Angiogenesis is a complex process, which is accomplished by reiteration of modules such as sprouting, elongation and bifurcation, that configures branching vascular networks. However, details of the individual and collective behaviors of vascular endothelial cells (ECs) during angiogenic morphogenesis remain largely unknown. Herein, we established a time-lapse imaging and computer-assisted analysis system that quantitatively characterizes behaviors in sprouting angiogenesis. Surprisingly, ECs moved backwards and forwards, overtaking each other even at the tip, showing an unknown mode of collective cell movement with dynamic ‘cell-mixing’. Mosaic analysis, which enabled us to monitor the behavior of individual cells in a multicellular structure, confirmed the ‘cell-mixing’ phenomenon of ECs that occurs at the whole-cell level. Furthermore, an in vivo EC-tracking analysis revealed evidence of cell-mixing and overtaking at the tip in developing murine retinal vessels. In parametrical analysis, VEGF enhanced tip cell behavior and directed EC migration at the stalk during branch elongation. These movements were counter-regulated by EC-EC interplay via γ-secretase-dependent Dll4-Notch signaling, and might be promoted by EC-mural cell interplay. Finally, multiple regression analysis showed that these molecule-mediated tip cell behaviors and directed EC migration contributed to effective branch elongation. Taken together, our findings provide new insights into the individual and collective EC movements driving angiogenic morphogenesis. The methodology used for this analysis might serve to bridge the gap in our understanding between individual cell behavior and branching morphogenesis.
Morphogenetic cell movement gives rise to various types of geometry observed in the living world. The processes are diverse, and recently the notion of collective cell movement has increasingly become a focus of research. Known collective cell movements include branching morphogenesis of the mammary gland, cluster movement of detached cancer cells, border cell migration in Drosophila eggs and lateral line primordium migration in zebrafish (Montell, 2008; Friedl and Gilmour, 2009).
Angiogenesis is a multicellular phenomenon whereby new blood vessels emerge from an existing vascular network in developmental and (patho)physiological contexts. Current understanding of angiogenesis encompasses tip cell selection by lateral inhibition, elongation, branching, anastomosis, vessel stabilization and lumen formation (Holderfield and Hughes, 2008; De Smet et al., 2009). Endothelial cell (EC) signaling involving the vascular endothelial growth factor (VEGF)-receptor (VEGFR), angiopoietin-Tie2 and Ephrin-Eph pathways has been intensively elucidated its role in these events (Armulik et al., 2005; Holderfield and Hughes, 2008; Gaengel et al., 2009). In addition, the interaction between ECs and mural cells (MCs), including vascular smooth muscle cells and pericytes, has also been implicated in the maintenance of the angiogenic process for appropriate organization (Hellstrom et al., 2001; Lafleur et al., 2001; Armulik et al., 2005; Gaengel et al., 2009; Liu et al., 2009).
Despite the extensive past studies, how the spatiotemporal regulation of molecules affects morphogenetic cell movement and what type of collective cell movement is involved in angiogenesis remain elusive, largely owing to the lack of a stable methodology for visualizing and assessing EC movements during angiogenesis. It is only recently that research has begun to focus on and assess real-time EC behavior during these processes (Murakami et al., 2006; Perryn et al., 2008; Vitorino and Meyer, 2008). In order to clarify the relationship between individual cell movements and angiogenic morphology, and to dissect the underlying molecular and cellular mechanisms, we first established a system in which dynamic cell behavior is visualized using time-lapse microscopy and set out to identify patterns of cellular behavior in an angiogenesis model through computational data processing. Surprisingly, cell movements were far more dynamic and heterogeneous in movement than previously thought. In elongating branches, ECs moved changing their relative positional relationships to each other at the tip. This ‘cell-mixing’ phenomenon was also confirmed in developing murine retinal vessels in vivo.
We further established a computer-assisted quantitative analysis system with which we can assess molecular and cellular mechanisms underlying complex EC movements. After validation of this system by analyzing the angiogenic effect of VEGF, we applied it to determining the role of Notch-related signaling in angiogenesis.
Recently, roles of Delta-like 4 (Dll4)-Notch signaling in angiogenesis have been demonstrated by a series of reports (Ridgway et al., 2006; Hellstrom et al., 2007; Siekmann and Lawson, 2007; Suchting et al., 2007). Murine retina heterozygous for a null mutation of the Dll4 gene showed excessive branching and this was recapitulated by administering a γ-secretase inhibitor, DAPT. Zebrafish in which Dll4 signaling was abolished using morpholino knockdown had increased migratory cells in intersomitic vessels. Blockade of Dll4 resulted in deregulated angiogenesis with a rather paradoxical reduction in tumor size. All of these findings have been attributed to a signaling cascade in which Dll4, by acting through Notch1, regulates tip cell differentiation. However, reports focusing on how individual EC movements are integrated into vascular morphology remain scarce.
The present analysis revealed that EC-EC interplay via Dll4-Notch signaling counter-regulated not only VEGF-induced tip cell behavior but also directed migration of stalk cells. In addition, EC-MC interplay might positively regulate EC behaviors, implying a novel role for MCs in the early stage of angiogenesis.
