Need for Speed: Examining Protein Behavior during CryoEM Grid Preparation at Different Timescales

Summary A host of new technologies are under development to improve the quality and reproducibility of cryoelectron microscopy (cryoEM) grid preparation. Here we have systematically investigated the preparation of three macromolecular complexes using three different vitrification devices (Vitrobot, chameleon, and a time-resolved cryoEM device) on various timescales, including grids made within 6 ms (the fastest reported to date), to interrogate particle behavior at the air-water interface for different timepoints. Results demonstrate that different macromolecular complexes can respond to the thin-film environment formed during cryoEM sample preparation in highly variable ways, shedding light on why cryoEM sample preparation can be difficult to optimize. We demonstrate that reducing time between sample application and vitrification is just one tool to improve cryoEM grid quality, but that it is unlikely to be a generic “silver bullet” for improving the quality of every cryoEM sample preparation.


44
Single particle cryo-electron microscopy (cryoEM) has emerged as a major structural 45 biology technique during the last decade (Kuehlbrandt 2014

47
and automated data acquisition (Thompson et al. 2019) have streamlined the technique, 48 sample preparation remains a major bottleneck for many projects. Single particle cryo-EM 49 sample preparation requires the specimen to be spread as a thin (≤ 20 to 80 nm) liquid film 50 (Rice et al. 2018) before being rapidly vitrified by plunging into a cryogen liquid such as ethane 51 (Dubochet & Lepault 1984). The formation of this thin film has commonly been achieved by 52 applying a relatively large sample volume (3-4 μL) to a cryoEM grid, and then blotting away 53 excess liquid with filter paper. The cryoEM grid, a 3 mm diameter metal (commonly copper) 54 disk with square windows, has a support layer (typically amorphous carbon), with small, 55 usually circular perforations (~1-2 μm diameter) in a regular array. The typical blotting process 56 removes almost all of the liquid applied to the grid, leaving a thin film of sample suspended 57 across the holes in the support where imaging can occur. This procedure was pioneered over 58 30 years ago by Dubochet and colleagues (Dubochet & Lepault 1984).

59
Formation of a thin film using blotting paper followed by vitrification can be achieved 60 through manual and home-built devices, as well as using commercially available devices such 61 as the Vitrobot™ (Thermo Fisher Scientific), EM GP (Leica Microsystems) and CP3 (Gatan),

62
where the general concept remains the same as when the method was first conceived. While 63 there can be problems with reproducibility of thin film formation through a blotting approach it 64 is undeniably successful, resulting in its application to a broad range of specimens, and it has 65 consequently come to underpin the vast majority of single particle structures to date.

66
Over the years, and across different fields of research, it has been shown that the air-

75
A recent systematic study of particle localisation on cryo-EM grids prepared with

88
The grid making process is currently a major focus in the cryoEM field with a number 89 of approaches in various stages of development, all seeking to improve access, quality, and/or 90 reproducibility of cryoEM sample preparation. The Spotiton system uses an inkjet piezo 91 dispenser to directly deposit samples onto self-wicking grids to create a thin film, and is chameleon. Since each of these sample preparation devices exposes particles to different 105 environments, forces, and timescales, we will briefly describe the specifics of each device.

106
The Vitrobot™ involves the application of 3-4 μL of sample volume onto an EM grid 107 held in a temperature-and humidity-controlled chamber. Subsequently it is blotted between 108 two sheets of filter paper, for 3-10 seconds, removing the vast majority of the sample volume, 109 before the blotting paper is withdrawn and the sample is plunged into the cryogen. Grids can 110 be prepared on a timescale of 5-15 seconds from sample application, and typical grids will 111 have a gradient containing some areas that are too thick and some that are too thin, with a 112 large number of suitable grid squares for imaging ( Figure 1A) (Thompson et al. 2019). While 113 this device has been used to successfully vitrify a wide range of specimens, there is evidence 114 that the irregular pattern of fibres in the filter paper causes non-uniform alterations in surface-115 to-volume ratio across the grid, and this may be a root cause of the irreproducibility often 116 reported for blotting paper-based vitrification techniques, as well as being detrimental to 117 samples (Armstrong et al. 2020).

