Arhgef15 Promotes Retinal Angiogenesis by Mediating VEGF-Induced Cdc42 Activation and Potentiating RhoJ Inactivation in Endothelial Cells

Background Drugs inhibiting vascular endothelial growth factor (VEGF) signaling are globally administered to suppress deregulated angiogenesis in a variety of eye diseases. However, anti-VEGF therapy potentially affects the normal functions of retinal neurons and glias which constitutively express VEGF receptor 2. Thus, it is desirable to identify novel drug targets which are exclusively expressed in endothelial cells (ECs). Here we attempted to identify an EC-specific Rho guanine nucleotide exchange factor (GEF) and evaluate its role in retinal angiogenesis. Methodology/Principal Findings By exploiting fluorescence-activated cell sorting and microarray analyses in conjunction with in silico bioinformatics analyses, we comprehensively identified endothelial genes in angiogenic retinal vessels of postnatal mice. Of 9 RhoGEFs which were highly expressed in retinal ECs, we show that Arhgef15 acted as an EC-specific GEF to mediate VEGF-induced Cdc42 activation and potentiated RhoJ inactivation, thereby promoting actin polymerization and cell motility. Disruption of the Arhgef15 gene led to delayed extension of vascular networks and subsequent reduction of total vessel areas in postnatal mouse retinas. Conclusions/Significance Our study provides information useful to the development of new means of selectively manipulating angiogenesis without affecting homeostasis in un-targeted tissues; not only in eyes but also in various disease settings such as cancer.


Introduction
Angiogenesis is the process of the formation of vascular networks characterized by sprouting, branching, and regression of new blood vessels [1]. Because vascular endothelial growth factor (VEGF, also known as VEGF-A) plays predominant roles in this complex process by promoting proliferation, migration, and survival of endothelial cells (ECs), drugs inhibiting VEGF signaling have been globally administered to suppress deregulated angiogenesis in a variety of eye diseases, including age-related macular degeneration [2,3] and retinopathy of prematurity [4]. However, adverse effects of VEGF deprivation have been indicated in the normal functions of retinal neurons and Müller glias which constitutively express VEGF receptor 2 (VEGFR2) [5][6][7]. Thus, it is desirable to develop an alternative modality which can selectively target abnormal vessels, without affecting homeostasis in neural tissues.
The small GTPase Cdc42, which cycles between an active, GTP-bound state and an inactive, GDP-bound state, facilitates actin polymerization in various types of cells and is critically involved in diverse cell processes, such as cell motility [8]. In ECs, Cdc42 is activated by binding of VEGF to VEGFR2, whereas binding of semaphorin 3E (Sema3E) to PlexinD1 receptor inactivates Cdc42 [9,10]. By contrast, RhoJ, which displays 55% homology to Cdc42 in its amino acid sequences, is inactivated by VEGF and activated by Sema3E in ECs [10]. Intriguingly, while RhoJ binds to effector proteins of Cdc42 such as p21-activated kinase (PAK) and neural Wiskott-Aldrich syndrome protein [11,12], RhoJ induces actin depolymerization in ECs [10]. Thus, the inverse regulation of the activation status of Cdc42 and RhoJ downstream of VEGF-VEGFR2 and Sema3E-PlexinD1 signals are the pivotal intracellular events to mediate the cytoskeletal reorganization in ECs. Because small molecule inhibitors targeting guanine nucleotide exchange factors (GEFs) are expected to have therapeutic value [13,14], an endothelial GEF which activates Cdc42 or RhoJ would be a promising target for novel antiangiogenic therapies.
Here, by utilizing fluorescence-activated cell sorting (FACS) and microarray transcriptome profiling in conjunction with in silico bioinformatics analyses, we show that Arhgef15 (also known as Vsm-RhoGEF [15] and Ephexin5 [16]) acts as an EC-specific GEF to mediate VEGF-induced Cdc42 activation and further potentiates RhoJ inactivation, thereby promoting actin polymerization. Inactivation of the Arhgef15 gene resulted in retardation of retinal vascular growth, indicating Arhgef15 as a potential drug target.

