Now that we've introduced tomography and talked about data collection and reconstruction, now let's talk about how to identify objects of interest within tomograms. The problem is that while tomography may produce beautiful three-dimensional reconstructions of objects as large, even, as an intact cell,. Still, how do we know what we're looking at inside? For instance, here is a slice through that reconstruction that I showed earlier. And we see lots of objects of interest, but how do we know what we're looking at? Well, the first line of answer is that some of the objects are simply obvious. So for instance, the outer membrane, the inner membrane, and then over here, the flagellum, these are objects that are easy to recognize because of all that's known about them, and this is true of other structures, as well. Here are shown tomographic slices through the very thin edge of human cells. And you can see for instance, actin fibers running through here. And then there are other fibers that are just a little bit thicker that must be the intermediate filaments, because they're thicker than the actin fibers. And the actin fibers are also intimately associated with the plasma membrane. And then microtubules look very different. Microtubules always look like two parallel lines. Sometimes you can see the individual protofilaments within a microtubule. But in a particular cross-section, they'll look like two parallel lines 25 nanometers apart. So some things can be recognized by their size and shape and positions. Here's another example of a tomographic slice through the very thin edge of a human cell. And so here's the boundary of the cell, and here's the material inside the cell. And if we zoom up on that, we find in this field of view, two very interesting objects here, and then there's another one that looks quite like it over here. And it turns out that there's a macromorphic complex that's been called a vault or a vaultizone and it's structure is known. It looks like this, and so you can recognize vaults within the thin edge of mammalian cells simply by their shape. And so the first way that objects in tomograms can be identified is that some things are just obvious from their shape. The next strategy is correlated light and electron microscopy, often called CLEM. The idea here is to put a fluorescent tag on and object of interest and then image, a cell, for instance, with a tagged object of interest in it, by fluorescence microscopy. And then, after the fluorescence microscopy, take another picture of the object with an electron microscope, and then correlate the light and the electron images. And see where the fluorescent spot lies in the electron microscope image. Now to illustrate that I'm going to use an example when we were searching for the chemo receptor arrays of bacterial cells. And what I'm showing here is a fluorescence image of an EM grid, and you recognize the circular holes on the EM grid an irregular pattern. And the on this EM grid we've deposited a thin layer of caulobacter crescentus cells. And so the caulobacter cells are the dark spots that span about the same size as the hole. And their shape is something like a banana. And so they're the slightly curved, ru, otherwise rod shaped objects across these holes. And in these caulobacter cells, there's been a fluorescent tag fused to a chemoreceptor. And so the fluorescent tag is mCherry, and so, in very many of these cells, you see a bright red dot at one tip of the cell. And that's because that's where that chemoreceptor is found. Now it's important to note that when this grid was prepared we applied a chemical that would make the grid sticky so that when the caulobacter cells hit the grid they would stick into one location. So these cells won't be moving around on the grid anymore. After that, we put it into the fluorescence microscope and recorded this image. Once that image was recorded, we took a pipette with buffer and we put the tip just underneath the cover slip and squirted additional buffer underneath the cover slip. And that caused the cover slip to rise a little bit and then slip off the grid. Then we're able to reach in with tweezers and retrieve the grid and take it out of that situation. And of course it still had a drop of fluid around it, so the cells remained in a native state. And then we took that grid and plunge froze it into liquid ethane. Once it was plunge frozen, then we stuck that same grid into the electron microscope and looked with low magnification to find the same grid square that we had imaged here by fluorescence microscopy. And here is a low mag EM image of that same grid square. Now in the electron microscope the holes in the Quantifoil film appear bright because at least in this hole there's very little ice, if any, over that hole, and so lots of the electrons pass through. Then in other holes they look darker because they're covered in ice and the cells retain water around them. So the cells are the darkest objects of all because the ice is thicker right around the dark cells. Now in this movie we start with that EM image and we gradually morph into the fluorescence image. To illustrate how there's a clear correlation between cells that you see in the EM image, for instance, this one, and the cells that are seen on the grid in the fluorescence image. You see that's the very same cell, and this is true of other cells. For instance, this one down here, we can see in both the EM image, and now the fluorescence image. And here in another example, you can see a cell in the electron microscope image and clearly identify which cell is the same one in the fluorescence image and you can see which tip of the cell has the fluorescent markers. And then we could look cell by cell at where was the fluorescent marker within the cell, compared to what were the structure that were seen in the tomogram of that same cell? And we found that every time we saw this nicely ordered array that was parallel to the inner membrane here, that is were we saw the fluorescent spot. So here's an example from this cell. Here's another example from another cell where you see the fluorescent spot here. And then sure enough right at that side of the tip of the cell again we see the array structure. And this allowed us to identify this array. As containing the receptor that had been fused to the mCherry tag. Now to give another example of this CLEM strategy, there is a structure in caulobacter crescentus cells that looks like this. It's called a cross-band. So caulobacter has