On the damage done to the structure of the Thermoplasma acidophilum proteasome by electron radiation

Abstract

It has long been known that proteins are damaged when they are exposed to the electron beam in an electron microscope. Here we show that exposure to electrons under cryo‐EM conditions leads to a small change in the quaternary structure of the Thermoplasma acidophilum proteasome, and that backbones atoms belonging to the α‐helices in this molecule appear to be particular prone to chemical damage. A chemical mechanism is proposed for this damage. Both this local chemical effect and the more global quaternary structure effect appear to heterogenize samples leading to a radiation dose‐dependent degradation of the resolution of the EM maps obtained from this molecule.

Introduction

As a result of the development of devices that detect electrons directly, cryo‐electron microscopy (cryo‐EM) has recently emerged as the method of choice for determining the structures of large biological macromolecules.1, 2, 3 Even though these detectors are far more sensitive than their predecessors, it is still true that the electron doses required to produce images of macromolecules that have decent signal‐to‐noise ratios (SNRs) are high enough to damage them significantly. However, little is known about how electron radiation alters the structures of biological macromolecules chemically, and how those chemical alterations affect cryo‐EM images.

In addition to providing investigators the means to correct cryo‐EM images for the blurring caused by the tendency of molecules to move when exposed to the electron beam, direct electron detectors can also be used to assess the damage caused by the beam. The output produced by such a detector is a succession of short‐exposure, low‐SNR micrographs that can be summed to yield a single, high‐SNR micrograph. Even though the SNR is low in each frame in such a stack of micrographs, it is high enough so that the locations of the images of every molecule in the field of view can be determined using correlation function techniques, and corrections made for any changes in location that may have occurred during exposure before images are summed.4 It is a simple matter to use data of this sort to generate pairs of images for each of the molecules in any given micrograph, one that is the motion‐corrected average image obtained from the early frames in each stack, and second the corresponding average obtained from the late frames. If samples are being damaged by exposure to the electron beam, the three‐dimensional electrostatic potential (ESP) map obtained using the early images will differ from that extracted from the late images.

Here we report the results of analysis of this kind that was carried out on the cryo‐EM data used to determine the structure of the Thermoplasma acidophilum proteasome that is described in EMD‐3456 (Ref. 5). The early (T1) and the late (T2) ESP maps obtained from the data are not identical. The resolution of the T1 map is significantly higher than that of the T2 map (~ 2.6 Å vs. ~ 3.4 Å). In addition, exposure to electron radiation induces a small change in the quaternary structure of this α₇β₇β₇α₇ assembly, as well as a host of other, more local changes in its chemical structure that affect its α‐helices more than its β‐sheets. Thus, in this case, it may not be advisable to merge the early time images with the late time images. Furthermore, it is likely that the reason the resolution of the T2 map is inferior to that of the T1 map is that the population of molecules it represents is more heterogeneous chemically.

Methods

In order to carry out such an early‐late comparison, we divided the cryo‐EM images associated with EMD‐3456 into two sets.5 The first set, the T1 set, consists of molecular images obtained by summing the early time frames (i.e. Frames 3–12) from the stacks of all of the proteasome images used to generate the EMD‐3456. The second set, the T2 set, is similar to the first, but consists of images generated by summing the late time frames (i.e. Frames 11–24) from the same collection of images. The first two frames were not used because the severity of the drift that was evident in them.6

