In vivo evidence for short- and long-range cell communication in cranial neural crest cells

The cranial neural crest is a highly invasive subpopulation of cells that pathfind from the neural tube to the branchial arches during development. The neural crest contributes to many peripheral structures, including the facial skeleton and the nervous system. Intriguingly, cranial neural crest cells do not invade all areas lateral to the neural tube, but rather sort into and migrate along stereotypical routes, creating a striking pattern of three segregated streams (Lumsden and Keynes,1989). The segregation of the streams is thought to prevent cell mixing and cranial ganglionic fusions to ensure proper establishment of anteroposterior structures in the branchial arches(Kontges and Lumsden 1996; Graham and Begbie, 2000). Failure of the cranial neural crest cells to reach the branchial arches can lead to a wide range of facial birth defects(Mooney and Siegel, 2002; Helms and Schneider, 2003),making it an important motivating factor for studying neural crest cell guidance mechanisms.

Because neural crest cells arise, undergo extensive migration and contribute to many different derivatives(LeDouarin and Kalchiem,1999), this system is an excellent model for studying induction,cell guidance and cell differentiation(Santagati and Rijli, 2003). Here, our focus is on the formation of the cranial neural crest cell migratory pattern, which in vertebrates is a consistent, stereotypical pattern of three segregated streams (Kulesa et al.,2004). From many cell labeling, tissue transplant and molecular studies, several distinct hypotheses have emerged on how neural crest cell streams form, but most agree that the pattern develops from a combination of intrinsic and extrinsic factors (Trainor and Krumlauf, 2000; Halloran and Berndt, 2003; Graham et al., 2004). However, despite advances in identifying molecules that induce premigratory cells and specify the fate of a cranial neural crest cell (Knect and Bronner-Fraser,2002; Anderson,2000; Santagati and Rijli,2003), relatively little is known about the molecular mechanisms that direct migrating cranial neural crest cells. Thus, one crucial step in analyzing the pattern formation is to determine the extent to which cranial neural crest cells interact with each other and the environment along the migratory routes in vivo.

Cranial neural crest cells migrate along a dorsolateral route and, for the most part, within one of three streams of cells that develop between the neural tube and the branchial arches. The subregions of the neural tube from which the majority of the neural crest cells emerge correlate with specific segments [rhombomeres (r)] of the hindbrain, namely r1+r2, r4, and r6(Lumsden and Keynes, 1989). Early studies suggest that after receiving a particular molecular identity at the neural tube, neural crest cells emerge and target a peripheral region with a similar molecular identity (Hunt et al.,1991), carrying cues for the patterning of the arch from neural-tube-derived signals (Noden,1983). The need for individual neural crest cell pathfinding in this scenario is simplified if the neural tube controls cell exit points(Lumsden et al., 1991) such that cells diffuse laterally from high populations to low populations in segregated streams (LeDouarin,1982; Newgreen et al.,1979; Rovasio et al.,1983). Studies in chick(Graham et al., 1993; Smith and Graham, 2001) and other amniotes (Knabe et al.,2004) show increased levels of cell death in specific rhombomeres,particularly r3 and r5. However, there is no apoptosis specific to r3 and r5 in Xenopus, zebrafish, and mouse(Schilling and Kimmel, 1994; Ellies et al., 1997; Hensey and Gautier, 1998; Del Pino and Medina, 1998; Kulesa et al., 2004),suggesting that cell death is not solely responsible for segregating the streams. Thus, in a prepattern-type model the neural tube is thought to endow destination instructions and control exit points such that there are few directional cues necessary to produce the cranial neural crest cell pattern.

