O-GlcNAcylation alters the selection of mRNAs for translation and promotes 4E-BP1–dependent mitochondrial dysfunction in the retina

Diabetes promotes the posttranslational modification of proteins by O-linked addition of GlcNAc (O-GlcNAcylation) to Ser/Thr residues of proteins and thereby contributes to diabetic complications. In the retina of diabetic mice, the repressor of mRNA translation, eIF4E-binding protein 1 (4E-BP1), is O-GlcNAcylated, and sequestration of the cap-binding protein eukaryotic translation initiation factor (eIF4E) is enhanced. O-GlcNAcylation has also been detected on several eukaryotic translation initiation factors and ribosomal proteins. However, the functional consequence of this modification is unknown. Here, using ribosome profiling, we evaluated the effect of enhanced O-GlcNAcylation on retinal gene expression. Mice receiving thiamet G (TMG), an inhibitor of the O-GlcNAc hydrolase O-GlcNAcase, exhibited enhanced retinal protein O-GlcNAcylation. The principal effect of TMG on retinal gene expression was observed in ribosome-associated mRNAs (i.e. mRNAs undergoing translation), as less than 1% of mRNAs exhibited changes in abundance. Remarkably, ∼19% of the transcriptome exhibited TMG-induced changes in ribosome occupancy, with 1912 mRNAs having reduced and 1683 mRNAs having increased translational rates. In the retina, the effect of O-GlcNAcase inhibition on translation of specific mitochondrial proteins, including superoxide dismutase 2 (SOD2), depended on 4E-BP1/2. O-GlcNAcylation enhanced cellular respiration and promoted mitochondrial superoxide levels in WT cells, and 4E-BP1/2 deletion prevented O-GlcNAcylation–induced mitochondrial superoxide in cells in culture and in the retina. The retina of diabetic WT mice exhibited increased reactive oxygen species levels, an effect not observed in diabetic 4E-BP1/2–deficient mice. These findings provide evidence for a mechanism whereby diabetes-induced O-GlcNAcylation promotes oxidative stress in the retina by altering the selection of mRNAs for translation.

