The in vivo functions of lymphatic endothelial cells depend on their microenvironment, which cannot be fully reproduced in vitro. Because of technical limitations, gene expression in uncultured, “ex vivo” lymphatic endothelial cells has not been characterized at the molecular level. We combined tissue micropreparation and direct cell isolation with DNA chip experiments to identify 159 genes differentiating human lymphatic endothelial cells from blood vascular endothelial cells ex vivo. The same analysis performed with cultured primary cells revealed that only 19 genes characteristic for lymphatic endothelium ex vivo retained this property upon culture, while 27 marker genes were newly induced. In addition, a set of panendothelial genes could be recognized. The propagation of lymphatic endothelial cells in culture stimulated transcription of genes associated with cell turnover, basic metabolism, and the cytoskeleton. On the other hand, there was downregulation of genes encoding extracellular matrix components, signaling via transmembrane tyrosine kinase pathways and the chemokine (C-C) ligand 21. Direct ex vivo analysis of the lymphatic endothelial cell transcriptome is helpful for the understanding of the physiology of the lymphatic vascular system and of the pathogenesis of its diseases.
- lymphatic capillary
- in vivo
lymphatic and blood capillaries are located in close proximity in the human dermis and are lined by lymphatic and blood vascular endothelial cells (LECs and BECs), respectively. The thin, discontinuous monolayer of LECs rests on an incomplete basement membrane made up of extracellular matrix proteins. The lymphatic capillaries contribute to regulation of tissue fluid pressure and body temperature, transport cells and molecules, and modulate functions of the immune system (8). The integration of lymphatic capillaries into a proper microenvironment is essential for maintaining these capacities (6, 24, 27, 30). Many pathological conditions are associated with malfunctioning of LECs, including lymphedema, inflammation, infectious and immune diseases, fibrosis, as well as solid tumors (10, 15, 22, 23, 26).
Most studies on lymphatic vessels use either fixed or frozen tissue samples, in vitro culture, or animal models, while little knowledge has been acquired from native human tissue (9, 11, 20, 21, 29). It has been difficult to generate pure preparations of quiescent LECs from tissues in quantities large enough for molecular screens without expansion in cell culture (3, 5). Studies of rat endothelial cells fuel expectations that the differences between in or ex vivo and in vitro samples are significant, but experimental evidence on human LECs is still limited (2).
Here, we report our first attempts to describe the LEC transcriptome as closely as possible to the human in vivo state compared with in vitro LEC cultures.
MATERIALS AND METHODS
Endothelial cell micropreparation and cell sorting.
In vitro LECs and BECs were separated from cultured human dermal microvascular endothelial cells (HDMECs, PromoCell) by immunomagnetic purification using antipodoplanin antibodies (12). HDMECs represent microvascular endothelial cells isolated from primary skin cultures by immunomagnetic purification using anti-CD31 antibodies. HDMECs, as well as pure LECs and BECs, were cultured in endothelial cell growth medium MV supplemented with 5% fetal calf serum; endothelial cell growth supplement with heparin, an extract of mixed-sex bovine hypothalamic tissue; as well as EGF; bFGF; and hydrocortisone (all from PromoCell). Purified LECs and BECs were cultured on fibronectin-coated cell culture plates before lysis. In addition, LECs were cultured in the presence of recombinant VEGF-C, while BECs were cultured without added VEGF-C. BECs secrete VEGF-C to the culture medium, and the secreted VEGF-C is sufficient to stimulate the growth of LECs in the HDMEC coculture. Ex vivo BECs and LECs were prepared according to a mechanical and enzymatic micropreparation protocol (11, 28), approved by the local ethics committee (no. 449/2001). LECs and BECs were further separated by fluorescence activated cell sorting (FACStar Plus, Becton Dickinson, NJ) using CD31, CD45, and podoplanin antibodies in a three-step procedure on ice with intervening washing steps, and lysed directly.
