Cryo-electron microscopy breaks a key barrier
nature.com
nature.com
Cryo EM is a very hot method right now to determine the three-dimensional structure of large molecules like e.g. proteins or protein complexes. Something like 10-15 years ago the best you could get was maybe 1.5 - 2 nm (these are rough numbers from memory), the results described in the article are 0.17 nm.
This is phenomenal! The title still feels a bit sensationalized, but don't they all?
One part of the CEITEC research center does "dry" tech - chemistry, robotics, motor and machine control and chips. Of course, they have quite a lot of cutting edge electron microscopes to image various nano structures. All looking super high tech as you watch their shiny metal over the clean room window. All the machines stand on concrete slabs embedded into the hill below the building to avoid vibrations and have ultra high vacuum inside.
The other part of CEITEC does "wet" research - proteins, cells, bacteria, microorganisms. Of course also they have cutting edge, cry electron microscopes. These looked quite different to the other ones - it was a huge black cube, reaching almost the high ceiling of the room. You could not see all the glistening internal components like in the previous case. In this case they even let us to the room itself - apparently given all the volatiles in the samples, the cleanliness requirements in the room containing the machine are not as strict & it likely (my guess) needs to keep quite a bit part of the machine at very low temperatures, explaining all the covers. IIRC the software controlling the camera runs on Linux & engineers from Brno working for one of the big EM manufacturers were involved in building this enormous cryo EM machine, which was a pretty nice touch. :)
I did want to point out though that this title is misleading. The researchers didn't get an image of an atom. Instead, they reconstructed hundreds of thousands of images of the protein to determine where the atoms are almost exactly. So there's no actual image that shows an atom, as cool as that would be.
Single particle gives you the opportunity to see different conformations, but only if the data is discrete. If there's a continuous amount of conformations (think a molecular motor that's rotating) you would need nearly infinite data to resolve a nearly infinite number of conformations. If the data is less than continuous, you can image enough particles to see all the different conformations by constructing multiple models in parallel and using 3D angular searching to bin them by what conformation they are in. This is a computationally exhausting process, however.
All that aside, crystallization certainly biases it towards specific conformations, which single-particle EM does not.
Exposure is a separate variable because it also causes sample degradation. Ideally, a microscope is only shooting one electron at a time at the sample (that never happens obviously), but the fewer electrons, the better. On overexposed micrographs, you can see the burn marks on the sample.
I'd recommend, if that's what you wanted to do, to start with biophysics or molecular biology as those are the hardest aspects. A good computer science background is helpful as well, because a huge part is processing terabytes upon terabytes of micrographs into a finished, ab initio protein model. There's always room for new techniques there.
Personally, if I were to do it over, I'd major in the closest thing to biophysics/structural biology I could, and spend as much time in a lab doing undergraduate molecular biology research as I could.
You could say the same about any image produced by a scientific instrument these days. Any astronomical image has a huge amount of data behind it, but we don't sit around saying "gee, there are no actual images, isn't that a shame".
If you don't want to call something an image, you can call it a diagram produced by a data processing program, but whatever you call it, it's disappointing if it's not provided.
(Submitted title was "Cryo-electron microscopy technique sees individual atoms for first time".)
With the presented method the structure sampling time seems to be O(10 s) which easily is about 10 orders of magnitudes slower than the dynamics we're interested in seeing.
A direct consequence of this is that for achieving atomic spatial resolution they needed to use a "rock-solid" protein -- one that has an exceptionally stiff structure (one that does not wiggle a lot). The presented method is great and cool, but this is a pretty severe limitation. Most proteins wiggle a lot :-).
Background: protein structure-function relationship can often be well understood only when considering the structural dynamics of the protein (key words: conformational changes, the entropic contribution to free energy).
That is, in the ideal case we would be able to measure molecular structure not only at high spatial resolution, but also at high temporal resolution.
How can one make such a measurement much faster, by ~10 orders of magnitude? By irradiating a lot of light. Via X-ray free electron lasers (XFEL).
