A hierarchical electrical synaptic circuit mechanism for integrative parallel visual processing in the retina

SUMMARY

Parallel visual processing begins with retinal bipolar cells, traditionally regarded as independent chemical synaptic channels. However, the circuit-level synaptic integration of chemical and electrical synapses within this network remains unclear. Using dual patch-clamp recordings and two-photon imaging in whole-mount retina, we systematically characterized synaptic transmission across 13 mouse and 2 human cone bipolar cell (CBC) types, revealing two distinct modes: a fast, direct chemical pathway and a slower, serial electrical-chemical circuit among both ON and OFF CBCs. In mouse, the slow mode generates spatially dispersed glutamate “clouds” that facilitate integration across CBC types. We discovered specific “driver” CBCs that distribute robust, sustained signals through a hierarchical, functionally rectified network, enhancing sensitivity to small, low-contrast stimuli in downstream retinal cells and thalamic neurons in awake mice. Our findings challenge the classical view of independent CBC channels, revealing an integrative, hierarchical electrical-chemical synaptic architecture that enhances visual detection and coding efficiency.

INTRODUCTION

Parallel information processing enables the nervous system to extract, process, and transmit sensory inputs through multiple channels simultaneously. In the vertebrate retina, light signals captured by photoreceptors are distributed to over a dozen bipolar cell (BC) types in the outer plexiform layer (OPL), which process and transmit visual information in parallel to distinct sublayers of the inner plexiform layer (IPL) ^1,2. This channel- and layer-specific organization is fundamental to visual computation and a canonical a model for neural information processing. In mice, one rod bipolar cell (RBC) type and 14 cone bipolar cell (CBC) types have been identified anatomically and genetically ^3-7. Despite substantial imaging and electrophysiological studies ^8-18, a systematic understanding of CBC type-specific synaptic function and circuitry—particularly the interplay between chemical and electrical synapses—is still lacking ^19-22. It remains unresolved whether parallel CBC channels operate independently or interdependently and whether they encode discrete or integrated visual information.

BC chemical synapses contain well-defined presynaptic ribbons that facilitate connectomic reconstructions ^4,23. In contrast, the functional roles of CBC gap junctions remain elusive^7,24. These electrical synapses are difficult to resolve with serial block-face EM ^4,23, but have been identified by transmission EM ²⁵. Gap junctions between AII amacrine and ON CBCs are essential for rod signal transmission under scotopic conditions ^26,27, and BC coupling has been linked to motion sensitivity ^21,28-31. However, the circuit mechanisms and functional roles of electrical coupling among specific ON CBC types under photopic conditions—and among OFF CBCs—remain largely unexplored.

Here, using dual patch-clamp recordings and two-photon (2P) imaging, we systematically characterize the electrophysiological properties and synaptic mechanisms of morphologically identified CBC types in flat-mount mouse and human retinas. We uncovered profound physiological diversity yet strong functional interdependence among CBC types, revealing a robust, hierarchically organized electrical synaptic network that integrates parallel CBC channels and enhances size and contrast sensitivity, spatiotemporal coding, and downstream signaling, including in the dorsal lateral geniculate nucleus (dLGN) of awake mice.

RESULTS

Electrophysiological signatures of morphologically defined CBC types in the flat-mount mouse retina

To investigate synaptic physiology and circuitry across all morphologically defined CBC types in an intact retina, we developed a flat-mount preparation that enables single and dual patch-clamp recordings combined with 2P imaging (Figure 1A). From >600 randomly sampled, morphologically reconstructed CBCs, we systematically characterized excitatory and inhibitory receptive field properties, background synaptic inputs, intrinsic current profiles, and input resistance across all 13 dendrite-bearing CBC types (Figures 1A-1E, S1, and S2). Using center–surround stimulation and voltage-step protocols, we identified distinct physiological signatures for each CBC type, enabling the establishment of a functional identification flowchart for all CBC types (Figure S3). For example, although BC5o, BC5t, and BC5i stratify at similar IPL depths (Figures 1D-E, S2A, and S2C), BC5o displayed a transient inward voltage-gated current followed by oscillatory activity (n=12), BC5t exhibited a transient inhibitory input at the onset of a 100-μm light spot (n=11), and BC5i lacked this feature (n=13) (Figures 1C and S1). BC3b was similarly distinguished from BC3a by its large, frequent spontaneous baseline IPSCs (n=10, Figures 1B and S1B). Together, this approach provides a functional framework for linking CBC physiology with structural organization.

Figure 1. Physiological signatures and two distinct synaptic transmission modes of CBC types in flat-mount mouse retina.

(A) Dual patch-clamp recording of BC4 (red) and W3 (green) under 2P imaging (top: z-projection; bottom: reconstructed cross-section). (B, C) Representative excitatory (green) and inhibitory (red) currents (evoked by center light spots of various diameters) and intrinsic voltage-activated currents (black, leak-subtracted) from OFF (B) and ON (C) CBCs. (D) Average axon terminal stratification profiles of CBC types reconstructed from 2P stacks (sample sizes: BC1,16; BC2, 19; BC3a, 12; BC3b, 11; BC4, 5; BC5o, 13; BC5t, 29; BC5i, 29; XBC, 7; BC6, 26; BC7, 16; BC8, 4; and RBC, 16). (E) Reconstructed cross-sections of CBCs (red) dual-recorded with W3 (green, top), ON SAC (red, bottom left), and ON-S alpha cells (green, bottom right). Scale bars: 25 μm. (F-H) Representative EPSCs (at −70 mV, E_Cl) from paired recordings with W3 (F), ON SAC (G), and ON-S alpha (H) cells, showing individual trial traces (F, G) and 3-5 trial averages (H). Arrows indicate fast ESPC components. (I) Normalized EPSCs from BC5o-W3 (blue, n=6) and BC5i-W3 (orange, n=5) pairs; thick lines, mean; thin lines, individual traces. (J-L) Onset delay (J), rise time (K), and decay time (L) for BC5i-W3 (red) and BC5o-W3 (blue) pairs (I); Mann-Whitney U test. (M) Distribution of fast (blue, peak within 7 ms) and slow (orange, mean over 25-250 ms window) EPSC components in CBC-W3 pairs. Data are mean ± SEM; n.s: not significant; *: p<0.05; **: p< 0.01; ***: p<0.001. See also Figures S1-S4.

A Noncanonical Mode of Signal Transmission by CBC Types

To directly measure CBC output, we performed paired recordings from CBCs and W3 ganglion cells in flat-mount TYW3 mouse ³² retinas (Figures 1E and 1F). W3 (or W3b) cells are identified as the most brightly labeled ganglion cells in this mouse line, with small, diffusely ramifying dendritic fields and light responses corresponding to the HD2 and UHD ganglion cell types ^33,34. The vast majority of W3 cells recorded in this study more closely resembled the HD2 type, distinguished by a slightly larger dendritic field (~150 μm, Figure 1A) and a stronger inhibitory surround compared to UHD cells (data not shown).

Depolarizing CBCs (−55 to −25 mV) evoked glutamatergic excitatory postsynaptic currents (EPSCs) in W3 cells for most CBC types (20-150 PA; Figures 1F and 1M), which were abolished by CNQX + CPP (Figures S4), and were substantially smaller than RBC inputs to AII amacrines (200-1000 pA) ³⁵. Unlike the fast, transient EPSCs from RBCs to AII amacrines, CBC-evoked EPSCs exhibited two distinct kinetic components: a fast, transient component and a slow, sustained component (Figure 1F). The fast component activated within ~2 ms, peaked rapidly, and decayed to half-maximal amplitude within ~20 ms, whereas the slow component emerged after ~10 ms and peaked at ~100 ms (Figures 1I-1L), often persisting throughout prolonged (1-s) depolarization (Figure 1F). Brief (2-ms) presynaptic pulses evoked only the fast component (Figure S4A), whereas longer (>10 ms) depolarizations elicited both. The slow component showed distinct presynaptic voltage dependence (Figures S4D and S4E) and lacked the paired-pulse depression typical of the fast component ¹⁶, indicating a mechanistically separable transmission mode (Figures S4).

Different CBC types produced characteristic fast/slow EPSC mixtures in W3 cells (Figures 1F and 1M). For example, BC3b and BC5o elicited predominantly the fast component, with only a minor slow component; BC3a, BC5i, BC6, and BC7 cells produced prominent slow EPSCs; and BC5t and BC2 cells generated both, although the fast component from BC2 was small. Similar dual-component transmission was observed in paired recordings between ON CBCs and other postsynaptic neurons, including ON starburst amacrine (SAC) (Figures 1E, 1G, and S4B), ON sustained alpha (ON-S α) ganglion (Figures 1E, 1H, and S4C), and ON-OFF direction-selective ganglion cells (data not shown), indicating a general feature of most CBC types.

Fast EPSCs appeared only when presynaptic CBC terminals co-stratified with and directly contacted postsynaptic dendrites (e.g., among BC5o-W3, BC5t-W3, BC7-ON SAC, and BC6-ON-S alpha pairs). Consistent with chemical synaptic connectivity found with SBEM ^23,36, ON SAC received fast EPSC inputs from BC7 (n = 5), BC5i (n=10), BC5t (n=10), and rarely from BC5o (one out of the 10 pairs recorded). In contrast, slow components were often detected without direct pre- and postsynaptic apposition (e.g., among BC6-W3, BC7-W3, and BC5i-ON-S alpha pairs, Figures 1E-1H), indicating transmission through an indirect, serial synaptic pathway.

