Changing Patterns of Vasculature in the Developing Amphibian Retina

Adult vertebrates display a wide range in the density distributions of neural cells within the retina, patterns that reflect the visual constraints placed on an animal by its lifestyle and environment (reviewed by Hughes, 1977; Collin, 1997). Examples of such specialised distributions include the high-cell-density fovea, the area centralis and the visual streak. The vascular demands of the outer retina are met by the choroidal circulation (Alder and Cringle, 1990; Cringle et al. 1990; Pournaras et al. 1989; Duke-Elder, 1958). Species in which the inner retina is vascularised display a wide range of blood vessel patterns which appear to be dictated by the density distributions, or topography, of neural cells in the ganglion cell layer (Michaelson, 1954; Johnson, 1968; Francois and Neetens, 1974). Capillary beds are concentrated in regions of high cell density, presumably reflecting the high metabolic demand of the latter within the neural retina. For example, in the Florida garfish Lepisosteus platyrhincus, the ventrally located regions of high cell density, the area centralis and visual streak, are fed by a rich capillary bed (Collin and Northcutt, 1993). A recent study examining the retinal vascular tree in the adult frog Litoria moorei (Tennant et al. 1993) confirmed earlier reports in ranid frogs (Virchow, 1881) showing a concentration of capillaries along the naso-temporal axis, which thus mirrors the topography of the area centralis and visual streak (Humphrey and Beazley, 1985). Similarly, in the adult cat Felis domesticus, squirrel monkey Saimiri sciureus and man, the area centralis and fovea are both supplied by a concentration of capillaries (Chan-Ling et al. 1990; Snodderly and Weinhaus, 1990; Duke-Elder, 1958). The major blood vessels differ from capillaries in that they avoid regions of high cell density. In this way, visual acuity is not compromised and, as a result of the wide variety in the topography of cells in the retinal ganglion cell layer, species display a matching variety in the patterns of their vascular trees (Copeland, 1976; Collin, 1989; Collin and Northcutt, 1993; Michaelson, 1954; Johnson, 1968; Francois and Neetens, 1974).

Development of retinal vasculature has been described only for eutherian mammals (Ashton, 1970; Shakib et al. 1968; Henkind and De Oliveira, 1967; Flower et al. 1985) and is best understood in the cat (Chan-Ling et al. 1990). The retinal vasculature begins to form 3 weeks before birth (gestation is 65 days) and is complete by approximately 1 month after birth, coincident with eye opening (Blakemore and Cummings, 1975), clearing of the optics (Bonds and Freeman, 1978; Thorn et al. 1976) and functional maturation within the primary visual system (Rusoff and Dubin, 1977; Friedlander et al. 1985). Despite the completion of retinal vascularisation in the cat at a time coincident with the onset of functional vision after birth, the area centralis is established almost 2 weeks before birth (Stone et al. 1982; Robinson, 1987), i.e. at least 5–6 weeks prior to its vascularisation (Chan-Ling et al. 1990).

In contrast to mammals (Rusoff and Dubin, 1977; Friedlander et al. 1985), the onset of visual function in amphibians occurs as early as stage 43, i.e. only 4 days after fertilization when animals emerge as free-swimming larvae (Gaze et al. 1974) that use visually guided behaviour for feeding and avoidance of predators. Another difference between mammals and amphibians is that, whereas cell density gradients are established early in the cat (Stone et al. 1982; Robinson, 1987), in amphibians these are present only from metamorphic climax as animals change from an aquatic to a terrestrial habitat (Bousfield and Pessoa, 1980; Dunlop and Beazley, 1981, 1984a; Coleman et al. 1984; Nguyen and Straznicky, 1989). Furthermore, retinal neurogenesis is complete early in mammalian development (Harman and Beazley, 1989; Harman et al. 1992; Rapaport and Stone, 1983; Young, 1985), but in amphibians it continues throughout life as new retinal cells are added at the ciliary margin (Straznicky and Gaze, 1971; Coleman et al. 1984).

