Expansion Microscopy: Choosing a Protocol and Proving the Expansion Was Isotropic

The thing that sinks most expansion microscopy (ExM) results is not the gel chemistry. It is that the author reports a distance in nanometres after dividing by an expansion factor they never measured in the region they imaged. ExM converts an optical resolution problem into a metrology problem: every post-expansion measurement is a real measurement divided by a number, and that number is a property of your specific sample, in that specific place, on that specific day. A protocol that nominally expands fourfold can deliver 3.7× in one organelle and 4.3× in the next, and nothing in the image tells you which.

So the useful way to think about ExM is: the chemistry is the easy part, the validation is the work. This guide covers what physically happens to the specimen, how to measure and report the expansion factor honestly, the controls that actually catch anisotropic expansion, how the main protocol families differ, and what expansion irreversibly destroys.

What physically happens to the specimen

Chen, Tillberg and Boyden introduced ExM in Science in 2015 with a deceptively simple idea: synthesise a swellable polyelectrolyte network inside a fixed specimen, covalently anchor the labels to that network, then dissolve away the mechanical resistance and let the network swell. Molecules that were closer together than the diffraction limit are physically pulled apart until an ordinary confocal can resolve them. Their original demonstration reported an apparent ~70 nm lateral resolution in cultured cells and brain tissue, and three-colour imaging of ~10⁷ cubic micrometres of mouse hippocampus on a conventional confocal microscope.

Nearly every variant since follows the same seven stages, laid out explicitly in the Nature Protocols guide to X10 optimisation: immunostaining, anchoring, polymerisation, homogenisation, expansion, imaging, and — listed as a stage in its own right, which is the point — validation.

• Anchoring. A bifunctional reagent gives the molecules you care about a polymerisable handle. In protein-retention ExM (proExM) this is Acryloyl-X SE (AcX), the succinimidyl ester of 6-((acryloyl)amino)hexanoic acid, which reacts with free amines on proteins and installs an acrylamide group. Other chemistries in use include methacrylate esters, glutaraldehyde, tetrafluorophenyl esters, and an acrylamide/formaldehyde mixture that anchors proteins during fixation.
• Polymerisation. The specimen is infused with a monomer solution of sodium acrylate, acrylamide and a small amount of N,N’-methylenebisacrylamide crosslinker, then gelled with ammonium persulfate (APS) and TEMED. The negatively charged acrylate is what does the swelling.
• Homogenisation. The specimen’s own mechanical integrity must be destroyed, or the gel cannot expand. This is done with a broad-specificity protease such as proteinase K, or with heat plus detergent, or a combination.
• Expansion. The gel is washed into deionised water. Removing the counter-ions unshields the acrylate charges, they repel, and the network stretches in all three dimensions.

Two consequences follow immediately and neither is optional. Homogenisation is indiscriminate, so anything that is not covalently anchored is lost. And because expansion is three-dimensional, a fourfold linear expansion is a 64-fold increase in volume — your fluorophores are spread over 64 times the space, so brightness per voxel falls accordingly.

The expansion factor is a measurement, not a protocol constant

The single most common reporting failure is quoting the nominal factor from the paper you copied the recipe from. There are three different numbers here and they are not interchangeable:

• Macroscopic gel expansion. Measure the gel disc with a ruler before and after. Easy, and the least relevant — it averages over the whole gel including regions with no specimen in them.
• Local expansion in the imaged region. What your nanometre claim actually depends on. In the ten-fold robust ExM (TREx) paper, nuclear pore complexes measured after a 9.5× macroscopic gel expansion came out about 8% smaller than the macroscopic factor predicted — the local factor and the gel factor genuinely differ.
• Per-structure expansion. Different organelles in the same cell have been reported to expand by different factors. There is no single scale bar that is correct everywhere in the image.

Practical ways to pin the local factor down, all of which appear in the published methodology literature:

• Image the same region of interest before gelation and after expansion, and register the two. Comparing a well-defined feature such as nuclear cross-sectional area is the cheapest version.
• Photobleach a defined geometric pattern (a square in the nucleus, for instance) before expansion, then look at that pattern after. This is the control that catches distortion you would otherwise miss — in one published example a nucleus looked convincingly isotropic until the photobleached square revealed a clearly deformed corner.
• Include an external size standard: DNA origami nanorulers, or seed cells on a fabricated reference pattern whose true geometry is known.
• Measure a structure of known dimension in your own sample. Nuclear pore complexes, microvilli and clathrin-coated pits are the usual internal rulers because independent electron-microscopy values exist for them.

Report the factor you measured, the method you measured it with, and the spread. “~4×” with no measurement behind it is not a result.