Electrical coupling of CBCs underlies the slow transmission mode

The properties of the slow EPSC component suggested mediation by gap junctions, which are present in many CBC types but absent in RBCs ⁷. To test this, we selectively knocked out connexin 36 (Cx36), the predominant connexin type in the retina ²⁶, from ON BCs by crossing mGluR6-Cre ³⁷ and Cx36^fl mice ³⁸. Dual recording revealed a near-complete loss of slow transmission from BC5t, BC6, and BC7 cells to W3 cells after this conditional Cx36KO, while fast transmission remained intact (Figures 2B, 2D, and 2E, Table S1). In contrast, slow transmission from BC5i and BC5o to W3 cells was not significantly affected (Figure 2B), likely reflecting residual or compensatory expression of other connexin types, such as Cx45 ^39,40. As expected, slow transmission from OFF CBCs (e.g., BC2, Figures 2B and 2E) to W3 cells was unaffected, confirming the ON-BC specificity of the knockout. By comparison, gap junction blockers (MFA, 100 μM, or 18β-GA, 50 μM; 25 min preincubation or 5-15 min bath perfusion) abolished slow transmission from both ON and OFF CBCs to W3 cells, while sparing the fast component (Figures 2C-2E, S4J, and S4K).

Figure 2. Electrical coupling underlies a slow, sustained mode of synaptic transmission in CBCs.

(A-C) Fast and slow EPSCs in W3 cells (−70 mV) evoked by CBCs under control (WT, A), conditional Cx36KO (B), and WT with gap-junction blockers (MFA, 100 μM or 18-βGA, 50 μM, 25 min, C). (D, E) Summary of fast (D) and slow (E) EPSC components (Kruskal-Wallis with Dunnett’s post hoc test). (F, G) Dual voltage-clamp recordings from ON (F) and OFF (G) CBC pairs showing bidirectional, MFA (100 μM)-sensitive gap-junction currents (at −70 mV). (H-I) Intracellular BAPTA (10 mM) selectively blocked the fast EPSC (BC5o–W3) but spared the slow EPSC (BC5i–W3), indicating an indirect, coupling-mediated origin of the slow component. (J) Summary of BAPTA effects on fast (BC5o–W3) and slow (BC5i–W3) EPSCs. (K–N) Dual recordings from AII amacrine–W3 pairs (K, L), showing AII-evoked slow EPSCs in W3, blocked by MFA (100 μM) in a time-dependent manner (M), summarized in (N). Error bars, mean ± SEM. n.s., not significant; *p < 0.05; **p < 0.01; ***p < 0.001. See also Figure S4.

To directly demonstrate functional coupling between CBCs, we performed dual voltage-clamp recordings from adjacent CBCs. Gap junctional currents were detected between ON CBCs (Figure 2F; mean conductance 102 ± 23 pS; n=3 pairs: BC6-BC5t, BC5i-BC5o, and BC6-BC5o). Notably, MFA-sensitive gap junctional currents were also detected between OFF CBCs (Figure 2G, mean conductance 110 ± 24 pS; n=3 pairs: one BC2-BC1 pair and two BC2-BC2 pairs), providing the first direct physiological evidence of OFF CBC coupling in the mammalian retina.

Using membrane input conductance as a proxy for total coupling strength ^18,41, we compared input conductance across CBC types (Figure S4L). CBC types with high input conductance (e.g., BC2, BC5i, BC6, BC7) corresponded to those evoking large slow EPSCs in W3 cells (Figure 1M). Across all BC5i–W3 recordings, BC5i input conductance correlated strongly with slow EPSC amplitude (adjusted R² = 0.88; p < 0.0001; Figure S5M). Conditional Cx36KO significantly reduced input conductance in BC5t, BC6, and BC7, but only modestly in BC5i and BC5o, mirroring the residual slow transmission from these two types (Figure 2B). MFA reduced input conductance across all CBCs, but not RBC (Figure S4L), indicating widespread CBC coupling.

To isolate the slow transmission under paired recording, we included 10 mM BAPTA in the whole-cell pipette solution for CBCs to chelate presynaptic calcium and block synaptic release from the recorded cell. Five minutes of presynaptic BAPTA dialysis abolished the fast EPSC (e.g., in BC5o-W3 pairs) but left the slow EPSC intact (e.g., in BC5i-W3 pairs; Figures 2H-2J and S4N). Prominent MFA-sensitive slow EPSCs were also evoked in W3 cells by depolarizing AII amacrine in AII-W3 paired recordings (Figures 2K-2N), demonstrating the involvement of AII amacrines in gap-junction-coupled pathways.

Together, the above results identified two distinct CBC output pathways: a direct chemical pathway mediating fast transmission and a serial electrical-chemical pathway underlying the slow transmission mode.

Electrical coupling among parallel CBC channels in the human retina.

To test whether the slow CBC transmission mode is conserved across mammalian species, we performed dual patch-clamp recordings from CBC-RGC pairs in short-postmortem-interval human retinal explants (see STAR Methods). In an ON CBC-ON RGC (parasol-like) pair in peripheral flat-mount retina, both fast and slow EPSC components were observed (Figures 3A-3D). The fast component peaked at ~50 pA (at −70 mV) with rapid onset (<2 ms) and decay (~10 ms) and ran down within ~12 min under whole-cell recording (Figure 3B). The slow component peaked at ~40 pA, showed delayed onset (~10 ms) and prolonged decay (>50 ms), remained stable for > 17 min, and was abolished by MFA (100 μM, 5 min bath application) (Figure 3B), consistent with a serial electrical-chemical pathway similar to that in mouse (Figures 2H-2J). The slow component increased with presynaptic depolarization (Figure 3C) and exhibited paired-pulse facilitation (Figure 3D), matching its mouse counterpart (Figures S4D-S4G).

Figure 3. Fast and slow modes of CBC transmission in human retina.

(A) Two-photon images of a dual-patch-clamped ON CBC (red) and a putative ON RGC (green) in peripheral postmortem human retina. Top: maximum-intensity z-projection in IPL-b. Bottom: reconstructed cross-section. (B) EPSCs at −70 mV from the pair in (A), showing a fast component that ran down over time (orange arrows) and a persistent slow component blocked by MFA (100 μM). MFA acted faster in human than mouse retina, likely due to reduced tissue thickness. Right: expanded view of boxed region. (C) From the same recording as in (B), five minutes after break-in, presynaptic depolarization to −40 mV evoked primarily a fast EPSC (top), whereas depolarization to −20 mV evoked both fast and slow components. (D) Paired-pulse facilitation of the slow EPSCs in the cell pair from (B), recorded 15 min after break-in. (E) Dual recording from an OFF CBC (red) and a putative OFF midget RGC (green) in another postmortem human retina. (F) Fast and slow EPSCs from (E); MFA (100 μM,10 min bath perfusion) abolished the slow component, while the fast component persisted with rundown. (G) Dual recording from an ON CBC (red) and ON RGC (green) in yet another postmortem human retina, showing no detectable contact between the pair. Top: reconstructed cross-section. Bottom: same cross-section after digital tracing and reconstruction using ImageJ. (H, I) EPSCs from (G). Only a slow component was observed at −70 mV (H), exhibiting paired-pulse facilitation (I). (J) Kinetic comparison of normalized fast and slow EPSCs between mouse and human CBC-RGC pairs.

A similar pattern was observed in another donor retina between an OFF CBC and an OFF RGC (midget-like), where MFA blocked the slow EPSC before the fast component fully ran down (Figures 3E and 3F). In a third donor retina, only a slow (no fast) EPSC was recorded from an ON CBC-ON RGC (midget-like) pair lacking visible direct contact (Figures 3G and 3H). This slow EPSC also exhibited paired-pulse facilitation and conserved kinetics (Figures 3I and 3J), supporting a conserved serial electrical-chemical synaptic pathway from CBCs to RGCs in both mice and humans.

Notably, these experiments represent the first dual-patch recordings of synaptic transmission in flat-mount human (or primate) retina, demonstrating preserved or recovered synaptic transmission up to 2-6 hours postmortem. These findings extend previous reports of revived neuronal activity in postmortem pig brains and human retinas ^42,43 by demonstrating well-defined electrical and chemical synaptic communications in a specific circuit within the postmortem human retina.

Electrical coupling among CBCs mediates a slow and dispersed (“cloud”-like) glutamate signaling pattern.

To examine how the two CBC transmission modes shape glutamate signaling in the IPL, we combined single-cell patch-clamp stimulation with 2P iGluSnFR imaging in flat-mount mouse retinas transfected intraocularly with AAV2/2-SF-iGluSnFR.A184S (Figure 4A). Depolarizing a single RBC (−55 to −25 mV) elicited localized iGluSnFR “hot spots” at axon boutons, whereas stimulating individual CBCs (e.g., BC3a) generated dispersed glutamate “clouds” extending up to ~30 μm beyond the axon arbor (Figure 4B). Most CBC types evoked clouds with type-dependent spatial extent (Figures 4C, 4D, S5A). The largest clouds were produced by BC2, BC3a, BC5i, and BC7— types with high input conductance (Figure S4L) and strong slow EPSCs in W3 cells (Figure 1M). As with slow EPSCs, glutamate cloud formation required sustained depolarization (Figures S5B and S5C).

Figure 4. CBC coupling mediates dispersed (cloud-like) glutamate signaling in the IPL.

