OFF bipolar cell density varies by subtype, eccentricity and along the dorsal ventral axis in the mouse retina

Abstract

The neural retina is organized along central-peripheral, dorsal-ventral and laminar planes. Cellular density and distributions vary along the central-peripheral and dorsal-ventral axis in species including primates, mice, fish and birds. Differential distribution of cell types within the retina is associated with sensitivity to different types of damage that underpin major retinal diseases, including macular degeneration and glaucoma. Normal variation in retinal distribution remains unreported for multiple cell types in widely used research models, including mouse. Here we map the distribution of all known OFF bipolar cell (BC) populations and horizontal cells. We report significant variation in the distribution of OFF BC populations and horizontal cells along the dorsal-ventral and central-peripheral axes of the retina. Distribution patterns are much more pronounced for some populations of OFF BC cells than others and may correspond to the cell type’s specialized functions.

Graphical Abstract:

Density of OFF bipolar cells vary by subtype, eccentricity and along the dorsal ventral axis. (a) Retinas were disected into four quadrants labeled with short-wave opsin to determine orientation. (b) Cells were marked and saved as series of individual coordinates which underwent Voronoi tessellation to analyze spatial heterogeneity. (c) Volumetric datasets were counted do determine cell count. (d) OFF-bipolar cell density contour maps by subtype 1a, 1b, 2, 3a, 3b, 4.

INTRODUCTION

The neural retina of many species is organized with regional variation in cellular distribution. In humans, the cone rich macula has a markedly different organization than the rod rich peripheral retina (Curcio & Sloan, 1992; Curcio, Sloan, Kalina, & Hendrickson, 1990; Elsner et al., 2017; Yamada, 1969). Long and medium wavelength cones in zebrafish have different densities based on retinal eccentricity (distance from the optic disc) and the dorsal-ventral axis (Branchek & Bremiller, 1984; Mackin et al., 2019; Takechi & Kawamura, 2005; Vihtelic, Doro, & Hyde, 1999). Other examples include a rod free zone in the chicken retina and the visual streak, a concentration of retinal ganglion cells (RGCs) along the horizontal field of predatory mammals’ retina (Bruhn & Cepko, 1996; Peichl, 1992). In mouse, the most pronounced difference in regional retinal organization is an enrichment of blue/green cones in the ventral retina and green cones in the dorsal retina (Haverkamp et al., 2005; Wassle, Puller, Muller, & Haverkamp, 2009). Other dorsal-ventral and central-peripheral gradients in cell distribution in the mouse retina have been reported, including a ventral enrichment in RGCs, both in general and for specific types of RGCs (Bleckert, Schwartz, Turner, Rieke, & Wong, 2014; Salinas-Navarro, Mayor-Torroglosa, et al., 2009; Yonehara et al., 2008; Yu et al., 2018).

A better understanding of cellular distribution is important to understand disease mechanisms because specific retinal domains have greater sensitivities to selected diseases (Bernstein & Wong, 1998; Paz & Anderson, 1992; Velez et al., 2018). This is also an important knowledge gap to address from a basic science perspective. For example, mapping cell density in different inbred mouse strains has identified many genes that regulate cellular differentiation and organization, and information on regional distribution will increase the power of such approaches (Kautzman et al., 2018; Keeley et al., 2014; Stincic, Keeley, Reese, & Taylor, 2018).

The goal of this study is to expand our understanding of retinal organization and regional variation. While the distribution of photoreceptors and some populations of retinal ganglion cells have been mapped, there is little information on the distribution of the bipolar cells (BCs) connecting photoreceptors and ganglion cells. Previous work has identified a peripheral to central gradient in bipolar cell distribution, indicating that bipolar cells may have regional variation like photoreceptors and ganglion cells (Jeon, Strettoi, & Masland, 1998). Here we aim to provide a baseline reference for OFF BCs and horizontal cells in the mouse retina. OFF BCs respond to light decrement and make flat synaptic contacts at the base of the cone terminal invaginations made by ON BC dendrites (Boycott & Hopkins, 1991; De Robertis & Franchi, 1956; Puller, Arbogast, Keeley, Reese, & Haverkamp, 2017; Puller, Ivanova, Euler, Haverkamp, & Schubert, 2013). We mapped the distribution six known populations of OFF BCs and horizontal cell interneurons (HCs) in the mouse retina with respect to retinal domain, eccentricity and mouse sex (Fox & Sanes, 2007; Haverkamp et al., 2008; Jeon et al., 1998; Mataruga, Kremmer, & Muller, 2007; Shekhar et al., 2016) (Figure 1). We report extensive dorsal-ventral and peripheral-central regional variation in OFF BC and HC density. We did not observe differences between nasal and temporal retinal regions, left and right eye, or in the distribution of bipolar cells in male and female mice.

FIGURE 1:

Horizontal and OFF bipolar cell populations with markers. Horizontal cells are labelled with calbindin (Calb). Type 1a/1b, 2, 3a, 3b, and 4 are labeled by distinct immunohistochemical markers/reporter: cyan fluorescent protein (CFP), Syt2, HCN4, PKARIIβ and calsenilin (Csen), respectively.

MATERIALS AND METHODS

Animal care

Mice were housed in the University of Idaho or University of Louisville vivariums on a 12-hour light/dark cycle and were fed ad libitum. Two strains of mice were used in this study: C57Bl/6J mice and the C57Bl/6 mice carrying the Thy1-mitoCFP-P transgene that expresses a CFP reporter in type 1 BCs and neither glycinergic nor GABAergic (nGnG) amacrine cells (Schubert, et al., 2008; Kay, Voinescu, Chu, & Sanes, 2011). Right and left eyes or the sex of mice provided by the U of L vivarium (Thy1-mitoCFP-P mice) were not marked and separately tracked. All mice were taken for study between 60 and 120 days of age.

