Top

iBiology

Bringing the World's Best Biology to You

Single Particle Cryo-EM

Part 1: Single Particle Cryo-EM

00:00:08.01 Hi.
00:00:09.01 My name is Yifan Cheng.
00:00:10.01 I'm a professor of Biochemistry and Biophysics at the University of California, San Francisco.
00:00:15.05 I'm also an Investigator of the Howard Hughes Medical Institute.
00:00:18.13 Today, I'm going to give you a very brief introduction about the single particle cryo-EM.
00:00:23.24 This by no means is a comprehensive introduction.
00:00:27.17 For a more comprehensive introduction about electron microscopy or about cyro-EM, you
00:00:33.10 can find other more comprehensive material to read and watch.
00:00:38.03 Now, here, showing on the left, is a picture of an electron microscope and... which is
00:00:45.02 the one that we are still using in our laboratory.
00:00:48.07 So, this is not the latest state-of-the-art microscope, but is still the one that is giving
00:00:53.23 us the very best results.
00:00:55.21 And, on the right is a schematic drawing shows that... how the electron beam is moving through
00:01:01.12 the optical column and on the top is an electron gun, then it passes through the condensed
00:01:06.24 lens and objective lens, aperture lens, and then shooting on the specimen, then through
00:01:12.06 the projection lens and intermediate lens, and finally forming the image on the... on
00:01:17.00 the viewing screen or recording in the camera.
00:01:20.10 Now, to explain the electron microscope, one can go back to the wave-particle duality of
00:01:28.18 moving electrons.
00:01:30.18 And this can be described by the De Broglie's hypothesis, which one can view a... view a....
00:01:37.23 moving electrons as either as a particle or as a wavefunction.
00:01:43.08 As a wave, it can be described by the wavelength, and as a particle as a... as a momentum.
00:01:48.04 So, De Broglie's hypothesis linked these two things together.
00:01:51.19 Now, from this formula, one can easily calculate... calculate the wavelength of an electron passing
00:01:59.08 through... accelerated with a certain acceleration voltage and... using this formula, here.
00:02:05.09 And to make a little bit more complicated, one has to also take into the consideration
00:02:10.02 of wave relativity and... which also shows that the mass of an electron increases with
00:02:16.05 the highest... when the speed increases and... as described by Einstein.
00:02:20.00 Now, you... when taking those into the consideration, the wavelengths can be calculated by this
00:02:26.00 formula.
00:02:27.00 And, showing here, in the bottom here, is that, for an electron microscope operating
00:02:31.11 at 120 kV, the wavelength is about 0.03 Angstroms.
00:02:37.07 For an instrument operating at, say, 300 kV, then the wavelength becomes shorter, at about
00:02:43.01 0.02 Angstroms.
00:02:45.18 Now, if we compare electrons with an X-ray, which is also commonly used, for X-ray crystallography,
00:02:52.17 there are some major differences there.
00:02:55.01 And, as a particle, if we think of both as a particle, the electron inter... interacts
00:03:00.10 with an atom's shared electrons and also the nucleus.
00:03:04.05 And, for X-ray, it only interacts with the shared electrons.
00:03:08.05 Therefore, for X-ray crystallography, the density map is often called
00:03:12.04 electron density map.
00:03:13.17 But, for EM, the density map... an EM density map should actually precisely be called a
00:03:19.16 Coulomb potential map, or we simply call it the density map for short.
00:03:24.02 Now, also, an electron has a much larger cross-section than X-rays.
00:03:30.04 Therefore, electrons scatter much stronger than X-ray, in a way that it is also causing
00:03:34.18 more severe radiation damage than X-ray can... can do.
00:03:38.10 Now, as a wave... thinking of the electron as a wave, or X-ray as a wave, the electron
00:03:45.11 is a charged particle.
00:03:46.23 Therefore, they can be manipulated by the electromagnetic lens, which can change the
00:03:52.01 direction of electron beam so that it can... electron beam can be easily focused by using
00:03:57.14 electromagnetic lens, which is very hard to do for X-rays.
00:04:00.23 Because it's not a charged particle, there's no simple lens which can simply focus and
00:04:07.01 manipulate it by... for the X-ray.
00:04:10.04 That's why it is easy to make an electron microscope, but it's very hard to make an
00:04:14.18 X-ray microscope.
00:04:15.24 Now, taking... these are the simple differences of the... between X-ray and an electron.
00:04:22.22 Now, when you have a specimen, put it into the election microscope, what we're taking
00:04:29.04 the picture of is a two-dimensional projection image of a three-dimensional object.
00:04:36.08 And then... then... the question, then, is, how do you reconstruct back a three-dimensional
00:04:40.12 object from a two-dimensional projection image?
00:04:43.07 Now, mathematically, this... the principle underlying such a reconstruction has been
00:04:48.14 outlined long ago.
00:04:49.17 And... which has also been further demonstrated by David DeRosier and Aaron Klug in their
00:04:54.20 early work, back in 1968, to reconstruct a tobacco mosaic virus from
00:04:59.15 two-dimensional projection images.
00:05:01.08 Now, the principle can be illustrated as so-called central section theorem.
00:05:06.19 Now, in this theorem, what it says is that if you have a 3D object, a projection of this
00:05:13.02 3D object into a two-dimensional image will give you a view of this 3D object in certain
00:05:19.01 specific views.
00:05:20.22 If you do a Fourier transform of these 2D projection images, what it gives you is a
00:05:26.00 2D Fourier transform, which is equivalent to the central section of a Fourier transform
00:05:31.23 directly calculated from the 3D object.
00:05:34.14 So, in this way, if you have multiple views of the same object in different views, in
00:05:40.09 two-dimensional projection images, that if you can calculate Fourier transform, and if
00:05:44.21 you know the orien... the relative orientation of all these views, then you can combine them
00:05:49.13 together and reconstruct this one in real space, in 3D.
00:05:54.04 And then, in this Fourier reconstruction, it contains both amplitude and the phase.
00:05:58.18 The reverse Fourier transform will give you back the 3D object.
00:06:02.13 Now, this is the basic idea of 3D reconstruction in a mathematical term... way.
00:06:08.21 Now that... we also talked about that... the electron has a much stronger Rayleigh scattering
00:06:16.04 than X-ray.
00:06:17.07 This poses a couple of challenges, particularly for biological specimens.
00:06:21.17 The first one is that a high electron... high scattering making us... that we will have
00:06:28.02 to put the electron beam into a high vacuum.
00:06:30.04 That's why all the electron microscopes, the electron beam is traveling inside the
00:06:34.07 microscope column in a high vacuum.
00:06:36.08 Then, putting any biologic sample into the... this vacuum, the first thing you encounter
00:06:42.05 is the dehydration.
00:06:43.19 And the dehydration caused by high vacuum will causing... damage to the integrity of
00:06:49.06 the structural information... the structures.
00:06:51.19 The second thing is this radiation damage.
00:06:53.21 We... so, I also mentioned that an electron has a much stronger scattering than X-ray,
00:06:58.18 so therefore it causes much more severe radiation damage to biological samples than the
00:07:03.22 X-ray beam usually does.
00:07:06.12 Now, traditionally, classically, there is... there are a number of different ways to prepare
00:07:10.18 biologic specimen for... to be visualized in the electron microscope, including rotary
00:07:16.15 shadowing and various kinds of staining, starting from positive stains to the negative stains.
00:07:22.11 But all this technology messes... could not preserve biologic samples, it could not preserve
00:07:29.13 the structural integrity of a biologic sample to the atomic resolution level.
00:07:33.10 Therefore, these messes are not suitable for studying biologic samples to the atomic resolution.
00:07:39.17 How do we do that?
00:07:40.22 This was the first demonstrated by Ken Taylor and Bob Glaeser back in 1974.
00:07:47.07 They demonstrated that using frozen hydration is a way to preserve biologic samples into
00:07:53.20 a... to preserve the structural integrity to atomic resolution.
00:07:57.00 The idea is that, if you freeze a biologic sample and then that... therefore, the water
00:08:02.21 molecules can be kept in a solid state, then you can put this sample into the electron
00:08:07.03 microscope and this frozen hydration can preserve... can prevent dehydration.
00:08:14.10 Showing here is a diffraction pattern of a catalyst crystal prepared by frozen hydration.
00:08:20.00 And you can see that, here, the diffraction spots goes to 3 Angstroms, demonstrating that,
00:08:24.03 using this approach, one can actually preserve the structural integrity of a biological specimen
00:08:31.13 to the atomic resolution level.
00:08:34.10 And that... there is an also different way to preserve biologic samples, to prevent dehydration,
00:08:39.14 instead of frozen hydration.
00:08:41.09 That is also using sugar embedding.
00:08:43.05 Basically, the idea is using the sugar molecule to substitute the water molecules so that
00:08:47.22 it can be preventing dehydration, to maintain the structural integrity.
00:08:52.03 And Nigel Unwin and the Richard Henderson were the first ones, using this approach,
00:08:56.14 to study a... a... a two-dimensional crystalline array.
00:09:01.04 And then, in this case, they studied bacteriorhodopsin.
00:09:02.23 The idea is all these bacteriorhodopsins form a two-dimensional crystal.
00:09:08.13 Then this two-dimensional crystal enables one to record both electron diffraction and
00:09:13.20 also the image.
00:09:15.04 From electron diffraction, one can obtain the information of amplitude
00:09:19.05 for every diffraction spot.
00:09:21.01 Now, from the image, because this is calculated from the image, that... these diffraction
00:09:25.20 spots give you also the phase information.
00:09:28.16 By combining the amplitude and the phase, they can calculate the projection images,
00:09:34.07 actually projection structures of this one in the 2D.
00:09:38.00 Then, combining with tilting the specimen to different angles and, using these central
00:09:42.20 sections, here, because you know how much, precisely, you tilted the specimen, you know
00:09:46.18 where this second image, the tilted image, where they come from.
00:09:50.02 By combining all the tilted images, they were able to demonstrate that it is possible to
00:09:54.09 actually obtain the 3D structure of a 2D crystal.
00:09:59.19 What I'm showing is a... a 3D structural reconstruction of bacteriorhodopsin at around 7 Angstrom
00:10:06.23 resolution, or really the first time visualizing the transmembrane helices of an
00:10:11.08 integral membrane protein.
00:10:13.10 This was done in... back in 1975.
00:10:15.19 Now, around the same time, Joachim Frank was the... proposed the idea that... what happens
00:10:21.18 if we don't have a crystal?
00:10:23.19 Can one image individual molecules, but then computationally, using cross-correlation as
00:10:29.21 a measure, to align those images together with each other, therefore computationally
00:10:34.19 averaging them together to generate a high-resolution picture.
00:10:38.06 He outlined this... this idea in a paper published in 1975.
00:10:44.05 And he further demonstrated this one in a work in 1978, by showing that you can indeed,
00:10:51.23 by averaging individual molecules to get a higher resolution struc... higher... better
00:10:57.14 structural information by averaging individual molecules... image of individual molecules.
00:11:04.18 This was demonstrated by this, a glutamine synthetase that he published in 1978.
00:11:10.05 And then, again, back then he also proposed the idea of tilting the specimen to obtain
00:11:15.04 the third dimension.
00:11:17.13 And this was further demonstrated with an E coli... demonstrated with an E coli ribosome
