Mammalian retinal horizontal cells are unconventional GABAergic neurons


  • Sercan Deniz,

    1. Department of Neurobiology and Genetics, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS UMR_7104, Inserm U 964, Université de Strasbourg, Illkirch, France
    2. Institut de la Vision, INSERM UMR_S968, CNRS UMR_7210, Université Pierre et Marie Curie Paris 06, Paris, France
    Search for more papers by this author
    • These authors contributed equally to this study.

  • Eric Wersinger,

    1. Department of Neurobiology and Genetics, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS UMR_7104, Inserm U 964, Université de Strasbourg, Illkirch, France
    2. Institut de la Vision, INSERM UMR_S968, CNRS UMR_7210, Université Pierre et Marie Curie Paris 06, Paris, France
    Search for more papers by this author
    • These authors contributed equally to this study.

    • The present address of Eric Wersinger is The Johns Hopkins University School of Medicine, Department of Otolaryngology Head and Neck Surgery, 720 Rutland Ave. Ross 818, Baltimore, MD 21205, USA.

  • Yannick Schwab,

    1. Imaging Center, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS UMR_7104, Inserm U 964, Université de Strasbourg, Illkirch, France
    Search for more papers by this author
  • Carole Mura,

    1. Eurexpress Platform, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS UMR_7104, Inserm U 964, Université de Strasbourg, Illkirch, France
    Search for more papers by this author
  • Ferenc Erdelyi,

    1. Department of Gene Technology and Developmental Neurobiology, Institute of Experimental Medicine, Budapest, Hungary
    Search for more papers by this author
  • Gábor Szabó,

    1. Eurexpress Platform, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS UMR_7104, Inserm U 964, Université de Strasbourg, Illkirch, France
    Search for more papers by this author
  • Alvaro Rendon,

    1. Institut de la Vision, INSERM UMR_S968, CNRS UMR_7210, Université Pierre et Marie Curie Paris 06, Paris, France
    Search for more papers by this author
  • José-Alain Sahel,

    1. Institut de la Vision, INSERM UMR_S968, CNRS UMR_7210, Université Pierre et Marie Curie Paris 06, Paris, France
    2. Centre d’Investigation Clinique 503, Inserm-Centre Hospitalier National d’Ophtalmologie des Quinze-Vingts, Paris, France
    3. Fondation Ophtalmologique Adolphe de Rothschild, Paris, France
    Search for more papers by this author
  • Serge Picaud,

    1. Institut de la Vision, INSERM UMR_S968, CNRS UMR_7210, Université Pierre et Marie Curie Paris 06, Paris, France
    2. Fondation Ophtalmologique Adolphe de Rothschild, Paris, France
    Search for more papers by this author
  • Michel J. Roux

    1. Department of Neurobiology and Genetics, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS UMR_7104, Inserm U 964, Université de Strasbourg, Illkirch, France
    Search for more papers by this author

Address correspondence and reprint requests to Michel J. Roux, Department of Neurobiology and Genetics, Institut de Génétique et de Biologie Moléculaire et Cellulaire, CNRS UMR 7104, Inserm U 964, Université de Strasbourg, 1 rue Laurent Fries BP10142, 67404 Illkirch Cedex, France. E-mail:


J. Neurochem. (2011) 116, 350–362.


Lateral interactions at the first retinal synapse have been initially proposed to involve GABA by transporter-mediated release from horizontal cells, onto GABAA receptors expressed on cone photoreceptor terminals and/or bipolar cell dendrites. However, in the mammalian retina, horizontal cells do not seem to contain GABA systematically or to express membrane GABA transporters. We here report that mouse retinal horizontal cells express GAD65 and/or GAD67 mRNA, and were weakly but consistently immunostained for GAD65/67. While GABA was readily detected after intracardiac perfusion, it was lost during classical preparation for histology or electrophysiology. It could not be restored by incubation in a GABA-containing medium, confirming the absence of membrane GABA transporters in these cells. However, GABA was synthesized de novo from glutamate or glutamine, upon addition of pyridoxal 5′-phosphate, a cofactor of GAD65/67. Mouse horizontal cells are thus atypical GABAergic neurons, with no functional GABA uptake, but a glutamate and/or glutamine transport system allowing GABA synthesis, probably depending physiologically from glutamate released by photoreceptors. Our results suggest that the role of GABA in lateral inhibition may have been underestimated, at least in mammals, and that tissue pre-incubation with glutamine and pyridoxal 5′-phosphate should yield a more precise estimate of outer retinal processing.

Abbreviations used:

artificial CSF








glutalic acid decarboxylase


green fluorescent protein


horizontal cell


inner nuclear layer


inner plexiform layer


normal goat serum


outer plexiform layer


phosphate buffer saline




pyridoxal 5′-phosphate

Retinal horizontal cells (HCs) are long known to be involved in the modulation of information transfer between photoreceptors and bipolar cells (Baylor et al. 1971; reviewed in Kamermans and Spekreijse 1999; Piccolino 1995) but the mechanism by which they do so is still highly debated. It was initially thought to be mediated by GABA, as in lower vertebrates HCs have been demonstrated to release GABA [via reverse uptake and not by a Ca2+-dependent, vesicular mechanism (Schwartz 1987)] and that cones were responding to externally applied GABA (Tachibana and Kaneko 1984). However, subsequent work has put forward alternative hypothesis for the basis of center-surround receptive fields, through a modulation of the synaptic cleft pH (Hirasawa and Kaneko 2003; Cadetti and Thoreson 2006; Davenport et al. 2008) or by an ephaptic mechanism implicating hemigap junctions (Kamermans et al. 2001; Pottek et al. 2003; Fahrenfort et al. 2004).