Tissue preparation and immunofluorescence

Mice were anesthetized by an intraperitoneal injection with 1 mL 2.5% (w/v) 2,2,2-tribromoethanol in 1X phosphate buffered saline (PBS) (137 mM NaCl; 10 mM Na₂HPO₄; 1.8 mM KH₂PO₄, 2.7 mM KCl; 7.4 pH). Mice underwent transcardial perfusions with 1X PBS to prevent non-specific secondary antibody binding to blood vessels. Eyes were removed and an incision was made near the cornea to allow affective permeation of fixing agent. The eyes were fixed in 4.0% Paraformaldehyde (PFA) in 1X PBS for 50 minutes at 4° C. After fixation, the eyes were washed 3 times for 10 minutes each with 1X PBS. Eyes were hemisected (Figure 2a) and the retina was removed from the back of the eye cup and incubated in a blocking buffer (0.4% Triton, 7.5% normal donkey serum, 0.02% sodium azide diluted in 1X PBS) at 4° C for 4 hours. To identify eye orientation, we labeled blue cones with an antibody to short-wave opsin 1 (opnsw1) (Figure 2b)(Sondereker, Stabio, Jamil, Tarchick, & Renna, 2018). Primary antibodies (example Figure 2c) were added (see Antibodies section below) in blocking buffer and incubated at 4° C for 3 days. After incubation, retinas were washed 3 times with 1X PBS. Secondary antibodies were added at a concentration of 1:500 and incubated for 2 days at 4° C. Retinas were then washed again 3 times for 10 minutes with 1X PBS then mounted on glass slides. Each retina region was labeled respectively; dorsal, ventral, nasal, temporal. The dorsal-ventral axis of the retina was identified by the density of blue cones (Haverkamp et al., 2005). To prevent the tissue from being compressed by the cover slip, a small strip of electrical tape was placed on each side of the sample. Retinas were mounted in 80% glycerol.

FIGURE 2:

Retina dissection, labelling and sample analysis. (a) Each eye was hemisected around the edges of the cornea. The retinas were extracted and incised into quadrants for mounting. (b) The known density variance in blue cones comparing dorsal and ventral retina was used to consistently maintain the retina’s orientation when generating density profiles. (c) OFF-bipolar cell (BC) populations were each labelled with immunohistochemical markers. (d) Each retina was then sub-divided into ~24 regions of interests, comprising of ~6 sub-fields per respective quadrant. Each field has a corresponding relative cartesian coordinate, all using a reference origin at the optic nerve. Each field size used for approximating densities were ~ 300 μm × ~400 μm in size. (e) Image acquisition was taken in a volumetric fashion, each z-stack encompassing the OPL, INL and IPL with step sizes ranging between 0.9–1.2 μm. Each volume was assessed manually by looking at unique features of each respective cell type to verify accurate counting. (f) For each field’s volume, cell counts were saved as series of individual coordinates which underwent Voronoi tessellation to analyse spatial heterogeneity.

Antibodies

The following antibodies were used in this study: mouse (IgG2a) anti-Syt2 (type 2 BCs; Zebrafish International Resource Center; AB10013783; 1:50), rabbit anti-HCN4 (type 3a BCs; Alomone Labs; APC-052; 1:250), mouse (IgG1) anti-PKARIIβ (type 3b BCs; BD Transduction Laboratories; 610625 1:250), mouse (IgG1) anti-calsenilin (type 4 BCs; Millipore; 05–756; 1:500), goat anti-opnsw1 (blue cones; Santa Cruz Biotechnology; sc-14363; 1:500), mouse (IgG2b) anti-Chx10 (All BCs; Santa Cruz Biotechnology; sc-374151; 1:500), rabbit anti-calbindin D-28k (horizontal cells; Swant; CB-38a; 1:500), rabbit anti-cone arrestin (cones; Millipore; AB15282; 1:500), rabbit anti-PPP1R17 (nGnG amacrine cells; Atlas Antibodies; HPA047819), mouse (IgG2a) anti-GFP clone N86/38 (MitoP line CFP reporter amplification in type 1 BC, UC Davis/ NIH NeuroMab; P42212). Secondary Antibodies were acquired from Jackson Immuno-Research and used at a concentration of 1:500. Isotype specific secondary antibodies were used when possible. Specificity of all antibodies used in this study have been previously documented in accordance with journal policy (Andrade, Long, Fleming, Li, & Fuerst, 2014; Shekhar et al., 2016).

Antibody Specificity:

The specificity of antibodies to Syt2, HCN4, PKARIIβ, calsenilin, CHX10, calbindin, cone arrestin were described according to journal policy (Andrade et al., 2014). Verification of the following antibodies was confirmed by the following:

OPNSW1: SWS Mouse (Blue opsin).

The anti-SWS1 antibody was used in this study to label a population of UV opsin-positive photoreceptor cells. The epitope maps to human origin using the N-terminus of OPN1SW gene. Its specificity was verified by western blot with a molecular weight of 40 kDa (manufacture’s website). The staining pattern matches the well-known distribution pattern of blue cones in the mouse retina and fish retina (Sukeena et al., 2016)

PPP1R17:

This antibody was generated using the epitope corresponding to amino acids 15–72 of human PPP1R17. Antibody specificity was confirmed by western blot analysis, which identified a single band of the expected size and by immunohistochemistry, in which it labeled cells shown to express the gene (Shekhar et al., 2016).

GFP:

This antibody was generated using full length GFP protein (amino acids 1–238). Specificity was confirmed by specific labeling in multiple GFP transfected cells and transgenic tissue compared to controls and by comparing antibody specificity to native GFP fluorescence (Firl et al., 2015).

