2. Methods 2.1. Isolation of Mouse Lung Endothelial Cells and Cell CultureMouse primary lung endothelial cells were isolated after digesting lung tissue from adult C57B/L6 mice (aged 6–8 weeks) with collagenase II by magnetic separation using bead-coated antibodies for endothelial cell surface markers CD31 and ICAM1, according to the method described in [9]. six-well cell culture dishes were precoated with 2% gelatin and 10 μg/mL fibronectin, and cells were plated in M199 medium (ThermoFisher, Waltham, MA, USA), 20% fetal calf serum superior (Biochrom GmbH, Berlin, Germany), 30 μg/mL endothelial cell growth supplement from bovine pituitary (Sigma-Aldrich, Burlington, MA, USA), 5 U/mL heparin, and antibiotics. Cells were expanded for 2 weeks in vitro and repurified by magnetic separation prior to seeding for gas exposure. Three individual cell cultures (prepared from 30 mice each) were used for label-free proteomic analysis. Another three independent cell cultures (prepared from 15 mice each) were used for mRNA expression analysis by qRT-PCR. 2.2. Cell Exposure to Different Oxygen ConditionsTrypsinized cells were plated into 6-well plates containing gas-permeable membranes (imaging plates; Zellkontakt, Nörten-Hardenberg, Germany) and transferred to custom-made boxes 24 h after plating (as described in [10]). Cells were exposed to different O2-conditions supplied by premixed gas bottles: (1) 21% O2—5% CO2—74% N2; (2) 95% O2, 5% CO2; (3) 0–95% O2 oscillations—5% CO2—rest N2, with a frequency of 6 oscillations per hour (as described in [11]). 2.3. Sample Preparation for Proteomic AnalysisAfter 24 h and 72 h of exposure to different oxygen conditions, cells were lysed with RIPA buffer (25 mM Tris/HCl pH 7.4; 150 mM NaCl; 0.5% sodium deoxycholate; 0.1% SDS; 1% NP-40; PIERCE protein inhibitor tablet). Proteins were quantified by a bicinchoninic acid-based protein assay (PIERCE- ThermoFisher, Waltham, MA, USA) and precipitated according to a modified method of Wessel and Fluegge [12]. 200 μg of protein was redissolved in 50 μL of 0.5% RapiGest (Waters Corporation, Milford, MA, USA) in 50 mM ammonium bicarbonate pH 7.8 and used for analysis. 2.4. Protein Identification and Semi-Quantitative Comparison by 2D-Liquid Chromatography/Mass Spectrometry (2DLC-MS)We have described our proteomic methodology in a previous paper from our lab [13]. Individual steps and settings are summarized here. 2.5. Enzymatic Digestion

Proteomic investigations were performed in three biological replicates. Protein samples were digested in-solution after the reduction in disulfide bridges and alkylation (reduction: 2 μL 100 mM DTT; 56 °C for 30 min; alkylation: 4 μL of 100 mM iodoacetamide; 40 min room temperature in the dark; quenching: 0.2 μL 100 mM DTT). Digestion was performed with 2.4 μL of 1 μg/μL trypsin at 37 °C overnight, and terminated by adding 12 μL of 10% formic acid.

2.6. 2D LC-MSMS Analysis

High pH reversed phase (RP) fractionation was performed using an Eldex micro HPLC pump (Sunchrom, Friedrichsdorf, Germany) on a RP column (Phenomenex, Kinetex 5u EVO C18 100A, 2.1 × 100 mm; Cat# 00D-4622-AN). Mobile phases were: A: 10 mM (NH4)HCO3, pH = 10; B:10% A in acetonitrile. Elution was performed at a flow rate of 150 μL/min with the following gradient: 5–40% B in 10 min, 40–95% B in 2 min, 95% B for 3 min 95–5% B in 2 min and 5% B for 7 min. 48 fractions were collected from 1 to 25 min and 4-4 fractions were combined (1, 13, 24, 37; 2, 14, 26, 38, etc.). This finally resulted in 12 fractions. Each of them was dried in a vacuum centrifuge, resolved in 0.1% formic acid in water and subjected to nanoLC-MSMS analysis on an LTQ-Orbitrap Elite (Thermo Fisher Scientific, Waltham, MA, USA) mass spectrometer. NanoUPLC runs were performed on a Waters nanoAcquity ULC system (Waters, Milford, MA, USA), using gradient elution after trapping the samples onto the trap column (186007238 Waters Symmetry C18, 0.180 mm × 20 mm, 5 μm, 100 Å) with 3% of B (mobile phase A: 0.1% formic acid in water; mobile phase B: 0.1% formic acid in acetonitrile) for 2 min with 10 μL/min flow rate. The analytical separation was performed with the following gradient elution: 3–10% B in 5 min and to 40% in 32 min, followed by double wash at the end of the gradient in order to reduce any carry over for the next run. The flow rate was 200 nl/min, the column (186003545 Waters BEH130 C18, 0.075 mm × 250 mm, 1.7 μm, 130 Å) was kept at 60 °C.

