Vesicular Release of GABA by Mammalian Horizontal Cells Mediates Inhibitory Output to Photoreceptors

Feedback inhibition by horizontal cells regulates rod and cone photoreceptor calcium channels that control their release of the neurotransmitter glutamate. This inhibition contributes to synaptic gain control and the formation of the center-surround antagonistic receptive fields passed on to all downstream neurons, which is important for contrast sensitivity and color opponency in vision. In contrast to the plasmalemmal GABA transporter found in non-mammalian horizontal cells, there is evidence that the mechanism by which mammalian horizontal cells inhibit photoreceptors involves the vesicular release of the inhibitory neurotransmitter GABA. Historically, inconsistent findings of GABA and its biosynthetic enzyme, L-glutamate decarboxylase (GAD) in horizontal cells, and the apparent lack of surround response block by GABAergic agents diminished support for GABA's role in feedback inhibition. However, the immunolocalization of the vesicular GABA transporter (VGAT) in the dendritic and axonal endings of horizontal cells that innervate photoreceptor terminals suggested GABA was released via vesicular exocytosis. To test the idea that GABA is released from vesicles, we localized GABA and GAD, multiple SNARE complex proteins, synaptic vesicle proteins, and Cav channels that mediate exocytosis to horizontal cell dendritic tips and axonal terminals. To address the perceived relative paucity of synaptic vesicles in horizontal cell endings, we used conical electron tomography on mouse and guinea pig retinas that revealed small, clear-core vesicles, along with a few clathrin-coated vesicles and endosomes in horizontal cell processes within photoreceptor terminals. Some small-diameter vesicles were adjacent to the plasma membrane and plasma membrane specializations. To assess vesicular release, a functional assay involving incubation of retinal slices in luminal VGAT-C antibodies demonstrated vesicles fused with the membrane in a depolarization- and calcium-dependent manner, and these labeled vesicles can fuse multiple times. Finally, targeted elimination of VGAT in horizontal cells resulted in a loss of tonic, autaptic GABA currents, and of inhibitory feedback modulation of the cone photoreceptor Cai, consistent with the elimination of GABA release from horizontal cell endings. These results in mammalian retina identify the central role of vesicular release of GABA from horizontal cells in the feedback inhibition of photoreceptors.

Feedback inhibition by horizontal cells regulates rod and cone photoreceptor calcium channels that control their release of the neurotransmitter glutamate. This inhibition contributes to synaptic gain control and the formation of the center-surround antagonistic receptive fields passed on to all downstream neurons, which is important for contrast sensitivity and color opponency in vision. In contrast to the plasmalemmal GABA transporter found in non-mammalian horizontal cells, there is evidence that the mechanism by which mammalian horizontal cells inhibit photoreceptors involves the vesicular release of the inhibitory neurotransmitter GABA. Historically, inconsistent findings of GABA and its biosynthetic enzyme, L-glutamate decarboxylase (GAD) in horizontal cells, and the apparent lack of surround response block by GABAergic agents diminished support for GABA's role in feedback inhibition. However, the immunolocalization of the vesicular GABA transporter (VGAT) in the dendritic and axonal endings of horizontal cells that innervate photoreceptor terminals suggested GABA was released via vesicular exocytosis. To test the idea that GABA is released from vesicles, we localized GABA and GAD, multiple SNARE complex proteins, synaptic vesicle proteins, and Ca v channels that mediate exocytosis to horizontal cell dendritic tips and axonal terminals. To address the perceived relative paucity of synaptic vesicles in horizontal cell endings, we used conical electron tomography on mouse and guinea pig retinas that revealed small, clear-core vesicles, along with a few clathrin-coated vesicles and endosomes in horizontal cell processes within photoreceptor terminals. Some small-diameter vesicles were adjacent to the plasma membrane and plasma membrane specializations. To assess vesicular release, a functional assay involving incubation of retinal slices in luminal VGAT-C antibodies demonstrated vesicles fused with the membrane in a depolarization-and calcium-dependent manner, and these labeled vesicles can fuse multiple times. Finally, targeted elimination of VGAT in horizontal cells resulted in a loss of tonic, autaptic GABA currents, and of inhibitory feedback modulation of the cone photoreceptor Ca i , consistent with the elimination of GABA

INTRODUCTION
Horizontal cells receive synaptic input from thousands of photoreceptors and feedback this broad spatial information back to photoreceptors as well as feeding it forward to bipolar cells to generate receptive field surrounds (Thoreson and Mangel, 2012). In 1970, Baylor et al. demonstrated that turtle retinal horizontal cells contribute a negative feedback signal to the cone photoreceptor light response. When our studies began over 30 years later, the proposed cellular mechanisms of horizontal cell neurotransmission were multiple, controversial, and unconventional: voltage-and sodium-dependent, calciumindependent plasmalemmal γ-aminobutyric acid (GABA) transporter (GAT) activity, as characterized in non-mammalian vertebrates (Schwartz, 2002), ephaptic coupling between photoreceptor calcium channel gating and current flow in horizontal cell glutamate receptors and hemichannels shown in fish retina (Byzov and Shura-Bura, 1986;Kamermans et al., 2001), and photoreceptor calcium current regulation by synaptic cleft pH (Hirasawa and Kaneko, 2003;Vessey et al., 2005;Cadetti and Thoreson, 2006;Kreitzer et al., 2012;Wang et al., 2014;Kramer and Davenport, 2015;Tchernookova et al., 2018;Grove et al., 2019). The apparent lack of the cone surround response block by GABAergic pharmacological agents in turtle, goldfish, mouse, and primate retinas (Thoreson and Burkhardt, 1990;Verweij et al., 1996Verweij et al., , 2003Endeman et al., 2012;Kemmler et al., 2014) was used to argue against a direct role for GABA in feedback inhibition. In contrast, early studies reported GABA in horizontal cells (Lam et al., 1978;Mosinger et al., 1986) suggesting that it may be a neurotransmitter used by horizontal cells. However, GABA immunoreactivity in horizontal cells was not consistently observed in adult mammalian retinas (Lam et al., 1978;Schnitzer and Rusoff, 1984;Mosinger et al., 1986;Chun and Wässle, 1989;Wässle and Chun, 1989) raising doubts about its role as a feedback transmitter. Further, unlike other GABAergic neurons, horizontal cells in adult mammalian retina did not take up 3 H-GABA or 3 H-muscimol (Blanks and Roffler-Tarlov, 1982;Wässle and Chun, 1989) and GATs were not expressed in these cells (Honda et al., 1995;Johnson et al., 1996;Casini et al., 2006;Guo et al., 2009Guo et al., , 2010. In contrast, horizontal cells in cat and monkey retinas showed GABA immunoreactivity (Agardh and Ehinger, 1982;Chun and Wässle, 1989;Pourcho and Owczarzak, 1989;Wässle and Chun, 1989;Grünert and Wässle, 1990). The GABA synthetic enzymes, glutamic acid decarboxylase65, and 67 (GAD65 and GAD67), were localized to mammalian horizontal cells Johnson and Vardi, 1998). Although like GABA, the GABA synthetic enzymes, GAD65 and GAD67 were observed in horizontal cells during development, and they were not consistently detected in adult horizontal cells (Brandon et al., 1979;Schnitzer and Rusoff, 1984;Mosinger and Yazulla, 1985). In contrast to the lack of GATs, the vesicular GABA transporter (VGAT/VIAAT, vesicular inhibitory amino acid transporter), which loads inhibitory amino acid transmitters into synaptic vesicles (McIntire et al., 1997;Sagné et al., 1997), was observed in amacrine and horizontal cells in multiple mammalian species Cueva et al., 2002;Jellali et al., 2002;Guo et al., 2010;Lee and Brecha, 2010;Hirano et al., 2011). The presence of VGAT in horizontal cell synaptic endings suggested that these unconventional neurons may release the neurotransmitter GABA via vesicular release. When VGAT was deleted from horizontal cells, these cells failed to feedback to photoreceptors (Hirano et al., 2016a) and the same mouse line revealed a lack of autaptic GABA reception by horizontal cells and no influence on cone calcium channels (Grove et al., 2019;Barnes et al., 2020), ending debate, at least in mammalian retinas, about whether horizontal cells utilize vesicular GABA release to send feedback to photoreceptors. Here we marshal evidence for the hypothesis that mammalian horizontal cells possess the cellular structures and proteins that mediate vesicular transmitter release. These include the presence and synthesis of GABA as a neurotransmitter, the essential molecular machinery for vesicular release, the structural basis of vesicular release, namely synaptic vesicles, and the regulated fusion and recycling of synaptic vesicles in mammalian horizontal cells. These findings show that the cellular mechanism underlying feedback inhibition in mammals involves vesicular GABA release by horizontal cells, and this stands to support a new GABA-pH hybrid model wherein autaptic reception of GABA by horizontal cells regulates pH in the synaptic cleft via depolarization and the bicarbonate permeability of the GABA receptors, resulting in the modulation of presynaptic calcium channels in photoreceptors (Grove et al., 2019;Barnes et al., 2020).

