- Record: found
- Abstract: found
- Article: found
Gently does it for submicron crystals
article-commentary
Publication date (Electronic):
19 November 2013
Journal:
eLife
Publisher:
eLife Sciences Publications, Ltd
Read this article at
There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.
Abstract
A protein structure has been refined with electron diffraction data obtained by using a very weak electron beam to collect large numbers of diffraction patterns from a few sub-micron-sized three-dimensional crystals.
Related collections
Most cited references7
- Record: found
- Abstract: found
- Article: not found
The potential and limitations of neutrons, electrons and X-rays for atomic resolution microscopy of unstained biological molecules.
R. Henderson (1995)
Radiation damage is the main problem which prevents the determination of the structure of a single biological macromolecule at atomic resolution using any kind of microscopy. This is true whether neutrons, electrons or X-rays are used as the illumination. For neutrons, the cross-section for nuclear capture and the associated energy deposition and radiation damage could be reduced by using samples that are fully deuterated and 15N-labelled and by using fast neutrons, but single molecule biological microscopy is still not feasible. For naturally occurring biological material, electrons at present provide the most information for a given amount of radiation damage. Using phase contrast electron microscopy on biological molecules and macromolecular assemblies of approximately 10(5) molecular weight and above, there is in theory enough information present in the image to allow determination of the position and orientation of individual particles: the application of averaging methods can then be used to provide an atomic resolution structure. The images of approximately 10,000 particles are required. Below 10(5) molecular weight, some kind of crystal or other geometrically ordered aggregate is necessary to provide a sufficiently high combined molecular weight to allow for the alignment. In practice, the present quality of the best images still falls short of that attainable in theory and this means that a greater number of particles must be averaged and that the molecular weight limitation is somewhat larger than the predicted limit. For X-rays, the amount of damage per useful elastic scattering event is several hundred times greater than for electrons at all wavelengths and energies and therefore the requirements on specimen size and number of particles are correspondingly larger. Because of the lack of sufficiently bright neutron sources in the foreseeable future, electron microscopy in practice provides the greatest potential for immediate progress.
- Record: found
- Abstract: found
- Article: found
Radiation damage in macromolecular crystallography: what is it and why should we care?
Elspeth Garman (2010)
1. Introduction The advent of highly intense wiggler and undulator beamlines fed from synchrotron sources has reintroduced the age-old problem of X-ray radiation damage in macromolecular crystallography (MX) even for crystals held at cryogenic temperatures (100 K). Unfortunately, such damage to macromolecular crystalline samples during the experiment is a problem that is inherent in using ionizing radiation to obtain diffraction patterns and has presented a challenge to MX since the beginning of the field. For room-temperature (RT) data collections, it often necessitates the use of many crystals to assemble a complete data set, because the crystalline order of the sample is damaged and decreases during the experiment and thus the diffracted intensity fades. The root cause of this damage is the energy lost by the beam in the crystal owing to either the total absorption or the inelastic scattering of a proportion of the X-rays as they pass through the crystal. The measure of this energy loss is the ‘dose’ measured per mass of the sample, given in SI units of grays (Gy; 1 Gy = 1 J kg−1). Dose may also be quoted in terms of the non-SI unit rad (radiation absorbed dose; 1 rad = 10 mGy). In MX, dose measurements are generally of the order of a million grays (1 MGy or 100 Mrad). The earliest investigation of radiation damage at RT in MX was carried out nearly 50 years ago by Blake & Phillips (1962 ▶) on a sealed-tube (copper) X-ray source. By making seven sets of successive measurements, they monitored the decay in the diffraction intensity of a particular set of reflections from crystals of sperm-whale myoglobin over a period of 300 h. They concluded that the damage was proportional to the irradiation time, which they assumed was linearly proportional to the absorbed dose. They deduced that a single 8 keV X-ray photon disrupts around 70 protein molecules and disorders a further 90 protein molecules for doses up to about 20 Mrad (0.2 MGy) absorbed after 100 h of X-ray exposure. The observed form of the decay with dose could be described by an exponential function representing a first-order process, where I t corresponds to the measured intensity at a particular time, I 0 is the initial intensity, B is a measure of disorder, θ is the angle of diffraction and λ is the incident X-ray wavelength. According to their model, after any irradiation the crystal consists of three components: (i) an undamaged fraction (A 1) which is entirely responsible for the remaining diffraction at high angles, (ii) a highly disordered fraction (A 2) which only contributes to the diffraction at low angles and (iii) a thoroughly disorganized or amorphous part [1 − (A 1 + A 2) = A 3] which no longer contributes to the single-crystal diffraction at all. From their plot of A 1, A 2 and A 3 against time derived from the seven successive sets of measurements and using their dose estimates it can be deduced that half of the crystal volume became amorphous after a dose of 0.59 MGy. Blake & Phillips (1962 ▶) also suggested that the protein molecules suffered specific structural damage. This conclusion was reached without knowledge of either the sequence or the three-dimensional structure of the protein, and the postulate was only confirmed many years later when radiation damage to disulfide bridges was noted in electron-density difference maps calculated from data collected from des-pentapeptide insulin crystals (Helliwell, 1988 ▶) as well as the opening of aromatic side chains in maps of ribonuclease (unpublished results from Burley and coworkers referred to in Helliwell, 1988 ▶). As early as 1958, it was postulated that covalent bonds in proteins provided a migratory route for ionizing energy from absorbed incident radiation to break weaker bonds (Augenstein, 1958 ▶). Breakage of disulfide bonds had been reported following the irradiation of solutions of the proteins trypsinogen and chymotryspinogen by 186 keV electrons (produced by the decay of the radionuclide 35S; Pechère et al., 1958 ▶). The presence of sulfur radicals and the subsequent formation of —SH groups was confirmed by ESR measurements (Gordy & Shields, 1958 ▶). Following the work of Blake & Phillips (1962 ▶), various researchers (Hendrickson et al., 1973 ▶; Hendrickson, 1976 ▶; Fletterick et al., 1976 ▶) investigated the radiation-damage problem in protein crystals both theoretically and experimentally at RT and made modifications to the initial model presented above. A detailed description of these developments can be found in the literature (Southworth-Davies et al., 2007 ▶) and will not be repeated here. The resulting working model for RT damage which fitted all the available data was that there appeared to be no direct pathway between states A 1 and A 3 and thus the rate constant for transition from undamaged to amorphous was zero. Additionally, it was found necessary to include an intermediate dose-dependent stage labelled A 1′ between the undamaged and the damaged stages as shown in (2). This state conformationally resembled the undamaged state and thus still contributed to diffraction at all angles (Sygusch & Allaire, 1988 ▶), Up until the 1990s, MX data were almost exclusively collected at RT, where the recommended practice was to monitor the intensity I 0 of a strong reflection as the experiment proceeded and to discard the crystal once the intensity had dropped to 0.85I 0, or at the very worst 0.70I 0 if the particular crystals were in very short supply (Blundell & Johnson, 1976 ▶). Much earlier, improved resolution of diffraction had been observed for crystals held at 246 K (King, 1958 ▶), although at the time this was not understood in terms of reduced radiation damage. Systematic measurements comparing the decay of two particular reflections for crystals held at 198 and 298 K (Haas & Rossmann, 1970 ▶) and efforts to import small-molecule crystallography cooling techniques into MX (Hope, 1988 ▶) showed that this would be an effective radiation-damage mitigation strategy. By irradiating the crystal while holding it at a reduced temperature, its lifetime should be significantly improved, since many of the radical species produced by the energy loss of the beam would diffuse much more slowly or not at all and would thus not further interact, so reducing the collateral damage. The cryocooling technique blossomed and