00:11:22.16 structure by the so-called random conical tilt approach.
00:11:25.17 This was also the first time that... using a single particle averaging approach to obtain
00:11:31.23 a 3D structure of a 3D object.
00:11:35.00 Now, this is a different approach of a... still, using negative stain, here... which...
00:11:41.10 however, to getting the 3D structure of molecules with forming a crystal.
00:11:47.20 Then, combining this single particle EM approach with the frozen hydration, that requires a
00:11:54.10 different technique of preparing a biologic sample.
00:11:57.05 Now, this came from Jacques Dubochet.
00:11:59.05 He invented this... he developed this method back in 19... early 1980s to prepare a biologic
00:12:06.09 sample in the frozen... individual biologic particles in a frozen hydrated way.
00:12:11.24 Now, showing here is a picture of icosahedral virus embedded in a very thin layer of ice.
00:12:18.14 The method he developed back then is still the one that we are using nowadays.
00:12:22.06 Now, nowadays we are using these kind of EM grids, that... which is copper grids, typically,
00:12:28.01 covered by a continuous carbon film.
00:12:30.02 In this carbon film, there are regularly spaced holes.
00:12:32.22 The diameter of these holes are about 2...
00:12:36.06 1 to 2 micrometers.
00:12:38.12 And the way, nowadays, we'll prepare a biologic sample for the frozen hydrated sample is that
00:12:43.23 we apply a droplet of our purified protein sample, typically about 2 micrometer... microliters
00:12:52.03 and... into the... into these EM grids.
00:12:55.16 Then, we use a filter paper to blot away most of the solution.
00:12:59.15 The remaining liquid in the... left in the grids, because of surface tension, will form
00:13:05.01 this continuous liquid layer across the mem... these holes.
00:13:09.07 And then you rapidly plunge freeze this one into the liquid ethane cooled by liquid nitrogen.
00:13:14.01 The temperature of liquid ethane is typically around minus 200 degrees C. And the cooling
00:13:21.07 rate has to be very fast, to about 1000 to 10,000... sorry, about from... about 10,000
00:13:27.17 to 100,000 degrees C per second.
00:13:30.06 In such a high cooling rate, that the specimen... the liquid will be frozen in this amorphous
00:13:36.09 state without time to crystallize.
00:13:39.09 And if this liquid became crystal, the crystal will crush the protein.
00:13:43.11 There, the high cooling rate is really required to preventing such a crystallization.
00:13:50.10 So, the end is what... what we call vitreous ice, that what happens is you have a... it's
00:13:57.09 a liquid forming across the mem... these holes will be frozen in vitreous ice and... with
00:14:04.01 the protein sample embedded in this thin layer of vitreous ice in a free-standing state.
00:14:10.03 And then in... in this... more or less in the native conformation.
00:14:13.14 Then, we take this one into the electron microscope and then take the image of it.
00:14:18.06 And then, if we look at each individual particle's image, this image can be defined by... by
00:14:25.04 three different... by a number of geometric parameters and, particularly, by a... three
00:14:34.06 parameters that defines its spatial orientation, and two other parameters defines the XY positioning,
00:14:42.07 and then the last one defines the Z direction, which is typically referred to as defocus.
00:14:48.22 And, if we can refine all these parameters as accurately as possible, then we were able
00:14:54.05 to get a higher... a higher resolution structure.
00:14:56.16 Now, putting all these things into a simple slide which shows you how this actually works,
00:15:02.21 which I can show you here.
00:15:04.11 Here, this is a typical electron micrograph of a... a biologic sample embedded in
00:15:11.19 a thin layer of vitreous ice.
00:15:13.05 In this case, the sample is an archaeal 20S proteasome, which is a specimen that we, in
00:15:18.09 our lab... that we use often as a tech... as a specimen for our
00:15:22.17 microscopy development work.
00:15:24.15 Now, we take many, many of such images.
00:15:27.00 And, in each image, we either computationally or visually identify individual particles.
00:15:32.14 And, as I circled here... and then we extract them out of this large micrograph and form
00:15:39.08 a stack of all individual particle images.
00:15:42.08 Then, we align them together with each other and then we... after alignment, then we can
00:15:49.02 average them together and then to... generating a 3D reconstruction, just as I outlined in
00:15:55.13 the earlier slides.
00:15:56.20 Now, in this case, this is a 3D reconstruction of an archaeal 20S proteasome to the resolution
00:16:02.09 about 5-6 Angstrom resolution.
00:16:07.03 At this resolution, one can visualize the secondary structural features,
00:16:11.15 including alpha helices, beta sheets, and loops.
00:16:14.12 However, at this resolution, we could not visualize the sidechain density.
00:16:19.05 In other words, that, at this resolution, it's insufficient for us to build an
00:16:22.23 atomic model de novo.
00:16:25.02 In order to do that, one really needs to improve the resolution from this kind of sub-nanometer...
00:16:30.02 Angstrom resolution to near atomic resolution level.
00:16:33.18 And, what is limiting the resolution of single particle EM?
00:16:37.06 This is not because of the electron microscope itself.
00:16:40.19 The electron microscope actually has a very high resolution.
00:16:44.00 Showing here is a single electron micrograph of a graphene.
00:16:48.14 And you can see that, out of this single electron micrograph, that graphene... and the carbon
00:16:53.17 atom is already resolved.
00:16:55.07 So, the typical electron microscope has a resolution down to almost 2 Angstroms or even better,
00:17:00.24 and it can resolve individual electron... individual atoms.
00:17:04.14 Here, the reason that they can get this high resolution is because they use a very high
00:17:09.24 electron dose because these kinds of specimen are not radiation sensitive.
00:17:13.19 So, therefore, you can get this high resolution.
00:17:15.24 But for a biologic sample, we could not take this similar approach using high electron
00:17:20.09 dose to getting the... getting these high resolution images.
00:17:25.07 And this has been a major limitation.
00:17:28.16 And then, related to this one, is actually how we're recording the image.
00:17:32.20 In the biologic sample... in cryo-EM, typically, traditionally, we are using either photographic film
00:17:38.07 or CCD camera.
00:17:39.14 Now, CCD camera gives you instant digital imaging, so that enables one to do this on-the-fly
00:17:46.05 data collection or automated data collection.
00:17:48.08 However, CCD camera has a problem; the CCD camera does not detect the electron --
00:17:54.03 it only detects the photon.
00:17:55.13 In order to use CCD camera to record electron micrograph, one has to put a thin layer what
00:18:01.18 we call a scintillator on top of the CCD camera.
00:18:04.17 Now, this scintillator converts the electron signal from... to photon so the photon can
00:18:10.18 be detected by the CCD camera.
00:18:12.17 This conversion, however, has a problem, which...
00:18:15.17 generally, converted from a point event to a block, because it... one electron generates
00:18:20.05 many photons, became converted from one point event to a block.
00:18:24.22 Now, this can deter it in the performance of the... of the camera.
00:18:29.02 And... and the camera performance can be characterized by using something we call a detectable quantum
00:18:37.00 efficiency -- DQE -- curve to describe it.
00:18:40.17 Showing here, you can see that the red curve corresponds to the DQE
00:18:44.07 of a typical photographic film.
00:18:47.06 And, showing here in the green, is a typical DQE curve of a CCD camera,
00:18:53.04 which we used to use.
00:18:55.17 And you can see that the DQE curve, compared with the film of a...
00:18:58.05 CCD camera compared with film, is significant... significantly poorer.
00:19:01.11 And then, for the CCD camera, at high frequency, we can almost... only able to gather about
00:19:07.03 10% of signal -- about 90% of the signal is lost.
00:19:11.00 This is a known problem and... for many years... and using this type of imaging to record...
00:19:16.24 this digital camera to record image.
00:19:19.07 And this is what we typically gather, here.
00:19:21.15 Showing on the left is, again, an electron micrograph of a frozen hydrated archaeal 20s
00:19:27.12 proteasome, and showing on the right, here, is a Fourier transform of this image.
00:19:32.17 And you can see there's a lot of rings surrounding the central beam and these rings are what
00:19:37.00 we call the Thon.
00:19:38.17 What we want is that we want to see these rings extended to as high resolution as possible.
00:19:44.00 And this is similar as X-ray crystallographers putting a crystal into electron... into an
00:19:49.09 X-ray beam; how far this crystal can diffract gives you the information about resolution
00:19:55.07 you can expect getting from this structure.
00:19:57.12 It's similar, how far these Thon rings go, giving us a certain expectation of what we
00:20:02.14 can get out of this kind of image.
00:20:04.03 And then, from this type of image,
00:20:05.24 we were able to get the structure I just showed you earlier.
00:20:09.09 Now, how to improve that?
00:20:10.23 And then, the field already new this for years, and that requires a new generation of camera
00:20:16.00 that can directly detect the electron signal.
00:20:19.07 This is called the... a direct electron detection camera.
00:20:24.01 Now, this type of camera became commercially available around 2011 and 2012.
00:20:29.03 This type of camera no longer needs a scintillator; it can directly detect the electrons,
00:20:33.12 convert the electric signal to the charge, which can be picked up by the camera.
00:20:39.00 And then the charge distribution generated by each electron event is also very small,
00:20:44.17 considerably small compared with the... this... the point spread function generated by the...
00:20:52.16 by the scintillator film converting from electron signal to the photons.
00:20:57.07 Now, if we... on the left, here, shows, again, the blue curve is the DQE of a camera that...
00:21:04.12 developed by Gatan called the K2 camera.
00:21:07.13 Now, you can see that this camera, now, at the... can be actually... the performance
00:21:13.11 of this camera is actually quite comparable or slightly better than photographic film.
00:21:17.11 This is very good.
00:21:18.14 Now we can have a camera that works as good as the photographic film, but also gives you
00:21:24.12 a digital image right away.
00:21:26.15 But this still not enough, right?
00:21:28.14 So, we wanted something even better.
00:21:31.01 And this comes from what we call a single election counter.
00:21:34.19 The idea is that this new generation of camera has a very high frame rate; for the K2 camera,
00:21:40.11 it can operate at a total 400 frames per second.
00:21:44.02 And in such a high frame rate, in each individual frame, those electron events are very sparse.
00:21:50.04 Now, if each... for each individual frame, if the... for every charge event, every electron
00:21:55.21 event, is still distributing a charge over several pixels.
00:21:59.19 However, if there's no other events nearby, one can identify the individual events, not
00:22:06.09 only identify where they are, but also can center that data to an exact pixel where this happens,
00:22:12.10 or even to a quarter of a pixel.
00:22:14.12 So, in this way, you gain two things.
00:22:16.11 One is that your... your point spread... point spread function becomes a single pixel size.
00:22:22.05 The second is that you remove all the readout noise, so that this becomes more or less like
00:22:27.18 a noiseless camera.
00:22:29.07 And now this made a huge impact to the... to the performance of the camera.
00:22:34.03 You can see the... showing in the red curve is the DQE curve of the same K2 camera, but
00:22:39.23 operating in the single electron counting mode -- the DQE is significantly improved.