Image reconstructions were carried out using RELION 2.0 (Ref. 7). For each of the several reconstructions that were done, the images of individual particles obtained from these micrographs were arbitrarily assigned numbers, and then divided into two groups,8 one consisting of the even‐numbered images, and the other comprised of the odd‐numbered images. The total number of particle images used was 114,366. The even and odd sets of images were merged for the purposes of orientation determination so that the ESP maps obtained from the two half data sets would be sampled on the same grid in real‐space, and would be oriented the same way. In every other respect, the even and odd image sets were treated separately during the reconstruction process. Even though the particles that contributed images to the T1 and T2 data sets were exactly the same, the image reconstructions done with them were fully independent. As a result, the maps obtained initially from the T1 and T2 data sets differed in chirality, and were not oriented the same way with respect to their real‐space grids. After inverting one of the maps so that it would have the same hand as its mate, the rotation matrix that would best superimpose the T2 map on the T1 map was determined using a real‐space map alignment procedure, and then applied to the T2 map so that it could be resampled on the T1 map grid. These operations were carried out using Chimera.9 Structure factors were calculated from maps using Phenix,10 and fractional amplitude differences were analyzed, and isomorphous T2–T1 difference Fourier maps were calculated using CCP4 (Ref. 11).

Because the “even” and “odd” ESP half maps are on exactly the same scale, for any given value of (h, k, l), structure factors for the average map could be computed as follows:

$$\overrightarrow{F_{\mathit{Ave}}\left(h,k,l\right)}=\left(\left(,\overrightarrow{F_{\mathit{\text{Even}}}\left(h,k,l\right)}+\overrightarrow{F_{\mathit{Odd}}}\left(h,k,l\right),\right)\right)/2,$$ (1)

where $\overrightarrow{F_{\mathit{\text{Even}}}}$ and $\overrightarrow{F_{\mathit{Odd}}}$ (or $\overrightarrow{F_E}$ and $\overrightarrow{F_O}$) are the corresponding structure factors obtained from the two half maps. After merging, the amplitudes (F _Ave) and phases (α_Ave) of the averaged structure factors, standard uncertainties of amplitudes [σ_F(Ave)], and the cosine of phase differences (or phase errors, figure‐of‐merit or fom‐like quantity as defined in crystallography) are as follows:

$$F_{\mathit{Ave}}=\sqrt{{\left[F_{E}cos\left({\alpha }_E\right)+F_{O}cos\left({\alpha }_O\right)\right]}^2+{\left[F_{E}sin\left({\alpha }_E\right)+F_{O}sin\left({\alpha }_O\right)\right]}^2}/2$$ (2)

$$a_{\mathit{Ave}}={tan}^{-1}\frac{F_{E}sin\left({\alpha }_E\right)+F_{O}sin\left({\alpha }_O\right)}{F_{E}cos\left({\alpha }_E\right)+F_{O}cos\left({\alpha }_O\right)}$$ (3)

$${\sigma }_{F\left(\mathit{Ave}\right)}\approx \sqrt{{\left(F_E-F_{\mathit{Ave}}\right)}^2+{\left(F_O-F_{\mathit{Ave}}\right)}^2}$$ (4)

$$\mathit{fom}\approx cos\left({\alpha }_E-{\alpha }_O\right)$$ (5)

The Fourier shell correlation (FSC) method is regularly used to assess the consistency of pairs of ESP half maps, and within each resolution shell it is calculated as follows:

$$\mathit{FSC}=\frac{\sum \overrightarrow{F_E}\overrightarrow{F_O^{*}}}{\sqrt{\sum {‖F_E‖}^2\sum {‖F_O‖}^2}}=\frac{\sum F_E{F}_{O}cos\left({\alpha }_E-{\alpha }_O\right)}{\sqrt{\sum F_E^2\sum F_O^2}}$$ (6)

Note that the imaginary component in the numerator of this expression will be zero when the summation called for runs over an entire shell in reciprocal space because of Friedel's law. Structure factor differences can be divided into phase differences, which are best represented using the cosines of phase differences (5), and amplitude differences, which can be quantitated using either the crystallographic R _diff (7) or noise to signal ratio, R _(σ):

$$R_{\mathit{\text{diff}}}=\frac{\sum \left|‖F_E‖-‖F_O‖\right|}{\sum \left[\left(‖F_E‖+‖F_O‖\right)/2\right]},$$ (7)

$$R_{\left(\sigma \right)}=\frac{\sum {\sigma }_{\mathit{Ave}}}{\sum ‖F_{\mathit{Ave}}‖}.$$ (8)