XFEL-based techniques are expected to revolutionize structural biology (as always, also still a long way to go):
> XFEL protein crystallography not only determines high resolution structures of proteins, but also reveals the time-stamped conformational changes of proteins.
- https://www.nature.com/articles/nmeth.3070.pdf?origin=ppub
- https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6678726/
(PhD in structural biology/bioinformatics, investigated dynamics of proteins with molecular dynamics simulations)
(I work on instrumentation improvements for XFELs and UED.)
There exists a race in the structural biology community about the next big method that allows to determine structures of proteins that were hard to crystalize and it seems that CryoEM is becoming the winner in this race.
As an alternative approach, people are building large X-ray lasers that have extremly high intensity and short pulse lengths which they plan to shoot at single particles and resolve individual scattering images. This method can very likely also achieve atomic resolution of non-crystaline particles, even without the need to freeze them down.
It will be exciting times for the whole bio-chemical physics community. Congrats to Holger and the team for another great publication in Nature.
Relevant; they used Cyro-EM to elucidate the structure of the SARS-CoV-2 Spike protein [1], which is a big step towards the development of a vaccine. This is a resolution of 3.5 ångström, but still a very nice feat.
They link the full labelled 3D model in the article's footer: https://www.rcsb.org/structure/6vsb - very cool.
If I'm wrong, it would be great if someone could ELI5.
But, well, I'm having one of those days. :(
"Everything is everything."
Because the outer shell of electrons is fuzzy, there are multiple ways to define radius: covalent radius, Van der Waals radius and ionic radius. In molecules atoms overlap in different ways because they share electrons and electrons interact.
There still exists fundamental uncertainty about the location of atoms and molecules. Electrons, atoms and molecules have de Broglie wavelength. For molecules with tens of atoms the wavelength is in the order of picometers.
This is a bit of a long winded list of discoveries. There is no simple sophism you can utilize and say "atoms are this or that" unless you actually understand something of the background.
This Khan academy lecture is as good resource as any:
https://www.khanacademy.org/science/chemistry/electronic-str...
The electron distribution falls with an exponential function on large distances, what is much cleaner than a tree surface anyway.
There are naturally a lot of views about this topic, but one that resonated with me (perhaps recently discussed in HN) is that science can't necessarily make things intuitive, because it has this nasty habit of telling us when our intuition is wrong.
If those line segments are probability distributions of shared electrons, that means our high school mental model of a covalent bond (drawing a line between two atoms) is not too bad.
https://kpwulab.wordpress.com/2018/12/13/a-1-96-a-cryoem-map...
https://cryoem.med.ubc.ca/image-gallery/
So it's like the x-ray version, just a different approach / energy scale? Or is that comparison ... doesn't make sense?
- In X-ray crystallography, you're collecting the amplitudes of the Fourier transform of the electron density in the repeating unit of the crystal. The fact that it's a well-ordered (usually) crystal effectively "amplifies" the FT, and you get a series of images like this (from one orientation of the crystal): https://cdn.britannica.com/02/147302-050-3F732246/X-ray-diff...
- In EM, you're collecting direct images of particles randomly distributed on a grid, but they're 2D and individually very noisy, like this: https://www.pnas.org/content/pnas/110/45/18037/F3.large.jpg?...
In either case, there's a lot of complex math that goes into reconstructing the electron density - but the end result, the electron density map, is very similar (aside from the absence of crystal packing interactions in the EM map).
(For the record, there is also something called neutron crystallography, which works like X-ray crystallography except it's the nuclei that diffract, not the electron clouds - which allows you to visualize hydrogen atoms more directly, and even resolve the difference between hydrogen and deuterium. But it's another speciality technique, in part because neutron sources are so much weaker.)
EDIT: since I actually did some work on visualizing electron density maps on the web, here's an interactive view of an older X-ray structure (from 2003, at only 3Å resolution) that shows how a protein molecule fits into the density: http://natechols.github.io/xtal.js/map_viewer.html (click and hold the middle mouse button to pan through the molecule)
The grids are isosurfaces of electron density. Analogous to isolines on a map illustrating mountain terrain height.