Here, we have examined the structure of the vascular tree from tadpole life through to adulthood in the frog L. moorei. Compared with mammals, the early onset of function and the continual development of the retina in amphibians present ever-changing demands to its vascular supply. Our findings on the changing vascular tree are related to the formation of the area centralis and visual streak within the ganglion cell layer, a process reported here to match that for other amphibians (Bousfield and Pessoa, 1980; Dunlop and Beazley, 1981, 1984a; Coleman et al. 1984; Nguyen and Straznicky, 1989). Specifically, we wished to determine whether, as in adult L. moorei, the pattern of vasculature during development reflected the topography of cells in the ganglion cell layer. Furthermore, given the earlier onset of visual function in tadpoles compared with mammals, we investigated whether specialisations within the retinal vasculature also developed at early stages, i.e. prior to the formation of the area centralis and visual streak. Part of this work has appeared in abstract form (Dunlop et al. 1996).

Litoria moorei (Copland) at Nieuwkoop and Faber (1956) stages 47, 51, 54 and 57 were selected from a breeding colony; others were raised to metamorphic climax or to 1-to 2-month-old juveniles. There were 2-to 3-week intervals between stages 47, 51, 54 and 57, whereas only 1–2 weeks lapsed between stage 57 and metamorphic climax. We collected frogs with snout–vent lengths of 5 cm, 7 cm and 9 cm locally under license from the Department of Conservation and Land Management. The smaller frogs were approximately 1 and 2 years post-metamorphosis, respectively, and the 9 cm animals represented the largest size reported for this species, being probably at least 3–4 years old (Barker and Grigg, 1977). Animals were maintained on a 12 h:12 h L:D cycle at 22±2 °C. Pre-metamorphic stages were fed Biorell fish food and boiled lettuce; post-metamorphic animals received mealworms three times a week and were provided with water ad libitum. Tadpoles were killed by immersion in MS222 (0.2 % in distilled water), adults by intramuscular injection of Saffan (1.8 mg g⁻¹ Alphaxalone; 0.6 mg g⁻¹ Alphadolone acetate, Pittman Moore).

Pilot studies showed that perfusion with Indian ink gave reproducible filling of the entire vascular tree. However, subsequent Cresyl staining revealed that the ink-filled vascular tree directly overlaid, and thus masked, significant numbers of cells in the ganglion cell layer (Fig. 1A,B). The retinal vasculature and density of cells in the ganglion cell layer were therefore examined using separate series of retinal whole-mounts.

Animals were perfused intracardially with heparinised saline followed by Indian ink using a peristaltic pump. A glass micropipette was used for the young stages and a 26 gauge needle for post-metamorphic animals. Retinae (N=4 for each of the tadpole stages 47–57, N=6 at metamorphic climax, N=8 in the 1–2 month juveniles and the 5 and 7 cm frogs, and N=2 in the 9 cm adults) were dissected in saline, removing the sclera, pigment epithelium, cornea and lens. The vitreous humour was dissolved by application of hyaluronidase (Sigma), and the retina was flattened by making radial cuts, taking care not to sever major blood vessels. Retinal whole-mounts were dried onto subbed slides, dehydrated, cleared in xylene and coverslipped.

To measure retinal area, whole-mounts were viewed using a light microscope and images were frame-grabbed onto a computer at low magnification (×2) using an Optimas imaging system. Measurements were taken directly from the computer images using Optimas image-analysis software (version 4.2). For analysis of the vascular tree at pre-metamorphic stages, retinae were viewed using a light microscope, images were frame-grabbed onto a computer at high power (×20) and printed to a final magnification of ×227 to generate photomontages of the entire retina; at post-metamorphic stages, camera lucida drawings of the complete vascular tree were made. For each retina, all the branch points within the vascular tree were counted directly from the photomontages and camera lucida drawings. The average density of branch points per mm2 across the entire retina was also calculated by dividing the total number of branch points by the retinal area. Changes with age in retinal area, the number of branch points and the density of branch points per mm2 were evaluated using a one-way analysis of variance (ANOVA) followed by the Fisher protected least significance difference test using Statview SE Plus Graphics software.

