Multi-omics study identifies novel signatures of DNA/RNA, amino acid, peptide, and lipid metabolism by simulated diabetes on coronary endothelial cells

Coronary artery endothelial cells (CAEC) exert an important role in the development of cardiovascular disease. Dysfunction of CAEC is associated with cardiovascular disease in subjects with type 2 diabetes mellitus (T2DM). However, comprehensive studies of the effects that a diabetic environment exerts on this cellular type are scarce. The present study characterized the molecular perturbations occurring on cultured bovine CAEC subjected to a prolonged diabetic environment (high glucose and high insulin). Changes at the metabolite and peptide level were assessed by Liquid Chromatography–Mass Spectrometry (LC–MS2) and chemoinformatics. The results were integrated with published LC–MS2-based quantitative proteomics on the same in vitro model. Our findings were consistent with reports on other endothelial cell types and identified novel signatures of DNA/RNA, amino acid, peptide, and lipid metabolism in cells under a diabetic environment. Manual data inspection revealed disturbances on tryptophan catabolism and biosynthesis of phenylalanine-based, glutathione-based, and proline-based peptide metabolites. Fluorescence microscopy detected an increase in binucleation in cells under treatment that also occurred when human CAEC were used. This multi-omics study identified particular molecular perturbations in an induced diabetic environment that could help unravel the mechanisms underlying the development of cardiovascular disease in subjects with T2DM.

www.nature.com/scientificreports/ were then incubated with a polyclonal antibody against the vWf (1:400, 3% BSA in PBS) overnight at 4 °C and thereafter washed 3 × with PBS. Alexa Fluor 488-labeled anti-rabbit (1:400 in PBS) was then used as a secondary antibody for 1 h at RT and washed 3 × with PBS. As a negative control, cells were incubated only with secondary antibody to assess for non-specific binding. Cell nuclei were stained with Hoechst 33258 (2 µg/mL in HBSS) for 30 min and washed 3 × with PBS. Fluorescent images were taken in at least three random fields per condition using an EVOS FLoid Cell Imaging Station with a fixed 20 × air objective. Image analysis was performed through ImageJ software (version 2.0.0).
Peptide and non-peptide metabolite extraction. Cells were seeded at 300,000 cells per well in 6-well plates (Corning CellBIND) and treated as above. After HG and HI conditions, metabolites were extracted following a published protocol for adherent cells with some modifications 21 (Fig. 1). In brief, after washing the cells 3 × with PBS, 500 µL of a cold mixture of methanol: ethanol (50:50, v:v) were added to each well, covered with aluminum foil, and incubated at − 80 °C for 4 h. Cells were then scrapped using a lifter (Fisher Scientific, Hampton, NH, USA), and the suspension was transferred to Eppendorf tubes before centrifugation for 10 min at 14,000 rpm at 4 °C. The supernatant was transferred to another tube and dried down by SpeedVac System (Thermo Fisher Scientific, Waltham, MA, USA). Samples were reconstituted in water/acetonitrile 95:5 v/v with 0.1% formic, centrifuged at 14,000 rpm for 10 min at 4 °C. The particle free supernatant was recovered for further LC-MS 2 analysis. LC-MS 2 data acquisition for untargeted metabolomics and peptidomics. Metabolites were loaded into an Eksigent nanoLC 400 system (AB Sciex, Foster City, CA, USA) with a HALO Phenyl-Hexyl column (0.5 × 50 mm, 2.7 μm, 90 Å pore size, Eksigent AB Sciex, Foster City, CA, USA) for data acquisition using the LC-MS parameters previously described with some modifications 22 . In brief, the separation of metabolites was performed using gradient elution with 0.1% formic acid in water (A) and 0.1% formic acid in ACN (B) as mobile phases at a constant flow rate of 5 μL/min. The gradient started with 5% B for 1 min followed by a stepped increase to 100%, B over 26 min and held constant for 4 min. Solvent composition was returned to 5% B for 0.1 min. Column re-equilibration was carried out with 5% mobile phase B for 4 min. Potential carryover was minimized with a blank run (1 μL buffer A) between sample experimental samples. The eluate from the LC was delivered directly to the TurboV source of a TripleTOF 5600 + mass spectrometer (AB Sciex, Foster City, CA, USA) using electrospray ionization (ESI) under positive mode. ESI source conditions were set as follows: IonSpray Voltage