Morphological Classification of Parvalbumin-Containing Retinal Ganglion Cells in Mouse: Single-Cell Injection after Immunocytochemistry

Abstract

purpose. Matching the neuron’s morphology with its expression of a particular protein cannot be easily achieved by immunocytochemistry alone, as many proteins are expressed too weakly. In this study, a newly developed method was adopted to match mouse retinal ganglion cell (RGC) morphology with its expression of parvalbumin.

methods. Parvalbumin-containing ganglion cells were first identified by immunocytochemistry and then were iontophoretically injected with a lipophilic dye DiI. Then confocal microscopy was used to characterize the morphologic classification of the parvalbumin-immunoreactive (IR) ganglion cells on the basis of the dendritic field size, branching pattern, and stratification within the inner plexiform layer.

results. The results indicate that at least eight different types of ganglion cells express parvalbumin in the mouse retina. They were heterogeneous in morphology: monostratified to bistratified, small-to-large dendritic field size, and sparse-to-dense dendritic arbors.

conclusions. Single-cell injection, after immunocytochemistry, provided the first means to identify the detailed functional anatomy of parvalbumin-containing RGCs in the mouse retina. The combined approach of cell morphology and the selective expression of parvalbumin will not only provide useful data for further correlation of physiological properties of the RGCs, but it will also provide a useful strategy for matching a neuron’s morphology with its expression of a particular protein.

Knowledge of the diverse morphologies and dynamic functions of retinal ganglion cells (RGCs) is the cornerstone of our understanding of retinal function. As the morphology of RGCs is a direct reflection of the underlying neural connectivity and as it remains the most common way of identifying and classifying retinal neurons, great attention has been focused on the development of diverse methods to identify individual neurons in isolation from their neighbors. ¹ Since Cajal’s great contribution to the recognition of individual neurons using Golgi stain, many advanced techniques have been developed to detail the morphology of single types of neurons. One is the microinjection with neurobiotin, horseradish peroxidase (HRP), or fluorescent dyes. ² The second is the particle-mediated introduction of lipophilic carbocyanine dyes by particle bombardment. ³ The third is photofilling with photooxidized rhodamine introduced either by diffusion or retrograde tracer. ^{4 5 6} The fourth is production of genetically encoded fluorescent or histochemical reporters, alkaline phosphatase, or green fluorescent protein. ^{7 8 9 10} All these methods show individual neurons in isolation from their neighbors for the morphologic analysis of neurons. 

The morphologic and functional properties of RGCs have been studied in extensive detail in monkeys and cats. ^{11 12 13 14} Two major types of ganglion cells, more than any other ganglion cell types, are distinguished in the monkey retina: Large magnocellular (parasol) and smaller parvocellular (midget) ganglion cells. The parvocellular ganglion cells have very small dendritic fields, whereas the magnocellular ganglion cells have much larger ones. Recently, however, at least eight previously unknown, morphologically distinct RGCs that project into the lateral geniculate nucleus were revealed in the monkey, by using an in vitro photostaining method. ⁶ Ganglion cells similar to magnocellular and parvocellular ganglion cells are found in other animals. In the cat, approximately 45% are small β cells (physiologically X cells), and 5% are large α cells (Y cells). At least eight different morphologic types of non-β and non-α types of ganglion cells have been revealed, by using intracellular staining. ^{15 16 17 18} Moreover, as it is expected that RGCs are heterogeneous, Rockhill et al. ¹⁹ meticulously distinguished at least 11 types of RGCs in rabbit retina, by using a combination of modern anatomic techniques. Recently, more than 10 morphologically distinct ganglion cell types have been revealed in the mouse retina by the introduction of particle-mediated gene transfer ^{20 21} and by the expression of the gene encoding an alkaline phosphatase. ¹⁰ The morphologic classification of the RGCs was achieved by using three parameters: level of dendritic stratification within the inner plexiform layer, extent of the dendritic field, and density of branching. 

