In this episode of Expert Insights for the Research Training Community, Dr. Bridget Carragher, co-director of the Simons Electron Microscopy Center, discusses the current and future applications of cryo-electron microscopy, or cryo-EM, and training opportunities for researchers involving cryo-EM.
The original recording of this episode took place as a webinar on May 14, 2020, with NIGMS host and director Dr. Jon Lorsch. A Q&A session with webinar attendees followed Carragher’s talk.
Recorded on May 14, 2020
View Transcript Download Recording [MP3]
Announcer:
Welcome to Expert Insights for the Research Training Community—A podcast from the National Institute of General Medical Sciences. Adapted from our webinar series, this is where the biomedical research community can connect with fellow scientists to gain valuable insights.
Dr. Jon Lorsch:
Hi everybody, looks like it’s right about time to start. I’m Jon Lorsch, director of NIGMS.
I want to thank all of you for attending this, which I think is our fifth webinar series for trainees. We started this series in order to support trainees—both those supported by NIGMS and also others throughout the country and throughout the world—during these difficult times where many of you are working from home, aren’t able to go into a lab and do what you would like to be doing, and we’re trying to find ways to make this time productive and interesting as well for you, even though you can’t actually be in a lab conducting science.
We also are going to have a question-and-answer session after Dr. Carragher gives a fairly brief presentation, where we want to get questions from you and have Bridget answer those. So thank you very much, Bridget, for coming.
Let me just introduce her. Bridget Carragher was born in South Africa. She had her education at the University of the Witwatersrand in South Africa in the field of physics. In 1982, she graduated from Northwestern University in Illinois with a master’s in physics and started her PhD at the University of Chicago, where she defended in 1987.
She has had a variety of private-sector and academic positions, including as the chief operating officer of NanoImaging Services, a company that she cofounded, and as a professor at the Scripps Research Institute. Together with Clint Potter, she is the director of the National Resource for Automated Molecular Microscopy or NRAMM. She’s the director of the Simons Electron Microscopy Center at the New York Structural Biology Center, in New York City, and the principal investigator of the National Center for CryoEM Access and Training, which is an NIH-funded cryo-EM service center that she will tell you more about. She is also an adjunct professor at Columbia University. She has numerous publications and numerous patents and numerous awards.
And so with that, Bridget, thank you so much for doing this and I’ll turn it over to you.
Dr. Bridget Carragher:
Thank you so much, Jon, for that kind introduction. So a quick, 15-minute introduction to cryo-EM.
It’s come a long way since 1985, when it sort of started, and now it’s definitely structural method du jour. And in the future, I think it’s got even further and much more exciting places to go.
So why should you care about cryo-EM?
One reason is because it’s become very much a structural technique to pay attention to. Over the first many years of cryo-EM lifetime—you can see we started in 1985. Not until about the year 2000 did we even have a structure in the PDB, and now it is growing exponentially and it’s on track to catch up with X-ray in the next five years or so. But it’s on a very steep exponential growth.
Another reason you might want to care about cryo-EM. These should look familiar. These are, of course, the infamous SARS-CoV-2 spike proteins. These were two cryo-EM papers that came out three weeks after the gene was delivered to these people. Three weeks from gene to structure is very, very rapid.
I think it’s fair to say without cryo-EM we wouldn’t have this knowledge, and I’m going to talk a little bit more about that at the end, because we are certainly paying a lot of attention to COVID-19 projects at the moment.
So where it all started was about 1985, at least for many of us. This was the very first 3DEM Gordon Research Conference, where cryo-EM was brand new and people weren’t quite sure even what it was, but there were three very important people at that meeting. That was Jacques Dubochet, Joachim Frank, and Richard Henderson, and they got the Nobel Prize in chemistry in 2017 for the development of this technique.
So what is cryo-EM? I think it’s actually two things at the moment. It’s single-particle cryo-EM, and that’s the thing that’s responsible for these exquisitely high-resolution structures of these macromolecules and macromolecular complexes. It’s also cryo-electron tomography, sometimes called in situ cryo-EM, and these are responsible for looking inside the cell and examining what these molecules are doing inside the cell.
And I want to very quickly, in the next few minutes, just explain what those are and what the difference is. So single-particle cryo-EM. These slides were made by Gabe Lander very many years ago, but I think they still do a good job of explaining what it is that we do.
So we start with a purified solution of either macromolecules we care about in solution, and they are tumbling about in random orientations. We put a tiny drop of that on an electron microscope grid, which is a few millimeters across, remove most of it to make it into a thin film, and we drop it very fast into a liquid cryogen which snaps that thin film into a solid state, trapping the molecules in this vitrified sample. That means it’s not crystallized; it’s just like solid water. We can then take that layer of vitrified ice, and we put it inside a transmission electron microscope—that means the electrons are transmitted all the way through the sample to form a projection image, and that’s what we’re imaging.
And so this is more or less what those images look like. What you see is a bunch of shadowy images of your particles, and they look different one from the other. And that’s simply because they happened to be snapped into the solid state while they were tumbling around and then they are presenting different views of themselves. And so now the electrons go all the way through these different orientations of these agents and form these projection images below, and that makes these different-shaped projection images.
You’ll also notice it’s not a great image, it’s very grainy, and that’s because we can use very few electrons in order to take this image because our biological samples in this vitrified condition are exquisitely sensitive to electron radiation. So can’t use much radiation to take pictures of them; it’s as if you opened the shutter on your camera for a very brief instant to get a grainy image, not a very well-developed one.
And if we opened the shutter for longer on the electron microscope we’d simply cook our protein and damage it. But we can get around that by using averaging. While there’s lots of these projection images in the same orientation, what we do is we chunk them out of the image.