Floating, 5500 V; Source temperature, 350 °C; Curtain gas, 20 psi; Ion source gases 1 and 2 were set to 40 and 45 psi; Declustering potential, 100 V. Data was acquired using data-dependent acquisition (DDA) with high sensitivity mode selected, automatically switching between full-scan MS and MS/MS. The accumulation time for TOF MS was 0.25 s/spectra over the m/z range 100-1500 Da and for MS/MS scan was 0.05 s/spectra over the m/z 50-1500 Da. The DDA settings were as follows: charge state + 1 to + 2, intensity 125 cps, exclude isotopes within 6 Da, mass tolerance 50 mDa, and a maximum number of candidate ions 20. Under IDA settings, the ''exclude former target ions'' was set as 15 s after two occurrences and ''dynamic background subtract'' was selected. Manufacturer rolling collision energy (CE) option was used based on the size and charge of the www.nature.com/scientificreports/ precursor ion using formula CE = m/z × 0.0575 + 9. The instrument was automatically calibrated by the batch mode using appropriate positive TOF MS and MS/MS calibration solutions before sample injection and after injection of two samples (< 3.5 working hours) to ensure a mass accuracy of < 5 ppm for both MS and MS/MS data. Instrument performance was monitored during data acquisition by including QC samples (pooled samples of equal volume) every 4 experimental samples. Data acquisition of experimental samples was also randomized.

Processing of LC-MS 2 for untargeted metabolomics.
To identify peptide and non-peptide metabolites (hereafter referred as to untargeted metabolomics), we followed a workflow comprised of open-access software packages and on-line platforms (for library spectral matching) commonly used for untargeted metabolomics. Three complementary informatic approaches were utilized to analyze the LC-MS 2 datasets: (1) feature extraction, alignment, normalization, and univariate statistical analysis was performed using the XCMS (version 2.7.2) online platform (https:// xcmso nline. scrip ps. edu) 23 ; (2) MS 2 spectral data extraction for metabolite identification (Metabolomics Standards Initiative (MSI) classification level 2 and 3) 24 was performed with MZmine (version 2.53) 25 , the Global Natural Products Social Molecular Networking web platform (GNPS, https:// gnps. ucsd. edu) 26,27 , and in-silico integrated tools within SIRIUS software (version 4.9.12) 28 , and; (3) multivariate statistical analysis and heatmap visualization was done using Metaboanalyst 5.0 (https:// www. metab oanal yst. ca) 29 . For approach 1, raw proprietary .wiff files were uploaded into the XCMS online platform to perform mass detection, chromatogram building and deconvolution, isotopic assignment, feature alignment, and gapfilling (to detect features missed during the initial alignment). For approach 2, .wiff files were first converted to .mzML using ProteoWizard version 3.0 and then imported into MZmine to perform the peak extraction steps as in the XCMS online platform. To identify or annotate the metabolites at the chemical structure and class level, the MS 2 -containing features extracted with MZmine were further analyzed by molecular networking using the GNPS platform and associated in-silico dereplication tools, Network Annotation Propagation (NAP) 30 and MS2LDA 31 , as well as the automated chemical classification by Classyfire 32 , as previously described 22 . The confidences of such annotations are level 2 (probable structure by library spectrum match) and level 3 (tentative candidates) in agreement with the Metabolomics Standards Initiative (MSI) classification 24 . Molecular networking, NAP, and Classyfire outputs were integrated using the MolNetEnhancer workflow 33 . Molecular networks were visualized using Cytoscape version 3.8.2 34 . In addition, chemical substructures (co-occurring fragments and neutral losses referred to as "mass2motifs" [M2M]) were recognized using the MS2LDA web pipeline (http:// www. ms2lda. org) to further annotate metabolites (level 3, MSI). For select metabolites, high-confidence annotations (at the substructure and structure pattern, level 3, MSI) were made using in-silico integrated tools within SIRIUS software (version 4.9.12) 28 . For approach 3, normalized peak abundance data (.txt) retrieved by XCMS analysis was imported into Metaboanalyst and filtered for Principal Component Analysis (PCA) and HeatMap analysis. The detailed processing parameters for all the pipelines are found in supplemental experimental methods in supporting information.
