Scientists love to decorate their favourite biomolecules with fluorescent tags. Attaching light-emitting labels to a protein can reveal when and where in a cell it functions, but usually the details are fuzzy. Optical microscopes use light with wavelengths between 350 and 750 nanometres, and structures smaller than about 200 nm cannot be seen clearly. That is much bigger than the thickness of a cell membrane and is about half as long as the mitochondria that supply cells' energy. At this scale, many cellular secrets are invisible. The protein machinery that allows a virus to invade a cell is blurry, as are the synapses across which neurons communicate.

The past few years have seen the rise of a suite of techniques, collectively known as super-resolution microscopy, that can use light to reveal structures much smaller than the theoretical limit. The trick is to control fluorescent labels, or fluorophores, so that not all of them signal at once. Light from each individual fluorophore creates a blur, but as long as blurs don't overlap, they can be resolved into individual points at their centres. This allows the position of the fluorophore to be identified precisely, revealing features as small as 20 nm. “The super-resolution that we have developed doesn't rely on changing the wave nature of light,” says Stefan Hell, director of nanobiophotonics at the Max Planck Institute for Biophysical Chemistry in Göttingen, Germany. “It relies on turning dyes on and off.”

Mitochondria in a cell, imaged by conventional microscopy (left), and super-resolution microscopy colour-coded by depth (middle) and in cross-section (right). Credit: ZHUANG LAB/HARVARD UNIV.

Although advances in instrumentation and informatics should not be overlooked, many researchers believe that it is better-performing fluorescent labels that will allow super-resolution microscopy to continue to move forward. “That's an area where the field will see the biggest advances,” says Jan Liphardt, a biophysicist at the University of California, Berkeley. “That's been limiting all of us.”

Sam Hess: "You make a map of where the molecules were. That's your image". Credit: G. MANLOVE/UNIV. MAINE

Electron microscopes can resolve features less than a nanometre long — even smaller than super-resolution. But electron microscopy requires elaborate preparation of samples: usually, cells must be 'fixed' with preservatives and then embedded in resin or frozen. By contrast, many forms of super-resolution microscopy can be done with live cells. And with fixed cells, labels for optical microscopy can identify proteins more specifically than can those available for electron microscopy.

Most super-resolution techniques fall into two categories. In one, sometimes called illumination-based super-resolution, precise geometric patterns of light shine repeatedly across a sample to control which fluorophores are active. In the other, sometimes called probe-based super-resolution, conditions are tuned so that just a few fluorophores emit light at a time.

Whereas illumination-based super-resolution microscopy requires specialized optical equipment, probe-based techniques do not. Experiments using the latter technique are relatively easy to set up (see 'Starting up in super-resolution'). However, only a few dozen of the hundreds of extant fluorescent proteins and dyes have the requisite properties for probe-based super-resolution microscopy: the ability to change from one 'spectral state' to another when exposed to certain wavelengths of light (see 'Fluorescent proteins for super-resolution microscopy'). Some remain dark until they are activated; others go from one colour to another.

Table 1 Fluorescent proteins for super-resolution microscopy
Full size table

Acronym uproar

Live-cell super-resolution images showing how actin and membrane proteins associate. Credit: HESS LAB/UNIV. OF MAINE

Three labs independently developed the first probe-based techniques in 2006: fluorescence photoactivation localization microscopy (fPALM) was described1 by Sam Hess, a physicist at the University of Maine in Orono; photoactivated localization microscopy (PALM) was described2 by Eric Betzig and Harald Hess, physicists at the Howard Hughes Medical Institute's Janelia Farm Research Campus in Ashburn, Virginia; and stochastic optical reconstruction microscopy (STORM) was described3 by Xiaowei Zhuang, a physicist at Harvard University in Cambridge, Massachusetts. Perhaps one of the most confusing aspects of these and other probe-based techniques is what to call them. Commonly used terms include fPALM/STORM and variations such as single-molecule localization microscopy (SMLM) and single-molecule active-control microscopy (SMACM). But the underlying concepts behind all the probe-based techniques are the same, says Sam Hess. “You somehow control the molecules so you only have a few visible at a time, you find their position, you cycle through a whole bunch of molecules, and you make a map of where the molecules were. That's your image.”

These techniques require more control over fluorophores than most scientists are used to, says Michael Davidson, director of the optical microscopy division at the National High Magnetic Laboratory in Tallahassee, Florida. “A lot of people are jumping into this. I get the question of what probes to use probably ten times a week,” he adds.

