A noncanonical inhibitory circuit dampens behavioral sensitivity to light

See allHide authors and affiliations

Science  01 May 2020:
Vol. 368, Issue 6490, pp. 527-531
DOI: 10.1126/science.aay3152

Retinal ganglion cells that release GABA

Retinal ganglion cells (RGCs) communicate light signals from the retina to the brain and were previously considered to signal exclusively through release of excitatory neurotransmitters. There have been earlier hints of RGCs producing the inhibitory neurotransmitter γ-aminobutyric acid (GABA), but the specific cells have never been identified and their function was entirely unknown. Sonoda et al. found that a subpopulation of intrinsically photosensitive RGCs (ipRGCs) releases GABA (see the Perspective by Ding and Wei). Removal of GABA signaling from ipRGCs led to an increased light sensitivity of the pupillary light reflex and of circadian photoentrainment. GABA release thus moved the dynamic range of these non–image-forming behaviors to bright light intensities. These results explain why these behaviors are so much less sensitive to environmental lighting conditions than conscious visual perception.

Science, this issue p. 527; see also p. 471


Retinal ganglion cells (RGCs) drive diverse, light-evoked behaviors that range from conscious visual perception to subconscious, non–image-forming behaviors. It is thought that RGCs primarily drive these functions through the release of the excitatory neurotransmitter glutamate. We identified a subset of melanopsin-expressing intrinsically photosensitive RGCs (ipRGCs) in mice that release the inhibitory neurotransmitter γ-aminobutyric acid (GABA) at non–image-forming brain targets. GABA release from ipRGCs dampened the sensitivity of both the pupillary light reflex and circadian photoentrainment, thereby shifting the dynamic range of these behaviors to higher light levels. Our results identify an inhibitory RGC population in the retina and provide a circuit-level mechanism that contributes to the relative insensitivity of non–image-forming behaviors at low light levels.

Retinal ganglion cells (RGCs) relay light information to >40 brain areas (13), giving rise to both conscious visual perception (image-forming vision) and subconscious non–image-forming functions, such as circadian photoentrainment and the pupillary light reflex (PLR) (4, 5). RGCs are thought to mediate these functions by the synaptic release of the excitatory neurotransmitter glutamate onto their postsynaptic targets. Previous immunohistochemical evidence has suggested the presence of inhibitory, γ-aminobutyric acid (GABA)–releasing (GABAergic) RGCs in several mammalian species (69). However, the identity of GABAergic RGCs and their function in visual behavior have remained elusive.

To label GABAergic projections from the retina to the brain, we used a mouse line, Gad2-IRES-Cre, in which Cre recombinase is coexpressed with the GABA synthesis enzyme Gad2 (10). We then performed unilateral eye injections of an adeno-associated virus (AAV) that drives Cre-dependent expression of the fluorescent reporter tdTomato (AAV2/hSyn-FLEX-Chrimson-tdTomato) (Fig. 1A). Because RGCs are the only retinal neurons that project to the brain, any labeled tdTomato+ axons in the brain indicate the presence of Gad2-expressing RGCs. We observed tdTomato+ axons that were largely confined to non–image-forming brain areas, including the suprachiasmatic nucleus (SCN), the intergeniculate leaflet (IGL), and the ventral lateral geniculate nucleus (vLGN), which are involved in circadian entrainment, as well as the ipsilateral shell of the olivary pretectal nucleus (OPN), which is involved in the PLR (14 of 14 animals; Fig. 1, B to D) (1115). No labeling was observed in the brains of wild-type animals injected with the same Cre-dependent virus (fig. S1). We observed less-frequent labeling in image-forming structures such as the medial posterior superior colliculus (10 of 14 animals) and projections to the shell of the contralateral dorsal LGN (dLGN) (9 of 14 animals) (fig. S2).

Fig. 1 ipRGCs are a potential source of inhibitory input to non–image-forming visual brain areas.