00:22:44.24 This is a very big deal for the single particle cryo-EM and for the cryo-EM in general.
00:22:51.09 Showing here, again, is an electron micrograph of a... a frozen hydrated archaeal 20S proteasome,
00:22:58.10 however this imaging, now, is recorded by the K2 camera operating
00:23:02.23 in the election counting mode.
00:23:05.00 Showing on the right side, here, is the Fourier transform.
00:23:07.03 You can see this... all these rings, now, all the way from... goes to... extended to
00:23:11.22 the... about 3 Angstrom level.
00:23:14.13 And this shows that this information in this type of image, now, is retained to almost
00:23:20.20 near-atomic resolution level.
00:23:22.13 And, using this type of image, now, will hopefully enable us to getting the structure to a much
00:23:28.04 higher resolution.
00:23:29.16 But this is also not everything this camera brings us.
00:23:33.17 As I mentioned that, this camera actually can operate in a very high frame rate.
00:23:37.15 Therefore, this type of camera now enables people to record an image as a movie instead
00:23:43.11 of a single steady picture.
00:23:45.22 And the power of recording images as a movie was first demonstrated
00:23:49.19 by Niko Grigorieff's lab.
00:23:51.16 Showing on the right... on the left, here, is a stack of a movie, from just adding together
00:23:58.07 of a picture recorded from ico... icosahedral virus.
00:24:02.16 You can see that all the virus particles show these kind of streaks across them.
00:24:06.24 This is because the particle has a motion when the beam hit the specimen.
00:24:11.21 And this is what is... what we know as a beam-induced image motion.
00:24:16.13 And this has been known for many years and this is a major factor that deters the quality
00:24:21.05 of the image.
00:24:22.05 Now, Niko Grigorieff's lab showed that, using this movie, that one can actually potentially
00:24:27.24 computationally correct for this motion and generating a motion-corrected image, which
00:24:32.18 you can see here on the right is much sharper than that.
00:24:35.24 So, we also further want to extend this approach for more general specimens, instead of just
00:24:41.03 only virus, because virus is very large, it's relatively easy to deal with it.
00:24:46.02 But for the protein that we want to work on, that are often a very small size, and we want
00:24:50.15 to have a different app... different approach,
00:24:52.10 but using the same principle to correct for motion.
00:24:55.24 Now, this is what we're doing, now.
00:24:57.19 So, this... showing here, the first image is what we call the perfect image.
00:25:01.21 The reason we call it a perfect image is because a Fourier transform of this image shows Thon
00:25:07.04 rings to a very high resolution, to close to 3 Angstroms.
00:25:10.01 However, this type of image is very rare and out of a hundred images we can probably get
00:25:16.04 about a handful or less images with such quality.
00:25:20.00 The more typical images are shown here... while either looks almost the same, but if
00:25:24.16 you look at the Fourier transform it shows that there is a severe cutoff.
00:25:28.06 This is because of the image-induced motion, as I described earlier.
00:25:33.13 Now, with a... the way to overcome this problem is to record the image as a movie, also as
00:25:40.04 I showed earlier.
00:25:41.10 But here, the problem is that, in this... each individual frame, the images... we could
00:25:46.05 not really see any signal.
00:25:47.24 The image really has a relatively... a much poorer signal-to-noise ratio.
00:25:50.24 We could not see any features by eye.
00:25:53.19 And therefore we have to develop some algorithms to track the motion between every frame.
00:25:58.24 Now, with this algorithm that developed in our laboratory, that we are able to track
00:26:05.22 the motion between every frame, then now we can
00:26:08.08 track how every frame moves relative to another.
00:26:12.04 With this information, we can computationally move every frame back to its origin and...
00:26:17.02 generating what we call a corrected image.
00:26:19.15 And then, in this corrected image, you can see that... in the Fourier transform, all
00:26:23.21 the information is completely restored to the high resolution.
00:26:26.20 This is very powerful and enables us to getting almost every image, nowadays, we're recording
00:26:34.12 has... able to getting this kind of quality, and for almost any type of sample.
00:26:40.01 And... but, also, down here... that...
00:26:42.10 I also want to mention that, in this algorithm here, I just showed you, here, every... for
00:26:48.04 every frame, we're treating every individual frame as a rigid body.
00:26:52.11 And there's... we did consider any local motion.
00:26:55.13 So, out of... anyway, out of this kind of approach, we're already able to getting the
00:27:00.06 structure of archaeal 20S proteasome from a... what I showed you earlier, about 5-6
00:27:05.15 Angstrom resolution, to 3.3 Angstrom resolution.
00:27:08.03 At this resolution, now, we can start to see sidechain densities.
00:27:13.16 Showing on the left, here, is a... an EM density map determined to... by this single-particle
00:27:21.15 cyro-EM to about 3.3 Angstrom resolution.
00:27:24.15 Showing on the right... right side, here, is the same... the... the density map from
00:27:29.19 an X-ray structure of the same protein at about a similar resolution, 3.3, 3.4 Angstrom resolution.
00:27:36.06 You can see that the density maps are very comparable and that this demonstrated to us
00:27:40.01 that, if we can get any structure to this level of resolution, we will be able to build
00:27:46.00 a de novo atomic model based on sequence.
00:27:48.16 This is very exciting.
00:27:50.08 And... but also, that we showed you earlier that... when we treat the... image motion,
00:27:58.02 we treat the entire micrograph as a one rigid body.
00:28:01.23 But, of course, this is not the case, and, instead, that every, actually, local region,
00:28:07.03 the particle motion has slight differences.
00:28:10.07 And this was actually able to track down by the... tracking individual particles.
00:28:15.20 And this work was demonstrated by Sjors Scheres' lab, that they were able to show that, actually,
00:28:20.15 in different locations, the particle moved differently.
00:28:23.14 By tracking individual particles, they were able to... they will be able to... to actually
00:28:29.17 correct the local motion as well.
00:28:31.12 So, nowadays, that... a lot of labs was int... doing is that they combined the two methods
00:28:35.21 to get... including ourselves... that would first use the global motion... the optics
00:28:40.10 that I described to you earlier, to corr... to correct for the... for the overall global
00:28:46.13 motion, then followed by tracking individual particles to correct for the local motion.
00:28:50.19 Now, these two processes also works very well, but it's still kind of tedious.
00:28:56.05 Can we actually do the things that all these two combined... this two-step of motion correction
00:29:01.08 together?
00:29:02.08 Now, this a question we want to also tackle.
00:29:05.09 And then, this... the clue comes from the early... still, from Niko Grigorieff's work,
00:29:09.01 when he looked the virus particle motion.
00:29:10.24 He found that the virus particle motion... it not only has lateral motion, but also has
00:29:16.08 this rotation.
00:29:17.17 The... the way this particle rotates is suggesting that the ice film has actually a... moving...
00:29:23.19 there is an upwards movement, z-direction motion, of the ice film when the electron
00:29:29.01 beam hits the specimen.
00:29:30.14 Now, he proposed this idea that when the beam hits a sample the specimen can move upwards.
00:29:36.10 Now, based on this idea, that we can think about that... what actually... what we see
00:29:42.18 in the local motion is actually a projection of this 3D motion projecting into the image.
00:29:48.19 When particularly this is more obvious is when the specimen was sighted tilted.
00:29:53.14 And now, Shawn Zheng from David Agard's lab demonstrated, further demonstrated this one,
00:29:58.14 by showing that looking at the specimen, at a highly tilted specimen, you can see that...
00:30:03.19 that almost all the motion is a per... perpendicular to the tilt axis, really supporting this idea
00:30:10.14 of this z motion.
00:30:12.09 Now, based on this idea, that we think that we can actually model the deformation of the
00:30:20.16 ice and then use this model to actually be describing the motion of every point.
00:30:26.06 So, nowadays, that... what we are doing now is to developed a new program called MotionCor2.
00:30:31.04 In this program, the way we are dealing with motion is that we first cor... correct for
00:30:35.16 the global motion, then we divide this entire image into different patches, and we can also
00:30:42.17 determine the average motion of each patch.
00:30:44.23 Then, we can use this same patch to fit to... into this... polynomial function, which is
00:30:50.14 a function we used to describing the shape of this deformation of the ice.
00:30:56.03 And by... by modeling this area... this... the shape of the ice deformation, we were
00:31:02.20 now able to correct the motion of every pixel in the image.
00:31:06.15 Now, this... basically, this approach combined both global motion and the local motion together
00:31:12.15 to do all the correction, and in one step, before we actually doing any further imaging
00:31:17.18 process.
00:31:19.09 Now, back to the same dataset.
00:31:22.07 We, early, on, were able to get a 3.3 Angstrom resolution.
00:31:26.09 Now, using the same dataset, but using this, a motion correction combining both the global
00:31:32.04 and the local motion, we were further able to improve the resolution of the same dataset
00:31:37.06 to almost 2 and a half Angstrom resolution.
00:31:39.01 This is a significantly improvement, and it's coming from the... combining this global and
00:31:44.22 local motion, mostly.
00:31:46.10 But, also, umm... umm... with... using the lat... later,
00:31:52.13 more advanced imaging process programs.
00:31:55.03 Now, with all these things together that I showed you today is the major technology breakthrough
00:32:01.08 in single particle cryo-EM, particularly facilitated by the camera technology, and then, particularly,
00:32:08.04 with the motion correction related to this new tech... camera technology.
00:32:12.02 But, of course, in the last few years, besides of this breakthrough, the other major breakthrough
00:32:18.15 was brought on by the software, particularly, how do we classify images into different groups?
00:32:24.08 How do we generate a more homogeneous subset of particles
00:32:28.16 from a more heterogeneous particle dataset?
00:32:32.00 And then, this also greatly facilitating us to get into high resolution structure.
00:32:36.15 And, by putting all these things together, what the cryo-EM field experienced in the
00:32:41.05 last few years was really a resolution revolution, as mentioned by Werner Kuhlbrandt in the paper
00:32:47.24 outlined... by his paper published, actually, in 2014.
00:32:52.07 Now, this was further also emphasized by another feature article in Nature that's showing that...
00:33:00.12 that... what's the impact of single particle cryo-EM to the entire structural biology community.
00:33:06.15 Now, to putting all this together, we can now say that single particle cryo-EM really
00:33:11.19 has become the method of choice when we started many challenging
00:33:15.24 biological macromolecule structures.
00:33:18.21 And then, so, this now concludes my talk today, for this part.
00:33:23.04 And... and finally I will acknowledge people involving the work.
00:33:27.19 These are people... a list, here, are people from my... current members of my lab and also
00:33:32.07 former members.
00:33:33.07 Also, I want to mention specifically that all our methodology development works are
00:33:37.12 in collaboration David Agard's lab.
00:33:38.21 The MotionCor2 program was developed by Shawn Zheng.
00:33:43.08 And also that our work was supported by a number of funding agencies, listed here.
00:33:46.23 Finally, thank you for your attention.

