Associations of human retinal very long-chain polyunsaturated fatty acids with dietary lipid biomarkers

The human retina is well-known to have unique lipid profiles enriched in long-chain polyunsaturated fatty acids (LC-PUFAs) and very long-chain polyunsaturated fatty acids (VLC-PUFAs) that appear to promote normal retinal structure and function, but the influence of diet on retinal lipid profiles in health and disease remains controversial. In this study, we examined two independent cohorts of donor eyes and related their retinal lipid profiles with systemic biomarkers of lipid intake. We found that serum and red blood cell lipids, and to a lesser extent orbital fat, are indeed excellent biomarkers of retinal lipid content and n-3/n-6 ratios in both the LC-PUFA and VLC-PUFA series. Eyes from age-related macular degeneration (AMD) donors have significantly decreased levels of VLC-PUFAs and low n-3/n-6 ratios. These results are consistent with the protective role of dietary n-3 LC-PUFAs against AMD and emphasize the importance of monitoring systemic biomarkers of lipid intake when undertaking clinical trials of lipid supplements for prevention and treatment of retinal disease.

suggested that a retinal defi ciency of LC-PUFAs and VLC-PUFAs infl uenced by diet and/or a dietary imbalance of n-3/n-6 LC-PUFA ratios may be involved in AMD pathology, but our study was limited by the low number of AMD eyes (n = 8) and lack of availability of biomarkers of dietary lipid intake. We now report follow-up studies showing that dietary intake of LC-PUFA and VLC-PUFA precursors clearly infl uence retinal lipid profi les in retina, and they confi rm that AMD eyes have numerous abnormalities in these profi les.

Sample collection
The studies reported here were conducted on two distinct collections of human donor eye tissues. All experimental procedures including tissue procurement were conducted according to the tenets of the Declaration of Helsinki. In the fi rst-phase study, human donor eyes with no history of eye disease were obtained from the Utah Lions Eye Bank. The time between donor death and enucleation was <4 h. Dissections of donor eyes were the action of the elongation of very long-chain fatty acids elongase 4 (ELOVL4) enzyme ( Fig. 1 ) ( 20 ). Autosomal dominant mutations in this enzyme lead to a rare form of Stargardt macular dystrophy (STGD), STGD3 ( 21,22 ), but the mechanisms underlying this degeneration (VLC-PUFA defi ciency vs. mutant protein mislocalization) remain controversial (23)(24)(25). Our study of dietary biomarkers of lipid consumption has shown that members of a Utah family with STGD3 who consume large amounts of fi sh have a milder phenotype than those who rarely consume fi sh ( 26 ), and an open-label clinical intervention trial with fi sh oil supplementation is in progress for this family (ClinicalTrials.gov, #NCT00420602).
In order to clarify the roles of LC-PUFAs and VLC-PU-FAs in retinal health and disease, we previously undertook a study of human donor eyes to determine whether abnormalities in lipid profi les are present in AMD eyes. We found that DHA and many VLC-PUFA levels are significantly lower in AMD eyes relative to age-matched controls and that n-3/n-6 ratios are signifi cantly lower for both LC-PUFAs and VLC-PUFAs ( 27 ). Because ELOVL4 genetic variants are not associated with AMD risk ( 28 ), these fi ndings

Lipid extraction and purifi cation of lipids with solid-phase extraction
Serum, RBCs, orbital adipose tissue, and retina punches were extracted using the procedure previously adopted in our laboratory ( 30 ). The samples and internal standards (50 µg of tridecanoic acid and 1.15 µg of hentriacontanoic acid) were added in 2 ml stainless steel vials and then homogenized with 0.7 ml silica carried out 6-24 h after donor death under dim light in cold temperature (4°C) to minimize lipid oxidation and autolysis. Eyes with large drusen, severe macular atrophy, macular hemorrhage, or any grossly visible chorioretinal pathologic abnormalities were excluded. After the cornea, iris, ciliary body, and lens were removed, a 4 mm peripheral retina (PR) region in the nasal mid-PR was punched with a trephine because previous studies from our laboratory have already indicated that lipid profi les of PR do not differ substantially from lipid profi les of macular tissue ( 29,30 ). All punched human ocular tissues were stored in tubes fi lled with argon gas and kept at Ϫ 80°C. Serum, red blood cells (RBCs), and orbital fat samples were also collected from these donors and stored at Ϫ 80°C.