Now I’ve got these individual projection images of particles, and we rearrange them until they are all lined up with each other and add them together and that improves the signal to noise. And that’s because when we add noisy images, the signal is always adding—the noise is sometimes positive, sometimes negative, so over the course of the averaging, the noise goes away, and the signal improves.
So we can do it for that one projection view. We can do it for all the other projection views. And we know if we have a lot of projection views of a three-dimensional object we can combine that using tomography methods to form that 3D object—and that’s just what happens in a CAT scan or an MRI scan. You take lots of projection views of a 3D object, and then you can put them together without physically slicing up the sample.
So that’s single-particle cryo-EM. What about cryo-electron tomography?
Sometimes called that, sometimes called in situ cryo-EM, sometimes called cellular cryo-electron tomography. And that comes about when you say, I do want to know about these molecules, but they’re inside a cell. And a cell’s a very large object. It’s not a bunch of molecules tumbling around.
We can’t reduce it to a thin film by blotting away most of the liquid. So in order to vitrify cells that are large, we have to use a fancy device called a high-pressure freezer, and that’s exactly what it says. It uses high pressure to very rapidly freeze something so it’s still vitrified, it’s still perfectly well preserved, but we can do it on much larger objects then just dumping it into the liquid cryogen.
So now this is what it looks like when it comes out of the high-pressure freezer. It’s a big clump of cells. But how are we going to find what we need to find inside that big clump of cells? And what we do then is we use a cryo light microscope.
So we take this clump of cells that has been vitrified in the high-pressure freezer and we put it in a cryo light microscope, and if it’s been labeled in the appropriate way, we can find these things we’re interested in. But that thing is very, very large. It is not suitable for transmission electron microscope where electrons have to go all the way through, so it’s way too thick.
So how do we thin it down in order to make it suitable for transmission electron microscope? We put it in another fancy instrument called a focused ion beam scanning electron microscope. And what that does is use the focused ion beam as a knife. It cuts away most of the sample, chisels it away, leaving only a thin window of your sample left. And this is a nice little animation made courtesy of Thermo Fisher.
What it looks like in reality is something like this. This is a bunch of yeast cells that were high-pressure frozen and then taken into this focused ion beam, and we chiseled away everything except this thin layer of yeast cells. Now that’s thin enough, ready to go into the TEM. We can stick it in the TEM, but we need three-dimensional views of this, so what we do is we tilt the sample. We tilt it to an angle, take a picture, tilt it to the next angle, take a picture, tilt it to the next angle, take a picture, which is very similar to what happens when you get a CAT scan.
And then this tilt series is shown below, and we can take that because we have projection images now for many different views, we can turn that into a three-dimensional volume. And these are very complex-looking volumes. Here’s a typical one. There’s a lot of grainy, noisy stuff there, but you can see things. There’s a nuclear core complex, the gateway between the cytoplasm and nucleus. Here’s some microtubules cut end on, sliced through. Here’s a ribosome in the cytoplasm. And here’s a ribosome in the mitochondria.
But you can imagine it’s quite a hard job to find these, hunt them down, and see them, and we can’t see much because, again, we had to use very low doses of electrons—otherwise we would have cooked the sample. Or what you can do is pick many, many, many, many of them and average them together, just like we do in single-particle cryo-EM and then you can get a better view, not this grainy view but an actual domain view of what those molecules are like inside the cell. And that’s, of course, extremely exciting.
So I told you cryo-EM developed about 1985. Where have we been for the last 30 years?
Well, prior to about 2012, 2014, we did a lot of cryo-EM, but it was described as blob-ology. It was very low resolution for the most part. And that’s because we didn’t have very good cameras. Then around 2012, 2014 came along this next generation of cameras.
And they were good for two reasons: They directly detect electrons, not photons, so they are much better for us. And they also allow us to acquire movies so we can take out any blur in our images as we are acquiring them. Just like your iPhone does. It goes snap, snap, snap, and it deblurs things for you and you get these nice, crisp images. And the cameras did the same for us.
And as a result of that, cryo-EM entered the near-atomic structural realm. So now on our very best specimens we get sub-2 angstrom resolution. That means you can see every side chain, you can see well-ordered water molecules, and I think somebody has even got this perfect sample. This is the lysozyme of electron microscopy, but this is now at something like 1.2 angstroms, so it’s a very high-resolution technique for a well-behaved sample.
So the resolution revolution, as we’ve called it because it was so revolutionary for us, a lot of credit goes to this new generation of detectors, and they well deserve that credit. But so, too, do the fabulous microscopes that we use now.
These are exceptionally stable, robust, well-behaved instruments. We can put samples into them and collect data for days and days and days. They really are well behaved. They produce beautiful data on every day of the week.
We also have a lot to thank for computers, and that’s nothing to do with our field, that’s hundreds of millions of gamers all over the world reducing the cost of computing and especially GPUs, and really the cost of computing these structures that we have to do all that alignment and reconstruction, that reduced by thousands of times over the last few years.
So overall in the last few years the whole field became better, faster, and cheaper. You hear this better, faster, cheaper thing in industry all the time, and certainly that’s exactly what happened to our field. The microscopes got better, the cameras got much faster, and the computers got much cheaper. And along with that came a ton of really excellent software that allowed us to control these instruments and do the analysis.
And the software is great because it allows us to collect enormous, huge data sets very automatically. It then allows us to sort out millions of images and pull out the few percent that are really the good ones and that go into the map. It also allows us to sort out from one data set many, many different structures.
One of the huge advantages of cryo-EM is that we can find many structures in the same sample. We don’t have to turn them all into identical copies of themselves into a crystal. We can have a look at the conformational landscape, if you like. OK, so that’s what cryo-EM is.
