Transient expression of a GABA receptor subunit during early development is critical for inhibitory synapse maturation and function

SUMMARY

Developing neural circuits, including GABAergic circuits, switch receptor types. But the role of early GABA receptor expression for establishment of functional inhibitory circuits remains unclear. Tracking the development of GABAergic synapses across axon terminals of retinal bipolar cells (BCs) we uncovered a crucial role of early GABA_A receptor expression for the formation and function of presynaptic inhibitory synapses. Specifically, early α3-subunit-containing GABA_A (GABA_Aα3) receptors are a key developmental organizer. Before eye-opening, GABA_Aα3 gives way to GABA_Aα1 at individual BC presynaptic inhibitory synapses. The developmental downregulation of GABA_Aα3 is independent of GABA_Aα1 expression. Importantly, lack of early GABA_Aα3 impairs clustering of GABA_Aα1 and formation of functional GABA_A synapses across mature BC terminals. This impacts the sensitivity of visual responses transmitted through the circuit. Lack of early GABA_Aα3 also perturbs aggregation of LRRTM4, the organizing protein at GABAergic synapses of rod BC terminals, and their arrangement of output ribbon synapses.

eTOC Blurb

Sinha et al. show that GABA_A synapses on axon terminals of retinal bipolar cells alter receptor types before eye-opening. GABA_Aα3 receptors present during early development regulate the assembly and function of these synapses by promoting the clustering of GABA_Aα1 receptors and the organizing protein LRRTM4.

INTRODUCTION

Formation of accurate and efficient synapses relies on a collaborative effort of molecular and activity-dependent processes¹. A common feature of developing neural circuits is a change in pre- and postsynaptic functional properties as circuits mature. For instance, developing inhibitory brainstem circuits switch neurotransmitter types (GABA->Glycine) and developing excitatory NMDA postsynapses switch receptor composition (GluN2B->2A) during maturation^{2, 3}. Although these developmental alterations have been documented for brain regions and excitatory and inhibitory neurons across the CNS^{2, 4-8}, the organizational role of these alterations have been more extensively studied for excitatory than inhibitory circuits.

Inhibitory circuits in the developing CNS are known to switch receptor composition and distinct receptor-types dominate in the immature vs mature CNS^{9, 10}. Of note, α3-subunit containing GABA_A receptors (GABA_Aα3Rs) are widely expressed in the CNS at the time of birth^{6, 11, 12}, with a much restricted expression in the mature CNS^13-15. But the role of this early receptor expression for the formation and function of individual inhibitory synapses remains unclear. Here, we focused on presynaptic inhibitory circuits of the mammalian retina to determine the role of early GABA receptors (GABARs) for the formation of functional inhibitory synapses.

Presynaptic inhibition onto axon terminals of neurons is a common circuit-motif that regulates neurotransmitter release. It is found in the spinal motor neuronal circuit¹⁶, the olfactory glomeruli circuit¹⁷ and the mammalian retinal circuit¹⁸. In the retina, amacrine cell (AC) interneurons provide presynaptic inhibition onto axon terminals of bipolar cells (BCs) to regulate glutamate release from BCs^{19, 20}. This regulation of glutamate release serves several functions, including adjusting the dynamic range of operation of these neurons^{21, 22} and regulating the threshold of retinal visual responses²³. BCs are glutamatergic second-order neurons that relay photoreceptor input from the outer retina to ACs and retinal output neurons (ganglion cells) in the inner retina (Figure 1A). Specific rod and cone BCs transfer dim-light (rod photoreceptor) and bright-light (cone photoreceptor) information to inner retinal neurons. BCs are classified as ‘ON’ if they depolarize to light increments, or ‘OFF’ if they depolarize to light decrements^{18, 24, 25}. Rod BCs (RBCs) are ON-BCs, whereas cone BCs are either ON or OFF-BCs. BCs receive GABAergic presynaptic inhibition at their terminals, with both GABA_A and GABA_CRs mediating inhibition onto ON-BC boutons^{18, 26-29}. Presynaptic inhibition is generated in two ways: (1), an AC receives excitatory input from the same BC that it provides feedback inhibition onto (reciprocal connection; Figure 1A’), and (2), an AC receives excitatory input from a different BC (non-reciprocal inhibition). Both reciprocal and non-reciprocal synapses can be GABAergic and utilize ionotropic GABA_ARs^30-32, composed of 2α-2β-1γ subunits³³. In the retina, specific α-subunit (α1-α3) containing GABA_ARs are localized at distinct non-overlapping postsynapses^{34, 35}.

Figure 1: Retinal BC terminals replace GABA_Aα3 for GABA_Aα1 during development.

(A) Schematic of the vertebrate retina, showing rod and cone photoreceptors (Pr) providing input to rod and cone bipolar cells (BCs). Cone BCs transfer visual signals to output ganglion cells (GC). BC terminals receive inhibitory input from amacrine cells (ACs) at the inner plexiform layer (IPL) of the retina (shaded grey box).

(A’) Glutamate (Glu) release at BC ribbon synapses is modulated by presynaptic inhibition from ACs. Enlarged view of a reciprocal GABAergic AC (rAC) which receives BC input and provides feedback inhibition, mediated by pentameric GABA_ARs composed of α, β and γ subunits.

(B) Single confocal image planes at the level of the IPL where rod BC axons stratify, co-labeled for GABA_Aα3 and GABA_Aα1Rs across development (P:postnatal day).

(C) Expression of GABA_Aα3 and GABA_Aα1Rs (pixels above background shown in white) within terminals (red) of developing rod and T6 BCs.

(D) % GABA_Aα3 and GABA_Aα1 occupancy within developing BC axons. For each time-point n>4 cells and N>3 animals; GABA_Aα1 P12-P30 data from Hoon et al.²⁸. Data plotted as mean±SEM for all Figures.

See also Figure S1 and S7.

Here, we determined the role of early GABA_ARs for the developmental organization and function of GABAergic synapses at ON-BC terminals. We chose the mouse retina for our study due to the wide availability of genetic tools that target specific BC and AC-types and retinal GABA_AR populations. Combining these genetic tools with electrophysiology, high-resolution light and electron microscopy (EM) we identified a GABA_AR rearrangement at developing BC presynaptic inhibitory synapses and uncovered a critical role of early GABA_AR expression for the formation of functional presynaptic inhibitory synapses.

RESULTS

Early GABA_Aα3 receptors are replaced with GABA_Aα1Rs at retinal BC axon terminals

We first determined which GABA_AR-types are expressed at developing BC terminals to identify the receptor-types that provide early presynaptic inhibition. We immunostained for two GABA_AR-types in the developing mouse retina: GABA_Aα3R and GABA_Aα1R, because these two types have been found at adult BC terminals^{26, 28, 36}. In postnatal day 7 (P7) retina, GABA_Aα3R labeling was abundant at the outer portion of the retinal inner plexiform layer (IPL), where ON-BC axons stratify. GABA_Aα1R clusters were, however, hardly noticeable in this layer at this early age (Figure 1B). Just before eye-opening, at P12, GABA_Aα1 clustering increased in this region of the IPL, increasing further by P30. In contrast, GABA_Aα3R labeling in the ON-lamina of the IPL decreased with age (Figure 1B).

The developmental downregulation of retinal GABA_Aα3R expression is also reflected in the promoter accessibility of retinal GABA_ARs. Promoter accessibilities of GABA_Aα3 and GABA_Aα1 were compared using a previous retinal ATAC-seq (Assay for Transposase-Accessible Chromatin using sequencing) dataset³⁷ that spanned different time-points (embryonic day 17.5, P0/day of birth, P3-10, P14/eye-opening, and P21). Whole retina analyses showed high promoter accessibility for GABA_ARs at early time-points with diminished accessibility after eye-opening (Figure S1A). To compare GABA_A promoter accessibility specifically across retinal BCs, we analyzed the Jorstad et al.³⁸ ATAC-seq dataset from adult ON-BCs. This dataset utilized FACS-purified BCs from the adult Grm6-tdTomato transgenic line where ON-BCs selectively express the fluorescent protein tdTomato³⁹. Comparing GABA_Aα3 and GABA_Aα1 promoter accessiblity from adult ON-BCs revealed that whereas the GABA_Aα1 promoter was still accessible in adult ON-BCs, the GABA_Aα3 promoter accessiblity was undetectable (Figure S1A). We also compared GABA_A mRNA levels between P6 vs P50 retinas and found an almost 2.5 fold increase in the mRNA levels of GABA_Aα1 between these time-points, whereas the relative GABA_Aα3 mRNA levels at P50 reduced to about half their P6 level (Figure S1B).