(A) Schematic of simultaneous patch-clamp stimulation of individual CBCs (red) and 2P iGluSnFR imaging (green) in flat-mount mouse retina, showing an example BC7 in cross-section. Dashed yellow line indicates the imaging plane. (B) Left: iGluSnFR signals evoked by 1-s depolarization (−55 to −25 mV) of RBC, BC3a, or BC6. Right: ΔF/F₀ measured from boutons (ROI1, blue circle) and surrounding annular region (ROI2, orange). (C) Lateral extent of iGluSnFR signals evoked by CBC depolarization in WT, mGluR6-Cre x Cx36fl/fl, and MFA-treated (100 μM, 25 min) WT mouse retinas. (D) Summary of iGluSnFR cloud sizes. Dunnett’s multiple comparisons test following Kruskal-Wallis test was performed (if n>3 per group). BC5i cloud size showed considerable variability in conditional Cx36 KO, likely due to heterogeneous residual or compensatory expression of non-Cx36 gap junctions. (E) Time course of glutamate cloud spread following 1-s BC5i depolarization (−55 to −25 mV), shown as snapshots (averaged over 50 ms, top) and low-pass–filtered (25 Hz) rising phases at increasing distances from boutons (bottom). Imaging rate: ~100 frames/s. (F) Summary of onset delays between boutons and sites 10 μm away. Paired Mann-Whitney U test performed. (G) Vertical (radial) spread of iGluSnFR signals evoked by CBC depolarization, reconstructed from z-stack images of responses to time-locked depolarization repeats. Yellow braces indicate approximate radial signal spread. (H) Summary of the extent of radial spread (% IPL width) for CBC-evoked iGluSnFR signals in WT (black), mGluR6-Cre x Cx36fl/fl (red), and MFA-treated WT (purple) retinas. Dunnett’s multiple comparisons test following Kruskal-Wallis test was performed for BC2, BC6. Results of individual multiple comparisons were labeled in color. (I) Lateral extent of iGluSnFR responses evoked by a 3-s light spot (25-μm diameter), showing 3-s signal averages during light ON in IPL-b (near S3/S4) and light OFF in IPL-a (near S2). (J) Summary of light-evoked cloud sizes WT (black) and mGluR6-Cre x Cx36^fl/fl (red) retinas. For all experiments in (A)-(H), L-AP4 (20 μM) and ACET (10 μM) were included in the bath. Error bars indicate mean ± SEM. n.s: not significant; *: p<0.05; **: p< 0.01; ***: p<0.001. See also Figure S5.

The lateral expansion of glutamate clouds showed a delayed onset (~90 ms) at sites 10–20 μm from the voltage-stimulated CBC boutons (Figures 4E, 4F), reminiscent of the slow EPSC kinetics under dual recording (Figures 1I-1K). Notably, in many cases, no significant difference in rise time was observed between neighboring ROIs at 10 and 20 μm distances (Figure 4E), suggesting glutamate release from electrically coupled CBC boutons at those sites, rather than passive glutamate diffusion from voltage-clamped CBC boutons. Moreover, iGluSnFR signals tended to form discrete spatial clusters in the clouds rather than a continuous gradient, again consistent with localized release rather than long-range diffusion.

MFA (100 μM) markedly reduced the glutamate cloud, restricting iGluSnFR signals to voltage-clamped bouton hot spots (Figures 4C, 4D). Conditional Cx36KO similarly reduced ON-CBC clouds in IPL-b but spared OFF-CBC clouds in IPL-a. BC5i clouds were variably reduced, likely reflecting residual or compensatory non-Cx36 coupling (Figures 2B, 2E, S4L).

Single-CBC stimulation also produced radial (inter-IPL sublayer) cloud spread (Figure 4G; Supplementary Data Videos S1-S4) in a CBC type-dependent manner (Figure 4H). BC6, which displayed high input conductance and a strong slow EPSCs (Figures 1F-H, 2A, and S4L), generated minimal lateral spread at its axon terminal boutons (Figures 4B and 4D), but prominent radial spread into other ON sublayers, where secondary lateral expansion was often observed (Figure 4G), presumably via coupled BC5 subtypes. However, glutamate clouds never crossed the ON-OFF IPL boundary during stimulation, indicating a lack of ON-OFF coupling (Figure 4G). At the offset of ON CBC (particularly BC6) stimulation, we occasionally observed OFF glutamate clouds in IPL-a at the offset of an ON CBC stimulation, presumably due to crossover disinhibition from ON CBCs to OFF CBCs (data not shown).

Conditional Cx36KO significantly reduced BC6-evoked radial spread of glutamate clouds but had little effect on BC2-evoked ones, whereas MFA (100 μM) suppressed radial spread from both BC6 and BC2 (Figure 4H). The broad radial spread observed from BC6, BC7, and BC5i (Figure 4H) matched their ability to evoke slow EPSCs in non-costratified targets (Figure 1F-H).

Interestingly, radial cloud spread showed directional bias: BC6/BC7 readily drove clouds into BC5t/BC5o IPL strata, whereas BC5t/BC5o showed weak spread toward BC6/BC7 strata (Figure 4G). This asymmetry mirrors the electrophysiological results—BC6 elicited robust slow EPSCs in W3 cells which receive direct inputs (fast EPSCs) from BC5t/BC5o (Figure 1M), whereas BC5t/BC5o failed to induce slow EPSCs in ON-S alpha cells which receive direct input from BC6 (Figure 1H), revealing directional rectification within the CBC electrical network (see Discussion).

To access physiological relevance, we imaged iGluSnFR responses to a small 25-μm light spot, a size selected to minimize lateral input from horizontal and amacrine cells. Light onset and offset triggered extended clouds in IPL-b (Figures 4I, imaged near S3/S4 border) and IPL-a (Figure 4J, at S2), respectively. Fast imaging from a narrow, elongated ROI spanning the clouds revealed a 93 ± 20 ms delay in rise phase between cloud center and periphery (80 μm from center) (n=6, Figures S5D and S5E). Compared with single-CBC–evoked clouds, light-evoked clouds spread farther and faster, likely due to recruitment of multiple electrically coupled CBCs. In conditional Cx36KO retinas, light-evoked cloud size was significantly reduced in IPL-b but not IPL-a (Figures 4I and 4J), demonstrating a critical role for CBC coupling in spatial integration of small visual signals.

Existence of “driver” CBCs

To determine how electrical coupling shapes the light response of individual CBCs, we compared the voltage responses of ON CBC types to small light spots (25-100 μm in diameter) between WT and conditional Cx36KO mice, yielding three main findings.

First, conditional Cx36KO markedly reduced the response amplitudes of multiple ON CBC types to small center flashes (25-μm, 600% contrast), indicating loss of excitatory drive normally supplied through electrical coupling (Figure 5A). The extent of the reduction varied by CBC type, with BC5t showing nearly complete suppression to small, low-contrast light spots (Figures 5A, 5C, and 5G). In contrast, BC6 responses increased two- to threefold after conditional Cx36KO (Figures 5A, 5C, and 5G), identifying BC6 as a “driver” cell that provides excitatory input to other ON CBC types via direct or serial gap-junction coupling (e.g., through AII amacrines). Consistently, BC6 receives the largest light-evoked current input among all ON CBC types tested (Figures 5D and 5H), exhibits strong coupling to other ON CBCs (Figure 1M, 4H and S4L) and AII ACs ⁷, and expresses high levels of TRPM1 and Cx36 transcripts ^5,6. This reveals a hierarchical and functionally interdependent organization among parallel CBC channels.

Figure 5. Effects of CBC coupling on CBC light responses.

(A-C) Voltage responses (current-clamped with no holding current) of example ON CBC types in whole-mount WT (black) and mGluR6-Cre x Cx36^fl/fl (red) mouse retinas, evoked by: (A) a 25-μm-diameter light flash (600% contrast) at varying distances from the receptive field (RF) center, (B) a 50-μm-dia light flash at RF center, and (C) a 50-μm-dia light spot (100% contrast) at RF center. Traces in (A) and (B) are from individual trials, while (C) shows averages from 3-5 trials. In mGluR6-Cre x Cx36^fl/fl retinas (B), BC7 typically responded to a 50-μm flash with either a single (thick red trace) or multiple bursts of regenerative events (thin red overlay), while BC5t responded with either a weak transient or no response (red trace). Yellow shading indicates the 3-s light ON period. (D) Light-evoked excitatory input currents recorded under voltage clamp (at −70 mV). (E-H) Summary of results from (A-D), comparing between WT (black) and mGluR6-Cre x Cx36^fl/fl (red) mice: (E) center RF size (half-width of Gaussian fit), (F) Sustainedness index (ratio of average response amplitude in the last second of a 50-μm flash to initial peak), (G) ΔV (average depolarization to a 50-μm, 100% contrast light spot over 3 s), and (H) current input (peak current from a 100-μm flash at −70 mV). BC5t responses of less than 1 mV (in mGluR6-Cre x Cx36^fl/fl retinae) were excluded from RF size and plateau index calculations. Mann-Whitney U test was performed. Error bars indicate mean ± SEM. n.s: not significant; *: p<0.05; **: p< 0.01; ***: p<0.001. See also Figure S6.

Second, in conditional Cx36 KO retinas, most ON CBC types exhibited more transient voltage responses to small center spots, losing the sustained component present in WT—a component presumably driven primarily by BC6 via electrical coupling (Figures 4A, 4B, and 4F). This effect was most pronounced in transient CBC types such as BC5o (Figures 5A, 5B, and 5F), suggesting that electrical coupling normally dampens response transience in these cells. Notably, BC7, previously reported to generate sustained Ca responses to light¹¹, showed an initial transient depolarization to a light spot in WT mice (also see ref. ¹³) and frequently displayed repetitive bursts of transient depolarization in conditional Cx36KO mice (Figures 5A and 5B). Moreover, BC7 cells exhibited distinct regenerative voltage responses to current injections in conditional Cx36KO, but not WT, mice (Figure S6).

Interestingly, whereas most ON CBC types became more transient and displayed a diminished sustained component after conditional Cx36KO, BC6 showed the opposite pattern—its light responses became slightly more sustained with less decay (Figures 5A-5C and 5F). These findings indicate reciprocal kinetic modulation between sustained and transient CBC types through cross-type electrical interactions, providing a novel mechanism for tuning response dynamics.

