A common cause for nystagmus in different congenital stationary night blindness mouse models

In Nyxnob mice, a model for congenital nystagmus associated with congenital stationary night blindness (CSNB), synchronous oscillating retinal ganglion cells (RGCs) lead to oscillatory eye movements, i.e. nystagmus. Given the specific expression of mGluR6 and Cav1.4 in the photoreceptor to bipolar cell synapses, as well as their clinical association with CSNB, we hypothesize that Grm6nob3 and Cav1.4‐KO mutants show, like the Nyxnob mouse, oscillations in both their RGC activity and eye movements. Using multi‐electrode array recordings of RGCs and measurements of the eye movements, we demonstrate that Grm6nob3 and Cav1.4‐KO mice also show oscillations of their RGCs as well as a nystagmus. Interestingly, the preferred frequencies of RGC activity as well as the eye movement oscillations of the Grm6nob3, Cav1.4‐KO and Nyxnob mice differ among mutants, but the neuronal activity and eye movement behaviour within a strain remain aligned in the same frequency domain. Model simulations indicate that mutations affecting the photoreceptor–bipolar cell synapse can form a common cause of the nystagmus of CSNB by driving oscillations in RGCs via AII amacrine cells.


Introduction
Patients with congenital stationary night blindness (CSNB) often have congenital nystagmus as well (Bijveld et al., 2013;Pieh et al., 2008).These involuntary, oscillating eye movements occur shortly after birth and vary from pendular to jerk nystagmus (Optican & Zee, 1984;Pieh et al., 2008).Previously, we reported that Nyx nob mice, a CSNB mouse model (Gregg et al., 2003;Pardue et al., 1998), have a pendular nystagmus that results from the oscillatory output of retinal ganglion cells (RGCs; 0 Maj-Britt Hölzel obtained both her Bachelor of Science in Biology (major) and Chemistry (minor) and her Master of Science in Biology at the Carl von Ossietzky University Oldenburg (Germany), where she discovered her passion for the retina and electrophysiology.She is now a PhD student in the Retinal Signal Processing department at the Netherlands Institute for Neuroscience in Amsterdam, where she investigates the retinal origin of congenital nystagmus.Wouter Kamermans obtained his bachelor's degree in physics at Fontys University of Applied Sciences.He then pursued a master's degree in computational modelling and control system engineering at HAN University of Applied Sciences, where he further developed his interest in complex control systems and discovered a passion for control systems in biological contexts.Currently, he is working towards a PhD in neuroscience at the Netherlands Institute for Neuroscience, where he is studying the control systems of eye movements.Winkelman et al., 2019).This oscillatory firing of RGCs appears to be driven by A II amacrine cells (A II ACs), which are probably depolarized outside of their normal working range due to the lack of ON-bipolar cell (ON-BC) input (Winkelman et al., 2019).
The photoreceptor to ON-BC synapse is a highly specialized metabotropic glutamatergic synapse (Fig. 1).Depolarization of a photoreceptor opens its Ca v 1.4 calcium channels and the resulting influx of Ca 2+ into its synaptic terminal ultimately leads to glutamate release (Schmitz & Witkovsky, 1997).The released glutamate diffuses across the synaptic cleft and binds to the postsynaptic metabotropic glutamate receptors (mGluR6; Masu et al., 1995).Activation of this receptor triggers a G-protein-coupled cascade, which leads to the closure of TRPM1 channels resulting in ON-BC hyperpolarization (Morgans et al., 2009;Shen et al., 2009).The function of this synapse depends crucially on the close interaction of all the synaptic proteins involved.
Mutations of the various proteins involved in this synapse will each affect the ON-BC differently.Postsynaptically, there are varying degrees of interdependency between the elements of the signalling cascade.For instance, elimination of mGluR6 disrupts both the expression levels and subcellular targeting of most signalling molecules so far examined.By contrast, when the scaffolding protein nyctalopin is lost, TRPM1 is no longer correctly trafficked and localized to the dendritic tips but other elements of the signalling cascade remain unaffected (Cao et al., 2009(Cao et al., , 2011;;Gregg et al., 2007Gregg et al., , 2014;;Martemyanov & Sampath, 2017;Pardue et al., 1998;Pearring et al., 2011;Ray et al., 2014).In Grm6 nob3 mice (Maddox et al., 2008), the mGluR6 receptors are not functional, which leads to open TRPM1 channels depolarization of the ON-BCs (Ishii et al., 2009;Koike et al., 2010;O'Connor et al., 2006;Tagawa et al., 1999).On the other hand, mutations that affect the Ca v 1.4 channel impair presynaptic signalling while leaving postsynaptic signalling intact (Specht et al., 2009), photoreceptors do not release glutamate and, as under light conditions, the mGluR6 receptor is not activated and the TRPM1 channels are open.This leads to a depolarization of the ON-BCs.These different conditions will probably lead to different membrane potentials of the ON-BC and possibly to slightly different phenotypes of CSNB.
There are several CSNB mouse models, each with a different photoreceptor to ON-BC synapse-specific protein mutated (Zeitz et al., 2015).As mutations in each of these different proteins probably affect the ON-BC differently, it is possible that each mutation presents with a phenotypically distinct nystagmus.To test this idea, we studied eye movements and RGC activity in two CSNB mouse models: Ca v 1.4-KO and Grm6 nob3 mice and compared the results with those of Nyx nob mice.We found that all CSNB mouse models had a disturbed optokinetic response (OKR), nystagmus and oscillating RGCs.However, each mouse model displayed phenotypically distinct features including different eye movements and RGC oscillation frequencies.Furthermore, we found that flashes of light at various intensities, which should polarize the A II ACs to differing degrees, changed the oscillation frequency of Nyx nob RGCs.Model simulations show that these phenotypical changes can indeed be generated by exclusively changing the ON-BC input.

