Retinal Development and Visual Sensitivity of Young Pacific Sockeye Salmon (Oncorhynchus Nerka)

Retinal development in some fish species appears to differ from that of higher vertebrates in two major ways. These are (1) the delayed maturation of rods after hatching (Raymond, 1985; Hagedorn and Fernald, 1992), as opposed to their delayed appearance in higher vertebrates before birth (in general, cones appear before rods in both lower and higher vertebrates, but hatching may not be the equivalent to birth in terms of retinal development; Reh, 1991; Carter-Dawson and La Vail, 1979), and (2) the age-dependent density of accessory cones presumed to be sensitive to ultraviolet light (Bowmaker and Kunz, 1987; Hawryshyn et al. 1989; Loew and Wahl, 1991; Beaudet et al. 1993). One major group of fish possessing this second feature is the salmonids, family Salmonidae (Lyall, 1957; Kunz, 1987; Beaudet et al. 1993), which also possess cone combinations maximally sensitive to short-(blue), middle-(green) and long-wavelength (red) radiation (Bowmaker and Kunz, 1987; Kusmic et al. 1993; Hawryshyn and Hárosi, 1994).

The development of salmonid retinas was first studied using light microscopy (Lyall, 1957; Ali, 1959). These authors classified photoreceptor types and described trends in the growth of several retinal layers (Lyall, 1957; Ali, 1959).

Although the reported growth trends were later supported by studies on other fish species (Blaxter, 1975; Ali and Wagner, 1980), additional photoreceptor types have since been found in early ontogeny of salmonids. For instance, reports describing the absence of rods at hatching among Oncorhynchus and Salmo species (Ali, 1959, 1963) differ from more recent findings using transmission electron microscopy (TEM) on rainbow trout (Schmitt and Kunz, 1989). In addition, previous light microscopy studies did not reveal any clear distinction between cone photoreceptor types at the embryonic and early post-hatching developmental stages (Lyall, 1957). Hence, the presence of accessory corner cones at these stages in salmonids has not been established [see radial sections in Schmitt and Kunz (1989) up to 3–4 weeks post-hatching, stages 1–3 in the eleutheroembryo (Balon, 1975)].

However, a number of recent findings have suggested that accessory corner cones may be important in the life of planktivorous fish and they are therefore likely to develop early in life. First, ultraviolet-mediated vision enhances the foraging performance in young of various fish species (yellow perch, Loew et al. 1993; rainbow trout and pumpkinseed, Browman et al. 1994). Second, the process of exogenous feeding in salmonids can start before full absorption of the yolk sac (Thomas and Shelton, 1968; Hurley and Brannon, 1969). Hence, if the time of appearance of accessory corner cones is developmentally linked to the initiation of exogenous feeding, it may occur very soon after hatching.

The goals of this study were twofold: (1) to document the growth of all retinal layers and cone types found during early ontogeny of sockeye salmon, and (2) to measure the spectral and polarized light sensitivities of the animal prior to and after smoltification. In addition to these physiological measurements, the theoretical visual acuity of the animal was calculated as an indicator of spatial resolution and foraging capabilities.

The sockeye salmon was chosen as a study subject because of its complex life history which involves several years of residence in a nursery lake as a juvenile. This habitat is shared by rainbow trout (Oncorhynchus mykiss), a species that possesses a full square mosaic arrangement with accessory corner cones and ultraviolet vision at the parr stage (pre-smolt, Beaudet et al. 1993; see van der Meer, 1992, for possible cone configurations in fish retinas). In contrast to the rainbow trout, which remains in fresh water and undergoes a transformation somewhat similar to smoltification, the sockeye salmon smolts completely before reaching the ocean, where it resides for 1–4 years before migrating back to the natal river to spawn and die (Groot and Margolis, 1991). Since ultraviolet-mediated vision may be involved in foraging and/or orientation during migration of young salmon (Hawryshyn et al. 1989; Parkyn and Hawryshyn, 1993; Novales Flamarique and Hawryshyn, 1993), sockeye salmon provides an appropriate model for studying retinal development and visual sensitivities in this behavioural and ecological context.

