Single-cell analysis of endothelial morphogenesis in vivo

The proper formation of functional blood vessel networks is essential for vertebrate organogenesis and physiological homeostasis (Chung and Ferrara, 2011). Insufficient or abnormal vessel growth frequently results in organ malfunction and is associated with diseases such as tumor growth and age-related macular degeneration (AMD) (Chung and Ferrara, 2011). During angiogenesis, new vessels assemble from vascular sprouts emerging from previously formed parental blood vessels. Once neighboring sprouts interconnect, in a process called anastomosis, newly formed vessel segments stabilize, lumenize, and form a functional network with directional blood circulation. At the sub-tissue level, different regions of sprouting vessels exhibit unique characteristics, including heterogeneous phenotypes and distinct gene expression signatures (Aird, 2012). Endothelial cell (EC) populations are not homogenous in different vascular beds or between normal physiological and pathological conditions (Aird, 2007a,b; Nolan et al., 2013). Characterization of ECs in different regions of sprouting vessels demonstrates that vascular morphogenesis is a dynamic process with the spatial and temporal differentiation of a variety of coordinated cell behaviors, including directional polarization, collective migration, cell division and lumen formation (Wacker and Gerhardt, 2011). During sprouting, a leading ‘tip cell’ directs migration, while ‘stalk cells’ behind undergo nascent lumen formation. Although these are not defined differential EC fates [tip and stalk cells can and do frequently interchange (Jakobsson et al., 2010; Pelton et al., 2014)], tip and stalk cells have been shown to display differential gene expression and distinct morphological features (del Toro et al., 2010; Gerhardt et al., 2003; Strasser et al., 2010).

Most of the available data on vascular morphology and dynamics are from in vitro and ex vivo experimental models. Technical challenges in high-resolution optical imaging of deep tissues and lack of suitable genetic tools for imaging ECs at single-cell resolution have hampered study of the cell biology of ECs in vivo. This is certainly true for the process of vascular tubulogenesis, a multistep interwoven developmental process involving initiation, expansion and stabilization of the nascent lumen. Recent reports have highlighted a number of different mechanisms potentially involved in the assembly of vascular tubes (Charpentier and Conlon, 2014). Cord hollowing with cellular rearrangements or cell hollowing involving fusion of intracellular vacuoles have both been proposed as mechanisms for lumen formation based on observations from in vitro cell culture, tissue sections, and live imaging of animal models (Folkman and Haudenschild, 1980b; Iruela-Arispe and Beitel, 2013; Lubarsky and Krasnow, 2003). Adhesion to surrounding extracellular matrix (ECM), loss of some EC-EC contacts, or repulsive interaction between EC surfaces have all been suggested as important forces promoting expansion of nascent lumenal spaces (Eilken and Adams, 2010; Lampugnani et al., 2010; Strilic et al., 2010,, 2009; Wang et al., 2010; Zovein et al., 2010). Other work, primarily in vitro, has uncovered molecular regulators crucial for lumenogenesis (Bayless and Davis, 2002; Bayless et al., 2000; Egginton and Gerritsen, 2003; Iruela-Arispe and Davis, 2009; Koh et al., 2009,, 2008; Popson et al., 2014; Sacharidou et al., 2010,, 2012).

The zebrafish has recently emerged as an important model organism for studying the growth and morphogenesis of blood vessels during development (Butler et al., 2011; Gore et al., 2012; Isogai et al., 2003; Kamei et al., 2006; Zovein et al., 2010). A variety of recent studies have taken advantage of the optical clarity of zebrafish embryos and larvae to carry out live imaging of tubulogenesis in developing vessels within living animals (Ellertsdottir et al., 2010; Herbert and Stainier, 2011; Herwig et al., 2011; Jin et al., 2005,, 2007; Kalen et al., 2009; Kamei et al., 2006; Lenard et al., 2013; Wang et al., 2010). These studies have provided in vivo evidence substantiating both the cell hollowing and the cord hollowing models for lumenogenesis. However, difficulties in distinguishing closely apposed ECs using available transgenic tools have made it difficult to comprehensively and accurately assess the contributions of individual ECs to lumen assembly and morphogenesis of vessels in general. In order to monitor the morphology and dynamic behaviors of individual ECs, we have developed new transgenic tools that simultaneously label both the nuclei and the plasma membranes or tight junctions of single ECs with different fluorescent proteins. Using high-speed confocal and two-photon imaging of these transgenes in injected mosaics, we are able to identify individual ECs and image and parse some of the complex cellular and subcellular dynamics of individual cells that contribute to vessel sprouting and lumen formation.

