Abstract
The ability to encode the direction of image motion is fundamental to our sense of vision. Direction selectivity along the four cardinal directions is thought to originate in direction-selective ganglion cells (DSGCs), due to directionally-tuned GABAergic suppression by starburst cells. Here, by utilizing two-photon glutamate imaging to measure synaptic release, we reveal that direction selectivity along all four directions arises earlier than expected, at bipolar cell outputs. Thus, DSGCs receive directionally-aligned glutamatergic inputs from bipolar cell boutons. We further show that this bouton-specific tuning relies on cholinergic excitation and GABAergic inhibition from starburst cells. In this way, starburst cells are able to refine directional tuning in the excitatory visual pathway by modulating the activity of DSGC dendrites and their axonal inputs using two different neurotransmitters.
Results
Retinal bipolar cells are excitatory glutamatergic interneurons that transfer visual information from photoreceptors to the inner retina, where ganglion cells form the main output to the brain (1). There are 14 molecularly distinct types of bipolar cell in the mouse retina, each carrying different types of visual information, including luminance change, temporal kinetics, and chromatic sensitivity (1–3). Information about the direction of moving stimuli is thought to be encoded by a specialized class of retinal ganglion cells, DSGCs, which receive inputs from GABAergic/cholinergic starburst amacrine cells as well as glutamatergic bipolar cells. DSGCs respond to one of the four cardinal directions (dorsal, ventral, nasal, and temporal) with maximal spiking in their preferred direction, but only minimal spiking in the opposite (null) direction (4). Thus, DSGCs embody the retinal coordinates of motion computation (5). A body of literature supports the consensus that cardinal direction selectivity emerges first at DSGC dendrites, due to directionally-tuned GABAergic inhibition from starburst cells (4, 6–8). Indeed, studies that have imaged extracellular glutamate at DSGC dendrites without identifying the types of input bipolar cells (9, 10), or intracellular calcium in axon terminals of a small range of specific bipolar cell types (9, 11), have failed to identify directionally-tuned activity in bipolar cells. However, there is anecdotal evidence to the contrary (12), and none of the reports have incorporated a comprehensive analyses of functionally classified bipolar cell types.
Motivated by two recent findings that 1) DSGC dendrites receive locally tuned cholinergic inputs from starburst cells (13) and 2) alpha-7 nicotinic acetylcholine receptors (α7-nAChRs) are selectively expressed in type 7 (T7) ON and type 2 (T2) OFF bipolar cell (14), we tested the potential role of cholinergic signaling in the motion-related modulation of these specific bipolar cell types outputs. Subretinal injection of an adeno-associated virus (AAV) vector of serotype 8BP/2 (15), containing a CAG promoter, preferentially labeled T7, T2, and rod bipolar cells (Xin Duan, personal communication on CAG promoter tropism; Fig. 1A). This was confirmed by the depths to which labeled bipolar cell axon terminals projected in mice in which starburst cell processes were labeled with tdTomato (Fig. 1, B and C; fig. S1) (1, 3, 14, 16). This strategy allowed us to target these bipolar cell types with a glutamate sensor SF-iGluSnFR.A184S for imaging light-evoked glutamate release from axon terminals of T7 and T2 bipolar cells (Fig. 1, D and E; fig. S2, A to D). A number of neighboring axonal boutons exhibited correlated noise during a static flash (Fig. 1D; fig. S2, I to K), indicating that they belonged to the same bipolar cell (2, 17). Strikingly, motion stimuli revealed directional tuning in a fraction of boutons (Fig. 1, E to G), which displayed heterogenous direction preference (Fig. 1H; direction selective index (DSI), 0.29 ± 0.19, 1108 ON boutons; fig. S2E). There was no bias towards a specific direction (Fig. 1, I and J; fig. S2, F and G; p > 0.1, Hodges-Ajne test), but instead clusters along the four cardinal directions associated with ON-OFF DSGC firing patterns (Fig. 1J; p = 0.037, number of modes = 4, Silverman’s test; fig. S2H) (4, 5).