these long stalks at the tip of the cell. And it's been observed by cryo tomography that the stalks are sometimes divided by these cross-bands. And once there was a good hypothesis about which proteins formed those cross-bands, investigators fused fluorescent tags to those proteins and this is an example of an image that we took, this very small image inset in the corner, is a florescence image of this same stalk on a caulobacter cell with two red spots where the mCherry fluorescent tags. Were found. And by superimposing this fluorescence image onto this EM image, over here we see that the fluorescence spots perfectly superimpose at least to the resolution of light microscopy with the crossbands here present in the stocks. So there's one example. And there's one more crossband here in this structure. And there it is, also marked by a red fluorescent spot. Now the localization precision of this florescence microscopy was very good, because we were able to do the light microscopy on a sample that was a room temperature. And when it's at room temperature, one can bring the objective lens into, almost into contact with the cover slip and put oil in between them. So you have an oil immersion objective lens with very high numerical aperture up to about 1.4. And because of the very high numerical aperture, you can get high resolution fluorescence images like these. But this strategy won't work for objects inside of a cell that are dynamic and might move in-between the fluorescence microscopy and the electron microscopy. For such dynamic structures, we would prefer to have a technique where we could first freeze the cell, then image it in the fluorescence microscope. And then because it was frozen, no molecular rearrangements would occur in between the fluorescence microscopy and the electron microscopy. However, to image a cell that's already frozen in a fluorescence microscope, prohibits us from using an oil immersion, high numerical aperture lens at room temperature. Because we can't have a frozen sample at about 80 Kelvin in direct contact with oil, which is in direct contact with the objective lens if the objective lens and the oil are at room temperature and our sample is at 80 Kelvin. In an attempt to overcome this problem, we pioneered a so-called super resolution microscopy technique that could be applied to frozen samples held at 80 Kelvin and we called it cryo-PALM. In the following movie, we'll see correlated cryo-PALM and electron cryotomography. So here we see a bacterial cell frozen across an EM grid, you'll recognize the quanta foil holes. And so here's the bacterial cell, it's already frozen on the grid. And now we're going to see a series of images that were recorded in a cryo-light microscope. On the sides, there's fluorescent bead that acts as fiducial markers. Okay. There's four fluorescent beads and then here is the cell in the middle. Now what you see is individual blinking of individual photoactivatable GFP molecules being activated and then fluorescing and then bleaching away. And from that kind of an image, you can find the centroid of each fluorescent signal and build a composite super resolution image. After that sequence of images was recorded, we then moved the cell into the electron microscope, where here you can see the polystyrene beads. And if we superimpose the cryo-PALM image on the EM image, you can see where in the cell the bright fluorescence was. Once the cell was in the electron microscope, we could record an entire tilt series. And so here's the tear, tilt series of that cell. Now you see some very dark spots blinking around the cell here, because some of the ice crystallized from the laser illumination. Nevertheless, the structure inside the cell was very well preserved. So here's a slice by slice view of the tomogram that resulted. And the tomogram has a lot of interesting features storage granules. There are ribosomes that you can see, fibers of the nucleoid, cytoskeletal bundles are present. And one of the objects of interest was a type six secretion system and so the, the photoactivatable GFP tag had actually been fused to a protein of the type six secretion system. And here, you can see those secretion systems. These tubes that they form inside the cell. And note that the cryo-palse, PALM signal, nicely correlates with the positions of the tubes inside the cell. And these tubes had recognizably the ultrastructure of a type six secretion system in both the extended and then the contracted state. And then one of the most striking examples of this correlated cryo-PALM electron cryotomography. The cryo-PALM signal was a nice round spot that almost exactly superimposed on a nascent type six secretion system that was just beginning to assemble. So coming back to the question of how can objects in tomograms be identified? We now also add to the list, correlated light and electron microscopy techniques. The next method is perturbing the abundance or the structure of something inside the cell. The example that I'll show concerns a cytoskeletal filament formed of CTP synthase and CTP synthase forms filaments that in caulobacter crescentus cells, line the inner curvature of the cell. So this is caulobacter crescentus cell. It's a tomographic slice through the middle of such a cell. And like I mentioned, caulobacter cells are roughly banana shaped. And along the inner curvature of these cells from the far outside, we see first a surface layer, then the outer membrane and then the inner membrane. And inside the inner membrane, we see a series of filaments that are stacked up on top of each other. And we hypothesize that these were made of the protein CTP synthase. Now, it turns out that there's a small molecule called DON that depolymerizes CTP synthase. And so we added DON to the cells and we re-imaged them. And in these DON treated cells, we didn't see any of the filaments present on the inner curvature. On the other hand, if we mildly over-express CTP synthase, then we saw much longer filaments and the stack of filaments was much deeper. And finally, if we over-express CTP synthase in the extreme, now we saw 100 of filaments. In fact, so many that they no longer lay in neat stacks of ribbons, but rather the filaments flowed out in a less structured pattern out into the cytoplasm. And so because the abundance of this filamentous structure correlated well with the expression or the depletion of the CTPA's filaments, we concluded that these filaments were, in fact, composed of CTP synthase. Now this same strategy