(A) Intravitreal injections of AAV2/hSyn-FLEX-Chrimson-tdTomato in Gad2-IRES-Cre mice to label GABAergic cells in the retina. Coronal brain sections were made 1 to 2 months after infection. (B to D) tdTomato+ axons were consistently observed (14 of 14 animals) in the IGL (B), vLGN (B), SCN (C), and OPN (D). Scale bars, 200 μm. (E) Intravitreal injections of AAV2/hSyn-DIO-mCherry in Gad2-IRES-Cre mice to label GABAergic cells in the retina. Retinas were immunostained for melanopsin to label ipRGCs. (F) Melanopsin and mCherry labeling in dorsal-temporal (top) and ventral-nasal (bottom) quadrants of retinas in (E). Solid white circles indicate Gad2+ ipRGCs and dotted green circles indicate Gad2− ipRGCs. Scale bars, 30 μm. (G) Percentage of melanopsin immunoreactive cells that were Gad2+. n = 7 to 8 retinas per quadrant. Data are means ± SD. VN, ventral nasal; VT, ventral temporal; DN, dorsal nasal; DT, dorsal temporal.

The melanopsin-expressing intrinsically photosensitive RGCs (ipRGCs) are the major type of RGCs that innervate non–image-forming structures—like the SCN, IGL, vLGN, and OPN shell—and are required for circadian photoentrainment and the PLR (11, 16, 17). To test whether Gad2+ RGCs are ipRGCs, we intravitreally injected Gad2-IRES-Cre mice with an AAV expressing a Cre-dependent mCherry reporter (AAV2/hSyn-DIO-mCherry) and immunolabeled these retinas for melanopsin (Fig. 1E). Of the melanopsin immunoreactive cells, 12% were mCherry+ (167 of 1437 cells from seven retinas), and the proportion of mCherry+ ipRGCs was highest (31%) in the dorsal-temporal quadrant of the retinas (Fig. 1, F and G). The individual densities of Gad2+ cells or of melanopsin immunopositive cells alone were not higher in the dorsal-temporal retinas compared with other quadrants (fig. S3). Additionally, Gad2 labeling was enriched in ipRGCs relative to the total RGC population because only 1% of the total RGCs and only 0.6% of the Brn3a+ (a marker of non-ipRGCs) RGCs were mCherry+ (fig. S4).

We performed RNA fluorescence in situ hybridization for Gad2 mRNA in ipRGCs to determine whether we could detect it in ipRGCs. We identified ipRGCs in retinal sections by probing for Opn4 mRNA (Fig. 2, A and B) and counting the number of Gad2 puncta in that region of interest (figs. S5 to S7 and methods). We did not observe any labeling in the retinas of Opn4 knockout (KO) animals (fig. S6). To establish a threshold for Gad2+ ipRGCs, we performed the same experiment in animals lacking Gad2 in ipRGCs (Opn4Cre/+; Gad2fx/fx). We estimate that Gad2 mRNA is detectable above background in 26% of ipRGCs (fig. S7). We next immunolabeled ipRGC terminals in the SCN for GAD65 protein, which is encoded by the Gad2 gene, in ipRGCs. Of the ipRGC nerve terminals in the SCN, 12% were GAD65 immunoreactive (Fig. 2, C to E). Less than 2% of ipRGC nerve terminals were GAD65 immunoreactive when we rotated the GAD65 channel (Fig. 2E), which indicates that colocalization levels were better than those that would be produced by chance. Likewise, <2% of ipRGC nerve terminals in Opn4Cre/+; Gad2fx/fx animals were GAD65 immunoreactive.

Fig. 2 ipRGCs express Gad2.