118
The TED was primarily designed to perform time-resolved experiments by rapidly 119 mixing constituents before vitrification on the millisecond timescale. However, in this study we 120 only make use of its ability to deposit a single sample and vitrify on a very fast timescale (≥ 6 121 ms) (Kontziampasis et al. 2019). A conventional EM grid is placed on a plunging arm which 122 has an adjustable speed within a high humidity chamber at room temperature. The liquid system (syringes, tubing and nozzle) is then equilibrated with ~40 µL sample, which is 124 deposited by spraying directly onto the grid as it plunges into the cryogen. A typical experiment 125 requires between 4 and 32 µL of sample volume per grid, depending mainly on the liquid flow 126 rate. Exposure time to the AWI is determined by the time of flight for the spray droplets (from 127 nozzle to grid) and the grid plunge time (from spray to ethane). A typical grid has a random 128 droplet pattern, with some thick regions corresponding to the centre of a droplet, and thinner 129 edges (which sometimes cover ~½ of a grid square) where the ice is sufficiently thin for 130 imaging ( Figure 1B). With the current design, dispense-to-plunge times can be set from 6 ms 131 to seconds.

132
The chameleon is a fully automated instrument which dispenses controlled droplets 133 onto a self-wicking grid as it plunges into the cryogen. Self-wicking grids and 5 µL of sample 134 are manually placed into the instrument as input. Workflows guide the user through system

143
50S, 70S, all C1 symmetry). Apoferritin was chosen as it is a common test specimen in 144 cryoEM, HSPD1 because when prepared using standard cryoEM methods it adopts an 145 extremely preferred orientation, and ribosomes because they are considered to be a very 146 robust macromolecular complex, and are also asymmetric, unlike the other two specimens.

150
Partitioning of particles to the AWI

151
The speed of grid making has been reported to influence the particle distribution at the 152 AWI, with ~100 ms showing a change in partitioning and angular orientation relative to slower 153 speeds (Noble, Dandey, et al. 2018). We used cryo-electron tomography (cryoET) to 154 investigate differences in particle partitioning in the thin ice layer at different time points for 155 various macromolecular complexes, using the Vitrobot™, TED, and chameleon ( Figure 2A).

156
Areas for tomogram acquisition were selected without prior investigation of particle distribution 157 in that area, and based upon ice thicknesses that would be deemed most suitable for data    The 'fast' TED data demonstrate that even on the fastest timescales we could 174 investigate using this device and in thick ice (up to ~180 nm), the interaction with the AWI is not eliminated. This is perhaps unsurprising given that calculations suggest particles will 176 interact 10-100 times with the AWI within 1 ms and for some proteins this interaction results 177 in sequestering at the AWI (Naydenova & Russo 2017). It should be noted that TED generally 178 produces thicker ice, especially at faster dispense-to-plunge times, as the TED relies on 179 droplet spreading upon contact with the grid to produce areas sufficiently thin to image (Table   180   S1). For the apoferritin grids prepared using the TED, we observed interesting trends in

194
Previous studies suggested that there is a variation in the concentration of the 195 necessary amount of sample required to achieve similar particle numbers in frozen grids when 196 using the Vitrobot™, TED and chameleon. We used tomograms to calculate particle 197 concentration within the thin film on frozen grids, which was compared to the concentration of

220
Orientation and angular distribution of HSPD1 221 HSPD1 is known to adopt strong preferred orientation when prepared using standard 222 blot-freezing methods. We examined HSPD1 angular orientation using the TED at 6 and 50 223 ms, the chameleon at 54 ms and the Vitrobot TM (Figure 4). Single particle datasets for each 224 timepoint and device were collected and combined after pre-processing. 2D-and 3D-225 classifications were performed on the combined data to impose the same class selection 226 criteria on all datasets and the consensus structure was determined. From this, the angular assignments for particles that were frozen using each device at the specific timepoints, were 228 extracted to analyse trends in preferred orientation ( Figure 4C, Figure S4).

229
As expected for HSPD1, strong preferred orientation was seen, with the 'top' and 230 'bottom' projections dominating the particle views present in all data collected. The quality of 231 the consensus 3D reconstruction suffered from the anisotropy of views, as seen in the Z-232 directional FSC ( Figure S5). The Vitrobot™ blotted sample ( Figure 4C) showed the strongest 233 preferred orientation. By increasing the speed of grid making using either the chameleon or 234 TED, more angular distributions were available compared to the standard blotted grid.

