Confessions of an icosahedral virus crystallographer

Abstract

This is a personal history of my structural studies of icosahedral viruses that evolved from crystallographic studies, to hybrid methods with electron cryo-microscopy and image reconstruction (cryoEM) and then developed further by incorporating a variety of physical methods to augment the high resolution crystallographic studies. It is not meant to be comprehensive, even for my own work, but hopefully provides some perspective on the growth of our understanding of these remarkable biologic assemblies. The goal is to provide a historical perspective for those new to the field and to emphasize the limitations of any one method, even those that provide atomic resolution information about viruses.

Early icosahedral virus crystallography

Icosahedral virus crystallography began, in earnest, in the late 1960s when it was possible to record and measure diffraction patterns of tomato bushy stunt virus (TBSV) with precession photography [1,2]. Achieving these results and other characterizations of TBSV constituted a significant portion of Stephen Harrison's Ph.D. thesis that was done at Harvard University with Donald Caspar as his major professor. These precession patterns, that extended to ∼15 Å resolution, were analyzed and initial phases determined from small angle X-ray (SAX) scattering patterns that were recorded from TBSV solutions. This provided a quantitative description of the spherically averaged TBSV electron density distribution and significantly extended our understanding of virus structure. Incremental progress was made by Harrison and colleagues [3–5], but it would take a decade of effort and advances in computation, X-ray sources, data collection methodology and phasing methods, until Harrison and colleagues could produce a near atomic resolution structure of TBSV in 1978 [6]. Their structure revealed, for the first time, the viral jellyroll protein fold [7] and a mechanism of molecular switching that allows the formation of pentamer and hexamer subunit distributions on quasi-equivalent surface lattices [8].

My start in icosahedral virus crystallography

In September of 1972 (slightly more than 40 years ago), I joined the laboratory of Michael Rossmann at Purdue University as a postdoctoral fellow to do virus crystallography. I had just obtained a Ph.D. in inorganic and physical chemistry at Iowa State University and all that I knew about the viruses of interest to Rossmann was that they had icosahedral symmetry and obeyed the same rotational point group symmetry operators as the borane B₁₂H₁₂⁻², a molecule I studied in a course on molecular orbital theory. Southern bean mosaic virus (SBMV) was chosen for study after also considering cowpea chlorotic mottle virus (CCMV). Mary Ann Wagner, Dr Ira Smiley and Rossmann had produced the first crystals of both of the viruses prior to my arrival. Mary Ann taught me how to grow and infect black eye pea plants and purify virus from the leaves. I enjoyed this change from a chemistry laboratory and was thrilled when we produced a new crystal form of SBMV. This crystal form was an accident in that we were draining pellets of virus, when I noticed glistening in the pellet (a typical preparation produced up to a gram of virus from a kilogram of leaves and in the final purification step, it was spread over 12 large translucent pellets in ultra centrifuge tubes). When I put this under a low-power microscope, I could see crystals growing in the pellet, some as large as 0.5 mm. After a variety of experiments to do this in a more controlled manner, Wagner, Smiley and I found that we could grow crystals by liquid diffusion through dialysis tubing that sealed a volume in Plexiglas ‘buttons’ that were made in our machine shop. The virus solution was initially at 100 mg/ml in the button, and the concentration was increased by liquid diffusion. Crystals in excess of 1 mm were routinely grown with this method. In parallel with the virus work, I spent significant effort trying to get an Elliot rotating anode X-ray generator operational. Everything converged in late 1973, when we had an intense rotating anode X-ray source and crystals that diffracted to 2.8 Å. The crystals, which we designated type II from their order of discovery in the Rossmann laboratory, were rhombohedral with lattice constants a = 318 Å, α = 64° [9]. We proudly showed these diffraction patterns at an East Coast Crystallography Meeting in the spring of 1974, attended by Professor Don Caspar who was well known for his co-development, with Aaron Klug, of the quasi-equivalence description of virus capsid structure [8]. He thought that the crystal quality was excellent, but that the collimated, 200 μm X-ray beam was poor. He invited me to Brandeis University in the summer of 1974 to learn how to use the double-focusing mirrors developed by Steve Harrison for the study of TBSV [1]. Dr Walter Phillips, a senior scientist in the structural biology group at Brandeis, taught me how to set up the mirrors and provided sage advice on alignment and focusing, as well as the machine drawings, so we could construct the mirror benders at Purdue. We recorded many SAX patterns of SBMV solutions, for practice, that Caspar thought were wonderful. I also spent many hours of discussion with Caspar during that week and learned an extraordinary amount about structural virology.

