Abstract
Activation of anaplerosis takes away glutamine from the biosynthetic pathways to the energy-producing TCA cycle. Especially, induction of hyperoxia driven anaplerosis in neurovascular tissues such as the retina during early stages of development could deplete biosynthetic precursors from newly proliferating endothelial cells impeding physiological angiogenesis and leading to vasoobliteration. Using an oxygen-induced retinopathy (OIR) mouse model, we investigated the metabolic differences between OIR-resistant BALB/cByJ and OIR susceptible C57BL/6J strains at system levels to understand the molecular underpinnings that potentially contribute to hyperoxia-induced vascular abnormalities in the neural retina. Our systems level in vivo RNA-seq, proteomics, and lipidomic profiling and ex-vivo retinal explant studies show that the medium-chain fatty acids serves as an alternative source to feed the TCA cycle. Our findings strongly implicate that medium-chain fatty acids could suppress glutamine-fueled anaplerosis and ameliorate hyperoxia-induced vascular abnormalities in conditions such as retinopathy of prematurity.
Similar content being viewed by others
Introduction
Premature infants are born with underdeveloped lungs, which necessitates the use of supplemental oxygen to prevent mortality1,2,3,4,5,6. Supplemental oxygen prevents mortality but also leads to the development of hyperoxia-induced retinal vascular abnormality-retinopathy of prematurity (ROP). The early stage (phase I) is characterized by stalled normal developmental angiogenesis and vaso-obliteration leading to late stage (phase II) abnormal leaky neovascular tufts resulting in vision loss6. Although limited therapies are available for phase II ROP, studies report incidences of ROP reactivation7,8,9. Phase I of ROP is a precursor to Phase II. Therefore, early therapeutic intervention at phase I of ROP is ideal to prevent disease progression and preserve vision. Every year, ROP claims the eyesight of nearly 30,000–40,000 prematurely born infants10. Multiple biochemical pathways have been implicated in the ROP and mouse oxygen-induced retinopathy (OIR) model, including serine-dependent one-carbon, glutamine metabolism, amino acid metabolism, polyamine pathway, and fatty acid metabolism6. These metabolic changes in the oxygen-induced retinopathy model and the ROP indicate a biosynthetic deficit. Proliferating endothelial cells have high biosynthetic demand to proliferate and migrate to develop new blood vessels. Inhibition of endothelial cell (EC)-specific glutaminase (GLS) has been demonstrated to decrease proliferation and induce senescence in ECs11,12. In addition, ablation of glutamine synthetase in Müller cells has been shown to cause EC proliferation defects13. This shows that glutamine production and utilization by the retina need to be balanced during angiogenesis. Similarly, loss of serine transporter from the retina demonstrates aberrant angiogenesis in the retina14. In addition, polyamine pathway metabolites have also been shown to cause neovascular defects in the retina and retinal cell types15,16. These findings demonstrate that biosynthetic pathways are essential during angiogenesis. Glycolysis and OXPHOS provide ATP necessary for cellular function; however, biosynthetic pathways like serine-one carbon, glutamine and polyamine pathways are necessary for cells to divide. At the cellular level, serine-one carbon metabolites are essential for glutathione production, NADPH synthesis, and heme production2. Serine-derived one-carbon is the only source of methylation in the early developmental stages17. At the same time, glutamine is an essential nutrient for cellular proliferation13. Glutamine and downstream urea cycle metabolites arginine and ornithine also serve as precursors for polyamine pathway metabolites18. Glutamine is present in millimolar ranges, both in humans19 and mice20, and because of its high abundance, it also serves as an alternative source of energy. However, glutamine filling into TCA cycle is bad for proliferating tissues as this lowers the amount available for biosynthetic reactions and produces ammonium, a toxic end product21. Carbon sources that can produce energy via the TCA cycle, other than glycolysis and glutaminolysis, are fatty acids and branched amino acids. Newborns rely on fatty acids as their main energy source, as fats account for 40–50% of calories in the mother’s milk22. Free fatty acid also contributes to the majority of energy generation during fetal development22. During development, the fetus is supplied with free fatty acids from the maternal circulation. Bitman et. al. have demonstrated that the fatty acid composition of the mother’s milk in premature and full-term birth differs in medium chain fatty acids (MCFAs)23. The highest amounts of MCFAs were observed in the mother’s milk of very preterm-born infants22. MCFAs are a very good energy source as they don’t need a carnitine-based transport to be oxidized, unlike required for long and very long chain fatty acids in many tissues, except for a few tissue types24. In addition, MCFAs are broken down by the action of lingual, milk, and gastric lipases and don’t need to rely on pancreatic lipase, which is low during the early stages of life22. Moreover, MCFAs don’t need bile acids for their absorption25. Bile acid systems are not fully developed in a new born26. This makes MCFAs a preferred energy source during the early stages of life.
The liver plays an important role in managing fatty acids in the body, both the production and the transport to different organs. We have demonstrated that the liver-based HIF1α is important to produce serine-one carbon and remotely provide biosynthetic products to the retina2. In addition, signaling molecules like FGF21, a hepatokine in adults, have been demonstrated to regulate physiological and pathological angiogenesis27. FGF21 is abundant in the mother’s milk and is transferred to the infant for its development. Although FGF21 association to retinal angiogenesis and ROP have been established, the source of FGF21 which contributes to the pathophysiology of OIR and ROP are unknown. Another anabolic hepatokine, IGF-1, has also been demonstrated to play a vital role in retinal angiogenesis. These findings demonstrate a systems level of metabolic and biochemical signaling in retinal angiogenesis during the early stages of life and demonstrate the importance of liver-retina axis in early stages of development. Unlike genetic disorders, where the gene product of one or more pathways contributes to pathology, the ROP develops due to changes in oxygenation level. Although prior studies extensively investigated the ROP pathophysiology locally in the retina, the systemic contribution of ROP complications in preterm infants remains underexplored. We here used mouse model of oxygen-induced retinopathy and determined the systems-level liver-retina metabolic exchanges in response to hyperoxia.
Results
Strain differences in susceptibility to oxygen-induced retinopathy
BALB/cByJ mice are reported to be resistant to OIR when exposed to hyperoxia28. The differences in genetic propensity to OIR have also been shown in other mouse and rat strains29,30,31. To understand the molecular underpinnings that protect BALB/cByJ from hyperoxia induced OIR, we first evaluated the retinal vasculature of C57BL/6J and BALB/cByJ mice at postnatal day (P) 12 (vasoobliterative-phase 1) and P17 (neovascular- phase 2).
At phase 1 (P12) of OIR, hyperoxia susceptible C57BL/6J strain showed more pronounced avascular zone in the retina in comparison to BALB/cByJ strain (Fig. 1a, b, d, e). At phase 2 (P17), C57BL/6J retinas displayed enhanced growth of neovascular tufts (Fig. 1c). In contrast, the BALB/cByJ retina exhibited diminished neovascular tufts at P17 (Fig. 1f). Consistent with prior findings28, the difference in susceptibility to OIR demonstrates the genetic basis of protection in BALB/cByJ mice strains.
a, d normoxic condition P12; b, e mice exposed to hyperoxia from P7 to P12; c, f mice exposed to hyperoxia from P7 to P12 and then room air until P17. Red arrows indicate pathological neovascular tufts.
Medium-chain fatty acid utilization pathway is upregulated in the retina of OIR resistant strain
We previously reported that hyperoxia alters cellular metabolism2,6,16,32. Therefore, we sought to determine if the genetic basis of protection is related to altered metabolic pathways in OIR-resistant (BALB/cByJ) mouse strain. RNA-seq was performed on retinas of OIR-susceptible (C57BL6J) and OIR-resistant (BALB/cByJ) strains at P12 (phase 1) and investigated the changes in biosynthetic pathways. Interestingly, we found significant differences in the transcripts levels derived from genes involved in the glycolytic regulator PDK2 which controls the entry of glucose into TCA cycle (Fig. 2a), polyamine pathway genes (Supplementary Fig. S1) and serine/one-carbon uptake and utilization (Supplementary Fig. S1). Since hyperoxia leads to glutamine-fueled anaplerosis32, we looked at the fatty acid breakdown pathways that can feed into the TCA cycle to replete metabolites in TCA cycle in the absence of flux from glycolysis or glutamine breakdown. There are only a few pathways that can feed into TCA cycle other than glycolysis which includes fatty acid oxidation, branched-chain amino acids breakdown, glutamine breakdown, and proline breakdown. Of all these pathways, medium-chain fatty acids (MCFAs) have been shown to differ in the mother’s milk of preterm infants, implicating a role in ROP23. We examined all types of acyl-coA dehydrogenases that are essential for the breakdown of each fatty acid types (short, medium, long and very long chain). Intriguingly, we found statistically significantly increased expression of Acadm, Ech1 and Acaa2 transcripts in the OIR-resistant strain than that of the susceptible strain (Fig. 2b). Interestingly, the retinal Pparα negatively correlated to the expression of MCFA utilization genes (Fig. 2c). We confirmed the expression levels of Acadm and Pparα in an additional qPCR experiment (Supplementary Fig. S2). The identified genes play a key role in the breakdown of MCFAs into their corresponding Acyl CoA derivative, which can then feed into TCA cycle and provide energy when the entry of carbon from other sources into TCA cycle is limited.
