scRNA-Seq reveals changes in retinal composition and transcriptional profiles. To investigate the impact of the ND6 mutation on retinal cell composition and transcriptional profiles, we conducted droplet-based scRNA-Seq on retinal cells from 7-month-old ND6P25L and WT mice. These analyses produced high-quality profiles for 83,672 cells. Following quality filtering, we integrated these profiles using canonical correlation analysis, performed differential expression analysis, and categorized the cells into 14 distinct clusters based on known gene markers (Figure 1, A and B, and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.188962DS1) (35–37). These markers included Arr3+ (cones), Vsx2+ (bipolar cells), Sncg+ (RGCs), Ptn+ (retinal progenitor cells), Snhg11+ (horizontal cells), Gria3+ (amacrine cells), Slc6a9+ (AC_Gly), Gad1+ (AC_GABA), C1qa+ (macrophages), Rpe65+ (RPE cells), Glul+ (Müller cells), and Rho+ (rods). Transcription factor activities in cell clusters were verified using SCENIC (Figure 1C) (38).

Mouse retinal compositional and transcriptional profiles. (A) Uniform manifold approximation and projection (UMAP) map of 83,672 merged retinal cells from WT and mutant (MT) mice color coded for the indicated cell type. BC, bipolar cell; RPC, retina progenitor cell; HC, horizontal cell; AC_Gly, glycinergic amacrine cell; AC_GABA, GABAergic amacrine cell. (B) UMAP plots showing the expression of marker genes for each cell type. (C) The expression of transcription factors for each cell type. (D) UMAP plots of retina cells from WT (left) and MT (right) mice. (E) Box plots showing the fractions of cells for each cell type in each mouse. Box plots show the interquartile range, median (line), and minimum and maximum (whiskers). (F and H) Pseudotime trajectories of Rod1, Rod2 (F), Muller1, and Muller2 (H) based on Slingshot and gene expression dynamics along the trajectory. Genes clustered into 5 gene sets, each characterized by specific expression profiles, as depicted by a selection of marker genes characteristic for each cluster. (G and I) Plots showing cell density of Rod1, Rod2 (G), Muller1, and Muller2 (I) along the trajectory comparing WT and MT.

To examine if the ND6 mutation affected retinal subset composition, we analyzed the relative proportions of each cell cluster among mutant and WT mouse retina. Figure 1, D and E, summarize marked variations in the cell type composition over 14 cell clusters between mutant and WT mouse retina. The mutant retina exhibited various reductions in neuronal cell types, including cones, bipolar cells, horizontal cells, and Muller2 cells, with especially pronounced effects on Rod2 cells, but significantly increased proportions of Muller1 cells, as compared with those in WT retina.

The transcriptomic changes in the rod and Müller cells were further evaluated using pseudotime trajectory analysis with the Slingshot package (Figure 1, F–I). The mutant retinas revealed a higher density of Rod1 and Muller1 cells but a lower density of Rod2 and Muller2 cells, as compared with WT retinas (Figure 1, G and I). Rod2 cells exhibited higher expression of photoreceptor differentiation transcription factors, such as Otx2 and Crx, but lower expression of OXPHOS markers, including ND4 and CYTB, than in Rod1 cells, indicating that Rod2 cells may be more differentiated and have specific mitochondrial features (Figure 1F and Supplemental Figure 1, B and C) (39–41). Moreover, Muller1 cells expressed higher levels of progenitor transcription factors Ascl1 and Olig1, but lower levels of Otx2 and Crx, than in Muller2 cells, suggesting that Muller1 cells may function as the progenitors for major retinal cell types (Figure 1H and Supplemental Figure 1, D and E) (42, 43).

Retinal cell–specific defects in transcriptional regulation of OXPHOS pathways. We performed differential expression analysis and scored REACTOME pathway activity for each cell subset in the mutant and WT retinas. The pathway analysis revealed the dysregulation of OXPHOS pathways among the majority of cell clusters in mutant retina. In particular, upregulated pathways included mitochondrial transfer RNA (tRNA) processing, tricarboxylic acid (TCA) cycle, and transcriptional regulation of small RNAs, while downregulated pathways were primarily involved in OXPHOS biogenesis, such as electron transport chains, ATP synthase, and TCA cycle (Figure 2, A and B). Furthermore, several phototransduction cascade pathways were upregulated in the C3_Müller1, C6_BC, and C10_AC_Gly (Figure 2A), suggesting that these cell types were particularly vulnerable to mitochondrial dysfunctions.