To examine the topography of the blood vessels, we generated isodensity maps of branch points per mm². The wide range in retinal area with age necessitated varying the sample size to allow sufficient sample sites per retina; these were square grids of 125 μm×125 μm at stages 47, 51 and 54, 250 μm×250 μm at stage 57 and 500 μm×500 μm from metamorphic climax onwards. For direct comparison between retinae, the numbers of branch points at these sample sites were expressed per mm2. Furthermore, we mapped the entire population of lacunae (Burri and Tarek, 1990; Patan et al. 1996) and blind-ending branches (Tennant et al. 1993) within the capillary network. The maps were generated by scanning the photomontages of the entire retina in a serpentine fashion in adjacent but non-overlapping fields. Blind-ending branches were measured from their blood vessel of origin by aligning the vessel with a line on an eye-piece graticule and taking the measurement from that line to the tip of the branch. Using this technique, we included in the measurement the broad junction between the blind-ending branch and its parent vessel. Blind-ending vessels were included only if they were a minimum of 25 μm long. By using these criteria, we excluded small protrusions along blood vessel walls of whose identity we were uncertain.

A series of whole-mounts (N=2 per stage) was prepared to map cell density in the ganglion cell layer, i.e. the population of ganglion cells plus displaced amacrine cells (Humphrey and Beazley, 1985). Eyes were removed, placed in saline, dissected, treated with hyaluronidase to remove the vitreous humour as above and defatted free-floating in a series of graded alcohols, followed by treatment in xylene for 15 min. After rehydration, retinae were whole-mounted, stained with Cresyl Violet, dehydrated, cleared in xylene, coverslipped with Depex and examined at ×1000 magnification. Retinae were drawn at ×20 magnification to generate an outline on squared paper, each square having a side of 2 mm. Cells in the retinal ganglion cell layer of L. moorei are arranged as a monolayer and thus could be mapped accurately using whole-mount preparations. The total cell density in the retinal ganglion cell layer was counted at ×1000 using an eye-piece graticule with a side of 100 μm. The outline of the retina was used to locate regions within the sample squares and the retina thus mapped in columns and rows to cover between 1 and 4 % of retinal area, the sampling frequency being highest where cell density gradients were steepest. Comparison of retinal area before and after staining revealed that shrinkage was less than 10 % and was confined to the cut edges. Post-staining retinal areas were used for our calculations.

At stage 47, retinal area was only 2.2±0.4 mm2 (mean ± S.D.; Fig. 2A). The ventrally located hyaloid vein was the sole major blood vessel and extended between the ventral periphery and the optic disk (Fig. 3); this vessel remained the most prominent vessel during development and adult life. There was a condensation of capillaries in the nasal retina which we assume represents the initial stages of major blood vessel formation in this quadrant. The remainder of the retina was covered by a honeycomb-like network of capillaries (Figs 3, 4A), spreading between the hyaloid vein and the peripheral collecting vessels. Small numbers of lacunae and of blind-ending vessels were seen (Figs 3, 4E–J). The topography of branch points at stage 47 (Fig. 5) was concentric, with the highest densities occurring centrally (1500 mm⁻²) and the lowest peripherally (250 mm⁻²). Thus, the map of the vascular tree was the reverse of that in the ganglion cell layer (Fig. 5), in which there was a shallow concentric distribution of cells with the lowest densities centrally (10×103 cells mm⁻²) and the highest peripherally (15×103 cells mm⁻²).

By stage 51, the retina had enlarged to 4.6±0.9 mm2 (Fig. 2A). In addition to the prominent hyaloid vein, major blood vessels had formed both nasally and temporally (Fig. 3). Other major vessels also appeared to be in the process of condensing within the capillary bed at approximately equally spaced intervals around the radius of the retina. The honeycomb appearance of the capillary bed was maintained, and both lacunae and blind-ending branches were observed in small numbers (Fig. 3). The topography of branch points showed two zones of high density (1000 mm⁻²), these being in the mid-temporal and mid-nasal retina (Fig. 5). Elsewhere, branch point densities had decreased somewhat compared with stage 47, at 500 mm⁻² centrally and 250 mm⁻² peripherally. Within the ganglion cell layer, the concentric pattern of cell topography seen at stage 47 was maintained and densities remained similar, at 10×10³ cells mm⁻² centrally and 16×10³ cells mm⁻² peripherally (Fig. 5). Therefore, regional specialisations were apparent within the vascular tree, whereas an area centralis and visual streak had not yet formed in the ganglion cell layer.

