Abstract
The altered function of peripheral sensory neurons is an emerging mechanism for symptoms of autism spectrum disorders. Visual sensitivities are common in autism, but whether differences in the retina might underlie these sensitivities is not well understood. This includes fragile X syndrome (FXS), which is the most common syndromic cause of autism. We explored retinal function in the Fmr1 knock-out mouse model of FXS. We focused on a specific type of retinal neuron homologous with primate ganglion cells, the “sustained On alpha” retinal ganglion cell, which plays roles in contrast sensing and binocular vision in mice. We found that these cells exhibit changes in dendritic structure and dampened responses to light in male Fmr1 knock-out mice. We show that decreased light sensitivity is due to increased inhibitory input and reduced E–I balance. The change in E–I balance supports the maintenance of circuit excitability similar to what has been observed in the cortex. However, this maintenance also reshapes the tuning of this retinal ganglion cell type. These results show that loss of Fmr1 in the mouse retina affects the sensory function of one retinal neuron type. As other retinal cell types also express Fmr1, FXS may affect the tuning of retinal cells more broadly. Our findings suggest that the retina may be relevant for understanding visual function in FXS.
Significance Statement
Atypical sensory processing underlies some symptoms and experiences of people with autism spectrum disorders. These symptoms may include differences in vision, audition, and sense of touch. In recent years, evidence has emerged that these differences start with the atypical function of neurons in the periphery. However, not much is known about how autism spectrum disorder (ASD) affects the function of the retina. Here, we explored retinal function in a mouse model of a disease strongly linked to ASD, fragile X syndrome (FXS). Our experiments demonstrate that a cell type in the retina has dampened responses to light in the mouse model of FXS. Our work suggests that atypical processing in the retina may contribute to sensory symptoms in FXS.
Introduction
Symptoms of autism spectrum disorder (ASD) arise from many different modules of the brain. Recent findings suggest that some symptoms stem from atypical sensory processing (Robertson and Baron-Cohen, 2017; Falck-Ytter and Bussu, 2023), including in the sensory periphery (Orefice et al., 2016; McCullagh et al., 2020; Yu and Wang, 2023). However, research into vision in ASD has primarily focused on the cerebral cortex (Simmons et al., 2009; Robertson and Baron-Cohen, 2017). There has been less evaluation of earlier processing stages, such as in the retina.
One example where the retina is understudied is fragile X syndrome (FXS). FXS is caused by loss-of-function of the Fmr1 gene on the X chromosome and is the most common genetic cause of autism. In FXS, several visual symptoms could be related to changes in the function of the retina. These include reduced temporal resolution of vision in infants (Farzin et al., 2011), lower contrast sensitivity (Kogan, 2003; Kogan et al., 2004), visual hypersensitivity (Raspa et al., 2018), and sleep/circadian disturbances (Hagerman et al., 2017). Some symptoms are homologous in the mouse model of FXS, which supports the idea that retinal processing could shape symptoms (Saré et al., 2017; Goel et al., 2018; Felgerolle et al., 2019; Perche et al., 2021; Yang et al., 2022).
The retina is a multilayered set of neural circuits. Its intricate wiring comprises over 100 different neuron types (Baden et al., 2018; Vlasits et al., 2019). These circuits support not only conscious visual experience but also reflexive and nonreflexive eye movements, circadian rhythms, mood, and affect (LeGates et al., 2014; Mahoney and Schmidt, 2024). These different roles are accomplished by circuits that convey different channels of light information—such as motion, contrast, and time of day—to the brain. The outputs of the retina are different types of retinal ganglion cells (RGCs). RGCs are axon-bearing neurons that project from the retina to many different brain areas. There are 20–40 different types of RGCs, depending on the species. Each of these carries distinct light information to the brain (Kerschensteiner, 2022). Excitatory and inhibitory inputs to RGCs shape their tuning for specific types of light information.
Loss of ASD-linked genes leads to changes in an array of neuronal properties. This includes morphological and physiological changes in excitatory and inhibitory neurons (Contractor et al., 2021; Deng and Klyachko, 2021; Zhao et al., 2022). Much of the neuronal machinery is shared between neurons in the retina and brain. This includes ASD-linked genes, which are widely expressed in the retina (Fig. 1). Gross physiology of the retina in people with ASD or FXS has revealed slower and lower amplitude retinal responses (Constable et al., 2020; Perche et al., 2021). This biomarker of inner retinal dysfunction was also observed in a mouse model of FXS (Rossignol et al., 2014; Perche et al., 2021). More recently, researchers found that many synaptic genes exhibit altered expression levels in the absence of Fmr1 (Attallah et al., 2024). However, whether and how the function of individual types of RGCs is altered in FXS has not, to our knowledge, been evaluated. Because different RGC types play distinct roles in vision, understanding how individual types are affected by loss of autism-linked genes is important to understanding vision in FXS and ASD.
Here, we found widespread expression of ASD-linked genes in RGCs. We hypothesized that RGCs exhibit atypical function in mouse models of ASD. To test this, we assessed RGC morphology and function in a mouse model of FXS, the Fmr1 knock-out. We focused on the “sustained On alpha” RGC (sOn-α cell) because this cell type is homologous with a primate ganglion cell type. Using electrophysiological recordings and computational modeling, we found changes in cell morphology and a shift in E–I balance. These alter the tuning of sOn-α cells while maintaining excitability at a set maximum. These findings indicate that the atypical function of RGCs may contribute to differences in visual function in FXS.
Materials and Methods
Animals
All procedures were approved by the Animal Care and Use Committee at Northwestern University. For most experiments, male mice on a mixed B6/129 background were bred with female C57/Bl6J mice of either wild type (WT) or Fmr1−/− (B6.129P2-Fmr1tm1Cgr/J, The Jackson Laboratory catalog #003025; Consortium, 1994) genotypes to produce the WT and Fmr1−/y male animals used in this study. We also performed experiments on littermate WT and Fmr1−/y mice produced from crosses of male C57/Bl6J and female Fmr1+/− animals on a C57/Bl6J background and observed no differences between electrophysiological responses according to background strain or littermate status. Therefore, all animals are combined in the Results. Tail biopsies were used to verify the genotype. All mice were between P50 and P75 except where otherwise noted.
Retina dissection
All mice were dark adapted for at least 1 h prior to being killed by CO2 asphyxiation followed by cervical dislocation. Under dim red light, the eyes were enucleated, and retinas were dissected in carbogenated (95% O2, 5% CO2) Ames medium (Sigma-Aldrich or United States Biological). Retinas were aligned using the subretinal vasculature (Wei et al., 2010) and cut into dorsotemporal (DT) and ventrotemporal (VT) sectors, which were mounted on a membrane filter (0.45 µm pore size, Millipore HABG01300) with a less than 1 mm2 hole cut in it. Retinas were maintained in carbogenated Ames media at room temperature (RT) for 30 min before transfer to the recording chamber.