Atoms are sticky little fuzzy balls. The fuzz, electron density, falls off exponentially. They don't have an "edge" in a non-fuzzy-object sense. Analogous to a stereotype volcano having an arbitrary perimeter. Where would I place it? Where you need climbing gear? Where biking up gets hard? In the foothills? Somewhere on the sloped plain? You place the perimeter wherever is convenient for what you're trying to use it for. People are usually interested in the densities between sticking atoms, which are several orders of magnitude down from peak density. Analogous to drawing an isoline at "the towering volcano is here high enough to start stubbing your toe on it". The picture draws isosurfaces of electron density at something like "around here one atom would 'bump' another, and you can see bonds" and at "I wrap a single atom only, but still have an interesting shape".
To show the various features of interest, one draws multiple isosurfaces. As with a terrain map. But an exponential electron density scale is harder to represent than linear height scale. And it's hard to clearly draw more than a couple of nested 3D isosurfaces. So instead of placing many iso's at linear steps, here there are just a couple, more like an order of magnitude apart.
Isosurfaces and nonlinear color gradients are educationally problematic. And fuzz gradients are hard to visually stereo fuse. And on near-term AR displays, black is transparent, and background environment colors can't be manipulated, which limits the space of attainable colorizations.
My current thought is to approach it as a volumetric light source. Composed of several linear scales, each a different color, with density as transparency. And respecting their true density when combined, so higher density scales "shine through" lower. We'll see. Maybe overlay random sample "sparkles" to maybe improve sense of spatial extent.
I've not seen much similar work. If you know of any, I'd love to hear of it. Also, it seems there was a push some years back to emphasize electron density when teaching intro chemistry. A push that doesn't seem to have gelled. I've guesses, but I'd value thoughts on why.
Random asides: 1. once you're used to it, the simple isosurface mesh is super easy to work with (i.e. build molecules into), and for structural biologists the broad approximation is perfectly adequate 2. even so, the amount of unexplained blobs - some of which are definitely not noise - in my (xray) electron density maps was always both fascinating and frustrating. that's one reason why we don't spend much time on fancy renderings of experimental density at anything worse than subatomic resolution, because you're just sharpening the features that your atomic model doesn't explain.
> once you're used to it, the simple isosurface mesh is super easy to work with
One wouldn't know it from the state of content, or even of much education research, but educational representation is much harder than professional. Students are unable to untangle features reflecting careful correctness, from artistic license. So both seed mis/conceptions. Professionals can downregulate misconceptions... though that can be surprisingly localized - asking first-tier astronomy graduate students "What color is the Sun?"... gets you a common misconception.
Consider that ball-and-stick wrapped with electron density in Fig 9. Imagine coming at it cold. What is that stick? Well, maybe it's a ridge in electron density, thus symbolizing the surrounding not-so-stick-shaped region of increased density which constitutes a bond. But then what are those balls? They're way too similarly sized to be electron density. Ok, maybe they're spherical crosshairs for nucleus location, and the stick is a linear crosshair for the ridge. Variously sized because... something. Oh, no, some of the sticks are doubled, and density certainly doesn't have two ridges, or (here) increased density. So we've a paper notation that badly misrepresents actual electron distribution, blended into a physical representation with a... my head hurts.
I saw a professional chem ed content discussion yesterday, around a misconception, that in a two-species ionic solid, one atom bonds to another, in pairs. Rather than to "4 to 6" neighboring atoms... because one common printed diagram draws 4 neighbors, and another 6. It was like they were non-scientists, thinking they could wordsmith their way to correctness, without the slightest need to examine the actual characteristics of the real physical systems being described, or to consult with someone deeply familiar with them. The focus remained on models, decoupled from reality. I see a lot of that.
Chemistry education research describes chemistry education content using adjectives like "incoherent". Maybe XR can serve as an excuse to do better?
