Dying to know “What’s the deal with cryo?” Cryo-Electron Microscopy (CryoEM) has transformed the structural biology field in recent years, and even garnered some wider fame when Jacques Dubochet, Joachim Frank and Richard Henderson won the Nobel Prize for it in 2017. So what is it? How does it compare the the “traditional” x-ray crystallography? And is crystallography dead? (hint: NO!) Each of these techniques has pros and cons. So let’s take a look at what’s going on. Starting with “What is structural biology anyway?” (I didn’t really even hear this term until late in undergrad and people kinda just expected me to know and never really explained, so here’s my attempt at it…)
“Structural biology” is a sub-field of biology that’s focused on figuring out what molecules “look like” and how their shape is connected to their function. A little more on this later, but for now just think of a silver fork and spoon. They’re made up of the same material so just figuring out “hey these are made of silver” isn’t going to tell you much about what they do (but feel free to pat yourself on the back – I’m all for celebrating little wins!). If you play around with them blindly (and carefully), you might be able to figure out that you can eat soup with a spoon but not a fork but it’s easier to stab and eat pasta noodles with your fork. And if you can see them, it’s easy to see how their distinct shapes allow them to perform differently at these tasks.
Structural biology works to connect the “what it does” to the “what it looks like” to help determine “how it does it” and it uses techniques including X-ray crystallography and cryo-EM to get looks, combined with biochemical & biophysical techniques (measuring activity, binding, etc.) to further investigate the functional importance of different parts that are seen (though sometimes this functional stuff is discovered first but makes a lot more sense once you can see it). It’s kinda like how if you make a hole in the spoon-y part of the spoon it makes your soup drip before it gets to your lip & if you see the spoon shape you can predict where hole-making might make you dribble. But, if you make a hole in the handle you might not know.
And speaking of things you might not know, let’s discuss “cryo” – “cryo” (which my computer insists on de-correcting to cry, so apologies if I don’t catch all the “corrections”) is short for “cryogenic” meaning really really cold (which we need to keep the molecules safe(ish) and still(ish)). And when structural biologists say “cryo” they’re usually referring to single particle cryo-electron microscopy (cryoEM) as opposed to things like cryo-tomography, which looks at “slabs” like sections of cells or tissues.
You might sometimes hear it called single *molecule* cryoEM, but the things you’re looking at don’t have to be individual molecules. A molecule is something where all the atoms (the individual carbons, hydrogens, etc.) are connected through strong covalent bonds (like one protein chain) as opposed to “complexes” where multiple molecules can interact with one another through weaker interactions (like multiple protein subunits and RNA pieces working together to make a ribosome). So “particles” is a broader term that encompasses single proteins, multi-protein complexes, protein/DNA complexes, etc. – any sort of individual “particles” in solution (i.e. each particle surrounded by its own full water coat).
Disclaimer – even though I’m right down the hall from a top-of-the-line cryo-electron microscope, I personally don’t use cryo-EM, but have done some crystallography and will talk more about this at the end.
So imagine you have a bunch of copies of one of these particles. And you want to figure out what they look like…
If you want to look at something small, your first thought might be – let’s use a microscope. Microscopes make things look bigger by taking advantage of the wave properties of light (visible light is a form of ElectroMagnetic Radiation (EMR) – which can be thought of as packets of energy called photons traveling as waves). When a wave interacts with things, the wave’s path can get altered. So a visible microscope can shine light through something and have that thing alter the light waves’ paths. Thanks to that altering, you now you have a bunch of waves “out of phase” (out of step with one-another “peak-wise”). So the signal from your thing is kinda jumbled.
Visible light microscopes then re-focus the light before you see it – but in such a way that the image looks bigger than the real thing was, so it’s easier to see it and tell different parts of it apart (resolve details). This is possible because the phenomenon of “refraction” causes visible light to bend going when it goes through different media (like going from air to glass to air). So, before you see the light, microscopes have those jumbled waves travel through glass lenses which, because of their curves & thicknesses, bend the waves back to a focal plane where the signal appears as a (bigger) image.
That works for things that are “small” in the sense most people think about small things, but not “small” at the level that structural biologists think about things. Structural biologists usually talk in terms of angstroms. An Angstrom (Å) is 10⁻¹⁰ meter, or 0.1 nanometers (nm) and the average protein has a diameter of ~5nm, so 50Å. Problem is, microscopes are limited by the wavelength of the light – waves are only useful for looking at things that on the same order of magnitude or bigger than their wavelength. The shortest-wavelength visible light is ~700 nanometers, so ~7000Å, whereas interatomic distances are closer to 1Å. So we can’t use visible light to resolve what we want to see.