Part 2: Single Particle Cyro-EM of Membrane Proteins

00:00:07.17 Hi.
00:00:08.17 My name is Yifan Cheng.
00:00:09.19 I'm a professor of Biochemistry and Biophysics at the University of California, San Francisco.
00:00:15.24 I'm also an Investigator of the Howard Hughes Medical Institute.
00:00:20.17 This is the second part of my talk today, and here I'm going to be mostly focused on
00:00:27.07 the work related to my own laboratory, particularly our eff... our efforts toward using
00:00:33.14 single particle cyro-EM to study integral membrane proteins.
00:00:38.09 And for a number of years that... our effort is trying to bring the single particle EM
00:00:45.01 of membrane proteins from a blob-ology level to the atomic resolution, in a way that we
00:00:52.04 try to bring something that looks like this, which is showing blogs of a molecule, to something
00:00:58.12 like this one, which has a secondary structural feature, and to... finally to something that
00:01:04.04 looks like this, which has atomic details.
00:01:07.09 So, this is our overall goal, our mission of the lab, and in this talk I'm going to
00:01:15.07 take you through our... one of our recent works, which is illustrating the transition
00:01:21.11 from a blob-ology to atomic resolution.
00:01:24.21 So, a major target of our laboratory's work is the ion channel called TRPV1.
00:01:31.06 And this is a sensor for the capsaicin and noxious heat.
00:01:36.03 And, in this work, we have a collaboration with David Julius' lab,
00:01:39.21 back since around 2009.
00:01:42.19 David Julius' lab was the one that initially identified, indeed discovered
00:01:47.10 this TRPV1 ion channel.
00:01:50.03 And the TRPV1 ion channel is an excitatory ion channel in the pain pathway.
00:01:55.02 It can be activated by a natural compound from capsaicin, which is in the... it's an
00:02:00.13 ingredient which makes your mouth feel hot when you eat chili peppers.
00:02:05.09 And this channel can also be activated by noxious heat, by the temperature.
00:02:10.04 When temperature rises above 43 degrees C, the channel will be activated,
00:02:14.00 and below this threshold the channel will remain inactive.
00:02:17.01 So, in a way, that... this channel is an indicator of different thermal and chemical signals.
00:02:22.16 And now, this... in my doctorate, we also used this channel as a model system for our
00:02:27.21 methodology development work.
00:02:30.22 Now, using single... structural biology of membrane proteins is a particularly challenging
00:02:37.03 area in structural biology.
00:02:39.01 And, over the last 10-15 years, tremendous progress has been made by X-ray crystallography
00:02:44.17 of... studying integral membrane proteins.
00:02:47.14 Particularly, the progress was made in the membrane protein production using overexpression
00:02:53.19 systems, and development of new detergents, and new meth... methods for...
00:02:58.11 for crystallizing membrane proteins.
00:03:01.13 Now, it's... cyro-EM, as I mentioned also earlier, has also been used by... to study
00:03:08.08 membrane proteins but, clearly, in the early time that electron crystallography has...
00:03:13.22 able to determine membrane protein structures to atomic resolution level.
00:03:18.00 And, in this case, that... which...
00:03:20.03 membrane protein was crystallized into a two-dimensional crystal and, using... still using a crystallogra...
00:03:26.10 crystallographic approach.
00:03:27.10 Now, the... however, the bottleneck of any crystallographic approach is that you need
00:03:32.13 to first form a very well-ordered or well-diffracting either 3D or 2D crystal, and of sufficient size
00:03:40.12 so that you can use a crystallographic approach.
00:03:42.21 Now, this is a major bottleneck of any crystallographic approach, that if you don't have a crystal
00:03:48.07 you don't have anything.
00:03:49.10 So, therefore, our goal is trying to use single particle cyro-EM to study membrane proteins.
00:03:55.12 However, this is a... it's a promising approach, which was demonstrated very early on, in the
00:04:02.17 1990s by Joachim Frank's lab and Irina Serysheva that they look at the... using the
00:04:08.08 single particle EM approach and... to study membrane proteins.
00:04:13.04 And particularly, in this case, the ryanodine receptor to... also, at the low resolution level,
00:04:19.06 but showing the correct structural shape, how this molecule looks like.
00:04:24.20 And then, around like 2009, 2010, the... the best single particle cyro-EM structure of
00:04:32.06 membrane proteins are these two structures -- it's the proton-driven ATP synthase and also the
00:04:39.08 IP3 receptors, both around the nanometer resolution level.
00:04:42.20 And it's in the resolution, just about resolving the secondary structural features about...
00:04:48.05 particularly, the transmembrane helices.
00:04:50.03 Now, back specifically to the TRPV1 ion channel, around this time, the resolution was still
00:04:55.17 really limited to the blob-ology level at a much poorer resolution.
00:05:00.06 This is not because of the... this is mostly because of, really, the technology of
00:05:07.04 single particle cyro-EM, particularly to study membrane proteins, was very challenging, was very limited,
00:05:11.22 so therefore the resolution of a number of these structures show the relatively low resolution.
00:05:16.13 Now, around the same time, the crystallography approach of the TRP channel can only resolve
00:05:21.24 the... determine the structure of a specific domain called ankyrin repeats domain.
00:05:26.05 This particular domain has been solved by the crystallography and the structure.
00:05:30.13 Now, our goal is trying, using single particle cyro-EM, and without forming crystals, determine
00:05:36.11 a high resolution structure of the TRPV1 ion channel to atomic resolution level, so that
00:05:41.14 we can understand the mechanism of how this channel is gated and atomic details.
00:05:46.13 Now, this all started by making the protein, and this work was done actually... carried
00:05:53.01 out very... for a number of years by Erhu Cao, a very talented postdoc from David Julius' lab.
00:05:58.21 Now, he expressed the... he made this construct that... which is a so-called minimum contract,
00:06:03.08 in which he truncated a little bit of C-terminal and N-terminal domains which were predicted
00:06:07.12 to be flexible.
00:06:08.12 So, that... this is mostly for better expression and also, initially, this construct was designed
00:06:15.09 for crystallization.
00:06:16.09 So, the idea is to remove all these, say, flexible domains so that it can facilitate
00:06:21.03 and be better for crystallization.
00:06:22.05 Now, this channel can be expressed in... in... in the HEK cell and using
00:06:28.14 a mammalian expression system.
00:06:31.01 And then the... once the protein was produced and purified, Erhu was able to reconstitute
00:06:36.00 this protein into the... back into liposomes and then use a patch clamp to measure the
00:06:41.19 activity of this protein, and showing that it... actually, the purified protein has the
00:06:46.16 same character as the wild-type protein.
00:06:51.04 So, that's a... already a tremendous progress for the making of the protein.
00:06:55.14 This is already, actually, quite a number of years’ efforts into this...
00:06:58.23 making this protein.
00:07:00.11 Then, the next step is putting this protein into the electron microscope to look at it.
00:07:04.04 So, the TRPV1 ion channel was first expressed in the mammalian expression system, and then
00:07:10.15 sol... solubilized and purified in the presence of detergent.
00:07:15.20 Showing here is a size-exclusion chromatography of a detergent-solubilized and purified TRPV1.
00:07:24.10 It's a very nice single peak, suggesting that the protein is a... is in a... mono-dispersed
00:07:32.07 and in... in a good shape.
00:07:34.21 Then, the next thing we did is that we put this membrane... membrane protein into...
00:07:39.06 to look at... visualize them in the negative stain EM.
00:07:42.18 Showing here, that indeed the proteins are monodisperse.
00:07:46.00 However, the 2D class averages of these proteins... these particles, showing that they are actually,
00:07:51.11 still... the conformation is actually quite heterogeneous.
00:07:55.19 And this actually posed us a tremendous challenge for starting to... to try to determine structure
00:08:00.23 of this protein to high resolution structure.
00:08:04.00 So... and another challenging imposed by this one is the presence of detergent, making it
00:08:10.01 very hard to... for us to make a thin layer of ice and, because the presence of detergent
00:08:15.15 changes the surface property, the surface tension, making it much harder to make a thin
00:08:20.12 layer of ice and with the protein embedded in it.
00:08:23.11 And, instead of that, we tried something else -- that is using amphipols.
00:08:28.00 Amphipol is also a... can of course... amphipathic molecules, different from detergent.
00:08:35.01 This one was first used, actually, by Werner Kuhlbrandt's lab, and then... to study a large
00:08:40.03 membrane protein.
00:08:41.14 And then, in this way, the idea that you can... you can remove the detergent and an amphipol,
00:08:46.13 in a way, is similar to detergent, but different from detergent.
00:08:50.16 You can actually put in the substitute detergent with amphipols and then run this protein through
00:08:57.02 the size-exclusion chromatography using a buffer without a... completely without detergent
00:09:01.16 or amphipols.
00:09:02.16 So, therefore it's very stable, actually, for... for membrane protein.
00:09:07.13 So, we tried this one with the TRPV1 and then you can see that, after substituting detergent
00:09:15.07 with amphipols, the protein also remained as a monomer, dispersed as a single peak.
