Transcriptome Analysis Reveals Altered Expression of Genes Involved in Hypoxia, Inflammation and Immune Regulation in Pdcd10-Depleted Mouse Endothelial Cells

Cerebral cavernous malformations (CCM) are capillary malformations affecting the central nervous system and commonly present with headaches, epilepsy and stroke. Treatment of CCM is symptomatic, and its prevention is limited. CCM are often sporadic but sometimes may be multifocal and/or affect multiple family members. Heterozygous pathogenic variants in PDCD10 cause the rarest and apparently most severe genetic variant of familial CCM. We carried out an RNA-Seq and a Q-PCR validation analysis in Pdcd10-silenced and wild-type mouse endothelial cells in order to better elucidate CCM molecular pathogenesis. Ninety-four differentially expressed genes presented an FDR-corrected p-value < 0.05. A functionally clustered dendrogram showed that differentially expressed genes cluster in cell proliferation, oxidative stress, vascular processes and immune response gene-ontology functions. Among differentially expressed genes, the major cluster fell in signaling related to inflammation and pathogen recognition, including HIF1α and Nos2 signaling and immune regulation. Validation analysis performed on wild-type, Pdcd10-null and Pdcd10-null reconstituted cell lines was consistent with RNA-Seq data. This work confirmed previous mouse transcriptomic data in endothelial cells, which are recognized as a critical tissue for CCM formation and expands the potential molecular signatures of PDCD10-related familial CCM to alterations in inflammation and pathogen recognition pathways.


Introduction
Cerebral cavernous malformations (CCM) are common vascular malformations derived from capillaries and small vessels of the central nervous system (CNS) [1]. Major clinical manifestations include intracranial haemorrhage, seizures and headache. Given the clinical unpredictability of CCM, surgery, stereotactic radiosurgery, pain medications and pharmacological prevention of seizures are the only therapeutic resources after neuroimaging detection of an otherwise unexpected lesion or, more commonly, after abrupt or subacute manifestations. Disease prevalence is estimated at 0.16-0.5% in the general population and often occurs sporadically [2]. More rarely, CCM may be multifocal and/or aggregate in families (familial CCM-FCCM) [3]. FCCM are caused by heterozygous, deleterious variants in either one of three genes encoding for interacting proteins, comprising Krev1 Interaction Trapped 1 (KRIT1; CCM1; MIM#604214), Malcavernin (alias MGC4607; CCM2; MIM#607929) and Programmed Cell Death 10 (PDCD10; CCM3; MIM#609118). Loss-of function is the prevalent molecular mechanism in FCCM. Genotype-phenotype correlations in FCCM are poor, and molecular data have limited clinical applications to date. More recently, the identification of a deleterious variant in either one of the known genes was considered mandatory for clinical trial enrolment in FCCM [4]. A better understanding of the biological diversity underpinning clinical variability in FCCM will improve prognostication, management planning and treatment approaches for future patients.
Alterations of PDCD10 are the rarest genetic cause of FCCM and tend to associate with a more aggressive phenotype with an earlier age of onset [5]. The encoded protein is identified as a key molecule for intracranial angiogenesis and endothelial cell homeostasis in both in vitro studies and animal disease models. In particular, studies in isolated endothelial cells show that Pdcd10-mediated pathways include Notch signaling, VEGF signaling and the ERK/MAPK pathway [6,7]. Zebrafish models reveal that Pdcd10 plays an essential role in early embryonic angiogenesis and cardiovascular development [8][9][10][11]. Furthermore, the murine Pdcd10 model shows that the Pdcd10 protein takes part in different intracellular signaling, which affects cell junction, apoptosis and stress responses [12]. Despite the many collected biochemical in vitro and in vivo data on PDCD10, the molecular pathogenesis of PDCD10-related FCCM remains only partially understood, and this lack of knowledge impacts the development of tailored patient's management.
Here, we explored the consequences of Pdcd10 silencing in mouse endothelial cells (ECs) by employing a transcriptomic analysis. This study allowed us to identify novel Pdcd10-controlled molecular pathways and offered the possibility of providing novel insights into FCCM pathogenesis and therapeutic targets.

