Ординатура / Офтальмология / Английские материалы / Ocular Neuroprotection_Levin, Polo _2003

III.WHOLE MOUNTS VERSUS SECTIONS: WHICH SHOULD YOU USE?

Morphological assessment typically involves histological sections prepared using conventional techniques such as paraffin embedding, cryosectioning, or epoxy embedding for electron microscopy. However, unlike most tissues, the retina can be studied as a whole mount, offering distinct advantages for quantitative studies. In a retinal whole mount, topography is preserved, with major landmarks still in place. With the exceptions noted below (Sec. VIII), the total population of a homogeneous cell type occupies layers that are only one to two cells deep, permitting its visualization within a small range of focus. Cells in individual layers can be visualized using type-specific antibodies or by distinctive fluorescence patterns achieved with nucleic acid dyes such as DAPI or ethidium homodimer (Molecular Probes). Shrinkage can be negligible in whole mounts except around cut edges. Large samples are easily attainable, so that the study of minority cell populations and the detection of small effects are possible. High-contrast images of specific cell types that are achievable in a limited focal plane should be the goal, as these images may be conducive to either automated counting or counting that can be done by relatively low-level personnel.

Photoreceptors present a challenge to whole-mount studies, because their extreme compartmentalization is reflected in the composition of the retinal layers viewed en face.

1.Inner segments. In human retina, we counted photoreceptors at the level of the inner segments, the cross-sections of which form a mosaic in a single focal plane. Cone inner segments are distinguishable from rods by their threefold larger diameter and their more highly refractile properties in most parts of human retina. However, in retinal degenerations, the morphology of photoreceptors changes as outer segments are lost and the inner segment shortens and broadens. Therefore, in our study of eyes with age-related maculopathy [5], we validated counts of rods and cones made on the basis of size differences of inner segments using carbonic anhydrase histochemistry [7], which stains cone nuclei even in degenerated retina. In other species, the difference between rod and cone inner segments is not as prominent as it is in human retina (e.g., rabbit [8]) or inner segments may form tiers rather than a mosaic (e.g., pig [9]). In these species, examination of multiple focal planes is required to distinguish between rod and cone inner segments.

2.Nuclei. Cone nuclei occupy a single layer in the outer nuclear layer of most mammalian retinas, but rod nuclei form multiple rows that require many focal planes for analysis.

3.Outer segment. The first widely used photoreceptor-specific antibodies were directed against proteins associated with transduction in the outer segment.

Other investigators have used lectins that bind cone and rod sheaths. The validity of these markers is beyond the scope of this chapter. However, it is worth mentioning that outer segments are fragile and could be easily lost in the preparation of a whole mount. Therefore, with outer segment–specific markers one should verify that the vast majority of inner segments have an attached outer segment, which we did when counting short-wavelength–sensitive cones in human retina. Some cone markers like calbindin stain cones in their entirety [l0].

The major challenge to accurate counts of ganglion cells is the substantial minority population of displaced amacrine cells, which may overlap in size and morphology with ganglion cells. The proportion of ganglion cell layer neurons that are amacrine cells varies among species and within a species, with retinal region [11]. The sine qua non of a retinal ganglion cell is the presence of a projection axon to the brain. Some markers (e.g., BRN3A and BRN3B [12]) have been used in conjunction with injections of a retrograde tracer in the brain to identify ganglion cells. However, tracer injections only reach some of retinorecipient targets in the brain, and within a given target, not all ganglion cell terminals may take up and transport tracer. Therefore, the gold standard for validating retinal ganglion cell counts is comparison to optic nerve axon counts in the same eye [13]. If the total number of ganglion cells is the only desired outcome variable (and not overall topography, or counts of a particular subtypes), one should count optic nerve axons rather than ganglion cell bodies. It is possible to obtain unbiased counts of optic nerve axons using high-throughput morphometric techniques developed for neurogenetic studies by Williams and colleagues [14]. These authors determined total axon number from the total cross-sectional area of the nerve and the number of axons within unbiased counting frames on electron micrographs that were systematically placed within the nerve. Axons were counted directly on the negatives.

Despite the many advantages of retinal whole mounts for quantification, sections will continue to play an important role in assessing the retina in neuroprotection studies. Stained sections currently afford greater detail than wholemounts and are amenable to multiple marker studies in single animals. Epoxyembedded tissues can be used for electron as well as light microscopy. There is a wide range of routine, special, and advanced stains and high-throughput processing technologies designed for paraffin sections, the mainstay of clinical pathology laboratories. Finally, it should be stressed that a complete census of retinal cells currently requires a combination of techniques. In a study that has provided the most complete enumeration of the neurons in the C57B1/6 mouse retina, photoreceptors were counted in wholes mounts, interneurons and Mu¨ller cells were counted in vertically oriented ultrathin sections examined by electron microscopy and optical sections examined by confocal microscopy, and ganglion cells were counted in electron micrographs of optic nerve sections [13].

