A Comparison of Liquid Nitrogen and Liquid Helium As Cryogens for Electron Cryotomography

Journal of Structural Biology 153 (2006) 231–240 www.elsevier.com/locate/yjsbi

A comparison of liquid nitrogen and liquid helium as cryogens for electron cryotomography

Cristina V. Iancu, Elizabeth R. Wright, J. Bernard Heymann 1, Grant J. Jensen ¤

Division of Biology, California Institute of Technology, 1200 E. California Blvd., Pasadena, CA 91125, USA

Received 29 June 2005; received in revised form 16 November 2005; accepted 7 December 2005 Available online 4 January 2006

Abstract

The principal resolution limitation in electron cryomicroscopy of frozen-hydrated biological samples is radiation damage. It has long been hoped that cooling such samples to just a few kelvins with liquid helium would slow this damage and allow statistically better- deWned images to be recorded. A new “G2 Polara” microscope from FEI Company was used to image various biological samples cooled by either liquid nitrogen or liquid helium to »82 or »12 K, respectively, and the results were compared with particular interest in the doses (10–200 e¡/Å2) and resolutions (3–8 nm) typical for electron cryotomography. Simple dose series revealed a gradual loss of contrast at »12 K through the Wrst several tens of e¡/Å2, after which small bubbles appeared. Single particle reconstructions from each image in a dose series showed no diVerence in the preservation of medium-resolution (3–5 nm) structural detail at the two temperatures. Tomo- graphic reconstructions produced with total doses between 10 and 350 e¡/Å2 showed better results at »82 K than »12 K for every dose tested. Thus disappointingly, cooling with liquid helium is actually disadvantageous for cryotomography.  2006 Elsevier Inc. All rights reserved.

Keywords: Electron cryomicroscopy; Tomography; Helium cooling; Radiation damage; CryoEM

1. Introduction specimens to just a few kelvins with liquid helium, but early studies were highly variable and “unable to Wnd deWnite Biological materials can be imaged in transmission elec- evidence that there is an improvement in radiation resis- tron microscopes in a life-like, “frozen-hydrated” state tance on going from liquid nitrogen to liquid helium” through the use of specialized cryostages that keep samples (International Study Group, 1986). More recently, how- frozen while they are inside the microscope column. For ever, Stark et al. found that the so-called “cryoprotection these samples, radiation damage is the principal resolution factor” at 4 K (under liquid helium cooling) was 1.4 and 2.5 limitation, far exceeding others such as electron optical per- times better than at 98 K (under liquid nitrogen cooling) for formance. The Wrst cryostages cooled samples to »90 K 7 and 3 Å spacings, respectively, in two-dimensional protein through thermal contact with liquid nitrogen. After it was crystals (Stark et al., 1996). Part of the ambiguity in past observed that radiation damage proceeded much more work stemmed from the fact that previous microscopes slowly at low temperature, it was hoped that additional used either liquid nitrogen or liquid helium exclusively, so dose resilience might be realized through further cooling that comparisons had to be made between diVerent micro- (Glaeser, 1971). New microscopes were engineered to cool scopes in diVerent labs or settings. All previous reports have focused on the doses (»10 e¡/ Å2 or less) and resolutions (»7 Å or better) of interest to * Corresponding author. Fax: +1 626 395 5730. electron crystallography, and used the same basic measure- E-mail address: [email protected] (G.J. Jensen). ment—the fading of crystal diVraction patterns. We wish to 1 Present address: Laboratory of Structural Biology Research, National Institute of Arthritis, Musculoskeletal and Skin Diseases, National Insti- emphasize that the question of whether liquid helium cool- tutes of Health, Bethesda, MD 20892, USA. ing is actually advantageous or not depends, of course, on