Axon terminals of bipolar cells are modulated by different types of amacrine cells (2, 18, 19), which might be responsible for the direction-selective glutamate release that we observed from bipolar cell terminals. Given the specific expression and conductance of α7-nAChRs in T7 and T2 bipolar cells (14), one potential mechanism is enhancement of glutamate release in response to the preferred direction by cholinergic starburst cells (20). Alternatively, GABAergic inhibition from wide-field amacrine cells (21), driven by voltage-gated sodium channels (NaV), could give rise to suppression in response to the null direction motion. To test these possibilities, we mapped the effects of pharmacological manipulation on individual boutons along the axon terminals of single T7 bipolar cells (Fig. 2, A and B). We found that blockade of both α7-nAChR by α-Bungarotoxin (α-Bgtx) and NaV by tetrodotoxin (TTX) diminished the tuning of direction-selective bipolar cell boutons.
These experiments revealed that boutons could be classified into four groups based on their sensitivity to pharmacological manipulation (Fig. 2C; fig. S3, A to D): 24% were sensitive to both α-Bgtx and TTX; 6% and 36% were sensitive only to α-Bgtx or TTX, respectively; and 34% of were sensitive to neither. Boutons that were sensitive to both α7-nAChR and NaV blockade showed significantly higher directional tuning under control conditions (mean DSI, 0.29 ± 0.19; Fig. 2D), compared to other boutons. In TTX-sensitive boutons, the effect of NaV blockade was occluded by pre-application of SR95531 and TPMPA, which block GABAA and GABAC receptors, respectively (Fig. 2E). This reveals that GABAergic inhibition by NaV-expressing wide-field cells, rather than NaV-expressing bipolar cell types (18, 22, 23), mediates directional tuning.
Directional tuning in α7-nAChR-blocked boutons was further reduced by subsequent application of SR95531 and TPMPA (fig. S3E) or TTX (fig. S3F), suggesting that GABAergic inhibition and cholinergic excitation are mechanistically independent. Furthermore, glutamate release from axonal boutons belonging to the same bipolar cell was highly correlated during static stimuli, but decorrelated during moving stimuli (fig. S3G). This motion-induced decorrelation was occluded by α-Bgtx and TTX, suggesting that glutamate release from T7 and T2 bipolar cells can be driven by dendritic inputs from photoreceptors, and modulated by the cholinergic excitation and GABAergic inhibition (fig. S3, G and H).
To examine the causal contribution of starburst cells to direction-selective glutamate release from bipolar cells, we selectively ablated starburst cells by intravitreal injection of diphtheria toxin to Rosa-iDTR × ChAT-IRES-Cre mice (Fig. 2F; fig S4, A to F). Direction selectivity in both T7 and T2 bipolar cells reduced significantly (Fig. 2, G and H, p = 6.82×10−37; fig S4, J to L). Blockade of α7-nAChRs did not affect direction selectivity in these ablated retinas, confirming that starburst cells mediate the cholinergic-dependent component of direction selectivity in bipolar cell terminals (Fig. 2I, cyan; fig S4, G and H). We further found that starburst cell ablation also occluded the effect of TTX on direction selectivity (Fig. 2I, orange; fig S4, H and I), suggesting that the contribution of wide-field cells to directional tuning is dependent on starburst cells.
To anatomically confirm the involvement of starburst and wide-field amacrine cells in bipolar cell computation of motion stimuli, we used a serial block-face electron microscopy (SBEM) dataset (24). We first reconstructed all ON and OFF bipolar cells that formed synapses with ON-OFF DSGC cells (fig. S5, A to D) (13, 24, 25), then focused on one of the reconstructed T7 bipolar cells (fig. S5E) for further tracing of synapses from amacrine cells onto this T7 bipolar cell. We identified a total of 112 amacrine cell-to-bipolar cell synapses and reconstructed the amacrine cell dendritic processes (Fig. 3, A to C). These reconstructed amacrine cells fell into two categories: wide-field cells (31%) with long axon-like processes showing little, if any, branching (Fig. 3B and D), and non-wide-field cells (69%) with highly-branched processes (Fig. 3, C and D; fig. S5, H to J). None of the non-wide-field cells were identified as starburst cells, which is consistent with previous reports (11, 26) and suggests that cholinergic signaling from starburst cells to T7 bipolar cells might be mediated by recently-identified non-synaptic, locally-tuned forms of transmission (13) rather than by clear synaptic structures. Of the terminal branches with output ribbon synapses that we observed, 20% received synapse from wide-field cells, 20% from both wide-field and non-wide-field cells, and 12% from only non-wide-field cells (Fig. 3E). 48% branches did not receive any inputs.