IV. VISUALIZING TISSUE FOR COUNTS

The investigator has several choices for imaging retinal tissue for cell counts. These include making counts directly from a microscope using an ocular reticule to delimit counting fields, making counts from a video image generated by a camera mounted on a microscope, producing photographs that can be printed or scanned for off-line analysis, and capturing digitized images for off-line analysis. In our studies of photoreceptors and ganglion cell layer neurons in human retina, we viewed tissue with a computer-video-microscope system that included a 60 1.4 N.A. oil-immersion objective, Nomarski differential interference contrast optics, a digitizing tablet, and a computer-controlled stepper motor stage. We viewed the tissue on a video monitor that provided 2000–3000 final magnification, depending on the objective used. The video image was combined with custom graphical overlays that were manipulated using a digitizing tablet to mark counted cells. This arrangement allowed us to optimize the focus for counting for photoreceptor inner segments and to count while focusing through the thickness of the human macular ganglion cells layer. A computer controlled the movements of the stage so that the retinal sheet could be sampled systematically in a manner that substantially eliminated the operator error inherent with manual stage control. Many of these functions, which required custom software in our studies, are now available in commercial stand-alone packages (Image Lab, Metamorph) or are bundled with image processing software available for confocal microscopy.

V.SAMPLING CONSIDERATIONS

The past 15 years have seen significant advances in stereology, the branch of applied mathematics that seeks to deduce unbiased information about threedimensional tissue structure from two-dimensional sections. These methods eliminate bias (i.e., systematic overor under-counts) and have practical implications for investigators in neuroprotection research. A thorough discussion of the mathematical foundations of spatial sampling is beyond the scope of this chapter.

A. Sampling Windows and Exclusion Rules

A counting window is an artificial boundary on the visible part of the tissue in which counts are made. In order to ensure statistical independence of samples, cells should be countable in only one window. The science of counting windows developed for sections through three-dimensional tissues (see Sec. VII) also apply to two-dimensional retinal whole mounts. For instructional purposes, it is useful to imagine a retinal whole mount that is completely covered by contiguous count-

Figure 2 Unbiased counting frames. A restricted field of view of tissue, delimited by thin dashed lines, contains profiles of cells cut in cross-section. In addition to the profiles completely within the frame, one counts all profiles partially within the frame as long as they do not touch or cross the solid exclusion lines. Thus, only the hatched cells are counted as “in.” (From Ref. 15.)

ing windows, like wallpaper. In this example, many cells fall on boundaries between adjacent windows, but they can be considered “in” only one window in order to avoid overcounts, and thus rules for determining what is “in” are required. In the unbiased counting window shown in Figure 2 [15], one counts all the profiles completely inside the frame provided that they do not in any way touch or intersect the full-drawn exclusion edges or their extensions. In our studies, cells that crossed the left and bottom borders of the counting window were considered “in,” and cells that crossed the right and top borders of the counting window were considered “out.” These rules apply regardless of whether counting windows are contiguous or widely spaced. The size of the counting window in tissue dimensions is determined by calibrating the optical and video components of the system with a calibrated micrometer slide.

B.Systematic Random Sampling for Determining Total Cell Number

For statistically meaningful counts, the sites where counts are made must be chosen independently of their content [15]. That is, sites cannot be chosen because they have features of interest. The choice of sampling pattern across the retina is dictated by whether totals or topography is the outcome variable of interest.

If the total number of cells is desired, then the method used in a recent study of cone survival in rd/rd mouse retina is appropriate [16]. These investigators randomly positioned a regular grid over the tissue, and they counted cells in windows located at 50 evenly spaced sites within the grid. In this approach, the regional specializations containing higher density of cells are sampled in proportion to their retinal area. A regional specialization such as a fovea or area centralis is small relative to the entire retina and contributes little to the total number of cells. In human retina, the small and densely packed foveal cones constitute only 0.7% of the total number of cones (32,000 vs. 4.6 million), because they occupy only a tiny area. It would therefore be possible to skip the entire fovea in a systematic sampling scheme and not appreciably change the total number of cones. However, topographic information is clearly lost by such an approach.

C.Systematic Sampling for Visualizing Topography

If analysis of retinal topography is planned, then the sampling scheme must take into account the presence of regional specializations. An efficient sampling scheme is one that balances the competing goals of avoiding artifacts caused by interpolating densities between widely spaced points (undersampling) and to avoid collecting more data than necessary to achieve a density estimate with tolerable error (oversampling). Cell densities in primate retina are approximately radially symmetric around the fovea and change most rapidly near the fovea. In our topography studies we counted photoreceptors and/or ganglion cells at 100– 120 locations that were closely spaced in the fovea and less closely spaced away from the fovea. We used a spiral pattern that evenly tesselated the retinal surface (Fig. 3).