I’m going to stop in a minute, but I’m sure one of the questions you’re going to ask is where you can get cryo-EM access and training. And as Jon mentioned, thanks to the foresight of NIH and the Common Fund initiative, we now have three major cryo-EM access and training centers in the country. One of them is in New York—that’s the one I help direct. There’s another one in the Pacific Northwest and one at Stanford. And aiding those are four other centers that are developing training materials. So let me tell you a little bit about our center. It’s called the National Center for Cryo-EM Access and Training. And what we provide is four KRIOSs.
These are very expensive, very high-end instruments, so they work very well in a major center. We have screening microscopes so we can train our users as they come in to get training, and then we have a bunch of fancy vitrification devices, etc. And we do a lot of onsite training. we are doing virtual training at the moment, and then we do a lot of providing access to these high-end instruments for people who have samples ready for structure.
Right now, as you might imagine, the center is on pause because of New York, but we are allowed to do essential work, and essential work is anything to do with COVID-19. So we have 10 projects on the go at the moment. Many of those have to do with examining the trimer complex, the infectious machinery of the virus, and seeing what antibodies are attached to it.
Some of the work has to do with examining what drugs will bind to the polymerase that’s in the virus, and there’s a few others as well. So we’ve 10 of these on the go. We’ve done 28 data collection sessions, many, many maps have been produced, and there’s information starting to come out about how these antibodies are interacting with the trimers. Also just wanted to mention we are doing a lot of remote cross-training at the moment, and if any of you are interested, these sessions are provided all the time.
There’s anything from one-on-one office hours if you have an ongoing project and need some help. Or just general roundtables to discuss things. And you can see of the 34 people who have joined in so far, many of them are beginners and they are from all stages of the work. So we welcome you to go to this website and sign up for any of these training things. And of course once we get back online, that will be onsite as well.
So that’s all I had to say. This is the amazing group I work with in New York. Lots of exceptionally good research scientists and technicians that keep all of those machines running absolutely perfectly. A lot of people developing new technology to automate things and develop new ways of going about this. And then lots of IT staff. And I’ll stop there and be very happy to answer questions.
Thank you, Bridget, for that fantastic talk. That was really a tour de force.
Just to emphasize to everyone out there that training is a key aspect of the centers that Bridget mentioned, so I hope that you all think about availing yourself of that, and we’ll probably talk a little bit more about that in a minute.
Bridget, why don’t we get started by a question about you and how did you get involved in cryo-EM research? You’ve been in it since 1985, early days. How’d you get there?
Even before that, as it turns out. I was a physicist, and then I switched to biophysics, and the first thing I came across was electron microscopy. It was the very early days.
This was even before that first Gordon conference. I got fascinated by it. It’s very visceral. You get images and you get to process them and then you get something from it. You get a structure from it. So I started at the very earliest time you probably could start this field. I was at that very first Gordon conference. I went to every other year, we only had them every two years in those days, went every other year for years and years and years. So I’ve been in it forever. And I love it.
That’s great. What advice do you have for a trainee or postdoc who wants to get into cryo-EM? They know their project would advance, but they have no idea how to do it.
Well, I can’t say enough how valuable these new centers are that the Common Fund has funded. It was a massive problem before that. People were always wanting to get into this, and individual research labs could do it on an occasional basis as a favor or as a collaboration, but now these centers are simply available to you.
You want to get trained in cryo-EM; you send in an application. The applications are reviewed because we want to make them very transparent and open. So the applications get reviewed and then you get time allocated. And we have various forms of training. Some of it’s workshop, but some of it is embedded training.
That means you come and live with us for a few months, get completely embedded into how it is to do cryo-EM in a lab—how to make samples, how to find out whether they are any good, how to take images—and all of that, after about three months you could go home and use it.
We have several examples of embedded trainees who’ve come, done beautiful work, now they’re back at their own universities teaching everybody else. That’s really what we want. We call it cross-training, actually, because we want those people to go back and take that information back to their centers.
So I would encourage you to talk to the centers. Talk to us. We’re happy to have a meeting one on one telling you where we think you’re at with your project and where it might be ready to go to next.
This is something someone could start even now working from home. They could contact you.
Exactly. We’ve had a lot of very interesting one on ones where people are showing us their data, and some of it is great data. They are agonizing, is it ready? Yes, it’s ready. People don’t really know what they’re looking at, but we’re very happy to have those consultations and tell you where we think you are and how you should do it.
We also are going to start virtually training prior to the whole country opening up. The embedded training is really you working side by side with a scientist who’s been doing it for 20 years. We’re going to see if we can do that with a GoPro and a cell phone—a GoPro at our site showing you what we’re doing, cell phone at your site showing us what you’re doing, and then we can pass advice back and forth.
So that’s our plan to get that started even prior to the whole country opening up. So contact us and let us know what you’d like.
That is terrific. We’ve already got a lot of great questions. Some of them are technical, some of them are more general. Here’s one. What’s the difference between cellular cryo-EM and traditional tomography EM using high-pressure freezer?
There isn’t a difference. We’re terrible at naming things in our field. We call this method single-particle EM that takes millions of particles to be averaged together and we haven’t got a single particle, so I think we’re bad at naming it. But people have sort of started talking cryo-electron tomography as if it’s the whole cellular pipeline. But you can do cryo-ET, you can do electron tomography even on single particles if you like. You can do cryo-ET on anything. The cellular side, though, if it’s a big, fat cell, you have to slim it down if you want high resolution from the EM.
And that’s where you have to start cutting these windows into the cell and needing a cryo-LM maybe to investigate that. So cryo-ET is sort of the end of that cellular in situ pipeline. You cryo pressure freeze, you cut thin windows, and then you get it into the electron microscope, and you do cryo-ET.
