Using migrating cells as probes to illuminate features in live embryonic tissues

See allHide authors and affiliations

Science Advances  04 Dec 2020:
Vol. 6, no. 49, eabc5546
DOI: 10.1126/sciadv.abc5546


The biophysical and biochemical properties of live tissues are important in the context of development and disease. Methods for evaluating these properties typically involve destroying the tissue or require specialized technology and complicated analyses. Here, we present a novel, noninvasive methodology for determining the spatial distribution of tissue features within embryos, making use of nondirectionally migrating cells and software we termed “Landscape,” which performs automatized high-throughput three-dimensional image registration. Using the live migrating cells as bioprobes, we identified structures within the zebrafish embryo that affect the distribution of the cells and studied one such structure constituting a physical barrier, which, in turn, influences amoeboid cell polarity. Overall, this work provides a unique approach for detecting tissue properties without interfering with animal’s development. In addition, Landscape allows for integrating data from multiple samples, providing detailed and reliable quantitative evaluation of variable biological phenotypes in different organisms.


Embryonic development relies on specification of diverse cell types and tissues that dynamically organize in three dimensions (3D) to form organ primordia and other structures. The architecture, movement, and function of embryonic tissues are controlled by various biophysical properties and biochemical activities (1, 2). Accordingly, the physical properties of the tissue and the extracellular matrix (ECM) can influence cell fate specification and proliferation (3, 4). Specific tissue properties such as viscoelasticity and stiffness have also been shown to play an important role in large-scale morphogenetic processes such as body axis elongation and germ layer formation (57), as well as in the migration of individual cells (8, 9). Furthermore, tissues’ biophysical properties are relevant in the context of pathological conditions such as cancer development and metastasis, where cell identity, division, and migration are misregulated (10).

Although tissue properties play critical roles in the processes mentioned above, only few methods exist for evaluating these properties in live organisms. For example, using atomic force microscopy to measure tissue stiffness requires direct contact between the tissue and a physical probe, which is challenging to achieve for deep tissues (8, 11). Other approaches such as optical tweezers and microdroplet deformation can be used, but these methods require inserting particles into the tissue, offer relatively low throughput, typically use custom-made specialized technology, and require more complicated analyses of the results (11, 12). One emerging relevant technology is Brillouin microscopy, a nondestructive, label-free, and contact-free method that can report on cells’ and tissues’ viscoelastic properties (13, 14). However, this methodology requires relatively long acquisition time and, in the context of studying live embryos, is more suitable for restricted domains that have been highlighted by other methods as regions of interest. Thus, it remains challenging to characterize tissue properties in vivo with high resolution in 3D and with minimal intervention.

To address this challenge, we present here a noninvasive approach for examining tissue properties and assessing their relevance for cell migration in vivo. Our approach involves using nondirectionally migrating cells, in this study, primordial germ cells (PGCs) (15), as bioprobes. In many vertebrates and invertebrates, PGCs migrate within the early embryo, using an evolutionarily conserved robust migration mode similar to that of the single-celled amoeba (16, 17); amoeboid-type migration, which is also observed in immune cells, is characterized by extensive cell body deformations that enable invasion through gaps in the environment. In addition, as amoeboid cells do not form stable and specific connections with neighboring cells or with the ECM, they are less restricted in their migration through different tissues (18). Thus, amoeboid migration represents a simple, flexible, and efficient mode of moving through tissues, which is consistent with the finding that this type of movement is common among metastatic cancer cells (19). Following their specification at four equidistant positions around the embryo, zebrafish PGCs acquire amoeboid cell motility and are guided toward the region where the gonad develops by the chemokine Cxcl12a (20, 21). In the absence of the Cxcl12a receptor Cxcr4b, the PGCs are motile but migrate nondirectionally (21, 22). Accordingly, since they start their migration from different points along the circumference of the early zebrafish embryos (20), the nondirectionally migrating PGCs, which we hereafter call naïve PGCs (nPGCs), can interact with and probe every developing tissue. The distribution of the nPGCs and their behavior within different regions of the embryo represent a new strategy for defining tissue and tissue border properties, in particular those relevant for cell migration in vivo. This method makes use of a software named here “Landscape,” an automated 3D image registration-based platform that we developed. Landscape allows us to integrate 3D data from a large number of wild-type or mutant embryos, it can be applied to different species and provides quantitative distribution maps of cells within them. Using zebrafish nPGCs as bioprobes, we established a unique map of the tissue properties of early zebrafish embryos and present our analysis of one of the structures that we identified: a tissue boundary acting as a physical barrier for cell migration.


Establishing a large-scale 3D image registration pipeline to identify and map structures that influence cell migration in vivo

Unlike their highly organized migration in wild-type embryos, nPGCs that lack the guidance receptor Cxcr4b fail to reach their migration targets and instead migrate nondirectionally, eventually arriving at positions in the embryo where they are not usually found (Fig. 1A and movie S1) (21, 22). In the context of this work, we consider the nPGCs to be sensors for tissue properties that are relevant for single-cell migration in vivo. Accordingly, when nPGCs are distributed in nonrandom patterns, this can reflect features in the environment that affect cell migration and reveal the degree to which amoeboid-type migration is robust with respect to specific biophysical and molecular tissue properties.

Fig. 1 Application of Landscape in 3D registration of embryos and generation of precise heatmaps.

(A) Images of a wild-type (left) and three example cxcr4b−/− embryos (right). In the wild-type embryos, PGCs (red) form two clusters, while in the mutant embryos, they are distributed in a nondistinguishable pattern. Nuclei are labeled in blue. Scale bar, 100 μm. (B) Zebrafish embryo used for the 3D image registration, with the structures used by the Landscape software: nuclei, landmark, and cells of interest (here, nPGCs) labeled in blue, green, and red, respectively. Scale bar, 100 μm. (C) Workflow of the Landscape pipeline. Embryos are normalized, registered with respect to size and orientation, and overlayed. The virtual 3D sphere is unfolded into a 2D projection map using conformal map projections. Then, the heatmap of the nPGC positions is generated. See fig. S1E for more details. (D) Demonstration of 3D image registration of zebrafish embryos. Two images of the same embryo were captured at two different orientations, with the size of one orientation reduced. Images were overlayed directly (no 3D image registration) and after correction (3D image registration). White represents the overlap of the signals between two embryo orientations. See movie S3. Scale bar, 100 μm. (E) Heatmaps of PGC positions in wild-type embryos (presented as whole-embryo projection) before (left, nonregistered) and following registration (right, registered). The abundance value in the scale represents the proportion of all registered cells (see Materials and Methods). A close-up view of the center of the clusters is presented in the lower panels. N = 97 embryos and n = 2106 PGCs.

To examine this idea, we focused on the 10 hours post-fertilization (hpf) stage, when PGCs actively migrate within the developing embryo, a stage when gastrulation has just been completed and early morphogenesis events occur. The average number of PGCs at this stage is about 22 per embryo, a number that is too small for comprehensive sampling of a substantial portion of the embryonic tissues (Fig. 1A). Assuming a random distribution of the nPGCs, full coverage of the cellular portion of the embryo would require about 8000 cells (equivalent to 360 embryos, fig. S1, A and B; see the “Volume measurements” section in Materials and Methods and Discussion). This would entail integration of data from a large number of 3D datasets, which is complicated by the fact that embryos differ in size and precise imaging orientation (fig. S1, C and D, and movie S2). This general issue, concerning the need to combine large sets of data for the purpose of determining average phenotypes, is of major importance in the field of developmental biology, where considerable variability among individual samples exists (2325).

