Synaptic Convergence Patterns onto Retinal Ganglion Cells Are Preserved despite Topographic Variation in Pre- and Postsynaptic Territories.

Sensory processing can be tuned by a neuron's integration area, the types of inputs, and the proportion and number of connections with those inputs. Integration areas often vary topographically to sample space differentially across regions. Here, we highlight two visual circuits in which topographic changes in the postsynaptic retinal ganglion cell (RGC) dendritic territories and their presynaptic bipolar cell (BC) axonal territories are either matched or unmatched. Despite this difference, in both circuits, the proportion of inputs from each BC type, i.e., synaptic convergence between specific BCs and RGCs, remained constant across varying dendritic territory sizes. Furthermore, synapse density between BCs and RGCs was invariant across topography. Our results demonstrate a wiring design, likely engaging homotypic axonal tiling of BCs, that ensures consistency in synaptic convergence between specific BC types onto their target RGCs while enabling independent regulation of pre- and postsynaptic territory sizes and synapse number between cell pairs.

INTRODUCTION

The sensory integration area of a neuron is defined by its spatial receptive field, which often corresponds to the dendritic territory of the neuron. Other properties of the physiological receptive field, such as its temporal dynamics, are shaped in part by the combination of presynaptic partner types and by the wiring patterns with these partners (Awatramani and Slaughter, 2000; Niell, 2015; Roska and Werblin, 2001). The presynaptic cells themselves also have receptive fields, which in some circuits form a substructure composed of ‘‘subunits’’ within the postsynaptic cell’s integration area (Demb et al., 2001; Freeman et al., 2015; Rust et al., 2005; Schwartz et al., 2012). These functional sub-units provide finer spatial sampling that is engaged in the detection of additional stimulus features (Gollisch, 2013; Schwartz and Rieke, 2011).

The spatial receptive fields of sensory neurons have been found to modify in size systematically across the sensory surface. For example, receptive fields and corresponding dendritic territories of many retinal ganglion cell (RGC) types are relatively more compact at locations that are involved in high acuity vision, such as the fovea (Dacey, 1993; Hammond, 1974; Polyak, 1941). Likewise, neurons in the somatosensory cortex with the smallest receptive fields show enhanced discrimination between adjacent points of contact (Kaas et al., 1979; Knibestöl and Vallbo, 1970). These receptive field variations reflect diverse spatial sampling needs across the sensory field. Such topographic variation within a given neuronal cell type raises the question of whether topological changes in the cell’s spatial receptive field are accompanied by similar alterations in its subunits. An answer to this question will reveal whether receptive field subunits are regulated separately or together with that of the postsynaptic neuron. To answer this question, we thus turn to the vertebrate retina, where the clearest examples of receptive fields with functionally defined subunits can be found.

Previous recordings from RGCs revealed the presence of nonlinear computation at a spatial scale finer than that of the RGC’s receptive field center (Demb et al., 1999; Enroth-Cugell and Robson, 1966; Hochstein and Shapley, 1976; Petrusca et al., 2007; Stone and Pinto, 1993). Nonlinear subunits have been found to permit RGCs to detect fine texture within its spatial receptive field center (Demb et al., 2001; Hochstein and Shapley, 1976; Schwartz et al., 2012) to facilitate detection of the movement of an object in the presence of a moving background (Olveczky et al., 2007) and to generate the response to looming stimuli (Münch et al., 2009). Studies using different size gratings or textured stimuli show that, in some RGCs, the size of the sub-unit corresponds to the size of the receptive field of the bipolar cell (BC), providing the major drive to the recorded RGC, implicating that, for these RGCs, the BC is the structural correlate of the nonlinear subunit (Crook et al., 2008; Demb et al., 2001; Schwartz et al., 2012). Using the mouse retina, where previous work has shown that dendritic arbors of some RGC types exhibit topographic variations in size (Bleckert et al., 2014; El-Danaf and Huberman, 2018; Hughes et al., 2013; Warwick et al., 2018; Zhang et al., 2012), we asked whether BCs also vary in size across the retina. We then asked: do the changes in the BCs vary together with that of their partner RGCs? To obtain a measure of BC size, we quantified the axonal territory area of the BC, which is proportional to the areas of its dendritic arbor and its corresponding receptive field (Berntson and Taylor, 2000; Dunn and Wong, 2012; Zhang and Wu, 2009).

Mouse RGCs, like other CNS neurons (Berck et al., 2016; Callaway, 2002; Fişek and Wilson, 2014), receive input from more than one type of excitatory neuron i.e., each RGC type is contacted by more than one BC type (Dunn and Wong, 2014). Moreover, connectivity with different BC types appears stereotypic (Morgan et al., 2011). Such stereotyped synaptic convergence begs the following additional questions: is connectivity maintained with the same BC types at different retinal locations, and if so, is the proportion of synapses from each BC partner type (synaptic convergence ratio) maintained? The answers will provide insight into whether the balance of synaptic drive from distinct BC input types is maintained as the postsynaptic RGC’s receptive field expands or shrinks.

