Retinal polyunsaturated fatty acid supplementation reverses aging-related vision decline in mice

Abstract

The retina is uniquely enriched in polyunsaturated fatty acids (PUFAs), primarily localized in cell membranes, where they govern membrane biophysical properties. During aging, alterations in lipid metabolism lead to reduced content of very long–chain PUFAs (VLC-PUFAs) in the retina, which is associated with normal age-related reductions in contrast sensitivity, diminished photoreceptor function and delayed rod-mediated dark adaptation recovery, and pathological age-related macular degeneration (AMD). ELOVL2 (elongation of very long chain fatty acids-like 2) encodes a transmembrane protein that produces precursors to docosahexaenoic acid (DHA) and VLC-PUFAs. The methylation status of the ELOVL2 promoter is currently one of the best predictors of chronological age. Here, we show that lower VLC-PUFA abundance in the aged mouse retina is accompanied by a reduction in visual function. Similarly, mice lacking ELOVL2-specific enzymatic activity (Elovl2^C234W) demonstrate reduced contrast sensitivity and slower rod-mediated dark adaptation. Intravitreal supplementation with the direct product of ELOVL2, 24:5n-3, in aged animals improved visual function for up to 4 weeks and reduced accumulation of APOE- and C3d-positive sub-RPE deposits. The gene expression pattern observed in supplemented retinas exhibited a partial rejuvenation profile, including decreased expression of aging-related genes and a transcriptomic signature resembling younger retinas. Last, human genetic data from the IAMDGC and UK Biobank linked two variants in the ELOVL2 locus with the onset of intermediate AMD, underlining the translational importance of our findings. Our work highlights VLC-PUFA supplementation as a potential therapeutic opportunity and defines ELOVL2 as a promising target for interventions to prevent age-related vision loss.

INTRODUCTION

The specific composition of lipids within membranes dictates their biophysical properties, such as diffusion, permeability, domain formation, and curvature generation. Age-related changes in membrane lipid composition have been postulated to be one of the hallmarks of aging (1–3). In the aged retina, polyunsaturated fatty acids (PUFAs), essential components of cellular membranes, show lower abundance, and this decrease is further exacerbated in retinas affected by age-related macular degeneration (AMD) (4, 5). Retinal tissue is particularly enriched in long- and very long–chain PUFAs (LC-PUFAs and VLC-PUFAs, respectively), which are integral components of photoreceptor disk membranes (6), and their depletion or reduction has been postulated to be one of the hallmarks of AMD (7).

Strategies aimed at preserving or replenishing VLC-PUFAs in the aging retina, such as dietary interventions with n-3 PUFAs and LC-PUFAs, are being investigated as potential approaches to maintain retinal health and function in older individuals (8, 9). Although some studies indicate improvement of vision after supplementation with docosahexaenoic acid (DHA) or both DHA and eicosapentaenoic acid (EPA) (10, 11), others, such as the Age-Related Eye Disease Study 2 (AREDS2), report no such correlation (12), deeming DHA and EPA supplementation trials inconclusive. Dietary and oral supplementation of animals with very high doses of VLC-PUFAs or with fish oil enriched in n-3 C24–28 VLC-PUFAs have shown promising results (10, 11). However, because of the high cost of synthesizing VLC-PUFAs and the limited accessibility to enriched fish oil, coupled with their relatively low efficacy, these methods currently lack practical applicability for preventing age-related decline in human vision.

The exact mechanisms underlying the decrease in VLC-PUFAs in aging and disease are not yet fully understood. The key enzyme involved in the elongation of LC- and VLC-PUFAs is ELOVL2 (elongation of very long chain fatty acids protein 2), which is an endoplasmic reticulum membrane-resident protein that produces precursors to DHA and VLC-PUFAs. Elovl2 is primarily expressed in tissues with high metabolic demands, such as the liver, retina, and brain (13, 14). Increasing methylation of the ELOVL2 regulatory region has been shown to be one of the best biomarkers of chronological aging (15–17). Recent work in our laboratory demonstrated that the expression and activity of ELOVL2 are closely linked to the aging process in the eye (14).

In this study, we identified and described critical molecular, structural, and functional changes in the aging retina and correlated these phenotypes with the lack of ELOVL2 activity and disturbed lipid composition. We show that intravitreal injection of the direct product of ELOVL2 elongation, 24:5n-3, improves visual function, reduces the severity of aging phenotypes, and decreases accumulation of sub–retinal pigment epithelial (RPE) deposits in aged mice. No other PUFA demonstrated such a profound effect. Last, we present human genetic data showing a correlation between two variants in the ELOVL2 locus and the onset of intermediate AMD. Our data underscore the importance of ELOVL2 activity in maintaining healthy vision and suggest a potential new therapy to reverse the symptoms of aging in the eye and prevent age-related eye diseases such as AMD.

RESULTS

Age-related vision decline is associated with decreased VLC-PUFA abundance in the mouse retina

To investigate changes in the lipid composition of aging retinas, we performed a series of lipidomic analyses on 3-, 6-, 12-, 18-, and 23-month-old dissected mouse tissues. Lipids were extracted using the Bligh-Dyer method (18), and fatty acids were released from lipids using the acid hydrolysis method and extracted using hexanes (19). Analysis was performed by liquid chromatography–mass spectrometry (LC-MS), as described in Materials and Methods. Our data revealed a progressive decline in abundance of DHA (22:6) and VLC-PUFAs in 18- and 23-month-old retinas compared with 3-month-old retinas (Fig. 1A).

Fig. 1. Age-related decreased VLC-PUFA abundance in the retina is associated with vision decline in mice.