00:09:20.22 And, then, we visualized this by negative stain EM.
00:09:23.22 It turned out that this one, now, in the... in the presence of amphipols, the protein
00:09:29.01 remains not only monodispersed, but also much more in a better homogeneous conformation,
00:09:34.00 shown here in the 2D class averages of this one.
00:09:37.18 This facilitated us to have a much more homogeneous sample to further study them in.... using
00:09:43.12 cryo-EM.
00:09:44.14 Now, showing here, on the left, is a typical electron micrograph of frozen hydrated TRPV1
00:09:51.13 in amphipol, embedded in a thin layer of ice.
00:09:54.15 Now, this image is recorded using a scintillator-based CCD... -based camera.
00:10:00.03 On the right, what shows here is a Fourier transform of this image.
00:10:03.00 It shows that the Thon rings go almost to 8 Angstroms.
00:10:05.21 This was very exciting when we first got this image, because this type of data enabled us
00:10:11.05 to getting a structure of TRPV1 at about 8 Angstrom resolution.
00:10:15.04 At this resolution, we were finally able to resolve the transmembrane helices and...
00:10:21.01 with a... showing in this movie, here, is the 3D density map of TRPV1.
00:10:29.00 You can see that we can nicely visualize the transmembrane helices and these helices can
00:10:34.13 be interpreted very similar as a vot... any... these tetrameric ion channels.
00:10:43.11 Now, this is... was a tremendous... a uhh... excitement for us, that this was the first
00:10:48.23 time we started visualizing the membrane protein in the... in the... its transmembrane helices
00:10:54.23 by single particle cryo-EM.
00:10:56.19 But this was also around the same time that we also worked out the technology of using
00:11:03.04 a direct electron detection camera to getting structure of a
00:11:08.22 macromolecule complex to high-resolution structure.
00:11:11.19 Therefore, around this time, we immediately... instead of trying to publish this structure,
00:11:15.12 we immediately trans... started using the new technology to study the same thing, and
00:11:20.18 then tried to aim to get a high-resolution structure.
00:11:22.15 Now, here, on the... on the left is an electron micrograph of, again, this TRPV1 embedded
00:11:30.22 in a layer of ice.
00:11:33.06 And... but this image, now, is recorded with this K2 camera in counting mode and after
00:11:38.17 motion correction.
00:11:40.02 Showing right here is a Fourier transform of this image.
00:11:42.14 You can see the Thon rings, all the way...
00:11:44.18 goes to 3 Angstrom, suggesting this image really has a quality enable... that could
00:11:48.23 enable us to getting an atomic resolution structure.
00:11:51.12 Now, from this type of data, we were finally getting structure of TRPV1 at about 3.3 or
00:11:58.03 3.4 Angstrom resolution.
00:11:59.17 You can see here, on this density map, that is showing that... a really nicely resolved
00:12:03.12 sidechain.
00:12:04.17 Based on this density map, we were able to build an atomic model de novo based on the
00:12:10.10 sequence of the structure.
00:12:12.15 And... just illustrating how the camera technology enabled us from a low... a poor...
00:12:21.00 relatively poor resolution level to a high resolution level.
00:12:23.15 Showing here on the left, now, is a 2D class averages of TRPV1 images...
00:12:28.22 TRPV1 ion channel recorded by using that scintillator-based camera.
00:12:33.20 You can see that we're resolv... resolving the... the shape of the molecule, you can
00:12:37.19 see the transmembrane domain, and also the soluble domain.
00:12:41.14 However, if we have the same sample, now recorded with a ser... with a K2 camera after motion correction,
00:12:47.21 now you can see there is a great more detail of these images.
00:12:51.09 Now, this is the one that enabled us to getting this structure.
00:12:54.05 As I showed in another talk earlier on, that we also now have a new version of motion correction,
00:13:02.00 called MotionCor2, which we can use to further improve the resolution and correct the...
00:13:06.01 both global and the local motion.
00:13:08.11 And then apply this... the new algorithm to this same data set, that we can see that we
00:13:13.00 can get a sample to a significant, even further improved, resolution.
00:13:16.18 Actually, now, we're actually able to improve this same data set, the resolution, to close
00:13:21.11 to 3 Angstrom.
00:13:23.14 Now, from this structure... this work was actually done by most... two very talented
00:13:29.01 postdocs in our laboratory, Erhu Cao from the Julius Lab and Maofu Liao from my laboratory.
00:13:35.20 And both are faculty members now and starting their own independent labs.
00:13:39.19 Now, back to this structure.
00:13:42.07 That... because we don't need to form a crystal, this enables us also to capture the structure
00:13:47.02 in different conformations.
00:13:49.05 Because for single particle cryo-EM, we do not need to form crystals, therefore this
00:13:53.17 enables us to using a ligand to... if we can trap the protein in different conformations
00:14:00.01 using ligand, this will enable us to actually determine structures of the same protein in
00:14:05.07 different conformations.
00:14:06.19 In this case, here, showing here... that we actually have the TRPV1 ion channel determined
00:14:12.14 in three different conformations.
00:14:14.13 Showing on the left is the structure of TRPV1 apo structure, without ligand, showing that
00:14:20.13 the channel is in a closed conformation.
00:14:22.23 So, important here is the single channel recording corr... corresponding to this... uhh... in
00:14:28.23 this conformation, in the apo state, showing the channel is in a completely closed conformation.
00:14:33.06 We can see that, in the ion permeation pathway, the channel has two major restrictions -- and
00:14:38.00 one is up, one is in the bottom -- but both are in a closed conformation, showing the
00:14:42.16 channel is in... in the... channel is in a closed conformation.
00:14:46.01 Now, in the middle, here, is the structure of TRPV1 in complex with capsaicin, which
00:14:52.11 is the ingredient that... in the chili pepper that makes your mouth feel hot.
00:14:55.23 Now, showing in the... you can see that the... the ion permeation pathway... showing that
00:15:01.17 the lower part of the channel is in the open conformation and... corresponding to the single
00:15:08.05 channel recording, shows that the channel is actually, in the presence of capsaicin,
00:15:13.09 is a... flipping back and forth between open and closed conformations.
00:15:16.20 This suggesting that... suggests that the opening of the lower gate by the capsaicin
00:15:24.09 could allosterically destabilize the upper gate, and then... and making the upper gate
00:15:28.23 also transition back and forth between the open and closed conformations.
00:15:34.14 On the far right, here, is the structure of TRPV1 in the person of a capsaicin-like molecule
00:15:40.17 bound in the bottom, in which you can see that the lower gate is in an open conformation,
00:15:45.01 but, also, in the outer pore, there's a double knot toxin binding to the channel, which also
00:15:51.03 opens the channel in... opens the channel.
00:15:54.23 You can see that, now, the channel, and the... both outer gate and the lower gate are both
00:15:58.09 in the open conformation -- the channel is in a complete... a completely open state.
00:16:02.16 Now, single channel recording shows that the channel indeed is in a completely open conformation.
00:16:08.03 This is suggesting that different ligands aiding... acting on the channel have different
00:16:13.11 effects on the either out... the outer pore or in the lower pore.
00:16:18.12 And that's suggestion that a channel can res... integrate different signals... input from
00:16:22.06 different ligand binding.
00:16:24.12 But, for us, technically, we want to also move this one to a next level.
00:16:29.11 And then, so... one thing we want to see is the... how can we put the channel into a...
00:16:34.23 study the channel in a lipid bilayer environment?
00:16:36.22 This is actually meaningful.
00:16:38.17 If we look back to the structure of TRPV1, the density map of TRPV1 in the pres... in
00:16:43.23 complex with the double knot toxin and a... and a vanilloid compound binding, then this
00:16:52.01 structure also gives us the conformation of the whole structure.
00:16:55.18 However, the resolution or the quality of the density map of this channel, actually
00:17:00.12 is unsatisfying for... in this one.
00:17:03.07 If we compare this channel with the apo structure, we can immediately see that the channel actually
00:17:09.19 did... in the... in the complex with... when the channel was in complex with double knot toxin,
00:17:14.08 the resolution is relatively poor, compared with the apo structure.
00:17:18.06 Now, this is the same protein and the structure was determined using the same approach, so
00:17:24.23 the resolution, the poor resolution, is not because of... the sample differs... it's not
00:17:30.13 because of the sample difference, or it's not because of the difference in the technology,
00:17:36.17 and it must have come from other reasons, why this resolution is poor.
00:17:40.05 So, one of the ideas, we're thinking, is that because in this case the channel... or...
00:17:44.21 is in the... as I mentioned earlier, is first purified in detergent and then substituted
00:17:50.18 with amphipol.
00:17:51.23 In this case, the... the lipid molecules are all removed during the sample preparation.
00:17:57.19 Now, it is from early work... it is known that the double knot toxic binding to the
00:18:02.13 channel, it first requires the presence of lipids.
00:18:06.00 So, remove the lipids and then that may be the major reason of stable...
00:18:11.24 destabilizing this complex, and then it leads to a relatively poor resolution.