To visualize GABA_Aα3 and GABA_Aα1R clusters specifically on developing BC terminals, we used Grm6-tdTomato mice to visualize individual ON-BCs, and immunolabeled these retinas for GABA_Aα3R and GABA_Aα1R at different ages (Figure 1C). We focused on two ON-BC types: RBCs that convey dim-light signals and Type 6 ON-cone BCs (T6) that stratify in the same plexus as RBCs but convey high-luminance information. We quantified the percent receptor occupancy of each GABA_AR-type relative to axon terminal volume, as a measure of receptor levels. We performed this analysis on P9, P12, P16 and P30 BC terminals (Figure 1C-D). For both RBCs and T6s, GABA_Aα3 levels were more than 10 fold higher at P9 compared to P30 (RBC P9 GABA_Aα3= 6.21±0.65; RBC P30 GABA_Aα3= 0.28±0.07; p-value_9vs30 = 0.000083: T6 P9 GABA_Aα3= 5.67±0.16; T6 P30 GABA_Aα3= 0.48±0.14; p-value_9vs30 = 0.0075). A rapid decline in GABA_Aα3R levels was observed around P12 just before eye-opening (Figure 1D). Conversely, axonal GABA_Aα1R levels of RBCs and T6s increased from P9 until P16 (two days after eye-opening) and thereafter reduced relative to the mature/P30 BC terminal volume (Figure 1D). Together, our observations reveal a developmental downregulation of GABA_Aα3 with a concomitant increase in GABA_Aα1 expression during retinal circuit maturation, at both the mRNA and protein level.

To determine whether or not the developmental transition from GABA_Aα3Rs to GABA_Aα1Rs on BC axons occurs at the same synapse during maturation, we focused on the A17->RBC GABAergic synapse because of the stereotyped arrangement of this connection across species and the availability of transgenic tools to directly visualize this contact. GABAergic A17 ACs provide reciprocal feedback inhibition onto RBC terminals^{23, 40}. Using the Ai9/slc6a5-cre double transgenic line in which A17 ACs express tdTomato (Figure S2A), individual A17s were targeted and filled with the dye, lucifer yellow, at two developmental time-points: P11 (before eye-opening) and >P30 (adult/mature). A17 dye-filled retinas were immunolabeled for GABA_Aα3R or GABA_Aα1R together with the RBC marker, protein kinase C (PKC)⁴¹, enabling identification of GABA_AR-types at A17->RBC synapses (Figure 2A). To quantify GABA_AR levels at A17->RBC synapses, the volume overlap between A17 varicosities and RBC terminals (i.e. synaptic contact volume) was digitally isolated, and thereafter the pixel volume of GABA_AR immunoreactivity relative to the synaptic overlap volume was expressed as ‘percent occupancy’ (Figure 2B). Before eye-opening (P11), A17->RBC synapses were enriched with GABA_Aα3Rs. Mature A17->RBC synapses instead largely contained GABA_Aα1Rs (Figure 2B), demonstrating a developmental alteration in the GABA_AR-type enriched at A17->RBC synapses.

Figure 2: Individual A17-rod BC inhibitory synapses alter their GABA_AR-type before eye-opening.

(A) Top-down view of lucifer yellow (LY/red) filled A17s in adult and P11 retinas co-immunolabeled with PKC (blue) and GABA_Aα3 or GABA_Aα1Rs (yellow). The GABA_A signal above background and within the A17-RBC volume-overlap is depicted in yellow pixels. Higher-magnification views of three regions per A17, and side-view of the A17-fills are provided.

(B) % GABA_A occupancy at A17-rod BC appositions in P11 (top) and mature retina (>P30; bottom plot). n=number of A17s analyzed. Each retina-piece contained a single A17-fill; N=3 animals.

See also Figure S2 and S7.

GABA_Aα3 receptor clustering at BC terminals show a developmental decline even in the absence of GABA_Aα1

We next determined whether elimination of GABA_Aα1R from ON-BCs would favor maintenance of high GABA_Aα3R clusters across adult ON-BC terminals. Crossing the GABA_Aα1 floxed line⁴² with the ON BC-specific Grm6-Cre²⁸ line and the Ai9 reporter line (Ai9/Grm6-Cre/GABA_Aα1 cKO) enabled specific deletion of GABA_Aα1 from ON BCs (GABA_Aα1 conditional knockout or cKO) and visualization of the GABA_Aα1-deficient BCs (Figure S2B). To confirm a lack of GABA_A currents from adult RBCs in the GABA_Aα1cKO, we performed single-cell patch-clamp recordings from RBCs and recorded their response to GABA-puffs. Previous studies using this technique demonstrated that the GABA-evoked response from RBCs is mediated by both GABA_C and GABA_ARs^{19, 28, 36, 43}. We thus measured the RBC response to GABA-puffs at axon terminals before and after application of the GABA_CR antagonist, TPMPA (Figure S3A). The difference between these responses, i.e., the GABA_A-specific response component, was strikingly lower in GABA_Aα1cKO RBCs compared to controls (Figure S3A-B), confirming the absence of GABA_Aα1Rs across adult GABA_Aα1cKO RBC terminals.

We next analyzed GABA_Aα3R clustering on adult RBCs and T6s in the GABA_Aα1cKO. GABA_Aα3R levels across GABA_Aα1cKO RBC and T6s were comparable (Figure 3A) to control BCs of the same type. Quantification of GABA_Aα1R clusters in GABA_Aα1cKO RBCs and T6s confirmed the downregulation of GABA_Aα1Rs across both axons and dendrites of KO BCs compared to control (Figure 3B). Quantification of GABA_Aα3R clusters within GABA_Aα1cKO RBCs and T6s and control BCs of the same type did not reveal any differences (Figure 3C). Of note, GABA_Aα3 immunolabeling is restricted to BC axons (Hoon et al.²⁸ and Figure 3C) with negligible GABA_Aα3R clustering in BC dendrites across genotypes (Figure 3C). Thus, elimination of GABA_Aα1 from ON-BCs does not alter mature levels of GABA_Aα3 at BC terminals, underscoring that increasing GABA_Aα1 expression does not serve as a developmental cue to downregulate GABA_Aα3 at maturing BC terminals. GABA_Aα3R levels within GABA_Aα1cKO RBC terminals were also comparable to controls at P12 (Figure S3C), ruling out compensatory regulation in the GABA_Aα1cKO early in development, and highlighting a GABA_Aα1R-independent regulation of GABA_Aα3R clustering at inhibitory synapses onto ON-BC terminals.

Figure 3: Absence of GABA_Aα1 on BC axons does not impact GABA_Aα3R clustering within adult terminals.

(A) Examples of GABA_Aα3 signal (white pixels represent signal above background) within adult rod BC (top) and T6 (bottom panel: red) terminals in GABA_Aα1cKO-littermate control (Ctrl) retina.

% occupancy of GABA_Aα1 (B) and of GABA_Aα3 (C) within adult α1cKO-Ctrl BC axons and dendrites (top plots: rod BC; bottom plots:T6). N≥4 GABA_Aα1 cKO-Ctrl pairs. For all Figures: Number in parenthesis indicates cells analyzed and p-value listed for two-tailed unpaired T-test.

See also Figure S2-3 and S7.

Early presence of GABA_Aα3 is necessary for accruing GABA_Aα1Rs at BC terminals

Although GABA_Aα3R clustering on ON-BCs does not rely on the presence on GABA_Aα1, the developmental increase in GABA_Aα1R clusters may depend on the early presence of GABA_Aα3. To test this hypothesis, we utilized mice that lack expression of GABA_Aα3 subunits globally⁴⁴. We confirmed that GABA_Aα3R clustering was abolished in the retinal IPL of the GABA_Aα3KO (Figure S4A). We next crossed GABA_Aα3KO mice with Grm6-tdTomato mice to enable visualization of individual RBCs and T6s in the KO background. We first quantified the occupancy of GABA_Aα3 within RBC and T6 terminals in the GABA_Aα3KO (Figure S4B). GABA_Aα3 was significantly diminished across terminals of both BCs in the GABA_Aα3KO compared to control (Figure S4B). RBC and T6 dendrites showed negligible GABA_Aα3 clustering across genotypes (Figure S4B).