(A) Relative abundances of 18:0, 20:4n-6, 22:6n-3, and VLC-PUFAs in retinas of 3-, 6-, 12-, 18-, and 23-month-old mice (n = 4 or 5 mice per age group). (B) Relative abundances of major phospholipid classes in retinas of 3- and 18-month-old mice (n = 5 per age group). PI, phosphatidylinositol. (C) Quantification of lipid classes in aging POS (n = 10 per age group). AcCa, acyl carnitine; Cer, ceramides; ChE, cholesterol ester; Hex1Cer, simple Glc series; Hex2Cer, simple Glc series; LPE, lysophosphatidylethanolamine; LPS, lipopolysaccharide; MG, monoglyceride; WE, wax esters. (D) LION analysis of significantly changed lipids in aging (18-month-old) POS compared with young (3-month-old) POS (n = 10 per group). Darker shades represent lower P values, whereas lighter shades indicate larger P values. (E) Heatmap showing the change in VLC-PC abundance in POS of 18-month-old compared to 3-month-old mice (n = 10 per age group). (F) Scotopic and (G) photopic ERG responses in 18-month-old compared with 3-month-old mice (n = 10 per age group). (H) Rates of rod-mediated dark adaptation recovery in 3- and 17-month-old mice (n = 14 per age group). NS, not significant. (I) Oscillatory potential (OP) amplitudes of dark-adapted scotopic ERG in 17- and 3.5-month-old mice (n = 14 per age group). (J) Concentration of PDE6A, PDE6B, and transducin proteins in 3- and 18-month-old retinas quantified using a SIL peptide–based method (n = 5 per age group). (K) Immunostaining and fluorescent quantification (right) of C3d and APOE (markers of AMD in 5- and 26-month-old) retinas. Scale bar, 10 μm. Ch, choroid; OS, outer segments. Data are presented as mean ± SEM. Two-way ANOVA followed by a Geisser-Greenhouse correction [(A), (F), (G), and (H)] or unpaired t tests [(B), (C), (E), (I), (J), and (K)] were used to evaluate statistical significance. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Analysis of complex lipids demonstrated that among the 528 lipids quantified in retinas, the abundance of 15 lipids was significantly reduced, and the abundance of 53 lipids was significantly increased in aged retinas [fold change (FC) > |1.5| and P < 0.05] (fig. S1A). Principal components analysis revealed clear clustering based on differences in lipid composition between 3- and 18-month-old retinas (fig. S1B). The two major membrane lipid classes, phosphatidylcholine (PC) and phosphatidylethanolamine (PE), were altered in opposite directions in 18-month-old retinas. Specifically, PC was significantly decreased (P = 0.0039) and PE was increased (P = 0.0329) in aged retinas (Fig. 1B). Next, we isolated photoreceptor outer segments (POS) from 3- and 18-month-old mouse retinas. Upon lipid extraction, we found a significant (5%; P < 0.01) decrease in PC and increase in phosphatidylglycerol (PG), sphingomyelin (SM), and triglycerides (TGs) (18, 26, and 25%, respectively) in aged retinas (Fig. 1C). On the basis of all the significantly changed lipids in aging POS (fig. S1C; FC > |1.5| and P < 0.05), we performed Bioinformatics Methodology for Pathway Analysis (20), which revealed that the synthesis of TG from diglycerides (DGs) was highly activated (z-score = 2.957) in 18-month-old POS, where- as the formation of DG from TG was inhibited (z-score = −3.165) (fig. S1D). Analysis of our previous transcriptomic data from young and old retinas (21) revealed that, although the expression of the gene encoding the enzyme with triacylglycerol lipase and transacylase activity (Pnpla5) remained unchanged, the expression of the gene encoding diacylglycerol O-acyltransferase 2 (Dgat2)—the enzyme responsible for catalyzing the final reaction in TG synthesis—was elevated in aged retinas (fig. S1E). To further analyze the lipidomic dataset on purified POS from 3- and 18-month-old retinas, we performed lipid ontology (LION) enrichment analysis using LION/web, a software that serves as an ontology database for lipid metabolism, linking more than 50,000 lipid molecular species to their biophysical, chemical, and cell biological characteristics (22). The LION terms for significantly (P < 0.05) up-regulated biological processes revealed highly enriched changes in membrane components, decreased bilayer thickness, and high lateral diffusion in the membranes (Fig. 1D). Next, we focused on the amount of VLC-PUFAs incorporated in complex lipids and their impact on membranes in the aging retina. PC-VLC-PUFA and PE-VLC-PUFA abundances in POS were analyzed using Lipid Data Analyzer with a customized database, demonstrating a decrease in PC (50:12) and PE (48:12) in the aging POS (Fig. 1E and fig. S1F).

To correlate age-related lipid changes with visual function, visual tests were performed on 3- and 18-month-old mice. Optomotor response (OMR) analysis revealed significantly (P < 0.001) reduced scotopic contrast sensitivity in 18-month-old mice (fig. S1G). Electroretinography (ERG) demonstrated significant (P < 0.0001) decreases in both scotopic a- and b-wave amplitudes (Fig. 1F) and reduced photopic green/blue a- and b-waves in aged mice (Fig. 1G and fig. S1H). Rod-mediated dark adaptation was also impaired with a 34% reduction in maximal scotopic a-wave in 18-month-old mice (166 ± 5 μV) as compared with that in 3-month-old animals (250 ± 12 μV; ***P < 0.001) (fig. S1I). Although rods in both age groups showed recovery of their photoresponses after bleaching 90% of their visual pigment, aged mice had a slower recovery rate (24.3 ± 2.7 min versus 18.9 ± 0.7 min; P < 0.001) but reached similar final amplitudes (81 ± 4%) (Fig. 1H). Oscillatory potentials were consistently reduced in older mice under both dim flash and bright flash scotopic responses (Fig. 1I). However, the recovery of rod-driven ERG a-wave sensitivity (S_f) after the same bleach was unaffected (fig. S1J), suggesting an intact visual cycle.

To address whether age-related visual decline involves changes in the phototransduction cascade, we quantified proteins involved in signal transduction in POS, [rhodopsin (RHO), Rod cGMP-specific 3′,5′-cyclic phosphodiesterase subunit alpha (PDE6A), Rod cGMP-specific 3′,5′-cyclic phosphodiesterase subunit beta (PDE6B), guanine nucleotide-binding protein G(t) subunit alpha-1 (GNAT1), and guanine nucleotide-binding protein G(I)/G(S)/G(T) subunit beta-1 (GBB1)] using a stable isotope–labeled (SIL) peptide-based absolute quantification method. SIL peptides were spiked into 3- and 18-month-old retina samples, which were then trypsinized and analyzed by a targeted parallel-reaction monitoring method using LC–tandem MS. Protein amounts were calculated from endogenous/SIL peptide peak area ratio, normalized to RHO, and displayed as the mean ± SEM. Absolute quantification of these proteins showed slight decreases in Pde6a and Pde6b in 18-month-old retinas (Fig. 1J); however, relative amounts of all measured proteins (PDE6A, PDE6B, GNAT1, and GBB1, normalized to RHO) remained unchanged (fig. S1K).

To assess age-related changes in the expression of proteins associated with retinal pathologies, we performed immunofluorescence staining on retinal cross sections from 5- and 26-month-old mice. Cross sections were stained with antibodies targeting complement factor 3d (C3d) and apolipoprotein E (APOE)—both of which are more abundant in the aged retina and correlate with increased AMD risk (23–25). Our data also demonstrated increased expression of these proteins in the RPE cell layer of 26-month-old animals when compared with 5-month-old retinas (Fig. 1K). Staining with rod bipolar and Müller cell–specific antibodies (protein kinase Cα and glutamine synthetase, respectively) also revealed age-related morphological changes in the outer plexiform layer, including overgrowth of dendrites into the outer nuclear layer (ONL) (fig. S1, L and M).

Age-related changes in Elovl2 expression in the mouse retina.

ELOVL2 encodes a key enzyme in the biosynthesis of LC-PUFAs (Fig. 2A). Specifically, it elongates 22:5n-3 to 24:5n-3, which is then elongated to VLC-PUFAs by other elongases, including ELOVL4, or further processed in peroxisomes to produce 22:6n-3 (DHA). Our previous data indicated higher methylation of the Elovl2 regulatory region in the aged retina (14). Here, we analyzed methylation status at each relevant CpG site in Elovl2 (Fig. 2B) (15). We isolated DNA from middle-aged (<12-month-old) and old (24-month-old) mouse retinas and treated it with sodium bisulfite to distinguish methylated from unmethylated cytosines. Using both converted and nonconverted DNA, we amplified target regions with methylation-insensitive primers. Polymerase chain reaction products were sequenced using Sanger sequencing, and methylation abundance was quantified by comparing the area under the sequencing peak in both DNA samples. Our data revealed five CpGs with higher methylation in older retinas, whereas the four CpGs near the transcription start site remained unchanged. The CpG sites that exhibited higher methylation in mice are direct homologs of CpGs methylated in aging human blood (26).

Fig. 2. Age-related decrease in Elovl2 expression in mice.