a Glycolytic inhibitor Pdk2 which inhibits Pdh and inhibits flux from glycolysis entering TCA cycle showed statistically significant changes between the two mouse strains. Mut is an additional entry point of propionyl-CoA and branch-chain amino acids. Mut showed statistically significant changes between the two strains. b Changes in medium chain fatty acid utilization enzymes Acadm, Ech1, and Acaa2. c Retinal Pparα negatively correlates with the expression of medium-chain fatty acid utilization genes. d Retinal expression of Oxct1, rate limiting ketone body utilization enzyme, showed statistically significant upregulation in the retinas of hyperoxia exposed OIR resistant strain. P-values were calculated using unpaired one-way ANOVA with Bonferroni corrections applied for multiple comparison, *p-value ≤ 0.05, **p-value ≤ 0.01, ***p-value ≤ 0.001 and ****p-value ≤ 0.0001. Legends: CN, C57BL/6J normoxia; BN, BALB/cByJ normoxia; CH, C57BL/6J hyperoxia; BH,BALB/cByJ hyperoxia. n = 4, all samples. Line at mean with standard deviations.
In addition, we also found upregulation of methylmalonyl-CoA mutase- Mut, which facilitates the entry of odd-chain fatty acids and branched-chain amino acids into the TCA cycle (Fig. 2a). Moreover, liver also processes MCFAs into ketone bodies that are broken down by the peripheral tissue to facilitate energy production24. The utilization of ketone bodies by the peripheral tissues is mediated by Oxct1, enzyme critical for ketolysis33,34. Our RNA-seq data shows statistically significant upregulation of the rate-limiting ketone body utilization enzyme Oxct1 in the retinas of hyperoxia exposed OIR resistant strain (Fig. 2d).
Secretome data reveal differential changes in circulating growth factors in lipid metabolism at systems levels in OIR-resistant and -susceptible strains
Growth factors like FGF27,35 and IGF36,37,38 have been shown to correlate with developmental angiogenesis and have been shown to correlate with phase 1 pathogenesis in oxygen-induced retinopathy. These growth factors in circulation also play a central role in interorgan regulation of lipid metabolism39,40. We investigated if the changes in the circulating growth factors contribute to altered metabolism in OIR-susceptible and OIR-resistant strains, using the proteomics platform SomaScan. This platform utilizes aptamer-based technology, which binds protein targets with the structural specificity, thus avoiding non-specific binding to misfolded proteins. In addition, this technique has a better dynamic range as compared to other proteomics techniques. This dynamic range is specifically beneficial for plasma samples in which albumin and other high abundance proteins hinder the measurement of low abundance targets.
The proteomic analysis show that the hyperoxia significantly downregulated IGF1, its binding partner IGF ALS, IGFBP-1, IGFBP-3 and IGFBP-5 levels in OIR-susceptible C57BL/6J mice in comparison OIR-resistant BALB/cByJ (Fig. 3). There was no significant change in the expression level of IGFBP-2 in response to hyperoxia in both the C57BL/6J and BALB/cByJ strains (Fig. 3). Moreover, we also measured various FGF and FGF binding proteins that includes FGF1/bECGF, FGF2/bFGF, FGF3, FGF4, FGF5, FGF6, FGF7, FGF8, FGF8A, FGF8B, FGF8F, FGF9, FGF10, FGF12, FGF16, FGF17, FGF18, FGF19, FGF20, FGF22, FGF23, FGFP1 and FGFP3 using SomaScan. Among the various FGF and FGF binding proteins, only FGF8 showed statistically significant difference in hyperoxic condition (Fig. 3).
Circulating IGFs, IGFBPs, FGFs show significant difference between OIR-susceptible and -resistant strains. Adiponectin negatively correlated with the protection seen in BALBc/ByJ. Adiponectin also didn’t correlate with FGF21 levels in two mouse strains. p-values were calculated using unpaired one-way ANOVA with Bonferroni corrections applied for multiple comparison, *p-value ≤ 0.05, **p-value ≤ 0.01, ***p-value ≤ 0.001 and ****p-value ≤ 0.0001. Legends: NC, P12 C57BL/6J normoxia; NB, P12 BALB/cByJ normoxia; HC, P12 C57BL/6J hyperoxia; HB, P12 BALB/cByJ hyperoxia; HCP12, P12 C57BL/6J hyperoxia; HBP12, P12 BALB/cByJ hyperoxia; NCP12, P12 C57BL/6J normoxia; NBP12, P12 BALB/cByJ normoxia; HBP17, P17 BALB/cByJ hyperoxia; HCP17, P17 C57BL/6J hyperoxia. n = 3 SomaScan analysis IGF-1, IGF-ALS, IGFBP-5, IGFBP-2, IGFBP-1, IGFBP-3, FGF-8; n = 4 ELISA analysis Adiponectin, n = 4 all samples except n = 3 for NB; FGF21 ELISA, n = 6 HCP12, n = 4 HBP12, n = 5 NCP12, n = 9 NBP12, n = 3 HBP17, n = 3 HCP17. Line at mean with standard deviations.
FGF21 has been shown to promote vascularization in phase 1 ROP via adiponectin dependent lipid oxidation in hyperglycemic OIR model27. As FGF21 was not among the SomaScan covered targets, we measured FGF21 levels separately by ELISA because of its regulatory role in metabolism41. The data showed drastic differences in FGF21 level between C57BL/6J and BALB/cByJ mice with and without exposure to hyperoxia at P12 (Fig. 3). Of interest, FGF21 levels were very low at phase II (P17) and did not show any significant change, implying FGF21 potential role in the early phases of OIR pathophysiology.
The two most important tissue targets of FGF21 are Adipose and CNS42. In adipose, FGF21 triggers Adiponectin expression leading to the activation of multiple downstream targets, including the acute insulin sensitivity42. Adiponectin is demonstrated to inhibit proliferation in vascular proliferative retinopathies such as phase II of OIR, AMD and Diabetes43. We investigated the plasma adiponectin levels to check if FGF21 and Adiponectin protein levels correlate in our OIR models. Adiponectin negatively correlated with the FGF21, and its levels were lower in OIR-resistant strain as compared to susceptible strain (Fig. 3). This suggest that the FGF21 dependent, but adiponectin independent mechanism likely provide BALB/cByJ resistance to OIR.
Genes required for ketogenesis and β-oxidation in the liver were upregulated in OIR-resistant strain, and correlates with plasma FGF21 protein level
Medium chain fatty acids are known to be cleaved in the liver to produce ketone bodies44. We investigated for genes involved in the uptake of fatty acids from circulation. Fatty acids are taken up by the liver-specific fatty acid binding proteins, and then broken down into acetyl-CoA molecules via β-oxidation and subsequently converted to ketone bodies for their transport into the peripheral tissues24,44. Liver is a ketogenic organ whereas peripheral tissues are ketolytic. These ketone bodies produced by the liver can serve as a source of energy for peripheral tissues. This process also releases non-esterified fatty acids, produced from long-chain or very long-chain fatty acids, which can leak through the liver into the circulation and serve as an additional source of energy. Transcript levels of Fabp1, the liver specific fatty acid binding protein, which imports and activates Pparα, was found to be higher in BALB/cByJ as compared to C57BL/6J (Fig. 4). Downstream, liver specific Pparα and Ppparβ/δ showed higher expression in BALB/cByJ exposed to hyperoxia as compared to C57BL/6J (Fig. 4). In addition, targets downstream of Pparα, involved in beta oxidation, ketone body production and its export, demonstrated higher expression in the BALB/cByJ exposed to hyperoxia in comparison to C57BL/6J (Fig. 4). HMG-CoA synthase, one the most important regulators of ketogenesis, was also higher in BALB/cByJ as compared to C57BL/6J (Fig. 4).
