Endocardium is necessary for cardiomyocyte movement during heart tube assembly

and Embryonic heart formation requires the union of bilateral populations of cardiomyocytes and their reorganization into a simple tube. Little is known about the morphogenetic mechanisms that coordinate assembly of the heart tube and determine its dimensions. Using time-lapse confocal microscopy to track individual cardiomyocyte movements in the zebrafish embryo, we identify two morphologically and genetically separable phases of cell movement that coordinate heart tube assembly. First, all cardiomyocytes undergo coherent medial movement. Next, peripherally located cardiomyocytes change their direction of movement, angling toward the endocardial precursors and thereby establishing the initial circumference of the nascent heart tube. These two phases of cardiomyocyte behavior are independently regulated. Furthermore, we find that myocardial-endocardial interactions influence the second phase by regulating the induction, direction and duration of cardiomyocyte movement. Thus, the endocardium plays a crucial early role in cardiac morphogenesis, organizing cardiomyocytes into a configuration appropriate for heart tube assembly. Together, our data reveal a dynamic cellular mechanism by which tissue interactions establish organ architecture.


INTRODUCTION
Organogenesis requires the precise arrangement of numerous cells into a specific three-dimensional architecture that is essential for effective organ function. Cardiogenesis, for example, involves a series of morphogenetic steps that must be carefully orchestrated in order to insure the proper dimensions of the heart tube. In all vertebrate embryos, heart tube assembly begins with the medial movement of bilateral groups of cardiomyocytes (Glickman and Yelon, 2002;Harvey, 2002;Moorman and Christoffels, 2003). Once they reach the embryonic midline, the bilateral populations merge through a process called cardiac fusion. Cardiac fusion produces an intermediate structure, referred to as a cardiac crescent in amniotes (Harvey, 2002;Moorman and Christoffels, 2003) and a cardiac cone in zebrafish (Glickman and Yelon, 2002), that gradually transforms into a primitive heart tube. This simple cylinder immediately becomes responsible for driving embryonic circulation; therefore, the precise coordination of heart tube assembly has significant consequences for embryonic physiology.
Previous studies of heart tube assembly in the zebrafish embryo have provided an overview of this dynamic process. Cardiac fusion initiates with contacts between posterior subsets of contralateral cells, followed by interactions between anterior subsets of contralateral cells (Glickman and Yelon, 2002;Yelon et al., 1999) (Fig. 1A). These connections create a ring of cardiomyocytes surrounding the centrally located precursors of the vascular endocardium (Stainier et al., 1993;Trinh and Stainier, 2004) (Fig.  1A). Next, the myocardial ring, which is also called the cardiac cone, converts into a tube (Glickman and Yelon, 2002;Stainier et al., 1993;Yelon et al., 1999). Through gradual extension of the length of its axis, the once shallow cone becomes an elongated cylinder, and the inner and outer circumferences of the cone become the arterial and venous apertures of the nascent heart tube (Glickman and Yelon, 2002;Stainier et al., 1993;Yelon et al., 1999). At the same time, the endocardial precursors spread out to create the endothelial lining of the myocardial cylinder (Stainier et al., 1993). Thus, the cardiac cone provides an essential foundation for the dimensions and composition of the zebrafish heart tube.
Despite the importance of creating a strong foundation for heart tube formation, the morphogenetic mechanisms that regulate cardiomyocyte behavior during this process remain relatively undefined. Work in multiple model organisms has indicated that interactions between the myocardium and neighboring gut endoderm are crucial for the recruitment of cardiomyocytes toward the embryonic midline (e.g. Dickmeis et al., 2001;Kikuchi et al., 2001;Li et al., 2004;Narita et al., 1997). Medial recruitment also appears to require the organization of cardiomyocytes into polarized epithelia. During cardiac fusion, cardiomyocytes begin to form the junctional complexes typical of epithelia and exhibit molecular characteristics of apicobasal polarity (Trinh and Stainier, 2004). Mutations that interfere with epithelialization, such as natter (also known as fibronectin 1), prevent cardiomyocytes from moving properly toward the midline (Trinh and Stainier, 2004). However, it is not clear whether the proximity of the contralateral myocardial populations at the midline is sufficient to coordinate their integration into a heart tube. Does tube construction simply result from a continuation of the initial medial movement of cardiomyocytes, or does it require a distinct mode of cell behavior subject to independent regulation?
To distinguish between these models, we have used time-lapse imaging to resolve the direction, rate and organization of individual cardiomyocyte movements during cardiac fusion in wild-type and mutant zebrafish embryos. Our data indicate that two morphologically and genetically separable phases of cardiomyocyte movement underlie cardiac fusion: a first phase that recruits cells toward the midline and a second phase that creates the morphology of the cardiac cone. Additionally, we find that cardiomyocyte Endocardium is necessary for cardiomyocyte movement during heart tube assembly movement during the second phase is regulated by the presence of the endocardium, thereby revealing a previously unappreciated role for myocardial-endocardial interactions during heart tube assembly.