(A) Illustration of VLC-PUFA elongation pathways. VLC-PUFAs are produced from essential fatty acid elongation by enzymes including ELOVL2, ELOVL4, and ELOVL5. (B) Left: Example of Sanger sequencing chromatogram. Right: Quantification of DNA methylation. Bottom: Locus organization and localization of CpGs relative to the transcription (Txn) start site of Elovl2. (C) Representative RNA in situ hybridization images of mouse retinal cross sections from 5-month-old (top panels) and 26-month-old (bottom panels) animals stained with RNAscope probes designed for Elovl2 (purple), Elovl4 (green), and Elovl5 (red), counterstained with Hoechst (blue). Scale bar, 25 μm. ONL, outer nuclear layer; INL, inner nuclear layer. (D) Uniform manifold approximation and projection (UMAP) visualization of snRNA-sequenced retinal tissue from 3-month-old (left) and 18-month-old (right) mice. The different retinal cell types are color coded. The scaled expression of Elovl2 in cone photoreceptors is shown in the bottom graphs. (E) GSEA of pathways enriched in cones from 3- and 18-month-old mice. (F) Interaction strength, representing ligand-receptor–mediated intercellular communication probability quantified by the law of mass action, was calculated using CellChat and visualized across young (left) and old (right) retinas. Each plot represents interactions among cones, rods, and Müller glial cells. The thickness of the lines between the cells indicates the interaction strength, with thicker lines representing stronger interactions. Data are presented as mean ± SEM, and unpaired t tests (B) were used to evaluate statistical significance. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

To visualize cell type– and age-specific expression of key enzymes in the PUFA elongation pathway, we performed RNA in situ hybridization on retinal cross sections from 5- and 26-month-old mice using probes targeting Elovl2, Elovl4, and Elovl5 (Fig. 2C). Elovl2 expression was predominantly observed in the photoreceptor layer, whereas Elovl4 was expressed throughout all retinal layers, and Elovl5 was expressed in the inner nuclear and ganglion cell layers (GCLs). Elovl5 expression remained relatively unchanged with age, but Elovl2 and Elovl4 expression decreased in 26-month-old mice, as previously reported (14).

To identify specific cell types expressing Elovl2 and other PUFA elongation enzymes, we performed single-nucleus RNA sequencing (snRNA-seq) on retinas isolated from 3- and 18-month-old mice. After normalizing to TATA-box binding protein (Tbp) mRNA, Elovl2 expression was significantly decreased in aged retinas (0.080 versus 0.100 in young retinas; P = 3.59 × 10⁻⁶). Elovl4 expression was also significantly lower (0.551 versus 0.643 in young retinas; P = 2.2 × 10⁻⁶), whereas Elovl5 showed no age-related change. As in human retinas (27), expression of Elovl2 was highest in cones and barely detectable in other cell types (Fig. 2D). Our data are in agreement with those obtained by the CZI CELL×GENE Discover program (28) from mouse and human retinas (fig. S2A), confirming the enrichment of Elovl2 in cones, whereas other enzymes were less restricted (Fig. 2D and fig. S2A). Differentially expressed genes (DEGs) [log FC > 0.1 and adjusted P (padj) < 0.05] were identified in cones from 3- and 18-month-old mice, which were then used in Gene Ontology (GO) functional enrichment analysis (Fig. 2E). In young cones, enriched pathways were related to visual perception and light detection, whereas in aged cones, they involved synaptic transmission, ion channels, and membrane assembly.

To better understand changes in intercellular interactions during retinal aging, we used CellChat (29) to analyze communication between cones, rods, and Müller glia (Fig. 2F and fig. S2, B and C). CellChat infers signaling from snRNA-seq data by assessing ligand-receptor expression, including soluble agonists, antagonists, and stimulatory or inhibitory membrane-bound coreceptors. In young mouse retinas, the strongest interactions occurred between Müller glia and cones, which intensified with age, along with increased signaling between rods and Müller glia. Cone-rod interactions were absent in young retinas but emerged in 18-month-old tissues. The analysis of specific ligand-receptor pairs within these signaling interactions revealed the emergence of links between neuronal adhesion molecules (Negr1-Negr1) and metabotropic glutamate receptors (Glu-Grm8) between rods and cones, specifically in aged retinas. In addition, several interactions between metabotropic glutamate receptors (Glu-Grm8 and Glu-Grm7) and ionotropic glutamate receptors (Glu-Gria4) emerged within Müller glia in aged retinas (fig. S2, B and C). In addition, our analysis showed previously unidentified cyclosporin A and CD147 (Ppia-Bsg) interactions between Müller and photoreceptor cells, which are recognized as an infection sensor and an inflammation initiation system (fig. S2B) (30).

Lack of ELOVL2 activity accelerates age-related phenotypic changes in the mouse retina

In our previous work, we generated Elovl2^C234W mice lacking ELOVL2 enzymatic activity because of impaired substrate binding (14). We performed comprehensive lipidomic and fatty acid analysis on retinas from 18-month-old Elovl2^C234W mice, which showed that the abundance of PUFAs synthesized from ELOVL2 product, including 22:6, 24:6, 32:6, 34:6, and 36:6, was lower in 18-month-old Elovl2^C234W than in age-matched wild-type retinas (Fig. 3A). Untargeted lipidomic analysis identified 595 lipids, with 28 significantly lower and 84 significantly higher abundances in Elovl2^C234W retinas compared with wild-type tissues (FC > |1.5| and P < 0.05) (fig. S3A). In addition, Elovl2^C234W tissue displayed a decrease in PC and an increase in PE abundance (Fig. 3B), similar to the changes detected in aging retinas (Fig. 1B). LION enrichment analysis based on all significantly changed lipids suggested highly enriched changes in plasma membrane components, decreased bilayer thickness, and high lateral diffusion (Fig. 3C), resembling the changes in aging POS. Further VLC-PCs analysis showed decreased PC (46:12), PC (48:12), PC (50:12), PC (54:12), and PC (56:12) (Fig. 3D), which contributed to a reduced total amount of VLC-PCs in the 18-month-old Elovl2^C234W retina (fig. S3B). An analysis of free fatty acids (FFAs) also showed a decrease in LC- and VLC-PUFAs in 18-month-old Elovl2^C234W retinas, as compared with age-matched wild-type samples (fig. S3C).

Fig. 3. Changed lipid composition in Elovl2^C234W mouse retinas is correlated with vision loss.

(A) Quantification of total fatty acid products of ELOVL2 elongation in retinas of Elovl2^C234W mice compared with age-matched wild-type mice (n = 4 per group). WT, wild-type. (B) Changes in major phospholipid classes in Elovl2^C234W retinas (n = 2) compared with age-matched wild-type retinas (n = 3). (C) LION analysis of changed lipids in Elovl2^C234W retinas (n = 2 per group). (D) VLC-PC species in Elovl2^C234W retinas (n = 2). (E) Quantification of OMRs to a rotating grating stimulus of differing contrast (n = 6 12-month-old wild-type, n = 7 12-month-old Elovl2^C234W, and n = 7 18-month-old mice), (F) scotopic ERG responses, and (G) ERG a-wave amplitudes of 12-month-old Elovl2^C234W retinas compared with those of 18-month-old wild-type retinas (n = 6 12-month-old wild-type, n = 7 12-month-old Elovl2^C234W, n = 13 18-month-old wild-type, and n = 14 18-month-old Elovl2^C234W). (H) Rod-mediated dark adaptation of Elovl2^C234W mice (n = 10) compared with age-matched wild-type mice (n = 11). (I) Quantification of retinoid abundance in wild-type and Elovl2^C234W eyeballs (n = 3 per group). (J) Quantification of the ONL thickness between 12-month-old wild-type (n = 5) and Elovl2^C234W retinas measured by OCT (n = 5) and 18-month-old Elovl2^C234W (n = 4) and wild-type retinas (n = 4). ONH, optic nerve head. (K) Comparison of GSEA NES of selected GO terms in 12-month-old Elovl2^C234W versus 12-month-old wild-type retinas with the NES of GO terms in 18-month-old versus 12-month-old retinas (n = 3 12-month-old wild-type, n = 3 12-month-old Elovl2^C234W, and n = 4 18-month-old wild-type). Data are presented as mean ± SEM. Two-way ANOVA followed by a Geisser-Greenhouse correction [(E), (F), and (H)] or unpaired t tests [(B), (C), (E), (I), (J), and (K)] were used to evaluate statistical significance. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

A comparison of visual function in 12- and 18-month-old Elovl2^C234W mice to age-matched wild-type mice detected that scotopic contrast sensitivity in both ages was lower in Elovl2^C234W animals compared to age-matched wild-type mice (Fig. 3E). Scotopic a- and b-wave ERG responses were also reduced at both ages compared with wild-type controls (Fig. 3F and fig. S3D). Furthermore, in wild-type mice, scotopic a- and b-wave ERG responses decreased markedly from 12 to 18 months (Fig. 3F and fig. S3D, blue lines). In contrast, Elovl2^C234W mice showed minimal change in ERG amplitudes between these ages (Fig. 3F and fig. S3D, red lines). The maximum ERG a-wave amplitudes in 12-month-old Elovl2^C234W (120 μV) mice were decreased and comparable to those in 18-month-old control (150 μV; P = 0.103) animals (Fig. 3G). There were no differences in photopic ERG a- and b-wave amplitudes between wild-type and Elovl2^C234W mice (fig. S3E and data file S1).

