Incident electrons can excite atoms by depositing their energy and charges to the sample (Jiang, 2016). The deposited energy allows the rearrangement of the inter- or intramolecular bonds, resulting in changes of the molecular structure or the formed crystal structure by radiolysis (Grubb, 1974; Egerton, 2021). The study of the tobacco mosaic virus showed an example of the modification of covalent bonds in macromolecules by radiolysis (Fromm et al., 2015). Radiolysis initiates the decarboxylation process by breaking the hydrogen–oxygen bond and carbon–carbon bond in carboxylate residues. The breaking of the carboxylate residue damages nearby side chains, which can further change the secondary structure and flexibility of the molecule (Fromm et al., 2015).

The degree of radiation damage by radiolysis has three levels. 1) Local damage: the incident electron directly excites the local electrons of an atom, breaking its covalent bonds (shown in Figure 3A) (Kato et al., 2021). The damage leads to the appearance of free radicals in biological specimens (Mishyna et al., 2017). 2) Nearby damage: some free radicals can trigger a cascade of chemical reactions in nearby atoms (Grant and Grigorieff, 2015), and then transfer the free radicals to nearby atoms (shown in Figure 3B). 3) Structure crash: as energy and free radicals spread to nearby atoms through chain reactions, the chain reaction can terminate by generating new bonds between neighboring atoms and permanently changing the structure. For example, when the water molecule is dissociated to form a hydrogen and hydroxyl radical (H₂O ⇒H• + OH•), some free radicals are converted back into H₂O, while other free radicals close to the protein molecule react with the hydrogen atoms in the protein molecule to form hydrogen gas and a new free radical (OH•+ R-H ⇒ RO + H₂). The released hydrogen coalesces into gas bubbles, which permanently damage the organic specimen (Leapman and Sun, 1995; Hankamer et al., 2007).

The electron beam causes electrostatic charging on the surface of the specimen. Negative charges can be caused by the accumulation of deposited incident electrons (Cazaux, 1986), while positive charges can be caused by the emission of secondary electrons of the insulating vitreous specimen (Egerton, 2021). These accumulated charges can cause electrostatic lensing by changing the trajectory of the incident electron, blurring the image and even generating a reaction force that induces mechanical movement of the sample, which in turn bends and distorts the thin specimen (Vinothkumar and Henderson, 2016).

Cryo-EM imaging often causes an increase in temperature of the specimen, which manifests as melting or boiling (Dubochet et al., 1988). Heating is caused by the electron beam during the process when its kinetic energy is partially converted into thermodynamic energy of the atoms in the specimen via electron-phonon scattering (one type of inelastic electron scattering) (Johnston-Peck et al., 2016). Two factors can affect the local temperature rise of the specimen, that is, the total thermal energy absorbed by the specimen and the thermal conductivity between the specimen and the specimen-supporting film (Voss et al., 2021).

The mass of the specimen can be reduced by the incident electron beam (Aronova et al., 2010) via two types of processes, that is, knockout and sublimation. For knockout, the incident electron directly hits the atom and kicks it out from the specimen (Kelly, 2002; Egerton et al., 2004; Gu et al., 2017; Su et al., 2019; Chen et al., 2020). Sublimation can be classified into direct sublimation and indirect sublimation. In direct sublimation, the energy directly increases the temperature of small molecules transforming the molecules from their solid state to their gaseous state, and releases them from the specimen (Dubochet et al., 1988). For indirect sublimation, radiolysis breaks the chemical bonds inside the macromolecules and transforms the fragments into small molecules, which are then released from the specimen surface in the gas phase (Egerton, 1999). For instance, water molecules in ice can be directly heated to a gas phase via sublimation into the vacuum environment, thus reducing the thickness of the specimen. Alternatively, the hydrogen–oxygen bond of water molecules can be also broken by radiolysis, generating many chemical intermediates (e.g., hydrogen radical, hydroxyl radicals, and hydrated electrons). These chemical intermediates can react with nearby damaged molecules released from the original macromolecules and form smaller new molecules, which then sublimate from the specimen (Le Caër, 2011). Both processes reduce the mass of specimens.