To pool and compare 3D datasets derived from a large number of biological samples, we developed a high-performance image processing software platform that we named Landscape (can be downloaded from In the context of this work, the software can align data derived from a large number of individual embryos within a normalized coordinate system, and this alignment is established relative to a reference structure, or a landmark. As a landmark, we made use of the tissue-specific expression of green fluorescent protein (GFP) driven by the goosecoid promoter, which labels mesodermal tissues at the midline of the embryo (Fig. 1, B and C). Using this alignment strategy, Landscape automatically performs the 3D image registration and generates a heatmap of nPGC positions. Landscape follows a processing pipeline that includes multiple interdependent steps (fig. S1E). First, morphological filters are applied to preprocess all imaged microscopy channels. This step reduces background signal noise, thereby enhancing the contrast of the imaged data. Second, the software performs 3D image segmentation of the whole embryo (based on the nucleus signal), the nPGCs, as well as the landmark, as visualized in the corresponding microscopy channels. On the basis of the segmentation of the nuclei, shape analysis of the embryo is performed, considering that the embryo shape can be approximated by a 3D ellipsoid. In this step, the parameters that best fit an ellipsoid for every embryo shape are calculated. The shape and dimension data of each embryo are then transformed to fit a normalized coordinate system, resulting in embryos of the same shape. In the next step, the software aligns the data, using a geodesic registration algorithm, we developed that orients the embryos with respect to the signal of the segmented landmark. The imaging data are thus normalized and registered and can be pooled with the mapped nPGC center coordinates obtained from the segmented images. To visualize the 3D nPGC distribution within the reference system, we made use of conformal map projections, commonly used in cartography (26) and in the visualization of data from 3D fluorescence microscopy (23, 24). Thus, similar to Earth’s projection onto cartographic maps, we unfolded the 3D data derived from the imaged embryos to obtain 2D cell distribution maps. As outputs, we obtained heatmaps that correspond to the underlying probability distribution of nPGCs at different depths within the reference embryo (see Materials and Methods for more details).

As a first test for the 3D image registration algorithm, we acquired multiple images of the same embryo and manipulated them concerning size and imaging orientation. Specifically, we imaged one embryo at two different imaging orientations and computationally generated a smaller-sized version of the rotated dataset. As manifested by the precise overlap of the landmark and the germ cell signals, the algorithm effectively corrected for both imaging orientation and embryo size (Fig. 1D and movie S3).

Next, we examined whether the Landscape software could highlight patterns of cell distribution within the tissue. To this end, we first registered wild-type embryos, in which the PGCs cluster at specific domains within the embryo in response to the guidance cue (Fig. 1A, left). Mapping the distribution of about 2100 PGCs (derived from 97 registered embryos) over the whole embryo revealed a distinct distribution pattern, highlighting the normal position of the PGCs at this stage (Fig. 1E, right). The observed cell clustering pattern is in agreement with previously described PGC distribution in individual embryos at this stage (20, 21). In contrast with a simple overlay of nonregistered data, the 3D registered data define sharper borders for cell clusters, as well as subdomains of higher cell density within each cluster (Fig. 1E, left), underlining the importance of precise alignment of the data to reveal fine details in cell positioning. These tests show that the established algorithm is a robust and reliable tool for automated 3D image registration of large datasets, as demonstrated by the distribution map of single cells.

nPGCs migrate throughout the embryo and highlight structures that influence cell migration within the live tissue

To determine the distribution of the nPGCs relative to the mediolateral and anteroposterior axes of the embryo, we constructed a distribution heatmap made of about 21,000 cells derived from more than 900 embryos. As expected from PGCs lacking the guidance receptor, the nPGCs did not cluster at the positions where they normally do at this stage. Notably, however, the nPGCs exhibited a distinct distribution pattern that cannot be recognized by monitoring single embryos (Figs. 1A and 2, A and B). The distribution pattern became clearer and more defined as the number of nPGCs was increased (Fig. 2, A and B, and movie S4) and the datasets registered (fig. S2A). Along the mediolateral axis, the highest cell density is found on either side of the midline (Fig. 2B, arrows), with medial structures devoid of cells (head; Fig. 2B, asterisk) or containing much fewer cells (Fig. 2B, arrowhead). In addition, two domains lateral to the head were populated by fewer cells (Fig. 2B, encircled areas). Thus, the distribution of nPGCs throughout the embryo can report on nonhomogeneous features of embryonic tissues, potentially biochemical and biomechanical properties. In this context, we provide experimental evidence that the known guidance cues Cxcl12a and Cxcl12b are not involved in controlling the distribution of the nPGCs (fig. S2B).

Fig. 2 Distribution pattern of nPGCs within the zebrafish embryo.

(A) Construction of whole-embryo nPGC distribution maps made of different number of cells randomly selected from the total registered cells (21,520 cells; see also movie S4). Cell abundance represents the proportion of all registered cells within each heatmap. (B) A detailed heatmap of nPGC distribution within the cxcr4b−/− embryos. The asterisk marks the head area; the arrowhead marks the midline, the domains that nPGCs avoided. The encircled areas (referred to as region A in fig. S6B) highlight the region with reduced number of nPGCs. Arrows point to regions of nPGC accumulation. Cell abundance represents the proportion of all registered cells. N = 934 embryos and n = 21,520 nPGCs. (C) PGC distribution within tissues belonging to specific germ layers of wild-type and cxcr4b−/− embryos. PGC enrichment was calculated as the percentage of PGCs found within the germ layer derivatives divided by the percentage of the volume occupied by the germ layer derivatives (see also fig. S2, C and D). (D) 3D nPGC distribution relative to the landmark. The distribution of the cells within three virtual shells: at the level of landmark, above landmark, and below landmark. Arrowheads point at the midline area, and the arrows point at cell accumulation domains. The abundance value in the scale represents the proportion of all registered cells. N = 934 embryos and n = 21520 nPGCs.

Next, we examined the distribution of nPGCs relative to germ layer–specific markers (fig. S2, C and D) (2732). Considering the volume that the labeled tissues occupy, the localization of the PGCs within wild-type embryos is biased toward the endoderm (the enrichment value is 3.6 times higher relative to mesodermal tissues and 18 times higher than in the ectodermal derivatives, where PGCs are rarely found; Fig. 2C and table S1). These results are in line with the observation that during early development of wild-type embryos, migrating PGCs are found in close proximity to endodermal cells (33). In contrast to the wild-type scenario, nPGCs are more evenly distributed across derivatives of the different germ layers (Fig. 2C and table S1). These findings further indicate that nPGCs are robust and can reach and migrate within germ layers that they normally do not encounter.

To combine the results regarding the location of the cells along the mediolateral and anteroposterior axes (Fig. 2B) with their depth within the embryo, we examined their position in this dimension relative to the landmark (goosecoid expression domain; Fig. 1B). To this end, we divided the embryo into three virtual shells: at the “level of landmark,” “above landmark,” and “below landmark” (Fig. 2D). The distribution of the nPGCs was different among the different shells. While in the layer above the landmark the midline area was occupied by the cells (Fig. 2D, left, arrowhead), at the level of landmark the same region was free of cells (Fig. 2D, middle, arrowhead). The layer below the landmark had few cells, but it also showed an area devoid of cells in the midline (Fig. 2D, right, arrowhead).

The position and shape of the domain the nPGCs avoided resemble the shape of the landmark that we used, which labels the prechordal plate and the notochord tissues (see overlap of gsc:GFP with the expression pattern of the notochord marker no tail (ntl) in Fig. 3A). To test whether indeed the region highlighted by the cells corresponds to these tissues, we specifically perturbed the development of the notochord using morpholino antisense oligonucleotides (MO) to knock down the transcription factor protein Noto (fig. S3, A to C) (34). In embryos in which the development of the notochord was compromised (noto KD), nPGCs could be found at midline positions that were normally devoid of the migrating cells (Fig. 3B, the level of landmark shell, bottom). In contrast, the positioning of wild-type PGCs that express Cxcr4b (and therefore can respond to Cxc12a) is not affected in noto mutants (fig. S3D). These findings indicate that the region of the early developing notochord affects nPGC localization. At the same time, we did not observe obvious differences between control and noto KD embryos concerning the distribution of nPGCs in the region of the head (Fig. 3B), which develops normally in noto mutants (34).

Fig. 3 Identification of the notochord region as an area avoided by the nPGCs.

