SILAC-based proteomics of human primary endothelial cell morphogenesis unveils tumor angiogenic markers

MS proteomic studies, CLEC14A immunoprecipitation and immunofluorescence staining were isolated from umbilical cords and cultured for 3-4 passages in M199 supplemented with 100 mg/l heparin, 20% FBS (Gibco), and bovine brain extract (referred to as M199 complete). For all other experiments, HUVECs were cultured in EGM-2 (Lonza). For the proteome study, three experiments were performed; for replicate 1 (exp1) and 2 (exp2) HUVECs at the second passage (P2) were used; for replicate 3 (exp3) at P3. study of ECs morphogenesis and spreading provides valuable information that can help to unravel molecular mechanisms regulating endothelial cells behavior. Our data may contain the basis for further exciting insights to be investigated in angiogenesis in vivo, especially with a view to interfering with this process in pathologies.


Summary
Proteomics has been successfully used for cell culture on dishes, but more complex cellular systems have proven to be challenging and so far poorly approached with proteomics. Because of the complexity of the angiogenic program, we still do not have a complete understanding of the molecular mechanisms involved in this process, and there have been no in depth quantitative proteomic studies. Plating endothelial cells on matrigel recapitulates aspects of vessel growth, and here we investigate this mechanism by using a spike-in SILAC quantitative proteomic approach. By comparing proteomic changes in primary human endothelial cells morphogenesis on matrigel to general adhesion mechanisms in cells spreading on culture dish, we pinpoint pathways and proteins modulated by endothelial cells. The cell-extracellular matrix adhesion proteome depends on the adhesion substrate, and a detailed proteomic profile of the extracellular matrix secreted by endothelial cells identified CLEC14A as a matrix component, which binds to MMRN2. We verify deregulated levels of these proteins during tumor angiogenesis in models of multi-stage carcinogenesis. This is the most in depth quantitative proteomic study of endothelial cell morphogenesis, which shows the potential of applying high accuracy quantitative proteomics to in vitro models of vessel growth to shed new light on mechanisms that accompany pathological angiogenesis. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium with the data set identifier PXD000359.
Angiogenesis, the process of new blood vessels growth from existing ones, is crucial to provide oxygen and nutrients to the body during development and in diseases such as cancer, where primary tumors need blood supply to fuel their growth [1]. During this process, endothelial cells (ECs) sprout, digest the extracellular matrix (ECM), differentiate, migrate, proliferate and change morphology to form a new capillary with a lumen. Finally, the ECs mature cell-cell adhesion structures, secrete new ECM and recruit mural cells to stabilize the newly formed vessel. In pathological conditions these cellular processes are deregulated. For example, tumor blood vessels have abnormal ECM structure and composition [2,3], which may results in leakiness and chaotic blood flow, and negatively impact conventional anti-cancer treatments [4]. Extensive work is therefore devoted to the characterization of molecular mechanisms regulating ECs in angiogenesis and to discover therapeutic targets with improved efficacy to interfere with this process [5,6].
In vitro models of angiogenesis have been established in which ECs cultured on or embedded in a thick layer of ECM assemble into tubular structures complete with a lumen-like formation. Particularly, ECs plated on matrigel form a tubular network within hours. With this system some of the cellular processes required in the in vivo situation can be investigated, such as cell-ECM and ECM remodeling, morphogenesis and proliferation [7].
Several studies of transcriptome changes associated with in vitro models of angiogenesis, such as matrigel and collagen I assays, have profiled gene regulation and begun to unravel protein functions associated with endothelial morphogenesis [8,9]. However, mRNA levels do not always correlate with protein levels and the cell-ECM interaction is mediated by protein complexes. This makes global MS-based proteomics an attractive approach, but so far this technology has not been explored much in the angiogenesis field [10] and the EC proteome during morphogenesis has been investigated only at a low depth [11]. We reasoned that an unbiased and comprehensive proteomic study might uncover novel proteins and cellular processes important for angiogenesis.
In the last decade MS-based technology has improved enormously and it is becoming a versatile tool to investigate proteomes in depth, and to accurately quantify their dynamics. In this context, high resolution mass analyzers used in combination with quantitative strategies, such as SILAC, are becoming invaluable to explore biological systems in vitro and in vivo [12]. However, the SILAC-labeling of primary ECs is still challenging and the several passages needed for the full incorporation of the labeled amino acids have so far prevented the use of the cells at early passages, which is critical to avoid interference of cultureinduced differentiation of primary cells.
In this study we have employed SILAC in an indirect "spike-in" fashion to accurately quantify a large proportion of the proteome dynamics of early passage human umbilical vein endothelial cells (HUVECs), a cell model that has been largely characterized and used to study angiogenesis in vitro, forming tubules on matrigel or spreading on different ECMs. We provide a detailed portrait of the proteomic changes associated to EC morphogenesis and, in combination with MS-proteomic analyses of protein-protein interaction and profiling of ECM proteins secreted by HUVECs, we discovered CLEC14A and MMRN2 as ECM components highly regulated and functional in EC morphogenesis, and deregulated in specific stage of tumor angiogenesis. This shows the power of combining MS quantitative proteomics to in vitro models of angiogenesis for a deeper understanding of the molecular mechanisms regulating ECs in angiogenesis, and relevant in pathology.