We fixed 5-μm cryosections obtained in parallel to skin resections according to standard protocols (1) in acetone and used them for immunofluorescence. Cell cultures were fixed in 4% paraformaldehyde and subjected to immunofluorescence. Nuclei were counterstained with 4′,6′-diamidino-2-phenylindole hydrochloride (SERVA). Images were captured with an Axiophot epifluorescence system (Zeiss) and processed with Adobe Photoshop 7.0 (Adobe Systems). We used 1- to 2-μm paraffin sections of healthy human skin samples in immunohistochemistry with specific primary antibodies and horseradish peroxidase-coupled secondary antibodies (Axell) using diaminobenzidine (Pierce, IL; 34065) as a chromogen.
Equal numbers of ex vivo LECs and BECs were lysed in sample buffer (2% SDS, 60 mM Tris·HCl pH 6.8, 0.02% DTT). The lysates were separated in 10% SDS-PAGE and transferred to a nitrocellulose membrane (Schleicher and Schüll, Protran BA83). Proteins were detected using anti-CD31 and antipodoplanin antibodies, horseradish peroxidase-conjugated secondary antibodies, followed by ECL (Amersham, RPN2106), and autoradiography (Kodak, no. 5087838).
The following antibodies were used: mouse monoclonal anti-CD31 antibody (DAKO Cytomation, no. M0823; immunofluorescence, Western blot), FITC-conjugated anti-CD31 (Becton Dickinson Pharmingen; no. 555445; FACS), RPE-Cy5.1-conjugated mouse monoclonal anti-CD45 (Beckman Coulter, no. PM IM2653), an IgG fraction of a polyclonal antipodoplanin antiserum, donkey anti-rabbit IgG RPE (Jackson ImmunoResearch, no. 711-116-152), monoclonal anti-PAL-E (Harlan SERA-LAB, MAB3356f), monoclonal antimannose receptor (BD Pharmingen, no. 555953), monoclonal antireelin (Chemicon, no. MAb5366), and monoclonal anti-CCL21 antibody (R&D Systems, no. AF366), secondary Alexa Fluor 594 and Alexa Fluor 488 conjugated goat anti-mouse and goat anti-rabbit antibodies, respectively, (Molecular Probes, nos. A-11020 and A-11034), horseradish peroxidase-conjugated rabbit anti-mouse and goat anti-rabbit antibodies (Axell, nos. JZM035046 and SGZ034047), respectively, and rabbit dysferlin antiserum (generously provided by Jan Bauer, Institute for Brain Research, Medical University of Vienna).
Isolation of total RNA, generation of cRNA, and DNA chip hybridization were performed as described (28). DNA chip quality parameters largely met the requirements (25). The raw values of a subset of endothelial housekeeping genes were used to exclude experiments with inefficient, nonuniform hybridization (28), and only DNA chip experiments with sufficient quality were used (Table 1 and Supplemental Table 1; the online version of this article contains supplemental data). Ex vivo samples were linearly amplified in a two-round and in vitro samples in a one-round protocol. Details including MIAME criteria and quality controls are described in the supplemental data. Data were submitted into ArrayExpress with accession number E-MEXP-455. Ex vivo cRNA was used for nonquantitative RT-PCR to amplify endothelial cell markers (von Willebrand factor: cgc tcc ttc tcg att att gg, ccg gac agc ttg tag tac cc), LEC markers (podoplanin: caa cgg gaa cga tgt gga ag, cgt tgg cag cag ggc gta ac; LYVE-1: gcc agg tgc ttc agc ctg gtg, ctt cag ctt cca ggc atc gca cgg; prox-1: aca agc cga agc gag aag g, aac aag ggt ggt ggc tca g), nonendothelial transcripts (CD45: tca acc aca aca ata gct act, gtc tcc att gtg aaa ata ggc; smooth muscle cell actin: gtg tgt gac aat ggc tct gg, tga tga tgc cat gtt cta tcg, keratin: aga cca aag gtc gct act gc, aga act ggg agg agg aga gg), and markers of RNA integrity (intronic sequence of neurofibromatosis 2: ggt gtc ttt tcc tgc tac ct, ggg agg aaa gag aac atc ac; housekeeping genes: breakpoint cluster region: gag aag agg gcg aac aag, ctg tgc tta aat cca gtg gc, and beta 2-microglobulin: att tcc tga att gct atg tg, gaa ttc act caa tcc aaa tg). The quality of in vitro derived RNA was analyzed by gel electrophoresis. For TaqMan analysis (Applied Biosystems) assay numbers Hs00357525_m1, Hs00170014_m1, Hs00826129_m1, Hs00415006_m1, and Hs00249890_m1 were used and showed the same amplification efficiencies (data not shown).