Cell Lines
An immortalized mouse aortic EC line was generously gifted by Prof. Francesca Boccafoschi (Health Science Department, University of Piemonte Orientale, Novara, Italy). Cells were cultured in D-MEM with Glutamax supplemented with 20% FBS, 1% penicillin (100 U/mL) and streptomycin (100 µg/mL) (Thermo Fisher Scientific, Waltham, MA, USA) and grown in a 5% CO 2 incubator at 37 • C. For validation studies, immortalized mouse lungderived endothelial cell lines of either wild-type or knocked out for Pdcd10 (here named as EC-Ctrl, Pdcd10iEC-KO, respectively) and endothelial cell lines from Pdcd10 knockout mice, to which the human PDCD10 (here named as Pdcd10iEC-KO +Pdcd10 ) were re-added, were cultured as described in [13]. In brief, to generate Pdcd10 −/− cells re-expressing mGFP-tagged PDCD10, Pdcd10 −/− were transduced with the recombinant lentivirus Lenti ORF clone mGFP-tagged PDCD10 (OriGene Technologies Inc., Rockville, MD, USA). The human PDCD10 aminoacid sequence presents a single substitution (p.V192I) compared to the Pdcd10 mouse protein. The Lenti ORF clone mGFP-tagged PDCD10 vector was already used in mouse cells as reported in [13]. The recombinant lentiviruses were resuspended in serum-free MCDB-131 medium and added to the cells for 1 h at 37 • C. To increase the number of the cells, the cells were then passaged four times.

RNA Interference
Stealth RNAi duplexes designed against Pdcd10 (Thermo Fisher Scientific, Waltham, MA, USA) or stealth RNAi negative control (Thermo Fisher Scientific, Waltham, MA, USA) were transfected in EC cells (here named as siPdcd10-ECs and siCNT-ECs, respectively) using Lipofectamine RNAiMAX (Thermo Fisher Scientific, Waltham, MA, USA) and according to the manufacturer's protocol.

RNA Extraction
Total RNA was extracted using a mini RNase kit reagent (Qiagen, Hilden, Germany). The quality of nucleic acids was assessed using Nanodrop ND1000 (EuroClone, Milan, Italy). The RNA quantity was evaluated by Qubit RNA BR Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA). The RNA integrity was assessed by the RNA Integrity Number (RIN) using the Agilent RNA 6000 Nano Kit on the BioAnalyzer 2100 (Agilent, Boulder, CO, USA). All analyzed samples displayed a RIN above 9.50.

Library Preparation
Total RNA of siPdcd10-EC and siCNT-EC lines from three replicas of each cell type was quantified using the Qubit 2.0 fluorimetric Assay (Thermo Fisher Scientific, Waltham, MA, USA). A poly-A enriched library was generated with the TruSeq RNA-Seq Library Preparation Kit v2 (#RS-122-2001, Illumina, San Diego, CA, USA) according to the manufacturer's instructions. Library quality control was performed using the Agilent 2100 Bioanalyzer. Indexed libraries were sequenced at the CRS4 Next Generation Sequencing facility with the HiSeq 3000 instrument to generate~40 M 50 bp single-end reads per sample. Read and library quality was assessed by running FastQC (RRID:SCR_014583) and RSeQC (RRID:SCR_005275) [19] on FASTQ and aligned BAM generated with STAR. Transcript abundance was estimated with Kallisto [20], and differentially expressed genes (DEGs) were identified using DeSeq2 (RRID:SCR_015687) [21] R package with an FDR corrected p-value < 0.05. Enrichment analysis was performed with ToppCluster (RRID:SCR_001503) [22].