Liver genes expression changes in fatty acid breakdown and ketogenic pathways in P12 old mice shows higher expression of enzymes in fatty acid oxidation and transport of ketone bodies in OIR-resistant strain of mice, whereas urea cycle genes demonstrate higher expression in OIR-susceptible mice indicative of difference in substrate utilization for energy. qPCR data p-values were calculated using unpaired two-tailed t-test. RNAseq data p-values were calculated using unpaired one-way ANOVA with Bonferroni corrections applied for multiple comparison. Legends: Balbc_H, P12 BALB/cByJ hyperoxia; C57_H, P12 C57BL/6J hyperoxia. n = 4 qPCR experiments. n = 4 RNAseq experiment except Balbc_N, n = 3 RNAseq Balbc_N. Line at mean with standard deviations.
In addition, urea cycle genes, Arg1 and Cps1, that are inhibited by Pparα showed lower expression in BALB/cByJ as compared to C57BL/6J (Fig. 4), implying the lower utilization of nitrogenous compounds for energy production. Nitrogenous compounds, like glutamine, release ammonium when deaminated or deamidated to produce energy via TCA cycle, which necessitates the use of liver-based urea cycle to remove toxic ammonium, and in response to which the expression of urea cycle genes is upregulated. Our data suggests that the BALB/cByJ uses non-nitrogenous compounds for its energy production in hyperoxia, and therefore has lower levels of urea cycle transcripts in hyperoxic conditions.
Moreover, liver RNA sequencing data further validated our gene expression studies and displayed similar expression levels for all the genes, except for Arg1, PPARβ/δ and Bdh1 which were not statistically significant in RNA sequencing data (Supplementary Fig. S3). Although change in Bdh1 was not statistically significant, it did show trend similar to that in found in qPCR analysis. We looked at the Acot2 and Acot3, enzymes that breakdown LCFA-CoA and MCFA-CoA into their corresponding non-esterified free fatty acids. Both these enzymes showed higher expression in OIR-resistant strains in hyperoxic condition (Fig. 4). Since hepatic Fgf21 has been shown to control lipid metabolism in the body via liver-brain axis45, we looked at hepatic Fgf21 expression levels in the hepatic RNA sequencing data to determine if FGF21 protein levels in the plasma correlate with the hepatic Fgf21 transcript levels in the OIR model at P12. Liver and milk are the two main sources of Fgf21 in the early stages of life46. Since the source of FGF21 protein in the plasma in OIR context is not clear, we performed correlation analysis of hepatic Fgf21 transcript and plasma FGF21 proteins levels. We found a very strong correlation between hepatic Fgf21 transcript levels measured using RNA sequencing and plasma levels of FGF21 protein levels measured with an ELISA (Fig. 5). We additionally looked at some of the known inducers of Fgf21 in the liver that have been previously reported. Pparα is a known master regulator of Fgf2147. High fat diet, ketogenic diet, and fasting are the main activators of Pparα dependent Fgf21 induction47. Crebh shows some cross talk with the Pparα during overnutrition and excess lipid availability48. Transcript levels of Fgf21 in the liver strongly correlated with the transcript levels of Pparα and Crebh in our RNA seq data (Fig. 5). Some of the other known transcriptional inducers of Fgf21 in the liver are Chrebp, Atf4, Nurp1, and Xbp148. Chrebp is a carbohydrate sensitive transcription factor, and it induces hepatic Fgf21 in response to glucose, fructose, and ethanol. Whereas Atf4 and Nurp1 activate Fgf21 in response to changes in amino acid concentrations in the liver49,50,51. Some studies have suggested that the ER stress involves Xbp1 to induce Fgf21, however, recent findings demonstrate Xbp1 is dispensable and may not be the direct regulator of Ffg21 in the mouse liver50. In our correlation analysis, Chrebp, Atf4, Nurp1, and Xbp1 transcript negatively correlated with the Fgf21 transcript levels (Fig. 5). Correlation analysis suggests that the Pparα and Crebh contribute to the FGF21 protein induction in OIR context.
Liver based FGF21 is secreted into the plasma in P12 old mice, and its induction involves hepatic PPARα and CREBH. Hepatic FGF21 transcript levels strongly correlate with FGF21 protein levels in plasma. Hepatic FGF21 transcript levels also strongly correlate with transcript levels of PPARα and CREBH. Hepatic transcript levels of ChREBP, ATF4, Nurp1, and XBP1, negatively correlated with the transcript levels of FGF21 in the liver.
Genome-scale modeling reveals metabolism-wide differences between BALB/cByJ and C57BL/6J mice
To gain further insight into systemic metabolic changes associated with hyperoxia in the two conditions, genome-scale metabolic modeling was performed. Briefly, genome-scale modeling uses manually curated, molecularly detailed genome-scale reconstructions to simulate metabolic fluxes for an organism of interest52. Genome-scale models can be contextualized through the integration of nutritional and omics data52. A genome-scale reconstruction of mouse53 was tailored condition-specific with the RNA sequencing data from the retina (Methods), resulting in 16 sample-specific models containing on average 3437.94 +/− 50.41 reactions, 2475.13 +/− 20.77 metabolites, and 1290.63 +/− 25.16 genes. Subsequently, the sample-specific metabolic fluxes were predicted for BALB/cByJ and C57BL/6J mice under hyperoxic and normoxic conditions. The entire range of feasible metabolic fluxes was accessed by retrieving the minimal and maximal flux through each reaction (Methods, Supplementary Fig. S4), with positive and negative flux values corresponding to fluxes in the forward and reverse direction, respectively.
Metabolic fluxes predicted for the four conditions were compared and reaction fluxes significantly differing between strains and between normoxia vs. hyperoxia were retrieved. Supplementary data 2a–d shows the analyzed enzymatic and transport reactions by reaction ID and annotated by metabolic subsystem as also available on the Virtual Metabolic Human (VMH) database54. We then grouped reactions with significantly different flux by subsystem. A variety of reactions from different metabolic subsystems were initially significantly different in metabolic flux in the retina (Supplementary Fig. S4), though the differences were no longer significant after more strict statistical correction for multiple testing. This implies the need to have a large study cohort in future, however, the initially significant differences are still enough to highlight the important differences between the two strains in normoxia vs. hyperoxia. In retina under hyperoxic conditions, the BALB/cByJ strain was enriched in fluxes in starch and sucrose metabolism, pentose phosphate pathway, sphingolipid metabolism, and carbohydrate metabolism compared with C57BL/6J (Supplementary Fig. S4, Supplementary Data 2). Note that reactions assigned to starch and sucrose metabolism in this case encompassed glycogen synthesis and utilization, e.g., glycogen synthase (VMH ID GLGNS1), glycogen phosphorylase (VMH ID GLPASE1) (Supplementary Data 2c). Glycogen is the only energy reserve in the retina55 suggesting lower available energy levels in the C57BL/6J strain under hypoxia compared with BALB/cByJ. In the BALB/cByJ under hyperoxia, retinal metabolism was depleted mainly in transporter flux with 384 affected reactions (Supplementary Fig. S4, Supplementary Data 2). The C57BL/6J strain showed a depletion in fatty acid oxidation (highest of all) and synthesis under hyperoxic conditions (Supplementary Fig. S4, Supplementary Data 2). Although simulations are impressive tools to map complex data to reduce complexity in interpreting the expression findings, it should be noted that genome-scale reconstructions in their basic form lack regulatory mechanisms such as enzyme kinetics and transcriptional regulation. In future studies, our in-silico predictions regarding affected pathways in retina could be validated experimentally.
In summary, a wide variety of pathways were altered in in-silico simulations. In retina, most changes were observed in transport reactions in the BALB/cByJ strains between the compared conditions, and fatty acid oxidation was depleted in C57BL6J exposed to hyperoxia.
Predicted metabolic fluxes correlate with lipidomics data
We next investigated if metabolic fluxes in retina that had been computed in silico were linked to systemic metabolism, as indicated by experimentally measured plasma lipidomics for the same animals. To this end, we calculated the Spearman correlations between all predicted metabolic reaction fluxes in retina and lipidomics measurements in plasma that were significantly different between groups. After correction for multiple testing, significant correlations remained between 129 metabolic model reactions and 162 lipids in the plasma (Fig. 6a). Shown are metabolic model reactions by VMH ID54 and annotated by metabolic subsystem. Lipids are shown by feature as determined through lipidomics. As expected, lipids correlated with reactions in lipid metabolism, but surprisingly, high correlations were also found with reactions in glycolysis, carbohydrate metabolism, vitamin metabolism, and amino acids metabolism (Fig. 6a). For instance, correlations were seen with retinol dehydrogenase (VMH reaction IDs RDH1, RDH2, RDH3), NAD metabolism (VMH reaction IDs NMNS, NMNDA, NAPRT), and pentose phosphate pathway (VMH reaction IDs G6PDH2rer, RPE, TALA). Hence, the modeling predicted that metabolic fluxes in retina were functionally linked to the plasma lipidome, which demonstrates that the retinal lipidome is dependent on circulating lipid species in the plasma. Remarkably, the plasma lipidome was additionally correlated with a variety of non-lipid pathways in retina indicating a link with systemic retinal metabolism.
a Spearman correlations between metabolic reaction fluxes (mmol*g dry weight−1*hr−1) and plasma lipid concentrations that were significant after correction for multiple testing. (Methods). Rows shows metabolic reactions in genome-scale models with VMH54 reaction IDs, and annotated by metabolic subsystem. “Forward” indicate reaction fluxes in the direction of reaction stoichiometry as shown on VMH, while “reverse” indicate reversible reactions that carried flux in the counter direction. Columns indicate plasma lipid features identified through lipidomics. Only reaction fluxes and lipids that correlated significantly after correction for multiple testing in at least one case are shown. b PLS-DA score plot of lipidomics data. c VIP plot resulting from PLS-DA analysis of lipidomics data depicting top 15 features from PLS-DA component 1. d Free fatty acid octanoate in the mouse plasma. Octonate data p-values were calculated using unpaired two-tailed t-test. n = 5 per group Octonate measurement. Line at mean with standard deviations.