In mammalian retinas, expression of GABA receptors on cone terminals is controversial: GABAA and GABAC receptors were described in cultured pig cones (Picaud et al. 1998b) and in mouse cones from degenerated rd1 retinas (Pattnaik et al. 2000), but responses to GABA were not detected in cones from macaque (Verweij et al. 2003) or grey squirrel (S. DeVries, personal communication) retinas. While this argues against a major role for HC feedback onto cones, data obtained in the last decade points toward an implication of this inhibitory neurotransmitter in the mammalian outer retinal processing. HCs express the vesicular GABA and glycine amino-acid transporter (VIAAT or VGAT) in mouse, macaque and guinea pig retinas (Haverkamp et al. 2000; Cueva et al. 2002; Jellali et al. 2002; Guo et al. 2010) as well as in cultured human HCs (Jellali et al. 2002). In addition, HCs were shown to express the synaptic proteins required for vesicular release in the rabbit (Hirano et al. 2005, 2007) and guinea pig (Lee and Brecha 2010) retinas – though there is no well-defined pre-synaptic specialization in HC terminals, nor post-synaptic density in the contacted photoreceptors and bipolar cell dendrites. On the post-synaptic side, a role for GABA in the triadic synapse formed by photoreceptors, bipolar and horizontal cells, is suggested by the expression of GABA receptor in bipolar cells dendrites (reviewed in Wässle et al. 1998). Moreover, the differential expression of the two Cl co-transporters, NKCC and KCC2, in ON and OFF bipolar cell dendrites (Vardi et al. 2000), allowing GABA to be depolarizing in the ON pathway (Varela et al. 2005a; Duebel et al. 2006) is compatible with HC influencing the center-surround receptive fields of bipolar cells through feed-forward GABAergic signaling.

However, heterogeneous results were published regarding the GABAergic nature of mammalian HCs. Notably, in nocturnal rodents, GABA and/or glutamic acid decarboxylase (GAD) were detected in HCs only during development (Schnitzer and Rusoff 1984; Agardh et al. 1986; Versaux-Botteri et al. 1989; Fletcher and Kalloniatis 1997; Yamasaki et al. 1999; Schubert et al. 2010). Furthermore, by contrast with classic GABAergic neurons, mammalian HCs do not express membrane GABA transporters, not even the low affinity betaine/GABA transporter BGT-1 (Honda et al. 1995; Johnson et al. 1996; Guo et al. 2009, 2010).

As green fluorescent protein (GFP) expression was driven by the GAD65 promoter in a subset of adult HCs in a transgenic mouse line, we re-investigated the presence of GABA in these cells. GAD65 and/or GAD67 were detected both by immunohistochemistry and in situ hybridization, however at a lower level than in amacrine cells. GABA was present in HCs after fixation by intracardiac perfusion, especially when vesicular release was blocked by addition of antagonists of α-amino-3-hydroxy-5-methylisoxazole-4-propionate (AMPA)/kainate receptors and voltage-dependent calcium channels. GABA was however absent from HCs in retinas prepared for electrophysiology, even after incubation at 37°C in Ames medium. As this is a prerequisite to determine the physiological role of HC GABA release, retinas were incubated in various conditions with the aim to favor this restoration. GABA was not taken up by HCs, but addition of the GAD cofactor pyridoxal 5′-phosphate (PLP) increased significantly GABA restoration in HC, in the presence of either glutamate or glutamine, at physiological temperature. This indicates that, though they don’t possess membrane GABA transporters, mouse HCs can synthesize GABA and contain GABA in vivo, and therefore appear to be true GABAergic neurons. Assessments of lateral inhibition in the outer retina – or of the global retinal processing – on slices or whole-mount mammalian retinas should thus be performed in the presence of glutamine and PLP, to get closer to in vivo conditions. More generally, the neurotransmitter status of a given tissue prepared for electrophysiology recordings may differ significantly from the one reported by immunohistology performed after intracardiac perfusion.

Materials and methods


We used adult mice (> 8 weeks) from several strains (Balb/cJ, C57Bl/6J, 129SV/Pas, CD1, GAD65-eGFP transgenic mice, n = 5), either purchased from Charles River (L’Arbresle, France) or bred locally in the Institut Clinique de la Souris animal facility. Animals were housed under a 12 h light/dark cycle and had access to food and water ad libitum. Procedures involving animals and their care were conducted in agreement with the French Ministry of Agriculture and the European Community Council Directive no. 86/609/EEC, OJL 358, 18 December 1986. Generation of the transgenic mouse line GAD65_3e/gfp5.5 30, expressing GFP under the control of the GAD65 promoter was previously described (López-Bendito et al. 2004). Briefly, a 6.5 kb segment of the GAD65 gene that includes 5.5 kb of the 5′-upstream region, the first two exons and a portion of the third exon and the introns in between drives the expression of GFP. Expression was confirmed by visualizing GFP fluorescence in vivo, using GFP goggles (BLS Ltd, Budapest, Hungary), through the fluorescence from neonate brains.