Sockeye salmon (Oncorhynchus nerka) were obtained at different developmental stages from Rosewall Creek hatchery (Vancouver Island, British Columbia, Canada) (Table 1). All the fish were the progeny of the same parents that had been caught in Pinkut Creek spawning channel (Babine Lake, British Columbia, Canada); eggs were fertilized at the hatchery. Embryos were raised in the dark until hatching; all post-hatching development occurred under a 12 h:12 h L:D cycle (see Fig. 3 for the spectral composition of the lamp). Fish were kept in a circular tank with flowing water at 9 °C and fed a daily diet of BioDiet Grower pellets (Bio-Products Inc., Warrenton, Oregon, USA).

Seven fish from each developmental stage (except for the embryo, parr and smolt stages, Table 1) were anaesthesized by brief immersion in a 0.1 g l⁻¹ solution of tricaine methanesulphonate (MS-222). They were then decapitated and the heads immersed in primary fixative (2.5 % glutaraldehyde, 1 % paraformaldehyde in 0.06 mol l⁻¹ phosphate buffer, pH 7.2 at 4 °C, Ali and Anctil, 1976) for light and transmission electron microscopy. Six eggs, the yolks of which were perforated to enable rapid penetration of the fixative, were also fixed. Eyes from specimens at the parr and smolt stages were fixed following electrophysiological recordings.

Fixed retinas from six specimens at each stage were removed from the eyes while the remaining eye was treated in its entirety. The tissue was then post-fixed with osmium tetroxide, dehydrated through an ethanol series and embedded in Epon media (Ali and Anctil, 1976). Three of the retinas at each stage were sectioned radially (along the long axis of the photoreceptors) while the remaining three retinas were cut tangentially (across the width of the photoreceptors). Tissue shrinkage was estimated at 15 % by measuring differences in retinal size prior to fixation and after embedding. However, no corrections were made for shrinkage in the data presented except in the visual acuity calculations.

The retina of the sockeye salmon can be divided into eight layers from the sclera to the vitreous humour (Figs 1A,B, 2): retinal pigment epithelium (RPE), visual cell layer (VCL), outer nuclear layer (ONL), outer plexiform layer (OPL), inner nuclear layer (INL), inner plexiform layer (IPL), ganglion cell layer (GCL) and nerve fibre layer (NFL). A photoreceptor square mosaic can be observed from tangential sections cut across or near the ellipsoids of the cones (Fig. 2B,C). More vitreally, the INL is made up of at least four types of cells and their connections. Oriented parallel to the retinal surface are the horizontal cells (Fig. 2E); these are followed by bipolar cells (Fig. 2F) and, at the inner border of the INL, by amacrine cells (Fig. 2G). Mueller cells extend the entire length of the retina. The GCL exhibits large pale-staining ganglion cell bodies (Fig. 2H); neural dendrites and axons exiting the GCL form the NFL (Fig. 1B).

Tangential sections at the ellipsoid level of photoreceptors enabled the density of each type of cone (accessory corner cone, central cone and double cone) and the diameter of each individual type of cone to be measured. This was done by counting, for each retina, the numbers of each cone type in a 0.01 mm² piece of central retina adjacent to the optic nerve head (except for the smolt retina where dorsal, nasal and temporal regions were also examined). Similarly, the diameter of 10 cones of each type was measured for each retina. Averages from the three sets of values were pooled and mean densities and diameters for each type of cone computed for each developmental stage. The counts and measurements were made from semi-thin sections using a Zeiss microscope (model Universal R, 100× overall magnification) and a calibrated ocular micrometer. A cone was counted if at least half of its ellipsoid was within the perimeter of the 0.01 mm² area. Only the long axis of the cross section of double cones was measured.

The thickness of the retinal layers was measured from radial cross sections. The width of the RPE was measured from the outermost part of the RPE layer to the average tips of the pigment projections (see Ali, 1963). The VCL covered the distance from the external limiting membrane (elm, Fig. 1) to the tips of the double cone outer segments. The limits for the other measurements follow from the previous description of each layer and Fig. 1. Averages from sections of three retinas were calculated for each developmental stage.

where f (the focal length in mm) was approximated from Matthiessen’s ratio (f=2.55r, r being the radius of the lens), S is the degree of shrinkage, and n is the density of cones (in total) per 0.01 mm² area (Blaxter and Pattie Jones, 1967). The minimum separable angle gives an indication (given a good-quality lens, Sivak, 1990) of the visual acuity of the animal. This is an important variable as it is closely associated with the development of the retina and the brain (Rahman et al. 1979; Jeserich and Rahman, 1979), and thus to the potential foraging performance of the animal.