retina of diabetic WT mice exhibited increased reactive oxygen species levels, an effect not observed in diabetic 4E-BP1/2deficient mice. These findings provide evidence for a mechanism whereby diabetes-induced O-GlcNAcylation promotes oxidative stress in the retina by altering the selection of mRNAs for translation.
Despite recent advances in therapeutics, diabetic retinopathy remains the most frequent cause of new cases of blindness among adults aged 20 -74 in developed countries (1). The principal underlying cause of diabetic retinopathy is hyperglycemia, as intensive glycemic control is associated with a reduction in both the onset and progression of neurovascular complications in the retina (2). Nevertheless, the molecular mechanisms whereby hyperglycemia causes neurovascular dysfunction are incompletely understood. One potential mechanism that is largely unexplored in the retina is altered gene expression in response to enhanced flux through the nutrient-and stresssensing hexosamine biosynthetic pathway (HBP). 3 Hyperglycemia and diabetes promote flux through the HBP, which converts glucose to UDP GlcNAc. Posttranslational modification of cellular proteins by enzymatic O-linked addition of the monosaccharide GlcNAc to Ser or Thr residues (O-GlcNAcylation) contributes to the pathophysiology of diabetes (3), and evidence supports a role for O-GlcNAc signaling in diabetic retinopathy (4). O-GlcNAcylation of proteins has been observed to alter function through a number of mechanisms, including changes in subcellular localization, proteinprotein interactions, enzymatic activity, and degradation rates (5)(6)(7)(8)(9). Protein O-GlcNAcylation is catalyzed by the enzyme O-GlcNAc transferase (OGT), whereas the enzyme O-GlcNAcase (OGA) catalyzes removal of the modification. These enzymes dynamically cycle GlcNAc residues on and off of proteins in a manner that is more reminiscent of protein phosphorylation compared with more stable forms of glycosylation (10).
Enhanced O-GlcNAcylation of retinal proteins has been observed in both diabetic Akita mice (11) and streptozotocintreated mice (12). Moreover, both of these mouse models of type 1 diabetes exhibit elevated expression of OGT. OGT-positive cells localize predominantly in the inner nuclear layer of the retina but also in the inner plexiform layer, ganglion cell layer, and photoreceptor inner segment. The O-GlcNAc cycling enzymes OGT and OGA strongly associate with cytosolic ribosomes, suggesting that they play an important role in regulating mRNA translation (13). In retinal pericytes exposed to hyperglycemic conditions, O-GlcNAcylated proteins are most heavily represented in the functional category "Protein Synthesis and Processing" (14). Indeed, O-GlcNAcylation has been detected on 34 of the ϳ80 proteins that compose the mammalian ribosome (13). Among translation initiation factors, subunits of eIF2 (15), eIF3 (15,16), eIF4 (17,18), eIF5 (15), poly(A)-binding protein (19), and 4E-BP1 (20) are all O-GlcNAcylated; however, the functional consequence of these modifications on gene expression remains to be established.
Recruitment of ribosomes to mRNA is altered in response to hyperglycemic conditions (21) or overexpression of OGT (13). In the retina of diabetic rodents, global rates of protein synthesis are reduced; however, all mRNAs are not affected equally (22). Diabetes promotes ribosome association for a number of retinal mRNAs, including the synaptic proteins Snap25 and snaptophysin 1 (22). Importantly, hyperglycemia also promotes translation of the mRNA encoding the proangiogenic cytokine vascular endothelial growth factor (VEGF) (23). VEGF concentrations are elevated in the eyes of patients with diabetic macular edema, and the cytokine is considered to play a causal role in the proliferative stages of diabetic retinopathy. In the retina of mice lacking the translational repressor 4E-BP1, VEGF expression remains unchanged in response to diabetes (24), and diabetes-induced visual dysfunction is delayed (25).
4E-BP1 regulates the selection of mRNAs for translation through binding and sequestration of the cap-binding protein eIF4E, which recognizes the 7-methyl-GTP cap structure found at the 5Ј terminus of all eukaryotic mRNAs (26). In the retina of diabetic mice, hyperglycemia promotes 4E-BP1 O-GlcNAcylation and binding to eIF4E (25). Variation in the availability of eIF4E serves as a critical regulator of gene expression patterns, as mRNAs exhibit a range of dependence on eIF4E for translation. In fact, some messages, such as the one that encodes VEGF, contain RNA elements that facilitate ribosome recruitment and translation independent of eIF4E and the 5Ј cap structure (i.e. cap-independent translation) (27). Although 4E-BP1 reduces eIF4E availability and cap-dependent translation, mRNAs that are capable of recruiting ribosomal subunits independent of the 5Ј cap often exhibit an increase in their relative rates of translation because of reduced competition for binding (21). Thus, altered selection of mRNAs for translation represents an intriguing target for novel clinical therapies that attempt to address the underlying molecular cause of DR.
In this study, we evaluated the hypothesis that enhanced O-GlcNAcylation alters retinal gene expression. We used nextgeneration sequencing to evaluate changes in retinal mRNA abundance and performed ribosome profiling to reveal which of those mRNAs were actually being translated into proteins. This technique goes beyond more commonly used methods for assessing gene expression by isolating and sequencing nuclease-resistant 28-nt ribosome-protected mRNA fragments (RPFs). Although a few genes exhibited altered mRNA abundance, the principal effect of enhanced O-GlcNAcylation on retinal gene expression was observed at the level of mRNA translation. Overall, the findings provide new insights into the impact of O-GlcNAcylation on retinal gene expression and identify a molecular network of translationally regulated mRNAs that potentially underlie dysfunctional mitochondrial respiration and superoxide production in diabetic retinopathy.

O-GlcNAcase inhibition alters mRNA translation
To evaluate the effect of enhanced protein O-GlcNAcylation on retinal gene expression, mice were administered the O-GlcNAcase inhibitor thiamet G (TMG). Twenty-four hours later, global protein O-GlcNAcylation levels were enhanced in the retina (Fig. 1A). To evaluate the impact of TMG on gene expression, we used next-generation sequencing to assess changes in retinal mRNA abundance and ribosome-bound mRNAs undergoing translation via RNA-Seq and ribosome profiling (Ribo-Seq), respectively (Fig. 1B). Overall, sequencing reads in the total mRNA and RPF that was obtained following nuclease digestion and ribosome isolation from whole retina were mapped to principal transcripts of 19,411 genes (Table  S1). For both TMG and PBS administration, replicates from two independent runs were highly similar in total mRNA abundance (R 2 Ն 0.95, Fig. S1A) and RPF (R 2 Ն 0.95, Fig. S1B). In response to TMG, the majority of transcripts observed in the retina did not exhibit a change in total mRNA abundance or RPF. We observed minimal differences in mean mRNA abundances from the retina of PBS control versus TMG-exposed mice (Fig. 1C, R 2 ϭ 0.96). In total, 148 retinal mRNAs exhibited a change in abundance 24 h after administration of TMG compared with vehicle. Alternatively, the principal effect of TMG on retinal gene expression was observed at the level of mRNA translation (R 2 ϭ 0.74, Fig. 1D). Specifically, ϳ19% of the transcriptome showed significant changes in ribosome density, as 1683 mRNAs exhibited increased and 1912 mRNAs exhibited reduced ribosome association with TMG administration (Fig.  1E, Table S1). To assess the effects on mRNA translational efficiency (TE), we compared the relative RPF and total mRNA abundance observations between retinas of mice receiving TMG and vehicle (Fig. 1F, Table S1) using the Riborex package (28). Although the TE of most mRNAs was not different, 3185 mRNAs exhibited altered TE in the retina of mice receiving TMG compared with vehicle (adjusted p Յ 0.01, Table S1). Specifically, 1360 mRNAs had enhanced TE, and 1825 mRNAs had attenuated TE in the retinas of mice receiving TMG compared with vehicle. For orthogonal validation, we performed polysome fractionation by sucrose density gradient centrifuga-  (D), respectively. E, changes in mRNA abundance and mRNA translation following TMG administration were compared. Translational efficiency as assessed by mean ribosome density was compared in mice administered either TMG or PBS vehicle. In C, D, and F, significant differences (adjusted p Յ 0.01) are indicated in red. In E, significant differences are indicated in purple (translation), green (transcription), or blue (both translation and transcription). TPM, transcripts per million mapped reads.

