Hexokinase-1 mitochondrial dissociation and protein O-GlcNAcylation drive heart failure with preserved ejection fraction

Heart failure with preserved ejection fraction (HFpEF) is a common cause of morbidity and mortality worldwide, but the underlying pathophysiology is not well-understood and treatment options are limited. Hexokinase-1 (HK1) mitochondrial-binding and protein O-GlcNAcylation are both altered in conditions with risk factors for HFpEF. Here we report a novel mouse model of HFpEF and show that HK1 mitochondrial-binding in endothelial cells (EC) is critical for the development of HFpEF. We demonstrate increased mitochondrial dislocation of HK1 in ECs from HFpEF mice. Mice with deletion of the mitochondrial-binding-domain of HK1 spontaneously develop HFpEF, and their ECs display impaired angiogenic potential. Mitochondrial-bound HK1 associates with dolichyl-diphosphooligosaccharide-protein-glycosyltransferase (DDOST) and its mitochondrial dislocation decreases protein N-glycosylation. We also show that the spatial proximity of dislocated HK1 and O-linked N-acetylglucosamine-transferase (OGT) increases protein O-GlcNAcylation by shifting the balance of the hexosamine-biosynthetic-pathway intermediate supply into the O-GlcNAcylation machinery. Pharmacological inhibition of OGT or EC-specific overexpression of O-GlcNAcase reverses angiogenic defects in ECs and the HFpEF phenotype, indicating that increased protein O-GlcNAcylation is responsible for the development of HFpEF. Our study demonstrates a new mechanism for HFpEF through HK1 cellular localization and resultant protein O-GlcNAcylation in ECs, and provides a potential new therapy for this disorder.


Introduction
Heart failure (HF) with preserved ejection fraction (HFpEF) is associated with higher all-cause mortality 1,2,3 , however, the molecular mechanisms that drive the development of HFpEF pathology remain poorly understood. HFpEF is more prevalent with older age 4 , and patients with HFpEF tend to have multiple comorbidities, such as hypertension, renal dysfunction, and diabetes 5,6 . There are few treatment options for HFpEF, and the current therapies relieve symptoms and target comorbid conditions. Evidence suggests that endothelial cell (EC) dysfunction plays a critical role in the development of HFpEF 7,8 . A recent prospective multi-center study reported that microvascular dysfunction (MVD) is prevalent in 69-81% of HFpEF patients 9 , and MVD is associated with worse diastolic dysfunction and 2.47 times increase in the major adverse cardiovascular events 10 . A positive correlation between EC dysfunction and exercise intolerance in HFpEF patients is also reported 11,12 . However, the mechanism underlying the observed EC dysfunction and the causal relationship between EC dysfunction and the development of HFpEF are not currently known. The two major risk factors for the development of HFpEF, diabetes and hypertension, are both associated with increased protein O-GlcNAcylation in ECs 13,14 . Thus, it is possible that HFpEF may be associated with increased protein O-GlcNAcylation in ECs due to the risk factors associated with the condition.
ECs highly rely on glycolysis to produce energy and to keep their physiological functions 15 , as they generate >85% of their ATP through glycolysis even in oxygen-replete conditions 16 . Hexokinases (HKs) carry out the first step in glycolysis by phosphorylating glucose to glucose-6-phosphate (G6P), committing it to various downstream metabolic pathways [17][18][19][20][21] . While its flux through glycolysis yields energy-producing intermediates, it can also be shunted into other pathways, including the pentose phosphate pathway (PPP), hexosamine biosynthesis pathway (HBP), or be converted to glucose-1phosphate for glycogen synthesis 22 . In mammals, five HK isozymes (HK1, HK2, HK3, HK4/glucokinase, and HK5/HKDC1) have been identified, each with distinct tissue expression, subcellular localization, kinetics, and substrate specificities [23][24][25] . HK1 and HK2 contain a mitochondrial binding domain (MBD) at their N-terminus, which enables outer mitochondrial membrane (OMM) binding 26 . Prior studies have proposed several HK1/HK2 mitochondrial binding functions, including regulation of opening of the permeability transition pore 27 and recruitment of apoptotic mediators 28 .
Recently, we showed that dislocation of HK1 from mitochondria in macrophages shunts glycolytic metabolites into PPP and facilitates the production of proinflammatory cytokines 29 .
In this paper, we proposed that since HFpEF is more common in patients with diabetes and hypertension, protein O-GlcNAcylation in ECs is increased in this condition. Since ECs highly rely on glycolysis for energy production and HBP is a side branch of glycolysis, we initiated our studies by assessing the levels of glycolytic enzymes in ECs. We demonstrated that HK1 is predominantly expressed in ECs compared to other cell types in the heart, and dislocates from mitochondria in the setting of HFpEF. Furthermore, we show that HK1 dislocation from mitochondria is sufficient to cause HFpEF in mice, with microvascular rarefaction preceding other pathological changes in the hearts. Isolated ECs from mice with deletion of the MBD (ΔE1HK1) display defects in their angiogenic potential, caused by rerouting the HBP intermediate supplies from N-glycosylation to the O-GlcNAcylation machinery. We also demonstrate that pharmacological inhibition of O-linked Nacetylglucosamine transferase (OGT) reverses the HFpEF phenotype. Our findings highlight the significant role of HK1 mitochondrial binding in the fate of HBP metabolites and the end products of this pathway, and provide a potential therapy for HFpEF by targeting this machinery. We next assessed the protein level of HK1 in ECs from hearts of HFpEF mouse model, and noted no difference ( Figure 1G). Since HK1 cellular localization has significant effects on cellular physiology and progression of some disorders 29,31 , we next studied the localization of HK1 in ECs in the hearts of the HFpEF mouse model. Freshly isolated ECs from NC mice showed clear colocalization with mitochondria, but a distinctly non-colocalized HK1 signal was observed in mice treated with HFD and L-NAME (Figure 1, H-J), suggesting that HK1 is dislocated from mitochondria in ECs of HFpEF hearts. To further investigate the localization of HK1 in ECs in HFpEF hearts, we performed immunogold electron microscopy (IEM) with an anti-HK1 antibody. We first confirmed that IEM worked properly by the observation of gold particles on the OMM in CMs (Supplemental Figure 3), and then showed that the number of gold particles not present on mitochondria (i.e., in the cytoplasm and nucleus) was significantly increased in ECs from mice treated with HFD_L-NAME (Figure 1, K and L). Thus, although HK1 expression levels are unaltered in HFpEF, its cellular localization changes from a predominantly mitochondrial-bound form to cytosolic and nuclear localization.

