Retinal ganglion cell topography and spatial resolution in the Japanese smelt Hypomesus nipponensis (McAllister, 1963)

Abstract

Vision plays a crucial role in the life of the vast majority of vertebrate species. The spatial arrangement of retinal ganglion cells has been reported to be related to a species’ visual behavior. There are many studies focusing on the ganglion cell topography in bony fish species. However, there are still large gaps in our knowledge on the subject. We studied the topography of retinal ganglion cells (GCs) in the Japanese smelt Hypomesus nipponensis, a highly visual teleostean fish with a complex life cycle. DAPI labeling was used to visualize cell nuclei in the ganglion cell and inner plexiform layers. The ganglion cell layer was relatively thin (about 6‐8 μm), even in areas of increased cell density (area retinae temporalis), and was normally composed of a single layer of cells. In all retinal regions, rare cells occurred in the inner plexiform layer. Nissl‐stained retinae were used to estimate the proportion of displaced amacrine cells and glia in different retinal regions. In all retinal regions, about 84.5% of cells in the GC layer were found to be ganglion cells. The density of GCs varied across the retina in a regular way. It was minimum (3990 and 2380 cells/mm² in the smaller and larger fish, respectively) in the dorsal and ventral periphery. It gradually increased centripetally and reached a maximum of 14,275 and 10,960 cells/mm² (in the smaller and larger fish, respectively) in the temporal retina, where a pronounced area retinae temporalis was detected. The total number of GCs varied from 177 × 10³ (smaller fish) to 212 × 10³ cells (larger fish). The theoretical anatomical spatial resolution (the anatomical estimate of the upper limit of visual acuity calculated from the density of GCs and eye geometry and expressed in cycles per degree) was minimum in the ventral periphery (smaller fish, 1.46 cpd; larger fish, 1.26 cpd) and maximum in area retinae temporalis (smaller fish, 2.83 cpd; larger fish, 2.75 cpd). The relatively high density of GCs and the presence of area retinae temporalis in the Japanese smelt are consistent with its highly visual behavior. The present findings contribute to our understanding of the factors affecting the topography of retinal ganglion cells and visual acuity in fish.

We studied the topography of retinal ganglion cells in the Japanese smelt. The spatial density of ganglion cells was minimum (2,360‐3,960 cells/mm²) in the dorsal and ventral periphery. It gradually increased centripetally to reach a peak of 10,960‐14,275 cells/mm² in the temporal retina, where a pronounced area retinae temporalis was detected. The theoretical anatomical spatial resolution was minimum in the ventral periphery (1.26‐1.47 cpd) and maximum in area retinae temporalis (2.70‐2.83 cpd).

1. INTRODUCTION

Vision plays a crucial role in the life of the vast majority of vertebrate species. The emergence of the visual system has contributed much to the evolution and adaptive radiation of vertebrates and allowed them to occupy a wide variety of ecological niches (Rodieck, 1998). All vertebrates share the same general organization of the visual system. There are, however, differences related to both overall visual processing complexity and adaptations to the visual environment. In particular, the number and topography of retinal ganglion cells (GCs) are to a great degree determined by a species’ visual ecology (Lythgoe, 1979). GCs are the output retinal neurons. They participate in the analysis of visual information and transmit it further to the brain. Their distribution may vary considerably between even related species reflecting differences in the visual environment. In accordance with the terrain theory of Hughes (Hughes, 1977), the retinae of terrestrial and aquatic vertebrates dwelling in open spaces often have a horizontal band of increased photoreceptor and GC density, so‐called visual streak. This allows an animal to scan the horizon with higher visual acuity without noticeable eye movements (Collin, 1999). In contrast, species found in heterogeneous environment without a substantial panoramic sight feature a more or less concentric increase in the GC density, so‐called area retinae. It may be an area retinae centralis, area retinae temporalis (art), area retinae dorsalis, and etc., depending on its exact location in the retina. This specialization allows an animal to spot the respective region of the visual field in greater detail.

Fishes with their huge ecological diversity provide many striking examples of the perfect match between the retinal specializations and visual environment. Thus, Collin & Partridge revealed (1996) a prominent area centralis in the centro‐lateral portion of the "main" (ventral) retina in two hatchetfish species (deep‐sea fishes with tubular eyes). The authors propose that this specialization may facilitate prey detection in the dorsal visual field. In adult European hakes, Merluccius merluccius, a visual streak and three areae were found in the nasal, temporal, and central parts of the retina (Bozzano & Catalán, 2002). The presence of the visual streak correlates with the benthopelagic niche of adult hakes and their diet including both pelagic and benthic prey. Altogether, the visual streak and three areae presumably allow the hake to detect prey in both the rostral, central, and caudal parts of the visual field while monitoring the environment for potential predators (Bozzano & Catalán, 2002).

