Chapter 12 Nanoscale Biological Fluorescence Imaging

Abstract

Biological imaging has been limited by the finite resolution of light microscopy. Recent developments in ultra‐high‐resolution microscopy methods, many of which are based on fluorescence, are breaking the diffraction barrier; it is becoming possible to image intracellular protein distributions with resolution of tens of nanometers or better. Fluorescence photoactivation localization microscopy (FPALM) is an example of such an ultra‐high‐resolution method which can image living or fixed cells with demonstrated lateral resolution of better than 20 nm. A detailed description of the methods involved in FPALM imaging of biological samples is presented here, accompanied by comparison with existing methods from the literature.

Introduction

The image of a point source has finite size R₀, even if the source is infinitesimal. Therefore, distinguishing multiple point sources from one another is possible when those point sources are separated by more than R₀, but increasingly more difficult when sources are numerous and separated by less than R₀. In most biological samples, it is advantageous to image large numbers of molecules of interest, and in most cases these molecules are observed within a region containing even larger numbers of water molecules, ions, proteins, nucleic acids, and lipids. Simultaneous resolution of all of these molecules from one another becomes virtually impossible. Thus, resolution limits the size of structures that can be imaged using light microscopy to sizes of order R₀. In wide‐field microscopy, the resolution has been quantified by the Rayleigh criterion (Born and Wolf, 1997):

$$R_0=.61\lambda /\text{NA}$$

where λ is the wavelength of the detected photons and NA is the numerical aperture of the lens system.

In laser‐scanning microscopes, the resolution is directly related to the properties of the observation volume O(r), the region in which fluorescence is both excited and detected, defined as O(r) = I(r) · C(r), where I(r) is the illumination point‐spread function, and C(r) is the detection profile. I(r) depends on the laser illumination wavelength, objective NA, the laser profile in the objective back aperture, objective aberrations, fluorescence saturation effects, and a number of other variables (Pawley, 1995, Sandison and Webb, 1994, Sandison et al., 1995). Illumination by high‐NA lenses can produce a diffraction‐limited illumination volume with full width at half maximum (FWHM) of ∼0.55λ/NA (Pawley, 1995). In confocal microscopy, epifluorescence detection using the same high‐NA lens as for illumination and a detector aperture placed in the image plane result in significant improvement in the resolution, but the detector aperture also reduces the fraction of light collected. For an infinitesimal detector aperture, the collection point spread function approaches that of the diffraction‐limited illumination profile in the case of an overfilled back aperture (i.e., the 1/e² radius of the laser beam is larger than the radius of the back aperture of the objective lens) and incoherent (uncorrelated) emission (Hess and Webb, 2002, Sandison and Webb, 1994, Sandison et al., 1995) and the resolution can be improved by as much as a factor of $\sqrt{2}$, neglecting the Stokes' shift of the fluorescence relative to the excitation wavelength, distortion of the illumination profile by polarization effects, and lens aberrations. However, the resolution is still limited to some fraction of a wavelength.

Two‐photon microscopy provides numerous advantages for imaging biological samples, including reduced out‐of‐plane photobleaching, excitation of multiple fluorescent probes using the same illumination wavelength, and excellent signal‐to‐background ratio (Denk et al., 1990, Xu et al., 1996). However, because of the longer wavelengths used to excite fluorescence, and despite the intensity‐squared dependence of the excitation rate, in practice the resolution is similar to that of a confocal microscope with detector aperture size optimized for maximum signal‐to‐noise ratio. Thus, a number of standard far‐field methods are limited in resolution to approximately λ/2n, where n is the refractive index of the medium (Hell, 2007).

One method for resolution enhancement is to extract information about the structure of the object using the “near‐field” electromagnetic waves found at distances less than λ from the object. Near‐field optics often use an aperture or optical fiber with diameter significantly smaller than λ to create an excitation volume which has a width much less than λ near the aperture. Because coupling the emitted light back through the tiny aperture or fiber tip is often inefficient, a standard objective lens is often used to collect fluorescence excited from single molecules in such applications. The tip can be scanned to image an area with resolution of at least 12 nm under visible illumination (Betzig et al., 2006, Betzig et al., 1991), or better than λ/40. Near‐field interactions between a sharp probe and sample can be used to image at even higher (∼12 nm) resolution (Betzig and Trautman, 1992). However, the requirement for proximity between the probe tip and the sample does pose a significant limitation for many biological applications (Hell, 2007).

Electron microscopy offers tremendous (near‐atomic) resolution and has been used extensively to image a variety of biological samples. Unfortunately, requirements for sample fixation, freezing, or other preparation methods, the addition of heavy metals to improve contrast, and the reduced ambient pressure or vacuum for electron beam propagation have limited successful imaging of living biological specimens. Since understanding dynamics is crucial to the understanding of biological processes, improved noninvasive methods based on far‐field visible‐light optics promise to reveal a great deal about biological function.

Development of “super‐resolution” methods, namely, techniques that break the diffraction barrier and image samples at length scales much less than a wavelength, is currently of great interest. These methods can be grouped into several categories and compared to highlight advantages and disadvantages.