O-GlcNAcase inhibition alters retinal mRNA translation
tion and analyzed mRNA distribution profiles. Actively translating mRNAs associate with multiple ribosomes to form heavy polysomes, whereas poorly translated mRNAs are associated with few or no ribosomes and localize to the subpolysomal fraction. Consistent with the results of next-generation sequencing ( Fig. 2A), mRNAs encoding Nr6a1 (nuclear receptor subfamily 6 group A member 1) and Fam122a (family with sequence similarity 122A) dramatically shifted from the polysome fraction so that they were almost exclusively observed in the subpolysomal fraction in the retinas of mice administered TMG (Fig. 2, B-D). In addition, mRNAs encoding Elmsan1 (ELM2 and Myb/SANT domain-containing 1), Mettl14 (methyltransferase-like 14), Nat8l (N-acetyltransferase 8 -like), and IL1Rap (interleukin 1 receptor accessory protein) exhibited shifts with polysome profiling that supported the results obtained from ribosome profiling (Fig. 2D).

O-GlcNAcase inhibition alters the translation of mitochondrial proteins
To assess the potential impact of the observed gene expression changes, functional relationships between genes that exhibited altered TE in the retina following O-GlcNAcase inhibition were investigated. The top canonical pathways identified by Ingenuity Pathways Analysis as being altered by TMG included acute phase response signaling, pattern recognition receptors, coagulation system, and LXR/RXR activation (Table  S3). This analysis also identified toxicologic list pathways associated with mitochondrial dysfunction and oxidative stress (Fig.  S2). GO enrichment analysis showed our gene set to be enriched 1.66-fold for genes associated with the annotation mitochondrion organization and 1.55-fold for genes associated with the annotation mitochondrial membrane (Table S3). Topscoring mRNAs encoding mitochondrial proteins with TMGinduced changes in translation are listed in Fig. 3A and Table  S4.

O-GlcNAcase inhibition regulates translation of mitochondrial proteins via 4E-BP1
4E-BP-dependent translational control has been implicated previously in regulating the expression of mitochondrion-re-lated proteins (29). Moreover, we previously demonstrated that elevated O-GlcNAcylation promotes sequestration of eIF4E by 4E-BP1 in cells in culture (20). To determine whether enhanced O-GlcNAcylation had a similar effect in the retina, we performed immunoprecipitation of eIF4E from retinal lysates. Retinas from mice administered TMG exhibited enhanced interaction of 4E-BP1 and reduced interaction of eIF4G with eIF4E ( Fig. 3, B-D). We previously demonstrated that hyperglycemia promotes 4E-BP1 O-GlcNAcylation and alters the translation of specific mRNAs in a manner that depends on 4E-BP1 (21). To evaluate the role of 4E-BP1 O-GlcNAcylation in regulating the translation of mRNAs encoding mitochondrial proteins, WT and 4E-BP1/2 (Eif4ebp1; Eif4ebp2) knockout mice were administered TMG. In the retinas of both WT and 4E-BP1/2-deficient mice, TMG enhanced global protein O-GlcNAcylation (Fig. 3E). Consistent with Fig. 3, B-D, TMG promoted 4E-BP1 co-immunoprecipitation with eIF4E in the retina of WT mice, whereas the protein could not be detected in 4E-BP1/2-deficient mice (Fig. 3, E and F). Next-generation sequencing demonstrated down-regulated translation of mRNAs encoding the mitochondrial proteins Tufm (Tu translation elongation factor, mitochondrial), Mrpl47 (mitochondrial ribosomal protein L47), and SOD2 (superoxide dismutase 2) in the retina of WT mice with enhanced protein O-GlcNAcylation (Fig. 3A, Table S1). In polysome profiles, Tufm and Mrpl47 were observed in the polysomal fraction obtained from the retina of WT mice receiving PBS (Fig. 3, G and H, respectively). TMG caused both mRNAs to shift almost exclusively into the subpolysomal fraction. In the retina of 4E-BP1/2-deficient mice, the ratio of Tufm and Mrpl47 observed in the polysomal fraction was similar with PBS and TMG administration. Similarly, the mRNA encoding SOD2 shifted out of the polysomal fraction with enhanced O-GlcNAcylation in WT mice, but a similar effect was not observed in mice deficient for 4E-BP1/2 (Fig. 3I).