Reciprocal of R_(σ) is the mean < SNR> value in each thin resolution shell or cumulative value to the given resolution as follows:

$$〈\mathit{SNR}〉=1/R_{\left(\sigma \right)}=\frac{\sum ‖F_{\mathit{Ave}}‖}{\sum {\sigma }_{\mathit{Ave}}}.$$ (9)

It is a simple matter to extend the definitions of FSC, R _diff, R(σ), and <fom> so that they can be used to compare sets of structure factors that have been obtained independently for the same structure provided the corresponding maps have been aligned and resampled on the same grid, and the structure factors scaled appropriately.

Six different kinds of maps were examined in the course of this study: (1) ESP maps that had been sharpened by RELION in a post‐refinement step, (2) unsharpened ESP maps, (3) unsharpened charge density (CD) maps, and (4, 5, 6) difference Fourier maps of all three kinds. None of the conclusions reported below are sensitive to the kind of map chosen for examination. Difference Fourier maps were computed using Fourier amplitudes that had been scaled to eliminate features that might otherwise be caused by differences in resolution between T1 and T2 maps. The experimental CD maps shown were obtained by applying the negative Laplacian operation to the original unsharpened ESP maps.12 Finally, in addition to examining T1 and T2 ESP maps, a third kind of ESP map was also considered, namely a map obtained by processing the images produced when Frames 3–24 from every stack were merged. That map is referred to as the T12 ESP map.

Results

Resolution of the T1 ESP map is superior to that of the T2 ESP map

The resolutions of the unsharpened versions of the T1 and T2 ESP maps were estimated using the standard FSC method, with and without a mask (Table 1 and Fig. 1).13 Using a mask,7 the resolution of the T1 map was found to be ~ 2.56 Å and that of the T2 map was ~ 3.37 Å. The resolutions of both estimated without a mask were somewhat lower: 2.91 Å and 3.50 Å, respectively (Table 1 and Fig. 1). Finally, the resolution of the merged, T12 ESP map was estimated to 2.46 Å with a mask, which is comparable to the resolution of the published map of the proteasome that was obtained from the same micrographs.5 Similar differences in resolution estimates were obtained when R _diff or <fom> were used as the criterion for determining it [Fig. 1(B) and (C)]. An important reason why the T1 map is higher in resolution than the T2 map is that its SNR is higher [Fig. 1(E) and (F)].

Table 1.

Data Statistics Between Two‐Half Maps for Each of T1, T2, and T12 ESP Maps and Netween T1 and T2 Maps

Resolution (Å) or SNR	T1	T2	T12	T1/T2
FSC‐based resolution	2.91	3.50	2.87	3.30
Cumulative SNR	4.7	4.1	5.4
Masked FSC‐based resolution	2.56	3.36	2.46
Cumulative SNR	3.5	3.8	3.9

Figure 1.

Reciprocal space statistics for T1, T2, and merged T12 maps. (A) Fourier shell correlation (FSC) as a function of reciprocal resolution (Å⁻¹) for the T1 (even, and odd or E, O) (black), T2 (E,O) (blue) maps, T12 (E,O) (green), and cross FSC between T1 and T2 maps (magenta). Gold standard FSC value is shown in forest‐green dashed line. (B) For T1 map, other three statistics are shown, R(σ) (red), R(diff) (green), <fom > (blue) in addition to FSC T1(E,O) (black). Wilson limit is shown in magenta dashed line. (C) The same as (B) but for T2 map. (D) The same as (B) and (C) but for cross statistics between T1 and T2 maps. (E) SNR in thin resolution shell (solid lines) and cumulative resolution (dashed lines) for T1 (black), T2 (green), T12 (orange) maps as a function of reciprocal resolution. When T1 structure factor amplitudes are sharpened with ΔB = −95 Ų, cumulative <SNR> values decrease (red dashed line). (F) Wilson‐like plot of logarithm mean amplitudes (black and cyan lines) and standard uncertainties (blue and green lines) in thin resolution shell for T1 (black and blue), T2 (green and cyan), and T12 (orange and magenta) maps as well as those for the sharpened T1 map (dashed lines) with ΔB = −95 Ų as a function of resolution squares.