Presence of GABA in Horizontal Cells
Several convergent findings show that GABA is the mammalian horizontal cell transmitter. Mammalian retinas contain typically two morphological types of horizontal cells, an axonless A-type whose dendrites contact only cones and an axon-bearing Btype whose dendrites contact cones and the axonal terminal system, the rods. Some rodents, including mouse and rat, possess only the B-type (Peichl and González-Soriano, 1994). The lack of immunoreactivity for GABA and its synthetic enzymes GAD65 and GAD67 in adult horizonal cells in some studies was used to argue against a role for GABA in horizontal cell neurotransmission. However, many studies have shown evidence for GABA in horizontal cells of cat, rabbit, rat, mouse, guinea pig, and primate retina (Nishimura et al., 1985;Mosinger et al., 1986;Osborne et al., 1986;Agardh et al., 1987;Mosinger and Yazulla, 1987;Chun, 1988, 1989;Chun and Wässle, 1989;Pourcho and Owczarzak, 1989;Grünert and Wässle, 1990;Vardi and Auerbach, 1995;Kalloniatis et al., 1996;Johnson and Vardi, 1998;Koulen et al., 1998b;Guo et al., 2010;Deniz et al., 2011;Herrmann et al., 2011), albeit at lower levels than in amacrine cells (Pourcho and Owczarzak, 1989;Wässle and Chun, 1989;Johnson and Vardi, 1998;Marc et al., 1998). In cat and monkey, horizontal cells in peripheral retina lacked GABA immunoreactivity, whereas they were immunoreactive in central retina Grünert and Wässle, 1990). Unlike non-mammalian horizontal cells in which not all subtypes contained GABA (Marc, 1992;Schwartz, 2002;Yang, 2004) both mammalian subtypes appeared to show GABA immunoreactivity Grünert and Wässle, 1990;Johnson and Vardi, 1998;Guo et al., 2010). In mouse and rabbit, horizontal cells exhibited high levels of GABA during early retinal development, which then dropped with maturation (Schnitzer and Rusoff, 1984;Osborne et al., 1986;Messersmith and Redburn, 1993;. An example of the GABA immunolabeling is shown in horizontal cells of the adult guinea pig retina, which contains both A-and B-types (Figure 1, Guo et al., 2010) similar to cat and macaque retinas (Pourcho and Owczarzak, 1989;Wässle and Chun, 1989;Grünert and Wässle, 1990). GABA immunoreactivity, like the punctate staining of neurotransmitter receptors in retina Greferath et al., 1995) was highly sensitive to fixation conditions, favoring weak fixation (e.g., shorter fixation times, lower aldehyde concentrations) for visualization Deniz et al., 2011). This lability as well as antibody specificity differences may account for reports of little to no immunostaining observed in well-fixed tissue (Agardh et al., 1986;Osborne et al., 1986;Versaux-Botteri et al., 1989;Messersmith and Redburn, 1992;Yamasaki et al., 1999;Loeliger and Rees, 2005).

