Venkatraman Ramakrishnan (2015) - Seeing is Believing - A Hundred Years of Visualizing Molecules

Kurt Wuethrich said, "You aren't 120 years old“ because my talk was called "120 Years of Visualizing Molecules“. So clearly, it's not going to be about my work. But after hearing 2 of the last 3 talks, I should say it's "100 years of visualizing molecular structure", because those are 2 somewhat different things. The first thing is, how did we even get to the idea that there are molecules? That's an amazing intellectual journey made by humans in the 18th and 19th century. And for those of you who are interested in knowing how we went from knowing nothing about the organization of ordinary matter to understanding molecular structure, I highly recommend this book. The point is that even before we could see molecules, we really knew a lot about what molecules were like and what their structures would be like. How do we actually see molecules? How do we directly visualize molecules? You heard 2 really amazingly brilliant talks about how to resolve fluorescent molecules beyond the Abbe limit and even in live cells, and that's really transforming biology. But now we come to the reality check that was discussed in the last talk. These techniques can show us where molecules are but not what their structure itself is. Because most of the atoms are not fluorescent, they simply scatter light, and so for the vast majority of atoms in any sample, the Abbe limit unfortunately still applies. In fact, it's ironic that if you want to know the structure of a fluorescent molecule or GFP, you can't use these super-resolution microscopy. You have to use something else. And what that something else is for the last 100 years was x-rays. X-rays are waves, and for that Max von Laue, for that discovery, won the Nobel Prize in, I believe, 1914. He showed that when he hit x-rays on a crystal, he would get these spots. Actually, he misinterpreted these spots, so he didn't get the analysis of these spots correct. That was done by a graduate student, Lawrence Bragg who, for his PhD work, went on to become the youngest ever Nobel laureate and still is the youngest one in science. In fact, because it was his original work, his supervisor didn't get the Nobel Prize because these were all published on his own. And that still remains something of a Cambridge tradition, not always honoured, I'm afraid, but still there. How did this work? What Bragg realized is that if you take crystals, which are regular 3-dimensional arrays of molecules, and you hit them with a beam of x-rays, then, because you have these arrays, you'll get interference. And that interference reinforces the x-rays only in certain directions and that gives rise to spots, which are now called Bragg reflections. If you collect these spots, then that's what this looks like. The point is that these scattered rays are there, whether there's a lens there or not. So the light rays - you can have a lens that recombines these scattered rays to form an image. With x-rays, you don't have a convenient lens, and so you have to recombine them in some other way. What you do is you measure them and you recombine them mathematically, which is equivalent to doing a Fourier transform, and you then get an image of the object. Since x-rays have wavelengths of about 1.5 Angstroms or now even shorter, you can resolve things, which are less than 1 Angstrom apart. And so you can actually resolve the atoms apart in a molecule. The first such structures were determined using guess work because there were only a few spots, so you could guess at the structure of your molecule and you could see if it agreed with the pattern of spots. Using that was the first surprise. The first molecule structure that Bragg discovered was was sodium chloride, just has 2 atoms, sodium and chlorine, and he found there was no sodium chloride molecule. This caused a big fight with the chemist who said that he's a physicist and he doesn't understand any chemistry and he's talking nonsense. Of course, this is one of the many ways in which crystallography has actually influenced chemistry. Now, subsequently, as it got more complex, people had to use mathematics to recombine those spots using Fourier methods to reconstruct an image of the molecule in order to be able to interpret it. In those days, there were no computer graphics. And so what they had to do was to take the Fourier density and contour it on these plastic plates and stack them up, That's how they'd look at their three-dimensional image of the molecule, But you can see here that these round blobs represent the atoms, you can actually see separate atoms, and so you can see what the atomic structure of the molecule is. This particular molecule was penicillin determined by Dorothy Hodgkin who is shown here on the postage stamp. She's the only scientist, I believe, in Britain who has 2 postage stamps at 2 different times after her. This shows the structure of penicillin, which again was a little bit of a controversy because of this rather square beta-lactam ring, which some chemist didn't believe existed because it would cause strain. Again, it's an example of how molecular structure has