Expanding the limits of optical resolution

An image of neurons in a mouse hippocampus taken with expansion microscopy. Credit Ed Boyden, Fei Chen, Paul Tillberg/MIT

Editor's Introduction

Expansion microscopy

annotated by
Bob Kao

How do scientists examine organ development and diseases at subcellular resolution? Recent cutting-edge imaging modalities—called superresolution microsopy—enable scientists to examine molecules at nanoscale resolution. The authors of this article discovered a new set of chemical and physical tricks called ExM: expanding preserved fluorescently labeled specimens using conventional light microscopy. ExM provides comparable subcelullar resolution with superresolution microscopy, and neuroscientists are now armed with a new arsenal to meet the challenge of mapping neuronal connections in human brains.  

Paper Details

Original title
Expansion microscopy
Edward S. Boyden et al.
Original publication date
Vol. 347, Issue 6221, pp. 543-548
Issue name


In optical microscopy, fine structural details are resolved by using refraction to magnify images of a specimen. We discovered that by synthesizing a swellable polymer network within a specimen, it can be physically expanded, resulting in physical magnification. By covalently anchoring specific labels located within the specimen directly to the polymer network, labels spaced closer than the optical diffraction limit can be isotropically separated and optically resolved, a process we call expansion microscopy (ExM). Thus, this process can be used to perform scalable superresolution microscopy with diffraction-limited microscopes. We demonstrate ExM with apparent ~70-nanometer lateral resolution in both cultured cells and brain tissue, performing three-color superresolution imaging of ~107 cubic micrometers of the mouse hippocampus with a conventional confocal microscope.


Microscopy has facilitated the discovery of many biological insights by optically magnifying images of structures in fixed cells and tissues. We here report that physical magnification of the specimen itself is also possible.

We first set out to see whether a well-known property of polyelectrolyte gels—namely, that dialyzing them in water causes expansion of the polymer network into extended conformations (Fig. 1A) (1)—could be performed in a biological sample. We infused into chemically fixed and permeabilized brain tissue (Fig. 1B) sodium acrylate, a monomer used to produce superabsorbent materials (23), along with the comonomer acrylamide and the cross-linker N-N′-methylenebisacrylamide. After triggering free radical polymerization with ammonium persulfate (APS) initiator and tetramethylethylenediamine (TEMED) accelerator, we treated the tissue-polymer composite with protease to homogenize its mechanical characteristics. After proteolysis, dialysis in water resulted in a 4.5-fold linear expansion, without distortion at the level of gross anatomy (Fig. 1C). Digestion was uniform throughout the slice (fig. S1). Expanded specimens were transparent (fig. S2) because they consist largely of water. Thus, polyelectrolyte gel expansion is possible when the polymer is embedded throughout a biological sample.


Fig. 1. Expansion microscopy (ExM) concept. (A) Schematic of (i) collapsed polyelectrolyte network, showing crosslinker (dot) and polymer chain (line), and (ii) expanded network after H2O dialysis. (B) Photograph of fixed mouse brain slice. (C) Photograph, post-ExM, of the sample (B) under side illumination. (D) Schematic of label that can be anchored to the gel at site of a biomolecule. (E) Schematic of microtubules (green) and polymer network (orange). (F) The label of (D), hybridized to the oligo-bearing secondary antibody top (top gray shape) bound via the primary (bottom gray shape) to microtubules (purple), is incorporated into the gel (orange lines) via the methacryloyl group (orange dot) and remains after proteolysis (dotted lines). Scale bars, (B) and (C) 5 mm. Schematics are not to scale.

Diapers and more diapers

Explore the different applications of polyelectrolyte gels. See this JoVE article: http://www.jove.com/visualize/abstract/22650419/polyelectrolyte-gels-comprising-lipophilic-cost-effective-aluminate

Another article exploring changes in volume transitions of polyectrolyte gels and their effects on ionic size (expanding on Tanaka’s classical paper) with JoVE related citations: http://www.ncbi.nlm.nih.gov/pubmed/25217949.

What's in a gel?