By stages 54 and 57, retinal area had increased to 7.4±3.1 and 13.0±0.3 mm² respectively (Fig. 2A). At least six major vessels were condensing between the central and peripheral retina (Figs 3, 6). Arteries could be distinguished from veins for the first time by a surrounding corridor approximately 50–100 μm wide which was becoming free from capillaries (Figs 3, 6), a feature maintained into maturity. The entire capillary network, however, still retained a honeycomb appearance (Figs 3, 4B, 6) and was characterised by both lacunae and blind-ending vessels (Fig. 3). The topography of blood vessel branch points showed that the two high-density patches seen at stage 51 were maintained with values of 1000–1500 mm⁻² (Fig. 5). These patches were surrounded by a horizontally aligned band of intermediate density which ranged between 500 and 1000 mm⁻²; densities were lowest peripherally, at 250 mm⁻² ventrally and 100 mm⁻² dorsally. Within the ganglion cell layer (Fig. 5), cell topography had changed to give the lowest densities in the dorsal retina (6×103 cells mm⁻²) and intermediate values centrally and ventrally (10×10³ and 12×10³ cells mm⁻² respectively); most notably, there was an increase in the temporal and nasal peripheries (18×103 cells mm⁻²), presumably due to greater cell division at these retinal poles (Coleman et al. 1984). Concentrations of blood vessel branch points in the mid-temporal and mid-nasal retina were not matched by high cell densities in the ganglion cell layer.

By metamorphic climax, the retina measured 19.7±1.6 mm² (Fig. 2A). For the first time, arteries appeared narrower than veins and their capillary-free zone was more pronounced than at stage 57 (Fig. 3). Major vessels lay orthogonally across the horizontal axis. The honeycomb-like arrangement of capillaries had given way to a simpler, more linear arrangement of parallel vessels without numerous interconnecting capillaries, appearing as if most of the rungs of a ‘ladder’ had been removed (compare Fig. 4B,C). Numbers of both lacunae and blind-ending vessels had decreased to low levels (Fig. 3). The topography of blood vessel branch points showed high-density patches in the mid-temporal and mid-nasal retina but values had fallen more than 10-fold compared with stage 57, being only 100 mm⁻² (Fig. 5). These regions were flanked by a horizontally aligned band of intermediate density with values of 50–75 mm⁻²; densities were lowest in the dorsal and ventral peripheries, having fallen to 10 mm⁻². Within the ganglion cell layer (Fig. 5), an area centralis (23×10³ cells mm⁻²) was visible for the first time in the mid-temporal retina; there was also a high-density patch nasally (20×10³ cells mm⁻²) and a horizontally aligned visual streak with 16×10³–18×10³ cells mm⁻². Thus, for the first time in development, comparable specialisations were seen across the naso-temporal axis for both the vasculature and the underlying cells in the ganglion cell layer.

The major characteristics seen in the vascular tree at metamorphic climax were maintained in the 1–2 month juveniles and in the 5, 7 and 9 cm stages as retinal area continued to increase to reach 81.1±4.7 mm2 in the 9 cm adults (Fig. 2A). The capillary bed was composed largely of parallel vessels (Fig. 4D); lacunae were not observed and the numbers of blind-ending vessels were low (Fig. 3). The streak-like topography of the vasculature appeared even more pronounced at progressively later stages (Figs 3, 7). At the area centralis, densities were 100 mm⁻² in the 1–2 month juveniles, 50 mm⁻² in the 5 cm frogs, 40 mm⁻² in the 7 cm frogs and 75 mm⁻² in the 9 cm frogs. Mid-nasally, values declined with age from 100 mm⁻² in the 1–2 month juveniles to 75 mm⁻² in the 5 cm frogs, 40 mm⁻² in the 7 cm frogs and to 30 mm⁻² in the 9 cm stage. Values across the visual streak also declined, being 50 mm⁻² in the 1–2 month juveniles, 40 mm⁻² in the 5 cm stage and 20 mm⁻² in the 7 and 9 cm stages. Reductions in branch point density were also seen in the peripheral retina so that, for example, a larger region was represented by the <10 mm⁻² isodensity line in the 9 cm than in the earlier stages.