illuminated chemistry. Now, these kinds of methods were useful for molecules for a few up to a hundred atoms or even a couple of hundred atoms. But when you get to proteins, they consist of thousands of atoms, and that required another way of calculating the image. That was done by 2 people who founded the lab where I work, which is Max Perutz, who was Bragg's protégé, and John Kendrew, who was Bragg's PhD student originally. These are the molecules that they looked at at the bottom. This is haemoglobin and this is myoglobin. These are 2 oxygen-carrying proteins in our blood. That person in the middle is Gisela, who is Max's wife, and she's pinning a carnation on his coat doing the Nobel ceremony. Now, when you get a typical protein structure, you don't have enough detail to see individual atoms as little spheres as you saw in that picture of penicillin. So the question is, how would you go about looking at a molecule at an image and actually getting an atomic structure out of it. It's a little bit like solving a jigsaw puzzle. This would be what a 3-dimensional image of the molecule looks like. This happens to be the small subunit of the ribosome. It has about 200,000 atoms in it. What you would do is you would zoom in on it and look for things you recognize. If you see here, there are 2 ridges, 1 here and 1 here, and if you know what you're doing, you realize that these are 2 chains of nucleic acid that are wrapped around each other to form a double helix. And so suddenly you can start building in your molecular pieces into this image just like you would fit in a jigsaw puzzle, except it's a 3-dimensional puzzle, and you don't know the answer beforehand because it's not on the cover of the box. Now you suddenly have a molecular structure from an image. But it's only part of the structure, so what you have to do is figure out where it is, and you have to keep on building until you've essentially interpreted this entire molecule. Basically, until you've got no density, which is no part of the image left to interpret, you just keep on going and you end up with a complete structure. Now, this is a sort of process by which every molecule has been directly imaged for the last 100 years, just about. There are other ways of determining molecular structure. Kurt Wuethrich, as you know, invented a way of determining molecular structure by NMR, but this is by a set of constraints on a chain, which tells you how the chain will be folded and how groups in the chain will be arranged. It's not a direct visualization, or imaging, like crystallography or like optical microscopy. This technique went from sodium chloride, which is 2 atoms, and it's capable of determining the complete atomic structure of something like a whole ribosome, which is half a million atoms. And in fact, now we have a structure of the ribosome from higher organisms, which is close to a million atoms. So you can see it's an incredibly powerful technique that's transformed chemistry, and it's one reason why so many chemistry Nobel Prizes have gone for molecular structure. The problem is that when you're studying more and more complex molecules, it's harder and harder to get them to crystallize. Many of my people have spent years trying to crystallize something unsuccessfully. The other problem is many molecular assemblies are transient. They don't stay stable enough and they're not confirmationally homogeneous enough to be able to crystallize them and solve by x-ray crystallography. It leads to a question. I'm a ribosome biologist. I'm not really what you would call an x-ray crystallographer, and so it made me realize, what exactly is my goal? If you don't understand what your goal is, you can be in trouble. When I was a child, if someone had told me that Kodak would go bankrupt, I would've thought they were completely insane. And yet, after 131 years, Kodak became bankrupt because it didn't realize that its goal, it was in the imaging business, not in the film business. In fact, although they made one of the first really good digital chips for photography, some guy in Kodak apparently decided that they didn't want to push that too much because it would cannibalize their film business. Bad mistake. Okay. I didn't want to make that sort of mistake. So looking around, there is another technique, which is electron microscopy, which has the right sort of wavelength to look at molecular structure. This is Ernst Ruska on the left and that's his electron microscope on the right. What you'll notice is that he had to wait 53 years after his discovery to get his Nobel Prize, and he only died 2 years after he got his Nobel Prize. It's something of a record for the amount he had to wait for the Nobel Prize. Material scientists have also always been able to determine atomic structures using electron microscopy, almost since its inception. But the contrast and complexity of biological molecules meant you couldn't do it that for biological molecules. How did electron microscopy get used in biology? A classic way is to cut sections of cells and stain them with some heavy atom and look at what the cells look like, so these are ultra-structures of cells that are familiar from every