Flash melting amorphous ice

Water is a poor glass former.^1,2 In order to achieve vitrification, it has to be cooled so rapidly that crystallization in so-called no man’s land¹ can be outrun, a range of deeply supercooled temperatures between 160 and 232 K, where water crystallizes within tens of microseconds.³ Critical cooling rates of 3 × 10⁵ K/s (Ref. 4) and 10⁶–10⁷ K/s have been determined with different methods.⁵ In fact, it had long been thought impossible that aqueous solutions could be vitrified,⁶ making the demonstration of successful vitrification a stunning achievement.^7,8 Arguably, one of the most important applications of this process is the preparation of samples for cryo-electron microscopy (cryo-EM),⁹ which is now set to become the preferred method in structural biology.¹⁰ Vitrification allows biological specimens to be preserved in a frozen-hydrated state, in which they can be imaged in the vacuum of an electron microscope while the damage inflicted by the electron beam is mitigated.^11–13 

Fast crystallization in no man’s land had long made it impossible to map out the structural evolution of water during the vitrification process.^3,14–18 Recently, we have used time-resolved electron diffraction to show that water evolves smoothly from a high- to a low-temperature structure between 260 and 220 K.³ Below 200 K, its diffraction pattern converges to that of hyperquenched glassy water (HGW), a glassy form of water with a glass transition temperature of 136 K.⁴ Such smooth evolution had previously also been inferred from the infrared spectra of transiently heated amorphous ices.¹⁹ Here, we study the reverse of the vitrification process by characterizing the structural evolution of different amorphous ices during impulsive laser melting. Laser heating of amorphous ice has frequently been used to prepare deeply supercooled water.^16,19–23 For example, such experiments have provided evidence of a liquid–liquid phase separation in no man’s land that can be observed after laser heating of high and low density amorphous ice.^22,23 Recently, we have also introduced a microsecond time-resolved approach to cryo-EM that involves rapidly melting a cryo sample with a microsecond laser pulse.^11,12,24 Once the sample is liquid, a suitable stimulus is used to initiate the dynamics of the embedded proteins. As they unfold, the heating laser is switched off, and the sample revitrifies within microseconds, trapping the proteins in their transient configurations, in which they are subsequently imaged. For example, we have used this approach to image the microsecond motions that the capsid of the cowpea chlorotic mottle virus (CCMV) virus undergoes as part of its infection mechanism.²⁴ Previous experiments have raised the question of whether cryo samples may partially crystallize during laser melting despite heating rates of about 10⁷ K/s.^12,25 This is consistent with previous studies that have shown that for aqueous solutions of cryoprotectants, the critical heating rate for outrunning crystallization is several orders of magnitude higher than the critical cooling rate.^26,27 However, these experiments were performed with high weight fractions of cryoprotectants to slow down crystallization, whereas no data are available for dilute solutions. Here, we characterize the melting process in amorphous ice samples of pure water with time-resolved electron diffraction.

Experiments are performed with a transmission electron microscope that we have modified for time-resolved experiments (supplementary material, Note 1).^28,29 As illustrated in Fig. 1(a), a 263 nm thick layer of amorphous solid water (ASW) is deposited in situ onto a sheet of few-layer graphene that is supported by a holey gold film (2 µm holes) on a 600 mesh gold grid held at 101 K. We then use a 30 µs laser pulse (532 nm) to melt the sample in the center of a grid square and probe its structural evolution by capturing a diffraction pattern with an intense, high-brightness electron pulse of 2 µs duration, containing about 10⁵ electrons [Fig. 1(b)].

Figure 1(c) shows a micrograph of a typical ASW sample, with the corresponding diffraction pattern in the inset and the positions of the diffraction maxima indicated with blue lines (supplementary material, Note 12). When we heat the sample with a 30 µs laser pulse to about 185 K, it devitrifies and forms stacking disordered ice³⁰ [Fig. 1(d)], while at higher temperatures, the crystallization of hexagonal ice is favored^16,31 [Fig. 1(e), ∼235 K, supplementary material, Note 11]. At even higher laser power, the sample melts and remains liquid for the duration of the laser pulse. Once the laser is switched off, the sample cools rapidly as the heat is efficiently dissipated to the surroundings, which have remained at cryogenic temperature.^3,11 With a cooling rate of over 10⁷ K/s [see discussion in Fig. 2(a)], the sample vitrifies and forms HGW [Fig. 1(f)] after heating to 281 K. Note that the positions of the diffraction maxima of ASW in Fig. 1(c) are slightly shifted with respect to HGW. Such shifts are usually associated with the formation of microporous ASW, which is typically obtained at lower temperatures than in our experiment.^32,33 Contrast variations in the micrograph of Fig. 1(c) suggest that even under our deposition conditions, the samples exhibit a small amount of nanoscale porosity.