Spatial resolution (SR) is an important physiological parameter related to visual object detection and/or discrimination. It is affected by many factors such as topography of retinal neurons, eye geometry, and visual signal processing in the retina and brain. It also depends on visual conditions (light intensity, contrast, etc.) and may change with time (Kohn, 2007). There are two traditional approaches to SR estimation, the anatomical and behavioral ones. In the former, the density of photoreceptors and/or GCs and eye optical geometry are used to calculate the upper theoretical limit of SR. GCs are the output retinal neurons. Their density is normally inferior to that of photoreceptors, although there are some exceptions. For instance, in the human fovea, two midget GCs correspond to each cone (Jusuf et al., 2006). GCs therefore act as a "bottleneck" to visual signal flow from the retina to the brain. For this reason the density of GCs has long been considered a more accurate anatomical measure of visual acuity (VA) (Caves et al., 2017; Wagner, 1990). There is, however, sound behavioral evidence that the SR in fish may be higher than that suggested by the GC distribution and may approach the theoretical limit imposed by the density of photoreceptors (Haug et al., 2010; Miller et al., 1993; Northmore et al., 2007). Recent evidence suggests that this may be due to non‐linear signal processing in GCs (Maximov et al., 2013). It is not clear at the moment how universal this mechanism is in terms of both phylogenetic occurrence and GC types involved.

The disadvantage of the anatomical approach to SR estimation is that it cannot account for optical imperfections of eye and higher‐order visual signal processing (Caves et al., 2017), as well as for the heterogeneity of cell types and functional circuits in the retina (Stone, 1983; see below). Behavioral SR estimates do not have these constraints but may vary substantially depending on light intensity, stimuli used, etc. (Parker et al., 2017). Behavioral SR estimates for a particular species are usually somewhat lower than anatomical ones (Donner, 1951; Parker et al., 2017; Pettigrew et al., 1988). One reason for this is that the number and density of the GC type(s) involved in a certain visual behavior type are lower than those of all GCs (Maturana et al., 1960; Witpaard & Terkeurs, 1975). Nonetheless, at least in fish, there is a strong and statistically significant correlation between anatomical and behavioral SR estimates (R = 0.7 at p = 0.08; derived from the data in Table 1 of Parker et al., 2017).

Table 1.

Spatial arrangement of retinal ganglion cells and spatial resolution in Hypomesus nipponensis.

Retinal wholemount	Fish standard length, mm	Equatorial eye diameter, mm	Retinal area, mm2	Total number of ganglion cells (x 103 cells)	Minimum cell density (cells/mm2)	Maximum cell density (cells/mm2)	PND, mm	Minimum TASR (cpd)	Maximum TASR (cpd)
HN 2R	59	4.4	26.58	177	3960	13170	2.57	1.46	2.70
HN 3L	61	4.5	27.64	189	3960	14275	2.60	1.47	2.83
HN 1L	68	5.5	35.16	212	2360	10960	2.90	1.26	2.75

The retinal area values are given for fresh retinae.

Abbreviations: cpd, cycles per degree; PND, posterior nodal distance; TASR, theoretical anatomical spatial resolution.

There are tens of studies dealing with GC topography in bony fish species belonging to different systematic groups and occupying different ecological niches. However, there are still large gaps in our knowledge on the issue which is not a surprise taking into account a huge systematic and ecological diversity of bony fishes. We studied the topography of GCs and estimated SR in the Japanese smelt, Hypomesus nipponensis [McAllister, 1963]. This is a relatively small smelt, with adults reaching 12 cm total length in the Primorsky region (Russian Far East) (Parpura & Kolpakov, 2001). There exist freshwater (lake and river‐lake) and migratory (anadromous) Japanese smelt forms (Hamada, 1961; McAllister, 1963). This species is a traditional object of recreational and commercial fishery. Anadromous smelts dwell in brackish or coastal waters and go to near‐shore rivers for spawning. Freshwater smelts live in lakes; they may spawn in situ or migrate to inflow rivers for spawning (Hiroki, 2004). The Japanese smelt is a diurnal and crepuscular benthopelagic predator with a broad feeding spectrum (Riede, 2004; Zhou et al., 2013). Its diet includes cladocerans, rotifers, copepods, benthic animals, surface insects, amphipods, and algae. It has highly mobile lateral eyes (Figure 1; IP, SK, unpublished observations). Its coloration is typical of pelagic fishes. It has a dark back and silvery flanks and belly with pale fins and a narrow dark stripe along the midline. The Japanese smelt displays a pronounced sexual dimorphism in many traits (Parpura & Kolpakov, 2001). As is seen from above, vision plays an important role in the biology of the species. However, to date, the visual system in osmerids was addressed in few studies, all dealing with the mosaics and spectral sensitivity of retinal photoreceptors (Ali & Anctil, 1976; Kondrashev et al., 2017; Montgomery & Carton, 2008; Reckel et al., 2003).