What are polyelectrolyte gels?

How does the addition of an initiator and accelerator aid in forming the polymer network?

What are the chemical components of the fluorescent tags?

Brain expansion

Figure 1B is a coronal section of a mouse brain slice; Figure 1C is an expanded image of 1B using ExM. A widefield microscope was used by illuminating the side of the specimen.


Chen and colleagues describe key facets of ExM.

Figures (Ai) and (Aii) illustrate the expansion of the polymer network after water dialysis, whereas B and C show before and after ExM of a brain slice.

Panels D and F illustrate the concept behind engineering the fluorescently labeled tags, that will prove useful in visualizing components of neurons and cells, as will be seen in later figures of this groundbreaking paper.

We developed a fluorescent labeling strategy compatible with the proteolytic treatment and subsequent tissue expansion described above, to see whether fluorescence nanoscopy would be possible. We designed a custom fluorescent label (Fig. 1D) that can be incorporated directly into the polymer network and thus survives the proteolytic digestion of endogenous biomolecules. This label is trifunctional, comprising a methacryloyl group capable of participating in free radical polymerization, a chemical fluorophore for visualization, and an oligonucleotide that can hybridize to a complementary sequence attached to an affinity tag (such as a secondary antibody) (Fig. 1, E and F). Thus, the fluorescent tag is targeted to a biomolecule of interest yet remains anchored covalently with high yield (table S1) to the polymer network. The entire process of labeling, gelation, digestion, expansion, and imaging we call expansion microscopy (ExM).

We performed fluorescence imaging using ExM, examining microtubules in fixed human embryonic kidney (HEK) 293 cells labeled with the trifunctional label and imaged with confocal laser scanning microscopy pre- versus post-ExM processing. The post-ExM image (Fig. 2B) was registered to the pre-ExM image (Fig. 2A) via a similarity transformation, resulting in visually indistinguishable images. To quantify the isotropy of ExM, we calculated the deformation vector field between the images via a nonrigid registration process (fig. S3). From this vector field, we quantified the root-mean-square (RMS) error of feature measurements post-ExM. The errors in length were small (<1% of distance, for errors larger than the imaging system point spread function size; n = 4 samples) (Fig. 2C). Throughout the paper, all distances measured in the post-expansion specimen are reported divided by the expansion factor (supplementary materials, materials and methods).


Fig. 2. Expansion microscopy physically magnifies, with nanoscale isotropy. We compared images acquired via conventional microscopy (blue scale bars) versus images acquired post-expansion (orange scale bars). (A) Confocal image of microtubules in HEK293 cells. (B) Post-expansion confocal image of sample (A). (C) RMS length measurement error of pre- versus post-ExM confocal images of cultured cells (blue line, mean; shaded area, standard deviation; n = 4 samples). (D) SR-SIM image of microtubules. (E) Post-expansion confocal image of the sample of (D). (F and G) Magnified views of boxed regions of (D) and (E), respectively. (H) Profiles of microtubule intensity taken along the blue and orange dotted lines in (F) and (G). (I) RMS length measurement error of ExM versus SR-SIM images (blue line, mean; shaded area, standard deviation; n = 4 samples). (J) Transverse profile of a representative microtubule (blue line), with Gaussian fit (black dotted line). (K) SR-SIM image of clathrin-coated pits (CCPs) in HEK293 cells. (L) Post-expansion confocal image of the sample of (K). (M and N) Magnified views of a single CCP in the boxed regions of (K) and (L), respectively. (O) Scatterplot of radii of CCPs measured via ExM versus SR-SIM (n = 50 CCPs from 3 samples). Green line, y = x line; shaded green region, half-pixel width of digitization error about the y = x line. Scale bars for pre- versus post-ExM images, (A) 20 μm; (B) 20 μm (physical size post-expansion, 81.6 μm); (D) 2 μm; (E) 2 μm (9.1 μm); (F) 500 nm; (G) 500 nm (2.27 μm); (K) 2 μm; (L) 2 μm (8.82 μm); (M) 100 nm; (N) 100 nm (441 nm).