Cell-specific mitochondrial dysfunctions in the retina. (A) Dot plots showing the significance (−log10P value) and overlap (percentage of differentially expressed genes; DEGs) of upregulated REACTOME pathways in the majority of cell clusters in MT retina by gene set enrichment analysis on DEGs, as compared with WT. (B) Dot plots showing the significance (–log10P value) and overlap of downregulated REACTOME pathways in the majority of cell clusters in MT retina, as compared with WT. (C) Volcano plots showing DEGs comparing MT and WT mice in 8 retina subsets (Supplemental Figure 2A shows the remaining 6 retina subsets). Dots on the volcano plot: gray, no significant change; red, P < 0.05 and log2FC ≥ 0.5 (FC, fold change); blue, P < 0.05 and log2FC ≤ –0.5. The significantly differentially expressed OXPHOS genes were labeled. The counts of significantly up- and downregulated OXPHOS genes for each cluster were labeled. (D) Dot plots showing the average fold-change of expression of mtDNA-encoded and nDNA-encoded OXPHOS genes in each OXPHOS complex across retina subsets, comparing MT and WT mice. (E and F) Heatmap displaying expression of genes involved in the antioxidant pathway across subsets (E) and dot plot showing the quantification of average expression levels of the antioxidant pathway across subsets (F).

To further assess the effects of ND6 mutation on the OXPHOS pathways, we focused on 152 (13 mtDNA and 139 nuclear DNA [nDNA], MitoCarta3.0 database) genes involved in OXPHOS biogenesis in the differential expression analysis of retinal subsets (44). As shown in Figure 2C and Supplemental Figure 2A, 13 mtDNA-encoded and 38 nDNA-encoded genes in the mutant retina exhibited differential expression across 14 subsets as compared with WT retina (absolute logFC > 0.5 and P value < 0.05). Notably, RGCs exhibited many more downregulated genes than upregulated genes. More downregulated than upregulated genes were also observed in Rod1, RPE, and macrophages. Conversely, more upregulated than downregulated genes occurred in the cone, HC, GABAergic AC, Rod2, and Muller2, respectively.

We then examined the expression levels of mtDNA and nDNA genes encoding components for each of the OXPHOS complexes (Supplemental Table 1). Similar expression patterns of mtDNA-encoded subunits, including upregulation of ND2, CYTB, and ATP8 but downregulation of ND3, ND4L, CO1, CO2, CO3, and ATP6, occurred among 14 cell types in the mutant retina (Figure 2D and Supplemental Figure 3). These mRNA expressions were further confirmed by quantitative PCR analysis (Supplemental Figure 2B). By contrast, the expression levels of nDNA-encoded OXPHOS subunits and assembly factors varied greatly across the 14 subsets in the mutant retina. Further expression analysis of genes associated with complex I assembly modules showed only downregulation of the ND2 module, including ND6, but upregulation of other 5 modules (Supplemental Figure 3).

ROS production acts as a major mitochondrial function involved in cellular adaptation and stress resistance (45). Mitochondrial dysfunctions increase the ROS production and therefore regulate the expression of antioxidant pathways (23, 24, 46). We then examined the effects of ND6 mutation on antioxidant pathways in mouse retina. As shown in Figure 2, E and F, 11 clusters, including cones and Rod2, exhibited variable upregulation of antioxidant pathways, such as SOD2, with particularly pronounced upregulation in the RGCs, whereas 3 clusters, including RPE, showed downregulation of these pathways. These data indicated that the ND6 mutation caused cell-specific mitochondrial defects in retinal subsets.