00:18:16.14 Now, the goal, therefore, is how to put this membrane protein back in the lipid bilayer environment,
00:18:21.17 and we can still study the high resolution structure of them in the lipid environment.
00:18:28.14 Now, traditionally, there are a number of different approaches, again, that this can
00:18:31.19 be studied by electron crystallography, because membrane protein was reconstituted into the
00:18:36.16 2D crystals in the presence of lipids.
00:18:39.16 Therefore, in the... in that way, that one can actually put a membrane protein into...
00:18:44.09 back into a lipid environment, getting a high resolution structure.
00:18:47.01 However, this approach again requires the formation of a crystal, and if we don't have
00:18:51.18 a crystal, you know, you're not able to get a structure.
00:18:54.10 So, if we want to use a single particle EM approach, now, we could not take this one.
00:18:58.17 But, using a single particle EM approach, there are... there's a method to... to put
00:19:02.20 a membrane protein in the lipid environment, bilayer environment... bilayer environment,
00:19:06.23 is using the so-called lipid nanodisc.
00:19:09.13 Now, their early works demonstrated that... that this approach is actually feasible to
00:19:13.16 use in the lipid nanodiscs.
00:19:15.00 Of... in both cases, the resolution they were able to get in was in...
00:19:19.02 a relatively limited resolution.
00:19:20.22 Now, then... using the lipid nanodiscs can also be, actually, a... not using it... but
00:19:27.16 another approach is putting membrane proteins into liposomes and then trying to look at
00:19:31.21 the structure of membrane proteins in the liposome.
00:19:33.20 Now, this method was first, actually... was carried out... demonstrated by Helen Saibil's lab
00:19:40.15 and followed by Fred Sigworth's lab, and they're now using this approach to studying
00:19:45.20 membrane proteins in the... in the lipid environment.
00:19:48.09 The resolution of this approach is still, right now, limiting in the... relatively around,
00:19:54.03 like, a nanometer or sub-nanometer resolution, not able to get atomic resolution yet.
00:19:59.10 Now, the... one other approach is still using liposomes and then... approach... reconstituting
00:20:06.05 membrane proteins into liposomes but taking electron tomography using a sub-volume averaging
00:20:12.09 technique to try to obtain the resolution, try to obtain the 3D structure of a membrane
00:20:18.21 protein in a lipid environment.
00:20:20.16 The resolution of this approach currently is also limited around, like,
00:20:24.16 a nanometer resolution.
00:20:26.04 Now, for our laboratory, we thought that we want to try the approach of using lipid nanodiscs.
00:20:31.12 And, then, this work was carried out by a very talented joint graduate student, Yuan Gao,
00:20:36.09 from my laboratory and the Julius lab.
00:20:37.24 Now, Yuan was finally able to, after a lot of... a lot of hard work, he was finally able
00:20:42.07 to reconstitute TRPV1 into lipid nanodiscs.
00:20:47.11 Showing here is a size exclusion chromatography; showing in the middle, here, is the peak corresponding
00:20:51.22 to the TRPV1 reconstituted into the lipid nanodiscs.
00:20:55.08 Now, if we examine this sample by negative stain EM, one can see that they can, indeed,
00:21:01.05 be reconstituted into nanodiscs.
00:21:03.02 Yuan was able to, using two different type of scaffolding proteins, which formed nanodiscs
00:21:08.02 of two different diameters... on the top, here, is the 2D class averages of a membrane
00:21:15.15 protein in a relatively larger... a membrane scaffolding protein... or, in other words...
00:21:20.21 can be reconstituted in a relatively larger nanodisc.
00:21:23.24 In the bottom, here, is a slightly smaller nanodisc, and you can see from these two images
00:21:29.24 that the bottom one shows that the spec... the specimen... the actual particles are much
00:21:34.19 more... more homogeneous in a... in a way.
00:21:38.05 And, because there is a... not a bigger nanodisc floating around it.
00:21:44.09 And the common wisdom will be saying that this bottom one, using these smaller nanodiscs,
00:21:49.09 will probably be the better one to enable us to getting high resolution structures.
00:21:53.01 However, our concern is that it... when you have a smaller nanodisc, it may impose a....
00:21:58.21 it might impose certain constraints to the transmembrane domain, to preventing it to
00:22:05.00 undergo conformation change.
00:22:06.16 So, for that reason, we still choose to use the larger nanodisc, so that, in this case,
00:22:12.00 the membrane proteins are really floating around inside a lar... relatively large nanodisc,
00:22:17.15 without being constrained by the membrane scaffolding protein.
00:22:21.15 Now, showing here, again, this is a typical electron micrograph, on the left, of TRPV1
00:22:27.12 in the nanodisc and recorded by... again, by the K2 camera in the... operating in the
00:22:32.20 counting mode.
00:22:33.20 On the right side, here, is a Fourier transform showing these, say, Thon rings always go to
00:22:37.22 3 Angstroms, indicating the quality of the images.
00:22:41.07 And the bottom, here, is the 2D class averages of these images, showing high quality of the...
00:22:46.19 of the... this data.
00:22:49.02 And from this data, we're... we're almost able to getting a structure of TRPV1 in nanodiscs,
00:22:55.13 together with the double knot toxin and vanilloid compound binding.
00:23:00.23 Showing here, you can see the large disc is a... is from the nanodisc.
00:23:06.13 The membrane protein is floating inside the disc, here, and his blue density should mark
00:23:11.23 the boundary of the membrane.
00:23:13.08 This is from... is from the headgroup of the lipid bilayer lipid molecules, that mark the
00:23:18.15 boundary of the... of the bilayer.
00:23:20.24 And it's the blue density, here, it shows here, are actually specific bound lipids...
00:23:25.24 lipids that are specifically bound to the membrane protein, that they are very well
00:23:30.16 resolved -- we can model them.
00:23:32.00 You can see a number of lipids in the outer leaflet and also in the inner leaflet.
00:23:36.17 The resolution of this uhh... reconstruction is significantly improved.
00:23:40.21 There are some densities, for example, this particular one, shown here in the mesh, was...
00:23:45.12 we actually observed this same density in the structure determined from amphipol...
00:23:49.11 amphipol structure early on.
00:23:52.03 The resolution there, however, was relatively poor; we could not model this density.
00:23:56.13 Now, with this improved density, we are now able to model this one as a phosphatidylcholine.
00:24:02.12 Now, more interestingly, that... whether the question is that... whether nanodiscs actually
00:24:07.16 can improve the resolution of a... a...
00:24:11.24 a single particle reconstruction of a membrane protein.
00:24:15.04 Now, this is a... not exactly the case.
00:24:17.17 Now, we... here, we actually have the three different structures of TRPV1, both in nanodisc
00:24:24.07 and also in amphipol.
00:24:26.02 On the left, here, is the apo structure, and in the middle, here,
00:24:30.15 is a TRPV1 with agonist bound.
00:24:33.21 On the right, here, is the TRPV1 with antagonist bound.
00:24:37.13 You can see that, in the middle one is the one that really had the resolution improved
00:24:42.10 quite significantly, from about 3.8 Angstrom to about 3 Angstrom.
00:24:47.02 On the left, here, in the apo structure, the two structures show you... the apo... nanodisc
00:24:52.09 or in amphipol gives you almost... gives us almost identical, normal resolution, showing
00:24:57.05 that the structure actually, when we compared the two, they look almost identical.
00:25:00.22 However, there's still a little bit of subtle difference.
00:25:04.10 And then... which shows that the structure... the nanodisc, actually, is a good thing for
00:25:09.19 the membrane protein.
00:25:10.20 Here, if we take out the most outer alpha helices and then, by comparing the density
00:25:16.13 from the nanodisc amphipol, here... from nanodisc and amphipol, here... and we can see that,
00:25:23.00 in one side of these helices, by looking at the sidechain density facing inside of the
00:25:29.19 protein, that one can see that this den... this sidechain density looks almost identical,
00:25:34.24 even the side chains can be all positioned in an almost identical way.
00:25:39.13 And this is suggesting that the structure... or, there's no change.
00:25:42.12 However, if we look at the sidechain facing the lipid part, we can see that, in the nanodisc case,
00:25:48.13 that these sidechain densities are significantly better resolved than the ones in the amphipol,
00:25:54.01 suggesting that, even for the membrane proteins that are facing the lipid part... that when
00:25:59.00 the protein is in the lipid nanodisc, that this is still providing a better environment
00:26:05.13 for the... for... for structure determination.
00:26:10.10 So, this is actually suggesting to us that the lipid nanodisc is probably a generally
00:26:15.14 good platform to study membrane proteins.
00:26:19.13 And not necessarily that it will give you a different conformation, but it probably
00:26:25.12 most likely will give you a better environment for membrane protein to sit there and then
00:26:30.06 for the... the better structural studies.
00:26:33.08 Now, in here, we also visualized some very interesting features, for example, we can
00:26:37.18 now see that this double knot toxin binding to the channel is really stabilized by the
00:26:45.16 presence of this lipid.
00:26:47.01 You can see that this lipid has its headgroup interact... forming an interaction with the...
00:26:53.19 with the channel protein.