In cryo-EM imaging, the incident electron beam often causes contamination on the specimen surface through the calcination of organic molecules, which degrades the images of the targeted molecules (Isabell et al., 1999). The cause of contamination is very complex. One theory suggests that the organic molecules on the specimen surface are polymerized by the electron beams (Egerton et al., 2004). These polymers may attract each other and form an aggregation, increasing the thickness of the local area as irradiation continues (Ashfaq et al., 2020). Another explanation could be that the newly formed small molecules from the abovementioned chemical intermediates induced by radiolysis are absorbed onto the cold specimens with an uneven distribution of surface charge and then aggregated in the illuminated area of the specimen (Dubochet et al., 1988).

The total number of incident electrons illuminating the unit area of the specimen, defined as the dose, has long been believed to be the only experimental parameter that governs the degree of radiation damage of the sample (Frank, 2009; de la Mora et al., 2020). However, the measured dose limit keeps changing over time (Wang et al., 2021a). For example, in the 1970s, the measured dose was ∼1 e⁻ Å⁻² for the 7 Å reconstruction of the purple membrane 2D crystals (Egerton, 2014). In the 1980s, 4.0 to 4.5 Å resolution reconstructions can be obtained using the same limited dose (∼1 e⁻ Å⁻²) on n-paraffin and purple membrane samples (Henderson and Glaeser, 1985). Subsequently, even higher resolution images (3.5 Å and 2.8 Å) were reported using a dose of 20 e⁻ Å⁻² on the purple membrane sample (Henderson et al., 1986a; Baldwin et al., 1988). In the 1990s, the 3.5 Å image of bacteriorhodopsin 2D crystals was acquired using a dose of 10–15 e⁻ Å⁻² (Henderson et al., 1990). 30–35 Å 3D reconstruction of the HSV-1 capsid was obtained using 5–10 e⁻ Å⁻² (Conway et al., 1993). In the 2000s, ∼11.5 Å 3D structure of 70S ribosome was achieved with a dose of ∼20 e⁻ Å⁻² (Gao et al., 2003; Rawat et al., 2003). ∼8 Å structures of bacteriophage P22 capsid and ε 15 bacteriophage were solved with a dose of ∼30 e⁻ Å⁻² (Jiang et al., 2003; Chen et al., 2008a). In the 2010s, ∼3.2 Å resolution structure of β-galactosidase was determined by a total accumulated dose of 45 e⁻ Å⁻² (Bartesaghi, et al, 2014). ∼4 Å resolution structure of Slo2.2 Na⁺-activated K⁺ channel and human transcription factor IIH (TFIIH) were solved with a total dose of 40 e⁻ Å⁻² (Hite et al., 2015; Greber et al., 2017). In the 2020s, 1.7 Å resolution 3D structure of human membrane protein, the β3 GABA_A receptor homopentamer, was solved with a total dose of 13.2 e⁻ Å⁻², while a 1.22-Å resolution reconstruction of mouse apoferritin was solved with a dose of 40 e⁻ Å⁻² (Nakane et al., 2020). Maps of RNA replication complexes were determined up to 8.5 Å resolution by cryo-ET subtomogram averaging with a total dose of 180 e⁻ Å⁻² (Unchwaniwala et al., 2020), and the atomic resolution structure of ferritin was achieved with a total dose of ∼50 e⁻ Å⁻² (Yip et al., 2020). Recently, the full-length Tetrahymena ribozyme was resolved at a resolution of 3.1 Å with a dose of up to 75 e⁻ Å⁻² (Su et al., 2021); the GagT8I and apoferritin were determined up to 5.0 Å and 2.8 Å with the corresponding total doses of 122 e⁻ Å⁻² and 102 e⁻ Å⁻², respectively (Ni et al., 2022). Furthermore, the 3.6 Å resolution dNTPase was resolved with a total dose of 120 e⁻ Å⁻² by cryo-ET subtomogram averaging (Bouvette et al., 2021). A summary of selected cryo-EM publications that contains the achieved 3D map resolution and their total dose is listed in Table 1. A summary of earlier publications can be found in a published article (Kudryashev et al., 2012).