Reagents
Matrigel basement membrane matrix and cell recovery solution were from BD Biosciences. Bovine plasma FN, gelatin, Engelbreth-Holm-Swarm murine sarcoma laminin, sheep IgG, and rabbit anti-MMRN2 used for mouse tissues immunofluorescence were from Sigma; sheep polyclonal anti-CLEC14A (AF4968) and anti-CD31/PECAM-1 (AF806) were from R&D Systems; goat polyclonal (K-17) anti-MMRN2, used for immunoblotting, and mouse monoclonal (6D365) anti-MMRN2 used for HUVECs immunofluorescence were from Santa Cruz Biotechnology; purified rat monoclonal anti-Panendothelial cell antigen (clone Meca32) was from BD Pharmingen; rabbit polyclonal anti-FN1, and rabbit anti-NG2 (chondroitin sulphate proteoglycan polyclonal) were from Chemicon; rabbit anti-LAMA4 was kindly provided by Prof. Lydia Sorokin; rabbit anti-CPT1A was from ProteinTech Group. All secondary antibodies used for immunofluorescence were Alexa Fluor 555 or 488 from Molecular Probes. DAPI nucleic acid stain was from Invitrogen; trypsin, sequencing grade modified was from Promega.

Endothelial cells culture and transgenic mouse models
SILAC HUVECs were grown for four passages in custom M199 (Gibco), without arginine and lysine, and supplemented with 42 mg/l of 13 C 6 15 N 4 L-arginine (that we refer to as heavy arginine), 73 mg/l of 13  The generation of RIP-Tag2 mice has been previously described [13]. RIP-Tag2 mice were maintained in the C57Bl/6J background (The Jackson Laboratory). From 12 weeks of age, all RIP-Tag2 mice received 50% sugar food (Harlan Teklad) and 5% sugar water to relieve hypoglycemia induced by the insulinsecreting tumors. Generation of K14-HPV16 transgenic mice has been previously reported [14]. K14-HPV16 mice were maintained in the FVB/n background (The Jackson Laboratory

Extracellular matrix and supernatant isolation and analysis.
The ECM for MS analysis was isolated according to [19] with minor modifications. Briefly, HUVECs were cultured for 7 days in M199 complete (see above). Cells were then washed with PBS and incubated with a solution 0.5% Triton X-100, 20mM NH 4 OH and 1mM EDTA. Detached cells were aspirated and the remaining ECM thoroughly washed with PBS. The ECM was then solubilized in SDS lysis buffer: 4% SDS, 100mM DTT and 100mM TrisHCl pH 7.6, followed by incubation at 95°C. For MS analysis, proteins from three replicates were in-gel digested and loaded onto StageTips, as described above, and analyzed by LC MS/MS (LTQ-Orbitrap Velos). Only proteins with an intensity measured in the three replicates were considered as components of the HUVEC ECM proteome. See below for more details.
For the analysis of CLEC14A distribution in cells, ECM and supernatant ( Figure 4C), HUVECs were cultured confluent for several days in EGM-2. ECM was prepared as described above; for the supernatant, cells were washed with PBS and incubated in EBM-2 for 6h. The medium was centrifuged, 12,000 rfc for 20 min at 4 C. To enrich for secreted proteins, the cleared supernatant was incubated with Strataclean beads (Agilent technologies) according to manufacturer protocol. For the cells, HUVECs were harvested with accutase and lysed in SDS lysis buffer.