00:26:56.04 And its hydrophobic tail forms a hydrophobic interaction
00:26:59.23 with the hydrophobic part of the toxin.
00:27:03.04 Now, this sort of a tripartite complex is probably required to really stabilize that;
00:27:09.02 it's critical for this whole complex to be in a stable conformation.
00:27:13.17 Now, it's probably...
00:27:14.18 most likely is that this kind of stabilization, and facilitated by the presence of lipid,
00:27:20.18 is the major reason that we were able to getting a better resolution of this particular conformation.
00:27:26.14 Now, we also see that a very interesting thing is that, for example, you can see on the left
00:27:31.11 is a unliganded structure showing two lipids molecule.
00:27:35.02 Its headgroup is more or less in a similar level, height level.
00:27:38.13 And when the double knot toxin is binding, here, it partitions into the bilayer, it will
00:27:43.02 actually move the lipids up and down to accommodate the binding of the double knot toxin.
00:27:47.11 This is also quite reasonable to... to understand.
00:27:51.11 Now, the further thing that... which is quite interesting, is that we now have a much better resolution,
00:27:56.22 compared with previously, in this vanilloid binding pocket.
00:28:00.24 In the apo structure, we see a very clearly defined density map, which can be modeled
00:28:05.22 very nicely by the phosphatidyl... phosphatidylinositide molecules.
00:28:13.03 And this can be... and we actually see this same density in the amphipol structure, but
00:28:18.07 not so clearly resolved.
00:28:20.01 So, suggesting that this molecule, this lipid molecule, is actually a... in... mostly likely
00:28:26.10 is an endogenous one, but not added during the purification or nanodisc reconstitution.
00:28:34.11 Now, in the same pocket in the agonist-bound structure, with a different toxin, which is
00:28:41.00 a very potent agonist, TRPV1 agonist, then we can actually see... also improve the density.
00:28:46.12 We can now model this one, agonist, how it binds there and how it's inducing a very subtle
00:28:53.04 conformation change to the channel.
00:28:56.02 And umm... so, understanding... how this channel...
00:28:58.09 how this agonist is actually acting on the channel.
00:29:02.03 Now, together, if we're putting all these things together, we can actually have
00:29:05.00 a much better understanding of how vanilloid is acting on the channel.
00:29:11.09 Showing here, you can... on the left side... on the left side, what you can see is that
00:29:15.05 the density of the vanilloid actually overlaps with the density of the... of the... this
00:29:23.02 lipid molecule, suggesting that the binding of agonist displaced this lipid molecule.
00:29:30.05 And such binding of agonist is promoting a salt bridge formation, which moved the S6 helices
00:29:35.24 backwards and which, in turn, also moves... sorry, moved S5 helices backwards,
00:29:42.09 which coupled to the S6 helices also moving backwards, and leads the channel to be open.
00:29:47.15 And this is how the agonist actions on the... on the channel.
00:29:51.24 Now, in the presence for... of antagonist, and this is the... the structure of a TRPV1
00:29:58.00 with capsazepine, which is a very potent antagonist, we can see that this antagonist molecule is
00:30:03.11 also binding to the same vanilloid binding pocket, it also displaced this lipid molecule,
00:30:09.00 which originally occupied in this pocket.
00:30:10.24 And... however, this binding does not promote or induce any conformation change in the channel.
00:30:17.24 And then... suggesting that this is... the antagonist binding is probably stabilizing
00:30:23.00 the channel in, actually, a closed conformation, then providing the mechanism of how the antagonist
00:30:31.04 is working.
00:30:32.04 So, this actually gave us an idea about how a channel is further... more detailed mechanistic
00:30:40.08 information about how agonists and antagonists are acting on the channel.
00:30:44.17 Now, back to the general question about how membrane proteins... can we study membrane
00:30:50.04 proteins in the lipid environment?
00:30:51.18 As we demonstrated here, the membrane proteins can be put into lipid nanodiscs to study the
00:30:58.09 structure of them, and facilitating the structure determination of a membrane protein
00:31:02.19 in a lipid environment.
00:31:04.07 Now, in... this is our structure, showing here, of TRPV1 in a lipid nanodisc.
00:31:09.19 On the left, here, is a... is a ryanodine receptor, which has a very large soluble domain,
00:31:14.09 also embedded in the lipid nanodisc.
00:31:16.15 In both cases, one can see... think that the structure, the image refinement was really facilitated,
00:31:25.23 probably, by this large or relatively large soluble domain, because in the transmembrane domain
00:31:33.11 we have a large chunk of unstructured mass that was added to the protein, which
00:31:38.07 would, on the other hand, interference... or interference with image alignment.
00:31:45.18 And, therefore, it seems like that... to getting the structure of... high resolution structure
00:31:51.15 of a membrane protein in lipid nanodiscs does require a certain size of soluble domain.
00:31:58.11 And this imposes a question about whether... how do we study a membrane protein which does
00:32:04.22 not have any soluble domain?
00:32:06.22 And can we still, using the lipid nanodisc approach, to study such membrane proteins?
00:32:12.00 And then, so, there are a number of different ways that we can think about it, but for...
00:32:16.00 in our laboratory, for a number of years, we have been taking this approach to artificially
00:32:21.20 increase the soluble domain by using an Fab, monoclonal Fab, particularly using a conformational-specific
00:32:30.02 monoclonal Fab that are... to bind to the integral membrane protein.
00:32:34.16 We can use this one as a fiducial marker to facilitate image alignment.
00:32:39.09 Showing here is an example of such a case.
00:32:42.02 Now, this is a membrane protein of an ABC transporter.
00:32:47.14 It's a bacterial ABC transporter.
00:32:49.20 It's a heterodimer and that... the two monomers are very similar.
00:32:53.13 So, therefore, this protein actually is a... has a pseudo two-fold symmetry... without
00:32:58.16 an actual precise two-fold symmetry, but a pseudo two-fold symmetry.
00:33:02.00 And, in this case, it does have some soluble domain, but because the two are very similar
00:33:08.00 domains that may interfere with the image alignment, we could not distinguish
00:33:12.13 which one is which at low resolution.
00:33:14.01 So, that's what's the concern.
00:33:16.01 So, in this case, we generated a monoclonal Fab by using a phage display library approach
00:33:21.24 that... obtaining of a membrane protein... generating this specific...
00:33:26.11 conformation-specific monoclonal Fab, which binds to one specific monomer.
00:33:30.19 And, using this Fab as a fiducial marker, we were able to getting the structure of this one,
00:33:36.09 a membrane protein, to a sub-nanometer resolution.
00:33:39.00 I have to say that this work was done in the earlier time, when the K2 camera just became available
00:33:44.03 -- this is one of the reasons that this resolution was still in the sub-nanometer
00:33:48.09 resolution, but not higher.
00:33:49.24 But, nevertheless, this actually demonstrated an approach, that using monoclonal Fabs as
00:33:55.07 an approach to study integral membrane proteins, and particularly to study membrane proteins
00:34:00.20 without a big soluble domain.
00:34:02.19 Now, to finally make a conclusion, here, that we're now able to demonstrate that actually...
00:34:08.10 also showed that the single particle cryo-EM... using single particle cryo-EM to study membrane
00:34:13.17 proteins actually has become a versatile or very powerful technique, particularly for
00:34:18.15 membrane proteins, for ion channels, for a large group of membrane proteins that are
00:34:22.11 particularly difficult to... for crystallization.
00:34:26.08 And then, for those membrane proteins, nowadays that... it is probably a more efficient way,
00:34:31.15 now, to study them, to get... obtain the high resolution structure,
00:34:34.15 by using single particle cryo-EM.
00:34:36.17 We also demonstrated that it is possible, now, to put a membrane protein into a
00:34:40.09 lipid bilayer environment using a lipid nanodisc approach to getting high resolution structure.
00:34:46.08 And then, also, specifically to visualizing the membrane protein-membrane lipid interactions,
00:34:51.08 which would... may be better preserved in the presence of the lipid environment, here.
00:34:56.09 And then, in the last, we also showed that, using a Fab approach, that we were able to...
00:35:02.02 most likely, to study membrane proteins that... with a very small or even no soluble domain.
00:35:07.16 And we hope that using Fab, it could be also another general approach that are able to
00:35:12.08 enable us to study such membrane proteins.
00:35:15.02 And, finally, I want to conclude this one, to acknowledge people who contributed work.
00:35:20.04 These are shown here.
00:35:21.07 Listed here are current lab members and former lab members.
00:35:24.01 And, in the TRPV1 work, we have a very tight and close collaboration with
00:35:28.08 the David Julius lab.
00:35:30.05 And these are the people from Julius lab which also contributed to our... to this work I
00:35:34.16 presented today.
00:35:36.14 And I've also noted that we have also a collaboration with Charlie Craik's lab and Bob Stroud's
00:35:41.01 lab at UCSF to develop this technique of using Fab to facilitate the
00:35:48.00 single particle cryo-EM study of a small membrane proteins.
00:35:51.01 Finally, these are the funding agencies which support our work.
00:35:54.02 Finally, I want to thank you for your attention.