We subsequently searched for connections between wide-field cells and starburst cells by scanning the axons of wide-field cells that formed synapses with the reference T7 bipolar cell axon terminals. We found synapses from starburst cells to the axons of wide-field cells, but not in the vicinity of the ribbon synapse between the bipolar cell and DSGC (Fig. 3, F to H). Interestingly, the estimated preferred directions of wide-field cells varied among the terminal branches of the T7 bipolar cell arbor (Fig. 3I), indicating that GABAergic inhibition of T7 cell boutons is tuned to different directions (Fig. 1I).
Together, these anatomical, genetic, and pharmacological analyses suggest two potential circuit mechanisms that could establish direction-selective glutamate release by T7 bipolar cells (Fig. 3J). The first is preferred direction enhancement of release via α7-nAChR activation at the bipolar cell axon terminal (Fig. 3J, left). The second is null direction inhibition of release via GABAergic NaV-expressing wide-field cells (Fig. 3J, right). In the null direction, the lack of ACh input to the bipolar cell terminal and the induction of GABAergic inhibition from wide-field cells would result in suppressed glutamate release.
We next sought to test whether the directional tuning in T7 and T2 bipolar cell terminal boutons is received by ON-OFF DSGCs by targeting DSGCs labeled in the retinas of Cart-IRES-Cre mice (27) with Cre-dependent AAVs expressing iGluSnFR. Strikingly, two-photon imaging of iGluSnFR-labeled DSGC dendrites (Fig. 4, A and B; fig. S6, A to F) revealed that DSGCs receive both directionally-tuned and untuned glutamate inputs (Fig. 4, B to D). To examine T7 and T2 bipolar cell inputs, we initially performed unsupervised clustering to identify response kinetics (2, 18, 28, 29) (fig. S6, G to K). We then identified six ON groups along the ON arbor (Fig. 4E, F; G1-G6) and six OFF groups along the OFF arbor (fig. S6, I to K; G7-G12) of DSGC dendrites. Of these, groups G3 (ON) and G11 (OFF) were selectively modulated by α7-nAChR block (Fig. 4F; fig. S6L), suggesting that they correspond to inputs from T7 and T2 bipolar cells, respectively.
Among the twelve groups, distributions of DSI were skewed toward higher values in groups G3 and G11 (Fig. 4G; skewness and median: 0.42, 0.21 in G3; 0.44, 0.26 in G11; 0.12, 0.13 in all inputs; fig. S6, M to P), indicating that they include subsets of inputs with higher directional tuning. These directional tunings were diminished by α7-nAChR and NaV blockade (fig. S7A). In particular, the G3 group displayed a heterogeneous response to these blockers, similar to that of T7 bipolar cells (fig. S7, B and C): 21% of inputs were sensitive to both TTX and α-Bgtx (compared to 24% of T7 bipolar axonal boutons; Fig. 2C), 7% were sensitive only to α-Bgtx, 28% only to TTX, and 44% were insensitive to both. Intriguingly, the DSI along the preferred direction for each DSGC was significantly higher in G1 and G3, compared to the low DSI along a randomized direction (Fig. 4H; fig. S7, D to F). When considered alongside the results from axon terminal imaging, these data suggest that DSGCs selectively receive glutamatergic inputs matched to their preferred direction from among a heterogeneously-tuned range of T7 and T2 bipolar cell boutons.
Finally, we tested the contribution of T7 and T2 bipolar cell inputs to the tuning of DSGCs using two-photon targeted patch-clamp recordings from EYFP-labeled DSGCs in retinas isolated from Oxtr-T2A-Cre; Thy1-STOP-EFYP mice (13) (Fig. 4, I and K; fig. S7, G and H). Blockade of NaV and α7-nAChR reduced the direction-selectivity of excitatory postsynaptic currents (EPSCs) recorded in DSGCs. TTX-sensitive tuning reduction was occluded by pre-application of GABA receptor blockers (Fig. 4, J and K), revealing that this inhibitory effect was mediated by NaV-dependent wide-field cells. The amplitude of inhibitory inputs to DSGCs in the null direction was also decreased by α-Bgtx (fig. S7, I and J), suggesting that the dendritic processes of starburst cells are excited by T7 and T2 bipolar cells, as previously shown (30, 31).
