Cone bipolar cell synapses generate transient versus sustained signals in parallel ON pathways of the mouse retina

Parallel processing is a fundamental organizing principle in the nervous system, and understanding how parallel neural circuits generate distinct outputs from common inputs is a key goal of neuroscience. In the mammalian retina, divergence of cone signals into multiple feed-forward bipolar cell pathways forms the initial basis for parallel retinal circuits dedicated to specific visual functions. Here, we used patch-clamp electrophysiology, electron microscopy and two photon imaging of a fluorescent glutamate sensor to examine how kinetically distinct responses arise in transient versus sustained ON alpha RGCs (ON-T and ON-S RGCs) of the mouse retina. We directly compared the visual response properties of these RGCs with their presynaptic bipolar cell partners, which we identified using 3D electron microscopy reconstruction. Different ON bipolar cell subtypes (type 5i, type 6 and type 7) had indistinguishable light-driven responses whereas extracellular glutamate signals around RGC dendrites and postsynaptic excitatory currents measured in ON-T and ON-S RGCs in response to the identical stimuli used to probe bipolar cells were kinetically distinct. Anatomical examination of the bipolar cell axon terminals presynaptic to ON-T and ON-S RGCs suggests bipolar subtype-specific differences in the size of synaptic ribbon-associated vesicle pools may contribute to transient versus sustained kinetics. Our findings indicate bipolar cell synapses are a primary point of divergence in kinetically distinct visual pathways.


Introduction
Divergence into parallel pathways allows neural circuits to generate diverse outputs from a common set of inputs.This functional diversity can be quite striking; in vision, for example, light inputs initially encoded by the same collection of cone photoreceptors mediate sensitivity to several kinds of motion, to color and to spatial structure, all in the same region of space.These diverse functional sensitivities arise through parallel processing of light inputs, first in the retina and subsequently in downstream visual areas.Here we focus on the mechanisms producing transient and sustained responses -a key differentiating characteristic of retinal output cells across species (Baden et al., 2016;Bae et al., 2018;Cleland et al., 1973;Ikeda and Wright, 1972;Roska and Werblin, 2001;Werblin, 1970).
The retina provides a unique opportunity to study the mechanisms that differentiate signals in parallel pathways.We have a near-complete list of retinal cell types and substantial information about which of these cell types participate in specific parallel pathways (Helmstaedter et al., 2013;Masland, 2012).Under daylight conditions, retinal processing starts with cone photoreceptors transducing light inputs and providing divergent output to ~13 types of bipolar cell (Euler et al., 2014;Greene et al., 2016;Helmstaedter et al., 2013;Shekhar et al., 2016;Tsukamoto and Omi, 2017).The resulting bipolar cell signals are transmitted to ~20-40 retinal ganglion cell (RGC) types (Baden et al., 2016;Bae et al., 2018;Goetz et al., 2022).RGC inputs are strongly shaped by the activity of a large and diverse collection of amacrine cells that receive input from bipolar cells and provide presynaptic inhibition to bipolar cells and postsynaptic inhibition to RGCs (Diamond, 2017).This circuitry produces highly distinct responses across RGC types -with cells that sample inputs on a variety of spatial and temporal scales, cells with distinct chromatic properties, cells that sense the direction of moving objects, and cells that separate local from global motion (Gollisch and Meister, 2010;Kerschensteiner, 2022).Several mechanisms contribute to these functional differences, starting with the division of signals into ON and OFF pathways at the cone-to-bipolar synapse (Euler et al., 2014;Slaughter and Miller, 1981).Temporal properties of responses of different bipolar cells can also differ for some stimuli (Awatramani and Slaughter, 2000;Baden et al., 2013;Euler et al., 2014;Euler and Masland, 2000;Ichinose et al., 2014;Ichinose and Hellmer, 2016;Wu et al., 2000).In the inner retina, integration of excitatory and inhibitory synaptic inputs produces directionally-selective responses in specific RGC types (Taylor and Vaney, 2002;Vaney et al., 2012) and differences in spatial weighting functions in other RGCs (Cafaro and Rieke, 2013;Roska and Werblin, 2001;Zhang et al., 2012).Properties of dendritic integration and RGC spike generation also differentiate responses of some RGCs (Margolis et al., 2010;Ran et al., 2020;Wienbar and Schwartz, 2022).
A fundamental difference among RGCs is whether their responses are sustained or transient (Baden et al., 2016;Bae et al., 2018;Cleland et al., 1973;Ikeda and Wright, 1972;Roska and Werblin, 2001).Cone photoreceptors respond to changes in light intensity with graded potentials that have both a transient and sustained component (Angueyra and Rieke, 2013;Cangiano et al., 2012;Szikra et al., 2014).Both kinetic components of the cone response are maintained in the firing rates of sustained cells, which track the prevailing contrast.However, only the initial component of the cone response is preserved in the spiking response of transient cells, which therefore respond primarily to changes in contrast.Where and how these differences originate is not clear.Here, we show that the distinct kinetics of responses of ONsustained alpha and ON-transient alpha RGCs (1) persist for stimuli that elicit near-identical responses among the relevant bipolar cells, (2) do not depend on amacrine cell-mediated inhibition and, (3) are correlated with anatomical differences in output synapses of bipolar cell subtypes that are presynaptic to these RGCs.Together, these findings indicate that bipolar cell synapses are an important point of divergence in kinetically distinct visual pathways.

Results
We focused on two ON RGC cell types, ON-transient alpha (ON-T) and ON-sustained alpha (ON-S) RGCs, that could be readily identified in a flat-mount preparation of the isolated mouse retina (Fig. 1A) by their characteristic spiking response to a step increment in light intensity (Fig. 1B) (Bae et al., 2018;Goetz et al., 2022;Krieger et al., 2017;Kuo et al., 2016).Based on their light response, dendritic field size and stratification pattern in the inner plexiform layer (IPL) (see below; Fig. S4), the ON-T RGCs we studied here likely correspond to the 6sw RGC type of Bae et al. (2018) and the medium-field ON-transient RGC of Goetz et al. (2022).
We characterized the temporal spiking response properties of ON-T and ON-S RGCs by using an extracellular cell-attached patch electrode to measure action potential firing from either cell type while presenting a spatially uniform, randomly fluctuating light stimulus (Gaussian noise) that allowed us to efficiently probe a variety of stimulus contrasts and temporal frequencies.
These and all subsequent physiological recordings were conducted using a photopic mean background light level (~3300 R*/rod/s; ~2000 P*/Scone/s) to suppress rod photoreceptor responses and therefore focus on cone-mediated signaling (Grimes et al., 2018;Ke et al., 2014).Figure 1C shows the spiking response of example ON-T and ON-S RGCs when we presented multiple repetitions of the same random stimulus contrast sequence; these RGC types clearly responded differently to temporally modulated light input.In particular, noiseevoked ON-T RGC spike responses were comprised primarily of large amplitude, less frequent spike bursts whereas ON-S RGC responses consisted of a more even distribution of small to large amplitude events (Fig. 1C and Fig S1A).We summarized the response of each ON-T and ON-S RGC by computing the average stimulus preceding each recorded spike (the spike-triggered average or STA stimulus) when presenting a non-repeated Gaussian noise stimulus (Fig. 1D) (Chichilnisky, 2001) and we quantified response kinetics from these STA waveforms by measuring the time of zero-crossing (between trough and peak) as well as the ratio of amplitudes of the trough and peak (biphasic index).ON-T RGC STAs were significantly more biphasic (biphasic index = 0.63 ± 0.07; mean ± S.D.; n = 15) than those measured for ON-S RGCs (biphasic index = 0.43 ± 0.11; mean ± S.D.; n = 13) (Fig. 1E, right), consistent with their more transient response kinetics that are apparent with step stimuli (Fig. 1B).ON-T RGC responses were also slightly delayed (zero-cross time = -0.080± 0.0063 sec; mean ± S.D.; n = 15) relative to those of ON-S RGCs (zero-cross time = -0.072± 0.0052 sec; mean ± S.D.; n = 13) (Fig. 1C, bottom and 1E, left).We used the STA to quantify spiking behavior because it provides a convenient model of RGC temporal response properties.
However, it should be noted that the STA alone does not capture all differences between the RGC spike responses -e.g. the discreteness of the ON-T responses in Figure 1C.

