Data collection (imaging)

Posted 2022-12-11 klausage

Now that we‘ve talked about sample prep, let‘s talk about imaging.

0:05

In a single particle project, the images can either be collected manually or 

automatically by a program written to do that. 

For instance, the one I‘m going to use as an example is Leginon, a program for 

automatic data collection. 

The Leginon imaging approach begins by recording 

what is called an atlas of an entire grid, and that‘s shown here in the first panel. 

And what you see is a bunch of green squares here in a lattice, 

and each of these squares was an individual image 

recorded by the microscope at a magnification of approximately 120 times, 

where you can see multiple individual grid squares in this image. 

And then lots of images like these are tiled together 

to form the atlas of the complete grid. 

So using the atlas one can inspect the whole grid and

detect regions where the ice is just right, or where the ice is too thin or 

too thick, and then select regions of interest. 

And so here if we consider this one of the regions of interest, in this image we can 

then further select specific grid squares where it looks the ice is just right. 

So this grid square looks promising, and this grid square is promising and 

then so is this one and perhaps this one. 

See these have some ice around them. 

You can see because some of the grid square is covered by thicker ice. 

For instance this is a grid square where the ice is very thin along the outside but 

then there‘s something very dark in the middle, and so 

this is not a promising grid square. 

But these look just right for further imaging. 

And so one of these particular grid squares can be chosen, 

and then it is imaged at a higher magnification, say like 500 times. 

And so here in the picture of a particular grid square you can now 

see individual holes. 

And here we have some region of the grid where the ice looks too thick for 

high resolution imaging, so those aren‘t the best. 

But over here these holes look just right. 

And so one can choose several holes for 

high resolution imaging, and in addition to the holes that will be imaged for 

high resolution information on your particle structure, another hole can be 

chosen to reset eucentric height during the imaging procedure.

2:36

Okay, once that is done then a higher 

magnification image can be recorded of a single hole. 

So here this hole has been imagined and 

the blue square shows the picture that you see here in panel D. 

And this image was recorded at more like 5,000 times magnification. 

So here you see a single hole in the carbon film full of particles and 

ice that nicely covers that hole, 

there‘s a little bit of contamination here that you obviously want to miss, and 

the computer can select target areas for final high resolution imaging. 

So, for instance, one target area is chosen right here, 

another target area is right here, and the third target area is right there. 

In addition to that, one might choose a focus region next to those holes 

shown here in purple, and the beam can be deflected to this focus region and 

a very high dose image can be recorded. 

If you look at the power spectrum of that high dose image here, that‘s what‘s being 

shown here and inside the purple square in Panel E, and you can see

3:49

in the power spectrum of that image the various tawn rings in the power spectrum. 

And that can be fit to theoretical CTF curves to determine the focus and 

set it just right. 

Once the focus is set just right at the focus spot, 

then high resolution lotis images can be recorded on these targets. 

And so for instance, 

the image recorded here, shown in green, is now presented as panel F. 

So this is a high magnification final image of the region being recorded at say, 

50,000 times magnification, and this is the image 

where individual particles will be picked and further processed. 

Now you‘ll remember that the principle resolution limitation in cryo-EM 

is radiation damage, meaning that the electrons break the sample, and so we have 

to be very limited in the dose that we apply to a sample before the final image. 

And so an important principle is to look at 

how many times have we already imaged this hole? 

Well, once in the 5,000 times mag image, again we imaged it 

in the 500 times mag image, we also imaged it at 120 times mag. 

And so you can ask how damaged are the particles here 

before we ever get the final image? 

Well, remember that the dose rate falls off as the square of the magnification. 

So if you take a one second exposure at 50,000 times magnification, 

and you take another one second exposure even at the same beam intensity, 

but this time the image is only 5000 times magnification, 

the dose delivered here is only one percent of the total dose delivered here.

5:36

And the dose delivered to record this image at 500 times magnification 

is one ten thousandth of the total dose used for this final image. 

And so because these are low magnification images 

they can be captured without significant radiation damage to your sample. 

And so in a typical single particle project 

a grid will be inserted into the microscope, and then Leginon or 

some other program for automatic data collection, will record an atlas, and 

a user might select specific squares that they think are best for further imaging. 

And then the computer can go ahead and pick holes by itself, and 

record hundreds or thousands of high magnification images 

at just the focus values that are desired of the sample automatically. 

Now, of course, while this is happening, the investigator will want to 

follow the process and ensure that the right things are happening along the way. 

And for this, Leginon has what it calls the Leginon observer utility. 

