Skip to Content

Analytical Chemistry

The Electrons Continue to Beam In

I had the chance yesterday to attend a one-day symposium on Cryo-EM (and MicroED) techniques here in Cambridge. The whole thing was co-hosted by ThermoFisher, whom I gather are having a glorious time selling these instruments and want to extoll their virtues as much as possible, and by MIT. It helps that there are a lot of virtues to extoll. I’ve been writing about these techniques here on the blog as they’ve been developing over the years, and what we’ve been seeing is the rise of a completely new analytical technique – first slowly, then quickening, and now moving at a really impressive clip. Some of that can be seen at the statistics page of the EMDB. That cumulative count of maps released is on pace to keep up the exponential growth shown so far: there have been 761 released so far this year (as I write), and that alone is more than were released in all of 2015. If there really are >2200 maps added this year, as it looks like there will be, the increase during 2019 by itself will be more than the total cumulative number that existed at the end of 2012. The increase in resolution of the deposited maps is impressive as well – the higher-resolution ones used to be a distinct minority, but the gap is closing rapidly.

Last fall I mentioned the applications to structure-based drug design, noting that at the time the review under discussion was written, that there were only five (total) cryo-EM structures at or below 3 Angstrom resolution with small molecule ligands in them. But I just went over to the EMDB and counted four released so far this year (an example), with still more if you count back to the publication date of the review itself, and my impression of yesterday’s meeting was that there are a lot more coming over the horizon. The idea of looking for such ligands has gone from “Don’t bother” to “Let’s try it” and is moving smartly towards “Sure, why would you even wonder” territory. It’s not there yet, but as this equipment gets more capable and more widely distributed, and as people get more experience with it, that’s where we’re headed for sure.

Also last fall, I wrote about the related MicroED technique, which uses the electron microscope equipment to do diffraction instead of imaging. Its applications to things that crystallize are immediately apparent, because the size of the crystals needed to produce a structure are ridiculously small. I had the chance yesterday to hear Tamir Gonen of UCLA (a leader in this field) speak on their recent results and it was quite impressive. Here’s a recent publication from the group to give you the idea. Gonen mentioned that some of the protein crystals that they’ve been working with are only about ten protein molecules on a side, and they’re pulling up to two-Angstrom resolution structures out of them. Try that with X-rays, if you want, but prepare for disappointment. Here’s another paper from the group from last year, showing a 0.75A structure (!) of a prion protofibril, and it has structural features that are invisible to X-ray crystallography in any form, including some electron density in the hydrogen-bonding network that is hard to explain at all with existing models. The small-molecule ED work described in that October blog post is continuing, too, an area that I’m following with great interest.

So where is the field going? One theme that kept coming up was the need to get all of these techniques out of the “artisinal handwork” mode that they’ve been in. Gonen is setting up a center at UCLA that’s trying to automate things as much as possible, and that same thought came up many times. Sample handling and tracking, evaluating particles on the grid (for cryo-EM), doing tiny crystallization experiments on the grid itself (for microED), making sure that you’re collecting useful data in general and not just shooting junk, automatically processing as much of the real data as possible: all of these need help. But my impression is that these problems, though not solved yet, are actively being hammered on by the best practicioners in the field, and there’s no reason that things can’t get a lot easier than they are now.

As Gonen himself put it, though, “I don’t want you going away from this saying “Oh, everything’s done, everything’s peachy, Tamir has it all figured out, we’re going to go get all our structures in an afternoon” We’re not there yet. But what’s been accomplished so far is startling, and there’s a lot more to come.

I should mention, for people in the Boston/Cambridge area, that you’ll have another chance to hear about this stuff soon. The MassBio folks are planning an event on cryo-EM in drug discovery the morning of July 9 (8-10 AM) at their venue in Tech Square. Registration will be open on their web site on May 1, or you can email them for more details.

16 comments on “The Electrons Continue to Beam In”

  1. MrXYZ says:

    I heard a talk several years ago about setting up a similar sort of facility at Genentech in South San Francisco. A very impressive talk with similar goals, particularly with regards to automating the process. The discussion of how you design such a facility on top of an earthquake zone was interesting.

    1. Derek Lowe says:

      Yeah, MIT has just opened a new building dedicated to this kind of thing, and they seem to have put a lot of effort into reducing the vibrations in the whole place. I gather that having the Red Line rumbling along underground doesn’t improve things, but at least we’re not on a fault line (that I know of) around here. . .

      1. Icefox says:

        No major faults I’m aware of, but you’re still moving about 2cm/yr no matter where you are.