Imaging and image processing

Images were captured using a confocal Nikon Andor spinning disk microscope with a Zyla monochromatic camera. A Nikon 4X objective was used to take a montage composite image (4×4) of each whole retina. Each retina was sampled in ~24 locations (~6 per dorsal, ventral, nasal and temporal regions) (Figure 2d) using a Nikon 20X objective. Due to damaged incurred during the dissection process that made it impossible to image some retinal regions of a small number of retinas, field sample sizes between cell populations varied slightly. The number of retinas used from the Thy1-mitoCFP-P line were reduced by half because male/female and left/right comparisons were not made. One channel was dedicated to S-opsin for sampling blue cones with opnsw1. Cartesian coordinates were recorded for each field sample taken in their respective regions. A script (see: data visualization and analysis) was written with Geometer’s Sketchpad using trigonometric identities to normalize the x, y coordinates across all tissue samples to calculate a correction angle to align the dorsal and ventral axis of the retina, θ_adjust (“Geometer’s Sketchpad, Version 5.06,” 2012).

To perform cell counts, each image was counted manually with a marker tool and assessed volumetrically (Figure 2e) (Rueden et al., 2017). In order to use ImageJ with the native Nikon file format, the Bio-Formats plugin was used (Hiner, Rueden, & Eliceiri, 2016). Spatial analysis of respective cell to cell spacing arrangement was analyzed with Voronoi tessellation. Voronoi analysis was performed with Ka-me: Voronoi Image Analyzer 1.0 (Jungck, 2015) (Figure 2f). Border cells were excluded from analysis by the software. The coefficient of variance (CV) was used to assess the homogeneity of spatial arrangement of each BC sub-type. Each CV was derived by taking the standard deviation of the Voronoi domain area divided by the arithmetic mean of all domain areas in each respective 20X field (see Data visualization and analysis section below).

Data visualization and analysis

The Cartesian coordinates, x_initial and y_initial, from each 20X imaged field taken were normalized by calculating an angle θ_adjust. This angle was calculated by rotating each 4X montaged image labeled for s-opsin cones until the direction of the density gradient was aligned with the y-axis. From these parameters, x_final and y_final were calculated respectively from the following equations:

$$x_{\mathit{\text{final}}}=x_{\mathit{\text{initial}}}\text{cos}\left({\text{tan}}^{-1}\left({\theta }_{\mathit{\text{adjust}}}\right)\right)-y_{\mathit{\text{initial}}}\text{sin}\left({\text{tan}}^{-1}\left({\theta }_{\mathit{\text{adjust}}}\right)\right)$$

$$y_{\mathit{\text{final}}}=x_{\mathit{\text{initial}}}\text{cos}\left({\text{tan}}^{-1}\left({\theta }_{\mathit{\text{adjust}}}\right)\right)+y_{\mathit{\text{initial}}}\text{sin}\left({\text{tan}}^{-1}\left({\theta }_{\mathit{\text{adjust}}}\right)\right)$$

Density contour plots were generated from each paired x_final and y_final (position) and a z value (cell density) for each field, creating a composite map for each respective BC population from all retinal samples. OriginPro 2020 was used to generate these contour plots (“Origin(Pro), Version 2020,” 2020). In order to compute spatial heterogeneity within each cell population, the variance (v_area) was used to calculate the standard deviation (σ_area) and mean (μ_area). The standard deviation and mean were recorded for each field to formulate the CV_area from the following relationship:

$$v_{\mathit{\text{area}}}={\sigma }_{\mathit{\text{area}}}{}^2\therefore C{V}_{\mathit{\text{area}}}=\frac{{\sigma }_{\mathit{\text{area}}}}{{\mu }_{\mathit{\text{area}}}}=\frac{\sqrt{v_{\mathit{\text{area}}}}}{{\mu }_{\mathit{\text{area}}}}$$

One-way ANOVA with Tukey-Kramer post-hoc tests were used to analyze density differences between cell subtypes and comparisons between dorsal, ventral, nasal and temporal regions of the retina. Kruskal–Wallis post hoc tests were used to analyze divergence between cell subtype spatial heterogeneity. The male vs. female and left vs. right comparisons were made with the Student’s t-test. Linear regressions were performed on each cell subtype to look at the relationship between both cell density vs. spatial heterogeneity as well has eccentricity vs spatial heterogeneity using Pearson’s coefficient (ρ). Pearson ‘s correlation coefficient values of ± .0 90 ≤ ρ ≤ 1.0, 0.70 ≤ ρ < 0.90, 0.50 ≤ ρ < 0.70, 0.30 ≤ ρ < 0.50, 0.00 ≤ ρ < 0.30 were considered very high, high, moderate, weak, and negligible, respectively.

RESULTS

BCs make up approximately 41% of the cells in the INL, at ~30,000 cells/mm² (Wassle et al., 2009). We measured the density of BCs, all cones and horizonal cells (Figure 3a–c). We did not observe differences in bipolar cell or cone distribution across the retina (Figure 3d–e). Horizontal cell distributions were significantly different when comparing different retinal domains (Figure 3f), most prominently a decrease in the density of HCs in the dorsal retina. We also compared the density of BCs, cones and HCs with respect to retinal eccentricity from the central retina around the optic nerve head to the peripheral retina. A decrease in cell number was observed for all three populations with respect to retinal eccentricity (Figure 3g–i). For the full list of p-values, see Table S1.