ΔE1HK1 mice spontaneously develop HFpEF with aging
To better define the consequences of HK1 dislocation from mitochondria on the progression of HFpEF, we used a mouse model with the MBD of the endogenous HK1 gene replaced with a FLAG sequence (ΔE1HK1 mice), described previously 29 . As expected, HK1 was mostly in the non-mitochondrial fraction of the ΔE1HK1 hearts (Figure 2A). Echocardiographic analysis of the hearts of these mice at 40 weeks of age showed no significant change in the systolic function or LV structure ( Figure 2B), but significantly lower e' and increased E/A and E/e' (Figure 2, C and D), increased LA area, diminished exercise tolerance (Figure 2, E and F), and no significant change in BP (Supplemental Figure 4). ΔE1HK1 mice also displayed the increased fluid volume in their lungs ( Figure 2G). Additionally, the heart of ΔE1HK1 mice displayed a higher percentage of fibrosis at 40 weeks of age ( Figure 2H). Finally, although HFD_L-NAME treatment leads to HFpEF only in male mice 30 , we noted a significant decrease in e' and exercise tolerance and an increase in E/e' and LA area in ΔE1HK1 female mice, while E/A ratio was not changed (Supplemental Figure 5). These results indicate that dislocation of HK1 from mitochondria results in overt HFpEF phenotype in male mice and most of the features of this disorder in female mice.
Additionally, we treated WT and ΔE1HK1 mice at 8 weeks of age with HFD for 20 weeks. The ΔE1HK1 mice treated with HFD displayed increased E/e', E/A, increased expression of HF markers Nppa and Nppb, and reduced exercise tolerance (Supplemental Figure 6, A-G). Thus, in addition to developing HFpEF with advanced age, ΔE1HK1 mice are also more prone to develop HFpEF in response to HFD.
To investigate the mechanism by which ΔE1HK1 developed HFpEF, we assessed isolated CM relaxation by measuring the half-time decay of Ca 2+ transient and microvascular density in the hearts of ΔE1HK1 mice at the age of 20 weeks (i.e., before they develop overt HFpEF phenotype). Although Ca 2+ decay times were comparable between isolated CMs of WT and ΔE1HK1 (Figure 2I), decreased microvascular density was noted in ΔE1HK1 hearts (Figure 2J), which indicate that microvascular rarefaction occurs before the development of HFpEF and damage to the other cell types in the hearts of ΔE1HK1 mice. Thus, EC dysfunction is likely the primary driver in the pathogenesis of the disease in these mice.

Isolated ECs from hearts of ΔE1HK1 mice display reduced angiogenic potential
We next studied the angiogenic potential of ECs from the hearts of ΔE1HK1 mice, we first showed that although the level of HK1 protein is not different (Figure 3A), it is dislocated from mitochondria in ECs from ΔE1HK1 hearts (Figure 3, B and C). Tube formation assay demonstrated that ECs from ΔE1HK1 mice have a reduced number of branching points, segments of meshes formed by tubes, and less total length of branching compared to ECs from WT hearts ( Figure 3D). We also observed less proliferation and motility in the isolated ECs from ΔE1HK1 mice, as assessed by bromodeoxyuridine (BrdU) and transwell migration assay (Figure 3, E and F). However, mitochondrial membrane potential (MMP) was preserved in EC from ΔE1HK1 mice ( Figure 3G). Together, these results suggest that ECs from ΔE1HK1 mice have reduced angiogenic potential, likely due to defects in proliferation and motility independent of MMP.

Levels of metabolites in HBP are reduced in isolated ECs from hearts of ΔE1HK1 mice
To study the mechanism for reduced angiogenesis in ECs from ΔE1HK1 mice, we next studied energy state of ECs from WT and ΔE1HK1 mice, and noted no difference, as assessed by total cellular ATP levels, energy charge calculated by the levels of adenine nucleotides, and the ratio of protein levels of the phosphorylated form of AMPKα to total AMPKα (Figure 4, A-C). Additionally, extracellular acidification rate (ECAR) and oxygen consumption rate (OCR) were similar among the groups, suggesting that glycolytic flux and the activity of the TCA cycle and electron transport chain in mitochondria are comparable between isolated ECs from WT and ΔE1HK1 mice (Figure 4

, D and E).
We next performed metabolomics in isolated ECs from WT and ΔE1HK1 mice and demonstrated that the levels of metabolites in HBP were consistently reduced in ECs from ΔE1HK1 mice, while the levels of metabolites in PPP were slightly increased (Figure 4, F-H). Most of the metabolites in the glycolysis and TCA cycle showed similar levels between WT and ΔE1HK1. To study the mechanism for the decrease in HBP metabolites in the ECs from ΔE1HK1 mice, we next performed isotope tracing metabolomics using U-13 C6-glucose, which showed similar labeling of HBP metabolites in the ECs of WT and ΔE1HK1 mice ( Figure 4I and Supplemental Figure 7). Additionally, the mRNA and protein levels of genes involved in HBP were not altered (Figure 4, J and K). Carbon-tracing studies and analysis of genes in HBP together suggest that the flux of HBP itself is not altered in ECs from WT and ΔE1HK1 mice, however, the decrease in the HBP metabolites is likely due to changes downstream of HBP, including steps related to protein N-glycosylation and O-GlcNAcylation.