  2. At the CHI DDC meeting a couple weeks ago speakers from both Astex and Merck mentioned that they are using Cryo-EM for FBLD, and showed some pretty impressive maps for fragments.

  3. Mrinal Shekhar says:

    I wish in the meeting they had discussed on conformational heterogeneity in the protein sample and how to leverage the information of protein dynamics from CryoEM. The fact that practitioners discard a large amount of data, calling it as junk, without ever characterizing as to what junk really means bothers me.Any comments?

    1. Wavefunction says:

      At the very least, that discarded data may provide a good dataset for machine learning.

      1. John Wayne says:

        This is both hilarious and a good idea at the same time.

    2. AGMMGA says:

      Short answer:
      Currently we discard “junk” because we cannot extract any information from it and it would reduce the resolution of our maps. Methods are being developed that will hopefully solve this.

      Long answer:
      For protein structure, cryoEM relies on taking millions of images (projections) of particles in random orientation (i.e. viewed from all possible angles), and then reconstructing the 3D shape. Unfortunately, each single particle has a very poor signal to noise ratio, so current algorithms rely on classify + average algorithms to boost the signal.
      In brief, (1) group all particles by similarity (with some algorithm), (2) superpose similar particles together (rotate+translate) and (3) average them.
      If the particles being averaged are similar, signal is boosted and noise is cancelled. If the particles are “not similar” or not correctly superposed, the averaging step will reduce the signal (i.e. your structural information) and result in (parts of) the electron map having lower resolution.
      In this context, “not similar” can mean that the particles in the class are simply broken by the freezing process, are not viewed from the same angle, are not made of the same subunits, or represent different conformational states of the molecule.
      Broken particles are very different from each other, so we discard those in the very first steps. All of the other “similarity” problems can in principle be solved by a more accurate classification step. The catch is that (1) particles representing rare conformations will not be usually detected by the classification algorithm as being worthy of their own class and (2) we need quite a few particles per class to get a good signal boost in the averaging step.

      Rare conformational states of the protein (rare meaning less than maybe 10% of the total particles) will be difficult to detect because of (1) and will anyway yield low resolution maps because of (2). Therefore, at present, we tend to discard these so that they do not “pollute” the maps of the more common conformational states. We do not discard all protein dynamics information, though: common (maybe >30% of total) conformations are routinely detected at the 3D classification step.

      As algorithms are developed that rely less on classification + averaging approaches, and as detector technology improves signal to noise in our images, we will be able to work with protein dynamics more and more. As an example, Vendruscolo et al., Biophysical Journal 114, 1604–1613, April 10, 2018 is the first result that pops up on google for “cryoEM conformational dynamics”

  4. em says:

    regarding microED, very few people can get this to work. In the groups that do get it to work, only a few scientists can do it, and they are very cagey, and protective of their process. Makes the whole thing seem very suspicious.

    1. HX says:

      Welcome to Stockholm and we will share all the techniques and tricks with you.

      1. em2 says:

        microed should only be worthy of Stockholm only if it delivers on its theranos-like claims of a super-easy and straight forward method to get high rez structures from tiny crystals using cheap scopes.. If only one or two people can do it (if they really can do it), there is limited applicability and just hype.

        1. em3 says:

          Dear em2,

          I believe the above poster is not implying that MicroED already warrants the Nobel Prize by invoking “Stockholm.” Rather, he is saying that there is a group literally based in Stockholm at Stockholm University (X. Zou) doing MicroED who are apparently willing to share all their secrets with you.

          With utmost sincerity,
          em3

        2. HY says:

          We are a group based in Stockholm trying to make 3D ED methods easier for everyone to use. We would love to spread the methods to other labs. I agree that some of the claims made by others are a little exaggerate. ED data is not as accurate nor as precise as X-ray data yet. However, electron as a radiation source for diffraction, does have its unique advantages. ED will become a complementary method to XRD, cryo-EM, XFEL and neutron diffraction. This is why there are many groups (other than the UCLA team) working on the development of 3D ED based methods.
          I literally mean you are welcome to visit us in Stockholm.

  5. Mrinal Shekhar says:

    @AGMGA. I think one really promising algorithm is MAP manifold based continnum classification ( for the lack of better terminology)
    https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4848141/

  6. HY@SU says:

    It is great to see that Gonen’s Lab and friends have spread the message to a wider research community and caught many attentions that are well deserved for the development of electron diffraction techniques. Our goal at Zou’s lab has always been to develop electron crystallography based methods into an user friendly package for chemists and biologists to use.

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.