Animals
All animal experiments were carried out under the responsibility of the ethical committee of the Royal Netherlands Academy of Arts and Sciences (KNAW) acting in accordance with the European Communities Council Directive of 22 September 2010 (2010/63/EU).The experiments were performed under licence numbers AVD-801002016517 and AVD-80100202115698, issued by the Central Committee for Animal Experiments of the Netherlands.
Grm6 nob3 and Nyx nob mice were obtained from the McCall lab (University of Louisville, Louisville, KY, USA) and Ca v 1.4-KO mice from the Koschak lab (University of Innsbruck, Innsbruck, Austria).All mice were in a C57BL/6JRj background either with or without a GFP label on ON-dsRGCs coding for upward motion (SPIG1 + mice; Yonehara et al., 2009).Since the Nyx mutation is X-linked, only male mice in the age range 5-71 weeks were used for the experiments.Room lights were timed on a 12/12 h light-dark schedule, and experiments took place during daytime.Mice had ad libitum access to food and water.
M.-B.Hölzel and others J Physiol 601.23

Eye movement recordings
Surgical preparation.Before the start of experiments, adult animals were equipped with a head-fixation pedestal, an anodized aluminium bit with an integrated magnet, attached to the parietal and frontal bones of the skull using dental cement (Super Bond C&B, Sun Medical, Japan).Surgery was performed under general isoflurane/O 2 and topical anaesthesia (bupivacaine).Analgesia was offered by subcutaneous injection of carprofen (5 mg/kg).The recovery time was at least 2 days after pedestal surgery.During experiments, the animal was placed head-fixed in the experimental setup using a custom-made adapter, which allowed panoramic vision.
Optokinetic stimulus setup.Optokinetic stimuli were displayed on two Benq XL2420t high-performance monitors (120 fps, gamma-1.745) that were placed in V-formation around the animal.The closest distance between the screen surface and the mouse head was 16.5 cm.Screen dimensions were 56.9 × 33.8 cm (combined field of view: 240 × 50°).For measurements in darkness, the displays were switched off.Mickelson contrast of the grayscale grating stimuli was 99.67%.Visual stimuli consisted of sine wave gratings (mean luminance: 71.6 cd/m 2 ) and homogeneous grayscale images (mean luminance: 71.6 cd/m 2 ).
Eye movement recordings.Eye movements were recorded using an infrared video tracking system (JAI RM-6740CL monochrome CCD camera, 200 fps, 850 nm illumination).Pilocarpine (2%) eye drops were used to reduce pupil size in case the pupil was too large to track.Online image analysis extracted the location of pupil edges and corneal light reflections from each frame using custom-built software for LabVIEW (National Instruments, Austin, TX, USA).Eye position was computed from the relative distance between pupil centre and corneal reflections of the infrared LEDs (Stahl et al., 2006) and pupil size (Stahl, 2002;Stahl et al., 2000).Epochs containing saccades, eye blinks and motion artefacts were excluded from analysis.Eye velocity was smoothed using a Gaussian smoothing kernel with an SD of 7.5 ms (25 Hz cut-off).All OKR measurements were done monocularly, during which the contralateral eye was covered by a miniature blackout cap.
Analysis of eye movement recordings.The power spectral density (PSD) was computed from angular eye velocity using Welch's method with a 4 s window length, 75% overlap between windows and a Hann window function.

Multielectrode RGC recordings
Retinal dissection.Multielectrode RGC recordings were performed as described previously (Hölzel et al., 2022;Winkelman et al., 2019).After 60-75 min of dark adaptation, mice were sedated using a mixture of CO 2 /O 2 and ultimately killed by cervical dislocation.All procedures were carried out under dim red light.The eyes were extracted from the eye socket and placed in room temperature Ames' medium (Sigma-Aldrich, St Louis, MO, USA).Next, the cornea and lens were removed by making an insertion around the ora serrata using fine spring scissors.As much vitreous humour as possible as well as the sclera were removed using fine forceps.Four small insertions were made and the retina was flat mounted on a filter paper annulus (1 mm inner radius; 0.8 μm hydrophilic MCE MF-Millipore membrane filter, Merck Millipore, Tullagreen, Ireland).The retina was then mounted photoreceptor cell side up on a perforated 60-electrode MEA chip (60pMEA200/30iR-Ti, Multichannel systems, Reutlingen, Germany) in a recording chamber mounted on a Nikon Optiphot-2 upright microscope and viewed under infrared light with an Olympus 2× objective and video camera (Abus TVCC 20530).During the experiment the retina was continuously superfused with Ames' medium gassed with a mixture of O 2 and CO 2 at a pH of 7.4 and a temperature of 35-36°C at a flow rate of 1.5 ml/min.Before the recording started, an acclimatization period in the dark for 15 min was given to ensure stable recordings.
Data acquisition.The extracellular RGC activity was acquired using MC rack (Multichannel systems) at a sampling frequency of 25 kHz.The data were then zero-phase bandpass filtered (250-6250 Hz) with a fourth-order Butterworth filter in Matlab (Mathworks, Natick, MA, USA).Subsequently the spiking activity was sorted manually into single unit activity using the Plexon offline sorter (Plexon, Dallas, TX, USA) based on the first two principal components versus time.For the extraction of spikes from the background, a spike detection amplitude threshold of >4σ n was used as criteria.Here, σ n is defined as an estimation of the standard deviation of background noise and x is the bandpass-filtered signal (Quiroga et al., 2004): Optical stimulator.Full field white light flashes were generated using Psychophysics Toolbox Version 3 (Brainard, 1997) and presented to the photoreceptor side of the retina with a custom-modified DLP projector (for details of modification see Winkelman et al., 2019;Light Crafter 4500, Wintech, Carlsbad, CA, USA).For wild-type (WT), Nyx nob and Ca v 1.4-KO retinas, the full field light flash had a duration of 500 ms.For Grm6 nob3 mice, the flash duration was extended to 2000 ms to accommodate for the lower RGC oscillatory frequencies present in this mouse model.In all cases the light flash was preceded by a 500 ms period of darkness and followed by a 1000 ms period of darkness and the stimulus was repeated 100 times.White light stimuli consisted of equal quantal output of the red (625 nm), green (530 nm) and blue (455 nm) LEDs.The maximal light intensity was 8.60 10 19 quanta/m 2 /s.When studying the effect of different light intensities on the Nyx nob RGC oscillation frequency, two light intensities were used: 9.02 × 10 18 and 8.60 × 10 19 quanta/m 2 /s.Data analysis.To identify spontaneous oscillatory RGC activity, the firing activity in the dark was recorded for a period of at least 15 min.For each cell the activity was binned in 1 ms bins, and then divided into 5 s non-overlapping epochs.Each epoch was baseline subtracted and the autocorrelation and PSD determined.Based on these data the average autocorrelation and Welch's PSD were calculated for each cell.The RGCs of one of the five Ca v 1.4-KO retinas oscillated at much lower frequencies than the other four and was therefore excluded from the analysis.Mean light responses were calculated over the 100 stimulus repetition for each cell.Five hundred milliseconds of the mean light response of each cell that occurred after 50 ms after the light onset was zero padded to 2 s and used to calculate Welch's PSD.
The median oscillation frequency of the eyes was plotted against the median oscillation frequency of the RGCs and fitted with a linear regression model.
Spike train synchrony.Synchronization between spike trains was quantified using SPIKE-Synchronization.This time scale-independent and parameter-free metric uses Inter-Spike-Interval derived coincidence windows to generate a similarity score.A similarity score of 1 occurs only if each spike in a train has a matching spike in the other train(s) whereas a score of 0 occurs only if the spike trains do not contain any coincidences (for more details see: Kreuz et al., 2015;Satuvuori et al., 2017).The analysis routine was used within cSPIKE, a publicly available Matlab-based spike train analysis software package (https: //thomas-kreuz.complexworld.net/source-codes/cspike).
We generated two distributions of similarity scores for each retina.The first distribution was of similarity scores between spike trains of spontaneous activity occurring in the dark for units within the same retina.The spontaneous activity was divided into 5 s long non-overlapping periods and for each of these periods the similarity scores for each unit vs. every other unit wee determined (e.g. a 15 min recording of 30 units resulted in 30 × 30 × 180 similarity scores).For these same units we also generated a second distribution of similarity scores, this time comparing their spike trains with those of units from different retinas in the other treatment groups.For example, units from a single Ca v 1.4-KO retina were compared with unrelated units from all the Nyx nob , WT and Grm6 nob3 retinas, again using 5 s long non-overlapping periods.This second population of similarity scores was used to estimate the chance distribution of similarity scores occurring for units whose activity is unrelated.
For each retina, we quantified the distance between their two distributions of similarity scores using the Hellinger distance.The relative probability distribution of the similarity scores was distributed over 51 equally spaced bins ranging from 0 to 1 in 0.02 increments, and for the two discrete probability distributions P = (p 1 ,….p k ) and Q = (q 1 ,….q k ), their Hellinger distance was then defined as where 1 indicates there is no overlap of the populations and 0 indicates the two populations fully overlap.

