Imaging the single-cell dynamics of the immune system within an intact environment requires the ability to look deep inside intact tissues and organisms with spatial and temporal resolution adequate to track cell morphology, motility, and signaling processes, all the while minimizing perturbation of the system under study. Optical microscopy employing fluorescence techniques is highly suited for this purpose, permitting both labeling of specific cells, organelles, or proteins and functional readout of physiological events (4
). However, conventional (single-photon) techniques such as wide-field and confocal microscopy suffer severe disadvantages, principally because the short wavelengths required for fluorescence excitation are subject to strong scattering in biological tissue and exacerbate phototoxicity. Nonlinear microscopy differs fundamentally from conventional techniques in that the elementary process involves near simultaneous interactions of two (or more) photons, so that the signal varies as the square (or higher power) of incident light intensity, rather than linearly. This nonlinear relationship leads to qualitatively new imaging modalities, of which fluorescence excitation by two-photon absorption and second harmonic generation have proved most useful for immunoimaging.
The essence of two-photon microscopy is that a fluorophore is excited by the near-simultaneous absorption of energy from two photons, each of which contributes one half of the energy required to induce fluorescence. Because excitation then increases as the square of the incident light intensity, fluorescence is essentially confined to the focal spot formed by a microscope objective, thereby providing an inherent optical sectioning effect, analogous (though involving a completely different mechanism) to confocal microscopy. To achieve practicable fluorescence signals, the photon density in the focal spot must be incredibly high, yet not so high as to damage the specimen. This is achieved by using femtosecond or picosecond pulsed lasers, which concentrate their output into brief bursts with enormous instantaneous power, yielding a two-photon advantage of ~105 as compared with a continuous beam of the same average power. The continuing improvement of ultrafast pulsed lasers has been a key factor driving the adoption of two-photon microscopy by immunologists.
In addition to its inherent optical sectioning effect, two-photon excitation has other major advantages for immunoimaging because the excitation wavelengths are roughly twice as long as would be used for conventional linear excitation by widefield or confocal microscopy. This use of long wavelength excitation is particularly advantageous for imaging deep into highly scattering biological tissues (5
) because scattering decreases with increasing wavelength (6
) and because absorption by hemoglobin and other proteins is minimized. Thus, the infrared wavelengths used for two-photon imaging enable a fivefold or deeper tissue penetration than does confocal imaging employing visible wavelengths (2
) and cause negligible photodamage or photobleaching. Long-term imaging is thereby facilitated because bleaching and damage processes (induced by nonlinear processes) are largely confined to those cells lying at the focal plane, whereas cells above and below experience only the innocuous infrared light. Nevertheless, the laser power must be kept below some sharp threshold value in order to maintain long-term viability of the preparation. Finally, the two-photon excitation spectra of most fluorophores are appreciably broader than for one-photon excitation (10
), so a single excitation wavelength can be used efficiently to excite multiple probes simultaneously with distinct emission wavelengths. Moreover, the ease of tuning of the latest generations of femtosecond lasers facilitates selection of a wavelength that balances excitation of different fluorophores. In instances in which highly divergent excitation wavelengths are required, two independently tuned lasers may be used simultaneously (8
In addition to multiphoton absorption, another nonlinear interaction that becomes prominent at very high light intensities is that of optical-harmonic generation, in which two (or more) photons are almost simultaneously scattered to generate a single photon with exactly twice (or higher multiples of) the incoming energy (11
). Second harmonic generation is produced by spatially ordered molecules and has proved especially useful for imaging ordered structural proteins such as collagen fibers (12
) and microtubules (13
) by detecting emitted light through a bandpass filter of twice the wavelength of the excitation light without the need for fluorescent labeling.