These findings, along with our time-lapse quantitative analyses, provide new insights into EC movements during angiogenesis, and the imaging and analysis system that we established should prove to be a powerful tool for linking individual cellular behaviors, molecular events and morphogenetic collective cell movements.
MATERIALS AND METHODS
ICR mice were purchased from Charles River Laboratories Japan and used in all experiments except for those on mice carrying the EdnraEgfp (EGFP-knock-in) allele (Asai et al., 2010). All animal experiments were reviewed and approved by The University of Tokyo Animal Care and Use Committee and were performed in accordance with the institutional guidelines.
The following antibodies were used: rat anti-CD31 (BD Pharmingen); rabbit anti-Dll4 and rat anti-platelet-derived growth factor receptor β (PDGFRβ) (Biolegend); rabbit anti-GFP (MBL); rabbit anti-NG2 and rabbit anti-phospho-histone H3 (PH3) (Millipore); and rat anti-PDGFRβ for neutralization (a gift from Dr Shin-ichi Nishikawa, RIKEN, Kobe, Japan).
Aortic ring assay and time-lapse live imaging
The aortic ring assay was performed as previously described (Blacher et al., 2001). Aortic rings were embedded in type I collagen gel (Nitta Gelatin) on eight-well chambered coverglasses (Nunc) and were cultured in medium-199 containing 5% FCS, 10 μg/ml streptomycin, 100 units/ml penicillin and 50 ng/ml human recombinant VEGF (R&D) (complete medium).
Time-lapse live imaging was started on the 5th or 6th day of the assay. For EC tracking, we labeled cells with SYTO-16 or SYTO-61 (0.1 μg/ml, Molecular Probes). Dynamic cell behavior was analyzed using a confocal laser scanning microscope (FluoView FV10i Olympus). Time-lapse live images were taken at 8-10 μm intervals for the z-axis every 15 minutes over 36 hours (10× 0.4 NA air objective), and for some assays, at 3 μm intervals every 5-7 minutes over 5-12 hours (60× 1.2 NA water-immersion objective). Obtained images were processed with analysis software FLUOVIEW (Olympus).
For mosaic analysis, the aortic ring was infected with adenovirus carrying EGFP (1×105 p.f.u./ml) (a gift from Dr Hideyuki Sakoda, The University of Tokyo, Japan) before cultivation. If needed, cells were incubated with 5 μg/ml of BS-1 lectin to visualize ECs.
To examine the effect of VEGF on sprouting angiogenesis, culture media were replaced with medium-199 containing 5% FCS, 10 μg/ml streptomycin, 100 units/ml penicillin and the indicated dose of VEGF 3 hours prior to time-lapse imaging. To inhibit Dll4-Notch signaling, sprouts were pre-treated with 50 μg/ml anti-Dll4-Ab (Yamanda et al., 2009) or 3 μM N-[N-(3,5-difluorophenacetyl-L-alanyl)]-S-phenylglycine t-butyl ester (DAPT, Sigma), a γ-secretase inhibitor, 24 hours prior to time-lapse live imaging. For morphological analysis, anti-Dll4-Ab or DAPT was added throughout the assay. To inhibit the PDGFRβ signal, cells were treated with anti-PDGFRβ-Ab (10 μg/ml) in complete medium throughout the assay (Uemura et al., 2002).
In vivo EC-tracking analysis
To label ECs of the retinal vasculature intravascularly, 3-5 μg of fluorescent-conjugated BS-1 lectin (Sigma) was injected into the cardiac chamber of neonatal mice beginning on postnatal day 1 using glass capillary pipettes. Retinas were obtained at the indicated time after the injection.
In silico analysis
Cell tracking, data extraction and data analysis were carried out using ImageJ, MTrackJ and MATLAB.
Nuclei of each EC were manually selected, and whenever two or more nuclei overlapped, general rules were applied to define the next points of the current track: (1) when there is obvious flow of cells in one direction, a cell is likely to move on that axis either anterograde or retrograde, albeit with some deviation; (2) in the setting of (1) and when there are two or more candidate axes, cells are prone to move on their previous axis and in their prior direction; (3) in the setting of (2), when there are more than two candidate positions for the next slice, either the nearest possible and/or the one most likely regarding the previous velocity is chosen; and (4) normally, cells avoid bumping into each other by fine-tuning their velocities and directions.
The term ‘tip cells’ refers to ‘cells at the tip of a certain time point’. The other, non-tip cells, were termed ‘stalk cells’. A branch was defined as a multicellular protrusion with more than two ECs. A ‘junction’ was defined as a point of divergence, juncture or intersection of several vessels, and a ‘stalk’ was the region distal to a junction. The area of analysis was defined by setting four coordinates that surround all nuclei in a stalk or a junction. The latter was set to exclude as the cells of distal stalks as much as possible.