The second-phase study utilized samples from the Utah Center for Translational Medicine Donor Eye Repository, a large collection of rigorously characterized ocular tissues from AMD and control donors collected in a manner similar to the collection described above. Demographic information of the subjects and their AMD grades using a modifi ed Rotterdam scale ( 31 ) are provided in Table 1 . Midperipheral 6 mm retinal punches were analyzed from age-matched AMD and control eyes along with serum samples. RBCs and orbital fat were not available from these donors. method but with a larger injection volume of 5 µl from a 20 µl sample that had been concentrated from a 200 µl original volume. The column temperature program was similar to the LC-PUFA method but held at 290°C for 35 min. MS conditions were similar to the LC-PUFA method, but the detector delay time was 20 min.
The low amounts of VLC-PUFAs that are present in the mammalian retina elute very late from the GC/MS, and standards are not available commercially, which means that their quantitation can be particularly challenging. To achieve this, two separate GC/MS runs linked by common C-24 PUFAs were necessary. With method A, the complete set of long-chain fatty acids (LC-FAs) up to 22 carbons in length and two C-24 FAs (24:1n-9 and 24:0) can be quantifi ed under full scan mode, and when the chromatogram is reanalyzed under selective ion mode ( m/z 79, 108, and 150), we can identify and quantify all n-3 and n-6 LC-PUFAs, and even the C-24 VLC-PUFAs become detectable and can be quantifi ed by comparing their mole percentages relative to the C-22 PUFAs. The C-24 VLC-PUFAs can then be used as the common link between method A and method B because they are present in both GC/MS chromatograms. All the VLC-PUFAs could be subsequently quantifi ed relative to the total LC-FAs determined by method A after correcting for the effects of carbon chain length and the degree of unsaturation on the response of the mass spectrometer ( 30 ).

Genotyping
Genotyping was done using the TaqMan platform (Applied Biosystems, Grand Island, NY). Amplifi cation and genotype assignments were conducted using the 7900HT and SDS 2.4 software. The SNPs, rs381253 and rs10753929, used in this study were selected based on their association with ELOVL4 and adiponectin receptor 1 ( AdipoR1 ) as reported in earlier literature ( 32,33 ).

Statistical analyses
Statistical analyses were performed using ANOVA, linear regressions, Chi-square tests, and t -tests on Prism software (Graph-Pad Software Inc., La Jolla, CA). Data are represented as the mean ± SD. Signifi cance is indicated by P value measurements, with P < 0.05 considered signifi cant.

Phase I study
Forty-four normal patient samples were collected (average age ± SD = 71.4 ± 13.4 years), and their serum, RBCs, orbital fat, and retinal punches were analyzed using GC/ MS techniques. We studied the correlations between the retinal lipid profi le and those of serum, RBCs, and fat, which are validated biomarkers of short-term (weeks), medium-term (months), and long-term (years) dietary fat intake ( 26,(34)(35)(36) ( Table 2 ). Individual key LC-PUFAs (EPA, DHA, and AA) in serum, RBCs, and fat generally showed signifi cant positive correlations with corresponding levels in retinal lipids. We observed strong positive correlations between EPA levels in serum, RBCs, and fat with EPA levels in retina (all P < 0.001), while retinal DHA levels correlate relatively weakly only with DHA levels of RBCs ( P < 0.05). The levels of the major n-6 fatty acid in retina, AA, were also in positive correlation with AA levels in beads and 1 ml hexane-isopropanol (3:2 v/v) by a Mini Bead Beater-16 (BioSpec Products Inc., Bartlesville, OK) and a Sonic Dismembrator Model 100 (Fisher Scientifi c). The homogenized samples were bath sonicated for 5 min in an ice water bath. After centrifugation at 10,000 rpm for 5 min, the extracted solution supernatant was transferred to a clean vial and then dried under a stream of nitrogen. The dried fi lm was dissolved in 200 µl hexane, and 2 ml of 4% HCl in methanol was added. The tubes were fl ushed with argon and incubated at 80°C for 4 h to form FAMEs ( 30 ) and then allowed to cool. The FAME mixture was extracted three times with 1 ml distilled water and 2 ml hexane. The hexane layers were combined and dried under nitrogen gas.