MS data acquisition and analysis
Digested peptides were analyzed by EASY-nLC system (Thermo Fisher Scientific) coupled on line to a LTQ-Orbitrap XL (for the EC morphogenesis and the spreading, and immunoprecipitation studies) or Velos (for the ECM study) (Thermo Fisher Scientific) via a nanoelectrospray ion source (Thermo Fisher Scientific). Chromatographic peptide separation was done in a 15 cm fused silica emitter (Thermo Fisher Scientific) packed in house with reversed-phase Reprosil (Dr. Maisch GmbH) and eluted with a flow of 250 nl/min from 5% to 70% ACN in 0.5% acetic acid, in a 140 min gradient. The full scan MS spectra were acquired with a resolution of 30,000 at m/z 400 in the Orbitrap. The top 5-10 most intense ions were sequentially isolated for fragmentation using CID (for the EC morphogenesis and spreading, and immunoprecipitation studies) or high-energy collision dissociation (for the ECM study), and recorded in the LTQ or Orbitrap, respectively. In the determination of CLEC14A phosphorylation sites, the neutral loss algorithm in the Xcalibur software was enabled for each MS/MS spectrum [20]. Data were acquired with Xcalibur software (Thermo Fisher Scientific). The MS files were processed with the MaxQuant software version 1.2.6.20 [21] and searched with Andromeda search engine [22] against the human UniProt database [23] (release-2012 01, 81,213 entries). To search parent mass and fragment ions, an initial mass deviation of 6 ppm and 0.5 Da (CID) or 20 ppm (HCD), respectively, were required. The minimum peptide length was set to 7 amino acids and strict specificity for trypsin cleavage was required, allowing up to two missed cleavage sites. Carbamidomethylation (Cys) was set as fixed modification, whereas oxidation (Met) and N-acetylation were considered as variable modifications. No labeling or double SILAC labeling was defined accordingly. The false discovery rates (FDRs) at the protein and peptide level were set to 1%. Scores were calculated in MaxQuant as described previously [21]. The reverse and common contaminants hits (in the ECM proteome analysis, KRT1 and KRT9 were additionally included), were removed from MaxQuant output. Only proteins identified with at least one peptide uniquely assigned to the respective sequence were considered for the analysis (for proteins identified with single peptide ID, see annotated spectra in Table S9). For SILAC protein quantification, the re-quantification feature was enabled, and the relative quantification of the peptides against their SILAC-labeled counterparts was performed by MaxQuant. Only unique peptides were used for quantification and we required proteins being quantified with at least two ratio counts.
For the immunoprecipitation and ECM analyses, proteins were quantified according to the MaxQuant label-free algorithm [24]; unique and razor (=most likely belonging to the protein group) peptides were used for protein quantification.
The .raw MS files and search/identification files obtained with MaxQuant have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org/cgi/GetDataset) via the PRIDE partner repository [25] with the dataset identifier PXD000359.

MS data normalization
For the SILAC data, the H/L normalized ratios in the analyzed MaxQuant output file were inverted and log 2 transformed. The distribution of the ratios was further normalized to the median (Table S2). For quantification, the protein groups were filtered for those with at most one missing quantification value within the three replicates, in each experimental condition (3,723 protein groups in Table S2).