Dislocation of HK1 in ECs causes a reduction in protein N-glycosylation
We first assessed protein N-glycosylation, one of the predominant consequences of HBP, in the ECs of ΔE1HK1 mice. N-glycosylation of proteins was significantly reduced in ECs from ΔE1HK1 mice, based on lectin blot analysis using Concanavalin A (ConA)-horseradish peroxidase (HRP) (Supplemental Figure 8A), and by assessing the levels of glycan on membrane proteins based on wheat germ agglutinin (WGA) intensity of fluorescence in capillaries (Supplemental Figure 8B). Despite the reduction in N-glycosylated proteins, the mRNA levels of genes involved in unfolded protein response (UPR) were not altered in the ECs of WT and ΔE1HK1 mice (Supplemental Figure  8C), suggesting the proper protein folding is preserved in spite of decreased protein N-glycosylation. We also overexpressed either full length HK1 (FL-HK1) or truncated HK1 lacking the MBD (Tr-HK1) conjugated with eGFP in HUVECs, but noted no effect on protein N-glycosylation (Supplemental Figure 8D), likely due to the saturation of the glycosylation process under basal conditions. Analysis of proteomic studies on immunoprecipitated samples with an anti-GFP antibody in HepG2 cells overexpressing either FL-HK1 or Tr-HK1 29 indicated that ribophorin 1 (RPN1), RPN2, and dolichyl-diphosphooligosaccharide-protein glycosyltransferase non-catalytic subunit (DDOST), subunits of oligosaccharyl transferase, are possible binding partners of FL-HK1, but not Tr-HK1 (Supplemental Figure 8E). We first confirmed that there is no significant difference in the protein levels of RPN1, RPN2, and DDOST in the ECs of WT and ΔE1HK1 mice (Supplemental Figure 8F). Co-immunoprecipitation (co-IP) experiments using HUVECs stably overexpressing eGFP, FL-HK1, and Tr-HK1 (Supplemental Figure 9, A and B) showed no clear evidence of binding of HK1 and RPN1 or RPN2 (Supplemental Figure 10, A and B). In contrast, DDOST was detected to interact mostly with FL-HK1, and not Tr-HK1, in HUVECs (Supplemental Figure 11, A and B). These results indicate that mitochondrial localization of HK1 is crucial for the interaction between HK1 and DDOST, and that the N-glycosylation observed in the ECs of ΔE1HK1 mice is likely due to the lack of this interaction.
Given the importance of mitochondria and endoplasmic reticulum (ER) interaction in physiological cellular processes and glucose metabolism 32,33 , we assessed whether the interaction between HK1 and DDOST and reduced N-glycosylation has an effect on the angiogenic potential of ECs. We first assessed tube formation of HUVECs with tunicamycin, an agent that inhibits the N-glycosylation machinery 34 , by conducting preliminary dose-curve experiments, and confirmed reduced protein Nglycosylation at 20 nM of tunicamycin with the plateau at 50 nM (Supplemental Figure 11C). The mRNA levels of genes related to UPR acted in the same manner as observed in N-glycosylation experiments (Supplemental Figure 11, D and F). Based on the data, we treated HUVECs with 0, 30, 50, and 100 nM of tunicamycin, but noted no significant effects on tube formation (Supplemental Figure 11F). Thus, localization of HK1 plays a significant role in the regulation of protein Nglycosylation, but this process is not responsible for the defect in the angiogenic potential of ECs when HK1 is not bound to the mitochondria.

Increased O-GlcNAcylation in ECs causes a defect in angiogenesis
Since our data indicated that the reduction in N-glycosylation was not responsible for the defect in angiogenic potential, we then focused on the other major end-product of HBP, i.e., protein O-GlcNAcylation. Immunoblots with anti-O-GlcNAcylated protein antibody in total cell lysate showed increased protein O-GlcNAcylation in ECs from ΔE1HK1 mice compared to WT mice ( Figure 5A). Protein O-GlcNAcylation was also increased in isolated ECs from C57BL/6J mice treated with HFD and L-NAME ( Figure 5B). We next measured protein levels of 2 major enzymes in O-GlcNAcylation machinery, OGT and O-GlcNAcase (OGA), but did not observe a significant difference in their levels in ECs from WT and ΔE1HK1 hearts ( Figure 5C). To study whether the hyper-O-GlcNAcylation was induced by the gain-of-function of ΔE1HK1, we overexpressed FL-HK1 and Tr-HK1 in HUVECs and showed reduced tube formation in the cells overexpressing Tr-HK1 compared to FL-HK1overexpressing cells (Figure 5, D and E). We next assessed whether the increase in protein O-GlcNAcylation in the ECs from the ΔE1HK1 mice is the cause of decreased angiogenesis noted in these mice. Treatment of ECs from ΔE1HK1 mice with OGT inhibitor ST045849 resulted in a significant increase in tube formation markers, including the number of junctions and total length of branching (Figure 5, F and G), indicating that the mechanism for the reduced angiogenesis in the ECs from ΔE1HK1 mice is due to increased protein O-GlcNAcylation in the cells.
The machinery of O-GlcNAcylation predominantly localizes to the cytosol and nucleus 35 and is spatially different from the N-glycosylation system. Since dislocated HK1 localizes to the cytosol and nucleus (Figure 1, H-L and 3, B and C and Supplemental Figure 9), we hypothesized that HBP intermediates are shuttled into protein O-GlcNAcylation from N-glycosylation by HK1 mitochondrial dislocation. We did not observe direct binding of OGT and HK1 lacking the MBD (Supplemental Figure 12), however, immunofluorescent imaging demonstrated increased colocalization of Tr-HK1 and OGT ( Figure 5H). Additionally, we produced HUVECs stably overexpressing FL-HK1 or Tr-HK1 conjugated with BioID2 36 (Supplemental Figure 13) to assess the proximity of HK1 lacking the MBD and enzymes related to O-GlcNAcylation machinery at the molecular level. We administered biotin to the cells for 18 hours and pulled down biotinylated proteins using streptavidin-conjugated sepharose beads, which demonstrated that OGT is significantly biotinylated more in the HUVECs overexpressing Tr-HK1 conjugated with BioID2 than FL-HK1 ( Figure 5I). Together, these data indicate that HK1 localizes in close proximity to OGT when it is dislocated from the mitochondria, and this spatial proximity modifies the balance of HBP intermediate supply into the O-GlcNAcylation machinery from N-glycosylation.