The trunk intersegmental vessels (ISVs) of the developing zebrafish provide a valuable model for investigating sprouting angiogenesis (Isogai et al., 2003; Lawson and Weinstein, 2002). The relatively simple cell composition, repeating pattern and stereotypical assembly (supplementary material Fig. S1) of these vessels make them ideal for examining in vivo cellular behaviors during sprouting angiogenesis and lumenization (Fig. 1A,B). However, available transgenic tools do not permit reliable identification of single ECs and tracking and imaging of their morphology and behavior, making it difficult to assess how individual ECs contribute to the assembly of the functional vessels. ECs migrate collectively and are extensively intertwined with one another in vascular sprouts. They also display highly dynamic behaviors in growing vessel segments.

We used two-photon time-lapse imaging to examine the movements of entire populations of ECs in developing ISVs in Tg(fli1a:nls-egfp)^y7; Tg(kdrl:mcherry-caax)^y171 double-transgenic animals with green fluorescent EC nuclei and red fluorescent EC membranes (Fig. 1C,D). ECs in growing ISVs are highly dynamic, undergoing cell division, passing one another, and exchanging positions within extending vascular sprouts (Fig. 1C,D; supplementary material Movie 1). As visualized in the Tg(fli1a:egfp-F)^y288 transgenic line, ISV sprouts also extend numerous protrusions that appear and disappear, and the overall morphology of the extending ISV sprouts can change dramatically from one time point to the next (supplementary material Fig. S2A). Using these and other available transgenic lines it is not possible to accurately distinguish boundaries between ECs in growing vessel segments, or to assess the morphology and dynamics of individual ECs. By creating mosaics by injecting Tg(fli1a:egfp-F)^y288 endothelial expression constructs into Tg(kdrl:mcherry-caax)^y171 germline transgenic zebrafish, it is possible to image subportions of the developing vasculature in isolation (supplementary material Fig. S2B, arrows). However, it is impossible to determine with certainty whether these represent individual ECs or multiple adjacent cells. In order to parse complicated, interwoven EC behaviors we designed a novel transgenic approach to identify and analyze the morphology of individual ECs.

To investigate the detailed architecture and dynamic behaviors of individual cells within developing blood vessels, we developed transgene constructs to reliably mark single ECs. We prepared Tol2 transgene constructs in which the fli1a endothelial promoter (Lawson and Weinstein, 2002) drives expression of H2B-TagBFP (nuclear-localized blue fluorescence) followed by eGFP-farnesyl (membrane-localized green fluorescence), linked together by the self-cleaving p2A peptide (Kim et al., 2011) permitting stoichiometric co-translation of both fluorescent proteins (Fig. 2A). This Tol2(fli1a:h2b-tagBFP-2A-egfp-f) transgene is used to generate mosaics by injecting it into Tg(kdrl:mRFP-F)^y286 germline transgenic animals, allowing simultaneous observation of the nucleus (blue fluorescence) and membranes (green fluorescence) of the same EC and, therefore, conclusive determination of the number of ECs being observed and the identification of single cells, all in the context of the complete vasculature (red fluorescence) (Fig. 2B).

ECs expressing the injected transgene (blue/green/red) do not show any evidence of growth delay or altered morphology or behavior compared with adjacent non-transgene-expressing ECs (red only), and injected animals are otherwise indistinguishable from their uninjected siblings (data not shown). The membrane-tethered eGFP-farnesyl protein permits clear identification of cell membranes and fine membrane-based subcellular structures, such as intracellular vesicles and extending processes (Fig. 2B). Thus, this approach allows us to investigate the morphology and dynamic behaviors of single ECs during vascular development in the living animal.