Bipolar cell to RGC synapses are an important site of signal diversification in ON pathways
A variety of mechanisms could potentially contribute to the distinct temporal response characteristics of ON-T versus ON-S RGCs, including differences in excitatory and/or inhibitory input conveyed by presynaptic circuits.Parallel visual channels are initially established by the divergence of cone signals to different cone bipolar cell subtypes, which have axons that terminate at distinct depths within the IPL and can correspondingly contact distinct RGC types (Euler et al., 2014;Wässle et al., 2009).We therefore sought to understand to what extent kinetic differences in bipolar cell-mediated excitatory synaptic input could account for transient versus sustained responses in these RGCs.
To do this, we used somatic whole-cell voltage-clamp recordings to measure excitatory synaptic currents from both RGC types while stimulating the retina with the same spatially uniform Gaussian noise stimulus we used when measuring RGC spike output (Fig. 2A).We used these measurements to construct linear-nonlinear (LN) models comprised of a linear temporal filter (Fig. 2B and Fig. S2A) and a time-invariant (static) nonlinearity (Fig. S2B) that provide a simple estimate of the transformation between visual input and excitatory synaptic input to each RGC type (Kim and Rieke, 2001).LN models for each RGC type accurately predicted measured responses to noise stimuli that were not used to construct the models, although LN models of ON-S RGC responses performed better than those for ON-T RGCs (correlation coefficient between average response to repeated noise stimulus and model prediction: 0.61 ± 0.057 (ON-S, n = 9) and 0.50 ± 0.13 (ON-T, n =11), mean ± S.D.; p = 0.028 Wilcoxan rank-sum test).This may in part be because ON-T stimulus-evoked excitatory currents were biased towards large amplitude events compared to those of ON-S RGCs (Fig. 2A, Fig. S1B), similar to the distinct amplitude distributions of spike events in these cell types noted above (Fig. 1C and Fig. S1A).
The linear filter captures the kinetics of the RGC response and is analogous to the time-reversed STA derived from spike responses (Chichilnisky, 2001).As with the STAs in Figure 1D, excitatory linear filters were slower and more biphasic in ON-T (zero-cross time = 0.10 ± 0.013 s; biphasic ind.= 0.45 ± 0.090; mean ± S.D.; n = 13) compared to ON-S RGCs (zero-cross time = 0.083 ± 0.0074 s; biphasic ind.= 0.31 ± 0.10; mean ± S.D.; n = 15) (Fig. 2C).This latter difference is consistent with more transient light-evoked excitatory synaptic input to ON-T RGCs.Indeed, light step-evoked excitatory currents to ON-T RGCs decayed rapidly to baseline levels within a few hundred milliseconds, even in the continued presence of light input, whereas excitatory inputs to ON-S RGCs had a sustained component that persisted for the duration of the light step (Fig. 2D-E).
These results show that transient versus sustained kinetics are already present in the excitatory pathways that provide input to ON-T and ON-S RGCs.These findings do not rule out additional contributions to the distinct spiking behavior of these RGC types from amacrine cell-mediated inhibitory synaptic input to the RGCs or from different intrinsic spiking properties.However, the clear differences we observed in excitatory synaptic input between ON-T and ON-S RGCs indicates that bipolar cell pathways have a central role in establishing transient versus sustained kinetics.

ON-T and ON-S RGCs receive input from distinct bipolar types
In general, retinal circuits dedicated to transient versus sustained signaling are located at the center and borders of the IPL, respectively (Baden et al., 2016;Bae et al., 2018;Franke et al., 2017;Roska and Werblin, 2001).We compared stratification patterns of ON-T and ON-S RGCs in a subset of whole-cell recordings by filling cells with a fluorescent dye and imaging the dendrites of physiologically-characterized RGCs using two-photon microscopy.We assessed IPL depth in live tissue by using retinas from a transgenic mouse line in which type 5 bipolar cells strongly express EGFP (Gjd2-EGFP mice).These EGFP+ bipolar cells have axons that occupy a narrow band just above and overlapping with the dendrites of ON starburst amacrine cells (Fig. S3A) and express calcium binding protein 5 (CaBP5) (Fig. S3B-D) (Ghosh et al., 2004).Based on the regular spacing of their somas, these cells likely correspond to one of the three subtypes of type 5 bipolar cells previously identified in mouse retina (Greene et al., 2016;Rodieck, 1991) (Fig. S3E-F).Imaging of RGCs together with EGFP+ axon terminals in Gjd2-EGFP mice revealed that ON-T and ON-S RGC dendrites occupy distinct, mostly non-overlapping regions of the IPL (Fig. S4).
The distinct IPL stratification of ON-T and ON-S RGCs indicates they receive input from different bipolar cell populations.A recent connectomic analysis that used a published electron microscopy data set from mouse retina (Ding et al., 2016) found that type 6 and type 7 bipolar cells account for ~50% and 46% of the excitatory synapses formed upon ON-S RGC dendrites, respectively (Sabbah et al., 2018).However, the bipolar cell subtype(s) presynaptic to ON-T RGCs have not previously been identified.Our observation that ON-T RGC dendrites occupy a narrow band in the IPL that very closely matches the stratification pattern of EGFP+ type 5 bipolar cell axons in Gjd2-EGFP mice (Fig. S5) suggests that these bipolar cells likely provide a substantial fraction of the excitatory input to ON-T RGCs.To test this possibility, we transfected retinas from Gjd2-EGFP mice with a plasmid encoding PSD-95 fused to the red fluorescent reporter mCherry to label postsynaptic sites of excitatory input (Fig. S5).We quantified the fraction of putative synapses, defined as sites with overlapping PSD-95 and EGFP fluorescent signals, that arose from Gjd2-EGFP+ axon terminals using confocal microscopy (Schwartz et al., 2012).From this analysis, we concluded that the labeled type 5 bipolar cells in Gjd2-EGFP mice provide ~40% of the ribbon synaptic input to ON-T RGCs (Fig. S5D).
We next performed correlative fluorescence imaging and serial block-face scanning electron microscopy (SBFSEM) to identify which type 5 bipolar subtype is labeled in Gjd2-EGFP mice and to further characterize what other bipolar cell types provide the remaining input to ON-T RGCs.
We first used 2P microscopy to image the dendritic morphology of two dye-filled ON-T RGCs that were identified from their light-evoked spike responses in flat-mounted retina from Gjd2-EGFP transgenic mice (Fig. 3A).After 2P imaging, we fixed the retina and used the near infrared branding (NIRB) technique to create fiducial marks in the tissue around the dye-filled ON-T RGCs (Bishop et al., 2011).After acquiring the SBFSEM volume, we located the somas of the two ON-T RGCs using these fiducial marks and reconstructed the dendritic arbors of both cells (e.g.Fig. 3B).The dendritic morphologies from our EM reconstruction matched those from the light microscopy images.This suggests that we successfully identified the two ON-T RGCs recorded physiologically in our EM volume.We chose the two longest dendritic arbors from the RGC soma to the dendritic tip and identified all the synapses from bipolar cells along these dendrites (Fig. 3C).We then reconstructed all the presynaptic bipolar cells and classified them according to their axon morphology and stratification (Fig. 3D).Careful examination of axonal tiling among type 5i, 5o and 5t subtypes provided additional confirmation of the classification of these type 5 bipolar cells (Fig. S6A) (Greene et al., 2016).When comparing the EM images with light microscopy images, it is clear that GJD2-GFP+ bipolar cells are type 5i bipolar cells (Fig. S6B).