We performed a principal component analysis of the lipidomics data resulting for all the four strains. PCA analysis shows a clear grouping of all the replicates into four different clusters (Supplementary Fig. S5). Normoxic and hyperoxic C57BL6J showed clear difference with very clear separation on the PCA plot, which demonstrate that the lipidome of C57BL6J changes to a very great extent between hyperoxia and normoxia. However, normoxic and hyperoxic BALBc/ByJ strain showed very little difference between the two groups demonstrating their lipidome changes very little in response to hyperoxia. We next performed supervised multivariate analysis- partial least square discriminant analysis to force separation between the four groups and find variables importance in projections. Component plot resulting from PLS-DA showed major differences in the lipidome between mouse strains in component 1 (Fig. 6b). Based on these findings, we evaluated VIP score plot for component 1 (Fig. 6c). The main class of lipids which change in response to hyperoxia are phosphatidylcholines. Inague et. al. has seen similar classes of lipids in their thorough investigation of retinal lipidome for the OIR model56. Interestingly, PCs containing AA and DHA ranked among the top hits in our analysis. Since AA and DHA have already been shown to have an important role in development57, our finding is not much of a surprise, however, it is interesting finding that some of these either showed changes that were either opposite between two strains in hyperoxic conditions or were unchanged in BALBc/ByJ mouse strain. We also found many lipids which changed in both the strains in response to hyperoxia, indicating those lipids to be oxygen sensitive, however implying those changes are not responsible for protection seen in BALBc/ByJ strain. The key finding from this analysis was that the nature of systemic changes we found in our study were like those reported earlier for the retina. Our data demonstrates that the changes are systemic in nature rather than local to the retina.
In addition to lipidomics, we also measured octanoate levels in the plasma of both the mice strains. Since octanoate is known to be the most abundant free fatty acid in the mother’s milk, we looked at octanoate levels in the plasma of the pups. There were higher levels of octanoate in the plasma of BALBc/ByJ mouse as compared to C57BL/6J mouse strain (Fig. 6d).
Stable isotope labeling of retinal explant demonstrates octanoate can replenish TCA cycle in response to hyperoxia
Based on our findings from the RNA sequencing data and known differences in MCFA levels in the milk of preterm vs. full-term infants, we hypothesized that retina may be able to utilize medium-chain fatty acids to fill its TCA cycle in hyperoxic conditions if we provide supplementation of MCFAs in the media. This is because MCFAs like octanoate and heptanoate have been shown to anaplerotic in multiple systems58,59,60,61. Although octanoate is not a direct anaplerotic compound, it induces anapleorsis through an unknown mechanism in multiple systems58. We incubated retinal explants from C57BL/6J mice in media containing [13C8] labeled octanoate. Retinal explants were able to use octanoate as a substrate to feed into their TCA cycle and in addition, this process was accelerated in hyperoxic conditions (Fig. 7). This implies that medium-chain fatty acid octanoate can fuel the TCA cycle to replenish the glutamate pool. Total quantities of citrate and its M6 form were higher in the hyperoxic retinal explants as compared to normoxic explants (Fig. 7). Since, hyperoxia up regulate glutamine fueled anaplerosis, we hoped that the MCFA supplementation will prevent glutamine breakdown and normalize glutamate levels. Interestingly, total glutamate levels were normalized and M5 form of glutamate, produced from labeled octanoate were higher in hyperoxic retinal explants as compared to normoxic explants (Fig. 7), indicative of higher conversion of [13C8] octanoate into citrate and subsequently into glutamate.
Total citrate and total glutamate are the sum of all MIDs. M6 citrate and M5 glutamate are fully labeled forms of citrate and glutamate, respectively. p-values were calculated using unpaired two-tailed t-test. n = 3 per group.
Discussion
In this study, we utilized multiple omics approaches and demonstrated that the systemic effect of hyperoxia contributes to ROP pathophysiology. We determined systems level changes in lipid metabolism pathways between OIR-resistant and OIR-susceptible strains. We investigated both the previously known OIR relevant lipid/fatty acid metabolic pathways and unknown contributors, to find genetic basis of protection against OIR. Liver is the central organ that regulates lipid metabolism in the body, converting dietary fatty acids or de novo produced fatty acids into transportable VLDL and LDL. Nutritional exogenous lipids are emulsified into chylomicrons in the small intestine and transported either to the liver, or to the peripheral organs for their metabolic needs. The tissue specific lipase catalyzes breakdown of both the chylomicrons packed and VLDLs containing TGs and helps release these fatty acids to the tissues. We investigated the lipid/fatty acid metabolic pathways in the liver (producer and transporter of lipids), plasma (transporter of lipids) and the retina (consumer of lipids for biomass and biosynthetic process), to have a comprehensive systems levels understanding of dietary and de novo production of lipids, and their utilization by the retina in OIR. We investigated all three omics layers, transcriptomics, proteomics and lipidomics, to evaluate these changes in OIR.
Using this comparative multiomics data from OIR-resistant and susceptible strains, we confirm our previous findings that the OIR is a biosynthetic deficit. In our previous work, we demonstrated that hyperoxia rewires retinal TCA cycle to use glutamine as an anaplerotic substrate, in response to inhibition of flux from glycolysis into TCA32. We have also demonstrated that this rewiring of TCA cycle also impacts other pathways like polyamine biosynthesis in the endothelial cells16 and serine derived one-carbon metabolite exchanges between the liver and the retina2. In this comparative multiomics study, we saw higher expression of serine pathway genes, polyamine biosynthesis genes in the retina of OIR-resistant strain as compared to OIR-susceptible strain of mice. In addition, we also observed lower expression of Pdk2, which inhibits PDH enzyme and blocks entry of glycolytic carbon into TCA cycle, in the retina of OIR-resistant strains as compared to OIR-susceptible strain. Moreover, we also observed lower expression of polyamine breakdown enzyme Smox in OIR resistant mice in comparison of OIR-susceptible mice. This implies that the OIR resistant mice can preserve its biosynthetic molecules for proliferation of endothelial cells in developing vasculature in the retina. This may be in response to availability of alternative sources of energy other than using these biosynthetic pathway metabolites for energy production via TCA cycle. Some of the alternative sources of TCA cycle filling are fatty acids and branched chain amino acids. Since, fatty acids are highly abundant in milk and 40–50% of calories a newborn consumes comes from the fatty acids present in mother’s milk22, we focused on systems levels analysis of production, utilization, and transcriptional regulation of lipids/fatty acids. Quantities of medium chain fatty acids in mother’s milk differs between a preterm and full-term infant implying they might have an important role in OIR and ROP. In addition, medium chain fatty acid metabolism by astrocytes in the brain has been shown to modulate glutamine supply62. Our analysis shows that the retina of OIR resistant strain has higher expression of MCFA breakdown enzymes, which breakdown MCFAs via retinal β-oxidation. We have found higher availability of MCFA-octanoate in the plasma of hyperoxic OIR-resistant mice in comparison to OIR-susceptible strain. Moreover, enzymes required to use medium chain fatty acids show higher expression in the liver and the retina of OIR-resistant mouse strains. Moreover, ketogenic enzymes that breakdown fatty acids in the ketones for their transport into the retina, and the ketolytic enzymes required to use the ketones in the retina showed higher expression in the OIR-resistant strain. In addition, [13C8] octanoate labeling of retinal explant demonstrates that the retina can use octanoate to replenish its TCA cycle metabolite pools. These findings suggest that MCFAs may cure biosynthetic deficit seen in phase 1 OIR. Triheptanoin, a triglyceride of 7 carbon containing medium chain fatty acid, has successfully been applied for many applications to replenish TCA cycle metabolite and cure energy crisis63,64. Our data suggests that this approach may also help to prevent energy crisis seen in OIR.
