Main

The ability to take light microscopy beyond the diffraction limit and resolve structures such as vesicles or molecules separated by less than 200 nanometers (nm) is the gift of super-resolution microscopy to cell biology.

Numerous super-resolution microscopy techniques have been developed to the point that scientists who are not in the habit of designing and building custom microscopes can decide how to use them to their best advantage1,2,3,4,5.

Nature Methods spoke with practitioners to find out how they address a question: How do you resolve structures closer to one another than 200 nm, obtain oodles of spatial and temporal data and perform speedy imaging—all with low phototoxicity? Their short answer: you cannot have it all. The longer answer: you can negotiate trade-offs for both fixed-cell and live-cell imaging.

Walls that have separated different super-resolution techniques and the groups developing them are coming down; commercial offerings are growing; ways to combine the techniques are emerging. New developments are spreading the techniques beyond a small circle of microscopists to a wider circle of scientists. At the same time, researchers say there are still plenty of misunderstandings and disagreements about super-resolution microscopy—even about the name.

Super or not super?

If only the community could avoid the word 'super' because it is so imprecise, physicist Stefan Hell, one of the directors of the Max Planck Institute for Biophysical Chemistry, says of the term 'super'-resolution microscopy. Instead, he says, the field should be called diffraction-unlimited fluorescence microscopy. Other scientists agree, but the term super-resolution microscopy has stuck. Whatever descriptor one chooses, optical imaging is now possible in ways unthinkable in the mid-1990s.

Among the main super-resolution techniques are photoactivated localization microscopy (PALM)/stochastic optical reconstruction microscopy (STORM), stimulated emission depletion (STED) microscopy, reversible saturable optical fluorescence transitions (RESOLFT) and structured illumination microscopy (SIM).

Structured illumination, which uses patterned light grating, is usually “limited to a factor of 2 beyond the theoretical diffraction barrier,” says Eric Betzig, who directs a lab at Janelia Farm Research Campus. He thinks SIM is not quite a member of the super-resolution family. In many respects, says Hell, SIM is not much better than a well-tuned confocal microscope, although some biologists might have been led to believe otherwise. SIM requires data processing, and confocal microscopy can be augmented with the same type of data processing. SIM is diffraction limited and not conceptually diffraction unlimited, he says.

Ernst Abbe comes to dinner

Ernst Abbe (1840–1905), physicist and innovator. Credit: Image Asset Management Ltd./Alamy

Hell, Betzig and others have found ways to work around Ernst Abbe's 19th century findings about light microscopes' resolution limits: the diffraction limit, he determined, is a function of the wavelength of light used to form the image and the angular aperture of the microscope objective.

3D SIM image of cytoskeletal proteins in a single optical section through a fixed human cell. Credit: L. Shao/E. Betzig, Janelia Farm; D. Burnette/J. Lippincott-Schwartz, NICHD

If Abbe came around today, Hell says, he would discover his findings unquestioned. Optical imaging cannot resolve a sample's individual features if they are closer than 200 nm and if a scientist relies on light reflecting from the sample.

Over dinner, Hell would tell the visiting Abbe how insight about the molecular nature of objects helps with the diffraction challenge—discoveries that emerged with early 20th century quantum mechanics. Special molecules called fluorophores can reveal features of objects. If two fluorophores are closer than 200 nm, the confocal microscope shows one fuzzy spot. But molecules can have different states: they may fluoresce for short periods and be 'on' or 'off'. In Abbe's day, the focusing abilities of the objective lens set the technical limitations. Better objectives led to better images. Today, says Hell, it's the best fluorophores that lead to the best microscopy images.

Leveraging the multiple states of fluorophores lets researchers capture details of neighboring structures. “That approach breaks the diffraction barrier and delivers in-focus images—in principle—all the way down to the level of individual molecules,” says Hell.

Just as is true for fluorescence microscopy in general, super-resolution microscopy methods do not image structures themselves, be they filaments or proteins. What light up are the fluorophores hitched to the structures, Hell says. “So the picture that you get represents the fluorophore distribution in the sample—nothing less, nothing more.”

The available super-resolution techniques differ in the way the fluorophores are made to light up sequentially, even though the fluorophores are exposed to excitatory light at the same time. Some fluorophores in a diffraction-limited zone are made to stay dark while neighbors shine, and then the ones that had been dark can be made to fluoresce while the other set stays dark. These steps allow scientists to discern the fluorophores and, through them, structures. “At one point all details have to have been lit up,” Hell says.