The axis and direction of elongation (elongation vector, E) was defined as E=r(end)-r(0), where r(t) stood for the coordinates of the tip at time=t (‘end’ stands for the last observed time point). When a branch sprouted out after the start, t=0 meant the start of elongation and thus of the sprouting event. To this end, the tip was tracked independently. The coordinates of each cell were calculated by orthogonal projection to the axis of elongation. Therefore, e(i,t)=dot[E,(r(i,t)-e0)]/|E|, where e(i,t) stood for the one-dimensional coordinates of a cell with an ID=i at time=t obtained by orthogonal projection. r(i,t) was the original coordinates of i-th cell at time=t. 0 on this axis meant the initial coordinates of the tip (e0) projected to the elongation vector, and dot[A,B] was the internal product of vectors A and B. The coordinates of a cell with an ID=i at time=t were represented as vector r(i,t). |A| stood for the absolute value of vector A; thus, |A|=sqrt(dot[A,A]), where sqrt(a) was the square root of a. The coordinates were plotted against time.
The point at which a cell was overtaken was visually determined as when the centroid of the pre-existing tip cell fell behind that of the cell coming from behind for at least two frames. For tip cell parameters, ‘elongation drive’ (net elongation of a branch)/(number of tip cells during the total observed time) and ‘tip duration’ (overall time)/(number of tip cells during the total observed time) were introduced. For parameters at stalks and junctions, the ‘coordination’ index was set. The angle [θ(i,t)] between the direction of elongation or the bisector (elongation direction, D) and the moving direction of a cell [velocity vector, v(i,t)] was calculated and plotted against time. D was set so as to face the direction of net cell flow, parallel to the axis of the distal stalk, or the bisector of distal stalks if there were two outflows, at junctions. The angle was calculated by the equation, θ=Arccos(dot[D,v]/|D||v|), where θ ranges from 0 to π and v=v(i,t)=r(i,t+15)-r(i,t). Thus, in coordinated junctions, θs were expected to be narrowly distributed around the mean. Quantification was performed by pooling the time series and taking the standard deviation (Std) for anterograde (θ<π/2) and retrograde (θ>π/2) separately. The ‘orientation’ index was defined as (net displacement of the cell)/(total length of the track of a cell). Net displacement of a cell [R(i)] was calculated by R(i)=|r(i,end)-r(i,start)|, where ‘start’ and ‘end’ referred to the time points at which the cell enters (‘start’) and moves out of (‘end’) the analyzed area, and the total length of the track of a cell [L(i)] was calculated by L(i)=Σ|v(i,t)|, where sigma (Σ) was the summation for time (t=start to t=end). A value near 1 means the cell has moved straight and with no change in direction in the analyzed area. Percentages of ECs moving forwards (anterograde, θ<π/2), backwards (retrograde, θ>π/2) or remaining (still, v=0) were also calculated as the ratio to the total number of cells observed in a particular stalk area throughout the period from t=0 to t=end (‘Directional motility’). ECs that moved fewer than 0.5 pixels were counted as stopped. Finally, mean absolute velocity (<v(i)>) at a specific area of analysis was calculated as <v(i)>=Σ|v(i,t)|/Δt, where Δt stood for the time interval between the start and the end.
Whole-mount staining and imaging
For murine aortic assays, whole-mount staining was performed as described previously (Yamashita et al., 2000). If needed, nuclei were counterstained with TOPRO3 (Invitrogen). For live cell imaging, SYTO-dye-stained cells were immunostained with fluorescent-conjugated anti-CD31- and PDGFRβ-Abs for 30 minutes prior to imaging. In the murine retina, whole-mount immunostaining was performed as previously described (Uemura et al., 2002). Fluorescent signals were visualized with a computer-assisted confocal microscope (Nikon D-ECLIPSE C1). Photomicrographs were obtained at 1-10 μm intervals as necessary and reconstituted using EZ-C1 software (Nikon). Signals were also visualized by an enzymatic reaction with 3,3′-diaminobenzidine (Dojin) chromogen, and photographs were obtained with a phase-contrast microscope (Nikon TE300, Tokyo, Japan).
Mitotic events were detected on movies (see Fig. S2A in the supplementary material). The percentage was calculated by dividing the number of ECs undergoing mitosis at specified places by the total number of ECs appearing in an observed movie. Mean observed time per cell was also calculated. In addition, angiogenic sprouts were immunostained with PH3, a marker of mitosis (Hendzel et al., 1997; Dai et al., 2005), at day 7. The number of PH3+ ECs was counted in randomly selected areas (including the sprout from the bottom to the tip), and the rate of PH3+ ECs to total ECs per ring was calculated. Furthermore, observed areas were divided into two specified areas (‘distal 400’, 400 μm area proximal to the tip; ‘proximal’, the area more proximal than the ‘distal 400’). Similar analyses were performed for murine retinal angiogenesis.
Z-stack images of CD31+ endothelial sprouts were obtained using a confocal microscope as described above. After optimization on Photoshop, endothelial sheet-forming areas and parts of the explanted ring were manually eliminated. The total length of angiogenic sprouts was then measured on the converted binary images using the NeuronJ plug-in for ImageJ (‘total length’). The total number of branching points was manually counted (‘total branch points’) and the ‘relative branch point’ was defined as (‘total branch points’)/(‘total length’). The ‘mean width’ was calculated by dividing total area of the sprouts by ‘total length’.