Processing of LC-MS 2 for peptidomics. To expand the identification of endogenous peptide metabolites not contained in standards-enriched spectral libraries (including GNPS and collaborators), we performed in-silico peptide identification using the Bos taurus proteome as a database to retrieve amino acid sequences for spectral matching (hereafter referred as to peptidomics). The raw files (derived from the same experiments) used for untargeted metabolomics (.wiff and .wiff.scan files) from the experimental and control groups were analyzed separately using ProteinPilot software version 4.2 (Ab Sciex, Foster City, CA, USA) with the Paragon algorithm. MS 1 and MS 2 data were searched against the Bos taurus SwissProt sequence database (6006 reviewed proteins + common protein contaminants, February 2019 release). The parameters input was: sample type, identification; digestion, none; Cys alkylation, none; instrument, TripleTOF 5600; special factors, none; species, Bos taurus; ID focus, biological modifications, and amino acid substitutions; search effort, thorough ID. False discovery rate analysis was also performed. All peptides were exported and those with a > 90% confidence were linked to the corresponding feature extracted by the XCMS algorithm using their accurate mass and retention time information. For peptide quantification, we employed the normalized feature abundances (MS 1 level) generated by XCMS. A significance threshold of p < 0.05 (Welch's t test) was utilized.
Processing of LC-SWATH-MS 2 for proteomics. The SWATH-based proteomics data (identifier PXD013643), hosted in ProteomeXchange consortium via PRIDE 35 , was reanalyzed with some modifications. This dataset derived from independent biological replicates as those used in this study. The parameters used to build the spectral library remained the same 16 37,38 . First, a protein-protein interaction (PPI) molecular network (first-order network containing query or seeds molecules and their immediate interacting partners) with minimum network filtering selected using STRING PPI database was built 39 , and then pathway enrichment analysis was performed using the built-in REACTOME 40 and the Kyoto Encyclopedia of Genes and Genomes (KEGG) 41 databases. Hypergeometric test was used to compute p values.
Integrative bioinformatic analysis of proteomics and metabolomics data. The molecular interactions between the proteins and metabolites differentially abundant between HG + HI and NG were determined in OmicsNet. The lists of proteins (EntrezGene ID) and metabolites (HMDB ID) were loaded to build a composite network using protein-protein (STRING database selected) and metabolite-protein (KEGG database selected) interaction types. The primary network relied on the metabolite input using the minimum network filtering option. Pathway enrichment analysis was performed using the built-in REACTOME and KEGG databases. Hypergeometric test was used to compute p values.

Statistical analysis.
All experiments were performed with biological triplicates. Three independent wells per condition were pooled to generate an independent biological replicate. For all the data, excluding proteomics and metabolomics, statistical analysis was performed by either using one-way ANOVA followed by the Dunnett's post hoc test or unpaired student's t test, as appropriate. A p value ≤ 0.05 was considered statistically significant. For the proteomics data, based on the high reproducibility of SWATH-based quantification 42,43 and fold-change compression phenomenon (i.e., lower measured fold-change vs. real biological fold-change) 43,44 , proteins with a fold change ≥ 1.2 or ≤ 1/1.2 and a p value < 0.05 (Welch's t-test) were considered as differentially abundant between NG and HG + HI conditions. For the metabolomics data (DDA), also taking into account the fold-change compression phenomenon 44 -but lower accuracy of DDA than DIA-SWATH, features with a fold change ≥ 1.3 or ≤ 1/1.3 and a p value < 0.05 (Welch's t-test) were considered as differentially abundant. We did not apply multiple-test corrections to calculate adjusted p values, because this process could obscure proteins or metabolites with real changes (true-positives) 45 . Instead, the analysis was focused on top-enriched signaling pathways (adjusted p value < 0.01) that allowed us to determine a set of interacting proteins and metabolites with relevant biological information and contributes in reducing false positives. For multivariate statistical analysis and heatmap visualization, Metaboanalyst 4.0 (https:// www. metab oanal yst. ca) was utilized. PCA on log-transformed data was used to assess for sample clustering behavior and inter-group variation. No scaling was used for PCA and heatmap analysis. Software PRISM 6.0 (GraphPad Software, San Diego, CA) was used for the creation of volcano plots and column graphs.