(A) Left: Schematic depiction of the gsc:GFP-labeled region, with the notochord outlined in red and prechordal plate outlined in green. Right: Confocal microscope image of wild-type transgenic (tg) gsc:GFP, ntl:CFP embryo at 10 hpf. Scale bar, 100 μm. (B) nPGC distribution in control and in notochord-deficient embryos. Expression of the gsc:GFP transgene in control (cxcr4b−/−, control) and notochord-deficient (cxcr4b−/−, noto KD) embryos (left). Distribution heatmaps of nPGC above the landmark, at the level of the landmark and below the landmark (right panels) are shown. The abundance value in the scale represents the proportion of all registered cells. Control, N = 431 embryos and n = 8802 nPGCs; noto KD, N = 415 embryos and n = 8507 nPGCs. Scale bar, 100 μm.

Next, to determine the basis for the inhomogeneous distribution of the nPGCs, we first followed the behavior of nPGCs located near the midline between 9 and 11 hpf in control and noto KD embryos (Fig. 4A and movies S5 and S6). In control embryos, nPGCs rarely crossed the region where the notochord develops (Fig. 4, A and B) and rather bounced off of the border of the forming tissue upon contact (Fig. 4, A and B, and movie S5). In contrast, in noto KD embryos, the nPGCs readily entered the midline region and frequently crossed it (Fig. 4, A and B, and movie S6), suggesting that the region of the normally developing notochord has properties that prevent nPGCs from invading it. Consistent with the cell distribution map of notochord-deficient embryos (noto KD; Fig. 3B, bottom panels), we did not observe significant differences in crossing frequency between control and knocked-down embryos in the region of the head (Fig. 4B). Together, these results demonstrate that as live probes, nPGCs reveal structures that affect single-cell migration in vivo, for example, the border of the notochord, which acts as a barrier for cell entry.

Fig. 4 Migration of nPGCs in wild-type and in notochord-deficient embryos.

(A) The behavior of nPGCs in the region of the developing notochord in control (cxcr4b−/−, control, see movie S5) and notochord-deficient embryos (cxcr4b−/−, noto KD; see movie S6). Scale bar, 10 μm. (B) Quantification of crossing events of nPGCs in the presence (cxcr4b−/−, control) and absence of the notochord (cxcr4b−/−, noto KD). Cells coming in contact with the notochord (control) or entering the domain where the notochord should have developed (noto KD) were considered. Control, N = 20 embryos and n = 207 nPGCs; noto KD, N = 20 embryos and n = 208 nPGCs. Fisher’s exact test (two tailed; confidence interval, 95%) was used for statistical analysis; P > 0.9999 [not significant (ns)] and ****P < 0.0001.

The interaction of nPGCs with tissue barriers

To characterize the dynamic behavior of the migrating nPGCs at the border of the notochord, a domain that they do not invade, we followed the cells at high spatiotemporal resolution, monitoring their migration path and polarity. Tracking the cells for at least 30 min in the region of the developing notochord, we observed two distinct cell behaviors. The cells either kept moving along the area marked by GFP (fig. S4A), referred to as nonreflective interactions, or deflected and moved away from it (Fig. 5A), referred to as reflective interactions. We found that cells showing nonreflective behavior approached the notochord at shallow angles (25° to 55°), while cells with reflective behaviors approached the notochord with intermediate (45° to 65°) or steep angles (70° to 90°) (Fig. 5A; fig. S4, A and B; and movies S7 to S9). Focusing on the reflective behavior, we found that the angles of approach and deflection were similar for cells approaching the notochord at intermediate angles, while cells that approached the notochord with steep angles moved away from it at more shallow angles (Fig. 5B and fig. S4C). These findings show that the impact angle affects the response of the cells to the barrier, as observed in the ensuing path of migration.

Fig. 5 The effect of the orientation of interaction with the barrier on nPGCs behavior.

(A) Impact angle–dependent response of nPGCs following interaction with the notochord. Intermediate angle (45° to 60°, blue) and steep angle (70° to 90°, orange). Scale bars, 10 μm. (B) Ratio between angle of deflection and angle of impact in the two cell populations. (C) Duration that cells in each group maintained contact with the barrier. (D) Effect of the interaction with the barrier on actin distribution at the cell front. (B to D) Intermediate, N = 24 embryos and n = 26 nPGCs; steep, N = 14 embryos and n = 13 nPGCs. The statistical significance was evaluated using the Mann-Whitney U test (two tailed; confidence interval, 95%): (B) ****P < 0.0001; (C) **P = 0.0024; (D) 0 min, P = 0.3183 (ns); 1 min, ****P < 0.0001; 2 min, ****P < 0.0001; 3 min, ****P < 0.0001; 4 min, *P = 0.0118; 5 min, P = 0.9888 (ns). Data are presented (B and C) as boxplots, and error bars designate minimum to maximum range of the data points; (D) as median ± interquartile range. (E) Measurements of the Brillouin shift in the xz planes at the level of notochord (N) and presomitic mesoderm (PSM) in 10-hpf embryo. Sample images of GFP signal and Brillouin shift are presented in the left panel. Right panel shows the average plot of the Brillouin shift across the borders between the notochord and PSM. Arrows point at the Brillouin shift drop. Gray dashed lines represent the SD. N = 13 embryos. Scale bar, 15 μm.

Next, we monitored the duration that the cells with reflective behaviors maintained their interaction with the developing notochord tissue. While cells with an intermediate impact angle remained in contact for about 10 min, cells that approach the barrier tissue at steep angles maintained their interaction for a significantly longer time (16 min on average; Fig. 5C). We hypothesized that the different durations of interaction with the barrier result from different effects on the polarity of the nPGCs. To examine this possibility, we characterized the distribution of actin in the cells upon impact. We found that the level of actin at the cell front, where blebs are more common (fig. S4D), decreased upon interaction at an intermediate angle (Fig. 5D and movie S8). Notably, a steep interaction angle resulted in a much more pronounced decrease in actin level (Fig. 5D) and loss of a defined cell front (Fig. 5, A and D, and movie S9). Intrigued by these findings, we sought to analyze this behavior in silico. Taking into account a reduction of actin-dependent protrusive activity at the site of the impact and suppression of protrusive activity at the rear end (fig. S4D) was sufficient to recapitulate the behavior that nPGCs exhibited upon interaction with the barrier (fig. S4E and movie S10). Together, the orientation of the interaction of the migrating cells (nPGCs in this case) with a migration barrier (the notochord in this case) determined the precise reaction with respect to the distribution of actin within the nPGCs and their further migration path.

To define the physical basis for the migration-relevant tissue features revealed by the migrating cells, we examined whether the position of the barrier is characterized by an increased cell density that restricted integration of nPGCs into it. To this end, we labeled the somatic nuclei and examined their distribution with respect to the localization of nPGCs. While we observed a negative correlation between the density of nuclei and the presence of nPGCs within the notochord (fig. S5A, see selected area 1 and the corresponding line plot 1), in general, when considering other regions in the embryo, positive correlations can be observed as well (e.g., fig. S5A, see selected area 2 and the corresponding line plot 2). These observations suggest that cell density alone cannot explain the distribution of nPGCs within the tissue.

An additional tissue feature that could affect the migrating nPGCs is tissue stiffness, a property that was shown to affect cell migration in vivo, where it could either facilitate (8) or inhibit (9) the migratory behavior per se. To examine whether the altered behavior of the nPGCs at the border of the notochord resulted from differences in stiffness, we used Brillouin microscopy, an emerging technique in the field of mechanobiology (14) and currently the only method that allows noninvasively probing tissue elasticity in 3D. Notably, when focusing on the distinct position defined by the nPGCs, we found that the boundary between the notochord and the early differentiating somitic tissue was characterized by a clear drop in the Brillouin shift (cross section in Fig. 5E, arrows and plot; see fig. S5B for Brillouin microscopy verification). These observations are consistent with the idea that the nPGCs identified this boundary as an embryonic structure with different biophysical properties that constituted a barrier from which they deflected.