Overexpression of OGA reverses the HFpEF phenotype
We next assessed whether a reduction of protein O-GlcNAcylation in ECs would reverse the HFpEF phenotype. For these studies, we used a doxycycline (DOX)-inducible, endothelial-specific OGA overexpression mouse model (Tie2-Te t -OGA) 13 , and subjected them to either NC, HFD_L-NAME, or HFD_L-NAME plus DOX to induce OGA overexpression ( Figure 6A). We first showed that total protein O-GlcNAcylation was reduced and OGA was overexpressed in the ECs of these mice ( Figure  6, B and C), and that they display no change in EF, LV diameter in diastole (LVDd) and in systole (LVDs) (Figure 6, D and E). However, induction of OGA expression in the ECs by treating mice with DOX resulted in a reversal of HFpEF markers, including E/A, e', E/e', and LA area (Figure 6, F and G), and a reduction in HW/TL ( Figure 6H). Gross examination of the hearts also demonstrated less cardiac hypertrophy in mice treated with DOX ( Figure 6I). WGA fluorescent images of hearts confirmed that hypertrophy of CMs, microvascular rarefaction and decreased N-glycan on capillaries induced by HFD and L-NAME were reversed by DOX administration (Figure 6, J-M). EC-specific OGA overexpression preserved the intensity of WGA fluorescence on capillaries compared to HFD and L-NAME group, indicating that UDP-GlcNAc produced by OGA from O-GlcNAcylated proteins was shuttled back to protein N-glycosylation ( Figure 6L). Together, these results indicate that reversal of protein O-GlcNAcylation is sufficient to reverse the HFpEF phenotype. Additionally, since the ΔE1HK1 mice have global deletion of the MBD of HK1, the studies on the Tie2-Te t -OGA mice provide tissue-specificity data and indicate that activation of O-GlcNAcylation pathway specifically in ECs is essential for the HFpEF phenotype.

Pharmacological inhibition of OGT reverses the HFpEF phenotype
We next studied whether the pharmacological inhibition of OGT reverses the HFpEF phenotype. A number of OGT inhibitors have been introduced, but among them, only one has shown to have significant in vivo activity, 5S-GlcNHex 37 . Thus, we subjected mice to 5 weeks of NC or HFD_L-NAME, followed by 3 weeks of treatment with vehicle or 5S-GlcNHex ( Figure 7A). Assessment of the hearts of these mice showed that EC O-GlcNAcylation is significantly reduced starting 16 hours after treatment and continued for 48 hours, before rebounding at 72 hours ( Figure 7B). Thus, treatment during the 3-week period was repeated every 2 days. This treatment did not cause a change in BW of WT mice when given with NC diet (Figure 7C). While 5S-GlcNHex did not alter cardiac systolic function ( Figure 7D, E), it tended to decrease HW ( Figure 7F) and reversed the HFpEF phenotype in these mice (as assessed by E/A, e', E/e', and LA area) ( Figure 7G). These results provide a novel pharmacological approach to reverse HFpEF phenotype with potential clinical implications for this common disorder.