Third, conditional Cx36KO reduced the excitatory receptive field sizes of ON CBC types, particularly BC5t, BC6, and BC7, which no longer responded to 25-μm light spots outside their dendritic centers (Figure 5A). This effect was weaker in BC5o and BC5i, possibly due to residual non-Cx36 coupling (Figures 5A and 5E). Together, these findings demonstrate the contribution of CBC electrical coupling to spatial integration and help explain why CBC receptive fields exceed their dendritic fields ^21,44.

Functional effects of CBC coupling on downstream retinal neurons

To assess how CBC coupling shapes downstream processing, we compared light-evoked responses in W3, ON SACs, and ON-S alpha cells between WT and conditional Cx36KO mice (Figures 6A and 6B). In W3 cells, ON EPSCs were markedly reduced by Cx36KO in a spot size- and contrast-dependent manner, with the largest reduction near response threshold. For example, ON responses to a 50-μm diameter, 100% contrast spot decreased from 30.48 ± 3.3 pA (WT, n = 4) to 4.24 ± 1.6 pA (Cx36KO, n=4; p = 0.029), whereas OFF responses were not significantly affected (25.7 ± 2.3 pA, n = 4, in WT vs 34.6 ± 11.5 pA, n = 4, in Cx36KO, p=1) (Figures 6A and 6B). These reductions parallel diminished voltage responses (Figures 5A, 5C, and 5G) of BC5t and BC5o cells —the primary direct ON CBC inputs to W3 cells (Figures 1F, 1M, and S4A)—and became less pronounced with increasing spot size and contrast (Figures 6A and 6B). Conditional Cx36KO also significantly accelerated ON-EPSC decay (sustainedness index to a 100-μm flash: 0.089 ± 0.009, WT, n=17; 0.035 ± 0.004, Cx36KO, n=15; p<0.001), consistent with the loss of sustained voltage components in BC5t and BC5o (Figures 5B and 5F). On-cell recordings confirmed the loss of onset, but not offset, spike responses to small, low-contrast stimuli (Figures 6C and 6D). These results demonstrate that CBC coupling enhances the sensitivity to small stimuli at low contrast (detection acuity) and shapes temporal integration and coding in W3 cells.

Figure 6. Effects of CBC coupling on light responses of retinal ganglion and amacrine Cells.

(A) Light-evoked EPSCs from W3, ON SAC and ON-S alpha cells in response to light spots of varying sizes and contrasts in whole-mount retinas of WT (black) and mGluR6-Cre x Cx36^fl/fl (red) mice. Thin traces show individual cell responses (each averaged over 4–6 trials), and thick traces show pooled averages. (B) Heatmaps of light-evoked EPSCs (averaged during light-ON) across different spot sizes and contrasts in WT and mGluR6-Cx36KO mice. (C) Light-evoked spike responses from W3 and ON-S alpha cells under on-cell loose patch recordings in WT (black) and mGluR6-Cre x Cx36^fl/fl (red) retinas. (D) Quantification of ON spike counts (W3) and increase in spike rate (Δrate) (ON-S alpha cells) during light-ON. Mann-Whitney U test performed. Yellow shading indicates timing of the 3-s light ON period. All light spot sizes are shown in diameter. Error bars indicate mean ± SEM (D). n.s: not significant; *: p<0.05; **: p< 0.01; ***: p<0.001. See also Figure S7.

ON SACs showed similar impairments after conditional Cx36KO: ON EPSCs to a 50-μm-diameter, 100%-contrast spot decreased from 37.9 ± 16.85 pA (n=4, WT) to 6.4 ± 2.2 pA (Cx36KO, n=4; p = 0.029), and responses to a 100-μm-diameter flash became more transient (sustainedness index: 0.40 ± 0.02, n=7, WT; 0.17 ± 0.01, n=8, Cx36KO; p<0.001) (Figures 6A and 6B). These changes align with reduced voltage responses in BC5t, BC5i, and, to a lesser extent, BC7 cells (Figures 5C and 5E)—the major direct inputs to ON SACs (Figures 1G and S4B).

In contrast, ON-S alpha cells exhibited a more sustained EPSC response to a 100-μm-diameter flash after conditional Cx36KO (sustainedness index: 0.34 ± 0.03, n=11, WT; 0.43 ± 0.03, n=9, Cx36KO; p<0.05), consistent with the more sustained light responses of BC6, their primary direct input (Figure 1H). Notably, despite BC6’s enhanced response to a 50-μm, 100%-contrast spot in conditional Cx36KO (Figures 5A-5C, and 5G), the corresponding light-evoked EPSC in ON-S alpha cells was not enhanced, but markedly reduced (32.7 ± 8.8 pA, n = 4, WT; 8.5 ± 2.9 pA, n=6, Cx36KO; p=0.0095, Figures 6A and 6B), likely due to a depolarized BC6 background membrane potential (Figures S7A and S7B) caused by decreased input conductance (Figure S5C) and strong photoreceptor input (Figures 5D and 5H). This depolarization increased tonic glutamate release from BC6, elevating background EPSCs, holding currents, and spontaneous spiking in postsynaptic ON-S alpha cells (Figures S7E-S7H), but reducing light-evoked EPSCs and spike rate change (Δrate) (Figures 6C, 6D), even though the total EPSC amplitude (background + light-evoked) remained comparable to that in WT (Figure S7G). Thus, rather than sacrificing its sensitivity by supplying gap junction–mediated excitation to non-driver CBC types, BC6 benefits from this output (“leakage current”), which lowers its background membrane potential and noise level, placing it in a more sensitive operating range.

Importantly, the sustained (non-adapting) light-evoked voltage response of BC6 (Figures 5A-5C), together with the corresponding sustained EPSC component in ON-S alpha cells (Figures 6A and S7G), provides electrophysiological evidence for a critical role of BC6 in encoding ambient light intensity ⁴⁵. Moreover, through electrical coupling, BC6 distributes this intensity signal to other ON CBCs (Figure 5), enabling shared coding across CBC channels.

Impact of CBC coupling on visual responses of dLGN neurons in awake mice.

To investigate the role of CBC coupling on central visual processing, we used in vivo two-photon calcium imaging to examine light response properties of GCaMP8s-expressing neurons in the shell region of the dorsal lateral genicular nucleus (dLGN) in head-fixed, awake mice (Figure 7A; see STAR Methods for details). We focused on neurons that exhibited increased activity during either the ON or OFF phase of a light spot. After determining the retinotopic map of the field-of-view (FOV), we presented white light spots of varying sizes and contrasts at the center of the map and recorded calcium responses from these neurons.

Figure 7. Effects of CBC coupling on light responses of dLGN neurons in awake mice.

(A) Schematic of two-photon imaging of visually evoked dLGN calcium responses in awake mice. (B, C) Example traces of visually evoked calcium responses to white flash spots of varying sizes and 2° spots of different contrasts in WT mice (B) and mGluR6-Cre x Cx36^fl/fl mice (C). Yellow shading indicates timing of the 3-s light-ON period. (D, E) Example 2P calcium response images from FOVs of WT (D) and mGluR6-Cre x Cx36^fl/fl (E) mice. (F, G) Normalized size-tuning (F) and contrast-tuning (G) curves for ON-phase calcium responses of dLGN neurons. Normalization was performed to minimize the effect of variation in GCaMP8s signal levels across neurons. Statistical comparisons between WT and mGluR6-Cre x Cx36^fl/fl were performed using linear mixed-effects models (see STAR methods for details). At spot size of 0.5°, p=0.1 (F). (H, I) Similar analysis as (F, G) but for OFF-phase responses. Error bars indicate mean ± SEM (F-I). n.s: not significant; *: p<0.05; **: p< 0.01; ***: p<0.001.

To assess size tuning, neurons that were responsive to an 8° white spot flash (1000% contrast) were identified and then tested for their responses to light spots of varying diameters. In WT mice, the size-tuning curve of ON responses, normalized to each neuron’s maximal ON response, showed that responses progressively increased with light spot sizes ranging from 0.5° to 8° in diameter (Figures 7B and 7D). A similar trend was observed in conditional Cx36KO mice (Figures 7C and 7E). However, the normalized ON responses (Figure 7F), but not the OFF responses (Figure 7H), to small light spots (1° and 2°) were significantly reduced compared to WT mice (see STAR Methods for details of the linear mixed-effect model).

We next measured contrast-tuning curves from all neurons responsive to 2°-diameter, 200%-contrast spots, testing their responses to a range of contrast levels (Figure 7C). In conditional Cx36KO mice, we observed a significant reduction in the normalized responses to 50%-contrast and 100%-contrast spots in the ON phase compared to WT mice, but no significant change in the OFF phase (Figures 7G and 7I), suggesting a specific reduction of the ON responses to low-contrast objects. Notably, the reduction in the ON-phase calcium responses of dLGN neurons to 100%-contrast, 2° spots in conditional Cx36KO mice mirrors the reduction in the ON-phase spike responses of retinal ganglion cells to 100%-contrast, 50-μm spots (Figures 6C and 6D), consistent with the ~30 μm/° visual angle conversion in adult mouse retina ⁴⁶. Together, these findings underscore an important role of CBC electrical coupling in enhancing the central representation of small and low-contrast visual stimuli.

DISCUSSION

This study systematically characterized 13 morphologically defined CBC types in the whole-mount mouse retina—covering all conventional types ^3-5 except the dendrite-lacking BC1a type ⁴⁷—by correlating morphology with RF properties, intrinsic physiology, and electrical and chemical synaptic connectivity. Each CBC type was found to exhibit distinct electrophysiological signatures, enabling functional identification and detailed synaptic characterization under paired patch-clamp recording and 2P imaging.

