key: cord-0018276-bv16nzlb
authors: Decano, Julius L.; Singh, Sasha A.; Gasparotto Bueno, Cauê; Ho Lee, Lang; Halu, Arda; Chelvanambi, Sarvesh; Matamalas, Joan T.; Zhang, Hengmin; Mlynarchik, Andrew K.; Qiao, Jiao; Sharma, Amitabh; Mukai, Shin; Wang, Jianguo; Anderson, Daniel G.; Ozaki, C. Keith; Libby, Peter; Aikawa, Elena; Aikawa, Masanori
title: Systems Approach to Discovery of Therapeutic Targets for Vein Graft Disease: PPARα Pivotally Regulates Metabolism, Activation, and Heterogeneity of Macrophages and Lesion Development
date: 2021-04-06
journal: Circulation
DOI: 10.1161/circulationaha.119.043724
sha: 6e35fdc985d55f469b889f53499e92597b901c28
doc_id: 18276
cord_uid: bv16nzlb

Vein graft failure remains a common clinical challenge. We applied a systems approach in mouse experiments to discover therapeutic targets for vein graft failure. METHODS: Global proteomics and high-dimensional clustering on multiple vein graft tissues were used to identify potential pathogenic mechanisms. The PPARs (peroxisome proliferator-activated receptors) pathway served as an example to substantiate our discovery platform. In vivo mouse experiments with macrophage-targeted PPARα small interfering RNA, or the novel, selective activator pemafibrate demonstrate the role of PPARα in the development and inflammation of vein graft lesions. In vitro experiments further included metabolomic profiling, quantitative polymerase chain reaction, flow cytometry, metabolic assays, and single-cell RNA sequencing on primary human and mouse macrophages. RESULTS: We identified changes in the vein graft proteome associated with immune responses, lipid metabolism regulated by the PPARs, fatty acid metabolism, matrix remodeling, and hematopoietic cell mobilization. PPARα agonism by pemafibrate retarded the development and inflammation of vein graft lesions in mice, whereas gene silencing worsened plaque formation. Pemafibrate also suppressed arteriovenous fistula lesion development. Metabolomics/lipidomics, functional metabolic assays, and single-cell analysis of cultured human macrophages revealed that PPARα modulates macrophage glycolysis, citrate metabolism, mitochondrial membrane sphingolipid metabolism, and heterogeneity. CONCLUSIONS: This study explored potential drivers of vein graft inflammation and identified PPARα as a novel potential pharmacological treatment for this unmet medical need.

What Is New?

• Using proteomics, network analysis, and high-resolution ultrasonography in the experimental vein graft disease model, we established a discovery platform to identify novel therapeutic targets. • Peroxisome proliferator-activated receptor α activation suppresses the development of vein graft and arteriovenous fistula lesions. • Peroxisome proliferator-activated receptor α reduces macrophage activation by influencing macrophage heterogeneity, mitochondrial integrity, and metabolome.

What Are the Clinical Implications?

• Peripheral artery disease and chronic kidney disease prevalences are increasing, warranting a need for vein grafts and arteriovenous fistula. • Vein graft and arteriovenous fistula failure lack effective therapeutic options. Our target discovery platform is applicable to such diseases. 

Two-group comparisons between tests and control groups were made using a t test after assessing their normality distribution (GraphPad). Multigroup comparison in proteomics was made using ANOVA filtering for P values and false discovery rates <0.05 (Qlucore Omics Explorer). Spearman correlation assessed a 2-parameter interrelationship. Time course proteomics was evaluated using XINA, an R package that provides a statistical workflow to investigate the trend clusters and coabundance patterns of proteins. 15 Network building and pathway enrichment using MetaCore and R created a set of network modules associated with input proteomic data or single-cell transcriptomic data. The algorithm was evaluated for creating modules that have higher than random saturation with the genes/proteins of interest. MetaCore calculated P values for the networks generated on the basis of hypergeometric distribution and evaluated its relevance to gene ontology biological processes. Benjamini-Hochberg correction was done in the network comparisons to avoid "the multicomparison problem" by adjusting individual P values from every pairwise comparison for significance (false discovery rate <0.001). Metabolomics data were tested (Metabolon) using standard statistical analyses (t test and ANOVA) in ArrayStudio on log-transformed data. Single-cell data were analyzed using principal component analysis using the genes accounting for the high variability across all the cells. The resulting top principal components accounting for the 95% variability between cells were used for tSNE dimensionality reduction (SeqGeq).