O-GlcNAcase inhibition promotes mitochondrial ROS via 4E-BP1
To determine whether increased O-GlcNAcylation could affect mitochondrial respiration in a manner that depends on

O-GlcNAcase inhibition alters retinal mRNA translation
4E-BP1, cultures of WT and 4E-BP1/2-deficient mouse embryonic fibroblasts (MEFs) were exposed to TMG. In both WT and 4E-BP1/2-deficient MEFs, exposure to TMG enhanced protein O-GlcNAcylation (Fig. 4A) and altered cellular respiration (Fig. 4B). TMG increased the basal oxygen consumption rate (OCR) (Fig. 4C), maximal OCR (Fig. 4D), and ATP production (Fig. 4E) in both WT and 4E-BP1/2-deficient MEFs. Although TMG enhanced cellular respiration in both WT and 4E-BP1/2-deficient cells, the maximal OCR was attenuated in 4E-BP1/2-deficient MEFs in both the presence and absence of TMG compared with the WT (Fig. 4D). In addition to changes in oxidative phosphorylation, mitochondrial dysfunction can also involve the generation of reactive oxygen species (ROS). To determine whether TMG altered cellular ROS production, we assessed mitochondrial superoxide. Superoxide levels were similar in vehicle-treated WT and 4E-BP1/2-deficient MEF cultures (Fig. 4, F and G). TMG exposure increased superoxide in WT MEFs. However, the effect of TMG was absent in 4EBP1/2deficient MEFs, as superoxide levels were similar to those observed in both WT and 4E-BP1/2-deficient MEFs exposed to vehicle alone. This supports that acute elevations in O-GlcNAcylation promote mitochondrial superoxide levels via a 4E-BP1/2-dependent mechanism. To further investigate the role of 4E-BP1 in O-GlcNAcylation-induced mitochondrial superoxide levels, we restored 4E-BP1 expression in 4E-BP1/2-deficient MEFs in culture (Fig. 5A). TMG enhanced mitochondrial superoxide in cells expressing WT 4E-BP1 (Fig. 5, A and B). However, in cells expressing a 4E-BP1 variant (Y54A) that is deficient in binding to eIF4E (30), TMG did not enhance mitochondrial superoxide levels (Fig.  5, A and B). In contrast, expression of a 4E-BP1 variant with an F113A substitution that prevents mTORC1-dependent phosphorylation and promotes eIF4E binding (31) was sufficient to enhance mitochondrial superoxide levels in WT MEFs (Fig. 5, C and D).

4E-BP1/2 deletion prevents O-GlcNAc-and diabetes-induced ROS in the retina
To evaluate the role of 4E-BP1 in the effect of O-GlcNAcyaltion on the retina, WT and 4E-BP1/2-deficient mice were administered TMG. Retinal cryosections were exposed to the fluorescent ROS indicator 2,7-dichlorofluoroscein (DCF) to evaluate oxidative stress. In the retina of WT mice, TMG administration enhanced DCF fluorescence compared with vehicle (Fig. 6, A and B). Enhanced ROS levels were most obvious in the photoreceptor layers but could also be observed throughout the entire retina. DCF fluorescence was similar in the retina of WT and 4E-BP1/2-deficient mice receiving vehicle. Moreover, ROS levels were similar in the retina of 4E-BP1/2-deficient mice administered either PBS or

O-GlcNAcase inhibition alters retinal mRNA translation
TMG. Importantly, ROS levels were reduced in 4E-BP1/2deficient mice administered TMG compared with WT controls administered TMG. We previously demonstrated that 4E-BP1 O-GlcNAcylation was enhanced in the retina of streptozotocin (STZ)-induced diabetic mice (25). Consistent with previous reports (e.g. 32), STZ-induced diabetes enhanced DCF fluorescence in the retina of WT mice compared with nondiabetic controls (Fig. 6, C and D). In nondiabetic 4E-BP1/2-deficient mice, DCF fluorescence was similar to that observed in diabetic 4E-BP1/2-deficient mice. Moreover, ROS levels were reduced in diabetic 4E-BP1/2-deficient mice compared with diabetic WT mice. Overall, the results support a model wherein diabetes-induced 4E-BP1 O-GlcNAcylation alters mRNA translation in a manner that causes mitochondrial oxidative stress in the retina. Gene expression is the sum of events that regulate cellular protein concentrations, including transcription and translation, as well as mRNA and protein degradation. Historically, gene expression studies have utilized techniques such as microarrays and, more recently, RNA-Seq to assess changes in mRNA abundance. However, mRNA abundance is a relatively poor correlate for protein expression (33,34), and some studies suggest that they only account for ϳ40% of the global variation in protein expression (35,36). Translational control facilitates