Deep tissue imaging is currently limited by several factors, including the requirement for exponentially increasing excitation power at increasing depths to compensate for increasing scattering loss. Use of infrared-emitting dyes may be advantageous in reducing signal loss owing to scattering, and the advent of ever more powerful lasers extends the depth range but will be limited by thermal effects and excitation near the surface (5
). Moreover, image quality (as well as brightness) degrades with transit through tissues with strongly varying refractive index but may be mitigated by correcting for wavefront distortions with deformable mirrors (14
) or by applying depth-dependent deconvolution algorithms to acquired images (15
). Imaging yet deeper into thick tissues and organs will necessitate mechanical techniques, such as penetration or removal of overlying tissue, use of needle-like gradient index lenses to extend the range of otherwise bulky objectives (13
), and development of miniaturized two-photon endoscopes (16
Fluorescent Labels and Probes
The imaging techniques reviewed here rely primarily on the introduction of a fluorescent probe or label into cells or structures of interest, although useful information may also be gained from intrinsic signals such as autofluorescence and second harmonic generation. Extrinsic labeling may be accomplished in vitro using isolated cells that are then adoptively transferred into a recipient animal by introducing fluorophores directly into the animal or organ being imaged (e.g., to stain vasculature) or by genetically engineering expression of fluorescent proteins.
The first in situ imaging studies of lymphocytes and dendritic cells (DCs) employed cell tracker fluorescent dyes, including CSFE (5,6-carboxyfluoresceine diacetate succinimidyl ester; green fluorescence) and CMTMR (chloromethylbenzoyl aminotetramethylrhodamine; red) for staining of adoptively transferred cells, and this method remains a mainstay. Such dyes have been used for many years by immunologists for flow cytometry and have the advantages of providing bright fluorescence signals while being relatively benign. Adoptive transfer of labeled cells typically results in <1% of cells being labeled in dense tissue populated by a vast excess of unlabeled cells. Although this is a huge advantage for tracking single cells and visualizing their morphological characteristics, it is obviously important that the unseen cells may exert significant effects on the labeled cells. The proportion of labeled cells can be adjusted to vary the density of labeled cells. Differently labeled cells can be coinjected and then distinguished by color. This approach has several advantages: It creates an internal control population for comparison; it allows labeling of a tissue landmark or compartment, such as the follicle, for orientation while imaging; and it allows comparison of cells by molecular perturbation with a drug treatment or a genetic deletion.
Dyes are available with a wide range of emission spectra separated sufficiently to permit independent visualization of at least three cell populations separately labeled with, for example, blue-, green-, and red-emitting dyes, by use of appropriate dichroic mirrors and bandpass filters before three photomultiplier detectors. Moreover, using dyes such as CFSE, one can track the number of times that lymphocytes divide in vitro or in vivo by taking advantage of the fact that the original dye content is split equally into the daughter cells each time the cell divides. Thus, comparison of relative fluorescence intensities among cells can be an indicator of how many times they have divided. However, this approach is complicated by the need to correct for variation in signal intensity with imaging depth into the tissue, and after a few cycles the remaining fluorescence becomes too faint to detect so that it is not possible to visualize labeled lymphocytes for more than one or two days after antigen stimulation or to follow them into peripheral sites of effector function. Significant disadvantages of in vitro dye labeling approaches arise because cells must first be extracted from their normal environment in the donor animal before adoptive transfer into the recipient host.
Expression of fluorescent proteins by target cells has been the other main method for visualizing target cells. The original fluorescent proteins derived from the Aequorea green fluorescent protein (GFP) suffered from low brightness and closely overlapping emission spectra. Subsequent developments, including the identification of a red fluorescent protein from the coral Discosoma
and extensive mutagenesis work, have now made available a wide range of fluorescent proteins with peak emissions ranging from blue (475 nm) to red (610 nm), which have greatly improved properties in terms of brightness, lack of pH sensitivity, faster maturation, and lack of oligomerization (reviewed in 18
). The two-photon excitation maxima for fluorescent proteins generally lie at longer wavelengths (approximately 900 nm) (20
) than for organic dyes (approximately 780 nm). Simultaneous imaging of both types of label with a single femtosecond laser thus requires selection of an intermediate wavelength that provides an optimal compromise, whereas application of a dual, independently tuned laser system may offer appreciable advantage. The fluorescence brightness obtained from a cell obviously depends on the level of protein expression and must be balanced against the possibility of cellular disruption with overexpression systems, particularly when the fluorescent protein is tagged to some other protein of interest.