After the explanted rings had been manually extracted from type I collagen gels, the gels were incubated in medium-199 containing 1 mg/ml collagenase (Sigma) at 37°C for 30 minutes, followed by incubation with 0.05% trypsin-EDTA (Sigma). The cells were immunostained with an anti-CD31-Ab conjugated with PE on ice. Analyses were performed on a FACS VantageSE flow-cytometer (BD Biosciences), and data were analyzed with CellQuest software (BD Biosciences). In the assay, nonviable cells stained with propidium iodide (Sigma) were excluded by electronic gating.
Cells were harvested as described above. Extracted total RNA was reverse-transcribed using a conventional method. For semi-quantitative RT-PCR, the resultant cDNAs were amplified with Taq polymerase (Takara) in a thermocycler. The primer pairs used were as follows: for murine Hes-1, 5′-TCATGGAGAAGAGGCGAAGG/GTATTTCCCCAACACGCTCG-3′; for murine Hey2, 5′-CCGAGAGTGCTTGACAGAAGTG/TGTGGGGAGATGGTGGAAA-3′; for murine CD31, 5′-AGGACAGACCCTTCCACCAA/AATGACAACCACCGCAATGA-3′; and for murine GAPDH, 5′-GGTGTGAACCACGAGAAATAT/AGATCCACGACGGACACATT-3′. Thermal cycling was performed for 18-30 cycles to maintain PCR conditions within the linear range of amplification before reaching saturation. Each cycle consisted of 30 seconds of denaturation at 94°C, 30 seconds of annealing at 58°C and 1 minute of extension at 70°C.
All data were compared between groups using the Mann-Whitney test and Kruskal-Wallis test accompanied by multi-comparison analysis where necessary. Pearson correlations were used to evaluate the relations between vessel elongation and other parameters. The relationships between vessel elongation and other parameters were studied using a linear regression analysis, which was performed separately for models involving single or multiple independent variables. The variance inflation factor was also estimated for all independent parameters in order to evaluate co-linearity. P<0.05 was considered significant. Data in bar graphs were represented as means±s.e.m.
Fluorescent imaging of ECs during in vitro angiogenesis
To characterize the behavior of vascular ECs during angiogenic morphogenesis, we visualized the dynamics of cell movement during in vitro angiogenic processes in a murine aortic ring assay (Blacher et al., 2001), using time-lapse microscopy. In this assay, various modules of the angiogenic processes such as endothelial sheet formation, sprouting, circumferential growth, branching and the resultant formation of a characteristic dendritic architecture were observed under VEGF stimulation (Fig. 1A). The sprouts of CD31+ ECs were covered with NG2+, PDGFRβ+ MCs (Ozerdem et al., 2001; Armulik et al., 2005), and also characterized by endothelin receptor type A (Ednra)-EGFP knock-in gene expression (Asai et al., 2010) (Fig. 1B-D).
To identify and track individual ECs in the multicellular structure, the nuclei of living cells were labeled with a cell-permeable fluorescent SYTO dye, by which we succeeded in selectively labeling EC during angiogenic morphogenesis (Fig. 1E-G; see Fig. S1 in the supplementary material; see also Movies 1 and 2 in the supplementary material). Although nuclear staining was not specific for ECs, we found that ECs could be distinguished from MCs covering endothelial sprouts by their signal intensity: whereas ECs showed intense staining with SYTO dyes, morphologically distinguishable or Ednra-EGFP+ MCs were minimally stained (Fig. 1F; see Fig. S1C and Movie 2 in the supplementary material). In addition, this was supported by the results of live cell staining and imaging, which demonstrated that cells intensely stained with SYTO dyes were positive for CD31, whereas faintly stained cells were negative for CD31 or positive for PDGFRβ (Fig. 1G; see Fig. S1A,B in the supplementary material). The difference in SYTO dye stainability was further confirmed by flow-cytometric analysis (see Fig. S1D in the supplementary material). Thus, differences in nuclear staining as well as in morphology were used to visually distinguishing ECs from MCs.
Time-lapse live imaging reveals complexity of EC behavior during in vitro angiogenesis
In the aortic ring assay, elongation and bifurcation of an existing branch were evident. However, cellular behaviors underlying these events were far more complex and heterogeneous than expected (Fig. 2A; see Movie 1 in the supplementary material). Individual ECs moved forwards and backwards along the path of the elongation in stalks and junctions, changing their velocity. At the tip region, fast cells overtaking other ECs become the new tip, indicating replacement of tip cells (‘overtaking of the tip cell’). As a result, relative positional relationship of ECs was dynamically changed (‘cell-mixing’). This ‘cell-mixing’ phenomenon was clearly shown by labeling cells segmentally with different colors (Fig. 2A; see Movie 1 in the supplementary material). Although the elongating process was dynamic, the frequency of cell division was very low at around 5% of total ECs in the observed time (mean observed time per cell was 22.4 hours), even in the presence of VEGF (see Table S1 in the supplementary material). Consistently, we observed that the number of PH3+ ECs was quite low (0.8% of total observed ECs), with lower frequency in the proximal region (see Fig. S2B,C in the supplementary material). Observations were similar for murine retinal angiogenesis at P1 and P3 (see Fig. S2B,C in the supplementary material), also consistent with a previous report (del Toro et al., 2010). These results suggest that elongation and branching in the observed area are attributable mainly to EC migration rather than to proliferation.