FIGURE 3:

Density distributions of bipolar cells, cones and horizontal cells. (a) All-type Bipolar cells (All-BCs) were identified by labelling with visual system homeobox 2 (Chx10) antibodies. Somas counted were tightly clustered and brightly labelled in the inner nuclear layer. (b) Cones were identified by labelling cone arrestin (CAR) with antibodies. Cones were marked at the start of the inner segment, below the synaptic terminal. (c) Horizonal cells were identified by labelling with calbindin (Calb) antibodies. Somas were brightly labelled within the inner nuclear layer, proximal to the outer nuclear layer. (d) Density by region of All-BCs, cones, and horizontal cells, respectively. All-BC density does not vary by region. Regional All-BC densities are as follows: Dorsal: 26,815 ± 2,336 cells/mm2; Ventral: 27,247±1,837 cells/mm2; Nasal: 27,242 ± 1,001 cells/mm2; Temporal: 26,208 ± 1,869 cells/mm2 (mean ± SD; N= 2 retinas, n = 43 fields, n = 128,898 cells). (e) Cone density does not vary by region. Regional cone densities are as follows: Dorsal: 14,912 ± 1,587 cells/mm2; Ventral: 14,423 ± 1,913 cells/mm2; Nasal: 13,788 ± 1,247 cells/mm2; Temporal: 14,710 ± 1,539 cells/mm2 (N = 2 retinas, n = 38 fields, n = 54,541 cells). (f) Horizontal cell density change is most prominent along the dorsal-ventral axis (p= 2.23 E-8). Horizontal cell densities are as follows: Dorsal: 1,295 ± 247 cells/mm2; Ventral: 1,664 ± 294 cells/mm2; Nasal: 1,442 ± 147 cells/mm2; Temporal: 1,466 ± 165 cells/mm2. (N = 5 retinas, n = 113 fields, n = 26,651 cells). When comparing between regions, an ANOVA with Tukey’s post-hoc indicated significant difference between regions. Difference between regions are denoted with asterisks (*). (g) BC density has negligible correlation with eccentricity (distance from the optic nerve) (ρ = −0.26771; R² = 0.0717). (h) Cone density has negligible correlation with eccentricity (ρ = −0.21338; R² = 0.0455). (i) HC density has a weak negative correlation with eccentricity (ρ = −0.43039; R² = 0.1852). For the full list of p-values, see Table S1. p ≤ 0.05; p ≤ 0.01; p ≤ 0.001 is indicated by *, **, and *** respectively. Pearson’s coefficient is denoted as ρ.

We next mapped the density of OFF BC types, starting with BC1s. Recent studies have determined BC1s are composed of two sub-populations with distinct morphology and connectivity patterns: BC1a and BC1b (Kay, Voinescu, Chu, & Sanes, 2011; Schubert et al., 2008; Shekhar et al., 2016). BC1bs have amacrine cell-like morphology, yet do not express pan-amacrine cell markers (Pax6) and make glutamatergic ribbon synapses in the IPL (Della Santina et al., 2016). BC1s were identified using the mitoCFP-P line, which expresses cyan fluorescent protein (CFP) in BC1s and nGnG amacrine cells. Though confirmed to be reliable from previous studies (Kay et al., 2011; Schubert et al., 2008; Shekhar et al., 2016) for identifying BC1s, it should be noted that mouse transgenes may reduce expression due to methylation (Garrick, Fiering, Martin, & Whitelaw, 1998; Ingelbrecht, Van Houdt, Van Montagu, & Depicker, 1994; Matzke, Neuhuber, Park, Ambros, & Matzke, 1994).

BC1as were identified as CFP⁺, Ppp1r17⁻ with a dendritic process extending towards the IPL, while nGnG amacrine cells were CFP⁺, Ppp1r17⁺ (Figure 4a). BC1as had a mean density of 2,261 ± 589 cells/mm². The density gradients of BC1as was strongest comparing the dorso-ventral axis (Figure 4b). The lowest densities of BC1as was observed in the dorsal region, 1,904 ± 801 cells/mm². The highest densities were observed in the ventral region: 2,471 ± 329 cells/mm² (p = 0.002) (Figure 4c). Type 1a cell spatial heterogneity shows neglegible positive correlation with eccentricity (ρ = 0.26393) and a weak negative correlation correlation with density (ρ = −0.45597) (Figure 4d–f).

FIGURE 4:

Type 1a BC mapping and spatial heterogeneity. (a) Type 1a BC (BC1as) were identified by using the MitoP line expressing cyan fluorescent protein (CFP) in BC1as and nGnG amacrine cells. CFP fluorescence was amplified using anti-GFP. BC1as were identified as CFP⁺, Ppp1r17⁻ in conjunction with having a dendrite process extending into the OPL. (b) Density contour map of BC1as centred around the optic nerve (N = 4 retinas, n = 91 fields, n = 28,588 cells). (c) Density of BC1as by region. Dorsal: 1,904 ± 801 cells/mm²; Ventral: 2,471 ± 329 cells/mm²; Nasal/Temporal: 2,349 ± 473 cells/mm². When comparing between regions, an ANOVA with Tukey’s post-hoc showed significant difference between regions. Difference between regions are denoted with asterisks (*). (d) Voronoi tessellation of BC1as at low, medium and high densities. (e) Negligible positive correlation with eccentricity (distance from the optic nerve. (e) and BC1a spatial heterogeneity (ρ = 0.2537; R² = 0.0644). (f) BC1a spatial heterogeneity has weak negative correlation with cell density (ρ = −0.4591; R² = 0.2107). For the full list of p-values, see Table S1. p ≤ 0.05; p ≤ 0.01; p ≤ 0.001 is indicated by *, **, and *** respectively. Pearson’s coefficient is denoted as ρ.