Results
Mitochondrial membrane potential. To define a simulated diabetes model, we tested the effects (time and concentration-dependent) of various HG concentrations with and without a single concentration of HI (100 nmol/L) on the endothelial mitochondrial function 19 . We noted differential effects of HG depending on the incubation time (3 vs. 6 vs. 9 days) and the presence of HI (Fig. S1). Our results showed that 20 mmol/L glucose with 100 nmol/L insulin at day 9 significantly reduced the mitochondrial membrane potential (Fig. S1). Hyperosmolar controls at day 9 (using mannitol) did not significantly affect the mitochondrial endpoint. Therefore, our simulated diabetes model was defined as an environment of 20 mmol/L glucose + 100 nmol/L insulin for 9 days, while the control groups consisted of cells treated with 5.5 mmol/L glucose for the same period.
Untargeted metabolomics. Overall 5571 features or potential metabolites were detected using XCMS and MZmine, wherein 957 (~ 18%) features were commonly identified in both platforms ( Fig. 2A). Based on the relative quantification using XCMS, 140 and 82 features were detected with reduced and increased abundances respectively in the experimental group compared to the control group (Fig. 2B). The effects of HG and HI in the experimental group are observed by PCA analysis wherein the experimental samples clustered away from the control group (Fig. 2C). The consistency of the LC-MS equipment is apparent by the tight-clustering of the QC samples (Fig. 2C). Further, the heatmap visualization of the top 100-modulated metabolites exhibited the different distribution patterns among groups (Fig. 2D). Using the GNPS platform for automatic metabolite annotation, 106 compounds (excluding duplicates and contaminants) were putatively annotated with a level 2 confidence annotation (MS 2 spectral match) (Table S1) in agreeance with the MSI classification 24 . Some metabolites identified by the GNPS platform could not be quantified because they were not detected by the XCMS algorithm during feature area normalization and quantification. Moreover, GNPS Molecular Networking aligned the MS 2 -containing features (n = 1013) based on their structural similarity, creating 118 independent networks or clusters with at least two connected nodes (Fig. 3A). The use of MolNetEnhancer workflow allowed to putatively identify chemical classes (level 3, MSI) for 56 of the 118 independent networks. The top-10 most abundant annotated chemical classes and associated metabolites are shown in Fig. 3A. A focused analysis on the chemical subclasses associated with the dysregulated metabolites by simulated diabetes, revealed that amino acids, peptides, and analogues were principally affected (Fig. 3B). Three-clusters from the network were further analyzed because they contained annotated metabolites by spectral matching, which facilitates the annotation of other cluster's nodes. Cluster 1 revealed two metabolites linked to the organonitrogen compounds class with reduced abundance in the experimental group (Fig. 3C). Library spectral match (level 2, MSI) suggest PC(16:0/18:1(9Z)) and PC(18:0/18:2(9Z,12Z)) as putative candidates, which was supported by MS2LDA phosphocholine-substructure recognition (Fig. 3D). www.nature.com/scientificreports/ In cluster 2, glutathione-based metabolites (MSI level 3) were detected through fragments m/z 308.0925, 233.0575, 179.0475, and 162.0225 retrieved by the M2M_453 substructure and associated with glutathione structure using mzCloud in silico predictions (Fig. 4A). The precursor ion at m/z 713.1472 and glutathione (annotated at level 2, MSI) were detected with increased abundance in the experimental group. MS2LDA visualization, at the M2M level, correlated with the GNPS molecular networking clustering (Fig. 4B). In cluster 3, various phenylalanine-based metabolites were putatively annotated aided by MS2LDA substructure recognition ( Fig. 4C and D). Within this cluster, glutamyl-phenylalanine, leucine-phenylalanine and the precursor ion at m/z 487.1548 (unknown) presented with increased abundance in the experimental versus control group. On the other hand, various amino acids were annotated (level 2, MSI) by GNPS spectral matching and manual inspection of data (Table S2). Threonine, valine, proline, leucine, serine, glutamic acid, methionine, and tyrosine presented increased abundance (fold change range 1.3-1.7, p < 0.05) in the experimental versus control group. Particularly, metabolites linked to the catabolism of tryptophan via the serotonin and kynurenine pathway 46 were annotated (level 2, MSI), including melatonin, acetyl serotonin, and kynurenine (Table S2). However, only kynurenine was significantly elevated in the experimental group. The full list of annotated metabolites, differential abundances and another relevant feature information is shown in Table S2.