We next evaluated GABA_Aα1R clustering in GABA_Aα3KO retinas (Figure 4). Figure 4A shows a retinal volume of an adult GABA_Aα3KO and control, immunolabeled for GABA_Aα1. The outer plexiform layer (OPL) where BC dendrites stratify had comparable GABA_Aα1 levels across genotypes (Figure 4A). However, the IPL where BC terminals stratify, displayed a striking reduction in GABA_Aα1 immunolabeling in the GABA_Aα3KO, compared to control (Figure 4A). The drastic reduction in axonal GABA_Aα1R clustering was confirmed when we quantified GABA_Aα1 occupancy at axon terminals of individual RBCs and T6s (Figure 4B-C; GABA_Aα1 occupancy at BC dendrites was unchanged). Thus, GABA_Aα1R clustering on RBC and T6 terminals requires the early presence of GABA_Aα3. GABA_Aα1R clusters were also significantly downregulated at inhibitory synapses onto another cone BC-type (Type1 OFF cone BC) in the GABA_Aα3KO (Figure S5A). This BC-type laminates at the opposite border of the IPL and the GABA_Aα3KO line was crossed to the Vsx1-cerulean line²⁸ to visualize Type1s in the KO. In contrast, glycinergic inhibitory synapses on Type1 terminals, as evaluated by presence of α1-subunit containing glycine receptors at these synapses²⁸, remained unaltered in the GABA_Aα3KO (Figure S5B). Thus, the role of early GABA_Aα3 in establishing GABA_Aα1R presynaptic inhibitory synapses appears to be conserved across diverse BC-types.

Figure 4: GABA_Aα1R clustering across BC terminals requires early GABA_Aα3 expression.

(A) Volume views of GABA_Aα1 immunofluorescence in the outer and inner plexiform layers (OPL; IPL) of an adult GABA_Aα3KO and littermate control retina.

(B) GABA_Aα1 expression (white pixels) within adult rod (left) and T6 (right) BC axons and dendrites in GABA_Aα3KO-littermate control (Ctrl) retina.

(C) % GABA_Aα1 occupancy within adult GABA_Aα3KO-Ctrl BC axons and dendrites (left: rod BC; right plot:T6). N>5 GABA_Aα3KO-Ctrl pairs.

(D) % occupancy of GABA_Aα1 within P12 GABA_Aα3KO-Ctrl BC axons and dendrites (left: rod BC; right plot:T6). N>4 GABA_Aα3KO-Ctrl pairs.

See also Figures S4-7.

Presynaptic inhibition at RBC and T6 terminals is mediated by both GABA_A and GABA_CRs²⁸. Blocking vesicular release of inhibitory neurotransmitters by elimination of the retinal vesicular inhibitory amino acid transporter (VIAAT) impairs maintenance of both GABA_A and GABA_CRs at RBC and T6 terminals²⁸. We thus immunolabeled GABA_CRs in GABA_Aα3KO retina to determine whether or not early expression of GABA_Aα3 impacts GABA_CR clustering at RBC and T6 axons (Figure S6A). GABA_CR occupancy remained unperturbed in both RBC and T6 terminals (Figure S6A), suggesting that early GABA_Aα3 expression specifically regulates maturational changes at GABA_A synapses. GABA_CR immunoreactivity on dendrites of RBC and T6s was negligible in both GABA_Aα3KO and control (Figure S6A), in keeping with previous findings that GABA_CR are sparse on ON-BC dendrites²⁸.

To test whether early GABA_Aα3 plays a role in the establishment or maintenance of GABA_Aα1Rs across BC terminals, we quantified GABA_Aα1R occupancy across P12 GABA_Aα3KO ON-BC terminals. We observed a substantial reduction of GABA_Aα1R clustering in P12 GABA_Aα3KO RBC and T6 terminals relative to control (Figure 4D). Early GABA_Aα3 expression is thus necessary for the developmental increase of GABA_Aα1Rs on ON-BC terminals, unlike a lack of VIAAT, which impairs maintenance but not the initial accruing of GABA_Aα1Rs across RBC and T6 terminals before eye-opening²⁸. As an important internal control, GABA_Aα1R levels across dendrites of the same BCs were not altered in the GABA_Aα3KO (Figure 4D). We confirmed that GABA_Aα3 expression was abolished across both axons and dendrites of P12 GABA_Aα3KO BCs (Figure S4C), and also determined that GABA_CR levels at ON-BC axons were unchanged in the P12 GABA_Aα3KO (Figure S6B). Together, our observations reveal a critical role for GABA_Aα3 in establishing GABA_Aα1 inhibitory synapses at developing BC axon terminals.

GABA_ARs do not aggregate at BC axons in GABA_Aα3KO and failure of GABA_Aα1 to cluster may be due to impaired receptor trafficking

The lack of GABA_Aα1R clustering on RBC terminals prompted us to evaluate whether other GABA_AR subunits are also affected in the GABA_Aα3KO. GABA_AR pentamers are comprised of 2α-2β-1γ subunits³³, and our previous work has shown that α1-subunits come together with γ2-subunits at GABA_AR synapses across ON-BC terminals²⁸; see also Wässle et al.³⁴ and Greferath et al.⁴⁵. GABA_Aβ2/3 expression is enriched at the outermost lamina of the IPL, where RBC terminals stratify^{34, 45, 46} and specifically mRNA for GABA_Aβ3 is found in the layer where BC somata reside⁴⁵. We immunolabeled for GABA_Aγ2 and GABA_Aβ3-subunits in retinal slices together with the RBC marker, PKC (Figure 5A) and found both receptor subunits to be robustly expressed on wildtype RBC terminals (control panel in Figure 5A). However, in the GABA_Aα3KO, levels of both GABA_Aγ2 and GABA_Aβ3-subunits on adult RBC terminals were drastically reduced compared to controls (Figure 5A-B), suggesting a lack of clustering of all GABA_AR subunits on adult BC terminals in the absence of early GABA_Aα3.

Figure 5: GABA_AR subunits are downregulated in GABA_Aα3KO but GABA_Aα1 total protein expression and promoter accessibility are unchanged in α3KO retina.

(A) Slices from adult GABA_Aα3KO-littermate control retinas co-immunolabeled for PKC (red) and GABA_Aγ2 (left) or GABA_Aβ3 (right panel: yellow). Bottom panels show higher-magnification views of the selected regions (white box) showing receptor immunostaining within PKC-labeled boutons.

(B) % occupancy of GABA_Aγ2 and GABA_Aβ3 within adult rod BC boutons across genotypes (N>4 GABA_Aα3KO-Ctrl pairs; ≥6 sections analyzed per condition).

(C) Western blot showing total protein levels of GABA_Aα1 compared to the reference Actin, of retinas from two pairs of adult GABA_Aα3KO-littermate control animals (KO-WT respectively).

(D) Normalized ATAC-Seq tracks from adult GABA_Aα3KO and littermate control (Ctrl) retinas showing promoter accessibility for the Gabra1 locus. Red box highlights similar Gabra1 promoter accessibility between genotypes.

See also Figures S6 and S7.

Immunolabeling for GABA_Aα1 reveals the amount of receptor clustered at the postsynaptic surface. Receptor clustering depends not only on total receptor protein expression but also on efficient trafficking from the protein synthesis machinery to the postsynaptic membrane. To determine which of these two possibilities account for decreased GABA_Aα1R clustering in the GABA_Aα3KO, we measured the total retinal GABA_Aα1 protein levels by western blot analyses. Total protein levels of GABA_Aα1 were equivalent between GABA_Aα3KO and control retinas (Figure 5C), indicating that GABA_Aα1 net protein levels are unaltered in the GABA_Aα3KO. Although we could not measure protein levels specifically in RBCs, the pan-reduction in GABA_Aα1 clustering in the IPL of the α3KO suggests that it is unlikely that GABA_Aα1 protein levels were differentially affected across cell-types. We also compared GABA_Aα1 mRNA levels in GABA_Aα3KO retinas compared to control and did not observe any change (Figure S6C). As a control, we determined mRNA levels of the RBC specific gene PCP2⁴⁷ and GABA_C in GABA_Aα3KO retinas compared to control (Figure S6D). The mRNA levels for both these genes remained comparable across genotypes (Figure S6C-D). As a final assay on the integrity of the GABA_Aα1 transcriptional machinery in GABA_Aα3KO retinas, we performed ATAC-seq analyses on adult GABA_Aα3KO-control retina pairs. We did not observe any noticeable differences in the GABA_Aα1 promoter accessibility (Gabra1 locus, Figure 5D) in the GABA_Aα3KO retina compared to control. Together, our observations suggest that transcription and translation of retinal GABA_Aα1 is unlikely to be perturbed in the GABA_Aα3KO, raising the possibility that a deficit in the trafficking of GABA_Aα1 to the postsynaptic membrane underlies the lack of GABA_Aα1R clusters on BC axons in the GABA_Aα3KO.

