Animals

Animal protocols were reviewed by the institutional preclinical core facility at the Institute of Science and Technology Austria (ISTA). All breeding and experimentation were performed under a license approved by the Austrian Federal Ministry of Science and Research in accordance with the Austrian and EU animal laws (BMF-66.018/0017-WF/V/3b/2017). During the experimental phase, mice were housed individually in standard macrolon cages with red plastic houses, running wheels and enrichment consisting of wood chips and nesting material, on an inverted 12-h light cycle. Experiments were done during the dark phase of the light cycle.

For in vivo tracing (n = 15, 8 males, 7 females), ex vivo patch-clamp (n = 11, 7 males, 4 females), in vivo opto/electrophysiology (n = 9, 5 males, 4 females), in vivo vLGN terminal imaging (n = 6, 4 males, 2 females) and in vivo TeLC experiments (n = 12, 6 males, 6 females), Gad2-IRES-Cre (JAX, cat. no. 010802) mice, aged 8 weeks (5–12 weeks for vLGN bouton imaging) at viral injection, were used. For in vivo anterograde transsynaptic experiments, Ai75D (JAX, cat. no. 025106, n = 6, 4 males, 2 females) mice, aged 8 weeks at viral injection, were used. For in vivo retrograde transsynaptic experiments, NTSR1-GN209-Cre (MMRRC, cat. no. 030780, n = 4, 2 males, 2 females), GRP-KH288-cre (MMRRC, cat. no. 031183, n = 4, 2 males, 2 females) and Rorb-Cre (JAX, cat. no. 023526, n = 4, 2 males, 2 females) mice, aged 8 weeks at viral injection, were used. The mice for retinal terminal imaging experiments were C57BL/6J (JAX, cat. no. 000664; n = 5, 3 males, 2 females), aged 6–11 weeks at eye injection. Of those, three mice have been used to record previously published separate datasets70.

Statistics and reproducibility

No statistical methods were used to pre-determine sample sizes but our sample sizes are similar to those reported in previous publications15,16,29. Extracellularly recorded units and imaged bouton regions of interest (ROIs) were selected or excluded based on quality and response criteria as described below. No other data points or animals were excluded in this study. For behavioral experiments, animals were randomly assigned to control and experimental groups and tested in random order. Experimenters were not blinded to the assignment of animals to experiments. However, the procedure for behavioral testing and data collection was automated and pooled for batch analysis. For all visual stimulation experiments, stimulus presentation was randomized.

Viral vectors

Anterograde transsynaptic expression was done with AAV1-cre (AAV1.CamKII0.4.Cre.SV40, 7 × 1012 genome copies per milliliter, Addgene). Retrograde transsynaptic expression was performed with starter vector (AAV-DIO-Ef1a-TVA-FLAG-2A-N2C_G)30 and pseudotyped rabies vector (N2C(Enva)-EGFP, ~2–5 × 108 genome copies per milliliter)30. Calcium indicator expression in vLGN/IGL neurons was achieved with AAV-hSynapsin1-FLEx-axon-GCaMP6s (1 × 1012 genome copies per milliliter, Addgene, cat. no. 112010-AAV5) for retinal expression AAV2.7M8-syn-GCaMP8m viral vectors (1 × 1013 genome copies per milliliter), generated at ISTA viral facility. TeLC viruses were generated using AAV5-hSyn-FLEX-TeLC-P2A-dTomato (Addgene, cat. no. 159102, 1 × 1013 genome copies per milliliter) at ISTA viral facility. ChR2 (AAV5-EF1a-doubleFloxed-hChR2(H134R)-EYFP-WPRE-HGHpa (cat. no. 20298-AAV5), 1 × 1013 genome copies per milliliter), mCherry control (AAV5-hSyn-DIO-mCherry (cat. no. 50459-AAV5), 7 × 1012 genome copies per milliliter) and eNpHR3.0 (AAV5-hSyn-eNpHR3.0-EYFP (cat. no. 26972-AAV5), 1013 genome copies per milliliter) viruses were purchased from Addgene.