(E) Bar plot showing the proportion of input synapses to ON-T RGCs from each BC type. Mean ± SEM (n=3
RGCs, 2 from this study and one from Ding et al. (2016)).
Results from the SBFSEM reconstructions agreed closely with those from the light microscopy approach described above.Type 5i bipolar cell axon terminals provided the largest fraction of synapses onto the two ON-T RGCs within our SBFSEM volume (~30-35%).Other inputs arose from type 5o (~10%), type 5t (~5-15%), XBC (~15-30%), and, somewhat surprisingly given the differences in stratification, from type 6 (~20%) bipolar cells (Fig. 3E).We confirmed this connectivity pattern by additionally analyzing bipolar cell synaptic inputs to a putative ON-T RGC (RGC7089) in the same publicly available SBFSEM volume (Ding et al., 2016) that was used by Sabbah et al., (2018) to map inputs to ON-S RGCs (Fig. 3E).

Bipolar cell light responses differ minimally across types
Are transient versus sustained excitatory inputs inherited from the response properties of the different bipolar cells presynaptic to ON-T and ON-S RGCs?Or is the distinct excitatory synaptic input these RGCs receive primarily due to differences at the bipolar cell to RGC synapse?To answer these questions, we compared the light-driven response properties of the bipolar cell subtypes that comprise most of the excitatory input to ON-T or ON-S RGCs.We made targeted electrophysiological recordings from type 5i, type 6 and type 7 bipolar cells in retinal slices from three transgenic mouse lines in which these different bipolar cell subtypes express fluorescent proteins (see Methods).In these experiments, we used somatic whole-cell perforated-patch recordings to limit perturbation of the G-protein signaling cascade that mediates ON bipolar cell light responses.
Somatic resting membrane potentials did not differ significantly among these bipolar cell types (mean ± S.D. V rest : -38 ± 6.3 mV in type 5i (n= 11); -33.7 ± 6.3 mV in type 6 (n= 7); -36.5 ± 4.7 mV in type 7 (n=6); Kruskal-Wallis test, p= 0.35).We characterized light responses using the same spatially uniform Gaussian noise stimulus (Fig. 4B) and linear-nonlinear model analysis (Fig. 4C and Fig. S7) as in RGC recordings to facilitate direct comparison of results from bipolar cells and RGCs (Kim and Rieke, 2001;Rieke, 2001).LN models of bipolar cell responses accurately predicted responses to noise stimuli (correlation coefficient between average response to repeated noise stimulus and LN model prediction = 0.75 ± 0.15 for n = 18 bipolar cells (n = 13 type 5; n = 3 type 6; n = 2 type 7)).Unlike for ON-T and ON-S RGCs, temporal characteristics of type 5i, type 6 and type 7 bipolar cell linear filters were indistinguishable from each other under our recording conditions (Fig 4B -D).Although there was some variability in the temporal response properties across the population of bipolar cells we recorded, these differences were not associated with morphologically and molecularly defined bipolar cell types.Rather, we observed variability in response kinetics within each bipolar cell subtype and temporal responses thus overlapped substantially across different type 5i, type 6 and type 7 bipolar cells (Fig. 4 and Fig. S7).
Responses to light steps measured in a subset of type 5i and type 6 bipolar cell recordings confirmed our conclusion from Gaussian noise stimuli that different bipolar cell types exhibit nearly identical light response kinetics under our recording conditions (Fig. 4E-F).Unlike the clear differences in light step-evoked excitatory currents we measured in ON-T and ON-S RGCs (Fig. 2D-E), type 5i and type 6 bipolar cells both had sustained responses to light steps.These findings indicate that mechanisms upstream of the generation of the cone bipolar cell voltage response, such as synaptic transmission from cones to cone bipolar cells and intrinsic response properties of the cone bipolar cells, cannot explain the kinetically distinct responses in parallel ON channels mediated by type 5i, type 6 or type 7 bipolar cells.

Presynaptic inhibition does not establish kinetic differences in bipolar cell-mediated excitation
Bipolar cells participate in diverse inhibitory circuits with amacrine cells, which are comprised of >40 morphologically and functionally distinct cell types (Diamond, 2017;Helmstaedter et al., 2013).Recent work suggests that amacrine cell-mediated presynaptic inhibition is critical to bipolar cell subtype-specific patterns of light-evoked glutamate release (Franke et al., 2017).
Our bipolar cell voltage recordings, which should reflect a combination of light-evoked excitatory and inhibitory input, argue against the involvement of presynaptic inhibition in generating differential bipolar cell synaptic output between type 5i, type 6 and type 7 bipolar cells.However, the slicing procedure we used to prepare tissue for bipolar cell recordings may damage some inhibitory connections, especially long-range amacrine cell inputs.Distance-dependent attenuation of voltage signals between bipolar cell axon terminals and the somatic recording site could also limit our ability to accurately measure the influence of presynaptic inhibition in these recordings.We therefore performed additional anatomical and pharmacological experiments to investigate a possible role for presynaptic inhibition in generating kinetically distinct excitation to ON-T and ON-S RGCs.Many bipolar cell output synapses are organized into a "dyad" arrangement in which a bipolar cell ribbon synapse releases glutamate onto the processes of two different postsynaptic cells.At dyads, bipolar cell output synapses can be presynaptic to the processes of two amacrine cells, two RGCs, or an amacrine cell and a RGC (Cohen and Sterling, 1990;Dowling and Boycott, 1966;McGuire et al., 1984;Yu et al., 2023).Dyad synapses comprised of an amacrine cell and a RGC may shape bipolar cell glutamate release dynamics onto the RGC partner because glutamatergic excitation of the postsynaptic amacrine cell can elicit release of inhibitory neurotransmitters (GABA or glycine) from the amacrine cell processes that feed back onto the bipolar cell axon terminal.This form of inhibitory feedback control of bipolar cell output, termed reciprocal inhibition, could potentially contribute to differences in glutamate release onto ON-T and ON-S RGCs.

