Snakes (n = 33) were provided by the Butantan Institute, São Paulo, Brazil, and were euthanized with a lethal injection of sodium thiopental (100 mg/kg). Animal procedures were in accordance with ethical principles of animal management and experimentation established by the Brazilian Animal Experiment College (COBEA). The project was approved by the Ethics Committee of Animal Research of the Psychology Institute, University of São Paulo, Brazil. Individuals were classified as adults or juveniles based on the snout-vent length (SVL) and body mass (Supplementary Table 1; Sazima and Abe, 1991; Almeida-Santos, 2005; de Moraes, 2008; Barros, 2011; Fiorillo et al., 2020a).

Following euthanasia, the eyes were enucleated, their axial lengths were measured, and a small radial incision was made in the dorsal region for later orientation. The corneas were removed and the eyecups were fixed in 4% paraformaldehyde (PFA) diluted in phosphate buffer (PB) 0.1 M, for 3 h. The retinas were carefully dissected and maintained in PB 0.1 M at 4°C. When needed, the retinas were bleached in 10% hydrogen peroxide diluted in PB 0.1 M, for 24–48 h, at room temperature, prior to immunohistochemistry or Nissl procedures.

For immunohistochemistry, free-floating retinas were preincubated in 10% normal goat serum (Jackson ImmunoResearch, West Grove, United States) or 10% normal donkey serum (Sigma-Aldrich, St. Louis, MO, EUA), diluted in PB 0.1 M with 0.3% Triton X-100, for 1 h, at room temperature. The retinas were incubated with primary antibodies (Table 1) diluted in PB 0.1 M with 0.3% Triton X-100, for 3 days, at 4°C. The following antibodies were used: rabbit anti-SWS1 opsin (Sigma-Aldrich, AB5407; 1:200), rabbit anti-LWS opsin (Sigma-Aldrich, AB5405; 1:200), and for double immunofluorescence labeling, a mixture of the antibodies goat anti-SWS1 opsin (Santa Cruz Biotechnology, sc-14363; 1:200) and rabbit anti-LWS opsin (Sigma-Aldrich; AB5407; 1:200). The retinas were washed in PB 0.1 M with 0.3% Triton X-100 and incubated for 2 h with the secondary antibodies, at room temperature, protected from light: tetramethylrhodamine (TRITC)-conjugated goat anti-rabbit (immunoglobulin G, whole molecules; Jackson Immunoresearch Laboratories; 1:200), and for double-labeled retinas, a combination of TRITC-conjugated donkey anti-goat with Alexa Fluor^® 488-conjugated donkey anti-rabbit (immunoglobulin G, whole molecule; Jackson Immunoresearch Laboratories; 1:200). The retinas were rinsed in 0.1 M PB, flat-mounted on glass slides with the photoreceptor layer facing up, mounted with Vectashield (Vector Laboratories Inc. California, United States), and observed under a fluorescent microscope (Leica DM5500B), with a set of filters for TRITC (excitation green, emission red) and Alexa Fluor^® 488 (excitation blue, emission green).

Immunohistochemistry procedures were performed with polyclonal antibodies raised in rabbits against the last 42 amino acids of the C-terminal of human blue opsin (Sigma-Aldrich; AB5407), against the last 38 amino acids of the C-terminal of human red/green opsins (Sigma-Aldrich; AB5405), or raised in goats against a synthetic peptide with 20 amino acids of human blue opsin (Santa Cruz Biotechnology; sc-14363) (Table 1). The specificity of the antibodies for snakes was described previously (Hauzman et al., 2014, 2017; Bittencourt et al., 2019; Gower et al., 2019). Double labeling with the antibodies against SWS1 and against LWS opsins showed differential labeling of distinct photoreceptor populations (Figure 2), further indicating the specificity of both antibodies for particular types of cones. We also assessed the specificity of the two anti-SWS1 antibodies, by incubating 12 μm retinal sections of B. jararaca and C. durissus obtained at −25°C with a cryostat (Leica, CM1100; Nussloch, Germany), with a mixture of both antibodies, rabbit anti-SWS1 (AB5407; 1:200) and goat anti-SWS1 (sc14363; 1:200). Immunofluorescence visualization showed a small number of small single cones labeled by both antibodies (data not shown), indicating the specificity of both antibodies against SWS1 cones.