The importance of calcium in the regulation of several intracellular metabolic processes is hard to overemphasize. Calcium-binding proteins are thought to play important roles in regulating intracellular calcium in the central nervous system. Considerable attention has been focused on the localization of calcium-binding proteins, because they serve as good markers to distinguish subpopulations of neurons ^{22 23} and because their distribution is altered in many neurodegenerative disorders. ^{24 25 26 27 28 29} Among the many calcium-binding proteins, parvalbumin is one that is expressed usually in specific fast-firing neurons. It consists of a single, unbranched chain of linked amino acids and has a molecular mass near 12,000 Da. ³⁰ Although there is some degree of species-specific variation, subpopulations of ganglion, amacrine, bipolar, and horizontal cells in the vertebrate retina are immunoreactive for parvalbumin. ^{31 32 33} However, there have been no reports showing the individual parvalbumin-containing retinal ganglion neurons in isolation to confirm a unique cell type. 

Morphologic types of RGC types can be identified neurochemically by using an array of different markers. ³⁴ Although these markers allowed specification of a molecular phenotype of each ganglion cell and a comprehensive classification, they do not reveal the entire cell’s arborization. At present, there are no reports of a survey of detailed RGC morphology that match them with the expression of a particular protein in any animal. Thus, the selective expression of the protein in combination with RGC structure remains unanswered. 

Antibodies against α-melanocyte-stimulating hormone and heavy neurofilament subunit have been labeled in a Golgi-like subpopulation of ganglion cells in the retina. ^{35 36} However, we are the first to attempt systematic surveys of the characterization and classification in shape of particular protein-containing RGCs in the retina. We used the mouse, as it is the mainstay of transgenic technology. Our approach was somewhat different from that used in previous studies. In the present study, we adopted a newly developed method to match a neuron’s morphology with its expression of a particular protein. ³⁷ We first identified parvalbumin-containing mouse RGCs by immunocytochemistry and then injected them iontophoretically with the lipophilic dye DiI. We finally used confocal microscopy to characterize the detailed morphologic classification of the parvalbumin-immunoreactive (IR) ganglion cells in the mouse retina. Our approach achieved a matching of the RGC morphology of mouse with its expression of parvalbumin. 

Materials and Methods

Animal Procedures

Adult mice (C57BL/6J) were used in these experiments. Animals were anesthetized with a mixture of ketamine hydrochloride (30–40 mg/kg) and xylazine (3–6 mg/kg). Proparacaine HCl (100–200 μL) was applied to the cornea to suppress blink reflexes. Eyes were quickly enucleated after a reference point was taken to label the superior pole, and they were immersed in 0.1 M phosphate buffer (pH 7.4). The animals were euthanatized by an overdose of the same anesthetics. All investigations involving animals conformed to the guidelines of the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. 

Immunocytochemistry

Immediately after enucleation, the retinas were carefully isolated from the eye cup in 0.1 M phosphate buffer (pH 7.4) and mounted onto a black, nonfluorescent filter membrane (HABP; Millipore, Bedford, MA) ganglion cell side up. The filter membrane and the attached retina were fixed for 0.5 hours in a dish of 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). All the retinas were then rinsed three times with ice-cold 0.1 M phosphate buffer before the immunocytochemical study. For immunocytochemistry, the tissues were incubated in a 1:200 dilution of the primary antiserum, a monoclonal antibody against parvalbumin from Sigma-Aldrich (Temecula, CA), in 0.1 M phosphate buffer for 2 hours. After three rinses in ice-cold 0.1 M phosphate buffer, the tissues were incubated in a 1:50 dilution of FITC-conjugated horse anti-mouse IgG (Vector Laboratories, Burlingame, CA) in 0.1 M phosphate buffer for 1 to 2 hours. 