In the paper on ExM, isotropy refers to the physical expansion of an object in all directions.

For example, if a golf ball is physically blown up in all directions, then this is isotropy.

Here are some additional examples:

(1) Blowing up a regular balloon is isotropic—it expands in all directions, whereas blowing up a balloon for making balloon animals is not isotropic because it expands in the cross-sectional directions first, then fills out the length of the balloon.

(2) A lump of dough rising appears to be isotropic on the macroscale, but if you look closer there are air pockets that open up within the dough—that is an example of distortion (anisotropy) on smaller scales.

(3) Stretching a rubber band is not isotropic because it will stretch in one dimension while contracting in the other two.

Superresolution microscopy

Here are some links regarding superresolution microscopy:



What's in the pits?

Explore and compare and contrast the ExM visualized clathrin-coated pits versus the ones visualized by a different type of microscopy method, called TIRF microscopy on JoVE:


Notice any differences between these different imaging modalities?


Figure panels C, D, E, and I illustrate that the images obtained by ExM compared with superresolution imaging methods are similar qualitatively and quantitatively.

Note that the error is within the point-spread-function size of a special superresolution method, called superresolution structured illumination microscopy (SR-SIM).

We next compared pre-ExM conventional superresolution images to post-ExM confocal images. We labeled features traditionally used to characterize the performance of superresolution microscopes, including microtubules (45) and clathrin coated pits (6), and imaged them with a superresolution structured illumination microscope (SR-SIM) pre-ExM, and a spinning disk confocal post-ExM. Qualitatively (Fig. 2, D and E), the images were similar, and quantitatively (Fig. 2I), measurement errors were again on the order of 1% and well within the point spread function size of the SR-SIM microscope (n = 4 samples). Microtubule networks were more sharply resolved in ExM (Fig. 2G) than with SR-SIM (Fig. 2F). ExM resolved individual microtubules that could not be distinguished with SR-SIM (Fig. 2H). Microtubules imaged with ExM presented a full-width at half-maximum (FWHM) (Fig. 2J) of 83.8 ± 5.68 nm (mean ± SD, n = 24 microtubules from 3 samples). This FWHM reflects the effective resolution of ExM convolved by the width of the labeled microtubule. To estimate the effective resolution of ExM, we deconvolved [as in (7)] our observed microtubule FWHM by the known immunostained microtubule width [55 nm (6)], conservatively ignoring the width of the trifunctional label, and obtained an effective resolution for ExM of ~60 nm. This conservative estimate is comparable with the diffraction-limited confocal resolution [~250-nm lateral resolution (8)] divided by the expansion factor (~4.5).

Clathrin-coated pits were also well resolved (Fig. 2, K and L). ExM resolved the central nulls of the pits better than SR-SIM (Fig. 2, M and N). Clathrin-coated pit radii measured via ExM and SR-SIM were highly correlated, with a slope of 1.001 (total least squares regression, confidence interval 0.013 with P <0.05, n = 50 pits from three samples) (Fig. 2O). Forty-nine of the 50 points lay within a half-pixel distance of the unity slope line, suggesting that variation in the ExM versus SR-SIM comparison was within the digitization error of the measurement.

We next applied ExM to fixed brain tissue. Slices of brain from Thy1-YFP-H mice expressing cytosolic yellow fluorescent protein (YFP) under the Thy1 promoter in a subset of neurons (9) were stained with a trifunctional label bearing Alexa 488, using primary antibodies to green fluorescent protein (GFP) (which also bind YFP). Slices expanded fourfold, similar to the expansion factor in cultured cells. We compared pre- versus post-ExM images taken on an epifluorescence microscope. As with cultured cells, the post-ExM image (Fig. 3B) was registered to the pre-ExM image (Fig. 3A) via a similarity transformation. The registered images closely matched, although some features moved in or out of the depth of field because of the axial expansion post-ExM. Quantitatively, post-ExM measurement errors (Fig. 3Cn = 4 cortical slices) were 2 to 4%.