We used Nissl staining technique in whole-mounted retinas to analyze the population of GCL cells. Small radial incisions were made in the dissected retinas to allow them to be flat-mounted onto gelatinized glass slides, with the GCL facing up. The retinas were incubated with paraformaldehyde vapors overnight, at room temperature, for enhancing the adherence to the slide and to increase the differentiation of ganglion cells during staining (Coimbra et al., 2006). The tissues were rehydrated in decreasing ethanol series (95, 70, 50%), rinsed in distilled water acidified with glacial acetic acid, stained in aqueous solution of 2% cresyl violet for approximately 1 min, dehydrated in increasing concentrations of ethanol, cleared in xylene, and coverslipped with DPX (Sigma-Aldrich. St. Louis, MO, EUA).

To analyze the density and distribution of GCL cells, we used the cytological criteria proposed by Wong (1989) to distinguish ganglion cells from amacrine and glial cells. The ganglion cells were differentiated by their larger polygonal soma, abundant Nissl substance in the cytoplasm, and a prominent nucleolus (Figure 3; Hart et al., 2012). Glial cells were differentiated by smaller soma size and a round and darkly stained profile, and amacrine cells were identified by a smaller and circular profile with a more darkly stained nucleus compared to ganglion cells (Hart et al., 2012). The glial and amacrine cells were not counted.

The density and distribution of retinal neurons were accessed using a stereological approach based on the optical fractionator method (West et al., 1991), modified for retinal whole-mounts (Coimbra et al., 2009, 2012), using a motorized microscope (DM5500B, Leica Microsystems, Germany), connected to a computer running the Stereo Investigator software (MicroBrightField, Colchester, VT). The coordinates of the outer edges of the retinas were obtained using a 5x/NA 0.15 objective. Approximately 200 counting frames were positioned in a random fashion covering the entire area of the retina. Cells were counted when laying entirely within the counting frame or when intersected the acceptance lines, without touching the rejection lines (Gundersen, 1997). The area of the counting frame and the sampling grids varied according to the cell types and the retinal area, and were defined in pilot experiments, in order to obtain an acceptable Scheaffer coefficient of error (CE) (<0.10) (Scheaffer et al., 1996). The stereological parameters used to estimate the number of photoreceptors and GCL cells of each retina are described in Supplementary Tables 2, 3. To estimate the total population of neurons (N_total), we considered the area of sampling fraction (asf) according to the algorithm: N_total = ∑Q×1/asf, where ∑Q is the sum of the total number of neurons counted and the area of sampling fraction is the ratio between the counting frame and the sampling grid (Coimbra et al., 2009).

We estimated the upper limits of spatial resolving power based on the peak density of GCL cells and the presumed focal distance of the eyes. The focal lengths were estimated by freezing and sectioning eyes from one adult and one juvenile individual of each species. The fresh eyes were enucleated and rapidly frozen, embedded in Tissue-Tek OCT compound (Sakura Finetechnical Co., Tokyo, Japan), and the blocks were sectioned at −25°C on a cryostat (Leica CM1100; Nussloch, Germany). Photographs of the blocks were taken every 12 μm, with a camera (Axio CamMR, Carl ZeissVision, Germany) coupled to a stereomicroscope (SMZ775-T, NIKON, Japan), and a computed running the Axio Vision 4.1 software (Carl Zeiss, Germany). The areas of the lens and eyes were measured using the ImageJ software (NIH, Bethesda, United States) and the photographs that had the larger lens diameters were identified and used for optical measurements. Measurements of the axial length of the eye, lens axial diameter, and posterior nodal distance (PND) that represents the focal length and corresponds to the distance from the center of the lens to the retina-choroid border were taken along the optical axis, which was located by connecting the geometric centers of the optical components (Lisney and Collin, 2008). In B. jararaca, the focal length corresponded to 50% of the adult eye axial length of the adult (Figure 4A) and 67% of the juvenile. In C. durissus the focal distance corresponded to 60% of eyes axial length of both adult (Figure 4B) and juvenile. Those values were used to estimate the focal distance of the eyes used for GCL cell counts.

