptions could be if the binding of compound induces a conformation change that can be detected on micrographs, or the compound blocks a protein-protein interaction (so, forgive me for the term, you’ll see two “blobs” instead of one).

      Another issue comes about with resolution. You’ll need a large, symmetrical protein and/or complex to approach sub-3 A. I think most would agree that you’re going to be limited in doing some structure-based design. Perhaps you can get one or two protein-inhibitor complexes from CryoEM, but you’re not going to churn out multiple protein-inhibtor complexes as you could from your crystallography set-up.

    2. Oli Clarke says:

      Respectfully, this is not how cryoEM works – one does not require the molecule of interest to bind to the exterior of the target protein, any more than one would require such for X-ray crystallography.

      Each particle image is a projection of the density – a transmission image – not like a photograph, which I suspect is where the confusion arises.

      Small molecules and ions can easily be identified in the 3-4Å range that has become fairly commonplace, although limited resolution often makes it difficult to unambiguously assign the binding mode.

      The highest resolution cryoEM structures are now in the 1.5-2Å range, and once we approach that range more routinely the technique will be vastly more useful for structure based drug design.

      Where cryoEM really comes into its own is in the characterization of heterogeneity – identifying different ligand bound or unbound classes in a single dataset. This is a major advantage for some projects.

      Throughput, however, is still a limitation – collecting (and processing) cryoEM data is not going to approach the speed of X-ray data collection and processing for the foreseeable future.


  4. Kelvin says:

    What if protein structures are not the rate limiting step in drug discovery, nor even screening, med chem and lead optimization, but finding viable new drug targets that have not already been fully exploited? How much difference could any improvements in cryogenic-EM and other techniques and technologies really make if there are no viable new drug targets left to exploit?

  5. curious says:

    Almost all big pharma have set up their own Cryo-EM capability. I would like to hear from their medicinal chemists or anyone with knowledge. How useful has it been for them? In which way it helped their project team?

    1. anon3 says:

      As stated above, its useful for antibody design but not structure based med chem.

      1. curious says:

        Thanks! We are considering. I have my doubts, but want to be open-minded about it.

  6. Wavefunction says:

    Has anyone tried to use data from cryo EM and used it to improve local x-ray models, essentially through a bootstrapping process?

    1. Nat says:

      Don’t you mean the reverse (improving EM structures with Xray models)?

      1. curious says:

        Both can be sort of true. People are using EM envelop to help solve the phase problem in crystallography.

  7. Robert-S says:

    Hi Derek,
    I have been following your site for quite some time now. Very useful and intersting. I enjoy reading the comments. It is obvious very experienced people in the DD business visit and contribute.
    Regarding the applications of Cryo-EM in structure-based drug discovery, please have a look at the paper we just published in one of the ACS journals:
    Our experience is that, even if the resolution is not be great (3.5-4 Angstr.), one can still derive very useful information to guide medicinal chemistry efforts if the right tools and a lot of compute resources are available.


Comments are closed.