Dynamics of cell movement during branch elongation
To further elucidate cell behavior in an elongating branch, we developed a trajectory analysis, which enabled us to evaluate visually and quantitatively single- and multi-cellular movements, as well as to compare the dynamics of cell movement between branches elongating in different directions (Fig. 2B; see Fig. S3 in the supplementary material). Trajectory analysis revealed that branch elongation was a repeated event in which an EC at the stalk overtook the tip cell to become the new tip (Fig. 2B, part a). The new tip cell became less motile after this overtaking (arrow), followed by being overtaken by another cell (arrowhead). Thus, branch elongation appeared to be promoted by repetitive overtaking (a dark-blue trajectory in Fig. 2B, part a). In some cases, a tip cell was overtaken by another EC (arrow) and then retreated to become a tip cell again (arrowhead) (purple trajectory in Fig. 2B, part b). Furthermore, overtaking (arrow) by a fast-moving EC from far behind (fast EC, a pink trajectory in Fig. 2B, part c) seemed to contribute to effective branch elongation. In a poorly elongating branch, overtaking by a fast EC was rare (arrowhead in Fig. 2B, part c). However, overtaking by a fast EC alone was not sufficient for elongation because some branches with overtaking by a fast EC did not show effective elongation (arrows in Fig. 2B, part d), suggesting that other conditions in which overtaking by a fast EC leads to effective elongation are also necessary.
Mosaic analysis confirms ‘cell-mixing’ phenomenon during in vitro angiogenesis
To confirm whether the identified phenomena such as ‘cell-mixing’ based on nuclear tracking occur on a whole-cell basis, we performed a mosaic analysis by visualizing the cytoplasm with EGFP. In this analysis, ECs could be identified by differences in the intensity of SYTO dye staining and, in some cases, staining for BS-1 lectin (Fig. 3). Mosaic analysis revealed that high EGFP-labeled EC (red arrows in Fig. 3A) appeared to ‘slip’ forward on the surface of a tip cell, overtaking and then being replaced by a follower EC without the EGFP signal (yellow arrows), which confirmed overtaking of the tip cell at a cellular level (see Movie 3 in the supplementary material). Mosaic analysis further showed that highly motile ECs displayed unidirectional forward-rear cell polarity, depending on the direction of movement, whereas minimally motile ECs were spindle-shaped without cell polarity (Fig. 3A-D; see Movies 3 and 4 in the supplementary material), and that the polarity of individual ECs was dynamically changed in conjunction with the motility (Fig. 3E; see Movie 5 in the supplementary material). These results provide a cellular basis for the phenomenon of angiogenic sprouting with ‘cell-mixing’.
Identification of ‘cell-mixing’ phenomenon during murine retinal angiogenesis in vivo
Next, we verified whether the ‘cell-mixing’ phenomenon occurs in vivo. We labeled ECs by intravascular injection of a BS-1 lectin at postnatal day 1.0 and tracked them during mouse retinal angiogenesis (Fig. 4A). At this stage, sprouting angiogenesis takes place on the retinal surface (Fig. 4B), and tip and distal stalk cells do not form a lumen (Gerhardt et al., 2003). Consistent with previous findings, the proximal (lumen-forming) regions of CD31+ endothelial sprouts were uniformly and reproducibly labeled by lectin shortly after its injection, whereas the distal regions (including the tip) were not (Fig. 4C). As expected, 6 hours after the injection, lectin-positive and -negative cells were colocalized at the distal regions in a mixed pattern. Typically, a CD31+ EC at the tip and some follower ECs were lectin positive in the distal regions (Fig. 4C; see Fig. S4A in the supplementary material), indicating that the ECs around the tip came from the proximal regions forming the lumen. Similar mosaic patterns were also observed at the distal and proximal positions of sprouts 12 hours after the lectin injection (see Fig. S4B in the supplementary material). Collectively, these results strongly suggest that the ‘cell-mixing’ phenomenon occurs during in vivo retinal angiogenesis.
Quantification of EC behaviors driving angiogenic morphogenesis
In order to unravel the cell-based mechanism that drive angiogenic morphogenesis and dissect the underlying molecular networks, we first set out parameters to quantify different angiogenic modules. VEGF also promoted angiogenesis in our setting (Fig. 5A; see Movie 6 in the supplementary material). One of the pro-angiogenic effects of VEGF is to elongate vessel branch, characterized herein by an increase in the index ‘vessel elongation’ (Fig. 5C), owing to the increased number of ECs in the branch (Fig. 5D). The tip of an elongating branch overtaking ECs was identified (Fig. 5B). Because overtaking of tip cells occurred regardless of the presence or absence of VEGF, we established ‘elongation drive’ (displacement per tip cell) and ‘tip duration’ (the time spent as a tip cell) parameters to quantify events at the tip further (Fig. 5B). Interestingly, VEGF-induced vessel elongation was closely associated only with greater elongation drive (Fig. 5C). Overtaking of tip cells occurred similarly at interval ranging from ∼350 to ∼900 minutes, whether or not VEGF was present. These results indicate that VEGF positively affects tip cell behaviors following overtaking without changing the frequency of overtaking.