Transcriptome Analysis in ECs of Living Mouse Retinas
In order to identify RhoGEFs which are highly expressed in retinal ECs, we performed transcriptome analyses in mouse retinas, in which new blood vessels begin to grow radially from the optic disc shortly after birth, and subsequently form a network in the most superficial layer [17,18]. Until the growing blood vessels reach the retinal periphery around postnatal day 9 (P9), new blood vessels continuously sprout at the leading fronts, whereas perpendicular vessels sprout from the preformed veins and capillaries around P8 to form the deep and intermediate vascular layers. To comprehensively analyze endothelial gene expression at these angiogenic stages, we dissociated retinas of P8 Tie2GFP transgenic (Tg) mice into single cells, and purified GFP-positive ECs by FACS ( Figure 1A, B, and Figure S1A). Because retinas of hemizygous Tie2GFP Tg mice showed incomplete separation of GFP-positive and GFP-negative cell fractions ( Figure S1B), we employed homozygous Tie2GFP Tg mice in all following experiments. In retinas of P8 Tie2GFP Tg mice, only 0.1% of the cell population represented GFP-positive cells. Accordingly, we obtained less than 500 GFP-positive cells per retina. However, reanalysis of sorted GFP-positive cells ensured nearly 100% purity ( Figure 1B), which was further confirmed by the expression of the Tie2 transcript exclusively in the GFP-positive cell fraction ( Figure 1C). We then performed microarray analysis on these cells. Using the Significance Analysis of Microarrays (SAM) algorithm [19] with a fold change .2 and a median false discovery rate (FDR) ,0.01, we identified 1,623 probe sets out of 36,701 probe sets as having higher signals in the GFP-positive cells than in the GFP-negative cells in P8 retinas (Dataset S1).
To identify endothelial genes encoding guanine nucleotide binding (G) proteins, we used the online NIH-DAVID software to extract probe sets whose gene ontology terms included 'GTP', such as 'GTP binding', 'GTPase activity', 'GTPase activator activity', 'small GTPase regulator activity', and 'small GTPase mediated signal transduction'. Of the 1,623 endothelial probe sets, 96 probe sets were related to the 'GTP' gene ontology terms (Dataset S2).
To ensure the integrity of microarray data, the exclusion of false-positive and false-negative errors is inevitably required [20]. In the present study, we performed in silico validation of the microarray data using the eXintegrator system, which displays the individual probe intensities in distinct probe sets across multiple samples [21]. We inspected microarray data sets obtained from entire P8 retinas (un-sorted), GFP-negative and GFP-positive P8 retinal cells, as well as GFP-positive cells from adult retinas (n = 3 for each population, Figure 1D). Importantly, microarray data with similar fold-change and FDR estimates showed variable degrees of internal correlation among probe-set member probes ( Figure 1E). Therefore, we evaluated the internal probe-set consistency by visually inspecting the identified probe sets ( Figure  S2) and extracted probe sets having f-ratios (variance between sample means to variance within member probes) higher than 1.0 using the eXintegrator system ( Figure 1F). Thus we determined 56 endothelial genes encoding G proteins in postnatal mouse retinas (Table S1).

Arhgef15 Facilitates Actin Polymerization and Cell Motility in ECs
To assess the roles of Arhgef15 in ECs, we transfected siRNA for Arhgef15 in cultured HUVECs. Arhgef15 knockdown did not affect the cell proliferation and apoptosis ( Figure S5), but induced depolymerization of actin filaments and a 28% decrease in cell surface areas, leading to cell collapse in 5.9% of siRNA-treated cells ( Figure 3A-D). Similar effects were observed by Cdc42 knockdown or Sema3E stimulation, but with higher rates of cell collapse ( Figure 3A-D). In addition to HUVECs, actin depolymerization by Arhgef15 knockdown was also induced in cultured human retinal microvascular ECs (HRECs, Figure S6). By contrast, a small population of Arhgef15-overexpressing HUVECs formed thickened actin fibers, which was more frequently observed in HUVECs overexpressing constitutively-active Cdc42 ( Figure 3E). Together with the disruption of actin fibers in RhoJoverexpressing HUVECs ( Figure 3E), these results indicate that Arhgef15 facilitates actin polymerization by activating Cdc42 and inactivating RhoJ in ECs.
Because reorganization of actin cytoskeletons is prerequisite for cell motility [23], we further performed scratch-wound assay and tube formation assay after siRNA transfection. Arhgef15 knockdown significantly reduced EC migration ( Figure 3F, G) and capillary-like network formation ( Figure 3H, I), which was more