was made technically more accessible for routine use in MX because of two pivotal developments: the loop-mounting method (Teng, 1990 ▶), in which the protein crystal is held by surface tension in a film of liquid ‘cryo-buffer’ across a small-diameter (1 mm down to 0.1 mm) nylon or fibre loop, and the availability of a reliable open-flow unpressurized cryostat with flexible stainless-steel hosing (Cosier & Glazer, 1986 ▶) to supply a stream of cooled gaseous nitrogen at a stable temperature of around 100 K with which to surround the sample during data collection. Initially, problems with the technique included ice formation within and outside the crystal and an increase in mosaic spread, particularly when cryocooling protocols were not optimized. Methods for improving the data quality obtainable were soon developed (Rodgers, 1997 ▶; Garman & Schneider, 1997 ▶; Garman, 1999 ▶; Pflugrath, 2004 ▶; Garman & Owen, 2006 ▶) and there was widespread adoption of the technique. In fact it has been estimated that over 90% of all protein structures are now determined at cryo-temperatures (Garman, 2009 ▶). The advantages of cryocooling for MX are a reduction in the rate of radiation damage; the use of a mounting technique (the loop) that is usually more gentle than the capillary method historically used for RT collection; the fact that higher resolution data can more easily be obtained because the crystal order is preserved for longer; a lower background in the diffraction experiment as it is not necessary to enclose the crystal in a glass, quartz or plastic tube to prevent dehydration; that fewer crystals (and thus a lower quantity of protein) are required for a project; that crystals can be shipped ahead of time to the synchrotron (more or less) safely; and that crystals can be flash-cooled when in peak condition for future use before they start to degrade in the crystallization drop. These positive aspects of cryocooling commonly outweigh the disadvantages. The latter include the requirement for expensive cryostat cooling equipment, a frequent increase in crystal mosaic spread (but not necessarily if the cryoprotection concentration and crystal handling are carefully optimized), the need to invest time for optimization of cryo-buffers and cooling protocols, and the fact that there are as yet no protocols that guarantee success, although progress is being made in this direction (see, for example, Alcorn & Juers, 2010 ▶). The improvement in dose tolerance for a crystal held at 100 K compared with a crystal irradiated at RT has been estimated to be approximately a factor of 70 on average (Nave & Garman, 2005 ▶). Thus, cryocooling is clearly a highly effective mitigation strategy. However, radiation damage is now routinely observed at synchrotrons in cryocooled crystals and the experimenter would be wise to be aware of the artefacts that can be produced. Below, the symptoms of radiation damage at cryotemperatures and the basic physical processes involved are described, the reasons why the crystallographer should care about this issue are addressed, and our current knowledge, as reflected in the published literature, is collated. The interested reader is also referred to Garman & Owen (2006 ▶) and Ravelli & Garman (2006 ▶), and to a recent article entitled A beginner’s guide to radiation damage (Holton, 2009 ▶). 2. What are the symptoms of radiation damage at cryotemperatures? Systematic studies of this phenomenon have identified two separate indicators of damage as a function of dose: global (Fig. 1 ▶) and specific (Fig. 2 ▶) damage. The former results in a loss of the measured reflection intensities (particularly at high resolution), expansion of the unit-cell volume, increasing values of the measure of the internal consistency of the data which quantifies the difference between reflection intensities that should ideally be the same (R meas), an increase in both the scaling B factors for the data and the atomic B values of the refined structure, rotation of the molecule within the unit cell and often (but not always) an increase in mosaicity. Visible differences in the samples as the experiment proceeds, including colour changes, are also observed. On warming of the sample following irradiation, bubbles of gas, now proposed to be hydrogen (Meents et al., 2009 ▶, 2010 ▶) and perhaps some CO2, are emitted and discolouration of the sample is common (see Fig. 3 ▶). Various metrics have been suggested and used for monitoring global damage, among which are the following. (i) I D /I 1, where I D is the summed mean intensity (I mean) of a complete data set (or equivalent sections of data) after a dose D and I 1 is the mean intensity of the first data set. Note that using I D /σ D (where σ D is the standard deviation of the signal, i.e. the ‘noise’) normalized to the intensity I 1/σ1 of the first data set is not a robust metric since the noise σ D increases with dose and thus I D /σ D reduces by an amount that more than represents the true loss of diffracting power. (ii) R d, the pairwise R factor between identical and symmetry-related reflections occurring on different diffraction images, plotted against the difference in dose, ΔD, between the images on which the reflections were collected (Diederichs, 2006 ▶). The plot of R d against ΔD is a straight line parallel to the x axis if there is no damage, but rises linearly in the presence of damage (see Fig. 4 ▶). This plot can be used to correct the intensity values of the reflections back to their ‘zero-dose’ values to improve the data quality (Diederichs et al., 2003 ▶). (iii) The isotropic B factor (B rel) has been found to be a robust measure of radiation damage at 100 K and to be linearly dependent on it (Kmetko et al., 2006 ▶). An example of B rel plotted against dose is given in Fig. 5 ▶. The relative B factors can be interpreted as proportional to the change in the mean-squared atomic displacements. A coefficient of sensitivity to absorbed dose, S AD, was also defined, S AD = ΔB rel/ΔD8π2, where ΔB rel/8π2 is the change in relative isotropic B factor and ΔD is the change in dose as above, i.e. S AD is the slope of the line in a graph such as that shown in Fig. 5 ▶. This metric relates the increase in mean-squared atomic displacements to the dose and it has been postulated that it is similar within quite a narrow range of values for most protein crystals (Kmetko et al., 2006 ▶). (iv) The volume of the unit cell increases more or less linearly with dose and was originally thought to be a possible metric for judging the extent of radiation damage; however, systematic work (Murray & Garman, 2002 ▶; Ravelli et al., 2002 ▶) has shown that it is not a reliable indicator since crystals of the same size and type expand at different rates with increasing absorbed dose. (v) Although mosaicity commonly increases with dose, it is not a reliable metric for quantization of radiation damage, since it does not behave in a reproducible or predictable manner. Of more direct relevance to the biological interpretation of structures than the global indicators detailed above is the fact that specific structural damage to particular covalent bonds is observed to occur in a reproducible order in many proteins (Weik et al., 2000 ▶; Burmeister, 2000 ▶; Ravelli & McSweeney, 2000 ▶): first disulfide bridges elongate and then break (Weik et al., 2002 ▶), then glutamates and aspartates are decarboxylated, tyrosine residues lose their hydroxyl group and subsequently the carbon–sulfur bonds in methionines are cleaved. Such damage is illustrated in Fig. 2 ▶, which shows damage to glutamate and methionine residues in a cryocooled crystal of apoferritin during sequential data sets collected on beamline ID14-4 at the ESRF. Covalent bonds to heavier atoms such as C—Br, C—I and S—Hg are also ruptured (see, for example, Ramagopal et al., 2005 ▶). Clearly, it is not feasible to monitor the specific structural damage during the experiment, since the refined structures are required. However, it is known that this damage often occurs well before there is any obvious degradation of the diffraction pattern. The global effects of radiation damage at 100 K are thought to be independent of dose rate up to the flux densities currently used (1015 photons s−1 mm−2; Sliz et al., 2003 ▶). Another study concurred with this finding but, following an analysis of electron-density difference maps, indicated that there might be a second-order dose-rate effect since specific damage was slightly more severe at higher dose rates (Leiros et al., 2006 ▶). Conversely, however, Owen et al. (2006 ▶) reported a small (10%) reduction in D 1/2 (the dose required to halve the original diffraction intensity) for a dose-rate increase from 4 × 103 to 40 × 103 Gy s−1 at flux densities of 4 × 1012 and 4 × 1013 photons s−1 mm−2, respectively. The manifestations of radiation damage in the diffraction experiment can now be monitored over a range of time scales and doses (illustrated in Fig. 6 ▶). For instance, the formation of the disulfide-anion radical, , can be observed in real time using UV/UV–vis microspectrophotometry after a few tens of milliseconds of X-ray irradiation as a 400 nm absorption peak, and solvated electrons have a maximal absorbance at 550–600 nm (McGeehan et al., 2009 ▶). This specific structural damage is often apparent in electron-density maps calculated using the structure factors of a data set that took around 30 s to collect and the resulting structure represents a time and space average over the 30 s of irradiation and over all the molecules in the crystal (Fig. 6 ▶). Metal centres are also reduced very swiftly by the X-ray beam and increasingly this can be monitored on-line during the X-ray experiment (see, for example, Hough et al., 2008 ▶). The global intensity loss owing to radiation damage is clearly evident following the collection of several data sets in succession from the same crystal when the summed intensity for each data set is plotted normalized to the intensity of the first data set (Fig. 6 ▶, right). 