Distinct developmental mechanisms act independently to shape biased synaptic divergence from an inhibitory neuron

SUMMARY

Neurons often contact more than one postsynaptic partner type and display stereotypic patterns of synaptic divergence. Such synaptic patterns usually involve some partners receiving more synapses than others. The developmental strategies generating ‘biased’ synaptic distributions remain largely unknown. To gain insight, we took advantage of a compact circuit in the vertebrate retina, whereby the AII amacrine cell (AII AC) provides inhibition onto cone bipolar cell (BC) axons and retinal ganglion cell (RGC) dendrites, but makes the majority of its synapses with the BCs. Using light and electron microscopy, we reconstructed the morphology and connectivity of mouse retinal AII ACs across postnatal development. We found that AII ACs do not elaborate their presynaptic structures, the lobular appendages, until BCs differentiate about a week after RGCs are present. Lobular appendages are present in mutant mice lacking BCs, implying that although synchronized with BC axonal differentiation, presynaptic differentiation of the AII ACs is not dependent on cues from BCs. With maturation, AII ACs maintain a constant number of synapses with RGCs, preferentially increase synaptogenesis with BCs, and eliminate synapses with wide-field amacrine cells. Thus, AII ACs undergo partner type-specific changes in connectivity to attain their mature pattern of synaptic divergence. Moreover, AII ACs contact non-BCs to the same extent in bipolarless retinas, indicating that AII ACs establish partner-type specific connectivity using diverse mechanisms that operate in parallel but independently.

eTOC

Gamlin et al. use serial blockface electron microscopy and confocal microscopy to map the synaptic divergence of an inhibitory interneuron in normal and mutant mouse retinas across development. They show that connections onto distinct partner types of the interneuron are established independently, engaging different developmental strategies.

INTRODUCTION

Vertebrate and invertebrate nervous systems share two major motifs of neuronal connectivity, synaptic convergence and synaptic divergence. Synaptic convergence involves multiple neurons or neuronal cell types providing input onto a common postsynaptic partner, and synaptic divergence entails the distribution of synapses from an individual neuron onto multiple postsynaptic partners or partner types. The importance of converging inputs in the integration of information is evident [1, 2]. Synaptic divergence has also been shown to play a key role in signal processing, such as in the parallel processing of visual information and noise reduction within circuits [1, 3-6]. Developmental mechanisms that shape synaptic convergence have been studied extensively [7-11]. Although there is increasing insight into the strategies that shape stereotypic patterns of synaptic divergence [12, 13], the strategies that establish circuits in which postsynaptic partners receive unequal numbers of synapses from a given presynaptic neuron remain unclear. Elucidating these mechanisms necessitates the ability to identify and reconstruct all the output synapses and synaptic partner types of a neuron, a challenging task for many well-studied circuits involving long-range axonal projections. Here, we take advantage of a compact and functionally well characterized circuit to better understand the cellular mechanisms that result in differential connectivity with distinct postsynaptic partner types.

Several major developmental strategies have been found to establish proper synaptic convergence in neural circuits in which some presynaptic partners provide more connections than others. For example, in zebrafish, H3 horizontal cells largely contact ultraviolet (UV) cones and a handful of blue cones. This biased connectivity pattern is set up by UV cones suppressing synaptogenesis with blue cones upon maturation [14], and preferential synaptogenesis with the ‘preferred’ UV cone partners. Biased synaptic convergence can also be achieved by closer physical proximity to one compared to another presynaptic partner type, providing a higher probability of contact with the partner type whose axons overlap the most with the postsynaptic cell [10]. Elimination of a presynaptic partner type also contributes to the final patterns of synaptic convergence during development [8, 10]. Which of these developmental strategies are engaged in determining synaptic divergence across different postsynaptic partner types is not yet evident. Furthermore, whether only a singular strategy (e.g. preferential pruning) is used, or whether several mechanisms are recruited in parallel, and if so how, remain to be determined. To answer these questions, we chose a circuit in the vertebrate retina for which all synaptic partner types can be identified, and the output synapses of the presynaptic neuron with each partner type can be mapped completely.

The rod pathway of the retina mediates scotopic, or nighttime vision [15], and is highly conserved across mammals [16-22]. Within the rod pathway, light is first detected by rod photoreceptors, which synapse onto rod bipolar cells (rod BCs). Rod BCs in turn provide excitatory input to the distal dendrites of AII amacrine cells (AII ACs), which are glycinergic inhibitory interneurons (Figure 1A). AII ACs form output synapses at presynaptic endings, called lobular appendages, onto distinct classes of retinal neurons, including cone bipolar cells (BCs) and retinal ganglion cells (RGCs) in the OFF sublamina of the inner plexiform layer (IPL) (Figure 1A). Previous electron microscopy (EM) reconstructions of rabbit and mouse retina showed that individual AII ACs form a greater number of synapses onto cone BC partners and fewer onto RGC partners [23-25]. The cellular strategies by which the AII AC achieves its biased pattern of synaptic divergence is unknown.

Figure 1. Presynaptic neurites of AII ACs elaborate later than their input processes.