Electrophysiology
Electrophysiological recordings were performed using a previously described setup (Sonoda et al., 2020b). Retinas were perfused with carbogenated Ames media at 33–35°C and visualized using infrared illumination under differential interferance contrast optics to minimize photobleaching of photoreceptors. sOn-α cells were identified by their large, square-ish cell somas (∼20 µm), sustained responses to dim green flashes, and prolonged responses to bright blue flashes (Fig. 1). tOff-α cells were identified by their transient off responses to dim green flashes and large somas. Post hoc staining and imaging confirmed alpha cell identity (see below, Retinal histology) and stratification in the On or Off layer relative to the ChAT bands. Boroscillate pipettes (Sutter Instrument, 3–5 MΩ) were used for all recordings, and whole-cell electrophysiology recordings were performed using a MultiClamp 700B amplifier (Molecular Devices). Data were acquired using a Digidata 1550B amplifier and were collected using pClamp 10 acquisition software (Molecular Devices, RRID:SCR_011323).
For loose cell-attached recordings, pipettes were filled with Ames media, and spikes were recorded in MultiClamp’s voltage-clamp configuration, achieving a minimum seal resistance of at least 30 MΩ. For whole-cell current-clamp recordings, the K+ internal contained the following (in mM): 125 K-gluconate, 2 CaCl2, 2 MgCl2, 10 EGTA, 10 HEPES, 2 ATP-Na2, 0.5 GTP-Na. For K+ internal, KOH was added to achieve a pH of 7.22. For whole-cell voltage-clamp recordings, the Cs+ internal contained the following (in mM): 110 CsMeSO4, 2.8 NaCl, 20 HEPES, 4 EGTA, 5 TEA-Cl, 4 ATP-Mg, 0.3 GTP-Na3, 10 phosphocreatine-Na2, 5 QX-314-Br. For Cs+ internal, CsOH was added to achieve a pH of 7.2, and osmolarity was verified to be 290 mOsm/kg H2O. For both internal solutions, 0.3% neurobiotin and 10 µM Alexa Fluor 594 (Thermo Fisher Scientific) were added to the internal solution for post hoc visualization. For most experiments, one cell was recorded per retina piece to minimize contamination of light responses by repeat stimulation. For cell-attached recordings, if a cell responded to one dim green flash with a response other than a sustained On response (i.e., the cell was not a sustained On cell), the experimenter targeted a second cell in an area of the retina far from that site. Due to the time-consuming nature of blind-patching the alpha cells for electrophysiology, the electrophysiologist was not blinded to genotype during recording.
Visual stimulation and recording protocols
A set of LED lights was used to stimulate the retina through the 60× water-immersion objective, achieving a stimulus circle with a diameter of 440 µm. The shutter was controlled via pClamp, and neutral density filters were used to control the light intensity. For the “dim green” stimulus, the wavelength was 560 nm, the irradiance was 109 photons/cm2/s, and the light was flashed for a duration of 1 s. For the “bright blue” stimulus, the wavelength was 450 nm, the irradiance was 1013.8 photons/cm2/s, and the light was flashed for a duration of 0.250 s. In most cases, both dim green and bright blue light responses were recorded from each cell. The dim green light was always presented first.
For voltage-clamp recordings to collect data for conductance analysis, cells were held at four or five holding potentials spanning from the reversal potential for Cl− (−55 mV) to the reversal potential for cations (0 mV), and the light was flashed for 1 s at each holding potential. Data were acquired at 10 kHz and low-pass filtered at 2 kHz. For bright blue light recordings, the cells were held at −73 mV. For current-clamp recordings, a holding current was applied so that cells were resting at −60 mV, which typically required applying −100 to −200 pA of current.
Retinal histology
For electrophysiology and counting cell densities, retinas were fixed in 4% paraformaldehyde (Electron Microscopy Sciences) in PBS for 30 min at room temperature (RT), followed by three 30 min washes in PBS at RT. Retinas were then blocked overnight at 4°C in PBS with 6% normal donkey serum and 0.3% Triton (blocking solution). Then retinas were incubated in a blocking solution with primary antibodies for 3–4 nights at 4°C. Primary antibodies following electrophysiology were 1 :1,000 streptavidin 546, 1 : 1,000 mouse anti-smi-32 (BioLegend catalog #801701), and 1 : 500 goat anti-ChAT (Sigma-Aldrich catalog #AB144P). Primary antibodies for counting cell densities were 1 : 1,000 mouse anti-smi-32, 1 : 1,000 rabbit anti-RBPMS, and 1 : 500 goat anti-calbindin. Next, retinas were washed three times for 30 min in PBS at RT, followed by overnight incubation at 4°C in a blocking solution with secondary antibodies. Secondary antibodies for electrophysiology were 1 : 1,000 streptavidin 546, 1 : 1,000 Alexa Fluor 647 donkey anti-mouse, and 1 : 500 Alexa Fluor 488 donkey anti-goat. Secondary antibodies for cell counts were 1 : 500 Alexa Fluor 488 Donkey anti-goat, 1 : 1,000 Alexa Fluor 647 donkey anti-mouse, and 1 : 500 Alexa Fluor 546 donkey anti-rabbit. Finally, retinas were washed three times for 30 min in PBS and mounted on slides using Fluoromount (Sigma-Aldrich). Retinas were imaged on a confocal microscope (Leica DM5500 SPE, Leica Microsystems) under a 20× objective. See detailed list of materials in Table 1.
Materials for immunohistochemistry
RNA fluorescence in situ hybridization
For combined immunohistochemistry and in situ hybridization, retinas were prepared either as whole mounts or sections for staining. For sections, tissue was prepared as previously described (Sonoda et al., 2020a). Briefly, the eye was enucleated and fixed in 4% nuclease-free PFA for 6 h at room temperature. Eyes were washed in PBS and then immersed in 30% sucrose in PBS overnight at 4°C. Tissue was frozen in optimal cutting temperature on dry ice and stored at −20°C until sectioning. Subsequently, 20 µm sections were made using a Leica CM1950 cryostat. Sections were dried overnight at room temperature. The tissue was processed according to the RNAscope Multiplex Fluorescent v2 Assay (Advanced Cell Diagnostics) instructions provided by the manufacturer.
For wholemount retinas, retinas were fully dissected in PBS and incubated in 4% PFA for 1 h, followed by three washes in PBS. Tissue was dehydrated through a graded MeOH/PBT series: 50 and 75% for at least 5 min each at room temperature. Then tissue was transferred to 100% MeOH and stored overnight at −20°C for storage. Tissue was rehydrated by a reverse MeOH/PBT series and then washed in PBS. Protease Plus was applied three times for 7 min at 40°C and then washed again in PBS. Then tissue was processed according to the RNAscope Multiplex Fluorescent v2 Assay (Advanced Cell Diagnostics) instructions provided by the manufacturer.
Following the completion of the RNAscope protocol, immunohistochemistry staining was performed as described above. For wholemount retinas, the probes used were Fmr1 (Advanced Cell Diagnostics Bio catalog #496391) and Opn4 (Advanced Cell Diagnostics catalog #438061). smi-32 antibody (BioLegend catalog #805501) at 1 : 400 was used for primary staining followed by 1 : 400 Alexa Fluor 647 donkey anti-mouse secondary. For sections, probes used were Fmr1 (Advanced Cell Diagnostics Bio catalog #496391), Spp1 (Advanced Cell Diagnostics Bio catalog #435191), and Opn4 (Advanced Cell Diagnostics catalog #438061). Retinas were imaged on a confocal microscope (Leica DM5500 SPE, Leica Microsystems) under a 40× objective.