There are several advantages to using expressed fluorescent proteins over other extrinsic fluorophores: (a
) The label is not diluted by successive cell divisions, allowing tracking of cell progeny after clonal expansion. Using a mouse with GFP under the control of a ubiquitous promoter such as β-actin or ubiquitin, T or B cells can be purified and adoptively transferred into recipients. For longer-term tracking, the recipient of choice would be one that has been tolerized by tissue-specific expression of GFP elsewhere, avoiding immune responses to GFP (21
) Fluorescent proteins can be expressed under the control of specific promoters to obtain selective expression of a particular cell type. For example, the CD11c promoter has been used to drive expression of YFP in DCs (22
). The advantage of being able to visualize every cell of a particular cell type must be weighed against the potential difficulty of tracking individual cell behavior in densely populated tissue. An analogy may be made with Golgi staining of the nervous system, which reveals the morphological complexity of sparsely labeled individual neurons in the brain. (c
) Specific proteins may be directly tagged with a fluorescent protein to visualize in real time their subcellular expression and/or relocalization, rather than merely expressing free cytosolic fluorescent protein as a cellular marker. (d
) Transgenic expression of fluorescent proteins enables in situ labeling of cells, so that for many experiments the need for ex vivo manipulations is eliminated. (e
) The percentage of a particular cell type that expresses a fluorescent protein can be adjusted by making mixed bone marrow chimeras (23
), thus facilitating tracking of individual cell behavior. (f
) Immune cells can be selectively ablated by irradiation in a mouse that ubiquitously expresses fluorescent protein and then reconstituted from bone marrow of a nonfluorescent donor, allowing radiation-resistant stromal and vascular cells to be visualized (25
Analyzing Cell Motility and Interactions
Multiphoton imaging yields 4-D (x, y, z, t
) information of cell morphology and motility, but the very wealth of data (gigabytes) presents appreciable problems of visualization and analysis. The results are almost impossible to convey in static images on the printed page but are better appreciated as time-lapse movies, usually presented as maximum-intensity 2-D projections along the z
-axis. Depth information is lost in this process (so it is impossible to discern whether cells touch or pass over one another), but it may be communicated by methods such as color-encoding the z
-axis location of cells (26
) or employing specialized 3-D visualization software.
The instant replay of time-lapse videos is, however, only a first step in immunoimaging, and deriving quantitative information by tracking cells is crucial (). Quantitative tracking is most easily achieved by 2-D analysis after compression of image stacks, a procedure that circumvents errors that are due to the inherently lower z-axis resolution. 3-D tracking is essential in cases in which motility may be anisotropic in the z-axis. Cell tracks may be superimposed directly on the imaging field () or, if there are no reasons to suspect regional differences, by overlaying tracks from several cells after normalizing their starting coordinates to generate a flower plot (). Various parameters () can then be extracted from such tracking data to characterize cell motility and morphology (). Instantaneous velocity is a simple and widely used parameter, but it is subject to errors. For example, jitter in measurements at brief time steps overestimates velocities and can give the appearance of motion in even stationary cells, whereas time steps longer than the persistence time for linear motion will tend to underestimate velocities.
Figure 1 Basics of motility analysis. (a) Snapshot visualization of the 3-D locations of fluorescently labeled cells throughout the imaging volume at a single time point. Tracks of three cells are superimposed. The cells are CFSE-labeled allogeneic CD8+ T cells (more ...)
Parameters used to characterize cell motility and migration
Early imaging studies (2
) suggested that T cell motility in the lymph node approximates a random walk, and several approaches have subsequently been adopted to characterize whether motility is indeed random or is directed (e.g., along chemokine gradients). For a random-walk process, the mean displacement from origin increases as a square root function with time. Analogous to the diffusion coefficient for Brownian motion, the slope of this relationship can be used to calculate a motility coefficient. This has generally been done by plotting displacement versus square root of time () (2
), but a plot of displacement squared versus time is more appropriate when data are appreciably scattered. A practical difficulty with this method is that cells exit the imaging volume at longer times, resulting in increasing standard error as fewer cells remain and possible bias as the remaining cells are likely to have lower velocities. Deviations from linearity of the displacement versus square root time plot point to nonrandom processes (): (a
) At intervals shorter than the persistence time, the relationship will curve upward, as cells migrate without turning; (b
) transition from a linear relationship to a plateau is consistent with migration confined by physical or biological barriers; (c
) a steeper-than-linear slope indicates directed motion (27
). The motility coefficient is a function of both the directionality of motion and the cell velocity, whereas the chemotactic index (displacement/path length) provides a measure of directionality that is independent of velocity but must be stated for some given time interval ().