235
Reducing the time-delay further, from 50 ms to 6 ms on the TED, provided further minor 236 improvements in angular distribution, although the data were still dominated by preferred 237 views.

238
Due to variations between datasets, such as ice thickness and particle number, it is 239 not possible to draw comparisons between the freezing devices used and resolution 240 outcomes. Instead we limit comparisons to the range of angular distributions. For example, 241 the reconstruction from the 6 ms TED data, which had a greater angular distribution, is limited 242 in resolution to approximately 7 Å. This is likely due to increased ice thickness compared to 243 the other datasets (Table S1); other reconstructions are likely resolution-limited due to low 244 particle numbers or ice thickness ( Figure S4).

245
Orientation and angular distribution of ribosomes 246 A sample containing the 30S, 50S and 70S ribosomes was used to investigate the 247 angular distributions of three related specimens in one dataset to keep as many parameters 248 constant as possible (ice quality etc.). Applying the same approach used to examine HSPD1 249 angular distribution, we collected single particle datasets for ribosome samples prepared with

252
The 30S subunit showed a clear correlation between speed of grid preparation and 253 improved angular distribution ( Figure 5B). This trend was also present in the 50S subunit data, 254 although it is not as pronounced ( Figure 5C). Interestingly, this trend is not present for the full 255 ribosome; instead the greatest angular distribution was observed from grids prepared using 256 the Vitrobot™ ( Figure 5D). Taking the datasets through the processing pipeline, none of the 257 ribosome reconstructions appear to be limited in resolution by angular orientations, and the   The physics of diffusion and AWI interactions cannot be outrun using technology 275 currently available (to the best of our knowledge) for cryoEM sample preparation. Even in the 276 fastest cases of grid vitrification in our study (6 ms) and using different approaches (blotting 277 vs spraying), the majority of particles still partitioned to the AWI. Considering AWI partitioning data from the three specimens we examined, apoferritin, HSPD1 and ribosomes, conflicting 279 lessons can be learnt from each. HSPD1 data suggest that the faster grids are prepared, the 280 fewer particles partition to the AWI (Figure 2, Figure S1). The ribosome data suggest the 281 precise opposite; the faster the grids are prepared, the more particles partition to the AWI 282 ( Figure 2, Figure S1). The apoferritin data are the most variable and provide the least clear 283 picture across different timescales, which may be partially explained by the propensity of 284 apoferritin to form 'rafts' at the AWI (discussed below in 'changes in particle concentration').

285
Overall, altering speed of grid preparation could be one mechanism to influence AWI

296
and how this varies depending on the sample. These data suggest that for a given specimen, 297 there may be a fast (<10 ms) stage where the protein initially partitions to the AWI, followed 298 by a slower stage where the particle explores its energy landscape before settling into a local 299 energy minimum. For some specimens, there may be a distinct orientation (leading to 300 preferred orientation), and for other specimens it may be a variety of orientations (Figure 7).

301
The timescale in this second, slower stage is likely to vary from specimen to specimen.

317
Concentrating effect of Vitrobot™ blotting 318 One of the most striking results was the change in concentration due to blotting as 319 compared with spraying ( Figure 3). These data clearly demonstrate that for the specimens we 320 have examined, the Vitrobot™ blotting approach greatly enriches the thin film with particles, 321 for the specimens examined here. And indeed, that the AWI may be responsible for the 322 concentration of particles in the thin film, which in many systems is required to achieve a viable 323 number of particles per micrograph. It should also be noted that the degree of concentration 324 is sample-dependent. This may go some way towards explaining the experience of many 325 cryoEM researchers in ascertaining the 'right' concentration of protein to use for their system.

326
Adsorption to the grid support may also have a significant impact on apparent particle 327 concentration in the imageable areas which requires further investigation.