Figure 5. Presynaptic inhibition does not establish kinetic differences between bipolar cell inputs to ON-T versus ON-S RGCs. (A) Example EM micrograph showing a ribbon dyad (red arrowhead) between a BC and an ON-T RGC and an amacrine cell, with an adjacent feedback inhibitory synapse (yellow asterisk). (B) Bar plots showing the percentage of RGC/AC ribbon dyads with feedback inhibition. Mean ± SEM (n=2 ON-S RGCs, 2 ON-T RGCs) (C) Excitatory
We analyzed bipolar cell dyads with postsynaptic amacrine cell processes and either ON-T or ON-S RGC dendrites in SBFSEM volumes to examine whether reciprocal inhibition is likely to generate more transient glutamate release onto ON-T RGCs.We defined putative reciprocal synapses as bipolar cell/amacrine cell/RGC dyads in which we observed clustered vesicles within the amacrine cell process that were within close proximity of the bipolar cell ribbon synapse (Fig. 5A).Our ultrastructural analysis revealed that the prevalence of reciprocal synapses was similar at dyad synapses with postsynaptic ON-T versus ON-S RGCs (Fig. 5A-B).
This finding is not consistent with an important role for feedback/reciprocal inhibition in generating differential bipolar cell output onto the different RGC types.
We also performed pharmacological experiments to further examine what role, if any, inhibition may have in establishing transient glutamate release onto ON-T RGC dendrites.We used ACSF in place of Ames solution for the dissection and maintenance of tissue for these experiments.ACSF is often used in retina experiments, including the glutamate release measurements describe below, and we wanted to check that our results generalized across conditions.Light step-evoked currents in both ON-T and ON-S RGCs were more sustained than responses measured using Ames' solution (Fig. S8A-B).Nonetheless, the relative differences between RGC types appear to be maintained across different experimental conditions.
Blockade of GABA A and GABA C receptors via bath co-application of SR95531 and TPMPA enhanced both the initial transient component and the sustained component of light step-evoked excitatory synaptic currents in ON-T RGCs (Fig. 5C).The strong enhancement of the initial phase of the response we observed is consistent with prior work showing that GABA C receptormediated inhibition limits multivesicular release from bipolar cells (Sagdullaev et al., 2006).On average, block of GABA receptors did not significantly change in the ratio of sustained versus transient components of excitatory inputs to ON-T RGCs compared to control conditions (Fig 5D).If anything, these drugs tended to make excitatory currents more transient, which is the opposite effect expected if presynaptic GABA receptors contribute to generating transient responses in these cells.Thus, GABAergic inhibition appears to control response gain but not kinetics of bipolar cell synaptic output to ON-T RGCs.Similarly, we found that glycinergic inhibition is not likely to contribute importantly to more transient glutamate release from bipolar cells presynaptic to ON-T RGCs; bath application of the glycine receptor antagonist strychnine did not significantly alter the kinetics of light step-evoked excitatory currents in these cells (Fig. 5E-F).Taken together with our ultrastructural analysis, these results show that amacrine cellmediated inhibition of bipolar cell axon terminals does not contribute to kinetic differences in excitatory synaptic input to ON-T and ON-S RGCs.; n = 307 ROIs) (Fig. 6).These results show that differences in glutamate release from presynaptic bipolar cells contribute to transient versus sustained ON RGC responses.However, these experiments do not rule out the possibility that additional postsynaptic mechanisms, for example differential glutamate receptor expression, may also contribute to kinetic differences between RGC types.
Although whole-cell excitatory currents in ON-T and ON-S RGCs (Fig. 2, Fig. S8) and iGluSnFR signals measured from ON-T and ON-S RGC dendrites (Fig. 6) both exhibited clear kinetic differences between the RGC types, there were quantitative differences in the response to light steps between the different measurement approaches.As described above, these differences appear to arise from differences in the external solution (Ames vs ACSF).Importantly, the kinetics of glutamate release measured using iGluSnFR are much more similar to the excitatory synaptic inputs to ON-T and ON-S RGCs than they are to the relevant bipolar voltage responses (Fig 4).This indicates that the difference between the ON-T and ON-S responses originates at least in part in the conversion of bipolar voltage to glutamate release.

Ribbon size and vesicle replenishment
Prior work has shown that bipolar cells can exhibit profound short-term depression of glutamate release due to use-dependent depletion of the readily-releasable pool of synaptic vesicles (Burrone and Lagnado, 2000;Singer andDiamond, 2006, 2003;von Gersdorff and Matthews, 1997).Flash-evoked excitatory synaptic currents were suppressed in both RGC types at short interflash intervals, consistent with incomplete recovery from vesicle depletion at the time of the second flash (Fig. 7A top/middle, B).However, the extent of suppression was much stronger for currents measured in ON-T RGCs compared to ON-S RGCs.We quantified the extent of suppression using a paired flash ratio, which was calculated as the peak amplitude of the response to the second flash divided by the peak amplitude of the response to the first flash.At the shortest interval we tested (50 ms), the response to a second flash was almost completely suppressed in ON-T RGCs (paired flash ratio = 0.04 ± 0.01; mean ± SEM) whereas the response to a second flash in ON-S was suppressed considerably less (paired flash ratio = 0.29 ± 0.04).To rule out the possibility that a reduction in bipolar cell voltage response to the second flash could account for the strong paired flash suppression of excitatory input to RGCs, we also

(A) 3D reconstructions showing segmented axon stalk and axonal terminal (gray) and all the synaptic ribbons within the axonal terminal a type 5i (cyan) and a type 6 (magenta) BC. (B) Violin plot showing distribution of ribbon volumes (each data point is a ribbon) for the type 5i and type 6 BC in (A). The boxes indicate the interquartile range (thick lines, 25%-75%), median value (white dot), and 1.5 times interquartile range (thin line). Shaded gray areas represent rotated kernel density plots. (C) 3D reconstructions showing segmented dendritic arbor (gray) and all the synaptic ribbons presynaptic to these segments of ON-T (cyan) or ON-S (magenta) RGC dendrites. (D) Violin plot with scattered dots showing ribbon volumes for each ribbon presynaptic to a stretch of reconstructed ON-T or ON-S RGC dendrite. Each ribbon (dot) is colored differently for each BC type. (E) An EM micrograph (left) and 3D reconstruction (right) of a ribbon and its adjacent vesicles. The ribbon is indicated by the red arrow in the EM image. In the 3D reconstruction, yellow dots indicate the adjacent vesicles surrounding the presynaptic ribbon (red) in the BC terminals (gray). Postsynaptic ON-T RGC is in blue. (F) Scatter plot showing ribbon size in volume in relationship with the number of adjacent vesicles. Gray line represents the linear
regression model fitted to the data.Linear regression: β = 6.98e−06, p < 0.001, r 2 = 0.66.
We next looked for ultrastructural correlates of differential synapse function in bipolar cell axon terminals reconstructed from SBFSEM volumes.Interestingly, type 6 bipolar cell synaptic ribbons were ~2-3 fold larger than those of types 5i bipolar cells (Fig. 8A-B).These differences did not depend on the identity of the postsynaptic RGC partner (Fig. 8D).Ribbons in XBCs, type 5o and type 5t bipolar cells were also smaller than type 6 bipolar cell ribbons (Fig. 8D).This bipolar cell subtype-specific difference in ribbon size and the difference in ratio of type 6 to type 5i input between RGC types meant that ribbons were on average smaller at synapses made on ON-T compared to ON-S dendrites (Fig. 8C-D).
We found that ribbon size of several ON bipolar cell types is strongly correlated with the total number of vesicles that are in close contact with the synaptic ribbon across different bipolar cell types (Fig. 8E-F).Ribbon-associated vesicles are thought to correspond to the releasable pool of vesicles, comprised of both a readily-releasable pool that are docked at the presynaptic active zone as well as a reserve pool of vesicles that are bound to the ribbon at more distal sites (Lagnado and Schmitz, 2015).These two vesicle pools exhibit distinct release kinetics, with the readily-releasable pool corresponding to the fastest phase of glutamate release and the reserve pool contributing to slower glutamate release that occurs with prolonged stimuli (Mennerick and Matthews, 1996;Neves and Lagnado, 1999).Our observations that the average excitatory synaptic input to ON-T RGCs is smaller and that most excitatory input to ON-T cells originates from bipolar synapses with fewer ribbon-associated vesicles are consistent with a role for vesicle depletion in establishing transient glutamate release onto these RGCs.The larger pool of ribbon-associated vesicles in type 6 bipolar cell terminals, which comprise ~ 55% of the synaptic input to ON-S RGCs, may support sustained synaptic transmission onto these RGCs.