We also set the following parameters to evaluate the complex cell movements at junctions and stalks: mean migratory velocity, coordination (whether or not ECs moved in the same direction (coordinated)), orientation (whether or not a single cell moved straightly), and directional motility (whether or not ECs migrated toward the direction of branch elongation) (Fig. 5E). In the presence of VEGF, ECs exhibited higher velocity and orientation at junctions and stalks (Fig. 5E-a,E-b). Coordination tended to be improved (P=0.051, for VEGF 0 ng/ml versus 50 ng/ml, Fig. 5E-c). At stalks, there was a higher proportion of ECs showing anterograde movement (Fig. 5E-d). Taken together, these data suggest that VEGF enhanced directed migration of ECs in the elongating direction at junctions and stalks. Parameters presented here are summarized in Fig. S6A in the supplementary material.
Dll4-Notch signaling pathway in collective EC movement driving angiogenic morphogenesis
To further define the molecular machinery involved in collective EC movements during angiogenesis, we interrupted Dll4-Notch signaling, which is crucial for tip cell regulation and proper angiogenesis (Hellstrom et al., 2007; Suchting et al., 2007). Consistent with previous findings (Claxton and Fruttiger, 2004; Hellstrom et al., 2007; Hofmann and Luisa Iruela-Arispe, 2007), Dll4 was expressed in ECs at the tip and, to a lesser extent, at stalks of angiogenic sprouts, but never in MCs (Fig. 6A). Dll4-Notch signaling was inhibited using a neutralizing Dll4-Ab (Yamanda et al., 2009). The inhibitory effect was confirmed by downregulation of Hey2, a Notch downstream target gene (Fig. 6B). In the aortic ring assay, Dll4-Ab treatment resulted in morphological changes with increased vessel branching and total sprout length (see Fig. S5A in the supplementary material), resembling the vascular phenotype in the retina and tumors of Dll4-mutant mice (Noguera-Troise et al., 2006; Hellstrom et al., 2007; Suchting et al., 2007).
In parametrical analysis, Dll4-Ab treatment enhanced vessel elongation with a concomitant increase in the EC number in each branch in the presence of VEGF (Fig. 6D,E), but did not affect the pattern of tip cell overtaking (Fig. 6C). The increase in vessel elongation was due to increased mean elongation drive without changes in mean tip duration (Fig. 6D). Dll4-Ab treatment also affected stalk cell behavior: mean EC velocity was increased at junctions and stalks, and the ‘still’ cell population was decreased, whereas other parameters were not changed by Dll4-Ab treatment (Fig. 6F). Increases in vessel elongation and mean velocity of ECs were also obvious in moving images (see Movie 7 in the supplementary material).
Dll4-Notch signaling was also inhibited by treatment with DAPT (Hellstrom et al., 2007; Suchting et al., 2007). This inhibitory effect was confirmed by downregulation of Hes1, another Notch downstream target gene (Fig. 6B). In addition, DAPT treatment eventually reduced Dll4 protein in ECs, suggesting inhibition Notch signaling was occurring (Jakobsson et al., 2010) (Fig. 6A). DAPT treatment enhanced branching formation in the same way as Dll4-Ab treatment, although there was no change in the total length of angiogenic sprouts, and sprouts were also wider, reflecting in vivo morphological changes induced by DAPT (see Fig. S5B in the supplementary material) (Hellstrom et al., 2007; Suchting et al., 2007). In parametrical analysis, unexpectedly, DAPT treatment recapitulated only the two Dll4-Ab-induced parametric changes in mean EC velocity and directional motility profiles (Fig. 6C-F; see Movie 8 in the supplementary material). Comparisons of parametric changes between Dll4-Ab and DAPT treatments are summarized in Table 1. The parametric similarities clearly suggest that the Dll4-Notch signaling pathway negatively regulates mean velocity in collective EC movements at stalks and junctions.
Role of EC-MC interaction in collective EC movement driving angiogenic morphogenesis
DAPT can inhibit not only Notch signaling (Hellstrom et al., 2007; Suchting et al., 2007) but also other γ-secretase-dependent pathways (Boulton et al., 2008), some of which are involved in interplays between EC and MC (Gaengel et al., 2009; Liu et al., 2009). Indeed, the number of NG2+, Ednra-EGFP+ MCs covering endothelial sprouts was decreased (Fig. 7A). To test whether the differences between the effects of Dll4-Ab and DAPT are attributable to Dll4-independent EC-MC interactions, the cells were treated with PDGFRβ-Ab, which interrupts proper covering of endothelial sprouts by MCs and the resulting EC-MC interactions (Uemura et al., 2002; Jin et al., 2008; Liu et al., 2009). As expected, the association between ECs and NG2+, Ednra-EGFP+ MCs was diminished by PDGFRβ-Ab, while endothelial sprouting was evident (Fig. 7A). Interestingly, moving images showed that PDGFRβ-Ab treatment apparently increased ECs moving in the direction opposite branch elongation, resulting in retarded branch elongation (see Movie 9 in the supplementary material). Accordingly, parametric analysis revealed that both mean tip duration and elongation drive were decreased by PDGFRβ-Ab treatment with a preserved pattern of tip cell overtaking (Fig. 7B,C). In addition, PDGFRβ-Ab treatment increased retrograde directional motility and decreased the orientation index without changes in velocity or coordination (Fig. 7D), indicating disturbed directional movement. These results recapitulated a portion of the DAPT-induced parametric changes, which had not been observed with Dll4-Ab (Table 1). The similarities in parametrical changes between PDGFRβ-Ab and DAPT suggest that MCs may affect behaviors of tip cells and directed migration at stalks with collective EC movement in a γ-secretase-dependent manner. However, the parametrical changes in response to DAPT treatment might also be due to alterations of EC-EC interplay via γ-secretase-dependent pathways other than the Dll4-Notch pathway.