(A) In situ hybridization for Opn4 (green) and Gad2 (magenta). ONL, outer nuclear later; INL, inner nuclear layer; GCL, ganglion cell layer; DAPI, 4′,6-diamidino-2-phenylindole. (B) Gad2+ (top) and Gad2− (bottom) ipRGCs from (A). See figs. S5 and S7 for how the proportion of Gad2+ ipRGCs was estimated. (C) Strategies for labeling ipRGC axons in the SCN. Black text indicates that reporter mice, in which the synaptophysin-tdT fusion protein was expressed in the presence of Cre recombinase (Syp-tdT or Ai34), were intravitreally injected with the pgk-Cre virus. Gray text indicates that Opn4Cre/+ animals were intravitreally injected with a virus driving Cre-dependent expression of Chrimson-tdT. (D) ipRGC terminals of Syp-tdt mice intravitreally injected with pgk-Cre virus (top) and immunolabeled for GAD65 (magenta) and synapsin to label axon terminals (cyan). Bottom panels show magnified images of ipRGC axon terminals that were GAD65 immunoreactive (boxes 1 and 2) and GAD65 negative (box 3). (E) Percentage of ipRGC terminals that were GAD65 immunoreactive (IR). n = 5 Syp-tdT mice injected with pgk-Cre virus, n = 4 Opn4Cre/+ mice injected with FLEX-Chrimson-tdT virus, and n = 3 Opn4Cre/+; Gad2fx/fx mice injected with FLEX-Chrimson-tdT virus. Error bars indicate SD.

We next investigated whether ipRGCs functionally release GABA. We expressed channelrhodopsin-2 (ChR2) in RGCs by delivering AAVs that drive Cre expression (AAV2/pgk-Cre) to the eyes of Ai32 mice (which drives Cre-dependent expression of ChR2) (18). We then made acute brain slices containing the SCN, which receives retinal input exclusively from ipRGCs in mice (11, 17). We photoactivated the ipRGC axons (Fig. 3A) and voltage-clamped SCN neurons at −60 mV (ECl) to isolate excitatory postsynaptic currents (EPSCs) and then at 0 mV (Ecation) to isolate inhibitory postsynaptic currents (IPSCs). Tetrodotoxin (TTX) and 4-aminopyridine (4-AP) were included in the extracellular solution to ensure that the elicited postsynaptic currents were monosynaptic (19). Photoactivation of ipRGCs evoked synaptic currents in 32 of 79 (40%) SCN neurons including EPSCs only (18 of 32 cells), IPSCs but not EPSCs (3 of 32 cells), and both IPSCs and EPSCs (11 of 32 cells). Therefore, just >43% of SCN neurons receiving direct input from ipRGCs receive inhibitory ipRGC input (Fig. 3D). The synaptic latency of EPSCs and IPSCs elicited in SCN neurons was not significantly different, which suggests that both types of postsynaptic currents arise from monosynaptic input from ipRGCs (Fig. 3E). When possible, we mapped the location of recorded SCN neurons (fig. S8) and immunolabeled recorded cells for vasoactive intestinal peptide (VIP) (Fig. 3, B and C). A higher proportion of VIP+ neurons in the SCN received monosynaptic ipRGC input (59%) compared with VIP− neurons (33%) (Fig. 3F). A larger percentage of VIP+ neurons received purely excitatory ipRGC input (41%) compared with VIP− neurons (15%), and a similar proportion of VIP+ and VIP− neurons received inhibitory ipRGC input (Fig. 3F).

Fig. 3 Functional GABA release by ipRGCs.