Animals were anaesthetized by immersion in MS-222 (0.1 g l⁻¹) and paralysed by Pavulon injection (pancuronium bromide, 0.038 mg g⁻¹ body mass). The fish were irrigated with MS-222 solution (0.005 %) during surgery to expose the optic tectum. A local anaesthetic (tetracaine, 0.5 %) was applied to the surgical area.

A Teflon-coated silver wire electrode was inserted rostro-ventrally through the optic tectum into the optic nerve. This electrode monitored compound action potentials from the optic nerve while a reference electrode was inserted into one of the nares of the fish and its signal subtracted from that of the recording electrode. The resulting difference signal was amplified using a Grass preamplifier (P50 series), with bandpass between 0.3 Hz and 0.1 kHz, and was displayed on an oscilloscope while simultaneously acquired by computer.

The optical system used in this study consisted of a stimulus channel (with a 300 W xenon light source, Oriel) and two background channels (using 250 W tungsten–halogen sources, EJH Spectro). The stimulus energy and spectral output were controlled by an Inconnel neutral density wedge and a monochromator. Using Corion interference filters, five types of background lighting conditions were produced so that the different cone mechanisms present in sockeye salmon could be selectively isolated and an average polarization sensitivity curve obtained (Fig. 3).

During spectral sensitivity experiments, the output from the three optical channels converged onto a diffuser placed in front of the fish’s eye. This ensured an even illumination from all the channels across the entire retina. In the case of polarization sensitivity measurements, each channel had its own diffuser, and a rotatable polarizer was placed immediately following the diffuser in the stimulus channel.

The channels were positioned so that the stimulus polarized light was incident on the central retina, where background radiances overlapped as much as possible. Furthermore, no light from the background channels crossed the polarizer.

Prior to testing, the fish was adapted to a particular background light for 1 h. Then 13 stimulus wavelengths ranging from 350 to 660 nm were presented to the fish in an order that precluded the adaptation of any particular cone mechanism. The stimulus duration was 750 ms. Each wavelength was presented in an increasing series of 0.2 log unit intensities. The time between stimuli in each series was 30 s. This procedure was slightly different for polarization sensitivity measurements. In this case, one wavelength was chosen for the entire experiment (λ=380 nm) and the polarizer was rotated from 0 ° to 180 ° in 30 ° increments.

Compound action potentials (CAPs) were observed at the onset and at the end of the light stimulus (ON and OFF signals, Fig. 4A). For each wavelength, the voltage amplitude of these responses was plotted against increasing radiance (Fig. 4B). A third-order polynomial function was fitted to the data points and the radiance required to elicit a 20 μV response was chosen as the threshold response. This signal level was chosen for two reasons: (1) it was very close to the absolute response threshold for all wavelengths tested (below which no response was obtained), and (2) it was in the linear part of the response curve (ensuring a regular and predictable response). Further details of the procedure may be obtained from Beaudet et al. (1993).

All animal experimentation was in accordance with guidelines set by the Canadian Council on Animal Care.

where R is the response curve, A_i(λ) is the absorbance of pigment i at light of wavelength λ, and p is an exponent resulting from the mathematical requirement that the function describing the spectral sensitivity curve is differentiable at the origin (see Sirovich and Abramov, 1977). The variables k_i are differential coefficients resulting from the previous assumption. They are coupling constants that are derived from the best fit of the model to the data. The absorbance values used in the model are those of photopigments with λ_max at 365 nm (ultraviolet), 434 nm (short), 531 nm (middle) and 576 nm (long), as determined using microspectrophotometry (MSP) on rainbow trout cones by Hawryshyn and Hárosi (1994) (MSP data are not available for sockeye salmon). All response curves that are a function of the absorbance of one or more pigments must conform to the above equation. The coupling coefficients and the exponent p are determined by the inputs of each photopigment to the spectral sensitivity curve and the complex interactions between photoreceptor mechanisms at various processing levels.

The procedure of fitting one best template to an isolated mechanism may give an erroneous approximation to the peak wavelength absorption of the isolated mechanism. Since isolation is never perfect, one must consider the inputs of other mechanisms that act in the spectral region examined and determine that these are significantly lower than the input of the mechanism being isolated. Thus, fitting individual polynomial templates to spectral sensitivity curves may be an oversimplification.