Discussion
Previous work has established the view that starburst cell-to-ganglion cell synapses are the only sites involved in computation of cardinal motion direction selectivity in the retina (9–11). Our results show instead that cardinal direction selectivity first emerges in the axon terminals of bipolar cells, and that this axonal direction selectivity is due to cholinergic excitation from starburst amacrine cells and GABAergic null direction suppression from wide-field amacrine cells. These fine microcircuit mechanisms are implemented at individual synapses in such a precise way that directionally-tuned outputs can be transmitted to functionally aligned DSGC types. Future studies could investigate the developmental mechanisms that allow starburst cells to align the tuning of both input axon terminals and recipient dendrites.
Although synapses from T7 and T2 bipolar cells to DSGCs are not abundant (3.4% and 15.3%, respectively), our findings suggest that they contribute to direction selectivity in ganglion cells using at least two distinct mechanisms. First, the tuning of these bipolar cell boutons shapes directionally-sensitive excitatory inputs to DSGCs (Fig. 4, I to K). Second, these bipolar cells drive GABA releases from starburst cells (fig. S7, I and J) (30, 31). This concept is supported by a previous study showing that genetic perturbation of T7 bipolar cells weakens the tuning of ONOFF DSGCs (32).
Intriguingly, previous DSGC glutamate imaging studies and bipolar cell axon terminal calcium imaging studies have failed to identify directionally-tuned bipolar terminals. This discrepancy could be due to imaging populations versus specific cell types, use of larger regions of interest that blend different tuning types (9, 10) compared to those in this study (fig. S6F), or underestimates of the tuning strength due to potential nonlinear Ca2+ signals in the axon terminals (33) in a previous study (11). Importantly, the response amplitudes in our glutamate imaging experiments were set within a linear range to faithfully capture release tuning (fig. S2M; R2, 0.98).
What is the advantage of encoding all of four cardinal directions, rather than only one, in individual T7 and T2 bipolar cells? Because visual space is represented by a mosaic of bipolar cells of the same type (1), multiplexed axonal direction selectivity in T7 and T2 bipolar cells would be an efficient way of encoding all four directions for both ON and OFF responses at every point in the visual space without having extra cell types.
Funding
VELUX FONDEN Postdoctoral Ophthalmology Research Fellowship (27786) to A.M., Lundbeck Foundation (DANDRITE-R248-2016-2518), Novo Nordisk Foundation (NNF15OC0017252), Carlsberg Foundation (CF17-0085), and European Research Council Starting (638730) grants to K.Y.
Author contributions
A.M. and K.Y. conceived and designed the experiments and analyses. A.M. performed all physiological experiments. A.M. and S.S.N. performed immunohistochemistry and confocal scanning. K.Y. designed viral vectors and performed eye injections. A.M. analyzed the physiological and confocal data. W.A., R.A., H.L., and S.S. performed the connectomic analyses. A.M., S.S., and K.Y. interpreted the data and wrote the paper.
Competing interests
The authors declare no competing interests.
Data and materials availability
Correspondence and requests for materials should be addressed to S.S (shai.sabbah{at}mail.huji.ac.il) for connectome data or K.Y. (keisuke.yonehara{at}dandrite.au.dk) for physiological data.
Supplementary Materials
Materials and Methods
Animals
Wild-type mice (C57BL/6J) were obtained from Janvier Labs. ChAT-IRES-Cre (strain: Chattm2(cre)Lowl/MwarJ, Jackson laboratory stock: 028861) and Rosa-STOP-tdTomato (Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J, Jackson laboratory stock: 007905) were used for bipolar cell imaging. Cart-IRES-Cre (strain: Cartpttm1.1(cre)Hze/J, Jackson laboratory stock: 028533) was used for DSGC dendrite imaging. Oxtr-T2A-Cre; Thy1-STOP-EYFP (strain: Cg-Oxtrtm1.1(cre)Hze/J, Jackson laboratory stock: 031303; strain: Cg-Tg(Thy1-EYFP)15Jrs/J, Jackson laboratory stock: 005630) was used for electrophysiological recordings. Rosa-iDTR (strain: Gt(ROSA)26Sortm1(HBEGF)Awai/J, Jackson laboratory stock: 007900) crossed with ChAT-Cre was used for the genetic ablation of starburst cells. These mice were purchased from Jackson laboratory and maintained in a C57BL/6J background. We used 8- to 16-week-old mice of either sex. Mice were group-housed throughout and maintained in a 12-hour/12-hour light/dark cycle with ad libitum access to food and water. All animal experiments were performed according to standard ethical guidelines and were approved by the Danish National Animal Experiment Committee (Permission No. 2015−15−0201−00541 and 2020-15-0201-00452).