In contrast to a prepattern hypothesis, studies of cell migratory behaviors and the local environment adjacent to the neural tube suggest that the cranial neural crest cell pattern emerges when cells encounter and respond to environmental cues and interactions with other neural crest cells. Novel culture and imaging techniques, combined with Nomarski optics and labeling of premigratory neural crest cells, has allowed cell migratory behaviors to be visualized in tissue culture (Abercrombie,1970; Bard and Hay,1975; Erickson et al.,1980; Krull et al.,1995), in 2D and 3D gel substrates(Newgreen et al., 1979; Rovasio et al., 1983; Thomas and Yamada, 1992), in whole embryo culture (Spieth and Keller,1984; Kulesa and Fraser,1998), and in ovo (Kulesa and Fraser, 2000). Analyses of cell movements suggest the mechanisms that sculpt the pattern are more complex than would be expected from a purely directed diffusion model and may include cell-cell and cell-environment interactions in the form of chemotaxis, contact inhibition, contact guidance and haptotaxis. From cell labeling studies in a variety of animal model systems, it was learned that cranial neural crest cells emerge and emigrate from all rhombomeres, rather than preferentially exiting from only the even-numbered ones (Sechrist et al.,1993; Schilling and Kimmel,1994; Birgbauer et al.,1995; Trainor et al.,2002). Regions lateral to r3 and r5 inhibit neural crest cell movements; cells stop and collapse filopodia or dramatically change direction into a neighboring stream (Kulesa and Fraser, 1998). Avian grafting experiments suggest the microenvironment adjacent to the neural tube may be important for maintaining the proper segregation of the neural crest cell streams(Farlie et al., 1999). When even-numbered quail rhombomeres are grafted lateral to and adjacent to chick r3, cells from the transplant diverge toward the even-numbered streams rather than migrate further laterally. There are clues that the repulsion is caused by a secreted factor at the neural tube midline. When either the r3 chick neuroepithelium or r5 surface ectoderm is removed, neural crest cells invade the area immediately adjacent to r3 and r5, respectively(Golding et al., 2002; Golding et al., 2004). Thus,in a self-organizing model, the neural crest cell pattern emerges from multiple factors and regional differences.

The following study was guided by our interest in learning more about the nature of neural crest cell pathfinding. Here, we take advantage of the accessibility of chick embryos to perform direct observations. The proximity of the migratory routes (just underneath the surface ectoderm and relatively short time period over which neural crest cells migrate to the branchial arches) permit us to use a whole embryo explant culture method and perform hi-resolution static and confocal time-lapse imaging of individual cell migratory behaviors. We examined in detail to what extent neural crest cells interact with each other and the environment and whether this may influence cell directionality. Using a set of fusion protein constructs targeted to the cell membrane and nucleus, we were able to distinguish the lamellipodia and filopodia of individual neural crest cells. The time-lapse movies capture the dynamics of the cell-cell interactions and outline a chronology of the downstream movements of individual neural crest cells within the migratory streams exiting from r4 and r6. Some of the cell-cell interactions demonstrate obvious opportunities for cell-cell communication and directional guidance. Our data reveal an exciting level of detail to in vivo neural crest cell pathfinding and suggest that local and long-range cell-cell interactions play an important role in cell guidance.

Fertile white leghorn chick eggs were acquired from a local supplier (Ozark Hatchery) and incubated at 38°C until approximately the 5-8 somite stage(ss) of development. Eggs were rinsed with 70% alcohol and 3 ml of albumin was removed before cutting a window in the shell. A solution of 10% India ink(Pelikan Fount; PLK 51822A143) in Howard Ringer's solution was injected below the blastodisc to visualize the embryos. Embryos were staged according to the criteria of Hamburger and Hamilton(Hamburger and Hamilton,1951), denoted as stage 10, for example; in other cases, the embryos were staged by their number of somites, denoted as `10 ss'.