Fig. 3. ExM imaging of mammalian brain tissue. (A) Widefield fluorescence (white) image of Thy1-YFP mouse brain slice. (B) Post-expansion widefield image of sample (A). (C) RMS length measurement error for pre- versus post-ExM images of brain slices (blue line, mean; shaded area, SD; n = 4 samples). (D and E) Confocal fluorescence images of boxed regions in (A) and (B), respectively, stained with presynaptic (anti-Bassoon, blue) and postsynaptic (anti-Homer1, red) markers, in addition to antibody to GFP (green), pre- (D) versus post- (E) expansion. (F and G) Details of boxed regions in (D) and (E), respectively. (H) Single representative synapse highlighted in (G). (I) Staining intensity for Bassoon (blue) and Homer1 (red) of the sample of (H) along white box long axis. Dotted black lines, Gaussian fits. a.u., arbitrary units. (J) Bassoon-Homer1 separation (n = 277 synapses from four cortical slices). Scale bars for pre-versus post-ExM images, (A) 500 μm; (B) 500 μm (physical size post-expansion 2.01 mm); (D) 5 μm; (E) 5 μm (20.1 μm); (F) 2.5 μm; (G) 2.5 μm (10.0 μm); and (H) 250 nm (1.00 μm).

What's in an image of the brain?

For panels A through G and panel H, can you describe what part of the brain the authors are showing?

Consider browsing the Allen Institute’s Brain Atlas: http://developingmouse.brain-map.org/

Compare errors

How does the root-mean-square error length measurement percent compared with Figure 2C and 2I? Keep in mind the specimen types used in each figure.

Intensity line scans

Explore how distribution line-scan plots like in Figure 3H are generated from panels F and G:


You can also try measuring line-scan intensity plots by using images already available in the link to Fiji/ImageJ software:



Bassoon and Homer 1 label the pre- and postsynapse that express green fluorescent protein driven by the Thy1 promoter in transgenic mice.

Can you determine the importance of Bassoon and Homer 1’s function, and their role in neural function?

We synthesized trifunctional labels with different colors and oligonucleotides (supplementary materials, materials and methods) to enable multicolor ExM. We obtained pre- (Fig. 3D) versus post-ExM (Fig. 3E) images of Thy1-YFP-H mouse cortex with ExM labels directed against YFP (Fig. 3E, green) and the pre- and postsynaptic scaffolding proteins Bassoon (Fig. 3E, blue) and Homer1 (Fig. 3E, red). In the pre-ExM image, Bassoon and Homer1 staining form overlapping spots at each synapse (Fig. 3F), whereas the post-ExM image (Fig. 3G) shows clearly distinguishable pre- and postsynaptic labeling. We quantified the distance between the Bassoon and Homer1 scaffolds, as measured with ExM. We fit the distributions of Bassoon and Homer1 staining intensity, taken along the line perpendicular to the synaptic cleft (Fig. 3H, boxed region), to Gaussians (Fig. 3I). The Bassoon-Homer1 separation was 169 ± 32.6 nm (Fig. 3Jn = 277 synapses from four cortical slices), similar to a previous study using stochastic optical reconstruction microscopy (STORM) in the ventral cortex and olfactory bulb, which obtained ~150 nm separation (10). We also imaged other antibody targets of interest in biology (fig. S4).

To explore whether expanded samples, scanned on fast diffraction-limited microscopes, could support scalable superresolution imaging, we imaged a volume of the adult Thy1-YFP-H mouse brain spanning 500 by 180 by 100 μm (tissue slice thickness), with three labels (antibody to GFP, green; antibody to Homer1, red; antibody to Bassoon, blue) (Fig. 4A). The diffraction limit of our confocal spinning disk microscope (with 40×, 1.15 NA, water immersion objective), divided by the expansion factor, yields an estimated effective resolution of ~70 nm laterally and ~200 nm axially. Shown in Fig. 4A is a three-dimensional (3D) rendered image of the data set (an animated rendering is provided in movie S1). Zooming into the raw data set, nanoscale features emerge (Fig. 4, B to D). We performed a volume rendering of the YFP-expressing neurons in a subset of CA1 stratum lacunosum moleculare (slm), revealing spine morphology (Fig. 4B and movie S2). Focusing on a dendrite in CA1 slm, we observed the postsynaptic protein Homer1 to be well localized to dendritic spine heads, with the presynaptic molecule Bassoon in apposition (Fig. 4Cand movie S3). Examination of a mossy fiber bouton in the hilus of the dentate gyrus reveals invaginations into the bouton by spiny excrescences of the opposing dendrite, as observed previously via electron microscopy (Fig. 4D) (11). Thus, ExM enables multiscale imaging and visualization of nanoscale features, across length scales relevant to understanding neural circuits.