Embryo collection and preparation
Wild-type fish and heterozygotes for the mutations clo s5 and mil te273 (Chen et al., 1996;Qian et al., 2005) carrying Tg(cmlc2:egfp) (Huang et al., 2003) were mated to generate experimental embryos. Embryos were screened for robust egfp expression using a Zeiss M2Bio microscope. For mil mutants, the genotype was assessed prior to time-lapse imaging: these mutants exhibit a clear cardia bifida phenotype at the 18-somite stage (Yelon et al., 1999). Selected embryos were mounted in a drop of 3% methylcellulose within a larger drop of 2% low-melting point agarose in embryo medium. Mounting slides were constructed by placing a single layer of electrical tape on a standard microscope slide and cutting a small square out of the center to create a well. Embryos were oriented to permit a dorsal view of the heart fields, using the tip of the notochord as a guide. Finally, a drop of embryo medium with tricaine was placed on top of each agar-mounted embryo.

Data collection and processing
Embryos were maintained at 28.5°C during data collection, using an upright Zeiss LSM510 confocal microscope equipped with a heated stage, a 25ϫ Plan-Neofluor multi-immersion objective and LSM510 software. In order to generate 3D reconstructions, 30-40 focal planes were collected at intervals of 1.5-2.5 m, a distance sufficient to accommodate sample movement during the course of the time-lapse. Images were collected at 2-3 minute intervals. Typically, time-lapses began around the 15-to 18-somite stage and ended at the completion of fusion, around the 20-to 21-somite stage. Time of initial imaging was often dependent upon GFP intensity, which varies at early stages. After imaging, embryos were coaxed from their mounting agar and transferred to fresh embryo medium. Embryonic stage was determined both before and after mounting; manipulation and imaging did not delay development. Additionally, embryonic health and genotype were assessed at 24 hours post-fertilization (hpf). Only data collected from embryos that appeared completely normal at 24 hpf were analyzed.
For 4D analysis, LSM510 databases were imported into a Volocity (Improvision) library. An image sequence composed of a linear array of 3Drendered time points was then generated for each data set. As appropriate, image appearance (brightness, contrast, opacity) was modified to optimize visualization of cells of interest. Optimization of image appearance was a key step: GFP intensity varies widely among individual cardiomyocytes and increases over time, so resolution of each cell required different adjustments of image settings. Image settings shown in Figs 1-4 allow resolution of many, but not all, tracked cells. Images were exported as QTmov files for visualization of a time-lapse from a preset view.