To account for low samples sizes we applied two bioinformatical strategies: A: after normalization, the data were imported into GeneSpring 5.1 (Silicon Genetics, CA, USA), and two main methods were used for comparing two groups of chip measurements to find genes preferentially expressed in one of the groups: 1) If two groups consisted of measurements, which could be related to each other pairwise, as in the case of LEC and BEC measurements from the same subject, a paired t-test could be performed (ex vivo LECs and BECs). If this condition was not fulfilled (in vitro BEC vs. in vitro LEC, and ex vivo vs. in vitro comparisons), we used a two-sample t-test. For both, the confidence levels were P < 0.01. 2) For selected questions, t-tests were complemented by three types of similarity measures: standard correlation, Pearson correlation, and (Euclidian) distance. For similarity measurements, we included 40 genes, which showed best behavior as matches (thresholds were at least 0.89 for Pearson correlation, 0.895 for standard correlation, and maximally 5.925 for distance). Only genes with postnormalization values above the BioB control in all samples of a group were included in calculations. Threshold of expression ratio was set at twofold. B: in addition, four ex vivo LEC and BEC as well as four in vitro LEC and BEC samples were normalized in GeneSpring 7.1 using the robust multiarray expression measure using sequence information function (GC-RMA) from the Bioconductor package. The normalized data were filtered for flags according to the criterion that a gene had to be present in three out of four LEC/BEC samples in either of the in vitro or ex vivo data set; 10,742 genes from the in vitro data set and 10,284 genes from the ex vivo data set passed this criterion. In total, 12,251 genes were included in the analysis. The filtered data were analyzed in a two-factor ANOVA test (parametric, assuming variances equal) using as a first parameter the cell type (LEC vs. BEC) and as a second parameter the tissue type (in vitro vs. ex vivo cells). The statistical significance (P < 0.05) was determined using the Benjamini-Hochberg false discovery rate for multiple testing correction.
GeneOntology (GO; http://www.geneontology.org/) was used as a datasource for the “pathway” information describing biological associations for a set of genes. Affymetrix Probe Set IDs were first translated to Ensembl Gene Stable IDs (Ensembl release 29). GO annotations were retrieved from Ensembl. Due to the hierarchical character of the GO annotations, all the pathways from the annotated most specific terms were iterated to the GO tree root term along all annotated paths.
The GC-RMA normalized DNA chip expression data and medium values of sample groups were used to calculate the fold changes, which where used to create the ordered gene list. Once the genes were associated with pathway information, an iterative hypergeometric distribution-based statistical analysis was performed. Here, using the ordered gene lists, we calculated the hypergeometric probabilities iteratively for each occasion of a gene belonging to the given pathway. The iterative process minimized the P value, i.e., with all occurrences of a gene belonging to a given pathway. The P value for this pathway was used, if it was the lowest P value for the pathway. This approach had the benefit that the gene lists did not need to be divided into “regulated” and “nonregulated” genes by any arbitrary limit, such as fold change. Since the obtained P values from the pathway analysis were not corrected for multiple testing problem, a conservative P value limit was used. All the above mentioned annotation and data analysis steps were done by tailor-made software (14). Two different comparisons were generated: in vitro LECs vs. BECs and ex vivo LECs vs. BECs to detect “overexpressed” and “underexpressed” pathways.