We identified a serial electrical-chemical transmission mode that forms a hierarchical network integrating signals across parallel CBC channels. A similar mechanism operates in the human retina, where functional electrical and chemical synaptic transmission persist for several hours postmortem, providing a valuable model for human retinal physiology and disease. This electrical-chemical synaptic integration enhances sensitivity to small and low-contrast visual stimuli and shapes the spatiotemporal coding in downstream retinal and dLGN neurons in vivo.

Driver CBCs

A central finding of this study is the identification of “driver” cells among CBC types. BC6 cells receive strong cone input and generate large, sustained voltage responses that, through direct and serial coupling, enhance the responses of other ON CBCs. Consistent with this role, BC6 cells express high levels of TRPM1 and Cx36 transcripts ^5,6 and are extensively coupled to AII amacrines ⁷. BC7 cells, which also receive substantial cone input and exhibit an initial transient peak in voltage response to center light stimulation in WT as well as transient bursting in conditional Cx36 KO mice, may serve a related function by providing transient excitatory boosts to other ON CBCs, although this possibility requires further study.

With its characteristically sustained kinetics, BC6 strongly shapes the response dynamics of other ON CBCs through low-pass gap-junction coupling, revealing a novel mechanism for regulating response kinetics and temporal coding. The sustained background activity of BC6 also modulates baseline activity in coupled CBCs and in its direct postsynaptic targets, such as ON-S alpha cells, via tonic glutamate release (Figure S7) ⁴⁸. Notably, the background activity of BC6 is itself regulated by coupling with other CBCs and AII amacrines (Figure 5) ¹⁹. Moreover, electrical coupling enables BC6’s intensity-coding signals to be distributed across ON CBC types, offering a plausible explanation for the widespread presence of intensity coding across CBC types ⁴⁵ and suggesting an efficient strategy for reducing redundancy across parallel circuits.

Whether equivalent “driver” cells exist among OFF CBCs remains unresolved. Here, we provide the first functional evidence of electrical coupling between OFF CBC types. In particular, BC2 and BC3a evoked prominent slow EPSCs in W3 cells and generated strong glutamate clouds, raising the possibility that they play roles analogous to BC6 and BC7/BC5i. BC2 cells are coupled to each other and to other OFF CBCs ⁷, suggesting that a BC2/BC3a-centered gap junction network could support OFF-pathway electrical integration even in the absence of a known AII counterpart. Alternatively, BC6 may indirectly influence OFF CBCs via crossover disinhibition through AII amacrines at light offset ^7,24,49,50. These possibilities warrant further investigation.

Glutamate clouds

We uncovered a three-dimensional “cloud-like” pattern of glutamate signaling by CBCs. Lateral cloud expansion likely reflects coupling among CBC types that stratify within overlapping IPL sublayers, promoting spatiotemporal integration that enhances contrast and motion sensitivity of downstream targets ^19-21. This mechanism operates across multiple CBC types, including OFF CBCs. The largest lateral glutamate clouds arise from BC2, BC3a, BC5i, and BC7—types prominently engaged in motion-detection circuits, including those of SACs ^36,51 and DSGCs ^4,52.

Radial glutamate spread reflects crosstalk among different CBC types through direct or serial electrical coupling, integrating signals across parallel CBC channels. Although BC6 produces minimal lateral spread within its own terminal layer, it evokes strong radial spread to other ON CBC types, which then propagated signals laterally within their respective strata. Together, the integration of lateral and radial signal spread reveals a complex electrical network that adds a crucial layer of presynaptic computation at the very first stage of parallel visual processing.

Hierarchical and rectified organization of the coupled CBC network

The identification of “driver” CBCs reveals a hierarchical organization among parallel CBC channels, with signals flowing from driver cells to less responsive CBC types in an electrical synaptic network. Despite apparently symmetric (non-rectified) gap-junction conductance, single CBC-evoked signal propagation within this network is directionally biased. For example, BC6 and BC7 drive broad radial glutamate spread from IPL S4/5 to S3 sublayers (Figures 4G and 4H), whereas BC5t and BC5o fail to propagate signals from IPL S3 to S4/5 (Figures 4G and 4H). This asymmetry likely arises from differences in gap-junctional connectivity, membrane impedance, and cellular excitability among CBC types. Because BC6/BC7 and AII amacrines are extensively coupled ⁷ and have lower membrane impedance than BC5t (Figure S4L), gap-junctional current from BC6 to BC5t via AII would produce greater depolarization in BC5t than the reverse current would in BC6, owing to impedance mismatch (Figure S5F, also see ref. ²²). Similarly, gap-junctional current from BC6 to AII is expected to excite BC5 subtypes more effectively than neighboring BC6 cells, generating broader lateral glutamate clouds at BC5 stratification levels (S3/4) than at BC6 terminal level (S4/5). During light stimulation, the strong photoreceptor input and large depolarizations of BC6 further bias radial signal flow from IPL S4/5 toward S3. Together, these properties endow the CBC network with hierarchical, rectified signal propagation and enhanced computational capability.

Integrative parallel processing by CBC channels

Parallel visual processing is a fundamental organizational principle in the CNS, enabling simultaneous encoding of distinct visual features across anatomically segregated channels. Our findings suggest, however, that this divide-and-conquer strategy becomes suboptimal for weak signals. Dividing already small inputs across independent channels reduces the likelihood of adequate downstream activation. The presence of driver CBCs and their coupling with other CBC types mitigates this limitation by redistributing signals across parallel channels to enhance network sensitivity.

Our findings provide a mechanistic explanation for prior observation that small stimuli evoke highly correlated responses across CBC types ⁹, indicating that such correlations largely arise from electrical coupling rather than intrinsic similarities among channels. As stimulus contrast and size increase, the influence of CBC coupling likely diverges among CBC types through integrations with amacrine cell circuits, contributing to functional diversity. Recent work shows that amacrine input can impose highly localized functional specialization at individual CBC synapses ^53,54. Together, electrical coupling spreads voltage signals across cells, whereas amacrine-mediated chemical interactions refine transmitter release at local terminal domains—two complementary mechanisms operating across different spatial scales.

Collectively, our results show that the CBC electrical network amplifies small-spot responses and shapes contrast sensitivity, luminance representation, and spatiotemporal dynamics. Acting as an integrative nexus, this network dynamically rebalances visual signals before their transmission to third-order neurons, enabling parallel yet interdependent processing and improving efficiency through shared functional specialization. Future studies should define the detailed functional circuitry of this network and examine its regulation under different lighting conditions.

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANTS DETAILS

Animals

All animal procedures were approved by Yale University Institutional Animal Care and Use Committee. Male and female mice, aged 6-16 weeks, were used for experiments. TYW3 ²⁶, mGluR6-Cre ²⁸, Cx36-floxed (Cx36^fl) ²⁹, Kcng4-Cre x YFP^{fl 45} mouse lines were kindly provided by In-Jung Kim (Yale University), Robert Duvoisin (Oregon Health Sciences University), Gautam Awatramani (University of Victoria), and Xin Duan (University of California-San Francisco), respectively. ChAT-cre mouse line was obtained from the Jackson Laboratory (ChAT^tm2(cre)Lowl, catalog #: 018957). All strains had been back-crossed to mice with C57BL/6J background for at least 5 generations. We cross-bred these lines to generate mGluR6-Cre x Cx36^fl/fl and mGluR6-Cre x Cx36^fl/fl x TYW3⁺ mouse stains (both referred to as mGluR6-Cre x Cx36^fl/fl). We used mGluR6⁻/Cx36^+/+, mGluR6⁺/Cx36fl^+/+, mGluR6⁻/Cx36fl^fl/+ and mGluR6⁻/Cx36fl^fl/fl littermates as controls for mGluR6-Cre x Cx36^fl/fl, with no significant difference observed among these controls.

Postmortem human retinal tissues

Postmortem human eyes were obtained through Yale Legacy Tissue Donation Program from deidentified donors (n=3, aged between 69 and 82). The acquisition and use of postmortem human tissue samples were approved by Yale University Human Research Protection Program’s Institutional Review Board. We complied with all relevant ethical regulations for work with human participants. All postmortem human tissue samples were obtained with informed consent prior to tissue collection from participants if enrolled antemortem or legal guardians if postmortem. Donor eyes were received within 2-4 postmortem. The retinas were then immediately dissected and stored for up to 12 h in ACSF (see below for composition), saturated with 95% O₂–5% CO₂ at room temperature. Dual patch clamp recordings were made from CBC-RGC pairs in a flat-mount retinal preparation under 2P imaging and constantly perfused with ACSF (saturated with 95% O₂–5% CO₂) at 35°C.

METHOD DETAILS

Experiments in ex vivo flat-mount retinas

Tissue preparation

For electrophysiological recording and two-photon imaging, mice were dark adapted for 30-60 min before euthanasia with CO₂ and decapitation. Retinas were dissected under dim red light in ACSF containing (in mM): 120 NaCl, 3.1 KCl, 1.1 CaCl₂, 0.5 KH₂PO₄, 1.24 MgSO₄, 26 NaHCO₃, 0.5 L-glutamine, 0.1 ascorbic acid, 0.1 Na-pyruvate, and 20 glucose (saturated with 95% O₂–5% CO₂, pH ~7.4). Ventral retinas were used for most electrophysiological recordings. All chemicals used in ACSF were purchased from Sigma-Aldrich.

Intravitreal viral transfection in mice

Approximately 1.5 μl viral construct (AAV2/2.CAG.SF-iGluSnFR.A184S, prepared by Yale Vision Core) was injected intravitreally in one eye of 4-8-week-old mice under anesthesia. Injections were performed with a glass pipette connected to a micro syringe. Glutamate (iGluSnFR) imaging was conducted 4-8 weeks post injection (Figure 4).

