Cryo-electron microscopy is a technique that’s been coming on for years and is continuining to take new territory in the world of structural biology. But there’s a related (and even more intense) imaging method that’s coming up behind it. The advances in cryo-EM (sensitivity, contrast, deconvolution of the data, etc.) are all directly applicable to this one: cryo-electron tomography.
The name tips you off. Just as X-ray tomography in medical imaging is the combination of numerous “flat” X-ray data sets into a three-dimensional picture, this method assembled cryo-EM data into pictures of the three-dimensional interior of the living cell. It’s an obvious thing to try – indeed, it’s an extension of how cryo-EM deals with individual proteins already, since you get a mixure of orientations on the sample grid that allow you to reconstruct a common structure. But dealing with that mixture of orientations in the context of flash-frozen cell innards is something else again. You’re obviously not looking at a purified protein sample any more, for starters – you have the biggest gemisch of stuff possible, with untold numbers of different biomolecules all dumped together. At first it seems like an intractable challenge, and progress in the early days of the idea wasn’t swift.
But there are some ways to cut the data set down to (semi-) manageable chunks. We know a lot about the large-scale organization of the living cell, thanks to traditional visible and electron microscopy and techniques such as freeze-fracture images. And we also have a lot of data about what sorts of proteins are localized to these different regions and compartments, giving you a place to start with the larger, more abundant, and more well-defined ones. In some of these cases, we have the significant advantage of having X-ray (or cryo-EM) structures already, so we know how things are likely to look. First among these is the mighty ribosome. There are a lot of ribosomes in any living cell, and they all look very similar, with large (50S) and small (30S) subunits, along with RNA polymerase. A vast amount of work has gone into the details of ribosome structural biology, culminating in the 2009 Nobel Prize, and they stand out in any high-resolution images of the cellular interior.
You can see some standing out in the left panel of the image, actually, which is a look at a flash-frozen Mycoplasma pneumonia cell. Taking those ribosome poses together and assembling the sub-tomograms gives you the image in the center, which is now a cryo-EM structure extracted from the interior of a living cell – well, living up until the time it was plunged into liquid nitrogen, but you know what I mean. And at right is the derived model, fitting in what we already know about ribosomal sequence and structure. This work actually helped illuminate the role of the NusA protein in this sort of transcriptional/translational coupling mechanism that you see in prokaryotes. DNA is going into the RNApol complex, mRNA is coming out the other end of that, and it’s being fed right into the ribosome for protein synthesis with the help of the NusA protein, which this work showed in detail.
Now this is clearly an optimized, cutting-edge case. We’re not going to be reading out the entire architecture of living cells any time soon, but at the same time, this shows that that (while very difficult) is not impossible per se. As it stands now, it’s quite a bit of work. That look into the mycoplasma was obtained by focused-gallium-ion beam milling, a technique that’s come on within the last ten years or so that prepares thin samples from the frozen cells by basically blasting away layers to give you a better view. That’s some on a separate instrument stage, and the resulting samples are then transferred (carefully!) to the cryo-EM rig. If we’re going to speed this workflow up (and we’re going to have to), those processes need to be even more automated and more closely integrated compared to their current artisanal formats. Collecting the cryo-EM data is a notoriously time-consuming process as well, although we have to keep in mind that we’re complaining about the time used to do things that not long ago were considered completely out of reach. As the short review paper linked at the end of the first paragraph also notes, there’s a built-in sensitivity problem with cry-ET, in that you cannot just blast away with the electron beam without significantly degrading the complex sample that you took so much time to prepare. So we need better detection methods to raise the signal-to-noise.
All of these things, and more besides (such as advanced techniques for working up the data) are the subject of active research, because the prize at the end of this is so huge. Over the decades we’ve learned more and more detail about what really goes on inside living cells, but there are still huge dark areas in our understanding, not to mention dark areas that we don’t even realize are dark areas yet at all. The idea of atomic-resolution pictures of how proteins and other biomolecules are assembled in real life is exciting and terrifying (man, is that ever going to be a lot of data to work through). But there is no substitute for the real system, the real cell.
This gets back to the subject of the recent AlphaFold protein structure predictions, which I wrote about here for the Royal Society’s Chemistry World magazine. My take (which has shown up on the blog as well, of course) is that this is a great achievement, but that protein structure per se does not wildly accelerate drug discovery – at least not in the way it’s practiced now. There are so many levels of knowledge to deal with on your way to treating a disease! There’s a biochemical pathway, full of proteins and cofactors and partners. All of these things have sequences, three-dimensional structures, and post-translational modifications. Those structures are dynamic, though, and will be powerfully influenced by those PTMs and the presence of various protein partners, and they will also vary in real time as the proteins carry out their functions. AlphaFold predictions are not going to tell you about those things; the only thing that will is something like this cryo-ET technique or a few other very advanced biophysical methods. Where AlphaFold’s predictions really will help is in sorting out what proteins we’re looking at – for X-ray and for electron microscopy, you really need some starting models to try to fit to the data. They won’t fit perfectly, because the models aren’t perfect, but they can get you started and let you see how the reality of protein structure matches up to your expectations.
And as we get more and more “ground truth” about protein structure and function in a real-cell context, that data will of course be fed back into future generations of protein prediction models. They will get better and better at extending to more dynamic states, to the wide universe of protein partner interactions, and so on – but this process is just starting. We’re not there yet, and we won’t be there for quite a while, because collecting solid data isn’t easy and there’s an insane amount of it out there to collect. But maybe, as we get clearer pictures, we can get to the highest level. We’ll know what proteins to look at. We’ll know when we’ve found them, and where. We’ll see their detailed structures, and watch how those change as the protein does its job (or fails to!)
And that highest level is understanding what we’re seeing when we get to that point. We’re going to have a lot of situations where we will be able look at some complex dance of interlocked proteins and say “Well, that really has to mean something, but I sure don’t know what it is yet”. That’s the hardest work of all, but we’re getting down to it. That’s the grand work of structural biology from here on out.
