henhao
Sign in
Molecular BiologybioRxiv
Preprint

Thousandfold Expansion Microscopy

Helena Hu, Donatus KrahiD, Antonios NtolkerasiD, Sushovan ChandaiD, Alina Heimbrodt, Milton MondaliD, Jonas AltendorfiD, Bowen Jing, Bonnie BergeriD, Ali ShaibiD, Silvio O. RizzoliiD, Edward S. BoydeniD

Abstract

Biological macromolecules, such as proteins, are made of concatenated building blocks. We hypothesized that individual protein residues could be imaged by anchoring their side chains to a swellable polymer, cleaving backbone amide bonds, and expanding residues away from each other to a degree that enables them to be visualized separately. We introduce thousandfold expansion microscopy (1000ExM), a four-network interpenetrating hydrogel architecture that enables successive expansion from [~]18-fold to >1000-fold (one billion-fold in volume). Protein and peptide structures are maintained across these expansion factors, as verified by analyses of proteins with known structures (nanobodies, GFP) and a well-studied peptide (mCLING). Computational analysis indicates that 1000ExM resolves adjacent amino acid residues, thereby achieving sub-nanometer precision on conventional light microscopes. We anticipate that 1000ExM will find wide utility in protein visualization and identification, potentially even in intact cells and tissues.

PDFDOICode
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
5
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (21 Jul 2026) · CC BY 4.0

A four-network sodium acrylate/dimethylacrylamide hydrogel, cast ionic-within-ionic without neutral re-embedding, reaches about 1000-fold linear expansion and resolves individual peptide residues on standard confocal microscopes.

Iterative expansion microscopy variants that stabilize gels with a neutral re-embedding step (iExM, iU-ExM, pan-ExM) reach only 16- to 25-fold linear expansion, and ONE microscopy recovers protein shape but not individual amino acids. Casting charged networks directly inside expanded charged gels compounds expansion to about 1000-fold, lifting the ~0.38 nm spacing of adjacent alpha carbons above the diffraction limit of conventional light microscopes.

Key results

  1. ShownFour charged SA/DMAA networks cast one after another, with no neutral re-embedding gels between them, build up to about 1000-fold linear expansion. Each new monomer bath roughly halves the gel's current expansion instead of collapsing it back to its original size.

    Representative gel ~18x, ~100x, ~500x, ~1500x over four rounds (~18x shrank to ~10x on monomer infiltration); median ~1000x after PBS equilibration (n = 11, 9, 7, 4 gelations for rounds 1-4). Two-incubation protocol: median ~1900x, up to 2534x; dry-surface drift 0.3 ± 0.5% over 98 hours. · Development of 1000ExM; Multiple rounds of gelation and expansion; Figure legends

  2. ShownIn 3D confocal volumes of mCLING expanded about 1000x, the dye coordinates of single putative peptides fit molecular dynamics (MD) conformations to within a nanometer. The error is similar along x, y and z, which is consistent with isotropic expansion at the residue scale.

    6,284 putative peptides (from 27,426 Atto647N localizations; 8 regions, 3 gels) vs a 1001-conformation MD ensemble: RMSD 0.573 ± 0.151 nm; per-axis 0.317 ± 0.121 (x), 0.316 ± 0.120 (y), 0.325 ± 0.126 nm (z); single-molecule expansion factor 1055.92 ± 21.554. kNN ranks 1-5 differ from randomized coordinates (KS p < 0.001). · Evaluation of expansion isotropy at the…; Identification of putative peptides; RMSD analysis of putative peptides

  3. ShownTwo molecular rulers show that distances scale with physical expansion. A dual-labeled nanobody keeps its ~4 nm fluorophore spacing at every expansion factor, and averaged mCLING images resolve successive fluorescein peaks at the residue positions the MD model predicts.

    Nanobody spacing normalizes to ~4 nm at ~18x, ~100x, ~500x and ~1000x (735 nanobodies at 1000x confocal), matching 1000 simulated nanobodies. mCLING vs MD: R2 = 0.86 for peak count (p = 0.0028, n = 7 gel types), R2 = 0.85 for positions (p<0.0001, n = 22); Atto647N-to-first-peak and adjacent-peak distances not different (p = 0.216, p = 0.383, Mann-Whitney). · Proportional scaling of molecular distances…; mCLING 2D line-scan analysis; Figure legends

  4. ShownGFP 3D reconstructions, built by aligning and averaging ONE microscopy images of many single molecules, get sharper as expansion rises from 10x to 100x. They approach the resolution limit set by the spacing between anchorable residues, not between atoms.

    FSC resolution (0.143 threshold) relative to GFP crystal structure PDB 1EMA: 17.37 Å at 10x (n = 885, 3 gels), 14.54 Å at 50x (n = 5,590, 3 gels), 11.63 Å at 100x (n = 5,324, 3 gels); anchored-residue sampling sets a bound near 10 Å. · GFP 3D reconstruction approaches amino acid…; GFP 3D reconstruction; Figure legends

  5. SuggestedSimulations suggest that 3D fingerprints of labeled residue classes, under the distortion measured at 1000x, could uniquely identify nearly all human proteins using only three chemically addressable residue classes.