In parallel with the streak-like arrangement of the retinal vasculature becoming more accentuated with age, so the area centralis and visual streak became more distinct in the ganglion cell layer during post-metamorphic life (Fig. 7). At the area centralis, cell densities remained high, at 23×103 cells mm⁻² in the 1–2 month juveniles and the 5 cm stage, but fell slightly to 20×10³ cells mm⁻² in the 7 and 9 cm stages. A slight decline in density was seen in the mid-nasal retina, maximal values being 20×10³ cells mm⁻² in the 1–2 month juveniles, falling to 18×103 cells mm⁻² in the 5 cm stage and 13×10³ cells mm⁻² in the 7 and 9 cm stages. Within the visual streak, values were 14×103 cells mm⁻² in the 1–2 month juveniles, 13×10³ to 15×10³ cells mm⁻² in the 5 cm stage and 6×10³ to 10×10³ cells mm⁻² thereafter. Cell densities decreased to a greater extent with age in the dorsal and ventral peripheries from 6×103 cells mm⁻² at metamorphic climax to fewer than 6×10 cells mm⁻² in the 9 cm adult. Thus, at all post-metamorphic stages, the topography of the vasculature was mirrored by that of cells in the ganglion cell layer.

The above results suggest that development of the retinal vasculature is biphasic, with an early exuberant pre-metamorphic stage followed by an extensive refinement before a post-metamorphic phase of growth that is closely linked to that of the underlying retina. This interpretation was supported by comparing between stages both the total number of blood vessel branch points and the average density of branch points across the entire retina as well as changes in the relationship between the topographies of the vasculature and the cells in the ganglion cell layer.

The total number of branch points (Fig. 2B) within the vasculature showed a biphasic pattern, increasing as a result of the addition of new vessels to reach a peak at stage 57 and then decreasing as vessels were lost to a minimum at 1–2 months post-metamorphosis. Thereafter, there was a steady increase, but values never approached the pre-metamorphic peak. The total number of branch points increased steadily from 472±62 at stage 47 to 889±133 and 1539±181 at stages 51 and 54, respectively, peaking at 2835±384 at stage 57 and then declining precipitously to 1087±56 at metamorphic climax and to 606±136 in the 1–2 month juveniles (mean ± S.D.). The total number of branch points then increased gradually to 1125±255, 1211±73 and 1384± 37 in the 5, 7 and 9 cm stages, respectively.

The average density of branch points per mm2 also showed a biphasic pattern (Fig. 2C); values were consistently high until stage 57, fell dramatically by metamorphic climax and declined only slightly thereafter. Values were similar at 220±18 mm⁻², 196±28 mm⁻², 228±67 mm⁻² and 217±25 mm⁻² at stages 47, 51, 54 and 57 respectively. The maintenance of these high values despite a sixfold increase in retinal area indicated a continuous addition of new vessels to the vascular tree matching the increase in retinal area. Between stage 57 and metamorphic climax, the mean density of branch points fell fourfold to 56±5 mm⁻². This decline far exceeded the 1.4-fold increase in retinal area between these stages, indicating a period during which many blood vessels were removed. Had the vascular tree merely become stretched as retinal area increased, the density of branch points would have decreased by only 1.4-fold.

Mean branch point density continued to decrease to 36±9 mm⁻², 28±9 mm⁻², 19±4 mm⁻² and 17±0 mm⁻² in the 1–2 month juveniles and the 5, 7 and 9 cm stages, respectively. During the early part of this second phase, between the 1–2 month juveniles and the 5 cm frogs, the decline in branch point density was only slight (1.4-fold) despite a 2.5-fold increase in retinal area, indicating that, although the vascular tree was presumably being stretched by retinal growth, new blood vessels were nevertheless still being formed. Between both the 5 and 7 cm stages and the 7 and 9 cm stages, the decline in mean branch point density (1.4-fold and 1.1-fold, respectively) approximately balanced the increase in retinal area (1.5-fold and 1.2-fold, respectively), indicating that branch points were added at a rate that matched the continuing expansion of the retina.

The biphasic nature of vascular development was also shown by the changes in the number of blood vessel branch points relative to the number of cells in the ganglion cell layer (Fig. 2D). Two retinal regions were considered, the mid-temporal retina and the dorsal periphery, destined to become the regions of highest cell density, the area centralis, and of lowest cell density, respectively. Using data from the representative isodensity maps (Figs 5, 7), we calculated the number of branch points per 1000 cells in these regions (Fig. 2D). For the area centralis, values were highest up to stage 57 at 83–100 branch points per 1000 cells, decreased 20-fold to 4 branch points per 1000 cells by metamorphic climax and then declined to 2.5 per 1000 cells by the 9 cm stage. In the dorsal periphery, values were also highest up to stage 57, being 17–25 per 1000 cells and also decreased 20-fold by metamorphic climax to a value of 0.8 per 1000 cells which was maintained thereafter.