Discussion
In this study, we sought to understand how kinetically distinct visual pathways are established in the retina by directly comparing the visual response properties of transient and sustained ON RGCs with their presynaptic bipolar cells.Our results show that bipolar cell synapses have a key role in establishing different temporal channels in ON circuits of the retina.Further, amacrine cell-mediated presynaptic inhibition does not appear to account for transient versus sustained bipolar cell synaptic output.Rather, bipolar cell subtype-specific differences in synaptic release properties likely generate kinetically distinct output to postsynaptic targets.Our anatomical analysis suggests that these differences may at least in part be due to differences in synaptic vesicle pool size.
A key aspect of our work is that we measured RGC and bipolar cell responses using identical stimuli and under the same experimental recording conditions.To our knowledge, a direct comparison of RGC responses with those of their presynaptic bipolar cells has not previously been reported.The importance of measuring RGC and bipolar cell responses under the same conditions is highlighted by our observation that excitatory synaptic currents measured in ON alpha RGCs exhibited quantitatively different kinetics, although relative differences between types were preserved, when measurements were made from retinas maintained in ACSF versus Ames' bath solution (compare Fig. 2 (Ames') and Figs.6 and S8 (ACSF)).
We found that the bipolar cell subtypes that provide the major input to ON-T (type 5i) or ON-S RGCs (types 6 and 7) had indistinguishable light-evoked voltage responses (Figure 4).This was not entirely surprising given that most or all ON cone bipolar cell subtypes are indirectly coupled to other nearby ON cone bipolar cells via electrical synapses between bipolar cell axon terminals and the distal processes of AII amacrine cells.These electrical synapses form the basis for the primary rod pathway that mediates vision under dim light conditions, but lateral signals can also flow among neighboring bipolar cells under conditions where cone signaling dominates (Demb and Singer, 2012).Indeed, a substantial fraction (~30-50%) of light-evoked responses in the bipolar cell types we investigated here can be attributed to lateral signals that arrive via electrical synapses at photopic light levels (Kuo et al., 2016).Each AII amacrine cell is electrically coupled to multiple ON cone bipolar subtypes as well as to other AII amacrine cells (Marc et al., 2014;Tsukamoto and Omi, 2017;Veruki andHartveit, 2002a, 2002b).Different ON cone bipolar cell subtypes may also be directly coupled via gap junctions to each other (Sigulinsky et al., 2020;Tsukamoto and Omi, 2017).
Direct or indirect (via AII) coupling may limit the extent to which different bipolar cells can generate subtype-specific voltage responses.However, our findings differ from a previous study that found some differences among mouse ON cone bipolar cell subtypes.Ichinose et al., (2014) reported that a subset of type 5 bipolar cells, which the authors termed type 5f, and type 7 bipolar cells had more transient responses than type 6 bipolar cells.Methodological differences may account for the discrepancy between our results and the previous work by Ichinose et al., (2014).For example, the previous study measured bipolar cell responses using whole-cell recordings and included inhibitory receptor antagonists in their experimental solutions whereas we used gramicidin perforated-patch recordings and left inhibitory signaling intact.
Regardless of the reasons for the different findings, our observation that bipolar cell subtypes exhibited similar responses under the same experimental conditions for which we observed obvious differences among postsynaptic RGCs argues that mechanisms upstream of bipolar cell glutamate release (e.g.cone to bipolar cell synapses or intrinsic voltage response properties of bipolar cells) are unlikely to have a major role in shaping RGC type-specific responses for the stimuli we used here.
A limitation of our work is that we did not record from every one of the bipolar cell subtypes that are presynaptic to ON-T RGCs.Instead, we measured responses from type 5i and type 6 bipolar cells, which together comprise >~50% of the synaptic input to ON-T RGC dendrites.It is possible that the other presynaptic bipolar cell types (type 5o, type 5t and XBC) may have more transient light-evoked voltage responses that could contribute to the transient excitation measured in ON-T RGCs.In fact, a previous study in mouse retina found that XBCs and a subset of type 5 bipolar cells express voltage-gated Na + channels (Hellmer et al., 2016).These channels may enhance response transience or might even contribute to the generation of lightdriven all-or-none spiking responses in these bipolar cells (Ichinose et al., 2014;Puthussery et al., 2013;Saszik and DeVries, 2012).Despite the possibility that intrinsic response properties of the bipolar cell types we did not record from could contribute to the generation of transient responses in ON-T RGCS, we consider it unlikely that this alone could not account for the kinetics of the excitatory synaptic currents we measured in these RGCs.Type 5i and type 6 bipolar cells have a significant sustained component of their light step-evoked voltage response (Fig. 4E-F), whereas excitatory currents measured for the same stimulus and under the same conditions in ON-T RGCs had little to no sustained component (Fig. 2D-E).Even if the synapses formed by type 5o, type 5t and XBC provide a major fraction of the functional synaptic input to ON-T RGCs (despite our anatomical estimate that together they comprise <50% of the total ribbon input), we would still expect to observe some sustained excitatory input if type 5i and type 6 bipolar cell synaptic output follows the voltage response of these cells.Thus, a marked transformation of the sustained bipolar cell membrane voltage response to a transient synaptic current must take place.Our measurements of extracellular glutamate at ON-T and ON-S dendrites (Fig. 5) suggest much of this transformation occurs at the level of bipolar cell glutamate release.It is interesting to note in this context that ON-T and ON-S RGCs both receive input from type 6 bipolar cells (comprising ~20% and ~50% of total synaptic input, respectively).Due to differences in dendrite stratification, type 6 bipolar cell synapses onto ON-T RGCs arise from a region of the axon that is located more proximal to the bipolar cell soma than the synapses formed onto ON-S RGCs, which are found within the distal axon lobules.However, we did not observe any differences in ribbon size at distal versus proximal locations in type 6 bipolar cell axons (not shown).Whether postsynaptic target-specific synapse function, as has previously been reported in salamander retina (Asari andMeister, 2014, 2012), contributes to signal diversification in circuits involving type 6 bipolar cells will require further investigation.
Our results indicate that amacrine cell-mediated inhibition at bipolar cell synaptic terminals is not necessary for generating transient glutamate release kinetics onto ON-T RGC dendrites.This is seemingly at odds with prior work that has noted an essential role for surround inhibition in establishing type-specific bipolar cell glutamate release dynamics (Franke et al., 2017).
However, we note that our analysis here focused only on whether presynaptic inhibition shapes the kinetics of excitatory inputs to ON-T RGCs.Our anatomical and pharmacological experiments together argue that presynaptic inhibition is not an important factor in the transformation from sustained bipolar cell voltage responses to transient excitatory synaptic currents in ON-T RGCs.This is not to say that amacrine cell-mediated inhibition does not have a role in generating subtype-specific response properties for other aspects of bipolar cell function.Indeed, inhibition does have in important role in establishing differential spatial tuning of excitatory inputs to the RGC types we studied; excitatory synaptic currents in ON-T RGCs exhibit stronger surround inhibition than ON-S RGCs, and this surround inhibition is reduced by bath application of GABA receptor antagonists (Fig. S8C).