The rod dark adaptation in ELOVL2-deficient mice was then evaluated after a nearly complete RHO bleach with green light. In wild-type mice, the averaged scotopic ERG a-wave maximal response, A_max, recovered with the time constant of 28.6 ± 2.6 min to about 90 ± 7% of the prebleach value after 60 min (Fig. 3H, black symbols). In contrast, the average rate of rod A_max recovery in Elovl2^C234W mice was about 1.5 times (42.7 ± 7.3 min) slower than in wild-type animals (P = 8 × 10⁻⁶) and reached only 69 ± 5% of its prebleached status by the end of 60-min recordings (Fig. 3H, red symbols). The dark-adapted a-wave photosensitivities (S_f) were also slightly lower (16%) in the same group of mice lacking ELOVL2 (1.50 ± 0.05 m² cd⁻¹ s⁻¹ versus 1.79 ± 0.08 m² cd⁻¹ s⁻¹ in controls; P < 0.05) (fig. S3F, left). However, the recovery of rod-driven S_f after the same bleach was not suppressed in ELOVL2-deficient animals (fig. S3F, right), suggesting the normal recycling of the visual chromophore in the mutant mice. To confirm this finding, we extracted retinoids from dark-adapted 18-month-old Elovl2^C234W and age-matched wild-type animals. Separation by normal phase high-performance liquid chromatography and quantification of 11-cis-retinal and all-trans-retinal species revealed no difference between Elovl2^C234W and wild-type animals (Fig. 3I).

To determine whether the reduction in ERG amplitude in Elovl2^C234W mice could be due to cell loss or retinal degeneration, we measured ONL thickness by optical coherence tomography (OCT) and found no significant difference between 12-month-old wild-type and Elovl2^C234W and 18-month-old wild-type and Elovl2^C234W eyes (Fig. 3J and fig. S3G). Therefore, the reduction in rod function in 12-month-old Elovl2^C234W animals preceded any detectable photoreceptor cell loss or retinal degeneration.

Given that Elovl2^C234W mice exhibit accelerated age-related retinal changes, we investigated whether their transcriptomic profiles resemble those of the aged retina. Bulk RNA-seq was performed on retinas from 12- and 18-month-old wild-type mice and compared to 12-month-old Elovl2^C234W retinas to identify transcriptomic similarities associated with the absence of the ELOVL2 enzyme and in aging. Gene set enrichment analysis (GSEA) of DEGs (FC > |1| and padj < 0.05) using GO revealed several shared patterns. Down-regulated pathways in both aged and Elovl2^C234W retinas included telomere maintenance, transcription by RNA polymerase I, negative regulation of gene expression, and regulation of gene silencing (Fig. 3K). Conversely, synapse assembly, transmembrane transporter activity, target of rapamycin complex 1 (TORC1) signaling, DNA damage response, signal transduction resulting in transcription, and ceramide metabolic process were up-regulated in both 12-month-old Elovl2^C234W and 18-month-old wild-type tissues when compared with 12-month-old wild-type retinas (Fig. 3K and fig. S3H).

Improved visual function in aged mice 5 days after intravitreal fatty acid injection

We hypothesized that the lack of direct ELOVL2 product, 24:5n-3, in the aging retina is one of the main culprits of age-related visual decline and that supplementation with this fatty acid may improve vision in aged animals. To test this hypothesis, we pursued an intravitreal supplementation strategy, which, in contrast with oral gavage, enabled us to precisely control the amount of lipid administered to the eye.

First, to evaluate the extent to which intravitreally injected fatty acid crosses the inner limiting membrane (ILM) and reaches photo-receptors, we injected biotinylated PUFA (linoleic acid) intravitreally and followed its localization 45, 90, and 120 min after injection. To visualize the lipid, fluorescently labeled streptavidin was applied to retinal cross sections (fig. S4A). Forty-five minutes after injection, the lipid was mostly detectable in the GCL. Next, the biotinylated PUFA was detected in all retinal cell layers, including the RPE. After 120 min, a considerable amount of the injected lipid was accumulated in the RPE cell layer.

To assess the retinal toxicity of 24:5n-3, intravitreal injections were administered unilaterally to 3-month-old mice, and visual function was tested at several time points postinjection (fig. S4B). The potential optimal dose of supplementation was estimated as half of the difference in the total amount of VLC-PUFAs between young and old retinas (0.36 nmol). Our analysis shows no changes in scotopic ERG a- or b-wave amplitudes on day 2 (fig. S4C) or ERG amplitudes (fig. S4D) and rod-mediated dark adaptation recovery (fig. S4E) on day 5 between vehicle- and 24:5-injected eyes. Furthermore, no differences in OMR responses (visual acuity and contrast sensitivity) on day 3 or 10 were detected (fig. S4, F and G). Structurally, no obvious retinal degeneration or inflammation was observed 2 weeks after 24:5n-3 injection, as quantified by ONL thickness (fig. S4H) and the absence of autofluorescent spots indicative of inflammation on fundus photography (fig. S4I).

To investigate the short-term effect of 24:5n-3 supplementation on visual function in aged mice, we administered the compound intravitreally into one eye of each 18-month-old animal, whereas the contralateral eye received an injection of a vehicle as a control. Five days postinjection, visual function was assessed, and retinas were collected for lipid profiling and molecular analysis (Fig. 4A). First, we tested different doses of 24:5n-3 to determine the optimal amount that could improve vision in 18-month-old mice. We observed that 24:5n-3–treated eyes showed a considerable improvement in both scotopic and photopic ERG a- and b-wave responses with an injection dose of 0.36 nmol (Fig. 4B, and data file S1), and no changes were observed when other doses were used (fig. S4J). Then, to test whether PUFAs up- or downstream of 24:5n-3, such as 20:5n-3, 22:6n-3 (DHA), and 32:6n-3, have similar effects, we replicated the procedure using the optimal dose (0.36 nmol) (fig. S4K and data file S1). Injection of 22:6n-3 and 20:5n-3 resulted in no functional improvement, but marginal improvement in photopic green and blue light–generated ERG b-wave responses was observed in eyes treated with 32:6n-3 (fig. S4K).

Fig. 4. Intravitreal supplementation of 24:5n-3 in aged mice improves visual function.

(A) Timeline for intravitreal 24:5n-3 supplementation study in 18-month-old mice. IVT, intravitreal; D1, day 1. (B) Measurements of scotopic a- and b- (n = 7 per group) and photopic green (n = 9 per group) and blue (n = 8 per group) b-wave ERG responses 5 days postinjection of 24:5n-3 or vehicle. Veh, vehicle. (C) Recovery of scotopic ERG maximal a-wave amplitudes (A_max) after bleaching > 90% of RHO at time 0 in vehicle- and 24:5-injected mice (n = 9 per group). For each time point, A_max was normalized to corresponding prebleach dark-adapted value (A_max^DA). Averaged data points were fitted with the equation A*[1 − exp(−x/tau)]. (D) Average time traces (with SEM) of VEP in the superior colliculus (SC) 5 days after intravitreal supplementation of 0.36-nmol 24:5n-3 (n = 16) and vehicle (n = 14). (E) Population average of the peak-to-peak (minimum and maximum values) evoked potential amplitudes between time 0 and 0.5 s calculated from single electrode (single dot) traces for the on response. Data are presented as mean ± SEM. Two-way ANOVA followed by a Geisser-Greenhouse correction [(B) and (C)] or unpaired t test (E) was used to evaluate statistical significances. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

The effect of intraocular 24:5n-3 supplementation on rod dark adaptation was then assessed by in vivo ERG. In 17-month-old mice, this treatment improved the final postbleach fraction of recovered rod A_max response by 30% (Fig. 4C), as compared with that in vehicle-injected contralateral eyes, without affecting their sensitivity recovery (fig. S4L). In contrast, a similar administration of 24:5n-3 did not affect the recovery of either of the two parameters in 3.5-month-old animals (fig. S4E), thus indicating the specificity of the therapeutic effect of the lipid to aged mice.