Cell Injection

The dish containing the immunolabeled retina was placed on a microscope stage, and the FITC-labeled neurons were viewed with a microscope with a water-immersion lens (40× Plan Achroplan, NA 0.80; Carl Zeiss Meditec, Inc., Dublin, CA), using a 100-W mercury source and a filter set (set 09; Carl Zeiss Meditec, Inc.; excitation, 450–490 nm; emission, 515 nm). Single cells in the RGC layer were selected by using an indexed grid reticle which referred to randomly generated grid coordinates. The stage was repositioned so that the randomly selected cell was in the center of the field of view. Cells were filled iontophoretically by passing positive current of 5 to 20 nA with 1% lipophilic dye DiI (1% DiI in 100% ethanol; Invitrogen, Eugene, OR). ³⁷ The DiI-filled cells were viewed using another filter set (set 20; Carl Zeiss Meditec, Inc.; excitation, 540–552 nm; emission, 575–640 nm). The optimal filling time was usually 15 to 30 minutes (Fig. 1) . When the selected cell was filled, we systematically moved the grid reticle 350 μm laterally and injected another cell into the center of the grid. We injected cells only in the midperiphery of the ventral retina, 1.0- to 1.7-mm inferior to the optic disc, as the dendritic field size varies according to retinal eccentricity. After several cells in the retina were filled, the tissue was rinsed briefly in 0.1 M phosphate buffer, and fixed for 2 hours in 4% paraformaldehyde. To stain all cells, the retina was immersed for 1 hour in a nucleic acid dye (10–20 nM SYTO13; Invitrogen), in 0.1 M phosphate buffer (pH 7.4), washed three times in 0.1 M phosphate buffer, and then coverslipped in mounting medium (Vectashield; Vector Laboratories) and sealed with nail polish. Every injected cell was imaged and recorded as part of the database, as shown in Table 1 . 

Data Analysis

The cells were imaged with a laser scanning confocal microscope (MRC 1024; Bio-Rad, Hercules, CA) equipped with a 40× objective (NA 0.75; Plan Fluor; Nikon, Tokyo, Japan) for small cells and 20× objective (NA 0.50; Plan Fluor; Nikon) for large cells, to show the entire dendritic field. Two sets of images were collected of each cell. A through-focus of the DiI-filled cells was imaged using an emission filter (580DF32; Chroma Technology Corp., Rockingham, VT). Subsequently, a through-focus of the nucleic acid-stained (SYTO13; Invitrogen) nuclei was collected using an emission filter 515DF30, to measure the thickness of the inner plexiform layer and to determine the level of stratification of the DiI-filled cells. The dendritic field area and diameter were computed with a digital camera (AxioVison 4; Carl Zeiss Meditec, Inc.) by connecting the distalmost tips of the dendrites. ¹⁴ Images were adjusted in terms of brightness and contrast (Photoshop CS; Adobe Systems, Mountain View, CA). 

Estimates of Parvalbumin-IR Cell Density

Cell density was expressed as the number of parvalbumin-IR cells per square millimeter of retinal surface. In three fluorescence-reacted wholemount retinas, parvalbumin-IR cells in the ganglion cell layer were viewed on the computer monitor with a 20× objective (Plan-Apochromat; Carl Zeiss Meditech, Inc.) and AxioCam HRc digital camera (Carl Zeiss Meditec, Inc.) at 300-μm intervals along the central dorsoventral and nasotemporal axes. The area sampled was 200 × 200 μm. A transparency sheet was placed on the computer monitor, and labeled cells were circled with a pen. In one fluorescence-reacted wholemount retina, in 323 sample areas across the retina, parvalbumin-IR cells were counted. The samples were taken at evenly distributed positions across the retina. Isodensity contours were fitted to the data and coded by a gray scale so that regions of highest density were black, with areas of decreasing density being increasingly lighter gray. The total number of cells was determined in each sample area and expressed as the number of cells per square millimeter. The cell density was multiplied by retinal area to determine the total number of parvalbumin-IR cells. 

Results

Types of Parvalbumin-IR Cells

Two-hundred and 93 ganglion cells from 45 retinas were analyzed in the present study. Ganglion cells were identified by the presence of an axon. For simplicity and to avoid the controversy of ganglion cell types or clusters within the mouse and between species, ^{10 19 20 21 38} we did not depend on preexisting names. Instead, we classified the cells by dendritic field size, branching pattern, and stratification within the inner plexiform layer (IPL) and we used a simple name, parvalbumin cell types 1 through 8 (PV1-8; Table 1 ). It should be noted, however, that each cell type corresponds to the recent classification of Sun et al. ²⁰ A description of cells follows in the order of their dendritic field diameter. 

PV1 cells possessed small, dense dendritic arbors (Fig. 2) . The dendritic field diameter of the cells was 142 ± 25 μm. They were the smallest cells in our sample, and they comprised 11.9% of our sample. The dendritic arbors branched narrowly in the outer part of the IPL (36.9% ± 7.75%; Fig. 3 ). The primary dendrites were thick and tapered from the soma to the periphery. They exhibited abrupt bends and turns along their course. The dendritic field contained many short branches and many small protrusions. Many of the dendrite terminals formed “y” shapes. The dendritic arbor was often asymmetric. It is clear that PV1 cells belong to type B4 (Table 1) . ²⁰  

PV2 cells possessed sparse dendritic arbors that were larger than those of PV1. The dendritic field diameter of the cells was 171 ± 30 μm. They made up 20.1% of our sample. PV2 cells in our sample came in ON and OFF varieties. The dendritic arbors of the ON variety occupied a depth of 62.31% ± 5.67%, whereas the dendritic arbors of the OFF variety occupied a depth of 33.63% ± 7.18%. As in the case of PV1, some PV2 cells had asymmetrical dendritic arbors. Dendritic processes branch from the primary processes that extend from the soma and spread more sparsely within the wider area than the PV1 cells. The dendritic branches were moderately wavy and were mostly longer than those of PV1. PV2 cells appeared to belong to type B3. ²⁰  

PV3 cells possessed medium and medium-dense dendritic arbors. The dendritic field diameter of the cells was 176 ± 28 μm. They made up 6.2% of our sample. These cells were bistratified in the IPL at a depth of 65.92% ± 6.12% and 29.95% ± 7.12%. These cells had thin, recursive, loop-forming dendrites. PV3 cells are very similar to the ON-OFF direction-selective cells that have been extensively described in the rabbit retina. ³⁹ PV3 cells correspond to type D2. ²⁰  

PV4 cells possessed medium-to-large and medium-dense dendritic arbors. The dendritic field diameter of the cells was 194 ± 19 μm. They made up 12.3% of our sample. The dendritic arbors barely branched in the outer part of the IPL (44.02% ± 9.15%). They had three to five primary dendrites. The primary dendrites tapered little from the soma to the periphery. The primary and secondary dendrites exhibited frequent smooth curves. The dendritic arbor covered the dendritic field evenly. PV4 cells correspond to type C5. ²⁰  

PV5 cells possessed medium-to-large, relatively dense dendritic arbors. The dendritic field diameter of the cells was 197 ± 27 μm. They made up 11.6% of our sample. The dendritic arbors branched in the outer part of the IPL (38.52% ± 7.95%). They had three to six primary dendrites, which were generally smooth, with compact and densely branched dendritic arbors. Many branches were curvy, and some recurved. The dendritic arbors overlapped in many fields of the cells. PV5 cells correspond to type C4. ²⁰  

PV6 cells possessed medium-to-large and sparse dendritic arbors. The dendritic field diameter of the cells was 222 ± 27 μm. They made up 18.1% of our sample. PV6 cells in our sample came in ON and OFF varieties. The dendritic arbors of OFF variety occupied a depth of 40.97% ± 8.62%, whereas the dendritic arbors of ON variety occupied a depth of 63.78% ± 7.00%. These cells were characterized by irregular-looking dendritic arbors that lacked an orderly, radiating pattern. They emitted from the cell body at any distance and they often crossed one another. PV6 cells appeared to belong to type C2. ²⁰  

PV7 cells possessed medium-to-large, medium-dense dendritic arbors. The dendritic field diameter of the cells was 240 ± 28 μm. They made up 3.7% of our sample. The dendritic arbors branched narrowly in the inner part of the IPL (61.36% ± 7.27%). These cells had three to five thick, primary dendrites, and the dendritic processes rarely crossed one another. They tended to be distributed uniformly across the dendritic field, giving the impression of a maze. PV7 cells appeared to belong to type C1. ²⁰  