When the cross‐FSC was calculated between aligned T1 and T2 maps, its values largely split the difference between those of the T1 and T2 maps at most resolutions, as expected, but with two exceptions [Fig. 1(A)]. The exceptions are (1) the cross‐FSC values between T1 and T2 maps at resolution ~ 6 Å are smaller than FSC values for either T1 or T2 map, and (2) the cross‐FSC values in the highest resolution shells are mainly determined by the difference in the limiting resolution of the two maps. When the two halves of either T1 or T2 map are compared, one finds that the average value of the differences of both phases and amplitudes approach zero asymptotically with decreasing resolution, as one would anticipate [Fig. 1(B) and (C)]. However, the average phase difference between the T1 and T2 maps approaches 18° at low resolution, not zero [Fig. 1(D)]. Similarly, at low resolution, the R _diff values between the two maps are ~ 0.25–0.35 rather than zero [Fig. 1(D)]. Both observations suggest that exposure to electron radiation induces a global change in the structure of the T. acidophilum proteasome.

At the highest resolution shell, where FSC drops to the “gold standard” limit (FSC ~ 0.143), <fom> approaches FSC (also ~ 0.143), and R _diff goes asymptotically to 0.587, the value theory expects one should obtain when non‐centrosymmetric structures are compared that are completely unrelated (Fig. 1).14, 15 These observations show that measurement errors affect the accuracies of both the amplitudes and phases of structure factors the same way, and that the estimates of the information content of sets of structural data derived from the three quantities of FSC, <fom>, and R_diff are similar, as one would expect, given how closely they are related.15

The programs used to produce ESP maps routinely sharpen the images they generate by applying a negative B‐factor correction to the amplitudes of their Fourier coefficients, and then recovering the sharpened image by inverse Fourier transformation (a procedure which is known as post‐refinement). The value of that correction is usually chosen so that plots of ln[<F(s)_corrected>] versus s ² will be as close to flat as possible. The sharpening B‐factor for the sharpened T1 map shown here was −96 Ų, and for the T2 map, it was −130 Ų, consistent with their difference in resolution. To minimize the effects that Fourier series termination is likely to have on maps that have been sharpened this way, the amplitudes are also multiplied by resolution‐dependent <fom>−like correction factors that diminish in magnitude as the resolution limit is approached. When the same sharpening was applied to two half maps, it reduced the cumulative SNR value, and increased the overall fractional amplitude difference between them [Fig. 1(E) and (F)].

Comparison of sharpened T1 and T2 ESP maps

In Figures 2 and S1, the sharpened T1 and T2 ESP maps for the entire α₇β₇β₇α₇ proteasome assembly are compared at two different contour levels (Figs. 2 and S1). At high contour levels, the long rod‐like features that represent β‐strands are much more visible in both the α and β subunits of the T2 map than they are in the T1 map, but the visibility of detailed features of α‐helices is slightly better in the T1 map. This observation suggests that electron radiation may affect the appearance of α‐helices more than it does the appearance of β‐strands.

Figure 2.

Comparison of T1 (left) and T2 (right) ESP functions after map sharpening in post‐refinement step. (A) Histograms to define two contouring levels (low and high) used in (B, C). See Figure S1 for stereodiagrams of the maps for each subunit.

In addition, short, narrow, rod‐like features 3–5 voxels long that are perfectly aligned along the three principal axes are clearly visible everywhere in the sharpened T1 ESP map [Fig. 2(C)], particularly at the highest positive and negative contour levels (data not shown), but not the corresponding T2 map. These high‐frequency features are an artifact of over‐sharpening.