To ascertain the topographic relationship between the RGC dendritic territories and BC axonal territories, we chose two types of mouse RGCs, OFF-sustained (OFF-S) and OFF-transient (OFF-T) alpha cells for our analysis. Alpha and alpha-like RGCs are found in all vertebrate species and are generally found to be motion sensitive and non-direction selective (Sanes and Masland, 2015). These ganglion cell types have previously been suggested to show systematic variations in their dendritic field size across the retina (Bleckert et al., 2014). The gradients in cell arbor size occur along different axes for these two RGC types. Both types are depolarized by light decrement (light-OFF), which suggests that they receive direct excitatory input from glutamatergic OFF cone BCs. There are five to six types of OFF cone BCs whose axonal terminals stratify at different depths of the inner plexiform layer (IPL) (Euler et al., 2014; Tsukamoto and Omi, 2014). Because of their distinct dendritic stratification patterns in the IPL, OFF-S and OFF-T alpha RGCs are presumed to connect differently to the OFF cone BCs (Bleckert et al., 2014; van Wyk et al., 2009). However, these synaptic patterns have not been identified or mapped systematically for the OFF-S and OFF-T RGCs across retinal locations. We thus determined the BC connectivity patterns of these RGC types across the retina using biolistic transfection methods to mark glutamatergic postsynaptic sites (PSD95). The dendritic arbors of the RGCs were visualized in the background of the OFF BC types, identified either by immunostamning or in GFP-expressing transgenic mouse lines. Comparison of the BC-RGC connectivity maps revealed uncorrelated topographic changes in axonal territory sizes among some BC types and differences in the degree of BC axonal and RGC dendritic territory scaling involving the two RGC types. Regardless of the axonal and dendritic arbor size distributions, the synaptic convergence ratios of BC synapses of the populations of each BC type were found to remain constant for both RGC types across the retina. However, some, but not all, of the BC types demonstrated topographic changes in axonal arbor size, and these variations did not always scale along the same retinal axis as the dendritic arbors of their postsynaptic ganglion cell.

RESULTS

Distinct Topographic Distributions of Two Types of OFF Alpha RGCs

OFF-S alpha RGCs labeled in Thy1-YFP-H transgenic mice appear to display a nasal to temporal gradient in their dendritic arbor size (Bleckert et al., 2014). We quantified here the dendritic arbor sizes of the OFF-S RGCs along the nasal-temporal axis after biolistic transfection with plasmids encoding a fluorescent protein to label cells and their glutamatergic postsynaptic sites. Examples of OFF-S RGCs in the nasal and temporal retina labeled biolistically are shown in Figures 1A and 1B. The dendritic arbor sizes obtained using this approach were within the size range of arbors at the same retinal eccentricities, obtained by conventional intracellularly dye filling (Figures 1C and 1D). The dendritic field diameter of the OFF-S alpha RGCs decreased from nasal to temporal retina (Figures 1C and 1D; linear regression: y = ɑ + βx; β = −56.00; p = 0.006; r2 = 0.26).

Figure 1.

Dendritic Morphology and PSD95 Puncta Density of OFF-S and OFF-T RGCs across the Retina

(A and B) Single-cell morphology of OFF-S ganglion cells within the (A) nasal (N) or the (B) temporal (T) retina labeled by biolistic transfection.

(C) Average dendritic field size at different retinal locations; p = 0.012.

(D) Dendritic diameter as a function of percent distance along the N-T axis. Data collected in total from 12 biolistic transfected retinas and 4 dye-injected retinas are shown.

(E and F) Dendritic morphology for OFF-T RGCs in the (E) dorsal (D) or (F) ventral (V) retinas.

(G) Average dendritic diameters of OFF-T RGCs at different retinal locations; p = 0.007.

(H) Dendritic diameter as a function of percent distance along the D-V axis.(D and H) Regression line fitted to the pooled data. Data collected in total from 17 biolistic transfected retinas and 19 dye-injected retinas are shown.

(I–L) Whole-mount views of an OFF-S (I) and OFF-T (K) RGC biolistically labeled by BFP and PSD95-RFP and the respective distributions of PSD95 puncta across the arbors (J and L). Magnified views of a stretch of dendrite (boxed regions) are shown in the lower panels of (I) and (K). PSD95-RFP puncta identified for quantification shown by gray dots (J and L, lower panels).

(M and O) Average linear density of PSD95 puncta for OFF-S (M) and OFF-T (O) at different retinal locations.

(N and P) Scatterplots show linear density of PSD95 puncta as a function of dendritic field size of OFF-S (N) and OFF-T (P).

Linear regression: OFF-S: β = 6.890e–07, p = 0.348, r2 = 0.04; OFF-T: b = −1.615e–06, p = 0.025, r2 = 0.14. * indicates p < 0.05 in (C) and (G). Error bars are SEM. See also Figure S4.