Figure 1.

The snout of a 6.8 cm (total length) Japanese smelt, Hypomesus nipponensis. a — the intact (left) eye. b — the right eye with the cornea and lens removed. Note the ventronasal orientation of the embryonic fissure (arrowheads). The location of the optic disc is marked with an asterisk. Scale bar =2 mm

2. MATERIALS AND METHODS

2.1. Material and specimen preparation

Adult fish (Hypomesus nipponensis McAllister, 1963) 5.8‐6.8 cm long (total length) were caught in December 2017 in the Volchanka river (Nakhodka suburbs, the very south of the Russian Far East). The fish were kept in aerated aquaria before the experiments. Seven fish were used in the study. Three fish 5.9‐6.8 cm long were used for the analysis of GC topography (one eye from each fish) and eye optics examination by computerized tomography (the other eye from each fish). Two fish 5.8 and 6.6 cm long were used for the estimation of the proportion of non‐ganglion cells in the GCL and IPL (one eye from each fish). Two fish 6.1 and 6.6 cm long were used for the evaluation of eye tissue shrinkage during eye preparation for computerized tomography.

A fish was deeply anesthetized and sacrificed with an overdose of 0.1% solution of 3‐aminobenzoic acid ethyl ester, methanesulfonate salt (MS‐222, Sigma, St. Louis, MO). MS‐222 is widely used as a single‐drug anesthetic for surgical interventions in anamniotes. The eyes were enucleated, the retinae were isolated, and fixed with 2.5% paraformaldehyde in 0.1 M PBS (pH 7.4) at 4°C for 50 min. They were then washed in several changes in the same PBS, stained with 4’,6‐diamidino‐2‐phenylindole (DAPI; Thermo Fisher Scientific Cat#D1306), and wholemounted onto a slide in FluorSave (Calbiochem‐Novabiochem, USA) with the nerve fiber layer uppermost. The linear shrinkage was measured for each wholemount. It varied from 2.8 to 3.3% and was approximately uniform across the wholemount.

The embryonic fissure was used as a landmark to note and confirm the orientation of the retina. The native location of the fissure was determined as follows. A fish was sacrificed and fixed as described above. The front wall and lens of each eye were removed to expose the embryonic fissure. The position of the fissure relative to the frontal plane was documented using a Carl Zeiss Stereo Binocular Microscope with a digital camera.

All applicable international, national, and institutional guidelines for the care and use of animals were followed. The animals were treated in strict accordance with the EU Directive 2010/63/EU for animal experiments and the rules for work with test animals established by the Ethics Committee of the National Scientific Center of Marine Biology, FEB RAS.

2.2. Cell topography analysis

The retinae were examined using laser confocal microscopy (Zeiss LSM 700 inverted scanning microscope, Zeiss 20/1.0 Plan‐Apochromat objective). The ganglion cell layer (GCL) was outlined close to the ora serrata and along the radial cuts excluding more sclerad retinal layers. The optic disc was also outlined and its area was subtracted from the total retinal area.

The number of cells in the GCL and inner plexiform layer (IPL) were calculated from series of confocal images obtained in 425 x 425 µm frames covering the entire wholemount area. The IPL was included into analysis since in most, if not all, fishes studied, a portion of GCs are displaced to this layer (e.g., Bailes et al., 2006; Cook et al., 1992; Pushchin & Karetin, 2014). Each frame was scanned at 405 nm excitation wavelength, which is within the absorption range of the DAPI/DNA complex. The number of optical sections for each field was adjusted so that the respective z‐stack spanned the whole thickness of the GCL and IPL. The scanning was conducted using the ZEN 2010 software suite. Average intensity projections of the stacks onto the retinal plane were obtained using the ZProjection tool of the ImageJ Software (https://imagej.net/; RRID:SCR_003070).

The scanned 425 × 425 µm frames were split into smaller (106.25 x 106.25 μm) frames. In each scanned frame, the number of cells was counted using the CellProfiler software (https://cellprofiler.org; RRID:SCR_007358) (Bray et al., 2015). To avoid double‐counting and make the counts eligible for subsequent stereological analysis (see below), only those cells were counted that were entirely within the counting frame or intersected the acceptance line (left and bottom frame boundaries) without touching the rejection line (right and top frame boundaries) (Gundersen, 1986). The counts were corrected manually to ensure that occasionally overlapping nuclei were correctly recognized and counted. Cell‐size retinal debris, etc., misidentified and counted as cells was manually excluded from the counts. The debris was identified by its irregular, bizarre shape and fluorescence level exceeding considerably that of neighboring cell nuclei. Further adjustments were made including accounting for retinal shrinkage (Stone, 1981) and the presence of amacrine and glial cells (see Section 2.4).