Retinal preparation
Retinas were isolated from the left eye of mice dark-adapted for 1 hour before experiments. The isolated retina was mounted on a small piece of filter paper (MF-membrane, Millipore), in which a 2 × 2 mm window had been cut, with the ganglion cell side up. During the procedure, the retina was illuminated by dim red light (KL 1600 LED, Olympus) filtered with a 650 ± 45 nm band-pass optical filter (ET650/45×, Chroma) and bathed in Ringer’s medium (in mM): 110 NaCl, 2.5 KCl, 1 CaCl2, 1.6 MgCl2, 10 D-glucose, 22 NaHCO3 bubbled with 5 % CO2, 95 % O2. The retina was kept at 35-36°C and continuously superfused with oxygenated Ringer’s medium during recordings.
Electrophysiology
Electrophysiological recordings were conducted with an Axon Multiclamp 700 B amplifier (Molecular Devices). Signals were acquired using customized software on LabVIEW (National Instruments) developed by Zoltan Raics (SENS Software) and digitized at 10 kHz. Borosilicate glass micropipettes pulled by a micropipette puller (P-97, Sutter Instrument) were used for recordings. The firing discharges were recorded in cell-attached mode using pipettes filled with Ringer’s medium. To visualize the dendrites of recorded neurons, Alexa 594 was added in intracellular solution (in mM): 112.5 CsCH3SO3, 1 MgSO4, 7.8 × 10−3 CaCl2, 0.5 BAPTA, 10 HEPES, 4 ATP-Na2, 0.5 GTP-Na3, 5 QX314-Br, 7.5 neurobiotin chloride. pH was adjusted to 7.2 with CsOH. The equilibrium potential for chloride was calculated to be ~ −60 mV. The resistance of pipettes was 5-8 mOhm for cell-attached recording. The labeled cells were targeted using a two-photon microscope equipped with a mode-locked Ti: sapphire laser (Mai Tai DeepSee, Spectra-Physics), set to 940 nm, integrated into the physiological recording setup (SliceScope, Scientifica), as described previously (18). The two-photon fluorescence image was overlaid on the infra-red (IR) image acquired by a CCD camera (RT3, SPOT Imaging). The IR light was generated by a digital light projector (NP-V311X, NEC) with a 750 ± 25 nm filter.
For pharmacological experiments, we used SR95531 (50 μM, Sigma) to bock GABAA receptors, TPMPA (100 μM, Sigma) to block GABAC receptors, alpha-Bungarotoxin (0.1 μM, Tocris) to block α7-nACh receptors, and tetrodotoxin (1 μM, Tocris) to block Na+ channels. These agents were bath-applied during recordings.
AAV production
The production plasmid for AAV-8BP/2-shortCAG.SF-iGluSnFR.A184S-WPRE-SV40p(A) for bipolar cell imaging was developed by Zurich Viral Vector Core by transferring the insert from pAAV.CAG.SF-iGluSnFR.A184S (Addgene #106198). The plasmid encoding the AAV capsid 2/8BP2 (14) was kindly provided by Jean Bennett (University of Pennsylvania). The AAV was produced by Zurich Viral Vector Core (v454; 3.5 × 1012 GC/ml). AAV9.hSyn.Flex.iGluSnFr.WPRE.SV40 for ganglion cell imaging was obtained from Penn Vector Core (#98931; 7.73 × 1013 GC/ml).