(A) Schematic of the rod pathway in the mammalian retina. Plus sign indicates excitatory input, negative sign indicates inhibitory input. The inner plexiform layer (IPL) is divided into ON and OFF sublayers [58]. (B) Rotated confocal volume of maximum intensity projection of AII ACs in the CDH1-GFP mouse retina. The orange plane isolates a single row of AII ACs within the volume. (C) Side view of a digitally isolated mature AII AC in a CDH1-GFP retina. (D) Examples of digitally isolated AII ACs at different developmental time points. The images are maximum intensity projections of the entire cell from the side, and en face views exclude the distal dendrites that ramify in the ON sublayer. (E) Quantification of the volume of the lobular appendages of AII ACs at each time point. Volumes were measured by digitally isolating individual lobular structures using Amira and then exporting the pixel volumes (voxels) the binary pixels to MATLAB for summation across the cell [56] (STAR methods). Lobular appendage volume increases significantly with age (One-Way Anova, p = 4.9147e-7). Using the Wilcoxon Rank Sum test, there are significant changes from P14 to P21 and from P21 to ≥P31 (P31 to 3.5 months of age). Wilcoxon Rank Sum tests: **P14-21, p = 0.002; *P21 to ≥P31 p = 0.0431. n = 5 cells, 2 animals (p8); 10 cells,3 animals (p14); 7 cells, 3 animals (p21); 10 cells, 3 animals (p ≥ 31).

The morphological development of AII ACs also raises additional questions concerning how patterned synaptic divergence is attained. In mice, RGCs and amacrine cells are born and differentiate prenatally, whereas BCs differentiate postnatally, nearly a week after synapses between amacrine cells and RGCs have begun to form [26-29]. Thus, AII ACs initially develop in a landscape where their minor postsynaptic partners, the RGCs, are present first, before their major partners, the cone BCs, are generated. We therefore asked whether the presence of BCs is necessary to trigger AII AC synaptogenesis, and whether the temporal disparity in the availability of the different partner types influences synaptic divergence of the AII ACs. To do so, we used mutant mice in which BCs (rod and cone BCs) are never produced during development [30]. We present here reconstructions of AII ACs and their changing ‘connectome’ during normal development, and in the absence of their primary postsynaptic targets.

RESULTS

AII AC output structures develop after input structures are formed.

Previous studies using immunolabeling to identify AII ACs in rabbit [31] suggested that these amacrine cells extend distal dendrites before elaborating their lobular appendages, and that mouse AII ACs may exhibit a similar sequence [32]. To confirm that mouse AII ACs establish their input processes before elaborating lobular appendages, and to determine the time-course of these developmental changes, we followed the morphological changes of mouse AII ACs across postnatal development using transgenic animals in which only these ACs are labeled. We used CDH1-GFP mice in which the majority of AII ACs express GFP, to visualize these cells from postnatal day 4 (P4) to adulthood (Figure 1B, C, D). Because lobular appendages of neighboring AII ACs do not overlap, we were able to digitally isolate these structures on individual cells across ages [see STAR Methods] (Figure 1D). Although distal dendrites were present, no clear neurite extensions in the OFF IPL were observed at P4. Thin and short GFP-labeled processes were first visible in the inner IPL at P6, became more evident with age, and lobular endings appeared during the second postnatal week (Figure 1D). Thus, like the rabbit retina, mouse AII ACs also first form distal dendrites prior to the emergence of lobular appendages. Quantitation of the volume of the lobular appendages [see STAR Methods], which included any lateral neurite extensions from the AII AC soma or central stalks (see Figure 1C), further demonstrated rapid growth of the lobules with age, especially after the time of eye-opening, which occurs around P14 (Figure 1E).

AII AC synaptic divergence is shaped by partner type-specific changes in connectivity.

We next sought to determine when synapses from AII ACs onto postsynaptic partners are established. In addition, we determined the patterns of connectivity between developing AII ACs and each of its major postsynaptic cell classes. To do so, we performed serial blockface scanning electron microscopy (SBFSEM) on P11 (before eye-opening) and P24 retinas. We identified 4 AII ACs at each age from the characteristic dyad synapse between rod BCs, AII ACs, and A17 amacrine cells [33]. From these synapses, we tracked AII AC distal dendrites to their somata and traced the lobular appendages emerging from the cells. Postsynaptic partner classes (i.e. RGCs, BCs, and amacrine cells) were identified by their morphology and synapse specializations [see STAR Methods, Supplemental Figure S1A]. We then quantified all output synapses of each AII AC and their postsynaptic partner types (Figure 2A, B), noting their locations on the cell (Figure 2A; see Figure S2 for higher magnification views of synapses with major partners, BCs and GCs). We also reconstructed the processes of postsynaptic partners within the imaged volume (Figure 2C).

Figure 2. Biased AII AC synaptic divergence emerges early in development.

(A) Left panel: Examples of synapses with each postsynaptic partner identified from EM (See also Figures S1A, S2). Magenta asterisk indicates a synapse onto a BC partner. Yellow asterisk: a RGC partner. Cyan asterisk: an AC partner. The second panels are EM reconstructions of individual AII ACs at P11 and P24. Each synapse is shown on the AII ACs themselves, color-coded to indicate the identity of the postsynaptic partner at that synapse. (B) Top: Quantification of the number of total output synapses of four AII ACs at each time point, and the number of output synapses onto each postsynaptic partner cell class. Bottom: Percentage of the total output synapses that are formed onto each postsynaptic partner cell class at P11 and at P24. (C) EM reconstructions of the same AII ACs in Figure 1A, and traced skeletons of all postsynaptic partners, color-coded by cell class. (D) Number (top) and percentage (bottom) of synapses made with each postsynaptic partner class at P11 (lighter colors) and at P24 (darker colors). * p< 0.05; Wilcoxon rank sum test. n = 4 cells, 1 animal (at each age) Error bars = SEM.