We used single-cell analysis to examine the morphology of ECs contributing to ISVs during their initial assembly in the zebrafish trunk. The trunk vessel network forms in a stereotypic fashion (Isogai et al., 2003). Angiogenic sprouts emerge from the dorsal aorta, extending dorsally along the intersomitic boundaries until they reach the dorsal-lateral surface of the neural tube (supplementary material Fig. S1) (Isogai et al., 2003). Here, they branch rostrally and caudally and fuse with similar vascular sprouts from adjacent ISV segments to form continuous dorsal longitudinal anastomotic vessels (DLAVs).

We generated mosaics containing identifiable single ECs by injecting the Tol2(fli1a:h2b-tagBFP-2A-egfp-f) transgene into Tg(kdrl:mRFP-F)^y286 germline transgenic animals as noted above. By examining a large number of individual ECs within the ISV at stages just post-DLAV interconnection, as assembly of the ISV/DLAV network is completed and lumenization is beginning, we are able to identify a number of different morphological classes (Fig. 3). Half of the ECs are found exclusively within the dorsally extending ISV segments, with a limited number extending the entire length of the ISV (Class I, 4.9%; Fig. 3A) but most extending only part way along the ISV (Class II, 45.1%; Fig. 3B). Other cells are found exclusively within the DLAV (Class III, 11.7%; Fig. 3C) or in an ‘L’ shape bridging the DLAV and ISV (Class IV, 16.7%; Fig. 3D) or bridging the ISV and dorsal aorta (Class V, 21.6%; Fig. 3E). The five classes and their relative proportions are summarized in Fig. 3F,G.

Higher magnification imaging of ECs within newly formed ISVs shows that they are often extensively interdigitated with one another, frequently extending fine processes along the vessel well beyond where the nucleus and main body of the cell are situated (Fig. 3H,I). This shows that cell-cell contacts between ECs in the trunk vascular network are complex, making it difficult to accurately discern EC boundaries and morphology without methods that reliably permit the identification and visualization of single ECs.

Vessel sprouting involves the collective migration of a cohesive group of asymmetrically polarized ECs, with more and longer protrusions at their leading ends and fewer and shorter protrusions at their trailing ends. Although a recent study reported that partial inhibition of actin polymerization and EC protrusion does not limit ISV growth (Phng et al., 2013), the correlation between protrusive activity and directional migration of ECs in vivo remains largely unexplored and difficult to measure quantitatively. We used our single-cell imaging tools to investigate spatial and temporal morphodynamics of endothelial tip and stalk cell protrusions in growing ISVs of either normal animals or those defective in trunk ISV guidance and assembly due to loss of plexin D1 (plxnd1).

In a previous study, we showed that growing ISV sprouts in zebrafish lacking the endothelial-specific semaphorin receptor Plxnd1 are unable to respond to somite-derived semaphorin guidance cues that normally restrict their growth to intersomitic boundaries, resulting in loss of proper directionality in their growth and the formation of disorganized trunk ISV networks (Torres-Vazquez et al., 2004) (Fig. 4A-D). Since plxnd1 morpholino-injected animals exhibit identical phenotypes to plxnd1^fov01b mutants, without additional off-target effects (Torres-Vazquez et al., 2004), we used morpholino knockdown to examine and quantitate the morphology and dynamics of verified individual plxnd1-deficient Tol2(fli1a:h2b-tagBFP-2A-egfp-f) ECs in Tg(kdrl:mRFP-F)^y286 germline transgenic animals (Fig. 4E,F). Visual inspection suggested that ECs indeed display loss of directional growth, but to determine whether this could be validated using objective metrics and to gain insights into the relationship between endothelial protrusiveness and loss of directionality, we performed a rigorous quantitative analysis of single EC protrusions. For each of six individual randomly chosen trunk ECs found above the level of the dorsal aorta, we measured the length of all of their protrusions (filopodia). The average total length of endothelial protrusions was similar between control and plxnd1 morphant animals (Fig. 4G). To examine the directionality of these protrusions, we grouped the direction of measured protrusions into three categories depending on whether the filopodia were oriented in a dorsal (120° arc), lateral (60° arcs, ‘side’) or ventral (120° arc) direction (Fig. 4H). We measured protrusions in tip cells and stalk cells separately in both control and plxnd1 morphants (Fig. 4I,J). There was a small but statistically significant increase in side and ventral protrusiveness along with a less significant decrease in dorsal protrusiveness in the tip cells of plxnd1 morphants compared with controls (Fig. 4I). For stalk cells, plxnd1 morphants showed more substantial increases in protrusiveness compared with controls, particularly dorsal and ventral protrusions (Fig. 4J). The lack of change in average protrusion length (Fig. 4G) and changes in the distribution of protrusions (Fig. 4I,J) together suggest that the spatial distribution and not total amount of protrusiveness is altered in plxnd1-deficient endothelial tip cells. The more pronounced effect on stalk cells compared with tip cells also hints that reduced EC-EC contact inhibition could be a contributing factor to the increased sprouting and branching of trunk vessels observed in the plxnd1-deficient animals.