The pattern of retinal vasculature in the frog L. moorei reflects the topography of cells in the ganglion cell layer throughout development. Furthermore, specialisations within the retinal vasculature appear before the emergence of the area centralis and visual streak. The retinal vasculature develops in two distinct phases. The first takes place pre-metamorphically as the retina enlarges, and the second occurs post-metamorphically against a backdrop of continuing increases in retinal area. During the first phase, a continuous addition of blood vessels maintains their high density and a topography emerges which resembles an area centralis and visual streak. However, comparable specialisations do not begin to form amongst the underlying cells in the ganglion cell layer until later, i.e. by metamorphic climax. By the end of the first phase, the vasculature is remodelled by the removal of numerous blood vessels, although the specialised topography remains intact. During the second phase, the addition of new blood vessels ensures that their density falls only slightly as the retina continues to grow. The topography of blood vessels becomes accentuated with age as does that of the area centralis and visual streak in the ganglion cell layer.

The functional integrity of neural tissue is critically dependent on an appropriate vascular supply and, during development, numerous factors probably influence vascularisation. A major factor is oxygen levels, with high metabolic demand driving the formation of new vessels and low demand resulting in vessel retraction (Michaelson, 1948; Ashton, 1966; Ashton and Pedler, 1962; Weiter et al. 1981; Phelps, 1990; Rosen et al. 1991; Chan-Ling et al. 1992, 1995).

The vertebrate retina is initially vascularised by hyaloid vessels which enter the ventrally located optic fissure shortly after invagination of the optic cup (Mann, 1964; Grant et al. 1980). In some fish and amphibians, the hyaloid system persists, spreading as a monolayer apposed to the ganglion cell layer to give rise to the retinal vessels. In contrast, the developing mammalian retina is supplied only transiently by the hyaloid vein (Duke-Elder, 1958), with the ophthalmic vessels arising later and becoming embedded within the ganglion cell layer. The factors that control the persistence or regression of the hyaloid system have yet to be determined. However, before stage 47 in L. moorei, retinal thickness is approximately 200 μm (Grant et al. 1980; Dunlop and Beazley, 1984b) and thus greater than the theoretical maximum of 140 μm for oxygen diffusion from the choroidal supply (Chase, 1982; Dunlop et al. 1994). At the earliest larval stages in amphibians, it is likely that a high oxygen demand stimulates the transition of the hyaloid into the retinal circulation.

By the first stage at which we could examine the vasculature in L. moorei (stage 47), a pan-retinal capillary supply was evident. A process likely to underlie its formation is sprouting angiogenesis, also termed capillary budding, in which new vessels sprout from existing ones by the temporary breakdown of capillary walls followed by the migration and alignment of newly generated endothelial cells; the sprouts anastomose and only then do the vessels become patent (Folkman, 1985; Risau, 1997). In mammals, development of the ophthalmic vessels takes place by a different sequence of events. There are two processes (Ashton, 1970; Shakib et al. 1968; Henkind and De Oliveira, 1967; Flower et al. 1985), best documented in the cat (Chan-Ling et al. 1990). In the first, mesenchymal spindle cells invade the eye via the optic disk, migrate across the retina and become transformed into blood vessels within the ganglion cell layer. The now pan-retinal ophthalmic vessels undergo capillary budding to supply the deeper retinal layers, a process that starts at the area centralis and spreads peripherally.

Once the capillary bed is established in L. moorei, there is a pan-retinal addition of blood vessels during the first phase of vascularisation which maintains their density in the face of increasing retinal area. We are uncertain which process is responsible for the addition of blood vessels, although it is unlikely that spindle cell migration and transformation occur. At no stage in our Cresyl-stained whole-mounts could we detect the long, radially oriented spindle cells that in the cat were closely associated with the nerve fibre layer (Chan-Ling et al. 1990). Furthermore, in L. moorei, the entire retina was already covered by the vascular network, whereas in the cat blood vessels form in a strict centre-to-periphery gradient.