Ten days prior to hatching (98 % of the fish hatched within a 2 day period 10 days after our first sampling), sockeye salmon embryos possess all five photoreceptor types present in the alevin or parr conditions (ultraviolet-, short-, middle-and long-wavelength-absorbing cones and a rod, Fig. 5A). Similarly all the retinal layers are found (Fig. 5B). Cones are arranged in a square mosaic throughout the central retina with, on average, four rods evenly dispersed around the centre cone. Photoreceptor outer segments are present in both cones and rods (Fig. 5A,C). The retina is least developed along two previously described growth zones: the embryonic fissure and near the periphery (Kunz and Callaghan, 1989).

The early growth rate of the retina is linear (Fig. 6A) and is the result of different trends followed by each retinal layer (Fig. 6B,C). The INL and GCL decrease in size during development, while the remaining layers increase in size. Cone diameters increase in size [Fig. 7A, ANOVA (model: diameter = stage conetype stage × conetype), r²=0.997, Pr>F=0.0001, week 1 diameters are significantly different from those at hatching at α=0.05 significance level, Student–Newman–Keuls and Waller–Duncan K-ratio t-tests of means). Single cones (cc and acoc) are very similar in diameter but their tangential growth is slower than that of double cones. Cone density decreases during development [Fig. 7B, week 1 densities are significantly different from those at hatching, ANOVA (model: density = stage conetype stage × conetype), r²=0.99, Pr>F=0.0001, Student–Newman–Keuls and Waller–Duncan K-ratio t-tests for means at α=0.05). This decrease is very rapid during the early stages of development but slows down from approximately week 2 post-hatching onwards. During this time, the cone pattern remains the same; however, the rod pattern is lost (see Lyall, 1957).

By the smolt stage, sockeye salmon have fewer accessory corner cones throughout most of the retina (Fig. 8, and Kunz, 1987). Areas still possessing these cones were found near the periphery and in the central retina close to the embryonic fissure. In addition to the usual cone patterns, doublet, triplet and quadruplet cone associations were found as part of incomplete square mosaics (Fig. 8A–F, see similar findings in Ahlbert, 1975). The cone mosaic, although square in most of the areas examined (especially when approaching the central retina), varied towards a row pattern in more peripheral areas and was more random immediately dorsal to the embryonic fissure (Fig. 8F). We know that these fish had already smolted because 30 individuals were subjected to a seawater challenge that resulted in zero mortality (see also natural smoltification ages in Groot and Margolis, 1991, p. 45).

The theoretical visual acuity of sockeye salmon improves with age [i.e. the minimum angle for stimulation of two non-neighbouring cones decreases; Fig. 9, assuming good lens quality; ANOVA (model: stage = minimum separable angle), r²=0.938, Pr>F=0.0001, week 6 minimum separable angles are significantly different from those at hatching, Student–Newman–Keuls and Duncan’s multiple-range tests at α=0.05]. This result parallels a decrease in total cone density and a simultaneous increase in lens radius (Fig. 9).

Under the spectral isolating backgrounds described previously, sockeye smolt ON responses revealed four types of cone mechanisms with peaks of sensitivity to ultraviolet light (λ_max=380 nm), blue or short-wavelength light (λ_max=425 nm), green or middle-wavelength light (λ_max=520 nm) and red or long-wavelength light (λ_max=635 nm) (Fig. 10). Parr exhibited similar sensitivities except for the dominance of the ultraviolet cone mechanism under ultraviolet isolation conditions (Fig. 11). When ultraviolet light was added to the ultraviolet isolation background (Fig. 10A), the peak in ultraviolet sensitivity diminished (Fig. 10B) leaving mainly a short-wavelength mechanism input (Table 2). The OFF responses follow the general shapes of the ON responses, although their spectral maxima are slightly shifted (Figs 10, 11). The OFF responses under ultraviolet isolation are driven primarily by a short-wavelength mechanism input (Figs 10A, 11; Table 2). Under scotopic conditions, the rod-driven curve showed a maximum at 540 nm (Fig. 10E); however, the maximum absorbance for the mechanism is probably lower (around 525–530 nm) given the symmetry of photopigment α absorption bands. Under a white background mimicking dusk conditions in a mesotrophic lake (Novales Flamarique et al. 1992), sockeye salmon smolts exhibited polarization sensitivity for the ON response with maximum at a 90 ° E-vector angle using a 380 nm stimulus (Fig. 10F).