3. What is it? Radiation damage to the sample is a result of it absorbing photons from the beam by either the photoelectric effect (total absorption of the photon and ejection of an inner shell electron) or Compton scattering (inelastic scattering of the photon, which then escapes following a varying amount of energy loss to an atomic electron, which can also be ejected). At the incident energies used for MX, the former effect has a much higher cross-section and dominates the absorption, accounting for over 90% of the energy deposited by the beam. Each photoelectron has enough energy to subsequently induce up to ∼500 further ionization events, which in turn can result in the formation of radical species in the crystal. In protein crystals, the presence of anything between 20% and 80% solvent means that the radiolysis of water and other components of the solvent is an important contributor to the creation of these species. Some of the energy deposited by the beam during these processes is converted into heat and induces a temperature rise in the sample. The diffracted photons are scattered elastically and thus do not contribute to the damage. These processes are illustrated diagrammatically in Figs. 7 ▶(a), 7 ▶(b) and 7 ▶(c). It is worth noting that for a 100 µm thick protein crystal only 2% of the incident photons of a 12.4 keV (1 Å) X-ray beam will interact in any way with it [∼1.7% (i.e. 84% of interacting photons) by the photoelectric effect and ∼0.15% (8%) by the Compton effect, with only ∼0.15% (8%) actually diffracting]. The usage of the terms ‘primary’, ‘secondary’ and ‘tertiary’ damage has become somewhat inconsistent in the literature and is largely a matter of semantics, but the definitions that will be adopted here are as follows. (i) Primary damage is the ionization of an atom owing to photoelectric absorption or Compton scattering. The primary photoelectron has a mean track length of a few micrometres (for 12 keV photons; O’Neill et al., 2002 ▶). (ii) Secondary damage is that arising from the formation of up to 500 low-energy secondary electrons per primary absorption event, which are able to diffuse and induce further ionization and excitation events (e.g. electronic and vibrational). The secondary electrons gradually become thermalized (that is, they have the distribution of energies expected at the equilibrium temperature of the sample) and chemical reactions between the radiation-induced moieties and the crystal components then become important. (iii) Tertiary damage is defined as the effect on the crystal lattice and other mechanical consequences of the energy deposition in the crystal. Damage can also be classified as direct, if the primary absorption event occurs at an atom in the protein molecule, or indirect, if the radiation is absorbed by the surrounding solvent and the reactive species formed subsequently interact with the protein. Energy deposition in the water in and around the crystal results in a cascade of reactions as shown below (Ward, 1988 ▶), giving hydroxyl radicals, hydrated electrons and H atoms, the relative amounts of which depend on the temperature, pH and other factors, As described in §2 above, the knock-on effects of the energy absorbed by the crystal manifest themselves as both a reduction of crystalline order (global damage) and specific structural damage, and over the last 10 years MX researchers have sought to identify mechanisms that explain these observations. At RT the products have thermal energy and can diffuse through the crystal, causing more secondary damage as they go. However, at cryotemperatures below 110 K nearly all the radical species, including radicals (Mike Sevilla, private communication), are immobilized, with the notable exception of electrons. These can quantum-mechanically tunnel along the amino-acid backbone and have been shown by ESR measurements to be mobile at 77 K (Jones et al., 1987 ▶). They migrate and seek out the most electron-affinic sites in the protein which, if there are no bound metals, are the disulfide bonds. This phenomenon accounts for the ‘pecking order’ of amino acids susceptible to specific structural damage. This mechanism also explains why the observed damage does not occur in the order of the largest to smallest X-ray absorption cross-sections of the atoms, as would be expected if there were no mobile species. Away from absorption edges, X-ray absorption cross-sections rise swiftly with the atomic number of an atom, so if the specific structural damage arose from primary processes alone the C—S bond in methionine should be the second most susceptible bond (after the disulfide bond). The reason for the global damage to crystalline order observed in MX was until recently thought to be the consequence of direct damage to the protein molecules. However, new results show that the loss of diffractive power may instead be attributable to the production of hydrogen gas in the crystal. At 100 K it is likely that the hydrogen gathers at interfaces between crystal domains, which would account for the commonly observed increase in mosaicity with dose. However, at 50 K it has been found that the rate of the specific structural damage with dose was reduced by a factor of four, although the global damage was not much slower than at 100 K. It is thus more likely that rather than collecting at grain boundaries the hydrogen is trapped within the unit cell at 50 K, accounting for the larger unit-cell increase at this temperature (Meents et al., 2010 ▶). An important experimental consideration is that at cryotemperatures the damage does not usually spread along the crystal. The crystal can thus be translated with respect to the beam so that fresh undamaged crystal can be irradiated at multiple positions. The exception is when heavy-atom clusters are present. The high absorption of these atoms can result in local heating of the crystal above 110 K, at which diffusion of radicals becomes more probable (e.g. for Na+,K+-ATPase crystals soaked with Ta6Br12 2+; Poul Nissen and J. Preben Morth, private communication at the Petra-III Workshop 2007). In this case, translation of the crystal to a new position can be unsuccessful as a strategy for obtaining more data, since the damage can spread several tens of micrometres along the crystal from the irradiated position. With the advent of X-ray microbeams, the question arises as to how close sequential irradiations can be made while ensuring that ‘fresh’ material is in the beam, and there is ongoing systematic research to investigate this (Robert Fischetti, private communication). 4. Why should we care? Radiation damage in MX is an increasingly important and limiting problem for several reasons. Firstly, as the diffraction experiment proceeds, creeping non-isomorphism occurs on three simultaneous fronts: the unit-cell volume increases, there is often movement of the protein molecule within the unit cell, and structural changes are induced by the damage, so that the protein conformation is changing during the measurements. This non-isomorphism is thought to be a major cause of unsuccessful MAD (multiple-wavelength anomalous dispersion) structure determinations, since by the time the second or third wavelength is collected, the cell and atomic structure can have changed such that the reflection intensities are significantly altered. This effect can obscure the anomalous signal required for structure solution. It has been calculated that a 0.5% change in all three dimensions of a 100 Å3 unit cell would change the intensity of a 3 Å reflection by 15% (Crick & Magdoff, 1956 ▶) so the MAD/SAD phasing signals would be completely destroyed. An empirical rule of thumb for successful MIR phasing has been proposed for the absolute shift in unit-cell dimensions (X) that can be tolerated as a function of the resolution limit of the data set (d min): X = d min/4 (Drenth, 1999 ▶). Secondly, the radiation-sensitivity of some crystals at 100 K means that it is not possible to collect a complete data set from a single crystal and data must be merged from several (or many) of them to measure all the unique reflection intensities. Although this was routinely the case when data were collected at RT, most crystallographers have become accustomed to being able to measure all unique reflections from just one cryocooled crystal. Use of multiple crystals to assemble a complete data set in general increases the errors arising from