Table 1 shows that 1) the dose limitation measured in recent years is nearly hundred times higher than that measured earlier and 2) the dose limitation measured by SPA is much higher than that measured by 2D crystals. One may ask whether the radiation damage measured on the same protein in different forms (SPA vs. 2D crystal) could be different. In crystals, the chemical bonds, such as ionic and hydrogen bonds, formed between the protein molecules should have no fundamental difference from that formed within a protein for its secondary structure. The radiation damage on the bonds should be the same regardless of whether it is between the protein molecules or within the protein molecules. However, the fact that the observed dose limit measured from the crystal structure is lower than that measured from SPA could be explained as follows. The radiation damage on the protein surface can be detected sensitively by the quality of diffraction spots of crystals but cannot be precisely detected by SPA. The SPA accuracy in classification and alignment is limited by the computer algorithms. The resulted resolution is the averaged resolution, with the surface of the protein usually at a lower resolution than the inside, which is the so-called anisotropic resolution (Aiyer et al., 2021). Under this condition, the SPA resolution is difficult to be linked precisely to the dose limit. Additionally, a higher dose is often used for high-resolution 3D reconstruction via multiple-frame imaging using the direct electron detector (DED). One should note that only the first few frames of each image are used for the final high-resolution reconstruction, while the total frame/dose is used only for the initial low-resolution refinement/reconstruction (Cheng et al., 2015). In other words, the dose used for image acquisition is not equal to the dose for high-resolution 3D reconstruction. The actual dose used for the reported resolution (corresponding level of radiation damage) cannot be well-tracked from the publications.

Despite the abovementioned details, the changing dose in different experiments at different times suggests that dose limitation is a very complex function with respect to the radiation damage. In addition to the measuring methods (such as 2D crystal and SPA) and sample types, many untracked experimental parameters (such as ice thickness and buffer) may also contribute to radiation damage.

The question whether the dose-rate correlates with radiation damage has been debated for decades. The experimental results have been controversial in the fields of both cryo-EM and X-ray crystallography. The experiments that support the argument that dose-rate is unrelated to radiation damage are in the majority based on the measurement of the fading of crystal diffraction spots against the total electron dose under various dose-rates (Glaeser, 1971; Taylor and Glaeser, 1976; Baker et al., 2010). In these experiments, a longer exposure time under the same total dose, that is, a lower dose rate, did not benefit high-resolution imaging (Glaeser, 1979), in which the fading of the critical diffraction spots was only related to the cumulative electron dose (Jeng and Chiu, 1984). Due to the fact that the maximum dose was less than 10 e⁻ Å⁻², it was predicted that the structure of an individual protein molecule would never be possible (Henderson, 1992). A radiation damage experiment by X-ray diffraction intensities showed a similar conclusion, that is, the dose rate did not affect radiation damage at the same total dose. In this experiment, the changes in diffracted intensities between the first image and the third image, which showed radiation damage at the macroscopic level, are proportional to the total dose, but independent of the X-ray dose rate at flux densities up to 10¹⁵ photons s⁻¹ mm⁻² (Sliz et al., 2003). As the underlying principles of radiation damage to protein molecules remain unclear (Dubochet et al., 1988; Glaeser, 2008), the conclusion that the radiation damage is only dependent on the total dose, rather than the dose-rate is still controversial (Chen et al., 2008b; Karuppasamy et al., 2011; Hattne et al., 2018). At a higher dose rate, a higher temperature will be generated. The amount of increase in temperature is dependent on the conductivity of the supporting film. A higher temperature means a higher level of thermal vibration of the atoms, weaker chemical bonds, and more difficulty to heal the broken chemical bonds, and because of this line of reasoning, radiation damage has been thought to correlate with the dose-rate (Ravelli et al., 2002; Hattne et al., 2018).

The experiments supporting the importance of dose-rate on radiation damage include the following. The dose-rate effects were investigated by cryo-EM by collecting several series of cryo-EM images under different dose-rate and acquisition times, but with a same cumulative dose (Chen et al., 2008b; Karuppasamy et al., 2011). For instance, an individual-particle averaging cryo-EM study of the tobacco mosaic virus (TMV) showed less radiation damage in the 3D reconstruction with data acquired at the lower dose-rate based on images acquired under a total dose of 15 e⁻ Å⁻² at dose-rates of 1.5 and 15 e⁻ Å⁻² s⁻¹, respectively (Chen et al., 2008b). This result is consistent with another cryo-EM experiment with a higher dose-rate range of 5–50 e⁻ Å⁻² s⁻¹ (Karuppasamy et al., 2011), suggesting that the secondary radiolytic effects, that is, radical recombination, play a role in the dose-rate effects.