Stereotaxic viral injections

Anesthesia was induced with 3% isoflurane and intraperitoneal (i.p.) ketamine and xylazine (100 mg kg−1, 10 mg kg−1). As an analgesic, meloxicam (20 mg kg−1) was subcutaneously injected. Mice were placed in a stereotaxic apparatus (Kopf) and body temperature was controlled with a heating pad at 37 °C throughout the whole procedure. Stereotaxic target coordinates relative to bregma were: −2.3 mm anterior-posterior (AP), 2.5 mm medial-lateral (ML), 3.4 mm dorso-ventral (DV) for vLGN; −2.3 mm (AP), 2.3 mm (ML), 3.4 (DV) for medial thalamus; −3.8 mm (AP), 0.8 mm (ML), 1.3 mm (DV) for SC; and 4 mm (AP), 2.6 mm (ML), 0.5 mm (DV) for visual cortex injection. Glass electrodes were pulled with a one-stage puller (DMZ-Zeitz-Puller) to produce a tip opening ~30 μm. The pipette was filled with mineral oil, then attached to a Nanoliter 2010 (World Precision Instruments) and loaded with the respective vector. Pipettes were slowly lowered to the target region (vLGN 150 nl/300 nl, SC 200 nl/300 nl, primary visual cortex 60 nl) and the solution was injected at a rate of 45 nl min−1. Once the volume was delivered, pipettes remained in place for 15 min before being carefully withdrawn and the incision closed with VetBond (3M). For retrograde transsynaptic tracing, the pseudotyped rabies vector was injected 7 d after the first starter vector injection. Otherwise, animals were recovering and awaiting viral expression for at least 3 weeks, before further experiments were conducted. For vLGN terminal imaging experiments, in four of six mice viral infection was immediately followed by cranial window implantation in the same surgery. The remaining two mice were implanted 4 weeks after the injection surgery.

In vitro electrophysiology

Mice were deeply anesthetized via i.p. injection of ketamine (95 mg kg−1) and xylazine (4.5 mg kg−1), followed by transcardial perfusion with ice-cold, oxygenated (95% O2, 5% CO2) artificial cerebrospinal fluid (ACSF) containing (in mM): 118 NaCl, 2.5 KCl, 1.25 NaH2PO4, 1.5 MgSO4, 1 CaCl2, 10 glucose, 3 myo-inositol, 30 sucrose, 30 NaHCO3; pH 7.4. The brain was rapidly excised and coronal sections of 300-µm thickness containing the SC were cut using a Linear-Pro7 vibratome (Dosaka). Slices were left to recover for 20 min at 35 °C, followed by a slow cool down to room temperature over 40–60 min. After recovery, one slice was transferred to the recording chamber (cat. no. RC-26GLP, Warner Instruments) and superfused with ACSF containing 2 mM CaCl2 at a rate of 3–4 ml min−1 at room temperature (21.0–23.0 °C). Glass pipettes (cat. no. B150-86-10, Sutter Instrument) with resistances of 3–4 MΩ were crafted using a P1000 horizontal pipette puller (Sutter Instrument) and filled with internal solution containing (in mM): 140 K-gluconate, 2 MgCl2, 2 MgATP, 0.2 NaGTP, 0.5 EGTA, 10 HEPES; pH 7.4 adjusted with KOH. Biocytin (0.2–0.3%) was added to the internal solution for post hoc morphological reconstruction. Electrical signals were acquired at 20–50 kHz and filtered at 4 kHz using a Multiclamp 700B amplifier (Molecular Devices) connected to a Digidata 1,440 A digitizer (Molecular Devices) with pClamp10 software (Molecular Devices). For optogenetically evoked inhibitory postsynaptic currents, neurons were held at −60 mV and blue light (λ = 465 nm, 10–20-mW cm−2 intensity, 5-ms pulse duration, 0.1–0.2-Hz stimulation frequency) was emitted through a mono fiber-optic cannula (5-mm length, fiber diameter 200 μm, total diameter 230 μm, Doric Lenses) connected to a PlexBright LED 644 (Plexon) with an optical patch cable (fiber diameter 200 μm, total diameter 230 μm, 0.48 numerical aperture (NA)). To block GABAA receptors, ACSF containing 20 µM bicuculline was bath-applied for 20–30 s followed by immediate washout. Access resistance was constantly monitored between protocols, and recordings with access resistances exceeding 20 MΩ or with changes in access resistance or holding current of more than 20% were discarded. After recordings, the pipette was carefully withdrawn and the slice was transferred to 4% paraformaldehyde in PBS solution.

Viral eye injections

For expression of calcium indicators in retinal neurons, C57BL/6J mice were anesthetized with ketamine/xylazine (100 mg kg−1, 10 mg kg−1) by i.p. injection. A small hole in the temporal eye, below the cornea, was cut with a 1/2-inch, 30-gauge needle. Subsequently, 1 μl of vitreous fluid was withdrawn and 1 μl of AAV2.7M8-syn-GCaMP8m viral vector solution was injected into the subretinal space with a Hamilton syringe and a 33-gauge blunt-ended needle. Mice were left to recover and viral expression was to commence for 2–4 weeks before implantation of the cranial window.