Statistical analysis
Statistical analysis was performed using SPSS Statistics 28 (IBM Corp., Armonk, NY, USA) and Origin Pro 8 (Northampton, MA, USA).Since the distribution of the frequencies of eye movement oscillations violated the assumption of normality (Shapiro Wilk), statistical differences were assessed using an independent-samples Mann-Whitney U test.All values are given as median ± MAD (median absolute deviation) and a statistically significant difference was assumed at P < 0.05.Statistical differences in the Hellinger distances were tested by a one-way ANOVA with a post hoc Bonferroni test.To test whether there is a statistically significant effect of light intensity on oscillation frequency we used a Wilcoxon signed-rank test.

Computational model
The schematic shown in Fig. 7Aa below gives an overview of the BC, A II AC and RGC network we simulated.An A II AC is electrically coupled to an ON-CBC and to other A II ACs via gap junctions (Marc et al., 2014).A II ACs receive glutamatergic input from ON-RBCs (Marc et al., 2014) and are inhibited by the OFF-CBC-driven OFF-ACs (Marc et al., 2014).This latter pathway is a so-called crossover inhibition pathway.Light stimulation leads to depolarization of the ON-BCs and thus to depolarization of the A II ACs.In addition, the OFF-CBCs and subsequently the OFF ACs hyperpolarize upon light stimulation and this will lead to a decrease of the inhibition of the A II ACs, further depolarizing the A II ACs.ON-CBCs generate an output to ON RGCs and OFF-BCs generate an output to OFF RGCs.All of the described cell types are distributed in a hexagonal grid.
To reduce the number of free parameters and computation time, we simplified the model (see Fig. 7Ab).The ON-RBCs are simulated by steady current injections since we simulate photopic conditions.The inhibitory input from the OFF pathway is implemented as a steady current injection into the A II Acs.Similarly, the photoreceptor input to the ON-CBCs was simulated as a steady current injection into the ON-CBCs.The output of the ON-CBCs is the input to ON-RGCs and the OFF-RGCs are the inverse of ON-RGCs.
Each ON-CBC contacts one A II AC, and each A II AC in turn contacts one RGC and six A II ACs.The literature shows that one A II AC is contacted by 10 ON-CBCs (Sigulinsky et al., 2020).This physiological connectivity was implemented by scaling the coupling conductances.Next, we present the equations describing the various cell types (see Fig. 7Ab).
A II AC model.The A II AC was simulated as a three-compartment model (Choi et al., 2014) consisting of a soma, an initiation site and a connecting cable (see Fig. 7Aa).The arboreal dendrites are part of the soma.The ionic currents are described by Hodgkin-Huxley type equations.
The sodium current (I Na ) is given by where the gating variables are The M-type potassium current (I M ) is given by where the gating variables are The A-type potassium current (I A ) is given by where the gating variables are Next to these ionic currents there is a leak current The ON-CBCs are coupled to A II ACs by gap junctions.For computational reasons, we simulated only one ON-CBC per A II AC.To account for the fact that 10 ON-CBCs are coupled to one A II AC, we made the coupling conductance seen by the A II ACs 10 times larger than the conductance seen by the ON-CBC.
A II ACs are coupled by gap junctions to each other in a hexagonal network such that each A II AC is coupled to six other A II ACs.The gap junction currents are given by and summed in a total gap junction current per cell I GJ-total .
The membrane potentials of the three compartments were calculated as For the ON-CBC, we used the model from Usui et al. (1996).This model consists of a delayed rectifying potassium current (I Kv ), a transient potassium current (I A ), a hyperpolarization activated current (I h ), a calcium current (I Ca ) and a calcium-dependent potassium current (I KCa ).A two-compartment model was used for the internal calcium concentration of the ON-CBC, which is needed to calculate I KCa .One compartment was close to the membrane and consisted of a fast and a slow calcium buffer, and two exchange currents.The other compartment was within the cell and consisted of only the two calcium buffers.Calcium will flow from one compartment into the other by diffusion.All ionic currents are described with Hodgkin-Huxley-type equations, except for I h which is described with a Markov chain-type model.
The delayed rectifying potassium current (I Kv ) current is described by where the gating variables are calculated as The transient potassium current (I A ) is modelled with Hodgkin-Huxley type equations, and is defined as where 15 The hyperpolarization activated current (I h ) is described with a Marko chain-like model which described the transition between open and closed states.The current is given by where the total open probability is the sum of the open states The open and closed states are calculated as with the transition rate constants defined as The calcium current (I Ca ) is given by with the gating variables where The calcium-dependent potassium current (I KCa ) is given by where the gating variables are described as − 1 The change in calcium concentration close to the membrane ([Ca 2+ ] s ) is described as bhs and the calcium concentration inside the cell ([Ca 2+ ] d ) as with the exchange currents defined as Finally, we added a leak current The gap junction between the ON-CBC and the A II AC is given by We only simulate one ON-CB per A II cell.The membrane potential of the BC is given by

C
dV BC dt = − (I Kv + I A + I h + I Ca + I KCa + I leak ) RGC model.The RGC model is a simple spike-generating mechanism consisting of three ion channels, a sodium current, an M-type potassium current, an A-type potassium channel and a leak current.Additionally to these ion channels there is a synaptic current which is modelled as a current that is dependent on the calcium current of the bipolar cell.The sodium channel, M-type potassium channel and A-type potassium channels are described with the same equation used for the A II ACs.The synaptic current is given by The membrane potential of the RGC is given by For the OFF RGC, the voltage dependency of I synapse is inverted and the leak current is adjusted in order to depolarize the RGC membrane potential: Model parameters for the ion channels and the various capacitances are given in Table 1.The parameters for the calcium buffers are presented in Table 2. Table 3 gives the current injections for the different mouse models, the bias current and the light step current.