    23,391 human structures, surface residues, 75%-100% labeling efficiency: lysine only 89.3-93.3% identifiable at 5 Å noise, 78.0-83.8% at 10 Å; plus cysteine 93.9-96.5% and 84.2-89.2%; plus acidic residues 99.3-99.5% (5 Å) and 96.4-97.8% (10 Å); more classes gave diminishing gains. · Simulated identifiability of the human proteome…; Figure legends

Methods, data and reproducibility

Proteins or peptides are anchored through amines with AcX, embedded in a sodium acrylate/DMAA gel, digested with proteinase K and expanded; three more charged networks are then cast in the swollen gel before NHS-fluorescein labeling and confocal or ONE imaging, with fidelity scored against MD ensembles, crystal structures and nanoruler models.

  • mCLING-Atto647N peptide (Synaptic Systems) in about 1000x gels; 3D confocal volumes used for single-molecule matching · n = 6,284 putative peptides; 8 regions across 3 gels
  • mCLING peptides imaged in 2D from 18x to 1000x (ONE or confocal) for averaged line-scan analysis · n = 18900 at 1000x from 5 gel replicates
  • FluoTag-X2 anti-ALFA nanobody with AbberiorStar635P (NanoTag) as a ~4 nm nanoruler, 18x to 1000x · n = 735 at 1000x/confocal; 4 replicates per condition
  • His-tagged GFP purified from NebExpress E. coli, imaged by ONE microscopy at 10x, 50x and 100x · E. coli (expression host) · n = 885, 5,590, 5,324 molecules (10x/50x/100x), 3 gels each
  • GFP crystal structure used as FSC reference · PDB 1EMA
  • Molecular dynamics ensemble of mCLING-Atto647N (GROMACS, 1 us NPT production) · n = 1001 conformations
  • Human proteome structures with surface-exposed residues for identifiability simulation · human · n = 23,391 canonical proteins
  • Gel expansion factor measurements per casting round (Fig. 1B; updated protocol in Table S18) · n = 11, 9, 7, 4 gelations (rounds 1-4)
Methods & tools
1000ExM four-network sodium acrylate/DMAA hydrogel, Acryloyl-X SE (AcX) anchoring with proteinase K digestion, NHS-ester fluorescein post-expansion labeling, ONE microscopy Java plugin (built on SRRF NanoJ core), Leica STELLARIS 8 confocal, HyD-X detectors, 100x NA 1.51 oil objective, GROMACS with Espaloma parameters (RDKit ETKDGv2, MMFF94s), SciPy cKDTree for kNN distances, Two-sample Kolmogorov-Smirnov test against randomized emitter coordinates, Similarity-transform RMSD alignment; Kabsch alignment for proteome simulation, Convolutional VAE ab initio 3D reconstruction with Fourier shell correlation (0.143), MATLAB 2023b and SigmaPlot 14, Mann-Whitney rank-sum test
Code
https://github.com/Rizzoli-Lab/ONE-Microscopy-Java-Plugin
Software & environment
MATLAB 2023b; SigmaPlot 14; GROMACS with Espaloma parameters, TIP3P water, 1 us NPT production; ONE plugin settings TRAC = 4, radiality = 10; SciPy cKDTree. Full gel protocols in Supplementary Notes 1 and 2.
Reproducibility gaps
Only the ONE plugin repository is linked; no repository is given for the emitter detection, peptide matching, RMSD, proteome simulation or VAE reconstruction code, and no deposition of raw images or MD trajectories is reported.
Limitations (6)
  • Only chemically anchored side chains (lysine amines here) are reported, not all atoms; side-chain length and rotational freedom (lysine about 1 nm, amine sampling about 2 nm) cap reconstruction precision near 10 Å. · stated by the authors
  • After about 1000x expansion each fluorophore's signal is diluted, so samples are very dim and hard to locate, and the enlarged specimen (a ~10 µm cell becomes ~10 mm) makes imaging volume and acquisition time limiting. · stated by the authors
  • Multicolor residue identification needs trivalent residue-selective anchoring and labeling reagents that were not built, and assigning fragments of touching proteins to their parents is unsolved; proteome identifiability is therefore only simulated. · stated by the authors
  • Expansion factors vary from gel to gel over four sequential castings, so each gel must be measured. The minimized RMSD is only an upper bound because it mixes expansion distortion with incomplete MD conformational sampling. · stated by the authors
  • Putative peptides are grouped with kNN percentile cut-offs chosen to agree with MD distance ranges, and each takes its best match among 1001 conformations with free scaling, so the 0.573 nm RMSD may partly reflect selection rather than fidelity alone. · noted in this summary
  • Residue-level 3D validation at about 1000x rests on one flexible peptide (mCLING) in 3 gels; the folded-protein benchmark (GFP) stopped at 100x with ONE, and the proteome simulation assumes independent Gaussian noise on one static structure per protein. · noted in this summary

Who should read:Read this if you develop expansion or super-resolution microscopy, or want residue-level 3D protein mapping on standard confocal hardware as a route toward imaging-based in situ protein identification.

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
THISGao '19Liu '26Damstra '21Truckenbrod '18Venkatarama '26Pernal '19
74% semantic matchBioengineering
R. Gao, C.-C. Yu, L. Gao+4 more·bioRxiv·Oct 22, 2019