Dysregulation of visual signaling pathways and VA metabolism. We then assessed the cell-specific responses to mitochondrial dysfunctions caused by the ND6 mutation in the retinal subsets (Figure 3A). Phototransduction cascade and its activation pathways were specifically dysregulated in Müller1, BC, AC_Gly, and macrophages, indicating that visual function is cell-specifically susceptible to mitochondrial defects. To further investigate the consequences of the ND6 mutation on retinal function, we focused on the assessment of expression of vision-related signaling pathways in mutant and WT mice (Supplemental Table 2). As shown in Figure 3B, ND6 mutant retina exhibited marked dysregulation of the phototransduction cascade, canonical retinoid (VA) cycle, and neurotransmission pathways, including neurotransmitter uptake, glutamate release, and GABA release and reuptake.

Abnormal visual signaling pathways and VA metabolism. (A) Dot plots showing the significance (–log10P value) and overlap of differentially expressed cell-specific REACTOME pathways in MT retina, as compared with WT. (B) Dot plot showing the significance (−log10P value) and overlap of vision-related signaling pathways (rows) in retina clusters (columns) between MT and WT mice. (C) Compass-based exploration of metabolic change in MT retina, compared with WT retina. Reactions (dots) are partitioned by Recon2 pathways and colored by the sign of their Cohen’s d statistic. Red asterisks indicate the significantly dysregulated metabolic pathways. BH, Benjamini-Hochberg. (D) Schematic illustration of VA metabolism (visual cycle) in retina. (E) Compass score differential activity test of VA metabolism pathways between MT and WT retina. Each dot represents a single biochemical reaction. (F) Relative levels of retinol in eyeballs of MT and WT mice using ELISA. Data are shown as mean ± SEM. *P < 0.05; n = 4 mice per group; 2-tailed unpaired Student’s t test. Normality was assessed using the Shapiro-Wilk test (P = 0.8986), and equal variance was confirmed using the F test (P = 0.8697). (G) Immunoblot analysis of proteins involved in VA metabolism. Total cellular proteins in WT and MT retina were electrophoresed with PAGE and hybridized with RPE65 and STRA6 antibodies and GAPDH as a loading control, respectively. Values on left represent kilodaltons.

To systematically evaluate metabolic changes in ND6 mutants, we employed the Compass algorithm for comprehensive single-cell metabolism characterization (47). Using flux balance analysis, we showed significant differences in metabolic profiles, including up- and downregulated reactions across pathway boundaries between mutant and WT retinas (Figure 3C). In particular, the mutant retina exhibited the downregulation of mitochondrial and cytosolic bioenergetic pathways, including glycolysis, CoA synthesis, fatty acid oxidation, NAD metabolism, and TCA cycle, but upregulation of ROS detoxification enzymes (Figure 3C and Supplemental Figure 4).

Strikingly, the VA metabolism was significantly upregulated in the mutant retina (Figure 3C). We then further assessed the impact of ND6 mutation–induced abnormal metabolism of VA on retina function. In fact, VA is delivered from the blood to RPE cells via retinoid binding protein receptor STRA6 (28). It then undergoes a visual cycle in RPE, photoreceptor cells, and Müller cells to regenerate the visual chromophore of rhodopsin for light reception (Figure 3D) (28–30). Compass analysis exhibited significant upregulation of retinol isomerase and reductase in the mutant retina (Figure 3E). Aberrant VA metabolism was further verified via ELISA and Western blot assays. As shown in Figure 3F, the levels of retinol in the mutant retina were decreased by 31%, as compared with WT retina. As illustrated in Figure 3G, mutant retina displayed marked decreases in the levels of STRA6 (retinal vitamin A transport protein) but marked increases in levels in the RPE65 (retinol isomerase), compared with those in the WT retina (Figure 3G) (48–50). These may provide new insights into pathophysiology of LHON-linked mtDNA mutations via the dysregulation of deficient VA metabolism.

