henhaobio
Sign in

NatureBiophysicsPeer-reviewed journal article

Complex Water Networks Visualized through 2.2-2.3 Angstrom Cryogenic Electron Microscopy of RNA

Rachael C. KretschiD, Shanshan LiiD, Grigore D PintilieiD, Michael Z. PaloiD, David Andrew CaseiD, Rhiju DasiD, Kaiming ZhangiD, Wah ChiuiD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

The stability and function of biomolecules are directly influenced by their myriad interactions with water. In this study, we investigated water through cryogenic electron microscopy (cryo-EM) on a highly solvated molecule, the Tetrahymena ribozyme, determined at 2.2 and 2.3 [A] resolutions. By employing segmentation-guided water and ion modeling (SWIM), an approach combining resolvability and chemical parameters, we automatically modeled and cross-validated water molecules and Mg2+ ions in the ribozyme core, revealing the extensive involvement of water in mediating RNA non-canonical interactions. Unexpectedly, in regions where SWIM does not model ordered water, we observed highly similar densities in both cryo-EM maps. In many of these regions, the cryo-EM densities superimpose with complex water networks predicted by molecular dynamics (MD), supporting their assignment as water and suggesting a biophysical explanation for their elusiveness to conventional atomic coordinate modeling. Our study demonstrates an approach to unveil both rigid and flexible waters that surround biomolecules through cryo-EM map densities, statistical and chemical metrics, and MD simulations.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (24 Jan 2025) · CC BY 4.0

Two Tetrahymena ribozyme cryo-EM maps (2.2 and 2.3 Å) reveal cross-validated ordered waters and Mg2+ ions, plus reproducible diffuse solvent density resembling molecular dynamics water networks.

RNA cores are heavily solvated, yet RNA-only cryo-EM had not reached the resolution to resolve water (only ions were modeled in prior catalytic-domain structures), and molecular dynamics (MD) hydration predictions rest on force fields of uncertain accuracy. These maps, the highest resolution reported for RNA-only cryo-EM, allow cross-validated water modeling and an experimental benchmark for MD water and ion models.

Key results

  1. ShownA stricter SWIM (5σ density plus Q-score above 0.7 in the full map and both half-maps) automatically placed waters and Mg2+ ions in the first solvent shell, and about half of the waters were reproduced in both independent maps.

    255 and 281 waters plus 47 and 47 Mg2+ ions in the 2.2 and 2.3 Å maps; 134 consensus waters (53% and 48% of each model's waters); consensus waters had significantly higher Q-scores; only 0.6% of ~89,000 random solvent-shell positions passed all criteria. · Automated modeling of water with SWIM; Segment-guided Water and Ion Modeling (SWIM); Figure legends

  2. ShownSolvent-shell density is highly reproducible across the two maps and goes well beyond what SWIM models. MD with TIP4P-D water reproduces some diffuse networks (e.g., along P6a and between P9.1 and P7) but has a large accuracy gap, performing notably poorly in the catalytic core.

    Against 2.2 Å map solvent-shell voxels above 3σ, the 2.3 Å map gave 71% recall, 63% precision, 5.6% false-positive rate; MD at 71% recall gave 35% false positives, 22% precision. SWIM waters: 96% precision at 10% recall (MD 68%), plateau near 25% recall. MD normalized AU-PRC 0.58 (tetraloop receptor) vs 0.34 (catalytic core). · Complex and diffuse water networks; Density comparison; Figure legends

  3. SuggestedWaters modeled in only one map mostly pass the density and resolvability criteria in the other map, so they are unlikely to be noise. Their peak density is lower than that of consensus waters, which MD attributes partly to lower occupancy and greater positional spread, causing inconsistent SWIM calls.

    Placed into the other map, 84% of waters kept >5σ density and Q-score >0.7 (1.2% of random solvent-shell positions); 75% also passed half-map criteria (0.6% random); 30% of non-consensus waters had a shifted nearby peak. Cryo-EM water sites averaged 90 M predicted MD water vs 55 M bulk; 10% fell below bulk. · Exploring non-consensus waters; Figure legends

  4. ShownOrdered waters directly bridge RNA atoms at tertiary contacts throughout the conserved core without needing site-bound ions. Hydration per nucleotide at the substrate-free active site matches that of other regions.

    Previously unmodeled waters bridge C255 and G272 (P3/P6a contact), the G254 and U273 sugars, the C216 sugar and U106 phosphate, and the G215-U258 wobble minor groove; at least 5 waters form a network at the P4/P5c/P14 junction; two waters mediate A306 contact with the A308-U267 pair. · Automated modeling of water with SWIM; Discussion

  5. ShownConformational classification of one dataset gave 2.2 and 2.3 Å maps whose well-resolved cores nearly coincide and that differ mainly in flexible peripheral domains. Per-nucleotide resolvability tracks MD-estimated flexibility.

    18,365 movies; 1,181,331 particles split by 3D variability into 708,006 and 473,325; RMSD 0.59 Å in well-resolved regions; differences concentrated in P9.2 and P6; significant negative correlation of nucleotide Q-score with RMSF from 30 MD runs of 400 ns. · Structure determination; Structure quality assessment; Exploring non-consensus waters; Image processing

Methods, data and reproducibility

Single-particle cryo-EM of the apo L-21 ScaI ribozyme gave two maps (2.2 and 2.3 Å). A stricter SWIM (segment-guided water and ion modeling) placed waters and Mg2+ ions, cross-validated between maps, and solvent-shell density was compared with 30 explicit-solvent 400 ns MD runs across three force fields.