The inferior vena cava of syngeneic donor mice was anastomosed end-to-end to the carotid artery to create vein grafts ( Figure 1A) . 16, 17 The vein graft wall at 4 weeks after implantation was thicker in fat-fed Ldlr -/mice compared with normal chow-fed C57BL/6 wild-type mice ( Figure 1B , Figure 

Global proteomics for target discovery used neointimal (NEO) and adventitial (ADV) layers dissected from 4-week vein grafts from 2 Ldlr -/mice along with their equal-length nonarterialized endogenous vein controls, the inferior vena cava (IVC) of each mouse ( Figure 1E ). These were paired with 2 age-matched wild-type mouse vein grafts and IVC controls whose relatively thinner layers prevented accurate dissection. A total of 1357 proteins were quantified (Expanded Methods in the Data Supplement), then filtered using a multigroup comparison, resulting in 729 proteins ( Figure 1F ). Cluster 1 ( Figure 1F , upper) included proteins that were either predominantly expressed in the IVC samples and in 1 wild-type vein graft with minimal to nonexistent plaque formation ( Figure IG in the Data Supplement) or diminished in the NEO and ADV samples. Conversely, Cluster 2 ( Figure 1F , lower) includes proteins that are predominant in the NEO and ADV of Ldlr -/vein grafts (thick plaque, Figure IH in the Data Supplement) or diminished in the IVC and "nonplaque" wild-type vein graft sample. Although NEO and ADV layers are distinct and expected to have different proteome profiles, multigroup comparison analysis revealed them as part of 1 major cluster on the basis of principal component analysis ( Figure IIA in the Data Supplement), and first-level hierarchical clustering because of their proteome similarity when compared against IVC samples (along with 1 lesion-free wild-type vein graft specimen). Even when considering only Ldlr -/samples, NEO and ADV proteomes are still clustered together, apart from the IVC ( Figure IIB and IIC in the Data Supplement). However, only by omitting the IVC can group comparison show a statistical difference between NEO and ADV as depicted in principal component analysis and hierarchical clustering, with only 33 proteins accounting for variability between the 2 sample types ( Figure IID and IIE in the Data Supplement). Hence, we considered NEO and ADV to be 1 group compared with the IVC counterparts.

Because the vein graft lesion increases gradually across 4 weeks ( Figure IIIA in the Data Supplement), the kinetics of these molecular signatures was monitored. We performed a time course proteome profile of developing vein grafts at day 1, day 3, week 2, and week 4 after implantation in Ldlr -/animals. IVCs were time point-matched ( Figure 1G ; Figure To parse proteins on the basis of kinetic profiles, we performed cluster analysis XINA established by us 15 week-old male mice and fat-fed (2 weeks prefed) low-density lipoprotein receptor (Ldlr -/-) 12-week-old male mice (C57BL6 background) vein grafts in ultrasound imaging (long-axis view) at week 4 after operation. (n=6 versus n=6). Near wall (ventral) = NW, far wall (dorsal) = FW. Scale =1 mm. C, Long axis view (scale bar 1 mm) of a representative vein graft (VG) lesion showing an increase of lesion size (blue arrowheads) from 1 week to 3 weeks after operation. Luminal stenosis evident at 3 weeks (yellow arrows) D, Immunofluorescence of vein graft at 4 weeks after operation using AF488-anti-CD68 (macrophages, green color), Cy3-α-smooth muscle actin, SMA (vascular smooth muscle cells, VSMCs, red color), and DAPI (4′,6-diamidino-2-phenylindole, nucleus, blue color). Co-loc. indicates co-localization. Scale=100 µm. E, VG tissue layer dissection of neointimal (NEO) and adventitial (ADV) layers of Ldlr -/-VG samples. WT VG samples were not microdissected. IVC was used as controls. Representative immunofluorescence staining of Ldlr -/versus WT VG tissue (n=2 biological replicates, n=2 technical replicates). F, Tissue layer proteomics (n=2 mice, 2 technical replicates per tissue) by multigroup comparison, false discovery rate ≤0.05. Proteins that are relatively increased in IVC samples and VG of 1 WT animal are also relatively decreased in Ldlr -/-VG tissues in both NEO and ADV layers and in 1 VG of another WT animal, and vice versa. G, Time course proteomics of Ldlr -/-VGs: 1 day (D1), 3 days (D3), 14 days (W2), and 28 days (W4) after VG surgery (n=12 VG donors, 3 biological replicates per time point). Vein grafts were processed as a whole (no layer dissection) and paired with matching IVC controls from the same animals. H, Multiplexed analysis of proteome across time points showing "coabundance" proteins in time (see Expanded Methods in the Data Supplement). that permits combining protein kinetic profiles from both Ldlr -/vein grafts and IVC tissues for a single clustering step and output ( Figure 1H ). Within each cluster trend line containing both vein graft and IVC proteins, only the proteins uniquely present in vein grafts, but not on IVCs, maybe the relevant molecular signatures for the diseased vein (vein grafts). Day 1 and day 3 time points may include immediate changes and adaptations of vein grafts and exposure to arterial flow. Early thrombosis and inflammatory damage may come into play during these early time points. Clusters 1, 3, 5, 6, 9, 10, 12, 13, 15, 16, 17, 20, 21, 23 , and 24 represent proteins that are abundant during these early time points in both the vein graft and the IVC samples and are depicted in purple line graphs ( Figure 1H ). Later time points may characterize factors responsible for accelerated lesion development, thickening, or adverse remodeling of the plaque that is evident during imaging ( Figure 