textbook. But that's not what I mean. What I mean is how do you get at molecular structures using electron microscopy in biology? One of the first big advances was done by David De Rosier and Aaron Klug. This is a picture, a diagram from their paper in Nature in 1968, which was reproduced in Aaron Klug's Nobel Lecture in 1982. The idea is that if you have a sample here and you shine a beam of electrons, what you get when you have an image, when you focus it to form an image, is a 2-dimensional projection of the molecule or of the object. Now, if you were to look at it from a different direction, you would get a different 2-dimensional projection and same here. If you were able to combine all these 2-dimensional projections, you might be able to get a 3-dimensional structure of the object. Of course, in an electron microscope you don't change the direction of the beam. What you would do is you would tilt the sample, you would rotate the sample and that way you would sample different views. One of the things that De Rosier and Klug realized is that if you take a Fourier transform of the 2-dimensional image of the object, it would be like a plane in Fourier space. And so if you collected all these planes and then did a reverse transform of the Fourier planes, you would get back a 3-dimensional image of the object. Those of you who are more in the cell biology medical side, I can tell you this is exactly how CAT scan works. You get different projections of your object, and there's a way to recombine the projections to get a 3-dimensional image. The problem with this is that biological samples have very low contrast, so you have to stain them in order to see them very well. That's one problem. The other problem is if you don't stain them, the contrast is so low that you have to hit them with enough electrons to see them. And electrons are damaging, so if you rotate your sample, you have to collect a whole series of images. By the time you finish collecting your series, your sample is dead. You've destroyed your molecular structure. There are a couple of ways around that. One is if you have a mixture of the same molecule and you're looking at all of them at the same time, these molecules will all be in different orientations, and each of them will give rise to a different projection. Now, the trick is to find out which orientation of this molecule gave rise to this projection and same for this molecule and same for this molecule. If you can figure that out, then you know that if you have a molecule, you know that this projection here corresponds to this view, this projection here corresponds to this view, this projection corresponds to this view, so you get thousands of views at the same time. What this means also is that you don't have to hit the molecules very hard. You get the power of averaging from all these thousands of molecules, and so you can get a 3-dimensional structure. The other really major advance that has come out is that, by embedding these molecules in ice that has been cooled so rapidly that it doesn't crystallize to form crystalline ice (so it's what we call vitreous ice), you can look at these molecules in effectively their native state. And so you can look at very large complexes in their native state. The problem is that even when you do that, they're incredibly noisy. This is a field of ribosomes, which are fairly large molecules, and people in this field joke that the ribosomes are a particularly easy sample to work with. But nevertheless, if you look at what an individual projection of the molecule looks like, you can see it's barely visible. You can hardly see what it looks like and let alone get some sort of atomic structure. And yet, remarkably, this technique was used to get something at about 27 Angstroms resolution of the ribosome. It's the first time in the ribosome that you could actually visualize tRNA molecules inside the ribosome. Even though crystals of the ribosome had been around for about 15 years at this point, these were in 1995 by far the best images we had of ribosomes. Of course, they were soon superseded in the next few years by crystallographic methods. But nevertheless, it showed that you could get something about the shape of molecules by using this. But could you use this method to get structures of biological molecules? No one thought it was possible at the time, in 1995, because a very large molecule, which should be easy to see in a line, could only give you about 27 Angstroms resolution. But that very same year, there was actually a seminal paper that was published by another person who works in my lab - who was the director of the lab, who actually hired me, although I'm now his boss, I should say. He published this paper, which was essentially a theoretical paper that had come out of studies he had done on bacteriorhodopsin. And what he suggested was that with electrons, if you had molecules which were only about 100,000 Daltons or bigger, you should be able to determine the atomic structure with only 10,000 molecules. What did he mean by atomic structure? He meant 3 Angstroms resolution. Now, 15 years later, there were still no atomic structures except for viruses, which