Analysis
Analysis was performed using Excel (Microsoft) and Jupyter Notebook running Python (v 3.9.7). For plotting, the seaborn package was used (Waskom, 2021). For statistical analysis, we used scipy (Virtanen et al., 2020) and pingouin (Vallat, 2018) packages. Unless otherwise stated, reported p-values are the result of two-tailed Student's t tests. Power analysis was performed using pingouin. Where possible, minimum effect sizes were estimated from data available empirically.
mRNA puncta counting
To estimate the number of puncta present in our fluorescence in situ hybridization images, we performed difference of Gaussians filtering on images to isolate puncta between 0.3 and 2 µm. Then we applied a standard intensity threshold across all images and used the ImageJ “Analyze Particles” function to count the number of puncta.
Cell counting
To estimate the density of cells in the sOn-α cells in WT and Fmr1−/y retinas, we sampled at eight locations throughout the retina—four locations close to the optic nerve and four locations distal to the optic nerve, equally distributed around the retina. Image stacks were collected using a 20× objective. Cells that were double positive for calbindin, and smi-32 were counted by a blinded observer (Sonoda et al., 2020b).
Electrophysiology
Cells were discarded from analysis if they did not meet the criteria to be considered α cells, if they were not responsive to light, or if the access resistance changed during the recording or was higher than 50 MΩ. pClamp abf files were imported into Python using the pyABF package (Harden, 2022). Custom scripts were used for spike detection, membrane property estimation, FWHM measurements, spike threshold and amplitude measures, and conductance analysis (based on Vlasits et al., 2014). Series resistance compensation was performed post hoc for all voltage-clamp recordings and voltages were corrected for the liquid junction potential (−10.05 mV).
The E–I ratio was calculated as Ge / (Gi + Ge) at each timepoint after conductances were denoised using a Savitzky–Golay filter. In some cases, the conductance values went below zero due to the high spontaneous baseline in retinal cells. Thus, we artificially shifted the baseline conductance by a factor of 4 to reduce the presence of negative values in the E–I ratio calculation.
Transcriptomic datasets
We collected a list of ASD-linked genes by searching PubMed for reviews on ASD-linked genes and from the Simons Foundation Autism Research Initiative list of mouse models of ASD (https://www.sfari.org/resource/mouse-models/). Most of the transcriptome datasets were accessed from singlecell.broadinstitute.org, except the single-cell RNAseq dataset, which was provided by that study's authors (Goetz et al., 2022). Data for ASD-linked genes were imported into Python using pandas (Reback and Team, 2020) and plotted using seaborn.
Anatomical tracing and analysis
Analysis of anatomical data was performed by experimenters blinded to genotype. Confocal image stacks of neurobiotin-labeled sOn-α and tOff-α cells were imported into Fiji (Schindelin et al., 2012) for analysis. We used the neurite tracer toolbox SNT to trace dendritic arbors, perform Sholl analysis, and measure the total dendritic length (Arshadi et al., 2021). The maximum radius and the dendritic diameter were manually measured in Fiji. To localize cells relative to the optic nerve, the coordinates of the optic nerve, cell, and cut edges of the retina were located and used to measure and align each retina piece using custom scripts in Python.
Parallel conductance model
Light-evoked postsynaptic potentials (PSPs) were predicted with a parallel conductance model (Antoine et al., 2019), which we implemented in Python using a Jupyter Notebook. The model predicted Vm using the parallel conductance equation:
Plateau index
The plateau index was defined as follows:
Results
Widespread expression of ASD-linked genes including Fmr1 in the retina
To examine which neuronal cell types in the retina might be most affected by loss of ASD-linked genes, we examined previously published transcriptomic data from the mouse and primate retina (Shekhar et al., 2016; Tran et al., 2019; Yan et al., 2020; Goetz et al., 2022). We identified a list of ASD-linked genes from the literature and explored the mRNA expression of those genes across retinal cell types (Fig. 1A,B; Extended Data Fig. 1-1). We noted from this analysis that one ASD-linked gene, Tbr1, has already been identified as a regulator of type-specific retinal ganglion cell (RGC) development (Liu et al., 2018). Several other commonly studied genes implicated in autism and related disorders, including Fmr1, Cntnap2, and Mecp2, exhibited high expression in retinal ganglion cells. Other genes, like Shank3 and Syngap1, exhibited sparser expression. Overall, we observed type specificity in the relative amount of expression of each ASD-linked gene and percent of cells expressing a given gene. Retinal ganglion cells and amacrine cells expressed more autism-linked genes than bipolar cells. This suggests that loss of ASD-linked genes may have differential effects on the function of different cell types in the retina.
Widespread expression of ASD-linked genes including Fmr1 in the retina. See also Extended Data Figure 1-1. A, Type-specific expression of ASD-linked genes in mouse retinal ganglion cells from a published single-cell RNAseq dataset (Tran et al., 2019). Color: relative expression level within each gene. Dot size: percent of cells of each type expressing that gene. Arrow: the sOn-α RGC. B, Type-specific expression of ASD-linked genes in mouse retinal ganglion cell types from a published patch-cell RNAseq dataset (Goetz et al., 2022). Color: relative expression level within each gene. Dot size: percent of cells of each type expressing that gene. C, Left, Comparison of Fmr1 expression in two different mouse transcriptomes from A and B. Dots: RGC types' relative Fmr1 expression in the Goetz et al. (2022) versus Tran et al. (2019) datasets. Color indicates whether the cell type is in the On, Off, or On–Off family. Right, List of cell types expressing Fmr1 in both datasets. Each row shows the cell type nomenclature for each dataset for matched cell types. D, Top, Confocal images of the ganglion cell layer of wholemount retinas showing both fluorescence in situ hybridization (FISH) of Fmr1 (magenta) and antibody staining against smi-32 (cyan), which labels alpha ganglion cells and vasculature. First row: WT mouse, second row: Fmr1−/y. Bottom, # of Fmr1 puncta per 100 µm2 for a WT (black) and Fmr1−/y (cyan) retina at different depths in the ganglion cell layer (GCL) and inner plexiform layer (IPL). E, Confocal image of sliced retina showing FISH for Fmr1 (magenta), Opn4 (green), and Spp1, a marker of sOn-α cells and M2 cells (cyan). A putative sOn-α cell weakly expresses Opn4 and is positive for Spp1 (arrow). A putative M1 ganglion cell strongly expresses Opn4 and does not express smi-32 (asterisk). Right, Insets of the putative sOn-α cell. F, Confocal images of wholemount retinas at three different developmental ages. Images show FISH for Fmr1 (magenta) and Opn4 (green) and antibody staining against smi-32 (cyan).
Figure 1-1
Autism-linked gene expression in the mouse and primate retina. Related to Figure 1.
A) Type-specific expression of ASD-linked genes in mouse bipolar cell types from a published bipolar cell transcriptome (Shekhar et al. 2016).
B) Type-specific expression of ASD-linked genes in mouse amacrine cell types from a published amacrine cell transcriptome(Yan et al. 2020).
C) Type-specific expression of ASD-linked genes in primate retinal ganglion cell types from a published retinal ganglion cell transcriptome (Peng et al. 2019).
Download Figure 1-1, TIF file.