Raw and calculated data including Supplemental Figs. 1–5 and Supplemental Tables 1–12 can be accessed at http://www.meduniwien.ac.at/complex-systems/supplementary2005/.
Sample preparation and quality control.
Both BECs and LECs express the glycoprotein CD31 at their plasma membranes, whereas only LECs are positive for the highly glycosylated sialoprotein podoplanin (Refs. 1, 19; Fig. 1, A–D). Using mechanical and enzymatic treatment to release single cells from human skin resections into suspension and CD31 and podoplanin antibodies, we purified BECs and LECs by fluorescence activated cell sorting (Ref. 11; Fig. 1, E–J). Both cell types were negative for CD45 glycoprotein, indicating exclusion of leukocytes. We lysed the purified cells without an intervening culture step to minimize any artifacts at this ex vivo stage, and isolated total RNA. The purity and integrity of the isolated RNA were analyzed by RT-PCR and gel electrophoresis (Fig. 1, K and O). We detected transcripts for von Willebrand factor in both LECs and BECs, while a panel of LEC-specific transcripts, including podoplanin, LYVE-1, and prox-1, was not present in BECs. Contamination of endothelial cells with keratinocytes, smooth muscle cells, or leukocytes could be excluded by unsuccessful amplification of keratin, smooth muscle cell actin, and CD45 transcripts, respectively. Genomic DNA contribution was negative, while two housekeeping genes were expressed in both endothelial subpopulations. At protein level, CD31 was detected in both LECs and BECs, while podoplanin was restricted to LECs (Fig. 1L). To learn about the cell culture effect on LECs, we isolated LECs and BECs also from cultured HDMECs by using magnetic beads coated with antipodoplanin antibodies. Staining for podoplanin confirmed that the LECs and BECs were >97% pure cell populations (Fig. 1, M and N), in line with the in vivo and ex vivo situations.
For comparison of genes expressed by ex vivo and in vitro LECs and BECs, we used the annotated portion of human expressed sequences on the U133A GeneChip from Affymetrix (Table 1). We subjected equal amounts of total RNA samples to linear amplification and generated labeled cRNA. Ex vivo endothelial cells from healthy donors were small and quiescent and contained only minute amounts of total RNA (data not shown). This was consistent with previous studies on endothelial cell turnover in vivo (3, 5). Thus, a second round of linear amplification was necessary to generate sufficient amounts of cRNA for DNA chip analysis (>5 μg). In contrast, total RNA from in vitro preparations was sufficient for a one-round amplification protocol. To exclude a systematic error resulting from the comparison of samples that had undergone amplification to different extents, we performed DNA chip experiments of cultured LECs and BECs subjected to one- or two-round amplifications. Candidate genes (see below) within these plots are highlighted in blue in Supplemental Fig. 1 (see http://www.meduniwien.ac.at/complex-systems/supplementary2005/). The experiments showed a faithful distinction of candidate genes with potential functional roles from artifact genes, i.e., those that were falsely misregulated due to preferential amplification of certain transcripts. Specifically, candidate genes were in the high intensity signal range, while artifact genes had low fluorescence signals. Importantly, we excluded the latter by using the lowest external control (BioB) value as a discrimination threshold. Moreover, we checked overall hybridization success of all samples (28). These steps ensured that contamination of ex vivo vs. in vitro LEC and BEC candidate genes was negligible. We then characterized transcriptomes of both LECs and BECs derived from the ex vivo and in vitro preparations (Fig. 2, see supplemental data for raw and calculated data files).
Global endothelial transcriptomes.