Embryos at 5-8 ss were injected with fusion protein constructs [GAP-43 EGFP and H2B-MRFP (Okada et al.,1999) (gifts from Dr Rusty Lansford/Caltech)] to label premigratory cranial neural crest. After windowing the eggshell, a hole was cut in the vitelline membrane above the neural tube at the cranial end using a fine tungsten needle. In embryos labeled for fluorescence imaging, constructs were injected into the lumen of the neural tube, filling the hindbrain region,using a Picospritzer III (Parker Hannfin Corporation). A small amount (10 mg/ml) of Fast Green FCF (Fisher 42053) was added to the construct for easier visualization during injection. The eggs were then electroporated using platinum electrodes and an Electro Square Porator ECM 830 (BTX, a division of Genetronics) with 20 V, 55 milliseconds pulse length, 5 pulses, at 1 second intervals. Injected eggs were resealed with adhesive tape and re-incubated at 38°C. For static imaging, embryos were evaluated after 12-15 hours. Embryos for time-lapse imaging were evaluated after 6 hours. A fluorescence dissecting microscope (Leica MZFL III) was used to evaluate each embryo for health, uniformity of labeling and brightness.

For static imaging, individual embryos were removed from the egg with paper rings (Whatman #1), cleansed in Ringer's solution and placed dorsal side up within a thin ring of high vacuum grease (Dow Corning 79810-99) on 22×75 mm microslides (VWR 48312-024). The embryo did not come in contact with the vacuum grease. A small amount of Ringer's solution was pipetted away from the embryo and the embryo positioning was adjusted with forceps. A 22×22 mm glass coverslip (VWR 48312-024) was placed on top of the ring of grease,creating a sealed, humidified chamber. The embryos were then imaged using one of three microscopes (Zeiss Axiovert 200M, Zeiss LSM 510 META, and Zeiss LSM 5 PASCAL) and a wide range of objectives.

Whole-embryo explant cultures were prepared according to the method described in Krull and Kulesa (Krull and Kulesa, 1998). Briefly, embryos were removed from the egg by placing an oval ring of filter paper (Whatman #1) around the circumference of the embryo and then cutting around the outer edges of the ring. The ring with the embryo attached was placed in Ringer's solution for rinsing. The paper rings are approximately 1.5 cm along the major axis with holes wide enough to provide ample space between the inner side of the rings and the embryos. This method leaves the entire embryo, as well as the surrounding blastoderm,intact. The excised embryos were cleansed of yolk platelets and India ink by gently pipeting Ringer's solution across them with a P-200 pipetman.

For short-term (<5 hours), hi-resolution (>20×), in vivo time-lapse imaging, embryos were mounted directly dorsal side down into glass bottom microwell dishes (MatTek corporation, P35G-0-14-C). An embryo's blastoderm was spread out using fine forceps. Excess Ringer's was removed to stabilize the embryo while maintaining its 3D morphology. Once positioned, a Teflon membrane ring was sealed on top of the embryo with a bead of silicone grease. The membrane is designed to keep the embryo moist while allowing gas exchange. The membrane ring was made from a Teflon membrane (YSI incorporated,97L0038) stretched over a plastic ring (OD=1 inch, ID=7/8 inch) and sealed in place using beeswax.

To maintain embryos at temperature, a single dish was placed on a heating plate (Lyon Electric, TX7115-020) kept at 38°C using a temperature controller (Cell Temp Bionomic Controller, BC-100). A heat sink compound(#276-1372A-RadioShack) was spread on the heating plate prior to the microwell dish placement to help heat convection to the plastic dish and its contents. Imaging was performed on an LSM 510 META (Carl Zeiss) or an Axiovert 200M inverted fluorescence compound microscope (Carl Zeiss) using a 40×(NA=0.75) plan-neofluar or 40× C-Apochromat W objective (Carl Zeiss). The GFP proteins were excited with the 488 nm laser line using the filter set(Chroma) intended for GFP. The RFP proteins were excited with the 543 nm laser line using the filter set (Chroma) intended for rhodamine. Images were recorded every 1-5 minutes and analyzed using either the AIM (Carl Zeiss) or Axiovision software (Carl Zeiss).