Glutamic Acid Decarboxylase (GAD)
The GABA-synthesizing enzyme L-glutamate decarboxylase (GAD) exists as two principal isoforms, GAD65 and GAD67 (Erlander et al., 1991;Kaufman et al., 1991). One or both of the GAD isoforms are found in mammalian horizontal cells at both the mRNA (Sarthy and Fu, 1989;Guo et al., 2010;Deniz et al., 2011) and protein levels (Schnitzer and Rusoff, 1984;Vardi and Auerbach, 1995;Johnson and Vardi, 1998;Yamasaki et al., 1999;Dkhissi et al., 2001;Guo et al., 2010;Deniz et al., 2011). In rabbit retina, GAD65 and GAD67 immunoreactivities were detected in horizontal cells (Johnson and Vardi, 1998). Several studies report GAD67 immunostaining is present at high levels in horizontal cells of the developing and juvenile mouse, rat, and rabbit retina (Schnitzer and Rusoff, 1984;Osborne et al., 1986;Versaux-Botteri et al., 1989;Schubert et al., 2010), but at low or non-detectable levels in adult horizontal cells (Brandon et al., 1979;Schnitzer and Rusoff, 1984;Brandon, 1985;Osborne et al., 1986;Wässle and Chun, 1989;Brecha et al., 1991;Yazulla et al., 1997;Koulen et al., 1998b), including mouse Schubert et al., 2010;Herrmann et al., 2011). GAD65 immunostaining (Figure 2) and mRNA were detected in adult guinea pig horizontal cells . Note the concentration of GAD65 immunoreactivity in the horizontal cell endings (Figure 2, arrows) and the scleral portion of the cell body. In rabbit horizontal cells, there are different subcellular localizations of GAD65 and GAD67 protein (Johnson and Vardi, 1998): GAD67 immunolabeling occurred in the dendritic terminals of A type and the dendritic and axonal terminals of the B type horizontal cells; whereas, GAD65 immunolabeling was found in A type somata and primary dendrites within the visual streak. In mouse, horizontal cells appear to express both GAD65 and GAD67 mRNA and protein (Deniz et al., 2011), but whether there is subcellular distribution difference between the two GAD isoform remains an open question.
The Gad1 gene, encoding GAD67, is highly transcriptionally regulated by DNA methylation of the promoter, and exhibits alternative promoter usage and alternative splicing (Martin and Rimvall, 1993;Tao et al., 2018;Lee et al., 2019), that may account for some of the detection variability. Alternative splicing of Gad1 produces proteins of differing molecular weights: the GAD67, GAD44, and GAD25 isoforms (Behar et al., 1993;Trifonov et al., 2014). Whereas GAD67 is thought to be constitutively active, GAD65 activity can be induced by neuronal activity (Lee et al., 2019). In the CNS, GAD65 is enriched in axonal terminals of GABAergic neurons (Esclapez et al., 1994). It is possible that the state of light adaptation and visual experience before collection of the tissue may influence the levels of protein detected (Connaughton et al., 2001). A transiently expressed GAD25/ES isoform was reported in retina (Connaughton et al., 2001;Dkhissi et al., 2001) and may account for the observed loss of GAD67 immunolabeling with retinal maturation. In addition to GAD67, there are at least 10 alternatively spliced isoforms of the full-length Gad1 gene comprised of 19 exons, producing a GAD44 isoform that has enzymatic activity and several GAD25s that do not (Chessler and Lernmark, 2000;Liu et al., 2010;Trifonov et al., 2014;Tao et al., 2018). The Gad2 gene encoding GAD65 appears to produce two splice variants, including a fulllength mRNA and a truncated version of undefined function (Davis et al., 2016).
There is also post-transcriptional regulation of GAD, including palmitoylation, phosphorylation, and protein cleavage (Baekkeskov and Kanaani, 2009;Lee et al., 2019) that alters GAD protein activity and conformation, intracellular protein localization, and possibly antibody-targeted epitopes. GAD65 and GAD67 can form heterodimers, during targeting of GAD65 and GAD67 to synaptic vesicles in presynaptic terminals (Dirkx et al., 1995;Kanaani et al., 2010). GAD65 can form a complex with the synaptic vesicle proteins, VGAT, cysteine string protein, and heat shock protein 70 (Wei and Wu, 2008), and thus influence GABA loading into synaptic vesicles (Wei and Wu, 2008;Lee et al., 2019). Weak, yet distinct, GABA immunolabeling occurs in the outer retina, in contrast to the strong GABA immunolabeling distributed to amacrine cell and displaced amacrine cell bodies and processes in the inner plexiform layer (IPL). (C) Merged image shows the co-localization in horizontal cell bodies and processes. (D-F) Higher magnification views of the outer plexiform layer (OPL) show the GABA immunoreactivity (D) in the calbindin-identified horizontal cells (E) in the merged image (F). Images are maximum intensity projections of 6 optical sections, z = 5 µm. Scale bar, 20 µm in C (applies to A-C), (F) (applies to D-F). ONL, outer nuclear layer; INL, inner nuclear layer; GCL, ganglion cell layer. [Modified from (Guo et al., 2009)].
The detection of GAD or GABA in the adult retina may be influenced by numerous factors, including the differential expression of GAD isoforms, regulations of levels of Gad transcripts and GAD proteins, and GABA synthesis in horizontal cells, as well as technical issues related to fixation composition, fixation protocols (perfusion or immersion) and antibody specificity Vardi and Auerbach, 1995;Kalloniatis et al., 1996;Johnson and Vardi, 1998;Deniz et al., 2011). Schubert et al. (2010) confirmed expression of GAD67 during neonatal development in mouse, but never detected GAD65 in horizontal cells. Some investigators observed the volatility of GABA immunoreactivity (Kalloniatis et al., 1996;Deniz et al., 2011) and suggested that it may be due to technical issues with harvesting the retina . GABA immunolabeling in mice was maintained by cardiac perfusion, but not postdissection, fixation, and under physiological conditions that promoted GAD activity with L-glutamate/glutamine incubation with co-factor pyridoxal phosphate) or intracardiac perfusion with CNQX and cadmium to inhibit transmitter release from horizontal cells prior to fixation (Deniz et al., 2011).
There is evidence that GAD activity and/or level of expression may be regulated by light (Herrmann et al., 2011) and light adaptation (Connaughton et al., 2001) and this may contribute to inconsistencies in detection of GABA in horizontal cells. The GABA immunostaining in horizontal cells increased as mice were subjected to increasing intensity of background light (Herrmann et al., 2011), indicating light increased GABA immunoreactivity. In addition to changes in GAD activity, light stimulation of the retina would result in membrane (F) Merged image shows the co-expression of GAD65 and calbindin immunoreactivities in the OPL, indicating that GAD65 immunoreactivity is localized to horizontal cell bodies, processes, and tips. Arrows in (D-F) point to GAD65 in horizontal cell endings. Images are maximum intensity projections of 6 optical sections, z = 5 µm. Scale bar, 20 µm in (C) (applies to A-C), (F) (applies to D-F). [Modified from (Guo et al., 2009)].
hyperpolarization of horizontal cells and presumably less release of transmitter. In fish, the levels of the full-length GAD67 mRNA and protein (Connaughton et al., 2001) and GABA were increased in light-adapted retina (Lam, 1972;Starr, 1973;Connaughton et al., 2001). Finally, GAD65 and GAD67 mRNA expression in mouse horizontal cells is consistent with the GFP expression in GAD65-eGFP and GAD67-GFP adult reporter mice (Deniz et al., 2011). These findings suggest expression of both GAD65 and GAD67 in adult mouse horizontal cells occurs (Deniz et al., 2011), but see (Schubert et al., 2010).

GABA Receptors in Outer Retina
The localization of GABA receptors in the outer retina to photoreceptors, bipolar cells, and horizontal cells (Brecha, 1992;Yang, 2004) is congruent with both feedback and feedforward roles for GABA released from horizontal cells. In non-mammalian retina, such as turtle, fish and salamander, photoreceptors clearly possess functional GABA A receptors, as GABA application generated a chloride conductance (Wu and Dowling, 1980;Tachibana and Kaneko, 1984;Kaneko and Tachibana, 1986;Yazulla et al., 1989;Wu, 1992). Reports of clear-cut expression of GABA A receptors in mammalian photoreceptors are scant, although there are reports of GABA A receptor subunit mRNAs by photoreceptors by in situ hybridization (ISH), single-cell RT-PCR, and GABA A receptor subunit immunohistochemistry (Greferath et al., 1993;Grigorenko and Yeh, 1994;Vardi et al., 1998). In rat retina, GABA A receptor subunit α2 is reported to be expressed at cone photoreceptor terminals and the β1, δ, γ2 mRNAs are expressed in the outer nuclear layer (ONL) (Greferath et al., 1995). However, the α1 subunit mRNA was not detected in the ONL of rat retina (Brecha et al., 1991), consistent with the lack of α1 and ρ1 immunoreactivities in mouse cone pedicles Frontiers in Cellular Neuroscience | www.frontiersin.org by immunoelectron microscopy (Kemmler et al., 2014). In neonatal rabbit retina, cone photoreceptors transiently express GABA A receptor subunits α1 and β2/3 (Mitchell and Redburn, 1996;Mitchell et al., 1999), when GABA and GAD67 levels are high in horizontal cells (Schnitzer and Rusoff, 1984). Cone terminals of pig and rat were reported to show GABA A ρ subunit (ρ subunit) immunoreactivity suggesting the presence of a GABA A ρ receptor (Picaud et al., 1998b;Pattnaik et al., 2000). However, Deniz et al. (2019) reported bicuculinesensitive, but not TPMPA-sensitive, GABA evoked currents in mouse cone photoreceptors in retinal slices, suggesting the presence of ionotropic GABA A receptors, but not those comprising ρ-subunits. Rod photoreceptors from cultured pig retina and in mouse retinal slices were reported to exhibit no response to GABA (Picaud et al., 1998b;Deniz et al., 2019).
Evidence for a horizontal cell feed-forward role includes the expression of GABA A receptor immunoreactivity on bipolar cell dendrites Vardi et al., 1992;Greferath et al., 1993;Brecha and Weigmann, 1994;Enz et al., 1996;Wässle et al., 1998;Hoon et al., 2015). GABA A receptor subunit immunoreactivity is localized to bipolar cell membranes adjacent to horizontal cell endings in cone pedicles and underneath the photoreceptor terminals (Greferath et al., 1994;Koulen et al., 1998a;Puller et al., 2014). The extrasynaptic GABA A receptor α6 subunit is expressed on rod bipolar cell dendrites (Figures 3A-C) (Hirano et al., 2016b), which suggests a role for tonic GABA A receptor currents in feedforward signaling.
As functional evidence of a feedforward input, full-field light stimulation, applied in the presence of L-AP4 to block direct photoreceptor input, reduced a gabazine-sensitive current in ON-cone bipolar cells (Yang and Wu, 1991;Chaffiol et al., 2017). This feedforward input results from GABA A receptor activation at ON cone bipolar cell dendrites, which is reduced by horizontal cell hyperpolarization. GABA may evoke responses of opposite polarities in ON and OFF bipolar cells as a result of differing internal chloride concentrations in their dendrites (Duebel et al., 2006). GABA elicited depolarizing inward currents when applied to dendrites of mouse rod bipolar cells and hyperpolarizing currents when applied to OFF-bipolar cells, congruent with feedforward input from horizontal cells (Satoh et al., 2001;Duebel et al., 2006). The basis of the differential intracellular chloride is the expression of Na + -K + -Cl − co-transporter (NKCC), which transports chloride into the cellular compartment, which is prominent in ON bipolar cell dendrites and horizontal cells (Vardi et al., 2000;Dmitriev et al., 2007;Puller et al., 2014). NKCC promotes accumulation of intracellular chloride and generates a chloride equilibrium potential above the resting membrane potential and thus a depolarization when ionotropic GABA receptor chloride channels are opened. In contrast, K + -Cl − co-transporter (KCC2), a chloride extruder, is expressed in OFF bipolar cell dendrites and axonal terminals of ON and OFF bipolar cells (Vardi et al., 2000), where a GABA-activated chloride conductance would elicit a hyperpolarization. Finally, GABA released by horizontal cells appears to act back on the horizontal cells themselves (Kamermans and Werblin, 1992;Blanco et al., 1996;Feigenspan and Weiler, 2004;Varela et al., 2005;Thoreson and Mangel, 2012). In non-mammalian horizontal cells, GABA elicited currents by activating ionotropic GABA A receptors, including GABA Aρ receptors, or electrogenic transporters [fish: (Wu and Dowling, 1980;Schwartz, 1982;Gilbertson et al., 1991;Kamermans and Werblin, 1992;Cammack and Schwartz, 1993;Qian and Dowling, 1993;Takahashi et al., 1994Takahashi et al., , 1995Jung et al., 1999) salamander: (Yang and Wu, 1993;Dong and Werblin, 1994;Yang et al., 1999;Wang et al., 2000)]. GABA elicited ionotropic GABA A receptor-mediated currents in mammalian (rabbit, mouse, rat, human) horizontal cells, but not a transporter-mediated current (Blanco et al., 1996;Picaud et al., 1998a;Feigenspan and Weiler, 2004;Varela et al., 2005;Liu et al., 2013). GABA and/or muscimol application activated ionotropic GABA A receptors and elicited chloride currents, blocked by bicuculline and picrotoxin, in whole-cell recordings of isolated rabbit, mouse, and rat horizontal cells (Blanco and de la Villa, 1999;Feigenspan and Weiler, 2004;Liu et al., 2013). In mouse horizontal cells, we showed distinct immunolabeling for GABA A ρ ρ2 subunit localized predominantly to their endings at its axon terminals within rod spherules and at its dendrites at cone pedicles (Grove et al., 2019;Barnes et al., 2020), indicating the presence of GABA A ρ receptors. Notable characteristics of GABA A ρ receptors include high affinity for GABA and nondesensitizing currents, capable of producing tonic currents at ambient levels of interstitial GABA, similar to extrasynaptic GABA receptors in other areas of the CNS (Bormann and Feigenspan, 1995;Bormann, 2000;Farrant and Nusser, 2005). Horizontal cells, recorded in rodent (mouse, rat, guinea pig) retinal slices, maintained a tonic GABA current in the cone terminal synaptic cleft that was sensitive to TPMPA, a GABA A ρ receptor blocker, and this tonic current proved critical for feedback inhibition of cone calcium current (Grove et al., 2019). Recordings in a horizontal cell conditional knockout of VGAT showed this tonic GABA current was abolished in these horizontal cells (Grove et al., 2019), suggesting that horizontal cells were the source of the GABA. In addition to ionotropic GABA receptors, metabotropic GABA B receptors have been reported on rat horizontal cell processes (Koulen et al., 1998b). Taken together, these studies indicate multiple targets for GABA exist in the OPL, which could mediate the action of horizontal cells in the outer retina.