The quaternary structure of the T. acidophilum proteasome is affected by radiation

Structures for the proteasome were produced by rigid‐body fitting of the X‐ray structures of its α and β subunits into the T1 and T2 maps. The two structures that emerged were then superimposed using a least‐squares method, and once this was done it became evident that their quaternary structures are not identical. The same result was obtained when this analysis was repeated using both unsharpened ESP and CD maps. The difference can be described as a rotation of the α subunit by ~ 1.0° around an axis that is perpendicular to the seven‐fold symmetry axis of the structure, accompanied by a translation of Δt _axis = 0.27 Å along that axis (Fig. 3). This axis is tangential to the circle that connects the centers of mass of the seven α subunits within the α₇ ring, and it comes close to passing through the ring's center of mass [Fig. 3(B) and (C)]. The result of this motion is that in the T2 structure, the pore region of the α subunit, including Y126, is ~ 0.50 Å further away from the central chamber of the proteasome than it is in the T1 structure, but its outer edge is closer.

Figure 3.

Electron radiation‐induced internal subunit rotations in T2 relative T1. (A) Rotation of T2 α subunit (salmon) relative T1 α‐subunit (cyan) of ~ 0.99° around an axis (a large blue‐to‐red arrow) with screw translation of 0.27 Å along it. Small arrows indicate the direction of rotation. The pore residue Y126 is also shown. (B) Rotation of α subunit in reference of the α₇ ring with the pore residues rotating away from the central chamber and its outer edge in opposite direction, viewed from inside to outside of the central chamber along the seven‐fold axis. (C) The same as (B) but viewed from outside to inside along the seven‐fold axis. (D) Rotation of T2 β subunit (salmon) relative to T1 β subunit (cyan) of 0.61° with screw translation of 0.07 Å along an axis located in a plane bisecting the two halves of the proteasome. (E) Rotation of β subunit in reference of rotation axes for α subunits in the α₇ ring. (F) Comparison of rotation axes for β subunits with those of α subunits.

To bring the T2 structure into perfect alignment with the T1 structure, a much smaller screw displacement of its β subunits is also required [~ 0.61°; Δt _axis = 0.07 Å, Fig. 3(D–F)]. The axes for these β subunit rotation/displacements are nearly anti‐parallel to the corresponding axes for the α subunits, and they are perpendicular to the seven‐fold axis of the complex [Fig. 3(F)]. They lie in the plane that bisect the two halves of the proteasome. Because these two displacements rotate the two subunits in opposite directions around nearly‐parallel axes, the α‐β interactions in the two structures are about the same.

T2–T1 difference Fourier maps

Because the sharpened T1 and T2 maps differ both in resolution and <SNR>, and were sharpened using different ΔB values, side‐by‐side comparisons of the two sharpened maps were unlikely to be informative. In order to reduce the impact of the systematic differences between the two maps, their structure factor amplitudes were scaled properly, the resolution of both maps set was set at ~ 3.4 Å, which is a conservative estimate of the resolution of the T2 map, and the two maps were then recalculated (Fig. S2). Differences between the T1 and T2 structures were identified using difference Fourier maps computed using the same scaled data sets. Given the differences in quaternary structural that exist between the T1 and T2 versions of the proteasome, the difference Fourier maps examined were computed using the structure factors obtained from two maps after either α or β subunits had been individually aligned. Analyses of the same kind were also done using unsharpened ESP and CD maps (Fig. S3).

Side‐by‐side comparisons of ESP maps will be misleading if the maps are not contoured in a comparable manner. It would make no sense to contour the sharpened T1 and T2 maps at, say, 3.0 σ above zero because they were not sharpened by the same amount, and the signal‐to‐noise ratios of the data sets from which they derive are not the same. This problem was addressed in two ways. First, histograms of the ESP values in both maps were constructed so that their ESP distributions could be compared [Fig. 2(A)], and the contour levels for the two maps were set so that the contours used would include the same fraction of the total positive density in both (Fig. S4). Second, difference Fourier maps computed using scaled data were also used to pinpoint regions where the two maps differ.