non-isomorphism, thereby potentially reducing the ease of structure solution as well as increasing the mounting/dismounting time burden. Even using a robot for this operation can be slow and in fact is sometimes the most time-consuming part of the experiment. It can also present some pitfalls during processing. For instance, space group I4 can be indexed with the b axis pointing in either direction, so that when data are merged care must be taken that each section is indexed in the same convention. Finally, the radiation-damage-induced structural changes can affect the apparent biological properties of the macromolecule under study. Enzyme mechanisms can involve redox-susceptible residues, so special care is required when interpreting structures that may have been modified by X-ray damage during the data collection. For instance, irradiation can change the oxidation state of metal ions in structural/active sites from that in their native state (Carugo & Djinovic Carugo, 2005 ▶; Yano et al., 2005 ▶) and cause the decarboxylation of glutamate and aspartate residues. X-ray-induced structural changes can also be misleading in studies of intermediates (e.g. Takeda et al., 2004 ▶). In such circumstances, separating radiation damage from an enzymatic mechanism can be extremely difficult and can cast doubt on the validity of biological conclusions drawn from crystal structures (Ravelli & Garman, 2006 ▶). In summary, radiation damage ultimately results in lower resolution structures, failed MAD structure solutions and sometimes the inaccurate interpretation of biological results if no control experiments are carried out to account for radiation-damage artefacts. It is thus an issue to be taken seriously by the structural biologist. 5. What is ‘dose’ and the ‘dose limit’? As already stated, the universal metric against which the decay indicators of a crystal are conveniently measured is the absorbed dose, which is defined as the energy absorbed per unit mass of the sample (Gy = J kg−1) in the irradiated volume. The fact that the amount of damage at 100 K is indeed proportional to the absorption coefficient and thus to the dose has been shown in elegant experiments by Kmetko et al. (2006 ▶) on lysozyme crystals soaked in a range of concentrations of various heavy-atom solutions. The ‘dose postulate’ states that there exists a universal ‘dose limit’, which is the maximum energy/mass that a macromolecular crystalline sample can tolerate before the diffraction will fade to a given level (traditionally half) of its original intensity. A crystal might not survive until the limit is reached (e.g. if there were susceptible residues at crystal contacts; Murray et al., 2005 ▶), but it would not be expected to survive beyond it. From observations made of the dose which generally caused biological samples at 77 K to lose half of their diffracting power (D 1/2) during two-dimensional diffraction experiments in electron microscopy, Henderson (1990 ▶) estimated a ‘dose limit’ (known as the ‘Henderson limit’) for three-dimensional macromolecular X-ray crystallography of 20 MGy. This was later measured experimentally in a series of experiments on apoferritin and holoferritin crystals (see Fig. 6 ▶), the absorption coefficients of which differ by a factor of two (Owen et al., 2006 ▶). The composition of the crystals was determined using proton-induced X-ray emission (PIXE; Garman & Grime, 2005 ▶) in order to obtain as accurate values as possible of, in particular, the iron content. This minimized the errors in the dose calculations. The dose limit (D 1/2) was found to be 43 MGy, although the recommended maximum dose was only 30 MGy in order to avoid compromising the biological information extracted from deduced structures. This dose limit corresponded to a drop in diffraction intensity to 70% (D 0.7) of the initial value (Owen et al., 2006 ▶). A number of other studies have corroborated this dose limit. An analysis of all the various experimental measurements has been made by Howells et al. (2009 ▶), who concluded that the resolution-dependent D 1/2 was 10d MGy (where d is the resolution in Å: thus for a 2 Å reflection D 1/2 = 20 MGy). This issue is described in detail later in this volume (Holton & Frankel, 2010 ▶). As noted above, this limit is thought to be largely independent of dose rate at cryotemperatures at the flux densities currently used in MX. It is also worth reiterating that structural damage generally occurs well before visible degradation of the diffraction pattern is observed. Thus, it is inadvisable to plan an experiment which requires collecting data beyond the time when the dose limit (which was determined from intensity decay) is reached. In the RT model developed by Blake & Phillips (1962 ▶), damage is directly proportional to dose and no dose-rate effect is included. Despite anecdotal reports from the early days of synchrotron use with crystals irradiated at RT that they had longer lifetimes at higher dose rates, this was only systematically investigated recently, when an inverse dose-rate effect was measured in-house at RT between dose rates of 6 and 10 Gy s−1, the higher rate giving four times the dose tolerance (i.e. four times the dose required to halve the total diffraction intensity, D 1/2) for hen egg-white lysozyme crystals (Southworth-Davies et al., 2007 ▶). For irradiation at a dose rate of 2800 Gy s−1 at a synchrotron at RT, ten times the dose tolerance has been recorded (Barker et al., 2009 ▶). The explanation of this phenomenon is that at high dose rates radicals produced in the crystal neutralize one another and thus do not cause further damage, whereas at lower dose rates they travel further, interacting with protein and solvent to produce additional damage. Interestingly, the RT exponential intensity decay with dose, which is typical of a first-order process (where the decay rate depends on the amount of material left), can be modified by the addition of scavenger molecules to become a zeroth-order dependence (where the rate of decay is a constant). This effect is not yet completely understood. At RT, the dose tolerance of HEWL crystals (as measured by the change in D 1/2) has been shown to be improved by factors of ∼2 and ∼9 by the addition of the scavengers ascorbate and 1,4-benzoquinone, respectively (Barker et al., 2009 ▶). To calculate the available time in the beam before the crystal reaches the experimental dose limit, knowledge of the sample size and composition (i.e. the number of each atom type in the unit cell) is required so that absorption coefficients can be computed, as well as detailed information about the incident beam [energy, size, shape and flux (in photons s−1)]. For MX, this can be conveniently carried out by means of the program RADDOSE (Murray et al., 2004 ▶; Paithankar et al., 2009 ▶; Paithankar & Garman, 2010 ▶), version 3 of which includes both the probability of fluorescent X-ray escape (non-negligible for heavy-atom-containing crystals) and the energy loss owing to Compton scattering (non-negligible above 20 keV). The calculations rely on accurate flux measurements being available for the X-ray beam at the particular beamline being used (Owen, Holton et al., 2009 ▶). However, RADDOSE does not yet give accurate results for crystals larger than the beam size where a fresh unirradiated crystal is continually being rotated into the beam. The time before the experimental limit is reached is thus underestimated in these cases. Currently, developments are under way that aim to provide on-line digitization of both the crystal shape and its position relative to the rotation axis of the goniometer. These efforts are being largely driven by the need for improved absorption corrections, but when the crystal information can be incorporated into RADDOSE they should also make possible the further improvement of dose estimates. 