Even the developers of these visualizations would probably agree that it's dangerous to rely on them too much, and they are only a way to convey specific information in a way our eyes and brain can quickly process. Crystallographers in particular tended to over-rely on those chicken wire views and that plus software limitations yielded a lot of very poor-quality structures with poor atomic packing. The developer of those ribbon diagrams (Jane Richardson) has done a lot of other work to educate the field about how to visualize packing and other molecular properties and avoid screwing up the analysis. Over the long term, I think constant self-criticism makes up for the occasional sloppiness in scientific research.
Consider an illustration of the solar system in some introductory astronomy content. Let say, that not atypically, it misrepresents sizes, positions, orientations, lighting, and Sun color. Thus creating and reinforcing misconceptions known to be a problem in K-12, in undergrad, and even on into astronomy graduate school. In contrast, consider a line representing an Earth-Mars transfer orbit. It seems unlikely that students will say think there's a material long pole there, and worry about it hitting satellites and cities. Students and teachers are both clear on the aphysicality of the line, but not of the Sun color. Hmm, though using a minimum-energy line to represent an energy landscape is a source of misconceptions.
Similarly, common representation of atoms and molecules are known to be causatively associated with the stew of misconceptions that pervade students and teachers of chemistry. Ribbons, perhaps not so much. Hmm, though also similarly, they're often used without indication of regional flexibility, and so perhaps contribute to the underappreciation of configuration landscapes, of the importance of tuned floppiness. Perhaps.
Consider a currently implausible goal, of science education which accessibly describes the physical world, and conveys a transferable understanding of it. Arguably its content would look much more like scientific visualizations than content does at present. But being for education, it faces additional constraints, challenges and tradeoffs, distinct from those of professional scientific visualization. Creating such would require a collaboration of both deep scientific and educational expertise. For which very little incentive exists at present.
I'm trying to come up with an XR-compatible visual representation of electron density, that is physically correct, accessible, and bears in mind patterns of misconceptions in chemistry education. That task should probably be more than one random software dev's lockdown hobby hack. But as far as I know, that's were we're at. The related NSF-funded work I've seen... doing this bit well wasn't their focus. Same with XR ed tech side. Same with chem ed apps. And scientific visualization programs. And chem ed research. There may well be something nice out there, but I've not yet seen it. And big ed side... I was chatting with a leading textbook publisher, which onboards content creators with the indoctrination that their liberal arts background, and complete unfamiliarity with science and tech, is not a problem... because there's "a scientist" on call. Deriving electron density, with current nice python libraries, and GPUs, has surely gotten vastly easier than it was with old fortran messes, but still... it seems something more is needed here. Some societal staffing seems missing. No?
Last year, Aricescu’s team used cryo-EM to map the protein to 2.5 ångströms. But with the new kit, the researchers attained a 1.7-ångström resolution, with even better resolution in some key parts of the protein. “It was like peeling off a blur over your eyes,” Aricescu says. “At this resolution, every half ångström opens up a whole universe.”
[1] https://en.wikipedia.org/wiki/Scanning_tunneling_microscope
http://pubsapp.acs.org/cen/coverstory/83/8348atoms.html?
> On Oct. 11, 1955, Pennsylvania State University physics professor Erwin W. Müller and Kanwar Bahadur, who at the time was a Ph.D. student working with Müller, made history by being the first people to image individual atoms. The scientists were using a relatively simple and inexpensive instrument, and with it they directly observed individual tungsten atoms at the tip of a sharply pointed tungsten specimen.
When the microscope was invented, people made the same question: What is the use of it? Just watching things we already know bigger?
That is a recurring question. So much that Aristotle created the book MetaPhysics(meta meaning different from the Physics book) with the idea that this particular book did not need to be useful to be created. It was a compilation of non useful things.
This non useful compilation helped create science as we know it.
I still like NMR, because you're looking at proteins in solution. But if the freezing is good in cryo-EM, maybe even that becomes moot.
The development goes real fast with both SW and HW.