Gain control by sparse, ultra-slow glycinergic synapses

SUMMARY

In the retina, ON starburst amacrine cells (SACs) play a crucial role in the direction-selective circuit, but the sources of inhibition that shape their response properties remain unclear. Previous studies demonstrate that ~95% of their inhibitory synapses are GABAergic, yet we find that the light-evoked inhibitory currents measured in SACs are predominantly glycinergic. Glycinergic inhibition is extremely slow, relying on non-canonical glycine receptors containing α4 subunits, and is driven by both the ON and OFF retinal pathways. These attributes enable glycine inputs to summate and effectively control the output gain of SACs, expanding the range over which they compute direction. Serial electron microscopic reconstructions reveal three specific types of ON and OFF narrow-field amacrine cells as the presumptive sources of glycinergic inhibition. Together, these results establish an unexpected role for specific glycinergic amacrine cells in the retinal computation of stimulus direction by SACs.

In brief

The role of glycine inhibition in controlling starburst function has often been overlooked, owing to the weak anatomical representation of glycinergic synapses in the circuit. Jain et al. demonstrate that the sparse population of glycinergic synapses uses specialized Glyα4 receptors to mediate a slow inhibition, which powerfully controls starburst output gain and direction selectivity.

Graphical Abstract

INTRODUCTION

From as early as the work of Ramón y Cajal, it has been clear that, in many brain regions, inhibitory neurons constitute a more heterogeneous collection of types than do the excitatory neurons (Maccaferri and Lacaille, 2003). Advances in molecular, neurochemical, and/or anatomical techniques have led to an explosion in the number of known interneuron types (Markram et al., 2004). Convergent evidence from such methods increasingly permits elaborate classification schemes, in which diverse inhibitory neurons are grouped into distinct, biologically meaningful types (Ascoli et al., 2008; DeFelipe et al., 2013). Understanding the implications of such diversity requires a more complete understanding of their specific roles in intact and functioning neural circuits.

The retina is a particularly favorable region of the central nervous system for such an effort. Although there are more than 60 types of retinal amacrine cells, the function of most is poorly understood (MacNeil and Masland, 1998; Masland, 2001, 2012; Yan et al., 2020). One exception is the starburst amacrine cell (SAC), which plays important, well-defined roles in retinal development and visual processing. In the mature retina, the radiating dendrites of starbursts appear to be the first elements where direction selectivity is generated. Directional information computed in SAC dendrites is relayed through direction-selective ganglion cells (DSGCs) to subcortical areas, where it contributes to reflexive behaviors (e.g., image stabilization through optokinetic nystagmus), and cortical visual areas, where it contributes to motion perception ((Rasmussen and Yonehara, 2020); reviewed by Murphy-Baum et al., 2021). Silencing SACs by pharmacological, genetic, or pharmacogenetic means disrupts direction selectivity in downstream DSGCs and in the optokinetic reflex, underscoring the critical role for SACs in this computation (Taylor and Smith, 2012; Murphy-Baum et al., 2021).

Recent studies have made great strides in characterizing the anatomy and function of synaptic inputs to SACs, providing valuable insights into how they compute direction. For example, serial electron microscopy (SEM) reconstructions show that SACs make extensive inhibitory connections with each other, accounting for the vast majority of the inhibitory inputs (Briggman et al., 2011; Ding et al., 2016). Moreover, the spacing of the SAC somas approximates the length of their dendrites, resulting in a circuit in which dendrites preferring opposite directions of motion inhibit each other at regions near the soma, away from their distal output sites (Ding et al., 2016). “Anti-parallel” inhibitory connections enhance SAC directional tuning by suppressing dendritic responses to centripetal motion (i.e., dendritic tip to soma; Chen et al., 2016; Ding et al., 2016; Lee and Zhou, 2006; Munch and Werblin, 2006). The remaining few presumptive GABAergic synapses to SACs derive from wide-field amacrine cells (WACs) (Briggman et al., 2011; Ding et al., 2016). These mediate a long-range suppression that impart size selectivity and context specificity to the SAC light responses (Hoggarth et al., 2015; Huang et al., 2019).

Anatomical studies also reveal that a small fraction (<5%) of the inhibitory connections to ON SACs derive from narrow-field amacrine cells (NACs) (Briggman et al., 2011; Ding et al., 2016), which are glycinergic (Heinze et al., 2007; Weiss et al., 2008). However, electrophysiological studies provide an inconsistent picture of a role for glycinergic inhibition in ON SACs. Blocking glycine receptors throughout the retina has little or no effect on the spontaneous or light-evoked inhibitory postsynaptic currents (IPSCs) measured in SACs (Lee and Zhou, 2006; Majumdar et al., 2009). Similarly, blocking all glycinergic inhibition does not appear to affect the direction selectivity of SAC dendrites (Hausselt et al., 2007). By contrast, results from other studies suggest a more prominent role for glycinergic inhibition of SACs. Local application of glycine evokes large currents in ON, but not OFF, SACs (Ishii and Kaneda, 2014; Majumdar et al., 2009). Moreover, in classic bulk-release studies, blockade of glycine receptors (GlyRs) greatly increased light-evoked acetylcholine (ACh) release (a measure of SAC output; Cunningham and Neal, 1983), although this could be mediated by network effects. Taken together, in the current view, it is believed that glycinergic inhibition plays a minor role in the direction-selective (DS) circuit and is often not even considered (Wei, 2018; Murphy-Baum et al., 2021).

To probe more deeply the specific contributions of glycine to SAC function, we assessed the functional properties of the inhibitory inputs to SACs in the mouse retinas in vitro. We assessed SAC function by patch recording their synaptic currents, imaging their intracellular Ca²⁺, and monitoring their synaptic output with a two-photon ACh sensor. We reassessed the circuit in a variety of mouse knockouts (KOs), some with deletions of genes coding for specific glycine-receptor subunits (including a novel Glra4 KO), others are known to disrupt SAC-to-SAC GABAergic inhibition. We also document the anatomical properties of NACs and their synaptic associations with SACs by mining the same publicly available SEM dataset used in the earlier analysis of NAC inputs to SACs (Ding et al., 2016). Together, our findings indicate that three types of NACs provide sparse glycinergic inhibition throughout the SAC arbor. Using GlyRs that mostly include the GlyRα4 subunit, these relatively sparse NAC inputs unexpectedly dominate the inhibitory input to ON SACs and modulate the processing of moving stimuli according to the recent history of ongoing activity.

RESULTS

ON SACs receive powerful glycinergic inputs

We targeted ON SACs for patch recordings using two-photon imaging of td-Tomato fluorescence in ChAT-Cre × Ai9 mice. Voltage-clamped ON SACs exhibited robust light-evoked IPSCs (V_HOLD ~ 0 mV, the excitatory reversal potential) in response to positive contrast spots of light (diameter 100 μm). At the onset of the light stimulus (Figure 1A, top trace), IPSCs exhibited a double peak. A fast initial transient lasting only ~100 ms was followed by a slow secondary rise, which peaked a full second post-onset. The secondary phase decayed slowly, with some inhibition still present at the end of the 2-s stimulus. The IPSC evoked at light offset resembled the slow component of the ON response but completely lacked the rapid transient component (Figure 1A).

Figure 1. Glycine, not GABA receptors, mediates the major fraction of the inhibitory charge to ON SACs.

(A) IPSCs recorded from an ON SAC (at ~0 mV) evoked by small stationary spots (2 s:100 μm diameter) in control (black), in the presence of GlyR antagonist (1 μM strychnine, red), and following the addition of the GABA_AR antagonist, gabazine (gray). Inset shows the transient response in the boxed region at higher magnification.

(B) Average ON (top and middle) and OFF (bottom) inhibitory responses measured across six cells (black, mean ± SEM; gray, individual cells) in three conditions as shown in (A). The middle panel plots the same data as the top panel and is magnified to highlight the significant reduction of ON inhibitory charge in gabazine (*p < 0.05; t test).

(C) ON SAC IPSCs measured in Gabra2 KO mouse (GABA_A α2 receptors are knocked out in SACs) and Vgat KO mouse (vesicular GABA transporter is knocked out from SACs) lack transient ON component. IPSCs in these mice are completely blocked by strychnine (left-top and middle). By contrast, SAC IPSCs in Glra4 KO mice mouse exhibited an ON transient response (but not sustained responses), which are blocked by gabazine. Right ON and OFF responses measured for a population of cells in different KO mice are shown (mean ± SEM; *p < 0.001).