GABA Transporter (GAT)
Earlier models of GABA release from non-mammalian horizontal cells posited a central role for a Ca-independent, Na-dependent GABA transporter, GAT-1 (Schwartz, 1987;. GABA uptake or release from the cytoplasm (Schwartz, 2002) is unlikely in mammalian horizontal cells based on several findings. First, uptake studies using radiolabeled GABA or GABA analogs have not reported high affinity uptake of these molecules by adult horizontal cells, although high affinity uptake was readily observed in amacrine cells (Ehinger, 1977;Agardh and Ehinger, 1982;Blanks and Roffler-Tarlov, 1982;Mosinger et al., 1986;Pow et al., 1996). In addition, GABA transporter currents  have not been detected in isolated mouse and rabbit horizontal cells (Feigenspan and Weiler, 2004;Varela et al., 2005). These findings are consistent with the failure to detect GAT mRNAs and immunostaining in horizontal cells of mouse, rat, and guinea pig retinas (Brecha and Weigmann, 1994;Honda et al., 1995;Johnson et al., 1996;Guo et al., 2009). In mammalian retinas, GAT-1 and GAT-3 instead are expressed by Müller cells (Johnson et al., 1996;Guo et al., 2009) that take up [ 3 H]-GABA (Marshall and Voaden, 1975;Blanks and Roffler-Tarlov, 1982).
In mammals, a preponderance of evidence shows that GABA meets the criteria for being a neurotransmitter of horizontal cells. There is the synthetic machinery for GABA in horizontal cells, detectable GABA immunoreactivity, and a plethora of GABA receptors in the OPL that would mediate the action of the released GABA (Wässle et al., 1998;. While mammalian horizontal cells do not express GATs, GABA uptake occurs in Müller cell processes that surround photoreceptor terminals, producing a honeycomb pattern in the outer plexiform layer (OPL) (Burris et al., 2002;Guo et al., 2009).
We found syntaxin-4, another isoform that targets vesicles to the plasma membrane (Teng et al., 2001) is highly expressed in horizontal cells at axonal terminals and dendrites (Figure 8), where it is concentrated beneath cone pedicles (Figure 8, arrows), and in the lateral elements at photoreceptor terminals (Hirano et al., 2007). Figures 8A-C shows syntaxin-4 immunolabeling in the OPL of mouse (A), rat (B), and rabbit (C) retina, which co-localizes with the horizontal cell marker, calbindin (Hirano et al., 2007). Syntaxin-4 co-localizes with SNAP-25 in the endings of horizontal cells (Figures 8E-G, Hirano et al., 2007). Immunoelectron microscopy places syntaxin-4 immunoreactivity in the lateral elements at photoreceptor synapses (Figures 6C,D). In other neuronal systems, syntaxin-4 is found in postsynaptic membranes and marks a domain for ionotropic glutamate receptor exocytosis in dendritic spines in hippocampus (Kennedy et al., 2010;Bin et al., 2019) and NGF release from Schwann cells (Lin et al., 2017). At the Drosophila neuromuscular junction, syntaxin-4 is postsynaptic and is involved in retrograde signaling to motoneurons (Harris et al., 2016) to regulate neurotransmitter release and the number of presynaptic active zones and Ca channels (Harris et al., 2018).
We observed VAMP-1, rather than VAMP-2, in horizontal cell endings by double label immunohistochemistry (Figures 9A-C, (Bitzer and Brecha, 2006;Lee and Brecha, 2009). VAMP-2 is the more common VAMP/synaptobrevin isoform in SNARE complexes at conventional synapses, with VAMP-1 occurring to a lesser degree (Elferink et al., 1989;Brunger et al., 2019). In well-fixed mouse retina, the VAMP-1 labeling was reported to be weaker than that of VAMP-2, in the plexiform layers . The strong fixation may have resulted in difficulties in interpretation of VAMP-1 immunostaining, as VAMP-1 immunoreactivity did not appear to label synaptic structures .

