Remodeling of Retinal Fatty Acids in an Animal Model of Diabetes: A Decrease in Long-Chain Polyunsaturated Fatty Acids Is Associated With a Decrease in Fatty Acid Elongases Elovl2 and Elovl4

High-performance liquid chromatography (HPLC)-grade acetonitrile, acetic acid, methanol, chloroform, streptozotocin (STZ), and commonly used chemicals and reagents were from Sigma-Aldrich Chemical (St. Louis, MO).

Male Sprague-Dawley rats weighing 237–283 g were made diabetic with a single intraperitoneal injection of 65 mg STZ per kg body wt. Body weight gains and blood glucose for the control and STZ-induced diabetic groups were monitored biweekly. At 3–6 weeks after STZ injection, the animals were killed and blood plasma, liver, and retina were recovered for analyses of fatty acid profiles and/or fatty acid elongase and desaturase expression levels. To isolate the retina, the optic nerve was cut out; the eye was opened; the coronary, cornea, and lens were discarded; and the retina was separated from choroid, washed in PBS, and frozen. Rats were maintained on Harlan-Teklad laboratory diet (no. 8,640) and water ad libitum. The fatty acid composition of the diet was analyzed by reverse-phase HPLC (RP-HPLC, see below) and found to be 16:0, 20.0%; 18:0, 1.8%; 18:1n9, 21.9%; 18:2n6, 50.8%; 18:3n3, 5.6%; 18:3n6, 0.1%; 20:4n6, 0.1%; 20:5n3, 0.4%; 22:5n3, 0.1%; and 22:6n3, 0.3%. Animal protocol was approved by the Michigan State University Institutional Animal Care and Use Committee. All experiments followed the guidelines set forth by the Association for Research in Vision and Ophthalmology Resolution on Treatment of Animals in Research.

Rat tissues were homogenized in Trizol reagent (Invitrogen), and RNA was isolated according to manufacturer instructions. After adding chloroform, the upper aqueous phase was separated and RNA was precipitated with isopropyl alcohol, washed with 75% ethanol, and redissolved in RNase-free water. Proteins from the same samples were isolated by washing in Tris buffer (30 mmol/l Tris-HCl, pH 6.8, with 0.1% SDS) followed by concentration on Amicon-15 (Millipore) centrifugal filters. After concentration, protease/phosphatase inhibitor cocktail was added and samples were frozen until further analysis.

Transcript-specific primers were designed using Primer3 software (available at http://frodo.wi.mit.edu/primer3/). First-strand cDNA was synthesized using the SuperScript II RNase H Reverse Transcriptase (Invitrogen, Carlsband, CA), and PCRs were performed in triplicate as previously described (25). Transcripts of interest were normalized to the abundance of cyclophilin mRNA. Rat gene–specific primers used in this study were Elovl4: GAAGTGGATGAAAGACCGAGA (sense) and GCGTTGTATGATCCCATGAA (antisense); Elovl7: TGGCGTTCAGCGATCTTAC and GATGATGGTTTGTGGCAGAG; IL-6: CCAGGAAATTTGCCTATTGA and GCTCTGAATGACTCTGGCTTT; VEGF A: GCTCTCTTGGGTGCACTGG and CACCACTTCATGGGCTTTCT; and ICAM-1: CCACCATCACTGTGTATTCGTT and ACGGAGCAGCACTACTGAGA. All other primers for rat elongases and desaturases were described previously (25).

Protein concentration was determined by a Qubit fluorometer (Invitrogen), according to manufacturer's instructions, and equivalent amounts of protein were loaded on the SDS-polyacrylamide (10%) minigels for SDS-PAGE separation. The separated proteins were electrophoretically transferred to a nitrocellulose membrane (Bio-Rad, Hercules, CA), blocked for 30 min at room temperature, and probed with primary rabbit anti-Elovl4 (Abcam) and mouse anti-α tubulin (Sigma) antibody followed by fluorescent secondary antibody (Invitrogen). The blots were analyzed by the Licor Odyssey scanner and quantitated using Licor Odyssey software.