The widespread expression of Fmr1 across cell classes and in both mice and primates was of particular interest because Fmr1 is the gene involved in fragile X syndrome, the most common genetic cause of autism. We compared Fmr1 RNA expression across two mouse transcriptomic datasets, one obtained using the “dropseq” method (Tran et al., 2019) and the other the “patchseq” method (Goetz et al., 2022), which is more conservative due to lower sample sizes and hit rates. We found that a large group of On and On–Off ganglion cell types exhibited Fmr1 expression in both datasets (13 of the ∼40 ganglion cell types; Fig. 1C). These include multiple types of direction-selective ganglion cells as well as the sustained On α retinal ganglion cell (sOn-α cell).
For the remainder of this study, we decided to focus on the sOn-α cell. This cell type has recently been identified as the mouse ganglion cell type homologous to the On midget ganglion cell type (Hahn et al., 2023). The midget cell types are the predominant ganglion cell types in the primate fovea. Thus, in primates, this cell type's function is crucial for high-resolution vision. In addition, this type exhibited relatively high expression of most ASD-linked genes, including Fmr1 (Fig. 1A). Beyond these reasons, the sOn-α cell is a well-studied RGC type (Krieger et al., 2017) with identified roles in visual behaviors (Schmidt et al., 2014; Kim et al., 2020; Johnson et al., 2021), which is easy to identify due to its large cell soma (Extended Data Fig. 2-1; Bleckert et al., 2014). We confirmed that the sOn-α cell expresses Fmr1 RNA using fluorescence in situ hybridization. We observed widespread Fmr1 puncta in the ganglion cell layer of the wild-type (WT) mouse retina that were low or absent in an Fmr1−/y negative control (Fig. 1D). In WT retinas, we observed puncta in sOn-α cells (Fig. 1E). Furthermore, we observed Fmr1 RNA in sOn-α cells during development (both before eye opening at P9 and after eye opening at P35) and in adulthood (P59; Fig. 1F). This suggests that Fmr1 could play a role in the development and function of the sOn-α cell.
sOn-α cells have denser dendrites in Fmr1−/y mice
To determine whether loss of Fmr1 affects the development of sOn-α cells, we first analyzed the dendritic morphology of sOn-α cells in isolated retinas from WT and Fmr1−/y mice. We identified sOn-α cells based on physiological and molecular features that are unique to this cell type (see Materials and Methods; Extended Data Fig. 2-1; Contreras et al., 2023). We filled these cells with neurobiotin to visualize their dendrites (Fig. 2A,B). sOn-α cells in Fmr1−/y retinas show denser arbors as measured by Sholl analysis, even though their dendritic diameters were not different between genotypes (Fig. 2C–E, n = 16 WT cells from 10 mice, 13 Fmr1−/y cells from 5 mice). The maximum number of crossings was higher (Fig. 2F; WT, 24 ± 3 crossings; Fmr1−/y, 29 ± 5 crossings; p = 0.005), and the total dendritic length was higher in Fmr1−/y retinas compared with WT (Fig. 2H,I; WT, 3,942 ± 923 µm; Fmr1−/y, 4,799 ± 1,043 µm; p = 0.032), while the maximum radii were not significantly different between the two populations of cells (Fig. 2G; WT, 182 ± 36 µm; Fmr1−/y, 186 ± 30 µm). Occasionally, we filled other ganglion cell types with dye while looking for sOn-α cells. The type encountered most frequently was the “transient Off alpha” (tOff-α) cell, a cell type with lower expression of Fmr1 (Fig. 1). We traced the dendrites of our samples of this cell type (Fig. 2J; n = 8 WT cells from six mice, 4 Fmr1−/y cells from four mice). We did not detect a statistical difference in the dendritic density between WT and Fmr1−/y (Fig. 2K,L; WT, 28 ± 7 crossings; Fmr1−/y, 26 ± 11 crossings; p = 0.64; statistical power, 0.39); however, given the low statistical power of our sample, further data are needed to reach a conclusion regarding tOff-α cells. Whether other cell types are affected by loss of Fmr1 or not remains to be determined.
sOn-α retinal ganglion cells have denser dendrites in Fmr1−/y mice. See also Extended Data Figure 2-1. A, Z-projection of a confocal image showing neurobiotin-filled dendrites of an sOn-α cell in an Fmr1−/y retina. B, Traced skeletons of two example sOn-α cells from an Fmr1−/y mouse and WT mouse. C, Sholl analysis of the example cells in B. D, Dendritic diameter of a population of sOn-α cells plotted as a function of the nasal–temporal position on the retina. Dots with circles around them: example cells from B. Gray line: linear regression to the entire population, R2 = 0.37. E, Mean Sholl analysis for populations of sOn-α cells from WT and Fmr1−/y retinas. Shaded regions: standard error. Dotted lines: region used for analysis in F. F, Maximum number of crossings in sOn-α cells from WT and Fmr1−/y mice. Dots are individual cells, line is the mean. * indicates statistical significance. G, Maximum radii of sOn-α cells from WT and Fmr1−/y mice. n.s. = not significant. H, sOn-α cells plotted on their retina coordinates, with dot size indicating the total dendritic length. Circled dots: example cells from B. I, Total dendritic lengths of cells within each retinal quadrant: dorsotemporal (DT), ventrotemporal (VT), dorsonasal (DN), and ventronasal (VN). J, Traced skeletons of two tOff-α cells from Fmr1−/y and WT mice. K, Mean Sholl analysis for populations of sOn-α cells from WT and Fmr1−/y retinas. Shaded regions: standard error. Dotted lines: region used for analysis in O. L, Maximum number of crossings in tOff-α cells from WT and Fmr1−/y mice. Dots are individual cells, line is the mean. t test, not significant. M, Confocal images of wholemount retinas stained against calbindin (magenta) and smi-32 (cyan) to label sOn-α cells. Dotted circles: identified double-labeled cells. N, Density of sOn-α cells as a function of temporal (T)-to-nasal (N) position in the retina. Each dot is a sampled position. N = 2 WT, 4 Fmr1−/y mice. Linear fits exhibited little difference between WT and Fmr1−/y. O, Maximum density in each retina. n.s.: not significant.
Figure 2-1
Identification of sOn-α cells. Related to Figures 1, 2 We identified putative sOn-α cells based on four features that, together, uniquely identify them.
A) Cell attached electrophysiological recording from an sOn-α cell during presentation of a dim green light stimulus presented for 1 s. sOn-α cells exhibit characteristic high baseline firing rates and a sustained increase in firing when the light is turned on. Same data as in Fig. 3B.
B) sOn-α cells are a type of intrinsically photosensitive melanopsin-positive ganglion cell. They exhibit prolonged firing in response to brief flashes of bright blue light, which optimally triggers the melanopsin expressed on their membranes. We checked for their melanopsin-dependent intrinsic photosensitivity by presenting a brief flash of bright blue light (blue rectangles) in either cell-attached (top two rows) to record the prolonged firing (note timescale) or voltage clamp (bottom row) to record the prolonged inward current. Recordings from three different cells are shown.
C) sOn-α cells have large somas and are positive for smi-32. We filled recorded cells with neurobiotin and fixed and stained for smi-32. Here, a confocal image of an example sOn-α cell filled with neurobiotin (NB) and labeled with an antibody for smi-32.