Arhgef15 Promotes Extension of Retinal Vascular Networks
To evaluate the contributions of Arhgef15 to retinal angiogenesis, we generated Arhgef15 knock-out (KO) mice, in which endogenous Arhgef15 expression can be monitored by lacZ expression ( Figure S7). In agreement with our ISH experiments, lacZ was expressed in ECs over the developing retinal vasculature in Arhgef15-KO mouse ( Figure 4A). Higher magnification images further demonstrated lacZ expression in ECs both at the tips and stalks of sprouting vessels ( Figure 4B and Figure S8A). Moreover, lacZ expression was undetectable in neurons, astrocytes, and vascular smooth muscle cells (vSMCs), ensuring the endothelial specificity of Arhgef15 expression in mouse retinas ( Figure 4C and Figure S8B-D). At later postnatal stages, while Arhgef15 was uniformly expressed in ECs of deeper vascular layers, its expression was faint in arterial ECs of the superficial vasculature. ( Figure 4D). Although pups homozygous for the Arhgef15 null alleles were viable without any gross abnormalities, the radial extension of the retinal vascular network was retarded by 18.5% at P5 and by 9.9% at P8, compared to wild-type (WT) mice ( Figure 5A, B). Because the growing blood vessels in the Arhgef15-KO mice reached the retinal periphery at P10 and formed deeper vascular layers as in WT mice ( Figure S9), the retinal vascular retardation at early postnatal stages is likely to be ascribable to the impaired EC motility due to the lack of Arhgef15. In P5 Arhgef15-KO retinas, while the vessel density and the number of vessel branch points were unaffected, the delayed vascular extension resulted in 28.3% reduction of the total vessel area ( Figure 5C). Interestingly, the degree of vascular growth retardation in Arhgef15-KO retinas was equivalent to that caused by endothelial RhoJ overexpression ( Figure 5D, E), indicating that Arhgef15 and exclusively up-regulated in all GFP-positive samples. The x axis represents samples (with 3 replicates of each cell type) and the y axis signal intensity. Vertical red lines indicate borders between sample groups. (E) Representative images showing the discrepancy between statistics values and probe-pair signals. Despite the equivalent fold change and FDR values, the eXintegrator analysis for 93450_at (Centromere protein e, Cenpe), 161025_f_at (Sry-box containing gene 18, Sox18), and 161099_at (Src homology 2 domain-containing transforming protein e, She) indicates a high variability in internal correlation of individual probes. This can also be inferred from f-ratios calculated as the ratio of variance between sample means to variance within probe set member probes after individual z-score normalization of member probes across the sample series. (F) A scheme for the identification of endothelial G proteins. doi:10.1371/journal.pone.0045858.g001

Discussion
In order to identify novel target molecules for the treatment of neovascular eye diseases, we performed comprehensive transcriptome analyses in postnatal mouse retinas, because a considerable number, if not all, of endothelial genes in developing retinal vessels are also expected to be expressed in pathological angiogenesis. To date, a series of high-throughput analyses exploiting cultured or tissue-derived ECs have shown that endothelial gene expression fluctuates depending on microenvironments [24][25][26]. This notion prompted us to isolate retinal ECs from Tie2GFP Tg mice, which can reduce alterations in gene expression by omitting timeconsuming antibody labeling steps. Indeed, the bioinformatics analysis of our microarray data using the NIH-DAVID software demonstrated that most of the gene ontology terms (level 3 biological processes with p-value ,10 25 ) enriched in the 1,623 endothelial genes were associated with dynamic cell behavior such as ''cell motility'', ''cell migration'', and ''angiogenesis'' (Table S2). However, we should be aware of drawbacks inherent in our FACS and microarray analyses. Firstly, the expression levels of short-life genes, such as those up-regulated under hypoxia, may be affected  during the sample preparation. Secondly, endothelial genes expressed at limited sites, such as sprouting vascular tips [27,28], might be masked by the lack of expression in the majority of ECs. Thirdly, the relative comparison of gene expression levels between ECs and non-ECs does not necessarily extract EC-specific genes. To overcome these obstacles, the exact expression pattern of particular genes should be determined by highly sensitive procedures, such as the use of reporter mouse strains.
While 80 RhoGEFs comprising the Dbl family and the Dock family have been identified [22], we successfully discovered that Arhgef15 acts as an EC-specific GEF to activate Cdc42 downstream of VEGF signaling. Although endothelial Arhgef15 expression was also detected in embryonic vasculature ( Figure  S3B), previous literature reported Arhgef15 expression in vSMCs [15] and in neurons [16]. Thus, the Arhgef15 expression should further be examined in various tissues during development and in adults. In addition to Cdc42 activation, Arhgef15 potentiated RhoJ inactivation, which may underlie VEGF-induced RhoJ inactivation [10]. In this process, a distinct GTPase activating protein [13] downstream of Arhgef15 or Cdc42, may directly inactivate RhoJ. The full understanding of the Arhgef15-mediated signaling cascades will provide further information for the discovery of novel drug targets.
Based on the impaired vascular growth in Arhgef15-KO retinas, we propose that Arhgef15 is a potential molecular target for selectively manipulating angiogenesis without affecting tissue homeostasis. Given the lower collapse rate of HUVECs induced by Arhgef15 knockdown than by Cdc42 knockdown, it seems likely that multiple GEFs, such as Fgd5 [29], are involved in Cdc42 activation in ECs. It is also possible that Arhgef15 modulates the activity of alternative Rho small GTPases, such as RhoA [15,16]. Therefore, enhanced anti-angiogenic effects may be achieved by targeting multiple endothelial G proteins. For this purpose, our transcriptome profiling is a useful source of information to identify novel target molecules for the treatment of neovascular eye diseases, as well as diseases such as cancer characterized by deregulated angiogenesis.