We estimated the upper limits of spatial resolution considering two possible arrangements of the ganglion cells, a hexagonal and a square array. The distance d subtended as 1 degree on the retina was determined from the PND and the equation d = (2πPND)/360 (Pettigrew et al., 1998). Considering a hexagonal array, we estimated the average spacing between cells (S) using the formula S2=2/(D3), where D is the peak density of GCL cells in cells/mm². The maximum spatial frequency (v) (Nyquist) of a sinusoidal grating (Snyder and Miller, 1977) that has resolution with this cellular arrangement was calculated as v=1/S3. This value was multiplied by the distance d, to obtain the spatial resolution in cycles per degree (cpd). In a second approach, considering that the ganglion cells might be organized in a square distribution, we estimated the linear density of GCL cells (cells/mm²) from the square root of the peak density (D), and divided the linear cell density by 2 (because at least 2 cells are required to detect 1 cycle of a given spatial frequency). The resulting value was multiplied by the distance d to obtain the visual acuity in cpd (Pettigrew et al., 1998; Coimbra et al., 2013).

To infer the behavioral significance of the estimated visual acuity in an ecological context, we predicted the minimum size of objects that the viperid snakes can spatially resolve (Coimbra et al., 2017). To do so, we estimated the angular distance in the retina that corresponds to one cycle by calculating the inverse of the spatial resolving power (cycles/degree). This value was divided by 2 to obtain the minimum angle of resolution (MAR), which represents the angular distance of the smallest resolvable detail on the retina. Subsequently, using the trigonometric relationship between the MAR and a presumed distance (D), relevant, for instance, for foraging a prey or for predator detection, we estimated the minimum object size (obj), according to the equation: D = obj/tanMar. According to the Nyquist sampling theorem, because an object needs to be twice the threshold to be spatially resolved, we multiplied the minimum object size by 2 (Marshall, 2000).

Statistical analyzes were performed with R (version 4.0.2)¹ and the RStudio software (1.3.959), to compare the total population and mean densities of retinal neurons among the four sampled groups, adults and juveniles of B. jararaca and of C. durissus, as well as the retinal area, eyes axial length and visual acuity. The normality of the distribution of values in each group was checked using the Shapiro-Wilk test, and homogeneities among groups were analyzed using the Levene test. The non-parametric test of Mann-Whitney for independent samples was applied for comparisons, even when values had normal distribution, due to the low sampling size in each group. Differences were considered significant when p < 0.05.

The total population of photoreceptors was estimated from the whole-mounted retinas of B. jararaca (n = 7) and C. durissus (n = 8). Adults and juveniles of both species had similar average total photoreceptors values, ranging from 3,092,787 ± 335,242 (median: 3,084,595) to 3,527,539 ± 134,480 (median: 3,585,315) cells in juveniles and adults of C. durissus, respectively (Figure 5 and Table 2). The mean density of photoreceptors, however, was higher in juveniles of both species (137,354 ± 28,330 cells/mm²; median: 128,376 cells/mm² in B. jararaca and 132,071 ± 13,230 cells/mm²; median: 125,533 cells/mm² in C. durissus) compared to adults (73,581 ± 16,205 cells/mm²; median: 74,724 cells/mm² in B. jararaca and 76,488 ± 10,400 cells/mm²; median: 75,540 cells/mm² in C. durissus) (Figure 5 and Table 2).