ANOVA test and one dimensional analysis
For the ANOVA test, the three replicates for each adhesion condition were grouped together and the statistical test corrected with permutation based FDR of 0.5% in a similar way as described for t-tests by Tusher and coworkers [26] (Table S2, "Quantified proteins" datasheet).
The one-dimensional analysis to identify the regulated KEGG categories [27] was performed on the subset of protein ratios that passed the ANOVA test. For each of the KEGG pathways ("KEGG" column in Table S2) it was tested whether the corresponding ratios (column "Sample/0h") had a preference to be systematically larger or smaller than the global distribution of protein abundance values. Multiple hypothesis testing was controlled by using a Benjamini-Hochberg FDR threshold of 5%. For categories that passed the test (Table S5) the median ratio of the proteins belonging to the category was calculated and shown as heat map in Figure3. For details see [28]. For the tissues, frozen sections were air-dried and fixed with Zinc-fixative (calcium acetate 0.5 g, zinc acetate 5.0 g, zinc chloride 5.0 g, 0.1M Tris buffer for 1000 ml of solution, pH 6.5-7). Samples were blocked with 1% BSA and 5% donkey serum in PBS for 1h at room temperature. Tissues were incubated with the primary antibodies. After washing with PBS, samples were incubated with Alexa-conjugated secondary antibodies and counterstained with DAPI. Stainings were analyzed by using confocal laserscanning microscope (TCS SP2 AOBS, Leica Microsystems).

Immunofluorescence
To analyze Clec14a and Mmrn2 protein levels (red channel) during RIP-Tag2 and HPV/E 2 tumorigenesis the immunofluorescence image acquisition was performed maintaining the same laser power, gain and offset settings. Ten different fields for each mouse at different stage of tumor progression (n=5) were considered. In each 400x power picture we selected a region of interest (ROI) close to each blood vessel (green channel). We quantified the mean fluorescence intensity of red and green channels by means of the Leica Confocal Software Histogram Quantification Tool. We calculated the ratio of red to green channel mean fluorescence intensity. Then, the quantification of Clec14a and Mmrn2 abundance was normalized on vessel density and total tumor vascularity.

3D reconstruction of mouse vessel
High-resolution confocal image stacks of ECs, pericytes and Clec14a were reconstructed by isosurface rendering using Imaris software (version 6.4.0, Bitplane, AG, Zurich, Switzerland). Isosurface rendering is a computer generated representation of a specified range of fluorescence intensities in a data set that allow the creation of an artificial solid object of a specific area. Selected confocal images were imported into Imaris to obtain a precise three-dimensional vessel reconstruction and to better point out the colocalization between ECs, pericytes and Clec14a.
Immunostaining was performed with LSAB+ Kit (Dako) according to the manufacturer's instructions.
Slides were counterstained with haematoxylin. Analysis of the samples followed an informed consent approved by the local ethics committee (Department of Pathology, Wroclaw Medical University, Poland).

Matrigel and 3D tubulogenesis
The matrigel assay performed with siRNA HUVECs was carried on in 96 well plate. Briefly, 6.5-10 E3 HUVECs were seeded on 30µl of solidified matrigel in 100 µl of EGM-2 with or without stimulus and let grown at the indicated time.
For the 3D tubulogenesis assay, collagen-coated beads were covered with HUVECs, embedded into fibrin gel and co-cultured with fibroblasts (CAF2, [29]) as previously described [30]. Quantifications were performed after 11 days of co-culture.

siRNA
The day before oligofection, HUVECs were seeded in six-well plates at a concentration of 2.0 E4 cells/well. Oligofection of the siRNA duplexes was performed according to the manufacturer's protocol.
After 48h since the oligofection, HUVECs were used for experiments.

Statistical analysis
For Clec14a and Mmrn2 protein levels in mouse vessels and 3D tubulogenesis p-values were calculated using a two-tail Mann Whitney test. For other experiments, unless indicated otherwise, p-values were calculated using a two tail unpaired t-test. Calculations were performed using GraphPad Prism. * = p < 0.05; ** p < 0.01; *** p<0.001.