To determine which genes were regulated globally in LECs and BECs upon culture, we compared all ex vivo experiments to all in vitro experiments. Out of 22,229 human transcripts represented on U133A, 470 were altered upon culture in LECs and 888 in BECs. This corresponded to 2.1% and 4.0%, respectively. Among these, 140 genes (0.6%) were regulated in both LECs and BECs to the same extent (Table 2, Supplemental Fig. 2). Eighty-eight transcripts were induced in both LECs and BECs upon culture, whereas 52 mRNAs were repressed in both cell types. To facilitate data handling, thresholds were manually set at 6.5-fold for expression ratio and 65% for variation coefficient. Induced genes included long-term cell cycle regulators (cyclins B and G, mitogen-activated protein kinase), components involved in RNA-polymerase II-mediated transcription (polypeptide E), players of glycolysis (pyruvate kinase, glyceraldehydes-3-phosphate dehydrogenase), or gene products of major cytoskeletal systems (actin, spectrin, microtubule-associated protein 4). Around 10% of transcripts were constantly expressed in both ex vivo and in vitro settings.
LEC characteristic transcriptomes.
Next, we wanted to analyze whether candidate gene profiles differentiating LECs from BECs were altered upon in vitro conditions. Therefore, ex vivo and in vitro transcriptomes of LECs and BECs were first analyzed separately (Supplemental Fig. 3). A t-test using the mean fluorescence values of ex vivo LECs and BECs identified LEC-specific genes and BEC-specific genes. In addition, genes expressed in both endothelial cell populations, but more pronounced in either LECs or BECs, were identified.
To compare expression levels of all genes to ideal LEC- or BEC-specific candidates, which were used as virtual inputs, we performed similarity measurements. The 40 most similar genes were considered real candidates on basis of visual inspection of the resulting profiles. Finally, gene lists based on the t-test and similarity measurements were fused and ex vivo as well as in vitro LEC and BEC candidate gene sets were obtained.
To investigate whether cell culture per se resulted in shifts of the LEC candidate gene sets, we identified in vitro LEC candidates in graphs, where ex vivo LEC candidates were plotted (Supplemental Fig. 4). The overlap between the ex vivo and in vitro sets was small (37 genes). A subset of 334 genes was downregulated in vitro, while 255 genes were induced. Altogether 218 genes out of the 235 LEC characteristic candidates (1.1%) were regulated in culture, while 19 remained unchanged (Table 3, Supplemental Table 5). Evaluation of BECs showed similar results (Supplemental Fig. 5). Out of 199 (0.9%) mRNAs characteristic for ex vivo BECs, 181 were changed, while 18 remained stable in culture (Supplemental Table 6, threshold for the expression ratio 2.0). The functional categories in the LEC transcriptome included genes involved in cell growth and adhesion, cytokines and chemokines, ion channels, GTPases and vesicle proteins, cell metabolism, including lipid- or carbohydrate metabolism, as well as a panel of nonclassifiable or unknown genes (Table 3). Known LEC marker gene products such as podoplanin, lyve-1, vegfr3/flt4, and mannose receptor C were classified as both ex vivo and in vitro LEC candidate genes. In addition, we specifically observed in vitro downregulation of chemokine (C-C) ligand 21 and induction of matrix Gla protein that both have been considered characteristic of LECs. Eleven selected genes were reconfirmed by immunohistological analysis or semiquantitative RT-PCR (Fig. 3). Supplemental Table 5). Evaluation of BECs showed similar results (Supplemental Fig. 5). Out of 199 (0.9%) mRNAs characteristic for ex vivo BECs, 181 were changed, while 18 remained stable in culture (Supplemental Table 6, threshold for the expression ratio 2.0). The functional categories in the LEC transcriptome included genes involved in cell growth and adhesion, cytokines and chemokines, ion channels, GTPases and vesicle proteins, cell metabolism, including lipid- or carbohydrate metabolism, as well as a panel of nonclassifiable or unknown genes (Table 3). Known LEC marker gene products such as podoplanin, lyve-1, vegfr3/flt4, and mannose receptor C were classified as both ex vivo and in vitro LEC candidate genes. In addition, we specifically observed in vitro downregulation of chemokine (C-C) ligand 21 and induction of matrix Gla protein that both have been considered characteristic of LECs. Eleven selected genes were reconfirmed by immunohistological analysis or semiquantitative RT-PCR (Fig. 3). For example, in blood capillaries, which were identified by the expression of the marker molecule PAL-E, pericytes were reactive with an antibody to mannose receptor, while BECs were negative. On contrary, there was distinct reactivity of endothelial cells of lymphatic capillaries, which by definition lack pericytic coverage. Similarly, dysferlin was predominantly observed on lymphatic vessels, while blood capillaries, characterized by pericytes, showed much diminished endothelial positivity. In addition, CD36 and complement factor H could be confirmed as additional LEC characteristic proteins on human tissue sections (data not shown). Reelin as well as mannose receptor proteins were detected in cultured, podoplanin-positive LECs. Moreover, loss of CCL 21 protein expression could be confirmed by comparing human paraffin sections of human dermis with LEC and BEC cultures.