We applied data-dependent analyses: the 20 most intense peaks were selected for ion-trap CID after each survey scan. The survey spectra were measured in the Orbitrap (mass range: 380–1400 m/z; resolution: 120,000 @ 400 m/z), while the CID MS2 spectra were detected in the ion trap. Selected precursor masses were dynamically excluded for 15 s to facilitate more comprehensive analysis of the samples.

2.7. Data Evaluation

We used Proteome Discoverer (version 1.3, Thermo Fisher Scientific, Waltham, MA, USA) to generate msms peaklist files and our in-cloud ProteinProspector (version 5.18.0, UCSF, San Francisco, CA, USA) database search engine for protein identification. Peaklist files related to one sample were merged and used for searches against the bovine and mouse proteins from UniProtKB.2015.4.16. database, supplemented with their random sequences (109,150 sequences). Only fully specific tryptic peptides with a maximum of one missed cleavage site were considered. Carbamidomethyl cysteine as constant, methionine oxidation, peptide N-terminal pyro-glutamine formation from glutamine, and protein N-terminal acetylation were set as variable modifications. Error tolerance for precursor ions was set to 10 ppm and 0.6 Da for fragment masses. Maximum 2 modifications per peptides were allowed.

2.8. Semi-Quantitative Evaluation by Spectral Counting

Because there are a lot of proteins with a high degree of homology between mouse and bovine (source of bovine proteins: fetal calf serum of cell culture medium), it is challenging to decide the origin of the proteins. Database search results were acquired without homology and set MOUSE as the preferred species. The calculated false discovery rates (FDR) were less than 1%. In at least one of the groups, the protein from all of the three replicates had to be identified and at least 5 unique peptides had to match the protein in 2 cases from the 3 replicates. The semi-quantitative analysis relied on the number of spectra identifying the particular protein (peptide count/total number of peptide spectrum match) of the sample. To avoid the calculation problem when a protein showed spectral count zero in one of the samples, the following formula was used to calculate the relative spectral counts: RPC = (n + f)/(t + f), where RPC is the relative peptide count, n is the number of spectra identifying the protein, t is the total number of identified spectra in the sample, f is a correction factor, set to 1.

2.9. Functional Protein-Interaction and Pathway Enrichment AnalysisThe lists of proteins that were identified as significantly changed in expression after treatment with constant or intermittent hyperoxia were uploaded, including fold change values, to the web-based analysis platform NetworkAnalyst (3.0) [14] (www.networkanalyst.ca (accessed on 7 April 2022) using MOUSE as species and ID (Uniprot) as an identifier. Using this platform, differentially expressed proteins were compared with the protein interactome of the STRING database (cut-off: 700) generating first- and higher-order networks with the differentially regulated proteins from the list as seeds. The networks were downloaded as graphml file and imported into Cytoscape 3.8.2 (The Cytoscape Consortium, San Diego, CA, USA) [15]. The stringApp of Cytoscape was used to “STRINGify” the network for the species mus musculus followed by functional enrichment of pathways and gene ontologies. The STRING enrichment table was ranked according to FDR values and filtered for Reactome, KEGG pathways and Gene Ontologies. Detailed results of the individual analysis are provided as Supplemental Information (Supplementary Data S2). 2.10. Quantitative Real-Time PCRFor quantification of protein gene expression, mRNA was isolated from cells after exposure to different oxygen conditions using the Rneasy plus kit (Qiagen, Hilden, Germany). Reverse transcription was performed using qScript Supermix (Quanta Biosciences, Gaithersburg, MD, USA), and resulting cDNA was analyzed by quantitative real-time PCR on a RotorGene Q (Qiagen, Hilden, Germany). Changes in gene expression were calculated relative to the control condition (21% O2) using the ΔΔCt method and Actb or Gadph as house-keeping gene. Primer sequences are provided in Supplementary Data S1. 2.11. Statistical Analysis

Proteomic data were generated from three independent cell preparations. Fold changes in protein levels were statistically evaluated by Student’s t-test (two sample t-test, assuming unequal variances) to decide whether the difference between the two groups (with 3 replicates) was significant (p < 0.05). Comparing different groups meant levels of individual proteins under constant or intermittent hyperoxia relative to protein levels after culture under normoxia (21% O2) for the same period of time (24 h or 72 h).