Lack of GABA_Aergic presynaptic inhibition at GABA_Aα3-deficient rod BC terminals alters rod BC output

To determine how the absence of GABA_Aα1R clusters in GABA_Aα3KO retina impacts GABA-evoked currents of RBCs, we performed whole-cell patch-clamp recordings from these BCs in slice preparations of adult retina, and recorded currents in response to GABA-puffs at their axon terminals. GABA-evoked responses in RBCs are mediated by GABA_A and GABA_CRs^{19, 28, 36, 43}, with near equal contributions of both receptor-types to the total evoked current amplitude^{28, 36, 43}. Comparison of evoked responses of RBCs in GABA_Aα3KO-control retinas suggests a ~50% reduction in the total GABA-evoked response in the KO (Figure 6A). Application of a GABA_CR antagonist (TPMPA) isolated the GABA_A-component of the response and revealed that GABA_A-mediated currents were substantially reduced in GABA_Aα3KO RBCs compared to control (Figure 6A). The response remaining in TPMPA was eliminated upon application of the GABA_AR antagonist, GABAzine (Figure 6A). The amplitudes of both the total GABA-evoked current and the GABA_AR-mediated component were significantly reduced in GABA_Aα3KO RBCs compared to control (Figure 6B). These recordings corroborate the lack of GABA_AR clusters on axon terminals of adult GABA_Aα3KO RBCs.

Figure 6: GABA_A-mediated response from GABA_Aα3KO rod BC terminals is attenuated leading to increased AII sensitivity.

(A) Responses from adult rod BCs after GABA-puff application at their terminals. Exemplar trace depicting net GABA-evoked responses from GABA_Aα3KO and littermate control (Control) rod BCs, and responses after application of the GABA_CR antagonist, TPMPA, and GABA_AR antagonist, GABAzine.

(B) Quantification of rod BC current amplitude before and after TPMPA application from adult GABA_Aα3KO-Control retinas. Numbers within the histograms represent recorded cells in each condition.

(C) Exemplar recordings of responses to brief light flashes from an AII in adult GABA_Aα3KO (right) and Control (middle) whole-mount retinas. Thick lines represent the average response (shaded region=SEM) to ten repeats of a dim flash (grey arrow/trace; 0.025 R*/Rod/flash) and a flash that was ~100 times brighter (purple arrow/trace; 3.2 R*/Rod/flash).

(D) Normalized flash responses across adult AIIs for Control (closed markers/Ctrl, N=5 animals) and GABA_Aα3KO (open markers/α3KO, N=4 animals) retinas. Insets: Example responses to a range of flash strengths.

(E) Quantification of the flash strength at which the amplitude of the AII response reaches 50% of its maximum value (R_50%) across adult GABA_Aα3KO-Ctrl.

A lack of GABA_A inhibition at RBC terminals could affect the efficacy of visual information transfer from RBCs. RBC terminals release glutamate at ribbon synaptic sites apposed to two postsynaptic AC partners. One of the AC partners, the AII, is responsible for conveying visual information from RBCs to ganglion cells^{18, 24, 25}. To determine how lack of GABA_A presynaptic inhibition impacts RBC->AII transmission, we performed voltage clamp recordings from AII ACs in GABA_Aα3KO-control retinas in a wholemount preparation and evaluated responses of AII cells across genotypes. We isolated the excitatory RBC-driven synaptic input across the dim-light regime (Figure 6C). Across flash strengths we observed exaggerated responses from GABA_Aα3KO AIIs compared to control (Figure 6C). Normalization of the AII response across the flash strengths probed showed a leftward shift of the AII response amplitude vs flash intensity curve in GABA_Aα3KO compared to control (Figure 6D), indicating an increased light sensitivity of GABA_Aα3KO AIIs (Figure 6D). We quantified the sensitivity of AIIs across genotypes by fitting the stimulus-response data of individual AIIs with a sigmoid and extracting the flash strength at which the response amplitude reached 50% of its maximum value (R_50%). The R_50% value was significantly reduced for GABA_Aα3KO AIIs compared to control (Figure 6E), confirming an increased light sensitivity of GABA_Aα3KO AIIs. Our observations thus reveal an abnormal RBC->AII transmission when GABA_A presynaptic inhibition is impaired in the GABA_Aα3KO.

Synaptic dyad assembly at rod BC terminals is perturbed in the GABA_Aα3KO

Our observations thus far led us to ask whether clustering of LRRTM4, the organizing protein at GABAergic synapses of RBC axons⁴³ could also be affected in GABA_Aα3KOs. We immunolabeled for LRRTM4 in adult GABA_Aα3KO-control retinas (Figure 7A) and found that LRRTM4 occupancy on RBC terminals was significantly reduced by almost half in GABA_Aα3KO compared to control (Figure 7B). As GABAergic presynaptic inhibition onto RBC axons is mediated almost equally by GABA_A and GABA_CRs^{28, 36, 43}, and because GABA_CR occupancy is unchanged in the GABA_Aα3KO (Figure S6A-B), our observations imply that LRRTM4 associated with GABA_A synapses could be selectively altered in GABA_Aα3KO RBC terminals.

Figure 7: LRRTM4 is downregulated in GABA_Aα3KO rod BC terminals with perturbed synaptic dyad arrangements.

(A) LRRTM4 immunofluorescence (white) localized within adult rod BC terminals (red) in a littermate control (Ctrl) and GABA_Aα3KO retina. Insets show side-views.

(B) % occupancy of LRRTM4 within adult GABA_Aα3KO-Ctrl rod BC boutons. N=4 animal pairs.

(C) Single-plane electron micrograph of an adult rod BC terminal in the GABA_Aα3KO retina with two ribbon sites apposed to an A17 and an AII process characteristic of a normal dyad arrangement. The right panel is pseudo-colored for visualization of synaptic elements/partners. A17->RBC inhibitory synapse denoted as Inh Syn.

(D) 3D view of an entire rod BC terminal (gray) from an adult GABA_Aα3KO showing ribbons (yellow), inhibitory sites (red), AII (magenta) and A17 (green) partners at each ribbon. Inset shows magnified view of a ribbon synapse with dyad components.

(E-E”) Single-plane electron micrograph of an adult GABA_Aα3KO rod BC with ribbons mis-localized opposite a single AII process or present at a triadic arrangement with an AII and two different A17 processes. Bottom panels are pseudo-colored. E” is the 3D rendering of GABA_Aα3KO triad arrangement shown in E’.

(F) Fraction of total rod BC ribbon sites with indicaed postsynaptic arrangements in adult wildtype-GABA_Aα3KOs. (3 wildtype, 3 GABA_Aα3KO reconstructions from 2 animal pairs). For AII/A17 arrangements, the values for GABA_Aα3KO include AII/A17 dyads with lateral/horizontal and vertical ribbons. AII/A17 dyads with lateral ribbons were never observed in wildtype. Wildtype data from Sinha et al.⁴³.

See also Figure S6.

Presynaptic inhibition at RBC terminals regulates assembly of output synapses at these terminals which occur at specialized ribbon sites apposed to two postsynaptic ‘dyad’ partners⁴³. Ribbon synapses are sites of glutamate release stereotypically organized such that a single RBC ribbon is apposed to an A17 and AII AC process with the A17 partner providing reciprocal inhibition (Figure 7C). Both loss of inhibitory neurotransmission and LRRTM4 expression were previously found to disrupt the assembly of the RBC dyad, seen as a reduction in the number of ribbons correctly apposed to one AII and one A17 process⁴³. We thus performed serial blockface scanning EM and reconstructed RBC terminals in GABA_Aα3KO retina to determine the ribbon synapse arrangements of adult GABA_Aα3KO RBC terminals (Figure 7D). On average, we observed 53.67±4.26 ribbons in the GABA_Aα3KO RBC terminals (n=3) we reconstructed, which was similar to the ribbon numbers in wildtype RBC terminals (50.0±2.08; p-value = 0.48). We then determined the postsynaptic partners at each ribbon site and found several RBC ribbons in the KO erroneously localized across a single AII AC or localized at a three-partner ‘triad’ junction (Figure 7E-E”); such erroneous contacts were not observed in wildtype. These anomalies led to a significant reduction in the number of GABA_Aα3KO RBC terminal ribbons correctly apposed across a pair of AII-A17 partners (Figure 7F; correct ribbon assembly rate reduced to 76±3% in the α3KO compared to 91±1% in wildtype). This impairment was less severe than that produced by reduction of both GABA_A and GABA_C-mediated inhibition across RBC terminals (e.g. in the LRRTM4KO where the correct dyad assembly is reduced to ~52%⁴³). These observations suggest a ‘dose-dependent’ effect of GABAergic inhibition in influencing RBC dyad assembly. The lack of GABA_A-mediated inhibition in GABA_Aα3KO, however, did not impair the formation of inhibitory synapses across RBC terminals because the number of both reciprocal (A17-mediated) and non-reciprocal inhibitory synapses in the KO was comparable to wildtype (reciprocal synapses at wildtype terminals = 34.00±1.15; reciprocal synapses at GABA_Aα3KO terminals = 31.33±1.33; p-value = 0.21; nonreciprocal synapses across wildtype terminals = 42.33±1.20; nonreciprocal synapses across GABA_Aα3KO terminals = 43.00±1.53; p-value = 0.75; n= 3 BCs per genotype).