Ethics Statement
All experimental protocols adhered to the Association for Research in Vision and Ophthalmology (ARVO) Statement for the Use of Animals in Ophthalmic and Vision Research, and were reviewed and approved by the Institutional Animal Care and Use Committee of RIKEN CDB (permit number H15-9).

FACS and Microarray
Retinas of homozygous Tie2GFP Tg mice were dissected in staining solution (phosphate buffered saline (PBS) containing 2 mM ethylenediaminetetraacetic acid (EDTA), 0.01% NaN 3 , 5% fetal bovine serum (FBS), 50 U/ml penicillin, and 0.05 mg/ml streptomycin), rinsed in PBS, and digested in papain solution (PBS containing 33 U/ml papain (Sigma-Aldrich), 125 U/ml DNaseI (Sigma-Aldrich), 0.4 mg/ml L-cysteine (Nacalai Tesque), and 0.5 mM EDTA) for 60 min at 37uC. Dissociated cells were treated twice with ovomucoid solution (PBS containing 2 mg/ml ovomucoid (Sigma-Aldrich), 1 mg/ml bovine serum albumin (BSA), and 125 U/ml DNaseI) for 60 min at 37uC and filtered through nylon mesh. After centrifugation, cells were incubated again in ovomucoid solution for 60 min at 37uC, rinsed twice in the staining solution, and resuspended in the staining solution containing 5 mg/ml of propidium iodide. After filtration through nylon mesh, cells were separated into GFP-positive and GFP-negative fractions by FACSAria (BD). Doublet cells were excluded using the 2dimensional profile of forward versus side scatter. Using total RNA extracted from more than 2610 4 sorted cells by RNeasy Micro Kit (QIAGEN) and amplified by MessageAmpII aRNA Amplification Kit (Ambion), biotin-labeled cRNA was prepared with BioArray RNA Transcript Labeling Kit (Affymetrix) and hybridized onto the Affymetrix oligonucleotide arrays MGU74v2 as described previously [34]. Primary microarray data have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO series accession number GSE27238 (http://www. ncbi.nlm.nih.gov/geo/query/acc.cgi?acc = GSE27238).

Bioinfomatics Analyses
The dChip software [35] (version 1.3) was used to normalize the CEL files at probe level and compute model-based expression values using the PM only model. A two-sample comparison was performed using the SAM algorithm [19] with 800 permutations. We used a minimum two-fold change in expression and a d-value set to give a median FDR ,0.01. Gene ontology annotations and classifications were generated using the DAVID (Database for Annotation, Visualization and Integrated Discovery) software Arhgef15 Activates Cdc42 and Inactivates RhoJ PLOS ONE | www.plosone.org (version 6.7) from the NIH (http://david.abcc.ncifcrf.gov/). The eXintegrator system was used to manually validate probe sets identified by the above statistical method as well as to calculate fratios for individual probe sets (http://www.cdb.riken.jp/scb/ documentation/).

In situ Hybridization
Whole-mount in situ hybridization (ISH) in P4 ICR retinas and 10.5 dpc ICR embryos was performed as previously described [36]. The template for Arhgef15 cRNA probe (1203-2343 of NM_177566) was obtained by reverse transcription of total RNA from P4 ICR mouse retinas with subsequent PCR amplification. After the ISH protocol was completed, the retinal samples were further labeled with rabbit anti-collagen IV polyclonal Ab (Cosmo Bio LSL) and Cy3 anti-rabbit secondary Ab (Jackson ImmunoResearch). Images were taken with an Axioplan2 microscope (Zeiss) equipped with associated software (AxioVision, version 3.1; Zeiss).