can be applied even to a specific macromolecular complex. In this example, this is a sub-tomogram average of a flagellar motor from campylobacter jejuni. And as you can see, it's a complex structure that involves 40 or 50 different proteins. And one of the densities at the bottom of this structure, we thought was the protein fly eye. And so to test this hypothesis, we imaged cells where the gene for fly eye had been deleted. And when we imaged those motors and produced another subtomogram average, sure enough, the only major difference in the structure was the absence of that density. And so this allowed us to conclude that that density was, in fact, the product of the fly eye gene. This next example comes from a project, in which we were hoping to visualize the higher order structure of DNA inside a living cell. And so, we recorded this tomogram. And this is of a cell. And it's the thin neck region between a mother cell and a daughter cell. So the way these cells divide is there's a mother cell, for instance on this side, a very large body of a mother cell. And then, when it comes time to divide, it extends this neck region and then on the other side of the neck, the daughter cell forms. And because of this configuration, we knew that a copy of the chromosome had to flow between the mother cell to the daughter cell, through this very thin neck. And so we plunge froze these necks and we imaged them by tomography, and this is the structure that we saw. So in it's a gram negative, so this is the inner membrane. And the inner membrane forms, of course, a tube here in the inside. And this is the outer membrane, and so this is a double membraned tube, and passing through that tube, we saw this helical structure that we thought might be DNA. Now, it looks like a duplex DNA. But the scale is not appropriate for that. The scale actually is appropriate for two duplex DNAs that are super coiled around each other as our, is known for DNA inside bacterial cells. The scale is more like two duplex DNA supercoiled. Nevertheless, to probe in an attempt to either confirm or falsify the hypothesis, that this was in fact a DNA structure, we added a DNA intercalator, ethidium bromide. That we thought would disrupt the structures of DNA inside the cell and reimaged it. And so here was the wild type unperturbed structure in which the object looked like helical twists. Then after adding ethidium bromide, we saw a similar structure, at least in its dimensions. But there is no longer a twist. Instead, it looked like a series of lines. And, again in another example, the twist was removed, and it looked like, like a ladder. So, while not conclusive, this at least supports the hypothesis that this structure did contain DNA, because adding ethidium bromide, which is a known intercalator, this changed the structure. And so the third strategy on our list of ways that objects in tomograms can be identified, we had perturbing the abundance or the structure of that object. The next method we'll discuss is template matching. And the idea of template matching was well-illustrated in this figure from Achilleas Frangakis. The idea is that assuming we have a tomogram of, for instance, a cell, and we wanted to search it for objects of interest. If we had a known structure of some of those objects, these are the templates, and we might know the structure from, for instance, x-ray crystallography. So we know the structure of some of the complexes that we're looking for. We could take these known structures and search the tomogram for densities in that tomogram that looked like our templates. Now this search would be performed through a cross-correlation function. So we would need to do the cross-correlation for each of the possible orientations of the templates. Because the template would need to be rotated in all possible orientations. And for each rotation, we do cross-correlation to find where within the volume are there objects that look like that template, in that orientation. Now, we might do this for a number of templates. Obviously, this is computationally very intensive, and so you would do this on a powerful, parallel computer. But then you could take for each position within the tomogram. If there were any high correlations between your templates and the object in the tomogram, you could pause it that, that particular template lay in your reconstruction in that orientation and build up a map of where all the objects were. A pioneering application of this strategy was reported by Kuhner and his colleagues a few years ago. They recorded cryotomograms of mycoplasma cells, which are very small and have simplified genomes. And these cells are characterized by an attachment organelle that it protrudes from the main body of the cell. And here in the main body of the cell, these authors apply template matching. And the templates that they looked for included the very largest, included the largest macromolecular complexes that they expected inside the cell, for instance, GroEL, the chaperone GroEL, and RNA polymerase. They looked for bacterial ribosomes and the pyruvate dehydrogenase structural core. So cross-correlation searches through the interior of these cells for these templates pointed to certain objects inside that were likely individual molecules of these templates. Now, the most ambitious application of template matching might be to take an entire cell, such as this. And instead of looking for specific templates of interest, perhaps one day, we'll be able to simply classify all of the densities inside this cell into different classes and then recognize which classes are which macromolecules. And this effort to discovery which macromolecules exist inside the cell and all their spatial relationships by direct imaging has been called visual proteomics. Now the last method that belongs on our list of methods for identifying objects in tomograms is the possibility that some day we'll be able to introduce heavy metal tags onto an object of interest. For instance, people have tried to fuse metallothionein to an object of interest and then have the metallothionein collect metal atoms inside the cell into a metal ball. And then by visualizing where the metal ball is, we would know where that object of interest is. Other ideas have been to use ferritin this way. And to date, this approach has only met limited success. But nevertheless, the hope remains that better metal tags will be developed that could potentially allow us to pinpoint with high certainty objects of interest.