In a small (~6%) portion of frames, the number of cells could not be obtained as described above because of retinal damage. In these frames, it was estimated by averaging the numbers of cells in adjacent frames. Based on the counts, an isodensity map was rendered by hand for each retina by linking areas of similar cell density (Stone, 1981).

2.3. Stereology analysis and estimation of the total cell number

We used the optical fractionator method to estimate the total cell number in the GCL and IPL (Sterio, 1984):

$$\mathrm{N}=\sum Q\times \left(t/h\right)\times \left(1/\text{asf}\right)\times \left(1/\text{ssf}\right)$$

where N is the total cell number in the GCL and IPL; Q, number of cells in each counting frame; t, section cut thickness; h, counting frame height; asf, area sampling fraction; ssf, section sampling fraction. The entire retina was considered a section, so ssf was taken to be 1. Since the height of each frame (i.e., the thickness of the respective z stack) was adjusted to span the local thickness of the GCL and IPL (see section 2.2), the t/h ratio was also taken to be 1. The accuracy of the total cell number estimation was estimated using Schaeffer coefficient of error (SCE).

This estimator was used in many studies and validated elsewhere (Glaser & Wilson, 1998). An acceptable SCE value is expected to contribute little to the observed group variance; according to an often used "rule of thumb", the variance introduced by the stereological analysis should not account for more than a 50% of the observed group variance (Slomianka & West, 2005). The counting grid size was, therefore, adjusted to achieve an SCE value that would satisfy the inequality SCE²/CV² < 0.5, where CV² is the observed group variance (squared variation coefficient of total cell number estimates). The final counting grid size was 212.5 × 212.5 µm. The counting frame size was 106.25 × 106.25 µm. The counting frames were regularly spaced across the retina. The area sampling fraction was 0.25. It was the same for all eye specimens. The SCE value was 0.034 and accounted for 12% of the observed group variance. The total number of counting frames was 582, 631, and 914 for the 26.58, 27.64, and 35.16 mm² retinal wholemounts, respectively.

2.4. Estimation of the proportion of amacrine cells and glia in the ganglion cell and inner plexiform layers

The major drawback of DAPI staining is non‐selective labeling of cell nuclei, leaving no way to distinguish between ganglion cells and other cell types in the GCL and IPL, namely, displaced amacrine cells (DACs) and glia. We, therefore, used two Nissl‐stained Japanese smelt retinae to estimate the proportion of non‐ganglion cells in these layers. For Nissl staining, a fish was prepared and the retinae were isolated as described above. The retinae were then flat mounted on a gelatinized slide, dried, stained for 5 minutes in cold 0.5% cresyl violet, dehydrated, cleared, and mounted in Canada balsam. The wholemounts were examined using a Zeiss LSM 700 inverted scanning microscope with a Zeiss 63/1.4 (Oil) Plan‐Apochromat objective. Large, granular, irregularly shaped profiles were identified as ganglion cells; darkly stained small round profiles, as amacrine cells; and cigar‐shaped or filamentous cells, as glial cells (Collin, 1988; Collin & Pettigrew, 1988a) (Figure 2). In each retina, a total of twenty‐five 135 × 135 µm sites were sampled, five, from the ventral periphery; five, from the dorsal periphery; five, from nasal midperiphery; five, from art; and five, from adjacent region of increased GC density. In each retinal region, the exact location of counting sites was chosen at random. In each site, the proportion of DACs and glia was determined manually using the above‐mentioned criteria, and the mean proportions were estimated for each region.

Figure 2.

Light micrograph of the Nissl‐stained ganglion cell layer of Hypomesus nipponensis (nasal periphery). The image is composed of several focal planes to bring neighboring retinal fragments to focus. gc, ganglion cell; ac, amacrine cell; gl, glial cell. Scale bar =50 μm

2.5. Eye anatomy and computerized tomography

A fish was sacrificed as described above. The eyes were enucleated and fixed in 2.5% paraformaldehyde in 0.1 M PBS (pH 7.4) at 4°C for 6 h, washed in several changes in PBS, dehydrated through a series of ethanol solutions (10%, 30%, 50%, 70%), contrasted in 0.5% phosphomolybdic acid solution in 70% ethanol for 3 days, and scanned using a Bruker SkyScan 1272 high‐resolution micro‐CT scanner (Bruker, Belgium). Computerized tomographic slices in the axial plane of the eye were obtained with 0.05 mm increments. The slices were examined using CTVox software (Bruker, Belgium). The slice with the maximum lens projection was used to estimate the posterior nodal distance (PND, the distance from the posterior nodal point of the lens to the photoreceptor layer of the retina) (Lisney & Collin, 2008) (Figure 3). A 3D eye representation was rendered using CTVox software. To estimate the PND for different retinal regions, a series of tomographic slices were obtained, each passing through the nodal point of the lens and the point of interest in the retina. In the Japanese smelt, the lens is near‐spherical, so the center of the lens was taken as a nodal point. The PND measured for different retinal regions was similar varying within 2%. The tissue shrinkage during eye preparation for computerized tomography was estimated as follows. The vertical, horizontal, and axial diameters of four eyes obtained from two fish 6.1 and 6.6 cm long (see section 2.1) were measured using a Leica M80 B microscope. The eyes were then fixed, dehydrated, and contrasted as described above, and their diameters were measured again. The horizontal, vertical, and axial eye capsule shrinkage was 5.36 ± 0.55, 6.60 ± 0.89, and 8.68 ± 0.82% (mean ± SEM, N = 4), respectively. The calculated PND values were corrected for shrinkage. To be noted, in 3D eye representations, one can occasionally see the retina and choroid partially detached from the sclera, apparently, due to different shrinkage coefficients. However, in this case, the original position of the retina can be easily "seen" (reconstructed) and the PND can be correctly measured.