Quantitative PCR (qPCR)
Total RNA from siPdcd10-EC and siCNT-EC and from Pdcd10iEC-KO, EC-Ctrl and Pdcd10iEC-KO +Pdcd10 was reverse transcribed using the RT2 First Strand Kit (Qiagen, Hilden, Germany), according to the manufacturer's instructions. Oligos for the quantitative real-time PCR (Q-PCR) were designed using the Primer express program (RRID:SCR_014326) [23] with default parameters (Table S1). Gapdh and Actin were used as reference genes. The reactions were run in triplicate in 10 µL of final volume with 10 ng of sample cDNA, 0.3 mM of each primer and 1XPower SYBR Green PCR Master Mix (Thermo Fisher Scientific-Applied Biosystems, Carlsbad, CA, USA). Reactions were set up in a 384-well plate format with a Biomeck 2000 (Beckmann Coulter, Carlsbad, CA, USA) and run in an ABI Prism7900HT (Thermo Fisher, Scientific-Applied Biosystems, Carlsbad, CA, USA) with default amplification conditions. Raw Ct values were obtained using SDS 2.4 (Applied Biosystems, Carlsbad, CA, USA). Calculations were carried out by the comparative Ct method as reported in [24]. Significance was determined by a two-tailed unpaired t-test for means [24].
Clustering of genes for the final heatmap of differentially expressed genes was carried out using the PAM (Partitioning Around Medoids) method using the fpc R library (https://cran.r-project.org/web/packages/fpc/index.html published 6 December 2020). Enrichment analysis for Gene Onthology was conducted using the topGO package [26].
Several database sources were referenced for enrichment analysis, including Interpro, NCBI, MSigDB, REACTOME and WikiPathways. Enrichment was calculated relative to a set of background genes relevant to the experiment. The top 50 biological process terms for Elim p-value were analyzed with Revigo [27].
Confocal microscopy was performed using a confocal microscope (TCS SP5, Leica, Wetzlar, Germany), with the ImageJ software (NIH, New York, NY, USA) used for image analysis.

Pdcd10-Related Transcriptomic Profile
In order to identify novel molecular pathways potentially altered by Pdcd10 silencing, we carried out RNA-sequencing (RNA-seq) analysis in wild-type (i.e., siCNT-EC) and Pdcd10-silenced lines (i.e., siPdcd10-EC) from aortic murine immortalized ECs. An in vitro culture of mouse ECs were previously used for exploring the molecular pathogenesis of FCCM, as these cells can be considered the counterpart of human endothelial tissue [28]. We first silenced Pdcd10 in the EC line by the transfection of specific Stealth RNAi for Pdcd10. We found a reduction of~80% protein level in siPdcd10-EC compared to siCNT-EC by Western blot assay (Figure 1a  levels of protein intensity related to Pdcd10/β-Actin was quantified by densitometry using Image J analysis software, and the mean of each quantification was reported in the graph. Graphs show averages calculated on three different biological experiments represented by three points (green, yellow and orange). Scale bars represent standard errors. Values are expressed as mean ± SEM (* p < 0.05, n = 3). (c) Heatmap of gene ontology enrichment analysis of functional differences between siPdcd10-EC and siCNT-EC lines. The statistical significance in the heatmap was calculated and presented based on the -log10 false discovery rate (FDR) corrected p-values (blue indicates significant upregulated genes; red indicates significant downregulated genes). The colored scale bar below shows the color scaling with FDR values. The horizontal or vertical bars (violet, blue, orange and green) represented the different clusters of genes coming from a gene ontology analysis generated by Rosalind analysis. On the right-hand side of the Heatmap, a list of DEGs was reported. (d) Volcano plot showing the differentially expressed genes (violet points represent downregulated genes, green points represent upregulated genes, and the adjusted p-value threshold plotted on the Y-axis is 1.3). (e) Treemap representing over-represented biological functions, grouped into processes. Sizes of rectangles are proportional to the number of genes involved in a specific biological process. On the right of the Treemap, the more representative biological function for each cluster is indicated.

Validation Study of Differentially Expressed Genes in Mouse Endothelial Cells
To validate the transcriptomic data, we performed Q-PCR analysis in the lung-derived siPdcd10-EC line. We confirmed the upregulation of a set of genes prioritized based on their functional classification that was significantly perturbed in siPdcd10-EC lines, including ADAM Metallopeptidase Domain 8 (Adam8), Colony Stimulating Factor 2 Receptor Subunit beta(Csf2rb), Gbe1, Glycogen Synthase 1 (Gys1), Heme Oxygenase 1 (Hmox1), Nitric Oxide Synthase 2 (Nos2) and Serpin Family E Member 1 (Serpin1), compared to control cell line (Figure 3a). Csf2rb, Hmox1, Nos2 and Serpin1 resulted the most upregulated genes. Transcriptome data were also validated by an independent Q-PCR assay performed on mRNA from either Pdcd10iEC-KO and EC-Ctrl lines and from Pdcd10iEC-KO +Pdcd10 , in which the human PDCD10 was over-expressed [13] (Figure 3b). Through analysis of different biological pathways databases, we selected a set of the most representative biological processes (Table  S3). Firstly, we stained the Pdcd10iEC-KO and EC-Ctrl cells with the endothelial cell marker PECAM1 in order to verify the endothelial profile (Supplementary Figure S1). Then, we measured the expression of a set of DEGs associated with the main significant deregulated pathways, including hypoxia, HIF-1α, NOD2 and immunological-associated signaling, for which the functional association with PDCD10/Pdcd10 has not been established yet. We showed an upregulation of all tested genes in Pdcd10iEC-KO compared with EC-Ctrl lines, of which 11 resulted upregulated. Furthermore, we also addressed a rescue by a reduction of gene expression in Pdcd10iEC-KO +Pdcd10 cells (Figure 3c). Among them, Serpin 1 resulted more upregulated than the other analyzed genes.