Silica gel, glass-encased, solid-phase extraction cartridges were subsequently used to clean the FAME extracts. The cartridge was activated with 6 ml of hexane before loading samples. The crude FAME extract was dissolved in 200 µl of hexane and loaded onto the activated cartridge. The cartridge was washed with 6 ml hexane, and the eluate was discarded. FAMEs were eluted from the cartridge with 5 ml hexane-ether (4:1), and the eluate was evaporated under nitrogen gas. The dry fi lm was dissolved in 200 µl of hexane and centrifuged for 3 min at 14,000 rpm to remove particles prior to GC/MS analysis. One microliter of sample was injected into the GC/MS instrument for LC-PUFA analysis. For VLC-PUFA analysis, the sample was dried with nitrogen again and redissolved in 20 l of n-nonane, and 5 µl samples were injected into the GC/MS instrument.

GC/MS instrumentation and chromatographic conditions
The Thermo Trace GC-DSQ II system (ThermoFisher Scientifi c, Waltham, MA) consists of an automatic sample injector (AS 3000), gas chromatograph, single quadrupole mass detector, and an analytical workstation. The chromatographic separation was carried out with an Rxi-5MS-coated 5% diphenyl/95% dimethyl polysiloxane capillary column (30 m × 0.25 mm inner diameter , 0.25 µm fi lm thickness) (Restek, Bellefonte, PA). Two methods (A and B) were used for detection and quantitation of LC-PUFAs and VLC-PUFAs, respectively.
For LC-PUFA analyses, the following MS conditions were used (method A): 1 µl from a 200 µl sample was injected into the GC/ MS using a splitless mode, the septum purge was on, and the injector temperature was set at 200°C. The column temperature was programmed as follows: initial temperature 60°C, 5 degrees/ min to 170°C, 1 degree/min to 180°C, 2 degrees/min to 240°C, 4 degrees/min to 290°C, and a hold at 290°C for 5 min. Transfer line temperature was 290°C. Helium was used as the carrier gas at a fl ow rate of 1.0 ml/min. MS conditions were as follows: electron ionization mode with full scan and selected ion monitoring (SIM, m/z 79, 108, and 150) because m/z 79, 108, and 150 are typical ions of PUFAs, and n-3 and n-6 PUFAs can be distinguished by comparing the ratio of ions of m/z 108 and 150; ion source temperature, 200°C; multiplier voltage, 1,182 V; and detector delay, 10 min. For peak identifi cation, the data were obtained by collecting the full-scan mass spectra within the scan range of 50-650 amu, and these peaks were identifi ed by comparing their mass spectra with those in the standard solution and the National Institute of Standards and Technology library. For the quantifi cation of LC-PUFAs, the data were obtained by SIM. Authentic reference compounds were used to calculate the mol percentage of each peak.
VLC-PUFAs analyses (method B) were conducted as described previously ( 30 ). Bovine retina VLC-PUFAs were extracted and used as VLC-PUFA standards to establish retention times because commercial standards are not available, and identifi cation of each VLC-PUFA in retinal samples was achieved as described in prior work from our laboratory ( 27 ). For the quantifi cation of VLC-PUFAs, we used MS conditions similar to the LC-PUFA VLC-PUFAs, and their precursors ( Table 4 and Fig. 2A , B ). We also noted that systemic EPA/AA ratios correlated well with retina EPA/AA ratios ( P < 0.001 for serum, RBCs, and fat), but DHA/AA ratios did not correlate signifi cantly. In fact, the simple EPA/AA ratios in serum, RBCs, or fat were highly predictive biomarkers of n-3/n-6 ratios of retinal VLC-PUFAs ( Table 4 and Fig. 2C ).