We constructed our pathways networks ( Figure 2 ) from proteins filtered from the week 4 end point vein graft tissue layer-static proteome ( Figure 1F ) and from the kinetic proteome profile of the whole vein graft lesion development ( Figure 1G ). For the first tissue layer network ( Figure 2A , Table I in the Data Supplement), we input proteins elevated in lesion-positive Ldlr -/-NEO and ADV layers or wild-type whole vein grafts (cluster 2 in Figure 1F ). The lesion-positive static proteome which defined, for instance, the Ldlr -/-NEO and ADV layers, showed enriched biological processes associated with inflammation and extracellular matrix remodeling ( Figure 2A , Figure IVA in the Data Supplement, in a higher resolution).

The second tissue network ( Figure 2B , Table II in the Data Supplement) was generated from proteins higher in the IVC controls of both Ldlr -/and the wild-type mouse with minimal lesion development (cluster 1, Figure 1F ). This IVC proteome enriched mainly mitochondrial metabolic processes such as nonglycolytic bioenergetics (ie, PPARα regulated lipid metabolism, tricarboxylic acid (TCA) cycle, and amino acid metabolism) ( Figure 2B , Figure IVB in the Data Supplement). Similarly, we constructed pathway network modules for the early-phase and late-phase filtered proteins ( Figure 2C Enriched pathways (nodes) in each group were connected through shared proteins to form a network. Intermediary pathways in central-most network positions act as a primary conduit for "passing" information between the "nonshared proteins" pathways. By this "traffic" conduit logic, 18 the 3 top-ranked central pathways/nodes for each sample-condition network may contain the most desirable target(s). Pathways with high centrality may serve as key pathobiological roles in disease processes, as we demonstrated. 19 The top 3 biological processes for each dataset and the most central protein within each top-ranked pathway were identified ( Table V in . The entire study was conducted blindly until all analyses were finalized ( Figure 3A , upper). The ultrasound 3-dimensional measurements showed that vein grafts of siPPARα (small interfering RNA of PPARa)treated mice had higher wall volume and thickness than controls ( Figure 3B and 3C). Enhanced glucose uptake, a feature of inflamed tissue, appeared increased in the PPARα siRNA versus control as well ( Figure 3D ).