Toshio Akimoto joined the laboratory in late 1973 and collected a complete 11 Å dataset of the type II crystals with precession photography (without the benefit of mirrors) during the next year. We processed the films with an Optronics rotating drum scanner that was interfaced to the Control Data 6400 computer housed in the Purdue central computer center about three blocks from the laboratory. This machine provided a large fraction of the campus-wide computing. We could only access the scanner between midnight and 6 A.M. because using it significantly reduced the performance for the rest of the campus. Toshio collected the data; I scanned the data and processed it with a program written by Geoffrey Ford in the Rossmann group. At the same time, I was also developing the computer code for refining phases with the 10-fold non-crystallographic symmetry present in the rhombohedral crystals. We all got plenty of exercise because we had to carry the punched cards (or transport them in a shopping cart donated by a local grocery store) three blocks to the computer center and then picked up the cards and printed output at a later time, which depended on the size of the job. Rossmann and Ford had shown that averaging with the 222 non-crystallographic symmetry and Fourier transforming the averaged map in glyceraldehyde 3-phosphate dehydrogenase (GPD) dramatically improved the multiple isomorphous replacement phases and the resulting electron density map [10]. The GPD structure [11], together with that of lactate dehydrogenase [12], also defined the, now famous, nucleotide-binding ‘Rossmann Fold’ [13]; work that was going on in parallel with the virus structure work.

We assumed that applying 10-fold symmetry would dramatically improve the starting phases as proposed by Rossmann and Blow in the famous rotation function paper [14]. It was impossible to use the approach of Ford for averaging the 5-fold symmetry because he took advantage of the orthogonal nature of the 222 symmetry in GPD and placed the crystallographic density in 222 related ‘skew planes’ by interpolation and that allowed straight forward averaging of equivalent grid points over the non-crystallographic point group.

Employing solution scattering analysis of the particles, we had accurate spherical parameters for the hydrated SBMV to 22 Å resolution. We computed phases based on this sphere, attaching the spherical (centrosymmetric) phases computed for each structure factor. The resulting map was, of course, dominated by the spherical phases and had a center of symmetry. I was initially baffled as to the approach for averaging virus density that was related by icosahedral symmetry because only a very small fraction of the electron density could be stored in core. One evening in 1973, working at home, it occurred to me how to do this with the so-called double sorting method [15]. This was a method developed independently by Gerard Bricogne [16] and also used for phase refinement of TBSV. Briefly, given the particle center and orientation (which are fixed by the R32 space group and the single particle in the unit cell), 10-fold-related positions were generated in orthogonal coordinates for each crystallographic grid position, and the same tag number was placed with each set of 10 coordinates. These were then converted to crystallographic coordinates and sorted on the Z crystallographic coordinate. Planes of electron density with constant Z were stored as sections. The sections of the map were read in two at a time (the most we could store in core), and the electron density at each coordinate position was evaluated, interpolating in three dimensions among the closest grid points. The evaluated densities were then sorted on the tag, gathering the 10-fold related positions into a group. These were averaged, and the averaged density was placed at each coordinate position in the crystallographic coordinates through another sort on Z. This electron density map was Fourier transformed with the resulting phases ‘symmetrized’ by the averaging process. These phases were then associated with the equivalent observed structure amplitudes, and a new electron density map was calculated. The process was cycled until the phases converged. An R-factor, computed between the F_calc from the symmetrized map and F_obs, provided a metric for convergence as well. The process was successfully tested with data at 22 Å resolution, generating our first electron density map of SBMV in 1976 [17]. The map was interesting. First, it was centrosymmetric because icosahedral averaging would not break the center, but we could see for the first time that SBMV had a structure reminiscent of CCMV determined by negative-stained electron micrographs. SBMV had clear hexamer and pentamer capsomers in the crystallographic electron density of the T = 3 shell, but in negative stain, they always appeared smooth because helices occupied the quasi 3-fold axes and prevented stain penetration and the definition seen in CCMV.

Ivan Rayment and Dietrich Suck joined the laboratory in that time frame and were able to get a derivative of the type II crystals. Toshio collected another dataset to 11 Å resolution with precession photography, and I was able to get the averaging to work at higher resolution. Together, we produced a map at 11 Å, the limit of the precession data [15].

