A mechanism for pathological oscillations in mouse retinal ganglion cells in a model of night blindness

Retinitis pigmentosa (RP), an inherited disorder characterized by progressive degeneration of photoreceptors, has been studied using model mice with retinal degeneration (rd) (Grover et al., 1999; Hartong et al., 2006; Margolis et al., 2008; Goo et al., 2011). In the case of rd1 mouse, rods are almost completely lost by P30–P40 (Carter-Dawson et al., 1978; Jiménez et al., 1996). Retinal ganglion cells (RGCs) of rd mice show pathological oscillatory spontaneous firing (Margolis et al., 2008, 2014; Stasheff, 2008; Borowska et al., 2011; Menzler and Zeck, 2011; Yee et al., 2012; Choi et al., 2014; Poria and Dhingra, 2015). The oscillations in postsynaptic RGCs are driven by oscillations in presynaptic AII amacrine cells (ACs) (Borowska et al., 2011; Trenholm et al., 2012; Choi et al., 2014; Margolis et al., 2014). RGC oscillations, often referred to as “noise,” may disrupt visual information processing. Thus, understanding the mechanisms underlying RGC oscillations is crucial for finding effective therapeutic strategies for RP.

It is widely assumed that RGC oscillations in the rd1 mouse retina may be caused primarily by photoreceptor degeneration, and that the following AII AC oscillations may be induced by intrinsic membrane properties of AII ACs (Choi et al., 2014) or by network properties of AII ACs and ON cone bipolar cells (ON CBCs) (Trenholm et al., 2012). In addition to RGC oscillations, the rd1 mouse retina undergoes significant anatomical remodeling. The sprouting neurites from cones develop ectopic synapses with rod bipolar cell (RBC) soma (Haq et al., 2014). Horizontal cells also extend abnormal processes toward the inner plexiform layer (IPL), whereas the dendrites of ON BCs retract (Strettoi and Pignatelli, 2000; Strettoi et al., 2002, 2003; Chen et al., 2012). Various forms of anatomical remodeling hinder our understanding of how photoreceptor degeneration leads to oscillations in the inner retina of the rd mice.

In the present study, we focus on the Trpm1 knockout (KO) mouse, a model for congenital stationary night blindness (CSNB). TRPM1, a member of the transient receptor potential cation channel subfamily M, is a critical visual transduction channel regulated by mGluR6 on the dendrites of ON BCs. Both Trpm1 KO and mGluR6 KO mice serve as models of CSNB and exhibit a loss of ON responses despite grossly normal retinal structures (Masu et al., 1995; Koike et al., 2009; Morgans et al., 2009; Koike et al., 2010; Nakamura et al., 2010; Xu et al., 2012). However, a notable difference is that pathological spontaneous RGC oscillations are prominently observed in the Trpm1 KO mouse, whereas oscillations are infrequent or absent in the mGluR6 KO mouse (Takeuchi et al., 2018; Hasan et al., 2020). Oscillatory activity occurs in other CSNB models, such as Nyx^nob, Lrit3^emrgg1, and Cav1.4 KO mice (Hasan et al., 2020; Hölzel et al., 2023). Notably, disruption of these genes leads to impaired membrane localization of TRPM1 (Pearring et al., 2011; Neuillé et al., 2015; Maddox et al., 2020). In particular, Nyx and Lrit3 play crucial roles in ensuring proper membrane targeting of TRPM1 (Pearring et al., 2011; Neuillé et al., 2015). Taken together, these consistent associations raise the possibility that TRPM1 loss is a key contributing factor in generating oscillatory activity in CSNB—a hypothesis we investigate using Trpm1 KO mouse retinas.

In this study, we used whole-cell recordings, morphological and immunohistochemical analyses, and computational modeling to investigate the mechanisms underlying RGC oscillations in the Trpm1 KO mouse retina. We compared Trpm1 KO and rd1 mouse retinas and observed shared morphological alterations in RBCs. Using in silico models, we examined how TRPM1 loss and these morphological alterations might together contribute to the emergence of oscillatory activity.

All animal experimental protocols were approved by the Animal Research Committee of Ritsumeikan University and conducted in accordance with local guidelines and the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. For electrophysiological experiments, 1- to 2-mo-old wild-type (WT) and Trpm1 KO mice (129 Sv/Ev background) were used. For immunohistochemical experiments, 1-mo-old WT, Trpm1 KO, and mGluR6 KO mice (129 Sv/Ev background), as well as 1-mo-old WT, mGluR6 KO, and Pde6b^rd1-2J/J (stock #004766; JAX, referred to as rd1) mice (C57BL/6J background), were used. Mice of either sex were used. The Trpm1 KO mouse line, generated by homologous recombination–mediated disruption of the Trpm1 gene, was provided by Dr. Takahisa Furukawa, Osaka University, Suita, Japan. The mGluR6 KO mouse line, generated by homologous recombination–mediated disruption of the mGluR6 gene, was provided by Dr. Shigetada Nakanishi, Kyoto University, Kyoto, Japan. The rd1 mouse retinas were provided by Dr. Michiko Mandai, Research Center, Kobe City Eye Hospital, Kobe, Japan. Mice were housed in a temperature-controlled room under a 12-h light/12-h dark cycle. Fresh water and diet were always available.

Mice were dark-adapted overnight prior to electrophysiological experiments. Each mouse was sacrificed by cervical dislocation under a dim red light, and eyes were enucleated. Under a stereomicroscope equipped with an infrared (IR) image converter (V6833P; Hamamatsu Photonics) and IR illumination (HVL-IRM; Sony), retinas were dissected in a dish filled with Ames’ medium (A1372; US Biological or A1420; Sigma-Aldrich) bubbled with 95% O₂/5% CO₂ at room temperature. The retina isolated from the sclera and pigment epithelium was transferred to a recording chamber, and flat-mounted with the ganglion cell layer (GCL) facing upward. The retina was anchored by a platinum horseshoe with fine nylon threads.

Mice were dark-adapted overnight prior to the procedure. Using an IR image converter and IR illumination, we performed retinal dissection in ice-cold HEPES-buffered solution (130 mM NaCl, 10 mM HEPES, 2.5 mM KCl, 2.5 mM CaCl₂, 1 mM MgCl₂, 28 mM glucose, pH 7.4, with NaOH) continuously bubbled with 100% O₂. Following enucleation, the cornea and lens were carefully removed to create an eyecup. The retina, with the attached retinal pigment epithelium (RPE) and sclera, was cut into quarters and placed RGC side down onto a membrane filter (GSWP04700; Millipore). The RPE and sclera were then gently peeled away from the retina. The retina was subsequently sliced into 200-µm-thick sections using a tissue slicer (EDMS14-234; NARISHIGE). Individual sections were secured in the recording chamber using a platinum horseshoe with fine nylon threads.