Videos in this Talk
  • Part 1: Single Particle Cryo-EM
    Part 1: Single Particle Cryo-EM
    Audience:
    • Student
    • Researcher
    • Educators of H. School / Intro Undergrad
    • Educators of Adv. Undergrad / Grad
  • Part 2: Single Particle Cyro-EM of Membrane Proteins
    Part 2: Single Particle Cyro-EM of Membrane Proteins
    Audience:
    • Researcher
    • Educators of Adv. Undergrad / Grad
Speaker: Yifan Cheng
Total Duration: 1:10:21
Recorded: May 2017
All Talks in Biophysics

Talk Overview

Cryo-Electron Microscopy (Cryo-EM) is a form of transmission electron microscopy in which a beam of electrons is transmitted through a frozen-hydrated sample, and is used to study the molecular structure of biological samples. In this lecture, Dr. Yifan Cheng overviews the principles of Cryo-EM, and describes how advances in this technique have allowed scientists to solve biological structures to atomic resolution.

In his second lecture, Cheng shows how his laboratory has used Cryo-EM to study the atomic resolution of membrane proteins. It is challenging to use conventional methods to study membrane protein structure, given that the 3D structure of most membrane proteins is dependent on their interaction with the phospholipid bilayer. Cheng describes how his laboratory has overcome these challenges to successfully solve the protein structure of the TRPV1 ion channel in different conformations at atomic resolution. He describes the benefits of using of amphipols, lipid nanodiscs, and Fab-assisted approaches to facilitate structural studies.

Speaker Bio

Yifan Cheng

Yifan Cheng

Dr. Yifan Cheng is a professor of biochemistry and biophysics at the University of California, San Francisco, and a Howard Hughes Medical Institute Investigator. Cheng obtained his bachelor’s degree in Physics at Wuhan University in 1982, and completed his doctoral degree in Physics at the Chinese Academy of Sciences in 1991.  In 2006, he joined… Continue Reading

Playlist: Imaging in Biology

Related Resources

Cheng, Y, et. a. (2015). A Primer to Single-Particle Cryo-Electron Microscopy. Cell 161(3), 438–449

Bai, XC, McMullan, G, & Scheres, SHW (2015) How cryo-EM is revolutionizing structural biology. Trends in Biochemical Sciences. 40(1), 49–57

Reader Interactions

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *