Cones Regenerate from Retinal Stem Cells Sequestered in the Inner Nuclear Layer of Adult Goldfish Retina

Abstract

purpose. To determine whether retinal progenitor cells in the inner nuclear layer give rise to regenerated cones after laser ablation of photoreceptors in adult goldfish retina.

methods. Using a technique developed previously in this laboratory, photoreceptors in the retina of adult goldfish were ablated with an argon laser. The mitotic marker, bromodeoxyuridine, was used to label proliferating and regenerated cells, which were identified with cell-specific markers.

results. Cells proliferating locally within lesion included microglia, Müller glia, and retinal progenitors in the inner nuclear layer (INL). The nuclei of both Müller glia and associated retinal progenitors migrated from the inner to the outer nuclear layer. The proliferating retinal progenitors, which express Notch-3 and N-cadherin, regenerated cone photoreceptors and then rod photoreceptors.

conclusions. Previous work has demonstrated that photoreceptors in the goldfish retina regenerate selectively after laser ablation, but the source of regenerated cones has not been identified. The results reported here provide support for the existence of retinal stem cells within the adult fish retina that are capable of regenerating cone photoreceptors. The data also support the involvement of Müller glia in the production of regenerated cones.

The neural retina in adult teleost fish contains several distinct populations of proliferating retinal progenitor cells, and neurogenesis continues in adult fish as part of an ongoing growth process. ^{1 2} These retinal progenitor cells have been identified in the adult retina of several different species of teleost fish, ^{3 4 5 6 7 8 9} by labeling their nuclei with an antibody against proliferating cell nuclear antigen (PCNA) or tagging them with a pulse-label of [³H]thymidine or the thymidine analogue bromodeoxyuridine (BrdU). At the circumferential (ciliary) margin of the retina, a germinal zone consisting of multipotent retinal progenitors produces annuli of new retina that are added appositionally, similar to the rings of new growth in a tree trunk. ^{1 10} The number of neurons added postembryonically to the growing retina is substantial. For example, the retina generated during embryonic stages in goldfish occupies less than 5% of the retinal area present in a 2-year-old fish. ¹¹  

Rod photoreceptors are the last retinal neurons to be generated in the fish retina, and they are added in a second wave of mitotic activity that occurs in specialized rod precursors, not in the circumferential germinal zone, but located within the outer nuclear layer (ONL). ^{3 12 13 14} Rod precursors are most abundant near the germinal zone, in the region of newly differentiated retina, but they are also found within the ONL and scattered across the entire differentiated retina. The purpose of interstitial addition of rod photoreceptors is to maintain a constant planar density of rods as the retina enlarges by stretching. ^{4 13 15} These mitotically active rod precursor cells in the ONL have a restricted lineage, in that they give rise only to rods in the intact retina. ^{3 4 16} Given this readily available source of retinal progenitors dedicated to the production of rods, it is not surprising that rods can regenerate when they are selectively destroyed in goldfish retina and that the selective regeneration of rod photoreceptors is due to enhanced mitotic activity of rod precursors in the ONL. ¹⁷  

Although it has been 20 years since proliferating rod precursors were described in the outer nuclear layer in the adult goldfish retina, the literature also contains repeated references to rare, mitotically active cells in the inner nuclear layer (INL) of the intact (undamaged) retina in larval or juvenile fish, ^{7 13} but only recently have these cells been identified with certainty in adult fish. Because they divide very slowly, they can be reliably labeled only by sustained exposure to mitotic markers such as BrdU. ^{8 18 19} The mitotic progeny of these dividing cells migrate radially outward into the ONL, where they contribute to the ongoing neurogenesis in the adult retina by replenishing the population of rod precursors. ⁸ These dividing retinal progenitors sequestered in the INL represent a continuation, but on a slower time scale, of the final stages of neurogenesis in the larval retina. Earlier studies in larval goldfish retina using longitudinal analysis of[ ³H]thymidine-labeled cells and electron microscopic autoradiography with serial reconstruction showed that rod precursors originate from residual multipotent progenitor cells in the INL, whose progeny migrate along the radial fibers of Müller glia to reach the ONL. ²⁰ Previous studies have also shown that some retinal cells in the INL proliferate in response to mechanical damage or cytotoxic insult, and they have often been observed in studies of retinal damage and/or regeneration. ^{21 22 23 24 25 26 27} Although some of these are Müller cells, ²⁸ the study of Julian et al., ⁸ which demonstrated the existence of progenitor cells in the INL of the undamaged adult fish retina, suggests that some of these may be retinal stem cells. 

The proliferating cells in the INL have many of the characteristics of stem cells found in other adult tissues. ^{29 30} For example, they divide slowly to both self-renew and replenish a population of more specialized progenitor cells (rod precursors) that have a higher rate of mitotic activity, and they respond to tissue damage and cell loss by enhanced proliferation, the purpose of which is to restore the lost cells. Another feature characteristic of adult stem cells is their multipotency—that is, their ability to generate a diversity of specific cell types. However, the only certain fates of the progenitor cells in the INL of the intact, undamaged retina are that they give rise to the rod precursors in the ONL and they self-renew. ⁸ Although their existence was not known at the time of earlier studies on retinal regeneration, ^{17 21 22 25} and the rod precursors were originally suspected to be the retinal progenitors responsible for regenerating the neural retina, in retrospect, these putative retinal stem cells in the INL are the more likely candidates to give rise to the regeneration blastema at the wound margin of the surgically lesioned retina and the neurogenic clusters in retinas damaged by cytochemical toxins or laser lesions. ^{31 32}  

To test the hypothesis that the retinal progenitor cells in the INL have an important property expected in stem cells—that is, that they can regenerate damaged retinal tissue by replacing retinal neurons—we challenged them in a selective regeneration paradigm, in which rods and cones in the adult goldfish retina were destroyed with an argon laser. ¹⁷ Both cones and rods regenerated in the lesioned area, and we asked whether the regenerated cone photoreceptors derive from mitotic cells resident in the INL and whether these mitotic cells express specific markers characteristic of the multipotent retinal progenitor cells in the embryonic retina and in the germinal zone at the ciliary margin. 

Methods

Animals

We obtained adult goldfish (Carassius auratus), with body lengths of 6 to 8 cm and eye diameters of 5.0 to 6.5 mm, from a commercial fishery (Ozark Fisheries, Stoutland, MO). All procedures involving animals conformed to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. 

Lentectomies

Because the goldfish lens is optically adapted to aquatic environments (and all the refractive power of the eye is in the lens), when the fish is taken out of the water, the retina is easier to visualize after lentectomy. ¹⁷ Fish were anesthetized in 0.1% tricaine methane sulfonate (Sigma, St Louis, MO) and an incision was made in the nasal cornea with a 1.5-mm, 15° microscalpel (Becton-Dickinson, Franklin Lakes, NJ), and extended along a crescent by cutting the cornea on the dorsal and ventral margins with microscissors. The temporal edge of the cornea remained attached to the eye, and the cornea was reflected away, revealing the lens. The ligaments holding the lens in place were carefully snipped with microscissors, and the lens was freed from its attachments and extracted with forceps. The cornea was then folded back over the pupil, and the fish was returned to water. Suturing was not necessary, and when attempted, it interfered with corneal clarity. A postoperative healing period of at least 2 weeks was allowed for the cornea to seal and clear. 

Laser Lesions

We used an argon laser (System 920; Coherent, Palo Alto, CA) to produce photocoagulatory lesions in the retinas, as described previously. ¹⁷ Fish were anesthetized and manually held in position, to view the fundus with an ophthalmoscope. We made a series of four lesions in the nasal–ventral quadrant of the retina, with the closest lesion positioned less than 1 mm away from the optic disc. The lesions were arranged in a square with sides that were approximately four lesion-diameters in length. The parameters of the laser were set at a duration of 0.1 second, a spot size of 500 μm, and a power setting of 80, 100, or 130 mW. The goal was to produce lesions that destroyed photoreceptors but spared the inner retina (inner nuclear and ganglion cell layers). ¹⁷ Most of the laser energy is absorbed by the melanin in the pigmented retinal epithelium, and the proximity of the photoreceptors makes them more vulnerable. We chose the laser power to use for each eye after examining corneal clarity, because we had determined empirically that greater powers were needed to produce the desired lesion when the cornea was less transparent. In the ophthalmoscope, lesions appeared as white patches on the retina, and the more powerful burns produced larger and whiter patches. The actual tissue damage caused by the laser, as assessed later by histology, varied with corneal clarity, vitreal clarity, and other unidentified sources of individual variability. Retinas were selected for study when the retinal damage from the laser was largely restricted to the ONL. 

Intraocular Injections

To identify mitotically active cells and their progeny, the thymidine analogue 5-bromo-2′-deoxyuridine (BrdU; Sigma) was injected intraocularly (1 mM in 0.9% NaCl). A small incision was made in the nasal cornea, and a blunt-tipped, 33-gauge needle attached to a microsyringe (Hamilton Company, Reno, NV) was inserted into the vitreal cavity. The amount injected was calculated to produce an approximate concentration of 50 μM BrdU in the vitreous, based on estimates of eye volume. ²² Tissue was processed from 1 hour to 115 days after the BrdU injections. 

Tissue Processing

Fish were anesthetized and decapitated. The eyes were enucleated and processed for cryosectioning as described previously. ³³ For wholemount preparations, fish were dark adapted for 2 hours before death to facilitate separation of the neural retina from the pigmented retinal epithelium. 

Immunocytochemistry

For immunofluorescence, cryosections (5-μm thickness) were prepared as described previously. ¹⁷ Antibodies used included: RET1 (1:500), a monoclonal antibody generated against goldfish retina that labels an unidentified nuclear epitope found in cones (but not rods), horizontal cells, a subset of INL neurons, Müller glia, and ganglion cells ³⁴ ; NN2, a monoclonal antibody (1:1000) generated against goldfish retina that labels an unknown cell-surface antigen on microglia and endothelial cells ²⁸ ; rabbit polyclonal antibodies against goldfish glial fibrillary acidic protein (FGP1, 1:500), a generous gift of Michal Schwartz (Weizmann Institute, Rehovot, Israel); rat anti-BrdU (1:20; Accurate Chemical, Westbury, NY); anti-glutamine synthetase (GS, 1:50) a generous gift of Paul Linser (University of Florida, Gainesville); and the zpr1 monoclonal antibody (1:500) that recognizes an uncharacterized surface epitope on double cones (from the zebrafish monoclonal stock center at the University of Oregon Institute of Neuroscience, Eugene, OR). 

After overnight incubation in primary antibody at 4°C, slides were rinsed and then incubated overnight at 4°C in secondary antibodies conjugated to fluorochromes including 7-amino-4-methylcoumarin-3-acetic acid (AMCA; blue), fluorescein isothiocyanate (FITC; green), and CY3 (red-orange); secondary antibodies were from Jackson ImmunoResearch Laboratories (Westgrove, PA). Alternatively, primary antibodies were visualized with an ABC peroxidase kit (Vectastain; Vector Laboratories, Burlingame, CA), according to the manufacturer’s instructions, except that incubation times were increased to overnight for the secondary antibody and to 4 to 6 hours for the avidin-biotin complex. The enzyme substrate was 3,3′-diaminobenzidine tetrahydrochloride (Sigma). For detection of BrdU in the nuclei, sections were labeled with rat anti-BrdU antibody, as described previously, ³⁴ except that secondary antibodies were conjugated to CY3. Wholemount retinas were also processed using this protocol, except that the primary antibody concentrations were 1:20, primary incubations were 40 to 50 hours in duration, secondary incubations were overnight, and rinse steps were increased to at least 30 minutes. 

In some tissue sections, the number of cells expressing these antibody markers was quantified. All cells designated as double-labeled with two different antibody probes were verified by examination of each fluorescent signal under single channel illumination. An analysis of variance (ANOVA), with a Tukey-Kramer posttest (SAS Institute, Cary, NC) was used to compare data. A schematic drawing of a retinal section (Fig. 1A ) illustrates the criteria used for defining the bounds of the lesion (see the figure legend for more information). Labeled cells were counted in 23 laser-lesioned eyes. 

In Situ Hybridization

Cryosections were processed for in situ hybridization with digoxigenin-labeled cRNA probes, as described previously, ^{35 36} with proteinase K treatment for 2 minutes and 45 seconds. The signals were visualized by alkaline phosphatase histochemistry with 4-nitroblue tetrazolium chloride and 5-bromo-4-chloro-3-indolyl phosphate solution (NBT-BCIP) as the substrate. The cRNA probes were generated according to protocols in the instruction manual from Roche Molecular Biochemicals (formerly Boehringer-Mannheim Biochemicals; Indianapolis, IN). The zebrafish N-cadherin (cdh2) antisense probe was generated from a full-length cDNA in the plasmid pBS (the generous gift of Benjamin Geiger, Weizmann Institute, Israel), linearized with the restriction enzyme HndIII, and transcribed with T7 RNA polymerase. The goldfish Notch-3 antisense cRNA probe was a mixture of two probes generated from partial-length cDNA clones (Gotch TM and Gotch CDC), as described previously. ³⁷ Sense control probes were prepared for all clones. 

Results

Identification of the Lesioned Area

The argon laser permits the accurate placement of focused photocoagulation lesions that can be limited to the outer retinal layers. ^{17 38 39} Figure 1B shows a wholemount preparation of a retina at 115 days after lesion, mounted vitreal surface up. Circular patches of BrdU immunoreactivity mark the four lesions. In this preparation, the proliferating cells were labeled with BrdU during the first week after the lesion. A higher magnification of one of the lesions, double immunolabeled for BrdU and glial fibrillary acidic protein (GFAP), is shown in Figure 1C . The end feet of the radial processes of the Müller glia at the inner limiting membrane can be visualized with the FGP1 polyclonal antibodies prepared against goldfish GFAP. The BrdU-labeled cone nuclei, out of focus because they were at a deeper focal plane, mark the lesion site. In the uninjured retina surrounding the lesion, Müller glial end feet were distributed in a regular array with uniform spacing, but within the lesion the end feet were disorganized. The radial processes of the Müller cells were pulled toward the center of the lesion, a distortion that is consistent with the defect in the outer retina that resulted from the photocoagulation (Fig. 1A) . 

Proliferation of Cone Progenitors after Laser Lesion

To identify the progenitors of the regenerated cones, we first had to determine at what point the regenerated cones were produced after the lesion and then to systematically identify the progenitor cells that were proliferating in the lesion area during that period and follow them at subsequent intervals to determine which may give rise to cones. Our earlier study of cone regeneration in laser-lesioned goldfish retina ¹⁷ had shown that BrdU injected within the first week after laser lesioning (3–8 days) is incorporated into cells that differentiate into cones. The results of the present study confirm the previous findings. Figure 1D illustrates BrdU-labeled, regenerated cones from a retina in which BrdU was injected into the eye twice, at 5 and 8 days after laser lesion and the retina processed for immunocytochemistry at 20 days, at which time, the regenerating cones had begun to differentiate. The section was triple-labeled with RET1 (blue), anti-GFAP (green), and anti-BrdU (red). The RET1 monoclonal antibody labeled the nuclei of some retinal cells in all three cellular layers of the retina, including cones (but not rods), subsets of neurons in the INL, Müller glia, and retinal ganglion cells. ³⁴ The GFAP antibody labeled the radial processes of the Müller glia, as described. Within the lesion, the nuclei of regenerated cones were double-labeled with RET1 and BrdU, which were both localized to the nucleus, and the double-labeled nuclei were therefore pink to pale violet (inset, Fig. 1D ). Additional evidence that the RET1/BrdU-labeled nuclei were regenerated cones is that these cells had begun to elaborate apical processes (inner and outer segments) that were characteristic of photoreceptors, but smaller than the adjacent (surviving) cones (data not shown; see reference ¹⁷ ). In summary, both histologic and immunocytochemical observations suggest that regenerated cones were produced by mitotic progenitors that were dividing during the first week after laser ablation. 

A few BrdU-labeled nuclei were also located in the INL of these retinas at 20 days after laser lesioning (Fig. 1D) . Some of these nuclei were double labeled with RET1 and were associated with GFAP-positive fibers (Fig. 1D , and data not shown), suggesting that they were nuclei of Müller cells. ²⁸ However, other BrdU-labeled nuclei in the INL showed no RET1 immunoreactivity, suggesting that they were not Müller glia. These cells may have been regenerated INL neurons, because we cannot exclude the possibility that the laser damaged cells in the INL, some of which do not express the RET1 antigen (discussed later). Alternatively, they may have been undifferentiated INL progenitors, because one of the hallmarks of stem cells is a capacity to self-renew. 

Some support for this latter suggestion comes from the following observations. When BrdU was injected once at either 3 or 5 days after laser lesion and the retina was processed at 25 days after lesion, cone nuclei were also labeled with BrdU. However, the BrdU label was weaker than it was after two BrdU injections, in that the label did not fill the nuclei but was instead granular in appearance (data not shown). This observation suggests that the cells giving rise to cones at 3 to 5 days underwent several cell divisions before differentiating into cones. In contrast, some cells in the INL at 20 to 25 days had strong BrdU labeling after a single injection at 3 or 5 days, suggesting that they underwent one or at most only a few mitotic divisions, which would be typical of stem cell self-renewal (data not shown). 

Regeneration of Rods

Although most of the progeny of cells proliferating during the first week differentiated as cones, a few regenerated rods were also labeled with BrdU, and the magnitude of cone genesis relative to rod genesis decreased with time. For example, Figure 1E shows a retina processed at 30 days after the lesion, after BrdU injections at 14 and 17 days. Fewer cone than rod nuclei were BrdU labeled. The rods did not express the RET1 epitope and were therefore not double labeled. They were identified by their position in the ONL, their small oval nuclei, and the absence of RET1 staining (inset, Fig. 1E ). This temporal pattern of photoreceptor genesis during regeneration is reminiscent of the order of cell production during normal development, in that cones are born before rods. ^{17 20 40}  

Progenitors Giving Rise to Regenerated Cones

The foregoing results are consistent with our earlier study, ¹⁷ which demonstrated that most regenerated cones are produced within the first 2 weeks after the lesion. To identify the cells that were dividing at the beginning of cone genesis, we injected BrdU at 2 to 5 days after the laser lesion and processed the retinas 1 to 5 hours after the injection. This paradigm allowed sufficient time for incorporation of BrdU, but not enough time for complete cell division or differentiation, and it therefore labeled the cells that were proliferating at the time of injection. To trace the origin of the regenerated cones, we injected BrdU during the first week after the lesion and then observed the locations of the labeled cells at intervals over the next 2 weeks. The BrdU-labeled nuclei in each layer were counted, and the proportion that expressed the RET1 antigen was determined. 

From previous studies we knew that retinal damage produced by a number of different causes (e.g., mechanical trauma, chemical toxins, ischemia) produces a progressive series of cellular reactions. These include invasion and activation of phagocytic cells (microglia and blood-borne macrophages), activation of Müller cells, enhanced proliferation of neural progenitor cells, including rod precursors and cells in the germinal zone. We used the RET1 monoclonal antibody to identify proliferating Müller glia (Figs. 1F 1G) , and the NN2 monoclonal antibody to distinguish them from microglia and endothelial cells (Fig. 2A ). We assumed that proliferating cells in the INL that stained with neither RET1 nor NN2 were retinal progenitor cells, and those in the ONL were rod precursors. We know that the latter do not express RET1, and we assume that the former also do not, because RET1 is not expressed at detectable levels in any other proliferating neural progenitor cells in the embryonic or adult retina. 

Figures 1F and 1G show the retina from an eye injected with BrdU at 5 days and examined 5 hours later. Similar results were obtained with survival times of 1 hour and/or injections at 3 days, although proportionately more of the labeled nuclei were in the INL than in the ONL at 3 days than at 5 days (discussed later). In the region of the lesion, the laser had typically destroyed all the photoreceptors and, we presume, all or most of the rod precursors (photoreceptors and rod precursors are the only nuclei normally resident in the ONL). Within the lesion there were many proliferating cells (Fig. 1G) , in contrast to the surrounding undamaged retina, in which only rare BrdU-labeled nuclei were seen. The proliferating cells (BrdU-labeled nuclei) that were present at 3 to 5 days in the INL, in the ONL, and between these two layers were heterogeneous in level of expression of RET1 immunoreactivity. Some were clearly RET1 positive (consistent with an identity as Müller glia), but some had little or no RET1 expression (insets, Figs. 1F 1G ). 

During the first week after the lesion (3–7 days), the majority of BrdU-labeled nuclei in the INL expressed RET1 (Table 1 , column 7). The RET1-positive, BrdU-labeled nuclei were presumed to be Müller cells (Figs. 1F 1G) . Many of these cells had irregular, spindle-shaped nuclei, consistent with their designation as Müller cells. ^{28 41} Subsequent studies with Müller-specific antibodies supported this inference and demonstrated additional cytological responses to the damage produced by the laser lesion. Müller cells in the goldfish retina expressed a basal level of GFAP, which was upregulated in the lesioned area, as we had observed previously. ¹⁷ Antibodies to GS can also be used to demonstrate Müller cell processes in the retina. ⁴² By 5 days after the lesion, the level of expression of GS in Müller cells in the region of the lesion was downregulated (Fig. 2C) , and this local decrease in expression levels was still apparent at 25 days (Fig. 2D) . These alterations in expression levels of GS are consistent with previously observed behavior of Müller cells that have lost their neuronal contacts. ^{42 43}  

Many other proliferating cells in the INL during the first week after the lesion were RET1 negative. To determine whether any of these were putative neural progenitors (i.e., neither RET1-labeled Müller cells nor NN2-labeled microglia), we examined lesioned retinas labeled with BrdU at 2, 3, or 5 days after the lesion and processed them for double immunocytochemistry after a 5-hour survival with a cocktail of RET1 and NN2 monoclonal antibodies. At each time point, we were able to find substantial numbers of BrdU-labeled nuclei in the INL that were not labeled with the RET1 and NN2 antibodies, suggesting that these cells were neither Müller glia nor microglia. Quantification of these data showed that proliferating neural progenitor cells in the INL were more frequent in retinas at 5 days than at 2 days after lesion: Of the BrdU-labeled nuclei in the INL at 5 days, 44% (2 lesions, 3 sections counted, 3–9 BrdU-labeled nuclei per section) were RET1-NN2–negative compared with only 6% (4 lesions, 8 sections counted, 2 to 17 BrdU-labeled nuclei per section) at 2 days. This observation is consistent with the delay in mitotic activation that would be expected of stem cells, which are rare and cycle slowly in the undamaged tissue. ^{29 30} In contrast, mitotic activation of microglia and Müller glia in damaged adult goldfish retina is known to be rapid. ²⁸  

Movement of Proliferating Cells from the INL to the ONL

We next examined changes in the spatiotemporal distribution of proliferating retinal cells during the first 2 weeks of regeneration. During the first week we found a tendency for BrdU-labeled nuclei in the lesioned area to increase in the ONL and decrease in the INL. Of the BrdU-labeled nuclei, 41% were in the INL at 3 days but only 21% at 9 days (weighted average of 19% and 27%; Table 1 , column 5). Over the same interval, the proportion of BrdU-labeled nuclei in the ONL increased from 59% to 78% (weighted average of 81% and 72%; Table 1 , column 6). The biggest incremental change occurred between 3 and 5 days (Table 1 , columns 5 and 6). These changes in the distribution of BrdU-labeled nuclei are consistent with a shift in the distribution of proliferating cells from INL to ONL (although the apparent changes in percentages between 3 and 9 days were not statistically significant). However, Müller nuclei (identified as those double-labeled with BrdU and RET1) showed a significant (P < 0.01) shift in distribution. At 5 days, 71% of the BrdU-labeled nuclei in the ONL expressed RET1 (weighted average of 75% and 61%), compared with only 16% at 3 days (Table 1 , column 8). In contrast, the fraction of BrdU-labeled nuclei in the INL that expressed RET1 between 3 and 7 days was more constant (range: 51%–72%, P > 0.05; Table 1 , column 7). These data are consistent with migration of Müller nuclei from the INL to the ONL. 

If the Müller cells proliferating in the region of the lesion produced progeny that survived, a local increase would be expected in the density of end feet in the regenerated region. To test this, we counted the glial end feet in the four lesions seen in Figure 1A and compared these data to counts in an adjacent region of comparable area, in which end feet were regularly spaced. Four separate counts were made of each lesion and control area, and the ratios were averaged. The number of end feet within the lesions ranged from 50 to 104. There were fewer end feet within the regenerated region in all four lesions (18%, 26%, 31%, and 37%), suggesting that the lesion-induced proliferation of Müller cells does not result in persistent gliosis—that is, the increase in number of Müller cells in the lesioned area was transient. 

Figure 1H shows the retina from a fish that was injected with BrdU at 5 days and fixed at 7 days. Most of the nuclei that had incorporated BrdU at 5 days were in the ONL (Table 1 , column 6). They were not labeled with NN2, although occasional BrdU-NN2–labeled microglia persisted in the INL in the region of the lesion (Fig. 2B) . Although RET1-BrdU double-labeled nuclei were still found in the INL (inset, Fig. 1H ; Table 1 , column 7), the intensity of the RET1 labeling in most of the proliferating cells within the lesion at 7 days (Fig. 1H) was not as strong as it was at 5 days (Figs. 1F 1G) , consistent with downregulation in levels of RET1 expression in the proliferating Müller glia and/or increased representation of the RET1-negative proliferating progenitor cells after the first week. Only 24% of the total BrdU-labeled nuclei that had been dividing at 5 to 7 days expressed RET1 at 9 days (sum of the weighted averages of 5% and 8% in the INL and 18% and 17% in the ONL), compared with 54% at 7 days (13% in the INL and 41% in the ONL), and 73% at 5 days (sum of the weighted average of 22% and 21% in the INL and 56% and 41% in the ONL; Table 1 , columns 9 and 10). Although these counts show a consistent decrease in the overall proportion of double-labeled nuclei with increased time after lesion, the differences were not statistically significant. It is unlikely that the apparent loss of RET1 expression in proliferating nuclei in the ONL is entirely accounted for by migration of Müller glial nuclei back into the INL, because the proportion of RET1-BrdU double-labeled nuclei in the INL decreased, rather than increased, during this period (Table 1 , column 9). 

When fish were injected with BrdU at 6 and 7 days and examined at 12 days, we found faintly RET1-labeled nuclei in the ONL that appeared to be young, regenerating cones, some of which were labeled with BrdU (data not shown). We confirmed this inference by labeling these immature (regenerated) cone photoreceptors with a cone-specific marker, the zpr1 monoclonal antibody (data not shown). 

In summary, both the qualitative and quantitative data suggest several concurrent events. First, RET1-positive Müller cells in the INL began to proliferate within a few days of the laser lesion. At 3 days, the majority of dividing cells in the INL expressed RET1, and the majority of RET1-BrdU double-labeled nuclei resided in the INL. The high percentage of double-labeled nuclei in the BrdU-labeled population within the INL up to 7 days suggests that Müller cells continued to divide during the first week after lesion. Second, RET1-BrdU double-labeled Müller nuclei quickly migrated into the ONL within the first few days after the lesion. Radially elongated, ectopic nuclei with varying levels of RET1 expression spanned the INL–ONL boundary within the first week, concomitant with a sudden increase in RET1-BrdU double-labeled cells in the ONL (between 3 and 5 days), and a gradual decline in double-labeled cells in the INL (between 3 and 9 days). Third, although the displaced Müller cell nuclei disappeared from the ONL during the second week, they did not migrate back into the INL. Fourth, proliferating, RET1-NN2–negative, retinal progenitors in the ONL and the INL increased sharply in number from 5 to 10 days. Taken together, these data are consistent with injury-induced proliferation of RET1-positive Müller cells and (slightly delayed) of RET1-negative neural progenitor cells and migration of nuclei of both cell types from the INL to the ONL during the period of cone regeneration within the first week after the laser lesion. 

Expression of Cell Surface Markers Characteristic of Retinal Stem Cells

The foregoing results showed that cone photoreceptors regenerated from proliferating retinal progenitors whose nuclei migrated from the INL into the “gap” in the ONL that was produced by the photocoagulation lesion. We next asked whether these retinal progenitors express markers characteristic of the multipotent retinal progenitors known to generate cones—that is, the progenitors in the embryonic retina and in the circumferential germinal zone of the adult retina. Two developmentally regulated genes that code for cell surface receptors involved in cell adhesion and cell–cell signaling—N-cadherin (cdh2) and Notch-3—are strongly expressed by the embryonic and germinal zone retinal progenitors, as demonstrated by in situ hybridization with cRNA probes: zebrafish cdh2 (Fig. 2E) and goldfish G-Notch-3 ^{36 37} (Fig. 2G) . The goldfish cdh2 gene has not been cloned, but the zebrafish cdh2 probe showed specific hybridization in goldfish retina consistent with the pattern in zebrafish retina (Liu Q, Barthel LK, Raymond PA, et al., unpublished observations, 1999). In addition to the strong level of expression of both these genes in the germinal zone, we found that scattered cells in the INL, usually in or near the amacrine cell stratum on the inner side of the INL, also expressed both cdh2 (Fig. 2E) and Notch-3 ³⁶ (Fig. 2G) . In the ONL outside the region of damage, there was no expression of cdh2 or Notch-3 in differentiated photoreceptors or in rod precursors ³⁷ (Figs. 2E 2F 2G 2H) . However, within the lesion, the expression of both genes was very strong in progenitor cells that had migrated into the ONL (Figs. 2F 2H) . 

Discussion

The present results confirm and extend previous work from this laboratory demonstrating cone regeneration in the adult goldfish retina. ¹⁷ Mitotic divisions that occurred in neural progenitor cells within the first 2 weeks after laser ablations produced progeny that differentiated preferentially into cones, whereas regenerated rod photoreceptors appeared later, and they were produced by specialized rod precursors. 

During the first week after laser lesion, many different types of proliferative cells infiltrated the ONL in response to the ablation of photoreceptors. Some of these were vascular-derived cells (microglia) recruited to dispose of the cellular debris. These were identified with a specific cell surface marker NN2. Others were Müller cells, identified by cell-specific antibody markers, including RET1, which recognizes an unknown nuclear epitope not specific to Müller glial markers and the glial-specific markers, GFAP and GS. It was surprising, however, that proliferation of Müller cells did not generate extra glial cells in the region of the lesion, nor was there evidence of Müller cell proliferation or turnover in the uninjured adult goldfish retina. For example, a recent study ⁴¹ found that the level of GS immunoreactivity in the retina remained constant in adult cichlid fish as the retina grew in size, whereas the density of Müller cells themselves declined. These data suggest that Müller cells in teleost fish respond to growth-related retinal expansion by increasing their size, rather than by adding new cells interstitially. In the present study, we found that the level of GS was reduced locally in the region of the lesion for up to 25 days, and the density of Müller cells was lower in the regenerated retina than in the surrounding, intact retina, despite evidence of substantial proliferation of Müller cells in the region of the lesion. 

The fate of new cells produced by the proliferating Müller cells is unclear. Because they did not produce extra Müller cells, they must have either died or transformed into another cell type. We cannot exclude the possibility that the progeny of the dividing Müller cells dedifferentiated or transdifferentiated into cone progenitors, but we have no direct evidence for this. The gradual loss of RET1 immunoreactivity that we observed in the proliferating progenitors in the ONL, where the Müller nuclei migrated, could reflect their dedifferentiation. The Müller glial cell phenotype is unstable, at least in cell culture, where Müller cells can transdifferentiate into lentoid cells that express crystalline proteins. ⁴⁴ However, in the mammalian retina, Müller cells also proliferate in response to various pathologic conditions, including photocoagulation lesions, ^{45 46 47} but they are not able to restore lost neurons and photoreceptors. Instead, the Müller cell response is associated with a pathologic, reactive gliosis. ^{38 48 49}  

Although there was no evidence of transformation of Müller cells into neuronal progenitors in the injured mammalian retina, the possibility that this might occur in the fish retina is worthy of serious consideration, especially given surprising new findings about neural stem cells in the adult mammalian brain. Two recent reports have suggested that neural stem cells in the adult mammalian brain are actually a subclass of glial cell—either specialized astrocytes in the subventricular zone ⁵⁰ or ependymal cells at the ventricular surface. ⁵¹ Both of these glial cell types express the intermediate filament protein, GFAP. ^{30 52} Müller cells are a specialized type of radial glia, ⁵³ and radial glial cells in the developing cerebral cortex in mammals can also behave as neural progenitors, even as they continue to express glial-specific markers such as GFAP. ⁵⁴ In addition, lineage-tracing studies in developing brain have shown clones of neurons and glia associated with a single radial glial cell, ^{55 56} and in the retina, similar studies have shown that Müller cells and retinal neurons, especially rod photoreceptors, are produced from a common retinal progenitor. ^{57 58}  

The results reported herein show that presumptive retinal progenitors in the INL responded to the loss of photoreceptors with behavior similar to Müller cells: increased mitotic activity and migration of their nuclei to the ONL. Julian et al. ⁸ have demonstrated conclusively that slowly cycling retinal progenitor cells persist in the INL of juvenile rainbow trout (Onchoryncus mykiss). They used a mitotic labeling paradigm optimized to detect slowly cycling cells, in which the thymidine analogues IdU (iododeoxyuridine) and BrdU were administered continuously for up to 10 days. By exposing fish sequentially to IdU and then BrdU, they showed that the labeled nuclei moved from the INL to the ONL and that the purpose of this migration was to replenish the rapidly cycling rod precursor pool in the ONL, which eventually differentiated into rods. The radially elongated shape of these cells, and their frequent position straddling the plexiform layer between the INL and the ONL, are also consistent with outward radial migration. We had previously described the analogous behavior of retinal progenitor cells in the INL of the larval goldfish retina, and we also showed that these cells that originated in the INL were the source of the rod precursors, which did not appear in the ONL until early larval stages. ²⁰  

Both our earlier study ²⁰ and the more recent work ⁸ have demonstrated that the only progeny of the retinal progenitors in the INL of the uninjured retina are the rod precursors, and the only progeny of rod precursors are rod photoreceptors. The results of the present study demonstrate that in the lesioned adult goldfish retina, the rate of proliferation was locally upregulated in the INL progenitors at the time when cones were being generated and that their nuclei migrated into the ONL, in association with the Müller cell nuclei. These results are consistent with the proposal that the INL progenitors are not, in fact, restricted to the rod lineage, but can also generate cone photoreceptors. Previous studies have reported regeneration of the complete array of retinal neurons in adult goldfish from clusters of proliferating cells with behavior and morphology similar to that described here. ^{31 32 59} We therefore conclude that the neural progenitor cells in the INL are likely to be multipotent retinal stem cells, with a capacity similar to that of the primitive, multipotent retinal progenitor cells in the germinal zone at the retinal margin. 

The regeneration of cones was associated with migration of the nuclei of retinal progenitors to establish contact with the outer limiting membrane (OLM), which represents the apical (ventricular) surface of the retinal epithelium. We have shown previously that in other experimental paradigms, regeneration of neurons in the adult fish retina takes place only when proliferating retinal progenitors reach the OLM. ^{17 22 34 60} This cellular arrangement mimics the cytological organization of the primitive retinal neuroepithelium and the circumferential germinal zone at the retinal margin, and it may reflect the presence of obligatory environmental factors that perhaps regulate the multipotent capacity of the retinal progenitors—that is, their ability to produce multiple types of retinal neurons. An association with the apical surface of the neuroepithelium may be a fundamental feature of stem cells. Neural stem cells in the embryonic and adult mammalian brain are also associated with the ventricular surface. ^{61 62}  

Although there are no known markers specific for retinal stem cells, the proliferating retinal progenitors within the lesioned retina expressed high levels of two developmentally regulated genes that have been implicated in retinal neurogenesis: N-cadherin (cdh2), a calcium-dependent, homophilic adhesion molecule that is expressed by multipotent neural progenitor cells, including those in the retina, ^{63 64} and G-Notch-3, a member of the Notch family of cell surface receptors that regulate contact-mediated, lateral inhibitory interactions and cell fate in developing tissues, ^{65 66} including the retina. ^{37 67} Rod precursors express neither of these genes, and their upregulation in association with regeneration of cone photoreceptors suggests that the expression of Notch and N-cadherin may be associated with multipotent retinal stem cells. Neural stem cells derived from the ependymal layer of the brain also express Notch, and antibodies to this cell surface marker have been used to enrich neural stem cells from the adult mammalian brain. ^{30 51} However, under some circumstances, activation of the Notch signaling pathway promotes gliogenesis, ⁶⁸ including differentiation of Müller glia. ⁶⁹  

If the INL progenitors are multipotent retinal stem cells, they would be expected to express other developmentally regulated genes that are characteristic of the primitive retinal progenitors in the embryonic retina and in the germinal zone at the retinal margin. ^{70 71} This includes transcriptional regulators in the homeodomain family (e.g., Rx, Pax6, Chx10/vsx1, Six3) and others in the basic helix-loop-helix family (atonal and achaete scute homologues), all of which have been implicated in retinal development. ^{67 72} The multipotent retinal progenitors in the regeneration blastema at the edge of a surgical lesion in adult goldfish retina express both vs.x1 and pax6, ^{73 74} and recently, Otteson et al. ^{18 19} showed that the slowly proliferating, INL progenitors in the undamaged adult goldfish retina could be immunostained with antibodies directed against recombinant zebrafish pax6.1 protein. Efforts are under way to determine what other genes are expressed by the retinal stem cells in the adult fish retina. 

³  Present address: 20696 Bond Road NE, Poulsbo, WA.

⁴  Present affiliation: Burgett Eye Care, Lafayette, Indiana.

⁵  Present affiliation: Department of Ophthalmology, Medical College of Georgia, Augusta.

 Pamela A. Raymond has published previously as Pamela R. Johns.

 Supported by National Institutes of Health Grant EY04318 (PAR).

 Submitted for publication January 5, 2001; revised March 9, 2001; accepted March 29, 2001.

 Commercial relationships policy: N.

 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.

 