Fig. 4. Scalable 3D superresolution microscopy of mouse brain tissue. (A) Volume rendering of a portion of hippocampus showing neurons (expressing YFP, shown in green) and synapses [marked with anti-Bassoon (blue) and antibody to Homer1 (red)]. (B) Volume rendering of dendrites in CA1 slm. (C) Volume rendering of dendritic branch in CA1 slm. (D) Mossy fiber bouton in hilus of the dentate gyrus. (i) to (iii), selected z-slices. Scale bars, (A) 100 μm in each dimension; (B) 52.7 μm (x); 42.5 μm (y); and 35.2 μm (z); (C) 13.5 μm (x); 7.3 μm (y); and 2.8 μm (z); (D), (i) to (iii) 1 μm.

3D renderings

Explore how 3D rendering enables scientists to extract spatial information of biological cells and tissues:


What do you notice in the morphologies of neurons located along each of the layers of the hippocampus?

Synapses and dendrites

Explore more of the rendering of hippocampal neurons in Figure 4C here: http://www.jove.com/visualize/abstract/22930273/synaptic-integration-in-dendrites-exceptional-need-for-speed

Homer distributions

What do you notice about the Homer1 distribution in panel C and in optical slices in D (i) through (iii)?

Can you posit the possible functional significance of Homer1 distribution along and around the dendrites?

Explore the Homer family of proteins here: http://www.jove.com/visualize/abstract/23554128/homer-proteins-in-ca-entry

Mossy fibers

Explore the function of mossy fibers imaged by ExM images in panels D(i), D(ii), and D(iii) here: http://www.jove.com/visualize/abstract/20472133/hippocampal-mossy-fiber-synaptic-transmission-and-its-modulation

We report the discovery of a new modality of magnification, namely that fixed cells and tissues, appropriately labeled and processed, can be physically magnified, with isotropic nanoscale resolution (effective ~60-nm lateral resolution). Although acrylate esters have been used for antigen-preserving embedding for electron microscopy (1213), ExM represents the first use of an embedded polyelectrolyte gel, used here to expand the specimen. Superresolution imaging methods are slower than their diffraction-limited counterparts because they must resolve more voxels per unit volume. ExM achieves this by expanding the voxels physically. ExM achieves the same voxel throughputs as a diffraction-limited microscope, but at the voxel sizes of a superresolution microscope. Ongoing technology trends for faster diffraction-limited microscopy (14) will continue to boost ExM speed.


The physical magnification of ExM enables superresolution imaging with several fundamental new properties. The axial effective resolution is improved by the same factor as the lateral effective resolution. ExM can achieve superresolution with standard fluorophores, and on a diffraction-limited microscope. Superresolution imaging is often performed within ~10 μm of the sample surface because of low signal-to-noise, scattering, and refractive index mismatch. We were able to perform three-color superresolution imaging of a large volume of brain tissue over an axial extent of 100 μm with a spinning disk confocal microscope. Because the ExM-processed sample is almost entirely water, eliminating scattering, ExM may empower fast methods such as light-sheet microscopy (15) to become superresolution methods. ExM potentially enables labels to be situated within a well-defined, in vitro–like environment, facilitating in situ analysis (16). Because the sample is physically larger, any mechanical errors in post-expansion sectioning, or stage drift, are divided by the expansion factor.