Similar to Shekhar, et al. 2016, we found that BC1bs resided in the amacrine cell layer (ACL) and lacked dendrite process extending into the outer plexiform layer (OPL). Ppp1r17 was used as an amacrine cell marker to determine relative location of BC1a and BC1bs within the INL. (Figure 5a). BC1bs had a mean density of 1,019 ± 265 cells/mm². Density gradients were most pronounced along the dorso-ventral axis (Figure 5b), although a region of lower density was observed in multiple retinas in the central ventral retina. The lowest densities of BC1bs were observed in the dorsal region with an average of 839 ± 340 cells/mm². The highest densities observed were in the ventral region: 1099 ± 167 cells/mm² (p = 0.001 when comparing dorsal vs. ventral) (Figure 5c). BC1b spatial heterogeneity had negligible correlation with both eccentricity (ρ = 0.07911) and density (ρ = −0.14677) (Figure 5d–f).

FIGURE 5:

Type 1b BC density mapping and spatial heterogeneity. (a) Type 1b BCs (BC1bs) were identified by using the MitoP line expressing cyan fluorescent protein (CFP) in BC1bs and nGnG amacrine cells. CFP fluorescence was amplified using anti-GFP. (i) BC1bs were identified as CFP⁺, Ppp1r17⁻ in conjunction with lacking a dendrite process projecting into the OPL and absent in the INL (ii-iii) residing mostly in the ACL (denoted with arrows), whereas BC1as had a dendrite process (denoted with arrows). (b) Density contour map BC1bs centred around the optic nerve (N = 4 retinas, n = 91 fields, n = 13,018 cells). (c) Density of BC1bs by region. Dorsal: 839 ± 340 cells/mm²; Ventral: 1099 ± 167 cells/mm²; Nasal/Temporal: 1077 ± 214 cells/mm². When comparing between regions, an ANOVA with Tukey’s post-hoc test showed significant difference between regions. Difference between regions are denoted with asterisks (*). (d) Voronoi tessellation of BC1bs at low, medium and high densities. (e) Negligible positive correlation with eccentricity (distance from the optic nerve) and spatial heterogeneity (ρ = 0.07911; R² = 0.00277). (f) As BC1b cell spatial heterogeneity has negligible negative correlation with cell density (ρ = −0.14677; R² = 0.0321). For the full list of p-values, see Table S1. p ≤ 0.05; p ≤ 0.01; p ≤ 0.001 is indicated by *, **, and *** respectively. Pearson’s coefficient is denoted as ρ.

BC2s were identified by expression of synaptotagmin 2 (syt2) in the ONL (Fox & Sanes, 2007; Wassle et al., 2009). Syt2 brightly labeled axons and faintly labeled somas. Cells were marked as the axon stalk connects to the soma (Figure 6a). The composite density map revealed cell densities ranging from 3,200–1800 cells/mm² with a sharp decrease in density at 1.5 mm from the optic nerve in the dorsal retina (Figure 6b). BC2s had a mean density of 2528 ± 524 cells/mm² (N = 8 retinas, n = 169 fields; n = 74,264 cells). When assesing BC2s binned by region, the most dense region was the ventral retina. The ventral region contained 2,723 ± 378 cells/mm² and had the lowest SD whereas the dorsal retina was least dense and averaged 2,200 ± 619 cells/mm² and had the highest SD (p = 8.86E-6). Nasal and temporal regions had densities of 2,592 ± 455 and 2,626 ± 511 cells/mm² respectively. Nasal, temporal and ventral regions were not statistically different from one another. Density of BC2s in the dorsal region is lower than all other retinal regions (Figure 6c). BC2 spatial heterogeneity had negligible correlation (ρ = 0.18898) with eccentricity and a weak negative correlation (ρ = −0.31994) with respect to density (Figure 6d–f).

FIGURE 6:

Type 2 BC density mapping and spatial heterogeneity. (a) Type 2 BCs (BC2s) were identified by labelling with Synaptotagmin 2 (Syt2) antibodies. Faint somas with brightly labelled dendrite branches were identified as BC2s. (b) Density contour map of BC2s centred around the optic nerve (N = 8 retinas, n = 169 fields, n = 74,264 cells). (c) Density of BC2s by region are as follows: Dorsal: 2,200 ± 619 cells/mm²; Ventral: 2,723 ± 378 cells/mm²; Nasal: 2,592 ± 455 cells/mm²; Temporal: 2,626 ± 511 cells/mm². When comparing between regions, an ANOVA with Tukey’s post-hoc test showed significant difference between regions. Significance is denoted with asterisks (*). (d) Voronoi tessellation of BC2s at low, medium and high densities. (e) Negligible positive correlation with eccentricity (distance from the optic nerve) and Type 2 BC spatial heterogeneity (ρ = 0.18898; R² = 0.0066). (f) BC2 cell spatial heterogeneity has a weak negative correlation with cell density (ρ = −0.31994; R² = 0.0425). For the full list of p-values, see Table S1. p ≤ 0.05; p ≤ 0.01; p ≤ 0.001 is indicated by *, **, and *** respectively. Pearson’s coefficient is denoted as ρ.