Whole-cell clamp recordings were performed from αRGCs (soma diameter >15 µm) in the flat-mount preparation and from RBCs in the retinal slice preparation. Patch pipettes were filled with either K⁺-based internal solution (125 mM K gluconate, 10 mM KCl, 10 mM HEPES, 10 mM phosphocreatine, 0.5 mM EGTA, 0.05 mM CaCl₂, 2 mM MgCl₂, 5 mM ATP-Na₂, 0.5 mM GTP-Na₃, and 0.2% neurobiotin [NB], pH 7.4, with KOH) for current-clamp recordings in RGCs, or Cs⁺-based internal solution (102 mM CsMeSO₃, 20 mM HEPES, 10 mM phosphocreatine, 5 mM EGTA, 0.5 mM CaCl₂, 2.5 mM MgCl₂, 5 mM ATP-Na₂, 0.5 mM GTP-Na₃, 5 mM QX-314, and 0.2% NB, pH 7.4, with CsOH [E_Cl ∼ −60 mV]) for voltage-clamp recordings in RGCs. For RBC recordings, only the K⁺-based internal solution was used, in which 50 µM sulforhodamine was included instead of NB.

Recordings were obtained using an EPC 10 amplifier (HEKA; Multi Channel Systems MCS GmbH) controlled by PATCHMASTER software (version 2.73.5; HEKA). Current and voltage records were sampled at 16 kHz and low-pass–filtered at 2.9 kHz. Patch pipettes were pulled from borosilicate glass capillaries (CNC 1.5; Ken Enterprise, or B150-86-10; Sutter Instrument), using a puller (P97; Sutter Instrument). The resistance of pipettes in Ames’ medium was 4–9 MΩ for αRGCs and 9–15 MΩ for RBCs. A correction was made for liquid junction potential (∼10 mV). The retina was continuously superfused with Ames’ medium bubbled with 95% O₂/5% CO₂ at the rate of 2.5 ml/min at 32°C.

In recordings from flat-mount preparation, a light stimulus was projected from a DLP projector (L51W, refresh rate 60 Hz, 1,280 × 1,024 pixels; NEC), in which the projection lens was replaced with an achromatic lens (f = 120 mm). The image was focused onto the photoreceptor layer through an objective lens (4× Plan Achromat; Nikon), after removing the condenser lens of an upright microscope (Axio Examiner D1; Zeiss). The retina was stimulated with a white spot (diameter: 300 µm) for 2 s (0.26 cd/m²) on a dark background (0.0012 cd/m²) every 10 s. Luminance was measured by a luminance meter (CS-150; KONICA MINOLTA).

In recordings from slice preparation, retinal images were observed using the upright microscope with IR differential interference contrast (DIC) optics.

Following recordings, retinas were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer for 1 h at room temperature. They were then rinsed several times in phosphate-buffered saline (PBS) and incubated overnight at 4°C in a blocking solution (4% normal donkey serum, 0.5% Triton X-100, and 0.1% sodium azide in PBS). Retinas were subsequently incubated with primary antibodies in the blocking solution for 7 days at 4°C. After washing with PBS, retinas were incubated with secondary antibodies for 2 days at 4°C. After washing with PBS, retinas were mounted onto microscope slides. Primary antibodies were goat anti-choline acetyltransferase (ChAT, 1:100, AB144P; Sigma-Aldrich) and mouse anti-SMI-32 (1:1,000, 801701; BioLegend). Secondary antibodies were streptavidin-Alexa 488 (1:500, S32354; Invitrogen), donkey anti-goat Alexa 568 (1:500, A-11057; Invitrogen), and donkey anti-mouse Alexa 647 (1:500, A-31571; Invitrogen).

Z-stack images (1,024 × 1,024 pixels, 0.3 µm/slice) were acquired by a confocal microscope (LSM900; Zeiss) equipped with 40× objective lens (Plan-Apochromat 40×/0.95 Corr M27; Zeiss). Images were analyzed using ImageJ (version 1.53t; National Instutes of Health) and custom codes written in Python. Among recorded RGCs, SMI-32–positive cells with a large soma size (>15 μm) were identified as αRGCs (Krieger et al., 2017). These αRGCs were further subclassified into OFF or ON types based on their dendritic stratification relative to ChAT bands in the IPL.

The cell intracellularly filled with sulforhodamine was identified as RBCs based on the soma location within the inner nuclear layer (INL) and on the position and morphology of axon terminals in the IPL.

Pharmacological agents dissolved in Ames’ medium were bath-applied: DNQX (100 µM; Sigma-Aldrich) for blocking AMPA/KA glutamate receptors; d-AP5 (50 µM; Tocris Bio-Techne) for blocking NMDA glutamate receptors; MFA (100 µM; Sigma-Aldrich) for blocking gap junctions; strychnine (10 µM; Sigma-Aldrich) for blocking glycine receptors.

All data analyses were performed using Python. The power spectral density (PSD) was calculated from the autocorrelogram of spike events recorded during a 145-s membrane potential trace under the current-clamp mode. For the PSD and the autocorrelogram, 0.5-Hz bin width and 1-ms bin width were used, respectively. Oscillation frequency was defined as the frequency corresponding to the peak power between 1.5 and 30 Hz in the PSD. The cross-correlogram (CCR) was calculated based on the spike timing between a pair of RGCs. The time lag was converted to phase by referencing the fundamental frequency of oscillations in each RGC pair.

The amplitude of spontaneous synaptic currents under the voltage-clamp mode was analyzed as follows. The current trace was band-pass–filtered (Bessel filter; 1–100 Hz), and the trace during the last 2.5 s of a 3-s voltage step was used to generate a current histogram with 5-pA bin width. A cumulative distribution function (CDF) was then calculated from the histogram. The current amplitude was defined as the distance between 0.05 and 0.95 in the CDF. The polarity of synaptic current was determined by the value of skewness of the current histogram (i.e., the inward and outward currents were less than −0.15 and >0.15, respectively, whereas the values in between were not used for analysis). The reversal potential of the synaptic current was estimated from the current amplitude–voltage relationship. The PSD of spontaneous synaptic current was calculated using a bin width of 0.5 Hz from the band-pass–filtered trace.

The membrane potential of RBCs was recorded for 40 s under the current-clamp mode, and the voltage histogram was constructed using a bin width of 1 mV. The resting membrane potential was defined as the membrane potential corresponding to the peak value of the histogram. For voltage-clamp recordings, cells were held at −60 mV and stepped from −80 to 0 mV in +10 mV increments to construct current–voltage (I–V) curves. The slope conductance was subsequently calculated from the linear portion of the I–V curve, specifically near the resting membrane potential.