Oscillating eye movements in Grm6 nob3 and Ca v 1.4-KO mice compared to those in Nyx nob mice
To investigate whether there is a common cause for nystagmus in CSNB we first measured the eye movements of WT, Grm6 nob3 , Nyx nob and Ca v 1.4-KO mice.Figure 2 shows example raw traces of the horizontal eye position for each of these mice in response to a monocularly presented horizontal moving vertical grating.WT mice show a normal OKR.They successfully follow the grating in the temporal to nasal direction (6-10 s), keep their eyes still when the stimulus is stationary (4-6 s) and do not follow the grating when it moves in the nasal to temporal direction (0-4 s).
On the other hand, the OKRs of Grm6 nob3 , Nyx nob and Ca v 1.4-KO mice were severely impaired and had different characteristics in each mutant.Grm6 nob3 and Nyx nob mice initially follow the gratings moving in the temporal to nasal direction (arrows in Fig. 2), but were unable to maintain this eye movement (also see Winkelman et al., 2019).Furthermore, their eyes tended to drift when the grating stopped moving.The Ca v 1.4-KO mice lacked an OKR altogether but did exhibit a random slow drift in eye position.However, all mutants showed the presence of small-amplitude high-frequency eye movement oscillations, i.e. pendular nystagmus.In this study we will focus only on these small-amplitude eye movement oscillations and not discuss the OKR further.
Using PSD plots, we quantified the nystagmus under four stimulus conditions: darkness, homogeneous illumination, vertical gratings moving horizontally and horizontal gratings moving horizontally (Fig. 3).While the eyes of Grm6 nob3 and Nyx nob mice only oscillate when stimulated with a vertical grating, Ca v 1.4-KO mice show oscillatory eye movements under all stimulus conditions (arrows).Furthermore, the frequency of the eye movement oscillations during the vertical grating condition differs between the three genotypes [median ± MAD; Grm6 nob3 : 3.4 ± 0.38 Hz (N = 12); Nyx nob : 5.0 ± 0.25 Hz (N = 7); Ca v 1.4-KO; 8.0 ± 0.38 Hz (N = 6); Mann-Whitney U test, P Grm6nob3,Nyxnob = 0.001; P Grm6nob3,Cav1.4-KO = 1.8 × 10 −4 ; P Nyxnob,Cav1.4-KO= 0.001).Additionally, the power spectra of all knockout mice also show a low-frequency peak representing a slow eye movement drift, which will not be discussed further.WT mice on the other hand do not exhibit comparable eye movement oscillations although a very small peak at around 5 Hz may be observed sometimes in their power spectrum.
Taken together, the eye movement data shown in Fig. 3 reveal phenotypical differences between the various CSNB mutants.How does this correlate with RGC activity?
Spontaneous oscillation frequencies of Grm6 nob3 , Ca v 1.4-KO and Nyx nob mice differ Previously, we have shown that the RGCs of Nyx nob mice fire in an oscillatory fashion with a frequency closely matching that of their pendular nystagmus.The causality of this relationship was shown using retina-specific pharmacological treatments that changed the oscillation frequency of RGC firing resulting in a similar change in frequency of the nystagmus (Winkelman et al., 2019).To test whether a similar relationship between RGC oscillations and nystagmus exists in the various mutants, we determined their RGC oscillatory firing frequency.

Light-induced oscillations in Grm6 nob3 RGCs
As we showed previously, in Nyx nob mice, individual RGCs oscillate at similar frequencies, but their activity is asynchronous in the dark.Light stimulation phase resets the RGC oscillations, and then this synchronized oscillatory retinal output induces oscillating eye movements (Winkelman et al., 2019).For Grm6 nob3 mice, could light stimulation still induce synchronized RGC oscillations even though their activity is not spontaneously oscillatory in the dark?To answer this question, we studied the responses of RGCs to a light flash (Fig. 5).
As expected, a light flash stimulus induced typical ON and OFF light responses in WT mice without RGC oscillations (Fig. 5A-C).This was not the case for Grm6 nob3 and Nyx nob mice, which lack the RGC ON light response but showed oscillatory activity briefly after the light stimulation onset.In the Grm6 nob3 mouse only 6.2 ± 4.95% (mean ± SD, N = 3 mice) of the RGCs oscillated, in Nyx nob 36.5 ± 13.95% (N = 7 mice) and in Ca v 1.4-KO 22.2 ± 12.89% (N = 4 mice) oscillated.As is also the case for Nyx nob mice, RGCs in the Grm6 nob3 mice showed oscillations directly after stimulus onset.Two groups of RGCs were found that oscillated in counterphase presumably originating from ON-and OFF-RGCs (Fig. 4,5B and C).

RGCs and eye movements oscillated at similar frequencies
Next, we determined how well the oscillation frequency of RGC activity matched that of the eye movements for the Grm6 nob , Nyx nob and Ca v 1.4-KO animals.For this we used the oscillation frequency of RGC activity during the light flash, and of eye movements during the vertical grating, as these were the only conditions in which oscillatory responses occurred in each of the three models.Fig. 5E shows the relationship between the eye movement oscillation frequency and the RGC oscillation frequency.In general, the RGC oscillation frequency was linearly related to the eye movement oscillation frequency, but the RGC oscillation frequency was slightly higher than that of the eye movements.