For long-term (>5 hours), low resolution (<20×), in vivo time-lapse imaging, cultures were set up using Millicell culture inserts(Millipore PICM 030-50) and six-well culture plates (Falcon 3046), similar to the protocol described in Krull et al.(Krull et al., 1995). The culture insert membranes were precoated with 200 μl of fibronectin (Sigma F-2006, diluted 1:50 in phosphate buffer) with the excess pipetted away. The dorsal surface of the embryos was placed on the coated culture insert, leaving the ventral surface exposed to the atmosphere. Excess Ringer's solution was pipetted from the membrane surfaces at the rostral and caudal ends of the explants, such that the flow of solution straightened the rostrocaudal axis of the embryos. This naturally spread the explants without flattening the embryos and mimicked the tension of the blastoderm normally created by the stretching of the yolk sac. Each explant covered approximately two-thirds (∼2.8 cm²) of the area of a culture insert. Each individual culture insert was then placed in separate wells of a six-well plate. The membranes were underlain with a Neurobasal medium (Gibco 21103-031), supplemented with B27 (Gibco 17504-036) and 0.5 mmol/l L-glutamine (Sigma G-3126). Sterile water was added to the any unfilled wells to minimize dehydration during time-lapse acquisition. The edges of the six-well plate were then sealed with parafilm.

Fluorescently-labeled explants were visualized using an inverted confocal microscope (LSM 510 META or LSM 5 Pascal; Carl Zeiss Inc.) using a 10×Neofluar (NA=0.30) lens with a zoom=2. This allowed observation of the entire r6 and r7 streams. For better image resolution, plastic was removed from the optical path by making holes in the bottom of the wells into which round 22 mm glass coverslips (VWR 48380-080) were sealed using silicone grease (Dow Corning 79810-99). The membrane of the Millipore culture insert becomes transparent when moist.

The microscope is surrounded by an incubating box fashioned from cardboard(4 mm thick) and covered with thermal insulation (Reflectix Co.; 5/16 inch thick). An enclosed heater (Lyon Electric Co. 115-20) maintained the cultures at 38°C for the duration of the time-lapse acquisition, with only mild temperature fluctuations. Images were recorded every 5 minutes and analyzed using the AIM software (Carl Zeiss).

Several features from the AIM software were used, including 2D and 3D visualization, projection and depth coding (Carl Zeiss). The depth code feature provided the ability to recognize which focal planes the cells were in throughout a z-stack by assigning a color code to the pixel intensity as a function of the z-depth. This was important for determining whether two cells were in contact with one another. Stacks of images were manipulated for analysis using VisArt (Carl Zeiss), Volocity (Improvision) and Image J (NIH).

To determine the extent to which neural crest cells interact with each other and the environment during migration to the branchial arches, we investigated intact cranial neural crest cells within living chick embryos. Previously, fine cellular processes, such as filopodia (∼10-20 μm in length) and lamellipodia of neural crest cells, were detected by scanning electron microscopy (Tosney,1978) and in organ culture(Bard and Hay, 1975). Using a set of fusion-protein-expressing constructs targeted to the cell membrane and nucleus, we electroporated the constructs into young chick embryos to label premigratory cranial neural crest cells. Compared with the conventional methods of vital dye labeling, DiI or non-targeted cytoplasmic GFPs,constructs targeted to specific cytoskeletal elements permit surprising neural crest cell features to be visualized in vivo, including very long and spiney filopodial extensions and thin cellular processes. Through a series of detailed static images, we describe the features of intact neural crest cells in terms of cell shape and number and length of filopodia and lamellipodia. We show how these aspects vary depending on the position of a cell within a migratory stream. We analyze the entire extent of the cranial neural crest cell migratory routes from the dorsal neural tube to the branchial arches,focusing mainly on streams emanating from the mid- and caudal-hindbrain. We demonstrate the dynamics of the lamellipodia and filopodia and the extent to which neural crest cells contact one another and the environment during pathfinding in vivo from time-lapse confocal recordings.