Cell tracking and analysis
Examination of high-resolution confocal images indicated that the myocardium is a relatively flat sheet of cells, approximately 8 m thick, from the 15-to 20-somite stages (N.G.H. and D.Y., unpublished). During these stages, most cardiomyocyte movement is along the x (mediolateral) and y (anteroposterior) axes, rather than along the z (dorsoventral) axis. The z-axis displacement of tracked cardiomyocytes did not exceed one cell diameter over the course of our time-lapses. Therefore, when tracking cardiomyocyte movements, we treated the myocardium as a 2D tissue and focused on tracking the x and y coordinates of the center of each cell. At time points following our window of analysis, we have observed significant dorsal cardiomyocyte movements as the cardiac cone extends into a tube (N.G.H. and D.Y., unpublished). These dorsal movements are consistent with prior histological observations (Stainier et al., 1993;Trinh and Stainier, 2004), although we have not observed lateral displacement of dorsal cardiomyocytes as suggested by Trinh and Stainier (Trinh and Stainier, 2004).
For each time-lapse, we examined all z-stack images and tracked as many cells as were resolvable in any focal plane throughout the entire period of imaging. Figs 1-4 show reconstructed projections of z-stacks and include a representative subset of tracks from the time-lapses depicted. To track a cell, we selected a pixel at its visual center at each time point, modifying the image intensity as necessary, and recorded each pixel's coordinates using Volocity. Each series of cell locations was then converted to a track using Volocity's 'manual track' function. This function automatically generates a measurement file that includes the distance traveled, velocity and a series of xy positions for each cell. These files were exported into Microsoft Excel for further analysis, including calculation of net direction of movement. To visually depict cell tracks (as in Fig. 1E-I, Fig. 2C-E and Fig. 3C,D) we generated lines connecting each series of cell locations, added arrowheads to indicate the direction of movement, and overlaid these tracks on processed images of individual time points. As a control for accuracy of tracking, all of the cells in one wild-type embryo were tracked twice. The two sets of distances, velocities and directions of movement did not vary by more than 5%.
For quantitative analysis of the net direction of movement for each tracked cell, we calculated the percentage of the y-axis displacement relative to the x-axis displacement in radians, using the formula: (⌬y/⌬x)ϫ100. This can also be expressed as an angle (°) using the formula: atan(⌬y/⌬x)ϫ57.295. The resulting value reflects the cell's degree of displacement along the anterior-posterior axis: a value of 0° corresponds to direct medial movement, and a value of 90° corresponds to direct anterior or posterior movement. Examination of the degree of displacement of wild-type cardiomyocytes during the first phase of cardiac fusion demonstrated that more than 80% of these cells were displaced by less then 30° (Fig. 1J). We therefore defined 'medial movement' as a displacement of less than 30° and 'angular movement' as a displacement of greater than 30°. Some tracked cells exhibited little movement: we defined 'no net displacement' as movement of less than 10% of a cell diameter over the course of a 20-minute period (as in Fig. 1I). Additionally, we defined 'no net directed movement' as failure to move beyond a cell diameter from the starting position over the course of a time-lapse lasting at least 90 minutes (as in Fig. 4C,D). To compare regional differences in the patterns of cardiomyocyte movement, we divided the bilateral cardiomyocyte populations into three regions: anterior, central and posterior. In wild-type embryos, we defined regions retrospectively, based on the positions of individual medial cardiomyocytes at the completion of cardiac fusion (Fig. 1M). Cells located at the anterior or posterior interfaces between contralateral populations were considered to be in anterior or posterior regions, respectively. Cells located between these areas were considered to be in the central region. In clo, mil and mil;clo mutant embryos, we defined regions differently, as cardiac cone morphology is aberrant. In these cases, we divided populations into equivalent thirds along the anterior-posterior axis, based on the number of cells present at the medial edge of the population at the beginning of the time-lapse ( Fig. 2G and Fig. 3F).

In situ hybridization
Single-color whole-mount in situ hybridization for cmlc2 and flk1 was performed according to a standard protocol (Liao et al., 1997;Yelon et al., 1999). Two-color fluorescent whole-mount in situ hybridization was conducted according to a protocol that will be reported separately (J.J.S., B. R. Keegan and D.Y., unpublished). Riboprobes for cmlc2 and fli1a (Thompson et al., 1998) were labeled with dinitrophenol (Mirus) and digoxygenin (Roche) haptens and detected by deposition of Cy3 and fluorescein tyramides (PerkinElmer), respectively.