These findings focused our attention on the basement membrane, which is made of different ECM proteins and is located at this position in zebrafish (35, 36) and was also described in mouse (37). This tissue structure is crucial for shaping the developing notochord by forming a clear boundary with neighboring tissues (38, 39). To determine whether a basement membrane structure is present at this early stage of notochord development in zebrafish, we performed antibody staining against laminin, a key component of the basement membrane. Notably, while laminin protein appears in a dotted pattern throughout most of the embryo, we found a structured distribution of the protein along the notochord, suggesting that the basement membrane surrounding it could constitute a barrier that the nPGCs identified (fig. S5C, arrowheads). In noto mutants in which the nPGCs can invade the midline of the embryo, the notochord area is devoid of a laminin sheet, showing a correlation between the presence of structured ECM and the ability of nPGCs to enter the area.

In addition to the notochord, the nPGCs revealed other structures that affected their migration, such as the prechordal plate tissue (Fig. 2B, asterisk) and two anterior domains lateral to the prechordal plate (Fig. 2B, encircled areas, hereafter referred to as region A). As the prechordal plate tissue was devoid of nPGCs to a similar degree as the notochord (Fig. 2D, asterisk and arrowhead), we considered that deposition of ECM might have prevented the nPGCs from invading this part of the embryo. However, in contrast to structured laminin at the border of the notochord (fig. S5C, bottom, arrowheads), laminin was expressed in a dotted pattern within the prechordal plate (fig. S5C, top). Consistently, we did not observe Brillouin shift drops at the border of the prechordal tissue to the adjacent somitic tissue (fig. S6A), suggesting that differential deposition of ECM may not be involved in this case. Our initial experiments show that region A is accessible for the nPGCs (fig. S6B), contrary to other depleted regions, suggesting the involvement of another mechanism. Collectively, by using nPGCs as bioprobes and the Landscape software, we could define domains in the vertebrate embryo that are characterized by specific biophysical and molecular properties relevant for tissue morphogenesis and the formation of tissue borders.

The Landscape framework facilitates mapping of embryonic structures and registration in different experimental models

The results above show that Landscape can be used for mapping the positions of single cells in embryos. As a next step, we set out to examine whether Landscape could also be used in other contexts for quantitative analysis of phenotypes where large structures are affected. To this end, we disrupted the development of the paraxial mesoderm tissue in zebrafish embryos and determined the consequences of this manipulation on the development of the neighboring axial mesoderm (Fig. 6A). Registering 40 control embryos and 40 embryos in which the paraxial mesoderm was disrupted (tbx16 KD), the heatmap of the axial mesoderm tissue reveals mediolateral extension of the perturbed tissue as compared with the control (Fig. 6A). Intriguingly, the heatmap also reveal that the length of the axial mesoderm tissue along the anteroposterior axis was shorter in tbx16 KD embryos. These observations (broader and shorter axial mesoderm tissue) are in agreement with previous studies (40, 41). In contrast with the previous studies, however, analyzing the phenotype with Landscape allowed for the rapid integration of results from many embryos, thereby allowing for a more comprehensive, unbiased, and quantitative analysis.

Fig. 6 Quantitative analysis of tissue patterning in zebrafish and in Drosophila using Landscape.

(A) Effect of paraxial mesoderm disruption (tbx16 KD) on axial mesoderm patterning. Left: Representative images from control and tbx16 KD embryos at 10 hpf. Axial mesoderm was visualized using tg(gsc:GFP) zebrafish embryos. N = 40 embryos for each condition. Scale bar, 100 μm. Right: Tissue maps generated using the Landscape program that represent distribution of the axial mesoderm. Overlay shows the segmented axial mesoderm tissues; control (magenta) and tbx16 KD (green), revealing differences in the length and width between the control and treated embryos. (B) Quantitative analysis of Drosophila tracheal branching using Landscape. Left: Representative images of embryonic stage 15 control embryos (wild type, y w) and embryos in which the Branchless (Bnl, fibroblast growth factor) protein was misexpressed using the 69B::GAL4 driver line (Bnl misexpression). Tracheae (red) were visualized by chitin staining. N = 37 embryos were imaged for each condition. Scale bar, 50 μm. Middle: Tissue maps generated using the Landscape program that represent distribution of the tracheal branches. Right: An overlay of trachea distribution in wild-type and Bnl-misexpressing embryos.

As presented above, the Landscape framework allowed us to accurately register and analyze sphere-shaped zebrafish embryos. To examine the robustness of the method, we sought to use the Landscape framework on specimens whose shapes differ from sphere-like zebrafish embryos.

As a proof of principle, we tested Landscape on Drosophila embryos (stage 14) that assume an elongated morphology. Analogous to the goosecoid expression used in zebrafish, we made use of the embryonic nerve cord, highlighted by phalloidin staining as a landmark (fig. S7A). Nuclei of glial cells, visualized by the glial-specific Reversed polarity (Repo) antibody (42), were considered as the points of interest to be registered. Images of the same Drosophila embryo were acquired at two different orientations, and the images were superimposed with or without 3D image registration using the Landscape program (fig. S7A). The positions of Repo-positive glial cell nuclei that are misaligned are effectively corrected by the Landscape program, such that in the overlay, each of the Repo-positive glial cell nuclei occupies the same position (fig. S7A and movie S11). This proof-of-principle result validates the robustness of the Landscape program, as it can be applied to embryos of different shapes from different species.

The success of the Landscape program in registering Drosophila embryos allowed us to generate a heatmap presenting the distribution of glial cell nuclei derived from multiple wild-type embryos (fig. S7, B and C). The resulting heatmap shows the distribution of glial cells along nerves emanating from the center of the embryo (fig. S7C, arrowhead) and reveals a strong accumulation of glial cells along the ventral nerve cord at the center of the embryo (fig. S7C, asterisk), a quantitative cell distribution feature that is less obvious when visualizing single embryos (fig. S7, B and C).

As a further demonstration of the functionality of the Landscape framework for analyzing Drosophila embryonic development, we investigated the establishment of the tracheal system. Tracheal branching is guided by the fibroblast growth factor homolog branchless (Bnl), which is produced by clusters of cells surrounding the tracheal primordia (43). To characterize the effect of ectopic expression of Bnl on the patterning of the tracheal system, we expressed it using the pan-ectodermal Gal4 driver 69B-Gal4 (Fig. 6B). We analyzed the distribution of tracheal branches marked by luminal chitin in 69B>Bnl and in wild-type control (y w) embryos (Fig. 6B).

In individually imaged control embryos, the tracheal system is stereotypically organized, with a single ganglionic branch in each hemisegment extending toward the ventral nerve cord (Fig. 6B top left). In contrast, Bnl misexpression leads to excessive branching and misguided migration of the tracheae (Fig. 6B, bottom left) (43). Notably, the distribution maps of the tracheae derived from multiple embryos revealed patterns that were not apparent from observing individual embryos. Specifically, the head region was more frequently invaded by tracheal branches in control embryos as compared with the Bnl misexpression situation (Fig. 6B, right, asterisks). In addition, the extent of the outgrowth of tracheae toward the ventral nerve cord was shortened in Bnl-misexpressing embryos (Fig. 6B, right, overlay of distribution maps, arrows), a feature that is more challenging to appreciate and quantitate when studying single embryos. Collectively, these results demonstrate the power and versatility of the data-driven approach for quantitative investigation of 3D cell distribution patterns and tissue properties at a high resolution.


The development and maintenance of embryonic structures are influenced by various biophysical and chemical properties of the tissues (1, 2). To expand the methods available for examining these tissue properties [reviewed in (11)], we present a noninvasive, biological property–based strategy for determining the spatial distribution of tissue properties, which makes use of an automated simple software tool we developed. Specifically, using nPGCs as bioprobes for tissue features, we map structures within the developing live embryo on the basis of how they influence cell migration. At the same time, distinguishing between regions in the embryo that can or cannot be populated by migrating cells provides unique insights regarding the robustness of particular modes of migration in different biological contexts. The method that we present here relies on localization of many cells that highlight domains containing more or less data points. Such analysis provides increased resolution of distribution patterns as more cells are included (see Fig. 2A and movie S4).