Mice were euthanized by cervical dislocation, and the eyes were enucleated. In a dish filled with PBS, the cornea and lens were removed, and the retinas were isolated from the eyecups. Retinas were fixed in 4% paraformaldehyde in 0.1 M phosphate buffer for 20 min at room temperature. After washing with PBS, the retinas were embedded in 4% low-melting agarose (NIPPON GENE) in PBS. Retinas were sectioned at 50-µm thickness using a vibratome (PRO7; Dosaka EM) and incubated in PBS containing either 5% normal goat serum or 5% normal donkey serum, both with 0.1% Triton X-100, for 1 h at room temperature. After blocking, sections were incubated with primary antibodies overnight at 4°C. After washing with PBS, the sections were then incubated with secondary antibodies. Sections were rinsed several times with PBS and mounted on microscope slides. The primary and secondary antibodies used are listed in Table 1.

Z-stack images (1,024 × 1,024 pixels, 0.2 µm/slice) were acquired by the confocal microscope equipped with a 63× oil immersion lens (Plan-Apochromat 63×/1.4 Oil DIC M27, Zeiss). Voxel size was 0.1 × 0.1 × 0.2 µm per pixel. Image analyses were performed using ImageJ, Imaris (ver. 9.9.1; Oxford Instruments), and Python. The volume of RBC axon terminals was measured from the merged volume of PKCα and VGLUT1 signals. The distributions of RBC axon terminals and type 6 BC axon terminals in the IPL were defined as the starting point (INL side) and ending point (GCL side) of PKCα signal distribution and Syt2 signal distribution by eye, and it was evaluated relative to the ChAT bands (OFF band: 0%; ON band: 100%).

Retinal neural circuit models were constructed to reproduce not only the normal firing of RGCs in WT retinas but also the spontaneous oscillatory firing of RGCs observed in pathological retinas. The following three steps were involved in the simulation. (1) Construction of a normal retinal neural circuit model (the WT circuit model): based on the experimental data obtained from WT retinas, we constructed a model that responded to both the offset and onset of light stimuli (Dunn et al., 2006; Margolis and Detwiler, 2007; Arman and Sampath, 2012). (2) Application of known pathological conditions: pathological conditions identified in previous studies (Chua et al., 2009; Koike et al., 2009; Križaj et al., 2010; Takeuchi et al., 2018) were applied to the WT mouse model constructed in step 1). Additionally, certain pathological conditions observed in the present study were also incorporated. (3) Exploration of conditions leading to RGC oscillations: based on the pathological model constructed in step 2, we explored the conditions under which RGCs exhibited pathological spontaneous oscillatory firing.

For comparison with experimental results, the retinal neural circuit models were modified by setting the input currents to BCs:

• -

  The WT circuit model: the input currents representing photoreceptor signals were applied to all BC types.

• -

  The Trpm1 KO circuit model: TRPM1 channels are expressed only on the dendrites of ON BCs (RBCs and ON CBCs). We hypothesized that light signals are not transmitted to downstream cells in the ON pathway due to the absence of TRPM1 function. Consistent with this hypothesis, we found that the membrane potential of RBCs is hyperpolarized in these mice (Fig. 4, A and B). Therefore, sustained hyperpolarizing currents were applied to RBCs to induce hyperpolarization, whereas no input currents were applied to ON CBCs. Input currents representing photoreceptor signals comparable to those in the WT circuit model were applied to OFF CBCs.

• -

  The mGluR6 KO circuit model: mGluR6 is also expressed only on the dendrites of ON BCs. Loss of mGluR6 leads to hyperpolarization of RBCs (Xu et al., 2012). Therefore, we assume that BCs in the mGluR6 KO model behave similar to those in the Trpm1 KO circuit model. As in the Trpm1 KO circuit model, sustained hyperpolarizing currents were applied to RBCs to induce hyperpolarization, whereas no input currents were applied to ON CBCs. Input currents representing photoreceptor signals comparable to those in the WT circuit model were applied to OFF CBCs.

• -

  The rd1 circuit model: in the rd1 mouse retina, photoreceptor degeneration leads to the loss of glutamate release from photoreceptors and the loss of TRPM1 channels in ON BC dendrites (Fig. 3 N) (Križaj et al., 2010). Based on these alterations, the hyperpolarization observed in RBCs in the Trpm1 KO mouse retina (Fig. 4), and a previous report showing a similar state in the rd1 mouse retina (Borowska et al., 2011), we hypothesized that RBCs are also hyperpolarized in the rd1 mouse retina. As in the Trpm1 KO and mGluR6 KO circuit models, sustained hyperpolarizing currents were applied to RBCs to induce hyperpolarization. No input currents were applied to OFF and ON CBCs.

In addition, we imposed the conditions based on the experimental evidence (2) and explored the conditions where oscillatory firing was evoked in appropriate RGCs (3).

A detailed description of the computational simulation is provided in the supplemental text at the end of the PDF.

Data were presented as the mean ± SEM, except for immunohistochemical data, which were presented as the mean ± SD. Error bars indicate SEM unless otherwise noted. For pharmacological data, statistical significance was evaluated using a two-sided paired t test. For immunohistochemical data, multiple comparisons were performed using one-way ANOVA with post hoc Tukey’s test. Other comparisons were made using two-sided unpaired Student’s t test. A P value <0.05 was considered statistically significant (n.s., not significant; *P < 0.05; **P < 0.01; ***P < 0.001).

Fig. S1 illustrates the states of neurotransmitters of a ribbon synapse and the transition between them. Fig. S2 displays time courses of response recovery of the ribbon synapse model. Table S1 summarizes parameters values of the ribbon synapse model. Table S2 summarizes parameters of synaptic connectivity between cells and synaptic conductance, and the values of the synaptic conductance (the upper row in each cell) and the connectivity ratio (pre:post; the bottom row in each cell). Supplemental text at the end of the PDF contains a detailed description of the computational simulation used in this paper.

Previous MEA recordings demonstrated that RGCs of the Trpm1 KO mouse retina generate spontaneous oscillatory firing (Takeuchi et al., 2018). To elucidate the mechanism underlying these oscillations, we performed whole-cell recordings from αRGCs (soma size >15 µm) in flat-mount retinas of Trpm1 KO mice. It should be noted that ON BCs do not respond to light illumination due to the deletion of TRPM1 channels. Whole-cell current-clamp recordings revealed that spot illumination elicited OFF responses in some αRGCs (Fig. 1 A, left), while others showed no light responses (Fig. 1 B, left). To confirm that the light-responsive and nonresponsive cells corresponded to OFF and ON αRGCs, respectively, we examined the morphology of RGCs filled with NB through a recording pipette. RGC subtypes were determined by their dendritic stratification relative to the ChAT bands (dendritic stratification of OFF and ON starburst ACs) in the IPL. The OFF- and ON-ChAT bands in the IPL were defined as 0% and 100%, respectively. Consistent with their light responses, RGCs with light-Off responses had dendrites that stratified in the OFF sublamina (Fig. 1 A, center), whereas those without responses stratified their dendrites in the ON sublamina (Fig. 1 B, center). Furthermore, immunohistochemical staining for SMI-32, an αRGC marker, confirmed that the recorded RGCs were indeed αRGCs (Fig. 1, A and B, right, arrowheads). Therefore, in the following experiments, the subtype of recorded RGCs was determined based on this postrecording morphological analysis.