Total lipids from retina, blood plasma, and liver were extracted with chloroform-methanol (2:1, vol/vol), normalized to tissue weight, dried, and resuspended as previously described (29). Acidification of total lipid extracts was omitted, as low pH has been demonstrated to destroy acid-labile plasmalogen lipids (30), which are abundant in neural tissues. No significant decrease of recovery of abundant lipid classes was observed in the absence of pH modification (29). Blood plasma lipids were normalized to protein, as measured by Qubit assay. Erythrocyte total lipids were extracted from 100 mg of packed cells by a modified Rose and Oklander method (31). Briefly, lysed cells were combined with 6.8 ml of 80% 2-propanol, vortexed, and incubated for 1 h on ice with occasional mixing. Lipids were extracted twice by addition of 3.2 ml of 100% chloroform and 1-h incubation on ice, with phase separation after each chloroform addition achieved by centrifugation at 2,000g for 30 min. Acidification of the extraction mixture was omitted, as it was found that low pH destroys acid-labile plasmalogen lipids (30), which are abundant in neural tissues. No significant decrease of recovery of abundant lipid classes was observed in the absence of pH modification (29). Pooled lipid extracts were dried, resuspended, and stored as described above.

Lipid extracts were introduced to a triple-quadrupole mass spectrometer (Thermo Scientific model TSQ Quantum Ultra, San Jose, CA) for nano-electrospray ionization tandem mass spectrometry (nESI-MS/MS) analysis of lipid species as previously described (29). Identification of phospholipid species by precursor ion and neutral loss scan mode MS/MS was performed according to published methods (29,32). Assignment of phosphatidylcholine (GPCho) acyl substituents was achieved by negative ion mode analysis of corresponding GPCho [M+Cl]⁻ ions byproduct ion scan mode MS/MS, as well as by precursor ion scanning for m/z corresponding to specific deprotonated fatty acyl ions. Peak finding and correction for ¹³C isotope effects was performed using the Lipid Mass Spectrum Analysis (LIMSA) software version 1.0 peak model fit algorithm (33). Quantitative analysis of the relative changes in GPCho lipid abundances between control and diabetic samples was achieved by normalization of the peak area of each detected GPCho m/z to that of the GPCho_(16:0/16:0) lipid present in each of the samples, after correction for ¹³C isotope contributions.

An aliquot of total lipids from tissues and blood fractions was saponified (0.4 N KOH in 80% methanol, 50°C for 1 h). Saponified fatty acids were acidified and extracted with diethyl ether (according to Wang et al. [25]) and stored in methanol containing 1 mmol/l butylated hydroxytoluene. Saponified free fatty acids were fractionated and quantitated by RP-HPLC using a YMC J-Sphere (ODS-H80) column and a sigmoidal gradient starting at 86.5% acetonitrile + acetic acid (0.1%) and ending at 100% acetonitrile + acetic acid (0.1%) over 50 min with a flow rate of 1.0 ml/min using a Waters 600 controller. Fatty acids were introduced to the HPLC by injection in methanol and detected using ultraviolet absorbance and evaporative light scatter as previously described (25). Authentic fatty acid standards (Nu-Chek Prep) were used to generate calibration curves for verification and quantification of fatty acids.

Data are expressed as the means ± SD. Student's t test was used for comparing data obtained from independent samples. Significance was established at P < 0.05.

As presented in Table 1, body weight gain was significantly slower in diabetic animals compared with control animals. Blood glucose levels were almost five times higher in diabetic animals compared with controls. As this was a short-term diabetes study, A1C levels were not measured.

The gene expression levels of elongases and desaturases in control retinas were determined by quantitative RT-PCR and compared with the levels found in the livers of the same animals. Retinal-specific elongase, Elovl4, had the highest expression level among all the elongases in the retina and was not expressed in the liver. Retinas also had high levels of Elovl2 expression (Fig. 2,A). Livers exhibited higher levels of Elovl5 and Δ5-, Δ6-, and Δ9-desaturases than retina (Fig. 2 A), and the liver profile of all elongases and desaturases agreed with our previous study (25).

In the liver, diabetes induced a 25% decrease in Elovl2 and a 33% decrease in Elovl6 expression, as well as an 85% decrease in Δ9-desaturase (Fig. 2,B) compared with controls. In the retina, diabetes induced a 40% reduction in Elovl4 and 50% reduction in Elovl2 expression levels (Fig. 2,C). There was no significant effect of diabetes on the retinal desaturases (Fig. 2,C). A decrease in Elovl4 protein level was confirmed by Western blot, as shown in Fig. 2 D.