Tracking cardiomyocyte movement during heart tube assembly
To examine dynamic cell movements during heart tube assembly, we took advantage of a reporter transgene, Tg(cmlc2:egfp), that drives egfp expression in cardiomyocytes (Huang et al., 2003). Tg(cmlc2:egfp) expression mimics endogenous cardiac myosin light chain 2 (cmlc2; myl7 -ZFIN) expression, beginning before cardiomyocytes approach the embryonic midline and continuing throughout development (Huang et al., 2003;Yelon et al., 1999) ( Fig. 1B-D). We visualized cell movements over time by collecting confocal z-stacks at 2-minute intervals and rendering a threedimensional time-lapse. To quantify patterns of cell behavior, we

RESEARCH ARTICLE
Development 134 (12) measured the displacement of individual cells between time points, creating a track representative of each cell's movement over time (detailed imaging and tracking protocols are provided in the Materials and methods). Using this strategy, we resolved the direction, rate and organization of individual cardiomyocyte movements throughout the process of cardiac fusion. Although we focused our analysis on cells at the medial edge of each cardiomyocyte population, we also found that lateral cells exhibit comparable patterns of movement (data not shown).

Coherent medial movement brings cardiomyocytes to the midline
Previous gene expression studies and fate-map analyses in zebrafish, chick and mouse embryos have suggested that cardiomyocytes advance toward the midline in an orderly fashion (Dehaan, 1963;Keegan et al., 2004;Lyons et al., 1995;O'Brien et al., 1993;Stalsberg and DeHaan, 1969;Yelon et al., 1999). In accordance with these previous studies, our time-lapse examination of wild-type embryos demonstrated that cardiac fusion begins with a phase of coherent medial movement that brings the cardiomyocytes into position for the initiation of cardiac cone formation (Fig. 1B,C,E,F and see Movie 1 in the supplementary material). All cardiomyocytes exhibited similar patterns of behavior, following relatively parallel paths of similar length with general maintenance of neighbor relationships (Fig.  1E,F). Examination of cell tracks revealed that most cardiomyocytes move directly toward the midline, exhibiting an angular displacement of less than 30° along the anterior-posterior axis ( Fig. 1J and Table 1). We classify this behavior as 'medial movement' (red arrows/bars), in contrast to 'angular movement' (yellow arrows/bars), which we define as an angular displacement of greater than 30° (Fig. 1E,F,J). Overall, the observed coherence  Table  1 for additional data.
of cardiomyocyte movement is consistent with the status of the migrating myocardium as a maturing, polarized epithelium (Trinh and Stainier, 2004).

Regionally restricted angular movement facilitates cone formation
We observed a striking transition in cardiomyocyte behavior once the contralateral populations initiate contact at the midline. Specifically, we found that many cardiomyocytes switch from medial movement to angular movement (Fig. 1C,D,G,H,L, Table  1 and see Movie 1 in the supplementary material). Most of the cells exhibiting angular movement are located in the anterior or posterior region of each heart field (angular movement observed in 20 of 26 anterior cells and 32 of 37 posterior cells, Fig. 1M and Table 1), with anterior cells angling toward the posterior and posterior cells angling toward the anterior (Fig. 1G,H). By contrast, cells located in the central region of each field tended to continue to move medially (medial movement observed in 45 of 55 central cells, Fig. 1M and Table 1). This pattern of behavior is a significant departure from the first phase of cardiac fusion, during which very few cardiomyocytes in anterior or posterior regions exhibit angular movement (Fig. 1K,M and Table 1). Thus, these regional changes in direction of movement distinguish a second phase of cardiac fusion. As a consequence of regional transitions to angular movement, the bilateral populations of cardiomyocytes formed mirror-image arcs, ultimately encircling the endocardial precursors and creating the cardiac cone (Fig. 1C,D,G,H). Regional differences in the duration of movement also contribute to arc formation. Although all cardiomyocytes moved at comparable rates during the second phase of cardiac fusion (0.36±0.03 m/minute), cells in central regions tended to stop moving earlier than cells in anterior or posterior regions. Examination of cell movements during the final 20 minutes of cone formation (20 minutes before the 20-somite stage, Fig. 1I) revealed that most centrally located cells exhibit no net displacement during this interval [64% (35 of 55) of cells examined; white asterisks]. By contrast, most cells in anterior and posterior regions continued moving during this time interval, with only 33% (21 of 63) of these cells coming to a halt. Together, our data indicate that regionally restricted transitions in the direction and duration of cardiomyocyte movement facilitate cardiac cone formation.