Cranial window implantation surgery

For cranial window implantation, mice were injected with meloxicam (20 mg per kg body weight, subcutaneous (s.c.), 3.125 mg ml−1 solution) and dexamethasone (0.2 mg per kg body weight, i.p., 0.02 mg ml−1 solution). Anesthesia was induced by 2.5% isoflurane in oxygen in an anesthesia chamber and maintained at 0.7% to 1.2% in a stereotaxic device (Kopf), while body temperature was controlled by a heating pad to 37.5 °C. After exposing and cleaning the cranium, a 4-mm circular craniotomy was drilled above the left SC, the dura mater was removed and the left transverse sinus was sutured twice with 9-0 monofil surgical suture material (B. Braun) and cut between the sutures. Cortical areas covering the left SC were aspirated with a cell culture vacuum pump (Accuris), and a 3-mm circular coverslip, glued (Norland optical adhesives 61) to a stainless-steel conical ring, was inserted with the glass flush on the surface of the SC. After filling the surrounding cavity with Dura-Gel (Cambridge Neurotech), the insert was fixed in place with VetBond (3M). Finally, a custom-designed TiAl6V4 head-plate was affixed to the cranium by sequential application and curing of (1) All-in-One Optibond (Kerr), (2) Charisma Flow (Kulzer) and (3) Paladur (Kulzer). Mice were given 300 µl of saline and 20 mg per kg body weight meloxicam (s.c.) before removing them from the stereotaxic frame and letting them wake up, while keeping them warm on a heating pad. Further doses of 20 mg per kg body weight meloxicam (s.c.) and 0.2 mg per kg body weight i.p. dexamethasone were injected 24 h after the conclusion of the surgery. After the implantation surgery, mice were allowed to recover for at least 1 week.

Setups for head-fixed in vivo recordings

For awake, behaving experiments, two similar setups were used, with the difference that one was coupled to a custom-built multiphoton setup, and the other allowed for silicon probe/neuropixels recordings. In short, mice were head-fixed while awake using a custom-manufactured clamp (for imaging: connected to a three-axis motorized stage (cat. no. 8MT167-25LS, Standa)) and could run freely on a custom-designed spherical treadmill (20-cm diameter). Running behavior was recorded by a pair of ADNS-3080 (iHaospace, Amazon) optical flow sensor modules, focused with 25-mm lenses (cat. no. AC127-025-AB-ML, Thorlabs) on a small patch at orthogonal locations of the Styrofoam ball and illuminated by an 850-nm light-emitting diode (LED). The alternating sensor readout was controlled at 50 frames per second by an Arduino Uno running custom scripts. The four signal channels from the sensor were linearly mapped to movement speed in the forward, sideways and rotational axes based on regular calibration with synchronous measurement of image translations and rotation at the ball’s apex. Eye and body movements were recorded at 50 frames per second with infrared illumination (850 nm) with a camera (cat. no. acA1920-150um, Basler) and an 18–108-mm macro zoom objective (MVL7000, Thorlabs) for multiphoton imaging or a fixed focal length objective for electrophysiology (Edmund Optics, f = 50 mm, cat. no. 59-873), pointed at the right side of the mouse via an infrared mirror. Eye position and saccades were determined post hoc as previously reported70, by first labeling eight points around the pupil with DeepLabCut71, which were fitted to an ellipse, and the center position was transformed to rotational coordinates. Fast eye position changes of more than 45° s−1 and at least 3° amplitude on a 0.7-s median filtered trace were defined as saccades. The ellipse area in mm2 was determined as pupil size.

Visual stimuli were projected by a modified LightCrafter (Texas Instruments) at 60 Hz (Multiphoton setup: DLP LightCrafter evaluation module; e-phys setup: DLP LightCrafter 4500, Texas Instruments), reflected by a quarter-sphere mirror (Modulor) below the mouse and presented on a custom-made spherical dome (80 cm in diameter) with the mouse’s head at its center. For imaging experiments, a double bandpass filter (387/480 HD Dualband Filter, Semrock) was positioned in front of the projector to minimize light contamination during imaging. In both setups, the blue LED in the projector was replaced by ultraviolet (cat. no. LZ1-00UB00-01U6, Osram) and, in addition, in the multiphoton setup, the green LED was replaced by a cyan LED (cat. no. LZ1-00DB00-0100, Osram) not to interfere with the calcium imaging wavelengths. The reflected red channel of the projector was used for synchronization and captured by a trans-impedance photo-amplifier (cat. no. PDA36A2, Thorlabs) and digitized. Stimuli were designed and presented with Psychtoolbox-3 (ref. 72), running on MATLAB (MathWorks) on Microsoft Windows 10 systems. Stimulus frames were morphed on the GPU using a customized projection map and an OpenGL shader to counteract the distortions resulting from the spherical mirror and dome. In both setups, the dome allows the presentation of mesopic stimuli from circa 100° on the left to circa 135° on the right in azimuth and from circa 50° below to circa 50° above the equator in elevation. In between dynamic stimuli presented in randomized order, the screen was set to a homogeneous gray (green and ultraviolet light) at scotopic level for at least 30 s. To determine behavioral coupling, these stimuli were interspersed with 5-min gray screens, that is, at visual baseline.