(D) ON SAC IPSCs evoked by stationary spots of increasing diameter. (Bottom) Normalized ON transient (ON-T), ON sustained (ON-S), and OFF responses are plotted as a function of spot size.

(E) IPSCs evoked by a moving spot typically have multiple components, each dominated by a single source (WAC, SAC, and NAC) with sensitivity to different drugs (as indicated). EPSC (V_hold ~ −60 mV, last trace) measured in the control condition provides an indication of the excitatory receptive field for reference. The average responses (mean ± SEM) for three inhibitory components in different drugs quantified across a population of ON SACs are shown.

Bath application of a GlyR antagonist (1 μM strychnine; Figure 1A, middle trace) significantly suppressed the sustained component of the ON IPSCs (p < 0.01; t test), as observed previously in rabbit ON SACs (Lee and Zhou, 2006). Strychnine also virtually eliminated the OFF IPSCs. The residual transient ON component was abolished upon additional application of the GABA_A receptor antagonist (5 μM SR-95331: gabazine; Figure 1A, bottom trace). These results suggest that the transient inhibition at light onset (ON) is mediated by GABAergic SACs, whereas the sustained ON/OFF IPSCs are mediated by glycinergic NACs. Though GABA release from WACs could potentially contribute to the transient IPSC, inputs from WACs are not expected to be evoked with the small stimuli used here. Here out, for simplicity, we will call the transient, GABA-mediated ON response the SAC component and the more sustained ON and OFF IPSCs the NAC component. A surprising aspect of these results is that, at least in this simple light-step experiment, most of the inhibitory charge in ON SACs is carried by GlyRs. This is unexpected, given strong evidence that the vast majority of inhibitory synaptic inputs to ON SACs are GABAergic (Ding et al., 2016).

To test the hypothesis that mutual SAC inhibition mediates the transient SAC-mediated component of the inhibitory light response, we repeated these experiments in two lines of genetically modified mice in which mutual SAC-SAC GABAergic inhibition is selectively suppressed in distinct ways. In the first, SAC-SAC GABAergic inhibition was compromised postsynaptically by conditional deletion of GABA_A α2 subunits receptor expression in SACs (Gabra2^fl/fl × ChAT-Cre; Chen et al., 2016). In the second, GABA release from SACs was reduced by deletion of the GABA vesicular transporter (Vgat^fl/fl × ChAT-Cre; Bleckert et al., 2018; Pei et al., 2015).

In both mouse lines in which GABAergic mutual inhibition was genetically disrupted, SACs lacked the transient IPSC component observed at the onset of the stimulus in wild-type SACs (compare Figure 1C, top two panels, with Figure 1A, top). By contrast, sustained ON and OFF IPSCs persist, matching the NAC component in wild-type (WT) SACs in amplitude, kinetics, and pharmacology. These results confirm that GlyRs mediate a slow, sustained ON/OFF postsynaptic inhibition to ON SACs, while GABA_ARs mediate a fast, transient ON inhibition.

Past studies suggest that GlyRα4 subunits are likely to be a key component of the receptors mediating glycinergic inhibition of ON SACs. Immunolabeling shows that GlyRα4 subunits costratify with ON SAC dendrites (Figure S2A). Miniature IPSCs in ON SACs appear normal in three mouse lines, each lacking one of the other three glycine-receptor alpha subunits (KOs of GlyRα1, α2, or α3; Heinze et al., 2007; Majumdar et al., 2009). To test directly whether GlyRα4 subunits contribute to glycinergic inhibition of SACs, we generated a Glra4 KO mouse and validated the absence of GlyRα4 expression using immunohistochemistry (Figures S1 and S2).

Indeed, Glra4 KO SACs lacked sustained ON and OFF inhibitory currents and only displayed the fast transient ON response (Figure 1C, bottom panel), similar to WT SACs when GlyRs are blocked by strychnine (6 ± 2 pC; p > 0.1; Figure 1A, middle red trace). As expected for a response based on mutual SAC inhibition, the transient ON SAC component was completely blocked by application of gabazine (Figure 1C, bottom panel, red trace). This result demonstrates that GABAergic SAC-SAC inhibition remains despite the absence of GlyRα4 expression during development. Taken together, these results support and extend the proposal made more than a decade ago that α4-containing GlyRs mediate glycinergic inhibition in SACs (Heinze et al., 2007; Majumdar et al., 2009).

Next, we examined the spatial extent of glycinergic inhibition to ON SACs using spot stimuli of different diameters (Figure 1D). Both the ON and OFF sustained glycinergic IPSCs had a clear center summation and surround antagonism (Figure 1D). However, the ON transient GABAergic IPSCs only had center summation, as their responses did not vary across large spots. This may reflect complementary decreases in SAC-SAC inhibition and increases in WAC inhibition as a function of increasing spot size (Hoggarth et al., 2015).

Moving stimuli (200-μm-diameter spot; 1,000 μm/s) evoked a characteristic sequence of inhibitory currents, which we interpret as reflecting individually, and sequentially, the contributions of WACs, SACs, and NACs (Figure 1E; green, gray, and blue rectangles, respectively). The first IPSC component (green rectangle) was evoked as the stimulus approached to within 500–700 μm of the dendritic field of the recorded SAC, and it was eliminated when action potentials were blocked (tetrodotoxin [TTX], 1 μM; Figure 1E; compare second and third traces). This result is consistent with the idea that this initial component was mediated by wide-field spiking amacrine cells. However, the WAC-mediated component to moving spot stimuli was evoked inconsistently, so we did not analyze it further.

The second IPSC component (gray rectangle) was transient, TTX-insensitive (Figure 1E, third trace), gabazine-sensitive (fourth trace), and synchronized with the excitatory postsynaptic current (EPSC) (Figure 1E; V_HOLD −60 mV; bottom trace). This component represents inhibition from neighboring SACs. The third IPSC component (blue rectangle) was sustained and emerged only after the termination of the SAC’s excitatory response, presumably as the spot left the receptive field. This component was insensitive to gabazine and TTX but was blocked by the GlyR antagonist strychnine (Figure 1E, fourth trace), suggesting that it arises from NACs. These slow responses to moving stimuli presumably represent a blend of ON and OFF components, but their individual contributions cannot be separated in these recordings. One inference to be drawn from these findings is that, though glycine evokes more total inhibitory charge compared with GABA for flashed stimuli, it presumably has little impact on the SAC response to moving spots, since its slow kinetics dictate that it lags SAC excitation. Under these conditions, it is SAC-to-SAC GABA inhibition, which is faster and coincident with the excitatory drive, that appears to be important for controlling starburst function (Ding et al., 2016).

Glycinergic unitary events exhibit slow kinetics

To better understand the properties of glycinergic transmission, we next examined the kinetics of spontaneous IPSCs. Under control conditions, spontaneous IPSCs (sIPSCs) varied widely in their kinetics and peak amplitudes (Figure 2A). Glycinergic sIPSCs, isolated by addition of gabazine, were slow to rise and extremely slow to decay (τ_rise = 2.8 ± 1.2 ms; τ_decay = 64 ± 34 ms; n = 146 events from 5 cells; Figure 2B). They are among the slowest sIPSCs observed in the central nervous system to date (Majumdar et al., 2009). Slow glycinergic sIPSCs were relatively small in amplitude (12 ± 4 pA) and may contribute to a tonic current (see Figure 3C). By contrast, GABAergic sIPSCs, isolated in the presence of strychnine, were significantly larger (p < 0.01) and faster (p < 0.01; compare Figures 2B and 2C). The decay kinetics of sIPSCs measured in strychnine were like those recorded in ON SACs of Glra4 KO mice (p > 0.05; compare Figures 2B and 2D). Gabazine completely blocked the fast sIPSCs in five of seven ON SACs. In the remaining two SACs, the residual sIPSCs had intermediate kinetics, suggesting that other faster types of GlyRs may also contribute to glycinergic modulation of SACs. Overall, though, transmission at unitary glycinergic synapses to ON SACs is extremely slow, presumably due to the unique biophysical properties of GlyRα4 subunits (see Discussion).

Figure 2. GlyRα4 receptors in SACs mediate slow and delayed sIPSCs.

(A) sIPSCs measured from ON SACs in wild-type mice at 0 mV in control (left panel). Normalized average fast (blue) and slow (red) sIPSCs are shown (middle panel). Frequency distribution of rise, decay time, and amplitude of the population events is shown (right panels; n = 5 SACs).

(B and C) Same as (A) but in gabazine (B) and strychnine (C) in wild-type SACs.