A different process of capillary formation, termed non-sprouting angiogenesis or intussusceptive growth, has not previously been reported in the retina but has been described in foetal mammalian lung and chick chorio-allantoic membrane (Burri and Tarek, 1990; Patan et al. 1996; Risau, 1997). Existing capillaries are divided by the insertion of pillars of tissue or lacunae which, as they expand, take on the appearance of a new mesh within a capillary bed. We have evidence of intussusceptive growth within the retina of L. moorei in that lacunae were observed during the first, but not the second, phase of vascular development (see Fig. 3). The small number of lacunae may indicate that their expansion to a branched capillary occurs rapidly, thus underlying the addition of many vessels. Alternatively, intussusceptive growth may give rise to only a proportion of new blood vessels and, if so, the vasculature must also expand by capillary budding. During the second phase, the vasculature is presumably growing solely by capillary budding, a suggestion supported by the slight but steady increase in the total number of branch points. Indeed, pilot data using tritiated thymidine autoradiography to label dividing cells (Fig. 1C) have shown that mitosis occurs within the vasculature even in the 9 cm adults.

In amphibians, the retinal vessels remain as a monolayer and are thus unlike those of eel-fishes (Hanyu, 1959) or some species of mammal in which one or more layers of capillaries bud off into the deeper retinal layers (Engerman and Meyer, 1965; Chan-Ling et al. 1990; McMenamin and Krause, 1993; Snodderly and Weinhaus, 1990). The stimulus for the formation of additional capillary beds must be absent from the amphibian retina. In mammals, the overlap between capillary beds and high levels of cytochrome oxidase staining suggests a correlation between retinal vasculature and functional demand (Kageyama and Wong-Riley, 1984; Snodderly and Weinhaus, 1990) rather than retinal thickness per se (Chase, 1982).

During the final part of the first phase of retinal vascularisation in L. moorei, the density of blood vessels declines dramatically, although the major vessels stay intact. As in the developing cat retina (Chan-Ling et al. 1990), capillaries in L. moorei appear to be removed actively by a degenerative process in which the lumen narrows and finally collapses. The highest number of blind-ending vessels, considered to represent degenerating capillaries, is seen at stage 57 before the precipitous decline in blood vessel numbers by metamorphic climax. Thyroxine is an essential trigger during metamorphosis for the degeneration of many tissues including the gut and respiratory surfaces, both of which are highly vascularised (Fox, 1983). An investigation using in situ labelling of endogenous endonuclease activity associated with programmed cell death (Gavrieli et al. 1992) would identify degenerating endothelial cells and determine the extent of cell death during this process. In addition, thyroxine also triggers changes in haemoglobin structure such that the oxygen-carrying capacity of blood increases from 7 volume % in larvae to 11 volume % in adults (Fox, 1983). Presumably, the lower blood vessel densities in the second phase are in part compensated by the greater oxygen-carrying capacity of the adult form of haemoglobin. The small numbers of blind-ending vessels after metamorphic climax suggest that only limited vessel degeneration takes place in the maturing vasculature.

A different pattern of exuberance and remodelling from that seen in L. moorei has been reported in the cat (Chan-Ling et al. 1990). Exuberance is confined to the leading edge of the vasculature, where spindle cells transform into the capillary plexus. More centrally, vessels have matured, presumably already pruned of excess capillaries. However, the absence of exuberance and remodelling from the developing vasculature which supplies the deeper retinal layers in the cat illustrates that remodelling is not essential for establishing all mature vascular beds.

In L. moorei, a process must occur whereby the post-metamorphic vascular tree is expanded with increasing retinal area. Expansion of the retina is pronounced in the second phase of vascular development, and it presumably triggers the rise in the total number of branch points. However, the slight but continuing decline in branch point density with post-metamorphic age indicates that the growth of the vasculature is outstripped by that of the retina itself. Within the area centralis and across the naso-temporal axis, growth of new blood vessels must, however, match that of the neural retina since a high concentration of blood vessels is maintained within this region. Elsewhere, for example in the dorsal periphery, the density of blood vessels decreases. In mammals, the retina grows non-uniformly (Mastronarde et al. 1984; Sengelaub et al. 1986; Lia et al. 1987; McCall et al. 1987; Robinson, 1987; Robinson et al. 1989). Although the mechanisms that drive such growth are unclear (Coulombre, 1956; Liu et al. 1986), the retinal vasculature presumably undergoes a similar asymmetric expansion.