Individual AII ACs showed a significant increase in their total number of output synapses between P11 to P24 (43.3 ± 2.8 to 66 ± 4.1, mean ± SEM, p = 0.029, Wilcoxon Rank Sum test). At P11, the AII ACs have already formed synapses with BCs, RGCs, and other amacrine cells. Comparison of the number of synapses that individual AII ACs formed with each partner class at P11 and P24 revealed cell class-specific changes in connectivity. The number of synapses formed with RGCs did not change significantly between these two ages (11 ± 1.8 to 8.0 ± 0.7, p = 0.289). However, the number of synapses made with BCs increased by almost three-fold (20.8 ± 1.5 to 57 ± 4.6, p = 0.029). This increase matches well with findings from previous single cell electrophysiological recordings showing a three-fold increase in AII AC membrane capacitance jump evoked by Ca²⁺ influx, representing the fusion of synaptic vesicles with the cell membrane, between P10 and P25 [34]. The number of synapses with amacrine cells, however, decreased significantly (10.8 ± 1.7 to 1.00 ± 0.7, p = 0.029) (Figure 2B). These disparate changes in synapse number with each cell class result in a disproportionate increase in connectivity with BCs with age (Figure 2B). Our P24 measurements are in close agreement with previous EM reconstructions in mouse retina, which revealed that AII ACs make on average 64 (vs our 66) total output synapses with approximately 57 synapses onto non-RGCs (vs 57 here) and 7 synapses onto RGCs (vs our 8) [35]. Such agreement in both total synapse number and distribution across postsynaptic partner types supports the idea that divergence from the AII AC is stereotyped both within and across retinas.

To determine whether these synaptic changes reflect an increase in the number of postsynaptic cells, we quantified the number of synaptic partners of each class (BC, RGC, AC) at both ages (Figure 2D). We found no significant difference in the total number of postsynaptic cells (25 ± 3.1 to 19.8 ± 0.9, p = 0.3), or the number of RGC partners between P11 and P24 (8 ± 0.7 to 5.5 ± 0.9, p = 0.086). Although not statistically significant, there is a trend towards an increase in the number of BC partners with age. We did find a significant decrease in the number of amacrine cell partners with maturation (9.5 ± 1.9 to 1.00 ± 0.7, p = 0.029). Our EM reconstructions showed that BCs became the major AII AC target at maturity (30.09% ± 1.79 to 66.65% ± 6.2, p = 0.029), in line with previous findings [25, 35].

Collectively, our observations demonstrate that AII ACs synapse with distinct postsynaptic partner classes well before eye-opening, and that the pattern of synaptic divergence refines with maturation in a class-specific manner.

Synaptogenesis onto BCs increases non-uniformly across BC partners.

Although AII ACs significantly increase their number of synapses with BCs with maturation, this increase can be attributed to only a modest recruitment of BC partners with age (Figure 2B, D). We next sought to determine whether increasing number of synapses were formed onto all or only a subset of BC partners by determining the number of synapses between all AII AC - BC pairs at the two ages (Figure 3A, B). We found that on average, BCs received 1-2 more synapses from an AII AC with maturation though this change was not significant (P11: 2.8 ± 0.4; P24: 4.3 ± 0.7, p = 0.5). However, the maximum number of synapses formed between an AII AC-BC pair increased by more than three-fold between P11 and P24 (6 ± 1.2 to 20.5 ± 2.9, p = 0.03, Wilcoxon Rank Sum test). These findings imply that during maturation, AII ACs increased synaptogenesis with a few, but not all BC partners. One reason that may account for this differential synapse formation is that synapses are preferentially increased with BC partners with axons that branch in the vicinity of the AII AC neurites. To test this hypothesis, we traced the appositions of AII ACs with postsynaptic BC axonal arbors at P24 and generated a measure of neurite overlap (see STAR Methods, Figure S1B). We found that the number of synapses a postsynaptic BC partner receives is positively correlated with the amount of apposition between the connected BC axon terminal and the AII AC neurites (R² = 0.9, Linear Fit: y = 1.7e-07*x – 0.088), supporting a variant of Peter’s rule [36] (Figures 3C, S1B). When we examined the morphology of the most highly connected BC partner of each of the three neighboring AII ACs, we found that the BCs show similar axonal morphology (the axon begins to branch near the top of the AII AC soma and extends processes into the IPL) and that their axons do not overlap but rather ‘tile’ (Figure 3D). These anatomical features suggest that these BCs are of the same type.

Figure 3. An AII AC does not connect uniformly with all BC partners.

(A) EM reconstructions of AII ACs and skeletons of all BC partners in the leftmost panels. BCs are color-coded by the number of synapses each receives as indicated by the heatmap. The right panels are EM reconstructions of the same AII AC with the skeleton of the BC that receives the most synapses from that AII AC. (B) Histogram of the number of synapses that individual BCs receive at P11 and at P24. (C) A plot of the number of synapses that each BC receives and the overlap of the BC terminal makes with the AII AC’s lobular appendages. R² = 0.9 (See also Figure S1B). (D) Reconstructed axon terminals from the most connected BC partners from 3 neighboring AII ACs (one pictured). Note their similar branching pattern (top) and non-overlapping arbors of these BCs (en face view).