Molecular and cellular mechanisms driving vessel elongation
Relationships between parametrical changes and the angiogenic factors studied herein are summarized in Fig. 7E. To identify which compartment of the collective EC movement most efficiently contributes to branch elongation, we analyzed correlations between the vessel elongation index and other parameters. Preliminary analyses highlighted elongation drive and velocity at junctions as possible positive explanatory variables and retrograde directional motility as a possible negative explanatory variable (see Fig. S6B,C in the supplementary material). Then, we tested whether the above three parameters can explain vessel elongation by using a multi-regression model. The results suggested elongation drive, velocity at junctions and retrograde directional motility to be significant determinants of vessel elongation in the present setting (Fig. 7F).
Therefore, the parametrical analyses collectively suggested a model possibly explaining the molecular and cellular mechanisms underlying branch elongation (Fig. 7G); VEGF induces vessel elongation by enhancing the elongation drive, velocity and percentage of anterograde directional motility, with elongation drive and velocity being counteracted by Dll4-Notch signaling. MCs may act on ECs to promote elongation drive at the tip and the directional motility that drives branch elongation.
Herein, we established a system for spatiotemporally assessing collective EC movements involved in in vitro angiogenesis using live-cell imaging with time-lapse microscopy and computational analysis. This system revealed complex and heterogeneous EC behaviors during angiogenesis: individual ECs migrate forwards and backwards at different velocities, changing their relative positions within the branch network (‘cell-mixing’), even at the tip (‘overtaking of the tip cell’). Based on an EC-tracking analysis of murine retina, the ‘cell-mixing’ phenomenon also very likely to occur during in vivo angiogenesis. Parametrical and statistical analyses further demonstrated the molecular and cellular mechanisms underlying collective EC movement. EC-EC interplay via Dll4-Notch signaling counteracts VEGF-induced tip cell elongation and directed migration. In addition, EC-MC interplay might positively regulate the EC elongation and directional motility, implying a novel role of MCs in the early stage of angiogenesis.
The ‘cell-mixing’ phenomenon during angiogenesis was unexpected because angiogenic elongation was assumed to be a quasi-static phenomenon in which tip cells with filopodia lead the way as pioneer cells and a chain of followers serve as stalk cells, which are connected to each other by cell-cell junctions, elongating the sprout via proliferation (Holderfield and Hughes, 2008; De Smet et al., 2009). The present findings indicate that the two distinct cellular phenotypes, tip and stalk, are interchangeable and branches elongate via a mixture of migratory ECs supplied from upstream vascular beds. Accordingly, gene expression patterns characteristic of tip and stalk cells (Phng and Gerhardt, 2009) may depend on positional context within an elongating branch. As for the phenomenon of tip cell overtaking, Jakobsson et al. reported similar findings in both in vitro and in vivo settings (Jakobsson et al., 2010) while we were preparing this manuscript. Taken together with theirs, our findings strongly suggest that cellular behaviors during angiogenesis may be more variable and complex than previously thought.
What is the biological significance of overtaking of tip cell? After one overtaking event, the new tip cell slows and elongation stops after a while, and the cell is then overtaken by the next one, leading us to speculate that the major role of a tip cell may be not only to elongate in itself but also to create a milieu suitable for the next cell. ECs express various matrix metalloproteinases which degrade the surrounding extracellular matrix (ECM) and thereby create space for invasion (van Hinsbergh and Koolwijk, 2008). Furthermore, a recent report (del Toro et al., 2010) on murine retinal neovessels showed that genes involved in ECM remodeling, such as urokinase-plasminogen-activated receptor and nidogen 2, were enriched in the tip cells. Indeed, a tip cell seems to serve as a ‘rail’ for the following new cell, and this machinery may also explain the ‘cell-mixing’ phenomenon observed at stalks and junctions.
Interestingly, some ECs migrated backwards when the branch was elongating forwards. Previous reports have shown that gradients of local VEGF appear to provide tip cell migration directionality in the retina (Stone et al., 1995; Provis et al., 1997; Gerhardt et al., 2003). Because the local VEGF gradient would presumably be canceled with addition of a large amount of VEGF to the culture medium in our experimental system, retrograde EC movement may merely reflect a loss of the VEGF gradient. However, administration of VEGF resulted in increased anterograde EC migration at stalks without any gradient. This suggests that there is an additional force driving movement towards the tip. In fact, the present parametrical analysis showed that inhibition of both Dll4-Notch signaling and EC-MC interaction also affected the directionality of EC migration. Thus, mechanisms determining directionality are complex, and VEGF may function to enhance directional movement independently of its regional gradient. At the moment, whether the retrograde EC migration occurs in in vivo angiogenesis is unknown. Even though our observations may not represent a physiological phenomenon, they may provide clues to the mechanism responsible for directionality.