We next investigated whether the observed improvement in photoreceptor activity was correlated with an increased amount of information transmitted to the brain by recording visually evoked potentials (VEPs). Our recordings in the superior colliculus showed an improvement in VEP signal (Fig. 4D). Detailed analysis showed a significant increase (P = 0.00082) in the population average of the peak to peak (minimum and maximum values)–evoked potential amplitudes between time 0 and 0.5 s calculated from single-electrode traces in 18-month-old animals supplemented with 24:5n-3 compared with the vehicle-treated eyes (Fig. 4E). Specifically, the VEP amplitude increased from 44.15 ± 5.95 μV in 18-month-old animals injected with a vehicle (n = 4 animals; n = 14 recording sites) to 86.67 ± 8.61 μV in supplemented eyes (n = 4; n = 16 recording sites).

Partial reversal of age-related transcriptomic changes in the mouse retina after intravitreal 24:5n-3 injection

To investigate the mechanism underlying vision improvement after 24:5n-3 treatment, we first examined whether the lipid supplementation altered the lipid composition of the retina. Retinas from treated 18-month-old mice were collected 5 days after lipid or vehicle injection. We analyzed the abundance of all VLC-PUFA–containing phospholipids from isolated POS and found that all were elevated in the 24:5n-3 injected group. Among them, VLC-PUFA–incorporated PE, PC, lysophosphatidylcholine (LPC), phosphatidylserine (PS), and plasmalogen PE increased by 25% (P = 0.035), 26% (P = 0.070), 20% (P = 0.017), 22% (P = 0.025), and 40% (P = 0.007) in the 24:5n-3–supplemented group, respectively (Fig. 5A). Overall, phospholipids containing VLC-PUFAs were elevated by 25% (P = 0.033) in the POS after fatty acid supplementation. In addition, DHA-containing TG accumulated considerably (fig. S5A).

Fig. 5. Reversal of molecular aging phenotypes in 18-month-old 24:5n-3–supplemented eyes.

(A) Heatmap showing the changed expression of possible classes of VLC-PUFA–incorporated phospholipids in isolated POS after intravitreal supplementation of 24:5n-3 (vehicle: n = 17; 24:5: n = 16). (B) Volcano plot showing DEGs in 24:5(n-3)– compared with vehicle-injected retinas from 18-month-old mice (n = 3). (C) GSEA enrichment plots for complement and oxidative stress and redox pathways. Red indicates enrichment in 24:5-injected retinas, and blue indicates enrichment in vehicle-injected retinas (n = 3). (D) Metascape analysis demonstrated down-regulation of immune response, inflammation, microglial phagocytosis, and cell migration pathways after 24:5n-3 supplementation. The network plot visualizes functionally grouped enriched terms, where each node represents a biological pathway or process. Node size reflects the number of genes associated with each term, and node color indicates statistical significance, with darker colors representing lower P values. Edges (connecting lines) indicate a similarity score > 0.3 between terms, representing shared genes or functional overlap (n = 3). (E) STRING functional protein-protein interaction network of top 20 TFs involved in gene regulation after 24:5n-3 supplementation. (F) Heatmap illustrating changed expression of genes selected from the aging cerebellum gene set in the retinas of 18-month-old mice after 24:5n-3 supplementation (n = 3). (G) Representative immunofluorescence staining of mouse RPE cell layer stained for C3d (green), APOE (red), and Hoechst counterstaining (blue) from retinas supplemented with 24:5n-3 or vehicle (left). The intensity-based quantification of C3d and APOE expression is shown on the right (n = 3). Scale bar, 10 μm. (H) GSEA analysis of up-regulated pathways in 18-month-old vehicle and 24:5n-3–supplemented retinas compared with 4-month-old retinas (n = 3). Data are presented as mean ± SEM. Unpaired t tests [(A), (F), and (G)] were used to evaluate statistical significance. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

To assess molecular changes occurring after fatty acid supplementation, we performed bulk RNA-seq analysis of 18-month-old 24:5n-3– versus vehicle-injected retinas isolated 5 days after intravitreal injection. Among 234 significantly altered genes (FC > |1.5| and padj < 0.05), expression of key PUFA elongation enzymes—Elovl2, Elovl4, and Elovl5—remained unchanged after supplementation (fig. S5B). Nearly all DEGs were down-regulated in the 24:5n-3–supplemented retinas (Fig. 5B). GSEA indicated significant down-regulation of complement [normalized enrichment scores (NES) = −1.77 and false discovery rate (FDR) q = 0.0002] and oxidative stress and redox pathways (NES = −1.70 and FDR q = 0.006) in 24:5n-3–supplemented retinas (Fig. 5C). The visualization of enriched terms as a network using Metascape (31) demonstrated down-regulation of pathways involved in immune response, inflammation, microglial phagocytosis, and cell migration (Fig. 5D and fig. S5C).

To identify transcription factors (TFs) driving gene expression changes after 24:5n-3 supplementation, we used ChIP-X Enrichment Analysis 3, a tool that predicts TFs associated with DEGs on the basis of TF-target gene set libraries curated from human, mouse, and rat (32). Using mean rank output, which averaged integrated ranks across libraries, we found down-regulation of Sp100, Lyl1, Sp140, Irf5, Stat6, Fli1, Ikzf1, Spi1, Relb, Elf4, and Sp110 and up-regulation of Trafd1 (fig. S5D). To visualize protein-protein interactions among these associated TFs, we created a functional interaction network of the top 20 TFs using STRING (v12.0) (Fig. 5E) (33). Low-expressing TFs, as defined by normalized read count (DESeq2 baseMean) < 10, were removed, and the difference in mean z-scores after 24:5n-3 supplementation of the remaining 16 TFs was calculated. Our data demonstrated down-regulation of nearly all top TFs, including several interferon regulatory factors (IRFs)—Irf1, Irf2, Irf5, Irf7, and Irf8—responsible for activating immune response (34).

Further analysis demonstrated significant down-regulation of genes from the “aging cerebellum” gene set (35) (NES = −1.82 and FDR = 5.83 × 10⁻⁴) in retinas from 18-month-old mice after supplementation with 24:5n-3 compared with control, including significant down-regulation of apolipoprotein E (Apoe) transcript (log₂FC = −0.79 and padj = 0.001) (Fig. 5F). To corroborate our RNA-seq data showing decreased expression of Apoe and genes related to the complement cascade after supplementation, we performed immunohistochemical analysis of retinal cross sections (Fig. 5G). First, we observed a marked reduction of complement component C3d, the final degradation product of C3, which has been shown to be elevated in AMD (24, 25), in the RPE of 24:5n-3–supplemented eyes. We also observed decreased accumulation of APOE, a major component of age-related sub-RPE deposits (36).