Previously, we showed that OFF-T alpha RGCs demonstrate a small change in cell density from dorsal to ventral retina (Bleckert et al., 2014). Because RGC dendritic arbor size scales inversely with their cell density (Wässle and Boycott, 1991), we expected that OFF-T RGCs would show a corresponding shallow gradient in their arbor size. We measured directly the arbor sizes of OFF-T alpha RGCs using biolistics and by dye-filling GFP-expressing OFF-T RGCs (Figures 1E–1H) in CB2-GFP mice (Huberman et al., 2008). We confirmed a small but significant decrease along the dorsal-ventral axis (Figure 1H; linear regression: y = ɑ + βx; β = −29.41; p = 0.006; r2 = 0.06) for the pooled data. Thus, both OFF-S and OFF-T alpha RGCs show dendritic arbor size gradients across the retina, which occur along different retinal axes: OFF-S varies along the nasal-temporal axis whereas OFF-T varies along the dorsal-nasal axis.

Glutamatergic Synapse Densities on OFF Alpha RGC Dendritic Arbors Are Topographically Invariant

The topographic variation in OFF alpha RGC dendritic arbor size prompted us to ask whether the density of excitatory post-synaptic sites on the dendrites of the OFF-S and OFF-T RGCs also change with retinal location. We thus quantified the density of excitatory postsynaptic sites on OFF alpha RGCs that were biolistically colabeled with BFP and PSD95-RFP (Figures 1I–1P). PSD95-RFP puncta appeared relatively evenly distributed on both OFF-S and OFF-T alpha RGC dendrites (Figures 1I–1L). Overall, OFF-T alpha RGCs had a slightly higher PSD95 density compared to OFF-S alpha RGCs (OFF-T alpha RGCs: 0.41 ± 0.01 [number of puncta/μm]; OFF-S alpha RGCs:0.34 ± 0.01 [number/μm]; p = 0.001). For either RGC type, we found no significant difference in PSD95 density between retinal locations where arbor sizes differed (Figures 1M and 1O; OFF-S: p = 0.884; OFF-T: p = 0.352). PSD95 density was invariant of the dendritic arbor size of the OFF alpha RGCs (Figures 1N and 1P).

Topographic Variations in BC Axonal Territories Are Not Consistent across BC Types

To determine whether the potential presynaptic partners of these RGCs also showed topographic variation, we examined the axonal territories of BCs. We measured the axonal territories of various OFF BC types (T1, T2, T3a, and T4 OFF BCs) that could be presynaptic partners of the OFF alpha RGCs and compared these measures at different locations. OFF bipolar cells were labeled in retinas from Vsx1-cerulean transgenic mice, of which there are two lines: Vsx1-cerulean1 (line 1) shows sparse labeling of types 1–4 (T1–4) BCs (Della Santina et al., 2016; Hoon et al., 2015), whereas the majority of bipolar cells labeled in Vsx1-cerulean2 (line 2; see STAR Methods) are T1 BCs, which do not immunolabel for synaptotagmin-2 (Syt2) (Fox and Sanes, 2007), a marker for T2 BCs (Figure 2A). Because T2 BCs are very sparse in the Vsx1-cerulean1 line, we identified the T2 population by imunostaining for Syt2. Additionally, populations of T4 BCs were visualized using the 5HTR2a-GFP line. Axonal terminal sizes of the T1, T3a, and T4 OFF BCs were obtained directly using transgenic mouse lines (Figures 2A, 2C, and 2D). For T2 BCs immunolabeled with Syt2, we inferred their axonal territory size from their cell distribution because the arbors of neighboring cells could not be separated (Figure 2B). This was achieved by generating and measuring the sizes of the Voronoi domains based on the spatial distribution of the Syt2-labeled axonal stalks (Figure 2B). The Voronoi domains are a reasonable estimate (Figure S1) of the axon territories because the axon terminals of each BC type do not overlap, i.e., they tile (Wä ssle et al., 2009).

Figure 2.

Variations in BC Axonal Size with Retinal Location

(A–D) BC axonal terminal sizes were determined using transgenic lines in which a single BC type expresses fluorescent protein (A, T1, Vsx1-cerulean2; C, T3a,Vsx1-cerulean1; D, T4, 5HTR2a-EGFP) or estimated from their Voronoi domains based on their axon stalk positions (B, T2, white dots).

(E–H) Axonal sizes at four retinal locations are shown here for T1 (E), T2 (F), T3a (G), and T4 (H) bipolar cells.

D, dorso-nasal; DT, dorso-temporal; VN, ventro-nasal; VT, ventro-temporal. **p < 0.001; ***p < 0.0001. Numbers indicate number of cells from 2–5 retinas. Error bars are SEM. See also Figures S1 and S4.

There were three main observations. First, some, but not all, BC types exhibited topographic variations in axonal territory size (Figures 2E–2H). Among the four BC types we examined, only T1 BCs showed a relatively constant axon terminal size across the retina (Figure 2E; Kruskal-Wallis test; p = 0.790). Topographic variations in axonal size differed in extent among the other three BCs types (Figures 2B–2D; Kruskal-Wallis test; T2: p = 0.038; T3a: p = 0.006; T4: p < 0.001). Second, for T2, T3a, and T4 BCs, their axon terminal sizes are relatively smaller in ventral compared to dorsal retina. Third, comparison of Figures 1 and 2 suggests that topographic variation in dendritic arbor sizes in RGCs and in axonal territory sizes in OFF BCs could be matched or unmatched, depending on which BC types connect with which RGC types.