(D) Same as (A) and (B) except SAC recordings were made in Glra4 KO retina.

Figure 3. Glycinergic inhibition suppresses SAC output under sustained visual simulation.

(A) ON SAC IPSCs and EPSCs under control (black) and strychnine (red) evoked by drifting sinusoidal gratings (SF 0.1 cycles/degree; TF 1 Hz, indicated on top; also see Figure S3 for responses evoked by 0.01 cycles/degree) presented through a 500-μm-diameter aperture. Dark traces are the average of three trials shown in a lighter shade.

(B) Transient IPSC evoked at each cycle of the stimulus, in control (black) and strychnine (red). Note the slow, glycinergic component was digitally subtracted to show the transient SAC components more clearly.

(C) Net change in the baseline holding current after the application of strychnine.

(D) Strychnine has distinct effects on the NAC- (slow IPSC; left) and SAC-mediated IPSCs (middle) and EPSCs (right) at each of the five cycles of the stimulus (control, black; strychnine, red). Data are represented as mean ± SEM.

NACs control SAC output during sustained stimulation

The impact of NAC inhibition on SAC output was more pronounced when a drifting grating replaced the spot stimulus. To minimize WAC-mediated inhibition from the surround, the grating was presented within an aperture only slightly larger than the SAC dendritic field (500 μm diameter; spatial frequency [SF] 0.1 cycles/degree; temporal frequency [TF] 1 Hz). Responses to such local grating stimuli were dominated by slow, sustained IPSCs (Figures 3A and 3D, black traces). Superimposed upon these slow currents were more transient ones that were stimulus locked at the fundamental frequency of the grating. The properties of the transient currents are more easily appreciated after digital subtraction of the sustained phase (see STAR Methods; Figure 3B). The transient IPSCs were suppressed by ~70% after the first cycle of the stimulus (Figures 3B and 3D).

Bath application of strychnine reduced a small outward current that is observed at baseline in the absence of visual stimulation (Figure 3C), suggesting that, even at rest, NACs help to set the inhibitory tone of ON SACs. GlyR blockade also reduced the sustained phase of the light-evoked IPSCs (Figure 3A). The absence of glycinergic input greatly enhanced the amplitude of the transient component of the inhibitory current, indicating that NAC inhibition strongly suppresses SAC output under control conditions (Figures 3A and 3D). The response to the first cycle of the moving grating appeared to be less affected by NAC input (Figures 3B and 3D). This is consistent with the evidence that glycinergic inhibition of SACs is slow to activate (Figure 1E). In the presence of strychnine, the transient, SAC-mediated component did not depress strongly over successive grating cycles (Figure 3B, red), as it did in control.

Similar effects of strychnine were observed for gratings of lower spatial frequency (Figure S3). Blocking GlyRs had no significant effect on the excitatory light-evoked currents in ON SAC (measured at V_HOLD ~ −60 mV; p > 0.05). This indicated that glycinergic modulation of SAC excitability occurs almost entirely postsynaptically, through direct NAC-to-SAC synapses (Figures 3A and 3D), rather than presynaptically on the bipolar terminals contacting SACs.

One caveat in voltage-clamp experiments is that the efficacy of the voltage clamp may change when sustained inhibition is eliminated. This would lead to underestimation of the relative contribution of the SAC component and exaggerate the degree to which SAC output is controlled by NACs. To independently verify the conclusions of our electrophysiological measurements, next, we optically monitored the light-evoked release of ACh from SACs, using a genetically encoded fluorescent probe (G-protein-coupled receptor-activation-based sensor GRAB ACh 3.0 or simply ACh 3.0, kindly provided by Dr. Yulong Li, Peking University). The probe, packaged in Cre-dependent viral vector (adeno-associated virus [AAV]), was expressed in SACs and/or DSGC dendrites using transgenic reporter lines in which these have been shown to be labeled (ChAT-Cre, CART-Cre, and/or Oxtr-Cre; (Kay et al., 2011; Park et al., 2014; Sethuramanujam et al., 2021; Figure S4). Two weeks after intravitreal injection, similar ACh responses to drifting gratings could be observed in SAC/DSGC dendrites (Figures 4A–4C). Brief exposure to strychnine doubled the ACh3.0 response (Figures 4B–4D), and these effects were reversed when the drug was washed out. Augmentation of the grating response was evident even for the first cycle of the grating. This stands in contrast to the observations on the SAC-mediated fast inhibitory conductance, where glycinergic inhibition had little impact on the response to the first grating cycle. This is presumably because laser scanning prior to the presentation of the gratings activated NACs and thus produced tonic glycinergic currents. Together with results from our electrophysiological experiments, these data demonstrate that glycinergic inhibition suppresses the SAC output during sustained visual stimulation.

Figure 4. Glycinergic modulation of starburst cholinergic output.

(A) 2P image stack showing the expression of ACh3.0 in dendrites of starburst and DSGC dendrites labeled in the Oxtr-Cre.

(B) The effects of a brief bath application of 1 μM strychnine on the ACh3.0 responses evoked by drifting gratings (SF 0.1 cycles/degree; TF 1 Hz; duration 5 s; presented at 1-min intervals). The line on top indicates the duration for which the tissue was exposed to strychnine.

(C) Selected ACh3.0 responses shown in (B), under control (left), strychnine (middle), and recovery from strychnine (wash; right).

(D) Relative change in ACh3.0 fluorescence under the conditions shown in (C) in six fields of view from a variety of transgenic retinas in which the sensor was expressed in starburst and/or DSGCs (see text for details; *p = 0.0053). Data are represented as mean ± SEM.

Glycinergic inhibition ensures that SACs maintain direction selectivity throughout their response range

While our experiments clearly demonstrate that NACs control SAC gain, they do not provide a clear indication of whether they affect SAC DS. To address this question, we examined the effect of strychnine on the DS IPSCs measured in superior-motion-encoding DSGCs (labeled in the Hb9-GFP mouse line), which receive their inhibitory input primarily from SACs (note, other DSGCs may receive both SAC and non-SAC input; (Park et al., 2015); S.S., unpublished data). SAC DS output was quantified by computing the ratio of the peak IPSCs evoked by gratings, drifting in the DSGC’s “preferred” and “null” directions. We found that blocking GlyRs increased inhibitory responses in both the null direction and preferred direction. This effect was more prominent on weaker IPSCs evoked at non-optimal spatiotemporal frequencies (p < 0.01; t test; Figure 5B versus 5D). The parallel increase in preferred and null-direction IPSCs resulted in a weaker DS. For example, in gratings with low spatial frequency (0.01 cycles/degree) where ON inhibitory currents are weak (Figures 5C and 5D), application of strychnine increased responses in both null (control 25 ± 10 pA and strychnine 59 ± 12 pA; n = 11; p < 0.001) and preferred directions (control 8 ± 4 pA and strychnine 34 ± 7 pA; n = 11; p < 0.005). This resulted in a significant loss of directional inhibition (direction-selective index [DSI]: control 0.73 ± 0.08 and strychnine 0.28 ± 0.07; p < 0.005; Figures 5C and 5D). By contrast, the OFF IPSCs in DSGCs remained unaltered in both strength and directionality following GlyR blockade.

Figure 5. Glycinergic inhibition affects SAC DS output.

(A) IPSCs measured in superior coding DSGCs evoked by gratings (SF 0.1 cycles/degree, TF 1 Hz) drifting in the null (top) or preferred directions (bottom; control, black; strychnine, red). For high-frequency gratings, the ON and OFF components of IPSCs could not be separated.

(B) Average currents evoked by preferred and null stimuli and the associated direction selectivity index (DSI = (PD − ND)/(PD + ND)), measured across the population (n = 11).

(C) Same as (A), except gratings were of lower spatial frequency (0.01 cycles/degree). Note the weak ON responses in control (black) and increase in strychnine for both null and preferred direction stimuli.

(D) Average current and DSI during the ON responses (n = 11).

(E) Same as (D), except for OFF IPSCs. Note that, on average, the OFF IPSCs did not change significantly following strychnine application. Data are represented as mean ± SEM.

For gratings with high spatial frequency (0.1 cycles/degree), the changes in DS were less dramatic, as null inhibition was still ~4× larger than the preferred response (DSI control 0.86 ± 0.07; strychnine 0.58 ± 0.07; p < 0.005; Figures 5A and 5B). Although the ON and OFF components of inhibitory currents could not be distinguished at high spatial frequencies, the observed effects are likely to be mediated at the level of ON SACs because strychnine did not affect the OFF pathway.