To clarify in vivo evidence for the suppressive role of PPARα in vein graft lesion development, the gain-offunction study ( Figure 3A , lower) used the first-in-class highly selective PPARα modulator pemafibrate in a blind fashion. 23, 24 Pemafibrate (0.2 mg/kg body weight per day) admixed with the diet lessened the volume of the developing lesion up to the third week (3-dimensional ultrasonography, Figure 3E ). At the 4-week time point, although the control group had a higher mean wall volume increase compared with the drug-treated group, this difference was statistically insignificant. The control group showed increased weight gain after the second week postoperatively ( Figure VIIIC in the Data Supplement) despite similar food consumption ( Figure VIIID in the Data Supplement). The neointima of the pemafibrate-treated grafts at the 4-week time point contained fewer macrophages than those of control grafts, whereas smooth muscle cell content did not differ ( Figure 3F ). MMP-9 and MMP-13 staining was less in pemafibrate-treated mice (Figure IXA in the Data Supplement). We chose the dose of pemafibrate that would not affect plasma triglyceride, cholesterol, and glucose levels to examine whether the effects of PPARα activation are independent of changes in the blood lipid profile. There were no differences between nonfasted control and drug treatment groups at 4 weeks after surgery ( Figure IXB in the Data Supplement). However, in a separate no-surgery experiment, in high fat-fed Ldlr -/mice, fasting plasma triglyceride levels decreased in the pemafibrate-treated group but not the fasting plasma cholesterol or glucose levels ( Figure  IXC in the Data Supplement). Circulating Ly6C ++ monocytes decreased in the pemafibrate-treated group but not the B cells or T cells (whole blood flow cytometry, Figure 3G , Figure 

To highlight clinical translatability of this target discovery platform, we evaluated the network proximity of the vein graft disease module to the gene modules for 4 linked or allied human vascular diseases 25 : atherosclerosis/coronary artery disease, AVF failure, chronic kidney disease with diabetes, and PAD. Because of scant information about human gene-associated vein graft disease or failure, the human vein graft disease-gene module remained unbuilt. An N × N plot ( Figure 4A ) showed false discovery rate values <0.05 (Benjamini-Hochberg correction) in dataset modules that have close associations by "first neighbor" proteins or shared proteins. The NEO/ ADV predominant proteins network module has close associations with the early-and late-phase proteins but not the IVC module from vein graft tissue. The NEO/ADV and late time point vein graft modules also have close associations with the human AVF failure. It indicates that our experimental vein grafts share some similar pathological processes with AVF failure, another arterialized vein disease.

Experimental AVF construction in fat-fed Ldlr -/mice used a previously reported technique. 26 An end (of vein)to-side (of artery) anastomosis of the left external jugular vein to the midportion of the left carotid artery was performed in the same mouse ( Figure 4B ). The in vivo study, carried out blindly, tested whether PPARα activation by pemafibrate would improve AVF patency and retard lesion progression ( Figure 4C ). Three weeks after surgery, pemafibrate-treated mice had more patent AVF and better blood flow by color Doppler (Figure 4D and 4E). At 7 weeks after surgery, pemafibrate-treated animals had more patent arterio-venous fistulas as determined histologically ( Figure 4F and 4G ).

In vitro gain-of-function and loss-of-function assays on mouse bone marrow-derived macrophages demonstrated that pemafibrate-induced PPARα activation reduced expression of proinflammatory factors Tlr4, Tlr2, Tnfα, Il6, Il1β, Cxcl9, Cxcl10, and Cxcl11 in lipopolysaccharide-elicited macrophages ( Figure 4H , Figure XA -XC in the Data Supplement). Patterns of differential gene expression after PPARα silencing mimicked the lipopolysaccharide (alone) condition, despite pemafibrate treatment. Arginase 1 and chitinase like-3 (Ym1), molecules associated with reparative macrophage polarization, 27, 28 Decano et al PPARα as Therapeutic Target for Vein Graft Disease . This observation that activated macrophages remain heterogeneous rather than simply polarized was consistent with our previous single-cell analysis data in primary human macrophages elicited with interferon-γ or indoxyl sulfate. 20, 29 This cluster harbor cells with the most expression of proinflammatory genes ( Figure 5B and 5C, Figure XIB and XIC in the Data Supplement). The M(-) population also exhibits a distinct cluster, M(-) cluster 1, that does not appear to contain any M(LPS) cells or have a high expression of proinflammatory genes, but has a high expression of PPARα ( Figure 5D ) and genes associated with the TCA cycle ( Figure XID in the Data Supplement). On the basis of the differential gene expression between these 2 clusters, we identified which genes are at least 1.5-fold increased in 1 cluster versus the other ( Figure XIB in the Data Supplement). MetaCore analysis on each set of differentially increased genes reveals each cluster's enriched biological processes. The top-ranked processes in the M(LPS) cluster 1, relative to the M(-) cluster 1, are associated mainly with proinflammatory signaling ( Figure 5E ), whereas M(-) cluster 1, relative to M(LPS) cluster 1, was related with phagocytosis and chemotaxis ( Figure 5F ). Single-cell quantitative polymerase chain reaction of primary human macrophages measured the expression of 32 genes corroborating cell heterogeneity reported above. We also confirmed the presence of the inflam- 

