The Jensen laboratory uses state-of-the-art electron cryomicroscopy techniques to understand the structure and function of large protein machines and their arrangement within cells. Projects range from theoretical studies on the mathematics of three-dimensional reconstructions to direct imaging of individual protein complexes to tomography of viruses and cells.
The "big picture" is to eventually combine these results with the vast data emerging from genomics, proteomics, and structural genomics to enable authentic, whole-cell simulations that allow the rational design of entirely new species for the production of clean fuels, water desalination, bioremediation, medical applications, etc. Electron cryomicroscopy fills a critical gap in what we call the "structural biology continuum".
Modern electron microscopes can deliver information with atomic resolution. The movie shows images of a gold cluster where the stacked planes of gold atoms are easily resolved. We want to apply this kind of resolving power to biological specimens such as proteins and cells, but face at least three fundamental challenges:
preserving native structures within the high vacuum of the EM,
extracting three-dimensional information from the projections recorded by EMs, and
recording enough image data before the sample is destroyed by the high energy electron beam.
The first challenge can be accomplished, at least for thin samples, by quick-freezing the specimen by plunging into liquid ethane and then keeping them frozen throughout the imaging process. This is the meaning of the prefix "cryo" in "cryo-electron microscopy." Quick-freezing causes the water to form vitreous ice around the proteins, preserving their native structure but solidifying the sample so it can withstand the vacuum of the EM. The second challenge can be achieved by recording projection images of a single object or a set of identical objects from a large number of directions and computationally merging the data into a three-dimensional reconstruction. One algorithm to do this is called a "back-projection." The final challenge can be overcome to varying degrees depending on the nature of the sample: when the sample is a single, unique object such as a cell, the tolerable dose is distributed through a series of images at different tilt angles-this is called "tomography," and the ultimate resolution appears to be limited to ~2nm. When the sample is not unique, the best strategy is to purify a large number of identical copies and image each one with a small dose and then average those images. This strategy is called "single particle analysis," and because there is no limit to the number of images that can be included and averaged, theoretically atomic resolutions are possible. Below is an image of whole bacterial cells frozen within vitreous ice.
Four frozen-hydrated Mesoplasma florum cells, suspended in vitreous ice around a lacey carbon support
"Tomography" refers to an imaging strategy in which projections are recorded of an object from various directions and then a higher dimensional reconstruction is calculated from those projections. In "electron" tomography, a series of two-dimensional projections are recorded in an electron microscope (EM) while incrementally tilting a sample along one or two axes. A three-dimensional reconstruction is then calculated. This allows the structure of individual macromolecules, organelles, and even whole cells to be imaged in 3D.
Tomography schematic: in tomography, a tilt series of projection images are recorded of a unique object, then the three-dimensional structure of the object is reconstructed by "back-projection." From Baumeister et al., Trends in Cell Biology 9:81.
EM samples must be thin to allow transmittance of the electron beam and must be solid so that they can maintain their integrity within the high vacuum required in the microscope column. These requirements lead to two basic paths in sample preparation depending on the size of the specimen: if the specimen is less than ~.5 microns, it can be frozen in vitreous ice and examined directly without artifact. If the specimen is larger, it must be sectioned and imaged piece by piece. Generally sectioning requires that the sample be chemically fixed and embedded in a plastic resin, processes which can introduce artifacts. Sections are then usually stained with heavy atom salts to enhance contrast. Nevertheless, this makes it possible to image even very large cells by "serial sections."
What we seek by electron tomography is to understand the ultrastructure of cells at a "molecular" resolution, or in other words understand where each protein is localized and how chains of sequentially-acting protein machines are organized within the cell. To achieve this we computationally search the three-dimensional reconstructions of cells to identify macromolecular complexes and cytoskeletal scaffolds of interest. Of course our resolution will be insufficient to locate small proteins and molecules. We are working to develop specific electron-dense markers to tag small objects, but in many cases it is not necessary: in regions of the cell where small enzymes and molecules are freely diffusing, it doesn't matter where they are-we can already model diffusion. It is only the limits to diffusion and those elements of the cell that are spatially localized that must be found experimentally, and those are precisely the large, continuous cellular components which electron tomography is most likely to reveal.