OFF BCs Maintain Their Stereotypic Synaptic Convergence onto OFF RGCs across the Retina

Because both RGC dendritic arbors and BC axonal territories can vary in size across the retina, it is possible that the synaptic convergence ratio (i.e., the proportion of synapses) of the different BC partner types with their postsynaptic RGC type may also alter with location. To determine whether bipolar cell convergence onto the OFF alpha RGCs is invariant with retinal location, we first identified the OFF BC types that synapse with each of the two OFF alpha cell types. We then compared the BC convergence ratio onto OFF-S and OFF-T alpha RGCs at two locations of the retina, where each RGC type exhibits different dendritic arbor sizes (Figure 3). RGCs were labeled with a plasmid encoding BFP and PSD95-RFP. T1 and T4 BCs were identified using transgenic lines, and T2 and T3a BCs were labeled by Syt2 and HCN4 (Mataruga et al., 2007), respectively.

Figure 3.

OFF-S and OFF-T Alpha RGCs Are Differentially Connected to 4 Types of OFF BCs

(A and B) Maps of PSD95-FP apposed (yellow dots) or not apposed (magenta dots) to axon terminals of each BC type are shown for OFF-S (A) or OFF-T alpha RGCs (B). Magnified views of an example of PSD95 (yellow) puncta on the RGC dendrite apposed (solid arrowhead, 1) or not apposed (open arrowhead, 2) to axon terminals of the BC are shown on the left. Because labeling of T1 and T4 BCs was not complete, we only determined the synaptic contacts on the dendritic segments of the RGCs within an area where BCs were labeled. Grey segments of the dendrites indicate regions where there were no labeled BCs.

(C) Quantification (mean ± SEM) showing the pro-portion of the total number of synapses on the dendritic arbor of OFF-S (left) and OFF-T (right) RGCs contributed by each BC type at different retinal locations. Number of cells in each region is provided.

See also Figures S2, S3, and S4.

We found that each ganglion cell type has distinct convergence ratios from the OFF BC types. OFF-S alpha RGCs received the majority of their glutamatergic synapses from T2, followed by T3a, T1, and T4 BCs (Figures 3A and3C). This is apparent from the connectivity maps provided in Figure 3A. The proportions of synapses from each BC partner type were 50.6% ± 1.3% from T2, 6.4% ± 2.3% from T3a, 4.4% ± 1.1% from T1, and 3.7% ± 0.6% from T4 BCs. In contrast, OFF-T alpha RGCs were largely connected to T3a, followed by T4, T2, and T1 BCs (Figures 3B and 3C). The proportions of synapses from each BC partner type were 39.8% ± 1.4% (T3a), 17.6% ± 0.9% (T4), 2.1% ± 0.5% (T2), and 0.07% ± 0.07% (T1; Figure 3C). Note that the sum of the BC inputs here falls short of 100% because some inputs are provided by yet undefined cell types (see Discussion). Nevertheless, our results thus far demonstrate that the two OFF alpha RGCs receive stereotypic and distinct combinations of excitatory input from separate OFF BC types.

What factors could give rise to the different patterns of BC connectivity with OFF-S and OFF-T alpha RGCs? We explored whether the extent of axon-dendrite overlap (Peters and Feldman, 1976) is a major contributing factor. We also asked whether mechanisms are required to limit partners whose processes are in close proximity (Williams et al., 2010; Zhang et al., 2017; Zipursky and Sanes, 2010). We thus considered two scenarios: first, we examined synaptic contacts between an individual BC type and the two OFF alpha RGCs and, second, we examined connectivity between a single OFF alpha RGC type and two BC types whose axon terminals arborize within the depth of the RGC’s dendritic arbor. We discovered that co-stratification increases the likelihood of (Figures S2A–S2C), but does not guarantee, connectivity between a pair of OFF alpha RGCs and a BC type (Figures S2D and S2E). Although the dendrites of OFF-S alpha RGCs intermingled with axon terminals of both T1 and T2 BCs at the same depth, OFF-S alpha RGCs form more synaptic contacts with T2 than with T1 BCs (Figures S2D and S2E; for quantification, see Figure 3C). RGC dendritic arbors stratify prior to the differentiation of BCs (Coombs et al., 2007; Morgan et al., 2006). Our observations here thus suggest that BC axons use cues apart from dendritic lamination of RGCs to target these partners.

Do BC convergence ratios vary across the retina? We found that, irrespective of RGC or BC arbor size, the BC convergence pattern of each OFF alpha RGC type remained the same (Figure 3C). Moreover, there is no significant correlation between the percentage of synapses formed with the major (Figure S3A) or minor (Figure S3B) BC partner type and the RGC dendritic arlbor size. Thus, the pattern of BC synaptic convergence onto the OFF alpha RGCs is maintained despite topographic changes in the size of BC or RGC arbors.