Peptidomics. Experimental and control datasets were analyzed separately to identify the peptides and their biological modifications. The complete list of peptides identified by ProteinPilot between the experimental and control groups are described in Table S3. Proline oxidation was the most frequent biological modification detected in the experimental group datasets. We identified 8 and 12 peptides with a confidence of > 90% in the control and experimental group, respectively. Differential abundance of 2 proline-rich peptides was observed in www.nature.com/scientificreports/ the experimental group compared to the control group. An additional tripeptide was manually annotated with a LPP sequence (Table S4).
Proteomics. The re-analysis of the SWATH data (PXD013643 dataset) facilitated the identification of 952 quantifiable proteins (717 proteins with at least 2 unique peptides, 1% false discovery rate) and no missing values among technical and biological replicates (Table S5). Sample datasets were normalized using 8 different methods to select the most appropriate based on quantitative and qualitative parameters on our dataset. Quantile normalization produced a better qualitative and quantitative profile and was selected to further process our data (Fig. S1). PCA analysis of normalized data denoted a clear separation of the groups suggesting overall differences in their proteomes (Fig. 5A). Differential abundance analysis revealed 32 and 33 proteins with increased and decreased abundance in the experimental group (Fig. 5B). Further, the heatmap visualization of the top 50-modulated proteins exhibited the different distribution patterns among the experimental and control groups (Fig. 5C). To obtain a molecular insight we performed a functional enrichment analysis using a networkbased approach. First, we created a composite network comprising PPI between the modulated proteins by simulated diabetes (seed proteins) and their immediate interacting partners (highest confidence > 0.9) retrieved from STRING Database (incorporated in OmicsNet platform). The principal network using the up-modulated proteins consisted of 91 proteins, 137 edges and 40 seed proteins (nodes with black shadow) and is illustrated in Fig. 5D. The 10 most significant (adjusted p value < 0.05) REACTOME and KEGG pathways on the global network are shown in Table 1. Generally speaking, signaling pathways linked to DNA/RNA metabolism, mitochondria and apoptosis were significantly enriched within the PPI network (Fig. 5E).
Integration of metabolomics and proteomics. The signaling pathways perturbed by simulated diabetes were identified by a composite network of interacting metabolites and proteins using OmicsNet built-in databases. Figure 6 illustrates the composite metabolite-PPI network generated, using the dysregulated molecules  Table 2. Generally speaking, signaling pathways linked to the metabolism of amino acids were significantly enriched within the protein-metabolite network. The pathways containing more hits (proteins) are highlighted (bue tones) in Fig. 6. We also noted a smaller interaction between the Acyl-protein thioesterase 1 (LYPLA1,UniProtKB:Q3MHR0) protein and a phosphatidylcholine metabolite.

Cellular morphology.
To better understand the effects that simulated diabetes exerts on endothelial cells the changes on cellular structure endpoints were evaluated. The endothelial nuclei morphology in the BCAEC control and experimental groups were evaluated using fluorescent-staining and image analysis. We also evaluated the presence of vWF (marker of endothelial cells) in BCAEC and HCAEC, to reveal the cellular boundary and to demonstrate their endothelial phenotype 47 . We noted an increase in the percentage of binucleated BCAEC in the experimental group compared to the control group (top panel Fig. 7A and B). A similar result with larger nuclei, was observed when using HCAEC as a human in vitro model (bottom panel Fig. 7A and B). Finally, as expected, we observed a typical intracellular localization of vWF and a 100% positivity in endothelial cells. www.nature.com/scientificreports/

Discussion
Endothelial cells cover the inner surface of blood vessels and are distributed across the body. Their functions include: acting as a mechanical barrier between the circulating blood and adjacent tissues as well as modulating multiple functions in distinct organs 48 . These regulatory functions vary according to localization and vascular bed-origin 49 . HG blood levels are detrimental to endothelial cells function in T2DM leading to coronary endothelial dysfunction and development of CVD 50,51 . The molecular effects of HG on endothelial cells have been previously characterized 4,6,7,10,11 ; nevertheless, the endothelial cell types used in these studies are not intrinsically involved in CVD. The present study used an in-vitro model involving endothelial cells that