Synaptic Vesicle Proteins
Given the prevalent view at the time that there were few or no synaptic vesicles in the horizontal cell endings [ (Schwartz, 2002), but see (Dowling and Boycott, 1966;Dowling, 1970;Raviola and Gilula, 1975;Spiwoks-Becker et al., 2001;Zampighi et al., 2011)] we checked whether there were other key synaptic vesicle proteins in addition to VGAT. There are at least 40 different families of vesicle and synaptic proteins, including the synaptotagmins, synapsins, GTP-binding Rab proteins and complexins, that have critical roles in Ca 2+ -dependent transmitter release, including Ca 2+ sensing, vesicle trafficking, and vesicle fusion (Jahn and Scheller, 2006;Takamori et al., 2006). Most of these proteins have multiple isoforms that are differentially expressed in the nervous system (Linial, 1997;Hong, 2005). From this screen, we localized several synaptic vesicle proteins to horizontal cell endings (Hirano et al., 2005(Hirano et al., , 2007(Hirano et al., , 2011Lee and Brecha, 2010), supporting the hypothesis that horizontal cells contain synaptic vesicles, and transmitter is released by a vesicular mechanism.
SV2A is a ubiquitous synaptic vesicle transporter protein in the brain (Buckley and Kelly, 1985;Bajjalieh et al., 1992;Feany et al., 1992;Janz and Südhof, 1999) and is involved in sensing presynaptic calcium levels to prime synaptic vesicles for calcium-dependent exocytosis Chang and Südhof, 2009;Wan et al., 2010). Knockout of SV2A resulted in a reduction in hippocampal GABAergic neurotransmission (Crowder et al., 1999). In outer retina, SV2A co-localized with VGAT in horizontal cell endings in likely synaptic vesicles . Figures 9D-F shows SV2A colocalized with calbindin in horizontal cell endings, as well as in photoreceptor terminals . SV2A was reported earlier to be transiently expressed in horizontal cells and cone photoreceptors during mouse retina development, but not in adult retina . The lack of double labeling for calbindin to clearly identify horizontal cell processes in the OPL in the relatively low-power magnification images makes it difficult to rule out horizontal cell labeling. In well-fixed adult mouse retina, SV2A was reported to be in cone ribbon synapses and a subset of conventional synapses; whereas, SV2B was in photoreceptor and bipolar cell ribbon synapses and SV2C, to sparse conventional synapses in the outer retina and starburst amacrine cells .
Synapsins are a family of 4 abundant synaptic vesicleassociated phosphoproteins that regulate synaptic vesicle availability (Hilfiker et al., 1999) and are markers of conventional synapses in retina, but not of ribbon synapses (Mandell et al., 1990). Synapsin I was expressed at low levels in rabbit horizontal cells (Hirano et al., 2005), consistent with the likely horizontal cell labeling in ferret retina (Karne et al., 1997) and guinea pig horizontal cells, which show strong immunolabeling . Ultrastructurally, synapsin I immunolabeling was found in the horizontal cell axonal endings at rod photoreceptor synapses ( Figure 6E, Hirano et al., 2005). Consistent with the immunolabeling, synapsin mRNA localized to presumed horizonal cells in developing rat retina (Haas et al., 1990).

Presence of a Ca 2+ Sensor and Voltage-Gated Ca Channels
The localization of synaptotagmin-2 to horizontal cells indicated that a calcium sensor for neurotransmitter release is present in these terminals (Figures 10D-F, Fox and Sanes, 2007;Lee and Brecha, 2010). Rabbit, cat and mouse horizontal cells express L-type voltage-dependent Ca 2+ channels (Ueda et al., 1992;Löhrke and Hofmann, 1994;Schubert et al., 2006;Liu et al., 2016), which are known to regulate sustained transmitter release in photoreceptor and bipolar cells and to modulate transmitter release smoothly and continuously with changes in membrane potential that accompany changing levels of illumination (Corey et al., 1984;Wilkinson and Barnes, 1996;de la Villa et al., 1998;Barnes and Kelly, 2002;Morgans et al., 2005;Mercer and Thoreson, 2011;Van Hook et al., 2019). The minimal voltage-dependent inactivation, characteristic of L-type Ca 2+ channels, is wellsuited for maintaining constant output at these tonic synapses (Juusola et al., 1996). Figures 10A-C shows immunolabeling for Ca v 2.2 and horizontal cell marker calbindin , and the co-localization of Ca v 2.2 to the horizontal cell axonal terminals and at cone pedicle dendritic contacts suggest N-type Ca channels may play a role in transmitter release. In rat, Ca v 1.2 (L-type, α1C), Ca v 2.1 (P/Qtype, α1A), and Ca y 2.2 (N-type, α1B) were localized by immunohistochemistry to horizontal cell endings (Liu et al., 2013). These findings are consistent with the physiological data supporting three types of voltage-gated Ca channels in mouse horizontal cells based on pharmacological discrimination using nifedipine/verapamil, ω-agatoxin IVA and ω-conotoxin GVIA, respectively (Schubert et al., 2006;Liu et al., 2013).  Bitzer and Brecha, 2006;Brecha et al., 2010).