Rods were predominant and accounted for approximately 90% of the photoreceptors in retinas of both species (Table 2). The total population of rods was similar between adults and juveniles, varying from 2,779,855 ± 291,932 (median: 2,751,033) cells in juveniles of C. durissus and 3,182,738 ± 126,107 (median: 3,249,205) cells in adults (Figure 5 and Table 2). The average density of rods was higher in juveniles (128,528 ± 26,347 cells/mm²; median: 119,446 cells/mm² in B. jararaca and 118,070 ± 12,448 cells/mm²; median: 111,759 cells/mm² in C. durissus) compared to adults (69,488 ± 14,290 cells/mm²; median: 69,425 cells/mm² in B. jararaca and 68,844 ± 9,297 cells/mm²; median: 68,248 cells/mm² in C. durissus) (Figure 5 and Table 2).

The total population of cones represented about 7% of the photoreceptors in B. jararaca, and about 10% in C. durissus (Table 2), and was similar between adults and juveniles, varying from 233,400 ± 22,82 (median: 230,245) cells in juveniles of B. jararaca and 355,582 ± 15,403 (median: 359,130) cells in adults of C. durissus (Figure 5 and Table 2). The average density of cones was higher in juveniles (10,113 ± 1,044 cells/mm²; median: 10,559 cells/mm² in B. jararaca and 14,471 ± 695 cells/mm²; median: 13,947 cells/mm² in C. durissus) compared to adults (4,874 ± 765 cells/mm²; median: 4,993 cells/mm² in B. jararaca and 7,704 ± 1,132 cells/mm²; median: 7,654 cells/mm² in C. durissus) (Figure 5 and Table 2). The mean density of cones was significantly higher in adults of C. durissus compared with adults of B. jararaca (Figure 5).

The population of SWS1 cones accounted for approximately 10% of the cones in retinas of both species (Table 3). The total population of SWS1 cones was similar between adults and juveniles, and lower in B. jararaca (21,655 ± 0,466 cells; adults median: 17,529 cells; juveniles median: 23,007 cells) compared with C. durissus (33,354 ± 1,366 cells; adults median: 33,563 cells; juveniles median: 30,126 cells) (Figure 6 and Table 3). The average density was higher in juveniles (960 ± 242 cells/mm²; median: 943 cells/mm² in B. jararaca and 1,378 ± 195 cells/mm²; median: 1,340 cells/mm² in C. durissus) compared with adults (414 ± 123 cells/mm²; median: 364 cells/mm² in B. jararaca and 765 ± 83 cells/mm²; median: 803 cells/mm² in C. durissus) (Figure 6 and Table 3).

The LWS cones accounted for approximately 85% of the cones (Table 3). The total population of LWS cones was similar between adults and juveniles of both species, and lower in B. jararaca (190,022 ± 6,927 cells; adults median: 216,110 cells; juveniles median: 194,195 cells) compared with C. durissus (292,585 ± 17,63 cells; adults median: 301,990 cells; juveniles median: 269,625 cells) (Figure 6 and Table 3). In both species the mean density of LWS cones was higher in juveniles (8,466 ± 814 cells/mm²; median: 8,159 cells/mm² in B. jararaca and 11,628 ± 1,207 cells/mm²; median: 12,037 cells/mm² in C. durissus) compared with adults (4,131 ± 952 cells/mm²; median: 4,062 cells/mm² in B. jararaca and 6,683 ± 623 cells/mm²; median: 6,608 cells/mm² in C. durissus) (Figure 6 and Table 3).