In the mature retina, Type 2 BCs are the major postsynaptic targets of AII ACs [25, 35]. Because our reconstructed volume did not extend beyond the INL such that we could unequivocally identify the BC types, we used immunocytochemistry to identify Type 2 BCs under light microscopy. Past electrophysiological studies based on mouse mutants with loss of specific glycine receptor subunits suggested that the glycine receptor subunit α3 (GlyRα3) is at inhibitory synapses formed onto AII ACs [37], but glycine receptor subunit α1 (GlyRαl) is at synapses from AII ACs onto OFF BCs [38]. We thus immunolabeled the CDH1-GFP retina for GlyRα1 (Figure 4A) to determine whether AII ACs preferentially synapse with Type 2 BCs during development. Light microscopy also enabled us to sample many more ages across development compared to SBFSEM. We validated our light microscopy data by comparing synapses defined by GlyRα1 puncta (see Methods) with the total number of output synapses formed by individual AII ACs numbers observed from SBFSEM, and found that measures from both datasets were comparable (Figure S1C).

Figure 4. AII AC synaptogenesis with different BC types proceed at different rates.

(A) Light microscopy images of AII ACs, Type 2 BCs, and glycine receptor α1 (GlyRα1; marks AII AC postsynapse) across ages. (B) Quantification of the number of synapses that individual AII ACs form onto Type 2 BCs across ages (magenta) identified as in (A), and the number of synapses formed onto all BCs measured by EM at P11 and P24 (filled grey) (See also Figure S1C). n = 5 cells, 2 animals (P8); 5 cells, 2 animals (P14); 5 cells, 3 animals (P21); 11 cells, 4 animals (P ≥ 30).

We compared the number of synapses an AII AC forms onto Type 2 BCs under light microscopy with the total number of synapses onto BCs from our EM reconstructions (Figure 4B). At P24, our EM reconstructions reveal that an AII AC forms on average 57 synapses with BCs, and from the light microscopy observations, approximately 30 synapses are formed with Type 2 BCs at >P30 (Figure 4B). Thus, we estimate that approximately 53% (30/57) of the synapses we observe onto BCs in EM are with Type 2 BCs. This is in keeping with previous EM reconstructions from Tsukamoto et al. (2017) [25], who found that in the adult mouse retina, 46% of synapses an AII AC formed with BCs were with Type 2 BCs. However, our estimates from the immunolabeling approach and Tsukamoto et al.’s EM reconstructions are somewhat lower than the percentage (70%) obtained by Graydon et al. (2018) [35], perhaps reflecting sampling in different parts of the retina. Nevertheless, all studies find that Type 2 BCs are the dominant BC partner type for AII ACs. Our P11 EM reconstructions revealed that approximately 21 synapses are formed with BCs by each AII AC. We estimate from the light microscopy observations that Type 2 BCs receive approximately 14 synapses (or 66% of the total BC synapses) at this early age. Thus, AII ACs increase synaptogenesis with BCs throughout development, but before eye-opening, synaptogenesis appears biased towards it’s preferred (Type 2) BC partner type.

Distribution of synapses across AII AC - RGC pairs is invariant with age.

Although the number of synapses an AII AC makes with RGCs does not change with maturation, it remains possible that there is a redistribution of synapses across RGCs (i.e. some RGCs receive more synapses with age, whereas others show a decrease). Thus, we compared the distribution of AII synapses onto individual RGCs in the EM reconstructions (Figure 5A). We found that there is no significant difference between the average number of synapses that an AII AC provides onto individual RGCs at the two ages (P11: 1.3 ± 0.2; P24: 1.5 ± 0.2, p = 0.33; Kolmogorov-Smirnov test could not reject the null hypothesis) (Figure 5B). Thus, synaptic connectivity between AII ACs and their RGC partners appears established early in development and maintained with maturation.

Figure 5. AII connectivity with RGCs is established early.

(A) Skeletons of EM reconstructions of RGC partners of an individual AII AC at P11 and P24. Each RGC partner is represented by a separate color. (B) Top: Histogram of the number of pairwise synapses formed with individual RGCs at P11 and at P24. Bottom: Histograms normalized by the number of instances of pairwise connectivity show the frequency with which an AII AC forms a specific number of synapses with an individual RGC partner. For example, approximately 70% of all connected AII AC and RGC pairs share a single synapse at P24.

Synapse elimination refines AII AC connectivity with other ACs.

AII ACs initially form many synapses with other ACs, but connectivity with these cells decreases with maturation (Figure 2B). At P11, the neurites of most ACs postsynaptic to AII ACs laminated narrowly in the IPL (Figure 6A). The processes of these ACs rarely branched (within the reconstructed volume) and resembled those of GABAergic wide-field ACs (Figure 6A). Each contacted AC process received a single synapse from the reconstructed AII AC (Figure 6B, C). Furthermore, like another described wide-field AC [39], the AC partners of the AII AC form output synapses (Figure 6C) along the length of the same process where they receive input with no clear spatial segregation of inputs and outputs (Figure 6D). The morphological similarities and stratification levels of the postsynaptic AC processes suggest that they likely belong to a single type of AC. Synapses with such wide-field ACs were observed in all four reconstructed P11 AII ACs. Surprisingly, these wide-field processes were rarely encountered postsynaptic to the P24 AII ACs (Figure 6A). In one of four reconstructed P24 AII ACs, the AII AC synapsed with another type of AC that stratified broadly throughout the IPL. Thus, our results suggest that the developmental loss of synapses from AII ACs onto other amacrine cells is largely due to the elimination of synapses with wide-field AC partners.