Meanwhile, the oscillation method of photography was developing, and we obtained our first Nonius oscillation camera in late 1974. We had also constructed our own set of double-focusing mirror benders based on the Brandeis design, and this dramatically improved the quality of the diffraction patterns, allowing order-to-order resolution of diffraction maxima to 2.8 Å resolution. Rossmann developed a superb oscillation photograph-processing program that was used to process data from virus crystals in the laboratory for the next decade [18]. Rayment was primarily responsible for collecting the oscillation data and making and solving additional derivatives, and we had a map at 5 Å resolution by the end of 1978 [19], when Harrison and collaborators published their near-atomic resolution map of TBSV [6]. I was still responsible for molecular replacement phase refinement with the icosahedral symmetry and kept expanding the program to the limits of the available mainframe computers at Purdue. Cele-Abad Zapatero, Tomitake Tsukihara and Andrew Leslie joined the project for the final push to 2.8 Å and our near atomic model. We were delayed significantly because of a mistake in the input lattice constants to the averaging program. Tsukihara discovered this problem, and with the right lattice constants, the phases converged well and the electron density was fully interpretable. Leslie and Rossmann traced the chain through electron density stacked in Plexiglas sheets, and they immediately realized that the fold of the shell-forming domain was the same jellyroll determined by Harrison and colleagues for TBSV 2 years earlier. Abad-Zapatero and Rossmann built metal physical models of all three subunits in the icosahedral asymmetric unit in Richard's Boxes.

In this day of anticipated common protein folds, it is difficult to appreciate the surprise that we experienced when these folds turned out to be so similar. Indeed, even the molecular switching that gave rise to quasi-equivalence was virtually identical (Fig. 1). The only significant difference in the subunit tertiary structures was the additional domain in the TBSV subunit that was added to the C-terminal end of the protein. The SBMV structure was published in the journal Nature on July 4, 1980 [20] and was the culmination of my postdoctoral work with Rossmann.

Fig. 1.

Comparison of the subunit tertiary structure (ramped in color from blue at the N-terminus to red at the C-terminus) and T = 3 particle quaternary structure of SBMV (top) and TBSV. Although there was no detectable sequence homology, the structures of the shell-forming domain were virtually superimposable. TBSV has an additional domain at the C-terminus of the subunit that adds the surface protrusions to the particle. The quaternary structures of the contiguous shell domains in SBMV and TBSV are also nearly identical. The maximum dimension of the SBMV particle is approximately 300 Å.

An independent faculty position

During the latter stages of the SBMV work, I began to study the structures of cowpea mosaic virus (CPMV, in collaboration with George Bruening at the University of California, Davis) and black beetle virus (BBV, in collaboration with Roland Rueckert at the University of Wisconsin). These were now my own projects pursued as an assistant professor at Purdue; an appointment that started January 1, 1978. While my effort was focused on CPMV and BBV, Rossmann and Rueckert began working on the structure of the human common cold virus.

The world of virus crystallography changed in 1981 as synchrotron X-ray radiation became available. My first successful experience with a synchrotron was in Orsay, France at LURE. Roger Fourme hosted me, and the results were fantastic, with intensity 10× that of a rotating anode and a square spot shape that was 100 μm. I collected a portion of a dataset from CPMV, working from 8 A.M. to 2 A.M. (the French were civilized and did not run 24 h/day) for 5 days. The unit cell parameters of the P6₁22 unit cell were a = 450; c = 1038 Å [21], and the order-to-order resolution of the C axis was extraordinary [22].

The second big event for virus crystallography at Purdue occurred in 1983, when the University purchased a Cyber 205 computer. I rewrote the molecular replacement programs to make use of the vector processors, and the phase refinement process went from taking 6 weeks to do a single cycle of icosahedral phase refinement at 2.8 Å resolution to being able to do six cycles in a day. The change was breathtaking and, together with the rate of data collection at synchrotrons, allowed us to tackle many virus structures, making it possible for graduate students to complete a high resolution structure as a thesis project.