Fading of the critical spot was often observed in the diffraction experiments of crystals of catalase 2D crystals (Taylor and Glaeser, 1974; Glaeser and Taylor, 1978) and 3D microcrystals (Nannenga et al., 2018). The fading phenomenon has been attributed only to the total dose-only–related radiation damage process, and not the dose-rate. One may argue against this viewpoint by the following logic. In the total dose-only model, each electron must permanently damage the local structure of a molecule, and the destroyed local structure should not change the location of molecule, that is, the overall order of the crystal should remain as it is. The undamaged molecules or undamaged portion of the molecules should still contribute to the high-resolution diffraction spots based on their perfect locations because diffraction is defined by the global order rather than the local structural disorder. In other words, the area that has not received any electrons would still contribute to the high-resolution spots during the process. In this case, the high-resolution spot intensities should be weaker instead of disappearing faster than the low-resolution spots. Thus, the dose-only model cannot fully explain the experimental observation, in which the reduction of high-resolution spots SNR is faster than that of the low-resolution spots.

An alternative explanation to the observed fading phenomena can be that the radiation damage is related to both the total dose and the dose-rate, as explained in the following. The incident beam on the illuminated area generates a higher temperature in the crystal. The higher the dose rate, the higher the thermal vibration of the crystal, the easier the disruption of the higher order crystallinity, and the faster the fading of high-resolution spots. The faster fading observed for the high-resolution spots than the low-resolution spots gives a clue that the temperature increase is the first effect before molecular damage. If so, allowing some time for the illuminated area to cool by using a pulsed beam instead of a continuous beam would give a higher dose toleration. A pulsed beam experiment has shown that the radiation damage is dependent on the time between electron pulses (VandenBussche and Flannigan, 2019). The other experiments support the notion that the dose rate correlated with the radiation damage involves the inorganic materials (Jiang and Spence, 2012), organic soft-materials (Kuei et al., 2020), and cells (Gordon Steel et al., 1986; Matsuya et al., 2018), in which self-healing has been proposed for the recovery of the damaged structure under a damaging threshold. The thresholds related to the dose-rate were studied in inorganic materials, biomolecules, and tissues (VandenBussche and Flannigan, 2019). The dose rate–dependent damage of cerium dioxide was found by scanning transmission electron microscopy (Johnston-Peck et al., 2016). The study of eight lysozyme crystals at room temperature and cryotemperature (100 K) showed that the tolerated dose is dependent on the dose-rate in a positive linear relationship (Southworth-Davies et al., 2007). The study of the disulfide bond Cys6-Cys127 showed that the electron density of the bond is dependent on the dose rate (de la Mora et al., 2020).

Due to the abovementioned controversial results from different experiments, whether radiation damage depends on the dose rate in cryo-EM or whether ultrafast electron pulses for TEM imaging is useful to mitigate radiation damage in soft biological materials remains under debate. A summary of selected cryo-EM publications that contain the achieved 3D electron density map resolution and their total dose and/or dose-rate is listed in Table 1 for readers to contemplate.