Next, we directly evaluated the impact of NAC inhibition on SAC DS using two-photon Ca²⁺ imaging (Figure 6A). We tested the effects of strychnine on the dendritic Ca²⁺ responses evoked by large drifting gratings (0.01 cycles/degree), where their effects are maximum. Consistent with previous reports (Euler et al., 2002; Hausselt et al., 2007), stronger Ca²⁺ responses were observed during centrifugal motion (soma to dendrite) compared with centripetal motion (Figures 6B, 6C, and 6F). However, the application of strychnine strengthened responses to centripetal motion, significantly decreasing the DS of Ca²⁺ responses in SAC varicosities (Figures 6B, 6C, and 6F). This effect was more prominent for grating stimuli. When DS of Ca²⁺ responses was probed with a moving spot, blocking GlyRs had no significant effect (Figures 6D–6F).

Figure 6. Glycine inhibition is required for the maintenance of DS in SAC dendrites during sustained activity.

(A) Two-photon image stack of ON SAC dendrites filled with Oregon green BAPTA-1 (Ca²⁺ indicator dye).

(B) Change in fluorescence measured at a single distal varicosity (yellow box in A) evoked by sinusoidal gratings, moving in the preferred (centrifugal) and null (centripetal) directions, under control (black) and in the presence of strychnine (red). Thicker line in each panel represents the average response of the trials shown.

(C) The ΔF/F (F1 component, averaged over three trials) measured in individual regions of interest (ROIs) for responses evoked in the null direction are plotted against those measured in the preferred direction (Pref), under control (left) and strychnine (right; n = 95 ROIs from 6 ON SACs).

(D) Same as (B), except for a 200-μm moving spot.

(E) Change in peak fluorescence for a 200-μm spot moving in preferred and null direction, plotted against each other. Data are pooled from 71 ROIs across six cells.

(F) DSI calculated for gratings and spot under control and strychnine. Note the loss of directional Ca²⁺ responses for gratings (*p < 0.01; t test), but not for spots (ns, not significant; p > 0.05; t test). Data are represented as mean ± SEM.

(G) Calcium influx at a distal varicosity of a model ON SAC for a drifting grating in the presence (black) and absence (red) of sustained glycine conductance. Note the directional Ca²⁺ response for the first cycle in absence of glycine inhibition (arrow).

To better understand how glycine affects SAC DS, we extended the computational models previously used to explore SAC-SAC interactions (Ding et al., 2016) to include the NAC-SAC circuitry. In this model, non-linear processing of inputs by regenerative voltage-dependent Ca²⁺ channels plays a key role in amplifying DS within a restricted range of SAC EPSP amplitudes (Hausselt et al., 2007; Tukker et al., 2004). A sinewave grating drifting at 2 Hz was shown to all the cells in the 7-SAC model (Figure 6G, top). The stimulus turned on at 0.5 s, started drifting right for approximately five cycles, and stopped at 3 s. The model was reset back to the original equilibrium state, and the stimulus was again turned on and started drifting left at 3.5 s for another approximately five cycles. The glycine inputs and Ca²⁺ responses in the central SAC, which is shown in Figure 6G, were delayed from the stimulus start at 0.5 s because the grating peak took ~250 ms to drift to the bipolar cells innervating the distal dendrite. In this model, rightward motion evoked large excitatory postsynaptic potentials (EPSPs) that initiated regenerative Ca²⁺ responses. The glycine-mediated inhibition shunted the SAC dendrites so that the smaller EPSPs from leftward motion did not evoke a regenerative Ca²⁺ response. When the model was run with glycinergic input removed from all the NACs, EPSPs from both directions initiated regenerative events. The lack of glycinergic inhibition allowed the stimulus to further depolarize the neighboring SACs, increasing their GABAergic inhibition of the central SAC. However, this effect was more than offset by the lack of the relatively strong glycinergic inhibition. Thus, recruiting a sustained glycinergic inhibition from an array of NACs limited the depolarization of the distal SAC dendrites and kept the response within the correct operating range, enabling SACs to produce robust DS (~0.9; Figure 6G), as we observed experimentally.

Three types of narrow-field amacrine cells synapse upon ON SACs

In the retina, glycinergic inhibition arises from narrow-field amacrine cells (Heinze et al., 2007; Weiss et al., 2008). A comprehensive survey of mouse retinal neuronal types using SEM reconstructions identified 10 types of NACs (Helmstaedter et al., 2013), some of which appear to contact SACs (Ding et al., 2016). Exploiting the previous SEM dataset of Ding et al. (2016), we reconstructed NACs that made synaptic contacts onto ON SACs and recovered three distinct populations (Figure 7). We also reconstructed other types of NACs in the same volume and scrutinized their connectivity but failed to identify any contacts with ON SACs. A survey of NAC types and their relationship to the types defined by Helmstaedter et al. (2013) is provided in Figure S5.

Figure 7. Sources and distribution of presumptive glycinergic (NAC) synapses onto SACs as revealed by serial electron microscopic reconstructions.

(A) En face views showing where NACs make synapses (dots) onto dendritic arbors of ON and OFF SACs. These two composite views were assembled by offsetting individual SAC processes and their synapses in the retinal plane so as to register their somas. Dots showing presumptive glycinergic synapses onto these SACs are color-coded by NAC type (H19, H22, or H23 amacrine cells). H17 and H19 cells or just H19 for simplicity (red dots) are by far the dominant source of presumed glycinergic NAC input to SACs. ON SACs (left) receive much more NAC input than do OFF SACs (right). These cells were analyzed in a previous SBEM data set (volume k0725; (Ding et al., 2016)).

(B–D) Anatomical properties of the three NAC types that synapse onto SACs. (B) The H19 amacrine cell type (n = 11); (C) the H23 type, also known as the Muller-coupled amacrine cell (MAC) ((Grimes et al., 2021); n = 6); and (D) the H22 amacrine type (n = 4) are shown. The left half of each panel shows an en face view of all cells of that type reconstructed from the k0725 volume (shown to scale and in their proper positions, though volume boundaries have been cropped). Mosaics of each type are incomplete here, but nearly complete ones generated for the same types by Helmstaedter et al. (2013) are reproduced in Figure S6, along with other comparisons with our data. The right half of each panel (B–D) shows the colored dendritic profiles of cells of each type in the vertical (side) view. The gray bands represent the ON and OFF SAC plexuses as reconstructed in this volume. All cells are shown from the same viewpoint and in their original positions. H19 cells (B) were arbitrarily divided into two groups and shown one above the other to provide a clearer view of individual arbors. Each dot overlaid on these dendritic profiles marks site of synaptic contact onto a SAC. Dots follow the same color scheme as for those in (A). The far right of each panel provides quantitative data for the type, plotting the relative density of dendritic processes (actually, the density of skeleton nodes) as a function of depth in the inner plexiform layer oriented and scaled vertically to match the side views of the cells to the left. Data from this study (red curves) are overlaid on similar plots for the corresponding type in Helmstaedter et al. (2013) and for the two SAC plexuses (paired black curves). H22 cells and H23/MAC amacrine cells arborize almost entirely in the ON sublayer and get ON bipolar input (Table S1) and are thus ON NACs. H19 cells span the ON and OFF SAC plexuses, but their bipolar input reveals them as OFF NACs (Table S1). Note, besides SACs, these NACs make stereotypical connections with other third-order neurons (Table S1).

By far the largest source of these presumptive glycinergic synapses (n = 92) was a narrow-field type that we will call H19, after the type 19 amacrine cell identified by Helmstaedter and colleagues (Helmstaedter et al., 2013). These NACs form a single mosaic, consistent with the idea that they belong to single type (Figure 7B). The H19s had compact profusely branching arbors 30 ± 5 μm in diameter (n = 8), with processes stratifying mostly within and between the ON and OFF starburst plexuses (Figure 7B, right). The SAC, however, is only one of many postsynaptic targets of H19 cells (Table S1).

To determine whether H19s drive ON or OFF inhibition to SACs, we examined whether they received their input from ON or OFF bipolar cells, which utilize “ribbon” synapses. Starting from their ribbon contacts, we reconstructed large numbers of bipolar cells synapsing on H19 cells (n = 223). These were derived overwhelmingly from OFF cone bipolar cells (n = 209), accounting for 94% of its ribbon input. Type 3b OFF bipolar cells supply by far most of the excitatory synaptic contacts onto the H19 cell, accounting for 85% of the OFF bipolar input (for more details, see Table S1). Thus, the H19 is effectively an OFF NAC. The other two SAC-contacting NACs are ON cells (see below), so H19 cells presumably mediate virtually all the glycinergic inhibition evoked in SACs by dimming stimuli.