I continued to contribute to the molecular replacement phase refinement for Rhinovirus structure determination and was a co-author on that landmark paper in 1985 [23]. The first high resolution virus structure from my laboratory was BBV, determined primarily by M.V. Hosur and in collaboration with Rueckert [24]. Our structure of beanpod mottle virus (a relative of CPMV, but in a more tractable space group and unit cell), determined primarily by Zhongguo Chen and Cynthia Stauffacher, was published in Science in 1989 and provided the first example of partially ordered RNA in a virus [25]. The computing power was now increasing rapidly, and the VAX Mini computers and individual workstations were appearing in the mid-to-late 1980s that competed well with mainframe computers in terms of speed, memory and disk space. We were collecting data at CHESS at Cornell in a wonderful facility established by Keith Moffat called MacCHESS. The intensity of the X-ray beam at CHESS was more than 100× that of a rotating anode. Over the next two decades, our group solved the structures of 16 additional, unique mammalian, plant, yeast, bacterial and insect viruses. Our data collection moved to the Advanced Photon Source at Argonne National Laboratory in 1996, where Keith Moffat set up another world-class facility called BioCARS on sector 14. These were halcyon days for icosahedral virus crystallography.

Crystallography and electron cryomicrosopy (cryoEM): a wonderful marriage

Tim Baker came to Purdue in 1983 intending to study membrane proteins with electron microscopy following the structure determination of bacteriorhodopsin [26]. It was not meant to be! He had worked with Ivan Rayment and Don Caspar at Brandeis University on the structure of polyomavirus and together they had shown that all the capsomers of the virus were pentamers even when occupying hexavalent sites in the T = 7 quasi-equivalent lattice [27]. An early project for Baker at Purdue was the cryoEM reconstruction of SV40 in collaboration with Minou Bina, also at Purdue [28]. From that point onward, much of Baker's work has involved virus structure and function with cryoEM.

My first exposure to cryoEM was with Baker and to the structure determinations of Nudaurelia capensis β virus (NβV) and Nudaurelia capensis ω virus (NωV), viruses that we had crystallized [29]. We needed a phasing model for these virus crystals, and we felt that a 20 Å resolution map would be adequate. Rossmann and Hogle had both shown that not only could you refine phases with non-crystallographic symmetry, but you could also extend phases to higher resolution [23,30]. Getting an atomic model of NβV and NωV would take awhile, but I was smitten with cryoEM. Baker and I worked on a cryoEM reconstruction of a monoclonal antibody bound to CPMV (Fig. 2). This was a glorious success, and the paper was published in Nature in 1992 [31]. We were able to use cryoEM with CCMV to build a pseudo-atomic model of the swollen form of the virus, after we determined an atomic model of the compact form of the virus [32]. These successes excited me about cryoEM and what it could do for particles that could not be crystallized. Indeed, when moderate resolution was adequate, structures could be quickly determined by cryoEM and interpreted in terms of atomic models of related forms of the particles.

Fig. 2.

Surface-shaded representations of the 23 Å cryoEM reconstructions of native CPMV and CPMV with a monoclonal antibody bound to it. The known crystal structure of CPMV allowed the identification of the ‘footprint’ of the antibody on to the surface residues of the virus [31].

During this time, most of the cryoEM reconstructions were determined at ∼20 Å resolution and without an atomic model were limited to interpreting ‘blobs’ of density, and some crystallographers derided cryoEM as ‘blobology’. Through mid 1995, Baker and I pushed on together and enjoyed a variety of successes, where we combined virus crystallography and cryoEM. Mapping bulk RNA within nodaviruses and its interactions with the capsid (Fig. 3) was particularly satisfying [33]. In parallel, Baker worked with Rossmann, mapping monoclonal antibody binding to HRV14 and the ICAM1 receptor to the region near the ‘canyon’ at the pentamers of the virion [34].

Fig. 3.

The structure of Pariocota virus, a typical Nodavirus. At left is the fold of the A subunit (blue subunit in the quaternary structure and clustered about the 5-fold symmetry axes of the icosahedron). The structure is ramped in color as in Fig. 1. The T = 3 quaternary structure has characteristic protrusions at the quasi 3-fold symmetry axes (middle). At right is a cutaway of the capsid showing the ordered duplex RNA and those portions of the subunit polypeptide that interact with the RNA. The gamma peptides are in blue and can be seen to lie in cavities within the protein shell, shown as density from a cryoEM reconstruction.