Download Figure 2-1, TIF file.
The density of sOn-α cells depends on retinal location, with a higher density of cells in the temporal retina supporting high-resolution binocular vision (Bleckert et al., 2014; Johnson et al., 2021). We therefore evaluated whether the density of ganglion cells is altered by loss of Fmr1 (Fig. 2M). We found that the nasal–temporal gradient is similar in WT and Fmr1−/y retinas (Fig. 2N) and that the maximum density observed was not significantly different between WT and Fmr1−/y retinas (Fig. 2O; p = 0.92).
Overall, these results suggest that sOn-α cells in the Fmr1−/y mouse are developing denser dendrites within overall typical arbor sizes and cell density, which could lead to altered synaptic wiring and function of these cells.
Reduced spiking responses to dim light flashes in sOn-α cells
The increased dendritic density of sOn-α cells in Fmr1−/y mice prompted the hypothesis that sOn-α cells might have more synaptic drive and therefore stronger responses to light. To test this, we performed cell-attached recordings of sOn-α cells and presented dim-photopic green-spectrum LED light (wavelength, 560 nm; “dim green light”) to measure rod and cone-mediated light responses (Fig. 3A). Because sOn-α cells vary in their dendritic field size and visual response properties depending on their location in the retina (Bleckert et al., 2014; Sonoda et al., 2020b; Fig. 2D), we confined our recordings to the temporal half of the retina. As suggested by their name, sOn-α cells in WT retinas exhibit an initial peak response to an increase in luminance (“On” response) followed by sustained firing for the duration of the light stimulus (Fig. 3B,C; n = 17 WT cells from 7 mice, 29 Fmr1−/y cells from 10 mice). Fmr1−/y cells still exhibited an overall On-sustained firing pattern and had similar baseline firing rates to WT (WT, 28 ± 20 Hz; Fmr1−/y, 31 ± 21 Hz; two-sided t test, p = 0.600). However, surprisingly, during the light response, the Fmr1−/y cells had lower firing rates compared with WT cells during both the peak (WT, 162 ± 69 Hz; Fmr1−/y, 92 ± 68 Hz; two-sided t test, p = 0.002) and sustained (WT, 60 ± 29 Hz; Fmr1−/y, 31 ± 33 Hz; p = 0.005) period of the response (Fig. 3D). Lower firing rates were observed in the Fmr1−/y cells compared with WT in both the dorsal and ventral retina (Fig. 3E,F), despite differences in green light sensitivity between these retinal regions (Szatko et al., 2020). These results demonstrate that the sOn-α cell light response is weaker in Fmr1−/y mice despite having more elaborate dendrites.
sOn-α cells have reduced responses to dim light flashes. A, Experimental setup showing microscope objective, electrophysiology electrode, perfusion chamber with the retina, and the path of the visual stimulus. The dim green stimulus was an LED source with the listed wavelength and intensity. B, Cell-attached responses of sOn-α cells to dim green light. Left, Wild type (WT). Right, Fmr1−/y knock-out. Top, Example recordings. Bottom, Firing rate of four repetitions (light gray) and their mean (black). Timing of stimulus: green boxes. C, Population mean of firing rates for WT and Fmr1−/y cells. Shaded region is s.d. D, Left, Population mean firing rates for WT (black) and Fmr1−/y (turquoise) overlaid to indicate regions used to calculate summary data. Right, Baseline, peak, and sustained firing rates for individual cells (dots) and the mean (lines). * indicates significantly different means. E, sOn-α cells plotted on their retinal coordinates, with dot size indicating the peak firing rate. All cells were located in the temporal retina. F, Peak firing rate as a function of retinal location.
Subtle differences in sOn-α cell-intrinsic properties in Fmr1−/y mice
The reduced light sensitivity of sOn-α cells in Fmr1−/y mice could be the result of changes in a variety of different physiological properties of these cells or changes in their neural circuits. First, we tested whether excitability of sOn-α is altered in Fmr1−/y mice and found no difference between the current-spiking functions of WT versus Fmr1−/y sOn-α cells (Sonoda et al., 2018; Fig. 4A–C; n = 11 WT cells from six mice, 14 Fmr1−/y cells from nine mice; two-way repeated measure ANOVA, p < 0.01 for injected current, not significant for genotype or interaction). Second, we examined the properties of the action potentials themselves (Fig. 4D–H; n = 10 WT cells from six mice, 13 Fmr1−/y cells from nine mice). While the spike width [full-width at half-maximum (FWHM), Fig. 4E] was not different between WT and Fmr1−/y retinas (WT, 0.423 ± 0.080 ms; Fmr1−/y, 0.465 ± 0.096 ms; p = 0.335), the spike amplitude was significantly lower (Fig. 4F; WT, 67.6 ± 5.73 mV; Fmr1−/y, 58.4 ± 9.08 mV; p = 0.008). However, the resting potential and spike threshold were not different between WT and Fmr1−/y (Fig. 4G; Vrest: WT, −55.3 ± 3.27 mV; Fmr1−/y, −54.0 ± 4.45 mV; Vthresh: WT, −49.9 ± 3.38 mV; Fmr1−/y, −48.4 ± 5.11 mV). In sOn-α cells, spike amplitude increases with developmental age (Lucas and Schmidt, 2019). Here, we found that, while spike amplitude increases with age in WT mice, this relationship is less clear in the Fmr1−/y cells (Fig. 4H).
Subtle differences in intrinsic properties of sOn-α cells. A, Current-clamp recordings from two example sOn-α cells showing spiking in response to increasing current injections indicated by the protocol at the top. B, Firing rate as a function of the amount of current injected for the two example cells in A. C, Average firing rates as a function of current injected for sOn-α cells in WT and Fmr1−/y mice. Vertical lines: s.d. D, Example action potentials from two sOn-α cells. Multiple action potentials from each cell are overlaid. The vertical line indicates the full-width at half-maximum (FWHM) measured in E. E, Average FWHM for sOn-α cells from WT and Fmr1−/y retinas. Population mean: lines. F, Spike amplitude from baseline for sOn-α cells from WT and Fmr1−/y mice. * indicates a significant difference. G, Resting membrane potential (Vrest) and threshold potential for spiking (Vthresh) for sOn-α cells from WT and Fmr1−/y retinas. H, Spike amplitude as a function of postnatal age of the animal. Black line: fit to the WT data. R2 = 0.53. Circles: example cells from D. I, Voltage-clamp recordings of bright blue stimulus-evoked currents in two example sOn-α cells (top, middle) and the population mean (bottom). Early and late epochs used for analysis are indicated. Stimulus period: blue rectangle. See also Extended Data Figure 2-1. J, Peak response during the stimulus period for sOn-α cells from WT and Fmr1−/y retinas. n.s. = not significant. Lines: population mean. K, Mean current change during early and late periods indicated in H. L, Mean current change during the early epoch (dot size) plotted on retina coordinates. M, Mean current change during the early epoch in the ventrotemporal (VT) and dorsotemporal (DT) regions of the retina.