In addition to the total dose and dose-rate as radiation damage–related parameters, one may ask whether the nature of biological specimens is related to the dose limit for radiation damage (Martynowycz et al., 2021); in other words, whether the intrinsic molecular properties, such as the molecular mass, diameter, density, stiffness, and local structure, affect the dose limit for radiation damage. The experiments that support the independence of radiation damage on the type of protein include the following. Radiation damage studies of protein crystals by cryo-EM reported that the dose limit is roughly the same for a very wide variety of organic and biological specimens (Henderson, 1990). In contrast, X-ray crystallography study suggests that there is no universally applicable dose limit for all types of protein crystals (Liebschner et al., 2015), which means that the dose limit depends on different protein crystals. The experiments that support the view that radiation damage depends on the type of protein or the molecule size or its physical environment (Kempner and Schlegel, 1979; Glaeser, 2016) include the following. The critical dose for the purple membrane at 7 Å resolution was ∼1.0 electrons Å⁻² (Egerton, 2014), while 2D crystals of Connexin26 complexes were resolved at the resolution of 7 Å with an electron dose of 25 electrons Å⁻² (Oshima et al., 2007). Moreover, in the SPA reconstruction, the dose for P22 bacteriophage capsid at 3 Å resolution was ∼37.5 e⁻ Å⁻² (Hryc et al., 2017), while a helical reconstruction of TMV was revealed at 3.3 Å resolution, with a dose of 25 electrons e⁻ Å⁻² (Ge and Zhou, 2011). One explanation that radiation damage may depend on protein types is that different proteins contain different percentages of negatively charged carboxylate residues, which are most susceptible to radiation damage. The radiolysis of carboxylate residues may lead to the deterioration of other intermediate and bulky side chains, as shown in Figure 4 (Grant and Grigorieff, 2015; Fromm et al., 2015; Bartesaghi, et al, 2014), resulting in a variety of radiation damage. As a matter of fact, to protect the charged carboxylate residues, a small amount of staining reagent, for example, uranyl formate, that can penetrate the molecular surface and bind to the carboxyl groups by the uranyl cation, may be introduced to reduce the radiation sensitivity, increase the tolerated dose, create more scattering from the heavy elements, and raise the image contrast of the organic soft-/biocompounds (shown in Figure 5) (Zhang et al., 2012b; Rames et al., 2014; Lei et al., 2018).

Whether the dose limit is related to the operating voltage (high tension) of cryo-EM has been discussed before (Martynowycz et al., 2021). Some people believed that a higher voltage (300–400 kV) in cryo-EM can reduce the radiation damage with a higher dose limit (Schmid et al., 1992; Yalcin et al., 2006). The study of the electron elastic scattering cross-section from atoms and molecules in the 1.0 keV to 1.0 MeV energy range (Yalcin et al., 2006) showed that the higher the electron velocity (shorter incident electron wavelength) (Henderson and Unwin, 1975), the fewer interactions it has with the atoms of the specimen, and thereby a higher incident electron dose is tolerated for the same radiation damage.

The studies showed that the scattering cross-section decreases under higher tension (Peng et al., 1996b; Colliex et al., 2006a), resulting in decreased percentage of incident electrons being scattered; in other words, the signal containing structural information is decreased, especially for a thin sample. Increasing the incident electron beam will not contribute to the structural information, but rather to the background, which reduces the image contrast. Cryo-EM imaging at low-voltage (e.g., 30–80 kV) can increase the image contrast by increasing the scattering cross-section, in which a higher percentage of incident electrons contribute to the structural images, although the total tolerated incident dose is reduced. Moreover, low-voltage imaging reduces knockout damage (Egerton, 2014). However, due to the increased scattering cross-section, the percentage of inelastic scattering and multiple scattering are also increased. Under this condition, super-thin specimens (e.g., ∼30 nm), energy filters, and special cameras for low accelerating voltages become necessary for cryo-EM imaging.

Although the lower-voltage can increase the image contrast, this benefit may be canceled by the need to use a lower defocus value for imaging. The need arises because the lower the operating voltage (longer weave length) of the electron beam (Carter and Williams, 2016), the more oscillations there are in the contrast transfer function (CTF). As a result, 1) a higher percentage of structural factors are permanently eliminated (at the frequencies that CTF crosses zero) and 2) the envelope function drops faster at higher frequency, which means that the weight of the high-resolution information is reduced faster (the SNR of the high-resolution portion decreases). Under this condition, a lower defocus is required during cryo-EM imaging, which adversely contributes to the image contrast. Thus, the benefit from the lower voltage is limited.

The success in imaging the soft-/ biomolecules by cryo-EM is due to the low-temperature of the specimen. Based on thermodynamics, cooling the sample slows down the motion of the molecules, reducing the secondary damage of radical diffusion in radiolysis. Thus, the radiation damage is also reduced under the same radiation condition. Cooling the specimens to liquid nitrogen temperature (∼77 K) significantly limits the radiation damage from the electron beam (Glaeser, 2008). The question about whether an even lower temperature such as liquid helium temperature (∼4 K) would be better for preventing radiation damage has been studied a decade ago, and the results were controversial.