PRESENCE OF VESICLES IN HORIZONTAL CELL PROCESSES AND ENDINGS AS POTENTIAL VESICULAR SOURCES OF GABA RELEASE
Initial electron microscopic studies of horizontal cell endings of cat, rabbit, and primate retina (Dowling and Boycott, 1966;Raviola and Gilula, 1975) reported infrequent small, clear-core vesicles using different fixation protocols, with the most detailed report in the rat retina (Gray and Pease, 1971). Clear-core vesicles represent a type typically containing small molecule transmitters, such as GABA, glutamate, or acetylcholine, and not catecholamines or peptides. These vesicles are similar in appearance to the small, clear-core vesicles in adjacent photoreceptor terminals ( Figure 12A).
We have used conical tomography electron microscopy (Zampighi et al., 2008(Zampighi et al., , 2011 to evaluate horizontal cell dendritic and axonal endings in mouse and guinea pig photoreceptor invaginations. Conical electron microscopy is a high resolution, electron microscopic technique with ∼3 nm isotropic resolution in the x-, y-, and z-planes. Essentially, this resolution eliminates the projection artifact common in thicker conventional and scanning block-face electron microscopic images that obscures fine cytoplasmic and membrane detail (Zampighi et al., 2008).
We have identified numerous small, clear-core vesicles, clathrin-coated vesicles, and patches of plasma membrane thickenings with prominent cytoplasmic specializations in the mouse horizontal cell terminals (Figures 12A, 13A, Zampighi et al., 2011). The small, clear-core vesicles have several fine fibrils that are readily seen in the conical tomograms, although they are not seen in conventional electron micrographs. These vesicles are similar in appearance to descriptions of synaptic vesicles in neurons (Peters et al., 1991). A preliminary comparison in mouse horizontal cells indicates a greater number of vesicles in axonal endings compared to dendritic endings. Vesicle diameters in these endings range between 37 and 62 nm with 2 major peaks at 46 and 53 nm, and a smaller peak at 40-41 nm . Overall, horizontal cell vesicle size is larger than the rod vesicle size (Figures 12B,C; N = 120; 6 endings). Interestingly, inspection of vesicle sizes in a primate cone terminal and adjacent horizontal cell dendrite (Raviola and Gilula, 1975) also shows that the vesicles in the cone cytoplasm are smaller overall than the vesicles in the horizontal cell dendritic ending and similarly in horizontal cell axon terminals (Moser et al., 2020). In addition, to numerous small vesicles, the horizontal cell terminal occasionally contained endocytotic (Figure 13A, red arrow) and clathrincoated vesicles (Figure 13B2, Zampighi et al., 2011). Some larger and irregular shaped vesicles were also seen in horizontal cell terminals of rat or guinea pig retina (Gray and Pease, 1971). The presence of both endosomes (Figure 13A, red arrow) and clathrin-coated vesicles is indicative of active processes occurring in these terminals.
Horizontal cell membranes that are opposite and flanking the arciform density of the mouse photoreceptor terminal are characterized by membrane specialization in conventional electron microscopic preparations (Dowling and Boycott, 1966;Gray and Pease, 1971;Raviola and Gilula, 1975;Linberg and Fisher, 1988). Plasma membrane specializations (arrowheads) also occur along different infoldings of the horizontal cell plasma membrane within the invagination (Figure 12A, Zampighi et al., 2011). There are examples of small vesicles connected by thin tethers to the plasma membrane or are closely associated with FIGURE 11 | Schematic of synaptic proteins found in mammalian horizontal cells. The diagram depicts a synaptic vesicle studded with synaptic vesicle proteins, VGAT, a neurotransmitter transporter, SV2A, Synaptotagmin-2, a calcium sensor, Synapsin I, and SNARE protein, VAMP-1. The other 2 SNARE proteins that form the minimal complex are Syntaxin-1,−4, and SNAP-25, that brings the synaptic vesicle close to the plasma membrane for fusion. Finally, complexin-1/2 is a SNARE-associated protein. The yellow circles represent GABA that is accumulated inside synaptic vesicles by VGAT.
these plasma membrane specializations (Figure 12A arrowheads, Figure 13A). In addition, small vesicles are near and adjacent to the plasma membrane in different parts of the horizontal cell terminal (Figures 12A, 13A). Together, these observations suggest the possibility that vesicle fusion and transmitter release sites are located at multiple sites within the horizontal cell terminals.
Vesicle clustering at membrane thickenings typical of many neuronal central synapses was not observed in early reports on primate, cat, rabbit, and rat horizontal cells (Dowling and Boycott, 1966;Raviola and Gilula, 1975;Kolb, 1977;Schaeffer et al., 1982;Peters et al., 1991). These findings may reflect a sampling issue of synapses that are sparsely distributed, as other ultrastructural studies on cat, rabbit, mouse, primate, mudpuppy, salamander, catfish, and turtle retinas demonstrated small clusters of synaptic vesicles in horizontal cell processes adjacent to membrane thickenings in bipolar cell dendrites, suggestive of horizontal cells feedforward synapses Olney, 1968;Dowling and Werblin, 1969;Dowling, 1970;Lasansky, 1973;Fisher and Boycott, 1974;Raviola and Gilula, 1975;Kolb and Jones, 1984;Sakai and Naka, 1986;Linberg and Fisher, 1988;Greferath et al., 1994). In human retina, horizontal cells were shown to make synaptic contacts with rod bipolar cell dendrites and the rod spherule within the invagination (Linberg and Fisher, 1988). Infrequent horizontal cell synapses with interplexiform processes were found in cat and rabbit also (Kolb, 1974;Kolb and West, 1977;Greferath et al., 1994).
The relative dearth and scattered distribution of synaptic vesicles in horizontal cell endings are similar to the observations of dopaminergic neurons that signal by extrasynaptic somatodendritic release, where it has been difficult to unequivocally identify the organelles (small clear-core vesicles, tuberovesicles, and large dense-core vesicles) that mediate dopamine release (Puopolo et al., 2001;Fortin et al., 2006;Hirasawa et al., 2012Hirasawa et al., , 2015Ludwig et al., 2016). Moreover, the dopaminergic amacrine cell perikaryon does not contain active zones; although, active zones were observed at their dendritic synapses with AII amacrine cells (Puopolo et al., 2001).

DEPOLARIZATION-AND CALCIUM-DEPENDENT SYNAPTIC VESICLE FUSION AND RECYCLING
Ca 2+ -regulated transmitter release is a well-established mechanism in the CNS (Südhof, 2013;Kaeser and Regehr, 2014;Rizo, 2018). In the mammalian retina, evidence supports the idea that horizontal cell transmitter release is regulated by Ca 2+ . The support includes demonstration of voltage-gated Ca 2+ currents (I Ca ) in horizontal cells (Schubert et al., 2006;Liu et al., 2016) and the localization of L-, N-, and P/Qtype Ca 2+ channels (Liu et al., 2013) and the Ca 2+ sensor, synaptotagmin-2 (Figures 10D-F, Hirano and Brecha, 2010;Lee and Brecha, 2010) to horizontal cell terminals. N-type Ca 2+ channels are of particular interest, since they mediate vesicle release at central synapses (Catterall, 2011). Somatodendritic secretion of dopamine and peptides relies on L-type Ca channels primarily (Ludwig et al., 2016). In striatum, dopamine release can involve N-, Q-, T-, and L-type voltage-gated Ca channels, depending on neuronal activity, diverse calcium dependence, and calcium buffering in different cellular domains (Brimblecombe et al., 2015).
Using a luminal VGAT-C antibody in a retinal slice assay, we show that the voltage-gated Ca channels participate in Ca 2+ -mediated vesicular release from horizontal cells Figure 14A. We developed a retinal slice assay (Lee, 2010;Vuong et al., 2011) to monitor VGAT-expressing vesicles, based on topological studies that showed the C-terminus of VGAT is located within the vesicle lumen and using a fluorophore-conjugated, C-terminal directed VGAT (VGAT-C) antibody (Martens et al., 2008). Depolarization resulted in an Oyster550-VGAT-C terminus antibody labeling of the internal face of exocytosed synaptic vesicles, now exposed to the extracellular milieu containing the Oyster550-VGAT-C antibod ( Figure 14A). In retinal slices, depolarization with high [K + ] or 50 µM kainate ( Figure 14D) in the presence of the VGAT-C antibody resulted in punctate VGAT-C labeling of horizontal cell endings in the OPL (Figures 14B,C), indicative of synaptic vesicles fusion with the plasma membrane. Vesicle fusion is only detected with the VGAT-C antibody and not with a N-terminal, cytoplasmically directed VGAT antibody (Figure 14D), indicating the labeling was not non-specific uptake. Labeling is absent or below detection in control experiments [e.g., basal 3 mM [K + ] (Figures 14C' ,D), Oyster550-VGAT-N antibodies (Figure 14D)]. We showed the VGAT-C antibody uptake in horizontal cell processes occurred in basal 2 and 10 mM [Ca 2+ ] o ; whereas, no labeling occurred in nominally 0 mM [Ca 2+ ] o (Figure 14E, Supplementary Figure 1) or in the presence of general (Cd 2+ , Co 2+ ) and voltage-gated Ca channel subtype-specific blockers (ω-agatoxin, ω-conotoxin, nifedipine) (Supplementary Figure 2). These data indicate that the vesicle fusion in horizontal cell endings was depolarization-and calcium-dependent. Further, multiple rounds of labeling with depolarization, depicted in the schematic in Figure 14A, could be visualized using Alexa488-conjugated secondary antibodies to the VGAT-C primary antibodies (Figures 14B,B' ,B",F), suggesting that the initially labeled vesicles are capable of recycling (Lee, 2010;Vuong et al., 2011).

