Scientific insight is often conceived at frontiers where diverse disciplines meet. In biology, several such fields are converging: new ways to label tissues at the molecular level, new techniques in optical and electron microscopy, powerful electronics to collect terabytes of data, and computer algorithms that can reveal complex patterns and relationships. At JFRC, I lead a team that seeks to understand bioimaging challenges, to harness and introduce new disciplines or technologies that are relevant, and to promote a coherent systems approach for optimal benefit to biological research. Three initial projects will characterize this effort: higher resolution novel microscopy, high-throughput microscopy, and correlative microscopy.
Optical Microscopy with Resolution of Electron Microscopy
The detail that can be seen by optical microscopy is limited to ~0.25 microns (about 0.25 percent the diameter of a human hair). Although this level of detail can resolve some of the organelles of a cell, it falls short of seeing an organelle's details and even farther from seeing any nanometer-sized structure that is responsible for so much of molecular biology. Combining three insights has led us to a new kind of optical microscopy that can peer into this regime: (1) Single fluorescent molecules (which might label a protein of interest) can be imaged and their centers can be localized to a fraction of the size of the fuzzy spot that corresponds to their image. (2) Closely spaced, optically overlapping fluorescent molecules can be separated, and each can be localized if there is a distinguishing characteristic. For example, if two molecules light up separately in different image frames, the center of each can be inferred to a fraction of the spot sizes. (3) A new class of activatable fluorescent proteins has been developed in the past several years, and a distinguishing subset of these proteins can be turned into a fluorescing state while the remainder remain dark. This last insight led to a new kind of microscope that my co-inventor, Eric Betzig (HHMI, Janelia Farm Research Campus), and I have dubbed PALM (photoactivated localization microscopy). This microscope can activate, sample, and localize the centers of a very small subset of closely spaced label molecules. After bleaching the first subset, this process is repeated for a new sparse subset to collect new centroid locations, and iterated thousands of times until a significant fraction of fluorescent label molecules have been sampled. (See the movie for an illustration of this principle.)
If we draw only the centers (and not the whole fuzzy ball) of all these fluorescent molecules that have each been imaged individually, we can generate a high-resolution image. The images in Figure 1 show a fluorescent proteinlabeled Golgi apparatus in a cell seen by regular microscopy and by PALM.
With colleagues at the National Institutes of Health and Florida State University, we are broadening the utility of this technique and demonstrating ever more compelling applications to biological problems. Developing and streamlining sample preparation techniques, that preserve cell structure and optimize photophysical properties of fluorescent labels, is key for expanding the application. Furthermore, instead of labeling only one protein, two or even more proteins can be labeled, each with a distinguishing trait such as color. This could paint a picture of how one protein works in relation to another.
PALM in Three Dimensions
Interferometry is a technique that can measure positions to nanometer accuracy. To gain access to the third axial dimension, we have recently combined interferometry with PALM in a method we call iPALM that can resolve individual fluorescent protein locations in three dimensions. Requirements of self-calibration and tolerance to fluctuating brightness of the labels can be met with simultaneous multiphase interferometry and inspired the invention of a custom three-way beam splitter. Figure 2 illustrates how light from a single-molecule source interferes into three different beams, depending on the axial location of the molecule. The nanometer accuracy can decipher the three-dimensional structure of protein complexes such as focal adhesion, an assembly of more than 100 proteins that the cell uses to attach its cytoskeleton to an external environment and is used in motion or to respond to forces. (For more details, see Super-Resolution Microscopy Takes on a Third Dimension.)
PALM is only an initial example of what I hope will be a series of broadly useful innovations in microscopy.
High-Throughput Electron Microscopy for Three Dimensions
Biological systems are intrinsically three dimensional. Capturing this third dimension with both good fidelity and high throughput is particularly useful for neural circuit reconstruction. Electron microscopes are the time-tested way to resolve the thin membranes, synapse, and fine processes that define a circuit. There are different electron microscope technologies, each with advantages and trade-offs in throughput, lateral resolution, vertical resolution, defects of missing volumes, and ease of automation. Furthermore, since the true limitations are in image processing, these data attributes critically influence ease of registration, segmentation, tracing, annotation, etc. Obtaining the usable data requires a tight interaction between sample preparation, image processing, and the data acquisition for best and timely reconstruction. Sample preparation, algorithms, and acquisition development go hand in hand.
We are optimizing and evaluating two approaches for this specific application: tilt STEM (scanning transmission electron microscopy) and FIB-SEM (focused ion beam with scanning electron microscopy). Since each transmitted primary electron can impart a greater signal-to-noise ratio than backscattered electrons, transmission microscopy offers a greater throughput for a given number of primary electrons. Furthermore, STEM allows one to scan large areas with fine pixilation (minimizing the need to splice multiple images), can dynamically focus on tilted surfaces (which is required on large areas), and is a convenient platform for automation. The sections used in transmission are cut 4050 nm, which defines the z resolution for normal imaging with serial sections. This is inadequate; however, tilted imaging, at a few angles, can give valuable depth resolution within one section. How this projects out to reconstruction quality again rests on the interplay between data acquisition parameters and algorithms for processing it, and we are exploring this with Mitya Chklovskii (HHMI, Janelia Farm Research Campus).
Focused ion beams were developed more than 20 years ago for fine cross-sectioning in materials science and for semiconductor chip analysis. A fine atomic beam with <10-nm diameter mills out trenches or polishes walls. These newly exposed surfaces, such as a cross section of a transistor, can then be imaged with a scanning electron microscope (SEM) that detects the scattered electrons. This is now being applied to biological samples, particularly neural circuits. A sequence of fine polishing steps of 10 nm or less, each followed by imaging of each new surface, can give a stack of 3D data with isotropic resolution. Such continuous milling/imaging also gives excellent registration and avoids many of the defects, such as tears and folds associated with cutting the thin sections required for transmission microscopy. However, this comes at the cost of slower imaging speeds. Again, a tight interaction from sample preparation to image processing is evolving this approach and defining how it will best serve open biological questions.
Such ambitious projects can only succeed with the collective participation of researchers at JFRC. The availability of genetically engineered samples, customized labels, instrumentation support, computation infrastructure, image processing, and theoretical modelingas well as biologists to give inspiration and directionis critical. I hope to bridge diverse fields, nucleate and drive collaborations, and be driven by new research opportunities identified by scientists at JFRC.
As of June 14, 2010