Fluorescent proteins are commonly used for super-resolution microscopy. The genes that code for them, often taken from jellyfish or other sea creatures, are fused with the genes for the proteins being studied, so that when the proteins are produced, they too are joined. Thus, when the fluorophore probe lights up, it allows researchers to locate the studied protein. The most popular protein for probe-based super-resolution microscopy is probably mEos2 (ref. 4). When first expressed, it fluoresces green, but a burst of ultraviolet light turns it red. Such 'photoconvertible' fluorophores offer certain advantages over those that start out in a dark, non-fluorescent state: they allow researchers to image the protein before experiments begin, and so more easily pick out healthy cells that are producing high levels of the labelled protein. What is more, newly produced proteins are different colours from those that have already been imaged, so researchers can follow pools of proteins over time and get a sense of their rates of production and destruction (see 'How to build a fluorescent protein').

Everything is illuminated

Often, though, one fluorophore per experiment is not enough. “Most of the outstanding questions that people want nanometric accuracy for are in the relationship of two or more different proteins relative to each other,” explains Jennifer Lippincott-Schwartz, a cell biologist at the National Institutes of Health in Bethesda, Maryland, and part of the team that invented PALM. “The only way that you can address that is using different markers at the same time.”

A lot of people are jumping into super resolution. I get the question of what probes to use probably ten times a week.

Gleb Shtengel, a physicist at Janelia Farm, says that getting two labels to work together inside a cell is difficult, partly because the optimal conditions for each are not always the same. Fluorophores always prove trickier to work with than the imaging apparatus. “You have to add another laser, but that's the simplest part,” says Shtengel. Putting the brighter label on the less-expressed protein can help to make sure that enough data can be collected on each of the proteins of interest to fix their locations definitively; expression levels must also be sufficient and reliable for both proteins.

And then there are the spectral considerations. If researchers want to use a second label alongside mEos2, for example, they have to find one unaffected by both red and green wavelengths of light. A protein described5 this year could be a big help: it converts from orange to far-red, a much desired colour that is distinct from both the natural fluorescence of cells and that of other popular fluorophores. “The palette is so small right now that any addition is a big step forward, especially if you add a colour in part of the spectrum that's empty,” says Shtengel.

But researchers are succeeding in using two-colour super-resolution microscopy. That has allowed them to address questions such as whether cell-surface receptors implicated in cancer are randomly assorted or are co-localized on the plasma membrane. Lippincott-Schwartz has described6,7 a general technique that allows researchers to quantify how proteins cluster together on plasma membranes, and to assess the size, abundance and density of clusters.

Even illumination-based techniques are benefiting from new fluorophores. A technique called stimulated emission depletion works by pairing lasers: one excites a spot to fluoresce, and the other shrinks the area of fluorescence by further exciting fluorophores on its periphery into a special dark state. To collect an image, the paired laser beams scan across the sample, repeatedly applying intense beams of light that force fluorophores into the appropriate state but can also damage cells. Hell and his colleagues last month described8 a fluorescent protein that can enable illumination-based super-resolution miscroscopy in extremely low light levels. Although most fluorescent proteins bleach out, or lose their fluorescence, with repeated imaging, this new protein can be switched on and off more than 1,000 times. The researchers were able to image dendritic spines (signal-receiving outgrowths on neurons) at light levels one million times lower than had previously been documented, and the technique can work with a standard confocal microscope, says Hell.

Conventional (a) and super-resolution (b, c) microscopy of microtubules and clathrin protein clusters. Credit: ZHUANG LAB/HARVARD UNIV.

Lippincott-Schwartz and others are working out ways to make conventional fluorophores amenable to probe-based super-resolution microscopy. Instead of lighting up just a few molecules at once, they activate an entire population and wait for the fluorescent proteins to slowly turn off. The analysis identifies the loss of signal, she explains. “As they bleach, molecules switch off and leave a hole that can be fit to determine where the molecule was.” Data for localizing dark holes are much noisier than those for localizing bright points, but the technique allows researchers to work with several labels at once. In unpublished work, Lippincott-Schwartz has been able to visualize as many as four fluorophores in a single fixed sample, and she thinks that the technique can also be made to work in live cells.

Desirable dyes

Xiaowei Zhuang of Harvard University looks for tiny details using fluorescent proteins and dyes. Credit: FREDFIELD.COM

To many cell biologists, the term fluorescent label is synonymous with fluorescent protein, but there are also small-molecule fluorescent dyes. Dyes tend to be more photostable than fluorescent proteins, so they can emit an order of magnitude more photons, which means that dye molecules can be detected and pinpointed more reliably, explains Markus Sauer, a biophysicist at the University of Würzburg in Germany. “The higher photon yield goes in hand with higher localization precision and thus a higher optical resolution,” he says.

The speed at which dyes turn on and off is also an advantage. In the first demonstration of live-cell, three-dimensional STORM, Zhuang used six different probes: four dyes and two proteins9. One of the dyes, Alexa 647, allowed an image to be taken in one second; proteins required substantially longer, at 30 seconds per image. Collecting more images in less time is a practical advantage for all samples, particularly for live cells, says Zhuang. “If you can't switch the probes fast, you can only image slow processes,” she adds.