6. What do we know and what would we like to know? There are many parameters that can be varied in an MX experiment, some of which can affect the rate of radiation damage to (or ‘dose tolerance’ of) a crystal. There are two challenges for researchers seeking to understand and trying to mitigate radiation damage. The first is to truly isolate the variable to be tested and to only change one experimental condition at a time so that definite conclusions can be reached. The second is to use a reliable metric(s) of radiation damage so that the effect of protocol modifications can be properly assessed. In addition, to reach a statistically significant result the same experiments must be repeated and reproduced on several crystals of the same protein and ideally extended to check the validity of the results in a more general way by conducting the same tests for a number of different proteins. Over the last 10 years there has been an extensive search for reliable metrics of global and structurally specific radiation damage. Since structural changes occur even before degradation of diffraction quality is apparent, intensity loss cannot be used as a yardstick with which to judge damage to specific amino acids, which is only obvious when the electron-density maps have been calculated once enough data have been collected. This can be understood because the diffraction loss occurs in reciprocal space and the specific damage in real space, and one point in real space contributes to all reflections in reciprocal space and vice versa. The parameter space of an MX experiment is composed of variables that can be categorized as follows (Garman, 2003 ▶). (i) The crystal in the cryo-loop: heavy-atom content (Se, S etc.), solvent content, solvent composition, crystal size and surface-to-volume ratio, the amount of residual liquid around the crystal prior to flash-cooling, the choice and concentration of cryoprotectant agent, the time spent in the cryobuffer prior to cooling, the flash-cooling method (stream or liquid), the cryogen used to flash-cool, the amount of crystal manipulation, the local humidity and the speed of the experimenter when flash-cooling from the cryobuffer drop. (ii) The X-ray beam: the flux density, the energy (wavelength), the beam size compared with the crystal size, the dose and the dose rate. (iii) The cryostat: the cold gas flow rate, the temperature and the cryogen (N2 or He). Systematic experiments to address the dependence of the rate of radiation damage on all these factors would take many years and be very labour-intensive in terms of data collection and processing, as well as requiring many hours of synchrotron beamtime. However, some of these variables have been investigated and studies can be broadly categorized as follows. (i) Crystal related. What is the minimum crystal size? What affects X-ray absorption? Can the unit-cell expansion be used as a metric? What is the effect of temperature (e.g. 100, 16, 40 K)? Does the addition of free-radical scavengers increase dose tolerance? What are the susceptibilities of particular amino acids to specific damage and why? (ii) X-ray beam related. What is the effect of changing the incident wavelength? Is it beneficial to change/regulate the dose/dose-rate regime? What is the effect of the beam size compared with the crystal size? Does the beam heat the crystal? (iii) Method developments and applications. Development of convenient flux calibration of beamlines. Development of on-line and off-line spectroscopy (UV–vis, Raman, XAS, EPR). Studying the effect on the success of MAD/SAD phasing. Development of RIP/RIPAS. Application/effect of radiation damage to/on the study of biological mechanisms. Phase transitions and/or radiation-induced changes with temperature-controlled cryocrystallography to study macromolecular function. Software developments. Finding strategies to minimize radiation damage in data collections. Extending the understanding of radiation damage in RT data collections. Studying RNA/DNA damage. The many experiments on the above topics that have been reported to date will not be detailed here, but a summary of the currently available literature is presented in Table 1 ▶. Useful collections of research papers addressing different aspects of radiation damage in MX can be found in four special issues of the Journal of Synchrotron Radiation, which each contain eight or more research papers presented at the Second, Third, Fourth and Fifth International Workshops on X-ray Radiation Damage to Biological Crystalline Samples [Journal of Synchrotron Radiation, Vol. 9 (2002), pp. 327–381, Vol. 12 (2005 ▶), pp. 257–328, Vol. 14 (2007), pp. 1–132 and Vol. 16 (2009), pp. 129–216, respectively]. These each have a brief introduction placing the contributions into the wider context of research in the field (Garman & Nave, 2002 ▶, 2009 ▶; Nave & Garman, 2005 ▶; Garman & McSweeney, 2007 ▶). The real question to which experimenters would like an answer is: what can I do to obtain the largest amount of data with the highest signal-to-noise ratio from my crystal in the beam? The current advice would include the following: (i) backsoaking of crystals to remove any nonspecifically bound heavy atoms in the mother liquor (e.g. the arsenic in cacodylate) or in a soaking solution for heavy-atom phasing, since these heavy atoms can contribute a lot to the dose owing to their high absorption but do not provide useful phasing information; (ii) sacrificing a crystal (if more than one crystal of a protein exists) to obtain a data set where the aim is to assess the radiation-sensitivity so that a suitable data-collection protocol can be designed; (iii) matching the beam size to the crystal size; (iv) if possible using a beam with a top-hat profile (or by careful slit setting selecting the central peak portion of a Gaussian-shaped beam) so that the crystal does not suffer differential radiation damage across its irradiated volume; (v) using BEST to optimize the data-collection strategy taking radiation damage into account (Bourenkov & Popov, 2010 ▶) and (vi) being satisfied with a 3 Å comparatively undamaged data set for phasing rather than chasing the 2.5 Å diffraction which fades as you watch and will thus be less useful. Above all, experimenters should make themselves aware of the parameters known to affect the rate of radiation damage, so that intelligent choices/compromises can be made. 7. Conclusions Since systematic research into MX radiation damage at 100 K began in earnest in the late 1990s, significant progress has been made in our knowledge and understanding of the phenomenon and much anecdotal evidence has been replaced by solid experimental results. We understand better how to perform investigations to identify suitable metrics and the importance of routinely measuring the X-ray flux so that the absorbed dose can be calculated. The research has also prompted some exciting new approaches such as RIP, UV-RIP/RIPAS, ‘time-resolved’ cryocrystallography and on-line spectroscopy. However, there are still many areas where systematic investigations are required to improve our understanding of the radiation chemistry within an irradiated protein crystal held at either RT or at various cryotemperatures so that better strategies for minimizing damage can be developed. The most useful contribution to be made by MX radiation-damage research is in identifying concrete experimental protocols for everyday use on synchrotron beamlines so that researchers can ensure that they obtain the maximum possible high-quality data from their crystals. This would firstly facilitate structure solution and secondly avoid compromising the biological information extracted from the structure once obtained.
- Record: found
- Abstract: found
- Article: found
A Medipix quantum area detector allows rotation electron diffraction data collection from submicrometre three-dimensional protein crystals
Igor Nederlof, Eric van Genderen, Yao-Wang Li … (2013)
1. Introduction Protein crystallography is a major justification for large-scale X-ray facilities such as synchrotrons and free-electron lasers. However, three-dimensional protein crystals that are smaller than about 0.5 µm are too small for standard X-ray crystallography, although XFEL sources are expanding the method towards smaller crystals (Chapman et al., 2011 ▶). This is a serious bottleneck, as about 30% of proteins that crystallize do not grow crystals of sufficient size or quality for X-ray structure determination (Rupp, 2004 ▶). In particular, membrane proteins and large (dynamic) protein–nucleic acid complexes fail to grow into crystals of sufficient size. Structural information on these important drug targets is therefore severely lacking. Electrons are less damaging to proteins than X-rays by several orders of magnitude per elastically diffracted quantum (Henderson, 1995 ▶). This property of electrons explains the successes of two-dimensional electron crystallography. For instance, 45 Å thick submicrometre patches of two-dimensional bacteriorhodopsin crystals yielded images with a resolution of 2.8 Å (Baldwin et al., 1988 ▶). Three-dimensional crystals with an equivalent volume would measure approximately 150 × 150 × 150 nm. Recent data demonstrate that useful high-resolution electron diffraction data (up to 2.5 Å resolution) can be obtained from nanosized three-dimensional protein crystals, where synchrotron X-rays fail (Jiang et al., 2009 ▶). However, only single diffraction shots could be collected. Collection of rotation data from protein nanocrystals was not possible because the signal-to-noise ratio and dynamic range of CCD detectors and image plates was insufficient. Very recently, direct electron detectors have become available which have a better signal-to-noise ratio and which may be better suited. These new detectors are very expensive and are probably not sufficiently radiation-hard to be routinely exposed to the direct electron beam. Electrons can also be detected by quantum area detectors such as the Medipix2 (Llopart Cudié et al., 2002 ▶; Faruqi & Henderson, 2007 ▶). The Medipix2 detector is more radiation-hard than other direct electron detectors such as the Falcon because its read-out electronic circuitry (which is sensitive to radiation damage and interference) is shielded by a semiconductor sensor layer, to which it is bump-bonded (Llopart Cudié et al., 2002 ▶). The detector has a very high signal-to-noise ratio because each pixel has its own readout electronics that measures the hole-charges that are produced by an incident electron hitting the sensor layer within 10−5 s. If