I’ll just add to that, that a follow-on to the cryo-EM centers of the Common Fund is going to be cryo-ET centers and resources, and that will be coming soon for those of you interested in doing cryo-ET using a very similar model. If I want to visualize my protein of interest in cryo-EM, this is an interesting question, does the sample provided need to be a purified protein or can it be a cell lysate sample?
It can be cell lysate. In fact, we did a very interesting project recently at our center where somebody just lysed cells that were making viruses and we could pull out enough of them to actually get to a 3-angstrom structure.
The only difficulty with that is if there’s enough of them. A big virus-like particle, that’s great; you can find it easily. Ribosomes you can find very easily, which is why even in situ there’s a ribosome structure now at 3.7 angstroms, which is amazing.
If it’s very, very, very tiny and there isn’t much of it, you saw what those noisy images look like. How are you going to find them? How are you going to find enough of them? But having said that, I wouldn’t be astonished if we can do that sometime in the next few years, because the methods are proceeding so rapidly, and things like machine learning are coming online that are making a huge impact in enabling us to find very small, noisy things in these cluttered environments.
Great segue into the next couple of questions, actually. Very good. What do you think will be the next revolution in cryo-EM and cryo-ET? Where are we headed next?
I think exactly that, Jon, into the cell. I mean, I think sooner or later we will have solved all high-resolution structures in solution, one hopes, between X-ray, NMR, and electron microscopy, we should be able to solve them all. Then we want to know what is happening inside the cell. What happens if you change conditions, if you mutate something? Where are these regions of interest in the cell? How are things going in and out of compartments?
All of that is going to be really fascinating. And that’s super hard at the moment. There are only a few places in the world that do it well. And as Jon just mentioned, again, NIH in their wisdom has decided to fund centers to help the rest of the country get going. There will be something like five or six specimen prep centers—that’s really the hard thing—and then one big center to get cryo-electron tomography done. So that’s going to be extremely fun and interesting.
A little bit selfishly, I would say I think time-resolved cryo-EM is also very exciting. We’ve just started projects to do that. And being able to mix two things together and watch what happens in the first hundred milliseconds of interactions, I think that’s quite interesting. And just making everything better, faster, and cheaper so that everybody can do it. And it’s just a method available to you rather than a postdoc lifetime to get something done.
I keep imagining, with the current crisis, when we get to the point where we can actually see maybe in time-resolved fashion, as you said, intermediates of viral infection on a cell and then viral budding out of a cell. We’re starting to see things with phage, but imagine if we could see SARS-CoV-2 doing that.
Going back to what you said before, we have a question on machine learning. What’s the role of machine learning in cryo-EM image processing, and what areas are in urgent need—for instance, particle picking, class selection, what else? This is someone, I think, who wants to get involved maybe in that.
All of the above.
And in fact we had a fabulous, fun time in the lab the last two weeks because we had these COVID-19 projects coming in, we had these trimers come in with antibodies on them, and it was a very, very messy sample. But some people in the lab—Alex Noble, working with Bonnie Berger’s group, Kristen Becht at MIT, have had this big machine-learning project.
So they had a nice particle picker, but this new COVID-19 project drove that to excellence because we were forced to make it better because they had this very messy sample and we were desperate to get the high-resolution structures out of it, so that machine-learning program got worked and worked and worked and improved and improved until we managed to solve these structures very rapidly.
We can do them now within hours or at most a day of it going into the microscope. All of that nice improvement now went back into that program, which is in CryoSPARC, which is one of the well-used packages, and now that’s available to everyone. So particle picking certainly was one of them.
The other one, as you saw, I showed this picture of the cell, a section of the cell. How are you going to find ribosomes and microtubules and nuclear pores in that? You can sit there and pick them out one at a time, but it’s super tedious, super-high burnout, and also you won’t do it 100 times in a row because you can’t. So machine learning is going to have a huge role to play in that and being able to be good at that will be great.
And don’t forget, a lot of this stuff has been done by hand, so there is a lot of annotated data sets out there to learn from. So I’d encourage anybody doing machine learning, go into imaging. That’s where we need most of it. And we’re an imaging method.
That’s great. Let’s see. Talk a little bit about the big data aspect of cryo-EM. How are we dealing with that? Because the data sets are getting bigger and bigger, that’s going to become an issue soon. How are you thinking about that?
I think it is an issue, but it’s not as bad as we thought. We always manage to outrun it, or computers manage to outrun it. At the beginning we thought, “Oh, this will be terrible.” And then of course discs just got bigger and bigger and cheaper and cheaper and cheaper, and you can get an amazing amount of disc for not that much money anymore.
But having said that, we do not keep every movie anymore. The movies are huge. The movies can be 100 frames and then they get compressed down to a single frame by realigning everything. And long term, we store those aligned ones, and that’s easy enough to do because disc is cheap enough.
If we had to store 100 times more data than that, that would be a problem and that would become a major expense and we’d have to do something. What we are doing, though, is talking to the cloud people. If we distribute that cost, we as a center, if we had to pay for every single byte that we put into the cloud, that would be expensive, but if a user just wants one data set and they’re only going to have that and work on that for a while, that’s not so expensive to keep in the cloud for a while, so that’s what we’re…and also to process in the cloud, by the way.
So we’re working partly with NIH and other people to try and put the processing in the cloud and the storage in the cloud so people who do this occasionally and don’t want to set up a whole EM IT center to do it can just go do it and then shut it down and then come back to it whenever they want. So that’s working fairly well.
Here’s a question that’s perfect for you, as are these others. What about automation and increasing throughput in cryo-EM? What are the prospects?
Great prospects and great question. Thank you. I love that question, because our whole lives have been spent on trying to automate everything, partly because it’s burdensome on the operator and partly because you can get more control that way.