Endothelial membrane protrusions are not static structures, but transient membranous extensions with highly dynamic behaviors. We used time-lapse two-photon imaging of Tol2(fli1a:h2b-tagBFP-2A-egfp-f)-injected Tg(kdrl:mRFP-F)^y286 animals to capture and evaluate the temporal dynamics of filopodia in single plxnd1-deficient or control trunk ECs (Fig. 4K,L). Time-lapse recording revealed that membrane protrusions have a significantly longer average lifetime in plxnd1-deficient than in control ECs (Fig. 4M,N). Longer protrusive lifetimes were observed in both tip (Fig. 4M) and stalk (Fig. 4N) cells.

As noted in the introduction, recent studies have suggested a number of possible mechanisms contributing to vascular lumen formation (Charpentier and Conlon, 2014). Lumen formation is thought to occur mainly by either cell hollowing (the formation of intracellular lumenal spaces by fusion of intracellular vacuoles) or cord hollowing (the generation of intercellular lumenal spaces by the rearrangement and creation of spaces between multiple ECs), although the extent to which lumen formation initiates intracellularly versus between cells is still not clear. As noted above, transgenic and other tools used to date for analysis of lumen formation in vivo have not permitted conclusive determination of how many cells are being examined, and therefore of where one cell ends and another cell begins, complicating interpretation.

As a first attempt to address these limitations, we began by using our single-cell analysis methods to determine the proportion of ECs in ISVs immediately after initial lumenization that contain enclosed lumens (i.e. lumenal spaces completely surrounded by a single EC). We captured images of single Tol2(fli1a:h2b-tagBFP-2A-egfp-f) ECs in Tg(kdrl:mRFP-F)^y286 germline transgenic animals and then carried out 3D image reconstruction using active contour-based image segmentation to examine the cross-sectional profiles of the vessels that the ECs were incorporated in, to determine what proportion of each identified single EC was entirely enclosing the lumen (Fig. 5A). Frequently, cross-sectional profiles of identified single ECs revealed that the lumenal spaces adjacent to the ‘ends’ of the cell were surrounded by multiple cells (Fig. 5B; see cross-sections 1 and 4), as might be expected given the extensive interdigitation of ECs in assembling trunk vascular networks noted earlier, whereas the lumen adjacent to the middle of an EC was entirely enclosed by that single cell (Fig. 5B; see cross-sections 2 and 3). Lumen-enclosing ECs are detected in all positions in the ISV (Fig. 5C,D). Approximately one-third of all ISV/DLAV ECs are lumen enclosing (Fig. 5D, ‘Sum’), with substantially more lumen-enclosing ECs in the DLAV (Fig. 5D, Classes III and IV) and fewer in the more ventral ISV segments (Fig. 5D, Classes II and V). Of the lumen-enclosing cells, on average one-third of the length of each of these cells completely encircles the lumen at the time of initial lumenization (supplementary material Fig. S3), with additional cells abutting the lumen at the ends (see cross-sections 1 and 4 in Fig. 5B).

These observations suggest the coexistence of both unicellular and multicellular lumens in newly formed trunk ISVs. However, as discussed further below, the presence of unicellular lumens does not necessarily imply that these form ‘seamlessly’.