1047-8477/$ - see front matter  2006 Elsevier Inc. All rights reserved. doi:10.1016/j.jsb.2005.12.004 232 C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240 the type of sample and goals of the project. For two-dimen- The temperature of the sample when liquid helium is pres- sional crystallographic studies probing near-atomic resolu- ent in the inner dewar was estimated by mounting a silicon tion, for instance, specimen atoms must remain in their diode thermometer on a special cartridge while the column original locations, and doses of just a few e¡/Å2 are typi- was open in the factory. For our microscope, that tempera- cally used. For single particle analysis studies seeking to ture was measured to be 11.5 K. During operation, a ther- resolve secondary structure, structures the size of helices mocouple provides real-time measurements of the and sheets must remain largely intact, and doses of »10– temperature of the “cryobox” immediately surrounding 20 e¡/Å2 are standard. Finally, for tomographic studies and in thermal contact with the cartridge. When the inner seeking to visualize shapes of individual proteins, radiolytic dewar is Wlled with liquid helium, this reads 10 K. When fragments of protein domains must remain only approxi- cooled with liquid nitrogen, it reads 82 K. mately in place, and doses of »20 to 200 e¡/Å2 are common. To begin our studies, simple dose series of several fro- Furthermore, while lower temperatures may or may not zen-hydrated biological samples were recorded at the two help reach any of these goals, potential disadvantages such temperatures, and the radiation damage trajectories were as increased charging or drift must also be considered. Thus observed. The Wrst sample tested was intact, frozen- successes or failures with liquid helium cooling in one con- hydrated cells of a small bacterium, Mesoplasma Xorum. text are not necessarily predictive in others. Cells were plunge frozen, inserted into the microscope, and Here, our focus is tomography. We compare liquid cooled to »82 K with liquid nitrogen. A suitable cell was nitrogen and liquid helium cooling with methods that are centered under the beam with less than 1 e¡/Å2 dose, and diVerent from all previous work in the nature of the sample, then approximately 70 exposures were recorded with 10 e¡/ the doses used, the type of data collected, the measures of Å2 each. Several key images are shown in the Wrst column quality, and the degree of control. Whole cells and individ- of Fig. 1. Next the grid was retracted into the liquid nitro- ual, large protein complexes are imaged through doses of gen cooled “multispecimen transfer system,” which resides 10–350 e¡/Å2. Image quality is measured both qualitatively within the column vacuum, and the liquid nitrogen in the and quantitatively in the resolution range of 3–6 nm. The inner dewar was replaced with liquid helium. After the tem- instrument used was one of the new series of “Polara” perature of the sample area settled to »12 K, the grid was microscopes from FEI, which allows liquid nitrogen and re-threaded onto the stage, allowed to cool (to »12 K), and helium to be exchanged repeatedly, all while a single grid is another Me. Xorum cell several microns away from the Wrst being imaged in a single microscope, without other con- was centered under the beam. An identical dose series was founding variables. Disappointingly, in this context we Wnd recorded (Fig. 1, second column). liquid helium cooling to be disadvantageous. In the com- At »82 K, the contrast from the cell membrane and mac- panion paper, we report several additional observations romolecular complexes inside the cell appeared qualita- that help explain this result (Wright et al., companion tively similar throughout the dose series, but did spread and paper). After conWrming that vitreous ice collapses from a smear gradually. In the Wrst image, the cell was punctuated low to a high density state when irradiated at »12 K, we go with small and distinct densities, which are the projections on to show that the collapse alone is not the immediate of internal macromolecules, and the membrane appeared as problem. Instead, we speculate that the key issue is how a dark, single band. In subsequent images, the internal den- radiolytic fragments aggregate within the high density ice. sities became less distinct and slightly rearranged. Never- theless, densities the size of single large protein complexes 2. Results could easily be traced through the images (circles in Fig. 1 column 1), suggesting that their basic shapes were pre- A 300 keV “G2 Polara” transmission electron micro- served, until small “bubbles” appeared after »160 e¡/Å2 scope from the FEI Company was installed in our lab at the and the internal structures were catastrophically perturbed. California Institute of Technology in the fall of 2003. It is At »12 K, the Wrst image was qualitatively indistinguish- equipped with a new cartridge-based sample holder speciW- able from the Wrst image at »82 K, but in subsequent cally designed to allow liquid helium cooling. An external images the membrane contrast faded quickly. The mem- dewar system is permanently mounted to the side of the brane became largely invisible after »40 e¡/Å2, and then column consisting of an inner “helium” dewar surrounded split into two dark bands separated by a light interior band. by an outer “nitrogen” dewar. The inner dewar is thermally This trilaminar structure had a slightly larger total width connected to the tip of the stage, so that when a cartridge is than the original membrane when it Wrst appeared (»80 e¡/ threaded onto the stage, it is cooled to near the temperature Å2), and then expanded non-uniformly as the dose series of the cryogen in the inner dewar within just a few minutes. continued. At the end (700 e¡/Å2), large bubbles appeared The Polara design is diVerent from some previous liquid between the dark outer layers (not shown). The relative helium cooled microscopes in that there is no special contrast between the layers increased steadily with dose. requirement for helium per se—the inner dewar can be Concerning the appearance of internal macromolecular Wlled with either liquid helium or nitrogen without compli- complexes, they seemed to smear and rearrange slightly cation, and in fact the cryogen can be exchanged back with dose just as those imaged at »82 K, until again small and forth all while a single cryosample is being imaged. bubbles disrupted their structure grossly after »160 e¡/Å2. C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240 233