BC Ribbon Number Increases Systematically with Axonal Terminal Volume

How do OFF alpha RGCs maintain their stereotypic connectivity patterns with different presynaptic OFF BC types across the retina? One possible explanation is that each BC population forms a constant density of synapses across the retina. If so, we predict that the number of presynaptic sites (ribbons) of a BC type scales with their axon terminal volume. We thus quantified the number of ribbons within individual BC axon terminals (Figure 4) upon crossing Ribeye-ribeye-tagRFP mice (see STAR Methods) with either Vsx1-cerulean1, Vsx1-cerulean2, or 5HTR2a-EGFP mice. As reported previously, OFF BCs are labeled sparsely Vsx1-cerulean1 mice. Although there are very few T2 BCs in the retinas of vsx1-cerulean1 mice, this cell type was more frequently observed in offspring of these mice bred with Ribeye-ribeye-tagRFP mice. We confirmed the identity of isolated T2 or T3a BCs by immunolabeling for Syt2 or HCN4, respectively. For all BC types, we found the ribbon density per BC to be consistent across retinal location (Figure 4B; T1: p = 0.573; T2: p = 0.282; T3a: p = 0.534; T4: p = 0.132). Rib-bon number and axon terminal volume of each BC type clearly show a positive correlation (Figure 4C). Our data therefore suggest that the convergence ratio of the different BC types onto the OFF alpha RGCs may be the same across the retina because the density of synapses provided by each BC type is invariant with retinal location, despite changes in the size of their axonal terminals.

Figure 4.

Ribbon Densities of 4 OFF BC Types Are Unchanged across the Retina

(A) Maximum intensity projections of confocal image stacks of isolated axon terminals and ribbons (Ribeye-ribeye-tagRFP) of four types of BCs at dorso-nasal (DN) and ventro-temporal (VT) locations.

(B) Quantification of ribbon density (mean ± SEM) per axon terminal at two retinal locations. Numbers indicate number of cells from 2–4 retinas.

(C) Scatterplot of ribbon number as a function of axon terminal volume. Solid lines represent linear regression lines. T1: β= 0.5209, p = 0.039, r2 = 0.2515; T2:β = 0.3761, p = 0.007, r2 = 0.4671; T3a: p = 0.534; b = 0.4256, p = 0.0112, r2 = 0.4078; T4: β= 0.3136, p < 0.001, r2 = 0.6865.

See also Figure S4.

DISCUSSION

RGC connectivity with BCs has been ascertained from serial EM reconstructions (Dunn and Wong, 2014; Helmstaedter et al., 2013), using synapse-specific markers in light microscopy and by electrophysiological recordings from BC and RGC pairs (Morgan et al., 2011; Schwartz et al., 2012; Tien et al., 2017). Here, we used synaptic markers to assess and map the connectivity of OFF alpha RGCs and OFF BCs across the retina. At least 40%–50% of the total synapses on OFF-S and OFF-T alpha RGCs are made with a single OFF BC type (Figure S4). As in a previous study (Neumann et al., 2016), we found that T2 BCs provide the major BC input to OFF-S alpha RGC. In contrast, T1 BCs do not appear to be a major partner of the OFF-S, as suggested previously by an electron microscopy (EM) study based on apposition of cell membranes (Helmstaedter et al., 2013). This discrepancy may be because the RGC type identified in Helmstaedter et al. (2013) may not be an OFF-S RGC (Bae et al., 2018). Our current findings are consistent with the possibility that a non-conventional glutamatergic neuron, the GlMMIs, provides significant synaptic input to the OFF-S alpha cells (Della Santina et al., 2016). Putative OFF-T alpha RGCs (Bae et al., 2018) identified from the serial EM reconstructions are contacted largely by T3a, T3b, and T4 BCs compared to T2 BCs (Helmstaedter et al., 2013). This is consistent with our observations here, except that we could not assess the connectivity of T3b BCs because their axon terminals cannot yet be visualized using cell-specific markers and light microscopy. T3b could, therefore, account for some of the remaining unidentified synaptic inputs on the OFF-T RGC (Figure S4C). In addition, VGlut3 amacrine cells may also provide significant input onto the OFF-T RGC (Lee et al., 2016).

Previous studies suggest that OFF alpha RGCs show distinct topographic variations in their dendritic arbor size (Figure S4A). Here, we confirmed that OFF-S alpha RGCs possess a nasal-temporal decrease in dendritic arbor size, whereas OFF-T alpha RGCs display a decrease along the dorsal-ventral axis (Figure S4A). We found that three of the four types of OFF BCs we examined (T2, T3a, and T4) exhibited relatively larger axon terminals in dorsal compared to ventral retina (Figure S4B). In contrast, T1 axonal arbor sizes were unchanged across the retina. Therefore, OFF BC types, like the OFF alpha RGCs, show differences in topographic variation in the size of their arbors. We found that OFF-T alpha RGCs and T3a and T4 BCs, their major BC input types, vary along the same retinal axis. The relatively smaller sizes of the OFF-T RGC and their major BC partners in the ventral retina would ensure better spatial resolution for the detection, by these RGCs, of objects looming above the animal (Münch et al., 2009). In contrast, OFF-S alpha RGC dendritic and T2 BC axonal arbors do not vary in parallel and in fact alter along different retinal axes. Such uncoordinated changes were also apparent for ON-S alpha RGCs and their major input, the T6 BCs (Bleckert et al., 2014). Such uncorrelated topographic changes in BC and RGC arbors suggest that, for some RGCs, BC receptive field subunit sizes are regulated separately from that of the RGC spatial receptive field. Moreover, because the number of BC output sites (ribbons) is proportional to the size of the BC axonal arbor, pairwise BC-RGC connectivity could alter with retinal location. Such variation in connectivity between individual BCs and their target RGCs could provide a means for fine adjustments of receptive field subunits (Schwartz et al., 2012). Future experiments are, however, needed to uncover the functional benefits for matching or non-matching topographic changes between RGCs and their presynaptic BC partners.