BC3as were labled using antibodies to hyperpolarization activated cyclic nucleotide gated potassium channel 4 (HCN4). HCN4 protein is localized throughout BC3a membranes (Mataruga et al., 2007; Muller et al., 2003; Wassle et al., 2009) (Figure 7a). We observed a sharp dorsal-ventral gradient of BC3as with a sharp reduction in cell count aproximately 1 mm from the optic nerve in the dorsal region. BC3as in the venteral region show a band between 0.50–1.00 mm from the optic nerve with densities upwards of 3,300 cells/mm² (Figure 7b). BC3as had a mean density of 2,532 ± 647 cells/mm² (N = 8 retinas, n = 181 fields, n = 66,774 cells). The ventral region had the highest density of 2,958 ± 349 cells/mm² and had the lowest SD. The dorsal retina was least dense and averaged 1976 ± 668 cells/mm², having the highest SD. Nasal and temporal regions had densities of 2,579 ± 455 and 2,661 ± 533 cells/mm² respectively. BC3a density in the dorsal region was statistically different from the ventral, nasal and temporal regions (p = <E-15, p = 1.50E-6, p = 6.44E-4 respectively). Nasal and temporal regions were not statistically different from one another. (Figure 7c). BC3a spatial heterogeneity had a weak positive correlation (ρ = 0.44308) with eccentricity and a moderate negative correlation (ρ = −0.69318) with respect to density (Figure 7d–f).

FIGURE 7:

Type 3a BC density mapping and spatial heterogeneity. (a) Type 3a BC (BC3as) were identified by labelling with hyperpolarization activated cyclic nucleotide gated potassium channel 4 (HCN4) antibodies. Brightly labelled somas with dendritic process were denoted as BC3as. Faintly labelled cells without dendritic process extending into the outer plexiform layer were denoted as amacrine cells. (b) Density contour map of BC3as centred around the optic nerve (N= 8 retinas, n = 181 fields, n = 66,774 cells). (c) Density of type BC3as by region. Dorsal: 1976 ± 668 cells/mm²; Ventral: 2,958 ± 349 cells/mm²; Nasal: 2,579 ± 455 cells/mm²; Temporal: 2,661 ± 533 cells/mm². When comparing between regions, an ANOVA with Tukey’s post-hoc test showed significant difference between regions. Significance is denoted with asterisks (*). (d) Voronoi tessellation of BC3as at low, medium and high densities. (e) Weak positive correlation with eccentricity (distance from the optic nerve) and BC3a spatial heterogeneity (ρ = 0.44308; R² = 0.2602). (f) BC3a cell spatial heterogeneity has a moderate negative correlation with cell density (ρ = −0.69318; R² = 0.397). For the full list of p-values, see Table S1. p ≤ 0.05; p ≤ 0.01; p ≤ 0.001 is indicated by *, **, and *** respectively. Pearson’s coefficient is denoted as ρ.

BC3bs were labeled using an antibody to protein kinase A, regulatory subunit IIβ (PKARIIβ). BC3bs somas show immunoreactivity in their somas and brightly labeled dendrites that extended into the OPL (Mataruga et al., 2007; Wassle et al., 2009) (Figure 8a). A population of amacrine cells was also faintly immunorective. Cells were marked and counted where the dendrite stalk extended from the soma immediately before the dendrites began branching to avoid mislabeling of amacrine cells. The composite density map revealed that BC3bs have a dense ring around the optic nerve, but otherwise a fairly flat density distribution apart from the dorsal-ventral gradient. This distribution remained constant throughout the retina with a slight dorsal gradient at 1.5 mm from the optic nerve. A high density ring of BC3bs reaching upwards of ~4,700 cells/mm² at ~0.25 mm from the optic nerve (Figure 8b). BC3b had a mean density of 3,592 ± 622 cells/mm² (N = 8 retinas, n = 174 fields, n = 69,464 cells). When considering regional localization of BC3bs, we found densities of 3,335 ± 734, 3,785 ± 488, 3,579 ± 569, 3,700 ± 591 cells/mm² in the dorsal, ventral, nasal, temporal regions respectively. BC3b density in the dorsal region differed from ventral and temporal regions (p = 0.002, p = 0.036 respectively) (Figure 8c). BC3b spatial heterogeneity had a weak negligible correlation with both eccentricity (ρ = −0.01487) and density (−0.12835) (Figure 8d–f).

FIGURE 8:

Type 3b BC density mapping and spatial heterogeneity. (a) Type 3b BC (BC3bs) were identified by labelling with protein kinase CAMP-dependent type II regulatory subunit beta (PKARIIβ) antibodies. (b) Density contour map of BC3bs centred around the optic nerve (N= 8 retinas, n = 174 fields, n = 69,464 cells). (c) Density of BC3bs by region. Dorsal: 3,335 ± 734 cells/mm²; Ventral: 3,785 ± 488 cells/mm²; Nasal: 3,579 ± 569 cells/mm²; Temporal: 3,700 ± 591 cells/mm². When comparing between regions, an ANOVA with Tukey’s post-hoc test showed significant difference between regions. Significance is denoted with asterisks (*). (d) Voronoi tessellation of BC3bs at low, medium and high densities. (e) Negligible negative correlation with eccentricity (distance from the optic nerve) and BC3b spatial heterogeneity (ρ = −0.01487; R²= 6E-05). (f) BC3b cell spatial heterogeneity has a negligible negative correlation with cell density (ρ = −0.12835; R² = 0.0231). For the full list of p-values, see Table S1. p ≤ 0.05; p ≤ 0.01; p ≤ 0.001 is indicated by *, **, and *** respectively. Pearson’s coefficient is denoted as ρ.

BC4s were labeled with an antibody to calsenilin (csen). Csen is expressed throughout the BC4 membrane and brightly labled the soma and dendrite projections (Haverkamp et al., 2008; Puller et al., 2013) (Figure 9a). BC4s had a high density in the nasal and temporal regions with downward dorsal gradient 1.0 mm from the optic nerve (Figure 9b). BC4s had a mean density of 2,369 ± 687 cells/mm² (N = 6 retinas, n = 135 fields, n = 48,339 cells). BC4 densities varied by region. The most dense region was the ventral retina, which also had the lowest SD with 2,529 ± 477 cells/mm². In contrast, the dorsal region was the least dense and had the highest SD 2,051 ± 857 cells/mm² (p = 0.022). Nasal and temporal regions had densities of 2,370 ± 673 and 2,503 ± 623 cells/mm² respectively. Nasal, temporal and ventral regions were not found to be statistically different from one another (Figure 9c). BC4 spatial heterogeneity has a negligible correlation (ρ = 0.28006) with eccentricity and a weak negative correlation (ρ = −0.47239) with respect to density (Figure 9d–f).