The second type of NAC making contacts to ON SACs closely matches the type 22 amacrine cell (H22; Helmstaedter et al., 2013). Dendritic fields of H22 cells, though clearly “narrow field” (40 ± 5 μm; n = 3), were slightly larger than H19 cells. Their processes arborized mainly in the ON sublayer of the inner plexiform layer (IPL), especially within and near the ON SAC plexus (Figure 7C, right), although their targets include diverse non-SAC amacrine cells (Table S1). Our sample is too sparse to demonstrate a tiled mosaic, but the denser reconstructions of the same type certainly do so (Helmstaedter et al., 2013; Figure S5). The excitatory ribbon synaptic input to H22 derived almost exclusively from ON cone bipolar cells (Table S1). Thus, H22 is expected to drive glycinergic inputs to ON SACs almost exclusively during light increments.

The third and last NAC type among those making significant contacts onto ON SACs (Figure 7B) corresponds to type 23 of Helmstaedter et al. (2013), which is an ON NAC that was recently shown to be electrically coupled to Muller glia (and thus referred to as Muller-coupled amacrine cells [MACs]: (Grimes et al., 2021)). Contacts between H23 cells and ON SACs appeared to be rare, however (n = 8 of 106 total synapses: 8%). Several virtually complete skeletons of this type were carefully scanned for contacts onto ON SACs without finding any (Figure 7C). Taken together, these data suggest that H22 and H23 are the major sources of glycinergic inputs to ON SACs that are activated during light increments.

In summary, we find that the NACs making presumptive glycinergic synapses onto ON SACs were heterogeneous and are a relatively uncommon source of input comprising 6% of all inhibitory input (whereas 89% of the input arises from other SACs), in agreement with the earlier study (Ding et al., 2016). NAC synapses were sparsely distributed widely across the entire SAC arbor (Figure 7A), which is different from SAC-SAC synapses that cluster around the soma (Ding et al., 2016). This synaptic distribution was true for the sparse samples of H22 and H23 cells, as well as the more abundant H19 inputs (Figure 7A). Being located nearer to the distal tips from which SACs release transmitter is likely to be one important reason why NAC input can powerfully control SAC output despite their sparsity.

DISCUSSION

Glycinergic NAC inputs: the final pieces of the DS circuit

Several specializations of glycinergic NAC input to ON SACs enable them to powerfully inhibit ON SACs despite their meager anatomical representation (Ding et al., 2016; Figure 7). First, glycinergic transmission at NAC-SAC synapses occurs on an ultra-slow timescale compared with the GABAergic inputs of SACs and WACs. Second, presynaptic NACs derive excitatory drive from both ON and OFF cone bipolar cells. This provides glycinergic inhibition to SACs at both phases of a grating stimulus. By contrast, SAC and WAC inhibition is triggered only during the ON phase. Third, NAC axonal arbors tile the retinal plane with orderly mosaics and distribute their glycinergic inputs uniformly throughout the SAC’s dendritic field, whereas SAC-to-SAC contacts are strongly biased to the proximal dendrites (Ding et al., 2016). This implies GABA/glycine balance exhibits a spatial gradient across the starburst arbor, with NACs contributing a larger fraction of all inhibitory input in the distal SAC arbor, the site of synaptic output. Together, these factors enable sparse glycinergic synapses to sum their contributions and mediate a slow, sustained, and widespread inhibition that can outweigh inhibition mediated by the much more abundant GABAergic synaptic contacts.

In recent studies, glycinergic contributions to the DS circuit have been largely overlooked. An early Ca²⁺-imaging study found that DS tuning of SAC dendrites was unaffected by strychnine (Hausselt et al., 2007). However, that study and many subsequent studies usually utilize expanding and contracting annular rings presented through a “doughnut-shaped” window, which by design would not be expected to optimally stimulate NACs, whose receptive fields are centered around the SAC soma (Figure 1D). Thus, in retrospect, the minor effects of strychnine on SAC Ca²⁺ responses that were previously observed are not surprising. On the other hand, in classic bulk-release studies, strychnine was found to increase light-evoked ACh release (Cunningham and Neal, 1983), which here we show is likely to be mediated by well-defined glycinergic inputs that are evenly distributed over SAC dendrites.

Classic studies also showed that blockade of GlyRs by strychnine had little if any effect on DS of DSGCs (Caldwell et al., 1978; Wyatt and Day, 1976). However, the simple moving spots used to probe DS mechanisms in these studies only briefly activate the circuit. Thus, here too, the effects of slow NAC inhibition would have been overlooked (Figure 1E). However, even when grating stimuli were used, which effectively engages the NAC inhibition, the effect of strychnine on the DS of ganglion cells was difficult to discern (Figure S6). We reason that this is because the DS of DSGC is shaped by several factors beyond directional tuning of SAC dendrites (reviewed by Murphy-Baum et al., 2021). For example, temporal differences between GABA and ACh input from SACs are enough to drive DS in DSGCs, even when SACs are rendered completely non-DS (Hanson et al., 2019; Schachter et al., 2010).

Our SEM reconstructions identify three distinct types of NACs as sources of glycinergic inhibition of SACs. H19 cells are by far the dominant source (87% of identified NAC synapses). This amacrine type was initially introduced by Helmstaedter et al. (2013), as two types (17 and 19), but based on their non-overlapping mosaics and similar stratification, we argue that they in fact belong to a single type (Figure S5). These cells thus appear to mediate a textbook example of crossover (OFF-to-ON) inhibition, with bipolar excitatory input coming mostly in the OFF sublayer, with outputs in the ON sublayer that include ON SACs.

Thus, t3b bipolar cells, through their ribbon contacts onto H19 cells, are presumably the major driver of the IPSC recorded in ON SACs at light offset. The ON IPSC seems to have a more diverse origin, including sparse inputs from two NAC types with exclusive ON bipolar input—H22 and H23/MAC. However, the pre-eminent source of NAC input to SACs, the H19 type, receives sparse ON bipolar input along with its dominant OFF input. Thus, the ON glycinergic IPSCs may reflect a blend of inputs from all three of these NAC types. The H23/MAC amacrine cell is linked to the glycinergic-SAC circuit in two ways. It makes sparse, presumed glycinergic synapses directly onto ON SACs. However, its larger role seems likely to stem from its dense inhibitory contacts onto the H19 amacrine cells, the dominant glycinergic input to SACs. Both influences are potentially subject to Muller-cell modulation, through coupling between Muller glia and H23/MACs (Grimes et al., 2021).

Complementary roles for multiple amacrine cells in the DS circuit

SACs have the remarkable ability to compute direction of moving images spanning a wide range of spatiotemporal frequencies, relying on multiple circuit mechanisms. For example, WACs control the spatiotemporal selectivity of the DS circuit acting on bipolar cells as well as directly on SACs (Hoggarth et al., 2015; Huang et al., 2019). Interestingly, NACs and WACs are likely to be reciprocally connected (D.B., unpublished data; also see (Werblin, 2010)). Presumably, WAC contributions were minimized in this study because stimuli were restricted to the vicinity of the SAC excitatory receptive field. Future investigations are required to probe NAC-WAC interactions and determine how these three distinct amacrine cell types interact to enable direction coding under natural stimulus conditions.

Reciprocal GABAergic inhibition arising from neighboring SACs plays an important role in shaping the directional tuning properties of SAC dendrites (Chen et al., 2016; Ding et al., 2016; Lee and Zhou, 2006; Munch and Werblin, 2006; Poleg-Polsky et al., 2018; but see Euler et al., 2002; Hausselt et al., 2007). Synaptic contacts between SAC dendrites are biased in an anti-parallel manner, such that dendrites selective for opposite directions inhibit each other more strongly than for any other orientation. These contacts lie in the central zone of the arbor, around the soma, where excitatory inputs from bipolar cells are made (Ding et al., 2016; Lee and Zhou, 2006). This ensures that moving objects only activate SAC-mediated surround inhibition on one side of an SAC dendritic field, the side from which they enter, as demonstrated in our electrophysiological recordings (Figure 1E). Thus, SAC-mediated inhibition cooperates with other intrinsic DS mechanisms to ensure that objects moving from the dendritic tip to the soma (centripetal motion) produce weak responses. Reciprocal GABAergic inhibition is especially important when stimulus contrast is high, suppressing the strong depolarizing excitatory inputs that would otherwise drive spiking during non-preferred centripetal motion (Ding et al., 2016). It is important to note that SAC-mediated GABAergic inhibition only occurs when SACs themselves are strongly stimulated.