Feedback to Photoreceptors
Finally, we showed that feedback inhibition to photoreceptors occurs in a GABA-dependent manner to modulate the photoreceptor calcium current (Figure 15). To assay feedback, photoreceptors in slices ( Figure 15E) were loaded with the calcium indicator Fluo-4 (green) in a Cx57-tdTomato retina, where the horizontal cells express the red fluorescent reporter tdTomato (converted to magenta), to show the relationship between horizontal cell processes and the photoreceptor cell bodies that were imaged. The increase in photoreceptor intracellular Ca 2+ in response to pulses of 30 mM K + was evaluated using drugs that depolarized or hyperpolarized horizontal cells (Vessey et al., 2005;Liu et al., 2013;Hirano et al., 2016a). A pulse of 30 mM K + drove Ca influx through the voltage-gated calcium channels in the photoreceptors in control conditions, and then, when a second pulse was applied in the presence of kainate to depolarize the horizontal cells, the second pulse produced a smaller peak in intracellular Ca 2+ , showing that horizontal cell depolarization produced an inhibitory signal on the photoreceptor calcium channels ( Figure 15A). Conversely, when the horizontal cells were hyperpolarized with 2,3-Dioxo-6-nitro-1,2,3,4,-tetrahydrobenzo[f ]quinoxaline-7-sulfonamide (NBQX), via blockade of ionotropic glutamate receptors during the second pulse, the calcium signal in photoreceptors is increased, indicating decreased feedback inhibition from horizontal cells to the photoreceptors ( Figure 15C). These findings are consistent with reports in mouse retina (Babai and Thoreson, 2009). Kainate did not produce a change in photoreceptor calcium signal upon superfusion prior to the high-K + pulse, consistent with a lack of ionotropic glutamate receptors on photoreceptors (Babai and Thoreson, 2009). To evaluate the role of vesicular GABA release in this feedback, we conditionally knocked out VGAT by crossing the horizontal cell-specific Cx57-iCre mouse (Hirano et al., 2016a) with a floxed VGAT mouse line (Tong et al., 2008). With the VGAT gene deleted, the neurotransmitter, most likely GABA, cannot be packaged into synaptic vesicles and released. Immunostaining for VGAT confirmed that the VGAT was selectively knocked out in horizontal cells (Hirano et al., 2016a). Whole-cell recordings of VGAT −/− horizontal cells showed that the voltage-gated K + and Ca 2+ membrane currents were normal. In the horizontal cell VGAT knockout, kainate did not produce increased feedback inhibition and NBQX did not result in decreased feedback inhibition (Figures 15B,D,F, Hirano et al., 2016a). These data show that the loss of horizontal cell VGAT eliminated feedback inhibition onto photoreceptors.

DISCUSSION
In mammals, a preponderance of experimental findings indicates that retinal horizontal cells utilize a vesicular mechanism of transmitter release. The evidence for GABA as the horizontal cell neurotransmitter is the presence of GABA immunoreactivity and of GABA synthesizing enzymes (GAD65 and/or GAD67), and postsynaptic targets bearing GABA receptors (photoreceptors, bipolar cells) as well as autoreceptors on horizontal cells. Neonatal rabbit horizontal cells show 3 H-GABA uptake that is downregulated after P5 (Redburn and Madtes, 1986); however, adult mammalian horizontal cells are atypical GABAergic neurons, in that they do not express plasmalemmal GABA transporters. The GABA transporters are expressed by Müller cells, whose processes ensheath photoreceptor synapses (Guo et al., 2009). The horizontal cells express SNARE proteins required for membrane fusion of synaptic vesicles . The fusion of VGAT-bearing synaptic vesicles with the plasma membrane is depolarizationand calcium-dependent and show the capacity for multiple rounds of vesicle fusion in retinal slice preparations. The selective knockout of VGAT from horizontal cells that resulted in the loss of the tonic GABA current (Grove et al., 2019) and disrupted feedback inhibition to photoreceptors, showing that GABA release plays an integral role in these cells' neurotransmission (Hirano et al., 2016a). When VGAT-N antibodies that recognize a cytosolic epitope not exposed to extracellular milieu were used, no specific labeling of horizontal cell endings was (Continued) FIGURE 14 | observed. This finding indicated that the VGAT-C immunolabeling was not a result of non-specific uptake of antibody *p < 0.02. (E) Quantification of VGAT-C immunofluorescence under different extracellular Ca concentrations. Significant increases in VGAT-C immunolabeling were observed in basal (2 mM) and high (10 mM) [Ca 2+ ] o conditions. In contrast, little to no labeling occurred in nominally calcium-free media (0 mM) *p < 0.01, **p < 0.005. (F) Quantification of VGAT-C immunofluorescence during multiple rounds of depolarization, marked by different fluorophores (Oyster550 vs. Alexa488). Goat or donkey anti-rabbit IgG-Alexa488 recognized the VGAT-C primary antibody from the initial round of immunolabeling. In contrast when goat anti-rabbit-Alexa488 IgG was present during the subsequent incubation period in basal [K + ] o , significantly less immunolabeling was observed. Similarly, when a goat anti-mouse IgG-Alexa488 was used, little to no immunolabeling was observed ****p < 0.001. (B,B',B",C,C',C,") maximum intensity projections, z = 5.0 µm. Scale bars, 20 µm. (Modified from Lee, 2010;Vuong et al., 2011). The low numbers and the scattered appearance of synaptic vesicles in horizontal cell terminals as well as the absence of clearcut active zones (Figures 12, 13) are morphological features found in dendrites that are known to release transmitter; for example, somatodendritic dopamine and GABA release from dopaminergic neurons in the CNS (Hirasawa et al., 2015;Ludwig et al., 2016). In striatum, only a third of the dopaminergic boutons expressed a minimal active zone-like cluster of RIM, bassoon and ELKS (Liu et al., 2018). Ultrastructural analysis of dopaminergic amacrine cell somata revealed no active zones and few synaptic vesicles and tubulovesicular organelles (Puopolo et al., 2001). Nevertheless, somatodendritic release exhibits properties of regulated exocytosis, such as calcium dependence (Chen et al., 2006), the involvement of SNARE protein isoforms SNAP-25, VAMP2, and syntaxin 3b (Fortin et al., 2006;Witkovsky et al., 2009;Rice and Patel, 2015;Ludwig et al., 2016), voltage-gated Ca channels (Puopolo et al., 2001;Brimblecombe et al., 2015;Rice and Patel, 2015), and quantal release (Jaffe et al., 1998;Puopolo et al., 2001). The subtypes of voltage-gated Ca channels involved can differ in dopaminergic neurons and may reflect different modes of release (e.g., somatodendritic vs. axonal, firing patterns) (Ludwig et al., 2016). In retina, dopamine acts at synapses as well as by volume transmission (Witkovsky, 2004). The abundance of GABA A receptors in the OPL along with the relatively few synapses found in horizontal cells to bipolar cell dendrites suggests that horizontal cell GABA may be acting by volume transmission. From the robust presence of syntaxin 4 and SNAP-25 throughout horizontal cell processes, it would be interesting to know if GABA release occurred extrasynaptically as well as synaptically from horizontal cells. Also, these SNARE proteins may function in the regulation of GABA or ionotropic glutamate receptor exocytosis, as syntaxin-4 is reported to be important for postsynaptic dendritic exocytosis in hippocampal neurons (Kennedy et al., 2010;Ovsepian and Dolly, 2011;Gu and Huganir, 2016;Bin et al., 2019).