Neural crest cells that traveled within the dense r4 and r6 streams maintained numerous contacts with neighboring cells(Fig. 1). The r4 stream was composed of cells originating from r3-r5 and extended laterally to r4 around the anterior portion of the otic vesicle(Fig. 1A). The r4 stream was densely packed with neural crest cells, as evidenced by the number of nuclear stained cells (Fig. 1B). Many of the cells within the r4 stream had a bipolar shape with two processes extending in opposite directions, one directed toward the branchial arch destination (Fig. 1B). Cells closer to the front of the r4 stream had many more filopodia extended in a variety of directions (described in a separate section below). Cells from the r5 region are known to migrate and join the r4 and r6 streams. Near the mid-r5 region, cells migrated laterally and then turned in either the anterior or posterior direction and moved in a perpendicular fashion to the r4 or r6 streams until making contact with cells in those streams(Fig. 1C).

In a similar manner, the r6 migratory stream was as dense as the r4 stream with as many neural crest cells in close contact with each other(Fig. 1A,D). By contrast to the bipolar shape of the cells within the stream, cells closer to the migratory fronts had a more polygonal shape (Fig. 1D; distal part of the stream). Individual cells tended to be in contact with at least one other cell and often were in contact with several neighbors (Fig. 1D). In higher resolution, neural crest cells indeed had several processes extended in many different directions, such that one cell simultaneously contacted several neighboring neural crest cells (Fig. 1E).

An in-depth look at individual cells within migratory streams revealed the extent to which an individual cell had several contacts with neighboring cells(Fig. 1E, Fig. 2; see Movie 1 in the supplementary material). Color coding cellular features based on z-depth helped to confirm whether filopodial extensions between cells were in the same z-plane (i.e. in contact) with neighboring neural crest cells (Fig. 2). For example, two neighboring neural crest cells were in contact(Fig. 2A,B; arrow), but other filopodia may have actually been separated in z-height(Fig. 2A,C; arrowhead). This contact could not be determined precisely from a projected image(Fig. 1E). The ability to rotate and render a 3D z-stack of confocal images around different axes of rotation (Fig. 2D-F)revealed that there may be other filopodial processes emanating from cells that were not visible in 2D projections. For example, an individual neural crest cell may have been in contact with at least two neighboring fluorescently-labeled cells (Fig. 2D). The first contact is clearly visible(Fig. 2D-E, arrow). The second point of contact is via a branch of the filopodium(Fig. 2F, arrowhead) that is not visible in either the projected image(Fig. 1E) or the depth coding(Fig. 2C, arrowhead).

By contrast to the single or bipolar neural crest cells, numerous neural crest cells displayed filopodia around the entire circumference of the cell body (Fig. 3A; see Movie 2 in the supplementary material). The filopodia did not appear to be distributed or aligned in any specific direction and consisted of both long and short lengths, ranging from approximately 20 μm to 100 μm(Fig. 3A). Cells with filopodia distributed around their entire circumference were usually located at or near the fronts of dense migratory streams (r4 or r6 streams) or adjacent to the neural tube in sparsely populated regions (near r1 or r7). Near the stream fronts, no labelled neural crest cells were usually observed distal to the hairy cells.

Many neural crest cells had a smaller number of filopodia that extended from the cell body in two different directions(Fig. 3B). These extended processes were usually aligned in the direction along the cell's trajectory,toward the branchial arches (at the front of the cell), or in the reverse direction toward the neural tube (at the back of the cell). Within a typical migratory stream, the filopodia may extend and weave around neighboring neural crest cells to contact non-local neighbors(Fig. 3B; see Movie 3 in the supplementary material). The filopodium itself may have a distinct shape(Fig. 3C). The process may be thick near the cell body (∼5 μm in diameter) while tapering away from the cell to approximately 1 μm in diameter. Forward extending filopodia varied in length, typically 50-60 μm, but could extend up to 100 μm in length. Along the length of the filopodium were multiple short processes (5-10μm in length) that extended in orthogonal directions from the filopodium(Fig. 3C; arrowheads). These shorter protrusions were distributed at random points along a filopodium,maintaining spacing with each other (Fig. 3C) and extended at different angles to the filopodium(Fig. 3C). Some filopodia had a wider fan-shape along the length or at the end of the filopodium(Fig. 3C, bottom right corner).

