Measurement of charges and chemical bonding in a cryo-EM structure

The study investigates whether single-particle cryogenic electron microscopy (cryo-EM) at atomic and sub-atomic resolutions can measure not only atomic coordinates but also chemical information such as hydrogen positions, bond polarity, and charge states in proteins. Advances in electron beam coherence using a cold field emission (CFE) gun and monochromator have enabled reconstructions of apoferritin at near 1.2 Å resolution, where individual atoms appear and hydrogen signals can be detected. Unlike X-rays, electrons are scattered by the Coulomb potential, making them sensitive to charge states; electron scattering factors differ markedly between neutral and charged atoms at low spatial frequency. Prior charge measurements have been achieved by CBED and 3D electron diffraction (ED), but single-particle analysis (SPA) is influenced by factors such as CTF, beam tilts, and alignment errors. With improvements in hardware and processing, the authors hypothesize that SPA cryo-EM now has potential to extract chemical properties related to charges within macromolecules and to localize hydrogen atoms, addressing a long-standing challenge due to hydrogen’s weak X-ray/electron visibility.

The paper builds on prior demonstration of atomic-resolution SPA cryo-EM maps of apoferritin (1.22–1.25 Å) and observations of hydrogen signals. It contrasts electron and X-ray scattering: X-rays scatter from electrons; electrons scatter from Coulomb potential (nuclei plus electrons), yielding strong sensitivity to charge states and hydrogen. Charge distributions have been probed in inorganic crystals via CBED and in proteins/organics via rotation 3D ED, with partial charge modeling challenges and differences between neutral and charged scattering factors emphasized (Yonekura & Maki-Yonekura 2016; Yonekura et al. 2018). Ultra-high-resolution X-ray and neutron crystallography have mapped charge density in select protein crystals but require exceptional crystals. Previous 3D ED studies reported diminished density for anionic atoms at low resolution, and resolution truncation recovered densities, consistent with charged scattering behavior. The authors situate their work as extending SPA cryo-EM beyond coordinates to chemical information, addressing limitations in amplitude accuracy and charge factor assignment seen in previous methods.

Samples of mouse apoferritin were expressed in E. coli BL21-Gold(DE3), purified (heat treatment, ammonium sulfate precipitation, gel filtration), and prepared on Quantifoil holey carbon grids (R1.2/1.3, 200 mesh). Grids were plunge-frozen at 4°C and 100% humidity. Data were collected on a JEOL CRYO ARM 300 microscope at 300 kV with a cold field emission gun, in-column energy filter (20 eV slit), parallel illumination, specimen temperature ~93 K, and a Gatan K3 camera (super-resolution). Two datasets were acquired: Dataset A with JAFIS Tool-assisted coma/astigmatism correction and Dataset B without; both used SerialEM automation and multi-hole imaging. Dataset A: 12,114 stacks, CDS mode, dose rate 3.819 e−/s per physical pixel, 0.0585 s/frame, total 2.34 s exposure, defocus −0.5 to 1.0 µm. Dataset B: 2,173 stacks, non-CDS mode, dose rate 15.872 e−/s per pixel, 0.016 s/frame, total 0.8 s exposure, defocus 0.5 to 1.5 µm. Beam tilt and shift compensations were calibrated; axial coma-free alignment was verified; periodic gun flashing was set to 8 h. Image processing used RELION-3.1: removal of micrographs collected too close spatially (via distpos.py), motion correction (MotionCor2-like), dose weighting, CTF estimation (CTFFIND), manual rejection of poor micrographs, template-based autopicking from a low-pass filtered apoferritin reference, iterative 2D classification, extraction at 0.495 Å/pixel, 3D auto-refinement with octahedral symmetry, anisotropic magnification and beam-tilt refinement, per-micrograph CTF refinement, particle motion correction, and Ewald-sphere curvature correction. From 7,852 micrographs (A) and 1,122 (B), 2,235,864 and 311,583 particles were selected, yielding maps at 1.21 Å (A) and 1.49 Å (B) by gold-standard FSC. Dataset A particles were re-extracted at 0.396 Å/pixel and re-refined to a final 1.19 Å map with Ewald-sphere correction. Filtering of micrographs by refined scale and beam tilt (rlnGroupScaleCorrection < 0.5; x,y tilt deviation > 0.15 mrad) yielded 2,104,187 particles, with negligible resolution change (~10−3 Å). Rosenthal–Henderson plots were computed on particle subgroups. Model building/refinement: PDB 7KOD (mouse apoferritin) was fitted in UCSF Chimera, refined in ISOLDE and REFMAC5 without hydrogens. Unfiltered half maps were boxed (3203) and used to compute weighted Fourier Fo–Fc difference maps (Servalcat, CCP-EM) against a hydrogen-omitting model. Peaks for hydrogens and waters at ≥2σ or ≥4σ were detected (PEAKMAX, CCP4), then manually curated using riding positions. Refinement retained hydrogens on parent atoms with average B ≤ 20 Å2 and removed hydrogens on waters or multi-conformational residues; hydrogen positions were fixed; non-H atoms refined anisotropically, H atoms isotropically. Validation used CCP-EM; Q-scores computed; figures prepared in PyMOL and Chimera. RMSD1/2 for bond-length uncertainty was estimated via independent restrained MD refinements against each half map in ISOLDE and RMSD between resultant models. Dose-dependence test: Particles were re-extracted from early frames 1–2 (Frame 2), 1–3 (Frame 3), and 1–20 (Frame 20); reconstructions used full-frame alignment parameters. Resolutions were 1.37 Å (Frames 2 and 3) and 1.19 Å (Frame 20). Weighted difference maps, limited to 3.0 Å, were computed from these half maps and compared to the full frame (1–40) datasets.