PV8 cells possessed large, sparse dendritic arbors. The dendritic field diameter of the cells was 289 ± 34 μm and they were the largest group of cells in our sample. They made up 11.3% of our sample. The dendritic arbors branched in the inner part of the IPL (70.01% ± 5.78%). The three to six thick, primary process extended radially from a large cell body, emitting several long branches along the way. PV8 cells are the ON variety of α-cells of most mammalian retinas. PV8 cells corresponded to type A1. ²⁰  

In the present study, there were 14 unclassifiable parvalbumin-IR RGCs. A negligible kind of unclassifiable cell is represented by scarce cells in our sample. These cells were very rare and could not be reproduced by morphologic characteristics. Mostly, we had only a single clear example in each case. We think they developed by accident or in transitional cases and thus could not be included in the present analysis. These 14 unclassifiable parvalbumin-IR RGCs varied considerably in dendritic field size and in dendritic arbors. The dendritic field diameter of the cells was 105.39 to 234.42 μm and the dendritic field area was 8,633.19 to 42,967.21 μm². Examples are shown in Figure 4 . The cells in Figure 4A(dendritic field diameter, 133.05 μm) and 4B (dendritic field diameter, 130.76 μm) possessed very small and very spare dendritic arbors. These cells are smaller than our smallest parvalbumin-IR cell (PV1) and had much sparser dendritic arbors than any other cells in the present study. Another example is shown in Figure 4C . This cell possessed medium-to-large but sparsely branched dendritic arbors (210.89 μm). Thus, the cell in Figure 4Cresembled the PV5 or PV6 in dendritic field size, but had distinctive density and dendrites. These kinds of distinctive features and their scarcity made us hesitate to force them into a specific single classification. If these 14 unclassifiable parvalbumin-IR RGCs are not developmental accidents or transitional cases, a sample as large as ours should have revealed at least several examples. Similar to the present result, unclassifiable scarce and transitional forms of RGCs and amacrine cells were found in rabbit retina in recent classification studies. ^{4 19}  

Density of Parvalbumin-IR Cells

The estimated total number of parvalbumin-IR RGCs in the retina varied from 14,328 to 15,717 cells among the four retinas sampled. Table 2and Figure 5show the result. There were 15,586 cells in retina 1, 14,896 cells in retina 2, 15,717 cells in retina 3, and 14,328 cells in retina 5. Therefore the average number of parvalbumin-IR cells per retina was 15,131 ± 645 (mean ± SD; n = 4; Table 2 ). The distribution of total parvalbumin-IR neurons are shown in Figure 5 . In Figure 5A , parvalbumin-IR cell density is given as isodensity lines which encircle the area of higher density. The two graphs (Figs. 5B 5C)show the number of cells encountered along the dorsoventral and nasotemporal axes intersecting the optic nerve head. In addition to RGCs, however, some displaced amacrine cells were also labeled by parvalbumin. ³² In the present study, 14.07% (48/341) of cells were axonless cells and 85.93% (293/341) were ganglion cells identified by the unequivocal presence of an axon. Thus, among the 15,131 ± 645 parvalbumin-IR cells in the ganglion cell layer, 2,128 ± 91 cells were displaced amacrine cells and 13,003 ± 554 cells were ganglion cells. Since the total ganglion cell density was 44,857 in the mouse retina, ⁴⁰ 28.98% (13,003/44,857) of RGCs were parvalbumin-containing RGCs. The parvalbumin-IR ganglion cell density was 724 cells/mm² in the mouse retina. The predicted number of each cell type is shown in Table 1 . 

On the basis of the previously known percentage of each ganglion cell type of Sun et al. ²⁰ (Table 1) , we estimated the proportion of the ganglion cell subtypes that express parvalbumin. The estimated proportion of parvalbumin-IR RGCs in each subtype varied, 11.59% to 72.77%. The proportion of the RGCs that express parvalbumin was 60.50% in PV1, 54.95% in PV2, 11.59% in PV3, 31.84% in PV4, 56.98% in PV5, 51.43% in PV6, 34.59% in PV7, and 72.77% in PV8. The highest number was found in PV8, and the lowest number was found in PV3 (Table 1) . The precision of these estimates is limited by the small number of examples of each cell type, but the general result is clear. 