D. How Much Is Enough?

It is possible to obtain smooth contour maps, reflective of low variance data, with remarkably few samples. In our studies of aging and degeneration, we counted cells within a 12-mm-diameter area centered on the fovea. We counted photoreceptors in a single 39- m-square counting window and ganglion cell layer neurons in adjacent 39- m-square windows until a total of 15 cells was obtained. Approximately 1385 cones (0.17% of the total within central 12 mm diameter), and 1815 ganglion cell layer neurons (0.3% of the total) were counted in these studies. We found empirically that this sample size was adequate to create a smooth map (see Sec. VI.B). In a recent study of mouse retina [16], 2.85% of the total number of counts were sampled. The validity of the sample size in this case was determined

Figure 3 Sampling points used to produce smooth maps of cell density in human retina. (A) Cells were counted at each point. Center of spiral pattern is the point of highest cone density in the fovea. (B) Central portion of A, showing a denser grid of sampling points in the fovea. (From Ref. 17.)

by showing that counting twice or three times as many cells did not change the outcome. It is possible that even fewer cells could have been counted.

VI. METHODS FOR ANALYZING TOPOGRAPHIC

CHANGES

A.Retinal Model

The retina, being part of sphere, must be cut to be flattened. This is not a problem for generating counts of total numbers [16]. On the other hand, the cut edges may interfere with topographic analysis. For our topography studies, we developed a method for reconstructing the human retina by aligning the cut edges of a threepiece whole mount using major retinal vessels as landmarks [17]. Our studies of aging and degeneration used only the macula, and reconstruction of the entire retina was not necessary. In both cases (whole retina or macula), we found it convenient to use a digital model consisting of locations on the retinal sphere that were indexed by spherical coordinates and an associated cell density [17]. The spherical coordinates were referenced by the center of highest cone density in the foveal center, and the directions nasal [l80°] and temporal [0°] referred to the appropriate sides of a line passing through the foveal center and perpendicular to a line through the fovea and optic disc center of a standard left eye. This coordinate system allowed us to combine maps from eyes of different individuals. Note that

our coordinate system was ideally suited for human retina, because the fovea, the point of highest cone density, and the optic disc were readily identified internal landmarks. However, a coordinate system for a species without grossly visible retinal landmarks should be done in reference to eye position in the head, a process that begins with marking directions on the eyes when they are still in the skull.

Data points in our retinal model were connected into a mesh of triangular patches which closely approximated a sphere. A value at any point within a patch could be determined by calculating a weighted average of the values at its three vertices, yielding an acceptable approximation to the value found in the tissue itself at that location. Computer graphic display techniques were used to produce a variety of images. The values associated with vertices of the triangular patches were linearly interpolated across the patch. Video look-up tables were manipulated to produce false-colored smooth surfaces, terraced surfaces in false color (Fig. 6, Ref. 4) and gray scale (Fig. 4), and traditional black-and-white contour maps [18]. Plots of average cells per square millimeter as a function of eccentricity along selected meridians and maps of average density were created by resampling models of individual eyes at a set of standard locations, in which retinal directions were preserved. The total number of cells in selected regions was calculated by multiplying the mean density of each triangular patch on the model by its area on the spherical surface of the retina.

B. Graphical Methods

A map is the most direct way to inspect retinal topography. We displayed our data in the polar azimuthal equidistant projection, familiar from human perimetry, which preserves radial distances and distorts circumferential distances. In addition to maps of individual eyes we created maps of mean cell density in a group of eyes by resampling data from individual eyes at standard locations that reflected the weighting of the original sample points—that is, with more locations near the fovca (Fig. 5A,B) [4].

A difference map is a comprehensive and effective way of seeing overall differences in retinal topography. We created maps of the difference between groups using a measure of local variability, so that a difference in a location where variability was high would be displayed less prominently than a difference in a location where variability was low. In our study of aging human macula [4], cell density in an older age group (test population) was compared to a younger age group (reference population). We computed the difference between log (density) at each standard location in all possible pairs of test and reference eyes. The mean of the pairwise differences at each location was used to create a colorcoded map. Figure 4C shows a gray-scale version of the difference map, and a color version is shown in Figure 6G of Ref. 4. This approach was used to demonstrate the exquisite localization of age-related rod loss in the human macula.

Figure 5 Rod and cone loss in exudative age-related macular degeneration. (A) Sampling pattern for photoreceptor counts in eyes with exudative age-related maculopathy (ARM). The small oval indicates the optic disc. The large irregular pattern indicates an area of almost complete photoreceptor loss overlying fibrovascular scars and atrophy of the retinal pigment epithelium (RPE). To quantify rod and cone loss, counts were made at 0.1 mm intervals along 4–8 arbitrarily placed meridians (hatched lines). Counts were also made at comparable locations in 3–4 age-matched control eyes. (B) Rods but not cones are lost around the margins of fibrovascular scar and RPE atrophy in an eye with exudative ARM. (Dark circles, rods; gray circles, cones.) (From Ref. 5.)

Figure 6 Disector. A block of retinal tissue (A) containing particles a–e is cut into serial sections (B). Hatched profiles in B–D are cells that are transected by the sectioning plane. A pair of sections, distance t apart, is drawn from the stack. The top (reference) plane contains an unbiased counting frame of area A (C), and the lower (look-up) section does not (D). The volume of the disector is A t. Only cell b is counted as “in,” because it falls within the counting frame in the reference section (C), and it is not present in the look-up section (D). This diagram shows a physical disector (i.e., identifying cells in separate sections) An optical disector involves counting cells identified in the top and bottom focal planes of the same thick section. (From Ref. 23.)