Validation Study of Differentially Expressed Genes in Mouse Endothelial Cells
To validate the transcriptomic data, we performed Q-PCR analysis in the lung-de rived siPdcd10-EC line. We confirmed the upregulation of a set of genes prioritized based on their functional classification that was significantly perturbed in siPdcd10-EC lines, in cluding ADAM Metallopeptidase Domain 8 (Adam8), Colony Stimulating Factor 2 Recep tor Subunit beta(Csf2rb), Gbe1, Glycogen Synthase 1 (Gys1), Heme Oxygenase 1 (Hmox1) Nitric Oxide Synthase 2 (Nos2) and Serpin Family E Member 1 (Serpin1), compared to con trol cell line (Figure 3a). Csf2rb, Hmox1, Nos2 and Serpin1 resulted the most upregulated genes. Transcriptome data were also validated by an independent Q-PCR assay per formed on mRNA from either Pdcd10iEC-KO and EC-Ctrl lines and from Pdcd10iEC KO +Pdcd10 , in which the human PDCD10 was over-expressed [13] (Figure 3b). Through anal ysis of different biological pathways databases, we selected a set of the most representa tive biological processes (Table S3). Firstly, we stained the Pdcd10iEC-KO and EC-Ctrl cells with the endothelial cell marker PECAM1 in order to verify the endothelial profile (Suppl Figure). Then, we measured the expression of a set of DEGs associated with the main sig nificant deregulated pathways, including hypoxia, HIF-1α, NOD2 and immunological associated signaling, for which the functional association with PDCD10/Pdcd10 has no been established yet. We showed an upregulation of all tested genes in Pdcd10iEC-KO compared with EC-Ctrl lines, of which 11 resulted upregulated. Furthermore, we also ad dressed a rescue by a reduction of gene expression in Pdcd10iEC-KO +Pdcd10 cells (Figure 3c) Among them, Serpin 1 resulted more upregulated than the other analyzed genes.

Discussion
Here, we carried out a transcriptome profiling analysis in mouse endothelial Pdcd10 silenced cells and validated our findings in ECs obtained from Pdcd10 knockdown mice and from Pdcd10 knockdown mice re-expressing the human PDCD10 in a subset of se lected genes by choosing the genes associated with enriched signaling. Novel findings included pathway alterations of hypoxia, HIF-1α, NOD2 signaling, specific immunologi cal pathways, glycogen biosynthesis, End-MT and TNFα signaling.
PDCD10 encodes for an evolutionarily conserved protein physiologically involved in different intracellular signaling pathways such as cell junction, angiogenesis, apoptosis End-MT and stress responses [12,28]. PDCD10 is highly expressed in the neurovascular unit, and this explains the organ-specific manifestations of FCCM due to heterozygous loss-of-function variants in PDCD10. While current management of FCCM is sympto matic, the growing insights into the FCCM molecular pathogenesis are opening the path to innovative therapies aimed at preventing complications. From this perspective, there are two drug-repurposing clinical trials exploring the efficacy of propranolol and atorvas tatin in reducing disease manifestations in adults with CCM [4,29]. Hopefully, a deeper understanding of the subcellular and cellular mechanisms leading to CCM formation and rupture in FCCM will ease the identification of further candidate targets for known and novel molecules.