Quantitative proteomics of HUVECs during tubulogenesis and spreading
To follow proteomic changes that contribute to vessel growth, we performed a time-resolved quantitative proteomic study of HUVECs plated on matrigel (Matr) for 0h (cells harvested but not plated on the matrix), 12h, 24h and 30h and on growth factor reduced (GFR) matrigel. Network structures were clearly evident after only 12h, and by 30h there were fully formed tubule. To contrast the morphogenetic process to general adhesion mechanisms, we seeded HUVECs for 24h on culture dishes coated with ECMs For accurate, in-depth characterization of the proteomes in these different environments we employed SILAC in an indirect 'spike-in' format [31]. HUVECs in culture were metabolically labeled with heavy arginine and lysine as a reference, and then spiked into the lysate of each of the samples described above.
This approach allowed us to profile the proteome of HUVECs at early passage (P), P2-P3. Mixed proteomes were then analyzed by online liquid chromatography mass spectrometry (LC MS/MS). We employed a linear ion trap Orbitrap instrument and processed the MS data in the MaxQuant environment [21] ( Figure 1A).
MS analysis was performed on three independent replicates, which unambiguously identified 6,678 proteins. Of those, 3,723 were accurately quantified in all adhesive conditions (Tables S1 and S2).
These data demonstrate that HUVECs grown on matrigel regulate part of the molecular machinery involved in physiological and pathological angiogenesis in vivo, and that the morphogenesis time course analyzed here reflects late aspects of the angiogenic process. We thus show the validity of combining quantitative proteomics and an in vitro model of angiogenesis to investigate vessel growth.

Diverse proteome regulation in tubule-forming and spreading HUVECs
Statistical test using ANOVA revealed that 1,401 of the 3,723 accurately quantified proteins were regulated during morphogenesis and spreading (Table S2) with excellent reproducibility (Spearman rank correlation=0.90, Table S3). We then calculated changes in protein abundance over the time as the median SILAC ratio between replicates in the specific adhesion condition divided by the median SILAC ratio measured at the start of the experiment (0h), subsequently referred to as "ratio" (Sample/0h). To subdivide the regulated proteins in up and downregulated and assign them to the different adhesion conditions, we required a minimum ratio higher than one standard deviation (SD) from the mean of the all calculated ratios (mean 1,401prot =0.01; SD=0.534) (Table S4).
Our analysis revealed that an increasing number of proteins were regulated as tubulogenesis progressed, which was greater than that found during spreading (Figure 2A). Furthermore, the Venn diagram analysis of the regulated proteins during morphogenesis showed an almost complete overlap between subsequent time points (Figure 2B), suggesting a progressive differentiation of the cells, as also indicated by increasing levels of several EC markers ( Figure 1B,C). Conversely, clear differences were identified between cell morphogenesis and spreading. As representative examples, less than 20% of proteins were commonly regulated between Matr 24h and Matr dil, or Matr 24h and LAM ( Figure 2C). This distinct proteome remodeling was confirmed by an unbiased principle component analysis (PCA) of the ratio profiles of the differently cultured HUVECs, which clearly separated them according to their macroscopic appearance ( Figure 2D). Accordingly, a close similarity was detected between cell morphogenesis on normal and GFR matrigel, as well as between cell spreading on different ECMs. Of note, the component number two separated the earlier, Matr 12h, from the later time points, Matr 24h and 30h. This suggests that our time course experiment may contain distinct proteomic signatures for early and late morphogenesis.
These results provide clear evidence that major proteomic differences accompany EC morphogenesis and spreading. Using this dataset, specific proteins and cellular processes needed by the cells to regulate EC morphogenesis can be pinpointed.