Tissue preparation for immunohistochemistry

Depending on experiments, animals were either fixed by intracardiac perfusion (see below) or killed by cervical dislocation. In the latter case, eyes were enucleated, globes were cut along the ora serrata and the anterior part of the eyeball was removed, and remaining vitreous humor separated from the retina. Retina was then detached from the sclera in 0.1 M phosphate buffer saline (PBS), pH 7.4, eventually incubated (see GABA restoration below) and fixed by immersion for 2 h in 4% paraformaldehyde (PFA) in PBS. Retinas were then processed either to 14 μm cryosections, after cryoprotection in 30% sucrose PBS and embedding in Shandon Cryomatrix (Anatomical Pathology International, Runcorn, UK), using CM3000 cryostat (Leica Microsystems GmbH, Wetzlar, Germany), or embedded in 3% agarose and sliced (200 μm) with a VT1000S vibratome (Leica Microsystems GmbH).

Where indicated, tissue fixation was performed by intracardiac perfusion. Animals were anesthetized with an intraperitoneal injection of ketamine (70 mg/kg) and xylazine (5 mg/kg) and then perfused with cold PBS supplemented with 50 μM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) and 100 μM cadmium ions (Cd2+, from CdCl2), then with 4% PFA in cold PBS also with CNQX and Cd2+. The eyes were removed and retinas isolated before post-fixation (2 h in 4% PFA) then included and vibratome sliced.

GABA restoration

For GABA restoration experiments, the retina was dissected out conventionally and then placed for at least 1 h 30 min at 37°C in an incubating medium, constantly bubbled with a 95% O2/5% CO2 mixture, both for oxygenation and pH buffering. Depending on the experiment, 200 μm thick vibratome-slices of retina were cut prior to the incubation, and in other cases whole-mount retinas were incubated before inclusion and slicing. Incubating media are composed of artificial CSF (aCSF) with (in mM) NaCl 126, KCl 2.5, MgCl2 1.2, NaH2PO4 1.2, NaHCO3 18, CaCl2 2.4, Glucose 11, supplemented with PLP (150 μM) and glutamine (1 mM) or glutamate (1 mM). We also used as a comparison Ames’s medium (Sigma-Aldrich, St-Louis, MO, USA) supplemented with glutamate (1 mM) and CNQX (50–250 μM) or GABA (1 mM) and bicuculline (100 μM). After incubation, the tissue was transferred to a glutamine/glutamate-free medium (for 30 min to 1 h 30 min) and then fixed by immersion in 4% PFA. Room temperature was 22°C.


All antibodies used in this study have been previously characterized, or have staining patterns that match those previously reported in the mammalian retina. All details regarding their characterization, validation and use can be found under the experimental methods section in Appendix S1 and in Table S1.

We used mainly the indirect fluorescence method to perform immunohistochemical labeling of the retina. Retinal slices were blocked for 2 h in a solution containing 10% normal goat serum (NGS) and 0.5% Triton-X-100 in PBS. Primary antibodies were diluted in 5% NGS and 0.5% Triton-X-100 in PBS and applied overnight at 4°C. After washing in PBS, secondary antibodies [goat anti-mouse or goat anti-rabbit Fab’2 coupled to Cy3 (Jackson Laboratories, West Grove, PA, USA) or Alexa 488 (Molecular Probes, Eugene, OR, USA), diluted 1/500 in 0.5% Triton-X-100 and 4′,6-diamidino-2-phenylindole (DAPI) 1 μg/mL in PBS] were applied for 2 h at 22°C. For both primary and secondary antibody incubations, we made double-labeling with incubation solutions containing two antibodies. Fluorescence pictures were acquired with a Leica TCS SP2 confocal microscope with a 63× oil immersion objective. Z-projections of stacks were made using the ImageJ software (NIH, In all immunohistochemistry figures except Fig. 1(b), the 4′,6-diamidino-2-phenylindole (DAPI) staining (blue) was added to the merge images, to facilitate the identification of the retinal layers.

Figure 1.

 GFP expression is driven by the GAD65 promoter in a subset of mouse horizontal cells. Confocal images from cryosections of retinas from transgenic GAD65-eGFP mice on a C57Bl/6J background, immunostained for calbindin (center panels and red in merged images, right). The GFP signal is shown in the left panels and in green in the merged images. (a) General view of the retinal layers. (b) Detailed view of the horizontal cell processes. ONL, outer nuclear layer; OPL, outer plexiform layer; INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer. Scale bars: 40 μm (a) and 10 μm (b).

For GAD65 and GAD67 immunostaining, we used in addition a peroxidase/3,3′-diaminobenzidine (DAB) amplification step. Retinal slices were immersed into 3% H2O2 and 50% ethanol for 15 min at 22°C to quench endogenous peroxidases. After washing in PBS, slices were blocked for 1 h in PBS containing 5% NGS and 0.5% Triton-X-100. Primary antibodies were diluted in 0.5% Triton-X-100 in PBS and applied overnight at 22°C. A biotinylated secondary antibody (goat anti-mouse, BA-9200, Vector Labs, Burlingame, CA, USA), diluted 1/500 in 0.5% Triton-X-100, was applied for 2 h at 22°C. Slices were incubated for 2 h in VECTASTAIN® Elite® ABC Reagent (PK 6100, Vector Labs), washed in PBS, then incubated in a DAB solution (D4168-50SET, Sigma-Aldrich).