Experiments supporting an even lower temperature include the following. The study of the dose tolerance of Caulobacter crescentus cells at liquid helium temperature versus liquid nitrogen temperature showed that bubbling within the cell is slower at helium temperature (Comolli and Downing, 2005). Studies of 2D protein crystals showed that higher-order diffraction spots fade slower under near liquid helium temperature than under near liquid nitrogen temperature, suggesting the lower temperature further inhibits the radiation damage (Stark et al., 1996; Fujiyoshi, 1998; Naydenova et al., 2022). Cryo-ET study showed a modest improvement in preventing the radiation damage at intermediate temperatures (25 or 42 K) as shown in Figure 6 (Bammes et al., 2010).

However, a study by cryo-EM SPA reconstructions of heavy-chain apoferritin at near liquid nitrogen (85 K) and near liquid helium (17 K) temperatures showed no difference between these two temperatures (Pfeil-Gardiner et al., 2019), which is consistent with the cryo-EM study of the bacteriorhodopsin 2D crystals at liquid helium temperature (Stark et al., 1996). The result is also consistent with another cryo-EM study at medium resolution (3–5 nm) under the temperature of ∼12 and ∼82 K (Iancu et al., 2006). In this study, cryo-ET reconstructions of various biological samples (e.g., mesoplasma florum and liposomes) under a total dose between 10 and 350 e⁻ Å⁻² showed worse results at ∼12 K than at ∼82 K (Iancu et al., 2006). Similarly, a study based on X-ray diffraction of biological samples showed no difference in the radiation damage at 50 K and at 5 K, although the damage can be reduced by a factor of ∼4 compared to 100 K (Meents et al., 2010).

The image contrast at near liquid helium temperature is lower than that at a relatively higher temperature (Iancu et al., 2006; Wright et al., 2006). The reason behind this phenomenon is unknown. An explanation could be the transformation of the amorphous ice at near liquid helium temperature into high-density amorphous (HAD) ice (Mishima et al., 1984), resulting in a reduction of the difference between the ice density and the protein density, which gives rise to a lower image contrast. Moreover, the collapse of vitreous water at a higher density state (Wright et al., 2006) can exert mechanical stresses on the protein or crystal, resulting in increased beam-induced motion with liquid helium cooling (Wright et al., 2006; Pfeil-Gardiner et al., 2019).

Radiation damage induced by the accumulated charges on the specimen during imaging has been discussed in a published article (Jiang, 2016). After the incident electron beam leaves the specimens, charges created by the interaction may accumulate on the specimen surface (Jiang, 2016), especially for insulating specimens. When the accumulated charge reaches a steady state, image may be degraded by the electrostatic perturbation. Moreover, the accumulated charges produce local electric fields, and the electrostatic force may induce the release of mechanical strain, which results in the movement of the frozen specimen or destroying the thin layer of vitrified ice (Glaeser, 2008). Considering that the charge build-up may perturb the electron beam and cause specimen motion, pre-exposure was often used to reduce charging before imaging. Based on the abovementioned factors, a higher electrical conductivity of the substrate (vitreous ice or supporting film) should contribute to the prevention of radiation damage by eliminating the accumulated charges (Taylor and Glaeser, 1974). The conductivity of the specimen can be improved by imbedding conductive nanomaterials (Pantelic et al., 2012; Kleinerman et al., 2015), such as graphene nanosheets and carbon nanotubes in the sample, or using a supporting film with better conductivity than amorphous carbon supporting film, such as gold holey/lacey film (Russo and Passmore, 2014a) and graphene (Russo and Passmore, 2014b).

Additionally, the increased specimen/substrate electrical conductivity can also increase its thermal conductivity based on the Wiedemann–Franz law in physics. The higher the specimen conductivity, the less electrons are accumulated, and less energy is deposited in the local illuminated area. A higher electrical conductivity also leads to a smaller increase in the local temperature, with less thermal expansion and less beam-induced deformation and specimen drift/motion. Although, large-scale beam-induced deformations (Frank et al., 2002; Zheng et al., 2017) and specimen motion (Li et al., 2013a) can be partially corrected by computer algorithms, such as the “unbending” method in cryo-crystallography (Henderson et al., 1986b; Gil et al., 2006), the motion-correction methods based on the correlation of DED frames (Li et al., 2013a; Bai et al., 2013; Zheng et al., 2017; Zivanov et al., 2019), or the focused electron tomography reconstruction (FETR) method to reduce the deformation influence to 3D cryo-ET reconstruction by reducing the reconstruction image size (Zhang and Ren, 2012), minimal beam-induced deformation during data collection will still have superior image quality. Thus, a specimen with higher conductivity is still useful.