Figure 6. Transient connections are made with wide-field ACs.

(A) Left: Skeletonized EM reconstructions of wide-field amacrine cell (WF AC) partners of three AII ACs at P11. At P11, each AC partner of the AII AC indicated by an asterisk is traced in non-cyan colors. WF AC partners of the other two AII ACs are shown in cyan. Numbers indicate WF ACs featured below in (D). Right: Skeletons of the postsynaptic partners of an individual AII AC at P24. (B) Left: Histograms of the number of synapses formed onto individual ACs at P11 and at P24. Right: Normalized histograms showing the frequency with which an AII AC forms a specific number of synapses onto an individual AC partner. (C) Top: Synapses (blue arrowheads) from AII AC onto the WF AC processes in (A). Bottom: Examples of output synapses from each of the WF ACs above. (D) EM reconstruction of the two WF processes in (A), showing the location of their output synapses (green patches, white asterisks) and the sites of synaptic input from the AII AC (cyan arrowhead) indicated by the white asterisk in (A).

Synaptic connectivity with BC and RGC partners is regulated independently.

As described above, AII ACs do not elaborate lobular appendages until their major postsynaptic partners (BCs) extend axonal branches into the IPL. This finding is surprising given that other targets of the AII ACs, the RGCs, are present and have elaborated their dendrites in the IPL at least a week before BCs differentiate. The timing of AII AC lobular appendage growth raises the possibility of a role for BCs in triggering the growth of AII lobular appendages. To test this hypothesis, we examined AII AC morphology and connectivity in a mutant mouse line, Vsx2-SE knockout [30] in which BCs never develop (Figure 7A). In these mutant mice, all other retinal cell classes are present, and their lamination patterns persist. We found that in the bipolarless retinas, mature AII ACs have lobular appendages, although the total volume of the lobular appendages is significantly smaller on average (151.3 ± 13.4 μm³ versus 43.8 ± 4.5 μm³, p < 0.001) (Figure 7A, B). The loss of total lobular appendage volume is due to the loss of approximately 2/3 of a cell’s lobular appendages (27.8 ± 1.4 versus 10.3 ± 1.0, p <0.001). We also found a small, but significant reduction in the average volume per lobular appendage (5.5 ± 0.3 μm³ versus 4.4 ± 0.3 μm³, p < 0.05) as quantified by the total volume/number of lobular appendages (Figure 7B). Despite a reduction in total volume and number, AII AC lobular appendages continue to form, on average, 10 ± 0.7 output synapses (Figure 7B lower left). While we were unable to determine the postsynaptic targets in the mutant under light microscopy, our EM reconstructions indicate that at P24, AII ACs normally form 9 ± 0.8 output synapses onto non-BC targets (Figure 7C) of which most synapses (8 ± 0.7) are with RGCs. Thus, AII ACs form the appropriate number of non-BC associated synapses in the absence of the major postsynaptic target (the OFF BCs), suggesting that synaptogenesis onto distinct target cell classes occurs independently.

Figure 7: Connections of AII ACs with different partner types are regulated separately.

(A) On the left panels are fluorescence images of AII ACs in wildtype control and ‘bipolarless’ (Vsx2-SE KO) retinas. On the right panels are the digitally isolated lobular appendages of the AII ACs indicated by color-matched asterisks on the left. Inset: Examples of GlyRα1 synapses apposed to AII AC lobular appendages. (B) Quantification of total lobular appendage volume, number of lobules, and average lobule volume in littermate control and mice lacking bipolar cells (KO). * p<0.05; **, p < 0.001 Wilcoxon-Rank Sum Test. Also shown (bottom right) is the quantification of non-BC associated output synapses that individual control AII ACs form at P24 (data from EM) compared to total output synapses of individual AII ACs in bipolarless mice. No significant difference in synapse number was observed between control and bipolarless mice (p = 0.50, Wilcoxon-Rank Sum Test). Lobular appendage volume: n = 13 cells, 3 animals (littermate control); 15 cells, 3 animals (KO). Synapse number: n = 4 cells, 1 animal (EM control), 16 cells, 3 animals (KO).

DISCUSSION

Structural differentiation of AII ACs and synaptic partner availability.

In many neural circuits, axons elaborate prior to the availability of their final postsynaptic targets. For example, axons from the entorhinal cortex enter the contralateral hippocampus before the dendrites of their pyramidal cell targets develop [40], and thalamocortical neurons extend their axons into the cortex before Layer IV cells are born [41]. Here we described a cell that does not elaborate its presynaptic processes even when a postsynaptic partner type (GC) is present and extending dendrites within reach. Rather, the presynaptic neurites of AII ACs elaborate only when their dominant partner type (BC) is available. Such a developmental sequence in morphological differentiation has also been observed previously in the drosophila olfactory system, where dendrites of projection neurons (PN) neurons become restricted to specific regions of the antennal lobe, prior to the arrival and targeting of presynaptic olfactory sensory neuron axons (OSN) [42]. Our analysis of AII ACs in bipolarless mice suggest that their major partner, the BCs, are not necessary for AII AC presynaptic process outgrowth. Whether such neurite outgrowth is regulated intrinsically by the AII AC or dependent on cues from non-BCs, remains to be elucidated.