We also quantified inhibitory synapse number across P12 RBC terminals in GABA_Aα3KO-control retinas by determining the number of VIAAT-immunoreactive boutons that were spatially apposed to P12 RBC terminals (Figure S6E). We did not observe any difference across genotypes. We verified that the number of inhibitory synapses obtained through our analyses of VIAAT-immunolabeled appositions in wildtype retina (44.13±0.696; n=15 RBC terminals, 7 animals) are comparable to the total number of inhibitory synapses at RBC axonal arbors (reciprocal + non-reciprocal), determined by a serial EM dataset from P12 wildtype retina (45±2; n= 2 RBC terminals reconstructed; P12-EM dataset from Sinha et al.⁴³). Thus, the total number of inhibitory synapses onto a RBC terminal is not altered in the GABA_Aα3KO.

DISCUSSION

Role of early GABARs in the development of inhibitory circuits

Our study unveiled an essential role of early GABA_Aα3 for inhibitory synapse formation and function at axo-axonic synapses. GABA_AR clustering was reduced across BC axons in the GABA_Aα3KO (Figure S7); clustering of several GABA_AR subunits, GABA_Aα1, GABA_Aβ3 and GABA_Aγ2, were also diminished. LRRTM4, an organizing protein at GABAergic synapses of RBC axons⁴³, was also significantly reduced in GABA_Aα3KOs. These changes are specific to GABA_AR synapses; GABA_CR synapses adjacent to GABA_AR synapses on BC axons²³ were not affected by a loss of GABA_Aα3. The absence of GABA_C upregulation in GABA_Aα3KO BCs also underscores a lack of engagement of homeostatic mechanisms to compensate for the loss of GABA_Aα1 and reduced inhibition. This contrasts with findings in retinas with impaired GABA synthesis where RBCs homeostatically adjust their output³⁶. The absence of compensation in GABA_Aα3KO BC terminals is similar to observations from other brain regions where a loss of GABA_Aα3 does not cause an upregulation of other GABA_AR subunits⁴⁸. Our current observations also suggest that early GABA_Aα3Rs have a separate role in GABAergic synapse development, distinct from that of inhibitory neurotransmission, which influences clustering of both GABA_A and GABA_CRs at ON-BC terminals²⁸. In somatic and dendritic synapses of the thalamic reticular nucleus, GABA_Aγ2 subunit expression is disrupted in the absence of GABA_Aα3 and leads to the mis-localization of the inhibitory scaffolding protein, gephyrin, to non-synaptic sites⁴⁸. Thus, early GABA_Aα3 expression may have common organizational role(s) for inhibitory synapse development at both axons and dendrites and across diverse CNS circuits.

Our estimates of total GABA_Aα1 protein expression, mRNA levels and promoter accessibility did not yield significant differences between GABA_Aα3KO-control retinas. The reduced clustering of GABA_Aα1 at BC axons may therefore be due to perturbed trafficking of GABA_Aα1 to postsynaptic sites when GABA_Aα3 is not expressed earlier in development. However, future experiments assessing protein levels specifically from BCs will be needed to test this hypothesis. Whether the physical presence of GABA_Aα3 is needed to guide GABA_Aα1 and the scaffolding protein LRRTM4 to GABA_A postsynapses of BC axons, or whether early GABA_Aα3-mediated transmission is key for recruiting GABA_Aα1 and LRRTM4 to these synapses is unknown. But our previous observations from retinas deficient in inhibitory neurotransmission (VIAAT or Glutamic acid decarboxylase KO) may help distinguish between these possibilities. We had found that inhibitory neurotransmission regulates the maintenance but not the initial accumulation of GABA_Aα1Rs on BC terminals^{28, 36}. Because GABA_Aα3 is critical for the initial accruing of GABA_Aα1Rs at BC axons, it is likely that GABA_Aα3 regulates the formation of inhibitory synapses onto BC axons through an activity-independent mechanism.

We demonstrate that early GABA_Aα3Rs are critical for the functional development of GABA_A synapses at RBC terminals. The absence of this receptor-type during development leads to hypersensitivity of AII AC visual responses, and perturbed RBC->AII transmission. The deficiency of GABA_A presynaptic inhibition at GABA_Aα3KO terminals could underlie the hypersensitivity of AII visual responses. Furthermore, the ultrastructural misarrangements of RBC output ribbon synapses could also contribute towards AIIs receiving abnormal RBC-mediated input. The number of RBC ribbons correctly apposed to an A17-AII pair in the GABA_Aα3KO was reduced by ~16%. This is, however, not as severe as the disruption to dyad assembly (~37% reduction in AII/A17 dyads) at RBC axons when both GABA_A and GABA_C-mediated presynaptic inhibition are impaired⁴³. Because GABA_A and GABA_CRs are expressed by RBCs in near equal proportions and contribute equally to the net GABA-evoked responses^{28, 36, 43}, the extent of error in RBC dyad assembly may correlate with the ‘amount’ of presynaptic inhibition onto BC axons during development. Because GABA_Aα3KO RBCs still have GABA_C-mediated inhibition, the ultrastructural mis-arrangements in the assembly of their output synapses may be less severe.

Taken together, our observations underscore a critical role of early GABA_Aα3 for the formation and function of inhibitory synapses at retinal BC axons. Given the extensive expression of GABA_Aα3 across developing brain regions, our findings raise the possibility that this receptor-type may have a more widespread developmental role in shaping inhibitory circuits than previously appreciated.

Receptor composition changes at synapses during development

Subunit composition of receptors at the postsynapse is a key determinant of response type and response kinetics. Thus, receptor composition has a direct functional correlate. It is well known that glutamate receptor composition in the CNS changes with circuit maturation. The GluN2B->GluN2A receptor subunit switch results in faster kinetics of NMDA receptor-mediated responses³. GluN2A-containing receptors exhibit faster rising and decaying currents compared to GluN2B-containing receptors^{3, 7, 49}. The GluN2B->2A switch has also been suggested to support the acquisition or enhancement of learning capabilities^{3, 50}, with the shift from GluN2B->2A controlling the threshold for modifying synaptic strength⁵¹.

Developing GABAergic neurons have similarly been found to change their GABAR composition to support faster inhibitory responses with maturation. The α–subunit of the GABA_AR is an important determinant of response kinetics⁶. In the thalamic reticular nucleus, GABA_Aα5-containing receptors mediate prolonged early postnatal tonic inhibitory current and a switch to GABA_Aα3 allows shortening of inhibitory postsynaptic currents and generation of faster rhythmic oscillations⁵. This receptor-type switch enables facilitation of the spontaneous network activity in the thalamic reticular nucleus, which is important for supporting increased external environment awareness of the organism⁵. GABA_AR subunit changes also occur in the rat visual cortex during development and have been correlated with an accelerated decay of spontaneous inhibitory currents⁵². Like these past findings, our current observations in the retina also reveal a change in the expression of GABA_Aα subunits, transitioning from GABA_Aα3 to GABA_Aα1, during retinal circuit maturation. However, our findings further demonstrate that such GABA_Aα-subunit changes at inhibitory synapses occur not only on dendrites of CNS neurons but also on their axons. The decline in GABA_Aα3 and increase in GABA_Aα1 at BC axons would ensure faster inhibitory modulation of BC excitatory transmission in the adult. This is because GABA_Aα3Rs are known to confer inhibition with slower response kinetics compared to GABA_Aα1Rs^{6, 42, 53, 54}. This is likely important for both rod and cone BCs because the GABA_Aα3->GABA_Aα1 transition occurs in both BC types. We demonstrate a functional impact when GABA_Aα3, and thus GABA_Aα1, are absent in the rod pathway, but the functional consequences of the GABA_Aα3->GABA_Aα1 dominance on the cone pathway remains to be determined.

Whether or not the change in GABA_Aα-subunit expression is also accompanied by modifications of other GABAergic postsynaptic components at BC terminals remains as yet unknown. Developmental alterations in the expression of both receptor subunits and associated scaffolding proteins are known to occur at glutamatergic postsynapses. NMDA receptors switch receptor-types (GluN2B->GluN2A) together with associated scaffolding proteins at a time coincident with synapse maturation^{3, 50}. SAP102, a scaffold that forms complexes with GluN2B receptors is replaced with PSD95, a scaffold of GluN2A receptors^{3, 55}. We previously found that LRRTM4, a postsynaptic organizing protein, is present at mature RBC GABA_Aα1 synapses⁴³. In the absence of LRRTM4, GABA_Aα1 accumulation at these synapses is much reduced⁴³. In the CNS, LRRTM4 expression is steeply upregulated from the first postnatal week until P30⁵⁶ matching the timeline we observe for GABA_Aα1 expression across RBC terminals. As GABA_Aα3R clustering on ON-BC terminals is highest around the first postnatal week it is unlikely that LRRTM4 is a common organizing protein of GABA_Aα3 and GABA_Aα1-containing BC synapses. Future studies, probably relying on single-BC transcriptomics from early developmental time-points are needed to uncover the identity of organizing protein(s) at early GABA_Aα3 synapses.