Discussion
Although several clinical studies suggest that EC dysfunction has a significant role in the development of HFpEF 9,10 , their causal relationship and the underlying molecular mechanisms are not well understood. In our studies, we experimentally demonstrate that EC dysfunction is a major player in the development of HFpEF. Our studies on ΔE1HK1 mice also demonstrate that microvascular rarefaction precedes other pathological changes, suggesting that the EC dysfunction is a primary driver of the pathogenesis of HFpEF. We also demonstrate that dislocation of HK1 was increased in ECs in the setting of HFpEF although its expression was not altered. The significance of HK1 dislocation from mitochondria in the progression of several diseases are reported, including its role in the development of Alzheimer disease through reduced ATP production 31 , and in conditions of lowgrade inflammation like aging or diabetes mellitus in macrophages 29 . Here, we reported a role for HK1 mitochondrial binding in another disease, i.e., HFpEF. Overall, our findings unveil a critical molecular mechanism underlying the pathogenesis of HFpEF, demonstrating that HK1 is dislocated from mitochondria in ECs in the setting of HFpEF and it is sufficient to induce HFpEF through increased protein O-GlcNAcylation and EC dysfunction.
Protein O-GlcNAcylation regulates multiple cellular processes in metazoans 35 . OGT and OGA are the predominant controllers of O-GlcNAcylation and are tightly regulated by multiple mechanisms, including protein abundance and glucose availability. However, we did not observe significant differences in their protein levels between WT and ΔE1HK1 ECs. Availability of UDP-GlcNAc is also one of the crucial factors to control O-GlcNAcylation, and since multiple nutrients (e.g. Fructose 6phosphate, L-glutamine, acetyl-CoA, and UTP) are required to produce UDP-GlcNAc, it is considered as a sensor of nutrient deficiency 38,39 . Our metabolomic and gene expression data indicate that increased O-GlcNAcylation is not induced by the modification of canonical pathways, but by altering HK1 subcellular localization and subsequently redirecting supplies of HBP-intermediates away from the N-Glycosylation machinery and towards O-GlcNAcylation. Indeed, several previous studies suggest that changes in the subcellular localization of metabolic enzymes enable the supply of metabolic substrates to specific structures or machineries and have critical effects on cellular processes. Examples include HK2 association with actin filaments in the cytosol to supply glycolytic ATP for cellular motility and angiogenesis 16 , and transient movement of TCA cycle enzymes into the nucleus for efficient supply of acetyl-CoA and α-KG to epigenetic machinery 40 .
The N-glycosylation machinery is located at the ER and the golgi apparatus, which is spatially separated from where the O-GlcNAcylation machinery reside. We demonstrate an interaction between mitochondrial HK1 and DDOST, a subunit of OST which transfers glycans on translated proteins in the ER. ER and mitochondria interact and form a structure called mitochondria-associated membranes (MAM) 33 through a number of proteins, including voltage-dependent anion channel (VDAC) 41 . Through its interaction with VDAC, HK1 may be in close proximity to the ER and efficiently supply sugar substrates to the N-glycosylation machinery. Consistent with this model, itraconazole, an-anti fungi drug that dislocates HK1 from the mitochondria, also inhibits the N-glycosylation of VEGFR2 in ECs 42 .
We also show that dislocated HK1 colocalizes with OGT in the nucleus and cytosol more than mitochondrial HK1 in ECs, and that Tr-HK1 conjugated with BioID2 is in proximity to OGT at the molecular level, supporting that dislocated HK1 localizes with the O-GlcNAcylation machinery and supplies more substrates for the reaction. The fact that overexpression of Tr-HK1 augmented O-GlcNAcylation levels and MMP was preserved in ΔE1HK1 ECs support that this increase in protein O-GlcNAcylation is not through disturbed mitochondrial function, but the gain of function of dislocated HK1. Overall, our results indicate that HK1 subcellular localization may be a key player in the balance of HBP substrate supply into downstream glycosylation machinery and protein N-glycosylation versus O-GlcNAcylation in ECs.
Developing animal models which recapitulate the clinical feature of HFpEF is of great interest to the field. Recent models of HFpEF use multiple perturbations like aging, HFD and compounds like L-NAME, AngII or desoxycorticosterone pivalate 43,44,45 . The multiple hit approach enables the models to recapitulate the HFpEF comorbidities, but it also makes it difficult to characterize the underlying molecular mechanisms because of the systemic and synergistic effects of these interventions. The ΔE1HK1 mice provides several advantages for studying HFpEF, including spontaneous development of HFpEF with aging, accelerated development of HFpEF with HFD, and occurrence of the disorder in female mice, in addition to male animals.
Our studies have limitations, including the fact that ΔE1HK1 mice are not endothelial specific and there may be off-target effects of HK1 dislocation in other cell types. However, it is important to note that mice with EC-specific overexpression of OGA reversed the HFpEF, indicating that increased O-GlcNAcation in ECs is necessary for the development of HFpEF. Also, we previously showed that macrophages from ΔE1HK1 mice produce increased cytokines in response to lipopolysaccharide (LPS) administration 29 . However, macrophage cytokine production is limited to LPS treatment and does not occur at baseline (which we confirmed in the hearts of ΔE1HK1 mice), suggesting that it is unlikely that macrophages play a major role in the development of HFpEF in these mice.
Another limitation of our study is that we have not identified specific O-GlcNAcylated proteins responsible for the EC dysfunction. Increased O-GlcNAcylation of AKT in ECs is shown to inhibit its signaling cascade and reduced angiogenesis 46,47 , and endothelial nitrogen oxide synthase (eNOS) is O-GlcNAcylated and its activity is downregulated in response to hyperglycemia 48 , suggesting that hyper-O-GlcNAcylation of certain proteins in ECs leads to detrimental effects on angiogenesis. Since dislocated HK1 from mitochondria localizes to the nucleus in addition to the cytoplasm, it is possible that specific nuclear proteins are O-GlcNAcylated by HK1 nuclear localization to orchestrate the machinery of angiogenesis in ECs. This is supported by previous reports that OGT interacts with multiple epigenetic factors and regulates epigenetics and gene expressions 49,50,51,52 . Finally, given the difficulty to measure HK1 mitochondrial localization in ECs in human hearts, our paper does not include human data.
Identifying molecular mechanisms underlying the development of HFpEF and compounds targeting the pathway to treat this devastating disease is of great interest. Our studies demonstrate a significant role for HK1 mitochondrial localization in the fate of HBP intermediates into either N-glycosylation or O-GlcNAcylation machinery in ECs. Hyper-O-GlcNAcylation mediated by dislocated HK1 causes defects in the angiogenic potential of ECs, which plays a critical role in the pathogenesis of HFpEF ( Figure 7H). These results suggest the causal relationship of endothelial hyper-O-GlcNAcylation and the development of HFpEF, and emphasize the significance of this pathway in the pathogenesis of the disease. Observation of higher protein O-GlcNAcylation in HFpEF is not surprising since this modification occurs frequently in the setting of diabetes and hypertension 13,14 , which by themselves are risk factors for the development of HFpEF. In addition, we demonstrate that EC specific overexpression of OGA is sufficient to prevent the development of HFpEF, and that pharmacological inhibition of OGT reverses the HFpEF phenotype in vivo. The majority of currently available OGT inhibitors are limited for in vivo application because of their inability to get into the cell or insolubility 53 . We show that the OGT inhibitor, 5S-GlcNHex, reduces protein O-GlcNAcylation in ECs in vivo and reverses HFpEF phenotype after exposure to risk factors (and not just prevent the development of the disease before exposure to risk factors), providing a novel potential treatment strategy for HFpEF that targets a new molecular pathway in the disease.

Reverse Transcription and Quantitative Realtime PCR
RNA was isolated from cells or heart samples using RNA-STAT60 (Tel-Test), and reverse transcribed with Taqman Reverse Transcription Reagents (Invitrogen) according to manufacturers' instructions. Quantification of relative gene expression was done using Fast SYBR Green Master Mix (Applied Biosystems) and run on 7500 Fast Real-Time PCR system (Applied Biosystems). The relative gene expression was determined using differences in Ct values between the gene of interest and housekeeping control genes, which are Actb, Hprt1, and/or 18S, depending on the experiment.