To examine the properties of spontaneous oscillatory discharges in αRGCs, cells were whole-cell current-clamped using a patch pipette filled with K⁺-based solution. Spike discharges were evoked when the membrane potential fluctuations reached a threshold (Fig. 1 C, left). An autocorrelogram was calculated from the timing of these spike discharges (Fig. 1 C, center). The PSD derived from the autocorrelogram revealed that the fundamental frequency of oscillations was ∼8–9 Hz (Fig. 1 C, right) in both OFF and ON αRGCs (Fig. 1 D). These results may be explained either by common inputs from a presynaptic oscillator or by their own similar intrinsic oscillators.

To examine the phase relationship of oscillations, we performed simultaneous whole-cell current-clamp recordings from pairs of αRGCs. CCRs were calculated from their spike discharges, and the time lag was converted to phase using the fundamental frequency of oscillations in each RGC pair (Fig. 1 E). The phase of oscillations was anti-phase between a pair of OFF and ON αRGCs, whereas it was in-phase between a pair of the same αRGC type. This difference in phase relationship was statistically significant (Fig. 1 F). These findings are consistent with those previously reported in rd1 and Nyx^nob mouse retinas (Margolis et al., 2014; Winkelman et al., 2019), and suggest the idea that oscillations in OFF and ON αRGCs are driven by synaptic inputs from common presynaptic oscillators.

To determine whether αRGC oscillations were driven by synaptic inputs rather than intrinsic voltage-dependent membrane properties, αRGCs were whole-cell voltage-clamped using a patch pipette filled with Cs⁺-based solution to suppress K⁺ channels. Under these conditions, oscillatory membrane currents were evident, indicating that the oscillations were indeed driven by synaptic inputs rather than intrinsic properties of αRGCs. The mean frequency of these synaptic current oscillations was 3.1 ± 0.2 Hz (n = 41 cells from 28 mice), which was significantly lower than that observed during current-clamp recordings with K⁺-based pipette solution (8.5 ± 0.3 Hz, n = 27 cells from 9 mice) (P = 4.64 × 10⁻²⁶, unpaired Student’s t test). This frequency difference might be attributable to the blockade of I_h channels (Kaneko and Tachibana, 1985; Müller et al., 2003; Trenholm et al., 2012) and/or M-type K⁺ channels (Cembrowski et al., 2012; Choi et al., 2014) by Cs⁺ extruded from a patch pipette before establishing the cell-attached configuration.

To further characterize the oscillatory synaptic inputs, synaptic currents were recorded at various holding potentials in both OFF and ON αRGCs (Fig. 1 H). The amplitudes of these synaptic currents (see Materials and methods) were plotted against their corresponding holding potentials (Fig. 1 I). The reversal potential of the synaptic currents differed significantly between OFF and ON αRGCs (Fig. 1 J). These results suggest that oscillations in OFF αRGCs are driven by inhibitory (glycinergic and/or GABAergic) synaptic inputs, while those in ON αRGCs are driven by excitatory (glutamatergic) synaptic inputs. The observed differences in synaptic inputs are consistent with the known retinal circuitry involving AII ACs. AII ACs form gap junctions with both ON CBCs and other AII ACs and make glycinergic synapses onto OFF CBCs and some OFF RGCs (Fig. 1 G) (Feigenspan et al., 2001; Murphy and Rieke, 2006; Murohy and Rieke, 2008; Münch et al., 2009; Tsukamoto and Omi, 2017). Therefore, it is likely that OFF αRGCs receive direct glycinergic inputs from AII ACs, whereas ON αRGCs may receive glutamatergic inputs from ON CBCs, which are connected to AII ACs through gap junctions (Fig. 1 G).

We conducted pharmacological experiments to investigate the synaptic pathways responsible for transmitting oscillatory signals to αRGCs. OFF αRGCs were voltage-clamped at 0 mV to isolate oscillatory inhibitory postsynaptic currents (IPSCs). Bath application of strychnine (10 µM), a glycine receptor antagonist, significantly reduced both the amplitude of IPSCs and the peak power in the PSD (Fig. 2, A and D). In contrast, application of iGluR antagonists (a mixture of 100 µM DNQX and 50 µM d-AP5) affected neither the amplitude of IPSCs nor the peak power in the PSD (Fig. 2, B and D). This result suggests that oscillatory IPSCs in OFF αRGCs may be driven directly by AII ACs rather than via other ACs through iGluR-mediated synapses. Application of MFA (100 µM), a nonspecific gap junction blocker, also significantly reduced both the amplitude of IPSCs and the peak power in the PSD (Fig. 2, C and D). Based on the circuit diagram shown in Fig. 1 G, the effects of glycine receptor and iGluR antagonists were consistent with expectations. However, the effect of the gap junction blocker was unexpected. It is possible that gap junction blockade hyperpolarizes AII ACs and abolished oscillatory activity in AII ACs themselves (see Fig. 7 C) (Trenholm et al., 2012; Choi et al., 2014).

ON αRGCs were voltage-clamped at −60 mV to record oscillatory excitatory postsynaptic currents (EPSCs). Application of strychnine (10 µM) affected neither the amplitude of EPSCs nor the peak power in the PSD (Fig. 2, E and H). In contrast, application of iGluR antagonists (a mixture of 100 µM DNQX and 50 µM d-AP5) abolished both the oscillatory EPSCs and the peak power in the PSD (Fig. 2, F and H). MFA (100 µM) also significantly reduced the amplitude of EPSCs and the peak power in the PSD (Fig. 2, G and H). These findings are consistent with the synaptic connections outlined in Fig. 1 G.