To further validate the results, we subjected the data sets to a second, independent approach of data analysis outlined in Fig. 2. The ex vivo and in vitro datasets were normalized by GC-RMA included in the Bioconductor package of the GeneSpring software. These data sets were then analyzed in a two-factor ANOVA test (Fig. 4). The expression of 941 genes was found to be significantly different between LECs and BECs (cell type effect, Fig. 5), while 6,613 genes were differentially expressed between ex vivo and in vitro conditions (cell culture effect). Of the genes present in the latter group, 4,410 changed more than twofold between the ex vivo and in vitro conditions (Supplemental Table 9). Specifically, 1,973 genes were expressed at a higher level ex vivo than in vitro, while 2,437 genes showed a higher expression in vitro. Thus, consistent with the previous mode of data analysis, the number of genes affected by culture conditions exceeded the number of genes differentially expressed between the LECs and BECs. In addition, as found in the previous analysis, genes such as c-fos or egr1 were downregulated in both cell types upon culture, while for example the cyclins were upregulated, consistent with our previous data analysis (Fig. 4, Table 2). Of note, 641 genes out of 941 showing differential expression between the LECs and BECs also differed in expression between the ex vivo and in vitro conditions (Supplemental Table 11), leaving 300 LEC- or BEC-specific genes that did not change significantly upon cell culture (Supplemental Table 10). Of these, 82 genes were more highly expressed in BECs, and 111 in LECs. This latter group contained the LEC marker genes for macrophage mannose receptor 1, prox1, lyve-1, podoplanin, and vegfr3/flt4. In total, 193 genes passed the criteria of P < 0.05 and fold change >2, indicating that these genes differentiated LECs from BECs both ex vivo and in vitro.
LEC- and BEC-specific pathways.
To analyze the biological pathways and processes differentiating LECs from BECs, the GC-RMA normalized data sets were subjected to pathway analysis using the GO database. Pairwise comparisons were made between LECs and BECs ex vivo as well as in vitro. Many of the pathways upregulated in ex vivo BECs were also upregulated in in vitro BECs, compared with LECs ex vivo and in vitro, respectively (Table 4). These included the chemokine and inflammatory responses, while, e.g., the pathway for exogenous antigen presentation was lost upon transition of BECs into cell culture. On the other side, several pathways, e.g., growth factor activity, were active in ex vivo LECs, compared with ex vivo BECs, while no pathways were upregulated in in vitro LECs, compared with in vitro BECs, using a very conservative P value. Thus, although in vitro cultivation affected gene expression of both cell lineages, the LECs seemed to be more sensitive to the loss of their microenvironment.
We have here made the first comprehensive gene expression analysis of lymphatic and blood vascular endothelial cells directly isolated from tissues without expansion in cell culture. This large-scale transcriptomal analysis of ex vivo and in vitro LECs and BECs indicated that cell culture introduced a substantial change in gene expression in these cells. The ideal scenario of comparing ex vivo and in vitro preparations deriving from a longitudinal experimental setup was impossible due to restrictions of skin sample size. However, characteristic marker genes and pathways for BECs and LECs could be identified based on careful scrutiny of the data using two methods of bioinformatic analysis. We identified at least 235 genes that were characteristic of LECs relative to BECs (Supplemental Table 5), while all previously published data concerns in vitro cultured LECs, which we here find unreliably representative to the in vivo gene expression (4, 16, 18).