328
Across both TED and chameleon, higher concentrations of specimen were necessary 331 at faster timepoints. However, specifically with TED at the 'fast' timescales, a depletion of 332 particles for apoferritin and HSPD1 was observed (Figure 3). The apoferritin data from TED 333 display greater variability relative to the other samples, which could be linked to the formation 334 of surface aggregates that were also observed in these data ( Figure S2). Surface aggregates, 335 or particle 'rafts', may begin to form while the droplet is traveling from nozzle to grid in TED

346
A major unexplained aspect of these data is that for HSPD1 and apoferritin at the 'fast'

373
Formation of a sacrificial layer of denatured protein has been shown for apoferritin 374 (Yoshimura, Hideyuki, et al. 1994), but the timescale of complete particle denaturation on 375 cryoEM grids remains an open question. There may be alternative explanations for these data, 376 and the hypotheses presented are not mutually exclusive. It is likely that there is an interplay 377 between multiple mechanisms on a specimen-dependent basis. It is only with additional 378 information on these trends across many specimens, added to these initial data that a better 379 understanding of particle behaviour in thin films can be achieved.
In conclusion these data go some way to offering an explanation to those cryoEM 381 researchers who have experienced huge variability in cryoEM sample preparation between 382 biological specimens. General trends indicate speed may ameliorate some of the adverse 383 effects of the AWI, thus providing a significant improvement in intact or non-preferentially 384 oriented particles. However, this speed may come at the price of a higher required sample 385 concentration, with data suggesting that, the faster grids are prepared, the higher the 386 concentration of protein required. This effect may seem exacerbated given the concentrating 387 effect currently enjoyed when using Vitrobot™ blotting to prepare samples.

388
While much is still unknown about the behaviour of particle in thin films, a general 389 model can be used to summarise the aforementioned ideas (Figure 7). First, diffusion dictates 390 the rate at which a particle interacts with the AWI. This is an initial fast phase, occurring within 391 ≤ 1 ms of the thin film forming. Each specimen will then have its own on-off rate and local 392 energy minima at the AWI, determining how likely it is for the protein to disassociate back into 393 bulk solution. Next, negative aspects of the AWI may take place with partial denaturation or         Table S1. lower AWI, respectively. Then, the closest distance was determined between either of the 500 AWIs and each particle. Particles which were at a distance ≤ 10 nm to an AWI were classed 501 as 'bound' to the AWI. For the majority of tomograms collected, the 10 nm threshold 502 adequately allowed characterisation of the data, but for tomograms on areas of thick ice (> 80 503 nm)/where the AWI is not clearly defined, a threshold of 20 nm was more suitable. Ideal 504 particle behaviour was modelled using the experimentally determined AWIs and randomly 505 generating particle coordinates (number/volume corresponding to the respective 506 concentration) in between the experimental ice layer. Then, distances between modelled 507 particles and AWIs were determined.

508
Single particle cryoEM data collection and processing All single particle cryoEM data was collected in the Astbury Biostructure Laboratory in Leeds 510 on Titan Krios I, equipped with a FEI Falcon III detector and operated in integrating mode.

511
Data collection parameters are listed in Tables S3 and S3.

517
All HSPD1 datasets were combined after particle extraction (rescaled to 2.13 Å pixel size).

518
One round each of 2D-and 3D-classification were used to clean the dataset. Consensus 519 reconstructions with particles from all datasets were generated in C1 and C7 symmetry and 520 used to determine angular distributions. Finally, the dataset was split into its original subsets 521 and each subset of particles and used to generate a reconstruction using the assigned angles 522 from the C7 consensus reconstruction.

523
Similarly, all ribosome datasets were combined after extraction and subjected to one round of 524 2D classification to remove 'junk' particles. Then, 3D classification was performed to separate 525 the combined datasets into 70S, 50S and 30S subsets. Those subsets were cleaned up by an 526 additional round of 2D classification (2 rounds for 30S) and a consensus reconstruction was 527 generated including data from all 4 datasets for the three species (70S, 50S and 30S). The 528 subset for each species was then further split into the original datasets, resulting in 529 reconstructions for 70S, 50S and 30S for each timepoint.

532
The probability density function was estimated using kernel density estimation with a gaussian 533 kernel at a fixed bandwidth of 10°, wider than the estimated angular accuracy in all cases (to 534 avoid overinterpretation of angular distribution maps).

536
We thank Dr Hao Shao and Dr Jason Gestwicki (UCSF) for providing the HSPD1 plasmid