Preparation of heart sections for histology and immunofluorescence
Mice were anesthetized with a 250 mg/kg dose of freshly prepared Tribromoethanol. After making a small incision in the inferior vena cava, the beating heart was perfused with relaxing buffer (100mM KCl, 5mM EGTA, and 5mM Na pyrophosphate) until termination of beating, followed immediately with 30ml of 4% paraformaldehyde (PFA) using the peristaltic pump (7ml/min) from the apex of the LV through 22G needle. The excised heart tissue was further fixed overnight in 4% PFA. The heart was cut in half, and one of them was dehydrated in 70%, 80%, 90%, and 100% ethanol. The tissue sample was further dehydrated with xylene and embedded into paraffin. Sections were stained with hematoxylin and eosin for evaluation of general cardiac morphology and tissue organization. Masson's Trichrome (MT) staining was used to visualize cardiac fibrosis. Another half of the fixed heart was incubated with 15% and 30% sucrose in PBS for cryoprotection, then was embedded into OCT compound, followed by freezing down in 2-methyl butane cooled in dry ice and liquid nitrogen. Frozen sections were subsequently used for immunofluorescent experiments.
immunofluorescence Cells grown on glass-bottom plates were fixed with 4% PFA for 15 min. The cells or frozen sections of hearts were then washed with PBS, blocked with 5% goat serum and 0.1% Triton X in PBS for 60 min, and incubated overnight with antibodies for HK1 (2024, Cell Signaling Technology), CD31 (553370, BD), ATP synthase beta conjugated with Alexa 555 (MA1930A555, Thermo Fisher Scientific) and OGT (11576-2-AP, Proteintech) in 1% BSA, 0.1% Triton X in PBS or 5 μg/ml WGA conjugated with Alexa TM Fluor 594 (Invitrogen) overnight at 4°C. The bound antibodies were labeled with an Alexa Fluor(R) 488 or 594 anti-rabbit secondary antibodies (111-545-144 and 111-585-144, Jackson ImmunoResearch) or an Alexa Fluor(R) 647 anti-rat secondary antibody (111-605-003, Jackson ImmunoResearch). Nuclei were stained with DAPI in ProLong® Gold Antifade Mountant (P36931, Invitrogen). Fluorescent images were visualized using a Nikon W1 Dual Cam Spinning Disk Confocal with TIRF. Colocalization analysis was performed using a plug-in of Coloc2 in ImageJ, and Pearson's R-value was calculated. A histogram of gray value on a line of region of interest was also obtained using ImageJ.

Protein fractionation
Protein samples of cytosolic and membrane fractions were obtained as described previously 57 . In brief, 40 mg of fresh heart tissue was homogenized in 250 μL of PBS by a tissue homogenizer, followed by centrifugation in QIAshredder homogenize (Qiagen) at 500 xg for 10 min at 4°C. The pellet in the filtrate was resuspended in 250 μL of Lysis buffer A (150 mM NaCl, 50 mM HEPES pH 7.4, 25 μg/mL digitonin [Sigma-Aldrich], 1M hexylene glycol, and 1% protease arrest), and incubated on an end-over-end rotator for 10 min at 4°C. The samples were centrifuged at 4000 xg for 10 min at 4°C, and the supernatant was stored as samples containing cytosolic proteins. The pellet was resuspended in 200 μL of Lysis buffer B (150 mM NaCl, 50 mM HEPES pH 7.4, 1% igepal [Sigma-Aldrich], 1M hexylene glycol, and 1% protease arrest) and incubated on an end-over-end rotator for 3 hr at 4°C. The suspension was centrifuged at 6800 xg for 10 min at 4°C, and the supernatant was stored as samples containing proteins from membrane-bound organelles.

Echocardiography
Mice were anesthetized using isoflurane via a nasal cone, the chest was shaved and the temperature of the mice was maintained at 37 °C. The heart rate was continually monitored. Transthoracic echocardiography was performed using a VisualSonics Vevo 2100 system equipped with an MS400 transducer (FUJIFILM). LVEF and other indices of systolic function and thickness of LV walls were obtained from short-axis M-mode scans at the midventricular level, as indicated by the presence of papillary muscles. Apical four-chamber views were obtained for diastolic function measurements using pulsed-wave and tissue doppler imaging at the level of the mitral valve. During echocardiogram acquisition, isoflurane was maintained in 1.0-1.5% to maintain a heart rate in the range of 400-450 beats per min. All parameters were measured at least three times, and means are presented.

Exercise exhaustion test
The exercise exhaustion test was performed as previously described 30 . After three days of acclimatization to treadmill exercise, an exhaustion test was performed in the experimental groups of mice. Mice ran uphill (20°) on the treadmill (Columbus Instruments) starting at a warm-up speed of 5 m/min for 3 min after which speed was increased to 14 m/min for 2 min. Every subsequent 2 min, the speed was increased by 2 m/min until the mouse was exhausted. Exhaustion was defined as the inability of the mouse to return to running within 10 s of direct contact with an electric-stimulus grid. Running time was measured and running distance calculated.

BP measurement
BP was measured noninvasively in conscious mice using the tail-cuff method and a CODA instrument (Kent Scientific). Animals were placed in holders on a temperature-controlled platform (37°C), and recordings were performed under steady-state conditions. Before testing, all mice were trained to become accustomed to short-term restraint. On the day of the experiments, BP was recorded for at least 30 cycles, and readings were averaged from at least 20 measurements.

Lung fluid volume measurement
Lungs were harvested and weighed immediately as wet lung weight. They were air dried for 60 hours, followed by dry lung measurement. Lung fluid volume was calculated by subtracting dry lung weight from wet lung weight.

IEM
For IEM, hearts of mice were perfused with 10 ml of PBS, followed by the perfusion of 30 ml of 4% PFA from the apex of them. The hearts were excised and incubated in 4% PFA overnight at 4°C. Small pieces of heart tissue were excised and incubated in the fixative overnight at 4°C. After post-fixation in 3% uranyl acetate, samples were dehydrated in series of ethanol dilutions, embedded in LRWhite resin and polymerized for 48 hours in 50°C. Then ultrathin sections were made using Ultracut UC7 Ultramicrotome (Leica Microsystems) and incubated with HK1 antibody (CST, #2024) overnight at 4°C, followed by anti-rabbit secondary antibody conjugated with 12 nm gold particles (Jackson Immunoresearch). Sections were contrasted with 3% uranyl acetate and Reynolds's lead citrate. Samples were imaged using a FEI Tecnai Spirit G2 transmission EM (FEI Company, Hillsboro, OR) operated at 80 kV. Images were captured by Eagle 4k HR 200kV CCD camera.