Last, we investigated how transcriptional changes after 24:5n-3 supplementation in 18-month-old retinas compared with the transcriptional profile of 4-month-old retinas (Fig. 5H). Comparing 18-month-old vehicle-injected to 4-month-old retinas, GSEA revealed significant up-regulation of inflammatory response (NES = 1.67 and P < 10 × 10⁻⁵), interleukin-6 (IL-6)/Janus kinase (JAK)/signal transducer and activator of transcription 3 (STAT3) (NES = 1.61 and P < 10 × 10⁻⁵), complement (NES = 1.31 and P = 0.02), and apoptosis (NES = 1.29 and P = 0.04), consistent with our previous findings (21). After 24:5n-3 supplementation, several pathways, specifically, IL-6/JAK/STAT3, complement, and apoptosis, were no longer significantly up-regulated, whereas inflammatory response enrichment was lowered to NES = 1.44 (P < 10 × 10⁻⁵).

Sustained improvement of visual function in aged mice after a single 24:5n-3 intravitreal injection

To assess the persistence of visual improvement after 24:5n-3 supplementation in aged mice, two approaches were applied: (i) Long-term follow-up strategy: An intravitreal injection of 24:5n-3 at the optimal dose (0.36 nmol) was administered into one eye of each 18-month-old animal, whereas the contralateral eye received a vehicle as a control. Visual function was assessed by ERG at 2 and 4 weeks postinjection, followed by tissue collection and retinal dissection for lipid analysis (Fig. 6A). (ii) Repeated injection strategy: Starting at 16 months of age, one eye of each animal received an intravitreal injection of 24:5n-3 at the optimal dose (0.36 nmol) every 3 weeks, whereas the contralateral eye received a vehicle injection as a control. Visual function was assessed by ERG 3 weeks after each injection. Upon reaching 18 months of age (after the third injection), the retinas were collected for lipid analysis (fig. S6A).

Fig. 6. Long-term supplementation of 24:5n-3 in aged mice rescues visual function.

(A) Timeline for long-term follow-up study. (B) Scotopic a- and b-wave (top panels) and photopic b-wave (bottom panels) ERG responses 2 weeks (left panels; n = 7 per group) and 4 weeks (right panels; n = 10 per group) postinjection of 24:5n-3 or vehicle. (C) Heat maps showing changes in the abundance of possible classes of VLC-PUFA–incorporated phospholipids, VLC-PC species, TG species, and FFAs (all normalized to the control eye) in isolated POS 4 weeks postinjection of 24:5n-3 compared with age-matched vehicle-injected retinas (n = 5 per group). Data are presented as mean ± SEM. Two-way ANOVA followed by a Geisser-Greenhouse correction (B) or unpaired t tests [(C)] were used to evaluate statistical significance. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Using the long-term follow-up strategy, we observed that after a single 24:5n-3 injection, 18-month-old animals showed an improvement in both scotopic a- and b-wave and photopic blue b-wave responses at day 15 postinjection (Fig. 6B, left) and increased photopic green and blue b-wave amplitudes at day 29 postinjection (Fig. 6B, right). Lipidomic analysis of isolated POS 4 weeks postinjection revealed a slight increase in VLC-PUFA–incorporated phospholipids, particularly VLC-PC species, such as PC56:12 (increased by 50.6%; P = 0.04) (Fig. 6C). In addition, repeated intravitreal injections of 24:5n-3 in 16-month-old mice conducted every 3 weeks only found improvement of photopic blue b-wave responses after the first two injections. The third injection did not result in any visual improvement of either scotopic or photopic ERG assessments (fig. S6B and data file S1). A lipidomic analysis of isolated POS after the third injection still showed an increase in VLC-PUFA–incorporated phospholipids and particularly VLC-PC species, such as PC48:12 (31.8%; P = 0.0073) and PC58:12 (38.9%; P = 0.015). We noticed a significant elevation in free 32:6n-3 (55.0%; P = 0.0047) and accumulation of several DHA-containing TG species in the eyes that received three repeated doses of 24:5n-3 injection (fig. S6C). In contrast, retinas that received only one injection showed no accumulation of free VLC-PUFAs, and only one DHA-containing TG was elevated at 4 weeks postinjection (Fig. 6C).

Genetic variants in ELOVL2 locus associate with earlier onset of intermediate AMD

To investigate the association between genetic variants within the ELOVL2 gene (chr6: 10980992–11044624) and the age of onset of intermediate AMD, we used two independent cohorts: unrelated, European individuals with intermediate AMD from the International AMD Genomics Consortium (IAMDGC; n = 2407; mean age at diagnosis: 74.1 years) (37) and incident AMD cases from the UK Biobank (n = 1309; mean age of onset: 62.8 years) (38). The strongest association was observed for rs911196 in the fifth intron of ELOVL2. The minor G allele (allele frequency: 25% in Europeans) resulted in 4.7 months (95% confidence interval: 2.1 to 7.3 months) earlier onset of intermediate AMD in our analyses (Fig. 7; P = 0.0003). This effect was seen in both cohorts (IAMDGC: 5.7 months earlier onset; UK Biobank: 4.5 months earlier). The G allele of the correlated variant rs9468304 in the first intron also showed a significant association; however, the effect was weaker (31% allele frequency, 4.32 months earlier onset; P = 0.0009). None of the remaining variants had a statistically significant correlation (P > 0.05).

Fig. 7. Variants in the ELOVL2 gene locus correlate with the onset of AMD.

(Top) Lolliplot of ELOVL2 gene variants correlated to the age of onset of AMD. The negative decadic P value of the correlation between the alleles and age of onset/diagnosis is depicted on the y axis. The lead variant rs911196 is highlighted in purple, and the correlated variants are color coded according to their genetic correlation (R²) to the lead variant. (Bottom) CpGs found in the ELOVL2 locus and their effect on aging expressed as the slope (higher methylation in those sites was correlated to older age). In purple, CpGs positively correlated with the G allele at rs911196 causing earlier onset of AMD are highlighted.

DISCUSSION

In this work, we identified a potential therapeutic approach aimed at restoring the health of aged retinas. First, we performed a comprehensive analysis of aging phenotypes and molecular pathways specific to the aging mouse retina. Concurrently, we systematically characterized both vision and molecular alterations in the retinas of animals deficient in ELOVL2 function. We found that Elovl2^C234W animals exhibit an acceleration of retinal aging phenotypes. Next, we designed a treatment strategy involving the intravitreal injection of the elongation product of ELOVL2 (24:5n-3) and demonstrated its efficacy in enhancing vision in aged animals. We show that the proposed therapy reverses several functional and structural aging phenotypes, including the accumulation of age-related sub-RPE deposits enriched in APOE and C3d. Furthermore, at the molecular level, the treatment exhibited the ability to down-regulate the innate immune response, complement activation, and oxidative stress response—key factors contributing to vision loss in several age-related eye diseases. Last, our genetic analysis revealed genetic variants in the human ELOVL2 locus that correlate with an earlier onset of intermediate AMD, indicating the translational relevance of the proposed 24:5n-3 treatment.

The ELOVL2 gene encodes a transmembrane enzyme involved in the synthesis of long-chain (C22 and C24) n-3 and n-6 PUFAs (Fig. 2A). Specifically, ELOVL2 elongates docosapentaenoic acid (DPA) (22:5n-3) to 24:5n-3, which is a precursor for VLC-PUFAs and 22:6n-3 (DHA). Total lipid extracts from the photoreceptors of patients with AMD contain lower amounts of DHA and VLC-PUFAs compared with those of age-matched normal donors (4). The regulatory element of ELOVL2 becomes increasingly methylated with age, a phenomenon observed across various tissues and animal species, including rodents and humans (15, 17, 39, 40). This reproducible direct relationship has been shown to be one of the most reliable DNA methylation markers associated with chronological age of organisms (15, 16, 41–45). Our snRNA-seq data confirmed that Elovl2 expression is highest in cones and decreases with age in the retina. Rods are much more numerous in the mouse retina. Consequently, in cases where PUFA delivery from the blood is reduced, as well as in regions with sub-RPE deposits or drusen, the synthesis of VLC-PUFAs and DHA may rely more on cones. Similarly, with aging and the decline in the efficiency of nutrient delivery pathways (choroid, RPE, and retinal vasculature), cones may become the primary source of these lipids, potentially supplying them to rods. Supporting this hypothesis, our CellChat analysis indicates that, in aging, rods and cones interact more closely at the molecular level. Furthermore, both cell types exhibit increased interactions and exchange inflammatory signals with Müller glia (30), a cell type that highly expresses Elovl5 (fig. S2A), encoding an enzyme that provides substrates for ELOVL2 (Fig. 2A).