So one of the things we’ve worked very hard on the last few years is to automate the specimen preparation. And that’s still done by pushing blotting paper against the grid and it’s a bit hit or miss. So we want to automate that whole aspect. The same with the in situ, the cutting those thin sections. That’s starting to become automated, and that’s important because again it’s very time consuming and very tedious.
The data collection, we sped up data collection through automation by a factor of five over the last few years. And in case you don’t think that’s a big deal, those microscopes cost $7 million. So if you can speed up data collection by five, that means it’s like you’ve just bought five $7 million microscopes and housed them and fed them. So that’s a huge deal. And that will keep going, I think.
And then the software processing is already highly automated, but it needs to become more so in these very complicated environments—lots of heterogeneity, lots of complexity, lots of different states of these proteins—and I think automating that and understanding the sort of landscape of conformations that these proteins will go through could stand for a lot more automation too.
Great. Here’s a very specific question. Compared to other software, has the use of CryoSPARC Live made data collection faster or requiring less micrographs?
Yes, it has. We have always evaluated our data on the fly. So for those who don’t know, this is a sophisticated question. Somebody is using a lot of nice software.
CryoSPARC Live is a processing engine that gobbles up your data and does it while you’re collecting, but in our lab, we’ve always done that. We’ve always evaluated the data as it comes in and been able to make a decision. But it certainly is a good idea to make a decision while the grid is in the scope, so you should evaluate your data, say, “Is this good data? Do I need more data? Can I stop right here?”
Because our scopes are extremely expensive and extremely in demand, so you do not want to waste lots and lots of hours if the sample isn’t looking any good. So CryoSPARC Live is great. We love it. We use it.
There are several other packages as well, though, and many, many on-the-fly processing packages available, and I think they’re absolutely essential to maximize the efficiency of your very expensive instruments.
That’s great. Here’s another technical question. Why is the resolution lower at the surface of a protein while it’s higher in the core, in general?
It’s not always, actually, but if you’ve noticed that it’s probably because the bits on the outside are wiggling around, so the stuff in the core may be locked into a very rigid conformation and the bits on the outside are able to move and able to do things, but that’s not always the case.
Something like a proteasome or something that has an inner space, sometimes you get a lot of variability inside, so it’s just a matter of where things can move and where they’re not locked into rigid conformation.
Here’s an interesting question. What’s it like working and running a national research resource?
The most fun you can ever have. It really is tremendous fun. This field has been around since 1985, and I was always happy in it. But I’ve got to say the last five years has been just a complete crazy trip.
The field burgeoned out, hundreds and hundreds of people have poured into it. It’s had these astonishing successes. We never imagined a ribosome at 2.5 angstroms with every water molecule, structures of 60 kilodaltons being solved, fantastic membrane proteins, CFTR and all these other high targets for disease, for drug discovery. So that’s been fantastic.
And running a center means you have a huge amount of expertise and people all being synergistic. As I said, the other day we were trying to solve this protein. We have the people doing the machine learning, we have people doing the processing, we have the experts collecting the data, and then we have the biologists who desperately want that answer, all working together. And that sort of driving force is super fun. I can recommend it.
Apply for a job there. That sounds good.
Here’s an interesting question too. Someone’s working with a conformationally flexible macromolecule and complex. What advice to you have for this person on how to make that amenable for cryo-EM work?
First, just go take a look at it. The best thing about microscopy is you can just go and take a look. First in stain, maybe. Negative stain is a method we use, quick and dirty. And then in ice.
You may be surprised how much you can do even with the nonflexible bits. And that’s another big area of software development. People are trying to sort of examine their flexibility, capture it, unravel it so to speak. If that all fails and it’s still completely flexible, you may try to bind it to something—bind it to a fab or bind it to something it complexes with, and that might stabilize it a bit. But there’s nothing like just go take a look. It’s not that hard to do. It’s quick and you might find out something anyway.
Here’s someone who’s interested in maybe getting involved in technology development and wants to know what are the major barriers in cryo-EM that someone could get in and try to solve and make better?
So on the single-particle cryo-EM specimen prep is still a bear, and it’s a problem because of something we ignored for the first 20 years, and that’s because we have our samples in very thin layers of ice, so that means they almost certainly are seeing the air/water interface. So they bump into the air/water interface and proteins don’t like to be at the very hydrophobic interface, so we need to solve that problem.
So if you’re a hardware person or if you’re somebody who understands surfaces, if you do physics or you’re a materials scientist, we would love to talk to you. Because we think there are ways to approach this, but we really need to understand the physics, the materials science of what those surfaces look like, what is happening at those surfaces much better in order to sort of be rational about how we approach how to solve it.
So on that hardware side, that’s certainly a lot of work to be done. And then on the processing side, the machine learning is going to be critical. Speeding everything up, being able to figure out these conformational landscapes is going to be important, so it depends on where your interests lie. There’s room for everything to be done, but it depends on what you like to do.
That’s great. Here’s a question again about how do you get started. Say you’ve got a sample; you think this might be amenable for cryo-EM. You’re not sure, you have no experience. What’s the person going to do now?
Come talk to us. Call us up and talk to us. We would suggest you apply to be trained, because then you can come to us, or now we can do it virtually. So one way or the other. But we will help you through that process. We will get you through the staining.
You have to have a microscope available, so if you can’t come to us then you need to find a local microscope, and we’ll try and help you find that too. We have various contacts and we know where the core centers are. So we’ll help you have a first look at it, evaluate whether it’s ready for cryo; if it’s ready for cryo, we’ll help you get it frozen.
These days with COVID-19 maybe that means you ship it to us, and we give it a shot and tell you what’s going on. Otherwise, if you have access to equipment in your own hometown, we will help you get that through. So it’s a step-by-step process. Just come talk, start a conversation. Let’s think about where you’re at. What does your gel look like? What does your size exclusion look like? And we’ll tell you whether we think it’s ready for prime time.