The distribution of photoreceptors showed differences between and within species. In adults of B. jararaca, rods and cones were distributed in poorly defined horizontal streaks (Figure 7 and Supplementary Figures 1, 2). In juveniles, rods were concentrated in an anisotropic area centralis in the dorsal retina, while cones were concentrated in the ventral retina (Figure 7 and Supplementary Figures 1, 2). Mean density values were estimated from retinal sectors (dorsal, ventral, temporal, and nasal). In adults and juveniles of B. jararaca, higher densities of rods were located in the temporal retina. Higher density of cones were located in the ventral retina of juveniles and in the temporal region in adults (Supplementary Figure 3 and Supplementary Table 4). In C. durissus, the isodensity maps of rods and cones were similar between adults and juveniles. Rods were concentrated in the dorsal retina in an anisotropic area centralis, and cones were organized in poorly defined visual streaks (Figure 7 and Supplementary Figures 1, 2). Density estimates of retinal sectors indicated higher density of rods in the temporal retina of juveniles and in the dorsal retina of adults. Cones had higher densities in the temporal region in adults and in the nasal region in juveniles (Supplementary Figure 3 and Supplementary Table 4). In both species, the SWS1 cones showed diffuse distributions, with peak densities located in the central or ventral retina (Figure 8 and Supplementary Figure 4). The distribution of LWS cones in the retinas of adults and juveniles of both species was similar to that observed for the distribution of total cones, with poorly defined visual streaks in adults of both species and in juveniles of C. durissus, and a higher concentration of LWS cones in the central and ventral retina of juveniles of B. jararaca (Figure 8 and Supplementary Figure 5).

The population of GCL cells was estimated from 11 retinas of B. jararaca (adults: n = 4, juveniles: n = 7) and 9 retinas of C. durissus (adults: n = 4, juveniles: n = 5) (Table 4). The total population of GCL cells was similar between adults and juveniles of both species, ranging from 169,569 ± 13,730 (median: 163,733) cells in juveniles of B. jararaca and 215,141 ± 22,291 (median: 221,798) cells in adults of C. durissus (Figure 9 and Table 4). The average density of GCL cells was significantly higher in retinas of juveniles (9,871 ± 900 cells/mm²; median: 9,496 cells/mm² in B. jararaca and 9,774 ± 1,078 cells/mm²; median: 9,951 cells/mm² in C. durissus) compared with adults (4,880 ± 1,092 cells/mm²; median: 4,657 cells/mm² in B. jararaca and 4,429 ± 448 cells/mm²; median: 4,517 cells/mm² in C. durissus) (Figure 9 and Table 4). In the four groups analyzed, the isodensity maps of GCL cells did not show defined distribution patterns (Figure 10 and Supplementary Figure 6). In retinas of adults and juveniles of C. durissus and adults of B. jararaca, we observed diffuse distributions and peak density of cells in the temporal retina (Figure 10 and Supplementary Figure 6). In juveniles of B. jararaca the GCL cells were concentrated in the ventral retina with a decreasing ventral-dorsal gradient and peak density located in the ventral area (Figure 10 and Supplementary Figure 6). Density estimates in retinal sectors showed higher values in the temporal retina of adults and juveniles of C. durissus and adults of B. jararaca (Supplementary Figure 3 and Supplementary Table 5). In juveniles of B. jararaca, higher densities were located in the ventral retinas (Supplementary Figure 3 and Supplementary Table 5), in agreement with the observed isodensity maps (Figure 10 and Supplementary Figure 6).

The theoretical upper limits of spatial resolving power were estimated based on the ganglion cell peak densities using two approaches, one considering that the ganglion cells are organized in a hexagonal array and the other considering a square array. Adults and juveniles of B. jararaca had similar estimated visual acuity values, with 2.2 ± 0.2 cycles per degree (cpd) (median: 2.3 cpd) and 2.4 ± 0.2 cpd (median: 2.3 cpd), respectively (Figure 9 and Table 4). In C. durissus the estimated visual acuity was significantly higher in adults, with 2.8 ± 0.2 cpd (median: 2.9 cpd), compared to juveniles, with 2.3 ± 0.1 cpd (median: 2.3 cpd) (Figure 9 and Table 4). The estimated acuity values of adults of C. durissus were also significantly higher than of adults of B. jararaca (Figure 9 and Table 4).