involved very large symmetric molecules that combined millions and millions of copies of the molecules. It wasn't clear that this thing would ever materialize, although no one could ever find a mistake in Richard's paper. In fact, everybody agreed that he was right and it wasn't clear what was happening, although Richard, himself, knew what the limitations were and was working on them for the last 20 years. Now, we have another Resolution Revolution - and that's not my term, that's the term of Werner Kuhlbrandt who wrote this perspective that accompanied our paper in Science last year on the mitochondrial ribosome. If you want to know in detail what's happened, I suggest that you read this very short perspective, which gives you an idea of what's happened in the field. But I'll just briefly tell you. There are 2 main advances. Firstly, we've had better microscopes over the last 20 years. We have better stages, more stable microscopes, et cetera, but that didn't take us to this point. It improved things but not to the point where we could get atomic structures. What has happened in the last few years is that we've had better detectors and better ways of processing the data. The first thing are better detectors. Classically, the way to detect electrons was by using film. Film is a reasonably good electron detector. It is actually a direct detector. To call these new electronic detectors 'direct detectors' is a slight misnomer, because so is film. Film is also slow and laborious. You have to scan it. It has its own problems of uniformity and noise and so on. Along the way, CCD detectors came out. But CCD detectors are actually a very bad electron detector because they take electrons, convert them to fluorescent light, and then that light is detected via fibre-optic into a CCD chip. Richard used to tell people that if you want to degrade your data and get worse data, then you should abandon film and go to CCDs. In the meantime, a new class of detectors based on CMOS chips have combined the advantages of film in that they directly detect electrons with the convenience of CCDs in that you don't have to develop and scan films and so on. They have better signal to noise than film, significantly better DQE than film, so they're more sensitive than film. The other thing that's happened is better image processing, better algorithms to look at these very noisy images and get the most out of them. The way we got into EM is because this post-doc, Israel Sanchez, spent 3 years trying to crystallize a ribosomal complex and was getting nowhere. And out of frustration, he decided to collaborate with another post-doc, Xiao-Chen Bai, who was working for my colleague, Sjors Scheres, and these 2 guys were doing electron microscopy. What they did was to figure out what his samples looked like. But to do that, first he thought, let me just put a test sample of ribosomes and see what I can get now, and then I'll look at my complex. Now, there are 2 things that happened. One is that it turns out that whenever electrons hit a sample, they ionize the sample. And if you're in a non-conductive medium like ice, vitreous ice, this creates very large local fields. It's a phenomenon called charging. And the molecules start to move because they're charged and they're experiencing these strong local fields. The moment you hit your sample, from the minute you start looking at it, the samples are moving around. This is just to show you what beam-induced movements look like. I'm showing it to you after the fact because now we can actually track them, but this is a problem. Now, the thing about these new detectors is that they're very fast, so we may be able actually be able to do something about these beam-induced movements. The first thing that these new algorithms do is that if you look at this image here, it's extremely noisy. And what you have to do is ask, what is the orientation of the molecule that gave rise to this particular projection? In order to do that, you do a mathematical thing. You calculate all possible projections and you decide which projection matches the observed projection. That's basically the essence of the idea. But if it's noisy, an incorrect projection can sometimes give you better agreement than the real orientation of the molecule. This is a classic problem with noisy data that you can converge on the wrong minimum. What Sjors Scheres did was apply a very well-known statistical method called Bayesian likelihood to this problem. So rather than asking what's the best align orientation of the molecule that matches this projection, he asked, By storing that entire Bayesian probability and then using all of that to do the recombination, you can arrive with a very robust and much more accurate 3-dimensional reconstruction of the molecule. The other thing I alluded to was the beam movement. Now, it turns out that these detectors are so fast, that what you're seeing here is a 1-second image, and that's what a 1-second projection of a particle looks like. But this 1-second image is actually 16 frames that are shown here. You might ask, if I could just watch this molecule move along these frames, then I might be able to track the movement