We observed that GABA_Aα3->GABA_Aα1R expression transitioned around eye-opening, a correlation that has been observed previously in other CNS regions. The GABA_Aα5-GABA_Aα3 switch in thalamic reticular nucleus occurs around the time of eye-opening⁵ as does the GABA_AR subunit switch in the visual cortex⁵². This common timeline in the expression changes of the α-subunits of GABA_ARs may suggest distinct roles for the early-expressing and late-expressing GABA_AR-types in synapse formation versus plasticity. Indeed, in visual cortical circuits, GABA_Aα1 is pivotal for experience-dependent plasticity but not for circuit development; the reverse is true for GABA_Aα3Rs^{6, 57}. For the retina, however, future experiments are needed to determine whether GABA_Aα1 and GABA_Aα3R-types could play distinct roles in supporting plasticity in the inner retina and to uncover the underlying developmental mechanisms regulating the GABA_Aα3->GABA_Aα1 dominance at inner retinal synapses.

By examining the stereotypically arranged RBC-A17 synapse, we were able to show that GABA_Aα1R clustering increases at this synapse during development as GABA_Aα3 clustering declines. This decline at most RBC-A17 synapses is unlikely to be due to a major loss of early connections, but rather reflects a change in receptor composition at the same synapses. Whether or not receptor subunit composition changes at the same synapse in other parts of the CNS and whether GABAR composition at the mature synapse always depends on the earlier presence of another receptor-type, have yet to be explored in detail. Such studies at the resolution of individual inhibitory synapses would augment our understanding of the process of inhibitory synapse maturation. Our findings here support a mechanism by which already established inhibitory synapses could alter receptor composition during circuit maturation to support functional demands of the emerging circuit.

STAR METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Mrinalini Hoon (mhoon@wisc.edu).

Materials availability

This study did not generate any unique reagents.

Data and code availability

The ATAC-seq dataset from GABA_Aα3KO-littermate control retina has been deposited (GEO: GSE180163). The remaining datasets supporting the current study have not been deposited in a public repository because of extremely large file sizes but are available from the corresponding author on request. This study did not generate a unique code.

EXPERIMENTAL MODELS AND SUBJECT DETAILS

All experiments were performed in accordance with the guidelines of the Institutional Animal Care and Use Committees (IACUC) of the University of Washington and the University of Wisconsin-Madison and the National Institutes of Health. Developing rod BCs and Type 6 ON bipolar cells were visualized in the Grm6-tdTomato mouse line³⁹ in which the metabotropic glutamate receptor-6 (mGluR6) promoter drives tdTomato expression in ON bipolar cells. Age matched littermate control and GABA_Aα3 KO⁴⁴ mice were utilized for analyses. GABA_Aα3 KO-littermate mice were analyzed at the adult time-point (> 6 weeks) and at P12 (before eye-opening). To visualize rod BCs and Type 6 ON bipolar cells in the GABA_Aα3 KO, the GABA_Aα3 KO line was crossed into the Grm6-tdTomato background. To visualize Type 1 OFF cone bipolar cells in the GABA_Aα3 KO, the GABA_Aα3 KO line was crossed into the Vsx1-cerulean mouse line where OFF bipolar cells are specifically labeled by expression of cerulean fluorescent protein²⁸. To target A17 amacrine cells, the Ai9 reporter line (Jackson Laboratory, B6.Cg-Gt(ROSA)26Sor^{tm9(CAG-tdTomato)Hze}/J) was crossed into the slc6a5-cre transgenic (GENSAT) to generate the Ai9/slc6a5-cre line. In this line amacrine cells including A17 amacrine cells are fluorescently labeled and can thus be visualized. The Ai9/slc6a5-cre line was used to target and fill A17 before eye-opening (P11) and at a mature time point (>P30). To eliminate GABA_Aα1 receptors from rod BCs (GABA_Aα1 cKO), the GABA_Aα1 floxed line (B6.129(FVB)-Gabra1^tm1Geh/J) was crossed into the ON bipolar cell specific Grm6-Cre²⁸ line and the Ai9 reporter line. Cre-expressing BCs in a wildtype background served as littermate controls. GABA_Aα1 cKO-littermate mice were analyzed at the adult time-point (> 6 weeks) and at P12 (before eye-opening). Mice of both sexes were utilized. For gene expression experiments P6 and P50 C57BL6/J wildtype animals (Jackson Laboratory) were utilized. 2 adult C57BL6/J wildtype animals were used in the Control data-set for AII electrophysiological experiments.

METHOD DETAILS

Immunohistochemistry

Immunolabeling was performed on whole-mount retinas isolated in cold oxygenated mouse artificial cerebrospinal fluid (mACSF, pH 7.4) containing (in mM): 119 NaCl, 2.5 KCl, 2.5 CaCl2, 1.3 MgCl2, 1 NaH2PO4, 11 glucose, and 20 HEPES. Retinas were flattened on a filter paper (Millipore, HABG01300), fixed for 15 mins in 4% paraformaldehyde prepared in mACSF, rinsed in phosphate buffer (PBS) and incubated with primary antibody in blocking solution containing 5% donkey serum and 0.5 Triton X-100 at 4°C for 3-4 days. Antibodies utilized were as follows: anti-PKC (1:1000, mouse, Sigma); anti-lucifer yellow (1:500, rabbit, Invitrogen), anti-LRRTM4 (BC-262) (1:500, rabbit⁵⁶); anti-GABA_Aα1 (1:5000, guinea pig, J.M Fritschy¹³); anti-GABA_Cρ (1:500, rabbit, R. Enz, H. Wassle and S. Haverkamp³⁴); anti-GABA_Aα3 (1:3000, guinea pig, J.M Fritschy¹³), anti-GABA_Aγ2 (1:1000, rabbit, Synaptic Systems), anti-GABA_Aβ3 (1:500, guinea pig, Synaptic Systems), anti-GlyRα1 (1:500, mouse monoclonal mAb2b, Synaptic Systems), anti-RFP (mouse monoclonal, 1:1000, Abcam), anti-VIAAT (rabbit polyclonal, 1:1000, Synaptic Systems), anti-DsRed (rabbit polyclonal, 1:1000, Clontech). After incubation with primary antibodies, retinas were rinsed in PBS and incubated with anti-isotypic Alexa Fluor (1:1000, Invitrogen) or DyLight (1:1000, Jackson Immunoresearch) secondary antibodies overnight at 4°C. Thereafter retinas were rinsed in PBS and mounted on slides with Vectashield antifade mounting medium (Vector Labs). For generating retina slices, fixed retinas were embedded in agarose (Sigma, low gelling temperature) and sectioned (120 μm) at a Leica Vibratome (VT1000S). The slices were collected in PBS and subsequently processed for immunohistochemistry.

Confocal microscopy and Image analyses

Images were acquired with an Olympus FV 1000 laser scanning confocal microscope and a 1.35 NA 60X oil immersion objective or a Leica SP8 confocal microscope and a 1.4 NA 63X oil immersion objective. Voxel size for acquired images was around 0.05-0.05-0.3 μm (x-y-z). Image stacks were processed using Image J (NIH) and Amira (ThermoFisher Scientific) software. Individual bipolar cell processes were isolated in 3D using the LabelField function in Amira. To isolate the receptor signal within a BC process, the receptor channel was multiplied with the BC mask using the Arithmetic function in Amira. To quantify the amount of receptor expressed within the BC process, we determined the volume occupied by the receptor signal within the BC terminal and expressed it relative to the volume of the BC terminal (% volume occupancy) as previously described^{28, 43, 58}. A threshold was applied to eliminate background pixels and the total volume of receptor signal/pixels above background was thereafter expressed as % occupancy relative to the BC volume^{28, 43, 58}.

To determine contacts between PKC positive rod BCs and A17 amacrine cell varicosities, the A17 amacrine cell varicosities and rod BC axonal boutons were first masked in 3D using the LabelField function in Amira. Using the Arithmetic function, the A17 varicosities mask and the rod BC axon terminal mask were then multiplied to determine the synaptic overlap. The GABA receptor channel was thereafter multiplied with the A17- rod BC synaptic overlap channel to isolate the receptor signal specifically within synaptic overlap. % occupancy of receptor signal within the overlap was determined as described above.