Discussion

In the present study, we identified parvalbumin-IR RGCs in the fixed retina of the mouse by using a combined method of immunocytochemistry and DiI injection. 

The current evidence strongly indicates that the family members of each major cell type in the retina have all been identified. It is clear that mammalian retinas contain a variety of cell types. ¹ The systematic surveys indicate that the human, monkey, cat, rabbit, and mouse each contain approximately 10 to 15 types of RGCs. These data are mainly confirmed by using a variety of modern cell-filling methods: A single cell injection with diffusible markers, such as Lucifer yellow or neurobiotin; a sparse delivery of lipophilic dyes by particle bombardment; expression of genetically encoded histochemical or fluorescent reporters; and a photofilling method. ^{10 41} RGCs contain many proteins, and immunocytochemistry is a powerful technique used to assess the presence of a specific protein in the retina. Many proteins, however, do not express enough to reveal the cell’s morphology. In many cases, immunocytochemistry after dye-filling usually blocks immunostaining. In addition, a neuron filled with an injection randomly is unlikely to be a molecule of particular concern. In the present study, we identified parvalbumin-IR cells by immunocytochemistry and then injected these cells with the lipophilic dye DiI. This strategy is especially useful for matching a neuron’s morphology with its expression of the protein of interest. In the experiments presented herein, we have chosen the first identified calcium-binding protein to demonstrate the method in the mouse retina. An extension of this method would be to use cell labels other than parvalbumin. The combined approach of cell structure and selective expression of protein is a more robust method of cell-type classification than strictly morphologic methods. 

This method, however, has some limitations. First, the resolution of cell morphology sometimes does not equal that of the intracellular filling in the living retina, as the injection occurs after immunocytochemical treatment of the fixed retina. Second, it will be difficult to apply this method to the staining and injection of cells in the inner nuclear layer of wholemounts. To retain cell morphology, we did not use any detergents such as Triton X-100, which will prevent antibody penetration from being labeled in the inner nuclear layer. Rigorously permeabilizing tissue with a harsh detergent prevented the dye from spreading in the neuronal membrane. ³⁷ However, the present method has been used for bipolar cells in the inner nuclear layer in vibratome-cut sections. Targeted injection with lipophilic dye identified the morphology of GABAergic bipolar cells. ³⁷ Third, it is difficult to obtain precise information about the density and percentage labeling of the cell population of each parvalbumin-IR cell if the spatial density of each RGC type is absent. ^{10 20 38}  

Previous work ^{31 32 33} and the present study on localization of parvalbumin-IR cells in the various mammalian retinas have revealed some non—parvalbumin-labeled large cell somas (e.g., α-cells). These data support the idea that parvalbumin-IR RGCs, if not all types, probably belong to the different chemical subpopulation from each type. 

Our present study provides evidence that all the eight types of parvalbumin-IR RGCs are subpopulations of each RGC type. The estimated proportion of parvalbumin-IR cells of each RGC type revealed a considerable mismatch of parvalbumin expressional pattern. Only approximately one tenth of cells of one type expressed parvalbumin type, whereas approximately three fourths of cells of another type expressed parvalbumin. Although the precision of the estimated parvalbumin-IR cell density among each ganglion cell type is limited by the small number of each cell type, these results clearly suggest that each RGC type in mouse differentially express parvalbumin. The results also suggest that the RGCs can be subdivided on the basis of protein properties. One possible reason for the differential expression of parvalbumin in each type could be differences in physiological diversity. Different proteins can have different physiological properties and could be involved in different synaptic circuits. ¹ Another possible reason could be differences in the retinal projection of ganglion cell subtypes. Neuropeptide Y presents in a subpopulation of approximately 2000 γ-type RGCs that project into the superior colliculus and C layers of the lateral geniculate nucleus. ⁴² Most calretinin-IR fibers in the superior colliculus originate from small γ-type cells. ⁴³ In general, calcium-binding proteins are selectively expressed in the subpopulation of neurons. ²³  