Previous studies have shown that both Trpm1 KO and rd1 mouse retinas exhibit prominent spontaneous oscillations in RGCs, whereas the mGluR6 KO mouse retina shows infrequent or absent spontaneous oscillations (Margolis et al., 2008; Margolis et al., 2014; Stasheff, 2008; Borowska et al., 2011; Menzler and Zeck, 2011; Yee et al., 2012; Choi et al., 2014; Poria and Dhingra, 2015; Takeuchi et al., 2018; Hasan et al., 2020). In the WT mouse retinas, RBCs, which express both TRPM1 and mGluR6 on their dendritic membranes, provide glutamatergic output to AII ACs. Thus, it is interesting to examine whether the morphological features of RBC differ among these mouse retinas. Firstly, we focused on the position and size of RBC axon terminals. We examined the distribution of RBC terminals relative to ChAT bands in the IPL in each mouse retina. The starting point of terminal distribution in the distal IPL did not significantly differ among WT, Trpm1 KO, and mGluR6 KO mouse retinas (Fig. 3, A–C). However, the ending point of terminal distribution in the proximal IPL was significantly closer to the ON-ChAT band (100%) in the Trpm1 KO mouse retinas compared with both WT and mGluR6 KO mouse retinas (Fig. 3, A–C).

To determine whether this change in terminal distribution is specific to RBCs or common among ON BCs, we compared the distribution of type 6 BC terminals between WT and Trpm1 KO mouse retinas. No significant difference was found in the terminal distribution of type 6 BCs between the two genotypes. In contrast, RBC terminals in the Trpm1 KO mouse retinas showed a significantly shorter span, with their ending point located closer to the ON-ChAT band than in the WT mouse retinas (Fig. 3, D–F). These results suggest that the altered distribution of RBC terminals in the Trpm1 KO mouse retinas is cell type–specific and not simply due to a general change in the thickness of the ON sublamina. In the rd1 mouse retinas, the RBC terminal distribution was shifted toward the INL side (Fig. 3, G–I).

We previously reported that the volume of individual RBC axon terminals is reduced in the Trpm1 KO mouse retinas relative to the mGluR6 KO and WT mouse retinas, which are indistinguishable (Takeuchi et al., 2018). In the rd1 mouse retina, a similar reduction in RBC axon terminal volume was observed (Fig. 3, J and K) (see also Strettoi et al., 2002, 2003; Chua et al., 2009). Together, these results demonstrate that prominent alterations in RBC axon terminals are present in both Trpm1 KO and rd1 mouse retinas, but not in the mGluR6 KO mouse retinas. It is tempting to speculate that these morphological alterations are associated with a reduction in signaling from RBCs to AII ACs in both Trpm1 KO and rd1 mouse retinas.

In the outer plexiform layer (OPL), knocking out Trpm1 affected neither the intensity of mGluR6 puncta nor their apposition to the presynaptic rod ribbon marker rib eye (CtBP2) (Fig. 3 L), consistent with previous reports (Koike et al., 2009; Morgans et al., 2009). Similarly, knocking out Trpm1 did not alter the colocalization of mGluR6 with GluK1 (a kainate receptor subunit: OFF CBC marker) and arrestin at cone terminals (Fig. 3 M). In contrast, the localization of TRPM1 in the OPL differed among WT and three models (Trpm1 KO, mGluR6 KO, and rd1) (Fig. 3 N). In the mGluR6 KO mouse retinas, TRPM1 puncta were present but displayed slightly reduced intensity compared with the WT mouse retinas. In the rd1 mouse retinas, TRPM1 puncta on RBC dendrites were markedly reduced (Fig. 3 N). This finding was consistent with the loss of rod spherules, although some cone pedicles were preserved (Fig. 3 O). The Trpm1 KO mouse retina lacks puncta by definition (Koike et al., 2009; Morgans et al., 2009). These results suggest a three-way association with RGC oscillations: the extent of TRPM1 loss at dendrites, the altered distribution of RBC terminals, and the shrinkage of RBC axon terminals.

Our immunohistochemical analyses revealed morphological alterations of RBCs in mouse retinas that exhibit RGC oscillations. Based on these findings, we next investigated potential functional changes, specifically focusing on the synaptic input from RBCs to AII ACs. It seems likely that the neurotransmitter release from neurons, including RBCs, depends on their membrane potential (Snellman et al., 2009; Graydon et al., 2018). Thus, using retinal slice preparation, we measured the resting membrane potential of RBCs. The resting membrane potential of RBCs in the Trpm1 KO mouse retinas was significantly hyperpolarized compared with those in the WT mouse retinas (Fig. 4, A and B). This may result from the loss of the TRPM1 channels. To test this possibility, we analyzed the I–V relationship of RBCs. We found that the slope conductance, calculated from the linear range of the I–V curve (Fig. 4 C) near the resting membrane potential, was significantly lower in the Trpm1 KO mouse retinas than in the WT mouse retinas (Fig. 4 D). Therefore, the hyperpolarization of RBCs in the Trpm1 KO mouse retinas may cause reduced glutamate release from RBCs to AII ACs.

We constructed retinal circuit models to examine whether reduced synaptic input from RBCs to AII ACs can generate oscillatory firing in RGCs. First, we developed a normal retinal circuit model (the WT circuit model) as described in Materials and methods (see also supplemental text at the end of the PDF). Second, we modified the input currents to BCs in each mouse circuit model to reproduce the light-evoked responses, as detailed in Materials and methods (see also supplemental text at the end of the PDF). Third, to reflect the shrinkage of RBC axon terminals observed in the Trpm1 KO and rd1 mouse retinas (Fig. 3) (Takeuchi et al., 2018), we reduced the synaptic conductance from RBCs to AII ACs in those circuit models.

In the mammalian retina, RBCs provide excitatory drive to AII ACs, which in turn are electrically coupled to ON CBCs via gap junctions (Fig. 1 G). In the Trpm1 KO and rd1 mouse retinas, the loss of TRPM1 channels induces severe defects in RBCs, including its absence or a marked reduction of the channel at their dendritic tips (Koike et al., 2009) (Fig. 3 N), altered axon terminal morphology (Takeuchi et al., 2018) (Fig. 3), and a hyperpolarized resting membrane potential (Fig. 4) (Borowska et al., 2011). Collectively, these deficits are expected to reduce the excitatory drive to AII ACs, consequently leading to a hyperpolarizing shift in the resting membrane potential of the coupled ON CBCs. Furthermore, the loss of the light-evoked responses in ON CBCs in these mouse retinas may also contribute to their hyperpolarization (Lagali et al., 2008; Koike et al., 2009). Since ON CBCs express T-type Ca²⁺ channels (de la Villa et al., 1998; Pan, 2000; Zhang et al., 2022), we hypothesized that these channels might allow glutamate release even at hyperpolarized membrane potentials. This condition was reproduced in those specific models by lowering the half-activation voltage of Ca²⁺ channels (see V_th in Eq. 5 in Materials and methods and additional details in the supplemental text at the end of the PDF) in the ON CBC terminals.

Using these retinal circuit models, we simulated the membrane potential changes of each cell. In the WT circuit model, all cell types (RBCs, OFF and ON CBCs, AII ACs, and OFF and ON RGCs) responded to light stimulation, and no membrane oscillations were observed (Fig. 5, A and B).