FIGURE 9:

Type 4 BC density mapping and spatial heterogeneity. (a) Type 4 BCs (BC4s) were identified by labelling with calsenilin (Csen) antibodies. Somas labelled with calsenilin that did not have a dendritic process extending into the outer plexiform layer were denoted as amacrine cells. (b) Density contour map of BC4s centred around the optic nerve (N= 6 retinas, n = 134 fields, n = 48,339 cells). (c) Density of BC4 by region. Dorsal: 2,051 ± 857 cells/mm²; Ventral: 2,529 ± 477 cells/mm²; Nasal: 2,370 ± 673 cells/mm²; Temporal: 2,503 ± 623 cells/mm². When comparing between regions, an ANOVA with Tukey’s post-hoc test showed significant difference between regions. Significance is denoted with asterisks (*). (d) Voronoi tessellation of BC4s at low, medium and high densities. (e) Negligible positive correlation with eccentricity (distance from the optic nerve) and BC4 spatial heterogeneity (ρ= 0.28006; R²=0.1008). (f) BC4 cell spatial heterogeneity has a weak negative correlation with cell density (ρ= −0.47239; R²=0.1997). For the full list of p-values, see Table S1. p ≤ 0.05; p ≤ 0.01; p ≤ 0.001 is indicated by *, **, and *** respectively. Pearson’s coefficient is denoted as ρ.

OFF BCs and HCs had varied densities (BC3b > BC4 = BC3a = BC2 = BC1a > HC > BC1b) (Figure 10a) and spatial heterogeneities (BC1b > BC4 = BC2 = BC1 = BC1a > BC3a > BC3b > HC) (Figure 10b). Spatial heterogeneity was inversely related to density, consistent with the constraints an increased cell density places on measurements of their organization (Keeley, Eglen, & Reese, 2020). We did not detect differences between overall retinal area (not shown), in cell densities when comparing the left and right eyes or comparing BC densities in male and female mice (Figure 11a–d).

FIGURE 10:

OFF bipolar cell (BC) density and spatial heterogeneity by subtype. (a) OFF BC average densities by subtype (mean ± SD). HC: 1,478 ± 261 cells/mm² (N = 5 retinas, n = 113 fields, n = 26,651 cells), BC1a: 2,261 ± 589 cells/mm² (N = 4 retinas, n = 90 fields, n = 41,606 cells), BC1b: 1,019 ± 265 cells/mm² (N = 4 retinas, n = 91 fields, n = 41,606 cells), BC2: 2528 ± 524 cells/mm² (N = 8 retinas, n = 169 fields, n = 74,264 cells), BC3a: 2,532 ± 647 cells/mm² (N= 8 retinas, n = 181 fields, n = 66,774 cells), BC3b: 3,592 ± 622 cells/mm² (N= 8 retinas, n = 174 fields, n = 69,464 cells), BC4: 2,369 ± 687 cells/mm² (N= 6 retinas, n = 134 fields, n = 48,339 cells). (b) OFF-BC average spatial heterogeneity by subtype (median coefficient of variation (CV)). HC: 0.35476, BC1a: 0.57736, BC1b: 0.68261, BC2: 0.5613, BC3a: 0.52814, BC3b: 0.43294, BC4: 0.58713. Significant differences were found comparing populations. Statistical values from ANOVA with Tukey’s post-hoc and Kruskal-Wallis ANOVA with Dunn’s post-hoc can be found in S1: Table summary of statistics.

FIGURE 11:

OFF bipolar cell (BC) 2, 3a, 3b, and 4 density by sex and eye. (a) BC2 density by sex and eye. Male: 2,549 ± 558 cells/mm²; Female: 2,504 ± 487 cells/mm²; Left: 2,480 ± 539 cells/mm²; Right: 2,571 ± 510 cells/mm² (N = 4 male, N = 4 female retinas; N = 4 left, N = 4 right retinas). (b) BC3a density by sex. Male 2,587 ± 682 cells/mm²; Female: 2,474± 608 cells/mm²; Left: 2,558 ± 643 cells/mm²; Right: 2,505 ± 654 cells/mm² (N = 4 male, N = 4 female retinas; N = 4 left, N = 4 right retinas). (c) BC3b density by sex and eye. Male: 3,673 ± 570 cells/mm²; Female: 3,512 ± 663 cells/mm²; Left: 3,599 ± 652 cells/mm²; Right: 3,585 ± 595 cells/mm² (N = 4 male, N = 4 female retinas, N = 4 left, N = 4 right retinas). (d) BC4 density by sex and eye. Male: 2,307 ± 686 cells/mm²; Female: 2,437 ± 687 cells/mm²; Left: 2,472 ± 717 cells/mm²; BC4 ± Right: 2,268 ± 644 cells/mm² (N = 3 male, N = 3 female retinas, N = 3 left, N = 3 right retinas). No significance was found between sex and eye. Statistical values from student’s t-test can be found in S1: Table summary of statistics.

DISCUSION

In this study we mapped the density of all known populations of OFF BCs in the mouse retina. We found marked regional heterogeneity, the extent of which varied by cell type. Our findings are consistent with what has been described for other cell types in the mouse retina, such as RGCs, which also have a dorsal-ventral distribution gradient (Dräger & Olsen, 1981; Jakobs, Libby, Ben, John, & Masland, 2005; Salinas-Navarro, Jiménez-López, et al., 2009).