We conclude that transient versus sustained excitatory inputs to ON-T and ON-S RGCs most likely arise from bipolar cell subtype-specific differences in synaptic release properties.
Excitatory inputs to ON-T RGCs exhibited stronger paired flash depression than those to ON-S RGCs (Fig. 7), which indicates distinct vesicle cycling dynamics at the bipolar cell axon terminals presynaptic to the different RGC types.Our ultrastructural analysis suggests that bipolar subtype-dependent differences in ribbon-associated vesicle pool size could be a factor in establishing transient versus sustained glutamate release kinetics (Fig. 8).This finding is consistent with the idea that sustained transmission from ribbon synapses is determined by the rate of replenishment of the readily-releasable pool of synaptic vesicles following exocytosis (Lagnado and Schmitz, 2015).It is important to note, however, that type 7 bipolar cells (~45% of excitatory input to ON-S RGCs) have small synaptic ribbons, yet iGluSnFR imaging of type 7 bipolar cell axon terminals found that these cells exhibit sustained glutamate release (Srivastava et al., 2022).Additional related mechanisms that we did not directly investigate here may contribute as well.For example, presynaptic resting potential has a strong effect on steady-state vesicle release and thus stimulus-evoked release dynamics.Exocytosis that occurs when bipolar cells rest at voltages near release threshold can lead to a depletion of readily-releasable vesicles that limits the amount of release that occurs upon further light-induced depolarizations (Grimes et al., 2014;Jarsky et al., 2011;Oesch and Diamond, 2011).Thus, an initial phasic component of release can be reduced by depolarized resting potentials whereas hyperpolarized potentials can facilitate greater occupancy of vesicles in the releasable pool, which can enhance response transience.Bipolar subtype-dependent differences in release threshold due to differential Ca 2+ channel expression/regulation (Pangrsic et al., 2018) and/or coupling between Ca 2+ channel activity and vesicle release (Burrone et al., 2002;Burrone and Lagnado, 2000) could also impact release dynamics.Several aspects of excitatory synaptic currents in the two RGC types we studied here are consistent with a situation in which bipolar cell inputs to ON-T RGCs have a higher threshold for glutamate release compared to those at ON-S RGCs; ON-T RGCs receive less spontaneous excitatory synaptic input than ON-S RGCs (Kuo et al., 2016) and stimulus-evoked EPSCs are slower, sparser and more biased towards larger amplitude events in ON-T RGCs compared to those in ON-S RGCs (Fig. 2, Fig. S1B).Further investigation will be required to determine whether this is due to differences in axon terminal resting potential and/or other presynaptic factors.We did not find evidence for subtype-specific differences in resting membrane potential among type 5i, type 6 and type 7 bipolar cells, but our somatic voltage recordings likely have limited ability to resolve small voltage differences at the bipolar cell axon terminals.
Our results do not rule out additional roles for postsynaptic mechanisms in establishing transient versus sustained RGC spike output.For example, delayed amacrine cell-mediated inhibitory synaptic input to RGC dendrites could truncate spiking responses (Nirenberg and Meister, 1997).RGC cell-intrinsic mechanisms, such as RGC subtype-specific differences in voltagegated Na + channel properties and/or expression pattern could also contribute to distinct response kinetics (Brombas et al., 2022;Chang et al., 2023;Werginz et al., 2020).While a combination of pre-and postsynaptic factors may ultimately act in concert to generate transient versus sustained RGC spike output (Warwick et al., 2018;Werginz et al., 2020), our results establish a fundamental role for bipolar cell synapses in the segregation of visual input into kinetically distinct channels in ON pathways of the retina.
Transgenic mouse strains were backcrossed into the same genetic background as wild-type mice (C57BL/6J; Jackson Laboratory).Animal care and handling followed procedures approved by the Institutional Animal Care and Use Committee of the University of Washington or the University of Victoria Animal Care Committee.
Electrical signals were acquired at 10 kHz and low-pass filtered at 3 kHz using a Multiclamp 700B or Multiclamp 700A amplifier (Molecular Devices) and Symphony Data Acquisition software (http://symphony-das.github.io).The somas of different ON bipolar cell subtypes were targeted for gramicidin perforated-patch current-clamp recordings using 2P laser scanning excitation (960nm) in retinal slices prepared from transgenic mouse lines in which bipolar cell subtypes expressed fluorescent proteins.Tissue from Grm6-tdTomato mice, in which a sparse subset of ON bipolar cell types (type 6, 7 and 8) express tdTomato was used to target recordings to type 6 bipolar cells (Morgan et al., 2011;Schwartz et al., 2012).Type 7 bipolar cells were targeted for recording in Gus8.4-EGFP mice, which express EGFP in a type 7 bipolar cells and rod bipolar cells (Huang et al., 2003).Type 5i bipolar cells were targeted in retinas from Gjd2-EGFP mice, which we found express EGFP selectively in these bipolar cells (Fig. S6) as well as less robustly in an uncharacterized amacrine cell type (not shown).For bipolar cell recordings, we minimized light exposure from the 2P laser by restricting scan regions to the inner retina, away from photoreceptors, and by keeping laser exposure brief (<~30 sec/scan region; <2 mW post-objective laser power) (Kuo et al., 2016).After identification of fluorescent cell bodies, we acquired recordings using infrared differential interference contrast microscopy.
Bipolar recordings used an internal solution containing 123 mM K-aspartate, 10 mM KCl, 10 mM HEPES, 1 mM MgCl 2 , 1 mM CaCl 2 , 2 mM EGTA, 4 mM Mg-ATP, 0.5 mM Tris-GTP, 100 µM Alexa Fluor 594-hydrazide and 25-30 µg/mL gramicidin D (280 mOsm/L; pH 7.2).Pipettes (~7-9 Ω tip resistance) were front-filled with gramicidin-free internal solution to facilitate formation of high resistance (>1 GΩ) seals between patch pipettes and the cell membrane.After seal formation, series resistance (R s ) was monitored in voltage-clamp mode by applying a 10 mV voltage step across the pipette tip.Experiments were initiated once R s declined to <50 MΩ, which typically took 20-30 min.During recordings, we occasionally confirmed the integrity of the cell membrane under the recording electrode by monitoring Alexa Fluor 594 fluorescence using 2P excitation; if any dye was visible in the cell body, we abandoned the recording.No holding current was applied during current-clamp recordings from bipolar cells.At the conclusion of each bipolar cell recording, the membrane under the patch pipette was intentionally ruptured and fluorescence images of Alexa Fluor 594-filled cells were acquired using 2P microscopy to confirm bipolar cell subtype identities (Wässle et al., 2009).RGC recordings were performed in a flat-mount preparation of the retina from wild-type mice or Gjd2-EGFP transgenic mice following previously described methods (Kuo et al., 2016) Excitatory synaptic currents were isolated in whole-cell voltage-clamp recordings by holding RGCs near the reversal potential for Cl --mediated conductances (-68.5 mV).Patch pipettes (2.5-4 MΩ) were filled with an internal solution containing (in mM): 105 Cs-methanesulfonate, 10 Tetraethylammonium-Cl, 20 HEPES, 10 EGTA, 2 QX-314, 5 Mg-ATP, 0.5 Tris-GTP (~280 mOsm; pH 7.2).Series resistance (<15 MΩ) was compensated by 70% using the amplifier circuitry.Inhibitory receptor antagonists (SR-95531, TPMPA, strychnine) were obtained from Tocris or Sigma Aldrich and added to the bath solution from concentrated aqueous stock solutions.