Fig. 1. Dose series of intact, frozen-hydrated bacterial cells. Me. Xorum cells were plunge frozen and imaged in the Polara electron cryomicroscope. Seventy images were recorded of each cell with 10 e¡/Å2/image. The Wrst and second columns show images recorded at »82 and »12 K, using liquid nitrogen and liquid helium as cryogen, respectively. One particular cluster of density is circled in each image of the Wrst column to facilitate comparison. The third and fourth columns show images recorded again at »12 K, but where the sample was brieXy warmed up in the liquid nitrogen cooled multispecimen holder, as if for rotation in the Xip-Xop rotation stage, and then re-cooled in the column to »12 K. In the third column, the sample was warmed once after a cumula- tive dose of 30 e¡/Å2. In the fourth column the sample was warmed and re-cooled after every 30 e¡/Å2. Note how the membrane contrast fades and is then replaced by bubbles at »12 K but not »82 K. This eVect is delayed by one warming cycle, and prevented indeWnitely by iterative warming cycles. (Scale bar 250 nm.) 234 C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240

Fig. 2. Dose series of liposomes. Dose series were recorded as in Fig. 1, but this time of liposomes. Pure lipids exhibit the same contrast eVects as the cell membrane, but at approximately twice the dose. While bubbles form on the carbon support at both temperatures, they are much larger and coalesce to a greater extent at »82 K than »12 K. Bubbling on the carbon is apparently prevented by iterative warming cycles. (Scale bar 200 nm.) C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240 235

Fig. 3. Dose series of puriWed protein complexes. Dose series of a puriWed protein complex, hemocyanin, are shown. At »82 K, the proteins become smeared as their Wne structure is destroyed, but their general shapes are still evident even after 300 e¡/Å2. As for lipids, at »12 K the contrast from puriWed protein fades and is eventually replaced by small bubbles. A single warming cycle delays the eVect slightly, but iterative warming cycles prevent it indeW- nitely. (Scale bar 200 nm.) 236 C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240

Unlike the cellular components imaged at »82 K, however, their contrast faded noticeably and the nature of the bubbling was diVerent in that the bubbles were more numerous but smaller, and did not coalesce as quickly. Our Polara was recently equipped with the prototype “Xip-Xop” rotation stage, which allows grids to be rotated 90° within the column vacuum to enable dual-axis cryotomography (Iancu et al., 2005). As designed, it had one major potential liability, which was that the grid is rotated in the so-called multispecimen transfer system, which can only be liquid nitrogen cooled. Thus, if a Wrst tilt-series were recorded at »12 K, before the second, orthogonal tilt-series could be recorded, the grid would have to be warmed up to »82 K for rotation and then re-cooled to »12 K. If liquid Fig. 4. Information loss as a function of dose at »82 and »12 K. From the hemocyanin dose series, 128 particles were chosen and used to produce helium cooling did in fact constrain radiation damage as independent single particle reconstructions from each image. As a control, hoped, such warming might release those constraints before an equivalent number of arbitrary positions not containing particles were the second tilt-series. To investigate this possibility, we aligned and merged, yielding the resolution labeled as “noise alone.” The recorded dose series at »12 K wherein the samples were rate of information loss at »82 and »12 K was assessed by tracking the retracted into the multispecimen holder as if for rotation resolution of the reconstructions. Two curves for each temperature are » » shown. The notable diVerences were that the Wrst image at »12 K was (and thus warmed to 82 K and re-cooled to 12 K, here- often blurred, and essentially no useful information was recovered from after referred to as a “brief warmup/cooldown cycle”) after ¡ the images at »12 K after about 150 e¡/Å2. 30 e /Å2 to mimic collection of a typical dual-axis tilt-series

Fig. 5. Tomogram quality as a function of dose at »82 and »12 K. Full tilt-series were recorded and tomographic reconstructions calculated of Welds of frozen-hydrated hemocyanin using total doses from 10 to 350 e¡/Å2 at both »82 and »12 K. From each tomogram, 100 hemocyanin particles were chosen and compared to a known higher resolution structure by cross correlation. The overall quality of each tomogram was then assessed by calculating the average cross-correlation coeYcient (plotted) and the standard deviation (shown as half error bars for clarity). As a control, an equivalent number of arbitrary positions not containing particles were also selected and analyzed (labeled as ‘noise’). Above or below each data point, inset panels show isosurface rendi- tions at 2.5 standard deviations above the mean of the best reconstructed particle after denoising from “top” and “side” views. At every dose, the results at »82 K were superior. For scale, hemocyanin is barrel-shaped, »30 nm in diameter and »35 nm long. C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240 237