Briefly, how do the techniques differ?

Today it's the best fluorophores that lead to the best microscopy images, says Stefan Hell. Credit: S. Hell/MPI for Biophysical Chemistry

PALM/STORM, STED and RESOLFT share this on-off approach to fluorescence but implement them in different ways, says Hell. PALM/STORM are spatially stochastic, meaning the on and off states are created at random, on a single-molecule basis. By analyzing the diffraction pattern of each probe, the instruments help to localize the particular molecule sending the signal. Successive switching makes it possible to localize every molecule.

STED and RESOLFT are coordinate targeted in that a light pattern such as a standing wave handles the on-off switching. For example, with STED, a donut-shaped light pattern is used to scan and map the sample. The molecules that switch to 'on' become visible and deliver their coordinates, tightly controlled by the donut, Hell says.

Is super-resolution microscopy hard?

“None of these techniques are as easy as what you are probably used to,” says Hari Shroff, who runs the high-resolution optical imaging lab at the US National Institute of Biomedical Imaging and Bioengineering, referring to potential users with experience in confocal microscopy.

Although the super-resolution microscopy field is still heavily populated by specialists, “it has broadened since these techniques first came out,” he says. Commercialization allows more labs to try these techniques. Prices of instruments have come down, though these instruments are still significant expenses. And life scientists are beginning to learn how to handle the techniques, says Betzig.

Biologists must keep asking of their imaging results: “How do I know this is real?” says Hari Shroff. Credit: H. Shroff, NIH/NIBIB

Sample prep in super-resolution microscopy is serious and time consuming. Scientists cannot just transfer an approach from confocal microscopy, Shroff says.

Building your own instrument for PALM is relatively easy, whereas it is fairly demanding for STED or RESOLFT, says Hell. But using a STED microscope is as simple as using a confocal microscope. “Push a button, and out comes the image,” he says. There is no image processing: the raw data are the image. In PALM, molecules are localized by fitting the photon measurements with a Gaussian function. STED microscope users do not need to know exactly how the technique works, he says, but should know enough to assure that the STED wavelength of light matches the fluorophore being used.

With PALM/STORM, users have to pay attention to what is switched on in the 200-nm zone, otherwise the image will contain artifacts, says Hell. But PALM/STORM take the user directly to the level of individual molecules, which simplifies counting molecules in the image. For fixed cells, the technique enables resolutions beyond 30 nm pretty easily, he says.

Betzig views STED as more unwieldy than other techniques such as PALM/STORM. It's very difficult to get STED working properly, he says. Hell highlights the potential speed advantages of STED and RESOLFT, which read out the fluorescence of a particular structure or detail rather than the individual fluorescent molecules in PALM/STORM. Users can tune the resolution in a controlled fashion between 20 and 200 nm. If less 'super-resolution' is needed, fewer molecules need to be 'off' on average, and the imaging speed proceeds faster, Hell says.

What should biologists heed?

The diffraction limit of light is 100 times the size of structures that cell biologists study as they characterize events in organelles or membranes, says Shroff. Below the 200-nm resolution limit is where most cellular action is.

These new imaging techniques intrigue ever more life scientists, he says. But awareness of the consequences of using these techniques has not kept up with the curiosity about them.

Certainly, biologists need to keep their skepticism levels high. When imaging in the 100-nm range, biologists will want to remember how the overexpression of fluorescent proteins might bias the image, and they should recall how the sample can be perturbed by the light intensity of super-resolution microscopy, Shroff says.

Also, if scientists seek an image with 20-nm resolution, the pixel size must be at least half of that, so around 10 nm. At that resolution, a researcher is also making many more measurements per experiment, given that there are more data points—pixels—in the image than there would be in a diffraction-limited image, in which the pixel size may be 200 nm, he says.

PALM/STORM and STED ask more of dyes. Given the increased number of pixels, these imaging techniques must “hit the dye harder” to obtain the same per-pixel signal-to-noise ratio as that of conventional microscopy. And then there is the speed factor: “You have to be able to get that fluorescence faster in order to move on to the next pixel,” Shroff says.

A lower marker density might be acceptable in confocal microscopy, but with super-resolution techniques, dense labeling is a must. Super-resolution microscopy interrogates a given sample more, taking more measurements than diffraction-limited techniques do.

PtK2 cells imaged with the super-resolution technique RESOLFT. Credit: Hell lab/MPI for Biophysical Chemistry

Imaging microtubules with confocal microscopy allows the eye to connect the lines of the structure even if labeling is not so dense. The diffraction limit “fills in the gap,” he says. That possibility disappears with crisp super-resolution. Artifacts can confound an image, particularly when the structure being imaged is one a researcher might not know well.