On the other hand, NACs are driven by a semi-independent pathway mediated by an overlapping set of ON/OFF bipolar cells. This enables NACs to provide inhibition over a significantly broader range of contrast, compared with the SACs. Moreover, SAC output was enhanced in both preferred and null directions by glycinergic block, indicating that the glycinergic inhibition itself is non-directional. Importantly, DS in SAC dendrites was significantly degraded when NAC inhibition was blocked. Under these conditions, SAC GABA release is increased, but the SAC-SAC GABAergic inhibition is not sufficient to significantly impact SAC DS. Thus, the powerful NAC-mediated glycine inhibition appears to be well suited and critically required for optimal DS SAC output when the circuit encounters repetitive contrast stimuli. Further, the effects of NACs on DS function were most pronounced at low spatial frequencies, where DS tended to be weaker (Figures 5B and 5D). Thus, NAC inhibition appears to broaden the range over which SACs can perform their computation.

Glycinergic inhibition shapes direction selectivity in SAC dendrites

The mechanisms by which glycinergic inhibition impacts SAC function can be best understood in the context of existing models proposed for SAC DS. These suggest that SACs rely on differences in temporal summation of EPSPs to produce directional responses. From the perspective of their distal dendrites (where output occurs), optimal summation occurs for centrifugal motion. Importantly, these anisotropies in temporal summation are strongly amplified by voltage-gated Ca²⁺ channels, in a manner that is steeply dependent on voltage. Thus, in these models, an effective gain control mechanism is a key requirement to keep the dendritic response within a restricted range, regardless of the strength of the stimulus. Without gain control, the mechanisms underlying DS would easily saturate. Reciprocal inhibition within the SAC network has been previously proposed to serve as a source of gain control (Ding et al., 2016). Consistent with this hypothesis, blocking GABA receptors significantly affects the ability of SACs to compute DS when stimulated with high-contrast, but not low-contrast, spot stimuli (Ding et al., 2016).

However, in the context of sustained activity, like that driven by drifting grating stimuli, we find blocking glycine receptors dramatically increases the SAC GABA output and compromises the ability of SAC dendrites to compute direction. Under GlyR blockade, the increased SAC network inhibition does not appear to offer adequate levels of gain control for maintaining DS under sustained stimulation. This is likely because continuous stimulation produces a significant continuous (“DC”) component that is not easily canceled by transient SAC GABAergic inhibition. The slow, temporally summating NAC-mediated glycinergic inhibition appears better suited for canceling the sustained excitatory components. By strongly shunting SAC dendrites, including their distal output varicosities, and lowering their input resistance, the NAC inhibition to SACs lowers the gain of SAC dendritic EPSPs and thus allows SAC dendrites to compute directional differences more effectively without the saturating effects of DC signals. In addition, the strong glycinergic shunt may also help preserve the electrical isolation of individual SAC dendrites. This would prevent excitation from spreading throughout the arbor, which would degrade its ability to compute direction. The precise degree to which this glycinergic input helps in the functional compartmentalization of signals remains to be assessed experimentally.

Other synapses expressing Glyα4Rs subtype

Slow inhibition appears to be mainly mediated by GlyRs containing α4 subunits, as spontaneous and light-evoked slow IPSCs were mostly absent from SACs in the Glra4 KO mouse. However, in a few SACs, we did note residual GlyR-mediated spontaneous IPSCs, suggesting that other subunits, such as GlyRα2, may also participate in SAC function (Heinze et al., 2007; Majumdar et al., 2009). ON DSGCs have been recently reported to receive both fast and slow forms of postsynaptic glycinergic inhibition, but which GlyR subtypes and NACs are involved remains unknown (Sivyer et al., 2019). Nevertheless, synaptic transmission at other α4-containing glycinergic synapses is also consistent with our conclusions. Some of the WACs and other types of NACs (non-AII ACs) are reported to express α4 receptors and exhibit slow kinetic spontaneous events (decay constant τ ~30 ms) in conjunction with α2 glycine receptors (Heinze et al., 2007; Majumdar et al., 2009). Knocking out GlyRα2 receptors increased the τ_decay of IPSCs to ~70 ms (similar to what we have measured in ON SACs), providing further evidence for the role of α4 in slow synaptic inhibition (Weiss et al., 2008). Furthermore, expression studies demonstrate that mouse α4β receptors mediate slow sIPSCs (τ_decay ~80 ms; Leacock et al., 2018), comparable to the decay of sIPSCs we measured in SACs. Importantly, during light-evoked activity, the unitary glycinergic events sum effectively to produce a sustained glycinergic inhibition that builds and relaxes over many hundreds of milliseconds, persisting long after the moving stimulus has left the SAC receptive field.

Caveats and limitations

We have observed ON and OFF components of glycinergic IPSC and found that the NACs that make sparse connections to SACs can be classified into ON and OFF types. While the simplest model suggests that the identified NACs are the major sources of glycine in the DS circuit, our study did not directly test whether they underlie the light-evoked IPSCs in SACs. Future studies using silencing and activation type experiments will help determine to what extent the identified NACs contribute to glycine inhibition. The molecular typing of glycinergic amacrine cells using high-throughput single-cell RNA sequencing methods (Yan et al., 2020) paves the way for gaining genetic access to specific NAC types required for such investigations.

Why ON, but not OFF, SACs require a glycinergic inhibition for their computation also remains unclear (Ishii and Kaneda, 2014). Though we find OFF SACs do receive some NAC input (Figure 7), the anatomy appears to confirm the functional observation of a bias toward ON SACs. The observed difference in glycinergic input adds to the growing list of differences between ON and OFF SAC properties. Previous reports have shown that ON and OFF SACs receive excitatory inputs with different kinetics (Borghuis et al., 2013). Furthermore, excitation to OFF, but not ON, SACs is partly mediated by NMDA receptors (Fransen and Borghuis, 2017). Purinergic receptor modulation also appears to be dominant in the OFF SAC pathway (Kaneda et al., 2004, 2008). Thus, although ON and OFF SACs are often considered as mirror symmetrical populations based on their anatomy, they appear to differ in the mechanisms that fine-tune their output. Understanding whether and how such regulatory mechanisms enable ON and OFF SACs to compute different aspects of motion within the complex visual scene remains a challenge for future investigations.

Conclusions

In summary, our results demonstrate an important role for NACs in mediating DS responses in SACs. They do so in a way that is distinct from other forms of inhibition mediated by SACs and WACs, highlighting the role of multiple interneurons within a given neural circuit. It is interesting to note that, although the GlyRs mediating NAC responses are ultra-slow, they participate in a rapid computation. Interestingly, in zebrafish, GlyRα4 appears to be involved in the rapid escape and startle responses (Leacock et al., 2018), raising the possibility that tonic inhibition may be used to maintain computational efficacy in other circuits. Thus, neural circuits appear to employ interneurons with diverse anatomical and functional properties for their complex computations.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Dr. Gautam B. Awatramani (gautam@uvic.ca).

Materials availability

The ES cells used to generate the Glra4 KO mouse came from the Knockout Mouse Project (KOMP; RRID:MMRRC_056080-UCD). These mice are available for distribution upon request maureen.mccall@louisville.edu and may require a Material Transfer Agreement.

Data and code availability

• All the data reported in this paper will be shared by the lead contact upon request.