From the estimated visual acuity values we calculated the minimum angle of resolution. The estimated distance relevant for predatory behavior was calculated considering the tangent of the minimum angle of resolution and minimum target size. For species with the lowest (2.2 cpd) and highest (2.8 cpd) spatial resolving power, we estimated a minimum angle of resolution of 0.45° and 0.36°. In an ecological context, snakes with lower estimated spatial resolution can observe an object with a minimum size of 10 cm, such as a small mammal, at a distance of approximately 13 m. Snakes with the higher presumed spatial resolution might be able to observe an object of the same size at a distance of about 17 m. These estimates indicate that at these predicted distances objects larger than the minimum target size can be spatially detected.

The retinas of B. jararaca and C. durissus had a high proportion of rods (about 90% of the photoreceptors) (Figure 5 and Table 2), which indicates high sensitivity to light, in agreement with their nocturnal or crepuscular activity pattern (Martins et al., 2001; Fiorillo et al., 2020b). Previous studies described the predominance of rods in retinas of viperid snakes based on analysis of retinal sections (Walls, 1934, 1942; Underwood, 1967b; Bittencourt et al., 2019; Gower et al., 2019) and fragments of flat-mounted retinas (Gower et al., 2019). Snakes stand out among vertebrates by their highly variable patterns of photoreceptor morphology (Walls, 1942). In the Caenophidia group (“advanced” snakes), nocturnal species from different families, including viperids, have four types of photoreceptors, with a unique type of double cone that differs from double cones of other vertebrates, with a large principal member and an extremely slender and attached accessory member (Walls, 1942). On the other hand, diurnal caenophidian snakes have “pure-cone” retinas, with the absence of typical rods, the presence of a transmuted cone-like rod, and lower photoreceptor density compared to nocturnal species (Walls, 1942; Schott et al., 2016; Hauzman et al., 2017; Hauzman, 2020).

The comparison of the populations of cones and rods in B. jararaca and C. durissus showed a higher proportion of cones in retinas of C. durissus (10% in juveniles and adults) compared to B. jararaca (7.5% in juveniles and 6.5% in adults) (Figure 5 and Table 2). This difference, although subtle, might reflect functional differences in temporal and spatial resolution between both species, which agree with the habitat occupied. C. durissus inhabit predominantly open areas of the Cerrado, with higher incidence of light (Miranda et al., 1997) compared to the closed forested areas predominantly occupied by B. jararaca (McWilliam et al., 1993; de Paula and Lemos Filho, 2001). In both species, the population of cones is dominated by large single cones and double cones that contain the LWS photopigment (about 80–90%) (Figure 6 and Table 3), with spectral sensitivity peak (λ_max) predicted at 555 nm (Bittencourt et al., 2019). Small single cones with the SWS1 photopigment comprise about 10% of the cone population (Figure 6 and Table 3), with λ_max predicted at the UV range (360–370 nm) in both species (Bittencourt et al., 2019). Gower et al. (2019) identified double cones with the SWS1 photopigment in retinas of two viperid snakes, Echis coloratus and C. durissus, a unique type of photoreceptor described for the first time in vertebrates. However, in our analysis of whole-mounted retinas, this type of cone was not identified, and all double cones observed were found to contain only the LWS photopigment (Figure 2).

The distribution of photoreceptors differed between C. durissus and B. jararaca and between juveniles and adults of B. jararaca (Figure 7). In C. durissus, the visual streak formed by cones might reflect a better panoramic view of the environment under photopic conditions (Figure 7). This specialization might benefit scanning of the environment while searching for shelters for body temperature control during the day, a frequent behavior of snakes that occupy open environments where they are subject to overheating (Tozetti and Martins, 2008). On the other hand, the higher density of rods in the dorsal retina (Figure 7) might improve light sensitivity in the lower visual field, possibly favoring foraging behavior and searching for rodents during twilight and at night. These patterns of distribution of cones and rods were found in both, juveniles and adults of C. durissus (Figures 7, 11).