The presence and size of the various photoreceptor types reported in this study for pre-hatched sockeye salmon markedly differ from past findings using this and other salmonid species. Previous investigations of sockeye salmon retinal development at the light microscope level reported the absence of rods at hatching (Ali, 1959). Similar results were obtained with other species of salmon and trout (Ali, 1959, 1963; Lyall, 1957). More recently, using transmission electron microscopy, Schmitt and Kunz (1989) showed that both developing cones and rods were present in rainbow trout at hatching. These authors also reported the appearance of a cone square mosaic within the first 2 weeks after hatching (stage 3, Balon, 1975), after the cones had differentiated completely. Our results, however, show developing cones and rods arranged in a square mosaic in the central retina of the embryo (Fig. 5A,C). In addition, the photoreceptor outer segments observed at this stage are much more developed than previous ones reported for very early post-hatching stages in this and other Oncorhynchus species (Ali, 1959, 1963; Schmitt and Kunz, 1989). These differences in maturation rates may be population-specific as other environmental parameters in our study are similar to those reported previously (Ali, 1959; Schmitt and Kunz, 1989). However, it should be noted that O₂ concentrations, which are an important variable affecting the rate of salmonid development (Groot and Margolis, 1991), were not reported in previous studies; O₂ concentrations in this study were 11.25±0.25 p.p.m.

Rod outer segments were observed at the embryonic stage (Fig. 5A), although they appeared less developed than cone outer segments (Fig. 5C for cone outer segment lengths, rod outer segment lengths not shown in radial section). Such results also differ from those reported by Schmitt and Kunz (1989) for rainbow trout, in which rod outer segments first appeared after hatching. Nonetheless, together with observations from rainbow trout (Schmitt and Kunz, 1989), our results show that the delayed post-hatching appearance of rods reported for other teleosts does not occur in salmonids [post-hatching rod appearance occurs in the herring Clupea harengus pallasi (Blaxter and Pattie Jones, 1967), the zebrafish Brachydanio rerio (Branchek and Bremiller, 1984), the goldfish Carassius auratus (Raymond, 1985), the cichlid Haplochromis burtoni (Hagedorn and Fernald, 1992) and the walleye Stizostedion vitreum (Wahl, 1994)]. Pre-hatching appearance of rods has also been reported in the yellow perch (Ahlbert, 1973), as well as in some amphibians (Grant et al. 1980). In the European eel Anguilla anguilla, rods are present before cones (Pankhurst, 1984).

The growth rate of the sockeye salmon retina is linear during the early development of the animal (Fig. 6A; see also Ali, 1963, for this developmental period). The growth trend slows thereafter (Fig. 8 legend) and possibly reaches a plateau in the adult condition (Ali, 1963). The variations in thickness reported for some of the retinal layers agree with previous results for other salmonids (Lyall, 1957; Ali, 1959, 1963), other teleosts (e.g. Branchek and Bremiller, 1984; Raymond, 1985) and small mammals (Reichenbach et al. 1991). While the various plexiform and fibre layers thicken because of the increasing complexity of neural connections being formed, the INL decreases in thickness as the ONL and VCL grow. It is thought that ‘rod-precursor’ cells located in the INL migrate to the ONL, perhaps guided by Mueller cells (Raymond and Rivlin, 1987; Hagedorn and Fernald, 1992), where they give rise to rods (Raymond, 1985; Raymond and Rivlin, 1987). This mechanism would explain the inverse relationship between the thickness of the ONL and the INL during early growth (Fig. 6B). Fast incorporation of rods into the VCL and stretching of the retina during the first week post-hatching would also agree well with the precipitous drop in cone density observed during this period (Fig. 7B). However, the possibility that stretching of the retina to incorporate rods explains our results (see Hagedorn and Fernald, 1992) remains a hypothesis to be tested. This could be done by measuring the numbers of rods in square mosaics and the distances between cones in the square mosaics prior to and after rod incorporation. Nevertheless, we favour the idea of simultaneous stretching of the retina (due to eye growth) and rod incorporation in the space created.