Figure 3.

Micro‐computerized tomography scan of the left eye of Hypomesus nipponensis in the equatorial projection with the maximum lens diameter. r, retina; pnd, posterior nodal distance (shown by an arrow). Scalebar =1 mm

2.6. Theoretical anatomical spatial resolution

The theoretical anatomical spatial resolution (TASR) was estimated based on the density of the cells in the GCL and IPL as described by Collin and Pettigrew (1989). The angle subtending 1 mm of the retina was obtained as the arc tangent of the reciprocal of the PND for the respective retinal region:

$$\alpha =arctan\left(1/\text{PND}\right)$$

The TASR was calculated as half the number of cells subtended by one degree of visual arc:

$$\text{TASR}=\text{sqrt}\left(\mathrm{D}\right)/2\alpha ,$$

where D is the density of GCs.

2.7. Statistical analysis

All statistics (means and standard errors of mean) were obtained using the Statsoft Statistica 6.0 package (STATISTICA; RRID:SCR_014213; URL: http://www.statsoft.com/Products/STATISTICA/Product‐Index).

3. RESULTS

3.1. Cell topography

The Japanese smelt has lateral eyes (Figure 1a). The optic nerve head is oval shaped or elongated. It is slightly displaced ventrotemporally from the retinal center, so that the nasal hemiretina is larger than the temporal one. The embryonic fissure is oriented ventronasally (Figure 1b).

The majority of DAPI‐stained nuclei in the GCL did not overlap or touch each other (Figure 4). The GCL was relatively thin even in areas of increased cell density (art; see below) spanning 6‐8 μm (Figure 5a,c). It was normally composed by a single layer of cells. In all retinal regions, nuclei were observed infrequently in the inner plexiform layer (Figure 5b,c).

Figure 4.

Fragments of different retinal regions of Hypomesus nipponensis with DAPI‐stained cell nuclei. Projections of the scans spanning the ganglion cell and inner plexiform layers are shown. a, c — area retinae temporalis. b, d — retinal midperiphery. The images in a and b are the same as in c and d, respectively, with white circles representing image fragments recognized as cell nuclei. Scale bar =50 μm

Figure 5.

A fragment of area retinae temporalis of Hypomesus nipponensis. a, b — projections of the scans spanning the ganglion cell layer (a) and inner plexiform layer (b). c — radial projection of the same fragment. The nucleus of a cell in the inner plexiform layer is shown with an arrowhead. GCL, ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer. Note DAPI‐stained cell nuclei. Scale bars: a, b = 50 μm; c = 25 μm

The proportion of DACs (as estimated from Nissl‐stained retinae) was 11.32 ± 0.85 and 13.48 ± 0.95% (hereinafter, mean ± SEM, N = 5) in art and ventral retinal periphery, respectively; the respective proportions of glial cells were 3.70 ± 0.18 and 2.57 ± 0.09%. Thus, 85‐86% of cells in the GCL and IPL are RGCs.

The data on the cell distribution are summarized in Table 1. The density of GCs (corrected by subtracting the 15.5% of non‐GCs in the GCL and IPL) varied across the retina (Figure 6). It was minimum (3990 and 2380 cells/mm² in the smaller and larger fish, respectively) in the dorsal and ventral periphery. It gradually increased centripetally and reached a maximum of 14275 cells/mm² (smaller fish) and 10960 cells/mm² (larger fish) in the temporal retinal quadrant, where a pronounced art was detected (Figure 6).

Figure 6.

Topographic maps of the distributions of ganglion cells in the ganglion cell and inner plexiform layers in three wholemounts obtained from different Hypomesus nipponensis individuals. D, dorsal; V, ventral; N, nasal, T, temporal. The optic disc area is striped

The total number of GCs varied from 177 × 10³ (smaller fish) to 212 × 10³ cells (larger fish).