Pull-down Assay
HEK-293T cells cultured in Dulbecco's modified Eagle medium (Sigma-Aldrich) containing 10% FBS were transfected with indicated plasmids with Lipofectamine 2000, and incubated for 48 h before cell lysis. HUVECs pre-treated with siRNA for 72 h were serum-starved with Endothelial Basal Medium EBM-2 (Lonza) containing 1% BSA for 3 h, followed by stimulation with 50 ng/ml human VEGF (HumanZyme) for the indicated periods. The cells were washed with ice-cold PBS, lysed at 4uC in pulldown lysis buffer (20 mM Tris-HCl at pH 7.5, 100 mM NaCl, 10 mM MgCl 2 , 1% TritonX-100, 1 mM ethylene glycol tetraacetic acid, and 1 mM dithiothreitol), and centrifuged for 10 min at 4uC. The supernatants were incubated with GST-PAK-PBD (Cytoskeleton Inc.) conjugated to glutathione agarose beads for 40 min at 4uC, subjected to SDS-PAGE, and transferred onto a polyvinylidene difluoride membrane (Millipore). After blocking with Blocking One (Nacalai Tesque), the membranes were incubated with anti-Cdc42 (clone 44/CDC42; BD Transduction Laboratories) or anti-HA (clone 3F10) Ab, washed, and further incubated with horseradish peroxidase-conjugated anti-mouse (Zymed) or anti-rat (GE healthcare) secondary Abs and Immu-noStar LD (Wako). Aliquots of total cell lysates were also immunoblotted with anti-Cdc42 or anti-HA Ab. Signals were quantified using NIH ImageJ software. The activity of Cdc42 and RhoJ was assessed by calculating the ratio of their GTP-bound forms relative to their total protein amounts. The resulting values were expressed as fold changes compared with the control values (Mock in Figure 2A, si-Ctrl with VEGF 0 min in Figure 2B, and RhoJ-WT in Figure 2C) which were normalized to 1.0. All experiments were repeated three times.

Cell Proliferation and Apoptosis Assay
At 24 h after siRNA transfection, 1610 4 HUVECs were seeded on a 24-well plate and incubated in EGM-2 for 24 h. After serumstarvation with EBM-2 containing 1% BSA for 3 h, cells were incubated with 30 mg/ml 5-bromo-29-deoxy-uridine (BrdU, Nacalai Tesque) with or without 50 ng/ml human VEGF for 24 h. After fixation with 4% PFA in PBS, cells were labeled with FITCconjugated anti-BrdU (Roche) or anti-cleaved Caspase-3 (Cell Signaling Technology) Ab. Fluorescence images were taken with an Axio Observer microscope (Zeiss) equipped with associated software (AxioVision, version 4.8.1; Zeiss). The proliferating and apoptotic cells were calculated by averaging the number of BrdUand Caspase-3-positive cells counted in 2 randomly-selected fields (0.57 mm 2 ) per well using WinROOF software (version 6.5, Mitani Corp.). All experiments were repeated three times.

Scratch-wound Assay
At 24 h after siRNA transfection, 1610 5 HUVECs were seeded on a 24-well plate and incubated in EBM-2 containing 1% BSA and 50 ng/ml human VEGF overnight. After the cell monolayers were wounded with 200 ml plastic pipette tips, cells were incubated in EBM-2 containing 1% BSA, 50 ng/ml human VEGF, and 10 mg/ml mitomycin C (Sigma-Aldrich). Images were taken at 0 and 6 h using an Olympus IX-81 inverted microscope. The wound areas were measured using NIH ImageJ software. To quantify wound closure, the wounds at 0 h were considered to be 0% closed.

Tube Formation Assay
At 48 h after siRNA transfection, 1.5610 4 HUVECs were seeded on a 96-well plate pre-coated with the growth factor reduced Matrigel (BD Biosciences). After 10 h incubation in EGM-2, images were taken with an Axio Observer microscope. The total tube length in a fixed area (2.4 mm 2 ) of each well was measured using WinROOF software.

Morphometric Analysis of the Retinal Vasculature
The radius of the retinal vasculature was calculated by averaging the distances from the optic disc to the sprouting vascular fronts in four quadrants per one retina. The vessel area was calculated by measuring the PECAM-1-positive area (except for the optic disc) in whole-mount retina using WinROOF software. The vascular density was assessed by calculating the proportion of the total vessel area to the vascularized retinal area encircled by the sprouting vascular edges. The vessel branch point was calculated by averaging the number of vascular branch points in four fixed retinal areas (1.5610 5 mm 2 ) per one retina.

Statistics
Statistical analysis was performed with JMP software (version 8.0.2, SAS Institute Inc.) using an unpaired two-tailed Student's t test, one-way ANOVA with Dunnett's post-hoc test, or Kruskal-Wallis with Dunn's post-hoc test. P values of less than 0.05 were considered statistically significant.