The diffraction barrier has been broken using stimulated emission depletion (STED) fluorescence microscopy (Hell and Wichmann, 1994, Klar et al., 2000). STED causes molecules excited at the edges of a normal diffraction‐limited volume to be driven to the ground state without fluorescence by illuminating them with an annular beam at a frequency that causes stimulated emission from the excited state. Only those molecules at the null (center) of the donut‐shaped STED beam remain in the excited state long enough to fluoresce, resulting in emission from a highly confined volume. Focal plane resolution of 15–20 nm has been achieved in fixed biological samples using STED with nonlinear deconvolution (Donnert et al., 2006). The concept of STED has been generalized to other reversible saturable optical fluorescence transition (RESOLFT) (Hell et al., 2003) techniques, which exploit optically driven transitions between states with drastically different emission properties, such as photoswitching of fluorescent proteins (Hofmann et al., 2005) to achieve subdiffraction‐limited resolution.

4Pi microscopy (Schrader and Hell, 1996) and I⁵M (Gustafsson et al., 1999) both use two opposing objective lenses to illuminate a sample and collect fluorescence with improved axial resolution. Structured illumination (Gustafsson et al., 1999) and saturated structured illumination microscopy (SSIM) use a spatially modulated sample illumination profile to, upon deconvolution, extract information at higher spatial frequencies and thereby improve resolution. SSIM resolution is in principle limited only by signal‐to‐noise ratio and the photobleaching properties of the probes, with demonstrated resolution of better than 50 nm (Gustafsson, 2005).

A recent example of hyperlens imaging uses a multilayered anisotropic material with hyperbolic dispersion to convert scattered nonpropagating evanescent electromagnetic waves containing high spatial frequency information about a sample smaller than λ, into propagating far‐field electromagnetic waves that were imaged using a high‐NA objective (Liu et al., 2007). However, such methods have not yet been demonstrated on biological samples.

Single‐molecule localization and image reconstruction is the basis for several super‐resolution methods. Localization, namely determination of the position of an object using its image, has been achieved with precision as high as 1.5 nm (Yildiz et al., 2003) for the diffraction‐limited image of a single fluorescent molecule. Single‐molecule detection methods provide additional information about absolute numbers of molecules, motion, and brightness of individual molecules, which can reveal population heterogeneities inaccessible to methods that image an ensemble of molecules. Ultrahigh‐resolution colocalization (UHRC) (Lacoste et al., 2000) is a scanning confocal microscopy technique capable of localizing multiple fluorophores that are excitable with a single laser source and of differing emission properties. While UHRC allows for simultaneous imaging of multiple probes, it is still difficult to use to resolve identical probe molecules separated by less than R₀. Fluorescence intermittency has been used to localize single‐molecules and single‐quantum dots with a precision on the order of tens of nanometers (Lagerholm et al., 2006, Lidke et al., 2005), as well as quantify velocities of individual biomolecules and protein assemblies by fluorescent speckle microscopy (Ponti et al., 2005, Salmon et al., 2002, Waterman‐Storer et al., 1998). Single‐particle localization in three dimensions has been achieved with ∼20 nm resolution and ∼30 ms time resolution (Levi et al., 2005a, Levi et al., 2005b). Other methods have exploited the photobleaching characteristics of fluorophores to localize single molecules. Single‐molecule high‐resolution imaging with photobleaching (SHRImP) (Gordon et al., 2004) and nanometer‐localized multiple single‐molecule (NALMS) fluorescent microscopy (Qu et al., 2004) both take advantage of the stepwise photobleaching of single molecules to localize their positions with precision on the order of a few nanometers. So far, such photobleaching methods have required that relatively few fluorophores reside within a cross‐sectional area of radius R₀. The points accumulation for imaging in nanoscale topography (PAINT) (Sharonov and Hochstrasser, 2006) method localizes single molecules that fluoresce as they bind to a target object of interest and then photobleach. While the methods described above provide a means of subdiffraction localization‐based resolution, control over the density of fluorescent molecules in the field of view requires adjustment of the concentration of fluorophores.

Techniques such as fluorescence photoactivation localization microscopy (FPALM) (Hess et al., 2006), and photoactivated localization microscopy (PALM) (Betzig et al., 2006), which use photoactivatable fluorescent proteins (PA‐FPs) or other photoactivatable fluorophores, allow for direct optical control over the number of fluorescent molecules by adjusting the rates of photoactivation and photobleaching (Fig. 1). In a similar manner, stochastic optical reconstruction microscopy (STORM) (Rust et al., 2006) uses photoswitchable combinations of organic fluorophores to control the number of molecules fluorescing at a given time. Images are then reconstructed from the coordinates and intensities of localized molecules.

Compared to confined and modulated illumination methods, FPALM can be used in a standard wide‐field microscope, does not require ultra‐fast pulsed lasers or image deconvolution, and does not rely on nonlinear excitation. Compared to other single‐molecule localization techniques, FPALM, PALM, and STORM also have the advantage that they rely on photophysical properties to control the number of fluorescent molecules in the field of view. The flexibility of using genetically encoded fluorescent markers such as green fluorescent proteins (GFPs) is both powerful and efficient, allowing existing GFP‐constructs to be converted into PA versions using standard molecular biology procedures.