One of the known transcriptional factors that induces fatty acid oxidation is PPARα65. Interestingly expression of retinal PPARα negatively correlated with expression MCFA utilization enzymes in our study. This may be due to higher availability of ketones from the liver to the retina in the OIR-resistant mouse strain. Interestingly, we saw higher expression of enzyme and transporter involved in ketogenesis in the liver of OIR resistant strain of mouse. PPARα in the liver positively correlated with the protection and MCFA utilization in the retina. FGF21, which is another well-known inducer of β-oxidation66,67, has been demonstrated to induce retinal angiogenesis in hyperglycemic model of OIR via adiponectin27. Interestingly, we observed higher expression of PPARα in the liver of P12 old OIR-resistant mice, which correlates with the fatty acid import, export, and oxidation enzymes in the liver of BALB/cByJ mice; hepatic PPARα expression correlates positively with the β-oxidation enzymes in the retina in our study. It is possible that during the early stages of life, the liver based PPARα controls MCFA oxidation in the retina by inducing FGF21 in the liver and secreting it into the blood. Our correlation analysis demonstrated a very strong correlation between hepatic FGF21 transcript levels and plasma FGF21 protein levels, indicating FGF21 necessary for protection is secreted from the liver. Adiponectin showed a negative correlation with hepatic and plasma FGF21. In addition, adiponectin showed negative correlation with the protection in OIR resistant strain. Our findings that the hepatic PPARα is higher in the OIR-resistant mice, and it is induced in response to hyperoxia, and displays strong correlation with hepatic FGF21 transcript levels and plasma FGF21 levels, demonstrates that hepatic PPARα may be the stimulus for FGF21 secretion in OIR context. Our finding that FGF21 only peaks up during the early stages of life, i.e higher in P12 mice as compared to that in P17 mice, suggests that the liver derived FGF21 is a developmental hepatokine required during early development of retina. FGF21 has been shown to be present in mother’s milk. The concentration of FGF21 in plasma and the milk in mouse and rat are almost comparable. In comparison, human milk has approximately half the quantity of FGF21 in milk compared to plasma. Our data demonstrates that the FGF21 protein levels in the plasma and FGF21 transcript levels (data not shown) in the retina showed no correlation. However, there was a very strong correlation between hepatic FGF21 expression in the liver and plasma FGF21 protein levels, indicative of hepatic origin of FGF21 in early stages of life. FGF21 is induced by many different stimuli, for example ChREBP based induction in response to reductive stress68. Our correlation analysis further suggests PPARα and CREBH based induction of hepatic FGF21 expression and secretion into the plasma. One of the known endogenous activators of PPARα is 7(S)-Hydroxydocosahexaenoic acid which is a derivative of DHA69. DHA levels have been demonstrated to negatively correlate with severity of ROP57. Free fatty acids are also one of the very potent activators of PPARα. We saw higher availability of free fatty acid octanoate in hyperoxic OIR-resistant mice that may be upstream signal to hepatic PPARα based protection in OIR. We don’t yet know if this octanoate is of maternal origin or de novo produced. Based on the multiomics findings, we believe pharmaceutical treatment with PPARα agonist fibrates or hepatic overexpression of PPARα using genetic tools like AAVs may prevent OIR by modulating downstream FGF21 and retinal β-oxidation. Fenofibrate has been demonstrated to protect against DR and NV in the phase II of the OIR rat model70. Since PPARα induces ketone body formation in liver, and our data suggests upregulation of ketone body production and secretion genes, a ketone body transport from liver to the retina may be an alternative mechanism of protection in OIR. Additionally, our data suggests that there are physiological differences in FGF21 signaling between pups and adult mice. FGF21 is not the only molecule that changes with age. There are other examples like OAT, which serves as a bidirectional enzyme in neonates and unidirectional in later stages of life71. The reason for this reversibility in neonatal OAT is low levels of arginine in the mother’s milk, which makes it impossible to support polyamine pathway flux for proliferative processes. This example is another reason to believe that maternal nutrition may be one of the main drivers of metabolic processes in the pups in OIR context. Given that the mother milk contains MCFA, FGF21, and their trend changes right after birth, it warrants looking into the role of maternal nutrition in OIR and ROP.
Methods
OIR model and tissue collection
All the animals experiments in this work were performed as per Massachusetts General Hospital Institutional Animal Care and Use Committee approved protocol. The OIR protocol used was as described in the publication by Smith et al.72. Briefly, mice were maintained in room air from P0-P7, on P7 mice were exposed to hyperoxia (75% oxygen) until P12 which caused vaso-obliteration Phase I of OIR. P12 hyperoxic treated mice, when maintained in room air on P17 demonstrated retinal neovascularization. Control mice were kept in room air until P12 or P17. All the mice were anesthetized with isoflurane and euthanized by cervical dislocation before the removal of tissues.
Blood was collected from retro-orbital route into heparin tubes and spun down at 1000 × g for 20–30 min at 4 °C to separate plasma. Plasma was flash-frozen in liquid nitrogen and kept at −80 °C until use. Livers were removed from euthanized mice using forceps and immediately frozen in liquid nitrogen. Livers were ground in a mortar pestle in the presence of liquid nitrogen, and then powdered livers were stored at −80 °C for later use.
Retina staining with Isolectin
Retinal flat mounts were prepared and stained as described in Jidigam et al.73. Isolectin GS-IB4 from Griffonia simplicifolia, Alexa Flourtm 568 conjugate (Invitrogen) was used to label blood vessels. Images were acquired on Nikon 90i upright epifluorescence microscope at MGH microscopy core facility.
RNA sequencing
Total RNA was extracted using TRIzol reagent (Invitrogen) as per instructions provided with the reagent. Following extraction, RNA was cleared of DNA using DNA-free DNase Removal kit (Invitrogen). The sample quality was determined on an Agilent 5400 fragment analyzer instrument. The library was prepared with the help of a NEBNext Ultra II library prep kit for Illumina. The samples were sequenced on NovaSeq 6000 with the help of NovaSeq 6000 S4 Illumina kit in PE150 (Paired-end with 6 G raw data per sample) mode. Fastq formatted files were processed to remove reads containing adapters, lower quality reads and poly-N reads. Reference genome index was prepared using Hisat2 v2.0.574. Mus Musculus reference genome GRCm38/mm10 was used as reference. Quantification of gene expression was performed with the help of featureCounts v1.5.0-p375. FPKM was calculated for each gene. Differential analysis was performed with the help of DESeq276 version 1.20.0, with features filtered for log2 (fold change) ≥ 1 and padj ≤0.05. The sequencing project post RNA extraction was performed by Novogene Corporation Inc., Sacramento, CA.
SomaScan proteomics
Blood samples were collected via retroorbital route into heparin tubes. Samples were spun at 1000 × g at 4 °C for 20–30 min and flash frozen in liquid nitrogen. Samples were later thawed and 35 µl sample volume was used to measure proteins on SOMAscan7K v4.1 proteomics platform as described previously as in Haslam et al.77.
Metabolite labeling and measurement
Eyes were enucleated from P12 old mice after euthanasia. Eyes were maintained in Astrocyte media (A1261301) on an ice pack during dissection. After dissection, retinas were removed from the media, washed with normal saline and then incubated in media containing [13C8] octanoate. To prepare this custom stable isotope media, first BSA bound [13C8] octanoate was prepared by dissolving 16 µl of octanoate in 1 ml of ethanol and then adding 200 µl of above in 2.5 ml of fatty acid free BSA solution (1.133 mg/ml prepared in 150 mM NaCl aqueous solution) made to 10 ml total volume with 150 mM NaCl. The BSA bound [13C8] octanoate was added in 1:5 dilution with minimal media (DMEM catalogue number A14430-01) containing 5 mM glucose, 2.5 mM glutamine, 2 mM carnitine and 1% PenStrep (10,000U/ml stock solution, Invitrogen). Two retinas were incubated per well in 500 µl of media volume for 24 h. One plate was exposed to hyperoxia 75% oxygen and other plate was exposed to normoxia 21%, in 5% CO2 incubator at 37 °C.
Metabolites were extracted by adding 1 ml of 80% methanol per tube. Tubes were centrifuged at 15,000 × g for 5 min at 4 °C, and the supernatant was dried for metabolite measurement. GCMS measurements were performed on 8890B GC coupled to EI/CI 5977 MSD mass spectrometer using an electron impact ionization extractor ion source (Agilent) as previously described16.