Once domains that affect cell migration are identified, one can analyze the relevant molecular and structural features of the tissue and its boundary and characterize the immediate cellular response to the interaction. Here, we conducted such an analysis on one of the structures that we identified: the border between the developing somitic mesoderm and the developing notochord; a basement membrane–surrounded tissue that was suggested to act as an early barrier for signal diffusion in the mouse and as a physical barrier at advanced stages of somitogenesis in the chick (44, 45). Focusing on this region, we found that it is characterized by elastic properties that are notably different from those of neighboring tissues, as determined by a lower Brillouin shift (Fig. 5E). The physical properties of the structure surrounding the developing notochord differ between this study and a previous one (46), where it was found to exhibit a higher Brillouin shift as compared with surrounding tissues (see also fig. S5B for data reproduction). We attribute the difference to the fact that the study presented here was conducted during early embryogenesis when extensive tissue remodeling is required, which is different from the more advanced stages studied by Bevilacqua et al. (46). This difference underscores the unique information that our cell behavior–based strategy for tissue mapping provides as compared with other alternatives. Characterizing the specific migration barrier that we focused on, we found that it consists of a sheet of laminin, in contrast with neighboring tissues that contain nonorganized spots of this protein (fig. S5C). The involvement of laminin in barrier formation was also shown in the establishment of a morphological boundary around the developing notochord in ascidians (38). Similarly, zebrafish laminin mutants are defective in notochord differentiation and fail to form a proper basement membrane surrounding the notochord (47). It would be interesting to compare the behavior of the migrating cells when encountering the dispersed spots of laminin with the response that we present when interacting with the same material organized as a sheet-like structure.

The behavior that nPGCs exhibited when encountering the ECM-containing physical barrier could be related to their amoeboid migration mode (17, 18). This robust ancestral mechanism of locomotion is used by germ cells in different organisms and is characteristic of unicellular eukaryotes such as Dictyostelium discoideum. Single Dictyostelium cells can aggregate, adhere to one another, and form a migrating multicellular slug, where ECM material was suggested to act as a physical barrier that prevents loss or entry of new cells into the aggregate (48). The response of the PGCs at the interface between the notochord and the developing somites could thus reflect a primary, evolutionarily conserved response of amoeboid cells to ECM-based cell-mixing borders.

The observation that upon physical contact, the nPGCs migrate away from the notochord border in an impact angle–dependent manner (Fig. 5A and movies S8 and S9) is reminiscent of a behavior termed contact inhibition of locomotion (CIL) (49). Accordingly, CIL was shown to control cell behaviors concerning angles of collision and deflection, for example, in the context of migratory behavior and dissemination of Drosophila hemocytes and vertebrate neural crest cells (50, 51). E-cadherin, which is a key cell adhesion molecule expressed within PGCs (52) could potentially be involved in the behaviors that we observed, as it was shown to either enable (53) or inhibit CIL (54).

As zebrafish PGCs are not thought to use cell-ECM adhesion for their motility (52), part of the response that we observed could reflect migration toward domains where the cells can form cell-cell adhesion. It is also possible that the cellular response could have been elicited by nPGCs sensing the differences in stiffness in a mechanism that involves mechanosensitive channels (1). Consistently, mechanosensitive channels such as trpm4a, trpm7, trpv4, piezo1, and piezo2a2 are expressed in the PGCs at the stages that we studied (fig. S4F).

In addition to studying the tissue barrier in detail, we identified other domains of differential cell accumulation. These are domains that nPGCs avoided such as the axial head mesoderm (Fig. 2B, asterisk) and anterior domains lateral to the head (Fig. 2B, encircled area; region A in fig. S6B). Our initial analysis suggests that those domains have distinct properties regarding their effect on nPGC migration. The prechordal plate tissue prevents nPGC entry, albeit with a different efficiency as compared with the notochord (Fig. 4B). Unlike the notochord, the prechordal plate is not bordered by structured ECM staining (fig. S5C) and does not present a distinct Brillouin shift drop at its border (fig. S6A). The detailed behavior of nPGC upon interaction with the prechordal plate remains to be determined. In contrast to the notochord and the prechordal plate, the regions lateral to the head mesoderm (region A in fig. S6B) are more accessible for nPGCs, such that the precise underlying mechanisms responsible for the relative depletion of cells within them could differ.

In contrast with the differential localization of the nPGCs described above, an important finding was that, on the basis of their distribution, the cells are able to move freely across the germ layers. This finding is unexpected considering the differences in gene expression and in certain biophysical properties characteristic of these developing tissues (5). Following the behavior of the cells within those compartments and determining the localization and dynamics of polarity markers (55) would shed light on the mechanism responsible for the observed robustness of the amoeboid migration that nPGCs exhibit. Considering the robustness of the migration of nPGCs, it is likely that structures that they cannot cross play key roles in defining developing tissue borders where cell mixing is strongly suppressed.

In this work, we used nPGCs for probing tissue properties in zebrafish embryos. However, other cell types [e.g., metastatic cells (56)] or cell extensions [e.g., in neuronal cells (57)], when those migrate or advance in the absence of guidance cues, could analogously serve as bioprobes to reveal tissue properties that could differ from those highlighted here.

In addition to the biological data derived from the experimental scheme, in this work, we present the Landscape software—a simple, robust, automated, high-performance 3D image registration–based processing platform—that we developed and used for determining the distribution of the nPGCs and mapping tissue features. As demonstrated using Drosophila embryos, this platform can be adapted for other experimental models of different shapes (Fig. 6B and fig. S7, A and C). In these biological samples, the Landscape registration pipeline can be used for following the position of specific tissues (e.g., axial mesoderm and tracheal branches; Fig. 6, A and B), as well as that of different cell types such as glial cells (fig. S7, A to C). The Landscape software allows for quantitative analysis of both prominent and subtle phenotypes in the 3D context (Fig. 6, A and B) and for directing and focusing further biological and biophysical studies to specific parts of the tissue (Fig. 3B). Thus, we expect that the experimental paradigm and experimental tools that we present in this work will be of general use in the fields of developmental biology, cell biology, and biomedicine.


Zebrafish strains and maintenance

Zebrafish (Danio rerio) embryos from the following lines were used: wild type of the AB background, cxcr4bt26035 (22), nototk241 (34), Tg(kop:mcherry-F-nos3′UTR) (58), Tg(gsc:GFP) (21), Tg(-1ntla:CFP) (35), and Tg(βactin:h2amCherry) (59). Zebrafish were maintained on a 14-hour light/10-hour dark cycle, and fertilized eggs were collected and raised at 25°, 28°, or 31°C. Embryos were kept in 0.3× Danieau’s solution [17.4 mM NaCl, 0.21 mM KCl, 0.12 mM MgSO4·7H2O, 0.18 mM Ca(NO3)2, and 1.5 mM Hepes (pH 7.6)]. The general zebrafish maintenance was performed in compliance with the German, North-Rhine-Westphalia state law, following the regulations of the Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen and was supervised by the veterinarian office of the city of Muenster.

Drosophila strains and maintenance

The following Drosophila stocks, obtained from the Bloomington Drosophila Stock Center, were used: wild-type strains w1118 and y1 w1118, Gal4 driver line 69B-Gal4. UAS-Bnl-SBP-GFP was generated by inserting a streptavidin-binding peptide (SBP) tag followed by the enhanced GFP (eGFP) open reading frame after amino acid 432 of Bnl protein (isoform A). The UAS-Bnl-SBP-GFP transgene is inserted within the attP40 landing site on the second chromosome.