Because fatty acid oxidation is related to oxidative phosphorylation, PPARα activation may shift macrophage metabolic state by reprogramming the M(LPS) cells from a hyperglycolytic state to an oxidative phosphorylation-dependent state. We profiled metabolomic changes in peripheral blood mononuclear cell-derived human macrophages during proinflammatory activation and subsequent PPARα activation using the HD4 and CLP + metabolomics platforms (Metabolon). Peak extracellular acidification rate (≈ glycolytic rate) of bone marrow-derived macrophages occurred between 60 and 65 minutes after lipopolysaccharide stimulation ( Figure  XIIIA in the Data Supplement). Given this information, we designed an in vitro timed assay for the changing metabolome (3 time points: 0 hour, 1 hour, and 4 hours, Figure 6A ). We analyzed whole-cell metabolites, whole-cell lipids, and metabolites from isolated mitochondria.

Glucose intracellular uptake and glycolysis involve several proteins regulated by PPARα: GLUT1, HXK, and TPI1 ( Figure 6B ). Whole-cell metabolomics showed elevated intracellular glucose both 1 and 4 hours after LPS stimulation of human primary macrophages but decreased on PPARα activation, whereas hexose diphosphates and dihydroxyacetone phosphate (DHAP) increased (Figure 6C) . Citrate was elevated in M(LPS), similar to reports of LPS-primed mouse macrophages, 30, 31 and decreased on PPARα activation. An increase of isocitrate, a TCA cycle intermediate, accompanied this shift (Figure 6D) , suggesting a preferred feed-forward mechanism from glycolysis to TCA rather than the expected cytosolic escape of excess mitochondrial citrate during LPS activation ( Figure 6E ). 31 In both primary human and mouse macrophages, on PPARα activation, the extracellular Decano et al PPARα as Therapeutic Target for Vein Graft Disease acidification rate remained low after LPS injection (arrow in Figure 6F , Figure However, the addition of pemafibrate, given nonsilenced PPARα expression, increased glycolytic reserve ( Figure  XIIID and XIIIE in the Data Supplement). Oxidative respiration related oxygen consumption rate and maximal respiratory reserve increased on additional PPARα activation (pemafibrate-treated M(LPS), light blue band in Figure 6G ). PPARα silencing negated metabolic effects of pemafibrate on M(LPS) (Figure 6F and 6G). Thus, PPARα shifts bioenergetic preferences of M(LPS) from a highly glycolytic state to a lesser one accompanied by increased oxidative respiration (oxidative phosphorylation). A PPARα-mediated feed-forward mechanism to TCA among M(LPS) is demonstrated using the Mitoplate S assay (Biolog). TCA substrate utilization of fumarate, succinate, glutamate, malate, isocitrate, and α-ketobutyrate appear to show a trend of increasing rate on pemafibrate treatment and reversal on PPARα silencing ( Figure 6H ). However, only glutamate and succinate consumption showed statistically significant differences ( Figure 

PPARα activation decreases nonmitochondrial oxygen consumption in LPS-treated mouse bone marrow-derived macrophages ( Figure XIIIB in the Data Supplement), which may be a result of nicotinamide adenine dinucleotide phosphate, reduced form, oxidases and nitric oxide synthase (NOS) activity. PPARα activation increased asymmetric and symmetric dimethylarginine (ADMA and SDMA) levels ( Figure XVIA and XVIB in the Data Supplement). ADMA can inhibit inducible NOS. 32 ADMA and SDMA mediate how PPARα activation reduces NOS activity in a gain-and loss-of-function NOS activity assay in human primary macrophages ( Figure  XVIA -XVIC in the Data Supplement). NOS activity increases pro-oxidant stressors like reactive oxygen species that damages the mitochondrial membranes, which may be mitigated by PPARα activation through ADMA and SDMA, and by increasing the presence of antioxidants like reduced glutathione ( Figure 6I and 6J) Mitochondria lipidomics showed that PPARα activation reduces sphingolipid degradation molecules such as ceramides in human M(LPS) ( Figure 6K ). It may indicate reduced degradation of sphingolipids from the mitochondrial membranes, preserving membrane integrity and fluidity. A resulting less "leaky" membrane maintains high mitochondrial membrane potential and "fitness. " 33, 34 M(-) macrophages contained more TMRM-high than TMRM-low cells ( Figure 7A ), similar to M(LPS) treated with pemafibrate, with no PPARα silencing ( Figure 7A ). In contrast, there is a lesser frequency of TMRM-high than TMRM-low cells in M(LPS) when PPARα is silenced or pemafibrate withheld ( Figure 7A ).