When data from individual participants were examined, one of the 44 subjects was noted to be an outlier with unusually high levels of EPA and DHA in serum, RBCs, and fat ( Fig. 3 ). Likewise, this donor's n-3/n-6 ratios were much higher than the others ( Fig. 2 ). When we contacted the subject's family, we learned that he had been consuming a high dose of fi sh oil (7 g/day) for at least 18 months prior to death. Excluding his data from statistical analysis did not change the statistical signifi cance of any of our fi ndings reported in Tables 2-4 , and his unusual systemic lipid levels and n-3/n-6 ratios corresponded well with elevated n-3 retinal VLC-PUFA levels and very high n-3/n-6 VLC-PUFA ratios ( Fig. 2, Fig. 3, and Fig. 4 ).

Phase II study
In the second phase of this project, we examined an independently collected cohort of AMD (n = 15) and control (n = 21) subjects who had donated eyes and serum ( Table 1 ). We sampled normal appearing PR for these analyses in order to rule out macular fi brosis, atrophy, or other advanced AMD pathology as an explanation for the differences between AMD and control eyes. This time, serum LC-PUFAs and VLC-PUFA precursors were significantly associated with retina VLC-PUFA levels for controls, while the associations had similar trends in AMD donors, but they were not signifi cant. Serum LC-PUFAs and serum VLC-PUFA precursors were signifi cantly correlated with retinal VLC-PUFAs for controls ( P < 0.05), and a similar trend was seen for the AMD subjects, but it did not reach statistical signifi cance ( Table 5 ). Serum n-3/n-6 LC-PUFA ratios signifi cantly correlated with retinal n-3/n-6 VLC-PUFA ratios for AMD and control eyes ( P values ranging from 0.01 to 0.006), and similar trends were also seen when comparing serum n-3/n-6 VLC-PUFA precursor ratios versus retinal n-3/n-6 VLC-PUFA ratios, but statistical signifi cance was reached only when AMD and control data were combined ( P = 0.04). Serum EPA/AA ratios were in signifi cant positive correlation with the retinal n-3/n-6 VLC-PUFA ratios in controls but not AMD subjects.
Most organisms do not interconvert lipids between the n-3 and n-6 series, so the n-3/n-6 ratios of various lipids are important to ascertain, especially because n-6 series lipids are generally pro-infl ammatory, while the n-3 lipids are not. We found many strong and consistent correlations between serum, RBCs, and fat n-3/n-6 ratios in systemic LC-PUFAs and VLC-PUFA precursors with retinal LC-PUFAs,   clinical signifi cance of their roles in retinal health and disease. Epidemiological studies generally support the recommendation that consumption of foods rich in n-3 LC-PUFAs is associated with a lower risk of AMD (14)(15)(16), but clinical intervention studies with n-3 LC-PUFA supplements have been either negative or equivocal (17)(18)(19).
Even when there are genetic mutations in the VLC-PUFA synthesis pathway at the fatty acid elongation step at ELOVL4, which are clearly associated with an early onset form of macular dystrophy known as STGD3 ( 21,22 ), the mechanistic basis (protein aggregation vs. VLC-PUFA deficiency) remains controversial (23)(24)(25). Part of the problem lies with the challenges of linking dietary lipid consumption patterns with actual levels in the tissue of interest, the human retina. Dietary surveys are imprecise DNA was also available for these phase II subjects, and we examined whether variants in ELOVL4 and AdipoR1 [a regulator of retinal VLC-PUFA levels ( 38 )] have any infl uence on AMD risk or lipid profi les in this cohort, but we found no signifi cant relationships, nor did we fi nd any statistically signifi cant infl uences for AMD grade ( Table 1 ; all P values for comparisons were >0.05).