Since there was a significant variance between groups, the averages of the relative peptide counts were compared and at least two-fold changes were considered as real differences. As it was assumed that most of the proteins did not show significant changes, the relative peptide count ratio normalization by the median value was performed.

qPCR analysis was performed from three independent experiments. The graphs show the results from one representative experiment analyzed from 3 individual wells (triplicates) of gas-exposed cell culture dishes. qPCR data were statistically evaluated by one-way ANOVA followed by Dunnett’s multiple comparison test using GraphPad Prism 8.0 software (GraphPad Software, San Diego, CA, USA). Mean expression levels were further compared to the value 1.0 by using a one-sample t-test. In all statistical tests p < 0.05 was regarded as the threshold for significant results.

4. DiscussionThe special properties of the pulmonary vascular beds allow adaptation to different requirements with regard to flow rates depending on cardiac output, thereby optimizing perfusion and gas exchange. The pulmonary endothelium regulates barrier function, vascular tone and immune responses, is involved in various signaling pathways, counteracts thrombosis, and has a special active metabolism [17]. Barrier function is maintained by multi-protein complexes, which form adherent junctions, tight junctions, and gap junctions, that control flow of fluids and transmigration of proteins and cells. Barrier disintegration leads to pulmonary edema, which is a hallmark of several lung diseases, including acute lung injury (ALI) and its worse form, acute respiratory distress syndrome (ARDS).In accordance with being a syndrome, ARDS is triggered by different (direct and indirect) insults and appears in different phenotypes with various degrees of hypoxemia, endothelial and epithelial injury, inflammation and aberrant coagulation. Accordingly, there have been a lot of efforts to sub-characterize the condition with regard to sub-phenotypes and endo-types in order to provide better suitable therapies and prognosis [18]. Precision medicine approaches, for instance, have been helpful in distinguishing a hyperinflammatory versus an uninflamed endo-type characterized by different plasma levels of inflammatory biomarkers, such as IL-6, IL-8, sICAM-1, and sTNFRI, that exhibited different response to treatments and outcome [19].Central to all ARDS treatment regimens is providing sufficient oxygenation. There has been considerable argumentation regarding whether this can only be achieved by very high oxygen saturation, and at the same time taking the risk of oxygen toxicity. A recent clinical trial compared liberal (target PaO2: 90–105 mm Hg; SpO2 >96%) and conservative (SpO2: 88–92%) oxygen therapy in ARDS patients, implying a possible worse outcome in the conservative-oxygenation strategy with regard to 90-days mortality [20]. However, responses to oxygen therapy can also be different in ARDS sub-phenotype groups depending on genetic factors and oxidative stress levels, which are linked to inflammation.The pulmonary vasculature has the special capability of sensing oxygen. Hypoxic conditions lead to the vasoconstriction of small pulmonary arteries, while systemic arteries dilate (=hypoxic pulmonary vasoconstriction, HPV) [21]. This mechanism redirects blood flow to better ventilated areas. Hyperoxia, on the other hand, leads to vasoconstriction in systemic microcirculation and high concentrations of oxygen further induce toxicity in the lungs, which is an issue in patients ventilated with supraphysiological oxygen. Translational animal experiments have shown that high oxygen aggravates ventilation-induced lung injury (VILI) with regard to pulmonary edema and inflammation [6]. Baboons exposed to hyperoxia revealed the destruction of endothelial cells and alveolar type I cells, interstitial edema, and activation of neutrophils [22]. Despite a number of animal experiments using hyperoxia exposure followed by proteomic analysis [23,24,25], detailed studies of molecular effects of hyperoxia on isolated cell types have only been described for alveolar type II cells [26], but not for the pulmonary endothelium.

In this study, we therefore aimed to decipher changes of the proteome in pulmonary endothelial cells in response to chronic constant and intermittent hyperoxia, which might help to better understand detrimental impact of oxygen on the organ lung.