  • Cryo-EM movies of apo Tetrahymena L-21 ScaI ribozyme (no P1 substrate or ωG) in 50 mM Na-HEPES pH 8.0 with 10 mM MgCl2 · Tetrahymena (ribozyme RNA) · n = 18,365 movie stacks; 3,804,753 picked particles · EMPIAR 11844
  • Two final reconstructions separated by cryoSPARC 3D Variability Analysis (2.2 Å and 2.3 Å maps) · Tetrahymena (ribozyme RNA) · n = 708,006 and 473,325 particles · EMD-42499 (2.2 Å); EMD-42498 (2.3 Å)
  • Refined atomic models carrying consensus-only or all SWIM-modeled waters and Mg2+ ions for each map · Tetrahymena (ribozyme RNA) · n = 255 and 281 waters; 47 and 47 Mg2+ ions (all-SWIM models) · PDB 9CBU, 9CBX (2.2 Å); PDB 9CBW, 9CBY (2.3 Å)
  • Explicit-solvent all-atom MD from the 2.2 Å model in TIP4P-D water; DESRES and parmBSC0χOL3 (mMg or nMg) force fields, 10 runs each · Tetrahymena (ribozyme RNA) · n = 30 simulations of 400 ns · Stanford Digital Repository doi 10.25740/sw275qs6749
  • Earlier Tetrahymena ribozyme cryo-EM models and X-ray structures of ribozyme subdomains (including P4-P6) used for water and Mg2+ overlap comparison · Tetrahymena (ribozyme RNA) · PDB 7ez0, 7ez2, 7r6l, 7xd5, 7xd6, 7xd7, 7yg8, 7yg9, 7yga, 7ygb, 7ygc, 8i7n, 1gid, 1hr2, 1x8w
  • Earlier 3.1 Å apo ribozyme cryo-EM map and model, used as the starting model and as a lower-resolution comparison · Tetrahymena (ribozyme RNA) · EMDB-31385; PDB 7ez0
  • MgRNA curated set of X-ray RNA structures for comparing Mg2+ binding motifs · n = 489 X-ray PDB structures; 15,334 Mg2+ ions
Methods & tools
SWIM (segment-guided water and ion modeling, Segger v2.9.7), Q-score (MapQ), MotionCor2 and CTFFIND4, EMAN2 (NeuralNet particle picking), RELION (2D classification), cryoSPARC (Non-Uniform Refinement, 3D Variability Analysis), Phenix (auto_sharpen, real_space_refine), ISOLDE in ChimeraX v1.6.1, MolProbity, Amber20 PMEMD (CUDA) with DESRES and parmBSC0χOL3 force fields, TIP4P-D water, MDAnalysis and CPPTRAJ, Rosetta rna_motif
Software & environment
EPU 2.7; ChimeraX v1.6.1 with ISOLDE; SWIM in Segger v2.9.7; Amber20 CUDA PMEMD; TIP4P-D water with DESRES or parmBSC0χOL3 plus in-house mMg/nMg parameters; 2 fs step, SHAKE, 9 Å cutoff, 310 K, 1 bar NPT.
Assay conditions & compute
Each MD run used a single GPU (Amber20 CUDA PMEMD), 30 runs of 400 ns; data collected on a Titan Krios G3i (300 kV, Gatan K3) at ~420 movies per hour. GPU model, runtimes and memory not reported.
Reproducibility gaps
Authors advise using only the consensus-water PDB models (9CBU, 9CBW) for confident water and ion placement; the all-SWIM models (9CBX, 9CBY) also contain sites not reproduced across the two maps.
Limitations (6)
  • At ~2 Å resolution some peaks remain ambiguous in exact position and identity: SWIM cannot reliably tell water from cations, may miss or mis-assign monovalent ions (Na+, Cl-), and does not model Mg2+ bound to RNA only through water. · stated by the authors
  • Conservative criteria leave much reproducible solvent density unmodeled: SWIM recall plateaus near 25% of significant solvent-shell voxels, and the modeled ions offset only about one-quarter of the RNA charge. · stated by the authors
  • MD agrees poorly with the cryo-EM solvent shell in key regions (normalized AU-PRC 0.34 in the catalytic core), and MD Mg2+ placement was biased by 20 pre-positioned ions, so MD-based explanations of diffuse density remain hypotheses. · stated by the authors
  • Effects of radiation damage and vitrification on RNA-bound water were not examined, and alternate partial-occupancy water positions were not modeled, leaving open how cryo-EM water ensembles relate to solution behavior. · stated by the authors
  • Water modeling was limited to well-resolved core regions and the first solvent shell; flexible peripheral domains (P9.2, P6) lacked resolvability, and cytosine and uracil could not be reliably distinguished at this resolution. · stated by the authors
  • Both maps come from one sample and one dataset split by 3D variability classification and handled by the same pipeline, so the consensus cross-validation does not capture variation from grid preparation or processing. · noted in this summary

Read this if you model waters and ions in high-resolution cryo-EM maps, study how hydration stabilizes RNA tertiary structure, or need experimental benchmarks for MD water and Mg2+ force fields.

Citations
26
FWCI
9.37
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISKretsch '25Posani '25Sun '25Liu '21Kieft '23Haack '25
70% semantic matchBiophysics
Rachael C. Kretsch, Elisa Posani, Eugene F. Baulin+48 more·Proteins Structure Function and Bioinformatics·Nov 8, 2025· 5 citations