Major vessels are considered to condense from capillaries (Folkman, 1985). In L. moorei, some vessels appear from the centre outwards while others originate from the peripheral collecting vessels (see Fig. 3), which together results in arteries and veins being equally spaced around the retinal circumference. In the cat, the pattern differs from that of L. moorei, with condensation occurring within a trilobed capillary bed giving rise to the three main artery–vein pairs (Chan-Ling et al. 1990) which radiate from the optic disk (Hughes, 1975).

The factors which drive the condensation of major vessels must vary widely between species to result in the strikingly different vascular trees that are observed within the vertebrates (Johnson, 1968; Michaelson, 1954; Francois and Neetens, 1974). Furthermore, an absence of major vessels within the deeper retinal layers of mammals (Chan-Ling et al. 1990) suggests that such factors also vary between retinal layers. It is possible that in amphibians, as in the cat, interactions with the underlying nerve fibre layer are important in determining the arrangement of major blood vessels. In the cat, spindle cells migrate across the nerve fibre layer, avoiding the presumptive area centralis from as early as embryonic day 29 (Chan-Ling et al. 1990). The cells seem able to avoid this region prior to the appearance of cell density gradients at embryonic day 54 (Lia et al. 1987; Robinson, 1987). Spindle cells may follow axons in the arcuate raphe, an arrangement that, in primates, is present before the inception of the fovea (Steineke and Kirby, 1993). In amphibians, the development of the area centralis and visual streak by metamorphic climax and the accentuation of these specialisations thereafter suggest that dynamic interactions between the vasculature and the neural retina continue throughout life.

During the first phase in L. moorei, the overall density of the retinal vasculature is higher than subsequently, although cell densities are within the adult range. Presumably, the high density of blood vessels is necessary to serve the greater demands of immature cells which are completing mitosis and migration, elaborating axons and dendrites and establishing functional synaptic connections (Grant et al. 1980; Holt, 1989; Frank and Hollyfield, 1987; Gaze et al. 1974). Moreover, these events follow a centro-peripheral gradient (Straznicky and Gaze, 1971). The high density of blood vessels in the central retina by stage 47 presumably reflects a higher metabolic activity in this more developmentally advanced region than in the peripheral retina, even though cell densities are greater peripherally than centrally.

At stage 51 and 57, there are two high-density patches of vasculature within the mid-temporal and mid-nasal retina which appear several weeks before the area centralis and visual streak and may indicate a locally raised oxygen demand. Because the retina grows continually at its edge (Straznicky and Gaze, 1971; Coleman et al. 1984), the area centralis and visual streak will form in regions peripheral to the two high-density vascular patches. These patches may act as a source of endothelial cells which seed newly forming capillaries as the area centralis and visual streak form within the ganglion cell layer. This chronological sequence ensures that the area centralis and visual streak will be highly vascularised, and therefore able to function, as soon as they are formed. In contrast, in the cat, the formation of capillaries at the area centralis is complete only by postnatal day 28 (Chan-Ling et al. 1990), i.e. 5–6 weeks after it can be identified by cell density gradients (Lia et al. 1987; Robinson, 1987). In L. moorei, therefore, vascularisation occurs before the area centralis forms, whereas in the cat the area centralis forms before its capillary bed. Nevertheless, in both L. moorei and the cat, the area centralis is vascularised from the onset of functional vision.

From metamorphic climax onwards, it is clear that the density of blood vessels in frog retinae is directly related to neural cell density. However, the density of the vasculature per 1000 cells is greater at the area centralis than at the dorsal periphery. The difference suggests that each cell in the area centralis imposes a greater metabolic load than its counterpart in the peripheral retina. We do not yet understand the retinal distribution of the numerous ganglion cell types present in the amphibian retina (Frank and Hollyfield, 1987). It may be that those with the highest metabolic demand are found in the area centralis whilst those cells with lower metabolic demands dominate more peripheral regions.

S.A.D. is a Research Fellow, National Health and Medical Research Council (NH&MRC, Australia). We thank W. M. Ross, R. Roberts and T. Moroney for expert technical assistance and Dr J. D. Roberts for statistical advice. This work was funded by the NH&MRC.