In the Trpm1 KO circuit model (Fig. 5, C and D), RBCs were hyperpolarized and showed no ON response. OFF CBCs exhibited weak membrane oscillations and a hyperpolarizing light response. ON CBCs and AII ACs were hyperpolarized due to reduced inputs from RBCs to AII ACs and displayed membrane oscillations without ON responses. Both OFF and ON RGCs showed oscillatory firing, but only OFF RGCs responded to light stimulation.

In the mGluR6 KO circuit model (Fig. 5, E and F), RBCs were hyperpolarized and lacked ON responses. OFF CBCs exhibited OFF responses to light stimulation but no obvious membrane oscillations. ON CBCs and AII ACs showed neither ON responses nor membrane oscillations. OFF and ON RGCs exhibited spontaneous firing, but only OFF RGCs responded to light stimulation. Such firing is unlikely to represent true oscillations, as they were not synchronized with fast membrane fluctuations of AII ACs. Rather, they may reflect intrinsic membrane properties of RGCs, which were not observed in our physiological recordings under the whole-cell voltage-clamp condition.

Notably, the circuit model reproduced RGC oscillations only in the Trpm1 KO circuit model but not in the mGluR6 KO circuit model. The critical difference between Trpm1 KO and mGluR6 KO circuit models is in the synaptic conductance from RBCs to AII ACs. In the Trpm1 KO circuit model, this conductance was reduced to 0 due to axon terminal shrinkage, whereas it remained normal in the mGluR6 KO circuit model, comparable to the WT levels. These findings suggest that reducing RBC-to-AII AC synaptic conductance may be essential for inducing membrane oscillations in AII ACs. Supporting this conclusion, the rd1 circuit model, which also included reduced synaptic conductance from RBCs to AII ACs, reproduced membrane oscillations (Fig. 5, G and H).

To evaluate how the membrane potential of RBC and ON CBC influences AII AC oscillations, we systematically manipulated parameters of the Trpm1 KO circuit model. Although our original Trpm1 KO circuit model exhibited membrane oscillations in AII ACs (Fig. 5 C), these oscillations disappeared when the synaptic conductance from RBC to AII AC (g_{RBC to AII AC}) was set to 1.0 (a value comparable to the WT circuit model) and the membrane potential of RBC and ON CBC was set to −68 and −52 mV, respectively (Fig. 6 A1). In contrast, marked membrane oscillations were observed in AII ACs when the ON CBC membrane potential was hyperpolarized to −58 mV, while the RBC membrane potential and g_{RBC to AII AC} were kept at −68 mV and 1.0, respectively (Fig. 6 A2). Furthermore, marked membrane oscillations were also observed in AII ACs when g_{RBC to AII AC} was reduced to 0.1, while the membrane potential of RBC and ON CBC remained the same (RBC: −68 mV, ON CBC: −52 mV; Fig. 6 A3). A summary of the effects of the membrane potential of RBC and ON CBC on the AII AC activity is shown in Fig. 6 B. These simulations suggest that the membrane potential of ON CBC may play a more critical role in generating AII AC oscillations than that of RBC (Fig. 6, A1 and A2). Moreover, reducing the synaptic conductance from RBC to AII AC also had a significant impact on generating AII AC oscillations (Fig. 6, A1 and A3). Taken together, these results suggest that both reduction of the synaptic conductance from RBC to AII AC and hyperpolarization of ON CBC are key factors for generating AII AC oscillations.

Using the Trpm1 KO circuit model, we simulated the effects of pharmacological blockers by selectively modifying the relevant synaptic or gap junctional conductance. First, to mimic the effect of a glycine receptor antagonist, we set g_glycine to 0 nS. This manipulation eliminated inhibitory input from AII AC to OFF RGC (Fig. 7 A3), resulting in tonic excitatory input from OFF CBC to OFF RGC (Fig. 7 A2). Consequently, this tonic excitatory input but not periodic input evoked continuous firing in OFF RGC during the epoch without light stimulation (Fig. 7 A1). OFF RGCs themselves or other GABAergic ACs may contribute to OFF RGC continuous firing. In response to light stimulation, OFF CBC reduced the excitatory input to OFF RGC (Fig. 7 A2), resulting in an OFF response (Fig. 7 A1). Notably, this parameter change did not affect the ON pathway (Fig. 7, A4 and A5).

Second, the effects of iGluR antagonist were mimicked by setting g_iGluR to 0 nS. Under this condition, oscillations in both OFF and ON RGCs disappeared (Fig. 7, B1 and B4). Excitatory inputs were eliminated from OFF CBC to OFF RGC (Fig. 7 B2) and from ON CBC to ON RGC (Fig. 7 B5), although inhibitory inputs from AII AC to OFF RGC were maintained (Fig. 7 B3).

Finally, to mimic the effects of gap junction blockers, we set g_{gap junction} to 0 nS. This manipulation reduced membrane oscillations in AII AC (Fig. 7 C3) and rendered inhibitory inputs from AII AC to OFF CBC and OFF RGC tonic (Fig. 7, C2 and C3). This resulted in reduced periodicity of OFF RGCs (Fig. 7 C1). Furthermore, membrane oscillations in both ON CBC and AII AC disappeared, and tonic excitatory inputs from ON CBC to ON RGC (Fig. 7 C5) led to repetitive firing (Fig. 7 C4). These simulations demonstrate that our Trpm1 KO circuit model successfully reproduced the blocker effects comparable to our electrophysiological results (Fig. 2).

In animal models of RP, including the rd1 mouse, severe malformations in the inner retina complicate our understanding of the mechanisms underlying pathological RGC oscillations. However, the Trpm1 KO mouse retina, which retains intact photoreceptors, also displays pathological RGC oscillations. Thus, the Trpm1 KO mouse retina provides a valuable system for investigating the mechanisms of RGC oscillations in a structurally less compromised context.

In the present study, we investigated the mechanisms underlying RGC oscillations in the Trpm1 KO mouse retina. We found that the RGC oscillations in OFF αRGCs are driven by inhibitory inputs mediated by glycine receptors, while oscillations in ON αRGCs arise from excitatory inputs mediated by glutamate receptors (Figs. 1 and 2). Notably, a gap junction blocker eliminated oscillations in both RGC types, suggesting a common upstream origin within an electrically coupled network (Fig. 2). Furthermore, Trpm1 KO mouse retina displayed morphological alterations in RBCs, similar to the rd1 mouse retina (Fig. 3). The membrane potential of RBCs in the Trpm1 KO mouse retina was significantly hyperpolarized (Fig. 4). Finally, our mathematical modeling reproduced the RGC oscillations by incorporating these structural changes and the loss of ON signals, suggesting that TRPM1 loss, hyperpolarization, and axon shortening are precursors to oscillations in the AII AC network (Figs. 5 and 6).