- Achieved a 1.19 Å single-particle cryo-EM reconstruction of apoferritin (Dataset A; CFE beam, K3 super-resolution), with 2,104,187 particles contributing to the final map after filtering. Dataset B reached 1.49 Å. - Weighted Fo–Fc difference maps (hydrogen-omitting model) revealed widespread positive densities for hydrogen atoms, especially in the protein core. A total of 905 hydrogen atoms were identified, ~70% of all possible, at ≥2σ, with assignments guided by riding positions. - Clear identification of the amino and oxo termini of asparagine and glutamine side chains via presence/absence of hydrogen densities at their termini. - Several water molecules exhibited resolvable hydrogen densities; six waters showed two distinct hydrogen densities, forming hydrogen-bonded water clusters. - Negative densities in weighted difference maps were observed around acidic residues (Asp 171, Glu 17), consistent with negatively charged carboxylates. When truncating data to lower resolution (100–2.5 Å), strong negative densities localized to acidic side chains emerged and diminished progressively as lower-resolution shells were removed; at higher resolution (2.5–1.19 Å), these negative features were reduced, matching theoretical expectations that neutral and anionic scattering factors converge at ≲2.5 Å while diverging at low resolution. - Dose-frame analysis (frames 1–2, 1–3, 1–20, 1–40; total doses ~2, 3, 20, 40 e− Å−2) with maps limited to 3.0 Å showed that negative densities did not diminish in early, lower-dose frames, arguing against radiation damage as the source and supporting assignment to charge effects. - Simulations assigning partial negative charge (e.g., −0.3 e) to terminal oxygens of Asp/Glu reproduced the observed negative difference densities at low resolution, reinforcing charge attribution. - Hydrogen peak positions in the Coulomb potential map depended on bond type: average peak distances (Å) at higher density thresholds approached nuclear positions, with N–H shorter than C–H and O–H; at lower thresholds peaks appeared 0.1–0.3 Å longer, influenced by B-factors and peak height. Table 1 reported representative averages (mean (SD), counts): C–H alkyl ~1.15 (0.12) Å at >2σ and 1.09 (0.09) Å at >4σ (n=185, 39); aromatic C–H ~1.14 (0.15) and 1.10 (0.06) Å (n=43, 4); N–H ~1.06 (0.09) and 1.03 (0.07) Å (n=187, 83); O–H ~1.04 (0.07) and 1.07 (0.10) Å (n=13, 3). Comparisons to neutron/X-ray positions (nuclear/electron) showed consistency with electron sensitivity to nuclear potential and bond polarity. - Estimated standard uncertainties in bond lengths using RMSD1/2 were ~0.05–0.07 Å, sufficient to support statistically significant differences between average N–H and C–H bond peak positions (1–2σ level).

The findings demonstrate that high-resolution SPA cryo-EM maps can provide chemical information beyond atomic coordinates in proteins. The robust detection of hydrogen densities and their bond-type-dependent positions indicates sensitivity to bond polarity in Coulomb potential maps, consistent with electron scattering being dominated by nuclear charge. The emergence of strong negative difference densities on acidic side chains when emphasizing low-resolution data and their persistence across dose frames supports the interpretation that SPA cryo-EM can sense charge states within proteins. By leveraging weighted difference maps, resolution selection, and dose-frame analyses, the study addresses the challenge of distinguishing true charge-related signals from artifacts such as radiation damage or CTF/amplitude modifications. These results position SPA cryo-EM as a complementary approach to X-ray/neutron crystallography and 3D ED for probing hydrogen bonding networks, side-chain protonation states, and intra-protein charge distributions in non-crystalline samples, expanding the scope of Coulomb potential analysis in structural biology.

The study presents a sub-1.2 Å SPA cryo-EM structure of apoferritin that enables measurement of hydrogen positions, bond polarity, and signatures of negative charges on acidic residues through weighted difference mapping. Electrons provide relatively higher sensitivity to hydrogen and to charge states, and with careful control of phase errors and beam tilts, SPA can distinguish average hydrogen positions by bond type. Resolution-truncated and dose-frame series difference maps reveal negative densities attributable to deprotonated carboxylates, while hydrogen densities clarify amide side-chain termini and water hydrogen-bond networks. These results highlight SPA cryo-EM’s potential as a powerful tool for analyzing chemical bond properties and charge states in macromolecules. Future work should develop quantitative frameworks for partial charge assignment, refine treatment of charged scattering factors, and extend analyses to functional sites in diverse proteins.

The charge analysis is qualitative; while negative difference densities correlate with deprotonated acidic residues and theory, precise attribution of partial charges and use of appropriate charged scattering factors remain challenging. Interpretation of residual difference densities is complicated by distributions of partial charges and suboptimal charge modeling. Amplitude information in SPA is less accurate due to CTF effects and standard sharpening; although mitigated here, it may influence density magnitudes. Hydrogen peak statistics are limited for certain bond types (notably O–H and aromatic C–H), and peak distance distributions are noisy with dependence on B-factors and peak height. Radiation damage was assessed via dose frames, but subtle damage effects cannot be entirely excluded. Generalizability to proteins with greater flexibility or lower resolution maps may be limited.