To determine whether there are differences in the proximal (back of the stream) to distal (front of the stream) distribution of hairy versus bipolar cells, we analyzed cell shapes in typical neural crest cell streams(n=8). We then plotted the percent of hairy versus bipolar cells at the front, middle and back of typical neural crest cell streams as a percentage of the total number of cells in the stream(Fig. 4A). We found that a larger percentage of hairy cells tend to be at the stream front (distal)(Fig. 4A). In a typical stream,there were approximately three times as many hairy cells near the front compared with bipolar cells. At the back portion of a stream, the percentage of hairy and bipolar cells was nearly equal. By contrast, a much larger percentage of bipolar cells was found in the midstream region; there were three times as many bipolar cells as hairy cells in a typical migratory stream(Fig. 4A). Thus, a typical neural crest cell stream has the striking feature of hairy cells at the front and bipolar cells distributed throughout the middle of the stream.

To determine whether there is a bias to the directional distribution and length of the filopodia on hairy versus bipolar cells, we measured the average filopodial lengths and spatial distributions on a sample of hairy and bipolar cells (n=16) within typical migratory streams. Bipolar cells had the feature of having few long filopodia that typically extended into only two quadrants (Fig. 4B). The forward extending filopodium stretched in the direction of travel (toward the branchial arch destination) and the trailing filopodium extended parallel, but in the opposite direction (Fig. 4B). In comparison, hairy cells had a large number of filopodia extending from nearly all aspects of the cell's circumference(Fig. 4B). The filopdia were longer in the directions toward the branchial arches and the neural tube.

Long, extended filopodial processes at the leading edge of cells appeared to act as active cell-cell contacts that influenced the direction in which a cell subsequently moved. There were two typical neural crest cell migratory behaviors that occurred when a follower cell contacted a downstream cell. In one case, the filopodium made contact with the downstream cell, and then retracted before the trailing cell began to move in the direction of the downstream cell (Fig. 5). The sequence of events was as follows. First, a cell extended numerous filopodia in different directions throughout the local microenvironment[Fig. 5 (t=0); red cell]. Second, an extended filopodium made contact with a neighboring cell(Fig. 5; t=1 hour), and elicited a response (Fig. 5;t=1 hour 5 minutes). The trailing cell then retracted its filopodium(Fig. 5; t=1.5 hours) and began to move in the direction of the contact with the lead cell(Fig. 5; t=3 hours).

By contrast to this, a neural crest cell filopodium may contact and track the position of a downstream cell (see Movie 4 in the supplementary material). In this sequence of events, a cell first extended a filopodium and contacted a downstream cell. The filopodium remained extended and followed the direction of the back of the downstream cell. At the back of the trailing cell,filopodia continued to be extended in the reverse direction toward the neural tube and to a small extent into other directions adjacent to the cell. As the lead cell continued to migrate further downstream, the filopodium grew as more of the body of the trailing cell followed. Interestingly, quantitative analysis of the area of the trailing cell, plotted as a function of time,revealed that the area of the follower cell did not increase monotonically in the direction of motion, but oscillated slightly(Fig. 4C). The oscillation of cell area versus time appears to result from filopodia that continued to extend in the reverse direction (back toward the neural tube) as the cell body moved in the forward direction (see Movie 4 in the supplementary material).

As described previously, neural crest cells can be connected by thin, long filopodia that are only 1-3 μm in diameter, but extend up to 100 μm in length (Fig. 6; see Movie 5 in the supplementary material). These long cellular processes typically did not lie in a single focal plane, but transcended large x, y and z distances, such that two cells within a migratory stream, although located far apart, may have been connected(Fig. 6B,C). Intriguingly, the long filopodial connections wound in between neighboring neural crest cells(Fig. 6B,C; arrow). An individual cell may have maintained several concurrent connections between neighboring cells, with connections of various lengths(Fig. 6).