the integrated energy after amplification is above a set threshold corresponding to the energy of a 200 keV electron the incident quantum is counted as a ‘hit’. Thus, the Medipix2 chip only counts 200 keV electrons and, unlike many other detectors, is blind to soft X-rays of lower energy that are also produced in great abundance inside any electron microscope. In this fashion its noise is almost exclusively determined by the counting statistics of the electrons. This gives a significant improvement over conventional CCD cameras in electron microscopes (Faruqi & McMullan, 2011 ▶; Georgieva et al., 2011 ▶). We recently showed that a Medipix2 detector with a 500 µm sensor layer can detect 200 keV electrons with a signal-to-noise ratio that is at least an order of magnitude better than that of image plate (Georgieva et al., 2011 ▶). The Medipix detector is highly sensitive at low count rates, allowing accurate measurement of the high-resolution terms. In addition, with a 500 µm sensor layer it is sufficiently radiation-hard to routinely be exposed to a direct beam of 200 keV electrons. It also has a high dynamic range, allowing accurate measurement of the intense highly peaked dose that it receives in the low-resolution diffraction spots. At 200 keV, the point spread of the detector is increased, but if the point spread is not much higher than the spread of the Bragg spots this would not matter for collecting diffraction data. In view of these desirable characteristics of the Medipix2 detector, we investigated whether it would be possible to reduce the electron dose per diffraction frame by an order of magnitude and still measure high-resolution diffraction data, as this would allow the collection of rotation diffraction data from single protein crystals for the first time. 2. Materials and methods 2.1. Crystallization Crystallization experiments were carried out using the sitting-drop vapour-diffusion technique in Innovadyne SD-2 plates. The Rock Maker software (Formulatrix) was used to design the experiments. A Genesis (Tecan) robot was used to dispense the screening solutions in the reservoirs of MRC2 plates (Swissci). An Oryx 6 (Douglas Instruments) crystallization robot was used to transfer 500 nl reservoir solution and 500 nl protein solution into sitting-drop wells. Plates were stored at 291 K and imaged using a Rock Imager automated imaging system (Formulatrix). Lysozyme (8 mg ml−1) formed needle-shaped crystals after 48 h when mixed in a 1:1 ratio with well solution consisting of 0.1 M sodium acetate pH 3.8, 1.0 M potassium nitrate (Fig. 1 ▶). 2.2. Vitrification Protein crystals were vitrified using a Vitrobot (FEI). 3 µl well solution was mixed with the drop containing nanocrystals and transferred to a 3 mm glow-discharged holey carbon grid (AGAR). Excess liquid was blotted away (blot time 3 s, blot force 5) and the sample was plunge-frozen in liquid ethane cooled by liquid nitrogen. 2.3. Rotation electron diffraction Diffraction data were collected at 200 keV on a CM200FEG (Philips) transmission electron microscope at the National Center for High Resolution Electron Microscopy in Delft. Samples were cooled to 93 K in liquid nitrogen in an in-house-modified cryo-transfer holder (Gatan). Diffraction patterns were collected using a Medipix2 detector. We created a narrow highly parallel electron beam with limited intensity by using a small (10 µm) condenser aperture and spot size 11 (which controls the first condensor lens). The beam was hardly convergent, as indicated by the data analysis described below (<0.4°). An automated compu-stage allowed crystal alignment and rotation. 2.4. Measuring electrons using a Medipix2 detector Diffraction patterns were collected with a CMOS Medipix2 detector mounted on a Nikhef carrier board (Fig. 2 ▶) and the data were transferred to a Windows PC using USB1.1 read-out electronics (Vykydal et al., 2006 ▶). Four abutting Medipix2 ASICs, each with 256 × 256 pixels and a pixel size of 55 µm, make a Medipix2 Quad. The Quad was covered by a single custom-made 500 µm semiconductor sensor chip. The distance between two neighbouring single Medipix2 ASICs is approximately 250 µm, so the edge pixels are about 125 µm wide. Therefore, they will capture more signal than the other pixels, resulting in a bright cross that quarters the raw images. This cross can be corrected for, as discussed below. The detector has a large dynamic range, where each pixel consists of a separate 14-bit pseudo-random counter. The test circuitry and four-bit trimming system are able to compensate for most fabrication variations and therefore the overall global threshold has a variation of 95 eV. The electronics noise can account for another 110 eV. These two values combined result in a dynamic range of no lower then 900 eV between channels. In silicon ∼3.6 eV is required to produce one electron–hole pair; therefore, a 200 keV electron can produce ∼55 000 pairs. This means that the dynamic range of the electronic noise accounts for less than 1.6% of the total deposited energy per electron incident and even less when there is more than one electron incident per clock cycle (McMullan et al., 2007 ▶; Plackett et al., 2010 ▶). The thickness of the sensor layer (500 µm) is larger than for stock Medipix2 chips (300 µm). We selected this larger thickness to prevent the 200 keV electrons from penetrating through the sensor layer and damaging the underlying electronics. Monte Carlo simulations (300 µm Si sensor layer; 120, 200 and 300 keV) predicting such events have been described previously (Faruqi & Henderson, 2007 ▶; McMullan et al., 2009 ▶; Turecek et al., 2011 ▶). Most importantly, they also describe events where these higher energy electrons scatter in the silicon layer to neighbouring pixels. As long as this scattering stays close to the boundaries of the Bragg spots, this would give no major point-spread problems. There is a need to set the threshold settings such that the gain of the detector is close to or slightly lower than one. Setting the threshold too low will result in multiple pixels recording a single high-energy electron hit; setting the threshold too high will not count electrons or (at sub-optimally high thresholds) will only count electrons that deposit all of their energy into a single pixel. It is important that the threshold is chosen in such a way that when the electron partly scatters to a neighbouring pixel it still is counted. The chance of a single high-energy electron scattering in the sensor layer to neighbouring pixels increases with higher energies (Faruqi & McMullan, 2011 ▶). To obtain a higher electron-counting gain, a compromise threshold level was chosen where lower energy particles would also be counted in each pixel. This, however, is not of much influence in electron-diffraction studies with the Medipix, since most other particles (X-rays) will stay well below an equivalent 100 keV threshold setting. This would be an undesirable setting in any case, because it would enable double counting in neighbouring pixels for a single-electron hit. To minimize radiation damage the experiment was conducted under such low-dose conditions that the chance of a two-electron hit within the same pixel within a clock cycle of the camera is almost non-existent (e.g. less then 0.1%). We validated our gain estimate with MOSFLM. The spot-profile statistics predicted a gain of 0.816 counts per electron. However, since the measurement was performed at very low dose conditions (the average count per pixel per frame is around five) it is no longer possible to estimate the gain by fitting a Gaussian (the MOSFLM method) instead of a Poisson distribution. Therefore, the true gain must be higher than 0.816 but not higher than 1.00. The vacuum pod in which the Medipix2 was mounted on the on-axis port of the CM-200 FEG electron microscope is shown in Fig. 3 ▶. It houses both the carrier and the USB1.1 readout board. Damage to the electronics of these boards by the direct electron beam was prevented by the small window of the pod, which only allowed illumination of about 80% of the total area of the Medipix2. Out-gassing of the electronics at 1 × 10−6 Pa vacuum did not prove to be a problem for either the electronics or the TEM. The bias voltage on the Medipix2 was set to 100 V to be able to count the holes after the electron incidents (Georgieva et al., 2011 ▶). The four chips are set to nonparallel readout by the Pixelman v.2.0 software (Turecek et al., 2011 ▶) on a PC with Windows 7 in Java mode. The Medipix2 Quad was calibrated using the standard tools that are included in Pixelman. (i) DACs were scanned to determine the noise edge of the different ASICs. (ii) Threshold equalization was performed to remove small standard discrepancies between individual pixels (3 bits). This procedure adjusts each pixel so that its threshold is as close as possible to the average and was performed with the standard settings supplied with the Pixelman package. (iii) Each lower threshold (THL) was set to a value close to the noise edge in order to acquire as much signal as possible from the incident electrons. (iv) Flat-field images obtained by illumination with a uniform beam allowed equalization of the four chips. 