We found greater dendritic density in sOn-α cells in the Fmr1−/y mice (Fig. 2). Therefore, we hypothesized that intrinsic properties that depend on membrane area could be atypical. sOn-α cells are intrinsically photosensitive retinal ganglion cells expressing the photopigment melanopsin (type M4, Fig. 1; Schmidt et al., 2014). We wondered whether the amount of melanopsin-mediated intrinsic photosensitivity in these cells could be increased given the cells' longer dendritic lengths (Fig. 2). The melanopsin photocurrent is different from rod/cone evoked synaptic currents in that its duration is very prolonged, even when presented with a very short light pulse (Extended Data Fig. 2-1; Contreras et al., 2023). We observed that sOn-α cells in Fmr1−/y retinas had slightly larger, more prolonged blue light-evoked currents (Fig. 4I–L, n = 8 WT cells from six mice, 8 Fmr1−/y cells from six mice; WT, −69 ± 34 pA; Fmr1−/y, −142 ± 87 pA; p = 0.045). However, this increased current in response to bright blue light cannot explain the lower firing rates in dimmer light conditions (Fig. 3).
Increased synaptic inhibition to sOn-α cells in Fmr1−/y mice
We examined whether differences in excitatory or inhibitory inputs could contribute to the reduced firing in response to light flashes. We performed voltage-clamp recordings from sOn-α cells to extract the time course of the excitatory (Ge) and inhibitory (Gi) conductances (Fig. 5A,B). We observed that while Ge was similar in WT and Fmr1−/y cells, Gi in the Fmr1−/y cells was significantly larger than the typical inhibitory conductance in WT (Fig. 5B–D; n = 8 WT cells from five mice, 11 KO cells from seven mice; for Ge, WT 160 ± 125 nS*s, Fmr1−/y 189 ± 187 nS*s; p = 0.69; for Gi, WT 313 ± 310 nS*s, Fmr1−/y 1,037 ± 447 nS*s; p = 0.0006). The difference in Gi persisted throughout the time course of the response, with both the transient (Fig. 5E; WT, 5.80 ± 4.85 nS; Fmr1−/y, 17.47 ± 7.33 nS; p = 0.0007) and sustained (WT, 2.81 ± 2.86 nS; Fmr1−/y, 9.81 ± 4.70 nS; p = 0.0009) periods exhibiting a significantly greater inhibitory conductance in Fmr1−/y cells compared with WT. We also observed a small, but still significant, increase in the excitatory conductance in sOn-α cells in Fmr1−/y retinas, which was restricted to the Off (WT, −1.32 ± 0.77 nS; Fmr1−/y, 0.30 ± 1.17 nS; p = 0.002) portions of the response. Neither the transient nor the sustained component of the On response was affected (transient, WT 3.75 ± 2.06 nS, Fmr1−/y 3.80 ± 3.15 nS; p = 0.97; sustained, WT 1.27 ± 1.10 nS, Fmr1−/y 1.69 ± 1.71 nS; p = 0.53). Most sOn-α cells in the Fmr1−/y retinas exhibited larger inhibitory conductances compared with excitatory conductances in response to light flashes (Fig. 5F), suggesting that Fmr1−/y sOn-α cells may have altered excitatory–inhibitory (E–I) ratio. Overall, these results suggest that the decreased spiking in response to light flashes in sOn-α cells in the Fmr1−/y mice is due to increased inhibitory input to these cells.
Heightened synaptic inhibition in sOn-α cells from Fmr1−/y mice. A, Voltage-clamp recordings from an example sOn-α cell in an Fmr1−/y retina showing response to dim green light stimulus (green rectangle) at multiple holding potentials. B, Results of conductance analysis for two example sOn-α cells from WT (black) and Fmr1−/y (teal) retinas. Left, Excitatory conductance (Ge). Right, Inhibitory conductance (Gi). C, Mean Ge and Gi for a population of sOn-α cells from WT and Fmr1−/y retinas. Epochs for transient (“trans.”), sustained (“sus.”), and light off (“off”) periods are indicated. Lines: group means. D, Integrated Ge (top) and Gi (bottom) for WT and Fmr1−/y sOn-α cells. n.s. = not significant, * indicates statistical significance. E, Summary values for Ge (top) and Gi (bottom) in WT and Fmr1−/y cells. “Trans.”: maximum average conductance during the transient period. “Sus.”: mean during the sustained period. “Off”: mean during the off period. F, Relationship between integrated Ge and integrated Gi in WT and Fmr1−/y cells. G, The mean excitatory–inhibitory ratio (E–I ratio) over the time course of the stimulus for WT and Fmr1−/y cells. H, Population summary of the E–I ratio during transient and sustained periods of the stimulus response. I, Spontaneous activity during the baseline period in example sOn-α cells from WT and Fmr1−/y retinas at holding potentials that isolate excitatory (−60 mV) and inhibitory (0 mV) currents. J, Standard deviation (s.d.) of spontaneous excitatory and inhibitory recordings in sOn-α cells from WT and Fmr1−/y retinas. K, Input resistance (Rin), membrane capacitance (Cm), holding current (Ih) in cesium (Cs+) internal, and holding current (Ih) in potassium (K+) internal for sOn-α cells in WT and Fmr1−/y retinas.
Altered E–I balance is a physiological change commonly described in models of ASD (Rubenstein and Merzenich, 2003; Sohal and Rubenstein, 2019). To explore this further, we calculated the E–I ratio of the light response (Fig. 5G; see Materials and Methods). We found that the E–I ratio was lower (indicating more inhibition relative to excitation) during both transient (WT, 0.58 ± 0.04; Fmr1−/y, 0.51 ± 0.08; p = 0.030) and sustained (WT, 0.46 ± 0.07; Fmr1−/y, 0.32 ± 0.11; p = 0.006) periods of the response to the light flash (Fig. 5H). The decreased E–I ratio in sOn-a cells suggests that there may be an increased number of inhibitory synaptic inputs onto these cells in Fmr1−/y retinas and consequent increase in the frequency of spontaneous currents. Spontaneous synaptic activity is very high in retinal neurons, so we could not isolate individual miniature currents in our recordings. As a measure of spontaneous activity, we therefore quantified spontaneous activity by measuring the standard deviation of the spontaneous currents (Vlasits et al., 2014; n = 11 WT cells from five mice, 12 KO cells from seven mice). We found that spontaneous excitatory currents were not significantly different in WT versus Fmr1−/y mice (WT, 50.91 ± 15.05 pA; Fmr1−/y, 44.82 ± 23.18; p = 0.464) but that spontaneous inhibitory currents had higher standard deviations in Fmr1−/y cells compared with WT cells (Fig. 5I,J; WT, 35.44 ± 14.86 pA; Fmr1−/y, 71.72 ± 41.35; p = 0.013). These results suggest that there may be more inhibitory synapses on sOn-α cells in Fmr1−/y. This interpretation is supported by a decrease in input resistance in sOn-α cells (Fig. 5K; n = 18 WT cells from eight mice, 23 Fmr1−/y cells from nine mice; Rin, WT 164 ± 63 MΩ, Fmr1−/y 113 ± 46; p = 0.007; Cm, WT 26 ± 16 pF, Fmr1−/y 22 ± 11; p = 0.464), which would be expected in the presence of increased inhibitory input and opening of chloride channels. We also observed high variability in the current required to hold cells at −60 mV; however, there was no significant difference between genotypes (Fig. 5K; for cesium internal, n = 18 WT cells from eight mice, 23 Fmr1−/y cells from nine mice; for potassium internal, n = 9 WT cells from five mice, 13 Fmr1−/y cells from seven mice; Ih in Cs+, WT −173 ± 95 pA, Fmr1−/y −154 ± 103; p = 0.528; Ih in K+, WT −131 ± 75 pA, Fmr1−/y −140 ± 69; p = 0.787). Overall, our results suggest that loss of Fmr1 in the retina alters circuits of sOn-α cells, leading to changes in their inhibitory synaptic inputs.