Isotropy: the control that actually catches distortion

Expansion being large is worthless if it is not uniform. The standard quantitative test is to image the same structure at high resolution before expansion — typically with structured illumination microscopy — then again by confocal after expansion, and plot root-mean-square (RMS) length-measurement error against measurement length. Published benchmarks give you something to compare against:

• proExM, cultured cells: RMS errors of roughly 1–2% of the measurement distance over length scales of 0–20 µm.
• proExM, tissue: 1–3% of measurement distance below 100 µm in pancreas, lung and spleen; 1–5% over 0–25 µm for nanoscale vimentin measurements in pancreas.
• TREx, large tiled acquisitions: measurement error behaved as a roughly constant fraction of measurement length, 3.2% ± 1.7%, across tiles of about 650 × 750 µm post-expansion.

If your own numbers are much worse than these, suspect incomplete homogenisation before you suspect the chemistry. Under-digested samples do not just expand less — they expand unevenly, and at higher expansion factors they tear. Tears are the visible failure; anisotropy is the invisible one, which is why it is more dangerous.

Some structures are known to misbehave. Chromosomal substructure has been reported to distort under ExM, plausibly because DNA molecules survive protease homogenisation intact and therefore cannot stretch in every dimension the way a digested proteome can; nuclease treatment before expansion has been used to address this. Whole organisms with mechanically heterogeneous regions — cuticle, cell wall, calcified tissue — are the hardest case, and are the reason cyclic-digestion protocols exist.

Choosing between the protocol families

proExM — ~4×, standard reagents, protein-retaining

proExM is the default entry point because it works with conventional immunostaining and off-the-shelf reagents. Proteins rather than labels are anchored to the gel with AcX, so ordinary fluorescent secondary antibodies and genetically encoded fluorescent proteins survive. The published monomer solution is 1× PBS, 2 M NaCl, 8.625% (w/w) sodium acrylate, 2.5% (w/w) acrylamide and 0.15% (w/w) bis-acrylamide, with APS and TEMED each to 0.2% (w/w); for tissue slices, 4-hydroxy-TEMPO is added up to 0.01% (w/w) to stop the gel setting before the monomer has diffused in. Gelation is two hours at 37°C. Digestion is proteinase K at 8 units/mL in 50 mM Tris pH 8, 1 mM EDTA, 0.5% Triton X-100, 1 M NaCl — overnight at room temperature, or four hours at 37°C. Expansion is 0.25–2 hours in excess deionised water, repeated three to five times.

Signal retention through that digestion is the reason proExM is usable: GFP fluorescence was preserved at 65 ± 5%, most tested fluorescent proteins retained more than 50% of their live intensity, and commercial dye-conjugated secondary antibodies retained roughly 50% of initial brightness. Denser tissues (pancreas, spleen, lung, which carry more connective tissue) resisted room-temperature digestion and needed 60°C for four hours.

U-ExM — ~4×, ultrastructure-preserving, post-expansion labelling

Ultrastructure ExM anchors with an acrylamide/formaldehyde mixture and stains after expansion rather than before. The payoff is structural fidelity good enough to resolve features previously accessible only by transmission electron microscopy — the original report resolved centriolar chirality. Protocol chapters describing U-ExM place its resolution in the 50–70 nm range and note the second advantage of post-expansion staining: it decrowds the sample, so antibodies reach epitopes they could not previously access, including in organisms with cell walls, cuticles or dense extracellular matrix.

TREx and X10 — ~10× in a single round

TREx reaches ten-fold expansion in one gelation step by moving to a very lightly crosslinked, high-monomer gel: 1.1 M sodium acrylate, 2.0 M acrylamide and 50 ppm bis-acrylamide (30–100 µg/mL is the usable range), with 1.5 mg/mL each of APS and TEMED. Homogenisation is a hybrid — reduced proteinase K digestion followed by three hours at 80°C in 5% SDS, 200 mM NaCl, 50 mM Tris pH 7.5. X10, the other well-documented ten-fold route, is reported at ~25 nm resolution against ~60–80 nm for a classic ~4× recipe.

The trade is signal. Ten-fold linear expansion is a thousand-fold volume increase, so labelling density and detector sensitivity, not resolution, become the limiting factor.

Very high expansion — 15× and beyond

Historically, factors above ~10× required iterating the whole process: digest the first gel, re-embed the already-expanded specimen in a second gel, expand again, for roughly 20× total. Single-shot alternatives now exist — a 2024 Nature Methods protocol (20ExM) reports ~20× in one expansion step, yielding sub-20 nm resolution on a conventional microscope while supporting post-expansion staining in brain tissue. Combining ~10× expansion with fluorescence fluctuation analysis (one-step nanoscale expansion, or ONE microscopy) has been used to image the shapes of individual proteins at around 1 nm, including conformational change in the ~17 kDa protein calmodulin.