In our previous report, ⁴⁰ we found that the total population of mouse RGCs is approximately 45,000. Our present data provide that 28.98% (13,003/44,587) of RGCs in the mouse express parvalbumin. It is thought that the RGCs of the mouse are diverse. ^{10 20 21} Recently, Sun et al. ²⁰ classified at least 14 types of mouse RGCs based on a large random-sampling method, using the sparse delivery of lipophilic dyes by particle bombardment. They classified monostratified cells into three groups: RG_{A1- A2} cells (large soma, large dendritic field), RG_{B1- B4} cells (small-to-medium-sized soma, small-to-medium-sized dendritic field), and RG_{C1- C6} cells (small-to-medium-sized soma, medium-sized to large dendritic field). Bistratified cells were classified as RG_{D1- D2}. Among these 14 types, our results indicate that at least eight different types express parvalbumin. These were RG_A1, RG_B3, RG_B4, RG_C1, RG_C2, RG_C4, RG_C5, and RG_D2. ²⁰ Our results, however, show that no single type of parvalbumin-IR RGCs dominated. In addition, there was no systematic correlation between parvalbumin-IR RGCs with cell morphology. With respect to the cell types, at least one conclusion can be drawn from these studies: the morphologic heterogeneity of parvalbumin-containing RGCs in the mouse retina. They were monostratified to bistratified, small-to-large dendritic field size, and had sparse-to-dense dendritic arbors. 

Although the physiological role of parvalbumin is still unclear, it is thought to play a major role in buffering the intracellular calcium level and regulation of various enzymes. Impaired regulation of calcium by parvalbumin is closely related to many neurodegenerative disorders. ^{23 24 30} Recent studies reflect that parvalbumin is closely linked to activity-dependent changes in retina. For example, the intensity of parvalbumin immunoreactivity in the retina showed an underlying circadian rhythm that is enhanced by cyclic light. ⁴⁴ In the retina of diabetics, the expression of parvalbumin is also increased. ⁴⁵ In contrast, a decrease in parvalbumin is associated with the differentiation of retinoblastoma cells. ⁴⁶ A recent study showed that there is no preference for the degeneration of any one type of mouse RGC in glaucoma. ⁴⁷ However, it is questionable whether the vulnerability of parvalbumin-containing RGCs is different from ganglion cells that do not contain parvalbumin. The current data showed the selectivity of the expression of parvalbumin in subpopulations of RGCs in the mouse. This may reflect the functional needs of chemically defined subpopulations of RGCs in different visual behavioral contexts. It may be possible that the parvalbumin-containing ganglion cells are especially resistant or vulnerable to degeneration, at least in a temporal way. 

In conclusion, matching the neuron’s morphology with its expression of a particular protein cannot be easily achieved by immunocytochemistry alone, as many proteins are expressed too weakly. Single-cell injection, after immunocytochemistry, provides the first means of identifying the detailed functional anatomy of parvalbumin-containing RGCs in the mouse retina. Parvalbumin-containing mouse RGCs were heterogeneous in morphology. Our results indicate that at least eight different types of mouse RGCs express parvalbumin. The combined approach of cell morphology and the selective expression of parvalbumin will not only provide useful data for further correlation of physiological properties of the RGCs, but it will also provide a useful strategy in identifying cell types and for matching a neuron’s morphology with its expression of a particular protein in the future. 

 Supported by Grant R01-2003-000-10505-0 from the Basic Research Program of the Korea Science and Engineering Foundation.

 Submitted for publication November 9, 2005; revised February 13, 2006; accepted April 28, 2006.

 Disclosure: T.-J. Kim, None; C.-J. Jeon, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Figure 1.