MitoSox staining, which indicates mitochondrial membrane oxidative stress damage, demonstrated that pemafibrate-treated M(LPS) without PPARα silencing (siControl) have minimal difference (count frequency) between high versus low MitoSox staining fractions compared with M(LPS) ( Figure 7A ). This difference in counts between high versus low MitoSox staining cells in pemafibrate-treated M(LPS) was abolished when PPARα was silenced (lower right quadrant of right panels in Figure 7A ). High TMRM staining indicates high mitochondrial membrane gradient potential typically seen in healthy mitochondria with a large capacity for oxidative phosphorylation. 34 Low TMRM staining is concordant with high MitoSox staining after lipopolysaccharide stimulation, as oxidative damage decreases the membrane gradient. TMRM staining decreased to minimal levels after 10 minutes in a RAW264.7 cell primed with lipopolysaccharide for an hour ( Figure 7B ).

Lipidomic profiling revealed that treating proinflammatory M(LPS) macrophages with pemafibrate increased intracellular loading of cholesterol esters, mono-, di-, and triacylglycerols ( Figure We therefore constructed a small "directed" gene regulatory network ( Figure XX in the Data Supplement) demonstrating how PPARα may increase expression of the following enzymes: ACON (aconitase/aconitate hydratase), IDH (isocitrate dehydrogenase), PMRT1 (protein arginine-N-methyl transferase), GPX (glutathione peroxidase), GSR (glutathione reductase), GSHB (glutathione synthetase), and HXK1 and HXK2 (hexokinases). The network also shows how PPARα may decrease expression of the proteins GLUT1 (glucose transporter 1), TPI1 (triosephosphate isomerase), and ASM (acid sphingomyelinase), explaining the changing metabolites seen in M(LPS) treated with pemafibrate. In the mitochondrial lipidomic survey, sphingomyelin degradation (via ASM) by-products N-behenoyl-sphingadienine (d18:2/22:0) and N-palmitoyl-sphingosine decreased with PPARα activation in M(LPS) ( Figure 6K ). Intracellular glucose availability for glycolysis decreased in M(LPS) treated with pemafibrate. Glucose uptake in mouse vein grafts that received PPARα siRNA increased ( Figure 3D ). Increased hexose diphosphates and DHAP accumulation in the PPARα-activated M(LPS) over vehicle-treated M(LPS) ( Figure 6C ) suggest increased HXK and decreased TPI1 activity. Vein graft proteomics show HXK2 protein increased whereas TPI1 protein decreased in nondiseased veins (IVC and wild-type tissues) versus Ldlr -/vein grafts (neointimal + adventitial layers) ( 

This study used a systems approach to profile vein graft lesion development, a major clinical problem promoted by maladaptive responses to changes from the venous flow to the arterial flow environment ("arterialization"). This platform, involving proteomics and network analysis, analyzed the proteome kinetics of experimental vein grafts in mice 16, 17 to increase understanding of vein graft disease. The specific goals were to establish a systems approach to identify potential targets for vein graft disease and verify this new platform via in vitro and in vivo studies involving loss-of-function/gain-of-function experiments. To accomplish the second goal, we chose the well-known PPARα pathway because it would be hard to support a new approach by examining the causal role of lesserknown pathways. In addition, we chose PPARα, for which specific drugs are available. Gain-and loss-of-function experiments substantiated our discovery platform by demonstrating that PPARα indeed exerts antiatherogenic and anti-inflammatory actions during vein graft lesion development. PPARα activation also attenuated lesion development in AVF, another vein maladaptation disorder. Multiple in vitro studies demonstrated PPARα modulation of macrophage metabolism, muting its inflamma-tory properties. The beneficial effects of PPARα may not necessarily depend on triglyceride lowering. [35] [36] [37] A low dose of pemafibrate, a novel potent and selective PPARα modulator, 24 suppressed vein graft lesion development. Metabolomic analyses revealed that pemafibrate may exert beneficial effects on vein grafts by modulating intracellular metabolism.