Measurement of Ca 2+ half decay time
Isolated CMs were incubated with 14 μM of the calcium indicator dye Cal520-AM (ATT Bioquest) for 25 min. Cells were added to the tissue bath on the stage of an inverted Zeiss LSM 510 (25 × objective, N/A 1.2). The bath was perfused with Tyrode's solution (143 mM NaCl, 2.5 mM KCl, 25 mM NaHCO3, 2 mM CaCl2, 2 mM MgCl2, 11 mM glucose, pH 7.4) and cells were paced at 1000, 500, 400, 300, and 200 ms cycle lengths 58 . Calcium transients were analyzed using custom software in MatLab 2015b, and WT and ΔE1HK1 mice were compared. Ca 2+ half decay time was measured as the time in which a peak Ca 2+ concentration in cytosol gets halved.

Tube formation assay
The formation of tube networks was assessed as described before 59 . HUVECs and isolated ECs were seeded at 20,000 and 45,000cells per well in a 96-well plate coated with 75 mL Matrigel (Fisher Scientific) or Cultrex (R & D Systems) reduced growth factor basement membrane matrix. The cells were treated with EGM-2 medium containing tunicamycin (Sigma-Aldrich) or ST045849 (Sigma-Aldrich) wherever mentioned. Following an 18 hour-incubation, tube networks from each biological replicate were analyzed in at least three random fields by light microscopy. The number of branch points (junctions), segments and meshes, and total length of tubule networks (total length of branching) were quantified by Fiji software (Angiogenesis Analyzer).

Proliferation assay
Isolated ECs were seeded at 150,000 cells on 35 mm glass-bottom cell culture plates (Thermo Fisher Scientific) coated with 0.2% gelatin. After overnight incubation, 10 μM BrdU was added to the media, and cells were incubated for 4 hours. They were then fixated by 4% PFA incubation for 15 minutes, followed by 2N HCl incubation for 30 minutes, followed by blocking with 5% goat serum and 0.1% Triton X in PBS for 60 min, and incubated overnight with antibodies for BrdU antibody (5292, Cell Signaling Technology) in 1% BSA, 0.1% Triton X in PBS. Subsequent incubation with a secondary antibody and mounting media containing DAPI was performed as described in the method section of immunofluorescence. BrdU positive nuclei were counted and divided by the number of DAPI positive nuclei to calculate %BrdU positive.

Transwell migration assay
Transwell migration assays were performed as described previously 60 . Transwells with PET membrane with 8.0 μm pores (353182, Corning) were coated with 20 μg/ml of fibronectin (Sigma-Aldrich). Isolated ECs were seeded (100,000 cells) in the upper compartment of the transwell. EGM2 media was added to the bottom well to serve as a chemo-attractant for isolated ECs. ECs were allowed to migrate for 18 hours, after which cells were fixed, stained with 0.5% crystal violet with 2% Ethanol and the number of stained cells was quantified.

Oxygen Consumption and Extracellular Acidification Analysis
Isolated ECs were plated at 30,000 cells per well onto a 10 μg/mL fibronectin-coated 96-well Seahorse cell culture plate and grown for 18 hours in EGM-2 media. At least 18 hours before running a plate, the Seahorse sensor cartridge was incubated with Seahorse Calibrant solution according to the manufacturer's protocol in a 37°C, CO2-free incubator. On the day of an assay, cells were washed and incubated with DMEM-based assay media. The sensor cartridge was fitted onto the cell culture plate, which was then placed into a 37°C, CO2-free incubator for one hour. During the assay, which was run on the Seahorse XF96 Analyzer, the following inhibitors were injected sequentially, as is standard for the Mito Stress Test: oligomycin (1.5 μM), FCCP (1 μM), and rotenone/antimycin (1 μM for each).

13
Carbon glucose tracing and steady-state metabolomics Cultured isolated ECs were treated with 5.5 mM of 13 C6-glucose for 30 mins or 2 hours. Massspectrometry and metabolite identification were performed on 80% methanol & 20% ultrapure water extracted metabolites. Metabolomics services were performed by the Metabolomics Core Facility at Robert H. Lurie Comprehensive Cancer Center of Northwestern University. Samples were analyzed by High-Performance Liquid Chromatography and High-Resolution Mass Spectrometry and Tandem Mass Spectrometry (HPLC-MS/MS). The system consisted of a Thermo Q-Exactive in line with an electrospray source and an Ultimate3000 (Thermo) series HPLC consisting of a binary pump, degasser, and auto-sampler outfitted with an Xbridge Amide column (Waters; dimensions of 4.6 mm 3 100 mm and a 3.5 mm particle size). The mobile phase A contained 95% (vol/vol) water, 5% (vol/vol) acetonitrile, 20 mM ammonium hydroxide, 20 mM ammonium acetate, pH = 9.0; B was 100% Acetonitrile. The gradient was as following: 0min, 15% A; 2.5min, 30% A; 7min, 43% A; 16min, 62% A; 16.1-18min, 75% A; 18-25min, 15% A with a flow rate of 400 mL/min. The capillary of the ESI source was set to 275°C, with sheath gas at 45 arbitrary units, auxiliary gas at 5 arbitrary units, and the spray voltage at 4.0 kV. In positive/negative polarity switching mode, an m/z scan range from 70 to 850 was chosen, and MS1 data were collected at a resolution of 70,000. The automatic gain control (AGC) target was set at 1 × 106 and the maximum injection time was 200 ms. The top 5 precursor ions were subsequently fragmented, in a data-dependent manner, using the higher energy collisional dissociation (HCD) cell set to 30% normalized collision energy in MS2 at a resolution power of 17,500.
The sample volumes of 10ml were injected. Data acquisition and analysis were carried out by Xcalibur 4.0 software and Tracefinder 2.1 software, respectively (both from Thermo Fisher Scientific).