Below are two movies: one of a tilt series recorded of a frozen-hydrated Caulobacter crescentus cell, and the other of the three-dimensional reconstruction that resulted.
In single particle analysis, a large number of identical, purified copies of the target macromolecule are frozen in solution across EM grids in random orientations. In this case instead of recording a series of tilted images of the same object (traditional tomography), the entire tolerable dose is delivered in a single projection. Thousands of identical molecules are imaged individually, each providing a unique projection direction, and then all the images are combined computationally to produce a three-dimensional reconstruction. Because the effective cumulative dose scales with the number of molecules included in the average, which is theoretically unlimited, atomic resolution reconstructions should be possible. If the sample exhibits conformational flexibility, individual particles can be sorted into classes representing the different conformations and the structure of each defined state can be determined independently. Above are images of individual 20S proteasomes, negatively stained on the left and frozen within vitreous ice on the right, with an atomic model between.
To better understand and test the limits of cryo-electron microscopy, as well as our image processing algorithms, we use sophisticated methods to simulate EM images. Currently the best approach is called a "multi-slice" algorithm, which treats the sample as a series of slices to track the cumulative phase shifts experienced by the incident electron beam as it traverses the sample. At the beginning the electron beam is represented by a plane wave. The sample is divided into an arbitrary number of "slices," and the projected electron potential of each slice is calculated from atomic scattering factors. The phase shifts each slice causes in the electron wave are calculated sequentially, with additional phase shifts introduced between slices to represent the propagation of the wave through space to the next slice. Finally the "exit" wave function is multiplied in Fourier space with the contrast transfer function, and a real image is simulated by squaring wave amplitudes. An image of the 20S proteasome, simulated using this algorithm, is shown below.
Images of 20S proteasome. Actual images at one defocus value are on the left, simulated images with varying defocus using the multi-slice algorithm are on the right.
The prospect of understanding simple cells in complete molecular detail is truly exciting. Coupled with the possibility of rewriting genomes and designing entirely new species, the future of microbiology is very bright indeed. After over a century of escalating research in the field, however, crowned today by the availability of nearly 500 complete microbial genome sequences, our persistent ignorance about many of the most fundamental processes in a bacterial cell cycle is sobering. We still don’t know, for instance, how bacteria generate and maintain their characteristic shapes. While we know shape determination involves cytoskeletal filaments and the patterned construction of a peptidoglycan shell, we are only beginning to understand molecular architectures and mechanisms, and we are still far from being able to swap “shape” modules of genes between species or modify sizes at will. Nor do we fully understand how prokaryotes establish or maintain polarity, how they organize their genomes, or how they segregate their chromosomes during cell division.
Scientific progress on questions like these often comes through the careful testing and revision of specific hypotheses, but it can also jump forward less methodically when major technological advances open entirely new windows into nature. Electron cryotomography is just such a technology, as it has begun to provide wholly new insight into the structures and functions of the (1) cytoskeleton, (2) cell wall, (3) genome, (4) motility machines, (5) intracellular compartments, (6) chemoreceptor arrays, and various other supramolecular assemblies.
The prokaryotic cytoskeleton
While the complete absence of any cytoskeleton was long thought to be one of the distinguishing characteristics of prokaryotes (Kürner et al., 2004), bacterial homologs of actin, tubulin, and intermediate filament proteins have now been identified (Michie and Lowe, 2006). Fluorescent light microscopy (fLM) has further shown that many proteins localize in filament-like patterns in-vivo. Although decades of traditional thin-section EM studies largely failed to reveal any such filaments (Bermudes et al., 1994), using ECT, we have now directly visualized many different cytoskeletal filaments and filament bundles in a variety of species. In collaboration with Dianne Newman (Caltech) we visualized, identified, and determined the function of MamK, a novel actin homolog responsible for organizing the magnetosomes of a magnetotactic bacterium (Komeili et al., 2006). In order to address the bacterial cytoskeleton more generally, we then imaged the classic model cell Caulobacter crescentus, which is the only prokaryote known to contain representatives of all three major classes of cytoskeletal filaments. We found five distinct classes of filament bundles with characteristic ultrastructures and locations (Briegel et al., 2006), and named them "inner curvature," "cytoplasmic," "polar," "ring-like," and "paired" bundles (the paired class was not described in the publication). Simply seeing these filaments for the first time raised a number of important questions. First, while some of the filaments appeared in expected locations (Ausmees et al., 2003), others were unanticipated and we did not see anything like the helical patterns suggested for many proteins by fLM images (Gitai et al., 2004). Second, none of the bundles were common, and their appearances were not correlated with the cell cycle. Third, most of the filaments were short, and did not obviously bridge meaningful destinations.