Microinjection into zebrafish embryos

Capped sense mRNAs were synthesized using the mMESSAGE mMACHINE (Thermo Fisher Scientific). One nanoliter of RNAs and/or morpholinos (Gene Tools, OR) were injected into the yolk of one-cell-stage embryos unless stated otherwise. In the experiments, the following RNAs were used: egfp-f’-nos3′UTR (30 ng/μl) (60), ezrin-ypet-nos3′UTR (30 ng/μl) (55), lifeact-mCherry-nos3′UTR (20 ng/μl) (55), mCherry-h2b-globin3′UTR (100 ng/μl) (33), and h2a-tagBFP-SV40polyA (100 ng/μl) (35). Morpholinos used were as follows: 800 μM MO-cxcr4a (AGACGATGTGTTCGTAATAAGCCAT, ZFIN-ID ZDB-MRPHLNO-070427-1), 100 μM MO-noto (GGGAATCTGCATGGCGTCTGTTTAG, ZFIN-ID ZDB-MRPHLNO-100514-1), and 350 μM MO-tbx16 (CTCTGATAGCCTGCATTATTTAGCC, ZFIN-ID ZDB-MRPHLNO-141217-1) and MO-control (CCTCTTACCTCAGTTACAATTTATA) at an 800, 100, and 350 μM concentration.

Landscape software

The Landscape software is a custom-made MATLAB-based (MathWorks, MA, version R2018a) data-driven pipeline. It allows to integrate and register 3D microscopy data derived from multiple tissue samples and automatically generates 2D distribution maps of cell positions. The software design and modules are described in the Supplementary Materials. All computational analyses were performed on a computer workstation (Linux Ubuntu 16.04.4) with the following hardware components: 16 Intel Xeon central processing units E5-2630 running at 2.40 GHz, 128 GB RAM, and 12 GB GeForce GTX Titan X graphic card. The Landscape software was also tested with MATLAB (MathWorks, MA, version R2018a) on computers with MacOS 10.13 and Windows 10. It can be run either directly in MATLAB or as a standalone executable application for Microsoft Windows operating system.

Generation of zebrafish heatmaps

Zebrafish embryos were incubated at 31°C until they reached tailbud stage (equivalent to 10 hpf at 28°C). Each embryo was individually staged as judged by its morphology, and we estimate the differences in age among embryos to be lower than 15 to 20 min. At the 10-hpf stage, the embryos were fixed in 4% phosphate-buffered paraformaldehyde (PFA) for 2 hours at room temperature (RT), washed three times in phosphate-buffered saline (PBS) (3 × 5 min) and dechorionated. For staining the nuclei, embryos were incubated in Hoechst 33342 (Thermo Fisher Scientific, 62249) at 1:10,000 in PBS + 0.1% Tween 20 (PBT) overnight at +4°C. Embryos were imaged using spinning disk confocal microscopy (for details, see the “Image acquisition and microscopy” section). The images were processed and registered with the Landscape software (see above and the Supplementary Materials). The whole-embryo heatmaps (Figs. 1E and 2, A and B, and fig. S2, A and B) were computed by considering the entire nPGC population within the 3D virtual embryo. To generate the shell heatmaps (Figs. 2D and 3B), the spatial intensity profile and morphology of the landmark [i.e., gsc:GFP expression domain (see Fig. 1B)] were used to subdivide the virtual embryo into concentric spherical shells. On the basis of this information, three heatmaps reflecting the nPGC distribution at the level of the landmark, above the landmark, and below the landmark were computed. Cell abundance value in the heatmaps represents the proportion of cells within a specific pixel relative to the total number of registered cells. The calculation was performed as follows: The heatmap was partitioned into a 90 × 90 pixel grid, and the number of cell centers within each pixel was counted and then divided by the total number of registered cells.

Preparation of Drosophila embryos

For the investigation of glial cells nucleus distribution, wild-type w1118 flies were allowed to lay eggs for 1 hour, and the eggs were aged for 10.5 hours at 25°C to enrich for dorsal closure stage (late stage 14). The embryos were dechorionated and fixed according to previously described protocols (61). Fixed embryos were stored in absolute ethanol at 4°C. Several rounds of these embryo collections were done, and pooled embryos were used within 5 to 6 days of fixation. Immunostaining was performed according to previously described protocols (61). Primary antibody mouse anti-Repo [1:10; DSHB (Developmental Studies Hybridoma Bank) 8D12, (62)] was used. Rhodamine phalloidin (1:100; Invitrogen) was added together with fluorophore-conjugated secondary antibody mouse Alexa Fluor 488 (1:200; Invitrogen). Last, a few drops of VECTASHIELD with 4′,6-diamidino-2-phenylindole (DAPI) (Vector Laboratories) were added, and the embryos were stored at 4°C for at least 8 hours before imaging. The embryos were imaged using a spinning disk microscope (for details, see the “Image acquisition and microscopy” section). Images were processed and registered using the Landscape program (see above and the Supplementary Materials).

For the investigation of the trachea, control (y w) flies and w;; 69B-Gal4 males crossed with w; UAS-Bnl-SBP-eGFP female flies were allowed to lay eggs for 1 hour at 25°C. Embryos were aged for 15 hours at 22°C and subsequently fixed in 4% formaldehyde in PBS/heptane for 20 min and devitellinized by shaking in methanol/heptane. Anti–horseradish peroxidase–Alexa Fluor 647 (1:500; Dianova) was used to label neurons, DAPI (1:1000; Sigma-Aldrich) was used to label DNA, and luminal chitin was detected using the chitin-binding domain from Bacillus circulans chitinase A1 conjugated with SNAP-Surface Alexa Fluor 488 produced as described previously (63). Embryos were imaged using a confocal microscope (for details, see the “Image acquisition and microscopy” section).

Proof of principle for zebrafish and Drosophila registration

The embryos were fixed and prepared for imaging as described above. The zebrafish embryo was imaged in two different imaging orientations. A smaller-sized version of the datasets was computationally generated using the Fiji software (64). For the Drosophila embryo, the image was acquired once. The alternative orientation was computationally generated by transforming the original image with the Fiji software (64). All orientations were independently registered with the Landscape program. Then, the images were transformed using the registration matrix and overlaid.

Mapping PGC positions to specific germ layer derivatives

Tg(gsc:GFP) embryos were fixed at the 10-hpf stage in 4% PFA for 2 hours in RT. After dechorionation, the RNAscope procedure was performed as in (65). The embryos were imaged using an LSM 710 confocal microscope (Zeiss). As the embryos had nuclei stained with Hoechst and GFP-positive landmark, the signal from different germ layers was segmented and registered similarly as for heatmap generation (see above). The segmented signal was masked on the registered PGC distribution, such that only the PGCs in this volume were counted and visualized. The following RNAscope probes were used: sox2 (28) (catalog no. 494861-C3) and tp63 (27) (catalog no. 475511-C3) to label ectoderm derivatives; sox17 (29) (catalog no. 494711-C3) to label endoderm derivatives; and pcdh8 (30) (catalog no. 494741-C3), pax8 (31) (catalog no. 494721-C3), and ntla (32) (catalog no. 4835511-C2) to label mesodermal derivatives.

RNAscope of wild-type and noto embryos

Tg(gsc:GFP) nototk241+/− fish were incrossed, and the embryos were fixed at the 10-hpf stage in 4% PFA for 2 hours in RT. The RNAscope procedure was performed as in (65). The embryos were imaged with a spinning disk confocal microscope. The following RNAscope probes were used: cxcl12a (65) (catalog no. 406481) and nanos (65) (catalog no. 404521-C2).

Image acquisition and microscopy

Before image acquisition and microscopy of live samples, embryos were dechorionated, and those older than 24 hpf were anesthetized with Tricaine (A5040, Sigma-Aldrich) in 0.3× Danieau’s solution. Embryos were placed onto agarose-coated ramps covered with Danieau’s solution and manually oriented. If not stated otherwise, embryos were oriented with the dorsal side up, except for 24-hpf embryos, which were oriented laterally. Samples were maintained at 28°C using a heated stage (PeCon, TempController 2000-2).