(A) Schematic of criteria used to define the lesion area for counting immunolabeled nuclei: RET1 (green), BrdU (red), and double-labeled with RET1 and BrdU (orange). In the flanking, unlesioned retina, RET1-labeled cone nuclei formed a continuous row in the ONL. Gray bracket: lesioned region, in which cones were largely missing. Occasionally, a surviving cone nucleus was seen within the lesion (a), but other RET1-labeled nuclei in the ONL (b) were probably misplaced Müller cells. Labeled nuclei within the trapezoidal region outlined by blue lines were counted. The sides of the imagined trapezoid (dashed lines) are parallel to the long axis of cones adjacent to the lesion boundary (arrows). Top of the trapezoid is defined by the RET1-labeled cells at the OLM; bottom is the inner edge of the INL. Shown are six double-labeled (orange) nuclei in the ONL and one in the INL. Horizontal cell layer (hcl) is outer boundary of the INL. Two nuclei in the INL were labeled only by BrdU (red). BrdU-labeled nuclei outside the trapezoid (c) and (d) in the subretinal space were not counted. (B) Wholemount preparation of retina from a fish injected with BrdU at 5 and 7 days after lesion and killed 115 days after lesion. Four clusters of BrdU-labeled nuclei (CY3, red; arrows) were regenerated cells within the lesions. (C) Higher magnification of one of the lesions in (B). Müller glial end feet were labeled with anti-GFAP (FITC, green), and the (out-of-focus) regenerated nuclei were labeled with BrdU (red-orange). Müller fibers showed even spacing in the unlesioned retina and a disrupted pattern within the lesion. (D–H) Immunolabeled retinal cryosections. (D) Section labeled with RET1 (AMCA, blue), GFAP (FITC, green), and BrdU (CY3, red-orange). Thin white lines: lesion’s boundaries in this and subsequent panels. Fish received BrdU injections at 5 and 8 days after lesion and the retina was fixed 20 days after lesion. Small arrows: nuclei of regenerated cones (double-labeled with BrdU and RET1). Two other nuclei in the ONL (on), which incorporated BrdU but did not express RET1, were rods or rod precursors (arrowhead). Left inset: higher magnification showing the double-labeled cone nuclei indicated by the central two arrows. The nuclei are pink, representing the combination of the AMCA (blue) RET1 signal and the CY3 (red-orange) BrdU. Also, note the single, round nucleus of a rod or rod precursor that was BrdU labeled but RET1 negative and is therefore red-orange. Right inset: RET1 (AMCA, blue) signal alone. The rod nucleus was not labeled with RET1. In the INL (in), the BrdU-labeled nucleus of a regenerated Müller cell was associated with a GFAP-labeled (FITC, green), radial glial fiber (wide arrow). However, some of the BrdU-labeled nuclei in the INL did not express RET1 (★). (E) Section immunolabeled as in (D). Fish received BrdU injections 10, 13, and 16 days after lesion and the retina was fixed 21 days after lesion. Small arrows: BrdU-labeled rod precursors or rod photoreceptor nuclei in the ONL. Arrowheads: double-labeled (BrdU-RET1) regenerated cone nuclei. Inset: higher magnification of region indicated by arrowhead at left. Most of the BrdU-labeled nuclei were rods or rod precursors (not double-labeled). Other BrdU-labeled cells were in the INL (★). (F, G) Section immunolabeled with RET1 (FITC, green) and BrdU (CY3, red-orange). (F) Single-channel, FITC image; (G) double-exposure image. Fish received a single BrdU injection 5 days after laser lesion and the retina was fixed 5 hours after injection. Arrows: BrdU-labeled nuclei in the INL. (F, G, insets) BrdU-labeled nuclei (from the region indicated by the fourth arrow) at higher magnification; the nucleus on the far left of the inset was strongly RET1-positive, but the three on the right had weak or no RET1-immunoreactivity (inset; F). (H) Section immunolabeled as in (G). Fish received a single BrdU injection 5 days after laser lesion and the retina was fixed 7 days after lesion. Arrow: pair of spindle-shaped, double-labeled nuclei. Right inset, the lower of the two at higher magnification; left inset: same nucleus showed strong RET1 immunoreactivity. The other BrdU-labeled nuclei in the INL expressed little or no RET1. Insets: one of these nuclei at higher magnification. (★) BrdU-labeled microglia in the subretinal space. Scale bars, (B, D–H) 50 μm; (C) 100 μm.

Figure 2.

 

Immunolabeled retinal cryosections. (A) Section immunolabeled with microglial surface marker NN2 (brown) and BrdU (bright-field exposure; CY3 fluorescence is bright, but pale pink). Fish received a single BrdU injection 5 days after laser lesion and the retina was fixed 5 hours after injection. Arrowheads: spindle-shaped nuclei that have incorporated BrdU but not NN2. Examples of proliferating microglial cells were present within the neural retina (black arrow) and in the subretinal space (white arrow). (B) Section immunolabeled as in (A). Fish received a single BrdU injection 5 days after laser lesion and the retina was fixed 9 days after laser lesion. Most BrdU-labeled nuclei (arrowheads) were in the ONL (on) and were not microglia. Black arrow: Single, double-labeled microglial cell. (C) Section labeled with anti-GS (red) and BrdU (green). Fish received a single BrdU injection 5 days after laser lesion and the retina was fixed 5 hours after injection. The region of the lesion (between the white arrows) shows a marked deficiency in GS. (D) Section immunolabeled as in (C). Fish received BrdU injections 5 and 7 days after laser lesion and the retina was fixed 25 days after lesion. The deficiency in GS within the area of the lesion persisted. (E, F) Sections hybridized with a zebrafish N-cadherin cRNA probe. (E) At the peripheral margin of the neural retina, the circumferential germinal zone expressed N-cadherin (arrow). Scattered cells in the INL also hybridized with the probe (arrowhead), as did some of the ganglion cells (gc), but none of the photoreceptors in the ONL (on). Thin black line: OLM, in this and subsequent panels. (F) In the region of the lesion (black arrows), cells expressing high levels of N-cadherin clustered along the OLM (arrowhead). (G, H) Sections hybridized with goldfish Notch3 cRNA probe and immunolabeled with anti-GFAP (pale green). (G) The retinal progenitors in the circumferential germinal zone expressed high levels of Notch3, as did some cells in the INL (arrowhead). (H) In the region of the lesion (black arrows), cells expressing high levels of Notch3 clustered along the OLM (arrowhead). Scale bars, 50 μm.

Table 1.

 

Quantification of BrdU- and RET1-Labeled Nuclei

1	2	3	4	5	6	7	8	9	10
BrdU Injections*	Survival Time*	Number of Sections	Total BrdU Nuclei, † per Section	INL BrdU, ‡/Total BrdU*	ONL BrdU, ‡/Total BrdU*	INL Double, §/INL BrdU, ‡	ONL Double, §/ONL BrdU, ‡	INL Double, §/Total BrdU*	ONL Double, §/ Total BrdU*
3	3	8	2–29	0.41	0.59	0.62	0.16	0.30	0.12
5	5	12	8–27	0.26	0.74	0.72	0.75	0.22	0.56
5	5	5	14–31	0.33	0.67	0.51	0.61	0.21	0.41
5	7	8	28–64	0.23	0.77	0.56	0.55	0.13	0.41
5	9	8	19–61	0.19	0.81	0.33	0.22	0.05	0.18
6, 7	9	4	25–52	0.27	0.73	0.29	0.27	0.08	0.17

*  Days after laser treatment.

†  BrdU-labeled nuclei in the ONL and INL within the lesion (Fig. 1A) .

‡  BrdU-labeled nuclei in the INL (or ONL) within the lesion.

§  Double-labeled nuclei (BrdU- and RET1-positive).

The authors thank the Kellogg Eye Center of the Department of Ophthalmology and Visual Sciences at the University of Michigan Medical School for the use of the argon laser and Kathleen Welch, University of Michigan Center for Statistical Consultation and Research, for help with statistical analysis.