No angular movement in the absence of endocardium
Our time-lapse analyses in wild-type embryos indicated that cardiac fusion involves two morphologically distinct phases of cell behavior. We hypothesized that these two phases could represent separately regulated morphogenetic processes, with endoderm-derived cues driving the initial medial movement (Alexander et al., 1999;Dickmeis et al., 2001;Kikuchi et al., 2001;Kupperman et al., 2000) and independent cues triggering the transition from medial to angular movement. Previous studies have suggested that the endocardial precursors cluster at the future center of the cardiac cone (Stainier et al., 1993;Trinh and Stainier, 2004). To test whether the endocardium influences the redirection of anterior and posterior cardiomyocytes toward a central point, we tracked cardiomyocyte movements in cloche (clo) mutant embryos (Stainier et al., 1995). Although the identity of the clo gene has not yet been reported, prior analyses have established that clo is required cell-autonomously for formation of endothelial precursors, and clo mutants therefore exhibit a general loss of endothelium, including the endocardium (Liao et al., 1997;Stainier et al., 1995).
Examination of clo mutants demonstrated that cardiac fusion begins normally in the absence of endocardium. The medial direction of movement exhibited by clo mutant cardiomyocytes ( Fig. 2A-D and see Movie 2 in the supplementary material) resembled the initial behavior of wild-type cardiomyocytes (Fig.  1E,F). Additionally, the mean rate of clo mutant cardiomyocyte movement (0.62±0.12 m/minute) was similar to the mean rate observed during the first phase of wild-type cardiac fusion (0.68±0.04 m/minute). However, in contrast to wild-type behavior, clo mutant cardiomyocytes failed to make a transition to angular movement (Fig. 2C,D,F,G,H and Table 1). Instead, anterior and posterior clo mutant cardiomyocytes continued to move medially throughout cardiac fusion. Additionally, central cardiomyocytes in clo mutants remained in motion over a longer duration than their wild-type counterparts (compare Fig. 2E with Fig. 1I). During the last 20 minutes of cone formation, only 31% (10 of 26) of the centrally located cells tracked in clo mutant embryos came to a halt. Together, these abnormal movement patterns result in the formation of a dysmorphic cardiac cone (Fig.  2B). Thus, the clo mutant phenotype suggests that the endocardium exerts significant influence on the direction and duration of cardiomyocyte movement during the second phase of cardiac fusion. Subsequently, heart tube extension proceeds slowly and aberrantly in clo mutants, such that the misshapen cardiac cone transforms into an abnormally short and wide tube (Fig. 2I,J). The aberrant dimensions of the clo mutant cardiac cone might be directly responsible for the dimensions of the clo mutant heart tube. However, it is also possible that the morphology of the clo mutant heart tube reflects a separate role of the endocardium in regulating heart tube extension.