Although PPARα is a relatively known molecule, our findings on its role in vein grafts are novel. The present study successfully reports that our new platform identified previously unknown targets for vein graft disease and further provided new findings demonstrating that specific inhibition or activation indeed worsened or attenuated vein graft lesion development. We realize the value of the other targets, particularly SIRT6, which are subjects of our future studies.

Network analysis linked the mouse vein graft proteomics data with AVF disease, indicating similar mechanisms shared by these 2 "arterialized" vein diseases. The network closeness implied therapies for vein graft disease may also benefit AVF disease. In this regard, activation of PPARα, a target derived from vein graft proteomics, indeed reduced AVF lesion development.

Although tissue proteomics and network analysis provided these pathways and targets, an experimental limitation lies in the inherent tissue heterogeneity of mouse vein grafts. It is also complicated by minimal starting vein graft materials to attempt cell sorting before proteomics because of their small sizes. The varying proportions of macrophages and smooth muscle cells may be a concern for the skewed contribution of proteome origins. However, macrophage and smooth muscle cell content among syngeneic, diet-controlled, genetically modified proatherogenic Ldlr -/mice with postsurgical vascular disease modeling tend to have similar graft cellular compositions. Yet with the inability to correct for cell distribution during proteomic processing, we therefore resorted to network and pathway analysis to explore the possible significant source of the resulting vein graft proteome.

Metabolic and inflammatory pathways enriched in our in vivo analysis suggest macrophages being key agents in vein grafts. In vitro, the expected glycolytic slant and proinflammatory activation of M(LPS) macrophages 38 were countered by PPARα activation.

LPS-stimulated macrophages exhibit a "blocked" mitochondrial IDH activity, causing citrate accumulation and escape to the cytoplasm. 31 It results in increased inducible NOS transcription and reactive oxygen species generation. 39, 40 PPARα activation, through gene regulatory control of IDH, may relax the block, easing citrate to fuel into the TCA cycle.

PPARα activation increased both inducible NOS inhibitor ADMA and reactive oxygen species scavenger reduced-form of glutathione (GSH) in macrophages, mitigating oxidative damage. 32 Consequently, lysophospholipids and ceramides, the reactive oxygen species

Comparison of global estimates of prevalence and risk factors for peripheral artery disease in 2000 and 2010: a systematic review and analysis

Low-density lipoprotein cholesterol lowering with evolocumab and outcomes in patients with peripheral artery disease: insights from the FOURIER Trial (Further Cardiovascular Outcomes Research With PCSK9 Inhibition in Subjects With Elevated Risk)

Predictive factors for better bypass patency and limb salvage after prosthetic above-knee bypass reconstruction

Randomized trial of stent versus surgery for asymptomatic carotid stenosis

Bypass versus Angioplasty in Severe Ischaemia of the Leg (BA-SIL) and the (hoped for) dawn of evidence-based treatment for advanced limb ischemia

Contemporary outcomes for open infrainguinal bypass in the endovascular era

Outcome of infrainguinal single-segment great saphenous vein bypass for critical limb ischemia is superior to alternative autologous vein bypass, especially in patients with high operative risk

PREVENT III Investigators. Results of PREVENT III: a multicenter, randomized trial of edifoligide for the prevention of vein graft failure in lower extremity bypass surgery

Saphenous vein graft failure after coronary artery bypass surgery: pathophysiology, management, and future directions

Macrophages in atherosclerosis: a dynamic balance

Inflammation in vein graft disease

Vein graft failure: from pathophysiology to clinical outcomes

Quantifying reproducibility for differential proteomics: noise analysis for protein liquid chromatographymass spectrometry of human serum

Comparison of label-free methods for quantifying human proteins by shotgun proteomics

XINA: a workflow for the integration of multiplexed proteomics kinetics data with network analysis

Rationale and practical techniques for mouse models of early vein graft adaptations

Macrophage notch ligand delta-like 4 promotes vein graft lesion development: implications for the treatment of vein graft failure