To determine appositions of VIAAT-positive terminals (inhibitory presynapses) across PKC positive rod BC boutons, individual rod BC terminals were first masked in 3D (volume isolation) using the LabelField function in Amira. Thereafter regions of interest were created for each VIAAT positive puncta that had volume overlap (in 3D) with the rod BC mask. The number of such VIAAT positive regions of interest were thereafter summed for each complete rod BC terminal.

Electrophysiology recordings and A17 cell fills

Experiments were conducted on dark-adapted GABA_Aα3 KO and littermate control mice in a whole-mount preparation for AII amacrine cell recording or a slice (200 μm thick) preparation for determining rod BC puff responses from GABA_Aα3 KO-littermate control mice and GABA_Aα1 cKO-littermate control mice. For AII recordings 3 littermate control (7 AII cells) and 2 non-littermate wildtype (C57BL6/J) animals (2 AII cells) were used. For recording light responses from AII amacrine cells, isolated retinas were stored in oxygenated (95% O₂/5% CO₂) bicarbonate Ames medium (Sigma) at ~32°C. Retinas were mounted photoreceptor-side down on poly-L-lysine coated cover slips. The mounted retina was continuously super fused with warm oxygenated Ames (~8mL/min). Retinal dissections and mounting were conducted exclusively under infrared illumination (>900nm) to preserve visual sensitivity.

Voltage-clamp whole-cell recordings from AII amacrine cells were conducted with electrodes (5-6 MΩ) containing (in mM): 105 Cs methanesulfonate, 10 TEA-Cl, 20 HEPES, 10 EGTA, 2 QX-314, 5 Mg-ATP, 0.5 Tris-GTP and 0.1 Alexa (594) hydrazide (~280 mOsm; pH ~7.3 with CsOH). To isolate excitatory synaptic input, cells were held at the estimated reversal potential for inhibitory input (−68.5 mV). Absolute voltage values were corrected for liquid junction potentials (−8.5 mV). Full field illumination (diameter: 500-560 μm) was delivered to the photoreceptors through a customized condenser from short wavelength (peak power at 405 or 460 nm) LEDs. Light intensities (photons/μm²/s) were converted to photoisomerization rates (R*/photoreceptor/s) using the estimated collecting area of rods (0.5 μm²; Field and Rieke⁵⁹, the LED emission spectra and the photoreceptor absorption spectra⁶⁰). Flashes were 10 ms in duration. Electrophysiology example traces in the figures represent the average of ~10 raw responses to the same stimuli. Responses (peak) from individual AII amacrine cells were normalized to the response amplitude at the brightest flash strength (3.2 R*/Rod/flash) and fit with the sigmoid fit function in Igor Pro (Wavemetrics). The flash strength that generates 50% of the maximum AII response amplitude (R₅₀%) was estimated from these fits.

Retinas were embedded in agarose and sliced as previously described^{28, 43} for rod BC recordings. For rod BC puff recordings 5 adult GABA_Aα3 KO animals and 6 littermate control animals were used and 3 adult GABA_Aα1 cKO and 3 littermate control animals were used. Voltage-clamp recordings from rod BCs used pipettes (10–14 MΩ) filled with an intracellular solution containing (in mM) the following: 105 Cs methanesulfonate, 10 TEA-Cl, 20 HEPES, 10 EGTA, 2 QX-314, 5 Mg-ATP, 0.5 Tris-GTP, and 0.1 Alexa-594 hydrazide (~280 mOsm; pH ~7.2 with KOH). To isolate the GABA_A and GABA_C mediated currents, (1,2,5,6-Tetrahydropyridin-4-yl) methylphosphinic acid (TPMPA, 50 μM; Tocris, Bristol, United Kingdom) and GABAzine (20 μM; Sigma) were added to the perfusion solution. To isolate the GABA mediated inhibitory currents cells were held at the estimated reversal potential for excitatory input ~+10 mV. Absolute voltage values were not corrected for liquid junction potentials (−8.5 mV). GABA was applied using a Picospritzer II (General Valve) connected to a patch pipette (resistance, ~5–7 MΩ). GABA (200 μM) was dissolved in Hepes-buffered Ames medium with 0.1 mM Alexa-488 hydrazide and applied with the puff pipette. The puffing direction and the duration of the 50 ms puff were selected such that the GABA puff completely covered the axon of the rod BC being recorded from. Our previous studies using this technique have confirmed that the GABA puff application at rod BC axon terminals only activates GABA receptors specifically at the axon terminal^{28, 43}. To quantify GABA-evoked currents, peak amplitude was calculated by subtracting the pre-stimulus baseline current from the peak response and averaging across several trials.

A17 cells were targeted with 2-photon microscopy (980 nm) in whole mount retinas⁶¹ from the Ai9/slc6a5-cre transgenic line. Two different developmental time points, postnatal day 11 and adult (>4 weeks) were used. A17 cells were injected with 2% Lucifer yellow (LY) prior to fixation with 4% paraformaldehyde (prepared in mACSF) for 15 mins. The retinas were thereafter rinsed with PBS and processed for immunohistochemistry as described above.

3D serial block face serial scanning electron microscopy and rod BC reconstructions

Retinas were immersion fixed in 4% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) and were processed and embedded in Durcupan resin as previously described⁶². A Zeiss 3View Serial block face scanning electron microscope was used to image retinal regions comprising a 2x2 montage of tiles (8600x8600 pixels; ~48μmx48μm) at a section thickness of 50nm. Image stacks were aligned, and rod BCs reconstructed using TrakEM2 module of Image J (NIH). Amira (ThermoFisher Scientific) was used for display of reconstructed cell profiles and associated synaptic profiles.

Rod BC terminals were determined in the EM stack by their characteristic morphology^{18, 24, 43, 63}. Ribbons at rod BC terminals are apposed to two postsynaptic amacrine interneurons, one of which (the A17) makes a reciprocal feedback inhibitory synapse back onto the same rod BC terminal it receives input from^{18, 24, 43, 63}. This feature was used to distinguish A17 profiles at each rod BC ribbon site. The AII partner was recognized through ultrastructural features as previously described^{43, 64}. Electron dense ribbons were easily recognized in EM images allowing demarcation of dyad sites and apposed AII and A17 synaptic partners. Inhibitory synapses were determined by the characteristic accumulation of synaptic vesicles along a defined release site and the presence of a thickening of the pre- and postsynaptic membranes as described previously⁶⁵. 3 GABA_Aα3 KO rod BC terminals were reconstructed from 2 adult animal pairs. A P12 wildtype retina dataset from Sinha et al.⁴³ was used to determine the total inhibitory synapse number across rod P12 BC terminals.

Real Time - Quantitative PCR (qPCR) from retina samples

Retinas were collected from 3 adult GABA_Aα3 KO and littermate control animals. For the developmental comparison, retinas were collected from 3 P6 and P50 C57BL/6J animal pairs. For each animal, both retinas were pooled into one sample. RNA was extracted using DNA/RNA/Protein extraction kit (IBI Scientific). cDNA was reverse-transcribed using M-MLV reverse transcriptase (Promega) and Oligo(dT) Primers (5′–AAGCAGTGGTATCAACGCAGAGTACT30VN-3)⁶⁶. The same concentration of total RNA was used for each sample. Real-Time qPCRs were run in 10 μL fast reactions (PowerUp SYBR™ Green Master Mix, Applied Biosystems) in a Quant Studio 7 Flex machine (Life Technologies). PCR’s were run in pair-matched batches with GAPDH as a reference gene. Relative Gene Expression was calculated using the ΔΔ Ct method with the Pfaffl correction⁶⁷. Primers were used from published literature or designed with the NCBI BLAST tool. New primer pairs were verified by band size and Sanger Sequencing of the PCR product. Product sequences obtained were analyzed with UGENE. Determined primer efficiencies are as follows; GAPDH 1.09, GABA_Aα1 0.91, GABA_Aα3 0.93, GABA_Cρ 1.05, PCP2 1.03. Primer sequences are as follows: GAPDH⁶⁸ Forward: GGCCGGTGCTGAGTATGTCG Reverse: TTCTGGGTGGCAGTGATGGC, GABA_Aα1 Forward: CACCATGAGGTTGACCGTGA Reverse: CTACAACCACTGAACGGGCT, GABA_Aα3 Forward: GTGACACTCGATCTCACAGGT Reverse: ATATCTGGGGCATGCTTGGG, GABA_Cρ Forward: GAGTTTCCCTGGGGATCACG Reverse: GCCATGGCTTGAACAGCATC, PCP 2 Forward: CAGACCTTCTAGACAAGGCAGG Reverse: TCGTTTCTGCATTCCATCCTTG.