Electrophysiological recording and two-photon imaging

Patch clamp recordings were performed in flat-mount mouse and human retinas as previously described ^55-58 in ACSF saturated with 95% O₂ - 5% CO₂ at 32-35°C. Pipette solutions used were: (1) on-cell loose-patch solution: ACSF; (2) voltage-clamp solution for bipolar cells (in mM): 105 CsMeSO₃, 10 TEA-Cl, 2 EGTA, 5 Na₂-phosphocreatine, 3 ATP-Mg, 0.5 GTP-Na₂, 2 ascorbic acid and 10 HEPES, pH 7.2, with 20-30 CsOH; (3) voltage-clamp solution for ganglion and amacrine cells (in mM): 105 CsMeSO₃, 0.5 CaCl₂, 10 HEPES, 5 EGTA, 5 Na₂-phosphocreatine, 2 ATP-Mg, 0.5 GTP-Na₂, 2 ascorbic acid, 8 QX314-Cl, pH 7.2, with 20-30 CsOH; and (4) current-clamp solution (in mM): 105 potassium gluconate or KMeSO₄, 5 KCl, 0.5 CaCl₂, 2 MgCl₂, 2 EGTA, 10 HEPES, 10 Na₂-phosphocreatine, 2 ATP-Na₂, 0.5 GTP-Na₂, 2 ascorbic acid, pH 7.2 with 20-30 KOH. To chelate intracellular calcium in bipolar cells (Figure 2H-J, Figure S4N), solution (2) was modified by including 10 mM BATPA-Cs₄ and reducing CsMeSO₃ to 80 mM. The calcium buffer concentrations used in these experiments, which are expected to have an important impact on synaptic release under whole-cell patch clamp, were based on previous studies of retinal bipolar cells ^59,60.

Cells in the flat-mount retina were viewed with IR light (>800 nm wavelength, Wratten 2 filter 87C, Kodak) through IR DIC optics under an up-right microscope (BX51WI, Olympus) with a 60x, NA/1.0, water-immersion objective (LUMPlanFL/IR, Olympus) and a CCD camera (CoolSNAP HQ, Teledyne Photometrics). The microscope was configured with a two-photon imaging system (Ultima, Bruker) and a Ti:Sapphire pulsed laser (MaiTai, Newport Corp.) tuned to 870 nm wavelength. The fluorescence signals were bandpass-filtered at 520 ± 18 nm (FF01-520/35–25; Semrock) and 603 ± 27 nm (HQ607/45; Chroma) for green and red channels, respectively. Data acquisition was controlled by Prairie View 7.0 software (Bruker).

Patch-clamp recording used a dual-channel amplifier (MultiClamp 700B, Molecular Devices) and a data interface board (Digidata1440a, Molecular Devices). Patch pipettes had a resistance of 6-8 MΩ. Access resistance was < 40 MΩ and not compensated, given the relatively high input resistance of BCs (0.25–2 GΩ) ^41,61. Capacitance was compensated using MultiClamp 700B software. Current and voltage data were filtered at 2 kHz (eight-pole Bessel, MultiClamp 700B) and stored in a data acquisition system (Powerlab 8/30, AD Instrument) at 4 KHz. Liquid junction potentials were calculated and subtracted. Electrophysiological data were analyzed with MATLAB 2018a (MathWorks) and LabChart 7 (AD Instrument).

BCs were targeted for recording based on soma size and position in the outer INL. CBC types were randomly sampled. W3 cells were identified under two-photon imaging by their brightly YFP-labeled somas in the GCL of TYW3 or mGluR6-Cre x Cx36^fl/fl x TYW3⁺ mice. ON-S alpha ganglion cells were targeted based on their fluorescence labeling in the Kcng4-Cre x YFP^fl mouse line or their large soma size in the GCL in other mice. Displaced (ON) SACs were targeted based on their fluorescence labeling in the ChAT-Cre line or on their soma size, shape, and position in the GCL in other mice. All recorded cells were dye-filled (Alexa488 or Alexa594 hydrazide, Thermo Fisher Scientific) during recording and reconstructed in 3D from z-stacks of two-photon optical sections (imaged at 0.5- or 1-μm depth intervals at the end of recording) using ImageJ software and MATLAB 2018a. Axonal stratification profiles of recorded BC types (Figure 1D) were determined from the mean fluorescence intensity of Alexa594-filled axon arbors in each optical section (imaged by the red channel) as a function of IPL depth (% of IPL thickness) based on the strong iGluSnFR expression in the IPL (imaged by the green channel) of iGluSnFR-expressing retina (Figure S2).

Intrinsic currents in CBCs were measured approximately 5 minutes after breaking-in (Figure S1C), by which point most of the voltage-gated outward currents had been suppressed by intracellular Cs⁺ dialysis to a low and stable level. Light-evoked inhibitory currents were typically recorded ≥5 minutes after break-in and remained stable in both ON and OFF CBCs, showing minimal rundown for >30 minutes (Figure S1A). Light-evoked excitatory currents in OFF CBCs remained largely stable, whereas those in ON CBCs frequently exhibited significant rundown (~50% in 8 min of break-in) as also reported for RBCs ⁶². To minimize this effect, light-evoked excitatory currents and voltage responses in ON CBCs were typically measured within 3 minutes of establishing the whole-cell configuration (Figures S1A), and response amplitudes were quantitatively compared under same recording conditions immediately after break-in (Figures 5 and 6).

Voltage pulse- or light-evoked glutamate release (iGluSnFR signals) from BCs was recorded with two-photon imaging in time series, synchronized with simultaneous patch clamp or light stimulation. Data were analyzed with Prairie View 7.0 software and MATLAB 2022a.

Spatiotemporal analysis of glutamate signaling in IPL

3D pseudo-movies of voltage pulse-evoked iGluSnFR signals (Supplementary videos) were generated from vertical two-photon image stacks of 2D time series acquired at consecutive IPL depths in response to identical voltage stimulations of the patch-clamped BC. The MATLAB ‘imread’ function was used to read and store the images in 4D arrays (x-position, y-position, time, green/red channel).

The lateral spread (cloud size) of iGluSnFR signals evoked by single BC activation (Figure 4C and 4D, Figure S5B and S5C) was measured at the IPL depth where most axon boutons of the recorded BC were located. A 100μm x 100μm field of view (FOV) containing 128 x 128 pixels (pixel size: 0.78 μm x 0.78 μm) was imaged for 1.5 s at ~10 Hz. Total number of pixels with a mean fluorescence intensity 3x larger than background standard deviation (0.5 s before voltage pulse) was multiplied by one pixel area to determine cloud size. The same method was applied to light-evoked iGluSnFR signals. In sublamina b (IPL-b), the response signal was averaged over the 3s of light stimulation, while in sublamina a (IPL-a), it was averaged over the 3s after light offset. A significant light response was defined as 1.5 times the background (0.5 s before light onset/offset) standard deviation. We note that the cloud size measured by pixel counts is different (typically smaller) than the total area within the cloud boundary.

The spatiotemporal dynamics of single-BC-activated glutamate clouds (Figures 4E and 4F) was characterized by imaging a field of view (FOV) of 38 μm x 4 μm (pixel size: 0.28 μm x 0.28 μm) at a frame rate of ~100 Hz. This FOV was selected to capture an Alexa594-filled bouton near the edge of the recorded BC’s axon arbor and an extended area beyond the arbor. The temporal dynamics of iGluSnFR response was compared across various ROIs in the FOV (Figure 4F). The response rising phase (time between 10% and 25% of peak response) was back-extrapolated to determine the zero-crossing timepoint, defined as the onset time of the response. Similarly, the spatiotemporal dynamics of light-evoked iGluSnFR signals (Figures S5D and S5E) were analyzed from a FOV of 120 μm x 6 μm (pixel size, 0.76 μm x 0.76 μm) at a frame rate of ~100 Hz. This FOV included both the region directly activated by the 25-μm light spot and the surrounding areas.

To measure radial (vertical) spread of iGluSnFR signals (Figures 4G and 4H), a region of interest (ROI) covering all axon boutons of the recorded BC was chosen at the IPL depth where most boutons were present (the main stratification level). iGluSnFR responses (ΔF/F) to repeated stimulations of an identical voltage pulse were measured from this ROI as a function of IPL depths (in 2.5-μm or 5-μm depth increments) and fitted with a Gaussian distribution. The radial spread (standard deviation of the distribution, σ) was normalized to IPL thickness (%) (Figure 4H).

Visual stimulation of ex vivo retina

Visual stimuli for patch-clamp experiments were generated using a miniature transmissive TFT LCD (Model XGAP01, CRL OPTO, 36.9 mm x 27.6 mm, 1024 x 768 pixels) with Vision Works software (Vision Research Graphics, Inc.). The LCD image was trans-illuminated by collimated light from a halogen lamp and projected to the photoreceptor outer segment layer through the microscope condenser lens. The maximum intensity of the LCD image at the retina was 7.8x10⁻¹³ W/μm² (measured with a radiometer, Model S470, UDT Instruments).

Stimuli included stationary light spots with diameters ranging from 25 to 1000 μm. The following intensity and contrast levels at the retina were used: (1) Flash: a light spot with an intensity of 5.5×10⁵ Rh* rod⁻¹ s⁻¹ against a uniform background light intensity of 8×10⁴ Rh* rod⁻¹ s⁻¹, resulting a 600% contrast; (2) Variable intensity light spots ranging from 1.7 to 5.5×10⁵ Rh* rod⁻¹ s⁻¹ over a uniform background of 1.5×10⁵ Rh* rod⁻¹ s⁻¹, with contrasts ranging from 16%-200%.