The long, thin filopodial connections were visually distinguishable when two neighboring cells moved apart from one another or when one cell divided and one of the progeny moved away. The sequence of how this occurred is demonstrated in Fig. 7. Time-lapse analysis shows a typical pair of neighboring cells in contact(Fig. 7). As the lead cell moved away, a thin process between the cells became visible(Fig. 7). As the distance increased, the process lengthened until it broke at an arbitrary point along the process (Fig. 7). Remnants of a thin filopodium were visible along the cell's route[Fig. 6A,B (arrowhead); Fig. 7]. This behavior was most often observed as cells were just exiting the neural tube or between two progeny of a recently divided cell.

As mentioned before, neural crest cells can have a varied number of filopodia that extend and retract in multiple directions. A single cell can undergo numerous phenotypic changes as it migrates. For example, a cell may start out with a bi-directional shape, extend filopodia in numerous directions and then regain a bi-directional shape while moving forward. These cells typically had few (<5) lengthy (50-60 μm) filopodial extensions, some of which remained extended while the cell migrated; the filopodia responded to the cell's forward movements. Interestingly, although the lamellipodial extensions may be numerous and spread through the local environment, these processes appeared to be confined mostly to the xy-plane of migration, rather than completely around the cell body. In this way, neural crest cells appeared fairly flat in the z-plane (<30 μm) in comparison to the tremendous spatial spread in the z-plane (up to 100μm in certain directions) of the filopodia.

Cranial neural crest cells undergo an extensive migration in three stereotypical streams to pattern peripheral structures of the face and neck. Over the past 30 years, research on neural crest cell migration has focused on elucidating the molecular mechanisms underlying the directional migration cues. Culture and imaging techniques have allowed neural crest cell migratory behaviors to be visualized in cell and tissue culture, 3D matrices, and in intact embryos. However, the lack of resolving fine cellular structures, such as lamellipodia and filopodia, on individual migrating neural crest cells in a living embryo has limited the ability to test the relative roles of potential guidance cues. The approach used here provides insight into the structure and dynamics of neural crest cell interactions in live chick embryos. We found that cranial neural crest cells interact extensively with each other and the local environment through remarkably dynamic short- and long-range lamellipodia and filopodia. The number and directionality of the cell's extensions change dynamically; however, at specific locations within a migratory stream, cells share similar shape features. Long filopodial protrusions can extend up to 100 μm in length and wind in between neighboring cells to contact more distant, downstream neural crest cells. Some of the filopodial connections persist between neighboring cells that move apart. Time-lapse confocal imaging data reveal the spatiotemporal dynamics of the neural crest cell interactions with each other and the local environment.

The observations of lamellipodial and filopodial dynamics between migrating neural crest cells suggest a role for the cell-cell contacts in directional guidance. Short, wide lamellipodial and short, thin filopodial contacts were often exchanged between neighboring neural crest cells. A typical filopodial extension from a trailing neural crest cell contacted, and tracked the back end of a downstream cell (see Movie 4 in the supplementary material), or retracted after contact and then re-extended, with the cell body migrating toward the position of the contact (Fig. 5). These extensions appeared to provide a feedback to inform the cell body what is in the local environment. The cell may then integrate the information and alter its trajectory. We did not see any evidence of a trailing cell nudging a lead cell forward, as judged by the lack of blebbing at the front of the lead cell. Evidence of nudging has been reported in deep cells of the fish Fundulus(Tickle and Trinkaus, 1976). Our data support the growing evidence for the role of filopodia in directional guidance in other systems. During Drosophila neuromuscular synapse formation, embryonic muscles at the target site extend dynamic actin-based filopodia (myopodia) toward invading motoneurons(Ritzenthaler and Chiba,2003). The myopodial contacts are thought to lure the motorneurons to the proper synaptic targets. Also in Drosophila, during dorsal closure the lead cells of an epithelial sheet send out numerous filopodia that contact cells on the opposite side. When the assembly of protrusions is inhibited, the adhesion and fusion of the opposing epithelial partners is prevented, leading to segment misalignments(Jacinto et al., 2000). Thus,our data indicate that local cell-cell contacts influence the migration of neural crest cell direction. It will be interesting to test to what extent the neural crest cells can pathfind when the protrusive activity of the filopodia is modulated.