In vivo electrophysiology and optogenetics

Gad2-Cre mice, previously injected with AAV5-EF1a-doubleFloxed-hChR2(H134R)-EYFP, were anesthetized with isoflurane (1–1.5% in oxygen 0.8 l min−1) and injected with meloxicam (20 mg kg−1, s.c.) and placed in the stereotaxic apparatus. The skull was exposed and the periosteum and connective tissue removed. Thin crossed grooves over the bone were cut to increase the contact surface using a scalpel. The skull was first covered with a thin layer of cyanoacrylate (VetBond, 3M), then Charisma Flow (Kulzer) that was blue-light-cured for 45 s, before securing a head-plate with SuperBond dental adhesive resin cement (Sun Medical). A tapered optic λ-fiber with an active zone of 0.5 mm (NA 0.39, Optogenix) was implanted using the same vLGN coordinates and craniotomy as the injection. The tip of the fiber was slowly lowered to a depth of 3.4 mm from the dorsal surface and cemented to the skull.

At 1 d before the recording session, mice were anesthetized with isoflurane (1–1.5% in oxygen 0.8 l min−1) and injected with meloxicam (20 mg kg−1, s.c.). A small craniotomy was made in the rostral skull (bregma: 0.5 mm AP, 2 mm ML) for implanting an inverted gold pin as a reference electrode. A second rectangular craniotomy was made over the cortex/SC region (bregma: −3.5–3.8 mm AP, 0.5–1 mm ML), leaving the dura mater intact. The window was covered with silicone elastomer (Kwik-Cast, World Precision Instruments). The next day, Kwik-Cast was removed and the well around the craniotomy was constantly filled with ACSF throughout the whole recording session. Extracellular recordings were obtained using a single shank acute linear 32-channel silicon probe (ASSY-37 H4 with probe tip sharpening, Cambridge Neurotech) connected to an RHD 32-channel amplifier board and RHD2000 USB Interface Board (Intan Technologies) and Neuropixels 2.0 multishank probes (IMEC), using a Neuropixels data-acquisition system (see www.neuropixels.org for more detail). Before recording, the tip of the electrode was coated with DiI (Invitrogen) to allow post hoc recording site location. To access the sSC, the probe was slowly inserted through the cortex at a speed of 1 μm s−1 to a depth of ~1.7 mm using a stable micromanipulator (Luigs & Neumann Motorized). The electrode was left in place for 30 min before starting to record. Data were sampled at 20 kHz using Labview 2017 (National Instruments). Spike-sorting was performed with Kilosort 2 (https://github.com/cortex-lab/Kilosort)73. The automatic template of Kilosort 2 was manually curated on Phy2. The 473-nm laser (cat. no. SDL-473-XXXSFL-RA, Shanghai Dream Laser Technology) bursts for optogenetics were generated in Arduino Due (www.arduino.cc) in pulses of 40 Hz with an approximate power at the fiber tip of 2.5 mW mm−2.

Visual local flash and optogenetics

Before starting the experiment, the visual field was scanned with a dark disk with 10° radius to determine the approximate location of RFs. We used this location to present a white or dark disk of the same radius. The visual local flash was interleaved in time with the laser burst of the same length, stimulating optogenetically vLGN. The duration of the laser and visual stimulation was 200 ms (Fig. 2e–g,i) or 1 s (Fig. 2h and Extended Data Fig. 8a–f). Laser burst onsets relative to the onset of the visual flash were randomized and varied in 13 increments. The start of the burst i was set to −1.5 × tflash + i × tflash/4 for i in [0, 12], where tflash was the duration of the flash.

CSD analysis

To confirm the location of the silicone probe during the in vivo recordings, CSD analysis74,75 was applied. For this analysis, local flash stimuli that were at least 0.5 s after the laser burst were used. For each such repetition of the flash, CSD profile75 was computed on the raw voltage recorded values in the interval [−0.1, 0.2] s around the flash onset and averaged over multiple repetitions. The channel of the silicone probe corresponding to the current sink is defined as the channel where the current flow is the smallest. The closest channel above with positive current flow is the source. The depth of the source channel was set to 300 μm; the depth of the remaining channels was derived relative to the source using the 25-μm spacing between the channels. The response magnitude (Fig. 2f) was computed as the variance of the CSD profile of each channel across time. The normalized response (Fig. 2g) is the variance across all channels after the variance before the onset of the flash was subtracted and normalized to the maximum. The same procedure was applied to compute the CSD analyses around laser bursts, but selecting laser burst onsets that were at least 0.5 s after a visual flash.