Biologists need to keep firing self-critical questions: “How do I know this is real?” suggests Shroff. “What controls have I done to make sure I am not being fooled by the microscope?” The options are many: electron microscopy (EM) or an antibody-based strategy can help confirm a finding. Another kind of “sanity check” is to computationally blur a SIM or PALM image so that it resembles a confocal microscope image. Super-resolution just gives “one piece of evidence,” he says, and additional methods help to put together a more comprehensive picture.

“Every photon is precious,” says Paul Goodwin, who steers instrument development in cell biology at GE Healthcare Life Sciences. Scientists must care for all photons emanating from a sample, he says, jesting with a reference to a famous satirical musical sketch by Monty Python about the preciousness of every sperm.

In the super-resolution microscopy courses that Goodwin co-teaches with academics, he hears about envisioned projects to image signaling in three-dimensional (3D) environments or watch vesicular traffic in cells. In addition to seeking adequate spatial resolution, scientists want a temporal resolution that can capture a process that might unfold in a fraction of a second or over days. No matter the timescale, using super-resolution microscopy on live cells means compromise, he says.

Viewing a human heart muscle cell under a confocal microscope is like taking a cell and “throwing it on the sidewalk in Saudi Arabia,” says Goodwin. In their natural state, these cells never see a photon. When imaging with super-resolution methods, scientists need to remember that delivering light to a cell's every nook and cranny risks “drastically” changing the biology.

Imaging fixed cells

Between 200- and 100-nm resolution is the regime in which SIM will deliver the answer scientists will want, with a method that is versatile and fast, says Betzig. “The advantage of SIM is that you can use any off-the-shelf fluorescent label you have got.”

He is concerned about claims from some teams that have used a single object's measured width as evidence that they have obtained a 50-nm-resolution image. “People in the field should know better,” he says.

The “perfect metric” in super-resolution microscopy is the one that researchers may know from EM, he says. If an electron microscopist says an instrument can obtain 5-nm resolution, a scientist can be confident that objects can be seen at that resolution, no matter how densely packed that space may be.

In super-resolution microscopy, if an object is 'resolved' such that one can discriminate it and measure it, “that's not good enough,” he says. Resolution has to be such that researchers have the “confidence of knowing” that their view actually reaches down to the indicated level. Without that definition, scientists reside in the “squishy regime of not knowing that you really have seen at the limit that you claim to have seen,” Betzig says.

Offering a hypothetical super-resolution microscopy scenario, he says a sample might have a diffraction-limited region in which there are multiple structures that may be 50 nm apart. But they are not discernible because the hazy fluorescence background that each of the structures provides can swamp out the “tiny ripple” in the small signal that visually separates them, he says.

With fixed cells, there is less worry about damage to the cell through imaging. Certainly, EM is “never going away,” says Betzig, who extols EM's “talent” for global contrast. But unlike EM, optical methods bring to the table protein-specific contrast via fluorescent labels.

“The EM people keep eating away at that,” he says, noting work in the labs of Roger Tsien from the University of California at San Diego and Alice Ting at the Massachusetts Institute of Technology, where high-contrast EM labels are under development that could provide molecular specificity to the black of classic heavy-metal staining in the EM image. “It's a step forward, but it still isn't multicolor,” and it still isn't at the single-molecule sensitivity that can be obtained from fluorescent proteins, says Betzig.

The “perfect metric” in super-resolution microscopy is the one that researchers may know from EM, says Eric Betzig. Credit: E. Betzig/HHMI Janelia Farm Research Campus

Imaging fixed cells, he believes, is the “regime” of PALM, and it will remain the dominant technique until genetically encoded EM labels come along. Several labs are working to correlate PALM and EM. “It is going to take awhile before we get that correlative PALM-EM right,” he says. “It'll happen,” he says of work at Janelia and in other labs that requires EM and PALM scientists to work side by side.

EM sample-prep protocols will often kill all the fluorescence, he says, and chemical fixation can cross-link proteins almost to the point that all results must carry a “big fat asterisk” about trustworthiness, says Betzig. Scientists must look at their cells in both live and fixed states in order to be sure of the structures they are seeing at high resolution.