EC morphogenesis and spreading involve different remodeling of the cell-ECM adhesion machinery
To gain bioinformatic insights into the processes involved in EC morphogenesis, we applied a 'onedimensional annotation distribution' analysis, which has been recently developed in our group [28], and identified the regulated KEGG pathways during EC morphogenesis and spreading. Figure 3A (see Table   S5 for details) shows their levels of regulation in the different experimental conditions.
Proliferation-related categories, DNA replication and cell-cycle, were downregulated in most of the experimental conditions. This may reflect a general decrease in cell proliferation due to the reduced amount of serum in the medium used during the experiment, which was half of the amount used during the culturing of the cells (see Experimental Procedures). Conversely, most of the KEGG categories had distinct regulation during EC morphogenesis and spreading, and we distinguished four major groups: categories regulated at early, medium and late time during EC morphogenesis, and categories regulated only during EC spreading ( Figure 3A and Table S5 for detailed list of the proteins).
After 12h on matrigel, HUVECs upregulated cell-ECM adhesion proteins involved in ECM-receptor interaction (p=1.5E-5), which include the ECM components and angiogenesis regulators FN1, HSPG2 and laminin 411 [38,40], and several integrin receptors. This regulation was maintained over the time.
Cell-ECM adhesion-related categories were found upregulated during EC spreading as well, though different ones compared to EC morphogenesis. Among them the category focal adhesion (best p=3.1E-4), which includes the protein kinases PAK2 and ERK1/2, and the myosin regulators MYL9 and MYL12A (Table S5). We investigated adhesion mechanism in more detail, and explored the regulation of proteins that have been manually annotated as component of the integrin adhesome [41]. This dataset includes two hundreds proteins involved in cell-ECM interaction mediated by integrin receptors [42]. Of those, 91 were quantified in our SILAC-based analysis, of which 33 were significantly regulated during EC morphogenesis and spreading (according to ANOVA test, see above). Heat map and hierarchical clustering based on the protein abundance profile of these 33 proteins revealed two main clusters with proteins distinctly regulated in EC morphogenesis and spreading ( Figure 3B). Fisher test-based category enrichment analysis revealed that the major difference between the proteins contained in the two clusters was their subcellular localization: membrane (p=1.7E-3; Benjamin Hochberg FDR=0.08) for the cluster of proteins with increased levels in morphogenesis only, while intracellular (p=2.3E-4, Benjamin Hochberg FDR=0.02) for the cluster of proteins with increased levels during EC spreading and decreased in morphogenesis ( Figure 3B).
This proteomic analysis provides a detailed portrait of the processes regulated in HUVECs during morphogenesis and spreading, and revealed the specific categories regulated depending on the adhesion substrate, and in the following we explore these aspects in more depth.

CLEC14A is an extracellular protein component of the HUVEC matrixome
Endothelial cell-ECM adhesion is a crucial mechanism in angiogenesis, because it regulates multiple steps of blood vessel growth, including vessel maturation. Moreover, a correct composition and structure of the ECM surrounding blood vessel is crucial to maintain its structure and functionality. Our analysis pointed out that ECM-receptor interaction proteins were significantly regulated during ECs morphogenesis, and we observed that amongst the most upregulated proteins (the 10% most upregulated proteins in Matr 30h are reported in Table S4) there were ECM components, such as LAMC1, NID1 and LAMB1, which regulate angiogenesis in vivo [43][44][45]. We therefore investigated ECM proteins further.
We set up a protocol to purify ECM produced by cells in culture (see Experimental Procedures) and performed an extensive MS profile of the ECM isolated from HUVECs. We identified 1,358 proteins, and estimation of the protein absolute abundance revealed that a subset of only 128 proteins made up 90% of the endothelial ECM protein mass, which we term "matrixome" ( Figure 4A, Table S6). The matrixome included known ECM components, such as FN1 and HSPG2, which were estimated to constitute more than 40% of the total matrixome mass. In contrast membrane proteins, such as the ECM receptors integrin α5 and β1, had low abundance ( Figure 4A). In order to identify ECM proteins relevant in angiogenesis in vivo, we searched for matrixome components highly abundant in HUVECs after 24h morphogenesis on matrigel and whose levels were higher in morphogenesis (Matr 24h) compared to spreading (adhesion on Matr dil or LAM for 24h). This pinpointed known secreted proteins [46] [47], such as the ECM component COL12A1, the growth factor TGFB1, the prelamin-A/C (LMNA) and a histone protein (HIST1H4A), but also proteins which have not been previously localized at the ECM ( Figure 4B, Table   S7). Among them we found the c-type lectin domain family XIV member A (CLEC14A), a single pass transmembrane glycoprotein member of the endosialin family ( Figure S1A). Another endosialin family member, the complement component C1q receptor (CD93), which has been previously shown to undergo shedding in the presence of inflammatory signals [48], was identified as a matrixome component with upregulation during morphogenesis ( Figure 4B) To investigate the relationship between CLEC14A and the ECM further we performed a quantitative MS proteomic interaction screen of endogenous CLEC14A immunoprecipitated from HUVECs seeded on matrigel. This identified 53 interaction candidates (Table S8). Of those, LAMA4, LAMC1, VIM, HSPG2, FN1, DBN1 and MMRN2 were upregulated during EC tubulogenesis on matrigel, similar to CLEC14A ( Figure 4D), and, except for DBN1 and LAMC1, identified as matrixome components. Strikingly, immunofluorescent confocal microscopy showed CLEC14A partial colocalizing with FN1, MMRN2 and LAMA4 ( Figure 4E and Figure S1B). MMRN2, an endothelial-specific matricellular glycoprotein [49], was the most enriched matrixome protein in the CLEC14A pull-downs. By immunoprecipitation of endogenous CLEC14A, we verified CLEC14A-MMRN2 interaction in HUVECs in culture ( Figure S1C).
Furthermore, as CLEC14A, MMRN2 was regulated at higher levels during morphogenesis compared to spreading. Altogether, these results prompted us to investigate CLEC14A and MMRN2 in vivo (see below).
Lastly, posttranslational modification analysis of immunoprecipitated CLEC14A uncovered a phosphorylation site at Ser 483 ( Figure 4F), in proximity of the PDZ domain-binding region at the Cterminus of the cytoplasmic tail. A functional role for the C-terminus PDZ binding domain was previously shown for CD93 [50], thus suggesting that Ser 483 may be of interest for future functional investigations.