iGluSnFR expression and two-photon imaging
Intravitreal injection of a cre-dependent viral plasmid (pAAV.hSyn.Flex.iGluSnFR.WPRE.SV40) was performed into Kcng4-cre mice to selectively express the glutamate sensor, iGluSnFR, in alpha RGCs, including ON-T and ON-S RGCs (Krieger et al., 2017).All injection procedures, tissue preparations, and subsequent two-photon imaging were performed as previously described in Srivastava et al., 2022.Briefly, for iGluSnFR imaging, the laser was tuned to 920 nm and all recordings were acquired at a frame rate of 58.25 Hz (256 x 256 pixels).In KCNG4-Cre mice, ON-S alpha RGCs were first identified by their spiking properties using extracellular cell-attached recordings.To identify the dendrites of ON-S RGCs, cells were filled with sulphorhodamine 101.ON-T RGC dendrites were then recorded at the same location but deeper into the retina (Krieger et al., 2017).

Visual stimulation
Visual stimuli were presented from a photopic mean background illumination (~3,300 rhodopsin isomerizations per rod per second (R*/rod/s)) to focus specifically on cone-mediated signaling (Grimes et al., 2018).Visual stimuli for electrophysiological recordings were generated using Stage Visual Stimulation Software (http://stage-vss.github.io).For all electrophysiological recordings except those in Figs.6C-F and S8, light stimuli consisted of a spatially uniform circular spot (520 µm diameter) delivered from a LED (405 nm peak output; Hosfelt) that was focused through the microscope substage condenser (Kuo et al., 2016).In the experiments shown in Figs. a Digital Light Processing (DLP) projector driven by a 405 nm LED (Wintech Digital) was used to project spots of various sizes through the microscope objective (Kuo et al., 2020).For iGluSnFR imaging (Fig. 5), background illuminance was kept around 1000 photons/µm 2 /s.For experiments, static spots of different sizes were generated and presented using StimGen software (https://github.com/benmurphybaum/StimGen).Stimuli were presented via a DLP through the substage condenser focused onto the photoreceptor layer of the retina.All visual stimuli were presented 4 s after the start of the recording for a duration of 2 s.

Data analysis and statistics
Spike triggered average (STA) waveforms (Fig. 1) and linear-nonlinear (LN) models of lightdriven RGC (Fig. 2) and bipolar cell (Fig. 4) responses were derived from responses to randomly fluctuating light stimuli (Gaussian distribution of light intensities with standard deviation = 50% of mean intensity; 0-60 Hz bandwidth) as described previously (Chichilnisky, 2001;Kim and Rieke, 2001;Rieke, 2001).For LN model construction, the linear filter was calculated by computing the cross-correlation of the light stimulus with the measured voltage (bipolar cell) or current (RGC) response and dividing the result by the power spectrum of the stimulus.The static nonlinearity was constructed from a point-by-point comparison of the predicted filter output with the measured response and binning the result along the prediction axis.In almost all cases, the linear filter had a biphasic shape with initial and secondary lobes that are opposite polarity.A small number of bipolar cells had triphasic filters (Fig. S7).We quantified the kinetics of each STA or linear filter by measuring the time at which the STA/filter waveform crossed zero ("zerocross time"; e.g.Fig. 1C, left) between the first and second lobes of the filter (e.g. between peak and trough of STA waveforms) and by calculating a biphasic index = |A| / |B|, where |B| is the absolute peak amplitude of the initial lobe of the filter and |A| is the amplitude of the peak amplitude of the secondary lobe (e.g.Fig. 2C, right).We quantified the kinetics of responses to light step stimuli by calculating the time between light step onset and the peak amplitude response ("time to peak"; e.g.Fig. 2E, left) and dividing the mean steady-state response (calculated from the last 100 ms light step increment) by the peak amplitude initial response at step onset (and multiplying by 100) (e.g.Fig. 2E, right).We note that alternative measures of step response or linear filter kinetics than those presented in Fig. 4 were also not statistically different between bipolar cell subtypes (not shown).Statistical tests were performed using MATLAB (Mathworks) or Igor Pro (WaveMetrics) software.

Near-infrared branding (NIRB) and serial block face scanning electron microscopy (SBFSEM)
Two neighboring ON-T RGCs were physiologically identified in Gjd2-EGFP mice.After identifying ON-T RGCs via their characteristic step-evoked extracellular spike response (e.g. Figure 1B), they were filled with Alexa 488 or Alexa 555.Dendritic arbors and T5i-BC axonal arbors were imaged using a confocal microscope.The retina was then fixed with 4% glutaraldehyde for 30 minutes at room temperature (RT).The retina was washed with 0.1M Na+ Cacodylate buffer for three times and flat-mounted with 0.1M Na+ Cacodylate buffer.To relocate the two RGCs under electron microscopy, we applied the near-infrared branding (NIRB) technique previously published to burn fiducial markers in the retina (Bishop et al.).After NIRBing, the retina tissue was unmounted and fixed in 4% glutaraldehyde overnight.The retina was processed for SBFSEM subsequently according to the protocol described previously (Glumi paper).Briefly, the tissue was initially treated with reduced osmium for 1 hour in the refrigerator.
Subsequently, after washing with double-distilled H2O (ddH2O), the tissue was placed in a freshly prepared thiocarbohydrazide solution for 20 minutes at RT.Following another round of ddH2O rinsing, the tissue was incubated in 2% osmium tetroxide for 30 minutes at RT.After rinsing, the sample was incubated in 1% uranyl acetate at 4°C overnight.Post-wash, the tissue was stained with Walton's lead aspartate for 30 minutes at 60°C.Following another wash, the retinal piece underwent dehydration using a graded alcohol series (20%, 50%, 70%, 90%, 100%, 100%), each step lasting 5 minutes, followed by two changes of 100% propylene oxide at RT for 10 minutes each.Finally, the tissue was embedded in Epon.The resulting block was then trimmed and mounted in the SBEM microscope (GATAN/Zeiss SIGMA, 3View), and image stacks were acquired at a voxel size of 5 x 5 x 50 nm.

Figure 1 .
Figure 1.Distinct temporal spiking response properties of ON-transient versus ON-sustained RGCs.(A) Maximum intensity projection z-stack fluorescence images of ON-T (left, cyan) and ON-S (right, magenta) RGCs dialyzed with Alexa Fluor 594 during whole-cell patch-clamp recording.Scale bar 50 µm.(B) Characteristic spiking responses to light step increment of example ON-T and ON-S RGCs.(C) Example spiking responses to repeated trials of spatially uniform (300 µm spot diameter) Gaussian noise stimulus (top black trace) presented with a mean luminance of ~2400 P*/Scone/s (~3000 R*/rod/s).ON-T (cyan) and ON-S (magenta) RGCs were recorded from sequentially in the same piece of retina.PSTHs (bottom) calculated from 20 repetitions of noise stimulus.(D) ON-T and ON-S STAs derived from responses to nonrepeated noise stimuli.Colored lines and shaded areas show mean ± SEM STA waveforms from ON-T (n= 15) and ON-S (n= 13) RGCs.Note time axis is reversed to facilitate comparison with linear filters in Figures 2 and 4. (E) Quantification of STA kinetics in different RGCs.Open circles show measurements from individual cells and filled circles with error bars show mean ± SEM values.Significance values from Wilcoxan rank sum tests.

Figure 2 .
Figure 2. Distinct excitatory synaptic input to ON-T and ON-S RGCs.(A) Excitatory synaptic currents measured in example ON-T (cyan) and ON-S (pink) RGCs in response to same Gaussian noise stimulus as in Fig. 1.Traces show average currents from ten repetitions of the same stimulus and are baseline subtracted to facilitate comparison.(B) Linear filters computed from responses to non-repeated noise stimulus presentation.Lines and shaded areas show mean ± SEM of responses from ON-T (n= 13) and ON-S (n= 15) RGCs.(C) Quantification of filter kinetics in different RGCs.(D) Excitatory currents measured from example ON-T and ON-S RGCs in response to a 0.5 second step increase in light intensity (50% contrast; mean response to five repetitions of stimulus).(E) Quantification of step response kinetics (n= 19 ON-T RGCs; n= 13 ON-S RGCs).In (C) and (E) open circles show measurements from individual cells and filled circles with error bars show mean ± SEM values.Significance values from Wilcoxan rank sum tests.