RGC oscillations in the Ca v 1.4-KO are synchronized under all stimulus conditions
Ca v 1.4-KO mice exhibited nystagmus in the four stimulus conditions (Fig. 3), whereas Grm6 nob3 and Nyx nob mice showed this behaviour only for the vertical grating.This suggests the Ca v 1.4-KO oscillatory eye movements occur independently from a visual stimulus.The Ca v 1.4-KO mice also lacked an OKR response (Fig. 2) and their RGCs did not respond to a light flash stimulus (Fig. 5A-C), which suggests that these mice are blind.How then can the RGCs oscillations in Ca v 1.4-KO mice be synchronized?A possibility is that unlike Grm6 nob3 and Nyx nob mice, the RGCs in Ca v 1.4-KO mice display a degree of inherent synchronization independent of the stimulus condition.
To test this, we assessed to what extent the spontaneous activity of RGCs was synchronized in the dark for each of the mouse models.Using the SPIKE-synchronization metric (https://thomas-kreuz.complexworld.net/sourcecodes/cspike),we determined two synchronization score distributions for each retina (Fig. 4D).The first set of scores measured the synchronization occurring between activity of RGCs from the same retina.The second set was generated by comparing the activity of these same RGCs with that of RGCs from the three other mouse models, giving an estimate of the chance distribution of synchronization scores for the retina.We then quantified the difference between these two distributions using the Hellinger distance (Fig. 4E).
Ca v 1.4-KO retinas had distributions that shifted towards higher synchronization scores for RGCs in the same retina versus unrelated retinas, compared to the three other mouse models (Fig. 4D).Indeed, the lowest Hellinger distance for the Ca v 1.4-KO retinas was greater than the highest Hellinger distance of any other group (Fig. 4E, one-way ANOVA, P WT, Cav1.4-KO = 0.0003, P Grm6nob3, Cav1.4-KO = 0.0001, P Nyxnob, Cav1.4-KO = 0.0003, P WT , Nyxnob = 1, P WT, Grm6nob3 = 1, P Grm6nob3, Nyxnob = 1).Furthermore, only the Ca v 1.4-KO group demonstrated elevated synchronization scores within the same retina.These outcomes indicate that there is a degree of inherent synchronization in the spontaneous activity of Ca v 1.4-KO RGCs that is not present in Grm6 nob3 , Nyx nob or WT conditions.A II ACs as a potential cause for the difference in eye movement oscillations Previously, we have shown evidence suggesting A II ACs are driving the oscillatory responses of RGCs (Winkelman et al., 2019).A II ACs oscillate when they are outside their normal membrane potential range and the oscillation frequency increases with depolarization (Choi et al., 2014).If this is the case then depolarizing Nyx nob A II ACs further should lead to an increase of the oscillation frequency of their RGCs.To test this hypothesis, we recorded the light-induced oscillations of Nyx nob RGCs at two different light intensities (Fig. 6).Consistent with our expectations, we found that when A II ACs are depolarized more strongly by a light flash of higher light intensity, the RGC oscillation frequency increases by about 1.1 ± 0.91 Hz (median ± MAD, n = 92 RGCs, N = 4 mice).This increase in oscillation frequency was statistically significant (Wilcoxon signed-rank test: Z = −4.179,P = 2.9 × 10 −5 , n = 92 RGCs, N = 4 mice).

Computational model
To determine whether the various phenotypes could be generated by only changes in ON-BC input to the A II AC, we constructed a computational model consisting of a grid of hexagonally arranged units consisting of an ON-CBC  A, raster plots comparing the extracellularly recorded spontaneous spiking activity for, in each case, one RGC from WT, Grm6 nob3 , Nyx nob (red) and Ca v 1.4-KO.B, autocorrelation of activity of the same RGC as in A showing oscillatory behaviour visible as periodic variations, only for the Nyx nob and Ca v 1.4-KO mice but not WT or Grm6 nob3 RGCs.C, mean power spectra of the spontaneous activity in WT (n = 209 RGCs from N = 5 mice), Grm6 nob3 (n = 543 RGCs from N = 11 mice), Nyx nob (n = 367 RGCs from N = 8 mice) and Ca v 1.4-KO (n = 135 RGCs from N = 4 mice).D, comparison of similarity scores as a measure for synchrony for spike trains from RGCs from the same versus from another retina.In the Ca v 1.4-KO retina RGCs are synchronized in the dark which is not the case for the other three mice.E, comparison of the Hellinger distance of WT and the three mouse models.* * * P < 0.001.[Colour figure can be viewed at wileyonlinelibrary.com] coupled electrically to an A II AC and synaptically to an ON-RGC.A detailed model description is presented in the Material and methods.
Figure 7Aa shows how the various relevant cell types are connected in the retina.The ON-and OFF-CBCs receive synaptic input from the cone photoreceptors, while the ON-RBC receives its input from the rod photoreceptor.The ON-CBCs are electrically coupled via gap junctions to the A II AC while the ON-RBCs give synaptic glutamatergic input to the A II ACs.The OFF-CBC projects directly to the OFF-RGC and via an inhibitory AC to the A II AC.We simulated a simplified version of this network (see Material and methods; Fig. 7Ab).
Fig. 7Ba-e gives an overview of the model response to a light flash.The heart of the model is the A II AC.When depolarized beyond their normal working range they start to oscillate.Previously, we showed that individual RCGs oscillate at slightly different frequencies (Winkelman et al., 2019).To simulate this we injected each of the A II ACs with a small sustained current, but varied the magnitude of each of the sustained currents randomly.This resulted in all A II ACs oscillating at slightly different frequencies.Fig. 7Ba shows example traces of six A II ACs, all slightly displaced along the y-axis for clarity.The A II ACs oscillations are transmitted via gap junctions to the ON-CBS.Fig. 7Bb shows the response to a light flash of a single ON-CBC (blue line) and a connected ON-RGC (red line).ON-CBCs show small oscillations and as a result the ON-RGC generates a burst of spikes on the upward flank of the CBC oscillations.Fig. 7Bc shows six CBCs, all oscillating with slightly different frequencies.A raster plot of the spikes and the sum of the spikes generated in an ON-RGC are shows in Fig. 7Bd and e.
In the dark, RGCs are oscillating but the oscillations are not synchronized.A simulated light flash with a duration of 4 s presented at t = 4 s causes most RGC oscillations to phase reset and synchronize for a brief period.This mimics the behaviour of the RGC recordings.Finally, the PSD plots in Fig. 7C show that RGCs oscillate at higher frequencies in the light.