This is why we use x-rays in x-ray crystallography – X-rays are “just” a more energetic form of visible light – they’re just different sections of the EMR spectrum. The only difference is that in x-rays the photons have more energy and the waves peak more often (higher frequency) and the peaks come closer together (shorter wavelength) so that, linearly, it’ll take a photon of an x-ray the same amount of time as a photon of visible light to go from one side of the room to another. X-rays have wavelengths that range from 0.01-10 nanometers (so 0.1-100Å) and for protein crystallography, we usually use x-rays with wavelengths of ~1 Å.
Great. We have waves we can use! We beam them at our sample and our sample scatters them. Now we just need to focus those scattered x-rays right? Um, there’s a problem. We can’t just focus x-rays like we could focus light! They’re too energetic and short-waved, so (even if they didn’t just fry the lens) they don’t wouldn’t refract when they went through glass – or any other material we could use. As a result, in order to get information from the scattered x-rays, we have to work with these “jumbled waves” directly (or at least evidence of where they hit a detector).
The jumbled waves interact with one another – due to the regular ordering of the crystal, most scattered waves cancel out, but some strengthen each other to produce a “diffraction pattern” – a series of spots on a detector. And these spots represent what we call “reciprocal space” – a weird place where things that were once close together are now far apart, along with other eccentricities I won’t go into here. But basically you can use a thing called a “Fourier transform” to go from reciprocal space to “real space.” But you have to know the “phases” of the waves – where they peaking or troughing when they hit? And we lose this information in crystallography so we have to use complicated tricks to try to find them.
That can be a lot of work… so, what if, instead of using x-rays, we use different type of super short wave. One that we *can* focus? Such an optical option is provided by electrons, but it comes with its own problems including tending to destroy your samples… Which is where the super-cooled-ness of cryo-EM will come in. But first, the electrons – as waves? I’m used to thinking about electrons as being one of 3 key subatomic particles – as in parts of the things we’re trying to look at?! Yes, there *are* electrons in our particle already – but they’re “in use” – and different from the electrons our scope will introduce!
All the stuff in our particle are constructed of atoms – of carbon, hydrogen, oxygen, etc. – And each atom is made up of even smaller parts – “subatomic particles.” At the core of every atom is a dense center called the atomic nucleus, and it’s home to positively-charged protons & neutral neutrons. Surrounding this nucleus is an “electron cloud” with negatively-charged electrons whizzing around. But the electrons we’re going to use in cryo-EM aren’t whizzing around a nucleus – they’ve broken free from the positive pull of protons in our “electron gun” and are instead traveling in a spiraling path through a tunnel towards our sample. And they’re wiggling a lot, which means that they have really short wavelengths, ~ 0.02 Å, so about 50-times shorter than the x-rays we use.
And, although we can’t focus electrons with conventional glass lenses, we *can* focus them with magnets because, for physic-y reasons, magnets are the “electron influencers” of the sub-atomic social media scene. Thanks to the “Lorentz force,” magnetic fields are able to direct the movement of charged things (like the negatively-charged electrons). So you can use electromagnets as “lenses” to focus electrons onto your sample and then “re-focus” them onto the detector after they’ve passed through.
Sounds easy, right? Wrong. I’ve glossed over a lot of points.
First of all, it’s not like we have nicely ordered particles sitting still for the camera. Each particle is swimming around randomly doing its own thing. It’s really hard to look at something really tiny – let alone something really tiny that’s also moving all around. So we need to get them to cut that out. In crystallography, you don’t have this problem, because you’ve gotten them to organize into an orderly 3D arrangement called a lattice (kinda like a brick wall) (though getting them to crystallize is a whole ‘other problem…)
In cryo-EM, you still need the particles to stay still, but you don’t have to get them to organize – in fact you don’t want them to – instead, you freeze them in place, in all of their random orientations. So you get views from all angles. These are “projection images.” Yes – you have *images* now because we’re able to focus the scattered electron waves – unlike the x-ray waves which we had to work with scattered. You then go through the images, pick out the individual particles, find the ones that happened to be frozen in the same position, average them together and, by using some math I am *not* going to get into, use them to create a 3D model of it.