Despite non-matching changes in BC and RGC territories in some RGC-BC circuits, the stereotypic synaptic convergence pattern between BCs and RGCs appears preserved regardless of retinal location. However, because the axonal arbors of neighboring BCs of the same type tile (Wässle et al., 2009) and BC ribbon density are constant, each BC type provides a consistent coverage of the retinal area and a constant density of output synapses from its population despite changes in the terminal sizes of individual BCs. Thus, along any length of RGC dendrite, the density of synapses formed by each BC type is unchanged, even though the number of synapses between a BC-RGC pair can vary (Figure S4D). Interestingly, recent EM reconstructions suggest that the dendritic density of an individual RGC type is uniform (Bae et al., 2018). The spatial arrangements of the populations of BC axonal territories and RGC dendritic arbors may thus both contribute to ensuring an unchanging synapse density provided by each BC type to its target RGC population.

Although our current work has focused on the contributions of BC connections to the RGC spatial receptive field, how temporal properties of RGC receptive fields vary with retinal location also requires attention. Previous work suggested there is matching of the temporal responses to light of the excitatory interneurons and their postsynaptic RGCs (Awatramani and Slaughter, 2000; Della Santina et al., 2016; Franke et al., 2017). However, stimulation of an individual BC can result in different temporal responses of distinct postsynaptic RGC partners (Asari and Meister, 2012). Moreover, differential glycinergic inhibition onto OFF-T alpha RGCs or their presynaptic inputs have recently been shown to modify the temporal response of these RGCs along the dorsalventral axis (Warwick et al., 2018). This inhibition is provided by the primary rod pathway and does not involve the cone pathway. Preserving the synaptic convergence patterns of BC-RGCs may not contribute to defining the ‘‘transient’’ or ‘‘sustained’’ nature of the RGC output. Stereotypic BC-RGC connectivity patterns, however, are likely to be important for shaping other output properties of RGCs, e.g., synaptic gain or signal-to-noise ratios. Notably, maintaining the proportion of input types across the RGC dendritic arbor would ensure properties, such as the input signal-to-noise ratio, at the majority of synapses across the arbor. In contrast, changes in the proportion of input types would change the balance of information from each BC channel, perhaps changing functional properties of RGCs across retinal location. Future work comparing the properties of synaptic transmission from an individual BC or BC type to different postsynaptic RGC types will provide insight into the role(s) these interneurons play in generating receptive fields that help define the visual processing features of their disparate RGC partners.

STAR★METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Rachel O.L. Wong (wongr2@uw.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

All experiments were conducted following animal protocols approved by the Institutional Animal Care and Use Committee at University of Washington and Stanford University School of Medicine. All procedures in these protocols are in compliance with the National Institute of Health Guidelines for the Care and Use of Laboratory Animals. Adult mice ranging in age between postnatal days (P) 30–60, of either sex were used. CB2-GFP mouse line was used to target OFF-T alpha RGCs (Huberman et al., 2008). Three different transgenic mouse lines were used to target specific fluorescently-labeled BC subtypes: vsx1-cerulean1 (Hoon et al., 2015), vsx1-cerulean2 and 5HTR2a-EGFP (Lu et al., 2009). Vsx1-cerulean2 mice were screened from the same pool of potential founder lines as vsx1-cerulean1. Whereas Types 1–4 OFF BCs are found sparsely labeled in vsx1-cerulean1 mice, the majority of labeled OFF BC cells in vsx1-cerulean2 retinas are Type 1 BCs. Patches of T4 BCs were expressed in 5HTR2a-EGFP (Lu et al., 2009). Ribeye-ribeye-tagRFP mice were generated to visualize ribbons in the retina. Briefly, these mice were produced by pronucleus injection of a DNA construct that contained the mouse Ribeye promotor placed upstream of an expression cassette that encoded for full-length rat Ribeye fused in frame to tagRFP-T (Shaner et al., 2008). Pronucleus injection which was performed by Frank Zimmermann/Sascha Dlugosz (IBF; University of Heidelberg). Positive founder mice were identified by genomic PCR using appropriate primer pairs. We crossed Ribeye-ribeye-tagRFP mice with either BC lines mentioned above to quantify ribbon numbers in individual BC axon terminals.