Analysis of the mRNA levels of selected proteins revealed dynamic changes of expression starting from as early as 4 h until 72 h of exposure. For quantitative proteomic analysis, however, we chose exposure times of 24 h and 72 h. These data showed that constant and intermittent hyperoxia induce different responses with regard to the proteome. A total of 24 h of constant severe hyperoxia upregulated pathways related to RNA metabolism, cell cycle, mRNA splicing, cellular stress response, interleukin signaling, and apoptosis, and downregulated translation, processes involving small GTPases, TCA cycle and respiratory electron transport, VEGF (vascular endothelial growth factor) signaling, platelet activation, and cell junction organization. After 72 h of constant severe hyperoxia, enriched pathways shift to an upregulation of protein metabolism, ribosomes, and intracellular transport, while cell cycle, DNA repair, TCA cycle, respiratory electron transport, and cellular stress response are downregulated. These dynamics might reflect the situation encountered by a relatively fast start-up response of mRNAs (as measured at 4 h by qRT-PCR), that in many cases is counter-regulated at later time points and is also translated into proteins with the necessary time delay. Here, feedback mechanisms might play an important role, as can be anticipated as an example of “cell cycle”, which is upregulated at 24 h and again downregulated at 72 h 95% O2.

24 h of intermittent hyperoxia (0–95% O2) downregulates calcium homeostasis, responses to oxygen compounds, phosphorylation, and platelet activation. After 72 h (0–95% O2), posttranslational modifications such as neddylation and endocytosis are upregulated, while cell cycle, DNA repair, and p53 signaling are downregulated. Neddylation is a post-translational conjugation of the ubiquitin-like molecule neural precursor cell-expressed developmentally downregulated protein 8 (Nedd8) to different substrates, such as cullins, Akt, Hdac2, Hif1α, Hif2α, IKKγ, Traf6, Myd88, PPARγ, and Pcna, and affects transcription factors such as Nrf2 and NF-ΚB, the expression of pro-inflammatory cytokines, and barrier function [27].It can be observed that under both conditions, long-term constant and intermittent hyperoxia, cells cease to proliferate, which is seen in a reduced cell count [28].From what is already known, both conditions—constant and intermittent hyperoxia—will induce oxidative stress in the pulmonary endothelium, but the mechanisms and sources of ROS might be different. Molecular mechanisms of intermittent hyperoxia have not been elucidated in detail so far. In some of our previous studies, our group found that intermittent hyperoxia blunts the inflammatory response elicited by constant hyperoxia [29], activates the renin-angiotensin-system (RAS), and generates large amounts of peroxynitrite in the pulmonary endothelium [30]. Interestingly, nitric oxide (NO) has been shown to inhibit the activation of NF-ΚB induced by hyperoxia in neonatal pulmonary microvascular endothelial cells [31]. There is a mechanistic concept for molecular events in intermittent hypoxia proposed by Nanduri [32], according to which NADPH oxidase-derived ROS activates PLCγ and produces a calcium signal via inositol-3-phosphate, that ultimately activates HIF-1α. Interestingly, our proteomic study also highlights the inositol-3-phosphate receptor as a protein affected by hypoxic/hyperoxic O2 oscillations, implicating the role of calcium signaling under these conditions, and KEGG pathway enrichment analysis implicates HIF-1 pathway involvement at 24 h 0–95% O2. Further detailed mechanistic studies into these issues are certainly required.Previous studies have shown that alternating oxygen between hypoxia and hyperoxia leads to the activation of signaling pathways that resemble hypoxic responses (such as HIF1 activation) without their detrimental side effects (“hyperoxic-hypoxic paradox”) [33]. Repeated oscillations between hyperoxia and hypoxia are believed to increase the ROS scavenger/ROS ratio, thereby protecting cells from ROS damage. Moreover, the oxygen oscillations present a therapeutical stress on mitochondria, encouraging the elimination of damaged organelles and the biogenesis of new, healthy mitochondria. The hypoxic periods allow the activation of HIF1, VEGF (angiogenesis, arteriogenesis), and stem cell proliferation. Clinical therapies making use of this phenomenon include Intermittent Hypoxic-Hyperoxic Treatment (IHHT) [34] and Hyperbaric Oxygen Treatment (HBOT) [35]. In a clinical setting, IHHT has been shown to improve the lipid profile and anti-inflammatory status in patients with metabolic syndrome [36]. An interesting issue for further mechanistic investigations might be the crosstalk between HIF and NF-ΚB transcription factors, which have been shown to be interlinked in a cell-type specific way [37].In context of such mechanistic insights, an old therapeutical concept obtrudes, suggesting the use of pharmaceuticals or nutraceuticals to tackle oxidative stress and inflammation. There are many “natural” candidates, such as antioxidants resveratrol, N-acetyl cysteine (NAC) or anti-inflammatory sulforaphane from broccoli, curcuminoids or short-chain fatty acids (SCFA) acetate, propionate and butyrate, which are metabolites of a healthy gut microbiome. In addition, there is a large number of pharmaceutical drugs available. Interestingly, despite some promising results from preclinical in vitro and animal studies, the translation into human medicine is frequently difficult [38]. Resveratrol has been shown to induce apoptosis and autophagy in cancer cells [39] and attenuates inflammation in allergic asthma [40]. Similarly, SCFAs have been shown to exhibit anti-cancer and anti-inflammatory activity in cell cultures [41]. There is plenty of evidence that gut dysbiosis underlies many diseases, including diabetes mellitus, atherosclerosis, depression, and pulmonary arterial hypertension [42]. Curcuminoids are known to inactivate NF-KB and thereby decrease the transcription of pro-inflammatory mediators. In addition, they are capable of modulating the immune response and are promising drug candidates in cancer therapy [43]. Due to the chemical properties of these compounds, they are not readily bioavailable and are poorly assimilated upon ingestion. These problems with efficacy are addressed by efforts to encapsulate the substances and to use to lipid carriers for delivery to targets [44].On the other hand, we are exposed to environmental toxins which have an impact on the antioxidant and inflammatory status of the respiratory system, thereby most likely also increasing the probability of developing ARDS during lung disease. One example are endocrine disruptors, which are part of plastics (“microplastics”, for example Bisphenol A) or are used as fungicides in agriculture (for example, Vinclozolin). These substances not only have an impact on the reproductive system, but also other organs including the lung by affecting Nrf2/NF-ΚB pathways [45,46].Oxygen therapy in cancer patients seems to be a double-edged sword, and its benefits and potential harmful effects are a matter of continuous debate. Tumor microenvironments are frequently hypoxic due to rapid cell growth, and, despite the neovascularization, due to limited oxygen supply. Switching the metabolism of cancer cells to hypoxia plays a role in metastasis [47]. Hypoxia also fosters resistance to cytotoxic CD8+ T-cell (CTL)-attack by different mechanisms, and upregulates the expression of PD-L1 to increase CTL apoptosis. Immune checkpoint inhibitors (ICIs) have been developed that block PD-L1, PD-1, or CTLA-4. These drugs have been shown to improve overall survival in cancer patients, but have also adverse (cardiac) side effects [48]. Mechanisms of action are an increase in NLRP3, MyD88, and interleukin signaling inducing a cytokine storm. In order to reduce tumor hypoxia and also to increase ROS that might support cancer therapy, the application of oxygen has been proposed as a supplemental measure [49]. Overall, several studies have resulted in mixed outcomes; therefore, a general recommendation has not been given. A reason for this might be the challenge of fully understanding the regulation of oxygen homeostasis in the organism in its complexity, ROS signaling function, and detrimental side effects and the network of differentially regulated antioxidative response genes. Additional oxygen not only alleviates hypoxia, but can also induce inflammation; therefore, the right dosage reaching the target is difficult to obtain. Moreover, respiratory oxygen application efforts are made to enclose oxygen in microcapsules, which was shown to improve immune checkpoint blockades in pre-clinical studies [50]. This might also be an issue, especially in the context of anti-cancer drugs such as doxorubicin and angiogenesis inhibitors, where inflammation and endotheliotoxicity are significant risks [51]. Limitations of the Study