So how do you get them to stay still? This takes us back to the “cryo” in cryo-EM – as in really really cold. As in liquid ethane cooled by liquid nitrogen to keep it at -195°C, cold. So cold that molecules don’t have enough energy to wander around, thus they stay stuck in place. And if you cool them really really quickly, they don’t have time to organize into their favorite poses before they get stuck. So they freeze in place like the freeze dance. This goes for the molecules you’re trying to look at and for the water molecules around them. So, instead of forming ice (the crystalline solid form of water which has an orderly arrangement of water molecules), the water “freezes” into a “vitreous glass” where the molecules of water are stuck in place but randomly oriented – it’s still a solid, but it’s “amorphous” (no shape having).
The freezing is usually done on a carbon-coated metal (often copper or gold) grid. These are really small wire meshes (kinda paper hole-punch size) – usually 3.05mm across with 200-400 squares per inch (so a “200-mesh” grid has 20 squares in each direction, each (hopefully) with a film of vitreous water with particles suspended in it.
The metal is often coated in a layer of carbon. This carbon coating is used because it’s unreactive and not very sticky – so you don’t have to worry about your molecule binding to the grid itself and only letting you see a couple angles. You don’t want your sample interacting with the grid itself at all – instead you want the particles to be suspending in the liquid films filling the grid holes (think kinda like a bunch of bubble wands). These films are really thin – a few hundred to a few thousand Å – you don’t want too thick because you only want a single particle per layer and you also don’t want a bunch of background signal. To get that thinness, you can use a Vitrobot machine that reminds me of one of those monkey cymbals things – it smooshes the grid with filter paper to remove excess.
So now you have your sample. You put it in the microscope and shoot electrons at it. Then what?
When an electron hits an atom, it has a few different fates.
- No scattering: Most of the time, since electrons are so damn energetic (and teeny), they just go straight through the atom, as if it weren’t even there. These beams (~80%) of what goes in, don’t tell us anything
- Inelastic scattering: The electron in the beam can run into another electron in the sample and that sample atom’s electron can absorb some of the energy from the collision. This can give it enough energy to break free from the atoms it’s hooked up to (electrons only normally stick around anyway because they don’t have enough energy to overcome the positive pull of the nucleus). But molecules rely on those electrons to keep atoms glued together, so if you lose those electrons, you can get broken bonds, so your sample gets messed up. And, to make things worse, bonds can break so that single electrons are left unpaired – we call these “radicals” and they’re really reactive. Thankfully, the cold helps make it more difficult for them to actually react, but you still can only take a limited number of pics of any single particle. Those things are bad – but, as a silver lining, because the incoming electron had some of its energy stolen by the electron it ran into, when it leaves the sample, it’s as slightly lower energy (longer wavelength, lower frequency) waves – and these can be filtered out. This makes up ~15% of what goes through.
- Elastic scattering: Finally – elastic scattering is like your classic “billiard ball bouncing off a wall” example – A wave hits one thing and just bounces off – its direction gets changed a little, but it doesn’t lose any energy – so the wavelength remains unchanged. But unfortunately this useful stuff is only like 5% of the electrons
So, exiting the sample you have electron waves that are either unchanged, of lower energy, or of the same energy, but deflected a teeny bit – and now you need to bend these jumbled waves back to hit the detector using other magnetic “lenses.”
And now things get (more) confusing. Electrons are actual physical things (teeny tiny though they may be) as opposed to merely energy as was the case with photons. But, due to electrons’ “wave-particle” duality they can still add through wave superposition – if 2 waves are traveling “in step” with one another (peaking and through-ing) at the same time), they add together constructively to give a stronger signal. But if 2 waves are traveling “out of step” with one another (one peaking while the other troughs) they cancel each other out in terms of the signal we see. We call this destructive interference and it can be complete (if the waves are 1/2 a wavelength out of step) or partial if they’re out of step by any other amount.
As a result, in addition to just hitting the detector at slightly different places, the waves of electrons can have different amplitudes (strengths). And as a result of this, you’re able to get 3D information instead of just 2D like you’d get from a shadow of something completely solid. Instead of a shadow, these are “projection images” – you get different projection images depending on the orientation of the particle. And in the “2D classification” step, you sort the ones with similar orientations before going through all the even harder work of getting out the 3D info.
But just finding the particles is hard. You need a lot of them in order to get enough information and they’re hard to find even when they’re there! One of the main problems in cryo-EM is “contrast” – turns out biochemical molecules are pretty good at “camouflage” – they only interact weakly with the electrons, so they blend into the background (and are super tiny) so it’s hard to tell signal from noise – there’s a poor signal to noise ration (SNR).