In situ hybridization

Digoxigenin-labeled riboprobes, generated by in vitro transcription (Dig-UTP from Roche, Mannheim, Germany) from DNA templates using T7 (antisense) and SP6 (sense)-bacteriophage RNA Polymerases (Promega, Madison, WI, USA), were used to detect GAD65 and GAD67 mRNAs. For GAD67, DNA template was a 634 bp cDNA fragment of mouse GAD1 (GenBank NM_0080077), gift of F. Guillemot, subcloned into a pGEM-T Easy vector. The templates for transcribing RNA were made by linearizing recombinant plasmids. For GAD65, DNA templates were kindly provided by the Genepaint organization. They were synthesized by RT-PCR and the initial PCR products were gel-purified and sequence–verified before the transcription (see detailed protocol and information about templates at

Hybridization and detection

Cryosections (12 μM) of non-fixed retina were collected on RNAse treated Super-Frost Plus Slides (Fisher Scientific, Illkirch, France). Fixation, carried out in a Leica Autosteiner XL, was in 4% paraformaldehyde for 20 min at 22°C. Slides were then washed for 5 min in PBS, acetylated and dried by passing them in successive ethanol solutions (70%, 90% and 100%). The hybridization steps were performed using a Tecan Genesis Freedom evo liquid handling platform (Tecan Deutschland Gmbh, Crailsheim, Germany). Briefly, slides carrying retina sections were integrated into flow-through chambers positioned into the Tecan Platform. After a first step of proteinase K digestion (0.01 μg/mL in 50 mM Tris, 5 mM EDTA, 0.05% Tween-20, pH 8.0) in PBS, sections were pre-hybridized for 30 min in a Hyb-mix solution (Ambion, Austin, TX, USA: B8807G). Sections were hybridized for 5 h 30 min in a humid chamber at 64°C in a solution consisting of the pre-hybridization solution with the addition of the digoxigenin-labeled RNA probe at a concentration of 300–600 ng/mL. Sections were processed for the immunodetection of the digoxigenin label using an anti-digoxigenin antibody coupled to peroxidase (Roche, Penzberg, Germany) and revealed with a tyramide-biotin amplification reaction. After the color reaction, sections were washed with blocking solutions and cover-slipped with hydro-matrix (Micro-Tech Lab, Graz, Austria).


GAD65 promoter drives GFP expression in mouse horizontal cells

Transgenic mice expressing fluorescent proteins under the control of cell-type specific promoters are useful tools to identify cell populations in living tissues, facilitating their studies using electrophysiological or imaging techniques. Though HCs have been reported to be GABA immunonegative in the adult mouse retina (Haverkamp and Wässle 2000; Schubert et al. 2010), we investigated the eventual GFP presence in mouse HCs in a transgenic GAD65-eGFP mouse line (López-Bendito et al. 2004). On retinal slices, a dense GFP signal was present in the inner half of the inner nuclear layer (INL) (Fig. 1a), corresponding to amacrine cells and probably some OFF bipolar cells (Kao et al. 2004), as there was no colocalisation with Goα, a specific marker of ON bipolar cells (Figure S1). This pattern was similar to the one reported in a GAD67-eGFP knock-in mouse line (May et al. 2008). In addition, some GFP expressing cells were also present in the outer half of the INL, next to the outer plexiform layer (OPL), with a dense arborization in this synaptic layer, with invaginating processes typical of HCs. Identification was confirmed by co-labeling with the HC marker calbindin (Fig. 1a and b). Colocalization of GFP and calbindin was not systematic, indicating that GFP was expressed in only a subset of HCs. We did not quantify precisely this population, but it was close to 10% of all HCs, with a tendency to have more GFP positive cells in the periphery than toward the center of the retina.

Horizontal cells express GAD65 and GAD67

As transgene expression may be ectopic, we checked whether HCs were expressing GABA synthesizing enzymes, using in situ hybridization with tyramide amplification to visualize low levels of mRNA. For short durations (≤ 45 min), GAD65 and GAD67 mRNA were detected only in cells in the inner half of the INL, as well as in a few cells in the ganglion cell layer (data not shown). This corresponds to amacrine and displaced amacrine cells, well-described populations of GABAergic neurons in the mouse retina (Haverkamp and Wässle 2000). For longer durations (1–2 h), GAD65 and GAD67 mRNAs were detected additionally in a few cells in the external part of the INL, in the two different mice strains tested, Balb/cJ (Fig. 2a and b) and C57Bl/6J (data not shown), with a pattern similar to the distribution of HCs.

Figure 2.

 GAD expression in adult mouse horizontal cells. GAD65 (a) and GAD67 (b) mRNAs expression on retinal slices from Balb/cJ mice detected by in situ hybridization. While most of the signal was located in the proximal part of the INL, corresponding to amacrine cells, and to the GCL, a discrete labeling (black arrowheads) was present in the distal part of the INL, with a distribution corresponding to the repartition of horizontal cells. (c) A pan-antibody recognizing both GAD65 and GAD67 (green) labels horizontal cells, identified by calbindin immunoreactivity (red), mostly in their perinuclear region (white arrowhead). A similar pattern is observed in amacrine cells (white arrows), which processes are in addition heavily labeled in the IPL. (d) Percentage of GAD immunoreactive cells reported to calbindin positive cells in the OPL in several mouse strains. Almost 70% of HCs were labeled with GAD antibody in all four strains (88.9% for Balb/cJ mice, see text for details). Error bars represent SEM. Scale bars = 20 μm.