Computer simulations suggest that RGC oscillations in the Trpm1 KO mouse retina arise from reducing inputs from RBC and ON CBC to AII AC (Figs. 5 and 6). This reduction may be ascribed to shrinkage of RBC axon terminals, which may be caused by the loss of TRPM1 localization at the dendritic tips of RBCs (Fig. 3 N) (Kozuka et al., 2017; Takeuchi et al., 2018). Such smaller RBC axon terminals may impair the functionality of the release machinery, for instance, through a reduction of Ca²⁺ channels and/or a decreased number of ribbons and synaptic vesicles. Therefore, future detailed experiments focusing on the synaptic transmission from BCs (particularly RBCs) to AII ACs are warranted.

Furthermore, the loss of TRPM1 localization at the dendritic tips of ON BCs may lead to their hyperpolarization. Indeed, our data show that the resting membrane potential of RBCs in the Trpm1 KO mouse retinas is more hyperpolarized than that in the WT mouse retinas (Fig. 4). This result suggests that ON CBCs may also be hyperpolarized in the Trpm1 KO mouse retina. These changes could, in turn, contribute to a modest hyperpolarization and membrane oscillations of AII ACs.

The Trpm1 KO mouse retina exhibits prominent spontaneous RGC oscillations. In contrast, mGluR6-mutant mice, one of CSNB models, exhibit infrequent or absent RGC oscillations (Rentería et al., 2006; Takeuchi et al., 2018; Hasan et al., 2020; Hölzel et al., 2023). For instance, Hasan et al. (2020) found that only ∼4% of RGCs in mGluR6 KO mice oscillate in darkness. Hölzel et al. (2023) also reported that oscillatory activity can be evoked by light stimulation in a small proportion of RGCs (∼5%), but absent oscillations are observed in the dark. These infrequent or stimulus-dependent oscillations differ markedly from the robust and spontaneous oscillations observed in Trpm1 KO mouse retinas, where ∼50% of RGCs exhibit oscillatory activity even in the absence of light stimulation (Takeuchi et al., 2018).

In addition to functional differences, the mGluR6 KO retina also exhibits distinct morphological features compared with the Trpm1 KO mouse retina. For example, TRPM1 localization at the ON BC dendritic tips is reduced but still detectable in the mGluR6 KO mouse retinas (Fig. 3 N) (Xu et al., 2012), and the distribution and size of RBC axon terminals are comparable to those in WT mouse retinas (Fig. 3, A–C) (Kozuka et al., 2017; Takeuchi et al., 2018). These observations suggest that the synaptic transmission from RBC to AII AC may be relatively unimpaired in the mGluR6 KO mouse retina.

Although the loss of mGluR6 is theoretically expected to render TRPM1 channels constitutively open, RBCs in both Trpm1 KO (Fig. 4, A and B) and mGluR6 KO mouse retinas are hyperpolarized (Xu et al., 2012). We measured a 0.20 ratio between the slope conductance in the voltage range around the resting membrane potential in the Trpm1 KO and WT, consistent with a marked reduction in membrane conductance (Trpm1 KO: 0.17 ± 0.05 nS; WT: 0.86 ± 0.13 nS) (Fig. 4 D). In comparison, Xu et al. (2012) found a ratio of 0.59 for mGluR6 KO versus WT mice under similar recording conditions (mGluR6 KO: 1.09 ± 0.08 nS; WT: 1.84 ± 0.26 nS). This difference suggests that the residual TRPM1 channels at the dendritic tips of RBCs in the mGluR6 KO mouse retina may remain functional and exhibit spontaneous gating. This preserved channel activity may contribute to the structural maturation of RBCs in the mGluR6 KO mouse retina, thereby explaining their near-normal morphology and a lack of robust network oscillations.

The neural oscillations in the Trpm1 KO mouse retina revealed in this study share striking similarities to the phenomenon reported in the rd1 mouse retina. This suggests that despite their distinct genetic backgrounds, both mouse retinas form a functionally equivalent pathological oscillatory network.

The fundamental physiological characteristics of the oscillations are almost identical in both mouse models. The oscillation frequency determined from the extracellular potential of RGCs was ∼8 Hz in the Trpm1 KO mouse retina (Takeuchi et al., 2018), which is comparable to the ∼9 Hz in the rd1 mouse retina (Menzler and Zeck, 2011). Furthermore, the phase relationships of oscillations between αRGCs (anti-phase between different-type cell pairs, in-phase between same-type cell pairs) and the pattern of synaptic inputs, in which OFF αRGCs receive oscillatory inhibitory inputs and ON αRGCs receive excitatory inputs (Fig. 1), were in agreement with reports from the rd1 mouse retina (Margolis et al., 2008, 2014). This convergence of multiple physiological features strongly suggests that a common underlying circuit is in operation.

These functional similarities appear to stem from shared functional and morphological alterations in RBCs. Although RBCs in the rd1 mouse retina were predicted to be depolarized due to a decrease in mGluR6 (Strettoi and Pignatelli, 2000; Strettoi et al., 2002, 2003), direct recordings have shown that they are hyperpolarized (Borowska et al., 2011), similar to the Trpm1 KO mouse retina (Fig. 4). This phenomenon is consistent with the marked reduction of TRPM1 at the dendritic tips of RBCs in the rd1 mouse retina (Fig. 3 N) and aligns with previous reports establishing a link between TRPM1 reduction and RBC hyperpolarization (Ou et al., 2015). Moreover, our study reveals that RBCs in Trpm1 KO and rd1 mouse retinas share morphological alterations, including reduced TRPM1 localization at dendritic tips, changes in axonal terminal distribution, and shrinkage of axonal terminals (Fig. 3) (Takeuchi et al., 2018). These shared pathologies likely serve as a critical trigger for driving the downstream oscillatory network.

Taken together, the multiple commonalities, including the oscillation frequency, phase relationships, synaptic input patterns, and the upstream hyperpolarization and morphological alterations in RBCs, strongly support the view that the oscillations in the Trpm1 KO and rd1 retinas are essentially equivalent phenomena driven by a common circuit mechanism.