The problem is that dyes are often less convenient than fluorescent proteins. Whereas researchers can label proteins with fluorescent proteins by introducing genes into cultured cells, dyes have to be attached in a separate step. The most common technique is to combine them with antibodies against a protein of interest. Usually, researchers label 'secondary antibodies', which themselves attach to antibodies against the protein of interest — a practice that allows the same reagents to be used in multiple experiments. However, because it is the antibody rather than the protein that is visualized, the dyes are somewhat distant from the protein of interest.

Antibodies can usually be used only on fixed cells, but they do have advantages. Relevant techniques are in common use, and antibodies work in samples that can't be transfected, such as human biopsies. What is more, the target proteins are produced naturally, rather than from introduced genes, which can have aberrant expression. Last year, Zhuang and her colleagues reported10 that they had used labelled antibodies with STORM to interrogate the locations of ten different proteins within synapses, distinguishing which occurred on the signal-sending (pre-synaptic) side and signal-receiving (post-synaptic) side — something that would be impossible in conventional microscopy because the synapse is so small.

There are also ways to use dyes without antibodies: 'soluble ligands', or secreted proteins that attach to cell surfaces can be produced, labelled and then added to cell cultures directly. Intracellular proteins can be labelled using a 'hybrid-fusion' approach. Instead of being fused to a fluorescent protein, a protein of interest is joined to a 'protein hook' that can attach to the dye molecules. A variety of tags are in use, and the technique can even work with commercially available chemical-tag kits made for conventional microscopy11. But the dye can sometimes attach to biomolecules other than the target, says Robert Campbell, a protein engineer at the University of Alberta in Edmonton. “That raises up background fluorescence, and that limits the level at which you can see the protein.”

Improved analysis will also help scientists to get more from their labels. Researchers led by Sam Hess showed12 that three fluorophores that all emit in the orange–red wavelengths could be distinguished from each other. Conventional microscopy would not be able to separate, say, a greenish-yellow label that emits two green photons for each red one from an orangish-yellow label that emits one green photon for each red one, but super-resolution microscopy can distinguish such signals because emitted photons are attributed to individual proteins. In such a case, “it's okay to have the emission spectrum overlap because you are imaging individual molecules”, says Hess. His team was able to use three labels with overlapping spectra to simultaneously image two membrane proteins and a cytoskeleton protein, showing how these different components of the cell interact.

Better resolution through computation

Better analysis and more sophisticated algorithms should also help researchers who are using only one label at a time. To speed imaging, researchers would like to increase the number of fluoropohores that emit light at any given time. But if too many fluorophores emit too close together, their signals overlap and cannot be resolved into individual points. Several groups are working on software that lets scientists image more labels in smaller spaces. For example, researchers at the University of Oxford, UK, adapted13 an algorithm originally developed to study crowded star systems, and used it in probe-based super-resolution microscopy. They showed that it could detect more fluorophores than could two imaging algorithms commonly used in microscopy.

Super-resolution imaging reveals the molecular architecture that enables cellular adhesion. Credit: H. HESS/C. WATERMAN/M. DAVIDSON

Aleksandra Radenovic, a biophysicist at the Swiss Federal Institute of Technology in Lausanne, has designed computational approaches to mitigate artefacts caused when 'bleached' proteins, which have supposedly lost their fluorescence permanently, revert to a state in which they can be activated14. The effort grew out of another project, exploring dense protein clusters on the cell membrane. After a protein fragment chosen as a negative control displayed unexpectedly high levels of clustering, Radenovic and her co-workers studied the activation times of individual molecules of mEos2. The data showed that signals from similar locations clustered together in time. The sequence of signalling molecules should be random across a sample, so these results indicated that the same protein was signalling more than once and was being misinterpreted as multiple proteins, explains Radenovic. “Just looking at the time domain, you can get rid of those artefacts,” she says.

Although probe-based super-resolution microscopy can be done using standard fluorescence microscopes, several manufacturers offer systems built specifically for this purpose, along with software for analysing the data. Such microscopes are designed to optimize the activation of probes. Licensing agreements restrict which acronyms each manufacturer uses in marketing, but a machine that works for one form of probe-based microscopy generally works for other forms as well. Tokyo-based company Nikon has installed its system in dozens of labs; Leica Microsystems of Wetzlar and Zeiss of Oberkochen, both in Germany, have also introduced systems. And small start-up companies, such as Vutara in Salt Lake City, Utah, are getting into the market as well. Applied Precision of Issaquah, Washington (acquired in April by GE Healthcare of Fairfield, Connecticut), plans to roll out its probe-based super-resolution system, Monet, later this year. Most of these companies also make instruments for illumination-based microscopy, which require specialized components.

With or without dedicated instruments, researchers are keen to try their hand at super-resolution microscopy. So far, most papers demonstrate proof of principle for microscope methods rather than fundamental new biology uncovered by the techniques, but the balance is shifting, says Davidson. “It's going to be an explosive field. It's just now raising its head, and it's about to take off like a bat out of hell.”