To verify if GAD65 and/or GAD67 could also be detected at the protein level in HCs, we performed immunohistochemistry with an antibody recognizing the two isoforms, on both the C57Bl/6J and Balb/cJ mice, as well as on two additional strains, the 129SV/Pas and CD1, all these strains being commonly used in physiological experiments and/or in the construction of genetically modified mice. In all four strains, GAD65/67 immunostaining was stronger in the inner plexiform layer (IPL) as well as around the soma of cells in the proximal part of the INL and, to a lesser extent, in the ganglion cell layer (Fig. 2c for Balb/cJ, and Figure S2 for the other strains). HCs, identified with an anti-calbindin antibody, were also labeled with the GAD65/67 antibody, the staining being mainly detected in their perinuclear region (Fig. 2c for Balb/cJ, and Figure S2 for the other strains), probably corresponding to the Golgi apparatus, as previously reported in neurons (Oertel et al. 1981), and visible also in amacrine cells. Based on the perinuclear labeling, at least 70% of calbindin-positive cells were also immunoreactive for GAD65/67, depending on the strain considered: C57Bl/6J: 70.0 ± 1.2% (one mouse, 27 slices, 1330 out of 1900 calbindin positive cells); Balb/cJ: 88.9 ± 0.9% (two mice, 36 slices, 1762 out of 1982 calbindin positive cells); CD1: 70.5 ± 2.1% (three mice, 15 slices, 642 out of 910 calbindin positive cells); 129 SV/Pas: 72.1 ± 1.5% (one mouse, 18 slices, 841 out of 1167 calbindin positive cells) (Fig. 2d). In addition, we did not see any obvious difference in the percentage of HCs immunoreactivity for GAD65/67 between the central and peripheral region of the retina.

Horizontal cells contain GABA

The presence of GABA synthesizing enzymes in adult mouse HCs raised the question of whether GABA could be present in these cells, despite previous reports of its absence (Haverkamp and Wässle 2000). A possible origin of this apparent discrepancy is the fact that most of retinal immunohistochemistry is classically performed on tissue fixed postmortem. It was reported that in the zebrafish, HC GABA content was lost rapidly upon depolarization (Kalloniatis et al. 1996); this could happen similarly in the mouse, during the ischemic period of enucleation and anterior segment dissection, prior to fixation. To avoid this phenomenon and thus to be closer to in vivo conditions, we performed immunohistochemical analysis on tissue collected after intracardiac perfusion of the animal with 4% PFA. Immunolabeling for calbindin and GABA of retinas fixed according to this protocol was performed on 16 animals (Balb/cJ: n = 7; C57Bl/6J: n = 3; 129SV/Pas: n = 3; CD1: n = 3). Results are shown in Fig. 3 for C57Bl/6J and Balb/cJ mice, and in Figure S3 for CD1 and 129SV/Pas. As expected, amacrine cells and the sublaminae in the INL were strongly immunopositive for GABA. While HCs were not as strongly labeled, GABA could readily be seen in both HC somas and processes, especially in the Balb/cJ mice (Fig. 3b), in which 78.2 ± 2.8% HCs were GABA immunopositive (558 out of 713 calbindin positive cells, on 12 slices from three mice). For C57Bl/6J, only 34.4 ± 2.7% HCs contained GABA (77 out of 223 calbindin positive cells, on three slices from a single mouse). GABA immunostainings in 129SV/Pas and CD1, though not quantified, were similar to what was observed in C57Bl/6J.

Figure 3.

 GABA is detected in horizontal cells after fixation by intracardiac perfusion in both C57Bl/6J (a) and Balb/cJ (b) mouse retinas. While most of the GABA immunoreactivity (left panels, and green in merged right images) was in the proximal part of the INL, in the IPL and in the GCL, horizontal cells (arrowheads), identified by calbindin immunoreactivity (center panels, and red in merged right images), were also stained, to a higher degree in Balb/cJ than in C57Bl/6J retinas. Scale bars: 50 μm.

To favor further GABA detection in HCs in Balb/cJ retina, we tried to limit HC depolarization and GABA vesicular release during tissue preparation, by supplementing the fixative with 100 μM Cd2+, an inhibitor of voltage-dependent Ca2+ channels, and 50 μM CNQX, an inhibitor of AMPA/kainate glutamate receptors to limit HC depolarization. This resulted in a greater number of GABA-immunopositive HCs, with staining both in somas and processes (Fig. 4a). GABA was present homogeneously all along the OPL, to a level closer to the one in amacrine cells (n = 4). In addition, some Müller glial cells were also labeled, though not homogeneously along the retina. The high ratio of GABA-containing HCs can be appreciated on a flatmount retina. On the 750 μm × 750 μm zone of intermediate retina (at mid distance between the optic nerve head and the retinal periphery) shown in Fig. 4(b), 602 HCs were GABA immunopositive out of 678.

Figure 4.

 Inhibiting vesicular release increase GABA immunoreactivity in horizontal cells. Confocal images from a vibratome slice (a) or flatmount retina (b) from a Balb/cJ mouse, immunolabeled for GABA (top, and green in merged bottom images) and calbindin (center, and red in merged bottom images), after intracardiac perfusion of 4% paraformaldehyde in PBS, supplemented with 50 μM CNQX, to hyperpolarize HCs, and 100 μM CdCl2 to block voltage-dependent calcium channels and thus vesicular release. Scale bars: 100 μm in (a), 200 μm in (b).