2.5. Preparing diffraction patterns for data processing The Medipix2 images need some pre-processing before they can be read into MOSFLM. Three problems needed to be addressed: (i) the cross needed to be removed, (ii) dead pixels needed to be corrected and (iii) the image needed to be centred, as there was a substantial drift of the direct beam. With this aim, we wrote a program that solved these issues as described below. 2.5.1. Removing the cross A bright cross appears in the raw Medipix2 Quad images as it is a tiled assembly of four single Medipix2 ASICs. The edge pixels of each single Medipix2 ASIC that abuts another in the Quad assembly are larger than the other pixels (they are 125 µm wide rather than 55 µm). Therefore, the pixels at the interface between single Medipix2 ASICs capture more electrons. This results in spatial distortion and non-uniformity of the measured signal. In order to correct for the spatial distortion, our program shifts pixels with an x value smaller than 256 by one pixel to the left and shifts pixels with a value of 256 or higher by one pixel to the right. It applies vertical shifts in a similar fashion. This procedure duplicates the pixels adjacent to the horizontal and vertical bisecting lines (and quadruples the four centre pixels). Their value was divided by 2.3 (or 5.3 for the quadrupled centre pixels) to correct the increased measured intensities of these pixels. 2.5.2. Correcting bad pixels Despite tuning the pixels of the Medipix2 detector, it can still have dead pixels (or bright pixels). Since the positions of dead pixels are fixed and their values do not follow a Poisson distribution over time, it is possible to identify these dead pixels. For each pixel position, we calculated its variance over a range of images and compared this variance with the variances at the other positions. ‘Good’ pixels will have a much higher variance than ‘bad’ pixels since the values of the latter tend to be relatively constant and largely independent of the signal, although an occasional bad pixel may be hypervariable with a large variance. If the variance value was below a set threshold the pixel was labelled as ‘bad’. After having identified the bad pixels, every image was corrected. For every bad pixel in every image, its value will be replaced by the mean value of the good pixels that surround it. If a bad pixel had no good neighbours the procedure was iterated after reclassification of the pixels. 2.5.3. Centring the images The position of the beam centre shifted as a function of the tilt of the sample holder. The shift was not huge (15 pixels maximally), but was sufficiently large to interfere with data processing. Images were centred by calculating their cross-correlations with a two-dimensional Gaussian function peaking at pixel (256, 256) and applying the calculated shift. After centring the images, they were written to disk in the CCP4 .pck format (Abrahams, 1993 ▶) as 1200 × 1200 pixel images so that they could be processed as small MAR images by MOSFLM (Leslie & Powell, 2007 ▶). 3. Results 3.1. Collecting diffraction data Prior to installing the Medipix2 detector in the electron microscope, we collected diffraction data on an image plate, film and CCD, but with these detectors we were never able to collect multiple diffraction patterns from single protein crystals to high resolution because of beam damage (Jiang et al., 2009 ▶). The Medipix2 detector has an efficiency at low count rates that is at least an order of magnitude higher than that of an image plate (Faruqi & McMullan, 2011 ▶; Georgieva et al., 2011 ▶). This meant that for the first time we could collect rotation data over multiple images without severe beam damage. Unfortunately, the hardware prevented us from interfacing the detector with the electron illumination and the rotation of the sample holder. This meant in practice that we had to start rotating the crystal slowly, switch on the beam and start collecting images until the crystal had sustained too much damage. Because transferring a single Medipix2 frame into the computer takes 0.7 s (we were using a USB1 interface), this meant that we could only collect rotation data with angular gaps between adjacent images. By using fine-ϕ slicing, we reduced the systematic errors that were introduced by this unfortunate but in the circumstances inevitable experimental flaw. We collected rotation data from many crystals, improving the data-collection strategies and sample handling iteratively. Here, we describe representative results using two rotation data sets from the same single crystal but collected at different positions using a fresh part of the crystal. We rotated in the positive direction for the first wedge of data and in the negative direction for the second wedge. Both wedges started at the same goniometer setting, but processing revealed their orientations to be about 1.5° apart. The first wedge was measured with a rotation speed of 0.083° per image (0.05° s−1), reading out an image every second. This resulted in a data set in which each image had recorded about 0.050° of rotation data, with a gap of 0.033° until the next image. For the second data set the crystal was rotated at a speed of 0.2° s−1, reading out an image every second. This resulted in a data set with 0.15° per image and gaps between images of 0.11°. Both data sets had diffraction spots to a resolution of 1.8 Å in the early images (see Fig. 4 ▶). The diffraction spots were analysed and the profiles of two single spots are shown in Fig. 5 ▶. Low-intensity spots (15–20 electrons in the peak) that are far from the central beam clearly show up above the noise (Fig. 5 ▶ a). These spots still show a significant level over the surrounding background pixels, which on average count 2–3 electrons. For comparison, a bright spot with a maximum intensity of 100–120 electrons per pixel is shown in Fig. 5 ▶(b). 3.2. Processing the diffraction data with MOSFLM Several essential parameters had to be extracted from the diffraction patterns: the angle of the rotation axis relative to the detector coordinate frame, the beam centre, the beam divergence, the unit-cell parameters and the orientation of the crystal. The determination of each of these parameters will be discussed in more detail below. 3.2.1. Beam centre We observed the beam centre to shift considerably (about 15 pixels) between the starting and ending rotation angles. Since we did not use a backstop, this shift could readily be corrected after data collection by shifting back the images accordingly (as described in §2). At the time of data collection it was unclear what the cause of the beam shift was, but it later transpired that a screw of the EM stage close to the crystals had become slightly magnetized. 3.2.2. Angle of the rotation axis As the electrons spiral their way through the magnetic lenses of the microscope, the rotation axis that is observed on the detector is not necessarily an orthogonal projection of the physical rotation axis. It needs to be calibrated for every camera distance. Careful analysis of the diffraction patterns using the human eye, a ruler and a protractor indicated that at a virtual camera length of 700 mm and an electron energy of 200 kV the angle between the rotation axis and the x direction of the detector was around 115°. We could refine this initial estimate to 116.5° by overlaying predicted diffraction patterns on observed diffraction patterns once we had good estimates of the other parameters. When rotating in the negative direction, the rotation axis was redefined to be −63.5°. 3.2.3. Unit-cell parameters Using our knowledge of the virtual crystal-to-film distance and the wavelength (which is 0.0251 Å for electrons at 200 keV energy), we interactively estimated the two spacings in the plane of the detector using the ‘measure’ facility in MOSFLM. These parameters (34.2 and 25.5 Å with an angle of 90°) were consistently found in both rotation ranges. No low-index spacings of the third axis were present in the images, and its magnitude could only be estimated from the location of the lunes in the second data set. A reasonable, but by no means perfect, overlay between observed and predicted spot positions could be obtained with a unit cell of a = 125, b = 34.2, c = 25.5 Å, with all cell angles 90° (Fig. 4 ▶). In the past we found orthorhombic nanocrystals of lysozyme to have P212121 symmetry, with a unit cell of around a = 31.5, b = 52.5, c = 89 Å (Jiang et al., 2009 ▶). However, we could not index the rotation data with the latter unit-cell parameters. 3.2.4. Determining the orientation of the crystal Owing to radiation damage, we could only measure small wedges of data. Auto-indexing routines could not cope with the data. Therefore, we indexed by eye and let MOSFLM refine the orientation using post-refinement, which, despite the angular gaps between images, managed to improve the fit between the observed and predicted patterns. The wedges contained sufficient data for orientation refinement, but did not allow refinement of the unit-cell parameters. 