O-GlcNAcase inhibition alters retinal mRNA translation
the selective recruitment of ribosomes to specific mRNAs to provide a rapid and reversible response in the expression of specific proteins to changes in nutrient supply, hormones, and stress (37, 38). Thus, an assessment of ribosome-protected mRNA fragments accounts for both transcriptional and translational changes that lead to variation in protein expression. A previous study from our laboratory demonstrated that hyperglycemic conditions alter the selection of mRNAs for translation (21). In this study, we performed genome-wide quantitative analysis of in vivo translation in the retina in combination with RNA-Seq to assess changes in retinal gene expression. This technique utilized nuclease footprinting to produce mRNA fragments that indicate which mRNAs were associated with ribosomes. The effect of O-GlcNAcase inhibition on retinal gene expression was largely observed at the level of mRNA translation. Remarkably, only a few genes exhibited changes in mRNA abundance (Ͻ1% of the observed transcripts) following TMG administration. However, ϳ19% of the transcriptome exhibited altered mRNA translation. The extremely limited transcriptional effect of O-GlcNAcase inhibition on retinal gene expression was surprising, as the activity of a number of transcription factors is regulated by O-GlcNAcylation (40). An important caveat is that retinal gene expression was assessed 24 h after administration of the O-GlcNAcase inhibitor, and thus transcriptional effects on gene expression may be more obvious after prolonged disruption of O-GlcNAc cycling. Another important consideration of our analysis is that the contribution of protein degradation to gene expression has not been considered. Protein concentrations in a cell are determined by the relative rates of synthesis and degradation. Enhanced rates of retinal protein degradation have been observed in diabetic rodents (22); however, compared with rates of synthesis, the impact of protein stability on gene expression is relatively minor (35).
In this study, changes in ribosome density were interpreted as reflecting changes in mRNA translation. Although this is a common assumption in the field, it is only accurate when there are no systemic pauses or stalls of ribosomes on transcripts. To ensure that the effects of O-GlcNAcase inhibition on ribosome density were not due to pausing/stalls, polarity scores (41) were generated to determine whether TMG skewed ribosome density toward the 5Ј or 3Ј ends of mRNA. Importantly, there were no significant changes in polarity scores between the two treatments (Wilcoxon signed-rank test with continuity correction, p ϭ 0.09, Fig. S3), indicating that the large-scale effect of TMG on mRNA translation was due to changes in initiation rates.
One of the best-characterized mechanisms for regulating translation initiation involves 4E-BP1 binding to eIF4E to prevent eIF4F complex assembly at the 5Ј end of an mRNA. We initially discovered that 4E-BP1 is directly modified by O-GlcNAcylation in the liver of diabetic rats (20). A variety of conditions that increase O-GlcNAc levels (e.g. hyperglycemia, O-GlcNAcase inhibition, or glucosamine addition to the culture medium) promote binding of 4E-BP1 with eIF4E (20,21,25). Furthermore, 4E-BP1 O-GlcNAcylation inhibits its degradation via an E3 ubiquitin ligase complex containing CUL3. Specifically, we have provided evidence for a model wherein O-GlcNAcylation and GSK3-dependent phosphorylation competitively modify 4E-BP1 at Thr-82/Ser-86 (25). Of importance to this study is that we reported previously that 4E-BP1 O-GlcNAcylation (25) and binding to eIF4E (24) are enhanced in the retina of diabetic rodents. Here we found that, in the retina of mice receiving TMG, O-GlcNAcylation levels were enhanced concomitant with an increase in co-immunoprecipitation of 4E-BP1 with eIF4E.
TMG-induced O-GlcNAcylation altered the translation of mRNAs encoding numerous mitochondrial proteins, including those involved in oxidative phosphorylation (OXPhos), mitochondrial ribosomal subunits, mitochondrion-specific tRNA