GABA content of horizontal cells is volatile, and lost in electrophysiological conditions

These results supported the idea that GABA is rapidly lost from HCs in the minutes after animals were killed. As the retina isolation for electrophysiology is longer than for immunohistochemistry, this suggested that GABA may be absent from HCs in recording conditions (whole mount retinas or retinal slices). We thus checked if GABA was present in flatmount retina, either right after preparation or after 1 h recovery at 37°C in classical electrophysiology solutions, aCSF or Ames medium (Ames and Nesbett 1981), supplemented with 50 μM CNQX to favor horizontal cell polarization, and thus limit their GABA release. Figure 5 represents an immunolabeling performed on a Balb/cJ mouse retina after dissection and 1-h incubation in Ames medium at 37°C. In these conditions, no GABA was detected in HCs, while amacrine cells and IPL were still strongly immunolabeled with GABA, as well as radial processes likely to be Müller glial cells (n = 3). Similar results were observed with incubation in aCSF at 22°C (n = 3) or at 37°C (n = 4) (data not shown).

Figure 5.

 Horizontal cells do not contain GABA in recording conditions. Confocal images of thick (200 μm) vertical slices from a Balb/cJ mouse retina, immunolabeled for GABA (left, and green in merged right images) and calbindin (center, and red in merged right images). After dissection and isolation, the retina was incubated for 1 h at 37°C in Ames medium, then fixed by immersion in PBS + 4% PFA and sliced. Arrowheads in the left panel indicate the positions of the calbindin positive somas from the center panel. Scale bar = 20 μm.

HC do not have a functional GABA uptake system

While the Ames medium contains many amino-acids, GABA is absent from its composition, which could hinder a restoration of GABA in HC after retinal isolation if uptake is a major source. To assess this hypothesis, we supplemented the incubation medium, either aCSF at 22°C or Ames at 37°C with 1 mM GABA. To limit HC depolarization, and thus limit vesicular release and increase a potential electrogenic GABA uptake, the medium also contained 50 μM CNQX to block AMPA/kainate receptors, and 100 μM bicuculline methiodide to block GABAA receptors, as GABA, though classically an inhibitory transmitter, has been reported to depolarize mammalian HCs (Blanco et al. 1996; Varela et al. 2005b). Following 1-h incubation at 22°C in supplemented aCSF (Figure S4a, n = 3) or at 37°C in supplemented Ames medium (n = 2), HCs were not immunopositive for GABA (Fig. 6a, Balb/cJ retina), while amacrine and Müller cells were strongly labeled, as expected as they express the GABA transporters GAT-1 and/or GAT-3 (Johnson et al. 1996).

Figure 6.

 Horizontal cells can synthetize GABA from glutamate. Confocal images from vertical vibratome sections (50 μm) from Balb/cJ mouse retinas immunolabeled for GABA (left, and green in merged right images) and calbindin (center, and red in merged right images). (a) When the isolated retina was incubated after dissection in Ames medium supplemented with 1 mM GABA, 100 μM bicuculline methiodide and 50 μM CNQX for 1 h at 37°C, the proximal retina and Müller cell soma (arrowheads) and processes (arrows) were immunoreactive for GABA, but HC somas and processes were not labeled, suggesting that there is no re-uptake of GABA in HCs. The outer limiting membrane (OLM), formed by Müller cell processes, was also strongly GABA immunoreactive. (b) When the incubating medium was aCSF + 1 mM glutamate + 50 μM CNQX, a faint GABA immunoreactivity was seen in HCs. This immunoreactivity was increased with the addition of 150 μM pyridoxal 5′-phosphate, both when the medium was aCSF (c) or Ames (d). Scale bars = 40 μm.

HC can synthesize GABA from glutamate or glutamine

Another way to refill HCs is GABA neosynthesis from glutamate, as GAD65/67 was detected in these cells. Glutamate can be taken up directly by HCs, which express membrane glutamate transporters (Rauen et al. 1996), or could be synthesized from glutamine if these cells possess a glutamine import system. Both amino-acids are present in the Ames medium, glutamate at 7 μM and glutamine at 500 μM. As this medium was not sufficient to allow for GABA restoration in HC even at 37°C, we added 1 mM glutamate in aCSF, as well as 50 μM CNQX to block the depolarizing effect of the excitatory amino acid. This allowed for a faint GABA immunoreactivity in HCs at 37°C (Fig. 6b, n = 2), but not at 22°C (Figure S4b).