(Fig. 1, column 3). Surprisingly, the radiation damage tra- multienzyme complexes manifest the same patterns in con- jectory seemed just like that seen at »12 K without any trast fading and bubbling as hemocyanin, as did hemocya- warming, except delayed by approximately 30 e¡/Å2. Spe- nin in thick ice (data not shown). ciWcally, the Wrst image after warming looked like the origi- To compare the usefulness of each image quantitatively nal, but the loss and subsequent splitting of the membrane as a function of dose, three-dimensional hemocyanin recon- contrast then followed. This prompted us to test a fourth structions were calculated from the particles found in each protocol, wherein the sample was iteratively retracted and image with standard “single particle” averaging methods, replaced (warmed to »82 K and then re-cooled) every 30 e¡/ using the known higher resolution model (Mouche et al., Å2 (Fig. 1, column 4). In this case the appearance of the 2003) of hemocyanin as the reference for alignment and res- membrane remains unchanged, and the results are like olution assessment (Fig. 4). Two dose series having the those recorded at a constant »82 K (column 1) except that same number of particles, defocus, and pixel size were ana- internal bubbling is further delayed. lyzed each for both »82 and »12 K. To assess the inXuence To investigate whether these membrane eVects were due of reference bias, an equal number of ice regions not con- to integral membrane proteins or the cellular environment, taining particles were also chosen, aligned to the known ref- we recorded similar dose series of pure liposomes (Fig. 2). erence, and merged to form a three-dimensional At »82 K, there was no visible change or bubbling in the reconstruction. This reconstruction of noise was also com- lipid itself, even to the end of the dose series (800 e¡/Å2). pared to the reference through Fourier shell correlation fol- The same loss of membrane contrast observed at »12 K lowing identical procedures, yielding an apparent was seen again here, followed by splitting into a trilaminar “resolution” of 4.3 nm. Thus, two factors contribute to res- structure, but importantly, the eVect came later in the dose olution here: the true information about particle shape series. While it is diYcult to assign speciWc numbers, the present in the images, and aligned noise. To track the pres- fading and splitting occurred at about twice the dose ervation of useful information and reduce the impact of required for the analogous eVect in the Me. Xorum cell aligned noise, the orientation angles of each particle were membrane (note the range of doses in Fig. 2 is nearly dou- estimated from the Wrst (or second in the case of »12 K) ble that shown in Fig. 1). Brief warmup/cooldown cycles image and then Wxed for the remainder of the dose series. had the same qualitative eVect as seen before in the cell. The Only particle positions were reWned in subsequent images. other interesting phenomenon in these dose series is the The results were highly reproducible, and showed no nature of the “bubbling” that occurs on the carbon sup- detectable diVerence in the cryoprotection factor for doses port. At »82 K, bubbles Wrst became visible at »50 e¡/Å2, from 10–120 e¡/Å2 in the 3.5–4.5 nm resolution range. After and then grew and coalesced. At »12 K, however, bubbling 120 e¡/Å2, however, the resolutions from the helium cooled was delayed, and the bubbles appeared much smaller and series degraded rapidly. This was expected, since the parti- remained distinct. Surprisingly, while the bubbles on the cles are essentially invisible. It was also observed that the carbon were unaVected by a single warming cycle, iterative very Wrst image recorded at »12 K was frequently blurred, warming cycles prevented visible bubbling, even through as evidenced by its anomalously poor resolution. Note that 600 e¡/Å2. images of the same 128 physical particles were used to Next we recorded analogous dose series of several large, produce each reconstruction. The experiment was repeated puriWed protein complexes including hemocyanin, the 20 S with various defocus values and pixel sizes, all with the proteasome, and the pyruvate dehydrogenase multienzyme same result. The same result was also obtained when the complex. Images from the hemocyanin dose series are orientation for each particle was taken from the image cor- shown in Fig. 3. At »82 K, the contrast from individual responding to its highest Wgure-of-merit in the alignment hemocyanin molecules gradually smeared through the Wrst routine, rather than the Wrst or second of the series. several hundred e¡/Å2, but did not fade, and