Transient connectivity in a divergent circuit.

Connectivity with “transient” partners has been found to play an important role in both the development and maturation of circuits across the CNS. For example, Cajal-Retzius cells and subplate neurons receive synaptic input from neurons whose final postsynaptic partners have not yet developed. They serve as an intermediate target and undergo apoptosis once the appropriate postsynaptic partners develop [40]. The wide-field amacrine cells postsynaptic to AII ACs likely serve a different role. AII AC lobular appendages do not elaborate until BCs are present, and therefore wide-field amacrine cells do not serve as an intermediate target. Furthermore, the wide-field amacrine cells likely remain in the retina as amacrine cell apoptosis has largely occurred by P7 [28, 43], and amacrine cells with similar morphology and stratification are observed within the mature mouse retina [44]. It seems more likely that the synapses onto but not the wide-field amacrine cells themselves are “transient”. Transient connectivity is not unusual in the CNS [45] or even within the retina [46, 47], and in fact, can play an important role in regulating spontaneous activity during development [48]. For example, cholinergic (or Stage II) retinal waves are spread via reciprocal cholinergic synapses between starburst amacrine cells (SACs)[49, 50]. While these synapses are only present during development, loss of these synapses disrupts eye-specific segregation of retinal ganglion cell projections [51]. AII ACs are known to depolarize during Stage III (glutamatergic) retinal waves [52, 53] and could potentially inhibit their widefield AC partners. Future experiments will be necessary to determine whether AII AC synapses onto widefield amacrine cells are involved in glutamatergic retinal wave generation and propagation.

Cellular strategies shaping synaptic divergence.

Divergence of signals is a common motif throughout the vertebrate retina, including the visual system’s first synapse (from photoreceptors onto bipolar cells). Synaptic divergence of cone photoreceptors to postsynaptic BCs, differs from that of AII ACs in that cone-BC synapses are glutamatergic, whereas AII AC synapses are glycinergic. Our findings here thus reveal for the first time, the developmental strategies engaged in establishing biased synaptic divergence from an inhibitory neuron. Cone-BC synaptogenesis, however, shares some features with AII synaptogenesis. For instance, the mature connectivity patterns associated with different partner types are finalized on separate time-scales. Connectivity between cones and small-field BCs are attained earlier than connections with large-field BCs [12]. We observed such ‘timing’ differences for AII AC circuits, whereby synaptogenesis with BCs increases steadily with age beyond the period when synapses with RGCs have reached their mature number. This increase in synapse formation with BCs is consistent with the increase in membrane capacitance jump between P10 and P25 [34], as well as the marked increase in glycinergic inhibitory postsynaptic currents (IPSCs) recorded in OFF BCs after P9 [54]. Also, synaptogenesis with the major BC partner type, the Type 2 BC, progresses earlier than that of other BC types. We also found that for cones and AII ACs, synapses are made and established with different partner types using different strategies, including circuit remodeling. Like some cone-BC synapses, AII AC synapses also undergo remodeling, whereby some inputs are maintained whereas others are eliminated. But, unlike cone-BC connections, AII ACs eliminate synapses with a specific cell class. Nevertheless, it is evident that divergence from cones and from AII ACs both engage distinct developmental mechanisms acting in parallel.

Our current study underscores the contribution of partner-type specific strategies to create a biased synaptic distribution pattern. In some neural circuits, the connectivity between cells can be regulated in a dependent manner. For example, in the zebrafish, the major presynaptic partner of the H3 horizontal cell (H3 HC) ‘dictates’ H3 HC connectivity with the minor partner (but not vice-versa)[14]. In contrast, AII ACs do not compensate for the loss of major partners by increasing synapses onto surviving partners. However, on average, the lobular appendages were slightly smaller in the bipolarless retina, raising the possibility that there may be functional changes in transmission from AII ACs onto postsynaptic targets in these retinas. Future electrophysiological experiments assessing AII-mediated transmission onto OFF GCs would be informative.

As mentioned earlier, neurons can achieve biased connectivity via preferential synaptogenesis and/or by selective pruning. Previous studies have shown that synaptic activity can be involved in both processes. For example, synaptic transmission from the dominant presynaptic partner can drive preferential synaptogenesis in the H3 horizontal cell circuit [14]. In the mouse sensory-motor system, individual sensory-motor neurons preferentially synapse onto homonymous motor neurons (which drive the same muscle from which the sensory neuron receives input) compared to heteronymous motor neurons. Silencing the sensory-motor neurons, however, leads to a sustained increase in the number of contacts onto heteronymous targets, without altering the number of synapses onto the preferred partner (homonymous targets) [13]. Therefore, activity of the presynaptic neuron can restrict the number of synapses formed onto the non-preferred target. In contrast, it seems unlikely that synaptic drive from rod BCs restricts AII AC connectivity with non-BC targets. In bipolarless mice, which also lack rod BCs, AIIs form the appropriate number of synapses onto non-BC targets. In models of retinal degeneration, loss of light driven rod BC transmission can lead AII ACs to burst spontaneously [55]. While it is not yet known whether AII ACs are spontaneous active, or show altered activity, in the bipolarless mice, we think that it is unlikely that AII transmission critically shapes its pattern of synaptic divergence. This is because GlyRα1 expression on Type 2 BCs is unaffected in mice lacking inhibitory neurotransmission due to knockout of the vesicular inhibitory amino-acid transporter [56]. Finally, because biased synaptic divergence is already apparent well before eye-opening, patterned vision is unlikely responsible.