Angular movement in the absence of prior medial movement
The aberrant patterns of cell behavior in clo mutants suggest that the endocardium might actively guide the direction of cardiomyocyte movement. Alternatively, the endocardium might play a more passive role in shaping the cardiac cone, perhaps by creating an obstacle for the moving cardiomyocytes. In this scenario, the central location of the endocardium, together with the intracellular junctions connecting the myocardial epithelium (Trinh and Stainier, 2004), could create physical constraints for the medially migrating cardiomyocytes, resulting in the conversion of medial movement into angular movement at the anterior and posterior ends of each heart field. To test whether the transition to angular movement requires an initial phase of medial movement, we tracked cardiomyocyte movement in mutants exhibiting cardia bifida, a phenotype in which the heart fields do not move toward the midline and instead form two lateral hearts. Zebrafish mutations that disrupt endoderm specification or morphogenesis, such as casanova (cas; sox32 -ZFIN) and miles apart (mil; edg5 -ZFIN), result in cardia bifida. The cas locus encodes a Sox-related transcription factor that is required cell-autonomously for endoderm specification (Alexander et al., 1999;Dickmeis et al., 2001;Kikuchi et al., 2001). The mil locus encodes a sphingosine-1phosphate receptor that plays a cell non-autonomous role in recruiting cardiomyocytes to the midline (Kupperman et al., 2000). Endoderm morphogenesis is disrupted in mil mutants (Kupperman et al., 2000), possibly as a consequence of defects in endoderm-extracellular matrix interactions (Matsui et al., 2007), and the mil myocardial defects are presumed to be a result of the observed endoderm defects. We tracked cardiomyocytes in both cas mutants and mil mutants. As we observed comparable phenotypes in both cases, we focus here on our mil mutant data. Examination of cardiomyocyte movements in mil mutants revealed a lack of initial medial movement, followed by a later phase of regionally restricted angular movement (  Table 1). Once this movement had begun, the mean rate of mil mutant cardiomyocyte movement (0.29±0.18 m/minute) was similar to the mean rate observed during the second phase of wild-type cardiac fusion (0.36±0.03 m/minute). Central cardiomyocytes typically exhibited no net directed movement over the course of the time-lapse (16 of 18 cells observed in mil mutants; Fig. 3C,D and Table 1). Continuation of this pattern of movement results in the formation of two lateral cardiac cones that extend into independent tubes (data not shown) (Yelon et al., 1999). Thus, the mil mutant phenotype indicates that the angular movements that establish the cardiac cone can occur even in the absence of initial medial movement.

Endocardium actively induces cardiomyocyte movement
Our data indicate that cone formation in cardia bifida mutants involves angular cell movements reminiscent of those observed during wildtype cone formation (Fig. 3C-G and Fig. 1G,H,L,M), suggesting that 2383 RESEARCH ARTICLE Endocardium and heart tube assembly  Fig. 1). Tracks displayed represent a subset of the tracked cells in this time-lapse. In clo mutants, cardiomyocytes exhibit medial movement throughout the duration of cardiac fusion. Scale bar: in E, 20 m for A-E. (F) Radial bar graph (see Fig. 1J,L) depicting degree of displacement for all tracked clo mutant cardiomyocytes. Few cells exhibit angular movement, resembling the pattern of cell behavior observed during the first phase of wild-type fusion (see Fig. 1J). (G,H) Location of cardiomyocytes moving angularly in clo mutants. In anterior (purple) and posterior (blue) regions ( ‡ ), the percentage of cells moving angularly is significantly reduced (t-test, P<0.01) compared with that observed during the second phase of wild-type fusion (see Fig. 1M). See Table 1 for additional data. (I,J) In situ hybridization depicts cmlc2 expression in wildtype (I) and clo mutant (J) embryos at 28 hpf. Dorsal views, anterior to the top; both images shown at the same magnification. Elongation of the heart tube is delayed and aberrant in clo mutant embryos.
lateral cone formation might also depend on the presence of endocardium. In mil mutants, the presumed endocardial precursors do not move to the midline (Fig. 3H-K). Instead, these cells remained clustered laterally, adjacent to the central regions of the mil mutant heart fields ( Fig. 3H-K). Thus, angular cardiomyocyte movements are directed toward centrally located endocardial cells in both wild-type and mil mutant embryos. To test whether endocardium is required for the angular movements observed in mil mutants, we tracked cardiomyocyte movements in mil;clo double mutants (Fig. 4A-D and see Movie 4 in the supplementary material). mil;clo double mutants displayed the expected lateral heart fields; however, unlike mil mutant cardiomyocytes, mil;clo double-mutant cardiomyocytes did not display any directed movement (Fig. 4C,D). Although cells appeared to jostle relative to their neighbors, they did not exhibit net medial or angular movement. As a consequence, the mil;clo double mutants did not form cardiac cones or tubes; instead, each heart field remained as a sheet of cells. Thus, the mil;clo double-mutant phenotype, like the clo mutant phenotype, suggests that the endocardium is required for angular cardiomyocyte movement during cone formation. Moreover, the requirement for endocardium in a cardia bifida scenario suggests that the endocardium is not merely an obstacle that diverts the medial progress of cardiomyocyte movement. Rather, the endocardium appears to play an active role in inducing the angularly directed cell movements that configure the cardiac cone.