And this is just a web page where you can show up to six different 

kinds of images that are being recorded during the procedure and 

just follow how the process is moving forward. 

A big advantage of using an automatic data collection software package is that your 

images are, without any extra effort from you, then organized into a database. 

And so in Leginon, Leginon has a database of all the images that it‘s ever 

recorded and their relationship to each other and 

other parameters about the data collection session along the way. 

For instance, it can track grid drifting during the session and

predict where samples will be moving, 

where targets will be moving during the data collection session. 

And afterwards, all these images are stored in a database, and so 

this is a web based viewing tool. 

So here is a list of the various images that the Leginon database has for 

a particular session, and you can click on different ones and see them. 

And for instance see the relationship between a high magnification data image. 

From the lower magnification images that show you where that was taken from and 

all of this is organized for you and can be processed and managed more easily.

8:10

Now, sometimes in a single particle project instead of just recording single 

images of a field of particles, sometimes focal pairs are recorded. 

And you‘ve seen this slide before, 

but it just shows a focal pair of a field of viral particles. 

So here‘s an icosahedral virus of this large round object. 

Here‘s another icosahedral virus, one of them is circled here. 

And this is another picture of the same field of viruses taken 

much further from focus. 

And below them you see the plot of the contrast transfer function. 

And we‘ve already gone over this slide before. 

But it just illustrates that there‘s different information content in this 

image taken at high d focus from this image which is taken closer to focus. 

And so, as we‘ll see later, in order to build a high-resolution model, 

sometimes investigators employ a strategy of recording 

focal pairs of the same field of particles, likewise, 

sometimes tilt pairs are recorded of the same field of particles. 

And so, here is a field of hemocyanin particles that were imaged first untilted. 

So the tilt angle is 0 degrees and 

here you can see all the various particles and I drew your attention 

to this image first because it‘s untilted unlike the others that we have discussed, 

but it‘s actually the second image that is typically recorded. 

The first image is typically recorded of a tilted sample, 

tilted to say maybe 45 degrees. 

And here is depicted the tilt axis and so what you can see, 

comparing the 45 degree tilt image to the untitled image and 

many of the particles have been circled. 

The same particles circled over here are the same ones circled over here. 

So you can see that in this tilted image, the particles have been compressed 

towards the tilt axis and here they‘re spread out more. 

And you can choose images of individual particles and show them, for 

instance here, as a 0 degree image, and then pair

10:24

them to the 45 degree tilted image here and you can pick this out and 

show what the particle looked at an orientation of 45 degrees tilted. 

And this allows one to produce a reconstruction using what‘s 

called the random conical tilt method that we‘ll discuss later.

10:45

Another variation on the same basic idea is to record the first image at, say, 

-45 degrees, and then a second image at +45 degrees. 

And another variation on that same basic idea is to record the first 

image at -45 degrees and then record the second image at +45 degrees. 

And as you might be able to guess, this tilt pair will have 

the advantage of being exactly 90 degrees apart from one another. 

And so, here‘s an enlarged version of the -45 degree image, 

here‘s an enlarged version of part of the +45 degree image. 

And as you can easily see, individual particles in this image can be 

paired with individual particles in the other image, 

giving us two different views of the same particle.

11:38

And so, in some variations of the basic single particle experiment, 

focal pairs are recorded or tilt pairs are recorded.

11:48

Now, unfortunately, when one takes a picture of a thin film of ice, 

the radiation causes changes in the specimen and 

causes some beam-induced specimen movement. 

In particular, vitreous ice within holes moves and 

flows and bulges with exposure. 

At one point my group was interested in the changes that happened to thin films 

of ice when they are radiated. 

And also when they‘re cooled either by liquid nitrogen or liquid helium. 

And when they‘re warmed up and transitioned between these. 

And what we observed, we had fiducial markers distributed throughout this ice.

And here‘s a picture of a particular hole and fiducial markers, and after 

one of these temperature transitions, we noted which fiducial markers had risen 

in the ice, and which ones marked in blue had actually lowered in the ice. 

So what you can see is that the fiducials in the middle 

bulged up with respect to the ones around the outside. 

And if we reverse that temperature transition 

we could get the opposite effect of the fiducials in the middle 

falling with respect to the ones on the outside. 

And these images are of a particular hole tilted to high angle, about 65 degrees. 

And here the ice is relatively flat and 

you see it from a high positive angle and a low angle. 

And then, after imaging and a temperature transition we see that now the ice 

has a dark bulge here right through the middle. 

See how this side of the ice is much darker? 

That‘s because it‘s bulging out away from that hole. 