Assay for Transposase-Accessible Chromatin (ATAC)-Seq analyses

One pair of GABA_Aα3 KO-littermate control animal was used for the ATAC-Seq experiment. Flash-frozen retinas were thawed by addition of 700μl of lysis buffer (10 mM Tris-HCl, 10 mM NaCl, 3 mM MgCl₂, 0.1% NP-40), homogenized by trituration 10x with a p1000 pipet set to 500 μl, dounced in an ice cold RNase-free 2mL glass dounce 10x with a loose pestle and 10x with tight pestle, and transferred back to original 1.5mL LoBind tube (Eppendorf). Dounce was washed with an additional 700μl ice cold lysis buffer and buffer was transferred to tube with sample for a final volume of 1.4mL. Sample was then centrifuged in a microcentrifuge at 4°C for 10min at 500 rcf. The pellet was re-eluted in 1.4mL ice cold lysis buffer, transferred to a pre-chilled dounce, homogenized again 12x with tight pestle, and transferred back to 1.5mL LoBind tube. Nuclei were counted on a hemocytometer and the volume of lysis buffer/nuclei suspension needed for 20,000 nuclei was aliquoted into a separate 1.5mL LoBind tube. Samples were centrifuged along with tube containing remaining nuclei (for nuclear localized RNA) at 4°C for 5 min at 500g (RCF). The supernatant was removed from samples and remaining nuclei. Subsequent ATAC libraries were generated with 20,000 nuclei per sample using the Nextera DNA library prep kit (FC-121-1030; Illumina, San Diego, CA USA) according to the protocol described in Buenrostro et al.^{69, 70}. ATAC-Seq libraries were prepared and sequenced on the Illumina NextSeq 500 platform with 75 bp single-end reads.

For ATAC-Seq Read processing and alignment. Demultiplexed FASTQ files were trimmed with Trimmomatic (version 0.33) using the parameter SLIDINGWINDOW:5:30⁷¹. Trimmed reads were indexed and aligned to the mouse genome (Mouse GRCm38/mm10 assembly, December 2011) using the Burrows-Wheeler Aligner (bwa) tool⁷² with the parameters: bwa aln -q 0 -t 4 -n 2 -k 2 -l 32 -e −1 -o 0 and bwa samse -n 5. Tag directories of reads were created using HOMER (version 4.6) makeTagDirectory⁷³. Bed files from processed and aligned sequence reads were extended to 200bp and normalized to 10M reads using BEDtools (version 2.23.0)⁷⁴ genomeCoverageBED using the -scale parameter before being converted into bigWig format for display on the UCSC genome browser (https://genome.ucsc.edu/).

Western Blot

Retinae from GABA_Aα3 KO and littermate control mice were homogenized in lysis buffer containing 50mM Tris, 100mM NaCl, 1% Triton X-100, 5mM EDTA, 0.1% SDS, 2.5% glycerol and 1x protease inhibitor cocktail. Both retinas were pooled per animal. Equal amounts of protein samples were run in a 10% SDS gel (Bio-Rad Labs), blotted onto nitrocellulose membranes, incubated with antibodies and visualized by ECL. Primary antibodies utilized were anti-GABA_Aα1 (clone N95/35 from NeuroMab at 1:5000) and anti-Actin (mouse monoclonal 1:5000 Chemicon).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical details of experiments including number of cells (denoted as n) and number of animals (denoted as N) analyzed is provided in the Figure legends. All data are presented as mean ± SEM (standard error of mean) and an unpaired two-tailed T-test was used to determine significance across genotypes.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-PKC clone MC5	Sigma	Catalog # P5704; RRID:AB_477375
Rabbit polyclonal anti-VIAAT	Synaptic Systems	Catalog # 131003; RRID:AB_887869
Rabbit polyclonal anti-GABAAγ2	Synaptic Systems	Catalog # 224003; RRID:AB_2263066
Guinea pig polyclonal anti-GABAAβ3	Synaptic Systems	Catalog # 224 404; RRID:AB_2619936
Rabbit polyclonal anti-Dsred	Clontech/Takara Bio	Takara Bio Cat# 632496; RRID:AB_10013483
Mouse monoclonal anti-RFP	Abcam	Catalog # ab65856
Rabbit polyclonal anti-lucifer yellow	Invitrogen	Catalog # A5750; RRID:AB_2536190
Mouse monoclonal anti-GlyRα1	Synaptic Systems	Catalog # 146111; RRID: AB_887723
Rabbit polyclonal anti-GABAC	34	Generated in Heinz Wässle and Joachim Bormann’s Lab.
Guinea pig polyclonal anti-GABAAα1	13	Generated in Jean-Marc Fritschy’s Lab
Guinea pig polyclonal anti-GABAAα3	13	Generated in Jean-Marc Fritschy’s Lab
Rabbit polyclonal anti-LRRTM4 (BC262)	56	Generated in Ann Marie Craig’s Lab
Mouse anti-GABAAα1	Neuromab	Catalog # 75-136; RRID:AB_2108811
Mouse anti-Actin	Chemicon/Millipore	Catalog # MAB1501; RRID:AB_2223041
Chemicals, peptides, and recombinant proteins
Ames	Sigma	A1420
Lucifer yellow	Sigma	L0259
GABAzine (SR-95531)	Sigma	S106
TPMPA	Tocris	1040
GABA	Sigma	A2129
Alexa 594-hydrazide	Invitrogen	A10438
Alexa 488-hydrazide	Invitrogen	A10436
Vectashield antifade mounting medium	Vector Labs	Catalog# H-1000
M-MLV reverse transcriptase	Promega	M1701
Critical commercial assays
DNA/RNA/Protein extraction kit	IBI Scientific	IB47702
PowerUp SYBR™ Green Master Mix	Applied Biosystems	A25741
Nextera DNA library prep kit	Illumina	FC-121-1030
Deposited data
ATAC-seq α3KO/control retina	This paper	GEO: GSE180163
Experimental models: Organisms/strains
Mouse: GABAAα3 knockout	U. Rudolph44	N/A
Mouse: C57BL/6J	Jackson Labs	JAX Stock No: 000664
Mouse: Grm6-tdtomato	Rachel Wong39	N/A
Mouse: Ai9 reporter; B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J strain	Jackson Labs	JAX Stock No: 007909
Mouse: slc6a5-Cre	Allen Brain Institute (GENSAT)	N/A
Mouse: Grm6-Cre	Rachel Wong28	N/A
Mouse: Vsx1-cerulean	Rachel Wong28	N/A
Mouse: GABAAα1 floxed; B6.129(FVB)-Gabra1tm1Geh/J	Jackson Labs	JAX Stock No: 004318
Oligonucleotides
GAPDH forward: GGCCGGTGCTGAGTATGTCG	68	N/A
GAPDH reverse: TTCTGGGTGGCAGTGATGGC	68	N/A
GABAAα1 Forward: CACCATGAGGTTGACCGTGA	This paper	N/A
GABAAα1 Reverse: CTACAACCACTGAACGGGCT	This paper	N/A
GABAAα3 Forward: GTGACACTCGATCTCACAGGT	This paper	N/A
GABAAα3 Reverse: ATATCTGGGGCATGCTTGGG	This paper	N/A
GABACρ Forward: GAGTTTCCCTGGGGATCACG	This paper	N/A
GABACρ Reverse: GCCATGGCTTGAACAGCATC	This paper	N/A
PCP2 Forward: CAGACCTTCTAGACAAGGCAGG	This paper	N/A
PCP2 Reverse: TCGTTTCTGCATTCCATCCTTG	This paper	N/A
Software and algorithms
BEDtools (version 2.23.0)	74	https://github.com/arq5x/bedtools2
HOMER (version 4.6)	73	http://homer.ucsd.edu/homer/
Trimmomatic (version 0.33)	71	http://www.usadellab.org/cms/?page=trimmomatic
IGOR Pro	WaveMetrics	https://www.wavemetrics.com/
MATLAB	Mathworks	https://ch.mathworks.com/products/matlab
Symphony	Symphony-DAS	https://github.com/symphony-das
ImageJ	NIH	https://imagej.nih.gov/ij/
Amira	ThermoFisher Scientific	https://www.fei.com/software/amira/

HIGHLIGHTS.

1. GABA_A synapses on bipolar cell axons change receptor types before eye-opening.

2. GABA_Aα3 recruits GABA_Aα1 and LRRTM4 to inhibitory synapses on bipolar cell axons.

3. Early GABA_Aα3 is required for functional GABA_A synapses on mature bipolar cell axons.

4. Early GABA_Aα3 expression regulates organization of rod bipolar cell ribbon synapses.