FGF21 ELISA
FGF21 ELISA was performed following the manufacturer’s guidelines (FGF21 ELISA kit, Millipore Sigma catalogue number- EZRMFGF21-26K). Briefly, we used 10 µl of plasma sample in 30 µl of assay buffer. Ten-microliters of detection antibody was added to the samples, followed by incubation for 2 h at room temperature, then washed 3 times with 300 µl wash buffer. Then 100 µl of enzyme solution was added to each well, incubated for 30 min at room temperature, followed by another washing step 3 times with 300 µl of wash buffer. This step was followed by an addition of 100 µl of substrate and then again incubated at room temperature for 20 min. Following this step 100 µl of stop solution was added to each well, and absorbance was recorded at 450 and 590 nm. Samples were quantified using a standard curve prepared with different concentrations of FGF21 standard provided in the kit.
Correlation analysis
Two targets, protein vs. transcript or transcript vs. transcript, were plotted and Pearson correlation coefficient was calculated. Statistical analysis was performed using GraphPad Prism 10 version 10.1.0.
qPCR experiment
cDNA was prepared with the help of iScript gDNA clear cDNA synthesis kit (Biorad). cDNA was added to iTaq Universal SYBR Green Supermix and gene specific primers. PCR settings used were as following- 50 °C for 2 min, 95 °C for 10 min, 40 cycles of 95 °C for 15 s and 60 °C for 1 min, followed by a melting curve 95 °C for 15 sec, 60 °C for 1 min, and 95 °C for 1 sec. ROX dye was used as an internal control for evaporation. Reagent volumes used per reaction were iTaq Universal SYBR Green Supermix-5µl, forward primer 0.5 µl (10 µM stock), reverse primer 0.5 µl (10 µM stock), DNase/RNase free water 2 µl, cDNA 2 µl per reaction. GAPDH was used as housekeeping gene in these experiments. qPCR experiments were run on a Thermo QuantStudio 6 pro equipped with 384-well plate setup.
Sequences of the PCR primers used
PPARα fwd 5’ ACGATGCTGTCCTCCTTGAT 3’
PPARα rev 5’ GATGTCACAGAACGGCTTCC 3’
ACADM fwd 5’ GGAGCCCGGATTAGGGTTTA 3’
ACADM rev 5’ CGAGTTCCCAGGCTCTTTTG 3’
PPAR β/δb/d fwd 5’ TCAAGTTCAATGCGCTGGAG 3’
PPAR β/δ rev 5’ TGTCCTGGATGGCTTCTACC 3’
CPS1 fwd 5’ GTGAAGGTCTTGGGCACATC 3’
CPS1 rev 5’ TTCCACTGCAAAACTGGGTG 3’
ARG1 fwd 5’ AGAGATTATCGGAGCGCCTT 3’
ARG1 rev 5’ AGTTTTTCCAGCAGACCAGC 3’
MOT7 fwd 5’ATACAGCCTCCTCTTCGTGG 3’
MOT7 rev 5’ CAGAGTGACTGCTTTCGGTG 3’
HMCS2 fwd 5’ GCATAGATACCACCAACGCC 3’
HMCS2 rev 5’ TCGGGTAGACTGCAATGTCA 3’
FABP1 fwd 5’ CTTCTCCGGCAAGTACCAAT 3’
FABP1 rev 5’ TCCCTTTCTGGATGAGGTCC 3’
BDH1 fwd 5’ GGATTTGGGTTCTCACTGGC 3’
BDH1 rev 5’ CCACCTCTTCACTGTTGCAG 3’
GAPDH fwd 5’ CAACGACCCCTTCATTGACC 3’
GAPDH rev 5’ TCCTGGAAGATGGTGATGGG 3’
Genome-scale modeling
Constraint-Based Reconstruction and Analysis (COBRA) was used to simulate metabolic fluxes in silico. Briefly, in the COBRA method, manually curated genome-scale reconstructions are converted into mathematical models through the implementation of constraints, such as physicochemical constraints, thermodynamics, and nutritional constraints52. These mathematical models can be further tailored condition-specific through the integration of omics data such as RNA sequencing data52. Genome-scale modeling was performed in MATLAB (Mathworks, Inc.) version R2020b, using IBM CPLEX (IBM, Inc.) version 12.10 as the linear programming solver. Simulations relied on functions implemented in the COBRA Toolbox78. A genome-scale reconstruction of mouse, iMM1865, was retrieved from53. Condition-specific models for retina samples were extracted from the mouse reconstruction through the rFASTCORMICS algorithm79. Subsequently, reaction bounds were adjusted according to the transcriptomics data by a modified version of the E-Flux algorithm80. To further contextualize the model, in silico retinal media were formulated by retrieving metabolites found to be present in retina81 and translating them to the respective compound IDs in the genome-scale reconstruction.
Minimal and maximal fluxes through all reactions were computed using flux variability analysis82. By convention, positive numbers correspond to fluxes in the forward direction and negative numbers correspond to fluxes in the reverse direction of a reaction.
Statistical analyses on fluxes were performed in MATLAB as follows: Wilcoxon rank sum test was performed on all 4,917 reactions carrying flux in at least one compared sample-specific model, followed by correction for false positive rate. Four comparisons were performed, namely, BALBc/ByJ HO vs. NO, C57BL/6J HO vs. NO, BALBc/ByJ HO vs. C57BL/6J HO, and BALBc/ByJ NO vs. C57BL/6J NO. Statistical analysis of lipidomics was performed through Kruskal-Wallis test followed by correction for false positive rate. The calculation of Spearman correlations between all 4,917 reactions carrying flux in at least one compared sample-specific model and lipidomics features that were significantly different after correction for multiple testing, and subsequent corrections for false positive rate were also performed in MATLAB.
Scripts to reproduce the in silico analysis can be found at https://github.com/almut-heinken/ngereSysBio/tree/main/retinopathyMouseModeling.
Statistics and reproducibility
n number provided in each figure corresponds to biological replicate (from different animals). Details of each statistical method used has been reported in the method section of each techniques and in the legends of the figures.
Lipidomics
Lipids were extracted as per published protocol68,83. The non-polar lipid samples were re-suspended in 35 µl of 1:1 LC/MS grade isopropanol:methanol prior to LC-MS/MS analysis, 5 µl were injected. A Cadenza 150 mm × 2 mm 3 μm C18 column (Imtakt) heated to 40 °C at 240 µl /min was used with a 1100 quaternary pump HPLC with room temperature autosampler (Agilent). Lipids were eluted over a 22 min. gradient from 32% B buffer (90% IPA/10% ACN/10 mM ammonium formate/0.1 formic acid) to 97% B. A buffer consisted of 59.9% ACN/40% water/10 mM ammonium formate/0.1% formic acid. Lipids were analyzed using a high-resolution hybrid QExactive HF Orbitrap mass spectrometer (Thermo Fisher Scientific) in DDA mode (Top 8) using positive/negative ion polarity switching. DDA data were acquired from m/z 225-1450 in MS1 mode, and the resolution was set to 70,000 for MS1 and 35,000 for MS2. MS1 and MS2 target values were set to 5e5 and 1e6, respectively. Lipidomic MS/MS data were analyzed for both identification and relative quantification using LipidSearch 4.2 software (Thermo Scientific) by searching internal and LipidMaps databases. Statistical analysis in Figs. 6b, c was performed on MetaboAnalyst 6.0 online tool84.
Free-fatty acid measurement on GCMS
Free fatty acids were extracted by adding together 10 µl of plasma, 10 µl of 1 mM tricarballylic acid prepared in water and 80 µl of methanol. The samples were incubated at −80 °C for at least 20 min followed by incubation of ice for 20 min, and then centrifugation at 15,000 × g at 4 °C for 5 min. Forty microliters of supernatant was dried in a fresh tube and 10 µl of [13C8] octanoate (stock prepared by adding 1.52 µl per 1 ml of 80% methanol) was added. Although the majority of octanoate extracted with this method is a free form of octanoate, there can still be traces of glycerol-bound octanoate that is freed during processing. Outmost care was take to avoid degradation of metabolites by keeping them on ice at all times during preparation. Dried samples were derivatized by adding together 25 µl of 2.5% of Pentafluorobenzyl bromide and 25 µl of 2.5% N-N-Diisopropylethylamine, both stocks prepared in acetonitrile. Samples were incubated at room temperature for 30 min, followed by drying in vacuum centrifuge. Samples were resuspended in 100 µl of isooctane and injected into the GCMS. Negative chemical ionization method was designed to prepare these samples. DB5ms-30 m × 250 µm × 0.25 µm, with column gradient of 60 °C for 1 min followed by 10 °C/min of temperature ramp to 325 °C, followed by 10 min hold at 325 °C, was used. Gain factor to amplify the single was used 6 min-5×, 14 min-1×, 14.40-5× was used.