3.2. Theoretical anatomical spatial resolution

The TASR was minimum in the ventral periphery (smaller fish, 1.46 cpd; larger fish, 1.26 cpd) and maximum in art (smaller fish, 2.83 cpd; larger fish, 2.75 cpd). These values correspond to a minimal separable angle of 0.68° and 0.78° (periphery) and 0.35° and 0.37° (art) in the smaller and larger fish, respectively.

4. DISCUSSION

4.1. Methodological considerations

Here, we used DAPI staining of cell nuclei to visualize and study GCs in the inner retina of the Japanese smelt. DAPI is a highly specific DNA stain widely used to label both live and fixed cells (Kapuscinski, 1995). The main advantage of DAPI staining is that the majority of stained nuclei are isolated from each other (as seen in their projection to the retinal plane). They can be easily identified against dark background facilitating their counting using automated algorithms such as those implemented in CellProfiler. The major drawback of using DAPI technique is non‐selective cell nuclei labeling leaving no way to distinguish between ganglion and non‐ganglion cells (DACs and glia). However, the proportion of non‐ganglion cells can be estimated otherwise, for example, from Nissl‐stained material (an approach used in the present study), and proper corrections can be made. To be noted, the only practicable way to selectively label GCs is their retrograde labeling or (at least in some species) immunochemical staining (Barnstable & Dräger, 1984; Collin & Pettigrew, 1988a). However, it has some inherent drawbacks; thus, none of a wide variety of retrograde tracers used thus far guarantees that all the GCs are labeled, and some of them may label non‐ganglion cells via gap junctions (Lanciego & Wouterlood, 2011; Marc et al., 2018; Vaney, 1991).

We counted cells all over the wholemount surface. This allowed us to examine the density of cells and render a respective topography map with due precision. We used stereological analysis to estimate the total number of cells in the GCL and IPL. This approach has become increasingly widespread in modern research of retinal neuron topography and retinal specializations (de Busseroles et al., 2014; Coimbra et al., 2017). Using it, therefore, allows for more consistent comparisons between different studies.

It is known that in some fish species, some GCs are displaced to the inner nuclear layer (INL). If it is the case in the Japanese smelt, then we missed these cells in the present analysis. Obviously, we could not include the INL to the present analysis as this would result in including orthotopic amacrine cells into the counts and a considerable bias in the GC density estimates. However, there is some evidence that most (if not all) GCs displaced to the INL in bony fishes belong to large‐field types (Pushchin, 2017; Pushchin & Karetin, 2014). In bony fishes, the density of large‐field GC types is very low (Cook & Becker, 1991; Cook et al., 1996; Pushchin & Kondrashev, 2003) suggesting that their proportion to all GCs is quite small. Therefore, omitting INL from the present analysis did not affect much of our GC density estimates.

4.2. Non‐ganglion cells in the inner retina of the Japanese smelt

The proportion of glia in the GCL and IPL of the Japanese smelt retina (2.57%‐3.70% depending on retinal region) is close to that in harlequin tusk fish, Lienardella fasciata, blue tusk fish, Choerodon albigena, and staghorn damselfish, Amblyglyphidodon curacao (about 5%; Collin & Pettigrew, 1988a), while that of DACs is less than in other fishes. Thus, in some deep‐sea teleosts, the proportion of DACs in the GCL exceeds 80% (Collin & Hoskins, 1997; Wagner et al., 1998); in several lanternfish species, it varies from 67 to 78% (de Busseroles et al., 2014); in the European hake, it ranges between 32 and 39% in juvenile fish to reach 50% in adults (Bozzano & Catalán, 2002); in three coral fish species, the average proportion of non‐ganglion cells is 24% (Collin & Pettigrew, 1988a); in the goldfish, Carassius auratus, a substantial portion of cells in the GCL are non‐ganglion cells (Mednick & Springer, 1988); in the Australian lungfish, Neoceratodus forsteri, up to 44% of the cells in the GCL represent putative amacrine cells (Bailes et al., 2006).

We found no sizeable variation in the proportion of DACs (as well as non‐ganglion cells as a whole) between art and retinal periphery in the Japanese smelt. Similar to this finding, in the European hake, the proportion of DACs was fairly uniform across the retina; however, it increased with age from 32‐39 to 50% (Bozzano & Catalán, 2002). On the contrary, other fishes displayed a substantial region‐dependent variation in the proportion of non‐ganglion cells in the GCL. Thus, in eight elasmobranch species, the proportion of DACs increased with eccentricity ranging between 0.4 and 12.3% in specialized retinal areas and between 8.2 and 48.1% in the periphery (Bozzano & Collin, 2000); in the Australian lungfish, it increased from 45 ± 6% in the central retina to 58 ± 12% in the periphery (Bailes et al., 2006); in deep‐sea lanternfishes, it varied considerably across the retina (de Busseroles et al., 2014); in three coral fishes, the proportion of non‐ganglion cells in the GCL increased from 8% in area retinae centralis to 34% in the periphery (Collin & Pettigrew, 1988a);