The most pronounced differences in regional cell localization in the mouse retina is the dorsal-ventral distribution of cone sub-populations. Total cone density is evenly distributed across the mouse retina, with around a 20% decrease moving from the central retina towards the periphery (Haverkamp et al., 2005; Ortin-Martinez et al., 2014). However, cones expressing predominantly short-wavelength sensitive opsin 1 (opnsw1; blue cones), localize mostly in the pan-ventral region of the retina and the dorsal retina has a low number of opnsw1 positive cones (Haverkamp et al., 2005). A subpopulation of blue cones, those that do not express any green opsin (true blue cones) are distributed throughout the dorsal and ventral retina, making up about 30% of cones in the mouse retina (Nadal-Nicolás et al., 2020). In contrast to the distribution of cones in the mouse retina, rod photoreceptor population has been found to have a more constant regional distribution with an estimated ~437,000 rods per mm² in the mouse retina, contributing 97.2% to the total photoreceptor population (Carter-Dawson & LaVail, 1979; Jeon et al., 1998).

Photoreceptors make synapses with both bipolar and horizontal cells. HC density ranges from 1,600–1,700 cells/mm² between peripheral and central regions of the retina, respectively (Raven & Reese, 2002). These studies did not consider retinal domain and in the current study we report a dorsal-ventral gradient of horizontal cells that correlates with the dorsal-ventral gradient we observe for many OFF BC types.

The BCs downstream of photoreceptors also have different distributions within the retina. Rod bipolar cell (RBC) density in the macaque retina shows dramatic variance with increasing distance from the optic disk. The central retina has more than 2-times the density of the peripheral retina (Grunert & Martin, 1991). ON and OFF-cone BCs in the macaque retina show an even sharper contrast in density gradients at increasing eccentricities. Peak density at 1 mm from the optic disk is nearly 7-times more dense than the far periphery (Martin & Grünert, 1992). A more recent study of bipolar cell connectivity with RGCs in the mouse retina identified differences in axonal arbor sizes in different retinal domains, predicting differences in bipolar cell density given the tiled organization of bipolar cell axonal and dendritic fields (Wassle et al., 2009; Yu et al., 2018).

Bipolar cells transmit visual stimulus information to amacrine and ganglion cells. Amacrine cells as a population follow a roughly flat distribution, but the ACs are highly heterogenous and sharper gradients exist for some of the many types of ACs (Perez de Sevilla Muller, Azar, de Los Santos, & Brecha, 2017; Yan et al., 2020). For example, cholinergic amacrine cells decrease in density by approximately 50% in both the dorsal and ventral regions of the retina with increasing distance from the optic nerve head (Jeon et al., 1998). A more extreme example of regional variation is the distribution of displaced somatostatin positive amacrine cells in the mouse retina, which are localized in the ventral but not dorsal retina (Vuong, Hardi, Barnes, & Brecha, 2015).

Ganglion cell distribution has been found to vary across the dorsal and ventral axis of the retina. A horizontal band of high cell density appears in the dorsal region near the optic nerve, then has a sharp decline with increasing distance from the center (Salinas-Navarro, Mayor-Torroglosa, et al., 2009). Sub-populations of ganglion cells are concentrated regionally, for example SPIG1-positive retinal ganglion cells are concentrated in the dorsal retina (Yonehara et al., 2008).

The functional consequences of biased cell distribution are still emerging. In the primate retina the gross distribution of cells in the fovea, macula and periphery are responsible for our high acuity central visual field surrounded by a less sensitive peripheral visual field (Apple, 1981). Different distribution of neurons across retinal domains could also assist in stabilizing ganglion cell output with respect to the visual field (Warwick, Kaushansky, Sarid, Golan, & Rivlin-Etzion, 2018). Surprisingly, the distribution of cones in the mouse retina does not appear to have a large impact on the ability to see in color, possibly due to contribution from the rod pathway and co-expression of blue opsin in most green cones (Denman et al., 2018; Nadal-Nicolás et al., 2020). Likewise convergence patterns can assist in maintaining input from BCs to RGCs despite differences in density and dendritic and axonal arbor sizes (Yu et al., 2018).

On a molecular level the developmental specification of the dorsal-ventral axis of the retina can be explained by differential distribution of signaling molecules and gene expression (Peters & Cepko, 2002). Detailed investigations into differences in cellular densities in the mouse retina have revealed that cell density is tightly regulated and sensitive to genetic modifiers (Reese & Keeley, 2016). The dorsal-ventral variance we report here will increase the power of this approach, which leverages recombinant inbred strains to discover genes influencing retinal organization.

We also measured the heterogeneity of spatial distribution. Consistent with findings by Reese and colleagues we find that spatial organization increases as cell density increases because the position of cell bodies constrains where additional cells can be located (Keeley & Reese, 2014). We found an exception to this pattern in type 3b BCs. We were able to measure BC3bs at the dendrite point before branching, which is a smaller space than the soma and could be offset from the soma due to dendritic tiling. This would suggest that factors that mediate spacing of BCs could promote dendrite spacing but not soma spacing, consistent with the localization of DSCAM protein, which regulates OFF BC tiling, on the dendrite tips of these cells (Simmons et al., 2017; Simmons et al., 2020).

In summary we find that OFF bipolar cells and horizontal cells are organized along the dorsal-ventral axis of the mouse retina, a pattern that has been reported for other cell types. We did not find an overall difference in BC density along the nasal-temporal axis and it will be interesting to determine if there is a similar dorsal-ventral gradient of the ON bipolar cell populations in general as has been documented for ON-type BC9s (Nadal-Nicolás et al., 2020).