A study of DNA damages and repairs in human cells showed that the detected DNA damages are completely repairable under 4 MeV electron beam irradiation by using a high dose-rate with laser-generated ultrashort pulses (Babayan et al., 2017). One may ask whether the cryo-EM sample has a self-healing capability from radiation damage. The capability of self-healing from radiation damage have been reported based on the studies of atomistic simulation (Bai et al., 2010) on hard materials (Zhang et al., 2020) as well as soft-materials (Perepelkin et al., 2019). In the process of self-healing, the radiolysis-generated radicals react to recover the original molecule spontaneously (Phaniendra et al., 2015). Thus, the capability of self-healing should depend on the recovery time and the dose-rate. One experimental strategy to benefit from self-healing is to use a pulsed electron beam, which can give the sample a timespan for molecular self-healing. The potential benefits of using a pulsed electron beam to mitigate radiation damage has been speculated for decades.

By allowing a short time for specimen self-healing, the higher tolerated dose may further increase the cryo-EM image contrast. However, one should note that the usage of ultrahigh dose pulsed imaging does not mean that a higher dose-rate must be used, a lower dose-rate can also be used. The choice of dose-rate could be limited by other parameters, such as limitations of the DED, in which case the dose-rate cannot be too low for enough information to be collected for frame alignment or motion correction, or higher than the frame readout rate (Sun et al., 2021). Nevertheless, whether cryo-EM samples have self-healing or recoverability capability from radiation damage is still an open question. We hope this information can be helpful to readers to design their own strategy for minimizing radiation damage by taking advantage of the prospective self-healing properties of their samples.

The physical size (thickness and diameter) of vitreous ice that spans across the holes in supporting carbon film should influence the radiation damage and dose limit (Zhang and Ren, 2012). However, its influence on the radiation damage has rarely been discussed. Two extreme cases can be used to understand why the physical dimension of the ice layer can influence the sample radiation damage. One extreme case is when the sample is imbedded in a bulk ice block, through which no incident electron can be transmitted. In this case, the bulk ice almost completely protects the protein from radiation damage. Although some electrons may carry the structural information of the proteins by elastic scattering from the sample, the transferred structure information is quickly destroyed by multiple scattering (Figure 2B), leaving no incident electrons for imaging. An opposite extreme case is when the ice layer is as thin as it is absent, in which case the electron beam interacts directly with the atoms of the sample. As a result, the sample can be easily damaged. In this case, all scattered electrons are 100% from the sample instead of partially from the sample and partially from the buffer/solution. In the normal case, when the sample is embedded in a thin layer of ice, the electrons scattered from the buffer/solution overlap with those from the sample, reducing the sample image contrast. Moreover, the thin sample also reduces multiscattering, which reduces the image blurring and increases the image sharpness.

In cryo-EM, the ice thickness is usually, in the abovementioned two cases, mostly within a range of 20–500 nm, in which the radiation damage is in between the abovementioned two extremes (shown in Figure 7) (Zhang et al., 2016). In other words, the radiation damage should be a function of ice thickness. The thinner the ice, the higher the image contrast we get, but with less tolerance to radiation damage. In contrast, the thicker the ice, the lower the image contrast we get, but with higher tolerance to radiation damage as more energy is needed to raise the ice temperature and to damage the imbedded proteins. Thin ice can be achieved by using lacey carbon film–coated grids instead of quantifoil grids, because various-sized holes (up to hundreds of microns) in lacey carbon grids provide a perfect match to the thinnest liquid film (i.e., largest size) that can be formed based on the buffer surface tension. To produce an even thinner ice film, a small amount of detergent (>0.1%) can be added into the sample solution to reduce the liquid surface tension if the detergent is compatible with the sample. Similarly, the area of the ice film should also be another function of radiation damage due to the higher heat capacity of the larger ice as described in the previous publications (shown in Figure 8) (Zhang and Ren, 2012; Lei et al., 2018; Wang et al., 2021b). Large area thin ice film can be achieved by using the lacey carbon film–coated TEM grid (Zhang and Ren, 2012; Segrest et al., 2015).