And the negative tilt angle shows that bulge on the opposite side. 

So one of the major changes in the ice during imaging is that 

the ice will bulge within the hole and also flow somewhat. 

Now, this process is depicted in this figure from 2012. 

Showing a carbon film with vitreous ice surrounding here icosahedral viruses and 

the idea is that the electron beam, when it hits the sample, 

can cause many changes. 

One is that the diameter of the carbon hole can decrease. 

The carbon can actually stretch inwards. 

In addition, radiolosis and 

charging within the hole can cause this material to want to expand and 

as a result there‘s a doming effect of the sample within the hole.

14:24

Now, unfortunately, while the vitreous ice is bulging during the exposure, 

particles within the ice change their orientation and they flow into different 

positions as illustrated in this experiment imaging icosahedral viruses.

14:43

Here, they had a field of purified icosahedral viruses. 

There‘s a virus. 

Here‘s another virus. 

Here‘s another virus. 

And these are imbedded in vitreous ice across a hole in the carbon film.

14:56

And using a direct detector which can record images in rapid sequence, 

40 times per second or more. 

These investigators recorded a sequence of images 

called subframes of a single long exposure.

15:13

And then, because each icosahedral virus, 

it‘s a massive object with lots of intrinsic signal and also symmetry.

15:23

They could find the center of the virus 

in each of the subframes of this exposure series. 

In addition, they could also detect the relative orientation 

of each virus because of it‘s high symmetry, you can do that. 

And track how the orientation of the particle changed during 

the single exposure. 

And so, in panel A, what is shown are vectors showing the direction and 

the magnitude of the orientation change of the virus, 

how it rolled during that single long exposure. 

And note that the scale bar here, 

this depicts a one degree change in the orientation. 

So you can see that some of the viruses, their orientation is pretty well fixed. 

But other viruses are rolling up to a degree or 

more, in an apparently random direction. 

This one is very different than this one, say. 

As that ice bulges, the translations that the virus 

has exhibited during the exposure are depicted over here in panel B. 

And so here, the scale bar is ten angstroms, and 

the translations are shown here on the figure. 

So obviously, the translation factors are not on the same scale as the image itself, 

because the viruses more like 100 angstroms itself. 

But this translation of that much is just ten angstroms. 

But what you can see is that the viruses are moving within the ice. 

Most of them here are moving in approximately the same direction, although 

some are a little different, by ten or more angstroms during the exposure series. 

And, unfortunately, the shifts and the rotations that 

each particle will experience during the exposure is unpredictable. 

Here on the left is a picture of mitoribsosmes recorded. 

And each of these vectors is again, 

showing the translation that each particle exhibited during an exposure series. 

And if you just look at it, you see that the whole field of particles seem to flow 

up and to the right during the exposure. 

But this is very different than another case of the complex-I protein. 

Here its vectors as you look at the vectors around the field, 

you see some of them seem to be flowing out towards the edge, others are flowing 

in a different direction, and these are doing something yet again different. 

So the take home message is that during an exposure the ice bulges and 

the particles can flow and roll. 

Now over the years there‘s been a couple ideas emerge to help reduce 

beam-induced specimen movement. 

The first is that it was observed that including some carbon in each of 

the exposures seemed to reduce beam-induced specimen movement. 

The second idea is to use a more rigid grid material and so 

there are efforts on going to explore different kinds of 

grid materials that might strengthen the samples and reduce this movement. 

And so this is why one typically chooses to take pictures on the edge 

of a hole like this one and this one and 

this one rather than right smack in the middle of the hole. 

It‘s because it‘s been observed that if some of the carbon next to the hole is 

also exposed during the image there‘s a last beam induced specimen movement, 

and the reasons for that are still unclear.

19:01

Now, the good news is that if your camera can record images very quickly 

in a so-called movie mode where it records many sub frames every second, 

then images can be motion corrected computationally after the fact. 

And this is illustrated in this figure. 

Here is a picture of three viruses. 

And in this image, 

all of the subframes were simply averaged together without any motion correction. 

And so you can see that particles are blurry and the details are missing. 

However, if you use cross-correlation to 

track how each particle moved during that exposure. 

And then motion corrected so 

that all the images are more coherently averaged together. 

Then the particles are much crisper, with much higher resolution detail present. 

So in summary, while data collection for single particle analysis might 

seem pretty simple and straight forward to just record a projection image 

of different fields of particles, in fact it could be more complicated.

Sometimes you‘d like to record focal pairs, 

at other times you record tilt pairs. 

And at all times you need to minimize beam-induced specimen movement. 

And if possible correct it through motion correction.