In x-ray crystallography, the diffraction comes from the electrons in your sample because the interaction is through electric perturbations of the electrons’ electric fields that cause them to absorb energy from the x-rays and use it to send out their own waves, like a golf ball causing ripples (but importantly of the same wavelength, so you have x-rays come in and x-rays get sent out). But in electron microscopy, in addition to the difference that it’s electrons getting sent in and “bounced out” it is actually the dense central nucleus, made up of protons and neutrons, that interacts with the electrons of the beamed wave the most. Because electron clouds are so diffuse, they’re easier to pass through without hitting anything, especially when the thing you’re passing through is as tiny as an electron. The nucleus is a lot harder to pass through, so that’s where the scattering comes from here. Thus, “heavier atoms” with lots of protons will scatter more. But most of the atoms in biological molecules (carbons, hydrogens, nitrogens, oxygens, etc.) aren’t very heavy, so they don’t scatter strongly.
One solution to this is “de-focusing” – basically you take some images slightly out of focus – meaning that instead of focusing the bent electron beams directly onto the focal plane of the detector, they focus them slightly in front of or behind the detector so that they’re a bit “blurry” when the image gets captured. This may seem pretty counterintuitive that you’re intentionally blurrifying something – but it improves the “contrast,” making it easier to tell the background apart from the true “particles.”
More on the “how/why” in a minute, but sticking to the workflow story for now, that de-focusing allows for “particle picking” – in this stage of the process, the computer (with help from the researchers) decides what is an actual particle (e.g. the protein, multi-protein complex, protein/DNA complex, etc.) versus what’s just the background. Because the signal from any one particle is so weak, it tries to pick a lot a lot of them.
Then it tries to sort them based on which look similar. Each image contains 3D info, but the computer at this point treats them as simple 2D images. So it goes through and groups the similar-looking ones together into a number of 2-D classes – usually these 2-D classes are different orientations of the molecules (e.g. top view, side view). The researcher tells the computer how many classes to make & it uses a translation-rotation function – basically it shifts and turns the picture to see how well it fits with the others in the class and finds the best match.
Then it’s on to 3D classification. The images in the 2D classes are used to build a rough “map” of the 3D protein they came from. The reason this is possible is because of something called the central projection theorem, which I’m not going to get into.
You can further refine and refine the map and get a final map and then build an atomic model into it.
So the basic workflow is:
specimen preparation: sample -> freeze it on a grid
imaging: take a LOT of images – go from hold to hole in the grid taking lots of pictures
particle picking: determine what’s actually a particle versus just background
alignment and 2D classification: group together the images that look similar to get 2D averages
3D reconstruction: use those 2D averages to generate an initial 3D map (helps if you have a low-res negative stain or something)
refinement: refine that initial map to get a refined 3D map
make it even better to get your final 3D map
stick an atomic model into it
Back to that contrast problem and how de-focusing can help. The technical explanation is complicated – involving a lot of math-y stuff I’m not going to get into, but the way of thinking about it that makes sense for me is something that a lot of you can probably relate to – an optometrist exam. You’re tasked with reading small letters across the room as the doctor swaps the lenses in front of your eyes (or do they? I always feel like they’re trying to trick me when they do the do you prefer “1” or “2” or “2” or “3”).
Your eye has its own biological lens, tasked with focusing light on receptors on your retina. But some people’s eyes focus the light in front of or behind the retina, making the image appear blurry. Thankfully lenses can work together, so by adding “corrective lenses” in the form or glasses or contacts, optometrists can fix the focus.
But they have to find the right lenses – through this trial and error testing – and when they err, things appear blurrier. At the eye doctor, this is bad – but in cryoEM, a little of this such blurring is good.
When the lens is slightly out of focus, it gets hard to read the tiny letters, but you can still read the big letters – and importantly, even though you might not be able to make out the individual little letters (you can’t resolve them) you can still tell that there are letters there – you have contrast between the black letters and the white background.
In cryoEM your background isn’t a nice clean canvas – instead its filled with water molecules and stuff which makes it “noisy” – but the signal from these molecules is weaker than the signal from the particles you’re interested in. And when you blur everything a little (defocus the light) that signal kinda gets blurred out, making it look like more of a blank canvas. But you still see the stronger signal from your particle, now with better contrast. Unfortunately, you also lose the weaker signals from your particles (which even more unfortunately is the highest-res stuff) so you don’t want to defocus too far. Usually what’s done is scientists collect images at a few different defocus levels to capture a range of information. These are taken at different locations in your grid because you’re limited as to how many pics of one particle you can based on how much radiation it can take before it’ll “break”
Speaking of that breaking – another problem cryo-EM has to worry about is radiation damage. The coldness helps protect your sample somewhat from “frying” – but there’s only so much it can do. Ultimately, your samples do suffer from the electron beating they’re subjected to, so you can only take a limited number of images from each spot on a grid.