Discussion
Here, we carried out a transcriptome profiling analysis in mouse endothelial Pdcd10 silenced cells and validated our findings in ECs obtained from Pdcd10 knockdown mice and from Pdcd10 knockdown mice re-expressing the human PDCD10 in a subset of selected genes by choosing the genes associated with enriched signaling. Novel findings included pathway alterations of hypoxia, HIF-1α, NOD2 signaling, specific immunological pathways, glycogen biosynthesis, End-MT and TNFα signaling.
PDCD10 encodes for an evolutionarily conserved protein physiologically involved in different intracellular signaling pathways such as cell junction, angiogenesis, apoptosis, End-MT and stress responses [12,28]. PDCD10 is highly expressed in the neurovascular unit, and this explains the organ-specific manifestations of FCCM due to heterozygous lossof-function variants in PDCD10. While current management of FCCM is symptomatic, the growing insights into the FCCM molecular pathogenesis are opening the path to innovative therapies aimed at preventing complications. From this perspective, there are two drugrepurposing clinical trials exploring the efficacy of propranolol and atorvastatin in reducing disease manifestations in adults with CCM [4,29]. Hopefully, a deeper understanding of the subcellular and cellular mechanisms leading to CCM formation and rupture in FCCM will ease the identification of further candidate targets for known and novel molecules.
In order to highlight novel potential genetic targets, several transcriptomic studies related to both coding and noncoding RNA were conducted on CCM patients' tissues without molecular characterization [30][31][32][33]. These studies showed dysregulation of several signaling which clustered in neuronal activity, angiogenesis, extracellular matrix signaling and vascular integrity. Abou-Fadel and co-authors provided a combination of proteomic and transcriptomic analysis from silencing CCM genes in endothelial cells and from Ccm1 and Ccm2-knockout zebrafish embryos, revealing a unique portrait detailing alterations in angiogenesis and endothelial permeability [34].
To date, three RNA-Seq analyses aiming to profile the molecular role of PDCD10 in CCM pathogenesis were reported. The first one consisted of a transcriptomic study from brain lesions of Pdcd10 knockdown mice and identified alterations in neurological signal transduction, postsynaptic signaling and oxidative stress [35]. A combination of transcriptomic analysis derived from mouse and C. elegans endothelial Pdcd10-silenced cells revealed a set of genes related to integrin-signaling and vesicle transportation [36]. Recently, Orsenigo and co-authors reported an in-depth single-cell RNA sequencing in a Pdcd10-mouse model mapping the transcriptional diversity of endothelial cells in vascular lesions [37]. The amount of transcriptomic data reported, if confirmed and accurately validated in other cell lines and/or disease models, will surely stimulate the development of novel therapeutic strategies.
In the present study, we first confirmed alterations in pathways identified as abnormal in previous RNA-Seq studies in different tissues and including oxidative stress, integrinsignaling, vesicle transportation, angiogenesis and vascular integrity [28,[36][37][38][39]. Our investigations also identified the involvement of novel pathways, including hypoxia and HIF-1α signaling, NOD2-related pathway and immune response.

Hypoxia and HIF-1α Signaling
Many DEGs in this study were related to the hypoxia regulatory network, which is one of the most crucial pathways implicated in the control of the immune response, tissue homeostasis and endothelial signaling in the vasculature. HIF-1α is the key regulator of tissue response to hypoxia [40]. HIF-1α is critical for the development of atherosclerosis through cell-specific responses by acting on endothelial cells, vascular smooth muscle cells and macrophages. HIF-1α controls different pathophysiological processes, including vascular dysfunction, atherosclerosis, myocardial infarction and stroke. In our study, DEGs with at least a 1.5-fold increase in expression linked to hypoxia included cytokines/growth factors (N-Myc Downstream Regulated 1 (Ndrg1), Hmox1, Inhibitor of DNA Binding 2 (Id2), Family With Sequence Similarity 162 Member A (Fam162A), Solute Carrier Family 2 Member 1 (Slc2a1)), receptors (Gbe, Gys1) and other signaling proteins (Serpine1, Nos2, Solute Carrier Family 2 Member 1 (Slca1), Selenium Binding Protein 1 (Selenbp1), Phosphofructokinase, Platelet (Pfkp), Endoplasmic Reticulum Oxidoreductase 1alpha (Ero1l), Prolyl 4-Hydroxylase Subunit alpha2 (P4ha2), Carbonic Anhydrase 12 (Car12), Gys1, Fam162A, and Glucosaminyl (N-Acetyl) Transferase 2 (Gcnt2)). Among them, Serpine1, which encodes for a member of the serine proteinase inhibitor superfamily, is interesting as it contributes to innate antiviral immunity, and its expression is influenced by HIF-1α as a result of stimulation of cellular migration and cell-adhesion markers expression. Both these mechanisms, if altered, might affect permeability, which appears defective in FCCM patients' cell lines [41].