Clec14a and Mmrn2 protein levels in blood vessels increase during tumorigenesis in multi-step mouse models
We assessed CLEC14A and MMRN2 function in tubule morphogenesis via small interfering RNA (siRNA) knock-down. CLEC14A or MMRN2 protein levels were silenced by siRNA, which compromised the formation of tubule structures when plated on matrigel, compared with control siRNAtransfected HUVECs ( Figure 5A and Figure S2A,B). We verified CLEC14A and MMRN2 function with an independent 3D tubulogenesis assay ( Figure 5B and Figure S2C,D).
Immunohistochemistry on a panel of human tumor tissues and adjacent normal tissue showed similar staining patterns between CLEC14A and the EC marker PECAM1 in arteries and veins ( Figure 5C and Figure S2E). Because tumor vessels strongly stained for CLEC14A, we further investigated whether CLEC14A levels were modulated during tumor progression. We employed two mouse models of spontaneous tumorigenesis that recapitulate the progressive development of human cancer. RIP-Tag2 mice develop pancreatic neuroendocrine tumors, which progress from hyperplastic islets to angiogenic and then tumor islets. HPV16/E 2 mice display cervical cancer through progression from low-grade cervical intraepithelial neoplasia, to high-grade dysplasia (CIN-3), and invasive squamous cell carcinoma (SCC). Fluorescence confocal microscopy showed partial colocalization between Clec14a and the EC marker Meca32 in blood vessels of normal, angiogenic and tumor pancreatic islets in RIP-Tag2 mice ( Figure 5D). Accurate fluorescence quantification revealed significantly higher Clec14a levels in blood vessels of tumor compared to normal and angiogenic islets ( Figure 5E). Similarly, higher levels of Clec14a were measured in blood vessels of CIN-3 and SCC of HPV16/E 2 mice, compared to nontransgenic E 2 -treated normal cervix ( Figure 5F). Similar to Clec14a, significantly higher levels of Mmrn2 were measured in tumor compared to normal vessels in RIP-Tag2 and HPV16/E 2 mice. However, in blood vessels of RIP-Tag2 tumor angiogenic islets, only Mmrn2 was found elevated and not Clec14a ( Figure 5E,F).
Finally, confocal analysis of tumor vessels in RIP-Tag2 and HPV16/E 2 mice showed partial colocalization between Clec14a, Mmrn2 and Lama4 ( Figure S3A,B), and 3D reconstruction of confocal images of the Rip-Tag2 tumor vasculature placed Clec14a at the interface between ECs and pericytes ( Figure 5G,H), thus supporting the presence of Clec14a in the ECM also in vivo.
Together, these results show that Clec14a and Mmrn2 are required in EC morphogenesis and that they are deregulated at specific stages of cancer progression in mouse models.