In order to identify filaments and gain insight into their functions and mechanisms, we are now pursuing a number of strategies. First, we are imaging expression mutants (over-expressions, deletions, and depletions) of suspected cytoskeletal genes. In the case of MamK, this worked beautifully because in a mamK deletion mutant, the filaments we had previously seen flanking the magnetosome chain were missing and the chain was no longer linear. When the mutant cells were complemented with mamK on a plasmid, however, the filaments and chain organization were restored, allowing us to conclude that the filaments we saw were in fact MamK and that one of their roles was to linearize the chain. We are now collaborating with a number of groups to image various expresssion mutants of many candidate cytoskeletal genes in a variety of species.
Second, we are attempting to identify cytoskeletal filaments by their inherent structural "signature" (periodicity, diameter, bundling patterns, twist, and perhaps eventually monomer shape). One of the filaments seen in C. crescentus, for instance, had a clear 8-nm periodicity, but this unfortunately did not match any candidate crystal structures of cytoskeletal filaments. We are now developing software to computationally find all the filaments present in our cells and quantify their structural properties.
Third, we are working to develop genetically encoded heavy-atom labels to identify specific proteins in cryotomograms.
Most motile bacteria propel themselves with flagella. We published the first structure of an entire flagellar motor, which showed among other things the number and shape of the stator "studs" in-situ(Murphy et al., 2006). We have now reconstructed motors from several different species, and the comparison is highlighting how each species has adapted the motor to its purposes. The T. primitia motor is, for example, wider in diameter and has more stator studs, adaptations that could "gear" it down to produce the higher torque it might need as a spirochete to rotate the entire cell. Non-flagellar ("gliding") motility mechanisms have also been described, including retraction of surface-attached pili, secretion of polysaccharides, or "treadmilling" movements of motors along surface protein tracks(McBride, 2001). We imaged the attachment organelle of Mycoplasma pneumoniae and proposed that it was a conformationally dynamic engine driving motility in that species(Henderson and Jensen, 2006). We are now extending these projects to look at a variety of motility machines and their mutants in different organisms.
Other internal compartments
While some enzymes diffuse freely, others are tethered or packaged into "super-enzyme" structures to streamline certain metabolic pathways. We recently solved the quaternary structure of E. coli's super-enzyme pyruvate hydrogenase by ECT(Murphy and Jensen, 2005). More recently we have been analyzing the carboxysome, an organelle-like polyhedral body that facilitates carbon fixation by sequestering ribulose 1,5-bisphosphate carboxylase/oxygenase (RuBisCO)(Cannon et al., 2001). Our cryotomograms have revealed the number, size, shape, and positions of carboxysomes inside H. neapolitanus as well as their surprising association with storage granules (see figures). The number, positions, and orientations of RuBisCOs inside purified carboxysomes are also visible(Iancu et al., 2007). Based on this structural data, we are now building spatially-explicit computational models of carboxysomes that will track individual RuBisCOs as well as their substrates and products to explore what metabolic advantages the carboxysome structure might confer.
In addition to imaging small cells, we have also made a major investment into the structural biology of HIV-1. In collaboration with Wes Sundquist's group at the Univ. of Utah, in January 2005 we presented the first three-dimensional reconstructions of the human immunodeficiency virus type 1 (HIV). HIV is unusual in that while each virus has the same basic membrane and protein layers, they are all unique, making standard methods such as X-ray crystallography or EM-based "single particle analysis" ineffective. Thus while there are literally hundreds of structures of individual HIV protein domains in the PDB, many questions remain about how these assemble into a virus and then mature. Our paper reported structures of mature viruses, and showed that the cone angle of the capsid shell does indeed match the predictions of the current "fullerene cone" model. We also found that some viruses had multiple and sometimes even nested capsid shells, arguing strongly against models of maturation that involve gradual collapse and reshaping of the spherical immature shell.