Neuronal responses

Both the zeta-test76 and a permutation test77 with subsampling were used to identify units that were responsive (P < 0.01) to the visual or optogenetic stimulation. The two tests detect complimentary response types: the zeta-test identifies event-locked responses, whereas the permutation test captures changes in the mean firing rate, including tonic changes of the firing rates, such as in the case of optogenetic stimulation of vLGN/IGL complex. For the permutation tests, the firing rate during the stimulus was compared with the baseline firing rate, estimated from random samples of 0.2-s intervals before the stimulus. Units were defined as visually responsive if they had either ON or OFF responses to the flash within 0.2 s after the start/end of the flash. A unit was considered optogenetically responsive if it exhibited a change of the spontaneous firing rate during 0.2 s after the start or end of the laser burst or if its visual ON/OFF responses were altered in the presence of optogenetic stimulation (permutation test). For analyses in Fig. 2h–l, only units responsive to both the visual flash and optogenetic stimulation were selected. To compute responses to visual flashes, optogenetic stimulation and both of the above (Fig. 2h–j and Extended Data Fig. 2b–d,f), we used the trials where the visual flash preceded optogenetic stimulation (visual responses, Fig. 2h and Extended Data Fig. 2b), or vice versa (optogenetic responses, Fig. 2h and Extended Data Fig. 2b), or where visual and optogenetic stimulation overlapped (visual and optogenetic responses, Fig. 2h and Extended Data Fig. 2b). In Fig. 2l and Extended Data Fig. 2e, the mean response of all units per laser offset was computed after normalizing the responses of each unit to their maximum across all laser offsets. To analyze RFs (Fig. 2m), a vertical bar (size 2° and 8°) was presented at a random horizontal position on the screen, and the location of the bar was updated with the frequency 15 Hz. At the same time, the pulses of optogenetic stimulation of LGN/IGL complex happened with the frequency of 1.4 Hz and pulse duration of 0.1 s, and this short duration of optogenetic pulses did not cause pupil dilation. To avoid the effect of rebound spiking, the spikes within 0.1 s after optogenetic pulses were removed from the analysis. We subsampled the spikes so that the numbers of spikes in conditions with and without optogenetic stimulation matched for each unit. To reconstruct RFs, we averaged the frames presented during [−0.5, 0.08] s around each spike, separating conditions into groups with and without optogenetic stimulation at the time of the spike. For further analysis, we excluded the units with noisy RF in either of the two conditions. For this, signal-to-noise ratio (SNR) was computed as the ratio of the variance of the RF in the time interval [T − 1, T + 1], to the variance outside of this interval, where T is the time of the maximal RF variance. The threshold for SNR was set to the 80th percentile of SNR of all units, at the value 3.44. The horizontal profile of an RF (Fig. 2m) was computed as the mean of three frames around T. We fit the one-dimensional Gaussian function (Fig. 2m) g(x) = A × exp((x − m)2/(2w2)) + b, where the parameters A, m, w and b are the amplitude, mean, width and the baseline. The fitting was done using the lsqcurvefit MATLAB function. The width of the RF (Fig. 5n) was estimated using the fitted parameter w.

Optogenetically triggered behavior analysis

For vLGN, tapered optic λ-fibers with an active zone of 0.5 mm (NA 0.39, Optogenix) were implanted using the same vLGN coordinates as vector injections. For SC, optic fibers (400-μm diameter, NA 0.39, ThorLabs) were implanted at 1,000 μm from the pial surface using the same AP and ML coordinates as vector injections. Both types of fiber were fixed using light-curing glue (Optibond Universal, Kerr Dental) and dental cement (SuperBond C&B Kit, Hentschel-Dental).

To analyze optogenetically triggered behaviors, only 1-s optogenetic stimulation pulses, where the offset preceded the visual flash, were included. To determine turning speed (Fig. 6k), mean speed in a window of 0.25 s before stimulation onset was subtracted and trials sorted by mean speed in a 0.5-s window after stimulation onset, for visualization. Maximum turning speed (Fig. 6l) is the maximum within 0.25 s after the laser onset. Change of azimuthal pupil position (Δaz) was normalized and sorted using the same windows, defining starting position as the mean pupil azimuth between 0.25 s before and at laser onset (Extended Data Fig. 8b,d,f). Pupil velocity (Extended Data Fig. 8a,c,e) was computed as a central difference of sequential pupil azimuth values vt = (azt + 1 − azt − 1)/2. Maximum pupil velocity (Fig. 6h) is the maximum of the velocity profile of each trial in the 0.25 s after the laser onset, and the mean of each group was computed from the trials in Extended Data Fig. 8a,c,e. Change in pupil azimuth (Fig. 6g) was defined as Δaz = azT − w/2 − azT + w/2, with the location T and width w of the maximum peak in the pupil velocity profile computed using the MATLAB function findpeaks. To determine changes of pupil size (Fig. 6i–j and Extended Data Fig. 8h,j), mean pupil diameter was computed during 0.25 s before and after laser offset; the difference between the two conditions was estimated using the Wilcoxon signed rank test (P = 10−74). Fig. 6j shows the pupil changes for all instances for the three experimental conditions in Fig. 6i and Extended Data Fig. 8h,j.