The computational model can account for the phenotypical differences
Next, we investigated whether we could simulate the behaviour of the various genetic mouse models by only changing the resting membrane potential of the ON-CBC.The mutations in Nyx nob and Grm6 nob3 mice both affect the photoreceptor to ON-BC synapses, leading to depolarization of the ON-BCs.The Ca v 1.4-KO mutation, on the other hand, blocks the output of the photoreceptors also leading to depolarization of the ON-BCs.Therefore, we hypothesized that the various phenotypical differences in the three mouse models are caused by differing degrees of ON-BC depolarization.Stronger depolarization of ON-BCs leads to depolarization of A II ACs, and increased oscillation frequency in both the A II ACs and RGCs.We tested this hypothesis using the computational model.
To change the resting membrane potential of the model ON-BC we injected a steady current into the model ON-BC.
Figure 8 gives an overview of the results for our simulations of the spontaneous activity occurring in the dark.The RGC spiking activity is shown as raster plots in Fig. 8A and the average spike activity of all RGCs in Fig. 8B.The autocorrelations of the RGC activity (Fig 8C) indicate oscillatory RGC firing for the Nyx nob and Ca v 1.4-KO mice.In contrast to the other mouse models, the Grm6 nob3 model shows no periodic bursting activity but only very high-frequency single spike activity.PSD analysis of RGC firing (Fig 8D ) shows clear peaks at 6 and 10 Hz for the Nyx nob and Ca v 1.4-KO conditions, respectively and no peak for Grm6 nob3 in that frequency range.The activity of RGCs is not synchronized for either Nyx nob or Grm6 nob3 , as evidenced by the lack of any clear periodic pattern in their averaged RGC activity (Fig 8B).However, for Ca v 1.4-KO a periodic pattern is apparent, indicating synchronized firing between the RGCs.This shows that the model can reproduce the behaviour of the RGCs in the dark in the various CSNB models adequately by only changing the ON-CBC resting membrane potential by current injection.
Figure 9 shows the RGC activity for Nyx nob and the Grm6 nob3 directly after a full field light stimulus.The full field light stimulus was simulated by injecting an equal current into all A II ACs. Figure 9A shows the raster plots of the RGCs for Nyx nob and Grm6 nob3 .No data for Ca v 1.4-KO in the light are presented as these retinas are not light responsive.It is clear that the RGCs oscillate in phase and that the oscillation frequency in Nyx nob mice is higher than for Grm6 nob3 .This is similar to what we found experimentally.The Nyx nob data were replotted from Winkelman et al. (2019).C, mean activity during a 500 ms/1 s light flash for the RGCs shown in B. D, mean power spectra of Grm6 nob3 (n = 17, N = 3), Nyx nob (n = 245, N = 7) and Ca v 1.4-KO (n = 95, N = 4) RGCs.E, relationship between the median RGC and eye movement oscillation frequencies for the three mouse models.The linear regression model fitted is y = −1.16+ 0.76x.[Colour figure can be viewed at wileyonlinelibrary.com]

Discussion
In the current study, we show that in all three mouse models with mutations associated with CSNB, i.e.Grm6 nob3 , Nyx nob ,and Ca v 1.4-KO mice, RGCs and eye movements oscillate.The oscillation frequencies of the RGCs and eye movements showed a strong positive linear relationship, with each genotype displaying consistent and aligned phenotypical characteristics.First, the oscillation frequencies associated with each genotype differed, with the lowest frequencies present in Grm6 nob3 mice and highest in Ca v 1.4-KO mice.Second, only vertical gratings induced eye movement oscillations in Grm6 nob3 and Nyx nob mice, whereas for Ca v 1.4-KO mice they occurred in all stimulus conditions.Third, in the dark the spontaneous firing of RGCs in Nyx nob and Ca v 1.4-KO mice oscillated, but did not for Grm6 nob3 mice.Fourth, light-flash stimuli led to synchronized RGC oscillations in Nyx nob and Grm6 nob3 mice, but not in Ca v 1.4-KO mice.Finally, in the dark Ca v 1.4-KO RGC firing displayed a degree of inherent synchronization that was not present for Nyx nob and Grm6 nob3 .Our modelling experiments indicate that these different outcomes may well arise from a single retinal source when taking the A II ACs as a common mediator.

Similar mechanism underlying the nystagmus in the three mouse models
Can our previously proposed retinal origin of nystagmus model account of the present results (Winkelman et al., 2019)?In short, the proposed mechanism works as follows.In Nyx nob mice A II ACs are depolarized outside their normal working range, which makes them oscillate spontaneously.However, due to small input differences in the various A II ACs, they all oscillate with slightly different frequencies and phases.These oscillations are transmitted to the RGCs, including the ON direction selective RGCs (ON-DSRGCs).In the dark, the signals from the ON-DSRGCs to the AOS are asynchronous and so the convergent signal generated is not large enough to induce an eye movement.However, as soon as a light stimulus occurs, the phase values of the activity of the various A II ACs are reset, which synchronizes their oscillations and subsequently those of the RGCs as well.When the AOS integrates these synchronized RGC inputs, the resulting signal is large enough to induce an eye movement.In turn, this eye movement induces a new visual stimulus, which again synchronizes the retinal oscillations and the loop repeats.
What causes the differences between the phenotypes of the three mouse models?
Can our proposal account for the distinct phenotypes of each of the three CSNB mouse models displayed?These distinct features include: (1) different oscillation frequency of each CSNB model, (2) the absence of spontaneous RGC oscillations in the dark for the Grm6 nob3 mice, and (3) the stimulus-independent synchronized oscillations of the RGCs in Ca v 1.4-KO mice?Differences in oscillation frequencies.The A II AC oscillation frequency depends on its membrane potential; the more depolarized A II ACs are, the higher is the oscillation frequency (Choi et al., 2014).This suggests that the A II AC membrane potential in the various CSNB mouse models is more depolarized than in WT, with Ca v 1.4-KOA II ACs being the most depolarized and Grm6 nob3 the least.A II ACs are ON cells meaning that they depolarize with a reduction of glutamate release by the photoreceptors either via a direct ON-BC input or via a crossover inhibitory input from the OFF-BC pathway.Blocking all glutamate release from the photoreceptors, as is that case in the Ca v 1.4-KO mice, would lead to a strong depolarization of the A II ACs and thus a high oscillation frequency.Indeed, the oscillation frequency in the Ca v 1.4-KO animals is highest.In Nyx nob animals the A II ACs seem to be less depolarized, because the crossover inhibition from the OFF-BCs is still intact, which keeps the A II ACs somewhat more hyperpolarized compared to the Ca v 1.4-KO mice.Our results also suggest that the A II ACs in Grm6 nob3 are less depolarized than they are in Nyx nob .We can only speculate why this may be so, but presumably the absence of the mGluR6 receptor affects the synapse differently than does the absence of the scaffolding protein nyctalopin.Unfortunately, the available literature regarding the exact resting membrane potentials of ON-BCs in the CSNB mouse models is contradictory (Ishii et al., 2009;O'Connor et al., 2006;Tagawa et al., 1999;Xu et al., 2012).However, following the arguments of Ishii et al. (2009), O'Conner et al. (2006) and Tagawa et al. (1999), it seems likely that ON-bipolar cells in the CSNB mouse models studied are constitutively depolarized.