Next we imaged immature HIV-1. This confirmed previous conclusions from projections that the Gag lattice is organized into concentric spherical shells with hexagonal packing, but imaging them in 3-D allowed us to see that the shells are incomplete: the Gag lattice was actually a patchwork of loosely connected hexagonal nets separated by regions of disorder or vacancy. Importantly, this meant that previous estimates of the total number of Gag molecules per virion were high by about a factor of 2. Furthermore, particle formation did not depend on inclusion of 12 or any specific number of pentameric Gag rings, and in fact the budding process probably bequeathed the virus with a significant amount of "extra" membrane. By computationally dissecting the different shells of the virus, we further confirmed that the subunit contacts that mediated the hexagonal packing were in the capsid and SP1 layers, as only they were hexagonally ordered. We then extracted individual unit cells from just the highly ordered patches and produced an average structure of a Gag hexamer, a process that could be described as "crystallization in-silico." The structure suggested that the SP1 spacer formed a lattice-stabilizing six-helix bundle, which explains why proteolysis between CA and SP1 during maturation triggers rearrangement, and why certain drugs that bind SP1 block maturation.
Our future efforts will be directed to more detailed structures of HIV-1 in various states and various macromolecular machines involved in its life cycle.
Electron cryomicroscopy is a (relatively) young field, and there are many opportunities for improvements. We need better samples, better images, and better image processing algorithms to extract as much information as possible from them.
It would be very helpful, for instance, to have a clonable, electron dense label that could be used to identify specific macromolecules or domains within reconstructions. Some of our recent and future efforts have been/will be devoted to this.
The principal resolution limitation in electron cryomicroscopy of frozen-hydrated biological samples is radiation damage. It has long been hoped that cooling such samples to just a few kelvins with liquid helium would slow this damage and allow statistically better-defined images to be recorded. Indeed the Polara was engineered specifically to allow liquid helium cooling, and was the first microscope model that allowed a single sample to be imaged while cooled with either liquid nitrogen or liquid helium, and have these cryogens exchanged repeatedly, without removing the sample from the column. Taking advantage of this unique opportunity, we performed comprehensive comparisons of the two cryogens. Disappointingly, liquid helium cooling actually proved disadvantageous, at least for the doses and resolutions of interest in cryotomography. Intrigued by this surprising result, we went further to explore the behavior of vitreous ice at the two temperatures. We confirmed and extended a then little-known result in the literature that vitreous ice has at least two distinct phases. In fact the plunge-frozen buffers and culture media of interest to cryoEM collapse into a higher density phase when irradiated at liquid helium temperatures. While the extent of radiation damage is presumably the same at either temperature, in the high density ice it manifests itself in a more troublesome way: bubbles gradually form that first negate the contrast of biological materials and then grossly distort the sample.
To get better images, better microscopes are needed. We motivated the construction of the prototype "flip-flop" cryorotation stage for routine dual-axis tomography and have now explored its use extensively. Collecting a second, orthogonal tilt-series can in fact improve the point-spread-function in cryotomograms.
Taking every opportunity to motivate and acquire better instrumentation, we have just purchased and installed one of the three prototype "UltraCAM" lens-coupled CCDs which are expected to have a significantly better point-spread-function and sensitivity than traditional fiber-optic-coupled cameras. In the coming months we will be working with the manufacturer (Gatan) to characterize this camera and test a new scintillator material.
Better image processing
Even with the best equipment and protocols, some tilt series are much better than others. We are therefore also striving to increase the number of tilt-series we can collect and process by implementing an automatic and integrated data acquisition, storage, and processing pipeline. In collaboration with David Agard at UCSF, we have just ported a "predictive" automatic tilt-series acquisition scheme(Zheng et al., 2004) into the Leginon microscope automation package(Suloway et al., 2005). This software now allows us to record a quick "atlas" overview of an entire EM grid, choose and prioritize the best targets present, and then automatically record tilt-series one after another through the day or night. We are currently optimizing and automating the calculation of reconstructions, CTF-correction and various segmentation processes.