The tight square mosaic cone arrangement present in the embryo progressively disperses as the retina grows, additional rods are added and new retina is formed along the embryonic fissure and the periphery (Fig. 8G, Kunz and Callaghan, 1989; Hagedorn and Fernald, 1992). Mean cone diameters increase during development but the entire retina expands at a faster rate. As a result, cone density decreases (Fig. 7B). Despite this, the theoretical visual acuity increases as a result of eye and lens growth which overcompensates for the decrease in cone density (Fig. 9, Blaxter and Pattie Jones, 1967; Neave, 1984; Fernald, 1991). The minimum angles of separation derived for sockeye salmon permit us to calculate the maximum distance at which a young fry (after absorption of its yolk sac, when it is approximately 3 cm in length) would locate a prey item. Considering the average prey item to be 1.20 mm in length (see Browman et al. 1994), the maximum distance at which it would be detected can be calculated to be 78 mm. In accordance with this prediction, maximum prey move distances for rainbow trout of this size feeding on 1.20 mm Daphnia pulex oscillate around 75 mm (Browman et al. 1994). Prey move distances refer to the average distances covered by the fish, between pauses, when it scans the surroundings looking for prey. These distances are strongly correlated with the distances at which prey are located (Bell, 1990).

The differences in tangential growth rates of the various cone types can be related to the visual ecology of sockeye salmon. Using geometrical methods, van der Meer (1992) has shown that square mosaic patterns with central elements, and with or without single ‘accessory’ elements (cone configurations PAT6C and PAT6CA in Fig. 4 of van der Meer, 1992), are well adapted for vision in bright and dim light. These conclusions are based on the larger area that the double cones occupy in the square mosaic (supposedly driving the luminosity system), the better cone packing that could be achieved with single and double cones as opposed to that with single cones alone, and the advantages for maximum photon catch in bright environments. In dim environments, maximization of the photon-catching abilities of double cones would lead to rearrangement of the area occupied by these at the expense of single cones, resulting in row mosaics (van der Meer, 1992). All these arguments support the contention that sockeye salmon, a fish experiencing many different light environments throughout its lifetime, should have a double cone mosaic arrangement. In the case of Babine lake salmon, limnological data suggest that these fish live in waters that can be classified as oligotrophic or mesotrophic (Narver, 1967; Hartman and Burgner, 1972). Thus, the light field to great depths should expand the full spectrum from ultraviolet to long wavelengths and be relatively bright (spectral irradiance >1015 photons m⁻² s⁻¹, Smith et al. 1973), while the lower photic zone should probably peak in the short to middle wavelengths (Novales Flamarique et al. 1992). Young salmon in this lake system exhibit a diel vertical migration: the fish stay at depths between 24 and 40 m during the day and within 3–5 m of the surface during crepuscular periods (Narver, 1967). Thus, the fish inhabit a variety of photic habitats where a square mosaic with predominant double cone surface area would maximize photon catch and, indirectly, the chances of survival due to prey location and predator avoidance.

Our observations for the smolt retina showing various types of cone arrangements agree with Ahlbert’s (1975) observations for other salmonid species. According to Ahlbert and other authors, the retinas of Atlantic salmon and trout shift from a full square mosaic retina at the parr stage to one with fewer accessory corner cones and more row patterns in the adult (Kunz and Callaghan, 1989; Beaudet et al. 1993). The cone densities found are also characteristic of the young smolt condition (Ali, 1963; Ahlbert, 1975), and their distribution implies a zooplanktivorous existence (highest densities in the ventro-temporal retina, Fig. 8, Ahlbert, 1969, 1975).

The peak sensitivities of the ON responses of cone mechanisms found for sockeye smolts are different from those found for rainbow trout and goldfish using behavioural methods (trout maxima at 440, 535 and 630 nm, Douglas, 1983; goldfish maxima at 365, 440, 540 and 640 nm, Hawryshyn, 1991). Using a similar optic nerve recording technique, Beaudet et al. (1993) showed that parr rainbow trout possess an ultraviolet cone mechanism peaking at 390 nm. These authors further showed that the ultraviolet cone mechanism was lost after the fish ‘pseudo-smolted’ (see also Hawryshyn et al. 1989) and that this was accompanied by a loss of accessory corner cones in the central retina. Our results provide evidence of ultraviolet sensitivity in smolt sockeye salmon from the presence of a residual maximum peaking at 380 nm under ultraviolet isolating conditions (Figs 10A, 11). This peak is indeed driven by the ultraviolet mechanism since it diminishes when ultraviolet light is introduced into the background illumination (Fig. 10B; Table 2). This conclusion is also supported mathematically as the fit becomes much worse if the model does not take into consideration the input from an ultraviolet cone mechanism (Fig. 10A, see SS in Table 2). In addition, γ band contributions from the middle-and long-wavelength mechanisms under ultraviolet isolating conditions are by comparison much smaller (Table 2), as would be expected since the main peaks of these two mechanisms are below the sensitivities observed in the ultraviolet range, and β bands are at least 0.5 log units below the main peaks (Fig. 10C,D; Beauchamp and Lovasik, 1973).