The control and diabetic blood plasma fatty acid profiles 3 weeks after STZ injection are presented in Table 2. There was a tendency toward higher total fatty acids level in diabetic versus control blood plasma. We observed changes in the plasma fatty acid profile consistent with inhibition of fatty acid remodeling in diabetes that leads to a lower end product–to–precursor fatty acid ratio. There was a decrease in two major end products of the PUFA synthesis pathway, arachidonic_20:4n6 acid and DHA_22:6n3, relative to their precursors, linoleic_18:2n6 and α-linolenic_18:3n3 acid, respectively (Table 2). As a result of these changes, we observed a decrease in unsaturation index (the number of double bonds per fatty acyl residue) and a decrease in long-chain–to–short-chain PUFA ratio in diabetic versus control animals (Table 2).

The control and diabetic liver fatty acid profiles 3 weeks after STZ injection are presented in Table 3. There was an increase in the linoleic acid_18:2n6 level in the livers of diabetic versus control animals that led to a decrease in long-chain–to–short-chain PUFA ratio (Table 3). There were no other significant changes in liver fatty acid profiles in diabetic versus control animals. The liver unsaturation index and PUFA synthesis pathway end product–to–precursor ratios did not change in diabetic versus control animals (Table 3).

Retina has a unique fatty acid profile, with the highest content of long-chain PUFAs in the body. In agreement with other studies, retinal profiles were very rich in DHA_22:6n3 and arachidonic_20:4n6 acid (Table 4). The levels of linoleic_18:2n6 and α-linolenic_18:3n3 acid in the retina were very low; thus we did not calculate the PUFA synthesis pathway end product–to–precursor ratios. Importantly, the retinas of diabetic animals had 28% less DHA_22:6n3 compared with controls. As a result, we observed a decrease in unsaturation index, a decrease in long-chain–to–short-chain PUFA ratio, and a decrease in the n3-to-n6 PUFA ratio in the retinas of diabetic versus control animals (Table 4). Representative RP-HPLC chromatograms of saponified fatty acids from control and diabetic retina are presented in Fig. 3 A.

In agreement with saponified fatty acid profile data, nESI-MS/MS analysis of the retina lipid extracts of diabetic animals (n = 3) showed a significant (up to 34%) decrease in the abundance of glycerophospholipids containing DHA_22:6n3 compared with the control animals (n = 3). For example, compare the abundance of the GPCho_(18:0/22:6) and GPCho_(22:6/22:6) lipids in the ratiometric analysis shown in Fig. 3,B. Similar decreases in the abundances of DHA_22:6n3 containing lipids were also observed for glycerophosphoethanolamine and glycerophosphoserine lipids (data not shown). In contrast, an increase (37%) in the abundance of the linoleic_18:2n6 acid–containing GPCho_(16:0/18:2) lipid was observed for the same samples as shown in Fig. 3 B. In addition to known fatty acids identified by HPLC analysis, nESI-MS/MS analysis revealed several VLCPUFAs, primarily 32:6n3 and 34:6n3, esterified to GPCho in the retina. Interestingly, GPCho_(32:6/22:6) was significantly decreased (24%) in diabetic retinas compared with controls, and there was a nonsignificant decrease (9%) of GPCho_(34:6/22:6).

In erythrocyte lipid extracts, an increase in the abundance of linoleic_18:2n6 acid–containing lipids, namely GPCho_(16:0/18:2) and GPCho_(18:0/18:2), was observed between the diabetic and control samples, consistent with the changes in retina lipids, as shown in Fig. 3 C. Erythrocytes had very low levels of DHA-containing phospholipid species, and there was no effect of diabetes on these species. There was no detectable GPCho_(32:6/22:6) or GPCho_(34:6/22:6) in the erythrocytes or in liver and blood plasma (data not shown).

As n3 PUFAs are known to have anti-inflammatory properties, we hypothesized that a decrease in n3 PUFAs would be associated with a proinflammatory state in diabetic retinas. As shown in Fig. 4, diabetic retinas had increased expression levels of several inflammatory markers including adhesion molecules (ICAM-1), cytokines (IL-6), and growth factors (VEGF).