FIGURE 15 |
Transmitter release by a regulated vesicular mechanism would be highly advantageous for fine control of feedback and feedforward action in the outer retina, as there are multiple molecular control points to modulate secretion from horizontal cells that utilizes the bicarbonate permeability of GABA A receptors to regulate cleft pH (Grove et al., 2019;Barnes et al., 2020). For instance, VGAT's dependence on a proton gradient for GABA uptake would influence vesicular GABA concentrations  and, by extension, postsynaptic responses. The possible complex of VGAT and GAD65 (Wei and Wu, 2008) on synaptic vesicles in horizontal cell synaptic endings and its regulation on demand (Buddhala et al., 2009) would also stimulate GABA loading of synaptic vesicles. The highly regulated cascade of SNARE protein interactions in exocytosis would allow for a precise control of the rate and level of transmitter secretion. Local modulation of membrane potential at different endings could also differentially influence presynaptic Ca 2+ channel dynamics and influence local GABA release.
The demonstration of GABA and its synthetic enzyme GAD65 and/or GAD67 in adult mammalian horizontal cells supports the notion that GABA is acting as a transmitter, despite not bearing GATs, notably like cerebellar Purkinje cells (Ribak et al., 1996;Guo et al., 2009). Müller cell processes that envelop photoreceptor terminals in the OPL are well placed to take up GABA (Guo et al., 2009), similar to Bergmann glia around the Purkinje cells (Ribak et al., 1996). The GABA receptors on horizontal cells and bipolar cell dendrites Wässle et al., 1998; indicate the receptor targets of the GABA released by horizontal cell are present.
The presence of GABA A receptor ρ2 (Grove et al., 2019) immunolabeling on horizontal cell terminals implies a significant role for tonic GABA modulation of horizontal cell membrane potential and conductance, signaling which is mediated by graded regulation rather than phasic synaptic transmission. Horizontal cells appear to be the primary source of GABA in the outer retina, as VGAT knockout resulted in a loss of the TPMPA-sensitive GABA-induced current in horizontal cells and feedback regulation of photoreceptor Ca channels (Grove et al., 2019). In primate retina, Haverkamp and colleagues Puller et al., 2014) described layers of horizontal cell processes under primate cone pedicles and GABA A receptorbearing bipolar cell dendrites sandwiched between the two layers, and postulated that a GABA tone may be present. There is an enrichment of syntaxin-4 in horizontal cell processes at Scones in primate retina and combined with expression of GABA A receptor α1 and ρ subunits and the chloride-accumulating transporter NKCC vitreal to S-cone pedicles is suggestive of HII horizontal cell to blue cone bipolar cells feedforward signal transmission (Puller et al., 2014). There are also interplexiform processes from GABAergic tyrosine hydroxylase (TH) amacrine cells that form synapses in the OPL onto bipolar cell processes (Dowling and Ehinger, 1975;Kolb and West, 1977;Linberg and Fisher, 1986;Chun and Wässle, 1989;Greferath et al., 1994), as well as a non-dopamine containing GABAergic interplexiform cell in mouse (Dedek et al., 2009), that might contribute to GABA levels in the OPL Witkovsky et al., 2008). Grove et al. (2019) showed that the high-affinity, non-desensitizing GABA A ρ receptors on horizontal cell endings generate a tonic GABA current in the outer retina, most notably within the photoreceptor terminal synapse. Changes in tonic inhibition can alter neuronal and network properties, due to a persistent increase in input conductance that will regulate membrane excitability and alter the gain of a neuron's input-output relationship, and thus a neuron's responsiveness (Semyanov et al., 2004;Walker and Semyanov, 2008;Lee and Maguire, 2014).
In addition to feedback to photoreceptors, GABA A receptors on bipolar cell dendrites relay the horizontal cell feedforward signal (Vardi et al., 1992;Enz et al., 1996;Wässle et al., 1998;Hoon et al., 2015;Chaffiol et al., 2017), although which bipolar cell type and the GABA A receptor subtypes used are not yet welldefined. The GABA A receptors are ρ-containing or extrasynaptic GABA A receptors containing δ subunits, such as those that contain GABA A receptor α6 subunits, are high affinity and nondesensitizing, and mediate tonic inhibition in other CNS areas (Farrant and Nusser, 2005;Glykys and Mody, 2007;Belelli et al., 2009;Brickley and Mody, 2012). Consistent with higher levels of GABA in the interstitial space of the OPL, the background labeling for GABA was observed to be higher in the OPL .
Conical electron tomography of mouse rod spherules and cone pedicles clearly demonstrate the presence of small, clearcore vesicles in horizontal cell axonal endings and dendritic endings, that are slightly larger compared to synaptic vesicles in rod and cone photoreceptors. Furthermore, the conical tomograms reveal putative active zones and membrane densities in the horizontal cell endings (Zampighi et al., 2011), endocytosis of clathrin-coated vesicles and vesicle tethers, indicative of vesicle specializations and vesicular activity that were not observed in thicker ultrathin sections due to projection artifacts. The presence of synaptic vesicles in horizontal cells is supported by conventional transmission electron micrographs from mouse (Spiwoks-Becker et al., 2001), cat and primates (Dowling and Boycott, 1966;Dowling, 1970;Raviola and Gilula, 1975;Linberg and Fisher, 1988), that depict small, clear-core vesicles in horizontal cell processes at both rod and cone photoreceptor synapses. The presence of synaptic vesicle proteins, such as VGAT and SV2A, in horizontal terminals further supports the conclusion of the presence of synaptic vesicles in these terminals.
These synaptic vesicles are the cellular substrate for the many synaptic vesicle proteins localized to horizontal cell endings, including VGAT, SV2A, synapsin I, complexin-1/2, synaptoporin (Brandstätter et al., 1996a;Hirano et al., 2005Hirano et al., , 2007Lee and Brecha, 2010). The SNARE complex proteins of syntaxin-1a and syntaxin-4, VAMP-1, and SNAP-25 (Hirano et al., 2005(Hirano et al., , 2007(Hirano et al., , 2011 along with the SNARE complex associated proteins (complexin-1/2 and synaptotagmin-2; Hirano et al., 2005) mediate and modulate vesicle fusion with the membrane. The use of less common isoforms, e.g. VAMP1, synaptotagmin-2, complexin-1/2, for synaptic vesicle release may reflect different kinetics and/or regulation at this tonic, graded potential GABAergic synapse (Reim et al., 2005;Hua et al., 2011;Chung and Raingo, 2013), similar to the usage of alternative synaptic protein isoforms at ribbon synapses (Moser et al., 2020). The subcellular localization of VAMP1 to horizontal cell terminals suggest it participates in synaptic transmission and that other VAMP isoforms may mediate vesicle trafficking between other cellular compartments within the cell. Although the precise role of syntaxin-4 remains to be determined, its high level of expression in horizontal cells, in addition to syntaxin-1a, likely reflect distinct pools of vesicles trafficked to the membrane. The calcium sensor, synaptotagmin-2, is preferentially expressed at cerebellar GABAergic synapses, where it is the fast Ca sensor and responsible for faster replenishment of the readily releasable pool, necessary for fast feedforward inhibition (Chen et al., 2017).
The functional VGAT-C antibody uptake studies indicated that the VGAT-containing synaptic vesicles fuse with the plasma membrane in a depolarization-and calcium-dependent manner, characteristic of vesicular exocytosis of transmitter (Südhof, 2013). Moreover, the synaptic vesicles recycle, as observed from the multiple rounds of labeling. Finally, the knockout of horizontal cell VGAT resulted in loss of GABA release and eliminated feedback inhibition of photoreceptor calcium channels (Hirano et al., 2016a;Grove et al., 2019). Together these findings show that the vesicular release of GABA from mammalian horizontal cells plays an essential role in horizontal cell synaptic transmission. Grove et al. (2019) demonstrated that cone photoreceptor calcium currents are modulated by picrotoxin and TPMPA (see Grove et al., Figures 1-2), and that they act at GABA A R-ρ receptors on the horizontal cell endings (see Grove et al., Figures  2I-K), not photoreceptors. This GABAergic modulation is absent in the presence of HEPES, indicating pH sensitivity. Grove et al. (2019) extended the findings in Hirano et al. (2016a) by showing that GABA release by horizontal cells acts back onto its own GABA receptors, and that the GABA release (and cone Ca channel modulation) is abolished in Cx57-VGAT −/− horizontal cells, concluding that the bicarbonate flux through these tonic GABA receptors regulates the synaptic cleft pH. The full hybrid GABA-pH model is much more detailed (see Figure 8, Grove et al., 2019; Figure 12, Barnes et al., 2020), including roles for sodium-proton exchangers (NHEs), the bicarbonate equilibrium potential and horizontal cell membrane potential excursions induced by GABA and glutamate (Grove et al., 2019;Barnes et al., 2020).

AUTHOR CONTRIBUTIONS
AH, SS, SB, and NB designed the experiments. HV and HK performed the VGAT-C functional labeling experiments. AH and HK performed the immunohistochemistry. HK performed PLA. CS performed the conical tomography. SB managed the calcium imaging experiments. AH, SB, and NB wrote the manuscript. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS
We gratefully thank our numerous collaborators and technical associates over the last two decades.