VA supplementation restores retinal deficiency. VA supplementation may be utilized as one therapeutic approach to restore retinal deficiency arising from LHON-linked mtDNA mutation. We investigated if that VA supplementation improved deficient VA metabolism, rescued mitochondrial morphology and function, and subsequently restored retinal deficiencies due to LHON-linked ND6 mutation in the mutant and WT mice. Starting from the age of 2 months, WT and ND6P25L mice were maintained on either a standard VA control chow (4 IU retinol/g) or VA excess chow (120 IU retinol/g). We then evaluated retinal phenotypes at the age of 2 months (baseline) and subsequently at 3 and 4 months. At the age of 2 months, fundus examinations revealed ocular lesions in the mutant retina, a typical phenotype of RPE abnormalities (Figure 4A and Supplemental Table 3) (5, 51). Corresponding optical coherence tomography (OCT) data verified structural abnormalities in the RPE and photoreceptor cells. Notably, the retinal deficiencies in the mutant mice worsened, such as enlarged areas of lesion and increased retina thickness, at the ages of 3 and 4 months (Figure 4, A–C). After supplementing VA for 1 or 2 months, retinal deficiency of ND6P25L mice was significantly restored, including decreased lesion area and retina thickness, as compared with those without VA. In contrast, there were no significant differences of retinal phenotype in the WT mice with and without VA supplementation.

The recovery of retinal deficiency with VA supplementation. (A) Image-guided OCT analysis was performed at 2 months (baseline), 3 months, and 4 months of age. The left panel for each age shows the fundus image, and the black arrowed line indicates the location of the OCT scans. The white arrow indicates the fundus lesion. The right panel for each age shows the corresponding OCT images, and the white arrow indicates abnormalities in the photoreceptor and RPE. (B) Quantification of changes (logFC) from baseline of lesion areas comparing MT (n = 7) and MT+VA mice (n = 6). (C) Quantification of retinal thickness changes from baseline in WT (n = 8), MT (n = 6), MT+VA (n = 5) mice. (D) Immunofluorescence staining of cryosection showing retinal section stained with Brn3a (red) with DAPI (blue) for RGC, Vimentin (green) with DAPI for MG, rhodopsin (red) with DAPI for rod. Scale bar: 50 μm (Brn3a), 100 μm (Vimentin and rhodopsin). GCL, ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer; MG, Müller glia. (E–G) Quantification of ratios of Brn3a-positive RGC (E), relative numbers of Müller dendrites in IPL (F), and relative number of photoreceptors (counts of DAPI-positive nuclei in ONL) (G) in WT, MT, and MT+VA mouse retina. n = 3–5 for Brn3a staining; n = 5–6 for Vimentin staining; n = 8–15 for photoreceptor counts. Data in B and C are shown as mean ± SEM. *P < 0.05, **P < 0.01 by 2-tailed unpaired Welch’s t test comparing MT and MT+VA mice. Data in E–G are shown as mean ± SEM. ***P < 0.001, ****P < 0.0001 by 1-way ANOVA followed by Tukey post hoc test.

We further investigated the cell-specific effects of VA supplementation on RGCs, Müller cells, and rods. Immunofluorescence staining analysis revealed a 53.4% decrease in Brn3a-positive (transcription factor essential for RGC survival) RGCs, a 95.4% reduction in Vimentin-positive Müller cell neurites in the inner plexiform layer, and a 19.4% decrease in the photoreceptor cells in the mutant mice at the age of 4 months, as compared with WT mice (Figure 4, D–G). Two months of VA supplementation markedly restored the degeneration of retinal cells in mutant mice to levels comparable to those in WT mice (Figure 4, D–G). In particular, the numbers of Brn3a-positive RGCs, Vimentin-positive Müller cell neurites, and the photoreceptor cells in the mutant mice after supplementation of VA were elevated to 233%, 1,204%, and 122% of those in untreated mutant mice. Furthermore, untreated mutant mice exhibited concentrated rhodopsin at the outer segment layer adjacent to the RPE, with a substantial reduction in rhodopsin within the outer nuclear layer (ONL) and inner segment layer (ISL). In contrast, rhodopsin distribution in VA-supplemented ND6P25L mice was similar to that in WT mice (Figure 4D). These results indicated that dietary VA supplementation remarkably ameliorated retinal abnormalities and cellular deficiencies in ND6 mutant mice.