Visual stimuli for eliciting iGluSnFR responses in the IPL (Figure 4I and 4J) were generated using a DLP projector (DLPCR4500EVN, contrast >1000%, EKB Technologies) and projected onto the photoreceptor outer segment layer of flat-mounted retina through the microscope condenser lens. To avoid contamination of iGluSnFR signals by stimulus light, collimated UV light (bandpass filtered at 380-400 nm) from a UV LED (A00X-UV4XX-XX, LED Supply) was used to illuminate the DLP projector via a light guide (LLG05-4H, Thorlabs). The light stimulus intensity was 2.3 ×10⁵ Rh* UV-cone⁻¹ s⁻¹ at the photoreceptor outer segment layer.

Pharmacology

The following drugs were used (in μM): CNQX (40) (Enzo Life Sciences), CPP (40) (Sigma-Aldrich), MFA (50-100) (Sigma-Aldrich), 18betaGA (50) (Sigma-Aldrich), L-AP4 (20) (Enzo LifeSciences) and ACET (10) (Tocris Bioscience or Hello Bio). All drugs were applied via bath perfusion. For experiments involving preincubation of MFA or 18betaGA (Figure 2), a 25-minute incubation period was allowed.

Experiments in in vivo mouse dLGN

dLGN Virus injections

To label neurons in the dLGN, mice were anesthetized with isoflurane in 100% O₂ (induction, 3%–5%; maintenance, 1%–2%). Buprenorphine extended-release injectable suspension (3.25 mg/kg; Fidelis Animal Health) was administered. Subcutaneous infiltration with lidocaine (less than 7 mg/kg) was performed five minutes before making the first skin incision. A total of 120 nL of AAV.syn.jGCaMP8s.WPRE (Addgene #162374-AAVrg) was stereotaxically injected into the left dLGN at two locations (2.25 mm lateral, 2.3 mm posterior to Bregma, at a depth of 2.6 mm from the skull surface; and 2.33 mm lateral, 2.7 mm posterior to Bregma, at a depth of 2.8 mm from the skull surface). This virus was used to label dLGN neurons locally. Finally, meloxicam (0.5 mg/kg) was administered, and the mouse was allowed to recover.

Headpost and cranial window implant for dLGN imaging

A headpost and a cranial window were implanted 1-3 weeks after the virus injection. A two-pronged headpost was attached to the skull, centered approximately 2.7 mm lateral and 2.1 mm posterior to Bregma over the left hemisphere and oriented tangentially to the curved skull surface. A 3-mm diameter craniotomy was performed at the center of the headpost. A 3 mm x 3.4 mm (diameter x height) stainless steel cylindrical cannula (MicroGroup) with a 3-mm diameter coverslip glued to the bottom by UV-cured Norland Optical Adhesive 71 was stereotaxically inserted into the craniotomy and lowered approximately 2.75 mm below the skull’s surface to slightly press against the thalamus surface. The cannula was affixed to the skull with C&B Metabond (Parkell). To facilitate imaging with a water-immersion objective and light shielding, a low-profile adaptor was created by gluing a neodymium ring magnet (Indigo^® Instruments, outer diameter, inner diameter, height: 7.5 mm, 5 mm, 1 mm) to the skull around the cannula. For additional details, see ref.⁶³.

dLGN two-photon calcium imaging

Two-photon calcium imaging was conducted on awake, head-restrained mice after they were habituated to head restraint while on a running wheel. Imaging of the dLGN was performed using a resonant-scanning two-photon microscope (Neurolabware), with a 20x, 1.0 NA, 5.6 mm WD objective (Zeiss) at 4 x (~154 x 208 μm²) digital zoom. Light shielding was positioned around the objective to prevent the monitor light from reaching the sample. Fields of view (FOVs) were imaged at depths ranging from 80 to 150 μm below the surface of the optic tract (roughly corresponding to the upper 20-90 μm of the dLGN shell). Calcium imaging of dLGN neurons was achieved at 960 nm using a tunable laser (80 MHz; Insight X3, Spectra-Physics). Emitted signals were detected by a GaAsP PMT after a 562nm long-pass dichroic mirror, a 510/84 band-pass filter. The laser power measured at the objective’s front aperture through the cannula ranged from 11 to 40 mW. Images were collected at a rate of 15.6 frames/s, 796 × 512 pixels/frame, using the Scanbox software (Neurolabware). Each imaging run lasted 15-30 minutes, and an imaging session consisted of 6 imaging runs. The same FOV was imaged throughout a session. When acquiring data from different FOVs from the same mouse, the FOVs were at least 25 μm apart along the dLGN depth.

dLGN visual stimulation

Visual stimuli for dLGN neuron imaging were generated using Psychtoolbox-3 ⁶⁴, and displayed on a luminance-calibrated LCD monitor (Dell, 19”,1280 x 1024 pixels, 60 Hz refresh rate) placed 23 cm from the mouse’s right eye and spanning 80° × 70° of visual space (azimuth: 3° – 83°; elevation: −10° – 60°). We estimated the mean luminance of the LCD to be 24.5 lux on the screen and 16.4 lux at the mouse eye. The brightness of all stimuli was gamma-corrected.

To map the retinotopy of the imaging field, we utilized a binarized version of a bandpass-filtered noise stimulus at 80% contrast. This noise stimulus had a spatial frequency corner at 0.05 cycles per degree (cpd), a cutoff of 0.32 cpd, and a temporal frequency cutoff of 4 Hz ⁶⁵. The visual stimulus was presented within 5° x 40° bars, arranged vertically at one of 8 azimuths and horizontally at one of 8 elevations. The stimuli were presented for 2 seconds in each trial, with a 2-second inter-stimulus interval at the mean luminance level. Visual stimulation also included a blank condition (mean luminance). The order of the stimuli was randomized within a single repeat, which consisted of a single presentation of each stimulus condition. A total of 20 repeats were delivered during one imaging session.

To evaluate the size tuning of each neuron, we used flashing spot stimulation. A white disk, with ~1000% Weber contrast (~ 6*10⁴ isomerization per M-cone per second) and a diameter of 0.5°, 1°, 2°, or 8° in visual angle ⁶⁶, was flashed on a black background (~ 5*10³ isomerization per M-cone per second) ⁶⁷. All spots were co-centered and interleaved randomly. A spot was presented for 3 seconds per trial, and the inter-trial interval was 3 seconds. A total of 100 repeats were delivered during one imaging session.

To evaluate the contrast tuning of each neuron, a white disk with a diameter of 2° in visual angle, at 50%, 100%, 150%, or 200% Weber contrast, was flashed on a grey background (~10⁴ isomerization per M-cone per second). All spots were co-centered and interleaved randomly. The spot was presented for 3 seconds per trial, and the inter-trial interval was 3 seconds. A total of 100 repeats were delivered during one imaging session.

dLGN image processing

Image preprocessing

To correct for x-y motion along the imaged plane, a series of image registration and data-cleaning steps were applied. Initially, the movies captured on each imaging day were registered to a common average field-of-view using efficient subpixel registration methods⁶⁸. Subsequently, denoising was performed using principal component analysis (PCA) on the concatenated movies across the entire imaging session. We used only the first 400 principal components (with the highest eigenvalues) out of a total of approximately 30,000 total to reconstruct the registered and downsampled movie while removing shot noise⁶⁹. The denoised images were then subjected to local normalization and image warping to derive the warping parameters, which were subsequently applied to the original raw movie. As a final step, PCA denoising was performed for the second time. Notably, the observed results were not dependent on this operation. For additional details, please see Liang et al ⁶³.

Cell mask identification

The automatic extraction of masks for dLGN cell bodies was performed using a custom implementation of an NMF-based neuron segmentation procedure ⁷⁰. The time scale of the deconvolution kernel is 0.7 s. The sampling rate is 15.6 Hz. Thresholding scaling was set at 0.7. All cell masks were manually verified and adjusted as necessary to ensure accuracy and reliability. Neuropil masks took the form of circular annuli with a width of 9 μm, positioned with their inner edge 4 μm beyond the outermost edge of the corresponding cell body mask.

Time course extraction and correction

To obtain raw fluorescence traces for cell body and neuropil masks, the fluorescence intensity value of a mask at each time point was defined as the average fluorescence across the pixels belonging to the mask.

To account for neuropil signals which might contaminate signals in the cell body trace, neuropil correction was applied by subtracting a scaled version of the corresponding neuropil trace from each cell body trace, with the scale factor individually derived from linear regression between the minimal bound of cell body trace and the corresponding neuropil trace ⁷¹.

To assess ON responses, the fractional change in fluorescence, ΔF/F₀(t), was calculated following each onset of the white spot presentation. The baseline F₀ was calculated as the average amplitude of the fluorescence 400 milliseconds before the stimulus onset. A single-trial ON response value was obtained by averaging the ΔF/F₀(t) response during the first 400-millisecond window after the stimulus onset. The ON response under a given stimulation condition was calculated as the average of single-trial ON response values across 100 repeats.

To assess OFF responses, the ΔF/F₀(t) was calculated following each offset of the white spot presentation. The baseline F₀ was calculated as the average amplitude of the fluorescence 400 milliseconds before the stimulus offset. Single-trial OFF response values were obtained by averaging the ΔF/F₀(t) response during the first 400-millisecond window after the stimulus offset. The OFF response under a given stimulation condition was calculated as the average of single-trial OFF response values across 100 repeats.