Virus injections
Mice were anesthetized with an i.p. injection of fentanyl (0.05 mg/kg body weight; Actavi), midazolam (5.0 mg/kg body weight; Dormicum, Roche), and medetomidine (0.5 mg/kg body weight; Domitor, Orion) mixture dissolved in saline. We made a small hole at the border between the sclera and the cornea with a 30-gauge needle. The AAV was delivered through a pulled borosilicate glass micropipette (30 μm tip diameter). All pressure injections were performed using a Picospritzer III (Parker) under a stereomicroscope (SZ61; Olympus). For targeting bipolar cells, 1 μl was pressure-injected through the hole into the subretinal space of the left eye. For targeting ganglion cells, 2 μl was pressure-injected into the vitreous of the left eye. Mice were returned to their home cage after anesthesia was antagonized by an i.p. injection of flumazenil (0.5 mg/kg body weight; Anexate, Roche) and atipamezole (2.5 mg/kg body weight; Antisedan, Orion Pharma) mixture dissolved in saline and, after recovering, were placed on a heating pad for one hour.
Diphtheria toxin injections
To genetically ablate starburst amacrine cells, we used Rosa-iDTR (+/−) × ChAT-IRES-Cre mice and control Rosa-iDTR (-/-) × ChAT-IRES-Cre mice (fig. S4, A to D). Diphtheria toxin stock solution was made from diphtheria toxin (D0564, Sigma), which was dissolved in PBS (1 μg/μl), and stored at −80°C. Immediately before the injection, the injection solution (1 ng/μl) was prepared by diluting the stock solution in PBS. First, we subretinally injected AAV-8BP/2-shortCAG.SF-iGluSnFR.A184S-WPRE-SV40p(A) to target T7 and T2 bipolar cells for glutamate imaging. Nine days later, 2 μl diphtheria toxin was intravitreally injected into each of both eyes. The eyes were re-injected with the same amount of diphtheria toxin two days after the initial injection.
Two-photon glutamate imaging
Three to four weeks after virus injection, we performed two-photon glutamate imaging (18). The isolated retina was placed under the microscope (SliceScope, Scientifica) equipped with a galvo-galvo scanning mirror system, a mode-locked Ti: Sapphire laser tuned to 940 nm (MaiTai DeepSee, Spectra-Physics), and an Olympus 60× (1.0 NA) or Olympus 25x (1.05 NA) objective. The retina was superfused with oxygenated Ringer’s medium. The iGluSnFR signals emitted were passed through a set of optical filters (ET525/50m, Chroma; lp GG495, Schott) and collected with a GaAsP detector. Images were acquired at 8-12 Hz using custom software developed by Zoltan Raics (SENS Software). Temporal information about scan timings was recorded by TTL signals generated at the end of each scan, and the scan timing and visual stimulus timing were subsequently aligned during off-line analysis.
Visual stimulation
The visual stimulation was generated via custom-made software (Python and LabVIEW) developed by Zoltan Raics (SENS Software). For electrophysiological recordings, the stimulus was projected through a DLP projector (NP-V311X, NEC). The stimulus was focused on the photoreceptor layer of the mounted retina through a condenser (WI-DICD, Olympus). The intensity was measured using a photodiode power meter (Thorlabs), and the power of the spectrum was measured using a spectrometer (Ocean Optics). The calculated photoisomerization rate ranged from 0.0025 to 0.01 × 107 photons absorbed per rod per second (R*/s) both for electrophysiological recordings and two-photon imaging. For glutamate imaging, the stimulus was projected using a DLP projector (LightCrafter Fiber E4500 MKII, EKB Technologies) coupled via a liquid light guide to an LED source (4-Wavelength High-Power LED Source, Thorlabs) with a 400 nm LED (LZ4-00UA00, LED Engin) through a band-pass optical filter (ET405/40×, Chroma). The stimuli were exclusively presented during the fly-back period of the horizontal scanning mirror (18). The contrast of visual stimulus (CS) was calculate as, in which LS and Lb indicate luminance intensity in stimulus and background, respectively.
Response measures
To evaluate sensitivity to luminance increments (ON) or decrements (OFF), we used static flash spots (300 μm in diameter, 2 s in duration, 100% positive contrast). To evaluate release kinetics, we used modulating spot (2,18). The stimulus (300 μm in diameter) had four phases: static flashing spot of 100% contrast, one of 50% contrast, one with increasing temporal frequency from 0.5 to 8 Hz, and one with increasing contrast from 5 to 80%.
To measure directional tuning and motion speed preference, we used a spot (300 μm in diameter, 100% positive contrast) moving in eight directions (0-315°, Δ45°) at 300 and 800 μm/s. The preferred direction was defined by a direction which elicited the maximum response, and the null direction was the opposite. To quantify the directional selectivity, we used a direction selectivity index (DSI) as, in which in which Rp and Rn indicate peak response amplitude in the preferred and null direction, respectively. DSI ranged from 0 to 1, with 0 indicating a perfectly symmetrical response, and 1 indicating a response only in the preferred direction.