In cryo-ET imaging, some other techniques also help in preventing radiation damage by reducing the total dose while achieving the same level of image contrast. The implementation of these techniques can change the experimental imaging strategy. One technique is using dark-field imaging, in which the nonscattered electrons are excluded, leaving an image formed solely from the scattered electrons (Chiu and Glaeser, 1975). The dark-field image contrast has been reported to be about four times higher than that from bright field imaging (Krivanek et al., 2010; Tsai et al., 2016) with the same total dose. However, due to the difficulty in designing a beam stopper and installing it in the center of the objective aperture to block the central unscattered beam without blocking the scattering beam, the application of this method is limited. The second technique is phase-plate imaging, in which the phase of the electron wave is modulated by a thin film placed in the beam pathway to improve the imaging contrast by slightly changing the phase (Danev and Nagayama, 2001; Fukuda et al., 2009; Murata et al., 2010; Fukuda and Nagayama, 2012; Danev et al., 2014; Egerton, 2019; Malac et al., 2021). Recently, a high-intensity laser beam has been used to replace the thin film to modulate the phase of the electron wave. This laser phase-plate can avoid the weakness of the electrostatic charging or contamination presented in the thin-film phase-plate (Schwartz et al., 2019).

The third technique is the DED, which can improve the detective quantum efficiency (DQE) for high-contrast imaging (Li et al., 2013b; Fromm et al., 2015). Moreover, the fast DED allows movies taken with dose fractionation for the correction of specimen drift/motion, which reduces influence on the image resolution. The last frames in DED movies can also provide additional low-resolution information after high-resolution information is degraded (Cheng et al., 2015). This feature can improve the final 3D resolution and provide a precise tool to set the dose limit for preventing radiation damage. In other words, increasing numbers of initial frames (thus increasing total dose) can be used to determine the dose when the resolution starts to deteriorate. With the total dose less than this critical dose, the actual radiation damage will be less than that which can be measured by any averaging method or tool because many parameters, such as the flexibility of the molecule, image blurring (caused by drift/motion, charge, inelastic scattering, beam coherence, and defocus measurement/correction), and the accuracy of the image alignment and image noise, can also reduce the 3D reconstruction resolution as radiation damage does. A reliable measured resolution should be lower than the resolution limited by radiation damage.

Additionally, many computational algorithms have also been developed to enhance the image contrast (Sigworth, 2016; Fan et al., 2019a). For example, an edge-preserving smoothing–based multiscale image decomposition algorithm can detect the object against a high-noise background and enhance the object image contrast for 3D reconstruction (without averaging from different particles) of an individual small molecule (<100 kDa, CETP ∼53 kDa) (Wu et al., 2018). The missing-wedge correction software can allow us to use a small-tilt angle range for cryo-ET imaging, in which a higher dose can be used for each tilt image with the same total dose. The missing information in the missing-wedge can be restored by computational programs (Paavolainen et al., 2014; Yan et al., 2019; Moebel and Kervrann, 2020), such as the low-tilt tomography 3D reconstruction method (LoTToR) (shown in Figure 9) (Zhai et al., 2020). The application of the missing-wedge correction can also change the imaging strategy. For instance, this LoTToR allowed 3D reconstruction from a smaller tilt range with bigger tilt steps, in which the missing information can be restored by computer algorithms. As fewer number of tilt images are acquired, a higher dose can be used on each image and higher image contrast can be generated without increasing the total dose (Zhai et al., 2020). Other than the traditional algorithms, deep learning has been applied to cryo-EM image to increase the signal-to-noise ratio (Zhu et al., 2017) and to correct for the missing-wedge in 3D reconstruction (Chen et al., 2017; Ede, 2021; Liu et al., 2021) by training deep neural networks (DNNs) with cryo-ET data to improve structural interpretability in resolving lattice defects in immature HIV particles (Liu et al., 2021). The combination of the optimized data collection strategy and computational improvements, high-contrast images of single molecules can be expected in the future for high-resolution 3D reconstruction by individual-particle electron tomography (IPET) (Zhang and Ren, 2012). The resolution should be better than the current resolution, which only reveals the shape of the individual particle, such as the polyhedral shape of an individual human plasma intermediate-density lipoproteins shown in Figure 10 (Lei et al., 2019).