modulate heart function, CAEC 52 . Noteworthy, most in-vitro published studies have arbitrarily selected a single concentration of HG and time point as a means to decipher the molecular mechanisms altered by simulated diabetes 4,6,7,10,11 . Before selecting a particular HG concentration, we first tested the effects of increasing concentrations of glucose with and without HI on the BCAEC mitochondria. The mitochondrial membrane potential was selected as an endpoint www.nature.com/scientificreports/ and surrogate marker for mitochondrial function and metabolism 19 , given that mitochondria are highly relevant for glucose metabolism and ATP production and because mitochondrial dysfunction is thought to contribute to T2DM 53 . Our approach consisted of pre-treating (before challenging them with HG) cells with HI for 3 days to create a prolonged hyperinsulinemic environment and then challenging cells with HG + HI for up to 9 days, trying to mimic the pathophysiological conditions that occur in T2DM subjects wherein hyperinsulinemia precedes hyperglycemia 18 . We noted a significant decrease in the mitochondrial membrane potential when using 20 mmol/L HG + 100 nmol/L HI until day 9 (Fig. S1), which is consistent with other reports using HG or HI on muscle cells 54 , endothelial cells 55 , epithelial cells 56 , and hepatocytes 57 . Mannitol, at equivalent concentrations on day 9, did not reduce the membrane potential; thus, we could rule out a hyperosmolar effect. We, therefore, for all our experiments, employed 20 mmol/L HG + 100 nmol/L HI for 9 days as the simulated diabetes model. Noteworthy, 9 days was the longest time point analyzed trying to mimic a chronic HG exposure and preventing measuring cell proliferation known to occur in early HG 10,12 .
Manual inspection of the proteomics data revealed dysregulated proteins involved in the signaling of insulin; a serine/threonine-protein phosphatase (PPP2R2B, [UniProtKB:Q5E9Q7]) and 14-3-3 adapter protein (YWHAB, [UniprotKB: P68250]). Elevated levels of the former protein, are linked to insulin resistance, reduced GLUT4  www.nature.com/scientificreports/ translocation and glucose transport by inhibition of AKT activity; a serine/threonine kinase responsible for the phosphorylation of the insulin receptor substrate (IRS-1, a direct intracellular effector of the insulin receptor) and mediator of down-stream insulin signaling 58 . On the other hand, 14-3-3 adapters are up-stream proteins essential in the signaling by insulin 59 . When insulin binds to its receptor, 14-3-3 interacts with the insulin receptor substrate 2 (IRS-2) and promotes the activation of PI3-kinase (activator of AKT) and subsequent down-stream signaling. Simulated diabetes augmented and reduced the protein levels of PPP2R2B and YWHAB in BCAEC, respectively, which suggest impaired insulin signaling in our model. This LC-MS 2 -based methodological pipeline that included appropriate controls during data acquisition (QC) and processing (e.g., normalization, filtering, annotation, dereplication, etc.), allowed the identification of global changes in the metabolome and peptidome of CAEC under HG + HI. Using the GNPS/NAP pipeline we retrieved biological information at the metabolome level. Specifically, we noted increased abundance of valine, leucine, tyrosine, serine, leucine, proline, methionine, and glutamic acid in cells under HG conditions; and this is consistent with reports on human aortic endothelial cells 60 . Notably, several clinical studies have established a direct relationship between prevalence/incidence of T2DM and increased levels of valine, leucine and tyrosine in serum and plasma [61][62][63][64][65] . Our results support the role of CAEC in contributing to the elevated pool of amino acids seen in circulation under a HG environment. We speculate that increased levels of these amino acids could result from either increased production or reduced utilization as suggested in endothelial cells (immortalized cell line, EA.hy 926) that transition from a glycolytic metabolism towards lipid and amino acid oxidation when challenged by HG 66 . Furthermore, evidence of increased tryptophan catabolism was identified through the kynurenine pathway. In this regard, a non-significant decrease of ~ 40% in the abundance of tryptophan was detected. However, a significant increase of ~ 450% in kynurenine (tryptophan's main metabolite) 67 between the HG + HI group and NG group was also observed, which is a key finding as elevated plasma levels of kynurenine are known to increase CVD risk 68,69 . This novel finding contributes to expanding the understanding of amino acid metabolism in endothelial cells under simulated diabetes. Acetyl serotonin and melatonin which are components of the serotonin pathway that degrades tryptophan 70 were also detected with only minor abundancy increases (20-30%) in the HG + HI group compared to control.