Software used: Figures were prepared using BioRender.com. Data was plotted using Graphpad prism. Fig. 6a was created using R version 4.3.0.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
All the sequencing has been deposited to GEO and have been assigned accession number GSE281854 and GSE281855. The numerical source data for all graphs in the manuscript can be found in supplementary data 1 file. Any additional questions pertaining to the data presented in this publication is available from the corresponding author. PCA and PLS-DA analysis were performed online using MetaboAnalyst 6.085.
References
Smith, L. E., Hard, A. L. & Hellström, A. The biology of retinopathy of prematurity: how knowledge of pathogenesis guides treatment. Clin. Perinatol. 40, 201–214 (2013).
Singh, C. et al. Serine and 1-carbon metabolism are required for HIF-mediated protection against retinopathy of prematurity. JCI Insight 4, e129398 (2019).
Sears, J. E., Hoppe, G., Ebrahem, Q. & Anand-Apte, B. Prolyl hydroxylase inhibition during hyperoxia prevents oxygen-induced retinopathy. Proc. Natl Acad. Sci. USA 105, 19898–19903 (2008).
Giusto, K., Wanczyk, H., Jensen, T. & Finck, C. Hyperoxia-induced bronchopulmonary dysplasia: better models for better therapies. Dis. Model Mech. 14, dmm047753 (2021).
Amadou, C. et al. Association between birth weight, preterm birth, and nonalcoholic fatty liver disease in a community-based cohort. Hepatology 76, 1438–1451 (2022).
Singh, C. Metabolism and Vascular Retinopathies: Current Perspectives and Future Directions. Diagnostics 12, 903 (2022).
Hu, J. et al. Reactivation of retinopathy of prematurity after bevacizumab injection. Arch. Ophthalmol. 130, 1000–1006 (2012).
Mintz-Hittner, H. A., Geloneck, M. M. & Chuang, A. Z. Clinical Management of Recurrent Retinopathy of Prematurity after Intravitreal Bevacizumab Monotherapy. Ophthalmology 123, 1845–1855 (2016).
Ling, K. P. et al. Rates and risk factors for recurrence of retinopathy of prematurity after laser or intravitreal anti-vascular endothelial growth factor monotherapy. Retina 40, 1793–1803 (2020).
Hansen, E. D. & Hartnett, M. E. A review of treatment for retinopathy of prematurity. Expert Rev. Ophthalmol. 14, 73–87 (2019).
Unterluggauer, H. et al. Premature senescence of human endothelial cells induced by inhibition of glutaminase. Biogerontology 9, 247–259 (2008).
Eelen, G., de Zeeuw, P., Simons, M. & Carmeliet, P. Endothelial cell metabolism in normal and diseased vasculature. Circ. Res. 116, 1231–1244 (2015).
Kim, B., Li, J., Jang, C. & Arany, Z. Glutamine fuels proliferation but not migration of endothelial cells. EMBO J. 36, 2321–2333 (2017).
Wang, Z. et al. Amino acid transporter SLC38A5 regulates developmental and pathological retinal angiogenesis. eLife 11, e73105 (2022).
Narayanan, S. P. et al. Arginase 2 deficiency reduces hyperoxia-mediated retinal neurodegeneration through the regulation of polyamine metabolism. Cell Death Dis. 5, e1075 (2014).
Singh, C. et al. Hyperoxia Inhibits Proliferation of Retinal Endothelial Cells in a Myc-Dependent Manner. Life 11, 614 (2021).
Mentch, S. J. & Locasale, J. W. One-carbon metabolism and epigenetics: understanding the specificity. Ann. N. Y. Acad. Sci. 1363, 91–98 (2016).
Pegg, A. E. Mammalian polyamine metabolism and function. IUBMB Life 61, 880–894 (2009).
de Vasconcelos, D. A. A. et al. Oral L-glutamine pretreatment attenuates skeletal muscle atrophy induced by 24-h fasting in mice. J. Nutritional Biochem. 70, 202–214 (2019).
Römisch-Margl, W. et al. Procedure for tissue sample preparation and metabolite extraction for high-throughput targeted metabolomics. Metabolomics 8, 133–142 (2012).
van der Valk, J. et al. Optimization of chemically defined cell culture media – Replacing fetal bovine serum in mammalian in vitro methods. Toxicol. Vitr. 24, 1053–1063 (2010).
Hamosh, M. Lipid metabolism in premature infants. Biol. Neonate 52, 50–64 (1987).
Bitman, J., Wood, L., Hamosh, M., Hamosh, P. & Mehta, N. R. Comparison of the lipid composition of breast milk from mothers of term and preterm infants. Am. J. Clin. Nutr. 38, 300–312 (1983).
Pereyra, A. S., McLaughlin, K. L., Buddo, K. A. & Ellis, J. M. Medium-chain fatty acid oxidation is independent of l-carnitine in liver and kidney but not in heart and skeletal muscle. Am. J. Physiol. Gastrointest. Liver Physiol. 325, G287–g294 (2023).
Send, S. R. Nutritional Management of Cholestasis. Clin. Liver Dis. 15, 9–12 (2020).
Lester, R. Bile acid metabolism in the fetus and newborn. Ciba Found Symp., 70, 99–115 (1979).
Fu, Z. et al. FGF21 via mitochondrial lipid oxidation promotes physiological vascularization in a mouse model of Phase I ROP. Angiogenesis 26, 409–421 (2023).
Ritter, M. R. et al. Myeloid progenitors differentiate into microglia and promote vascular repair in a model of ischemic retinopathy. J. Clin. Invest. 116, 3266–3276 (2006).
van Wijngaarden, P., Brereton, H. M., Coster, D. J. & Williams, K. A. Genetic influences on susceptibility to oxygen-induced retinopathy. Invest. Ophthalmol. Vis. Sci. 48, 1761–1766 (2007).
van Wijngaarden, P., Brereton, H. M., Gibbins, I. L., Coster, D. J. & Williams, K. A. Kinetics of strain-dependent differential gene expression in oxygen-induced retinopathy in the rat. Exp. Eye Res. 85, 508–517 (2007).
O’Bryhim, B. E., Radel, J., Macdonald, S. J. & Symons, R. C. The genetic control of avascular area in mouse oxygen-induced retinopathy. Mol. Vis. 18, 377–389 (2012).
Singh, C. et al. Hyperoxia induces glutamine-fuelled anaplerosis in retinal Müller cells. Nat. Commun. 11, 1277 (2020).
Maurer, G. D. et al. Differential utilization of ketone bodies by neurons and glioma cell lines: a rationale for ketogenic diet as experimental glioma therapy. BMC Cancer 11, 315 (2011).
Zhuang, Y.-S. et al. Neuroprotective mechanisms of OXCT1 via the SIRT3-SOD2 pathway after traumatic brain injury. Brain Res. 1808, 148324 (2023).
Holm, M. et al. Systemic Inflammation-Associated Proteins and Retinopathy of Prematurity in Infants Born Before the 28th Week of Gestation. Invest. Ophthalmol. Vis. Sci. 58, 6419–6428 (2017).
Smith, L. E. IGF-1 and retinopathy of prematurity in the preterm infant. Biol. Neonate 88, 237–244 (2005).
Yu, H. et al. Serum concentrations of cytokines in infants with retinopathy of prematurity. Apmis 122, 818–823 (2014).
Liegl, R., Löfqvist, C., Hellström, A. & Smith, L. E. IGF-1 in retinopathy of prematurity, a CNS neurovascular disease. Early Hum. Dev. 102, 13–19 (2016).
Struik, D., Dommerholt, M. B. & Jonker, J. W. Fibroblast growth factors in control of lipid metabolism: from biological function to clinical application. Curr. Opin. Lipido. 30, 235–243 (2019).
Clemmons, D. R. Metabolic actions of insulin-like growth factor-I in normal physiology and diabetes. Endocrinol. Metab. Clin. North Am. 41, 425–443 (2012).
Geng, L., Lam, K. S. L. & Xu, A. The therapeutic potential of FGF21 in metabolic diseases: from bench to clinic. Nat. Rev. Endocrinol. 16, 654–667 (2020).
BonDurant, L. D. et al. FGF21 Regulates Metabolism Through Adipose-Dependent and -Independent Mechanisms. Cell Metab. 25, 935–944.e934 (2017).
Fu, Z., Gong, Y., Löfqvist, C., Hellström, A. & Smith, L. E. Review: adiponectin in retinopathy. Biochim. Biophys. Acta 1862, 1392–1400 (2016).
Frost, S. C. & Wells, M. A. A comparison of the utilization of medium and long-chain fatty acids for oxidation and ketogenesis in the suckling rat: In vivo and in vitro studies. Arch. Biochem. Biophys. 211, 537–546 (1981).
Zhou, B. et al. Central FGF21 production regulates memory but not peripheral metabolism. Cell Rep. 40, 111239 (2022).