This study was performed using murine primary pulmonary endothelial cells, which—as discussed previously—are prone to relatively rapid dedifferentiation in vitro after isolation from the lungs. We tried to tackle this problem by repurifying cells in cultures prior to the experiment, using endothelial-specific surface markers in order to make sure the cells maintained the endothelial properties. We always quantified changes in protein expression related to the normoxic condition at the same time point in order to eliminate changes in protein expression attributed to in vitro culturing.

It could be argued that another issue is that our “normoxic” condition (=21% O2) is not “physoxia” for lung endothelial cells (5–10% O2), but rather already a slightly hyperoxic condition. The pulmonary endothelium and other microvascular endothelial cells sense oxygen and its metabolism in an organ-specific way [34]. Reiterer et al. [34] show that maintaining cells at supra-physiological O2 levels impairs a normal response to hypoxia. However, in our study we merely investigate responses to severe hyperoxia and oscillations around a hyperoxic mean value compared to ambient O2 concentrations (21% O2).

A technical limitation of this study is owed to the fact that the yield of endothelial cells from mouse lung is low. For the sake of reducing animal numbers in the sense of the “3R Principle (Replacement-Reduction-Refinement of animal experiments”, we limited our proteomic analysis to the minimum of necessary replicates for a (semi-)quantitative assessment of changes in protein expression. However, results have been supported by independent mRNA expression analysis and are in good accordance with previous findings, as discussed in this paper.