The long, thin filopodia that stretch between two neighboring neural crest cells as the cells move apart suggests a role for filopodia in long-range cell communication. We presented evidence that some neighboring cells maintain a contact as one of the cells moves away (Figs 6, 7; see Movies 5, 6 in the supplementary material). The process between the cells lengthens until it breaks. In some cases, the trailing cell migrates toward the direction of the fragmented contact, although we did not determine whether the cell precisely follows the trail of the broken contact. Neural crest cells have been shown to make numerous short (10-20 μm) filopodial contacts with neighboring neural crest cells during migration in the chick cornea(Bard and Hay, 1975) and in the mouse gut (Young et al.,2004). In this case, the cell contacts are thought to play a role in mediating the collective migration of the cells in chains. Our evidence of cell-cell contact in neural crest cell streams includes much longer connections (up to 100 μm) and is similar to the growing evidence in other embryonic systems that non-neighboring cells form long-range contacts(Rorth, 2003; Cohen, 2003). The discovery and analysis of long cellular extensions (cytonemes) in the Drosophilaimaginal disc revealed that cytonemes project from distant cells toward the signaling centers of the disc(Ramirez-Weber and Kornberg,1999). The cytonemes are thought to transport, deposit or retrieve signaling molecules and play a role in pattern formation. This idea of long-range cell communication was strengthened with recent work reporting that during Drosophila sense organ development, long filopodia on a single precursor cell convey signals at a distance to non-local neighbors to pattern the field (De Joussineau et al.,2003). Previously, it had been thought that the signal to pattern the sense organ was conveyed through the long-distance secretion of a signaling protein. Thus, it will be interesting to investigate the intracellular dynamics within the neural crest lamellipodia and filopodia during cell-cell contact for the possible role in cell communication.

Our studies have shown that there is extensive detail in neural crest cell pathfinding in the form of short- and long-range cell-cell contacts in vivo,pointing to a diverse set of directional guidance mechanisms for neural crest cells (Fig. 8). The number and length of the contacts, in the form of lamellipodia and filopodia, vary depending on the cell's position within a migratory stream. Time-lapse confocal imaging reveals that the long-range cell-cell contacts, mediated by filopodial extensions, play a role in directional guidance by allowing trailing cells to follow downstream leaders. Short-range contacts between neighboring migrating neural crest cells appear to inform the cells of the position and number of neighboring cells. Long, thin filopodial connections allow two neighboring cells to remain in contact as the cells move apart. Our results support the hypothesis that a combination of intrinsic and extrinsic cues sculpts the cranial neural crest cell migration pattern, but that lamellipodia and filopodia play a critical role in neural crest cell pathfinding in the local microenvironment. Our evidence of neural crest cell short- and long-range cell communication parallels with data in Drosophila, mediated through myopodia and cytonemes, and opens several exciting lines of investigation. The neural crest cell-cell contacts may involve signaling to communicate positional information or allow cells of a similar fate to maintain a relationship. By contrast to our previous view of neural crest cell streams consisting of compactly shaped cells with lamellipodia and relatively short filopodia, our new perspective is that neural crest cell streams are very densely packed with lamellipodia and filopodia in constant cell contact, intertwined around local and non-local migrating cells. Further dissection of the function of the cell-cell and cell-microenvironment interactions will probably bring unexpected insights into how neural crest cells navigate.

The authors would like to kindly thank Rusty Lansford (Caltech) for his generosity in providing the fusion protein constructs. We kindly thank Paul Rupp and Paul Trainor (Stowers Institute) for their careful reading of the manuscript.