That’s great. Related question to that was, what kind of approaches do you need to look at your sample before you start? You mentioned gels, size exclusion, anything else? And are there any no-no’s, like buffers you shouldn’t use? Glycerol, etc.
We don’t like a lot of glycerol, and that’s because it’s a contrast killer. We are looking at our sample in solution, and the only reason we can see it is because its density is slightly different to water. But if you start bringing up the density of the water by adding glycerol or sugars and things like that and you get very little contrast, that just makes your life harder. So 15 percent glycerol? No. Or very, very high sugar.
But as we always say to people, your protein has to be happy. If you make your protein unhappy by taking all these things out, then you’re nowhere. So you get your protein into the conditions we like, as happy as possible, and then we’ll work with you.
Sometimes what we do if it will not live without glycerol, we’ll dilute it just before we put it on the grid or tricks like that. So we can work with just about anything.
What was the first question?
What other techniques besides gels?
The usual biochemistry techniques. And then the first thing we’ll do is put it on a grid and stain. And that takes literally 10 minutes. It’s not a big deal, but you might as well have it cleaned up and as well known as possible before you do that.
It sounds as if talking to you early is a good idea. Before you invest all your time, talk to you guys.
Right, come talk to us and see where you’re at. We’ve seen hundreds and hundreds and hundreds of projects, so that helps.
A lot of experience. Is the virtual training at NCCAT for anybody or is it only for people who have a grant with NCCAT?
No grant with NCCAT. Sorry, I’m getting mixed up myself. It’s an NIH Common Fund-funded center, and there are no rules as to who can and cannot apply. Anybody, literally anybody can apply.
We tend to favor people from the United States, obviously, but we have served a few elsewhere. And all you have to do is write the application and we will review it. It gets reviewed, so that means we want to be open and transparent. We’re not just choosing our friends and relatives to give time to. So the review committee reviews it and then we evaluate those reviews and then allocate time.
Great. And just a reminder that there are two other centers too, so depending on where you are in the country or if the center is busy, there are three places you can go. Can you comment on the role of elastic and inelastic light scattering in cryo-EM?
Electron scatter, but if a sample is very thick, you’ll have a lot of inelastic scattering, and then the kind of formulas we use to do the reconstructions start to break down. So that’s why we want very thin samples that the electrons go through elastically rather than inelastically.
So there are things called energy filters where we can sweep away some of the inelastic scattering, and that helps if your sample’s a little bit thick. That’s probably quite a long technical question, but anybody is very free to call me up later or email me later, and I can refer you to some good literature if you’re wanting to know more about these topics.
Terrific. Can you give us some perspective on phase plates and how they might influence cryo-EM structure determination?
That’s a good question. Phase plates were very, very much in vogue, and we have a bunch of them at lab, but I am sad to say we don’t use them anymore. At first we thought they would be the game-changing thing that would allow us to see everything that we couldn’t see anyway. It turned out the cameras were so darn good we didn’t really need the phase plates.
That’s not to say I don’t think a phase plate would help if it was a perfect phase plate. And Bob Glaeser’s working in Berkeley on this laser light beam phase plate, and that would be better than the ones we have now. The ones we have now are problematical enough I think you don’t use them unless you’re desperate. We use them occasionally, but most of the time we can do it without.
But if we had a perfect one like Bob might invent, then we would use it all the time. It’s too much trouble at the moment, Jon.
We need technology development there, it sounds like.
Very much so. And what Bob’s doing is super cool but jolly hard. Very tough physics.
You mentioned this before. Why do you do negative stains before you make the cryo-EM grid? What’s that for?
Because it’s very easy to do and it’s very quick to tell what’s going on. Whereas, if you make a cryo-EM grid, if you’re not an expert and you look at the grid and you don’t see anything and you think, “Is it because my ice is too thin or is it too thick? Do I not have any protein there? Did the protein fall apart? Is the protein stuck to the carbon? What is going on here? Is my dilution wrong?”
Whereas, a negative stain, which you can make in 10 minutes, you can make 10 different dilutions, quick look at them and say, “Ah, I really do have particles there, so I should expect to see them.” But if you go straight to cryo, it takes longer to do that. It takes longer to get the grid into the scope.
You have to be better at it, you have to have more training, so it takes a bit of time. That’s not to say the experts, the people who do it every day of the week, wouldn’t just bang out a quick cryo grid and have a look at it, but if you can’t figure out what’s going on, then back to stain you go.
Interesting. Somebody’s thinking about cryo-ET and wants to know if… you mentioned the problem of how do you know which of the things that you see there are your complexes. Could they use gold-conjugated antibodies to detect which are their complexes in cryo-ET?
Sort of. One would like to say that’s true. The problem with gold-conjugated antibodies, the antibody is probably bigger than your molecule and the gold might be bigger than your molecule, so it’s kind of messing and cluttering up the view. Also, is it 100 percent effective at labeling?
When you’re labeling things in the lab microscope if 1/1,000 is labeled and you’re looking at a million of them in your field of view, you still see them. We have to hunt them down one by one. If 1/1,000 is labeled, well, we only have one in this field of view. We’d have to take hundreds of thousands of fields of view. So complications like that.
But we are desperate for a GFP for EM, and there have been some that have come along. I wouldn’t say there’s a perfect solution yet. Somebody else may jump in and contradict me because I’m not an expert on this labeling area, but I think there isn’t something that’s perfect yet. The thing that comes closest is fabs or nanobodies. You can stick those on, and they act like a fiducial at least, but they are still hard to see.
That’s great. Someone wants to know whether you can actually see—you talked about looking at drug-polymerase complexes in COVID-19—can you actually see that resolution with the drug and the specific amino acids it interacts with?