The skewness (Skw) in DSI histogram was calculated to quantify the potential biases in the distribution by mean-variance around the mode: where x is DSI of individual input.M0 indicates a mode of distribution. < > and V () indicate mean and variance, respectively. To quantify precisely the directional biases in motion responses, an angle of preferred direction (θ) was defined by the vector sum: where i denotes the motion direction, and Ri denotes the response amplitude. For the analysis, we used ON-OFF DS ganglion cells whose DSIs were higher than 0.3 (9 of 12 cells: fig. S6D).
To quantify the effects of pharmacological blockers in light-evoked responses, we calculated change index (CI): where Rcontrol Rblocker response amplitudes in control and after blocker application, respectively.
Regions of interests (ROIs) detection
ROIs for glutamate signals were determined by customized programs in MATLAB. First, the stack of acquired images was filtered with a Gaussian filter (3 × 3 pixels), and then each image was downsampled to 0.8 of the original using a MATLAB imresize function. The signals in each pixel were resampled using the MATLAB interp function with a rate of 2 and smoothed temporally by a moving average filter with a window size of 2 time-bin. Next, we calculated the temporal correlation among the pixels during static flash stimulus based on a raw cross-correlation : in which Fp and Fq indicate glutamate signals in pixel p and q, respectively. The noise correlation (ncp,q) was then given by a subtraction: in which CRRS indicates trials-shuffled cross-correlation. The noise correlation was normalized as (N Cp,q): in which, < > and V () indicate mean and variance, respectively. We set a threshold of the score at 0 time-lag as 0.5 to determine which pixels were to be included as a single ROI. Then the response of each ROI (ΔF(t)) was calculated as where F(t) is the fluorescent signal in arbitrary units, F0 is the baseline fluorescence measured as the average fluorescence in a 1-second window before the presentation of the stimulus. After the processing, responsive pixels were detected based on a response index (RI): where Ri is a peak response amplitude during motion stimulus to direction i in ROI P, and indicates glutamate signals before the stimulus (1 s period). The ROIs with the RI higher than 0.6 were determined as responsive.
Clustering
The clustering for classifying glutamate inputs to DS ganglion cells was based on the temporal kinetics in the responses to modulating spot, as described previously (2,18). We used a sparse principal component analysis (sPCA) to extract temporal features in response to a modulating flash based on the SpaSM toolbox on MATLAB (28). Next, we fitted a Gaussian mixture model based on the expectation-maximization algorithm using the MATLAB gmdistribution function to the dataset of detected sparse features. To determine the optimal number of clusters in the model, we calculated the Bayesian information criterion (BIC) score (29): in which L is the log-likelihood of the model, k is the number of dimensions in the model, and n is the number of datasets. To separate ON and OFF input groups, we first performed the clustering using responses to static flash (the first phase in modulating spot), then we repeated the clustering, for the dissected each ON and OFF group, using the responses to the whole stimulus phases in the modulating spot. We sorted the detected clusters based on the similarity calculated by hierarchical clustering analysis using a standard linkage algorithm by MATLAB linkage function (Fig. 4E; fig. S6, G and I). We identified 12 groups (G1-G12; 6 ON and 6 OFF groups) in the glutamate inputs to ON-OFF DS ganglion cells.