For fixed cells, PALM is superior to STED, Betzig says, who developed PALM and admits, “obviously, I have a bias.” He appreciates “the beauty of the concept of STED” but sees PALM's advantage in its ability to deliver single-molecule information without the high light intensity of STED. STED risks damaging a fixed sample with its 1 gigawatt per square centimeter of power, and similarly for RESOLFT's required 1–10 kilowatts, when trying to reach beyond 50-nm resolution.

Hell, who invented STED, disagrees and counters that STED and RESOLFT have a lower risk of artifacts. What has been missing for the techniques' broad applicability is a wealth of compact lasers at a variety of wavelengths, he says. A number of firms have stepped up development, and instruments are slated to be launched in 2014, all of which will advance STED.

He acknowledges that STED microscopes are considered among the priciest super-resolution instruments and that building your own calls for rather sophisticated optics experience. But the situation is about to change, he says. Hell has launched a start-up called Abberior Instruments, which will be offering kits that researchers can use to assemble their own microscopes next year.

Hell also likes PALM/STORM, which require expertise for correct use. Learning how to properly perform PALM/STORM imaging is worth the reward because of the high-quality results the techniques can yield, he says. Images of fixed cells with PALM and STORM “are among the most impressive” super-resolution images to date.

Harvard University physicist Xiaowei Zhuang, who developed STORM, uses super-resolution microscopy to study neurons and recently showed that actin filaments wrap around the circumference of axons in an evenly spaced, ring-like pattern6. In their paper, the authors suggest that these filaments may provide both elastic and mechanical support for the axonal membrane and help to organize membrane proteins along the axon.

Imaging live cells

Among the high-resolution optical methods, SIM offers the best trade-off for live-cell imaging, says Shroff. Researchers obtain a twofold resolution improvement in every imaging dimension, and they can use conventional dyes. The sample prep time is shorter than with other super-resolution techniques.

SIM coupled with total-internal-reflection fluorescence microscopy can take imaging down to around 80 nm at 1–10 frames per second, says Betzig. One can move further, even down to 60 nm, but that move involves going to another technique and is costly in terms of cell viability, he says, pointing to his team's previous work7. The researchers performed live PALM at a resolution of around 60 nm, in which a frame took 25 seconds. And then, he says, “you can collect around 30 frames before the cell is toast.”

A comparison of focal adhesions in a fixed cell with conventional, diffraction-limited total-internal-reflection microscopy (top) and PALM (bottom). Insets indicate magnified regions in subsequent images. PALM images contain much finer sampling—many more pixels or measurements—and greater resolution than the diffraction-limited images. Credit: H. Shroff, NIH/NIBIB

With live cells, the main trade-off is between imaging and damage to the sample. Super-resolution microscopy below 80 nm on live cells is, for now, a technical demonstration, Betzig says. “We've got a long way to go to make that ready for prime-time live imaging that people will really be able to use.” He sees promise in single-particle microscopy coupled with PALM (sptPALM) to track proteins moving through cells.

The advantage of sptPALM begins when scientists seek to track many molecules in the cell. Before sptPALM, researchers could track only around 100 molecules before the “fuzzy spots” ran together, he says. sptPALM offers the ability to track thousands of molecules per cell. A subset of molecules can be photoactivated, diffusion can be followed and then another subset can be photoactivated and followed. It is a powerful way to study live diffusion, he says.

But live-cell imaging calls for both spatial and temporal resolution. Doubling spatial resolution in two dimensions means quadrupling the number of measurements and octupling for three dimensions, which also takes eight times as long. Given the dynamics within cells, features can 'smear out' over the time it takes to capture the image. It is much like keeping the shutter open to take a photo of a merry-go-round, says Betzig.

Confocal microscopy is the workhorse in many labs, but super-resolution microscopy sees more. Here, nuclear pore complexes from amphibian epithelial cell lines show immune-labeled proteins pan-FG (green) and gp210 (red). Credit: Hell lab/ MPI for Biophysical Chemistry

Coping with that movement might also require 16 times as much label or light. The higher the spatial resolution that is sought, the more time it takes to obtain the data and the more light has to be put onto the sample. “These trade-offs just bite you in the ass big time at 100 nm and below in doing live imaging,” he says.

Commercial SIM is usually not fast enough for live-cell imaging, but this motivates him to keep exploring the technique. “If you could buy it, I certainly wouldn't be doing it in my lab,” he says.

Commercial trend-watchers

Goodwin believes the most promising super-resolution methods are those suited to 3D biology and live-cell imaging. As more complex 3D cultures are developed, viable super-resolution methods will need to be compatible with them, he says.