Differential requirements for two distinct phases of cardiac fusion
Together, our data provide a new resolution of the cellular mechanisms regulating heart tube assembly. We find that cardiac fusion involves two morphologically and genetically separable phases of directed cell movement (Fig. 5). During the first phase, coherent medial movement (Fig. 5A, red arrows) recruits cardiomyocytes toward the midline. Transition into the second phase features a significant change in the direction of movement of anterior and posterior cardiomyocytes; these angular movements (Fig. 5A, yellow arrows) integrate the contralateral populations into a properly shaped cardiac cone. These two modes of cell behavior are regulated independently: clo mutant cardiomyocytes undergo the first phase but do not progress into the second (Fig. 5B), whereas mil mutant cardiomyocytes proceed through the second phase without having experienced the first (Fig. 5C). These results suggest that the first phase of cardiac fusion, but not the second, requires myocardium-endoderm interactions (Fig. 5C), whereas the second phase of cardiac fusion, but not the first, requires myocardium-endocardium interactions (Fig. 5B,D).

Endocardium influences the induction, direction and duration of cardiomyocyte movement
Our data suggest a new function for the endocardium in regulating myocardial behavior: the endocardium actively provokes a critical transition in cardiomyocyte movement, inducing a phase of movement that is necessary to shape the cardiac cone. The endocardium may also actively dictate the angular direction of cardiomyocyte movement during the second phase of cardiac fusion. However, our data cannot rule out a more passive mechanism in which the physical position of the endocardial cells influences the angle of cardiomyocyte movement. In addition to its role in triggering cardiomyocyte movement, the endocardium also appears important for bringing cardiomyocytes to a halt: it is likely that the duration of central cardiomyocyte movement is limited by the physical impediment of the centrally located endocardial precursors.