A possible reason for the above discrepancies is that distinguishing between ganglion and non‐ganglion cells in the GCL in Nissl‐stained retinae may be difficult and not very reliable (Collin & Pettigrew, 1988a; Mednick & Springer, 1988; Fritsch, Collin, & Michiels, 2017). Therefore, the proportions of DACs and glia could be biased, at least, in some species. Another reason is that the proportion of non‐ganglion elements in the same retinal region may change with age (Mednick & Springer, 1988; Mack et al., 2004; Shand et al., 2000). It is therefore possible that in larger Japanese smelts, the ganglion to non‐ganglion cell ratio in the GCL and IPL is somewhat different from the one registered here. In the Primorsky region (Russian Far East), adult Japanese smelts reach 12 cm (total length) and the age of 2+ (Parpura & Kolpakov, 2001). The fish used in this study had the age of 0+ – 1+ (Parpura & Kolpakov, 2001).

The functional role of DACs in fish remains largely unknown. Mack et al. suggested (2004) that they may exert strong influence on the output activity of ganglion cells because of their proximity. Putative glycinergic DACs were found in the goldfish (Yazulla & Studholme, 1990). Parvalbumin‐immunoreactive DACs were revealed in the wolf fish, Hoplias malabaricus, tench, Tinca tinca, and several lanternfish species (Bejarano‐Escobar et al., 2009; Bonci et al., 2006; de Busseroles et al., 2014). Putative cholinergic DACs were found in the zebrafish, Danio rerio, Pacific Coast dogfish, Squalus acanthias, and goldfish (Brandon, 1991; Paulsen et al., 2010; Tumosa & Stell, 1986). Cholecystokinin‐immunoreactive cells occur in the zebrafish (Guerrera et al., 2018). Putative GABAergic DACs were revealed in the goldfish, channel catfish, Ictalurus punctatus, and Atlantic salmon, Salmo salar (Brandon, 1985; Ekström & Anzelius, 1998; Pow et al., 1996; Yazulla et al., 1986). Putative glutamatergic DACs were found in the goldfish (Marc et al., 1990). Wagner and Behrens (1993) found dopaminergic interplexiform cells displaced to the GCL in the rainbow trout, Oncorhynchus mykiss. This neurochemical diversity suggests that there may be several types of DACs in teleosts, each serving a particular function.

4.3. Evolutionary and ecological implications

The maximum density of GCs displays a considerable variation among fishes (Bozzano & Collin, 2000; Collin, 1999; Garza‐Gisholt et al., 2018). Thus, in the channel catfish it is as low as 1.5 x 10³ cells/mm² (Dunn‐Meynell & Sharma, 1987), but may reach 50 × 10³ cells/mm² or even more in some coral reef fishes (Collin & Pettigrew, 1988c). The peak density of GCs in the Japanese smelt (10.96‐14.28 × 10³ cells/mm²) falls within this range. It is rather low and comparable to those in predatory freshwater fishes of similar size (e.g., roach (6.3 × 10³ cells/mm²) (Zaunreiter et al., 1991) and goldfish (5.96 × 10³ cells/mm²) (Mednick & Springer, 1988).

The total number of GCs and maximum density depend on various factors. A species’ reliance on visual sensitivity or spatial resolution is one of them (Coimbra et al., 2016; Lisney et al., 2012). These aspects are discussed below. However, eye size and age also play a role. A larger eye requires a larger number of GCs for properly representing the visual environment (Bousfield & Pessoa, 1980; Johns & Easter, 1977); retinal specializations may change their size, shape, and prominence (with respective changes in total GC number and density) with age reflecting the changes in a particular species’ visual ecology (Bozzano & Catalán, 2002; Shand et al., 2000). In agreement with this, the peak density estimates obtained for smelts of different total length and apparent age were quite different (Table 1).