In vivo vLGN terminal imaging

Two-photon terminal imaging in SC was performed using a custom build system, controlled by ScanImage (Vidrio Technologies) on MATLAB 2020b (MathWorks) using a pulsed Ti:Sapphire laser (Mai-Tai DeepSee, Spectra-Physics) set at wavelengths between 920 and 950 nm. The beam was expanded to underfill the back-aperture of the objective (×16, 0.8-NA water-immersion, Nikon) and scanned through the tissue by a galvanometric-resonant (8 kHz) mirror combination (Cambridge Scientific) and a piezo actuator (cat. no. P-725.4CA, Physik Instrumente) controlling the objective. Emission light was measured with GaAsP photomultiplier tubes (cat. no. H10770B-40, Hamamatsu) following collection by a dichroic mirror (cat. no. FF775-Di01, Semrock) and channel splitting (580-nm long-pass, cat. no. FF580-FDi01, Semrock) as well as filtering (green: cat. no. FF03-525/50; red: cat. no. FF01-641/75; Semrock). The signals were then amplified by a TIA60 amplifier (Thorlabs) and digitized with a PXI system (PXIe-7961R NI FlexRIO FPGA, NI 5734 16-bit, National Instruments). Average laser output power at the objective ranged from 38 to 125 mW (median of 75 mW). A field of view of 0.13–1.85 mm2 (median of 0.77 mm2) was imaged over 3–12 planes (median of 6 planes) with a plane distance of 10–45 µm (median of 28 µm), at a pixel size of 0.6–1.9 µm (median of 1.3 µm) and a volume rate of 4.2–9.5 Hz (median of 4.8 Hz). The field of view varied between recordings, ranging from 0.2 to 1.8 mm2 (median = 0.7 mm2) of the SC surface for vLGN/IGL terminal imaging and from 0.1 to 1.6 mm2 (median = 0.7 mm2) for retinal bouton imaging. Each mouse was recorded in 1–9 (median of 7) imaging sessions on different days. In a subset of recordings (n = 15) in separate imaging sessions, the absence of substantial z-motion was verified by injecting 40 µl of Texas Red dextran (3,000 MW, 14.3 mg ml−1, diluted in saline, Thermo Fisher) subcutaneously and imaging brightly red-labeled blood vessels at 980 nm (ref. 78).

Visual stimuli for in vivo terminal response mapping

To measure sensitivity to luminance dynamics, repeated sequences of luminance chirps, as reported previously70, were used. The stimulus started at gray level, followed by a 1-s bright step and sinusoidal luminance changes over 8 s each, first with increasing amplitude at 2 Hz and then fixed full amplitude but frequency modulated (0 to 8 Hz). For determining direction selectivity, sinusoidal gratings of 0.1 cycles per degree spatial frequency, and 2 cycles per second temporal frequency, were presented at full contrast moving in 8 or 16 directions in randomized order. Gratings were presented stationarily for 3 s and then moved for 7 s in the current direction. To test for retinotopy, a dark bar with length spanning the screen and width of 25° was moved over gray background at 22.5° s−1 in 8 directions perpendicular to the bar orientation for 7 s with 3-s interval between presentations. Full-field flash responses were determined by presenting either dark or white 1-s full-field flashes from gray baseline at a pseudorandom interval of 5 to 10 s. Pseudosaccade stimuli consisted of vertical gratings with 0.08 to 0.25 cycles per degree (cpd) spatial frequency, or random checkerboard patterns with 4° to 12° visual angle checker size, which were presented on the screen. At pseudorandom intervals between 3 and 6 s, the full screen texture moved in a random horizontal direction over 0.08 s by a median of 5° (3° to 30°) visual angle. The distribution of such pseudosaccadic image shifts approximately matched those from actual saccades, as determined from head-fixed population spontaneous saccade statistics.

In vivo axonal terminal imaging analysis

Imaging data were motion corrected and ROI segmented with suite2p (v.0.10.0)79 followed by a manual curation step based on morphological and activity shape. Note that multiple axonal ROIs can originate from the same neuron. Further analysis was performed in MATLAB (MathWorks). dF/F0, where dF is the fluorescence change over time and F0 the baseline fluorescence, was estimated as done previously70,80, by subtracting neuropil contamination with a factor of 0.5, defining F0 baseline as the 8th percentile of a moving window of 15 s (ref. 81) and finally subtracting and then dividing the fluorescence trace by the median of the same 15-s window. The fluorescence SNR was defined for each ROI by dividing the 99th percentile of the dF/F0 trace (‘signal’) by the standard deviation of its negative values after baseline correction (‘noise’). Only axonal segments with a fluorescence SNR ≥ 5 were included in further analysis.