Spatiotemporal multi-omics mapping generates a molecular atlas of the aortic valve and reveals networks driving disease

Controllability in an islet specific regulatory network identifies the transcriptional factor NFATC4, which regulates type 2 diabetes associated genes

Uremic toxin indoxyl sulfate promotes proinflammatory macrophage activation via the interplay of OATP2B1 and Dll4-notch signaling

Systemic RNAi-mediated gene silencing in nonhuman primate and rodent myeloid cells

Therapeutic siRNA silencing in inflammatory monocytes in mice

Design and synthesis of highly potent and selective human peroxisome proliferator-activated receptor alpha agonists

Selective peroxisome proliferator-activated receptor α modulators (SPPARMα): the next generation of peroxisome proliferatoractivated receptor α-agonists

Chronic inflammation links cardiovascular, metabolic and renal diseases

Relaxin receptor deficiency promotes vascular inflammation and impairs outward remodeling in arteriovenous fistulas

Emergence of immunoregulatory Ym1+Ly6Chi monocytes during recovery phase of tissue injury

The clinical dynamic changes of macrophage phenotype and function in different stages of human wound healing and hypertrophic scar formation

PARP9 and PARP14 cross-regulate macrophage activation via STAT1 ADP-ribosylation

The mitochondrial citrate carrier: a new player in inflammation

A critical role for citrate metabolism in LPS signalling

Asymmetric dimethylarginine regulates the lipopolysaccharide-induced nitric oxide production in macrophages by suppressing the activation of NF-kappaB and iNOS expression

Sphingolipids and mitochondrial apoptosis

Anti-inflammatory effect of IL-10 mediated by metabolic reprogramming of macrophages. Science

Mouse models for atherosclerosis and pharmaceutical modifiers

Lowering of dietary advanced glycation endproducts (AGE) reduces neointimal formation after arterial injury in genetically hypercholesterolemic mice

Pyruvate kinase M2 regulates Hif-1α activity and IL-1β induction and is a critical determinant of the Warburg effect in LPSactivated macrophages

iNOS as a driver of inflammation and apoptosis in mouse skeletal muscle after burn injury: possible involvement of Sirt1 S-nitrosylation-mediated acetylation of p65 NF-κB and p53

Impaired mitophagy and protein acetylation levels in fibroblasts from Parkinson's disease patients

Ceramide and the mitochondrial respiratory chain

Target-decoy search strategy for increased confidence in large-scale protein identifications by mass spectrometry

United neighborhood closeness centrality and orthology for predicting essential proteins

A novel murine model of arteriovenous fistula failure: the surgical procedure in detail

Organelle isolation: functional mitochondria from mouse liver, muscle and cultured fibroblasts

Role of choline deficiency in the fatty liver phenotype of mice fed a low protein, very low carbohydrate ketogenic diet

The authors thank Jung Choi, Peter Mattson, Alexander Mojcher, Jennifer Wen, Anna Ha, and Michael Creager for their technical assistance. We especially thank William Oldham, MD, PhD, for his assistance and guidance with the Seahorse metabolic assays. Some figures were created with BioRender.com. 

This study was supported by research grants from Kowa Company Ltd, Nagoya, Japan (to M.A.), and the National Institutes of Health (R01HL107550, R01HL126901, and R01HL149302 to M.A.).

Outside the present study, M.A. has been supported by grants from Pfizer, Inc, and Sanofi US Services, Inc, on vein graft research. P.L. is an unpaid consultant to or involved in clinical trials for Amgen, AstraZeneca, Esperion Therapeutics, Ionis Pharmaceuticals, Kowa Pharmaceuticals, Novartis, Pfizer, Sanofi-Regeneron, and XBiotech, Inc. P.L. is a member of the scientific advisory board for Amgen, Corvidia Therapeutics, DalCor Pharmaceuticals, IFM Therapeutics, Kowa Pharmaceuticals, Olatec Therapeutics, Medimmune, Novartis, and XBiotech, Inc. P.L. serves on the board of XBiotech, Inc. The laboratory of P.L. has received research funding in the last 2 years from Novartis. P.L. has a financial interest in XBiotech, a company developing therapeutic human antibodies. His interests were reviewed and are managed by Brigham and Women's Hospital and Partners HealthCare in accordance with their conflict of interest policies. The other authors report no conflicts. Tables I-V  References 42-46