Another problem – when a particle gets hit by an electron, it “hurts” – The first couple of hits are the worst in terms of jolting the particle – it gets hit and shifts a bit, introducing the very kind of motion during imaging that we need to avoid.
This is where “movies” come in. When I first heard about cryo-EM movies I thought the point was to see the molecules actually moving and doing their thing – that is NOT what it means! Instead, it means that instead of just taking a single snapshot of a particle it takes lots of them – really fast – and then it can compare different movie frames & correct for movement of the particle that occurred during the taking of the movie during the “filming”
Speaking of film – in the old days, the images were captured on actual film that had to be developed and everything. But another major innovation driving the cryo-EM revolution was direct electron capture technology – basically detectors that have lots of pixels that can measure how many electrons hit them per frame of the movies. Unlike in crystallography, where the highest achievable resolution is limited by the wavelength of light, since the electron waves are so much smaller, the wavelength isn’t limiting – instead it’s the pixels that provide the maximum resolution limit – the smaller the pixel, the better able to tell tiny differences in where exactly an electron hits the detector. So scientists are frequently trying to up the pixels.
But, just like in crystallography, most of the time you’re limited most by the sample itself – not that it’s always the researcher’s “fault” – some molecules are just tricky to image (and it’s not their fault either – you ripped them from their natural environment, remember and now you’re complaining they’re not smiling pretty for the camera?!).
Which takes us to some more differences between cryo-EM & crystallography. It’s not like you can always just randomly take your pick, or that, now that cryo-EM technology has advanced you should give up that crystallography stuff. Cryo-EM works well for big things that provide lots of contrast – so it’s great for big complexes. Which is basically the opposite of what crystallography’s good for.
The major advantage of cryo-EM is that it doesn’t require crystallization. Because crystallography requires that perfect orderliness, the fewer potential particle to particle “variables” the better – and the more parts there are the more such “variables” you have to get to organize. But, for small, well-structured particles, crystallography can provide you detail at higher-res. And it’s great for things like showing drug binding, etc.
But getting something to crystallize usually means a lot a lot of screening & optimizing and requires a lot a lot of protein. Plus, even if you do manage to get something to crystallize, sometimes the molecules adopt slightly “unnatural” positions due to crystal packing interactions.
But it’s not like getting good cryo-EM data is a walk in the park – it takes a lot of optimization too – as I’ve seen my colleagues go through!
An example of cryo-EM in action is the telomerase complex, tasked with keeping your chromosomes from shortening every time they copy themselves. I wrote a piece for Massive Science a couple years ago about a cryo-EM structure of telomerase from the labs of Kathleen Collins and Eva Nogales (who I’ve met and think is awesome). Then-post-doc Thi Hoang Duong (Kelly) Nguyen (who now is a group leader at MRC LMB in the UK) carried out a lot of the crucial work, and if you want to learn more about it, here’s a link to their paper: https://go.nature.com/35Leh9M
and my article on it: https://massivesci.com/articles/dna-telomere-cancer-fountain-youth/
But in the pics, I’m just going to use it as an example to show you some of their workflow
As I mentioned before, I personally don’t use cryo-EM. I’ve done some crystallography, but my real passion is for the more biochemistry side of things – figuring out how molecules do what they do at the mechanistic level. Don’t get me wrong – I’m not trying to discount structural biology at all – and I think it’s highly under-appreciated. Often seeing structures of molecules can be a huge help – as we saw when we looked at Rosalind Franklin & Raymond Gosling’s “Photo 51,” that blurry X of a DNA x-ray fiber diffraction pattern that led to the discovery that DNA forms a double helix able to store and transmit genetic information.
And with proteins, even if you know the linear sequence of amino acid letters, you don’t know which letters are actually near each other in the folded protein – but “structural biology” techniques can provide answers, showing you how/which amino acids (protein letters) in a protein are near each other and other molecules they’re acting on. etc.) So I really am grateful for structural biology and techniques such as crystallography and cryo-EM. And really hope they keep up the great work! I just personally, in my own day-to-day research, prefer focusing on experiments to measure activity of the proteins and that sort of thing.
more on x-ray crystallography:
more on vitrification: http://bit.ly/flashfreezingdance