RESEARCH ARTICLE
Development 134 (12) (C,D) Paths traveled by cardiomyocytes in the right lateral heart field during the entire time-lapse (arrows and asterisks as described for Fig. 1). Tracks displayed represent a subset of the tracked cells in this time-lapse. Although initial medial movement is lost, mil mutant cardiomyocytes exhibit a phase of angular movement. Images in C,D are double the magnification of those in A,B. Scale bar: 20 m. (E) Radial bar graph (see Fig. 1J,L) depicting degree of displacement for all tracked mil mutant cardiomyocytes. Most cells exhibit angular movement.
(F,G) Location of cardiomyocytes moving angularly in mil mutants. In mil mutants, as in wild-type embryos, angular movement is regionally restricted. See Table 1  Dorsal views, anterior to the top; only the right lateral heart field is shown in K; both images shown at the same magnification. In both wild-type and mil mutant embryos, the presumed endocardial precursors are clustered adjacent to the central cardiomyocytes.
The molecular mechanisms by which the endocardium stimulates cardiomyocyte movement remain to be explored. The endocardium might communicate with the myocardium via direct cell-cell contacts or via secreted cues. Previous studies have demonstrated multiple requirements for myocardial-endocardial crosstalk during later phases of heart development. For example, myocardialendocardial communication mediated by Tgf-␤, Wnt, Notch, Calcineurin and Vegf signaling is crucial during atrioventricular valve formation (e.g. Armstrong and Bischoff, 2004;Beis et al., 2005;Chang et al., 2004;Hurlstone et al., 2003;Lee et al., 2006;Timmerman et al., 2004). Additionally, Neuregulin and Plexin-Semaphorin signaling are crucial during trabecular outgrowth (Meyer et al., 1997;Toyofuku et al., 2004), Neuregulin and Notch signaling are crucial during specification of the cardiac conduction system (Milan et al., 2006;Rentschler et al., 2002), and Heart of glass signaling is crucial during chamber wall thickening (Mably et al., 2003). In future work, it will be interesting to evaluate whether any of these signaling pathways are crucial for the myocardialendocardial interactions that coordinate heart tube assembly. Alternatively, endocardium could play a less direct role, perhaps contributing to the local deposition of extracellular matrix components (Trinh and Stainier, 2004) in a way that affects the available routes for cardiomyocyte movement.
The regulation of cardiomyocyte behavior by the endocardium reveals a dynamic cellular mechanism by which two populations interact to create the foundation for an embryonic organ. By directing cardiomyocyte movement during cone formation, the endocardium establishes key parameters of heart tube morphology. We speculate that the importance of myocardial-endocardial interactions during heart tube assembly might be broadly conserved, even though the specific geometry of tube formation varies between species (Glickman and Yelon, 2002;Harvey, 2002;Moorman and Christoffels, 2003). In this regard, it is interesting to compare our findings with data regarding cardiac morphogenesis in Ciona intestinalis (Davidson et al., 2005). The Ciona embryo contains a pair of cardiac progenitor cells that exhibit two separate phases of directed movement as they travel from their origin in the tail toward the ventral midline. First, the cardiac progenitor cells migrate anteriorly in close association with the endoderm. Then, the cells switch direction and move ventrally to facilitate cardiac fusion at the midline. Combining data from zebrafish and Ciona, we suggest that a two-phase model of cardiac fusion might be applicable throughout chordate phyla, with an initial phase of cardiomyocyte recruitment relying on interactions with the endoderm, and a second phase of cardiomyocyte integration relying on independent factors such as those supplied by the endocardium during zebrafish cardiac cone formation.
We thank members of the Yelon laboratory, Schier laboratory and Skirball Developmental Genetics Program for helpful discussions. This work was supported by an American Heart Association Heritage Affiliate Postdoctoral Fellowship (N.G.H.) and by grants from the National Institutes of Health and American Heart Association (D.Y.). J.J.S. received support from the NYU Graduate Training Program in Developmental Genetics (NIH T32 HD007520).

RESEARCH ARTICLE
Endocardium and heart tube assembly Paths traveled by cardiomyocytes in the right lateral heart field during the entire timelapse (asterisks as described in Fig. 1). Images in C,D are three times the magnification of those in A,B. Scale bar: 20 m. Example shown is representative of four heart fields analyzed. Cardiomyocytes in mil;clo double mutants exhibit no net directed movement. Fig. 5. Patterns of cell movement demonstrate differential requirements for two discrete phases of cardiac fusion in zebrafish. (A) Cardiac fusion begins with a phase of medial cardiomyocyte movement (red arrows). Cardiac cone formation is facilitated by a second phase of regionally restricted angular movement (yellow arrows). (B) In the absence of endocardium, as in clo mutants, medial movement proceeds, but angular movement is lost, resulting in a dysmorphic cardiac cone. (C) In cardia bifida mutants such as mil, medial movement is lost, but angular movement proceeds, resulting in the formation of bilateral cardiac cones. (D) In mil;clo double mutants, all cardiomyocyte movement is lost, indicating a requirement for endocardium for induction of the angular movement observed in mil mutants.