DISCUSSION
Considering the extraordinarily high concentrations of LC-PUFAs in retinal cell membranes and the largely unique presence of VLC-PUFAs in the human retina, there is a surprising amount of controversy regarding the Fig. 2. Correlation of n-3/n-6 LC-PUFA ratios in serum with the n-3/n-6 ratios of VLC-PUFAs in retina ( r = 0.91; P < 0.001) (A). Correlation of n-3/n-6 of VLC-PUFA precursor ratios in serum with the n-3/n-6 ratios of VLC-PUFAs in retina ( r = 0.76; P < 0.001) (B). Correlation of EPA/AA ratios in serum with the n-3/n-6 ratios of VLC-PUFAs in retina ( r = 0.92; P < 0.001) (C). The arrows correspond to data points from an outlier donor who consumed 7 g of fi sh oil daily for 18 months prior to death. represent what is going on at the tissue level. We used a high-sensitivity GC/MS analytical protocol developed in our laboratory ( 30 ) that requires minimal amounts of valuable retinal tissue; typically, a single 4-6 mm punch is suffi cient. Our GC/MS method readily allows for the distinction between n-3 and n-6 LC-PUFAs and VLC-PUFAs, which is in contrast to LC/MS methods ( 8,39 ), which do not permit distinction between these two important classes of lipids, one of which is generally anti-infl ammatory and associated epidemiologically with decreased AMD risk and cumbersome tools, and nutritional researchers have therefore developed and validated biomarkers of shortterm (weeks), medium-term (months), and long-term (years) dietary intake, but until now, there has been relatively little linkage between these biomarkers and retinal tissue levels in adult humans ( 7,8 ).
In the fi rst phase of this study, we prospectively collected serum, RBCs, orbital fat, and retinal tissue from donors with no known eye diseases to ascertain whether these systemic biomarkers of dietary lipid consumption truly  2-4 , his n-3/n-6 and EPA/AA ratios were several times higher systemically and in the VLC-PUFAs of the retina relative to all of the other donors. Although dietary histories were not part of this study, it is clear that his daily consumption of 7 g per day of fi sh oil is expected to be more than 10 times higher than what anyone would consume from diet alone and is 7 times higher than the dose tested in AREDS2. The outlier serum showed an increase of EPA over AA ( Fig. 3 ), thereby increasing the serum n-3/n-6 LC-PUFA ratio to 6 times higher than normal human serum. Simultaneously, the n-3/n-6 VLC-PUFA ratio in retina of the outlier patient was also elevated in comparison with normal human retina by 6 times, consistent with a strong effect of diet on n-3/n-6 VLC-PUFA ratios in retina.
In the second-phase study, we tested control and AMD samples from the Utah Center for Translational Medicine Donor Eye Repository, an independent cohort of donors with rigorously characterized clinical histories. As we have seen before, the AMD donors have signifi cantly lower retinal n-3 VLC-PUFAs (50%), VLC-PUFA levels (40%), and n-3/n-6 VLC-PUFA ratios than age-matched controls ( Fig.  5 ), and serum LC-PUFA, VLC-PUFA precursor, EPA/AA, and n-3/n-6 LC-PUFA ratios were likewise lower relative to (n-3 series) and one of which is proinfl ammatory and associated with increased AMD risk (n-6 series) ( 40,41 ).
Our results depicted in Table 2 demonstrate that serum and RBC lipid biomarkers are highly predictive of retinal lipid composition for total LC-PUFAs, EPA, and AA. Orbital fat is associated only with retinal EPA levels. On the other hand, retinal DHA levels have much lower associations with systemic lipid biomarkers, consistent with a more active and regulated uptake of DHA into the human retina relative to other LC-PUFAs. We examined whether LC-PUFA biomarkers correlate with retinal VLC-PUFAs, and we found that only RBCs were signifi cantly associated ( Table 3 ). When we excluded systemic DHA, which is not a major precursor of VLC-PUFAs, we could improve the correlations for RBCs and serum substantially, especially with regard to the n-3 series. Finally, we looked at n-3/n-6 ratios systemically and in the retina and found strong and consistent correlations for all three biomarkers as long as DHA was excluded ( Table 4 ). In fact, just a simple EPA/ AA ratio could accurately refl ect the n-3/n-6 ratio of the VLC-PUFAs in the retina.