Why are Grm6 nob3
RGCs not oscillating spontaneously in the dark?We propose that in the dark the Grm6 nob3 A II ACs are resting at potentials just below that required to initiate spontaneous membrane potential oscillations.However, when stimulated with light they depolarize further and start to oscillate.This could explain why we and others (Takeuchi et al., 2018) find so few oscillating RGCs in the Grm6 nob3 retina compared with the other CSNB mouse models.One prediction of our proposed mechanism is that the oscillation frequency of the RGCs depends on the light intensity.Indeed, the oscillation frequency in the Nyx nob RGC increases with a higher light intensity and therefore the membrane potential of the A II AC.
Why do RGCs and eyes in the Ca v 1.4-KO mice oscillate under all stimulus conditions?Ca v 1.4-KO mice are blind and our results indicate their A II ACs are strongly depolarized and oscillate with a degree of inherent synchronization.This suggests that the coupling strength of A II ACs is higher in the Ca v 1.4-KO retinas than in the other CSNB mouse models.The electrical coupling strength of neurons depends on the ratio of the gap junction conductance and the input conductance of the coupled neurons.
In Ca v 1.4-KO, crossover inhibition from the OFF pathway to the A II ACs is also impaired and thus the associated inhibitory conductance will be closed.Accordingly, the input resistance of the A II ACs in Ca v 1.4-KO is likely to be higher than for the other CSNB mouse models.The reciprocal decrease in input conductance shifts the ratio towards the gap junction conductance, increasing the electrical coupling strength between coupled A II ACs.This increase in coupling between oscillating A II ACs in Ca v 1.4-KO enables the occurrence of an entrained oscillation state across the A II AC network.

Model simulations show that the different oscillation frequencies in the various mutations originate in ON-BC membrane potential differences
We constructed a model consisting of a grid of hexagonally arranged units consisting of a ON-CBC that is electrically coupled to an A II AC and synaptically to an ON-RGC (Fig. 7Ab).To account for the slight differences in oscillation frequency of the RGCs within each genotype, each A II AC received a slightly different bias current such that their resting membrane potentials differed slightly.First, we tested whether it is possible to simulate the phenotypical differences in the spontaneous oscillatory behaviour of RGCs in the dark for the three mouse models.Variable amounts of current were injected into the ON-CBCs to manipulate their resting membrane potential.The resting membrane potential was depolarized the least to simulate the Grm6 nob3 condition and the most for the Ca v 1.4-KO simulation.Under these settings the Grm6 nob3 RGCs did not oscillate while the Nyx nob and Ca v 1.4-KO RGCs oscillated at 6 and 10 Hz respectively.Hence, by manipulating a single parameter, the model reproduced the experimentally observed RGC behaviour in the dark.
Next, we tested if the difference in resting membrane potential of ON-BCs could also reproduce the phenotypical differences in the RGC light flash responses of the various mouse models.As the experiment data show that light flash stimuli synchronize the RGC oscillations (Figs 5 and 6;Winkelman et al., 2019), we first confirmed that the model could replicate this behaviour.Figure 7B shows the activity of the various model cells before, and during, a stimulated light stimulus.Before the stimulus, the A II ACs are spontaneously oscillating independently because they are each at a different resting membrane potential and consequently the averaged activity of the RGCs shows no oscillations (Fig. 7Be, t < 1000 ms).As the ON pathway is not light sensitive in CSNB, we assumed A II AC light responses were generated by input from an inhibitory AC driven by OFF-BCs, which we simulated as a current injection.Injecting an equal current pulse into all A II ACs phase resets their own oscillations, and consequently leads to synchronized RGC activity.Now the mean activity of the RCS showed clear oscillations (Fig. 7Be, 1000 ms ≤ t ≤ 1500 ms).Furthermore, the oscillations in the RGC mean activity diminished over time (Fig. 7Be, t > 1500ms) indicating this synchronization is gradually lost.This occurs because though A II AC oscillations are initially in phase, their individual oscillation frequencies are still different and over time this leads to phase differences and loss of synchrony.This behaviour matches the experimental data.
Finally, we simulated the response to a light flash in the model phenotypes for Nyx nob and Grm6 nob3 (Fig. 9).This simulation replicated several key characteristics observed in the experimental data (Fig. 5).The Grm6 nob3 RGCs showed oscillatory firing during the light flash simulation, which dampened out sooner and oscillated at a lower frequency compared to the Nyx nob RGCs.Thus, the model indicates again that the phenotypical differences we observed between the different mouse models can be accounted for by only depolarizing the membrane potential of the ON-BCs.

Alternative mechanisms leading to RGC oscillations
Our data, both here and in earlier publications (Winkelman et al., 2019), and modelling work support our proposed role of A II ACs as the driver of the RGC oscillations we observe in CSNB mouse models.However, alternative mechanisms may also account for the RGC oscillations.In the following section we discuss possible alternatives and issues, as well as our rational for why our A II AC proposal is still the most parsimonious explanation.
Could each mutation have a separate and distinct working mechanism for the induction of nystagmus?
Since the genes studied are ON-BC-(Nyx and Grm6) or photoreceptor-specific (Ca v 1.4), alternative mechanisms should be based on a non-functional modulation of the TRPM1 channels, as this is the first common target in the system in all the mouse models.This strongly limits the number of alternative mechanisms leading to oscillating RGCs and consequently the nystagmus.Changes in activation of the TRPM1 channel lead to changes in the membrane potential of the ON-BC.Having separate mechanisms for the various mutations would imply that subtle changes of the ON-BC membrane potential would activate vastly different pathways.There is no evidence for such a demarcation point in the ON-BC responses to the RGCs.Furthermore, such a system would be very instable.It is therefore highly unlikely that the various mutations work via different mechanisms.
Could there be more than one oscillator in the retina?
What would these other oscillators be?The oscillations in both the retinal degeneration (rd) mouse models and the CSNB mouse models seem to be widespread over the various retinal layers, occurring in many ACs and BCs, making it unlikely that they originate in the RGCs themselves.Furthermore, to the best of our knowledge, there is no evidence, either functional or morphological, to suggest that RGCs feed back into the A II AC system.The most likely site of the oscillators would be the ACs or networks that involve ACs.Although we cannot exclude that other AC types are driving the oscillations as well, our data are consistent with A II ACs as the single origin of the oscillation.
Based on pharmacological experiments, Tu et al. (2016) suggested that the oscillations in the RGCs are driven by two distinct oscillators in the retina.Our work, on the other hand, suggests that A II ACs form a population of oscillators with slightly different oscillation frequencies that drive the RGC oscillations.In the dark the A II ACs oscillate asynchronously, all with their own frequency.Light stimulation can synchronize the A II ACs, which is a property of the ON-BC -A II AC network.The synchronized activity of the A II ACs is then transmitted to the ON-and OFF-BCs via two pharmacologically different pathways.The pathway to the ON-CBCs is routed via gap junctions to the ON-CBC terminal, which also receives strong GABAergic inhibition (Marc et al., 2014).The pathway to the OFF-BCs runs via glycinergic input to the OFF-BCs.So, some pharmacological experiments may block the output of the A II ACs to the ON-pathway while others may block the output to the OFF-pathway.Thus, the pharmacological experiments of Tu et al. (2016) can also be interpreted as nicely revealing the difference in pharmacology of the output pathways of the A II ACs and not per se as evidence for more than one oscillator.
Based on these considerations, the blocking/Cx36-knockout experiments we performed previously in Nyx nob mice (Winkelman et al., 2019) and the modelling work presented in this paper, we propose that the A II ACs are exclusively the oscillators that drive the nystagmus.Consistently, by solely changing the membrane potential of the ON-BCs in our model while keeping all other parameters the same, we can account for the Nyx nob , Grm6 −/− and the Ca v 1.4-KO results.

