Inhibition of ceramide accumulation in AdipoR1–/– mice increases photoreceptor survival and improves vision

Adiponectin receptor 1 (ADIPOR1) is a lipid and glucose metabolism regulator that possesses intrinsic ceramidase activity. Mutations of the ADIPOR1 gene have been associated with nonsyndromic and syndromic retinitis pigmentosa. Here, we show that the absence of AdipoR1 in mice leads to progressive photoreceptor degeneration, significant reduction of electroretinogram amplitudes, decreased retinoid content in the retina, and reduced cone opsin expression. Single-cell RNA-Seq results indicate that ADIPOR1 encoded the most abundantly expressed ceramidase in mice and one of the 2 most highly expressed ceramidases in the human retina, next to acid ceramidase ASAH1. We discovered an accumulation of ceramides in the AdipoR1–/– retina, likely due to insufficient ceramidase activity for healthy retina function, resulting in photoreceptor death. Combined treatment with desipramine/L-cycloserine (DC) lowered ceramide levels and exerted a protective effect on photoreceptors in AdipoR1–/– mice. Moreover, we observed improvement in cone-mediated retinal function in the DC-treated animals. Lastly, we found that prolonged DC treatment corrected the electrical responses of the primary visual cortex to visual stimuli, approaching near-normal levels for some parameters. These results highlight the importance of ADIPOR1 ceramidase in the retina and show that pharmacological inhibition of ceramide generation can provide a therapeutic strategy for ADIPOR1-related retinopathy.

individual samples with lysis buffer. Samples were subjected to SDS-PAGE electrophoresis, transferred to nitrocellulose membranes, and subsequently incubated for 1 hr at RT in trisbuffered saline containing 0.1% Tween20 (TBST) and 5% dry non-fat milk powder. One membrane was cut into two fragments (between the 37 kDa and 50 kDa protein standards), and the other identically prepared membrane was left intact. Membranes were incubated overnight at 4°C in TBST containing 2.5% dry non-fat milk powder with (i) ADIPOR1 Ab (1:200, #18993, IBL), or (ii) α-tubulin Ab (1:1000, #2144, Cell Signaling), used as a loading control. Membranes were washed 4x 5 min with TBST and incubated with HRP-labeled goat anti-rabbit Ab (1:3000, #7074, Cell Signaling) for 1 hr at RT. After washing (4 x 5 min with TBST), membranes were submerged in a chemiluminescent substrate (#34577, Thermo Fisher) for 5 min and imaged with Odyssey Fc System (LI-COR), which was also used for band quantification. Data were normalized to α-tubulin band intensities. The experiment was repeated two times.
Retinal and RPE-eyecup flatmounts. Mice were euthanized by CO2 inhalation followed by cervical dislocation. The superior position of the eye was marked on the cornea with a blue Sharpie; globes were enucleated and fixed in PBS-buffered 4% paraformaldehyde (PFA) for 10 min. After fixing, eyes were washed three times in PBS for 3 min and placed on a microscope slide. Muscles, fat, and optic nerve were removed from the globe, followed by a puncture in the center of the cornea with a 25-gauge needle. Cohan-Vannas spring scissors (Fine Science Tools) were used to make four symmetric radial incisions starting from the center of the cornea and ending directly before the optic nerve head. The lens and vitreous were removed, and the eyecup was oriented with the sclera side facing up and flattened. The corneal flaps were removed, and the superior position was marked by making a small triangular cut. The RPE-eyecup was gently peeled off the retina and flattened on a new slide, and the retina was flattened with the photoreceptors facing up. Next, both retina and RPE-eyecup were further fixed in PBS-buffered 4% PFA for 30 min. Finally, all flatmounts were washed three times in PBS for 5 min and used for immunofluorescence staining.
Retinal cross-sections. Mice were euthanized by CO2 inhalation followed by cervical dislocation. Eyes were enucleated and rinsed briefly with PBS. Using a 23G needle, two small punctures were made in each eyeball at the junction of the cornea and sclera, and the eyes were fixed in PBS-buffered 4% PFA for 15 min. The cornea was cut off, followed by the lens and vitreous removal, and the eyecup was further fixed for 15 min. Next, the eyecup was gently washed three times in PBS for 5 min and transferred to a sucrose solution in PBS (10% and 20% sucrose for 30 min each at room temp., then 30% sucrose overnight at 4°C). On the next day, the eyecup was embedded in O.C.T. (Tissue-Tek) and snap-frozen in liquid nitrogen vapors. The eyecup was cut into 10 µm thick serial sections using a cryostat and stored at -80°C.
Iba1/F-actin staining. RPE-eyecup flatmounts were incubated in a blocking buffer consisting of 3% BSA with 0.3% Triton X-100 in PBS for 2 hr at 22°C. Next, rabbit Iba1 Ab was added at 1:1000 dilution (#019-19741, Fujifilm Wako) in blocking buffer overnight at 4°C, in a humidified chamber. Then the primary Ab was removed with five rinses of PBS with 0.3% Triton X-100. Flatmounts were incubated with a secondary goat anti-rabbit IgG Alexa Fluor 594 Ab (#A11037, Invitrogen) at 1:500 dilution in blocking buffer for 2 hr at 22°C, and then rinsed four times with PBS containing 0.3% Triton X-100. The RPE-eyecup flatmounts were incubated in Phalloidin Alexa Fluor 488 (#A12379, ThermoFisher) at 1:1000 dilution in blocking buffer for 30 min at 22°C, and then rinsed four times with PBS with 0.3% Triton X-100 and once with PBS. Finally, specimens were mounted with Vectashield medium with DAPI (#H-2000; Vector Lab.) and protected with coverslips. Images were captured with a Z-stack mode using a Keyence BZ-X800 microscope (Keyence Corp.). The experiment was repeated two times. Lipid identification was performed with LipidSearch (ThermoFisher). Mass accuracy, chromatography, and peak integration of all identified lipids were verified with Skyline (5).
Skyline peak areas were used in data reporting, and data were normalized using internal standards and the protein content of the sample. Sequence data were demultiplexed using bcl2fastq software (Illumina), and their quality was assessed using FastQC (Babraham Bioinformatics). Reads were aligned to the reference mouse genome assembly (mm10) and quantified using the Rsubread software package with standard parameters (8), yielding on average 29 million counts per sample. Differential expression analysis was performed with EdgeR (v3.20.9)(9); and functional enrichment analysis was conducted on the identified differentially expressed genes (log2FC > 1 & FDR < 0.05), using the findGO.pl function from the HOMER suite (Hypergeometric Optimization of Motif Enrichment, v4.9.1)(6). To correct for multiple hypotheses testing, the Benjamini and Hochberg FDR correction was used, and we retained terms at FDR < 0.05.  (11). We performed downstream analysis using the R software package Seurat v3.2.2 (12). We removed low-quality cells with less than 500 genes detected and all genes expressed in less than 10 cells, leaving 26,764 cells and 19,324 genes for further analysis.