Our protocol made ex vivo analysis of normal human endothelial cells possible. It could provide a lead for the use of limited clinical samples in transcriptomal screens, even if heterogeneity of preparations existed. In fact, the definition of the minimal sample size is controversial at the moment: On the one hand, for tumor studies the use of at least six samples per condition has been recommended (25). However, mostly these are mixed tissue samples, whereas we have worked on highly purified cell preparations with known absolute cell numbers. On the other hand, successful transcriptomal analyses on endothelial cells have been performed using four paired samples (16). We followed the latter protocols and compared LECs with BECs derived from the same individual. Reliable inference of specific gene sets also depended on consideration of several technical control steps as well as the combination of two independent methods of data analysis (Ref. 28; Supplemental Table 1, Supplemental Figs. 1–3]. Still, we believe that these steps have to be optimized for each cell type, mode of micropreparation, and technology platform. Furthermore, the U133A GeneChip has only a half-complete representation of the expressed portion of the human genome. However, the proportion of genes regulated in different endothelial cell subpopulations relative to all sequences (LECs 2.1%, BECs 4.0%) serves as an orientation for the extent of transcriptional alteration occurring upon cell culture. This seemed significant in light of the background of stably expressed genes (∼10%). It is possible that some of the observed differences between ex vivo and in vitro endothelial cell transcriptomes may be due to the different site of the skin tissue used to derive ex vivo and in vitro LECs and BECs. However, in a different study we compared gene expression of LECs derived from two different tissue types and found that a very small proportion (in the range of 1/1,000) of genes is differentially expressed between two different tissue LEC types (T. Petrova, unpublished results).
Only around 1% of transcripts could be regarded as LEC candidates ex vivo, i.e., differentiating LECs from BECs. The majority of these candidates was altered upon culture. Only a small gene set of 19 LEC markers was identified, which could be considered LEC characteristic both ex vivo and in vitro. This confirmed an expected influence of cell culture on the transcriptional profile of LECs (Table 2, Supplemental Fig. 4). It was interesting to note, however, that two-factor ANOVA proposed 300 genes that were able to differentiate between LECs and BECs both ex vivo and in vitro. This could be explained by the less stringent thresholds chosen for this procedure (P < 0.05, three out of four positive hybridization results, no threshold in signal intensity), compared with two-sample t-test (P < 0.01, 4/4, > BioB). Importantly, after application of the twofold cutoff for the identification of LEC characteristic genes, the ratio between globally in vitro-regulated genes and stable LEC marker genes was 34.3 (6,613/193) and 56.9 (1,992/35) for two-factor ANOVA and two-sample t-test, respectively. These relative numbers were thus comparable between the two methods. Of importance, known LEC candidate genes, including podoplanin, lyve-1, vegfr3 or macrophage mannose receptor, were identified as stable LEC markers by both types of data analysis. Moreover, as reported, matrix Gla gene was induced in LECs in vitro (16), while glycoprotein M6A was an LEC marker restricted to the ex vivo conditions.