In B. jararaca, ontogenetic changes in the niche occupied seem to be associated with plasticity of the retinal architecture. In juveniles, a higher density of cones was observed in the ventral retina (Figures 7, 11), a specialization that might provide higher acuity in the upper visual field. Oppositely, a higher density of rods in the dorsal retina might benefit the view of the lower visual field under scotopic conditions (Figure 7). Juveniles of B. jararaca occupy the arboreal stratum and use sit-and-wait and caudal luring as hunting strategies. Therefore, it is plausible to speculate that the difference in the distribution of cones and rods might be associated with the direction of predation pressure. Cones located in a ventral area might favor the view of aerial predators, such as diurnal birds approaching from above during the day (Figure 11) (Sazima, 1992; Costa et al., 2014). Higher concentration of rods in the dorsal retina might benefit the view of terrestrial predators, such as marsupials, approaching from below, during the night (Figure 11) (Emmons, 1990; Sazima, 1992; Jared et al., 1998; Oliveira and Santori, 1999). Compared to terrestrial snakes, arboreal species display a higher number of defensive tactics, which is likely associated with greater exposure to predators approaching from a variety of directions (Martins et al., 2008).

Adults of B. jararaca occupy terrestrial environments, lose the caudal luring, and actively forage for endothermic prey (Sazima, 1992, 2006). Their retinas have cones and rods distributed in poorly defined visual streaks, a specialization that benefits the panoramic view of the terrestrial forested stratum under photopic and scotopic conditions (Figures 7, 11). A similar distribution of cones was described in terrestrial, arboreal and semiaquatic colubrids (Hauzman, 2014; Hauzman et al., 2014), indicating that this specialization is widely observed in snakes and might contribute to active foraging behavior in different environments. As far as we are aware, this is the first description of a horizontal streak formed by rods in the retinas of snakes.

The distribution of LWS cones (single and double) was similar to the distribution of total cones in both species, as expected based on their high proportion (Figure 8). The distribution of SWS1 cones did not show a defined pattern of specialization. Higher densities of SWS1 cones were found in the ventral and central retina in adults and juveniles of both species (Figure 8), as described for diurnal colubrid snakes (Hauzman et al., 2014). This distribution might favor the view of potential aerial predators approaching from the upper visual field, as many bird species are important predators of snakes (Martins and Oliveira, 1998; Tozetti, 2006; Specht et al., 2008; Costa et al., 2014). Higher densities of UV cones were also described in the ventral retina of mammals (Szél et al., 1992, 1994; Famiglietti and Sharpe, 1995; Peichl et al., 2005; Huber et al., 2010). In mice, UV cones contribute to chromatic discrimination of the upper visual field (Szatko et al., 2020), and might be relevant for visualizing the silhouette of aerial predators against the blue sky background (Calderone and Jacobs, 1995; Szél et al., 1992).

The mean density of GCL cells was similar between B. jararaca and C. durissus. However, juveniles had higher average density values compared to adults (Figure 9 and Table 4). This difference can be attributed to the increase in the area of the retina. Juveniles have smaller retinas and thus, higher cell packaging. As the animals grow, the eye increases, such as the retinal area, but the cell population remains constant, resulting in lower density values. A higher GCL cell density in juveniles compared to adults was also described in fish (Hagedorn and Fernald, 1992; Bailes et al., 2006), birds (Straznicky and Chehade, 1987), amphibians (Nguyen and Straznicky, 1989), and mammals (Robinson et al., 1989). Considering there is no neuron loss or additional generation of cells in the GCL, we suggest the increase of retinal area followed by the decrease of cell density is associated with interstitial growth, as described for chicken retinas (Straznicky and Chehade, 1987). It is notable that in Bothrops jararaca these changes in cell density are associated with rearrangements of the cell distribution.