To study the role of ELOVL2 in vision, we generated a homozygous mutant mouse, Elovl2^C234W, which lacks the ability to convert the 22:5n-3 PUFA (DPA) to 24:5n-3 (46, 47). Here, our comprehensive lipidomic analysis unveiled notable similarities in the lipid composition of the retina and POS between aging and Elovl2^C234W animals, which carry a mutation causing functional deficiency. Consistent with the finding that most DHA and VLC-PUFAs in the retina are acquired during embryonic development, both cases showed reduced, but not absent, abundance of DHA and VLC-PUFAs. Our longitudinal studies on vision revealed an earlier onset of vision decline in the mutant retinas at 12 months. Nevertheless, we did not detect any retinal degeneration in Elovl2^C234W animals between 12 and 18 months, consistent with reports indicating that retinal degeneration in aging is only detectable at 24 months (48).

Currently, there is no effective treatment to slow early or intermediate AMD progression. Whereas anti–vascular endothelial growth factor therapy has transformed care for neovascular AMD, no breakthrough exists for early dry, or nonneovascular, AMD. Although inhibitors of complement proteins C3 (pegcetacoplan) (49) and C5 (avacincaptad pegol) (50) are approved for geographic atrophy (GA) in the United States, therapies for earlier stages remain unavailable. Epidemiological studies have suggested that diets rich in n-3 PUFAs may be associated with lower occurrence of AMD, whereas low dietary intake of n-3 PUFAs may be associated with an opposite effect (51, 52). Two early surveys found that high plasma abundance of n-3 PUFAs was correlated with decreased risk of AMD (53, 54). Nutritional supplementation with EPA and DHA, however, has yielded inconclusive results. In two large prospective studies, the AREDS2 and the nutritional AMD study (Nutritional AMD Treatment-2, NAT-2), no difference was observed between oral supplementation with DHA and EPA and placebo in the progression of the disease (12, 55). In the AREDS2 study, a retrospective analysis into factors influencing progression of AMD, participants with the highest self-reported intake of foods rich in n-3 LC-PUFAs were 30% less likely to develop central GA and 50% less likely to develop AMD compared with those with the lowest self-reported intake (56). One possible explanation for these findings is that, given the advanced age of AMD participants, and therefore a loss of ELOVL2 expression in the eye, it may not be possible to process supplemented shorter lipids into VLC-PUFAs. Recent research from the Bernstein laboratory (10) supports this hypothesis, showing that Elovl4 mutant animals supplemented orally with a high dose of 32:6n-3 VLC-PUFA, rather than DHA and EPA, exhibited a modest improvement in vision.

Although several laboratories have undertaken studies with oral or dietary supplementation of PUFAs to prevent vision loss in aging or disease (10, 11), notable progress has only been achieved recently. In light of inconclusive results from EPA and DHA supplementation, focus has shifted to longer n-3 PUFAs. In addition to the short-term feeding study using 32:6n-3 (10), more recently, 8-week dietary supplementation of fish oil enriched in C24-C28 n-3 PUFAs in 9-month-old animals showed slightly improved ERG responses and visual performance, presenting promising alternatives to DHA- and EPA-enriched diets (11). Together, these data suggest that the beneficial impact of high fish intake observed in AREDS studies might be attributed to the requirement for PUFAs longer than EPA and DHA in the eye.

After our analysis of physiology and lipid metabolism in both aged and Elovl2^C234W mice, we decided to use a product of PUFA elongation facilitated by the ELOVL2 enzyme, namely, 24:5n-3, in a direct supplementation strategy. For this proof-of-concept study, we opted for the intravitreal route of delivery to precisely control the amount of fatty acid delivered to the retina. In addition, this approach offers direct insights into a potential clinical therapeutic strategy in lieu of nutrition-based approaches. By tracing the localization of intravitreally injected biotinylated PUFA, we observed that the fatty acid efficiently crossed the ILM and reached RPE within hours. A single intravitreal injection did not induce adverse effects on vision or retinal structure in either young or old animals. In contrast, improvements in scotopic and photopic ERG recordings, VEP readings, and recovery of rod maximal response after photobleach in aged animals underscored the beneficial effects of the treatment. Moreover, long-term follow-up experiments demonstrated improved ERG responses up to 4 weeks post–single dose administration.

Lipidomic analysis after 24:5n-3 supplementation revealed a modest increase in VLC-PUFA–containing phospholipid species in POS after both short- and long-term treatments. In addition, we observed an accumulation of multiple DHA-containing TGs 5 days after injection with only one 22:6n-3–containing TG species remaining elevated 4 weeks after 24:5n-3 administrations. These findings suggest two fates of the supplemented 24:5n-3: (i) A portion of the injected 24:5n-3 was elongated to VLC-PUFAs and incorporated into phospholipids, potentially integrating into the photoreceptor membrane to restore VLC-PUFA abundance and support sustained visual improvement with aging. (ii) Another fraction was converted to 22:6n-3 and stored in TGs, which can be facilitated by increased levels of Dgat2 in aging retinas (fig. S2D). The minimal changes in DHA-containing TGs 4 weeks after a single injection also suggest a limited benefit for vision improvement beyond this time frame. Last, three injections of the same dose of 24:5n-3 caused accumulation of several DHA-containing TGs and free VLC-PUFA accumulation, whereas no vision improvement was observed. Together with the observation that no changes in scotopic or photopic ERG responses were seen 5 days postadministration of high doses (0.72 and 2.5 nmol) (fig. S4J), these findings suggested that excessive lipid supplementation may diminish the beneficial effects of treatment. Studies using a labeled (deuterated) 24:5n-3 for supplementation should be performed to understand the fate of the injected fatty acid and its direct impact on vision as elongated molecules in the long term.

The modest increase in VLC-PUFA–containing lipids raises questions about the efficiency of the 24:5n-3 supplementation strategy. In vitro studies have shown that even minute increases in long omega-3 lipids in membranes can alter their biophysical properties, affecting membrane domain organization and viscosity (57, 58). These changes could potentially affect the function of all membrane-residing proteins, such as channels and G protein–coupled receptors. Consequently, excessive lipid supplementation, for example, through repeated administration, may lead to an accumulation of free VLC-PUFAs, which could counteract the beneficial effects of supplementation on vision during aging. Our detailed studies underscore the importance of carefully optimizing the dosage and regimen of supplementation to maximize therapeutic benefits of the treatment.

We cannot, however, exclude the possibility that 24:5n-3 plays an additional role in the retina, as suggested by the study involving fish oil enriched in C24-C28 n-3 PUFAs, where dietary supplementation favorably altered the transcriptional regulation of peroxisome proliferator–activated receptor (PPAR) targets in the liver (11). Our transcriptomic analysis of retinas collected 5 days after intravitreal 24:5n-3 supplementation did not demonstrate activation of PPAR pathways. However, it still revealed robust down-regulation of inflammation, including decreased activity of the complement pathway, suggesting that 24:5n-3 may act through mechanisms other than PPARs, such as reducing oxidative stress. In addition, we detected down-regulation of microglial phagocytosis and migration, hallmark processes affected in the aging retina (59–61). To further investigate the mechanisms underlying the improvement of retinal health in the aged mice after 24:5n-3 supplementation, we examined TFs that potentially regulate DEGs. Most of the down-regulated TFs have been previously shown to be key regulators of retinal microglial sensome and proinflammatory genes activated in microglia (60). Our data therefore underscore the role of microglia in retinal aging.