3.2.5. Beam divergence Once we had satisfactory predictions of the diffraction patterns, we adjusted the beam divergence (or mosaic spread; the two phenomena cannot easily be disentangled) in MOSFLM so that all spots were covered by the predicted pattern. The combined beam divergence/mosaic spread was about 0.3–0.4°. Typical profiles of the spots are shown in Fig. 6 ▶. There were a number of bad spots, mainly because of a high gradient in the background, but by setting the maximum gradient to five (instead of the default value of three) all spots were accepted. One likely cause of the high background gradient for some of the spots was the very intense zero-order reflection. Removing this reflection with a backstop was not possible without also removing all of the reflections with a resolution lower than about 5 Å. The reason for this is the low scattering angle and the position of the backstop in the CM-200 microscope. For protein crystallographic data collection the position of the backstop should have been much closer to the detector, but hardware restrictions prevented us from moving the backstop. Scaling the data was problematic in view of the high partiality, the small wedge size, the scarcity of data and the angular gaps between the frames. For this purpose, we used the CCP4 program SCALA (Evans, 2006 ▶). The statistics were poor because we had to scale the partials using their calculated partiality and because of the large missing gaps of data between adjacent images. Fig. 7 ▶ shows the merged and scaled h = 0 zones of both data sets. Note that we processed all of the data as P1 with anomalous signal. We did so in order to identify potential dynamic scattering, which results in a breakdown of point-group symmetry in the diffraction patterns (Abrahams, 2010 ▶). The symmetries shown in Fig. 7 ▶ are therefore intrinsic to the diffraction experiment and not imposed by symmetry averaging of corresponding reflections. 4. Discussion and conclusion Three-dimensional nanocrystals of lysozyme have successfully been grown, vitrified and transferred to an electron microscope at liquid-nitrogen temperature. Rotation electron-diffraction data were obtained at multiple positions of a single submicrometre crystal to resolutions better than 2 Å. For the first time, we could collect multiple frames from protein nanocrystals. This was made possible by the much higher sensitivity and signal-to-noise ratio of the Medipix2 detector, combined with its radiation hardness at 200 keV. By using a parallel electron beam, we could observe the curvature of the Ewald sphere in electron-diffraction patterns of protein crystals for the first time. At 200 keV, electrons have a wavelength only about 0.025 Å, causing the Ewald sphere to be extremely flat at low resolution. Yet the data show that the curvature is still significant. When the curvature of the Ewald sphere cannot be ignored, both members of a Friedel pair cannot simultaneously be in full diffraction. Geometric considerations imply that the rotation angle between the two reflections of a Friedel pair is always at least their diffraction angle. Consider, for instance, a reflection at 5 Å resolution and assume that it is optimally aligned. At λ = 0.025 Å the crystal would need to be rotated by almost sin−1(0.025/10) = ∼0.3° in order to move its Friedel mate into full diffraction. The fact that we can clearly see deviations in intensity between Friedel mates even at 5 Å resolution (Fig. 4 ▶) confirms the low beam divergence and mosaic spread suggested by the data processing. Clearly, the rocking curve of the crystal (with a maximal width of less than 0.4°) is sufficiently peaked to allow the observation of these fine angular differences in orientation. This implies that it should be possible to pinpoint areas of low mosaicity in the crystals. This may have an advantage if the crystalline order varies within a single nanocrystal, in which case its best parts can be selected. These results are encouraging enough to consider fixing the technical problems that prevent optimal data collection. The lack of a scheduling interface between the detector and the goniometer that controls the crystal orientation, combined with the relatively slow readout of the Medipix2 Quad (0.7 s), is an evident cause of many of these problems. As the rotation of the sample could not be interrupted during the readout time, there were angular gaps between subsequent frames. We anticipate fixing this problem by employing a novel Medipix2 detector on a Relaxd read-out board (Visser et al., 2011 ▶) with improved readout time (up to 200 frames per second) that can also read data while collecting the next frame. This would significantly decrease the illumination time and subsequent beam damage. We plan to mount the detector on a Titan Krios machine, so data collection will be improved owing to the presence of an autoloader, a programmable goniometer and a programmable beam blanker. For small-molecule crystals, it has been reported that the collection of data in STEM diffraction mode further reduces beam damage (Kolb et al., 2011 ▶), so we plan to incorporate the Medipix2 detector in a microscope that has this option. In view of the suboptimal data collection, we were happy being able to integrate the data with one of the standard programs for processing X-ray data: MOSFLM. However, we cannot exclude the possibility that the results that we report here do not reflect the true crystal structure. We noticed unexpected correlations between parameters. For instance, rotation of the crystal around ϕ was accompanied by an apparent rotation (as suggested by post-refinement) around ΦX, which is a rotation of the crystal around the beam. This may perhaps be caused by eucentric height variation of the crystal as it is being rotated. Since the electrons are focused spirally, rotation of the image or diffraction pattern with height is expected. These and other correlations, which were independent of the unit-cell or crystal-orientation parameters, may indicate that the indexing solution that we present here is wrong. However, we think that these correlations are caused at least in part by distortions specific to electron diffraction. These exist and we have unequivocally identified at least one. For instance, the correlation of the beam shift with rotation of the crystal, which we have observed (and corrected for), would never be encountered in X-ray diffraction. Together with beam damage, the angular gaps between adjacent images may explain the observed breakdown in point symmetry observed in the h = 0 Laue zones (Fig. 7 ▶). An alternative explanation may be found in the effects of dynamic scattering, which also causes such a breakdown in symmetry (Abrahams, 2010 ▶). With improved hardware we anticipate being able to distinguish between these effects. We have demonstrated that it is possible, at least in principle, to collect rotation electron-diffraction data from protein nanocrystals. This prompts the question concerning the usefulness of such data sets in solving protein crystal structures. Problems with dynamic scattering have been anticipated, yet these did not prevent structure solution of the aquareovirus to 3.3 Å resolution by single-particle analysis (Zhang et al., 2010 ▶) using crystals that also had a size of approximately 100 nm, just like those analyzed here. If at higher resolution the effects of dynamical scattering can no longer be ignored, they can be modelled by the multi-slice approach (Jansen et al., 1998 ▶). This is common practice in electron crystallography of small molecules. Furthermore, there are theoretical considerations that imply that dynamical scattering may yield phase information (Abrahams, 2010 ▶) and a limited amount of such dynamical scatter may therefore even be beneficial. Electron area detectors such as Medipix that allow the distinction of the signal of high-energy electrons from that of other types of radiation are likely to have a major impact on structural biology. Our results imply that protein crystallography also stands to benefit from these technical advances.
Author and article information
Journal
Journal ID (nlm-ta): eLife
Journal ID (iso-abbrev): Elife
Journal ID (hwp): eLife
Journal ID (publisher-id): eLife
Title:
eLife
Publisher:
eLife Sciences Publications, Ltd
ISSN
(Electronic):
2050-084X
Publication date
(Electronic):
19
November
2013
Publication date Collection: 2013
Volume: 2
Electronic Location Identifier: e01662
Affiliations
[1
]
Oliver B Zeldin is in the Department of Molecular and Cellular Physiology and the Howard Hughes Medical Institute, Stanford University , Stanford, United States
zeldin@
123456stanford.edu
[2
]
Axel T Brunger is in the Departments of Molecular and Cellular Physiology, Neurology and Neurological Sciences,
Structural Biology, Photon Science and the Howard Hughes Medical Institute, Stanford University , Stanford, United States
brunger@
123456stanford.edu
Article
Publisher ID: 01662DOI: 10.7554/eLife.01662
PMC ID: 3831941
SO-VID: 4f915a24-a6d0-4dba-a277-b20f4f3ad86c
Copyright © Copyright © 2013, Zeldin and Brunger
License:
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
History
Related
Categories
Subject:
Insight
Subject:
Biochemistry
Subject:
Biophysics and Structural Biology
Subject:
Protein Crystallography
ScienceOpen disciplines: Life sciences
Data availability:
ScienceOpen disciplines: Life sciences