The distribution of GCL cells was similar between adults and juveniles of C. durissus, with no defined type of specialization, and with peak density of cells in the temporal retina (Figures 10, 11). This specialization might benefit the view of the frontal field and favor strike performance of endothermic prey. The same topographic pattern of GCL cells was found in retinas of adults of B. jararaca (Figures 10, 11). The visual information arriving in the midbrain, from the projections of this temporal specialization might be combined with infrared information from the loreal pit, a thermosensory organ of Crotalinae snakes located between the eyes and the nostrils (Noble and Schmidt, 1973). The nerve terminals of the loreal pit cells project to the optic tectum in the midbrain (Newman and Hartline, 1982), where the visual and thermosensory inputs are combined, allowing an integrated perception from ultra-violet to infrared wavelengths (Newman and Hartline, 1981; Moiseenkova et al., 2003; Goris, 2011), important in both scotopic and photopic conditions. During the night, thermal detection enables the perception of temperature changes. During the day, discrimination of infrared information may be compromised due to higher temperatures (Chen et al., 2012), yielding predominance of perceptual information to visual inputs.

In juveniles of B. jararaca the higher density of GCL cells in the ventral retina is in agreement with the distribution of cones (Figures 8, 10, 11), which indicates a higher convergence from cones to GCs in this region (from approximately 4:1 in the periphery, to ∼1:1 in the area centralis), enabling higher spatial resolution in the upper visual field. A similar distribution pattern of GCL cells, with higher density in the ventral retina was observed in the dipsadid snakes Dipsas albifrons and Sibynomorphus neuwiedi (Hauzman et al., 2018). Both species are nocturnal, semi-arboreal, and feed on ectothermic prey (goo-eaters) (Maia et al., 2011; Sazima and Muscat, 2016), such as juveniles of B. jararaca. Thus, we suggest that this type of specialization might be associated with the niche occupied by snakes, the direction of predatory threats, and particularities of hunting strategies, which do not involve active foraging of fast moving prey (Figure 11). Reorganization of the distribution of GCL cells according to ontogenetic changes in ecology and behavior was also described in frogs (Dunlop and Beazley, 1981) and in fish (Shand, 1997; Bozzano and Catalan, 2002; Miyazaki et al., 2011). In the fish Acanthopagrus butcheri, the retinas of juveniles have a temporal area centralis, which favors object detection in the frontal visual field and might be important for feeding on plankton in pelagic waters. On the other hand, adults occupy the benthic environment. This ecological switch is accompanied by a displacement of the area toward the dorsal or dorso-temporal retina, which might increase acuity in the lower-frontal visual field and benefit their hunting strategies for the capture of small fish, polychaete worms and detritus obtained from the benthos (Shand et al., 2000).

The estimated upper limits of spatial resolving power based on the peak cell density of GCL cells of the viperid snakes varied from 2 to 3 cpd (Table 4). These values were similar to those estimated for diurnal dipsadid snakes of the genus Philodryas spp. (Hauzman et al., 2014) and for the marine elapid Aipysurus laevis (Hart et al., 2012), with spatial resolution between 2.3 and 2.6 cpd, and were higher than the values estimated for nocturnal dipsadids, with about 1.3 cpd (Hauzman et al., 2018). Our comparisons showed slightly higher estimated visual acuity in adults of C. durissus compared to the other groups analyzed (Figure 9). We predicted that the estimated visual acuity of ∼2.3 cpd of adults and juveniles of B. jararaca and juveniles of C. durissus would allow the view of potential prey such as rodents, at a minimum size of 10 cm, at a distance of about 13 m. On the other hand, adults of C. durissus with estimated visual acuity at ∼2.8 cpd, might be able to detect the same minimum object size at a higher distance, about 17 m. These estimates also indicate that these snakes can have a high capability to observe potential predators such as birds of prey (Martins and Oliveira, 1998; Tozetti, 2006; Specht et al., 2008; Costa et al., 2014) at longer distances. However, behavioral analyzes are necessary to confirm if the estimated values of spatial resolution based on anatomical data are reached, and to ascertain whether the differences imply functional significance, relevant to the species ecology and visual behavior.