The association of dyslipidemia with the development of diabetic retinopathy has been underscored by the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications cohort study (8). Despite this evidence, the experimental data on diabetes-induced changes in lipid profile and lipid metabolism in the retina are not available. This is the first comprehensive study to analyze retinal-specific fatty acid profiles and metabolism and to compare them to liver and blood plasma in control and diabetic animals.

In this study utilizing STZ-induced diabetic rats, we found a decreased level of DHA_22:6n3, the major retinal long-chain PUFA, in diabetic retina. This finding confirmed earlier studies showing a decrease in relative percentage of DHA_22:6n3 in the diabetic retina (34,35). In addition to DHA_22:6n3, VLCPUFAs including 32:6n3 and 34:6n3 were detected as substituents of retina GPCho. VLCPUFAs were not detected in lipid classes other than GPCho and were only detected in retina. Three weeks of diabetes reduced retinal levels of 32:6n3-GPCho compared with controls. As a result of these changes, the diabetic retina had a lower unsaturation index and lower long-chain–to–short-chain PUFA ratio. Moreover, there was a shift toward n3 PUFA–deficient, n6 PUFA–rich, profile in the diabetic retina.

In general, n6 PUFAs induce, while n3 PUFAs inhibit, inflammation, and the relative amount of these PUFAs plays an important role in the regulation of immunity (36). Our previous studies indicated that treatment of a cell type affected by diabetic retinopathy, HRECs, with n6 PUFA leads to a lipoxygenase-dependent increase in ICAM-1/vascular cell adhesion molecule-1 expression (37). Conversely, we have demonstrated that DHA_22:6n3 inhibited cytokine-induced activation of the NFκB signaling pathway and adhesion molecule expression in HRECs (28). Thus, a decrease in the n3-to-n6 PUFA ratio in the diabetic retina observed in this study would create proinflammatory conditions potentially contributing to the development of diabetic retinopathy. Indeed, previous studies demonstrated an upregulation in a number of inflammatory markers in the retina early in diabetes: VEGF (4,5), ICAM-1 (6,7), TNF-α (8), and IL-6 (9). ICAM-mediated leukostasis was detected within 1 week of diabetes in rats (38,39). VEGF was shown to increase ICAM expression in retinas of nondiabetic mice (40), and vitreal VEGF levels were found to be correlated with that of IL-6 and severity of diabetic retinopathy in diabetic patients (41).

In this study, we chose a cytokine (IL-6), a growth factor (VEGF), and an adhesion molecule (ICAM-1) as readout of an inflammatory status in the retinas of diabetic animals with decreased n3/n6 PUFAs. mRNA levels of all three markers were elevated in diabetic retinas compared with controls.

Importantly, diabetes induced the most pronounced changes in the retinal fatty acid profile, whereas liver fatty acid profile was only slightly affected, indicating that the disruption of retinal fatty acid metabolism in diabetes might not simply be a result of altered liver metabolism. Moreover, VLCPUFA-containing phospholipids detected in the retina were not present in the liver or erythrocyte total lipids. The fatty acid profile in a particular peripheral tissue depends on two factors: 1) the profile in circulation due to the diet and liver metabolism and 2) the ability of a local tissue to remodel fatty acids. Retina has a unique fatty acid profile characterized by one of the highest levels of DHA_22:6n3 in the body and by the presence of VLCPUFAs (27,42). While the expression level of retinal desaturases was relatively low compared with retinal elongases, it has been reported that retina can synthesize DHA_22:6n3 from α-linolenic_18:3n3 acid and EPA_20:5n3 (43). Although retina may obtain additional DHA_22:6n3 by uptake from the circulation, changes in the retinal fatty acid profiles of diabetic animals did not mirror changes observed in liver and plasma fatty acid profiles. Thus, a retina-specific decrease in DHA_22:6n3 in diabetes is likely to be due to changes in retinal fatty acid metabolism.