Absolutely. You need high resolution. Above 3 angstroms maybe you can see a blob of the drug; below 3 you can see the whole drug, and you can tell what conformation it’s in. Even in this amazing cryo-ET project, this was an in-situ ribosome, that thing had a drug on it.
I think it was a bacterial ribosome, and there was a drug on that ribosome; you could see the drug. So of course more resolution is always better if you want to exactly know where every atom of the drug is, but absolutely we do that all the time. And Jon mentioned that we run a company in San Diego. That’s what our clients want there. They want to look at targets with drugs on them, and we get that routinely for them.
That’s fantastic. Here’s another great question. We’re having really great questions. Can you talk a little bit about cryo-electron crystallography, MicroED?
MicroED. Great question. New kid on the block, MicroED. And I think it has huge potential, particularly for the small molecules. We’ve solved a lot of not-so-small molecules, actually, things up to 1,000 daltons, and it’s very good at doing that. You can get 1-angstrom resolution data off a pretty middle-of-the-road instrument. You don’t need a KRIOS for this. And if you get 1-angstrom data, you can solve the de novo using the brilliant methods that the X-ray crystallographers have been developing for years and years.
Protein crystallography, a little tougher because the proteins are harder to handle, harder to get on the grid, but that’s another area of good development. We need to develop better techniques to do that. And then if you’re only getting 3-angstrom data or so out of that protein crystal, solving it de novo, of course, is a little difficult.
But that may not be all that people want to do. They may want to solve something or have a homolog and then do it with many, many different drugs, and they may only be able to get small crystals and, therefore, MicroED is the perfect method for it. I think MicroED will be very synergistic with X-ray crystallography and the two will feed back and forth from each other, but it’s a very exciting new method and it works very well, actually.
That’s terrific. Exciting to watch that happening. OK, here’s another really great question. I was wondering if you could comment on possible issues that could arise during data processing. It’s a complicated process, so are there opportunities for data manipulation or misinterpretation that people should be aware of, so they don’t do it?
Yes, for sure there are. It used to be much worse.
People are probably remembering some 10 years ago there was quite a bit of scandal in our field because there was back and forth as to whether something was right or not. It used to be much easier to get this wrong at low resolution. So if you get a blob, you can interpret that blob just about any way you like. If you’re at atomic resolution where you can see every atom and you can see alpha helices and you can see beta sheets, it’s easier to evaluate and judge whether that thing is right or not.
So we do have all kinds of checks and balances. We don’t have the fancy R factor that the X-ray crystallographers have. Sadly, there isn’t something just ideal like that. But there are many different evaluations of it, and ultimately if you look at the thing and if it’s a high-enough resolution it’s pretty easy to tell whether something is going wrong.
Which is not to say that people couldn’t cheat, but we hope they don’t, and you shouldn’t be able to get it wrong just by doing the wrong thing. You should be able to get to the right results if you’re honest and you use the methods in the right way.
I suppose more and more computational approaches, machine learning, are taking the human out of the alignment in things like that. Is that correct?
Yes. And there’s a lot of checks and balances inside these packages. It’s a while since I’ve seen anything that went wrong in a way that wasn’t so obvious, like, “Oh, well that didn’t work. Let’s try again. Let’s do something else. I wonder what went wrong in the software.”
It’s not that you get to something and say, “Oh, that looks right,” and then a month later realize it’s wrong. That doesn’t happen anymore. It’s either horribly wrong, looks like some ghastly thing, or it’s kind of right. And then you can always make it better.
So here’s a really interesting question. Coming from a low-income country with a background in basic medical sciences in Rwanda, how can one get this kind of experience and training in cryo-EM so that they can advance this sort of research there?
Well that is an excellent question. We’d have to talk to NIH about how they feel about having trainees from other countries. Maybe there’s workshops that could be run. We’d have to think about that. It’s not obvious to me what resources we have. But we do work with people from other countries.
We have a very lovely project on Zika vaccine going at the moment with people in Mexico, because they do not have access to cryo-EM, so we said let’s start that as a collaboration. That’s really done with the NRAMM side of our lab, because it’s a driver of some interesting technology we’re developing. So we do collaborations with other countries.
I would encourage you to contact us and talk about it and we’ll talk to NIH and see what can and cannot be done.
And I would say the one place at NIH you could contact is the Fogarty International Center, which has a number of programs for fellowships for people from low-and middle-income countries to come to the NIH, to come to labs in the U.S. and study and learn these kinds of techniques.
That would be ideal.
Here’s someone who wants to know more about dynamic cryo-EM. What’s the cutting edge there of doing time-resolved cryo-EM and seeing things change and move over time?
Well that’s a nice question for us because we just sent a paper out—it’s under review, actually, at Nature Methods on time-resolved cryo-EM. And we’ve developed a fancy instrument that uses piezo dispensing onto self-wicking grids in order to try to make the specimen prep better. And once we’d done all that—that’s been commercialized now—we decided we could split two streams of piezo droplets at the grid and mix them on the grid.
And we just did some really fun experiments with that, mixing DNA promoter with RNA polymerase and watching that first hundred milliseconds of how that DNA melts into the promoter. Throwing ATPase at a dynamin tube and watching this unravel and things like that. So that’s super exciting.
Joachim Frank has done this for quite a while with ribosomes. So they mix two different ribosomes together or mix components of the ribosomes together and watch that. So they’ve been quite doing a lot of time-resolved on ribosomes up to now.
But this new time-resolved method that we have with Spotiton, with our specimen-making device, allows you to just mix very tiny quantities together, and that’s helpful. So we’re very excited about doing more of that.
Really cool. A couple of career questions. I think you’ve convinced at least one person that they should come work at the National Center. If you’re interested in working at the National Center in the future, what type of training would you recommend?