Connectomic reconstruction
To characterize the bipolar-cell inputs to a DS ganglion cell and the amacrine-cell input to bipolar cells, we analyzed a set of serial electron microscopic sections of the adult mouse retina. The volume (k0725) is described in detail elsewhere (24). The volume was obtained from a young adult mouse (C57BL/6; 30 days of age) and fixed for 2 hr at room temperature in 2% buffered glutaraldehyde. A 1 mm2 sample obtained roughly midway between the optic disk and retinal margin was excised, stained with heavy metals to reveal synaptic ribbons and vesicles and other intracellular detail, dehydrated, and embedded in Epon Hard. A trimmed block (~200 mm x 400 mm) was imaged in a scanning electron microscope with a field-emission cathode (QuantaFEG 200, FEI Company). Back-scattered electrons were detected using a custom-designed detector and custom-built current amplifier. The incident electron beam delivered about 10 electrons/nm2. Imaging was performed at a high vacuum. Sides of the block were evaporation-coated with gold. The block face was serially cut as described elsewhere. Using a 26 nm section thickness, 10112 consecutive block faces were imaged, yielding aligned data volumes of 4992 x 16000 x 10112 voxels (135 mosaics of 3584 x 3094 images). This corresponds to a spatial volume of approximately 66 x 211 x 263 mm. The smallest dimension corresponds to retinal depth, which ranged from the ganglion cell layer to the innermost part of the inner nuclear layer. The edges of neighboring mosaic images overlapped by ~1 mm. Mosaics and slices were aligned offline to subpixel precision by Fourier shift-based interpolation. The datasets were then split into cubes (128 x 128 x 128 voxels) for import into Knossos (http://knossostool.org), a freely available software package for exploration and skeletonization of cell profiles in SBEM datasets. We also used webKnossos (http://webknossos.org), a cloud- and browser-based 3D annotation tool for large-scale data analysis of SBEM data.
Histology and confocal imaging
After the two-photon imaging experiments, retinas were fixed for 30 minutes in 4% paraformaldehyde in PBS and washed with PBS overnight at 4 ? on a shaker. The retinas were incubated in 30% sucrose in PBS for at least 3 hours at room temperature (RT). To enhance the penetration of antibodies, retinas were transferred in the sucrose buffer and frozen and thawed three times. After washing with PBS, retinas were blocked for 3 hours in blocking buffer (1% bovine serum albumin [BSA], 10% normal donkey serum [NDS], 0.5% TritonX 100, 0.02% sodium azide in PBS) at RT. The retinas were incubated with primary antibodies (chicken anti-GFP 1:1000 [abcam, ab13970]; goat anti-ChAT 1:200 [Milipore, ABN1144P]) for 5 days at RT in antibody reaction buffer (1% BSA, 3% NDS, 0.5% TritonX 100, 0.02% sodium azide in PBS), and secondary antibodies (donkey anti-chicken IgY Alexa 488 1:200 [Jackson ImmunoResearch, AB 2340375]; donkey anti-goat IgG Alexa 568 1:200 [Invitrogen, A11057]) for one day at 4 ? in antibody reaction buffer. After a final washing in PBS, retinas were embedded in Fluoromount-G (eBioscience). The stained retinas were imaged using a confocal microscope (Zeiss LSM 780) using a 40x (1.4 NA) or 63x (1.2 NA) objectives. The images were acquired at 1024 × 1024 pixel (0.35 μm/pixel for 40x; 0.22 μm/pixels for 63x), and the optical thickness of each imaging plane in z-stack was 0.3 μm. The images were processed and analyzed using MATLAB.
Statistical analysis
All analyses and statistical tests were performed by MATLAB 2017b (Mathworks). Population data were shown as mean ± SD. To compare the differences in paired conditions, Wilcoxon singed-rank test was used. To compare the differences in different groups, Mann-Whitney-Wilcoxon test was used. To compare the differences in distribution, Kolmogorov-Smirnov test was used. To evaluate the uniformity in angler distribution, Hodges-Ajne test was used. Fitting of the Gaussian function was based on the least-square method in MATLAB (Fig. 4; fig. S2). The numbers of modality in the preferred direction of bipolar cell axonal boutons (Fig. 1; fig. S2) were evaluated by Silverman’s test using kernel density estimation. The kernel was a Gaussian kernel. The p-values were calculated by the Bootstrap method (2000 times replication; fig. S2).
Data and code availability
The datasets and code generated in this research are available from the corresponding author, Keisuke Yonehara (keisuke.yonehara{at}dandrite.au.dk) for imaging data or Shai Sabbah (shai.sabbah{at}mail.huji.ac.il) for EM data upon reasonable requests.
Acknowledgments
We thank David Berson for providing us with an SBEM volume with traced DSGCs and bipolar cells, Xin Duan for sharing unpublished findings on viral tropism, Zoltan Raics for developing our visual stimulation system, and Bjarke Thomsen and Misugi Yonehara for technical assistance. We also thank Antonia Drinnenberg, Karl Farrow, and Stuart Trenholm for critical comments on an early version, and Lesley Anson for comments on a later version, of the manuscript.