(A) SCN acute brain slices were prepared from ChR2-YFP (Ai32) mice intravitreally injected with pgk-Cre virus. Full field 470-nm light flashes were used to photoactivate ipRGC axons. (B) Neurobiotin-filled SCN neurons (magenta, indicated by arrows) in SCN slices labeled for VIP. ipRGC axons are labeled in green. (C) Magnified images of the VIP− (top panels) and VIP+ (bottom) neurons in (B). (D) EPSCs (black) and IPSCs (red) elicited in SCN neurons after photoactivating ipRGC axons in the presence of TTX and 4-AP (n = 79 cells). The blue line indicates delivery of a 1-ms light stimulus. TTX, tetrodotoxin; 4-AP, 4-aminopyridine. (E) Synaptic latency of EPSCs and IPSCs after photoactivation of ipRGC axons (n = 29 EPSCs and 14 IPSCs). n.s., not significant. (F) Proportion of VIP+ (left) and VIP− (right) SCN neurons that receive excitatory and/or inhibitory input from ipRGCs (n = 17 VIP+ and 55 VIP− SCN neurons). (G) Example recording from an SCN neuron that receives both excitatory and inhibitory ipRGC input. Bath application of NBQX and D-APV abolished the EPSC but did not affect the IPSC. Subsequent application of gabazine abolished the IPSC. (H) EPSC (black, left) and IPSC (red, right) amplitude in SCN neurons receiving both excitatory and inhibitory input from ipRGCs before and after application of NBQX and D-APV and then gabazine. n = 6 cells. All data are means ± SD.

Bath application of 2,3-Dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide (NBQX) and D-(-)-2-Amino-5-phosphonopentanoic acid (D-APV) abolished the evoked EPSCs (Fig. 3, G and H) in SCN neurons receiving excitatory and inhibitory ipRGC input, but it did not affect IPSC amplitudes. This further confirms that elicited IPSCs were likely not a result of a disynaptic inhibition arising from evoked glutamatergic ipRGC inputs onto a GABAergic interneuron. Subsequent bath application of gabazine (SR-95531) abolished the remaining light-evoked IPSCs in SCN neurons (Fig. 3, G and H).

To assess how GABA release by ipRGCs influences non–image-forming visual behavior, we crossed Opn4Cre mice (20) with Gad2fx/fx mice (21) to knock out Gad2 specifically in ipRGCs (Opn4Cre/+; Gad2fx/fx, referred to as Gad2 cKO). These animals showed normal ipRGC projections and visual acuity, which indicates that this manipulation does not affect the development of the visual system (figs. S9 and S10). We measured the PLR of Gad2 cKO animals compared with littermate controls (Opn4+/+; Gad2fx/fx) across a range of light intensities and measured the irradiance-response relationship (Fig. 4, A to C). PLR amplitude and kinetics were both unaffected at bright light intensities in Gad2 cKO animals (Fig. 4C and fig. S11), and baseline pupil size in darkness was also unchanged (Fig. 4, A and B). However, at low light levels, Gad2 cKO mice showed significantly stronger pupil constriction (Fig. 4C) than that of their littermate controls, although the kinetics were not significantly different (fig. S11).

Fig. 4 GABA release by ipRGCs influences non–image-forming behaviors.

(A) Representative PLR images from control (left panels, Opn4+/+; Gad2fx/fx) and Gad2 cKO (right panels, Opn4Cre/+; Gad2fx/fx) mice in darkness (top), dim light (middle, 10.9 log quanta cm−2 s−1), and bright light (bottom, 13.9 log quanta cm−2 s−1). (B) Control (n = 5) and Gad2 cKO (n = 7) pupil area in the dark. (C) Irradiance-response relationship of PLR in control and Gad2 cKO mice. (D) Representative double-plotted actograms from control (top) and Gad2 cKO (bottom) mice. Mice were initially exposed to a 12:12 LD cycle with 100 lux light during the light phase. The light level was subsequently lowered to 1.5 and 0.2 lux. The mice were exposed to a 6-hour phase advance each time the light level was lowered. (E and F) Circadian amplitude measured using the peak amplitude of the χ2 periodogram (E) and percent activity during the light phase (F) in control (n = 8) and Gad2 cKO (n = 9) mice. All data are means ± SD. n.s., not significant; *P < 0.05; **P < 0.01 (Mann-Whitney U test).