Many LEC candidate genes transcriptionally affected by cell culture were associated with interaction of LECs with their microenvironment (Table 3). Specifically, pdgf-c and igf1 were downregulated. The chemokine (C-C) ligand 21, which has been shown to colocalize with podoplanin in vivo and bind to it in vitro, was no longer produced by LECs in vitro (Ref. 10 and Fig. 3). A panel of extracellular matrix components lost their specificity for LECs in vitro, including distinct laminin chains, fibronectin 1, and tenascin XB isoform. In addition, an amyloid beta-binding protein was dramatically affected, as was the cytoskeletal protein sarcoglycan-ε. Although any functional inference is purely speculative at this stage of analysis, it is conceivable that LECs abandon mechanisms that depend on their spatial integration into a lymphatic capillary, substituting them with different molecules. In this context the appearance of new LEC candidates in culture could be of interest, including the transcription factors ID1 and 3, the lamin B receptor, and the matrix components nidogen and matrix Gla protein. Finally, there was a panel of transcripts with only limited or unclear annotation information, which were forfeited upon culture with respect to BECs and which await more detailed studies. It should be noted that, although the differential amplification protocol of ex vivo and in vitro LEC samples was compensated by sequential normalization to the selected housekeeping gene set and to the corresponding BEC preparations, only relative conclusions on transcriptome profiles are legitimate. Conservative pathway analysis revealed functional categories of LEC- and BEC-specific transcripts and demonstrated in a striking way the overall difference between transcriptomes of LECs and BECs upon culture.
The panendothelial comparison of all prenormalized LEC and BEC ex vivo (eight) with in vitro (ten) DNA chip results revealed profound differences between the ex vivo and in vitro states of ECs in general (Table 2); 1,504 genes representing 6.8% of all transcripts of U133A were expressed in common between HDMECs and cultured skin microvascular endothelial cells. In contrast, there were many genes significantly regulated in both LECs and BECs to the same extent (Supplemental Table 2). For example, induction of genes encoding long-term cell cycle regulators (cyclins B and G, mitogen-activated protein kinase), components involved in RNA-polymerase II-mediated transcription (polypeptide E), or in glycolysis (pyruvate kinase, glyceraldehydes-3-phosphate dehydrogenase) and genes encoding major cytoskeletal proteins (actin, spectrin, microtubule-associated protein 4) could be noted. This is in line with the common knowledge that bringing quiescent cells into a two-dimensional cell culture environment comes along with induction of basic cell functions like cell cycling, increased metabolism, and gene expression. On the other hand, intercellular adhesion molecule 1 was downregulated in vitro. Apparently, for endothelial cells in culture, gene programs are dedicated to single cell fate rather than to interaction with the cell environment are engaged (see Ref. 13 for review).
LEC morphology, transcription, and function are influenced by interaction with the surrounding matrix as well as by intra- and extravascular cells (6, 24, 27, 30). Our transcriptome analysis suggests that the two-dimensional cell culture has only limited capacity to maintain these attributes. LEC cultures also have other limitations; for example, VEGF-C treatment was necessary to maintain LECs, while the ex vivo LECs did not need such treatments (12). It should be mentioned here that we have also confirmed selected genes of our study in immortalized LECs and BECs as well (Fig. 3), adding confidence in our data (7).
Obviously, endothelial cell culturing will have to be improved to imitate the native conditions more closely. Our transcriptomal as well as proteomic data reported recently could contribute to this intention (2).
An in-depth understanding of the transcriptome of LECs may assist in defining their pathophysiological roles on a molecular level. In this regard, a coupled proteomic analysis would be of great value. However, with the current micropreparation of human dermal endothelial cells sufficient amounts of protein are not obtained, except for limited single protein experiments (Fig. 1L). Nevertheless, our results on the ex vivo LEC transcriptome should serve as one strong building block in the identification of and future studies on the molecular mechanisms of lymphatic endothelial identity and integrity.
This work was supported by grants to N. Wick from the Austrian Science Foundation, to N. Wick and S. Thurner from the Vienna Science and Technology Fund, to K. Alitalo from National Heart, Lung, and Blood Institute 5R01HL-075183-03, and to D. Kerjaschko and K. Alitalo from the EU-FP6 framework program Lymphangiogenomics.
We thank Jan Bauer for support with immunohistochemistry. Josef Bruck was helpful with bioinformatics. Anton Jäger provided graphical assistance.
Address for reprint requests and other correspondence: D. Kerjaschki, Clinical Inst. for Pathology, AKH and Medical Univ. of Vienna, Waehringer Guertel 18-20, A-1090 Vienna, Austria (e-mail:).
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).
- Copyright © 2007 the American Physiological Society