Hondares, E. et al. Hepatic FGF21 expression is induced at birth via PPARalpha in response to milk intake and contributes to thermogenic activation of neonatal brown fat. Cell Metab. 11, 206–212 (2010).
Flippo, K. H. & Potthoff, M. J. Metabolic Messengers: FGF21. Nat. Metab. 3, 309–317 (2021).
Martínez-Garza, Ú. et al. Fibroblast Growth Factor 21 and the Adaptive Response to Nutritional Challenges. Int. J. Mol. Sci. 20, 4692 (2019).
Maida, A. et al. A liver stress-endocrine nexus promotes metabolic integrity during dietary protein dilution. J. Clin. Invest. 126, 3263–3278 (2022).
Olivares, S. & Henkel, A. S. Induction of fibroblast growth factor 21 does not require activation of the hepatic X-box binding protein 1 in mice. Mol. Metab. 6, 1616–1624 (2017).
De Sousa-Coelho, A. L., Marrero, P. F. & Haro, D. Activating transcription factor 4-dependent induction of FGF21 during amino acid deprivation. Biochem J. 443, 165–171 (2012).
O’Brien, E. J., Monk, J. M. & Palsson, B. O. Using Genome-scale Models to Predict Biological Capabilities. Cell 161, 971–987 (2015).
Khodaee, S., Asgari, Y., Totonchi, M. & Karimi-Jafari, M. H. iMM1865: A New Reconstruction of Mouse Genome-Scale Metabolic Model. Sci. Rep. 10, 6177 (2020).
Noronha, A. et al. The Virtual Metabolic Human database: integrating human and gut microbiome metabolism with nutrition and disease. Nucleic Acids Res. 47, D614–D624 (2019).
Osorio-Paz, I., Sanchez-Chavez, G. & Salceda, R. Control of glycogen content in retina: allosteric regulation of glycogen synthase. PLoS One 7, e30822 (2012).
Inague, A. et al. Oxygen-induced pathological angiogenesis promotes intense lipid synthesis and remodeling in the retina. iScience 26, 106777 (2023).
Hellström, A. et al. Association of Docosahexaenoic Acid and Arachidonic Acid Serum Levels With Retinopathy of Prematurity in Preterm Infants. JAMA Netw. Open 4, e2128771 (2021).
Beauloye, C., Horman, S. & Bertrand, L. Even is better than odd: one fat may conceal another. Am. J. Physiol. Heart Circ. Physiol. 309, H1112–H1114 (2015).
Sklirou, E., Alodaib, A. N., Dobrowolski, S. F., Mohsen, A. A. & Vockley, J. Physiological Perspectives on the Use of Triheptanoin as Anaplerotic Therapy for Long Chain Fatty Acid Oxidation Disorders. Front Genet 11, 598760 (2020).
Kajimoto, M. et al. Differential effects of octanoate and heptanoate on myocardial metabolism during extracorporeal membrane oxygenation in an infant swine model. Am. J. Physiol. Heart Circ. Physiol. 309, H1157–H1165 (2015).
Brunengraber, H. & Roe, C. R. Anaplerotic molecules: Current and future. J. Inherit. Metab. Dis. 29, 327–331 (2006).
Ebert, D., Haller, R. G. & Walton, M. E. Energy contribution of octanoate to intact rat brain metabolism measured by 13C nuclear magnetic resonance spectroscopy. J. Neurosci. 23, 5928–5935 (2003).
Tan, K. N., McDonald, T. S. & Borges, K. in Bioactive Nutraceuticals and Dietary Supplements in Neurological and Brain Disease (eds. R.R. Watson & V.R. Preedy) 461–473 (Academic Press, 2015).
Roe, C. R. & Brunengraber, H. Anaplerotic treatment of long-chain fat oxidation disorders with triheptanoin: Review of 15 years Experience. Mol. Genet Metab. 116, 260–268 (2015).
Pearsall, E. A. et al. PPARα is essential for retinal lipid metabolism and neuronal survival. BMC Biol. 15, 113 (2017).
Tezze, C., Romanello, V. & Sandri, M. FGF21 as Modulator of Metabolism in Health and Disease. Front. Physiol. 10, 419 (2019).
Fisher, F. M. & Maratos-Flier, E. Understanding the Physiology of FGF21. Annu. Rev. Physiol. 78, 223–241 (2016).
Singh, C. et al. ChREBP is activated by reductive stress and mediates GCKR-associated metabolic traits. Cell Metab. 36, 144–158.e147 (2024).
Liu, J. et al. The omega-3 hydroxy fatty acid 7(S)-HDHA is a high-affinity PPARα ligand that regulates brain neuronal morphology. Sci. Signal. 15, eabo1857 (2022).
Chen, Y. et al. Therapeutic Effects of PPARα Agonists on Diabetic Retinopathy in Type 1 Diabetes Models. Diabetes 62, 261–272 (2012).
Ventura, G. et al. Overexpression of ornithine aminotransferase: consequences on amino acid homeostasis. Br. J. Nutr. 101, 843–851 (2009).
Smith, L. E. et al. Oxygen-induced retinopathy in the mouse. Invest Ophthalmol. Vis. Sci. 35, 101–111 (1994).
Jidigam, V. K. et al. Neuronal Bmal1 regulates retinal angiogenesis and neovascularization in mice. Commun. Biol. 5, 792 (2022).
Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat. Biotechnol. 37, 907–915 (2019).
Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2013).
Anders, S. & Huber, W. Differential expression analysis for sequence count data. Genome Biol. 11, R106 (2010).
Haslam, D. E. et al. Stability and reproducibility of proteomic profiles in epidemiological studies: comparing the Olink and SOMAscan platforms. Proteomics 22, e2100170 (2022).
Heirendt, L. et al. Creation and analysis of biochemical constraint-based models using the COBRA Toolbox v.3.0. Nat. Protoc. 14, 639–702 (2019).
Pacheco, M. P. et al. Identifying and targeting cancer-specific metabolism with network-based drug target prediction. EBioMedicine 43, 98–106 (2019).
Colijn, C. et al. Interpreting expression data with metabolic flux models: predicting Mycobacterium tuberculosis mycolic acid production. PLoS Comput. Biol. 5, e1000489 (2009).
Tomita, Y. et al. Vitreous metabolomics profiling of proliferative diabetic retinopathy. Diabetologia 64, 70–82 (2021).
Gudmundsson, S. & Thiele, I. Computationally efficient flux variability analysis. BMC Bioinforma. 11, 489 (2010).
Cajka, T. & Fiehn, O. LC-MS-Based Lipidomics and Automated Identification of Lipids Using the LipidBlast In-Silico MS/MS Library. Methods Mol. Biol. 1609, 149–170 (2017).
Pang, Z. et al. MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Res. 49, W388–W396 (2021).
Xia, J. & Wishart, D. S. Web-based inference of biological patterns, functions and pathways from metabolomic data using MetaboAnalyst. Nat. Protoc. 6, 743–760 (2011).
Acknowledgements
This project is supported by a National Institute of Health/National Eye Institute grants R21EY033046 (CS), R01EY032502 (GG) and by NIH grants 5P01CA120964 (J.M.A.), 5P30CA006516 (J.M.A.), 5P30DK043351 and 1P30DK135043. A.H. received funding from the Agence Nationale de la Recherche (ANR) under the decree 2021-1710. This publication was originally submitted while the corresponding author was affiliated to Mass General Hospital and Harvard Medical School. I want to thank my mentor Dr. Russell Goodman who helped immensely in making sure that I ran this project independently and yet supported in each step of this study. I want to thank Dr. Nirajan Shrestha for proofreading the manuscript and valuable suggestions. I want to additionally thank Dr. Fumito Ichinose and Dr. Eizo Marutani who allowed me to use their hyperoxia chamber for initial experiments. I want to thank Dr. Towia Libermann and Dr. Simon Dillon BIDMC for their help in SOMAscan measurements.
Author information
Authors and Affiliations
Contributions
A.H. performed constraint-based modeling and wrote manuscript. C.S. conceived the idea, performed wet-lab experiments, analyzed the data, and wrote the manuscript. G.G. interpreted the data and helped write manuscript. J.M.A. measured lipidomics samples on LCMS and analyzed the LCMS data. All authors approved the final version of the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Communications Biology thanks James Hurley and the other, anonymous, reviewer for their contribution to the peer review of this work. Primary Handling Editors: Joao Valente. A peer review file is available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Heinken, A., Asara, J.M., Gnanaguru, G. et al. Systemic regulation of retinal medium-chain fatty acid oxidation repletes TCA cycle flux in oxygen-induced retinopathy. Commun Biol 8, 25 (2025). https://doi.org/10.1038/s42003-024-07394-w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s42003-024-07394-w