Both GAD65 and GAD67 need PLP as a cofactor, GAD65 being predominantly found in the PLP-unbound inactive apo form (Kaufman et al. 1991), suggesting that the PLP concentration is not saturating in vivo. To favor GABA synthesis, we thus supplemented further the incubation medium with PLP. After 1 h at 37°C, with 1 mM glutamate, 50 μM CNQX and 150 μM PLP, the GABA content of HCs was increased compared to previous conditions, both with aCSF (Fig. 6c, n = 5) or with Ames medium (Fig. 6d, n = 3). As our goal was to develop a protocol to replenish HCs to perform electrophysiology afterwards, we then tried to replace glutamate and CNQX by glutamine (1 mM), as this could be added to the medium continuously, without blocking synaptic transmission. This allowed for a GABA refilling of HCs when added to aCSF at 37°C (Fig. 7a, n = 5). This refilling was stable in time, as GABA immunoreactivity was still detected in HCs even when the incubated retina was placed 1 h 30 min in plain aCSF at 22°C prior to fixation (Fig. 7b, n = 7). As Ames contains 500 μM of glutamine, we supplemented this medium with only PLP. This condition also favored the presence of GABA in HC (Fig. 7c, n = 2). Supplementation of aCSF with PLP alone at 37°C did also increase HC GABA content, though at a lesser degree (n = 3, Figure S5). These results indicates that HCs can synthesize GABA in in vitro conditions provided PLP is added to the incubation medium, while an additional supplementation with glutamate or its precursor glutamine favors further the GABA refilling of HCs.

Figure 7.

 Horizontal cells can synthetize GABA from glutamine in the presence of pyridoxal 5′-phosphate. Confocal images from vertical vibratome sections (50 μm) from Balb/cJ mouse retinas immunolabeled for GABA (left, and green in merged right images) and calbindin (center, and red in merged right images). (a) When the isolated retina was incubated after dissection in aCSF supplemented with 1 mM glutamine and 150 μm PLP for 1 h at 37°C, HC somas and processes were immunoreactive for GABA. (b) The presence of GABA was not transient, as HCs were still GABA immunopositive when incubated vibratome slices (200 μm) were placed for 1 h in glutamine and PLP-free medium prior to fixation. (c) In Ames medium, which contains 500 μm glutamine, adding 150 μm PLP for a 1 h incubation at 37°C was sufficient to restore GABA in HCs. Scale bars: 20 μm in (a), 15 μm in (b), 40 μm in (c).


Horizontal cells contain GABA in vivo

We report here that GABA was not detected in mouse adult HCs following a classical tissue preparation for electrophysiology, in line with the immunohistochemical characterization of the mouse retina (Haverkamp and Wässle 2000; Schubert et al. 2010). However, GABA-immunoreactivity in HCs was increased after intracardiac perfusion fixation, which should maintain the neurotransmitter pool closer to in vivo conditions. Adding pharmacological agents to the solution further improved GABA-immunostaining of HCs, likely by preventing vesicular release. These observations were specific to HCs, as GABAergic amacrine cells were GABA-immunopositive in all experimental conditions, thus ruling out an interaction between the different tissue preparations and the efficacy of the GABA antibody. This suggests that the in vivo GABA HC content is quickly lost after killing of animals.

More generally, these results point to a possible discrepancy between the neurotransmitter pools suggested by immunohistochemistry and those effectively present in the recording conditions. In addition to the ischemic stress during tissue preparation, differences in available metabolites (Rheims et al. 2009) and oxygenation (Hájos and Mody 2009) may bias the physiological results obtained from slices or isolated tissue.

Horizontal cell GABA comes essentially from neosynthesis, not from uptake

Mouse HCs are not able to take up GABA significantly, even at 37°C (Fig. 6a), confirming that they don’t express membrane GABA transporters, as suggested in previous reports (Honda et al. 1995; Johnson et al. 1996; Casini et al. 2006; Guo et al. 2009). This rules out a Ca2+-independent, transporter-mediated GABA release from HCs.

On the other hand, most adult mouse HCs express GAD65 and/or GAD67, though at a much lower level than amacrine cells (Figs 1 and S1). They are able to synthesize GABA when placed in media supplemented with glutamate or glutamine. This synthesis is increased when PLP, a cofactor of both GADs, is added to the incubation medium (Figs 6c,d and 7). GAD67 exists mostly in the PLP-bound active form, to the contrary of GAD65, which is mostly found in the PLP-unbound, inactive form (Kaufman et al. 1991). Another indirect argument in favor of GAD65 over GAD67 is that the GAD65/67 immunolabeling in HCs was concentrated in the cell somas, in a ring-like structure reminiscent of the Golgi apparatus, as well as some small punctas in the OPL (Fig. 2c). Such a distribution is more likely to correspond to GAD65, which is membrane-associated, while GAD67 is usually more widely distributed throughout the neuron (Kaufman et al. 1991; Soghomonian and Martin 1998), as can be seen in amacrine cells, which processes are strongly labeled in the IPL. However, antibodies specific for GAD65 and GAD67 did not allow confirming a higher expression of GAD65 over GAD67. Both antibodies strongly labeled the amacrine cells, but the signal in HCs could not be distinguished from the background, either with immunofluorescence (Figure S6a and b) or following peroxidase/DAB amplification (Figure S6c and d). An alternate source of GABA could be its synthesis from polyamines, as putrescine, spermine or ornithine, some enzymes in the polyamines to GABA pathways being dependent on PLP. However, polyamines do not seem to be present in HCs in the adult rabbit retina (Withrow et al. 2002), and the expression of ornithine decarboxylase, the limiting step in putrescine synthesis, is present only at very low levels in the adult rat retina (Yamasaki et al. 1999). As the presence of glutamate or glutamine in the incubation medium favored the restoration of GABA in HCs in the presence of PLP, neosynthesis from low amounts of GAD65 and/or GAD67 is the most probable source of GABA for HCs.