Move to the Scripps Research Institute

In 1995, I moved from Purdue to the Scripps Research Institute after a long recruiting courtship that started in 1989 and included a 7-month sabbatical with Ian Wilson in 1993. Arriving at Scripps, we renewed collaboration with Anette Schneeman, a virologist who obtained her Ph.D. with Roland Rueckert, when we were all working on the structure of BBV and the related flock house virus (FHV). Schneeman's expertise allowed the pursuit of many structure-based studies of FHV function [35]. I also began collaboration with Mark Yeager, a cardiac physician who did cryoEM, when not doing angioplasties. Our last crystal structure determined at Purdue was NωV at 2.8 Å resolution [36]. This was the first T = 4 virus structure determined at near-atomic resolution, and the result was satisfying because the subunit fold was closely similar to that of FHV, but an entire Ig domain was inserted between two strands of the jellyroll fold. The molecular switching was striking and controlled from the C-terminal portion of the subunits. In that time frame, we made VLPs of NωV by expressing the capsid protein gene in a baculovirus system. The subunits spontaneously assembled and formed particles that were readily purified. The cryoEM reconstruction of the VLP, done in collaboration with Yeager, showed a 500 Å spherical particle with T = 4 quasi-symmetry and a dimer organization of the subunits. The subunits were all in closely similar positions, in spite of different quaternary structure environments. We were stunned because the virus structure determined by crystallography was 400 Å with trimeric capsomers and icosahedral in shape. There was considerable hand wringing, and we were suspicious that the Yeager group had made mistakes during the reconstruction. It was an awkward start to our collaboration. What started as a problem turned into a wonderful opportunity as we found that we could convert the provirion to the virion size and shape, by lowering the pH from 7.5 to 5. The mature form displayed the structure observed in the crystallography experiment, illustrating virus maturation (Fig. 4). In addition to the structural transition, the subunits underwent an autocatalytic proteolysis, when the pH was lowered to 5 [37]. This was the first example, where maturation could be explicitly controlled in vitro and lead to a number of insights regarding maturation and the opportunity to discern the behavior of quasi-equivalent subunits during maturation [38–40].

Fig. 4.

The subunit tertiary structure (color ramped as in Fig. 1) for Nudaurelia Omega Capensis virus is shown at left. The transition from procapsid to capsid for the T = 4 particle is depicted on the right. The color of the quaternary structure is ramped by radius from red to blue. The procapsid particle is about 490 Å in diameter with the subunits clustered as dimmers. After maturation at pH 5.0, the particle size decreases to 410 Å, and the subunits are clustered as trimers. An autocatalytic cleavage is activated when the particles compact [37].

Much earlier, it was realized that the autocatalytic reaction occurred in FHV and other Nodaviruses and that the cleavage was required for infectivity [41]. The active sites for the reactions in Noda and Tetraviruses were nearly identical [36]. We proposed that the cleavage covalently released a lytic peptide that left the particle and facilitated the breaching of a membrane to deliver the RNA into the cytoplasm. The problem with the proposal was the internal location of the cleaved peptide. Clearly, crystallography would not provide support for the proposal, but possibly a solution-based method would. We collaborated with Gary Suizdak and Brian Bothner, mass spectroscopists at Scripps, to look for evidence of transient exposure of the peptide. Purified particles were exposed to trypsin for various periods of time, and the particles were analyzed with matrix-assisted laser desorption/ionization mass spectrometry. The experiments showed that the gamma peptide was, indeed, the most readily proteolyzed region of the subunit, supporting the concept of transient exposure and membrane activity [42]. These studies were followed by a variety of in vitro and in vivo investigations that mapped the membrane-active motifs of this 44 amino acid polypeptide, revealing considerable mechanistic detail of its function [43–46]).

Our work on NωV benefited greatly from combining cryoEM and crystallography as we investigated the stabilization of the particle gained by the autocatalytic cleavage, the kinetics of the autocatalysis and the functional significance of the order of cleavage of the four subunits in the icosahedral asymmetric unit [47–49]. Indeed, employing time-resolved cryoEM lead to a deep appreciation of the flexibility of cryoEM, when compared with crystallography [50].