Discussion
Our mass spectrometric analysis of HUVECs during tubules morphogenesis in matrigel shows that this process involves significant changes of about 20% of the proteome, which we measured to a depth of more than 3,700 proteins, the most in depth endothelial proteome generated so far. The accuracy of the quantitative approach allowed us to shed new light on EC regulation. This is illustrated by a very reproducible regulation in abundance of a remarkable number of cell adhesion and metabolism-related proteins and enzymes. Furthermore, we provide the first in depth characterization of the HUVEC matrixome, which we proved to be an excellent resource of proteins deregulated in tumor angiogenesis.
Cell-matrix adhesion regulates EC sprouting and migration during angiogenesis, and provides support and stability to the newly formed capillary. Importantly, adhesion mechanisms can be influenced by the environment surrounding the cell, and here we show that ECs differently modulate their adhesion machinery during morphogenesis and spreading ( Figure 3A). An important difference between the matrigel and spreading systems used for this study is the mechanical property of the matrix, where matrigel is soft and other ECMs are rigid. Accordingly, among the regulated cell adhesion proteins the levels of the cell contractility regulators MYL12A and MYL9 decreased in morphogenesis, while they were augmented during spreading. Because increasing ECM rigidity has been shown to support tumor progression [51], we speculate that the differentially regulated adhesion proteins (Table S5) may be relevant to understanding the role of ECM stiffness on EC behavior in tumors.
Tumor vessels often display absent or abnormally thick ECM, which impact their function [3]. Our MS analyses highlighted significantly increased levels of ECM proteins during HUVECs morphogenesis (Table S7), and we discovered that CLEC14A is an ECM component. We showed that CLEC14A binds to and co-localizes, in vitro and in vivo, with FN1 and LAMA4, which are master regulators of angiogenesis and previously found deregulated in tumor vessels [2,52] and with MMRN2 ( Figure 4). In contrast to our results, MMRN2 overexpression has been reported to inhibit angiogenesis by affecting VEGF signaling [53]. This suggests that MMRN2 effects on endothelial cells might be context and dosedependent and that further studies are needed to elucidate its function in angiogenesis. Furthermore, we show that CLEC14A and MMRN2 are EC marker, functional in HUVECs, and that CLEC14A is highly expressed in tumor compared to normal vessels. This confirmed previous literature [54,55]. Additionally, through multiple-stage mouse tumor models, Rip-Tag2 and HPV16/E 2 , we identified Clec14a and Mmrn2 deregulation at specific stages of tumorigenesis ( Figures 5,S2,S3). These mice will therefore be valuable models to further investigate Clec14a and Mmrn2 functions in tumor angiogenesis. Similarly to CLEC14A, other matrixome components contained in our screen (Table S7) and that were not previously known to be localized at the ECM may be interesting candidates for further investigation of their role in normal and pathological angiogenesis.
Our results show that applying modern proteomic technologies to established in vitro model of angiogenesis is a powerful approach to discover functional regulations in endothelial cells, relevant for the angiogenic process in vivo. Here we have functionally explored ECM proteins, but other processes were also found regulated. For instance, significantly altered abundance was measured for a remarkable number of metabolic proteins, a phenomenon that to our knowledge has not been observed in morphogenesis before ( Figures 3A, Table S5). In conclusion, we have demonstrated that our proteomic study of ECs morphogenesis and spreading provides valuable information that can help to unravel molecular mechanisms regulating endothelial cells behavior. Our data may contain the basis for further exciting insights to be investigated in angiogenesis in vivo, especially with a view to interfering with this process in pathologies.   Table S4.  Table S2.     Table S8).  (Table   S8), most of which are matrixome proteins (Table S7). The x axis represents the ratio Matr 30h/0h (Table   S2), the y axis the average ratio (CLEC14A/IgG) in two independent CLEC14A immunoprecipatates against IgG control (Table S8).