Upright spinning disk microscopy (10× and 40×) was performed using a Carl Zeiss Axio Imager Z1 microscope equipped with a Yokogawa CSUX1FW-06P-01 spinning disk unit. Imaging was performed using a Hamamatsu ORCA-Flash4.0 LT C11440 camera and Visitron Systems acquisition software (VisiView version Acquisition (10×) was conducted considering 400-μm z stacks (40 μm by 10 μm z planes) and 3-min time intervals between stacks (Fig. 4A and movies S5 and S6). Acquisition (40×) was performed considering 30-μm-thick z projections with 10-μm intervals and a 1-min time interval (Fig. 5A fig. S4A, and movies S7 to S9). Imaging of live zebrafish embryos was performed between 9 and 11 hpf. Recorded embryos were kept in a 28°C incubator until the next morning and phenotypically assessed for proper development using a Leica MZ7.5 stereomicroscope. For zebrafish heatmaps, embryos were recorded with a 10× water immersion objective, considering 400-μm z stacks (40 μm by 10 μm z planes) (Fig. 1, A, B, D, and E; Fig. 2, A, B, and D; Fig. 3B; fig. S1, C and D; fig. S2, A and B; and fig. S3A). For Drosophila heatmaps of glial cell nuclei, embryos were aligned ventrally on a coverslip (24 × 40, Thermo Fisher Scientific) and recorded in three partially overlapping parts with a 40× oil immersion objective, considering 200-μm z stacks (40 μm by 5 μm z planes) (fig. S7, A to C).

For heatmaps of Drosophila tracheae, embryos were mounted in ProLong Gold medium (Thermo Fisher Scientific) and covered with a coverslip (0.17 mm, grade no. 1.5). Embryos were imaged on a Leica SP8 confocal microscope using a 40×/1.3 numerical aperture (NA) oil immersion objective and Hybrid Detectors. Z stacks at 1-μm intervals over 60 μm from the ventral surface of the embryos were acquired (Fig. 7B and fig. S7D).

Confocal images were acquired using a Carl Zeiss LSM 710 microscope equipped with a 20× and 63× water immersion objective and controlled by the ZEN software (Zeiss, version 2010B SP1, 6.0). Acquisition (20×) was conducted considering 400-μm-thick confocal plane images with 10-μm intervals (Fig. 3A). RNAscope-labeled embryos were acquired either with a 20× water immersion objective, considering 400-μm z stacks (80 μm by 5 μm) (fig. S2D) or with 10× water immersion objective considering 400-μm z stacks (40 μm by 10 μm) (fig. S3D). Immunostained embryos were recorded with a 63× water immersion objective considering 2-μm-thin optical slices (fig. S5C).

Brillouin microscopy

Brillouin microscopy probes viscoelastic properties of cells and tissues through a light-matter interaction (“scattering”) (14). Analyzing the backscattered light’s spectrum allows interpretation of a material’s mechanical properties. In particular, the peak position of the frequency shifted light (“Brillouin shift”) is determined by the real part of the Longitudinal modulus, which accounts for the elastic behavior, that is, the stored elastic energy inside a sample. Therefore, we report the Brillouin shift as a direct metric of mechanical properties as it is the parameter measured in our experiment. Note that Brillouin scattering probes elasticity via a different physical process with respect to other mechanical measurements performed, for example, with an atomic force microscope. The “stiffness” measured by Brillouin scattering is therefore fundamentally different from the often used tensile (Young’s) modulus. In our experiments, 10-hpf stage embryos were dechorionated, mounted in 0.6% low–melting point agarose drop (A0701, Sigma-Aldrich) and imaged with a custom-made Brillouin microscope [for detailed description of the microscope, see (46)]. Images were acquired in the middle of the notochord (Fig. 5E) and at the widest point of the prechordal plate (fig. S6A) with a 40× (0.85 to 1.00 effective NA) objective, a step size of 0.75 μm (both in the x and z planes), 60-ms exposure time per pixel and 15-mW laser power. Before Brillouin measurement, a confocal image of GFP was acquired with the 0.75-μm step size (both in the x and z planes). As the Brillouin measurement is comparatively slow (12 min for the notochord and 20 min for the prechordal plate), the GFP signal (acquired in 10 s) is not perfectly overlapping with the measured tissue and serves as a general reference only. As a verification experiment, we imaged the notochord of 3 days post-fertilization (dpf) wild-type embryo 250 μm from the posterior end [fig. S5B, as in (46)] using the same imaging settings as for the 10-hpf embryos. To quantify the Brillouin shifts, 5-pixel-wide lines were drawn across the middle of notochord (for 10-hpf and 3-dpf embryos) and across the prechordal plate, 20 μm away from the yolk and parallel to the yolk.

Quantification of crossing behavior

The quantitative analysis of crossing behavior was based on the 10× spinning disk confocal microscopy spatiotemporally following the migration of PGCs toward the midline. The 4D data were analyzed with the Imaris software (Bitplane, versions 9.3.0 and 9.5.1), allowing a detailed analysis of nPGCs crossing the midline (Fig. 4 and movies S5 and S6). Crossing events were considered only for nPGCs crossing the midline at the same height as GFP-expressing cells. nPGCs crossing below or above GFP-expressing cells were not classified as crossing.

Quantification of impact and deflection angle

The angle of impact was defined as the angle between the direction of cell migration toward the notochord and the notochord length at the moment of impact. The angle of deflection was defined as the angle between the direction of cell migration away from the notochord and the notochord at the moment of deflection (Fig. 5B and fig. S4, B and C). Angles were measured using the angle tool of the Fiji software (64). The classification of cell behaviors upon impact with the notochord (fig. S4B) was performed using a MATLAB-based (MathWorks, MA, version R2018a) k-means clustering algorithm considering k = 3 groups.

Quantification of duration of contact

The duration of contact was defined as the time between the first and last time point of contact (i.e., moment of impact until moment of deflection) (Fig. 5C). The time lapse data of the nPGCs were analyzed using the Fiji software (64).

Quantification of actin dynamics

Actin level was estimated by the LifeAct-mCherry signal intensity as measured at the cell front using the free hand tool and region of interest (ROI) manager of the Fiji software (64). The front of the cell was defined as an ROI covering an area of approximately 30% away from the tip of the cell (i.e., reflects the area that is ahead of the nucleus). Defining the cell axis along the direction of migration, the cell front area was divided into two sectors, i.e., the part of the front that was in direct contact with the barrier and the other part of the front, away from the barrier.

Time lapse analysis

cxcr4b−/−, Tg(kop:mCherry-F-3′nos; gsc:GFP) embryos were injected with h2a-tagBFP RNA at the one-cell stage and imaged between 9.5- and 10.5-hpf stages at 3-min time intervals. The PGCs and somatic nuclei were tracked using the Imaris software (Bitplane, versions 9.3.0 and 9.5.1). For each PGC migrating within region A, a PGC migrating in region B on a similar longitude was selected from the same embryo. The PGC tracks were corrected using the tracks of somatic nuclei, and the corrected cell speed and track straightness were extracted.

Volume measurements

PGCs and somatic nuclei from fixed 10-hpf cxcrb−/−, Tg(kop:mCherry-F-3′nos; gsc:GFP) and Tg(βactin:h2amCherry, gsc:GFP) embryos, respectively, were imaged using a Zeiss Lightsheet microscope Z.1 equipped with a 20× water immersion objective and controlled by the ZEN software (Zeiss, version 2014 SP1, 9.0). Before imaging, embryos were ramped in 1% low–melting point agarose (Invitrogen). The volume of PGCs and the embryo (dorsal half, 400 μm from dorsal to ventral) was determined using the surface function of the Imaris software (Bitplane, versions 9.3.1 and 9.5.1). The average volume of the embryo, excluding the yolk, considered for all analyses is 2.98 × 107 μm3 (fig. S1A). The average volume of a PGC is 3.7 × 103 μm3 (fig. S1B). To achieve significant coverage of the embryo by PGCs, i.e., PGCs reaching similar volume as the embryo, 8058 cells would be required.

Cell density analysis

Nucleus and nPGC distribution maps at the level of the notochord tissue were considered when analyzing the correlation between the density of nuclei (and thus of cells) and of nPGCs. The embryos used for the nucleus distribution map (N = 8) were imaged using light-sheet microscopy. The data for the nPGC distribution map are the same as in Fig. 2D. The distribution maps were scaled in the same way to allow for proper analysis. Averaged distribution profiles were extracted from selected areas (10 pixels wide) along the mediolateral and anteroposterior axes (see fig. S5A, selected areas and corresponding line plots 1 and 2). A correlation test between the nucleus density and the nPGC distribution profiles was conducted by using a linear regression analysis; mediolateral (1) and anteroposterior (2) distributions revealed the following results: (i) y = − 0.572x + 0.443, R2 = 0.0975; and (ii) y = 0.7647x − 0.0365, R2 = 0.7099.