O-GlcNAcase inhibition alters retinal mRNA translation
synthetases, and subunits of the translocase of inner membrane (TIM)/translocase of outer membrane (TOM) complex. Importantly, we demonstrated that 4E-BP1/2 were critical in mediating the effects of O-GlcNAcylation on the translation of specific mRNAs encoding mitochondrial proteins and mitochondrial ROS production. This observation supports previous reports implicating 4E-BP-dependent translational control in the regulation of mitochondrial function (29,42). The mitochondrial electron transport chain compromises ϳ150 distinct proteins, including 13 hydrophobic inner membrane proteins that are encoded by the mitochondrial DNA (mtDNA) and synthesized by mitochondrial ribosomes (mitoribosomes). Alternatively, the remaining majority of OXPhos proteins are synthesized by cytosolic ribosomes and imported by the TIM/ TOM complex. Proper mitochondrial function requires rapid adaptation to metabolic conditions and involves coordinated expression from both the mtDNA and nuclear genome. Yeast ribosome profiling demonstrates that nuclear and mitochondrial encoded OXPhos mRNA abundances are not altered in concordance following a nutrient shift, but rather, synchronized protein expression is achieved via translational control (43). Thus it is likely that defects in mRNA translation would contribute to human pathology, as there are a variety of mitochondrial diseases that result from improper mitochondrial protein expression (reviewed in Ref. 44).
In recent years, O-GlcNAcylation has emerged as a critical regulator of mitochondrial function (45,46). Overexpression of the enzymes responsible for O-GlcNAc modification (i.e. OGT and OGA) alters mitochondrial protein levels, including specific proteins involved in OXPhos (45). Moreover, cells with elevated O-GlcNAc levels exhibit abnormal mitochondrial morphology and hyperpolarization (46). Numerous mitochondrial proteins are

O-GlcNAcase inhibition alters retinal mRNA translation
posttranslationally modified by O-GlcNAcylation directly in response to diabetes and hyperglycemic conditions (47,48). Tan et al. (46) have suggested previously that mitochondria undergo a two-stage adaptive process in response to O-GlcNAcase inhibition. Initially, mitochondrial proteins are O-GlcNAcylated, resulting in increased cellular respiration and enhanced ROS production (49). However, with sustained elevation in O-GlcNAcylation, mitochondria undergo transcriptional reprogramming as ATP production and ROS levels decline (46). Consistent with the previous studies, we observed increased cellular respiration and elevated mitochondrial superoxide levels in cells after 24 h of exposure to culture medium containing TMG. This study extends these previous reports by demonstrating that the synthesis of mitochondrial proteins is dramatically altered by enhanced O-GlcNAcylation in response to TMG-induced O-GlcNAcase inhibition. Although the effect of TMG on cellular respiration was observed in 4E-BP1/2-deficient cells, TMG-induced mitochondrial superoxide was absent in cells lacking 4E-BP1/2. This observation supports a role for 4E-BP1/2-dependent translational control in increased ROS production during the first stage of the adaptive process that is initiated upon disruption of O-GlcNAc cycling.
The retina is among the most metabolically active tissues in the body and therefore is heavily dependent on a constant supply of energy from oxidative phosphorylation in the mitochondria. However, in diabetes, retinal mitochondria contain damaged mtDNA (50) in association with increased capillary cell apoptosis, an abnormality that occurs prior to the onset of histopathological changes in the retina (51). The dynamic nature of mitochondria necessitates tight regulation of the organelle's shape, size, formation of cristae, and transport of molecules across the mitochondrial membranes (52). In retinas of diabetic rats, this tight control is impaired as the mitochondrial ultrastructure is disrupted (53), and the enzymes required for repairing damaged retinal mtDNA fail to access the mitochondria despite an increase in their mRNA levels (50). Furthermore, the expression of small-molecule transport proteins involved in the exchange of citric acid cycle substrates as well as transport of antioxidant molecules into the mitochondria are less abundant in the mitochondria of diabetic rats and diabetic donor eyes (53). Similarly, we report in our dataset a TMG-induced decrease in translation of the outer membrane translocase complex proteins Tomm6, Tomm40, and Tomm70a, the mitochondrial amino acid transporter Slc25a29, and the mitochondrial protein transporter Metaxin3. These findings are consistent with previous reports that suggest diabetes induces impairment in mitochondrial transport machinery (e.g. Ref. 53).
Hyperglycemic conditions provide an abundance of substrate for glucose oxidation, leading to increased electron transport chain flux and a higher voltage gradient across the mitochondrial membrane (54,55). When a critical threshold is reached, electron transfer inside complex III of the electron transport chain is obstructed, giving way to accumulation of electrons at coenzyme Q and the formation of superoxide (56,57). Studies have shown this to be problematic in diabetic retinopathy, as retinal mitochondrial superoxide levels are elevated in diabetic rodents (58), whereas the activities of the superox-ide-scavenging SOD2 (also known as Mn-SOD) and complex III of the electron transport chain are attenuated (58,59). Overproduction of superoxides by mitochondria links the principal pathways, such as the HBP, responsible for hyperglycemia-induced tissue damage (55). Notably, long-term administration of antioxidants inhibits the development of retinopathy in diabetic rats (60), as does overexpression of SOD2 (61). In the retina of mice administered TMG, SOD2 mRNA translation was attenuated compared with controls. However, in the retina of 4E-BP1/2-deficient mice administered TMG, relative SOD2 mRNA translation was similar to that observed in 4E-BP1/2deficient mice and WT controls administered vehicle alone. Thus, 4E-BP1 is necessary for the repressive effect of O-GlcNAcylation on SOD2 mRNA translation. ROS levels were elevated in the retina of diabetic WT mice or nondiabetic mice administered TMG. Moreover, 4E-BP1/2 deletion was sufficient to normalize ROS levels in the retina of diabetic mice as well as of those administered TMG to enhance O-GlcNAcylation.
Enhanced O-GlcNAcylation of retinal proteins has been observed in rodent models of both type 1 (11,12) and pre/type 2 (62) diabetes. This study provides new evidence that O-GlcNAcylation regulates mitochondrial function in the retina. We report that enhanced O-GlcNAcylation alters retinal gene expression by principally acting at the level of mRNA translation. We provide evidence that the effect of O-GlcNAcylation on translation of specific mitochondrial mRNAs depends on 4E-BP1/2 and provide functional evidence that O-GlcNAcylation enhances ROS levels in the retina in a 4E-BP1/2-dependent manner. Overall, the findings are consistent with a model wherein diabetes-induced O-GlcNAcylation acts to promote oxidative stress in the retina by altering the selection of mRNAs for translation. Because of the critical role oxidative stress plays in the development of diabetic retinopathy, pharmacological targeting of O-GlcNAcylation or 4E-BP1 binding to eIF4E may represent novel targets for therapeutic intervention.