Intracellular lumen formation would be expected to not only result in unicellular lumenal spaces, but also to form them transcellularly or ‘seamlessly’ without an autocellular junction (Fig. 6Ai). Autocellular junctions could result from lumen formation by single ECs ‘wrapping’ on themselves (Fig. 6Aii). The membrane-tethered fluorescent proteins used in the transgenes described thus far (egfp-f, mRFP-F) outline boundaries between cells but do not permit direct conclusive visualization of cell-cell or autocellular junctions. To address this issue, we developed a genetic approach to label EC junctions in vivo using fluorescent proteins. A fusion protein construct was generated (Fig. 6B) using the coding sequences for the endothelial-specific tight junction protein Claudin 5b and eGFP. This construct was placed under the control of the endothelial fli1a promoter in a Tol2 vector, and stable double germline transgenic Tg(fli1a:egfp-claudin 5b)^y287; Tg(kdrl:mRFP-F)^y286 animals were generated. The Tg(fli1a:egfp-claudin 5b)^y287; Tg(kdrl:mRFP-F)^y286 double-transgenic animals, and the vessels within them, develop indistinguishably from those of Tg(kdrl:mRFP-F)^y286 control animals. The eGFP-Claudin 5b fusion protein colocalizes with endogenous ZO-1 (Tjp1a – ZFIN), as shown by co-immunostaining with anti-eGFP and anti-ZO-1 antibodies (supplementary material Fig. S4), validating that this transgenic line permits detailed visualization of EC tight junction complexes in living animals (Fig. 6C,D).

We carried out extensive confocal and two-photon microscopy using this line to examine the junctional morphology of the newly assembled ISVs and DLAVs in the trunk of zebrafish embryos. We were able to identify a variety of junctional types suggestive of different types of EC-EC contacts in these vessels (Fig. 6D-G; supplementary material Movies 2-4). Interestingly, we found that approximately one-quarter of the length of all vessel segments lacked any obvious junctions (Fig. 6E,H,I; supplementary material Movie 2). The proportion of vessel length lacking eGFP-Claudin 5b junctional localization was greater in the DLAV (34.9±7.7%) than in the ascending ISV (17.9±5.6%). These ‘gaps’ in junctional localization along the vessels were not a result of lack of transgene expression in selected cells, as the gaps were also observed in mosaics generated by injecting a Tol2(kdrl:mcherry-p2A-egfp-claudin 5b) transgene simultaneously labeling EC bodies and tight junctions. In these animals approximately one-quarter of the length of transgene-labeled ECs showed ‘dark’ gaps in eGFP-Claudin 5b fusion protein localization despite expressing mCherry throughout the cell (Fig. 6J-M; supplementary material Movie 5). 3D examination of eGFP-Claudin 5b patterns in these mosaics also revealed the complexity of endothelial junctional morphology and the difficulty in interpreting morphology from static single-view images (compare supplementary material Movie 5 and Fig. 6M). Tight junction-free vessel segments are also detected by endogenous ZO-1 staining in the perfused ISV and DLAV (supplementary material Fig. S4). Tight junction-free vessel segments in Tg(fli1a:egfp-claudin 5b)^y287; Tg(kdrl:mRFP-F)^y286 double-transgenic animals persist for at least a day after ISV lumenization (supplementary material Fig. S5A-C), showing that these are not simply transient intermediates during vessel formation, although the overall length of these segments does decrease over time (supplementary material Fig. S5D,E).

Together, our results suggest that, at least for small-caliber early trunk vessels (ISV, DLAV), there are multiple types of tubular lumens and cell junction distributions, with lumens either hollowed inside the cell, enclosed by a single EC, or sealed by multiple ECs (Fig. 6A). These observations, in concert with the results from single EC 3D configuration analysis, suggest that multiple mechanisms of lumen formation coexist in the early developing vasculature.

Our laboratory and others have previously reported intracellular vacuoles contributing to lumen formation in vitro and in vivo (Bayless and Davis, 2002; Davis et al., 2011; Folkman and Haudenschild, 1980a; Kamei et al., 2006; Koh et al., 2009; Sacharidou et al., 2012). However, other recent work has highlighted the importance of cellular rearrangements and multicellular/extracellular lumenization (Wacker and Gerhardt, 2011), and the existence of intracellular vacuoles or nascent lumenal compartments has been questioned. This is in part because of the challenges in imaging these compartments in vivo and the difficulty of discerning whether these spaces are found within single ECs or between multiple ECs.