The performance of ExM suggests that despite statistical fluctuations in polymer chain length at the molecular scale, at the nanoscale distances here examined these fluctuations average out, yielding isotropy. Estimates of mesh size for comparable gels suggest that the distance between nearest-neighbor polymer chains are in the ~1 to 2 nm range (17, 18). By tuning the material properties of the ExM polymer, such as the density of cross-links, yet higher effective resolutions may be possible.

Supplementary Materials


Materials and Methods

Figs. S1 to S5

Tables S1 to S4

References (1928)

Movies S1 to S3

References and Notes

  1. T. Tanaka et al., Phys. Rev. Lett. 45, 1636–1639 (1980).

  2. I. Ohmine, J. Chem. Phys. 77, 5725 (1982).

  3. F. L. Buchholz, Superabsorbent Polymers 573, 27–38 (1994).

  4. B. Huang, S. A. Jones, B. Brandenburg, X. Zhuang, Nat. Methods 5, 1047–1052 (2008).

  5. E. H. Rego et al., Proc. Natl. Acad. Sci. U.S.A. 109, E135–E143 (2012).

  6. M. Bates, B. Huang, G. T. Dempsey, X. Zhuang, Science 317, 1749–1753 (2007).

  7. N. Olivier, D. Keller, P. Gönczy, S. Manley, PLOS One 8, (2013).

  8. R. W. Cole, T. Jinadasa, C. M. Brown, Nat. Protoc. 6, 1929–1941 (2011).

  9. G. Feng et al., Neuron 28, 41–51 (2000).

  10. A. Dani, B. Huang, J. Bergan, C. Dulac, X. Zhuang, Neuron 68, 843–856 (2010).

  11. A. Rollenhagen, J. H. R. Lübke, Front. Synaptic Neurosci. 2, 2 (2010).

  12. G. R. Newman, B. Jasani, E. D. Williams, Histochem. J. 15, 543–555 (1983).

  13. K. D. Micheva, S. J. Smith, Neuron 55, 25–36 (2007)

  14. B.-C. Chen et al., Science 346, 1257998 (2014).

  15. J. Huisken, J. Swoger, F. Del Bene, J. Wittbrodt, E. H. K. Stelzer, Science 305, 1007–1009 (2004).

  16. J. H. Lee et al., Science 343, 1360–1363 (2014).

  17. A. M. Hecht, R. Duplessix, E. Geissler, Macromolecules 18, 2167–2173 (1985).

  18. D. Calvet, J. Y. Wong, S. Giasson, Macromolecules 37, 7762–7771 (2004). 


E.S.B. was funded by NIH Director’s Pioneer Award 1DP1NS087724 and NIH Director’s Transformative Research Award 1R01MH103910-01, the New York Stem Cell Foundation-Robertson Investigator Award, the MIT Center for Brains, Minds, and Machines NSF CCF-1231216, Jeremy and Joyce Wertheimer, Google, NSF CAREER Award CBET 1053233, the MIT Synthetic Intelligence Project, the MIT Media Lab, the MIT McGovern Institute, and the MIT Neurotechnology Fund. F.C. was funded by an NSF Graduate Fellowship. P.W.T. was funded by a Fannie and John Hertz Graduate Fellowship. Confocal imaging was performed in the W. M. Keck Facility for Biological Imaging at the Whitehead Institute for Biomedical Research. Deltavision OMX SR-SIM imaging was performed at the Koch Institute Swanson Biotechnology Center imaging core. We acknowledge W. Salmon and E. Vasile for assistance with confocal and SR-SIM imaging. We acknowledge N. Pak for assistance with perfusions. We also acknowledge, for helpful discussions, B. Chow, A. Marblestone, G. Church, P. So, S. Manalis, J.-B. Chang, J. Enriquez, I. Gupta, M. Kardar, and A. Wissner-Gross. The authors have applied for a patent on the technology, assigned to MIT (U.S. Provisional Application 61943035). The order of co–first author names was determined by a coin toss. The imaging and other data reported in the paper are hosted by MIT (http://expansionmicroscopy.org/rawdataScience2014).