Bacteriophage structure and maturation

An extremely fruitful area of research on bacteriophage structure and function began, when we moved to Scripps. We collaborated with Roger Hendrix and Bob Duda at the University of Pittsburgh in our work on the crystallography of HK97, a lambda-like bacteriophage that they produced as tailless VLPs. CryoEM studies that they had done in collaboration with Alastair Steven and James Conway (at the National Institutes of Health) showed a large change in structure between the initial assembly product called prohead and the mature particle [51]. The transition changed the particle from 500 to 600 Å in diameter, altered the protein shell thickness from 40 to 18 Å and changed the shape from spherical to a facetted icosahedron. The opportunity to study these forms with crystallography was irresistible. William Wycoff, a graduate student in the laboratory at the start of the project, persevered for 5 years and produced a structure of the mature particle at 3.4 Å resolution, allowing the placement of every residue in the T = 7l quasi-equivalent surface lattice [52]. The subunits displayed a new fold, now known as the HK97 fold, and the quaternary structure was extraordinary, with subunits interleaving with each other in a manner not previously seen [53]. The topology of the inter-subunit, autocatalytic crosslinks was of particular note and generated considerable excitement. They formed catanated rings creating a chain mail fabric that made the mature particles impervious to denaturants as was previously shown by Duda in a remarkable paper that foresaw the chain mail explanation to the particle stability [54]. The subunit fold of HK97 appears to be the canonical tertiary structure of dsDNA bacteriophage as all subsequent structures of these subunits, as well as the herpes virus subunit, have displayed the HK97 fold. We followed up the work on the mature form with collaborative studies of the cryoEM structure of prohead and proposed a model for the large-scale transition associated with maturation [55]. The work continues into detailed studies of all known intermediates (Fig. 5) with a variety of methods [56–61].

Fig. 5.

The HK97 virus-like particle assembly and maturation pathway that is followed, when only the coat protein and protease are co-expressed in Escherichia coli. At the top are all the intermediates that have been characterized by crystallography and/or cryoEM. Below is shown the processing that occurs to the capsid protein, gp5, and the residues that form the autocatalytic crosslink. At right is an enlargement of the final mature particle indicating that the cross-linked gp5 proteins mechanically chain link the particle together. Each ring of the same color corresponds to five or six subunits chain linked together by the isopeptide bond formed by the side chains of Asn 356 and Lys169.

Sulfolobus-turreted icosahedral virus

Our recent cryoEM work, currently lead by Chi-Yu Fu and David Veesler in the laboratory, and done in collaboration with Mark Young and Martin Lawrence at Montana State University, has focused on structural studies of the 75 Mdalton Sulfolobus-turreted icosahedral virus that has been highly informative in understanding a virus that has an extracellular portion of its life cycle at 80°C and pH 3. We studied purified particles at 12 Å resolution [62], determined the crystal structure of the major capsid protein [63] and fitted the atomic coordinates to the cryoEM density, and then studied the virus inside the sulfolobus cell with cryoEM tomography [64], revealing steps in the virus assembly process and remarkable pyramids formed in the outer membrane of the sulfolobus cells that allow release of the virus particles. Recently, the particle structure was extended to 4.4 Å with cryoEM, and this structure is being interpreted with high resolution crystal structures of the two additional major gene products in the T = 31 d capsid solved by crystallography in the laboratory of Martin Lawrence at Montana State University.

The last chapter

At this point, I am on course to close my laboratory on June 30, 2015. That corresponds to 43 years of studying virus structure and function, and it has been a truly extraordinary experience. The future of virus work seems as exciting as ever, and the opportunity to determine structure at all levels is remarkable. Thanks to my close friend and collaborator Professor Peter Doerschuk at Cornell University, we are specifically identifying the variance that occurs at each pixel position of a cryoEM-reconstructed dataset [65]. This will allow higher resolution structures to be determined of the more stable regions of the particle by weighting reconstructions with the inverse variance, and it will help to identify functionally important regions of a structure because they are frequently more dynamic. Dr Andrew Routh, who is currently in thelaboratory, has developed an exceptionally informative procedure for studying RNA virus recombination and packaging ‘errors’ by deep sequencing the RNA packaged in FHV [66–68]. Finally, the arrival of the direct electron detector for electron microscopy [69,70] ranks with the major technical breakthroughs in the field and will change cryoEM fundamentally.

I believe that icosahedral virus crystallography will not be common in the future because cryoEM avoids the need for crystallization, and the ‘purification’ can be performed digitally with powerful clustering algorithms that generate discrete class averages for reconstruction. The new detectors allow explicit corrections for beam-induced motion, improving the resolution even further [71]. It seems like an appropriate time to direct my individual efforts toward challenging computing problems that confront cryoEM as it approaches the resolution of X-ray crystallography and that is how I hope to spend my ‘retirement.’