Trpm1 is one of the genes responsible for CSNB (Nakamura et al., 2010). Other genes implicated in CSNB include mGluR6 and Ca_v1.4. Additionally, genes such as Nyx and Lrit3, which encode proteins essential for the correct localization of TRPM1 to the dendritic tips of ON BCs, are also associated with CSNB (Pearring et al., 2011; Neuillé et al., 2015; Zeitz et al., 2015). With the notable exception of the mGluR6 KO mouse retinas, which retain TRPM1, spontaneous RGC oscillations comparable to those in the Trpm1 KO mouse retinas have been described in the Nyx^nob, Ca_v1.4 KO, and Lrit3^emrgg1 mouse retinas (Winkelman et al., 2019; Hasan et al., 2020; Hölzel et al., 2023). RGC oscillations have also been reported in the rd10 mouse retinas (Goo et al., 2011; Toychiev et al., 2013), other RP model mice. Immunohistochemical investigation, including our findings for the Trpm1 KO and rd1 mouse retinas, has revealed that TRPM1 localization is absent or reduced at the dendritic tips of ON BCs in the oscillated retinas, such as Trpm1 KO, Nyx^nob, Lrit3^emrgg1, Ca_v1.4 KO, rd1, and rd10 mouse retinas (Fig. 3 N and Table 2) (Koike et al., 2009; Morgans et al., 2009; Križaj et al., 2010; Pearring et al., 2011; Gayet-Primo and Puthussery, 2015; Hasan et al., 2019; Maddox et al., 2020). The rd1 and rd10 mice are models of RP that cause photoreceptor degeneration but relatively preserve BCs. This raises the question of why the rd retina loses TRPM1 from the dendritic tips of ON BCs despite no direct genetic defect in TRPM1 itself. As Hasan et al. (2019) have shown that LRIT3 is necessary for the localization of TRPM1 to the dendritic membrane of ON BCs and is located at photoreceptor terminals, it is plausible that the extent of photoreceptor degeneration affects LRIT3 expression and/or localization, which may influence the degree of TRPM1 loss at the dendritic tips of RBCs. Collectively, our results suggest that the loss of TRPM1 localization at RBC dendritic tips may precede and contribute to the observed alterations in RBC axon terminal distribution and shrinkage. Therefore, TRPM1 loss could also contribute to RGC oscillations in the rd mouse retinas.

In the degenerating rd10 mouse retina, it is reported that the membrane potential of RBCs is depolarized when some rod photoreceptors still survive (Li et al., 2024, Preprint), suggesting that some TRPM1 channels may remain functional in RBC dendrites during this stage. However, after complete rod photoreceptor degeneration in the rd10 mouse retina, TRPM1 localization at RBC dendrites is absent. Furthermore, RGC oscillations have also been observed at this stage of degeneration (Goo et al., 2011; Gayet-Primo and Puthussery, 2015). Our proposed mechanism involving TRPM1 loss at RBC dendritic tips as a driver of pathological oscillations may also explain the oscillations observed in the late stage of degeneration in the rd10 mouse.

It has been reported that the retinoic acid (RA) level increases in the rd mouse retinas due to photoreceptor degeneration, which, in turn, contributes to RGC hyperactivity (Telias et al., 2019, 2022). Given that the outer retinal layers remain intact in the Trpm1 KO retina, it is unlikely that RA levels are elevated in this model. In our present study, we identified common morphological alterations in RBCs from both Trpm1 KO and rd1 mouse retinas, including altered terminal distribution in the IPL, reduced axon terminal size, and loss of TRPM1 at dendritic tips. The absence or reduction of TRPM1 localization at RBC dendritic tips is a consistent feature across oscillating retinas, including those in RP models and certain CSNB models. Taken together, these results suggest that the loss of TRPM1 at the dendritic tips of ON BCs could be a significant contributor to RGC oscillations, potentially independent of RA signaling.

Several models have been proposed to account for spontaneous oscillatory activity in the mouse retina, including models for rd1 retina (Ly et al., 2022) and for CSNB retina (Hölzel et al., 2023). These models, as well as our model, share the common mechanism based on the properties of AII ACs reported by Choi et al. (2014). Fig. 8 presents a conceptual framework that synthesizes our interpretation of experimental results under various conditions (Fig. 2 in this paper; Trenholm et al., 2012; Choi et al., 2014; Takeuchi et al., 2018), illustrating the proposed relationship between the level of AII AC membrane potential and contributing inputs to AII AC. AII AC exhibits three distinct activity modes: high-frequency fluctuations caused by abundant sodium channels at the initiation site in the depolarized state (Fig. 8, zone A), low-frequency (∼10 Hz) and large-amplitude oscillations in the intermediate range of membrane potential (Fig. 8, zone B), and no oscillations due to sodium channel inactivation in the hyperpolarized state (Fig. 8, zone C). Since the low-frequency oscillations in the intermediate range (zone B) are regarded as the source of spontaneous oscillatory activity observed in degenerated retinas, and since the membrane potential of AII AC in the WT mouse retina is more depolarized than that in the rd1 mouse retina (Choi et al., 2014), AII ACs in the WT mouse retina are presumed to stay in the depolarized state (zone A). Therefore, AII ACs are thought to receive depolarizing inputs from other cell types, and a reduction in these inputs can cause oscillations in AII ACs, leading to spontaneous RGC oscillations. The assumptions of these models are primarily different in regard to the inputs to AII AC. Although other (or previous) models assumed only the gap junctional input from ON CBC to AII AC, our present model considers excitatory inputs from RBCs to AII ACs in addition to the gap junctional input. Considering two excitatory inputs to AII AC enables us to understand physiological results more consistently.

In conclusion, our study suggests a final common mechanism for pathological retinal oscillations, applicable to conditions ranging from severe degeneration to specific genetic alterations like the Trpm1 KO. This relatively simple model, capable of producing a full range of oscillatory behaviors, provides a robust framework for future investigations into secondary dysfunctions, such as pathological RGC oscillations in various retinal diseases.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Christopher J. Lingle served as editor.

We thank Drs. Takahisa Furukawa and Shigetada Nakanishi for the Trpm1 KO and mGluR6 KO mice, respectively, and Dr. Michiko Mandai for the rd1 mouse retina. We also thank Dr. Kirill A. Martemyanov, University of Florida, Gainesville, FL, USA for TRPM1 antibody.

This work was supported by Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research Grants 24H00747, 22KK0137, 19H01140, and 24390019, Takeda Science Foundation, the Kobayashi Foundation, JST PRESTO, and R-GIRO.

Author contributions: Sho Horie: conceptualization, data curation, formal analysis, investigation, methodology, project administration, resources, software, validation, visualization, and writing—original draft, review, and editing. Konan Sakuta: formal analysis and investigation. Keigo Tada: data curation, formal analysis, investigation, and software. Haruki Tokumoto: formal analysis, investigation, visualization, and writing—review and editing. Taketo Nishimoto: data curation, investigation, and validation. Katsunori Kitano: data curation, formal analysis, funding acquisition, investigation, methodology, software, supervision, validation, visualization, and writing—original draft, review, and editing. Masao Tachibana: conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, validation, visualization, and writing—original draft, review, and editing. Chieko Koike: conceptualization, funding acquisition, project administration, resources, supervision, visualization, and writing—original draft, review, and editing.