METHOD DETAILS

Tissue preparation

Mice were deeply euthanized with isoflurane followed by decapitation, and their eyes were enucleated. For immunohistochemistry and biolistic transfection, the eyes were then transferred to oxygenated mouse artificial cerebrospinal fluid (mACSF, pH 7.4) containing (in μm) 119 NaCl, 2.5 KCl, 2.5 CaCl2, 1.3 MgCl2, 1 NaH2PO4, 11 glucose and 20 HEPES at room temperature. After removing the anterior eye and vitreous, a deep cut was made on ventral side to mark the orientation of the eyes (Wei et al., 2010). The retina was isolated from the eyecup and mounted on a nitrocellulose membrane disc (Millipore) with retinal ganglion cell side up. The retinas were subsequently fixed for 15–20 minutes in 4% paraformaldehyde in mACSF. For intracellular dye-fill, the dorsal pole of the left and right eyes were marked before removing them from the animal, using waterproof color markers thus ensuring that knowledge about retinal location was preserved (Wei et al., 2010). Retinas were isolated and kept in oxygenated (95% O2/5% CO2) NaHCO3 (23 μm) containing Ames’ medium (Sigma-Aldrich).

Immunohistochemistry

The retinas were pre-incubated in blocking solutions containing 5% normal donkey serum for 2 hours under room temperature. Then, they were incubated in primary antibodies for 3 nights at 4°C. Secondary antibodies were antiisotypic DyLight (1:1000, Jackson ImmunoResearch) or Alexa conjugates (1:1000, Invitrogen). Primary antibodies used in this study were: anti-GFP (chicken polyclonal, 1:1000, Abcam), anti-Synaptotagmin-2 (mouse monoclonal, 1:1000, znp-1, Zebrafish International Resource Center), anti-HCN4 (rabbit polyclonal, 1:500, Alomone Labs) and anti-myc (mouse IgG1, 1:1000, deposited to the DSHB by Bishop, J.M.).

Biolistic transfection

Gold particles (1.6 μm diameter, 12.5 mg, Bio-Rad) were coated with DNA plasmids encoding cerulean or BFP-myc (24 mg), PSD95-YFP or PSD95-RFP (12 mg) under the control of the cytomegalovirus (CMV) promoter. The particles were delivered to retinal ganglion cells in whole-mount retinas using a Helios gene gun (40 psi, Bio-Rad). Transfected retinas were kept overnight in mACSF in a humid, oxygenated chamber at 33°C.

Intracellular dye-filling

Labeling of individual RGCs was performed using a previously described protocol (Beier et al., 2013; Cruz-Martı´n et al., 2014; Dhande et al., 2013; El-Danaf and Huberman, 2015). The fluorescent RGC somas were localized using differential interference contrast (DIC) and epifuorescence microscopy. RGCs were targeted with borosilicate glass electrodes (Sutter Instruments; 15–20 MU) containing Alexa Fluor 555 hydrizide dye (10 mM in 200 mM KCl; Invitrogen), and were completely filled by applying hyperpolarizing current pulses ranging between 0.1– 0.9 nA for ≤ 1 minute.

Image acquisition

Image acquisition protocols for dye-filled retinas were previously described (El-Danaf and Huberman, 2015). Individual RGCs were imaged using a laser scanning confocal microscope (Zeiss LSM 710) and 40X water immersion objective lens (LDC-Apochromat 40×/1.1). Z stacks were collected at a scanning resolution of 1024 × 1024 pixels with a 0.5 μm Z-step increment size, and a Kalman averaging of 2–4. To obtain the location of individual RGCs, images of the retinal halves were acquired at 5X using an epifluorescence microscope (Zeiss Axio Imager 2) equipped with an Axiocam HR camera, and were later stitched together into whole-retina, orientation-preserved images, using Adobe Photoshop software. Image stacks for other analyses were acquired using a laser scanning confocal microscope (Leica TCS LSP8). Image stacks for individual BC axon terminals and ribbons were acquired using a 63× oil immersion objective (NA 1.4) with a voxel size of 0.05 × 0.05 × 0.25 μm. To obtain the location of biolistic labeled RGCs, image tiles covering the whole-mount retina were acquired under a 20x oil objective (NA 0.75) and automatically stitched together using the tile scan function of the Leica acquisition software. The x-y resolution for each tile was 0.96–1.5 μm and z resolution was 4–5 μm. All other images were acquired using 60× oil objective (NA 1.4) with a voxel size of 0.1 × 0.1 × 0.3 μm.

QUANTIFICATION AND STATISTICAL ANALYSIS

Analysis of retinal ganglion cells

Each flat whole mount retina was reconstructed into a standard spherical retina space using Retistruct R package (Bleckert et al., 2014; Sterratt et al., 2013). The locations of analyzed RGCs were represented as longitudes (nasal to temporal axis) and latitudes (dorsal to ventral axis) in this spherical space. For individual RGCs, their dendrites were skeletonized using Imaris (Bitplane). Manual corrections were made whenever necessary. For mapping all the PSD95 puncta on the dendritic arbor of RGCs, space surrounding the dendrites was masked to limit the search for potential puncta, and puncta were detected using Imaris (Bitplane). Manual corrections were made after automatic detection. Information for both skeletonized dendrites and identified PSD95 puncta was exported to MATLAB (MathWorks). Colocalization of PSD95 puncta with BC axons was assessed plane-by-plane in 2-D using a custom-written program in MATLAB (Okawa et al., 2014). When FP-expressing BCs (T1 and T4) did not cover the entire dendritic arbors of RGC imaged, a 2D mask was created manually to outline the area occupied by the labeled BC axon terminals. Only PSD95 puncta on the RGC dendrites within these areas were included in the colocalization analysis. Contacts within the areas of 48 and 51 T1 BCs within OFF and OFF-T RGCs (n = 6 RGCs each), respectively, were analyzed.