In order to examine whether intracellular vacuoles are present during ISV/DLAV assembly, we used two-photon imaging to visualize subcellular compartments in single Tol2(fli1a:h2b-tagBFP-2A-egfp-f) ECs in Tg(kdrl:mRFP-F)^y286 animals. Using the h2b-tagBFP tag to verify that we were indeed examining single ECs, we were able to observe eGFP-farnesyl-labeled intracellular membrane structures of varying sizes that appeared to be entirely enclosed within the cytoplasm (Fig. 7A,B; supplementary material Movies 6 and 7). These were observed frequently within individual ECs, especially at the early stages before lumen formation. Using time-lapse imaging we were also able to observe fusion of multiple smaller vesicles to generate larger vacuolar compartments within single ECs (Fig. 7C,D; supplementary material Movie 8). We were also able to image the emergence of a vacuolar compartment that appeared to be enclosed within a single EC and its subsequent enlargement and eventual fusion with the nascent adjacent extracellular space (supplementary material Fig. S6, Movie 9). Intracellular vesicular/vacuolar compartments were observed more frequently in the unlumenized tip cells than in lumen-forming stalk cells.

Taken together with our results showing an increased proportion of seamless vessel segments in the newly formed DLAV, these findings suggest that intracellular membrane-bound compartments may contribute to the initial lumenization of angiogenic trunk vessels.

Developmental angiogenesis is a tightly controlled process that involves complex and diverse endothelial cellular behaviors. Over the past decade, a great deal of effort has been focused on understanding the molecular mechanisms regulating angiogenesis, but most of this work has been carried out using in vitro EC culture methods or by studying vessels in vivo at the tissue level. Our understanding of the cellular mechanisms regulating angiogenesis in vivo remains incomplete, at least in part due to the limitations of live imaging of ECs in vivo and lack of suitable genetic tools for analyzing the morphology and behavior of individual ECs in the context of the entire vasculature. Although studies of vessel formation at the organ and tissue levels have provided important insights into the underlying genetic basis for embryonic vascular development and organogenesis, endothelial morphology and/or dynamics have mostly been analyzed in groups of ECs or whole growing vessels rather than in individual ECs, which generally cannot be definitively distinguished from their neighbors.

In this study, we describe an approach for labeling and definitively identifying single ECs in vivo. We prepared transgene constructs driving simultaneous expression of a blue fluorescent tag in EC nuclei (H2B-TagBFP), to definitively identify individual cells by their nuclei, and a green fluorescent tag in EC membranes (eGFP-F), to visualize cellular and subcellular morphology. The H2B-TagBFP and eGFP-F proteins are stoichiometrically coexpressed from the same construct by linking them together using a viral 2A peptide sequence (Kim et al., 2011) permitting co-translational expression of both proteins (Provost et al., 2007; Wang et al., 2011). Injection of the Tol2(fli1a:h2b-tagBFP-2A-egfp-f) transgene into Tg(kdrl:mRFP-F)^y286 germline transgenic zebrafish results in mosaic coexpression of H2B-TagBFP and eGFP-F in a subset of ECs within globally mRFP-F-positive vasculature. By examining the nuclear H2B-TagBFP and cellular eGFP-F expression in the ECs of mosaic injected animals, it is straightforward to identify individually tagged ECs for morphological and behavioral characterization that have no immediately adjacent tagged neighbors. This makes it possible to examine the morphology and behavior of individual ECs within the context of multicellular vascular cords or tubes.