SIM methods strike him as “the most gentle on the cell,” he says. And if the STED camp could find a way to substantially reduce the photon load on the sample, this technique could play an important role in the area, he says. He also sees promise in an approach in Betzig's lab that combines SIM with selective-plane illumination systems.

Leanna Ferrand and Ian Clements work with customers using GE Healthcare Life Sciences' DeltaVision OMX super-resolution imaging system. In the last 2 years, Clements says he has seen scientists begin to move away from a technique focus and skepticism about whether they can make the technique work for their scientific question, shifting to an eagerness to pursue their particular research question on cellular mechanisms and behavior as well as in drug discovery.

Although the instruments are still in the “large capital equipment bracket,” prices have come down considerably, which has helped widen the user base, Clements says. Originally, the instruments were found in more “physics-heavy” labs whose members worked on proving the technique itself, says Ferrand. In an era of tight budgets, biologists, biophysicists and chemists might apply for a grant together to acquire a shared instrument so that they all benefit. “A lot of our super-resolution instruments are in core facilities these days,” she says. The ability of researchers to connect super-resolution microscopy to other types of data, including physiology results, is invigorating the field, says Ferrand.

The techniques are now starting to leave physics-focused labs. Super-resolution microscopy has become “less intimidating,” says Ferrand. Although there are many factors to consider, she increasingly sees nonmicroscopists sensing that they can learn it. For a light microscopist, moving on to EM is like an Englishman moving to Japan or Central China, where language and culture are unlike what he knows, says Clements; moving from confocal microscopy to super-resolution is more akin to the experience of a Brit moving to North America, where language and culture are similar.

As in other imaging fields, in super-resolution microscopy, users have to optimize parameters for their scientific question, says Clements. In balancing speed, resolution and sensitivity, “it's basically 'pick any two'.”

The field stands to benefit from the refinement of dyes: these can be modified to have the photochemical characteristics needed for a given technique. Dyes such as caged probes, which were previously set aside as less suitable for regular imaging, are now garnering new interest among super-resolution microscopy users, he says. These molecules have a chemical group—a caging group—that inhibits fluorescence. A pulse of light breaks off the caging group, and the molecule can then fluoresce.

There are no standards for imaging dyes and probes in this still-young field, says Ferrand: needs are different across techniques. When dyes don't work in one technique, scientists are willing to look at them “in a different light” in another technique, she says.

What's next?

The emerging cell culture architectures that let cells grow as spheroids, for example, appeal to scientists who seek protocols that are closer to the biology than cells growing as monolayers on a coverslip. But these architectures add to imaging challenges. Cells contain water, proteins, carbohydrates and lipids, all of which have different refractive indices. In these samples, light rays are scrambled, like the view through a windshield when rain drops collect before the wipers are on, says Betzig. “The super-resolution methods are exquisitely sensitive to aberration.”

Although he sees many claims about using super-resolution techniques to image in live tissues or 3D structures, he fears that the results may not be “real.” Without adaptive optics, super-resolution, live, multicellular imaging will not progress, he says.

Another issue with wide-field fluorescence is that all the illumination light is present throughout the sample, but only one plane of the sample is in focus. Besides hindering researchers' ability to see, the approach prematurely bleaches the sample before analysis can begin in the next focal plane.

One of the biggest disappointments that users of commercial SIM face occurs when they move from imaging single cells to groups of cells or tissue. The reaction, Betzig says, often goes from “Gee, this looks great” to horror about the “garbage” of multicellular live imaging. Noise drowns out the super-resolution information in the image. “In principle, that's one area where STED or RESOLFT could shine,” he says. In practice, though, it is hard because the illumination intensities required for these techniques can be phototoxic.

Hell explains that Leica has had an exclusive commercial license for STED, a situation that is changing in 2014, when the basic STED patent expires. STED and RESOLFT are going to be available in many new formats, and Leica instruments have improved, too, all of which will enable live-cell diffraction-unlimited imaging, he says. “Competition propels the development process,” Hell says.

“Every molecule can give out a finite number of photons before it's bleached,” Betzig says. Confocal microscopy is inefficient in handling that “finite photon budget.” He sees promise in light-sheet microscopy approaches, which do not expose the areas to light above and below the sheet plane. He hopes that work in his lab will succeed in extending plane illumination to achieve subcellular-resolution imaging inside cells and embryos.

This type of work in his and other labs to combine the advantages of plane illumination and super-resolution could make choosing a suitable super-resolution imaging technique easier in the years to come.