Lectin blot
Nitrocellulose membranes with blotted proteins were blocked with PBS containing 2% Tween 20 for 2 minutes at room temperature. They were rinsed with PBS twice, then Incubated with 1 ug/ml of ConA-HRP (Sigma-Aldrich) in PBS containing 0.05% Tween 20, 1 mM CaCl2, 1 mM MnCl2, and 1 mM MgCl2 for 16 hours at room temperature avoiding exposure to the light. The presence of proteins conjugated with the lectin was visualized using Super Surgical Western Pico ECL substrate (Pierce). Quantification of the images was done using ImageJ (NIH).

Generation of HK1 lentivirus gene overexpression constructs
Constructs to produce lentiviruses for overexpressing eGFP, FL-HK1 and Tr-HK1 were produced as described previously 29 . To produce constructs for overexpressing FL-HK1-BioID2-HA and Tr-HK1-BioID2-HA, PCR fragments of BioID2-HA were obtained from MCS-BioID2-HA plasmid (#74224, Addgene) using primers with a 15-bp homology portion to the 3'-end of either FL-HK1 or Tr-HK1 just before the stop codon of the genes to facilitate recombination with In-Fusion (Takara Bio) reaction.

Production of lentiviruses and stably overexpressing HUVECs
To overexpress the gene of interest, HUVECs within passage number of 3 in a 35 mm-well were incubated with 25 μl of concentrated viruses, 775 μl of EGM2, and 0.8 μl of polybrene (Millipore Sigma) for 8 hours, followed by addition of 1 ml of EGM2 media and incubation for another 16 hrs. The media was replaced with fresh media without lentivirus and the cells were used for subsequent experiments 48 hours after the change of media (72 hrs after the lentivirus incubation). To establish stably overexpressing HUVECs, 0.6 μg/ml of puromycin was added for 7 days to select the cells with the integration of the gene of interest into their genome. The successful selection was confirmed by immunoblot and immunofluorescence.

Co-IP
HUVEC cells stably overexpressing genes of interest were plated in 150 mm culture dishes and lysed using IP lysis buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 1mM EDTA, 1% NP-40, and 5% glycerol) with protease arrest (Fisher Scientific). The lysate was centrifuged at 15,000xg and supernatant was collected in a fresh tube, followed by making the concentration of samples among the groups consistent. 25 μg of Protein G magnetic beads (Thermo Fisher Scientific) was washed with IP lysis buffer, then mixed with the sample, and 4 μg of the antibody of DDOST (14916-1-AP, Proteintech) or OGT (11576-2-AP, Proteintech) was added and incubated overnight at 4°C on a rotator. The next day, beads were magnetically precipitated and washed 3x with IP lysis buffer. Bound protein from precipitated beads was eluted using sample buffer (4x Laemmli Sample Buffer [Bio-Rad], 10x NuPAGE sample reducing agent [Boston BioProducts] in ddH2O) and western blot analysis was performed.

Biotinylated protein pulldown
Experiments to pulldown biotinylated proteins were performed as described previously 61 . In brief, HUVEC stably overexpressing FL-HK1-BioID2-HA or Tr-HK1-BioID2-HA were seeded in 150 mm culture plates and treated with 50 μM of biotin (Sigma-Aldrich) in EGM2 for 18 hours. The cells were lysed using lysis buffer (8 M urea in 50 mM Tris pH7.4 with 1×protease arrest), followed by sonication and centrifugation at 15,000xg. The supernatant of the samples was collected, and the concentration of protein was made consistent among the samples. Preclearing of the samples with Gelatinconjugated Sepharose 4B (GE Healthcare) was performed, then the supernatant was incubated with 50 μL of Streptavidin Sepharose High-Performance Beads (GE Healthcare) for 4 hours at 4°C. After the incubation, the complex of beads and proteins was washed with Lysis buffer on a rotator at room temperature for 8 minutes 3 times. To extract proteins from the complex, sample buffer (same formulation with IP experiments above) was added and the supernatant was collected as protein samples 10 minutes after incubation at 98°C and used for subsequent experiments.

Analysis of scRNA-seq data
Expression levels of genes involved in glycolysis in CMs, FBs, and ECs in mice at the age of 9 weeks were analyzed using the scRNA-seq data available in the National Center for Biotechnology Information Gene Expression Omnibus with the accession ID GSE106118 62 . Relative TPM were obtained by dividing each value of TPM by the mean value of TPM in CMs, and they were plotted in the form of a violin plot.

Analysis of proteomics data
Proteomics data in a previous study was obtained from MassIVE (Accession number: MSV000088901) 29 . The numbers of detected fragments of peptides corresponding to the amino acid sequence of RPN1, RPN2, and DDOST in each group were analyzed.
Synthesis and administration of 5S-GlcNHex to mice 5S-GlcNHex (WuXi AppTec Co., Ltd, China) was synthesized as described previously 37 . 50 mg/kg of the chemical was dissolved in 100 μL of PBS and injected into the intraperitoneal cavities of mice. Isolated ECs were collected 0, 18, 24, 40, and 48 hours after injection, followed by purification of protein samples. Subsequently, levels of O-GlcNAcylation of proteins in the samples were analyzed by immunoblot experiments. For the assessment of the effect of 5S-GlcNHex on the development of HFpEF, C57BL/6J mice were treated with HFD and L-NAME for 5 weeks, followed by the injection of 50 mg/kg 5S-GlcNHex into the intraperitoneal cavities every 2 days for 3 weeks, and then echocardiography was performed.

Statistics
Data are presented as mean ± SEM. Single comparisons were assessed by an unpaired Student ttest. Statistical significance was assessed with ANOVA, with post hoc Tukey's test for multiple group comparison. For a comparison of non-parametric data, the Mann-Whitney test was performed. A Pvalue less than 0.05 was considered statistically significant. The analysis was conducted using GraphPad Prism9 software.