Figure 3 .
Figure 3. Identification of bipolar cell subtypes presynaptic to ON-T RGCs.(A) Maximumprojections z-stacks of two dye-filled ON-T RGCs (green) in wholemount retina from Gjd2-EGFP (red) transgenic mice.(B) Traced skeletons of dendritic arbors of the two ON-T RGCs from EM volume.(C) An EM micrograph showing a ribbon synapse (arrowhead) on a BC axonal terminal onto the dendrite of a ON-T RGC.(D) An example skeleton for each presynaptic BC type (types 5i, 5t, 5o, X, 6 and 7) of ON-T RGCs.Top: whole-mount view, Bottom: side view.(E) Bar plot Figs. 1-2) in example type 5i (cyan), type 6 (magenta) and type 7 (green) bipolar cells.Voltage traces are average responses to ten repetitions of the noise stimulus.(C) Linear filters computed from responses to non-repeated noise stimulus presentation.Lines and shaded areas show mean ± SEM of responses (type 5i, n= 11; type 6, n= 7; type 7, n= 6).(D) Quantification of response kinetics from linear filters.No significant differences across types (p = 0.19 zero-cross times; p= 0.89 biphasic index; Kruskal-Wallis test).(E) Responses to step light increment.Lines and shaded areas show peaknormalized mean ± SEM responses across type 5i bipolar cells (n= 8) and type 6 bipolar cells (n= 5).(F) Quantification of step response kinetics.Time to peak and steady-state response amplitude were not different between type 5i and type 6 bipolar cells (p= 0.72 and p= 0.35, respectively; Wilcoxan rank sum test).
currents measured in response to light step increment in control conditions (cyan) and in the presence of GABA receptor antagonists (10 µM SR95531 and 50 µM TPMPA; dashed gray) in an example ON-T RGC.Traces are mean response from six repeats of stimulus presentation.(D) Summary of steady-state amplitudes measured from step responses in control or with GABA receptor antagonists.No significant difference between conditions (p= 0.063; Wilcoxan signed rank test) (E) Control excitatory currents (cyan) or with addition of strychnine (1 µM; dashed gray) to bath in an example ON-T RGC.(F) Summary of strychnine experiments.No significant difference between conditions (p= 0.438; Wilcoxan signed rank test).In (D) and (F), open circles with lines show measurements from individual ON-T RGCs.Filled circles with error bars show mean ± SEM. amp.) c o n t r o l s t r y c h n i n e

Figure 6 .
Figure 6.Distinct glutamate input kinetics at ON-S vs ON-T RGC dendrites.(A) ON-S RGC dendrites expressing iGluSnFR in a KCNG-Cre mouse were filled with sulforhodamine 101 (left) after measuring their spiking response to a 200 µm spot extracellularly (Right).Yellow bands indicate stimulus duration.(B) iGluSnFR expressing ON-S and ON-T dendrites at different depths of the inner-plexiform layer (Left) were imaged using two-photon microscopy.Example iGluSnFR signals extracted from small regions of interest (ROIs 1-8) from ON-S and ON-T RGC dendrites illustrate the kinetic differences in their respective inputs.Black, mean responses; gray, individual trials.(C) Distribution of steady-state response amplitudes of the individual (gray) and average (black) ROIs from different recordings done from ON-S and ON-T dendritic field of views evoked by a 200 µm static spot (n = 7 FOVs, 6 retinas for ON-S and 5 FOVs, 5 retinas for ON-T RGC dendrites, *p<0.001;t-test).
The iGluSnFR experiments used an ACSF solution, like the patch-clamp experiments in Figs. 5 and S8.Both iGluSnFR responses and excitatory synaptic inputs measured in ACSF showed a larger sustained component of the step response at both ON-T and ON-S dendrites compared to currents measured in Ames (compare Fig. 2D-E, Fig. S8A-B and Fig.6B-C).This observation highlights the importance of comparing responses between different RGC types, as well as between RGCs and bipolar cells, under identical recording conditions.
Strong vesicle depletion at the cone bipolar cell synapses onto ON-T RGCs could generate more transient responses in these cells compared to ON-S RGCs.To examine this possibility, we used a paired flash stimulus paradigm, in which two brief light flashes separated by varying time intervals were presented while measuring excitatory synaptic currents in ON-T or ON-S RGCs.As in Figures 1, 2 and 4, we used Ames' medium for tissue preparation and maintenance in these experiments.
measured type 5i bipolar cell responses to the paired flash stimuli.The amplitude of flashevoked voltage responses in type 5i bipolar cells was unchanged by a preceding flash at all time intervals (Fig.7A bottom, C), which argues that paired flash suppression arises at the bipolar cell synaptic terminal.These results are consistent with the idea that differences in the extent of stimulus-evoked vesicle pool depletion in bipolar cells could underlie transient versus sustained excitatory input to ON-T and ON-S RGCs.

Figure 7 .Figure 8 .
Figure 7. Stronger flash-evoked suppression of excitatory currents in ON-T versus ON-S RGCs.(A) Paired flash-evoked excitatory currents measured at different inter-flash intervals (50, 100, 200, 400 ms) in an example ON-T (top; cyan) and an ON-S (middle, magenta) RGC.Bottom black traces show voltage responses to same stimulus measured in an example type 5i bipolar cell.Arrows at top indicate timing of first flash (black) and second flashes (gray).Flashes were 400% contrast and 10 ms duration.Traces show mean responses to five repeats of each paired flash interval.(B) Filled circles with error bars show mean ± SEM paired-flash ratios at different inter-flash intervals across all ON-T RGCs (cyan; n= 6), ON-S RGCs (magenta; n= 5) and type 5i bipolar cells (black; n= 6).

Figure S2 .Figure S3 .
Figure S2.Linear-nonlinear models of excitatory synaptic input to ON-T and ON-S RGCs.(A) Linear filters computed from responses to non-repeated noise stimulus presentation.Lines show filters from individual ON-T (cyan; n= 13) and ON-S (magenta; n= 15) RGCs.Filters were normalized to the peak amplitude of the primary inward component to facilitate visual comparison of temporal responses.Same cells as shown in average filters in Fig. 2B.(B)Nonlinearities derived by plotting the predicted response (convolution of linear filter and stimulus) against the measured excitatory currents for each RGC shown in (A).Linear predictions were calculated using non-normalized filters.

Figure S4 .Figure S5 .Figure S6 .
Figure S4.ON-T and ON-S RGC dendrites ramify in distinct IPL sublaminae.(A) Side projections from 2P z-stack images of dye-filled RGCs (ON-T: top, cyan; ON-S: bottom, magenta) and EGFP signal (yellow) in retinas from Gjd2-EGFP mice.RGCs are same cells as shown in en face view in Fig. 1A (B) Quantification of fluorescence signals from dye filled RGCs in Gjd2-EGFP mouse retina as in (A).Colored lines and shaded areas show mean ± SEM fluorescence intensity with respect to distance from peak EGFP (yellow) fluorescence intensity (n= 6 ON-T, cyan and n= 4 ON-S, magenta, RGCs).

Figure S7 .Figure S8 .
Figure S7.Linear-nonlinear models of bipolar cell voltage responses.(A) Individual linear filters for all type 5i (cyan; n= 11), type 6 (magenta; n= 7), and type 7 (green; n=6) bipolar cells computed from responses to non-repeated noise stimulus presentation.Filters were normalized to the peak amplitude of the primary inward component to facilitate visual comparison of temporal responses.Same cells as shown in average filters in Fig. 4B.(B) Nonlinearities derived by plotting the predicted response (convolution of linear filter and stimulus) against the measured voltage response for each bipolar cell shown in (A).Linear predictions were calculated using non-normalized filters.