Animals
At 4 weeks of age, male and female WT and 4E-BP1/2 double knockout C57BL/6J mice received 50 mg/kg TMG or phosphate-buffered saline as a control via intraperitoneal injection. Retinas were extracted 24 h after TMG injections. Alternatively, male mice were administered 50 mg/kg streptozotocin for 5 consecutive days to induce diabetes. Control mice were injected with equivalent volumes of sodium citrate buffer. Diabetic phenotype was confirmed by blood glucose concentration (Ͼ250 mg/dL) in fasted animals. Retinas were extracted after 4 weeks of diabetes. All procedures were approved by the Pennsylvania State College of Medicine Institutional Animal Care and Use Committee.

Protein analysis
Retinas were extracted, flash-frozen in liquid nitrogen, and later homogenized in 250 l of extraction buffer as described previously (24). The homogenate was centrifuged at 1000 ϫ g for 5 min at 4°C, and the supernatant was collected for

O-GlcNAcase inhibition alters retinal mRNA translation
analysis. A fraction of the supernatant was added to 1ϫ SDS sample buffer, boiled for 5 min, and analyzed by Western blot analysis as described previously (24). Antibodies used included O-GlcNAc (Cell Signaling Technology, catalog no. 9875), ␣-actin (Cell Signaling Technology, catalog no. 4970), and GAPDH (Santa Cruz Biotechnology, catalog no. sc-32233). Preparation of the eIF4E, 4E-BP1, and eIF4G antibodies has been described previously (63,64). Immunoprecipitations were performed by incubating supernatants of retina homogenates with monoclonal anti-eIF4E antibody as described previously (21).

Sequencing library preparation
Total RNA and RPF libraries for deep sequencing were prepared from mouse retina using the TruSeq Ribo Profile Kit (Illumina). Briefly, 10 retinas/group were Dounce-homogenized in 500 l of polysome buffer containing 10% Triton X-100, 100 mM DTT, 100 mM EGTA, 0.5 units of DNaseI, and 50 g of cycloheximide. For the RPF, 300 l of retinal homogenate was treated with 0.6 units of nuclease/optical density unit measured at A 260 and purified using an Illustra MicroSpin S-400 HR column (GE Healthcare Life Sciences). rRNA depletion was performed with a Ribo-Zero Gold rRNA Removal Kit (Illumina). Total RNA was prepared from the remaining retinal homogenate and heat-fragmented according to the manufacturer's instructions. RPF RNA and fragmented total RNA were end-repaired with TruSeq Ribo Profile PNK for 3Ј adapter ligation, and cDNA was reverse-transcribed with EpiScript RT. cDNA libraries were PAGE-purified, circularized with Circ-Ligase, and amplified by PCR. Libraries were purified with Agencourt AMPure XP beads (Beckman Coulter) and sequenced on an Illumina HiSeq 2500 to a depth of 15 million total reads/sample.

Sequence analysis
Sequencing reads were mapped to 19,411 principal transcripts for each gene in the APPRIS database (65) referenced by the GENCODE release M14 (GRCm38.p5) mouse genome. Principal transcripts for each gene were selected using the script_for_transcript_annotation from RiboViz toolkit package (65). Transcript and ribosome footprint abundances for each transcript/gene were estimated using Kallisto v0.45.0 (66). Given the small size of RPF and RNA-Seq inserts in our libraries (ϳ30 bp), Kallisto indices were generated using a lower k-mer value (Ϫk 19). Gene-specific -fold changes in RNA and footprint abundances were estimated using DESeq2 packages in R (39) using default log -fold change shrinkage options. Changes in ribosome densities (translation efficiencies) were estimated using the Riborex package in R (28). Unless otherwise specified, significant changes in RNA, RPF, and ribosome densities were estimated at an adjusted p Ͻ 0.01 (Benjamini and Hochbergadjusted p value). Functional analysis of GO categories and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were conducted using clusterProfiler packages in R (28). The data obtained from sequencing analysis were also analyzed using the Ingenuity Pathways Analysis software and Gene Ontology (GO) Enrichment Analysis (http://geneontology. org/; 4 67,68).