To determine the effect of diabetes on retinal fatty acid metabolism, we analyzed the level of fatty acid elongase and desaturase gene expression in control and diabetic animals. Retinas had a very high expression level of the retinal-specific elongase, Elovl4, as well as high expression levels of long-chain PUFA elongase Elovl2. Δ5-, Δ6-, and Δ9-desaturase levels were low compared with the liver expression levels. The high levels of Elovl4 and Elovl2 and low levels of desaturases suggest that the retina is preferentially involved in production of very-long-chain fatty acids and exhibits a low level of de novo lipogenesis. The retinal elongase expression profile that we observed likely explains the high level of long-chain PUFAs in the retina compared with liver and blood plasma levels. Elovl2 elongates C_20–22 fatty acids (44,–46). Elovl4 was recently shown to be involved in VLCPUFA synthesis with substrate specificity for C_26–36 fatty acids (15). The role of VLCPUFAs is not known. Because of their specific presence in tissues with high membrane curvature and their ability to span both leaflets of the lipid bilayer, VLCPUFAs are suggested to play the role of an anchor stabilizing high curvature cellular membranes (15). In the retina, VLCPUFAs are mainly present in the rod outer-segment membrane (15), where they are suggested to play a role in stabilizing the rims of photoreceptor disks. This specific localization might explain low abundance of VLCPUFAs in the total retinal lipids extracted in this study. At the same time, specific localization suggests that VLCPUFAs might play an important role in photoreceptor function. This study provides the first direct evidence that a significant decrease in Elovl4 in diabetic retina is indeed associated with a decrease in VLCPUFA (i.e., 32:6n3) synthesis. Despite lower abundance, diabetes-induced decrease in 32:6n3 containing GPCho (24%) was similar to the decrease in DHA_22:6n3 containing GPChos (15–34%). Elovl4 protein expression in diabetic retina was inhibited to a higher degree (73%) compared with mRNA expression (40%), suggesting control of Elovl4 expression at both transcriptional and translational levels. Although decrease of VLCPUFAs is most likely to arise from Elovl4 loss, another plausible explanation could be that this reduction was due to reduction in VLCPUFA precursor lipids, EPA_20:5n3, and/or DHA_22:6n3. This possibility can be tested in the future by determining whether downregulation of VLCPUFAs in diabetes persists in animals supplemented with high-EPA_20:5n3/DHA_22:6n3 diet.

Another possibility could be that high level of reactive oxygen species in diabetic retina leads to degradation of a highly oxidation-prone DHA molecule. Previous studies using the same STZ-induced diabetic model of similar duration, however, did not find oxidized DHA products in diabetic retina (47).

Several Elovl4 gene mutations have been recently identified in pathogenesis of another retinal disease, Stargardt-like macular dystrophy (19,–21). Stargardt-like macular dystrophy is an autosomal-dominant disorder due to a dominant-negative effect of the mutated Elovl4 on wild-type protein (19). As Elovl4 is highly expressed in the photoreceptors (19,21), it is not surprising that mutant Elovl4 transgenic mice are characterized by lipofuscin accumulation, abnormal electrophysiology, and photoreceptor degeneration (20). Although photoreceptors are not the primary site of diabetic retinopathy, several abnormalities in neural retina have been associated with the development of diabetic retinopathy (48,49). The decrease in Elovl4 observed in this study would not be expected to have as dramatic an effect on photoreceptor viability as the dominant-negative mutation in Elovl4. However, the reduction in Elovl4 in diabetic retina could be responsible for more subtle changes in photoreceptor/RPE cell function that could lead to metabolic changes in the whole retina and eventually contribute to the pathology characteristic of diabetic retinopathy.

In conclusion, a decrease in the expression level of retinal fatty acid elongases Elovl2 and Elovl4 and concomitant decrease in the major n3 PUFA, DHA_22:6n3, as well as the VLCPUFA_32:6n3, results in an increased n6-to-n3 PUFA ratio in the diabetic retina that likely creates a proinflammatory state contributing to the development of diabetic retinopathy. Increasing the gene expression of fatty acid elongases in the retina represents a potential therapeutic strategy for modulating fatty acid metabolism and altering the pathogenesis of diabetic retinopathy.

This work was supported by grants from the Juvenile Diabetes Research Foundation (2-2005-97 [to J.V.B.]), the National Institutes of Health (EY-016077 [to J.V.B.], R01RR025386 [to G.E.R. and J.V.B.], and DK 43220 [to D.B.J.]), and MAES MICL02163 (to J.V.B.).

No potential conflicts of interest relevant to this article were reported.