Proceeding from simple trigonometric ratios, with the TASR of 2.75‐2.83 cpd, the Japanese smelt could resolve a 1 mm sized prey or algae at a distance of 31‐32 cm. The bulk of the species’ food is about 1 mm or more in size (Riede, 2004; Zhou et al., 2013). To be noted, the actual spatial resolution at prey detection may be lower for two reasons. First, only a subset of GCs are obviously involved in this task, and their density is lower than that of all GCs (see above). Second, the spatial resolution depends on the environmental conditions, and the above estimates are obtained for the perfect conditions (sufficient light intensity, low water turbidity, perfect contrast, etc.), which rarely occur. Nonetheless, even if the actual spatial resolution is several times less that the present estimates, it seems to be sufficient for efficient food detection. The present findings are consistent with some basic relationships among VA, eye size, and visual ecology revealed in ray‐finned fishes. Eye size imposes one of the major constraints on VA (Caves et al., 2017). Thus, larger fishes with larger eyes have on average higher VA than smaller ones with smaller eyes. The habitat type and complexity also influence VA: fishes living in complex habitats (e.g., coral reefs) and featureless environment (pelagic realm) have significantly higher VA as compared to those dwelling in horizon‐dominated habitats (Caves et al., 2017). In agreement with these relationships, the Japanese smelt is a small‐sized benthopelagic predator with proportionally small eyes. It dwells in shallow waters with a prominent water‐bottom horizon, be it coastal waters or lakes. Its TASR is rather low (2.75‐2.83 cpd) and is comparable to that in other pelagic or benthopelagic fishes, diurnal predators of comparable size (e.g., goldfish (1.5‐2 cpd (behavioral acuity); Wilkinson, 1972; Neumeyer, 2003); common minnow (Phoxinus phoxinus; 5.45 cpd (both TASR and behavioral acuity); Brünner, 1934); and yellow dottyback (Pseudochromis fuscus; 3.2‐3.6 cpd (TASR) and 1.69 cpd (behavioral acuity); Parker et al., 2017). There is another possible factor affecting the VA in the Japanese smelt. A high VA is not always beneficial to predatory fishes. A prey item moving away from a predator may fade away before it becomes too small to recognize, especially, in turbid water (Johnsen, 2012). At a fixed eye size, increasing VA leads to reducing sensitivity and vice versa (Land, 1990). The Japanese smelt's habitats include near‐shore water of variable turbidity. It may be, therefore, that the optimum trade‐off between VA and sensitivity in this species imposes additional constraint on VA.

We found a prominent art in the Japanese smelt. Areae retinae temporales have been described in many pelagic, epi‐, and benthopelagic fish species (Collin, 1999, 2008; Peel, Collin, & Hart, 2020). They provide an increased acuity in the frontal visual field and facilitate binocular vision in predatory species or ones living in af structurally complex environment (Collin, 1999; Hughes, 1977). They may subserve numerous functions related to prey location, predator escape, and social behavior (Collin, 1999; Collin & Pettigrew, 1988b,c). In the Japanese smelt, art is located in the mid‐temporal retina, somewhat ventral to the equator. It may, therefore, partly subtend the dorso‐lateral (monocular) and upper frontal view of the visual field including a portion of the binocular zone as the Japanese smelt’ eyes are highly mobile in the frontal and transverse planes (SLK, IP, unpublished observations). In fishes with lateral eyes, the lens moves parallel to the plane of the pupil (Fernald & Wright, 1985). In the relaxed eye, the lens is situated closer to the nasal pole so that the temporal retina is myopic and the nasal retina, hypermetropic. This allows a fish to effectively detect prey in the frontal visual field using art, while tracing potential predators and conspecifics in other parts of the visual field. In agreement with this, the Japanese smelt is an active schooling predator hunting for prey as diverse as cladocerans, rotifers, copepods, benthic animals, surface insects, and amphipods (Riede, 2004; Zhou et al., 2013). The bulk of its diet is plankton and surface food. It is fed upon by many larger predatory fishes (Zhou et al., 2013).

According to the terrain theory (Hughes, 1977), the Japanese smelt might have been expected to have a visual streak as it lives in relatively open space with a water‐bottom horizon. However, we found no prominent visual streak in this species. There may be several explanations for this. First, the Japanese smelt is a schooling species, and the presence of numerous conspecifics masks the horizon and makes the visual environment more disruptive and complex. Second, as was mentioned above, the ventral location of art in this species suggests that it focuses more on the upper frontal view of the visual field rather than on the lower one comprising the water‐bottom horizon. Third, this may be an inherited characteristic, which does not necessarily represent an adaptive condition of a particular species’ lifestyle (Stone, 1983). The GC topography in the Japanese smelt may be, therefore, driven by both ecology and phylogeny.

In conclusion, we would like to stress the importance of research on the relationship between the characteristics of a fish species’ visual system, on the one hand, and its general anatomy, physiology, and ecology, on the other. Such studies allow us, first, to better understand factors driving the evolution and adaptation of the fish visual systems and, second, to quantify the relative contribution of ancestry and environment to the design of the visual system(s) in a particular species or a group of species.

CONFLICT OF INTEREST

The authors declare no conflict of interests.

AUTHOR CONTRIBUTIONS

All authors had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: IP and SK. Acquisition of data: IP, SK, and YK. Analysis and interpretation of data: IP, SK, and YK. Drafting of the manuscript: IP and SK. Critical revision of the manuscript for important intellectual content: IP and SK. Statistical analysis: IP. Obtained funding: IP. Administrative, technical, and material support: IP, SK, and YK. Study supervision: IP.

Pushchin I, Kondrashev S, Kamenev Y. Retinal ganglion cell topography and spatial resolution in the Japanese smelt Hypomesus nipponensis (McAllister, 1963). J. Anat. 2021;238:905–916. 10.1111/joa.13346

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.