Well we take people in at all levels. We’re always looking for great cryo-EM scientists, but a lot of our cryo-EM scientists who are at the National Center right now started as techs. They came out of undergraduate; they started out as technicians; they trained up and now they’re running KRIOSs and doing amazing work. So we regard our training mission as doing that too.
We don’t expect people to come in perfectly ready to work. We do a lot of training. So if you can learn some cryo-EM, great. If you are comfortable around computers, very good because we do a lot with computers. And if you love instrumentation, excellent. But other than that, you can be trained to do these things.
That’s great. Here’s a slightly more sociological question. What advice do you have for an outsider trying to break into a community when so much in science is built on pedigrees and on your CV on paper?
Again, I’d say start somewhere. Start as a tech in a lab or doing something useful in a lab and get to know some people and get a little bit of experience. Is it built around pedigrees? I mean, it’s certainly built around knowledge and competence. You can’t just walk in and say, “I’m going to now solve all these problems.”
You need to bring something to that table and maybe what you bring is a willingness to learn. We take in so many postdocs and students who come in and say, “I’d love to learn cryo-EM.” It’s very hard to say “no” in that case because you also want them to learn it and we’d love them to learn it too. So it’s very powerful to be willing to go in and learn and just tell someone, “I’d like to enter into this. Can you help me out?” Don’t be afraid to say that.
I think that’s good advice, just ask.
Yeah, just ask.
It’s good advice. People like to be asked, right?
Yes, and you’ll find PIs find it very hard to say no.
I think that was Tip O’Neill who said people like to be thanked and they like to be asked.
Yeah, that’s right.
What about smaller proteins? NMR, for instance, has a size limit—can’t go above it. Cryo-EM can solve huge things, but is it going to be good for really small proteins or other things?
For sure. We just solved with Filippo Mancia’s lab at Columbia University the PFCRT, which is a transporter in malaria, a parasite, that is responsible for inhibiting drugs. And that thing is 50 kilodaltons. It’s an integral membrane protein, and it’s 50 kilodaltons. That definitely would have been said to be impossible a few years ago.
We cheated a bit because we put a fab on it. [unintelligible] lab made a nice little fab and we added a fab and that made it a lot easier. But somebody has done one without a fab at about 50 kilodaltons. So 50 kilodaltons is definitely possible now, and I think NMR can do anything below 50 kilodaltons maybe. And I think that’s a pretty sweet spot anyway, 50 kilodaltons. Many, many things are that size and above, so we’re definitely making progress on that.
Do you think cryo-EM is going to replace crystallography?
No. Not at all. I think they’ll all be synergistic. I think NMR, crystallography—I think NMR is going to have a little comeback too by the way. I think all three of them are these three legs we build high-resolution structural biology on. All of them will feed off each other and go back and forth.
If you want an example, so cryo-EM is being hugely successful for COVID-19, but the fragment-screening people in Diamond did 1,500 samples in a few days I think—they at least did something like 600 in 72 hours—and in the end they did 1,500. It was a small piece of the protein and it was screened with fragments, but wow, 1,500 in a few days? We’re not there. We can’t do that. So the two of them will work well together.
Cryo-EM might do the entire trimer or the polymerase and say this is the working part. Crystallographers can pull that out, they can make a smaller protein out of it, and then screen it against thousands of fragments. That’s cool synergy between the two. And then NMR has its own role to play in dynamics and all kinds of things like that.
I’m being reminded by one of our great program directors, Paula Flicker, that I should mention that each of the centers does training in a different way and they have different workshops, so you may actually want to look at what all of them have to offer, because they can offer very complementary and synergistic things.
Let’s see. We still have lots of questions, but we only have two minutes left. And there are lots of compliments here too, I’ll tell you, Bridget, to your talk. Here’s a great question. What do you think the 200-kilovolt microscope is in the future—and this is versus 300, so maybe you could just give a little background on that, but then I’d love to hear your opinion too on this idea of lower-voltage microscopes that might be cheaper.
So Richard Henderson and Chris Russo had a very nice paper recently that was written about quite a lot. I was interviewed, toward democratizing EM, saying $7 million, an enormous room, and half a million dollars a year to look after it; that’s too expensive. We need lower end.
What they really want is 120K of a cheap instrument with a fig that’s capable of 3 to 4-ish-angstrom resolution. And the idea is that instrument would be less than $1 million, nice camera, nice gun, nice instrument. All small research centers, small universities, could afford to have it, could afford to house it, could afford to look after it, do most of their work at home, and then if they really need that sub-3 angstrom embedded, they’ll send those samples to the centers, and the centers should be able to cope with that.
So I think there is a future for that. We have to get the instrument companies to agree they would like to do that and build such an instrument, and I don’t think any of them is at the moment. So Richard and Chris are building it themselves. There are the 200 kV instruments, and they’re very good, but they are still quite up there in terms of price.
That’s half a KRIOS, it’s not less than $1 million or anything like that, but they are cheaper than the KRIOSs and they do very well. I would still say the KRIOS is the creme de la creme instrument, and if you really want the very best data, that’s where you all want to end up.
Great. Well, we are at the end of the hour. There are, unfortunately, still questions I could have asked you. Sorry I can’t ask everyone’s questions, but boy did you guys have great questions. Bridget, this was an absolutely terrific session and talk. I’m so grateful for you and grateful to all the trainees who tuned in and everyone who’s working so hard right now, either at home or on the front lines of this epidemic. Please, everyone, stay safe and keep doing your part because you’re doing great things.
Thank you so much, Jon. It was really fun, and those were terrific questions. And feel free to contact us so we can answer more questions offline.
Good. Thanks, everybody. Tune in to the next one.
Thank you. Bye.