The difference between these results and those of Beaudet et al. (1993) may be due to differences in the visual systems of the species studied. However, we believe that a more likely explanation resides in the fact that a diffuser was used to obtain spectral sensitivity curves in the present study. Beaudet et al. (1993) were stimulating preferentially the centro-ventral retina while, with a diffuser, we were stimulating all regions of the retina. Our results, therefore, suggest that a population of ultraviolet cones remains functional in the retina (Fig. 8), which may partly explain why ultraviolet single units are found in the torus semicircularis of large juvenile rainbow trout (Coughlin and Hawryshyn, 1994).

The use of a diffuser, however, does not explain why sockeye smolts exhibit polarization sensitivity while ‘pseudo-smolted’ rainbow trout do not (Fig. 10F, Hawryshyn et al. 1990; Parkyn and Hawryshyn, 1993). The differences in this instance may be due to the larger size of the post ‘pseudo-smolted’ fish tested by Parkyn and Hawryshyn (1993) (50–78 g), compared with our fish (Table 1, i.e. the retina may progressively get rid of more ultraviolet cones with time after smoltification). Alternatively, as with the spectral sensitivity results, the differences may reside within the visual genetic make-up of each species.

The peak OFF responses of the middle-and long-wavelength mechanisms differ from those of goldfish (DeMarco and Powers, 1991) but resemble those of rainbow trout (Beaudet et al. 1993). In contrast to results found for goldfish (DeMarco and Powers, 1991), the short-wavelength mechanism input of sockeye smolts dominates the OFF response under short-wavelength and ultraviolet isolation conditions. The Simplex model nevertheless predicts a smaller input from the middle-or long-wavelength mechanisms under ‘unfavourable’ ultraviolet or short-wavelength isolation conditions (Table 2), which suggests that these inputs may be important components of the OFF response under natural illumination. However, it is the peak OFF response of the middle-wavelength mechanism as well as the peak rod ON response that are within the spectral maxima of wavelengths penetrating meso-eutrophic lakes (Novales Flamarique et al. 1992). These neuronal pathways may therefore be the major constituents of a shadow detector mechanism permitting prey detection against the downwelling light in bright and dim environments, as suggested by Beaudet et al. (1993).

The ecological advantages of ultraviolet vision for prey detection and orientation have been discussed in various studies (e.g. Bowmaker and Kunz, 1987; Hawryshyn et al. 1989; Loew and Wahl, 1991). Some of these advantages probably apply to the Babine Lake system where the majority of sockeye salmon performing the diel migration are yearlings (supposedly ultraviolet-sensitive) following copepod and Daphnia prey (Scarsbrook et al. 1978). If our results also hold for adult fish, polarized ultraviolet light may be used by the fish in the open ocean to orient towards feeding grounds or during homeward migration (Kunz and Callaghan, 1989; Hawryshyn et al. 1990; Novales Flamarique and Hawryshyn, 1993). Such use is probably more meaningful than outward migration from lakes, given the scarcity of environmental cues in the former environment. However, it is only in lakes that migratory observations have been extensively investigated for correlation purposes (Johnson and Groot, 1963; Groot, 1965; Dill, 1971).

We wish to thank Mr Luc Beaudet for technical help and Mr William Harrower for the sockeye specimens. We would also like to thank Dr Dave Coughlin for his version of the Simplex program. Dr Christina Wahl-Loew, Luc Beaudet and Daryl Parkyn provided valuable comments on the manuscript. This research was supported by a NSERC operating grant to C.W.H. and graduate scholarships to I.N.F.; I.N.F. was supported by a NSERC postgraduate scholarship and by a University of Victoria President’s Research Award.