The present results provide clues as to how the spatiotemporal regulation of molecular interactions collectively affect EC behaviors, resulting in branch elongation. The VEGFR-Dll4-Notch circuit is reported to be crucial for tip cell function (Hellstrom et al., 2007; Suchting et al., 2007). Similarly, in the present study, interrupting VEGF and Dll4-Notch signaling pathways affected tip cell behavior, as analyzed by the ‘elongation drive’ index. More recently, Jakobsson et al. demonstrated that the signal circuit contributes to tip cell selection and the resultant tip cell replacement (Jakobsson et al., 2010). In the present study, however, the frequency of tip overtaking was unchanged regardless of the absence or presence of VEGF or inhibition of Dll4-Notch signaling. Our results imply that another mechanism may drive overtaking of the tip cell. The parametric analysis results suggest γ-secretase-dependent machinery other than the Dll4-Notch pathway and/or EC-MC interplay possibly to control the frequency of tip cell overtaking.
In conjugation with tip cell navigation, migratory behaviors of ECs following the tip cell affected branch elongation as previously assumed (Schmidt et al., 2007; Perryn et al., 2008). VEGF enhanced the velocity and directionality of EC migratory behaviors, resulting in efficient branch elongation, whereas Dll4-Notch signaling exerted inhibitory actions on both. The string of inhibitory actions exerted by the Dll4-Notch pathway may be due to downregulation of VEGFR2 (Ridgway et al., 2006; Williams et al., 2006). Another possibility is that Notch activation might inhibit EC migration via an Id1-dependent mechanism, which potently contributes to angiogenesis (Itoh et al., 2004; Nishiyama et al., 2005). Considering the mosaic expression pattern of Dll4 throughout angiogenic sprouts, Dll4-Notch interactions occur widely between ECs within sprouts, and scattered activation of the Dll4-Notch signal seems to partially explain the heterogeneity of EC movements. Collectively, these results raise the possibility of a yet-to-be defined role of the Dll4-Notch signaling pathway in controlling EC movements at stalks and junctions.
The present study demonstrated that EC-MC interaction might affect collective EC movement driving branch elongation. In particular, the suppressive action on backwards EC movement was marked. MCs are widely accepted as playing a pivotal role in the process of vessel maturation and remodeling at the late stage of angiogenesis. However, the biological significance of MCs in the early stage remained largely unknown, despite MCs having identified in proximity to sprouting ECs during angiogenesis of the developing brain and retina (Bauer et al., 1992; Ozerdem and Stallcup, 2003). Therefore, this observation suggests a novel role for MCs in angiogenesis. The present study further indicated that a γ-secretase-dependent pathway other than the Dll4-Notch pathway might contribute to the EC-MC interplay. γ-Secretase activates the Notch signal via cleavage of its receptor and a jagged 1 ligand is expressed in MCs (Domenga et al., 2004; Liu et al., 2009). γ-Secretase activity also regulates the dynamics of cadherins (Boulton et al., 2008), which mediate morphogenetic movements (Gumbiner, 2005; Theveneau et al., 2010). Thus, MCs may use the jagged 1-Notch pathway and/or the cadherin-mediated cell-adhesion system for the interplay with ECs, resulting in the control of collective EC movements during angiogenesis.
In conclusion, we have developed a time-lapse imaging and quantitative analysis system revealing cell-based mechanisms underlying elongation processes during in vitro angiogenesis. The physiological relevance of ‘cell mixing’, one of our major findings, was confirmed by analysis of in vivo retinal angiogenesis. The biological findings reported herein and the parameters that we defined can be incorporated in an agent-based model to simulate angiogenic cell behavior, which may in turn yield new insights for understanding the sprouting, elongating and branching aspects of angiogenesis. Furthermore, the methodology employed may allow the gaps among molecular, cellular and integrative bases of organic morphogenesis to be bridged.
This work was supported in part by the Global COE Program (Integrative Life Science Based on the Study of Biosignaling Mechanisms) from the Ministry of Education, Culture, Sports, Science and Technology (MEXT, Japan); by grants-in-aid for scientific research from MEXT (Japan) [23111505 to K.N.]; by grants-in-aid for scientific research from the Ministry of Health, Labour and Welfare of Japan [09156294 to H.K.]; and by grants-in-aid for scientific research from Japan Society for the Promotion of Science (JSPS, Japan) [23591099 to K.N., 21390238 to H.K.].
We thank Olympus Corporation for multiple trials on FluoView FV10i, Dr Shin-ichi Nishikawa and Dr Akiyoshi Uemura for antibodies, and Dr Hideyuki Sakoda for the adenovirus. We also thank Yuko Fujisawa, Takahiro Sato, Rieko Asai and Sakura Kushiyama for excellent technical assistance. MTrackJ was developed at the Biomedical Imaging Group Rotterdam of the Erasmus MC-University Medical Center Rotterdam (The Netherlands) by Erik Meijering.
Competing interests statement
The authors declare no competing financial interests.