Non-protein targets and scale-out

• RNA. Proteases do not cut nucleic acids, and RNA needs its own anchor. Expansion FISH (ExFISH) uses a small-molecule linker to attach RNA covalently to the gel, then performs fluorescence in situ hybridization after expansion with single-molecule precision, and is compatible with hybridization chain reaction amplification and multiplexed readout.
• Membranes. Ultrastructural membrane ExM (umExM) pairs a dense membrane label with an optimised protocol to image membranes and proteins at ~60 nm on a confocal, or ~35 nm when combined with fluctuation imaging or a second expansion round.
• Whole organisms. Whole-body ExM uses cyclic digestion — alternating two enzyme compositions over repeated rounds — to expand mouse embryos up to E18.5 and neonates fourfold despite their calcified tissue.
• Throughput. High-throughput ExM (HiExM) performs the whole workflow in a standard 96-well plate, with ~4.2× expansion reproducible within wells, across wells and between plates.

What expansion destroys, and what that costs

Plan around these before you start, not after the first failed run.

• Membranes do not survive as entities. Most lipids have no primary amine, so they are not anchored to the gel; they are only partially fixed under standard conditions and are then removed by the detergent used in homogenisation. If you need membranes you need a dedicated strategy — click chemistry, metabolic labelling, an amphiphilic probe such as mCLING, or a membrane-specific ExM variant.
• Signal dilutes with the cube of the expansion factor. Fourfold linear is 64-fold in volume; ten-fold linear is a thousand. Countermeasures are multiple dyes per target, post-staining amplification via biotin–streptavidin or DNA-based signal multiplication, and post-expansion labelling.
• Dyes are lost during polymerisation. Radical-mediated gelation is chemically aggressive and can irreversibly bleach some fluorophores; others are lost simply through incomplete grafting into the gel.
• Homogenisation denatures. This is the whole point, but it means epitopes are gone or altered. It is also why ExM is fundamentally a fixed-sample method — there is no live-cell version.
• ExM is not free resolution. The physics of the objective has not changed; you have made the sample bigger. Understanding the diffraction limit and what erodes resolution still governs what you get, and the effective resolution is roughly the instrument’s resolution divided by the measured expansion factor.

Pre- versus post-expansion labelling

This choice drives more of your result quality than the exact gel recipe does.

Pre-expansion labelling uses ordinary immunofluorescence on the intact specimen. It is familiar and works with antibodies validated for native epitopes, but the label then has to survive gelation and digestion, and antibodies cannot reach epitopes hidden in dense, crowded structures.

Post-expansion labelling introduces fluorophores after homogenisation. It consistently produces stronger signal — the label is not exposed to the polymerisation and digestion chemistry — and it exploits decrowding, so previously inaccessible epitopes become stainable. The catch is that the target protein is now denatured, so your primary antibody must recognise the denatured form. In practice that means favouring antibodies validated for Western blotting, and accepting that some antibodies simply will not work post-expansion. Published comparisons found post-expansion staining performance to be highly antibody-dependent, so test rather than assume.

Getting an expanded gel onto the microscope

An expanded gel is a large, fragile, mostly-water object, and it breaks the assumptions most imaging setups are built on.

• Immersion and depth. The sample expanded in Z as well as X and Y, so you are now imaging much deeper than before. Water immersion, or explicit refractive-index matching, is generally required; oil immersion accumulates optical distortion as you move away from the coverslip. Free working distance becomes a hard constraint — see how NA, magnification and immersion medium trade against each other when picking the objective.
• Ionic buffers shrink the gel. Any salt reintroduces charge shielding and the gel contracts. This is the specific reason expanded gels are incompatible with the photoswitching buffers used for single-molecule localisation microscopy; the workaround is to re-embed the expanded, stained gel in a neutral acrylamide gel that mechanically holds its dimensions.
• Mechanical stability. Use an imaging chamber. Gels drift, dehydrate and tear, and long tiled 3D acquisitions are exactly when that happens.
• Acquisition budget. A larger sample with dimmer voxels means longer exposures over more tiles, which puts you straight into the trade-off covered in managing photobleaching during extended acquisitions. Phototoxicity is moot on a fixed sample, but bleaching across a long tile scan will absolutely bias your measurements if you do not plan the acquisition order.

Frequently asked questions

Is expansion microscopy really super-resolution, or just magnification?

It is physical magnification that yields an effective resolution improvement. The microscope’s point spread function is unchanged; the specimen is larger, so features that were below the diffraction limit are now above it. Functionally the outcome resembles super-resolution — the original paper described it as scalable super-resolution microscopy on diffraction-limited instruments — but the mechanism is entirely sample preparation, which is why it needs no specialised hardware.