Difference Fourier maps that compare the sharpened T2 and T1 ESP maps were computed after their maps had been aligned on either their α or β subunits, and their structure factor amplitudes had been scaled. These maps have significant negative difference peaks that superimpose on the backbones of all of their α‐helices, and that have maximal values approximately every two residues (Figs. 4 and S5–S7). The same results are observed in the unsharped difference CD maps (Figs. 5 and 6). In addition, there are negative difference features on the side chains of positively charged residues (i.e., Arg and Lys), and hydrophobic residues, most noticeably those containing tertiary C atoms such as Cγ atoms of Leu side chains (Fig. 4). No negative difference features were observed on side chains of any negatively charged residue such as Asp or Glu (Fig. 4). The difference maps obtained are much quieter in the vicinity of β‐strands (Fig. 5).

Figure 4.

Stereodiagrams of sharpened T2 (right panels) and T1 (left panels) ESP maps contoured at +3.0σ (sky blue, and forest green) (A) and their subunit‐aligned T2‐T1 difference ESP Fourier maps (B) for an α‐helix in α subunit contoured at ±6σ (right panels, cyan and magenta) and ± 5σ (left panels). Two sharpened maps have already been scaled. See Figure S5 for the α₇β₇β₇α₇ assembly‐aligned T2–T1 ESP maps at approximately orthogonal views contoured at ±6σ (green and red), Figure S6 for a second α‐helix in α subunit, and S7 for a pair of α‐helices in β subunit in subunit‐aligned difference Fourier maps.

Figure 5.

Difference Fourier maps for unsharpened CD functions after subunit‐realignment for α (A) and β (B) subunit in stereodiagram at a positive (green) and negative (red) contouring levels as shown in map histogram for each subunits. Subunits are rainbow colored from N‐terminus to C‐terminus in blue to red.

Figure 6.

Close‐up views of unsharpened difference CD Fourier maps for two helices from α subunit (A, B) and from β subunit (C, D) with three gradually decreasing contour levels.

Discussion

Effects of electron dose on the resolution of proteasome ESP maps

The data described above demonstrate that the resolution of the ESP images obtained from frozen samples of T. acidophilum proteasomes degrade as the electron dose they are exposed to increases. Two processes contribute to that dose‐dependent loss of resolution: a change of the quaternary structure that is likely to be specific to the proteosome, and changes in the chemical structure of the molecule that are certain to be the same as those that occur in any other protein that is examined in the electron microscope. It is not clear whether for any given protesome molecule the conformational change described above is progressive, or all‐or‐none. Be that as it may, a population of proteasome molecules that has been exposed to electron radiation will be more heterogeneous conformationally at the quaternary level than one that has not, and that increase in heterogeneity is bound to make the T2 map inferior to the T1 map both in resolution and SNR. It also explains why the structures obtained from the two maps do not superimpose well.

It is not obvious why the quaternary structure of the proteasome is radiation‐sensitive, but it would not be surprising if it were caused by the changes in the chemical structure of the molecule that are responsible for the local changes in ESP evident in the difference maps shown above. As noted above, there are significant negative features in these difference maps that are associated with the backbones of nearly all the α‐helices in both the α and β subunits, but similar features are largely absent from the backbones of β‐strands. These negative features cannot be ascribed to a loss of backbone atoms because any chemical reaction that degraded a protein's backbone would surely affect its side chains too, and there is no evidence of side‐chain damage. If, on the other hand, one of the conjugated (NH‐C=O) bonds in the peptide backbone were to absorb an electron, a negative feature would appear in corresponding (T2–T1) difference Fourier maps. The hydrated electrons that form when electrons are scattered inelastically by water molecules might provide the reducing agent required. A similar hypothesis was recently advanced to explain a loss of ESP contrast in images of gold clusters, although the nature of charges is different.16