What mechanisms underlie the preferential increase in synapses onto the Type 2 BCs, which eventually leads to biased synaptic divergence? For BCs connected to AII ACs, the BCs with the greatest amount of axonal overlap with AII AC lobular appendages received more synapses from the AC, suggesting that proximity might be a major factor. This may not be surprising, given that both BC axonal arbors and the AII lobular ‘fields’ tile and are fairly narrow, and thus BC arbors further away from an AII AC will not receive many synapses from that AC. We cannot, however, rule out the influence of guidance cues that direct BC axonal growth towards AII processes or vice versa, or the role of adhesion molecules that might contribute to bipolar cell target specificity resulting in greater connectivity with Type 2 BCs compared to other BC types with overlapping arbors. Indeed, the guidance molecules, Semaphorin5 (A and B) and their receptors (Plexin 1 and 2) have been shown to direct neurite lamination of both AII AC processes and Type 2 BC processes [57]. Because expression of these proteins was disrupted in a non cell-autonomous manner, we cannot as yet distinguish between cues that commonly attract AIIs and Type 2 BC processes, and cues from one cell type directing outgrowth of the other. Nevertheless, it is clear that BCs are not instrumental in causing AII AC neurites to extend because lobular appendages form in the absence of BCs. In the ON-sublamina of the inner plexiform layer, synapses made between Type 6 and Type 7 ON BCs with ON-alpha GCs are in part determined by spatial proximity of the two BC types to the dendrites of these GCs. Loss of Type 6 BCs increases the opportunity and therefore the number of synapses that Type 7 BCs make with the ON-alpha GC [10]. Thus, it would be informative in the future to determine whether other OFF BCs increase their connectivity with the AII AC in the absence of Type 2 BCs. If so, then synaptic divergence from the AII AC may in part be regulated by the relative spatial arrangements of the AII AC and the BC arbors. If not, then there are likely factors that separately regulate connectivity between each type of OFF BC and the AII AC.

STAR METHODS

LEAD CONTACT AND MATERIALS AVAILABILITY

Further information and requests for resources and reagents should be directed to and fulfilled by the Lead Contact, Rachel Wong (wongr2@uw.edu). This study did not generate new unique reagents or mouse lines.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

All procedures were conducted in accordance with University of Washington Institutional Animal Care and Use Committee guidelines. C57BL/6 mice (JAX Stock No. 000664), CDH1-GFP mice [53] and Vsx2-SE knockout (bipolarless) mice [30] of both sexes were used in this study.

METHOD DETAILS

Immunohistochemistry

Mice were deeply anesthetized with Isoflurane (5%), cervically dislocated or decapitated, and enucleated. Retinas were dissected in room temperature oxygenated mouse artificial cerebral spinal fluid (mACSF, pH 7.4) containing (in mM): 119 NaCl, 2.5 KCl, 2.5 CaCl₂, 1.3 MgCl₂, 1 NaH₂PO₄, 11 glucose, and 20 HEPES. Retinas were flat mounted on filter paper (Millipore, HABP013) then fixed for 15 minutes in 4% paraformaldehyde in mACSF. Retinas were rinsed with PBS, 3-10 minute washes, and incubated in blocking solution containing 5% normal donkey serum and 0.5% Triton X-100 overnight at 4°C. The retinas were then incubated with primary antibodies in blocking solution for 3 nights at 4°C. After rinsing in PBS, 3-10 minutes each, and the retinas were incubated in secondary antibodies in PBS solution overnight at 4°C. The tissues were rinsed with PBS prior to mounting on slides with Vectashield for confocal imaging.

Confocal Image Acquisition

Image stacks were acquired on an Olympus FV1000 microscope using a 60x oil-immersion objective with 1.35 NA. Voxel sizes of the acquired images were 0.051-0.051-0.3 μm (x-y-z).

Image Analysis

Image stacks were median filtered in FIJI (NIH). Maximum intensity images were visualized in Amira (Thermo-Fisher). To quantify the volume and number of the lobular appendages, processes of individual AII ACs were traced in 3D using the ‘labelfield’ function (creates binary masks) in Amira. During development, we classified a ‘lobule’ or its ‘precursor’ if the neurite extended laterally from the AII AC cell soma or from the central stalk(s) until the stalks branched into the distal dendrites (see Figure 1C). In the mature animals, such neurites had lobular endings. The binary masks were exported to MATLAB and custom scripts were used for quantification of volume. The same masks were exported to FIJI to manually count the number of lobular appendages using ‘PointPicker’. To determine the total number of output synapses from an AII cell, entire AII AC somata and lobular appendages were masked in Amira using the labelfield function. The threshold for the receptor signal was determined by the image statistics of the channel as measured by FIJI. Thresholds were 6 standard deviations above the mode pixel intensity because the mode pixel intensity is an approximation of “noise” within the channel [59]. This binary mask was multiplied by the GlyRα1 immunolabel (receptor ‘signal’) using the Arithmetic function in Amira, such that only GlyRα1 puncta that overlapped with the lobular appendage mask remained [56]. Each remaining punctum was counted as a single synapse, and puncta were visually confirmed to be apposed to both the AII and the Type 2 BC before they were included in the quantification.