the gross To complete the comparison of cryogens in the full con- shapes of the molecules remained evident through more text of tomography, tilt-series were recorded of Welds of fro- than 300 e¡/Å2 (column 1). Bubbling was only observed zen-hydrated hemocyanin with diVerent total doses where the proteins were aggregated. At »12 K, the Wrst between 10 and 350 e¡/Å2 at both temperatures. One hun- image was quite similar to the Wrst image at »82 K, but dred particles from each tomogram were selected and thereafter the contrast of the proteins gradually faded, until aligned in three dimensions to the known 12 Å structure they were virtually invisible at »200 e¡/Å2. Past 200 e¡/Å2, (Mouche et al., 2003). The average cross-correlation coeY- small bubbles appeared in a strikingly similar pattern as the cient between the model and the reconstructed particles was domains of the original proteins, and then grew with addi- plotted as a function of total dose used for the tilt-series tional dose (column 2). The pattern of contrast fading at (Fig. 5). All other variables besides temperature and total »12 K was not visibly diVerent when the sample was brieXy dose were kept as similar as possible, including defocus, warmed and cooled just once after 30 e¡/Å2 (column 3), but magniWcation, ice thickness, tilt range, and tilt step. While when the sample was taken through iterative warmup/cool- images could be aligned and produced meaningful recon- down cycles (column 4), the contrast was actually pre- structions with as few as 10 e¡/Å2, the results improved with served, resembling at each dose the series recorded at dose for both temperatures up to 120 e¡/Å2, presumably »82 K. The 20S proteasome and pyruvate dehydrogenase due to reduction of shot noise. Past 120 e¡/Å2, however, the 238 C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240 results began to deteriorate, presumably because of radia- ids and diVerent proteins, even though they were in similar tion damage. At all doses, the results at »82 K were better or in some cases identical buVers which should change than those at »12 K, and the diVerence became more pro- phases at the same rate. nounced at higher doses, probably due to lost contrast in Instead, we speculate that the diVerences in contrast and the later images at »12 K. bubbling at »82 and »12 K are due to diVerences in how water molecules and radiolytic fragments diVuse in the two 3. Discussion ice phases. Various small molecules are expected to form when frozen-hydrated organic material is irradiated, Through dose series and tomograms of various frozen- including hydroxyl radicals, hydrogen atoms, hydrogen gas, hydrated samples including intact cells, liposomes, and sev- carbon monoxide, and carbon dioxide. There are reports eral puriWed protein complexes, we have shown empirically and/or claims in the literature that the mobility of each of that cooling samples with liquid helium rather than liquid these speciWc small molecules changes signiWcantly in the nitrogen is unexpectedly disadvantageous. The most temperature range of 4–90 K (see, for example, Glaeser important observation was that at »12 K, the contrast of et al., 1971; Heide, 1982, 1984; Jenniskens and Blake, 1994; proteins and lipids faded steadily with irradiation in the Jenniskens et al., 1995; Sandford and Allamandola, Wrst several tens of e¡/Å2, and was then replaced by high- 1993a,b). We can still only speculate, however, exactly how contrast but artifactual features such as trilaminar struc- such small molecules behave in the unknown local condi- tures in the place of membranes and small bubbles in the tions surrounding a protein, for instance, as it is being place of protein domains. destroyed by irradiation, cooled by a cryogen but locally Two types of quantitative measures were used to detect heated by inelastic scattering events, and embedded in vit- any diVerence in the cryoprotection factors at »82 and reous ices of variable composition (i.e. the diVerent buVers »12 K. In the Wrst, independent single particle reconstruc- and cell culture medias were used here). Furthermore, tions of hemocyanin were calculated from individual diVerent methods and conditions of formation seem to pro- images within a dose series. While the resolutions steadily duce diVerent high density vitreous ices (Angell, 2004), so it decreased with dose as expected, no diVerence was seen in is unclear which mobility studies are applicable in the pres- the rate of this decrease at the two temperatures. Instead, ent context and which are not. Finally, surprises such as the the only signiWcant diVerences were (1) that the Wrst image Wnding that