A conductance model predicts a reduced E–I ratio to stabilize postsynaptic potentials
We found that sOn-α cells in the Fmr1−/y mouse have larger inhibitory conductances and lower E–I ratio than WTs. This result contrasts with changes observed in other areas of the brain, where the E–I ratio is often found to be higher in ASD models (Contractor et al., 2015; Sohal and Rubenstein, 2019). Recently, Antoine et al. (2019) proposed that an increased E–I ratio in the cortex may serve as a compensatory mechanism to stabilize spiking. Using a parallel conductance model, they demonstrated that, as excitatory and inhibitory conductances scale down, the E–I ratio must increase to maintain stable postsynaptic potentials (PSPs).
We explored whether this model fits with our data by replicating the parallel conductance model using the average excitatory and inhibitory conductances from WT sOn-α cells (Fig. 6A–C). First, we compared the “Native” WT conductance scaled to two alternate cases: “Stable” scaling, in which the E–I ratio is maintained with both Gex and Gin scaled up three times, and “Decreased” scaling, in which the E–I ratio decreases through scaling Gin by three times and Gex by 1.1×, as more typically observed in our data for Fmr1−/y cells (Fig. 6D). Our model predicts that the PSPs in the “Native” and “Decreased” models have similar peak amplitudes, while the “Stable” scaled model has a higher peak amplitude (Fig. 6E). We explored the parameter space of conductance scaling and found that when both Gex and Gin are scaled up, the E–I ratio must decrease to maintain a stable PSP (Fig. 6F). Thus, the parallel conductance model provides evidence of a consistent trend in E–I ratios observed in the cortex and retina in ASD models, regardless of whether excitation and inhibition are scaling up or down. Loss of Fmr1 may have differential effects on synaptic scaling in different brain areas, but common compensatory mechanisms to stabilize postsynaptic potentials may be in place.
A conductance model predicts a reduced E–I ratio to stabilize postsynaptic potentials. A, Schematic of the parallel conductance model. B, Average excitatory (Gex) and inhibitory (Gin) input from WT sOn-α cells used in conductance model. C, Predicted membrane voltage from the conductance model for three conditions: the excitatory postsynaptic potential predicted from Gex, the inhibitory postsynaptic potential (IPSP) predicted from Gin, and the overall postsynaptic potential (PSP) using both Gex and Gin. D, Native and scaled excitatory and inhibitory conductances. “Stable” E–I ratio scales the conductances by the same scaling factor (3×), while “decreased” E–I ratio scales up Gin (3×) compared with Gex (1.1×). E, Model prediction for the membrane voltage Vm for each of the conductance conditions in D. F, Heat map of the change in the postsynaptic potential (PSP) peak for different scaling factors of Gex and Gin compared with the Native case. The three conditions in D are shown with white symbols. Circle: native; square: stable; triangle: decreased. Red dotted line: stable E–I ratio. Blue dots: where PSP is <0.5 mV different from the Native case. G, PSPs predicted from conductances from example WT and Fmr1−/y cells. H, The predicted peak PSP for a population of WT and Fmr1−/y cells. * indicates significantly different means. I, Current-clamp recordings from example WT and Fmr1−/y cells showing original recordings in gray and filtered PSPs in black/teal. J, Mean PSPs from WT and Fmr1−/y cells showing the time periods used to calculate summary data in K. K, Peak and sustained PSPs from WT and Fmr1−/y cells. * indicates significantly different means. n.s., not significant. L, Plateau index calculated from PSPs measures in K. * indicates significantly different means.
Next, we predicted PSPs from the cells for which we had measured synaptic conductance (Fig. 5) by running our model using the individual excitatory and inhibitory conductances measured from each of the WT and Fmr1−/y cells in our dataset. We found that overall, the predicted peak PSP is significantly reduced and more variable in Fmr1−/y cells compared with WT (Fig. 6G,H; n = 8 WT cells from five mice, 11 KO cells from seven mice; peak PSP, WT 23.51 ± 4.21 mV, Fmr1−/y 16.55 ± 8.45 mV; p = 0.032). These results show that even though inhibitory conductances scale up in Fmr1−/y cells compared with WT, the E–I balance limits the PSP amplitude to amplitudes at or below the WT amplitudes. In some cases, this comes along with losing the “sustained” character of the cell’s response (Fig. 6G), making responses predicted from Fmr1−/y conductances more transient.
We verified the results of our model by recording sOn-α cells’ light responses in the current-clamp mode (n = 8 WT cells, 13 Fmr1−/y cells; Fig. 6I,J). We found that the peak PSPs of Fmr1−/y cells were not significantly different from WTs (Fig. 6K; WT, 15.8 mV ± 5.1 mV; Fmr1−/y, 13.0 ± 5.6 mV; p = 0.254, power = 0.22), although this sample is underpowered. However, as predicted by our model, the sustained component of the PSP was significantly lower in Fmr1−/y cells (WT, 10.6 ± 4.3 mV; Fmr1−/y, 5.8 ± 5.3 mV; p = 0.04), although not as dramatically as observed in the model. These underestimations may simply be due to our underpowered sample. Alternatively, our approach to estimating the PSPs, which did not apply sodium channel blockers to block action potentials, may underestimate PSP amplitude when firing rates are high. Finally, we calculated the plateau index for each cell, where a higher value indicates a more sustained response, and found that Fmr1−/y cells have an overall lower plateau index and more variability (Fig. 6L; WT, 0.68 ± 0.16; Fmr1−/y, 0.45 ± 0.23; p = 0.01). Therefore, loss of Fmr1 may fundamentally change the light information sOn-α cells encode.
Discussion
Here, we found that visual deficits observed in fragile X syndrome may arise, at least in part, at the earliest stages of visual processing. We find that sOn-α RGCs in mouse Fmr1−/y retinas show changes in dendritic morphology and damped light responses. The atypical light responses arise due to changes in excitation–inhibition ratio. Collectively, our results open up new avenues to understand the origin of sensory symptoms in ASD. Future research should focus on exploring how additional retinal cell types are affected by loss of Fmr1 and the downstream effects on the brain and behavior. In addition, understanding how loss of Fmr1 affects visual development and the mechanisms of its effects on retinal development will provide important new insights into neurodevelopment in fragile X syndrome.