To estimate modulation by visual and behavioral stimuli, a suite of stimuli (full-field flashes, frequency modulation, moving bar, moving full-field grating) was presented and a range of behaviors (locomotion, pupil size, saccades) were sampled. Modulation indices were computed by (Fresponse − Fbaseline)/(Fresponse + Fbaseline), F being the average dF/F0 value in a baseline or response window, respectively. For moving bar and grating stimuli, the baseline window was defined as [1.5, 0.1] s before movement start, and the response window from 0.25 s after stimulus start until the end of the stimulus. For full-field luminance modulation, baseline was between 1.3 s before and until frequency modulation start, and response was the time of frequency modulation presentation. For 0.5-s black or white full-field flashes, baseline window was 1 s before flash onset, and response window was 1 s following flash onset, including OFF responses. For saccades and locomotion onset analysis (Fig. 3k), [1.5, 0.5] s before respective onset was used as baseline and [0, 1] s following onset was used as response window. Significance of modulation was determined by two-sided Wilcoxon signed rank tests. Significance of correlations was determined by 5,000 repeats of randomly shifting the behavioral trace between 120 s and 1,200 s and computing correlations for shuffled datasets. Significance was then determined by calculating the proportion of shuffles with more extreme correlation values than the actual data. Boutons from either vLGN or retina were included in further analysis if they showed Bonferroni-corrected significant (P < 0.01) modulation or correlation to at least one of eight tested conditions. Note that not all recordings included all stimulus sections. These inclusion criteria removed 102,833 of 322,854 vLGNs and 22,646 of 101,376 RGCs over all recorded boutons from further analysis (Figs. 3 and 5e–h and Extended Data Figs. 3 and 4). To illustrate the diversity of responses to ‘chirp’ stimuli (Fig. 3e and Extended Data Fig. 3e), only boutons with an SNR > 0.35 are shown70. SNR was calculated as previously reported as the temporal variance of the mean across trials divided by the mean of the temporal variance of each trial. To illustrate the behavioral modulation of vLGN boutons (Fig. 3k), only boutons with at least ten trials in each condition (saccades, locomotion, pupil) are shown.

For pseudosaccade analyses (Fig. 5e–h), the baseline window was defined as [0.5, 0.1] s before and the response window as [0, 0.5] s following the pseudosaccade or saccade. Only saccadic events separated by at least 0.75 s from other saccadic events were included to avoid cross-contamination. Population synchrony (Extended Data Fig. 4a) was determined as variance of the population mean divided by the population mean of individual bouton variance. Similarly, individual bouton variance explained (Extended Data Fig. 4b) was determined by z-scoring bouton activity, and computing remaining variance after subtracting population mean (varexp = 1 −var[act − <act>population]t, with act being z-scored bouton activity). To determine direction selectivity and preferred directions, 1 − circular variance and vector sums were used82. To demonstrate preferred direction distribution (Fig. 3f and Extended Data Fig. 3f), only boutons with significant direction tuning, P < 0.01 (10,000-fold shuffled direction label test), were plotted. To plot retinotopic alignment of bouton responses (Fig. 3g and Extended Data Fig. 3g), mean bouton responses to dark bars moving in nasal-temporal direction were determined. The centroids were projected into a set of one-dimensional axes, rotated at angles from 0° to 180° with the increment of 5° and binned at 20 µm. The responses of boutons within each bin were averaged. The axis that yielded the maximal correlation of the binned response peak-latency with the horizontal position of the bar was used for the alignment and the corresponding binned responses are shown in Fig. 3g and Extended Data Fig. 3g. To determine cross-correlation timing (Extended Data Fig. 4d,e), lag time of cross-correlation maximum was determined and the bouton included if peak lag was within [−3, 3] s and P < 0.01 (random shift test, see above). To disentangle independent locomotion and pupil size contributions to bouton correlations (Extended Data Fig. 4f–i), correlations to pupil size were separately computed for stationary periods (0.25-s window median filtered locomotion speed <1 cm s−1). Due to large sample sizes, comparisons between bouton populations yield arbitrarily low P values. In these cases, mean and standard deviation of the difference are reported alongside.