Zebrafish embryos were fixed in 4% PFA overnight at +4°C, rinsed in PBT, dechorionated, and dehydrated in 100% methanol (over the weekend at −20°C). Embryos were then rehydrated into PBT (75, 50, and 25% methanol in PBT, each 5 min at RT), briefly washed with PBT, digested in proteinase K (10 μg/ml) for 15 s, refixed in 4% PFA for 20 min at RT, briefly washed with PBT and blocked for at least 3 hours in BSA (2 mg/ml) and 2% goat serum in PBT at RT. Subsequently, embryos were incubated overnight at +4°C in rabbit anti-laminin (Sigma-Aldrich, L9393) 1:100 in blocking solution, washed in PBT, and incubated with Alexa Fluor 568 anti-rabbit immunoglobulin G (Invitrogen) secondary antibody at 1:1000 dilution in PBT overnight at +4°C. For staining the nuclei, embryos were incubated in Hoechst 33342 (Thermo Fisher Scientific, 62249) at 1:10,000 in PBT for 1 hour at RT. Before imaging, embryos were mounted in 1% low–melting point agarose (Invitrogen) for confocal imaging.

RNA in situ hybridization

Whole-mount in situ hybridization was performed as previously described (60). Digoxigenin-labeled ntl probe was synthetized according to the manufacturer’s protocol (Roche, Switzerland) (fig. S3B).

Reverse transcription polymerase chain reaction analysis

Reverse transcription polymerase chain reaction (RT-PCR) analysis was performed on the PGC complementary DNA library (9 to 12 hpf) using following primers: trpm4a (forward: CGTGATCACGATCGCAGGTTGAG, reverse: GAACACGTCGAAAAGCTGAAGCTCA), trpm7 (forward: ACATGCGGACGGTGAACTCTTACGC, reverse: TGGACTGAGCGGACTTCCCCA), trpv4 (forward: TTCTGAACACACACAGGGTGGACGG, reverse: CTGTTCTCCCAACGGTGCTGAGA), piezo1 (forward: CAAGGCACTCTTCAGCACCAGTCTG, reverse: AGGATCTTCCAGCGGGAGCCT), piezo2a2 (forward: GCGTGGACCGAATCCTCAAACTG, reverse: CCTCCGGGATCTATTGCACAGCTC), odc (forward: GACTGGCTGCTGTTCGAGA, reverse: GGGAACATCCAGACTGCTCT), and nanos (forward: ATGGCTTTTTCTCTTCTCCAAT, reverse: GTGTTCTGCTCCGGTGAGTC) (fig. S5F).

Statistical analyses

Statistical analysis was performed using GraphPad Prism software versions 6 and 8 (La Jolla, CA). Data were checked for a normal distribution using the D’Agostino-Pearson test. The specific statistical test used, including the confidence interval, sample size, and the P values are indicated in the figure legends. If not stated otherwise, all experiments were performed in at least three independent biological replicates. If applicable, multiple comparisons were statistically analyzed using Dunn’s multiple comparison test.

Data and code materials availability

The Landscape software and test datasets can be downloaded from All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data are available from authors upon request.


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We thank E.-M. Messerschmidt, I. Sandbote, U. Jordan, and I. Halbig for technical help, K. Tarbashevich for conducting the RT-PCR analysis (fig. S4F), and B. Wallmeyer for help in reconstructing the light-sheet data (fig. S5A). We thank the laboratory of C.-P. Heisenberg for sending Tg(-1ntla:CFP) embryos (Fig. 3A), the laboratory of L. Solnica-Krezel and of C.-P. Heisenberg for help in establishing the immunostaining protocol (fig. S5C). We thank N. Knubel for help with the artwork and K. Tarbashevich, M. Reichman, and C. Brennecka for critical reading of the manuscript. Funding: This work was supported by the Medical Faculty of the University of Muenster, the Cells-in-Motion Cluster of Excellence (EXC 1003, FF-2017-14, PP-2016-10), and the Deutsche Forschungsgemeinschaft (DFG, RA863/11-1, DFG, CRC1348) to E.R., by the DFG (DI 2205/2-1 and DI2205/3-1) to A.D.-M., and the European Molecular Biology Laboratory to A.D.-M. and R.P., by the DFG (CRC1348) to S.L. and C.C., by the DFG (MA 6726/1) to M.M., by the Bundesministerium für Bildung und Forschung under the project ID 05M16PMB (MED4D) and the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement no. 777826 (NoMADS) to M.B. and D.T. T.B. and B.E.V. were supported by the European Research Council (ERC, PolarizeMe, no. 771201). H.S.-I. is supported by an EMBO fellowship (ALTF 306-2018). J.P. is a member of the Medizinerkolleg (MedK) at the Medical Faculty of the University of Muenster. S.G.-T. is a fellowship holder and L.T. is a member of the CiM-IMPRS, the joint Graduate School of the Cells-in-Motion Cluster of Excellence (EXC 1003–CiM), University of Münster, and the International Max Planck Research School–Molecular Biomedicine, Münster. Author contributions: S.G.-T. and L.T. performed the experiments and analyzed the data shown in Figs. 1 (A, B, D, and E) and 2 figs. S2 (A and B) and S6B. S.G.-T. performed the experiments and analyzed the data shown in Figs. 3, 4, 5 (A to D), and 6 and figs. S1 (A to D), S3 (A to C), S4 (A to D), S5 (A and C), and S7. L.T. performed the experiments and analyzed the data shown in Figs. 2C and 5E and figs. S3D, S5B, and S6A. D.T. designed the overall Landscape software pipeline. D.T. with M.B. derived mathematical methods for each processing step and implemented the majority of the provided functionality using MathWorks MATLAB. H.S.-I. performed and analyzed, together with L.T., the data shown in Fig. 5E and figs. S5B and S6A. C.C. and S.L. designed the Drosophila experiments presented in Fig. 6B and fig. S7D. C.C. performed the imaging and the staining of Drosophila embryos shown in Fig. 6B and fig. S7D. K.J.W. collected data shown in Figs. 2 (A, B, and D) and 6A and fig. S3 (B and D). A.S. and M.M. designed the Drosophila experiments presented in fig. S7 (A to C). A.S. performed the imaging and the staining of Drosophila embryos shown in fig. S7 (A to C). P.M. collected and analyzed data presented in Figs. 1E, 3B, and 4 and fig. S2D. J.P. collected data shown in Fig. 2 (A, B, and D) and helped, together with S.G.-T., L.T., K.J.W., and D.T., to devise the idea of the 2D heatmap presentation provided by the Landscape software. P.L. and J.H. adapted, evaluated, and validated the MATLAB code for the analysis of the mutant embryo datasets (Fig. 3B). P.L. further refined the code for 3D image registration of Drosophila data (Fig 6B and fig. S7, A to C). J.H. implemented the necessary functionality for the cell distribution analysis across germ layers (Fig. 2C and fig. S2D). D.T., J.H., and F.G. developed the MATLAB graphical user interface for the Landscape software. R.S. and J.H. improved the implemented methodology for ellipsoid estimation. B.E.V. developed the simulation for the cell-barrier interaction shown in fig. S4E and movie S11, with help from T.B., E.R., and S.G.-T. R.P. built, together with A.D.-M., the Brillouin microscope. A.D.-M. supervised the Brillouin experiments and provided critical feedback concerning the analysis and interpretation of the Brillouin data shown in Fig. 5E and figs. S5B and S6A. S.G.-T., L.T., and E.R. designed the research study and interpreted the results. S.G.-T. coordinated and managed the entire project. S.G.-T., L.T., D.T., and E.R. wrote the manuscript. All authors read, commented, and approved the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Stay Connected to Science Advances

Navigate This Article