acoustic waves in room-temperature liquid recorded at »12 K was frequently blurred, and (2) that after water can produce small bubbles with local temperatures of »120 e¡/Å2, only the nitrogen cooled samples retained use- millions of kelvins (Crum and Matula, 1997) highlights ful information about the shapes of the proteins. The sec- how misguided simple intuition can be! ond quantitative measure compared full tomographic Nevertheless, two facts seem clear: (1) when covalent reconstructions of hemocyanin at diVerent total doses at bonds are broken by the beam, the newly unbonded atoms the two temperatures. At every total dose from 10 to 350 e¡/ will repel each other strongly through van der Waals and/or Å2, the results were better when liquid nitrogen was used Coulombic forces, and (2) the low density state of vitreous than liquid helium. The best reconstructions of all, as mea- ice must have larger or more intermolecular spaces than the sured by cross-correlation with the known higher-resolu- high density state (Jenniskens et al., 1995). At »82 K then, tion structure, were obtained with 120 e¡/Å2 and liquid radiolytic fragments and byproducts (like hydrogen gas) nitrogen cooling. likely diVuse away from each other gradually through these While the conclusion that liquid helium cooling was intermolecular spaces. The surrounding water and other inferior in these comparisons was clear, its explanation was fragments might rearrange slightly, as energy deposited by not. After observing the gradual loss of contrast at »12 K, the beam dissipates, and the net eVect in the image would our Wrst thought was that the ice was collapsing and simply be an apparent “smearing.” If the diVusion of small radio- “contrast-matching” the proteins and lipids. As reported in lytic fragments and byproducts is blocked, however, by the companion paper, however, extensive additional experi- some denser material like proteins in a crowded cytoplasm, ments were performed to understand the behavior of ice at a carbon support, or even just the denser vitreous ice at the two temperatures (Wright et al., companion paper). The »12 K, the accumulating repulsion of radiolytic fragments key results were that, in conWrmation of the existing litera- might eventually force open pockets of relatively lower ture (Heide, 1984), (1) vitreous ice does indeed collapse into density material, described here as “bubbles.” Thus, the a higher density phase when irradiated at »12 K, (2) this bubbles are likely to contain hydrogen gas and other small happens rapidly–within just the Wrst 2–3 e¡/Å2, and (3) the molecules. The formation and incremental growth of these high density ice spontaneously expands back to the low bubbles is likely responsible for the gradual loss of contrast density state when warmed to »82 K over a period of at seen. We also observed that when samples at »12 K are least several minutes. Thus collapse of the ice cannot alone irradiated and then brieXy warmed and cooled, as for explain the loss in contrast seen here for two reasons. First, instance is required by the Xip-Xop cryorotation holder, the rates of the two processes are diVerent by an order of bubbling is prevented and normal contrast is retained. Our magnitude (two versus several tens of e¡/Å2). Second, the interpretation is that while these cycles are not long enough loss of contrast proceeded at diVerent rates for diVerent lip- to cause global expansion of the ice into its low density C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240 239 form (Wright et al., companion paper), they are suYcient to chem) were diluted in 50 mM sodium phosphate, pH 7.2 relieve local strains like the forces that would otherwise to Wnal protein concentrations of 2, 8, and 3 mg/ml, result in bubble formation. respectively. Ten-nanometer colloidal gold (Ted Pella) was combined 4. Materials and methods with the samples and also applied to the grids separately. Protein solutions and cell suspensions were Xash frozen 4.1. Dose calibration onto R1.2/1.3 carbon grids (Quantifoil) in liquid ethane using a Vitrobot (FEI, The Netherlands) and the following All reported doses were measured using the 2048 £ 2048 typical conditions: 100% humidity, 23 °C, 4 l/grid if manu- pixel CCD camera of our Gatan (Pleasanton, CA) Imaging ally applied, 2.0–3.5 s blot time, ¡2 to ¡3 mm blot oVset. Filter. The number of counts reported by this camera per Grids with ice thickness and gold distribution suitable for primary electron was exhaustively calibrated using four data collection were reproducibly generated using these independent ammeters: an in-house Faraday cup; the conditions. GIF’s drift tube; the