• There are no original codes generated in this paper.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animals

All procedures were performed in accordance with the Canadian Council on Animal Care and approved by the University of Victoria’s Animal Care Committee or the University of Louisville Animal Care and Use Committee. Experiments were performed using adult (P30 or older) mice of either sex: C57BL/6J (JAX:000,664). In most experiments, SACs and DSGCs were identified in mouse lines in which they were fluorescently labelled, using 2-photon imaging. Superior coding DSGCs were targeted in Hb9^eGFP (RRID: MGI_109,160) line, in which they are fluorescently labelled. ON SACs were genetically accessed using the choline acetyltransferase (ChaT) cre-mouse line (RRID: MGI_5475195) or Oxtr-T2A-Cre (B6.Cg-Oxtr^{tm1.1(cre)Hze}/J; Jackson laboratory stock: 031303). To label SACs with fluorescent markers the ChAT-cre line was crossed with Ai9 (RRID:MGI_3809523); or floxed nGFP line (Kindly provided by Dr. Marla Feller, University of California at Berkeley). To reduce SAC GABA release the ChAT-cre line was crossed with a floxed allele of Slc32a1 (Slc32a1^tm1Lowl, JAX stock # 012,897), which also is commonly referred to as Vgat^flox/flox; JAX012897). To block SAC-SAC GABA transmission postsynaptically, the ChAT-cre line was crossed with Gabra2^tm2.2Uru. Glrα4 KO mice were generated from ES cells purchased from the KOMP Repository (Glra4^{tm1a(KOMP)Wild-types}; RRID:MMRRC_056080-UCD) and used by the Transgenic Core Facility at Cincinnati Children’s Hospital Medical Center to generate chimeric mice. Figure S1 shows the WILD-TYPE and targeting locus for Glra4. Founder mice were crossed to C57BL6/J mice to produce heterozygous animals carrying the targeted allele. To generate the Glra4 KO used in this study mice carrying the KOMP allele were crossed to B6N.Cg-Edil3^{Tg(Sox-cre)1Amc} (Jax stock # 014094), and offspring carrying the floxed allele were identified by PCR. This removed all the targeting construct between the LoxP sites, including exon 3. The genotype of the progeny was determined using PCR and the following primers, which differentiated the wild-type allele (201 bp) from the KO allele (271 bp).

Glra4–1 CAGGTGGACACCAATCTTCC

Glra4–3 TTTTGGTAGGCCAGGAGGTT

Because Glra4 is located on the X chromosome, female heterozygous mice were bred to C57BL/6J males, and the first generation produced Glra4 homozygous males. The progeny was backcrossed onto C57BL/6J for 8 generations and the animals then maintained as homozygous in the breeding colony thereafter.

METHOD DETAILS

Immunofluorescence

To characterize the changes in GlyRα4 expression, we use a previously published immunohistochemistry protocol (Nobles et al., 2012) on 20 μm cryostat sections. We examined and compared the expression pattern of GlyRα4 in wild-type, heterozygous and KO mouse retinal section (Figure S2), using a rabbit anti-Glra4 antibody (1:100; Chemicon Cat# AB9696) with a goat anti-rabbit Cy3 secondary (1:200, Jackson ImmunoResearch Cat# 705165003). We examined the pattern of GlyRα4 relative to the bands formed by the ON and OFF SAC dendrites in the IPL, using a goat anti-Choline Acetyltransferase antibody (1:200; Chemicon Cat# AB144) with a donkey anti-goat IgG Alexa 405 (1:200, Abcam Cat# ab175664). Confocal images were acquired on an Olympus FV1000 confocal microscope (Olympus, U.S.A.) and z-stack images were acquired with 0.2 μm z-steps using a 63× oil immersion lens (NA 1.4). In wild-type and heterozygous mouse retina sections, GlyRα4 expression is found only in the IPL and is most highly concentrated within the band formed by the ON SAC dendrites (immunopositive for ChAT) (Figure S2). The developing (P20) and mature (≥ P40) Glra4 KO retinal sections do not express GlyRα4. We attempted to use this same antibody in Western blots and found that a band of the same size is labeled in wild-type and Glra4KO retinal fractions, thus this antibody is not useful for detecting GlyRα4 in Western blots (data not shown).

To demonstrate the ACh 3.0 sensor is selectively targeted to SACs in the ChAT-Cre transgenic mouse line, we examined the co-expression of the ACh 3.0 sensor and Choline Acetyltransferase using immunochemisrty. A goat anti-GFP antibody (1:500; Abcam ab6673) labeled the tagged ACh 3.0 sensor, and a rabbit anti-Choline Acetyltransferase antibody (1:500; Abcam ab178850) labeled the ON and OFF SACs. Retinas were incubated in the secondary antibodies donkey anti-goat Alexa 488 (1:2000; ThermoFisher A11055) and donkey anti-rabbit Alexa 647 (1:500; Invitrogen A31573) in blocking solution for 60 min at room temperature. Confocal images with 0.6 μm z-steps using a 40× oil immersion lens (NA 1.3, Nikon Plan Fluor; Nikon Corp.) were acquired on a Nikon C2plus confocal microscope (Nikon Corp., Tokyo, Japan).

Retinal preparation and physiological recordings

Mice were dark-adapted for 30–60 minutes before being anesthetized with isoflurane and decapitated. Retinas of both eyes were extracted in standard Ringer’s solution under a dissecting microscope equipped with infrared optics. Isolated retinas were laid flat over a pre-cut window in 0.22 mm membrane filter paper (Millipore), with ganglion cell side up. Retinas were then placed in a recording chamber maintained at 33–35°C and continuously perfused with Ringer’s solution (110 mM NaCl, 2.5 mM KCl, 1mM CaCl₂, 1.6 mM MgCl₂, 10 mM glucose, 22 mM NaHCO₃) that was bubbled with 95% CO₂/5% O₂. Retinas were visualized using a BX-51 WI microscope (40× water-immersion objective; Olympus, Canada) with the help of an infrared CCD camera (Spot RT3 Diagnostic Instruments). Patch-clamp recordings were made with a MultiClamp 700B amplifier (Molecular Devices). Patch electrodes (5–8 MΩ) contained the following (in mM): 112.5 CH₃CsO₃S, 7.75 CsCl, 1 MgSO₄, 10 EGTA, 10 HEPES, 5 QX-314-bromide (Tocris). The pH was adjusted to 7.4 with CsOH. The reversal potential for chloride was calculated to be −56mV. The recordings were not corrected for series resistance or the junction potential. Signals were digitized at 10 kHz (PCI-6036E acquisition board, National Instruments) and acquired using custom software written in LabVIEW. Data was analyzed offline with custom written software in MATLAB. Unless otherwise noted, all reagents were purchased from Sigma-Aldrich Canada. TTX and gabazine were purchased from Alomone Labs and Tocris, respectively.

Virus injections

The single-stranded AAV vector ssAAV-9/2-hSyn1-chI-dlox-Igk_ACH4.3(rev)-dlox-WPRE (mut6)-SV40p(A) (1.2 × 10¹³ vg/mL) was produced using the Viral Vector Facility at University of Zurich (Addgene #121922). The AAV vectors were intravitreally injected in CART-Cre and Oxtr-Cre mice. To begin, mice were anesthetized with isoflurane (2–3% at 1–1.5 L/min), along with a single dose of subcutaneously injected buprenorphine (0.05–0.1 mg/kg body weight). In addition, a drop of 1% proparacaine was applied topically to the eye. Next, using a 30-gauge needle, a small hole was made at the margin of the cornea and sclera. The AAV preparation (~2 μL) was injected through this hole using a Hamilton injection system (syringe: 7633–01, needle: 7803–05, point style 3, length 10 mm). After injection, mice were returned to their home cage on a heating pad and monitored, until fully recovered.

Two-photon functional imaging

For ACh3.0 sensor imaging, experiments were performed three-to-four weeks after virus injection into the eyes of Oxtr-T2A-Cre, CART-Cre or ChAT-Cre mice (to label SAC/DSGC dendrites). For calcium imaging, SAC somas labelled in ChAT-cre x nGFP mouse line were electroporated with a calcium indicator Oregon Green 488 Bapta-1 (15mM in water; ThermoFisher Scientific) (Ding et al., 2016). Retinas were prepared following a procedure that was similar to that used for the electrophysiological recordings. SAC/DSGC dendrites were imaged using a custom two photon microscope equipped with Ti/Sapphire laser (Insight DeepSee⁺:Spectra Physics) tuned to 920 nm. The laser was guided by XY galvanometer mirrors (Cambridge Technology) controlled with software developed by Dr. Jamie Boyd (University of British Columbia) in the Igor Pro environment. Image scans were acquired at 10–15 Hz with a XY pixel size of 0.28 μm × 0.28 μm, and a dwell time of 1 μs/pixel. Imaging windows were typically 125 μm × 50 μm in area.

Visual stimulation

Visual stimuli were produced using a digital light projector and were focused onto the photoreceptor layer of the retina through the sub-stage condenser. The background luminance was measured to be 10 photoisomerisations/rod/second (R*/rod/s). Full contrast spot stimuli were used in most experiments. Experiments were performed in dark- and light-adapted states. Stimuli were designed and generated using Psychtoolbox (MATLAB, Mathworks). Spots were presented on a dark background, while gratings were presented on a grey background (400 R*/rod/s) through a 500 μm circular window. Sine wave gratings with spatial frequency 0.1 and 0.01 cycles/deg moving at 1Hz were used and were presented once every 10 seconds. Stimulus widths were converted to cycles/degree of visual angle assuming 1° degree ~30 μm on the mouse retina (Remtulla and Hallett, 1985).