As a proof-of-principle to demonstrate the usefulness of this method for qualitative and quantitative analysis of differences in EC morphology and behavior between normal and defective vasculature, we carried out single EC analysis in Plxnd1-deficient embryos. In previous studies, we and others have shown that growing ISVs lacking the endothelial-specific semaphorin receptor Plxnd1 due to genetic mutation or knockdown are unable to respond to somite-derived semaphorin guidance cues that normally restrict their growth to intersomitic boundaries, resulting in loss of proper directionality in their growth and the formation of disorganized trunk vascular networks (Gitler et al., 2004; Gu et al., 2005; Torres-Vazquez et al., 2004). By measuring the amount and dynamics of membrane protrusions in individual trunk ECs we were able to show that, although the total length of endothelial protrusions is similar in control and plxnd1 morpholino-injected animals, the directionality of protrusions is altered. Our results also revealed a prolonged persistence of endothelial membrane protrusions in plxnd1 morphants compared with controls, as might be expected given the inability of these ECs to respond to repulsive semaphorin cues produced by the adjacent somites. A similar prolongation of protrusions has been noted in axonal growth cones in the absence of semaphorin-plexin repulsion (Ayoob et al., 2004; Chauvet et al., 2007).

Vascular lumen formation is a complicated morphogenetic process involving the structural transition from a cord of ECs to a hollow tube. As discussed in detail in the introduction, recent reports have highlighted a number of different mechanisms potentially involved in endothelial tubulogenesis (Charpentier and Conlon, 2014). It has been proposed that vascular tubes form via either cord hollowing, involving cellular rearrangements, or cell hollowing, involving the formation and fusion of intracellular vacuoles (Charpentier and Conlon, 2014; Ellertsdottir et al., 2010; Kamei et al., 2006; Lampugnani et al., 2010; Lubarsky and Krasnow, 2003; Strilic et al., 2009; Wang et al., 2010). Mechanisms such as adhesion to surrounding ECM, loss of EC-EC contacts, repulsive interaction between EC surfaces, and the formation of intracellular vacuoles and their intra- or intercellular fusion have all been suggested as important forces promoting the expansion of nascent lumenal spaces. The cellular basis for endothelial tube formation in vivo has been difficult to study due to the challenge of observing individual EC behaviors within the vasculature. Our single-cell analysis in intact live animals suggests that endothelial tube formation occurs in a heterogeneous manner, using a combination of different cellular mechanisms (Fig. 8). We observe both single-cell-enclosed lumens and lumens lined by multiple ECs within newly formed tubes in the trunk vascular network. Live imaging of EC junctions reveals that many lumenal segments enclosed by single ECs lack apparent autocellular junctions, suggesting that these are seamless vascular tubes formed, at least in part, by a cell hollowing mechanism. The existence of seamless vascular tubes in mammals has also been suggested from electron microscopy analysis of capillary vessels from various vascular beds (Bar et al., 1984) and by VE-cadherin staining of the mouse retinal vasculature (Ehling et al., 2013; Fan et al., 2014). We also document the existence of small vesicles and larger vacuoles within individual ECs of angiogenic ISVs, although additional work will be needed to definitively conclude whether these vacuole-like structures fuse and enlarge intracellularly, directly form extracellular spaces between adjacent ECs, or a combination of both. Taken together, our observations and those of many other groups using a variety of different approaches strongly suggest that angiogenic sprouts, especially those forming small-caliber vessels such as capillaries, become lumenized through multiple distinct cellular mechanisms including both cord hollowing and cell hollowing. The coexistence of multiple cellular mechanisms for lumen formation might be advantageous for endothelial adaptation to the local challenges of distinct microenvironments.

EC abnormalities in form and function are common features of many conditions involving pathological angiogenesis, notably cancer (Baluk et al., 2005). Although there has been enormous progress recently in identifying the genetic and epigenetic factors that contribute to changes in blood vessel form and function at the tissue level, a thorough investigation of the underlying in vivo cellular mechanisms at high resolution is needed to understand how molecular alterations in ECs translate into changes in the vasculature at the tissue level. The new tools that we describe here should be useful in helping to bridge this knowledge gap.

Zebrafish were maintained and zebrafish experiments were performed according to standard protocols (Westerfield, 2000) and in conformity with the Guide for the Care and Use of Laboratory Animals of National Institutes of Health in an Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) accredited facility. Wild-type EK strain was obtained through natural spawning from Ekkwill (Ruskin, FL, USA). All transgenic lines were maintained in the EK strain. Transgenic lines Tg(fli1a:nls-egfp)^y7 and Tg(kdrl:mcherry-caax)^y171 were published previously (Fujita et al., 2011; Siekmann and Lawson, 2007).