small, insertable Xuorescent screen; and the large Xuorescent viewing screen. First, an in-house 4.3. Dose series protocols Faraday cup was installed at the level of the viewing screen and connected to an external ammeter in such as way that a All images were energy Wltered (20 eV slit width) and condensed beam could be deXected either to an exposed recorded on the 2048 £ 2048 CCD of the Polara TEM’s part of the viewing screen or into the Faraday cup. The cur- GIF. Successive images were collected at 30 s intervals; at a rent to either of the two Xuorescent screens is reported by dose/image of 5 e¡/Å2 (hemocyanin) or 10 e¡/Å2 (Me. the FluScreenMgr panel of the Feispy application based on Xorum; liposomes; mixture of hemocyanin, 20S proteasome factory calibrations. The measurements from the two and pyruvate dehydrogenase); at 6.7 Å/pixel; using defocus screens were in all cases essentially identical, and these were values of ¡6 m (hemocyanin), ¡8 m (liposomes), or found to agree well with measurements taken with the Far- ¡15 m (Me. Xorum) (Wrst zeroes of the contrast transfer aday cup, varying predictably by up to 30% (the worst case) function at 1/3.4, 1/4.0, and 1/5.4 nm¡1, respectively); either across two orders of magnitude in intensity (0.034– at »82 or »12 K. Typically, the areas imaged at the two 1.980 nA) and at voltages of 120, 200, and 300 keV. After temperatures came from the same grid square and were the Faraday cup was removed, beam currents were again separated by »10 m to ensure similarity in ice thickness, measured by deXecting the entire beam into the GIF drift sample, and gold distribution. tube and recording its output current with an external ammeter. These measurements were all »20% higher than 4.4. Single-particle reconstructions those from the viewing screens and the Faraday cup. The Faraday cup was judged to be the most accurate of the four For Fig. 4 the hemocyanin dose series were taken with methods, and was used to calibrate the sensitivity of the 10 e¡/Å2, a defocus of ¡4 m, and a pixel size of 6.7 Å. The GIF CCD camera as producing 7 counts/primary electron. experiment was repeated with a diVerent defocus value This was done by condensing the beam entirely within the (¡6 m), and a third time with a diVerent pixel size (9.8 Å), CCD, exposing for exactly 1 s, integrating the total counts all with the same result. All processing steps were carried reported on the CCD, and dividing by the total dose out using the Bsoft (Heymann, 2001) and Peach (Leong applied. The sensitivity of the CCD was conWrmed to be et al., 2005) packages. For each hemocyanin dose series linear throughout the range used in this study. analyzed, the images were mutually aligned, and then a set of 128 particles within the Quantifoil hole was selected and 4.2. Samples boxed out separately from each image in the series. The position and orientation of each particle were found by Mesoplasma Xorum was obtained from the ATCC comparison to projections of the published structure (Mou- (strain #33453) and cultured in Mycoplasma medium che et al., 2003). Independent, three-dimensional recon- (ATCC medium #243) with the addition of phenol red as structions for each image in the series were then calculated a growth indicator. Cells were allowed to reach late expo- by weighted back-projection using reWned positions for nential phase at which time they were centrifuged at each particle image. Particle orientations for the data at 10,000g for 3 min and resuspended in growth media. Lip- »12 K were Wxed at the values obtained from the second osomes were formed from either pure 1,2-dipalmitoyl-sn- image in the series, since the Wrst image was typically glycero-3-phosphocholine (DPPC) or a 1:1 mixture of smeared. The resolution of each reconstruction was DPPC and 1,2-dioleoyl-sn-glycero-phosphoethanol- assessed by the Fourier shell correlation method using a amine-N-[4-(p-maleimidophenyl) butyramide] (MPB PE) threshold of 0.5. An equal number of arbitrary positions by extrusion through 0.2 m Wlters. PuriWed Escherichia not containing particles were treated identically to produce coli pyruvate dehydrogenase complex, Megathura crenu- an “aligned-noise-only” reconstruction as a control to lata Keyhole Limpet hemocyanin (A.G. ScientiWc), and assess reference bias. The resolution of the aligned-noise- Methanosarcina thermophila 20S proteasome (Calbio- only reconstructions varied consistently with the number of 240 C.V. Iancu et al. / Journal of Structural Biology 153 (2006) 231–240 noise “particles” used as expected. For 128 particles, the References “resolution” was 4.25 nm. Angell, C.A., 2004. Amorphous water. Annu. Rev. Phys. Chem. 55, 559–583. 