Restoration of VA metabolism and visual function. We assessed if the VA supplementation restored the deficient VA metabolism due to ND6 mutation. The levels of VA metabolic pathway–related proteins, including STRA6, RPE65, retinol dehydrogenases (RDH5 and RDH12), lipoprotein ligase (LPL), and lipoprotein receptor (LRP1), in the presence and absence of VA supplementation were measured by Western blot analysis (28–30). As shown in Figure 5A and Supplemental Figure 5A, the VA supplementation levels made the levels of STRA6, RPE65, and LPL in the mutant retina comparable to those in WT mice but did not change the levels of RDH5, RDH12, and LRP1 in the mutant retina. Furthermore, the VA treatment elevated the levels of retinoid acid (RA) receptors RARα and RXRα in the mutant retina, indicating the recovery of VA signaling pathways (Figure 5A and Supplemental Figure 5A) (52). As shown in Figure 5B, ELISA revealed that VA supplementation increased VA content in the eyes of ND6P25L mice to 395.3% of the levels without VA supplementation.

VA metabolism and visual function. (A) Western blot analysis of proteins involved in VA metabolism. Total cellular proteins in WT, MT, and MT+VA retina were electrophoresed with PAGE and hybridized with STRA6, RDH5, LRP1, LPL, RPE65, RDH12, RARα, and RXRα antibodies and GAPDH as a loading control, respectively. (B) Relative levels of retinol in eyeballs of WT, MT, and MT+VA mice using ELISA. n = 4 mice per group; **P < 0.01, ****P < 0.0001 by 2-tailed unpaired Student’s t test of the differences between MT and WT, or MT and MT+VA mice. Normality was assessed using the Shapiro-Wilk test (P > 0.05), and equal variance was confirmed using the F test (P > 0.05). (C) Western blot analysis of proteins involved in phototransduction and neurotransmission. Total cellular proteins in WT, MT, and MT+VA retina were electrophoresed with PAGE and hybridized with RHO, PRKCQ, GluR1, PDE6B, VAMP2, GABBR1, NMDAR1, and GNB3 antibodies and GAPDH as a loading control, respectively. (D and E) Quantification of RHO, PDE6B, and PRKCQ for phototransduction (D) and VAMP2, GluR1, GABBR1, NMDAR1, and GNB3 for neurotransmission (E) in WT, MT, and MT+VA retina. The calculations were based on 3 independent determinations in each mouse. (F) Analysis of ffERG for WT (n = 5), MT (n = 3), or MT+VA (n = 3) mice. By dark adaptation for a night, mice were analyzed for scotopic response and then photopic response. Data in D–F are shown as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 by 1-way ANOVA followed by Tukey’s post hoc test.

We then investigated whether VA supplementation ameliorated the phototransduction and neurotransmitter transmission pathways in the mutant mice. The administration of VA resulted in various changes in the levels of components (RHO, PRKCQ, and PDE6B) of phototransduction pathways and components (GluR1, GABBR1, NMDAR1, GNB3, and VAMP2) of neurotransmitter transmission pathways (Figure 5, C–E, and Supplemental Figure 5B) (2, 53–55). In particular, VA administration resulted in decreased RHO levels and increased PRKCQ levels but no change in PDE6B level in the mutant retina (Figure 5, C and D). Moreover, markedly increased levels of GluR1, GABBR1, NMDAR1, and GNB3 occurred in the mutant retina, but VA administration reduced the levels of these proteins to be comparable to those in WT retina (Figure 5, C and E). However, VA treatment did not significantly change the levels of VAMP2, associated with the glutamate neurotransmitter release pathway in the mutant retina, as compared with those in WT retina (Figure 5, C and E).

We then evaluated if VA supplementation restored visual functions, focusing on photoreceptor deficits, by full-field electroretinography (ffERG). As shown in Figure 5F, the amplitudes of b-waves for scotopic (rod) responses and photopic (cone) responses in the mutant retina were 31.5% and 58.6% of those in WT mice. The supplementation of VA increased these amplitudes to 213.9% and 155.6% of unsupplemented levels for scotopic responses and photopic responses, respectively. These data demonstrated that the VA supplementation restored retinal failure arising from ND6 mutation.

Rescuing abnormal mitochondrial morphology and function. We then assessed if VA supplementation was able to rescue ND6 mutation–induced abnormal morphology and function of mitochondria in the mouse retina. As shown in Figure 6A, supplementation of VA significantly corrected abnormal mitochondrial morphologies, including swelling, cristae malformations, enlarged size, and reduced number in the RGCs and rods of mutant retina, as compared with WT retina.