Polysome fractionation by sucrose density gradient centrifugation and RNA isolation
Sucrose density gradient centrifugation was employed to separate the subpolysomal from the polysomal ribosome fractions. Retinas were harvested, and 10 retinas from the respective treatment groups were pooled. Retinas were homogenized in 600 l of buffer (50 mM HEPES, 75 mM KCl, 5 mM MgCl 2 , 250 mM sucrose, 2 mM DTT, 100 g/ml cycloheximide, and 20 units/l SUPERase-In TM RNase inhibitor (Invitrogen)). Homogenates were incubated on ice for 5 min, and 75 l of Tween-deoxycholate solution (1.34 ml of Tween 20, 0.66 g of deoxycholate, and 18 ml of sterile water) was added for 15 min. Lysates were centrifuged at 1000 ϫ g for 3 min at 4°C. The resulting supernatant (550 l) was layered on a 20 -47% linear sucrose gradient (10 mM HEPES, 75 mM KCl, 5 mM MgCl 2 , and 0.5 mM EDTA) and centrifuged in a SW41 rotor at 34,000 rpm for 190 min at 4°C. Following centrifugation, the gradient was displaced upward (3 ml/min) using Fluorinert (Sigma) through a spectrophotometer, and A 254 nm was continuously recorded (chart speed, 150 cm/h). Two sucrose fractions representing the subpolysomal and polysomal portions of the gradient were collected directly into an equal volume of TRIzol reagent (Invitrogen). To improve the recovery of RNA from dense sucrose portions of the gradient, the polysomal fraction was diluted twice with RNase-free water (HyClone), and the appropriate amount of TRIzol was added. RNA was extracted using the standard manufacturer's protocol and resuspended in 14 l of RNA storage solution. An equal amount of RNA from each fraction collected was converted into cDNA using a high-capacity cDNA reverse transcription kit (Applied Biosystems) and subjected to quantitative real-time PCR using QuantiTect SYBR Green Master Mix (Qiagen). Primer sequences can be found in Table S5.

Bioenergetics analysis of OCR
The Seahorse XF e 96 extracellular flux analyzer (Seahorse Bioscience, North Billerica, MA) was used to assess mitochondrial respiration. Prior to use of the extracellular flux analyzer, cells were seeded (30,000 cells/well, following optimization of cell seeding number) into the XF96 cell culture microplate, allowed to adhere, and then exposed to medium containing 50 nM TMG for 24 h. Cell culture medium was then replaced with XF medium (Seahorse Bioscience), and cells were placed in a non-CO 2 37°C incubator for 1 h prior to the start of the extracellular flux analysis. The Cell Mito Stress Kit was applied according to the manufacturer's instructions (Seahorse Bioscience). Prior to commencement of these experiments, doseresponse curves were carried out to establish the concentration of carbonyl cyanide p-trifluoromethoxyphenylhydrazone needed to achieve maximal OCR. Additionally, dose-response curves were carried out to establish the optimal concentration of oligomycin required to inhibit oxygen consumption, as cells did not respond to the manufacturer's recommended 1 M. Both cell lines achieved a plateau of maximal OCR at 0.5 M carbonyl cyanide p-trifluoromethoxyphenylhydrazone and inhibition of OCR at 2 M oligomycin.

ROS detection
The MitoSOX Red mitochondrial superoxide indicator (Invitrogen) was used in WT and 4E-BP1/2 DKO MEFs to detect mitochondrial superoxide according to the manufacturer's instructions. Briefly, cells were seeded at 300,000 cells/dish in 35-mm, no. 1.5 coverglass, poly-D-lysinecoated dishes (MatTek Corp.). Cells were then exposed to 50 nM TMG for 24 h. 1 ml of 5 M MitoSOX diluted in Hanks Balanced Salt Solution/Ca/Mg (Life Technologies) was applied to cells. The dye-loaded cells were incubated for 10 min at 37°C in the dark and then washed with HBSS buffer. To evaluate retinal ROS levels, whole eyes were harvested from mice, embedded in optimal cutting temperature compound (OCT, Sakura Finetek), and flash-frozen. Cryosections (10 m) were fixed with 2% paraformaldehyde in PBS (pH 7.4). Cryosections were stained with 1.6 M Hoechst and 10 M DCF. Live cells and cryosections were imaged using an inverted Leica TCS SP8 confocal microscope.