What resolution can I honestly claim?

Roughly your instrument’s resolution divided by your measured local expansion factor, and no better. Published anchor points: ~4× protocols are reported around 60–80 nm, U-ExM around 50–70 nm, X10 at ~25 nm, single-shot 20× below 20 nm, and ONE microscopy around 1 nm when combined with fluctuation analysis. Quote your own validated numbers rather than the headline figure from the source paper.

Do I need to buy a new microscope?

No — that is the central appeal. A conventional confocal is sufficient. What you may need is a longer-working-distance water-immersion objective and a stable sample chamber. Compared with the cost of dedicated super-resolution instrumentation, ExM is largely a reagent and labour cost; the relevant capital comparison is what a confocal system itself costs, since you are extending one you likely already have.

Can I do expansion microscopy on live cells?

No. The workflow requires fixation, chemical anchoring, gelation and proteolytic or thermal homogenisation. It is a terminal endpoint assay.

Why did my gel tear?

Almost always incomplete homogenisation. Regions that remain mechanically intact resist expansion while the surrounding gel pulls, and the stress concentrates until something rips. This gets worse at higher expansion factors, and it is worse in heterogeneous samples where grafting efficiency varies from region to region. Extend or intensify digestion, or switch to a heat-plus-detergent or hybrid homogenisation step.

Should I start at 4× or go straight to 10×?

Start at ~4×. It has the widest published validation across tissue types, the most forgiving signal budget, and the lowest chance of tearing. Move to ten-fold or higher only once you have demonstrated that your labelling density and detection sensitivity can survive a thousand-fold volume dilution, and that your homogenisation is complete enough to expand without tearing.

Can I combine expansion with other super-resolution methods?

Yes, and the gains compound — ExM has been combined with structured illumination, STED, single-molecule localisation and fluorescence fluctuation analysis. The practical obstacles are sample-specific: localisation microscopy needs photoswitching buffers that shrink the gel unless it is re-embedded, and 3D STED of an expanded volume demands acquisition times long enough that gel immobilisation becomes critical.

How do I image RNA or DNA in an expanded sample?

Nucleic acids need their own anchoring chemistry, because amine-reactive protein anchors do not attach them and proteases do not cut them. Expansion FISH supplies a dedicated RNA linker and performs hybridization after expansion. For DNA, be aware that intact chromosomal DNA can resist local stretching and produce distorted substructure; nuclease treatment before expansion has been used to mitigate this.

References

• Expansion microscopy. Science, 2015. doi:10.1126/science.1260088 (PMC4312537) — the founding method paper; source for the ~70 nm figure.
• Protein-retention expansion microscopy of cells and tissues labeled using standard fluorescent proteins and antibodies. Nature Biotechnology, 2016;34:987. PMC5068827 — source for the proExM monomer and digestion recipes, retention percentages and RMS error figures.
• Visualizing cellular and tissue ultrastructure using Ten-fold Robust Expansion Microscopy (TREx). eLife, 2022. elifesciences.org/articles/73775 — source for the TREx gel recipe, hybrid homogenisation and 3.2% ± 1.7% measurement error.
• Expanding boundaries — a cell biologist’s guide to expansion microscopy. Journal of Cell Science, 2024;137(7). PMC11058692 — source for validation strategies, membrane loss, signal dilution and gel-handling constraints.
• A practical guide to optimization in X10 expansion microscopy. Nature Protocols, 2019. PMID 30778205 — source for the seven-stage workflow and the ~4×/60–80 nm versus ~10×/~25 nm comparison.
• Imaging cellular ultrastructures using expansion microscopy (U-ExM). Nature Methods, 2019. PMC6314451
• Ultrastructure expansion microscopy (U-ExM). Methods in Cell Biology, 2021. doi:10.1016/bs.mcb.2020.05.006
• Nanoscale imaging of RNA with expansion microscopy (ExFISH). Nature Methods, 2016. PMC4965288
• Single-shot 20-fold expansion microscopy. Nature Methods, 2024. PMC11541206
• One-step nanoscale expansion microscopy reveals individual protein shapes. Nature Biotechnology, 2025. PMC7616833
• Dense, continuous membrane labeling and expansion microscopy visualization of ultrastructure in tissues (umExM). Nature Communications, 2025. PMC11821914
• High-throughput expansion microscopy enables scalable super-resolution imaging (HiExM). eLife, 2024. PMC11594540
• Nanoscale resolution imaging of whole mouse embryos using expansion microscopy. ACS Nano, 2025. PMID 39964913