The key insight of the study, led by Liming Zheng together with corresponding authors Hailin Peng and Hong-Wei Wang, is that the difficulty of ice control does not begin at the moment of freezing. It begins earlier, with the physics of the thin liquid film itself. Theoretical calculations show that once a liquid film thins below roughly 100 nanometers, the regime required for high-resolution cryo-EM, it becomes intrinsically unstable. The free energy of such a film depends on long-range van der Waals forces, which grow stronger as the film gets thinner. Thinner regions therefore experience a higher attractive force, prompting liquid to migrate spontaneously toward adjacent thicker areas. The film becomes increasingly prone to rupture, breaking into microscopic droplets in an unpredictable way. This instability, the authors argue, is the previously underappreciated determinant that has made ice thickness in cryo-EM a trial-and-error affair.

The team’s answer draws on the emerging physics of liquids under confinement. When water is squeezed into spaces only nanometers across, its molecular migration slows dramatically, with diffusion coefficients dropping two to eight orders of magnitude below bulk values, and the confining structure can spatially mold the liquid even at thicknesses of a few atomic layers. The researchers exploited a quirk of atomically thin graphene: when a graphene sheet is draped over a hole in a supporting film, van der Waals attraction pulls it against the hole’s sidewalls, causing the membrane to sag inward like a trampoline. This sagging creates a shallow, nanometer-scale basin, which the team calls a graphene reservoir, whose depth is set by the geometry of the hole.

Crucially, the depth of that basin can be engineered. By fabricating holes with smooth, rounded edges using industry-compatible deep-ultraviolet lithography, the researchers achieved highly uniform sagging depths of 31 plus or minus 2 nanometers in holes of 1,200-nanometer diameter, as measured by atomic force microscopy. Varying the hole diameter from 600 to 2,000 nanometers tuned the sagging depth from 14 plus or minus 1 nanometer up to 40 plus or minus 3 nanometers, in agreement with a theoretical model in which sagging depth scales with hole diameter, edge curvature radius, graphene-edge adhesion energy and the two-dimensional Young’s modulus of graphene. The strategy also worked on the backsides of several commercially available grids, including ANTcryo, C-flat, Quantifoil and UltrAufoil, meaning labs can adopt it without abandoning their existing consumables. Batch production of more than 500 homemade grids on a four-inch wafer demonstrated scalability.

The payoff shows up at blotting time. In conventional cryo-EM sample preparation, filter paper wicks away excess liquid for one to two seconds before the grid is plunged into liquid ethane, and the outcome is notoriously variable. On graphene reservoir grids, the confined liquid film remained stable even after 10 seconds of blotting, five to ten times longer than typical practice. Cryo-electron tomography revealed that the resulting vitreous ice matched the reservoir depth, about 30 nanometers, and that ice thickness plateaued at roughly 27 plus or minus 3 nanometers regardless of whether blotting lasted 10, 12, 15 or 20 seconds. Independent replicate experiments produced statistically comparable results, and overview atlas images showed uniform ice across entire grids. Numerical simulations explained why: liquid in thin peripheral regions flows faster than liquid inside the deeper reservoirs, so the reservoirs actively draw liquid in rather than losing it, inverting the usual instability.

Because the plateau thickness tracks the reservoir depth, ice thickness becomes a design parameter rather than a lottery. Grids with 600-, 800-, 1,200- and 2,000-nanometer holes yielded controllable ice thicknesses of 15 plus or minus 1, 20 plus or minus 2, 30 plus or minus 3 and 42 plus or minus 3 nanometers, closely matching the corresponding sagging depths. Thinner, more uniform ice translated directly into better images, with reduced B factors and decreased particle motion. By contrast, conventional grids with sharply protruding hole edges suffered liquid film rupture under the same 10-second blot, producing uneven ice, protein aggregation and denaturation, and a markedly lower fraction of usable particles for reconstruction.

The technique also cracks a long-standing problem with graphene sandwich specimens, in which a second graphene sheet is laid over the sample to encapsulate the liquid between two atomically thin membranes. Sandwiches promise to shield proteins from the damaging air-water interface, but conventional versions suffer from poorly controlled, heterogeneous ice thickness. In the new graphene reservoir sandwich, the reservoirs provide uniform dimensions, periodic arrangement and thermodynamically favorable liquid migration, so the encapsulated ice is far more homogeneous. Cryo-ET measurements showed a narrow thickness distribution centered at 30 nanometers, versus 52 plus or minus 18 nanometers for conventional sandwiches, and the thickness could be tuned downward to 19 plus or minus 2 nanometers simply by using trilayer instead of monolayer graphene, since the stiffer membrane sags less. Encapsulation efficiency exceeded 90 percent across 2,219 grid squares, compared with about 67 percent for the conventional method, and particles in the reservoir sandwich showed reduced beam-induced motion and faster convergence.

The structural biology results demonstrate what this control buys. The Ll.LtrB group II intron, an RNA molecule that typically fails on conventional grids because particles refuse to enter holes or aggregate, yielded a 2.7-angstrom reconstruction from graphene reservoir grids, with high contrast, well-distributed orientations and clearly resolved side chains. The SARS-CoV-2 spike protein encapsulated in graphene reservoir sandwiches reconstructed to 2.9 angstroms. Most strikingly, streptavidin, a small 52-kilodalton protein whose low contrast makes it a nightmare for cryo-EM, was reconstructed at 2.46 angstroms in the reservoir sandwich, with individual amino acid side chains identifiable in the density map. Small proteins under 100 kilodaltons are frequent drug-discovery targets, and the authors note that uniform thin ice is precisely what such targets demand.

The implications extend beyond single-particle imaging. The authors anticipate that graphene reservoir sandwiches will benefit laser flash-melting approaches that overcome preferred orientation, time-resolved cryo-EM, and in situ liquid electron microscopy, where precisely manipulated confined liquids could reveal biological and chemical processes in their natively hydrated states. The method also meshes with spray-based sample preparation: after droplets are deposited on the grid, liquid spontaneously spreads into the reservoirs, producing uniform ice where conventional sprayed droplets leave only thick, uneven patches. With fabrication procedures compatible with standard blotting workflows and published step-by-step protocols, the barrier to adoption is low. After decades in which ice thickness was the uncontrolled variable at the heart of cryo-EM, a shallow dip in a one-atom-thick membrane may finally have tamed it.

Subject of Research: Control of vitreous ice thickness in cryo-EM sample preparation using graphene liquid reservoirs

Article Title: Control of ice thickness in cryo-EM via confinement

Article References: Zheng, L., Song, J., Zhao, X., Cao, J., Xu, J., Wang, Z., Zhang, C., Sun, W., Chen, B., Gao, X., Liu, H., Yang, J., Xu, Y., Sun, L., Dai, Z., Wei, X., Liu, N., Peng, H., & Wang, H.-W. (2026). Control of ice thickness in cryo-EM via confinement. Nature Methods. https://doi.org/10.1038/s41592-026-03244-1

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03244-1

Keywords: cryo-EM, graphene, ice thickness, vitrification, sample preparation, structural biology, nanofluidics, van der Waals forces, graphene sandwich, single-particle analysis, proteins, RNA