Data analysis

Electrophysiology

Physiological data were analyzed using custom routines written in Matlab or Igor (Wavemetrics). For stationary spots, ON IPSCs measured in SACs were separated into transient and sustained components based on their timing. The strength of the transient component was estimated integrating IPSC over the first 100 ms of the response, whereas the sustained component was estimated by integrating the rest of the response. For the OFF IPSC, the entire response was considered to be sustained. For moving spots, the temporal windows over which responses were integrated were based on the pharmacological profile of the transient and sustained inhibitory components.

For grating stimuli presented to ON SACs for each cycle, first the total charge (area under the curve) was measured. Then, the transient component was estimated by baselining the current 200ms before the response, as shown in Figure 3B. Finally, the strength of sustained component was computed by subtracting the transient component from the total charge. For the sustained IPSCs, ON and OFF components could not be easily separated, thus the whole area under the sustained region was considered as the slow current mediated by glycine receptors. In ON-OFF DSGCs, ON and OFF components could not be separated for high spatial frequency gratings (SF 0.1 cycles/degree) but could be separated for lower spatial frequency gratings (SF 0.01 cycles/degree). A photodiode was used to measure the phase of the stimulus.

To estimate the tuning width and preferred direction of the DSGC responses, the data was fitted to a Von Mises fit defined by the following equation.

$$f(\theta )=A*e^{\frac{1}{{\sigma }^2}*\cos (\theta -\mu )}$$

where θ indicates the direction of motion, A refers to the amplitude, μ indicates the mean and σ the standard deviation. σ is indicated as tuning width and μ as the preferred direction in Figure S5.

Spontaneous IPSCs measured in SACs (V_HOLD ~ 0 mV) in dark-adapted retinas and analyzed using Taro tools software in Igor. Only events with amplitudes above 5 pA were selected for analysis. The rise of the sIPSC was estimated by measuring the time taken for the event to rise from 20 to 80% of its maximal amplitude. The decay kinetics were estimated using single exponential function (only non-overlapping events were chosen to measure decay kinetics).

Two photon imaging

For the ACh3.0 signals, light-evoked changes in fluorescence were measured over the whole field-of-view. In the case of Ca²⁺ signals, changes in fluorescence were measured in individual SAC varicosities. Light-evoked changes is fluorescence (ΔF) were normalized to the baseline fluorescence (F) measured before stimulus onset, corrected for background fluorescence (Fb). In the case of AC3.0 signals, the average ΔF/F over the duration of the grating stimuli was used to quantify the strength of the response. In the case of Ca²⁺ signals, for moving spots, the peak ΔF/F was used to quantify the strength of the response, while for sine wave gratings, the magnitude of the fundamental harmonic was used (F1 component). DSI was calculated as (PD−ND)/(PD + ND), where PD and ND refer to responses evoked by stimuli moving in the preferred and null direction, respectively.

Anatomical reconstruction

A previously published dataset acquired using scanning SBEM was analyzed (retina k0563 (Ding et al., 2016)). Voxel dimensions were 12 × 12 × 25 nanometer (nm) (x, y, and z, respectively). ON-starbursts were identified and back traced to identify the pre-synaptic NACs. Bipolar ribbon synapses were identified on the NACs, by using their synaptic ribbons. All analyses were performed by tracing skeletons and annotating synapses using the Knossos software package (https://knossostool.org/).

Computational model

We used a computational model to estimate membrane potential, intracellular Ca²⁺ or local synaptic conductances in SAC dendrites. The model comprised a 7-cell SAC network (1 cell in the center with 6 neighbors in a hexagonal array) incorporating mouse connectivity, utilizing several example morphologies with random rotations to minimize results peculiar to one morphology or orientation (Ding et al., 2016; Stincic et al., 2016). SAC dendritic tips made reciprocal GABAergic connections on their neighbors’ opposing dendrites proximal to the soma. The SAC morphologies were discretized with a compartment size of 0.1 lambda (space constant), giving 170–200 compartments per cell. Bipolar cells were placed by an algorithm in a semi-regular array according to the known regularity of their spacing and based on their proximity made synaptic contact with AMPA receptors on SAC dendrites within a radius of ~100 μm of the soma (Ding et al., 2016). Biophysical properties (Ri, Rm, channel conductances and densities, synaptic properties and distributions) were set from previous models (Ding et al., 2016; Stincic et al., 2016). During simulated motion, roughly 25 GABAergic SAC-SAC synapses were activated, each with a unitary synaptic conductance that ranged between 25–100 nS, producing a total inhibitory conductance that ranged between 500–1000 nS at the central model SAC. Glycinergic input to SACs came from narrow-field amacrine cells that made a total of ~20 synaptic connections to each SAC within a 40 μm radius of the soma. Each glycinergic synapse had a maximum conductance of 40–50 nS, contributing a total inhibitory conductance to each SAC of 500–1000 nS. The glycinergic synapses were given a time constant of 500 ms. Kv3 channels were included at the SAC soma (3 mS/cm²) and proximal dendrites (2 mS/cm²) for current clamp recordings from the soma but were removed for somatic voltage clamp recordings. Slowly inactivating calcium channels were included in medial and distal SAC dendrites (9 mS/cm²) and generated regenerative peaks to −20 mV when SAC dendritic EPSPs were depolarized above −45 mV. Ligand and voltage-gated ion channels were modeled using Markov sequential-state diagrams. Visual signals were simulated in the model’s bipolar cells by voltage clamping them with a spatio-temporal stimulus pattern specified as a resting potential that produced a low continuous vesicle release rate and contrast defined as a depolarization in the range of 1–10 mV. The 7-cell network was stimulated with a sine-wave grating (300 μm/cycle, 600 μm/s) at a background sufficient to depolarize (with the given synaptic input) the SACs to a somatic V_rest of −55 to −50 mV with voltage peaks to −40 mV. Models were run with the Neuron-C simulation language on an array of 12 3200 MHz Opteron servers (each with 2 CPUs, 8 cores) running Linux-64 under the Mosix parallel job-sharing system.

QUANTIFICATION AND STATISTICAL ANALYSIS

All population data has been expressed as mean ± SEM. Student’s t test was used to compare values under different conditions and the differences were considered significant when p ≤ 0.05.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
anti-Glra4	Chemicon	Cat# AB9696;RRID:AB_673015
anti-Choline Acetyltransferase	Chemicon	Cat# AB144; RRID:AB_11212843
anti-GFP	Abcam	ab6673; RRID:AB_305643
Bacterial and virus strains
GPCR activation based ACh sensor	Addgene	Addgene #121922
GRAB-ACh3.0
Chemicals, peptides, and recombinant proteins
SR-95331	Tocris	Cat# 1262
Strychnine	Sigma	Cat# S0532
TTX	Alomone Labs	T-550
Oregon green 488 BAPTA-1	ThermoFisher Scientific	O6798
Experimental models: Organisms/strains
Mouse: wild-type (C57BL/6J)	The Jackson Laboratory	RRID: IMSR_JAX:000664
Mouse: Hb9-EGFP	Gift from R. Brownstone (University of Dalhousie)	RRID: MGI_109160
Mouse: Oxtr-T2A-Cre	The Jackson Laboratory	RRID: IMSR_JAX: 031303
Mouse: Vgatflox/flox	The Jackson Laboratory	RRID: JAX 012897
Mouse: ChAT-Cre/nGFP/TrHr (CNT)	Gift from Dr. Marla Feller (University of California at Berkeley)	Morrie and Feller, 2018
Mouse: Gabra2flox/flox	The Jackson Laboratory	RRID: MGI_5140553
Mouse: Chat-IRES-Cre	The Jackson Laboratory	RRID: MGI_5475195
Mouse: Ai9	The Jackson Laboratory	RRID: MGI_3809523
Software and algorithms
MATLAB	Mathworks	http://www.mathworks.com; RRID: SCR_01622
LABVIEW	National Instruments	http://www.ni.com; RRID:SCR_014325
KNOSSOS	Max Planck Institute for Medical Research, Heidelberg, Germany	http://www.knossostool.org; RRID:SCR_003582

Highlights.

• ON starburst amacrine cells (SACs) receive a powerful ON/OFF glycinergic inhibition

• Glycinergic input to SACs is sparse and ultra-slow and relies on α4-containing GlyRs

• Glycinergic pathways control SAC output gain and direction selectivity

• SBEM reveals specific types of narrow-field amacrine cells connect to SACs