Visual cliff setup

The visual cliff paradigm was performed in a black walled 50 × 50-cm2 acrylic box with 80-cm height, covered with transparent (5-mm thickness) and surrounded by black acrylic walls (~25-cm height). The illusory platform (25 × 25 cm2) was created by gluing a paper-printed checkerboard pattern to the bottom and adding a matching black acrylic border (width: 1 cm, height: 0.5 cm) to the surface of the transparent acrylic surface. Interior walls beneath the transparent surface as well as the box floor were covered with a high-contrast checkerboard pattern (2.5 × 2.5-cm2 black and white squares) so that all edges were aligned. At 1 d before experiments, mice were shortly anesthetized with isoflurane (5%) and vibrissae clipped next to the mystacial pad with surgical scissors. Once recovered, mice were returned to the maintenance cage. A camera (Basler, cat. no. acA1920-150um) with a fixed focal length objective (Edmund Optics, f = 50 mm, cat. no. 59-873) was located above the middle point of the arena to cover all movements. Camera control and recordings were obtained using a custom-made script in Python. Each mouse was recorded for 30 min while freely roaming in the arena, only the first 10 min of which were included in further analysis.

Visual cliff analysis

The head and body were tracked using a custom-trained network in DeepLabCut71. Ear tag labels were used for trajectory analysis as they were the most reliable. Video frames were cropped to the arena size and scaled to 1,000 × 1,000 pixels for consistency. Platform area was defined as the bottom-left 25-cm quadrant of the arena, and the remainder as cliff area. The cliff avoidance index (Fig. 5d,e and Extended Data Fig. 5h,i) was computed as AI = (tplatform − tcliff)/(tplatform + tcliff), where t is the time spent in the platform or cliff normalized by the area of these regions. To compute the avoidance index, a 10-cm strip along the walls was excluded (Fig. 5b). Aborted exits (Fig. 5f,g and Extended Data Fig. 5j) were counted when a mouse crossed the platform boundary from inside the platform but then reversed the direction of movement normal to the platform boundary. Aborted exit trajectories were extracted for 4 s around these timepoints and multiple aborted exits removed.

vLGN inactivation physiology

To quantify responses to saccades (Fig. 4g–i), the recordings during oscillating random checker stimulus were used. A random checker pattern (8° per checker) was oscillating sinusoidally by 17° in the horizontal direction for 450 cycles per recording, analogous to previous reports52. Up to two sessions were recorded per mouse; however, due to synchronization problems, some recordings had to be discarded. Only the first available recording per animal was used for the analysis. Saccades were identified as described above. Zeta-test76 was used to identify saccade-responsive neurons. To test optokinetic reflexes (Fig. 7a–e), the same stimulus was used. The starting pupil position per cycle was subtracted from the pupil position traces (Fig. 7c). Saccadic events (Fig. 7d,e) were identified as pupil displacements above 2.5° per frame. To determine luminance change responses, 1-s full-field flashes of three different intensities (low, medium and high) were presented in a random order. Pupil size response speed (Fig. 7g,h) was determined by estimating the slope of a linear fit to the relative pupil area change (baseline subtracted, fitted to the pupil values 0.2 s after flash onset). Quiescent and locomoting states (Fig. 7i–k) were identified as forward locomoting speed below and above 5 cm s−1. Zeta-test76 was done to detect visually responsive neurons, and peristimulus time histogram for white and black flashes was computed (Extended Data Fig. 5d–i).

Euthanasia and histology

Mice were dosed with a 750–1,000 mg kg−1 mixture of ketamine/xylazine and transcardially perfused with PBS, followed by ice-cold 4% paraformaldehyde in PBS. Brains were carefully extracted and post-fixed overnight, cryoprotected with sucrose 30% and sectioned at 60 μm using a sliding microtome (Leica, cat. no. SM2010). Sections were collected in three series. The first series was used for signal amplification of the respective vector. Briefly, sections were incubated with PBST (Triton 0.3%) solution containing 5% donkey normal serum and one or more antibodies (goat Anti GFP, cat. no. ab6673, Abcam, diluted 1:2,000; rabbit anti-RFP, cat. no. 600-401-379, Rockland, diluted 1:1,000) overnight at 4 °C, followed by secondary fluorescent antibodies (Donkey anti-goat-488, cat. no. ab150129, Abcam, diluted 1:1,000; Donkey anti-rabbit-594, cat. no. R37119, Thermo Fisher, diluted 1:1,000) at room temperature for 1 h. Sections were mounted on slides and coverslips with custom-made mowiol.

Confocal microscopy

Brain sections were imaged with a Nikon CSU-W1 spinning disk confocal microscope. All images were processed with FIJI (ImageJ).

Statistics

Analyses were performed in custom-written MATLAB (MathWorks) and Python scripts. Nonparametric tests used are defined in the figure legends. All statistical tests are reported in the text and appropriate figure legends (*P < 0.05, **P < 0.01, ***P < 0.001). In bar plots the mean ± s.e.m. are shown, unless otherwise stated.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.