4.5. Hemocyanin tomograms Crum, L.A., Matula, T.J., 1997. Sonoluminescence: shocking revelations. Science 276 (5317), 1348–1349. Frangakis, A.S., Hegerl, R., 2001. Noise reduction in electron tomographic Energy-Wltered (20 eV slit width) image tilt-series with reconstructions using nonlinear anisotropic diVusion. J. Struct. Biol. total doses of 10, 25, 50, 80, 120, 200, and 350e¡/Å2 were 135 (3), 239–250. recorded from ¡69° to 69° with 3° step sizes, a defocus of Glaeser, R.M., 1971. Limitations to signiWcant information in biological ¡6m (Wrst zero of CTF at 1/3.4 nm¡1), and a pixel size of electron microscopy as a result of radiation damage. J. Ultrastruct. Res. 36 (3), 466–482. 9.8 Å using the “predictive” UCSF tomography package Glaeser, R.M., Cosslett, V.E., Valdre, U., 1971. Low temperature electron (Zheng et al., 2004). All holes imaged at both temperatures microscopy: radiation damage in crystalline biological materials. J. were chosen from one grid square on one grid with similar Microscopie 12, 133–138. ice thickness except one (80 e¡/Å2 at »12 K), which had to be Heide, H.G., 1982. On the irradiation of organic samples in the vicinity of repeated on a diVerent day and grid due to technical ice. Ultramicroscopy 7, 299–300. Y Heide, H.G., 1984. Observations on ice layers. Ultramicroscopy 14, di culties. Tomographic reconstructions were calculated 271–278. using IMOD (Mastronarde, 1997). One hundred particles Heymann, J.B., 2001. Bsoft: image and molecular processing in electron from each tomogram were manually selected and aligned to microscopy. J. Struct. Biol. 133 (2–3), 156–169. the published structure (Mouche et al., 2003) using the Bsoft Iancu, C.V., Wright, E.R., Benjamin, J., Tivol, W.F., Dias, D.P., Murphy, X X package (Heymann, 2001) and the Peach distributed compu- G.E., Heymann, J.B., Jensen, G.J., 2005. A “ ip- op” rotation stage for routine dual-axis electron cryotomography. J. Struct. Biol. 151, 288–297. tation system (Leong et al., 2005). The highest cross-correla- International Study Group I.E.S. 1986. Cryoprotection in electron micros- tion coeYcient, which identiWes the best three-dimensional copy. J. Microsc. 141(3), 385–391. alignment, was used as a quantitative measure of reconstruc- Jenniskens, P., Blake, D.F., 1994. Structural transitions in amorphous tion quality. An equal number of arbitrary positions not con- water ice and astrophysical implications. Science 265, 753–756. taining particles within the tomograms were picked and Jenniskens, P., Blake, D.F., Wilson, M.A., Pohorille, A., 1995. High-density amorphous ice, the frost on interstellar grains. Astrophys. J. 455, 389–401. aligned using identical calculations to assess the cross-corre- Leong, P.A., Heymann, J.B., Jensen, G.J., 2005. Peach: a simple Perl-based lation coeYcients from “noise.” For the surface rendition system for distributed computation and its application to cryoEM data insets in Fig. 5, particles were denoised with 10 cycles of non- processing. Structure 13, 1–7. linear anisotropic diVusion (Frangakis and Hegerl, 2001). Mastronarde, D.N., 1997. Dual-axis tomography: an approach with alignment methods that preserve resolution. J. Struct. Biol. 120 (3), 343–352. Mouche, F., Zhu, Y., Pulokas, J., Potter, C.S., Carragher, B., 2003. Auto- Acknowledgments mated three-dimensional reconstruction of keyhole limpet hemocyanin type 1. J. Struct. Biol. 144 (3), 301–312. We thank Y. He for providing liposomes; T. Wagenkn- Sandford, S.A., Allamandola, L.J., 1993a. Condensation and vaporization echt and J. Berkowitz for pyruvate dehydrogenase; J. Ben- studies of CH30H and NH3 ices: major implications for astrochemis- try. Astrophys. J. 417, 815–825. jamin, P. Leong, and J. Ding for help with data collection Sandford, S.A., Allamandola, L.J., 1993b. H2 in interstellar and extraga- and image processing; and W. Tivol for reading the manu- lactic ices: infrared characteristics, ultraviolet production, and implica- script. This work was supported in part by NIH Grant PO1 tions. Astrophys. J. 409, L65–L68. GM66521 to G.J.J., DOE Grant DE-FG02-04ER63785 to Stark, H., Zemlin, F., Boettcher, C., 1996. Electron radiation damage to V G.J.J., a Searle Scholar Award to G.J.J., the Beckman Insti- protein crystals of bacteriorhodopsin at di erent temperatures. Ultra- microscopy 63, 75–79. tute at Caltech, and gifts to Caltech from the Ralph M. Par- Zheng, Q.S., Braunfeld, M.B., Sedat, J.W., Agard, D.A., 2004. An improved sons Foundation, the Agouron Institute, and the Gordon strategy for automated electron microscopic tomography. J. Struct. and Betty Moore Foundation. Biol. 147 (2), 91–101.