We characterize the structural evolution of the ASW sample during laser melting with time-resolved electron diffraction, with the diffraction pattern of the ASW sample before the laser pulse shown in Fig. 2(b) (black curve). For comparison, a simulation of the temperature evolution during this process is presented in Fig. 2(a) (black solid curve, supplementary material, Note 8). Under laser irradiation, the sample heats up rapidly, reaching a temperature of 229 K at 6 µs. The corresponding diffraction pattern reveals that the supercooled liquid has traversed most of no man’s land without crystallization [Fig. 2(c)]. In contrast, additional diffraction features begin to appear at 7 µs (black arrows) that rapidly grow more intense [Fig. 2(e) and 9 µs], indicating the formation of stacking disordered ice and later, hexagonal ice (Fig. S7). Once the sample temperature reaches the melting point [16.8 µs in the simulation of Fig. 2(a)], the crystallites begin to melt, so that at long time delays, the diffraction pattern turns into that of stable water [Fig. 2(f), 25–30 µs]. We determine that at the end of the laser pulse, the sample reaches a temperature of 281 K (supplementary material, Note 7). Time-resolved diffraction patterns of the complete laser melting process are shown in Fig. S8. Note that, as discussed below, the simulation of the temperature evolution in Fig. 2(a) does not account for the heat released by the crystallization process during laser melting. This causes the sample to heat up more rapidly and approximately follow the temperature evolution indicated by the dashed line.

Our experiments reveal that crystallization occurs during melting but is avoided during vitrification, even though the respective heating and cooling rates are similar [Fig. 2(a)]. In fact, when crystallization sets in at about 6.8 µs (supplementary material, Note 6), the heating rate has not dropped below 5 × 10⁶ K/s. This suggests that the critical heating rate of pure water is higher than the critical cooling rate, similar to solutions of cryoprotectants.^26,27 As has been previously discussed,³⁴ this can be qualitatively understood by considering how the nucleation and growth rates^1,39 [Fig. 2(g)] evolve during the melting and vitrification process [green and purple curves in Fig. 2(a); see also the discussion in the supplementary material, Note 9]. As the sample heats up, a large number of nuclei are formed in no man’s land, where the nucleation rate goes through a maximum. At higher temperatures, nucleation slows significantly, but the growth rate picks up, causing the already existing nuclei to grow rapidly so that the sample crystallizes. In contrast, during the vitrification process, a large concentration of nuclei is only formed once the growth rate has already dropped. These nuclei will then grow only a little before the sample vitrifies and its structure is arrested.

Surprisingly, crystallization occurs significantly earlier when we repeat the experiment with the HGW sample that is obtained after the laser pulse [blue curves in Figs. 2(b)–2(f)]. Diffraction features indicating crystallization already appear at 6 µs [Fig. 2(c), blue arrows], after which they grow rapidly [Fig. 2(d)]. At the end of the laser pulse, the sample has fully melted, and its diffraction pattern is indistinguishable from that of the ASW sample [Fig. 2(f)]. The difference in crystallization kinetics is clearly evident in the evolution of the first diffraction maximum. Figure 3(a) shows that its intensity initially decreases as the sample heats up but rises sharply when crystallization sets in, which occurs at about 6.8 µs for ASW (black), but already at 5.8 µs for HGW (blue), 1 µs earlier (supplementary material, Note 6). This difference cannot be fully explained by the fact that the HGW sample is thinner due to evaporation during the first laser pulse and, therefore, heats up more rapidly. Simulations of the crystallization kinetics reveal that this effect can only account for a time delay of about 0.48 µs (supplementary material, Note 9). Indeed, when we laser melt the HGW sample a second time, it only crystallizes another 0.42 µs earlier [HGW₂, green data points in Fig. 3(a)]. The remaining time difference in the onset of crystallization between the ASW and HGW samples of over 0.5 µs must, therefore, result from a difference in the crystallization kinetics.