We next tested whether these changes in PLR were caused by a lack of Gad2 in ipRGCs or, potentially, by developmental effects of Gad2 excision early in development. We knocked out the Gad2 gene specifically in RGCs of adult Gad2fx/fx mice through intravitreal delivery of an AAV that drives Cre expression under an RGC-specific promoter (AAV2/SNCG-Cre-HA) (22) (fig. S12 and methods). Less than 5% of amacrine cells were labeled in well-infected areas (fig. S12B). We then measured the PLR in response to dim (10.4 log photons cm−2 s−1) and bright (13.4 log photons cm−2 s−1) light stimuli. Mice in which Gad2 was knocked out in RGCs exhibited significantly more sensitive PLR in response to dim light but not bright light (fig. S12, C and D) compared with control animals. These results suggest that these changes in PLR are not due to developmental defects from Gad2 gene excision.

To determine whether GABA release by ipRGCs also influences circadian photoentrainment, we tracked voluntary wheel-running activity of littermate controls and Gad2 cKO mice across multiple light levels. Mice were first exposed to a 12-hour–12-hour light-dark (LD) cycle with bright, 100-lux light during the light phase. We then performed a 6-hour phase advance of the LD cycle at 4-week intervals and simultaneously lowered the light intensity at each shift, first to 1.5 and then to 0.2 lux (Fig. 4D and fig. S13). The rate of re-entrainment in response to both 6-hour phase advances was the same in Gad2 cKO compared to control animals (fig. S14). However, Gad2 cKO animals had significantly higher circadian amplitudes in LD cycles at low (1.5 and 0.2 lux) but not high (100 lux) light levels (Fig. 4E). This indicates that the circadian photoentrainment of Gad2 cKO animals remains more robust at lower light levels and is relatively insensitive to decreases in environmental light levels. There were no significant differences between Gad2 cKO and control animals in total daily activity, activity onset time, or activity onset variability (fig. S15). Instead, Gad2 cKO animals exhibited significantly less activity in the light phase at 1.5 and 0.2 lux, which likely accounts for their increased circadian amplitude at low light levels relative to controls (Fig. 4F).

Our results reveal a GABAergic circuit originating in the retina that decreases the sensitivity of the non–image-forming visual system at low light levels. Recent reports have shown that the ipRGCs providing input to non–image-forming brain regions are highly sensitive to dim light (2325), and yet the non–image-forming behaviors driven by these inputs are relatively insensitive to light compared with image-forming vision (26, 27). The mechanisms underlying this discrepancy between sensitive cellular inputs to non–image-forming brain regions and relatively insensitive behavioral outputs had remained a mystery. Our results suggest that a subpopulation of ipRGCs may serve to actively dampen the sensitivity of the non–image-forming visual behaviors by releasing the inhibitory neurotransmitter GABA, providing a circuit mechanism for this discrepancy. For the PLR, these inputs serve to maximize light entry into the eye at low light levels. For circadian behaviors, these inputs likely prevent unnecessary adjustments of the body’s master clock to relatively minor perturbations in environmental light.

Supplementary Materials

Materials and Methods

Figs. S1 to S15

References (2830)

MDAR Reproducibility Checklist

References and Notes

Acknowledgments: We thank T. Bozza, R. Allada, and M. Gallio for helpful comments on the manuscript; S. Hattar for the gift of Opn4Cre mice; and Q. Wu for the gift of Gad2fx mice. Funding: This work was funded by a Klingenstein-Simons Fellowship in the Neurosciences to T.M.S., a Sloan Research Fellowship to T.M.S., NIH grant 1DP2EY022584 to T.M.S, NIH T32 EY025202 to support T.S., and NIH F31 EY030360-01 to T.S. Author contributions: Conceptualization, T.S., J.Y.L., and T.M.S.; Investigation, T.S., J.Y.L., N.W.H., Y.O., and T.M.S.; Formal analysis, T.S.; Software, J.C.C.; Writing – original draft, T.S. and T.M.S.; Writing – reviewing and editing, T.S., J.Y.L., and T.M.S.; Visualization, T.S., J.Y.L., N.W.H., J.C.C., and T.M.S.; Resources, J.C.C., S.B., and H.N.; and Funding acquisition, T.S. and T.M.S. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the manuscript or the supplementary materials.

Stay Connected to Science

Navigate This Article