Causal relationship between RGCs and eye movement oscillations
One hypothesis for the cause of nystagmus is a mal-adapted oscillating OKN loop (see Kamermans et al., 2023 for a discussion).Opening such a loop should stop oscillation throughout the loop and abolish nystagmus.Recently, we presented evidence that nystagmus in CSNB is not caused by such an oscillating loop (Winkelman et al., 2019), but instead by intrinsic retinal oscillations.The present paper adds to that conclusion in a rather direct way.
Consistent with previous reports (Michalakis et al., 2014), our results indicate Ca v 1.4-KO mice are blind and as such their visual impairment is greater than is found in human patients with similar mutations.Since the photoreceptor output is absent with this mutation, it is interesting to compare results of the present study with those of the DNQX + d-AP5 experiments in Nyx nob mice we described earlier (Winkelman et al., 2019).This drug cocktail blocks all RGC inputs.In that paper we showed that applying this drug cocktail abolished all RGC oscillations in the isolated retina.When this cocktail was injected into the eyes of Nyx nob mice, eye movement oscillations stopped.This result is consistent with our hypothesis of a retinal origin of nystagmus in CSNB.However, this experiment does not per se discriminate between two possible conditions.The result could have occurred because blocking RGC inputs opened the oscillating OKR loop or because a retinal oscillator was silenced.
For the Ca v 1.4-KO mice, since the photoreceptor output is blocked, their mutation also opens the OKR loop but now at the site of the photoreceptors, while leaving the RGC inputs intact.Hence, the output from oscillating A II ACs is not blocked and still reaches the RGCs.The finding that these mice still have oscillating eye movements shows that the oscillating A II ACs are driving the nystagmus.Thus, although we cannot exclude some secondary modifications of the eye movement system downstream, these experiments confirm the retinal origin of the nystagmus in the CSNB mouse models and a causal relationshipn between RGC and eye movement oscillations.

Differences in RGC and eye movement oscillation frequency
While there is a clear linear relationship between the RGC and eye movement oscillations, the RGC oscillation frequencies are slightly higher than the frequencies for eye movements.This difference is probably caused by the different physiological conditions of the in vitro and in vivo experiments.Many factors such as pH, the composition of the external solution and potential retinal damage during the dissection may have influenced the RGC oscillations during the in vitro electrophysiological recordings.Differences between in vivo and in vitro conditions have been reported previously by other research groups as well (Belle et al., 2018;Meijer et al., 1997).

Mechanism potentially underlying other forms of nystagmus
The retinal mechanism we propose to cause nystagmus is probably not only limited to CSNB since any condition that can depolarize A II ACs beyond their normal working range will lead them to oscillate, possibly introducing nystagmus.Interestingly, a specific mutation in Munc 18-1 leads also to nystagmus (Li et al., 2020).Munc 18-1 is involved in the docking of synaptic vesicles to the membrane.It does so by binding with syntaxin.In the photoreceptor synaptic terminal Munc 18-1 binds to syntaxin 3B (Curtis et al., 2010).The mutation leading to nystagmus enhances the binding of Munc 18-1 with syntaxin 3B (Li et al., 2020).This mutation probably leads to reduced glutamate release by the photoreceptors resulting in depolarization of the A II ACs, which in turn will start to oscillate and eventually lead to nystagmus.This example illustrates that it is essential to investigate whether novel genes implicated in nystagmus may have affected the A II AC membrane potential and in that way have induced nystagmus.This knowledge will allow us

Figure 1 .
Figure 1.Schematic overview of the photoreceptor to ON-bipolar cell synapse [Colour figure can be viewed at wileyonlinelibrary.com]

Figure 2 .
Figure 2. Example horizontal eye movement traces for a WT, Grm6 nob3 , Nyx nob and Ca v 1.4-KO mouse in response to a vertical sinusoidal grating presented to the left eye All three mouse models show small oscillatory eye movements.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 3 .
Figure 3. Power spectra comparing the horizontal eye movements of WT (N = 7), Grm6 nob3 (N = 12), Nyx nob (N = 7) and Ca v 1.4-KO (N = 7) mice in darkness, under homogenous light as well as with vertical and horizontal bar stimulations Only stimulation with a vertical grating induces oscillatory eye movements in Grm6 nob3 and Nyx nob mice while the eyes of Ca v 1.4-KO mice are oscillating under all stimulus conditions.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 4 .
Figure 4. Comparison of the spontaneous RGC activity of WT, Grm6 nob3 , Nyx nob and Ca v 1.4-KO mice

Figure 5 .
Figure 5.Comparison of the light responses of WT, Grm6 nob3 , Nyx nob and Ca v 1.4-KO RGCs A, raster plots comparing the light responses of, in each case, one RGC from WT (black), Grm6 nob3 (blue), Nyx nob (red) and Ca v 1.4-KO (green).B, example light response traces showing the oscillatory activity of the four genotypes.

Figure 6 .
Figure 6.RGC oscillation frequency depends on light intensity Left: raster plots of the light response of one example Nyx nob RGC for two different light intensities.Right: mean light response of this example cell over the 100 traces showing that the oscillation frequency increases with higher light intensity.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 8 .
Figure 8. Simulated RGC activity in the dark for the three different CSNB mouse models A, raster plots of RGCs.B, average spiking activity.C, cross-correlation of RGC activity.D, power spectral density plots.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 9 .
Figure 9. Simulated RGC activity after a light stimulus for Grm6 nob3 (blue) and Nyx nob (red) mice A, raster plots of the spiking activity of RGCs.B, average spiking activity for the cells shown in A. C, cross-correlation of RGC activity.D, power spectral density plots.[Colour figure can be viewed at wileyonlinelibrary.com]