For plasmid Tol2fli1a:H2B-TagBFP-p2A-eGFP-F, TagBFP-tagged human histone H2B was purchased from Evrogen (pTagBFP-H2B, #FP176); porcine 2A viral peptide (DNA sequence: 5′-GGAAGCGGAGCTACTAACTTCAGCCTGCTGAAGCAGGCTGGAGACGTGGAGGAGAACCCTGGACCT-3′) was fused between TagBFP-H2B and eGFP-farnesyl sequences; then H2B-TagBFP-p2A-eGFP-F was subcloned into miniTol2-containing vector harboring the endothelial promoter fli1a.

The zebrafish claudin 5b gene (NCBI accession NM_001006044) was amplified from a cDNA library from 48 hours post-fertilization (hpf) zebrafish embryos. claudin 5b was subcloned into pEGFP-C1 (Kamei et al., 2006) to generate eGFP-Claudin 5b. eGFP-Claudin 5b was digested by NheI and MluI and subcloned into miniTol2-containing vector harboring the endothelial promoter fli1a. For plasmid (I-SceI)kdrl:mCherry-p2A-eGFP-Claudin 5b, the constructs p5′E:mCherry-p2A, pME:eGFP-Claudin 5b and p3′E:polyA were cloned into vector pDEST:kdrl (Fujita et al., 2011) containing meganuclease I-SceI sequences using a MultiSite Gateway Construction Kit (Invitrogen).

Morpholinos targeting plxnd1 (5′-TGAGGGTATTTACAGTCGCTCCGC-3′) were injected at the one- to four-cell stage at a dose of 5 ng (Torres-Vazquez et al., 2004).

For mosaic single-cell analysis, 1 nl solution containing 25 ng/μl Tol2fli1a:H2B-TagBFP-p2A-eGFP-F DNA plasmid and 30 ng/μl miniTol2 RNA was injected into Tg(kdrl:mRFP-F)^y286 transgenic embryos at the one-cell stage. Injected embryos with membrane GFP fluorescence in the developing vasculature were selected for single EC morphology analysis under two-photon confocal microscopy. For stable transgenic germline generation, Tol2fli1a:eGFP-Claudin 5b or ISceIkdrl:mRFP-F DNA plasmid was co-injected with miniTol2 RNA or I-SceI meganuclease protein into wild-type EK embryos at the one-cell stage. Embryos with fluorescent blood vessels were selected as F0 founders and maintained for descendant generations.

Live embryos were mounted in 0.7% low-melting agarose gel and cultured in embryo water for confocal imaging using a long working distance objective (20×, water-immersion, 507701, Leica). Confocal images were acquired with a FluoView 1000 (Olympus) or TCS SP5II (Leica) confocal microscope equipped with an imaging chamber mounted with MatTek 60 mm glass-bottom dishes. z-series stacks were captured with 0.25 μm spacing between image planes. The active contour-based image segmentation and 3D reconstruction were performed using Volocity 6.0 (PerkinElmer) and Imaris 7.4 (Bitplane) software. 3D rotation movies were constructed using ImageJ (NIH) and MAYA (Autodesk). Lasers used in this study: 850 nm 2P for TagBFP, 488 nm for eGFP, 561 nm argon laser for mRFP and mCherry.

Statistical analysis was performed using Prism 6.0 (GraphPad). Box-and-whisker with minimum and maximum and geometric plots with mean±s.d. were generated from Prism graphs. The difference among indicated groups was evaluated by unpaired two-tailed Student's t-test. P<0.05 was considered as statistically significant between test groups. Polar graphs for filopodia protrusiveness and spatial distribution were generated by open-source software Gnumeric 1.12.16 (www.gnumeric.org).

We thank members of the B.M.W. laboratory for technical help and critical suggestions; Dr Roberto Weigert for providing XFP-F constructs; Ms Jingjing Wang for help in generating transgenic fish; and Jeremy Swan and Sam Farabaugh for assistance with image processing.