Single-cell RNA-Seq of
Principal Component Analysis (PCA) was performed on a submatrix of the top 1,000 most variable genes computed, using the function FindVariableGenes from the Seurat package. Then, we removed the batch effect between samples using the package Harmony (13). In particular, we removed variables associated with two separate sequencing runs (B1 and B2) and WT and AdipoR1 -/conditions to remove the non-cell-type-specific factors that impact cell clustering. We evaluated the number of top principal components by the elbow method keeping 17 PCs for clustering and data visualization. The cells were clustered using a shared nearest neighbor (SNN) modularity optimization-based clustering algorithm (FindClusters in the Seurat package). To visualize cells in two dimensions, we used Uniform Manifold Approximation and Projection (UMAP) (14). The cluster-specific genes were computed using FindAllMarkers from the Seurat package, using the MAST test with the number of UMIs detected as a latent variable (15). The same test was used to compute differentially expressed genes between WT and AdipoR1 -/mice within cell clusters. Cell clusters were annotated by assessing known cell-type-specific markers.
The gene set enrichment analysis was conducted as described in the bulk RNA-Seq data analysis section.

RNA-Seq data availability.
Single-cell and bulk RNA-Seq data are deposited in the NCBI Gene Expression Omnibus under data repository accession number GSE184902. Fig. S4 was performed as previously described (16).