NOD2 Signaling
Our transcriptomic data also reported a significant transcriptional activation of Nod2associated genes. NOD2/Nod2 is an intracellular pattern recognition receptor that stimulates the host immune response. A variety of extracellular stimuli can activate distinct signaling pathways that converge to initiate NOD2/Nod2 expression. Specific cell wall components of bacteria and fungi can trigger the innate immune signaling cascade and then lead to NOD2/Nod2 expression. Following activation, NOD2/Nod2 stimulates proinflammatory pathways such as NF-κB and MAPK signaling [42] and thereby contributes to host defence via the production of inflammatory cytokines, antimicrobial molecules [43] and mucins [44]. More specifically, NOD2/Nod2 acts as an immune sensor in the gut microbiota balance and the related microbiota-host interaction. Research into the role of the gut microbiome in modulating brain function has rapidly increased over the past 10 years.
Increasing clinical and preclinical evidence implicates the microbiome as a possible key susceptibility factor for neurological disorders, such as Alzheimer's disease, autism spectrum disorder, multiple sclerosis, Parkinson's disease and stroke [45]. Interestingly, a recent study showed that CCM lesions arise from an excess of MEKK3 signaling downstream of TLR4 stimulation by the gut microbiome. This suggests the existence of a gut-brain disease axis in FCCM [46,47].
We demonstrated dysregulation of several genes which converge to NOD2/Nod2 signaling and include Csfr2b, Ndrg1, Car12, Csf2rb2, Semaphoring A7 (SemaA7), Fam162A and Slc2a1 [48][49][50][51]. In light of the recent discoveries of a possible role of the microbiota in the pathogenesis of CCM, our preliminary findings could be interpreted as a link between CCM formation and altered gut-microbiota interactions via Nod2 pathway dysfunction in PDCD10-related FCCM.

Immunological Signatures
Human T cells, CD4 + T and CD8 + T cells coordinate adaptive immune responses and are essential for establishing protective immunity and maintaining immune homeostasis through the production of cytokines and effector molecules. CD4 + T cells secrete cytokines to recruit and activate other immune cells, while CD8 + T cells acquire cytotoxic functions to directly kill infected cells [52]. The CNS is recognized as immune-privileged. However, recent advances highlight interactions between the peripheral immune system and CNS in controlling infections and tissue homeostasis [53,54]. One study suggested the role of inflammation in the CCM pathogenesis by revealing a robust inflammatory cell infiltration in CCM [55]. In our work, DEG analysis identified genes involved in the immune and inflammation response, such as Adam8, Gys1 and Elastin Microfibril Interfacer 2 (Emilin2).
ADAM8 was described as a promoter of macrophage infiltration upon inflammation [56]. GYS1 might be a novel therapeutic strategy for chronic inflammatory arthritis since its expression deregulation was associated with chronic inflammation in patient cell lines [57]. Finally, EMILIN2 stimulates the production of a number of cytokines involved in angiogenesis and inflammation [58]. Overall, the significant overexpression of these genes in our study suggests a relationship between Pdcd10 expression and the immune and inflammatory responses. These findings suggest that the immunological profile may be closely implicated in the CCM pathogenesis at least in PDCD10-related FCCM.

Conclusions
This work first confirmed previous studies showing gene expression alterations of oxidative stress, integrin-signaling, vesicle transportation, angiogenesis and vascular integrity in selected tissues of the Pdcd10-knockdown mouse model. Our findings reinforce the significance of these data and localize them in ECs, which are considered critical tissue for CMM formation. We also documented the involvement of novel pathways, including hypoxia, HIF-1α and Nod2 signaling, as well as immune response. Hopefully, these findings, if supported by further investigations and confirmed in other disease models, will contribute to the identification of a more personalized approach to disease prevention and treatment.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/genes13060961/s1, Figure S1: immunofluorescence analyses of PECAM 1 protein; Table S1: sequences of mouse primers used in this study for qRT-PCR study; Table  S2: all differential expressed genes; Table S2: biological process annotation clustering by different databases.