Estimation of neurons with significant visual responses

We determined a neuron to be significantly responsive to flash spot stimuli by requiring the amplitude of ΔF/F₀(t) during the response window to be significantly different from the amplitude of ΔF/F₀(t) 400 milliseconds before the stimulation onset (using the ‘Ranksum’ function in MATLAB) and the quality index (QI) to be larger than 0.15. The quality index was calculated from the variance of the A (the amplitude of ΔF/F₀(t)) across time, var(·)_t, and the mean across stimulus repeats, 〈·〉_r, as follows⁷²:

$$QI=\frac{var{({\langle A\rangle }_r)}_t}{{\langle var{(A)}_t\rangle }_r}$$

Linear mixed-effects model

To address the interdependence among dLGN neurons from the same animal, we examined the significance between various variables using linear mixed-effects models ('fitlme' function in MATLAB) ⁷³. In each testing instance, we considered the animal (MouseNum) as a random effect and the mouse genotype (GroupNum; WT as 0 and mGluR6-Cre x Cx36^fl/fl as 1) as a fixed effect. Meanwhile, we used (GroupNum|MouseNum) to nest the genotype under the mouse ID. Finally, we wrote the function as ‘Data~GroupNum+(1∣MouseNum)+(GroupNum∣MouseNum)’, where Data represents the neuronal activity. This model has the following mathematical form:

$${\mathrm{y}}_{\text{ij}}={\beta }_0+{\beta }_1{\mathrm{x}}_{\mathrm{i}}+{\alpha }_{\mathrm{j}}+{\alpha }_{0j}+{\alpha }_{1\mathrm{j}}{\mathrm{x}}_{\text{ij}}+{\epsilon }_{\text{ij}}$$

In this function, ${\mathrm{y}}_{\text{ij}}$ is the ${\mathrm{i}}^{\text{th}}$ observation for the ${\mathrm{j}}^{\text{th}}$ mouse; ${\beta }_0$ is the fixed intercept; ${\beta }_1$ is the fixed effect coefficient for ${\mathrm{x}}_{\mathrm{i}}$; ${\mathrm{x}}_{\mathrm{i}}$ is the experimental group, either WT or mGluR6-Cre x Cx36^fl/fl for the ${\mathrm{i}}^{\text{th}}$ observation; ${\alpha }_{\mathrm{j}}$ is the random effect for the ${\mathrm{j}}^{\text{th}}$ mouse, with the distribution ${\alpha }_{\mathrm{j}}\sim \mathrm{N}(0,{{\sigma }_1}^2);{\alpha }_{0j}$ are random effects, with the distribution ${\alpha }_{0\mathrm{j}}\sim \mathrm{N}(0,{{\sigma }_2}^2);{\alpha }_{1\mathrm{j}}$ are random effects, with the distribution ${\alpha }_{1\mathrm{j}}\sim \mathrm{N}(0,{{\sigma }_3}^2);{\mathrm{x}}_{\text{ij}}$ is the experimental group for the ${\mathrm{i}}^{\text{th}}$ observation of the ${\mathrm{j}}^{\text{th}}$ mouse; and ${\epsilon }_{\text{ij}}$ represents the random error, with the distribution ${\epsilon }_{\text{ij}}\sim \mathrm{N}(0,{\sigma }^2)$.

QUANTIFICATION AND STATISTICAL ANALYSIS

Unpaired or paired Mann-Whitney U-test in MATLAB (functions Ranksum or Signrank) was used to compare two independent groups or related groups, respectively. Dunnett’s multiple comparison test following a Kruskal-Wallis test was used to compare three independent groups (WT, mGluR6-Cre x Cx36^fl/fl and MFA/18betaGA). Linear correlation (Figure S5D) was analyzed with functions Fitlm in MATLAB. Linear-mixed models was used to protect against grouped effect during mouse dLGN imaging (Figure 7). All Data were plotted as mean ± SEM, with individual data points displayed and p-value level indicated. Specifically, *p < 0.05; **p < 0.01; ***p < 0.001. Additional details on sample sizes, statistical test, significant levels for each experiment can be found in figures, figure legends, Results and Table S1. All acquired data was used for analysis.

Supplementary Material

Video S1

Videos S1. Pseudo-movies of the radial extent of glutamate release evoked by BC1, Related to Figure 4

Movies showing the cross-sectional view of radial extent of glutamate release evoked by BC1. iGluSnFR signals (green, ΔF/F₀) were detected only around BC1 axon terminals (red). The scale bar is 25 μm. Brightness and contrast were adjusted for optimal visualization. For more details, see STAR Methods.

Video S2

Videos S2. Pseudo-movies of the radial extent of glutamate release evoked by BC2, Related to Figure 4

Similar to Video S1, but for BC2. iGluSnFR signals were detected radially (vertically) beyond the layers of BC2 axon terminals.

Video S3

Videos S3. Pseudo-movies of the radial extent of glutamate release evoked by BC5t, Related to Figure 4

Similar to Video S1, but for BC5t. iGluSnFR signals were detected only around BC5t axon terminals.

Video S4

Videos S4. Pseudo-movies of the radial extent of glutamate release evoked by BC6, Related to Figure 4

Similar to Video S1, but for BC6. iGluSnFR signals were detected radially (vertically) beyond the layers of BC6 axon terminals.

Spreadsheet Table S1

Table S1. Statistical data used in various figures, Related to all figures.

Document S1. Figure S1-S7

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
AAV2/2.CAG.SF-iGluSnFR.A184S	This paper	N/A
AAVrg.syn.jGCaMP8s.WPRE	Zhang et al Janelia Research Campus, Online Resource (2020)doi:10.25378/janelia.13148243.	Addgene, Cat #: 162374-AAVrg
Biological samples
Human eye globe	Yale Legacy Tissue Donation Program	https://medicine.yale.edu/pathology/clinical/legacy-tissue-donation/
Chemicals, peptides, and recombinant proteins
Sodium chloride (NaCl)	Millipore Sigma	Cat #: S7653; CAS: 7647-14-5
Potassium chloride (KCl)	Millipore Sigma	Cat #: P9333; CAS: 7447-40-7
Calcium chloride (CaCl2)	Tocris Bioscience	Cat #: 3148; CAS: 10035-04-8
Potassium phosphate monobasic (KH2PO4)	Millipore Sigma	Cat #: P5655; CAS: 7778-77-0
Magnesium sulfate (MgSO4)	Millipore Sigma	Cat #: M7506; CAS: 7487-88-9
Sodium bicarbonate (NaHCO3)	Millipore Sigma	Cat #: S5761; CAS: 144-55-8
L-Glutamine	Millipore Sigma	Cat #: G8540; CAS: 56-85-9
L-Ascorbic acid	Millipore Sigma	Cat #: A5960; CAS: 50-81-7
Sodium pyruvate (Na-pyruvate)	Millipore Sigma	Cat #: P2256; CAS: 113-24-6
D-Glucose	Millipore Sigma	Cat#: G5767; CAS: 50-99-7
Cesium methanesulfonate (CsMeSO3)	Millipore Sigma	Cat#: C1426; CAS: 2550-61-0
Tetraethylammonium chloride (TEA-Cl)	Millipore Sigma	Cat#: T2265; CAS: 56-34-8
EGTA	Thermo Fisher	Cat#: 409911000; CAS: 67-42-5
Phosphocreatine disodium salt hydrate (Na2-phosphocreatine)	Millipore Sigma	Cat#: P7936; CAS: 19333-65-4
Adenosine 5’-triphosphate magnesium salt (ATP-Mg)	Millipore Sigma	Cat#: A9187; CAS: 74804-12-9
Guanosine 5’-triphophate disodium salt hydrate (GTP-Na2)	Thermo Fisher	Cat#: 226250010; CAS: 56001-37-7
HEPES	Millipore Sigma	Cat#: H3375; CAS: 7365-45-9
Cesium hydroxide (CsOH)	Millipore Sigma	Cat#: 516988; CAS: 35103-79-8
Lidocaine N-ethyl chloride (QX314-Cl)	Millipore Sigma	Cat#: L1663; CAS: 5369-03-9
Potassium gluconate	Millipore Sigma	Cat#: G4500; CAS: 299-27-4
Potassium methylsulfate (KMeSO4)	ICN Biomedicals	Cat#: 215481; CAS: 562-54-9
BAPTA, tetracesium salt	Thermo Fisher	Cat#: B1212
Potassium hydroxide (KOH)	J.T.Baker	Cat#: 3140-01; CAS: 1310-58-3
Alexa594 hydrazide, sodium salt	Thermo Fisher	Cat#: A10438
Alexa488 hydrazide, sodium salt	Thermo Fisher	Cat#: A10436
CNQX	Enzo Life Sciences	Cat#: ALX-550-042; CAS: 115066-14-3
CPP	Millipore Sigma	Cat#: C104; CAS: 100828-16-8
18β-Glycyrrhetinic acid	Millipore Sigma	Cat#: G10105; CAS: 471-53-4
Meclofenamic acid disodium salt (MFA)	Millipore Sigma	Cat#: M453; CAS: 6385-02-0
L-AP4	Enzo Life Sciences	Cat#: ALX-550-026; CAS: 23052-81-5
ACET	Thermo Fisher	Cat#: 2728; CAS: 936095-50-0
ACET	Hello Bio	Cat#: HB0102; CAS: 936095-50-0
Experimental models: Organisms/strains
Mouse: C57BL/6J	The Jackson Laboratory	JAX: 00664
Mouse: C57BL/6J, W3	Dr. In-Jung Kim	N/A
Mouse: C57BL/6J, mGluR6-Cre	Dr. Robert Duvoisin	N/A
Mouse: C57BL/6J, Cx36fl	Dr. Gautam Awatramani	N/A
Mouse: C57BL/6J, KCNG4-Cre x YFPfl	Dr. Xin Duan	N/A
Mouse: C57BL/6J, ChATtm2(cre)Lowl	The Jackson Laboratory	JAX: 018957
Software and algorithms
ImageJ	NIH	https://imagej.nih.gov/ij/download.html
MATLAB2018a	MathWorks	https://www.mathworks.com/products/matlab.html
MATLAB2022a	MathWorks	https://www.mathworks.com/products/matlab.html
Prairie View 7.0	Bruker	https://www.brukersupport.com/
pClamp10	Molecular Devices	https://www.moleculardevices.com
LabChart7	ADInstruments	https://www.adinstruments.com/support/software