Differences in GABA HC content in mammals

As retinas are frequently fixed postmortem for immunohistochemistry, the reported absence of GABA from adult HCs in some mammals, notably nocturnal rodents (Agardh et al. 1986; Mosinger et al. 1986; Versaux-Botteri et al. 1989; Vardi and Sterling 1994; Fletcher and Kalloniatis 1997; Johnson and Vardi 1998; Haverkamp and Wässle 2000; Loeliger and Rees 2005; Schubert et al. 2010), may find its origin in the volatility of the neurotransmitter within this neural population. Such volatility was also reported in goldfish HCs, in which GABA stores can be depleted in minutes following depolarization (Kalloniatis et al. 1996). An opposite variation was described in the rabbit, in which GABA was absent from HCs when postmortem fixation was performed rapidly (∼ 1 min) but present when fixation was performed 10–40 min after death (Pow and Crook 1994). It was tentatively explained by an ongoing GABA synthesis, as GAD can function in ischemic conditions, while GABA catabolism by transamination is dependent upon α-ketoglutarate availability, and thus aerobic conditions. Such a difference between mouse and rabbit could be explained by the fact that the former has a retinal circulation, while the rabbit retina is mostly avascular, and thus expected to be affected less rapidly by ischemia. GAD could remain active longer in the rabbit than in the mouse postmortem. Another unassessed source of discrepancy could be a differential GABA transaminase activity.

Physiological consequence of glutamate/glutamine dependence of GABA synthesis

Whether HCs synthetize their GABA physiologically from glutamate or glutamine, the common source should be glutamate release by photoreceptors, either directly captured by HCs via their glutamate transporters (Rauen et al. 1996), or subsequently as glutamine after transformation in Müller cells. As mouse HCs have a low capacity to synthesize GABA, because of a low expression of GADs, this suggests that GABA release could be influenced by the recent local activity. When photoreceptors are depolarized, as when adapted to a given level of illumination, their glutamate release could lead to a higher GABA content in HCs, while bright illumination or flickering stimuli for which photoreceptors are more polarized could lower it. This would provide a local, ‘recent history’ dependent, component to the information processing carried by HCs. Such a local role may depend on the local expressions of GAD65 and GAD67, and thus differ between species, being more prominent in GAD65 expressing zones, as in the macaque (Vardi et al. 1994), the guinea pig (Guo et al. 2010) or in the rabbit visual streak (Johnson and Vardi 1998).

Re-evaluating the role of GABA in the outer retina

The lateral inhibition mediated by HCs was initially thought to be carried by GABA, from work in lower Vertebrates, but inconsistent pharmacology has casted doubts on such a role (reviewed in Kamermans and Spekreijse 1999; Piccolino 1995), including in the macaque retina. This has led to the propositions that HCs transmit their information through changes in the extracellular pH in the monkey (Davenport et al. 2008), as was previously proposed in amphibians (Hirasawa and Kaneko 2003), or through an ephaptic communication involving hemigap junctions in the mouse (Xia and Nawy 2003), as previously proposed in lower Vertebrates (Kamermans et al. 2001; Kamermans and Fahrenfort 2004). This does not preclude, however, that GABA may play a role in lateral inhibition, in addition to those mechanisms. Indeed, with the presence of GABA in HCs, all the required elements for a GABA transmission are expressed in the triadic synapse formed by HCs, bipolar cells and photoreceptors. HCs express the vesicular GABA transporter VIAAT/VGAT in their terminals (Haverkamp et al. 2000; Cueva et al. 2002; Jellali et al. 2002), as well as the proteins required for vesicular release (Hirano et al. 2005, 2007). While ionotropic GABA receptors do not seem to be systematically expressed on mammalian cone terminals, GABAA receptor subunits were found at the dendritic tips of all bipolar cells in various mammalian retinas, including those of humans and macaques, with the strongest staining being in apposition to horizontal cell processes (Haverkamp et al. 2000; Vardi and Sterling 1994; reviewed in Wässle et al. 1998), supporting the hypothesis of a GABAergic feedforward from HC to bipolar cells. A criticism often raised against feed-forward was that such a signaling would modulate similarly the ON and OFF pathways, while a contrast-reinforcing process should exert an opposite action on the two channels. However, the dendrites of ON and OFF bipolar cells were demonstrated to express different chloride transporters (Vardi et al. 2000), allowing for a depolarizing action of GABA on ON dendrites (Varela et al. 2005a; Duebel et al. 2006). GABA could also act as an autocrine regulator as shown in the tiger salamander (Kamermans and Werblin 1992), as HC express GABAA (rabbit, mouse and human, Blanco et al. 1996; Feigenspan and Weiler 2004; Picaud et al. 1998a) and/or GABAB (rat, Koulen et al. 1998) receptors.

The presence of all these elements on the pre- and post-synaptic sites is of course not a proof of the implication of horizontal cell GABA release in lateral inhibition. Electrophysiological recordings of HCs, their GABA sensitive synaptic partners, and/or ganglion cell surround after incubation in glutamine/PLP containing medium should allow to re-evaluate the possible roles played by GABA in outer retinal processing.


This work was supported through HFSP grant RGY0004/2003 to MJR, ANR GABARET to SP, Association française contre les Myopathies (AFM) fellowship to EW, Université Paris 6 fellowship to SD, CNRS and Inserm. We thank Dr. Heinz Wässle and Nicholas Brecha for their comments on the manuscript, and the Eurexpress platform, especially Muriel Philipps and Violaine Alunni, for their assistance with in situ hybridization experiments. Authors have no conflict of interest.