Fmr protein (FMRP) regulates mRNA translation in dendrites. This affects key anatomical and physiological factors, especially at inhibitory synapses (Hagerman et al., 2017). Here, we observed denser dendritic arbors and increased synaptic inhibition in an RGC type in the Fmr1−/y mouse. Denser dendrites and altered synaptic development have also been observed in a variety of other brain areas (Qin et al., 2011; He and Portera-Cailliau, 2013). This suggests that FMRP may play similar roles in dendritic development in the retina as in other brain areas. Our findings demonstrate that both dendritic arbor development and synapse formation and/or pruning occur atypically in RGCs in the absence of FMRP. However, it is still an open question at which stage of development FMRP exerts its effects on dendrites and synapses. sOn-α cell development in the mouse is characterized by the elongation of their dendrites during the first 2 postnatal weeks, with the branching pattern established at the beginning of this developmental phase (Lucas and Schmidt, 2019). This suggests that FMRP during the first postnatal week may be important for establishing dendritic density. We found that Fmr1 RNA was high in the retina at P9, followed by weaker expression at P35 and a return to high expression in adulthood (P60). How FMRP shapes RGC function at these different developmental stages is not yet clear.
Altered synaptic signaling is a common theme across brain areas in fragile X syndrome and ASD more broadly (Coghlan et al., 2012; Contractor et al., 2021). Changes in the E–I ratio have been proposed as a common mechanism of dysfunction in ASD (Monday et al., 2023). However, more recent research in the cortex suggests that changes in the E–I ratio are a compensatory mechanism to stabilize spiking (Antoine et al., 2019). Here, we found that synaptic changes in sOn-α cells are the opposite of what is observed in pyramidal cells in the cortex in Fmr1−/y mice: both synaptic excitation and inhibition increase (rather than decrease in the cortex), and the E–I ratio goes down (rather than going up in the cortex; Fig. 5). However, these changes are still consistent with a model by Antoine et al. (2019) proposing that stabilized spiking requires nonlinear scaling of excitation and inhibition. Our results satisfy the specific prediction in their model. If excitation and inhibition increase, the E–I ratio should decrease to maintain stable spiking (Fig. 6). Therefore, these results extend the validity of that model to cases where excitation and inhibition have increased. The results show that circuits of sOn-α cells may increase their inhibitory input to compensate for slightly increased excitatory input. However, in maintaining this balance in overall excitability, the cell may dampen its “sustained” response property, leading to an overall change in visual encoding. Understanding how synaptic and intrinsic properties of cells relate to E–I balance and visual tuning is important for understanding how the brain adapts to different homeostatic setpoints.
Fmr1 is broadly expressed in the retina. Fmr1 is not only expressed in RGCs but also in the excitatory interneurons, bipolar cells, inhibitory interneurons, and amacrine cells (Fig. 1, Extended Data Fig. 1-1). Within each of these cell classes, Fmr1 appears to be expressed at type-specific levels. This suggests that loss of Fmr1 could affect retinal cell types to a different degree. Here, we found changes in the strength of synaptic input onto sOn-α cells and changes in dendritic density. These changes could be the result of loss of Fmr1 in sOn-α cells or changes in their presynaptic partners. Pinpointing how Fmr1 influences sOn-α cell development and function will be an important next step to understanding how loss of Fmr1 affects the retina.
We identified differential expression of Fmr1 and other ASD genes across ganglion cell types in transcriptomic data. Whether Fmr1 expression is meaningfully different across cell types is still an open question. Our fluorescence in situ hybridization experiments demonstrated widespread Fmr1 RNA and did not exhibit noticeable cell-type–specific differences in the number of puncta. Therefore, whether and how differences in expression levels may affect cell-type–specific functions yet unknown. We found that a high Fmr1-expressing cell type, the sOn-α cell, had noticeable differences in dendritic morphology in the Fmr1 KO. Meanwhile, a low Fmr1-expressing cell type, the tOff-α cell, lacked changes in dendritic morphology in the knock-out in our small sample. These results suggest that loss of Fmr1 in fragile X syndrome may not affect all ganglion cell types equally. Instead, loss of Fmr1 may leave some cell types functioning in the typical range while others operate atypically. Furthermore, even in high-expressing cell types, loss of Fmr1 resulted in relatively subtle changes in morphology and function. This matches the fact that people and mice lacking functionalFmr1 are by no means considered visually impaired, but instead have subtle changes in vision. The widespread expression of ASD genes in the retina points to an unexplored frontier for understanding neurodiversity and its effects on visual sensation and perception. Some people with ASD report strong visual symptoms while others have little-to-no reported symptoms. One explanation for these differences may lie in the periphery with distinct effects of ASD-linked genes on the development and function of the retina.
How loss of Fmr1 will relate to visual symptoms in fragile X syndrome is not yet obvious. Each RGC type may project to multiple brain areas (Kerschensteiner, 2022). This complicates inquiry into how changes in their signaling affect behavior. For example, sOn-α cells project to the superior colliculus, the lateral geniculate nucleus, and other targets. In mice, these cells provide information for behaviors including binocular-vision–guided hunting (Kim et al., 2020) and contrast sensitivity of the optokinetic reflex (Schmidt et al., 2014). Different brain areas may rely more or less heavily on the specific “sustained”-ness or light sensitivity of a given ganglion cell type. Therefore, these behaviors may not be equally affected by changes in the function of sOn-α cells. Beyond these effects, altered activity of sOn-α cells during visual development could influence the development of downstream circuits in the brain (Thompson et al., 2017). These activity-dependent mechanisms for establishing connections between ganglion cells and their downstream partners could ultimately compensate for differences in light sensitivity at the level of the retina.
The homologous cell type to the sOn-α cell in humans is thought to be the midget cells of the parvocellular visual pathway (Hahn et al., 2023). Midget cells are the dominant ganglion cell type of the primate fovea, important for high-resolution color vision. Whether loss of Fmr1 in the midget cells would result in similar changes in morphology or physiology as in sOn-α cells is not known. The transcriptomic dataset for primate retina shows lower relative expression of Fmr1 in midget cells compared with some other primate ganglion cell types (Extended Data Fig. 1-1). Given FMRP's role in regulating synapses and dendritic development, species-specific differences in neuronal function could be shaped by FMRP itself. One commonly reported visual symptom in ASD is visual hypersensitivity (Simmons et al., 2009), which does not overtly match with dampened light sensitivity in the retina. However, the retina does not develop in isolation. Given that the brain may compensate for changes in visual sensitivity of the retina, visual hypersensitivity could be a side-effect of the brain boosting the weaker signals arriving from the periphery. Indeed, increased “cortical noise” has been proposed as an explanation for symptoms of ASD (Sohal and Rubenstein, 2019). Perhaps this noise originates from weaker sensory input.
Footnotes
We thank Anis Contractor for providing Fmr1 knock-out mice, Greg Schwartz and Zach Jessen for providing the patchseq dataset, the staff of the Center for Comparative Medicine at Northwestern University for providing excellent animal care, and finally, the members of the Schmidt lab who provided training, support, and comments on the manuscript. This work was supported by the Knights Templar Eye Foundation Career Starter Award (to A.L.V.); Northwestern Summer Undergraduate Research Grant (to A.W.) and Summer Internship Grant Program (to P.G.); NIH 5T32HL007909; NIH R01EY030565 (to T.M.S.), NIH P30 EY001792 (A.L.V.), unrestricted departmental funding from Research to Prevent Blindness (A.L.V., E.L.), and a Research to Prevent Blindness Career Development Award (to A.L.V.).
A.L.V. is the lead contact.
The authors declare no competing financial interests.
- Correspondence should be addressed to Anna L. Vlasits at avlasits{at}uic.edu.