In endothelial cells, the tripeptide glutathione (cysteine-glutamic acid-glycine) is believed to be the most critical antioxidant thiol scavenging reactive oxygen species when oxidative insults (e.g., ambient HG) are triggered 71,72 . Here, we noted an increased abundance of glutathione and its metabolite precursor glutamic acid 71 , suggesting an increased response to oxidative stress by ambient HG and HI. In line with this evidence, previous research reported a glutathione-dependent reaction to ambient HG in artery-derived endothelial cells 73,74 but the same could not be observed in vein-derived endothelial cells 75,76 . That emphasizes the different responses to HG among endothelial phenotypes. The use of exhaustive and complementary dereplication tools further allowed us to provide evidence of up-regulated novel glutathione-based metabolites, suggesting a coordinated boost of antioxidant metabolite peptides. Likewise, antioxidant enzymes conforming the cells natural enzymatic defense 77 were found with increased abundance in the experimental group, including peroxiredoxin-2 (PRDX2, [UniProtKB:Q9BGI3], peroxiredoxin-6 [ PRDX6, UniProtKB:O77834]), and mitochondrial thioredoxin (TXN2, [UniProtKB:Q95108]). Our methodology also allowed us to identifying other peptide metabolites, including glutamic acid-and phenylalanine-based metabolites, presumably di-or tri-peptides, including the annotated metabolite glutamyl-phenylalanine and leucine-phenylalanine.
To further extract biological information at the peptidome level, we performed an in-silico analysis with PeakView software on the same LC-MS 2 datasets used for untargeted metabolomics. GNPS/NAP and PeakView pipelines use MS 2 spectral matching between experimental and reference spectral data for the identification of peptides, but they rely on different reference spectral databases. Regarding peptide metabolites, GNPS/NAP 26,27,30 pipeline is focused principally on the identification of peptidic natural products (as well as other types of metabolites), while the Paragon algorithm in PeakView software 78 allows identifying endogenous peptides (by in-silico prediction) derived from an organism's proteome when used as a reference (e.g., human, animals). The CAEC peptidome analysis by PeakView pipeline suggested an increase in proline-containing peptides. This type of peptide is of particular interest because of its resistance to non-specific proteolytic degradation, body distribution and remarkable biological effects [79][80][81][82] . The origin of phenylalanine-and proline-based peptides may be derived from protein degradation supported by the noted increased protein abundance of core and regulatory subunits from the proteasome complex, including PSMA4 [UniProtKB:Q3ZCK9] and PSMD3 [UniProtKB:Q2KJ46]). This complex is a central proteolytic system that degrades proteins, releasing peptides with 3 to 22 residues for further degradation into amino acids 83 . Yet, the question to be answered is if these peptides are biologically active or merely products of protein degradation. However, that is beyond the scope of our study.