The hypothesis that diet can profoundly infl uence retinal lipid composition was supported in dramatic fashion by an outlier in our group of 44 donors. As shown in Figs.   signifi cant correlations for only AA and n-3/n-6 ratios ( 7 ). Additional positive and negative correlations between RBC lipids with retinal LC-PUFAs and VLC-PUFAs were discernable when they used LC/ESI/MS, but this analytical technique is unable to distinguish between n-3 and n-6 fatty acids.
The results reported here have several important implications. First, they confi rm our prior fi ndings that VLC-PUFA levels and n-3/n-6 ratios are lower in AMD eyes ( 27 ). This suggests that further studies to understand the physiology of these unusual retinal lipids are warranted. Knockdown of VLC-PUFA levels in mice produces abnormalities only in extreme conditions (i.e., 90% knockdown) ( 24 ), but it must be noted that humans have much lower basal levels of retinal VLC-PUFAs than mice ( 30 ), so the lower levels that we observed in AMD eyes may indeed induce physiological and structural disruption in the photoreceptors. Second, our results emphasize the importance of assessing n-3/n-6 ratios systemically and in the retina when studying the potential role of dietary and supplemental lipids in modifying AMD risk, as systemic n-3/n-6 ratios show tight and consistent correlations with retinal health and are readily measured with GC/MS. Third, these data demonstrate that biomarkers of systemic lipid status accurately refl ect lipid status in the retina, especially with regard to VLC-PUFAs.
Future observational and interventional projects studying nutritional benefi ts of n-3 fatty acids should include easily obtained biomarkers of short-term and mediumterm lipid intake (serum and RBCs, respectively) to account for variability of background diet, intestinal absorption effi ciency, and participant compliance. Diet plays an important role in altering the levels and ratios of n-3/n-6 LC-PUFAs in serum, which in turn infl uence the n-3/n-6 LC-PUFA and VLC-PUFA ratios and levels in retina with possible benefi cial effects on macular physiology and protection against degeneration. Our results clarify the biological mechanisms underlying epidemiological studies that have shown that diets rich in n-3 fatty acids are protective against AMD.
age-matched controls ( Fig. 6 ). Table 5 shows that for control donors, serum n-3/n-6 ratios of LC-PUFA, VLC-PUFA precursors, and EPA/AA generally predicted retinal VLC-PUFA ratios, and for AMD donors similar trends were seen. Table 3 indicates that RBC lipids were better biomarkers than serum for retinal VLC-PUFAs in the phase I study, but unfortunately, only serum was available for the phase II study. We are therefore uncertain whether the significant associations of systemic lipid status with retinal VLC-PUFAs in controls versus the nonsignifi cant associations in AMD subjects seen in Table 5 are simply refl ections of serum biomarker variability or if there are true differences in retina lipid uptake and metabolism between normal and AMD eyes. We did explore the possibility that variants in ELOVL4 (the key enzyme in VLC-PUFA synthesis) ( 20 ) or AdipoR1 (a regulator of VLC-PUFA synthesis associated with AMD risk) ( 38 ) could explain the difference between AMD and control eyes, but we found no evidence for their involvement ( Table 1 ). In agreement with our results, others could not attain a statistically signifi cant correlation between ELOVL4 gene and macular degeneration ( 28,42 ). The association of AdipoR1 gene expression with AMD risk in our cohort was insignifi cant ( P > 0.1), which is in contrast with a recently published Finnish population study ( 33 ). In fact, the risk allele of AdipoR1 (CT) was found in only 1 of our 15 AMD subjects.
Few other studies have examined relationships between biomarkers of systemic lipid status and adult human retinal lipid composition. In 2008, a research group from France reported a negative correlation of retinal DHA with orbital fat DHA ( 8 ), which is in agreement with our present results. A follow-up study by this same group examined RBCs and retinal lipids from nine elderly donors and found that GC with fl ame ionization detection yielded