 

Video images of single-cell injection after immunocytochemistry on the whole-mounted retina (A) Parvalbumin-containing neurons were labeled with a monoclonal antibody against parvalbumin. (B, C) A DiI-filled micropipette targeted the parvalbumin-containing cell. (D–F) A DiI-filled parvalbumin-containing neuron revealed the distinctive morphologic pattern. Note the axon in (D). The axon and cell body were out of focus in (E) and (F). FITC-labeled parvalbumin-containing neurons (A–C) were viewed with a water-immersion lens, using a 100-W mercury source and a Zeiss filter set 09. The DiI-filled cells (D–F) were viewed with another filter set (Zeiss filter set 20).

Table 1.

 

Quantitative Data of Parvalbumin-Containing Mouse RGC in this Study and Their Homologues from Previous Data

Cell Type	Dendritic Field Position (% of IPL Depth; Mean ± SD)	Number in Sample	% of Total Ganglion Cell Filled	Dendritic Field Diameter (μm; Mean ± SD)	Dendritic Field Area (μm2)	Predicted Number (cells/mm2)	Sun et al.20					Estimated % of PV-IR RGC in Each Type
															Type	% of Total RGCs	Classification Scheme (Dendritic Field Size/density)
PV 1	36.90 ± 7.75	35	11.9	142 ± 25	16213	86	B4	5.7	Small/dense	60.50
	62.31 ± 5.67	30				74			Small to
PV 2*	33.36 ± 7.18	29	20.1	171 ± 30	23576	71	B3	10.6	medium/sparse	54.95
	65.92 ± 6.12
PV 3, †	29.95 ± 7.12	18	6.2	176 ± 28	24723	45	D2	15.5	Medium/medium	11.59
PV 4	44.02 ± 9.15	36	12.3	194 ± 19	29823	89	C5	11.2	Medium/medium	31.84
PV 5	38.52 ± 7.95	34	11.6	197 ± 27	31072	84	C4	5.9	Medium/dense	56.98
	63.78 ± 7.00	32				79			Medium to large/sparse
PV 6*	40.97 ± 8.62	21	18.1	222 ± 27	39034	52	C2	10.2		51.43
PV 7	61.36 ± 7.27	11	3.7	240 ± 28	45633	28	C1	3.1	Medium to large/medium	34.59
PV 8	70.01 ± 5.78	33	11.3	289 ± 34	66496	82	A1	4.5	Large/sparse	72.77
Unclassified		14	4.8	170 ± 40	23972	34
Total		293	100			724

*  Monostratified cell type with ON and OFF population.

†  Bistratified cell type.

Figure 2.

 

The eight types (PV1–PV8) of parvalbumin-containing mouse RGCs identified in our study. Boxed areas: the area from which dendrites are shown in higher magnification in PV1′–PV8′. Scale bars, 50 μm.

Figure 3.

 

The branching and levels of stratification of the eight types of parvalbumin-IR mouse RGCs identified in our study. Scale bar, 50 μm.

Figure 4.

 

Examples of unclassified cells. (A, B) Cells possessing very small and very spare dendritic arbors. (C) A cell possessing medium-to-large and sparse dendritic arbors. The cells were clearly filled but were rarely encountered in the present study. Scale bar, 50 μm.

Table 2.

 

Total Parvalbumin-IR Neurons in the Ganglion Cell Layer

Retina	Sampled Area (n)	Sampled Area (mm2)	Neurons Counted	Mean Density (cells/mm2)	Total Retinal Area (mm2)	Total PV-IR Neurons
1	28	1.12	968	864	18.04	15,586
2	29	1.16	980	844	17.65	14,896
3	29	1.16	1,004	865	18.17	15,717
5	323	12.92	10,344	800	17.91	14,328
Mean ± SD		4.09	3,324	843	17.94	15,131 ± 645

*  One sampled area = 200 × 200 μm.

Figure 5.

 

Parvalbumin-IR neurons of the ganglion cell layer. (A) Isodensity map of the distribution of parvalbumin-IR cells in mouse retina reconstructed from wholemount fluorescence immunocytochemistry. The map shows isodensity lines and the density values are given as cells/mm². (B, C) The number of cells encountered along two axes (dorsoventral and nasotemporal, respectively) intersecting the optic nerve head.

The authors thank Graham Harding for proofreading the paper and Richard Masland for helpful suggestions.