Mitochondrial morphology and function. (A) Representative transmission electron micrographs of mitochondria from GCL and ISL of retina in WT, MT, and MT+VA mice. Scale bars: 1 μm in GCL, 2 μm in ISL. The arrow indicates mitochondria. The right panel shows the quantification of mitochondrial size (n = 86–319 mitochondria) and relative mitochondrial number (n = 4–6 mice) in GCL and ISL of mouse retina. (B and C) Assessment of mitochondrial function by enzyme histochemistry staining for COX and SDH in the frozen sections of retinas (B) and brains (C) in WT, MT, and MT+VA mice. scale bar: 50 μm in retina; 100 μm in brain. OPL, outer plexiform layer. (D) In-gel activity of respiratory chain complexes I, II, IV, and V. Twenty micrograms of mitochondrial protein from brains of WT, MT, and MT+VA mice was used for BN-PAGE, and the activities of complexes were measured in the presence of specific substrates. Coomassie staining was used as a loading control. SC, super complexes. (E) ATP levels among brains of WT (n = 8), MT (n = 8), and MT+VA (n = 8) mice were measured using a luciferin/luciferase assay. Absolute level of total cellular ATP was shown. (F) Western blot analysis of antioxidant proteins. Total cellular proteins in WT, MT, and MT+VA retinas were electrophoresed with PAGE and hybridized with catalase, SOD1, and SOD2 antibodies and GAPDH as a loading control, respectively. (G) Quantification of catalase, SOD1, and SOD2 in WT, MT, and MT+VA retina. Representative of 3 to 4 independent experiments. Data in A, E, and G are represented as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 by 1-way ANOVA followed by Tukey’s post hoc test.

We then examined if the VA administration restored mitochondrial dysfunctions in the retina by enzyme histochemistry staining for succinate dehydrogenase (SDH) and cytochrome c oxidase (COX) in the frozen retinal sections. As shown in Figure 6B, mutant retina exhibited markedly decreased SDH activity, particularly in the ISL, but mild reductions in COX activity, as compared with those in WT retina. In contrast, the SDH and COX staining in the brains between mutant and WT mice was not significantly different (Figure 6C). The discrepancy of complex II between the brain and retina was further assessed by Western blot analysis of complex I and II subunits. As shown in Supplemental Figure 6A, the levels of SDHA and SDHB were decreased in retina but not changed in the brain in the mutant mice, as compared with those in WT mice. We then analyzed the activities of OXPHOS complexes I, II, IV, and V using in-gel activity assays. Due to the difficulty in obtaining enough mutant mouse retinas, we used brains, which belong to the central nervous system and may share some common mitochondrial features with retina (56), for these experiments. Mitochondrial membrane proteins isolated from brains were separated by blue native–PAGE (BN-PAGE) and stained with specific substrates for each complex. As shown in Figure 6D, the brains of mutant mice exhibited markedly decreased activities of complexes I and IV but no changes in those of complexes II and V, as compared with WT mice. Furthermore, the levels of total cellular ATP production in the mutant brains were 73% of those in the WT brains (Figure 6E). Strikingly, the supplementation of VA remarkably elevated SDH and COX activities in the mutant retina (Figure 6B) and increased activities of complexes I and IV as well as levels of total cellular ATP in the mutant brains (Figure 6, D and E).

To test whether VA supplementation reduced the overproduction of ROS in the mutant retina, we measured the levels of antioxidant-related proteins, including catalase, SOD1, and SOD2, in mouse retina (45). As shown in Figure 6, F and G, the levels of SOD1, SOD2, and catalase in the mutant retina were 188.6%, 186.7%, and 190.5% of those in WT retina. Notably, VA supplementation gave rise to significantly reduced levels of SOD1, SOD2, and catalase in the mutant retina, indicating that VA supplementation reduced overproduction of ROS. The levels of ROS production were further measured by dihydroethidium staining in the frozen brain sections. As shown in Supplemental Figure 6B, increased levels of dihydroethidium staining reflecting the levels of ROS were observed in the mutant brains, and VA supplementation reduced the levels of dihydroethidium staining in the mutant brains, as compared with WT brains.