Single unit and local field potential recordings and visual stimulation. Mice were
initially anesthetized with 2% isoflurane in a mixture of N2O/O2 (70%/30%), then placed into a stereotaxic apparatus. A small, custom-made plastic chamber was glued (3M Vetbond) to the exposed skull. After one day of recovery, re-anesthetized animals were placed in a custom-made hammock, maintained under isoflurane anesthesia (1-2% in a mixture of N2O/O2), and multiple tungsten electrodes were inserted into a small craniotomy above the visual cortex. Once the electrodes were inserted, the chamber was filled with sterile agar and sealed with sterile bone wax. During recording sessions, animals were sedated with chlorprothixene hydrochloride (1 mg/kg; IM (17); and kept under light isoflurane anesthesia (0.2 -0.4% in 30% O2). EEG and EKG recordings were monitored throughout the experiments, and body temperature was maintained with a warming pad (Harvard Apparatus). Data were acquired using a 32-channel Scout recording system (Ripple, USA). The local field potential (LFP) from multiple locations was band-pass filtered from 0.1 Hz to 250 Hz and stored together with spiking data on a computer with a 1 kHz sampling rate. The LFP signal was cut according to stimulus time stamps and averaged across trials for each recording location to calculate visually evoked potentials (VEP) (18,19). The spike signal was band-pass filtered from 500 Hz to 7 kHz and stored on a computer hard drive at 30 kHz sampling frequency. Spikes were sorted online in Trellis (Ripple, USA) during visual stimulation. Visual stimuli were generated in Matlab (Mathworks, USA) using the Psychophysics Toolbox (20,21) and displayed on a gamma-corrected LCD monitor (55", 60 Hz; 1920 x 1080 pixels; 52 cd/m 2 mean luminance). Stimulus onset times were corrected for LCD monitor delay, using a photodiode and microcontroller (in-house design) (22).
The vision quality was assessed using a protocol published previously by our laboratories (19,(22)(23)(24). For recordings of visually evoked responses, cells were first tested with 100 repetitions of a 500 msec bright flash stimulus (105 cd/m 2 ). Receptive fields for visually responsive cells were then located using square-wave drifting gratings, after which optimal orientation/direction, and spatial and temporal frequencies were determined using sine-wave gratings. Spatial frequencies were from 0.001 to 0.5 cycles/deg. Temporal frequencies tested were 0.1 to 10 cycles/sec. With these optimal parameters, size tuning was assessed using apertures of 1 to 110º at 100% contrast. With the optimal size, temporal, and spatial frequencies, and at high contrast, the orientation tuning of the cell was tested again using 8 orientations x 2 directions each, stepped by 22.5º increments. This was followed by testing the contrast response.
The LFP signal was normalized using z-score standardization (18,25). The response amplitude of LFP was calculated as a difference between the peak of the positive and negative components in the VEP wave. The response latency was defined as the time point where maximum response occurred. The maximum response was defined as the maximum of either the negative or positive peak.
Tuning curves were calculated based on the average spike rate. Optimal visual parameters were chosen as the maximum response value.
The orientation tuning bandwidth was measured in degrees as the half-width at halfheight (HWHH; 1.18 x σ) based on fits to Gaussian distributions (22,26,27), using: where Os is the stimulus orientation, ROs is the response to different orientations, Op is the preferred orientation, Rp and Rn are the responses at the preferred and non-preferred directions, σ is the tuning width, and 'baseline' is the offset of the Gaussian distribution. Gaussian fits were estimated without subtracting spontaneous activity (26).
Size tuning curves were fitted by a difference of Gaussian (DoG) function: where Rs is the response evoked by different aperture sizes. The free parameters, Ke and re, describe the strength and the size of the excitatory space, respectively; Ki and ri represent the strength and the size of the inhibitory space, respectively; and R0 is the spontaneous activity of the cell.
The optimal spatial and temporal frequency was extracted from the data fitted to Gaussian distributions using the following equation (22,28): Where RSF/TF is the estimated response, Rpref indicates the response at the preferred spatial or temporal frequency, SF/TF indicates spatial or temporal frequency, σ is the standard deviation of the Gaussian, and baseline is the Gaussian offset.
The contrast tuning was fitted by using the Naka-Rushton equation (29): where g is the gain (response), C50 is the contrast at mid response, and n is the exponent. For the contrast tuning fit, the background activity was subtracted from the response curve, and values below background standard deviation were changed to 0 (22,30). Average differences between animal groups were considered statistically significant at P ≤ 0.05 for two-tailed Mann-Whitney U-tests. Mean values given in the results include SEM. All offline data analysis and statistics were performed in Matlab.
Statistics. Statistical analyses, other than those described above for visual cortex responses, were performed using GraphPad Prism software, and data are shown as mean ± SEM, unless otherwise stated. Results are considered significant if P was less than 0.05. The significance of differences between 2 groups was determined using a 2-tailed Student's t test. For more than 2 groups, 1-or 2-way ANOVA was used with post hoc correction for multiple comparisons. When comparing multiple measurements within subjects, correction for repeated measures was performed. Figure S1. AdipoR1 knockout alters retinal ERGs and retinoid cycle in mice. Figure S2.   . Scotopic a-wave (A) and b-wave (B) amplitudes are dramatically attenuated in AdipoR1 -/mice. Photopic a-wave (C) and b-wave (D) amplitudes of AdipoR1 +/+ and AdipoR1 -/mice also show such attenuation, but the extent is animal-age sensitive (n = 4 for both genotypes and all ages). Repeated measures two-way ANOVA followed by Sidak's post hoc test was used; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.  The statistical significance was determined with two-way ANOVA followed by Tukey's multiple comparison test; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.     . Ceramidase activity of ADIPOR1. The enzymatic assay was performed to determine the relative amounts of sphingosine produced by purified mouse ADIPOR1 wild type (WT) or H191A,H337A-mutant (M) forms of ADIPOR1, using ceramide C24:1 as substrate. The samples were additionally treated with high-molecular-weight fractions of recombinant mouse adiponectin, recombinant human C1QTNF5, or native mouse C1Q to examine their effect on the ceramidase activity of ADIPOR1. Detected sphingosine (d18:1) values were normalized to internal standard (sphingosine-d7). The data (mean ± SEM) are representative of three independent experiments performed in three or four replicates. Statistical significance was determined by one-way ANOVA followed by Sidak's post hoc test; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.