Serial Block-face Electron Microscopy and image analysis

The P11 retina was from a C57BL/6 mouse. The P24 retina is from the control dataset acquired in our previous study [60]. Retinas were dissected as described above, and fixed with 4% glutaraldehyde in 0.1M sodium cacodylate buffer, pH 7.4, at room temperature overnight. The tissue was then processed for SBFSEM using methods similar to the those described previously [61]. Briefly, the tissue was rinsed 5x5 min in 0.1M sodium cacodylate buffer, pH7.4 and incubated in a solution containing 1.5% potassium ferrocyanide and 2% osmium tetroxide (OsO4) in 0.1M cacodylate buffer (0.66% lead in 0.03M aspartic acid, pH 5.5) for 1 hour on ice. After washing, the tissue was placed in a freshly made thiocarbohydrazide solution (0.1g TCH in 10 ml double-distilled H₂O heated to 60°C for 1 h) for 20 min at room temperature (RT). After another rinse at RT, the tissue was incubated in 2% OsO4 for 30 min at RT. The samples were rinsed again and stained en bloc in 1% uranyl acetate overnight at 4°C, washed and stained with Walton’s lead aspartate for 30 min at 60°C. After a final wash, the retinal pieces were dehydrated in a graded ice-cold alcohol series, and placed in propylene oxide at RT for 10 min. Finally, samples were embedded in Durcupan resin.

Image stacks were acquired using a GATAN 3View and Zeiss Sigma scanning electron microscope. Voxel dimensions for micrograph stacks were (x-y-z): 5-5-50 nm for the P24 and P11 blocks. EM micrographs were aligned and stitched using TrakEM2 software (FIJI). AII neuronal processes were traced using an interconnected series of nodes called a “Tree” in TrakEM2. AII ACs and the axon terminals of presumed Type 2 BCs were segmented and visualized using the AreaTree function, whereby an annotator traced the outline of the cell in each plane manually. All other processes were visualized using the Treeline function in TrakEM2 whereby the nodes were assigned a specific radius for clear visualization revealing the “skeleton” of the cell. Surface overlap between BCs and AII AC lobular appendages was determined by tracing the appositions of a single BC-AII AC pair in each plane of the EM stack in TrakEM using the “brush” tool at a set width (Figure S1B). The sum of lengths of all the traced lines for a given BC-AII AC pair were used as a proxy for the amount of overlap. Since all pairs were within the same EM block, the lack of interpolation between planes was consistent and therefore the measurements could be used as relative measures of overlap.

QUANTIFICATION AND STATISTICAL ANALYSIS

All pairwise comparisons are Wilcoxon tests, except for the use of a one-way Anova to determine if there are significant differences in lobular appendage volume with age (Figure 1) and a Kolmogorov-Smirnov test to determine whether there were significant differences in the distribution of synapses onto ganglion cells between the two ages (Section: Distribution of synapses across AII AC- RGC pairs is invariant with age). Statistics were performed in Matlab. Number of animals and cells are provided in figure legends.

DATA AND CODE AVAILABILITY

The EM datasets supporting the current study have not been deposited at a public repository because of the large size of the datasets and use in ongoing studies, but are available from the corresponding author upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Chicken polyclonal anti-GFP	Abcam	Cat#ab13970, RRID: AB_300798
Mouse monoclonal anti-GlyRα1	Synaptic Systems	Cat#146111, RRID: AB_887723
Mouse monoclonal anti-Synaptotagmin-2	Zebrafish International Resource Center	Cat#znp-1, RRID: AB_10013783
Donkey anti-chicken Dylight 488	Jackson Immunoresearch	Cat#703-545-155, RRID: AB_2340376
Goat anti-mouse IgG1 Alexa 568	Invitrogen	Cat#A-21124, RRID: AB_141611
Goat anti-mouse IgG2a Alexa 647	Jackson Immunoresearch	Cat#115-605-206, RRID: AB_2338917
Experimental Models: Organisms/Strains
Mouse: C57BL/6	The Jackson Laboratory	JAX Stock No: 000664
Mouse: CDH1-GFP	Joshua Singer [53]	RRID: MMRRC_011775-UCD
Mouse: Vsx2 CRC-SE KO	Michael Dyer [30]	N/A
Software and Algorithms
ImageJ	NIH	https://imagej.nih.gov/ij/ RRID: SCR_003070
TrakEM2 (ImageJ)	NIH	https://imagej.net/TrakEM2
Amira	ThermoFisher Scientific	https://www.fei.com/software/amira/ RRID: SCR_014305
MATLAB	Mathworks	www.mathworks.com RRID: SCR_001622

Highlights.

1. Mouse AII amacrine cells differentiate input neurites before output processes.

2. Partner-type specific connection strategies create biased synaptic divergence.

3. Partner-type specific synaptogenesis proceeds at different rates.

4. Synaptogenesis with distinct partners is regulated independently.