A kind of radiation damage we expected to see, but did not, was the decarboxylation of aspartic and glutamic acid side chains, which is frequently observed in protein crystals that have been exposed to X‐rays.17, 18, 19, 20, 21 However, if hydrated electrons are indeed responsible for most of the radiation damage suffered by macromolecules during electron microscopy, it would be surprising if carboxylate groups were much affected because carboxylate groups are negatively charged. It is more likely that the lack of visibility of these groups in EM maps, which has often been commented upon, is attributable to the effects that their negative charges have on local ESP distributions. Their high mobility also tends to wash out their contribution to ESP maps.22

Whatever the chemistry responsible for these local changes in ESP, they too will result in a dose‐dependent increase in the heterogeneity of protein samples that degrades the resolution and signal to noise ratios of EM images.

A proposal for mechanism of the reduction of peptide groups by hydrated electrons

The difference maps described above suggest that even though the chemical effects that electron radiation has on proteins are largely backbone‐associated, they do not alter backbone connectivity. Further, the negative sign of the most prominent features in these difference maps indicates that the reactions in question might be reductions. It is also the case that while the exposure of protein solutions to electron radiation leads to the production of H₂, the irradiation of water with electrons does not.23, 24 What kinds of reactions might account for these observations? We are far from being able to answer this question definitively, but think that reaction schemes like the one shown in Figure 7 are worth considering, which involves both highly reductive, mobile hydrated electrons generated by electron inelastic scattering events and immobile free radical intermediates.

Figure 7.

Proposed chemical reaction mechanism involving electron radiation‐induced reduction of protein backbone unit. Upon absorption of each hydrated electron, the conjugated π‐system, notable the C=O bond, becomes a π‐radical anion (II and IV). In the acidic conditions, external protons may make the peptide unit to be catalytic for H₂ generation during which the Cα center would become racemized.

Each peptide unit has a low energy π*‐orbital system that should be a better acceptor of electrons than water (Fig. 7). The product of the addition of an electron to this system can be schematized as a π‐radical anion, which should have a reduced ESP value due to the added electron (II and III in Fig. 7). The addition of a second electron to an adjacent carbonyl group will generate a second π‐radical anion that could undergo the electronic rearrangement shown in IV. That rearrangement would generate one H₂ molecule and an anionic peptide intermediate (V) capable of accepting a pair of protons from solution. Protonation of V to generate species VI (it is noted that tertiary Cα atom can exhibit some acidicity under certain conditions25, 26, 27, 28), followed by tautomerization would yield species VII. This mechanism predicts that the rate of H₂ generation should be proportional to the square of the electron dose, and that at moderate doses intermediates like II and III might accumulate, accounting for a radiation‐induced buildup of negative ESP in the vicinity of the backbones. We cannot yet rationalize why the β structure seems less sensitive to this kind of damage than the α structure.

Interestingly, the protein modifications caused by exposure to X‐rays have been studied extensively for over a century using a variety of chemical and physical methods.29, 30, 31, 32, 33 The predominant products detected are O insertions at tertiary C atoms that are bonded to H atoms, e.g., backbone Cα atoms, the Cβ of Val and Ile side chains, and the Cγ of Leu side chains. These products are favored because of the high stability of tertiary C free radicals and the strong oxidizing power of mobile, cascade‐amplifiable hydroxyl radicals. After O‐insertion reactions, the sp ³ tertiary C atoms remain sp ³ [Ref. 34]. H₂ molecules released during X‐ray radiation are likely to derive mainly from H₂O molecules.35 Therefore, the chemistry responsible for the radiation damage caused by X‐rays is not the same as the chemistry that leads to radiation damage caused by electrons.

Conflict of interest

Authors declare no conflict of interest in publishing the results of this study.

Supporting information

Appendix S1: Supplementary Material