of the molecules. In order to do that, you have to be able to detect the molecule within the frames. Now, clearly, if you add up all 16 frames, there's enough signal there to see the molecule and align it. Now, it turns out for ribosomes, you can do it for 8 frames, you can do it for 6 or 4 frames, but you can't do it for 2 frames - because it becomes so noisy that you get errors in the location and orientation of the molecule, so it's not good enough. What we can do is do a 4-frame average and then do a moving 4-frame window across those 16 frames. And what this does is it allows you to track the movement of the molecule during that 1 second. Once you can track it, what you can do is you can apply a correction for the movement to each one of those frames, and then add up the corrected frames. So now you have a projection that's corrected for the movement, so it's not blurred; it's effectively deblurred. Doing that what Sjors and his colleagues along with my post-doc, Israel, did was they took just 30,000 particles of the ribosome, and they could actually separate these molecular strands, which are called beta strands. And they could look at the helical structure of alpha helices and even see density for the side chains of helices. Suddenly, it looked like you could do atomic structures by electron microscopy. That indeed is what 2 other post-docs in my lab applied to a problem that had essentially defeated crystallographic methods for over a decade. And that is the structure of mitochondrial ribosomes. This is Alexey Amunts, who essentially started this mitochondrial project in my lab and really drove all of the biology. And this is Alan Brown, who did a brilliant job of interpreting what was an amazingly complex structure about which very little was known previously. These structures were published in 3 papers in Science over the last year, and that's just to tell you that the Nobel Prize doesn't have to mean the kiss of death. I know that there is a temptation after you get the prize to go around giving talks and being wined and dined. But if you say no to most of these invitations and really drive science in your lab and remember what actually got you into science in the first place, you can still be publishing important papers. Incidentally, Alexey Amunts is going to where Astrid works, which is the University of Stockholm, as a faculty member. The question is what are mitochondrial ribosomes and why are they interesting? Mitochondria are the energy powerhouses of our cell; they're organelles in our cell, But the way they are thought to have originated is that 2 billion years ago, 1 cell swallowed another cell. And the cell that it swallowed, which was now inside a bigger cell, eventually evolved to become mitochondria. But because, as species diverged, the mitochondria from 1 species could not recombine with the mitochondria from other species, the mitochondria of various species diverged in their own way and they're incredibly divergent. What Alexey did was to look at these mitochondrial ribosomes, and he found that there were a mixture of many different types of particles. And so this is not something that would ever have crystallized. Yet, using this method of electron microscopy, we can sort out mixtures of particles, and it can even sort out mixtures of different confirmations of the same particle in the object, which a crystallographic method could never do. He could get images that were detailed enough to build a molecular structure, so he and Alan Brown went on to build this molecule, which has almost a million atoms in its entirety. In fact, we have built 80 proteins, which have very little sequence homology with what was known. What this does is, it means that people who want to look at molecular structure are no longer at the mercy of having to crystallize something, which involves producing large amounts of stable homogeneous material. They can look at even mixtures of samples. They can look at molecules that assemble transiently and so on. And so just as these revolutions in optimal microscopy are going to transform cell biology, the revolution in electron microscopy is going to transform molecular structural biology. Thank you very much.

Venkatraman Ramakrishnan (2015)

Seeing is Believing - A Hundred Years of Visualizing Molecules

Venkatraman Ramakrishnan (2015)

Seeing is Believing - A Hundred Years of Visualizing Molecules

Abstract

It has been a hundred years since molecules were first visualized directly by using x-ray crystallography. That gave us our first look at molecules as simple as common salt to one as complex as the ribosome that has almost a million atoms. In the last few years, electron microscopy has offered an alternative to directly obtaining the structure of very large molecules. I will describe some highlights in this journey with an emphasis on the recent developments in electron microscopy and how it is creating a new range of possibilities for visualizing biological structures.

Content User Level

Beginner  Intermediate  Advanced 

Cite


Specify width: px

Share

COPYRIGHT

Content User Level

Beginner  Intermediate  Advanced 

Cite


Specify width: px

Share

COPYRIGHT