Quantification of bipolar cell arbor size and ribbon density

For quantification of T1 and T4 OFF BC axon arbor sizes, images were acquired in the middle of four quadrants (dorsal-temporal, dorsal-nasal, ventral-temporal and ventral-nasal) of each whole mount retina. Each image stack covered an area of 246 × 246 μm. Each axon’s volume was segmented using the surface function of Imaris (Bitplane). Their axonal territory area was defined as the area of the convex hull of the Z projection of the segmented axonal arbor. For T2 OFF BC, locations of individual T2 OFF BCs was marked by identifying their axon stalks in whole mount retinas immunolabeled with Syt2 (Fox and Sanes, 2007). Voronoi domains were generated using the identified axon stalk positions using R package ‘spatstat’ (Baddeley et al., 2015). The axon arbor sizes were estimated by the sizes of voronoi domains. For measuring ribbon density on BC axon terminals, masks of individual BC terminals were generated using the Labelfield function in Amira (FEI). The ribbon signal within the bipolar axon terminals was isolated by multiplying the original image by the mask. The ribbons were detected by ‘dot finding’ functions in Imaris (Bitplane) with manual correction.

Statistics

All statistics were performed using R version 3.4.1 (R Foundation for Statistical Computing; https://www.R-project.org). Unless noted otherwise, a two-tailed Mann-Whitney test was used to compare two groups of samples. Kruskal-Wallis one way analysis of variance was performed for more than two sample groups. Significance was determined at p < 0.05. Data are presented as mean ± SEM.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-synaptotagmin II	Zebrafish International Resource	Cat# znp-1; RRID: AB_10013783
Rabbit polyclonal anti-HCN4	Alomone Labs	Cat# APC-052, RRID:AB_2039906
Chicken polyclonal anti-GFP	Abcam	Cat# ab13970, RRID:AB_300798
Mouse monoclonal anti-myc	DSHB	Cat# 9E 10, RRID:AB_2266850
Donkey polyclonal anti-chicken IgY,Alexa Fluor 488	Jackson ImmunoResearch Labs	Cat# 703-545-155, RRID: AB_2340375
Donkey polyclonal anti-rabbit IgG,Alexa Fluor 647	Jackson ImmunoResearch Labs	Cat# 711-605-152, RRID: AB_2492288
Goat anti-mouse IgG2a, Alexa Fluor 647	Jackson ImmunoResearch Labs	Cat# 115-605-206, RRID: AB_2338917
Goat anti-mouse IgG1, DyLight 405	Jackson ImmunoResearch Labs	Cat# 115-477-185, RRID: AB_2632529
Chemicals, Peptides, and Recombinant Proteins
Alexa Fluor 555 hydrizide	Invitrogen	Cat# A20501MP
Normal Donkey Serum	Sigma-Aldrich	Cat# D9663
Vectashield	Vector Laboratories	Cat# H-1000, RRID: AB_2336789
Gold Particles/micro carriers(1.6 μm diameter)	Bio-Rad	Cat# 165-2264
Experimental Models: Organisms/Strains
Mouse model: Vsx1-cerulean2	This paper	N/A
Mouse model: Vsx1-cerulean1	(Hoon et al., 2015)	N/A
Mouse model: Ribeye-ribeye-tagRFP	This paper	N/A
Mouse model: CB2-GFP	(Huberman et al., 2008)	RRID:MMRRC_000283-MU
Mouse model: 5HTR2a-EGFP	(Lu et al., 2009)	RRID:MMRRC_010915-UCD
Mouse model: Grm6-tdTomato	(Kerschensteiner et al., 2009)	N/A
Mouse model: Gus8.4-GFP	The Jackson Laboratory	RRID: IMSR_JAX:026704
Recombinant DNA
Plasmid: pCMV-CFP	(Morgan et al., 2011)	N/A
Plasmid: pCMV-tagBFP2-myc	This paper (Modified from pEGFP-N1from Clonetech)	N/A
Plasmid: pCMV-PSD95-tagRFP	This paper (Modified from pCMV-PSD95-YFP)	N/A
Plasmid: pCMV-PSD95-YFP	A.M.Craig, University of British Columbia(Morgan et al., 2008)	N/A
Software and Algorithms
ImageJ	NIH	https://imagej.nih.gov/ij/, RRID:SCR_003070
Amira	Thermo-Fisher Scientific	https://www.fei.com/software/amira-avizo/,RRID: SCR_014305
Imaris	Bitplane	http://www.bitplane.com/, RRID:SCR_007370
MATLAB	MathWorks	https://www.mathworks.com/products/matlab.html,RRID: SCR_001622
R	R Core Team	https://www.r-project.org/,RRID:SCR_001905