The intensity of the first diffraction peak goes through a maximum once the melting point is reached and crystal growth ceases, which occurs at ∼8.0 µs for HGW₂, 8.5 µs for HGW, and 9.5 µs for ASW [Fig. 3(a)]. This is earlier than predicted by the simulation in Fig. 2(a) (solid line), which does not account for the heat released by the partial crystallization of the sample (16.8 µs). This heat release causes the sample to warm up more rapidly and follow a temperature evolution as sketched by the dashed line. From the time difference between the simulation and experiment for reaching the melting point, we can estimate that roughly a third of the sample crystallizes during laser melting (supplementary material, Note 10). Based on the relative diffraction intensities, we calculate that the maximum crystalline fraction of the HGW and HGW₂ samples is about 1.5 times as large as for ASW, a reflection of the faster crystallization kinetics of HGW.

The position of the first diffraction maximum provides insights into the structural evolution of the supercooled liquid before it crystallizes [Fig. 3(b)]. The first diffraction maximum of HGW (HGW₂) appears at 1.74 Å⁻¹, whereas that of ASW is shifted to 1.77 Å⁻¹ under our deposition conditions, as discussed above. During laser heating, the peak positions barely change until crystallization sets in, which is marked by a shift to lower momentum transfer, indicating the formation of stacking disordered ice, which features a strong reflection at 1.71 Å⁻¹ [Fig. 1(d)]. This suggests that the supercooled liquids obtained by laser heating ASW and HGW initially retain distinct structures. In particular, the ASW sample appears to partially preserve its nanoscale porosity. Nucleation, which predominantly takes place at lower temperatures [∼140–230 K, Fig. 2(g)], must, therefore, occur in non-equilibrium configurations of supercooled water in our experiment. Evidently, the larger surface area of the ASW sample and the associated changes in local structure slow crystallization. This is in contrast to previous experiments on the crystallization of slowly heated amorphous ices, which found similar crystallization rates for ASW and HGW samples that were heated to 150 K.³⁵ Increasing the porosity of ASW was even shown to increase the crystallization rate in the temperature range of 130–141 K,³⁶ an observation that was ascribed to the associated larger surface area and the fact that surface nucleation dominates the crystallization process.^37,38 In our experiment, the nanoporous ASW sample instead crystallizes more slowly, which is likely a consequence of the flash heating process, which causes nucleation to predominantly occur at higher temperatures.

Based on simulations of the crystallization kinetics, we estimate an average nucleation rate of HGW in no man’s land of 3.4 × 10²⁶ m⁻³ s⁻¹, in rough agreement with previous measurements, while that of ASW is 6.6 × 10²⁵ m⁻³ s⁻¹, about five times lower (supplementary material, Note 9). Note that our simulations can also reproduce the experimental observations if we instead assume a 1.7 fold higher growth rate for HGW but an identical nucleation rate. We, therefore, cannot exclude that the observed difference in crystallization kinetics is in part due to the different growth rates of the two supercooled liquids. This is, however, unlikely to be the dominant contribution since the growth rate scales with diffusivity,^14,20,39 which intuitively should only weakly depend on changes in local structure.

In conclusion, our experiments reveal that during the rapid melting of ASW and HGW samples with microsecond laser pulses, about a third of the sample crystallizes, despite a heating rate of more than 5 × 10⁶ K/s. In contrast, crystallization is avoided entirely under the same conditions during hyperquenching with a similar cooling rate of ∼10⁷ K/s. This can be understood by considering that as the sample heats up, it first undergoes fast nucleation before rapid growth sets in, whereas the opposite occurs during the vitrification process.³⁴ One can estimate an upper limit for the critical heating rate of 10¹⁰ K/s since crystallization can be outrun with 10 ns laser pulses.^19–21 

Our experiments open up the intriguing perspective of investigating how changes in local water structure affect the crystallization trajectory as well as the preference for the formation of different polymorphs.⁴⁰ For example, one would expect that preparing an amorphous ice precursor with an increased concentration of pentagonal rings would slow crystallization, which requires these rings to open and is, therefore, associated with a high energetic barrier.⁴¹ Such experiments promise to help further elucidate the crystallization mechanism of water, which has remained an active area of research.^1,42,43

We have previously shown that near-atomic resolution reconstructions can be obtained from revitrified cryo samples and that the structure of the proteins is not altered by the melting and revitrification processes.^25,44 The partial crystallization of the sample during the melting process, therefore, does not appear to negatively affect microsecond time-resolved cryo-EM experiments. If it did, it should always be possible to increase the heating rate and outrun crystallization by using a shaped laser pulse with an intense leading edge.⁴⁵ Transient crystallization during melting may, however, contribute to improving the sample quality by partially reshuffling the angular distribution of the particles²⁵ and, thus, help to overcome issues with preferred orientation that plague many cryo-EM projects.⁴⁶