Metabolomic profiling also revealed changes in the lipidome of CAEC challenged with HG + HI, wherein a reduction in phosphatidylcholine (PC) lipids and subsequent increase in phosphocholine were noted. Changes in the phospholipidomic profile of bovine aortic endothelial cells treated with HG for 24 h has been reported in a lipidome study 84 . Here, proteomics and metabolomics data were manually integrated and this allowed to determine critical roles for PAFAH1B2 (UniProtKB:P68401) and LYPLA1 in mediating the degradation of PC lipids (Fig. 8). PAFAH1B2 was found to be up-regulated in this study and it is known to be associated with inflammation and higher levels of lysoPC 85 . As a result, PAFAH1B2 could increase the pool of lysoPC lipids, further exacerbating inflammation in the cardiovascular system 86 . On the other hand, LYPLA1 has a lysophospholipase activity that can hydrolyze a range of lysophospholipids, including LysoPC, thereby generating a fatty acid and glycerophosphocholine as products 87 . Increased levels of phosphocholine (~ 460%) were detected in HG treated cells compared to control, that could be associated with the degradation of LysoPC lipids. It should be noted that the use of pathways databases such as KEGG and REACTOME possess some limitations when dealing with lipid metabolites because its chemical diversity is not well annotated/defined within the databases. For example, KEGG provides a chemical class identifier instead of individual identity to lipids, constricting their biological www.nature.com/scientificreports/ importance 88 . Thus, based on our manual inspection of the metabolomics-proteomics data and in line with the evidence, we suggest that simulated diabetes evokes inflammation on BCAEC and that PAFAH1B2 and LYPLA1 play a role in modulating such process. Previously, we reported the multinucleation of CAEC cultured under simulated diabetes 16 . In the present study, we replicated our previous findings of increased binucleation in BCAEC. Notably, the same outcome was obtained when using HCAEC as a human in vitro model ( Fig. 7A and B), validating the binucleation process in other CAEC, which indicates that such a process is not species specific but rather a true biological process. After refinement of LC-MS 2 data and bioinformatics re-processing of published SWATH-based datasets of BCAEC under simulated diabetes 16 , molecular signatures and pathways that could be linked to the binucleation process were found (Fig. 8). For instance, we noted an increased abundance of proteins, under simulated diabetes, with reported nuclei localization and linked to DNA metabolism, including ribosomal proteins RPS7 (UniProtKB:A6H769), RPS13 (UniProtKB:Q56JX8), and RPL9 (UniProtKB:Q3SYR7) 89 . Further, we observed an increased abundance of proteasome proteins, PSMA4 and PSMD3, which are linked to protein metabolism 90 . Hence, we infer that the CAEC binucleation occurs as a compensatory mechanism to increase the cell capacity to metabolize the excess of ambient glucose by increasing the cell metabolic machinery (transcription/translation processes).
Despite a lack of apparent increase in cell proliferation in the experimental group compared to control group after 12 days, an increase in overall protein abundance was inferred from the total ion chromatogram (TIC) of MS (Fig. S1A). Previous studies have shown reduced endothelial cell proliferation (mostly in HUVEC) after long-term (7-14 days) HG exposure 4,11,[91][92][93][94][95][96][97] , accompanied by an increase in protein synthesis 97 . Although an increase in cell proliferation could boost a coordinated increase of ribosomal and proteasome proteins, we do not believe this is the case here, as mentioned before. After 4-5 days of simulated diabetes, cells occupied 100% of the well's plate surface, thereby impeding to harbor more cells because endothelial cells grow as a monolayer. This is consistent with findings stating that when endothelial cells become highly confluent, they www.nature.com/scientificreports/ stop growing due to cell-cell contact, even in the presence of growth factors 98 . In support of this, up-stream (CTGF [UniProtKB:O18739]) and CD62 [UniProtKB:P98107]) 99,100 (Table S5) and down-stream proteins (FABP4 [UniProtKB:P48035]) 101 (Table S5) involved in angiogenesis and proliferation were down-regulated by simulated diabetes. Importantly, there is evidence (not in endothelial cells) of cellular processes contributing to the stimulation of cellular binucleation without increases in cell proliferation, including cellular enhancement of antimicrobial defenses 102 , senescence 103 , and malignancy 104 . Various mechanisms have been linked to the binucleation process, such as cytokinesis failure, cellular fusion, mitotic slippage, and endoreduplication 105 . The elucidation of the exact molecular mechanisms leading to the binucleation process of CAEC is beyond the scope of our study. The chosen period (9-days) to simulate diabetes may represent a limitation, as longer periods may better reflect chronicity and trigger more severe endothelial dysfunction. The findings observed in BCAEC may not be equivalent to those occurring in human cells. However, the binucletation process triggered in BCAEC by simulated diabetes was also observed in human endothelial cells.
In conclusion, integration of omics and bio/chemoinformatics data revealed dysregulations in the metabolism of amino acids, peptides, and phospholipids, impaired insulin signaling, reduced mitochondrial mass, angiogenesis, and increased apoptosis and oxidative stress when CAEC were subjected to simulated diabetes. The appearance of non-proliferative binucleated CAEC cells is thought to be a strategy to metabolize the excess ambient glucose was also reported. Collectively, we believe that these dysregulated factors contribute to the development of CAEC dysfunction and may be associated with critical mechanisms underlying the onset of CVD in subjects with T2DM. We, therefore, suggest a multi-target therapeutic modality when protecting diabetic patients from CVD.