Aging is one of the major risk factors for developing AMD. A specific look at the aging cerebellum transcriptional pathway (35) revealed down-regulation of many cytokines, growth factors, and other molecules. ApoE, encoding one of the main proteins accumulated in the aging eye and a risk factor for developing AMD, was down-regulated in the supplemented retinas. Our fluorescence immunostaining confirmed protein down-regulation, along with a reduction in complement cleavage product C3d. In addition, when comparing the transcriptomic changes between 24:5n-3-supplemented retinas and young retinas, we observed that supplementation partially restored a more youthful retinal transcriptomic profile. Together, our transcriptomic and immunostaining data suggest that the 24:5n-3 treatment induces a shift in the status of the retina toward a younger phenotype.

Previous studies have reported associations between variants of ELOVL2 and abundance of LC-PUFAs in plasma and colostrum (fig. S7). However, no ELOVL2 mutations or variants have been implicated in the risk of AMD. We hypothesize several possible explanations (27): (i) ELOVL2 is an essential gene for population survival. It has been shown that heterozygosity of the gene in a mouse model causes infertility in C57BL/6 mice (47). Therefore, variants that can be potentially correlated with the disease are rare and have yet to be found. (ii) The genetics of AMD is primarily composed of variants influencing complement activation with comparatively large effect sizes. Because those variants explain more than half of the genetics of AMD, a minor dysregulation of LC-PUFAs because of genetic mutations might not be a driving factor in disease risk. (iii) The variants in ELOVL2 only change the risk of AMD on specific genetic backgrounds or disease stages and progression. To address the latter possibility, we used genomic data from two large cohorts to investigate the modulation of AMD onset due to ELOVL2 genetics. Although the associated variants were not found to influence ELOVL2 gene expression in the retina or brain, the G allele of the lead variant rs911196 increases methylation of cg01799681, cg16867657, cg21572722, and cg23642061 in whole blood (62, 63), which also show increased methylation with advanced age (64). Last, the AREDS studies have indicated that higher intake of fish protects against progression of intermediate AMD, reinforcing the importance of our finding. To our knowledge, these data represent the first genetic evidence of the involvement of ELOVL2 in AMD. We also note that the A allele of the previously identified variant rs953413 [which reduces DHA levels in plasma (65–67)] is on the same haplotype as the G allele of rs911196 (D′ = 1.0). Thus, individuals with the DHA-reducing allele also carry the allele associated with earlier AMD onset. We were not able to reliably impute rs911196 in the IAMDGC dataset; thus, we could not further investigate this variant. In sum, we anticipate that future analyses will reveal genetic correlations between ELOVL2 and other age-related diseases of the central nervous system.

We acknowledge several limitations of our study. First, this proof-of-concept work used intravitreal injection, which is a less convenient route of administration for preventing aging of the eye in patients. Future animal studies with labeled 24:5n-3 will explore alternative administration routes, such as intraperitoneal injection, eye drops, or oral supplementation, to assess the effectiveness of the treatment. In addition, the duration of treatment efficacy will need to be optimized. Furthermore, although we demonstrate that 24:5n-3 contributes to changes in the abundance of VLC-containing phospholipids, we take into consideration that it may exert protective effects through additional mechanisms including a potential role of ELOVL2 in RPE. Lower concentrations of PUFAs in the retina can also drive many metabolic changes including photoreceptor energetic status, glucose metabolism, and mitochondrial function. These aspects require further experimental analysis. Last, although the intermediate AMD definition of the IAMDGC dataset is based on fundus imaging and thus includes individuals with pigmentary changes in the RPE and age at first diagnosis ≥ 50 years, the AMD phenotype ascertained in the UK Biobank cannot be stratified because of the lack of retinal imaging available in the cohort. Furthermore, the early age of onset for UK Biobank cases (mean age of onset of around 63 years) suggests that these individuals are likely in the early stages of the disease rather than the advanced stages. This is further highlighted in the observation that the observed effect sizes are similar between IAMDGC and UK Biobank cohorts. Therefore, independent replication is necessary to test whether the observed associations hold up in additional cohorts, preferably also with different AMD definitions and varying ethnic backgrounds.

At the biological level, some of our findings also warrant further investigation. For instance, although we observed high expression of ELOVL2 in cone photoreceptors, supplementation improved rod-mediated dark adaptation recovery. This is notable because rod-mediated dark adaptation is one of the earliest and most persistent functional impairments associated with aging and AMD (68). We propose two key possibilities: (i) It remains possible that cone-mediated dark adaptation is also an early indicator of aging, although studies in both animal models and human participants are currently lacking. (ii) Mechanistically, we hypothesize that cones are a major source of endogenous 24:5n-3 in the retina, which supports the renewal and function of both cone and rod outer segments. In aging, decreased abundance of cone-derived 24:5n-3—needed for endogenous synthesis of DHA and VLC-PUFAs—combined with the accumulation of sub-RPE deposits that impair PUFA delivery from circulation, may impose metabolic stress on rods, which do not produce sufficient amounts of 24:5n-3 on their own. This metabolic vulnerability could help explain why rods are the first photoreceptors to lose function with age and the first to degenerate in AMD.

In sum, our work demonstrates that targeted delivery of a product of ELOVL2 to the eye can reverse age-related molecular changes in the retina and associated visual function impairments. We have also found a genetic connection between the ELOVL2 locus and the onset of intermediate AMD in human patients. Our work is paving the way toward developing lipid-based therapeutic strategies aimed at preserving visual health in aging populations.

MATERIALS AND METHODS

Study design

This study was designed to investigate the role of retinal LC-PUFAs in age-related visual decline and to evaluate the therapeutic potential of 24:5n-3 supplementation in aging retinas. The experimental approach combined genetic, molecular, and functional analyses. For therapeutic intervention, short- and long-term intravitreal injections of 24:5n-3 were administered to aged wild-type male C57BL6/J mice. Sample sizes were based on prior experimental variance or preliminary data, and the minimum number of animals required for statistical power was used. For supplementation studies, the right eye of each mouse received 24:5n-3, whereas the left eye served as a contralateral vehicle-injected control. Mice with corneal or lens opacities were excluded from ERG or OMR measurements, where clear ocular media are essential for obtaining accurate results, but not from molecular or cellular analyses. Each experiment was independently repeated at least twice, and sample sizes and biological replicates are indicated in the figure legends. The University of California, Irvine Institutional Animal Care and Use Committee approved all mouse experiments under protocol no. AUP-23–064. The study complies with the ARRIVE 2.0 reporting guidelines for in vivo animal research.

Statistical analysis

Individual-level data are provided in data file S2. Statistical analyses were performed using GraphPad Prism 10. Data are presented as mean ± SEM. Differences between two groups were analyzed using Student’s unpaired t test, whereas comparisons among three or more groups were conducted using one-way analysis of variance (ANOVA). The Shapiro-Wilk test was used to assess normality, and results confirmed that the data followed a normal distribution; therefore, parametric tests such as the t test and ANOVA were applied. All statistical tests used are specified in the figure legends. Statistical significance was assessed using two-tailed tests, with thresholds defined as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Supplementary Material

The PDF file includes:

Materials and Methods

Figs. S1 to S7

Tables S1 to S4

References (69–85)

Other Supplementary Material for this manuscript includes the following:

Data files S1 to S3

MDAR Reproducibility Checklist

Data and materials availability:

All data associated with this study are present in the paper or the Supplementary Materials. Bulk RNA-seq data generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE303302. snRNA-seq data generated in this study have been deposited under accession number GSE303793. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD066430. The MS lipidomic data are included in data file S3. Elovl2^C234W animals are available to interested researchers through a material transfer agreement with University of California, Irvine.